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 */
891 /* With -DPURFIY we allocate everything directly, and don't use arenas.
892 This seems a rather elegant way to simplify some of the code below. */
893 #define HASARENA FALSE
895 #define HASARENA TRUE
897 #define NOARENA FALSE
899 /* Size the arenas to exactly fit a given number of bodies. A count
900 of 0 fits the max number bodies into a PERL_ARENA_SIZE.block,
901 simplifying the default. If count > 0, the arena is sized to fit
902 only that many bodies, allowing arenas to be used for large, rare
903 bodies (XPVFM, XPVIO) without undue waste. The arena size is
904 limited by PERL_ARENA_SIZE, so we can safely oversize the
907 #define FIT_ARENA0(body_size) \
908 ((size_t)(PERL_ARENA_SIZE / body_size) * body_size)
909 #define FIT_ARENAn(count,body_size) \
910 ( count * body_size <= PERL_ARENA_SIZE) \
911 ? count * body_size \
912 : FIT_ARENA0 (body_size)
913 #define FIT_ARENA(count,body_size) \
915 ? FIT_ARENAn (count, body_size) \
916 : FIT_ARENA0 (body_size))
918 /* Calculate the length to copy. Specifically work out the length less any
919 final padding the compiler needed to add. See the comment in sv_upgrade
920 for why copying the padding proved to be a bug. */
922 #define copy_length(type, last_member) \
923 STRUCT_OFFSET(type, last_member) \
924 + sizeof (((type*)SvANY((const SV *)0))->last_member)
926 static const struct body_details bodies_by_type[] = {
927 /* HEs use this offset for their arena. */
928 { 0, 0, 0, SVt_NULL, FALSE, NONV, NOARENA, 0 },
930 /* IVs are in the head, so the allocation size is 0. */
932 sizeof(IV), /* This is used to copy out the IV body. */
933 STRUCT_OFFSET(XPVIV, xiv_iv), SVt_IV, FALSE, NONV,
934 NOARENA /* IVS don't need an arena */, 0
939 STRUCT_OFFSET(XPVNV, xnv_u),
940 SVt_NV, FALSE, HADNV, NOARENA, 0 },
942 { sizeof(NV), sizeof(NV),
943 STRUCT_OFFSET(XPVNV, xnv_u),
944 SVt_NV, FALSE, HADNV, HASARENA, FIT_ARENA(0, sizeof(NV)) },
947 { sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur),
948 copy_length(XPV, xpv_len) - STRUCT_OFFSET(XPV, xpv_cur),
949 + STRUCT_OFFSET(XPV, xpv_cur),
950 SVt_PV, FALSE, NONV, HASARENA,
951 FIT_ARENA(0, sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur)) },
953 { sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur),
954 copy_length(XINVLIST, is_offset) - STRUCT_OFFSET(XPV, xpv_cur),
955 + STRUCT_OFFSET(XPV, xpv_cur),
956 SVt_INVLIST, TRUE, NONV, HASARENA,
957 FIT_ARENA(0, sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur)) },
959 { sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur),
960 copy_length(XPVIV, xiv_u) - STRUCT_OFFSET(XPV, xpv_cur),
961 + STRUCT_OFFSET(XPV, xpv_cur),
962 SVt_PVIV, FALSE, NONV, HASARENA,
963 FIT_ARENA(0, sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur)) },
965 { sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur),
966 copy_length(XPVNV, xnv_u) - STRUCT_OFFSET(XPV, xpv_cur),
967 + STRUCT_OFFSET(XPV, xpv_cur),
968 SVt_PVNV, FALSE, HADNV, HASARENA,
969 FIT_ARENA(0, sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur)) },
971 { sizeof(XPVMG), copy_length(XPVMG, xnv_u), 0, SVt_PVMG, FALSE, HADNV,
972 HASARENA, FIT_ARENA(0, sizeof(XPVMG)) },
977 SVt_REGEXP, TRUE, NONV, HASARENA,
978 FIT_ARENA(0, sizeof(regexp))
981 { sizeof(XPVGV), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV,
982 HASARENA, FIT_ARENA(0, sizeof(XPVGV)) },
984 { sizeof(XPVLV), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV,
985 HASARENA, FIT_ARENA(0, sizeof(XPVLV)) },
988 copy_length(XPVAV, xav_alloc),
990 SVt_PVAV, TRUE, NONV, HASARENA,
991 FIT_ARENA(0, sizeof(XPVAV)) },
994 copy_length(XPVHV, xhv_max),
996 SVt_PVHV, TRUE, NONV, HASARENA,
997 FIT_ARENA(0, sizeof(XPVHV)) },
1002 SVt_PVCV, TRUE, NONV, HASARENA,
1003 FIT_ARENA(0, sizeof(XPVCV)) },
1008 SVt_PVFM, TRUE, NONV, NOARENA,
1009 FIT_ARENA(20, sizeof(XPVFM)) },
1014 SVt_PVIO, TRUE, NONV, HASARENA,
1015 FIT_ARENA(24, sizeof(XPVIO)) },
1018 #define new_body_allocated(sv_type) \
1019 (void *)((char *)S_new_body(aTHX_ sv_type) \
1020 - bodies_by_type[sv_type].offset)
1022 /* return a thing to the free list */
1024 #define del_body(thing, root) \
1026 void ** const thing_copy = (void **)thing; \
1027 *thing_copy = *root; \
1028 *root = (void*)thing_copy; \
1032 #if !(NVSIZE <= IVSIZE)
1033 # define new_XNV() safemalloc(sizeof(XPVNV))
1035 #define new_XPVNV() safemalloc(sizeof(XPVNV))
1036 #define new_XPVMG() safemalloc(sizeof(XPVMG))
1038 #define del_XPVGV(p) safefree(p)
1042 #if !(NVSIZE <= IVSIZE)
1043 # define new_XNV() new_body_allocated(SVt_NV)
1045 #define new_XPVNV() new_body_allocated(SVt_PVNV)
1046 #define new_XPVMG() new_body_allocated(SVt_PVMG)
1048 #define del_XPVGV(p) del_body(p + bodies_by_type[SVt_PVGV].offset, \
1049 &PL_body_roots[SVt_PVGV])
1053 /* no arena for you! */
1055 #define new_NOARENA(details) \
1056 safemalloc((details)->body_size + (details)->offset)
1057 #define new_NOARENAZ(details) \
1058 safecalloc((details)->body_size + (details)->offset, 1)
1061 Perl_more_bodies (pTHX_ const svtype sv_type, const size_t body_size,
1062 const size_t arena_size)
1064 void ** const root = &PL_body_roots[sv_type];
1065 struct arena_desc *adesc;
1066 struct arena_set *aroot = (struct arena_set *) PL_body_arenas;
1070 const size_t good_arena_size = Perl_malloc_good_size(arena_size);
1071 #if defined(DEBUGGING) && defined(PERL_GLOBAL_STRUCT)
1074 #if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT)
1075 static bool done_sanity_check;
1077 /* PERL_GLOBAL_STRUCT cannot coexist with global
1078 * variables like done_sanity_check. */
1079 if (!done_sanity_check) {
1080 unsigned int i = SVt_LAST;
1082 done_sanity_check = TRUE;
1085 assert (bodies_by_type[i].type == i);
1091 /* may need new arena-set to hold new arena */
1092 if (!aroot || aroot->curr >= aroot->set_size) {
1093 struct arena_set *newroot;
1094 Newxz(newroot, 1, struct arena_set);
1095 newroot->set_size = ARENAS_PER_SET;
1096 newroot->next = aroot;
1098 PL_body_arenas = (void *) newroot;
1099 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot));
1102 /* ok, now have arena-set with at least 1 empty/available arena-desc */
1103 curr = aroot->curr++;
1104 adesc = &(aroot->set[curr]);
1105 assert(!adesc->arena);
1107 Newx(adesc->arena, good_arena_size, char);
1108 adesc->size = good_arena_size;
1109 adesc->utype = sv_type;
1110 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %" UVuf "\n",
1111 curr, (void*)adesc->arena, (UV)good_arena_size));
1113 start = (char *) adesc->arena;
1115 /* Get the address of the byte after the end of the last body we can fit.
1116 Remember, this is integer division: */
1117 end = start + good_arena_size / body_size * body_size;
1119 /* computed count doesn't reflect the 1st slot reservation */
1120 #if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE)
1121 DEBUG_m(PerlIO_printf(Perl_debug_log,
1122 "arena %p end %p arena-size %d (from %d) type %d "
1124 (void*)start, (void*)end, (int)good_arena_size,
1125 (int)arena_size, sv_type, (int)body_size,
1126 (int)good_arena_size / (int)body_size));
1128 DEBUG_m(PerlIO_printf(Perl_debug_log,
1129 "arena %p end %p arena-size %d type %d size %d ct %d\n",
1130 (void*)start, (void*)end,
1131 (int)arena_size, sv_type, (int)body_size,
1132 (int)good_arena_size / (int)body_size));
1134 *root = (void *)start;
1137 /* Where the next body would start: */
1138 char * const next = start + body_size;
1141 /* This is the last body: */
1142 assert(next == end);
1144 *(void **)start = 0;
1148 *(void**) start = (void *)next;
1153 /* grab a new thing from the free list, allocating more if necessary.
1154 The inline version is used for speed in hot routines, and the
1155 function using it serves the rest (unless PURIFY).
1157 #define new_body_inline(xpv, sv_type) \
1159 void ** const r3wt = &PL_body_roots[sv_type]; \
1160 xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \
1161 ? *((void **)(r3wt)) : Perl_more_bodies(aTHX_ sv_type, \
1162 bodies_by_type[sv_type].body_size,\
1163 bodies_by_type[sv_type].arena_size)); \
1164 *(r3wt) = *(void**)(xpv); \
1170 S_new_body(pTHX_ const svtype sv_type)
1173 new_body_inline(xpv, sv_type);
1179 static const struct body_details fake_rv =
1180 { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 };
1183 =for apidoc sv_upgrade
1185 Upgrade an SV to a more complex form. Generally adds a new body type to the
1186 SV, then copies across as much information as possible from the old body.
1187 It croaks if the SV is already in a more complex form than requested. You
1188 generally want to use the C<SvUPGRADE> macro wrapper, which checks the type
1189 before calling C<sv_upgrade>, and hence does not croak. See also
1196 Perl_sv_upgrade(pTHX_ SV *const sv, svtype new_type)
1200 const svtype old_type = SvTYPE(sv);
1201 const struct body_details *new_type_details;
1202 const struct body_details *old_type_details
1203 = bodies_by_type + old_type;
1204 SV *referent = NULL;
1206 PERL_ARGS_ASSERT_SV_UPGRADE;
1208 if (old_type == new_type)
1211 /* This clause was purposefully added ahead of the early return above to
1212 the shared string hackery for (sort {$a <=> $b} keys %hash), with the
1213 inference by Nick I-S that it would fix other troublesome cases. See
1214 changes 7162, 7163 (f130fd4589cf5fbb24149cd4db4137c8326f49c1 and parent)
1216 Given that shared hash key scalars are no longer PVIV, but PV, there is
1217 no longer need to unshare so as to free up the IVX slot for its proper
1218 purpose. So it's safe to move the early return earlier. */
1220 if (new_type > SVt_PVMG && SvIsCOW(sv)) {
1221 sv_force_normal_flags(sv, 0);
1224 old_body = SvANY(sv);
1226 /* Copying structures onto other structures that have been neatly zeroed
1227 has a subtle gotcha. Consider XPVMG
1229 +------+------+------+------+------+-------+-------+
1230 | NV | CUR | LEN | IV | MAGIC | STASH |
1231 +------+------+------+------+------+-------+-------+
1232 0 4 8 12 16 20 24 28
1234 where NVs are aligned to 8 bytes, so that sizeof that structure is
1235 actually 32 bytes long, with 4 bytes of padding at the end:
1237 +------+------+------+------+------+-------+-------+------+
1238 | NV | CUR | LEN | IV | MAGIC | STASH | ??? |
1239 +------+------+------+------+------+-------+-------+------+
1240 0 4 8 12 16 20 24 28 32
1242 so what happens if you allocate memory for this structure:
1244 +------+------+------+------+------+-------+-------+------+------+...
1245 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME |
1246 +------+------+------+------+------+-------+-------+------+------+...
1247 0 4 8 12 16 20 24 28 32 36
1249 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you
1250 expect, because you copy the area marked ??? onto GP. Now, ??? may have
1251 started out as zero once, but it's quite possible that it isn't. So now,
1252 rather than a nicely zeroed GP, you have it pointing somewhere random.
1255 (In fact, GP ends up pointing at a previous GP structure, because the
1256 principle cause of the padding in XPVMG getting garbage is a copy of
1257 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now
1258 this happens to be moot because XPVGV has been re-ordered, with GP
1259 no longer after STASH)
1261 So we are careful and work out the size of used parts of all the
1269 referent = SvRV(sv);
1270 old_type_details = &fake_rv;
1271 if (new_type == SVt_NV)
1272 new_type = SVt_PVNV;
1274 if (new_type < SVt_PVIV) {
1275 new_type = (new_type == SVt_NV)
1276 ? SVt_PVNV : SVt_PVIV;
1281 if (new_type < SVt_PVNV) {
1282 new_type = SVt_PVNV;
1286 assert(new_type > SVt_PV);
1287 STATIC_ASSERT_STMT(SVt_IV < SVt_PV);
1288 STATIC_ASSERT_STMT(SVt_NV < SVt_PV);
1295 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena,
1296 there's no way that it can be safely upgraded, because perl.c
1297 expects to Safefree(SvANY(PL_mess_sv)) */
1298 assert(sv != PL_mess_sv);
1301 if (UNLIKELY(old_type_details->cant_upgrade))
1302 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf,
1303 sv_reftype(sv, 0), (UV) old_type, (UV) new_type);
1306 if (UNLIKELY(old_type > new_type))
1307 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d",
1308 (int)old_type, (int)new_type);
1310 new_type_details = bodies_by_type + new_type;
1312 SvFLAGS(sv) &= ~SVTYPEMASK;
1313 SvFLAGS(sv) |= new_type;
1315 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of
1316 the return statements above will have triggered. */
1317 assert (new_type != SVt_NULL);
1320 assert(old_type == SVt_NULL);
1321 SET_SVANY_FOR_BODYLESS_IV(sv);
1325 assert(old_type == SVt_NULL);
1326 #if NVSIZE <= IVSIZE
1327 SET_SVANY_FOR_BODYLESS_NV(sv);
1329 SvANY(sv) = new_XNV();
1335 assert(new_type_details->body_size);
1338 assert(new_type_details->arena);
1339 assert(new_type_details->arena_size);
1340 /* This points to the start of the allocated area. */
1341 new_body_inline(new_body, new_type);
1342 Zero(new_body, new_type_details->body_size, char);
1343 new_body = ((char *)new_body) - new_type_details->offset;
1345 /* We always allocated the full length item with PURIFY. To do this
1346 we fake things so that arena is false for all 16 types.. */
1347 new_body = new_NOARENAZ(new_type_details);
1349 SvANY(sv) = new_body;
1350 if (new_type == SVt_PVAV) {
1354 if (old_type_details->body_size) {
1357 /* It will have been zeroed when the new body was allocated.
1358 Lets not write to it, in case it confuses a write-back
1364 #ifndef NODEFAULT_SHAREKEYS
1365 HvSHAREKEYS_on(sv); /* key-sharing on by default */
1367 /* start with PERL_HASH_DEFAULT_HvMAX+1 buckets: */
1368 HvMAX(sv) = PERL_HASH_DEFAULT_HvMAX;
1371 /* SVt_NULL isn't the only thing upgraded to AV or HV.
1372 The target created by newSVrv also is, and it can have magic.
1373 However, it never has SvPVX set.
1375 if (old_type == SVt_IV) {
1377 } else if (old_type >= SVt_PV) {
1378 assert(SvPVX_const(sv) == 0);
1381 if (old_type >= SVt_PVMG) {
1382 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic);
1383 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash);
1385 sv->sv_u.svu_array = NULL; /* or svu_hash */
1390 /* XXX Is this still needed? Was it ever needed? Surely as there is
1391 no route from NV to PVIV, NOK can never be true */
1392 assert(!SvNOKp(sv));
1406 assert(new_type_details->body_size);
1407 /* We always allocated the full length item with PURIFY. To do this
1408 we fake things so that arena is false for all 16 types.. */
1409 if(new_type_details->arena) {
1410 /* This points to the start of the allocated area. */
1411 new_body_inline(new_body, new_type);
1412 Zero(new_body, new_type_details->body_size, char);
1413 new_body = ((char *)new_body) - new_type_details->offset;
1415 new_body = new_NOARENAZ(new_type_details);
1417 SvANY(sv) = new_body;
1419 if (old_type_details->copy) {
1420 /* There is now the potential for an upgrade from something without
1421 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */
1422 int offset = old_type_details->offset;
1423 int length = old_type_details->copy;
1425 if (new_type_details->offset > old_type_details->offset) {
1426 const int difference
1427 = new_type_details->offset - old_type_details->offset;
1428 offset += difference;
1429 length -= difference;
1431 assert (length >= 0);
1433 Copy((char *)old_body + offset, (char *)new_body + offset, length,
1437 #ifndef NV_ZERO_IS_ALLBITS_ZERO
1438 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a
1439 * correct 0.0 for us. Otherwise, if the old body didn't have an
1440 * NV slot, but the new one does, then we need to initialise the
1441 * freshly created NV slot with whatever the correct bit pattern is
1443 if (old_type_details->zero_nv && !new_type_details->zero_nv
1444 && !isGV_with_GP(sv))
1448 if (UNLIKELY(new_type == SVt_PVIO)) {
1449 IO * const io = MUTABLE_IO(sv);
1450 GV *iogv = gv_fetchpvs("IO::File::", GV_ADD, SVt_PVHV);
1453 /* Clear the stashcache because a new IO could overrule a package
1455 DEBUG_o(Perl_deb(aTHX_ "sv_upgrade clearing PL_stashcache\n"));
1456 hv_clear(PL_stashcache);
1458 SvSTASH_set(io, MUTABLE_HV(SvREFCNT_inc(GvHV(iogv))));
1459 IoPAGE_LEN(sv) = 60;
1461 if (old_type < SVt_PV) {
1462 /* referent will be NULL unless the old type was SVt_IV emulating
1464 sv->sv_u.svu_rv = referent;
1468 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1469 (unsigned long)new_type);
1472 /* if this is zero, this is a body-less SVt_NULL, SVt_IV/SVt_RV,
1473 and sometimes SVt_NV */
1474 if (old_type_details->body_size) {
1478 /* Note that there is an assumption that all bodies of types that
1479 can be upgraded came from arenas. Only the more complex non-
1480 upgradable types are allowed to be directly malloc()ed. */
1481 assert(old_type_details->arena);
1482 del_body((void*)((char*)old_body + old_type_details->offset),
1483 &PL_body_roots[old_type]);
1489 =for apidoc sv_backoff
1491 Remove any string offset. You should normally use the C<SvOOK_off> macro
1497 /* prior to 5.000 stable, this function returned the new OOK-less SvFLAGS
1498 prior to 5.23.4 this function always returned 0
1502 Perl_sv_backoff(SV *const sv)
1505 const char * const s = SvPVX_const(sv);
1507 PERL_ARGS_ASSERT_SV_BACKOFF;
1510 assert(SvTYPE(sv) != SVt_PVHV);
1511 assert(SvTYPE(sv) != SVt_PVAV);
1513 SvOOK_offset(sv, delta);
1515 SvLEN_set(sv, SvLEN(sv) + delta);
1516 SvPV_set(sv, SvPVX(sv) - delta);
1517 SvFLAGS(sv) &= ~SVf_OOK;
1518 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
1523 /* forward declaration */
1524 static void S_sv_uncow(pTHX_ SV * const sv, const U32 flags);
1530 Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1531 upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1532 Use the C<SvGROW> wrapper instead.
1539 Perl_sv_grow(pTHX_ SV *const sv, STRLEN newlen)
1543 PERL_ARGS_ASSERT_SV_GROW;
1547 if (SvTYPE(sv) < SVt_PV) {
1548 sv_upgrade(sv, SVt_PV);
1549 s = SvPVX_mutable(sv);
1551 else if (SvOOK(sv)) { /* pv is offset? */
1553 s = SvPVX_mutable(sv);
1554 if (newlen > SvLEN(sv))
1555 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
1559 if (SvIsCOW(sv)) S_sv_uncow(aTHX_ sv, 0);
1560 s = SvPVX_mutable(sv);
1563 #ifdef PERL_COPY_ON_WRITE
1564 /* the new COW scheme uses SvPVX(sv)[SvLEN(sv)-1] (if spare)
1565 * to store the COW count. So in general, allocate one more byte than
1566 * asked for, to make it likely this byte is always spare: and thus
1567 * make more strings COW-able.
1569 * Only increment if the allocation isn't MEM_SIZE_MAX,
1570 * otherwise it will wrap to 0.
1572 if ( newlen != MEM_SIZE_MAX )
1576 #if defined(PERL_USE_MALLOC_SIZE) && defined(Perl_safesysmalloc_size)
1577 #define PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1580 if (newlen > SvLEN(sv)) { /* need more room? */
1581 STRLEN minlen = SvCUR(sv);
1582 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
1583 if (newlen < minlen)
1585 #ifndef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1587 /* Don't round up on the first allocation, as odds are pretty good that
1588 * the initial request is accurate as to what is really needed */
1590 STRLEN rounded = PERL_STRLEN_ROUNDUP(newlen);
1591 if (rounded > newlen)
1595 if (SvLEN(sv) && s) {
1596 s = (char*)saferealloc(s, newlen);
1599 s = (char*)safemalloc(newlen);
1600 if (SvPVX_const(sv) && SvCUR(sv)) {
1601 Move(SvPVX_const(sv), s, SvCUR(sv), char);
1605 #ifdef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1606 /* Do this here, do it once, do it right, and then we will never get
1607 called back into sv_grow() unless there really is some growing
1609 SvLEN_set(sv, Perl_safesysmalloc_size(s));
1611 SvLEN_set(sv, newlen);
1618 =for apidoc sv_setiv
1620 Copies an integer into the given SV, upgrading first if necessary.
1621 Does not handle 'set' magic. See also C<L</sv_setiv_mg>>.
1627 Perl_sv_setiv(pTHX_ SV *const sv, const IV i)
1629 PERL_ARGS_ASSERT_SV_SETIV;
1631 SV_CHECK_THINKFIRST_COW_DROP(sv);
1632 switch (SvTYPE(sv)) {
1635 sv_upgrade(sv, SVt_IV);
1638 sv_upgrade(sv, SVt_PVIV);
1642 if (!isGV_with_GP(sv))
1650 /* diag_listed_as: Can't coerce %s to %s in %s */
1651 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1653 NOT_REACHED; /* NOTREACHED */
1657 (void)SvIOK_only(sv); /* validate number */
1663 =for apidoc sv_setiv_mg
1665 Like C<sv_setiv>, but also handles 'set' magic.
1671 Perl_sv_setiv_mg(pTHX_ SV *const sv, const IV i)
1673 PERL_ARGS_ASSERT_SV_SETIV_MG;
1680 =for apidoc sv_setuv
1682 Copies an unsigned integer into the given SV, upgrading first if necessary.
1683 Does not handle 'set' magic. See also C<L</sv_setuv_mg>>.
1689 Perl_sv_setuv(pTHX_ SV *const sv, const UV u)
1691 PERL_ARGS_ASSERT_SV_SETUV;
1693 /* With the if statement to ensure that integers are stored as IVs whenever
1695 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
1698 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1700 If you wish to remove the following if statement, so that this routine
1701 (and its callers) always return UVs, please benchmark to see what the
1702 effect is. Modern CPUs may be different. Or may not :-)
1704 if (u <= (UV)IV_MAX) {
1705 sv_setiv(sv, (IV)u);
1714 =for apidoc sv_setuv_mg
1716 Like C<sv_setuv>, but also handles 'set' magic.
1722 Perl_sv_setuv_mg(pTHX_ SV *const sv, const UV u)
1724 PERL_ARGS_ASSERT_SV_SETUV_MG;
1731 =for apidoc sv_setnv
1733 Copies a double into the given SV, upgrading first if necessary.
1734 Does not handle 'set' magic. See also C<L</sv_setnv_mg>>.
1740 Perl_sv_setnv(pTHX_ SV *const sv, const NV num)
1742 PERL_ARGS_ASSERT_SV_SETNV;
1744 SV_CHECK_THINKFIRST_COW_DROP(sv);
1745 switch (SvTYPE(sv)) {
1748 sv_upgrade(sv, SVt_NV);
1752 sv_upgrade(sv, SVt_PVNV);
1756 if (!isGV_with_GP(sv))
1764 /* diag_listed_as: Can't coerce %s to %s in %s */
1765 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
1767 NOT_REACHED; /* NOTREACHED */
1772 (void)SvNOK_only(sv); /* validate number */
1777 =for apidoc sv_setnv_mg
1779 Like C<sv_setnv>, but also handles 'set' magic.
1785 Perl_sv_setnv_mg(pTHX_ SV *const sv, const NV num)
1787 PERL_ARGS_ASSERT_SV_SETNV_MG;
1793 /* Return a cleaned-up, printable version of sv, for non-numeric, or
1794 * not incrementable warning display.
1795 * Originally part of S_not_a_number().
1796 * The return value may be != tmpbuf.
1800 S_sv_display(pTHX_ SV *const sv, char *tmpbuf, STRLEN tmpbuf_size) {
1803 PERL_ARGS_ASSERT_SV_DISPLAY;
1806 SV *dsv = newSVpvs_flags("", SVs_TEMP);
1807 pv = sv_uni_display(dsv, sv, 32, UNI_DISPLAY_ISPRINT);
1810 const char * const limit = tmpbuf + tmpbuf_size - 8;
1811 /* each *s can expand to 4 chars + "...\0",
1812 i.e. need room for 8 chars */
1814 const char *s = SvPVX_const(sv);
1815 const char * const end = s + SvCUR(sv);
1816 for ( ; s < end && d < limit; s++ ) {
1818 if (! isASCII(ch) && !isPRINT_LC(ch)) {
1822 /* Map to ASCII "equivalent" of Latin1 */
1823 ch = LATIN1_TO_NATIVE(NATIVE_TO_LATIN1(ch) & 127);
1829 else if (ch == '\r') {
1833 else if (ch == '\f') {
1837 else if (ch == '\\') {
1841 else if (ch == '\0') {
1845 else if (isPRINT_LC(ch))
1864 /* Print an "isn't numeric" warning, using a cleaned-up,
1865 * printable version of the offending string
1869 S_not_a_number(pTHX_ SV *const sv)
1874 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1876 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1879 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1880 /* diag_listed_as: Argument "%s" isn't numeric%s */
1881 "Argument \"%s\" isn't numeric in %s", pv,
1884 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1885 /* diag_listed_as: Argument "%s" isn't numeric%s */
1886 "Argument \"%s\" isn't numeric", pv);
1890 S_not_incrementable(pTHX_ SV *const sv) {
1894 PERL_ARGS_ASSERT_NOT_INCREMENTABLE;
1896 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1898 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1899 "Argument \"%s\" treated as 0 in increment (++)", pv);
1903 =for apidoc looks_like_number
1905 Test if the content of an SV looks like a number (or is a number).
1906 C<Inf> and C<Infinity> are treated as numbers (so will not issue a
1907 non-numeric warning), even if your C<atof()> doesn't grok them. Get-magic is
1914 Perl_looks_like_number(pTHX_ SV *const sv)
1920 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1922 if (SvPOK(sv) || SvPOKp(sv)) {
1923 sbegin = SvPV_nomg_const(sv, len);
1926 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
1927 numtype = grok_number(sbegin, len, NULL);
1928 return ((numtype & IS_NUMBER_TRAILING)) ? 0 : numtype;
1932 S_glob_2number(pTHX_ GV * const gv)
1934 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1936 /* We know that all GVs stringify to something that is not-a-number,
1937 so no need to test that. */
1938 if (ckWARN(WARN_NUMERIC))
1940 SV *const buffer = sv_newmortal();
1941 gv_efullname3(buffer, gv, "*");
1942 not_a_number(buffer);
1944 /* We just want something true to return, so that S_sv_2iuv_common
1945 can tail call us and return true. */
1949 /* Actually, ISO C leaves conversion of UV to IV undefined, but
1950 until proven guilty, assume that things are not that bad... */
1955 As 64 bit platforms often have an NV that doesn't preserve all bits of
1956 an IV (an assumption perl has been based on to date) it becomes necessary
1957 to remove the assumption that the NV always carries enough precision to
1958 recreate the IV whenever needed, and that the NV is the canonical form.
1959 Instead, IV/UV and NV need to be given equal rights. So as to not lose
1960 precision as a side effect of conversion (which would lead to insanity
1961 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
1962 1) to distinguish between IV/UV/NV slots that have a valid conversion cached
1963 where precision was lost, and IV/UV/NV slots that have a valid conversion
1964 which has lost no precision
1965 2) to ensure that if a numeric conversion to one form is requested that
1966 would lose precision, the precise conversion (or differently
1967 imprecise conversion) is also performed and cached, to prevent
1968 requests for different numeric formats on the same SV causing
1969 lossy conversion chains. (lossless conversion chains are perfectly
1974 SvIOKp is true if the IV slot contains a valid value
1975 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1976 SvNOKp is true if the NV slot contains a valid value
1977 SvNOK is true only if the NV value is accurate
1980 while converting from PV to NV, check to see if converting that NV to an
1981 IV(or UV) would lose accuracy over a direct conversion from PV to
1982 IV(or UV). If it would, cache both conversions, return NV, but mark
1983 SV as IOK NOKp (ie not NOK).
1985 While converting from PV to IV, check to see if converting that IV to an
1986 NV would lose accuracy over a direct conversion from PV to NV. If it
1987 would, cache both conversions, flag similarly.
1989 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
1990 correctly because if IV & NV were set NV *always* overruled.
1991 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
1992 changes - now IV and NV together means that the two are interchangeable:
1993 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
1995 The benefit of this is that operations such as pp_add know that if
1996 SvIOK is true for both left and right operands, then integer addition
1997 can be used instead of floating point (for cases where the result won't
1998 overflow). Before, floating point was always used, which could lead to
1999 loss of precision compared with integer addition.
2001 * making IV and NV equal status should make maths accurate on 64 bit
2003 * may speed up maths somewhat if pp_add and friends start to use
2004 integers when possible instead of fp. (Hopefully the overhead in
2005 looking for SvIOK and checking for overflow will not outweigh the
2006 fp to integer speedup)
2007 * will slow down integer operations (callers of SvIV) on "inaccurate"
2008 values, as the change from SvIOK to SvIOKp will cause a call into
2009 sv_2iv each time rather than a macro access direct to the IV slot
2010 * should speed up number->string conversion on integers as IV is
2011 favoured when IV and NV are equally accurate
2013 ####################################################################
2014 You had better be using SvIOK_notUV if you want an IV for arithmetic:
2015 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
2016 On the other hand, SvUOK is true iff UV.
2017 ####################################################################
2019 Your mileage will vary depending your CPU's relative fp to integer
2023 #ifndef NV_PRESERVES_UV
2024 # define IS_NUMBER_UNDERFLOW_IV 1
2025 # define IS_NUMBER_UNDERFLOW_UV 2
2026 # define IS_NUMBER_IV_AND_UV 2
2027 # define IS_NUMBER_OVERFLOW_IV 4
2028 # define IS_NUMBER_OVERFLOW_UV 5
2030 /* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
2032 /* For sv_2nv these three cases are "SvNOK and don't bother casting" */
2034 S_sv_2iuv_non_preserve(pTHX_ SV *const sv
2040 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
2041 PERL_UNUSED_CONTEXT;
2043 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));
2044 if (SvNVX(sv) < (NV)IV_MIN) {
2045 (void)SvIOKp_on(sv);
2047 SvIV_set(sv, IV_MIN);
2048 return IS_NUMBER_UNDERFLOW_IV;
2050 if (SvNVX(sv) > (NV)UV_MAX) {
2051 (void)SvIOKp_on(sv);
2054 SvUV_set(sv, UV_MAX);
2055 return IS_NUMBER_OVERFLOW_UV;
2057 (void)SvIOKp_on(sv);
2059 /* Can't use strtol etc to convert this string. (See truth table in
2061 if (SvNVX(sv) <= (UV)IV_MAX) {
2062 SvIV_set(sv, I_V(SvNVX(sv)));
2063 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2064 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
2066 /* Integer is imprecise. NOK, IOKp */
2068 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
2071 SvUV_set(sv, U_V(SvNVX(sv)));
2072 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2073 if (SvUVX(sv) == UV_MAX) {
2074 /* As we know that NVs don't preserve UVs, UV_MAX cannot
2075 possibly be preserved by NV. Hence, it must be overflow.
2077 return IS_NUMBER_OVERFLOW_UV;
2079 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
2081 /* Integer is imprecise. NOK, IOKp */
2083 return IS_NUMBER_OVERFLOW_IV;
2085 #endif /* !NV_PRESERVES_UV*/
2087 /* If numtype is infnan, set the NV of the sv accordingly.
2088 * If numtype is anything else, try setting the NV using Atof(PV). */
2090 S_sv_setnv(pTHX_ SV* sv, int numtype)
2092 bool pok = cBOOL(SvPOK(sv));
2095 if ((numtype & IS_NUMBER_INFINITY)) {
2096 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -NV_INF : NV_INF);
2101 if ((numtype & IS_NUMBER_NAN)) {
2102 SvNV_set(sv, NV_NAN);
2107 SvNV_set(sv, Atof(SvPVX_const(sv)));
2108 /* Purposefully no true nok here, since we don't want to blow
2109 * away the possible IOK/UV of an existing sv. */
2112 SvNOK_only(sv); /* No IV or UV please, this is pure infnan. */
2114 SvPOK_on(sv); /* PV is okay, though. */
2119 S_sv_2iuv_common(pTHX_ SV *const sv)
2121 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
2124 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
2125 * without also getting a cached IV/UV from it at the same time
2126 * (ie PV->NV conversion should detect loss of accuracy and cache
2127 * IV or UV at same time to avoid this. */
2128 /* IV-over-UV optimisation - choose to cache IV if possible */
2130 if (SvTYPE(sv) == SVt_NV)
2131 sv_upgrade(sv, SVt_PVNV);
2133 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
2134 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
2135 certainly cast into the IV range at IV_MAX, whereas the correct
2136 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
2138 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2139 if (Perl_isnan(SvNVX(sv))) {
2145 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2146 SvIV_set(sv, I_V(SvNVX(sv)));
2147 if (SvNVX(sv) == (NV) SvIVX(sv)
2148 #ifndef NV_PRESERVES_UV
2149 && SvIVX(sv) != IV_MIN /* avoid negating IV_MIN below */
2150 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2151 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2152 /* Don't flag it as "accurately an integer" if the number
2153 came from a (by definition imprecise) NV operation, and
2154 we're outside the range of NV integer precision */
2158 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2160 /* scalar has trailing garbage, eg "42a" */
2162 DEBUG_c(PerlIO_printf(Perl_debug_log,
2163 "0x%" UVxf " iv(%" NVgf " => %" IVdf ") (precise)\n",
2169 /* IV not precise. No need to convert from PV, as NV
2170 conversion would already have cached IV if it detected
2171 that PV->IV would be better than PV->NV->IV
2172 flags already correct - don't set public IOK. */
2173 DEBUG_c(PerlIO_printf(Perl_debug_log,
2174 "0x%" UVxf " iv(%" NVgf " => %" IVdf ") (imprecise)\n",
2179 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2180 but the cast (NV)IV_MIN rounds to a the value less (more
2181 negative) than IV_MIN which happens to be equal to SvNVX ??
2182 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2183 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2184 (NV)UVX == NVX are both true, but the values differ. :-(
2185 Hopefully for 2s complement IV_MIN is something like
2186 0x8000000000000000 which will be exact. NWC */
2189 SvUV_set(sv, U_V(SvNVX(sv)));
2191 (SvNVX(sv) == (NV) SvUVX(sv))
2192 #ifndef NV_PRESERVES_UV
2193 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2194 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2195 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2196 /* Don't flag it as "accurately an integer" if the number
2197 came from a (by definition imprecise) NV operation, and
2198 we're outside the range of NV integer precision */
2204 DEBUG_c(PerlIO_printf(Perl_debug_log,
2205 "0x%" UVxf " 2iv(%" UVuf " => %" IVdf ") (as unsigned)\n",
2211 else if (SvPOKp(sv)) {
2214 const char *s = SvPVX_const(sv);
2215 const STRLEN cur = SvCUR(sv);
2217 /* short-cut for a single digit string like "1" */
2222 if (SvTYPE(sv) < SVt_PVIV)
2223 sv_upgrade(sv, SVt_PVIV);
2225 SvIV_set(sv, (IV)(c - '0'));
2230 numtype = grok_number(s, cur, &value);
2231 /* We want to avoid a possible problem when we cache an IV/ a UV which
2232 may be later translated to an NV, and the resulting NV is not
2233 the same as the direct translation of the initial string
2234 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2235 be careful to ensure that the value with the .456 is around if the
2236 NV value is requested in the future).
2238 This means that if we cache such an IV/a UV, we need to cache the
2239 NV as well. Moreover, we trade speed for space, and do not
2240 cache the NV if we are sure it's not needed.
2243 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2244 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2245 == IS_NUMBER_IN_UV) {
2246 /* It's definitely an integer, only upgrade to PVIV */
2247 if (SvTYPE(sv) < SVt_PVIV)
2248 sv_upgrade(sv, SVt_PVIV);
2250 } else if (SvTYPE(sv) < SVt_PVNV)
2251 sv_upgrade(sv, SVt_PVNV);
2253 if ((numtype & (IS_NUMBER_INFINITY | IS_NUMBER_NAN))) {
2254 if (ckWARN(WARN_NUMERIC) && ((numtype & IS_NUMBER_TRAILING)))
2256 S_sv_setnv(aTHX_ sv, numtype);
2260 /* If NVs preserve UVs then we only use the UV value if we know that
2261 we aren't going to call atof() below. If NVs don't preserve UVs
2262 then the value returned may have more precision than atof() will
2263 return, even though value isn't perfectly accurate. */
2264 if ((numtype & (IS_NUMBER_IN_UV
2265 #ifdef NV_PRESERVES_UV
2268 )) == IS_NUMBER_IN_UV) {
2269 /* This won't turn off the public IOK flag if it was set above */
2270 (void)SvIOKp_on(sv);
2272 if (!(numtype & IS_NUMBER_NEG)) {
2274 if (value <= (UV)IV_MAX) {
2275 SvIV_set(sv, (IV)value);
2277 /* it didn't overflow, and it was positive. */
2278 SvUV_set(sv, value);
2282 /* 2s complement assumption */
2283 if (value <= (UV)IV_MIN) {
2284 SvIV_set(sv, value == (UV)IV_MIN
2285 ? IV_MIN : -(IV)value);
2287 /* Too negative for an IV. This is a double upgrade, but
2288 I'm assuming it will be rare. */
2289 if (SvTYPE(sv) < SVt_PVNV)
2290 sv_upgrade(sv, SVt_PVNV);
2294 SvNV_set(sv, -(NV)value);
2295 SvIV_set(sv, IV_MIN);
2299 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2300 will be in the previous block to set the IV slot, and the next
2301 block to set the NV slot. So no else here. */
2303 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2304 != IS_NUMBER_IN_UV) {
2305 /* It wasn't an (integer that doesn't overflow the UV). */
2306 S_sv_setnv(aTHX_ sv, numtype);
2308 if (! numtype && ckWARN(WARN_NUMERIC))
2311 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2iv(%" NVgf ")\n",
2312 PTR2UV(sv), SvNVX(sv)));
2314 #ifdef NV_PRESERVES_UV
2315 (void)SvIOKp_on(sv);
2317 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2318 if (Perl_isnan(SvNVX(sv))) {
2324 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2325 SvIV_set(sv, I_V(SvNVX(sv)));
2326 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2329 NOOP; /* Integer is imprecise. NOK, IOKp */
2331 /* UV will not work better than IV */
2333 if (SvNVX(sv) > (NV)UV_MAX) {
2335 /* Integer is inaccurate. NOK, IOKp, is UV */
2336 SvUV_set(sv, UV_MAX);
2338 SvUV_set(sv, U_V(SvNVX(sv)));
2339 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2340 NV preservse UV so can do correct comparison. */
2341 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2344 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
2349 #else /* NV_PRESERVES_UV */
2350 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2351 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
2352 /* The IV/UV slot will have been set from value returned by
2353 grok_number above. The NV slot has just been set using
2356 assert (SvIOKp(sv));
2358 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2359 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2360 /* Small enough to preserve all bits. */
2361 (void)SvIOKp_on(sv);
2363 SvIV_set(sv, I_V(SvNVX(sv)));
2364 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2366 /* Assumption: first non-preserved integer is < IV_MAX,
2367 this NV is in the preserved range, therefore: */
2368 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2370 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);
2374 0 0 already failed to read UV.
2375 0 1 already failed to read UV.
2376 1 0 you won't get here in this case. IV/UV
2377 slot set, public IOK, Atof() unneeded.
2378 1 1 already read UV.
2379 so there's no point in sv_2iuv_non_preserve() attempting
2380 to use atol, strtol, strtoul etc. */
2382 sv_2iuv_non_preserve (sv, numtype);
2384 sv_2iuv_non_preserve (sv);
2388 #endif /* NV_PRESERVES_UV */
2389 /* It might be more code efficient to go through the entire logic above
2390 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2391 gets complex and potentially buggy, so more programmer efficient
2392 to do it this way, by turning off the public flags: */
2394 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2398 if (isGV_with_GP(sv))
2399 return glob_2number(MUTABLE_GV(sv));
2401 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2403 if (SvTYPE(sv) < SVt_IV)
2404 /* Typically the caller expects that sv_any is not NULL now. */
2405 sv_upgrade(sv, SVt_IV);
2406 /* Return 0 from the caller. */
2413 =for apidoc sv_2iv_flags
2415 Return the integer value of an SV, doing any necessary string
2416 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2417 Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2423 Perl_sv_2iv_flags(pTHX_ SV *const sv, const I32 flags)
2425 PERL_ARGS_ASSERT_SV_2IV_FLAGS;
2427 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2428 && SvTYPE(sv) != SVt_PVFM);
2430 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2436 if (flags & SV_SKIP_OVERLOAD)
2438 tmpstr = AMG_CALLunary(sv, numer_amg);
2439 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2440 return SvIV(tmpstr);
2443 return PTR2IV(SvRV(sv));
2446 if (SvVALID(sv) || isREGEXP(sv)) {
2447 /* FBMs use the space for SvIVX and SvNVX for other purposes, so
2448 must not let them cache IVs.
2449 In practice they are extremely unlikely to actually get anywhere
2450 accessible by user Perl code - the only way that I'm aware of is when
2451 a constant subroutine which is used as the second argument to index.
2453 Regexps have no SvIVX and SvNVX fields.
2458 const char * const ptr =
2459 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2461 = grok_number(ptr, SvCUR(sv), &value);
2463 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2464 == IS_NUMBER_IN_UV) {
2465 /* It's definitely an integer */
2466 if (numtype & IS_NUMBER_NEG) {
2467 if (value < (UV)IV_MIN)
2470 if (value < (UV)IV_MAX)
2475 /* Quite wrong but no good choices. */
2476 if ((numtype & IS_NUMBER_INFINITY)) {
2477 return (numtype & IS_NUMBER_NEG) ? IV_MIN : IV_MAX;
2478 } else if ((numtype & IS_NUMBER_NAN)) {
2479 return 0; /* So wrong. */
2483 if (ckWARN(WARN_NUMERIC))
2486 return I_V(Atof(ptr));
2490 if (SvTHINKFIRST(sv)) {
2491 if (SvREADONLY(sv) && !SvOK(sv)) {
2492 if (ckWARN(WARN_UNINITIALIZED))
2499 if (S_sv_2iuv_common(aTHX_ sv))
2503 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2iv(%" IVdf ")\n",
2504 PTR2UV(sv),SvIVX(sv)));
2505 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
2509 =for apidoc sv_2uv_flags
2511 Return the unsigned integer value of an SV, doing any necessary string
2512 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2513 Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
2515 =for apidoc Amnh||SV_GMAGIC
2521 Perl_sv_2uv_flags(pTHX_ SV *const sv, const I32 flags)
2523 PERL_ARGS_ASSERT_SV_2UV_FLAGS;
2525 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2531 if (flags & SV_SKIP_OVERLOAD)
2533 tmpstr = AMG_CALLunary(sv, numer_amg);
2534 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2535 return SvUV(tmpstr);
2538 return PTR2UV(SvRV(sv));
2541 if (SvVALID(sv) || isREGEXP(sv)) {
2542 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2543 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2544 Regexps have no SvIVX and SvNVX fields. */
2548 const char * const ptr =
2549 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2551 = grok_number(ptr, SvCUR(sv), &value);
2553 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2554 == IS_NUMBER_IN_UV) {
2555 /* It's definitely an integer */
2556 if (!(numtype & IS_NUMBER_NEG))
2560 /* Quite wrong but no good choices. */
2561 if ((numtype & IS_NUMBER_INFINITY)) {
2562 return UV_MAX; /* So wrong. */
2563 } else if ((numtype & IS_NUMBER_NAN)) {
2564 return 0; /* So wrong. */
2568 if (ckWARN(WARN_NUMERIC))
2571 return U_V(Atof(ptr));
2575 if (SvTHINKFIRST(sv)) {
2576 if (SvREADONLY(sv) && !SvOK(sv)) {
2577 if (ckWARN(WARN_UNINITIALIZED))
2584 if (S_sv_2iuv_common(aTHX_ sv))
2588 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2uv(%" UVuf ")\n",
2589 PTR2UV(sv),SvUVX(sv)));
2590 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
2594 =for apidoc sv_2nv_flags
2596 Return the num value of an SV, doing any necessary string or integer
2597 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2598 Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
2604 Perl_sv_2nv_flags(pTHX_ SV *const sv, const I32 flags)
2606 PERL_ARGS_ASSERT_SV_2NV_FLAGS;
2608 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2609 && SvTYPE(sv) != SVt_PVFM);
2610 if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) {
2611 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2612 the same flag bit as SVf_IVisUV, so must not let them cache NVs.
2613 Regexps have no SvIVX and SvNVX fields. */
2615 if (flags & SV_GMAGIC)
2619 if (SvPOKp(sv) && !SvIOKp(sv)) {
2620 ptr = SvPVX_const(sv);
2621 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
2622 !grok_number(ptr, SvCUR(sv), NULL))
2628 return (NV)SvUVX(sv);
2630 return (NV)SvIVX(sv);
2635 assert(SvTYPE(sv) >= SVt_PVMG);
2636 /* This falls through to the report_uninit near the end of the
2638 } else if (SvTHINKFIRST(sv)) {
2643 if (flags & SV_SKIP_OVERLOAD)
2645 tmpstr = AMG_CALLunary(sv, numer_amg);
2646 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2647 return SvNV(tmpstr);
2650 return PTR2NV(SvRV(sv));
2652 if (SvREADONLY(sv) && !SvOK(sv)) {
2653 if (ckWARN(WARN_UNINITIALIZED))
2658 if (SvTYPE(sv) < SVt_NV) {
2659 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2660 sv_upgrade(sv, SVt_NV);
2661 CLANG_DIAG_IGNORE_STMT(-Wthread-safety);
2663 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2664 STORE_LC_NUMERIC_SET_STANDARD();
2665 PerlIO_printf(Perl_debug_log,
2666 "0x%" UVxf " num(%" NVgf ")\n",
2667 PTR2UV(sv), SvNVX(sv));
2668 RESTORE_LC_NUMERIC();
2670 CLANG_DIAG_RESTORE_STMT;
2673 else if (SvTYPE(sv) < SVt_PVNV)
2674 sv_upgrade(sv, SVt_PVNV);
2679 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
2680 #ifdef NV_PRESERVES_UV
2686 /* Only set the public NV OK flag if this NV preserves the IV */
2687 /* Check it's not 0xFFFFFFFFFFFFFFFF */
2689 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
2690 : (SvIVX(sv) == I_V(SvNVX(sv))))
2696 else if (SvPOKp(sv)) {
2698 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2699 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
2701 #ifdef NV_PRESERVES_UV
2702 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2703 == IS_NUMBER_IN_UV) {
2704 /* It's definitely an integer */
2705 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
2707 S_sv_setnv(aTHX_ sv, numtype);
2714 SvNV_set(sv, Atof(SvPVX_const(sv)));
2715 /* Only set the public NV OK flag if this NV preserves the value in
2716 the PV at least as well as an IV/UV would.
2717 Not sure how to do this 100% reliably. */
2718 /* if that shift count is out of range then Configure's test is
2719 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2721 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2722 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2723 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
2724 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2725 /* Can't use strtol etc to convert this string, so don't try.
2726 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2729 /* value has been set. It may not be precise. */
2730 if ((numtype & IS_NUMBER_NEG) && (value >= (UV)IV_MIN)) {
2731 /* 2s complement assumption for (UV)IV_MIN */
2732 SvNOK_on(sv); /* Integer is too negative. */
2737 if (numtype & IS_NUMBER_NEG) {
2738 /* -IV_MIN is undefined, but we should never reach
2739 * this point with both IS_NUMBER_NEG and value ==
2741 assert(value != (UV)IV_MIN);
2742 SvIV_set(sv, -(IV)value);
2743 } else if (value <= (UV)IV_MAX) {
2744 SvIV_set(sv, (IV)value);
2746 SvUV_set(sv, value);
2750 if (numtype & IS_NUMBER_NOT_INT) {
2751 /* I believe that even if the original PV had decimals,
2752 they are lost beyond the limit of the FP precision.
2753 However, neither is canonical, so both only get p
2754 flags. NWC, 2000/11/25 */
2755 /* Both already have p flags, so do nothing */
2757 const NV nv = SvNVX(sv);
2758 /* XXX should this spot have NAN_COMPARE_BROKEN, too? */
2759 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2760 if (SvIVX(sv) == I_V(nv)) {
2763 /* It had no "." so it must be integer. */
2767 /* between IV_MAX and NV(UV_MAX).
2768 Could be slightly > UV_MAX */
2770 if (numtype & IS_NUMBER_NOT_INT) {
2771 /* UV and NV both imprecise. */
2773 const UV nv_as_uv = U_V(nv);
2775 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2784 /* It might be more code efficient to go through the entire logic above
2785 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2786 gets complex and potentially buggy, so more programmer efficient
2787 to do it this way, by turning off the public flags: */
2789 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2790 #endif /* NV_PRESERVES_UV */
2793 if (isGV_with_GP(sv)) {
2794 glob_2number(MUTABLE_GV(sv));
2798 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2800 assert (SvTYPE(sv) >= SVt_NV);
2801 /* Typically the caller expects that sv_any is not NULL now. */
2802 /* XXX Ilya implies that this is a bug in callers that assume this
2803 and ideally should be fixed. */
2806 CLANG_DIAG_IGNORE_STMT(-Wthread-safety);
2808 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2809 STORE_LC_NUMERIC_SET_STANDARD();
2810 PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2nv(%" NVgf ")\n",
2811 PTR2UV(sv), SvNVX(sv));
2812 RESTORE_LC_NUMERIC();
2814 CLANG_DIAG_RESTORE_STMT;
2821 Return an SV with the numeric value of the source SV, doing any necessary
2822 reference or overload conversion. The caller is expected to have handled
2829 Perl_sv_2num(pTHX_ SV *const sv)
2831 PERL_ARGS_ASSERT_SV_2NUM;
2836 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
2837 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
2838 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2839 return sv_2num(tmpsv);
2841 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2844 /* int2str_table: lookup table containing string representations of all
2845 * two digit numbers. For example, int2str_table.arr[0] is "00" and
2846 * int2str_table.arr[12*2] is "12".
2848 * We are going to read two bytes at a time, so we have to ensure that
2849 * the array is aligned to a 2 byte boundary. That's why it was made a
2850 * union with a dummy U16 member. */
2851 static const union {
2854 } int2str_table = {{
2855 '0', '0', '0', '1', '0', '2', '0', '3', '0', '4', '0', '5', '0', '6',
2856 '0', '7', '0', '8', '0', '9', '1', '0', '1', '1', '1', '2', '1', '3',
2857 '1', '4', '1', '5', '1', '6', '1', '7', '1', '8', '1', '9', '2', '0',
2858 '2', '1', '2', '2', '2', '3', '2', '4', '2', '5', '2', '6', '2', '7',
2859 '2', '8', '2', '9', '3', '0', '3', '1', '3', '2', '3', '3', '3', '4',
2860 '3', '5', '3', '6', '3', '7', '3', '8', '3', '9', '4', '0', '4', '1',
2861 '4', '2', '4', '3', '4', '4', '4', '5', '4', '6', '4', '7', '4', '8',
2862 '4', '9', '5', '0', '5', '1', '5', '2', '5', '3', '5', '4', '5', '5',
2863 '5', '6', '5', '7', '5', '8', '5', '9', '6', '0', '6', '1', '6', '2',
2864 '6', '3', '6', '4', '6', '5', '6', '6', '6', '7', '6', '8', '6', '9',
2865 '7', '0', '7', '1', '7', '2', '7', '3', '7', '4', '7', '5', '7', '6',
2866 '7', '7', '7', '8', '7', '9', '8', '0', '8', '1', '8', '2', '8', '3',
2867 '8', '4', '8', '5', '8', '6', '8', '7', '8', '8', '8', '9', '9', '0',
2868 '9', '1', '9', '2', '9', '3', '9', '4', '9', '5', '9', '6', '9', '7',
2872 /* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2873 * UV as a string towards the end of buf, and return pointers to start and
2876 * We assume that buf is at least TYPE_CHARS(UV) long.
2879 PERL_STATIC_INLINE char *
2880 S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
2882 char *ptr = buf + TYPE_CHARS(UV);
2883 char * const ebuf = ptr;
2885 U16 *word_ptr, *word_table;
2887 PERL_ARGS_ASSERT_UIV_2BUF;
2889 /* ptr has to be properly aligned, because we will cast it to U16* */
2890 assert(PTR2nat(ptr) % 2 == 0);
2891 /* we are going to read/write two bytes at a time */
2892 word_ptr = (U16*)ptr;
2893 word_table = (U16*)int2str_table.arr;
2895 if (UNLIKELY(is_uv))
2901 /* Using 0- here to silence bogus warning from MS VC */
2902 uv = (UV) (0 - (UV) iv);
2907 *--word_ptr = word_table[uv % 100];
2910 ptr = (char*)word_ptr;
2913 *--ptr = (char)uv + '0';
2915 *--word_ptr = word_table[uv];
2916 ptr = (char*)word_ptr;
2926 /* Helper for sv_2pv_flags and sv_vcatpvfn_flags. If the NV is an
2927 * infinity or a not-a-number, writes the appropriate strings to the
2928 * buffer, including a zero byte. On success returns the written length,
2929 * excluding the zero byte, on failure (not an infinity, not a nan)
2930 * returns zero, assert-fails on maxlen being too short.
2932 * XXX for "Inf", "-Inf", and "NaN", we could have three read-only
2933 * shared string constants we point to, instead of generating a new
2934 * string for each instance. */
2936 S_infnan_2pv(NV nv, char* buffer, size_t maxlen, char plus) {
2938 assert(maxlen >= 4);
2939 if (Perl_isinf(nv)) {
2941 if (maxlen < 5) /* "-Inf\0" */
2951 else if (Perl_isnan(nv)) {
2955 /* XXX optionally output the payload mantissa bits as
2956 * "(unsigned)" (to match the nan("...") C99 function,
2957 * or maybe as "(0xhhh...)" would make more sense...
2958 * provide a format string so that the user can decide?
2959 * NOTE: would affect the maxlen and assert() logic.*/
2964 assert((s == buffer + 3) || (s == buffer + 4));
2970 =for apidoc sv_2pv_flags
2972 Returns a pointer to the string value of an SV, and sets C<*lp> to its length.
2973 If flags has the C<SV_GMAGIC> bit set, does an C<mg_get()> first. Coerces C<sv> to a
2974 string if necessary. Normally invoked via the C<SvPV_flags> macro.
2975 C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
2981 Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
2985 PERL_ARGS_ASSERT_SV_2PV_FLAGS;
2987 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2988 && SvTYPE(sv) != SVt_PVFM);
2989 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2994 if (flags & SV_SKIP_OVERLOAD)
2996 tmpstr = AMG_CALLunary(sv, string_amg);
2997 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
2998 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
3000 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
3004 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
3005 if (flags & SV_CONST_RETURN) {
3006 pv = (char *) SvPVX_const(tmpstr);
3008 pv = (flags & SV_MUTABLE_RETURN)
3009 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
3012 *lp = SvCUR(tmpstr);
3014 pv = sv_2pv_flags(tmpstr, lp, flags);
3027 SV *const referent = SvRV(sv);
3031 retval = buffer = savepvn("NULLREF", len);
3032 } else if (SvTYPE(referent) == SVt_REGEXP &&
3033 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
3034 amagic_is_enabled(string_amg))) {
3035 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
3039 /* If the regex is UTF-8 we want the containing scalar to
3040 have an UTF-8 flag too */
3047 *lp = RX_WRAPLEN(re);
3049 return RX_WRAPPED(re);
3051 const char *const typestr = sv_reftype(referent, 0);
3052 const STRLEN typelen = strlen(typestr);
3053 UV addr = PTR2UV(referent);
3054 const char *stashname = NULL;
3055 STRLEN stashnamelen = 0; /* hush, gcc */
3056 const char *buffer_end;
3058 if (SvOBJECT(referent)) {
3059 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
3062 stashname = HEK_KEY(name);
3063 stashnamelen = HEK_LEN(name);
3065 if (HEK_UTF8(name)) {
3071 stashname = "__ANON__";
3074 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
3075 + 2 * sizeof(UV) + 2 /* )\0 */;
3077 len = typelen + 3 /* (0x */
3078 + 2 * sizeof(UV) + 2 /* )\0 */;
3081 Newx(buffer, len, char);
3082 buffer_end = retval = buffer + len;
3084 /* Working backwards */
3088 *--retval = PL_hexdigit[addr & 15];
3089 } while (addr >>= 4);
3095 memcpy(retval, typestr, typelen);
3099 retval -= stashnamelen;
3100 memcpy(retval, stashname, stashnamelen);
3102 /* retval may not necessarily have reached the start of the
3104 assert (retval >= buffer);
3106 len = buffer_end - retval - 1; /* -1 for that \0 */
3118 if (flags & SV_MUTABLE_RETURN)
3119 return SvPVX_mutable(sv);
3120 if (flags & SV_CONST_RETURN)
3121 return (char *)SvPVX_const(sv);
3126 /* I'm assuming that if both IV and NV are equally valid then
3127 converting the IV is going to be more efficient */
3128 const U32 isUIOK = SvIsUV(sv);
3129 /* The purpose of this union is to ensure that arr is aligned on
3130 a 2 byte boundary, because that is what uiv_2buf() requires */
3132 char arr[TYPE_CHARS(UV)];
3138 if (SvTYPE(sv) < SVt_PVIV)
3139 sv_upgrade(sv, SVt_PVIV);
3140 ptr = uiv_2buf(buf.arr, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
3142 /* inlined from sv_setpvn */
3143 s = SvGROW_mutable(sv, len + 1);
3144 Move(ptr, s, len, char);
3149 else if (SvNOK(sv)) {
3150 if (SvTYPE(sv) < SVt_PVNV)
3151 sv_upgrade(sv, SVt_PVNV);
3152 if (SvNVX(sv) == 0.0
3153 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
3154 && !Perl_isnan(SvNVX(sv))
3157 s = SvGROW_mutable(sv, 2);
3162 STRLEN size = 5; /* "-Inf\0" */
3164 s = SvGROW_mutable(sv, size);
3165 len = S_infnan_2pv(SvNVX(sv), s, size, 0);
3171 /* some Xenix systems wipe out errno here */
3180 5 + /* exponent digits */
3184 s = SvGROW_mutable(sv, size);
3185 #ifndef USE_LOCALE_NUMERIC
3186 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3192 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
3193 STORE_LC_NUMERIC_SET_TO_NEEDED();
3195 local_radix = _NOT_IN_NUMERIC_STANDARD;
3196 if (local_radix && SvCUR(PL_numeric_radix_sv) > 1) {
3197 size += SvCUR(PL_numeric_radix_sv) - 1;
3198 s = SvGROW_mutable(sv, size);
3201 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3203 /* If the radix character is UTF-8, and actually is in the
3204 * output, turn on the UTF-8 flag for the scalar */
3206 && SvUTF8(PL_numeric_radix_sv)
3207 && instr(s, SvPVX_const(PL_numeric_radix_sv)))
3212 RESTORE_LC_NUMERIC();
3215 /* We don't call SvPOK_on(), because it may come to
3216 * pass that the locale changes so that the
3217 * stringification we just did is no longer correct. We
3218 * will have to re-stringify every time it is needed */
3225 else if (isGV_with_GP(sv)) {
3226 GV *const gv = MUTABLE_GV(sv);
3227 SV *const buffer = sv_newmortal();
3229 gv_efullname3(buffer, gv, "*");
3231 assert(SvPOK(buffer));
3237 *lp = SvCUR(buffer);
3238 return SvPVX(buffer);
3243 if (flags & SV_UNDEF_RETURNS_NULL)
3245 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
3247 /* Typically the caller expects that sv_any is not NULL now. */
3248 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
3249 sv_upgrade(sv, SVt_PV);
3254 const STRLEN len = s - SvPVX_const(sv);
3259 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2pv(%s)\n",
3260 PTR2UV(sv),SvPVX_const(sv)));
3261 if (flags & SV_CONST_RETURN)
3262 return (char *)SvPVX_const(sv);
3263 if (flags & SV_MUTABLE_RETURN)
3264 return SvPVX_mutable(sv);
3269 =for apidoc sv_copypv
3271 Copies a stringified representation of the source SV into the
3272 destination SV. Automatically performs any necessary C<mg_get> and
3273 coercion of numeric values into strings. Guaranteed to preserve
3274 C<UTF8> flag even from overloaded objects. Similar in nature to
3275 C<sv_2pv[_flags]> but operates directly on an SV instead of just the
3276 string. Mostly uses C<sv_2pv_flags> to do its work, except when that
3277 would lose the UTF-8'ness of the PV.
3279 =for apidoc sv_copypv_nomg
3281 Like C<sv_copypv>, but doesn't invoke get magic first.
3283 =for apidoc sv_copypv_flags
3285 Implementation of C<sv_copypv> and C<sv_copypv_nomg>. Calls get magic iff flags
3286 has the C<SV_GMAGIC> bit set.
3292 Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
3297 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
3299 s = SvPV_flags_const(ssv,len,(flags & SV_GMAGIC));
3300 sv_setpvn(dsv,s,len);
3308 =for apidoc sv_2pvbyte
3310 Return a pointer to the byte-encoded representation of the SV, and set C<*lp>
3311 to its length. May cause the SV to be downgraded from UTF-8 as a
3314 Usually accessed via the C<SvPVbyte> macro.
3320 Perl_sv_2pvbyte_flags(pTHX_ SV *sv, STRLEN *const lp, const U32 flags)
3322 PERL_ARGS_ASSERT_SV_2PVBYTE_FLAGS;
3324 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
3326 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3327 || isGV_with_GP(sv) || SvROK(sv)) {
3328 SV *sv2 = sv_newmortal();
3329 sv_copypv_nomg(sv2,sv);
3332 sv_utf8_downgrade_nomg(sv,0);
3333 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3337 =for apidoc sv_2pvutf8
3339 Return a pointer to the UTF-8-encoded representation of the SV, and set C<*lp>
3340 to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3342 Usually accessed via the C<SvPVutf8> macro.
3348 Perl_sv_2pvutf8_flags(pTHX_ SV *sv, STRLEN *const lp, const U32 flags)
3350 PERL_ARGS_ASSERT_SV_2PVUTF8_FLAGS;
3352 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
3354 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3355 || isGV_with_GP(sv) || SvROK(sv)) {
3356 SV *sv2 = sv_newmortal();
3357 sv_copypv_nomg(sv2,sv);
3360 sv_utf8_upgrade_nomg(sv);
3361 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3366 =for apidoc sv_2bool
3368 This macro is only used by C<sv_true()> or its macro equivalent, and only if
3369 the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>.
3370 It calls C<sv_2bool_flags> with the C<SV_GMAGIC> flag.
3372 =for apidoc sv_2bool_flags
3374 This function is only used by C<sv_true()> and friends, and only if
3375 the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>. If the flags
3376 contain C<SV_GMAGIC>, then it does an C<mg_get()> first.
3383 Perl_sv_2bool_flags(pTHX_ SV *sv, I32 flags)
3385 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
3388 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
3394 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
3395 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv)))) {
3398 if(SvGMAGICAL(sv)) {
3400 goto restart; /* call sv_2bool */
3402 /* expanded SvTRUE_common(sv, (flags = 0, goto restart)) */
3403 else if(!SvOK(sv)) {
3406 else if(SvPOK(sv)) {
3407 svb = SvPVXtrue(sv);
3409 else if((SvFLAGS(sv) & (SVf_IOK|SVf_NOK))) {
3410 svb = (SvIOK(sv) && SvIVX(sv) != 0)
3411 || (SvNOK(sv) && SvNVX(sv) != 0.0);
3415 goto restart; /* call sv_2bool_nomg */
3425 RX_WRAPLEN(sv) > 1 || (RX_WRAPLEN(sv) && *RX_WRAPPED(sv) != '0');
3427 if (SvNOK(sv) && !SvPOK(sv))
3428 return SvNVX(sv) != 0.0;
3430 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
3434 =for apidoc sv_utf8_upgrade
3436 Converts the PV of an SV to its UTF-8-encoded form.
3437 Forces the SV to string form if it is not already.
3438 Will C<mg_get> on C<sv> if appropriate.
3439 Always sets the C<SvUTF8> flag to avoid future validity checks even
3440 if the whole string is the same in UTF-8 as not.
3441 Returns the number of bytes in the converted string
3443 This is not a general purpose byte encoding to Unicode interface:
3444 use the Encode extension for that.
3446 =for apidoc sv_utf8_upgrade_nomg
3448 Like C<sv_utf8_upgrade>, but doesn't do magic on C<sv>.
3450 =for apidoc sv_utf8_upgrade_flags
3452 Converts the PV of an SV to its UTF-8-encoded form.
3453 Forces the SV to string form if it is not already.
3454 Always sets the SvUTF8 flag to avoid future validity checks even
3455 if all the bytes are invariant in UTF-8.
3456 If C<flags> has C<SV_GMAGIC> bit set,
3457 will C<mg_get> on C<sv> if appropriate, else not.
3459 The C<SV_FORCE_UTF8_UPGRADE> flag is now ignored.
3461 Returns the number of bytes in the converted string.
3463 This is not a general purpose byte encoding to Unicode interface:
3464 use the Encode extension for that.
3466 =for apidoc sv_utf8_upgrade_flags_grow
3468 Like C<sv_utf8_upgrade_flags>, but has an additional parameter C<extra>, which is
3469 the number of unused bytes the string of C<sv> is guaranteed to have free after
3470 it upon return. This allows the caller to reserve extra space that it intends
3471 to fill, to avoid extra grows.
3473 C<sv_utf8_upgrade>, C<sv_utf8_upgrade_nomg>, and C<sv_utf8_upgrade_flags>
3474 are implemented in terms of this function.
3476 Returns the number of bytes in the converted string (not including the spares).
3480 If the routine itself changes the string, it adds a trailing C<NUL>. Such a
3481 C<NUL> isn't guaranteed due to having other routines do the work in some input
3482 cases, or if the input is already flagged as being in utf8.
3487 Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV *const sv, const I32 flags, STRLEN extra)
3489 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
3491 if (sv == &PL_sv_undef)
3493 if (!SvPOK_nog(sv)) {
3495 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3496 (void) sv_2pv_flags(sv,&len, flags);
3498 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3502 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
3506 /* SVt_REGEXP's shouldn't be upgraded to UTF8 - they're already
3507 * compiled and individual nodes will remain non-utf8 even if the
3508 * stringified version of the pattern gets upgraded. Whether the
3509 * PVX of a REGEXP should be grown or we should just croak, I don't
3511 if (SvUTF8(sv) || isREGEXP(sv)) {
3512 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3517 S_sv_uncow(aTHX_ sv, 0);
3520 if (SvCUR(sv) == 0) {
3521 if (extra) SvGROW(sv, extra + 1); /* Make sure is room for a trailing
3523 } else { /* Assume Latin-1/EBCDIC */
3524 /* This function could be much more efficient if we
3525 * had a FLAG in SVs to signal if there are any variant
3526 * chars in the PV. Given that there isn't such a flag
3527 * make the loop as fast as possible. */
3528 U8 * s = (U8 *) SvPVX_const(sv);
3531 if (is_utf8_invariant_string_loc(s, SvCUR(sv), (const U8 **) &t)) {
3533 /* utf8 conversion not needed because all are invariants. Mark
3534 * as UTF-8 even if no variant - saves scanning loop */
3536 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3540 /* Here, there is at least one variant (t points to the first one), so
3541 * the string should be converted to utf8. Everything from 's' to
3542 * 't - 1' will occupy only 1 byte each on output.
3544 * Note that the incoming SV may not have a trailing '\0', as certain
3545 * code in pp_formline can send us partially built SVs.
3547 * There are two main ways to convert. One is to create a new string
3548 * and go through the input starting from the beginning, appending each
3549 * converted value onto the new string as we go along. Going this
3550 * route, it's probably best to initially allocate enough space in the
3551 * string rather than possibly running out of space and having to
3552 * reallocate and then copy what we've done so far. Since everything
3553 * from 's' to 't - 1' is invariant, the destination can be initialized
3554 * with these using a fast memory copy. To be sure to allocate enough
3555 * space, one could use the worst case scenario, where every remaining
3556 * byte expands to two under UTF-8, or one could parse it and count
3557 * exactly how many do expand.
3559 * The other way is to unconditionally parse the remainder of the
3560 * string to figure out exactly how big the expanded string will be,
3561 * growing if needed. Then start at the end of the string and place
3562 * the character there at the end of the unfilled space in the expanded
3563 * one, working backwards until reaching 't'.
3565 * The problem with assuming the worst case scenario is that for very
3566 * long strings, we could allocate much more memory than actually
3567 * needed, which can create performance problems. If we have to parse
3568 * anyway, the second method is the winner as it may avoid an extra
3569 * copy. The code used to use the first method under some
3570 * circumstances, but now that there is faster variant counting on
3571 * ASCII platforms, the second method is used exclusively, eliminating
3572 * some code that no longer has to be maintained. */
3575 /* Count the total number of variants there are. We can start
3576 * just beyond the first one, which is known to be at 't' */
3577 const Size_t invariant_length = t - s;
3578 U8 * e = (U8 *) SvEND(sv);
3580 /* The length of the left overs, plus 1. */
3581 const Size_t remaining_length_p1 = e - t;
3583 /* We expand by 1 for the variant at 't' and one for each remaining
3584 * variant (we start looking at 't+1') */
3585 Size_t expansion = 1 + variant_under_utf8_count(t + 1, e);
3587 /* +1 = trailing NUL */
3588 Size_t need = SvCUR(sv) + expansion + extra + 1;
3591 /* Grow if needed */
3592 if (SvLEN(sv) < need) {
3593 t = invariant_length + (U8*) SvGROW(sv, need);
3594 e = t + remaining_length_p1;
3596 SvCUR_set(sv, invariant_length + remaining_length_p1 + expansion);
3598 /* Set the NUL at the end */
3599 d = (U8 *) SvEND(sv);
3602 /* Having decremented d, it points to the position to put the
3603 * very last byte of the expanded string. Go backwards through
3604 * the string, copying and expanding as we go, stopping when we
3605 * get to the part that is invariant the rest of the way down */
3609 if (NATIVE_BYTE_IS_INVARIANT(*e)) {
3612 *d-- = UTF8_EIGHT_BIT_LO(*e);
3613 *d-- = UTF8_EIGHT_BIT_HI(*e);
3618 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3619 /* Update pos. We do it at the end rather than during
3620 * the upgrade, to avoid slowing down the common case
3621 * (upgrade without pos).
3622 * pos can be stored as either bytes or characters. Since
3623 * this was previously a byte string we can just turn off
3624 * the bytes flag. */
3625 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3627 mg->mg_flags &= ~MGf_BYTES;
3629 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3630 magic_setutf8(sv,mg); /* clear UTF8 cache */
3640 =for apidoc sv_utf8_downgrade
3642 Attempts to convert the PV of an SV from characters to bytes.
3643 If the PV contains a character that cannot fit
3644 in a byte, this conversion will fail;
3645 in this case, either returns false or, if C<fail_ok> is not
3648 This is not a general purpose Unicode to byte encoding interface:
3649 use the C<Encode> extension for that.
3651 This function process get magic on C<sv>.
3653 =for apidoc sv_utf8_downgrade_nomg
3655 Like C<sv_utf8_downgrade>, but does not process get magic on C<sv>.
3657 =for apidoc sv_utf8_downgrade_flags
3659 Like C<sv_utf8_downgrade>, but with additional C<flags>.
3660 If C<flags> has C<SV_GMAGIC> bit set, processes get magic on C<sv>.
3666 Perl_sv_utf8_downgrade_flags(pTHX_ SV *const sv, const bool fail_ok, const U32 flags)
3668 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE_FLAGS;
3670 if (SvPOKp(sv) && SvUTF8(sv)) {
3674 U32 mg_flags = flags & SV_GMAGIC;
3677 S_sv_uncow(aTHX_ sv, 0);
3679 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3681 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3682 if (mg && mg->mg_len > 0 && mg->mg_flags & MGf_BYTES) {
3683 mg->mg_len = sv_pos_b2u_flags(sv, mg->mg_len,
3684 mg_flags|SV_CONST_RETURN);
3685 mg_flags = 0; /* sv_pos_b2u does get magic */
3687 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3688 magic_setutf8(sv,mg); /* clear UTF8 cache */
3691 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3693 if (!utf8_to_bytes(s, &len)) {
3698 Perl_croak(aTHX_ "Wide character in %s",
3701 Perl_croak(aTHX_ "Wide character");
3712 =for apidoc sv_utf8_encode
3714 Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3715 flag off so that it looks like octets again.
3721 Perl_sv_utf8_encode(pTHX_ SV *const sv)
3723 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3725 if (SvREADONLY(sv)) {
3726 sv_force_normal_flags(sv, 0);
3728 (void) sv_utf8_upgrade(sv);
3733 =for apidoc sv_utf8_decode
3735 If the PV of the SV is an octet sequence in Perl's extended UTF-8
3736 and contains a multiple-byte character, the C<SvUTF8> flag is turned on
3737 so that it looks like a character. If the PV contains only single-byte
3738 characters, the C<SvUTF8> flag stays off.
3739 Scans PV for validity and returns FALSE if the PV is invalid UTF-8.
3745 Perl_sv_utf8_decode(pTHX_ SV *const sv)
3747 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3750 const U8 *start, *c, *first_variant;
3752 /* The octets may have got themselves encoded - get them back as
3755 if (!sv_utf8_downgrade(sv, TRUE))
3758 /* it is actually just a matter of turning the utf8 flag on, but
3759 * we want to make sure everything inside is valid utf8 first.
3761 c = start = (const U8 *) SvPVX_const(sv);
3762 if (! is_utf8_invariant_string_loc(c, SvCUR(sv), &first_variant)) {
3763 if (!is_utf8_string(first_variant, SvCUR(sv) - (first_variant -c)))
3767 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3768 /* XXX Is this dead code? XS_utf8_decode calls SvSETMAGIC
3769 after this, clearing pos. Does anything on CPAN
3771 /* adjust pos to the start of a UTF8 char sequence */
3772 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3774 I32 pos = mg->mg_len;
3776 for (c = start + pos; c > start; c--) {
3777 if (UTF8_IS_START(*c))
3780 mg->mg_len = c - start;
3783 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3784 magic_setutf8(sv,mg); /* clear UTF8 cache */
3791 =for apidoc sv_setsv
3793 Copies the contents of the source SV C<ssv> into the destination SV
3794 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3795 function if the source SV needs to be reused. Does not handle 'set' magic on
3796 destination SV. Calls 'get' magic on source SV. Loosely speaking, it
3797 performs a copy-by-value, obliterating any previous content of the
3800 You probably want to use one of the assortment of wrappers, such as
3801 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3802 C<SvSetMagicSV_nosteal>.
3804 =for apidoc sv_setsv_flags
3806 Copies the contents of the source SV C<ssv> into the destination SV
3807 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3808 function if the source SV needs to be reused. Does not handle 'set' magic.
3809 Loosely speaking, it performs a copy-by-value, obliterating any previous
3810 content of the destination.
3811 If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
3812 C<ssv> if appropriate, else not. If the C<flags>
3813 parameter has the C<SV_NOSTEAL> bit set then the
3814 buffers of temps will not be stolen. C<sv_setsv>
3815 and C<sv_setsv_nomg> are implemented in terms of this function.
3817 You probably want to use one of the assortment of wrappers, such as
3818 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3819 C<SvSetMagicSV_nosteal>.
3821 This is the primary function for copying scalars, and most other
3822 copy-ish functions and macros use this underneath.
3824 =for apidoc Amnh||SV_NOSTEAL
3830 S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
3832 I32 mro_changes = 0; /* 1 = method, 2 = isa, 3 = recursive isa */
3833 HV *old_stash = NULL;
3835 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB;
3837 if (dtype != SVt_PVGV && !isGV_with_GP(dstr)) {
3838 const char * const name = GvNAME(sstr);
3839 const STRLEN len = GvNAMELEN(sstr);
3841 if (dtype >= SVt_PV) {
3847 SvUPGRADE(dstr, SVt_PVGV);
3848 (void)SvOK_off(dstr);
3849 isGV_with_GP_on(dstr);
3851 GvSTASH(dstr) = GvSTASH(sstr);
3853 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
3854 gv_name_set(MUTABLE_GV(dstr), name, len,
3855 GV_ADD | (GvNAMEUTF8(sstr) ? SVf_UTF8 : 0 ));
3856 SvFAKE_on(dstr); /* can coerce to non-glob */
3859 if(GvGP(MUTABLE_GV(sstr))) {
3860 /* If source has method cache entry, clear it */
3862 SvREFCNT_dec(GvCV(sstr));
3863 GvCV_set(sstr, NULL);
3866 /* If source has a real method, then a method is
3869 GvCV((const GV *)sstr) && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3875 /* If dest already had a real method, that's a change as well */
3877 !mro_changes && GvGP(MUTABLE_GV(dstr)) && GvCVu((const GV *)dstr)
3878 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3883 /* We don't need to check the name of the destination if it was not a
3884 glob to begin with. */
3885 if(dtype == SVt_PVGV) {
3886 const char * const name = GvNAME((const GV *)dstr);
3887 const STRLEN len = GvNAMELEN(dstr);
3888 if(memEQs(name, len, "ISA")
3889 /* The stash may have been detached from the symbol table, so
3891 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3895 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
3896 || (len == 1 && name[0] == ':')) {
3899 /* Set aside the old stash, so we can reset isa caches on
3901 if((old_stash = GvHV(dstr)))
3902 /* Make sure we do not lose it early. */
3903 SvREFCNT_inc_simple_void_NN(
3904 sv_2mortal((SV *)old_stash)
3909 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
3912 /* freeing dstr's GP might free sstr (e.g. *x = $x),
3913 * so temporarily protect it */
3915 SAVEFREESV(SvREFCNT_inc_simple_NN(sstr));
3916 gp_free(MUTABLE_GV(dstr));
3917 GvINTRO_off(dstr); /* one-shot flag */
3918 GvGP_set(dstr, gp_ref(GvGP(sstr)));
3921 if (SvTAINTED(sstr))
3923 if (GvIMPORTED(dstr) != GVf_IMPORTED
3924 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
3926 GvIMPORTED_on(dstr);
3929 if(mro_changes == 2) {
3930 if (GvAV((const GV *)sstr)) {
3932 SV * const sref = (SV *)GvAV((const GV *)dstr);
3933 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
3934 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
3935 AV * const ary = newAV();
3936 av_push(ary, mg->mg_obj); /* takes the refcount */
3937 mg->mg_obj = (SV *)ary;
3939 av_push((AV *)mg->mg_obj, SvREFCNT_inc_simple_NN(dstr));
3941 else sv_magic(sref, dstr, PERL_MAGIC_isa, NULL, 0);
3943 mro_isa_changed_in(GvSTASH(dstr));
3945 else if(mro_changes == 3) {
3946 HV * const stash = GvHV(dstr);
3947 if(old_stash ? (HV *)HvENAME_get(old_stash) : stash)
3953 else if(mro_changes) mro_method_changed_in(GvSTASH(dstr));
3954 if (GvIO(dstr) && dtype == SVt_PVGV) {
3955 DEBUG_o(Perl_deb(aTHX_
3956 "glob_assign_glob clearing PL_stashcache\n"));
3957 /* It's a cache. It will rebuild itself quite happily.
3958 It's a lot of effort to work out exactly which key (or keys)
3959 might be invalidated by the creation of the this file handle.
3961 hv_clear(PL_stashcache);
3967 Perl_gv_setref(pTHX_ SV *const dstr, SV *const sstr)
3969 SV * const sref = SvRV(sstr);
3971 const int intro = GvINTRO(dstr);
3974 const U32 stype = SvTYPE(sref);
3976 PERL_ARGS_ASSERT_GV_SETREF;
3979 GvINTRO_off(dstr); /* one-shot flag */
3980 GvLINE(dstr) = CopLINE(PL_curcop);
3981 GvEGV(dstr) = MUTABLE_GV(dstr);
3986 location = (SV **) &(GvGP(dstr)->gp_cv); /* XXX bypassing GvCV_set */
3987 import_flag = GVf_IMPORTED_CV;
3990 location = (SV **) &GvHV(dstr);
3991 import_flag = GVf_IMPORTED_HV;
3994 location = (SV **) &GvAV(dstr);
3995 import_flag = GVf_IMPORTED_AV;
3998 location = (SV **) &GvIOp(dstr);
4001 location = (SV **) &GvFORM(dstr);
4004 location = &GvSV(dstr);
4005 import_flag = GVf_IMPORTED_SV;
4008 if (stype == SVt_PVCV) {
4009 /*if (GvCVGEN(dstr) && (GvCV(dstr) != (const CV *)sref || GvCVGEN(dstr))) {*/
4010 if (GvCVGEN(dstr)) {
4011 SvREFCNT_dec(GvCV(dstr));
4012 GvCV_set(dstr, NULL);
4013 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4016 /* SAVEt_GVSLOT takes more room on the savestack and has more
4017 overhead in leave_scope than SAVEt_GENERIC_SV. But for CVs
4018 leave_scope needs access to the GV so it can reset method
4019 caches. We must use SAVEt_GVSLOT whenever the type is
4020 SVt_PVCV, even if the stash is anonymous, as the stash may
4021 gain a name somehow before leave_scope. */
4022 if (stype == SVt_PVCV) {
4023 /* There is no save_pushptrptrptr. Creating it for this
4024 one call site would be overkill. So inline the ss add
4028 SS_ADD_PTR(location);
4029 SS_ADD_PTR(SvREFCNT_inc(*location));
4030 SS_ADD_UV(SAVEt_GVSLOT);
4033 else SAVEGENERICSV(*location);
4036 if (stype == SVt_PVCV && (*location != sref || GvCVGEN(dstr))) {
4037 CV* const cv = MUTABLE_CV(*location);
4039 if (!GvCVGEN((const GV *)dstr) &&
4040 (CvROOT(cv) || CvXSUB(cv)) &&
4041 /* redundant check that avoids creating the extra SV
4042 most of the time: */
4043 (CvCONST(cv) || ckWARN(WARN_REDEFINE)))
4045 SV * const new_const_sv =
4046 CvCONST((const CV *)sref)
4047 ? cv_const_sv((const CV *)sref)
4049 HV * const stash = GvSTASH((const GV *)dstr);
4050 report_redefined_cv(
4053 ? Perl_newSVpvf(aTHX_
4054 "%" HEKf "::%" HEKf,
4055 HEKfARG(HvNAME_HEK(stash)),
4056 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr))))
4057 : Perl_newSVpvf(aTHX_
4059 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr))))
4062 CvCONST((const CV *)sref) ? &new_const_sv : NULL
4066 cv_ckproto_len_flags(cv, (const GV *)dstr,
4067 SvPOK(sref) ? CvPROTO(sref) : NULL,
4068 SvPOK(sref) ? CvPROTOLEN(sref) : 0,
4069 SvPOK(sref) ? SvUTF8(sref) : 0);
4071 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4072 GvASSUMECV_on(dstr);
4073 if(GvSTASH(dstr)) { /* sub foo { 1 } sub bar { 2 } *bar = \&foo */
4074 if (intro && GvREFCNT(dstr) > 1) {
4075 /* temporary remove extra savestack's ref */
4077 gv_method_changed(dstr);
4080 else gv_method_changed(dstr);
4083 *location = SvREFCNT_inc_simple_NN(sref);
4084 if (import_flag && !(GvFLAGS(dstr) & import_flag)
4085 && CopSTASH_ne(PL_curcop, GvSTASH(dstr))) {
4086 GvFLAGS(dstr) |= import_flag;
4089 if (stype == SVt_PVHV) {
4090 const char * const name = GvNAME((GV*)dstr);
4091 const STRLEN len = GvNAMELEN(dstr);
4094 (len > 1 && name[len-2] == ':' && name[len-1] == ':')
4095 || (len == 1 && name[0] == ':')
4097 && (!dref || HvENAME_get(dref))
4100 (HV *)sref, (HV *)dref,
4106 stype == SVt_PVAV && sref != dref
4107 && memEQs(GvNAME((GV*)dstr), GvNAMELEN((GV*)dstr), "ISA")
4108 /* The stash may have been detached from the symbol table, so
4109 check its name before doing anything. */
4110 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
4113 MAGIC * const omg = dref && SvSMAGICAL(dref)
4114 ? mg_find(dref, PERL_MAGIC_isa)
4116 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
4117 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
4118 AV * const ary = newAV();
4119 av_push(ary, mg->mg_obj); /* takes the refcount */
4120 mg->mg_obj = (SV *)ary;
4123 if (SvTYPE(omg->mg_obj) == SVt_PVAV) {
4124 SV **svp = AvARRAY((AV *)omg->mg_obj);
4125 I32 items = AvFILLp((AV *)omg->mg_obj) + 1;
4129 SvREFCNT_inc_simple_NN(*svp++)
4135 SvREFCNT_inc_simple_NN(omg->mg_obj)
4139 av_push((AV *)mg->mg_obj,SvREFCNT_inc_simple_NN(dstr));
4145 sref, omg ? omg->mg_obj : dstr, PERL_MAGIC_isa, NULL, 0
4147 for (i = 0; i <= AvFILL(sref); ++i) {
4148 SV **elem = av_fetch ((AV*)sref, i, 0);
4151 *elem, sref, PERL_MAGIC_isaelem, NULL, i
4155 mg = mg_find(sref, PERL_MAGIC_isa);
4157 /* Since the *ISA assignment could have affected more than
4158 one stash, don't call mro_isa_changed_in directly, but let
4159 magic_clearisa do it for us, as it already has the logic for
4160 dealing with globs vs arrays of globs. */
4162 Perl_magic_clearisa(aTHX_ NULL, mg);
4164 else if (stype == SVt_PVIO) {
4165 DEBUG_o(Perl_deb(aTHX_ "gv_setref clearing PL_stashcache\n"));
4166 /* It's a cache. It will rebuild itself quite happily.
4167 It's a lot of effort to work out exactly which key (or keys)
4168 might be invalidated by the creation of the this file handle.
4170 hv_clear(PL_stashcache);
4174 if (!intro) SvREFCNT_dec(dref);
4175 if (SvTAINTED(sstr))
4183 #ifdef PERL_DEBUG_READONLY_COW
4184 # include <sys/mman.h>
4186 # ifndef PERL_MEMORY_DEBUG_HEADER_SIZE
4187 # define PERL_MEMORY_DEBUG_HEADER_SIZE 0
4191 Perl_sv_buf_to_ro(pTHX_ SV *sv)
4193 struct perl_memory_debug_header * const header =
4194 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4195 const MEM_SIZE len = header->size;
4196 PERL_ARGS_ASSERT_SV_BUF_TO_RO;
4197 # ifdef PERL_TRACK_MEMPOOL
4198 if (!header->readonly) header->readonly = 1;
4200 if (mprotect(header, len, PROT_READ))
4201 Perl_warn(aTHX_ "mprotect RW for COW string %p %lu failed with %d",
4202 header, len, errno);
4206 S_sv_buf_to_rw(pTHX_ SV *sv)
4208 struct perl_memory_debug_header * const header =
4209 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4210 const MEM_SIZE len = header->size;
4211 PERL_ARGS_ASSERT_SV_BUF_TO_RW;
4212 if (mprotect(header, len, PROT_READ|PROT_WRITE))
4213 Perl_warn(aTHX_ "mprotect for COW string %p %lu failed with %d",
4214 header, len, errno);
4215 # ifdef PERL_TRACK_MEMPOOL
4216 header->readonly = 0;
4221 # define sv_buf_to_ro(sv) NOOP
4222 # define sv_buf_to_rw(sv) NOOP
4226 Perl_sv_setsv_flags(pTHX_ SV *dstr, SV* sstr, const I32 flags)
4231 unsigned int both_type;
4233 PERL_ARGS_ASSERT_SV_SETSV_FLAGS;
4235 if (UNLIKELY( sstr == dstr ))
4238 if (UNLIKELY( !sstr ))
4239 sstr = &PL_sv_undef;
4241 stype = SvTYPE(sstr);
4242 dtype = SvTYPE(dstr);
4243 both_type = (stype | dtype);
4245 /* with these values, we can check that both SVs are NULL/IV (and not
4246 * freed) just by testing the or'ed types */
4247 STATIC_ASSERT_STMT(SVt_NULL == 0);
4248 STATIC_ASSERT_STMT(SVt_IV == 1);
4249 if (both_type <= 1) {
4250 /* both src and dst are UNDEF/IV/RV, so we can do a lot of
4256 /* minimal subset of SV_CHECK_THINKFIRST_COW_DROP(dstr) */
4257 if (SvREADONLY(dstr))
4258 Perl_croak_no_modify();
4260 if (SvWEAKREF(dstr))
4261 sv_unref_flags(dstr, 0);
4263 old_rv = SvRV(dstr);
4266 assert(!SvGMAGICAL(sstr));
4267 assert(!SvGMAGICAL(dstr));
4269 sflags = SvFLAGS(sstr);
4270 if (sflags & (SVf_IOK|SVf_ROK)) {
4271 SET_SVANY_FOR_BODYLESS_IV(dstr);
4272 new_dflags = SVt_IV;
4274 if (sflags & SVf_ROK) {
4275 dstr->sv_u.svu_rv = SvREFCNT_inc(SvRV(sstr));
4276 new_dflags |= SVf_ROK;
4279 /* both src and dst are <= SVt_IV, so sv_any points to the
4280 * head; so access the head directly
4282 assert( &(sstr->sv_u.svu_iv)
4283 == &(((XPVIV*) SvANY(sstr))->xiv_iv));
4284 assert( &(dstr->sv_u.svu_iv)
4285 == &(((XPVIV*) SvANY(dstr))->xiv_iv));
4286 dstr->sv_u.svu_iv = sstr->sv_u.svu_iv;
4287 new_dflags |= (SVf_IOK|SVp_IOK|(sflags & SVf_IVisUV));
4291 new_dflags = dtype; /* turn off everything except the type */
4293 SvFLAGS(dstr) = new_dflags;
4294 SvREFCNT_dec(old_rv);
4299 if (UNLIKELY(both_type == SVTYPEMASK)) {
4300 if (SvIS_FREED(dstr)) {
4301 Perl_croak(aTHX_ "panic: attempt to copy value %" SVf
4302 " to a freed scalar %p", SVfARG(sstr), (void *)dstr);
4304 if (SvIS_FREED(sstr)) {
4305 Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p to %p",
4306 (void*)sstr, (void*)dstr);
4312 SV_CHECK_THINKFIRST_COW_DROP(dstr);
4313 dtype = SvTYPE(dstr); /* THINKFIRST may have changed type */
4315 /* There's a lot of redundancy below but we're going for speed here */
4320 if (LIKELY( dtype != SVt_PVGV && dtype != SVt_PVLV )) {
4321 (void)SvOK_off(dstr);
4329 /* For performance, we inline promoting to type SVt_IV. */
4330 /* We're starting from SVt_NULL, so provided that define is
4331 * actual 0, we don't have to unset any SV type flags
4332 * to promote to SVt_IV. */
4333 STATIC_ASSERT_STMT(SVt_NULL == 0);
4334 SET_SVANY_FOR_BODYLESS_IV(dstr);
4335 SvFLAGS(dstr) |= SVt_IV;
4339 sv_upgrade(dstr, SVt_PVIV);
4343 goto end_of_first_switch;
4345 (void)SvIOK_only(dstr);
4346 SvIV_set(dstr, SvIVX(sstr));
4349 /* SvTAINTED can only be true if the SV has taint magic, which in
4350 turn means that the SV type is PVMG (or greater). This is the
4351 case statement for SVt_IV, so this cannot be true (whatever gcov
4353 assert(!SvTAINTED(sstr));
4358 if (dtype < SVt_PV && dtype != SVt_IV)
4359 sv_upgrade(dstr, SVt_IV);
4363 if (LIKELY( SvNOK(sstr) )) {
4367 sv_upgrade(dstr, SVt_NV);
4371 sv_upgrade(dstr, SVt_PVNV);
4375 goto end_of_first_switch;
4377 SvNV_set(dstr, SvNVX(sstr));
4378 (void)SvNOK_only(dstr);
4379 /* SvTAINTED can only be true if the SV has taint magic, which in
4380 turn means that the SV type is PVMG (or greater). This is the
4381 case statement for SVt_NV, so this cannot be true (whatever gcov
4383 assert(!SvTAINTED(sstr));
4390 sv_upgrade(dstr, SVt_PV);
4393 if (dtype < SVt_PVIV)
4394 sv_upgrade(dstr, SVt_PVIV);
4397 if (dtype < SVt_PVNV)
4398 sv_upgrade(dstr, SVt_PVNV);
4402 invlist_clone(sstr, dstr);
4406 const char * const type = sv_reftype(sstr,0);
4408 /* diag_listed_as: Bizarre copy of %s */
4409 Perl_croak(aTHX_ "Bizarre copy of %s in %s", type, OP_DESC(PL_op));
4411 Perl_croak(aTHX_ "Bizarre copy of %s", type);
4413 NOT_REACHED; /* NOTREACHED */
4417 if (dtype < SVt_REGEXP)
4418 sv_upgrade(dstr, SVt_REGEXP);
4424 if (SvGMAGICAL(sstr) && (flags & SV_GMAGIC)) {
4426 if (SvTYPE(sstr) != stype)
4427 stype = SvTYPE(sstr);
4429 if (isGV_with_GP(sstr) && dtype <= SVt_PVLV) {
4430 glob_assign_glob(dstr, sstr, dtype);
4433 if (stype == SVt_PVLV)
4435 if (isREGEXP(sstr)) goto upgregexp;
4436 SvUPGRADE(dstr, SVt_PVNV);
4439 SvUPGRADE(dstr, (svtype)stype);
4441 end_of_first_switch:
4443 /* dstr may have been upgraded. */
4444 dtype = SvTYPE(dstr);
4445 sflags = SvFLAGS(sstr);
4447 if (UNLIKELY( dtype == SVt_PVCV )) {
4448 /* Assigning to a subroutine sets the prototype. */
4451 const char *const ptr = SvPV_const(sstr, len);
4453 SvGROW(dstr, len + 1);
4454 Copy(ptr, SvPVX(dstr), len + 1, char);
4455 SvCUR_set(dstr, len);
4457 SvFLAGS(dstr) |= sflags & SVf_UTF8;
4458 CvAUTOLOAD_off(dstr);
4463 else if (UNLIKELY(dtype == SVt_PVAV || dtype == SVt_PVHV
4464 || dtype == SVt_PVFM))
4466 const char * const type = sv_reftype(dstr,0);
4468 /* diag_listed_as: Cannot copy to %s */
4469 Perl_croak(aTHX_ "Cannot copy to %s in %s", type, OP_DESC(PL_op));
4471 Perl_croak(aTHX_ "Cannot copy to %s", type);
4472 } else if (sflags & SVf_ROK) {
4473 if (isGV_with_GP(dstr)
4474 && SvTYPE(SvRV(sstr)) == SVt_PVGV && isGV_with_GP(SvRV(sstr))) {
4477 if (GvIMPORTED(dstr) != GVf_IMPORTED
4478 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
4480 GvIMPORTED_on(dstr);
4485 glob_assign_glob(dstr, sstr, dtype);
4489 if (dtype >= SVt_PV) {
4490 if (isGV_with_GP(dstr)) {
4491 gv_setref(dstr, sstr);
4494 if (SvPVX_const(dstr)) {
4500 (void)SvOK_off(dstr);
4501 SvRV_set(dstr, SvREFCNT_inc(SvRV(sstr)));
4502 SvFLAGS(dstr) |= sflags & SVf_ROK;
4503 assert(!(sflags & SVp_NOK));
4504 assert(!(sflags & SVp_IOK));
4505 assert(!(sflags & SVf_NOK));
4506 assert(!(sflags & SVf_IOK));
4508 else if (isGV_with_GP(dstr)) {
4509 if (!(sflags & SVf_OK)) {
4510 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4511 "Undefined value assigned to typeglob");
4514 GV *gv = gv_fetchsv_nomg(sstr, GV_ADD, SVt_PVGV);
4515 if (dstr != (const SV *)gv) {
4516 const char * const name = GvNAME((const GV *)dstr);
4517 const STRLEN len = GvNAMELEN(dstr);
4518 HV *old_stash = NULL;
4519 bool reset_isa = FALSE;
4520 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
4521 || (len == 1 && name[0] == ':')) {
4522 /* Set aside the old stash, so we can reset isa caches
4523 on its subclasses. */
4524 if((old_stash = GvHV(dstr))) {
4525 /* Make sure we do not lose it early. */
4526 SvREFCNT_inc_simple_void_NN(
4527 sv_2mortal((SV *)old_stash)
4534 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
4535 gp_free(MUTABLE_GV(dstr));
4537 GvGP_set(dstr, gp_ref(GvGP(gv)));
4540 HV * const stash = GvHV(dstr);
4542 old_stash ? (HV *)HvENAME_get(old_stash) : stash
4552 else if ((dtype == SVt_REGEXP || dtype == SVt_PVLV)
4553 && (stype == SVt_REGEXP || isREGEXP(sstr))) {
4554 reg_temp_copy((REGEXP*)dstr, (REGEXP*)sstr);
4556 else if (sflags & SVp_POK) {
4557 const STRLEN cur = SvCUR(sstr);
4558 const STRLEN len = SvLEN(sstr);
4561 * We have three basic ways to copy the string:
4567 * Which we choose is based on various factors. The following
4568 * things are listed in order of speed, fastest to slowest:
4570 * - Copying a short string
4571 * - Copy-on-write bookkeeping
4573 * - Copying a long string
4575 * We swipe the string (steal the string buffer) if the SV on the
4576 * rhs is about to be freed anyway (TEMP and refcnt==1). This is a
4577 * big win on long strings. It should be a win on short strings if
4578 * SvPVX_const(dstr) has to be allocated. If not, it should not
4579 * slow things down, as SvPVX_const(sstr) would have been freed
4582 * We also steal the buffer from a PADTMP (operator target) if it
4583 * is ‘long enough’. For short strings, a swipe does not help
4584 * here, as it causes more malloc calls the next time the target
4585 * is used. Benchmarks show that even if SvPVX_const(dstr) has to
4586 * be allocated it is still not worth swiping PADTMPs for short
4587 * strings, as the savings here are small.
4589 * If swiping is not an option, then we see whether it is
4590 * worth using copy-on-write. If the lhs already has a buf-
4591 * fer big enough and the string is short, we skip it and fall back
4592 * to method 3, since memcpy is faster for short strings than the
4593 * later bookkeeping overhead that copy-on-write entails.
4595 * If the rhs is not a copy-on-write string yet, then we also
4596 * consider whether the buffer is too large relative to the string
4597 * it holds. Some operations such as readline allocate a large
4598 * buffer in the expectation of reusing it. But turning such into
4599 * a COW buffer is counter-productive because it increases memory
4600 * usage by making readline allocate a new large buffer the sec-
4601 * ond time round. So, if the buffer is too large, again, we use
4604 * Finally, if there is no buffer on the left, or the buffer is too
4605 * small, then we use copy-on-write and make both SVs share the
4610 /* Whichever path we take through the next code, we want this true,
4611 and doing it now facilitates the COW check. */
4612 (void)SvPOK_only(dstr);
4616 /* slated for free anyway (and not COW)? */
4617 (sflags & (SVs_TEMP|SVf_IsCOW)) == SVs_TEMP
4618 /* or a swipable TARG */
4620 (SVs_PADTMP|SVf_READONLY|SVf_PROTECT|SVf_IsCOW))
4622 /* whose buffer is worth stealing */
4623 && CHECK_COWBUF_THRESHOLD(cur,len)
4626 !(sflags & SVf_OOK) && /* and not involved in OOK hack? */
4627 (!(flags & SV_NOSTEAL)) &&
4628 /* and we're allowed to steal temps */
4629 SvREFCNT(sstr) == 1 && /* and no other references to it? */
4630 len) /* and really is a string */
4631 { /* Passes the swipe test. */
4632 if (SvPVX_const(dstr)) /* we know that dtype >= SVt_PV */
4634 SvPV_set(dstr, SvPVX_mutable(sstr));
4635 SvLEN_set(dstr, SvLEN(sstr));
4636 SvCUR_set(dstr, SvCUR(sstr));
4639 (void)SvOK_off(sstr); /* NOTE: nukes most SvFLAGS on sstr */
4640 SvPV_set(sstr, NULL);
4645 else if (flags & SV_COW_SHARED_HASH_KEYS
4647 #ifdef PERL_COPY_ON_WRITE
4650 ( (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4651 /* If this is a regular (non-hek) COW, only so
4652 many COW "copies" are possible. */
4653 && CowREFCNT(sstr) != SV_COW_REFCNT_MAX ))
4654 : ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4655 && !(SvFLAGS(dstr) & SVf_BREAK)
4656 && CHECK_COW_THRESHOLD(cur,len) && cur+1 < len
4657 && (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4661 && !(SvFLAGS(dstr) & SVf_BREAK)
4664 /* Either it's a shared hash key, or it's suitable for
4668 PerlIO_printf(Perl_debug_log, "Copy on write: sstr --> dstr\n");
4674 if (!(sflags & SVf_IsCOW)) {
4676 CowREFCNT(sstr) = 0;
4679 if (SvPVX_const(dstr)) { /* we know that dtype >= SVt_PV */
4685 if (sflags & SVf_IsCOW) {
4689 SvPV_set(dstr, SvPVX_mutable(sstr));
4694 /* SvIsCOW_shared_hash */
4695 DEBUG_C(PerlIO_printf(Perl_debug_log,
4696 "Copy on write: Sharing hash\n"));
4698 assert (SvTYPE(dstr) >= SVt_PV);
4700 HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)))));
4702 SvLEN_set(dstr, len);
4703 SvCUR_set(dstr, cur);
4706 /* Failed the swipe test, and we cannot do copy-on-write either.
4707 Have to copy the string. */
4708 SvGROW(dstr, cur + 1); /* inlined from sv_setpvn */
4709 Move(SvPVX_const(sstr),SvPVX(dstr),cur,char);
4710 SvCUR_set(dstr, cur);
4711 *SvEND(dstr) = '\0';
4713 if (sflags & SVp_NOK) {
4714 SvNV_set(dstr, SvNVX(sstr));
4716 if (sflags & SVp_IOK) {
4717 SvIV_set(dstr, SvIVX(sstr));
4718 if (sflags & SVf_IVisUV)
4721 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_NOK|SVp_NOK|SVf_UTF8);
4723 const MAGIC * const smg = SvVSTRING_mg(sstr);
4725 sv_magic(dstr, NULL, PERL_MAGIC_vstring,
4726 smg->mg_ptr, smg->mg_len);
4727 SvRMAGICAL_on(dstr);
4731 else if (sflags & (SVp_IOK|SVp_NOK)) {
4732 (void)SvOK_off(dstr);
4733 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_IVisUV|SVf_NOK|SVp_NOK);
4734 if (sflags & SVp_IOK) {
4735 /* XXXX Do we want to set IsUV for IV(ROK)? Be extra safe... */
4736 SvIV_set(dstr, SvIVX(sstr));
4738 if (sflags & SVp_NOK) {
4739 SvNV_set(dstr, SvNVX(sstr));
4743 if (isGV_with_GP(sstr)) {
4744 gv_efullname3(dstr, MUTABLE_GV(sstr), "*");
4747 (void)SvOK_off(dstr);
4749 if (SvTAINTED(sstr))
4755 =for apidoc sv_set_undef
4757 Equivalent to C<sv_setsv(sv, &PL_sv_undef)>, but more efficient.
4758 Doesn't handle set magic.
4760 The perl equivalent is C<$sv = undef;>. Note that it doesn't free any string
4761 buffer, unlike C<undef $sv>.
4763 Introduced in perl 5.25.12.
4769 Perl_sv_set_undef(pTHX_ SV *sv)
4771 U32 type = SvTYPE(sv);
4773 PERL_ARGS_ASSERT_SV_SET_UNDEF;
4775 /* shortcut, NULL, IV, RV */
4777 if (type <= SVt_IV) {
4778 assert(!SvGMAGICAL(sv));
4779 if (SvREADONLY(sv)) {
4780 /* does undeffing PL_sv_undef count as modifying a read-only
4781 * variable? Some XS code does this */
4782 if (sv == &PL_sv_undef)
4784 Perl_croak_no_modify();
4789 sv_unref_flags(sv, 0);
4792 SvFLAGS(sv) = type; /* quickly turn off all flags */
4793 SvREFCNT_dec_NN(rv);
4797 SvFLAGS(sv) = type; /* quickly turn off all flags */
4802 Perl_croak(aTHX_ "panic: attempt to undefine a freed scalar %p",
4805 SV_CHECK_THINKFIRST_COW_DROP(sv);
4807 if (isGV_with_GP(sv))
4808 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4809 "Undefined value assigned to typeglob");
4817 =for apidoc sv_setsv_mg
4819 Like C<sv_setsv>, but also handles 'set' magic.
4825 Perl_sv_setsv_mg(pTHX_ SV *const dstr, SV *const sstr)
4827 PERL_ARGS_ASSERT_SV_SETSV_MG;
4829 sv_setsv(dstr,sstr);
4834 # define SVt_COW SVt_PV
4836 Perl_sv_setsv_cow(pTHX_ SV *dstr, SV *sstr)
4838 STRLEN cur = SvCUR(sstr);
4839 STRLEN len = SvLEN(sstr);
4841 #if defined(PERL_DEBUG_READONLY_COW) && defined(PERL_COPY_ON_WRITE)
4842 const bool already = cBOOL(SvIsCOW(sstr));
4845 PERL_ARGS_ASSERT_SV_SETSV_COW;
4848 PerlIO_printf(Perl_debug_log, "Fast copy on write: %p -> %p\n",
4849 (void*)sstr, (void*)dstr);
4856 if (SvTHINKFIRST(dstr))
4857 sv_force_normal_flags(dstr, SV_COW_DROP_PV);
4858 else if (SvPVX_const(dstr))
4859 Safefree(SvPVX_mutable(dstr));
4863 SvUPGRADE(dstr, SVt_COW);
4865 assert (SvPOK(sstr));
4866 assert (SvPOKp(sstr));
4868 if (SvIsCOW(sstr)) {
4870 if (SvLEN(sstr) == 0) {
4871 /* source is a COW shared hash key. */
4872 DEBUG_C(PerlIO_printf(Perl_debug_log,
4873 "Fast copy on write: Sharing hash\n"));
4874 new_pv = HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr))));
4877 assert(SvCUR(sstr)+1 < SvLEN(sstr));
4878 assert(CowREFCNT(sstr) < SV_COW_REFCNT_MAX);
4880 assert ((SvFLAGS(sstr) & CAN_COW_MASK) == CAN_COW_FLAGS);
4881 SvUPGRADE(sstr, SVt_COW);
4883 DEBUG_C(PerlIO_printf(Perl_debug_log,
4884 "Fast copy on write: Converting sstr to COW\n"));
4885 CowREFCNT(sstr) = 0;
4887 # ifdef PERL_DEBUG_READONLY_COW
4888 if (already) sv_buf_to_rw(sstr);
4891 new_pv = SvPVX_mutable(sstr);
4895 SvPV_set(dstr, new_pv);
4896 SvFLAGS(dstr) = (SVt_COW|SVf_POK|SVp_POK|SVf_IsCOW);
4899 SvLEN_set(dstr, len);
4900 SvCUR_set(dstr, cur);
4910 =for apidoc sv_setpv_bufsize
4912 Sets the SV to be a string of cur bytes length, with at least
4913 len bytes available. Ensures that there is a null byte at SvEND.
4914 Returns a char * pointer to the SvPV buffer.
4920 Perl_sv_setpv_bufsize(pTHX_ SV *const sv, const STRLEN cur, const STRLEN len)
4924 PERL_ARGS_ASSERT_SV_SETPV_BUFSIZE;
4926 SV_CHECK_THINKFIRST_COW_DROP(sv);
4927 SvUPGRADE(sv, SVt_PV);
4928 pv = SvGROW(sv, len + 1);
4931 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4934 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4939 =for apidoc sv_setpvn
4941 Copies a string (possibly containing embedded C<NUL> characters) into an SV.
4942 The C<len> parameter indicates the number of
4943 bytes to be copied. If the C<ptr> argument is NULL the SV will become
4944 undefined. Does not handle 'set' magic. See C<L</sv_setpvn_mg>>.
4950 Perl_sv_setpvn(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4954 PERL_ARGS_ASSERT_SV_SETPVN;
4956 SV_CHECK_THINKFIRST_COW_DROP(sv);
4957 if (isGV_with_GP(sv))
4958 Perl_croak_no_modify();
4964 /* len is STRLEN which is unsigned, need to copy to signed */
4967 Perl_croak(aTHX_ "panic: sv_setpvn called with negative strlen %"
4970 SvUPGRADE(sv, SVt_PV);
4972 dptr = SvGROW(sv, len + 1);
4973 Move(ptr,dptr,len,char);
4976 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4978 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4982 =for apidoc sv_setpvn_mg
4984 Like C<sv_setpvn>, but also handles 'set' magic.
4990 Perl_sv_setpvn_mg(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4992 PERL_ARGS_ASSERT_SV_SETPVN_MG;
4994 sv_setpvn(sv,ptr,len);
4999 =for apidoc sv_setpv
5001 Copies a string into an SV. The string must be terminated with a C<NUL>
5002 character, and not contain embeded C<NUL>'s.
5003 Does not handle 'set' magic. See C<L</sv_setpv_mg>>.
5009 Perl_sv_setpv(pTHX_ SV *const sv, const char *const ptr)
5013 PERL_ARGS_ASSERT_SV_SETPV;
5015 SV_CHECK_THINKFIRST_COW_DROP(sv);
5021 SvUPGRADE(sv, SVt_PV);
5023 SvGROW(sv, len + 1);
5024 Move(ptr,SvPVX(sv),len+1,char);
5026 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5028 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
5032 =for apidoc sv_setpv_mg
5034 Like C<sv_setpv>, but also handles 'set' magic.
5040 Perl_sv_setpv_mg(pTHX_ SV *const sv, const char *const ptr)
5042 PERL_ARGS_ASSERT_SV_SETPV_MG;
5049 Perl_sv_sethek(pTHX_ SV *const sv, const HEK *const hek)
5051 PERL_ARGS_ASSERT_SV_SETHEK;
5057 if (HEK_LEN(hek) == HEf_SVKEY) {
5058 sv_setsv(sv, *(SV**)HEK_KEY(hek));
5061 const int flags = HEK_FLAGS(hek);
5062 if (flags & HVhek_WASUTF8) {
5063 STRLEN utf8_len = HEK_LEN(hek);
5064 char *as_utf8 = (char *)bytes_to_utf8((U8*)HEK_KEY(hek), &utf8_len);
5065 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
5068 } else if (flags & HVhek_UNSHARED) {
5069 sv_setpvn(sv, HEK_KEY(hek), HEK_LEN(hek));
5072 else SvUTF8_off(sv);
5076 SV_CHECK_THINKFIRST_COW_DROP(sv);
5077 SvUPGRADE(sv, SVt_PV);
5079 SvPV_set(sv,(char *)HEK_KEY(share_hek_hek(hek)));
5080 SvCUR_set(sv, HEK_LEN(hek));
5086 else SvUTF8_off(sv);
5094 =for apidoc sv_usepvn_flags
5096 Tells an SV to use C<ptr> to find its string value. Normally the
5097 string is stored inside the SV, but sv_usepvn allows the SV to use an
5098 outside string. C<ptr> should point to memory that was allocated
5099 by L<C<Newx>|perlclib/Memory Management and String Handling>. It must be
5100 the start of a C<Newx>-ed block of memory, and not a pointer to the
5101 middle of it (beware of L<C<OOK>|perlguts/Offsets> and copy-on-write),
5102 and not be from a non-C<Newx> memory allocator like C<malloc>. The
5103 string length, C<len>, must be supplied. By default this function
5104 will C<Renew> (i.e. realloc, move) the memory pointed to by C<ptr>,
5105 so that pointer should not be freed or used by the programmer after
5106 giving it to C<sv_usepvn>, and neither should any pointers from "behind"
5107 that pointer (e.g. ptr + 1) be used.
5109 If S<C<flags & SV_SMAGIC>> is true, will call C<SvSETMAGIC>. If
5110 S<C<flags & SV_HAS_TRAILING_NUL>> is true, then C<ptr[len]> must be C<NUL>,
5112 will be skipped (i.e. the buffer is actually at least 1 byte longer than
5113 C<len>, and already meets the requirements for storing in C<SvPVX>).
5115 =for apidoc Amnh||SV_SMAGIC
5116 =for apidoc Amnh||SV_HAS_TRAILING_NUL
5122 Perl_sv_usepvn_flags(pTHX_ SV *const sv, char *ptr, const STRLEN len, const U32 flags)
5126 PERL_ARGS_ASSERT_SV_USEPVN_FLAGS;
5128 SV_CHECK_THINKFIRST_COW_DROP(sv);
5129 SvUPGRADE(sv, SVt_PV);
5132 if (flags & SV_SMAGIC)
5136 if (SvPVX_const(sv))
5140 if (flags & SV_HAS_TRAILING_NUL)
5141 assert(ptr[len] == '\0');
5144 allocate = (flags & SV_HAS_TRAILING_NUL)
5146 #ifdef Perl_safesysmalloc_size
5149 PERL_STRLEN_ROUNDUP(len + 1);
5151 if (flags & SV_HAS_TRAILING_NUL) {
5152 /* It's long enough - do nothing.
5153 Specifically Perl_newCONSTSUB is relying on this. */
5156 /* Force a move to shake out bugs in callers. */
5157 char *new_ptr = (char*)safemalloc(allocate);
5158 Copy(ptr, new_ptr, len, char);
5159 PoisonFree(ptr,len,char);
5163 ptr = (char*) saferealloc (ptr, allocate);
5166 #ifdef Perl_safesysmalloc_size
5167 SvLEN_set(sv, Perl_safesysmalloc_size(ptr));
5169 SvLEN_set(sv, allocate);
5173 if (!(flags & SV_HAS_TRAILING_NUL)) {
5176 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5178 if (flags & SV_SMAGIC)
5184 S_sv_uncow(pTHX_ SV * const sv, const U32 flags)
5186 assert(SvIsCOW(sv));
5189 const char * const pvx = SvPVX_const(sv);
5190 const STRLEN len = SvLEN(sv);
5191 const STRLEN cur = SvCUR(sv);
5195 PerlIO_printf(Perl_debug_log,
5196 "Copy on write: Force normal %ld\n",
5202 # ifdef PERL_COPY_ON_WRITE
5204 /* Must do this first, since the CowREFCNT uses SvPVX and
5205 we need to write to CowREFCNT, or de-RO the whole buffer if we are
5206 the only owner left of the buffer. */
5207 sv_buf_to_rw(sv); /* NOOP if RO-ing not supported */
5209 U8 cowrefcnt = CowREFCNT(sv);
5210 if(cowrefcnt != 0) {
5212 CowREFCNT(sv) = cowrefcnt;
5217 /* Else we are the only owner of the buffer. */
5222 /* This SV doesn't own the buffer, so need to Newx() a new one: */
5227 if (flags & SV_COW_DROP_PV) {
5228 /* OK, so we don't need to copy our buffer. */
5231 SvGROW(sv, cur + 1);
5232 Move(pvx,SvPVX(sv),cur,char);
5237 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5245 const char * const pvx = SvPVX_const(sv);
5246 const STRLEN len = SvCUR(sv);
5250 if (flags & SV_COW_DROP_PV) {
5251 /* OK, so we don't need to copy our buffer. */
5254 SvGROW(sv, len + 1);
5255 Move(pvx,SvPVX(sv),len,char);
5258 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5265 =for apidoc sv_force_normal_flags
5267 Undo various types of fakery on an SV, where fakery means
5268 "more than" a string: if the PV is a shared string, make
5269 a private copy; if we're a ref, stop refing; if we're a glob, downgrade to
5270 an C<xpvmg>; if we're a copy-on-write scalar, this is the on-write time when
5271 we do the copy, and is also used locally; if this is a
5272 vstring, drop the vstring magic. If C<SV_COW_DROP_PV> is set
5273 then a copy-on-write scalar drops its PV buffer (if any) and becomes
5274 C<SvPOK_off> rather than making a copy. (Used where this
5275 scalar is about to be set to some other value.) In addition,
5276 the C<flags> parameter gets passed to C<sv_unref_flags()>
5277 when unreffing. C<sv_force_normal> calls this function
5278 with flags set to 0.
5280 This function is expected to be used to signal to perl that this SV is
5281 about to be written to, and any extra book-keeping needs to be taken care
5282 of. Hence, it croaks on read-only values.
5284 =for apidoc Amnh||SV_COW_DROP_PV
5290 Perl_sv_force_normal_flags(pTHX_ SV *const sv, const U32 flags)
5292 PERL_ARGS_ASSERT_SV_FORCE_NORMAL_FLAGS;
5295 Perl_croak_no_modify();
5296 else if (SvIsCOW(sv) && LIKELY(SvTYPE(sv) != SVt_PVHV))
5297 S_sv_uncow(aTHX_ sv, flags);
5299 sv_unref_flags(sv, flags);
5300 else if (SvFAKE(sv) && isGV_with_GP(sv))
5301 sv_unglob(sv, flags);
5302 else if (SvFAKE(sv) && isREGEXP(sv)) {
5303 /* Need to downgrade the REGEXP to a simple(r) scalar. This is analogous
5304 to sv_unglob. We only need it here, so inline it. */
5305 const bool islv = SvTYPE(sv) == SVt_PVLV;
5306 const svtype new_type =
5307 islv ? SVt_NULL : SvMAGIC(sv) || SvSTASH(sv) ? SVt_PVMG : SVt_PV;
5308 SV *const temp = newSV_type(new_type);
5309 regexp *old_rx_body;
5311 if (new_type == SVt_PVMG) {
5312 SvMAGIC_set(temp, SvMAGIC(sv));
5313 SvMAGIC_set(sv, NULL);
5314 SvSTASH_set(temp, SvSTASH(sv));
5315 SvSTASH_set(sv, NULL);
5318 SvCUR_set(temp, SvCUR(sv));
5319 /* Remember that SvPVX is in the head, not the body. */
5320 assert(ReANY((REGEXP *)sv)->mother_re);
5323 /* LV-as-regex has sv->sv_any pointing to an XPVLV body,
5324 * whose xpvlenu_rx field points to the regex body */
5325 XPV *xpv = (XPV*)(SvANY(sv));
5326 old_rx_body = xpv->xpv_len_u.xpvlenu_rx;
5327 xpv->xpv_len_u.xpvlenu_rx = NULL;
5330 old_rx_body = ReANY((REGEXP *)sv);
5332 /* Their buffer is already owned by someone else. */
5333 if (flags & SV_COW_DROP_PV) {
5334 /* SvLEN is already 0. For SVt_REGEXP, we have a brand new
5335 zeroed body. For SVt_PVLV, we zeroed it above (len field
5336 a union with xpvlenu_rx) */
5337 assert(!SvLEN(islv ? sv : temp));
5338 sv->sv_u.svu_pv = 0;
5341 sv->sv_u.svu_pv = savepvn(RX_WRAPPED((REGEXP *)sv), SvCUR(sv));
5342 SvLEN_set(islv ? sv : temp, SvCUR(sv)+1);
5346 /* Now swap the rest of the bodies. */
5350 SvFLAGS(sv) &= ~SVTYPEMASK;
5351 SvFLAGS(sv) |= new_type;
5352 SvANY(sv) = SvANY(temp);
5355 SvFLAGS(temp) &= ~(SVTYPEMASK);
5356 SvFLAGS(temp) |= SVt_REGEXP|SVf_FAKE;
5357 SvANY(temp) = old_rx_body;
5359 SvREFCNT_dec_NN(temp);
5361 else if (SvVOK(sv)) sv_unmagic(sv, PERL_MAGIC_vstring);
5367 Efficient removal of characters from the beginning of the string buffer.
5368 C<SvPOK(sv)>, or at least C<SvPOKp(sv)>, must be true and C<ptr> must be a
5369 pointer to somewhere inside the string buffer. C<ptr> becomes the first
5370 character of the adjusted string. Uses the C<OOK> hack. On return, only
5371 C<SvPOK(sv)> and C<SvPOKp(sv)> among the C<OK> flags will be true.
5373 Beware: after this function returns, C<ptr> and SvPVX_const(sv) may no longer
5374 refer to the same chunk of data.
5376 The unfortunate similarity of this function's name to that of Perl's C<chop>
5377 operator is strictly coincidental. This function works from the left;
5378 C<chop> works from the right.
5384 Perl_sv_chop(pTHX_ SV *const sv, const char *const ptr)
5395 PERL_ARGS_ASSERT_SV_CHOP;
5397 if (!ptr || !SvPOKp(sv))
5399 delta = ptr - SvPVX_const(sv);
5401 /* Nothing to do. */
5404 max_delta = SvLEN(sv) ? SvLEN(sv) : SvCUR(sv);
5405 if (delta > max_delta)
5406 Perl_croak(aTHX_ "panic: sv_chop ptr=%p, start=%p, end=%p",
5407 ptr, SvPVX_const(sv), SvPVX_const(sv) + max_delta);
5408 /* SvPVX(sv) may move in SV_CHECK_THINKFIRST(sv), so don't use ptr any more */
5409 SV_CHECK_THINKFIRST(sv);
5410 SvPOK_only_UTF8(sv);
5413 if (!SvLEN(sv)) { /* make copy of shared string */
5414 const char *pvx = SvPVX_const(sv);
5415 const STRLEN len = SvCUR(sv);
5416 SvGROW(sv, len + 1);
5417 Move(pvx,SvPVX(sv),len,char);
5423 SvOOK_offset(sv, old_delta);
5425 SvLEN_set(sv, SvLEN(sv) - delta);
5426 SvCUR_set(sv, SvCUR(sv) - delta);
5427 SvPV_set(sv, SvPVX(sv) + delta);
5429 p = (U8 *)SvPVX_const(sv);
5432 /* how many bytes were evacuated? we will fill them with sentinel
5433 bytes, except for the part holding the new offset of course. */
5436 evacn += (old_delta < 0x100 ? 1 : 1 + sizeof(STRLEN));
5438 assert(evacn <= delta + old_delta);
5442 /* This sets 'delta' to the accumulated value of all deltas so far */
5446 /* If 'delta' fits in a byte, store it just prior to the new beginning of
5447 * the string; otherwise store a 0 byte there and store 'delta' just prior
5448 * to that, using as many bytes as a STRLEN occupies. Thus it overwrites a
5449 * portion of the chopped part of the string */
5450 if (delta < 0x100) {
5454 p -= sizeof(STRLEN);
5455 Copy((U8*)&delta, p, sizeof(STRLEN), U8);
5459 /* Fill the preceding buffer with sentinals to verify that no-one is
5469 =for apidoc sv_catpvn
5471 Concatenates the string onto the end of the string which is in the SV.
5472 C<len> indicates number of bytes to copy. If the SV has the UTF-8
5473 status set, then the bytes appended should be valid UTF-8.
5474 Handles 'get' magic, but not 'set' magic. See C<L</sv_catpvn_mg>>.
5476 =for apidoc sv_catpvn_flags
5478 Concatenates the string onto the end of the string which is in the SV. The
5479 C<len> indicates number of bytes to copy.
5481 By default, the string appended is assumed to be valid UTF-8 if the SV has
5482 the UTF-8 status set, and a string of bytes otherwise. One can force the
5483 appended string to be interpreted as UTF-8 by supplying the C<SV_CATUTF8>
5484 flag, and as bytes by supplying the C<SV_CATBYTES> flag; the SV or the
5485 string appended will be upgraded to UTF-8 if necessary.
5487 If C<flags> has the C<SV_SMAGIC> bit set, will
5488 C<mg_set> on C<dsv> afterwards if appropriate.
5489 C<sv_catpvn> and C<sv_catpvn_nomg> are implemented
5490 in terms of this function.
5496 Perl_sv_catpvn_flags(pTHX_ SV *const dsv, const char *sstr, const STRLEN slen, const I32 flags)
5499 const char * const dstr = SvPV_force_flags(dsv, dlen, flags);
5501 PERL_ARGS_ASSERT_SV_CATPVN_FLAGS;
5502 assert((flags & (SV_CATBYTES|SV_CATUTF8)) != (SV_CATBYTES|SV_CATUTF8));
5504 if (!(flags & SV_CATBYTES) || !SvUTF8(dsv)) {
5505 if (flags & SV_CATUTF8 && !SvUTF8(dsv)) {
5506 sv_utf8_upgrade_flags_grow(dsv, 0, slen + 1);
5509 else SvGROW(dsv, dlen + slen + 3);
5511 sstr = SvPVX_const(dsv);
5512 Move(sstr, SvPVX(dsv) + dlen, slen, char);
5513 SvCUR_set(dsv, SvCUR(dsv) + slen);
5516 /* We inline bytes_to_utf8, to avoid an extra malloc. */
5517 const char * const send = sstr + slen;
5520 /* Something this code does not account for, which I think is
5521 impossible; it would require the same pv to be treated as
5522 bytes *and* utf8, which would indicate a bug elsewhere. */
5523 assert(sstr != dstr);
5525 SvGROW(dsv, dlen + slen * 2 + 3);
5526 d = (U8 *)SvPVX(dsv) + dlen;
5528 while (sstr < send) {
5529 append_utf8_from_native_byte(*sstr, &d);
5532 SvCUR_set(dsv, d-(const U8 *)SvPVX(dsv));
5535 (void)SvPOK_only_UTF8(dsv); /* validate pointer */
5537 if (flags & SV_SMAGIC)
5542 =for apidoc sv_catsv
5544 Concatenates the string from SV C<ssv> onto the end of the string in SV
5545 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5546 Handles 'get' magic on both SVs, but no 'set' magic. See C<L</sv_catsv_mg>>
5547 and C<L</sv_catsv_nomg>>.
5549 =for apidoc sv_catsv_flags
5551 Concatenates the string from SV C<ssv> onto the end of the string in SV
5552 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5553 If C<flags> has the C<SV_GMAGIC> bit set, will call C<mg_get> on both SVs if
5554 appropriate. If C<flags> has the C<SV_SMAGIC> bit set, C<mg_set> will be called on
5555 the modified SV afterward, if appropriate. C<sv_catsv>, C<sv_catsv_nomg>,
5556 and C<sv_catsv_mg> are implemented in terms of this function.
5561 Perl_sv_catsv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
5563 PERL_ARGS_ASSERT_SV_CATSV_FLAGS;
5567 const char *spv = SvPV_flags_const(ssv, slen, flags);
5568 if (flags & SV_GMAGIC)
5570 sv_catpvn_flags(dsv, spv, slen,
5571 DO_UTF8(ssv) ? SV_CATUTF8 : SV_CATBYTES);
5572 if (flags & SV_SMAGIC)
5578 =for apidoc sv_catpv
5580 Concatenates the C<NUL>-terminated string onto the end of the string which is
5582 If the SV has the UTF-8 status set, then the bytes appended should be
5583 valid UTF-8. Handles 'get' magic, but not 'set' magic. See
5589 Perl_sv_catpv(pTHX_ SV *const sv, const char *ptr)
5595 PERL_ARGS_ASSERT_SV_CATPV;
5599 junk = SvPV_force(sv, tlen);
5601 SvGROW(sv, tlen + len + 1);
5603 ptr = SvPVX_const(sv);
5604 Move(ptr,SvPVX(sv)+tlen,len+1,char);
5605 SvCUR_set(sv, SvCUR(sv) + len);
5606 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5611 =for apidoc sv_catpv_flags
5613 Concatenates the C<NUL>-terminated string onto the end of the string which is
5615 If the SV has the UTF-8 status set, then the bytes appended should
5616 be valid UTF-8. If C<flags> has the C<SV_SMAGIC> bit set, will C<mg_set>
5617 on the modified SV if appropriate.
5623 Perl_sv_catpv_flags(pTHX_ SV *dstr, const char *sstr, const I32 flags)
5625 PERL_ARGS_ASSERT_SV_CATPV_FLAGS;
5626 sv_catpvn_flags(dstr, sstr, strlen(sstr), flags);
5630 =for apidoc sv_catpv_mg
5632 Like C<sv_catpv>, but also handles 'set' magic.
5638 Perl_sv_catpv_mg(pTHX_ SV *const sv, const char *const ptr)
5640 PERL_ARGS_ASSERT_SV_CATPV_MG;
5649 Creates a new SV. A non-zero C<len> parameter indicates the number of
5650 bytes of preallocated string space the SV should have. An extra byte for a
5651 trailing C<NUL> is also reserved. (C<SvPOK> is not set for the SV even if string
5652 space is allocated.) The reference count for the new SV is set to 1.
5654 In 5.9.3, C<newSV()> replaces the older C<NEWSV()> API, and drops the first
5655 parameter, I<x>, a debug aid which allowed callers to identify themselves.
5656 This aid has been superseded by a new build option, C<PERL_MEM_LOG> (see
5657 L<perlhacktips/PERL_MEM_LOG>). The older API is still there for use in XS
5658 modules supporting older perls.
5664 Perl_newSV(pTHX_ const STRLEN len)
5670 sv_grow(sv, len + 1);
5675 =for apidoc sv_magicext
5677 Adds magic to an SV, upgrading it if necessary. Applies the
5678 supplied C<vtable> and returns a pointer to the magic added.
5680 Note that C<sv_magicext> will allow things that C<sv_magic> will not.
5681 In particular, you can add magic to C<SvREADONLY> SVs, and add more than
5682 one instance of the same C<how>.
5684 If C<namlen> is greater than zero then a C<savepvn> I<copy> of C<name> is
5685 stored, if C<namlen> is zero then C<name> is stored as-is and - as another
5686 special case - if C<(name && namlen == HEf_SVKEY)> then C<name> is assumed
5687 to contain an SV* and is stored as-is with its C<REFCNT> incremented.
5689 (This is now used as a subroutine by C<sv_magic>.)
5694 Perl_sv_magicext(pTHX_ SV *const sv, SV *const obj, const int how,
5695 const MGVTBL *const vtable, const char *const name, const I32 namlen)
5699 PERL_ARGS_ASSERT_SV_MAGICEXT;
5701 SvUPGRADE(sv, SVt_PVMG);
5702 Newxz(mg, 1, MAGIC);
5703 mg->mg_moremagic = SvMAGIC(sv);
5704 SvMAGIC_set(sv, mg);
5706 /* Sometimes a magic contains a reference loop, where the sv and
5707 object refer to each other. To prevent a reference loop that
5708 would prevent such objects being freed, we look for such loops
5709 and if we find one we avoid incrementing the object refcount.
5711 Note we cannot do this to avoid self-tie loops as intervening RV must
5712 have its REFCNT incremented to keep it in existence.
5715 if (!obj || obj == sv ||
5716 how == PERL_MAGIC_arylen ||
5717 how == PERL_MAGIC_regdata ||
5718 how == PERL_MAGIC_regdatum ||
5719 how == PERL_MAGIC_symtab ||
5720 (SvTYPE(obj) == SVt_PVGV &&
5721 (GvSV(obj) == sv || GvHV(obj) == (const HV *)sv
5722 || GvAV(obj) == (const AV *)sv || GvCV(obj) == (const CV *)sv
5723 || GvIOp(obj) == (const IO *)sv || GvFORM(obj) == (const CV *)sv)))
5728 mg->mg_obj = SvREFCNT_inc_simple(obj);
5729 mg->mg_flags |= MGf_REFCOUNTED;
5732 /* Normal self-ties simply pass a null object, and instead of
5733 using mg_obj directly, use the SvTIED_obj macro to produce a
5734 new RV as needed. For glob "self-ties", we are tieing the PVIO
5735 with an RV obj pointing to the glob containing the PVIO. In
5736 this case, to avoid a reference loop, we need to weaken the
5740 if (how == PERL_MAGIC_tiedscalar && SvTYPE(sv) == SVt_PVIO &&
5741 obj && SvROK(obj) && GvIO(SvRV(obj)) == (const IO *)sv)
5747 mg->mg_len = namlen;
5750 mg->mg_ptr = savepvn(name, namlen);
5751 else if (namlen == HEf_SVKEY) {
5752 /* Yes, this is casting away const. This is only for the case of
5753 HEf_SVKEY. I think we need to document this aberation of the
5754 constness of the API, rather than making name non-const, as
5755 that change propagating outwards a long way. */
5756 mg->mg_ptr = (char*)SvREFCNT_inc_simple_NN((SV *)name);
5758 mg->mg_ptr = (char *) name;
5760 mg->mg_virtual = (MGVTBL *) vtable;
5767 Perl_sv_magicext_mglob(pTHX_ SV *sv)
5769 PERL_ARGS_ASSERT_SV_MAGICEXT_MGLOB;
5770 if (SvTYPE(sv) == SVt_PVLV && LvTYPE(sv) == 'y') {
5771 /* This sv is only a delegate. //g magic must be attached to
5776 return sv_magicext(sv, NULL, PERL_MAGIC_regex_global,
5777 &PL_vtbl_mglob, 0, 0);
5781 =for apidoc sv_magic
5783 Adds magic to an SV. First upgrades C<sv> to type C<SVt_PVMG> if
5784 necessary, then adds a new magic item of type C<how> to the head of the
5787 See C<L</sv_magicext>> (which C<sv_magic> now calls) for a description of the
5788 handling of the C<name> and C<namlen> arguments.
5790 You need to use C<sv_magicext> to add magic to C<SvREADONLY> SVs and also
5791 to add more than one instance of the same C<how>.
5797 Perl_sv_magic(pTHX_ SV *const sv, SV *const obj, const int how,
5798 const char *const name, const I32 namlen)
5800 const MGVTBL *vtable;
5803 unsigned int vtable_index;
5805 PERL_ARGS_ASSERT_SV_MAGIC;
5807 if (how < 0 || (unsigned)how >= C_ARRAY_LENGTH(PL_magic_data)
5808 || ((flags = PL_magic_data[how]),
5809 (vtable_index = flags & PERL_MAGIC_VTABLE_MASK)
5810 > magic_vtable_max))
5811 Perl_croak(aTHX_ "Don't know how to handle magic of type \\%o", how);
5813 /* PERL_MAGIC_ext is reserved for use by extensions not perl internals.
5814 Useful for attaching extension internal data to perl vars.
5815 Note that multiple extensions may clash if magical scalars
5816 etc holding private data from one are passed to another. */
5818 vtable = (vtable_index == magic_vtable_max)
5819 ? NULL : PL_magic_vtables + vtable_index;
5821 if (SvREADONLY(sv)) {
5823 !PERL_MAGIC_TYPE_READONLY_ACCEPTABLE(how)
5826 Perl_croak_no_modify();
5829 if (SvMAGICAL(sv) || (how == PERL_MAGIC_taint && SvTYPE(sv) >= SVt_PVMG)) {
5830 if (SvMAGIC(sv) && (mg = mg_find(sv, how))) {
5831 /* sv_magic() refuses to add a magic of the same 'how' as an
5834 if (how == PERL_MAGIC_taint)
5840 /* Force pos to be stored as characters, not bytes. */
5841 if (SvMAGICAL(sv) && DO_UTF8(sv)
5842 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
5844 && mg->mg_flags & MGf_BYTES) {
5845 mg->mg_len = (SSize_t)sv_pos_b2u_flags(sv, (STRLEN)mg->mg_len,
5847 mg->mg_flags &= ~MGf_BYTES;
5850 /* Rest of work is done else where */
5851 mg = sv_magicext(sv,obj,how,vtable,name,namlen);
5854 case PERL_MAGIC_taint:
5857 case PERL_MAGIC_ext:
5858 case PERL_MAGIC_dbfile:
5865 S_sv_unmagicext_flags(pTHX_ SV *const sv, const int type, MGVTBL *vtbl, const U32 flags)
5872 if (SvTYPE(sv) < SVt_PVMG || !SvMAGIC(sv))
5874 mgp = &(((XPVMG*) SvANY(sv))->xmg_u.xmg_magic);
5875 for (mg = *mgp; mg; mg = *mgp) {
5876 const MGVTBL* const virt = mg->mg_virtual;
5877 if (mg->mg_type == type && (!flags || virt == vtbl)) {
5878 *mgp = mg->mg_moremagic;
5879 if (virt && virt->svt_free)
5880 virt->svt_free(aTHX_ sv, mg);
5881 if (mg->mg_ptr && mg->mg_type != PERL_MAGIC_regex_global) {
5883 Safefree(mg->mg_ptr);
5884 else if (mg->mg_len == HEf_SVKEY)
5885 SvREFCNT_dec(MUTABLE_SV(mg->mg_ptr));
5886 else if (mg->mg_type == PERL_MAGIC_utf8)
5887 Safefree(mg->mg_ptr);
5889 if (mg->mg_flags & MGf_REFCOUNTED)
5890 SvREFCNT_dec(mg->mg_obj);
5894 mgp = &mg->mg_moremagic;
5897 if (SvMAGICAL(sv)) /* if we're under save_magic, wait for restore_magic; */
5898 mg_magical(sv); /* else fix the flags now */
5907 =for apidoc sv_unmagic
5909 Removes all magic of type C<type> from an SV.
5915 Perl_sv_unmagic(pTHX_ SV *const sv, const int type)
5917 PERL_ARGS_ASSERT_SV_UNMAGIC;
5918 return S_sv_unmagicext_flags(aTHX_ sv, type, NULL, 0);
5922 =for apidoc sv_unmagicext
5924 Removes all magic of type C<type> with the specified C<vtbl> from an SV.
5930 Perl_sv_unmagicext(pTHX_ SV *const sv, const int type, MGVTBL *vtbl)
5932 PERL_ARGS_ASSERT_SV_UNMAGICEXT;
5933 return S_sv_unmagicext_flags(aTHX_ sv, type, vtbl, 1);
5937 =for apidoc sv_rvweaken
5939 Weaken a reference: set the C<SvWEAKREF> flag on this RV; give the
5940 referred-to SV C<PERL_MAGIC_backref> magic if it hasn't already; and
5941 push a back-reference to this RV onto the array of backreferences
5942 associated with that magic. If the RV is magical, set magic will be
5943 called after the RV is cleared. Silently ignores C<undef> and warns
5944 on already-weak references.
5950 Perl_sv_rvweaken(pTHX_ SV *const sv)
5954 PERL_ARGS_ASSERT_SV_RVWEAKEN;
5956 if (!SvOK(sv)) /* let undefs pass */
5959 Perl_croak(aTHX_ "Can't weaken a nonreference");
5960 else if (SvWEAKREF(sv)) {
5961 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is already weak");
5964 else if (SvREADONLY(sv)) croak_no_modify();
5966 Perl_sv_add_backref(aTHX_ tsv, sv);
5968 SvREFCNT_dec_NN(tsv);
5973 =for apidoc sv_rvunweaken
5975 Unweaken a reference: Clear the C<SvWEAKREF> flag on this RV; remove
5976 the backreference to this RV from the array of backreferences
5977 associated with the target SV, increment the refcount of the target.
5978 Silently ignores C<undef> and warns on non-weak references.
5984 Perl_sv_rvunweaken(pTHX_ SV *const sv)
5988 PERL_ARGS_ASSERT_SV_RVUNWEAKEN;
5990 if (!SvOK(sv)) /* let undefs pass */
5993 Perl_croak(aTHX_ "Can't unweaken a nonreference");
5994 else if (!SvWEAKREF(sv)) {
5995 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is not weak");
5998 else if (SvREADONLY(sv)) croak_no_modify();
6003 SvREFCNT_inc_NN(tsv);
6004 Perl_sv_del_backref(aTHX_ tsv, sv);
6009 =for apidoc sv_get_backrefs
6011 If C<sv> is the target of a weak reference then it returns the back
6012 references structure associated with the sv; otherwise return C<NULL>.
6014 When returning a non-null result the type of the return is relevant. If it
6015 is an AV then the elements of the AV are the weak reference RVs which
6016 point at this item. If it is any other type then the item itself is the
6019 See also C<Perl_sv_add_backref()>, C<Perl_sv_del_backref()>,
6020 C<Perl_sv_kill_backrefs()>
6026 Perl_sv_get_backrefs(SV *const sv)
6030 PERL_ARGS_ASSERT_SV_GET_BACKREFS;
6032 /* find slot to store array or singleton backref */
6034 if (SvTYPE(sv) == SVt_PVHV) {
6036 struct xpvhv_aux * const iter = HvAUX((HV *)sv);
6037 backrefs = (SV *)iter->xhv_backreferences;
6039 } else if (SvMAGICAL(sv)) {
6040 MAGIC *mg = mg_find(sv, PERL_MAGIC_backref);
6042 backrefs = mg->mg_obj;
6047 /* Give tsv backref magic if it hasn't already got it, then push a
6048 * back-reference to sv onto the array associated with the backref magic.
6050 * As an optimisation, if there's only one backref and it's not an AV,
6051 * store it directly in the HvAUX or mg_obj slot, avoiding the need to
6052 * allocate an AV. (Whether the slot holds an AV tells us whether this is
6056 /* A discussion about the backreferences array and its refcount:
6058 * The AV holding the backreferences is pointed to either as the mg_obj of
6059 * PERL_MAGIC_backref, or in the specific case of a HV, from the
6060 * xhv_backreferences field. The array is created with a refcount
6061 * of 2. This means that if during global destruction the array gets
6062 * picked on before its parent to have its refcount decremented by the
6063 * random zapper, it won't actually be freed, meaning it's still there for
6064 * when its parent gets freed.
6066 * When the parent SV is freed, the extra ref is killed by
6067 * Perl_sv_kill_backrefs. The other ref is killed, in the case of magic,
6068 * by mg_free() / MGf_REFCOUNTED, or for a hash, by Perl_hv_kill_backrefs.
6070 * When a single backref SV is stored directly, it is not reference
6075 Perl_sv_add_backref(pTHX_ SV *const tsv, SV *const sv)
6081 PERL_ARGS_ASSERT_SV_ADD_BACKREF;
6083 /* find slot to store array or singleton backref */
6085 if (SvTYPE(tsv) == SVt_PVHV) {
6086 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
6089 mg = mg_find(tsv, PERL_MAGIC_backref);
6091 mg = sv_magicext(tsv, NULL, PERL_MAGIC_backref, &PL_vtbl_backref, NULL, 0);
6092 svp = &(mg->mg_obj);
6095 /* create or retrieve the array */
6097 if ( (!*svp && SvTYPE(sv) == SVt_PVAV)
6098 || (*svp && SvTYPE(*svp) != SVt_PVAV)
6102 mg->mg_flags |= MGf_REFCOUNTED;
6105 SvREFCNT_inc_simple_void_NN(av);
6106 /* av now has a refcnt of 2; see discussion above */
6107 av_extend(av, *svp ? 2 : 1);
6109 /* move single existing backref to the array */
6110 AvARRAY(av)[++AvFILLp(av)] = *svp; /* av_push() */
6115 av = MUTABLE_AV(*svp);
6117 /* optimisation: store single backref directly in HvAUX or mg_obj */
6121 assert(SvTYPE(av) == SVt_PVAV);
6122 if (AvFILLp(av) >= AvMAX(av)) {
6123 av_extend(av, AvFILLp(av)+1);
6126 /* push new backref */
6127 AvARRAY(av)[++AvFILLp(av)] = sv; /* av_push() */
6130 /* delete a back-reference to ourselves from the backref magic associated
6131 * with the SV we point to.
6135 Perl_sv_del_backref(pTHX_ SV *const tsv, SV *const sv)
6139 PERL_ARGS_ASSERT_SV_DEL_BACKREF;
6141 if (SvTYPE(tsv) == SVt_PVHV) {
6143 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
6145 else if (SvIS_FREED(tsv) && PL_phase == PERL_PHASE_DESTRUCT) {
6146 /* It's possible for the the last (strong) reference to tsv to have
6147 become freed *before* the last thing holding a weak reference.
6148 If both survive longer than the backreferences array, then when
6149 the referent's reference count drops to 0 and it is freed, it's
6150 not able to chase the backreferences, so they aren't NULLed.
6152 For example, a CV holds a weak reference to its stash. If both the
6153 CV and the stash survive longer than the backreferences array,
6154 and the CV gets picked for the SvBREAK() treatment first,
6155 *and* it turns out that the stash is only being kept alive because
6156 of an our variable in the pad of the CV, then midway during CV
6157 destruction the stash gets freed, but CvSTASH() isn't set to NULL.
6158 It ends up pointing to the freed HV. Hence it's chased in here, and
6159 if this block wasn't here, it would hit the !svp panic just below.
6161 I don't believe that "better" destruction ordering is going to help
6162 here - during global destruction there's always going to be the
6163 chance that something goes out of order. We've tried to make it
6164 foolproof before, and it only resulted in evolutionary pressure on
6165 fools. Which made us look foolish for our hubris. :-(
6171 = SvMAGICAL(tsv) ? mg_find(tsv, PERL_MAGIC_backref) : NULL;
6172 svp = mg ? &(mg->mg_obj) : NULL;
6176 Perl_croak(aTHX_ "panic: del_backref, svp=0");
6178 /* It's possible that sv is being freed recursively part way through the
6179 freeing of tsv. If this happens, the backreferences array of tsv has
6180 already been freed, and so svp will be NULL. If this is the case,
6181 we should not panic. Instead, nothing needs doing, so return. */
6182 if (PL_phase == PERL_PHASE_DESTRUCT && SvREFCNT(tsv) == 0)
6184 Perl_croak(aTHX_ "panic: del_backref, *svp=%p phase=%s refcnt=%" UVuf,
6185 (void*)*svp, PL_phase_names[PL_phase], (UV)SvREFCNT(tsv));
6188 if (SvTYPE(*svp) == SVt_PVAV) {
6192 AV * const av = (AV*)*svp;
6194 assert(!SvIS_FREED(av));
6198 /* for an SV with N weak references to it, if all those
6199 * weak refs are deleted, then sv_del_backref will be called
6200 * N times and O(N^2) compares will be done within the backref
6201 * array. To ameliorate this potential slowness, we:
6202 * 1) make sure this code is as tight as possible;
6203 * 2) when looking for SV, look for it at both the head and tail of the
6204 * array first before searching the rest, since some create/destroy
6205 * patterns will cause the backrefs to be freed in order.
6212 SV **p = &svp[fill];
6213 SV *const topsv = *p;
6220 /* We weren't the last entry.
6221 An unordered list has this property that you
6222 can take the last element off the end to fill
6223 the hole, and it's still an unordered list :-)
6229 break; /* should only be one */
6236 AvFILLp(av) = fill-1;
6238 else if (SvIS_FREED(*svp) && PL_phase == PERL_PHASE_DESTRUCT) {
6239 /* freed AV; skip */
6242 /* optimisation: only a single backref, stored directly */
6244 Perl_croak(aTHX_ "panic: del_backref, *svp=%p, sv=%p",
6245 (void*)*svp, (void*)sv);
6252 Perl_sv_kill_backrefs(pTHX_ SV *const sv, AV *const av)
6258 PERL_ARGS_ASSERT_SV_KILL_BACKREFS;
6263 /* after multiple passes through Perl_sv_clean_all() for a thingy
6264 * that has badly leaked, the backref array may have gotten freed,
6265 * since we only protect it against 1 round of cleanup */
6266 if (SvIS_FREED(av)) {
6267 if (PL_in_clean_all) /* All is fair */
6270 "panic: magic_killbackrefs (freed backref AV/SV)");
6274 is_array = (SvTYPE(av) == SVt_PVAV);
6276 assert(!SvIS_FREED(av));
6279 last = svp + AvFILLp(av);
6282 /* optimisation: only a single backref, stored directly */
6288 while (svp <= last) {
6290 SV *const referrer = *svp;
6291 if (SvWEAKREF(referrer)) {
6292 /* XXX Should we check that it hasn't changed? */
6293 assert(SvROK(referrer));
6294 SvRV_set(referrer, 0);
6296 SvWEAKREF_off(referrer);
6297 SvSETMAGIC(referrer);
6298 } else if (SvTYPE(referrer) == SVt_PVGV ||
6299 SvTYPE(referrer) == SVt_PVLV) {
6300 assert(SvTYPE(sv) == SVt_PVHV); /* stash backref */
6301 /* You lookin' at me? */
6302 assert(GvSTASH(referrer));
6303 assert(GvSTASH(referrer) == (const HV *)sv);
6304 GvSTASH(referrer) = 0;
6305 } else if (SvTYPE(referrer) == SVt_PVCV ||
6306 SvTYPE(referrer) == SVt_PVFM) {
6307 if (SvTYPE(sv) == SVt_PVHV) { /* stash backref */
6308 /* You lookin' at me? */
6309 assert(CvSTASH(referrer));
6310 assert(CvSTASH(referrer) == (const HV *)sv);
6311 SvANY(MUTABLE_CV(referrer))->xcv_stash = 0;
6314 assert(SvTYPE(sv) == SVt_PVGV);
6315 /* You lookin' at me? */
6316 assert(CvGV(referrer));
6317 assert(CvGV(referrer) == (const GV *)sv);
6318 anonymise_cv_maybe(MUTABLE_GV(sv),
6319 MUTABLE_CV(referrer));
6324 "panic: magic_killbackrefs (flags=%" UVxf ")",
6325 (UV)SvFLAGS(referrer));
6336 SvREFCNT_dec_NN(av); /* remove extra count added by sv_add_backref() */
6342 =for apidoc sv_insert
6344 Inserts and/or replaces a string at the specified offset/length within the SV.
6345 Similar to the Perl C<substr()> function, with C<littlelen> bytes starting at
6346 C<little> replacing C<len> bytes of the string in C<bigstr> starting at
6347 C<offset>. Handles get magic.
6349 =for apidoc sv_insert_flags
6351 Same as C<sv_insert>, but the extra C<flags> are passed to the
6352 C<SvPV_force_flags> that applies to C<bigstr>.
6358 Perl_sv_insert_flags(pTHX_ SV *const bigstr, const STRLEN offset, const STRLEN len, const char *little, const STRLEN littlelen, const U32 flags)
6364 SSize_t i; /* better be sizeof(STRLEN) or bad things happen */
6367 PERL_ARGS_ASSERT_SV_INSERT_FLAGS;
6369 SvPV_force_flags(bigstr, curlen, flags);
6370 (void)SvPOK_only_UTF8(bigstr);
6372 if (little >= SvPVX(bigstr) &&
6373 little < SvPVX(bigstr) + (SvLEN(bigstr) ? SvLEN(bigstr) : SvCUR(bigstr))) {
6374 /* little is a pointer to within bigstr, since we can reallocate bigstr,
6375 or little...little+littlelen might overlap offset...offset+len we make a copy
6377 little = savepvn(little, littlelen);
6381 if (offset + len > curlen) {
6382 SvGROW(bigstr, offset+len+1);
6383 Zero(SvPVX(bigstr)+curlen, offset+len-curlen, char);
6384 SvCUR_set(bigstr, offset+len);
6388 i = littlelen - len;
6389 if (i > 0) { /* string might grow */
6390 big = SvGROW(bigstr, SvCUR(bigstr) + i + 1);
6391 mid = big + offset + len;
6392 midend = bigend = big + SvCUR(bigstr);
6395 while (midend > mid) /* shove everything down */
6396 *--bigend = *--midend;
6397 Move(little,big+offset,littlelen,char);
6398 SvCUR_set(bigstr, SvCUR(bigstr) + i);
6403 Move(little,SvPVX(bigstr)+offset,len,char);
6408 big = SvPVX(bigstr);
6411 bigend = big + SvCUR(bigstr);
6413 if (midend > bigend)
6414 Perl_croak(aTHX_ "panic: sv_insert, midend=%p, bigend=%p",
6417 if (mid - big > bigend - midend) { /* faster to shorten from end */
6419 Move(little, mid, littlelen,char);
6422 i = bigend - midend;
6424 Move(midend, mid, i,char);
6428 SvCUR_set(bigstr, mid - big);
6430 else if ((i = mid - big)) { /* faster from front */
6431 midend -= littlelen;
6433 Move(big, midend - i, i, char);
6434 sv_chop(bigstr,midend-i);
6436 Move(little, mid, littlelen,char);
6438 else if (littlelen) {
6439 midend -= littlelen;
6440 sv_chop(bigstr,midend);
6441 Move(little,midend,littlelen,char);
6444 sv_chop(bigstr,midend);
6450 =for apidoc sv_replace
6452 Make the first argument a copy of the second, then delete the original.
6453 The target SV physically takes over ownership of the body of the source SV
6454 and inherits its flags; however, the target keeps any magic it owns,
6455 and any magic in the source is discarded.
6456 Note that this is a rather specialist SV copying operation; most of the
6457 time you'll want to use C<sv_setsv> or one of its many macro front-ends.
6463 Perl_sv_replace(pTHX_ SV *const sv, SV *const nsv)
6465 const U32 refcnt = SvREFCNT(sv);
6467 PERL_ARGS_ASSERT_SV_REPLACE;
6469 SV_CHECK_THINKFIRST_COW_DROP(sv);
6470 if (SvREFCNT(nsv) != 1) {
6471 Perl_croak(aTHX_ "panic: reference miscount on nsv in sv_replace()"
6472 " (%" UVuf " != 1)", (UV) SvREFCNT(nsv));
6474 if (SvMAGICAL(sv)) {
6478 sv_upgrade(nsv, SVt_PVMG);
6479 SvMAGIC_set(nsv, SvMAGIC(sv));
6480 SvFLAGS(nsv) |= SvMAGICAL(sv);
6482 SvMAGIC_set(sv, NULL);
6486 assert(!SvREFCNT(sv));
6487 #ifdef DEBUG_LEAKING_SCALARS
6488 sv->sv_flags = nsv->sv_flags;
6489 sv->sv_any = nsv->sv_any;
6490 sv->sv_refcnt = nsv->sv_refcnt;
6491 sv->sv_u = nsv->sv_u;
6493 StructCopy(nsv,sv,SV);
6495 if(SvTYPE(sv) == SVt_IV) {
6496 SET_SVANY_FOR_BODYLESS_IV(sv);
6500 SvREFCNT(sv) = refcnt;
6501 SvFLAGS(nsv) |= SVTYPEMASK; /* Mark as freed */
6506 /* We're about to free a GV which has a CV that refers back to us.
6507 * If that CV will outlive us, make it anonymous (i.e. fix up its CvGV
6511 S_anonymise_cv_maybe(pTHX_ GV *gv, CV* cv)
6516 PERL_ARGS_ASSERT_ANONYMISE_CV_MAYBE;
6519 assert(SvREFCNT(gv) == 0);
6520 assert(isGV(gv) && isGV_with_GP(gv));
6522 assert(!CvANON(cv));
6523 assert(CvGV(cv) == gv);
6524 assert(!CvNAMED(cv));
6526 /* will the CV shortly be freed by gp_free() ? */
6527 if (GvCV(gv) == cv && GvGP(gv)->gp_refcnt < 2 && SvREFCNT(cv) < 2) {
6528 SvANY(cv)->xcv_gv_u.xcv_gv = NULL;
6532 /* if not, anonymise: */
6533 gvname = (GvSTASH(gv) && HvNAME(GvSTASH(gv)) && HvENAME(GvSTASH(gv)))
6534 ? newSVhek(HvENAME_HEK(GvSTASH(gv)))
6535 : newSVpvn_flags( "__ANON__", 8, 0 );
6536 sv_catpvs(gvname, "::__ANON__");
6537 anongv = gv_fetchsv(gvname, GV_ADDMULTI, SVt_PVCV);
6538 SvREFCNT_dec_NN(gvname);
6542 SvANY(cv)->xcv_gv_u.xcv_gv = MUTABLE_GV(SvREFCNT_inc(anongv));
6547 =for apidoc sv_clear
6549 Clear an SV: call any destructors, free up any memory used by the body,
6550 and free the body itself. The SV's head is I<not> freed, although
6551 its type is set to all 1's so that it won't inadvertently be assumed
6552 to be live during global destruction etc.
6553 This function should only be called when C<REFCNT> is zero. Most of the time
6554 you'll want to call C<sv_free()> (or its macro wrapper C<SvREFCNT_dec>)
6561 Perl_sv_clear(pTHX_ SV *const orig_sv)
6566 const struct body_details *sv_type_details;
6570 STRLEN hash_index = 0; /* initialise to make Coverity et al happy.
6571 Not strictly necessary */
6573 PERL_ARGS_ASSERT_SV_CLEAR;
6575 /* within this loop, sv is the SV currently being freed, and
6576 * iter_sv is the most recent AV or whatever that's being iterated
6577 * over to provide more SVs */
6583 assert(SvREFCNT(sv) == 0);
6584 assert(SvTYPE(sv) != (svtype)SVTYPEMASK);
6586 if (type <= SVt_IV) {
6587 /* See the comment in sv.h about the collusion between this
6588 * early return and the overloading of the NULL slots in the
6592 SvFLAGS(sv) &= SVf_BREAK;
6593 SvFLAGS(sv) |= SVTYPEMASK;
6597 /* objs are always >= MG, but pad names use the SVs_OBJECT flag
6598 for another purpose */
6599 assert(!SvOBJECT(sv) || type >= SVt_PVMG);
6601 if (type >= SVt_PVMG) {
6603 if (!curse(sv, 1)) goto get_next_sv;
6604 type = SvTYPE(sv); /* destructor may have changed it */
6606 /* Free back-references before magic, in case the magic calls
6607 * Perl code that has weak references to sv. */
6608 if (type == SVt_PVHV) {
6609 Perl_hv_kill_backrefs(aTHX_ MUTABLE_HV(sv));
6613 else if (SvMAGIC(sv)) {
6614 /* Free back-references before other types of magic. */
6615 sv_unmagic(sv, PERL_MAGIC_backref);
6621 /* case SVt_INVLIST: */
6624 IoIFP(sv) != PerlIO_stdin() &&
6625 IoIFP(sv) != PerlIO_stdout() &&
6626 IoIFP(sv) != PerlIO_stderr() &&
6627 !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6629 io_close(MUTABLE_IO(sv), NULL, FALSE,
6630 (IoTYPE(sv) == IoTYPE_WRONLY ||
6631 IoTYPE(sv) == IoTYPE_RDWR ||
6632 IoTYPE(sv) == IoTYPE_APPEND));
6634 if (IoDIRP(sv) && !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6635 PerlDir_close(IoDIRP(sv));
6636 IoDIRP(sv) = (DIR*)NULL;
6637 Safefree(IoTOP_NAME(sv));
6638 Safefree(IoFMT_NAME(sv));
6639 Safefree(IoBOTTOM_NAME(sv));
6640 if ((const GV *)sv == PL_statgv)
6644 /* FIXME for plugins */
6645 pregfree2((REGEXP*) sv);
6649 cv_undef(MUTABLE_CV(sv));
6650 /* If we're in a stash, we don't own a reference to it.
6651 * However it does have a back reference to us, which needs to
6653 if ((stash = CvSTASH(sv)))
6654 sv_del_backref(MUTABLE_SV(stash), sv);
6657 if (PL_last_swash_hv == (const HV *)sv) {
6658 PL_last_swash_hv = NULL;
6660 if (HvTOTALKEYS((HV*)sv) > 0) {
6662 /* this statement should match the one at the beginning of
6663 * hv_undef_flags() */
6664 if ( PL_phase != PERL_PHASE_DESTRUCT
6665 && (hek = HvNAME_HEK((HV*)sv)))
6667 if (PL_stashcache) {
6668 DEBUG_o(Perl_deb(aTHX_
6669 "sv_clear clearing PL_stashcache for '%" HEKf
6672 (void)hv_deletehek(PL_stashcache,
6675 hv_name_set((HV*)sv, NULL, 0, 0);
6678 /* save old iter_sv in unused SvSTASH field */
6679 assert(!SvOBJECT(sv));
6680 SvSTASH(sv) = (HV*)iter_sv;
6683 /* save old hash_index in unused SvMAGIC field */
6684 assert(!SvMAGICAL(sv));
6685 assert(!SvMAGIC(sv));
6686 ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index = hash_index;
6689 next_sv = Perl_hfree_next_entry(aTHX_ (HV*)sv, &hash_index);
6690 goto get_next_sv; /* process this new sv */
6692 /* free empty hash */
6693 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6694 assert(!HvARRAY((HV*)sv));
6698 AV* av = MUTABLE_AV(sv);
6699 if (PL_comppad == av) {
6703 if (AvREAL(av) && AvFILLp(av) > -1) {
6704 next_sv = AvARRAY(av)[AvFILLp(av)--];
6705 /* save old iter_sv in top-most slot of AV,
6706 * and pray that it doesn't get wiped in the meantime */
6707 AvARRAY(av)[AvMAX(av)] = iter_sv;
6709 goto get_next_sv; /* process this new sv */
6711 Safefree(AvALLOC(av));
6716 if (LvTYPE(sv) == 'T') { /* for tie: return HE to pool */
6717 SvREFCNT_dec(HeKEY_sv((HE*)LvTARG(sv)));
6718 HeNEXT((HE*)LvTARG(sv)) = PL_hv_fetch_ent_mh;
6719 PL_hv_fetch_ent_mh = (HE*)LvTARG(sv);
6721 else if (LvTYPE(sv) != 't') /* unless tie: unrefcnted fake SV** */
6722 SvREFCNT_dec(LvTARG(sv));
6724 /* SvLEN points to a regex body. Free the body, then
6725 * set SvLEN to whatever value was in the now-freed
6726 * regex body. The PVX buffer is shared by multiple re's
6727 * and only freed once, by the re whose len in non-null */
6728 STRLEN len = ReANY(sv)->xpv_len;
6729 pregfree2((REGEXP*) sv);
6730 SvLEN_set((sv), len);
6735 if (isGV_with_GP(sv)) {
6736 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
6737 && HvENAME_get(stash))
6738 mro_method_changed_in(stash);
6739 gp_free(MUTABLE_GV(sv));
6741 unshare_hek(GvNAME_HEK(sv));
6742 /* If we're in a stash, we don't own a reference to it.
6743 * However it does have a back reference to us, which
6744 * needs to be cleared. */
6745 if ((stash = GvSTASH(sv)))
6746 sv_del_backref(MUTABLE_SV(stash), sv);
6748 /* FIXME. There are probably more unreferenced pointers to SVs
6749 * in the interpreter struct that we should check and tidy in
6750 * a similar fashion to this: */
6751 /* See also S_sv_unglob, which does the same thing. */
6752 if ((const GV *)sv == PL_last_in_gv)
6753 PL_last_in_gv = NULL;
6754 else if ((const GV *)sv == PL_statgv)
6756 else if ((const GV *)sv == PL_stderrgv)
6765 /* Don't bother with SvOOK_off(sv); as we're only going to
6769 SvOOK_offset(sv, offset);
6770 SvPV_set(sv, SvPVX_mutable(sv) - offset);
6771 /* Don't even bother with turning off the OOK flag. */
6776 SV * const target = SvRV(sv);
6778 sv_del_backref(target, sv);
6784 else if (SvPVX_const(sv)
6785 && !(SvTYPE(sv) == SVt_PVIO
6786 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6791 PerlIO_printf(Perl_debug_log, "Copy on write: clear\n");
6796 if (CowREFCNT(sv)) {
6803 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6808 Safefree(SvPVX_mutable(sv));
6812 else if (SvPVX_const(sv) && SvLEN(sv)
6813 && !(SvTYPE(sv) == SVt_PVIO
6814 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6815 Safefree(SvPVX_mutable(sv));
6816 else if (SvPVX_const(sv) && SvIsCOW(sv)) {
6817 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6827 SvFLAGS(sv) &= SVf_BREAK;
6828 SvFLAGS(sv) |= SVTYPEMASK;
6830 sv_type_details = bodies_by_type + type;
6831 if (sv_type_details->arena) {
6832 del_body(((char *)SvANY(sv) + sv_type_details->offset),
6833 &PL_body_roots[type]);
6835 else if (sv_type_details->body_size) {
6836 safefree(SvANY(sv));
6840 /* caller is responsible for freeing the head of the original sv */
6841 if (sv != orig_sv && !SvREFCNT(sv))
6844 /* grab and free next sv, if any */
6852 else if (!iter_sv) {
6854 } else if (SvTYPE(iter_sv) == SVt_PVAV) {
6855 AV *const av = (AV*)iter_sv;
6856 if (AvFILLp(av) > -1) {
6857 sv = AvARRAY(av)[AvFILLp(av)--];
6859 else { /* no more elements of current AV to free */
6862 /* restore previous value, squirrelled away */
6863 iter_sv = AvARRAY(av)[AvMAX(av)];
6864 Safefree(AvALLOC(av));
6867 } else if (SvTYPE(iter_sv) == SVt_PVHV) {
6868 sv = Perl_hfree_next_entry(aTHX_ (HV*)iter_sv, &hash_index);
6869 if (!sv && !HvTOTALKEYS((HV *)iter_sv)) {
6870 /* no more elements of current HV to free */
6873 /* Restore previous values of iter_sv and hash_index,
6874 * squirrelled away */
6875 assert(!SvOBJECT(sv));
6876 iter_sv = (SV*)SvSTASH(sv);
6877 assert(!SvMAGICAL(sv));
6878 hash_index = ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index;
6880 /* perl -DA does not like rubbish in SvMAGIC. */
6884 /* free any remaining detritus from the hash struct */
6885 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6886 assert(!HvARRAY((HV*)sv));
6891 /* unrolled SvREFCNT_dec and sv_free2 follows: */
6895 if (!SvREFCNT(sv)) {
6899 if (--(SvREFCNT(sv)))
6903 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6904 "Attempt to free temp prematurely: SV 0x%" UVxf
6905 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6909 if (SvIMMORTAL(sv)) {
6910 /* make sure SvREFCNT(sv)==0 happens very seldom */
6911 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6920 /* This routine curses the sv itself, not the object referenced by sv. So
6921 sv does not have to be ROK. */
6924 S_curse(pTHX_ SV * const sv, const bool check_refcnt) {
6925 PERL_ARGS_ASSERT_CURSE;
6926 assert(SvOBJECT(sv));
6928 if (PL_defstash && /* Still have a symbol table? */
6934 stash = SvSTASH(sv);
6935 assert(SvTYPE(stash) == SVt_PVHV);
6936 if (HvNAME(stash)) {
6937 CV* destructor = NULL;
6938 struct mro_meta *meta;
6940 assert (SvOOK(stash));
6942 DEBUG_o( Perl_deb(aTHX_ "Looking for DESTROY method for %s\n",
6945 /* don't make this an initialization above the assert, since it needs
6947 meta = HvMROMETA(stash);
6948 if (meta->destroy_gen && meta->destroy_gen == PL_sub_generation) {
6949 destructor = meta->destroy;
6950 DEBUG_o( Perl_deb(aTHX_ "Using cached DESTROY method %p for %s\n",
6951 (void *)destructor, HvNAME(stash)) );
6954 bool autoload = FALSE;
6956 gv_fetchmeth_pvn(stash, S_destroy, S_destroy_len, -1, 0);
6958 destructor = GvCV(gv);
6960 gv = gv_autoload_pvn(stash, S_destroy, S_destroy_len,
6961 GV_AUTOLOAD_ISMETHOD);
6963 destructor = GvCV(gv);
6967 /* we don't cache AUTOLOAD for DESTROY, since this code
6968 would then need to set $__PACKAGE__::AUTOLOAD, or the
6969 equivalent for XS AUTOLOADs */
6971 meta->destroy_gen = PL_sub_generation;
6972 meta->destroy = destructor;
6974 DEBUG_o( Perl_deb(aTHX_ "Set cached DESTROY method %p for %s\n",
6975 (void *)destructor, HvNAME(stash)) );
6978 DEBUG_o( Perl_deb(aTHX_ "Not caching AUTOLOAD for DESTROY method for %s\n",
6982 assert(!destructor || SvTYPE(destructor) == SVt_PVCV);
6984 /* A constant subroutine can have no side effects, so
6985 don't bother calling it. */
6986 && !CvCONST(destructor)
6987 /* Don't bother calling an empty destructor or one that
6988 returns immediately. */
6989 && (CvISXSUB(destructor)
6990 || (CvSTART(destructor)
6991 && (CvSTART(destructor)->op_next->op_type
6993 && (CvSTART(destructor)->op_next->op_type
6995 || CvSTART(destructor)->op_next->op_next->op_type
7001 SV* const tmpref = newRV(sv);
7002 SvREADONLY_on(tmpref); /* DESTROY() could be naughty */
7004 PUSHSTACKi(PERLSI_DESTROY);
7009 call_sv(MUTABLE_SV(destructor),
7010 G_DISCARD|G_EVAL|G_KEEPERR|G_VOID);
7014 if(SvREFCNT(tmpref) < 2) {
7015 /* tmpref is not kept alive! */
7017 SvRV_set(tmpref, NULL);
7020 SvREFCNT_dec_NN(tmpref);
7023 } while (SvOBJECT(sv) && SvSTASH(sv) != stash);
7026 if (check_refcnt && SvREFCNT(sv)) {
7027 if (PL_in_clean_objs)
7029 "DESTROY created new reference to dead object '%" HEKf "'",
7030 HEKfARG(HvNAME_HEK(stash)));
7031 /* DESTROY gave object new lease on life */
7037 HV * const stash = SvSTASH(sv);
7038 /* Curse before freeing the stash, as freeing the stash could cause
7039 a recursive call into S_curse. */
7040 SvOBJECT_off(sv); /* Curse the object. */
7041 SvSTASH_set(sv,0); /* SvREFCNT_dec may try to read this */
7042 SvREFCNT_dec(stash); /* possibly of changed persuasion */
7048 =for apidoc sv_newref
7050 Increment an SV's reference count. Use the C<SvREFCNT_inc()> wrapper
7057 Perl_sv_newref(pTHX_ SV *const sv)
7059 PERL_UNUSED_CONTEXT;
7068 Decrement an SV's reference count, and if it drops to zero, call
7069 C<sv_clear> to invoke destructors and free up any memory used by
7070 the body; finally, deallocating the SV's head itself.
7071 Normally called via a wrapper macro C<SvREFCNT_dec>.
7077 Perl_sv_free(pTHX_ SV *const sv)
7083 /* Private helper function for SvREFCNT_dec().
7084 * Called with rc set to original SvREFCNT(sv), where rc == 0 or 1 */
7087 Perl_sv_free2(pTHX_ SV *const sv, const U32 rc)
7091 PERL_ARGS_ASSERT_SV_FREE2;
7093 if (LIKELY( rc == 1 )) {
7099 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
7100 "Attempt to free temp prematurely: SV 0x%" UVxf
7101 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
7105 if (SvIMMORTAL(sv)) {
7106 /* make sure SvREFCNT(sv)==0 happens very seldom */
7107 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
7111 if (! SvREFCNT(sv)) /* may have have been resurrected */
7116 /* handle exceptional cases */
7120 if (SvFLAGS(sv) & SVf_BREAK)
7121 /* this SV's refcnt has been artificially decremented to
7122 * trigger cleanup */
7124 if (PL_in_clean_all) /* All is fair */
7126 if (SvIMMORTAL(sv)) {
7127 /* make sure SvREFCNT(sv)==0 happens very seldom */
7128 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
7131 if (ckWARN_d(WARN_INTERNAL)) {
7132 #ifdef DEBUG_LEAKING_SCALARS_FORK_DUMP
7133 Perl_dump_sv_child(aTHX_ sv);
7135 #ifdef DEBUG_LEAKING_SCALARS
7138 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7139 if (PL_warnhook == PERL_WARNHOOK_FATAL
7140 || ckDEAD(packWARN(WARN_INTERNAL))) {
7141 /* Don't let Perl_warner cause us to escape our fate: */
7145 /* This may not return: */
7146 Perl_warner(aTHX_ packWARN(WARN_INTERNAL),
7147 "Attempt to free unreferenced scalar: SV 0x%" UVxf
7148 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
7151 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7161 Returns the length of the string in the SV. Handles magic and type
7162 coercion and sets the UTF8 flag appropriately. See also C<L</SvCUR>>, which
7163 gives raw access to the C<xpv_cur> slot.
7169 Perl_sv_len(pTHX_ SV *const sv)
7176 (void)SvPV_const(sv, len);
7181 =for apidoc sv_len_utf8
7183 Returns the number of characters in the string in an SV, counting wide
7184 UTF-8 bytes as a single character. Handles magic and type coercion.
7190 * The length is cached in PERL_MAGIC_utf8, in the mg_len field. Also the
7191 * mg_ptr is used, by sv_pos_u2b() and sv_pos_b2u() - see the comments below.
7192 * (Note that the mg_len is not the length of the mg_ptr field.
7193 * This allows the cache to store the character length of the string without
7194 * needing to malloc() extra storage to attach to the mg_ptr.)
7199 Perl_sv_len_utf8(pTHX_ SV *const sv)
7205 return sv_len_utf8_nomg(sv);
7209 Perl_sv_len_utf8_nomg(pTHX_ SV * const sv)
7212 const U8 *s = (U8*)SvPV_nomg_const(sv, len);
7214 PERL_ARGS_ASSERT_SV_LEN_UTF8_NOMG;
7216 if (PL_utf8cache && SvUTF8(sv)) {
7218 MAGIC *mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_utf8) : NULL;
7220 if (mg && (mg->mg_len != -1 || mg->mg_ptr)) {
7221 if (mg->mg_len != -1)
7224 /* We can use the offset cache for a headstart.
7225 The longer value is stored in the first pair. */
7226 STRLEN *cache = (STRLEN *) mg->mg_ptr;
7228 ulen = cache[0] + Perl_utf8_length(aTHX_ s + cache[1],
7232 if (PL_utf8cache < 0) {
7233 const STRLEN real = Perl_utf8_length(aTHX_ s, s + len);
7234 assert_uft8_cache_coherent("sv_len_utf8", ulen, real, sv);
7238 ulen = Perl_utf8_length(aTHX_ s, s + len);
7239 utf8_mg_len_cache_update(sv, &mg, ulen);
7243 return SvUTF8(sv) ? Perl_utf8_length(aTHX_ s, s + len) : len;
7246 /* Walk forwards to find the byte corresponding to the passed in UTF-8
7249 S_sv_pos_u2b_forwards(const U8 *const start, const U8 *const send,
7250 STRLEN *const uoffset_p, bool *const at_end)
7252 const U8 *s = start;
7253 STRLEN uoffset = *uoffset_p;
7255 PERL_ARGS_ASSERT_SV_POS_U2B_FORWARDS;
7257 while (s < send && uoffset) {
7264 else if (s > send) {
7266 /* This is the existing behaviour. Possibly it should be a croak, as
7267 it's actually a bounds error */
7270 *uoffset_p -= uoffset;
7274 /* Given the length of the string in both bytes and UTF-8 characters, decide
7275 whether to walk forwards or backwards to find the byte corresponding to
7276 the passed in UTF-8 offset. */
7278 S_sv_pos_u2b_midway(const U8 *const start, const U8 *send,
7279 STRLEN uoffset, const STRLEN uend)
7281 STRLEN backw = uend - uoffset;
7283 PERL_ARGS_ASSERT_SV_POS_U2B_MIDWAY;
7285 if (uoffset < 2 * backw) {
7286 /* The assumption is that going forwards is twice the speed of going
7287 forward (that's where the 2 * backw comes from).
7288 (The real figure of course depends on the UTF-8 data.) */
7289 const U8 *s = start;
7291 while (s < send && uoffset--)
7301 while (UTF8_IS_CONTINUATION(*send))
7304 return send - start;
7307 /* For the string representation of the given scalar, find the byte
7308 corresponding to the passed in UTF-8 offset. uoffset0 and boffset0
7309 give another position in the string, *before* the sought offset, which
7310 (which is always true, as 0, 0 is a valid pair of positions), which should
7311 help reduce the amount of linear searching.
7312 If *mgp is non-NULL, it should point to the UTF-8 cache magic, which
7313 will be used to reduce the amount of linear searching. The cache will be
7314 created if necessary, and the found value offered to it for update. */
7316 S_sv_pos_u2b_cached(pTHX_ SV *const sv, MAGIC **const mgp, const U8 *const start,
7317 const U8 *const send, STRLEN uoffset,
7318 STRLEN uoffset0, STRLEN boffset0)
7320 STRLEN boffset = 0; /* Actually always set, but let's keep gcc happy. */
7322 bool at_end = FALSE;
7324 PERL_ARGS_ASSERT_SV_POS_U2B_CACHED;
7326 assert (uoffset >= uoffset0);
7331 if (!SvREADONLY(sv) && !SvGMAGICAL(sv) && SvPOK(sv)
7333 && (*mgp || (SvTYPE(sv) >= SVt_PVMG &&
7334 (*mgp = mg_find(sv, PERL_MAGIC_utf8))))) {
7335 if ((*mgp)->mg_ptr) {
7336 STRLEN *cache = (STRLEN *) (*mgp)->mg_ptr;
7337 if (cache[0] == uoffset) {
7338 /* An exact match. */
7341 if (cache[2] == uoffset) {
7342 /* An exact match. */
7346 if (cache[0] < uoffset) {
7347 /* The cache already knows part of the way. */
7348 if (cache[0] > uoffset0) {
7349 /* The cache knows more than the passed in pair */
7350 uoffset0 = cache[0];
7351 boffset0 = cache[1];
7353 if ((*mgp)->mg_len != -1) {
7354 /* And we know the end too. */
7356 + sv_pos_u2b_midway(start + boffset0, send,
7358 (*mgp)->mg_len - uoffset0);
7360 uoffset -= uoffset0;
7362 + sv_pos_u2b_forwards(start + boffset0,
7363 send, &uoffset, &at_end);
7364 uoffset += uoffset0;
7367 else if (cache[2] < uoffset) {
7368 /* We're between the two cache entries. */
7369 if (cache[2] > uoffset0) {
7370 /* and the cache knows more than the passed in pair */
7371 uoffset0 = cache[2];
7372 boffset0 = cache[3];
7376 + sv_pos_u2b_midway(start + boffset0,
7379 cache[0] - uoffset0);
7382 + sv_pos_u2b_midway(start + boffset0,
7385 cache[2] - uoffset0);
7389 else if ((*mgp)->mg_len != -1) {
7390 /* If we can take advantage of a passed in offset, do so. */
7391 /* In fact, offset0 is either 0, or less than offset, so don't
7392 need to worry about the other possibility. */
7394 + sv_pos_u2b_midway(start + boffset0, send,
7396 (*mgp)->mg_len - uoffset0);
7401 if (!found || PL_utf8cache < 0) {
7402 STRLEN real_boffset;
7403 uoffset -= uoffset0;
7404 real_boffset = boffset0 + sv_pos_u2b_forwards(start + boffset0,
7405 send, &uoffset, &at_end);
7406 uoffset += uoffset0;
7408 if (found && PL_utf8cache < 0)
7409 assert_uft8_cache_coherent("sv_pos_u2b_cache", boffset,
7411 boffset = real_boffset;
7414 if (PL_utf8cache && !SvGMAGICAL(sv) && SvPOK(sv)) {
7416 utf8_mg_len_cache_update(sv, mgp, uoffset);
7418 utf8_mg_pos_cache_update(sv, mgp, boffset, uoffset, send - start);
7425 =for apidoc sv_pos_u2b_flags
7427 Converts the offset from a count of UTF-8 chars from
7428 the start of the string, to a count of the equivalent number of bytes; if
7429 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7430 C<offset>, rather than from the start
7431 of the string. Handles type coercion.
7432 C<flags> is passed to C<SvPV_flags>, and usually should be
7433 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7439 * sv_pos_u2b_flags() uses, like sv_pos_b2u(), the mg_ptr of the potential
7440 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7441 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7446 Perl_sv_pos_u2b_flags(pTHX_ SV *const sv, STRLEN uoffset, STRLEN *const lenp,
7453 PERL_ARGS_ASSERT_SV_POS_U2B_FLAGS;
7455 start = (U8*)SvPV_flags(sv, len, flags);
7457 const U8 * const send = start + len;
7459 boffset = sv_pos_u2b_cached(sv, &mg, start, send, uoffset, 0, 0);
7462 && *lenp /* don't bother doing work for 0, as its bytes equivalent
7463 is 0, and *lenp is already set to that. */) {
7464 /* Convert the relative offset to absolute. */
7465 const STRLEN uoffset2 = uoffset + *lenp;
7466 const STRLEN boffset2
7467 = sv_pos_u2b_cached(sv, &mg, start, send, uoffset2,
7468 uoffset, boffset) - boffset;
7482 =for apidoc sv_pos_u2b
7484 Converts the value pointed to by C<offsetp> from a count of UTF-8 chars from
7485 the start of the string, to a count of the equivalent number of bytes; if
7486 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7487 the offset, rather than from the start of the string. Handles magic and
7490 Use C<sv_pos_u2b_flags> in preference, which correctly handles strings longer
7497 * sv_pos_u2b() uses, like sv_pos_b2u(), the mg_ptr of the potential
7498 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7499 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7503 /* This function is subject to size and sign problems */
7506 Perl_sv_pos_u2b(pTHX_ SV *const sv, I32 *const offsetp, I32 *const lenp)
7508 PERL_ARGS_ASSERT_SV_POS_U2B;
7511 STRLEN ulen = (STRLEN)*lenp;
7512 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, &ulen,
7513 SV_GMAGIC|SV_CONST_RETURN);
7516 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, NULL,
7517 SV_GMAGIC|SV_CONST_RETURN);
7522 S_utf8_mg_len_cache_update(pTHX_ SV *const sv, MAGIC **const mgp,
7525 PERL_ARGS_ASSERT_UTF8_MG_LEN_CACHE_UPDATE;
7526 if (SvREADONLY(sv) || SvGMAGICAL(sv) || !SvPOK(sv))
7529 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7530 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7531 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, &PL_vtbl_utf8, 0, 0);
7535 (*mgp)->mg_len = ulen;
7538 /* Create and update the UTF8 magic offset cache, with the proffered utf8/
7539 byte length pairing. The (byte) length of the total SV is passed in too,
7540 as blen, because for some (more esoteric) SVs, the call to SvPV_const()
7541 may not have updated SvCUR, so we can't rely on reading it directly.
7543 The proffered utf8/byte length pairing isn't used if the cache already has
7544 two pairs, and swapping either for the proffered pair would increase the
7545 RMS of the intervals between known byte offsets.
7547 The cache itself consists of 4 STRLEN values
7548 0: larger UTF-8 offset
7549 1: corresponding byte offset
7550 2: smaller UTF-8 offset
7551 3: corresponding byte offset
7553 Unused cache pairs have the value 0, 0.
7554 Keeping the cache "backwards" means that the invariant of
7555 cache[0] >= cache[2] is maintained even with empty slots, which means that
7556 the code that uses it doesn't need to worry if only 1 entry has actually
7557 been set to non-zero. It also makes the "position beyond the end of the
7558 cache" logic much simpler, as the first slot is always the one to start
7562 S_utf8_mg_pos_cache_update(pTHX_ SV *const sv, MAGIC **const mgp, const STRLEN byte,
7563 const STRLEN utf8, const STRLEN blen)
7567 PERL_ARGS_ASSERT_UTF8_MG_POS_CACHE_UPDATE;
7572 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7573 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7574 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, (MGVTBL*)&PL_vtbl_utf8, 0,
7576 (*mgp)->mg_len = -1;
7580 if (!(cache = (STRLEN *)(*mgp)->mg_ptr)) {
7581 Newxz(cache, PERL_MAGIC_UTF8_CACHESIZE * 2, STRLEN);
7582 (*mgp)->mg_ptr = (char *) cache;
7586 if (PL_utf8cache < 0 && SvPOKp(sv)) {
7587 /* SvPOKp() because, if sv is a reference, then SvPVX() is actually
7588 a pointer. Note that we no longer cache utf8 offsets on refer-
7589 ences, but this check is still a good idea, for robustness. */
7590 const U8 *start = (const U8 *) SvPVX_const(sv);
7591 const STRLEN realutf8 = utf8_length(start, start + byte);
7593 assert_uft8_cache_coherent("utf8_mg_pos_cache_update", utf8, realutf8,
7597 /* Cache is held with the later position first, to simplify the code
7598 that deals with unbounded ends. */
7600 ASSERT_UTF8_CACHE(cache);
7601 if (cache[1] == 0) {
7602 /* Cache is totally empty */
7605 } else if (cache[3] == 0) {
7606 if (byte > cache[1]) {
7607 /* New one is larger, so goes first. */
7608 cache[2] = cache[0];
7609 cache[3] = cache[1];
7617 /* float casts necessary? XXX */
7618 #define THREEWAY_SQUARE(a,b,c,d) \
7619 ((float)((d) - (c))) * ((float)((d) - (c))) \
7620 + ((float)((c) - (b))) * ((float)((c) - (b))) \
7621 + ((float)((b) - (a))) * ((float)((b) - (a)))
7623 /* Cache has 2 slots in use, and we know three potential pairs.
7624 Keep the two that give the lowest RMS distance. Do the
7625 calculation in bytes simply because we always know the byte
7626 length. squareroot has the same ordering as the positive value,
7627 so don't bother with the actual square root. */
7628 if (byte > cache[1]) {
7629 /* New position is after the existing pair of pairs. */
7630 const float keep_earlier
7631 = THREEWAY_SQUARE(0, cache[3], byte, blen);
7632 const float keep_later
7633 = THREEWAY_SQUARE(0, cache[1], byte, blen);
7635 if (keep_later < keep_earlier) {
7636 cache[2] = cache[0];
7637 cache[3] = cache[1];
7643 const float keep_later = THREEWAY_SQUARE(0, byte, cache[1], blen);
7644 float b, c, keep_earlier;
7645 if (byte > cache[3]) {
7646 /* New position is between the existing pair of pairs. */
7647 b = (float)cache[3];
7650 /* New position is before the existing pair of pairs. */
7652 c = (float)cache[3];
7654 keep_earlier = THREEWAY_SQUARE(0, b, c, blen);
7655 if (byte > cache[3]) {
7656 if (keep_later < keep_earlier) {
7666 if (! (keep_later < keep_earlier)) {
7667 cache[0] = cache[2];
7668 cache[1] = cache[3];
7675 ASSERT_UTF8_CACHE(cache);
7678 /* We already know all of the way, now we may be able to walk back. The same
7679 assumption is made as in S_sv_pos_u2b_midway(), namely that walking
7680 backward is half the speed of walking forward. */
7682 S_sv_pos_b2u_midway(pTHX_ const U8 *const s, const U8 *const target,
7683 const U8 *end, STRLEN endu)
7685 const STRLEN forw = target - s;
7686 STRLEN backw = end - target;
7688 PERL_ARGS_ASSERT_SV_POS_B2U_MIDWAY;
7690 if (forw < 2 * backw) {
7691 return utf8_length(s, target);
7694 while (end > target) {
7696 while (UTF8_IS_CONTINUATION(*end)) {
7705 =for apidoc sv_pos_b2u_flags
7707 Converts C<offset> from a count of bytes from the start of the string, to
7708 a count of the equivalent number of UTF-8 chars. Handles type coercion.
7709 C<flags> is passed to C<SvPV_flags>, and usually should be
7710 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7716 * sv_pos_b2u_flags() uses, like sv_pos_u2b_flags(), the mg_ptr of the
7717 * potential PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8
7722 Perl_sv_pos_b2u_flags(pTHX_ SV *const sv, STRLEN const offset, U32 flags)
7725 STRLEN len = 0; /* Actually always set, but let's keep gcc happy. */
7731 PERL_ARGS_ASSERT_SV_POS_B2U_FLAGS;
7733 s = (const U8*)SvPV_flags(sv, blen, flags);
7736 Perl_croak(aTHX_ "panic: sv_pos_b2u: bad byte offset, blen=%" UVuf
7737 ", byte=%" UVuf, (UV)blen, (UV)offset);
7743 && SvTYPE(sv) >= SVt_PVMG
7744 && (mg = mg_find(sv, PERL_MAGIC_utf8)))
7747 STRLEN * const cache = (STRLEN *) mg->mg_ptr;
7748 if (cache[1] == offset) {
7749 /* An exact match. */
7752 if (cache[3] == offset) {
7753 /* An exact match. */
7757 if (cache[1] < offset) {
7758 /* We already know part of the way. */
7759 if (mg->mg_len != -1) {
7760 /* Actually, we know the end too. */
7762 + S_sv_pos_b2u_midway(aTHX_ s + cache[1], send,
7763 s + blen, mg->mg_len - cache[0]);
7765 len = cache[0] + utf8_length(s + cache[1], send);
7768 else if (cache[3] < offset) {
7769 /* We're between the two cached pairs, so we do the calculation
7770 offset by the byte/utf-8 positions for the earlier pair,
7771 then add the utf-8 characters from the string start to
7773 len = S_sv_pos_b2u_midway(aTHX_ s + cache[3], send,
7774 s + cache[1], cache[0] - cache[2])
7778 else { /* cache[3] > offset */
7779 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + cache[3],
7783 ASSERT_UTF8_CACHE(cache);
7785 } else if (mg->mg_len != -1) {
7786 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + blen, mg->mg_len);
7790 if (!found || PL_utf8cache < 0) {
7791 const STRLEN real_len = utf8_length(s, send);
7793 if (found && PL_utf8cache < 0)
7794 assert_uft8_cache_coherent("sv_pos_b2u", len, real_len, sv);
7800 utf8_mg_len_cache_update(sv, &mg, len);
7802 utf8_mg_pos_cache_update(sv, &mg, offset, len, blen);
7809 =for apidoc sv_pos_b2u
7811 Converts the value pointed to by C<offsetp> from a count of bytes from the
7812 start of the string, to a count of the equivalent number of UTF-8 chars.
7813 Handles magic and type coercion.
7815 Use C<sv_pos_b2u_flags> in preference, which correctly handles strings
7822 * sv_pos_b2u() uses, like sv_pos_u2b(), the mg_ptr of the potential
7823 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7828 Perl_sv_pos_b2u(pTHX_ SV *const sv, I32 *const offsetp)
7830 PERL_ARGS_ASSERT_SV_POS_B2U;
7835 *offsetp = (I32)sv_pos_b2u_flags(sv, (STRLEN)*offsetp,
7836 SV_GMAGIC|SV_CONST_RETURN);
7840 S_assert_uft8_cache_coherent(pTHX_ const char *const func, STRLEN from_cache,
7841 STRLEN real, SV *const sv)
7843 PERL_ARGS_ASSERT_ASSERT_UFT8_CACHE_COHERENT;
7845 /* As this is debugging only code, save space by keeping this test here,
7846 rather than inlining it in all the callers. */
7847 if (from_cache == real)
7850 /* Need to turn the assertions off otherwise we may recurse infinitely
7851 while printing error messages. */
7852 SAVEI8(PL_utf8cache);
7854 Perl_croak(aTHX_ "panic: %s cache %" UVuf " real %" UVuf " for %" SVf,
7855 func, (UV) from_cache, (UV) real, SVfARG(sv));
7861 Returns a boolean indicating whether the strings in the two SVs are
7862 identical. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7863 coerce its args to strings if necessary.
7865 =for apidoc sv_eq_flags
7867 Returns a boolean indicating whether the strings in the two SVs are
7868 identical. Is UTF-8 and S<C<'use bytes'>> aware and coerces its args to strings
7869 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get-magic, too.
7875 Perl_sv_eq_flags(pTHX_ SV *sv1, SV *sv2, const U32 flags)
7887 /* if pv1 and pv2 are the same, second SvPV_const call may
7888 * invalidate pv1 (if we are handling magic), so we may need to
7890 if (sv1 == sv2 && flags & SV_GMAGIC
7891 && (SvTHINKFIRST(sv1) || SvGMAGICAL(sv1))) {
7892 pv1 = SvPV_const(sv1, cur1);
7893 sv1 = newSVpvn_flags(pv1, cur1, SVs_TEMP | SvUTF8(sv2));
7895 pv1 = SvPV_flags_const(sv1, cur1, flags);
7903 pv2 = SvPV_flags_const(sv2, cur2, flags);
7905 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7906 /* Differing utf8ness. */
7908 /* sv1 is the UTF-8 one */
7909 return bytes_cmp_utf8((const U8*)pv2, cur2,
7910 (const U8*)pv1, cur1) == 0;
7913 /* sv2 is the UTF-8 one */
7914 return bytes_cmp_utf8((const U8*)pv1, cur1,
7915 (const U8*)pv2, cur2) == 0;
7920 return (pv1 == pv2) || memEQ(pv1, pv2, cur1);
7928 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7929 string in C<sv1> is less than, equal to, or greater than the string in
7930 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7931 coerce its args to strings if necessary. See also C<L</sv_cmp_locale>>.
7933 =for apidoc sv_cmp_flags
7935 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7936 string in C<sv1> is less than, equal to, or greater than the string in
7937 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware and will coerce its args to strings
7938 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get magic. See
7939 also C<L</sv_cmp_locale_flags>>.
7945 Perl_sv_cmp(pTHX_ SV *const sv1, SV *const sv2)
7947 return sv_cmp_flags(sv1, sv2, SV_GMAGIC);
7951 Perl_sv_cmp_flags(pTHX_ SV *const sv1, SV *const sv2,
7955 const char *pv1, *pv2;
7957 SV *svrecode = NULL;
7964 pv1 = SvPV_flags_const(sv1, cur1, flags);
7971 pv2 = SvPV_flags_const(sv2, cur2, flags);
7973 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7974 /* Differing utf8ness. */
7976 const int retval = -bytes_cmp_utf8((const U8*)pv2, cur2,
7977 (const U8*)pv1, cur1);
7978 return retval ? retval < 0 ? -1 : +1 : 0;
7981 const int retval = bytes_cmp_utf8((const U8*)pv1, cur1,
7982 (const U8*)pv2, cur2);
7983 return retval ? retval < 0 ? -1 : +1 : 0;
7987 /* Here, if both are non-NULL, then they have the same UTF8ness. */
7990 cmp = cur2 ? -1 : 0;
7994 STRLEN shortest_len = cur1 < cur2 ? cur1 : cur2;
7997 if (! DO_UTF8(sv1)) {
7999 const I32 retval = memcmp((const void*)pv1,
8003 cmp = retval < 0 ? -1 : 1;
8004 } else if (cur1 == cur2) {
8007 cmp = cur1 < cur2 ? -1 : 1;
8011 else { /* Both are to be treated as UTF-EBCDIC */
8013 /* EBCDIC UTF-8 is complicated by the fact that it is based on I8
8014 * which remaps code points 0-255. We therefore generally have to
8015 * unmap back to the original values to get an accurate comparison.
8016 * But we don't have to do that for UTF-8 invariants, as by
8017 * definition, they aren't remapped, nor do we have to do it for
8018 * above-latin1 code points, as they also aren't remapped. (This
8019 * code also works on ASCII platforms, but the memcmp() above is
8022 const char *e = pv1 + shortest_len;
8024 /* Find the first bytes that differ between the two strings */
8025 while (pv1 < e && *pv1 == *pv2) {
8031 if (pv1 == e) { /* Are the same all the way to the end */
8035 cmp = cur1 < cur2 ? -1 : 1;
8038 else /* Here *pv1 and *pv2 are not equal, but all bytes earlier
8039 * in the strings were. The current bytes may or may not be
8040 * at the beginning of a character. But neither or both are
8041 * (or else earlier bytes would have been different). And
8042 * if we are in the middle of a character, the two
8043 * characters are comprised of the same number of bytes
8044 * (because in this case the start bytes are the same, and
8045 * the start bytes encode the character's length). */
8046 if (UTF8_IS_INVARIANT(*pv1))
8048 /* If both are invariants; can just compare directly */
8049 if (UTF8_IS_INVARIANT(*pv2)) {
8050 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8052 else /* Since *pv1 is invariant, it is the whole character,
8053 which means it is at the beginning of a character.
8054 That means pv2 is also at the beginning of a
8055 character (see earlier comment). Since it isn't
8056 invariant, it must be a start byte. If it starts a
8057 character whose code point is above 255, that
8058 character is greater than any single-byte char, which
8060 if (UTF8_IS_ABOVE_LATIN1_START(*pv2))
8065 /* Here, pv2 points to a character composed of 2 bytes
8066 * whose code point is < 256. Get its code point and
8067 * compare with *pv1 */
8068 cmp = ((U8) *pv1 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
8073 else /* The code point starting at pv1 isn't a single byte */
8074 if (UTF8_IS_INVARIANT(*pv2))
8076 /* But here, the code point starting at *pv2 is a single byte,
8077 * and so *pv1 must begin a character, hence is a start byte.
8078 * If that character is above 255, it is larger than any
8079 * single-byte char, which *pv2 is */
8080 if (UTF8_IS_ABOVE_LATIN1_START(*pv1)) {
8084 /* Here, pv1 points to a character composed of 2 bytes
8085 * whose code point is < 256. Get its code point and
8086 * compare with the single byte character *pv2 */
8087 cmp = (EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1)) < (U8) *pv2)
8092 else /* Here, we've ruled out either *pv1 and *pv2 being
8093 invariant. That means both are part of variants, but not
8094 necessarily at the start of a character */
8095 if ( UTF8_IS_ABOVE_LATIN1_START(*pv1)
8096 || UTF8_IS_ABOVE_LATIN1_START(*pv2))
8098 /* Here, at least one is the start of a character, which means
8099 * the other is also a start byte. And the code point of at
8100 * least one of the characters is above 255. It is a
8101 * characteristic of UTF-EBCDIC that all start bytes for
8102 * above-latin1 code points are well behaved as far as code
8103 * point comparisons go, and all are larger than all other
8104 * start bytes, so the comparison with those is also well
8106 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8109 /* Here both *pv1 and *pv2 are part of variant characters.
8110 * They could be both continuations, or both start characters.
8111 * (One or both could even be an illegal start character (for
8112 * an overlong) which for the purposes of sorting we treat as
8114 if (UTF8_IS_CONTINUATION(*pv1)) {
8116 /* If they are continuations for code points above 255,
8117 * then comparing the current byte is sufficient, as there
8118 * is no remapping of these and so the comparison is
8119 * well-behaved. We determine if they are such
8120 * continuations by looking at the preceding byte. It
8121 * could be a start byte, from which we can tell if it is
8122 * for an above 255 code point. Or it could be a
8123 * continuation, which means the character occupies at
8124 * least 3 bytes, so must be above 255. */
8125 if ( UTF8_IS_CONTINUATION(*(pv2 - 1))
8126 || UTF8_IS_ABOVE_LATIN1_START(*(pv2 -1)))
8128 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8132 /* Here, the continuations are for code points below 256;
8133 * back up one to get to the start byte */
8138 /* We need to get the actual native code point of each of these
8139 * variants in order to compare them */
8140 cmp = ( EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1))
8141 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
8150 SvREFCNT_dec(svrecode);
8156 =for apidoc sv_cmp_locale
8158 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
8159 S<C<'use bytes'>> aware, handles get magic, and will coerce its args to strings
8160 if necessary. See also C<L</sv_cmp>>.
8162 =for apidoc sv_cmp_locale_flags
8164 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
8165 S<C<'use bytes'>> aware and will coerce its args to strings if necessary. If
8166 the flags contain C<SV_GMAGIC>, it handles get magic. See also
8167 C<L</sv_cmp_flags>>.
8173 Perl_sv_cmp_locale(pTHX_ SV *const sv1, SV *const sv2)
8175 return sv_cmp_locale_flags(sv1, sv2, SV_GMAGIC);
8179 Perl_sv_cmp_locale_flags(pTHX_ SV *const sv1, SV *const sv2,
8182 #ifdef USE_LOCALE_COLLATE
8188 if (PL_collation_standard)
8193 /* Revert to using raw compare if both operands exist, but either one
8194 * doesn't transform properly for collation */
8196 pv1 = sv_collxfrm_flags(sv1, &len1, flags);
8200 pv2 = sv_collxfrm_flags(sv2, &len2, flags);
8206 pv1 = sv1 ? sv_collxfrm_flags(sv1, &len1, flags) : (char *) NULL;
8207 pv2 = sv2 ? sv_collxfrm_flags(sv2, &len2, flags) : (char *) NULL;
8210 if (!pv1 || !len1) {
8221 retval = memcmp((void*)pv1, (void*)pv2, len1 < len2 ? len1 : len2);
8224 return retval < 0 ? -1 : 1;
8227 * When the result of collation is equality, that doesn't mean
8228 * that there are no differences -- some locales exclude some
8229 * characters from consideration. So to avoid false equalities,
8230 * we use the raw string as a tiebreaker.
8237 PERL_UNUSED_ARG(flags);
8238 #endif /* USE_LOCALE_COLLATE */
8240 return sv_cmp(sv1, sv2);
8244 #ifdef USE_LOCALE_COLLATE
8247 =for apidoc sv_collxfrm
8249 This calls C<sv_collxfrm_flags> with the SV_GMAGIC flag. See
8250 C<L</sv_collxfrm_flags>>.
8252 =for apidoc sv_collxfrm_flags
8254 Add Collate Transform magic to an SV if it doesn't already have it. If the
8255 flags contain C<SV_GMAGIC>, it handles get-magic.
8257 Any scalar variable may carry C<PERL_MAGIC_collxfrm> magic that contains the
8258 scalar data of the variable, but transformed to such a format that a normal
8259 memory comparison can be used to compare the data according to the locale
8266 Perl_sv_collxfrm_flags(pTHX_ SV *const sv, STRLEN *const nxp, const I32 flags)
8270 PERL_ARGS_ASSERT_SV_COLLXFRM_FLAGS;
8272 mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_collxfrm) : (MAGIC *) NULL;
8274 /* If we don't have collation magic on 'sv', or the locale has changed
8275 * since the last time we calculated it, get it and save it now */
8276 if (!mg || !mg->mg_ptr || *(U32*)mg->mg_ptr != PL_collation_ix) {
8281 /* Free the old space */
8283 Safefree(mg->mg_ptr);
8285 s = SvPV_flags_const(sv, len, flags);
8286 if ((xf = _mem_collxfrm(s, len, &xlen, cBOOL(SvUTF8(sv))))) {
8288 mg = sv_magicext(sv, 0, PERL_MAGIC_collxfrm, &PL_vtbl_collxfrm,
8303 if (mg && mg->mg_ptr) {
8305 return mg->mg_ptr + sizeof(PL_collation_ix);
8313 #endif /* USE_LOCALE_COLLATE */
8316 S_sv_gets_append_to_utf8(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8318 SV * const tsv = newSV(0);
8321 sv_gets(tsv, fp, 0);
8322 sv_utf8_upgrade_nomg(tsv);
8323 SvCUR_set(sv,append);
8326 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8330 S_sv_gets_read_record(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8333 const STRLEN recsize = SvUV(SvRV(PL_rs)); /* RsRECORD() guarantees > 0. */
8334 /* Grab the size of the record we're getting */
8335 char *buffer = SvGROW(sv, (STRLEN)(recsize + append + 1)) + append;
8342 /* With a true, record-oriented file on VMS, we need to use read directly
8343 * to ensure that we respect RMS record boundaries. The user is responsible
8344 * for providing a PL_rs value that corresponds to the FAB$W_MRS (maximum
8345 * record size) field. N.B. This is likely to produce invalid results on
8346 * varying-width character data when a record ends mid-character.
8348 fd = PerlIO_fileno(fp);
8350 && PerlLIO_fstat(fd, &st) == 0
8351 && (st.st_fab_rfm == FAB$C_VAR
8352 || st.st_fab_rfm == FAB$C_VFC
8353 || st.st_fab_rfm == FAB$C_FIX)) {
8355 bytesread = PerlLIO_read(fd, buffer, recsize);
8357 else /* in-memory file from PerlIO::Scalar
8358 * or not a record-oriented file
8362 bytesread = PerlIO_read(fp, buffer, recsize);
8364 /* At this point, the logic in sv_get() means that sv will
8365 be treated as utf-8 if the handle is utf8.
8367 if (PerlIO_isutf8(fp) && bytesread > 0) {
8368 char *bend = buffer + bytesread;
8369 char *bufp = buffer;
8370 size_t charcount = 0;
8371 bool charstart = TRUE;
8374 while (charcount < recsize) {
8375 /* count accumulated characters */
8376 while (bufp < bend) {
8378 skip = UTF8SKIP(bufp);
8380 if (bufp + skip > bend) {
8381 /* partial at the end */
8392 if (charcount < recsize) {
8394 STRLEN bufp_offset = bufp - buffer;
8395 SSize_t morebytesread;
8397 /* originally I read enough to fill any incomplete
8398 character and the first byte of the next
8399 character if needed, but if there's many
8400 multi-byte encoded characters we're going to be
8401 making a read call for every character beyond
8402 the original read size.
8404 So instead, read the rest of the character if
8405 any, and enough bytes to match at least the
8406 start bytes for each character we're going to
8410 readsize = recsize - charcount;
8412 readsize = skip - (bend - bufp) + recsize - charcount - 1;
8413 buffer = SvGROW(sv, append + bytesread + readsize + 1) + append;
8414 bend = buffer + bytesread;
8415 morebytesread = PerlIO_read(fp, bend, readsize);
8416 if (morebytesread <= 0) {
8417 /* we're done, if we still have incomplete
8418 characters the check code in sv_gets() will
8421 I'd originally considered doing
8422 PerlIO_ungetc() on all but the lead
8423 character of the incomplete character, but
8424 read() doesn't do that, so I don't.
8429 /* prepare to scan some more */
8430 bytesread += morebytesread;
8431 bend = buffer + bytesread;
8432 bufp = buffer + bufp_offset;
8440 SvCUR_set(sv, bytesread + append);
8441 buffer[bytesread] = '\0';
8442 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8448 Get a line from the filehandle and store it into the SV, optionally
8449 appending to the currently-stored string. If C<append> is not 0, the
8450 line is appended to the SV instead of overwriting it. C<append> should
8451 be set to the byte offset that the appended string should start at
8452 in the SV (typically, C<SvCUR(sv)> is a suitable choice).
8458 Perl_sv_gets(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8468 PERL_ARGS_ASSERT_SV_GETS;
8470 if (SvTHINKFIRST(sv))
8471 sv_force_normal_flags(sv, append ? 0 : SV_COW_DROP_PV);
8472 /* XXX. If you make this PVIV, then copy on write can copy scalars read
8474 However, perlbench says it's slower, because the existing swipe code
8475 is faster than copy on write.
8476 Swings and roundabouts. */
8477 SvUPGRADE(sv, SVt_PV);
8480 /* line is going to be appended to the existing buffer in the sv */
8481 if (PerlIO_isutf8(fp)) {
8483 sv_utf8_upgrade_nomg(sv);
8484 sv_pos_u2b(sv,&append,0);
8486 } else if (SvUTF8(sv)) {
8487 return S_sv_gets_append_to_utf8(aTHX_ sv, fp, append);
8493 /* not appending - "clear" the string by setting SvCUR to 0,
8494 * the pv is still avaiable. */
8497 if (PerlIO_isutf8(fp))
8500 if (IN_PERL_COMPILETIME) {
8501 /* we always read code in line mode */
8505 else if (RsSNARF(PL_rs)) {
8506 /* If it is a regular disk file use size from stat() as estimate
8507 of amount we are going to read -- may result in mallocing
8508 more memory than we really need if the layers below reduce
8509 the size we read (e.g. CRLF or a gzip layer).
8512 int fd = PerlIO_fileno(fp);
8513 if (fd >= 0 && (PerlLIO_fstat(fd, &st) == 0) && S_ISREG(st.st_mode)) {
8514 const Off_t offset = PerlIO_tell(fp);
8515 if (offset != (Off_t) -1 && st.st_size + append > offset) {
8516 #ifdef PERL_COPY_ON_WRITE
8517 /* Add an extra byte for the sake of copy-on-write's
8518 * buffer reference count. */
8519 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 2));
8521 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 1));
8528 else if (RsRECORD(PL_rs)) {
8529 return S_sv_gets_read_record(aTHX_ sv, fp, append);
8531 else if (RsPARA(PL_rs)) {
8537 /* Get $/ i.e. PL_rs into same encoding as stream wants */
8538 if (PerlIO_isutf8(fp)) {
8539 rsptr = SvPVutf8(PL_rs, rslen);
8542 if (SvUTF8(PL_rs)) {
8543 if (!sv_utf8_downgrade(PL_rs, TRUE)) {
8544 Perl_croak(aTHX_ "Wide character in $/");
8547 /* extract the raw pointer to the record separator */
8548 rsptr = SvPV_const(PL_rs, rslen);
8552 /* rslast is the last character in the record separator
8553 * note we don't use rslast except when rslen is true, so the
8554 * null assign is a placeholder. */
8555 rslast = rslen ? rsptr[rslen - 1] : '\0';
8557 if (rspara) { /* have to do this both before and after */
8558 /* to make sure file boundaries work right */
8562 i = PerlIO_getc(fp);
8566 PerlIO_ungetc(fp,i);
8572 /* See if we know enough about I/O mechanism to cheat it ! */
8574 /* This used to be #ifdef test - it is made run-time test for ease
8575 of abstracting out stdio interface. One call should be cheap
8576 enough here - and may even be a macro allowing compile
8580 if (PerlIO_fast_gets(fp)) {
8582 * We can do buffer based IO operations on this filehandle.
8584 * This means we can bypass a lot of subcalls and process
8585 * the buffer directly, it also means we know the upper bound
8586 * on the amount of data we might read of the current buffer
8587 * into our sv. Knowing this allows us to preallocate the pv
8588 * to be able to hold that maximum, which allows us to simplify
8589 * a lot of logic. */
8592 * We're going to steal some values from the stdio struct
8593 * and put EVERYTHING in the innermost loop into registers.
8595 STDCHAR *ptr; /* pointer into fp's read-ahead buffer */
8596 STRLEN bpx; /* length of the data in the target sv
8597 used to fix pointers after a SvGROW */
8598 I32 shortbuffered; /* If the pv buffer is shorter than the amount
8599 of data left in the read-ahead buffer.
8600 If 0 then the pv buffer can hold the full
8601 amount left, otherwise this is the amount it
8604 /* Here is some breathtakingly efficient cheating */
8606 /* When you read the following logic resist the urge to think
8607 * of record separators that are 1 byte long. They are an
8608 * uninteresting special (simple) case.
8610 * Instead think of record separators which are at least 2 bytes
8611 * long, and keep in mind that we need to deal with such
8612 * separators when they cross a read-ahead buffer boundary.
8614 * Also consider that we need to gracefully deal with separators
8615 * that may be longer than a single read ahead buffer.
8617 * Lastly do not forget we want to copy the delimiter as well. We
8618 * are copying all data in the file _up_to_and_including_ the separator
8621 * Now that you have all that in mind here is what is happening below:
8623 * 1. When we first enter the loop we do some memory book keeping to see
8624 * how much free space there is in the target SV. (This sub assumes that
8625 * it is operating on the same SV most of the time via $_ and that it is
8626 * going to be able to reuse the same pv buffer each call.) If there is
8627 * "enough" room then we set "shortbuffered" to how much space there is
8628 * and start reading forward.
8630 * 2. When we scan forward we copy from the read-ahead buffer to the target
8631 * SV's pv buffer. While we go we watch for the end of the read-ahead buffer,
8632 * and the end of the of pv, as well as for the "rslast", which is the last
8633 * char of the separator.
8635 * 3. When scanning forward if we see rslast then we jump backwards in *pv*
8636 * (which has a "complete" record up to the point we saw rslast) and check
8637 * it to see if it matches the separator. If it does we are done. If it doesn't
8638 * we continue on with the scan/copy.
8640 * 4. If we run out of read-ahead buffer (cnt goes to 0) then we have to get
8641 * the IO system to read the next buffer. We do this by doing a getc(), which
8642 * returns a single char read (or EOF), and prefills the buffer, and also
8643 * allows us to find out how full the buffer is. We use this information to
8644 * SvGROW() the sv to the size remaining in the buffer, after which we copy
8645 * the returned single char into the target sv, and then go back into scan
8648 * 5. If we run out of write-buffer then we SvGROW() it by the size of the
8649 * remaining space in the read-buffer.
8651 * Note that this code despite its twisty-turny nature is pretty darn slick.
8652 * It manages single byte separators, multi-byte cross boundary separators,
8653 * and cross-read-buffer separators cleanly and efficiently at the cost
8654 * of potentially greatly overallocating the target SV.
8660 /* get the number of bytes remaining in the read-ahead buffer
8661 * on first call on a given fp this will return 0.*/
8662 cnt = PerlIO_get_cnt(fp);
8664 /* make sure we have the room */
8665 if ((I32)(SvLEN(sv) - append) <= cnt + 1) {
8666 /* Not room for all of it
8667 if we are looking for a separator and room for some
8669 if (rslen && cnt > 80 && (I32)SvLEN(sv) > append) {
8670 /* just process what we have room for */
8671 shortbuffered = cnt - SvLEN(sv) + append + 1;
8672 cnt -= shortbuffered;
8675 /* ensure that the target sv has enough room to hold
8676 * the rest of the read-ahead buffer */
8678 /* remember that cnt can be negative */
8679 SvGROW(sv, (STRLEN)(append + (cnt <= 0 ? 2 : (cnt + 1))));
8683 /* we have enough room to hold the full buffer, lets scream */
8687 /* extract the pointer to sv's string buffer, offset by append as necessary */
8688 bp = (STDCHAR*)SvPVX_const(sv) + append; /* move these two too to registers */
8689 /* extract the point to the read-ahead buffer */
8690 ptr = (STDCHAR*)PerlIO_get_ptr(fp);
8692 /* some trace debug output */
8693 DEBUG_P(PerlIO_printf(Perl_debug_log,
8694 "Screamer: entering, ptr=%" UVuf ", cnt=%ld\n",PTR2UV(ptr),(long)cnt));
8695 DEBUG_P(PerlIO_printf(Perl_debug_log,
8696 "Screamer: entering: PerlIO * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%"
8698 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8699 PTR2UV(PerlIO_has_base(fp) ? PerlIO_get_base(fp) : 0)));
8703 /* if there is stuff left in the read-ahead buffer */
8705 /* if there is a separator */
8707 /* find next rslast */
8710 /* shortcut common case of blank line */
8712 if ((*bp++ = *ptr++) == rslast)
8713 goto thats_all_folks;
8715 p = (STDCHAR *)memchr(ptr, rslast, cnt);
8717 SSize_t got = p - ptr + 1;
8718 Copy(ptr, bp, got, STDCHAR);
8722 goto thats_all_folks;
8724 Copy(ptr, bp, cnt, STDCHAR);
8730 /* no separator, slurp the full buffer */
8731 Copy(ptr, bp, cnt, char); /* this | eat */
8732 bp += cnt; /* screams | dust */
8733 ptr += cnt; /* louder | sed :-) */
8735 assert (!shortbuffered);
8736 goto cannot_be_shortbuffered;
8740 if (shortbuffered) { /* oh well, must extend */
8741 /* we didnt have enough room to fit the line into the target buffer
8742 * so we must extend the target buffer and keep going */
8743 cnt = shortbuffered;
8745 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8747 /* extned the target sv's buffer so it can hold the full read-ahead buffer */
8748 SvGROW(sv, SvLEN(sv) + append + cnt + 2);
8749 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8753 cannot_be_shortbuffered:
8754 /* we need to refill the read-ahead buffer if possible */
8756 DEBUG_P(PerlIO_printf(Perl_debug_log,
8757 "Screamer: going to getc, ptr=%" UVuf ", cnt=%" IVdf "\n",
8758 PTR2UV(ptr),(IV)cnt));
8759 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* deregisterize cnt and ptr */
8761 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8762 "Screamer: pre: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf "\n",
8763 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8764 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8767 call PerlIO_getc() to let it prefill the lookahead buffer
8769 This used to call 'filbuf' in stdio form, but as that behaves like
8770 getc when cnt <= 0 we use PerlIO_getc here to avoid introducing
8771 another abstraction.
8773 Note we have to deal with the char in 'i' if we are not at EOF
8775 bpx = bp - (STDCHAR*)SvPVX_const(sv);
8776 /* signals might be called here, possibly modifying sv */
8777 i = PerlIO_getc(fp); /* get more characters */
8778 bp = (STDCHAR*)SvPVX_const(sv) + bpx;
8780 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8781 "Screamer: post: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf "\n",
8782 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8783 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8785 /* find out how much is left in the read-ahead buffer, and rextract its pointer */
8786 cnt = PerlIO_get_cnt(fp);
8787 ptr = (STDCHAR*)PerlIO_get_ptr(fp); /* reregisterize cnt and ptr */
8788 DEBUG_P(PerlIO_printf(Perl_debug_log,
8789 "Screamer: after getc, ptr=%" UVuf ", cnt=%" IVdf "\n",
8790 PTR2UV(ptr),(IV)cnt));
8792 if (i == EOF) /* all done for ever? */
8793 goto thats_really_all_folks;
8795 /* make sure we have enough space in the target sv */
8796 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8798 SvGROW(sv, bpx + cnt + 2);
8799 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8801 /* copy of the char we got from getc() */
8802 *bp++ = (STDCHAR)i; /* store character from PerlIO_getc */
8804 /* make sure we deal with the i being the last character of a separator */
8805 if (rslen && (STDCHAR)i == rslast) /* all done for now? */
8806 goto thats_all_folks;
8810 /* check if we have actually found the separator - only really applies
8812 if ((rslen > 1 && (STRLEN)(bp - (STDCHAR*)SvPVX_const(sv)) < rslen) ||
8813 memNE((char*)bp - rslen, rsptr, rslen))
8814 goto screamer; /* go back to the fray */
8815 thats_really_all_folks:
8817 cnt += shortbuffered;
8818 DEBUG_P(PerlIO_printf(Perl_debug_log,
8819 "Screamer: quitting, ptr=%" UVuf ", cnt=%" IVdf "\n",PTR2UV(ptr),(IV)cnt));
8820 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* put these back or we're in trouble */
8821 DEBUG_P(PerlIO_printf(Perl_debug_log,
8822 "Screamer: end: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf
8824 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8825 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8827 SvCUR_set(sv, bp - (STDCHAR*)SvPVX_const(sv)); /* set length */
8828 DEBUG_P(PerlIO_printf(Perl_debug_log,
8829 "Screamer: done, len=%ld, string=|%.*s|\n",
8830 (long)SvCUR(sv),(int)SvCUR(sv),SvPVX_const(sv)));
8834 /*The big, slow, and stupid way. */
8835 #ifdef USE_HEAP_INSTEAD_OF_STACK /* Even slower way. */
8836 STDCHAR *buf = NULL;
8837 Newx(buf, 8192, STDCHAR);
8845 const STDCHAR * const bpe = buf + sizeof(buf);
8847 while ((i = PerlIO_getc(fp)) != EOF && (*bp++ = (STDCHAR)i) != rslast && bp < bpe)
8848 ; /* keep reading */
8852 cnt = PerlIO_read(fp,(char*)buf, sizeof(buf));
8853 /* Accommodate broken VAXC compiler, which applies U8 cast to
8854 * both args of ?: operator, causing EOF to change into 255
8857 i = (U8)buf[cnt - 1];
8863 cnt = 0; /* we do need to re-set the sv even when cnt <= 0 */
8865 sv_catpvn_nomg(sv, (char *) buf, cnt);
8867 sv_setpvn(sv, (char *) buf, cnt); /* "nomg" is implied */
8869 if (i != EOF && /* joy */
8871 SvCUR(sv) < rslen ||
8872 memNE(SvPVX_const(sv) + SvCUR(sv) - rslen, rsptr, rslen)))
8876 * If we're reading from a TTY and we get a short read,
8877 * indicating that the user hit his EOF character, we need
8878 * to notice it now, because if we try to read from the TTY
8879 * again, the EOF condition will disappear.
8881 * The comparison of cnt to sizeof(buf) is an optimization
8882 * that prevents unnecessary calls to feof().
8886 if (!(cnt < (I32)sizeof(buf) && PerlIO_eof(fp)))
8890 #ifdef USE_HEAP_INSTEAD_OF_STACK
8895 if (rspara) { /* have to do this both before and after */
8896 while (i != EOF) { /* to make sure file boundaries work right */
8897 i = PerlIO_getc(fp);
8899 PerlIO_ungetc(fp,i);
8905 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8911 Auto-increment of the value in the SV, doing string to numeric conversion
8912 if necessary. Handles 'get' magic and operator overloading.
8918 Perl_sv_inc(pTHX_ SV *const sv)
8927 =for apidoc sv_inc_nomg
8929 Auto-increment of the value in the SV, doing string to numeric conversion
8930 if necessary. Handles operator overloading. Skips handling 'get' magic.
8936 Perl_sv_inc_nomg(pTHX_ SV *const sv)
8943 if (SvTHINKFIRST(sv)) {
8944 if (SvREADONLY(sv)) {
8945 Perl_croak_no_modify();
8949 if (SvAMAGIC(sv) && AMG_CALLunary(sv, inc_amg))
8951 i = PTR2IV(SvRV(sv));
8955 else sv_force_normal_flags(sv, 0);
8957 flags = SvFLAGS(sv);
8958 if ((flags & (SVp_NOK|SVp_IOK)) == SVp_NOK) {
8959 /* It's (privately or publicly) a float, but not tested as an
8960 integer, so test it to see. */
8962 flags = SvFLAGS(sv);
8964 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8965 /* It's publicly an integer, or privately an integer-not-float */
8966 #ifdef PERL_PRESERVE_IVUV
8970 if (SvUVX(sv) == UV_MAX)
8971 sv_setnv(sv, UV_MAX_P1);
8973 (void)SvIOK_only_UV(sv);
8974 SvUV_set(sv, SvUVX(sv) + 1);
8976 if (SvIVX(sv) == IV_MAX)
8977 sv_setuv(sv, (UV)IV_MAX + 1);
8979 (void)SvIOK_only(sv);
8980 SvIV_set(sv, SvIVX(sv) + 1);
8985 if (flags & SVp_NOK) {
8986 const NV was = SvNVX(sv);
8987 if (LIKELY(!Perl_isinfnan(was)) &&
8988 NV_OVERFLOWS_INTEGERS_AT != 0.0 &&
8989 was >= NV_OVERFLOWS_INTEGERS_AT) {
8990 /* diag_listed_as: Lost precision when %s %f by 1 */
8991 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8992 "Lost precision when incrementing %" NVff " by 1",
8995 (void)SvNOK_only(sv);
8996 SvNV_set(sv, was + 1.0);
9000 /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */
9001 if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv)))
9002 Perl_croak_no_modify();
9004 if (!(flags & SVp_POK) || !*SvPVX_const(sv)) {
9005 if ((flags & SVTYPEMASK) < SVt_PVIV)
9006 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV ? SVt_PVIV : SVt_IV));
9007 (void)SvIOK_only(sv);
9012 while (isALPHA(*d)) d++;
9013 while (isDIGIT(*d)) d++;
9014 if (d < SvEND(sv)) {
9015 const int numtype = grok_number_flags(SvPVX_const(sv), SvCUR(sv), NULL, PERL_SCAN_TRAILING);
9016 #ifdef PERL_PRESERVE_IVUV
9017 /* Got to punt this as an integer if needs be, but we don't issue
9018 warnings. Probably ought to make the sv_iv_please() that does
9019 the conversion if possible, and silently. */
9020 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
9021 /* Need to try really hard to see if it's an integer.
9022 9.22337203685478e+18 is an integer.
9023 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
9024 so $a="9.22337203685478e+18"; $a+0; $a++
9025 needs to be the same as $a="9.22337203685478e+18"; $a++
9032 /* sv_2iv *should* have made this an NV */
9033 if (flags & SVp_NOK) {
9034 (void)SvNOK_only(sv);
9035 SvNV_set(sv, SvNVX(sv) + 1.0);
9038 /* I don't think we can get here. Maybe I should assert this
9039 And if we do get here I suspect that sv_setnv will croak. NWC
9041 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_inc punt failed to convert '%s' to IOK or NOKp, UV=0x%" UVxf " NV=%" NVgf "\n",
9042 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
9044 #endif /* PERL_PRESERVE_IVUV */
9045 if (!numtype && ckWARN(WARN_NUMERIC))
9046 not_incrementable(sv);
9047 sv_setnv(sv,Atof(SvPVX_const(sv)) + 1.0);
9051 while (d >= SvPVX_const(sv)) {
9059 /* MKS: The original code here died if letters weren't consecutive.
9060 * at least it didn't have to worry about non-C locales. The
9061 * new code assumes that ('z'-'a')==('Z'-'A'), letters are
9062 * arranged in order (although not consecutively) and that only
9063 * [A-Za-z] are accepted by isALPHA in the C locale.
9065 if (isALPHA_FOLD_NE(*d, 'z')) {
9066 do { ++*d; } while (!isALPHA(*d));
9069 *(d--) -= 'z' - 'a';
9074 *(d--) -= 'z' - 'a' + 1;
9078 /* oh,oh, the number grew */
9079 SvGROW(sv, SvCUR(sv) + 2);
9080 SvCUR_set(sv, SvCUR(sv) + 1);
9081 for (d = SvPVX(sv) + SvCUR(sv); d > SvPVX_const(sv); d--)
9092 Auto-decrement of the value in the SV, doing string to numeric conversion
9093 if necessary. Handles 'get' magic and operator overloading.
9099 Perl_sv_dec(pTHX_ SV *const sv)
9108 =for apidoc sv_dec_nomg
9110 Auto-decrement of the value in the SV, doing string to numeric conversion
9111 if necessary. Handles operator overloading. Skips handling 'get' magic.
9117 Perl_sv_dec_nomg(pTHX_ SV *const sv)
9123 if (SvTHINKFIRST(sv)) {
9124 if (SvREADONLY(sv)) {
9125 Perl_croak_no_modify();
9129 if (SvAMAGIC(sv) && AMG_CALLunary(sv, dec_amg))
9131 i = PTR2IV(SvRV(sv));
9135 else sv_force_normal_flags(sv, 0);
9137 /* Unlike sv_inc we don't have to worry about string-never-numbers
9138 and keeping them magic. But we mustn't warn on punting */
9139 flags = SvFLAGS(sv);
9140 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
9141 /* It's publicly an integer, or privately an integer-not-float */
9142 #ifdef PERL_PRESERVE_IVUV
9146 if (SvUVX(sv) == 0) {
9147 (void)SvIOK_only(sv);
9151 (void)SvIOK_only_UV(sv);
9152 SvUV_set(sv, SvUVX(sv) - 1);
9155 if (SvIVX(sv) == IV_MIN) {
9156 sv_setnv(sv, (NV)IV_MIN);
9160 (void)SvIOK_only(sv);
9161 SvIV_set(sv, SvIVX(sv) - 1);
9166 if (flags & SVp_NOK) {
9169 const NV was = SvNVX(sv);
9170 if (LIKELY(!Perl_isinfnan(was)) &&
9171 NV_OVERFLOWS_INTEGERS_AT != 0.0 &&
9172 was <= -NV_OVERFLOWS_INTEGERS_AT) {
9173 /* diag_listed_as: Lost precision when %s %f by 1 */
9174 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
9175 "Lost precision when decrementing %" NVff " by 1",
9178 (void)SvNOK_only(sv);
9179 SvNV_set(sv, was - 1.0);
9184 /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */
9185 if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv)))
9186 Perl_croak_no_modify();
9188 if (!(flags & SVp_POK)) {
9189 if ((flags & SVTYPEMASK) < SVt_PVIV)
9190 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV) ? SVt_PVIV : SVt_IV);
9192 (void)SvIOK_only(sv);
9195 #ifdef PERL_PRESERVE_IVUV
9197 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), NULL);
9198 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
9199 /* Need to try really hard to see if it's an integer.
9200 9.22337203685478e+18 is an integer.
9201 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
9202 so $a="9.22337203685478e+18"; $a+0; $a--
9203 needs to be the same as $a="9.22337203685478e+18"; $a--
9210 /* sv_2iv *should* have made this an NV */
9211 if (flags & SVp_NOK) {
9212 (void)SvNOK_only(sv);
9213 SvNV_set(sv, SvNVX(sv) - 1.0);
9216 /* I don't think we can get here. Maybe I should assert this
9217 And if we do get here I suspect that sv_setnv will croak. NWC
9219 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_dec punt failed to convert '%s' to IOK or NOKp, UV=0x%" UVxf " NV=%" NVgf "\n",
9220 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
9223 #endif /* PERL_PRESERVE_IVUV */
9224 sv_setnv(sv,Atof(SvPVX_const(sv)) - 1.0); /* punt */
9227 /* this define is used to eliminate a chunk of duplicated but shared logic
9228 * it has the suffix __SV_C to signal that it isnt API, and isnt meant to be
9229 * used anywhere but here - yves
9231 #define PUSH_EXTEND_MORTAL__SV_C(AnSv) \
9233 SSize_t ix = ++PL_tmps_ix; \
9234 if (UNLIKELY(ix >= PL_tmps_max)) \
9235 ix = tmps_grow_p(ix); \
9236 PL_tmps_stack[ix] = (AnSv); \
9240 =for apidoc sv_mortalcopy
9242 Creates a new SV which is a copy of the original SV (using C<sv_setsv>).
9243 The new SV is marked as mortal. It will be destroyed "soon", either by an
9244 explicit call to C<FREETMPS>, or by an implicit call at places such as
9245 statement boundaries. See also C<L</sv_newmortal>> and C<L</sv_2mortal>>.
9247 =for apidoc sv_mortalcopy_flags
9249 Like C<sv_mortalcopy>, but the extra C<flags> are passed to the
9255 /* Make a string that will exist for the duration of the expression
9256 * evaluation. Actually, it may have to last longer than that, but
9257 * hopefully we won't free it until it has been assigned to a
9258 * permanent location. */
9261 Perl_sv_mortalcopy_flags(pTHX_ SV *const oldstr, U32 flags)
9265 if (flags & SV_GMAGIC)
9266 SvGETMAGIC(oldstr); /* before new_SV, in case it dies */
9268 sv_setsv_flags(sv,oldstr,flags & ~SV_GMAGIC);
9269 PUSH_EXTEND_MORTAL__SV_C(sv);
9275 =for apidoc sv_newmortal
9277 Creates a new null SV which is mortal. The reference count of the SV is
9278 set to 1. It will be destroyed "soon", either by an explicit call to
9279 C<FREETMPS>, or by an implicit call at places such as statement boundaries.
9280 See also C<L</sv_mortalcopy>> and C<L</sv_2mortal>>.
9286 Perl_sv_newmortal(pTHX)
9291 SvFLAGS(sv) = SVs_TEMP;
9292 PUSH_EXTEND_MORTAL__SV_C(sv);
9298 =for apidoc newSVpvn_flags
9300 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9301 characters) into it. The reference count for the
9302 SV is set to 1. Note that if C<len> is zero, Perl will create a zero length
9303 string. You are responsible for ensuring that the source string is at least
9304 C<len> bytes long. If the C<s> argument is NULL the new SV will be undefined.
9305 Currently the only flag bits accepted are C<SVf_UTF8> and C<SVs_TEMP>.
9306 If C<SVs_TEMP> is set, then C<sv_2mortal()> is called on the result before
9307 returning. If C<SVf_UTF8> is set, C<s>
9308 is considered to be in UTF-8 and the
9309 C<SVf_UTF8> flag will be set on the new SV.
9310 C<newSVpvn_utf8()> is a convenience wrapper for this function, defined as
9312 #define newSVpvn_utf8(s, len, u) \
9313 newSVpvn_flags((s), (len), (u) ? SVf_UTF8 : 0)
9315 =for apidoc Amnh||SVf_UTF8
9316 =for apidoc Amnh||SVs_TEMP
9322 Perl_newSVpvn_flags(pTHX_ const char *const s, const STRLEN len, const U32 flags)
9326 /* All the flags we don't support must be zero.
9327 And we're new code so I'm going to assert this from the start. */
9328 assert(!(flags & ~(SVf_UTF8|SVs_TEMP)));
9330 sv_setpvn(sv,s,len);
9332 /* This code used to do a sv_2mortal(), however we now unroll the call to
9333 * sv_2mortal() and do what it does ourselves here. Since we have asserted
9334 * that flags can only have the SVf_UTF8 and/or SVs_TEMP flags set above we
9335 * can use it to enable the sv flags directly (bypassing SvTEMP_on), which
9336 * in turn means we dont need to mask out the SVf_UTF8 flag below, which
9337 * means that we eliminate quite a few steps than it looks - Yves
9338 * (explaining patch by gfx) */
9340 SvFLAGS(sv) |= flags;
9342 if(flags & SVs_TEMP){
9343 PUSH_EXTEND_MORTAL__SV_C(sv);
9350 =for apidoc sv_2mortal
9352 Marks an existing SV as mortal. The SV will be destroyed "soon", either
9353 by an explicit call to C<FREETMPS>, or by an implicit call at places such as
9354 statement boundaries. C<SvTEMP()> is turned on which means that the SV's
9355 string buffer can be "stolen" if this SV is copied. See also
9356 C<L</sv_newmortal>> and C<L</sv_mortalcopy>>.
9362 Perl_sv_2mortal(pTHX_ SV *const sv)
9369 PUSH_EXTEND_MORTAL__SV_C(sv);
9377 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9378 characters) into it. The reference count for the
9379 SV is set to 1. If C<len> is zero, Perl will compute the length using
9380 C<strlen()>, (which means if you use this option, that C<s> can't have embedded
9381 C<NUL> characters and has to have a terminating C<NUL> byte).
9383 This function can cause reliability issues if you are likely to pass in
9384 empty strings that are not null terminated, because it will run
9385 strlen on the string and potentially run past valid memory.
9387 Using L</newSVpvn> is a safer alternative for non C<NUL> terminated strings.
9388 For string literals use L</newSVpvs> instead. This function will work fine for
9389 C<NUL> terminated strings, but if you want to avoid the if statement on whether
9390 to call C<strlen> use C<newSVpvn> instead (calling C<strlen> yourself).
9396 Perl_newSVpv(pTHX_ const char *const s, const STRLEN len)
9401 sv_setpvn(sv, s, len || s == NULL ? len : strlen(s));
9406 =for apidoc newSVpvn
9408 Creates a new SV and copies a string into it, which may contain C<NUL> characters
9409 (C<\0>) and other binary data. The reference count for the SV is set to 1.
9410 Note that if C<len> is zero, Perl will create a zero length (Perl) string. You
9411 are responsible for ensuring that the source buffer is at least
9412 C<len> bytes long. If the C<buffer> argument is NULL the new SV will be
9419 Perl_newSVpvn(pTHX_ const char *const buffer, const STRLEN len)
9423 sv_setpvn(sv,buffer,len);
9428 =for apidoc newSVhek
9430 Creates a new SV from the hash key structure. It will generate scalars that
9431 point to the shared string table where possible. Returns a new (undefined)
9432 SV if C<hek> is NULL.
9438 Perl_newSVhek(pTHX_ const HEK *const hek)
9447 if (HEK_LEN(hek) == HEf_SVKEY) {
9448 return newSVsv(*(SV**)HEK_KEY(hek));
9450 const int flags = HEK_FLAGS(hek);
9451 if (flags & HVhek_WASUTF8) {
9453 Andreas would like keys he put in as utf8 to come back as utf8
9455 STRLEN utf8_len = HEK_LEN(hek);
9456 SV * const sv = newSV_type(SVt_PV);
9457 char *as_utf8 = (char *)bytes_to_utf8 ((U8*)HEK_KEY(hek), &utf8_len);
9458 /* bytes_to_utf8() allocates a new string, which we can repurpose: */
9459 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
9462 } else if (flags & HVhek_UNSHARED) {
9463 /* A hash that isn't using shared hash keys has to have
9464 the flag in every key so that we know not to try to call
9465 share_hek_hek on it. */
9467 SV * const sv = newSVpvn (HEK_KEY(hek), HEK_LEN(hek));
9472 /* This will be overwhelminly the most common case. */
9474 /* Inline most of newSVpvn_share(), because share_hek_hek() is far
9475 more efficient than sharepvn(). */
9479 sv_upgrade(sv, SVt_PV);
9480 SvPV_set(sv, (char *)HEK_KEY(share_hek_hek(hek)));
9481 SvCUR_set(sv, HEK_LEN(hek));
9493 =for apidoc newSVpvn_share
9495 Creates a new SV with its C<SvPVX_const> pointing to a shared string in the string
9496 table. If the string does not already exist in the table, it is
9497 created first. Turns on the C<SvIsCOW> flag (or C<READONLY>
9498 and C<FAKE> in 5.16 and earlier). If the C<hash> parameter
9499 is non-zero, that value is used; otherwise the hash is computed.
9500 The string's hash can later be retrieved from the SV
9501 with the C<SvSHARED_HASH()> macro. The idea here is
9502 that as the string table is used for shared hash keys these strings will have
9503 C<SvPVX_const == HeKEY> and hash lookup will avoid string compare.
9509 Perl_newSVpvn_share(pTHX_ const char *src, I32 len, U32 hash)
9513 bool is_utf8 = FALSE;
9514 const char *const orig_src = src;
9517 STRLEN tmplen = -len;
9519 /* See the note in hv.c:hv_fetch() --jhi */
9520 src = (char*)bytes_from_utf8((const U8*)src, &tmplen, &is_utf8);
9524 PERL_HASH(hash, src, len);
9526 /* The logic for this is inlined in S_mro_get_linear_isa_dfs(), so if it
9527 changes here, update it there too. */
9528 sv_upgrade(sv, SVt_PV);
9529 SvPV_set(sv, sharepvn(src, is_utf8?-len:len, hash));
9536 if (src != orig_src)
9542 =for apidoc newSVpv_share
9544 Like C<newSVpvn_share>, but takes a C<NUL>-terminated string instead of a
9551 Perl_newSVpv_share(pTHX_ const char *src, U32 hash)
9553 return newSVpvn_share(src, strlen(src), hash);
9556 #if defined(PERL_IMPLICIT_CONTEXT)
9558 /* pTHX_ magic can't cope with varargs, so this is a no-context
9559 * version of the main function, (which may itself be aliased to us).
9560 * Don't access this version directly.
9564 Perl_newSVpvf_nocontext(const char *const pat, ...)
9570 PERL_ARGS_ASSERT_NEWSVPVF_NOCONTEXT;
9572 va_start(args, pat);
9573 sv = vnewSVpvf(pat, &args);
9580 =for apidoc newSVpvf
9582 Creates a new SV and initializes it with the string formatted like
9589 Perl_newSVpvf(pTHX_ const char *const pat, ...)
9594 PERL_ARGS_ASSERT_NEWSVPVF;
9596 va_start(args, pat);
9597 sv = vnewSVpvf(pat, &args);
9602 /* backend for newSVpvf() and newSVpvf_nocontext() */
9605 Perl_vnewSVpvf(pTHX_ const char *const pat, va_list *const args)
9609 PERL_ARGS_ASSERT_VNEWSVPVF;
9612 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
9619 Creates a new SV and copies a floating point value into it.
9620 The reference count for the SV is set to 1.
9626 Perl_newSVnv(pTHX_ const NV n)
9638 Creates a new SV and copies an integer into it. The reference count for the
9645 Perl_newSViv(pTHX_ const IV i)
9651 /* Inlining ONLY the small relevant subset of sv_setiv here
9652 * for performance. Makes a significant difference. */
9654 /* We're starting from SVt_FIRST, so provided that's
9655 * actual 0, we don't have to unset any SV type flags
9656 * to promote to SVt_IV. */
9657 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9659 SET_SVANY_FOR_BODYLESS_IV(sv);
9660 SvFLAGS(sv) |= SVt_IV;
9672 Creates a new SV and copies an unsigned integer into it.
9673 The reference count for the SV is set to 1.
9679 Perl_newSVuv(pTHX_ const UV u)
9683 /* Inlining ONLY the small relevant subset of sv_setuv here
9684 * for performance. Makes a significant difference. */
9686 /* Using ivs is more efficient than using uvs - see sv_setuv */
9687 if (u <= (UV)IV_MAX) {
9688 return newSViv((IV)u);
9693 /* We're starting from SVt_FIRST, so provided that's
9694 * actual 0, we don't have to unset any SV type flags
9695 * to promote to SVt_IV. */
9696 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9698 SET_SVANY_FOR_BODYLESS_IV(sv);
9699 SvFLAGS(sv) |= SVt_IV;
9701 (void)SvIsUV_on(sv);
9710 =for apidoc newSV_type
9712 Creates a new SV, of the type specified. The reference count for the new SV
9719 Perl_newSV_type(pTHX_ const svtype type)
9724 ASSUME(SvTYPE(sv) == SVt_FIRST);
9725 if(type != SVt_FIRST)
9726 sv_upgrade(sv, type);
9731 =for apidoc newRV_noinc
9733 Creates an RV wrapper for an SV. The reference count for the original
9734 SV is B<not> incremented.
9740 Perl_newRV_noinc(pTHX_ SV *const tmpRef)
9744 PERL_ARGS_ASSERT_NEWRV_NOINC;
9748 /* We're starting from SVt_FIRST, so provided that's
9749 * actual 0, we don't have to unset any SV type flags
9750 * to promote to SVt_IV. */
9751 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9753 SET_SVANY_FOR_BODYLESS_IV(sv);
9754 SvFLAGS(sv) |= SVt_IV;
9759 SvRV_set(sv, tmpRef);
9764 /* newRV_inc is the official function name to use now.
9765 * newRV_inc is in fact #defined to newRV in sv.h
9769 Perl_newRV(pTHX_ SV *const sv)
9771 PERL_ARGS_ASSERT_NEWRV;
9773 return newRV_noinc(SvREFCNT_inc_simple_NN(sv));
9779 Creates a new SV which is an exact duplicate of the original SV.
9782 =for apidoc newSVsv_nomg
9784 Like C<newSVsv> but does not process get magic.
9790 Perl_newSVsv_flags(pTHX_ SV *const old, I32 flags)
9796 if (SvTYPE(old) == (svtype)SVTYPEMASK) {
9797 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), "semi-panic: attempt to dup freed string");
9800 /* Do this here, otherwise we leak the new SV if this croaks. */
9801 if (flags & SV_GMAGIC)
9804 sv_setsv_flags(sv, old, flags & ~SV_GMAGIC);
9809 =for apidoc sv_reset
9811 Underlying implementation for the C<reset> Perl function.
9812 Note that the perl-level function is vaguely deprecated.
9818 Perl_sv_reset(pTHX_ const char *s, HV *const stash)
9820 PERL_ARGS_ASSERT_SV_RESET;
9822 sv_resetpvn(*s ? s : NULL, strlen(s), stash);
9826 Perl_sv_resetpvn(pTHX_ const char *s, STRLEN len, HV * const stash)
9828 char todo[PERL_UCHAR_MAX+1];
9831 if (!stash || SvTYPE(stash) != SVt_PVHV)
9834 if (!s) { /* reset ?? searches */
9835 MAGIC * const mg = mg_find((const SV *)stash, PERL_MAGIC_symtab);
9837 const U32 count = mg->mg_len / sizeof(PMOP**);
9838 PMOP **pmp = (PMOP**) mg->mg_ptr;
9839 PMOP *const *const end = pmp + count;
9843 SvREADONLY_off(PL_regex_pad[(*pmp)->op_pmoffset]);
9845 (*pmp)->op_pmflags &= ~PMf_USED;
9853 /* reset variables */
9855 if (!HvARRAY(stash))
9858 Zero(todo, 256, char);
9862 I32 i = (unsigned char)*s;
9866 max = (unsigned char)*s++;
9867 for ( ; i <= max; i++) {
9870 for (i = 0; i <= (I32) HvMAX(stash); i++) {
9872 for (entry = HvARRAY(stash)[i];
9874 entry = HeNEXT(entry))
9879 if (!todo[(U8)*HeKEY(entry)])
9881 gv = MUTABLE_GV(HeVAL(entry));
9885 if (sv && !SvREADONLY(sv)) {
9886 SV_CHECK_THINKFIRST_COW_DROP(sv);
9887 if (!isGV(sv)) SvOK_off(sv);
9892 if (GvHV(gv) && !HvNAME_get(GvHV(gv))) {
9903 Using various gambits, try to get an IO from an SV: the IO slot if its a
9904 GV; or the recursive result if we're an RV; or the IO slot of the symbol
9905 named after the PV if we're a string.
9907 'Get' magic is ignored on the C<sv> passed in, but will be called on
9908 C<SvRV(sv)> if C<sv> is an RV.
9914 Perl_sv_2io(pTHX_ SV *const sv)
9919 PERL_ARGS_ASSERT_SV_2IO;
9921 switch (SvTYPE(sv)) {
9923 io = MUTABLE_IO(sv);
9927 if (isGV_with_GP(sv)) {
9928 gv = MUTABLE_GV(sv);
9931 Perl_croak(aTHX_ "Bad filehandle: %" HEKf,
9932 HEKfARG(GvNAME_HEK(gv)));
9938 Perl_croak(aTHX_ PL_no_usym, "filehandle");
9940 SvGETMAGIC(SvRV(sv));
9941 return sv_2io(SvRV(sv));
9943 gv = gv_fetchsv_nomg(sv, 0, SVt_PVIO);
9950 if (SvGMAGICAL(sv)) {
9951 newsv = sv_newmortal();
9952 sv_setsv_nomg(newsv, sv);
9954 Perl_croak(aTHX_ "Bad filehandle: %" SVf, SVfARG(newsv));
9964 Using various gambits, try to get a CV from an SV; in addition, try if
9965 possible to set C<*st> and C<*gvp> to the stash and GV associated with it.
9966 The flags in C<lref> are passed to C<gv_fetchsv>.
9972 Perl_sv_2cv(pTHX_ SV *sv, HV **const st, GV **const gvp, const I32 lref)
9977 PERL_ARGS_ASSERT_SV_2CV;
9984 switch (SvTYPE(sv)) {
9988 return MUTABLE_CV(sv);
9998 sv = amagic_deref_call(sv, to_cv_amg);
10001 if (SvTYPE(sv) == SVt_PVCV) {
10002 cv = MUTABLE_CV(sv);
10007 else if(SvGETMAGIC(sv), isGV_with_GP(sv))
10008 gv = MUTABLE_GV(sv);
10010 Perl_croak(aTHX_ "Not a subroutine reference");
10012 else if (isGV_with_GP(sv)) {
10013 gv = MUTABLE_GV(sv);
10016 gv = gv_fetchsv_nomg(sv, lref, SVt_PVCV);
10023 /* Some flags to gv_fetchsv mean don't really create the GV */
10024 if (!isGV_with_GP(gv)) {
10028 *st = GvESTASH(gv);
10029 if (lref & ~GV_ADDMG && !GvCVu(gv)) {
10030 /* XXX this is probably not what they think they're getting.
10031 * It has the same effect as "sub name;", i.e. just a forward
10040 =for apidoc sv_true
10042 Returns true if the SV has a true value by Perl's rules.
10043 Use the C<SvTRUE> macro instead, which may call C<sv_true()> or may
10044 instead use an in-line version.
10050 Perl_sv_true(pTHX_ SV *const sv)
10055 const XPV* const tXpv = (XPV*)SvANY(sv);
10057 (tXpv->xpv_cur > 1 ||
10058 (tXpv->xpv_cur && *sv->sv_u.svu_pv != '0')))
10065 return SvIVX(sv) != 0;
10068 return SvNVX(sv) != 0.0;
10070 return sv_2bool(sv);
10076 =for apidoc sv_pvn_force
10078 Get a sensible string out of the SV somehow.
10079 A private implementation of the C<SvPV_force> macro for compilers which
10080 can't cope with complex macro expressions. Always use the macro instead.
10082 =for apidoc sv_pvn_force_flags
10084 Get a sensible string out of the SV somehow.
10085 If C<flags> has the C<SV_GMAGIC> bit set, will C<mg_get> on C<sv> if
10086 appropriate, else not. C<sv_pvn_force> and C<sv_pvn_force_nomg> are
10087 implemented in terms of this function.
10088 You normally want to use the various wrapper macros instead: see
10089 C<L</SvPV_force>> and C<L</SvPV_force_nomg>>.
10095 Perl_sv_pvn_force_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
10097 PERL_ARGS_ASSERT_SV_PVN_FORCE_FLAGS;
10099 if (flags & SV_GMAGIC) SvGETMAGIC(sv);
10100 if (SvTHINKFIRST(sv) && (!SvROK(sv) || SvREADONLY(sv)))
10101 sv_force_normal_flags(sv, 0);
10111 if (SvTYPE(sv) > SVt_PVLV
10112 || isGV_with_GP(sv))
10113 /* diag_listed_as: Can't coerce %s to %s in %s */
10114 Perl_croak(aTHX_ "Can't coerce %s to string in %s", sv_reftype(sv,0),
10116 s = sv_2pv_flags(sv, &len, flags &~ SV_GMAGIC);
10123 if (SvTYPE(sv) < SVt_PV ||
10124 s != SvPVX_const(sv)) { /* Almost, but not quite, sv_setpvn() */
10127 SvUPGRADE(sv, SVt_PV); /* Never FALSE */
10128 SvGROW(sv, len + 1);
10129 Move(s,SvPVX(sv),len,char);
10130 SvCUR_set(sv, len);
10131 SvPVX(sv)[len] = '\0';
10134 SvPOK_on(sv); /* validate pointer */
10136 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2pv(%s)\n",
10137 PTR2UV(sv),SvPVX_const(sv)));
10140 (void)SvPOK_only_UTF8(sv);
10141 return SvPVX_mutable(sv);
10145 =for apidoc sv_pvbyten_force
10147 The backend for the C<SvPVbytex_force> macro. Always use the macro
10154 Perl_sv_pvbyten_force(pTHX_ SV *const sv, STRLEN *const lp)
10156 PERL_ARGS_ASSERT_SV_PVBYTEN_FORCE;
10158 sv_pvn_force(sv,lp);
10159 sv_utf8_downgrade(sv,0);
10165 =for apidoc sv_pvutf8n_force
10167 The backend for the C<SvPVutf8x_force> macro. Always use the macro
10174 Perl_sv_pvutf8n_force(pTHX_ SV *const sv, STRLEN *const lp)
10176 PERL_ARGS_ASSERT_SV_PVUTF8N_FORCE;
10178 sv_pvn_force(sv,0);
10179 sv_utf8_upgrade_nomg(sv);
10185 =for apidoc sv_reftype
10187 Returns a string describing what the SV is a reference to.
10189 If ob is true and the SV is blessed, the string is the class name,
10190 otherwise it is the type of the SV, "SCALAR", "ARRAY" etc.
10196 Perl_sv_reftype(pTHX_ const SV *const sv, const int ob)
10198 PERL_ARGS_ASSERT_SV_REFTYPE;
10199 if (ob && SvOBJECT(sv)) {
10200 return SvPV_nolen_const(sv_ref(NULL, sv, ob));
10203 /* WARNING - There is code, for instance in mg.c, that assumes that
10204 * the only reason that sv_reftype(sv,0) would return a string starting
10205 * with 'L' or 'S' is that it is a LVALUE or a SCALAR.
10206 * Yes this a dodgy way to do type checking, but it saves practically reimplementing
10207 * this routine inside other subs, and it saves time.
10208 * Do not change this assumption without searching for "dodgy type check" in
10211 switch (SvTYPE(sv)) {
10226 case SVt_PVLV: return (char *) (SvROK(sv) ? "REF"
10227 /* tied lvalues should appear to be
10228 * scalars for backwards compatibility */
10229 : (isALPHA_FOLD_EQ(LvTYPE(sv), 't'))
10230 ? "SCALAR" : "LVALUE");
10231 case SVt_PVAV: return "ARRAY";
10232 case SVt_PVHV: return "HASH";
10233 case SVt_PVCV: return "CODE";
10234 case SVt_PVGV: return (char *) (isGV_with_GP(sv)
10235 ? "GLOB" : "SCALAR");
10236 case SVt_PVFM: return "FORMAT";
10237 case SVt_PVIO: return "IO";
10238 case SVt_INVLIST: return "INVLIST";
10239 case SVt_REGEXP: return "REGEXP";
10240 default: return "UNKNOWN";
10248 Returns a SV describing what the SV passed in is a reference to.
10250 dst can be a SV to be set to the description or NULL, in which case a
10251 mortal SV is returned.
10253 If ob is true and the SV is blessed, the description is the class
10254 name, otherwise it is the type of the SV, "SCALAR", "ARRAY" etc.
10260 Perl_sv_ref(pTHX_ SV *dst, const SV *const sv, const int ob)
10262 PERL_ARGS_ASSERT_SV_REF;
10265 dst = sv_newmortal();
10267 if (ob && SvOBJECT(sv)) {
10268 HvNAME_get(SvSTASH(sv))
10269 ? sv_sethek(dst, HvNAME_HEK(SvSTASH(sv)))
10270 : sv_setpvs(dst, "__ANON__");
10273 const char * reftype = sv_reftype(sv, 0);
10274 sv_setpv(dst, reftype);
10280 =for apidoc sv_isobject
10282 Returns a boolean indicating whether the SV is an RV pointing to a blessed
10283 object. If the SV is not an RV, or if the object is not blessed, then this
10290 Perl_sv_isobject(pTHX_ SV *sv)
10306 Returns a boolean indicating whether the SV is blessed into the specified
10307 class. This does not check for subtypes; use C<sv_derived_from> to verify
10308 an inheritance relationship.
10314 Perl_sv_isa(pTHX_ SV *sv, const char *const name)
10316 const char *hvname;
10318 PERL_ARGS_ASSERT_SV_ISA;
10328 hvname = HvNAME_get(SvSTASH(sv));
10332 return strEQ(hvname, name);
10336 =for apidoc newSVrv
10338 Creates a new SV for the existing RV, C<rv>, to point to. If C<rv> is not an
10339 RV then it will be upgraded to one. If C<classname> is non-null then the new
10340 SV will be blessed in the specified package. The new SV is returned and its
10341 reference count is 1. The reference count 1 is owned by C<rv>. See also
10342 newRV_inc() and newRV_noinc() for creating a new RV properly.
10348 Perl_newSVrv(pTHX_ SV *const rv, const char *const classname)
10352 PERL_ARGS_ASSERT_NEWSVRV;
10356 SV_CHECK_THINKFIRST_COW_DROP(rv);
10358 if (UNLIKELY( SvTYPE(rv) >= SVt_PVMG )) {
10359 const U32 refcnt = SvREFCNT(rv);
10363 SvREFCNT(rv) = refcnt;
10365 sv_upgrade(rv, SVt_IV);
10366 } else if (SvROK(rv)) {
10367 SvREFCNT_dec(SvRV(rv));
10369 prepare_SV_for_RV(rv);
10377 HV* const stash = gv_stashpv(classname, GV_ADD);
10378 (void)sv_bless(rv, stash);
10384 Perl_newSVavdefelem(pTHX_ AV *av, SSize_t ix, bool extendible)
10386 SV * const lv = newSV_type(SVt_PVLV);
10387 PERL_ARGS_ASSERT_NEWSVAVDEFELEM;
10389 sv_magic(lv, NULL, PERL_MAGIC_defelem, NULL, 0);
10390 LvTARG(lv) = SvREFCNT_inc_simple_NN(av);
10391 LvSTARGOFF(lv) = ix;
10392 LvTARGLEN(lv) = extendible ? 1 : (STRLEN)UV_MAX;
10397 =for apidoc sv_setref_pv
10399 Copies a pointer into a new SV, optionally blessing the SV. The C<rv>
10400 argument will be upgraded to an RV. That RV will be modified to point to
10401 the new SV. If the C<pv> argument is C<NULL>, then C<PL_sv_undef> will be placed
10402 into the SV. The C<classname> argument indicates the package for the
10403 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10404 will have a reference count of 1, and the RV will be returned.
10406 Do not use with other Perl types such as HV, AV, SV, CV, because those
10407 objects will become corrupted by the pointer copy process.
10409 Note that C<sv_setref_pvn> copies the string while this copies the pointer.
10415 Perl_sv_setref_pv(pTHX_ SV *const rv, const char *const classname, void *const pv)
10417 PERL_ARGS_ASSERT_SV_SETREF_PV;
10424 sv_setiv(newSVrv(rv,classname), PTR2IV(pv));
10429 =for apidoc sv_setref_iv
10431 Copies an integer into a new SV, optionally blessing the SV. The C<rv>
10432 argument will be upgraded to an RV. That RV will be modified to point to
10433 the new SV. The C<classname> argument indicates the package for the
10434 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10435 will have a reference count of 1, and the RV will be returned.
10441 Perl_sv_setref_iv(pTHX_ SV *const rv, const char *const classname, const IV iv)
10443 PERL_ARGS_ASSERT_SV_SETREF_IV;
10445 sv_setiv(newSVrv(rv,classname), iv);
10450 =for apidoc sv_setref_uv
10452 Copies an unsigned integer into a new SV, optionally blessing the SV. The C<rv>
10453 argument will be upgraded to an RV. That RV will be modified to point to
10454 the new SV. The C<classname> argument indicates the package for the
10455 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10456 will have a reference count of 1, and the RV will be returned.
10462 Perl_sv_setref_uv(pTHX_ SV *const rv, const char *const classname, const UV uv)
10464 PERL_ARGS_ASSERT_SV_SETREF_UV;
10466 sv_setuv(newSVrv(rv,classname), uv);
10471 =for apidoc sv_setref_nv
10473 Copies a double into a new SV, optionally blessing the SV. The C<rv>
10474 argument will be upgraded to an RV. That RV will be modified to point to
10475 the new SV. The C<classname> argument indicates the package for the
10476 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10477 will have a reference count of 1, and the RV will be returned.
10483 Perl_sv_setref_nv(pTHX_ SV *const rv, const char *const classname, const NV nv)
10485 PERL_ARGS_ASSERT_SV_SETREF_NV;
10487 sv_setnv(newSVrv(rv,classname), nv);
10492 =for apidoc sv_setref_pvn
10494 Copies a string into a new SV, optionally blessing the SV. The length of the
10495 string must be specified with C<n>. The C<rv> argument will be upgraded to
10496 an RV. That RV will be modified to point to the new SV. The C<classname>
10497 argument indicates the package for the blessing. Set C<classname> to
10498 C<NULL> to avoid the blessing. The new SV will have a reference count
10499 of 1, and the RV will be returned.
10501 Note that C<sv_setref_pv> copies the pointer while this copies the string.
10507 Perl_sv_setref_pvn(pTHX_ SV *const rv, const char *const classname,
10508 const char *const pv, const STRLEN n)
10510 PERL_ARGS_ASSERT_SV_SETREF_PVN;
10512 sv_setpvn(newSVrv(rv,classname), pv, n);
10517 =for apidoc sv_bless
10519 Blesses an SV into a specified package. The SV must be an RV. The package
10520 must be designated by its stash (see C<L</gv_stashpv>>). The reference count
10521 of the SV is unaffected.
10527 Perl_sv_bless(pTHX_ SV *const sv, HV *const stash)
10530 HV *oldstash = NULL;
10532 PERL_ARGS_ASSERT_SV_BLESS;
10536 Perl_croak(aTHX_ "Can't bless non-reference value");
10538 if (SvFLAGS(tmpRef) & (SVs_OBJECT|SVf_READONLY|SVf_PROTECT)) {
10539 if (SvREADONLY(tmpRef))
10540 Perl_croak_no_modify();
10541 if (SvOBJECT(tmpRef)) {
10542 oldstash = SvSTASH(tmpRef);
10545 SvOBJECT_on(tmpRef);
10546 SvUPGRADE(tmpRef, SVt_PVMG);
10547 SvSTASH_set(tmpRef, MUTABLE_HV(SvREFCNT_inc_simple(stash)));
10548 SvREFCNT_dec(oldstash);
10550 if(SvSMAGICAL(tmpRef))
10551 if(mg_find(tmpRef, PERL_MAGIC_ext) || mg_find(tmpRef, PERL_MAGIC_uvar))
10559 /* Downgrades a PVGV to a PVMG. If it's actually a PVLV, we leave the type
10560 * as it is after unglobbing it.
10563 PERL_STATIC_INLINE void
10564 S_sv_unglob(pTHX_ SV *const sv, U32 flags)
10568 SV * const temp = flags & SV_COW_DROP_PV ? NULL : sv_newmortal();
10570 PERL_ARGS_ASSERT_SV_UNGLOB;
10572 assert(SvTYPE(sv) == SVt_PVGV || SvTYPE(sv) == SVt_PVLV);
10574 if (!(flags & SV_COW_DROP_PV))
10575 gv_efullname3(temp, MUTABLE_GV(sv), "*");
10577 SvREFCNT_inc_simple_void_NN(sv_2mortal(sv));
10579 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
10580 && HvNAME_get(stash))
10581 mro_method_changed_in(stash);
10582 gp_free(MUTABLE_GV(sv));
10585 sv_del_backref(MUTABLE_SV(GvSTASH(sv)), sv);
10586 GvSTASH(sv) = NULL;
10589 if (GvNAME_HEK(sv)) {
10590 unshare_hek(GvNAME_HEK(sv));
10592 isGV_with_GP_off(sv);
10594 if(SvTYPE(sv) == SVt_PVGV) {
10595 /* need to keep SvANY(sv) in the right arena */
10596 xpvmg = new_XPVMG();
10597 StructCopy(SvANY(sv), xpvmg, XPVMG);
10598 del_XPVGV(SvANY(sv));
10601 SvFLAGS(sv) &= ~SVTYPEMASK;
10602 SvFLAGS(sv) |= SVt_PVMG;
10605 /* Intentionally not calling any local SET magic, as this isn't so much a
10606 set operation as merely an internal storage change. */
10607 if (flags & SV_COW_DROP_PV) SvOK_off(sv);
10608 else sv_setsv_flags(sv, temp, 0);
10610 if ((const GV *)sv == PL_last_in_gv)
10611 PL_last_in_gv = NULL;
10612 else if ((const GV *)sv == PL_statgv)
10617 =for apidoc sv_unref_flags
10619 Unsets the RV status of the SV, and decrements the reference count of
10620 whatever was being referenced by the RV. This can almost be thought of
10621 as a reversal of C<newSVrv>. The C<cflags> argument can contain
10622 C<SV_IMMEDIATE_UNREF> to force the reference count to be decremented
10623 (otherwise the decrementing is conditional on the reference count being
10624 different from one or the reference being a readonly SV).
10625 See C<L</SvROK_off>>.
10627 =for apidoc Amnh||SV_IMMEDIATE_UNREF
10633 Perl_sv_unref_flags(pTHX_ SV *const ref, const U32 flags)
10635 SV* const target = SvRV(ref);
10637 PERL_ARGS_ASSERT_SV_UNREF_FLAGS;
10639 if (SvWEAKREF(ref)) {
10640 sv_del_backref(target, ref);
10641 SvWEAKREF_off(ref);
10642 SvRV_set(ref, NULL);
10645 SvRV_set(ref, NULL);
10647 /* You can't have a || SvREADONLY(target) here, as $a = $$a, where $a was
10648 assigned to as BEGIN {$a = \"Foo"} will fail. */
10649 if (SvREFCNT(target) != 1 || (flags & SV_IMMEDIATE_UNREF))
10650 SvREFCNT_dec_NN(target);
10651 else /* XXX Hack, but hard to make $a=$a->[1] work otherwise */
10652 sv_2mortal(target); /* Schedule for freeing later */
10656 =for apidoc sv_untaint
10658 Untaint an SV. Use C<SvTAINTED_off> instead.
10664 Perl_sv_untaint(pTHX_ SV *const sv)
10666 PERL_ARGS_ASSERT_SV_UNTAINT;
10667 PERL_UNUSED_CONTEXT;
10669 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10670 MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10677 =for apidoc sv_tainted
10679 Test an SV for taintedness. Use C<SvTAINTED> instead.
10685 Perl_sv_tainted(pTHX_ SV *const sv)
10687 PERL_ARGS_ASSERT_SV_TAINTED;
10688 PERL_UNUSED_CONTEXT;
10690 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10691 const MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10692 if (mg && (mg->mg_len & 1) )
10698 #ifndef NO_MATHOMS /* Can't move these to mathoms.c because call uiv_2buf(),
10699 private to this file */
10702 =for apidoc sv_setpviv
10704 Copies an integer into the given SV, also updating its string value.
10705 Does not handle 'set' magic. See C<L</sv_setpviv_mg>>.
10711 Perl_sv_setpviv(pTHX_ SV *const sv, const IV iv)
10713 /* The purpose of this union is to ensure that arr is aligned on
10714 a 2 byte boundary, because that is what uiv_2buf() requires */
10716 char arr[TYPE_CHARS(UV)];
10720 char * const ptr = uiv_2buf(buf.arr, iv, 0, 0, &ebuf);
10722 PERL_ARGS_ASSERT_SV_SETPVIV;
10724 sv_setpvn(sv, ptr, ebuf - ptr);
10728 =for apidoc sv_setpviv_mg
10730 Like C<sv_setpviv>, but also handles 'set' magic.
10736 Perl_sv_setpviv_mg(pTHX_ SV *const sv, const IV iv)
10738 PERL_ARGS_ASSERT_SV_SETPVIV_MG;
10740 GCC_DIAG_IGNORE_STMT(-Wdeprecated-declarations);
10742 sv_setpviv(sv, iv);
10744 GCC_DIAG_RESTORE_STMT;
10749 #endif /* NO_MATHOMS */
10751 #if defined(PERL_IMPLICIT_CONTEXT)
10753 /* pTHX_ magic can't cope with varargs, so this is a no-context
10754 * version of the main function, (which may itself be aliased to us).
10755 * Don't access this version directly.
10759 Perl_sv_setpvf_nocontext(SV *const sv, const char *const pat, ...)
10764 PERL_ARGS_ASSERT_SV_SETPVF_NOCONTEXT;
10766 va_start(args, pat);
10767 sv_vsetpvf(sv, pat, &args);
10771 /* pTHX_ magic can't cope with varargs, so this is a no-context
10772 * version of the main function, (which may itself be aliased to us).
10773 * Don't access this version directly.
10777 Perl_sv_setpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10782 PERL_ARGS_ASSERT_SV_SETPVF_MG_NOCONTEXT;
10784 va_start(args, pat);
10785 sv_vsetpvf_mg(sv, pat, &args);
10791 =for apidoc sv_setpvf
10793 Works like C<sv_catpvf> but copies the text into the SV instead of
10794 appending it. Does not handle 'set' magic. See C<L</sv_setpvf_mg>>.
10800 Perl_sv_setpvf(pTHX_ SV *const sv, const char *const pat, ...)
10804 PERL_ARGS_ASSERT_SV_SETPVF;
10806 va_start(args, pat);
10807 sv_vsetpvf(sv, pat, &args);
10812 =for apidoc sv_vsetpvf
10814 Works like C<sv_vcatpvf> but copies the text into the SV instead of
10815 appending it. Does not handle 'set' magic. See C<L</sv_vsetpvf_mg>>.
10817 Usually used via its frontend C<sv_setpvf>.
10823 Perl_sv_vsetpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10825 PERL_ARGS_ASSERT_SV_VSETPVF;
10827 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10831 =for apidoc sv_setpvf_mg
10833 Like C<sv_setpvf>, but also handles 'set' magic.
10839 Perl_sv_setpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10843 PERL_ARGS_ASSERT_SV_SETPVF_MG;
10845 va_start(args, pat);
10846 sv_vsetpvf_mg(sv, pat, &args);
10851 =for apidoc sv_vsetpvf_mg
10853 Like C<sv_vsetpvf>, but also handles 'set' magic.
10855 Usually used via its frontend C<sv_setpvf_mg>.
10861 Perl_sv_vsetpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10863 PERL_ARGS_ASSERT_SV_VSETPVF_MG;
10865 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10869 #if defined(PERL_IMPLICIT_CONTEXT)
10871 /* pTHX_ magic can't cope with varargs, so this is a no-context
10872 * version of the main function, (which may itself be aliased to us).
10873 * Don't access this version directly.
10877 Perl_sv_catpvf_nocontext(SV *const sv, const char *const pat, ...)
10882 PERL_ARGS_ASSERT_SV_CATPVF_NOCONTEXT;
10884 va_start(args, pat);
10885 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10889 /* pTHX_ magic can't cope with varargs, so this is a no-context
10890 * version of the main function, (which may itself be aliased to us).
10891 * Don't access this version directly.
10895 Perl_sv_catpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10900 PERL_ARGS_ASSERT_SV_CATPVF_MG_NOCONTEXT;
10902 va_start(args, pat);
10903 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10910 =for apidoc sv_catpvf
10912 Processes its arguments like C<sprintf>, and appends the formatted
10913 output to an SV. As with C<sv_vcatpvfn> called with a non-null C-style
10914 variable argument list, argument reordering is not supported.
10915 If the appended data contains "wide" characters
10916 (including, but not limited to, SVs with a UTF-8 PV formatted with C<%s>,
10917 and characters >255 formatted with C<%c>), the original SV might get
10918 upgraded to UTF-8. Handles 'get' magic, but not 'set' magic. See
10919 C<L</sv_catpvf_mg>>. If the original SV was UTF-8, the pattern should be
10920 valid UTF-8; if the original SV was bytes, the pattern should be too.
10925 Perl_sv_catpvf(pTHX_ SV *const sv, const char *const pat, ...)
10929 PERL_ARGS_ASSERT_SV_CATPVF;
10931 va_start(args, pat);
10932 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10937 =for apidoc sv_vcatpvf
10939 Processes its arguments like C<sv_vcatpvfn> called with a non-null C-style
10940 variable argument list, and appends the formatted output
10941 to an SV. Does not handle 'set' magic. See C<L</sv_vcatpvf_mg>>.
10943 Usually used via its frontend C<sv_catpvf>.
10949 Perl_sv_vcatpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10951 PERL_ARGS_ASSERT_SV_VCATPVF;
10953 sv_vcatpvfn_flags(sv, pat, strlen(pat), args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10957 =for apidoc sv_catpvf_mg
10959 Like C<sv_catpvf>, but also handles 'set' magic.
10965 Perl_sv_catpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10969 PERL_ARGS_ASSERT_SV_CATPVF_MG;
10971 va_start(args, pat);
10972 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10978 =for apidoc sv_vcatpvf_mg
10980 Like C<sv_vcatpvf>, but also handles 'set' magic.
10982 Usually used via its frontend C<sv_catpvf_mg>.
10988 Perl_sv_vcatpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10990 PERL_ARGS_ASSERT_SV_VCATPVF_MG;
10992 sv_vcatpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10997 =for apidoc sv_vsetpvfn
10999 Works like C<sv_vcatpvfn> but copies the text into the SV instead of
11002 Usually used via one of its frontends C<sv_vsetpvf> and C<sv_vsetpvf_mg>.
11008 Perl_sv_vsetpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11009 va_list *const args, SV **const svargs, const Size_t sv_count, bool *const maybe_tainted)
11011 PERL_ARGS_ASSERT_SV_VSETPVFN;
11014 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, sv_count, maybe_tainted, 0);
11018 /* simplified inline Perl_sv_catpvn_nomg() when you know the SV's SvPOK */
11020 PERL_STATIC_INLINE void
11021 S_sv_catpvn_simple(pTHX_ SV *const sv, const char* const buf, const STRLEN len)
11023 STRLEN const need = len + SvCUR(sv) + 1;
11026 /* can't wrap as both len and SvCUR() are allocated in
11027 * memory and together can't consume all the address space
11029 assert(need > len);
11034 Copy(buf, end, len, char);
11037 SvCUR_set(sv, need - 1);
11042 * Warn of missing argument to sprintf. The value used in place of such
11043 * arguments should be &PL_sv_no; an undefined value would yield
11044 * inappropriate "use of uninit" warnings [perl #71000].
11047 S_warn_vcatpvfn_missing_argument(pTHX) {
11048 if (ckWARN(WARN_MISSING)) {
11049 Perl_warner(aTHX_ packWARN(WARN_MISSING), "Missing argument in %s",
11050 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11059 Perl_croak(aTHX_ "Integer overflow in format string for %s",
11060 (PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn"));
11064 /* Given an int i from the next arg (if args is true) or an sv from an arg
11065 * (if args is false), try to extract a STRLEN-ranged value from the arg,
11066 * with overflow checking.
11067 * Sets *neg to true if the value was negative (untouched otherwise.
11068 * Returns the absolute value.
11069 * As an extra margin of safety, it croaks if the returned value would
11070 * exceed the maximum value of a STRLEN / 4.
11074 S_sprintf_arg_num_val(pTHX_ va_list *const args, int i, SV *sv, bool *neg)
11088 if (UNLIKELY(SvIsUV(sv))) {
11089 UV uv = SvUV_nomg(sv);
11091 S_croak_overflow();
11095 iv = SvIV_nomg(sv);
11099 S_croak_overflow();
11105 if (iv > (IV)(((STRLEN)~0) / 4))
11106 S_croak_overflow();
11111 /* Read in and return a number. Updates *pattern to point to the char
11112 * following the number. Expects the first char to 1..9.
11113 * Croaks if the number exceeds 1/4 of the maximum value of STRLEN.
11114 * This is a belt-and-braces safety measure to complement any
11115 * overflow/wrap checks done in the main body of sv_vcatpvfn_flags.
11116 * It means that e.g. on a 32-bit system the width/precision can't be more
11117 * than 1G, which seems reasonable.
11121 S_expect_number(pTHX_ const char **const pattern)
11125 PERL_ARGS_ASSERT_EXPECT_NUMBER;
11127 assert(inRANGE(**pattern, '1', '9'));
11129 var = *(*pattern)++ - '0';
11130 while (isDIGIT(**pattern)) {
11131 /* if var * 10 + 9 would exceed 1/4 max strlen, croak */
11132 if (var > ((((STRLEN)~0) / 4 - 9) / 10))
11133 S_croak_overflow();
11134 var = var * 10 + (*(*pattern)++ - '0');
11139 /* Implement a fast "%.0f": given a pointer to the end of a buffer (caller
11140 * ensures it's big enough), back fill it with the rounded integer part of
11141 * nv. Returns ptr to start of string, and sets *len to its length.
11142 * Returns NULL if not convertible.
11146 S_F0convert(NV nv, char *const endbuf, STRLEN *const len)
11148 const int neg = nv < 0;
11151 PERL_ARGS_ASSERT_F0CONVERT;
11153 assert(!Perl_isinfnan(nv));
11156 if (nv != 0.0 && nv < UV_MAX) {
11162 if (uv & 1 && uv == nv)
11163 uv--; /* Round to even */
11166 const unsigned dig = uv % 10;
11168 } while (uv /= 10);
11178 /* XXX maybe_tainted is never assigned to, so the doc above is lying. */
11181 Perl_sv_vcatpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11182 va_list *const args, SV **const svargs, const Size_t sv_count, bool *const maybe_tainted)
11184 PERL_ARGS_ASSERT_SV_VCATPVFN;
11186 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, sv_count, maybe_tainted, SV_GMAGIC|SV_SMAGIC);
11190 /* For the vcatpvfn code, we need a long double target in case
11191 * HAS_LONG_DOUBLE, even without USE_LONG_DOUBLE, so that we can printf
11192 * with long double formats, even without NV being long double. But we
11193 * call the target 'fv' instead of 'nv', since most of the time it is not
11194 * (most compilers these days recognize "long double", even if only as a
11195 * synonym for "double").
11197 #if defined(HAS_LONG_DOUBLE) && LONG_DOUBLESIZE > DOUBLESIZE && \
11198 defined(PERL_PRIgldbl) && !defined(USE_QUADMATH)
11199 # define VCATPVFN_FV_GF PERL_PRIgldbl
11200 # if defined(__VMS) && defined(__ia64) && defined(__IEEE_FLOAT)
11201 /* Work around breakage in OTS$CVT_FLOAT_T_X */
11202 # define VCATPVFN_NV_TO_FV(nv,fv) \
11205 fv = Perl_isnan(_dv) ? LDBL_QNAN : _dv; \
11208 # define VCATPVFN_NV_TO_FV(nv,fv) (fv)=(nv)
11210 typedef long double vcatpvfn_long_double_t;
11212 # define VCATPVFN_FV_GF NVgf
11213 # define VCATPVFN_NV_TO_FV(nv,fv) (fv)=(nv)
11214 typedef NV vcatpvfn_long_double_t;
11217 #ifdef LONGDOUBLE_DOUBLEDOUBLE
11218 /* The first double can be as large as 2**1023, or '1' x '0' x 1023.
11219 * The second double can be as small as 2**-1074, or '0' x 1073 . '1'.
11220 * The sum of them can be '1' . '0' x 2096 . '1', with implied radix point
11221 * after the first 1023 zero bits.
11223 * XXX The 2098 is quite large (262.25 bytes) and therefore some sort
11224 * of dynamically growing buffer might be better, start at just 16 bytes
11225 * (for example) and grow only when necessary. Or maybe just by looking
11226 * at the exponents of the two doubles? */
11227 # define DOUBLEDOUBLE_MAXBITS 2098
11230 /* vhex will contain the values (0..15) of the hex digits ("nybbles"
11231 * of 4 bits); 1 for the implicit 1, and the mantissa bits, four bits
11232 * per xdigit. For the double-double case, this can be rather many.
11233 * The non-double-double-long-double overshoots since all bits of NV
11234 * are not mantissa bits, there are also exponent bits. */
11235 #ifdef LONGDOUBLE_DOUBLEDOUBLE
11236 # define VHEX_SIZE (3+DOUBLEDOUBLE_MAXBITS/4)
11238 # define VHEX_SIZE (1+(NVSIZE * 8)/4)
11241 /* If we do not have a known long double format, (including not using
11242 * long doubles, or long doubles being equal to doubles) then we will
11243 * fall back to the ldexp/frexp route, with which we can retrieve at
11244 * most as many bits as our widest unsigned integer type is. We try
11245 * to get a 64-bit unsigned integer even if we are not using a 64-bit UV.
11247 * (If you want to test the case of UVSIZE == 4, NVSIZE == 8,
11248 * set the MANTISSATYPE to int and the MANTISSASIZE to 4.)
11250 #if defined(HAS_QUAD) && defined(Uquad_t)
11251 # define MANTISSATYPE Uquad_t
11252 # define MANTISSASIZE 8
11254 # define MANTISSATYPE UV
11255 # define MANTISSASIZE UVSIZE
11258 #if defined(DOUBLE_LITTLE_ENDIAN) || defined(LONGDOUBLE_LITTLE_ENDIAN)
11259 # define HEXTRACT_LITTLE_ENDIAN
11260 #elif defined(DOUBLE_BIG_ENDIAN) || defined(LONGDOUBLE_BIG_ENDIAN)
11261 # define HEXTRACT_BIG_ENDIAN
11263 # define HEXTRACT_MIX_ENDIAN
11266 /* S_hextract() is a helper for S_format_hexfp, for extracting
11267 * the hexadecimal values (for %a/%A). The nv is the NV where the value
11268 * are being extracted from (either directly from the long double in-memory
11269 * presentation, or from the uquad computed via frexp+ldexp). frexp also
11270 * is used to update the exponent. The subnormal is set to true
11271 * for IEEE 754 subnormals/denormals (including the x86 80-bit format).
11272 * The vhex is the pointer to the beginning of the output buffer of VHEX_SIZE.
11274 * The tricky part is that S_hextract() needs to be called twice:
11275 * the first time with vend as NULL, and the second time with vend as
11276 * the pointer returned by the first call. What happens is that on
11277 * the first round the output size is computed, and the intended
11278 * extraction sanity checked. On the second round the actual output
11279 * (the extraction of the hexadecimal values) takes place.
11280 * Sanity failures cause fatal failures during both rounds. */
11282 S_hextract(pTHX_ const NV nv, int* exponent, bool *subnormal,
11283 U8* vhex, U8* vend)
11287 int ixmin = 0, ixmax = 0;
11289 /* XXX Inf/NaN are not handled here, since it is
11290 * assumed they are to be output as "Inf" and "NaN". */
11292 /* These macros are just to reduce typos, they have multiple
11293 * repetitions below, but usually only one (or sometimes two)
11294 * of them is really being used. */
11295 /* HEXTRACT_OUTPUT() extracts the high nybble first. */
11296 #define HEXTRACT_OUTPUT_HI(ix) (*v++ = nvp[ix] >> 4)
11297 #define HEXTRACT_OUTPUT_LO(ix) (*v++ = nvp[ix] & 0xF)
11298 #define HEXTRACT_OUTPUT(ix) \
11300 HEXTRACT_OUTPUT_HI(ix); HEXTRACT_OUTPUT_LO(ix); \
11302 #define HEXTRACT_COUNT(ix, c) \
11304 v += c; if (ix < ixmin) ixmin = ix; else if (ix > ixmax) ixmax = ix; \
11306 #define HEXTRACT_BYTE(ix) \
11308 if (vend) HEXTRACT_OUTPUT(ix); else HEXTRACT_COUNT(ix, 2); \
11310 #define HEXTRACT_LO_NYBBLE(ix) \
11312 if (vend) HEXTRACT_OUTPUT_LO(ix); else HEXTRACT_COUNT(ix, 1); \
11314 /* HEXTRACT_TOP_NYBBLE is just convenience disguise,
11315 * to make it look less odd when the top bits of a NV
11316 * are extracted using HEXTRACT_LO_NYBBLE: the highest
11317 * order bits can be in the "low nybble" of a byte. */
11318 #define HEXTRACT_TOP_NYBBLE(ix) HEXTRACT_LO_NYBBLE(ix)
11319 #define HEXTRACT_BYTES_LE(a, b) \
11320 for (ix = a; ix >= b; ix--) { HEXTRACT_BYTE(ix); }
11321 #define HEXTRACT_BYTES_BE(a, b) \
11322 for (ix = a; ix <= b; ix++) { HEXTRACT_BYTE(ix); }
11323 #define HEXTRACT_GET_SUBNORMAL(nv) *subnormal = Perl_fp_class_denorm(nv)
11324 #define HEXTRACT_IMPLICIT_BIT(nv) \
11326 if (!*subnormal) { \
11327 if (vend) *v++ = ((nv) == 0.0) ? 0 : 1; else v++; \
11331 /* Most formats do. Those which don't should undef this.
11333 * But also note that IEEE 754 subnormals do not have it, or,
11334 * expressed alternatively, their implicit bit is zero. */
11335 #define HEXTRACT_HAS_IMPLICIT_BIT
11337 /* Many formats do. Those which don't should undef this. */
11338 #define HEXTRACT_HAS_TOP_NYBBLE
11340 /* HEXTRACTSIZE is the maximum number of xdigits. */
11341 #if defined(USE_LONG_DOUBLE) && defined(LONGDOUBLE_DOUBLEDOUBLE)
11342 # define HEXTRACTSIZE (2+DOUBLEDOUBLE_MAXBITS/4)
11344 # define HEXTRACTSIZE 2 * NVSIZE
11347 const U8* vmaxend = vhex + HEXTRACTSIZE;
11349 assert(HEXTRACTSIZE <= VHEX_SIZE);
11351 PERL_UNUSED_VAR(ix); /* might happen */
11352 (void)Perl_frexp(PERL_ABS(nv), exponent);
11353 *subnormal = FALSE;
11354 if (vend && (vend <= vhex || vend > vmaxend)) {
11355 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11356 Perl_croak(aTHX_ "Hexadecimal float: internal error (entry)");
11359 /* First check if using long doubles. */
11360 #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE)
11361 # if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN
11362 /* Used in e.g. VMS and HP-UX IA-64, e.g. -0.1L:
11363 * 9a 99 99 99 99 99 99 99 99 99 99 99 99 99 fb bf */
11364 /* The bytes 13..0 are the mantissa/fraction,
11365 * the 15,14 are the sign+exponent. */
11366 const U8* nvp = (const U8*)(&nv);
11367 HEXTRACT_GET_SUBNORMAL(nv);
11368 HEXTRACT_IMPLICIT_BIT(nv);
11369 # undef HEXTRACT_HAS_TOP_NYBBLE
11370 HEXTRACT_BYTES_LE(13, 0);
11371 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_BIG_ENDIAN
11372 /* Used in e.g. Solaris Sparc and HP-UX PA-RISC, e.g. -0.1L:
11373 * bf fb 99 99 99 99 99 99 99 99 99 99 99 99 99 9a */
11374 /* The bytes 2..15 are the mantissa/fraction,
11375 * the 0,1 are the sign+exponent. */
11376 const U8* nvp = (const U8*)(&nv);
11377 HEXTRACT_GET_SUBNORMAL(nv);
11378 HEXTRACT_IMPLICIT_BIT(nv);
11379 # undef HEXTRACT_HAS_TOP_NYBBLE
11380 HEXTRACT_BYTES_BE(2, 15);
11381 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN
11382 /* x86 80-bit "extended precision", 64 bits of mantissa / fraction /
11383 * significand, 15 bits of exponent, 1 bit of sign. No implicit bit.
11384 * NVSIZE can be either 12 (ILP32, Solaris x86) or 16 (LP64, Linux
11385 * and OS X), meaning that 2 or 6 bytes are empty padding. */
11386 /* The bytes 0..1 are the sign+exponent,
11387 * the bytes 2..9 are the mantissa/fraction. */
11388 const U8* nvp = (const U8*)(&nv);
11389 # undef HEXTRACT_HAS_IMPLICIT_BIT
11390 # undef HEXTRACT_HAS_TOP_NYBBLE
11391 HEXTRACT_GET_SUBNORMAL(nv);
11392 HEXTRACT_BYTES_LE(7, 0);
11393 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN
11394 /* Does this format ever happen? (Wikipedia says the Motorola
11395 * 6888x math coprocessors used format _like_ this but padded
11396 * to 96 bits with 16 unused bits between the exponent and the
11398 const U8* nvp = (const U8*)(&nv);
11399 # undef HEXTRACT_HAS_IMPLICIT_BIT
11400 # undef HEXTRACT_HAS_TOP_NYBBLE
11401 HEXTRACT_GET_SUBNORMAL(nv);
11402 HEXTRACT_BYTES_BE(0, 7);
11404 # define HEXTRACT_FALLBACK
11405 /* Double-double format: two doubles next to each other.
11406 * The first double is the high-order one, exactly like
11407 * it would be for a "lone" double. The second double
11408 * is shifted down using the exponent so that that there
11409 * are no common bits. The tricky part is that the value
11410 * of the double-double is the SUM of the two doubles and
11411 * the second one can be also NEGATIVE.
11413 * Because of this tricky construction the bytewise extraction we
11414 * use for the other long double formats doesn't work, we must
11415 * extract the values bit by bit.
11417 * The little-endian double-double is used .. somewhere?
11419 * The big endian double-double is used in e.g. PPC/Power (AIX)
11422 * The mantissa bits are in two separate stretches, e.g. for -0.1L:
11423 * 9a 99 99 99 99 99 59 bc 9a 99 99 99 99 99 b9 3f (LE)
11424 * 3f b9 99 99 99 99 99 9a bc 59 99 99 99 99 99 9a (BE)
11427 #else /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) */
11428 /* Using normal doubles, not long doubles.
11430 * We generate 4-bit xdigits (nybble/nibble) instead of 8-bit
11431 * bytes, since we might need to handle printf precision, and
11432 * also need to insert the radix. */
11434 # ifdef HEXTRACT_LITTLE_ENDIAN
11435 /* 0 1 2 3 4 5 6 7 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11436 const U8* nvp = (const U8*)(&nv);
11437 HEXTRACT_GET_SUBNORMAL(nv);
11438 HEXTRACT_IMPLICIT_BIT(nv);
11439 HEXTRACT_TOP_NYBBLE(6);
11440 HEXTRACT_BYTES_LE(5, 0);
11441 # elif defined(HEXTRACT_BIG_ENDIAN)
11442 /* 7 6 5 4 3 2 1 0 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11443 const U8* nvp = (const U8*)(&nv);
11444 HEXTRACT_GET_SUBNORMAL(nv);
11445 HEXTRACT_IMPLICIT_BIT(nv);
11446 HEXTRACT_TOP_NYBBLE(1);
11447 HEXTRACT_BYTES_BE(2, 7);
11448 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_LE_BE
11449 /* 4 5 6 7 0 1 2 3 (MSB = 7, LSB = 0, 6:7 = nybble:exponent:sign) */
11450 const U8* nvp = (const U8*)(&nv);
11451 HEXTRACT_GET_SUBNORMAL(nv);
11452 HEXTRACT_IMPLICIT_BIT(nv);
11453 HEXTRACT_TOP_NYBBLE(2); /* 6 */
11454 HEXTRACT_BYTE(1); /* 5 */
11455 HEXTRACT_BYTE(0); /* 4 */
11456 HEXTRACT_BYTE(7); /* 3 */
11457 HEXTRACT_BYTE(6); /* 2 */
11458 HEXTRACT_BYTE(5); /* 1 */
11459 HEXTRACT_BYTE(4); /* 0 */
11460 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_BE_LE
11461 /* 3 2 1 0 7 6 5 4 (MSB = 7, LSB = 0, 7:6 = sign:exponent:nybble) */
11462 const U8* nvp = (const U8*)(&nv);
11463 HEXTRACT_GET_SUBNORMAL(nv);
11464 HEXTRACT_IMPLICIT_BIT(nv);
11465 HEXTRACT_TOP_NYBBLE(5); /* 6 */
11466 HEXTRACT_BYTE(6); /* 5 */
11467 HEXTRACT_BYTE(7); /* 4 */
11468 HEXTRACT_BYTE(0); /* 3 */
11469 HEXTRACT_BYTE(1); /* 2 */
11470 HEXTRACT_BYTE(2); /* 1 */
11471 HEXTRACT_BYTE(3); /* 0 */
11473 # define HEXTRACT_FALLBACK
11476 # define HEXTRACT_FALLBACK
11478 #endif /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) #else */
11480 #ifdef HEXTRACT_FALLBACK
11481 HEXTRACT_GET_SUBNORMAL(nv);
11482 # undef HEXTRACT_HAS_TOP_NYBBLE /* Meaningless, but consistent. */
11483 /* The fallback is used for the double-double format, and
11484 * for unknown long double formats, and for unknown double
11485 * formats, or in general unknown NV formats. */
11486 if (nv == (NV)0.0) {
11494 NV d = nv < 0 ? -nv : nv;
11496 U8 ha = 0x0; /* hexvalue accumulator */
11497 U8 hd = 0x8; /* hexvalue digit */
11499 /* Shift d and e (and update exponent) so that e <= d < 2*e,
11500 * this is essentially manual frexp(). Multiplying by 0.5 and
11501 * doubling should be lossless in binary floating point. */
11511 while (d >= e + e) {
11515 /* Now e <= d < 2*e */
11517 /* First extract the leading hexdigit (the implicit bit). */
11533 /* Then extract the remaining hexdigits. */
11534 while (d > (NV)0.0) {
11540 /* Output or count in groups of four bits,
11541 * that is, when the hexdigit is down to one. */
11546 /* Reset the hexvalue. */
11555 /* Flush possible pending hexvalue. */
11565 /* Croak for various reasons: if the output pointer escaped the
11566 * output buffer, if the extraction index escaped the extraction
11567 * buffer, or if the ending output pointer didn't match the
11568 * previously computed value. */
11569 if (v <= vhex || v - vhex >= VHEX_SIZE ||
11570 /* For double-double the ixmin and ixmax stay at zero,
11571 * which is convenient since the HEXTRACTSIZE is tricky
11572 * for double-double. */
11573 ixmin < 0 || ixmax >= NVSIZE ||
11574 (vend && v != vend)) {
11575 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11576 Perl_croak(aTHX_ "Hexadecimal float: internal error (overflow)");
11582 /* S_format_hexfp(): helper function for Perl_sv_vcatpvfn_flags().
11584 * Processes the %a/%A hexadecimal floating-point format, since the
11585 * built-in snprintf()s which are used for most of the f/p formats, don't
11586 * universally handle %a/%A.
11587 * Populates buf of length bufsize, and returns the length of the created
11589 * The rest of the args have the same meaning as the local vars of the
11590 * same name within Perl_sv_vcatpvfn_flags().
11592 * The caller's determination of IN_LC(LC_NUMERIC), passed as in_lc_numeric,
11593 * is used to ensure we do the right thing when we need to access the locale's
11596 * It requires the caller to make buf large enough.
11600 S_format_hexfp(pTHX_ char * const buf, const STRLEN bufsize, const char c,
11601 const NV nv, const vcatpvfn_long_double_t fv,
11602 bool has_precis, STRLEN precis, STRLEN width,
11603 bool alt, char plus, bool left, bool fill, bool in_lc_numeric)
11605 /* Hexadecimal floating point. */
11607 U8 vhex[VHEX_SIZE];
11608 U8* v = vhex; /* working pointer to vhex */
11609 U8* vend; /* pointer to one beyond last digit of vhex */
11610 U8* vfnz = NULL; /* first non-zero */
11611 U8* vlnz = NULL; /* last non-zero */
11612 U8* v0 = NULL; /* first output */
11613 const bool lower = (c == 'a');
11614 /* At output the values of vhex (up to vend) will
11615 * be mapped through the xdig to get the actual
11616 * human-readable xdigits. */
11617 const char* xdig = PL_hexdigit;
11618 STRLEN zerotail = 0; /* how many extra zeros to append */
11619 int exponent = 0; /* exponent of the floating point input */
11620 bool hexradix = FALSE; /* should we output the radix */
11621 bool subnormal = FALSE; /* IEEE 754 subnormal/denormal */
11622 bool negative = FALSE;
11625 /* XXX: NaN, Inf -- though they are printed as "NaN" and "Inf".
11627 * For example with denormals, (assuming the vanilla
11628 * 64-bit double): the exponent is zero. 1xp-1074 is
11629 * the smallest denormal and the smallest double, it
11630 * could be output also as 0x0.0000000000001p-1022 to
11631 * match its internal structure. */
11633 vend = S_hextract(aTHX_ nv, &exponent, &subnormal, vhex, NULL);
11634 S_hextract(aTHX_ nv, &exponent, &subnormal, vhex, vend);
11636 #if NVSIZE > DOUBLESIZE
11637 # ifdef HEXTRACT_HAS_IMPLICIT_BIT
11638 /* In this case there is an implicit bit,
11639 * and therefore the exponent is shifted by one. */
11641 # elif defined(NV_X86_80_BIT)
11643 /* The subnormals of the x86-80 have a base exponent of -16382,
11644 * (while the physical exponent bits are zero) but the frexp()
11645 * returned the scientific-style floating exponent. We want
11646 * to map the last one as:
11647 * -16831..-16384 -> -16382 (the last normal is 0x1p-16382)
11648 * -16835..-16388 -> -16384
11649 * since we want to keep the first hexdigit
11650 * as one of the [8421]. */
11651 exponent = -4 * ( (exponent + 1) / -4) - 2;
11655 /* TBD: other non-implicit-bit platforms than the x86-80. */
11659 negative = fv < 0 || Perl_signbit(nv);
11670 xdig += 16; /* Use uppercase hex. */
11673 /* Find the first non-zero xdigit. */
11674 for (v = vhex; v < vend; v++) {
11682 /* Find the last non-zero xdigit. */
11683 for (v = vend - 1; v >= vhex; v--) {
11690 #if NVSIZE == DOUBLESIZE
11696 #ifndef NV_X86_80_BIT
11698 /* IEEE 754 subnormals (but not the x86 80-bit):
11699 * we want "normalize" the subnormal,
11700 * so we need to right shift the hex nybbles
11701 * so that the output of the subnormal starts
11702 * from the first true bit. (Another, equally
11703 * valid, policy would be to dump the subnormal
11704 * nybbles as-is, to display the "physical" layout.) */
11707 /* Find the ceil(log2(v[0])) of
11708 * the top non-zero nybble. */
11709 for (i = vfnz[0], n = 0; i > 1; i >>= 1, n++) { }
11713 for (vshr = vlnz; vshr >= vfnz; vshr--) {
11714 vshr[1] |= (vshr[0] & (0xF >> (4 - n))) << (4 - n);
11728 U8* ve = (subnormal ? vlnz + 1 : vend);
11729 SSize_t vn = ve - v0;
11731 if (precis < (Size_t)(vn - 1)) {
11732 bool overflow = FALSE;
11733 if (v0[precis + 1] < 0x8) {
11734 /* Round down, nothing to do. */
11735 } else if (v0[precis + 1] > 0x8) {
11738 overflow = v0[precis] > 0xF;
11740 } else { /* v0[precis] == 0x8 */
11741 /* Half-point: round towards the one
11742 * with the even least-significant digit:
11750 * 78 -> 8 f8 -> 10 */
11751 if ((v0[precis] & 0x1)) {
11754 overflow = v0[precis] > 0xF;
11759 for (v = v0 + precis - 1; v >= v0; v--) {
11761 overflow = *v > 0xF;
11767 if (v == v0 - 1 && overflow) {
11768 /* If the overflow goes all the
11769 * way to the front, we need to
11770 * insert 0x1 in front, and adjust
11772 Move(v0, v0 + 1, vn - 1, char);
11778 /* The new effective "last non zero". */
11779 vlnz = v0 + precis;
11783 subnormal ? precis - vn + 1 :
11784 precis - (vlnz - vhex);
11791 /* If there are non-zero xdigits, the radix
11792 * is output after the first one. */
11800 zerotail = has_precis ? precis : 0;
11803 /* The radix is always output if precis, or if alt. */
11804 if ((has_precis && precis > 0) || alt) {
11809 #ifndef USE_LOCALE_NUMERIC
11812 if (in_lc_numeric) {
11814 WITH_LC_NUMERIC_SET_TO_NEEDED_IN(TRUE, {
11815 const char* r = SvPV(PL_numeric_radix_sv, n);
11816 Copy(r, p, n, char);
11831 if (zerotail > 0) {
11832 while (zerotail--) {
11839 /* sanity checks */
11840 if (elen >= bufsize || width >= bufsize)
11841 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11842 Perl_croak(aTHX_ "Hexadecimal float: internal error (overflow)");
11844 elen += my_snprintf(p, bufsize - elen,
11845 "%c%+d", lower ? 'p' : 'P',
11848 if (elen < width) {
11849 STRLEN gap = (STRLEN)(width - elen);
11851 /* Pad the back with spaces. */
11852 memset(buf + elen, ' ', gap);
11855 /* Insert the zeros after the "0x" and the
11856 * the potential sign, but before the digits,
11857 * otherwise we end up with "0000xH.HHH...",
11858 * when we want "0x000H.HHH..." */
11859 STRLEN nzero = gap;
11860 char* zerox = buf + 2;
11861 STRLEN nmove = elen - 2;
11862 if (negative || plus) {
11866 Move(zerox, zerox + nzero, nmove, char);
11867 memset(zerox, fill ? '0' : ' ', nzero);
11870 /* Move it to the right. */
11871 Move(buf, buf + gap,
11873 /* Pad the front with spaces. */
11874 memset(buf, ' ', gap);
11883 =for apidoc sv_vcatpvfn
11885 =for apidoc sv_vcatpvfn_flags
11887 Processes its arguments like C<vsprintf> and appends the formatted output
11888 to an SV. Uses an array of SVs if the C-style variable argument list is
11889 missing (C<NULL>). Argument reordering (using format specifiers like C<%2$d>
11890 or C<%*2$d>) is supported only when using an array of SVs; using a C-style
11891 C<va_list> argument list with a format string that uses argument reordering
11892 will yield an exception.
11894 When running with taint checks enabled, indicates via
11895 C<maybe_tainted> if results are untrustworthy (often due to the use of
11898 If called as C<sv_vcatpvfn> or flags has the C<SV_GMAGIC> bit set, calls get magic.
11900 It assumes that pat has the same utf8-ness as sv. It's the caller's
11901 responsibility to ensure that this is so.
11903 Usually used via one of its frontends C<sv_vcatpvf> and C<sv_vcatpvf_mg>.
11910 Perl_sv_vcatpvfn_flags(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11911 va_list *const args, SV **const svargs, const Size_t sv_count, bool *const maybe_tainted,
11914 const char *fmtstart; /* character following the current '%' */
11915 const char *q; /* current position within format */
11916 const char *patend;
11919 static const char nullstr[] = "(null)";
11920 bool has_utf8 = DO_UTF8(sv); /* has the result utf8? */
11921 const bool pat_utf8 = has_utf8; /* the pattern is in utf8? */
11922 /* Times 4: a decimal digit takes more than 3 binary digits.
11923 * NV_DIG: mantissa takes that many decimal digits.
11924 * Plus 32: Playing safe. */
11925 char ebuf[IV_DIG * 4 + NV_DIG + 32];
11926 bool no_redundant_warning = FALSE; /* did we use any explicit format parameter index? */
11927 #ifdef USE_LOCALE_NUMERIC
11928 bool have_in_lc_numeric = FALSE;
11930 /* we never change this unless USE_LOCALE_NUMERIC */
11931 bool in_lc_numeric = FALSE;
11933 PERL_ARGS_ASSERT_SV_VCATPVFN_FLAGS;
11934 PERL_UNUSED_ARG(maybe_tainted);
11936 if (flags & SV_GMAGIC)
11939 /* no matter what, this is a string now */
11940 (void)SvPV_force_nomg(sv, origlen);
11942 /* the code that scans for flags etc following a % relies on
11943 * a '\0' being present to avoid falling off the end. Ideally that
11944 * should be fixed */
11945 assert(pat[patlen] == '\0');
11948 /* Special-case "", "%s", "%-p" (SVf - see below) and "%.0f".
11949 * In each case, if there isn't the correct number of args, instead
11950 * fall through to the main code to handle the issuing of any
11954 if (patlen == 0 && (args || sv_count == 0))
11957 if (patlen <= 4 && pat[0] == '%' && (args || sv_count == 1)) {
11960 if (patlen == 2 && pat[1] == 's') {
11962 const char * const s = va_arg(*args, char*);
11963 sv_catpv_nomg(sv, s ? s : nullstr);
11966 /* we want get magic on the source but not the target.
11967 * sv_catsv can't do that, though */
11968 SvGETMAGIC(*svargs);
11969 sv_catsv_nomg(sv, *svargs);
11976 if (patlen == 3 && pat[1] == '-' && pat[2] == 'p') {
11977 SV *asv = MUTABLE_SV(va_arg(*args, void*));
11978 sv_catsv_nomg(sv, asv);
11982 #if !defined(USE_LONG_DOUBLE) && !defined(USE_QUADMATH)
11983 /* special-case "%.0f" */
11984 else if ( patlen == 4
11985 && pat[1] == '.' && pat[2] == '0' && pat[3] == 'f')
11987 const NV nv = SvNV(*svargs);
11988 if (LIKELY(!Perl_isinfnan(nv))) {
11992 if ((p = F0convert(nv, ebuf + sizeof ebuf, &l))) {
11993 sv_catpvn_nomg(sv, p, l);
11998 #endif /* !USE_LONG_DOUBLE */
12002 patend = (char*)pat + patlen;
12003 for (fmtstart = pat; fmtstart < patend; fmtstart = q) {
12004 char intsize = 0; /* size qualifier in "%hi..." etc */
12005 bool alt = FALSE; /* has "%#..." */
12006 bool left = FALSE; /* has "%-..." */
12007 bool fill = FALSE; /* has "%0..." */
12008 char plus = 0; /* has "%+..." */
12009 STRLEN width = 0; /* value of "%NNN..." */
12010 bool has_precis = FALSE; /* has "%.NNN..." */
12011 STRLEN precis = 0; /* value of "%.NNN..." */
12012 int base = 0; /* base to print in, e.g. 8 for %o */
12013 UV uv = 0; /* the value to print of int-ish args */
12015 bool vectorize = FALSE; /* has "%v..." */
12016 bool vec_utf8 = FALSE; /* SvUTF8(vec arg) */
12017 const U8 *vecstr = NULL; /* SvPVX(vec arg) */
12018 STRLEN veclen = 0; /* SvCUR(vec arg) */
12019 const char *dotstr = NULL; /* separator string for %v */
12020 STRLEN dotstrlen; /* length of separator string for %v */
12022 Size_t efix = 0; /* explicit format parameter index */
12023 const Size_t osvix = svix; /* original index in case of bad fmt */
12026 bool is_utf8 = FALSE; /* is this item utf8? */
12027 bool arg_missing = FALSE; /* give "Missing argument" warning */
12028 char esignbuf[4]; /* holds sign prefix, e.g. "-0x" */
12029 STRLEN esignlen = 0; /* length of e.g. "-0x" */
12030 STRLEN zeros = 0; /* how many '0' to prepend */
12032 const char *eptr = NULL; /* the address of the element string */
12033 STRLEN elen = 0; /* the length of the element string */
12035 char c; /* the actual format ('d', s' etc) */
12038 /* echo everything up to the next format specification */
12039 for (q = fmtstart; q < patend && *q != '%'; ++q)
12042 if (q > fmtstart) {
12043 if (has_utf8 && !pat_utf8) {
12044 /* upgrade and copy the bytes of fmtstart..q-1 to utf8 on
12048 STRLEN need = SvCUR(sv) + (q - fmtstart) + 1;
12050 for (p = fmtstart; p < q; p++)
12051 if (!NATIVE_BYTE_IS_INVARIANT(*p))
12056 for (p = fmtstart; p < q; p++)
12057 append_utf8_from_native_byte((U8)*p, (U8**)&dst);
12059 SvCUR_set(sv, need - 1);
12062 S_sv_catpvn_simple(aTHX_ sv, fmtstart, q - fmtstart);
12067 fmtstart = q; /* fmtstart is char following the '%' */
12070 We allow format specification elements in this order:
12071 \d+\$ explicit format parameter index
12073 v|\*(\d+\$)?v vector with optional (optionally specified) arg
12074 0 flag (as above): repeated to allow "v02"
12075 \d+|\*(\d+\$)? width using optional (optionally specified) arg
12076 \.(\d*|\*(\d+\$)?) precision using optional (optionally specified) arg
12078 [%bcdefginopsuxDFOUX] format (mandatory)
12081 if (inRANGE(*q, '1', '9')) {
12082 width = expect_number(&q);
12085 Perl_croak_nocontext(
12086 "Cannot yet reorder sv_vcatpvfn() arguments from va_list");
12088 efix = (Size_t)width;
12090 no_redundant_warning = TRUE;
12102 if (plus == '+' && *q == ' ') /* '+' over ' ' */
12129 /* at this point we can expect one of:
12131 * 123 an explicit width
12132 * * width taken from next arg
12133 * *12$ width taken from 12th arg
12136 * But any width specification may be preceded by a v, in one of its
12141 * So an asterisk may be either a width specifier or a vector
12142 * separator arg specifier, and we don't know which initially
12147 STRLEN ix; /* explicit width/vector separator index */
12149 if (inRANGE(*q, '1', '9')) {
12150 ix = expect_number(&q);
12153 Perl_croak_nocontext(
12154 "Cannot yet reorder sv_vcatpvfn() arguments from va_list");
12155 no_redundant_warning = TRUE;
12164 /* The asterisk was for *v, *NNN$v: vectorizing, but not
12165 * with the default "." */
12170 vecsv = va_arg(*args, SV*);
12172 ix = ix ? ix - 1 : svix++;
12173 vecsv = ix < sv_count ? svargs[ix]
12174 : (arg_missing = TRUE, &PL_sv_no);
12176 dotstr = SvPV_const(vecsv, dotstrlen);
12177 /* Keep the DO_UTF8 test *after* the SvPV call, else things go
12178 bad with tied or overloaded values that return UTF8. */
12179 if (DO_UTF8(vecsv))
12181 else if (has_utf8) {
12182 vecsv = sv_mortalcopy(vecsv);
12183 sv_utf8_upgrade(vecsv);
12184 dotstr = SvPV_const(vecsv, dotstrlen);
12191 /* the asterisk specified a width */
12196 i = va_arg(*args, int);
12198 ix = ix ? ix - 1 : svix++;
12199 sv = (ix < sv_count) ? svargs[ix]
12200 : (arg_missing = TRUE, (SV*)NULL);
12202 width = S_sprintf_arg_num_val(aTHX_ args, i, sv, &left);
12205 else if (*q == 'v') {
12216 /* explicit width? */
12221 if (inRANGE(*q, '1', '9'))
12222 width = expect_number(&q);
12232 STRLEN ix; /* explicit precision index */
12234 if (inRANGE(*q, '1', '9')) {
12235 ix = expect_number(&q);
12238 Perl_croak_nocontext(
12239 "Cannot yet reorder sv_vcatpvfn() arguments from va_list");
12240 no_redundant_warning = TRUE;
12253 i = va_arg(*args, int);
12255 ix = ix ? ix - 1 : svix++;
12256 sv = (ix < sv_count) ? svargs[ix]
12257 : (arg_missing = TRUE, (SV*)NULL);
12259 precis = S_sprintf_arg_num_val(aTHX_ args, i, sv, &neg);
12261 /* ignore negative precision */
12267 /* although it doesn't seem documented, this code has long
12269 * no digits following the '.' is treated like '.0'
12270 * the number may be preceded by any number of zeroes,
12271 * e.g. "%.0001f", which is the same as "%.1f"
12272 * so I've kept that behaviour. DAPM May 2017
12276 precis = inRANGE(*q, '1', '9') ? expect_number(&q) : 0;
12285 case 'I': /* Ix, I32x, and I64x */
12286 # ifdef USE_64_BIT_INT
12287 if (q[1] == '6' && q[2] == '4') {
12293 if (q[1] == '3' && q[2] == '2') {
12297 # ifdef USE_64_BIT_INT
12303 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
12304 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
12307 # ifdef USE_QUADMATH
12320 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
12321 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
12322 if (*q == 'l') { /* lld, llf */
12331 if (*++q == 'h') { /* hhd, hhu */
12348 c = *q++; /* c now holds the conversion type */
12350 /* '%' doesn't have an arg, so skip arg processing */
12359 if (vectorize && !strchr("BbDdiOouUXx", c))
12362 /* get next arg (individual branches do their own va_arg()
12363 * handling for the args case) */
12366 efix = efix ? efix - 1 : svix++;
12367 argsv = efix < sv_count ? svargs[efix]
12368 : (arg_missing = TRUE, &PL_sv_no);
12378 eptr = va_arg(*args, char*);
12381 elen = my_strnlen(eptr, precis);
12383 elen = strlen(eptr);
12385 eptr = (char *)nullstr;
12386 elen = sizeof nullstr - 1;
12390 eptr = SvPV_const(argsv, elen);
12391 if (DO_UTF8(argsv)) {
12392 STRLEN old_precis = precis;
12393 if (has_precis && precis < elen) {
12394 STRLEN ulen = sv_or_pv_len_utf8(argsv, eptr, elen);
12395 STRLEN p = precis > ulen ? ulen : precis;
12396 precis = sv_or_pv_pos_u2b(argsv, eptr, p, 0);
12397 /* sticks at end */
12399 if (width) { /* fudge width (can't fudge elen) */
12400 if (has_precis && precis < elen)
12401 width += precis - old_precis;
12404 elen - sv_or_pv_len_utf8(argsv,eptr,elen);
12411 if (has_precis && precis < elen)
12423 * "%...p" is normally treated like "%...x", except that the
12424 * number to print is the SV's address (or a pointer address
12425 * for C-ish sprintf).
12427 * However, the C-ish sprintf variant allows a few special
12428 * extensions. These are currently:
12430 * %-p (SVf) Like %s, but gets the string from an SV*
12431 * arg rather than a char* arg.
12432 * (This was previously %_).
12434 * %-<num>p Ditto but like %.<num>s (i.e. num is max width)
12436 * %2p (HEKf) Like %s, but using the key string in a HEK
12438 * %3p (HEKf256) Ditto but like %.256s
12440 * %d%lu%4p (UTF8f) A utf8 string. Consumes 3 args:
12441 * (cBOOL(utf8), len, string_buf).
12442 * It's handled by the "case 'd'" branch
12443 * rather than here.
12445 * %<num>p where num is 1 or > 4: reserved for future
12446 * extensions. Warns, but then is treated as a
12447 * general %p (print hex address) format.
12455 /* not %*p or %*1$p - any width was explicit */
12459 if (left) { /* %-p (SVf), %-NNNp */
12464 argsv = MUTABLE_SV(va_arg(*args, void*));
12465 eptr = SvPV_const(argsv, elen);
12466 if (DO_UTF8(argsv))
12471 else if (width == 2 || width == 3) { /* HEKf, HEKf256 */
12472 HEK * const hek = va_arg(*args, HEK *);
12473 eptr = HEK_KEY(hek);
12474 elen = HEK_LEN(hek);
12485 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
12486 "internal %%<num>p might conflict with future printf extensions");
12490 /* treat as normal %...p */
12492 uv = PTR2UV(args ? va_arg(*args, void*) : argsv);
12497 /* Ignore any size specifiers, since they're not documented as
12498 * being allowed for %c (ideally we should warn on e.g. '%hc').
12499 * Setting a default intsize, along with a positive
12500 * (which signals unsigned) base, causes, for C-ish use, the
12501 * va_arg to be interpreted as as unsigned int, when it's
12502 * actually signed, which will convert -ve values to high +ve
12503 * values. Note that unlike the libc %c, values > 255 will
12504 * convert to high unicode points rather than being truncated
12505 * to 8 bits. For perlish use, it will do SvUV(argsv), which
12506 * will again convert -ve args to high -ve values.
12509 base = 1; /* special value that indicates we're doing a 'c' */
12510 goto get_int_arg_val;
12519 goto get_int_arg_val;
12522 /* probably just a plain %d, but it might be the start of the
12523 * special UTF8f format, which usually looks something like
12524 * "%d%lu%4p" (the lu may vary by platform)
12526 assert((UTF8f)[0] == 'd');
12527 assert((UTF8f)[1] == '%');
12529 if ( args /* UTF8f only valid for C-ish sprintf */
12530 && q == fmtstart + 1 /* plain %d, not %....d */
12531 && patend >= fmtstart + sizeof(UTF8f) - 1 /* long enough */
12533 && strnEQ(q + 1, UTF8f + 2, sizeof(UTF8f) - 3))
12535 /* The argument has already gone through cBOOL, so the cast
12537 is_utf8 = (bool)va_arg(*args, int);
12538 elen = va_arg(*args, UV);
12539 /* if utf8 length is larger than 0x7ffff..., then it might
12540 * have been a signed value that wrapped */
12541 if (elen > ((~(STRLEN)0) >> 1)) {
12542 assert(0); /* in DEBUGGING build we want to crash */
12543 elen = 0; /* otherwise we want to treat this as an empty string */
12545 eptr = va_arg(*args, char *);
12546 q += sizeof(UTF8f) - 2;
12553 goto get_int_arg_val;
12564 goto get_int_arg_val;
12569 goto get_int_arg_val;
12580 goto get_int_arg_val;
12595 esignbuf[esignlen++] = plus;
12598 /* initialise the vector string to iterate over */
12600 vecsv = args ? va_arg(*args, SV*) : argsv;
12602 /* if this is a version object, we need to convert
12603 * back into v-string notation and then let the
12604 * vectorize happen normally
12606 if (sv_isobject(vecsv) && sv_derived_from(vecsv, "version")) {
12607 if ( hv_existss(MUTABLE_HV(SvRV(vecsv)), "alpha") ) {
12608 Perl_ck_warner_d(aTHX_ packWARN(WARN_PRINTF),
12609 "vector argument not supported with alpha versions");
12613 vecstr = (U8*)SvPV_const(vecsv,veclen);
12614 vecsv = sv_newmortal();
12615 scan_vstring((char *)vecstr, (char *)vecstr + veclen,
12619 vecstr = (U8*)SvPV_const(vecsv, veclen);
12620 vec_utf8 = DO_UTF8(vecsv);
12622 /* This is the re-entry point for when we're iterating
12623 * over the individual characters of a vector arg */
12626 goto done_valid_conversion;
12628 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
12638 /* test arg for inf/nan. This can trigger an unwanted
12639 * 'str' overload, so manually force 'num' overload first
12643 if (UNLIKELY(SvAMAGIC(argsv)))
12644 argsv = sv_2num(argsv);
12645 if (UNLIKELY(isinfnansv(argsv)))
12646 goto handle_infnan_argsv;
12650 /* signed int type */
12655 case 'c': iv = (char)va_arg(*args, int); break;
12656 case 'h': iv = (short)va_arg(*args, int); break;
12657 case 'l': iv = va_arg(*args, long); break;
12658 case 'V': iv = va_arg(*args, IV); break;
12659 case 'z': iv = va_arg(*args, SSize_t); break;
12660 #ifdef HAS_PTRDIFF_T
12661 case 't': iv = va_arg(*args, ptrdiff_t); break;
12663 default: iv = va_arg(*args, int); break;
12664 case 'j': iv = (IV) va_arg(*args, PERL_INTMAX_T); break;
12667 iv = va_arg(*args, Quad_t); break;
12674 /* assign to tiv then cast to iv to work around
12675 * 2003 GCC cast bug (gnu.org bugzilla #13488) */
12676 IV tiv = SvIV_nomg(argsv);
12678 case 'c': iv = (char)tiv; break;
12679 case 'h': iv = (short)tiv; break;
12680 case 'l': iv = (long)tiv; break;
12682 default: iv = tiv; break;
12685 iv = (Quad_t)tiv; break;
12692 /* now convert iv to uv */
12696 esignbuf[esignlen++] = plus;
12699 /* Using 0- here to silence bogus warning from MS VC */
12700 uv = (UV) (0 - (UV) iv);
12701 esignbuf[esignlen++] = '-';
12705 /* unsigned int type */
12708 case 'c': uv = (unsigned char)va_arg(*args, unsigned);
12710 case 'h': uv = (unsigned short)va_arg(*args, unsigned);
12712 case 'l': uv = va_arg(*args, unsigned long); break;
12713 case 'V': uv = va_arg(*args, UV); break;
12714 case 'z': uv = va_arg(*args, Size_t); break;
12715 #ifdef HAS_PTRDIFF_T
12716 /* will sign extend, but there is no
12717 * uptrdiff_t, so oh well */
12718 case 't': uv = va_arg(*args, ptrdiff_t); break;
12720 case 'j': uv = (UV) va_arg(*args, PERL_UINTMAX_T); break;
12721 default: uv = va_arg(*args, unsigned); break;
12724 uv = va_arg(*args, Uquad_t); break;
12731 /* assign to tiv then cast to iv to work around
12732 * 2003 GCC cast bug (gnu.org bugzilla #13488) */
12733 UV tuv = SvUV_nomg(argsv);
12735 case 'c': uv = (unsigned char)tuv; break;
12736 case 'h': uv = (unsigned short)tuv; break;
12737 case 'l': uv = (unsigned long)tuv; break;
12739 default: uv = tuv; break;
12742 uv = (Uquad_t)tuv; break;
12753 char *ptr = ebuf + sizeof ebuf;
12760 const char * const p =
12761 (c == 'X') ? PL_hexdigit + 16 : PL_hexdigit;
12766 } while (uv >>= 4);
12767 if (alt && *ptr != '0') {
12768 esignbuf[esignlen++] = '0';
12769 esignbuf[esignlen++] = c; /* 'x' or 'X' */
12776 *--ptr = '0' + dig;
12777 } while (uv >>= 3);
12778 if (alt && *ptr != '0')
12784 *--ptr = '0' + dig;
12785 } while (uv >>= 1);
12786 if (alt && *ptr != '0') {
12787 esignbuf[esignlen++] = '0';
12788 esignbuf[esignlen++] = c; /* 'b' or 'B' */
12793 /* special-case: base 1 indicates a 'c' format:
12794 * we use the common code for extracting a uv,
12795 * but handle that value differently here than
12796 * all the other int types */
12798 (!UVCHR_IS_INVARIANT(uv) && SvUTF8(sv)))
12801 assert(sizeof(ebuf) >= UTF8_MAXBYTES + 1);
12803 elen = uvchr_to_utf8((U8*)eptr, uv) - (U8*)ebuf;
12808 ebuf[0] = (char)uv;
12813 default: /* it had better be ten or less */
12816 *--ptr = '0' + dig;
12817 } while (uv /= base);
12820 elen = (ebuf + sizeof ebuf) - ptr;
12824 zeros = precis - elen;
12825 else if (precis == 0 && elen == 1 && *eptr == '0'
12826 && !(base == 8 && alt)) /* "%#.0o" prints "0" */
12829 /* a precision nullifies the 0 flag. */
12835 /* FLOATING POINT */
12838 c = 'f'; /* maybe %F isn't supported here */
12840 case 'e': case 'E':
12842 case 'g': case 'G':
12843 case 'a': case 'A':
12846 STRLEN float_need; /* what PL_efloatsize needs to become */
12847 bool hexfp; /* hexadecimal floating point? */
12849 vcatpvfn_long_double_t fv;
12852 /* This is evil, but floating point is even more evil */
12854 /* for SV-style calling, we can only get NV
12855 for C-style calling, we assume %f is double;
12856 for simplicity we allow any of %Lf, %llf, %qf for long double
12860 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12864 /* [perl #20339] - we should accept and ignore %lf rather than die */
12868 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12869 intsize = args ? 0 : 'q';
12873 #if defined(HAS_LONG_DOUBLE)
12886 /* Now we need (long double) if intsize == 'q', else (double). */
12888 /* Note: do not pull NVs off the va_list with va_arg()
12889 * (pull doubles instead) because if you have a build
12890 * with long doubles, you would always be pulling long
12891 * doubles, which would badly break anyone using only
12892 * doubles (i.e. the majority of builds). In other
12893 * words, you cannot mix doubles and long doubles.
12894 * The only case where you can pull off long doubles
12895 * is when the format specifier explicitly asks so with
12897 #ifdef USE_QUADMATH
12898 fv = intsize == 'q' ?
12899 va_arg(*args, NV) : va_arg(*args, double);
12901 #elif LONG_DOUBLESIZE > DOUBLESIZE
12902 if (intsize == 'q') {
12903 fv = va_arg(*args, long double);
12906 nv = va_arg(*args, double);
12907 VCATPVFN_NV_TO_FV(nv, fv);
12910 nv = va_arg(*args, double);
12917 /* we jump here if an int-ish format encountered an
12918 * infinite/Nan argsv. After setting nv/fv, it falls
12919 * into the isinfnan block which follows */
12920 handle_infnan_argsv:
12921 nv = SvNV_nomg(argsv);
12922 VCATPVFN_NV_TO_FV(nv, fv);
12925 if (Perl_isinfnan(nv)) {
12927 Perl_croak(aTHX_ "Cannot printf %" NVgf " with '%c'",
12928 SvNV_nomg(argsv), (int)c);
12930 elen = S_infnan_2pv(nv, ebuf, sizeof(ebuf), plus);
12939 /* special-case "%.0f" */
12943 && !(width || left || plus || alt)
12946 && ((eptr = F0convert(nv, ebuf + sizeof ebuf, &elen)))
12950 /* Determine the buffer size needed for the various
12951 * floating-point formats.
12953 * The basic possibilities are:
12956 * %f 1111111.123456789
12957 * %e 1.111111123e+06
12958 * %a 0x1.0f4471f9bp+20
12960 * %g 1.11111112e+15
12962 * where P is the value of the precision in the format, or 6
12963 * if not specified. Note the two possible output formats of
12964 * %g; in both cases the number of significant digits is <=
12967 * For most of the format types the maximum buffer size needed
12968 * is precision, plus: any leading 1 or 0x1, the radix
12969 * point, and an exponent. The difficult one is %f: for a
12970 * large positive exponent it can have many leading digits,
12971 * which needs to be calculated specially. Also %a is slightly
12972 * different in that in the absence of a specified precision,
12973 * it uses as many digits as necessary to distinguish
12974 * different values.
12976 * First, here are the constant bits. For ease of calculation
12977 * we over-estimate the needed buffer size, for example by
12978 * assuming all formats have an exponent and a leading 0x1.
12980 * Also for production use, add a little extra overhead for
12981 * safety's sake. Under debugging don't, as it means we're
12982 * more likely to quickly spot issues during development.
12985 float_need = 1 /* possible unary minus */
12986 + 4 /* "0x1" plus very unlikely carry */
12987 + 1 /* default radix point '.' */
12988 + 2 /* "e-", "p+" etc */
12989 + 6 /* exponent: up to 16383 (quad fp) */
12991 + 20 /* safety net */
12996 /* determine the radix point len, e.g. length(".") in "1.2" */
12997 #ifdef USE_LOCALE_NUMERIC
12998 /* note that we may either explicitly use PL_numeric_radix_sv
12999 * below, or implicitly, via an snprintf() variant.
13000 * Note also things like ps_AF.utf8 which has
13001 * "\N{ARABIC DECIMAL SEPARATOR} as a radix point */
13002 if (! have_in_lc_numeric) {
13003 in_lc_numeric = IN_LC(LC_NUMERIC);
13004 have_in_lc_numeric = TRUE;
13007 if (in_lc_numeric) {
13008 WITH_LC_NUMERIC_SET_TO_NEEDED_IN(TRUE, {
13009 /* this can't wrap unless PL_numeric_radix_sv is a string
13010 * consuming virtually all the 32-bit or 64-bit address
13013 float_need += (SvCUR(PL_numeric_radix_sv) - 1);
13015 /* floating-point formats only get utf8 if the radix point
13016 * is utf8. All other characters in the string are < 128
13017 * and so can be safely appended to both a non-utf8 and utf8
13019 * Note that this will convert the output to utf8 even if
13020 * the radix point didn't get output.
13022 if (SvUTF8(PL_numeric_radix_sv) && !has_utf8) {
13023 sv_utf8_upgrade(sv);
13032 if (isALPHA_FOLD_EQ(c, 'f')) {
13033 /* Determine how many digits before the radix point
13034 * might be emitted. frexp() (or frexpl) has some
13035 * unspecified behaviour for nan/inf/-inf, so lucky we've
13036 * already handled them above */
13038 int i = PERL_INT_MIN;
13039 (void)Perl_frexp((NV)fv, &i);
13040 if (i == PERL_INT_MIN)
13041 Perl_die(aTHX_ "panic: frexp: %" VCATPVFN_FV_GF, fv);
13044 digits = BIT_DIGITS(i);
13045 /* this can't overflow. 'digits' will only be a few
13046 * thousand even for the largest floating-point types.
13047 * And up until now float_need is just some small
13048 * constants plus radix len, which can't be in
13049 * overflow territory unless the radix SV is consuming
13050 * over 1/2 the address space */
13051 assert(float_need < ((STRLEN)~0) - digits);
13052 float_need += digits;
13055 else if (UNLIKELY(isALPHA_FOLD_EQ(c, 'a'))) {
13058 /* %a in the absence of precision may print as many
13059 * digits as needed to represent the entire mantissa
13061 * This estimate seriously overshoots in most cases,
13062 * but better the undershooting. Firstly, all bytes
13063 * of the NV are not mantissa, some of them are
13064 * exponent. Secondly, for the reasonably common
13065 * long doubles case, the "80-bit extended", two
13066 * or six bytes of the NV are unused. Also, we'll
13067 * still pick up an extra +6 from the default
13068 * precision calculation below. */
13070 #ifdef LONGDOUBLE_DOUBLEDOUBLE
13071 /* For the "double double", we need more.
13072 * Since each double has their own exponent, the
13073 * doubles may float (haha) rather far from each
13074 * other, and the number of required bits is much
13075 * larger, up to total of DOUBLEDOUBLE_MAXBITS bits.
13076 * See the definition of DOUBLEDOUBLE_MAXBITS.
13078 * Need 2 hexdigits for each byte. */
13079 (DOUBLEDOUBLE_MAXBITS/8 + 1) * 2;
13081 NVSIZE * 2; /* 2 hexdigits for each byte */
13083 /* see "this can't overflow" comment above */
13084 assert(float_need < ((STRLEN)~0) - digits);
13085 float_need += digits;
13088 /* special-case "%.<number>g" if it will fit in ebuf */
13090 && precis /* See earlier comment about buggy Gconvert
13091 when digits, aka precis, is 0 */
13093 /* check, in manner not involving wrapping, that it will
13095 && float_need < sizeof(ebuf)
13096 && sizeof(ebuf) - float_need > precis
13097 && !(width || left || plus || alt)
13101 WITH_LC_NUMERIC_SET_TO_NEEDED_IN(in_lc_numeric,
13102 SNPRINTF_G(fv, ebuf, sizeof(ebuf), precis)
13104 elen = strlen(ebuf);
13111 STRLEN pr = has_precis ? precis : 6; /* known default */
13112 /* this probably can't wrap, since precis is limited
13113 * to 1/4 address space size, but better safe than sorry
13115 if (float_need >= ((STRLEN)~0) - pr)
13116 croak_memory_wrap();
13120 if (float_need < width)
13121 float_need = width;
13123 if (float_need > INT_MAX) {
13124 /* snprintf() returns an int, and we use that return value,
13125 so die horribly if the expected size is too large for int
13127 Perl_croak(aTHX_ "Numeric format result too large");
13130 if (PL_efloatsize <= float_need) {
13131 /* PL_efloatbuf should be at least 1 greater than
13132 * float_need to allow a trailing \0 to be returned by
13133 * snprintf(). If we need to grow, overgrow for the
13134 * benefit of future generations */
13135 const STRLEN extra = 0x20;
13136 if (float_need >= ((STRLEN)~0) - extra)
13137 croak_memory_wrap();
13138 float_need += extra;
13139 Safefree(PL_efloatbuf);
13140 PL_efloatsize = float_need;
13141 Newx(PL_efloatbuf, PL_efloatsize, char);
13142 PL_efloatbuf[0] = '\0';
13145 if (UNLIKELY(hexfp)) {
13146 elen = S_format_hexfp(aTHX_ PL_efloatbuf, PL_efloatsize, c,
13147 nv, fv, has_precis, precis, width,
13148 alt, plus, left, fill, in_lc_numeric);
13151 char *ptr = ebuf + sizeof ebuf;
13154 #if defined(USE_QUADMATH)
13155 if (intsize == 'q') {
13159 /* FIXME: what to do if HAS_LONG_DOUBLE but not PERL_PRIfldbl? */
13160 #elif defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
13161 /* Note that this is HAS_LONG_DOUBLE and PERL_PRIfldbl,
13162 * not USE_LONG_DOUBLE and NVff. In other words,
13163 * this needs to work without USE_LONG_DOUBLE. */
13164 if (intsize == 'q') {
13165 /* Copy the one or more characters in a long double
13166 * format before the 'base' ([efgEFG]) character to
13167 * the format string. */
13168 static char const ldblf[] = PERL_PRIfldbl;
13169 char const *p = ldblf + sizeof(ldblf) - 3;
13170 while (p >= ldblf) { *--ptr = *p--; }
13175 do { *--ptr = '0' + (base % 10); } while (base /= 10);
13180 do { *--ptr = '0' + (base % 10); } while (base /= 10);
13192 /* No taint. Otherwise we are in the strange situation
13193 * where printf() taints but print($float) doesn't.
13196 /* hopefully the above makes ptr a very constrained format
13197 * that is safe to use, even though it's not literal */
13198 GCC_DIAG_IGNORE_STMT(-Wformat-nonliteral);
13199 #ifdef USE_QUADMATH
13201 const char* qfmt = quadmath_format_single(ptr);
13203 Perl_croak_nocontext("panic: quadmath invalid format \"%s\"", ptr);
13204 WITH_LC_NUMERIC_SET_TO_NEEDED_IN(in_lc_numeric,
13205 elen = quadmath_snprintf(PL_efloatbuf, PL_efloatsize,
13208 if ((IV)elen == -1) {
13211 Perl_croak_nocontext("panic: quadmath_snprintf failed, format \"%s\"", qfmt);
13216 #elif defined(HAS_LONG_DOUBLE)
13217 WITH_LC_NUMERIC_SET_TO_NEEDED_IN(in_lc_numeric,
13218 elen = ((intsize == 'q')
13219 ? my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv)
13220 : my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, (double)fv))
13223 WITH_LC_NUMERIC_SET_TO_NEEDED_IN(in_lc_numeric,
13224 elen = my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv)
13227 GCC_DIAG_RESTORE_STMT;
13230 eptr = PL_efloatbuf;
13234 /* Since floating-point formats do their own formatting and
13235 * padding, we skip the main block of code at the end of this
13236 * loop which handles appending eptr to sv, and do our own
13237 * stripped-down version */
13242 assert(elen >= width);
13244 S_sv_catpvn_simple(aTHX_ sv, eptr, elen);
13246 goto done_valid_conversion;
13254 /* XXX ideally we should warn if any flags etc have been
13255 * set, e.g. "%-4.5n" */
13256 /* XXX if sv was originally non-utf8 with a char in the
13257 * range 0x80-0xff, then if it got upgraded, we should
13258 * calculate char len rather than byte len here */
13259 len = SvCUR(sv) - origlen;
13261 int i = (len > PERL_INT_MAX) ? PERL_INT_MAX : (int)len;
13264 case 'c': *(va_arg(*args, char*)) = i; break;
13265 case 'h': *(va_arg(*args, short*)) = i; break;
13266 default: *(va_arg(*args, int*)) = i; break;
13267 case 'l': *(va_arg(*args, long*)) = i; break;
13268 case 'V': *(va_arg(*args, IV*)) = i; break;
13269 case 'z': *(va_arg(*args, SSize_t*)) = i; break;
13270 #ifdef HAS_PTRDIFF_T
13271 case 't': *(va_arg(*args, ptrdiff_t*)) = i; break;
13273 case 'j': *(va_arg(*args, PERL_INTMAX_T*)) = i; break;
13276 *(va_arg(*args, Quad_t*)) = i; break;
13284 Perl_croak_nocontext(
13285 "Missing argument for %%n in %s",
13286 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
13287 sv_setuv_mg(argsv, has_utf8 ? (UV)sv_len_utf8(sv) : (UV)len);
13289 goto done_valid_conversion;
13297 && (PL_op->op_type == OP_PRTF || PL_op->op_type == OP_SPRINTF)
13298 && ckWARN(WARN_PRINTF))
13300 SV * const msg = sv_newmortal();
13301 Perl_sv_setpvf(aTHX_ msg, "Invalid conversion in %sprintf: ",
13302 (PL_op->op_type == OP_PRTF) ? "" : "s");
13303 if (fmtstart < patend) {
13304 const char * const fmtend = q < patend ? q : patend;
13306 sv_catpvs(msg, "\"%");
13307 for (f = fmtstart; f < fmtend; f++) {
13309 sv_catpvn_nomg(msg, f, 1);
13311 Perl_sv_catpvf(aTHX_ msg,
13312 "\\%03" UVof, (UV)*f & 0xFF);
13315 sv_catpvs(msg, "\"");
13317 sv_catpvs(msg, "end of string");
13319 Perl_warner(aTHX_ packWARN(WARN_PRINTF), "%" SVf, SVfARG(msg)); /* yes, this is reentrant */
13322 /* mangled format: output the '%', then continue from the
13323 * character following that */
13324 sv_catpvn_nomg(sv, fmtstart-1, 1);
13327 /* Any "redundant arg" warning from now onwards will probably
13328 * just be misleading, so don't bother. */
13329 no_redundant_warning = TRUE;
13330 continue; /* not "break" */
13333 if (is_utf8 != has_utf8) {
13336 sv_utf8_upgrade(sv);
13339 const STRLEN old_elen = elen;
13340 SV * const nsv = newSVpvn_flags(eptr, elen, SVs_TEMP);
13341 sv_utf8_upgrade(nsv);
13342 eptr = SvPVX_const(nsv);
13345 if (width) { /* fudge width (can't fudge elen) */
13346 width += elen - old_elen;
13353 /* append esignbuf, filler, zeros, eptr and dotstr to sv */
13356 STRLEN need, have, gap;
13360 /* signed value that's wrapped? */
13361 assert(elen <= ((~(STRLEN)0) >> 1));
13363 /* if zeros is non-zero, then it represents filler between
13364 * elen and precis. So adding elen and zeros together will
13365 * always be <= precis, and the addition can never wrap */
13366 assert(!zeros || (precis > elen && precis - elen == zeros));
13367 have = elen + zeros;
13369 if (have >= (((STRLEN)~0) - esignlen))
13370 croak_memory_wrap();
13373 need = (have > width ? have : width);
13376 if (need >= (((STRLEN)~0) - (SvCUR(sv) + 1)))
13377 croak_memory_wrap();
13378 need += (SvCUR(sv) + 1);
13385 for (i = 0; i < esignlen; i++)
13386 *s++ = esignbuf[i];
13387 for (i = zeros; i; i--)
13389 Copy(eptr, s, elen, char);
13391 for (i = gap; i; i--)
13396 for (i = 0; i < esignlen; i++)
13397 *s++ = esignbuf[i];
13402 for (i = gap; i; i--)
13404 for (i = 0; i < esignlen; i++)
13405 *s++ = esignbuf[i];
13408 for (i = zeros; i; i--)
13410 Copy(eptr, s, elen, char);
13415 SvCUR_set(sv, s - SvPVX_const(sv));
13423 if (vectorize && veclen) {
13424 /* we append the vector separator separately since %v isn't
13425 * very common: don't slow down the general case by adding
13426 * dotstrlen to need etc */
13427 sv_catpvn_nomg(sv, dotstr, dotstrlen);
13429 goto vector; /* do next iteration */
13432 done_valid_conversion:
13435 S_warn_vcatpvfn_missing_argument(aTHX);
13438 /* Now that we've consumed all our printf format arguments (svix)
13439 * do we have things left on the stack that we didn't use?
13441 if (!no_redundant_warning && sv_count >= svix + 1 && ckWARN(WARN_REDUNDANT)) {
13442 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
13443 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
13449 /* =========================================================================
13451 =head1 Cloning an interpreter
13455 All the macros and functions in this section are for the private use of
13456 the main function, perl_clone().
13458 The foo_dup() functions make an exact copy of an existing foo thingy.
13459 During the course of a cloning, a hash table is used to map old addresses
13460 to new addresses. The table is created and manipulated with the
13461 ptr_table_* functions.
13463 * =========================================================================*/
13466 #if defined(USE_ITHREADS)
13468 /* XXX Remove this so it doesn't have to go thru the macro and return for nothing */
13469 #ifndef GpREFCNT_inc
13470 # define GpREFCNT_inc(gp) ((gp) ? (++(gp)->gp_refcnt, (gp)) : (GP*)NULL)
13474 /* Certain cases in Perl_ss_dup have been merged, by relying on the fact
13475 that currently av_dup, gv_dup and hv_dup are the same as sv_dup.
13476 If this changes, please unmerge ss_dup.
13477 Likewise, sv_dup_inc_multiple() relies on this fact. */
13478 #define sv_dup_inc_NN(s,t) SvREFCNT_inc_NN(sv_dup_inc(s,t))
13479 #define av_dup(s,t) MUTABLE_AV(sv_dup((const SV *)s,t))
13480 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
13481 #define hv_dup(s,t) MUTABLE_HV(sv_dup((const SV *)s,t))
13482 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
13483 #define cv_dup(s,t) MUTABLE_CV(sv_dup((const SV *)s,t))
13484 #define cv_dup_inc(s,t) MUTABLE_CV(sv_dup_inc((const SV *)s,t))
13485 #define io_dup(s,t) MUTABLE_IO(sv_dup((const SV *)s,t))
13486 #define io_dup_inc(s,t) MUTABLE_IO(sv_dup_inc((const SV *)s,t))
13487 #define gv_dup(s,t) MUTABLE_GV(sv_dup((const SV *)s,t))
13488 #define gv_dup_inc(s,t) MUTABLE_GV(sv_dup_inc((const SV *)s,t))
13489 #define SAVEPV(p) ((p) ? savepv(p) : NULL)
13490 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
13492 /* clone a parser */
13495 Perl_parser_dup(pTHX_ const yy_parser *const proto, CLONE_PARAMS *const param)
13499 PERL_ARGS_ASSERT_PARSER_DUP;
13504 /* look for it in the table first */
13505 parser = (yy_parser *)ptr_table_fetch(PL_ptr_table, proto);
13509 /* create anew and remember what it is */
13510 Newxz(parser, 1, yy_parser);
13511 ptr_table_store(PL_ptr_table, proto, parser);
13513 /* XXX eventually, just Copy() most of the parser struct ? */
13515 parser->lex_brackets = proto->lex_brackets;
13516 parser->lex_casemods = proto->lex_casemods;
13517 parser->lex_brackstack = savepvn(proto->lex_brackstack,
13518 (proto->lex_brackets < 120 ? 120 : proto->lex_brackets));
13519 parser->lex_casestack = savepvn(proto->lex_casestack,
13520 (proto->lex_casemods < 12 ? 12 : proto->lex_casemods));
13521 parser->lex_defer = proto->lex_defer;
13522 parser->lex_dojoin = proto->lex_dojoin;
13523 parser->lex_formbrack = proto->lex_formbrack;
13524 parser->lex_inpat = proto->lex_inpat;
13525 parser->lex_inwhat = proto->lex_inwhat;
13526 parser->lex_op = proto->lex_op;
13527 parser->lex_repl = sv_dup_inc(proto->lex_repl, param);
13528 parser->lex_starts = proto->lex_starts;
13529 parser->lex_stuff = sv_dup_inc(proto->lex_stuff, param);
13530 parser->multi_close = proto->multi_close;
13531 parser->multi_open = proto->multi_open;
13532 parser->multi_start = proto->multi_start;
13533 parser->multi_end = proto->multi_end;
13534 parser->preambled = proto->preambled;
13535 parser->lex_super_state = proto->lex_super_state;
13536 parser->lex_sub_inwhat = proto->lex_sub_inwhat;
13537 parser->lex_sub_op = proto->lex_sub_op;
13538 parser->lex_sub_repl= sv_dup_inc(proto->lex_sub_repl, param);
13539 parser->linestr = sv_dup_inc(proto->linestr, param);
13540 parser->expect = proto->expect;
13541 parser->copline = proto->copline;
13542 parser->last_lop_op = proto->last_lop_op;
13543 parser->lex_state = proto->lex_state;
13544 parser->rsfp = fp_dup(proto->rsfp, '<', param);
13545 /* rsfp_filters entries have fake IoDIRP() */
13546 parser->rsfp_filters= av_dup_inc(proto->rsfp_filters, param);
13547 parser->in_my = proto->in_my;
13548 parser->in_my_stash = hv_dup(proto->in_my_stash, param);
13549 parser->error_count = proto->error_count;
13550 parser->sig_elems = proto->sig_elems;
13551 parser->sig_optelems= proto->sig_optelems;
13552 parser->sig_slurpy = proto->sig_slurpy;
13553 parser->recheck_utf8_validity = proto->recheck_utf8_validity;
13556 char * const ols = SvPVX(proto->linestr);
13557 char * const ls = SvPVX(parser->linestr);
13559 parser->bufptr = ls + (proto->bufptr >= ols ?
13560 proto->bufptr - ols : 0);
13561 parser->oldbufptr = ls + (proto->oldbufptr >= ols ?
13562 proto->oldbufptr - ols : 0);
13563 parser->oldoldbufptr= ls + (proto->oldoldbufptr >= ols ?
13564 proto->oldoldbufptr - ols : 0);
13565 parser->linestart = ls + (proto->linestart >= ols ?
13566 proto->linestart - ols : 0);
13567 parser->last_uni = ls + (proto->last_uni >= ols ?
13568 proto->last_uni - ols : 0);
13569 parser->last_lop = ls + (proto->last_lop >= ols ?
13570 proto->last_lop - ols : 0);
13572 parser->bufend = ls + SvCUR(parser->linestr);
13575 Copy(proto->tokenbuf, parser->tokenbuf, 256, char);
13578 Copy(proto->nextval, parser->nextval, 5, YYSTYPE);
13579 Copy(proto->nexttype, parser->nexttype, 5, I32);
13580 parser->nexttoke = proto->nexttoke;
13582 /* XXX should clone saved_curcop here, but we aren't passed
13583 * proto_perl; so do it in perl_clone_using instead */
13589 /* duplicate a file handle */
13592 Perl_fp_dup(pTHX_ PerlIO *const fp, const char type, CLONE_PARAMS *const param)
13596 PERL_ARGS_ASSERT_FP_DUP;
13597 PERL_UNUSED_ARG(type);
13600 return (PerlIO*)NULL;
13602 /* look for it in the table first */
13603 ret = (PerlIO*)ptr_table_fetch(PL_ptr_table, fp);
13607 /* create anew and remember what it is */
13608 #ifdef __amigaos4__
13609 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE|PERLIO_DUP_FD);
13611 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE);
13613 ptr_table_store(PL_ptr_table, fp, ret);
13617 /* duplicate a directory handle */
13620 Perl_dirp_dup(pTHX_ DIR *const dp, CLONE_PARAMS *const param)
13624 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
13626 const Direntry_t *dirent;
13627 char smallbuf[256]; /* XXX MAXPATHLEN, surely? */
13633 PERL_UNUSED_CONTEXT;
13634 PERL_ARGS_ASSERT_DIRP_DUP;
13639 /* look for it in the table first */
13640 ret = (DIR*)ptr_table_fetch(PL_ptr_table, dp);
13644 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
13646 PERL_UNUSED_ARG(param);
13650 /* open the current directory (so we can switch back) */
13651 if (!(pwd = PerlDir_open("."))) return (DIR *)NULL;
13653 /* chdir to our dir handle and open the present working directory */
13654 if (fchdir(my_dirfd(dp)) < 0 || !(ret = PerlDir_open("."))) {
13655 PerlDir_close(pwd);
13656 return (DIR *)NULL;
13658 /* Now we should have two dir handles pointing to the same dir. */
13660 /* Be nice to the calling code and chdir back to where we were. */
13661 /* XXX If this fails, then what? */
13662 PERL_UNUSED_RESULT(fchdir(my_dirfd(pwd)));
13664 /* We have no need of the pwd handle any more. */
13665 PerlDir_close(pwd);
13668 # define d_namlen(d) (d)->d_namlen
13670 # define d_namlen(d) strlen((d)->d_name)
13672 /* Iterate once through dp, to get the file name at the current posi-
13673 tion. Then step back. */
13674 pos = PerlDir_tell(dp);
13675 if ((dirent = PerlDir_read(dp))) {
13676 len = d_namlen(dirent);
13677 if (len > sizeof(dirent->d_name) && sizeof(dirent->d_name) > PTRSIZE) {
13678 /* If the len is somehow magically longer than the
13679 * maximum length of the directory entry, even though
13680 * we could fit it in a buffer, we could not copy it
13681 * from the dirent. Bail out. */
13682 PerlDir_close(ret);
13685 if (len <= sizeof smallbuf) name = smallbuf;
13686 else Newx(name, len, char);
13687 Move(dirent->d_name, name, len, char);
13689 PerlDir_seek(dp, pos);
13691 /* Iterate through the new dir handle, till we find a file with the
13693 if (!dirent) /* just before the end */
13695 pos = PerlDir_tell(ret);
13696 if (PerlDir_read(ret)) continue; /* not there yet */
13697 PerlDir_seek(ret, pos); /* step back */
13701 const long pos0 = PerlDir_tell(ret);
13703 pos = PerlDir_tell(ret);
13704 if ((dirent = PerlDir_read(ret))) {
13705 if (len == (STRLEN)d_namlen(dirent)
13706 && memEQ(name, dirent->d_name, len)) {
13708 PerlDir_seek(ret, pos); /* step back */
13711 /* else we are not there yet; keep iterating */
13713 else { /* This is not meant to happen. The best we can do is
13714 reset the iterator to the beginning. */
13715 PerlDir_seek(ret, pos0);
13722 if (name && name != smallbuf)
13727 ret = win32_dirp_dup(dp, param);
13730 /* pop it in the pointer table */
13732 ptr_table_store(PL_ptr_table, dp, ret);
13737 /* duplicate a typeglob */
13740 Perl_gp_dup(pTHX_ GP *const gp, CLONE_PARAMS *const param)
13744 PERL_ARGS_ASSERT_GP_DUP;
13748 /* look for it in the table first */
13749 ret = (GP*)ptr_table_fetch(PL_ptr_table, gp);
13753 /* create anew and remember what it is */
13755 ptr_table_store(PL_ptr_table, gp, ret);
13758 /* ret->gp_refcnt must be 0 before any other dups are called. We're relying
13759 on Newxz() to do this for us. */
13760 ret->gp_sv = sv_dup_inc(gp->gp_sv, param);
13761 ret->gp_io = io_dup_inc(gp->gp_io, param);
13762 ret->gp_form = cv_dup_inc(gp->gp_form, param);
13763 ret->gp_av = av_dup_inc(gp->gp_av, param);
13764 ret->gp_hv = hv_dup_inc(gp->gp_hv, param);
13765 ret->gp_egv = gv_dup(gp->gp_egv, param);/* GvEGV is not refcounted */
13766 ret->gp_cv = cv_dup_inc(gp->gp_cv, param);
13767 ret->gp_cvgen = gp->gp_cvgen;
13768 ret->gp_line = gp->gp_line;
13769 ret->gp_file_hek = hek_dup(gp->gp_file_hek, param);
13773 /* duplicate a chain of magic */
13776 Perl_mg_dup(pTHX_ MAGIC *mg, CLONE_PARAMS *const param)
13778 MAGIC *mgret = NULL;
13779 MAGIC **mgprev_p = &mgret;
13781 PERL_ARGS_ASSERT_MG_DUP;
13783 for (; mg; mg = mg->mg_moremagic) {
13786 if ((param->flags & CLONEf_JOIN_IN)
13787 && mg->mg_type == PERL_MAGIC_backref)
13788 /* when joining, we let the individual SVs add themselves to
13789 * backref as needed. */
13792 Newx(nmg, 1, MAGIC);
13794 mgprev_p = &(nmg->mg_moremagic);
13796 /* There was a comment "XXX copy dynamic vtable?" but as we don't have
13797 dynamic vtables, I'm not sure why Sarathy wrote it. The comment dates
13798 from the original commit adding Perl_mg_dup() - revision 4538.
13799 Similarly there is the annotation "XXX random ptr?" next to the
13800 assignment to nmg->mg_ptr. */
13803 /* FIXME for plugins
13804 if (nmg->mg_type == PERL_MAGIC_qr) {
13805 nmg->mg_obj = MUTABLE_SV(CALLREGDUPE((REGEXP*)nmg->mg_obj, param));
13809 nmg->mg_obj = (nmg->mg_flags & MGf_REFCOUNTED)
13810 ? nmg->mg_type == PERL_MAGIC_backref
13811 /* The backref AV has its reference
13812 * count deliberately bumped by 1 */
13813 ? SvREFCNT_inc(av_dup_inc((const AV *)
13814 nmg->mg_obj, param))
13815 : sv_dup_inc(nmg->mg_obj, param)
13816 : (nmg->mg_type == PERL_MAGIC_regdatum ||
13817 nmg->mg_type == PERL_MAGIC_regdata)
13819 : sv_dup(nmg->mg_obj, param);
13821 if (nmg->mg_ptr && nmg->mg_type != PERL_MAGIC_regex_global) {
13822 if (nmg->mg_len > 0) {
13823 nmg->mg_ptr = SAVEPVN(nmg->mg_ptr, nmg->mg_len);
13824 if (nmg->mg_type == PERL_MAGIC_overload_table &&
13825 AMT_AMAGIC((AMT*)nmg->mg_ptr))
13827 AMT * const namtp = (AMT*)nmg->mg_ptr;
13828 sv_dup_inc_multiple((SV**)(namtp->table),
13829 (SV**)(namtp->table), NofAMmeth, param);
13832 else if (nmg->mg_len == HEf_SVKEY)
13833 nmg->mg_ptr = (char*)sv_dup_inc((const SV *)nmg->mg_ptr, param);
13835 if ((nmg->mg_flags & MGf_DUP) && nmg->mg_virtual && nmg->mg_virtual->svt_dup) {
13836 nmg->mg_virtual->svt_dup(aTHX_ nmg, param);
13842 #endif /* USE_ITHREADS */
13844 struct ptr_tbl_arena {
13845 struct ptr_tbl_arena *next;
13846 struct ptr_tbl_ent array[1023/3]; /* as ptr_tbl_ent has 3 pointers. */
13849 /* create a new pointer-mapping table */
13852 Perl_ptr_table_new(pTHX)
13855 PERL_UNUSED_CONTEXT;
13857 Newx(tbl, 1, PTR_TBL_t);
13858 tbl->tbl_max = 511;
13859 tbl->tbl_items = 0;
13860 tbl->tbl_arena = NULL;
13861 tbl->tbl_arena_next = NULL;
13862 tbl->tbl_arena_end = NULL;
13863 Newxz(tbl->tbl_ary, tbl->tbl_max + 1, PTR_TBL_ENT_t*);
13867 #define PTR_TABLE_HASH(ptr) \
13868 ((PTR2UV(ptr) >> 3) ^ (PTR2UV(ptr) >> (3 + 7)) ^ (PTR2UV(ptr) >> (3 + 17)))
13870 /* map an existing pointer using a table */
13872 STATIC PTR_TBL_ENT_t *
13873 S_ptr_table_find(PTR_TBL_t *const tbl, const void *const sv)
13875 PTR_TBL_ENT_t *tblent;
13876 const UV hash = PTR_TABLE_HASH(sv);
13878 PERL_ARGS_ASSERT_PTR_TABLE_FIND;
13880 tblent = tbl->tbl_ary[hash & tbl->tbl_max];
13881 for (; tblent; tblent = tblent->next) {
13882 if (tblent->oldval == sv)
13889 Perl_ptr_table_fetch(pTHX_ PTR_TBL_t *const tbl, const void *const sv)
13891 PTR_TBL_ENT_t const *const tblent = ptr_table_find(tbl, sv);
13893 PERL_ARGS_ASSERT_PTR_TABLE_FETCH;
13894 PERL_UNUSED_CONTEXT;
13896 return tblent ? tblent->newval : NULL;
13899 /* add a new entry to a pointer-mapping table 'tbl'. In hash terms, 'oldsv' is
13900 * the key; 'newsv' is the value. The names "old" and "new" are specific to
13901 * the core's typical use of ptr_tables in thread cloning. */
13904 Perl_ptr_table_store(pTHX_ PTR_TBL_t *const tbl, const void *const oldsv, void *const newsv)
13906 PTR_TBL_ENT_t *tblent = ptr_table_find(tbl, oldsv);
13908 PERL_ARGS_ASSERT_PTR_TABLE_STORE;
13909 PERL_UNUSED_CONTEXT;
13912 tblent->newval = newsv;
13914 const UV entry = PTR_TABLE_HASH(oldsv) & tbl->tbl_max;
13916 if (tbl->tbl_arena_next == tbl->tbl_arena_end) {
13917 struct ptr_tbl_arena *new_arena;
13919 Newx(new_arena, 1, struct ptr_tbl_arena);
13920 new_arena->next = tbl->tbl_arena;
13921 tbl->tbl_arena = new_arena;
13922 tbl->tbl_arena_next = new_arena->array;
13923 tbl->tbl_arena_end = C_ARRAY_END(new_arena->array);
13926 tblent = tbl->tbl_arena_next++;
13928 tblent->oldval = oldsv;
13929 tblent->newval = newsv;
13930 tblent->next = tbl->tbl_ary[entry];
13931 tbl->tbl_ary[entry] = tblent;
13933 if (tblent->next && tbl->tbl_items > tbl->tbl_max)
13934 ptr_table_split(tbl);
13938 /* double the hash bucket size of an existing ptr table */
13941 Perl_ptr_table_split(pTHX_ PTR_TBL_t *const tbl)
13943 PTR_TBL_ENT_t **ary = tbl->tbl_ary;
13944 const UV oldsize = tbl->tbl_max + 1;
13945 UV newsize = oldsize * 2;
13948 PERL_ARGS_ASSERT_PTR_TABLE_SPLIT;
13949 PERL_UNUSED_CONTEXT;
13951 Renew(ary, newsize, PTR_TBL_ENT_t*);
13952 Zero(&ary[oldsize], newsize-oldsize, PTR_TBL_ENT_t*);
13953 tbl->tbl_max = --newsize;
13954 tbl->tbl_ary = ary;
13955 for (i=0; i < oldsize; i++, ary++) {
13956 PTR_TBL_ENT_t **entp = ary;
13957 PTR_TBL_ENT_t *ent = *ary;
13958 PTR_TBL_ENT_t **curentp;
13961 curentp = ary + oldsize;
13963 if ((newsize & PTR_TABLE_HASH(ent->oldval)) != i) {
13965 ent->next = *curentp;
13975 /* remove all the entries from a ptr table */
13976 /* Deprecated - will be removed post 5.14 */
13979 Perl_ptr_table_clear(pTHX_ PTR_TBL_t *const tbl)
13981 PERL_UNUSED_CONTEXT;
13982 if (tbl && tbl->tbl_items) {
13983 struct ptr_tbl_arena *arena = tbl->tbl_arena;
13985 Zero(tbl->tbl_ary, tbl->tbl_max + 1, struct ptr_tbl_ent *);
13988 struct ptr_tbl_arena *next = arena->next;
13994 tbl->tbl_items = 0;
13995 tbl->tbl_arena = NULL;
13996 tbl->tbl_arena_next = NULL;
13997 tbl->tbl_arena_end = NULL;
14001 /* clear and free a ptr table */
14004 Perl_ptr_table_free(pTHX_ PTR_TBL_t *const tbl)
14006 struct ptr_tbl_arena *arena;
14008 PERL_UNUSED_CONTEXT;
14014 arena = tbl->tbl_arena;
14017 struct ptr_tbl_arena *next = arena->next;
14023 Safefree(tbl->tbl_ary);
14027 #if defined(USE_ITHREADS)
14030 Perl_rvpv_dup(pTHX_ SV *const dstr, const SV *const sstr, CLONE_PARAMS *const param)
14032 PERL_ARGS_ASSERT_RVPV_DUP;
14034 assert(!isREGEXP(sstr));
14036 if (SvWEAKREF(sstr)) {
14037 SvRV_set(dstr, sv_dup(SvRV_const(sstr), param));
14038 if (param->flags & CLONEf_JOIN_IN) {
14039 /* if joining, we add any back references individually rather
14040 * than copying the whole backref array */
14041 Perl_sv_add_backref(aTHX_ SvRV(dstr), dstr);
14045 SvRV_set(dstr, sv_dup_inc(SvRV_const(sstr), param));
14047 else if (SvPVX_const(sstr)) {
14048 /* Has something there */
14050 /* Normal PV - clone whole allocated space */
14051 SvPV_set(dstr, SAVEPVN(SvPVX_const(sstr), SvLEN(sstr)-1));
14052 /* sstr may not be that normal, but actually copy on write.
14053 But we are a true, independent SV, so: */
14057 /* Special case - not normally malloced for some reason */
14058 if (isGV_with_GP(sstr)) {
14059 /* Don't need to do anything here. */
14061 else if ((SvIsCOW(sstr))) {
14062 /* A "shared" PV - clone it as "shared" PV */
14064 HEK_KEY(hek_dup(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)),
14068 /* Some other special case - random pointer */
14069 SvPV_set(dstr, (char *) SvPVX_const(sstr));
14074 /* Copy the NULL */
14075 SvPV_set(dstr, NULL);
14079 /* duplicate a list of SVs. source and dest may point to the same memory. */
14081 S_sv_dup_inc_multiple(pTHX_ SV *const *source, SV **dest,
14082 SSize_t items, CLONE_PARAMS *const param)
14084 PERL_ARGS_ASSERT_SV_DUP_INC_MULTIPLE;
14086 while (items-- > 0) {
14087 *dest++ = sv_dup_inc(*source++, param);
14093 /* duplicate an SV of any type (including AV, HV etc) */
14096 S_sv_dup_common(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14101 PERL_ARGS_ASSERT_SV_DUP_COMMON;
14103 if (SvTYPE(sstr) == (svtype)SVTYPEMASK) {
14104 #ifdef DEBUG_LEAKING_SCALARS_ABORT
14109 /* look for it in the table first */
14110 dstr = MUTABLE_SV(ptr_table_fetch(PL_ptr_table, sstr));
14114 if(param->flags & CLONEf_JOIN_IN) {
14115 /** We are joining here so we don't want do clone
14116 something that is bad **/
14117 if (SvTYPE(sstr) == SVt_PVHV) {
14118 const HEK * const hvname = HvNAME_HEK(sstr);
14120 /** don't clone stashes if they already exist **/
14121 dstr = MUTABLE_SV(gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
14122 HEK_UTF8(hvname) ? SVf_UTF8 : 0));
14123 ptr_table_store(PL_ptr_table, sstr, dstr);
14127 else if (SvTYPE(sstr) == SVt_PVGV && !SvFAKE(sstr)) {
14128 HV *stash = GvSTASH(sstr);
14129 const HEK * hvname;
14130 if (stash && (hvname = HvNAME_HEK(stash))) {
14131 /** don't clone GVs if they already exist **/
14133 stash = gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
14134 HEK_UTF8(hvname) ? SVf_UTF8 : 0);
14136 stash, GvNAME(sstr),
14142 if (svp && *svp && SvTYPE(*svp) == SVt_PVGV) {
14143 ptr_table_store(PL_ptr_table, sstr, *svp);
14150 /* create anew and remember what it is */
14153 #ifdef DEBUG_LEAKING_SCALARS
14154 dstr->sv_debug_optype = sstr->sv_debug_optype;
14155 dstr->sv_debug_line = sstr->sv_debug_line;
14156 dstr->sv_debug_inpad = sstr->sv_debug_inpad;
14157 dstr->sv_debug_parent = (SV*)sstr;
14158 FREE_SV_DEBUG_FILE(dstr);
14159 dstr->sv_debug_file = savesharedpv(sstr->sv_debug_file);
14162 ptr_table_store(PL_ptr_table, sstr, dstr);
14165 SvFLAGS(dstr) = SvFLAGS(sstr);
14166 SvFLAGS(dstr) &= ~SVf_OOK; /* don't propagate OOK hack */
14167 SvREFCNT(dstr) = 0; /* must be before any other dups! */
14170 if (SvANY(sstr) && PL_watch_pvx && SvPVX_const(sstr) == PL_watch_pvx)
14171 PerlIO_printf(Perl_debug_log, "watch at %p hit, found string \"%s\"\n",
14172 (void*)PL_watch_pvx, SvPVX_const(sstr));
14175 /* don't clone objects whose class has asked us not to */
14177 && ! (SvFLAGS(SvSTASH(sstr)) & SVphv_CLONEABLE))
14183 switch (SvTYPE(sstr)) {
14185 SvANY(dstr) = NULL;
14188 SET_SVANY_FOR_BODYLESS_IV(dstr);
14190 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
14192 SvIV_set(dstr, SvIVX(sstr));
14196 #if NVSIZE <= IVSIZE
14197 SET_SVANY_FOR_BODYLESS_NV(dstr);
14199 SvANY(dstr) = new_XNV();
14201 SvNV_set(dstr, SvNVX(sstr));
14205 /* These are all the types that need complex bodies allocating. */
14207 const svtype sv_type = SvTYPE(sstr);
14208 const struct body_details *const sv_type_details
14209 = bodies_by_type + sv_type;
14213 Perl_croak(aTHX_ "Bizarre SvTYPE [%" IVdf "]", (IV)SvTYPE(sstr));
14214 NOT_REACHED; /* NOTREACHED */
14230 assert(sv_type_details->body_size);
14231 if (sv_type_details->arena) {
14232 new_body_inline(new_body, sv_type);
14234 = (void*)((char*)new_body - sv_type_details->offset);
14236 new_body = new_NOARENA(sv_type_details);
14240 SvANY(dstr) = new_body;
14243 Copy(((char*)SvANY(sstr)) + sv_type_details->offset,
14244 ((char*)SvANY(dstr)) + sv_type_details->offset,
14245 sv_type_details->copy, char);
14247 Copy(((char*)SvANY(sstr)),
14248 ((char*)SvANY(dstr)),
14249 sv_type_details->body_size + sv_type_details->offset, char);
14252 if (sv_type != SVt_PVAV && sv_type != SVt_PVHV
14253 && !isGV_with_GP(dstr)
14255 && !(sv_type == SVt_PVIO && !(IoFLAGS(dstr) & IOf_FAKE_DIRP)))
14256 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
14258 /* The Copy above means that all the source (unduplicated) pointers
14259 are now in the destination. We can check the flags and the
14260 pointers in either, but it's possible that there's less cache
14261 missing by always going for the destination.
14262 FIXME - instrument and check that assumption */
14263 if (sv_type >= SVt_PVMG) {
14265 SvMAGIC_set(dstr, mg_dup(SvMAGIC(dstr), param));
14266 if (SvOBJECT(dstr) && SvSTASH(dstr))
14267 SvSTASH_set(dstr, hv_dup_inc(SvSTASH(dstr), param));
14268 else SvSTASH_set(dstr, 0); /* don't copy DESTROY cache */
14271 /* The cast silences a GCC warning about unhandled types. */
14272 switch ((int)sv_type) {
14283 /* FIXME for plugins */
14284 re_dup_guts((REGEXP*) sstr, (REGEXP*) dstr, param);
14287 /* XXX LvTARGOFF sometimes holds PMOP* when DEBUGGING */
14288 if (LvTYPE(dstr) == 't') /* for tie: unrefcnted fake (SV**) */
14289 LvTARG(dstr) = dstr;
14290 else if (LvTYPE(dstr) == 'T') /* for tie: fake HE */
14291 LvTARG(dstr) = MUTABLE_SV(he_dup((HE*)LvTARG(dstr), 0, param));
14293 LvTARG(dstr) = sv_dup_inc(LvTARG(dstr), param);
14294 if (isREGEXP(sstr)) goto duprex;
14297 /* non-GP case already handled above */
14298 if(isGV_with_GP(sstr)) {
14299 GvNAME_HEK(dstr) = hek_dup(GvNAME_HEK(dstr), param);
14300 /* Don't call sv_add_backref here as it's going to be
14301 created as part of the magic cloning of the symbol
14302 table--unless this is during a join and the stash
14303 is not actually being cloned. */
14304 /* Danger Will Robinson - GvGP(dstr) isn't initialised
14305 at the point of this comment. */
14306 GvSTASH(dstr) = hv_dup(GvSTASH(dstr), param);
14307 if (param->flags & CLONEf_JOIN_IN)
14308 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
14309 GvGP_set(dstr, gp_dup(GvGP(sstr), param));
14310 (void)GpREFCNT_inc(GvGP(dstr));
14314 /* PL_parser->rsfp_filters entries have fake IoDIRP() */
14315 if(IoFLAGS(dstr) & IOf_FAKE_DIRP) {
14316 /* I have no idea why fake dirp (rsfps)
14317 should be treated differently but otherwise
14318 we end up with leaks -- sky*/
14319 IoTOP_GV(dstr) = gv_dup_inc(IoTOP_GV(dstr), param);
14320 IoFMT_GV(dstr) = gv_dup_inc(IoFMT_GV(dstr), param);
14321 IoBOTTOM_GV(dstr) = gv_dup_inc(IoBOTTOM_GV(dstr), param);
14323 IoTOP_GV(dstr) = gv_dup(IoTOP_GV(dstr), param);
14324 IoFMT_GV(dstr) = gv_dup(IoFMT_GV(dstr), param);
14325 IoBOTTOM_GV(dstr) = gv_dup(IoBOTTOM_GV(dstr), param);
14326 if (IoDIRP(dstr)) {
14327 IoDIRP(dstr) = dirp_dup(IoDIRP(dstr), param);
14330 /* IoDIRP(dstr) is already a copy of IoDIRP(sstr) */
14332 IoIFP(dstr) = fp_dup(IoIFP(sstr), IoTYPE(dstr), param);
14334 if (IoOFP(dstr) == IoIFP(sstr))
14335 IoOFP(dstr) = IoIFP(dstr);
14337 IoOFP(dstr) = fp_dup(IoOFP(dstr), IoTYPE(dstr), param);
14338 IoTOP_NAME(dstr) = SAVEPV(IoTOP_NAME(dstr));
14339 IoFMT_NAME(dstr) = SAVEPV(IoFMT_NAME(dstr));
14340 IoBOTTOM_NAME(dstr) = SAVEPV(IoBOTTOM_NAME(dstr));
14343 /* avoid cloning an empty array */
14344 if (AvARRAY((const AV *)sstr) && AvFILLp((const AV *)sstr) >= 0) {
14345 SV **dst_ary, **src_ary;
14346 SSize_t items = AvFILLp((const AV *)sstr) + 1;
14348 src_ary = AvARRAY((const AV *)sstr);
14349 Newx(dst_ary, AvMAX((const AV *)sstr)+1, SV*);
14350 ptr_table_store(PL_ptr_table, src_ary, dst_ary);
14351 AvARRAY(MUTABLE_AV(dstr)) = dst_ary;
14352 AvALLOC((const AV *)dstr) = dst_ary;
14353 if (AvREAL((const AV *)sstr)) {
14354 dst_ary = sv_dup_inc_multiple(src_ary, dst_ary, items,
14358 while (items-- > 0)
14359 *dst_ary++ = sv_dup(*src_ary++, param);
14361 items = AvMAX((const AV *)sstr) - AvFILLp((const AV *)sstr);
14362 while (items-- > 0) {
14367 AvARRAY(MUTABLE_AV(dstr)) = NULL;
14368 AvALLOC((const AV *)dstr) = (SV**)NULL;
14369 AvMAX( (const AV *)dstr) = -1;
14370 AvFILLp((const AV *)dstr) = -1;
14374 if (HvARRAY((const HV *)sstr)) {
14376 const bool sharekeys = !!HvSHAREKEYS(sstr);
14377 XPVHV * const dxhv = (XPVHV*)SvANY(dstr);
14378 XPVHV * const sxhv = (XPVHV*)SvANY(sstr);
14380 Newx(darray, PERL_HV_ARRAY_ALLOC_BYTES(dxhv->xhv_max+1)
14381 + (SvOOK(sstr) ? sizeof(struct xpvhv_aux) : 0),
14383 HvARRAY(dstr) = (HE**)darray;
14384 while (i <= sxhv->xhv_max) {
14385 const HE * const source = HvARRAY(sstr)[i];
14386 HvARRAY(dstr)[i] = source
14387 ? he_dup(source, sharekeys, param) : 0;
14391 const struct xpvhv_aux * const saux = HvAUX(sstr);
14392 struct xpvhv_aux * const daux = HvAUX(dstr);
14393 /* This flag isn't copied. */
14396 if (saux->xhv_name_count) {
14397 HEK ** const sname = saux->xhv_name_u.xhvnameu_names;
14399 = saux->xhv_name_count < 0
14400 ? -saux->xhv_name_count
14401 : saux->xhv_name_count;
14402 HEK **shekp = sname + count;
14404 Newx(daux->xhv_name_u.xhvnameu_names, count, HEK *);
14405 dhekp = daux->xhv_name_u.xhvnameu_names + count;
14406 while (shekp-- > sname) {
14408 *dhekp = hek_dup(*shekp, param);
14412 daux->xhv_name_u.xhvnameu_name
14413 = hek_dup(saux->xhv_name_u.xhvnameu_name,
14416 daux->xhv_name_count = saux->xhv_name_count;
14418 daux->xhv_aux_flags = saux->xhv_aux_flags;
14419 #ifdef PERL_HASH_RANDOMIZE_KEYS
14420 daux->xhv_rand = saux->xhv_rand;
14421 daux->xhv_last_rand = saux->xhv_last_rand;
14423 daux->xhv_riter = saux->xhv_riter;
14424 daux->xhv_eiter = saux->xhv_eiter
14425 ? he_dup(saux->xhv_eiter,
14426 cBOOL(HvSHAREKEYS(sstr)), param) : 0;
14427 /* backref array needs refcnt=2; see sv_add_backref */
14428 daux->xhv_backreferences =
14429 (param->flags & CLONEf_JOIN_IN)
14430 /* when joining, we let the individual GVs and
14431 * CVs add themselves to backref as
14432 * needed. This avoids pulling in stuff
14433 * that isn't required, and simplifies the
14434 * case where stashes aren't cloned back
14435 * if they already exist in the parent
14438 : saux->xhv_backreferences
14439 ? (SvTYPE(saux->xhv_backreferences) == SVt_PVAV)
14440 ? MUTABLE_AV(SvREFCNT_inc(
14441 sv_dup_inc((const SV *)
14442 saux->xhv_backreferences, param)))
14443 : MUTABLE_AV(sv_dup((const SV *)
14444 saux->xhv_backreferences, param))
14447 daux->xhv_mro_meta = saux->xhv_mro_meta
14448 ? mro_meta_dup(saux->xhv_mro_meta, param)
14451 /* Record stashes for possible cloning in Perl_clone(). */
14453 av_push(param->stashes, dstr);
14457 HvARRAY(MUTABLE_HV(dstr)) = NULL;
14460 if (!(param->flags & CLONEf_COPY_STACKS)) {
14465 /* NOTE: not refcounted */
14466 SvANY(MUTABLE_CV(dstr))->xcv_stash =
14467 hv_dup(CvSTASH(dstr), param);
14468 if ((param->flags & CLONEf_JOIN_IN) && CvSTASH(dstr))
14469 Perl_sv_add_backref(aTHX_ MUTABLE_SV(CvSTASH(dstr)), dstr);
14470 if (!CvISXSUB(dstr)) {
14472 CvROOT(dstr) = OpREFCNT_inc(CvROOT(dstr));
14474 CvSLABBED_off(dstr);
14475 } else if (CvCONST(dstr)) {
14476 CvXSUBANY(dstr).any_ptr =
14477 sv_dup_inc((const SV *)CvXSUBANY(dstr).any_ptr, param);
14479 assert(!CvSLABBED(dstr));
14480 if (CvDYNFILE(dstr)) CvFILE(dstr) = SAVEPV(CvFILE(dstr));
14482 SvANY((CV *)dstr)->xcv_gv_u.xcv_hek =
14483 hek_dup(CvNAME_HEK((CV *)sstr), param);
14484 /* don't dup if copying back - CvGV isn't refcounted, so the
14485 * duped GV may never be freed. A bit of a hack! DAPM */
14487 SvANY(MUTABLE_CV(dstr))->xcv_gv_u.xcv_gv =
14489 ? gv_dup_inc(CvGV(sstr), param)
14490 : (param->flags & CLONEf_JOIN_IN)
14492 : gv_dup(CvGV(sstr), param);
14494 if (!CvISXSUB(sstr)) {
14495 PADLIST * padlist = CvPADLIST(sstr);
14497 padlist = padlist_dup(padlist, param);
14498 CvPADLIST_set(dstr, padlist);
14500 /* unthreaded perl can't sv_dup so we dont support unthreaded's CvHSCXT */
14501 PoisonPADLIST(dstr);
14504 CvWEAKOUTSIDE(sstr)
14505 ? cv_dup( CvOUTSIDE(dstr), param)
14506 : cv_dup_inc(CvOUTSIDE(dstr), param);
14516 Perl_sv_dup_inc(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14518 PERL_ARGS_ASSERT_SV_DUP_INC;
14519 return sstr ? SvREFCNT_inc(sv_dup_common(sstr, param)) : NULL;
14523 Perl_sv_dup(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14525 SV *dstr = sstr ? sv_dup_common(sstr, param) : NULL;
14526 PERL_ARGS_ASSERT_SV_DUP;
14528 /* Track every SV that (at least initially) had a reference count of 0.
14529 We need to do this by holding an actual reference to it in this array.
14530 If we attempt to cheat, turn AvREAL_off(), and store only pointers
14531 (akin to the stashes hash, and the perl stack), we come unstuck if
14532 a weak reference (or other SV legitimately SvREFCNT() == 0 for this
14533 thread) is manipulated in a CLONE method, because CLONE runs before the
14534 unreferenced array is walked to find SVs still with SvREFCNT() == 0
14535 (and fix things up by giving each a reference via the temps stack).
14536 Instead, during CLONE, if the 0-referenced SV has SvREFCNT_inc() and
14537 then SvREFCNT_dec(), it will be cleaned up (and added to the free list)
14538 before the walk of unreferenced happens and a reference to that is SV
14539 added to the temps stack. At which point we have the same SV considered
14540 to be in use, and free to be re-used. Not good.
14542 if (dstr && !(param->flags & CLONEf_COPY_STACKS) && !SvREFCNT(dstr)) {
14543 assert(param->unreferenced);
14544 av_push(param->unreferenced, SvREFCNT_inc(dstr));
14550 /* duplicate a context */
14553 Perl_cx_dup(pTHX_ PERL_CONTEXT *cxs, I32 ix, I32 max, CLONE_PARAMS* param)
14555 PERL_CONTEXT *ncxs;
14557 PERL_ARGS_ASSERT_CX_DUP;
14560 return (PERL_CONTEXT*)NULL;
14562 /* look for it in the table first */
14563 ncxs = (PERL_CONTEXT*)ptr_table_fetch(PL_ptr_table, cxs);
14567 /* create anew and remember what it is */
14568 Newx(ncxs, max + 1, PERL_CONTEXT);
14569 ptr_table_store(PL_ptr_table, cxs, ncxs);
14570 Copy(cxs, ncxs, max + 1, PERL_CONTEXT);
14573 PERL_CONTEXT * const ncx = &ncxs[ix];
14574 if (CxTYPE(ncx) == CXt_SUBST) {
14575 Perl_croak(aTHX_ "Cloning substitution context is unimplemented");
14578 ncx->blk_oldcop = (COP*)any_dup(ncx->blk_oldcop, param->proto_perl);
14579 switch (CxTYPE(ncx)) {
14581 ncx->blk_sub.cv = cv_dup_inc(ncx->blk_sub.cv, param);
14582 if(CxHASARGS(ncx)){
14583 ncx->blk_sub.savearray = av_dup_inc(ncx->blk_sub.savearray,param);
14585 ncx->blk_sub.savearray = NULL;
14587 ncx->blk_sub.prevcomppad = (PAD*)ptr_table_fetch(PL_ptr_table,
14588 ncx->blk_sub.prevcomppad);
14591 ncx->blk_eval.old_namesv = sv_dup_inc(ncx->blk_eval.old_namesv,
14593 /* XXX should this sv_dup_inc? Or only if CxEVAL_TXT_REFCNTED ???? */
14594 ncx->blk_eval.cur_text = sv_dup(ncx->blk_eval.cur_text, param);
14595 ncx->blk_eval.cv = cv_dup(ncx->blk_eval.cv, param);
14596 /* XXX what do do with cur_top_env ???? */
14598 case CXt_LOOP_LAZYSV:
14599 ncx->blk_loop.state_u.lazysv.end
14600 = sv_dup_inc(ncx->blk_loop.state_u.lazysv.end, param);
14601 /* Fallthrough: duplicate lazysv.cur by using the ary.ary
14602 duplication code instead.
14603 We are taking advantage of (1) av_dup_inc and sv_dup_inc
14604 actually being the same function, and (2) order
14605 equivalence of the two unions.
14606 We can assert the later [but only at run time :-(] */
14607 assert ((void *) &ncx->blk_loop.state_u.ary.ary ==
14608 (void *) &ncx->blk_loop.state_u.lazysv.cur);
14611 ncx->blk_loop.state_u.ary.ary
14612 = av_dup_inc(ncx->blk_loop.state_u.ary.ary, param);
14614 case CXt_LOOP_LIST:
14615 case CXt_LOOP_LAZYIV:
14616 /* code common to all 'for' CXt_LOOP_* types */
14617 ncx->blk_loop.itersave =
14618 sv_dup_inc(ncx->blk_loop.itersave, param);
14619 if (CxPADLOOP(ncx)) {
14620 PADOFFSET off = ncx->blk_loop.itervar_u.svp
14621 - &CX_CURPAD_SV(ncx->blk_loop, 0);
14622 ncx->blk_loop.oldcomppad =
14623 (PAD*)ptr_table_fetch(PL_ptr_table,
14624 ncx->blk_loop.oldcomppad);
14625 ncx->blk_loop.itervar_u.svp =
14626 &CX_CURPAD_SV(ncx->blk_loop, off);
14629 /* this copies the GV if CXp_FOR_GV, or the SV for an
14630 * alias (for \$x (...)) - relies on gv_dup being the
14631 * same as sv_dup */
14632 ncx->blk_loop.itervar_u.gv
14633 = gv_dup((const GV *)ncx->blk_loop.itervar_u.gv,
14637 case CXt_LOOP_PLAIN:
14640 ncx->blk_format.prevcomppad =
14641 (PAD*)ptr_table_fetch(PL_ptr_table,
14642 ncx->blk_format.prevcomppad);
14643 ncx->blk_format.cv = cv_dup_inc(ncx->blk_format.cv, param);
14644 ncx->blk_format.gv = gv_dup(ncx->blk_format.gv, param);
14645 ncx->blk_format.dfoutgv = gv_dup_inc(ncx->blk_format.dfoutgv,
14649 ncx->blk_givwhen.defsv_save =
14650 sv_dup_inc(ncx->blk_givwhen.defsv_save, param);
14663 /* duplicate a stack info structure */
14666 Perl_si_dup(pTHX_ PERL_SI *si, CLONE_PARAMS* param)
14670 PERL_ARGS_ASSERT_SI_DUP;
14673 return (PERL_SI*)NULL;
14675 /* look for it in the table first */
14676 nsi = (PERL_SI*)ptr_table_fetch(PL_ptr_table, si);
14680 /* create anew and remember what it is */
14681 Newx(nsi, 1, PERL_SI);
14682 ptr_table_store(PL_ptr_table, si, nsi);
14684 nsi->si_stack = av_dup_inc(si->si_stack, param);
14685 nsi->si_cxix = si->si_cxix;
14686 nsi->si_cxsubix = si->si_cxsubix;
14687 nsi->si_cxmax = si->si_cxmax;
14688 nsi->si_cxstack = cx_dup(si->si_cxstack, si->si_cxix, si->si_cxmax, param);
14689 nsi->si_type = si->si_type;
14690 nsi->si_prev = si_dup(si->si_prev, param);
14691 nsi->si_next = si_dup(si->si_next, param);
14692 nsi->si_markoff = si->si_markoff;
14693 #if defined DEBUGGING && !defined DEBUGGING_RE_ONLY
14694 nsi->si_stack_hwm = 0;
14700 #define POPINT(ss,ix) ((ss)[--(ix)].any_i32)
14701 #define TOPINT(ss,ix) ((ss)[ix].any_i32)
14702 #define POPLONG(ss,ix) ((ss)[--(ix)].any_long)
14703 #define TOPLONG(ss,ix) ((ss)[ix].any_long)
14704 #define POPIV(ss,ix) ((ss)[--(ix)].any_iv)
14705 #define TOPIV(ss,ix) ((ss)[ix].any_iv)
14706 #define POPUV(ss,ix) ((ss)[--(ix)].any_uv)
14707 #define TOPUV(ss,ix) ((ss)[ix].any_uv)
14708 #define POPBOOL(ss,ix) ((ss)[--(ix)].any_bool)
14709 #define TOPBOOL(ss,ix) ((ss)[ix].any_bool)
14710 #define POPPTR(ss,ix) ((ss)[--(ix)].any_ptr)
14711 #define TOPPTR(ss,ix) ((ss)[ix].any_ptr)
14712 #define POPDPTR(ss,ix) ((ss)[--(ix)].any_dptr)
14713 #define TOPDPTR(ss,ix) ((ss)[ix].any_dptr)
14714 #define POPDXPTR(ss,ix) ((ss)[--(ix)].any_dxptr)
14715 #define TOPDXPTR(ss,ix) ((ss)[ix].any_dxptr)
14718 #define pv_dup_inc(p) SAVEPV(p)
14719 #define pv_dup(p) SAVEPV(p)
14720 #define svp_dup_inc(p,pp) any_dup(p,pp)
14722 /* map any object to the new equivent - either something in the
14723 * ptr table, or something in the interpreter structure
14727 Perl_any_dup(pTHX_ void *v, const PerlInterpreter *proto_perl)
14731 PERL_ARGS_ASSERT_ANY_DUP;
14734 return (void*)NULL;
14736 /* look for it in the table first */
14737 ret = ptr_table_fetch(PL_ptr_table, v);
14741 /* see if it is part of the interpreter structure */
14742 if (v >= (void*)proto_perl && v < (void*)(proto_perl+1))
14743 ret = (void*)(((char*)aTHX) + (((char*)v) - (char*)proto_perl));
14751 /* duplicate the save stack */
14754 Perl_ss_dup(pTHX_ PerlInterpreter *proto_perl, CLONE_PARAMS* param)
14757 ANY * const ss = proto_perl->Isavestack;
14758 const I32 max = proto_perl->Isavestack_max + SS_MAXPUSH;
14759 I32 ix = proto_perl->Isavestack_ix;
14772 void (*dptr) (void*);
14773 void (*dxptr) (pTHX_ void*);
14775 PERL_ARGS_ASSERT_SS_DUP;
14777 Newx(nss, max, ANY);
14780 const UV uv = POPUV(ss,ix);
14781 const U8 type = (U8)uv & SAVE_MASK;
14783 TOPUV(nss,ix) = uv;
14785 case SAVEt_CLEARSV:
14786 case SAVEt_CLEARPADRANGE:
14788 case SAVEt_HELEM: /* hash element */
14789 case SAVEt_SV: /* scalar reference */
14790 sv = (const SV *)POPPTR(ss,ix);
14791 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14793 case SAVEt_ITEM: /* normal string */
14794 case SAVEt_GVSV: /* scalar slot in GV */
14795 sv = (const SV *)POPPTR(ss,ix);
14796 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14797 if (type == SAVEt_SV)
14801 case SAVEt_MORTALIZESV:
14802 case SAVEt_READONLY_OFF:
14803 sv = (const SV *)POPPTR(ss,ix);
14804 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14806 case SAVEt_FREEPADNAME:
14807 ptr = POPPTR(ss,ix);
14808 TOPPTR(nss,ix) = padname_dup((PADNAME *)ptr, param);
14809 PadnameREFCNT((PADNAME *)TOPPTR(nss,ix))++;
14811 case SAVEt_SHARED_PVREF: /* char* in shared space */
14812 c = (char*)POPPTR(ss,ix);
14813 TOPPTR(nss,ix) = savesharedpv(c);
14814 ptr = POPPTR(ss,ix);
14815 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14817 case SAVEt_GENERIC_SVREF: /* generic sv */
14818 case SAVEt_SVREF: /* scalar reference */
14819 sv = (const SV *)POPPTR(ss,ix);
14820 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14821 if (type == SAVEt_SVREF)
14822 SvREFCNT_inc_simple_void((SV *)TOPPTR(nss,ix));
14823 ptr = POPPTR(ss,ix);
14824 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14826 case SAVEt_GVSLOT: /* any slot in GV */
14827 sv = (const SV *)POPPTR(ss,ix);
14828 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14829 ptr = POPPTR(ss,ix);
14830 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14831 sv = (const SV *)POPPTR(ss,ix);
14832 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14834 case SAVEt_HV: /* hash reference */
14835 case SAVEt_AV: /* array reference */
14836 sv = (const SV *) POPPTR(ss,ix);
14837 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14839 case SAVEt_COMPPAD:
14841 sv = (const SV *) POPPTR(ss,ix);
14842 TOPPTR(nss,ix) = sv_dup(sv, param);
14844 case SAVEt_INT: /* int reference */
14845 ptr = POPPTR(ss,ix);
14846 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14847 intval = (int)POPINT(ss,ix);
14848 TOPINT(nss,ix) = intval;
14850 case SAVEt_LONG: /* long reference */
14851 ptr = POPPTR(ss,ix);
14852 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14853 longval = (long)POPLONG(ss,ix);
14854 TOPLONG(nss,ix) = longval;
14856 case SAVEt_I32: /* I32 reference */
14857 ptr = POPPTR(ss,ix);
14858 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14860 TOPINT(nss,ix) = i;
14862 case SAVEt_IV: /* IV reference */
14863 case SAVEt_STRLEN: /* STRLEN/size_t ref */
14864 ptr = POPPTR(ss,ix);
14865 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14867 TOPIV(nss,ix) = iv;
14869 case SAVEt_TMPSFLOOR:
14871 TOPIV(nss,ix) = iv;
14873 case SAVEt_HPTR: /* HV* reference */
14874 case SAVEt_APTR: /* AV* reference */
14875 case SAVEt_SPTR: /* SV* reference */
14876 ptr = POPPTR(ss,ix);
14877 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14878 sv = (const SV *)POPPTR(ss,ix);
14879 TOPPTR(nss,ix) = sv_dup(sv, param);
14881 case SAVEt_VPTR: /* random* reference */
14882 ptr = POPPTR(ss,ix);
14883 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14885 case SAVEt_INT_SMALL:
14886 case SAVEt_I32_SMALL:
14887 case SAVEt_I16: /* I16 reference */
14888 case SAVEt_I8: /* I8 reference */
14890 ptr = POPPTR(ss,ix);
14891 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14893 case SAVEt_GENERIC_PVREF: /* generic char* */
14894 case SAVEt_PPTR: /* char* reference */
14895 ptr = POPPTR(ss,ix);
14896 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14897 c = (char*)POPPTR(ss,ix);
14898 TOPPTR(nss,ix) = pv_dup(c);
14900 case SAVEt_GP: /* scalar reference */
14901 gp = (GP*)POPPTR(ss,ix);
14902 TOPPTR(nss,ix) = gp = gp_dup(gp, param);
14903 (void)GpREFCNT_inc(gp);
14904 gv = (const GV *)POPPTR(ss,ix);
14905 TOPPTR(nss,ix) = gv_dup_inc(gv, param);
14908 ptr = POPPTR(ss,ix);
14909 if (ptr && (((OP*)ptr)->op_private & OPpREFCOUNTED)) {
14910 /* these are assumed to be refcounted properly */
14912 switch (((OP*)ptr)->op_type) {
14914 case OP_LEAVESUBLV:
14918 case OP_LEAVEWRITE:
14919 TOPPTR(nss,ix) = ptr;
14922 (void) OpREFCNT_inc(o);
14926 TOPPTR(nss,ix) = NULL;
14931 TOPPTR(nss,ix) = NULL;
14933 case SAVEt_FREECOPHH:
14934 ptr = POPPTR(ss,ix);
14935 TOPPTR(nss,ix) = cophh_copy((COPHH *)ptr);
14937 case SAVEt_ADELETE:
14938 av = (const AV *)POPPTR(ss,ix);
14939 TOPPTR(nss,ix) = av_dup_inc(av, param);
14941 TOPINT(nss,ix) = i;
14944 hv = (const HV *)POPPTR(ss,ix);
14945 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14947 TOPINT(nss,ix) = i;
14950 c = (char*)POPPTR(ss,ix);
14951 TOPPTR(nss,ix) = pv_dup_inc(c);
14953 case SAVEt_STACK_POS: /* Position on Perl stack */
14955 TOPINT(nss,ix) = i;
14957 case SAVEt_DESTRUCTOR:
14958 ptr = POPPTR(ss,ix);
14959 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14960 dptr = POPDPTR(ss,ix);
14961 TOPDPTR(nss,ix) = DPTR2FPTR(void (*)(void*),
14962 any_dup(FPTR2DPTR(void *, dptr),
14965 case SAVEt_DESTRUCTOR_X:
14966 ptr = POPPTR(ss,ix);
14967 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14968 dxptr = POPDXPTR(ss,ix);
14969 TOPDXPTR(nss,ix) = DPTR2FPTR(void (*)(pTHX_ void*),
14970 any_dup(FPTR2DPTR(void *, dxptr),
14973 case SAVEt_REGCONTEXT:
14975 ix -= uv >> SAVE_TIGHT_SHIFT;
14977 case SAVEt_AELEM: /* array element */
14978 sv = (const SV *)POPPTR(ss,ix);
14979 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14981 TOPIV(nss,ix) = iv;
14982 av = (const AV *)POPPTR(ss,ix);
14983 TOPPTR(nss,ix) = av_dup_inc(av, param);
14986 ptr = POPPTR(ss,ix);
14987 TOPPTR(nss,ix) = ptr;
14990 ptr = POPPTR(ss,ix);
14991 ptr = cophh_copy((COPHH*)ptr);
14992 TOPPTR(nss,ix) = ptr;
14994 TOPINT(nss,ix) = i;
14995 if (i & HINT_LOCALIZE_HH) {
14996 hv = (const HV *)POPPTR(ss,ix);
14997 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
15000 case SAVEt_PADSV_AND_MORTALIZE:
15001 longval = (long)POPLONG(ss,ix);
15002 TOPLONG(nss,ix) = longval;
15003 ptr = POPPTR(ss,ix);
15004 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
15005 sv = (const SV *)POPPTR(ss,ix);
15006 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
15008 case SAVEt_SET_SVFLAGS:
15010 TOPINT(nss,ix) = i;
15012 TOPINT(nss,ix) = i;
15013 sv = (const SV *)POPPTR(ss,ix);
15014 TOPPTR(nss,ix) = sv_dup(sv, param);
15016 case SAVEt_COMPILE_WARNINGS:
15017 ptr = POPPTR(ss,ix);
15018 TOPPTR(nss,ix) = DUP_WARNINGS((STRLEN*)ptr);
15021 ptr = POPPTR(ss,ix);
15022 TOPPTR(nss,ix) = parser_dup((const yy_parser*)ptr, param);
15026 "panic: ss_dup inconsistency (%" IVdf ")", (IV) type);
15034 /* if sv is a stash, call $class->CLONE_SKIP(), and set the SVphv_CLONEABLE
15035 * flag to the result. This is done for each stash before cloning starts,
15036 * so we know which stashes want their objects cloned */
15039 do_mark_cloneable_stash(pTHX_ SV *const sv)
15041 const HEK * const hvname = HvNAME_HEK((const HV *)sv);
15043 GV* const cloner = gv_fetchmethod_autoload(MUTABLE_HV(sv), "CLONE_SKIP", 0);
15044 SvFLAGS(sv) |= SVphv_CLONEABLE; /* clone objects by default */
15045 if (cloner && GvCV(cloner)) {
15052 mXPUSHs(newSVhek(hvname));
15054 call_sv(MUTABLE_SV(GvCV(cloner)), G_SCALAR);
15061 SvFLAGS(sv) &= ~SVphv_CLONEABLE;
15069 =for apidoc perl_clone
15071 Create and return a new interpreter by cloning the current one.
15073 C<perl_clone> takes these flags as parameters:
15075 C<CLONEf_COPY_STACKS> - is used to, well, copy the stacks also,
15076 without it we only clone the data and zero the stacks,
15077 with it we copy the stacks and the new perl interpreter is
15078 ready to run at the exact same point as the previous one.
15079 The pseudo-fork code uses C<COPY_STACKS> while the
15080 threads->create doesn't.
15082 C<CLONEf_KEEP_PTR_TABLE> -
15083 C<perl_clone> keeps a ptr_table with the pointer of the old
15084 variable as a key and the new variable as a value,
15085 this allows it to check if something has been cloned and not
15086 clone it again, but rather just use the value and increase the
15088 If C<KEEP_PTR_TABLE> is not set then C<perl_clone> will kill the ptr_table
15089 using the function S<C<ptr_table_free(PL_ptr_table); PL_ptr_table = NULL;>>.
15090 A reason to keep it around is if you want to dup some of your own
15091 variables which are outside the graph that perl scans.
15093 C<CLONEf_CLONE_HOST> -
15094 This is a win32 thing, it is ignored on unix, it tells perl's
15095 win32host code (which is c++) to clone itself, this is needed on
15096 win32 if you want to run two threads at the same time,
15097 if you just want to do some stuff in a separate perl interpreter
15098 and then throw it away and return to the original one,
15099 you don't need to do anything.
15104 /* XXX the above needs expanding by someone who actually understands it ! */
15105 EXTERN_C PerlInterpreter *
15106 perl_clone_host(PerlInterpreter* proto_perl, UV flags);
15109 perl_clone(PerlInterpreter *proto_perl, UV flags)
15112 #ifdef PERL_IMPLICIT_SYS
15114 PERL_ARGS_ASSERT_PERL_CLONE;
15116 /* perlhost.h so we need to call into it
15117 to clone the host, CPerlHost should have a c interface, sky */
15119 #ifndef __amigaos4__
15120 if (flags & CLONEf_CLONE_HOST) {
15121 return perl_clone_host(proto_perl,flags);
15124 return perl_clone_using(proto_perl, flags,
15126 proto_perl->IMemShared,
15127 proto_perl->IMemParse,
15129 proto_perl->IStdIO,
15133 proto_perl->IProc);
15137 perl_clone_using(PerlInterpreter *proto_perl, UV flags,
15138 struct IPerlMem* ipM, struct IPerlMem* ipMS,
15139 struct IPerlMem* ipMP, struct IPerlEnv* ipE,
15140 struct IPerlStdIO* ipStd, struct IPerlLIO* ipLIO,
15141 struct IPerlDir* ipD, struct IPerlSock* ipS,
15142 struct IPerlProc* ipP)
15144 /* XXX many of the string copies here can be optimized if they're
15145 * constants; they need to be allocated as common memory and just
15146 * their pointers copied. */
15149 CLONE_PARAMS clone_params;
15150 CLONE_PARAMS* const param = &clone_params;
15152 PerlInterpreter * const my_perl = (PerlInterpreter*)(*ipM->pMalloc)(ipM, sizeof(PerlInterpreter));
15154 PERL_ARGS_ASSERT_PERL_CLONE_USING;
15155 #else /* !PERL_IMPLICIT_SYS */
15157 CLONE_PARAMS clone_params;
15158 CLONE_PARAMS* param = &clone_params;
15159 PerlInterpreter * const my_perl = (PerlInterpreter*)PerlMem_malloc(sizeof(PerlInterpreter));
15161 PERL_ARGS_ASSERT_PERL_CLONE;
15162 #endif /* PERL_IMPLICIT_SYS */
15164 /* for each stash, determine whether its objects should be cloned */
15165 S_visit(proto_perl, do_mark_cloneable_stash, SVt_PVHV, SVTYPEMASK);
15166 PERL_SET_THX(my_perl);
15169 PoisonNew(my_perl, 1, PerlInterpreter);
15172 PL_defstash = NULL; /* may be used by perl malloc() */
15175 PL_scopestack_name = 0;
15177 PL_savestack_ix = 0;
15178 PL_savestack_max = -1;
15179 PL_sig_pending = 0;
15181 Zero(&PL_debug_pad, 1, struct perl_debug_pad);
15182 Zero(&PL_padname_undef, 1, PADNAME);
15183 Zero(&PL_padname_const, 1, PADNAME);
15184 # ifdef DEBUG_LEAKING_SCALARS
15185 PL_sv_serial = (((UV)my_perl >> 2) & 0xfff) * 1000000;
15187 # ifdef PERL_TRACE_OPS
15188 Zero(PL_op_exec_cnt, OP_max+2, UV);
15190 #else /* !DEBUGGING */
15191 Zero(my_perl, 1, PerlInterpreter);
15192 #endif /* DEBUGGING */
15194 #ifdef PERL_IMPLICIT_SYS
15195 /* host pointers */
15197 PL_MemShared = ipMS;
15198 PL_MemParse = ipMP;
15205 #endif /* PERL_IMPLICIT_SYS */
15208 param->flags = flags;
15209 /* Nothing in the core code uses this, but we make it available to
15210 extensions (using mg_dup). */
15211 param->proto_perl = proto_perl;
15212 /* Likely nothing will use this, but it is initialised to be consistent
15213 with Perl_clone_params_new(). */
15214 param->new_perl = my_perl;
15215 param->unreferenced = NULL;
15218 INIT_TRACK_MEMPOOL(my_perl->Imemory_debug_header, my_perl);
15220 PL_body_arenas = NULL;
15221 Zero(&PL_body_roots, 1, PL_body_roots);
15225 PL_sv_arenaroot = NULL;
15227 PL_debug = proto_perl->Idebug;
15229 /* dbargs array probably holds garbage */
15232 PL_compiling = proto_perl->Icompiling;
15234 /* pseudo environmental stuff */
15235 PL_origargc = proto_perl->Iorigargc;
15236 PL_origargv = proto_perl->Iorigargv;
15238 #ifndef NO_TAINT_SUPPORT
15239 /* Set tainting stuff before PerlIO_debug can possibly get called */
15240 PL_tainting = proto_perl->Itainting;
15241 PL_taint_warn = proto_perl->Itaint_warn;
15243 PL_tainting = FALSE;
15244 PL_taint_warn = FALSE;
15247 PL_minus_c = proto_perl->Iminus_c;
15249 PL_localpatches = proto_perl->Ilocalpatches;
15250 PL_splitstr = proto_perl->Isplitstr;
15251 PL_minus_n = proto_perl->Iminus_n;
15252 PL_minus_p = proto_perl->Iminus_p;
15253 PL_minus_l = proto_perl->Iminus_l;
15254 PL_minus_a = proto_perl->Iminus_a;
15255 PL_minus_E = proto_perl->Iminus_E;
15256 PL_minus_F = proto_perl->Iminus_F;
15257 PL_doswitches = proto_perl->Idoswitches;
15258 PL_dowarn = proto_perl->Idowarn;
15259 #ifdef PERL_SAWAMPERSAND
15260 PL_sawampersand = proto_perl->Isawampersand;
15262 PL_unsafe = proto_perl->Iunsafe;
15263 PL_perldb = proto_perl->Iperldb;
15264 PL_perl_destruct_level = proto_perl->Iperl_destruct_level;
15265 PL_exit_flags = proto_perl->Iexit_flags;
15267 /* XXX time(&PL_basetime) when asked for? */
15268 PL_basetime = proto_perl->Ibasetime;
15270 PL_maxsysfd = proto_perl->Imaxsysfd;
15271 PL_statusvalue = proto_perl->Istatusvalue;
15273 PL_statusvalue_vms = proto_perl->Istatusvalue_vms;
15275 PL_statusvalue_posix = proto_perl->Istatusvalue_posix;
15278 /* RE engine related */
15279 PL_regmatch_slab = NULL;
15280 PL_reg_curpm = NULL;
15282 PL_sub_generation = proto_perl->Isub_generation;
15284 /* funky return mechanisms */
15285 PL_forkprocess = proto_perl->Iforkprocess;
15287 /* internal state */
15288 PL_main_start = proto_perl->Imain_start;
15289 PL_eval_root = proto_perl->Ieval_root;
15290 PL_eval_start = proto_perl->Ieval_start;
15292 PL_filemode = proto_perl->Ifilemode;
15293 PL_lastfd = proto_perl->Ilastfd;
15294 PL_oldname = proto_perl->Ioldname; /* XXX not quite right */
15295 PL_gensym = proto_perl->Igensym;
15297 PL_laststatval = proto_perl->Ilaststatval;
15298 PL_laststype = proto_perl->Ilaststype;
15301 PL_profiledata = NULL;
15303 PL_generation = proto_perl->Igeneration;
15305 PL_in_clean_objs = proto_perl->Iin_clean_objs;
15306 PL_in_clean_all = proto_perl->Iin_clean_all;
15308 PL_delaymagic_uid = proto_perl->Idelaymagic_uid;
15309 PL_delaymagic_euid = proto_perl->Idelaymagic_euid;
15310 PL_delaymagic_gid = proto_perl->Idelaymagic_gid;
15311 PL_delaymagic_egid = proto_perl->Idelaymagic_egid;
15312 PL_nomemok = proto_perl->Inomemok;
15313 PL_an = proto_perl->Ian;
15314 PL_evalseq = proto_perl->Ievalseq;
15315 PL_origenviron = proto_perl->Iorigenviron; /* XXX not quite right */
15316 PL_origalen = proto_perl->Iorigalen;
15318 PL_sighandlerp = proto_perl->Isighandlerp;
15320 PL_runops = proto_perl->Irunops;
15322 PL_subline = proto_perl->Isubline;
15324 PL_cv_has_eval = proto_perl->Icv_has_eval;
15327 PL_cryptseen = proto_perl->Icryptseen;
15330 #ifdef USE_LOCALE_COLLATE
15331 PL_collation_ix = proto_perl->Icollation_ix;
15332 PL_collation_standard = proto_perl->Icollation_standard;
15333 PL_collxfrm_base = proto_perl->Icollxfrm_base;
15334 PL_collxfrm_mult = proto_perl->Icollxfrm_mult;
15335 PL_strxfrm_max_cp = proto_perl->Istrxfrm_max_cp;
15336 #endif /* USE_LOCALE_COLLATE */
15338 #ifdef USE_LOCALE_NUMERIC
15339 PL_numeric_standard = proto_perl->Inumeric_standard;
15340 PL_numeric_underlying = proto_perl->Inumeric_underlying;
15341 PL_numeric_underlying_is_standard = proto_perl->Inumeric_underlying_is_standard;
15342 #endif /* !USE_LOCALE_NUMERIC */
15344 /* Did the locale setup indicate UTF-8? */
15345 PL_utf8locale = proto_perl->Iutf8locale;
15346 PL_in_utf8_CTYPE_locale = proto_perl->Iin_utf8_CTYPE_locale;
15347 PL_in_utf8_COLLATE_locale = proto_perl->Iin_utf8_COLLATE_locale;
15348 my_strlcpy(PL_locale_utf8ness, proto_perl->Ilocale_utf8ness, sizeof(PL_locale_utf8ness));
15349 #if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
15350 PL_lc_numeric_mutex_depth = 0;
15352 /* Unicode features (see perlrun/-C) */
15353 PL_unicode = proto_perl->Iunicode;
15355 /* Pre-5.8 signals control */
15356 PL_signals = proto_perl->Isignals;
15358 /* times() ticks per second */
15359 PL_clocktick = proto_perl->Iclocktick;
15361 /* Recursion stopper for PerlIO_find_layer */
15362 PL_in_load_module = proto_perl->Iin_load_module;
15364 /* sort() routine */
15365 PL_sort_RealCmp = proto_perl->Isort_RealCmp;
15367 /* Not really needed/useful since the reenrant_retint is "volatile",
15368 * but do it for consistency's sake. */
15369 PL_reentrant_retint = proto_perl->Ireentrant_retint;
15371 /* Hooks to shared SVs and locks. */
15372 PL_sharehook = proto_perl->Isharehook;
15373 PL_lockhook = proto_perl->Ilockhook;
15374 PL_unlockhook = proto_perl->Iunlockhook;
15375 PL_threadhook = proto_perl->Ithreadhook;
15376 PL_destroyhook = proto_perl->Idestroyhook;
15377 PL_signalhook = proto_perl->Isignalhook;
15379 PL_globhook = proto_perl->Iglobhook;
15382 PL_last_swash_hv = NULL; /* reinits on demand */
15383 PL_last_swash_klen = 0;
15384 PL_last_swash_key[0]= '\0';
15385 PL_last_swash_tmps = (U8*)NULL;
15386 PL_last_swash_slen = 0;
15388 PL_srand_called = proto_perl->Isrand_called;
15389 Copy(&(proto_perl->Irandom_state), &PL_random_state, 1, PL_RANDOM_STATE_TYPE);
15391 if (flags & CLONEf_COPY_STACKS) {
15392 /* next allocation will be PL_tmps_stack[PL_tmps_ix+1] */
15393 PL_tmps_ix = proto_perl->Itmps_ix;
15394 PL_tmps_max = proto_perl->Itmps_max;
15395 PL_tmps_floor = proto_perl->Itmps_floor;
15397 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
15398 * NOTE: unlike the others! */
15399 PL_scopestack_ix = proto_perl->Iscopestack_ix;
15400 PL_scopestack_max = proto_perl->Iscopestack_max;
15402 /* next SSPUSHFOO() sets PL_savestack[PL_savestack_ix]
15403 * NOTE: unlike the others! */
15404 PL_savestack_ix = proto_perl->Isavestack_ix;
15405 PL_savestack_max = proto_perl->Isavestack_max;
15408 PL_start_env = proto_perl->Istart_env; /* XXXXXX */
15409 PL_top_env = &PL_start_env;
15411 PL_op = proto_perl->Iop;
15414 PL_Xpv = (XPV*)NULL;
15415 my_perl->Ina = proto_perl->Ina;
15417 PL_statcache = proto_perl->Istatcache;
15419 #ifndef NO_TAINT_SUPPORT
15420 PL_tainted = proto_perl->Itainted;
15422 PL_tainted = FALSE;
15424 PL_curpm = proto_perl->Icurpm; /* XXX No PMOP ref count */
15426 PL_chopset = proto_perl->Ichopset; /* XXX never deallocated */
15428 PL_restartjmpenv = proto_perl->Irestartjmpenv;
15429 PL_restartop = proto_perl->Irestartop;
15430 PL_in_eval = proto_perl->Iin_eval;
15431 PL_delaymagic = proto_perl->Idelaymagic;
15432 PL_phase = proto_perl->Iphase;
15433 PL_localizing = proto_perl->Ilocalizing;
15435 PL_hv_fetch_ent_mh = NULL;
15436 PL_modcount = proto_perl->Imodcount;
15437 PL_lastgotoprobe = NULL;
15438 PL_dumpindent = proto_perl->Idumpindent;
15440 PL_efloatbuf = NULL; /* reinits on demand */
15441 PL_efloatsize = 0; /* reinits on demand */
15445 PL_colorset = 0; /* reinits PL_colors[] */
15446 /*PL_colors[6] = {0,0,0,0,0,0};*/
15448 /* Pluggable optimizer */
15449 PL_peepp = proto_perl->Ipeepp;
15450 PL_rpeepp = proto_perl->Irpeepp;
15451 /* op_free() hook */
15452 PL_opfreehook = proto_perl->Iopfreehook;
15454 #ifdef USE_REENTRANT_API
15455 /* XXX: things like -Dm will segfault here in perlio, but doing
15456 * PERL_SET_CONTEXT(proto_perl);
15457 * breaks too many other things
15459 Perl_reentrant_init(aTHX);
15462 /* create SV map for pointer relocation */
15463 PL_ptr_table = ptr_table_new();
15465 /* initialize these special pointers as early as possible */
15467 ptr_table_store(PL_ptr_table, &proto_perl->Isv_undef, &PL_sv_undef);
15468 ptr_table_store(PL_ptr_table, &proto_perl->Isv_no, &PL_sv_no);
15469 ptr_table_store(PL_ptr_table, &proto_perl->Isv_zero, &PL_sv_zero);
15470 ptr_table_store(PL_ptr_table, &proto_perl->Isv_yes, &PL_sv_yes);
15471 ptr_table_store(PL_ptr_table, &proto_perl->Ipadname_const,
15472 &PL_padname_const);
15474 /* create (a non-shared!) shared string table */
15475 PL_strtab = newHV();
15476 HvSHAREKEYS_off(PL_strtab);
15477 hv_ksplit(PL_strtab, HvTOTALKEYS(proto_perl->Istrtab));
15478 ptr_table_store(PL_ptr_table, proto_perl->Istrtab, PL_strtab);
15480 Zero(PL_sv_consts, SV_CONSTS_COUNT, SV*);
15482 /* This PV will be free'd special way so must set it same way op.c does */
15483 PL_compiling.cop_file = savesharedpv(PL_compiling.cop_file);
15484 ptr_table_store(PL_ptr_table, proto_perl->Icompiling.cop_file, PL_compiling.cop_file);
15486 ptr_table_store(PL_ptr_table, &proto_perl->Icompiling, &PL_compiling);
15487 PL_compiling.cop_warnings = DUP_WARNINGS(PL_compiling.cop_warnings);
15488 CopHINTHASH_set(&PL_compiling, cophh_copy(CopHINTHASH_get(&PL_compiling)));
15489 PL_curcop = (COP*)any_dup(proto_perl->Icurcop, proto_perl);
15491 param->stashes = newAV(); /* Setup array of objects to call clone on */
15492 /* This makes no difference to the implementation, as it always pushes
15493 and shifts pointers to other SVs without changing their reference
15494 count, with the array becoming empty before it is freed. However, it
15495 makes it conceptually clear what is going on, and will avoid some
15496 work inside av.c, filling slots between AvFILL() and AvMAX() with
15497 &PL_sv_undef, and SvREFCNT_dec()ing those. */
15498 AvREAL_off(param->stashes);
15500 if (!(flags & CLONEf_COPY_STACKS)) {
15501 param->unreferenced = newAV();
15504 #ifdef PERLIO_LAYERS
15505 /* Clone PerlIO tables as soon as we can handle general xx_dup() */
15506 PerlIO_clone(aTHX_ proto_perl, param);
15509 PL_envgv = gv_dup_inc(proto_perl->Ienvgv, param);
15510 PL_incgv = gv_dup_inc(proto_perl->Iincgv, param);
15511 PL_hintgv = gv_dup_inc(proto_perl->Ihintgv, param);
15512 PL_origfilename = SAVEPV(proto_perl->Iorigfilename);
15513 PL_xsubfilename = proto_perl->Ixsubfilename;
15514 PL_diehook = sv_dup_inc(proto_perl->Idiehook, param);
15515 PL_warnhook = sv_dup_inc(proto_perl->Iwarnhook, param);
15518 PL_patchlevel = sv_dup_inc(proto_perl->Ipatchlevel, param);
15519 PL_inplace = SAVEPV(proto_perl->Iinplace);
15520 PL_e_script = sv_dup_inc(proto_perl->Ie_script, param);
15522 /* magical thingies */
15524 SvPVCLEAR(PERL_DEBUG_PAD(0)); /* For regex debugging. */
15525 SvPVCLEAR(PERL_DEBUG_PAD(1)); /* ext/re needs these */
15526 SvPVCLEAR(PERL_DEBUG_PAD(2)); /* even without DEBUGGING. */
15529 /* Clone the regex array */
15530 /* ORANGE FIXME for plugins, probably in the SV dup code.
15531 newSViv(PTR2IV(CALLREGDUPE(
15532 INT2PTR(REGEXP *, SvIVX(regex)), param))))
15534 PL_regex_padav = av_dup_inc(proto_perl->Iregex_padav, param);
15535 PL_regex_pad = AvARRAY(PL_regex_padav);
15537 PL_stashpadmax = proto_perl->Istashpadmax;
15538 PL_stashpadix = proto_perl->Istashpadix ;
15539 Newx(PL_stashpad, PL_stashpadmax, HV *);
15542 for (; o < PL_stashpadmax; ++o)
15543 PL_stashpad[o] = hv_dup(proto_perl->Istashpad[o], param);
15546 /* shortcuts to various I/O objects */
15547 PL_ofsgv = gv_dup_inc(proto_perl->Iofsgv, param);
15548 PL_stdingv = gv_dup(proto_perl->Istdingv, param);
15549 PL_stderrgv = gv_dup(proto_perl->Istderrgv, param);
15550 PL_defgv = gv_dup(proto_perl->Idefgv, param);
15551 PL_argvgv = gv_dup_inc(proto_perl->Iargvgv, param);
15552 PL_argvoutgv = gv_dup(proto_perl->Iargvoutgv, param);
15553 PL_argvout_stack = av_dup_inc(proto_perl->Iargvout_stack, param);
15555 /* shortcuts to regexp stuff */
15556 PL_replgv = gv_dup_inc(proto_perl->Ireplgv, param);
15558 /* shortcuts to misc objects */
15559 PL_errgv = gv_dup(proto_perl->Ierrgv, param);
15561 /* shortcuts to debugging objects */
15562 PL_DBgv = gv_dup_inc(proto_perl->IDBgv, param);
15563 PL_DBline = gv_dup_inc(proto_perl->IDBline, param);
15564 PL_DBsub = gv_dup_inc(proto_perl->IDBsub, param);
15565 PL_DBsingle = sv_dup(proto_perl->IDBsingle, param);
15566 PL_DBtrace = sv_dup(proto_perl->IDBtrace, param);
15567 PL_DBsignal = sv_dup(proto_perl->IDBsignal, param);
15568 Copy(proto_perl->IDBcontrol, PL_DBcontrol, DBVARMG_COUNT, IV);
15570 /* symbol tables */
15571 PL_defstash = hv_dup_inc(proto_perl->Idefstash, param);
15572 PL_curstash = hv_dup_inc(proto_perl->Icurstash, param);
15573 PL_debstash = hv_dup(proto_perl->Idebstash, param);
15574 PL_globalstash = hv_dup(proto_perl->Iglobalstash, param);
15575 PL_curstname = sv_dup_inc(proto_perl->Icurstname, param);
15577 PL_beginav = av_dup_inc(proto_perl->Ibeginav, param);
15578 PL_beginav_save = av_dup_inc(proto_perl->Ibeginav_save, param);
15579 PL_checkav_save = av_dup_inc(proto_perl->Icheckav_save, param);
15580 PL_unitcheckav = av_dup_inc(proto_perl->Iunitcheckav, param);
15581 PL_unitcheckav_save = av_dup_inc(proto_perl->Iunitcheckav_save, param);
15582 PL_endav = av_dup_inc(proto_perl->Iendav, param);
15583 PL_checkav = av_dup_inc(proto_perl->Icheckav, param);
15584 PL_initav = av_dup_inc(proto_perl->Iinitav, param);
15585 PL_savebegin = proto_perl->Isavebegin;
15587 PL_isarev = hv_dup_inc(proto_perl->Iisarev, param);
15589 /* subprocess state */
15590 PL_fdpid = av_dup_inc(proto_perl->Ifdpid, param);
15592 if (proto_perl->Iop_mask)
15593 PL_op_mask = SAVEPVN(proto_perl->Iop_mask, PL_maxo);
15596 /* PL_asserting = proto_perl->Iasserting; */
15598 /* current interpreter roots */
15599 PL_main_cv = cv_dup_inc(proto_perl->Imain_cv, param);
15601 PL_main_root = OpREFCNT_inc(proto_perl->Imain_root);
15604 /* runtime control stuff */
15605 PL_curcopdb = (COP*)any_dup(proto_perl->Icurcopdb, proto_perl);
15607 PL_preambleav = av_dup_inc(proto_perl->Ipreambleav, param);
15609 PL_ors_sv = sv_dup_inc(proto_perl->Iors_sv, param);
15611 /* interpreter atexit processing */
15612 PL_exitlistlen = proto_perl->Iexitlistlen;
15613 if (PL_exitlistlen) {
15614 Newx(PL_exitlist, PL_exitlistlen, PerlExitListEntry);
15615 Copy(proto_perl->Iexitlist, PL_exitlist, PL_exitlistlen, PerlExitListEntry);
15618 PL_exitlist = (PerlExitListEntry*)NULL;
15620 PL_my_cxt_size = proto_perl->Imy_cxt_size;
15621 if (PL_my_cxt_size) {
15622 Newx(PL_my_cxt_list, PL_my_cxt_size, void *);
15623 Copy(proto_perl->Imy_cxt_list, PL_my_cxt_list, PL_my_cxt_size, void *);
15626 PL_my_cxt_list = (void**)NULL;
15628 PL_modglobal = hv_dup_inc(proto_perl->Imodglobal, param);
15629 PL_custom_op_names = hv_dup_inc(proto_perl->Icustom_op_names,param);
15630 PL_custom_op_descs = hv_dup_inc(proto_perl->Icustom_op_descs,param);
15631 PL_custom_ops = hv_dup_inc(proto_perl->Icustom_ops, param);
15633 PL_compcv = cv_dup(proto_perl->Icompcv, param);
15635 PAD_CLONE_VARS(proto_perl, param);
15637 #ifdef HAVE_INTERP_INTERN
15638 sys_intern_dup(&proto_perl->Isys_intern, &PL_sys_intern);
15641 PL_DBcv = cv_dup(proto_perl->IDBcv, param);
15643 #ifdef PERL_USES_PL_PIDSTATUS
15644 PL_pidstatus = newHV(); /* XXX flag for cloning? */
15646 PL_osname = SAVEPV(proto_perl->Iosname);
15647 PL_parser = parser_dup(proto_perl->Iparser, param);
15649 /* XXX this only works if the saved cop has already been cloned */
15650 if (proto_perl->Iparser) {
15651 PL_parser->saved_curcop = (COP*)any_dup(
15652 proto_perl->Iparser->saved_curcop,
15656 PL_subname = sv_dup_inc(proto_perl->Isubname, param);
15658 #if defined(USE_POSIX_2008_LOCALE) \
15659 && defined(USE_THREAD_SAFE_LOCALE) \
15660 && ! defined(HAS_QUERYLOCALE)
15661 for (i = 0; i < (int) C_ARRAY_LENGTH(PL_curlocales); i++) {
15662 PL_curlocales[i] = savepv("."); /* An illegal value */
15665 #ifdef USE_LOCALE_CTYPE
15666 /* Should we warn if uses locale? */
15667 PL_warn_locale = sv_dup_inc(proto_perl->Iwarn_locale, param);
15670 #ifdef USE_LOCALE_COLLATE
15671 PL_collation_name = SAVEPV(proto_perl->Icollation_name);
15672 #endif /* USE_LOCALE_COLLATE */
15674 #ifdef USE_LOCALE_NUMERIC
15675 PL_numeric_name = SAVEPV(proto_perl->Inumeric_name);
15676 PL_numeric_radix_sv = sv_dup_inc(proto_perl->Inumeric_radix_sv, param);
15678 # if defined(HAS_POSIX_2008_LOCALE)
15679 PL_underlying_numeric_obj = NULL;
15681 #endif /* !USE_LOCALE_NUMERIC */
15683 PL_langinfo_buf = NULL;
15684 PL_langinfo_bufsize = 0;
15686 PL_setlocale_buf = NULL;
15687 PL_setlocale_bufsize = 0;
15689 /* utf8 character class swashes */
15690 PL_seen_deprecated_macro = hv_dup_inc(proto_perl->Iseen_deprecated_macro, param);
15692 if (proto_perl->Ipsig_pend) {
15693 Newxz(PL_psig_pend, SIG_SIZE, int);
15696 PL_psig_pend = (int*)NULL;
15699 if (proto_perl->Ipsig_name) {
15700 Newx(PL_psig_name, 2 * SIG_SIZE, SV*);
15701 sv_dup_inc_multiple(proto_perl->Ipsig_name, PL_psig_name, 2 * SIG_SIZE,
15703 PL_psig_ptr = PL_psig_name + SIG_SIZE;
15706 PL_psig_ptr = (SV**)NULL;
15707 PL_psig_name = (SV**)NULL;
15710 if (flags & CLONEf_COPY_STACKS) {
15711 Newx(PL_tmps_stack, PL_tmps_max, SV*);
15712 sv_dup_inc_multiple(proto_perl->Itmps_stack, PL_tmps_stack,
15713 PL_tmps_ix+1, param);
15715 /* next PUSHMARK() sets *(PL_markstack_ptr+1) */
15716 i = proto_perl->Imarkstack_max - proto_perl->Imarkstack;
15717 Newx(PL_markstack, i, I32);
15718 PL_markstack_max = PL_markstack + (proto_perl->Imarkstack_max
15719 - proto_perl->Imarkstack);
15720 PL_markstack_ptr = PL_markstack + (proto_perl->Imarkstack_ptr
15721 - proto_perl->Imarkstack);
15722 Copy(proto_perl->Imarkstack, PL_markstack,
15723 PL_markstack_ptr - PL_markstack + 1, I32);
15725 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
15726 * NOTE: unlike the others! */
15727 Newx(PL_scopestack, PL_scopestack_max, I32);
15728 Copy(proto_perl->Iscopestack, PL_scopestack, PL_scopestack_ix, I32);
15731 Newx(PL_scopestack_name, PL_scopestack_max, const char *);
15732 Copy(proto_perl->Iscopestack_name, PL_scopestack_name, PL_scopestack_ix, const char *);
15734 /* reset stack AV to correct length before its duped via
15735 * PL_curstackinfo */
15736 AvFILLp(proto_perl->Icurstack) =
15737 proto_perl->Istack_sp - proto_perl->Istack_base;
15739 /* NOTE: si_dup() looks at PL_markstack */
15740 PL_curstackinfo = si_dup(proto_perl->Icurstackinfo, param);
15742 /* PL_curstack = PL_curstackinfo->si_stack; */
15743 PL_curstack = av_dup(proto_perl->Icurstack, param);
15744 PL_mainstack = av_dup(proto_perl->Imainstack, param);
15746 /* next PUSHs() etc. set *(PL_stack_sp+1) */
15747 PL_stack_base = AvARRAY(PL_curstack);
15748 PL_stack_sp = PL_stack_base + (proto_perl->Istack_sp
15749 - proto_perl->Istack_base);
15750 PL_stack_max = PL_stack_base + AvMAX(PL_curstack);
15752 /*Newxz(PL_savestack, PL_savestack_max, ANY);*/
15753 PL_savestack = ss_dup(proto_perl, param);
15757 ENTER; /* perl_destruct() wants to LEAVE; */
15760 PL_statgv = gv_dup(proto_perl->Istatgv, param);
15761 PL_statname = sv_dup_inc(proto_perl->Istatname, param);
15763 PL_rs = sv_dup_inc(proto_perl->Irs, param);
15764 PL_last_in_gv = gv_dup(proto_perl->Ilast_in_gv, param);
15765 PL_defoutgv = gv_dup_inc(proto_perl->Idefoutgv, param);
15766 PL_toptarget = sv_dup_inc(proto_perl->Itoptarget, param);
15767 PL_bodytarget = sv_dup_inc(proto_perl->Ibodytarget, param);
15768 PL_formtarget = sv_dup(proto_perl->Iformtarget, param);
15770 PL_errors = sv_dup_inc(proto_perl->Ierrors, param);
15772 PL_sortcop = (OP*)any_dup(proto_perl->Isortcop, proto_perl);
15773 PL_firstgv = gv_dup_inc(proto_perl->Ifirstgv, param);
15774 PL_secondgv = gv_dup_inc(proto_perl->Isecondgv, param);
15776 PL_stashcache = newHV();
15778 PL_watchaddr = (char **) ptr_table_fetch(PL_ptr_table,
15779 proto_perl->Iwatchaddr);
15780 PL_watchok = PL_watchaddr ? * PL_watchaddr : NULL;
15781 if (PL_debug && PL_watchaddr) {
15782 PerlIO_printf(Perl_debug_log,
15783 "WATCHING: %" UVxf " cloned as %" UVxf " with value %" UVxf "\n",
15784 PTR2UV(proto_perl->Iwatchaddr), PTR2UV(PL_watchaddr),
15785 PTR2UV(PL_watchok));
15788 PL_registered_mros = hv_dup_inc(proto_perl->Iregistered_mros, param);
15789 PL_blockhooks = av_dup_inc(proto_perl->Iblockhooks, param);
15791 /* Call the ->CLONE method, if it exists, for each of the stashes
15792 identified by sv_dup() above.
15794 while(av_tindex(param->stashes) != -1) {
15795 HV* const stash = MUTABLE_HV(av_shift(param->stashes));
15796 GV* const cloner = gv_fetchmethod_autoload(stash, "CLONE", 0);
15797 if (cloner && GvCV(cloner)) {
15802 mXPUSHs(newSVhek(HvNAME_HEK(stash)));
15804 call_sv(MUTABLE_SV(GvCV(cloner)), G_DISCARD);
15810 if (!(flags & CLONEf_KEEP_PTR_TABLE)) {
15811 ptr_table_free(PL_ptr_table);
15812 PL_ptr_table = NULL;
15815 if (!(flags & CLONEf_COPY_STACKS)) {
15816 unreferenced_to_tmp_stack(param->unreferenced);
15819 SvREFCNT_dec(param->stashes);
15821 /* orphaned? eg threads->new inside BEGIN or use */
15822 if (PL_compcv && ! SvREFCNT(PL_compcv)) {
15823 SvREFCNT_inc_simple_void(PL_compcv);
15824 SAVEFREESV(PL_compcv);
15831 S_unreferenced_to_tmp_stack(pTHX_ AV *const unreferenced)
15833 PERL_ARGS_ASSERT_UNREFERENCED_TO_TMP_STACK;
15835 if (AvFILLp(unreferenced) > -1) {
15836 SV **svp = AvARRAY(unreferenced);
15837 SV **const last = svp + AvFILLp(unreferenced);
15841 if (SvREFCNT(*svp) == 1)
15843 } while (++svp <= last);
15845 EXTEND_MORTAL(count);
15846 svp = AvARRAY(unreferenced);
15849 if (SvREFCNT(*svp) == 1) {
15850 /* Our reference is the only one to this SV. This means that
15851 in this thread, the scalar effectively has a 0 reference.
15852 That doesn't work (cleanup never happens), so donate our
15853 reference to it onto the save stack. */
15854 PL_tmps_stack[++PL_tmps_ix] = *svp;
15856 /* As an optimisation, because we are already walking the
15857 entire array, instead of above doing either
15858 SvREFCNT_inc(*svp) or *svp = &PL_sv_undef, we can instead
15859 release our reference to the scalar, so that at the end of
15860 the array owns zero references to the scalars it happens to
15861 point to. We are effectively converting the array from
15862 AvREAL() on to AvREAL() off. This saves the av_clear()
15863 (triggered by the SvREFCNT_dec(unreferenced) below) from
15864 walking the array a second time. */
15865 SvREFCNT_dec(*svp);
15868 } while (++svp <= last);
15869 AvREAL_off(unreferenced);
15871 SvREFCNT_dec_NN(unreferenced);
15875 Perl_clone_params_del(CLONE_PARAMS *param)
15877 /* This seemingly funky ordering keeps the build with PERL_GLOBAL_STRUCT
15879 PerlInterpreter *const to = param->new_perl;
15881 PerlInterpreter *const was = PERL_GET_THX;
15883 PERL_ARGS_ASSERT_CLONE_PARAMS_DEL;
15889 SvREFCNT_dec(param->stashes);
15890 if (param->unreferenced)
15891 unreferenced_to_tmp_stack(param->unreferenced);
15901 Perl_clone_params_new(PerlInterpreter *const from, PerlInterpreter *const to)
15904 /* Need to play this game, as newAV() can call safesysmalloc(), and that
15905 does a dTHX; to get the context from thread local storage.
15906 FIXME - under PERL_CORE Newx(), Safefree() and friends should expand to
15907 a version that passes in my_perl. */
15908 PerlInterpreter *const was = PERL_GET_THX;
15909 CLONE_PARAMS *param;
15911 PERL_ARGS_ASSERT_CLONE_PARAMS_NEW;
15917 /* Given that we've set the context, we can do this unshared. */
15918 Newx(param, 1, CLONE_PARAMS);
15921 param->proto_perl = from;
15922 param->new_perl = to;
15923 param->stashes = (AV *)Perl_newSV_type(to, SVt_PVAV);
15924 AvREAL_off(param->stashes);
15925 param->unreferenced = (AV *)Perl_newSV_type(to, SVt_PVAV);
15933 #endif /* USE_ITHREADS */
15936 Perl_init_constants(pTHX)
15940 SvREFCNT(&PL_sv_undef) = SvREFCNT_IMMORTAL;
15941 SvFLAGS(&PL_sv_undef) = SVf_READONLY|SVf_PROTECT|SVt_NULL;
15942 SvANY(&PL_sv_undef) = NULL;
15944 SvANY(&PL_sv_no) = new_XPVNV();
15945 SvREFCNT(&PL_sv_no) = SvREFCNT_IMMORTAL;
15946 SvFLAGS(&PL_sv_no) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15947 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15950 SvANY(&PL_sv_yes) = new_XPVNV();
15951 SvREFCNT(&PL_sv_yes) = SvREFCNT_IMMORTAL;
15952 SvFLAGS(&PL_sv_yes) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15953 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15956 SvANY(&PL_sv_zero) = new_XPVNV();
15957 SvREFCNT(&PL_sv_zero) = SvREFCNT_IMMORTAL;
15958 SvFLAGS(&PL_sv_zero) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15959 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15963 SvPV_set(&PL_sv_no, (char*)PL_No);
15964 SvCUR_set(&PL_sv_no, 0);
15965 SvLEN_set(&PL_sv_no, 0);
15966 SvIV_set(&PL_sv_no, 0);
15967 SvNV_set(&PL_sv_no, 0);
15969 SvPV_set(&PL_sv_yes, (char*)PL_Yes);
15970 SvCUR_set(&PL_sv_yes, 1);
15971 SvLEN_set(&PL_sv_yes, 0);
15972 SvIV_set(&PL_sv_yes, 1);
15973 SvNV_set(&PL_sv_yes, 1);
15975 SvPV_set(&PL_sv_zero, (char*)PL_Zero);
15976 SvCUR_set(&PL_sv_zero, 1);
15977 SvLEN_set(&PL_sv_zero, 0);
15978 SvIV_set(&PL_sv_zero, 0);
15979 SvNV_set(&PL_sv_zero, 0);
15981 PadnamePV(&PL_padname_const) = (char *)PL_No;
15983 assert(SvIMMORTAL_INTERP(&PL_sv_yes));
15984 assert(SvIMMORTAL_INTERP(&PL_sv_undef));
15985 assert(SvIMMORTAL_INTERP(&PL_sv_no));
15986 assert(SvIMMORTAL_INTERP(&PL_sv_zero));
15988 assert(SvIMMORTAL(&PL_sv_yes));
15989 assert(SvIMMORTAL(&PL_sv_undef));
15990 assert(SvIMMORTAL(&PL_sv_no));
15991 assert(SvIMMORTAL(&PL_sv_zero));
15993 assert( SvIMMORTAL_TRUE(&PL_sv_yes));
15994 assert(!SvIMMORTAL_TRUE(&PL_sv_undef));
15995 assert(!SvIMMORTAL_TRUE(&PL_sv_no));
15996 assert(!SvIMMORTAL_TRUE(&PL_sv_zero));
15998 assert( SvTRUE_nomg_NN(&PL_sv_yes));
15999 assert(!SvTRUE_nomg_NN(&PL_sv_undef));
16000 assert(!SvTRUE_nomg_NN(&PL_sv_no));
16001 assert(!SvTRUE_nomg_NN(&PL_sv_zero));
16005 =head1 Unicode Support
16007 =for apidoc sv_recode_to_utf8
16009 C<encoding> is assumed to be an C<Encode> object, on entry the PV
16010 of C<sv> is assumed to be octets in that encoding, and C<sv>
16011 will be converted into Unicode (and UTF-8).
16013 If C<sv> already is UTF-8 (or if it is not C<POK>), or if C<encoding>
16014 is not a reference, nothing is done to C<sv>. If C<encoding> is not
16015 an C<Encode::XS> Encoding object, bad things will happen.
16016 (See F<cpan/Encode/encoding.pm> and L<Encode>.)
16018 The PV of C<sv> is returned.
16023 Perl_sv_recode_to_utf8(pTHX_ SV *sv, SV *encoding)
16025 PERL_ARGS_ASSERT_SV_RECODE_TO_UTF8;
16027 if (SvPOK(sv) && !SvUTF8(sv) && !IN_BYTES && SvROK(encoding)) {
16036 if (SvPADTMP(nsv)) {
16037 nsv = sv_newmortal();
16038 SvSetSV_nosteal(nsv, sv);
16047 Passing sv_yes is wrong - it needs to be or'ed set of constants
16048 for Encode::XS, while UTf-8 decode (currently) assumes a true value means
16049 remove converted chars from source.
16051 Both will default the value - let them.
16053 XPUSHs(&PL_sv_yes);
16056 call_method("decode", G_SCALAR);
16060 s = SvPV_const(uni, len);
16061 if (s != SvPVX_const(sv)) {
16062 SvGROW(sv, len + 1);
16063 Move(s, SvPVX(sv), len + 1, char);
16064 SvCUR_set(sv, len);
16069 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
16070 /* clear pos and any utf8 cache */
16071 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
16074 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
16075 magic_setutf8(sv,mg); /* clear UTF8 cache */
16080 return SvPOKp(sv) ? SvPVX(sv) : NULL;
16084 =for apidoc sv_cat_decode
16086 C<encoding> is assumed to be an C<Encode> object, the PV of C<ssv> is
16087 assumed to be octets in that encoding and decoding the input starts
16088 from the position which S<C<(PV + *offset)>> pointed to. C<dsv> will be
16089 concatenated with the decoded UTF-8 string from C<ssv>. Decoding will terminate
16090 when the string C<tstr> appears in decoding output or the input ends on
16091 the PV of C<ssv>. The value which C<offset> points will be modified
16092 to the last input position on C<ssv>.
16094 Returns TRUE if the terminator was found, else returns FALSE.
16099 Perl_sv_cat_decode(pTHX_ SV *dsv, SV *encoding,
16100 SV *ssv, int *offset, char *tstr, int tlen)
16104 PERL_ARGS_ASSERT_SV_CAT_DECODE;
16106 if (SvPOK(ssv) && SvPOK(dsv) && SvROK(encoding)) {
16117 offsv = newSViv(*offset);
16119 mPUSHp(tstr, tlen);
16121 call_method("cat_decode", G_SCALAR);
16123 ret = SvTRUE(TOPs);
16124 *offset = SvIV(offsv);
16130 Perl_croak(aTHX_ "Invalid argument to sv_cat_decode");
16135 /* ---------------------------------------------------------------------
16137 * support functions for report_uninit()
16140 /* the maxiumum size of array or hash where we will scan looking
16141 * for the undefined element that triggered the warning */
16143 #define FUV_MAX_SEARCH_SIZE 1000
16145 /* Look for an entry in the hash whose value has the same SV as val;
16146 * If so, return a mortal copy of the key. */
16149 S_find_hash_subscript(pTHX_ const HV *const hv, const SV *const val)
16155 PERL_ARGS_ASSERT_FIND_HASH_SUBSCRIPT;
16157 if (!hv || SvMAGICAL(hv) || !HvARRAY(hv) ||
16158 (HvTOTALKEYS(hv) > FUV_MAX_SEARCH_SIZE))
16161 array = HvARRAY(hv);
16163 for (i=HvMAX(hv); i>=0; i--) {
16165 for (entry = array[i]; entry; entry = HeNEXT(entry)) {
16166 if (HeVAL(entry) != val)
16168 if ( HeVAL(entry) == &PL_sv_undef ||
16169 HeVAL(entry) == &PL_sv_placeholder)
16173 if (HeKLEN(entry) == HEf_SVKEY)
16174 return sv_mortalcopy(HeKEY_sv(entry));
16175 return sv_2mortal(newSVhek(HeKEY_hek(entry)));
16181 /* Look for an entry in the array whose value has the same SV as val;
16182 * If so, return the index, otherwise return -1. */
16185 S_find_array_subscript(pTHX_ const AV *const av, const SV *const val)
16187 PERL_ARGS_ASSERT_FIND_ARRAY_SUBSCRIPT;
16189 if (!av || SvMAGICAL(av) || !AvARRAY(av) ||
16190 (AvFILLp(av) > FUV_MAX_SEARCH_SIZE))
16193 if (val != &PL_sv_undef) {
16194 SV ** const svp = AvARRAY(av);
16197 for (i=AvFILLp(av); i>=0; i--)
16204 /* varname(): return the name of a variable, optionally with a subscript.
16205 * If gv is non-zero, use the name of that global, along with gvtype (one
16206 * of "$", "@", "%"); otherwise use the name of the lexical at pad offset
16207 * targ. Depending on the value of the subscript_type flag, return:
16210 #define FUV_SUBSCRIPT_NONE 1 /* "@foo" */
16211 #define FUV_SUBSCRIPT_ARRAY 2 /* "$foo[aindex]" */
16212 #define FUV_SUBSCRIPT_HASH 3 /* "$foo{keyname}" */
16213 #define FUV_SUBSCRIPT_WITHIN 4 /* "within @foo" */
16216 Perl_varname(pTHX_ const GV *const gv, const char gvtype, PADOFFSET targ,
16217 const SV *const keyname, SSize_t aindex, int subscript_type)
16220 SV * const name = sv_newmortal();
16221 if (gv && isGV(gv)) {
16223 buffer[0] = gvtype;
16226 /* as gv_fullname4(), but add literal '^' for $^FOO names */
16228 gv_fullname4(name, gv, buffer, 0);
16230 if ((unsigned int)SvPVX(name)[1] <= 26) {
16232 buffer[1] = SvPVX(name)[1] + 'A' - 1;
16234 /* Swap the 1 unprintable control character for the 2 byte pretty
16235 version - ie substr($name, 1, 1) = $buffer; */
16236 sv_insert(name, 1, 1, buffer, 2);
16240 CV * const cv = gv ? ((CV *)gv) : find_runcv(NULL);
16243 assert(!cv || SvTYPE(cv) == SVt_PVCV || SvTYPE(cv) == SVt_PVFM);
16245 if (!cv || !CvPADLIST(cv))
16247 sv = padnamelist_fetch(PadlistNAMES(CvPADLIST(cv)), targ);
16248 sv_setpvn(name, PadnamePV(sv), PadnameLEN(sv));
16252 if (subscript_type == FUV_SUBSCRIPT_HASH) {
16253 SV * const sv = newSV(0);
16255 const char * const pv = SvPV_nomg_const((SV*)keyname, len);
16257 *SvPVX(name) = '$';
16258 Perl_sv_catpvf(aTHX_ name, "{%s}",
16259 pv_pretty(sv, pv, len, 32, NULL, NULL,
16260 PERL_PV_PRETTY_DUMP | PERL_PV_ESCAPE_UNI_DETECT ));
16261 SvREFCNT_dec_NN(sv);
16263 else if (subscript_type == FUV_SUBSCRIPT_ARRAY) {
16264 *SvPVX(name) = '$';
16265 Perl_sv_catpvf(aTHX_ name, "[%" IVdf "]", (IV)aindex);
16267 else if (subscript_type == FUV_SUBSCRIPT_WITHIN) {
16268 /* We know that name has no magic, so can use 0 instead of SV_GMAGIC */
16269 Perl_sv_insert_flags(aTHX_ name, 0, 0, STR_WITH_LEN("within "), 0);
16277 =for apidoc find_uninit_var
16279 Find the name of the undefined variable (if any) that caused the operator
16280 to issue a "Use of uninitialized value" warning.
16281 If match is true, only return a name if its value matches C<uninit_sv>.
16282 So roughly speaking, if a unary operator (such as C<OP_COS>) generates a
16283 warning, then following the direct child of the op may yield an
16284 C<OP_PADSV> or C<OP_GV> that gives the name of the undefined variable. On the
16285 other hand, with C<OP_ADD> there are two branches to follow, so we only print
16286 the variable name if we get an exact match.
16287 C<desc_p> points to a string pointer holding the description of the op.
16288 This may be updated if needed.
16290 The name is returned as a mortal SV.
16292 Assumes that C<PL_op> is the OP that originally triggered the error, and that
16293 C<PL_comppad>/C<PL_curpad> points to the currently executing pad.
16299 S_find_uninit_var(pTHX_ const OP *const obase, const SV *const uninit_sv,
16300 bool match, const char **desc_p)
16305 const OP *o, *o2, *kid;
16307 PERL_ARGS_ASSERT_FIND_UNINIT_VAR;
16309 if (!obase || (match && (!uninit_sv || uninit_sv == &PL_sv_undef ||
16310 uninit_sv == &PL_sv_placeholder)))
16313 switch (obase->op_type) {
16316 /* undef should care if its args are undef - any warnings
16317 * will be from tied/magic vars */
16325 const bool pad = ( obase->op_type == OP_PADAV
16326 || obase->op_type == OP_PADHV
16327 || obase->op_type == OP_PADRANGE
16330 const bool hash = ( obase->op_type == OP_PADHV
16331 || obase->op_type == OP_RV2HV
16332 || (obase->op_type == OP_PADRANGE
16333 && SvTYPE(PAD_SVl(obase->op_targ)) == SVt_PVHV)
16337 int subscript_type = FUV_SUBSCRIPT_WITHIN;
16339 if (pad) { /* @lex, %lex */
16340 sv = PAD_SVl(obase->op_targ);
16344 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
16345 /* @global, %global */
16346 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
16349 sv = hash ? MUTABLE_SV(GvHV(gv)): MUTABLE_SV(GvAV(gv));
16351 else if (obase == PL_op) /* @{expr}, %{expr} */
16352 return find_uninit_var(cUNOPx(obase)->op_first,
16353 uninit_sv, match, desc_p);
16354 else /* @{expr}, %{expr} as a sub-expression */
16358 /* attempt to find a match within the aggregate */
16360 keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16362 subscript_type = FUV_SUBSCRIPT_HASH;
16365 index = find_array_subscript((const AV *)sv, uninit_sv);
16367 subscript_type = FUV_SUBSCRIPT_ARRAY;
16370 if (match && subscript_type == FUV_SUBSCRIPT_WITHIN)
16373 return varname(gv, (char)(hash ? '%' : '@'), obase->op_targ,
16374 keysv, index, subscript_type);
16378 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
16380 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
16381 if (!gv || !GvSTASH(gv))
16383 if (match && (GvSV(gv) != uninit_sv))
16385 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
16388 return find_uninit_var(cUNOPx(obase)->op_first, uninit_sv, 1, desc_p);
16391 if (match && PAD_SVl(obase->op_targ) != uninit_sv)
16393 return varname(NULL, '$', obase->op_targ,
16394 NULL, 0, FUV_SUBSCRIPT_NONE);
16397 gv = cGVOPx_gv(obase);
16398 if (!gv || (match && GvSV(gv) != uninit_sv) || !GvSTASH(gv))
16400 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
16402 case OP_AELEMFAST_LEX:
16405 AV *av = MUTABLE_AV(PAD_SV(obase->op_targ));
16406 if (!av || SvRMAGICAL(av))
16408 svp = av_fetch(av, (I8)obase->op_private, FALSE);
16409 if (!svp || *svp != uninit_sv)
16412 return varname(NULL, '$', obase->op_targ,
16413 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
16416 gv = cGVOPx_gv(obase);
16421 AV *const av = GvAV(gv);
16422 if (!av || SvRMAGICAL(av))
16424 svp = av_fetch(av, (I8)obase->op_private, FALSE);
16425 if (!svp || *svp != uninit_sv)
16428 return varname(gv, '$', 0,
16429 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
16431 NOT_REACHED; /* NOTREACHED */
16434 o = cUNOPx(obase)->op_first;
16435 if (!o || o->op_type != OP_NULL ||
16436 ! (o->op_targ == OP_AELEM || o->op_targ == OP_HELEM))
16438 return find_uninit_var(cBINOPo->op_last, uninit_sv, match, desc_p);
16443 bool negate = FALSE;
16445 if (PL_op == obase)
16446 /* $a[uninit_expr] or $h{uninit_expr} */
16447 return find_uninit_var(cBINOPx(obase)->op_last,
16448 uninit_sv, match, desc_p);
16451 o = cBINOPx(obase)->op_first;
16452 kid = cBINOPx(obase)->op_last;
16454 /* get the av or hv, and optionally the gv */
16456 if (o->op_type == OP_PADAV || o->op_type == OP_PADHV) {
16457 sv = PAD_SV(o->op_targ);
16459 else if ((o->op_type == OP_RV2AV || o->op_type == OP_RV2HV)
16460 && cUNOPo->op_first->op_type == OP_GV)
16462 gv = cGVOPx_gv(cUNOPo->op_first);
16466 == OP_RV2HV ? MUTABLE_SV(GvHV(gv)) : MUTABLE_SV(GvAV(gv));
16471 if (kid && kid->op_type == OP_NEGATE) {
16473 kid = cUNOPx(kid)->op_first;
16476 if (kid && kid->op_type == OP_CONST && SvOK(cSVOPx_sv(kid))) {
16477 /* index is constant */
16480 kidsv = newSVpvs_flags("-", SVs_TEMP);
16481 sv_catsv(kidsv, cSVOPx_sv(kid));
16484 kidsv = cSVOPx_sv(kid);
16488 if (obase->op_type == OP_HELEM) {
16489 HE* he = hv_fetch_ent(MUTABLE_HV(sv), kidsv, 0, 0);
16490 if (!he || HeVAL(he) != uninit_sv)
16494 SV * const opsv = cSVOPx_sv(kid);
16495 const IV opsviv = SvIV(opsv);
16496 SV * const * const svp = av_fetch(MUTABLE_AV(sv),
16497 negate ? - opsviv : opsviv,
16499 if (!svp || *svp != uninit_sv)
16503 if (obase->op_type == OP_HELEM)
16504 return varname(gv, '%', o->op_targ,
16505 kidsv, 0, FUV_SUBSCRIPT_HASH);
16507 return varname(gv, '@', o->op_targ, NULL,
16508 negate ? - SvIV(cSVOPx_sv(kid)) : SvIV(cSVOPx_sv(kid)),
16509 FUV_SUBSCRIPT_ARRAY);
16512 /* index is an expression;
16513 * attempt to find a match within the aggregate */
16514 if (obase->op_type == OP_HELEM) {
16515 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16517 return varname(gv, '%', o->op_targ,
16518 keysv, 0, FUV_SUBSCRIPT_HASH);
16521 const SSize_t index
16522 = find_array_subscript((const AV *)sv, uninit_sv);
16524 return varname(gv, '@', o->op_targ,
16525 NULL, index, FUV_SUBSCRIPT_ARRAY);
16530 (char)((o->op_type == OP_PADAV || o->op_type == OP_RV2AV)
16532 o->op_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
16534 NOT_REACHED; /* NOTREACHED */
16537 case OP_MULTIDEREF: {
16538 /* If we were executing OP_MULTIDEREF when the undef warning
16539 * triggered, then it must be one of the index values within
16540 * that triggered it. If not, then the only possibility is that
16541 * the value retrieved by the last aggregate index might be the
16542 * culprit. For the former, we set PL_multideref_pc each time before
16543 * using an index, so work though the item list until we reach
16544 * that point. For the latter, just work through the entire item
16545 * list; the last aggregate retrieved will be the candidate.
16546 * There is a third rare possibility: something triggered
16547 * magic while fetching an array/hash element. Just display
16548 * nothing in this case.
16551 /* the named aggregate, if any */
16552 PADOFFSET agg_targ = 0;
16554 /* the last-seen index */
16556 PADOFFSET index_targ;
16558 IV index_const_iv = 0; /* init for spurious compiler warn */
16559 SV *index_const_sv;
16560 int depth = 0; /* how many array/hash lookups we've done */
16562 UNOP_AUX_item *items = cUNOP_AUXx(obase)->op_aux;
16563 UNOP_AUX_item *last = NULL;
16564 UV actions = items->uv;
16567 if (PL_op == obase) {
16568 last = PL_multideref_pc;
16569 assert(last >= items && last <= items + items[-1].uv);
16576 switch (actions & MDEREF_ACTION_MASK) {
16578 case MDEREF_reload:
16579 actions = (++items)->uv;
16582 case MDEREF_HV_padhv_helem: /* $lex{...} */
16585 case MDEREF_AV_padav_aelem: /* $lex[...] */
16586 agg_targ = (++items)->pad_offset;
16590 case MDEREF_HV_gvhv_helem: /* $pkg{...} */
16593 case MDEREF_AV_gvav_aelem: /* $pkg[...] */
16595 agg_gv = (GV*)UNOP_AUX_item_sv(++items);
16596 assert(isGV_with_GP(agg_gv));
16599 case MDEREF_HV_gvsv_vivify_rv2hv_helem: /* $pkg->{...} */
16600 case MDEREF_HV_padsv_vivify_rv2hv_helem: /* $lex->{...} */
16603 case MDEREF_HV_pop_rv2hv_helem: /* expr->{...} */
16604 case MDEREF_HV_vivify_rv2hv_helem: /* vivify, ->{...} */
16610 case MDEREF_AV_gvsv_vivify_rv2av_aelem: /* $pkg->[...] */
16611 case MDEREF_AV_padsv_vivify_rv2av_aelem: /* $lex->[...] */
16614 case MDEREF_AV_pop_rv2av_aelem: /* expr->[...] */
16615 case MDEREF_AV_vivify_rv2av_aelem: /* vivify, ->[...] */
16622 index_const_sv = NULL;
16624 index_type = (actions & MDEREF_INDEX_MASK);
16625 switch (index_type) {
16626 case MDEREF_INDEX_none:
16628 case MDEREF_INDEX_const:
16630 index_const_sv = UNOP_AUX_item_sv(++items)
16632 index_const_iv = (++items)->iv;
16634 case MDEREF_INDEX_padsv:
16635 index_targ = (++items)->pad_offset;
16637 case MDEREF_INDEX_gvsv:
16638 index_gv = (GV*)UNOP_AUX_item_sv(++items);
16639 assert(isGV_with_GP(index_gv));
16643 if (index_type != MDEREF_INDEX_none)
16646 if ( index_type == MDEREF_INDEX_none
16647 || (actions & MDEREF_FLAG_last)
16648 || (last && items >= last)
16652 actions >>= MDEREF_SHIFT;
16655 if (PL_op == obase) {
16656 /* most likely index was undef */
16658 *desc_p = ( (actions & MDEREF_FLAG_last)
16659 && (obase->op_private
16660 & (OPpMULTIDEREF_EXISTS|OPpMULTIDEREF_DELETE)))
16662 (obase->op_private & OPpMULTIDEREF_EXISTS)
16665 : is_hv ? "hash element" : "array element";
16666 assert(index_type != MDEREF_INDEX_none);
16668 if (GvSV(index_gv) == uninit_sv)
16669 return varname(index_gv, '$', 0, NULL, 0,
16670 FUV_SUBSCRIPT_NONE);
16675 if (PL_curpad[index_targ] == uninit_sv)
16676 return varname(NULL, '$', index_targ,
16677 NULL, 0, FUV_SUBSCRIPT_NONE);
16681 /* If we got to this point it was undef on a const subscript,
16682 * so magic probably involved, e.g. $ISA[0]. Give up. */
16686 /* the SV returned by pp_multideref() was undef, if anything was */
16692 sv = PAD_SV(agg_targ);
16694 sv = is_hv ? MUTABLE_SV(GvHV(agg_gv)) : MUTABLE_SV(GvAV(agg_gv));
16701 if (index_type == MDEREF_INDEX_const) {
16706 HE* he = hv_fetch_ent(MUTABLE_HV(sv), index_const_sv, 0, 0);
16707 if (!he || HeVAL(he) != uninit_sv)
16711 SV * const * const svp =
16712 av_fetch(MUTABLE_AV(sv), index_const_iv, FALSE);
16713 if (!svp || *svp != uninit_sv)
16718 ? varname(agg_gv, '%', agg_targ,
16719 index_const_sv, 0, FUV_SUBSCRIPT_HASH)
16720 : varname(agg_gv, '@', agg_targ,
16721 NULL, index_const_iv, FUV_SUBSCRIPT_ARRAY);
16724 /* index is an var */
16726 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16728 return varname(agg_gv, '%', agg_targ,
16729 keysv, 0, FUV_SUBSCRIPT_HASH);
16732 const SSize_t index
16733 = find_array_subscript((const AV *)sv, uninit_sv);
16735 return varname(agg_gv, '@', agg_targ,
16736 NULL, index, FUV_SUBSCRIPT_ARRAY);
16740 return varname(agg_gv,
16742 agg_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
16744 NOT_REACHED; /* NOTREACHED */
16748 /* only examine RHS */
16749 return find_uninit_var(cBINOPx(obase)->op_first, uninit_sv,
16753 o = cUNOPx(obase)->op_first;
16754 if ( o->op_type == OP_PUSHMARK
16755 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)
16759 if (!OpHAS_SIBLING(o)) {
16760 /* one-arg version of open is highly magical */
16762 if (o->op_type == OP_GV) { /* open FOO; */
16764 if (match && GvSV(gv) != uninit_sv)
16766 return varname(gv, '$', 0,
16767 NULL, 0, FUV_SUBSCRIPT_NONE);
16769 /* other possibilities not handled are:
16770 * open $x; or open my $x; should return '${*$x}'
16771 * open expr; should return '$'.expr ideally
16778 /* ops where $_ may be an implicit arg */
16783 if ( !(obase->op_flags & OPf_STACKED)) {
16784 if (uninit_sv == DEFSV)
16785 return newSVpvs_flags("$_", SVs_TEMP);
16786 else if (obase->op_targ
16787 && uninit_sv == PAD_SVl(obase->op_targ))
16788 return varname(NULL, '$', obase->op_targ, NULL, 0,
16789 FUV_SUBSCRIPT_NONE);
16796 match = 1; /* print etc can return undef on defined args */
16797 /* skip filehandle as it can't produce 'undef' warning */
16798 o = cUNOPx(obase)->op_first;
16799 if ((obase->op_flags & OPf_STACKED)
16801 ( o->op_type == OP_PUSHMARK
16802 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)))
16803 o = OpSIBLING(OpSIBLING(o));
16807 case OP_ENTEREVAL: /* could be eval $undef or $x='$undef'; eval $x */
16808 case OP_CUSTOM: /* XS or custom code could trigger random warnings */
16810 /* the following ops are capable of returning PL_sv_undef even for
16811 * defined arg(s) */
16830 case OP_GETPEERNAME:
16877 case OP_SMARTMATCH:
16886 /* XXX tmp hack: these two may call an XS sub, and currently
16887 XS subs don't have a SUB entry on the context stack, so CV and
16888 pad determination goes wrong, and BAD things happen. So, just
16889 don't try to determine the value under those circumstances.
16890 Need a better fix at dome point. DAPM 11/2007 */
16896 GV * const gv = gv_fetchpvs(".", GV_NOTQUAL, SVt_PV);
16897 if (gv && GvSV(gv) == uninit_sv)
16898 return newSVpvs_flags("$.", SVs_TEMP);
16903 /* def-ness of rval pos() is independent of the def-ness of its arg */
16904 if ( !(obase->op_flags & OPf_MOD))
16910 if (SvROK(PL_rs) && uninit_sv == SvRV(PL_rs))
16911 return newSVpvs_flags("${$/}", SVs_TEMP);
16916 if (!(obase->op_flags & OPf_KIDS))
16918 o = cUNOPx(obase)->op_first;
16924 /* This loop checks all the kid ops, skipping any that cannot pos-
16925 * sibly be responsible for the uninitialized value; i.e., defined
16926 * constants and ops that return nothing. If there is only one op
16927 * left that is not skipped, then we *know* it is responsible for
16928 * the uninitialized value. If there is more than one op left, we
16929 * have to look for an exact match in the while() loop below.
16930 * Note that we skip padrange, because the individual pad ops that
16931 * it replaced are still in the tree, so we work on them instead.
16934 for (kid=o; kid; kid = OpSIBLING(kid)) {
16935 const OPCODE type = kid->op_type;
16936 if ( (type == OP_CONST && SvOK(cSVOPx_sv(kid)))
16937 || (type == OP_NULL && ! (kid->op_flags & OPf_KIDS))
16938 || (type == OP_PUSHMARK)
16939 || (type == OP_PADRANGE)
16943 if (o2) { /* more than one found */
16950 return find_uninit_var(o2, uninit_sv, match, desc_p);
16952 /* scan all args */
16954 sv = find_uninit_var(o, uninit_sv, 1, desc_p);
16966 =for apidoc report_uninit
16968 Print appropriate "Use of uninitialized variable" warning.
16974 Perl_report_uninit(pTHX_ const SV *uninit_sv)
16976 const char *desc = NULL;
16977 SV* varname = NULL;
16980 desc = PL_op->op_type == OP_STRINGIFY && PL_op->op_folded
16982 : PL_op->op_type == OP_MULTICONCAT
16983 && (PL_op->op_private & OPpMULTICONCAT_FAKE)
16986 if (uninit_sv && PL_curpad) {
16987 varname = find_uninit_var(PL_op, uninit_sv, 0, &desc);
16989 sv_insert(varname, 0, 0, " ", 1);
16992 else if (PL_curstackinfo->si_type == PERLSI_SORT && cxstack_ix == 0)
16993 /* we've reached the end of a sort block or sub,
16994 * and the uninit value is probably what that code returned */
16997 /* PL_warn_uninit_sv is constant */
16998 GCC_DIAG_IGNORE_STMT(-Wformat-nonliteral);
17000 /* diag_listed_as: Use of uninitialized value%s */
17001 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit_sv,
17002 SVfARG(varname ? varname : &PL_sv_no),
17005 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit,
17007 GCC_DIAG_RESTORE_STMT;
17011 * ex: set ts=8 sts=4 sw=4 et: