3 * Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000,
4 * 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009 by Larry Wall
7 * You may distribute under the terms of either the GNU General Public
8 * License or the Artistic License, as specified in the README file.
13 * 'I wonder what the Entish is for "yes" and "no",' he thought.
16 * [p.480 of _The Lord of the Rings_, III/iv: "Treebeard"]
22 * This file contains the code that creates, manipulates and destroys
23 * scalar values (SVs). The other types (AV, HV, GV, etc.) reuse the
24 * structure of an SV, so their creation and destruction is handled
25 * here; higher-level functions are in av.c, hv.c, and so on. Opcode
26 * level functions (eg. substr, split, join) for each of the types are
39 /* Missing proto on LynxOS */
40 char *gconvert(double, int, int, char *);
44 # define SNPRINTF_G(nv, buffer, size, ndig) \
45 quadmath_snprintf(buffer, size, "%.*Qg", (int)ndig, (NV)(nv))
47 # define SNPRINTF_G(nv, buffer, size, ndig) \
48 PERL_UNUSED_RESULT(Gconvert((NV)(nv), (int)ndig, 0, buffer))
51 #ifndef SV_COW_THRESHOLD
52 # define SV_COW_THRESHOLD 0 /* COW iff len > K */
54 #ifndef SV_COWBUF_THRESHOLD
55 # define SV_COWBUF_THRESHOLD 1250 /* COW iff len > K */
57 #ifndef SV_COW_MAX_WASTE_THRESHOLD
58 # define SV_COW_MAX_WASTE_THRESHOLD 80 /* COW iff (len - cur) < K */
60 #ifndef SV_COWBUF_WASTE_THRESHOLD
61 # define SV_COWBUF_WASTE_THRESHOLD 80 /* COW iff (len - cur) < K */
63 #ifndef SV_COW_MAX_WASTE_FACTOR_THRESHOLD
64 # define SV_COW_MAX_WASTE_FACTOR_THRESHOLD 2 /* COW iff len < (cur * K) */
66 #ifndef SV_COWBUF_WASTE_FACTOR_THRESHOLD
67 # define SV_COWBUF_WASTE_FACTOR_THRESHOLD 2 /* COW iff len < (cur * K) */
69 /* Work around compiler warnings about unsigned >= THRESHOLD when thres-
72 # define GE_COW_THRESHOLD(cur) ((cur) >= SV_COW_THRESHOLD)
74 # define GE_COW_THRESHOLD(cur) 1
76 #if SV_COWBUF_THRESHOLD
77 # define GE_COWBUF_THRESHOLD(cur) ((cur) >= SV_COWBUF_THRESHOLD)
79 # define GE_COWBUF_THRESHOLD(cur) 1
81 #if SV_COW_MAX_WASTE_THRESHOLD
82 # define GE_COW_MAX_WASTE_THRESHOLD(cur,len) (((len)-(cur)) < SV_COW_MAX_WASTE_THRESHOLD)
84 # define GE_COW_MAX_WASTE_THRESHOLD(cur,len) 1
86 #if SV_COWBUF_WASTE_THRESHOLD
87 # define GE_COWBUF_WASTE_THRESHOLD(cur,len) (((len)-(cur)) < SV_COWBUF_WASTE_THRESHOLD)
89 # define GE_COWBUF_WASTE_THRESHOLD(cur,len) 1
91 #if SV_COW_MAX_WASTE_FACTOR_THRESHOLD
92 # define GE_COW_MAX_WASTE_FACTOR_THRESHOLD(cur,len) ((len) < SV_COW_MAX_WASTE_FACTOR_THRESHOLD * (cur))
94 # define GE_COW_MAX_WASTE_FACTOR_THRESHOLD(cur,len) 1
96 #if SV_COWBUF_WASTE_FACTOR_THRESHOLD
97 # define GE_COWBUF_WASTE_FACTOR_THRESHOLD(cur,len) ((len) < SV_COWBUF_WASTE_FACTOR_THRESHOLD * (cur))
99 # define GE_COWBUF_WASTE_FACTOR_THRESHOLD(cur,len) 1
102 #define CHECK_COW_THRESHOLD(cur,len) (\
103 GE_COW_THRESHOLD((cur)) && \
104 GE_COW_MAX_WASTE_THRESHOLD((cur),(len)) && \
105 GE_COW_MAX_WASTE_FACTOR_THRESHOLD((cur),(len)) \
107 #define CHECK_COWBUF_THRESHOLD(cur,len) (\
108 GE_COWBUF_THRESHOLD((cur)) && \
109 GE_COWBUF_WASTE_THRESHOLD((cur),(len)) && \
110 GE_COWBUF_WASTE_FACTOR_THRESHOLD((cur),(len)) \
113 #ifdef PERL_UTF8_CACHE_ASSERT
114 /* if adding more checks watch out for the following tests:
115 * t/op/index.t t/op/length.t t/op/pat.t t/op/substr.t
116 * lib/utf8.t lib/Unicode/Collate/t/index.t
119 # define ASSERT_UTF8_CACHE(cache) \
120 STMT_START { if (cache) { assert((cache)[0] <= (cache)[1]); \
121 assert((cache)[2] <= (cache)[3]); \
122 assert((cache)[3] <= (cache)[1]);} \
125 # define ASSERT_UTF8_CACHE(cache) NOOP
128 static const char S_destroy[] = "DESTROY";
129 #define S_destroy_len (sizeof(S_destroy)-1)
131 /* ============================================================================
133 =head1 Allocation and deallocation of SVs.
135 An SV (or AV, HV, etc.) is allocated in two parts: the head (struct
136 sv, av, hv...) contains type and reference count information, and for
137 many types, a pointer to the body (struct xrv, xpv, xpviv...), which
138 contains fields specific to each type. Some types store all they need
139 in the head, so don't have a body.
141 In all but the most memory-paranoid configurations (ex: PURIFY), heads
142 and bodies are allocated out of arenas, which by default are
143 approximately 4K chunks of memory parcelled up into N heads or bodies.
144 Sv-bodies are allocated by their sv-type, guaranteeing size
145 consistency needed to allocate safely from arrays.
147 For SV-heads, the first slot in each arena is reserved, and holds a
148 link to the next arena, some flags, and a note of the number of slots.
149 Snaked through each arena chain is a linked list of free items; when
150 this becomes empty, an extra arena is allocated and divided up into N
151 items which are threaded into the free list.
153 SV-bodies are similar, but they use arena-sets by default, which
154 separate the link and info from the arena itself, and reclaim the 1st
155 slot in the arena. SV-bodies are further described later.
157 The following global variables are associated with arenas:
159 PL_sv_arenaroot pointer to list of SV arenas
160 PL_sv_root pointer to list of free SV structures
162 PL_body_arenas head of linked-list of body arenas
163 PL_body_roots[] array of pointers to list of free bodies of svtype
164 arrays are indexed by the svtype needed
166 A few special SV heads are not allocated from an arena, but are
167 instead directly created in the interpreter structure, eg PL_sv_undef.
168 The size of arenas can be changed from the default by setting
169 PERL_ARENA_SIZE appropriately at compile time.
171 The SV arena serves the secondary purpose of allowing still-live SVs
172 to be located and destroyed during final cleanup.
174 At the lowest level, the macros new_SV() and del_SV() grab and free
175 an SV head. (If debugging with -DD, del_SV() calls the function S_del_sv()
176 to return the SV to the free list with error checking.) new_SV() calls
177 more_sv() / sv_add_arena() to add an extra arena if the free list is empty.
178 SVs in the free list have their SvTYPE field set to all ones.
180 At the time of very final cleanup, sv_free_arenas() is called from
181 perl_destruct() to physically free all the arenas allocated since the
182 start of the interpreter.
184 The function visit() scans the SV arenas list, and calls a specified
185 function for each SV it finds which is still live - ie which has an SvTYPE
186 other than all 1's, and a non-zero SvREFCNT. visit() is used by the
187 following functions (specified as [function that calls visit()] / [function
188 called by visit() for each SV]):
190 sv_report_used() / do_report_used()
191 dump all remaining SVs (debugging aid)
193 sv_clean_objs() / do_clean_objs(),do_clean_named_objs(),
194 do_clean_named_io_objs(),do_curse()
195 Attempt to free all objects pointed to by RVs,
196 try to do the same for all objects indir-
197 ectly referenced by typeglobs too, and
198 then do a final sweep, cursing any
199 objects that remain. Called once from
200 perl_destruct(), prior to calling sv_clean_all()
203 sv_clean_all() / do_clean_all()
204 SvREFCNT_dec(sv) each remaining SV, possibly
205 triggering an sv_free(). It also sets the
206 SVf_BREAK flag on the SV to indicate that the
207 refcnt has been artificially lowered, and thus
208 stopping sv_free() from giving spurious warnings
209 about SVs which unexpectedly have a refcnt
210 of zero. called repeatedly from perl_destruct()
211 until there are no SVs left.
213 =head2 Arena allocator API Summary
215 Private API to rest of sv.c
219 new_XPVNV(), del_XPVGV(),
224 sv_report_used(), sv_clean_objs(), sv_clean_all(), sv_free_arenas()
228 * ========================================================================= */
231 * "A time to plant, and a time to uproot what was planted..."
235 # define MEM_LOG_NEW_SV(sv, file, line, func) \
236 Perl_mem_log_new_sv(sv, file, line, func)
237 # define MEM_LOG_DEL_SV(sv, file, line, func) \
238 Perl_mem_log_del_sv(sv, file, line, func)
240 # define MEM_LOG_NEW_SV(sv, file, line, func) NOOP
241 # define MEM_LOG_DEL_SV(sv, file, line, func) NOOP
244 #ifdef DEBUG_LEAKING_SCALARS
245 # define FREE_SV_DEBUG_FILE(sv) STMT_START { \
246 if ((sv)->sv_debug_file) PerlMemShared_free((sv)->sv_debug_file); \
248 # define DEBUG_SV_SERIAL(sv) \
249 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%" UVxf ": (%05ld) del_SV\n", \
250 PTR2UV(sv), (long)(sv)->sv_debug_serial))
252 # define FREE_SV_DEBUG_FILE(sv)
253 # define DEBUG_SV_SERIAL(sv) NOOP
257 # define SvARENA_CHAIN(sv) ((sv)->sv_u.svu_rv)
258 # define SvARENA_CHAIN_SET(sv,val) (sv)->sv_u.svu_rv = MUTABLE_SV((val))
259 /* Whilst I'd love to do this, it seems that things like to check on
261 # define POISON_SV_HEAD(sv) PoisonNew(sv, 1, struct STRUCT_SV)
263 # define POISON_SV_HEAD(sv) PoisonNew(&SvANY(sv), 1, void *), \
264 PoisonNew(&SvREFCNT(sv), 1, U32)
266 # define SvARENA_CHAIN(sv) SvANY(sv)
267 # define SvARENA_CHAIN_SET(sv,val) SvANY(sv) = (void *)(val)
268 # define POISON_SV_HEAD(sv)
271 /* Mark an SV head as unused, and add to free list.
273 * If SVf_BREAK is set, skip adding it to the free list, as this SV had
274 * its refcount artificially decremented during global destruction, so
275 * there may be dangling pointers to it. The last thing we want in that
276 * case is for it to be reused. */
278 #define plant_SV(p) \
280 const U32 old_flags = SvFLAGS(p); \
281 MEM_LOG_DEL_SV(p, __FILE__, __LINE__, FUNCTION__); \
282 DEBUG_SV_SERIAL(p); \
283 FREE_SV_DEBUG_FILE(p); \
285 SvFLAGS(p) = SVTYPEMASK; \
286 if (!(old_flags & SVf_BREAK)) { \
287 SvARENA_CHAIN_SET(p, PL_sv_root); \
293 #define uproot_SV(p) \
296 PL_sv_root = MUTABLE_SV(SvARENA_CHAIN(p)); \
301 /* make some more SVs by adding another arena */
307 char *chunk; /* must use New here to match call to */
308 Newx(chunk,PERL_ARENA_SIZE,char); /* Safefree() in sv_free_arenas() */
309 sv_add_arena(chunk, PERL_ARENA_SIZE, 0);
314 /* new_SV(): return a new, empty SV head */
316 #ifdef DEBUG_LEAKING_SCALARS
317 /* provide a real function for a debugger to play with */
319 S_new_SV(pTHX_ const char *file, int line, const char *func)
326 sv = S_more_sv(aTHX);
330 sv->sv_debug_optype = PL_op ? PL_op->op_type : 0;
331 sv->sv_debug_line = (U16) (PL_parser && PL_parser->copline != NOLINE
337 sv->sv_debug_inpad = 0;
338 sv->sv_debug_parent = NULL;
339 sv->sv_debug_file = PL_curcop ? savesharedpv(CopFILE(PL_curcop)): NULL;
341 sv->sv_debug_serial = PL_sv_serial++;
343 MEM_LOG_NEW_SV(sv, file, line, func);
344 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%" UVxf ": (%05ld) new_SV (from %s:%d [%s])\n",
345 PTR2UV(sv), (long)sv->sv_debug_serial, file, line, func));
349 # define new_SV(p) (p)=S_new_SV(aTHX_ __FILE__, __LINE__, FUNCTION__)
357 (p) = S_more_sv(aTHX); \
361 MEM_LOG_NEW_SV(p, __FILE__, __LINE__, FUNCTION__); \
366 /* del_SV(): return an empty SV head to the free list */
379 S_del_sv(pTHX_ SV *p)
381 PERL_ARGS_ASSERT_DEL_SV;
386 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
387 const SV * const sv = sva + 1;
388 const SV * const svend = &sva[SvREFCNT(sva)];
389 if (p >= sv && p < svend) {
395 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
396 "Attempt to free non-arena SV: 0x%" UVxf
397 pTHX__FORMAT, PTR2UV(p) pTHX__VALUE);
404 #else /* ! DEBUGGING */
406 #define del_SV(p) plant_SV(p)
408 #endif /* DEBUGGING */
412 =head1 SV Manipulation Functions
414 =for apidoc sv_add_arena
416 Given a chunk of memory, link it to the head of the list of arenas,
417 and split it into a list of free SVs.
423 S_sv_add_arena(pTHX_ char *const ptr, const U32 size, const U32 flags)
425 SV *const sva = MUTABLE_SV(ptr);
429 PERL_ARGS_ASSERT_SV_ADD_ARENA;
431 /* The first SV in an arena isn't an SV. */
432 SvANY(sva) = (void *) PL_sv_arenaroot; /* ptr to next arena */
433 SvREFCNT(sva) = size / sizeof(SV); /* number of SV slots */
434 SvFLAGS(sva) = flags; /* FAKE if not to be freed */
436 PL_sv_arenaroot = sva;
437 PL_sv_root = sva + 1;
439 svend = &sva[SvREFCNT(sva) - 1];
442 SvARENA_CHAIN_SET(sv, (sv + 1));
446 /* Must always set typemask because it's always checked in on cleanup
447 when the arenas are walked looking for objects. */
448 SvFLAGS(sv) = SVTYPEMASK;
451 SvARENA_CHAIN_SET(sv, 0);
455 SvFLAGS(sv) = SVTYPEMASK;
458 /* visit(): call the named function for each non-free SV in the arenas
459 * whose flags field matches the flags/mask args. */
462 S_visit(pTHX_ SVFUNC_t f, const U32 flags, const U32 mask)
467 PERL_ARGS_ASSERT_VISIT;
469 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
470 const SV * const svend = &sva[SvREFCNT(sva)];
472 for (sv = sva + 1; sv < svend; ++sv) {
473 if (SvTYPE(sv) != (svtype)SVTYPEMASK
474 && (sv->sv_flags & mask) == flags
487 /* called by sv_report_used() for each live SV */
490 do_report_used(pTHX_ SV *const sv)
492 if (SvTYPE(sv) != (svtype)SVTYPEMASK) {
493 PerlIO_printf(Perl_debug_log, "****\n");
500 =for apidoc sv_report_used
502 Dump the contents of all SVs not yet freed (debugging aid).
508 Perl_sv_report_used(pTHX)
511 visit(do_report_used, 0, 0);
517 /* called by sv_clean_objs() for each live SV */
520 do_clean_objs(pTHX_ SV *const ref)
524 SV * const target = SvRV(ref);
525 if (SvOBJECT(target)) {
526 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning object ref:\n "), sv_dump(ref)));
527 if (SvWEAKREF(ref)) {
528 sv_del_backref(target, ref);
534 SvREFCNT_dec_NN(target);
541 /* clear any slots in a GV which hold objects - except IO;
542 * called by sv_clean_objs() for each live GV */
545 do_clean_named_objs(pTHX_ SV *const sv)
548 assert(SvTYPE(sv) == SVt_PVGV);
549 assert(isGV_with_GP(sv));
553 /* freeing GP entries may indirectly free the current GV;
554 * hold onto it while we mess with the GP slots */
557 if ( ((obj = GvSV(sv) )) && SvOBJECT(obj)) {
558 DEBUG_D((PerlIO_printf(Perl_debug_log,
559 "Cleaning named glob SV object:\n "), sv_dump(obj)));
561 SvREFCNT_dec_NN(obj);
563 if ( ((obj = MUTABLE_SV(GvAV(sv)) )) && SvOBJECT(obj)) {
564 DEBUG_D((PerlIO_printf(Perl_debug_log,
565 "Cleaning named glob AV object:\n "), sv_dump(obj)));
567 SvREFCNT_dec_NN(obj);
569 if ( ((obj = MUTABLE_SV(GvHV(sv)) )) && SvOBJECT(obj)) {
570 DEBUG_D((PerlIO_printf(Perl_debug_log,
571 "Cleaning named glob HV object:\n "), sv_dump(obj)));
573 SvREFCNT_dec_NN(obj);
575 if ( ((obj = MUTABLE_SV(GvCV(sv)) )) && SvOBJECT(obj)) {
576 DEBUG_D((PerlIO_printf(Perl_debug_log,
577 "Cleaning named glob CV object:\n "), sv_dump(obj)));
579 SvREFCNT_dec_NN(obj);
581 SvREFCNT_dec_NN(sv); /* undo the inc above */
584 /* clear any IO slots in a GV which hold objects (except stderr, defout);
585 * called by sv_clean_objs() for each live GV */
588 do_clean_named_io_objs(pTHX_ SV *const sv)
591 assert(SvTYPE(sv) == SVt_PVGV);
592 assert(isGV_with_GP(sv));
593 if (!GvGP(sv) || sv == (SV*)PL_stderrgv || sv == (SV*)PL_defoutgv)
597 if ( ((obj = MUTABLE_SV(GvIO(sv)) )) && SvOBJECT(obj)) {
598 DEBUG_D((PerlIO_printf(Perl_debug_log,
599 "Cleaning named glob IO object:\n "), sv_dump(obj)));
601 SvREFCNT_dec_NN(obj);
603 SvREFCNT_dec_NN(sv); /* undo the inc above */
606 /* Void wrapper to pass to visit() */
608 do_curse(pTHX_ SV * const sv) {
609 if ((PL_stderrgv && GvGP(PL_stderrgv) && (SV*)GvIO(PL_stderrgv) == sv)
610 || (PL_defoutgv && GvGP(PL_defoutgv) && (SV*)GvIO(PL_defoutgv) == sv))
616 =for apidoc sv_clean_objs
618 Attempt to destroy all objects not yet freed.
624 Perl_sv_clean_objs(pTHX)
627 PL_in_clean_objs = TRUE;
628 visit(do_clean_objs, SVf_ROK, SVf_ROK);
629 /* Some barnacles may yet remain, clinging to typeglobs.
630 * Run the non-IO destructors first: they may want to output
631 * error messages, close files etc */
632 visit(do_clean_named_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
633 visit(do_clean_named_io_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
634 /* And if there are some very tenacious barnacles clinging to arrays,
635 closures, or what have you.... */
636 visit(do_curse, SVs_OBJECT, SVs_OBJECT);
637 olddef = PL_defoutgv;
638 PL_defoutgv = NULL; /* disable skip of PL_defoutgv */
639 if (olddef && isGV_with_GP(olddef))
640 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olddef));
641 olderr = PL_stderrgv;
642 PL_stderrgv = NULL; /* disable skip of PL_stderrgv */
643 if (olderr && isGV_with_GP(olderr))
644 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olderr));
645 SvREFCNT_dec(olddef);
646 PL_in_clean_objs = FALSE;
649 /* called by sv_clean_all() for each live SV */
652 do_clean_all(pTHX_ SV *const sv)
654 if (sv == (const SV *) PL_fdpid || sv == (const SV *)PL_strtab) {
655 /* don't clean pid table and strtab */
658 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning loops: SV at 0x%" UVxf "\n", PTR2UV(sv)) ));
659 SvFLAGS(sv) |= SVf_BREAK;
664 =for apidoc sv_clean_all
666 Decrement the refcnt of each remaining SV, possibly triggering a
667 cleanup. This function may have to be called multiple times to free
668 SVs which are in complex self-referential hierarchies.
674 Perl_sv_clean_all(pTHX)
677 PL_in_clean_all = TRUE;
678 cleaned = visit(do_clean_all, 0,0);
683 ARENASETS: a meta-arena implementation which separates arena-info
684 into struct arena_set, which contains an array of struct
685 arena_descs, each holding info for a single arena. By separating
686 the meta-info from the arena, we recover the 1st slot, formerly
687 borrowed for list management. The arena_set is about the size of an
688 arena, avoiding the needless malloc overhead of a naive linked-list.
690 The cost is 1 arena-set malloc per ~320 arena-mallocs, + the unused
691 memory in the last arena-set (1/2 on average). In trade, we get
692 back the 1st slot in each arena (ie 1.7% of a CV-arena, less for
693 smaller types). The recovery of the wasted space allows use of
694 small arenas for large, rare body types, by changing array* fields
695 in body_details_by_type[] below.
698 char *arena; /* the raw storage, allocated aligned */
699 size_t size; /* its size ~4k typ */
700 svtype utype; /* bodytype stored in arena */
705 /* Get the maximum number of elements in set[] such that struct arena_set
706 will fit within PERL_ARENA_SIZE, which is probably just under 4K, and
707 therefore likely to be 1 aligned memory page. */
709 #define ARENAS_PER_SET ((PERL_ARENA_SIZE - sizeof(struct arena_set*) \
710 - 2 * sizeof(int)) / sizeof (struct arena_desc))
713 struct arena_set* next;
714 unsigned int set_size; /* ie ARENAS_PER_SET */
715 unsigned int curr; /* index of next available arena-desc */
716 struct arena_desc set[ARENAS_PER_SET];
720 =for apidoc sv_free_arenas
722 Deallocate the memory used by all arenas. Note that all the individual SV
723 heads and bodies within the arenas must already have been freed.
729 Perl_sv_free_arenas(pTHX)
735 /* Free arenas here, but be careful about fake ones. (We assume
736 contiguity of the fake ones with the corresponding real ones.) */
738 for (sva = PL_sv_arenaroot; sva; sva = svanext) {
739 svanext = MUTABLE_SV(SvANY(sva));
740 while (svanext && SvFAKE(svanext))
741 svanext = MUTABLE_SV(SvANY(svanext));
748 struct arena_set *aroot = (struct arena_set*) PL_body_arenas;
751 struct arena_set *current = aroot;
754 assert(aroot->set[i].arena);
755 Safefree(aroot->set[i].arena);
763 i = PERL_ARENA_ROOTS_SIZE;
765 PL_body_roots[i] = 0;
772 Here are mid-level routines that manage the allocation of bodies out
773 of the various arenas. There are 4 kinds of arenas:
775 1. SV-head arenas, which are discussed and handled above
776 2. regular body arenas
777 3. arenas for reduced-size bodies
780 Arena types 2 & 3 are chained by body-type off an array of
781 arena-root pointers, which is indexed by svtype. Some of the
782 larger/less used body types are malloced singly, since a large
783 unused block of them is wasteful. Also, several svtypes dont have
784 bodies; the data fits into the sv-head itself. The arena-root
785 pointer thus has a few unused root-pointers (which may be hijacked
786 later for arena type 4)
788 3 differs from 2 as an optimization; some body types have several
789 unused fields in the front of the structure (which are kept in-place
790 for consistency). These bodies can be allocated in smaller chunks,
791 because the leading fields arent accessed. Pointers to such bodies
792 are decremented to point at the unused 'ghost' memory, knowing that
793 the pointers are used with offsets to the real memory.
795 Allocation of SV-bodies is similar to SV-heads, differing as follows;
796 the allocation mechanism is used for many body types, so is somewhat
797 more complicated, it uses arena-sets, and has no need for still-live
800 At the outermost level, (new|del)_X*V macros return bodies of the
801 appropriate type. These macros call either (new|del)_body_type or
802 (new|del)_body_allocated macro pairs, depending on specifics of the
803 type. Most body types use the former pair, the latter pair is used to
804 allocate body types with "ghost fields".
806 "ghost fields" are fields that are unused in certain types, and
807 consequently don't need to actually exist. They are declared because
808 they're part of a "base type", which allows use of functions as
809 methods. The simplest examples are AVs and HVs, 2 aggregate types
810 which don't use the fields which support SCALAR semantics.
812 For these types, the arenas are carved up into appropriately sized
813 chunks, we thus avoid wasted memory for those unaccessed members.
814 When bodies are allocated, we adjust the pointer back in memory by the
815 size of the part not allocated, so it's as if we allocated the full
816 structure. (But things will all go boom if you write to the part that
817 is "not there", because you'll be overwriting the last members of the
818 preceding structure in memory.)
820 We calculate the correction using the STRUCT_OFFSET macro on the first
821 member present. If the allocated structure is smaller (no initial NV
822 actually allocated) then the net effect is to subtract the size of the NV
823 from the pointer, to return a new pointer as if an initial NV were actually
824 allocated. (We were using structures named *_allocated for this, but
825 this turned out to be a subtle bug, because a structure without an NV
826 could have a lower alignment constraint, but the compiler is allowed to
827 optimised accesses based on the alignment constraint of the actual pointer
828 to the full structure, for example, using a single 64 bit load instruction
829 because it "knows" that two adjacent 32 bit members will be 8-byte aligned.)
831 This is the same trick as was used for NV and IV bodies. Ironically it
832 doesn't need to be used for NV bodies any more, because NV is now at
833 the start of the structure. IV bodies, and also in some builds NV bodies,
834 don't need it either, because they are no longer allocated.
836 In turn, the new_body_* allocators call S_new_body(), which invokes
837 new_body_inline macro, which takes a lock, and takes a body off the
838 linked list at PL_body_roots[sv_type], calling Perl_more_bodies() if
839 necessary to refresh an empty list. Then the lock is released, and
840 the body is returned.
842 Perl_more_bodies allocates a new arena, and carves it up into an array of N
843 bodies, which it strings into a linked list. It looks up arena-size
844 and body-size from the body_details table described below, thus
845 supporting the multiple body-types.
847 If PURIFY is defined, or PERL_ARENA_SIZE=0, arenas are not used, and
848 the (new|del)_X*V macros are mapped directly to malloc/free.
850 For each sv-type, struct body_details bodies_by_type[] carries
851 parameters which control these aspects of SV handling:
853 Arena_size determines whether arenas are used for this body type, and if
854 so, how big they are. PURIFY or PERL_ARENA_SIZE=0 set this field to
855 zero, forcing individual mallocs and frees.
857 Body_size determines how big a body is, and therefore how many fit into
858 each arena. Offset carries the body-pointer adjustment needed for
859 "ghost fields", and is used in *_allocated macros.
861 But its main purpose is to parameterize info needed in
862 Perl_sv_upgrade(). The info here dramatically simplifies the function
863 vs the implementation in 5.8.8, making it table-driven. All fields
864 are used for this, except for arena_size.
866 For the sv-types that have no bodies, arenas are not used, so those
867 PL_body_roots[sv_type] are unused, and can be overloaded. In
868 something of a special case, SVt_NULL is borrowed for HE arenas;
869 PL_body_roots[HE_SVSLOT=SVt_NULL] is filled by S_more_he, but the
870 bodies_by_type[SVt_NULL] slot is not used, as the table is not
875 struct body_details {
876 U8 body_size; /* Size to allocate */
877 U8 copy; /* Size of structure to copy (may be shorter) */
878 U8 offset; /* Size of unalloced ghost fields to first alloced field*/
879 PERL_BITFIELD8 type : 4; /* We have space for a sanity check. */
880 PERL_BITFIELD8 cant_upgrade : 1;/* Cannot upgrade this type */
881 PERL_BITFIELD8 zero_nv : 1; /* zero the NV when upgrading from this */
882 PERL_BITFIELD8 arena : 1; /* Allocated from an arena */
883 U32 arena_size; /* Size of arena to allocate */
886 #define ALIGNED_TYPE_NAME(name) name##_aligned
887 #define ALIGNED_TYPE(name) \
892 } ALIGNED_TYPE_NAME(name);
894 ALIGNED_TYPE(regexp);
908 /* With -DPURFIY we allocate everything directly, and don't use arenas.
909 This seems a rather elegant way to simplify some of the code below. */
910 #define HASARENA FALSE
912 #define HASARENA TRUE
914 #define NOARENA FALSE
916 /* Size the arenas to exactly fit a given number of bodies. A count
917 of 0 fits the max number bodies into a PERL_ARENA_SIZE.block,
918 simplifying the default. If count > 0, the arena is sized to fit
919 only that many bodies, allowing arenas to be used for large, rare
920 bodies (XPVFM, XPVIO) without undue waste. The arena size is
921 limited by PERL_ARENA_SIZE, so we can safely oversize the
924 #define FIT_ARENA0(body_size) \
925 ((size_t)(PERL_ARENA_SIZE / body_size) * body_size)
926 #define FIT_ARENAn(count,body_size) \
927 ( count * body_size <= PERL_ARENA_SIZE) \
928 ? count * body_size \
929 : FIT_ARENA0 (body_size)
930 #define FIT_ARENA(count,body_size) \
932 ? FIT_ARENAn (count, body_size) \
933 : FIT_ARENA0 (body_size))
935 /* Calculate the length to copy. Specifically work out the length less any
936 final padding the compiler needed to add. See the comment in sv_upgrade
937 for why copying the padding proved to be a bug. */
939 #define copy_length(type, last_member) \
940 STRUCT_OFFSET(type, last_member) \
941 + sizeof (((type*)SvANY((const SV *)0))->last_member)
943 static const struct body_details bodies_by_type[] = {
944 /* HEs use this offset for their arena. */
945 { 0, 0, 0, SVt_NULL, FALSE, NONV, NOARENA, 0 },
947 /* IVs are in the head, so the allocation size is 0. */
949 sizeof(IV), /* This is used to copy out the IV body. */
950 STRUCT_OFFSET(XPVIV, xiv_iv), SVt_IV, FALSE, NONV,
951 NOARENA /* IVS don't need an arena */, 0
956 STRUCT_OFFSET(XPVNV, xnv_u),
957 SVt_NV, FALSE, HADNV, NOARENA, 0 },
959 { sizeof(NV), sizeof(NV),
960 STRUCT_OFFSET(XPVNV, xnv_u),
961 SVt_NV, FALSE, HADNV, HASARENA, FIT_ARENA(0, sizeof(NV)) },
964 { sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur),
965 copy_length(XPV, xpv_len) - STRUCT_OFFSET(XPV, xpv_cur),
966 + STRUCT_OFFSET(XPV, xpv_cur),
967 SVt_PV, FALSE, NONV, HASARENA,
968 FIT_ARENA(0, sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur)) },
970 { sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur),
971 copy_length(XINVLIST, is_offset) - STRUCT_OFFSET(XPV, xpv_cur),
972 + STRUCT_OFFSET(XPV, xpv_cur),
973 SVt_INVLIST, TRUE, NONV, HASARENA,
974 FIT_ARENA(0, sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur)) },
976 { sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur),
977 copy_length(XPVIV, xiv_u) - STRUCT_OFFSET(XPV, xpv_cur),
978 + STRUCT_OFFSET(XPV, xpv_cur),
979 SVt_PVIV, FALSE, NONV, HASARENA,
980 FIT_ARENA(0, sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur)) },
982 { sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur),
983 copy_length(XPVNV, xnv_u) - STRUCT_OFFSET(XPV, xpv_cur),
984 + STRUCT_OFFSET(XPV, xpv_cur),
985 SVt_PVNV, FALSE, HADNV, HASARENA,
986 FIT_ARENA(0, sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur)) },
988 { sizeof(XPVMG), copy_length(XPVMG, xnv_u), 0, SVt_PVMG, FALSE, HADNV,
989 HASARENA, FIT_ARENA(0, sizeof(XPVMG)) },
991 { sizeof(ALIGNED_TYPE_NAME(regexp)),
994 SVt_REGEXP, TRUE, NONV, HASARENA,
995 FIT_ARENA(0, sizeof(ALIGNED_TYPE_NAME(regexp)))
998 { sizeof(ALIGNED_TYPE_NAME(XPVGV)), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV,
999 HASARENA, FIT_ARENA(0, sizeof(ALIGNED_TYPE_NAME(XPVGV))) },
1001 { sizeof(ALIGNED_TYPE_NAME(XPVLV)), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV,
1002 HASARENA, FIT_ARENA(0, sizeof(ALIGNED_TYPE_NAME(XPVLV))) },
1004 { sizeof(ALIGNED_TYPE_NAME(XPVAV)),
1005 copy_length(XPVAV, xav_alloc),
1007 SVt_PVAV, TRUE, NONV, HASARENA,
1008 FIT_ARENA(0, sizeof(ALIGNED_TYPE_NAME(XPVAV))) },
1010 { sizeof(ALIGNED_TYPE_NAME(XPVHV)),
1011 copy_length(XPVHV, xhv_max),
1013 SVt_PVHV, TRUE, NONV, HASARENA,
1014 FIT_ARENA(0, sizeof(ALIGNED_TYPE_NAME(XPVHV))) },
1016 { sizeof(ALIGNED_TYPE_NAME(XPVCV)),
1019 SVt_PVCV, TRUE, NONV, HASARENA,
1020 FIT_ARENA(0, sizeof(ALIGNED_TYPE_NAME(XPVCV))) },
1022 { sizeof(ALIGNED_TYPE_NAME(XPVFM)),
1025 SVt_PVFM, TRUE, NONV, NOARENA,
1026 FIT_ARENA(20, sizeof(ALIGNED_TYPE_NAME(XPVFM))) },
1028 { sizeof(ALIGNED_TYPE_NAME(XPVIO)),
1031 SVt_PVIO, TRUE, NONV, HASARENA,
1032 FIT_ARENA(24, sizeof(ALIGNED_TYPE_NAME(XPVIO))) },
1035 #define new_body_allocated(sv_type) \
1036 (void *)((char *)S_new_body(aTHX_ sv_type) \
1037 - bodies_by_type[sv_type].offset)
1039 /* return a thing to the free list */
1041 #define del_body(thing, root) \
1043 void ** const thing_copy = (void **)thing; \
1044 *thing_copy = *root; \
1045 *root = (void*)thing_copy; \
1049 #if !(NVSIZE <= IVSIZE)
1050 # define new_XNV() safemalloc(sizeof(XPVNV))
1052 #define new_XPVNV() safemalloc(sizeof(XPVNV))
1053 #define new_XPVMG() safemalloc(sizeof(XPVMG))
1055 #define del_XPVGV(p) safefree(p)
1059 #if !(NVSIZE <= IVSIZE)
1060 # define new_XNV() new_body_allocated(SVt_NV)
1062 #define new_XPVNV() new_body_allocated(SVt_PVNV)
1063 #define new_XPVMG() new_body_allocated(SVt_PVMG)
1065 #define del_XPVGV(p) del_body(p + bodies_by_type[SVt_PVGV].offset, \
1066 &PL_body_roots[SVt_PVGV])
1070 /* no arena for you! */
1072 #define new_NOARENA(details) \
1073 safemalloc((details)->body_size + (details)->offset)
1074 #define new_NOARENAZ(details) \
1075 safecalloc((details)->body_size + (details)->offset, 1)
1078 Perl_more_bodies (pTHX_ const svtype sv_type, const size_t body_size,
1079 const size_t arena_size)
1081 void ** const root = &PL_body_roots[sv_type];
1082 struct arena_desc *adesc;
1083 struct arena_set *aroot = (struct arena_set *) PL_body_arenas;
1087 const size_t good_arena_size = Perl_malloc_good_size(arena_size);
1088 #if defined(DEBUGGING) && defined(PERL_GLOBAL_STRUCT)
1091 #if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT)
1092 static bool done_sanity_check;
1094 /* PERL_GLOBAL_STRUCT cannot coexist with global
1095 * variables like done_sanity_check. */
1096 if (!done_sanity_check) {
1097 unsigned int i = SVt_LAST;
1099 done_sanity_check = TRUE;
1102 assert (bodies_by_type[i].type == i);
1108 /* may need new arena-set to hold new arena */
1109 if (!aroot || aroot->curr >= aroot->set_size) {
1110 struct arena_set *newroot;
1111 Newxz(newroot, 1, struct arena_set);
1112 newroot->set_size = ARENAS_PER_SET;
1113 newroot->next = aroot;
1115 PL_body_arenas = (void *) newroot;
1116 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot));
1119 /* ok, now have arena-set with at least 1 empty/available arena-desc */
1120 curr = aroot->curr++;
1121 adesc = &(aroot->set[curr]);
1122 assert(!adesc->arena);
1124 Newx(adesc->arena, good_arena_size, char);
1125 adesc->size = good_arena_size;
1126 adesc->utype = sv_type;
1127 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %" UVuf "\n",
1128 curr, (void*)adesc->arena, (UV)good_arena_size));
1130 start = (char *) adesc->arena;
1132 /* Get the address of the byte after the end of the last body we can fit.
1133 Remember, this is integer division: */
1134 end = start + good_arena_size / body_size * body_size;
1136 /* computed count doesn't reflect the 1st slot reservation */
1137 #if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE)
1138 DEBUG_m(PerlIO_printf(Perl_debug_log,
1139 "arena %p end %p arena-size %d (from %d) type %d "
1141 (void*)start, (void*)end, (int)good_arena_size,
1142 (int)arena_size, sv_type, (int)body_size,
1143 (int)good_arena_size / (int)body_size));
1145 DEBUG_m(PerlIO_printf(Perl_debug_log,
1146 "arena %p end %p arena-size %d type %d size %d ct %d\n",
1147 (void*)start, (void*)end,
1148 (int)arena_size, sv_type, (int)body_size,
1149 (int)good_arena_size / (int)body_size));
1151 *root = (void *)start;
1154 /* Where the next body would start: */
1155 char * const next = start + body_size;
1158 /* This is the last body: */
1159 assert(next == end);
1161 *(void **)start = 0;
1165 *(void**) start = (void *)next;
1170 /* grab a new thing from the free list, allocating more if necessary.
1171 The inline version is used for speed in hot routines, and the
1172 function using it serves the rest (unless PURIFY).
1174 #define new_body_inline(xpv, sv_type) \
1176 void ** const r3wt = &PL_body_roots[sv_type]; \
1177 xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \
1178 ? *((void **)(r3wt)) : Perl_more_bodies(aTHX_ sv_type, \
1179 bodies_by_type[sv_type].body_size,\
1180 bodies_by_type[sv_type].arena_size)); \
1181 *(r3wt) = *(void**)(xpv); \
1187 S_new_body(pTHX_ const svtype sv_type)
1190 new_body_inline(xpv, sv_type);
1196 static const struct body_details fake_rv =
1197 { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 };
1200 =for apidoc sv_upgrade
1202 Upgrade an SV to a more complex form. Generally adds a new body type to the
1203 SV, then copies across as much information as possible from the old body.
1204 It croaks if the SV is already in a more complex form than requested. You
1205 generally want to use the C<SvUPGRADE> macro wrapper, which checks the type
1206 before calling C<sv_upgrade>, and hence does not croak. See also
1213 Perl_sv_upgrade(pTHX_ SV *const sv, svtype new_type)
1217 const svtype old_type = SvTYPE(sv);
1218 const struct body_details *new_type_details;
1219 const struct body_details *old_type_details
1220 = bodies_by_type + old_type;
1221 SV *referent = NULL;
1223 PERL_ARGS_ASSERT_SV_UPGRADE;
1225 if (old_type == new_type)
1228 /* This clause was purposefully added ahead of the early return above to
1229 the shared string hackery for (sort {$a <=> $b} keys %hash), with the
1230 inference by Nick I-S that it would fix other troublesome cases. See
1231 changes 7162, 7163 (f130fd4589cf5fbb24149cd4db4137c8326f49c1 and parent)
1233 Given that shared hash key scalars are no longer PVIV, but PV, there is
1234 no longer need to unshare so as to free up the IVX slot for its proper
1235 purpose. So it's safe to move the early return earlier. */
1237 if (new_type > SVt_PVMG && SvIsCOW(sv)) {
1238 sv_force_normal_flags(sv, 0);
1241 old_body = SvANY(sv);
1243 /* Copying structures onto other structures that have been neatly zeroed
1244 has a subtle gotcha. Consider XPVMG
1246 +------+------+------+------+------+-------+-------+
1247 | NV | CUR | LEN | IV | MAGIC | STASH |
1248 +------+------+------+------+------+-------+-------+
1249 0 4 8 12 16 20 24 28
1251 where NVs are aligned to 8 bytes, so that sizeof that structure is
1252 actually 32 bytes long, with 4 bytes of padding at the end:
1254 +------+------+------+------+------+-------+-------+------+
1255 | NV | CUR | LEN | IV | MAGIC | STASH | ??? |
1256 +------+------+------+------+------+-------+-------+------+
1257 0 4 8 12 16 20 24 28 32
1259 so what happens if you allocate memory for this structure:
1261 +------+------+------+------+------+-------+-------+------+------+...
1262 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME |
1263 +------+------+------+------+------+-------+-------+------+------+...
1264 0 4 8 12 16 20 24 28 32 36
1266 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you
1267 expect, because you copy the area marked ??? onto GP. Now, ??? may have
1268 started out as zero once, but it's quite possible that it isn't. So now,
1269 rather than a nicely zeroed GP, you have it pointing somewhere random.
1272 (In fact, GP ends up pointing at a previous GP structure, because the
1273 principle cause of the padding in XPVMG getting garbage is a copy of
1274 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now
1275 this happens to be moot because XPVGV has been re-ordered, with GP
1276 no longer after STASH)
1278 So we are careful and work out the size of used parts of all the
1286 referent = SvRV(sv);
1287 old_type_details = &fake_rv;
1288 if (new_type == SVt_NV)
1289 new_type = SVt_PVNV;
1291 if (new_type < SVt_PVIV) {
1292 new_type = (new_type == SVt_NV)
1293 ? SVt_PVNV : SVt_PVIV;
1298 if (new_type < SVt_PVNV) {
1299 new_type = SVt_PVNV;
1303 assert(new_type > SVt_PV);
1304 STATIC_ASSERT_STMT(SVt_IV < SVt_PV);
1305 STATIC_ASSERT_STMT(SVt_NV < SVt_PV);
1312 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena,
1313 there's no way that it can be safely upgraded, because perl.c
1314 expects to Safefree(SvANY(PL_mess_sv)) */
1315 assert(sv != PL_mess_sv);
1318 if (UNLIKELY(old_type_details->cant_upgrade))
1319 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf,
1320 sv_reftype(sv, 0), (UV) old_type, (UV) new_type);
1323 if (UNLIKELY(old_type > new_type))
1324 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d",
1325 (int)old_type, (int)new_type);
1327 new_type_details = bodies_by_type + new_type;
1329 SvFLAGS(sv) &= ~SVTYPEMASK;
1330 SvFLAGS(sv) |= new_type;
1332 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of
1333 the return statements above will have triggered. */
1334 assert (new_type != SVt_NULL);
1337 assert(old_type == SVt_NULL);
1338 SET_SVANY_FOR_BODYLESS_IV(sv);
1342 assert(old_type == SVt_NULL);
1343 #if NVSIZE <= IVSIZE
1344 SET_SVANY_FOR_BODYLESS_NV(sv);
1346 SvANY(sv) = new_XNV();
1352 assert(new_type_details->body_size);
1355 assert(new_type_details->arena);
1356 assert(new_type_details->arena_size);
1357 /* This points to the start of the allocated area. */
1358 new_body_inline(new_body, new_type);
1359 Zero(new_body, new_type_details->body_size, char);
1360 new_body = ((char *)new_body) - new_type_details->offset;
1362 /* We always allocated the full length item with PURIFY. To do this
1363 we fake things so that arena is false for all 16 types.. */
1364 new_body = new_NOARENAZ(new_type_details);
1366 SvANY(sv) = new_body;
1367 if (new_type == SVt_PVAV) {
1371 if (old_type_details->body_size) {
1374 /* It will have been zeroed when the new body was allocated.
1375 Lets not write to it, in case it confuses a write-back
1381 #ifndef NODEFAULT_SHAREKEYS
1382 HvSHAREKEYS_on(sv); /* key-sharing on by default */
1384 /* start with PERL_HASH_DEFAULT_HvMAX+1 buckets: */
1385 HvMAX(sv) = PERL_HASH_DEFAULT_HvMAX;
1388 /* SVt_NULL isn't the only thing upgraded to AV or HV.
1389 The target created by newSVrv also is, and it can have magic.
1390 However, it never has SvPVX set.
1392 if (old_type == SVt_IV) {
1394 } else if (old_type >= SVt_PV) {
1395 assert(SvPVX_const(sv) == 0);
1398 if (old_type >= SVt_PVMG) {
1399 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic);
1400 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash);
1402 sv->sv_u.svu_array = NULL; /* or svu_hash */
1407 /* XXX Is this still needed? Was it ever needed? Surely as there is
1408 no route from NV to PVIV, NOK can never be true */
1409 assert(!SvNOKp(sv));
1423 assert(new_type_details->body_size);
1424 /* We always allocated the full length item with PURIFY. To do this
1425 we fake things so that arena is false for all 16 types.. */
1426 if(new_type_details->arena) {
1427 /* This points to the start of the allocated area. */
1428 new_body_inline(new_body, new_type);
1429 Zero(new_body, new_type_details->body_size, char);
1430 new_body = ((char *)new_body) - new_type_details->offset;
1432 new_body = new_NOARENAZ(new_type_details);
1434 SvANY(sv) = new_body;
1436 if (old_type_details->copy) {
1437 /* There is now the potential for an upgrade from something without
1438 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */
1439 int offset = old_type_details->offset;
1440 int length = old_type_details->copy;
1442 if (new_type_details->offset > old_type_details->offset) {
1443 const int difference
1444 = new_type_details->offset - old_type_details->offset;
1445 offset += difference;
1446 length -= difference;
1448 assert (length >= 0);
1450 Copy((char *)old_body + offset, (char *)new_body + offset, length,
1454 #ifndef NV_ZERO_IS_ALLBITS_ZERO
1455 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a
1456 * correct 0.0 for us. Otherwise, if the old body didn't have an
1457 * NV slot, but the new one does, then we need to initialise the
1458 * freshly created NV slot with whatever the correct bit pattern is
1460 if (old_type_details->zero_nv && !new_type_details->zero_nv
1461 && !isGV_with_GP(sv))
1465 if (UNLIKELY(new_type == SVt_PVIO)) {
1466 IO * const io = MUTABLE_IO(sv);
1467 GV *iogv = gv_fetchpvs("IO::File::", GV_ADD, SVt_PVHV);
1470 /* Clear the stashcache because a new IO could overrule a package
1472 DEBUG_o(Perl_deb(aTHX_ "sv_upgrade clearing PL_stashcache\n"));
1473 hv_clear(PL_stashcache);
1475 SvSTASH_set(io, MUTABLE_HV(SvREFCNT_inc(GvHV(iogv))));
1476 IoPAGE_LEN(sv) = 60;
1478 if (old_type < SVt_PV) {
1479 /* referent will be NULL unless the old type was SVt_IV emulating
1481 sv->sv_u.svu_rv = referent;
1485 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1486 (unsigned long)new_type);
1489 /* if this is zero, this is a body-less SVt_NULL, SVt_IV/SVt_RV,
1490 and sometimes SVt_NV */
1491 if (old_type_details->body_size) {
1495 /* Note that there is an assumption that all bodies of types that
1496 can be upgraded came from arenas. Only the more complex non-
1497 upgradable types are allowed to be directly malloc()ed. */
1498 assert(old_type_details->arena);
1499 del_body((void*)((char*)old_body + old_type_details->offset),
1500 &PL_body_roots[old_type]);
1506 =for apidoc sv_backoff
1508 Remove any string offset. You should normally use the C<SvOOK_off> macro
1514 /* prior to 5.000 stable, this function returned the new OOK-less SvFLAGS
1515 prior to 5.23.4 this function always returned 0
1519 Perl_sv_backoff(SV *const sv)
1522 const char * const s = SvPVX_const(sv);
1524 PERL_ARGS_ASSERT_SV_BACKOFF;
1527 assert(SvTYPE(sv) != SVt_PVHV);
1528 assert(SvTYPE(sv) != SVt_PVAV);
1530 SvOOK_offset(sv, delta);
1532 SvLEN_set(sv, SvLEN(sv) + delta);
1533 SvPV_set(sv, SvPVX(sv) - delta);
1534 SvFLAGS(sv) &= ~SVf_OOK;
1535 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
1540 /* forward declaration */
1541 static void S_sv_uncow(pTHX_ SV * const sv, const U32 flags);
1547 Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1548 upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1549 Use the C<SvGROW> wrapper instead.
1556 Perl_sv_grow(pTHX_ SV *const sv, STRLEN newlen)
1560 PERL_ARGS_ASSERT_SV_GROW;
1564 if (SvTYPE(sv) < SVt_PV) {
1565 sv_upgrade(sv, SVt_PV);
1566 s = SvPVX_mutable(sv);
1568 else if (SvOOK(sv)) { /* pv is offset? */
1570 s = SvPVX_mutable(sv);
1571 if (newlen > SvLEN(sv))
1572 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
1576 if (SvIsCOW(sv)) S_sv_uncow(aTHX_ sv, 0);
1577 s = SvPVX_mutable(sv);
1580 #ifdef PERL_COPY_ON_WRITE
1581 /* the new COW scheme uses SvPVX(sv)[SvLEN(sv)-1] (if spare)
1582 * to store the COW count. So in general, allocate one more byte than
1583 * asked for, to make it likely this byte is always spare: and thus
1584 * make more strings COW-able.
1586 * Only increment if the allocation isn't MEM_SIZE_MAX,
1587 * otherwise it will wrap to 0.
1589 if ( newlen != MEM_SIZE_MAX )
1593 #if defined(PERL_USE_MALLOC_SIZE) && defined(Perl_safesysmalloc_size)
1594 #define PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1597 if (newlen > SvLEN(sv)) { /* need more room? */
1598 STRLEN minlen = SvCUR(sv);
1599 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
1600 if (newlen < minlen)
1602 #ifndef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1604 /* Don't round up on the first allocation, as odds are pretty good that
1605 * the initial request is accurate as to what is really needed */
1607 STRLEN rounded = PERL_STRLEN_ROUNDUP(newlen);
1608 if (rounded > newlen)
1612 if (SvLEN(sv) && s) {
1613 s = (char*)saferealloc(s, newlen);
1616 s = (char*)safemalloc(newlen);
1617 if (SvPVX_const(sv) && SvCUR(sv)) {
1618 Move(SvPVX_const(sv), s, SvCUR(sv), char);
1622 #ifdef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1623 /* Do this here, do it once, do it right, and then we will never get
1624 called back into sv_grow() unless there really is some growing
1626 SvLEN_set(sv, Perl_safesysmalloc_size(s));
1628 SvLEN_set(sv, newlen);
1635 =for apidoc sv_setiv
1637 Copies an integer into the given SV, upgrading first if necessary.
1638 Does not handle 'set' magic. See also C<L</sv_setiv_mg>>.
1644 Perl_sv_setiv(pTHX_ SV *const sv, const IV i)
1646 PERL_ARGS_ASSERT_SV_SETIV;
1648 SV_CHECK_THINKFIRST_COW_DROP(sv);
1649 switch (SvTYPE(sv)) {
1652 sv_upgrade(sv, SVt_IV);
1655 sv_upgrade(sv, SVt_PVIV);
1659 if (!isGV_with_GP(sv))
1667 /* diag_listed_as: Can't coerce %s to %s in %s */
1668 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1670 NOT_REACHED; /* NOTREACHED */
1674 (void)SvIOK_only(sv); /* validate number */
1680 =for apidoc sv_setiv_mg
1682 Like C<sv_setiv>, but also handles 'set' magic.
1688 Perl_sv_setiv_mg(pTHX_ SV *const sv, const IV i)
1690 PERL_ARGS_ASSERT_SV_SETIV_MG;
1697 =for apidoc sv_setuv
1699 Copies an unsigned integer into the given SV, upgrading first if necessary.
1700 Does not handle 'set' magic. See also C<L</sv_setuv_mg>>.
1706 Perl_sv_setuv(pTHX_ SV *const sv, const UV u)
1708 PERL_ARGS_ASSERT_SV_SETUV;
1710 /* With the if statement to ensure that integers are stored as IVs whenever
1712 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
1715 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1717 If you wish to remove the following if statement, so that this routine
1718 (and its callers) always return UVs, please benchmark to see what the
1719 effect is. Modern CPUs may be different. Or may not :-)
1721 if (u <= (UV)IV_MAX) {
1722 sv_setiv(sv, (IV)u);
1731 =for apidoc sv_setuv_mg
1733 Like C<sv_setuv>, but also handles 'set' magic.
1739 Perl_sv_setuv_mg(pTHX_ SV *const sv, const UV u)
1741 PERL_ARGS_ASSERT_SV_SETUV_MG;
1748 =for apidoc sv_setnv
1750 Copies a double into the given SV, upgrading first if necessary.
1751 Does not handle 'set' magic. See also C<L</sv_setnv_mg>>.
1757 Perl_sv_setnv(pTHX_ SV *const sv, const NV num)
1759 PERL_ARGS_ASSERT_SV_SETNV;
1761 SV_CHECK_THINKFIRST_COW_DROP(sv);
1762 switch (SvTYPE(sv)) {
1765 sv_upgrade(sv, SVt_NV);
1769 sv_upgrade(sv, SVt_PVNV);
1773 if (!isGV_with_GP(sv))
1781 /* diag_listed_as: Can't coerce %s to %s in %s */
1782 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
1784 NOT_REACHED; /* NOTREACHED */
1789 (void)SvNOK_only(sv); /* validate number */
1794 =for apidoc sv_setnv_mg
1796 Like C<sv_setnv>, but also handles 'set' magic.
1802 Perl_sv_setnv_mg(pTHX_ SV *const sv, const NV num)
1804 PERL_ARGS_ASSERT_SV_SETNV_MG;
1810 /* Return a cleaned-up, printable version of sv, for non-numeric, or
1811 * not incrementable warning display.
1812 * Originally part of S_not_a_number().
1813 * The return value may be != tmpbuf.
1817 S_sv_display(pTHX_ SV *const sv, char *tmpbuf, STRLEN tmpbuf_size) {
1820 PERL_ARGS_ASSERT_SV_DISPLAY;
1823 SV *dsv = newSVpvs_flags("", SVs_TEMP);
1824 pv = sv_uni_display(dsv, sv, 32, UNI_DISPLAY_ISPRINT);
1827 const char * const limit = tmpbuf + tmpbuf_size - 8;
1828 /* each *s can expand to 4 chars + "...\0",
1829 i.e. need room for 8 chars */
1831 const char *s = SvPVX_const(sv);
1832 const char * const end = s + SvCUR(sv);
1833 for ( ; s < end && d < limit; s++ ) {
1835 if (! isASCII(ch) && !isPRINT_LC(ch)) {
1839 /* Map to ASCII "equivalent" of Latin1 */
1840 ch = LATIN1_TO_NATIVE(NATIVE_TO_LATIN1(ch) & 127);
1846 else if (ch == '\r') {
1850 else if (ch == '\f') {
1854 else if (ch == '\\') {
1858 else if (ch == '\0') {
1862 else if (isPRINT_LC(ch))
1881 /* Print an "isn't numeric" warning, using a cleaned-up,
1882 * printable version of the offending string
1886 S_not_a_number(pTHX_ SV *const sv)
1891 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1893 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1896 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1897 /* diag_listed_as: Argument "%s" isn't numeric%s */
1898 "Argument \"%s\" isn't numeric in %s", pv,
1901 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1902 /* diag_listed_as: Argument "%s" isn't numeric%s */
1903 "Argument \"%s\" isn't numeric", pv);
1907 S_not_incrementable(pTHX_ SV *const sv) {
1911 PERL_ARGS_ASSERT_NOT_INCREMENTABLE;
1913 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1915 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1916 "Argument \"%s\" treated as 0 in increment (++)", pv);
1920 =for apidoc looks_like_number
1922 Test if the content of an SV looks like a number (or is a number).
1923 C<Inf> and C<Infinity> are treated as numbers (so will not issue a
1924 non-numeric warning), even if your C<atof()> doesn't grok them. Get-magic is
1931 Perl_looks_like_number(pTHX_ SV *const sv)
1937 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1939 if (SvPOK(sv) || SvPOKp(sv)) {
1940 sbegin = SvPV_nomg_const(sv, len);
1943 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
1944 numtype = grok_number(sbegin, len, NULL);
1945 return ((numtype & IS_NUMBER_TRAILING)) ? 0 : numtype;
1949 S_glob_2number(pTHX_ GV * const gv)
1951 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1953 /* We know that all GVs stringify to something that is not-a-number,
1954 so no need to test that. */
1955 if (ckWARN(WARN_NUMERIC))
1957 SV *const buffer = sv_newmortal();
1958 gv_efullname3(buffer, gv, "*");
1959 not_a_number(buffer);
1961 /* We just want something true to return, so that S_sv_2iuv_common
1962 can tail call us and return true. */
1966 /* Actually, ISO C leaves conversion of UV to IV undefined, but
1967 until proven guilty, assume that things are not that bad... */
1972 As 64 bit platforms often have an NV that doesn't preserve all bits of
1973 an IV (an assumption perl has been based on to date) it becomes necessary
1974 to remove the assumption that the NV always carries enough precision to
1975 recreate the IV whenever needed, and that the NV is the canonical form.
1976 Instead, IV/UV and NV need to be given equal rights. So as to not lose
1977 precision as a side effect of conversion (which would lead to insanity
1978 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
1979 1) to distinguish between IV/UV/NV slots that have a valid conversion cached
1980 where precision was lost, and IV/UV/NV slots that have a valid conversion
1981 which has lost no precision
1982 2) to ensure that if a numeric conversion to one form is requested that
1983 would lose precision, the precise conversion (or differently
1984 imprecise conversion) is also performed and cached, to prevent
1985 requests for different numeric formats on the same SV causing
1986 lossy conversion chains. (lossless conversion chains are perfectly
1991 SvIOKp is true if the IV slot contains a valid value
1992 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1993 SvNOKp is true if the NV slot contains a valid value
1994 SvNOK is true only if the NV value is accurate
1997 while converting from PV to NV, check to see if converting that NV to an
1998 IV(or UV) would lose accuracy over a direct conversion from PV to
1999 IV(or UV). If it would, cache both conversions, return NV, but mark
2000 SV as IOK NOKp (ie not NOK).
2002 While converting from PV to IV, check to see if converting that IV to an
2003 NV would lose accuracy over a direct conversion from PV to NV. If it
2004 would, cache both conversions, flag similarly.
2006 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
2007 correctly because if IV & NV were set NV *always* overruled.
2008 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
2009 changes - now IV and NV together means that the two are interchangeable:
2010 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
2012 The benefit of this is that operations such as pp_add know that if
2013 SvIOK is true for both left and right operands, then integer addition
2014 can be used instead of floating point (for cases where the result won't
2015 overflow). Before, floating point was always used, which could lead to
2016 loss of precision compared with integer addition.
2018 * making IV and NV equal status should make maths accurate on 64 bit
2020 * may speed up maths somewhat if pp_add and friends start to use
2021 integers when possible instead of fp. (Hopefully the overhead in
2022 looking for SvIOK and checking for overflow will not outweigh the
2023 fp to integer speedup)
2024 * will slow down integer operations (callers of SvIV) on "inaccurate"
2025 values, as the change from SvIOK to SvIOKp will cause a call into
2026 sv_2iv each time rather than a macro access direct to the IV slot
2027 * should speed up number->string conversion on integers as IV is
2028 favoured when IV and NV are equally accurate
2030 ####################################################################
2031 You had better be using SvIOK_notUV if you want an IV for arithmetic:
2032 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
2033 On the other hand, SvUOK is true iff UV.
2034 ####################################################################
2036 Your mileage will vary depending your CPU's relative fp to integer
2040 #ifndef NV_PRESERVES_UV
2041 # define IS_NUMBER_UNDERFLOW_IV 1
2042 # define IS_NUMBER_UNDERFLOW_UV 2
2043 # define IS_NUMBER_IV_AND_UV 2
2044 # define IS_NUMBER_OVERFLOW_IV 4
2045 # define IS_NUMBER_OVERFLOW_UV 5
2047 /* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
2049 /* For sv_2nv these three cases are "SvNOK and don't bother casting" */
2051 S_sv_2iuv_non_preserve(pTHX_ SV *const sv
2057 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
2058 PERL_UNUSED_CONTEXT;
2060 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_2iuv_non '%s', IV=0x%" UVxf " NV=%" NVgf " inttype=%" UVXf "\n", SvPVX_const(sv), SvIVX(sv), SvNVX(sv), (UV)numtype));
2061 if (SvNVX(sv) < (NV)IV_MIN) {
2062 (void)SvIOKp_on(sv);
2064 SvIV_set(sv, IV_MIN);
2065 return IS_NUMBER_UNDERFLOW_IV;
2067 if (SvNVX(sv) > (NV)UV_MAX) {
2068 (void)SvIOKp_on(sv);
2071 SvUV_set(sv, UV_MAX);
2072 return IS_NUMBER_OVERFLOW_UV;
2074 (void)SvIOKp_on(sv);
2076 /* Can't use strtol etc to convert this string. (See truth table in
2078 if (SvNVX(sv) <= (UV)IV_MAX) {
2079 SvIV_set(sv, I_V(SvNVX(sv)));
2080 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2081 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
2083 /* Integer is imprecise. NOK, IOKp */
2085 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
2088 SvUV_set(sv, U_V(SvNVX(sv)));
2089 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2090 if (SvUVX(sv) == UV_MAX) {
2091 /* As we know that NVs don't preserve UVs, UV_MAX cannot
2092 possibly be preserved by NV. Hence, it must be overflow.
2094 return IS_NUMBER_OVERFLOW_UV;
2096 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
2098 /* Integer is imprecise. NOK, IOKp */
2100 return IS_NUMBER_OVERFLOW_IV;
2102 #endif /* !NV_PRESERVES_UV*/
2104 /* If numtype is infnan, set the NV of the sv accordingly.
2105 * If numtype is anything else, try setting the NV using Atof(PV). */
2107 S_sv_setnv(pTHX_ SV* sv, int numtype)
2109 bool pok = cBOOL(SvPOK(sv));
2112 if ((numtype & IS_NUMBER_INFINITY)) {
2113 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -NV_INF : NV_INF);
2118 if ((numtype & IS_NUMBER_NAN)) {
2119 SvNV_set(sv, NV_NAN);
2124 SvNV_set(sv, Atof(SvPVX_const(sv)));
2125 /* Purposefully no true nok here, since we don't want to blow
2126 * away the possible IOK/UV of an existing sv. */
2129 SvNOK_only(sv); /* No IV or UV please, this is pure infnan. */
2131 SvPOK_on(sv); /* PV is okay, though. */
2136 S_sv_2iuv_common(pTHX_ SV *const sv)
2138 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
2141 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
2142 * without also getting a cached IV/UV from it at the same time
2143 * (ie PV->NV conversion should detect loss of accuracy and cache
2144 * IV or UV at same time to avoid this. */
2145 /* IV-over-UV optimisation - choose to cache IV if possible */
2147 if (SvTYPE(sv) == SVt_NV)
2148 sv_upgrade(sv, SVt_PVNV);
2150 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
2151 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
2152 certainly cast into the IV range at IV_MAX, whereas the correct
2153 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
2155 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2156 if (Perl_isnan(SvNVX(sv))) {
2162 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2163 SvIV_set(sv, I_V(SvNVX(sv)));
2164 if (SvNVX(sv) == (NV) SvIVX(sv)
2165 #ifndef NV_PRESERVES_UV
2166 && SvIVX(sv) != IV_MIN /* avoid negating IV_MIN below */
2167 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2168 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2169 /* Don't flag it as "accurately an integer" if the number
2170 came from a (by definition imprecise) NV operation, and
2171 we're outside the range of NV integer precision */
2175 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2177 /* scalar has trailing garbage, eg "42a" */
2179 DEBUG_c(PerlIO_printf(Perl_debug_log,
2180 "0x%" UVxf " iv(%" NVgf " => %" IVdf ") (precise)\n",
2186 /* IV not precise. No need to convert from PV, as NV
2187 conversion would already have cached IV if it detected
2188 that PV->IV would be better than PV->NV->IV
2189 flags already correct - don't set public IOK. */
2190 DEBUG_c(PerlIO_printf(Perl_debug_log,
2191 "0x%" UVxf " iv(%" NVgf " => %" IVdf ") (imprecise)\n",
2196 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2197 but the cast (NV)IV_MIN rounds to a the value less (more
2198 negative) than IV_MIN which happens to be equal to SvNVX ??
2199 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2200 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2201 (NV)UVX == NVX are both true, but the values differ. :-(
2202 Hopefully for 2s complement IV_MIN is something like
2203 0x8000000000000000 which will be exact. NWC */
2206 SvUV_set(sv, U_V(SvNVX(sv)));
2208 (SvNVX(sv) == (NV) SvUVX(sv))
2209 #ifndef NV_PRESERVES_UV
2210 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2211 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2212 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2213 /* Don't flag it as "accurately an integer" if the number
2214 came from a (by definition imprecise) NV operation, and
2215 we're outside the range of NV integer precision */
2221 DEBUG_c(PerlIO_printf(Perl_debug_log,
2222 "0x%" UVxf " 2iv(%" UVuf " => %" IVdf ") (as unsigned)\n",
2228 else if (SvPOKp(sv)) {
2231 const char *s = SvPVX_const(sv);
2232 const STRLEN cur = SvCUR(sv);
2234 /* short-cut for a single digit string like "1" */
2239 if (SvTYPE(sv) < SVt_PVIV)
2240 sv_upgrade(sv, SVt_PVIV);
2242 SvIV_set(sv, (IV)(c - '0'));
2247 numtype = grok_number(s, cur, &value);
2248 /* We want to avoid a possible problem when we cache an IV/ a UV which
2249 may be later translated to an NV, and the resulting NV is not
2250 the same as the direct translation of the initial string
2251 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2252 be careful to ensure that the value with the .456 is around if the
2253 NV value is requested in the future).
2255 This means that if we cache such an IV/a UV, we need to cache the
2256 NV as well. Moreover, we trade speed for space, and do not
2257 cache the NV if we are sure it's not needed.
2260 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2261 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2262 == IS_NUMBER_IN_UV) {
2263 /* It's definitely an integer, only upgrade to PVIV */
2264 if (SvTYPE(sv) < SVt_PVIV)
2265 sv_upgrade(sv, SVt_PVIV);
2267 } else if (SvTYPE(sv) < SVt_PVNV)
2268 sv_upgrade(sv, SVt_PVNV);
2270 if ((numtype & (IS_NUMBER_INFINITY | IS_NUMBER_NAN))) {
2271 if (ckWARN(WARN_NUMERIC) && ((numtype & IS_NUMBER_TRAILING)))
2273 S_sv_setnv(aTHX_ sv, numtype);
2277 /* If NVs preserve UVs then we only use the UV value if we know that
2278 we aren't going to call atof() below. If NVs don't preserve UVs
2279 then the value returned may have more precision than atof() will
2280 return, even though value isn't perfectly accurate. */
2281 if ((numtype & (IS_NUMBER_IN_UV
2282 #ifdef NV_PRESERVES_UV
2285 )) == IS_NUMBER_IN_UV) {
2286 /* This won't turn off the public IOK flag if it was set above */
2287 (void)SvIOKp_on(sv);
2289 if (!(numtype & IS_NUMBER_NEG)) {
2291 if (value <= (UV)IV_MAX) {
2292 SvIV_set(sv, (IV)value);
2294 /* it didn't overflow, and it was positive. */
2295 SvUV_set(sv, value);
2299 /* 2s complement assumption */
2300 if (value <= (UV)IV_MIN) {
2301 SvIV_set(sv, value == (UV)IV_MIN
2302 ? IV_MIN : -(IV)value);
2304 /* Too negative for an IV. This is a double upgrade, but
2305 I'm assuming it will be rare. */
2306 if (SvTYPE(sv) < SVt_PVNV)
2307 sv_upgrade(sv, SVt_PVNV);
2311 SvNV_set(sv, -(NV)value);
2312 SvIV_set(sv, IV_MIN);
2316 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2317 will be in the previous block to set the IV slot, and the next
2318 block to set the NV slot. So no else here. */
2320 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2321 != IS_NUMBER_IN_UV) {
2322 /* It wasn't an (integer that doesn't overflow the UV). */
2323 S_sv_setnv(aTHX_ sv, numtype);
2325 if (! numtype && ckWARN(WARN_NUMERIC))
2328 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2iv(%" NVgf ")\n",
2329 PTR2UV(sv), SvNVX(sv)));
2331 #ifdef NV_PRESERVES_UV
2332 (void)SvIOKp_on(sv);
2334 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2335 if (Perl_isnan(SvNVX(sv))) {
2341 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2342 SvIV_set(sv, I_V(SvNVX(sv)));
2343 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2346 NOOP; /* Integer is imprecise. NOK, IOKp */
2348 /* UV will not work better than IV */
2350 if (SvNVX(sv) > (NV)UV_MAX) {
2352 /* Integer is inaccurate. NOK, IOKp, is UV */
2353 SvUV_set(sv, UV_MAX);
2355 SvUV_set(sv, U_V(SvNVX(sv)));
2356 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2357 NV preservse UV so can do correct comparison. */
2358 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2361 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
2366 #else /* NV_PRESERVES_UV */
2367 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2368 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
2369 /* The IV/UV slot will have been set from value returned by
2370 grok_number above. The NV slot has just been set using
2373 assert (SvIOKp(sv));
2375 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2376 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2377 /* Small enough to preserve all bits. */
2378 (void)SvIOKp_on(sv);
2380 SvIV_set(sv, I_V(SvNVX(sv)));
2381 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2383 /* Assumption: first non-preserved integer is < IV_MAX,
2384 this NV is in the preserved range, therefore: */
2385 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2387 Perl_croak(aTHX_ "sv_2iv assumed (U_V(fabs((double)SvNVX(sv))) < (UV)IV_MAX) but SvNVX(sv)=%" NVgf " U_V is 0x%" UVxf ", IV_MAX is 0x%" UVxf "\n", SvNVX(sv), U_V(SvNVX(sv)), (UV)IV_MAX);
2391 0 0 already failed to read UV.
2392 0 1 already failed to read UV.
2393 1 0 you won't get here in this case. IV/UV
2394 slot set, public IOK, Atof() unneeded.
2395 1 1 already read UV.
2396 so there's no point in sv_2iuv_non_preserve() attempting
2397 to use atol, strtol, strtoul etc. */
2399 sv_2iuv_non_preserve (sv, numtype);
2401 sv_2iuv_non_preserve (sv);
2405 #endif /* NV_PRESERVES_UV */
2406 /* It might be more code efficient to go through the entire logic above
2407 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2408 gets complex and potentially buggy, so more programmer efficient
2409 to do it this way, by turning off the public flags: */
2411 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2415 if (isGV_with_GP(sv))
2416 return glob_2number(MUTABLE_GV(sv));
2418 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2420 if (SvTYPE(sv) < SVt_IV)
2421 /* Typically the caller expects that sv_any is not NULL now. */
2422 sv_upgrade(sv, SVt_IV);
2423 /* Return 0 from the caller. */
2430 =for apidoc sv_2iv_flags
2432 Return the integer value of an SV, doing any necessary string
2433 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2434 Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2440 Perl_sv_2iv_flags(pTHX_ SV *const sv, const I32 flags)
2442 PERL_ARGS_ASSERT_SV_2IV_FLAGS;
2444 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2445 && SvTYPE(sv) != SVt_PVFM);
2447 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2453 if (flags & SV_SKIP_OVERLOAD)
2455 tmpstr = AMG_CALLunary(sv, numer_amg);
2456 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2457 return SvIV(tmpstr);
2460 return PTR2IV(SvRV(sv));
2463 if (SvVALID(sv) || isREGEXP(sv)) {
2464 /* FBMs use the space for SvIVX and SvNVX for other purposes, so
2465 must not let them cache IVs.
2466 In practice they are extremely unlikely to actually get anywhere
2467 accessible by user Perl code - the only way that I'm aware of is when
2468 a constant subroutine which is used as the second argument to index.
2470 Regexps have no SvIVX and SvNVX fields.
2475 const char * const ptr =
2476 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2478 = grok_number(ptr, SvCUR(sv), &value);
2480 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2481 == IS_NUMBER_IN_UV) {
2482 /* It's definitely an integer */
2483 if (numtype & IS_NUMBER_NEG) {
2484 if (value < (UV)IV_MIN)
2487 if (value < (UV)IV_MAX)
2492 /* Quite wrong but no good choices. */
2493 if ((numtype & IS_NUMBER_INFINITY)) {
2494 return (numtype & IS_NUMBER_NEG) ? IV_MIN : IV_MAX;
2495 } else if ((numtype & IS_NUMBER_NAN)) {
2496 return 0; /* So wrong. */
2500 if (ckWARN(WARN_NUMERIC))
2503 return I_V(Atof(ptr));
2507 if (SvTHINKFIRST(sv)) {
2508 if (SvREADONLY(sv) && !SvOK(sv)) {
2509 if (ckWARN(WARN_UNINITIALIZED))
2516 if (S_sv_2iuv_common(aTHX_ sv))
2520 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2iv(%" IVdf ")\n",
2521 PTR2UV(sv),SvIVX(sv)));
2522 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
2526 =for apidoc sv_2uv_flags
2528 Return the unsigned integer value of an SV, doing any necessary string
2529 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2530 Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
2532 =for apidoc Amnh||SV_GMAGIC
2538 Perl_sv_2uv_flags(pTHX_ SV *const sv, const I32 flags)
2540 PERL_ARGS_ASSERT_SV_2UV_FLAGS;
2542 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2548 if (flags & SV_SKIP_OVERLOAD)
2550 tmpstr = AMG_CALLunary(sv, numer_amg);
2551 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2552 return SvUV(tmpstr);
2555 return PTR2UV(SvRV(sv));
2558 if (SvVALID(sv) || isREGEXP(sv)) {
2559 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2560 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2561 Regexps have no SvIVX and SvNVX fields. */
2565 const char * const ptr =
2566 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2568 = grok_number(ptr, SvCUR(sv), &value);
2570 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2571 == IS_NUMBER_IN_UV) {
2572 /* It's definitely an integer */
2573 if (!(numtype & IS_NUMBER_NEG))
2577 /* Quite wrong but no good choices. */
2578 if ((numtype & IS_NUMBER_INFINITY)) {
2579 return UV_MAX; /* So wrong. */
2580 } else if ((numtype & IS_NUMBER_NAN)) {
2581 return 0; /* So wrong. */
2585 if (ckWARN(WARN_NUMERIC))
2588 return U_V(Atof(ptr));
2592 if (SvTHINKFIRST(sv)) {
2593 if (SvREADONLY(sv) && !SvOK(sv)) {
2594 if (ckWARN(WARN_UNINITIALIZED))
2601 if (S_sv_2iuv_common(aTHX_ sv))
2605 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2uv(%" UVuf ")\n",
2606 PTR2UV(sv),SvUVX(sv)));
2607 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
2611 =for apidoc sv_2nv_flags
2613 Return the num value of an SV, doing any necessary string or integer
2614 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2615 Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
2621 Perl_sv_2nv_flags(pTHX_ SV *const sv, const I32 flags)
2623 PERL_ARGS_ASSERT_SV_2NV_FLAGS;
2625 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2626 && SvTYPE(sv) != SVt_PVFM);
2627 if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) {
2628 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2629 the same flag bit as SVf_IVisUV, so must not let them cache NVs.
2630 Regexps have no SvIVX and SvNVX fields. */
2632 if (flags & SV_GMAGIC)
2636 if (SvPOKp(sv) && !SvIOKp(sv)) {
2637 ptr = SvPVX_const(sv);
2638 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
2639 !grok_number(ptr, SvCUR(sv), NULL))
2645 return (NV)SvUVX(sv);
2647 return (NV)SvIVX(sv);
2652 assert(SvTYPE(sv) >= SVt_PVMG);
2653 /* This falls through to the report_uninit near the end of the
2655 } else if (SvTHINKFIRST(sv)) {
2660 if (flags & SV_SKIP_OVERLOAD)
2662 tmpstr = AMG_CALLunary(sv, numer_amg);
2663 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2664 return SvNV(tmpstr);
2667 return PTR2NV(SvRV(sv));
2669 if (SvREADONLY(sv) && !SvOK(sv)) {
2670 if (ckWARN(WARN_UNINITIALIZED))
2675 if (SvTYPE(sv) < SVt_NV) {
2676 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2677 sv_upgrade(sv, SVt_NV);
2678 CLANG_DIAG_IGNORE_STMT(-Wthread-safety);
2680 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2681 STORE_LC_NUMERIC_SET_STANDARD();
2682 PerlIO_printf(Perl_debug_log,
2683 "0x%" UVxf " num(%" NVgf ")\n",
2684 PTR2UV(sv), SvNVX(sv));
2685 RESTORE_LC_NUMERIC();
2687 CLANG_DIAG_RESTORE_STMT;
2690 else if (SvTYPE(sv) < SVt_PVNV)
2691 sv_upgrade(sv, SVt_PVNV);
2696 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
2697 #ifdef NV_PRESERVES_UV
2703 /* Only set the public NV OK flag if this NV preserves the IV */
2704 /* Check it's not 0xFFFFFFFFFFFFFFFF */
2706 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
2707 : (SvIVX(sv) == I_V(SvNVX(sv))))
2713 else if (SvPOKp(sv)) {
2715 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2716 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
2718 #ifdef NV_PRESERVES_UV
2719 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2720 == IS_NUMBER_IN_UV) {
2721 /* It's definitely an integer */
2722 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
2724 S_sv_setnv(aTHX_ sv, numtype);
2731 SvNV_set(sv, Atof(SvPVX_const(sv)));
2732 /* Only set the public NV OK flag if this NV preserves the value in
2733 the PV at least as well as an IV/UV would.
2734 Not sure how to do this 100% reliably. */
2735 /* if that shift count is out of range then Configure's test is
2736 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2738 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2739 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2740 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
2741 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2742 /* Can't use strtol etc to convert this string, so don't try.
2743 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2746 /* value has been set. It may not be precise. */
2747 if ((numtype & IS_NUMBER_NEG) && (value >= (UV)IV_MIN)) {
2748 /* 2s complement assumption for (UV)IV_MIN */
2749 SvNOK_on(sv); /* Integer is too negative. */
2754 if (numtype & IS_NUMBER_NEG) {
2755 /* -IV_MIN is undefined, but we should never reach
2756 * this point with both IS_NUMBER_NEG and value ==
2758 assert(value != (UV)IV_MIN);
2759 SvIV_set(sv, -(IV)value);
2760 } else if (value <= (UV)IV_MAX) {
2761 SvIV_set(sv, (IV)value);
2763 SvUV_set(sv, value);
2767 if (numtype & IS_NUMBER_NOT_INT) {
2768 /* I believe that even if the original PV had decimals,
2769 they are lost beyond the limit of the FP precision.
2770 However, neither is canonical, so both only get p
2771 flags. NWC, 2000/11/25 */
2772 /* Both already have p flags, so do nothing */
2774 const NV nv = SvNVX(sv);
2775 /* XXX should this spot have NAN_COMPARE_BROKEN, too? */
2776 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2777 if (SvIVX(sv) == I_V(nv)) {
2780 /* It had no "." so it must be integer. */
2784 /* between IV_MAX and NV(UV_MAX).
2785 Could be slightly > UV_MAX */
2787 if (numtype & IS_NUMBER_NOT_INT) {
2788 /* UV and NV both imprecise. */
2790 const UV nv_as_uv = U_V(nv);
2792 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2801 /* It might be more code efficient to go through the entire logic above
2802 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2803 gets complex and potentially buggy, so more programmer efficient
2804 to do it this way, by turning off the public flags: */
2806 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2807 #endif /* NV_PRESERVES_UV */
2810 if (isGV_with_GP(sv)) {
2811 glob_2number(MUTABLE_GV(sv));
2815 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2817 assert (SvTYPE(sv) >= SVt_NV);
2818 /* Typically the caller expects that sv_any is not NULL now. */
2819 /* XXX Ilya implies that this is a bug in callers that assume this
2820 and ideally should be fixed. */
2823 CLANG_DIAG_IGNORE_STMT(-Wthread-safety);
2825 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2826 STORE_LC_NUMERIC_SET_STANDARD();
2827 PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2nv(%" NVgf ")\n",
2828 PTR2UV(sv), SvNVX(sv));
2829 RESTORE_LC_NUMERIC();
2831 CLANG_DIAG_RESTORE_STMT;
2838 Return an SV with the numeric value of the source SV, doing any necessary
2839 reference or overload conversion. The caller is expected to have handled
2846 Perl_sv_2num(pTHX_ SV *const sv)
2848 PERL_ARGS_ASSERT_SV_2NUM;
2853 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
2854 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
2855 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2856 return sv_2num(tmpsv);
2858 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2861 /* int2str_table: lookup table containing string representations of all
2862 * two digit numbers. For example, int2str_table.arr[0] is "00" and
2863 * int2str_table.arr[12*2] is "12".
2865 * We are going to read two bytes at a time, so we have to ensure that
2866 * the array is aligned to a 2 byte boundary. That's why it was made a
2867 * union with a dummy U16 member. */
2868 static const union {
2871 } int2str_table = {{
2872 '0', '0', '0', '1', '0', '2', '0', '3', '0', '4', '0', '5', '0', '6',
2873 '0', '7', '0', '8', '0', '9', '1', '0', '1', '1', '1', '2', '1', '3',
2874 '1', '4', '1', '5', '1', '6', '1', '7', '1', '8', '1', '9', '2', '0',
2875 '2', '1', '2', '2', '2', '3', '2', '4', '2', '5', '2', '6', '2', '7',
2876 '2', '8', '2', '9', '3', '0', '3', '1', '3', '2', '3', '3', '3', '4',
2877 '3', '5', '3', '6', '3', '7', '3', '8', '3', '9', '4', '0', '4', '1',
2878 '4', '2', '4', '3', '4', '4', '4', '5', '4', '6', '4', '7', '4', '8',
2879 '4', '9', '5', '0', '5', '1', '5', '2', '5', '3', '5', '4', '5', '5',
2880 '5', '6', '5', '7', '5', '8', '5', '9', '6', '0', '6', '1', '6', '2',
2881 '6', '3', '6', '4', '6', '5', '6', '6', '6', '7', '6', '8', '6', '9',
2882 '7', '0', '7', '1', '7', '2', '7', '3', '7', '4', '7', '5', '7', '6',
2883 '7', '7', '7', '8', '7', '9', '8', '0', '8', '1', '8', '2', '8', '3',
2884 '8', '4', '8', '5', '8', '6', '8', '7', '8', '8', '8', '9', '9', '0',
2885 '9', '1', '9', '2', '9', '3', '9', '4', '9', '5', '9', '6', '9', '7',
2889 /* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2890 * UV as a string towards the end of buf, and return pointers to start and
2893 * We assume that buf is at least TYPE_CHARS(UV) long.
2896 PERL_STATIC_INLINE char *
2897 S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
2899 char *ptr = buf + TYPE_CHARS(UV);
2900 char * const ebuf = ptr;
2902 U16 *word_ptr, *word_table;
2904 PERL_ARGS_ASSERT_UIV_2BUF;
2906 /* ptr has to be properly aligned, because we will cast it to U16* */
2907 assert(PTR2nat(ptr) % 2 == 0);
2908 /* we are going to read/write two bytes at a time */
2909 word_ptr = (U16*)ptr;
2910 word_table = (U16*)int2str_table.arr;
2912 if (UNLIKELY(is_uv))
2918 /* Using 0- here to silence bogus warning from MS VC */
2919 uv = (UV) (0 - (UV) iv);
2924 *--word_ptr = word_table[uv % 100];
2927 ptr = (char*)word_ptr;
2930 *--ptr = (char)uv + '0';
2932 *--word_ptr = word_table[uv];
2933 ptr = (char*)word_ptr;
2943 /* Helper for sv_2pv_flags and sv_vcatpvfn_flags. If the NV is an
2944 * infinity or a not-a-number, writes the appropriate strings to the
2945 * buffer, including a zero byte. On success returns the written length,
2946 * excluding the zero byte, on failure (not an infinity, not a nan)
2947 * returns zero, assert-fails on maxlen being too short.
2949 * XXX for "Inf", "-Inf", and "NaN", we could have three read-only
2950 * shared string constants we point to, instead of generating a new
2951 * string for each instance. */
2953 S_infnan_2pv(NV nv, char* buffer, size_t maxlen, char plus) {
2955 assert(maxlen >= 4);
2956 if (Perl_isinf(nv)) {
2958 if (maxlen < 5) /* "-Inf\0" */
2968 else if (Perl_isnan(nv)) {
2972 /* XXX optionally output the payload mantissa bits as
2973 * "(unsigned)" (to match the nan("...") C99 function,
2974 * or maybe as "(0xhhh...)" would make more sense...
2975 * provide a format string so that the user can decide?
2976 * NOTE: would affect the maxlen and assert() logic.*/
2981 assert((s == buffer + 3) || (s == buffer + 4));
2987 =for apidoc sv_2pv_flags
2989 Returns a pointer to the string value of an SV, and sets C<*lp> to its length.
2990 If flags has the C<SV_GMAGIC> bit set, does an C<mg_get()> first. Coerces C<sv> to a
2991 string if necessary. Normally invoked via the C<SvPV_flags> macro.
2992 C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
2998 Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
3002 PERL_ARGS_ASSERT_SV_2PV_FLAGS;
3004 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
3005 && SvTYPE(sv) != SVt_PVFM);
3006 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
3011 if (flags & SV_SKIP_OVERLOAD)
3013 tmpstr = AMG_CALLunary(sv, string_amg);
3014 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
3015 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
3017 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
3021 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
3022 if (flags & SV_CONST_RETURN) {
3023 pv = (char *) SvPVX_const(tmpstr);
3025 pv = (flags & SV_MUTABLE_RETURN)
3026 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
3029 *lp = SvCUR(tmpstr);
3031 pv = sv_2pv_flags(tmpstr, lp, flags);
3044 SV *const referent = SvRV(sv);
3048 retval = buffer = savepvn("NULLREF", len);
3049 } else if (SvTYPE(referent) == SVt_REGEXP &&
3050 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
3051 amagic_is_enabled(string_amg))) {
3052 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
3056 /* If the regex is UTF-8 we want the containing scalar to
3057 have an UTF-8 flag too */
3064 *lp = RX_WRAPLEN(re);
3066 return RX_WRAPPED(re);
3068 const char *const typestr = sv_reftype(referent, 0);
3069 const STRLEN typelen = strlen(typestr);
3070 UV addr = PTR2UV(referent);
3071 const char *stashname = NULL;
3072 STRLEN stashnamelen = 0; /* hush, gcc */
3073 const char *buffer_end;
3075 if (SvOBJECT(referent)) {
3076 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
3079 stashname = HEK_KEY(name);
3080 stashnamelen = HEK_LEN(name);
3082 if (HEK_UTF8(name)) {
3088 stashname = "__ANON__";
3091 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
3092 + 2 * sizeof(UV) + 2 /* )\0 */;
3094 len = typelen + 3 /* (0x */
3095 + 2 * sizeof(UV) + 2 /* )\0 */;
3098 Newx(buffer, len, char);
3099 buffer_end = retval = buffer + len;
3101 /* Working backwards */
3105 *--retval = PL_hexdigit[addr & 15];
3106 } while (addr >>= 4);
3112 memcpy(retval, typestr, typelen);
3116 retval -= stashnamelen;
3117 memcpy(retval, stashname, stashnamelen);
3119 /* retval may not necessarily have reached the start of the
3121 assert (retval >= buffer);
3123 len = buffer_end - retval - 1; /* -1 for that \0 */
3135 if (flags & SV_MUTABLE_RETURN)
3136 return SvPVX_mutable(sv);
3137 if (flags & SV_CONST_RETURN)
3138 return (char *)SvPVX_const(sv);
3143 /* I'm assuming that if both IV and NV are equally valid then
3144 converting the IV is going to be more efficient */
3145 const U32 isUIOK = SvIsUV(sv);
3146 /* The purpose of this union is to ensure that arr is aligned on
3147 a 2 byte boundary, because that is what uiv_2buf() requires */
3149 char arr[TYPE_CHARS(UV)];
3155 if (SvTYPE(sv) < SVt_PVIV)
3156 sv_upgrade(sv, SVt_PVIV);
3157 ptr = uiv_2buf(buf.arr, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
3159 /* inlined from sv_setpvn */
3160 s = SvGROW_mutable(sv, len + 1);
3161 Move(ptr, s, len, char);
3166 else if (SvNOK(sv)) {
3167 if (SvTYPE(sv) < SVt_PVNV)
3168 sv_upgrade(sv, SVt_PVNV);
3169 if (SvNVX(sv) == 0.0
3170 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
3171 && !Perl_isnan(SvNVX(sv))
3174 s = SvGROW_mutable(sv, 2);
3179 STRLEN size = 5; /* "-Inf\0" */
3181 s = SvGROW_mutable(sv, size);
3182 len = S_infnan_2pv(SvNVX(sv), s, size, 0);
3188 /* some Xenix systems wipe out errno here */
3197 5 + /* exponent digits */
3201 s = SvGROW_mutable(sv, size);
3202 #ifndef USE_LOCALE_NUMERIC
3203 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3209 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
3210 STORE_LC_NUMERIC_SET_TO_NEEDED();
3212 local_radix = _NOT_IN_NUMERIC_STANDARD;
3213 if (local_radix && SvCUR(PL_numeric_radix_sv) > 1) {
3214 size += SvCUR(PL_numeric_radix_sv) - 1;
3215 s = SvGROW_mutable(sv, size);
3218 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3220 /* If the radix character is UTF-8, and actually is in the
3221 * output, turn on the UTF-8 flag for the scalar */
3223 && SvUTF8(PL_numeric_radix_sv)
3224 && instr(s, SvPVX_const(PL_numeric_radix_sv)))
3229 RESTORE_LC_NUMERIC();
3232 /* We don't call SvPOK_on(), because it may come to
3233 * pass that the locale changes so that the
3234 * stringification we just did is no longer correct. We
3235 * will have to re-stringify every time it is needed */
3242 else if (isGV_with_GP(sv)) {
3243 GV *const gv = MUTABLE_GV(sv);
3244 SV *const buffer = sv_newmortal();
3246 gv_efullname3(buffer, gv, "*");
3248 assert(SvPOK(buffer));
3254 *lp = SvCUR(buffer);
3255 return SvPVX(buffer);
3260 if (flags & SV_UNDEF_RETURNS_NULL)
3262 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
3264 /* Typically the caller expects that sv_any is not NULL now. */
3265 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
3266 sv_upgrade(sv, SVt_PV);
3271 const STRLEN len = s - SvPVX_const(sv);
3276 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2pv(%s)\n",
3277 PTR2UV(sv),SvPVX_const(sv)));
3278 if (flags & SV_CONST_RETURN)
3279 return (char *)SvPVX_const(sv);
3280 if (flags & SV_MUTABLE_RETURN)
3281 return SvPVX_mutable(sv);
3286 =for apidoc sv_copypv
3288 Copies a stringified representation of the source SV into the
3289 destination SV. Automatically performs any necessary C<mg_get> and
3290 coercion of numeric values into strings. Guaranteed to preserve
3291 C<UTF8> flag even from overloaded objects. Similar in nature to
3292 C<sv_2pv[_flags]> but operates directly on an SV instead of just the
3293 string. Mostly uses C<sv_2pv_flags> to do its work, except when that
3294 would lose the UTF-8'ness of the PV.
3296 =for apidoc sv_copypv_nomg
3298 Like C<sv_copypv>, but doesn't invoke get magic first.
3300 =for apidoc sv_copypv_flags
3302 Implementation of C<sv_copypv> and C<sv_copypv_nomg>. Calls get magic iff flags
3303 has the C<SV_GMAGIC> bit set.
3309 Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
3314 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
3316 s = SvPV_flags_const(ssv,len,(flags & SV_GMAGIC));
3317 sv_setpvn(dsv,s,len);
3325 =for apidoc sv_2pvbyte
3327 Return a pointer to the byte-encoded representation of the SV, and set C<*lp>
3328 to its length. If the SV is marked as being encoded as UTF-8, it will
3329 downgrade it to a byte string as a side-effect, if possible. If the SV cannot
3330 be downgraded, this croaks.
3332 Usually accessed via the C<SvPVbyte> macro.
3338 Perl_sv_2pvbyte_flags(pTHX_ SV *sv, STRLEN *const lp, const U32 flags)
3340 PERL_ARGS_ASSERT_SV_2PVBYTE_FLAGS;
3342 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
3344 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3345 || isGV_with_GP(sv) || SvROK(sv)) {
3346 SV *sv2 = sv_newmortal();
3347 sv_copypv_nomg(sv2,sv);
3350 sv_utf8_downgrade_nomg(sv,0);
3351 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3355 =for apidoc sv_2pvutf8
3357 Return a pointer to the UTF-8-encoded representation of the SV, and set C<*lp>
3358 to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3360 Usually accessed via the C<SvPVutf8> macro.
3366 Perl_sv_2pvutf8_flags(pTHX_ SV *sv, STRLEN *const lp, const U32 flags)
3368 PERL_ARGS_ASSERT_SV_2PVUTF8_FLAGS;
3370 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
3372 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3373 || isGV_with_GP(sv) || SvROK(sv)) {
3374 SV *sv2 = sv_newmortal();
3375 sv_copypv_nomg(sv2,sv);
3378 sv_utf8_upgrade_nomg(sv);
3379 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3384 =for apidoc sv_2bool
3386 This macro is only used by C<sv_true()> or its macro equivalent, and only if
3387 the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>.
3388 It calls C<sv_2bool_flags> with the C<SV_GMAGIC> flag.
3390 =for apidoc sv_2bool_flags
3392 This function is only used by C<sv_true()> and friends, and only if
3393 the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>. If the flags
3394 contain C<SV_GMAGIC>, then it does an C<mg_get()> first.
3401 Perl_sv_2bool_flags(pTHX_ SV *sv, I32 flags)
3403 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
3406 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
3412 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
3413 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv)))) {
3416 if(SvGMAGICAL(sv)) {
3418 goto restart; /* call sv_2bool */
3420 /* expanded SvTRUE_common(sv, (flags = 0, goto restart)) */
3421 else if(!SvOK(sv)) {
3424 else if(SvPOK(sv)) {
3425 svb = SvPVXtrue(sv);
3427 else if((SvFLAGS(sv) & (SVf_IOK|SVf_NOK))) {
3428 svb = (SvIOK(sv) && SvIVX(sv) != 0)
3429 || (SvNOK(sv) && SvNVX(sv) != 0.0);
3433 goto restart; /* call sv_2bool_nomg */
3443 RX_WRAPLEN(sv) > 1 || (RX_WRAPLEN(sv) && *RX_WRAPPED(sv) != '0');
3445 if (SvNOK(sv) && !SvPOK(sv))
3446 return SvNVX(sv) != 0.0;
3448 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
3452 =for apidoc sv_utf8_upgrade
3454 Converts the PV of an SV to its UTF-8-encoded form.
3455 Forces the SV to string form if it is not already.
3456 Will C<mg_get> on C<sv> if appropriate.
3457 Always sets the C<SvUTF8> flag to avoid future validity checks even
3458 if the whole string is the same in UTF-8 as not.
3459 Returns the number of bytes in the converted string
3461 This is not a general purpose byte encoding to Unicode interface:
3462 use the Encode extension for that.
3464 =for apidoc sv_utf8_upgrade_nomg
3466 Like C<sv_utf8_upgrade>, but doesn't do magic on C<sv>.
3468 =for apidoc sv_utf8_upgrade_flags
3470 Converts the PV of an SV to its UTF-8-encoded form.
3471 Forces the SV to string form if it is not already.
3472 Always sets the SvUTF8 flag to avoid future validity checks even
3473 if all the bytes are invariant in UTF-8.
3474 If C<flags> has C<SV_GMAGIC> bit set,
3475 will C<mg_get> on C<sv> if appropriate, else not.
3477 The C<SV_FORCE_UTF8_UPGRADE> flag is now ignored.
3479 Returns the number of bytes in the converted string.
3481 This is not a general purpose byte encoding to Unicode interface:
3482 use the Encode extension for that.
3484 =for apidoc sv_utf8_upgrade_flags_grow
3486 Like C<sv_utf8_upgrade_flags>, but has an additional parameter C<extra>, which is
3487 the number of unused bytes the string of C<sv> is guaranteed to have free after
3488 it upon return. This allows the caller to reserve extra space that it intends
3489 to fill, to avoid extra grows.
3491 C<sv_utf8_upgrade>, C<sv_utf8_upgrade_nomg>, and C<sv_utf8_upgrade_flags>
3492 are implemented in terms of this function.
3494 Returns the number of bytes in the converted string (not including the spares).
3498 If the routine itself changes the string, it adds a trailing C<NUL>. Such a
3499 C<NUL> isn't guaranteed due to having other routines do the work in some input
3500 cases, or if the input is already flagged as being in utf8.
3505 Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV *const sv, const I32 flags, STRLEN extra)
3507 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
3509 if (sv == &PL_sv_undef)
3511 if (!SvPOK_nog(sv)) {
3513 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3514 (void) sv_2pv_flags(sv,&len, flags);
3516 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3520 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
3524 /* SVt_REGEXP's shouldn't be upgraded to UTF8 - they're already
3525 * compiled and individual nodes will remain non-utf8 even if the
3526 * stringified version of the pattern gets upgraded. Whether the
3527 * PVX of a REGEXP should be grown or we should just croak, I don't
3529 if (SvUTF8(sv) || isREGEXP(sv)) {
3530 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3535 S_sv_uncow(aTHX_ sv, 0);
3538 if (SvCUR(sv) == 0) {
3539 if (extra) SvGROW(sv, extra + 1); /* Make sure is room for a trailing
3541 } else { /* Assume Latin-1/EBCDIC */
3542 /* This function could be much more efficient if we
3543 * had a FLAG in SVs to signal if there are any variant
3544 * chars in the PV. Given that there isn't such a flag
3545 * make the loop as fast as possible. */
3546 U8 * s = (U8 *) SvPVX_const(sv);
3549 if (is_utf8_invariant_string_loc(s, SvCUR(sv), (const U8 **) &t)) {
3551 /* utf8 conversion not needed because all are invariants. Mark
3552 * as UTF-8 even if no variant - saves scanning loop */
3554 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3558 /* Here, there is at least one variant (t points to the first one), so
3559 * the string should be converted to utf8. Everything from 's' to
3560 * 't - 1' will occupy only 1 byte each on output.
3562 * Note that the incoming SV may not have a trailing '\0', as certain
3563 * code in pp_formline can send us partially built SVs.
3565 * There are two main ways to convert. One is to create a new string
3566 * and go through the input starting from the beginning, appending each
3567 * converted value onto the new string as we go along. Going this
3568 * route, it's probably best to initially allocate enough space in the
3569 * string rather than possibly running out of space and having to
3570 * reallocate and then copy what we've done so far. Since everything
3571 * from 's' to 't - 1' is invariant, the destination can be initialized
3572 * with these using a fast memory copy. To be sure to allocate enough
3573 * space, one could use the worst case scenario, where every remaining
3574 * byte expands to two under UTF-8, or one could parse it and count
3575 * exactly how many do expand.
3577 * The other way is to unconditionally parse the remainder of the
3578 * string to figure out exactly how big the expanded string will be,
3579 * growing if needed. Then start at the end of the string and place
3580 * the character there at the end of the unfilled space in the expanded
3581 * one, working backwards until reaching 't'.
3583 * The problem with assuming the worst case scenario is that for very
3584 * long strings, we could allocate much more memory than actually
3585 * needed, which can create performance problems. If we have to parse
3586 * anyway, the second method is the winner as it may avoid an extra
3587 * copy. The code used to use the first method under some
3588 * circumstances, but now that there is faster variant counting on
3589 * ASCII platforms, the second method is used exclusively, eliminating
3590 * some code that no longer has to be maintained. */
3593 /* Count the total number of variants there are. We can start
3594 * just beyond the first one, which is known to be at 't' */
3595 const Size_t invariant_length = t - s;
3596 U8 * e = (U8 *) SvEND(sv);
3598 /* The length of the left overs, plus 1. */
3599 const Size_t remaining_length_p1 = e - t;
3601 /* We expand by 1 for the variant at 't' and one for each remaining
3602 * variant (we start looking at 't+1') */
3603 Size_t expansion = 1 + variant_under_utf8_count(t + 1, e);
3605 /* +1 = trailing NUL */
3606 Size_t need = SvCUR(sv) + expansion + extra + 1;
3609 /* Grow if needed */
3610 if (SvLEN(sv) < need) {
3611 t = invariant_length + (U8*) SvGROW(sv, need);
3612 e = t + remaining_length_p1;
3614 SvCUR_set(sv, invariant_length + remaining_length_p1 + expansion);
3616 /* Set the NUL at the end */
3617 d = (U8 *) SvEND(sv);
3620 /* Having decremented d, it points to the position to put the
3621 * very last byte of the expanded string. Go backwards through
3622 * the string, copying and expanding as we go, stopping when we
3623 * get to the part that is invariant the rest of the way down */
3627 if (NATIVE_BYTE_IS_INVARIANT(*e)) {
3630 *d-- = UTF8_EIGHT_BIT_LO(*e);
3631 *d-- = UTF8_EIGHT_BIT_HI(*e);
3636 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3637 /* Update pos. We do it at the end rather than during
3638 * the upgrade, to avoid slowing down the common case
3639 * (upgrade without pos).
3640 * pos can be stored as either bytes or characters. Since
3641 * this was previously a byte string we can just turn off
3642 * the bytes flag. */
3643 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3645 mg->mg_flags &= ~MGf_BYTES;
3647 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3648 magic_setutf8(sv,mg); /* clear UTF8 cache */
3658 =for apidoc sv_utf8_downgrade
3660 Attempts to convert the PV of an SV from characters to bytes.
3661 If the PV contains a character that cannot fit
3662 in a byte, this conversion will fail;
3663 in this case, either returns false or, if C<fail_ok> is not
3666 This is not a general purpose Unicode to byte encoding interface:
3667 use the C<Encode> extension for that.
3669 This function process get magic on C<sv>.
3671 =for apidoc sv_utf8_downgrade_nomg
3673 Like C<sv_utf8_downgrade>, but does not process get magic on C<sv>.
3675 =for apidoc sv_utf8_downgrade_flags
3677 Like C<sv_utf8_downgrade>, but with additional C<flags>.
3678 If C<flags> has C<SV_GMAGIC> bit set, processes get magic on C<sv>.
3684 Perl_sv_utf8_downgrade_flags(pTHX_ SV *const sv, const bool fail_ok, const U32 flags)
3686 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE_FLAGS;
3688 if (SvPOKp(sv) && SvUTF8(sv)) {
3692 U32 mg_flags = flags & SV_GMAGIC;
3695 S_sv_uncow(aTHX_ sv, 0);
3697 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3699 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3700 if (mg && mg->mg_len > 0 && mg->mg_flags & MGf_BYTES) {
3701 mg->mg_len = sv_pos_b2u_flags(sv, mg->mg_len,
3702 mg_flags|SV_CONST_RETURN);
3703 mg_flags = 0; /* sv_pos_b2u does get magic */
3705 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3706 magic_setutf8(sv,mg); /* clear UTF8 cache */
3709 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3711 if (!utf8_to_bytes(s, &len)) {
3716 Perl_croak(aTHX_ "Wide character in %s",
3719 Perl_croak(aTHX_ "Wide character");
3730 =for apidoc sv_utf8_encode
3732 Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3733 flag off so that it looks like octets again.
3739 Perl_sv_utf8_encode(pTHX_ SV *const sv)
3741 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3743 if (SvREADONLY(sv)) {
3744 sv_force_normal_flags(sv, 0);
3746 (void) sv_utf8_upgrade(sv);
3751 =for apidoc sv_utf8_decode
3753 If the PV of the SV is an octet sequence in Perl's extended UTF-8
3754 and contains a multiple-byte character, the C<SvUTF8> flag is turned on
3755 so that it looks like a character. If the PV contains only single-byte
3756 characters, the C<SvUTF8> flag stays off.
3757 Scans PV for validity and returns FALSE if the PV is invalid UTF-8.
3763 Perl_sv_utf8_decode(pTHX_ SV *const sv)
3765 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3768 const U8 *start, *c, *first_variant;
3770 /* The octets may have got themselves encoded - get them back as
3773 if (!sv_utf8_downgrade(sv, TRUE))
3776 /* it is actually just a matter of turning the utf8 flag on, but
3777 * we want to make sure everything inside is valid utf8 first.
3779 c = start = (const U8 *) SvPVX_const(sv);
3780 if (! is_utf8_invariant_string_loc(c, SvCUR(sv), &first_variant)) {
3781 if (!is_utf8_string(first_variant, SvCUR(sv) - (first_variant -c)))
3785 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3786 /* XXX Is this dead code? XS_utf8_decode calls SvSETMAGIC
3787 after this, clearing pos. Does anything on CPAN
3789 /* adjust pos to the start of a UTF8 char sequence */
3790 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3792 I32 pos = mg->mg_len;
3794 for (c = start + pos; c > start; c--) {
3795 if (UTF8_IS_START(*c))
3798 mg->mg_len = c - start;
3801 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3802 magic_setutf8(sv,mg); /* clear UTF8 cache */
3809 =for apidoc sv_setsv
3811 Copies the contents of the source SV C<ssv> into the destination SV
3812 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3813 function if the source SV needs to be reused. Does not handle 'set' magic on
3814 destination SV. Calls 'get' magic on source SV. Loosely speaking, it
3815 performs a copy-by-value, obliterating any previous content of the
3818 You probably want to use one of the assortment of wrappers, such as
3819 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3820 C<SvSetMagicSV_nosteal>.
3822 =for apidoc sv_setsv_flags
3824 Copies the contents of the source SV C<ssv> into the destination SV
3825 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3826 function if the source SV needs to be reused. Does not handle 'set' magic.
3827 Loosely speaking, it performs a copy-by-value, obliterating any previous
3828 content of the destination.
3829 If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
3830 C<ssv> if appropriate, else not. If the C<flags>
3831 parameter has the C<SV_NOSTEAL> bit set then the
3832 buffers of temps will not be stolen. C<sv_setsv>
3833 and C<sv_setsv_nomg> are implemented in terms of this function.
3835 You probably want to use one of the assortment of wrappers, such as
3836 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3837 C<SvSetMagicSV_nosteal>.
3839 This is the primary function for copying scalars, and most other
3840 copy-ish functions and macros use this underneath.
3842 =for apidoc Amnh||SV_NOSTEAL
3848 S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
3850 I32 mro_changes = 0; /* 1 = method, 2 = isa, 3 = recursive isa */
3851 HV *old_stash = NULL;
3853 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB;
3855 if (dtype != SVt_PVGV && !isGV_with_GP(dstr)) {
3856 const char * const name = GvNAME(sstr);
3857 const STRLEN len = GvNAMELEN(sstr);
3859 if (dtype >= SVt_PV) {
3865 SvUPGRADE(dstr, SVt_PVGV);
3866 (void)SvOK_off(dstr);
3867 isGV_with_GP_on(dstr);
3869 GvSTASH(dstr) = GvSTASH(sstr);
3871 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
3872 gv_name_set(MUTABLE_GV(dstr), name, len,
3873 GV_ADD | (GvNAMEUTF8(sstr) ? SVf_UTF8 : 0 ));
3874 SvFAKE_on(dstr); /* can coerce to non-glob */
3877 if(GvGP(MUTABLE_GV(sstr))) {
3878 /* If source has method cache entry, clear it */
3880 SvREFCNT_dec(GvCV(sstr));
3881 GvCV_set(sstr, NULL);
3884 /* If source has a real method, then a method is
3887 GvCV((const GV *)sstr) && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3893 /* If dest already had a real method, that's a change as well */
3895 !mro_changes && GvGP(MUTABLE_GV(dstr)) && GvCVu((const GV *)dstr)
3896 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3901 /* We don't need to check the name of the destination if it was not a
3902 glob to begin with. */
3903 if(dtype == SVt_PVGV) {
3904 const char * const name = GvNAME((const GV *)dstr);
3905 const STRLEN len = GvNAMELEN(dstr);
3906 if(memEQs(name, len, "ISA")
3907 /* The stash may have been detached from the symbol table, so
3909 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3913 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
3914 || (len == 1 && name[0] == ':')) {
3917 /* Set aside the old stash, so we can reset isa caches on
3919 if((old_stash = GvHV(dstr)))
3920 /* Make sure we do not lose it early. */
3921 SvREFCNT_inc_simple_void_NN(
3922 sv_2mortal((SV *)old_stash)
3927 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
3930 /* freeing dstr's GP might free sstr (e.g. *x = $x),
3931 * so temporarily protect it */
3933 SAVEFREESV(SvREFCNT_inc_simple_NN(sstr));
3934 gp_free(MUTABLE_GV(dstr));
3935 GvINTRO_off(dstr); /* one-shot flag */
3936 GvGP_set(dstr, gp_ref(GvGP(sstr)));
3939 if (SvTAINTED(sstr))
3941 if (GvIMPORTED(dstr) != GVf_IMPORTED
3942 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
3944 GvIMPORTED_on(dstr);
3947 if(mro_changes == 2) {
3948 if (GvAV((const GV *)sstr)) {
3950 SV * const sref = (SV *)GvAV((const GV *)dstr);
3951 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
3952 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
3953 AV * const ary = newAV();
3954 av_push(ary, mg->mg_obj); /* takes the refcount */
3955 mg->mg_obj = (SV *)ary;
3957 av_push((AV *)mg->mg_obj, SvREFCNT_inc_simple_NN(dstr));
3959 else sv_magic(sref, dstr, PERL_MAGIC_isa, NULL, 0);
3961 mro_isa_changed_in(GvSTASH(dstr));
3963 else if(mro_changes == 3) {
3964 HV * const stash = GvHV(dstr);
3965 if(old_stash ? (HV *)HvENAME_get(old_stash) : stash)
3971 else if(mro_changes) mro_method_changed_in(GvSTASH(dstr));
3972 if (GvIO(dstr) && dtype == SVt_PVGV) {
3973 DEBUG_o(Perl_deb(aTHX_
3974 "glob_assign_glob clearing PL_stashcache\n"));
3975 /* It's a cache. It will rebuild itself quite happily.
3976 It's a lot of effort to work out exactly which key (or keys)
3977 might be invalidated by the creation of the this file handle.
3979 hv_clear(PL_stashcache);
3985 Perl_gv_setref(pTHX_ SV *const dstr, SV *const sstr)
3987 SV * const sref = SvRV(sstr);
3989 const int intro = GvINTRO(dstr);
3992 const U32 stype = SvTYPE(sref);
3994 PERL_ARGS_ASSERT_GV_SETREF;
3997 GvINTRO_off(dstr); /* one-shot flag */
3998 GvLINE(dstr) = CopLINE(PL_curcop);
3999 GvEGV(dstr) = MUTABLE_GV(dstr);
4004 location = (SV **) &(GvGP(dstr)->gp_cv); /* XXX bypassing GvCV_set */
4005 import_flag = GVf_IMPORTED_CV;
4008 location = (SV **) &GvHV(dstr);
4009 import_flag = GVf_IMPORTED_HV;
4012 location = (SV **) &GvAV(dstr);
4013 import_flag = GVf_IMPORTED_AV;
4016 location = (SV **) &GvIOp(dstr);
4019 location = (SV **) &GvFORM(dstr);
4022 location = &GvSV(dstr);
4023 import_flag = GVf_IMPORTED_SV;
4026 if (stype == SVt_PVCV) {
4027 /*if (GvCVGEN(dstr) && (GvCV(dstr) != (const CV *)sref || GvCVGEN(dstr))) {*/
4028 if (GvCVGEN(dstr)) {
4029 SvREFCNT_dec(GvCV(dstr));
4030 GvCV_set(dstr, NULL);
4031 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4034 /* SAVEt_GVSLOT takes more room on the savestack and has more
4035 overhead in leave_scope than SAVEt_GENERIC_SV. But for CVs
4036 leave_scope needs access to the GV so it can reset method
4037 caches. We must use SAVEt_GVSLOT whenever the type is
4038 SVt_PVCV, even if the stash is anonymous, as the stash may
4039 gain a name somehow before leave_scope. */
4040 if (stype == SVt_PVCV) {
4041 /* There is no save_pushptrptrptr. Creating it for this
4042 one call site would be overkill. So inline the ss add
4046 SS_ADD_PTR(location);
4047 SS_ADD_PTR(SvREFCNT_inc(*location));
4048 SS_ADD_UV(SAVEt_GVSLOT);
4051 else SAVEGENERICSV(*location);
4054 if (stype == SVt_PVCV && (*location != sref || GvCVGEN(dstr))) {
4055 CV* const cv = MUTABLE_CV(*location);
4057 if (!GvCVGEN((const GV *)dstr) &&
4058 (CvROOT(cv) || CvXSUB(cv)) &&
4059 /* redundant check that avoids creating the extra SV
4060 most of the time: */
4061 (CvCONST(cv) || ckWARN(WARN_REDEFINE)))
4063 SV * const new_const_sv =
4064 CvCONST((const CV *)sref)
4065 ? cv_const_sv((const CV *)sref)
4067 HV * const stash = GvSTASH((const GV *)dstr);
4068 report_redefined_cv(
4071 ? Perl_newSVpvf(aTHX_
4072 "%" HEKf "::%" HEKf,
4073 HEKfARG(HvNAME_HEK(stash)),
4074 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr))))
4075 : Perl_newSVpvf(aTHX_
4077 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr))))
4080 CvCONST((const CV *)sref) ? &new_const_sv : NULL
4084 cv_ckproto_len_flags(cv, (const GV *)dstr,
4085 SvPOK(sref) ? CvPROTO(sref) : NULL,
4086 SvPOK(sref) ? CvPROTOLEN(sref) : 0,
4087 SvPOK(sref) ? SvUTF8(sref) : 0);
4089 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4090 GvASSUMECV_on(dstr);
4091 if(GvSTASH(dstr)) { /* sub foo { 1 } sub bar { 2 } *bar = \&foo */
4092 if (intro && GvREFCNT(dstr) > 1) {
4093 /* temporary remove extra savestack's ref */
4095 gv_method_changed(dstr);
4098 else gv_method_changed(dstr);
4101 *location = SvREFCNT_inc_simple_NN(sref);
4102 if (import_flag && !(GvFLAGS(dstr) & import_flag)
4103 && CopSTASH_ne(PL_curcop, GvSTASH(dstr))) {
4104 GvFLAGS(dstr) |= import_flag;
4107 if (stype == SVt_PVHV) {
4108 const char * const name = GvNAME((GV*)dstr);
4109 const STRLEN len = GvNAMELEN(dstr);
4112 (len > 1 && name[len-2] == ':' && name[len-1] == ':')
4113 || (len == 1 && name[0] == ':')
4115 && (!dref || HvENAME_get(dref))
4118 (HV *)sref, (HV *)dref,
4124 stype == SVt_PVAV && sref != dref
4125 && memEQs(GvNAME((GV*)dstr), GvNAMELEN((GV*)dstr), "ISA")
4126 /* The stash may have been detached from the symbol table, so
4127 check its name before doing anything. */
4128 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
4131 MAGIC * const omg = dref && SvSMAGICAL(dref)
4132 ? mg_find(dref, PERL_MAGIC_isa)
4134 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
4135 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
4136 AV * const ary = newAV();
4137 av_push(ary, mg->mg_obj); /* takes the refcount */
4138 mg->mg_obj = (SV *)ary;
4141 if (SvTYPE(omg->mg_obj) == SVt_PVAV) {
4142 SV **svp = AvARRAY((AV *)omg->mg_obj);
4143 I32 items = AvFILLp((AV *)omg->mg_obj) + 1;
4147 SvREFCNT_inc_simple_NN(*svp++)
4153 SvREFCNT_inc_simple_NN(omg->mg_obj)
4157 av_push((AV *)mg->mg_obj,SvREFCNT_inc_simple_NN(dstr));
4163 sref, omg ? omg->mg_obj : dstr, PERL_MAGIC_isa, NULL, 0
4165 for (i = 0; i <= AvFILL(sref); ++i) {
4166 SV **elem = av_fetch ((AV*)sref, i, 0);
4169 *elem, sref, PERL_MAGIC_isaelem, NULL, i
4173 mg = mg_find(sref, PERL_MAGIC_isa);
4175 /* Since the *ISA assignment could have affected more than
4176 one stash, don't call mro_isa_changed_in directly, but let
4177 magic_clearisa do it for us, as it already has the logic for
4178 dealing with globs vs arrays of globs. */
4180 Perl_magic_clearisa(aTHX_ NULL, mg);
4182 else if (stype == SVt_PVIO) {
4183 DEBUG_o(Perl_deb(aTHX_ "gv_setref clearing PL_stashcache\n"));
4184 /* It's a cache. It will rebuild itself quite happily.
4185 It's a lot of effort to work out exactly which key (or keys)
4186 might be invalidated by the creation of the this file handle.
4188 hv_clear(PL_stashcache);
4192 if (!intro) SvREFCNT_dec(dref);
4193 if (SvTAINTED(sstr))
4201 #ifdef PERL_DEBUG_READONLY_COW
4202 # include <sys/mman.h>
4204 # ifndef PERL_MEMORY_DEBUG_HEADER_SIZE
4205 # define PERL_MEMORY_DEBUG_HEADER_SIZE 0
4209 Perl_sv_buf_to_ro(pTHX_ SV *sv)
4211 struct perl_memory_debug_header * const header =
4212 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4213 const MEM_SIZE len = header->size;
4214 PERL_ARGS_ASSERT_SV_BUF_TO_RO;
4215 # ifdef PERL_TRACK_MEMPOOL
4216 if (!header->readonly) header->readonly = 1;
4218 if (mprotect(header, len, PROT_READ))
4219 Perl_warn(aTHX_ "mprotect RW for COW string %p %lu failed with %d",
4220 header, len, errno);
4224 S_sv_buf_to_rw(pTHX_ SV *sv)
4226 struct perl_memory_debug_header * const header =
4227 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4228 const MEM_SIZE len = header->size;
4229 PERL_ARGS_ASSERT_SV_BUF_TO_RW;
4230 if (mprotect(header, len, PROT_READ|PROT_WRITE))
4231 Perl_warn(aTHX_ "mprotect for COW string %p %lu failed with %d",
4232 header, len, errno);
4233 # ifdef PERL_TRACK_MEMPOOL
4234 header->readonly = 0;
4239 # define sv_buf_to_ro(sv) NOOP
4240 # define sv_buf_to_rw(sv) NOOP
4244 Perl_sv_setsv_flags(pTHX_ SV *dstr, SV* sstr, const I32 flags)
4249 unsigned int both_type;
4251 PERL_ARGS_ASSERT_SV_SETSV_FLAGS;
4253 if (UNLIKELY( sstr == dstr ))
4256 if (UNLIKELY( !sstr ))
4257 sstr = &PL_sv_undef;
4259 stype = SvTYPE(sstr);
4260 dtype = SvTYPE(dstr);
4261 both_type = (stype | dtype);
4263 /* with these values, we can check that both SVs are NULL/IV (and not
4264 * freed) just by testing the or'ed types */
4265 STATIC_ASSERT_STMT(SVt_NULL == 0);
4266 STATIC_ASSERT_STMT(SVt_IV == 1);
4267 if (both_type <= 1) {
4268 /* both src and dst are UNDEF/IV/RV, so we can do a lot of
4274 /* minimal subset of SV_CHECK_THINKFIRST_COW_DROP(dstr) */
4275 if (SvREADONLY(dstr))
4276 Perl_croak_no_modify();
4278 if (SvWEAKREF(dstr))
4279 sv_unref_flags(dstr, 0);
4281 old_rv = SvRV(dstr);
4284 assert(!SvGMAGICAL(sstr));
4285 assert(!SvGMAGICAL(dstr));
4287 sflags = SvFLAGS(sstr);
4288 if (sflags & (SVf_IOK|SVf_ROK)) {
4289 SET_SVANY_FOR_BODYLESS_IV(dstr);
4290 new_dflags = SVt_IV;
4292 if (sflags & SVf_ROK) {
4293 dstr->sv_u.svu_rv = SvREFCNT_inc(SvRV(sstr));
4294 new_dflags |= SVf_ROK;
4297 /* both src and dst are <= SVt_IV, so sv_any points to the
4298 * head; so access the head directly
4300 assert( &(sstr->sv_u.svu_iv)
4301 == &(((XPVIV*) SvANY(sstr))->xiv_iv));
4302 assert( &(dstr->sv_u.svu_iv)
4303 == &(((XPVIV*) SvANY(dstr))->xiv_iv));
4304 dstr->sv_u.svu_iv = sstr->sv_u.svu_iv;
4305 new_dflags |= (SVf_IOK|SVp_IOK|(sflags & SVf_IVisUV));
4309 new_dflags = dtype; /* turn off everything except the type */
4311 SvFLAGS(dstr) = new_dflags;
4312 SvREFCNT_dec(old_rv);
4317 if (UNLIKELY(both_type == SVTYPEMASK)) {
4318 if (SvIS_FREED(dstr)) {
4319 Perl_croak(aTHX_ "panic: attempt to copy value %" SVf
4320 " to a freed scalar %p", SVfARG(sstr), (void *)dstr);
4322 if (SvIS_FREED(sstr)) {
4323 Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p to %p",
4324 (void*)sstr, (void*)dstr);
4330 SV_CHECK_THINKFIRST_COW_DROP(dstr);
4331 dtype = SvTYPE(dstr); /* THINKFIRST may have changed type */
4333 /* There's a lot of redundancy below but we're going for speed here */
4338 if (LIKELY( dtype != SVt_PVGV && dtype != SVt_PVLV )) {
4339 (void)SvOK_off(dstr);
4347 /* For performance, we inline promoting to type SVt_IV. */
4348 /* We're starting from SVt_NULL, so provided that define is
4349 * actual 0, we don't have to unset any SV type flags
4350 * to promote to SVt_IV. */
4351 STATIC_ASSERT_STMT(SVt_NULL == 0);
4352 SET_SVANY_FOR_BODYLESS_IV(dstr);
4353 SvFLAGS(dstr) |= SVt_IV;
4357 sv_upgrade(dstr, SVt_PVIV);
4361 goto end_of_first_switch;
4363 (void)SvIOK_only(dstr);
4364 SvIV_set(dstr, SvIVX(sstr));
4367 /* SvTAINTED can only be true if the SV has taint magic, which in
4368 turn means that the SV type is PVMG (or greater). This is the
4369 case statement for SVt_IV, so this cannot be true (whatever gcov
4371 assert(!SvTAINTED(sstr));
4376 if (dtype < SVt_PV && dtype != SVt_IV)
4377 sv_upgrade(dstr, SVt_IV);
4381 if (LIKELY( SvNOK(sstr) )) {
4385 sv_upgrade(dstr, SVt_NV);
4389 sv_upgrade(dstr, SVt_PVNV);
4393 goto end_of_first_switch;
4395 SvNV_set(dstr, SvNVX(sstr));
4396 (void)SvNOK_only(dstr);
4397 /* SvTAINTED can only be true if the SV has taint magic, which in
4398 turn means that the SV type is PVMG (or greater). This is the
4399 case statement for SVt_NV, so this cannot be true (whatever gcov
4401 assert(!SvTAINTED(sstr));
4408 sv_upgrade(dstr, SVt_PV);
4411 if (dtype < SVt_PVIV)
4412 sv_upgrade(dstr, SVt_PVIV);
4415 if (dtype < SVt_PVNV)
4416 sv_upgrade(dstr, SVt_PVNV);
4420 invlist_clone(sstr, dstr);
4424 const char * const type = sv_reftype(sstr,0);
4426 /* diag_listed_as: Bizarre copy of %s */
4427 Perl_croak(aTHX_ "Bizarre copy of %s in %s", type, OP_DESC(PL_op));
4429 Perl_croak(aTHX_ "Bizarre copy of %s", type);
4431 NOT_REACHED; /* NOTREACHED */
4435 if (dtype < SVt_REGEXP)
4436 sv_upgrade(dstr, SVt_REGEXP);
4442 if (SvGMAGICAL(sstr) && (flags & SV_GMAGIC)) {
4444 if (SvTYPE(sstr) != stype)
4445 stype = SvTYPE(sstr);
4447 if (isGV_with_GP(sstr) && dtype <= SVt_PVLV) {
4448 glob_assign_glob(dstr, sstr, dtype);
4451 if (stype == SVt_PVLV)
4453 if (isREGEXP(sstr)) goto upgregexp;
4454 SvUPGRADE(dstr, SVt_PVNV);
4457 SvUPGRADE(dstr, (svtype)stype);
4459 end_of_first_switch:
4461 /* dstr may have been upgraded. */
4462 dtype = SvTYPE(dstr);
4463 sflags = SvFLAGS(sstr);
4465 if (UNLIKELY( dtype == SVt_PVCV )) {
4466 /* Assigning to a subroutine sets the prototype. */
4469 const char *const ptr = SvPV_const(sstr, len);
4471 SvGROW(dstr, len + 1);
4472 Copy(ptr, SvPVX(dstr), len + 1, char);
4473 SvCUR_set(dstr, len);
4475 SvFLAGS(dstr) |= sflags & SVf_UTF8;
4476 CvAUTOLOAD_off(dstr);
4481 else if (UNLIKELY(dtype == SVt_PVAV || dtype == SVt_PVHV
4482 || dtype == SVt_PVFM))
4484 const char * const type = sv_reftype(dstr,0);
4486 /* diag_listed_as: Cannot copy to %s */
4487 Perl_croak(aTHX_ "Cannot copy to %s in %s", type, OP_DESC(PL_op));
4489 Perl_croak(aTHX_ "Cannot copy to %s", type);
4490 } else if (sflags & SVf_ROK) {
4491 if (isGV_with_GP(dstr)
4492 && SvTYPE(SvRV(sstr)) == SVt_PVGV && isGV_with_GP(SvRV(sstr))) {
4495 if (GvIMPORTED(dstr) != GVf_IMPORTED
4496 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
4498 GvIMPORTED_on(dstr);
4503 glob_assign_glob(dstr, sstr, dtype);
4507 if (dtype >= SVt_PV) {
4508 if (isGV_with_GP(dstr)) {
4509 gv_setref(dstr, sstr);
4512 if (SvPVX_const(dstr)) {
4518 (void)SvOK_off(dstr);
4519 SvRV_set(dstr, SvREFCNT_inc(SvRV(sstr)));
4520 SvFLAGS(dstr) |= sflags & SVf_ROK;
4521 assert(!(sflags & SVp_NOK));
4522 assert(!(sflags & SVp_IOK));
4523 assert(!(sflags & SVf_NOK));
4524 assert(!(sflags & SVf_IOK));
4526 else if (isGV_with_GP(dstr)) {
4527 if (!(sflags & SVf_OK)) {
4528 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4529 "Undefined value assigned to typeglob");
4532 GV *gv = gv_fetchsv_nomg(sstr, GV_ADD, SVt_PVGV);
4533 if (dstr != (const SV *)gv) {
4534 const char * const name = GvNAME((const GV *)dstr);
4535 const STRLEN len = GvNAMELEN(dstr);
4536 HV *old_stash = NULL;
4537 bool reset_isa = FALSE;
4538 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
4539 || (len == 1 && name[0] == ':')) {
4540 /* Set aside the old stash, so we can reset isa caches
4541 on its subclasses. */
4542 if((old_stash = GvHV(dstr))) {
4543 /* Make sure we do not lose it early. */
4544 SvREFCNT_inc_simple_void_NN(
4545 sv_2mortal((SV *)old_stash)
4552 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
4553 gp_free(MUTABLE_GV(dstr));
4555 GvGP_set(dstr, gp_ref(GvGP(gv)));
4558 HV * const stash = GvHV(dstr);
4560 old_stash ? (HV *)HvENAME_get(old_stash) : stash
4570 else if ((dtype == SVt_REGEXP || dtype == SVt_PVLV)
4571 && (stype == SVt_REGEXP || isREGEXP(sstr))) {
4572 reg_temp_copy((REGEXP*)dstr, (REGEXP*)sstr);
4574 else if (sflags & SVp_POK) {
4575 const STRLEN cur = SvCUR(sstr);
4576 const STRLEN len = SvLEN(sstr);
4579 * We have three basic ways to copy the string:
4585 * Which we choose is based on various factors. The following
4586 * things are listed in order of speed, fastest to slowest:
4588 * - Copying a short string
4589 * - Copy-on-write bookkeeping
4591 * - Copying a long string
4593 * We swipe the string (steal the string buffer) if the SV on the
4594 * rhs is about to be freed anyway (TEMP and refcnt==1). This is a
4595 * big win on long strings. It should be a win on short strings if
4596 * SvPVX_const(dstr) has to be allocated. If not, it should not
4597 * slow things down, as SvPVX_const(sstr) would have been freed
4600 * We also steal the buffer from a PADTMP (operator target) if it
4601 * is ‘long enough’. For short strings, a swipe does not help
4602 * here, as it causes more malloc calls the next time the target
4603 * is used. Benchmarks show that even if SvPVX_const(dstr) has to
4604 * be allocated it is still not worth swiping PADTMPs for short
4605 * strings, as the savings here are small.
4607 * If swiping is not an option, then we see whether it is
4608 * worth using copy-on-write. If the lhs already has a buf-
4609 * fer big enough and the string is short, we skip it and fall back
4610 * to method 3, since memcpy is faster for short strings than the
4611 * later bookkeeping overhead that copy-on-write entails.
4613 * If the rhs is not a copy-on-write string yet, then we also
4614 * consider whether the buffer is too large relative to the string
4615 * it holds. Some operations such as readline allocate a large
4616 * buffer in the expectation of reusing it. But turning such into
4617 * a COW buffer is counter-productive because it increases memory
4618 * usage by making readline allocate a new large buffer the sec-
4619 * ond time round. So, if the buffer is too large, again, we use
4622 * Finally, if there is no buffer on the left, or the buffer is too
4623 * small, then we use copy-on-write and make both SVs share the
4628 /* Whichever path we take through the next code, we want this true,
4629 and doing it now facilitates the COW check. */
4630 (void)SvPOK_only(dstr);
4634 /* slated for free anyway (and not COW)? */
4635 (sflags & (SVs_TEMP|SVf_IsCOW)) == SVs_TEMP
4636 /* or a swipable TARG */
4638 (SVs_PADTMP|SVf_READONLY|SVf_PROTECT|SVf_IsCOW))
4640 /* whose buffer is worth stealing */
4641 && CHECK_COWBUF_THRESHOLD(cur,len)
4644 !(sflags & SVf_OOK) && /* and not involved in OOK hack? */
4645 (!(flags & SV_NOSTEAL)) &&
4646 /* and we're allowed to steal temps */
4647 SvREFCNT(sstr) == 1 && /* and no other references to it? */
4648 len) /* and really is a string */
4649 { /* Passes the swipe test. */
4650 if (SvPVX_const(dstr)) /* we know that dtype >= SVt_PV */
4652 SvPV_set(dstr, SvPVX_mutable(sstr));
4653 SvLEN_set(dstr, SvLEN(sstr));
4654 SvCUR_set(dstr, SvCUR(sstr));
4657 (void)SvOK_off(sstr); /* NOTE: nukes most SvFLAGS on sstr */
4658 SvPV_set(sstr, NULL);
4663 else if (flags & SV_COW_SHARED_HASH_KEYS
4665 #ifdef PERL_COPY_ON_WRITE
4668 ( (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4669 /* If this is a regular (non-hek) COW, only so
4670 many COW "copies" are possible. */
4671 && CowREFCNT(sstr) != SV_COW_REFCNT_MAX ))
4672 : ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4673 && !(SvFLAGS(dstr) & SVf_BREAK)
4674 && CHECK_COW_THRESHOLD(cur,len) && cur+1 < len
4675 && (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4679 && !(SvFLAGS(dstr) & SVf_BREAK)
4682 /* Either it's a shared hash key, or it's suitable for
4686 PerlIO_printf(Perl_debug_log, "Copy on write: sstr --> dstr\n");
4692 if (!(sflags & SVf_IsCOW)) {
4694 CowREFCNT(sstr) = 0;
4697 if (SvPVX_const(dstr)) { /* we know that dtype >= SVt_PV */
4703 if (sflags & SVf_IsCOW) {
4707 SvPV_set(dstr, SvPVX_mutable(sstr));
4712 /* SvIsCOW_shared_hash */
4713 DEBUG_C(PerlIO_printf(Perl_debug_log,
4714 "Copy on write: Sharing hash\n"));
4716 assert (SvTYPE(dstr) >= SVt_PV);
4718 HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)))));
4720 SvLEN_set(dstr, len);
4721 SvCUR_set(dstr, cur);
4724 /* Failed the swipe test, and we cannot do copy-on-write either.
4725 Have to copy the string. */
4726 SvGROW(dstr, cur + 1); /* inlined from sv_setpvn */
4727 Move(SvPVX_const(sstr),SvPVX(dstr),cur,char);
4728 SvCUR_set(dstr, cur);
4729 *SvEND(dstr) = '\0';
4731 if (sflags & SVp_NOK) {
4732 SvNV_set(dstr, SvNVX(sstr));
4734 if (sflags & SVp_IOK) {
4735 SvIV_set(dstr, SvIVX(sstr));
4736 if (sflags & SVf_IVisUV)
4739 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_NOK|SVp_NOK|SVf_UTF8);
4741 const MAGIC * const smg = SvVSTRING_mg(sstr);
4743 sv_magic(dstr, NULL, PERL_MAGIC_vstring,
4744 smg->mg_ptr, smg->mg_len);
4745 SvRMAGICAL_on(dstr);
4749 else if (sflags & (SVp_IOK|SVp_NOK)) {
4750 (void)SvOK_off(dstr);
4751 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_IVisUV|SVf_NOK|SVp_NOK);
4752 if (sflags & SVp_IOK) {
4753 /* XXXX Do we want to set IsUV for IV(ROK)? Be extra safe... */
4754 SvIV_set(dstr, SvIVX(sstr));
4756 if (sflags & SVp_NOK) {
4757 SvNV_set(dstr, SvNVX(sstr));
4761 if (isGV_with_GP(sstr)) {
4762 gv_efullname3(dstr, MUTABLE_GV(sstr), "*");
4765 (void)SvOK_off(dstr);
4767 if (SvTAINTED(sstr))
4773 =for apidoc sv_set_undef
4775 Equivalent to C<sv_setsv(sv, &PL_sv_undef)>, but more efficient.
4776 Doesn't handle set magic.
4778 The perl equivalent is C<$sv = undef;>. Note that it doesn't free any string
4779 buffer, unlike C<undef $sv>.
4781 Introduced in perl 5.25.12.
4787 Perl_sv_set_undef(pTHX_ SV *sv)
4789 U32 type = SvTYPE(sv);
4791 PERL_ARGS_ASSERT_SV_SET_UNDEF;
4793 /* shortcut, NULL, IV, RV */
4795 if (type <= SVt_IV) {
4796 assert(!SvGMAGICAL(sv));
4797 if (SvREADONLY(sv)) {
4798 /* does undeffing PL_sv_undef count as modifying a read-only
4799 * variable? Some XS code does this */
4800 if (sv == &PL_sv_undef)
4802 Perl_croak_no_modify();
4807 sv_unref_flags(sv, 0);
4810 SvFLAGS(sv) = type; /* quickly turn off all flags */
4811 SvREFCNT_dec_NN(rv);
4815 SvFLAGS(sv) = type; /* quickly turn off all flags */
4820 Perl_croak(aTHX_ "panic: attempt to undefine a freed scalar %p",
4823 SV_CHECK_THINKFIRST_COW_DROP(sv);
4825 if (isGV_with_GP(sv))
4826 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4827 "Undefined value assigned to typeglob");
4835 =for apidoc sv_setsv_mg
4837 Like C<sv_setsv>, but also handles 'set' magic.
4843 Perl_sv_setsv_mg(pTHX_ SV *const dstr, SV *const sstr)
4845 PERL_ARGS_ASSERT_SV_SETSV_MG;
4847 sv_setsv(dstr,sstr);
4852 # define SVt_COW SVt_PV
4854 Perl_sv_setsv_cow(pTHX_ SV *dstr, SV *sstr)
4856 STRLEN cur = SvCUR(sstr);
4857 STRLEN len = SvLEN(sstr);
4859 #if defined(PERL_DEBUG_READONLY_COW) && defined(PERL_COPY_ON_WRITE)
4860 const bool already = cBOOL(SvIsCOW(sstr));
4863 PERL_ARGS_ASSERT_SV_SETSV_COW;
4866 PerlIO_printf(Perl_debug_log, "Fast copy on write: %p -> %p\n",
4867 (void*)sstr, (void*)dstr);
4874 if (SvTHINKFIRST(dstr))
4875 sv_force_normal_flags(dstr, SV_COW_DROP_PV);
4876 else if (SvPVX_const(dstr))
4877 Safefree(SvPVX_mutable(dstr));
4881 SvUPGRADE(dstr, SVt_COW);
4883 assert (SvPOK(sstr));
4884 assert (SvPOKp(sstr));
4886 if (SvIsCOW(sstr)) {
4888 if (SvLEN(sstr) == 0) {
4889 /* source is a COW shared hash key. */
4890 DEBUG_C(PerlIO_printf(Perl_debug_log,
4891 "Fast copy on write: Sharing hash\n"));
4892 new_pv = HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr))));
4895 assert(SvCUR(sstr)+1 < SvLEN(sstr));
4896 assert(CowREFCNT(sstr) < SV_COW_REFCNT_MAX);
4898 assert ((SvFLAGS(sstr) & CAN_COW_MASK) == CAN_COW_FLAGS);
4899 SvUPGRADE(sstr, SVt_COW);
4901 DEBUG_C(PerlIO_printf(Perl_debug_log,
4902 "Fast copy on write: Converting sstr to COW\n"));
4903 CowREFCNT(sstr) = 0;
4905 # ifdef PERL_DEBUG_READONLY_COW
4906 if (already) sv_buf_to_rw(sstr);
4909 new_pv = SvPVX_mutable(sstr);
4913 SvPV_set(dstr, new_pv);
4914 SvFLAGS(dstr) = (SVt_COW|SVf_POK|SVp_POK|SVf_IsCOW);
4917 SvLEN_set(dstr, len);
4918 SvCUR_set(dstr, cur);
4928 =for apidoc sv_setpv_bufsize
4930 Sets the SV to be a string of cur bytes length, with at least
4931 len bytes available. Ensures that there is a null byte at SvEND.
4932 Returns a char * pointer to the SvPV buffer.
4938 Perl_sv_setpv_bufsize(pTHX_ SV *const sv, const STRLEN cur, const STRLEN len)
4942 PERL_ARGS_ASSERT_SV_SETPV_BUFSIZE;
4944 SV_CHECK_THINKFIRST_COW_DROP(sv);
4945 SvUPGRADE(sv, SVt_PV);
4946 pv = SvGROW(sv, len + 1);
4949 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4952 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4957 =for apidoc sv_setpvn
4959 Copies a string (possibly containing embedded C<NUL> characters) into an SV.
4960 The C<len> parameter indicates the number of
4961 bytes to be copied. If the C<ptr> argument is NULL the SV will become
4962 undefined. Does not handle 'set' magic. See C<L</sv_setpvn_mg>>.
4964 The UTF-8 flag is not changed by this function. A terminating NUL byte is
4971 Perl_sv_setpvn(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4975 PERL_ARGS_ASSERT_SV_SETPVN;
4977 SV_CHECK_THINKFIRST_COW_DROP(sv);
4978 if (isGV_with_GP(sv))
4979 Perl_croak_no_modify();
4985 /* len is STRLEN which is unsigned, need to copy to signed */
4988 Perl_croak(aTHX_ "panic: sv_setpvn called with negative strlen %"
4991 SvUPGRADE(sv, SVt_PV);
4993 dptr = SvGROW(sv, len + 1);
4994 Move(ptr,dptr,len,char);
4997 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4999 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
5003 =for apidoc sv_setpvn_mg
5005 Like C<sv_setpvn>, but also handles 'set' magic.
5011 Perl_sv_setpvn_mg(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
5013 PERL_ARGS_ASSERT_SV_SETPVN_MG;
5015 sv_setpvn(sv,ptr,len);
5020 =for apidoc sv_setpv
5022 Copies a string into an SV. The string must be terminated with a C<NUL>
5023 character, and not contain embeded C<NUL>'s.
5024 Does not handle 'set' magic. See C<L</sv_setpv_mg>>.
5030 Perl_sv_setpv(pTHX_ SV *const sv, const char *const ptr)
5034 PERL_ARGS_ASSERT_SV_SETPV;
5036 SV_CHECK_THINKFIRST_COW_DROP(sv);
5042 SvUPGRADE(sv, SVt_PV);
5044 SvGROW(sv, len + 1);
5045 Move(ptr,SvPVX(sv),len+1,char);
5047 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5049 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
5053 =for apidoc sv_setpv_mg
5055 Like C<sv_setpv>, but also handles 'set' magic.
5061 Perl_sv_setpv_mg(pTHX_ SV *const sv, const char *const ptr)
5063 PERL_ARGS_ASSERT_SV_SETPV_MG;
5070 Perl_sv_sethek(pTHX_ SV *const sv, const HEK *const hek)
5072 PERL_ARGS_ASSERT_SV_SETHEK;
5078 if (HEK_LEN(hek) == HEf_SVKEY) {
5079 sv_setsv(sv, *(SV**)HEK_KEY(hek));
5082 const int flags = HEK_FLAGS(hek);
5083 if (flags & HVhek_WASUTF8) {
5084 STRLEN utf8_len = HEK_LEN(hek);
5085 char *as_utf8 = (char *)bytes_to_utf8((U8*)HEK_KEY(hek), &utf8_len);
5086 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
5089 } else if (flags & HVhek_UNSHARED) {
5090 sv_setpvn(sv, HEK_KEY(hek), HEK_LEN(hek));
5093 else SvUTF8_off(sv);
5097 SV_CHECK_THINKFIRST_COW_DROP(sv);
5098 SvUPGRADE(sv, SVt_PV);
5100 SvPV_set(sv,(char *)HEK_KEY(share_hek_hek(hek)));
5101 SvCUR_set(sv, HEK_LEN(hek));
5107 else SvUTF8_off(sv);
5115 =for apidoc sv_usepvn_flags
5117 Tells an SV to use C<ptr> to find its string value. Normally the
5118 string is stored inside the SV, but sv_usepvn allows the SV to use an
5119 outside string. C<ptr> should point to memory that was allocated
5120 by L<C<Newx>|perlclib/Memory Management and String Handling>. It must be
5121 the start of a C<Newx>-ed block of memory, and not a pointer to the
5122 middle of it (beware of L<C<OOK>|perlguts/Offsets> and copy-on-write),
5123 and not be from a non-C<Newx> memory allocator like C<malloc>. The
5124 string length, C<len>, must be supplied. By default this function
5125 will C<Renew> (i.e. realloc, move) the memory pointed to by C<ptr>,
5126 so that pointer should not be freed or used by the programmer after
5127 giving it to C<sv_usepvn>, and neither should any pointers from "behind"
5128 that pointer (e.g. ptr + 1) be used.
5130 If S<C<flags & SV_SMAGIC>> is true, will call C<SvSETMAGIC>. If
5131 S<C<flags & SV_HAS_TRAILING_NUL>> is true, then C<ptr[len]> must be C<NUL>,
5133 will be skipped (i.e. the buffer is actually at least 1 byte longer than
5134 C<len>, and already meets the requirements for storing in C<SvPVX>).
5136 =for apidoc Amnh||SV_SMAGIC
5137 =for apidoc Amnh||SV_HAS_TRAILING_NUL
5143 Perl_sv_usepvn_flags(pTHX_ SV *const sv, char *ptr, const STRLEN len, const U32 flags)
5147 PERL_ARGS_ASSERT_SV_USEPVN_FLAGS;
5149 SV_CHECK_THINKFIRST_COW_DROP(sv);
5150 SvUPGRADE(sv, SVt_PV);
5153 if (flags & SV_SMAGIC)
5157 if (SvPVX_const(sv))
5161 if (flags & SV_HAS_TRAILING_NUL)
5162 assert(ptr[len] == '\0');
5165 allocate = (flags & SV_HAS_TRAILING_NUL)
5167 #ifdef Perl_safesysmalloc_size
5170 PERL_STRLEN_ROUNDUP(len + 1);
5172 if (flags & SV_HAS_TRAILING_NUL) {
5173 /* It's long enough - do nothing.
5174 Specifically Perl_newCONSTSUB is relying on this. */
5177 /* Force a move to shake out bugs in callers. */
5178 char *new_ptr = (char*)safemalloc(allocate);
5179 Copy(ptr, new_ptr, len, char);
5180 PoisonFree(ptr,len,char);
5184 ptr = (char*) saferealloc (ptr, allocate);
5187 #ifdef Perl_safesysmalloc_size
5188 SvLEN_set(sv, Perl_safesysmalloc_size(ptr));
5190 SvLEN_set(sv, allocate);
5194 if (!(flags & SV_HAS_TRAILING_NUL)) {
5197 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5199 if (flags & SV_SMAGIC)
5205 S_sv_uncow(pTHX_ SV * const sv, const U32 flags)
5207 assert(SvIsCOW(sv));
5210 const char * const pvx = SvPVX_const(sv);
5211 const STRLEN len = SvLEN(sv);
5212 const STRLEN cur = SvCUR(sv);
5216 PerlIO_printf(Perl_debug_log,
5217 "Copy on write: Force normal %ld\n",
5223 # ifdef PERL_COPY_ON_WRITE
5225 /* Must do this first, since the CowREFCNT uses SvPVX and
5226 we need to write to CowREFCNT, or de-RO the whole buffer if we are
5227 the only owner left of the buffer. */
5228 sv_buf_to_rw(sv); /* NOOP if RO-ing not supported */
5230 U8 cowrefcnt = CowREFCNT(sv);
5231 if(cowrefcnt != 0) {
5233 CowREFCNT(sv) = cowrefcnt;
5238 /* Else we are the only owner of the buffer. */
5243 /* This SV doesn't own the buffer, so need to Newx() a new one: */
5248 if (flags & SV_COW_DROP_PV) {
5249 /* OK, so we don't need to copy our buffer. */
5252 SvGROW(sv, cur + 1);
5253 Move(pvx,SvPVX(sv),cur,char);
5258 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5266 const char * const pvx = SvPVX_const(sv);
5267 const STRLEN len = SvCUR(sv);
5271 if (flags & SV_COW_DROP_PV) {
5272 /* OK, so we don't need to copy our buffer. */
5275 SvGROW(sv, len + 1);
5276 Move(pvx,SvPVX(sv),len,char);
5279 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5286 =for apidoc sv_force_normal_flags
5288 Undo various types of fakery on an SV, where fakery means
5289 "more than" a string: if the PV is a shared string, make
5290 a private copy; if we're a ref, stop refing; if we're a glob, downgrade to
5291 an C<xpvmg>; if we're a copy-on-write scalar, this is the on-write time when
5292 we do the copy, and is also used locally; if this is a
5293 vstring, drop the vstring magic. If C<SV_COW_DROP_PV> is set
5294 then a copy-on-write scalar drops its PV buffer (if any) and becomes
5295 C<SvPOK_off> rather than making a copy. (Used where this
5296 scalar is about to be set to some other value.) In addition,
5297 the C<flags> parameter gets passed to C<sv_unref_flags()>
5298 when unreffing. C<sv_force_normal> calls this function
5299 with flags set to 0.
5301 This function is expected to be used to signal to perl that this SV is
5302 about to be written to, and any extra book-keeping needs to be taken care
5303 of. Hence, it croaks on read-only values.
5305 =for apidoc Amnh||SV_COW_DROP_PV
5311 Perl_sv_force_normal_flags(pTHX_ SV *const sv, const U32 flags)
5313 PERL_ARGS_ASSERT_SV_FORCE_NORMAL_FLAGS;
5316 Perl_croak_no_modify();
5317 else if (SvIsCOW(sv) && LIKELY(SvTYPE(sv) != SVt_PVHV))
5318 S_sv_uncow(aTHX_ sv, flags);
5320 sv_unref_flags(sv, flags);
5321 else if (SvFAKE(sv) && isGV_with_GP(sv))
5322 sv_unglob(sv, flags);
5323 else if (SvFAKE(sv) && isREGEXP(sv)) {
5324 /* Need to downgrade the REGEXP to a simple(r) scalar. This is analogous
5325 to sv_unglob. We only need it here, so inline it. */
5326 const bool islv = SvTYPE(sv) == SVt_PVLV;
5327 const svtype new_type =
5328 islv ? SVt_NULL : SvMAGIC(sv) || SvSTASH(sv) ? SVt_PVMG : SVt_PV;
5329 SV *const temp = newSV_type(new_type);
5330 regexp *old_rx_body;
5332 if (new_type == SVt_PVMG) {
5333 SvMAGIC_set(temp, SvMAGIC(sv));
5334 SvMAGIC_set(sv, NULL);
5335 SvSTASH_set(temp, SvSTASH(sv));
5336 SvSTASH_set(sv, NULL);
5339 SvCUR_set(temp, SvCUR(sv));
5340 /* Remember that SvPVX is in the head, not the body. */
5341 assert(ReANY((REGEXP *)sv)->mother_re);
5344 /* LV-as-regex has sv->sv_any pointing to an XPVLV body,
5345 * whose xpvlenu_rx field points to the regex body */
5346 XPV *xpv = (XPV*)(SvANY(sv));
5347 old_rx_body = xpv->xpv_len_u.xpvlenu_rx;
5348 xpv->xpv_len_u.xpvlenu_rx = NULL;
5351 old_rx_body = ReANY((REGEXP *)sv);
5353 /* Their buffer is already owned by someone else. */
5354 if (flags & SV_COW_DROP_PV) {
5355 /* SvLEN is already 0. For SVt_REGEXP, we have a brand new
5356 zeroed body. For SVt_PVLV, we zeroed it above (len field
5357 a union with xpvlenu_rx) */
5358 assert(!SvLEN(islv ? sv : temp));
5359 sv->sv_u.svu_pv = 0;
5362 sv->sv_u.svu_pv = savepvn(RX_WRAPPED((REGEXP *)sv), SvCUR(sv));
5363 SvLEN_set(islv ? sv : temp, SvCUR(sv)+1);
5367 /* Now swap the rest of the bodies. */
5371 SvFLAGS(sv) &= ~SVTYPEMASK;
5372 SvFLAGS(sv) |= new_type;
5373 SvANY(sv) = SvANY(temp);
5376 SvFLAGS(temp) &= ~(SVTYPEMASK);
5377 SvFLAGS(temp) |= SVt_REGEXP|SVf_FAKE;
5378 SvANY(temp) = old_rx_body;
5380 SvREFCNT_dec_NN(temp);
5382 else if (SvVOK(sv)) sv_unmagic(sv, PERL_MAGIC_vstring);
5388 Efficient removal of characters from the beginning of the string buffer.
5389 C<SvPOK(sv)>, or at least C<SvPOKp(sv)>, must be true and C<ptr> must be a
5390 pointer to somewhere inside the string buffer. C<ptr> becomes the first
5391 character of the adjusted string. Uses the C<OOK> hack. On return, only
5392 C<SvPOK(sv)> and C<SvPOKp(sv)> among the C<OK> flags will be true.
5394 Beware: after this function returns, C<ptr> and SvPVX_const(sv) may no longer
5395 refer to the same chunk of data.
5397 The unfortunate similarity of this function's name to that of Perl's C<chop>
5398 operator is strictly coincidental. This function works from the left;
5399 C<chop> works from the right.
5405 Perl_sv_chop(pTHX_ SV *const sv, const char *const ptr)
5416 PERL_ARGS_ASSERT_SV_CHOP;
5418 if (!ptr || !SvPOKp(sv))
5420 delta = ptr - SvPVX_const(sv);
5422 /* Nothing to do. */
5425 max_delta = SvLEN(sv) ? SvLEN(sv) : SvCUR(sv);
5426 if (delta > max_delta)
5427 Perl_croak(aTHX_ "panic: sv_chop ptr=%p, start=%p, end=%p",
5428 ptr, SvPVX_const(sv), SvPVX_const(sv) + max_delta);
5429 /* SvPVX(sv) may move in SV_CHECK_THINKFIRST(sv), so don't use ptr any more */
5430 SV_CHECK_THINKFIRST(sv);
5431 SvPOK_only_UTF8(sv);
5434 if (!SvLEN(sv)) { /* make copy of shared string */
5435 const char *pvx = SvPVX_const(sv);
5436 const STRLEN len = SvCUR(sv);
5437 SvGROW(sv, len + 1);
5438 Move(pvx,SvPVX(sv),len,char);
5444 SvOOK_offset(sv, old_delta);
5446 SvLEN_set(sv, SvLEN(sv) - delta);
5447 SvCUR_set(sv, SvCUR(sv) - delta);
5448 SvPV_set(sv, SvPVX(sv) + delta);
5450 p = (U8 *)SvPVX_const(sv);
5453 /* how many bytes were evacuated? we will fill them with sentinel
5454 bytes, except for the part holding the new offset of course. */
5457 evacn += (old_delta < 0x100 ? 1 : 1 + sizeof(STRLEN));
5459 assert(evacn <= delta + old_delta);
5463 /* This sets 'delta' to the accumulated value of all deltas so far */
5467 /* If 'delta' fits in a byte, store it just prior to the new beginning of
5468 * the string; otherwise store a 0 byte there and store 'delta' just prior
5469 * to that, using as many bytes as a STRLEN occupies. Thus it overwrites a
5470 * portion of the chopped part of the string */
5471 if (delta < 0x100) {
5475 p -= sizeof(STRLEN);
5476 Copy((U8*)&delta, p, sizeof(STRLEN), U8);
5480 /* Fill the preceding buffer with sentinals to verify that no-one is
5490 =for apidoc sv_catpvn
5492 Concatenates the string onto the end of the string which is in the SV.
5493 C<len> indicates number of bytes to copy. If the SV has the UTF-8
5494 status set, then the bytes appended should be valid UTF-8.
5495 Handles 'get' magic, but not 'set' magic. See C<L</sv_catpvn_mg>>.
5497 =for apidoc sv_catpvn_flags
5499 Concatenates the string onto the end of the string which is in the SV. The
5500 C<len> indicates number of bytes to copy.
5502 By default, the string appended is assumed to be valid UTF-8 if the SV has
5503 the UTF-8 status set, and a string of bytes otherwise. One can force the
5504 appended string to be interpreted as UTF-8 by supplying the C<SV_CATUTF8>
5505 flag, and as bytes by supplying the C<SV_CATBYTES> flag; the SV or the
5506 string appended will be upgraded to UTF-8 if necessary.
5508 If C<flags> has the C<SV_SMAGIC> bit set, will
5509 C<mg_set> on C<dsv> afterwards if appropriate.
5510 C<sv_catpvn> and C<sv_catpvn_nomg> are implemented
5511 in terms of this function.
5513 =for apidoc Amnh||SV_CATUTF8
5514 =for apidoc Amnh||SV_CATBYTES
5515 =for apidoc Amnh||SV_SMAGIC
5521 Perl_sv_catpvn_flags(pTHX_ SV *const dsv, const char *sstr, const STRLEN slen, const I32 flags)
5524 const char * const dstr = SvPV_force_flags(dsv, dlen, flags);
5526 PERL_ARGS_ASSERT_SV_CATPVN_FLAGS;
5527 assert((flags & (SV_CATBYTES|SV_CATUTF8)) != (SV_CATBYTES|SV_CATUTF8));
5529 if (!(flags & SV_CATBYTES) || !SvUTF8(dsv)) {
5530 if (flags & SV_CATUTF8 && !SvUTF8(dsv)) {
5531 sv_utf8_upgrade_flags_grow(dsv, 0, slen + 1);
5534 else SvGROW(dsv, dlen + slen + 3);
5536 sstr = SvPVX_const(dsv);
5537 Move(sstr, SvPVX(dsv) + dlen, slen, char);
5538 SvCUR_set(dsv, SvCUR(dsv) + slen);
5541 /* We inline bytes_to_utf8, to avoid an extra malloc. */
5542 const char * const send = sstr + slen;
5545 /* Something this code does not account for, which I think is
5546 impossible; it would require the same pv to be treated as
5547 bytes *and* utf8, which would indicate a bug elsewhere. */
5548 assert(sstr != dstr);
5550 SvGROW(dsv, dlen + slen * 2 + 3);
5551 d = (U8 *)SvPVX(dsv) + dlen;
5553 while (sstr < send) {
5554 append_utf8_from_native_byte(*sstr, &d);
5557 SvCUR_set(dsv, d-(const U8 *)SvPVX(dsv));
5560 (void)SvPOK_only_UTF8(dsv); /* validate pointer */
5562 if (flags & SV_SMAGIC)
5567 =for apidoc sv_catsv
5569 Concatenates the string from SV C<ssv> onto the end of the string in SV
5570 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5571 Handles 'get' magic on both SVs, but no 'set' magic. See C<L</sv_catsv_mg>>
5572 and C<L</sv_catsv_nomg>>.
5574 =for apidoc sv_catsv_flags
5576 Concatenates the string from SV C<ssv> onto the end of the string in SV
5577 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5578 If C<flags> has the C<SV_GMAGIC> bit set, will call C<mg_get> on both SVs if
5579 appropriate. If C<flags> has the C<SV_SMAGIC> bit set, C<mg_set> will be called on
5580 the modified SV afterward, if appropriate. C<sv_catsv>, C<sv_catsv_nomg>,
5581 and C<sv_catsv_mg> are implemented in terms of this function.
5586 Perl_sv_catsv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
5588 PERL_ARGS_ASSERT_SV_CATSV_FLAGS;
5592 const char *spv = SvPV_flags_const(ssv, slen, flags);
5593 if (flags & SV_GMAGIC)
5595 sv_catpvn_flags(dsv, spv, slen,
5596 DO_UTF8(ssv) ? SV_CATUTF8 : SV_CATBYTES);
5597 if (flags & SV_SMAGIC)
5603 =for apidoc sv_catpv
5605 Concatenates the C<NUL>-terminated string onto the end of the string which is
5607 If the SV has the UTF-8 status set, then the bytes appended should be
5608 valid UTF-8. Handles 'get' magic, but not 'set' magic. See
5614 Perl_sv_catpv(pTHX_ SV *const sv, const char *ptr)
5620 PERL_ARGS_ASSERT_SV_CATPV;
5624 junk = SvPV_force(sv, tlen);
5626 SvGROW(sv, tlen + len + 1);
5628 ptr = SvPVX_const(sv);
5629 Move(ptr,SvPVX(sv)+tlen,len+1,char);
5630 SvCUR_set(sv, SvCUR(sv) + len);
5631 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5636 =for apidoc sv_catpv_flags
5638 Concatenates the C<NUL>-terminated string onto the end of the string which is
5640 If the SV has the UTF-8 status set, then the bytes appended should
5641 be valid UTF-8. If C<flags> has the C<SV_SMAGIC> bit set, will C<mg_set>
5642 on the modified SV if appropriate.
5648 Perl_sv_catpv_flags(pTHX_ SV *dstr, const char *sstr, const I32 flags)
5650 PERL_ARGS_ASSERT_SV_CATPV_FLAGS;
5651 sv_catpvn_flags(dstr, sstr, strlen(sstr), flags);
5655 =for apidoc sv_catpv_mg
5657 Like C<sv_catpv>, but also handles 'set' magic.
5663 Perl_sv_catpv_mg(pTHX_ SV *const sv, const char *const ptr)
5665 PERL_ARGS_ASSERT_SV_CATPV_MG;
5674 Creates a new SV. A non-zero C<len> parameter indicates the number of
5675 bytes of preallocated string space the SV should have. An extra byte for a
5676 trailing C<NUL> is also reserved. (C<SvPOK> is not set for the SV even if string
5677 space is allocated.) The reference count for the new SV is set to 1.
5679 In 5.9.3, C<newSV()> replaces the older C<NEWSV()> API, and drops the first
5680 parameter, I<x>, a debug aid which allowed callers to identify themselves.
5681 This aid has been superseded by a new build option, C<PERL_MEM_LOG> (see
5682 L<perlhacktips/PERL_MEM_LOG>). The older API is still there for use in XS
5683 modules supporting older perls.
5689 Perl_newSV(pTHX_ const STRLEN len)
5695 sv_grow(sv, len + 1);
5700 =for apidoc sv_magicext
5702 Adds magic to an SV, upgrading it if necessary. Applies the
5703 supplied C<vtable> and returns a pointer to the magic added.
5705 Note that C<sv_magicext> will allow things that C<sv_magic> will not.
5706 In particular, you can add magic to C<SvREADONLY> SVs, and add more than
5707 one instance of the same C<how>.
5709 If C<namlen> is greater than zero then a C<savepvn> I<copy> of C<name> is
5710 stored, if C<namlen> is zero then C<name> is stored as-is and - as another
5711 special case - if C<(name && namlen == HEf_SVKEY)> then C<name> is assumed
5712 to contain an SV* and is stored as-is with its C<REFCNT> incremented.
5714 (This is now used as a subroutine by C<sv_magic>.)
5719 Perl_sv_magicext(pTHX_ SV *const sv, SV *const obj, const int how,
5720 const MGVTBL *const vtable, const char *const name, const I32 namlen)
5724 PERL_ARGS_ASSERT_SV_MAGICEXT;
5726 SvUPGRADE(sv, SVt_PVMG);
5727 Newxz(mg, 1, MAGIC);
5728 mg->mg_moremagic = SvMAGIC(sv);
5729 SvMAGIC_set(sv, mg);
5731 /* Sometimes a magic contains a reference loop, where the sv and
5732 object refer to each other. To prevent a reference loop that
5733 would prevent such objects being freed, we look for such loops
5734 and if we find one we avoid incrementing the object refcount.
5736 Note we cannot do this to avoid self-tie loops as intervening RV must
5737 have its REFCNT incremented to keep it in existence.
5740 if (!obj || obj == sv ||
5741 how == PERL_MAGIC_arylen ||
5742 how == PERL_MAGIC_regdata ||
5743 how == PERL_MAGIC_regdatum ||
5744 how == PERL_MAGIC_symtab ||
5745 (SvTYPE(obj) == SVt_PVGV &&
5746 (GvSV(obj) == sv || GvHV(obj) == (const HV *)sv
5747 || GvAV(obj) == (const AV *)sv || GvCV(obj) == (const CV *)sv
5748 || GvIOp(obj) == (const IO *)sv || GvFORM(obj) == (const CV *)sv)))
5753 mg->mg_obj = SvREFCNT_inc_simple(obj);
5754 mg->mg_flags |= MGf_REFCOUNTED;
5757 /* Normal self-ties simply pass a null object, and instead of
5758 using mg_obj directly, use the SvTIED_obj macro to produce a
5759 new RV as needed. For glob "self-ties", we are tieing the PVIO
5760 with an RV obj pointing to the glob containing the PVIO. In
5761 this case, to avoid a reference loop, we need to weaken the
5765 if (how == PERL_MAGIC_tiedscalar && SvTYPE(sv) == SVt_PVIO &&
5766 obj && SvROK(obj) && GvIO(SvRV(obj)) == (const IO *)sv)
5772 mg->mg_len = namlen;
5775 mg->mg_ptr = savepvn(name, namlen);
5776 else if (namlen == HEf_SVKEY) {
5777 /* Yes, this is casting away const. This is only for the case of
5778 HEf_SVKEY. I think we need to document this aberation of the
5779 constness of the API, rather than making name non-const, as
5780 that change propagating outwards a long way. */
5781 mg->mg_ptr = (char*)SvREFCNT_inc_simple_NN((SV *)name);
5783 mg->mg_ptr = (char *) name;
5785 mg->mg_virtual = (MGVTBL *) vtable;
5792 Perl_sv_magicext_mglob(pTHX_ SV *sv)
5794 PERL_ARGS_ASSERT_SV_MAGICEXT_MGLOB;
5795 if (SvTYPE(sv) == SVt_PVLV && LvTYPE(sv) == 'y') {
5796 /* This sv is only a delegate. //g magic must be attached to
5801 return sv_magicext(sv, NULL, PERL_MAGIC_regex_global,
5802 &PL_vtbl_mglob, 0, 0);
5806 =for apidoc sv_magic
5808 Adds magic to an SV. First upgrades C<sv> to type C<SVt_PVMG> if
5809 necessary, then adds a new magic item of type C<how> to the head of the
5812 See C<L</sv_magicext>> (which C<sv_magic> now calls) for a description of the
5813 handling of the C<name> and C<namlen> arguments.
5815 You need to use C<sv_magicext> to add magic to C<SvREADONLY> SVs and also
5816 to add more than one instance of the same C<how>.
5822 Perl_sv_magic(pTHX_ SV *const sv, SV *const obj, const int how,
5823 const char *const name, const I32 namlen)
5825 const MGVTBL *vtable;
5828 unsigned int vtable_index;
5830 PERL_ARGS_ASSERT_SV_MAGIC;
5832 if (how < 0 || (unsigned)how >= C_ARRAY_LENGTH(PL_magic_data)
5833 || ((flags = PL_magic_data[how]),
5834 (vtable_index = flags & PERL_MAGIC_VTABLE_MASK)
5835 > magic_vtable_max))
5836 Perl_croak(aTHX_ "Don't know how to handle magic of type \\%o", how);
5838 /* PERL_MAGIC_ext is reserved for use by extensions not perl internals.
5839 Useful for attaching extension internal data to perl vars.
5840 Note that multiple extensions may clash if magical scalars
5841 etc holding private data from one are passed to another. */
5843 vtable = (vtable_index == magic_vtable_max)
5844 ? NULL : PL_magic_vtables + vtable_index;
5846 if (SvREADONLY(sv)) {
5848 !PERL_MAGIC_TYPE_READONLY_ACCEPTABLE(how)
5851 Perl_croak_no_modify();
5854 if (SvMAGICAL(sv) || (how == PERL_MAGIC_taint && SvTYPE(sv) >= SVt_PVMG)) {
5855 if (SvMAGIC(sv) && (mg = mg_find(sv, how))) {
5856 /* sv_magic() refuses to add a magic of the same 'how' as an
5859 if (how == PERL_MAGIC_taint)
5865 /* Force pos to be stored as characters, not bytes. */
5866 if (SvMAGICAL(sv) && DO_UTF8(sv)
5867 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
5869 && mg->mg_flags & MGf_BYTES) {
5870 mg->mg_len = (SSize_t)sv_pos_b2u_flags(sv, (STRLEN)mg->mg_len,
5872 mg->mg_flags &= ~MGf_BYTES;
5875 /* Rest of work is done else where */
5876 mg = sv_magicext(sv,obj,how,vtable,name,namlen);
5879 case PERL_MAGIC_taint:
5882 case PERL_MAGIC_ext:
5883 case PERL_MAGIC_dbfile:
5890 S_sv_unmagicext_flags(pTHX_ SV *const sv, const int type, MGVTBL *vtbl, const U32 flags)
5897 if (SvTYPE(sv) < SVt_PVMG || !SvMAGIC(sv))
5899 mgp = &(((XPVMG*) SvANY(sv))->xmg_u.xmg_magic);
5900 for (mg = *mgp; mg; mg = *mgp) {
5901 const MGVTBL* const virt = mg->mg_virtual;
5902 if (mg->mg_type == type && (!flags || virt == vtbl)) {
5903 *mgp = mg->mg_moremagic;
5904 if (virt && virt->svt_free)
5905 virt->svt_free(aTHX_ sv, mg);
5906 if (mg->mg_ptr && mg->mg_type != PERL_MAGIC_regex_global) {
5908 Safefree(mg->mg_ptr);
5909 else if (mg->mg_len == HEf_SVKEY)
5910 SvREFCNT_dec(MUTABLE_SV(mg->mg_ptr));
5911 else if (mg->mg_type == PERL_MAGIC_utf8)
5912 Safefree(mg->mg_ptr);
5914 if (mg->mg_flags & MGf_REFCOUNTED)
5915 SvREFCNT_dec(mg->mg_obj);
5919 mgp = &mg->mg_moremagic;
5922 if (SvMAGICAL(sv)) /* if we're under save_magic, wait for restore_magic; */
5923 mg_magical(sv); /* else fix the flags now */
5932 =for apidoc sv_unmagic
5934 Removes all magic of type C<type> from an SV.
5940 Perl_sv_unmagic(pTHX_ SV *const sv, const int type)
5942 PERL_ARGS_ASSERT_SV_UNMAGIC;
5943 return S_sv_unmagicext_flags(aTHX_ sv, type, NULL, 0);
5947 =for apidoc sv_unmagicext
5949 Removes all magic of type C<type> with the specified C<vtbl> from an SV.
5955 Perl_sv_unmagicext(pTHX_ SV *const sv, const int type, MGVTBL *vtbl)
5957 PERL_ARGS_ASSERT_SV_UNMAGICEXT;
5958 return S_sv_unmagicext_flags(aTHX_ sv, type, vtbl, 1);
5962 =for apidoc sv_rvweaken
5964 Weaken a reference: set the C<SvWEAKREF> flag on this RV; give the
5965 referred-to SV C<PERL_MAGIC_backref> magic if it hasn't already; and
5966 push a back-reference to this RV onto the array of backreferences
5967 associated with that magic. If the RV is magical, set magic will be
5968 called after the RV is cleared. Silently ignores C<undef> and warns
5969 on already-weak references.
5975 Perl_sv_rvweaken(pTHX_ SV *const sv)
5979 PERL_ARGS_ASSERT_SV_RVWEAKEN;
5981 if (!SvOK(sv)) /* let undefs pass */
5984 Perl_croak(aTHX_ "Can't weaken a nonreference");
5985 else if (SvWEAKREF(sv)) {
5986 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is already weak");
5989 else if (SvREADONLY(sv)) croak_no_modify();
5991 Perl_sv_add_backref(aTHX_ tsv, sv);
5993 SvREFCNT_dec_NN(tsv);
5998 =for apidoc sv_rvunweaken
6000 Unweaken a reference: Clear the C<SvWEAKREF> flag on this RV; remove
6001 the backreference to this RV from the array of backreferences
6002 associated with the target SV, increment the refcount of the target.
6003 Silently ignores C<undef> and warns on non-weak references.
6009 Perl_sv_rvunweaken(pTHX_ SV *const sv)
6013 PERL_ARGS_ASSERT_SV_RVUNWEAKEN;
6015 if (!SvOK(sv)) /* let undefs pass */
6018 Perl_croak(aTHX_ "Can't unweaken a nonreference");
6019 else if (!SvWEAKREF(sv)) {
6020 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is not weak");
6023 else if (SvREADONLY(sv)) croak_no_modify();
6028 SvREFCNT_inc_NN(tsv);
6029 Perl_sv_del_backref(aTHX_ tsv, sv);
6034 =for apidoc sv_get_backrefs
6036 If C<sv> is the target of a weak reference then it returns the back
6037 references structure associated with the sv; otherwise return C<NULL>.
6039 When returning a non-null result the type of the return is relevant. If it
6040 is an AV then the elements of the AV are the weak reference RVs which
6041 point at this item. If it is any other type then the item itself is the
6044 See also C<Perl_sv_add_backref()>, C<Perl_sv_del_backref()>,
6045 C<Perl_sv_kill_backrefs()>
6051 Perl_sv_get_backrefs(SV *const sv)
6055 PERL_ARGS_ASSERT_SV_GET_BACKREFS;
6057 /* find slot to store array or singleton backref */
6059 if (SvTYPE(sv) == SVt_PVHV) {
6061 struct xpvhv_aux * const iter = HvAUX((HV *)sv);
6062 backrefs = (SV *)iter->xhv_backreferences;
6064 } else if (SvMAGICAL(sv)) {
6065 MAGIC *mg = mg_find(sv, PERL_MAGIC_backref);
6067 backrefs = mg->mg_obj;
6072 /* Give tsv backref magic if it hasn't already got it, then push a
6073 * back-reference to sv onto the array associated with the backref magic.
6075 * As an optimisation, if there's only one backref and it's not an AV,
6076 * store it directly in the HvAUX or mg_obj slot, avoiding the need to
6077 * allocate an AV. (Whether the slot holds an AV tells us whether this is
6081 /* A discussion about the backreferences array and its refcount:
6083 * The AV holding the backreferences is pointed to either as the mg_obj of
6084 * PERL_MAGIC_backref, or in the specific case of a HV, from the
6085 * xhv_backreferences field. The array is created with a refcount
6086 * of 2. This means that if during global destruction the array gets
6087 * picked on before its parent to have its refcount decremented by the
6088 * random zapper, it won't actually be freed, meaning it's still there for
6089 * when its parent gets freed.
6091 * When the parent SV is freed, the extra ref is killed by
6092 * Perl_sv_kill_backrefs. The other ref is killed, in the case of magic,
6093 * by mg_free() / MGf_REFCOUNTED, or for a hash, by Perl_hv_kill_backrefs.
6095 * When a single backref SV is stored directly, it is not reference
6100 Perl_sv_add_backref(pTHX_ SV *const tsv, SV *const sv)
6106 PERL_ARGS_ASSERT_SV_ADD_BACKREF;
6108 /* find slot to store array or singleton backref */
6110 if (SvTYPE(tsv) == SVt_PVHV) {
6111 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
6114 mg = mg_find(tsv, PERL_MAGIC_backref);
6116 mg = sv_magicext(tsv, NULL, PERL_MAGIC_backref, &PL_vtbl_backref, NULL, 0);
6117 svp = &(mg->mg_obj);
6120 /* create or retrieve the array */
6122 if ( (!*svp && SvTYPE(sv) == SVt_PVAV)
6123 || (*svp && SvTYPE(*svp) != SVt_PVAV)
6127 mg->mg_flags |= MGf_REFCOUNTED;
6130 SvREFCNT_inc_simple_void_NN(av);
6131 /* av now has a refcnt of 2; see discussion above */
6132 av_extend(av, *svp ? 2 : 1);
6134 /* move single existing backref to the array */
6135 AvARRAY(av)[++AvFILLp(av)] = *svp; /* av_push() */
6140 av = MUTABLE_AV(*svp);
6142 /* optimisation: store single backref directly in HvAUX or mg_obj */
6146 assert(SvTYPE(av) == SVt_PVAV);
6147 if (AvFILLp(av) >= AvMAX(av)) {
6148 av_extend(av, AvFILLp(av)+1);
6151 /* push new backref */
6152 AvARRAY(av)[++AvFILLp(av)] = sv; /* av_push() */
6155 /* delete a back-reference to ourselves from the backref magic associated
6156 * with the SV we point to.
6160 Perl_sv_del_backref(pTHX_ SV *const tsv, SV *const sv)
6164 PERL_ARGS_ASSERT_SV_DEL_BACKREF;
6166 if (SvTYPE(tsv) == SVt_PVHV) {
6168 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
6170 else if (SvIS_FREED(tsv) && PL_phase == PERL_PHASE_DESTRUCT) {
6171 /* It's possible for the the last (strong) reference to tsv to have
6172 become freed *before* the last thing holding a weak reference.
6173 If both survive longer than the backreferences array, then when
6174 the referent's reference count drops to 0 and it is freed, it's
6175 not able to chase the backreferences, so they aren't NULLed.
6177 For example, a CV holds a weak reference to its stash. If both the
6178 CV and the stash survive longer than the backreferences array,
6179 and the CV gets picked for the SvBREAK() treatment first,
6180 *and* it turns out that the stash is only being kept alive because
6181 of an our variable in the pad of the CV, then midway during CV
6182 destruction the stash gets freed, but CvSTASH() isn't set to NULL.
6183 It ends up pointing to the freed HV. Hence it's chased in here, and
6184 if this block wasn't here, it would hit the !svp panic just below.
6186 I don't believe that "better" destruction ordering is going to help
6187 here - during global destruction there's always going to be the
6188 chance that something goes out of order. We've tried to make it
6189 foolproof before, and it only resulted in evolutionary pressure on
6190 fools. Which made us look foolish for our hubris. :-(
6196 = SvMAGICAL(tsv) ? mg_find(tsv, PERL_MAGIC_backref) : NULL;
6197 svp = mg ? &(mg->mg_obj) : NULL;
6201 Perl_croak(aTHX_ "panic: del_backref, svp=0");
6203 /* It's possible that sv is being freed recursively part way through the
6204 freeing of tsv. If this happens, the backreferences array of tsv has
6205 already been freed, and so svp will be NULL. If this is the case,
6206 we should not panic. Instead, nothing needs doing, so return. */
6207 if (PL_phase == PERL_PHASE_DESTRUCT && SvREFCNT(tsv) == 0)
6209 Perl_croak(aTHX_ "panic: del_backref, *svp=%p phase=%s refcnt=%" UVuf,
6210 (void*)*svp, PL_phase_names[PL_phase], (UV)SvREFCNT(tsv));
6213 if (SvTYPE(*svp) == SVt_PVAV) {
6217 AV * const av = (AV*)*svp;
6219 assert(!SvIS_FREED(av));
6223 /* for an SV with N weak references to it, if all those
6224 * weak refs are deleted, then sv_del_backref will be called
6225 * N times and O(N^2) compares will be done within the backref
6226 * array. To ameliorate this potential slowness, we:
6227 * 1) make sure this code is as tight as possible;
6228 * 2) when looking for SV, look for it at both the head and tail of the
6229 * array first before searching the rest, since some create/destroy
6230 * patterns will cause the backrefs to be freed in order.
6237 SV **p = &svp[fill];
6238 SV *const topsv = *p;
6245 /* We weren't the last entry.
6246 An unordered list has this property that you
6247 can take the last element off the end to fill
6248 the hole, and it's still an unordered list :-)
6254 break; /* should only be one */
6261 AvFILLp(av) = fill-1;
6263 else if (SvIS_FREED(*svp) && PL_phase == PERL_PHASE_DESTRUCT) {
6264 /* freed AV; skip */
6267 /* optimisation: only a single backref, stored directly */
6269 Perl_croak(aTHX_ "panic: del_backref, *svp=%p, sv=%p",
6270 (void*)*svp, (void*)sv);
6277 Perl_sv_kill_backrefs(pTHX_ SV *const sv, AV *const av)
6283 PERL_ARGS_ASSERT_SV_KILL_BACKREFS;
6288 /* after multiple passes through Perl_sv_clean_all() for a thingy
6289 * that has badly leaked, the backref array may have gotten freed,
6290 * since we only protect it against 1 round of cleanup */
6291 if (SvIS_FREED(av)) {
6292 if (PL_in_clean_all) /* All is fair */
6295 "panic: magic_killbackrefs (freed backref AV/SV)");
6299 is_array = (SvTYPE(av) == SVt_PVAV);
6301 assert(!SvIS_FREED(av));
6304 last = svp + AvFILLp(av);
6307 /* optimisation: only a single backref, stored directly */
6313 while (svp <= last) {
6315 SV *const referrer = *svp;
6316 if (SvWEAKREF(referrer)) {
6317 /* XXX Should we check that it hasn't changed? */
6318 assert(SvROK(referrer));
6319 SvRV_set(referrer, 0);
6321 SvWEAKREF_off(referrer);
6322 SvSETMAGIC(referrer);
6323 } else if (SvTYPE(referrer) == SVt_PVGV ||
6324 SvTYPE(referrer) == SVt_PVLV) {
6325 assert(SvTYPE(sv) == SVt_PVHV); /* stash backref */
6326 /* You lookin' at me? */
6327 assert(GvSTASH(referrer));
6328 assert(GvSTASH(referrer) == (const HV *)sv);
6329 GvSTASH(referrer) = 0;
6330 } else if (SvTYPE(referrer) == SVt_PVCV ||
6331 SvTYPE(referrer) == SVt_PVFM) {
6332 if (SvTYPE(sv) == SVt_PVHV) { /* stash backref */
6333 /* You lookin' at me? */
6334 assert(CvSTASH(referrer));
6335 assert(CvSTASH(referrer) == (const HV *)sv);
6336 SvANY(MUTABLE_CV(referrer))->xcv_stash = 0;
6339 assert(SvTYPE(sv) == SVt_PVGV);
6340 /* You lookin' at me? */
6341 assert(CvGV(referrer));
6342 assert(CvGV(referrer) == (const GV *)sv);
6343 anonymise_cv_maybe(MUTABLE_GV(sv),
6344 MUTABLE_CV(referrer));
6349 "panic: magic_killbackrefs (flags=%" UVxf ")",
6350 (UV)SvFLAGS(referrer));
6361 SvREFCNT_dec_NN(av); /* remove extra count added by sv_add_backref() */
6367 =for apidoc sv_insert
6369 Inserts and/or replaces a string at the specified offset/length within the SV.
6370 Similar to the Perl C<substr()> function, with C<littlelen> bytes starting at
6371 C<little> replacing C<len> bytes of the string in C<bigstr> starting at
6372 C<offset>. Handles get magic.
6374 =for apidoc sv_insert_flags
6376 Same as C<sv_insert>, but the extra C<flags> are passed to the
6377 C<SvPV_force_flags> that applies to C<bigstr>.
6383 Perl_sv_insert_flags(pTHX_ SV *const bigstr, const STRLEN offset, const STRLEN len, const char *little, const STRLEN littlelen, const U32 flags)
6389 SSize_t i; /* better be sizeof(STRLEN) or bad things happen */
6392 PERL_ARGS_ASSERT_SV_INSERT_FLAGS;
6394 SvPV_force_flags(bigstr, curlen, flags);
6395 (void)SvPOK_only_UTF8(bigstr);
6397 if (little >= SvPVX(bigstr) &&
6398 little < SvPVX(bigstr) + (SvLEN(bigstr) ? SvLEN(bigstr) : SvCUR(bigstr))) {
6399 /* little is a pointer to within bigstr, since we can reallocate bigstr,
6400 or little...little+littlelen might overlap offset...offset+len we make a copy
6402 little = savepvn(little, littlelen);
6406 if (offset + len > curlen) {
6407 SvGROW(bigstr, offset+len+1);
6408 Zero(SvPVX(bigstr)+curlen, offset+len-curlen, char);
6409 SvCUR_set(bigstr, offset+len);
6413 i = littlelen - len;
6414 if (i > 0) { /* string might grow */
6415 big = SvGROW(bigstr, SvCUR(bigstr) + i + 1);
6416 mid = big + offset + len;
6417 midend = bigend = big + SvCUR(bigstr);
6420 while (midend > mid) /* shove everything down */
6421 *--bigend = *--midend;
6422 Move(little,big+offset,littlelen,char);
6423 SvCUR_set(bigstr, SvCUR(bigstr) + i);
6428 Move(little,SvPVX(bigstr)+offset,len,char);
6433 big = SvPVX(bigstr);
6436 bigend = big + SvCUR(bigstr);
6438 if (midend > bigend)
6439 Perl_croak(aTHX_ "panic: sv_insert, midend=%p, bigend=%p",
6442 if (mid - big > bigend - midend) { /* faster to shorten from end */
6444 Move(little, mid, littlelen,char);
6447 i = bigend - midend;
6449 Move(midend, mid, i,char);
6453 SvCUR_set(bigstr, mid - big);
6455 else if ((i = mid - big)) { /* faster from front */
6456 midend -= littlelen;
6458 Move(big, midend - i, i, char);
6459 sv_chop(bigstr,midend-i);
6461 Move(little, mid, littlelen,char);
6463 else if (littlelen) {
6464 midend -= littlelen;
6465 sv_chop(bigstr,midend);
6466 Move(little,midend,littlelen,char);
6469 sv_chop(bigstr,midend);
6475 =for apidoc sv_replace
6477 Make the first argument a copy of the second, then delete the original.
6478 The target SV physically takes over ownership of the body of the source SV
6479 and inherits its flags; however, the target keeps any magic it owns,
6480 and any magic in the source is discarded.
6481 Note that this is a rather specialist SV copying operation; most of the
6482 time you'll want to use C<sv_setsv> or one of its many macro front-ends.
6488 Perl_sv_replace(pTHX_ SV *const sv, SV *const nsv)
6490 const U32 refcnt = SvREFCNT(sv);
6492 PERL_ARGS_ASSERT_SV_REPLACE;
6494 SV_CHECK_THINKFIRST_COW_DROP(sv);
6495 if (SvREFCNT(nsv) != 1) {
6496 Perl_croak(aTHX_ "panic: reference miscount on nsv in sv_replace()"
6497 " (%" UVuf " != 1)", (UV) SvREFCNT(nsv));
6499 if (SvMAGICAL(sv)) {
6503 sv_upgrade(nsv, SVt_PVMG);
6504 SvMAGIC_set(nsv, SvMAGIC(sv));
6505 SvFLAGS(nsv) |= SvMAGICAL(sv);
6507 SvMAGIC_set(sv, NULL);
6511 assert(!SvREFCNT(sv));
6512 #ifdef DEBUG_LEAKING_SCALARS
6513 sv->sv_flags = nsv->sv_flags;
6514 sv->sv_any = nsv->sv_any;
6515 sv->sv_refcnt = nsv->sv_refcnt;
6516 sv->sv_u = nsv->sv_u;
6518 StructCopy(nsv,sv,SV);
6520 if(SvTYPE(sv) == SVt_IV) {
6521 SET_SVANY_FOR_BODYLESS_IV(sv);
6525 SvREFCNT(sv) = refcnt;
6526 SvFLAGS(nsv) |= SVTYPEMASK; /* Mark as freed */
6531 /* We're about to free a GV which has a CV that refers back to us.
6532 * If that CV will outlive us, make it anonymous (i.e. fix up its CvGV
6536 S_anonymise_cv_maybe(pTHX_ GV *gv, CV* cv)
6541 PERL_ARGS_ASSERT_ANONYMISE_CV_MAYBE;
6544 assert(SvREFCNT(gv) == 0);
6545 assert(isGV(gv) && isGV_with_GP(gv));
6547 assert(!CvANON(cv));
6548 assert(CvGV(cv) == gv);
6549 assert(!CvNAMED(cv));
6551 /* will the CV shortly be freed by gp_free() ? */
6552 if (GvCV(gv) == cv && GvGP(gv)->gp_refcnt < 2 && SvREFCNT(cv) < 2) {
6553 SvANY(cv)->xcv_gv_u.xcv_gv = NULL;
6557 /* if not, anonymise: */
6558 gvname = (GvSTASH(gv) && HvNAME(GvSTASH(gv)) && HvENAME(GvSTASH(gv)))
6559 ? newSVhek(HvENAME_HEK(GvSTASH(gv)))
6560 : newSVpvn_flags( "__ANON__", 8, 0 );
6561 sv_catpvs(gvname, "::__ANON__");
6562 anongv = gv_fetchsv(gvname, GV_ADDMULTI, SVt_PVCV);
6563 SvREFCNT_dec_NN(gvname);
6567 SvANY(cv)->xcv_gv_u.xcv_gv = MUTABLE_GV(SvREFCNT_inc(anongv));
6572 =for apidoc sv_clear
6574 Clear an SV: call any destructors, free up any memory used by the body,
6575 and free the body itself. The SV's head is I<not> freed, although
6576 its type is set to all 1's so that it won't inadvertently be assumed
6577 to be live during global destruction etc.
6578 This function should only be called when C<REFCNT> is zero. Most of the time
6579 you'll want to call C<sv_free()> (or its macro wrapper C<SvREFCNT_dec>)
6586 Perl_sv_clear(pTHX_ SV *const orig_sv)
6591 const struct body_details *sv_type_details;
6595 STRLEN hash_index = 0; /* initialise to make Coverity et al happy.
6596 Not strictly necessary */
6598 PERL_ARGS_ASSERT_SV_CLEAR;
6600 /* within this loop, sv is the SV currently being freed, and
6601 * iter_sv is the most recent AV or whatever that's being iterated
6602 * over to provide more SVs */
6608 assert(SvREFCNT(sv) == 0);
6609 assert(SvTYPE(sv) != (svtype)SVTYPEMASK);
6611 if (type <= SVt_IV) {
6612 /* See the comment in sv.h about the collusion between this
6613 * early return and the overloading of the NULL slots in the
6617 SvFLAGS(sv) &= SVf_BREAK;
6618 SvFLAGS(sv) |= SVTYPEMASK;
6622 /* objs are always >= MG, but pad names use the SVs_OBJECT flag
6623 for another purpose */
6624 assert(!SvOBJECT(sv) || type >= SVt_PVMG);
6626 if (type >= SVt_PVMG) {
6628 if (!curse(sv, 1)) goto get_next_sv;
6629 type = SvTYPE(sv); /* destructor may have changed it */
6631 /* Free back-references before magic, in case the magic calls
6632 * Perl code that has weak references to sv. */
6633 if (type == SVt_PVHV) {
6634 Perl_hv_kill_backrefs(aTHX_ MUTABLE_HV(sv));
6638 else if (SvMAGIC(sv)) {
6639 /* Free back-references before other types of magic. */
6640 sv_unmagic(sv, PERL_MAGIC_backref);
6646 /* case SVt_INVLIST: */
6649 IoIFP(sv) != PerlIO_stdin() &&
6650 IoIFP(sv) != PerlIO_stdout() &&
6651 IoIFP(sv) != PerlIO_stderr() &&
6652 !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6654 io_close(MUTABLE_IO(sv), NULL, FALSE,
6655 (IoTYPE(sv) == IoTYPE_WRONLY ||
6656 IoTYPE(sv) == IoTYPE_RDWR ||
6657 IoTYPE(sv) == IoTYPE_APPEND));
6659 if (IoDIRP(sv) && !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6660 PerlDir_close(IoDIRP(sv));
6661 IoDIRP(sv) = (DIR*)NULL;
6662 Safefree(IoTOP_NAME(sv));
6663 Safefree(IoFMT_NAME(sv));
6664 Safefree(IoBOTTOM_NAME(sv));
6665 if ((const GV *)sv == PL_statgv)
6669 /* FIXME for plugins */
6670 pregfree2((REGEXP*) sv);
6674 cv_undef(MUTABLE_CV(sv));
6675 /* If we're in a stash, we don't own a reference to it.
6676 * However it does have a back reference to us, which needs to
6678 if ((stash = CvSTASH(sv)))
6679 sv_del_backref(MUTABLE_SV(stash), sv);
6682 if (HvTOTALKEYS((HV*)sv) > 0) {
6684 /* this statement should match the one at the beginning of
6685 * hv_undef_flags() */
6686 if ( PL_phase != PERL_PHASE_DESTRUCT
6687 && (hek = HvNAME_HEK((HV*)sv)))
6689 if (PL_stashcache) {
6690 DEBUG_o(Perl_deb(aTHX_
6691 "sv_clear clearing PL_stashcache for '%" HEKf
6694 (void)hv_deletehek(PL_stashcache,
6697 hv_name_set((HV*)sv, NULL, 0, 0);
6700 /* save old iter_sv in unused SvSTASH field */
6701 assert(!SvOBJECT(sv));
6702 SvSTASH(sv) = (HV*)iter_sv;
6705 /* save old hash_index in unused SvMAGIC field */
6706 assert(!SvMAGICAL(sv));
6707 assert(!SvMAGIC(sv));
6708 ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index = hash_index;
6711 next_sv = Perl_hfree_next_entry(aTHX_ (HV*)sv, &hash_index);
6712 goto get_next_sv; /* process this new sv */
6714 /* free empty hash */
6715 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6716 assert(!HvARRAY((HV*)sv));
6720 AV* av = MUTABLE_AV(sv);
6721 if (PL_comppad == av) {
6725 if (AvREAL(av) && AvFILLp(av) > -1) {
6726 next_sv = AvARRAY(av)[AvFILLp(av)--];
6727 /* save old iter_sv in top-most slot of AV,
6728 * and pray that it doesn't get wiped in the meantime */
6729 AvARRAY(av)[AvMAX(av)] = iter_sv;
6731 goto get_next_sv; /* process this new sv */
6733 Safefree(AvALLOC(av));
6738 if (LvTYPE(sv) == 'T') { /* for tie: return HE to pool */
6739 SvREFCNT_dec(HeKEY_sv((HE*)LvTARG(sv)));
6740 HeNEXT((HE*)LvTARG(sv)) = PL_hv_fetch_ent_mh;
6741 PL_hv_fetch_ent_mh = (HE*)LvTARG(sv);
6743 else if (LvTYPE(sv) != 't') /* unless tie: unrefcnted fake SV** */
6744 SvREFCNT_dec(LvTARG(sv));
6746 /* SvLEN points to a regex body. Free the body, then
6747 * set SvLEN to whatever value was in the now-freed
6748 * regex body. The PVX buffer is shared by multiple re's
6749 * and only freed once, by the re whose len in non-null */
6750 STRLEN len = ReANY(sv)->xpv_len;
6751 pregfree2((REGEXP*) sv);
6752 SvLEN_set((sv), len);
6757 if (isGV_with_GP(sv)) {
6758 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
6759 && HvENAME_get(stash))
6760 mro_method_changed_in(stash);
6761 gp_free(MUTABLE_GV(sv));
6763 unshare_hek(GvNAME_HEK(sv));
6764 /* If we're in a stash, we don't own a reference to it.
6765 * However it does have a back reference to us, which
6766 * needs to be cleared. */
6767 if ((stash = GvSTASH(sv)))
6768 sv_del_backref(MUTABLE_SV(stash), sv);
6770 /* FIXME. There are probably more unreferenced pointers to SVs
6771 * in the interpreter struct that we should check and tidy in
6772 * a similar fashion to this: */
6773 /* See also S_sv_unglob, which does the same thing. */
6774 if ((const GV *)sv == PL_last_in_gv)
6775 PL_last_in_gv = NULL;
6776 else if ((const GV *)sv == PL_statgv)
6778 else if ((const GV *)sv == PL_stderrgv)
6787 /* Don't bother with SvOOK_off(sv); as we're only going to
6791 SvOOK_offset(sv, offset);
6792 SvPV_set(sv, SvPVX_mutable(sv) - offset);
6793 /* Don't even bother with turning off the OOK flag. */
6798 SV * const target = SvRV(sv);
6800 sv_del_backref(target, sv);
6806 else if (SvPVX_const(sv)
6807 && !(SvTYPE(sv) == SVt_PVIO
6808 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6813 PerlIO_printf(Perl_debug_log, "Copy on write: clear\n");
6818 if (CowREFCNT(sv)) {
6825 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6830 Safefree(SvPVX_mutable(sv));
6834 else if (SvPVX_const(sv) && SvLEN(sv)
6835 && !(SvTYPE(sv) == SVt_PVIO
6836 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6837 Safefree(SvPVX_mutable(sv));
6838 else if (SvPVX_const(sv) && SvIsCOW(sv)) {
6839 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6849 SvFLAGS(sv) &= SVf_BREAK;
6850 SvFLAGS(sv) |= SVTYPEMASK;
6852 sv_type_details = bodies_by_type + type;
6853 if (sv_type_details->arena) {
6854 del_body(((char *)SvANY(sv) + sv_type_details->offset),
6855 &PL_body_roots[type]);
6857 else if (sv_type_details->body_size) {
6858 safefree(SvANY(sv));
6862 /* caller is responsible for freeing the head of the original sv */
6863 if (sv != orig_sv && !SvREFCNT(sv))
6866 /* grab and free next sv, if any */
6874 else if (!iter_sv) {
6876 } else if (SvTYPE(iter_sv) == SVt_PVAV) {
6877 AV *const av = (AV*)iter_sv;
6878 if (AvFILLp(av) > -1) {
6879 sv = AvARRAY(av)[AvFILLp(av)--];
6881 else { /* no more elements of current AV to free */
6884 /* restore previous value, squirrelled away */
6885 iter_sv = AvARRAY(av)[AvMAX(av)];
6886 Safefree(AvALLOC(av));
6889 } else if (SvTYPE(iter_sv) == SVt_PVHV) {
6890 sv = Perl_hfree_next_entry(aTHX_ (HV*)iter_sv, &hash_index);
6891 if (!sv && !HvTOTALKEYS((HV *)iter_sv)) {
6892 /* no more elements of current HV to free */
6895 /* Restore previous values of iter_sv and hash_index,
6896 * squirrelled away */
6897 assert(!SvOBJECT(sv));
6898 iter_sv = (SV*)SvSTASH(sv);
6899 assert(!SvMAGICAL(sv));
6900 hash_index = ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index;
6902 /* perl -DA does not like rubbish in SvMAGIC. */
6906 /* free any remaining detritus from the hash struct */
6907 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6908 assert(!HvARRAY((HV*)sv));
6913 /* unrolled SvREFCNT_dec and sv_free2 follows: */
6917 if (!SvREFCNT(sv)) {
6921 if (--(SvREFCNT(sv)))
6925 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6926 "Attempt to free temp prematurely: SV 0x%" UVxf
6927 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6931 if (SvIMMORTAL(sv)) {
6932 /* make sure SvREFCNT(sv)==0 happens very seldom */
6933 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6942 /* This routine curses the sv itself, not the object referenced by sv. So
6943 sv does not have to be ROK. */
6946 S_curse(pTHX_ SV * const sv, const bool check_refcnt) {
6947 PERL_ARGS_ASSERT_CURSE;
6948 assert(SvOBJECT(sv));
6950 if (PL_defstash && /* Still have a symbol table? */
6956 stash = SvSTASH(sv);
6957 assert(SvTYPE(stash) == SVt_PVHV);
6958 if (HvNAME(stash)) {
6959 CV* destructor = NULL;
6960 struct mro_meta *meta;
6962 assert (SvOOK(stash));
6964 DEBUG_o( Perl_deb(aTHX_ "Looking for DESTROY method for %s\n",
6967 /* don't make this an initialization above the assert, since it needs
6969 meta = HvMROMETA(stash);
6970 if (meta->destroy_gen && meta->destroy_gen == PL_sub_generation) {
6971 destructor = meta->destroy;
6972 DEBUG_o( Perl_deb(aTHX_ "Using cached DESTROY method %p for %s\n",
6973 (void *)destructor, HvNAME(stash)) );
6976 bool autoload = FALSE;
6978 gv_fetchmeth_pvn(stash, S_destroy, S_destroy_len, -1, 0);
6980 destructor = GvCV(gv);
6982 gv = gv_autoload_pvn(stash, S_destroy, S_destroy_len,
6983 GV_AUTOLOAD_ISMETHOD);
6985 destructor = GvCV(gv);
6989 /* we don't cache AUTOLOAD for DESTROY, since this code
6990 would then need to set $__PACKAGE__::AUTOLOAD, or the
6991 equivalent for XS AUTOLOADs */
6993 meta->destroy_gen = PL_sub_generation;
6994 meta->destroy = destructor;
6996 DEBUG_o( Perl_deb(aTHX_ "Set cached DESTROY method %p for %s\n",
6997 (void *)destructor, HvNAME(stash)) );
7000 DEBUG_o( Perl_deb(aTHX_ "Not caching AUTOLOAD for DESTROY method for %s\n",
7004 assert(!destructor || SvTYPE(destructor) == SVt_PVCV);
7006 /* A constant subroutine can have no side effects, so
7007 don't bother calling it. */
7008 && !CvCONST(destructor)
7009 /* Don't bother calling an empty destructor or one that
7010 returns immediately. */
7011 && (CvISXSUB(destructor)
7012 || (CvSTART(destructor)
7013 && (CvSTART(destructor)->op_next->op_type
7015 && (CvSTART(destructor)->op_next->op_type
7017 || CvSTART(destructor)->op_next->op_next->op_type
7023 SV* const tmpref = newRV(sv);
7024 SvREADONLY_on(tmpref); /* DESTROY() could be naughty */
7026 PUSHSTACKi(PERLSI_DESTROY);
7031 call_sv(MUTABLE_SV(destructor),
7032 G_DISCARD|G_EVAL|G_KEEPERR|G_VOID);
7036 if(SvREFCNT(tmpref) < 2) {
7037 /* tmpref is not kept alive! */
7039 SvRV_set(tmpref, NULL);
7042 SvREFCNT_dec_NN(tmpref);
7045 } while (SvOBJECT(sv) && SvSTASH(sv) != stash);
7048 if (check_refcnt && SvREFCNT(sv)) {
7049 if (PL_in_clean_objs)
7051 "DESTROY created new reference to dead object '%" HEKf "'",
7052 HEKfARG(HvNAME_HEK(stash)));
7053 /* DESTROY gave object new lease on life */
7059 HV * const stash = SvSTASH(sv);
7060 /* Curse before freeing the stash, as freeing the stash could cause
7061 a recursive call into S_curse. */
7062 SvOBJECT_off(sv); /* Curse the object. */
7063 SvSTASH_set(sv,0); /* SvREFCNT_dec may try to read this */
7064 SvREFCNT_dec(stash); /* possibly of changed persuasion */
7070 =for apidoc sv_newref
7072 Increment an SV's reference count. Use the C<SvREFCNT_inc()> wrapper
7079 Perl_sv_newref(pTHX_ SV *const sv)
7081 PERL_UNUSED_CONTEXT;
7090 Decrement an SV's reference count, and if it drops to zero, call
7091 C<sv_clear> to invoke destructors and free up any memory used by
7092 the body; finally, deallocating the SV's head itself.
7093 Normally called via a wrapper macro C<SvREFCNT_dec>.
7099 Perl_sv_free(pTHX_ SV *const sv)
7105 /* Private helper function for SvREFCNT_dec().
7106 * Called with rc set to original SvREFCNT(sv), where rc == 0 or 1 */
7109 Perl_sv_free2(pTHX_ SV *const sv, const U32 rc)
7113 PERL_ARGS_ASSERT_SV_FREE2;
7115 if (LIKELY( rc == 1 )) {
7121 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
7122 "Attempt to free temp prematurely: SV 0x%" UVxf
7123 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
7127 if (SvIMMORTAL(sv)) {
7128 /* make sure SvREFCNT(sv)==0 happens very seldom */
7129 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
7133 if (! SvREFCNT(sv)) /* may have have been resurrected */
7138 /* handle exceptional cases */
7142 if (SvFLAGS(sv) & SVf_BREAK)
7143 /* this SV's refcnt has been artificially decremented to
7144 * trigger cleanup */
7146 if (PL_in_clean_all) /* All is fair */
7148 if (SvIMMORTAL(sv)) {
7149 /* make sure SvREFCNT(sv)==0 happens very seldom */
7150 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
7153 if (ckWARN_d(WARN_INTERNAL)) {
7154 #ifdef DEBUG_LEAKING_SCALARS_FORK_DUMP
7155 Perl_dump_sv_child(aTHX_ sv);
7157 #ifdef DEBUG_LEAKING_SCALARS
7160 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7161 if (PL_warnhook == PERL_WARNHOOK_FATAL
7162 || ckDEAD(packWARN(WARN_INTERNAL))) {
7163 /* Don't let Perl_warner cause us to escape our fate: */
7167 /* This may not return: */
7168 Perl_warner(aTHX_ packWARN(WARN_INTERNAL),
7169 "Attempt to free unreferenced scalar: SV 0x%" UVxf
7170 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
7173 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7183 Returns the length of the string in the SV. Handles magic and type
7184 coercion and sets the UTF8 flag appropriately. See also C<L</SvCUR>>, which
7185 gives raw access to the C<xpv_cur> slot.
7191 Perl_sv_len(pTHX_ SV *const sv)
7198 (void)SvPV_const(sv, len);
7203 =for apidoc sv_len_utf8
7205 Returns the number of characters in the string in an SV, counting wide
7206 UTF-8 bytes as a single character. Handles magic and type coercion.
7212 * The length is cached in PERL_MAGIC_utf8, in the mg_len field. Also the
7213 * mg_ptr is used, by sv_pos_u2b() and sv_pos_b2u() - see the comments below.
7214 * (Note that the mg_len is not the length of the mg_ptr field.
7215 * This allows the cache to store the character length of the string without
7216 * needing to malloc() extra storage to attach to the mg_ptr.)
7221 Perl_sv_len_utf8(pTHX_ SV *const sv)
7227 return sv_len_utf8_nomg(sv);
7231 Perl_sv_len_utf8_nomg(pTHX_ SV * const sv)
7234 const U8 *s = (U8*)SvPV_nomg_const(sv, len);
7236 PERL_ARGS_ASSERT_SV_LEN_UTF8_NOMG;
7238 if (PL_utf8cache && SvUTF8(sv)) {
7240 MAGIC *mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_utf8) : NULL;
7242 if (mg && (mg->mg_len != -1 || mg->mg_ptr)) {
7243 if (mg->mg_len != -1)
7246 /* We can use the offset cache for a headstart.
7247 The longer value is stored in the first pair. */
7248 STRLEN *cache = (STRLEN *) mg->mg_ptr;
7250 ulen = cache[0] + Perl_utf8_length(aTHX_ s + cache[1],
7254 if (PL_utf8cache < 0) {
7255 const STRLEN real = Perl_utf8_length(aTHX_ s, s + len);
7256 assert_uft8_cache_coherent("sv_len_utf8", ulen, real, sv);
7260 ulen = Perl_utf8_length(aTHX_ s, s + len);
7261 utf8_mg_len_cache_update(sv, &mg, ulen);
7265 return SvUTF8(sv) ? Perl_utf8_length(aTHX_ s, s + len) : len;
7268 /* Walk forwards to find the byte corresponding to the passed in UTF-8
7271 S_sv_pos_u2b_forwards(const U8 *const start, const U8 *const send,
7272 STRLEN *const uoffset_p, bool *const at_end)
7274 const U8 *s = start;
7275 STRLEN uoffset = *uoffset_p;
7277 PERL_ARGS_ASSERT_SV_POS_U2B_FORWARDS;
7279 while (s < send && uoffset) {
7286 else if (s > send) {
7288 /* This is the existing behaviour. Possibly it should be a croak, as
7289 it's actually a bounds error */
7292 *uoffset_p -= uoffset;
7296 /* Given the length of the string in both bytes and UTF-8 characters, decide
7297 whether to walk forwards or backwards to find the byte corresponding to
7298 the passed in UTF-8 offset. */
7300 S_sv_pos_u2b_midway(const U8 *const start, const U8 *send,
7301 STRLEN uoffset, const STRLEN uend)
7303 STRLEN backw = uend - uoffset;
7305 PERL_ARGS_ASSERT_SV_POS_U2B_MIDWAY;
7307 if (uoffset < 2 * backw) {
7308 /* The assumption is that going forwards is twice the speed of going
7309 forward (that's where the 2 * backw comes from).
7310 (The real figure of course depends on the UTF-8 data.) */
7311 const U8 *s = start;
7313 while (s < send && uoffset--)
7323 while (UTF8_IS_CONTINUATION(*send))
7326 return send - start;
7329 /* For the string representation of the given scalar, find the byte
7330 corresponding to the passed in UTF-8 offset. uoffset0 and boffset0
7331 give another position in the string, *before* the sought offset, which
7332 (which is always true, as 0, 0 is a valid pair of positions), which should
7333 help reduce the amount of linear searching.
7334 If *mgp is non-NULL, it should point to the UTF-8 cache magic, which
7335 will be used to reduce the amount of linear searching. The cache will be
7336 created if necessary, and the found value offered to it for update. */
7338 S_sv_pos_u2b_cached(pTHX_ SV *const sv, MAGIC **const mgp, const U8 *const start,
7339 const U8 *const send, STRLEN uoffset,
7340 STRLEN uoffset0, STRLEN boffset0)
7342 STRLEN boffset = 0; /* Actually always set, but let's keep gcc happy. */
7344 bool at_end = FALSE;
7346 PERL_ARGS_ASSERT_SV_POS_U2B_CACHED;
7348 assert (uoffset >= uoffset0);
7353 if (!SvREADONLY(sv) && !SvGMAGICAL(sv) && SvPOK(sv)
7355 && (*mgp || (SvTYPE(sv) >= SVt_PVMG &&
7356 (*mgp = mg_find(sv, PERL_MAGIC_utf8))))) {
7357 if ((*mgp)->mg_ptr) {
7358 STRLEN *cache = (STRLEN *) (*mgp)->mg_ptr;
7359 if (cache[0] == uoffset) {
7360 /* An exact match. */
7363 if (cache[2] == uoffset) {
7364 /* An exact match. */
7368 if (cache[0] < uoffset) {
7369 /* The cache already knows part of the way. */
7370 if (cache[0] > uoffset0) {
7371 /* The cache knows more than the passed in pair */
7372 uoffset0 = cache[0];
7373 boffset0 = cache[1];
7375 if ((*mgp)->mg_len != -1) {
7376 /* And we know the end too. */
7378 + sv_pos_u2b_midway(start + boffset0, send,
7380 (*mgp)->mg_len - uoffset0);
7382 uoffset -= uoffset0;
7384 + sv_pos_u2b_forwards(start + boffset0,
7385 send, &uoffset, &at_end);
7386 uoffset += uoffset0;
7389 else if (cache[2] < uoffset) {
7390 /* We're between the two cache entries. */
7391 if (cache[2] > uoffset0) {
7392 /* and the cache knows more than the passed in pair */
7393 uoffset0 = cache[2];
7394 boffset0 = cache[3];
7398 + sv_pos_u2b_midway(start + boffset0,
7401 cache[0] - uoffset0);
7404 + sv_pos_u2b_midway(start + boffset0,
7407 cache[2] - uoffset0);
7411 else if ((*mgp)->mg_len != -1) {
7412 /* If we can take advantage of a passed in offset, do so. */
7413 /* In fact, offset0 is either 0, or less than offset, so don't
7414 need to worry about the other possibility. */
7416 + sv_pos_u2b_midway(start + boffset0, send,
7418 (*mgp)->mg_len - uoffset0);
7423 if (!found || PL_utf8cache < 0) {
7424 STRLEN real_boffset;
7425 uoffset -= uoffset0;
7426 real_boffset = boffset0 + sv_pos_u2b_forwards(start + boffset0,
7427 send, &uoffset, &at_end);
7428 uoffset += uoffset0;
7430 if (found && PL_utf8cache < 0)
7431 assert_uft8_cache_coherent("sv_pos_u2b_cache", boffset,
7433 boffset = real_boffset;
7436 if (PL_utf8cache && !SvGMAGICAL(sv) && SvPOK(sv)) {
7438 utf8_mg_len_cache_update(sv, mgp, uoffset);
7440 utf8_mg_pos_cache_update(sv, mgp, boffset, uoffset, send - start);
7447 =for apidoc sv_pos_u2b_flags
7449 Converts the offset from a count of UTF-8 chars from
7450 the start of the string, to a count of the equivalent number of bytes; if
7451 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7452 C<offset>, rather than from the start
7453 of the string. Handles type coercion.
7454 C<flags> is passed to C<SvPV_flags>, and usually should be
7455 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7461 * sv_pos_u2b_flags() uses, like sv_pos_b2u(), the mg_ptr of the potential
7462 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7463 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7468 Perl_sv_pos_u2b_flags(pTHX_ SV *const sv, STRLEN uoffset, STRLEN *const lenp,
7475 PERL_ARGS_ASSERT_SV_POS_U2B_FLAGS;
7477 start = (U8*)SvPV_flags(sv, len, flags);
7479 const U8 * const send = start + len;
7481 boffset = sv_pos_u2b_cached(sv, &mg, start, send, uoffset, 0, 0);
7484 && *lenp /* don't bother doing work for 0, as its bytes equivalent
7485 is 0, and *lenp is already set to that. */) {
7486 /* Convert the relative offset to absolute. */
7487 const STRLEN uoffset2 = uoffset + *lenp;
7488 const STRLEN boffset2
7489 = sv_pos_u2b_cached(sv, &mg, start, send, uoffset2,
7490 uoffset, boffset) - boffset;
7504 =for apidoc sv_pos_u2b
7506 Converts the value pointed to by C<offsetp> from a count of UTF-8 chars from
7507 the start of the string, to a count of the equivalent number of bytes; if
7508 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7509 the offset, rather than from the start of the string. Handles magic and
7512 Use C<sv_pos_u2b_flags> in preference, which correctly handles strings longer
7519 * sv_pos_u2b() uses, like sv_pos_b2u(), the mg_ptr of the potential
7520 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7521 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7525 /* This function is subject to size and sign problems */
7528 Perl_sv_pos_u2b(pTHX_ SV *const sv, I32 *const offsetp, I32 *const lenp)
7530 PERL_ARGS_ASSERT_SV_POS_U2B;
7533 STRLEN ulen = (STRLEN)*lenp;
7534 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, &ulen,
7535 SV_GMAGIC|SV_CONST_RETURN);
7538 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, NULL,
7539 SV_GMAGIC|SV_CONST_RETURN);
7544 S_utf8_mg_len_cache_update(pTHX_ SV *const sv, MAGIC **const mgp,
7547 PERL_ARGS_ASSERT_UTF8_MG_LEN_CACHE_UPDATE;
7548 if (SvREADONLY(sv) || SvGMAGICAL(sv) || !SvPOK(sv))
7551 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7552 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7553 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, &PL_vtbl_utf8, 0, 0);
7557 (*mgp)->mg_len = ulen;
7560 /* Create and update the UTF8 magic offset cache, with the proffered utf8/
7561 byte length pairing. The (byte) length of the total SV is passed in too,
7562 as blen, because for some (more esoteric) SVs, the call to SvPV_const()
7563 may not have updated SvCUR, so we can't rely on reading it directly.
7565 The proffered utf8/byte length pairing isn't used if the cache already has
7566 two pairs, and swapping either for the proffered pair would increase the
7567 RMS of the intervals between known byte offsets.
7569 The cache itself consists of 4 STRLEN values
7570 0: larger UTF-8 offset
7571 1: corresponding byte offset
7572 2: smaller UTF-8 offset
7573 3: corresponding byte offset
7575 Unused cache pairs have the value 0, 0.
7576 Keeping the cache "backwards" means that the invariant of
7577 cache[0] >= cache[2] is maintained even with empty slots, which means that
7578 the code that uses it doesn't need to worry if only 1 entry has actually
7579 been set to non-zero. It also makes the "position beyond the end of the
7580 cache" logic much simpler, as the first slot is always the one to start
7584 S_utf8_mg_pos_cache_update(pTHX_ SV *const sv, MAGIC **const mgp, const STRLEN byte,
7585 const STRLEN utf8, const STRLEN blen)
7589 PERL_ARGS_ASSERT_UTF8_MG_POS_CACHE_UPDATE;
7594 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7595 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7596 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, (MGVTBL*)&PL_vtbl_utf8, 0,
7598 (*mgp)->mg_len = -1;
7602 if (!(cache = (STRLEN *)(*mgp)->mg_ptr)) {
7603 Newxz(cache, PERL_MAGIC_UTF8_CACHESIZE * 2, STRLEN);
7604 (*mgp)->mg_ptr = (char *) cache;
7608 if (PL_utf8cache < 0 && SvPOKp(sv)) {
7609 /* SvPOKp() because, if sv is a reference, then SvPVX() is actually
7610 a pointer. Note that we no longer cache utf8 offsets on refer-
7611 ences, but this check is still a good idea, for robustness. */
7612 const U8 *start = (const U8 *) SvPVX_const(sv);
7613 const STRLEN realutf8 = utf8_length(start, start + byte);
7615 assert_uft8_cache_coherent("utf8_mg_pos_cache_update", utf8, realutf8,
7619 /* Cache is held with the later position first, to simplify the code
7620 that deals with unbounded ends. */
7622 ASSERT_UTF8_CACHE(cache);
7623 if (cache[1] == 0) {
7624 /* Cache is totally empty */
7627 } else if (cache[3] == 0) {
7628 if (byte > cache[1]) {
7629 /* New one is larger, so goes first. */
7630 cache[2] = cache[0];
7631 cache[3] = cache[1];
7639 /* float casts necessary? XXX */
7640 #define THREEWAY_SQUARE(a,b,c,d) \
7641 ((float)((d) - (c))) * ((float)((d) - (c))) \
7642 + ((float)((c) - (b))) * ((float)((c) - (b))) \
7643 + ((float)((b) - (a))) * ((float)((b) - (a)))
7645 /* Cache has 2 slots in use, and we know three potential pairs.
7646 Keep the two that give the lowest RMS distance. Do the
7647 calculation in bytes simply because we always know the byte
7648 length. squareroot has the same ordering as the positive value,
7649 so don't bother with the actual square root. */
7650 if (byte > cache[1]) {
7651 /* New position is after the existing pair of pairs. */
7652 const float keep_earlier
7653 = THREEWAY_SQUARE(0, cache[3], byte, blen);
7654 const float keep_later
7655 = THREEWAY_SQUARE(0, cache[1], byte, blen);
7657 if (keep_later < keep_earlier) {
7658 cache[2] = cache[0];
7659 cache[3] = cache[1];
7665 const float keep_later = THREEWAY_SQUARE(0, byte, cache[1], blen);
7666 float b, c, keep_earlier;
7667 if (byte > cache[3]) {
7668 /* New position is between the existing pair of pairs. */
7669 b = (float)cache[3];
7672 /* New position is before the existing pair of pairs. */
7674 c = (float)cache[3];
7676 keep_earlier = THREEWAY_SQUARE(0, b, c, blen);
7677 if (byte > cache[3]) {
7678 if (keep_later < keep_earlier) {
7688 if (! (keep_later < keep_earlier)) {
7689 cache[0] = cache[2];
7690 cache[1] = cache[3];
7697 ASSERT_UTF8_CACHE(cache);
7700 /* We already know all of the way, now we may be able to walk back. The same
7701 assumption is made as in S_sv_pos_u2b_midway(), namely that walking
7702 backward is half the speed of walking forward. */
7704 S_sv_pos_b2u_midway(pTHX_ const U8 *const s, const U8 *const target,
7705 const U8 *end, STRLEN endu)
7707 const STRLEN forw = target - s;
7708 STRLEN backw = end - target;
7710 PERL_ARGS_ASSERT_SV_POS_B2U_MIDWAY;
7712 if (forw < 2 * backw) {
7713 return utf8_length(s, target);
7716 while (end > target) {
7718 while (UTF8_IS_CONTINUATION(*end)) {
7727 =for apidoc sv_pos_b2u_flags
7729 Converts C<offset> from a count of bytes from the start of the string, to
7730 a count of the equivalent number of UTF-8 chars. Handles type coercion.
7731 C<flags> is passed to C<SvPV_flags>, and usually should be
7732 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7738 * sv_pos_b2u_flags() uses, like sv_pos_u2b_flags(), the mg_ptr of the
7739 * potential PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8
7744 Perl_sv_pos_b2u_flags(pTHX_ SV *const sv, STRLEN const offset, U32 flags)
7747 STRLEN len = 0; /* Actually always set, but let's keep gcc happy. */
7753 PERL_ARGS_ASSERT_SV_POS_B2U_FLAGS;
7755 s = (const U8*)SvPV_flags(sv, blen, flags);
7758 Perl_croak(aTHX_ "panic: sv_pos_b2u: bad byte offset, blen=%" UVuf
7759 ", byte=%" UVuf, (UV)blen, (UV)offset);
7765 && SvTYPE(sv) >= SVt_PVMG
7766 && (mg = mg_find(sv, PERL_MAGIC_utf8)))
7769 STRLEN * const cache = (STRLEN *) mg->mg_ptr;
7770 if (cache[1] == offset) {
7771 /* An exact match. */
7774 if (cache[3] == offset) {
7775 /* An exact match. */
7779 if (cache[1] < offset) {
7780 /* We already know part of the way. */
7781 if (mg->mg_len != -1) {
7782 /* Actually, we know the end too. */
7784 + S_sv_pos_b2u_midway(aTHX_ s + cache[1], send,
7785 s + blen, mg->mg_len - cache[0]);
7787 len = cache[0] + utf8_length(s + cache[1], send);
7790 else if (cache[3] < offset) {
7791 /* We're between the two cached pairs, so we do the calculation
7792 offset by the byte/utf-8 positions for the earlier pair,
7793 then add the utf-8 characters from the string start to
7795 len = S_sv_pos_b2u_midway(aTHX_ s + cache[3], send,
7796 s + cache[1], cache[0] - cache[2])
7800 else { /* cache[3] > offset */
7801 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + cache[3],
7805 ASSERT_UTF8_CACHE(cache);
7807 } else if (mg->mg_len != -1) {
7808 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + blen, mg->mg_len);
7812 if (!found || PL_utf8cache < 0) {
7813 const STRLEN real_len = utf8_length(s, send);
7815 if (found && PL_utf8cache < 0)
7816 assert_uft8_cache_coherent("sv_pos_b2u", len, real_len, sv);
7822 utf8_mg_len_cache_update(sv, &mg, len);
7824 utf8_mg_pos_cache_update(sv, &mg, offset, len, blen);
7831 =for apidoc sv_pos_b2u
7833 Converts the value pointed to by C<offsetp> from a count of bytes from the
7834 start of the string, to a count of the equivalent number of UTF-8 chars.
7835 Handles magic and type coercion.
7837 Use C<sv_pos_b2u_flags> in preference, which correctly handles strings
7844 * sv_pos_b2u() uses, like sv_pos_u2b(), the mg_ptr of the potential
7845 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7850 Perl_sv_pos_b2u(pTHX_ SV *const sv, I32 *const offsetp)
7852 PERL_ARGS_ASSERT_SV_POS_B2U;
7857 *offsetp = (I32)sv_pos_b2u_flags(sv, (STRLEN)*offsetp,
7858 SV_GMAGIC|SV_CONST_RETURN);
7862 S_assert_uft8_cache_coherent(pTHX_ const char *const func, STRLEN from_cache,
7863 STRLEN real, SV *const sv)
7865 PERL_ARGS_ASSERT_ASSERT_UFT8_CACHE_COHERENT;
7867 /* As this is debugging only code, save space by keeping this test here,
7868 rather than inlining it in all the callers. */
7869 if (from_cache == real)
7872 /* Need to turn the assertions off otherwise we may recurse infinitely
7873 while printing error messages. */
7874 SAVEI8(PL_utf8cache);
7876 Perl_croak(aTHX_ "panic: %s cache %" UVuf " real %" UVuf " for %" SVf,
7877 func, (UV) from_cache, (UV) real, SVfARG(sv));
7883 Returns a boolean indicating whether the strings in the two SVs are
7884 identical. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7885 coerce its args to strings if necessary.
7887 =for apidoc sv_eq_flags
7889 Returns a boolean indicating whether the strings in the two SVs are
7890 identical. Is UTF-8 and S<C<'use bytes'>> aware and coerces its args to strings
7891 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get-magic, too.
7897 Perl_sv_eq_flags(pTHX_ SV *sv1, SV *sv2, const U32 flags)
7909 /* if pv1 and pv2 are the same, second SvPV_const call may
7910 * invalidate pv1 (if we are handling magic), so we may need to
7912 if (sv1 == sv2 && flags & SV_GMAGIC
7913 && (SvTHINKFIRST(sv1) || SvGMAGICAL(sv1))) {
7914 pv1 = SvPV_const(sv1, cur1);
7915 sv1 = newSVpvn_flags(pv1, cur1, SVs_TEMP | SvUTF8(sv2));
7917 pv1 = SvPV_flags_const(sv1, cur1, flags);
7925 pv2 = SvPV_flags_const(sv2, cur2, flags);
7927 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7928 /* Differing utf8ness. */
7930 /* sv1 is the UTF-8 one */
7931 return bytes_cmp_utf8((const U8*)pv2, cur2,
7932 (const U8*)pv1, cur1) == 0;
7935 /* sv2 is the UTF-8 one */
7936 return bytes_cmp_utf8((const U8*)pv1, cur1,
7937 (const U8*)pv2, cur2) == 0;
7942 return (pv1 == pv2) || memEQ(pv1, pv2, cur1);
7950 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7951 string in C<sv1> is less than, equal to, or greater than the string in
7952 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7953 coerce its args to strings if necessary. See also C<L</sv_cmp_locale>>.
7955 =for apidoc sv_cmp_flags
7957 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7958 string in C<sv1> is less than, equal to, or greater than the string in
7959 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware and will coerce its args to strings
7960 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get magic. See
7961 also C<L</sv_cmp_locale_flags>>.
7967 Perl_sv_cmp(pTHX_ SV *const sv1, SV *const sv2)
7969 return sv_cmp_flags(sv1, sv2, SV_GMAGIC);
7973 Perl_sv_cmp_flags(pTHX_ SV *const sv1, SV *const sv2,
7977 const char *pv1, *pv2;
7979 SV *svrecode = NULL;
7986 pv1 = SvPV_flags_const(sv1, cur1, flags);
7993 pv2 = SvPV_flags_const(sv2, cur2, flags);
7995 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7996 /* Differing utf8ness. */
7998 const int retval = -bytes_cmp_utf8((const U8*)pv2, cur2,
7999 (const U8*)pv1, cur1);
8000 return retval ? retval < 0 ? -1 : +1 : 0;
8003 const int retval = bytes_cmp_utf8((const U8*)pv1, cur1,
8004 (const U8*)pv2, cur2);
8005 return retval ? retval < 0 ? -1 : +1 : 0;
8009 /* Here, if both are non-NULL, then they have the same UTF8ness. */
8012 cmp = cur2 ? -1 : 0;
8016 STRLEN shortest_len = cur1 < cur2 ? cur1 : cur2;
8019 if (! DO_UTF8(sv1)) {
8021 const I32 retval = memcmp((const void*)pv1,
8025 cmp = retval < 0 ? -1 : 1;
8026 } else if (cur1 == cur2) {
8029 cmp = cur1 < cur2 ? -1 : 1;
8033 else { /* Both are to be treated as UTF-EBCDIC */
8035 /* EBCDIC UTF-8 is complicated by the fact that it is based on I8
8036 * which remaps code points 0-255. We therefore generally have to
8037 * unmap back to the original values to get an accurate comparison.
8038 * But we don't have to do that for UTF-8 invariants, as by
8039 * definition, they aren't remapped, nor do we have to do it for
8040 * above-latin1 code points, as they also aren't remapped. (This
8041 * code also works on ASCII platforms, but the memcmp() above is
8044 const char *e = pv1 + shortest_len;
8046 /* Find the first bytes that differ between the two strings */
8047 while (pv1 < e && *pv1 == *pv2) {
8053 if (pv1 == e) { /* Are the same all the way to the end */
8057 cmp = cur1 < cur2 ? -1 : 1;
8060 else /* Here *pv1 and *pv2 are not equal, but all bytes earlier
8061 * in the strings were. The current bytes may or may not be
8062 * at the beginning of a character. But neither or both are
8063 * (or else earlier bytes would have been different). And
8064 * if we are in the middle of a character, the two
8065 * characters are comprised of the same number of bytes
8066 * (because in this case the start bytes are the same, and
8067 * the start bytes encode the character's length). */
8068 if (UTF8_IS_INVARIANT(*pv1))
8070 /* If both are invariants; can just compare directly */
8071 if (UTF8_IS_INVARIANT(*pv2)) {
8072 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8074 else /* Since *pv1 is invariant, it is the whole character,
8075 which means it is at the beginning of a character.
8076 That means pv2 is also at the beginning of a
8077 character (see earlier comment). Since it isn't
8078 invariant, it must be a start byte. If it starts a
8079 character whose code point is above 255, that
8080 character is greater than any single-byte char, which
8082 if (UTF8_IS_ABOVE_LATIN1_START(*pv2))
8087 /* Here, pv2 points to a character composed of 2 bytes
8088 * whose code point is < 256. Get its code point and
8089 * compare with *pv1 */
8090 cmp = ((U8) *pv1 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
8095 else /* The code point starting at pv1 isn't a single byte */
8096 if (UTF8_IS_INVARIANT(*pv2))
8098 /* But here, the code point starting at *pv2 is a single byte,
8099 * and so *pv1 must begin a character, hence is a start byte.
8100 * If that character is above 255, it is larger than any
8101 * single-byte char, which *pv2 is */
8102 if (UTF8_IS_ABOVE_LATIN1_START(*pv1)) {
8106 /* Here, pv1 points to a character composed of 2 bytes
8107 * whose code point is < 256. Get its code point and
8108 * compare with the single byte character *pv2 */
8109 cmp = (EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1)) < (U8) *pv2)
8114 else /* Here, we've ruled out either *pv1 and *pv2 being
8115 invariant. That means both are part of variants, but not
8116 necessarily at the start of a character */
8117 if ( UTF8_IS_ABOVE_LATIN1_START(*pv1)
8118 || UTF8_IS_ABOVE_LATIN1_START(*pv2))
8120 /* Here, at least one is the start of a character, which means
8121 * the other is also a start byte. And the code point of at
8122 * least one of the characters is above 255. It is a
8123 * characteristic of UTF-EBCDIC that all start bytes for
8124 * above-latin1 code points are well behaved as far as code
8125 * point comparisons go, and all are larger than all other
8126 * start bytes, so the comparison with those is also well
8128 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8131 /* Here both *pv1 and *pv2 are part of variant characters.
8132 * They could be both continuations, or both start characters.
8133 * (One or both could even be an illegal start character (for
8134 * an overlong) which for the purposes of sorting we treat as
8136 if (UTF8_IS_CONTINUATION(*pv1)) {
8138 /* If they are continuations for code points above 255,
8139 * then comparing the current byte is sufficient, as there
8140 * is no remapping of these and so the comparison is
8141 * well-behaved. We determine if they are such
8142 * continuations by looking at the preceding byte. It
8143 * could be a start byte, from which we can tell if it is
8144 * for an above 255 code point. Or it could be a
8145 * continuation, which means the character occupies at
8146 * least 3 bytes, so must be above 255. */
8147 if ( UTF8_IS_CONTINUATION(*(pv2 - 1))
8148 || UTF8_IS_ABOVE_LATIN1_START(*(pv2 -1)))
8150 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8154 /* Here, the continuations are for code points below 256;
8155 * back up one to get to the start byte */
8160 /* We need to get the actual native code point of each of these
8161 * variants in order to compare them */
8162 cmp = ( EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1))
8163 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
8172 SvREFCNT_dec(svrecode);
8178 =for apidoc sv_cmp_locale
8180 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
8181 S<C<'use bytes'>> aware, handles get magic, and will coerce its args to strings
8182 if necessary. See also C<L</sv_cmp>>.
8184 =for apidoc sv_cmp_locale_flags
8186 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
8187 S<C<'use bytes'>> aware and will coerce its args to strings if necessary. If
8188 the flags contain C<SV_GMAGIC>, it handles get magic. See also
8189 C<L</sv_cmp_flags>>.
8195 Perl_sv_cmp_locale(pTHX_ SV *const sv1, SV *const sv2)
8197 return sv_cmp_locale_flags(sv1, sv2, SV_GMAGIC);
8201 Perl_sv_cmp_locale_flags(pTHX_ SV *const sv1, SV *const sv2,
8204 #ifdef USE_LOCALE_COLLATE
8210 if (PL_collation_standard)
8215 /* Revert to using raw compare if both operands exist, but either one
8216 * doesn't transform properly for collation */
8218 pv1 = sv_collxfrm_flags(sv1, &len1, flags);
8222 pv2 = sv_collxfrm_flags(sv2, &len2, flags);
8228 pv1 = sv1 ? sv_collxfrm_flags(sv1, &len1, flags) : (char *) NULL;
8229 pv2 = sv2 ? sv_collxfrm_flags(sv2, &len2, flags) : (char *) NULL;
8232 if (!pv1 || !len1) {
8243 retval = memcmp((void*)pv1, (void*)pv2, len1 < len2 ? len1 : len2);
8246 return retval < 0 ? -1 : 1;
8249 * When the result of collation is equality, that doesn't mean
8250 * that there are no differences -- some locales exclude some
8251 * characters from consideration. So to avoid false equalities,
8252 * we use the raw string as a tiebreaker.
8259 PERL_UNUSED_ARG(flags);
8260 #endif /* USE_LOCALE_COLLATE */
8262 return sv_cmp(sv1, sv2);
8266 #ifdef USE_LOCALE_COLLATE
8269 =for apidoc sv_collxfrm
8271 This calls C<sv_collxfrm_flags> with the SV_GMAGIC flag. See
8272 C<L</sv_collxfrm_flags>>.
8274 =for apidoc sv_collxfrm_flags
8276 Add Collate Transform magic to an SV if it doesn't already have it. If the
8277 flags contain C<SV_GMAGIC>, it handles get-magic.
8279 Any scalar variable may carry C<PERL_MAGIC_collxfrm> magic that contains the
8280 scalar data of the variable, but transformed to such a format that a normal
8281 memory comparison can be used to compare the data according to the locale
8288 Perl_sv_collxfrm_flags(pTHX_ SV *const sv, STRLEN *const nxp, const I32 flags)
8292 PERL_ARGS_ASSERT_SV_COLLXFRM_FLAGS;
8294 mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_collxfrm) : (MAGIC *) NULL;
8296 /* If we don't have collation magic on 'sv', or the locale has changed
8297 * since the last time we calculated it, get it and save it now */
8298 if (!mg || !mg->mg_ptr || *(U32*)mg->mg_ptr != PL_collation_ix) {
8303 /* Free the old space */
8305 Safefree(mg->mg_ptr);
8307 s = SvPV_flags_const(sv, len, flags);
8308 if ((xf = _mem_collxfrm(s, len, &xlen, cBOOL(SvUTF8(sv))))) {
8310 mg = sv_magicext(sv, 0, PERL_MAGIC_collxfrm, &PL_vtbl_collxfrm,
8325 if (mg && mg->mg_ptr) {
8327 return mg->mg_ptr + sizeof(PL_collation_ix);
8335 #endif /* USE_LOCALE_COLLATE */
8338 S_sv_gets_append_to_utf8(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8340 SV * const tsv = newSV(0);
8343 sv_gets(tsv, fp, 0);
8344 sv_utf8_upgrade_nomg(tsv);
8345 SvCUR_set(sv,append);
8348 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8352 S_sv_gets_read_record(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8355 const STRLEN recsize = SvUV(SvRV(PL_rs)); /* RsRECORD() guarantees > 0. */
8356 /* Grab the size of the record we're getting */
8357 char *buffer = SvGROW(sv, (STRLEN)(recsize + append + 1)) + append;
8364 /* With a true, record-oriented file on VMS, we need to use read directly
8365 * to ensure that we respect RMS record boundaries. The user is responsible
8366 * for providing a PL_rs value that corresponds to the FAB$W_MRS (maximum
8367 * record size) field. N.B. This is likely to produce invalid results on
8368 * varying-width character data when a record ends mid-character.
8370 fd = PerlIO_fileno(fp);
8372 && PerlLIO_fstat(fd, &st) == 0
8373 && (st.st_fab_rfm == FAB$C_VAR
8374 || st.st_fab_rfm == FAB$C_VFC
8375 || st.st_fab_rfm == FAB$C_FIX)) {
8377 bytesread = PerlLIO_read(fd, buffer, recsize);
8379 else /* in-memory file from PerlIO::Scalar
8380 * or not a record-oriented file
8384 bytesread = PerlIO_read(fp, buffer, recsize);
8386 /* At this point, the logic in sv_get() means that sv will
8387 be treated as utf-8 if the handle is utf8.
8389 if (PerlIO_isutf8(fp) && bytesread > 0) {
8390 char *bend = buffer + bytesread;
8391 char *bufp = buffer;
8392 size_t charcount = 0;
8393 bool charstart = TRUE;
8396 while (charcount < recsize) {
8397 /* count accumulated characters */
8398 while (bufp < bend) {
8400 skip = UTF8SKIP(bufp);
8402 if (bufp + skip > bend) {
8403 /* partial at the end */
8414 if (charcount < recsize) {
8416 STRLEN bufp_offset = bufp - buffer;
8417 SSize_t morebytesread;
8419 /* originally I read enough to fill any incomplete
8420 character and the first byte of the next
8421 character if needed, but if there's many
8422 multi-byte encoded characters we're going to be
8423 making a read call for every character beyond
8424 the original read size.
8426 So instead, read the rest of the character if
8427 any, and enough bytes to match at least the
8428 start bytes for each character we're going to
8432 readsize = recsize - charcount;
8434 readsize = skip - (bend - bufp) + recsize - charcount - 1;
8435 buffer = SvGROW(sv, append + bytesread + readsize + 1) + append;
8436 bend = buffer + bytesread;
8437 morebytesread = PerlIO_read(fp, bend, readsize);
8438 if (morebytesread <= 0) {
8439 /* we're done, if we still have incomplete
8440 characters the check code in sv_gets() will
8443 I'd originally considered doing
8444 PerlIO_ungetc() on all but the lead
8445 character of the incomplete character, but
8446 read() doesn't do that, so I don't.
8451 /* prepare to scan some more */
8452 bytesread += morebytesread;
8453 bend = buffer + bytesread;
8454 bufp = buffer + bufp_offset;
8462 SvCUR_set(sv, bytesread + append);
8463 buffer[bytesread] = '\0';
8464 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8470 Get a line from the filehandle and store it into the SV, optionally
8471 appending to the currently-stored string. If C<append> is not 0, the
8472 line is appended to the SV instead of overwriting it. C<append> should
8473 be set to the byte offset that the appended string should start at
8474 in the SV (typically, C<SvCUR(sv)> is a suitable choice).
8480 Perl_sv_gets(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8490 PERL_ARGS_ASSERT_SV_GETS;
8492 if (SvTHINKFIRST(sv))
8493 sv_force_normal_flags(sv, append ? 0 : SV_COW_DROP_PV);
8494 /* XXX. If you make this PVIV, then copy on write can copy scalars read
8496 However, perlbench says it's slower, because the existing swipe code
8497 is faster than copy on write.
8498 Swings and roundabouts. */
8499 SvUPGRADE(sv, SVt_PV);
8502 /* line is going to be appended to the existing buffer in the sv */
8503 if (PerlIO_isutf8(fp)) {
8505 sv_utf8_upgrade_nomg(sv);
8506 sv_pos_u2b(sv,&append,0);
8508 } else if (SvUTF8(sv)) {
8509 return S_sv_gets_append_to_utf8(aTHX_ sv, fp, append);
8515 /* not appending - "clear" the string by setting SvCUR to 0,
8516 * the pv is still avaiable. */
8519 if (PerlIO_isutf8(fp))
8522 if (IN_PERL_COMPILETIME) {
8523 /* we always read code in line mode */
8527 else if (RsSNARF(PL_rs)) {
8528 /* If it is a regular disk file use size from stat() as estimate
8529 of amount we are going to read -- may result in mallocing
8530 more memory than we really need if the layers below reduce
8531 the size we read (e.g. CRLF or a gzip layer).
8534 int fd = PerlIO_fileno(fp);
8535 if (fd >= 0 && (PerlLIO_fstat(fd, &st) == 0) && S_ISREG(st.st_mode)) {
8536 const Off_t offset = PerlIO_tell(fp);
8537 if (offset != (Off_t) -1 && st.st_size + append > offset) {
8538 #ifdef PERL_COPY_ON_WRITE
8539 /* Add an extra byte for the sake of copy-on-write's
8540 * buffer reference count. */
8541 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 2));
8543 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 1));
8550 else if (RsRECORD(PL_rs)) {
8551 return S_sv_gets_read_record(aTHX_ sv, fp, append);
8553 else if (RsPARA(PL_rs)) {
8559 /* Get $/ i.e. PL_rs into same encoding as stream wants */
8560 if (PerlIO_isutf8(fp)) {
8561 rsptr = SvPVutf8(PL_rs, rslen);
8564 if (SvUTF8(PL_rs)) {
8565 if (!sv_utf8_downgrade(PL_rs, TRUE)) {
8566 Perl_croak(aTHX_ "Wide character in $/");
8569 /* extract the raw pointer to the record separator */
8570 rsptr = SvPV_const(PL_rs, rslen);
8574 /* rslast is the last character in the record separator
8575 * note we don't use rslast except when rslen is true, so the
8576 * null assign is a placeholder. */
8577 rslast = rslen ? rsptr[rslen - 1] : '\0';
8579 if (rspara) { /* have to do this both before and after */
8580 /* to make sure file boundaries work right */
8584 i = PerlIO_getc(fp);
8588 PerlIO_ungetc(fp,i);
8594 /* See if we know enough about I/O mechanism to cheat it ! */
8596 /* This used to be #ifdef test - it is made run-time test for ease
8597 of abstracting out stdio interface. One call should be cheap
8598 enough here - and may even be a macro allowing compile
8602 if (PerlIO_fast_gets(fp)) {
8604 * We can do buffer based IO operations on this filehandle.
8606 * This means we can bypass a lot of subcalls and process
8607 * the buffer directly, it also means we know the upper bound
8608 * on the amount of data we might read of the current buffer
8609 * into our sv. Knowing this allows us to preallocate the pv
8610 * to be able to hold that maximum, which allows us to simplify
8611 * a lot of logic. */
8614 * We're going to steal some values from the stdio struct
8615 * and put EVERYTHING in the innermost loop into registers.
8617 STDCHAR *ptr; /* pointer into fp's read-ahead buffer */
8618 STRLEN bpx; /* length of the data in the target sv
8619 used to fix pointers after a SvGROW */
8620 I32 shortbuffered; /* If the pv buffer is shorter than the amount
8621 of data left in the read-ahead buffer.
8622 If 0 then the pv buffer can hold the full
8623 amount left, otherwise this is the amount it
8626 /* Here is some breathtakingly efficient cheating */
8628 /* When you read the following logic resist the urge to think
8629 * of record separators that are 1 byte long. They are an
8630 * uninteresting special (simple) case.
8632 * Instead think of record separators which are at least 2 bytes
8633 * long, and keep in mind that we need to deal with such
8634 * separators when they cross a read-ahead buffer boundary.
8636 * Also consider that we need to gracefully deal with separators
8637 * that may be longer than a single read ahead buffer.
8639 * Lastly do not forget we want to copy the delimiter as well. We
8640 * are copying all data in the file _up_to_and_including_ the separator
8643 * Now that you have all that in mind here is what is happening below:
8645 * 1. When we first enter the loop we do some memory book keeping to see
8646 * how much free space there is in the target SV. (This sub assumes that
8647 * it is operating on the same SV most of the time via $_ and that it is
8648 * going to be able to reuse the same pv buffer each call.) If there is
8649 * "enough" room then we set "shortbuffered" to how much space there is
8650 * and start reading forward.
8652 * 2. When we scan forward we copy from the read-ahead buffer to the target
8653 * SV's pv buffer. While we go we watch for the end of the read-ahead buffer,
8654 * and the end of the of pv, as well as for the "rslast", which is the last
8655 * char of the separator.
8657 * 3. When scanning forward if we see rslast then we jump backwards in *pv*
8658 * (which has a "complete" record up to the point we saw rslast) and check
8659 * it to see if it matches the separator. If it does we are done. If it doesn't
8660 * we continue on with the scan/copy.
8662 * 4. If we run out of read-ahead buffer (cnt goes to 0) then we have to get
8663 * the IO system to read the next buffer. We do this by doing a getc(), which
8664 * returns a single char read (or EOF), and prefills the buffer, and also
8665 * allows us to find out how full the buffer is. We use this information to
8666 * SvGROW() the sv to the size remaining in the buffer, after which we copy
8667 * the returned single char into the target sv, and then go back into scan
8670 * 5. If we run out of write-buffer then we SvGROW() it by the size of the
8671 * remaining space in the read-buffer.
8673 * Note that this code despite its twisty-turny nature is pretty darn slick.
8674 * It manages single byte separators, multi-byte cross boundary separators,
8675 * and cross-read-buffer separators cleanly and efficiently at the cost
8676 * of potentially greatly overallocating the target SV.
8682 /* get the number of bytes remaining in the read-ahead buffer
8683 * on first call on a given fp this will return 0.*/
8684 cnt = PerlIO_get_cnt(fp);
8686 /* make sure we have the room */
8687 if ((I32)(SvLEN(sv) - append) <= cnt + 1) {
8688 /* Not room for all of it
8689 if we are looking for a separator and room for some
8691 if (rslen && cnt > 80 && (I32)SvLEN(sv) > append) {
8692 /* just process what we have room for */
8693 shortbuffered = cnt - SvLEN(sv) + append + 1;
8694 cnt -= shortbuffered;
8697 /* ensure that the target sv has enough room to hold
8698 * the rest of the read-ahead buffer */
8700 /* remember that cnt can be negative */
8701 SvGROW(sv, (STRLEN)(append + (cnt <= 0 ? 2 : (cnt + 1))));
8705 /* we have enough room to hold the full buffer, lets scream */
8709 /* extract the pointer to sv's string buffer, offset by append as necessary */
8710 bp = (STDCHAR*)SvPVX_const(sv) + append; /* move these two too to registers */
8711 /* extract the point to the read-ahead buffer */
8712 ptr = (STDCHAR*)PerlIO_get_ptr(fp);
8714 /* some trace debug output */
8715 DEBUG_P(PerlIO_printf(Perl_debug_log,
8716 "Screamer: entering, ptr=%" UVuf ", cnt=%ld\n",PTR2UV(ptr),(long)cnt));
8717 DEBUG_P(PerlIO_printf(Perl_debug_log,
8718 "Screamer: entering: PerlIO * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%"
8720 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8721 PTR2UV(PerlIO_has_base(fp) ? PerlIO_get_base(fp) : 0)));
8725 /* if there is stuff left in the read-ahead buffer */
8727 /* if there is a separator */
8729 /* find next rslast */
8732 /* shortcut common case of blank line */
8734 if ((*bp++ = *ptr++) == rslast)
8735 goto thats_all_folks;
8737 p = (STDCHAR *)memchr(ptr, rslast, cnt);
8739 SSize_t got = p - ptr + 1;
8740 Copy(ptr, bp, got, STDCHAR);
8744 goto thats_all_folks;
8746 Copy(ptr, bp, cnt, STDCHAR);
8752 /* no separator, slurp the full buffer */
8753 Copy(ptr, bp, cnt, char); /* this | eat */
8754 bp += cnt; /* screams | dust */
8755 ptr += cnt; /* louder | sed :-) */
8757 assert (!shortbuffered);
8758 goto cannot_be_shortbuffered;
8762 if (shortbuffered) { /* oh well, must extend */
8763 /* we didnt have enough room to fit the line into the target buffer
8764 * so we must extend the target buffer and keep going */
8765 cnt = shortbuffered;
8767 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8769 /* extned the target sv's buffer so it can hold the full read-ahead buffer */
8770 SvGROW(sv, SvLEN(sv) + append + cnt + 2);
8771 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8775 cannot_be_shortbuffered:
8776 /* we need to refill the read-ahead buffer if possible */
8778 DEBUG_P(PerlIO_printf(Perl_debug_log,
8779 "Screamer: going to getc, ptr=%" UVuf ", cnt=%" IVdf "\n",
8780 PTR2UV(ptr),(IV)cnt));
8781 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* deregisterize cnt and ptr */
8783 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8784 "Screamer: pre: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf "\n",
8785 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8786 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8789 call PerlIO_getc() to let it prefill the lookahead buffer
8791 This used to call 'filbuf' in stdio form, but as that behaves like
8792 getc when cnt <= 0 we use PerlIO_getc here to avoid introducing
8793 another abstraction.
8795 Note we have to deal with the char in 'i' if we are not at EOF
8797 bpx = bp - (STDCHAR*)SvPVX_const(sv);
8798 /* signals might be called here, possibly modifying sv */
8799 i = PerlIO_getc(fp); /* get more characters */
8800 bp = (STDCHAR*)SvPVX_const(sv) + bpx;
8802 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8803 "Screamer: post: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf "\n",
8804 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8805 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8807 /* find out how much is left in the read-ahead buffer, and rextract its pointer */
8808 cnt = PerlIO_get_cnt(fp);
8809 ptr = (STDCHAR*)PerlIO_get_ptr(fp); /* reregisterize cnt and ptr */
8810 DEBUG_P(PerlIO_printf(Perl_debug_log,
8811 "Screamer: after getc, ptr=%" UVuf ", cnt=%" IVdf "\n",
8812 PTR2UV(ptr),(IV)cnt));
8814 if (i == EOF) /* all done for ever? */
8815 goto thats_really_all_folks;
8817 /* make sure we have enough space in the target sv */
8818 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8820 SvGROW(sv, bpx + cnt + 2);
8821 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8823 /* copy of the char we got from getc() */
8824 *bp++ = (STDCHAR)i; /* store character from PerlIO_getc */
8826 /* make sure we deal with the i being the last character of a separator */
8827 if (rslen && (STDCHAR)i == rslast) /* all done for now? */
8828 goto thats_all_folks;
8832 /* check if we have actually found the separator - only really applies
8834 if ((rslen > 1 && (STRLEN)(bp - (STDCHAR*)SvPVX_const(sv)) < rslen) ||
8835 memNE((char*)bp - rslen, rsptr, rslen))
8836 goto screamer; /* go back to the fray */
8837 thats_really_all_folks:
8839 cnt += shortbuffered;
8840 DEBUG_P(PerlIO_printf(Perl_debug_log,
8841 "Screamer: quitting, ptr=%" UVuf ", cnt=%" IVdf "\n",PTR2UV(ptr),(IV)cnt));
8842 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* put these back or we're in trouble */
8843 DEBUG_P(PerlIO_printf(Perl_debug_log,
8844 "Screamer: end: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf
8846 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8847 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8849 SvCUR_set(sv, bp - (STDCHAR*)SvPVX_const(sv)); /* set length */
8850 DEBUG_P(PerlIO_printf(Perl_debug_log,
8851 "Screamer: done, len=%ld, string=|%.*s|\n",
8852 (long)SvCUR(sv),(int)SvCUR(sv),SvPVX_const(sv)));
8856 /*The big, slow, and stupid way. */
8857 #ifdef USE_HEAP_INSTEAD_OF_STACK /* Even slower way. */
8858 STDCHAR *buf = NULL;
8859 Newx(buf, 8192, STDCHAR);
8867 const STDCHAR * const bpe = buf + sizeof(buf);
8869 while ((i = PerlIO_getc(fp)) != EOF && (*bp++ = (STDCHAR)i) != rslast && bp < bpe)
8870 ; /* keep reading */
8874 cnt = PerlIO_read(fp,(char*)buf, sizeof(buf));
8875 /* Accommodate broken VAXC compiler, which applies U8 cast to
8876 * both args of ?: operator, causing EOF to change into 255
8879 i = (U8)buf[cnt - 1];
8885 cnt = 0; /* we do need to re-set the sv even when cnt <= 0 */
8887 sv_catpvn_nomg(sv, (char *) buf, cnt);
8889 sv_setpvn(sv, (char *) buf, cnt); /* "nomg" is implied */
8891 if (i != EOF && /* joy */
8893 SvCUR(sv) < rslen ||
8894 memNE(SvPVX_const(sv) + SvCUR(sv) - rslen, rsptr, rslen)))
8898 * If we're reading from a TTY and we get a short read,
8899 * indicating that the user hit his EOF character, we need
8900 * to notice it now, because if we try to read from the TTY
8901 * again, the EOF condition will disappear.
8903 * The comparison of cnt to sizeof(buf) is an optimization
8904 * that prevents unnecessary calls to feof().
8908 if (!(cnt < (I32)sizeof(buf) && PerlIO_eof(fp)))
8912 #ifdef USE_HEAP_INSTEAD_OF_STACK
8917 if (rspara) { /* have to do this both before and after */
8918 while (i != EOF) { /* to make sure file boundaries work right */
8919 i = PerlIO_getc(fp);
8921 PerlIO_ungetc(fp,i);
8927 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8933 Auto-increment of the value in the SV, doing string to numeric conversion
8934 if necessary. Handles 'get' magic and operator overloading.
8940 Perl_sv_inc(pTHX_ SV *const sv)
8949 =for apidoc sv_inc_nomg
8951 Auto-increment of the value in the SV, doing string to numeric conversion
8952 if necessary. Handles operator overloading. Skips handling 'get' magic.
8958 Perl_sv_inc_nomg(pTHX_ SV *const sv)
8965 if (SvTHINKFIRST(sv)) {
8966 if (SvREADONLY(sv)) {
8967 Perl_croak_no_modify();
8971 if (SvAMAGIC(sv) && AMG_CALLunary(sv, inc_amg))
8973 i = PTR2IV(SvRV(sv));
8977 else sv_force_normal_flags(sv, 0);
8979 flags = SvFLAGS(sv);
8980 if ((flags & (SVp_NOK|SVp_IOK)) == SVp_NOK) {
8981 /* It's (privately or publicly) a float, but not tested as an
8982 integer, so test it to see. */
8984 flags = SvFLAGS(sv);
8986 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8987 /* It's publicly an integer, or privately an integer-not-float */
8988 #ifdef PERL_PRESERVE_IVUV
8992 if (SvUVX(sv) == UV_MAX)
8993 sv_setnv(sv, UV_MAX_P1);
8995 (void)SvIOK_only_UV(sv);
8996 SvUV_set(sv, SvUVX(sv) + 1);
8998 if (SvIVX(sv) == IV_MAX)
8999 sv_setuv(sv, (UV)IV_MAX + 1);
9001 (void)SvIOK_only(sv);
9002 SvIV_set(sv, SvIVX(sv) + 1);
9007 if (flags & SVp_NOK) {
9008 const NV was = SvNVX(sv);
9009 if (LIKELY(!Perl_isinfnan(was)) &&
9010 NV_OVERFLOWS_INTEGERS_AT != 0.0 &&
9011 was >= NV_OVERFLOWS_INTEGERS_AT) {
9012 /* diag_listed_as: Lost precision when %s %f by 1 */
9013 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
9014 "Lost precision when incrementing %" NVff " by 1",
9017 (void)SvNOK_only(sv);
9018 SvNV_set(sv, was + 1.0);
9022 /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */
9023 if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv)))
9024 Perl_croak_no_modify();
9026 if (!(flags & SVp_POK) || !*SvPVX_const(sv)) {
9027 if ((flags & SVTYPEMASK) < SVt_PVIV)
9028 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV ? SVt_PVIV : SVt_IV));
9029 (void)SvIOK_only(sv);
9034 while (isALPHA(*d)) d++;
9035 while (isDIGIT(*d)) d++;
9036 if (d < SvEND(sv)) {
9037 const int numtype = grok_number_flags(SvPVX_const(sv), SvCUR(sv), NULL, PERL_SCAN_TRAILING);
9038 #ifdef PERL_PRESERVE_IVUV
9039 /* Got to punt this as an integer if needs be, but we don't issue
9040 warnings. Probably ought to make the sv_iv_please() that does
9041 the conversion if possible, and silently. */
9042 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
9043 /* Need to try really hard to see if it's an integer.
9044 9.22337203685478e+18 is an integer.
9045 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
9046 so $a="9.22337203685478e+18"; $a+0; $a++
9047 needs to be the same as $a="9.22337203685478e+18"; $a++
9054 /* sv_2iv *should* have made this an NV */
9055 if (flags & SVp_NOK) {
9056 (void)SvNOK_only(sv);
9057 SvNV_set(sv, SvNVX(sv) + 1.0);
9060 /* I don't think we can get here. Maybe I should assert this
9061 And if we do get here I suspect that sv_setnv will croak. NWC
9063 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_inc punt failed to convert '%s' to IOK or NOKp, UV=0x%" UVxf " NV=%" NVgf "\n",
9064 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
9066 #endif /* PERL_PRESERVE_IVUV */
9067 if (!numtype && ckWARN(WARN_NUMERIC))
9068 not_incrementable(sv);
9069 sv_setnv(sv,Atof(SvPVX_const(sv)) + 1.0);
9073 while (d >= SvPVX_const(sv)) {
9081 /* MKS: The original code here died if letters weren't consecutive.
9082 * at least it didn't have to worry about non-C locales. The
9083 * new code assumes that ('z'-'a')==('Z'-'A'), letters are
9084 * arranged in order (although not consecutively) and that only
9085 * [A-Za-z] are accepted by isALPHA in the C locale.
9087 if (isALPHA_FOLD_NE(*d, 'z')) {
9088 do { ++*d; } while (!isALPHA(*d));
9091 *(d--) -= 'z' - 'a';
9096 *(d--) -= 'z' - 'a' + 1;
9100 /* oh,oh, the number grew */
9101 SvGROW(sv, SvCUR(sv) + 2);
9102 SvCUR_set(sv, SvCUR(sv) + 1);
9103 for (d = SvPVX(sv) + SvCUR(sv); d > SvPVX_const(sv); d--)
9114 Auto-decrement of the value in the SV, doing string to numeric conversion
9115 if necessary. Handles 'get' magic and operator overloading.
9121 Perl_sv_dec(pTHX_ SV *const sv)
9130 =for apidoc sv_dec_nomg
9132 Auto-decrement of the value in the SV, doing string to numeric conversion
9133 if necessary. Handles operator overloading. Skips handling 'get' magic.
9139 Perl_sv_dec_nomg(pTHX_ SV *const sv)
9145 if (SvTHINKFIRST(sv)) {
9146 if (SvREADONLY(sv)) {
9147 Perl_croak_no_modify();
9151 if (SvAMAGIC(sv) && AMG_CALLunary(sv, dec_amg))
9153 i = PTR2IV(SvRV(sv));
9157 else sv_force_normal_flags(sv, 0);
9159 /* Unlike sv_inc we don't have to worry about string-never-numbers
9160 and keeping them magic. But we mustn't warn on punting */
9161 flags = SvFLAGS(sv);
9162 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
9163 /* It's publicly an integer, or privately an integer-not-float */
9164 #ifdef PERL_PRESERVE_IVUV
9168 if (SvUVX(sv) == 0) {
9169 (void)SvIOK_only(sv);
9173 (void)SvIOK_only_UV(sv);
9174 SvUV_set(sv, SvUVX(sv) - 1);
9177 if (SvIVX(sv) == IV_MIN) {
9178 sv_setnv(sv, (NV)IV_MIN);
9182 (void)SvIOK_only(sv);
9183 SvIV_set(sv, SvIVX(sv) - 1);
9188 if (flags & SVp_NOK) {
9191 const NV was = SvNVX(sv);
9192 if (LIKELY(!Perl_isinfnan(was)) &&
9193 NV_OVERFLOWS_INTEGERS_AT != 0.0 &&
9194 was <= -NV_OVERFLOWS_INTEGERS_AT) {
9195 /* diag_listed_as: Lost precision when %s %f by 1 */
9196 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
9197 "Lost precision when decrementing %" NVff " by 1",
9200 (void)SvNOK_only(sv);
9201 SvNV_set(sv, was - 1.0);
9206 /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */
9207 if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv)))
9208 Perl_croak_no_modify();
9210 if (!(flags & SVp_POK)) {
9211 if ((flags & SVTYPEMASK) < SVt_PVIV)
9212 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV) ? SVt_PVIV : SVt_IV);
9214 (void)SvIOK_only(sv);
9217 #ifdef PERL_PRESERVE_IVUV
9219 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), NULL);
9220 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
9221 /* Need to try really hard to see if it's an integer.
9222 9.22337203685478e+18 is an integer.
9223 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
9224 so $a="9.22337203685478e+18"; $a+0; $a--
9225 needs to be the same as $a="9.22337203685478e+18"; $a--
9232 /* sv_2iv *should* have made this an NV */
9233 if (flags & SVp_NOK) {
9234 (void)SvNOK_only(sv);
9235 SvNV_set(sv, SvNVX(sv) - 1.0);
9238 /* I don't think we can get here. Maybe I should assert this
9239 And if we do get here I suspect that sv_setnv will croak. NWC
9241 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_dec punt failed to convert '%s' to IOK or NOKp, UV=0x%" UVxf " NV=%" NVgf "\n",
9242 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
9245 #endif /* PERL_PRESERVE_IVUV */
9246 sv_setnv(sv,Atof(SvPVX_const(sv)) - 1.0); /* punt */
9249 /* this define is used to eliminate a chunk of duplicated but shared logic
9250 * it has the suffix __SV_C to signal that it isnt API, and isnt meant to be
9251 * used anywhere but here - yves
9253 #define PUSH_EXTEND_MORTAL__SV_C(AnSv) \
9255 SSize_t ix = ++PL_tmps_ix; \
9256 if (UNLIKELY(ix >= PL_tmps_max)) \
9257 ix = tmps_grow_p(ix); \
9258 PL_tmps_stack[ix] = (AnSv); \
9262 =for apidoc sv_mortalcopy
9264 Creates a new SV which is a copy of the original SV (using C<sv_setsv>).
9265 The new SV is marked as mortal. It will be destroyed "soon", either by an
9266 explicit call to C<FREETMPS>, or by an implicit call at places such as
9267 statement boundaries. See also C<L</sv_newmortal>> and C<L</sv_2mortal>>.
9269 =for apidoc sv_mortalcopy_flags
9271 Like C<sv_mortalcopy>, but the extra C<flags> are passed to the
9277 /* Make a string that will exist for the duration of the expression
9278 * evaluation. Actually, it may have to last longer than that, but
9279 * hopefully we won't free it until it has been assigned to a
9280 * permanent location. */
9283 Perl_sv_mortalcopy_flags(pTHX_ SV *const oldstr, U32 flags)
9287 if (flags & SV_GMAGIC)
9288 SvGETMAGIC(oldstr); /* before new_SV, in case it dies */
9290 sv_setsv_flags(sv,oldstr,flags & ~SV_GMAGIC);
9291 PUSH_EXTEND_MORTAL__SV_C(sv);
9297 =for apidoc sv_newmortal
9299 Creates a new null SV which is mortal. The reference count of the SV is
9300 set to 1. It will be destroyed "soon", either by an explicit call to
9301 C<FREETMPS>, or by an implicit call at places such as statement boundaries.
9302 See also C<L</sv_mortalcopy>> and C<L</sv_2mortal>>.
9308 Perl_sv_newmortal(pTHX)
9313 SvFLAGS(sv) = SVs_TEMP;
9314 PUSH_EXTEND_MORTAL__SV_C(sv);
9320 =for apidoc newSVpvn_flags
9322 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9323 characters) into it. The reference count for the
9324 SV is set to 1. Note that if C<len> is zero, Perl will create a zero length
9325 string. You are responsible for ensuring that the source string is at least
9326 C<len> bytes long. If the C<s> argument is NULL the new SV will be undefined.
9327 Currently the only flag bits accepted are C<SVf_UTF8> and C<SVs_TEMP>.
9328 If C<SVs_TEMP> is set, then C<sv_2mortal()> is called on the result before
9329 returning. If C<SVf_UTF8> is set, C<s>
9330 is considered to be in UTF-8 and the
9331 C<SVf_UTF8> flag will be set on the new SV.
9332 C<newSVpvn_utf8()> is a convenience wrapper for this function, defined as
9334 #define newSVpvn_utf8(s, len, u) \
9335 newSVpvn_flags((s), (len), (u) ? SVf_UTF8 : 0)
9337 =for apidoc Amnh||SVf_UTF8
9338 =for apidoc Amnh||SVs_TEMP
9344 Perl_newSVpvn_flags(pTHX_ const char *const s, const STRLEN len, const U32 flags)
9348 /* All the flags we don't support must be zero.
9349 And we're new code so I'm going to assert this from the start. */
9350 assert(!(flags & ~(SVf_UTF8|SVs_TEMP)));
9352 sv_setpvn(sv,s,len);
9354 /* This code used to do a sv_2mortal(), however we now unroll the call to
9355 * sv_2mortal() and do what it does ourselves here. Since we have asserted
9356 * that flags can only have the SVf_UTF8 and/or SVs_TEMP flags set above we
9357 * can use it to enable the sv flags directly (bypassing SvTEMP_on), which
9358 * in turn means we dont need to mask out the SVf_UTF8 flag below, which
9359 * means that we eliminate quite a few steps than it looks - Yves
9360 * (explaining patch by gfx) */
9362 SvFLAGS(sv) |= flags;
9364 if(flags & SVs_TEMP){
9365 PUSH_EXTEND_MORTAL__SV_C(sv);
9372 =for apidoc sv_2mortal
9374 Marks an existing SV as mortal. The SV will be destroyed "soon", either
9375 by an explicit call to C<FREETMPS>, or by an implicit call at places such as
9376 statement boundaries. C<SvTEMP()> is turned on which means that the SV's
9377 string buffer can be "stolen" if this SV is copied. See also
9378 C<L</sv_newmortal>> and C<L</sv_mortalcopy>>.
9384 Perl_sv_2mortal(pTHX_ SV *const sv)
9391 PUSH_EXTEND_MORTAL__SV_C(sv);
9399 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9400 characters) into it. The reference count for the
9401 SV is set to 1. If C<len> is zero, Perl will compute the length using
9402 C<strlen()>, (which means if you use this option, that C<s> can't have embedded
9403 C<NUL> characters and has to have a terminating C<NUL> byte).
9405 This function can cause reliability issues if you are likely to pass in
9406 empty strings that are not null terminated, because it will run
9407 strlen on the string and potentially run past valid memory.
9409 Using L</newSVpvn> is a safer alternative for non C<NUL> terminated strings.
9410 For string literals use L</newSVpvs> instead. This function will work fine for
9411 C<NUL> terminated strings, but if you want to avoid the if statement on whether
9412 to call C<strlen> use C<newSVpvn> instead (calling C<strlen> yourself).
9418 Perl_newSVpv(pTHX_ const char *const s, const STRLEN len)
9423 sv_setpvn(sv, s, len || s == NULL ? len : strlen(s));
9428 =for apidoc newSVpvn
9430 Creates a new SV and copies a string into it, which may contain C<NUL> characters
9431 (C<\0>) and other binary data. The reference count for the SV is set to 1.
9432 Note that if C<len> is zero, Perl will create a zero length (Perl) string. You
9433 are responsible for ensuring that the source buffer is at least
9434 C<len> bytes long. If the C<buffer> argument is NULL the new SV will be
9441 Perl_newSVpvn(pTHX_ const char *const buffer, const STRLEN len)
9445 sv_setpvn(sv,buffer,len);
9450 =for apidoc newSVhek
9452 Creates a new SV from the hash key structure. It will generate scalars that
9453 point to the shared string table where possible. Returns a new (undefined)
9454 SV if C<hek> is NULL.
9460 Perl_newSVhek(pTHX_ const HEK *const hek)
9469 if (HEK_LEN(hek) == HEf_SVKEY) {
9470 return newSVsv(*(SV**)HEK_KEY(hek));
9472 const int flags = HEK_FLAGS(hek);
9473 if (flags & HVhek_WASUTF8) {
9475 Andreas would like keys he put in as utf8 to come back as utf8
9477 STRLEN utf8_len = HEK_LEN(hek);
9478 SV * const sv = newSV_type(SVt_PV);
9479 char *as_utf8 = (char *)bytes_to_utf8 ((U8*)HEK_KEY(hek), &utf8_len);
9480 /* bytes_to_utf8() allocates a new string, which we can repurpose: */
9481 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
9484 } else if (flags & HVhek_UNSHARED) {
9485 /* A hash that isn't using shared hash keys has to have
9486 the flag in every key so that we know not to try to call
9487 share_hek_hek on it. */
9489 SV * const sv = newSVpvn (HEK_KEY(hek), HEK_LEN(hek));
9494 /* This will be overwhelminly the most common case. */
9496 /* Inline most of newSVpvn_share(), because share_hek_hek() is far
9497 more efficient than sharepvn(). */
9501 sv_upgrade(sv, SVt_PV);
9502 SvPV_set(sv, (char *)HEK_KEY(share_hek_hek(hek)));
9503 SvCUR_set(sv, HEK_LEN(hek));
9515 =for apidoc newSVpvn_share
9517 Creates a new SV with its C<SvPVX_const> pointing to a shared string in the string
9518 table. If the string does not already exist in the table, it is
9519 created first. Turns on the C<SvIsCOW> flag (or C<READONLY>
9520 and C<FAKE> in 5.16 and earlier). If the C<hash> parameter
9521 is non-zero, that value is used; otherwise the hash is computed.
9522 The string's hash can later be retrieved from the SV
9523 with the C<SvSHARED_HASH()> macro. The idea here is
9524 that as the string table is used for shared hash keys these strings will have
9525 C<SvPVX_const == HeKEY> and hash lookup will avoid string compare.
9531 Perl_newSVpvn_share(pTHX_ const char *src, I32 len, U32 hash)
9535 bool is_utf8 = FALSE;
9536 const char *const orig_src = src;
9539 STRLEN tmplen = -len;
9541 /* See the note in hv.c:hv_fetch() --jhi */
9542 src = (char*)bytes_from_utf8((const U8*)src, &tmplen, &is_utf8);
9546 PERL_HASH(hash, src, len);
9548 /* The logic for this is inlined in S_mro_get_linear_isa_dfs(), so if it
9549 changes here, update it there too. */
9550 sv_upgrade(sv, SVt_PV);
9551 SvPV_set(sv, sharepvn(src, is_utf8?-len:len, hash));
9558 if (src != orig_src)
9564 =for apidoc newSVpv_share
9566 Like C<newSVpvn_share>, but takes a C<NUL>-terminated string instead of a
9573 Perl_newSVpv_share(pTHX_ const char *src, U32 hash)
9575 return newSVpvn_share(src, strlen(src), hash);
9578 #if defined(PERL_IMPLICIT_CONTEXT)
9580 /* pTHX_ magic can't cope with varargs, so this is a no-context
9581 * version of the main function, (which may itself be aliased to us).
9582 * Don't access this version directly.
9586 Perl_newSVpvf_nocontext(const char *const pat, ...)
9592 PERL_ARGS_ASSERT_NEWSVPVF_NOCONTEXT;
9594 va_start(args, pat);
9595 sv = vnewSVpvf(pat, &args);
9602 =for apidoc newSVpvf
9604 Creates a new SV and initializes it with the string formatted like
9611 Perl_newSVpvf(pTHX_ const char *const pat, ...)
9616 PERL_ARGS_ASSERT_NEWSVPVF;
9618 va_start(args, pat);
9619 sv = vnewSVpvf(pat, &args);
9624 /* backend for newSVpvf() and newSVpvf_nocontext() */
9627 Perl_vnewSVpvf(pTHX_ const char *const pat, va_list *const args)
9631 PERL_ARGS_ASSERT_VNEWSVPVF;
9634 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
9641 Creates a new SV and copies a floating point value into it.
9642 The reference count for the SV is set to 1.
9648 Perl_newSVnv(pTHX_ const NV n)
9660 Creates a new SV and copies an integer into it. The reference count for the
9667 Perl_newSViv(pTHX_ const IV i)
9673 /* Inlining ONLY the small relevant subset of sv_setiv here
9674 * for performance. Makes a significant difference. */
9676 /* We're starting from SVt_FIRST, so provided that's
9677 * actual 0, we don't have to unset any SV type flags
9678 * to promote to SVt_IV. */
9679 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9681 SET_SVANY_FOR_BODYLESS_IV(sv);
9682 SvFLAGS(sv) |= SVt_IV;
9694 Creates a new SV and copies an unsigned integer into it.
9695 The reference count for the SV is set to 1.
9701 Perl_newSVuv(pTHX_ const UV u)
9705 /* Inlining ONLY the small relevant subset of sv_setuv here
9706 * for performance. Makes a significant difference. */
9708 /* Using ivs is more efficient than using uvs - see sv_setuv */
9709 if (u <= (UV)IV_MAX) {
9710 return newSViv((IV)u);
9715 /* We're starting from SVt_FIRST, so provided that's
9716 * actual 0, we don't have to unset any SV type flags
9717 * to promote to SVt_IV. */
9718 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9720 SET_SVANY_FOR_BODYLESS_IV(sv);
9721 SvFLAGS(sv) |= SVt_IV;
9723 (void)SvIsUV_on(sv);
9732 =for apidoc newSV_type
9734 Creates a new SV, of the type specified. The reference count for the new SV
9741 Perl_newSV_type(pTHX_ const svtype type)
9746 ASSUME(SvTYPE(sv) == SVt_FIRST);
9747 if(type != SVt_FIRST)
9748 sv_upgrade(sv, type);
9753 =for apidoc newRV_noinc
9755 Creates an RV wrapper for an SV. The reference count for the original
9756 SV is B<not> incremented.
9762 Perl_newRV_noinc(pTHX_ SV *const tmpRef)
9766 PERL_ARGS_ASSERT_NEWRV_NOINC;
9770 /* We're starting from SVt_FIRST, so provided that's
9771 * actual 0, we don't have to unset any SV type flags
9772 * to promote to SVt_IV. */
9773 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9775 SET_SVANY_FOR_BODYLESS_IV(sv);
9776 SvFLAGS(sv) |= SVt_IV;
9781 SvRV_set(sv, tmpRef);
9786 /* newRV_inc is the official function name to use now.
9787 * newRV_inc is in fact #defined to newRV in sv.h
9791 Perl_newRV(pTHX_ SV *const sv)
9793 PERL_ARGS_ASSERT_NEWRV;
9795 return newRV_noinc(SvREFCNT_inc_simple_NN(sv));
9801 Creates a new SV which is an exact duplicate of the original SV.
9804 =for apidoc newSVsv_nomg
9806 Like C<newSVsv> but does not process get magic.
9812 Perl_newSVsv_flags(pTHX_ SV *const old, I32 flags)
9818 if (SvTYPE(old) == (svtype)SVTYPEMASK) {
9819 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), "semi-panic: attempt to dup freed string");
9822 /* Do this here, otherwise we leak the new SV if this croaks. */
9823 if (flags & SV_GMAGIC)
9826 sv_setsv_flags(sv, old, flags & ~SV_GMAGIC);
9831 =for apidoc sv_reset
9833 Underlying implementation for the C<reset> Perl function.
9834 Note that the perl-level function is vaguely deprecated.
9840 Perl_sv_reset(pTHX_ const char *s, HV *const stash)
9842 PERL_ARGS_ASSERT_SV_RESET;
9844 sv_resetpvn(*s ? s : NULL, strlen(s), stash);
9848 Perl_sv_resetpvn(pTHX_ const char *s, STRLEN len, HV * const stash)
9850 char todo[PERL_UCHAR_MAX+1];
9853 if (!stash || SvTYPE(stash) != SVt_PVHV)
9856 if (!s) { /* reset ?? searches */
9857 MAGIC * const mg = mg_find((const SV *)stash, PERL_MAGIC_symtab);
9859 const U32 count = mg->mg_len / sizeof(PMOP**);
9860 PMOP **pmp = (PMOP**) mg->mg_ptr;
9861 PMOP *const *const end = pmp + count;
9865 SvREADONLY_off(PL_regex_pad[(*pmp)->op_pmoffset]);
9867 (*pmp)->op_pmflags &= ~PMf_USED;
9875 /* reset variables */
9877 if (!HvARRAY(stash))
9880 Zero(todo, 256, char);
9884 I32 i = (unsigned char)*s;
9888 max = (unsigned char)*s++;
9889 for ( ; i <= max; i++) {
9892 for (i = 0; i <= (I32) HvMAX(stash); i++) {
9894 for (entry = HvARRAY(stash)[i];
9896 entry = HeNEXT(entry))
9901 if (!todo[(U8)*HeKEY(entry)])
9903 gv = MUTABLE_GV(HeVAL(entry));
9907 if (sv && !SvREADONLY(sv)) {
9908 SV_CHECK_THINKFIRST_COW_DROP(sv);
9909 if (!isGV(sv)) SvOK_off(sv);
9914 if (GvHV(gv) && !HvNAME_get(GvHV(gv))) {
9925 Using various gambits, try to get an IO from an SV: the IO slot if its a
9926 GV; or the recursive result if we're an RV; or the IO slot of the symbol
9927 named after the PV if we're a string.
9929 'Get' magic is ignored on the C<sv> passed in, but will be called on
9930 C<SvRV(sv)> if C<sv> is an RV.
9936 Perl_sv_2io(pTHX_ SV *const sv)
9941 PERL_ARGS_ASSERT_SV_2IO;
9943 switch (SvTYPE(sv)) {
9945 io = MUTABLE_IO(sv);
9949 if (isGV_with_GP(sv)) {
9950 gv = MUTABLE_GV(sv);
9953 Perl_croak(aTHX_ "Bad filehandle: %" HEKf,
9954 HEKfARG(GvNAME_HEK(gv)));
9960 Perl_croak(aTHX_ PL_no_usym, "filehandle");
9962 SvGETMAGIC(SvRV(sv));
9963 return sv_2io(SvRV(sv));
9965 gv = gv_fetchsv_nomg(sv, 0, SVt_PVIO);
9972 if (SvGMAGICAL(sv)) {
9973 newsv = sv_newmortal();
9974 sv_setsv_nomg(newsv, sv);
9976 Perl_croak(aTHX_ "Bad filehandle: %" SVf, SVfARG(newsv));
9986 Using various gambits, try to get a CV from an SV; in addition, try if
9987 possible to set C<*st> and C<*gvp> to the stash and GV associated with it.
9988 The flags in C<lref> are passed to C<gv_fetchsv>.
9994 Perl_sv_2cv(pTHX_ SV *sv, HV **const st, GV **const gvp, const I32 lref)
9999 PERL_ARGS_ASSERT_SV_2CV;
10006 switch (SvTYPE(sv)) {
10010 return MUTABLE_CV(sv);
10020 sv = amagic_deref_call(sv, to_cv_amg);
10023 if (SvTYPE(sv) == SVt_PVCV) {
10024 cv = MUTABLE_CV(sv);
10029 else if(SvGETMAGIC(sv), isGV_with_GP(sv))
10030 gv = MUTABLE_GV(sv);
10032 Perl_croak(aTHX_ "Not a subroutine reference");
10034 else if (isGV_with_GP(sv)) {
10035 gv = MUTABLE_GV(sv);
10038 gv = gv_fetchsv_nomg(sv, lref, SVt_PVCV);
10045 /* Some flags to gv_fetchsv mean don't really create the GV */
10046 if (!isGV_with_GP(gv)) {
10050 *st = GvESTASH(gv);
10051 if (lref & ~GV_ADDMG && !GvCVu(gv)) {
10052 /* XXX this is probably not what they think they're getting.
10053 * It has the same effect as "sub name;", i.e. just a forward
10062 =for apidoc sv_true
10064 Returns true if the SV has a true value by Perl's rules.
10065 Use the C<SvTRUE> macro instead, which may call C<sv_true()> or may
10066 instead use an in-line version.
10072 Perl_sv_true(pTHX_ SV *const sv)
10077 const XPV* const tXpv = (XPV*)SvANY(sv);
10079 (tXpv->xpv_cur > 1 ||
10080 (tXpv->xpv_cur && *sv->sv_u.svu_pv != '0')))
10087 return SvIVX(sv) != 0;
10090 return SvNVX(sv) != 0.0;
10092 return sv_2bool(sv);
10098 =for apidoc sv_pvn_force
10100 Get a sensible string out of the SV somehow.
10101 A private implementation of the C<SvPV_force> macro for compilers which
10102 can't cope with complex macro expressions. Always use the macro instead.
10104 =for apidoc sv_pvn_force_flags
10106 Get a sensible string out of the SV somehow.
10107 If C<flags> has the C<SV_GMAGIC> bit set, will C<mg_get> on C<sv> if
10108 appropriate, else not. C<sv_pvn_force> and C<sv_pvn_force_nomg> are
10109 implemented in terms of this function.
10110 You normally want to use the various wrapper macros instead: see
10111 C<L</SvPV_force>> and C<L</SvPV_force_nomg>>.
10117 Perl_sv_pvn_force_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
10119 PERL_ARGS_ASSERT_SV_PVN_FORCE_FLAGS;
10121 if (flags & SV_GMAGIC) SvGETMAGIC(sv);
10122 if (SvTHINKFIRST(sv) && (!SvROK(sv) || SvREADONLY(sv)))
10123 sv_force_normal_flags(sv, 0);
10133 if (SvTYPE(sv) > SVt_PVLV
10134 || isGV_with_GP(sv))
10135 /* diag_listed_as: Can't coerce %s to %s in %s */
10136 Perl_croak(aTHX_ "Can't coerce %s to string in %s", sv_reftype(sv,0),
10138 s = sv_2pv_flags(sv, &len, flags &~ SV_GMAGIC);
10145 if (SvTYPE(sv) < SVt_PV ||
10146 s != SvPVX_const(sv)) { /* Almost, but not quite, sv_setpvn() */
10149 SvUPGRADE(sv, SVt_PV); /* Never FALSE */
10150 SvGROW(sv, len + 1);
10151 Move(s,SvPVX(sv),len,char);
10152 SvCUR_set(sv, len);
10153 SvPVX(sv)[len] = '\0';
10156 SvPOK_on(sv); /* validate pointer */
10158 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2pv(%s)\n",
10159 PTR2UV(sv),SvPVX_const(sv)));
10162 (void)SvPOK_only_UTF8(sv);
10163 return SvPVX_mutable(sv);
10167 =for apidoc sv_pvbyten_force
10169 The backend for the C<SvPVbytex_force> macro. Always use the macro
10170 instead. If the SV cannot be downgraded from UTF-8, this croaks.
10176 Perl_sv_pvbyten_force(pTHX_ SV *const sv, STRLEN *const lp)
10178 PERL_ARGS_ASSERT_SV_PVBYTEN_FORCE;
10180 sv_pvn_force(sv,lp);
10181 sv_utf8_downgrade(sv,0);
10187 =for apidoc sv_pvutf8n_force
10189 The backend for the C<SvPVutf8x_force> macro. Always use the macro
10196 Perl_sv_pvutf8n_force(pTHX_ SV *const sv, STRLEN *const lp)
10198 PERL_ARGS_ASSERT_SV_PVUTF8N_FORCE;
10200 sv_pvn_force(sv,0);
10201 sv_utf8_upgrade_nomg(sv);
10207 =for apidoc sv_reftype
10209 Returns a string describing what the SV is a reference to.
10211 If ob is true and the SV is blessed, the string is the class name,
10212 otherwise it is the type of the SV, "SCALAR", "ARRAY" etc.
10218 Perl_sv_reftype(pTHX_ const SV *const sv, const int ob)
10220 PERL_ARGS_ASSERT_SV_REFTYPE;
10221 if (ob && SvOBJECT(sv)) {
10222 return SvPV_nolen_const(sv_ref(NULL, sv, ob));
10225 /* WARNING - There is code, for instance in mg.c, that assumes that
10226 * the only reason that sv_reftype(sv,0) would return a string starting
10227 * with 'L' or 'S' is that it is a LVALUE or a SCALAR.
10228 * Yes this a dodgy way to do type checking, but it saves practically reimplementing
10229 * this routine inside other subs, and it saves time.
10230 * Do not change this assumption without searching for "dodgy type check" in
10233 switch (SvTYPE(sv)) {
10248 case SVt_PVLV: return (char *) (SvROK(sv) ? "REF"
10249 /* tied lvalues should appear to be
10250 * scalars for backwards compatibility */
10251 : (isALPHA_FOLD_EQ(LvTYPE(sv), 't'))
10252 ? "SCALAR" : "LVALUE");
10253 case SVt_PVAV: return "ARRAY";
10254 case SVt_PVHV: return "HASH";
10255 case SVt_PVCV: return "CODE";
10256 case SVt_PVGV: return (char *) (isGV_with_GP(sv)
10257 ? "GLOB" : "SCALAR");
10258 case SVt_PVFM: return "FORMAT";
10259 case SVt_PVIO: return "IO";
10260 case SVt_INVLIST: return "INVLIST";
10261 case SVt_REGEXP: return "REGEXP";
10262 default: return "UNKNOWN";
10270 Returns a SV describing what the SV passed in is a reference to.
10272 dst can be a SV to be set to the description or NULL, in which case a
10273 mortal SV is returned.
10275 If ob is true and the SV is blessed, the description is the class
10276 name, otherwise it is the type of the SV, "SCALAR", "ARRAY" etc.
10282 Perl_sv_ref(pTHX_ SV *dst, const SV *const sv, const int ob)
10284 PERL_ARGS_ASSERT_SV_REF;
10287 dst = sv_newmortal();
10289 if (ob && SvOBJECT(sv)) {
10290 HvNAME_get(SvSTASH(sv))
10291 ? sv_sethek(dst, HvNAME_HEK(SvSTASH(sv)))
10292 : sv_setpvs(dst, "__ANON__");
10295 const char * reftype = sv_reftype(sv, 0);
10296 sv_setpv(dst, reftype);
10302 =for apidoc sv_isobject
10304 Returns a boolean indicating whether the SV is an RV pointing to a blessed
10305 object. If the SV is not an RV, or if the object is not blessed, then this
10312 Perl_sv_isobject(pTHX_ SV *sv)
10328 Returns a boolean indicating whether the SV is blessed into the specified
10331 This does not check for subtypes or method overloading. Use C<sv_isa_sv> to
10332 verify an inheritance relationship in the same way as the C<isa> operator by
10333 respecting any C<isa()> method overloading; or C<sv_derived_from_sv> to test
10334 directly on the actual object type.
10340 Perl_sv_isa(pTHX_ SV *sv, const char *const name)
10342 const char *hvname;
10344 PERL_ARGS_ASSERT_SV_ISA;
10354 hvname = HvNAME_get(SvSTASH(sv));
10358 return strEQ(hvname, name);
10362 =for apidoc newSVrv
10364 Creates a new SV for the existing RV, C<rv>, to point to. If C<rv> is not an
10365 RV then it will be upgraded to one. If C<classname> is non-null then the new
10366 SV will be blessed in the specified package. The new SV is returned and its
10367 reference count is 1. The reference count 1 is owned by C<rv>. See also
10368 newRV_inc() and newRV_noinc() for creating a new RV properly.
10374 Perl_newSVrv(pTHX_ SV *const rv, const char *const classname)
10378 PERL_ARGS_ASSERT_NEWSVRV;
10382 SV_CHECK_THINKFIRST_COW_DROP(rv);
10384 if (UNLIKELY( SvTYPE(rv) >= SVt_PVMG )) {
10385 const U32 refcnt = SvREFCNT(rv);
10389 SvREFCNT(rv) = refcnt;
10391 sv_upgrade(rv, SVt_IV);
10392 } else if (SvROK(rv)) {
10393 SvREFCNT_dec(SvRV(rv));
10395 prepare_SV_for_RV(rv);
10403 HV* const stash = gv_stashpv(classname, GV_ADD);
10404 (void)sv_bless(rv, stash);
10410 Perl_newSVavdefelem(pTHX_ AV *av, SSize_t ix, bool extendible)
10412 SV * const lv = newSV_type(SVt_PVLV);
10413 PERL_ARGS_ASSERT_NEWSVAVDEFELEM;
10415 sv_magic(lv, NULL, PERL_MAGIC_defelem, NULL, 0);
10416 LvTARG(lv) = SvREFCNT_inc_simple_NN(av);
10417 LvSTARGOFF(lv) = ix;
10418 LvTARGLEN(lv) = extendible ? 1 : (STRLEN)UV_MAX;
10423 =for apidoc sv_setref_pv
10425 Copies a pointer into a new SV, optionally blessing the SV. The C<rv>
10426 argument will be upgraded to an RV. That RV will be modified to point to
10427 the new SV. If the C<pv> argument is C<NULL>, then C<PL_sv_undef> will be placed
10428 into the SV. The C<classname> argument indicates the package for the
10429 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10430 will have a reference count of 1, and the RV will be returned.
10432 Do not use with other Perl types such as HV, AV, SV, CV, because those
10433 objects will become corrupted by the pointer copy process.
10435 Note that C<sv_setref_pvn> copies the string while this copies the pointer.
10441 Perl_sv_setref_pv(pTHX_ SV *const rv, const char *const classname, void *const pv)
10443 PERL_ARGS_ASSERT_SV_SETREF_PV;
10450 sv_setiv(newSVrv(rv,classname), PTR2IV(pv));
10455 =for apidoc sv_setref_iv
10457 Copies an integer into a new SV, optionally blessing the SV. The C<rv>
10458 argument will be upgraded to an RV. That RV will be modified to point to
10459 the new SV. The C<classname> argument indicates the package for the
10460 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10461 will have a reference count of 1, and the RV will be returned.
10467 Perl_sv_setref_iv(pTHX_ SV *const rv, const char *const classname, const IV iv)
10469 PERL_ARGS_ASSERT_SV_SETREF_IV;
10471 sv_setiv(newSVrv(rv,classname), iv);
10476 =for apidoc sv_setref_uv
10478 Copies an unsigned integer into a new SV, optionally blessing the SV. The C<rv>
10479 argument will be upgraded to an RV. That RV will be modified to point to
10480 the new SV. The C<classname> argument indicates the package for the
10481 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10482 will have a reference count of 1, and the RV will be returned.
10488 Perl_sv_setref_uv(pTHX_ SV *const rv, const char *const classname, const UV uv)
10490 PERL_ARGS_ASSERT_SV_SETREF_UV;
10492 sv_setuv(newSVrv(rv,classname), uv);
10497 =for apidoc sv_setref_nv
10499 Copies a double into a new SV, optionally blessing the SV. The C<rv>
10500 argument will be upgraded to an RV. That RV will be modified to point to
10501 the new SV. The C<classname> argument indicates the package for the
10502 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10503 will have a reference count of 1, and the RV will be returned.
10509 Perl_sv_setref_nv(pTHX_ SV *const rv, const char *const classname, const NV nv)
10511 PERL_ARGS_ASSERT_SV_SETREF_NV;
10513 sv_setnv(newSVrv(rv,classname), nv);
10518 =for apidoc sv_setref_pvn
10520 Copies a string into a new SV, optionally blessing the SV. The length of the
10521 string must be specified with C<n>. The C<rv> argument will be upgraded to
10522 an RV. That RV will be modified to point to the new SV. The C<classname>
10523 argument indicates the package for the blessing. Set C<classname> to
10524 C<NULL> to avoid the blessing. The new SV will have a reference count
10525 of 1, and the RV will be returned.
10527 Note that C<sv_setref_pv> copies the pointer while this copies the string.
10533 Perl_sv_setref_pvn(pTHX_ SV *const rv, const char *const classname,
10534 const char *const pv, const STRLEN n)
10536 PERL_ARGS_ASSERT_SV_SETREF_PVN;
10538 sv_setpvn(newSVrv(rv,classname), pv, n);
10543 =for apidoc sv_bless
10545 Blesses an SV into a specified package. The SV must be an RV. The package
10546 must be designated by its stash (see C<L</gv_stashpv>>). The reference count
10547 of the SV is unaffected.
10553 Perl_sv_bless(pTHX_ SV *const sv, HV *const stash)
10556 HV *oldstash = NULL;
10558 PERL_ARGS_ASSERT_SV_BLESS;
10562 Perl_croak(aTHX_ "Can't bless non-reference value");
10564 if (SvFLAGS(tmpRef) & (SVs_OBJECT|SVf_READONLY|SVf_PROTECT)) {
10565 if (SvREADONLY(tmpRef))
10566 Perl_croak_no_modify();
10567 if (SvOBJECT(tmpRef)) {
10568 oldstash = SvSTASH(tmpRef);
10571 SvOBJECT_on(tmpRef);
10572 SvUPGRADE(tmpRef, SVt_PVMG);
10573 SvSTASH_set(tmpRef, MUTABLE_HV(SvREFCNT_inc_simple(stash)));
10574 SvREFCNT_dec(oldstash);
10576 if(SvSMAGICAL(tmpRef))
10577 if(mg_find(tmpRef, PERL_MAGIC_ext) || mg_find(tmpRef, PERL_MAGIC_uvar))
10585 /* Downgrades a PVGV to a PVMG. If it's actually a PVLV, we leave the type
10586 * as it is after unglobbing it.
10589 PERL_STATIC_INLINE void
10590 S_sv_unglob(pTHX_ SV *const sv, U32 flags)
10594 SV * const temp = flags & SV_COW_DROP_PV ? NULL : sv_newmortal();
10596 PERL_ARGS_ASSERT_SV_UNGLOB;
10598 assert(SvTYPE(sv) == SVt_PVGV || SvTYPE(sv) == SVt_PVLV);
10600 if (!(flags & SV_COW_DROP_PV))
10601 gv_efullname3(temp, MUTABLE_GV(sv), "*");
10603 SvREFCNT_inc_simple_void_NN(sv_2mortal(sv));
10605 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
10606 && HvNAME_get(stash))
10607 mro_method_changed_in(stash);
10608 gp_free(MUTABLE_GV(sv));
10611 sv_del_backref(MUTABLE_SV(GvSTASH(sv)), sv);
10612 GvSTASH(sv) = NULL;
10615 if (GvNAME_HEK(sv)) {
10616 unshare_hek(GvNAME_HEK(sv));
10618 isGV_with_GP_off(sv);
10620 if(SvTYPE(sv) == SVt_PVGV) {
10621 /* need to keep SvANY(sv) in the right arena */
10622 xpvmg = new_XPVMG();
10623 StructCopy(SvANY(sv), xpvmg, XPVMG);
10624 del_XPVGV(SvANY(sv));
10627 SvFLAGS(sv) &= ~SVTYPEMASK;
10628 SvFLAGS(sv) |= SVt_PVMG;
10631 /* Intentionally not calling any local SET magic, as this isn't so much a
10632 set operation as merely an internal storage change. */
10633 if (flags & SV_COW_DROP_PV) SvOK_off(sv);
10634 else sv_setsv_flags(sv, temp, 0);
10636 if ((const GV *)sv == PL_last_in_gv)
10637 PL_last_in_gv = NULL;
10638 else if ((const GV *)sv == PL_statgv)
10643 =for apidoc sv_unref_flags
10645 Unsets the RV status of the SV, and decrements the reference count of
10646 whatever was being referenced by the RV. This can almost be thought of
10647 as a reversal of C<newSVrv>. The C<cflags> argument can contain
10648 C<SV_IMMEDIATE_UNREF> to force the reference count to be decremented
10649 (otherwise the decrementing is conditional on the reference count being
10650 different from one or the reference being a readonly SV).
10651 See C<L</SvROK_off>>.
10653 =for apidoc Amnh||SV_IMMEDIATE_UNREF
10659 Perl_sv_unref_flags(pTHX_ SV *const ref, const U32 flags)
10661 SV* const target = SvRV(ref);
10663 PERL_ARGS_ASSERT_SV_UNREF_FLAGS;
10665 if (SvWEAKREF(ref)) {
10666 sv_del_backref(target, ref);
10667 SvWEAKREF_off(ref);
10668 SvRV_set(ref, NULL);
10671 SvRV_set(ref, NULL);
10673 /* You can't have a || SvREADONLY(target) here, as $a = $$a, where $a was
10674 assigned to as BEGIN {$a = \"Foo"} will fail. */
10675 if (SvREFCNT(target) != 1 || (flags & SV_IMMEDIATE_UNREF))
10676 SvREFCNT_dec_NN(target);
10677 else /* XXX Hack, but hard to make $a=$a->[1] work otherwise */
10678 sv_2mortal(target); /* Schedule for freeing later */
10682 =for apidoc sv_untaint
10684 Untaint an SV. Use C<SvTAINTED_off> instead.
10690 Perl_sv_untaint(pTHX_ SV *const sv)
10692 PERL_ARGS_ASSERT_SV_UNTAINT;
10693 PERL_UNUSED_CONTEXT;
10695 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10696 MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10703 =for apidoc sv_tainted
10705 Test an SV for taintedness. Use C<SvTAINTED> instead.
10711 Perl_sv_tainted(pTHX_ SV *const sv)
10713 PERL_ARGS_ASSERT_SV_TAINTED;
10714 PERL_UNUSED_CONTEXT;
10716 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10717 const MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10718 if (mg && (mg->mg_len & 1) )
10724 #ifndef NO_MATHOMS /* Can't move these to mathoms.c because call uiv_2buf(),
10725 private to this file */
10728 =for apidoc sv_setpviv
10730 Copies an integer into the given SV, also updating its string value.
10731 Does not handle 'set' magic. See C<L</sv_setpviv_mg>>.
10737 Perl_sv_setpviv(pTHX_ SV *const sv, const IV iv)
10739 /* The purpose of this union is to ensure that arr is aligned on
10740 a 2 byte boundary, because that is what uiv_2buf() requires */
10742 char arr[TYPE_CHARS(UV)];
10746 char * const ptr = uiv_2buf(buf.arr, iv, 0, 0, &ebuf);
10748 PERL_ARGS_ASSERT_SV_SETPVIV;
10750 sv_setpvn(sv, ptr, ebuf - ptr);
10754 =for apidoc sv_setpviv_mg
10756 Like C<sv_setpviv>, but also handles 'set' magic.
10762 Perl_sv_setpviv_mg(pTHX_ SV *const sv, const IV iv)
10764 PERL_ARGS_ASSERT_SV_SETPVIV_MG;
10766 GCC_DIAG_IGNORE_STMT(-Wdeprecated-declarations);
10768 sv_setpviv(sv, iv);
10770 GCC_DIAG_RESTORE_STMT;
10775 #endif /* NO_MATHOMS */
10777 #if defined(PERL_IMPLICIT_CONTEXT)
10779 /* pTHX_ magic can't cope with varargs, so this is a no-context
10780 * version of the main function, (which may itself be aliased to us).
10781 * Don't access this version directly.
10785 Perl_sv_setpvf_nocontext(SV *const sv, const char *const pat, ...)
10790 PERL_ARGS_ASSERT_SV_SETPVF_NOCONTEXT;
10792 va_start(args, pat);
10793 sv_vsetpvf(sv, pat, &args);
10797 /* pTHX_ magic can't cope with varargs, so this is a no-context
10798 * version of the main function, (which may itself be aliased to us).
10799 * Don't access this version directly.
10803 Perl_sv_setpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10808 PERL_ARGS_ASSERT_SV_SETPVF_MG_NOCONTEXT;
10810 va_start(args, pat);
10811 sv_vsetpvf_mg(sv, pat, &args);
10817 =for apidoc sv_setpvf
10819 Works like C<sv_catpvf> but copies the text into the SV instead of
10820 appending it. Does not handle 'set' magic. See C<L</sv_setpvf_mg>>.
10826 Perl_sv_setpvf(pTHX_ SV *const sv, const char *const pat, ...)
10830 PERL_ARGS_ASSERT_SV_SETPVF;
10832 va_start(args, pat);
10833 sv_vsetpvf(sv, pat, &args);
10838 =for apidoc sv_vsetpvf
10840 Works like C<sv_vcatpvf> but copies the text into the SV instead of
10841 appending it. Does not handle 'set' magic. See C<L</sv_vsetpvf_mg>>.
10843 Usually used via its frontend C<sv_setpvf>.
10849 Perl_sv_vsetpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10851 PERL_ARGS_ASSERT_SV_VSETPVF;
10853 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10857 =for apidoc sv_setpvf_mg
10859 Like C<sv_setpvf>, but also handles 'set' magic.
10865 Perl_sv_setpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10869 PERL_ARGS_ASSERT_SV_SETPVF_MG;
10871 va_start(args, pat);
10872 sv_vsetpvf_mg(sv, pat, &args);
10877 =for apidoc sv_vsetpvf_mg
10879 Like C<sv_vsetpvf>, but also handles 'set' magic.
10881 Usually used via its frontend C<sv_setpvf_mg>.
10887 Perl_sv_vsetpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10889 PERL_ARGS_ASSERT_SV_VSETPVF_MG;
10891 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10895 #if defined(PERL_IMPLICIT_CONTEXT)
10897 /* pTHX_ magic can't cope with varargs, so this is a no-context
10898 * version of the main function, (which may itself be aliased to us).
10899 * Don't access this version directly.
10903 Perl_sv_catpvf_nocontext(SV *const sv, const char *const pat, ...)
10908 PERL_ARGS_ASSERT_SV_CATPVF_NOCONTEXT;
10910 va_start(args, pat);
10911 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10915 /* pTHX_ magic can't cope with varargs, so this is a no-context
10916 * version of the main function, (which may itself be aliased to us).
10917 * Don't access this version directly.
10921 Perl_sv_catpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10926 PERL_ARGS_ASSERT_SV_CATPVF_MG_NOCONTEXT;
10928 va_start(args, pat);
10929 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10936 =for apidoc sv_catpvf
10938 Processes its arguments like C<sprintf>, and appends the formatted
10939 output to an SV. As with C<sv_vcatpvfn> called with a non-null C-style
10940 variable argument list, argument reordering is not supported.
10941 If the appended data contains "wide" characters
10942 (including, but not limited to, SVs with a UTF-8 PV formatted with C<%s>,
10943 and characters >255 formatted with C<%c>), the original SV might get
10944 upgraded to UTF-8. Handles 'get' magic, but not 'set' magic. See
10945 C<L</sv_catpvf_mg>>. If the original SV was UTF-8, the pattern should be
10946 valid UTF-8; if the original SV was bytes, the pattern should be too.
10951 Perl_sv_catpvf(pTHX_ SV *const sv, const char *const pat, ...)
10955 PERL_ARGS_ASSERT_SV_CATPVF;
10957 va_start(args, pat);
10958 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10963 =for apidoc sv_vcatpvf
10965 Processes its arguments like C<sv_vcatpvfn> called with a non-null C-style
10966 variable argument list, and appends the formatted output
10967 to an SV. Does not handle 'set' magic. See C<L</sv_vcatpvf_mg>>.
10969 Usually used via its frontend C<sv_catpvf>.
10975 Perl_sv_vcatpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10977 PERL_ARGS_ASSERT_SV_VCATPVF;
10979 sv_vcatpvfn_flags(sv, pat, strlen(pat), args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10983 =for apidoc sv_catpvf_mg
10985 Like C<sv_catpvf>, but also handles 'set' magic.
10991 Perl_sv_catpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10995 PERL_ARGS_ASSERT_SV_CATPVF_MG;
10997 va_start(args, pat);
10998 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
11004 =for apidoc sv_vcatpvf_mg
11006 Like C<sv_vcatpvf>, but also handles 'set' magic.
11008 Usually used via its frontend C<sv_catpvf_mg>.
11014 Perl_sv_vcatpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
11016 PERL_ARGS_ASSERT_SV_VCATPVF_MG;
11018 sv_vcatpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
11023 =for apidoc sv_vsetpvfn
11025 Works like C<sv_vcatpvfn> but copies the text into the SV instead of
11028 Usually used via one of its frontends C<sv_vsetpvf> and C<sv_vsetpvf_mg>.
11034 Perl_sv_vsetpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11035 va_list *const args, SV **const svargs, const Size_t sv_count, bool *const maybe_tainted)
11037 PERL_ARGS_ASSERT_SV_VSETPVFN;
11040 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, sv_count, maybe_tainted, 0);
11044 /* simplified inline Perl_sv_catpvn_nomg() when you know the SV's SvPOK */
11046 PERL_STATIC_INLINE void
11047 S_sv_catpvn_simple(pTHX_ SV *const sv, const char* const buf, const STRLEN len)
11049 STRLEN const need = len + SvCUR(sv) + 1;
11052 /* can't wrap as both len and SvCUR() are allocated in
11053 * memory and together can't consume all the address space
11055 assert(need > len);
11060 Copy(buf, end, len, char);
11063 SvCUR_set(sv, need - 1);
11068 * Warn of missing argument to sprintf. The value used in place of such
11069 * arguments should be &PL_sv_no; an undefined value would yield
11070 * inappropriate "use of uninit" warnings [perl #71000].
11073 S_warn_vcatpvfn_missing_argument(pTHX) {
11074 if (ckWARN(WARN_MISSING)) {
11075 Perl_warner(aTHX_ packWARN(WARN_MISSING), "Missing argument in %s",
11076 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11085 Perl_croak(aTHX_ "Integer overflow in format string for %s",
11086 (PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn"));
11090 /* Given an int i from the next arg (if args is true) or an sv from an arg
11091 * (if args is false), try to extract a STRLEN-ranged value from the arg,
11092 * with overflow checking.
11093 * Sets *neg to true if the value was negative (untouched otherwise.
11094 * Returns the absolute value.
11095 * As an extra margin of safety, it croaks if the returned value would
11096 * exceed the maximum value of a STRLEN / 4.
11100 S_sprintf_arg_num_val(pTHX_ va_list *const args, int i, SV *sv, bool *neg)
11114 if (UNLIKELY(SvIsUV(sv))) {
11115 UV uv = SvUV_nomg(sv);
11117 S_croak_overflow();
11121 iv = SvIV_nomg(sv);
11125 S_croak_overflow();
11131 if (iv > (IV)(((STRLEN)~0) / 4))
11132 S_croak_overflow();
11137 /* Read in and return a number. Updates *pattern to point to the char
11138 * following the number. Expects the first char to 1..9.
11139 * Croaks if the number exceeds 1/4 of the maximum value of STRLEN.
11140 * This is a belt-and-braces safety measure to complement any
11141 * overflow/wrap checks done in the main body of sv_vcatpvfn_flags.
11142 * It means that e.g. on a 32-bit system the width/precision can't be more
11143 * than 1G, which seems reasonable.
11147 S_expect_number(pTHX_ const char **const pattern)
11151 PERL_ARGS_ASSERT_EXPECT_NUMBER;
11153 assert(inRANGE(**pattern, '1', '9'));
11155 var = *(*pattern)++ - '0';
11156 while (isDIGIT(**pattern)) {
11157 /* if var * 10 + 9 would exceed 1/4 max strlen, croak */
11158 if (var > ((((STRLEN)~0) / 4 - 9) / 10))
11159 S_croak_overflow();
11160 var = var * 10 + (*(*pattern)++ - '0');
11165 /* Implement a fast "%.0f": given a pointer to the end of a buffer (caller
11166 * ensures it's big enough), back fill it with the rounded integer part of
11167 * nv. Returns ptr to start of string, and sets *len to its length.
11168 * Returns NULL if not convertible.
11172 S_F0convert(NV nv, char *const endbuf, STRLEN *const len)
11174 const int neg = nv < 0;
11177 PERL_ARGS_ASSERT_F0CONVERT;
11179 assert(!Perl_isinfnan(nv));
11182 if (nv != 0.0 && nv < UV_MAX) {
11188 if (uv & 1 && uv == nv)
11189 uv--; /* Round to even */
11192 const unsigned dig = uv % 10;
11194 } while (uv /= 10);
11204 /* XXX maybe_tainted is never assigned to, so the doc above is lying. */
11207 Perl_sv_vcatpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11208 va_list *const args, SV **const svargs, const Size_t sv_count, bool *const maybe_tainted)
11210 PERL_ARGS_ASSERT_SV_VCATPVFN;
11212 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, sv_count, maybe_tainted, SV_GMAGIC|SV_SMAGIC);
11216 /* For the vcatpvfn code, we need a long double target in case
11217 * HAS_LONG_DOUBLE, even without USE_LONG_DOUBLE, so that we can printf
11218 * with long double formats, even without NV being long double. But we
11219 * call the target 'fv' instead of 'nv', since most of the time it is not
11220 * (most compilers these days recognize "long double", even if only as a
11221 * synonym for "double").
11223 #if defined(HAS_LONG_DOUBLE) && LONG_DOUBLESIZE > DOUBLESIZE && \
11224 defined(PERL_PRIgldbl) && !defined(USE_QUADMATH)
11225 # define VCATPVFN_FV_GF PERL_PRIgldbl
11226 # if defined(__VMS) && defined(__ia64) && defined(__IEEE_FLOAT)
11227 /* Work around breakage in OTS$CVT_FLOAT_T_X */
11228 # define VCATPVFN_NV_TO_FV(nv,fv) \
11231 fv = Perl_isnan(_dv) ? LDBL_QNAN : _dv; \
11234 # define VCATPVFN_NV_TO_FV(nv,fv) (fv)=(nv)
11236 typedef long double vcatpvfn_long_double_t;
11238 # define VCATPVFN_FV_GF NVgf
11239 # define VCATPVFN_NV_TO_FV(nv,fv) (fv)=(nv)
11240 typedef NV vcatpvfn_long_double_t;
11243 #ifdef LONGDOUBLE_DOUBLEDOUBLE
11244 /* The first double can be as large as 2**1023, or '1' x '0' x 1023.
11245 * The second double can be as small as 2**-1074, or '0' x 1073 . '1'.
11246 * The sum of them can be '1' . '0' x 2096 . '1', with implied radix point
11247 * after the first 1023 zero bits.
11249 * XXX The 2098 is quite large (262.25 bytes) and therefore some sort
11250 * of dynamically growing buffer might be better, start at just 16 bytes
11251 * (for example) and grow only when necessary. Or maybe just by looking
11252 * at the exponents of the two doubles? */
11253 # define DOUBLEDOUBLE_MAXBITS 2098
11256 /* vhex will contain the values (0..15) of the hex digits ("nybbles"
11257 * of 4 bits); 1 for the implicit 1, and the mantissa bits, four bits
11258 * per xdigit. For the double-double case, this can be rather many.
11259 * The non-double-double-long-double overshoots since all bits of NV
11260 * are not mantissa bits, there are also exponent bits. */
11261 #ifdef LONGDOUBLE_DOUBLEDOUBLE
11262 # define VHEX_SIZE (3+DOUBLEDOUBLE_MAXBITS/4)
11264 # define VHEX_SIZE (1+(NVSIZE * 8)/4)
11267 /* If we do not have a known long double format, (including not using
11268 * long doubles, or long doubles being equal to doubles) then we will
11269 * fall back to the ldexp/frexp route, with which we can retrieve at
11270 * most as many bits as our widest unsigned integer type is. We try
11271 * to get a 64-bit unsigned integer even if we are not using a 64-bit UV.
11273 * (If you want to test the case of UVSIZE == 4, NVSIZE == 8,
11274 * set the MANTISSATYPE to int and the MANTISSASIZE to 4.)
11276 #if defined(HAS_QUAD) && defined(Uquad_t)
11277 # define MANTISSATYPE Uquad_t
11278 # define MANTISSASIZE 8
11280 # define MANTISSATYPE UV
11281 # define MANTISSASIZE UVSIZE
11284 #if defined(DOUBLE_LITTLE_ENDIAN) || defined(LONGDOUBLE_LITTLE_ENDIAN)
11285 # define HEXTRACT_LITTLE_ENDIAN
11286 #elif defined(DOUBLE_BIG_ENDIAN) || defined(LONGDOUBLE_BIG_ENDIAN)
11287 # define HEXTRACT_BIG_ENDIAN
11289 # define HEXTRACT_MIX_ENDIAN
11292 /* S_hextract() is a helper for S_format_hexfp, for extracting
11293 * the hexadecimal values (for %a/%A). The nv is the NV where the value
11294 * are being extracted from (either directly from the long double in-memory
11295 * presentation, or from the uquad computed via frexp+ldexp). frexp also
11296 * is used to update the exponent. The subnormal is set to true
11297 * for IEEE 754 subnormals/denormals (including the x86 80-bit format).
11298 * The vhex is the pointer to the beginning of the output buffer of VHEX_SIZE.
11300 * The tricky part is that S_hextract() needs to be called twice:
11301 * the first time with vend as NULL, and the second time with vend as
11302 * the pointer returned by the first call. What happens is that on
11303 * the first round the output size is computed, and the intended
11304 * extraction sanity checked. On the second round the actual output
11305 * (the extraction of the hexadecimal values) takes place.
11306 * Sanity failures cause fatal failures during both rounds. */
11308 S_hextract(pTHX_ const NV nv, int* exponent, bool *subnormal,
11309 U8* vhex, U8* vend)
11313 int ixmin = 0, ixmax = 0;
11315 /* XXX Inf/NaN are not handled here, since it is
11316 * assumed they are to be output as "Inf" and "NaN". */
11318 /* These macros are just to reduce typos, they have multiple
11319 * repetitions below, but usually only one (or sometimes two)
11320 * of them is really being used. */
11321 /* HEXTRACT_OUTPUT() extracts the high nybble first. */
11322 #define HEXTRACT_OUTPUT_HI(ix) (*v++ = nvp[ix] >> 4)
11323 #define HEXTRACT_OUTPUT_LO(ix) (*v++ = nvp[ix] & 0xF)
11324 #define HEXTRACT_OUTPUT(ix) \
11326 HEXTRACT_OUTPUT_HI(ix); HEXTRACT_OUTPUT_LO(ix); \
11328 #define HEXTRACT_COUNT(ix, c) \
11330 v += c; if (ix < ixmin) ixmin = ix; else if (ix > ixmax) ixmax = ix; \
11332 #define HEXTRACT_BYTE(ix) \
11334 if (vend) HEXTRACT_OUTPUT(ix); else HEXTRACT_COUNT(ix, 2); \
11336 #define HEXTRACT_LO_NYBBLE(ix) \
11338 if (vend) HEXTRACT_OUTPUT_LO(ix); else HEXTRACT_COUNT(ix, 1); \
11340 /* HEXTRACT_TOP_NYBBLE is just convenience disguise,
11341 * to make it look less odd when the top bits of a NV
11342 * are extracted using HEXTRACT_LO_NYBBLE: the highest
11343 * order bits can be in the "low nybble" of a byte. */
11344 #define HEXTRACT_TOP_NYBBLE(ix) HEXTRACT_LO_NYBBLE(ix)
11345 #define HEXTRACT_BYTES_LE(a, b) \
11346 for (ix = a; ix >= b; ix--) { HEXTRACT_BYTE(ix); }
11347 #define HEXTRACT_BYTES_BE(a, b) \
11348 for (ix = a; ix <= b; ix++) { HEXTRACT_BYTE(ix); }
11349 #define HEXTRACT_GET_SUBNORMAL(nv) *subnormal = Perl_fp_class_denorm(nv)
11350 #define HEXTRACT_IMPLICIT_BIT(nv) \
11352 if (!*subnormal) { \
11353 if (vend) *v++ = ((nv) == 0.0) ? 0 : 1; else v++; \
11357 /* Most formats do. Those which don't should undef this.
11359 * But also note that IEEE 754 subnormals do not have it, or,
11360 * expressed alternatively, their implicit bit is zero. */
11361 #define HEXTRACT_HAS_IMPLICIT_BIT
11363 /* Many formats do. Those which don't should undef this. */
11364 #define HEXTRACT_HAS_TOP_NYBBLE
11366 /* HEXTRACTSIZE is the maximum number of xdigits. */
11367 #if defined(USE_LONG_DOUBLE) && defined(LONGDOUBLE_DOUBLEDOUBLE)
11368 # define HEXTRACTSIZE (2+DOUBLEDOUBLE_MAXBITS/4)
11370 # define HEXTRACTSIZE 2 * NVSIZE
11373 const U8* vmaxend = vhex + HEXTRACTSIZE;
11375 assert(HEXTRACTSIZE <= VHEX_SIZE);
11377 PERL_UNUSED_VAR(ix); /* might happen */
11378 (void)Perl_frexp(PERL_ABS(nv), exponent);
11379 *subnormal = FALSE;
11380 if (vend && (vend <= vhex || vend > vmaxend)) {
11381 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11382 Perl_croak(aTHX_ "Hexadecimal float: internal error (entry)");
11385 /* First check if using long doubles. */
11386 #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE)
11387 # if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN
11388 /* Used in e.g. VMS and HP-UX IA-64, e.g. -0.1L:
11389 * 9a 99 99 99 99 99 99 99 99 99 99 99 99 99 fb bf */
11390 /* The bytes 13..0 are the mantissa/fraction,
11391 * the 15,14 are the sign+exponent. */
11392 const U8* nvp = (const U8*)(&nv);
11393 HEXTRACT_GET_SUBNORMAL(nv);
11394 HEXTRACT_IMPLICIT_BIT(nv);
11395 # undef HEXTRACT_HAS_TOP_NYBBLE
11396 HEXTRACT_BYTES_LE(13, 0);
11397 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_BIG_ENDIAN
11398 /* Used in e.g. Solaris Sparc and HP-UX PA-RISC, e.g. -0.1L:
11399 * bf fb 99 99 99 99 99 99 99 99 99 99 99 99 99 9a */
11400 /* The bytes 2..15 are the mantissa/fraction,
11401 * the 0,1 are the sign+exponent. */
11402 const U8* nvp = (const U8*)(&nv);
11403 HEXTRACT_GET_SUBNORMAL(nv);
11404 HEXTRACT_IMPLICIT_BIT(nv);
11405 # undef HEXTRACT_HAS_TOP_NYBBLE
11406 HEXTRACT_BYTES_BE(2, 15);
11407 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN
11408 /* x86 80-bit "extended precision", 64 bits of mantissa / fraction /
11409 * significand, 15 bits of exponent, 1 bit of sign. No implicit bit.
11410 * NVSIZE can be either 12 (ILP32, Solaris x86) or 16 (LP64, Linux
11411 * and OS X), meaning that 2 or 6 bytes are empty padding. */
11412 /* The bytes 0..1 are the sign+exponent,
11413 * the bytes 2..9 are the mantissa/fraction. */
11414 const U8* nvp = (const U8*)(&nv);
11415 # undef HEXTRACT_HAS_IMPLICIT_BIT
11416 # undef HEXTRACT_HAS_TOP_NYBBLE
11417 HEXTRACT_GET_SUBNORMAL(nv);
11418 HEXTRACT_BYTES_LE(7, 0);
11419 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN
11420 /* Does this format ever happen? (Wikipedia says the Motorola
11421 * 6888x math coprocessors used format _like_ this but padded
11422 * to 96 bits with 16 unused bits between the exponent and the
11424 const U8* nvp = (const U8*)(&nv);
11425 # undef HEXTRACT_HAS_IMPLICIT_BIT
11426 # undef HEXTRACT_HAS_TOP_NYBBLE
11427 HEXTRACT_GET_SUBNORMAL(nv);
11428 HEXTRACT_BYTES_BE(0, 7);
11430 # define HEXTRACT_FALLBACK
11431 /* Double-double format: two doubles next to each other.
11432 * The first double is the high-order one, exactly like
11433 * it would be for a "lone" double. The second double
11434 * is shifted down using the exponent so that that there
11435 * are no common bits. The tricky part is that the value
11436 * of the double-double is the SUM of the two doubles and
11437 * the second one can be also NEGATIVE.
11439 * Because of this tricky construction the bytewise extraction we
11440 * use for the other long double formats doesn't work, we must
11441 * extract the values bit by bit.
11443 * The little-endian double-double is used .. somewhere?
11445 * The big endian double-double is used in e.g. PPC/Power (AIX)
11448 * The mantissa bits are in two separate stretches, e.g. for -0.1L:
11449 * 9a 99 99 99 99 99 59 bc 9a 99 99 99 99 99 b9 3f (LE)
11450 * 3f b9 99 99 99 99 99 9a bc 59 99 99 99 99 99 9a (BE)
11453 #else /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) */
11454 /* Using normal doubles, not long doubles.
11456 * We generate 4-bit xdigits (nybble/nibble) instead of 8-bit
11457 * bytes, since we might need to handle printf precision, and
11458 * also need to insert the radix. */
11460 # ifdef HEXTRACT_LITTLE_ENDIAN
11461 /* 0 1 2 3 4 5 6 7 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11462 const U8* nvp = (const U8*)(&nv);
11463 HEXTRACT_GET_SUBNORMAL(nv);
11464 HEXTRACT_IMPLICIT_BIT(nv);
11465 HEXTRACT_TOP_NYBBLE(6);
11466 HEXTRACT_BYTES_LE(5, 0);
11467 # elif defined(HEXTRACT_BIG_ENDIAN)
11468 /* 7 6 5 4 3 2 1 0 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11469 const U8* nvp = (const U8*)(&nv);
11470 HEXTRACT_GET_SUBNORMAL(nv);
11471 HEXTRACT_IMPLICIT_BIT(nv);
11472 HEXTRACT_TOP_NYBBLE(1);
11473 HEXTRACT_BYTES_BE(2, 7);
11474 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_LE_BE
11475 /* 4 5 6 7 0 1 2 3 (MSB = 7, LSB = 0, 6:7 = nybble:exponent:sign) */
11476 const U8* nvp = (const U8*)(&nv);
11477 HEXTRACT_GET_SUBNORMAL(nv);
11478 HEXTRACT_IMPLICIT_BIT(nv);
11479 HEXTRACT_TOP_NYBBLE(2); /* 6 */
11480 HEXTRACT_BYTE(1); /* 5 */
11481 HEXTRACT_BYTE(0); /* 4 */
11482 HEXTRACT_BYTE(7); /* 3 */
11483 HEXTRACT_BYTE(6); /* 2 */
11484 HEXTRACT_BYTE(5); /* 1 */
11485 HEXTRACT_BYTE(4); /* 0 */
11486 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_BE_LE
11487 /* 3 2 1 0 7 6 5 4 (MSB = 7, LSB = 0, 7:6 = sign:exponent:nybble) */
11488 const U8* nvp = (const U8*)(&nv);
11489 HEXTRACT_GET_SUBNORMAL(nv);
11490 HEXTRACT_IMPLICIT_BIT(nv);
11491 HEXTRACT_TOP_NYBBLE(5); /* 6 */
11492 HEXTRACT_BYTE(6); /* 5 */
11493 HEXTRACT_BYTE(7); /* 4 */
11494 HEXTRACT_BYTE(0); /* 3 */
11495 HEXTRACT_BYTE(1); /* 2 */
11496 HEXTRACT_BYTE(2); /* 1 */
11497 HEXTRACT_BYTE(3); /* 0 */
11499 # define HEXTRACT_FALLBACK
11502 # define HEXTRACT_FALLBACK
11504 #endif /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) #else */
11506 #ifdef HEXTRACT_FALLBACK
11507 HEXTRACT_GET_SUBNORMAL(nv);
11508 # undef HEXTRACT_HAS_TOP_NYBBLE /* Meaningless, but consistent. */
11509 /* The fallback is used for the double-double format, and
11510 * for unknown long double formats, and for unknown double
11511 * formats, or in general unknown NV formats. */
11512 if (nv == (NV)0.0) {
11520 NV d = nv < 0 ? -nv : nv;
11522 U8 ha = 0x0; /* hexvalue accumulator */
11523 U8 hd = 0x8; /* hexvalue digit */
11525 /* Shift d and e (and update exponent) so that e <= d < 2*e,
11526 * this is essentially manual frexp(). Multiplying by 0.5 and
11527 * doubling should be lossless in binary floating point. */
11537 while (d >= e + e) {
11541 /* Now e <= d < 2*e */
11543 /* First extract the leading hexdigit (the implicit bit). */
11559 /* Then extract the remaining hexdigits. */
11560 while (d > (NV)0.0) {
11566 /* Output or count in groups of four bits,
11567 * that is, when the hexdigit is down to one. */
11572 /* Reset the hexvalue. */
11581 /* Flush possible pending hexvalue. */
11591 /* Croak for various reasons: if the output pointer escaped the
11592 * output buffer, if the extraction index escaped the extraction
11593 * buffer, or if the ending output pointer didn't match the
11594 * previously computed value. */
11595 if (v <= vhex || v - vhex >= VHEX_SIZE ||
11596 /* For double-double the ixmin and ixmax stay at zero,
11597 * which is convenient since the HEXTRACTSIZE is tricky
11598 * for double-double. */
11599 ixmin < 0 || ixmax >= NVSIZE ||
11600 (vend && v != vend)) {
11601 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11602 Perl_croak(aTHX_ "Hexadecimal float: internal error (overflow)");
11608 /* S_format_hexfp(): helper function for Perl_sv_vcatpvfn_flags().
11610 * Processes the %a/%A hexadecimal floating-point format, since the
11611 * built-in snprintf()s which are used for most of the f/p formats, don't
11612 * universally handle %a/%A.
11613 * Populates buf of length bufsize, and returns the length of the created
11615 * The rest of the args have the same meaning as the local vars of the
11616 * same name within Perl_sv_vcatpvfn_flags().
11618 * The caller's determination of IN_LC(LC_NUMERIC), passed as in_lc_numeric,
11619 * is used to ensure we do the right thing when we need to access the locale's
11622 * It requires the caller to make buf large enough.
11626 S_format_hexfp(pTHX_ char * const buf, const STRLEN bufsize, const char c,
11627 const NV nv, const vcatpvfn_long_double_t fv,
11628 bool has_precis, STRLEN precis, STRLEN width,
11629 bool alt, char plus, bool left, bool fill, bool in_lc_numeric)
11631 /* Hexadecimal floating point. */
11633 U8 vhex[VHEX_SIZE];
11634 U8* v = vhex; /* working pointer to vhex */
11635 U8* vend; /* pointer to one beyond last digit of vhex */
11636 U8* vfnz = NULL; /* first non-zero */
11637 U8* vlnz = NULL; /* last non-zero */
11638 U8* v0 = NULL; /* first output */
11639 const bool lower = (c == 'a');
11640 /* At output the values of vhex (up to vend) will
11641 * be mapped through the xdig to get the actual
11642 * human-readable xdigits. */
11643 const char* xdig = PL_hexdigit;
11644 STRLEN zerotail = 0; /* how many extra zeros to append */
11645 int exponent = 0; /* exponent of the floating point input */
11646 bool hexradix = FALSE; /* should we output the radix */
11647 bool subnormal = FALSE; /* IEEE 754 subnormal/denormal */
11648 bool negative = FALSE;
11651 /* XXX: NaN, Inf -- though they are printed as "NaN" and "Inf".
11653 * For example with denormals, (assuming the vanilla
11654 * 64-bit double): the exponent is zero. 1xp-1074 is
11655 * the smallest denormal and the smallest double, it
11656 * could be output also as 0x0.0000000000001p-1022 to
11657 * match its internal structure. */
11659 vend = S_hextract(aTHX_ nv, &exponent, &subnormal, vhex, NULL);
11660 S_hextract(aTHX_ nv, &exponent, &subnormal, vhex, vend);
11662 #if NVSIZE > DOUBLESIZE
11663 # ifdef HEXTRACT_HAS_IMPLICIT_BIT
11664 /* In this case there is an implicit bit,
11665 * and therefore the exponent is shifted by one. */
11667 # elif defined(NV_X86_80_BIT)
11669 /* The subnormals of the x86-80 have a base exponent of -16382,
11670 * (while the physical exponent bits are zero) but the frexp()
11671 * returned the scientific-style floating exponent. We want
11672 * to map the last one as:
11673 * -16831..-16384 -> -16382 (the last normal is 0x1p-16382)
11674 * -16835..-16388 -> -16384
11675 * since we want to keep the first hexdigit
11676 * as one of the [8421]. */
11677 exponent = -4 * ( (exponent + 1) / -4) - 2;
11681 /* TBD: other non-implicit-bit platforms than the x86-80. */
11685 negative = fv < 0 || Perl_signbit(nv);
11696 xdig += 16; /* Use uppercase hex. */
11699 /* Find the first non-zero xdigit. */
11700 for (v = vhex; v < vend; v++) {
11708 /* Find the last non-zero xdigit. */
11709 for (v = vend - 1; v >= vhex; v--) {
11716 #if NVSIZE == DOUBLESIZE
11722 #ifndef NV_X86_80_BIT
11724 /* IEEE 754 subnormals (but not the x86 80-bit):
11725 * we want "normalize" the subnormal,
11726 * so we need to right shift the hex nybbles
11727 * so that the output of the subnormal starts
11728 * from the first true bit. (Another, equally
11729 * valid, policy would be to dump the subnormal
11730 * nybbles as-is, to display the "physical" layout.) */
11733 /* Find the ceil(log2(v[0])) of
11734 * the top non-zero nybble. */
11735 for (i = vfnz[0], n = 0; i > 1; i >>= 1, n++) { }
11739 for (vshr = vlnz; vshr >= vfnz; vshr--) {
11740 vshr[1] |= (vshr[0] & (0xF >> (4 - n))) << (4 - n);
11754 U8* ve = (subnormal ? vlnz + 1 : vend);
11755 SSize_t vn = ve - v0;
11757 if (precis < (Size_t)(vn - 1)) {
11758 bool overflow = FALSE;
11759 if (v0[precis + 1] < 0x8) {
11760 /* Round down, nothing to do. */
11761 } else if (v0[precis + 1] > 0x8) {
11764 overflow = v0[precis] > 0xF;
11766 } else { /* v0[precis] == 0x8 */
11767 /* Half-point: round towards the one
11768 * with the even least-significant digit:
11776 * 78 -> 8 f8 -> 10 */
11777 if ((v0[precis] & 0x1)) {
11780 overflow = v0[precis] > 0xF;
11785 for (v = v0 + precis - 1; v >= v0; v--) {
11787 overflow = *v > 0xF;
11793 if (v == v0 - 1 && overflow) {
11794 /* If the overflow goes all the
11795 * way to the front, we need to
11796 * insert 0x1 in front, and adjust
11798 Move(v0, v0 + 1, vn - 1, char);
11804 /* The new effective "last non zero". */
11805 vlnz = v0 + precis;
11809 subnormal ? precis - vn + 1 :
11810 precis - (vlnz - vhex);
11817 /* If there are non-zero xdigits, the radix
11818 * is output after the first one. */
11826 zerotail = has_precis ? precis : 0;
11829 /* The radix is always output if precis, or if alt. */
11830 if ((has_precis && precis > 0) || alt) {
11835 #ifndef USE_LOCALE_NUMERIC
11838 if (in_lc_numeric) {
11840 WITH_LC_NUMERIC_SET_TO_NEEDED_IN(TRUE, {
11841 const char* r = SvPV(PL_numeric_radix_sv, n);
11842 Copy(r, p, n, char);
11857 if (zerotail > 0) {
11858 while (zerotail--) {
11865 /* sanity checks */
11866 if (elen >= bufsize || width >= bufsize)
11867 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11868 Perl_croak(aTHX_ "Hexadecimal float: internal error (overflow)");
11870 elen += my_snprintf(p, bufsize - elen,
11871 "%c%+d", lower ? 'p' : 'P',
11874 if (elen < width) {
11875 STRLEN gap = (STRLEN)(width - elen);
11877 /* Pad the back with spaces. */
11878 memset(buf + elen, ' ', gap);
11881 /* Insert the zeros after the "0x" and the
11882 * the potential sign, but before the digits,
11883 * otherwise we end up with "0000xH.HHH...",
11884 * when we want "0x000H.HHH..." */
11885 STRLEN nzero = gap;
11886 char* zerox = buf + 2;
11887 STRLEN nmove = elen - 2;
11888 if (negative || plus) {
11892 Move(zerox, zerox + nzero, nmove, char);
11893 memset(zerox, fill ? '0' : ' ', nzero);
11896 /* Move it to the right. */
11897 Move(buf, buf + gap,
11899 /* Pad the front with spaces. */
11900 memset(buf, ' ', gap);
11909 =for apidoc sv_vcatpvfn
11911 =for apidoc sv_vcatpvfn_flags
11913 Processes its arguments like C<vsprintf> and appends the formatted output
11914 to an SV. Uses an array of SVs if the C-style variable argument list is
11915 missing (C<NULL>). Argument reordering (using format specifiers like C<%2$d>
11916 or C<%*2$d>) is supported only when using an array of SVs; using a C-style
11917 C<va_list> argument list with a format string that uses argument reordering
11918 will yield an exception.
11920 When running with taint checks enabled, indicates via
11921 C<maybe_tainted> if results are untrustworthy (often due to the use of
11924 If called as C<sv_vcatpvfn> or flags has the C<SV_GMAGIC> bit set, calls get magic.
11926 It assumes that pat has the same utf8-ness as sv. It's the caller's
11927 responsibility to ensure that this is so.
11929 Usually used via one of its frontends C<sv_vcatpvf> and C<sv_vcatpvf_mg>.
11936 Perl_sv_vcatpvfn_flags(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11937 va_list *const args, SV **const svargs, const Size_t sv_count, bool *const maybe_tainted,
11940 const char *fmtstart; /* character following the current '%' */
11941 const char *q; /* current position within format */
11942 const char *patend;
11945 static const char nullstr[] = "(null)";
11946 bool has_utf8 = DO_UTF8(sv); /* has the result utf8? */
11947 const bool pat_utf8 = has_utf8; /* the pattern is in utf8? */
11948 /* Times 4: a decimal digit takes more than 3 binary digits.
11949 * NV_DIG: mantissa takes that many decimal digits.
11950 * Plus 32: Playing safe. */
11951 char ebuf[IV_DIG * 4 + NV_DIG + 32];
11952 bool no_redundant_warning = FALSE; /* did we use any explicit format parameter index? */
11953 #ifdef USE_LOCALE_NUMERIC
11954 bool have_in_lc_numeric = FALSE;
11956 /* we never change this unless USE_LOCALE_NUMERIC */
11957 bool in_lc_numeric = FALSE;
11959 PERL_ARGS_ASSERT_SV_VCATPVFN_FLAGS;
11960 PERL_UNUSED_ARG(maybe_tainted);
11962 if (flags & SV_GMAGIC)
11965 /* no matter what, this is a string now */
11966 (void)SvPV_force_nomg(sv, origlen);
11968 /* the code that scans for flags etc following a % relies on
11969 * a '\0' being present to avoid falling off the end. Ideally that
11970 * should be fixed */
11971 assert(pat[patlen] == '\0');
11974 /* Special-case "", "%s", "%-p" (SVf - see below) and "%.0f".
11975 * In each case, if there isn't the correct number of args, instead
11976 * fall through to the main code to handle the issuing of any
11980 if (patlen == 0 && (args || sv_count == 0))
11983 if (patlen <= 4 && pat[0] == '%' && (args || sv_count == 1)) {
11986 if (patlen == 2 && pat[1] == 's') {
11988 const char * const s = va_arg(*args, char*);
11989 sv_catpv_nomg(sv, s ? s : nullstr);
11992 /* we want get magic on the source but not the target.
11993 * sv_catsv can't do that, though */
11994 SvGETMAGIC(*svargs);
11995 sv_catsv_nomg(sv, *svargs);
12002 if (patlen == 3 && pat[1] == '-' && pat[2] == 'p') {
12003 SV *asv = MUTABLE_SV(va_arg(*args, void*));
12004 sv_catsv_nomg(sv, asv);
12008 #if !defined(USE_LONG_DOUBLE) && !defined(USE_QUADMATH)
12009 /* special-case "%.0f" */
12010 else if ( patlen == 4
12011 && pat[1] == '.' && pat[2] == '0' && pat[3] == 'f')
12013 const NV nv = SvNV(*svargs);
12014 if (LIKELY(!Perl_isinfnan(nv))) {
12018 if ((p = F0convert(nv, ebuf + sizeof ebuf, &l))) {
12019 sv_catpvn_nomg(sv, p, l);
12024 #endif /* !USE_LONG_DOUBLE */
12028 patend = (char*)pat + patlen;
12029 for (fmtstart = pat; fmtstart < patend; fmtstart = q) {
12030 char intsize = 0; /* size qualifier in "%hi..." etc */
12031 bool alt = FALSE; /* has "%#..." */
12032 bool left = FALSE; /* has "%-..." */
12033 bool fill = FALSE; /* has "%0..." */
12034 char plus = 0; /* has "%+..." */
12035 STRLEN width = 0; /* value of "%NNN..." */
12036 bool has_precis = FALSE; /* has "%.NNN..." */
12037 STRLEN precis = 0; /* value of "%.NNN..." */
12038 int base = 0; /* base to print in, e.g. 8 for %o */
12039 UV uv = 0; /* the value to print of int-ish args */
12041 bool vectorize = FALSE; /* has "%v..." */
12042 bool vec_utf8 = FALSE; /* SvUTF8(vec arg) */
12043 const U8 *vecstr = NULL; /* SvPVX(vec arg) */
12044 STRLEN veclen = 0; /* SvCUR(vec arg) */
12045 const char *dotstr = NULL; /* separator string for %v */
12046 STRLEN dotstrlen; /* length of separator string for %v */
12048 Size_t efix = 0; /* explicit format parameter index */
12049 const Size_t osvix = svix; /* original index in case of bad fmt */
12052 bool is_utf8 = FALSE; /* is this item utf8? */
12053 bool arg_missing = FALSE; /* give "Missing argument" warning */
12054 char esignbuf[4]; /* holds sign prefix, e.g. "-0x" */
12055 STRLEN esignlen = 0; /* length of e.g. "-0x" */
12056 STRLEN zeros = 0; /* how many '0' to prepend */
12058 const char *eptr = NULL; /* the address of the element string */
12059 STRLEN elen = 0; /* the length of the element string */
12061 char c; /* the actual format ('d', s' etc) */
12064 /* echo everything up to the next format specification */
12065 for (q = fmtstart; q < patend && *q != '%'; ++q)
12068 if (q > fmtstart) {
12069 if (has_utf8 && !pat_utf8) {
12070 /* upgrade and copy the bytes of fmtstart..q-1 to utf8 on
12074 STRLEN need = SvCUR(sv) + (q - fmtstart) + 1;
12076 for (p = fmtstart; p < q; p++)
12077 if (!NATIVE_BYTE_IS_INVARIANT(*p))
12082 for (p = fmtstart; p < q; p++)
12083 append_utf8_from_native_byte((U8)*p, (U8**)&dst);
12085 SvCUR_set(sv, need - 1);
12088 S_sv_catpvn_simple(aTHX_ sv, fmtstart, q - fmtstart);
12093 fmtstart = q; /* fmtstart is char following the '%' */
12096 We allow format specification elements in this order:
12097 \d+\$ explicit format parameter index
12099 v|\*(\d+\$)?v vector with optional (optionally specified) arg
12100 0 flag (as above): repeated to allow "v02"
12101 \d+|\*(\d+\$)? width using optional (optionally specified) arg
12102 \.(\d*|\*(\d+\$)?) precision using optional (optionally specified) arg
12104 [%bcdefginopsuxDFOUX] format (mandatory)
12107 if (inRANGE(*q, '1', '9')) {
12108 width = expect_number(&q);
12111 Perl_croak_nocontext(
12112 "Cannot yet reorder sv_vcatpvfn() arguments from va_list");
12114 efix = (Size_t)width;
12116 no_redundant_warning = TRUE;
12128 if (plus == '+' && *q == ' ') /* '+' over ' ' */
12155 /* at this point we can expect one of:
12157 * 123 an explicit width
12158 * * width taken from next arg
12159 * *12$ width taken from 12th arg
12162 * But any width specification may be preceded by a v, in one of its
12167 * So an asterisk may be either a width specifier or a vector
12168 * separator arg specifier, and we don't know which initially
12173 STRLEN ix; /* explicit width/vector separator index */
12175 if (inRANGE(*q, '1', '9')) {
12176 ix = expect_number(&q);
12179 Perl_croak_nocontext(
12180 "Cannot yet reorder sv_vcatpvfn() arguments from va_list");
12181 no_redundant_warning = TRUE;
12190 /* The asterisk was for *v, *NNN$v: vectorizing, but not
12191 * with the default "." */
12196 vecsv = va_arg(*args, SV*);
12198 ix = ix ? ix - 1 : svix++;
12199 vecsv = ix < sv_count ? svargs[ix]
12200 : (arg_missing = TRUE, &PL_sv_no);
12202 dotstr = SvPV_const(vecsv, dotstrlen);
12203 /* Keep the DO_UTF8 test *after* the SvPV call, else things go
12204 bad with tied or overloaded values that return UTF8. */
12205 if (DO_UTF8(vecsv))
12207 else if (has_utf8) {
12208 vecsv = sv_mortalcopy(vecsv);
12209 sv_utf8_upgrade(vecsv);
12210 dotstr = SvPV_const(vecsv, dotstrlen);
12217 /* the asterisk specified a width */
12220 SV *width_sv = NULL;
12222 i = va_arg(*args, int);
12224 ix = ix ? ix - 1 : svix++;
12225 width_sv = (ix < sv_count) ? svargs[ix]
12226 : (arg_missing = TRUE, (SV*)NULL);
12228 width = S_sprintf_arg_num_val(aTHX_ args, i, width_sv, &left);
12231 else if (*q == 'v') {
12242 /* explicit width? */
12247 if (inRANGE(*q, '1', '9'))
12248 width = expect_number(&q);
12258 STRLEN ix; /* explicit precision index */
12260 if (inRANGE(*q, '1', '9')) {
12261 ix = expect_number(&q);
12264 Perl_croak_nocontext(
12265 "Cannot yet reorder sv_vcatpvfn() arguments from va_list");
12266 no_redundant_warning = TRUE;
12275 SV *width_sv = NULL;
12279 i = va_arg(*args, int);
12281 ix = ix ? ix - 1 : svix++;
12282 width_sv = (ix < sv_count) ? svargs[ix]
12283 : (arg_missing = TRUE, (SV*)NULL);
12285 precis = S_sprintf_arg_num_val(aTHX_ args, i, width_sv, &neg);
12287 /* ignore negative precision */
12293 /* although it doesn't seem documented, this code has long
12295 * no digits following the '.' is treated like '.0'
12296 * the number may be preceded by any number of zeroes,
12297 * e.g. "%.0001f", which is the same as "%.1f"
12298 * so I've kept that behaviour. DAPM May 2017
12302 precis = inRANGE(*q, '1', '9') ? expect_number(&q) : 0;
12311 case 'I': /* Ix, I32x, and I64x */
12312 # ifdef USE_64_BIT_INT
12313 if (q[1] == '6' && q[2] == '4') {
12319 if (q[1] == '3' && q[2] == '2') {
12323 # ifdef USE_64_BIT_INT
12329 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
12330 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
12333 # ifdef USE_QUADMATH
12346 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
12347 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
12348 if (*q == 'l') { /* lld, llf */
12357 if (*++q == 'h') { /* hhd, hhu */
12374 c = *q++; /* c now holds the conversion type */
12376 /* '%' doesn't have an arg, so skip arg processing */
12385 if (vectorize && !memCHRs("BbDdiOouUXx", c))
12388 /* get next arg (individual branches do their own va_arg()
12389 * handling for the args case) */
12392 efix = efix ? efix - 1 : svix++;
12393 argsv = efix < sv_count ? svargs[efix]
12394 : (arg_missing = TRUE, &PL_sv_no);
12404 eptr = va_arg(*args, char*);
12407 elen = my_strnlen(eptr, precis);
12409 elen = strlen(eptr);
12411 eptr = (char *)nullstr;
12412 elen = sizeof nullstr - 1;
12416 eptr = SvPV_const(argsv, elen);
12417 if (DO_UTF8(argsv)) {
12418 STRLEN old_precis = precis;
12419 if (has_precis && precis < elen) {
12420 STRLEN ulen = sv_or_pv_len_utf8(argsv, eptr, elen);
12421 STRLEN p = precis > ulen ? ulen : precis;
12422 precis = sv_or_pv_pos_u2b(argsv, eptr, p, 0);
12423 /* sticks at end */
12425 if (width) { /* fudge width (can't fudge elen) */
12426 if (has_precis && precis < elen)
12427 width += precis - old_precis;
12430 elen - sv_or_pv_len_utf8(argsv,eptr,elen);
12437 if (has_precis && precis < elen)
12449 * "%...p" is normally treated like "%...x", except that the
12450 * number to print is the SV's address (or a pointer address
12451 * for C-ish sprintf).
12453 * However, the C-ish sprintf variant allows a few special
12454 * extensions. These are currently:
12456 * %-p (SVf) Like %s, but gets the string from an SV*
12457 * arg rather than a char* arg.
12458 * (This was previously %_).
12460 * %-<num>p Ditto but like %.<num>s (i.e. num is max width)
12462 * %2p (HEKf) Like %s, but using the key string in a HEK
12464 * %3p (HEKf256) Ditto but like %.256s
12466 * %d%lu%4p (UTF8f) A utf8 string. Consumes 3 args:
12467 * (cBOOL(utf8), len, string_buf).
12468 * It's handled by the "case 'd'" branch
12469 * rather than here.
12471 * %<num>p where num is 1 or > 4: reserved for future
12472 * extensions. Warns, but then is treated as a
12473 * general %p (print hex address) format.
12481 /* not %*p or %*1$p - any width was explicit */
12485 if (left) { /* %-p (SVf), %-NNNp */
12490 argsv = MUTABLE_SV(va_arg(*args, void*));
12491 eptr = SvPV_const(argsv, elen);
12492 if (DO_UTF8(argsv))
12497 else if (width == 2 || width == 3) { /* HEKf, HEKf256 */
12498 HEK * const hek = va_arg(*args, HEK *);
12499 eptr = HEK_KEY(hek);
12500 elen = HEK_LEN(hek);
12511 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
12512 "internal %%<num>p might conflict with future printf extensions");
12516 /* treat as normal %...p */
12518 uv = PTR2UV(args ? va_arg(*args, void*) : argsv);
12523 /* Ignore any size specifiers, since they're not documented as
12524 * being allowed for %c (ideally we should warn on e.g. '%hc').
12525 * Setting a default intsize, along with a positive
12526 * (which signals unsigned) base, causes, for C-ish use, the
12527 * va_arg to be interpreted as an unsigned int, when it's
12528 * actually signed, which will convert -ve values to high +ve
12529 * values. Note that unlike the libc %c, values > 255 will
12530 * convert to high unicode points rather than being truncated
12531 * to 8 bits. For perlish use, it will do SvUV(argsv), which
12532 * will again convert -ve args to high -ve values.
12535 base = 1; /* special value that indicates we're doing a 'c' */
12536 goto get_int_arg_val;
12545 goto get_int_arg_val;
12548 /* probably just a plain %d, but it might be the start of the
12549 * special UTF8f format, which usually looks something like
12550 * "%d%lu%4p" (the lu may vary by platform)
12552 assert((UTF8f)[0] == 'd');
12553 assert((UTF8f)[1] == '%');
12555 if ( args /* UTF8f only valid for C-ish sprintf */
12556 && q == fmtstart + 1 /* plain %d, not %....d */
12557 && patend >= fmtstart + sizeof(UTF8f) - 1 /* long enough */
12559 && strnEQ(q + 1, UTF8f + 2, sizeof(UTF8f) - 3))
12561 /* The argument has already gone through cBOOL, so the cast
12563 is_utf8 = (bool)va_arg(*args, int);
12564 elen = va_arg(*args, UV);
12565 /* if utf8 length is larger than 0x7ffff..., then it might
12566 * have been a signed value that wrapped */
12567 if (elen > ((~(STRLEN)0) >> 1)) {
12568 assert(0); /* in DEBUGGING build we want to crash */
12569 elen = 0; /* otherwise we want to treat this as an empty string */
12571 eptr = va_arg(*args, char *);
12572 q += sizeof(UTF8f) - 2;
12579 goto get_int_arg_val;
12590 goto get_int_arg_val;
12595 goto get_int_arg_val;
12606 goto get_int_arg_val;
12621 esignbuf[esignlen++] = plus;
12624 /* initialise the vector string to iterate over */
12626 vecsv = args ? va_arg(*args, SV*) : argsv;
12628 /* if this is a version object, we need to convert
12629 * back into v-string notation and then let the
12630 * vectorize happen normally
12632 if (sv_isobject(vecsv) && sv_derived_from(vecsv, "version")) {
12633 if ( hv_existss(MUTABLE_HV(SvRV(vecsv)), "alpha") ) {
12634 Perl_ck_warner_d(aTHX_ packWARN(WARN_PRINTF),
12635 "vector argument not supported with alpha versions");
12639 vecstr = (U8*)SvPV_const(vecsv,veclen);
12640 vecsv = sv_newmortal();
12641 scan_vstring((char *)vecstr, (char *)vecstr + veclen,
12645 vecstr = (U8*)SvPV_const(vecsv, veclen);
12646 vec_utf8 = DO_UTF8(vecsv);
12648 /* This is the re-entry point for when we're iterating
12649 * over the individual characters of a vector arg */
12652 goto done_valid_conversion;
12654 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
12664 /* test arg for inf/nan. This can trigger an unwanted
12665 * 'str' overload, so manually force 'num' overload first
12669 if (UNLIKELY(SvAMAGIC(argsv)))
12670 argsv = sv_2num(argsv);
12671 if (UNLIKELY(isinfnansv(argsv)))
12672 goto handle_infnan_argsv;
12676 /* signed int type */
12681 case 'c': iv = (char)va_arg(*args, int); break;
12682 case 'h': iv = (short)va_arg(*args, int); break;
12683 case 'l': iv = va_arg(*args, long); break;
12684 case 'V': iv = va_arg(*args, IV); break;
12685 case 'z': iv = va_arg(*args, SSize_t); break;
12686 #ifdef HAS_PTRDIFF_T
12687 case 't': iv = va_arg(*args, ptrdiff_t); break;
12689 default: iv = va_arg(*args, int); break;
12690 case 'j': iv = (IV) va_arg(*args, PERL_INTMAX_T); break;
12693 iv = va_arg(*args, Quad_t); break;
12700 /* assign to tiv then cast to iv to work around
12701 * 2003 GCC cast bug (gnu.org bugzilla #13488) */
12702 IV tiv = SvIV_nomg(argsv);
12704 case 'c': iv = (char)tiv; break;
12705 case 'h': iv = (short)tiv; break;
12706 case 'l': iv = (long)tiv; break;
12708 default: iv = tiv; break;
12711 iv = (Quad_t)tiv; break;
12718 /* now convert iv to uv */
12722 esignbuf[esignlen++] = plus;
12725 /* Using 0- here to silence bogus warning from MS VC */
12726 uv = (UV) (0 - (UV) iv);
12727 esignbuf[esignlen++] = '-';
12731 /* unsigned int type */
12734 case 'c': uv = (unsigned char)va_arg(*args, unsigned);
12736 case 'h': uv = (unsigned short)va_arg(*args, unsigned);
12738 case 'l': uv = va_arg(*args, unsigned long); break;
12739 case 'V': uv = va_arg(*args, UV); break;
12740 case 'z': uv = va_arg(*args, Size_t); break;
12741 #ifdef HAS_PTRDIFF_T
12742 /* will sign extend, but there is no
12743 * uptrdiff_t, so oh well */
12744 case 't': uv = va_arg(*args, ptrdiff_t); break;
12746 case 'j': uv = (UV) va_arg(*args, PERL_UINTMAX_T); break;
12747 default: uv = va_arg(*args, unsigned); break;
12750 uv = va_arg(*args, Uquad_t); break;
12757 /* assign to tiv then cast to iv to work around
12758 * 2003 GCC cast bug (gnu.org bugzilla #13488) */
12759 UV tuv = SvUV_nomg(argsv);
12761 case 'c': uv = (unsigned char)tuv; break;
12762 case 'h': uv = (unsigned short)tuv; break;
12763 case 'l': uv = (unsigned long)tuv; break;
12765 default: uv = tuv; break;
12768 uv = (Uquad_t)tuv; break;
12779 char *ptr = ebuf + sizeof ebuf;
12786 const char * const p =
12787 (c == 'X') ? PL_hexdigit + 16 : PL_hexdigit;
12792 } while (uv >>= 4);
12793 if (alt && *ptr != '0') {
12794 esignbuf[esignlen++] = '0';
12795 esignbuf[esignlen++] = c; /* 'x' or 'X' */
12802 *--ptr = '0' + dig;
12803 } while (uv >>= 3);
12804 if (alt && *ptr != '0')
12810 *--ptr = '0' + dig;
12811 } while (uv >>= 1);
12812 if (alt && *ptr != '0') {
12813 esignbuf[esignlen++] = '0';
12814 esignbuf[esignlen++] = c; /* 'b' or 'B' */
12819 /* special-case: base 1 indicates a 'c' format:
12820 * we use the common code for extracting a uv,
12821 * but handle that value differently here than
12822 * all the other int types */
12824 (!UVCHR_IS_INVARIANT(uv) && SvUTF8(sv)))
12827 assert(sizeof(ebuf) >= UTF8_MAXBYTES + 1);
12829 elen = uvchr_to_utf8((U8*)eptr, uv) - (U8*)ebuf;
12834 ebuf[0] = (char)uv;
12839 default: /* it had better be ten or less */
12842 *--ptr = '0' + dig;
12843 } while (uv /= base);
12846 elen = (ebuf + sizeof ebuf) - ptr;
12850 zeros = precis - elen;
12851 else if (precis == 0 && elen == 1 && *eptr == '0'
12852 && !(base == 8 && alt)) /* "%#.0o" prints "0" */
12855 /* a precision nullifies the 0 flag. */
12861 /* FLOATING POINT */
12864 c = 'f'; /* maybe %F isn't supported here */
12866 case 'e': case 'E':
12868 case 'g': case 'G':
12869 case 'a': case 'A':
12872 STRLEN float_need; /* what PL_efloatsize needs to become */
12873 bool hexfp; /* hexadecimal floating point? */
12875 vcatpvfn_long_double_t fv;
12878 /* This is evil, but floating point is even more evil */
12880 /* for SV-style calling, we can only get NV
12881 for C-style calling, we assume %f is double;
12882 for simplicity we allow any of %Lf, %llf, %qf for long double
12886 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12890 /* [perl #20339] - we should accept and ignore %lf rather than die */
12894 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12895 intsize = args ? 0 : 'q';
12899 #if defined(HAS_LONG_DOUBLE)
12912 /* Now we need (long double) if intsize == 'q', else (double). */
12914 /* Note: do not pull NVs off the va_list with va_arg()
12915 * (pull doubles instead) because if you have a build
12916 * with long doubles, you would always be pulling long
12917 * doubles, which would badly break anyone using only
12918 * doubles (i.e. the majority of builds). In other
12919 * words, you cannot mix doubles and long doubles.
12920 * The only case where you can pull off long doubles
12921 * is when the format specifier explicitly asks so with
12923 #ifdef USE_QUADMATH
12924 fv = intsize == 'q' ?
12925 va_arg(*args, NV) : va_arg(*args, double);
12927 #elif LONG_DOUBLESIZE > DOUBLESIZE
12928 if (intsize == 'q') {
12929 fv = va_arg(*args, long double);
12932 nv = va_arg(*args, double);
12933 VCATPVFN_NV_TO_FV(nv, fv);
12936 nv = va_arg(*args, double);
12943 /* we jump here if an int-ish format encountered an
12944 * infinite/Nan argsv. After setting nv/fv, it falls
12945 * into the isinfnan block which follows */
12946 handle_infnan_argsv:
12947 nv = SvNV_nomg(argsv);
12948 VCATPVFN_NV_TO_FV(nv, fv);
12951 if (Perl_isinfnan(nv)) {
12953 Perl_croak(aTHX_ "Cannot printf %" NVgf " with '%c'",
12954 SvNV_nomg(argsv), (int)c);
12956 elen = S_infnan_2pv(nv, ebuf, sizeof(ebuf), plus);
12965 /* special-case "%.0f" */
12969 && !(width || left || plus || alt)
12972 && ((eptr = F0convert(nv, ebuf + sizeof ebuf, &elen)))
12976 /* Determine the buffer size needed for the various
12977 * floating-point formats.
12979 * The basic possibilities are:
12982 * %f 1111111.123456789
12983 * %e 1.111111123e+06
12984 * %a 0x1.0f4471f9bp+20
12986 * %g 1.11111112e+15
12988 * where P is the value of the precision in the format, or 6
12989 * if not specified. Note the two possible output formats of
12990 * %g; in both cases the number of significant digits is <=
12993 * For most of the format types the maximum buffer size needed
12994 * is precision, plus: any leading 1 or 0x1, the radix
12995 * point, and an exponent. The difficult one is %f: for a
12996 * large positive exponent it can have many leading digits,
12997 * which needs to be calculated specially. Also %a is slightly
12998 * different in that in the absence of a specified precision,
12999 * it uses as many digits as necessary to distinguish
13000 * different values.
13002 * First, here are the constant bits. For ease of calculation
13003 * we over-estimate the needed buffer size, for example by
13004 * assuming all formats have an exponent and a leading 0x1.
13006 * Also for production use, add a little extra overhead for
13007 * safety's sake. Under debugging don't, as it means we're
13008 * more likely to quickly spot issues during development.
13011 float_need = 1 /* possible unary minus */
13012 + 4 /* "0x1" plus very unlikely carry */
13013 + 1 /* default radix point '.' */
13014 + 2 /* "e-", "p+" etc */
13015 + 6 /* exponent: up to 16383 (quad fp) */
13017 + 20 /* safety net */
13022 /* determine the radix point len, e.g. length(".") in "1.2" */
13023 #ifdef USE_LOCALE_NUMERIC
13024 /* note that we may either explicitly use PL_numeric_radix_sv
13025 * below, or implicitly, via an snprintf() variant.
13026 * Note also things like ps_AF.utf8 which has
13027 * "\N{ARABIC DECIMAL SEPARATOR} as a radix point */
13028 if (! have_in_lc_numeric) {
13029 in_lc_numeric = IN_LC(LC_NUMERIC);
13030 have_in_lc_numeric = TRUE;
13033 if (in_lc_numeric) {
13034 WITH_LC_NUMERIC_SET_TO_NEEDED_IN(TRUE, {
13035 /* this can't wrap unless PL_numeric_radix_sv is a string
13036 * consuming virtually all the 32-bit or 64-bit address
13039 float_need += (SvCUR(PL_numeric_radix_sv) - 1);
13041 /* floating-point formats only get utf8 if the radix point
13042 * is utf8. All other characters in the string are < 128
13043 * and so can be safely appended to both a non-utf8 and utf8
13045 * Note that this will convert the output to utf8 even if
13046 * the radix point didn't get output.
13048 if (SvUTF8(PL_numeric_radix_sv) && !has_utf8) {
13049 sv_utf8_upgrade(sv);
13058 if (isALPHA_FOLD_EQ(c, 'f')) {
13059 /* Determine how many digits before the radix point
13060 * might be emitted. frexp() (or frexpl) has some
13061 * unspecified behaviour for nan/inf/-inf, so lucky we've
13062 * already handled them above */
13064 int i = PERL_INT_MIN;
13065 (void)Perl_frexp((NV)fv, &i);
13066 if (i == PERL_INT_MIN)
13067 Perl_die(aTHX_ "panic: frexp: %" VCATPVFN_FV_GF, fv);
13070 digits = BIT_DIGITS(i);
13071 /* this can't overflow. 'digits' will only be a few
13072 * thousand even for the largest floating-point types.
13073 * And up until now float_need is just some small
13074 * constants plus radix len, which can't be in
13075 * overflow territory unless the radix SV is consuming
13076 * over 1/2 the address space */
13077 assert(float_need < ((STRLEN)~0) - digits);
13078 float_need += digits;
13081 else if (UNLIKELY(isALPHA_FOLD_EQ(c, 'a'))) {
13084 /* %a in the absence of precision may print as many
13085 * digits as needed to represent the entire mantissa
13087 * This estimate seriously overshoots in most cases,
13088 * but better the undershooting. Firstly, all bytes
13089 * of the NV are not mantissa, some of them are
13090 * exponent. Secondly, for the reasonably common
13091 * long doubles case, the "80-bit extended", two
13092 * or six bytes of the NV are unused. Also, we'll
13093 * still pick up an extra +6 from the default
13094 * precision calculation below. */
13096 #ifdef LONGDOUBLE_DOUBLEDOUBLE
13097 /* For the "double double", we need more.
13098 * Since each double has their own exponent, the
13099 * doubles may float (haha) rather far from each
13100 * other, and the number of required bits is much
13101 * larger, up to total of DOUBLEDOUBLE_MAXBITS bits.
13102 * See the definition of DOUBLEDOUBLE_MAXBITS.
13104 * Need 2 hexdigits for each byte. */
13105 (DOUBLEDOUBLE_MAXBITS/8 + 1) * 2;
13107 NVSIZE * 2; /* 2 hexdigits for each byte */
13109 /* see "this can't overflow" comment above */
13110 assert(float_need < ((STRLEN)~0) - digits);
13111 float_need += digits;
13114 /* special-case "%.<number>g" if it will fit in ebuf */
13116 && precis /* See earlier comment about buggy Gconvert
13117 when digits, aka precis, is 0 */
13119 /* check, in manner not involving wrapping, that it will
13121 && float_need < sizeof(ebuf)
13122 && sizeof(ebuf) - float_need > precis
13123 && !(width || left || plus || alt)
13127 WITH_LC_NUMERIC_SET_TO_NEEDED_IN(in_lc_numeric,
13128 SNPRINTF_G(fv, ebuf, sizeof(ebuf), precis)
13130 elen = strlen(ebuf);
13137 STRLEN pr = has_precis ? precis : 6; /* known default */
13138 /* this probably can't wrap, since precis is limited
13139 * to 1/4 address space size, but better safe than sorry
13141 if (float_need >= ((STRLEN)~0) - pr)
13142 croak_memory_wrap();
13146 if (float_need < width)
13147 float_need = width;
13149 if (float_need > INT_MAX) {
13150 /* snprintf() returns an int, and we use that return value,
13151 so die horribly if the expected size is too large for int
13153 Perl_croak(aTHX_ "Numeric format result too large");
13156 if (PL_efloatsize <= float_need) {
13157 /* PL_efloatbuf should be at least 1 greater than
13158 * float_need to allow a trailing \0 to be returned by
13159 * snprintf(). If we need to grow, overgrow for the
13160 * benefit of future generations */
13161 const STRLEN extra = 0x20;
13162 if (float_need >= ((STRLEN)~0) - extra)
13163 croak_memory_wrap();
13164 float_need += extra;
13165 Safefree(PL_efloatbuf);
13166 PL_efloatsize = float_need;
13167 Newx(PL_efloatbuf, PL_efloatsize, char);
13168 PL_efloatbuf[0] = '\0';
13171 if (UNLIKELY(hexfp)) {
13172 elen = S_format_hexfp(aTHX_ PL_efloatbuf, PL_efloatsize, c,
13173 nv, fv, has_precis, precis, width,
13174 alt, plus, left, fill, in_lc_numeric);
13177 char *ptr = ebuf + sizeof ebuf;
13180 #if defined(USE_QUADMATH)
13181 if (intsize == 'q') {
13185 /* FIXME: what to do if HAS_LONG_DOUBLE but not PERL_PRIfldbl? */
13186 #elif defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
13187 /* Note that this is HAS_LONG_DOUBLE and PERL_PRIfldbl,
13188 * not USE_LONG_DOUBLE and NVff. In other words,
13189 * this needs to work without USE_LONG_DOUBLE. */
13190 if (intsize == 'q') {
13191 /* Copy the one or more characters in a long double
13192 * format before the 'base' ([efgEFG]) character to
13193 * the format string. */
13194 static char const ldblf[] = PERL_PRIfldbl;
13195 char const *p = ldblf + sizeof(ldblf) - 3;
13196 while (p >= ldblf) { *--ptr = *p--; }
13201 do { *--ptr = '0' + (base % 10); } while (base /= 10);
13206 do { *--ptr = '0' + (base % 10); } while (base /= 10);
13218 /* No taint. Otherwise we are in the strange situation
13219 * where printf() taints but print($float) doesn't.
13222 /* hopefully the above makes ptr a very constrained format
13223 * that is safe to use, even though it's not literal */
13224 GCC_DIAG_IGNORE_STMT(-Wformat-nonliteral);
13225 #ifdef USE_QUADMATH
13227 if (!quadmath_format_valid(ptr))
13228 Perl_croak_nocontext("panic: quadmath invalid format \"%s\"", ptr);
13229 WITH_LC_NUMERIC_SET_TO_NEEDED_IN(in_lc_numeric,
13230 elen = quadmath_snprintf(PL_efloatbuf, PL_efloatsize,
13233 if ((IV)elen == -1) {
13234 Perl_croak_nocontext("panic: quadmath_snprintf failed, format \"%s\"", ptr);
13237 #elif defined(HAS_LONG_DOUBLE)
13238 WITH_LC_NUMERIC_SET_TO_NEEDED_IN(in_lc_numeric,
13239 elen = ((intsize == 'q')
13240 ? my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv)
13241 : my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, (double)fv))
13244 WITH_LC_NUMERIC_SET_TO_NEEDED_IN(in_lc_numeric,
13245 elen = my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv)
13248 GCC_DIAG_RESTORE_STMT;
13251 eptr = PL_efloatbuf;
13255 /* Since floating-point formats do their own formatting and
13256 * padding, we skip the main block of code at the end of this
13257 * loop which handles appending eptr to sv, and do our own
13258 * stripped-down version */
13263 assert(elen >= width);
13265 S_sv_catpvn_simple(aTHX_ sv, eptr, elen);
13267 goto done_valid_conversion;
13275 /* XXX ideally we should warn if any flags etc have been
13276 * set, e.g. "%-4.5n" */
13277 /* XXX if sv was originally non-utf8 with a char in the
13278 * range 0x80-0xff, then if it got upgraded, we should
13279 * calculate char len rather than byte len here */
13280 len = SvCUR(sv) - origlen;
13282 int i = (len > PERL_INT_MAX) ? PERL_INT_MAX : (int)len;
13285 case 'c': *(va_arg(*args, char*)) = i; break;
13286 case 'h': *(va_arg(*args, short*)) = i; break;
13287 default: *(va_arg(*args, int*)) = i; break;
13288 case 'l': *(va_arg(*args, long*)) = i; break;
13289 case 'V': *(va_arg(*args, IV*)) = i; break;
13290 case 'z': *(va_arg(*args, SSize_t*)) = i; break;
13291 #ifdef HAS_PTRDIFF_T
13292 case 't': *(va_arg(*args, ptrdiff_t*)) = i; break;
13294 case 'j': *(va_arg(*args, PERL_INTMAX_T*)) = i; break;
13297 *(va_arg(*args, Quad_t*)) = i; break;
13305 Perl_croak_nocontext(
13306 "Missing argument for %%n in %s",
13307 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
13308 sv_setuv_mg(argsv, has_utf8
13309 ? (UV)utf8_length((U8*)SvPVX(sv), (U8*)SvEND(sv))
13312 goto done_valid_conversion;
13320 && (PL_op->op_type == OP_PRTF || PL_op->op_type == OP_SPRINTF)
13321 && ckWARN(WARN_PRINTF))
13323 SV * const msg = sv_newmortal();
13324 Perl_sv_setpvf(aTHX_ msg, "Invalid conversion in %sprintf: ",
13325 (PL_op->op_type == OP_PRTF) ? "" : "s");
13326 if (fmtstart < patend) {
13327 const char * const fmtend = q < patend ? q : patend;
13329 sv_catpvs(msg, "\"%");
13330 for (f = fmtstart; f < fmtend; f++) {
13332 sv_catpvn_nomg(msg, f, 1);
13334 Perl_sv_catpvf(aTHX_ msg,
13335 "\\%03" UVof, (UV)*f & 0xFF);
13338 sv_catpvs(msg, "\"");
13340 sv_catpvs(msg, "end of string");
13342 Perl_warner(aTHX_ packWARN(WARN_PRINTF), "%" SVf, SVfARG(msg)); /* yes, this is reentrant */
13345 /* mangled format: output the '%', then continue from the
13346 * character following that */
13347 sv_catpvn_nomg(sv, fmtstart-1, 1);
13350 /* Any "redundant arg" warning from now onwards will probably
13351 * just be misleading, so don't bother. */
13352 no_redundant_warning = TRUE;
13353 continue; /* not "break" */
13356 if (is_utf8 != has_utf8) {
13359 sv_utf8_upgrade(sv);
13362 const STRLEN old_elen = elen;
13363 SV * const nsv = newSVpvn_flags(eptr, elen, SVs_TEMP);
13364 sv_utf8_upgrade(nsv);
13365 eptr = SvPVX_const(nsv);
13368 if (width) { /* fudge width (can't fudge elen) */
13369 width += elen - old_elen;
13376 /* append esignbuf, filler, zeros, eptr and dotstr to sv */
13379 STRLEN need, have, gap;
13383 /* signed value that's wrapped? */
13384 assert(elen <= ((~(STRLEN)0) >> 1));
13386 /* if zeros is non-zero, then it represents filler between
13387 * elen and precis. So adding elen and zeros together will
13388 * always be <= precis, and the addition can never wrap */
13389 assert(!zeros || (precis > elen && precis - elen == zeros));
13390 have = elen + zeros;
13392 if (have >= (((STRLEN)~0) - esignlen))
13393 croak_memory_wrap();
13396 need = (have > width ? have : width);
13399 if (need >= (((STRLEN)~0) - (SvCUR(sv) + 1)))
13400 croak_memory_wrap();
13401 need += (SvCUR(sv) + 1);
13408 for (i = 0; i < esignlen; i++)
13409 *s++ = esignbuf[i];
13410 for (i = zeros; i; i--)
13412 Copy(eptr, s, elen, char);
13414 for (i = gap; i; i--)
13419 for (i = 0; i < esignlen; i++)
13420 *s++ = esignbuf[i];
13425 for (i = gap; i; i--)
13427 for (i = 0; i < esignlen; i++)
13428 *s++ = esignbuf[i];
13431 for (i = zeros; i; i--)
13433 Copy(eptr, s, elen, char);
13438 SvCUR_set(sv, s - SvPVX_const(sv));
13446 if (vectorize && veclen) {
13447 /* we append the vector separator separately since %v isn't
13448 * very common: don't slow down the general case by adding
13449 * dotstrlen to need etc */
13450 sv_catpvn_nomg(sv, dotstr, dotstrlen);
13452 goto vector; /* do next iteration */
13455 done_valid_conversion:
13458 S_warn_vcatpvfn_missing_argument(aTHX);
13461 /* Now that we've consumed all our printf format arguments (svix)
13462 * do we have things left on the stack that we didn't use?
13464 if (!no_redundant_warning && sv_count >= svix + 1 && ckWARN(WARN_REDUNDANT)) {
13465 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
13466 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
13469 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
13470 /* while we shouldn't set the cache, it may have been previously
13471 set in the caller, so clear it */
13472 MAGIC *mg = mg_find(sv, PERL_MAGIC_utf8);
13474 magic_setutf8(sv,mg); /* clear UTF8 cache */
13479 /* =========================================================================
13481 =head1 Cloning an interpreter
13485 All the macros and functions in this section are for the private use of
13486 the main function, perl_clone().
13488 The foo_dup() functions make an exact copy of an existing foo thingy.
13489 During the course of a cloning, a hash table is used to map old addresses
13490 to new addresses. The table is created and manipulated with the
13491 ptr_table_* functions.
13493 * =========================================================================*/
13496 #if defined(USE_ITHREADS)
13498 /* XXX Remove this so it doesn't have to go thru the macro and return for nothing */
13499 #ifndef GpREFCNT_inc
13500 # define GpREFCNT_inc(gp) ((gp) ? (++(gp)->gp_refcnt, (gp)) : (GP*)NULL)
13504 /* Certain cases in Perl_ss_dup have been merged, by relying on the fact
13505 that currently av_dup, gv_dup and hv_dup are the same as sv_dup.
13506 If this changes, please unmerge ss_dup.
13507 Likewise, sv_dup_inc_multiple() relies on this fact. */
13508 #define sv_dup_inc_NN(s,t) SvREFCNT_inc_NN(sv_dup_inc(s,t))
13509 #define av_dup(s,t) MUTABLE_AV(sv_dup((const SV *)s,t))
13510 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
13511 #define hv_dup(s,t) MUTABLE_HV(sv_dup((const SV *)s,t))
13512 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
13513 #define cv_dup(s,t) MUTABLE_CV(sv_dup((const SV *)s,t))
13514 #define cv_dup_inc(s,t) MUTABLE_CV(sv_dup_inc((const SV *)s,t))
13515 #define io_dup(s,t) MUTABLE_IO(sv_dup((const SV *)s,t))
13516 #define io_dup_inc(s,t) MUTABLE_IO(sv_dup_inc((const SV *)s,t))
13517 #define gv_dup(s,t) MUTABLE_GV(sv_dup((const SV *)s,t))
13518 #define gv_dup_inc(s,t) MUTABLE_GV(sv_dup_inc((const SV *)s,t))
13519 #define SAVEPV(p) ((p) ? savepv(p) : NULL)
13520 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
13522 /* clone a parser */
13525 Perl_parser_dup(pTHX_ const yy_parser *const proto, CLONE_PARAMS *const param)
13529 PERL_ARGS_ASSERT_PARSER_DUP;
13534 /* look for it in the table first */
13535 parser = (yy_parser *)ptr_table_fetch(PL_ptr_table, proto);
13539 /* create anew and remember what it is */
13540 Newxz(parser, 1, yy_parser);
13541 ptr_table_store(PL_ptr_table, proto, parser);
13543 /* XXX eventually, just Copy() most of the parser struct ? */
13545 parser->lex_brackets = proto->lex_brackets;
13546 parser->lex_casemods = proto->lex_casemods;
13547 parser->lex_brackstack = savepvn(proto->lex_brackstack,
13548 (proto->lex_brackets < 120 ? 120 : proto->lex_brackets));
13549 parser->lex_casestack = savepvn(proto->lex_casestack,
13550 (proto->lex_casemods < 12 ? 12 : proto->lex_casemods));
13551 parser->lex_defer = proto->lex_defer;
13552 parser->lex_dojoin = proto->lex_dojoin;
13553 parser->lex_formbrack = proto->lex_formbrack;
13554 parser->lex_inpat = proto->lex_inpat;
13555 parser->lex_inwhat = proto->lex_inwhat;
13556 parser->lex_op = proto->lex_op;
13557 parser->lex_repl = sv_dup_inc(proto->lex_repl, param);
13558 parser->lex_starts = proto->lex_starts;
13559 parser->lex_stuff = sv_dup_inc(proto->lex_stuff, param);
13560 parser->multi_close = proto->multi_close;
13561 parser->multi_open = proto->multi_open;
13562 parser->multi_start = proto->multi_start;
13563 parser->multi_end = proto->multi_end;
13564 parser->preambled = proto->preambled;
13565 parser->lex_super_state = proto->lex_super_state;
13566 parser->lex_sub_inwhat = proto->lex_sub_inwhat;
13567 parser->lex_sub_op = proto->lex_sub_op;
13568 parser->lex_sub_repl= sv_dup_inc(proto->lex_sub_repl, param);
13569 parser->linestr = sv_dup_inc(proto->linestr, param);
13570 parser->expect = proto->expect;
13571 parser->copline = proto->copline;
13572 parser->last_lop_op = proto->last_lop_op;
13573 parser->lex_state = proto->lex_state;
13574 parser->rsfp = fp_dup(proto->rsfp, '<', param);
13575 /* rsfp_filters entries have fake IoDIRP() */
13576 parser->rsfp_filters= av_dup_inc(proto->rsfp_filters, param);
13577 parser->in_my = proto->in_my;
13578 parser->in_my_stash = hv_dup(proto->in_my_stash, param);
13579 parser->error_count = proto->error_count;
13580 parser->sig_elems = proto->sig_elems;
13581 parser->sig_optelems= proto->sig_optelems;
13582 parser->sig_slurpy = proto->sig_slurpy;
13583 parser->recheck_utf8_validity = proto->recheck_utf8_validity;
13586 char * const ols = SvPVX(proto->linestr);
13587 char * const ls = SvPVX(parser->linestr);
13589 parser->bufptr = ls + (proto->bufptr >= ols ?
13590 proto->bufptr - ols : 0);
13591 parser->oldbufptr = ls + (proto->oldbufptr >= ols ?
13592 proto->oldbufptr - ols : 0);
13593 parser->oldoldbufptr= ls + (proto->oldoldbufptr >= ols ?
13594 proto->oldoldbufptr - ols : 0);
13595 parser->linestart = ls + (proto->linestart >= ols ?
13596 proto->linestart - ols : 0);
13597 parser->last_uni = ls + (proto->last_uni >= ols ?
13598 proto->last_uni - ols : 0);
13599 parser->last_lop = ls + (proto->last_lop >= ols ?
13600 proto->last_lop - ols : 0);
13602 parser->bufend = ls + SvCUR(parser->linestr);
13605 Copy(proto->tokenbuf, parser->tokenbuf, 256, char);
13608 Copy(proto->nextval, parser->nextval, 5, YYSTYPE);
13609 Copy(proto->nexttype, parser->nexttype, 5, I32);
13610 parser->nexttoke = proto->nexttoke;
13612 /* XXX should clone saved_curcop here, but we aren't passed
13613 * proto_perl; so do it in perl_clone_using instead */
13619 /* duplicate a file handle */
13622 Perl_fp_dup(pTHX_ PerlIO *const fp, const char type, CLONE_PARAMS *const param)
13626 PERL_ARGS_ASSERT_FP_DUP;
13627 PERL_UNUSED_ARG(type);
13630 return (PerlIO*)NULL;
13632 /* look for it in the table first */
13633 ret = (PerlIO*)ptr_table_fetch(PL_ptr_table, fp);
13637 /* create anew and remember what it is */
13638 #ifdef __amigaos4__
13639 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE|PERLIO_DUP_FD);
13641 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE);
13643 ptr_table_store(PL_ptr_table, fp, ret);
13647 /* duplicate a directory handle */
13650 Perl_dirp_dup(pTHX_ DIR *const dp, CLONE_PARAMS *const param)
13654 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
13656 const Direntry_t *dirent;
13657 char smallbuf[256]; /* XXX MAXPATHLEN, surely? */
13663 PERL_UNUSED_CONTEXT;
13664 PERL_ARGS_ASSERT_DIRP_DUP;
13669 /* look for it in the table first */
13670 ret = (DIR*)ptr_table_fetch(PL_ptr_table, dp);
13674 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
13676 PERL_UNUSED_ARG(param);
13680 /* open the current directory (so we can switch back) */
13681 if (!(pwd = PerlDir_open("."))) return (DIR *)NULL;
13683 /* chdir to our dir handle and open the present working directory */
13684 if (fchdir(my_dirfd(dp)) < 0 || !(ret = PerlDir_open("."))) {
13685 PerlDir_close(pwd);
13686 return (DIR *)NULL;
13688 /* Now we should have two dir handles pointing to the same dir. */
13690 /* Be nice to the calling code and chdir back to where we were. */
13691 /* XXX If this fails, then what? */
13692 PERL_UNUSED_RESULT(fchdir(my_dirfd(pwd)));
13694 /* We have no need of the pwd handle any more. */
13695 PerlDir_close(pwd);
13698 # define d_namlen(d) (d)->d_namlen
13700 # define d_namlen(d) strlen((d)->d_name)
13702 /* Iterate once through dp, to get the file name at the current posi-
13703 tion. Then step back. */
13704 pos = PerlDir_tell(dp);
13705 if ((dirent = PerlDir_read(dp))) {
13706 len = d_namlen(dirent);
13707 if (len > sizeof(dirent->d_name) && sizeof(dirent->d_name) > PTRSIZE) {
13708 /* If the len is somehow magically longer than the
13709 * maximum length of the directory entry, even though
13710 * we could fit it in a buffer, we could not copy it
13711 * from the dirent. Bail out. */
13712 PerlDir_close(ret);
13715 if (len <= sizeof smallbuf) name = smallbuf;
13716 else Newx(name, len, char);
13717 Move(dirent->d_name, name, len, char);
13719 PerlDir_seek(dp, pos);
13721 /* Iterate through the new dir handle, till we find a file with the
13723 if (!dirent) /* just before the end */
13725 pos = PerlDir_tell(ret);
13726 if (PerlDir_read(ret)) continue; /* not there yet */
13727 PerlDir_seek(ret, pos); /* step back */
13731 const long pos0 = PerlDir_tell(ret);
13733 pos = PerlDir_tell(ret);
13734 if ((dirent = PerlDir_read(ret))) {
13735 if (len == (STRLEN)d_namlen(dirent)
13736 && memEQ(name, dirent->d_name, len)) {
13738 PerlDir_seek(ret, pos); /* step back */
13741 /* else we are not there yet; keep iterating */
13743 else { /* This is not meant to happen. The best we can do is
13744 reset the iterator to the beginning. */
13745 PerlDir_seek(ret, pos0);
13752 if (name && name != smallbuf)
13757 ret = win32_dirp_dup(dp, param);
13760 /* pop it in the pointer table */
13762 ptr_table_store(PL_ptr_table, dp, ret);
13767 /* duplicate a typeglob */
13770 Perl_gp_dup(pTHX_ GP *const gp, CLONE_PARAMS *const param)
13774 PERL_ARGS_ASSERT_GP_DUP;
13778 /* look for it in the table first */
13779 ret = (GP*)ptr_table_fetch(PL_ptr_table, gp);
13783 /* create anew and remember what it is */
13785 ptr_table_store(PL_ptr_table, gp, ret);
13788 /* ret->gp_refcnt must be 0 before any other dups are called. We're relying
13789 on Newxz() to do this for us. */
13790 ret->gp_sv = sv_dup_inc(gp->gp_sv, param);
13791 ret->gp_io = io_dup_inc(gp->gp_io, param);
13792 ret->gp_form = cv_dup_inc(gp->gp_form, param);
13793 ret->gp_av = av_dup_inc(gp->gp_av, param);
13794 ret->gp_hv = hv_dup_inc(gp->gp_hv, param);
13795 ret->gp_egv = gv_dup(gp->gp_egv, param);/* GvEGV is not refcounted */
13796 ret->gp_cv = cv_dup_inc(gp->gp_cv, param);
13797 ret->gp_cvgen = gp->gp_cvgen;
13798 ret->gp_line = gp->gp_line;
13799 ret->gp_file_hek = hek_dup(gp->gp_file_hek, param);
13803 /* duplicate a chain of magic */
13806 Perl_mg_dup(pTHX_ MAGIC *mg, CLONE_PARAMS *const param)
13808 MAGIC *mgret = NULL;
13809 MAGIC **mgprev_p = &mgret;
13811 PERL_ARGS_ASSERT_MG_DUP;
13813 for (; mg; mg = mg->mg_moremagic) {
13816 if ((param->flags & CLONEf_JOIN_IN)
13817 && mg->mg_type == PERL_MAGIC_backref)
13818 /* when joining, we let the individual SVs add themselves to
13819 * backref as needed. */
13822 Newx(nmg, 1, MAGIC);
13824 mgprev_p = &(nmg->mg_moremagic);
13826 /* There was a comment "XXX copy dynamic vtable?" but as we don't have
13827 dynamic vtables, I'm not sure why Sarathy wrote it. The comment dates
13828 from the original commit adding Perl_mg_dup() - revision 4538.
13829 Similarly there is the annotation "XXX random ptr?" next to the
13830 assignment to nmg->mg_ptr. */
13833 /* FIXME for plugins
13834 if (nmg->mg_type == PERL_MAGIC_qr) {
13835 nmg->mg_obj = MUTABLE_SV(CALLREGDUPE((REGEXP*)nmg->mg_obj, param));
13839 nmg->mg_obj = (nmg->mg_flags & MGf_REFCOUNTED)
13840 ? nmg->mg_type == PERL_MAGIC_backref
13841 /* The backref AV has its reference
13842 * count deliberately bumped by 1 */
13843 ? SvREFCNT_inc(av_dup_inc((const AV *)
13844 nmg->mg_obj, param))
13845 : sv_dup_inc(nmg->mg_obj, param)
13846 : (nmg->mg_type == PERL_MAGIC_regdatum ||
13847 nmg->mg_type == PERL_MAGIC_regdata)
13849 : sv_dup(nmg->mg_obj, param);
13851 if (nmg->mg_ptr && nmg->mg_type != PERL_MAGIC_regex_global) {
13852 if (nmg->mg_len > 0) {
13853 nmg->mg_ptr = SAVEPVN(nmg->mg_ptr, nmg->mg_len);
13854 if (nmg->mg_type == PERL_MAGIC_overload_table &&
13855 AMT_AMAGIC((AMT*)nmg->mg_ptr))
13857 AMT * const namtp = (AMT*)nmg->mg_ptr;
13858 sv_dup_inc_multiple((SV**)(namtp->table),
13859 (SV**)(namtp->table), NofAMmeth, param);
13862 else if (nmg->mg_len == HEf_SVKEY)
13863 nmg->mg_ptr = (char*)sv_dup_inc((const SV *)nmg->mg_ptr, param);
13865 if ((nmg->mg_flags & MGf_DUP) && nmg->mg_virtual && nmg->mg_virtual->svt_dup) {
13866 nmg->mg_virtual->svt_dup(aTHX_ nmg, param);
13872 #endif /* USE_ITHREADS */
13874 struct ptr_tbl_arena {
13875 struct ptr_tbl_arena *next;
13876 struct ptr_tbl_ent array[1023/3]; /* as ptr_tbl_ent has 3 pointers. */
13879 /* create a new pointer-mapping table */
13882 Perl_ptr_table_new(pTHX)
13885 PERL_UNUSED_CONTEXT;
13887 Newx(tbl, 1, PTR_TBL_t);
13888 tbl->tbl_max = 511;
13889 tbl->tbl_items = 0;
13890 tbl->tbl_arena = NULL;
13891 tbl->tbl_arena_next = NULL;
13892 tbl->tbl_arena_end = NULL;
13893 Newxz(tbl->tbl_ary, tbl->tbl_max + 1, PTR_TBL_ENT_t*);
13897 #define PTR_TABLE_HASH(ptr) \
13898 ((PTR2UV(ptr) >> 3) ^ (PTR2UV(ptr) >> (3 + 7)) ^ (PTR2UV(ptr) >> (3 + 17)))
13900 /* map an existing pointer using a table */
13902 STATIC PTR_TBL_ENT_t *
13903 S_ptr_table_find(PTR_TBL_t *const tbl, const void *const sv)
13905 PTR_TBL_ENT_t *tblent;
13906 const UV hash = PTR_TABLE_HASH(sv);
13908 PERL_ARGS_ASSERT_PTR_TABLE_FIND;
13910 tblent = tbl->tbl_ary[hash & tbl->tbl_max];
13911 for (; tblent; tblent = tblent->next) {
13912 if (tblent->oldval == sv)
13919 Perl_ptr_table_fetch(pTHX_ PTR_TBL_t *const tbl, const void *const sv)
13921 PTR_TBL_ENT_t const *const tblent = ptr_table_find(tbl, sv);
13923 PERL_ARGS_ASSERT_PTR_TABLE_FETCH;
13924 PERL_UNUSED_CONTEXT;
13926 return tblent ? tblent->newval : NULL;
13929 /* add a new entry to a pointer-mapping table 'tbl'. In hash terms, 'oldsv' is
13930 * the key; 'newsv' is the value. The names "old" and "new" are specific to
13931 * the core's typical use of ptr_tables in thread cloning. */
13934 Perl_ptr_table_store(pTHX_ PTR_TBL_t *const tbl, const void *const oldsv, void *const newsv)
13936 PTR_TBL_ENT_t *tblent = ptr_table_find(tbl, oldsv);
13938 PERL_ARGS_ASSERT_PTR_TABLE_STORE;
13939 PERL_UNUSED_CONTEXT;
13942 tblent->newval = newsv;
13944 const UV entry = PTR_TABLE_HASH(oldsv) & tbl->tbl_max;
13946 if (tbl->tbl_arena_next == tbl->tbl_arena_end) {
13947 struct ptr_tbl_arena *new_arena;
13949 Newx(new_arena, 1, struct ptr_tbl_arena);
13950 new_arena->next = tbl->tbl_arena;
13951 tbl->tbl_arena = new_arena;
13952 tbl->tbl_arena_next = new_arena->array;
13953 tbl->tbl_arena_end = C_ARRAY_END(new_arena->array);
13956 tblent = tbl->tbl_arena_next++;
13958 tblent->oldval = oldsv;
13959 tblent->newval = newsv;
13960 tblent->next = tbl->tbl_ary[entry];
13961 tbl->tbl_ary[entry] = tblent;
13963 if (tblent->next && tbl->tbl_items > tbl->tbl_max)
13964 ptr_table_split(tbl);
13968 /* double the hash bucket size of an existing ptr table */
13971 Perl_ptr_table_split(pTHX_ PTR_TBL_t *const tbl)
13973 PTR_TBL_ENT_t **ary = tbl->tbl_ary;
13974 const UV oldsize = tbl->tbl_max + 1;
13975 UV newsize = oldsize * 2;
13978 PERL_ARGS_ASSERT_PTR_TABLE_SPLIT;
13979 PERL_UNUSED_CONTEXT;
13981 Renew(ary, newsize, PTR_TBL_ENT_t*);
13982 Zero(&ary[oldsize], newsize-oldsize, PTR_TBL_ENT_t*);
13983 tbl->tbl_max = --newsize;
13984 tbl->tbl_ary = ary;
13985 for (i=0; i < oldsize; i++, ary++) {
13986 PTR_TBL_ENT_t **entp = ary;
13987 PTR_TBL_ENT_t *ent = *ary;
13988 PTR_TBL_ENT_t **curentp;
13991 curentp = ary + oldsize;
13993 if ((newsize & PTR_TABLE_HASH(ent->oldval)) != i) {
13995 ent->next = *curentp;
14005 /* remove all the entries from a ptr table */
14006 /* Deprecated - will be removed post 5.14 */
14009 Perl_ptr_table_clear(pTHX_ PTR_TBL_t *const tbl)
14011 PERL_UNUSED_CONTEXT;
14012 if (tbl && tbl->tbl_items) {
14013 struct ptr_tbl_arena *arena = tbl->tbl_arena;
14015 Zero(tbl->tbl_ary, tbl->tbl_max + 1, struct ptr_tbl_ent *);
14018 struct ptr_tbl_arena *next = arena->next;
14024 tbl->tbl_items = 0;
14025 tbl->tbl_arena = NULL;
14026 tbl->tbl_arena_next = NULL;
14027 tbl->tbl_arena_end = NULL;
14031 /* clear and free a ptr table */
14034 Perl_ptr_table_free(pTHX_ PTR_TBL_t *const tbl)
14036 struct ptr_tbl_arena *arena;
14038 PERL_UNUSED_CONTEXT;
14044 arena = tbl->tbl_arena;
14047 struct ptr_tbl_arena *next = arena->next;
14053 Safefree(tbl->tbl_ary);
14057 #if defined(USE_ITHREADS)
14060 Perl_rvpv_dup(pTHX_ SV *const dstr, const SV *const sstr, CLONE_PARAMS *const param)
14062 PERL_ARGS_ASSERT_RVPV_DUP;
14064 assert(!isREGEXP(sstr));
14066 if (SvWEAKREF(sstr)) {
14067 SvRV_set(dstr, sv_dup(SvRV_const(sstr), param));
14068 if (param->flags & CLONEf_JOIN_IN) {
14069 /* if joining, we add any back references individually rather
14070 * than copying the whole backref array */
14071 Perl_sv_add_backref(aTHX_ SvRV(dstr), dstr);
14075 SvRV_set(dstr, sv_dup_inc(SvRV_const(sstr), param));
14077 else if (SvPVX_const(sstr)) {
14078 /* Has something there */
14080 /* Normal PV - clone whole allocated space */
14081 SvPV_set(dstr, SAVEPVN(SvPVX_const(sstr), SvLEN(sstr)-1));
14082 /* sstr may not be that normal, but actually copy on write.
14083 But we are a true, independent SV, so: */
14087 /* Special case - not normally malloced for some reason */
14088 if (isGV_with_GP(sstr)) {
14089 /* Don't need to do anything here. */
14091 else if ((SvIsCOW(sstr))) {
14092 /* A "shared" PV - clone it as "shared" PV */
14094 HEK_KEY(hek_dup(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)),
14098 /* Some other special case - random pointer */
14099 SvPV_set(dstr, (char *) SvPVX_const(sstr));
14104 /* Copy the NULL */
14105 SvPV_set(dstr, NULL);
14109 /* duplicate a list of SVs. source and dest may point to the same memory. */
14111 S_sv_dup_inc_multiple(pTHX_ SV *const *source, SV **dest,
14112 SSize_t items, CLONE_PARAMS *const param)
14114 PERL_ARGS_ASSERT_SV_DUP_INC_MULTIPLE;
14116 while (items-- > 0) {
14117 *dest++ = sv_dup_inc(*source++, param);
14123 /* duplicate an SV of any type (including AV, HV etc) */
14126 S_sv_dup_common(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14131 PERL_ARGS_ASSERT_SV_DUP_COMMON;
14133 if (SvTYPE(sstr) == (svtype)SVTYPEMASK) {
14134 #ifdef DEBUG_LEAKING_SCALARS_ABORT
14139 /* look for it in the table first */
14140 dstr = MUTABLE_SV(ptr_table_fetch(PL_ptr_table, sstr));
14144 if(param->flags & CLONEf_JOIN_IN) {
14145 /** We are joining here so we don't want do clone
14146 something that is bad **/
14147 if (SvTYPE(sstr) == SVt_PVHV) {
14148 const HEK * const hvname = HvNAME_HEK(sstr);
14150 /** don't clone stashes if they already exist **/
14151 dstr = MUTABLE_SV(gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
14152 HEK_UTF8(hvname) ? SVf_UTF8 : 0));
14153 ptr_table_store(PL_ptr_table, sstr, dstr);
14157 else if (SvTYPE(sstr) == SVt_PVGV && !SvFAKE(sstr)) {
14158 HV *stash = GvSTASH(sstr);
14159 const HEK * hvname;
14160 if (stash && (hvname = HvNAME_HEK(stash))) {
14161 /** don't clone GVs if they already exist **/
14163 stash = gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
14164 HEK_UTF8(hvname) ? SVf_UTF8 : 0);
14166 stash, GvNAME(sstr),
14172 if (svp && *svp && SvTYPE(*svp) == SVt_PVGV) {
14173 ptr_table_store(PL_ptr_table, sstr, *svp);
14180 /* create anew and remember what it is */
14183 #ifdef DEBUG_LEAKING_SCALARS
14184 dstr->sv_debug_optype = sstr->sv_debug_optype;
14185 dstr->sv_debug_line = sstr->sv_debug_line;
14186 dstr->sv_debug_inpad = sstr->sv_debug_inpad;
14187 dstr->sv_debug_parent = (SV*)sstr;
14188 FREE_SV_DEBUG_FILE(dstr);
14189 dstr->sv_debug_file = savesharedpv(sstr->sv_debug_file);
14192 ptr_table_store(PL_ptr_table, sstr, dstr);
14195 SvFLAGS(dstr) = SvFLAGS(sstr);
14196 SvFLAGS(dstr) &= ~SVf_OOK; /* don't propagate OOK hack */
14197 SvREFCNT(dstr) = 0; /* must be before any other dups! */
14200 if (SvANY(sstr) && PL_watch_pvx && SvPVX_const(sstr) == PL_watch_pvx)
14201 PerlIO_printf(Perl_debug_log, "watch at %p hit, found string \"%s\"\n",
14202 (void*)PL_watch_pvx, SvPVX_const(sstr));
14205 /* don't clone objects whose class has asked us not to */
14207 && ! (SvFLAGS(SvSTASH(sstr)) & SVphv_CLONEABLE))
14213 switch (SvTYPE(sstr)) {
14215 SvANY(dstr) = NULL;
14218 SET_SVANY_FOR_BODYLESS_IV(dstr);
14220 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
14222 SvIV_set(dstr, SvIVX(sstr));
14226 #if NVSIZE <= IVSIZE
14227 SET_SVANY_FOR_BODYLESS_NV(dstr);
14229 SvANY(dstr) = new_XNV();
14231 SvNV_set(dstr, SvNVX(sstr));
14235 /* These are all the types that need complex bodies allocating. */
14237 const svtype sv_type = SvTYPE(sstr);
14238 const struct body_details *const sv_type_details
14239 = bodies_by_type + sv_type;
14243 Perl_croak(aTHX_ "Bizarre SvTYPE [%" IVdf "]", (IV)SvTYPE(sstr));
14244 NOT_REACHED; /* NOTREACHED */
14260 assert(sv_type_details->body_size);
14261 if (sv_type_details->arena) {
14262 new_body_inline(new_body, sv_type);
14264 = (void*)((char*)new_body - sv_type_details->offset);
14266 new_body = new_NOARENA(sv_type_details);
14270 SvANY(dstr) = new_body;
14273 Copy(((char*)SvANY(sstr)) + sv_type_details->offset,
14274 ((char*)SvANY(dstr)) + sv_type_details->offset,
14275 sv_type_details->copy, char);
14277 Copy(((char*)SvANY(sstr)),
14278 ((char*)SvANY(dstr)),
14279 sv_type_details->body_size + sv_type_details->offset, char);
14282 if (sv_type != SVt_PVAV && sv_type != SVt_PVHV
14283 && !isGV_with_GP(dstr)
14285 && !(sv_type == SVt_PVIO && !(IoFLAGS(dstr) & IOf_FAKE_DIRP)))
14286 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
14288 /* The Copy above means that all the source (unduplicated) pointers
14289 are now in the destination. We can check the flags and the
14290 pointers in either, but it's possible that there's less cache
14291 missing by always going for the destination.
14292 FIXME - instrument and check that assumption */
14293 if (sv_type >= SVt_PVMG) {
14295 SvMAGIC_set(dstr, mg_dup(SvMAGIC(dstr), param));
14296 if (SvOBJECT(dstr) && SvSTASH(dstr))
14297 SvSTASH_set(dstr, hv_dup_inc(SvSTASH(dstr), param));
14298 else SvSTASH_set(dstr, 0); /* don't copy DESTROY cache */
14301 /* The cast silences a GCC warning about unhandled types. */
14302 switch ((int)sv_type) {
14313 /* FIXME for plugins */
14314 re_dup_guts((REGEXP*) sstr, (REGEXP*) dstr, param);
14317 /* XXX LvTARGOFF sometimes holds PMOP* when DEBUGGING */
14318 if (LvTYPE(dstr) == 't') /* for tie: unrefcnted fake (SV**) */
14319 LvTARG(dstr) = dstr;
14320 else if (LvTYPE(dstr) == 'T') /* for tie: fake HE */
14321 LvTARG(dstr) = MUTABLE_SV(he_dup((HE*)LvTARG(dstr), 0, param));
14323 LvTARG(dstr) = sv_dup_inc(LvTARG(dstr), param);
14324 if (isREGEXP(sstr)) goto duprex;
14327 /* non-GP case already handled above */
14328 if(isGV_with_GP(sstr)) {
14329 GvNAME_HEK(dstr) = hek_dup(GvNAME_HEK(dstr), param);
14330 /* Don't call sv_add_backref here as it's going to be
14331 created as part of the magic cloning of the symbol
14332 table--unless this is during a join and the stash
14333 is not actually being cloned. */
14334 /* Danger Will Robinson - GvGP(dstr) isn't initialised
14335 at the point of this comment. */
14336 GvSTASH(dstr) = hv_dup(GvSTASH(dstr), param);
14337 if (param->flags & CLONEf_JOIN_IN)
14338 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
14339 GvGP_set(dstr, gp_dup(GvGP(sstr), param));
14340 (void)GpREFCNT_inc(GvGP(dstr));
14344 /* PL_parser->rsfp_filters entries have fake IoDIRP() */
14345 if(IoFLAGS(dstr) & IOf_FAKE_DIRP) {
14346 /* I have no idea why fake dirp (rsfps)
14347 should be treated differently but otherwise
14348 we end up with leaks -- sky*/
14349 IoTOP_GV(dstr) = gv_dup_inc(IoTOP_GV(dstr), param);
14350 IoFMT_GV(dstr) = gv_dup_inc(IoFMT_GV(dstr), param);
14351 IoBOTTOM_GV(dstr) = gv_dup_inc(IoBOTTOM_GV(dstr), param);
14353 IoTOP_GV(dstr) = gv_dup(IoTOP_GV(dstr), param);
14354 IoFMT_GV(dstr) = gv_dup(IoFMT_GV(dstr), param);
14355 IoBOTTOM_GV(dstr) = gv_dup(IoBOTTOM_GV(dstr), param);
14356 if (IoDIRP(dstr)) {
14357 IoDIRP(dstr) = dirp_dup(IoDIRP(dstr), param);
14360 /* IoDIRP(dstr) is already a copy of IoDIRP(sstr) */
14362 IoIFP(dstr) = fp_dup(IoIFP(sstr), IoTYPE(dstr), param);
14364 if (IoOFP(dstr) == IoIFP(sstr))
14365 IoOFP(dstr) = IoIFP(dstr);
14367 IoOFP(dstr) = fp_dup(IoOFP(dstr), IoTYPE(dstr), param);
14368 IoTOP_NAME(dstr) = SAVEPV(IoTOP_NAME(dstr));
14369 IoFMT_NAME(dstr) = SAVEPV(IoFMT_NAME(dstr));
14370 IoBOTTOM_NAME(dstr) = SAVEPV(IoBOTTOM_NAME(dstr));
14373 /* avoid cloning an empty array */
14374 if (AvARRAY((const AV *)sstr) && AvFILLp((const AV *)sstr) >= 0) {
14375 SV **dst_ary, **src_ary;
14376 SSize_t items = AvFILLp((const AV *)sstr) + 1;
14378 src_ary = AvARRAY((const AV *)sstr);
14379 Newx(dst_ary, AvMAX((const AV *)sstr)+1, SV*);
14380 ptr_table_store(PL_ptr_table, src_ary, dst_ary);
14381 AvARRAY(MUTABLE_AV(dstr)) = dst_ary;
14382 AvALLOC((const AV *)dstr) = dst_ary;
14383 if (AvREAL((const AV *)sstr)) {
14384 dst_ary = sv_dup_inc_multiple(src_ary, dst_ary, items,
14388 while (items-- > 0)
14389 *dst_ary++ = sv_dup(*src_ary++, param);
14391 items = AvMAX((const AV *)sstr) - AvFILLp((const AV *)sstr);
14392 while (items-- > 0) {
14397 AvARRAY(MUTABLE_AV(dstr)) = NULL;
14398 AvALLOC((const AV *)dstr) = (SV**)NULL;
14399 AvMAX( (const AV *)dstr) = -1;
14400 AvFILLp((const AV *)dstr) = -1;
14404 if (HvARRAY((const HV *)sstr)) {
14406 const bool sharekeys = !!HvSHAREKEYS(sstr);
14407 XPVHV * const dxhv = (XPVHV*)SvANY(dstr);
14408 XPVHV * const sxhv = (XPVHV*)SvANY(sstr);
14410 Newx(darray, PERL_HV_ARRAY_ALLOC_BYTES(dxhv->xhv_max+1)
14411 + (SvOOK(sstr) ? sizeof(struct xpvhv_aux) : 0),
14413 HvARRAY(dstr) = (HE**)darray;
14414 while (i <= sxhv->xhv_max) {
14415 const HE * const source = HvARRAY(sstr)[i];
14416 HvARRAY(dstr)[i] = source
14417 ? he_dup(source, sharekeys, param) : 0;
14421 const struct xpvhv_aux * const saux = HvAUX(sstr);
14422 struct xpvhv_aux * const daux = HvAUX(dstr);
14423 /* This flag isn't copied. */
14426 if (saux->xhv_name_count) {
14427 HEK ** const sname = saux->xhv_name_u.xhvnameu_names;
14429 = saux->xhv_name_count < 0
14430 ? -saux->xhv_name_count
14431 : saux->xhv_name_count;
14432 HEK **shekp = sname + count;
14434 Newx(daux->xhv_name_u.xhvnameu_names, count, HEK *);
14435 dhekp = daux->xhv_name_u.xhvnameu_names + count;
14436 while (shekp-- > sname) {
14438 *dhekp = hek_dup(*shekp, param);
14442 daux->xhv_name_u.xhvnameu_name
14443 = hek_dup(saux->xhv_name_u.xhvnameu_name,
14446 daux->xhv_name_count = saux->xhv_name_count;
14448 daux->xhv_aux_flags = saux->xhv_aux_flags;
14449 #ifdef PERL_HASH_RANDOMIZE_KEYS
14450 daux->xhv_rand = saux->xhv_rand;
14451 daux->xhv_last_rand = saux->xhv_last_rand;
14453 daux->xhv_riter = saux->xhv_riter;
14454 daux->xhv_eiter = saux->xhv_eiter
14455 ? he_dup(saux->xhv_eiter,
14456 cBOOL(HvSHAREKEYS(sstr)), param) : 0;
14457 /* backref array needs refcnt=2; see sv_add_backref */
14458 daux->xhv_backreferences =
14459 (param->flags & CLONEf_JOIN_IN)
14460 /* when joining, we let the individual GVs and
14461 * CVs add themselves to backref as
14462 * needed. This avoids pulling in stuff
14463 * that isn't required, and simplifies the
14464 * case where stashes aren't cloned back
14465 * if they already exist in the parent
14468 : saux->xhv_backreferences
14469 ? (SvTYPE(saux->xhv_backreferences) == SVt_PVAV)
14470 ? MUTABLE_AV(SvREFCNT_inc(
14471 sv_dup_inc((const SV *)
14472 saux->xhv_backreferences, param)))
14473 : MUTABLE_AV(sv_dup((const SV *)
14474 saux->xhv_backreferences, param))
14477 daux->xhv_mro_meta = saux->xhv_mro_meta
14478 ? mro_meta_dup(saux->xhv_mro_meta, param)
14481 /* Record stashes for possible cloning in Perl_clone(). */
14483 av_push(param->stashes, dstr);
14487 HvARRAY(MUTABLE_HV(dstr)) = NULL;
14490 if (!(param->flags & CLONEf_COPY_STACKS)) {
14495 /* NOTE: not refcounted */
14496 SvANY(MUTABLE_CV(dstr))->xcv_stash =
14497 hv_dup(CvSTASH(dstr), param);
14498 if ((param->flags & CLONEf_JOIN_IN) && CvSTASH(dstr))
14499 Perl_sv_add_backref(aTHX_ MUTABLE_SV(CvSTASH(dstr)), dstr);
14500 if (!CvISXSUB(dstr)) {
14502 CvROOT(dstr) = OpREFCNT_inc(CvROOT(dstr));
14504 CvSLABBED_off(dstr);
14505 } else if (CvCONST(dstr)) {
14506 CvXSUBANY(dstr).any_ptr =
14507 sv_dup_inc((const SV *)CvXSUBANY(dstr).any_ptr, param);
14509 assert(!CvSLABBED(dstr));
14510 if (CvDYNFILE(dstr)) CvFILE(dstr) = SAVEPV(CvFILE(dstr));
14512 SvANY((CV *)dstr)->xcv_gv_u.xcv_hek =
14513 hek_dup(CvNAME_HEK((CV *)sstr), param);
14514 /* don't dup if copying back - CvGV isn't refcounted, so the
14515 * duped GV may never be freed. A bit of a hack! DAPM */
14517 SvANY(MUTABLE_CV(dstr))->xcv_gv_u.xcv_gv =
14519 ? gv_dup_inc(CvGV(sstr), param)
14520 : (param->flags & CLONEf_JOIN_IN)
14522 : gv_dup(CvGV(sstr), param);
14524 if (!CvISXSUB(sstr)) {
14525 PADLIST * padlist = CvPADLIST(sstr);
14527 padlist = padlist_dup(padlist, param);
14528 CvPADLIST_set(dstr, padlist);
14530 /* unthreaded perl can't sv_dup so we dont support unthreaded's CvHSCXT */
14531 PoisonPADLIST(dstr);
14534 CvWEAKOUTSIDE(sstr)
14535 ? cv_dup( CvOUTSIDE(dstr), param)
14536 : cv_dup_inc(CvOUTSIDE(dstr), param);
14546 Perl_sv_dup_inc(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14548 PERL_ARGS_ASSERT_SV_DUP_INC;
14549 return sstr ? SvREFCNT_inc(sv_dup_common(sstr, param)) : NULL;
14553 Perl_sv_dup(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14555 SV *dstr = sstr ? sv_dup_common(sstr, param) : NULL;
14556 PERL_ARGS_ASSERT_SV_DUP;
14558 /* Track every SV that (at least initially) had a reference count of 0.
14559 We need to do this by holding an actual reference to it in this array.
14560 If we attempt to cheat, turn AvREAL_off(), and store only pointers
14561 (akin to the stashes hash, and the perl stack), we come unstuck if
14562 a weak reference (or other SV legitimately SvREFCNT() == 0 for this
14563 thread) is manipulated in a CLONE method, because CLONE runs before the
14564 unreferenced array is walked to find SVs still with SvREFCNT() == 0
14565 (and fix things up by giving each a reference via the temps stack).
14566 Instead, during CLONE, if the 0-referenced SV has SvREFCNT_inc() and
14567 then SvREFCNT_dec(), it will be cleaned up (and added to the free list)
14568 before the walk of unreferenced happens and a reference to that is SV
14569 added to the temps stack. At which point we have the same SV considered
14570 to be in use, and free to be re-used. Not good.
14572 if (dstr && !(param->flags & CLONEf_COPY_STACKS) && !SvREFCNT(dstr)) {
14573 assert(param->unreferenced);
14574 av_push(param->unreferenced, SvREFCNT_inc(dstr));
14580 /* duplicate a context */
14583 Perl_cx_dup(pTHX_ PERL_CONTEXT *cxs, I32 ix, I32 max, CLONE_PARAMS* param)
14585 PERL_CONTEXT *ncxs;
14587 PERL_ARGS_ASSERT_CX_DUP;
14590 return (PERL_CONTEXT*)NULL;
14592 /* look for it in the table first */
14593 ncxs = (PERL_CONTEXT*)ptr_table_fetch(PL_ptr_table, cxs);
14597 /* create anew and remember what it is */
14598 Newx(ncxs, max + 1, PERL_CONTEXT);
14599 ptr_table_store(PL_ptr_table, cxs, ncxs);
14600 Copy(cxs, ncxs, max + 1, PERL_CONTEXT);
14603 PERL_CONTEXT * const ncx = &ncxs[ix];
14604 if (CxTYPE(ncx) == CXt_SUBST) {
14605 Perl_croak(aTHX_ "Cloning substitution context is unimplemented");
14608 ncx->blk_oldcop = (COP*)any_dup(ncx->blk_oldcop, param->proto_perl);
14609 switch (CxTYPE(ncx)) {
14611 ncx->blk_sub.cv = cv_dup_inc(ncx->blk_sub.cv, param);
14612 if(CxHASARGS(ncx)){
14613 ncx->blk_sub.savearray = av_dup_inc(ncx->blk_sub.savearray,param);
14615 ncx->blk_sub.savearray = NULL;
14617 ncx->blk_sub.prevcomppad = (PAD*)ptr_table_fetch(PL_ptr_table,
14618 ncx->blk_sub.prevcomppad);
14621 ncx->blk_eval.old_namesv = sv_dup_inc(ncx->blk_eval.old_namesv,
14623 /* XXX should this sv_dup_inc? Or only if CxEVAL_TXT_REFCNTED ???? */
14624 ncx->blk_eval.cur_text = sv_dup(ncx->blk_eval.cur_text, param);
14625 ncx->blk_eval.cv = cv_dup(ncx->blk_eval.cv, param);
14626 /* XXX what to do with cur_top_env ???? */
14628 case CXt_LOOP_LAZYSV:
14629 ncx->blk_loop.state_u.lazysv.end
14630 = sv_dup_inc(ncx->blk_loop.state_u.lazysv.end, param);
14631 /* Fallthrough: duplicate lazysv.cur by using the ary.ary
14632 duplication code instead.
14633 We are taking advantage of (1) av_dup_inc and sv_dup_inc
14634 actually being the same function, and (2) order
14635 equivalence of the two unions.
14636 We can assert the later [but only at run time :-(] */
14637 assert ((void *) &ncx->blk_loop.state_u.ary.ary ==
14638 (void *) &ncx->blk_loop.state_u.lazysv.cur);
14641 ncx->blk_loop.state_u.ary.ary
14642 = av_dup_inc(ncx->blk_loop.state_u.ary.ary, param);
14644 case CXt_LOOP_LIST:
14645 case CXt_LOOP_LAZYIV:
14646 /* code common to all 'for' CXt_LOOP_* types */
14647 ncx->blk_loop.itersave =
14648 sv_dup_inc(ncx->blk_loop.itersave, param);
14649 if (CxPADLOOP(ncx)) {
14650 PADOFFSET off = ncx->blk_loop.itervar_u.svp
14651 - &CX_CURPAD_SV(ncx->blk_loop, 0);
14652 ncx->blk_loop.oldcomppad =
14653 (PAD*)ptr_table_fetch(PL_ptr_table,
14654 ncx->blk_loop.oldcomppad);
14655 ncx->blk_loop.itervar_u.svp =
14656 &CX_CURPAD_SV(ncx->blk_loop, off);
14659 /* this copies the GV if CXp_FOR_GV, or the SV for an
14660 * alias (for \$x (...)) - relies on gv_dup being the
14661 * same as sv_dup */
14662 ncx->blk_loop.itervar_u.gv
14663 = gv_dup((const GV *)ncx->blk_loop.itervar_u.gv,
14667 case CXt_LOOP_PLAIN:
14670 ncx->blk_format.prevcomppad =
14671 (PAD*)ptr_table_fetch(PL_ptr_table,
14672 ncx->blk_format.prevcomppad);
14673 ncx->blk_format.cv = cv_dup_inc(ncx->blk_format.cv, param);
14674 ncx->blk_format.gv = gv_dup(ncx->blk_format.gv, param);
14675 ncx->blk_format.dfoutgv = gv_dup_inc(ncx->blk_format.dfoutgv,
14679 ncx->blk_givwhen.defsv_save =
14680 sv_dup_inc(ncx->blk_givwhen.defsv_save, param);
14693 /* duplicate a stack info structure */
14696 Perl_si_dup(pTHX_ PERL_SI *si, CLONE_PARAMS* param)
14700 PERL_ARGS_ASSERT_SI_DUP;
14703 return (PERL_SI*)NULL;
14705 /* look for it in the table first */
14706 nsi = (PERL_SI*)ptr_table_fetch(PL_ptr_table, si);
14710 /* create anew and remember what it is */
14711 Newx(nsi, 1, PERL_SI);
14712 ptr_table_store(PL_ptr_table, si, nsi);
14714 nsi->si_stack = av_dup_inc(si->si_stack, param);
14715 nsi->si_cxix = si->si_cxix;
14716 nsi->si_cxsubix = si->si_cxsubix;
14717 nsi->si_cxmax = si->si_cxmax;
14718 nsi->si_cxstack = cx_dup(si->si_cxstack, si->si_cxix, si->si_cxmax, param);
14719 nsi->si_type = si->si_type;
14720 nsi->si_prev = si_dup(si->si_prev, param);
14721 nsi->si_next = si_dup(si->si_next, param);
14722 nsi->si_markoff = si->si_markoff;
14723 #if defined DEBUGGING && !defined DEBUGGING_RE_ONLY
14724 nsi->si_stack_hwm = 0;
14730 #define POPINT(ss,ix) ((ss)[--(ix)].any_i32)
14731 #define TOPINT(ss,ix) ((ss)[ix].any_i32)
14732 #define POPLONG(ss,ix) ((ss)[--(ix)].any_long)
14733 #define TOPLONG(ss,ix) ((ss)[ix].any_long)
14734 #define POPIV(ss,ix) ((ss)[--(ix)].any_iv)
14735 #define TOPIV(ss,ix) ((ss)[ix].any_iv)
14736 #define POPUV(ss,ix) ((ss)[--(ix)].any_uv)
14737 #define TOPUV(ss,ix) ((ss)[ix].any_uv)
14738 #define POPBOOL(ss,ix) ((ss)[--(ix)].any_bool)
14739 #define TOPBOOL(ss,ix) ((ss)[ix].any_bool)
14740 #define POPPTR(ss,ix) ((ss)[--(ix)].any_ptr)
14741 #define TOPPTR(ss,ix) ((ss)[ix].any_ptr)
14742 #define POPDPTR(ss,ix) ((ss)[--(ix)].any_dptr)
14743 #define TOPDPTR(ss,ix) ((ss)[ix].any_dptr)
14744 #define POPDXPTR(ss,ix) ((ss)[--(ix)].any_dxptr)
14745 #define TOPDXPTR(ss,ix) ((ss)[ix].any_dxptr)
14748 #define pv_dup_inc(p) SAVEPV(p)
14749 #define pv_dup(p) SAVEPV(p)
14750 #define svp_dup_inc(p,pp) any_dup(p,pp)
14752 /* map any object to the new equivent - either something in the
14753 * ptr table, or something in the interpreter structure
14757 Perl_any_dup(pTHX_ void *v, const PerlInterpreter *proto_perl)
14761 PERL_ARGS_ASSERT_ANY_DUP;
14764 return (void*)NULL;
14766 /* look for it in the table first */
14767 ret = ptr_table_fetch(PL_ptr_table, v);
14771 /* see if it is part of the interpreter structure */
14772 if (v >= (void*)proto_perl && v < (void*)(proto_perl+1))
14773 ret = (void*)(((char*)aTHX) + (((char*)v) - (char*)proto_perl));
14781 /* duplicate the save stack */
14784 Perl_ss_dup(pTHX_ PerlInterpreter *proto_perl, CLONE_PARAMS* param)
14787 ANY * const ss = proto_perl->Isavestack;
14788 const I32 max = proto_perl->Isavestack_max + SS_MAXPUSH;
14789 I32 ix = proto_perl->Isavestack_ix;
14802 void (*dptr) (void*);
14803 void (*dxptr) (pTHX_ void*);
14805 PERL_ARGS_ASSERT_SS_DUP;
14807 Newx(nss, max, ANY);
14810 const UV uv = POPUV(ss,ix);
14811 const U8 type = (U8)uv & SAVE_MASK;
14813 TOPUV(nss,ix) = uv;
14815 case SAVEt_CLEARSV:
14816 case SAVEt_CLEARPADRANGE:
14818 case SAVEt_HELEM: /* hash element */
14819 case SAVEt_SV: /* scalar reference */
14820 sv = (const SV *)POPPTR(ss,ix);
14821 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14823 case SAVEt_ITEM: /* normal string */
14824 case SAVEt_GVSV: /* scalar slot in GV */
14825 sv = (const SV *)POPPTR(ss,ix);
14826 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14827 if (type == SAVEt_SV)
14831 case SAVEt_MORTALIZESV:
14832 case SAVEt_READONLY_OFF:
14833 sv = (const SV *)POPPTR(ss,ix);
14834 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14836 case SAVEt_FREEPADNAME:
14837 ptr = POPPTR(ss,ix);
14838 TOPPTR(nss,ix) = padname_dup((PADNAME *)ptr, param);
14839 PadnameREFCNT((PADNAME *)TOPPTR(nss,ix))++;
14841 case SAVEt_SHARED_PVREF: /* char* in shared space */
14842 c = (char*)POPPTR(ss,ix);
14843 TOPPTR(nss,ix) = savesharedpv(c);
14844 ptr = POPPTR(ss,ix);
14845 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14847 case SAVEt_GENERIC_SVREF: /* generic sv */
14848 case SAVEt_SVREF: /* scalar reference */
14849 sv = (const SV *)POPPTR(ss,ix);
14850 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14851 if (type == SAVEt_SVREF)
14852 SvREFCNT_inc_simple_void((SV *)TOPPTR(nss,ix));
14853 ptr = POPPTR(ss,ix);
14854 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14856 case SAVEt_GVSLOT: /* any slot in GV */
14857 sv = (const SV *)POPPTR(ss,ix);
14858 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14859 ptr = POPPTR(ss,ix);
14860 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14861 sv = (const SV *)POPPTR(ss,ix);
14862 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14864 case SAVEt_HV: /* hash reference */
14865 case SAVEt_AV: /* array reference */
14866 sv = (const SV *) POPPTR(ss,ix);
14867 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14869 case SAVEt_COMPPAD:
14871 sv = (const SV *) POPPTR(ss,ix);
14872 TOPPTR(nss,ix) = sv_dup(sv, param);
14874 case SAVEt_INT: /* int reference */
14875 ptr = POPPTR(ss,ix);
14876 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14877 intval = (int)POPINT(ss,ix);
14878 TOPINT(nss,ix) = intval;
14880 case SAVEt_LONG: /* long reference */
14881 ptr = POPPTR(ss,ix);
14882 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14883 longval = (long)POPLONG(ss,ix);
14884 TOPLONG(nss,ix) = longval;
14886 case SAVEt_I32: /* I32 reference */
14887 ptr = POPPTR(ss,ix);
14888 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14890 TOPINT(nss,ix) = i;
14892 case SAVEt_IV: /* IV reference */
14893 case SAVEt_STRLEN: /* STRLEN/size_t ref */
14894 ptr = POPPTR(ss,ix);
14895 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14897 TOPIV(nss,ix) = iv;
14899 case SAVEt_TMPSFLOOR:
14901 TOPIV(nss,ix) = iv;
14903 case SAVEt_HPTR: /* HV* reference */
14904 case SAVEt_APTR: /* AV* reference */
14905 case SAVEt_SPTR: /* SV* reference */
14906 ptr = POPPTR(ss,ix);
14907 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14908 sv = (const SV *)POPPTR(ss,ix);
14909 TOPPTR(nss,ix) = sv_dup(sv, param);
14911 case SAVEt_VPTR: /* random* reference */
14912 ptr = POPPTR(ss,ix);
14913 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14915 case SAVEt_INT_SMALL:
14916 case SAVEt_I32_SMALL:
14917 case SAVEt_I16: /* I16 reference */
14918 case SAVEt_I8: /* I8 reference */
14920 ptr = POPPTR(ss,ix);
14921 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14923 case SAVEt_GENERIC_PVREF: /* generic char* */
14924 case SAVEt_PPTR: /* char* reference */
14925 ptr = POPPTR(ss,ix);
14926 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14927 c = (char*)POPPTR(ss,ix);
14928 TOPPTR(nss,ix) = pv_dup(c);
14930 case SAVEt_GP: /* scalar reference */
14931 gp = (GP*)POPPTR(ss,ix);
14932 TOPPTR(nss,ix) = gp = gp_dup(gp, param);
14933 (void)GpREFCNT_inc(gp);
14934 gv = (const GV *)POPPTR(ss,ix);
14935 TOPPTR(nss,ix) = gv_dup_inc(gv, param);
14938 ptr = POPPTR(ss,ix);
14939 if (ptr && (((OP*)ptr)->op_private & OPpREFCOUNTED)) {
14940 /* these are assumed to be refcounted properly */
14942 switch (((OP*)ptr)->op_type) {
14944 case OP_LEAVESUBLV:
14948 case OP_LEAVEWRITE:
14949 TOPPTR(nss,ix) = ptr;
14952 (void) OpREFCNT_inc(o);
14956 TOPPTR(nss,ix) = NULL;
14961 TOPPTR(nss,ix) = NULL;
14963 case SAVEt_FREECOPHH:
14964 ptr = POPPTR(ss,ix);
14965 TOPPTR(nss,ix) = cophh_copy((COPHH *)ptr);
14967 case SAVEt_ADELETE:
14968 av = (const AV *)POPPTR(ss,ix);
14969 TOPPTR(nss,ix) = av_dup_inc(av, param);
14971 TOPINT(nss,ix) = i;
14974 hv = (const HV *)POPPTR(ss,ix);
14975 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14977 TOPINT(nss,ix) = i;
14980 c = (char*)POPPTR(ss,ix);
14981 TOPPTR(nss,ix) = pv_dup_inc(c);
14983 case SAVEt_STACK_POS: /* Position on Perl stack */
14985 TOPINT(nss,ix) = i;
14987 case SAVEt_DESTRUCTOR:
14988 ptr = POPPTR(ss,ix);
14989 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14990 dptr = POPDPTR(ss,ix);
14991 TOPDPTR(nss,ix) = DPTR2FPTR(void (*)(void*),
14992 any_dup(FPTR2DPTR(void *, dptr),
14995 case SAVEt_DESTRUCTOR_X:
14996 ptr = POPPTR(ss,ix);
14997 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14998 dxptr = POPDXPTR(ss,ix);
14999 TOPDXPTR(nss,ix) = DPTR2FPTR(void (*)(pTHX_ void*),
15000 any_dup(FPTR2DPTR(void *, dxptr),
15003 case SAVEt_REGCONTEXT:
15005 ix -= uv >> SAVE_TIGHT_SHIFT;
15007 case SAVEt_AELEM: /* array element */
15008 sv = (const SV *)POPPTR(ss,ix);
15009 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
15011 TOPIV(nss,ix) = iv;
15012 av = (const AV *)POPPTR(ss,ix);
15013 TOPPTR(nss,ix) = av_dup_inc(av, param);
15016 ptr = POPPTR(ss,ix);
15017 TOPPTR(nss,ix) = ptr;
15020 ptr = POPPTR(ss,ix);
15021 ptr = cophh_copy((COPHH*)ptr);
15022 TOPPTR(nss,ix) = ptr;
15024 TOPINT(nss,ix) = i;
15025 if (i & HINT_LOCALIZE_HH) {
15026 hv = (const HV *)POPPTR(ss,ix);
15027 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
15030 case SAVEt_PADSV_AND_MORTALIZE:
15031 longval = (long)POPLONG(ss,ix);
15032 TOPLONG(nss,ix) = longval;
15033 ptr = POPPTR(ss,ix);
15034 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
15035 sv = (const SV *)POPPTR(ss,ix);
15036 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
15038 case SAVEt_SET_SVFLAGS:
15040 TOPINT(nss,ix) = i;
15042 TOPINT(nss,ix) = i;
15043 sv = (const SV *)POPPTR(ss,ix);
15044 TOPPTR(nss,ix) = sv_dup(sv, param);
15046 case SAVEt_COMPILE_WARNINGS:
15047 ptr = POPPTR(ss,ix);
15048 TOPPTR(nss,ix) = DUP_WARNINGS((STRLEN*)ptr);
15051 ptr = POPPTR(ss,ix);
15052 TOPPTR(nss,ix) = parser_dup((const yy_parser*)ptr, param);
15056 "panic: ss_dup inconsistency (%" IVdf ")", (IV) type);
15064 /* if sv is a stash, call $class->CLONE_SKIP(), and set the SVphv_CLONEABLE
15065 * flag to the result. This is done for each stash before cloning starts,
15066 * so we know which stashes want their objects cloned */
15069 do_mark_cloneable_stash(pTHX_ SV *const sv)
15071 const HEK * const hvname = HvNAME_HEK((const HV *)sv);
15073 GV* const cloner = gv_fetchmethod_autoload(MUTABLE_HV(sv), "CLONE_SKIP", 0);
15074 SvFLAGS(sv) |= SVphv_CLONEABLE; /* clone objects by default */
15075 if (cloner && GvCV(cloner)) {
15082 mXPUSHs(newSVhek(hvname));
15084 call_sv(MUTABLE_SV(GvCV(cloner)), G_SCALAR);
15091 SvFLAGS(sv) &= ~SVphv_CLONEABLE;
15099 =for apidoc perl_clone
15101 Create and return a new interpreter by cloning the current one.
15103 C<perl_clone> takes these flags as parameters:
15105 C<CLONEf_COPY_STACKS> - is used to, well, copy the stacks also,
15106 without it we only clone the data and zero the stacks,
15107 with it we copy the stacks and the new perl interpreter is
15108 ready to run at the exact same point as the previous one.
15109 The pseudo-fork code uses C<COPY_STACKS> while the
15110 threads->create doesn't.
15112 C<CLONEf_KEEP_PTR_TABLE> -
15113 C<perl_clone> keeps a ptr_table with the pointer of the old
15114 variable as a key and the new variable as a value,
15115 this allows it to check if something has been cloned and not
15116 clone it again, but rather just use the value and increase the
15118 If C<KEEP_PTR_TABLE> is not set then C<perl_clone> will kill the ptr_table
15119 using the function S<C<ptr_table_free(PL_ptr_table); PL_ptr_table = NULL;>>.
15120 A reason to keep it around is if you want to dup some of your own
15121 variables which are outside the graph that perl scans.
15123 C<CLONEf_CLONE_HOST> -
15124 This is a win32 thing, it is ignored on unix, it tells perl's
15125 win32host code (which is c++) to clone itself, this is needed on
15126 win32 if you want to run two threads at the same time,
15127 if you just want to do some stuff in a separate perl interpreter
15128 and then throw it away and return to the original one,
15129 you don't need to do anything.
15134 /* XXX the above needs expanding by someone who actually understands it ! */
15135 EXTERN_C PerlInterpreter *
15136 perl_clone_host(PerlInterpreter* proto_perl, UV flags);
15139 perl_clone(PerlInterpreter *proto_perl, UV flags)
15142 #ifdef PERL_IMPLICIT_SYS
15144 PERL_ARGS_ASSERT_PERL_CLONE;
15146 /* perlhost.h so we need to call into it
15147 to clone the host, CPerlHost should have a c interface, sky */
15149 #ifndef __amigaos4__
15150 if (flags & CLONEf_CLONE_HOST) {
15151 return perl_clone_host(proto_perl,flags);
15154 return perl_clone_using(proto_perl, flags,
15156 proto_perl->IMemShared,
15157 proto_perl->IMemParse,
15159 proto_perl->IStdIO,
15163 proto_perl->IProc);
15167 perl_clone_using(PerlInterpreter *proto_perl, UV flags,
15168 struct IPerlMem* ipM, struct IPerlMem* ipMS,
15169 struct IPerlMem* ipMP, struct IPerlEnv* ipE,
15170 struct IPerlStdIO* ipStd, struct IPerlLIO* ipLIO,
15171 struct IPerlDir* ipD, struct IPerlSock* ipS,
15172 struct IPerlProc* ipP)
15174 /* XXX many of the string copies here can be optimized if they're
15175 * constants; they need to be allocated as common memory and just
15176 * their pointers copied. */
15179 CLONE_PARAMS clone_params;
15180 CLONE_PARAMS* const param = &clone_params;
15182 PerlInterpreter * const my_perl = (PerlInterpreter*)(*ipM->pMalloc)(ipM, sizeof(PerlInterpreter));
15184 PERL_ARGS_ASSERT_PERL_CLONE_USING;
15185 #else /* !PERL_IMPLICIT_SYS */
15187 CLONE_PARAMS clone_params;
15188 CLONE_PARAMS* param = &clone_params;
15189 PerlInterpreter * const my_perl = (PerlInterpreter*)PerlMem_malloc(sizeof(PerlInterpreter));
15191 PERL_ARGS_ASSERT_PERL_CLONE;
15192 #endif /* PERL_IMPLICIT_SYS */
15194 /* for each stash, determine whether its objects should be cloned */
15195 S_visit(proto_perl, do_mark_cloneable_stash, SVt_PVHV, SVTYPEMASK);
15196 PERL_SET_THX(my_perl);
15199 PoisonNew(my_perl, 1, PerlInterpreter);
15202 PL_defstash = NULL; /* may be used by perl malloc() */
15205 PL_scopestack_name = 0;
15207 PL_savestack_ix = 0;
15208 PL_savestack_max = -1;
15209 PL_sig_pending = 0;
15211 Zero(&PL_debug_pad, 1, struct perl_debug_pad);
15212 Zero(&PL_padname_undef, 1, PADNAME);
15213 Zero(&PL_padname_const, 1, PADNAME);
15214 # ifdef DEBUG_LEAKING_SCALARS
15215 PL_sv_serial = (((UV)my_perl >> 2) & 0xfff) * 1000000;
15217 # ifdef PERL_TRACE_OPS
15218 Zero(PL_op_exec_cnt, OP_max+2, UV);
15220 #else /* !DEBUGGING */
15221 Zero(my_perl, 1, PerlInterpreter);
15222 #endif /* DEBUGGING */
15224 #ifdef PERL_IMPLICIT_SYS
15225 /* host pointers */
15227 PL_MemShared = ipMS;
15228 PL_MemParse = ipMP;
15235 #endif /* PERL_IMPLICIT_SYS */
15238 param->flags = flags;
15239 /* Nothing in the core code uses this, but we make it available to
15240 extensions (using mg_dup). */
15241 param->proto_perl = proto_perl;
15242 /* Likely nothing will use this, but it is initialised to be consistent
15243 with Perl_clone_params_new(). */
15244 param->new_perl = my_perl;
15245 param->unreferenced = NULL;
15248 INIT_TRACK_MEMPOOL(my_perl->Imemory_debug_header, my_perl);
15250 PL_body_arenas = NULL;
15251 Zero(&PL_body_roots, 1, PL_body_roots);
15255 PL_sv_arenaroot = NULL;
15257 PL_debug = proto_perl->Idebug;
15259 /* dbargs array probably holds garbage */
15262 PL_compiling = proto_perl->Icompiling;
15264 /* pseudo environmental stuff */
15265 PL_origargc = proto_perl->Iorigargc;
15266 PL_origargv = proto_perl->Iorigargv;
15268 #ifndef NO_TAINT_SUPPORT
15269 /* Set tainting stuff before PerlIO_debug can possibly get called */
15270 PL_tainting = proto_perl->Itainting;
15271 PL_taint_warn = proto_perl->Itaint_warn;
15273 PL_tainting = FALSE;
15274 PL_taint_warn = FALSE;
15277 PL_minus_c = proto_perl->Iminus_c;
15279 PL_localpatches = proto_perl->Ilocalpatches;
15280 PL_splitstr = proto_perl->Isplitstr;
15281 PL_minus_n = proto_perl->Iminus_n;
15282 PL_minus_p = proto_perl->Iminus_p;
15283 PL_minus_l = proto_perl->Iminus_l;
15284 PL_minus_a = proto_perl->Iminus_a;
15285 PL_minus_E = proto_perl->Iminus_E;
15286 PL_minus_F = proto_perl->Iminus_F;
15287 PL_doswitches = proto_perl->Idoswitches;
15288 PL_dowarn = proto_perl->Idowarn;
15289 #ifdef PERL_SAWAMPERSAND
15290 PL_sawampersand = proto_perl->Isawampersand;
15292 PL_unsafe = proto_perl->Iunsafe;
15293 PL_perldb = proto_perl->Iperldb;
15294 PL_perl_destruct_level = proto_perl->Iperl_destruct_level;
15295 PL_exit_flags = proto_perl->Iexit_flags;
15297 /* XXX time(&PL_basetime) when asked for? */
15298 PL_basetime = proto_perl->Ibasetime;
15300 PL_maxsysfd = proto_perl->Imaxsysfd;
15301 PL_statusvalue = proto_perl->Istatusvalue;
15303 PL_statusvalue_vms = proto_perl->Istatusvalue_vms;
15305 PL_statusvalue_posix = proto_perl->Istatusvalue_posix;
15308 /* RE engine related */
15309 PL_regmatch_slab = NULL;
15310 PL_reg_curpm = NULL;
15312 PL_sub_generation = proto_perl->Isub_generation;
15314 /* funky return mechanisms */
15315 PL_forkprocess = proto_perl->Iforkprocess;
15317 /* internal state */
15318 PL_main_start = proto_perl->Imain_start;
15319 PL_eval_root = proto_perl->Ieval_root;
15320 PL_eval_start = proto_perl->Ieval_start;
15322 PL_filemode = proto_perl->Ifilemode;
15323 PL_lastfd = proto_perl->Ilastfd;
15324 PL_oldname = proto_perl->Ioldname; /* XXX not quite right */
15325 PL_gensym = proto_perl->Igensym;
15327 PL_laststatval = proto_perl->Ilaststatval;
15328 PL_laststype = proto_perl->Ilaststype;
15331 PL_profiledata = NULL;
15333 PL_generation = proto_perl->Igeneration;
15335 PL_in_clean_objs = proto_perl->Iin_clean_objs;
15336 PL_in_clean_all = proto_perl->Iin_clean_all;
15338 PL_delaymagic_uid = proto_perl->Idelaymagic_uid;
15339 PL_delaymagic_euid = proto_perl->Idelaymagic_euid;
15340 PL_delaymagic_gid = proto_perl->Idelaymagic_gid;
15341 PL_delaymagic_egid = proto_perl->Idelaymagic_egid;
15342 PL_nomemok = proto_perl->Inomemok;
15343 PL_an = proto_perl->Ian;
15344 PL_evalseq = proto_perl->Ievalseq;
15345 PL_origenviron = proto_perl->Iorigenviron; /* XXX not quite right */
15346 PL_origalen = proto_perl->Iorigalen;
15348 PL_sighandlerp = proto_perl->Isighandlerp;
15349 PL_sighandler1p = proto_perl->Isighandler1p;
15350 PL_sighandler3p = proto_perl->Isighandler3p;
15352 PL_runops = proto_perl->Irunops;
15354 PL_subline = proto_perl->Isubline;
15356 PL_cv_has_eval = proto_perl->Icv_has_eval;
15359 PL_cryptseen = proto_perl->Icryptseen;
15362 #ifdef USE_LOCALE_COLLATE
15363 PL_collation_ix = proto_perl->Icollation_ix;
15364 PL_collation_standard = proto_perl->Icollation_standard;
15365 PL_collxfrm_base = proto_perl->Icollxfrm_base;
15366 PL_collxfrm_mult = proto_perl->Icollxfrm_mult;
15367 PL_strxfrm_max_cp = proto_perl->Istrxfrm_max_cp;
15368 #endif /* USE_LOCALE_COLLATE */
15370 #ifdef USE_LOCALE_NUMERIC
15371 PL_numeric_standard = proto_perl->Inumeric_standard;
15372 PL_numeric_underlying = proto_perl->Inumeric_underlying;
15373 PL_numeric_underlying_is_standard = proto_perl->Inumeric_underlying_is_standard;
15374 #endif /* !USE_LOCALE_NUMERIC */
15376 /* Did the locale setup indicate UTF-8? */
15377 PL_utf8locale = proto_perl->Iutf8locale;
15378 PL_in_utf8_CTYPE_locale = proto_perl->Iin_utf8_CTYPE_locale;
15379 PL_in_utf8_COLLATE_locale = proto_perl->Iin_utf8_COLLATE_locale;
15380 my_strlcpy(PL_locale_utf8ness, proto_perl->Ilocale_utf8ness, sizeof(PL_locale_utf8ness));
15381 #if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
15382 PL_lc_numeric_mutex_depth = 0;
15384 /* Unicode features (see perlrun/-C) */
15385 PL_unicode = proto_perl->Iunicode;
15387 /* Pre-5.8 signals control */
15388 PL_signals = proto_perl->Isignals;
15390 /* times() ticks per second */
15391 PL_clocktick = proto_perl->Iclocktick;
15393 /* Recursion stopper for PerlIO_find_layer */
15394 PL_in_load_module = proto_perl->Iin_load_module;
15396 /* Not really needed/useful since the reenrant_retint is "volatile",
15397 * but do it for consistency's sake. */
15398 PL_reentrant_retint = proto_perl->Ireentrant_retint;
15400 /* Hooks to shared SVs and locks. */
15401 PL_sharehook = proto_perl->Isharehook;
15402 PL_lockhook = proto_perl->Ilockhook;
15403 PL_unlockhook = proto_perl->Iunlockhook;
15404 PL_threadhook = proto_perl->Ithreadhook;
15405 PL_destroyhook = proto_perl->Idestroyhook;
15406 PL_signalhook = proto_perl->Isignalhook;
15408 PL_globhook = proto_perl->Iglobhook;
15410 PL_srand_called = proto_perl->Isrand_called;
15411 Copy(&(proto_perl->Irandom_state), &PL_random_state, 1, PL_RANDOM_STATE_TYPE);
15413 if (flags & CLONEf_COPY_STACKS) {
15414 /* next allocation will be PL_tmps_stack[PL_tmps_ix+1] */
15415 PL_tmps_ix = proto_perl->Itmps_ix;
15416 PL_tmps_max = proto_perl->Itmps_max;
15417 PL_tmps_floor = proto_perl->Itmps_floor;
15419 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
15420 * NOTE: unlike the others! */
15421 PL_scopestack_ix = proto_perl->Iscopestack_ix;
15422 PL_scopestack_max = proto_perl->Iscopestack_max;
15424 /* next SSPUSHFOO() sets PL_savestack[PL_savestack_ix]
15425 * NOTE: unlike the others! */
15426 PL_savestack_ix = proto_perl->Isavestack_ix;
15427 PL_savestack_max = proto_perl->Isavestack_max;
15430 PL_start_env = proto_perl->Istart_env; /* XXXXXX */
15431 PL_top_env = &PL_start_env;
15433 PL_op = proto_perl->Iop;
15436 PL_Xpv = (XPV*)NULL;
15437 my_perl->Ina = proto_perl->Ina;
15439 PL_statcache = proto_perl->Istatcache;
15441 #ifndef NO_TAINT_SUPPORT
15442 PL_tainted = proto_perl->Itainted;
15444 PL_tainted = FALSE;
15446 PL_curpm = proto_perl->Icurpm; /* XXX No PMOP ref count */
15448 PL_chopset = proto_perl->Ichopset; /* XXX never deallocated */
15450 PL_restartjmpenv = proto_perl->Irestartjmpenv;
15451 PL_restartop = proto_perl->Irestartop;
15452 PL_in_eval = proto_perl->Iin_eval;
15453 PL_delaymagic = proto_perl->Idelaymagic;
15454 PL_phase = proto_perl->Iphase;
15455 PL_localizing = proto_perl->Ilocalizing;
15457 PL_hv_fetch_ent_mh = NULL;
15458 PL_modcount = proto_perl->Imodcount;
15459 PL_lastgotoprobe = NULL;
15460 PL_dumpindent = proto_perl->Idumpindent;
15462 PL_efloatbuf = NULL; /* reinits on demand */
15463 PL_efloatsize = 0; /* reinits on demand */
15467 PL_colorset = 0; /* reinits PL_colors[] */
15468 /*PL_colors[6] = {0,0,0,0,0,0};*/
15470 /* Pluggable optimizer */
15471 PL_peepp = proto_perl->Ipeepp;
15472 PL_rpeepp = proto_perl->Irpeepp;
15473 /* op_free() hook */
15474 PL_opfreehook = proto_perl->Iopfreehook;
15476 #ifdef USE_REENTRANT_API
15477 /* XXX: things like -Dm will segfault here in perlio, but doing
15478 * PERL_SET_CONTEXT(proto_perl);
15479 * breaks too many other things
15481 Perl_reentrant_init(aTHX);
15484 /* create SV map for pointer relocation */
15485 PL_ptr_table = ptr_table_new();
15487 /* initialize these special pointers as early as possible */
15489 ptr_table_store(PL_ptr_table, &proto_perl->Isv_undef, &PL_sv_undef);
15490 ptr_table_store(PL_ptr_table, &proto_perl->Isv_no, &PL_sv_no);
15491 ptr_table_store(PL_ptr_table, &proto_perl->Isv_zero, &PL_sv_zero);
15492 ptr_table_store(PL_ptr_table, &proto_perl->Isv_yes, &PL_sv_yes);
15493 ptr_table_store(PL_ptr_table, &proto_perl->Ipadname_const,
15494 &PL_padname_const);
15496 /* create (a non-shared!) shared string table */
15497 PL_strtab = newHV();
15498 HvSHAREKEYS_off(PL_strtab);
15499 hv_ksplit(PL_strtab, HvTOTALKEYS(proto_perl->Istrtab));
15500 ptr_table_store(PL_ptr_table, proto_perl->Istrtab, PL_strtab);
15502 Zero(PL_sv_consts, SV_CONSTS_COUNT, SV*);
15504 /* This PV will be free'd special way so must set it same way op.c does */
15505 PL_compiling.cop_file = savesharedpv(PL_compiling.cop_file);
15506 ptr_table_store(PL_ptr_table, proto_perl->Icompiling.cop_file, PL_compiling.cop_file);
15508 ptr_table_store(PL_ptr_table, &proto_perl->Icompiling, &PL_compiling);
15509 PL_compiling.cop_warnings = DUP_WARNINGS(PL_compiling.cop_warnings);
15510 CopHINTHASH_set(&PL_compiling, cophh_copy(CopHINTHASH_get(&PL_compiling)));
15511 PL_curcop = (COP*)any_dup(proto_perl->Icurcop, proto_perl);
15513 param->stashes = newAV(); /* Setup array of objects to call clone on */
15514 /* This makes no difference to the implementation, as it always pushes
15515 and shifts pointers to other SVs without changing their reference
15516 count, with the array becoming empty before it is freed. However, it
15517 makes it conceptually clear what is going on, and will avoid some
15518 work inside av.c, filling slots between AvFILL() and AvMAX() with
15519 &PL_sv_undef, and SvREFCNT_dec()ing those. */
15520 AvREAL_off(param->stashes);
15522 if (!(flags & CLONEf_COPY_STACKS)) {
15523 param->unreferenced = newAV();
15526 #ifdef PERLIO_LAYERS
15527 /* Clone PerlIO tables as soon as we can handle general xx_dup() */
15528 PerlIO_clone(aTHX_ proto_perl, param);
15531 PL_envgv = gv_dup_inc(proto_perl->Ienvgv, param);
15532 PL_incgv = gv_dup_inc(proto_perl->Iincgv, param);
15533 PL_hintgv = gv_dup_inc(proto_perl->Ihintgv, param);
15534 PL_origfilename = SAVEPV(proto_perl->Iorigfilename);
15535 PL_xsubfilename = proto_perl->Ixsubfilename;
15536 PL_diehook = sv_dup_inc(proto_perl->Idiehook, param);
15537 PL_warnhook = sv_dup_inc(proto_perl->Iwarnhook, param);
15540 PL_patchlevel = sv_dup_inc(proto_perl->Ipatchlevel, param);
15541 PL_inplace = SAVEPV(proto_perl->Iinplace);
15542 PL_e_script = sv_dup_inc(proto_perl->Ie_script, param);
15544 /* magical thingies */
15546 SvPVCLEAR(PERL_DEBUG_PAD(0)); /* For regex debugging. */
15547 SvPVCLEAR(PERL_DEBUG_PAD(1)); /* ext/re needs these */
15548 SvPVCLEAR(PERL_DEBUG_PAD(2)); /* even without DEBUGGING. */
15551 /* Clone the regex array */
15552 /* ORANGE FIXME for plugins, probably in the SV dup code.
15553 newSViv(PTR2IV(CALLREGDUPE(
15554 INT2PTR(REGEXP *, SvIVX(regex)), param))))
15556 PL_regex_padav = av_dup_inc(proto_perl->Iregex_padav, param);
15557 PL_regex_pad = AvARRAY(PL_regex_padav);
15559 PL_stashpadmax = proto_perl->Istashpadmax;
15560 PL_stashpadix = proto_perl->Istashpadix ;
15561 Newx(PL_stashpad, PL_stashpadmax, HV *);
15564 for (; o < PL_stashpadmax; ++o)
15565 PL_stashpad[o] = hv_dup(proto_perl->Istashpad[o], param);
15568 /* shortcuts to various I/O objects */
15569 PL_ofsgv = gv_dup_inc(proto_perl->Iofsgv, param);
15570 PL_stdingv = gv_dup(proto_perl->Istdingv, param);
15571 PL_stderrgv = gv_dup(proto_perl->Istderrgv, param);
15572 PL_defgv = gv_dup(proto_perl->Idefgv, param);
15573 PL_argvgv = gv_dup_inc(proto_perl->Iargvgv, param);
15574 PL_argvoutgv = gv_dup(proto_perl->Iargvoutgv, param);
15575 PL_argvout_stack = av_dup_inc(proto_perl->Iargvout_stack, param);
15577 /* shortcuts to regexp stuff */
15578 PL_replgv = gv_dup_inc(proto_perl->Ireplgv, param);
15580 /* shortcuts to misc objects */
15581 PL_errgv = gv_dup(proto_perl->Ierrgv, param);
15583 /* shortcuts to debugging objects */
15584 PL_DBgv = gv_dup_inc(proto_perl->IDBgv, param);
15585 PL_DBline = gv_dup_inc(proto_perl->IDBline, param);
15586 PL_DBsub = gv_dup_inc(proto_perl->IDBsub, param);
15587 PL_DBsingle = sv_dup(proto_perl->IDBsingle, param);
15588 PL_DBtrace = sv_dup(proto_perl->IDBtrace, param);
15589 PL_DBsignal = sv_dup(proto_perl->IDBsignal, param);
15590 Copy(proto_perl->IDBcontrol, PL_DBcontrol, DBVARMG_COUNT, IV);
15592 /* symbol tables */
15593 PL_defstash = hv_dup_inc(proto_perl->Idefstash, param);
15594 PL_curstash = hv_dup_inc(proto_perl->Icurstash, param);
15595 PL_debstash = hv_dup(proto_perl->Idebstash, param);
15596 PL_globalstash = hv_dup(proto_perl->Iglobalstash, param);
15597 PL_curstname = sv_dup_inc(proto_perl->Icurstname, param);
15599 PL_beginav = av_dup_inc(proto_perl->Ibeginav, param);
15600 PL_beginav_save = av_dup_inc(proto_perl->Ibeginav_save, param);
15601 PL_checkav_save = av_dup_inc(proto_perl->Icheckav_save, param);
15602 PL_unitcheckav = av_dup_inc(proto_perl->Iunitcheckav, param);
15603 PL_unitcheckav_save = av_dup_inc(proto_perl->Iunitcheckav_save, param);
15604 PL_endav = av_dup_inc(proto_perl->Iendav, param);
15605 PL_checkav = av_dup_inc(proto_perl->Icheckav, param);
15606 PL_initav = av_dup_inc(proto_perl->Iinitav, param);
15607 PL_savebegin = proto_perl->Isavebegin;
15609 PL_isarev = hv_dup_inc(proto_perl->Iisarev, param);
15611 /* subprocess state */
15612 PL_fdpid = av_dup_inc(proto_perl->Ifdpid, param);
15614 if (proto_perl->Iop_mask)
15615 PL_op_mask = SAVEPVN(proto_perl->Iop_mask, PL_maxo);
15618 /* PL_asserting = proto_perl->Iasserting; */
15620 /* current interpreter roots */
15621 PL_main_cv = cv_dup_inc(proto_perl->Imain_cv, param);
15623 PL_main_root = OpREFCNT_inc(proto_perl->Imain_root);
15626 /* runtime control stuff */
15627 PL_curcopdb = (COP*)any_dup(proto_perl->Icurcopdb, proto_perl);
15629 PL_preambleav = av_dup_inc(proto_perl->Ipreambleav, param);
15631 PL_ors_sv = sv_dup_inc(proto_perl->Iors_sv, param);
15633 /* interpreter atexit processing */
15634 PL_exitlistlen = proto_perl->Iexitlistlen;
15635 if (PL_exitlistlen) {
15636 Newx(PL_exitlist, PL_exitlistlen, PerlExitListEntry);
15637 Copy(proto_perl->Iexitlist, PL_exitlist, PL_exitlistlen, PerlExitListEntry);
15640 PL_exitlist = (PerlExitListEntry*)NULL;
15642 PL_my_cxt_size = proto_perl->Imy_cxt_size;
15643 if (PL_my_cxt_size) {
15644 Newx(PL_my_cxt_list, PL_my_cxt_size, void *);
15645 Copy(proto_perl->Imy_cxt_list, PL_my_cxt_list, PL_my_cxt_size, void *);
15648 PL_my_cxt_list = (void**)NULL;
15650 PL_modglobal = hv_dup_inc(proto_perl->Imodglobal, param);
15651 PL_custom_op_names = hv_dup_inc(proto_perl->Icustom_op_names,param);
15652 PL_custom_op_descs = hv_dup_inc(proto_perl->Icustom_op_descs,param);
15653 PL_custom_ops = hv_dup_inc(proto_perl->Icustom_ops, param);
15655 PL_compcv = cv_dup(proto_perl->Icompcv, param);
15657 PAD_CLONE_VARS(proto_perl, param);
15659 #ifdef HAVE_INTERP_INTERN
15660 sys_intern_dup(&proto_perl->Isys_intern, &PL_sys_intern);
15663 PL_DBcv = cv_dup(proto_perl->IDBcv, param);
15665 #ifdef PERL_USES_PL_PIDSTATUS
15666 PL_pidstatus = newHV(); /* XXX flag for cloning? */
15668 PL_osname = SAVEPV(proto_perl->Iosname);
15669 PL_parser = parser_dup(proto_perl->Iparser, param);
15671 /* XXX this only works if the saved cop has already been cloned */
15672 if (proto_perl->Iparser) {
15673 PL_parser->saved_curcop = (COP*)any_dup(
15674 proto_perl->Iparser->saved_curcop,
15678 PL_subname = sv_dup_inc(proto_perl->Isubname, param);
15680 #if defined(USE_POSIX_2008_LOCALE) \
15681 && defined(USE_THREAD_SAFE_LOCALE) \
15682 && ! defined(HAS_QUERYLOCALE)
15683 for (i = 0; i < (int) C_ARRAY_LENGTH(PL_curlocales); i++) {
15684 PL_curlocales[i] = savepv("."); /* An illegal value */
15687 #ifdef USE_LOCALE_CTYPE
15688 /* Should we warn if uses locale? */
15689 PL_warn_locale = sv_dup_inc(proto_perl->Iwarn_locale, param);
15692 #ifdef USE_LOCALE_COLLATE
15693 PL_collation_name = SAVEPV(proto_perl->Icollation_name);
15694 #endif /* USE_LOCALE_COLLATE */
15696 #ifdef USE_LOCALE_NUMERIC
15697 PL_numeric_name = SAVEPV(proto_perl->Inumeric_name);
15698 PL_numeric_radix_sv = sv_dup_inc(proto_perl->Inumeric_radix_sv, param);
15700 # if defined(HAS_POSIX_2008_LOCALE)
15701 PL_underlying_numeric_obj = NULL;
15703 #endif /* !USE_LOCALE_NUMERIC */
15706 PL_mbrlen_ps = proto_perl->Imbrlen_ps;
15709 PL_mbrtowc_ps = proto_perl->Imbrtowc_ps;
15712 PL_wcrtomb_ps = proto_perl->Iwcrtomb_ps;
15715 PL_langinfo_buf = NULL;
15716 PL_langinfo_bufsize = 0;
15718 PL_setlocale_buf = NULL;
15719 PL_setlocale_bufsize = 0;
15721 /* Unicode inversion lists */
15723 PL_AboveLatin1 = sv_dup_inc(proto_perl->IAboveLatin1, param);
15724 PL_Assigned_invlist = sv_dup_inc(proto_perl->IAssigned_invlist, param);
15725 PL_GCB_invlist = sv_dup_inc(proto_perl->IGCB_invlist, param);
15726 PL_HasMultiCharFold = sv_dup_inc(proto_perl->IHasMultiCharFold, param);
15727 PL_InMultiCharFold = sv_dup_inc(proto_perl->IInMultiCharFold, param);
15728 PL_Latin1 = sv_dup_inc(proto_perl->ILatin1, param);
15729 PL_LB_invlist = sv_dup_inc(proto_perl->ILB_invlist, param);
15730 PL_SB_invlist = sv_dup_inc(proto_perl->ISB_invlist, param);
15731 PL_SCX_invlist = sv_dup_inc(proto_perl->ISCX_invlist, param);
15732 PL_UpperLatin1 = sv_dup_inc(proto_perl->IUpperLatin1, param);
15733 PL_in_some_fold = sv_dup_inc(proto_perl->Iin_some_fold, param);
15734 PL_utf8_foldclosures = sv_dup_inc(proto_perl->Iutf8_foldclosures, param);
15735 PL_utf8_idcont = sv_dup_inc(proto_perl->Iutf8_idcont, param);
15736 PL_utf8_idstart = sv_dup_inc(proto_perl->Iutf8_idstart, param);
15737 PL_utf8_perl_idcont = sv_dup_inc(proto_perl->Iutf8_perl_idcont, param);
15738 PL_utf8_perl_idstart = sv_dup_inc(proto_perl->Iutf8_perl_idstart, param);
15739 PL_utf8_xidcont = sv_dup_inc(proto_perl->Iutf8_xidcont, param);
15740 PL_utf8_xidstart = sv_dup_inc(proto_perl->Iutf8_xidstart, param);
15741 PL_WB_invlist = sv_dup_inc(proto_perl->IWB_invlist, param);
15742 for (i = 0; i < POSIX_CC_COUNT; i++) {
15743 PL_XPosix_ptrs[i] = sv_dup_inc(proto_perl->IXPosix_ptrs[i], param);
15744 if (i != _CC_CASED && i != _CC_VERTSPACE) {
15745 PL_Posix_ptrs[i] = sv_dup_inc(proto_perl->IPosix_ptrs[i], param);
15748 PL_Posix_ptrs[_CC_CASED] = PL_Posix_ptrs[_CC_ALPHA];
15749 PL_Posix_ptrs[_CC_VERTSPACE] = NULL;
15751 PL_utf8_toupper = sv_dup_inc(proto_perl->Iutf8_toupper, param);
15752 PL_utf8_totitle = sv_dup_inc(proto_perl->Iutf8_totitle, param);
15753 PL_utf8_tolower = sv_dup_inc(proto_perl->Iutf8_tolower, param);
15754 PL_utf8_tofold = sv_dup_inc(proto_perl->Iutf8_tofold, param);
15755 PL_utf8_tosimplefold = sv_dup_inc(proto_perl->Iutf8_tosimplefold, param);
15756 PL_utf8_charname_begin = sv_dup_inc(proto_perl->Iutf8_charname_begin, param);
15757 PL_utf8_charname_continue = sv_dup_inc(proto_perl->Iutf8_charname_continue, param);
15758 PL_utf8_mark = sv_dup_inc(proto_perl->Iutf8_mark, param);
15759 PL_InBitmap = sv_dup_inc(proto_perl->IInBitmap, param);
15760 PL_CCC_non0_non230 = sv_dup_inc(proto_perl->ICCC_non0_non230, param);
15761 PL_Private_Use = sv_dup_inc(proto_perl->IPrivate_Use, param);
15764 PL_seen_deprecated_macro = hv_dup_inc(proto_perl->Iseen_deprecated_macro, param);
15767 if (proto_perl->Ipsig_pend) {
15768 Newxz(PL_psig_pend, SIG_SIZE, int);
15771 PL_psig_pend = (int*)NULL;
15774 if (proto_perl->Ipsig_name) {
15775 Newx(PL_psig_name, 2 * SIG_SIZE, SV*);
15776 sv_dup_inc_multiple(proto_perl->Ipsig_name, PL_psig_name, 2 * SIG_SIZE,
15778 PL_psig_ptr = PL_psig_name + SIG_SIZE;
15781 PL_psig_ptr = (SV**)NULL;
15782 PL_psig_name = (SV**)NULL;
15785 if (flags & CLONEf_COPY_STACKS) {
15786 Newx(PL_tmps_stack, PL_tmps_max, SV*);
15787 sv_dup_inc_multiple(proto_perl->Itmps_stack, PL_tmps_stack,
15788 PL_tmps_ix+1, param);
15790 /* next PUSHMARK() sets *(PL_markstack_ptr+1) */
15791 i = proto_perl->Imarkstack_max - proto_perl->Imarkstack;
15792 Newx(PL_markstack, i, I32);
15793 PL_markstack_max = PL_markstack + (proto_perl->Imarkstack_max
15794 - proto_perl->Imarkstack);
15795 PL_markstack_ptr = PL_markstack + (proto_perl->Imarkstack_ptr
15796 - proto_perl->Imarkstack);
15797 Copy(proto_perl->Imarkstack, PL_markstack,
15798 PL_markstack_ptr - PL_markstack + 1, I32);
15800 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
15801 * NOTE: unlike the others! */
15802 Newx(PL_scopestack, PL_scopestack_max, I32);
15803 Copy(proto_perl->Iscopestack, PL_scopestack, PL_scopestack_ix, I32);
15806 Newx(PL_scopestack_name, PL_scopestack_max, const char *);
15807 Copy(proto_perl->Iscopestack_name, PL_scopestack_name, PL_scopestack_ix, const char *);
15809 /* reset stack AV to correct length before its duped via
15810 * PL_curstackinfo */
15811 AvFILLp(proto_perl->Icurstack) =
15812 proto_perl->Istack_sp - proto_perl->Istack_base;
15814 /* NOTE: si_dup() looks at PL_markstack */
15815 PL_curstackinfo = si_dup(proto_perl->Icurstackinfo, param);
15817 /* PL_curstack = PL_curstackinfo->si_stack; */
15818 PL_curstack = av_dup(proto_perl->Icurstack, param);
15819 PL_mainstack = av_dup(proto_perl->Imainstack, param);
15821 /* next PUSHs() etc. set *(PL_stack_sp+1) */
15822 PL_stack_base = AvARRAY(PL_curstack);
15823 PL_stack_sp = PL_stack_base + (proto_perl->Istack_sp
15824 - proto_perl->Istack_base);
15825 PL_stack_max = PL_stack_base + AvMAX(PL_curstack);
15827 /*Newxz(PL_savestack, PL_savestack_max, ANY);*/
15828 PL_savestack = ss_dup(proto_perl, param);
15832 ENTER; /* perl_destruct() wants to LEAVE; */
15835 PL_statgv = gv_dup(proto_perl->Istatgv, param);
15836 PL_statname = sv_dup_inc(proto_perl->Istatname, param);
15838 PL_rs = sv_dup_inc(proto_perl->Irs, param);
15839 PL_last_in_gv = gv_dup(proto_perl->Ilast_in_gv, param);
15840 PL_defoutgv = gv_dup_inc(proto_perl->Idefoutgv, param);
15841 PL_toptarget = sv_dup_inc(proto_perl->Itoptarget, param);
15842 PL_bodytarget = sv_dup_inc(proto_perl->Ibodytarget, param);
15843 PL_formtarget = sv_dup(proto_perl->Iformtarget, param);
15845 PL_errors = sv_dup_inc(proto_perl->Ierrors, param);
15847 PL_sortcop = (OP*)any_dup(proto_perl->Isortcop, proto_perl);
15848 PL_firstgv = gv_dup_inc(proto_perl->Ifirstgv, param);
15849 PL_secondgv = gv_dup_inc(proto_perl->Isecondgv, param);
15851 PL_stashcache = newHV();
15853 PL_watchaddr = (char **) ptr_table_fetch(PL_ptr_table,
15854 proto_perl->Iwatchaddr);
15855 PL_watchok = PL_watchaddr ? * PL_watchaddr : NULL;
15856 if (PL_debug && PL_watchaddr) {
15857 PerlIO_printf(Perl_debug_log,
15858 "WATCHING: %" UVxf " cloned as %" UVxf " with value %" UVxf "\n",
15859 PTR2UV(proto_perl->Iwatchaddr), PTR2UV(PL_watchaddr),
15860 PTR2UV(PL_watchok));
15863 PL_registered_mros = hv_dup_inc(proto_perl->Iregistered_mros, param);
15864 PL_blockhooks = av_dup_inc(proto_perl->Iblockhooks, param);
15866 /* Call the ->CLONE method, if it exists, for each of the stashes
15867 identified by sv_dup() above.
15869 while(av_tindex(param->stashes) != -1) {
15870 HV* const stash = MUTABLE_HV(av_shift(param->stashes));
15871 GV* const cloner = gv_fetchmethod_autoload(stash, "CLONE", 0);
15872 if (cloner && GvCV(cloner)) {
15877 mXPUSHs(newSVhek(HvNAME_HEK(stash)));
15879 call_sv(MUTABLE_SV(GvCV(cloner)), G_DISCARD);
15885 if (!(flags & CLONEf_KEEP_PTR_TABLE)) {
15886 ptr_table_free(PL_ptr_table);
15887 PL_ptr_table = NULL;
15890 if (!(flags & CLONEf_COPY_STACKS)) {
15891 unreferenced_to_tmp_stack(param->unreferenced);
15894 SvREFCNT_dec(param->stashes);
15896 /* orphaned? eg threads->new inside BEGIN or use */
15897 if (PL_compcv && ! SvREFCNT(PL_compcv)) {
15898 SvREFCNT_inc_simple_void(PL_compcv);
15899 SAVEFREESV(PL_compcv);
15906 S_unreferenced_to_tmp_stack(pTHX_ AV *const unreferenced)
15908 PERL_ARGS_ASSERT_UNREFERENCED_TO_TMP_STACK;
15910 if (AvFILLp(unreferenced) > -1) {
15911 SV **svp = AvARRAY(unreferenced);
15912 SV **const last = svp + AvFILLp(unreferenced);
15916 if (SvREFCNT(*svp) == 1)
15918 } while (++svp <= last);
15920 EXTEND_MORTAL(count);
15921 svp = AvARRAY(unreferenced);
15924 if (SvREFCNT(*svp) == 1) {
15925 /* Our reference is the only one to this SV. This means that
15926 in this thread, the scalar effectively has a 0 reference.
15927 That doesn't work (cleanup never happens), so donate our
15928 reference to it onto the save stack. */
15929 PL_tmps_stack[++PL_tmps_ix] = *svp;
15931 /* As an optimisation, because we are already walking the
15932 entire array, instead of above doing either
15933 SvREFCNT_inc(*svp) or *svp = &PL_sv_undef, we can instead
15934 release our reference to the scalar, so that at the end of
15935 the array owns zero references to the scalars it happens to
15936 point to. We are effectively converting the array from
15937 AvREAL() on to AvREAL() off. This saves the av_clear()
15938 (triggered by the SvREFCNT_dec(unreferenced) below) from
15939 walking the array a second time. */
15940 SvREFCNT_dec(*svp);
15943 } while (++svp <= last);
15944 AvREAL_off(unreferenced);
15946 SvREFCNT_dec_NN(unreferenced);
15950 Perl_clone_params_del(CLONE_PARAMS *param)
15952 /* This seemingly funky ordering keeps the build with PERL_GLOBAL_STRUCT
15954 PerlInterpreter *const to = param->new_perl;
15956 PerlInterpreter *const was = PERL_GET_THX;
15958 PERL_ARGS_ASSERT_CLONE_PARAMS_DEL;
15964 SvREFCNT_dec(param->stashes);
15965 if (param->unreferenced)
15966 unreferenced_to_tmp_stack(param->unreferenced);
15976 Perl_clone_params_new(PerlInterpreter *const from, PerlInterpreter *const to)
15979 /* Need to play this game, as newAV() can call safesysmalloc(), and that
15980 does a dTHX; to get the context from thread local storage.
15981 FIXME - under PERL_CORE Newx(), Safefree() and friends should expand to
15982 a version that passes in my_perl. */
15983 PerlInterpreter *const was = PERL_GET_THX;
15984 CLONE_PARAMS *param;
15986 PERL_ARGS_ASSERT_CLONE_PARAMS_NEW;
15992 /* Given that we've set the context, we can do this unshared. */
15993 Newx(param, 1, CLONE_PARAMS);
15996 param->proto_perl = from;
15997 param->new_perl = to;
15998 param->stashes = (AV *)Perl_newSV_type(to, SVt_PVAV);
15999 AvREAL_off(param->stashes);
16000 param->unreferenced = (AV *)Perl_newSV_type(to, SVt_PVAV);
16008 #endif /* USE_ITHREADS */
16011 Perl_init_constants(pTHX)
16015 SvREFCNT(&PL_sv_undef) = SvREFCNT_IMMORTAL;
16016 SvFLAGS(&PL_sv_undef) = SVf_READONLY|SVf_PROTECT|SVt_NULL;
16017 SvANY(&PL_sv_undef) = NULL;
16019 SvANY(&PL_sv_no) = new_XPVNV();
16020 SvREFCNT(&PL_sv_no) = SvREFCNT_IMMORTAL;
16021 SvFLAGS(&PL_sv_no) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
16022 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
16025 SvANY(&PL_sv_yes) = new_XPVNV();
16026 SvREFCNT(&PL_sv_yes) = SvREFCNT_IMMORTAL;
16027 SvFLAGS(&PL_sv_yes) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
16028 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
16031 SvANY(&PL_sv_zero) = new_XPVNV();
16032 SvREFCNT(&PL_sv_zero) = SvREFCNT_IMMORTAL;
16033 SvFLAGS(&PL_sv_zero) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
16034 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
16038 SvPV_set(&PL_sv_no, (char*)PL_No);
16039 SvCUR_set(&PL_sv_no, 0);
16040 SvLEN_set(&PL_sv_no, 0);
16041 SvIV_set(&PL_sv_no, 0);
16042 SvNV_set(&PL_sv_no, 0);
16044 SvPV_set(&PL_sv_yes, (char*)PL_Yes);
16045 SvCUR_set(&PL_sv_yes, 1);
16046 SvLEN_set(&PL_sv_yes, 0);
16047 SvIV_set(&PL_sv_yes, 1);
16048 SvNV_set(&PL_sv_yes, 1);
16050 SvPV_set(&PL_sv_zero, (char*)PL_Zero);
16051 SvCUR_set(&PL_sv_zero, 1);
16052 SvLEN_set(&PL_sv_zero, 0);
16053 SvIV_set(&PL_sv_zero, 0);
16054 SvNV_set(&PL_sv_zero, 0);
16056 PadnamePV(&PL_padname_const) = (char *)PL_No;
16058 assert(SvIMMORTAL_INTERP(&PL_sv_yes));
16059 assert(SvIMMORTAL_INTERP(&PL_sv_undef));
16060 assert(SvIMMORTAL_INTERP(&PL_sv_no));
16061 assert(SvIMMORTAL_INTERP(&PL_sv_zero));
16063 assert(SvIMMORTAL(&PL_sv_yes));
16064 assert(SvIMMORTAL(&PL_sv_undef));
16065 assert(SvIMMORTAL(&PL_sv_no));
16066 assert(SvIMMORTAL(&PL_sv_zero));
16068 assert( SvIMMORTAL_TRUE(&PL_sv_yes));
16069 assert(!SvIMMORTAL_TRUE(&PL_sv_undef));
16070 assert(!SvIMMORTAL_TRUE(&PL_sv_no));
16071 assert(!SvIMMORTAL_TRUE(&PL_sv_zero));
16073 assert( SvTRUE_nomg_NN(&PL_sv_yes));
16074 assert(!SvTRUE_nomg_NN(&PL_sv_undef));
16075 assert(!SvTRUE_nomg_NN(&PL_sv_no));
16076 assert(!SvTRUE_nomg_NN(&PL_sv_zero));
16080 =head1 Unicode Support
16082 =for apidoc sv_recode_to_utf8
16084 C<encoding> is assumed to be an C<Encode> object, on entry the PV
16085 of C<sv> is assumed to be octets in that encoding, and C<sv>
16086 will be converted into Unicode (and UTF-8).
16088 If C<sv> already is UTF-8 (or if it is not C<POK>), or if C<encoding>
16089 is not a reference, nothing is done to C<sv>. If C<encoding> is not
16090 an C<Encode::XS> Encoding object, bad things will happen.
16091 (See F<cpan/Encode/encoding.pm> and L<Encode>.)
16093 The PV of C<sv> is returned.
16098 Perl_sv_recode_to_utf8(pTHX_ SV *sv, SV *encoding)
16100 PERL_ARGS_ASSERT_SV_RECODE_TO_UTF8;
16102 if (SvPOK(sv) && !SvUTF8(sv) && !IN_BYTES && SvROK(encoding)) {
16111 if (SvPADTMP(nsv)) {
16112 nsv = sv_newmortal();
16113 SvSetSV_nosteal(nsv, sv);
16122 Passing sv_yes is wrong - it needs to be or'ed set of constants
16123 for Encode::XS, while UTf-8 decode (currently) assumes a true value means
16124 remove converted chars from source.
16126 Both will default the value - let them.
16128 XPUSHs(&PL_sv_yes);
16131 call_method("decode", G_SCALAR);
16135 s = SvPV_const(uni, len);
16136 if (s != SvPVX_const(sv)) {
16137 SvGROW(sv, len + 1);
16138 Move(s, SvPVX(sv), len + 1, char);
16139 SvCUR_set(sv, len);
16144 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
16145 /* clear pos and any utf8 cache */
16146 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
16149 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
16150 magic_setutf8(sv,mg); /* clear UTF8 cache */
16155 return SvPOKp(sv) ? SvPVX(sv) : NULL;
16159 =for apidoc sv_cat_decode
16161 C<encoding> is assumed to be an C<Encode> object, the PV of C<ssv> is
16162 assumed to be octets in that encoding and decoding the input starts
16163 from the position which S<C<(PV + *offset)>> pointed to. C<dsv> will be
16164 concatenated with the decoded UTF-8 string from C<ssv>. Decoding will terminate
16165 when the string C<tstr> appears in decoding output or the input ends on
16166 the PV of C<ssv>. The value which C<offset> points will be modified
16167 to the last input position on C<ssv>.
16169 Returns TRUE if the terminator was found, else returns FALSE.
16174 Perl_sv_cat_decode(pTHX_ SV *dsv, SV *encoding,
16175 SV *ssv, int *offset, char *tstr, int tlen)
16179 PERL_ARGS_ASSERT_SV_CAT_DECODE;
16181 if (SvPOK(ssv) && SvPOK(dsv) && SvROK(encoding)) {
16192 offsv = newSViv(*offset);
16194 mPUSHp(tstr, tlen);
16196 call_method("cat_decode", G_SCALAR);
16198 ret = SvTRUE(TOPs);
16199 *offset = SvIV(offsv);
16205 Perl_croak(aTHX_ "Invalid argument to sv_cat_decode");
16210 /* ---------------------------------------------------------------------
16212 * support functions for report_uninit()
16215 /* the maxiumum size of array or hash where we will scan looking
16216 * for the undefined element that triggered the warning */
16218 #define FUV_MAX_SEARCH_SIZE 1000
16220 /* Look for an entry in the hash whose value has the same SV as val;
16221 * If so, return a mortal copy of the key. */
16224 S_find_hash_subscript(pTHX_ const HV *const hv, const SV *const val)
16230 PERL_ARGS_ASSERT_FIND_HASH_SUBSCRIPT;
16232 if (!hv || SvMAGICAL(hv) || !HvARRAY(hv) ||
16233 (HvTOTALKEYS(hv) > FUV_MAX_SEARCH_SIZE))
16236 array = HvARRAY(hv);
16238 for (i=HvMAX(hv); i>=0; i--) {
16240 for (entry = array[i]; entry; entry = HeNEXT(entry)) {
16241 if (HeVAL(entry) != val)
16243 if ( HeVAL(entry) == &PL_sv_undef ||
16244 HeVAL(entry) == &PL_sv_placeholder)
16248 if (HeKLEN(entry) == HEf_SVKEY)
16249 return sv_mortalcopy(HeKEY_sv(entry));
16250 return sv_2mortal(newSVhek(HeKEY_hek(entry)));
16256 /* Look for an entry in the array whose value has the same SV as val;
16257 * If so, return the index, otherwise return -1. */
16260 S_find_array_subscript(pTHX_ const AV *const av, const SV *const val)
16262 PERL_ARGS_ASSERT_FIND_ARRAY_SUBSCRIPT;
16264 if (!av || SvMAGICAL(av) || !AvARRAY(av) ||
16265 (AvFILLp(av) > FUV_MAX_SEARCH_SIZE))
16268 if (val != &PL_sv_undef) {
16269 SV ** const svp = AvARRAY(av);
16272 for (i=AvFILLp(av); i>=0; i--)
16279 /* varname(): return the name of a variable, optionally with a subscript.
16280 * If gv is non-zero, use the name of that global, along with gvtype (one
16281 * of "$", "@", "%"); otherwise use the name of the lexical at pad offset
16282 * targ. Depending on the value of the subscript_type flag, return:
16285 #define FUV_SUBSCRIPT_NONE 1 /* "@foo" */
16286 #define FUV_SUBSCRIPT_ARRAY 2 /* "$foo[aindex]" */
16287 #define FUV_SUBSCRIPT_HASH 3 /* "$foo{keyname}" */
16288 #define FUV_SUBSCRIPT_WITHIN 4 /* "within @foo" */
16291 Perl_varname(pTHX_ const GV *const gv, const char gvtype, PADOFFSET targ,
16292 const SV *const keyname, SSize_t aindex, int subscript_type)
16295 SV * const name = sv_newmortal();
16296 if (gv && isGV(gv)) {
16298 buffer[0] = gvtype;
16301 /* as gv_fullname4(), but add literal '^' for $^FOO names */
16303 gv_fullname4(name, gv, buffer, 0);
16305 if ((unsigned int)SvPVX(name)[1] <= 26) {
16307 buffer[1] = SvPVX(name)[1] + 'A' - 1;
16309 /* Swap the 1 unprintable control character for the 2 byte pretty
16310 version - ie substr($name, 1, 1) = $buffer; */
16311 sv_insert(name, 1, 1, buffer, 2);
16315 CV * const cv = gv ? ((CV *)gv) : find_runcv(NULL);
16318 assert(!cv || SvTYPE(cv) == SVt_PVCV || SvTYPE(cv) == SVt_PVFM);
16320 if (!cv || !CvPADLIST(cv))
16322 sv = padnamelist_fetch(PadlistNAMES(CvPADLIST(cv)), targ);
16323 sv_setpvn(name, PadnamePV(sv), PadnameLEN(sv));
16327 if (subscript_type == FUV_SUBSCRIPT_HASH) {
16328 SV * const sv = newSV(0);
16330 const char * const pv = SvPV_nomg_const((SV*)keyname, len);
16332 *SvPVX(name) = '$';
16333 Perl_sv_catpvf(aTHX_ name, "{%s}",
16334 pv_pretty(sv, pv, len, 32, NULL, NULL,
16335 PERL_PV_PRETTY_DUMP | PERL_PV_ESCAPE_UNI_DETECT ));
16336 SvREFCNT_dec_NN(sv);
16338 else if (subscript_type == FUV_SUBSCRIPT_ARRAY) {
16339 *SvPVX(name) = '$';
16340 Perl_sv_catpvf(aTHX_ name, "[%" IVdf "]", (IV)aindex);
16342 else if (subscript_type == FUV_SUBSCRIPT_WITHIN) {
16343 /* We know that name has no magic, so can use 0 instead of SV_GMAGIC */
16344 Perl_sv_insert_flags(aTHX_ name, 0, 0, STR_WITH_LEN("within "), 0);
16352 =for apidoc find_uninit_var
16354 Find the name of the undefined variable (if any) that caused the operator
16355 to issue a "Use of uninitialized value" warning.
16356 If match is true, only return a name if its value matches C<uninit_sv>.
16357 So roughly speaking, if a unary operator (such as C<OP_COS>) generates a
16358 warning, then following the direct child of the op may yield an
16359 C<OP_PADSV> or C<OP_GV> that gives the name of the undefined variable. On the
16360 other hand, with C<OP_ADD> there are two branches to follow, so we only print
16361 the variable name if we get an exact match.
16362 C<desc_p> points to a string pointer holding the description of the op.
16363 This may be updated if needed.
16365 The name is returned as a mortal SV.
16367 Assumes that C<PL_op> is the OP that originally triggered the error, and that
16368 C<PL_comppad>/C<PL_curpad> points to the currently executing pad.
16374 S_find_uninit_var(pTHX_ const OP *const obase, const SV *const uninit_sv,
16375 bool match, const char **desc_p)
16380 const OP *o, *o2, *kid;
16382 PERL_ARGS_ASSERT_FIND_UNINIT_VAR;
16384 if (!obase || (match && (!uninit_sv || uninit_sv == &PL_sv_undef ||
16385 uninit_sv == &PL_sv_placeholder)))
16388 switch (obase->op_type) {
16391 /* undef should care if its args are undef - any warnings
16392 * will be from tied/magic vars */
16400 const bool pad = ( obase->op_type == OP_PADAV
16401 || obase->op_type == OP_PADHV
16402 || obase->op_type == OP_PADRANGE
16405 const bool hash = ( obase->op_type == OP_PADHV
16406 || obase->op_type == OP_RV2HV
16407 || (obase->op_type == OP_PADRANGE
16408 && SvTYPE(PAD_SVl(obase->op_targ)) == SVt_PVHV)
16412 int subscript_type = FUV_SUBSCRIPT_WITHIN;
16414 if (pad) { /* @lex, %lex */
16415 sv = PAD_SVl(obase->op_targ);
16419 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
16420 /* @global, %global */
16421 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
16424 sv = hash ? MUTABLE_SV(GvHV(gv)): MUTABLE_SV(GvAV(gv));
16426 else if (obase == PL_op) /* @{expr}, %{expr} */
16427 return find_uninit_var(cUNOPx(obase)->op_first,
16428 uninit_sv, match, desc_p);
16429 else /* @{expr}, %{expr} as a sub-expression */
16433 /* attempt to find a match within the aggregate */
16435 keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16437 subscript_type = FUV_SUBSCRIPT_HASH;
16440 index = find_array_subscript((const AV *)sv, uninit_sv);
16442 subscript_type = FUV_SUBSCRIPT_ARRAY;
16445 if (match && subscript_type == FUV_SUBSCRIPT_WITHIN)
16448 return varname(gv, (char)(hash ? '%' : '@'), obase->op_targ,
16449 keysv, index, subscript_type);
16453 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
16455 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
16456 if (!gv || !GvSTASH(gv))
16458 if (match && (GvSV(gv) != uninit_sv))
16460 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
16463 return find_uninit_var(cUNOPx(obase)->op_first, uninit_sv, 1, desc_p);
16466 if (match && PAD_SVl(obase->op_targ) != uninit_sv)
16468 return varname(NULL, '$', obase->op_targ,
16469 NULL, 0, FUV_SUBSCRIPT_NONE);
16472 gv = cGVOPx_gv(obase);
16473 if (!gv || (match && GvSV(gv) != uninit_sv) || !GvSTASH(gv))
16475 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
16477 case OP_AELEMFAST_LEX:
16480 AV *av = MUTABLE_AV(PAD_SV(obase->op_targ));
16481 if (!av || SvRMAGICAL(av))
16483 svp = av_fetch(av, (I8)obase->op_private, FALSE);
16484 if (!svp || *svp != uninit_sv)
16487 return varname(NULL, '$', obase->op_targ,
16488 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
16491 gv = cGVOPx_gv(obase);
16496 AV *const av = GvAV(gv);
16497 if (!av || SvRMAGICAL(av))
16499 svp = av_fetch(av, (I8)obase->op_private, FALSE);
16500 if (!svp || *svp != uninit_sv)
16503 return varname(gv, '$', 0,
16504 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
16506 NOT_REACHED; /* NOTREACHED */
16509 o = cUNOPx(obase)->op_first;
16510 if (!o || o->op_type != OP_NULL ||
16511 ! (o->op_targ == OP_AELEM || o->op_targ == OP_HELEM))
16513 return find_uninit_var(cBINOPo->op_last, uninit_sv, match, desc_p);
16518 bool negate = FALSE;
16520 if (PL_op == obase)
16521 /* $a[uninit_expr] or $h{uninit_expr} */
16522 return find_uninit_var(cBINOPx(obase)->op_last,
16523 uninit_sv, match, desc_p);
16526 o = cBINOPx(obase)->op_first;
16527 kid = cBINOPx(obase)->op_last;
16529 /* get the av or hv, and optionally the gv */
16531 if (o->op_type == OP_PADAV || o->op_type == OP_PADHV) {
16532 sv = PAD_SV(o->op_targ);
16534 else if ((o->op_type == OP_RV2AV || o->op_type == OP_RV2HV)
16535 && cUNOPo->op_first->op_type == OP_GV)
16537 gv = cGVOPx_gv(cUNOPo->op_first);
16541 == OP_RV2HV ? MUTABLE_SV(GvHV(gv)) : MUTABLE_SV(GvAV(gv));
16546 if (kid && kid->op_type == OP_NEGATE) {
16548 kid = cUNOPx(kid)->op_first;
16551 if (kid && kid->op_type == OP_CONST && SvOK(cSVOPx_sv(kid))) {
16552 /* index is constant */
16555 kidsv = newSVpvs_flags("-", SVs_TEMP);
16556 sv_catsv(kidsv, cSVOPx_sv(kid));
16559 kidsv = cSVOPx_sv(kid);
16563 if (obase->op_type == OP_HELEM) {
16564 HE* he = hv_fetch_ent(MUTABLE_HV(sv), kidsv, 0, 0);
16565 if (!he || HeVAL(he) != uninit_sv)
16569 SV * const opsv = cSVOPx_sv(kid);
16570 const IV opsviv = SvIV(opsv);
16571 SV * const * const svp = av_fetch(MUTABLE_AV(sv),
16572 negate ? - opsviv : opsviv,
16574 if (!svp || *svp != uninit_sv)
16578 if (obase->op_type == OP_HELEM)
16579 return varname(gv, '%', o->op_targ,
16580 kidsv, 0, FUV_SUBSCRIPT_HASH);
16582 return varname(gv, '@', o->op_targ, NULL,
16583 negate ? - SvIV(cSVOPx_sv(kid)) : SvIV(cSVOPx_sv(kid)),
16584 FUV_SUBSCRIPT_ARRAY);
16587 /* index is an expression;
16588 * attempt to find a match within the aggregate */
16589 if (obase->op_type == OP_HELEM) {
16590 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16592 return varname(gv, '%', o->op_targ,
16593 keysv, 0, FUV_SUBSCRIPT_HASH);
16596 const SSize_t index
16597 = find_array_subscript((const AV *)sv, uninit_sv);
16599 return varname(gv, '@', o->op_targ,
16600 NULL, index, FUV_SUBSCRIPT_ARRAY);
16605 (char)((o->op_type == OP_PADAV || o->op_type == OP_RV2AV)
16607 o->op_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
16609 NOT_REACHED; /* NOTREACHED */
16612 case OP_MULTIDEREF: {
16613 /* If we were executing OP_MULTIDEREF when the undef warning
16614 * triggered, then it must be one of the index values within
16615 * that triggered it. If not, then the only possibility is that
16616 * the value retrieved by the last aggregate index might be the
16617 * culprit. For the former, we set PL_multideref_pc each time before
16618 * using an index, so work though the item list until we reach
16619 * that point. For the latter, just work through the entire item
16620 * list; the last aggregate retrieved will be the candidate.
16621 * There is a third rare possibility: something triggered
16622 * magic while fetching an array/hash element. Just display
16623 * nothing in this case.
16626 /* the named aggregate, if any */
16627 PADOFFSET agg_targ = 0;
16629 /* the last-seen index */
16631 PADOFFSET index_targ;
16633 IV index_const_iv = 0; /* init for spurious compiler warn */
16634 SV *index_const_sv;
16635 int depth = 0; /* how many array/hash lookups we've done */
16637 UNOP_AUX_item *items = cUNOP_AUXx(obase)->op_aux;
16638 UNOP_AUX_item *last = NULL;
16639 UV actions = items->uv;
16642 if (PL_op == obase) {
16643 last = PL_multideref_pc;
16644 assert(last >= items && last <= items + items[-1].uv);
16651 switch (actions & MDEREF_ACTION_MASK) {
16653 case MDEREF_reload:
16654 actions = (++items)->uv;
16657 case MDEREF_HV_padhv_helem: /* $lex{...} */
16660 case MDEREF_AV_padav_aelem: /* $lex[...] */
16661 agg_targ = (++items)->pad_offset;
16665 case MDEREF_HV_gvhv_helem: /* $pkg{...} */
16668 case MDEREF_AV_gvav_aelem: /* $pkg[...] */
16670 agg_gv = (GV*)UNOP_AUX_item_sv(++items);
16671 assert(isGV_with_GP(agg_gv));
16674 case MDEREF_HV_gvsv_vivify_rv2hv_helem: /* $pkg->{...} */
16675 case MDEREF_HV_padsv_vivify_rv2hv_helem: /* $lex->{...} */
16678 case MDEREF_HV_pop_rv2hv_helem: /* expr->{...} */
16679 case MDEREF_HV_vivify_rv2hv_helem: /* vivify, ->{...} */
16685 case MDEREF_AV_gvsv_vivify_rv2av_aelem: /* $pkg->[...] */
16686 case MDEREF_AV_padsv_vivify_rv2av_aelem: /* $lex->[...] */
16689 case MDEREF_AV_pop_rv2av_aelem: /* expr->[...] */
16690 case MDEREF_AV_vivify_rv2av_aelem: /* vivify, ->[...] */
16697 index_const_sv = NULL;
16699 index_type = (actions & MDEREF_INDEX_MASK);
16700 switch (index_type) {
16701 case MDEREF_INDEX_none:
16703 case MDEREF_INDEX_const:
16705 index_const_sv = UNOP_AUX_item_sv(++items)
16707 index_const_iv = (++items)->iv;
16709 case MDEREF_INDEX_padsv:
16710 index_targ = (++items)->pad_offset;
16712 case MDEREF_INDEX_gvsv:
16713 index_gv = (GV*)UNOP_AUX_item_sv(++items);
16714 assert(isGV_with_GP(index_gv));
16718 if (index_type != MDEREF_INDEX_none)
16721 if ( index_type == MDEREF_INDEX_none
16722 || (actions & MDEREF_FLAG_last)
16723 || (last && items >= last)
16727 actions >>= MDEREF_SHIFT;
16730 if (PL_op == obase) {
16731 /* most likely index was undef */
16733 *desc_p = ( (actions & MDEREF_FLAG_last)
16734 && (obase->op_private
16735 & (OPpMULTIDEREF_EXISTS|OPpMULTIDEREF_DELETE)))
16737 (obase->op_private & OPpMULTIDEREF_EXISTS)
16740 : is_hv ? "hash element" : "array element";
16741 assert(index_type != MDEREF_INDEX_none);
16743 if (GvSV(index_gv) == uninit_sv)
16744 return varname(index_gv, '$', 0, NULL, 0,
16745 FUV_SUBSCRIPT_NONE);
16750 if (PL_curpad[index_targ] == uninit_sv)
16751 return varname(NULL, '$', index_targ,
16752 NULL, 0, FUV_SUBSCRIPT_NONE);
16756 /* If we got to this point it was undef on a const subscript,
16757 * so magic probably involved, e.g. $ISA[0]. Give up. */
16761 /* the SV returned by pp_multideref() was undef, if anything was */
16767 sv = PAD_SV(agg_targ);
16769 sv = is_hv ? MUTABLE_SV(GvHV(agg_gv)) : MUTABLE_SV(GvAV(agg_gv));
16776 if (index_type == MDEREF_INDEX_const) {
16781 HE* he = hv_fetch_ent(MUTABLE_HV(sv), index_const_sv, 0, 0);
16782 if (!he || HeVAL(he) != uninit_sv)
16786 SV * const * const svp =
16787 av_fetch(MUTABLE_AV(sv), index_const_iv, FALSE);
16788 if (!svp || *svp != uninit_sv)
16793 ? varname(agg_gv, '%', agg_targ,
16794 index_const_sv, 0, FUV_SUBSCRIPT_HASH)
16795 : varname(agg_gv, '@', agg_targ,
16796 NULL, index_const_iv, FUV_SUBSCRIPT_ARRAY);
16799 /* index is an var */
16801 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16803 return varname(agg_gv, '%', agg_targ,
16804 keysv, 0, FUV_SUBSCRIPT_HASH);
16807 const SSize_t index
16808 = find_array_subscript((const AV *)sv, uninit_sv);
16810 return varname(agg_gv, '@', agg_targ,
16811 NULL, index, FUV_SUBSCRIPT_ARRAY);
16815 return varname(agg_gv,
16817 agg_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
16819 NOT_REACHED; /* NOTREACHED */
16823 /* only examine RHS */
16824 return find_uninit_var(cBINOPx(obase)->op_first, uninit_sv,
16828 o = cUNOPx(obase)->op_first;
16829 if ( o->op_type == OP_PUSHMARK
16830 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)
16834 if (!OpHAS_SIBLING(o)) {
16835 /* one-arg version of open is highly magical */
16837 if (o->op_type == OP_GV) { /* open FOO; */
16839 if (match && GvSV(gv) != uninit_sv)
16841 return varname(gv, '$', 0,
16842 NULL, 0, FUV_SUBSCRIPT_NONE);
16844 /* other possibilities not handled are:
16845 * open $x; or open my $x; should return '${*$x}'
16846 * open expr; should return '$'.expr ideally
16853 /* ops where $_ may be an implicit arg */
16858 if ( !(obase->op_flags & OPf_STACKED)) {
16859 if (uninit_sv == DEFSV)
16860 return newSVpvs_flags("$_", SVs_TEMP);
16861 else if (obase->op_targ
16862 && uninit_sv == PAD_SVl(obase->op_targ))
16863 return varname(NULL, '$', obase->op_targ, NULL, 0,
16864 FUV_SUBSCRIPT_NONE);
16871 match = 1; /* print etc can return undef on defined args */
16872 /* skip filehandle as it can't produce 'undef' warning */
16873 o = cUNOPx(obase)->op_first;
16874 if ((obase->op_flags & OPf_STACKED)
16876 ( o->op_type == OP_PUSHMARK
16877 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)))
16878 o = OpSIBLING(OpSIBLING(o));
16882 case OP_ENTEREVAL: /* could be eval $undef or $x='$undef'; eval $x */
16883 case OP_CUSTOM: /* XS or custom code could trigger random warnings */
16885 /* the following ops are capable of returning PL_sv_undef even for
16886 * defined arg(s) */
16905 case OP_GETPEERNAME:
16952 case OP_SMARTMATCH:
16961 /* XXX tmp hack: these two may call an XS sub, and currently
16962 XS subs don't have a SUB entry on the context stack, so CV and
16963 pad determination goes wrong, and BAD things happen. So, just
16964 don't try to determine the value under those circumstances.
16965 Need a better fix at dome point. DAPM 11/2007 */
16971 GV * const gv = gv_fetchpvs(".", GV_NOTQUAL, SVt_PV);
16972 if (gv && GvSV(gv) == uninit_sv)
16973 return newSVpvs_flags("$.", SVs_TEMP);
16978 /* def-ness of rval pos() is independent of the def-ness of its arg */
16979 if ( !(obase->op_flags & OPf_MOD))
16985 if (SvROK(PL_rs) && uninit_sv == SvRV(PL_rs))
16986 return newSVpvs_flags("${$/}", SVs_TEMP);
16991 if (!(obase->op_flags & OPf_KIDS))
16993 o = cUNOPx(obase)->op_first;
16999 /* This loop checks all the kid ops, skipping any that cannot pos-
17000 * sibly be responsible for the uninitialized value; i.e., defined
17001 * constants and ops that return nothing. If there is only one op
17002 * left that is not skipped, then we *know* it is responsible for
17003 * the uninitialized value. If there is more than one op left, we
17004 * have to look for an exact match in the while() loop below.
17005 * Note that we skip padrange, because the individual pad ops that
17006 * it replaced are still in the tree, so we work on them instead.
17009 for (kid=o; kid; kid = OpSIBLING(kid)) {
17010 const OPCODE type = kid->op_type;
17011 if ( (type == OP_CONST && SvOK(cSVOPx_sv(kid)))
17012 || (type == OP_NULL && ! (kid->op_flags & OPf_KIDS))
17013 || (type == OP_PUSHMARK)
17014 || (type == OP_PADRANGE)
17018 if (o2) { /* more than one found */
17025 return find_uninit_var(o2, uninit_sv, match, desc_p);
17027 /* scan all args */
17029 sv = find_uninit_var(o, uninit_sv, 1, desc_p);
17041 =for apidoc report_uninit
17043 Print appropriate "Use of uninitialized variable" warning.
17049 Perl_report_uninit(pTHX_ const SV *uninit_sv)
17051 const char *desc = NULL;
17052 SV* varname = NULL;
17055 desc = PL_op->op_type == OP_STRINGIFY && PL_op->op_folded
17057 : PL_op->op_type == OP_MULTICONCAT
17058 && (PL_op->op_private & OPpMULTICONCAT_FAKE)
17061 if (uninit_sv && PL_curpad) {
17062 varname = find_uninit_var(PL_op, uninit_sv, 0, &desc);
17064 sv_insert(varname, 0, 0, " ", 1);
17067 else if (PL_curstackinfo->si_type == PERLSI_SORT && cxstack_ix == 0)
17068 /* we've reached the end of a sort block or sub,
17069 * and the uninit value is probably what that code returned */
17072 /* PL_warn_uninit_sv is constant */
17073 GCC_DIAG_IGNORE_STMT(-Wformat-nonliteral);
17075 /* diag_listed_as: Use of uninitialized value%s */
17076 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit_sv,
17077 SVfARG(varname ? varname : &PL_sv_no),
17080 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit,
17082 GCC_DIAG_RESTORE_STMT;
17086 * ex: set ts=8 sts=4 sw=4 et: