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 # pragma warning(push)
2091 # pragma warning(disable:4756;disable:4056)
2094 S_sv_setnv(pTHX_ SV* sv, int numtype)
2096 bool pok = cBOOL(SvPOK(sv));
2099 if ((numtype & IS_NUMBER_INFINITY)) {
2100 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -NV_INF : NV_INF);
2105 if ((numtype & IS_NUMBER_NAN)) {
2106 SvNV_set(sv, NV_NAN);
2111 SvNV_set(sv, Atof(SvPVX_const(sv)));
2112 /* Purposefully no true nok here, since we don't want to blow
2113 * away the possible IOK/UV of an existing sv. */
2116 SvNOK_only(sv); /* No IV or UV please, this is pure infnan. */
2118 SvPOK_on(sv); /* PV is okay, though. */
2122 # pragma warning(pop)
2126 S_sv_2iuv_common(pTHX_ SV *const sv)
2128 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
2131 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
2132 * without also getting a cached IV/UV from it at the same time
2133 * (ie PV->NV conversion should detect loss of accuracy and cache
2134 * IV or UV at same time to avoid this. */
2135 /* IV-over-UV optimisation - choose to cache IV if possible */
2137 if (SvTYPE(sv) == SVt_NV)
2138 sv_upgrade(sv, SVt_PVNV);
2140 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
2141 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
2142 certainly cast into the IV range at IV_MAX, whereas the correct
2143 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
2145 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2146 if (Perl_isnan(SvNVX(sv))) {
2152 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2153 SvIV_set(sv, I_V(SvNVX(sv)));
2154 if (SvNVX(sv) == (NV) SvIVX(sv)
2155 #ifndef NV_PRESERVES_UV
2156 && SvIVX(sv) != IV_MIN /* avoid negating IV_MIN below */
2157 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2158 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2159 /* Don't flag it as "accurately an integer" if the number
2160 came from a (by definition imprecise) NV operation, and
2161 we're outside the range of NV integer precision */
2165 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2167 /* scalar has trailing garbage, eg "42a" */
2169 DEBUG_c(PerlIO_printf(Perl_debug_log,
2170 "0x%" UVxf " iv(%" NVgf " => %" IVdf ") (precise)\n",
2176 /* IV not precise. No need to convert from PV, as NV
2177 conversion would already have cached IV if it detected
2178 that PV->IV would be better than PV->NV->IV
2179 flags already correct - don't set public IOK. */
2180 DEBUG_c(PerlIO_printf(Perl_debug_log,
2181 "0x%" UVxf " iv(%" NVgf " => %" IVdf ") (imprecise)\n",
2186 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2187 but the cast (NV)IV_MIN rounds to a the value less (more
2188 negative) than IV_MIN which happens to be equal to SvNVX ??
2189 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2190 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2191 (NV)UVX == NVX are both true, but the values differ. :-(
2192 Hopefully for 2s complement IV_MIN is something like
2193 0x8000000000000000 which will be exact. NWC */
2196 SvUV_set(sv, U_V(SvNVX(sv)));
2198 (SvNVX(sv) == (NV) SvUVX(sv))
2199 #ifndef NV_PRESERVES_UV
2200 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2201 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2202 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2203 /* Don't flag it as "accurately an integer" if the number
2204 came from a (by definition imprecise) NV operation, and
2205 we're outside the range of NV integer precision */
2211 DEBUG_c(PerlIO_printf(Perl_debug_log,
2212 "0x%" UVxf " 2iv(%" UVuf " => %" IVdf ") (as unsigned)\n",
2218 else if (SvPOKp(sv)) {
2221 const char *s = SvPVX_const(sv);
2222 const STRLEN cur = SvCUR(sv);
2224 /* short-cut for a single digit string like "1" */
2229 if (SvTYPE(sv) < SVt_PVIV)
2230 sv_upgrade(sv, SVt_PVIV);
2232 SvIV_set(sv, (IV)(c - '0'));
2237 numtype = grok_number(s, cur, &value);
2238 /* We want to avoid a possible problem when we cache an IV/ a UV which
2239 may be later translated to an NV, and the resulting NV is not
2240 the same as the direct translation of the initial string
2241 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2242 be careful to ensure that the value with the .456 is around if the
2243 NV value is requested in the future).
2245 This means that if we cache such an IV/a UV, we need to cache the
2246 NV as well. Moreover, we trade speed for space, and do not
2247 cache the NV if we are sure it's not needed.
2250 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2251 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2252 == IS_NUMBER_IN_UV) {
2253 /* It's definitely an integer, only upgrade to PVIV */
2254 if (SvTYPE(sv) < SVt_PVIV)
2255 sv_upgrade(sv, SVt_PVIV);
2257 } else if (SvTYPE(sv) < SVt_PVNV)
2258 sv_upgrade(sv, SVt_PVNV);
2260 if ((numtype & (IS_NUMBER_INFINITY | IS_NUMBER_NAN))) {
2261 if (ckWARN(WARN_NUMERIC) && ((numtype & IS_NUMBER_TRAILING)))
2263 S_sv_setnv(aTHX_ sv, numtype);
2267 /* If NVs preserve UVs then we only use the UV value if we know that
2268 we aren't going to call atof() below. If NVs don't preserve UVs
2269 then the value returned may have more precision than atof() will
2270 return, even though value isn't perfectly accurate. */
2271 if ((numtype & (IS_NUMBER_IN_UV
2272 #ifdef NV_PRESERVES_UV
2275 )) == IS_NUMBER_IN_UV) {
2276 /* This won't turn off the public IOK flag if it was set above */
2277 (void)SvIOKp_on(sv);
2279 if (!(numtype & IS_NUMBER_NEG)) {
2281 if (value <= (UV)IV_MAX) {
2282 SvIV_set(sv, (IV)value);
2284 /* it didn't overflow, and it was positive. */
2285 SvUV_set(sv, value);
2289 /* 2s complement assumption */
2290 if (value <= (UV)IV_MIN) {
2291 SvIV_set(sv, value == (UV)IV_MIN
2292 ? IV_MIN : -(IV)value);
2294 /* Too negative for an IV. This is a double upgrade, but
2295 I'm assuming it will be rare. */
2296 if (SvTYPE(sv) < SVt_PVNV)
2297 sv_upgrade(sv, SVt_PVNV);
2301 SvNV_set(sv, -(NV)value);
2302 SvIV_set(sv, IV_MIN);
2306 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2307 will be in the previous block to set the IV slot, and the next
2308 block to set the NV slot. So no else here. */
2310 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2311 != IS_NUMBER_IN_UV) {
2312 /* It wasn't an (integer that doesn't overflow the UV). */
2313 S_sv_setnv(aTHX_ sv, numtype);
2315 if (! numtype && ckWARN(WARN_NUMERIC))
2318 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2iv(%" NVgf ")\n",
2319 PTR2UV(sv), SvNVX(sv)));
2321 #ifdef NV_PRESERVES_UV
2322 (void)SvIOKp_on(sv);
2324 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2325 if (Perl_isnan(SvNVX(sv))) {
2331 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2332 SvIV_set(sv, I_V(SvNVX(sv)));
2333 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2336 NOOP; /* Integer is imprecise. NOK, IOKp */
2338 /* UV will not work better than IV */
2340 if (SvNVX(sv) > (NV)UV_MAX) {
2342 /* Integer is inaccurate. NOK, IOKp, is UV */
2343 SvUV_set(sv, UV_MAX);
2345 SvUV_set(sv, U_V(SvNVX(sv)));
2346 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2347 NV preservse UV so can do correct comparison. */
2348 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2351 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
2356 #else /* NV_PRESERVES_UV */
2357 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2358 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
2359 /* The IV/UV slot will have been set from value returned by
2360 grok_number above. The NV slot has just been set using
2363 assert (SvIOKp(sv));
2365 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2366 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2367 /* Small enough to preserve all bits. */
2368 (void)SvIOKp_on(sv);
2370 SvIV_set(sv, I_V(SvNVX(sv)));
2371 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2373 /* Assumption: first non-preserved integer is < IV_MAX,
2374 this NV is in the preserved range, therefore: */
2375 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2377 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);
2381 0 0 already failed to read UV.
2382 0 1 already failed to read UV.
2383 1 0 you won't get here in this case. IV/UV
2384 slot set, public IOK, Atof() unneeded.
2385 1 1 already read UV.
2386 so there's no point in sv_2iuv_non_preserve() attempting
2387 to use atol, strtol, strtoul etc. */
2389 sv_2iuv_non_preserve (sv, numtype);
2391 sv_2iuv_non_preserve (sv);
2395 #endif /* NV_PRESERVES_UV */
2396 /* It might be more code efficient to go through the entire logic above
2397 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2398 gets complex and potentially buggy, so more programmer efficient
2399 to do it this way, by turning off the public flags: */
2401 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2405 if (isGV_with_GP(sv))
2406 return glob_2number(MUTABLE_GV(sv));
2408 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2410 if (SvTYPE(sv) < SVt_IV)
2411 /* Typically the caller expects that sv_any is not NULL now. */
2412 sv_upgrade(sv, SVt_IV);
2413 /* Return 0 from the caller. */
2420 =for apidoc sv_2iv_flags
2422 Return the integer value of an SV, doing any necessary string
2423 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2424 Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2430 Perl_sv_2iv_flags(pTHX_ SV *const sv, const I32 flags)
2432 PERL_ARGS_ASSERT_SV_2IV_FLAGS;
2434 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2435 && SvTYPE(sv) != SVt_PVFM);
2437 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2443 if (flags & SV_SKIP_OVERLOAD)
2445 tmpstr = AMG_CALLunary(sv, numer_amg);
2446 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2447 return SvIV(tmpstr);
2450 return PTR2IV(SvRV(sv));
2453 if (SvVALID(sv) || isREGEXP(sv)) {
2454 /* FBMs use the space for SvIVX and SvNVX for other purposes, so
2455 must not let them cache IVs.
2456 In practice they are extremely unlikely to actually get anywhere
2457 accessible by user Perl code - the only way that I'm aware of is when
2458 a constant subroutine which is used as the second argument to index.
2460 Regexps have no SvIVX and SvNVX fields.
2465 const char * const ptr =
2466 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2468 = grok_number(ptr, SvCUR(sv), &value);
2470 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2471 == IS_NUMBER_IN_UV) {
2472 /* It's definitely an integer */
2473 if (numtype & IS_NUMBER_NEG) {
2474 if (value < (UV)IV_MIN)
2477 if (value < (UV)IV_MAX)
2482 /* Quite wrong but no good choices. */
2483 if ((numtype & IS_NUMBER_INFINITY)) {
2484 return (numtype & IS_NUMBER_NEG) ? IV_MIN : IV_MAX;
2485 } else if ((numtype & IS_NUMBER_NAN)) {
2486 return 0; /* So wrong. */
2490 if (ckWARN(WARN_NUMERIC))
2493 return I_V(Atof(ptr));
2497 if (SvTHINKFIRST(sv)) {
2498 if (SvREADONLY(sv) && !SvOK(sv)) {
2499 if (ckWARN(WARN_UNINITIALIZED))
2506 if (S_sv_2iuv_common(aTHX_ sv))
2510 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2iv(%" IVdf ")\n",
2511 PTR2UV(sv),SvIVX(sv)));
2512 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
2516 =for apidoc sv_2uv_flags
2518 Return the unsigned integer value of an SV, doing any necessary string
2519 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2520 Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
2526 Perl_sv_2uv_flags(pTHX_ SV *const sv, const I32 flags)
2528 PERL_ARGS_ASSERT_SV_2UV_FLAGS;
2530 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2536 if (flags & SV_SKIP_OVERLOAD)
2538 tmpstr = AMG_CALLunary(sv, numer_amg);
2539 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2540 return SvUV(tmpstr);
2543 return PTR2UV(SvRV(sv));
2546 if (SvVALID(sv) || isREGEXP(sv)) {
2547 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2548 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2549 Regexps have no SvIVX and SvNVX fields. */
2553 const char * const ptr =
2554 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2556 = grok_number(ptr, SvCUR(sv), &value);
2558 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2559 == IS_NUMBER_IN_UV) {
2560 /* It's definitely an integer */
2561 if (!(numtype & IS_NUMBER_NEG))
2565 /* Quite wrong but no good choices. */
2566 if ((numtype & IS_NUMBER_INFINITY)) {
2567 return UV_MAX; /* So wrong. */
2568 } else if ((numtype & IS_NUMBER_NAN)) {
2569 return 0; /* So wrong. */
2573 if (ckWARN(WARN_NUMERIC))
2576 return U_V(Atof(ptr));
2580 if (SvTHINKFIRST(sv)) {
2581 if (SvREADONLY(sv) && !SvOK(sv)) {
2582 if (ckWARN(WARN_UNINITIALIZED))
2589 if (S_sv_2iuv_common(aTHX_ sv))
2593 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2uv(%" UVuf ")\n",
2594 PTR2UV(sv),SvUVX(sv)));
2595 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
2599 =for apidoc sv_2nv_flags
2601 Return the num value of an SV, doing any necessary string or integer
2602 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2603 Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
2609 Perl_sv_2nv_flags(pTHX_ SV *const sv, const I32 flags)
2611 PERL_ARGS_ASSERT_SV_2NV_FLAGS;
2613 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2614 && SvTYPE(sv) != SVt_PVFM);
2615 if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) {
2616 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2617 the same flag bit as SVf_IVisUV, so must not let them cache NVs.
2618 Regexps have no SvIVX and SvNVX fields. */
2620 if (flags & SV_GMAGIC)
2624 if (SvPOKp(sv) && !SvIOKp(sv)) {
2625 ptr = SvPVX_const(sv);
2626 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
2627 !grok_number(ptr, SvCUR(sv), NULL))
2633 return (NV)SvUVX(sv);
2635 return (NV)SvIVX(sv);
2640 assert(SvTYPE(sv) >= SVt_PVMG);
2641 /* This falls through to the report_uninit near the end of the
2643 } else if (SvTHINKFIRST(sv)) {
2648 if (flags & SV_SKIP_OVERLOAD)
2650 tmpstr = AMG_CALLunary(sv, numer_amg);
2651 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2652 return SvNV(tmpstr);
2655 return PTR2NV(SvRV(sv));
2657 if (SvREADONLY(sv) && !SvOK(sv)) {
2658 if (ckWARN(WARN_UNINITIALIZED))
2663 if (SvTYPE(sv) < SVt_NV) {
2664 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2665 sv_upgrade(sv, SVt_NV);
2666 CLANG_DIAG_IGNORE_STMT(-Wthread-safety);
2668 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2669 STORE_LC_NUMERIC_SET_STANDARD();
2670 PerlIO_printf(Perl_debug_log,
2671 "0x%" UVxf " num(%" NVgf ")\n",
2672 PTR2UV(sv), SvNVX(sv));
2673 RESTORE_LC_NUMERIC();
2675 CLANG_DIAG_RESTORE_STMT;
2678 else if (SvTYPE(sv) < SVt_PVNV)
2679 sv_upgrade(sv, SVt_PVNV);
2684 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
2685 #ifdef NV_PRESERVES_UV
2691 /* Only set the public NV OK flag if this NV preserves the IV */
2692 /* Check it's not 0xFFFFFFFFFFFFFFFF */
2694 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
2695 : (SvIVX(sv) == I_V(SvNVX(sv))))
2701 else if (SvPOKp(sv)) {
2703 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2704 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
2706 #ifdef NV_PRESERVES_UV
2707 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2708 == IS_NUMBER_IN_UV) {
2709 /* It's definitely an integer */
2710 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
2712 S_sv_setnv(aTHX_ sv, numtype);
2719 SvNV_set(sv, Atof(SvPVX_const(sv)));
2720 /* Only set the public NV OK flag if this NV preserves the value in
2721 the PV at least as well as an IV/UV would.
2722 Not sure how to do this 100% reliably. */
2723 /* if that shift count is out of range then Configure's test is
2724 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2726 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2727 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2728 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
2729 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2730 /* Can't use strtol etc to convert this string, so don't try.
2731 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2734 /* value has been set. It may not be precise. */
2735 if ((numtype & IS_NUMBER_NEG) && (value >= (UV)IV_MIN)) {
2736 /* 2s complement assumption for (UV)IV_MIN */
2737 SvNOK_on(sv); /* Integer is too negative. */
2742 if (numtype & IS_NUMBER_NEG) {
2743 /* -IV_MIN is undefined, but we should never reach
2744 * this point with both IS_NUMBER_NEG and value ==
2746 assert(value != (UV)IV_MIN);
2747 SvIV_set(sv, -(IV)value);
2748 } else if (value <= (UV)IV_MAX) {
2749 SvIV_set(sv, (IV)value);
2751 SvUV_set(sv, value);
2755 if (numtype & IS_NUMBER_NOT_INT) {
2756 /* I believe that even if the original PV had decimals,
2757 they are lost beyond the limit of the FP precision.
2758 However, neither is canonical, so both only get p
2759 flags. NWC, 2000/11/25 */
2760 /* Both already have p flags, so do nothing */
2762 const NV nv = SvNVX(sv);
2763 /* XXX should this spot have NAN_COMPARE_BROKEN, too? */
2764 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2765 if (SvIVX(sv) == I_V(nv)) {
2768 /* It had no "." so it must be integer. */
2772 /* between IV_MAX and NV(UV_MAX).
2773 Could be slightly > UV_MAX */
2775 if (numtype & IS_NUMBER_NOT_INT) {
2776 /* UV and NV both imprecise. */
2778 const UV nv_as_uv = U_V(nv);
2780 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2789 /* It might be more code efficient to go through the entire logic above
2790 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2791 gets complex and potentially buggy, so more programmer efficient
2792 to do it this way, by turning off the public flags: */
2794 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2795 #endif /* NV_PRESERVES_UV */
2798 if (isGV_with_GP(sv)) {
2799 glob_2number(MUTABLE_GV(sv));
2803 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2805 assert (SvTYPE(sv) >= SVt_NV);
2806 /* Typically the caller expects that sv_any is not NULL now. */
2807 /* XXX Ilya implies that this is a bug in callers that assume this
2808 and ideally should be fixed. */
2811 CLANG_DIAG_IGNORE_STMT(-Wthread-safety);
2813 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2814 STORE_LC_NUMERIC_SET_STANDARD();
2815 PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2nv(%" NVgf ")\n",
2816 PTR2UV(sv), SvNVX(sv));
2817 RESTORE_LC_NUMERIC();
2819 CLANG_DIAG_RESTORE_STMT;
2826 Return an SV with the numeric value of the source SV, doing any necessary
2827 reference or overload conversion. The caller is expected to have handled
2834 Perl_sv_2num(pTHX_ SV *const sv)
2836 PERL_ARGS_ASSERT_SV_2NUM;
2841 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
2842 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
2843 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2844 return sv_2num(tmpsv);
2846 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2849 /* int2str_table: lookup table containing string representations of all
2850 * two digit numbers. For example, int2str_table.arr[0] is "00" and
2851 * int2str_table.arr[12*2] is "12".
2853 * We are going to read two bytes at a time, so we have to ensure that
2854 * the array is aligned to a 2 byte boundary. That's why it was made a
2855 * union with a dummy U16 member. */
2856 static const union {
2859 } int2str_table = {{
2860 '0', '0', '0', '1', '0', '2', '0', '3', '0', '4', '0', '5', '0', '6',
2861 '0', '7', '0', '8', '0', '9', '1', '0', '1', '1', '1', '2', '1', '3',
2862 '1', '4', '1', '5', '1', '6', '1', '7', '1', '8', '1', '9', '2', '0',
2863 '2', '1', '2', '2', '2', '3', '2', '4', '2', '5', '2', '6', '2', '7',
2864 '2', '8', '2', '9', '3', '0', '3', '1', '3', '2', '3', '3', '3', '4',
2865 '3', '5', '3', '6', '3', '7', '3', '8', '3', '9', '4', '0', '4', '1',
2866 '4', '2', '4', '3', '4', '4', '4', '5', '4', '6', '4', '7', '4', '8',
2867 '4', '9', '5', '0', '5', '1', '5', '2', '5', '3', '5', '4', '5', '5',
2868 '5', '6', '5', '7', '5', '8', '5', '9', '6', '0', '6', '1', '6', '2',
2869 '6', '3', '6', '4', '6', '5', '6', '6', '6', '7', '6', '8', '6', '9',
2870 '7', '0', '7', '1', '7', '2', '7', '3', '7', '4', '7', '5', '7', '6',
2871 '7', '7', '7', '8', '7', '9', '8', '0', '8', '1', '8', '2', '8', '3',
2872 '8', '4', '8', '5', '8', '6', '8', '7', '8', '8', '8', '9', '9', '0',
2873 '9', '1', '9', '2', '9', '3', '9', '4', '9', '5', '9', '6', '9', '7',
2877 /* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2878 * UV as a string towards the end of buf, and return pointers to start and
2881 * We assume that buf is at least TYPE_CHARS(UV) long.
2884 PERL_STATIC_INLINE char *
2885 S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
2887 char *ptr = buf + TYPE_CHARS(UV);
2888 char * const ebuf = ptr;
2890 U16 *word_ptr, *word_table;
2892 PERL_ARGS_ASSERT_UIV_2BUF;
2894 /* ptr has to be properly aligned, because we will cast it to U16* */
2895 assert(PTR2nat(ptr) % 2 == 0);
2896 /* we are going to read/write two bytes at a time */
2897 word_ptr = (U16*)ptr;
2898 word_table = (U16*)int2str_table.arr;
2900 if (UNLIKELY(is_uv))
2911 *--word_ptr = word_table[uv % 100];
2914 ptr = (char*)word_ptr;
2917 *--ptr = (char)uv + '0';
2919 *--word_ptr = word_table[uv];
2920 ptr = (char*)word_ptr;
2930 /* Helper for sv_2pv_flags and sv_vcatpvfn_flags. If the NV is an
2931 * infinity or a not-a-number, writes the appropriate strings to the
2932 * buffer, including a zero byte. On success returns the written length,
2933 * excluding the zero byte, on failure (not an infinity, not a nan)
2934 * returns zero, assert-fails on maxlen being too short.
2936 * XXX for "Inf", "-Inf", and "NaN", we could have three read-only
2937 * shared string constants we point to, instead of generating a new
2938 * string for each instance. */
2940 S_infnan_2pv(NV nv, char* buffer, size_t maxlen, char plus) {
2942 assert(maxlen >= 4);
2943 if (Perl_isinf(nv)) {
2945 if (maxlen < 5) /* "-Inf\0" */
2955 else if (Perl_isnan(nv)) {
2959 /* XXX optionally output the payload mantissa bits as
2960 * "(unsigned)" (to match the nan("...") C99 function,
2961 * or maybe as "(0xhhh...)" would make more sense...
2962 * provide a format string so that the user can decide?
2963 * NOTE: would affect the maxlen and assert() logic.*/
2968 assert((s == buffer + 3) || (s == buffer + 4));
2974 =for apidoc sv_2pv_flags
2976 Returns a pointer to the string value of an SV, and sets C<*lp> to its length.
2977 If flags has the C<SV_GMAGIC> bit set, does an C<mg_get()> first. Coerces C<sv> to a
2978 string if necessary. Normally invoked via the C<SvPV_flags> macro.
2979 C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
2985 Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
2989 PERL_ARGS_ASSERT_SV_2PV_FLAGS;
2991 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2992 && SvTYPE(sv) != SVt_PVFM);
2993 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2998 if (flags & SV_SKIP_OVERLOAD)
3000 tmpstr = AMG_CALLunary(sv, string_amg);
3001 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
3002 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
3004 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
3008 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
3009 if (flags & SV_CONST_RETURN) {
3010 pv = (char *) SvPVX_const(tmpstr);
3012 pv = (flags & SV_MUTABLE_RETURN)
3013 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
3016 *lp = SvCUR(tmpstr);
3018 pv = sv_2pv_flags(tmpstr, lp, flags);
3031 SV *const referent = SvRV(sv);
3035 retval = buffer = savepvn("NULLREF", len);
3036 } else if (SvTYPE(referent) == SVt_REGEXP &&
3037 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
3038 amagic_is_enabled(string_amg))) {
3039 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
3043 /* If the regex is UTF-8 we want the containing scalar to
3044 have an UTF-8 flag too */
3051 *lp = RX_WRAPLEN(re);
3053 return RX_WRAPPED(re);
3055 const char *const typestr = sv_reftype(referent, 0);
3056 const STRLEN typelen = strlen(typestr);
3057 UV addr = PTR2UV(referent);
3058 const char *stashname = NULL;
3059 STRLEN stashnamelen = 0; /* hush, gcc */
3060 const char *buffer_end;
3062 if (SvOBJECT(referent)) {
3063 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
3066 stashname = HEK_KEY(name);
3067 stashnamelen = HEK_LEN(name);
3069 if (HEK_UTF8(name)) {
3075 stashname = "__ANON__";
3078 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
3079 + 2 * sizeof(UV) + 2 /* )\0 */;
3081 len = typelen + 3 /* (0x */
3082 + 2 * sizeof(UV) + 2 /* )\0 */;
3085 Newx(buffer, len, char);
3086 buffer_end = retval = buffer + len;
3088 /* Working backwards */
3092 *--retval = PL_hexdigit[addr & 15];
3093 } while (addr >>= 4);
3099 memcpy(retval, typestr, typelen);
3103 retval -= stashnamelen;
3104 memcpy(retval, stashname, stashnamelen);
3106 /* retval may not necessarily have reached the start of the
3108 assert (retval >= buffer);
3110 len = buffer_end - retval - 1; /* -1 for that \0 */
3122 if (flags & SV_MUTABLE_RETURN)
3123 return SvPVX_mutable(sv);
3124 if (flags & SV_CONST_RETURN)
3125 return (char *)SvPVX_const(sv);
3130 /* I'm assuming that if both IV and NV are equally valid then
3131 converting the IV is going to be more efficient */
3132 const U32 isUIOK = SvIsUV(sv);
3133 /* The purpose of this union is to ensure that arr is aligned on
3134 a 2 byte boundary, because that is what uiv_2buf() requires */
3136 char arr[TYPE_CHARS(UV)];
3142 if (SvTYPE(sv) < SVt_PVIV)
3143 sv_upgrade(sv, SVt_PVIV);
3144 ptr = uiv_2buf(buf.arr, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
3146 /* inlined from sv_setpvn */
3147 s = SvGROW_mutable(sv, len + 1);
3148 Move(ptr, s, len, char);
3153 else if (SvNOK(sv)) {
3154 if (SvTYPE(sv) < SVt_PVNV)
3155 sv_upgrade(sv, SVt_PVNV);
3156 if (SvNVX(sv) == 0.0
3157 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
3158 && !Perl_isnan(SvNVX(sv))
3161 s = SvGROW_mutable(sv, 2);
3166 STRLEN size = 5; /* "-Inf\0" */
3168 s = SvGROW_mutable(sv, size);
3169 len = S_infnan_2pv(SvNVX(sv), s, size, 0);
3175 /* some Xenix systems wipe out errno here */
3184 5 + /* exponent digits */
3188 s = SvGROW_mutable(sv, size);
3189 #ifndef USE_LOCALE_NUMERIC
3190 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3196 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
3197 STORE_LC_NUMERIC_SET_TO_NEEDED();
3199 local_radix = _NOT_IN_NUMERIC_STANDARD;
3200 if (local_radix && SvCUR(PL_numeric_radix_sv) > 1) {
3201 size += SvCUR(PL_numeric_radix_sv) - 1;
3202 s = SvGROW_mutable(sv, size);
3205 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3207 /* If the radix character is UTF-8, and actually is in the
3208 * output, turn on the UTF-8 flag for the scalar */
3210 && SvUTF8(PL_numeric_radix_sv)
3211 && instr(s, SvPVX_const(PL_numeric_radix_sv)))
3216 RESTORE_LC_NUMERIC();
3219 /* We don't call SvPOK_on(), because it may come to
3220 * pass that the locale changes so that the
3221 * stringification we just did is no longer correct. We
3222 * will have to re-stringify every time it is needed */
3229 else if (isGV_with_GP(sv)) {
3230 GV *const gv = MUTABLE_GV(sv);
3231 SV *const buffer = sv_newmortal();
3233 gv_efullname3(buffer, gv, "*");
3235 assert(SvPOK(buffer));
3241 *lp = SvCUR(buffer);
3242 return SvPVX(buffer);
3247 if (flags & SV_UNDEF_RETURNS_NULL)
3249 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
3251 /* Typically the caller expects that sv_any is not NULL now. */
3252 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
3253 sv_upgrade(sv, SVt_PV);
3258 const STRLEN len = s - SvPVX_const(sv);
3263 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2pv(%s)\n",
3264 PTR2UV(sv),SvPVX_const(sv)));
3265 if (flags & SV_CONST_RETURN)
3266 return (char *)SvPVX_const(sv);
3267 if (flags & SV_MUTABLE_RETURN)
3268 return SvPVX_mutable(sv);
3273 =for apidoc sv_copypv
3275 Copies a stringified representation of the source SV into the
3276 destination SV. Automatically performs any necessary C<mg_get> and
3277 coercion of numeric values into strings. Guaranteed to preserve
3278 C<UTF8> flag even from overloaded objects. Similar in nature to
3279 C<sv_2pv[_flags]> but operates directly on an SV instead of just the
3280 string. Mostly uses C<sv_2pv_flags> to do its work, except when that
3281 would lose the UTF-8'ness of the PV.
3283 =for apidoc sv_copypv_nomg
3285 Like C<sv_copypv>, but doesn't invoke get magic first.
3287 =for apidoc sv_copypv_flags
3289 Implementation of C<sv_copypv> and C<sv_copypv_nomg>. Calls get magic iff flags
3290 has the C<SV_GMAGIC> bit set.
3296 Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
3301 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
3303 s = SvPV_flags_const(ssv,len,(flags & SV_GMAGIC));
3304 sv_setpvn(dsv,s,len);
3312 =for apidoc sv_2pvbyte
3314 Return a pointer to the byte-encoded representation of the SV, and set C<*lp>
3315 to its length. May cause the SV to be downgraded from UTF-8 as a
3318 Usually accessed via the C<SvPVbyte> macro.
3324 Perl_sv_2pvbyte(pTHX_ SV *sv, STRLEN *const lp)
3326 PERL_ARGS_ASSERT_SV_2PVBYTE;
3329 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3330 || isGV_with_GP(sv) || SvROK(sv)) {
3331 SV *sv2 = sv_newmortal();
3332 sv_copypv_nomg(sv2,sv);
3335 sv_utf8_downgrade(sv,0);
3336 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3340 =for apidoc sv_2pvutf8
3342 Return a pointer to the UTF-8-encoded representation of the SV, and set C<*lp>
3343 to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3345 Usually accessed via the C<SvPVutf8> macro.
3351 Perl_sv_2pvutf8(pTHX_ SV *sv, STRLEN *const lp)
3353 PERL_ARGS_ASSERT_SV_2PVUTF8;
3355 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3356 || isGV_with_GP(sv) || SvROK(sv))
3357 sv = sv_mortalcopy(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.
3655 Perl_sv_utf8_downgrade(pTHX_ SV *const sv, const bool fail_ok)
3657 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3659 if (SvPOKp(sv) && SvUTF8(sv)) {
3663 int mg_flags = SV_GMAGIC;
3666 S_sv_uncow(aTHX_ sv, 0);
3668 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3670 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3671 if (mg && mg->mg_len > 0 && mg->mg_flags & MGf_BYTES) {
3672 mg->mg_len = sv_pos_b2u_flags(sv, mg->mg_len,
3673 SV_GMAGIC|SV_CONST_RETURN);
3674 mg_flags = 0; /* sv_pos_b2u does get magic */
3676 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3677 magic_setutf8(sv,mg); /* clear UTF8 cache */
3680 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3682 if (!utf8_to_bytes(s, &len)) {
3687 Perl_croak(aTHX_ "Wide character in %s",
3690 Perl_croak(aTHX_ "Wide character");
3701 =for apidoc sv_utf8_encode
3703 Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3704 flag off so that it looks like octets again.
3710 Perl_sv_utf8_encode(pTHX_ SV *const sv)
3712 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3714 if (SvREADONLY(sv)) {
3715 sv_force_normal_flags(sv, 0);
3717 (void) sv_utf8_upgrade(sv);
3722 =for apidoc sv_utf8_decode
3724 If the PV of the SV is an octet sequence in Perl's extended UTF-8
3725 and contains a multiple-byte character, the C<SvUTF8> flag is turned on
3726 so that it looks like a character. If the PV contains only single-byte
3727 characters, the C<SvUTF8> flag stays off.
3728 Scans PV for validity and returns FALSE if the PV is invalid UTF-8.
3734 Perl_sv_utf8_decode(pTHX_ SV *const sv)
3736 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3739 const U8 *start, *c, *first_variant;
3741 /* The octets may have got themselves encoded - get them back as
3744 if (!sv_utf8_downgrade(sv, TRUE))
3747 /* it is actually just a matter of turning the utf8 flag on, but
3748 * we want to make sure everything inside is valid utf8 first.
3750 c = start = (const U8 *) SvPVX_const(sv);
3751 if (! is_utf8_invariant_string_loc(c, SvCUR(sv), &first_variant)) {
3752 if (!is_utf8_string(first_variant, SvCUR(sv) - (first_variant -c)))
3756 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3757 /* XXX Is this dead code? XS_utf8_decode calls SvSETMAGIC
3758 after this, clearing pos. Does anything on CPAN
3760 /* adjust pos to the start of a UTF8 char sequence */
3761 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3763 I32 pos = mg->mg_len;
3765 for (c = start + pos; c > start; c--) {
3766 if (UTF8_IS_START(*c))
3769 mg->mg_len = c - start;
3772 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3773 magic_setutf8(sv,mg); /* clear UTF8 cache */
3780 =for apidoc sv_setsv
3782 Copies the contents of the source SV C<ssv> into the destination SV
3783 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3784 function if the source SV needs to be reused. Does not handle 'set' magic on
3785 destination SV. Calls 'get' magic on source SV. Loosely speaking, it
3786 performs a copy-by-value, obliterating any previous content of the
3789 You probably want to use one of the assortment of wrappers, such as
3790 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3791 C<SvSetMagicSV_nosteal>.
3793 =for apidoc sv_setsv_flags
3795 Copies the contents of the source SV C<ssv> into the destination SV
3796 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3797 function if the source SV needs to be reused. Does not handle 'set' magic.
3798 Loosely speaking, it performs a copy-by-value, obliterating any previous
3799 content of the destination.
3800 If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
3801 C<ssv> if appropriate, else not. If the C<flags>
3802 parameter has the C<SV_NOSTEAL> bit set then the
3803 buffers of temps will not be stolen. C<sv_setsv>
3804 and C<sv_setsv_nomg> are implemented in terms of this function.
3806 You probably want to use one of the assortment of wrappers, such as
3807 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3808 C<SvSetMagicSV_nosteal>.
3810 This is the primary function for copying scalars, and most other
3811 copy-ish functions and macros use this underneath.
3817 S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
3819 I32 mro_changes = 0; /* 1 = method, 2 = isa, 3 = recursive isa */
3820 HV *old_stash = NULL;
3822 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB;
3824 if (dtype != SVt_PVGV && !isGV_with_GP(dstr)) {
3825 const char * const name = GvNAME(sstr);
3826 const STRLEN len = GvNAMELEN(sstr);
3828 if (dtype >= SVt_PV) {
3834 SvUPGRADE(dstr, SVt_PVGV);
3835 (void)SvOK_off(dstr);
3836 isGV_with_GP_on(dstr);
3838 GvSTASH(dstr) = GvSTASH(sstr);
3840 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
3841 gv_name_set(MUTABLE_GV(dstr), name, len,
3842 GV_ADD | (GvNAMEUTF8(sstr) ? SVf_UTF8 : 0 ));
3843 SvFAKE_on(dstr); /* can coerce to non-glob */
3846 if(GvGP(MUTABLE_GV(sstr))) {
3847 /* If source has method cache entry, clear it */
3849 SvREFCNT_dec(GvCV(sstr));
3850 GvCV_set(sstr, NULL);
3853 /* If source has a real method, then a method is
3856 GvCV((const GV *)sstr) && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3862 /* If dest already had a real method, that's a change as well */
3864 !mro_changes && GvGP(MUTABLE_GV(dstr)) && GvCVu((const GV *)dstr)
3865 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3870 /* We don't need to check the name of the destination if it was not a
3871 glob to begin with. */
3872 if(dtype == SVt_PVGV) {
3873 const char * const name = GvNAME((const GV *)dstr);
3874 const STRLEN len = GvNAMELEN(dstr);
3875 if(memEQs(name, len, "ISA")
3876 /* The stash may have been detached from the symbol table, so
3878 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3882 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
3883 || (len == 1 && name[0] == ':')) {
3886 /* Set aside the old stash, so we can reset isa caches on
3888 if((old_stash = GvHV(dstr)))
3889 /* Make sure we do not lose it early. */
3890 SvREFCNT_inc_simple_void_NN(
3891 sv_2mortal((SV *)old_stash)
3896 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
3899 /* freeing dstr's GP might free sstr (e.g. *x = $x),
3900 * so temporarily protect it */
3902 SAVEFREESV(SvREFCNT_inc_simple_NN(sstr));
3903 gp_free(MUTABLE_GV(dstr));
3904 GvINTRO_off(dstr); /* one-shot flag */
3905 GvGP_set(dstr, gp_ref(GvGP(sstr)));
3908 if (SvTAINTED(sstr))
3910 if (GvIMPORTED(dstr) != GVf_IMPORTED
3911 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
3913 GvIMPORTED_on(dstr);
3916 if(mro_changes == 2) {
3917 if (GvAV((const GV *)sstr)) {
3919 SV * const sref = (SV *)GvAV((const GV *)dstr);
3920 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
3921 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
3922 AV * const ary = newAV();
3923 av_push(ary, mg->mg_obj); /* takes the refcount */
3924 mg->mg_obj = (SV *)ary;
3926 av_push((AV *)mg->mg_obj, SvREFCNT_inc_simple_NN(dstr));
3928 else sv_magic(sref, dstr, PERL_MAGIC_isa, NULL, 0);
3930 mro_isa_changed_in(GvSTASH(dstr));
3932 else if(mro_changes == 3) {
3933 HV * const stash = GvHV(dstr);
3934 if(old_stash ? (HV *)HvENAME_get(old_stash) : stash)
3940 else if(mro_changes) mro_method_changed_in(GvSTASH(dstr));
3941 if (GvIO(dstr) && dtype == SVt_PVGV) {
3942 DEBUG_o(Perl_deb(aTHX_
3943 "glob_assign_glob clearing PL_stashcache\n"));
3944 /* It's a cache. It will rebuild itself quite happily.
3945 It's a lot of effort to work out exactly which key (or keys)
3946 might be invalidated by the creation of the this file handle.
3948 hv_clear(PL_stashcache);
3954 Perl_gv_setref(pTHX_ SV *const dstr, SV *const sstr)
3956 SV * const sref = SvRV(sstr);
3958 const int intro = GvINTRO(dstr);
3961 const U32 stype = SvTYPE(sref);
3963 PERL_ARGS_ASSERT_GV_SETREF;
3966 GvINTRO_off(dstr); /* one-shot flag */
3967 GvLINE(dstr) = CopLINE(PL_curcop);
3968 GvEGV(dstr) = MUTABLE_GV(dstr);
3973 location = (SV **) &(GvGP(dstr)->gp_cv); /* XXX bypassing GvCV_set */
3974 import_flag = GVf_IMPORTED_CV;
3977 location = (SV **) &GvHV(dstr);
3978 import_flag = GVf_IMPORTED_HV;
3981 location = (SV **) &GvAV(dstr);
3982 import_flag = GVf_IMPORTED_AV;
3985 location = (SV **) &GvIOp(dstr);
3988 location = (SV **) &GvFORM(dstr);
3991 location = &GvSV(dstr);
3992 import_flag = GVf_IMPORTED_SV;
3995 if (stype == SVt_PVCV) {
3996 /*if (GvCVGEN(dstr) && (GvCV(dstr) != (const CV *)sref || GvCVGEN(dstr))) {*/
3997 if (GvCVGEN(dstr)) {
3998 SvREFCNT_dec(GvCV(dstr));
3999 GvCV_set(dstr, NULL);
4000 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4003 /* SAVEt_GVSLOT takes more room on the savestack and has more
4004 overhead in leave_scope than SAVEt_GENERIC_SV. But for CVs
4005 leave_scope needs access to the GV so it can reset method
4006 caches. We must use SAVEt_GVSLOT whenever the type is
4007 SVt_PVCV, even if the stash is anonymous, as the stash may
4008 gain a name somehow before leave_scope. */
4009 if (stype == SVt_PVCV) {
4010 /* There is no save_pushptrptrptr. Creating it for this
4011 one call site would be overkill. So inline the ss add
4015 SS_ADD_PTR(location);
4016 SS_ADD_PTR(SvREFCNT_inc(*location));
4017 SS_ADD_UV(SAVEt_GVSLOT);
4020 else SAVEGENERICSV(*location);
4023 if (stype == SVt_PVCV && (*location != sref || GvCVGEN(dstr))) {
4024 CV* const cv = MUTABLE_CV(*location);
4026 if (!GvCVGEN((const GV *)dstr) &&
4027 (CvROOT(cv) || CvXSUB(cv)) &&
4028 /* redundant check that avoids creating the extra SV
4029 most of the time: */
4030 (CvCONST(cv) || ckWARN(WARN_REDEFINE)))
4032 SV * const new_const_sv =
4033 CvCONST((const CV *)sref)
4034 ? cv_const_sv((const CV *)sref)
4036 HV * const stash = GvSTASH((const GV *)dstr);
4037 report_redefined_cv(
4040 ? Perl_newSVpvf(aTHX_
4041 "%" HEKf "::%" HEKf,
4042 HEKfARG(HvNAME_HEK(stash)),
4043 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr))))
4044 : Perl_newSVpvf(aTHX_
4046 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr))))
4049 CvCONST((const CV *)sref) ? &new_const_sv : NULL
4053 cv_ckproto_len_flags(cv, (const GV *)dstr,
4054 SvPOK(sref) ? CvPROTO(sref) : NULL,
4055 SvPOK(sref) ? CvPROTOLEN(sref) : 0,
4056 SvPOK(sref) ? SvUTF8(sref) : 0);
4058 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4059 GvASSUMECV_on(dstr);
4060 if(GvSTASH(dstr)) { /* sub foo { 1 } sub bar { 2 } *bar = \&foo */
4061 if (intro && GvREFCNT(dstr) > 1) {
4062 /* temporary remove extra savestack's ref */
4064 gv_method_changed(dstr);
4067 else gv_method_changed(dstr);
4070 *location = SvREFCNT_inc_simple_NN(sref);
4071 if (import_flag && !(GvFLAGS(dstr) & import_flag)
4072 && CopSTASH_ne(PL_curcop, GvSTASH(dstr))) {
4073 GvFLAGS(dstr) |= import_flag;
4076 if (stype == SVt_PVHV) {
4077 const char * const name = GvNAME((GV*)dstr);
4078 const STRLEN len = GvNAMELEN(dstr);
4081 (len > 1 && name[len-2] == ':' && name[len-1] == ':')
4082 || (len == 1 && name[0] == ':')
4084 && (!dref || HvENAME_get(dref))
4087 (HV *)sref, (HV *)dref,
4093 stype == SVt_PVAV && sref != dref
4094 && memEQs(GvNAME((GV*)dstr), GvNAMELEN((GV*)dstr), "ISA")
4095 /* The stash may have been detached from the symbol table, so
4096 check its name before doing anything. */
4097 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
4100 MAGIC * const omg = dref && SvSMAGICAL(dref)
4101 ? mg_find(dref, PERL_MAGIC_isa)
4103 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
4104 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
4105 AV * const ary = newAV();
4106 av_push(ary, mg->mg_obj); /* takes the refcount */
4107 mg->mg_obj = (SV *)ary;
4110 if (SvTYPE(omg->mg_obj) == SVt_PVAV) {
4111 SV **svp = AvARRAY((AV *)omg->mg_obj);
4112 I32 items = AvFILLp((AV *)omg->mg_obj) + 1;
4116 SvREFCNT_inc_simple_NN(*svp++)
4122 SvREFCNT_inc_simple_NN(omg->mg_obj)
4126 av_push((AV *)mg->mg_obj,SvREFCNT_inc_simple_NN(dstr));
4132 sref, omg ? omg->mg_obj : dstr, PERL_MAGIC_isa, NULL, 0
4134 for (i = 0; i <= AvFILL(sref); ++i) {
4135 SV **elem = av_fetch ((AV*)sref, i, 0);
4138 *elem, sref, PERL_MAGIC_isaelem, NULL, i
4142 mg = mg_find(sref, PERL_MAGIC_isa);
4144 /* Since the *ISA assignment could have affected more than
4145 one stash, don't call mro_isa_changed_in directly, but let
4146 magic_clearisa do it for us, as it already has the logic for
4147 dealing with globs vs arrays of globs. */
4149 Perl_magic_clearisa(aTHX_ NULL, mg);
4151 else if (stype == SVt_PVIO) {
4152 DEBUG_o(Perl_deb(aTHX_ "gv_setref clearing PL_stashcache\n"));
4153 /* It's a cache. It will rebuild itself quite happily.
4154 It's a lot of effort to work out exactly which key (or keys)
4155 might be invalidated by the creation of the this file handle.
4157 hv_clear(PL_stashcache);
4161 if (!intro) SvREFCNT_dec(dref);
4162 if (SvTAINTED(sstr))
4170 #ifdef PERL_DEBUG_READONLY_COW
4171 # include <sys/mman.h>
4173 # ifndef PERL_MEMORY_DEBUG_HEADER_SIZE
4174 # define PERL_MEMORY_DEBUG_HEADER_SIZE 0
4178 Perl_sv_buf_to_ro(pTHX_ SV *sv)
4180 struct perl_memory_debug_header * const header =
4181 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4182 const MEM_SIZE len = header->size;
4183 PERL_ARGS_ASSERT_SV_BUF_TO_RO;
4184 # ifdef PERL_TRACK_MEMPOOL
4185 if (!header->readonly) header->readonly = 1;
4187 if (mprotect(header, len, PROT_READ))
4188 Perl_warn(aTHX_ "mprotect RW for COW string %p %lu failed with %d",
4189 header, len, errno);
4193 S_sv_buf_to_rw(pTHX_ SV *sv)
4195 struct perl_memory_debug_header * const header =
4196 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4197 const MEM_SIZE len = header->size;
4198 PERL_ARGS_ASSERT_SV_BUF_TO_RW;
4199 if (mprotect(header, len, PROT_READ|PROT_WRITE))
4200 Perl_warn(aTHX_ "mprotect for COW string %p %lu failed with %d",
4201 header, len, errno);
4202 # ifdef PERL_TRACK_MEMPOOL
4203 header->readonly = 0;
4208 # define sv_buf_to_ro(sv) NOOP
4209 # define sv_buf_to_rw(sv) NOOP
4213 Perl_sv_setsv_flags(pTHX_ SV *dstr, SV* sstr, const I32 flags)
4218 unsigned int both_type;
4220 PERL_ARGS_ASSERT_SV_SETSV_FLAGS;
4222 if (UNLIKELY( sstr == dstr ))
4225 if (UNLIKELY( !sstr ))
4226 sstr = &PL_sv_undef;
4228 stype = SvTYPE(sstr);
4229 dtype = SvTYPE(dstr);
4230 both_type = (stype | dtype);
4232 /* with these values, we can check that both SVs are NULL/IV (and not
4233 * freed) just by testing the or'ed types */
4234 STATIC_ASSERT_STMT(SVt_NULL == 0);
4235 STATIC_ASSERT_STMT(SVt_IV == 1);
4236 if (both_type <= 1) {
4237 /* both src and dst are UNDEF/IV/RV, so we can do a lot of
4243 /* minimal subset of SV_CHECK_THINKFIRST_COW_DROP(dstr) */
4244 if (SvREADONLY(dstr))
4245 Perl_croak_no_modify();
4247 if (SvWEAKREF(dstr))
4248 sv_unref_flags(dstr, 0);
4250 old_rv = SvRV(dstr);
4253 assert(!SvGMAGICAL(sstr));
4254 assert(!SvGMAGICAL(dstr));
4256 sflags = SvFLAGS(sstr);
4257 if (sflags & (SVf_IOK|SVf_ROK)) {
4258 SET_SVANY_FOR_BODYLESS_IV(dstr);
4259 new_dflags = SVt_IV;
4261 if (sflags & SVf_ROK) {
4262 dstr->sv_u.svu_rv = SvREFCNT_inc(SvRV(sstr));
4263 new_dflags |= SVf_ROK;
4266 /* both src and dst are <= SVt_IV, so sv_any points to the
4267 * head; so access the head directly
4269 assert( &(sstr->sv_u.svu_iv)
4270 == &(((XPVIV*) SvANY(sstr))->xiv_iv));
4271 assert( &(dstr->sv_u.svu_iv)
4272 == &(((XPVIV*) SvANY(dstr))->xiv_iv));
4273 dstr->sv_u.svu_iv = sstr->sv_u.svu_iv;
4274 new_dflags |= (SVf_IOK|SVp_IOK|(sflags & SVf_IVisUV));
4278 new_dflags = dtype; /* turn off everything except the type */
4280 SvFLAGS(dstr) = new_dflags;
4281 SvREFCNT_dec(old_rv);
4286 if (UNLIKELY(both_type == SVTYPEMASK)) {
4287 if (SvIS_FREED(dstr)) {
4288 Perl_croak(aTHX_ "panic: attempt to copy value %" SVf
4289 " to a freed scalar %p", SVfARG(sstr), (void *)dstr);
4291 if (SvIS_FREED(sstr)) {
4292 Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p to %p",
4293 (void*)sstr, (void*)dstr);
4299 SV_CHECK_THINKFIRST_COW_DROP(dstr);
4300 dtype = SvTYPE(dstr); /* THINKFIRST may have changed type */
4302 /* There's a lot of redundancy below but we're going for speed here */
4307 if (LIKELY( dtype != SVt_PVGV && dtype != SVt_PVLV )) {
4308 (void)SvOK_off(dstr);
4316 /* For performance, we inline promoting to type SVt_IV. */
4317 /* We're starting from SVt_NULL, so provided that define is
4318 * actual 0, we don't have to unset any SV type flags
4319 * to promote to SVt_IV. */
4320 STATIC_ASSERT_STMT(SVt_NULL == 0);
4321 SET_SVANY_FOR_BODYLESS_IV(dstr);
4322 SvFLAGS(dstr) |= SVt_IV;
4326 sv_upgrade(dstr, SVt_PVIV);
4330 goto end_of_first_switch;
4332 (void)SvIOK_only(dstr);
4333 SvIV_set(dstr, SvIVX(sstr));
4336 /* SvTAINTED can only be true if the SV has taint magic, which in
4337 turn means that the SV type is PVMG (or greater). This is the
4338 case statement for SVt_IV, so this cannot be true (whatever gcov
4340 assert(!SvTAINTED(sstr));
4345 if (dtype < SVt_PV && dtype != SVt_IV)
4346 sv_upgrade(dstr, SVt_IV);
4350 if (LIKELY( SvNOK(sstr) )) {
4354 sv_upgrade(dstr, SVt_NV);
4358 sv_upgrade(dstr, SVt_PVNV);
4362 goto end_of_first_switch;
4364 SvNV_set(dstr, SvNVX(sstr));
4365 (void)SvNOK_only(dstr);
4366 /* SvTAINTED can only be true if the SV has taint magic, which in
4367 turn means that the SV type is PVMG (or greater). This is the
4368 case statement for SVt_NV, so this cannot be true (whatever gcov
4370 assert(!SvTAINTED(sstr));
4377 sv_upgrade(dstr, SVt_PV);
4380 if (dtype < SVt_PVIV)
4381 sv_upgrade(dstr, SVt_PVIV);
4384 if (dtype < SVt_PVNV)
4385 sv_upgrade(dstr, SVt_PVNV);
4389 invlist_clone(sstr, dstr);
4393 const char * const type = sv_reftype(sstr,0);
4395 /* diag_listed_as: Bizarre copy of %s */
4396 Perl_croak(aTHX_ "Bizarre copy of %s in %s", type, OP_DESC(PL_op));
4398 Perl_croak(aTHX_ "Bizarre copy of %s", type);
4400 NOT_REACHED; /* NOTREACHED */
4404 if (dtype < SVt_REGEXP)
4405 sv_upgrade(dstr, SVt_REGEXP);
4411 if (SvGMAGICAL(sstr) && (flags & SV_GMAGIC)) {
4413 if (SvTYPE(sstr) != stype)
4414 stype = SvTYPE(sstr);
4416 if (isGV_with_GP(sstr) && dtype <= SVt_PVLV) {
4417 glob_assign_glob(dstr, sstr, dtype);
4420 if (stype == SVt_PVLV)
4422 if (isREGEXP(sstr)) goto upgregexp;
4423 SvUPGRADE(dstr, SVt_PVNV);
4426 SvUPGRADE(dstr, (svtype)stype);
4428 end_of_first_switch:
4430 /* dstr may have been upgraded. */
4431 dtype = SvTYPE(dstr);
4432 sflags = SvFLAGS(sstr);
4434 if (UNLIKELY( dtype == SVt_PVCV )) {
4435 /* Assigning to a subroutine sets the prototype. */
4438 const char *const ptr = SvPV_const(sstr, len);
4440 SvGROW(dstr, len + 1);
4441 Copy(ptr, SvPVX(dstr), len + 1, char);
4442 SvCUR_set(dstr, len);
4444 SvFLAGS(dstr) |= sflags & SVf_UTF8;
4445 CvAUTOLOAD_off(dstr);
4450 else if (UNLIKELY(dtype == SVt_PVAV || dtype == SVt_PVHV
4451 || dtype == SVt_PVFM))
4453 const char * const type = sv_reftype(dstr,0);
4455 /* diag_listed_as: Cannot copy to %s */
4456 Perl_croak(aTHX_ "Cannot copy to %s in %s", type, OP_DESC(PL_op));
4458 Perl_croak(aTHX_ "Cannot copy to %s", type);
4459 } else if (sflags & SVf_ROK) {
4460 if (isGV_with_GP(dstr)
4461 && SvTYPE(SvRV(sstr)) == SVt_PVGV && isGV_with_GP(SvRV(sstr))) {
4464 if (GvIMPORTED(dstr) != GVf_IMPORTED
4465 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
4467 GvIMPORTED_on(dstr);
4472 glob_assign_glob(dstr, sstr, dtype);
4476 if (dtype >= SVt_PV) {
4477 if (isGV_with_GP(dstr)) {
4478 gv_setref(dstr, sstr);
4481 if (SvPVX_const(dstr)) {
4487 (void)SvOK_off(dstr);
4488 SvRV_set(dstr, SvREFCNT_inc(SvRV(sstr)));
4489 SvFLAGS(dstr) |= sflags & SVf_ROK;
4490 assert(!(sflags & SVp_NOK));
4491 assert(!(sflags & SVp_IOK));
4492 assert(!(sflags & SVf_NOK));
4493 assert(!(sflags & SVf_IOK));
4495 else if (isGV_with_GP(dstr)) {
4496 if (!(sflags & SVf_OK)) {
4497 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4498 "Undefined value assigned to typeglob");
4501 GV *gv = gv_fetchsv_nomg(sstr, GV_ADD, SVt_PVGV);
4502 if (dstr != (const SV *)gv) {
4503 const char * const name = GvNAME((const GV *)dstr);
4504 const STRLEN len = GvNAMELEN(dstr);
4505 HV *old_stash = NULL;
4506 bool reset_isa = FALSE;
4507 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
4508 || (len == 1 && name[0] == ':')) {
4509 /* Set aside the old stash, so we can reset isa caches
4510 on its subclasses. */
4511 if((old_stash = GvHV(dstr))) {
4512 /* Make sure we do not lose it early. */
4513 SvREFCNT_inc_simple_void_NN(
4514 sv_2mortal((SV *)old_stash)
4521 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
4522 gp_free(MUTABLE_GV(dstr));
4524 GvGP_set(dstr, gp_ref(GvGP(gv)));
4527 HV * const stash = GvHV(dstr);
4529 old_stash ? (HV *)HvENAME_get(old_stash) : stash
4539 else if ((dtype == SVt_REGEXP || dtype == SVt_PVLV)
4540 && (stype == SVt_REGEXP || isREGEXP(sstr))) {
4541 reg_temp_copy((REGEXP*)dstr, (REGEXP*)sstr);
4543 else if (sflags & SVp_POK) {
4544 const STRLEN cur = SvCUR(sstr);
4545 const STRLEN len = SvLEN(sstr);
4548 * We have three basic ways to copy the string:
4554 * Which we choose is based on various factors. The following
4555 * things are listed in order of speed, fastest to slowest:
4557 * - Copying a short string
4558 * - Copy-on-write bookkeeping
4560 * - Copying a long string
4562 * We swipe the string (steal the string buffer) if the SV on the
4563 * rhs is about to be freed anyway (TEMP and refcnt==1). This is a
4564 * big win on long strings. It should be a win on short strings if
4565 * SvPVX_const(dstr) has to be allocated. If not, it should not
4566 * slow things down, as SvPVX_const(sstr) would have been freed
4569 * We also steal the buffer from a PADTMP (operator target) if it
4570 * is ‘long enough’. For short strings, a swipe does not help
4571 * here, as it causes more malloc calls the next time the target
4572 * is used. Benchmarks show that even if SvPVX_const(dstr) has to
4573 * be allocated it is still not worth swiping PADTMPs for short
4574 * strings, as the savings here are small.
4576 * If swiping is not an option, then we see whether it is
4577 * worth using copy-on-write. If the lhs already has a buf-
4578 * fer big enough and the string is short, we skip it and fall back
4579 * to method 3, since memcpy is faster for short strings than the
4580 * later bookkeeping overhead that copy-on-write entails.
4582 * If the rhs is not a copy-on-write string yet, then we also
4583 * consider whether the buffer is too large relative to the string
4584 * it holds. Some operations such as readline allocate a large
4585 * buffer in the expectation of reusing it. But turning such into
4586 * a COW buffer is counter-productive because it increases memory
4587 * usage by making readline allocate a new large buffer the sec-
4588 * ond time round. So, if the buffer is too large, again, we use
4591 * Finally, if there is no buffer on the left, or the buffer is too
4592 * small, then we use copy-on-write and make both SVs share the
4597 /* Whichever path we take through the next code, we want this true,
4598 and doing it now facilitates the COW check. */
4599 (void)SvPOK_only(dstr);
4603 /* slated for free anyway (and not COW)? */
4604 (sflags & (SVs_TEMP|SVf_IsCOW)) == SVs_TEMP
4605 /* or a swipable TARG */
4607 (SVs_PADTMP|SVf_READONLY|SVf_PROTECT|SVf_IsCOW))
4609 /* whose buffer is worth stealing */
4610 && CHECK_COWBUF_THRESHOLD(cur,len)
4613 !(sflags & SVf_OOK) && /* and not involved in OOK hack? */
4614 (!(flags & SV_NOSTEAL)) &&
4615 /* and we're allowed to steal temps */
4616 SvREFCNT(sstr) == 1 && /* and no other references to it? */
4617 len) /* and really is a string */
4618 { /* Passes the swipe test. */
4619 if (SvPVX_const(dstr)) /* we know that dtype >= SVt_PV */
4621 SvPV_set(dstr, SvPVX_mutable(sstr));
4622 SvLEN_set(dstr, SvLEN(sstr));
4623 SvCUR_set(dstr, SvCUR(sstr));
4626 (void)SvOK_off(sstr); /* NOTE: nukes most SvFLAGS on sstr */
4627 SvPV_set(sstr, NULL);
4632 else if (flags & SV_COW_SHARED_HASH_KEYS
4634 #ifdef PERL_COPY_ON_WRITE
4637 ( (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4638 /* If this is a regular (non-hek) COW, only so
4639 many COW "copies" are possible. */
4640 && CowREFCNT(sstr) != SV_COW_REFCNT_MAX ))
4641 : ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4642 && !(SvFLAGS(dstr) & SVf_BREAK)
4643 && CHECK_COW_THRESHOLD(cur,len) && cur+1 < len
4644 && (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4648 && !(SvFLAGS(dstr) & SVf_BREAK)
4651 /* Either it's a shared hash key, or it's suitable for
4655 PerlIO_printf(Perl_debug_log, "Copy on write: sstr --> dstr\n");
4661 if (!(sflags & SVf_IsCOW)) {
4663 CowREFCNT(sstr) = 0;
4666 if (SvPVX_const(dstr)) { /* we know that dtype >= SVt_PV */
4672 if (sflags & SVf_IsCOW) {
4676 SvPV_set(dstr, SvPVX_mutable(sstr));
4681 /* SvIsCOW_shared_hash */
4682 DEBUG_C(PerlIO_printf(Perl_debug_log,
4683 "Copy on write: Sharing hash\n"));
4685 assert (SvTYPE(dstr) >= SVt_PV);
4687 HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)))));
4689 SvLEN_set(dstr, len);
4690 SvCUR_set(dstr, cur);
4693 /* Failed the swipe test, and we cannot do copy-on-write either.
4694 Have to copy the string. */
4695 SvGROW(dstr, cur + 1); /* inlined from sv_setpvn */
4696 Move(SvPVX_const(sstr),SvPVX(dstr),cur,char);
4697 SvCUR_set(dstr, cur);
4698 *SvEND(dstr) = '\0';
4700 if (sflags & SVp_NOK) {
4701 SvNV_set(dstr, SvNVX(sstr));
4703 if (sflags & SVp_IOK) {
4704 SvIV_set(dstr, SvIVX(sstr));
4705 if (sflags & SVf_IVisUV)
4708 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_NOK|SVp_NOK|SVf_UTF8);
4710 const MAGIC * const smg = SvVSTRING_mg(sstr);
4712 sv_magic(dstr, NULL, PERL_MAGIC_vstring,
4713 smg->mg_ptr, smg->mg_len);
4714 SvRMAGICAL_on(dstr);
4718 else if (sflags & (SVp_IOK|SVp_NOK)) {
4719 (void)SvOK_off(dstr);
4720 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_IVisUV|SVf_NOK|SVp_NOK);
4721 if (sflags & SVp_IOK) {
4722 /* XXXX Do we want to set IsUV for IV(ROK)? Be extra safe... */
4723 SvIV_set(dstr, SvIVX(sstr));
4725 if (sflags & SVp_NOK) {
4726 SvNV_set(dstr, SvNVX(sstr));
4730 if (isGV_with_GP(sstr)) {
4731 gv_efullname3(dstr, MUTABLE_GV(sstr), "*");
4734 (void)SvOK_off(dstr);
4736 if (SvTAINTED(sstr))
4742 =for apidoc sv_set_undef
4744 Equivalent to C<sv_setsv(sv, &PL_sv_undef)>, but more efficient.
4745 Doesn't handle set magic.
4747 The perl equivalent is C<$sv = undef;>. Note that it doesn't free any string
4748 buffer, unlike C<undef $sv>.
4750 Introduced in perl 5.25.12.
4756 Perl_sv_set_undef(pTHX_ SV *sv)
4758 U32 type = SvTYPE(sv);
4760 PERL_ARGS_ASSERT_SV_SET_UNDEF;
4762 /* shortcut, NULL, IV, RV */
4764 if (type <= SVt_IV) {
4765 assert(!SvGMAGICAL(sv));
4766 if (SvREADONLY(sv)) {
4767 /* does undeffing PL_sv_undef count as modifying a read-only
4768 * variable? Some XS code does this */
4769 if (sv == &PL_sv_undef)
4771 Perl_croak_no_modify();
4776 sv_unref_flags(sv, 0);
4779 SvFLAGS(sv) = type; /* quickly turn off all flags */
4780 SvREFCNT_dec_NN(rv);
4784 SvFLAGS(sv) = type; /* quickly turn off all flags */
4789 Perl_croak(aTHX_ "panic: attempt to undefine a freed scalar %p",
4792 SV_CHECK_THINKFIRST_COW_DROP(sv);
4794 if (isGV_with_GP(sv))
4795 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4796 "Undefined value assigned to typeglob");
4804 =for apidoc sv_setsv_mg
4806 Like C<sv_setsv>, but also handles 'set' magic.
4812 Perl_sv_setsv_mg(pTHX_ SV *const dstr, SV *const sstr)
4814 PERL_ARGS_ASSERT_SV_SETSV_MG;
4816 sv_setsv(dstr,sstr);
4821 # define SVt_COW SVt_PV
4823 Perl_sv_setsv_cow(pTHX_ SV *dstr, SV *sstr)
4825 STRLEN cur = SvCUR(sstr);
4826 STRLEN len = SvLEN(sstr);
4828 #if defined(PERL_DEBUG_READONLY_COW) && defined(PERL_COPY_ON_WRITE)
4829 const bool already = cBOOL(SvIsCOW(sstr));
4832 PERL_ARGS_ASSERT_SV_SETSV_COW;
4835 PerlIO_printf(Perl_debug_log, "Fast copy on write: %p -> %p\n",
4836 (void*)sstr, (void*)dstr);
4843 if (SvTHINKFIRST(dstr))
4844 sv_force_normal_flags(dstr, SV_COW_DROP_PV);
4845 else if (SvPVX_const(dstr))
4846 Safefree(SvPVX_mutable(dstr));
4850 SvUPGRADE(dstr, SVt_COW);
4852 assert (SvPOK(sstr));
4853 assert (SvPOKp(sstr));
4855 if (SvIsCOW(sstr)) {
4857 if (SvLEN(sstr) == 0) {
4858 /* source is a COW shared hash key. */
4859 DEBUG_C(PerlIO_printf(Perl_debug_log,
4860 "Fast copy on write: Sharing hash\n"));
4861 new_pv = HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr))));
4864 assert(SvCUR(sstr)+1 < SvLEN(sstr));
4865 assert(CowREFCNT(sstr) < SV_COW_REFCNT_MAX);
4867 assert ((SvFLAGS(sstr) & CAN_COW_MASK) == CAN_COW_FLAGS);
4868 SvUPGRADE(sstr, SVt_COW);
4870 DEBUG_C(PerlIO_printf(Perl_debug_log,
4871 "Fast copy on write: Converting sstr to COW\n"));
4872 CowREFCNT(sstr) = 0;
4874 # ifdef PERL_DEBUG_READONLY_COW
4875 if (already) sv_buf_to_rw(sstr);
4878 new_pv = SvPVX_mutable(sstr);
4882 SvPV_set(dstr, new_pv);
4883 SvFLAGS(dstr) = (SVt_COW|SVf_POK|SVp_POK|SVf_IsCOW);
4886 SvLEN_set(dstr, len);
4887 SvCUR_set(dstr, cur);
4897 =for apidoc sv_setpv_bufsize
4899 Sets the SV to be a string of cur bytes length, with at least
4900 len bytes available. Ensures that there is a null byte at SvEND.
4901 Returns a char * pointer to the SvPV buffer.
4907 Perl_sv_setpv_bufsize(pTHX_ SV *const sv, const STRLEN cur, const STRLEN len)
4911 PERL_ARGS_ASSERT_SV_SETPV_BUFSIZE;
4913 SV_CHECK_THINKFIRST_COW_DROP(sv);
4914 SvUPGRADE(sv, SVt_PV);
4915 pv = SvGROW(sv, len + 1);
4918 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4921 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4926 =for apidoc sv_setpvn
4928 Copies a string (possibly containing embedded C<NUL> characters) into an SV.
4929 The C<len> parameter indicates the number of
4930 bytes to be copied. If the C<ptr> argument is NULL the SV will become
4931 undefined. Does not handle 'set' magic. See C<L</sv_setpvn_mg>>.
4937 Perl_sv_setpvn(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4941 PERL_ARGS_ASSERT_SV_SETPVN;
4943 SV_CHECK_THINKFIRST_COW_DROP(sv);
4944 if (isGV_with_GP(sv))
4945 Perl_croak_no_modify();
4951 /* len is STRLEN which is unsigned, need to copy to signed */
4954 Perl_croak(aTHX_ "panic: sv_setpvn called with negative strlen %"
4957 SvUPGRADE(sv, SVt_PV);
4959 dptr = SvGROW(sv, len + 1);
4960 Move(ptr,dptr,len,char);
4963 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4965 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4969 =for apidoc sv_setpvn_mg
4971 Like C<sv_setpvn>, but also handles 'set' magic.
4977 Perl_sv_setpvn_mg(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4979 PERL_ARGS_ASSERT_SV_SETPVN_MG;
4981 sv_setpvn(sv,ptr,len);
4986 =for apidoc sv_setpv
4988 Copies a string into an SV. The string must be terminated with a C<NUL>
4989 character, and not contain embeded C<NUL>'s.
4990 Does not handle 'set' magic. See C<L</sv_setpv_mg>>.
4996 Perl_sv_setpv(pTHX_ SV *const sv, const char *const ptr)
5000 PERL_ARGS_ASSERT_SV_SETPV;
5002 SV_CHECK_THINKFIRST_COW_DROP(sv);
5008 SvUPGRADE(sv, SVt_PV);
5010 SvGROW(sv, len + 1);
5011 Move(ptr,SvPVX(sv),len+1,char);
5013 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5015 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
5019 =for apidoc sv_setpv_mg
5021 Like C<sv_setpv>, but also handles 'set' magic.
5027 Perl_sv_setpv_mg(pTHX_ SV *const sv, const char *const ptr)
5029 PERL_ARGS_ASSERT_SV_SETPV_MG;
5036 Perl_sv_sethek(pTHX_ SV *const sv, const HEK *const hek)
5038 PERL_ARGS_ASSERT_SV_SETHEK;
5044 if (HEK_LEN(hek) == HEf_SVKEY) {
5045 sv_setsv(sv, *(SV**)HEK_KEY(hek));
5048 const int flags = HEK_FLAGS(hek);
5049 if (flags & HVhek_WASUTF8) {
5050 STRLEN utf8_len = HEK_LEN(hek);
5051 char *as_utf8 = (char *)bytes_to_utf8((U8*)HEK_KEY(hek), &utf8_len);
5052 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
5055 } else if (flags & HVhek_UNSHARED) {
5056 sv_setpvn(sv, HEK_KEY(hek), HEK_LEN(hek));
5059 else SvUTF8_off(sv);
5063 SV_CHECK_THINKFIRST_COW_DROP(sv);
5064 SvUPGRADE(sv, SVt_PV);
5066 SvPV_set(sv,(char *)HEK_KEY(share_hek_hek(hek)));
5067 SvCUR_set(sv, HEK_LEN(hek));
5073 else SvUTF8_off(sv);
5081 =for apidoc sv_usepvn_flags
5083 Tells an SV to use C<ptr> to find its string value. Normally the
5084 string is stored inside the SV, but sv_usepvn allows the SV to use an
5085 outside string. C<ptr> should point to memory that was allocated
5086 by L<C<Newx>|perlclib/Memory Management and String Handling>. It must be
5087 the start of a C<Newx>-ed block of memory, and not a pointer to the
5088 middle of it (beware of L<C<OOK>|perlguts/Offsets> and copy-on-write),
5089 and not be from a non-C<Newx> memory allocator like C<malloc>. The
5090 string length, C<len>, must be supplied. By default this function
5091 will C<Renew> (i.e. realloc, move) the memory pointed to by C<ptr>,
5092 so that pointer should not be freed or used by the programmer after
5093 giving it to C<sv_usepvn>, and neither should any pointers from "behind"
5094 that pointer (e.g. ptr + 1) be used.
5096 If S<C<flags & SV_SMAGIC>> is true, will call C<SvSETMAGIC>. If
5097 S<C<flags & SV_HAS_TRAILING_NUL>> is true, then C<ptr[len]> must be C<NUL>,
5099 will be skipped (i.e. the buffer is actually at least 1 byte longer than
5100 C<len>, and already meets the requirements for storing in C<SvPVX>).
5106 Perl_sv_usepvn_flags(pTHX_ SV *const sv, char *ptr, const STRLEN len, const U32 flags)
5110 PERL_ARGS_ASSERT_SV_USEPVN_FLAGS;
5112 SV_CHECK_THINKFIRST_COW_DROP(sv);
5113 SvUPGRADE(sv, SVt_PV);
5116 if (flags & SV_SMAGIC)
5120 if (SvPVX_const(sv))
5124 if (flags & SV_HAS_TRAILING_NUL)
5125 assert(ptr[len] == '\0');
5128 allocate = (flags & SV_HAS_TRAILING_NUL)
5130 #ifdef Perl_safesysmalloc_size
5133 PERL_STRLEN_ROUNDUP(len + 1);
5135 if (flags & SV_HAS_TRAILING_NUL) {
5136 /* It's long enough - do nothing.
5137 Specifically Perl_newCONSTSUB is relying on this. */
5140 /* Force a move to shake out bugs in callers. */
5141 char *new_ptr = (char*)safemalloc(allocate);
5142 Copy(ptr, new_ptr, len, char);
5143 PoisonFree(ptr,len,char);
5147 ptr = (char*) saferealloc (ptr, allocate);
5150 #ifdef Perl_safesysmalloc_size
5151 SvLEN_set(sv, Perl_safesysmalloc_size(ptr));
5153 SvLEN_set(sv, allocate);
5157 if (!(flags & SV_HAS_TRAILING_NUL)) {
5160 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5162 if (flags & SV_SMAGIC)
5168 S_sv_uncow(pTHX_ SV * const sv, const U32 flags)
5170 assert(SvIsCOW(sv));
5173 const char * const pvx = SvPVX_const(sv);
5174 const STRLEN len = SvLEN(sv);
5175 const STRLEN cur = SvCUR(sv);
5179 PerlIO_printf(Perl_debug_log,
5180 "Copy on write: Force normal %ld\n",
5186 # ifdef PERL_COPY_ON_WRITE
5188 /* Must do this first, since the CowREFCNT uses SvPVX and
5189 we need to write to CowREFCNT, or de-RO the whole buffer if we are
5190 the only owner left of the buffer. */
5191 sv_buf_to_rw(sv); /* NOOP if RO-ing not supported */
5193 U8 cowrefcnt = CowREFCNT(sv);
5194 if(cowrefcnt != 0) {
5196 CowREFCNT(sv) = cowrefcnt;
5201 /* Else we are the only owner of the buffer. */
5206 /* This SV doesn't own the buffer, so need to Newx() a new one: */
5211 if (flags & SV_COW_DROP_PV) {
5212 /* OK, so we don't need to copy our buffer. */
5215 SvGROW(sv, cur + 1);
5216 Move(pvx,SvPVX(sv),cur,char);
5221 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5229 const char * const pvx = SvPVX_const(sv);
5230 const STRLEN len = SvCUR(sv);
5234 if (flags & SV_COW_DROP_PV) {
5235 /* OK, so we don't need to copy our buffer. */
5238 SvGROW(sv, len + 1);
5239 Move(pvx,SvPVX(sv),len,char);
5242 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5249 =for apidoc sv_force_normal_flags
5251 Undo various types of fakery on an SV, where fakery means
5252 "more than" a string: if the PV is a shared string, make
5253 a private copy; if we're a ref, stop refing; if we're a glob, downgrade to
5254 an C<xpvmg>; if we're a copy-on-write scalar, this is the on-write time when
5255 we do the copy, and is also used locally; if this is a
5256 vstring, drop the vstring magic. If C<SV_COW_DROP_PV> is set
5257 then a copy-on-write scalar drops its PV buffer (if any) and becomes
5258 C<SvPOK_off> rather than making a copy. (Used where this
5259 scalar is about to be set to some other value.) In addition,
5260 the C<flags> parameter gets passed to C<sv_unref_flags()>
5261 when unreffing. C<sv_force_normal> calls this function
5262 with flags set to 0.
5264 This function is expected to be used to signal to perl that this SV is
5265 about to be written to, and any extra book-keeping needs to be taken care
5266 of. Hence, it croaks on read-only values.
5272 Perl_sv_force_normal_flags(pTHX_ SV *const sv, const U32 flags)
5274 PERL_ARGS_ASSERT_SV_FORCE_NORMAL_FLAGS;
5277 Perl_croak_no_modify();
5278 else if (SvIsCOW(sv) && LIKELY(SvTYPE(sv) != SVt_PVHV))
5279 S_sv_uncow(aTHX_ sv, flags);
5281 sv_unref_flags(sv, flags);
5282 else if (SvFAKE(sv) && isGV_with_GP(sv))
5283 sv_unglob(sv, flags);
5284 else if (SvFAKE(sv) && isREGEXP(sv)) {
5285 /* Need to downgrade the REGEXP to a simple(r) scalar. This is analogous
5286 to sv_unglob. We only need it here, so inline it. */
5287 const bool islv = SvTYPE(sv) == SVt_PVLV;
5288 const svtype new_type =
5289 islv ? SVt_NULL : SvMAGIC(sv) || SvSTASH(sv) ? SVt_PVMG : SVt_PV;
5290 SV *const temp = newSV_type(new_type);
5291 regexp *old_rx_body;
5293 if (new_type == SVt_PVMG) {
5294 SvMAGIC_set(temp, SvMAGIC(sv));
5295 SvMAGIC_set(sv, NULL);
5296 SvSTASH_set(temp, SvSTASH(sv));
5297 SvSTASH_set(sv, NULL);
5300 SvCUR_set(temp, SvCUR(sv));
5301 /* Remember that SvPVX is in the head, not the body. */
5302 assert(ReANY((REGEXP *)sv)->mother_re);
5305 /* LV-as-regex has sv->sv_any pointing to an XPVLV body,
5306 * whose xpvlenu_rx field points to the regex body */
5307 XPV *xpv = (XPV*)(SvANY(sv));
5308 old_rx_body = xpv->xpv_len_u.xpvlenu_rx;
5309 xpv->xpv_len_u.xpvlenu_rx = NULL;
5312 old_rx_body = ReANY((REGEXP *)sv);
5314 /* Their buffer is already owned by someone else. */
5315 if (flags & SV_COW_DROP_PV) {
5316 /* SvLEN is already 0. For SVt_REGEXP, we have a brand new
5317 zeroed body. For SVt_PVLV, we zeroed it above (len field
5318 a union with xpvlenu_rx) */
5319 assert(!SvLEN(islv ? sv : temp));
5320 sv->sv_u.svu_pv = 0;
5323 sv->sv_u.svu_pv = savepvn(RX_WRAPPED((REGEXP *)sv), SvCUR(sv));
5324 SvLEN_set(islv ? sv : temp, SvCUR(sv)+1);
5328 /* Now swap the rest of the bodies. */
5332 SvFLAGS(sv) &= ~SVTYPEMASK;
5333 SvFLAGS(sv) |= new_type;
5334 SvANY(sv) = SvANY(temp);
5337 SvFLAGS(temp) &= ~(SVTYPEMASK);
5338 SvFLAGS(temp) |= SVt_REGEXP|SVf_FAKE;
5339 SvANY(temp) = old_rx_body;
5341 SvREFCNT_dec_NN(temp);
5343 else if (SvVOK(sv)) sv_unmagic(sv, PERL_MAGIC_vstring);
5349 Efficient removal of characters from the beginning of the string buffer.
5350 C<SvPOK(sv)>, or at least C<SvPOKp(sv)>, must be true and C<ptr> must be a
5351 pointer to somewhere inside the string buffer. C<ptr> becomes the first
5352 character of the adjusted string. Uses the C<OOK> hack. On return, only
5353 C<SvPOK(sv)> and C<SvPOKp(sv)> among the C<OK> flags will be true.
5355 Beware: after this function returns, C<ptr> and SvPVX_const(sv) may no longer
5356 refer to the same chunk of data.
5358 The unfortunate similarity of this function's name to that of Perl's C<chop>
5359 operator is strictly coincidental. This function works from the left;
5360 C<chop> works from the right.
5366 Perl_sv_chop(pTHX_ SV *const sv, const char *const ptr)
5377 PERL_ARGS_ASSERT_SV_CHOP;
5379 if (!ptr || !SvPOKp(sv))
5381 delta = ptr - SvPVX_const(sv);
5383 /* Nothing to do. */
5386 max_delta = SvLEN(sv) ? SvLEN(sv) : SvCUR(sv);
5387 if (delta > max_delta)
5388 Perl_croak(aTHX_ "panic: sv_chop ptr=%p, start=%p, end=%p",
5389 ptr, SvPVX_const(sv), SvPVX_const(sv) + max_delta);
5390 /* SvPVX(sv) may move in SV_CHECK_THINKFIRST(sv), so don't use ptr any more */
5391 SV_CHECK_THINKFIRST(sv);
5392 SvPOK_only_UTF8(sv);
5395 if (!SvLEN(sv)) { /* make copy of shared string */
5396 const char *pvx = SvPVX_const(sv);
5397 const STRLEN len = SvCUR(sv);
5398 SvGROW(sv, len + 1);
5399 Move(pvx,SvPVX(sv),len,char);
5405 SvOOK_offset(sv, old_delta);
5407 SvLEN_set(sv, SvLEN(sv) - delta);
5408 SvCUR_set(sv, SvCUR(sv) - delta);
5409 SvPV_set(sv, SvPVX(sv) + delta);
5411 p = (U8 *)SvPVX_const(sv);
5414 /* how many bytes were evacuated? we will fill them with sentinel
5415 bytes, except for the part holding the new offset of course. */
5418 evacn += (old_delta < 0x100 ? 1 : 1 + sizeof(STRLEN));
5420 assert(evacn <= delta + old_delta);
5424 /* This sets 'delta' to the accumulated value of all deltas so far */
5428 /* If 'delta' fits in a byte, store it just prior to the new beginning of
5429 * the string; otherwise store a 0 byte there and store 'delta' just prior
5430 * to that, using as many bytes as a STRLEN occupies. Thus it overwrites a
5431 * portion of the chopped part of the string */
5432 if (delta < 0x100) {
5436 p -= sizeof(STRLEN);
5437 Copy((U8*)&delta, p, sizeof(STRLEN), U8);
5441 /* Fill the preceding buffer with sentinals to verify that no-one is
5451 =for apidoc sv_catpvn
5453 Concatenates the string onto the end of the string which is in the SV.
5454 C<len> indicates number of bytes to copy. If the SV has the UTF-8
5455 status set, then the bytes appended should be valid UTF-8.
5456 Handles 'get' magic, but not 'set' magic. See C<L</sv_catpvn_mg>>.
5458 =for apidoc sv_catpvn_flags
5460 Concatenates the string onto the end of the string which is in the SV. The
5461 C<len> indicates number of bytes to copy.
5463 By default, the string appended is assumed to be valid UTF-8 if the SV has
5464 the UTF-8 status set, and a string of bytes otherwise. One can force the
5465 appended string to be interpreted as UTF-8 by supplying the C<SV_CATUTF8>
5466 flag, and as bytes by supplying the C<SV_CATBYTES> flag; the SV or the
5467 string appended will be upgraded to UTF-8 if necessary.
5469 If C<flags> has the C<SV_SMAGIC> bit set, will
5470 C<mg_set> on C<dsv> afterwards if appropriate.
5471 C<sv_catpvn> and C<sv_catpvn_nomg> are implemented
5472 in terms of this function.
5478 Perl_sv_catpvn_flags(pTHX_ SV *const dsv, const char *sstr, const STRLEN slen, const I32 flags)
5481 const char * const dstr = SvPV_force_flags(dsv, dlen, flags);
5483 PERL_ARGS_ASSERT_SV_CATPVN_FLAGS;
5484 assert((flags & (SV_CATBYTES|SV_CATUTF8)) != (SV_CATBYTES|SV_CATUTF8));
5486 if (!(flags & SV_CATBYTES) || !SvUTF8(dsv)) {
5487 if (flags & SV_CATUTF8 && !SvUTF8(dsv)) {
5488 sv_utf8_upgrade_flags_grow(dsv, 0, slen + 1);
5491 else SvGROW(dsv, dlen + slen + 3);
5493 sstr = SvPVX_const(dsv);
5494 Move(sstr, SvPVX(dsv) + dlen, slen, char);
5495 SvCUR_set(dsv, SvCUR(dsv) + slen);
5498 /* We inline bytes_to_utf8, to avoid an extra malloc. */
5499 const char * const send = sstr + slen;
5502 /* Something this code does not account for, which I think is
5503 impossible; it would require the same pv to be treated as
5504 bytes *and* utf8, which would indicate a bug elsewhere. */
5505 assert(sstr != dstr);
5507 SvGROW(dsv, dlen + slen * 2 + 3);
5508 d = (U8 *)SvPVX(dsv) + dlen;
5510 while (sstr < send) {
5511 append_utf8_from_native_byte(*sstr, &d);
5514 SvCUR_set(dsv, d-(const U8 *)SvPVX(dsv));
5517 (void)SvPOK_only_UTF8(dsv); /* validate pointer */
5519 if (flags & SV_SMAGIC)
5524 =for apidoc sv_catsv
5526 Concatenates the string from SV C<ssv> onto the end of the string in SV
5527 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5528 Handles 'get' magic on both SVs, but no 'set' magic. See C<L</sv_catsv_mg>>
5529 and C<L</sv_catsv_nomg>>.
5531 =for apidoc sv_catsv_flags
5533 Concatenates the string from SV C<ssv> onto the end of the string in SV
5534 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5535 If C<flags> has the C<SV_GMAGIC> bit set, will call C<mg_get> on both SVs if
5536 appropriate. If C<flags> has the C<SV_SMAGIC> bit set, C<mg_set> will be called on
5537 the modified SV afterward, if appropriate. C<sv_catsv>, C<sv_catsv_nomg>,
5538 and C<sv_catsv_mg> are implemented in terms of this function.
5543 Perl_sv_catsv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
5545 PERL_ARGS_ASSERT_SV_CATSV_FLAGS;
5549 const char *spv = SvPV_flags_const(ssv, slen, flags);
5550 if (flags & SV_GMAGIC)
5552 sv_catpvn_flags(dsv, spv, slen,
5553 DO_UTF8(ssv) ? SV_CATUTF8 : SV_CATBYTES);
5554 if (flags & SV_SMAGIC)
5560 =for apidoc sv_catpv
5562 Concatenates the C<NUL>-terminated string onto the end of the string which is
5564 If the SV has the UTF-8 status set, then the bytes appended should be
5565 valid UTF-8. Handles 'get' magic, but not 'set' magic. See
5571 Perl_sv_catpv(pTHX_ SV *const sv, const char *ptr)
5577 PERL_ARGS_ASSERT_SV_CATPV;
5581 junk = SvPV_force(sv, tlen);
5583 SvGROW(sv, tlen + len + 1);
5585 ptr = SvPVX_const(sv);
5586 Move(ptr,SvPVX(sv)+tlen,len+1,char);
5587 SvCUR_set(sv, SvCUR(sv) + len);
5588 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5593 =for apidoc sv_catpv_flags
5595 Concatenates the C<NUL>-terminated string onto the end of the string which is
5597 If the SV has the UTF-8 status set, then the bytes appended should
5598 be valid UTF-8. If C<flags> has the C<SV_SMAGIC> bit set, will C<mg_set>
5599 on the modified SV if appropriate.
5605 Perl_sv_catpv_flags(pTHX_ SV *dstr, const char *sstr, const I32 flags)
5607 PERL_ARGS_ASSERT_SV_CATPV_FLAGS;
5608 sv_catpvn_flags(dstr, sstr, strlen(sstr), flags);
5612 =for apidoc sv_catpv_mg
5614 Like C<sv_catpv>, but also handles 'set' magic.
5620 Perl_sv_catpv_mg(pTHX_ SV *const sv, const char *const ptr)
5622 PERL_ARGS_ASSERT_SV_CATPV_MG;
5631 Creates a new SV. A non-zero C<len> parameter indicates the number of
5632 bytes of preallocated string space the SV should have. An extra byte for a
5633 trailing C<NUL> is also reserved. (C<SvPOK> is not set for the SV even if string
5634 space is allocated.) The reference count for the new SV is set to 1.
5636 In 5.9.3, C<newSV()> replaces the older C<NEWSV()> API, and drops the first
5637 parameter, I<x>, a debug aid which allowed callers to identify themselves.
5638 This aid has been superseded by a new build option, C<PERL_MEM_LOG> (see
5639 L<perlhacktips/PERL_MEM_LOG>). The older API is still there for use in XS
5640 modules supporting older perls.
5646 Perl_newSV(pTHX_ const STRLEN len)
5652 sv_grow(sv, len + 1);
5657 =for apidoc sv_magicext
5659 Adds magic to an SV, upgrading it if necessary. Applies the
5660 supplied C<vtable> and returns a pointer to the magic added.
5662 Note that C<sv_magicext> will allow things that C<sv_magic> will not.
5663 In particular, you can add magic to C<SvREADONLY> SVs, and add more than
5664 one instance of the same C<how>.
5666 If C<namlen> is greater than zero then a C<savepvn> I<copy> of C<name> is
5667 stored, if C<namlen> is zero then C<name> is stored as-is and - as another
5668 special case - if C<(name && namlen == HEf_SVKEY)> then C<name> is assumed
5669 to contain an SV* and is stored as-is with its C<REFCNT> incremented.
5671 (This is now used as a subroutine by C<sv_magic>.)
5676 Perl_sv_magicext(pTHX_ SV *const sv, SV *const obj, const int how,
5677 const MGVTBL *const vtable, const char *const name, const I32 namlen)
5681 PERL_ARGS_ASSERT_SV_MAGICEXT;
5683 SvUPGRADE(sv, SVt_PVMG);
5684 Newxz(mg, 1, MAGIC);
5685 mg->mg_moremagic = SvMAGIC(sv);
5686 SvMAGIC_set(sv, mg);
5688 /* Sometimes a magic contains a reference loop, where the sv and
5689 object refer to each other. To prevent a reference loop that
5690 would prevent such objects being freed, we look for such loops
5691 and if we find one we avoid incrementing the object refcount.
5693 Note we cannot do this to avoid self-tie loops as intervening RV must
5694 have its REFCNT incremented to keep it in existence.
5697 if (!obj || obj == sv ||
5698 how == PERL_MAGIC_arylen ||
5699 how == PERL_MAGIC_regdata ||
5700 how == PERL_MAGIC_regdatum ||
5701 how == PERL_MAGIC_symtab ||
5702 (SvTYPE(obj) == SVt_PVGV &&
5703 (GvSV(obj) == sv || GvHV(obj) == (const HV *)sv
5704 || GvAV(obj) == (const AV *)sv || GvCV(obj) == (const CV *)sv
5705 || GvIOp(obj) == (const IO *)sv || GvFORM(obj) == (const CV *)sv)))
5710 mg->mg_obj = SvREFCNT_inc_simple(obj);
5711 mg->mg_flags |= MGf_REFCOUNTED;
5714 /* Normal self-ties simply pass a null object, and instead of
5715 using mg_obj directly, use the SvTIED_obj macro to produce a
5716 new RV as needed. For glob "self-ties", we are tieing the PVIO
5717 with an RV obj pointing to the glob containing the PVIO. In
5718 this case, to avoid a reference loop, we need to weaken the
5722 if (how == PERL_MAGIC_tiedscalar && SvTYPE(sv) == SVt_PVIO &&
5723 obj && SvROK(obj) && GvIO(SvRV(obj)) == (const IO *)sv)
5729 mg->mg_len = namlen;
5732 mg->mg_ptr = savepvn(name, namlen);
5733 else if (namlen == HEf_SVKEY) {
5734 /* Yes, this is casting away const. This is only for the case of
5735 HEf_SVKEY. I think we need to document this aberation of the
5736 constness of the API, rather than making name non-const, as
5737 that change propagating outwards a long way. */
5738 mg->mg_ptr = (char*)SvREFCNT_inc_simple_NN((SV *)name);
5740 mg->mg_ptr = (char *) name;
5742 mg->mg_virtual = (MGVTBL *) vtable;
5749 Perl_sv_magicext_mglob(pTHX_ SV *sv)
5751 PERL_ARGS_ASSERT_SV_MAGICEXT_MGLOB;
5752 if (SvTYPE(sv) == SVt_PVLV && LvTYPE(sv) == 'y') {
5753 /* This sv is only a delegate. //g magic must be attached to
5758 return sv_magicext(sv, NULL, PERL_MAGIC_regex_global,
5759 &PL_vtbl_mglob, 0, 0);
5763 =for apidoc sv_magic
5765 Adds magic to an SV. First upgrades C<sv> to type C<SVt_PVMG> if
5766 necessary, then adds a new magic item of type C<how> to the head of the
5769 See C<L</sv_magicext>> (which C<sv_magic> now calls) for a description of the
5770 handling of the C<name> and C<namlen> arguments.
5772 You need to use C<sv_magicext> to add magic to C<SvREADONLY> SVs and also
5773 to add more than one instance of the same C<how>.
5779 Perl_sv_magic(pTHX_ SV *const sv, SV *const obj, const int how,
5780 const char *const name, const I32 namlen)
5782 const MGVTBL *vtable;
5785 unsigned int vtable_index;
5787 PERL_ARGS_ASSERT_SV_MAGIC;
5789 if (how < 0 || (unsigned)how >= C_ARRAY_LENGTH(PL_magic_data)
5790 || ((flags = PL_magic_data[how]),
5791 (vtable_index = flags & PERL_MAGIC_VTABLE_MASK)
5792 > magic_vtable_max))
5793 Perl_croak(aTHX_ "Don't know how to handle magic of type \\%o", how);
5795 /* PERL_MAGIC_ext is reserved for use by extensions not perl internals.
5796 Useful for attaching extension internal data to perl vars.
5797 Note that multiple extensions may clash if magical scalars
5798 etc holding private data from one are passed to another. */
5800 vtable = (vtable_index == magic_vtable_max)
5801 ? NULL : PL_magic_vtables + vtable_index;
5803 if (SvREADONLY(sv)) {
5805 !PERL_MAGIC_TYPE_READONLY_ACCEPTABLE(how)
5808 Perl_croak_no_modify();
5811 if (SvMAGICAL(sv) || (how == PERL_MAGIC_taint && SvTYPE(sv) >= SVt_PVMG)) {
5812 if (SvMAGIC(sv) && (mg = mg_find(sv, how))) {
5813 /* sv_magic() refuses to add a magic of the same 'how' as an
5816 if (how == PERL_MAGIC_taint)
5822 /* Force pos to be stored as characters, not bytes. */
5823 if (SvMAGICAL(sv) && DO_UTF8(sv)
5824 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
5826 && mg->mg_flags & MGf_BYTES) {
5827 mg->mg_len = (SSize_t)sv_pos_b2u_flags(sv, (STRLEN)mg->mg_len,
5829 mg->mg_flags &= ~MGf_BYTES;
5832 /* Rest of work is done else where */
5833 mg = sv_magicext(sv,obj,how,vtable,name,namlen);
5836 case PERL_MAGIC_taint:
5839 case PERL_MAGIC_ext:
5840 case PERL_MAGIC_dbfile:
5847 S_sv_unmagicext_flags(pTHX_ SV *const sv, const int type, MGVTBL *vtbl, const U32 flags)
5854 if (SvTYPE(sv) < SVt_PVMG || !SvMAGIC(sv))
5856 mgp = &(((XPVMG*) SvANY(sv))->xmg_u.xmg_magic);
5857 for (mg = *mgp; mg; mg = *mgp) {
5858 const MGVTBL* const virt = mg->mg_virtual;
5859 if (mg->mg_type == type && (!flags || virt == vtbl)) {
5860 *mgp = mg->mg_moremagic;
5861 if (virt && virt->svt_free)
5862 virt->svt_free(aTHX_ sv, mg);
5863 if (mg->mg_ptr && mg->mg_type != PERL_MAGIC_regex_global) {
5865 Safefree(mg->mg_ptr);
5866 else if (mg->mg_len == HEf_SVKEY)
5867 SvREFCNT_dec(MUTABLE_SV(mg->mg_ptr));
5868 else if (mg->mg_type == PERL_MAGIC_utf8)
5869 Safefree(mg->mg_ptr);
5871 if (mg->mg_flags & MGf_REFCOUNTED)
5872 SvREFCNT_dec(mg->mg_obj);
5876 mgp = &mg->mg_moremagic;
5879 if (SvMAGICAL(sv)) /* if we're under save_magic, wait for restore_magic; */
5880 mg_magical(sv); /* else fix the flags now */
5889 =for apidoc sv_unmagic
5891 Removes all magic of type C<type> from an SV.
5897 Perl_sv_unmagic(pTHX_ SV *const sv, const int type)
5899 PERL_ARGS_ASSERT_SV_UNMAGIC;
5900 return S_sv_unmagicext_flags(aTHX_ sv, type, NULL, 0);
5904 =for apidoc sv_unmagicext
5906 Removes all magic of type C<type> with the specified C<vtbl> from an SV.
5912 Perl_sv_unmagicext(pTHX_ SV *const sv, const int type, MGVTBL *vtbl)
5914 PERL_ARGS_ASSERT_SV_UNMAGICEXT;
5915 return S_sv_unmagicext_flags(aTHX_ sv, type, vtbl, 1);
5919 =for apidoc sv_rvweaken
5921 Weaken a reference: set the C<SvWEAKREF> flag on this RV; give the
5922 referred-to SV C<PERL_MAGIC_backref> magic if it hasn't already; and
5923 push a back-reference to this RV onto the array of backreferences
5924 associated with that magic. If the RV is magical, set magic will be
5925 called after the RV is cleared. Silently ignores C<undef> and warns
5926 on already-weak references.
5932 Perl_sv_rvweaken(pTHX_ SV *const sv)
5936 PERL_ARGS_ASSERT_SV_RVWEAKEN;
5938 if (!SvOK(sv)) /* let undefs pass */
5941 Perl_croak(aTHX_ "Can't weaken a nonreference");
5942 else if (SvWEAKREF(sv)) {
5943 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is already weak");
5946 else if (SvREADONLY(sv)) croak_no_modify();
5948 Perl_sv_add_backref(aTHX_ tsv, sv);
5950 SvREFCNT_dec_NN(tsv);
5955 =for apidoc sv_rvunweaken
5957 Unweaken a reference: Clear the C<SvWEAKREF> flag on this RV; remove
5958 the backreference to this RV from the array of backreferences
5959 associated with the target SV, increment the refcount of the target.
5960 Silently ignores C<undef> and warns on non-weak references.
5966 Perl_sv_rvunweaken(pTHX_ SV *const sv)
5970 PERL_ARGS_ASSERT_SV_RVUNWEAKEN;
5972 if (!SvOK(sv)) /* let undefs pass */
5975 Perl_croak(aTHX_ "Can't unweaken a nonreference");
5976 else if (!SvWEAKREF(sv)) {
5977 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is not weak");
5980 else if (SvREADONLY(sv)) croak_no_modify();
5985 SvREFCNT_inc_NN(tsv);
5986 Perl_sv_del_backref(aTHX_ tsv, sv);
5991 =for apidoc sv_get_backrefs
5993 If C<sv> is the target of a weak reference then it returns the back
5994 references structure associated with the sv; otherwise return C<NULL>.
5996 When returning a non-null result the type of the return is relevant. If it
5997 is an AV then the elements of the AV are the weak reference RVs which
5998 point at this item. If it is any other type then the item itself is the
6001 See also C<Perl_sv_add_backref()>, C<Perl_sv_del_backref()>,
6002 C<Perl_sv_kill_backrefs()>
6008 Perl_sv_get_backrefs(SV *const sv)
6012 PERL_ARGS_ASSERT_SV_GET_BACKREFS;
6014 /* find slot to store array or singleton backref */
6016 if (SvTYPE(sv) == SVt_PVHV) {
6018 struct xpvhv_aux * const iter = HvAUX((HV *)sv);
6019 backrefs = (SV *)iter->xhv_backreferences;
6021 } else if (SvMAGICAL(sv)) {
6022 MAGIC *mg = mg_find(sv, PERL_MAGIC_backref);
6024 backrefs = mg->mg_obj;
6029 /* Give tsv backref magic if it hasn't already got it, then push a
6030 * back-reference to sv onto the array associated with the backref magic.
6032 * As an optimisation, if there's only one backref and it's not an AV,
6033 * store it directly in the HvAUX or mg_obj slot, avoiding the need to
6034 * allocate an AV. (Whether the slot holds an AV tells us whether this is
6038 /* A discussion about the backreferences array and its refcount:
6040 * The AV holding the backreferences is pointed to either as the mg_obj of
6041 * PERL_MAGIC_backref, or in the specific case of a HV, from the
6042 * xhv_backreferences field. The array is created with a refcount
6043 * of 2. This means that if during global destruction the array gets
6044 * picked on before its parent to have its refcount decremented by the
6045 * random zapper, it won't actually be freed, meaning it's still there for
6046 * when its parent gets freed.
6048 * When the parent SV is freed, the extra ref is killed by
6049 * Perl_sv_kill_backrefs. The other ref is killed, in the case of magic,
6050 * by mg_free() / MGf_REFCOUNTED, or for a hash, by Perl_hv_kill_backrefs.
6052 * When a single backref SV is stored directly, it is not reference
6057 Perl_sv_add_backref(pTHX_ SV *const tsv, SV *const sv)
6063 PERL_ARGS_ASSERT_SV_ADD_BACKREF;
6065 /* find slot to store array or singleton backref */
6067 if (SvTYPE(tsv) == SVt_PVHV) {
6068 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
6071 mg = mg_find(tsv, PERL_MAGIC_backref);
6073 mg = sv_magicext(tsv, NULL, PERL_MAGIC_backref, &PL_vtbl_backref, NULL, 0);
6074 svp = &(mg->mg_obj);
6077 /* create or retrieve the array */
6079 if ( (!*svp && SvTYPE(sv) == SVt_PVAV)
6080 || (*svp && SvTYPE(*svp) != SVt_PVAV)
6084 mg->mg_flags |= MGf_REFCOUNTED;
6087 SvREFCNT_inc_simple_void_NN(av);
6088 /* av now has a refcnt of 2; see discussion above */
6089 av_extend(av, *svp ? 2 : 1);
6091 /* move single existing backref to the array */
6092 AvARRAY(av)[++AvFILLp(av)] = *svp; /* av_push() */
6097 av = MUTABLE_AV(*svp);
6099 /* optimisation: store single backref directly in HvAUX or mg_obj */
6103 assert(SvTYPE(av) == SVt_PVAV);
6104 if (AvFILLp(av) >= AvMAX(av)) {
6105 av_extend(av, AvFILLp(av)+1);
6108 /* push new backref */
6109 AvARRAY(av)[++AvFILLp(av)] = sv; /* av_push() */
6112 /* delete a back-reference to ourselves from the backref magic associated
6113 * with the SV we point to.
6117 Perl_sv_del_backref(pTHX_ SV *const tsv, SV *const sv)
6121 PERL_ARGS_ASSERT_SV_DEL_BACKREF;
6123 if (SvTYPE(tsv) == SVt_PVHV) {
6125 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
6127 else if (SvIS_FREED(tsv) && PL_phase == PERL_PHASE_DESTRUCT) {
6128 /* It's possible for the the last (strong) reference to tsv to have
6129 become freed *before* the last thing holding a weak reference.
6130 If both survive longer than the backreferences array, then when
6131 the referent's reference count drops to 0 and it is freed, it's
6132 not able to chase the backreferences, so they aren't NULLed.
6134 For example, a CV holds a weak reference to its stash. If both the
6135 CV and the stash survive longer than the backreferences array,
6136 and the CV gets picked for the SvBREAK() treatment first,
6137 *and* it turns out that the stash is only being kept alive because
6138 of an our variable in the pad of the CV, then midway during CV
6139 destruction the stash gets freed, but CvSTASH() isn't set to NULL.
6140 It ends up pointing to the freed HV. Hence it's chased in here, and
6141 if this block wasn't here, it would hit the !svp panic just below.
6143 I don't believe that "better" destruction ordering is going to help
6144 here - during global destruction there's always going to be the
6145 chance that something goes out of order. We've tried to make it
6146 foolproof before, and it only resulted in evolutionary pressure on
6147 fools. Which made us look foolish for our hubris. :-(
6153 = SvMAGICAL(tsv) ? mg_find(tsv, PERL_MAGIC_backref) : NULL;
6154 svp = mg ? &(mg->mg_obj) : NULL;
6158 Perl_croak(aTHX_ "panic: del_backref, svp=0");
6160 /* It's possible that sv is being freed recursively part way through the
6161 freeing of tsv. If this happens, the backreferences array of tsv has
6162 already been freed, and so svp will be NULL. If this is the case,
6163 we should not panic. Instead, nothing needs doing, so return. */
6164 if (PL_phase == PERL_PHASE_DESTRUCT && SvREFCNT(tsv) == 0)
6166 Perl_croak(aTHX_ "panic: del_backref, *svp=%p phase=%s refcnt=%" UVuf,
6167 (void*)*svp, PL_phase_names[PL_phase], (UV)SvREFCNT(tsv));
6170 if (SvTYPE(*svp) == SVt_PVAV) {
6174 AV * const av = (AV*)*svp;
6176 assert(!SvIS_FREED(av));
6180 /* for an SV with N weak references to it, if all those
6181 * weak refs are deleted, then sv_del_backref will be called
6182 * N times and O(N^2) compares will be done within the backref
6183 * array. To ameliorate this potential slowness, we:
6184 * 1) make sure this code is as tight as possible;
6185 * 2) when looking for SV, look for it at both the head and tail of the
6186 * array first before searching the rest, since some create/destroy
6187 * patterns will cause the backrefs to be freed in order.
6194 SV **p = &svp[fill];
6195 SV *const topsv = *p;
6202 /* We weren't the last entry.
6203 An unordered list has this property that you
6204 can take the last element off the end to fill
6205 the hole, and it's still an unordered list :-)
6211 break; /* should only be one */
6218 AvFILLp(av) = fill-1;
6220 else if (SvIS_FREED(*svp) && PL_phase == PERL_PHASE_DESTRUCT) {
6221 /* freed AV; skip */
6224 /* optimisation: only a single backref, stored directly */
6226 Perl_croak(aTHX_ "panic: del_backref, *svp=%p, sv=%p",
6227 (void*)*svp, (void*)sv);
6234 Perl_sv_kill_backrefs(pTHX_ SV *const sv, AV *const av)
6240 PERL_ARGS_ASSERT_SV_KILL_BACKREFS;
6245 /* after multiple passes through Perl_sv_clean_all() for a thingy
6246 * that has badly leaked, the backref array may have gotten freed,
6247 * since we only protect it against 1 round of cleanup */
6248 if (SvIS_FREED(av)) {
6249 if (PL_in_clean_all) /* All is fair */
6252 "panic: magic_killbackrefs (freed backref AV/SV)");
6256 is_array = (SvTYPE(av) == SVt_PVAV);
6258 assert(!SvIS_FREED(av));
6261 last = svp + AvFILLp(av);
6264 /* optimisation: only a single backref, stored directly */
6270 while (svp <= last) {
6272 SV *const referrer = *svp;
6273 if (SvWEAKREF(referrer)) {
6274 /* XXX Should we check that it hasn't changed? */
6275 assert(SvROK(referrer));
6276 SvRV_set(referrer, 0);
6278 SvWEAKREF_off(referrer);
6279 SvSETMAGIC(referrer);
6280 } else if (SvTYPE(referrer) == SVt_PVGV ||
6281 SvTYPE(referrer) == SVt_PVLV) {
6282 assert(SvTYPE(sv) == SVt_PVHV); /* stash backref */
6283 /* You lookin' at me? */
6284 assert(GvSTASH(referrer));
6285 assert(GvSTASH(referrer) == (const HV *)sv);
6286 GvSTASH(referrer) = 0;
6287 } else if (SvTYPE(referrer) == SVt_PVCV ||
6288 SvTYPE(referrer) == SVt_PVFM) {
6289 if (SvTYPE(sv) == SVt_PVHV) { /* stash backref */
6290 /* You lookin' at me? */
6291 assert(CvSTASH(referrer));
6292 assert(CvSTASH(referrer) == (const HV *)sv);
6293 SvANY(MUTABLE_CV(referrer))->xcv_stash = 0;
6296 assert(SvTYPE(sv) == SVt_PVGV);
6297 /* You lookin' at me? */
6298 assert(CvGV(referrer));
6299 assert(CvGV(referrer) == (const GV *)sv);
6300 anonymise_cv_maybe(MUTABLE_GV(sv),
6301 MUTABLE_CV(referrer));
6306 "panic: magic_killbackrefs (flags=%" UVxf ")",
6307 (UV)SvFLAGS(referrer));
6318 SvREFCNT_dec_NN(av); /* remove extra count added by sv_add_backref() */
6324 =for apidoc sv_insert
6326 Inserts and/or replaces a string at the specified offset/length within the SV.
6327 Similar to the Perl C<substr()> function, with C<littlelen> bytes starting at
6328 C<little> replacing C<len> bytes of the string in C<bigstr> starting at
6329 C<offset>. Handles get magic.
6331 =for apidoc sv_insert_flags
6333 Same as C<sv_insert>, but the extra C<flags> are passed to the
6334 C<SvPV_force_flags> that applies to C<bigstr>.
6340 Perl_sv_insert_flags(pTHX_ SV *const bigstr, const STRLEN offset, const STRLEN len, const char *little, const STRLEN littlelen, const U32 flags)
6346 SSize_t i; /* better be sizeof(STRLEN) or bad things happen */
6349 PERL_ARGS_ASSERT_SV_INSERT_FLAGS;
6351 SvPV_force_flags(bigstr, curlen, flags);
6352 (void)SvPOK_only_UTF8(bigstr);
6354 if (little >= SvPVX(bigstr) &&
6355 little < SvPVX(bigstr) + (SvLEN(bigstr) ? SvLEN(bigstr) : SvCUR(bigstr))) {
6356 /* little is a pointer to within bigstr, since we can reallocate bigstr,
6357 or little...little+littlelen might overlap offset...offset+len we make a copy
6359 little = savepvn(little, littlelen);
6363 if (offset + len > curlen) {
6364 SvGROW(bigstr, offset+len+1);
6365 Zero(SvPVX(bigstr)+curlen, offset+len-curlen, char);
6366 SvCUR_set(bigstr, offset+len);
6370 i = littlelen - len;
6371 if (i > 0) { /* string might grow */
6372 big = SvGROW(bigstr, SvCUR(bigstr) + i + 1);
6373 mid = big + offset + len;
6374 midend = bigend = big + SvCUR(bigstr);
6377 while (midend > mid) /* shove everything down */
6378 *--bigend = *--midend;
6379 Move(little,big+offset,littlelen,char);
6380 SvCUR_set(bigstr, SvCUR(bigstr) + i);
6385 Move(little,SvPVX(bigstr)+offset,len,char);
6390 big = SvPVX(bigstr);
6393 bigend = big + SvCUR(bigstr);
6395 if (midend > bigend)
6396 Perl_croak(aTHX_ "panic: sv_insert, midend=%p, bigend=%p",
6399 if (mid - big > bigend - midend) { /* faster to shorten from end */
6401 Move(little, mid, littlelen,char);
6404 i = bigend - midend;
6406 Move(midend, mid, i,char);
6410 SvCUR_set(bigstr, mid - big);
6412 else if ((i = mid - big)) { /* faster from front */
6413 midend -= littlelen;
6415 Move(big, midend - i, i, char);
6416 sv_chop(bigstr,midend-i);
6418 Move(little, mid, littlelen,char);
6420 else if (littlelen) {
6421 midend -= littlelen;
6422 sv_chop(bigstr,midend);
6423 Move(little,midend,littlelen,char);
6426 sv_chop(bigstr,midend);
6432 =for apidoc sv_replace
6434 Make the first argument a copy of the second, then delete the original.
6435 The target SV physically takes over ownership of the body of the source SV
6436 and inherits its flags; however, the target keeps any magic it owns,
6437 and any magic in the source is discarded.
6438 Note that this is a rather specialist SV copying operation; most of the
6439 time you'll want to use C<sv_setsv> or one of its many macro front-ends.
6445 Perl_sv_replace(pTHX_ SV *const sv, SV *const nsv)
6447 const U32 refcnt = SvREFCNT(sv);
6449 PERL_ARGS_ASSERT_SV_REPLACE;
6451 SV_CHECK_THINKFIRST_COW_DROP(sv);
6452 if (SvREFCNT(nsv) != 1) {
6453 Perl_croak(aTHX_ "panic: reference miscount on nsv in sv_replace()"
6454 " (%" UVuf " != 1)", (UV) SvREFCNT(nsv));
6456 if (SvMAGICAL(sv)) {
6460 sv_upgrade(nsv, SVt_PVMG);
6461 SvMAGIC_set(nsv, SvMAGIC(sv));
6462 SvFLAGS(nsv) |= SvMAGICAL(sv);
6464 SvMAGIC_set(sv, NULL);
6468 assert(!SvREFCNT(sv));
6469 #ifdef DEBUG_LEAKING_SCALARS
6470 sv->sv_flags = nsv->sv_flags;
6471 sv->sv_any = nsv->sv_any;
6472 sv->sv_refcnt = nsv->sv_refcnt;
6473 sv->sv_u = nsv->sv_u;
6475 StructCopy(nsv,sv,SV);
6477 if(SvTYPE(sv) == SVt_IV) {
6478 SET_SVANY_FOR_BODYLESS_IV(sv);
6482 SvREFCNT(sv) = refcnt;
6483 SvFLAGS(nsv) |= SVTYPEMASK; /* Mark as freed */
6488 /* We're about to free a GV which has a CV that refers back to us.
6489 * If that CV will outlive us, make it anonymous (i.e. fix up its CvGV
6493 S_anonymise_cv_maybe(pTHX_ GV *gv, CV* cv)
6498 PERL_ARGS_ASSERT_ANONYMISE_CV_MAYBE;
6501 assert(SvREFCNT(gv) == 0);
6502 assert(isGV(gv) && isGV_with_GP(gv));
6504 assert(!CvANON(cv));
6505 assert(CvGV(cv) == gv);
6506 assert(!CvNAMED(cv));
6508 /* will the CV shortly be freed by gp_free() ? */
6509 if (GvCV(gv) == cv && GvGP(gv)->gp_refcnt < 2 && SvREFCNT(cv) < 2) {
6510 SvANY(cv)->xcv_gv_u.xcv_gv = NULL;
6514 /* if not, anonymise: */
6515 gvname = (GvSTASH(gv) && HvNAME(GvSTASH(gv)) && HvENAME(GvSTASH(gv)))
6516 ? newSVhek(HvENAME_HEK(GvSTASH(gv)))
6517 : newSVpvn_flags( "__ANON__", 8, 0 );
6518 sv_catpvs(gvname, "::__ANON__");
6519 anongv = gv_fetchsv(gvname, GV_ADDMULTI, SVt_PVCV);
6520 SvREFCNT_dec_NN(gvname);
6524 SvANY(cv)->xcv_gv_u.xcv_gv = MUTABLE_GV(SvREFCNT_inc(anongv));
6529 =for apidoc sv_clear
6531 Clear an SV: call any destructors, free up any memory used by the body,
6532 and free the body itself. The SV's head is I<not> freed, although
6533 its type is set to all 1's so that it won't inadvertently be assumed
6534 to be live during global destruction etc.
6535 This function should only be called when C<REFCNT> is zero. Most of the time
6536 you'll want to call C<sv_free()> (or its macro wrapper C<SvREFCNT_dec>)
6543 Perl_sv_clear(pTHX_ SV *const orig_sv)
6548 const struct body_details *sv_type_details;
6552 STRLEN hash_index = 0; /* initialise to make Coverity et al happy.
6553 Not strictly necessary */
6555 PERL_ARGS_ASSERT_SV_CLEAR;
6557 /* within this loop, sv is the SV currently being freed, and
6558 * iter_sv is the most recent AV or whatever that's being iterated
6559 * over to provide more SVs */
6565 assert(SvREFCNT(sv) == 0);
6566 assert(SvTYPE(sv) != (svtype)SVTYPEMASK);
6568 if (type <= SVt_IV) {
6569 /* See the comment in sv.h about the collusion between this
6570 * early return and the overloading of the NULL slots in the
6574 SvFLAGS(sv) &= SVf_BREAK;
6575 SvFLAGS(sv) |= SVTYPEMASK;
6579 /* objs are always >= MG, but pad names use the SVs_OBJECT flag
6580 for another purpose */
6581 assert(!SvOBJECT(sv) || type >= SVt_PVMG);
6583 if (type >= SVt_PVMG) {
6585 if (!curse(sv, 1)) goto get_next_sv;
6586 type = SvTYPE(sv); /* destructor may have changed it */
6588 /* Free back-references before magic, in case the magic calls
6589 * Perl code that has weak references to sv. */
6590 if (type == SVt_PVHV) {
6591 Perl_hv_kill_backrefs(aTHX_ MUTABLE_HV(sv));
6595 else if (SvMAGIC(sv)) {
6596 /* Free back-references before other types of magic. */
6597 sv_unmagic(sv, PERL_MAGIC_backref);
6603 /* case SVt_INVLIST: */
6606 IoIFP(sv) != PerlIO_stdin() &&
6607 IoIFP(sv) != PerlIO_stdout() &&
6608 IoIFP(sv) != PerlIO_stderr() &&
6609 !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6611 io_close(MUTABLE_IO(sv), NULL, FALSE,
6612 (IoTYPE(sv) == IoTYPE_WRONLY ||
6613 IoTYPE(sv) == IoTYPE_RDWR ||
6614 IoTYPE(sv) == IoTYPE_APPEND));
6616 if (IoDIRP(sv) && !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6617 PerlDir_close(IoDIRP(sv));
6618 IoDIRP(sv) = (DIR*)NULL;
6619 Safefree(IoTOP_NAME(sv));
6620 Safefree(IoFMT_NAME(sv));
6621 Safefree(IoBOTTOM_NAME(sv));
6622 if ((const GV *)sv == PL_statgv)
6626 /* FIXME for plugins */
6627 pregfree2((REGEXP*) sv);
6631 cv_undef(MUTABLE_CV(sv));
6632 /* If we're in a stash, we don't own a reference to it.
6633 * However it does have a back reference to us, which needs to
6635 if ((stash = CvSTASH(sv)))
6636 sv_del_backref(MUTABLE_SV(stash), sv);
6639 if (PL_last_swash_hv == (const HV *)sv) {
6640 PL_last_swash_hv = NULL;
6642 if (HvTOTALKEYS((HV*)sv) > 0) {
6644 /* this statement should match the one at the beginning of
6645 * hv_undef_flags() */
6646 if ( PL_phase != PERL_PHASE_DESTRUCT
6647 && (hek = HvNAME_HEK((HV*)sv)))
6649 if (PL_stashcache) {
6650 DEBUG_o(Perl_deb(aTHX_
6651 "sv_clear clearing PL_stashcache for '%" HEKf
6654 (void)hv_deletehek(PL_stashcache,
6657 hv_name_set((HV*)sv, NULL, 0, 0);
6660 /* save old iter_sv in unused SvSTASH field */
6661 assert(!SvOBJECT(sv));
6662 SvSTASH(sv) = (HV*)iter_sv;
6665 /* save old hash_index in unused SvMAGIC field */
6666 assert(!SvMAGICAL(sv));
6667 assert(!SvMAGIC(sv));
6668 ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index = hash_index;
6671 next_sv = Perl_hfree_next_entry(aTHX_ (HV*)sv, &hash_index);
6672 goto get_next_sv; /* process this new sv */
6674 /* free empty hash */
6675 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6676 assert(!HvARRAY((HV*)sv));
6680 AV* av = MUTABLE_AV(sv);
6681 if (PL_comppad == av) {
6685 if (AvREAL(av) && AvFILLp(av) > -1) {
6686 next_sv = AvARRAY(av)[AvFILLp(av)--];
6687 /* save old iter_sv in top-most slot of AV,
6688 * and pray that it doesn't get wiped in the meantime */
6689 AvARRAY(av)[AvMAX(av)] = iter_sv;
6691 goto get_next_sv; /* process this new sv */
6693 Safefree(AvALLOC(av));
6698 if (LvTYPE(sv) == 'T') { /* for tie: return HE to pool */
6699 SvREFCNT_dec(HeKEY_sv((HE*)LvTARG(sv)));
6700 HeNEXT((HE*)LvTARG(sv)) = PL_hv_fetch_ent_mh;
6701 PL_hv_fetch_ent_mh = (HE*)LvTARG(sv);
6703 else if (LvTYPE(sv) != 't') /* unless tie: unrefcnted fake SV** */
6704 SvREFCNT_dec(LvTARG(sv));
6706 /* SvLEN points to a regex body. Free the body, then
6707 * set SvLEN to whatever value was in the now-freed
6708 * regex body. The PVX buffer is shared by multiple re's
6709 * and only freed once, by the re whose len in non-null */
6710 STRLEN len = ReANY(sv)->xpv_len;
6711 pregfree2((REGEXP*) sv);
6712 SvLEN_set((sv), len);
6717 if (isGV_with_GP(sv)) {
6718 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
6719 && HvENAME_get(stash))
6720 mro_method_changed_in(stash);
6721 gp_free(MUTABLE_GV(sv));
6723 unshare_hek(GvNAME_HEK(sv));
6724 /* If we're in a stash, we don't own a reference to it.
6725 * However it does have a back reference to us, which
6726 * needs to be cleared. */
6727 if ((stash = GvSTASH(sv)))
6728 sv_del_backref(MUTABLE_SV(stash), sv);
6730 /* FIXME. There are probably more unreferenced pointers to SVs
6731 * in the interpreter struct that we should check and tidy in
6732 * a similar fashion to this: */
6733 /* See also S_sv_unglob, which does the same thing. */
6734 if ((const GV *)sv == PL_last_in_gv)
6735 PL_last_in_gv = NULL;
6736 else if ((const GV *)sv == PL_statgv)
6738 else if ((const GV *)sv == PL_stderrgv)
6747 /* Don't bother with SvOOK_off(sv); as we're only going to
6751 SvOOK_offset(sv, offset);
6752 SvPV_set(sv, SvPVX_mutable(sv) - offset);
6753 /* Don't even bother with turning off the OOK flag. */
6758 SV * const target = SvRV(sv);
6760 sv_del_backref(target, sv);
6766 else if (SvPVX_const(sv)
6767 && !(SvTYPE(sv) == SVt_PVIO
6768 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6773 PerlIO_printf(Perl_debug_log, "Copy on write: clear\n");
6778 if (CowREFCNT(sv)) {
6785 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6790 Safefree(SvPVX_mutable(sv));
6794 else if (SvPVX_const(sv) && SvLEN(sv)
6795 && !(SvTYPE(sv) == SVt_PVIO
6796 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6797 Safefree(SvPVX_mutable(sv));
6798 else if (SvPVX_const(sv) && SvIsCOW(sv)) {
6799 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6809 SvFLAGS(sv) &= SVf_BREAK;
6810 SvFLAGS(sv) |= SVTYPEMASK;
6812 sv_type_details = bodies_by_type + type;
6813 if (sv_type_details->arena) {
6814 del_body(((char *)SvANY(sv) + sv_type_details->offset),
6815 &PL_body_roots[type]);
6817 else if (sv_type_details->body_size) {
6818 safefree(SvANY(sv));
6822 /* caller is responsible for freeing the head of the original sv */
6823 if (sv != orig_sv && !SvREFCNT(sv))
6826 /* grab and free next sv, if any */
6834 else if (!iter_sv) {
6836 } else if (SvTYPE(iter_sv) == SVt_PVAV) {
6837 AV *const av = (AV*)iter_sv;
6838 if (AvFILLp(av) > -1) {
6839 sv = AvARRAY(av)[AvFILLp(av)--];
6841 else { /* no more elements of current AV to free */
6844 /* restore previous value, squirrelled away */
6845 iter_sv = AvARRAY(av)[AvMAX(av)];
6846 Safefree(AvALLOC(av));
6849 } else if (SvTYPE(iter_sv) == SVt_PVHV) {
6850 sv = Perl_hfree_next_entry(aTHX_ (HV*)iter_sv, &hash_index);
6851 if (!sv && !HvTOTALKEYS((HV *)iter_sv)) {
6852 /* no more elements of current HV to free */
6855 /* Restore previous values of iter_sv and hash_index,
6856 * squirrelled away */
6857 assert(!SvOBJECT(sv));
6858 iter_sv = (SV*)SvSTASH(sv);
6859 assert(!SvMAGICAL(sv));
6860 hash_index = ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index;
6862 /* perl -DA does not like rubbish in SvMAGIC. */
6866 /* free any remaining detritus from the hash struct */
6867 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6868 assert(!HvARRAY((HV*)sv));
6873 /* unrolled SvREFCNT_dec and sv_free2 follows: */
6877 if (!SvREFCNT(sv)) {
6881 if (--(SvREFCNT(sv)))
6885 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6886 "Attempt to free temp prematurely: SV 0x%" UVxf
6887 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6891 if (SvIMMORTAL(sv)) {
6892 /* make sure SvREFCNT(sv)==0 happens very seldom */
6893 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6902 /* This routine curses the sv itself, not the object referenced by sv. So
6903 sv does not have to be ROK. */
6906 S_curse(pTHX_ SV * const sv, const bool check_refcnt) {
6907 PERL_ARGS_ASSERT_CURSE;
6908 assert(SvOBJECT(sv));
6910 if (PL_defstash && /* Still have a symbol table? */
6916 stash = SvSTASH(sv);
6917 assert(SvTYPE(stash) == SVt_PVHV);
6918 if (HvNAME(stash)) {
6919 CV* destructor = NULL;
6920 struct mro_meta *meta;
6922 assert (SvOOK(stash));
6924 DEBUG_o( Perl_deb(aTHX_ "Looking for DESTROY method for %s\n",
6927 /* don't make this an initialization above the assert, since it needs
6929 meta = HvMROMETA(stash);
6930 if (meta->destroy_gen && meta->destroy_gen == PL_sub_generation) {
6931 destructor = meta->destroy;
6932 DEBUG_o( Perl_deb(aTHX_ "Using cached DESTROY method %p for %s\n",
6933 (void *)destructor, HvNAME(stash)) );
6936 bool autoload = FALSE;
6938 gv_fetchmeth_pvn(stash, S_destroy, S_destroy_len, -1, 0);
6940 destructor = GvCV(gv);
6942 gv = gv_autoload_pvn(stash, S_destroy, S_destroy_len,
6943 GV_AUTOLOAD_ISMETHOD);
6945 destructor = GvCV(gv);
6949 /* we don't cache AUTOLOAD for DESTROY, since this code
6950 would then need to set $__PACKAGE__::AUTOLOAD, or the
6951 equivalent for XS AUTOLOADs */
6953 meta->destroy_gen = PL_sub_generation;
6954 meta->destroy = destructor;
6956 DEBUG_o( Perl_deb(aTHX_ "Set cached DESTROY method %p for %s\n",
6957 (void *)destructor, HvNAME(stash)) );
6960 DEBUG_o( Perl_deb(aTHX_ "Not caching AUTOLOAD for DESTROY method for %s\n",
6964 assert(!destructor || SvTYPE(destructor) == SVt_PVCV);
6966 /* A constant subroutine can have no side effects, so
6967 don't bother calling it. */
6968 && !CvCONST(destructor)
6969 /* Don't bother calling an empty destructor or one that
6970 returns immediately. */
6971 && (CvISXSUB(destructor)
6972 || (CvSTART(destructor)
6973 && (CvSTART(destructor)->op_next->op_type
6975 && (CvSTART(destructor)->op_next->op_type
6977 || CvSTART(destructor)->op_next->op_next->op_type
6983 SV* const tmpref = newRV(sv);
6984 SvREADONLY_on(tmpref); /* DESTROY() could be naughty */
6986 PUSHSTACKi(PERLSI_DESTROY);
6991 call_sv(MUTABLE_SV(destructor),
6992 G_DISCARD|G_EVAL|G_KEEPERR|G_VOID);
6996 if(SvREFCNT(tmpref) < 2) {
6997 /* tmpref is not kept alive! */
6999 SvRV_set(tmpref, NULL);
7002 SvREFCNT_dec_NN(tmpref);
7005 } while (SvOBJECT(sv) && SvSTASH(sv) != stash);
7008 if (check_refcnt && SvREFCNT(sv)) {
7009 if (PL_in_clean_objs)
7011 "DESTROY created new reference to dead object '%" HEKf "'",
7012 HEKfARG(HvNAME_HEK(stash)));
7013 /* DESTROY gave object new lease on life */
7019 HV * const stash = SvSTASH(sv);
7020 /* Curse before freeing the stash, as freeing the stash could cause
7021 a recursive call into S_curse. */
7022 SvOBJECT_off(sv); /* Curse the object. */
7023 SvSTASH_set(sv,0); /* SvREFCNT_dec may try to read this */
7024 SvREFCNT_dec(stash); /* possibly of changed persuasion */
7030 =for apidoc sv_newref
7032 Increment an SV's reference count. Use the C<SvREFCNT_inc()> wrapper
7039 Perl_sv_newref(pTHX_ SV *const sv)
7041 PERL_UNUSED_CONTEXT;
7050 Decrement an SV's reference count, and if it drops to zero, call
7051 C<sv_clear> to invoke destructors and free up any memory used by
7052 the body; finally, deallocating the SV's head itself.
7053 Normally called via a wrapper macro C<SvREFCNT_dec>.
7059 Perl_sv_free(pTHX_ SV *const sv)
7065 /* Private helper function for SvREFCNT_dec().
7066 * Called with rc set to original SvREFCNT(sv), where rc == 0 or 1 */
7069 Perl_sv_free2(pTHX_ SV *const sv, const U32 rc)
7073 PERL_ARGS_ASSERT_SV_FREE2;
7075 if (LIKELY( rc == 1 )) {
7081 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
7082 "Attempt to free temp prematurely: SV 0x%" UVxf
7083 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
7087 if (SvIMMORTAL(sv)) {
7088 /* make sure SvREFCNT(sv)==0 happens very seldom */
7089 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
7093 if (! SvREFCNT(sv)) /* may have have been resurrected */
7098 /* handle exceptional cases */
7102 if (SvFLAGS(sv) & SVf_BREAK)
7103 /* this SV's refcnt has been artificially decremented to
7104 * trigger cleanup */
7106 if (PL_in_clean_all) /* All is fair */
7108 if (SvIMMORTAL(sv)) {
7109 /* make sure SvREFCNT(sv)==0 happens very seldom */
7110 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
7113 if (ckWARN_d(WARN_INTERNAL)) {
7114 #ifdef DEBUG_LEAKING_SCALARS_FORK_DUMP
7115 Perl_dump_sv_child(aTHX_ sv);
7117 #ifdef DEBUG_LEAKING_SCALARS
7120 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7121 if (PL_warnhook == PERL_WARNHOOK_FATAL
7122 || ckDEAD(packWARN(WARN_INTERNAL))) {
7123 /* Don't let Perl_warner cause us to escape our fate: */
7127 /* This may not return: */
7128 Perl_warner(aTHX_ packWARN(WARN_INTERNAL),
7129 "Attempt to free unreferenced scalar: SV 0x%" UVxf
7130 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
7133 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7143 Returns the length of the string in the SV. Handles magic and type
7144 coercion and sets the UTF8 flag appropriately. See also C<L</SvCUR>>, which
7145 gives raw access to the C<xpv_cur> slot.
7151 Perl_sv_len(pTHX_ SV *const sv)
7158 (void)SvPV_const(sv, len);
7163 =for apidoc sv_len_utf8
7165 Returns the number of characters in the string in an SV, counting wide
7166 UTF-8 bytes as a single character. Handles magic and type coercion.
7172 * The length is cached in PERL_MAGIC_utf8, in the mg_len field. Also the
7173 * mg_ptr is used, by sv_pos_u2b() and sv_pos_b2u() - see the comments below.
7174 * (Note that the mg_len is not the length of the mg_ptr field.
7175 * This allows the cache to store the character length of the string without
7176 * needing to malloc() extra storage to attach to the mg_ptr.)
7181 Perl_sv_len_utf8(pTHX_ SV *const sv)
7187 return sv_len_utf8_nomg(sv);
7191 Perl_sv_len_utf8_nomg(pTHX_ SV * const sv)
7194 const U8 *s = (U8*)SvPV_nomg_const(sv, len);
7196 PERL_ARGS_ASSERT_SV_LEN_UTF8_NOMG;
7198 if (PL_utf8cache && SvUTF8(sv)) {
7200 MAGIC *mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_utf8) : NULL;
7202 if (mg && (mg->mg_len != -1 || mg->mg_ptr)) {
7203 if (mg->mg_len != -1)
7206 /* We can use the offset cache for a headstart.
7207 The longer value is stored in the first pair. */
7208 STRLEN *cache = (STRLEN *) mg->mg_ptr;
7210 ulen = cache[0] + Perl_utf8_length(aTHX_ s + cache[1],
7214 if (PL_utf8cache < 0) {
7215 const STRLEN real = Perl_utf8_length(aTHX_ s, s + len);
7216 assert_uft8_cache_coherent("sv_len_utf8", ulen, real, sv);
7220 ulen = Perl_utf8_length(aTHX_ s, s + len);
7221 utf8_mg_len_cache_update(sv, &mg, ulen);
7225 return SvUTF8(sv) ? Perl_utf8_length(aTHX_ s, s + len) : len;
7228 /* Walk forwards to find the byte corresponding to the passed in UTF-8
7231 S_sv_pos_u2b_forwards(const U8 *const start, const U8 *const send,
7232 STRLEN *const uoffset_p, bool *const at_end)
7234 const U8 *s = start;
7235 STRLEN uoffset = *uoffset_p;
7237 PERL_ARGS_ASSERT_SV_POS_U2B_FORWARDS;
7239 while (s < send && uoffset) {
7246 else if (s > send) {
7248 /* This is the existing behaviour. Possibly it should be a croak, as
7249 it's actually a bounds error */
7252 *uoffset_p -= uoffset;
7256 /* Given the length of the string in both bytes and UTF-8 characters, decide
7257 whether to walk forwards or backwards to find the byte corresponding to
7258 the passed in UTF-8 offset. */
7260 S_sv_pos_u2b_midway(const U8 *const start, const U8 *send,
7261 STRLEN uoffset, const STRLEN uend)
7263 STRLEN backw = uend - uoffset;
7265 PERL_ARGS_ASSERT_SV_POS_U2B_MIDWAY;
7267 if (uoffset < 2 * backw) {
7268 /* The assumption is that going forwards is twice the speed of going
7269 forward (that's where the 2 * backw comes from).
7270 (The real figure of course depends on the UTF-8 data.) */
7271 const U8 *s = start;
7273 while (s < send && uoffset--)
7283 while (UTF8_IS_CONTINUATION(*send))
7286 return send - start;
7289 /* For the string representation of the given scalar, find the byte
7290 corresponding to the passed in UTF-8 offset. uoffset0 and boffset0
7291 give another position in the string, *before* the sought offset, which
7292 (which is always true, as 0, 0 is a valid pair of positions), which should
7293 help reduce the amount of linear searching.
7294 If *mgp is non-NULL, it should point to the UTF-8 cache magic, which
7295 will be used to reduce the amount of linear searching. The cache will be
7296 created if necessary, and the found value offered to it for update. */
7298 S_sv_pos_u2b_cached(pTHX_ SV *const sv, MAGIC **const mgp, const U8 *const start,
7299 const U8 *const send, STRLEN uoffset,
7300 STRLEN uoffset0, STRLEN boffset0)
7302 STRLEN boffset = 0; /* Actually always set, but let's keep gcc happy. */
7304 bool at_end = FALSE;
7306 PERL_ARGS_ASSERT_SV_POS_U2B_CACHED;
7308 assert (uoffset >= uoffset0);
7313 if (!SvREADONLY(sv) && !SvGMAGICAL(sv) && SvPOK(sv)
7315 && (*mgp || (SvTYPE(sv) >= SVt_PVMG &&
7316 (*mgp = mg_find(sv, PERL_MAGIC_utf8))))) {
7317 if ((*mgp)->mg_ptr) {
7318 STRLEN *cache = (STRLEN *) (*mgp)->mg_ptr;
7319 if (cache[0] == uoffset) {
7320 /* An exact match. */
7323 if (cache[2] == uoffset) {
7324 /* An exact match. */
7328 if (cache[0] < uoffset) {
7329 /* The cache already knows part of the way. */
7330 if (cache[0] > uoffset0) {
7331 /* The cache knows more than the passed in pair */
7332 uoffset0 = cache[0];
7333 boffset0 = cache[1];
7335 if ((*mgp)->mg_len != -1) {
7336 /* And we know the end too. */
7338 + sv_pos_u2b_midway(start + boffset0, send,
7340 (*mgp)->mg_len - uoffset0);
7342 uoffset -= uoffset0;
7344 + sv_pos_u2b_forwards(start + boffset0,
7345 send, &uoffset, &at_end);
7346 uoffset += uoffset0;
7349 else if (cache[2] < uoffset) {
7350 /* We're between the two cache entries. */
7351 if (cache[2] > uoffset0) {
7352 /* and the cache knows more than the passed in pair */
7353 uoffset0 = cache[2];
7354 boffset0 = cache[3];
7358 + sv_pos_u2b_midway(start + boffset0,
7361 cache[0] - uoffset0);
7364 + sv_pos_u2b_midway(start + boffset0,
7367 cache[2] - uoffset0);
7371 else if ((*mgp)->mg_len != -1) {
7372 /* If we can take advantage of a passed in offset, do so. */
7373 /* In fact, offset0 is either 0, or less than offset, so don't
7374 need to worry about the other possibility. */
7376 + sv_pos_u2b_midway(start + boffset0, send,
7378 (*mgp)->mg_len - uoffset0);
7383 if (!found || PL_utf8cache < 0) {
7384 STRLEN real_boffset;
7385 uoffset -= uoffset0;
7386 real_boffset = boffset0 + sv_pos_u2b_forwards(start + boffset0,
7387 send, &uoffset, &at_end);
7388 uoffset += uoffset0;
7390 if (found && PL_utf8cache < 0)
7391 assert_uft8_cache_coherent("sv_pos_u2b_cache", boffset,
7393 boffset = real_boffset;
7396 if (PL_utf8cache && !SvGMAGICAL(sv) && SvPOK(sv)) {
7398 utf8_mg_len_cache_update(sv, mgp, uoffset);
7400 utf8_mg_pos_cache_update(sv, mgp, boffset, uoffset, send - start);
7407 =for apidoc sv_pos_u2b_flags
7409 Converts the offset from a count of UTF-8 chars from
7410 the start of the string, to a count of the equivalent number of bytes; if
7411 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7412 C<offset>, rather than from the start
7413 of the string. Handles type coercion.
7414 C<flags> is passed to C<SvPV_flags>, and usually should be
7415 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7421 * sv_pos_u2b_flags() uses, like sv_pos_b2u(), the mg_ptr of the potential
7422 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7423 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7428 Perl_sv_pos_u2b_flags(pTHX_ SV *const sv, STRLEN uoffset, STRLEN *const lenp,
7435 PERL_ARGS_ASSERT_SV_POS_U2B_FLAGS;
7437 start = (U8*)SvPV_flags(sv, len, flags);
7439 const U8 * const send = start + len;
7441 boffset = sv_pos_u2b_cached(sv, &mg, start, send, uoffset, 0, 0);
7444 && *lenp /* don't bother doing work for 0, as its bytes equivalent
7445 is 0, and *lenp is already set to that. */) {
7446 /* Convert the relative offset to absolute. */
7447 const STRLEN uoffset2 = uoffset + *lenp;
7448 const STRLEN boffset2
7449 = sv_pos_u2b_cached(sv, &mg, start, send, uoffset2,
7450 uoffset, boffset) - boffset;
7464 =for apidoc sv_pos_u2b
7466 Converts the value pointed to by C<offsetp> from a count of UTF-8 chars from
7467 the start of the string, to a count of the equivalent number of bytes; if
7468 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7469 the offset, rather than from the start of the string. Handles magic and
7472 Use C<sv_pos_u2b_flags> in preference, which correctly handles strings longer
7479 * sv_pos_u2b() uses, like sv_pos_b2u(), the mg_ptr of the potential
7480 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7481 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7485 /* This function is subject to size and sign problems */
7488 Perl_sv_pos_u2b(pTHX_ SV *const sv, I32 *const offsetp, I32 *const lenp)
7490 PERL_ARGS_ASSERT_SV_POS_U2B;
7493 STRLEN ulen = (STRLEN)*lenp;
7494 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, &ulen,
7495 SV_GMAGIC|SV_CONST_RETURN);
7498 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, NULL,
7499 SV_GMAGIC|SV_CONST_RETURN);
7504 S_utf8_mg_len_cache_update(pTHX_ SV *const sv, MAGIC **const mgp,
7507 PERL_ARGS_ASSERT_UTF8_MG_LEN_CACHE_UPDATE;
7508 if (SvREADONLY(sv) || SvGMAGICAL(sv) || !SvPOK(sv))
7511 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7512 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7513 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, &PL_vtbl_utf8, 0, 0);
7517 (*mgp)->mg_len = ulen;
7520 /* Create and update the UTF8 magic offset cache, with the proffered utf8/
7521 byte length pairing. The (byte) length of the total SV is passed in too,
7522 as blen, because for some (more esoteric) SVs, the call to SvPV_const()
7523 may not have updated SvCUR, so we can't rely on reading it directly.
7525 The proffered utf8/byte length pairing isn't used if the cache already has
7526 two pairs, and swapping either for the proffered pair would increase the
7527 RMS of the intervals between known byte offsets.
7529 The cache itself consists of 4 STRLEN values
7530 0: larger UTF-8 offset
7531 1: corresponding byte offset
7532 2: smaller UTF-8 offset
7533 3: corresponding byte offset
7535 Unused cache pairs have the value 0, 0.
7536 Keeping the cache "backwards" means that the invariant of
7537 cache[0] >= cache[2] is maintained even with empty slots, which means that
7538 the code that uses it doesn't need to worry if only 1 entry has actually
7539 been set to non-zero. It also makes the "position beyond the end of the
7540 cache" logic much simpler, as the first slot is always the one to start
7544 S_utf8_mg_pos_cache_update(pTHX_ SV *const sv, MAGIC **const mgp, const STRLEN byte,
7545 const STRLEN utf8, const STRLEN blen)
7549 PERL_ARGS_ASSERT_UTF8_MG_POS_CACHE_UPDATE;
7554 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7555 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7556 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, (MGVTBL*)&PL_vtbl_utf8, 0,
7558 (*mgp)->mg_len = -1;
7562 if (!(cache = (STRLEN *)(*mgp)->mg_ptr)) {
7563 Newxz(cache, PERL_MAGIC_UTF8_CACHESIZE * 2, STRLEN);
7564 (*mgp)->mg_ptr = (char *) cache;
7568 if (PL_utf8cache < 0 && SvPOKp(sv)) {
7569 /* SvPOKp() because, if sv is a reference, then SvPVX() is actually
7570 a pointer. Note that we no longer cache utf8 offsets on refer-
7571 ences, but this check is still a good idea, for robustness. */
7572 const U8 *start = (const U8 *) SvPVX_const(sv);
7573 const STRLEN realutf8 = utf8_length(start, start + byte);
7575 assert_uft8_cache_coherent("utf8_mg_pos_cache_update", utf8, realutf8,
7579 /* Cache is held with the later position first, to simplify the code
7580 that deals with unbounded ends. */
7582 ASSERT_UTF8_CACHE(cache);
7583 if (cache[1] == 0) {
7584 /* Cache is totally empty */
7587 } else if (cache[3] == 0) {
7588 if (byte > cache[1]) {
7589 /* New one is larger, so goes first. */
7590 cache[2] = cache[0];
7591 cache[3] = cache[1];
7599 /* float casts necessary? XXX */
7600 #define THREEWAY_SQUARE(a,b,c,d) \
7601 ((float)((d) - (c))) * ((float)((d) - (c))) \
7602 + ((float)((c) - (b))) * ((float)((c) - (b))) \
7603 + ((float)((b) - (a))) * ((float)((b) - (a)))
7605 /* Cache has 2 slots in use, and we know three potential pairs.
7606 Keep the two that give the lowest RMS distance. Do the
7607 calculation in bytes simply because we always know the byte
7608 length. squareroot has the same ordering as the positive value,
7609 so don't bother with the actual square root. */
7610 if (byte > cache[1]) {
7611 /* New position is after the existing pair of pairs. */
7612 const float keep_earlier
7613 = THREEWAY_SQUARE(0, cache[3], byte, blen);
7614 const float keep_later
7615 = THREEWAY_SQUARE(0, cache[1], byte, blen);
7617 if (keep_later < keep_earlier) {
7618 cache[2] = cache[0];
7619 cache[3] = cache[1];
7625 const float keep_later = THREEWAY_SQUARE(0, byte, cache[1], blen);
7626 float b, c, keep_earlier;
7627 if (byte > cache[3]) {
7628 /* New position is between the existing pair of pairs. */
7629 b = (float)cache[3];
7632 /* New position is before the existing pair of pairs. */
7634 c = (float)cache[3];
7636 keep_earlier = THREEWAY_SQUARE(0, b, c, blen);
7637 if (byte > cache[3]) {
7638 if (keep_later < keep_earlier) {
7648 if (! (keep_later < keep_earlier)) {
7649 cache[0] = cache[2];
7650 cache[1] = cache[3];
7657 ASSERT_UTF8_CACHE(cache);
7660 /* We already know all of the way, now we may be able to walk back. The same
7661 assumption is made as in S_sv_pos_u2b_midway(), namely that walking
7662 backward is half the speed of walking forward. */
7664 S_sv_pos_b2u_midway(pTHX_ const U8 *const s, const U8 *const target,
7665 const U8 *end, STRLEN endu)
7667 const STRLEN forw = target - s;
7668 STRLEN backw = end - target;
7670 PERL_ARGS_ASSERT_SV_POS_B2U_MIDWAY;
7672 if (forw < 2 * backw) {
7673 return utf8_length(s, target);
7676 while (end > target) {
7678 while (UTF8_IS_CONTINUATION(*end)) {
7687 =for apidoc sv_pos_b2u_flags
7689 Converts C<offset> from a count of bytes from the start of the string, to
7690 a count of the equivalent number of UTF-8 chars. Handles type coercion.
7691 C<flags> is passed to C<SvPV_flags>, and usually should be
7692 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7698 * sv_pos_b2u_flags() uses, like sv_pos_u2b_flags(), the mg_ptr of the
7699 * potential PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8
7704 Perl_sv_pos_b2u_flags(pTHX_ SV *const sv, STRLEN const offset, U32 flags)
7707 STRLEN len = 0; /* Actually always set, but let's keep gcc happy. */
7713 PERL_ARGS_ASSERT_SV_POS_B2U_FLAGS;
7715 s = (const U8*)SvPV_flags(sv, blen, flags);
7718 Perl_croak(aTHX_ "panic: sv_pos_b2u: bad byte offset, blen=%" UVuf
7719 ", byte=%" UVuf, (UV)blen, (UV)offset);
7725 && SvTYPE(sv) >= SVt_PVMG
7726 && (mg = mg_find(sv, PERL_MAGIC_utf8)))
7729 STRLEN * const cache = (STRLEN *) mg->mg_ptr;
7730 if (cache[1] == offset) {
7731 /* An exact match. */
7734 if (cache[3] == offset) {
7735 /* An exact match. */
7739 if (cache[1] < offset) {
7740 /* We already know part of the way. */
7741 if (mg->mg_len != -1) {
7742 /* Actually, we know the end too. */
7744 + S_sv_pos_b2u_midway(aTHX_ s + cache[1], send,
7745 s + blen, mg->mg_len - cache[0]);
7747 len = cache[0] + utf8_length(s + cache[1], send);
7750 else if (cache[3] < offset) {
7751 /* We're between the two cached pairs, so we do the calculation
7752 offset by the byte/utf-8 positions for the earlier pair,
7753 then add the utf-8 characters from the string start to
7755 len = S_sv_pos_b2u_midway(aTHX_ s + cache[3], send,
7756 s + cache[1], cache[0] - cache[2])
7760 else { /* cache[3] > offset */
7761 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + cache[3],
7765 ASSERT_UTF8_CACHE(cache);
7767 } else if (mg->mg_len != -1) {
7768 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + blen, mg->mg_len);
7772 if (!found || PL_utf8cache < 0) {
7773 const STRLEN real_len = utf8_length(s, send);
7775 if (found && PL_utf8cache < 0)
7776 assert_uft8_cache_coherent("sv_pos_b2u", len, real_len, sv);
7782 utf8_mg_len_cache_update(sv, &mg, len);
7784 utf8_mg_pos_cache_update(sv, &mg, offset, len, blen);
7791 =for apidoc sv_pos_b2u
7793 Converts the value pointed to by C<offsetp> from a count of bytes from the
7794 start of the string, to a count of the equivalent number of UTF-8 chars.
7795 Handles magic and type coercion.
7797 Use C<sv_pos_b2u_flags> in preference, which correctly handles strings
7804 * sv_pos_b2u() uses, like sv_pos_u2b(), the mg_ptr of the potential
7805 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7810 Perl_sv_pos_b2u(pTHX_ SV *const sv, I32 *const offsetp)
7812 PERL_ARGS_ASSERT_SV_POS_B2U;
7817 *offsetp = (I32)sv_pos_b2u_flags(sv, (STRLEN)*offsetp,
7818 SV_GMAGIC|SV_CONST_RETURN);
7822 S_assert_uft8_cache_coherent(pTHX_ const char *const func, STRLEN from_cache,
7823 STRLEN real, SV *const sv)
7825 PERL_ARGS_ASSERT_ASSERT_UFT8_CACHE_COHERENT;
7827 /* As this is debugging only code, save space by keeping this test here,
7828 rather than inlining it in all the callers. */
7829 if (from_cache == real)
7832 /* Need to turn the assertions off otherwise we may recurse infinitely
7833 while printing error messages. */
7834 SAVEI8(PL_utf8cache);
7836 Perl_croak(aTHX_ "panic: %s cache %" UVuf " real %" UVuf " for %" SVf,
7837 func, (UV) from_cache, (UV) real, SVfARG(sv));
7843 Returns a boolean indicating whether the strings in the two SVs are
7844 identical. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7845 coerce its args to strings if necessary.
7847 =for apidoc sv_eq_flags
7849 Returns a boolean indicating whether the strings in the two SVs are
7850 identical. Is UTF-8 and S<C<'use bytes'>> aware and coerces its args to strings
7851 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get-magic, too.
7857 Perl_sv_eq_flags(pTHX_ SV *sv1, SV *sv2, const U32 flags)
7869 /* if pv1 and pv2 are the same, second SvPV_const call may
7870 * invalidate pv1 (if we are handling magic), so we may need to
7872 if (sv1 == sv2 && flags & SV_GMAGIC
7873 && (SvTHINKFIRST(sv1) || SvGMAGICAL(sv1))) {
7874 pv1 = SvPV_const(sv1, cur1);
7875 sv1 = newSVpvn_flags(pv1, cur1, SVs_TEMP | SvUTF8(sv2));
7877 pv1 = SvPV_flags_const(sv1, cur1, flags);
7885 pv2 = SvPV_flags_const(sv2, cur2, flags);
7887 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7888 /* Differing utf8ness. */
7890 /* sv1 is the UTF-8 one */
7891 return bytes_cmp_utf8((const U8*)pv2, cur2,
7892 (const U8*)pv1, cur1) == 0;
7895 /* sv2 is the UTF-8 one */
7896 return bytes_cmp_utf8((const U8*)pv1, cur1,
7897 (const U8*)pv2, cur2) == 0;
7902 return (pv1 == pv2) || memEQ(pv1, pv2, cur1);
7910 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7911 string in C<sv1> is less than, equal to, or greater than the string in
7912 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7913 coerce its args to strings if necessary. See also C<L</sv_cmp_locale>>.
7915 =for apidoc sv_cmp_flags
7917 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7918 string in C<sv1> is less than, equal to, or greater than the string in
7919 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware and will coerce its args to strings
7920 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get magic. See
7921 also C<L</sv_cmp_locale_flags>>.
7927 Perl_sv_cmp(pTHX_ SV *const sv1, SV *const sv2)
7929 return sv_cmp_flags(sv1, sv2, SV_GMAGIC);
7933 Perl_sv_cmp_flags(pTHX_ SV *const sv1, SV *const sv2,
7937 const char *pv1, *pv2;
7939 SV *svrecode = NULL;
7946 pv1 = SvPV_flags_const(sv1, cur1, flags);
7953 pv2 = SvPV_flags_const(sv2, cur2, flags);
7955 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7956 /* Differing utf8ness. */
7958 const int retval = -bytes_cmp_utf8((const U8*)pv2, cur2,
7959 (const U8*)pv1, cur1);
7960 return retval ? retval < 0 ? -1 : +1 : 0;
7963 const int retval = bytes_cmp_utf8((const U8*)pv1, cur1,
7964 (const U8*)pv2, cur2);
7965 return retval ? retval < 0 ? -1 : +1 : 0;
7969 /* Here, if both are non-NULL, then they have the same UTF8ness. */
7972 cmp = cur2 ? -1 : 0;
7976 STRLEN shortest_len = cur1 < cur2 ? cur1 : cur2;
7979 if (! DO_UTF8(sv1)) {
7981 const I32 retval = memcmp((const void*)pv1,
7985 cmp = retval < 0 ? -1 : 1;
7986 } else if (cur1 == cur2) {
7989 cmp = cur1 < cur2 ? -1 : 1;
7993 else { /* Both are to be treated as UTF-EBCDIC */
7995 /* EBCDIC UTF-8 is complicated by the fact that it is based on I8
7996 * which remaps code points 0-255. We therefore generally have to
7997 * unmap back to the original values to get an accurate comparison.
7998 * But we don't have to do that for UTF-8 invariants, as by
7999 * definition, they aren't remapped, nor do we have to do it for
8000 * above-latin1 code points, as they also aren't remapped. (This
8001 * code also works on ASCII platforms, but the memcmp() above is
8004 const char *e = pv1 + shortest_len;
8006 /* Find the first bytes that differ between the two strings */
8007 while (pv1 < e && *pv1 == *pv2) {
8013 if (pv1 == e) { /* Are the same all the way to the end */
8017 cmp = cur1 < cur2 ? -1 : 1;
8020 else /* Here *pv1 and *pv2 are not equal, but all bytes earlier
8021 * in the strings were. The current bytes may or may not be
8022 * at the beginning of a character. But neither or both are
8023 * (or else earlier bytes would have been different). And
8024 * if we are in the middle of a character, the two
8025 * characters are comprised of the same number of bytes
8026 * (because in this case the start bytes are the same, and
8027 * the start bytes encode the character's length). */
8028 if (UTF8_IS_INVARIANT(*pv1))
8030 /* If both are invariants; can just compare directly */
8031 if (UTF8_IS_INVARIANT(*pv2)) {
8032 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8034 else /* Since *pv1 is invariant, it is the whole character,
8035 which means it is at the beginning of a character.
8036 That means pv2 is also at the beginning of a
8037 character (see earlier comment). Since it isn't
8038 invariant, it must be a start byte. If it starts a
8039 character whose code point is above 255, that
8040 character is greater than any single-byte char, which
8042 if (UTF8_IS_ABOVE_LATIN1_START(*pv2))
8047 /* Here, pv2 points to a character composed of 2 bytes
8048 * whose code point is < 256. Get its code point and
8049 * compare with *pv1 */
8050 cmp = ((U8) *pv1 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
8055 else /* The code point starting at pv1 isn't a single byte */
8056 if (UTF8_IS_INVARIANT(*pv2))
8058 /* But here, the code point starting at *pv2 is a single byte,
8059 * and so *pv1 must begin a character, hence is a start byte.
8060 * If that character is above 255, it is larger than any
8061 * single-byte char, which *pv2 is */
8062 if (UTF8_IS_ABOVE_LATIN1_START(*pv1)) {
8066 /* Here, pv1 points to a character composed of 2 bytes
8067 * whose code point is < 256. Get its code point and
8068 * compare with the single byte character *pv2 */
8069 cmp = (EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1)) < (U8) *pv2)
8074 else /* Here, we've ruled out either *pv1 and *pv2 being
8075 invariant. That means both are part of variants, but not
8076 necessarily at the start of a character */
8077 if ( UTF8_IS_ABOVE_LATIN1_START(*pv1)
8078 || UTF8_IS_ABOVE_LATIN1_START(*pv2))
8080 /* Here, at least one is the start of a character, which means
8081 * the other is also a start byte. And the code point of at
8082 * least one of the characters is above 255. It is a
8083 * characteristic of UTF-EBCDIC that all start bytes for
8084 * above-latin1 code points are well behaved as far as code
8085 * point comparisons go, and all are larger than all other
8086 * start bytes, so the comparison with those is also well
8088 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8091 /* Here both *pv1 and *pv2 are part of variant characters.
8092 * They could be both continuations, or both start characters.
8093 * (One or both could even be an illegal start character (for
8094 * an overlong) which for the purposes of sorting we treat as
8096 if (UTF8_IS_CONTINUATION(*pv1)) {
8098 /* If they are continuations for code points above 255,
8099 * then comparing the current byte is sufficient, as there
8100 * is no remapping of these and so the comparison is
8101 * well-behaved. We determine if they are such
8102 * continuations by looking at the preceding byte. It
8103 * could be a start byte, from which we can tell if it is
8104 * for an above 255 code point. Or it could be a
8105 * continuation, which means the character occupies at
8106 * least 3 bytes, so must be above 255. */
8107 if ( UTF8_IS_CONTINUATION(*(pv2 - 1))
8108 || UTF8_IS_ABOVE_LATIN1_START(*(pv2 -1)))
8110 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8114 /* Here, the continuations are for code points below 256;
8115 * back up one to get to the start byte */
8120 /* We need to get the actual native code point of each of these
8121 * variants in order to compare them */
8122 cmp = ( EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1))
8123 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
8132 SvREFCNT_dec(svrecode);
8138 =for apidoc sv_cmp_locale
8140 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
8141 S<C<'use bytes'>> aware, handles get magic, and will coerce its args to strings
8142 if necessary. See also C<L</sv_cmp>>.
8144 =for apidoc sv_cmp_locale_flags
8146 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
8147 S<C<'use bytes'>> aware and will coerce its args to strings if necessary. If
8148 the flags contain C<SV_GMAGIC>, it handles get magic. See also
8149 C<L</sv_cmp_flags>>.
8155 Perl_sv_cmp_locale(pTHX_ SV *const sv1, SV *const sv2)
8157 return sv_cmp_locale_flags(sv1, sv2, SV_GMAGIC);
8161 Perl_sv_cmp_locale_flags(pTHX_ SV *const sv1, SV *const sv2,
8164 #ifdef USE_LOCALE_COLLATE
8170 if (PL_collation_standard)
8175 /* Revert to using raw compare if both operands exist, but either one
8176 * doesn't transform properly for collation */
8178 pv1 = sv_collxfrm_flags(sv1, &len1, flags);
8182 pv2 = sv_collxfrm_flags(sv2, &len2, flags);
8188 pv1 = sv1 ? sv_collxfrm_flags(sv1, &len1, flags) : (char *) NULL;
8189 pv2 = sv2 ? sv_collxfrm_flags(sv2, &len2, flags) : (char *) NULL;
8192 if (!pv1 || !len1) {
8203 retval = memcmp((void*)pv1, (void*)pv2, len1 < len2 ? len1 : len2);
8206 return retval < 0 ? -1 : 1;
8209 * When the result of collation is equality, that doesn't mean
8210 * that there are no differences -- some locales exclude some
8211 * characters from consideration. So to avoid false equalities,
8212 * we use the raw string as a tiebreaker.
8219 PERL_UNUSED_ARG(flags);
8220 #endif /* USE_LOCALE_COLLATE */
8222 return sv_cmp(sv1, sv2);
8226 #ifdef USE_LOCALE_COLLATE
8229 =for apidoc sv_collxfrm
8231 This calls C<sv_collxfrm_flags> with the SV_GMAGIC flag. See
8232 C<L</sv_collxfrm_flags>>.
8234 =for apidoc sv_collxfrm_flags
8236 Add Collate Transform magic to an SV if it doesn't already have it. If the
8237 flags contain C<SV_GMAGIC>, it handles get-magic.
8239 Any scalar variable may carry C<PERL_MAGIC_collxfrm> magic that contains the
8240 scalar data of the variable, but transformed to such a format that a normal
8241 memory comparison can be used to compare the data according to the locale
8248 Perl_sv_collxfrm_flags(pTHX_ SV *const sv, STRLEN *const nxp, const I32 flags)
8252 PERL_ARGS_ASSERT_SV_COLLXFRM_FLAGS;
8254 mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_collxfrm) : (MAGIC *) NULL;
8256 /* If we don't have collation magic on 'sv', or the locale has changed
8257 * since the last time we calculated it, get it and save it now */
8258 if (!mg || !mg->mg_ptr || *(U32*)mg->mg_ptr != PL_collation_ix) {
8263 /* Free the old space */
8265 Safefree(mg->mg_ptr);
8267 s = SvPV_flags_const(sv, len, flags);
8268 if ((xf = _mem_collxfrm(s, len, &xlen, cBOOL(SvUTF8(sv))))) {
8270 mg = sv_magicext(sv, 0, PERL_MAGIC_collxfrm, &PL_vtbl_collxfrm,
8285 if (mg && mg->mg_ptr) {
8287 return mg->mg_ptr + sizeof(PL_collation_ix);
8295 #endif /* USE_LOCALE_COLLATE */
8298 S_sv_gets_append_to_utf8(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8300 SV * const tsv = newSV(0);
8303 sv_gets(tsv, fp, 0);
8304 sv_utf8_upgrade_nomg(tsv);
8305 SvCUR_set(sv,append);
8308 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8312 S_sv_gets_read_record(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8315 const STRLEN recsize = SvUV(SvRV(PL_rs)); /* RsRECORD() guarantees > 0. */
8316 /* Grab the size of the record we're getting */
8317 char *buffer = SvGROW(sv, (STRLEN)(recsize + append + 1)) + append;
8324 /* With a true, record-oriented file on VMS, we need to use read directly
8325 * to ensure that we respect RMS record boundaries. The user is responsible
8326 * for providing a PL_rs value that corresponds to the FAB$W_MRS (maximum
8327 * record size) field. N.B. This is likely to produce invalid results on
8328 * varying-width character data when a record ends mid-character.
8330 fd = PerlIO_fileno(fp);
8332 && PerlLIO_fstat(fd, &st) == 0
8333 && (st.st_fab_rfm == FAB$C_VAR
8334 || st.st_fab_rfm == FAB$C_VFC
8335 || st.st_fab_rfm == FAB$C_FIX)) {
8337 bytesread = PerlLIO_read(fd, buffer, recsize);
8339 else /* in-memory file from PerlIO::Scalar
8340 * or not a record-oriented file
8344 bytesread = PerlIO_read(fp, buffer, recsize);
8346 /* At this point, the logic in sv_get() means that sv will
8347 be treated as utf-8 if the handle is utf8.
8349 if (PerlIO_isutf8(fp) && bytesread > 0) {
8350 char *bend = buffer + bytesread;
8351 char *bufp = buffer;
8352 size_t charcount = 0;
8353 bool charstart = TRUE;
8356 while (charcount < recsize) {
8357 /* count accumulated characters */
8358 while (bufp < bend) {
8360 skip = UTF8SKIP(bufp);
8362 if (bufp + skip > bend) {
8363 /* partial at the end */
8374 if (charcount < recsize) {
8376 STRLEN bufp_offset = bufp - buffer;
8377 SSize_t morebytesread;
8379 /* originally I read enough to fill any incomplete
8380 character and the first byte of the next
8381 character if needed, but if there's many
8382 multi-byte encoded characters we're going to be
8383 making a read call for every character beyond
8384 the original read size.
8386 So instead, read the rest of the character if
8387 any, and enough bytes to match at least the
8388 start bytes for each character we're going to
8392 readsize = recsize - charcount;
8394 readsize = skip - (bend - bufp) + recsize - charcount - 1;
8395 buffer = SvGROW(sv, append + bytesread + readsize + 1) + append;
8396 bend = buffer + bytesread;
8397 morebytesread = PerlIO_read(fp, bend, readsize);
8398 if (morebytesread <= 0) {
8399 /* we're done, if we still have incomplete
8400 characters the check code in sv_gets() will
8403 I'd originally considered doing
8404 PerlIO_ungetc() on all but the lead
8405 character of the incomplete character, but
8406 read() doesn't do that, so I don't.
8411 /* prepare to scan some more */
8412 bytesread += morebytesread;
8413 bend = buffer + bytesread;
8414 bufp = buffer + bufp_offset;
8422 SvCUR_set(sv, bytesread + append);
8423 buffer[bytesread] = '\0';
8424 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8430 Get a line from the filehandle and store it into the SV, optionally
8431 appending to the currently-stored string. If C<append> is not 0, the
8432 line is appended to the SV instead of overwriting it. C<append> should
8433 be set to the byte offset that the appended string should start at
8434 in the SV (typically, C<SvCUR(sv)> is a suitable choice).
8440 Perl_sv_gets(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8450 PERL_ARGS_ASSERT_SV_GETS;
8452 if (SvTHINKFIRST(sv))
8453 sv_force_normal_flags(sv, append ? 0 : SV_COW_DROP_PV);
8454 /* XXX. If you make this PVIV, then copy on write can copy scalars read
8456 However, perlbench says it's slower, because the existing swipe code
8457 is faster than copy on write.
8458 Swings and roundabouts. */
8459 SvUPGRADE(sv, SVt_PV);
8462 /* line is going to be appended to the existing buffer in the sv */
8463 if (PerlIO_isutf8(fp)) {
8465 sv_utf8_upgrade_nomg(sv);
8466 sv_pos_u2b(sv,&append,0);
8468 } else if (SvUTF8(sv)) {
8469 return S_sv_gets_append_to_utf8(aTHX_ sv, fp, append);
8475 /* not appending - "clear" the string by setting SvCUR to 0,
8476 * the pv is still avaiable. */
8479 if (PerlIO_isutf8(fp))
8482 if (IN_PERL_COMPILETIME) {
8483 /* we always read code in line mode */
8487 else if (RsSNARF(PL_rs)) {
8488 /* If it is a regular disk file use size from stat() as estimate
8489 of amount we are going to read -- may result in mallocing
8490 more memory than we really need if the layers below reduce
8491 the size we read (e.g. CRLF or a gzip layer).
8494 int fd = PerlIO_fileno(fp);
8495 if (fd >= 0 && (PerlLIO_fstat(fd, &st) == 0) && S_ISREG(st.st_mode)) {
8496 const Off_t offset = PerlIO_tell(fp);
8497 if (offset != (Off_t) -1 && st.st_size + append > offset) {
8498 #ifdef PERL_COPY_ON_WRITE
8499 /* Add an extra byte for the sake of copy-on-write's
8500 * buffer reference count. */
8501 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 2));
8503 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 1));
8510 else if (RsRECORD(PL_rs)) {
8511 return S_sv_gets_read_record(aTHX_ sv, fp, append);
8513 else if (RsPARA(PL_rs)) {
8519 /* Get $/ i.e. PL_rs into same encoding as stream wants */
8520 if (PerlIO_isutf8(fp)) {
8521 rsptr = SvPVutf8(PL_rs, rslen);
8524 if (SvUTF8(PL_rs)) {
8525 if (!sv_utf8_downgrade(PL_rs, TRUE)) {
8526 Perl_croak(aTHX_ "Wide character in $/");
8529 /* extract the raw pointer to the record separator */
8530 rsptr = SvPV_const(PL_rs, rslen);
8534 /* rslast is the last character in the record separator
8535 * note we don't use rslast except when rslen is true, so the
8536 * null assign is a placeholder. */
8537 rslast = rslen ? rsptr[rslen - 1] : '\0';
8539 if (rspara) { /* have to do this both before and after */
8540 /* to make sure file boundaries work right */
8544 i = PerlIO_getc(fp);
8548 PerlIO_ungetc(fp,i);
8554 /* See if we know enough about I/O mechanism to cheat it ! */
8556 /* This used to be #ifdef test - it is made run-time test for ease
8557 of abstracting out stdio interface. One call should be cheap
8558 enough here - and may even be a macro allowing compile
8562 if (PerlIO_fast_gets(fp)) {
8564 * We can do buffer based IO operations on this filehandle.
8566 * This means we can bypass a lot of subcalls and process
8567 * the buffer directly, it also means we know the upper bound
8568 * on the amount of data we might read of the current buffer
8569 * into our sv. Knowing this allows us to preallocate the pv
8570 * to be able to hold that maximum, which allows us to simplify
8571 * a lot of logic. */
8574 * We're going to steal some values from the stdio struct
8575 * and put EVERYTHING in the innermost loop into registers.
8577 STDCHAR *ptr; /* pointer into fp's read-ahead buffer */
8578 STRLEN bpx; /* length of the data in the target sv
8579 used to fix pointers after a SvGROW */
8580 I32 shortbuffered; /* If the pv buffer is shorter than the amount
8581 of data left in the read-ahead buffer.
8582 If 0 then the pv buffer can hold the full
8583 amount left, otherwise this is the amount it
8586 /* Here is some breathtakingly efficient cheating */
8588 /* When you read the following logic resist the urge to think
8589 * of record separators that are 1 byte long. They are an
8590 * uninteresting special (simple) case.
8592 * Instead think of record separators which are at least 2 bytes
8593 * long, and keep in mind that we need to deal with such
8594 * separators when they cross a read-ahead buffer boundary.
8596 * Also consider that we need to gracefully deal with separators
8597 * that may be longer than a single read ahead buffer.
8599 * Lastly do not forget we want to copy the delimiter as well. We
8600 * are copying all data in the file _up_to_and_including_ the separator
8603 * Now that you have all that in mind here is what is happening below:
8605 * 1. When we first enter the loop we do some memory book keeping to see
8606 * how much free space there is in the target SV. (This sub assumes that
8607 * it is operating on the same SV most of the time via $_ and that it is
8608 * going to be able to reuse the same pv buffer each call.) If there is
8609 * "enough" room then we set "shortbuffered" to how much space there is
8610 * and start reading forward.
8612 * 2. When we scan forward we copy from the read-ahead buffer to the target
8613 * SV's pv buffer. While we go we watch for the end of the read-ahead buffer,
8614 * and the end of the of pv, as well as for the "rslast", which is the last
8615 * char of the separator.
8617 * 3. When scanning forward if we see rslast then we jump backwards in *pv*
8618 * (which has a "complete" record up to the point we saw rslast) and check
8619 * it to see if it matches the separator. If it does we are done. If it doesn't
8620 * we continue on with the scan/copy.
8622 * 4. If we run out of read-ahead buffer (cnt goes to 0) then we have to get
8623 * the IO system to read the next buffer. We do this by doing a getc(), which
8624 * returns a single char read (or EOF), and prefills the buffer, and also
8625 * allows us to find out how full the buffer is. We use this information to
8626 * SvGROW() the sv to the size remaining in the buffer, after which we copy
8627 * the returned single char into the target sv, and then go back into scan
8630 * 5. If we run out of write-buffer then we SvGROW() it by the size of the
8631 * remaining space in the read-buffer.
8633 * Note that this code despite its twisty-turny nature is pretty darn slick.
8634 * It manages single byte separators, multi-byte cross boundary separators,
8635 * and cross-read-buffer separators cleanly and efficiently at the cost
8636 * of potentially greatly overallocating the target SV.
8642 /* get the number of bytes remaining in the read-ahead buffer
8643 * on first call on a given fp this will return 0.*/
8644 cnt = PerlIO_get_cnt(fp);
8646 /* make sure we have the room */
8647 if ((I32)(SvLEN(sv) - append) <= cnt + 1) {
8648 /* Not room for all of it
8649 if we are looking for a separator and room for some
8651 if (rslen && cnt > 80 && (I32)SvLEN(sv) > append) {
8652 /* just process what we have room for */
8653 shortbuffered = cnt - SvLEN(sv) + append + 1;
8654 cnt -= shortbuffered;
8657 /* ensure that the target sv has enough room to hold
8658 * the rest of the read-ahead buffer */
8660 /* remember that cnt can be negative */
8661 SvGROW(sv, (STRLEN)(append + (cnt <= 0 ? 2 : (cnt + 1))));
8665 /* we have enough room to hold the full buffer, lets scream */
8669 /* extract the pointer to sv's string buffer, offset by append as necessary */
8670 bp = (STDCHAR*)SvPVX_const(sv) + append; /* move these two too to registers */
8671 /* extract the point to the read-ahead buffer */
8672 ptr = (STDCHAR*)PerlIO_get_ptr(fp);
8674 /* some trace debug output */
8675 DEBUG_P(PerlIO_printf(Perl_debug_log,
8676 "Screamer: entering, ptr=%" UVuf ", cnt=%ld\n",PTR2UV(ptr),(long)cnt));
8677 DEBUG_P(PerlIO_printf(Perl_debug_log,
8678 "Screamer: entering: PerlIO * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%"
8680 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8681 PTR2UV(PerlIO_has_base(fp) ? PerlIO_get_base(fp) : 0)));
8685 /* if there is stuff left in the read-ahead buffer */
8687 /* if there is a separator */
8689 /* find next rslast */
8692 /* shortcut common case of blank line */
8694 if ((*bp++ = *ptr++) == rslast)
8695 goto thats_all_folks;
8697 p = (STDCHAR *)memchr(ptr, rslast, cnt);
8699 SSize_t got = p - ptr + 1;
8700 Copy(ptr, bp, got, STDCHAR);
8704 goto thats_all_folks;
8706 Copy(ptr, bp, cnt, STDCHAR);
8712 /* no separator, slurp the full buffer */
8713 Copy(ptr, bp, cnt, char); /* this | eat */
8714 bp += cnt; /* screams | dust */
8715 ptr += cnt; /* louder | sed :-) */
8717 assert (!shortbuffered);
8718 goto cannot_be_shortbuffered;
8722 if (shortbuffered) { /* oh well, must extend */
8723 /* we didnt have enough room to fit the line into the target buffer
8724 * so we must extend the target buffer and keep going */
8725 cnt = shortbuffered;
8727 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8729 /* extned the target sv's buffer so it can hold the full read-ahead buffer */
8730 SvGROW(sv, SvLEN(sv) + append + cnt + 2);
8731 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8735 cannot_be_shortbuffered:
8736 /* we need to refill the read-ahead buffer if possible */
8738 DEBUG_P(PerlIO_printf(Perl_debug_log,
8739 "Screamer: going to getc, ptr=%" UVuf ", cnt=%" IVdf "\n",
8740 PTR2UV(ptr),(IV)cnt));
8741 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* deregisterize cnt and ptr */
8743 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8744 "Screamer: pre: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf "\n",
8745 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8746 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8749 call PerlIO_getc() to let it prefill the lookahead buffer
8751 This used to call 'filbuf' in stdio form, but as that behaves like
8752 getc when cnt <= 0 we use PerlIO_getc here to avoid introducing
8753 another abstraction.
8755 Note we have to deal with the char in 'i' if we are not at EOF
8757 i = PerlIO_getc(fp); /* get more characters */
8759 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8760 "Screamer: post: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf "\n",
8761 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8762 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8764 /* find out how much is left in the read-ahead buffer, and rextract its pointer */
8765 cnt = PerlIO_get_cnt(fp);
8766 ptr = (STDCHAR*)PerlIO_get_ptr(fp); /* reregisterize cnt and ptr */
8767 DEBUG_P(PerlIO_printf(Perl_debug_log,
8768 "Screamer: after getc, ptr=%" UVuf ", cnt=%" IVdf "\n",
8769 PTR2UV(ptr),(IV)cnt));
8771 if (i == EOF) /* all done for ever? */
8772 goto thats_really_all_folks;
8774 /* make sure we have enough space in the target sv */
8775 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8777 SvGROW(sv, bpx + cnt + 2);
8778 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8780 /* copy of the char we got from getc() */
8781 *bp++ = (STDCHAR)i; /* store character from PerlIO_getc */
8783 /* make sure we deal with the i being the last character of a separator */
8784 if (rslen && (STDCHAR)i == rslast) /* all done for now? */
8785 goto thats_all_folks;
8789 /* check if we have actually found the separator - only really applies
8791 if ((rslen > 1 && (STRLEN)(bp - (STDCHAR*)SvPVX_const(sv)) < rslen) ||
8792 memNE((char*)bp - rslen, rsptr, rslen))
8793 goto screamer; /* go back to the fray */
8794 thats_really_all_folks:
8796 cnt += shortbuffered;
8797 DEBUG_P(PerlIO_printf(Perl_debug_log,
8798 "Screamer: quitting, ptr=%" UVuf ", cnt=%" IVdf "\n",PTR2UV(ptr),(IV)cnt));
8799 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* put these back or we're in trouble */
8800 DEBUG_P(PerlIO_printf(Perl_debug_log,
8801 "Screamer: end: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf
8803 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8804 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8806 SvCUR_set(sv, bp - (STDCHAR*)SvPVX_const(sv)); /* set length */
8807 DEBUG_P(PerlIO_printf(Perl_debug_log,
8808 "Screamer: done, len=%ld, string=|%.*s|\n",
8809 (long)SvCUR(sv),(int)SvCUR(sv),SvPVX_const(sv)));
8813 /*The big, slow, and stupid way. */
8814 #ifdef USE_HEAP_INSTEAD_OF_STACK /* Even slower way. */
8815 STDCHAR *buf = NULL;
8816 Newx(buf, 8192, STDCHAR);
8824 const STDCHAR * const bpe = buf + sizeof(buf);
8826 while ((i = PerlIO_getc(fp)) != EOF && (*bp++ = (STDCHAR)i) != rslast && bp < bpe)
8827 ; /* keep reading */
8831 cnt = PerlIO_read(fp,(char*)buf, sizeof(buf));
8832 /* Accommodate broken VAXC compiler, which applies U8 cast to
8833 * both args of ?: operator, causing EOF to change into 255
8836 i = (U8)buf[cnt - 1];
8842 cnt = 0; /* we do need to re-set the sv even when cnt <= 0 */
8844 sv_catpvn_nomg(sv, (char *) buf, cnt);
8846 sv_setpvn(sv, (char *) buf, cnt); /* "nomg" is implied */
8848 if (i != EOF && /* joy */
8850 SvCUR(sv) < rslen ||
8851 memNE(SvPVX_const(sv) + SvCUR(sv) - rslen, rsptr, rslen)))
8855 * If we're reading from a TTY and we get a short read,
8856 * indicating that the user hit his EOF character, we need
8857 * to notice it now, because if we try to read from the TTY
8858 * again, the EOF condition will disappear.
8860 * The comparison of cnt to sizeof(buf) is an optimization
8861 * that prevents unnecessary calls to feof().
8865 if (!(cnt < (I32)sizeof(buf) && PerlIO_eof(fp)))
8869 #ifdef USE_HEAP_INSTEAD_OF_STACK
8874 if (rspara) { /* have to do this both before and after */
8875 while (i != EOF) { /* to make sure file boundaries work right */
8876 i = PerlIO_getc(fp);
8878 PerlIO_ungetc(fp,i);
8884 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8890 Auto-increment of the value in the SV, doing string to numeric conversion
8891 if necessary. Handles 'get' magic and operator overloading.
8897 Perl_sv_inc(pTHX_ SV *const sv)
8906 =for apidoc sv_inc_nomg
8908 Auto-increment of the value in the SV, doing string to numeric conversion
8909 if necessary. Handles operator overloading. Skips handling 'get' magic.
8915 Perl_sv_inc_nomg(pTHX_ SV *const sv)
8922 if (SvTHINKFIRST(sv)) {
8923 if (SvREADONLY(sv)) {
8924 Perl_croak_no_modify();
8928 if (SvAMAGIC(sv) && AMG_CALLunary(sv, inc_amg))
8930 i = PTR2IV(SvRV(sv));
8934 else sv_force_normal_flags(sv, 0);
8936 flags = SvFLAGS(sv);
8937 if ((flags & (SVp_NOK|SVp_IOK)) == SVp_NOK) {
8938 /* It's (privately or publicly) a float, but not tested as an
8939 integer, so test it to see. */
8941 flags = SvFLAGS(sv);
8943 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8944 /* It's publicly an integer, or privately an integer-not-float */
8945 #ifdef PERL_PRESERVE_IVUV
8949 if (SvUVX(sv) == UV_MAX)
8950 sv_setnv(sv, UV_MAX_P1);
8952 (void)SvIOK_only_UV(sv);
8953 SvUV_set(sv, SvUVX(sv) + 1);
8955 if (SvIVX(sv) == IV_MAX)
8956 sv_setuv(sv, (UV)IV_MAX + 1);
8958 (void)SvIOK_only(sv);
8959 SvIV_set(sv, SvIVX(sv) + 1);
8964 if (flags & SVp_NOK) {
8965 const NV was = SvNVX(sv);
8966 if (LIKELY(!Perl_isinfnan(was)) &&
8967 NV_OVERFLOWS_INTEGERS_AT != 0.0 &&
8968 was >= NV_OVERFLOWS_INTEGERS_AT) {
8969 /* diag_listed_as: Lost precision when %s %f by 1 */
8970 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8971 "Lost precision when incrementing %" NVff " by 1",
8974 (void)SvNOK_only(sv);
8975 SvNV_set(sv, was + 1.0);
8979 /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */
8980 if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv)))
8981 Perl_croak_no_modify();
8983 if (!(flags & SVp_POK) || !*SvPVX_const(sv)) {
8984 if ((flags & SVTYPEMASK) < SVt_PVIV)
8985 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV ? SVt_PVIV : SVt_IV));
8986 (void)SvIOK_only(sv);
8991 while (isALPHA(*d)) d++;
8992 while (isDIGIT(*d)) d++;
8993 if (d < SvEND(sv)) {
8994 const int numtype = grok_number_flags(SvPVX_const(sv), SvCUR(sv), NULL, PERL_SCAN_TRAILING);
8995 #ifdef PERL_PRESERVE_IVUV
8996 /* Got to punt this as an integer if needs be, but we don't issue
8997 warnings. Probably ought to make the sv_iv_please() that does
8998 the conversion if possible, and silently. */
8999 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
9000 /* Need to try really hard to see if it's an integer.
9001 9.22337203685478e+18 is an integer.
9002 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
9003 so $a="9.22337203685478e+18"; $a+0; $a++
9004 needs to be the same as $a="9.22337203685478e+18"; $a++
9011 /* sv_2iv *should* have made this an NV */
9012 if (flags & SVp_NOK) {
9013 (void)SvNOK_only(sv);
9014 SvNV_set(sv, SvNVX(sv) + 1.0);
9017 /* I don't think we can get here. Maybe I should assert this
9018 And if we do get here I suspect that sv_setnv will croak. NWC
9020 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_inc punt failed to convert '%s' to IOK or NOKp, UV=0x%" UVxf " NV=%" NVgf "\n",
9021 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
9023 #endif /* PERL_PRESERVE_IVUV */
9024 if (!numtype && ckWARN(WARN_NUMERIC))
9025 not_incrementable(sv);
9026 sv_setnv(sv,Atof(SvPVX_const(sv)) + 1.0);
9030 while (d >= SvPVX_const(sv)) {
9038 /* MKS: The original code here died if letters weren't consecutive.
9039 * at least it didn't have to worry about non-C locales. The
9040 * new code assumes that ('z'-'a')==('Z'-'A'), letters are
9041 * arranged in order (although not consecutively) and that only
9042 * [A-Za-z] are accepted by isALPHA in the C locale.
9044 if (isALPHA_FOLD_NE(*d, 'z')) {
9045 do { ++*d; } while (!isALPHA(*d));
9048 *(d--) -= 'z' - 'a';
9053 *(d--) -= 'z' - 'a' + 1;
9057 /* oh,oh, the number grew */
9058 SvGROW(sv, SvCUR(sv) + 2);
9059 SvCUR_set(sv, SvCUR(sv) + 1);
9060 for (d = SvPVX(sv) + SvCUR(sv); d > SvPVX_const(sv); d--)
9071 Auto-decrement of the value in the SV, doing string to numeric conversion
9072 if necessary. Handles 'get' magic and operator overloading.
9078 Perl_sv_dec(pTHX_ SV *const sv)
9087 =for apidoc sv_dec_nomg
9089 Auto-decrement of the value in the SV, doing string to numeric conversion
9090 if necessary. Handles operator overloading. Skips handling 'get' magic.
9096 Perl_sv_dec_nomg(pTHX_ SV *const sv)
9102 if (SvTHINKFIRST(sv)) {
9103 if (SvREADONLY(sv)) {
9104 Perl_croak_no_modify();
9108 if (SvAMAGIC(sv) && AMG_CALLunary(sv, dec_amg))
9110 i = PTR2IV(SvRV(sv));
9114 else sv_force_normal_flags(sv, 0);
9116 /* Unlike sv_inc we don't have to worry about string-never-numbers
9117 and keeping them magic. But we mustn't warn on punting */
9118 flags = SvFLAGS(sv);
9119 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
9120 /* It's publicly an integer, or privately an integer-not-float */
9121 #ifdef PERL_PRESERVE_IVUV
9125 if (SvUVX(sv) == 0) {
9126 (void)SvIOK_only(sv);
9130 (void)SvIOK_only_UV(sv);
9131 SvUV_set(sv, SvUVX(sv) - 1);
9134 if (SvIVX(sv) == IV_MIN) {
9135 sv_setnv(sv, (NV)IV_MIN);
9139 (void)SvIOK_only(sv);
9140 SvIV_set(sv, SvIVX(sv) - 1);
9145 if (flags & SVp_NOK) {
9148 const NV was = SvNVX(sv);
9149 if (LIKELY(!Perl_isinfnan(was)) &&
9150 NV_OVERFLOWS_INTEGERS_AT != 0.0 &&
9151 was <= -NV_OVERFLOWS_INTEGERS_AT) {
9152 /* diag_listed_as: Lost precision when %s %f by 1 */
9153 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
9154 "Lost precision when decrementing %" NVff " by 1",
9157 (void)SvNOK_only(sv);
9158 SvNV_set(sv, was - 1.0);
9163 /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */
9164 if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv)))
9165 Perl_croak_no_modify();
9167 if (!(flags & SVp_POK)) {
9168 if ((flags & SVTYPEMASK) < SVt_PVIV)
9169 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV) ? SVt_PVIV : SVt_IV);
9171 (void)SvIOK_only(sv);
9174 #ifdef PERL_PRESERVE_IVUV
9176 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), NULL);
9177 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
9178 /* Need to try really hard to see if it's an integer.
9179 9.22337203685478e+18 is an integer.
9180 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
9181 so $a="9.22337203685478e+18"; $a+0; $a--
9182 needs to be the same as $a="9.22337203685478e+18"; $a--
9189 /* sv_2iv *should* have made this an NV */
9190 if (flags & SVp_NOK) {
9191 (void)SvNOK_only(sv);
9192 SvNV_set(sv, SvNVX(sv) - 1.0);
9195 /* I don't think we can get here. Maybe I should assert this
9196 And if we do get here I suspect that sv_setnv will croak. NWC
9198 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_dec punt failed to convert '%s' to IOK or NOKp, UV=0x%" UVxf " NV=%" NVgf "\n",
9199 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
9202 #endif /* PERL_PRESERVE_IVUV */
9203 sv_setnv(sv,Atof(SvPVX_const(sv)) - 1.0); /* punt */
9206 /* this define is used to eliminate a chunk of duplicated but shared logic
9207 * it has the suffix __SV_C to signal that it isnt API, and isnt meant to be
9208 * used anywhere but here - yves
9210 #define PUSH_EXTEND_MORTAL__SV_C(AnSv) \
9212 SSize_t ix = ++PL_tmps_ix; \
9213 if (UNLIKELY(ix >= PL_tmps_max)) \
9214 ix = tmps_grow_p(ix); \
9215 PL_tmps_stack[ix] = (AnSv); \
9219 =for apidoc sv_mortalcopy
9221 Creates a new SV which is a copy of the original SV (using C<sv_setsv>).
9222 The new SV is marked as mortal. It will be destroyed "soon", either by an
9223 explicit call to C<FREETMPS>, or by an implicit call at places such as
9224 statement boundaries. See also C<L</sv_newmortal>> and C<L</sv_2mortal>>.
9229 /* Make a string that will exist for the duration of the expression
9230 * evaluation. Actually, it may have to last longer than that, but
9231 * hopefully we won't free it until it has been assigned to a
9232 * permanent location. */
9235 Perl_sv_mortalcopy_flags(pTHX_ SV *const oldstr, U32 flags)
9239 if (flags & SV_GMAGIC)
9240 SvGETMAGIC(oldstr); /* before new_SV, in case it dies */
9242 sv_setsv_flags(sv,oldstr,flags & ~SV_GMAGIC);
9243 PUSH_EXTEND_MORTAL__SV_C(sv);
9249 =for apidoc sv_newmortal
9251 Creates a new null SV which is mortal. The reference count of the SV is
9252 set to 1. It will be destroyed "soon", either by an explicit call to
9253 C<FREETMPS>, or by an implicit call at places such as statement boundaries.
9254 See also C<L</sv_mortalcopy>> and C<L</sv_2mortal>>.
9260 Perl_sv_newmortal(pTHX)
9265 SvFLAGS(sv) = SVs_TEMP;
9266 PUSH_EXTEND_MORTAL__SV_C(sv);
9272 =for apidoc newSVpvn_flags
9274 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9275 characters) into it. The reference count for the
9276 SV is set to 1. Note that if C<len> is zero, Perl will create a zero length
9277 string. You are responsible for ensuring that the source string is at least
9278 C<len> bytes long. If the C<s> argument is NULL the new SV will be undefined.
9279 Currently the only flag bits accepted are C<SVf_UTF8> and C<SVs_TEMP>.
9280 If C<SVs_TEMP> is set, then C<sv_2mortal()> is called on the result before
9281 returning. If C<SVf_UTF8> is set, C<s>
9282 is considered to be in UTF-8 and the
9283 C<SVf_UTF8> flag will be set on the new SV.
9284 C<newSVpvn_utf8()> is a convenience wrapper for this function, defined as
9286 #define newSVpvn_utf8(s, len, u) \
9287 newSVpvn_flags((s), (len), (u) ? SVf_UTF8 : 0)
9293 Perl_newSVpvn_flags(pTHX_ const char *const s, const STRLEN len, const U32 flags)
9297 /* All the flags we don't support must be zero.
9298 And we're new code so I'm going to assert this from the start. */
9299 assert(!(flags & ~(SVf_UTF8|SVs_TEMP)));
9301 sv_setpvn(sv,s,len);
9303 /* This code used to do a sv_2mortal(), however we now unroll the call to
9304 * sv_2mortal() and do what it does ourselves here. Since we have asserted
9305 * that flags can only have the SVf_UTF8 and/or SVs_TEMP flags set above we
9306 * can use it to enable the sv flags directly (bypassing SvTEMP_on), which
9307 * in turn means we dont need to mask out the SVf_UTF8 flag below, which
9308 * means that we eliminate quite a few steps than it looks - Yves
9309 * (explaining patch by gfx) */
9311 SvFLAGS(sv) |= flags;
9313 if(flags & SVs_TEMP){
9314 PUSH_EXTEND_MORTAL__SV_C(sv);
9321 =for apidoc sv_2mortal
9323 Marks an existing SV as mortal. The SV will be destroyed "soon", either
9324 by an explicit call to C<FREETMPS>, or by an implicit call at places such as
9325 statement boundaries. C<SvTEMP()> is turned on which means that the SV's
9326 string buffer can be "stolen" if this SV is copied. See also
9327 C<L</sv_newmortal>> and C<L</sv_mortalcopy>>.
9333 Perl_sv_2mortal(pTHX_ SV *const sv)
9340 PUSH_EXTEND_MORTAL__SV_C(sv);
9348 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9349 characters) into it. The reference count for the
9350 SV is set to 1. If C<len> is zero, Perl will compute the length using
9351 C<strlen()>, (which means if you use this option, that C<s> can't have embedded
9352 C<NUL> characters and has to have a terminating C<NUL> byte).
9354 This function can cause reliability issues if you are likely to pass in
9355 empty strings that are not null terminated, because it will run
9356 strlen on the string and potentially run past valid memory.
9358 Using L</newSVpvn> is a safer alternative for non C<NUL> terminated strings.
9359 For string literals use L</newSVpvs> instead. This function will work fine for
9360 C<NUL> terminated strings, but if you want to avoid the if statement on whether
9361 to call C<strlen> use C<newSVpvn> instead (calling C<strlen> yourself).
9367 Perl_newSVpv(pTHX_ const char *const s, const STRLEN len)
9372 sv_setpvn(sv, s, len || s == NULL ? len : strlen(s));
9377 =for apidoc newSVpvn
9379 Creates a new SV and copies a string into it, which may contain C<NUL> characters
9380 (C<\0>) and other binary data. The reference count for the SV is set to 1.
9381 Note that if C<len> is zero, Perl will create a zero length (Perl) string. You
9382 are responsible for ensuring that the source buffer is at least
9383 C<len> bytes long. If the C<buffer> argument is NULL the new SV will be
9390 Perl_newSVpvn(pTHX_ const char *const buffer, const STRLEN len)
9394 sv_setpvn(sv,buffer,len);
9399 =for apidoc newSVhek
9401 Creates a new SV from the hash key structure. It will generate scalars that
9402 point to the shared string table where possible. Returns a new (undefined)
9403 SV if C<hek> is NULL.
9409 Perl_newSVhek(pTHX_ const HEK *const hek)
9418 if (HEK_LEN(hek) == HEf_SVKEY) {
9419 return newSVsv(*(SV**)HEK_KEY(hek));
9421 const int flags = HEK_FLAGS(hek);
9422 if (flags & HVhek_WASUTF8) {
9424 Andreas would like keys he put in as utf8 to come back as utf8
9426 STRLEN utf8_len = HEK_LEN(hek);
9427 SV * const sv = newSV_type(SVt_PV);
9428 char *as_utf8 = (char *)bytes_to_utf8 ((U8*)HEK_KEY(hek), &utf8_len);
9429 /* bytes_to_utf8() allocates a new string, which we can repurpose: */
9430 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
9433 } else if (flags & HVhek_UNSHARED) {
9434 /* A hash that isn't using shared hash keys has to have
9435 the flag in every key so that we know not to try to call
9436 share_hek_hek on it. */
9438 SV * const sv = newSVpvn (HEK_KEY(hek), HEK_LEN(hek));
9443 /* This will be overwhelminly the most common case. */
9445 /* Inline most of newSVpvn_share(), because share_hek_hek() is far
9446 more efficient than sharepvn(). */
9450 sv_upgrade(sv, SVt_PV);
9451 SvPV_set(sv, (char *)HEK_KEY(share_hek_hek(hek)));
9452 SvCUR_set(sv, HEK_LEN(hek));
9464 =for apidoc newSVpvn_share
9466 Creates a new SV with its C<SvPVX_const> pointing to a shared string in the string
9467 table. If the string does not already exist in the table, it is
9468 created first. Turns on the C<SvIsCOW> flag (or C<READONLY>
9469 and C<FAKE> in 5.16 and earlier). If the C<hash> parameter
9470 is non-zero, that value is used; otherwise the hash is computed.
9471 The string's hash can later be retrieved from the SV
9472 with the C<SvSHARED_HASH()> macro. The idea here is
9473 that as the string table is used for shared hash keys these strings will have
9474 C<SvPVX_const == HeKEY> and hash lookup will avoid string compare.
9480 Perl_newSVpvn_share(pTHX_ const char *src, I32 len, U32 hash)
9484 bool is_utf8 = FALSE;
9485 const char *const orig_src = src;
9488 STRLEN tmplen = -len;
9490 /* See the note in hv.c:hv_fetch() --jhi */
9491 src = (char*)bytes_from_utf8((const U8*)src, &tmplen, &is_utf8);
9495 PERL_HASH(hash, src, len);
9497 /* The logic for this is inlined in S_mro_get_linear_isa_dfs(), so if it
9498 changes here, update it there too. */
9499 sv_upgrade(sv, SVt_PV);
9500 SvPV_set(sv, sharepvn(src, is_utf8?-len:len, hash));
9507 if (src != orig_src)
9513 =for apidoc newSVpv_share
9515 Like C<newSVpvn_share>, but takes a C<NUL>-terminated string instead of a
9522 Perl_newSVpv_share(pTHX_ const char *src, U32 hash)
9524 return newSVpvn_share(src, strlen(src), hash);
9527 #if defined(PERL_IMPLICIT_CONTEXT)
9529 /* pTHX_ magic can't cope with varargs, so this is a no-context
9530 * version of the main function, (which may itself be aliased to us).
9531 * Don't access this version directly.
9535 Perl_newSVpvf_nocontext(const char *const pat, ...)
9541 PERL_ARGS_ASSERT_NEWSVPVF_NOCONTEXT;
9543 va_start(args, pat);
9544 sv = vnewSVpvf(pat, &args);
9551 =for apidoc newSVpvf
9553 Creates a new SV and initializes it with the string formatted like
9560 Perl_newSVpvf(pTHX_ const char *const pat, ...)
9565 PERL_ARGS_ASSERT_NEWSVPVF;
9567 va_start(args, pat);
9568 sv = vnewSVpvf(pat, &args);
9573 /* backend for newSVpvf() and newSVpvf_nocontext() */
9576 Perl_vnewSVpvf(pTHX_ const char *const pat, va_list *const args)
9580 PERL_ARGS_ASSERT_VNEWSVPVF;
9583 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
9590 Creates a new SV and copies a floating point value into it.
9591 The reference count for the SV is set to 1.
9597 Perl_newSVnv(pTHX_ const NV n)
9609 Creates a new SV and copies an integer into it. The reference count for the
9616 Perl_newSViv(pTHX_ const IV i)
9622 /* Inlining ONLY the small relevant subset of sv_setiv here
9623 * for performance. Makes a significant difference. */
9625 /* We're starting from SVt_FIRST, so provided that's
9626 * actual 0, we don't have to unset any SV type flags
9627 * to promote to SVt_IV. */
9628 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9630 SET_SVANY_FOR_BODYLESS_IV(sv);
9631 SvFLAGS(sv) |= SVt_IV;
9643 Creates a new SV and copies an unsigned integer into it.
9644 The reference count for the SV is set to 1.
9650 Perl_newSVuv(pTHX_ const UV u)
9654 /* Inlining ONLY the small relevant subset of sv_setuv here
9655 * for performance. Makes a significant difference. */
9657 /* Using ivs is more efficient than using uvs - see sv_setuv */
9658 if (u <= (UV)IV_MAX) {
9659 return newSViv((IV)u);
9664 /* We're starting from SVt_FIRST, so provided that's
9665 * actual 0, we don't have to unset any SV type flags
9666 * to promote to SVt_IV. */
9667 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9669 SET_SVANY_FOR_BODYLESS_IV(sv);
9670 SvFLAGS(sv) |= SVt_IV;
9672 (void)SvIsUV_on(sv);
9681 =for apidoc newSV_type
9683 Creates a new SV, of the type specified. The reference count for the new SV
9690 Perl_newSV_type(pTHX_ const svtype type)
9695 ASSUME(SvTYPE(sv) == SVt_FIRST);
9696 if(type != SVt_FIRST)
9697 sv_upgrade(sv, type);
9702 =for apidoc newRV_noinc
9704 Creates an RV wrapper for an SV. The reference count for the original
9705 SV is B<not> incremented.
9711 Perl_newRV_noinc(pTHX_ SV *const tmpRef)
9715 PERL_ARGS_ASSERT_NEWRV_NOINC;
9719 /* We're starting from SVt_FIRST, so provided that's
9720 * actual 0, we don't have to unset any SV type flags
9721 * to promote to SVt_IV. */
9722 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9724 SET_SVANY_FOR_BODYLESS_IV(sv);
9725 SvFLAGS(sv) |= SVt_IV;
9730 SvRV_set(sv, tmpRef);
9735 /* newRV_inc is the official function name to use now.
9736 * newRV_inc is in fact #defined to newRV in sv.h
9740 Perl_newRV(pTHX_ SV *const sv)
9742 PERL_ARGS_ASSERT_NEWRV;
9744 return newRV_noinc(SvREFCNT_inc_simple_NN(sv));
9750 Creates a new SV which is an exact duplicate of the original SV.
9753 =for apidoc newSVsv_nomg
9755 Like C<newSVsv> but does not process get magic.
9761 Perl_newSVsv_flags(pTHX_ SV *const old, I32 flags)
9767 if (SvTYPE(old) == (svtype)SVTYPEMASK) {
9768 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), "semi-panic: attempt to dup freed string");
9771 /* Do this here, otherwise we leak the new SV if this croaks. */
9772 if (flags & SV_GMAGIC)
9775 sv_setsv_flags(sv, old, flags & ~SV_GMAGIC);
9780 =for apidoc sv_reset
9782 Underlying implementation for the C<reset> Perl function.
9783 Note that the perl-level function is vaguely deprecated.
9789 Perl_sv_reset(pTHX_ const char *s, HV *const stash)
9791 PERL_ARGS_ASSERT_SV_RESET;
9793 sv_resetpvn(*s ? s : NULL, strlen(s), stash);
9797 Perl_sv_resetpvn(pTHX_ const char *s, STRLEN len, HV * const stash)
9799 char todo[PERL_UCHAR_MAX+1];
9802 if (!stash || SvTYPE(stash) != SVt_PVHV)
9805 if (!s) { /* reset ?? searches */
9806 MAGIC * const mg = mg_find((const SV *)stash, PERL_MAGIC_symtab);
9808 const U32 count = mg->mg_len / sizeof(PMOP**);
9809 PMOP **pmp = (PMOP**) mg->mg_ptr;
9810 PMOP *const *const end = pmp + count;
9814 SvREADONLY_off(PL_regex_pad[(*pmp)->op_pmoffset]);
9816 (*pmp)->op_pmflags &= ~PMf_USED;
9824 /* reset variables */
9826 if (!HvARRAY(stash))
9829 Zero(todo, 256, char);
9833 I32 i = (unsigned char)*s;
9837 max = (unsigned char)*s++;
9838 for ( ; i <= max; i++) {
9841 for (i = 0; i <= (I32) HvMAX(stash); i++) {
9843 for (entry = HvARRAY(stash)[i];
9845 entry = HeNEXT(entry))
9850 if (!todo[(U8)*HeKEY(entry)])
9852 gv = MUTABLE_GV(HeVAL(entry));
9856 if (sv && !SvREADONLY(sv)) {
9857 SV_CHECK_THINKFIRST_COW_DROP(sv);
9858 if (!isGV(sv)) SvOK_off(sv);
9863 if (GvHV(gv) && !HvNAME_get(GvHV(gv))) {
9874 Using various gambits, try to get an IO from an SV: the IO slot if its a
9875 GV; or the recursive result if we're an RV; or the IO slot of the symbol
9876 named after the PV if we're a string.
9878 'Get' magic is ignored on the C<sv> passed in, but will be called on
9879 C<SvRV(sv)> if C<sv> is an RV.
9885 Perl_sv_2io(pTHX_ SV *const sv)
9890 PERL_ARGS_ASSERT_SV_2IO;
9892 switch (SvTYPE(sv)) {
9894 io = MUTABLE_IO(sv);
9898 if (isGV_with_GP(sv)) {
9899 gv = MUTABLE_GV(sv);
9902 Perl_croak(aTHX_ "Bad filehandle: %" HEKf,
9903 HEKfARG(GvNAME_HEK(gv)));
9909 Perl_croak(aTHX_ PL_no_usym, "filehandle");
9911 SvGETMAGIC(SvRV(sv));
9912 return sv_2io(SvRV(sv));
9914 gv = gv_fetchsv_nomg(sv, 0, SVt_PVIO);
9921 if (SvGMAGICAL(sv)) {
9922 newsv = sv_newmortal();
9923 sv_setsv_nomg(newsv, sv);
9925 Perl_croak(aTHX_ "Bad filehandle: %" SVf, SVfARG(newsv));
9935 Using various gambits, try to get a CV from an SV; in addition, try if
9936 possible to set C<*st> and C<*gvp> to the stash and GV associated with it.
9937 The flags in C<lref> are passed to C<gv_fetchsv>.
9943 Perl_sv_2cv(pTHX_ SV *sv, HV **const st, GV **const gvp, const I32 lref)
9948 PERL_ARGS_ASSERT_SV_2CV;
9955 switch (SvTYPE(sv)) {
9959 return MUTABLE_CV(sv);
9969 sv = amagic_deref_call(sv, to_cv_amg);
9972 if (SvTYPE(sv) == SVt_PVCV) {
9973 cv = MUTABLE_CV(sv);
9978 else if(SvGETMAGIC(sv), isGV_with_GP(sv))
9979 gv = MUTABLE_GV(sv);
9981 Perl_croak(aTHX_ "Not a subroutine reference");
9983 else if (isGV_with_GP(sv)) {
9984 gv = MUTABLE_GV(sv);
9987 gv = gv_fetchsv_nomg(sv, lref, SVt_PVCV);
9994 /* Some flags to gv_fetchsv mean don't really create the GV */
9995 if (!isGV_with_GP(gv)) {
10000 if (lref & ~GV_ADDMG && !GvCVu(gv)) {
10001 /* XXX this is probably not what they think they're getting.
10002 * It has the same effect as "sub name;", i.e. just a forward
10011 =for apidoc sv_true
10013 Returns true if the SV has a true value by Perl's rules.
10014 Use the C<SvTRUE> macro instead, which may call C<sv_true()> or may
10015 instead use an in-line version.
10021 Perl_sv_true(pTHX_ SV *const sv)
10026 const XPV* const tXpv = (XPV*)SvANY(sv);
10028 (tXpv->xpv_cur > 1 ||
10029 (tXpv->xpv_cur && *sv->sv_u.svu_pv != '0')))
10036 return SvIVX(sv) != 0;
10039 return SvNVX(sv) != 0.0;
10041 return sv_2bool(sv);
10047 =for apidoc sv_pvn_force
10049 Get a sensible string out of the SV somehow.
10050 A private implementation of the C<SvPV_force> macro for compilers which
10051 can't cope with complex macro expressions. Always use the macro instead.
10053 =for apidoc sv_pvn_force_flags
10055 Get a sensible string out of the SV somehow.
10056 If C<flags> has the C<SV_GMAGIC> bit set, will C<mg_get> on C<sv> if
10057 appropriate, else not. C<sv_pvn_force> and C<sv_pvn_force_nomg> are
10058 implemented in terms of this function.
10059 You normally want to use the various wrapper macros instead: see
10060 C<L</SvPV_force>> and C<L</SvPV_force_nomg>>.
10066 Perl_sv_pvn_force_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
10068 PERL_ARGS_ASSERT_SV_PVN_FORCE_FLAGS;
10070 if (flags & SV_GMAGIC) SvGETMAGIC(sv);
10071 if (SvTHINKFIRST(sv) && (!SvROK(sv) || SvREADONLY(sv)))
10072 sv_force_normal_flags(sv, 0);
10082 if (SvTYPE(sv) > SVt_PVLV
10083 || isGV_with_GP(sv))
10084 /* diag_listed_as: Can't coerce %s to %s in %s */
10085 Perl_croak(aTHX_ "Can't coerce %s to string in %s", sv_reftype(sv,0),
10087 s = sv_2pv_flags(sv, &len, flags &~ SV_GMAGIC);
10094 if (SvTYPE(sv) < SVt_PV ||
10095 s != SvPVX_const(sv)) { /* Almost, but not quite, sv_setpvn() */
10098 SvUPGRADE(sv, SVt_PV); /* Never FALSE */
10099 SvGROW(sv, len + 1);
10100 Move(s,SvPVX(sv),len,char);
10101 SvCUR_set(sv, len);
10102 SvPVX(sv)[len] = '\0';
10105 SvPOK_on(sv); /* validate pointer */
10107 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2pv(%s)\n",
10108 PTR2UV(sv),SvPVX_const(sv)));
10111 (void)SvPOK_only_UTF8(sv);
10112 return SvPVX_mutable(sv);
10116 =for apidoc sv_pvbyten_force
10118 The backend for the C<SvPVbytex_force> macro. Always use the macro
10125 Perl_sv_pvbyten_force(pTHX_ SV *const sv, STRLEN *const lp)
10127 PERL_ARGS_ASSERT_SV_PVBYTEN_FORCE;
10129 sv_pvn_force(sv,lp);
10130 sv_utf8_downgrade(sv,0);
10136 =for apidoc sv_pvutf8n_force
10138 The backend for the C<SvPVutf8x_force> macro. Always use the macro
10145 Perl_sv_pvutf8n_force(pTHX_ SV *const sv, STRLEN *const lp)
10147 PERL_ARGS_ASSERT_SV_PVUTF8N_FORCE;
10149 sv_pvn_force(sv,0);
10150 sv_utf8_upgrade_nomg(sv);
10156 =for apidoc sv_reftype
10158 Returns a string describing what the SV is a reference to.
10160 If ob is true and the SV is blessed, the string is the class name,
10161 otherwise it is the type of the SV, "SCALAR", "ARRAY" etc.
10167 Perl_sv_reftype(pTHX_ const SV *const sv, const int ob)
10169 PERL_ARGS_ASSERT_SV_REFTYPE;
10170 if (ob && SvOBJECT(sv)) {
10171 return SvPV_nolen_const(sv_ref(NULL, sv, ob));
10174 /* WARNING - There is code, for instance in mg.c, that assumes that
10175 * the only reason that sv_reftype(sv,0) would return a string starting
10176 * with 'L' or 'S' is that it is a LVALUE or a SCALAR.
10177 * Yes this a dodgy way to do type checking, but it saves practically reimplementing
10178 * this routine inside other subs, and it saves time.
10179 * Do not change this assumption without searching for "dodgy type check" in
10182 switch (SvTYPE(sv)) {
10197 case SVt_PVLV: return (char *) (SvROK(sv) ? "REF"
10198 /* tied lvalues should appear to be
10199 * scalars for backwards compatibility */
10200 : (isALPHA_FOLD_EQ(LvTYPE(sv), 't'))
10201 ? "SCALAR" : "LVALUE");
10202 case SVt_PVAV: return "ARRAY";
10203 case SVt_PVHV: return "HASH";
10204 case SVt_PVCV: return "CODE";
10205 case SVt_PVGV: return (char *) (isGV_with_GP(sv)
10206 ? "GLOB" : "SCALAR");
10207 case SVt_PVFM: return "FORMAT";
10208 case SVt_PVIO: return "IO";
10209 case SVt_INVLIST: return "INVLIST";
10210 case SVt_REGEXP: return "REGEXP";
10211 default: return "UNKNOWN";
10219 Returns a SV describing what the SV passed in is a reference to.
10221 dst can be a SV to be set to the description or NULL, in which case a
10222 mortal SV is returned.
10224 If ob is true and the SV is blessed, the description is the class
10225 name, otherwise it is the type of the SV, "SCALAR", "ARRAY" etc.
10231 Perl_sv_ref(pTHX_ SV *dst, const SV *const sv, const int ob)
10233 PERL_ARGS_ASSERT_SV_REF;
10236 dst = sv_newmortal();
10238 if (ob && SvOBJECT(sv)) {
10239 HvNAME_get(SvSTASH(sv))
10240 ? sv_sethek(dst, HvNAME_HEK(SvSTASH(sv)))
10241 : sv_setpvs(dst, "__ANON__");
10244 const char * reftype = sv_reftype(sv, 0);
10245 sv_setpv(dst, reftype);
10251 =for apidoc sv_isobject
10253 Returns a boolean indicating whether the SV is an RV pointing to a blessed
10254 object. If the SV is not an RV, or if the object is not blessed, then this
10261 Perl_sv_isobject(pTHX_ SV *sv)
10277 Returns a boolean indicating whether the SV is blessed into the specified
10278 class. This does not check for subtypes; use C<sv_derived_from> to verify
10279 an inheritance relationship.
10285 Perl_sv_isa(pTHX_ SV *sv, const char *const name)
10287 const char *hvname;
10289 PERL_ARGS_ASSERT_SV_ISA;
10299 hvname = HvNAME_get(SvSTASH(sv));
10303 return strEQ(hvname, name);
10307 =for apidoc newSVrv
10309 Creates a new SV for the existing RV, C<rv>, to point to. If C<rv> is not an
10310 RV then it will be upgraded to one. If C<classname> is non-null then the new
10311 SV will be blessed in the specified package. The new SV is returned and its
10312 reference count is 1. The reference count 1 is owned by C<rv>.
10318 Perl_newSVrv(pTHX_ SV *const rv, const char *const classname)
10322 PERL_ARGS_ASSERT_NEWSVRV;
10326 SV_CHECK_THINKFIRST_COW_DROP(rv);
10328 if (UNLIKELY( SvTYPE(rv) >= SVt_PVMG )) {
10329 const U32 refcnt = SvREFCNT(rv);
10333 SvREFCNT(rv) = refcnt;
10335 sv_upgrade(rv, SVt_IV);
10336 } else if (SvROK(rv)) {
10337 SvREFCNT_dec(SvRV(rv));
10339 prepare_SV_for_RV(rv);
10347 HV* const stash = gv_stashpv(classname, GV_ADD);
10348 (void)sv_bless(rv, stash);
10354 Perl_newSVavdefelem(pTHX_ AV *av, SSize_t ix, bool extendible)
10356 SV * const lv = newSV_type(SVt_PVLV);
10357 PERL_ARGS_ASSERT_NEWSVAVDEFELEM;
10359 sv_magic(lv, NULL, PERL_MAGIC_defelem, NULL, 0);
10360 LvTARG(lv) = SvREFCNT_inc_simple_NN(av);
10361 LvSTARGOFF(lv) = ix;
10362 LvTARGLEN(lv) = extendible ? 1 : (STRLEN)UV_MAX;
10367 =for apidoc sv_setref_pv
10369 Copies a pointer into a new SV, optionally blessing the SV. The C<rv>
10370 argument will be upgraded to an RV. That RV will be modified to point to
10371 the new SV. If the C<pv> argument is C<NULL>, then C<PL_sv_undef> will be placed
10372 into the SV. The C<classname> argument indicates the package for the
10373 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10374 will have a reference count of 1, and the RV will be returned.
10376 Do not use with other Perl types such as HV, AV, SV, CV, because those
10377 objects will become corrupted by the pointer copy process.
10379 Note that C<sv_setref_pvn> copies the string while this copies the pointer.
10385 Perl_sv_setref_pv(pTHX_ SV *const rv, const char *const classname, void *const pv)
10387 PERL_ARGS_ASSERT_SV_SETREF_PV;
10394 sv_setiv(newSVrv(rv,classname), PTR2IV(pv));
10399 =for apidoc sv_setref_iv
10401 Copies an integer into a new SV, optionally blessing the SV. The C<rv>
10402 argument will be upgraded to an RV. That RV will be modified to point to
10403 the new SV. The C<classname> argument indicates the package for the
10404 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10405 will have a reference count of 1, and the RV will be returned.
10411 Perl_sv_setref_iv(pTHX_ SV *const rv, const char *const classname, const IV iv)
10413 PERL_ARGS_ASSERT_SV_SETREF_IV;
10415 sv_setiv(newSVrv(rv,classname), iv);
10420 =for apidoc sv_setref_uv
10422 Copies an unsigned integer into a new SV, optionally blessing the SV. The C<rv>
10423 argument will be upgraded to an RV. That RV will be modified to point to
10424 the new SV. The C<classname> argument indicates the package for the
10425 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10426 will have a reference count of 1, and the RV will be returned.
10432 Perl_sv_setref_uv(pTHX_ SV *const rv, const char *const classname, const UV uv)
10434 PERL_ARGS_ASSERT_SV_SETREF_UV;
10436 sv_setuv(newSVrv(rv,classname), uv);
10441 =for apidoc sv_setref_nv
10443 Copies a double into a new SV, optionally blessing the SV. The C<rv>
10444 argument will be upgraded to an RV. That RV will be modified to point to
10445 the new SV. The C<classname> argument indicates the package for the
10446 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10447 will have a reference count of 1, and the RV will be returned.
10453 Perl_sv_setref_nv(pTHX_ SV *const rv, const char *const classname, const NV nv)
10455 PERL_ARGS_ASSERT_SV_SETREF_NV;
10457 sv_setnv(newSVrv(rv,classname), nv);
10462 =for apidoc sv_setref_pvn
10464 Copies a string into a new SV, optionally blessing the SV. The length of the
10465 string must be specified with C<n>. The C<rv> argument will be upgraded to
10466 an RV. That RV will be modified to point to the new SV. The C<classname>
10467 argument indicates the package for the blessing. Set C<classname> to
10468 C<NULL> to avoid the blessing. The new SV will have a reference count
10469 of 1, and the RV will be returned.
10471 Note that C<sv_setref_pv> copies the pointer while this copies the string.
10477 Perl_sv_setref_pvn(pTHX_ SV *const rv, const char *const classname,
10478 const char *const pv, const STRLEN n)
10480 PERL_ARGS_ASSERT_SV_SETREF_PVN;
10482 sv_setpvn(newSVrv(rv,classname), pv, n);
10487 =for apidoc sv_bless
10489 Blesses an SV into a specified package. The SV must be an RV. The package
10490 must be designated by its stash (see C<L</gv_stashpv>>). The reference count
10491 of the SV is unaffected.
10497 Perl_sv_bless(pTHX_ SV *const sv, HV *const stash)
10500 HV *oldstash = NULL;
10502 PERL_ARGS_ASSERT_SV_BLESS;
10506 Perl_croak(aTHX_ "Can't bless non-reference value");
10508 if (SvFLAGS(tmpRef) & (SVs_OBJECT|SVf_READONLY|SVf_PROTECT)) {
10509 if (SvREADONLY(tmpRef))
10510 Perl_croak_no_modify();
10511 if (SvOBJECT(tmpRef)) {
10512 oldstash = SvSTASH(tmpRef);
10515 SvOBJECT_on(tmpRef);
10516 SvUPGRADE(tmpRef, SVt_PVMG);
10517 SvSTASH_set(tmpRef, MUTABLE_HV(SvREFCNT_inc_simple(stash)));
10518 SvREFCNT_dec(oldstash);
10520 if(SvSMAGICAL(tmpRef))
10521 if(mg_find(tmpRef, PERL_MAGIC_ext) || mg_find(tmpRef, PERL_MAGIC_uvar))
10529 /* Downgrades a PVGV to a PVMG. If it's actually a PVLV, we leave the type
10530 * as it is after unglobbing it.
10533 PERL_STATIC_INLINE void
10534 S_sv_unglob(pTHX_ SV *const sv, U32 flags)
10538 SV * const temp = flags & SV_COW_DROP_PV ? NULL : sv_newmortal();
10540 PERL_ARGS_ASSERT_SV_UNGLOB;
10542 assert(SvTYPE(sv) == SVt_PVGV || SvTYPE(sv) == SVt_PVLV);
10544 if (!(flags & SV_COW_DROP_PV))
10545 gv_efullname3(temp, MUTABLE_GV(sv), "*");
10547 SvREFCNT_inc_simple_void_NN(sv_2mortal(sv));
10549 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
10550 && HvNAME_get(stash))
10551 mro_method_changed_in(stash);
10552 gp_free(MUTABLE_GV(sv));
10555 sv_del_backref(MUTABLE_SV(GvSTASH(sv)), sv);
10556 GvSTASH(sv) = NULL;
10559 if (GvNAME_HEK(sv)) {
10560 unshare_hek(GvNAME_HEK(sv));
10562 isGV_with_GP_off(sv);
10564 if(SvTYPE(sv) == SVt_PVGV) {
10565 /* need to keep SvANY(sv) in the right arena */
10566 xpvmg = new_XPVMG();
10567 StructCopy(SvANY(sv), xpvmg, XPVMG);
10568 del_XPVGV(SvANY(sv));
10571 SvFLAGS(sv) &= ~SVTYPEMASK;
10572 SvFLAGS(sv) |= SVt_PVMG;
10575 /* Intentionally not calling any local SET magic, as this isn't so much a
10576 set operation as merely an internal storage change. */
10577 if (flags & SV_COW_DROP_PV) SvOK_off(sv);
10578 else sv_setsv_flags(sv, temp, 0);
10580 if ((const GV *)sv == PL_last_in_gv)
10581 PL_last_in_gv = NULL;
10582 else if ((const GV *)sv == PL_statgv)
10587 =for apidoc sv_unref_flags
10589 Unsets the RV status of the SV, and decrements the reference count of
10590 whatever was being referenced by the RV. This can almost be thought of
10591 as a reversal of C<newSVrv>. The C<cflags> argument can contain
10592 C<SV_IMMEDIATE_UNREF> to force the reference count to be decremented
10593 (otherwise the decrementing is conditional on the reference count being
10594 different from one or the reference being a readonly SV).
10595 See C<L</SvROK_off>>.
10601 Perl_sv_unref_flags(pTHX_ SV *const ref, const U32 flags)
10603 SV* const target = SvRV(ref);
10605 PERL_ARGS_ASSERT_SV_UNREF_FLAGS;
10607 if (SvWEAKREF(ref)) {
10608 sv_del_backref(target, ref);
10609 SvWEAKREF_off(ref);
10610 SvRV_set(ref, NULL);
10613 SvRV_set(ref, NULL);
10615 /* You can't have a || SvREADONLY(target) here, as $a = $$a, where $a was
10616 assigned to as BEGIN {$a = \"Foo"} will fail. */
10617 if (SvREFCNT(target) != 1 || (flags & SV_IMMEDIATE_UNREF))
10618 SvREFCNT_dec_NN(target);
10619 else /* XXX Hack, but hard to make $a=$a->[1] work otherwise */
10620 sv_2mortal(target); /* Schedule for freeing later */
10624 =for apidoc sv_untaint
10626 Untaint an SV. Use C<SvTAINTED_off> instead.
10632 Perl_sv_untaint(pTHX_ SV *const sv)
10634 PERL_ARGS_ASSERT_SV_UNTAINT;
10635 PERL_UNUSED_CONTEXT;
10637 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10638 MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10645 =for apidoc sv_tainted
10647 Test an SV for taintedness. Use C<SvTAINTED> instead.
10653 Perl_sv_tainted(pTHX_ SV *const sv)
10655 PERL_ARGS_ASSERT_SV_TAINTED;
10656 PERL_UNUSED_CONTEXT;
10658 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10659 const MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10660 if (mg && (mg->mg_len & 1) )
10666 #ifndef NO_MATHOMS /* Can't move these to mathoms.c because call uiv_2buf(),
10667 private to this file */
10670 =for apidoc sv_setpviv
10672 Copies an integer into the given SV, also updating its string value.
10673 Does not handle 'set' magic. See C<L</sv_setpviv_mg>>.
10679 Perl_sv_setpviv(pTHX_ SV *const sv, const IV iv)
10681 /* The purpose of this union is to ensure that arr is aligned on
10682 a 2 byte boundary, because that is what uiv_2buf() requires */
10684 char arr[TYPE_CHARS(UV)];
10688 char * const ptr = uiv_2buf(buf.arr, iv, 0, 0, &ebuf);
10690 PERL_ARGS_ASSERT_SV_SETPVIV;
10692 sv_setpvn(sv, ptr, ebuf - ptr);
10696 =for apidoc sv_setpviv_mg
10698 Like C<sv_setpviv>, but also handles 'set' magic.
10704 Perl_sv_setpviv_mg(pTHX_ SV *const sv, const IV iv)
10706 PERL_ARGS_ASSERT_SV_SETPVIV_MG;
10708 sv_setpviv(sv, iv);
10712 #endif /* NO_MATHOMS */
10714 #if defined(PERL_IMPLICIT_CONTEXT)
10716 /* pTHX_ magic can't cope with varargs, so this is a no-context
10717 * version of the main function, (which may itself be aliased to us).
10718 * Don't access this version directly.
10722 Perl_sv_setpvf_nocontext(SV *const sv, const char *const pat, ...)
10727 PERL_ARGS_ASSERT_SV_SETPVF_NOCONTEXT;
10729 va_start(args, pat);
10730 sv_vsetpvf(sv, pat, &args);
10734 /* pTHX_ magic can't cope with varargs, so this is a no-context
10735 * version of the main function, (which may itself be aliased to us).
10736 * Don't access this version directly.
10740 Perl_sv_setpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10745 PERL_ARGS_ASSERT_SV_SETPVF_MG_NOCONTEXT;
10747 va_start(args, pat);
10748 sv_vsetpvf_mg(sv, pat, &args);
10754 =for apidoc sv_setpvf
10756 Works like C<sv_catpvf> but copies the text into the SV instead of
10757 appending it. Does not handle 'set' magic. See C<L</sv_setpvf_mg>>.
10763 Perl_sv_setpvf(pTHX_ SV *const sv, const char *const pat, ...)
10767 PERL_ARGS_ASSERT_SV_SETPVF;
10769 va_start(args, pat);
10770 sv_vsetpvf(sv, pat, &args);
10775 =for apidoc sv_vsetpvf
10777 Works like C<sv_vcatpvf> but copies the text into the SV instead of
10778 appending it. Does not handle 'set' magic. See C<L</sv_vsetpvf_mg>>.
10780 Usually used via its frontend C<sv_setpvf>.
10786 Perl_sv_vsetpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10788 PERL_ARGS_ASSERT_SV_VSETPVF;
10790 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10794 =for apidoc sv_setpvf_mg
10796 Like C<sv_setpvf>, but also handles 'set' magic.
10802 Perl_sv_setpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10806 PERL_ARGS_ASSERT_SV_SETPVF_MG;
10808 va_start(args, pat);
10809 sv_vsetpvf_mg(sv, pat, &args);
10814 =for apidoc sv_vsetpvf_mg
10816 Like C<sv_vsetpvf>, but also handles 'set' magic.
10818 Usually used via its frontend C<sv_setpvf_mg>.
10824 Perl_sv_vsetpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10826 PERL_ARGS_ASSERT_SV_VSETPVF_MG;
10828 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10832 #if defined(PERL_IMPLICIT_CONTEXT)
10834 /* pTHX_ magic can't cope with varargs, so this is a no-context
10835 * version of the main function, (which may itself be aliased to us).
10836 * Don't access this version directly.
10840 Perl_sv_catpvf_nocontext(SV *const sv, const char *const pat, ...)
10845 PERL_ARGS_ASSERT_SV_CATPVF_NOCONTEXT;
10847 va_start(args, pat);
10848 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10852 /* pTHX_ magic can't cope with varargs, so this is a no-context
10853 * version of the main function, (which may itself be aliased to us).
10854 * Don't access this version directly.
10858 Perl_sv_catpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10863 PERL_ARGS_ASSERT_SV_CATPVF_MG_NOCONTEXT;
10865 va_start(args, pat);
10866 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10873 =for apidoc sv_catpvf
10875 Processes its arguments like C<sv_catpvfn>, and appends the formatted
10876 output to an SV. As with C<sv_catpvfn> called with a non-null C-style
10877 variable argument list, argument reordering is not supported.
10878 If the appended data contains "wide" characters
10879 (including, but not limited to, SVs with a UTF-8 PV formatted with C<%s>,
10880 and characters >255 formatted with C<%c>), the original SV might get
10881 upgraded to UTF-8. Handles 'get' magic, but not 'set' magic. See
10882 C<L</sv_catpvf_mg>>. If the original SV was UTF-8, the pattern should be
10883 valid UTF-8; if the original SV was bytes, the pattern should be too.
10888 Perl_sv_catpvf(pTHX_ SV *const sv, const char *const pat, ...)
10892 PERL_ARGS_ASSERT_SV_CATPVF;
10894 va_start(args, pat);
10895 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10900 =for apidoc sv_vcatpvf
10902 Processes its arguments like C<sv_catpvfn> called with a non-null C-style
10903 variable argument list, and appends the formatted output
10904 to an SV. Does not handle 'set' magic. See C<L</sv_vcatpvf_mg>>.
10906 Usually used via its frontend C<sv_catpvf>.
10912 Perl_sv_vcatpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10914 PERL_ARGS_ASSERT_SV_VCATPVF;
10916 sv_vcatpvfn_flags(sv, pat, strlen(pat), args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10920 =for apidoc sv_catpvf_mg
10922 Like C<sv_catpvf>, but also handles 'set' magic.
10928 Perl_sv_catpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10932 PERL_ARGS_ASSERT_SV_CATPVF_MG;
10934 va_start(args, pat);
10935 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10941 =for apidoc sv_vcatpvf_mg
10943 Like C<sv_vcatpvf>, but also handles 'set' magic.
10945 Usually used via its frontend C<sv_catpvf_mg>.
10951 Perl_sv_vcatpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10953 PERL_ARGS_ASSERT_SV_VCATPVF_MG;
10955 sv_vcatpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10960 =for apidoc sv_vsetpvfn
10962 Works like C<sv_vcatpvfn> but copies the text into the SV instead of
10965 Usually used via one of its frontends C<sv_vsetpvf> and C<sv_vsetpvf_mg>.
10971 Perl_sv_vsetpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10972 va_list *const args, SV **const svargs, const Size_t sv_count, bool *const maybe_tainted)
10974 PERL_ARGS_ASSERT_SV_VSETPVFN;
10977 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, sv_count, maybe_tainted, 0);
10981 /* simplified inline Perl_sv_catpvn_nomg() when you know the SV's SvPOK */
10983 PERL_STATIC_INLINE void
10984 S_sv_catpvn_simple(pTHX_ SV *const sv, const char* const buf, const STRLEN len)
10986 STRLEN const need = len + SvCUR(sv) + 1;
10989 /* can't wrap as both len and SvCUR() are allocated in
10990 * memory and together can't consume all the address space
10992 assert(need > len);
10997 Copy(buf, end, len, char);
11000 SvCUR_set(sv, need - 1);
11005 * Warn of missing argument to sprintf. The value used in place of such
11006 * arguments should be &PL_sv_no; an undefined value would yield
11007 * inappropriate "use of uninit" warnings [perl #71000].
11010 S_warn_vcatpvfn_missing_argument(pTHX) {
11011 if (ckWARN(WARN_MISSING)) {
11012 Perl_warner(aTHX_ packWARN(WARN_MISSING), "Missing argument in %s",
11013 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11022 Perl_croak(aTHX_ "Integer overflow in format string for %s",
11023 (PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn"));
11027 /* Given an int i from the next arg (if args is true) or an sv from an arg
11028 * (if args is false), try to extract a STRLEN-ranged value from the arg,
11029 * with overflow checking.
11030 * Sets *neg to true if the value was negative (untouched otherwise.
11031 * Returns the absolute value.
11032 * As an extra margin of safety, it croaks if the returned value would
11033 * exceed the maximum value of a STRLEN / 4.
11037 S_sprintf_arg_num_val(pTHX_ va_list *const args, int i, SV *sv, bool *neg)
11051 if (UNLIKELY(SvIsUV(sv))) {
11052 UV uv = SvUV_nomg(sv);
11054 S_croak_overflow();
11058 iv = SvIV_nomg(sv);
11062 S_croak_overflow();
11068 if (iv > (IV)(((STRLEN)~0) / 4))
11069 S_croak_overflow();
11075 /* Returns true if c is in the range '1'..'9'
11076 * Written with the cast so it only needs one conditional test
11078 #define IS_1_TO_9(c) ((U8)(c - '1') <= 8)
11080 /* Read in and return a number. Updates *pattern to point to the char
11081 * following the number. Expects the first char to 1..9.
11082 * Croaks if the number exceeds 1/4 of the maximum value of STRLEN.
11083 * This is a belt-and-braces safety measure to complement any
11084 * overflow/wrap checks done in the main body of sv_vcatpvfn_flags.
11085 * It means that e.g. on a 32-bit system the width/precision can't be more
11086 * than 1G, which seems reasonable.
11090 S_expect_number(pTHX_ const char **const pattern)
11094 PERL_ARGS_ASSERT_EXPECT_NUMBER;
11096 assert(IS_1_TO_9(**pattern));
11098 var = *(*pattern)++ - '0';
11099 while (isDIGIT(**pattern)) {
11100 /* if var * 10 + 9 would exceed 1/4 max strlen, croak */
11101 if (var > ((((STRLEN)~0) / 4 - 9) / 10))
11102 S_croak_overflow();
11103 var = var * 10 + (*(*pattern)++ - '0');
11108 /* Implement a fast "%.0f": given a pointer to the end of a buffer (caller
11109 * ensures it's big enough), back fill it with the rounded integer part of
11110 * nv. Returns ptr to start of string, and sets *len to its length.
11111 * Returns NULL if not convertible.
11115 S_F0convert(NV nv, char *const endbuf, STRLEN *const len)
11117 const int neg = nv < 0;
11120 PERL_ARGS_ASSERT_F0CONVERT;
11122 assert(!Perl_isinfnan(nv));
11125 if (nv != 0.0 && nv < UV_MAX) {
11131 if (uv & 1 && uv == nv)
11132 uv--; /* Round to even */
11135 const unsigned dig = uv % 10;
11137 } while (uv /= 10);
11147 /* XXX maybe_tainted is never assigned to, so the doc above is lying. */
11150 Perl_sv_vcatpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11151 va_list *const args, SV **const svargs, const Size_t sv_count, bool *const maybe_tainted)
11153 PERL_ARGS_ASSERT_SV_VCATPVFN;
11155 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, sv_count, maybe_tainted, SV_GMAGIC|SV_SMAGIC);
11159 /* For the vcatpvfn code, we need a long double target in case
11160 * HAS_LONG_DOUBLE, even without USE_LONG_DOUBLE, so that we can printf
11161 * with long double formats, even without NV being long double. But we
11162 * call the target 'fv' instead of 'nv', since most of the time it is not
11163 * (most compilers these days recognize "long double", even if only as a
11164 * synonym for "double").
11166 #if defined(HAS_LONG_DOUBLE) && LONG_DOUBLESIZE > DOUBLESIZE && \
11167 defined(PERL_PRIgldbl) && !defined(USE_QUADMATH)
11168 # define VCATPVFN_FV_GF PERL_PRIgldbl
11169 # if defined(__VMS) && defined(__ia64) && defined(__IEEE_FLOAT)
11170 /* Work around breakage in OTS$CVT_FLOAT_T_X */
11171 # define VCATPVFN_NV_TO_FV(nv,fv) \
11174 fv = Perl_isnan(_dv) ? LDBL_QNAN : _dv; \
11177 # define VCATPVFN_NV_TO_FV(nv,fv) (fv)=(nv)
11179 typedef long double vcatpvfn_long_double_t;
11181 # define VCATPVFN_FV_GF NVgf
11182 # define VCATPVFN_NV_TO_FV(nv,fv) (fv)=(nv)
11183 typedef NV vcatpvfn_long_double_t;
11186 #ifdef LONGDOUBLE_DOUBLEDOUBLE
11187 /* The first double can be as large as 2**1023, or '1' x '0' x 1023.
11188 * The second double can be as small as 2**-1074, or '0' x 1073 . '1'.
11189 * The sum of them can be '1' . '0' x 2096 . '1', with implied radix point
11190 * after the first 1023 zero bits.
11192 * XXX The 2098 is quite large (262.25 bytes) and therefore some sort
11193 * of dynamically growing buffer might be better, start at just 16 bytes
11194 * (for example) and grow only when necessary. Or maybe just by looking
11195 * at the exponents of the two doubles? */
11196 # define DOUBLEDOUBLE_MAXBITS 2098
11199 /* vhex will contain the values (0..15) of the hex digits ("nybbles"
11200 * of 4 bits); 1 for the implicit 1, and the mantissa bits, four bits
11201 * per xdigit. For the double-double case, this can be rather many.
11202 * The non-double-double-long-double overshoots since all bits of NV
11203 * are not mantissa bits, there are also exponent bits. */
11204 #ifdef LONGDOUBLE_DOUBLEDOUBLE
11205 # define VHEX_SIZE (3+DOUBLEDOUBLE_MAXBITS/4)
11207 # define VHEX_SIZE (1+(NVSIZE * 8)/4)
11210 /* If we do not have a known long double format, (including not using
11211 * long doubles, or long doubles being equal to doubles) then we will
11212 * fall back to the ldexp/frexp route, with which we can retrieve at
11213 * most as many bits as our widest unsigned integer type is. We try
11214 * to get a 64-bit unsigned integer even if we are not using a 64-bit UV.
11216 * (If you want to test the case of UVSIZE == 4, NVSIZE == 8,
11217 * set the MANTISSATYPE to int and the MANTISSASIZE to 4.)
11219 #if defined(HAS_QUAD) && defined(Uquad_t)
11220 # define MANTISSATYPE Uquad_t
11221 # define MANTISSASIZE 8
11223 # define MANTISSATYPE UV
11224 # define MANTISSASIZE UVSIZE
11227 #if defined(DOUBLE_LITTLE_ENDIAN) || defined(LONGDOUBLE_LITTLE_ENDIAN)
11228 # define HEXTRACT_LITTLE_ENDIAN
11229 #elif defined(DOUBLE_BIG_ENDIAN) || defined(LONGDOUBLE_BIG_ENDIAN)
11230 # define HEXTRACT_BIG_ENDIAN
11232 # define HEXTRACT_MIX_ENDIAN
11235 /* S_hextract() is a helper for S_format_hexfp, for extracting
11236 * the hexadecimal values (for %a/%A). The nv is the NV where the value
11237 * are being extracted from (either directly from the long double in-memory
11238 * presentation, or from the uquad computed via frexp+ldexp). frexp also
11239 * is used to update the exponent. The subnormal is set to true
11240 * for IEEE 754 subnormals/denormals (including the x86 80-bit format).
11241 * The vhex is the pointer to the beginning of the output buffer of VHEX_SIZE.
11243 * The tricky part is that S_hextract() needs to be called twice:
11244 * the first time with vend as NULL, and the second time with vend as
11245 * the pointer returned by the first call. What happens is that on
11246 * the first round the output size is computed, and the intended
11247 * extraction sanity checked. On the second round the actual output
11248 * (the extraction of the hexadecimal values) takes place.
11249 * Sanity failures cause fatal failures during both rounds. */
11251 S_hextract(pTHX_ const NV nv, int* exponent, bool *subnormal,
11252 U8* vhex, U8* vend)
11256 int ixmin = 0, ixmax = 0;
11258 /* XXX Inf/NaN are not handled here, since it is
11259 * assumed they are to be output as "Inf" and "NaN". */
11261 /* These macros are just to reduce typos, they have multiple
11262 * repetitions below, but usually only one (or sometimes two)
11263 * of them is really being used. */
11264 /* HEXTRACT_OUTPUT() extracts the high nybble first. */
11265 #define HEXTRACT_OUTPUT_HI(ix) (*v++ = nvp[ix] >> 4)
11266 #define HEXTRACT_OUTPUT_LO(ix) (*v++ = nvp[ix] & 0xF)
11267 #define HEXTRACT_OUTPUT(ix) \
11269 HEXTRACT_OUTPUT_HI(ix); HEXTRACT_OUTPUT_LO(ix); \
11271 #define HEXTRACT_COUNT(ix, c) \
11273 v += c; if (ix < ixmin) ixmin = ix; else if (ix > ixmax) ixmax = ix; \
11275 #define HEXTRACT_BYTE(ix) \
11277 if (vend) HEXTRACT_OUTPUT(ix); else HEXTRACT_COUNT(ix, 2); \
11279 #define HEXTRACT_LO_NYBBLE(ix) \
11281 if (vend) HEXTRACT_OUTPUT_LO(ix); else HEXTRACT_COUNT(ix, 1); \
11283 /* HEXTRACT_TOP_NYBBLE is just convenience disguise,
11284 * to make it look less odd when the top bits of a NV
11285 * are extracted using HEXTRACT_LO_NYBBLE: the highest
11286 * order bits can be in the "low nybble" of a byte. */
11287 #define HEXTRACT_TOP_NYBBLE(ix) HEXTRACT_LO_NYBBLE(ix)
11288 #define HEXTRACT_BYTES_LE(a, b) \
11289 for (ix = a; ix >= b; ix--) { HEXTRACT_BYTE(ix); }
11290 #define HEXTRACT_BYTES_BE(a, b) \
11291 for (ix = a; ix <= b; ix++) { HEXTRACT_BYTE(ix); }
11292 #define HEXTRACT_GET_SUBNORMAL(nv) *subnormal = Perl_fp_class_denorm(nv)
11293 #define HEXTRACT_IMPLICIT_BIT(nv) \
11295 if (!*subnormal) { \
11296 if (vend) *v++ = ((nv) == 0.0) ? 0 : 1; else v++; \
11300 /* Most formats do. Those which don't should undef this.
11302 * But also note that IEEE 754 subnormals do not have it, or,
11303 * expressed alternatively, their implicit bit is zero. */
11304 #define HEXTRACT_HAS_IMPLICIT_BIT
11306 /* Many formats do. Those which don't should undef this. */
11307 #define HEXTRACT_HAS_TOP_NYBBLE
11309 /* HEXTRACTSIZE is the maximum number of xdigits. */
11310 #if defined(USE_LONG_DOUBLE) && defined(LONGDOUBLE_DOUBLEDOUBLE)
11311 # define HEXTRACTSIZE (2+DOUBLEDOUBLE_MAXBITS/4)
11313 # define HEXTRACTSIZE 2 * NVSIZE
11316 const U8* vmaxend = vhex + HEXTRACTSIZE;
11318 assert(HEXTRACTSIZE <= VHEX_SIZE);
11320 PERL_UNUSED_VAR(ix); /* might happen */
11321 (void)Perl_frexp(PERL_ABS(nv), exponent);
11322 *subnormal = FALSE;
11323 if (vend && (vend <= vhex || vend > vmaxend)) {
11324 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11325 Perl_croak(aTHX_ "Hexadecimal float: internal error (entry)");
11328 /* First check if using long doubles. */
11329 #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE)
11330 # if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN
11331 /* Used in e.g. VMS and HP-UX IA-64, e.g. -0.1L:
11332 * 9a 99 99 99 99 99 99 99 99 99 99 99 99 99 fb bf */
11333 /* The bytes 13..0 are the mantissa/fraction,
11334 * the 15,14 are the sign+exponent. */
11335 const U8* nvp = (const U8*)(&nv);
11336 HEXTRACT_GET_SUBNORMAL(nv);
11337 HEXTRACT_IMPLICIT_BIT(nv);
11338 # undef HEXTRACT_HAS_TOP_NYBBLE
11339 HEXTRACT_BYTES_LE(13, 0);
11340 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_BIG_ENDIAN
11341 /* Used in e.g. Solaris Sparc and HP-UX PA-RISC, e.g. -0.1L:
11342 * bf fb 99 99 99 99 99 99 99 99 99 99 99 99 99 9a */
11343 /* The bytes 2..15 are the mantissa/fraction,
11344 * the 0,1 are the sign+exponent. */
11345 const U8* nvp = (const U8*)(&nv);
11346 HEXTRACT_GET_SUBNORMAL(nv);
11347 HEXTRACT_IMPLICIT_BIT(nv);
11348 # undef HEXTRACT_HAS_TOP_NYBBLE
11349 HEXTRACT_BYTES_BE(2, 15);
11350 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN
11351 /* x86 80-bit "extended precision", 64 bits of mantissa / fraction /
11352 * significand, 15 bits of exponent, 1 bit of sign. No implicit bit.
11353 * NVSIZE can be either 12 (ILP32, Solaris x86) or 16 (LP64, Linux
11354 * and OS X), meaning that 2 or 6 bytes are empty padding. */
11355 /* The bytes 0..1 are the sign+exponent,
11356 * the bytes 2..9 are the mantissa/fraction. */
11357 const U8* nvp = (const U8*)(&nv);
11358 # undef HEXTRACT_HAS_IMPLICIT_BIT
11359 # undef HEXTRACT_HAS_TOP_NYBBLE
11360 HEXTRACT_GET_SUBNORMAL(nv);
11361 HEXTRACT_BYTES_LE(7, 0);
11362 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN
11363 /* Does this format ever happen? (Wikipedia says the Motorola
11364 * 6888x math coprocessors used format _like_ this but padded
11365 * to 96 bits with 16 unused bits between the exponent and the
11367 const U8* nvp = (const U8*)(&nv);
11368 # undef HEXTRACT_HAS_IMPLICIT_BIT
11369 # undef HEXTRACT_HAS_TOP_NYBBLE
11370 HEXTRACT_GET_SUBNORMAL(nv);
11371 HEXTRACT_BYTES_BE(0, 7);
11373 # define HEXTRACT_FALLBACK
11374 /* Double-double format: two doubles next to each other.
11375 * The first double is the high-order one, exactly like
11376 * it would be for a "lone" double. The second double
11377 * is shifted down using the exponent so that that there
11378 * are no common bits. The tricky part is that the value
11379 * of the double-double is the SUM of the two doubles and
11380 * the second one can be also NEGATIVE.
11382 * Because of this tricky construction the bytewise extraction we
11383 * use for the other long double formats doesn't work, we must
11384 * extract the values bit by bit.
11386 * The little-endian double-double is used .. somewhere?
11388 * The big endian double-double is used in e.g. PPC/Power (AIX)
11391 * The mantissa bits are in two separate stretches, e.g. for -0.1L:
11392 * 9a 99 99 99 99 99 59 bc 9a 99 99 99 99 99 b9 3f (LE)
11393 * 3f b9 99 99 99 99 99 9a bc 59 99 99 99 99 99 9a (BE)
11396 #else /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) */
11397 /* Using normal doubles, not long doubles.
11399 * We generate 4-bit xdigits (nybble/nibble) instead of 8-bit
11400 * bytes, since we might need to handle printf precision, and
11401 * also need to insert the radix. */
11403 # ifdef HEXTRACT_LITTLE_ENDIAN
11404 /* 0 1 2 3 4 5 6 7 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11405 const U8* nvp = (const U8*)(&nv);
11406 HEXTRACT_GET_SUBNORMAL(nv);
11407 HEXTRACT_IMPLICIT_BIT(nv);
11408 HEXTRACT_TOP_NYBBLE(6);
11409 HEXTRACT_BYTES_LE(5, 0);
11410 # elif defined(HEXTRACT_BIG_ENDIAN)
11411 /* 7 6 5 4 3 2 1 0 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11412 const U8* nvp = (const U8*)(&nv);
11413 HEXTRACT_GET_SUBNORMAL(nv);
11414 HEXTRACT_IMPLICIT_BIT(nv);
11415 HEXTRACT_TOP_NYBBLE(1);
11416 HEXTRACT_BYTES_BE(2, 7);
11417 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_LE_BE
11418 /* 4 5 6 7 0 1 2 3 (MSB = 7, LSB = 0, 6:7 = nybble:exponent:sign) */
11419 const U8* nvp = (const U8*)(&nv);
11420 HEXTRACT_GET_SUBNORMAL(nv);
11421 HEXTRACT_IMPLICIT_BIT(nv);
11422 HEXTRACT_TOP_NYBBLE(2); /* 6 */
11423 HEXTRACT_BYTE(1); /* 5 */
11424 HEXTRACT_BYTE(0); /* 4 */
11425 HEXTRACT_BYTE(7); /* 3 */
11426 HEXTRACT_BYTE(6); /* 2 */
11427 HEXTRACT_BYTE(5); /* 1 */
11428 HEXTRACT_BYTE(4); /* 0 */
11429 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_BE_LE
11430 /* 3 2 1 0 7 6 5 4 (MSB = 7, LSB = 0, 7:6 = sign:exponent:nybble) */
11431 const U8* nvp = (const U8*)(&nv);
11432 HEXTRACT_GET_SUBNORMAL(nv);
11433 HEXTRACT_IMPLICIT_BIT(nv);
11434 HEXTRACT_TOP_NYBBLE(5); /* 6 */
11435 HEXTRACT_BYTE(6); /* 5 */
11436 HEXTRACT_BYTE(7); /* 4 */
11437 HEXTRACT_BYTE(0); /* 3 */
11438 HEXTRACT_BYTE(1); /* 2 */
11439 HEXTRACT_BYTE(2); /* 1 */
11440 HEXTRACT_BYTE(3); /* 0 */
11442 # define HEXTRACT_FALLBACK
11445 # define HEXTRACT_FALLBACK
11447 #endif /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) #else */
11449 #ifdef HEXTRACT_FALLBACK
11450 HEXTRACT_GET_SUBNORMAL(nv);
11451 # undef HEXTRACT_HAS_TOP_NYBBLE /* Meaningless, but consistent. */
11452 /* The fallback is used for the double-double format, and
11453 * for unknown long double formats, and for unknown double
11454 * formats, or in general unknown NV formats. */
11455 if (nv == (NV)0.0) {
11463 NV d = nv < 0 ? -nv : nv;
11465 U8 ha = 0x0; /* hexvalue accumulator */
11466 U8 hd = 0x8; /* hexvalue digit */
11468 /* Shift d and e (and update exponent) so that e <= d < 2*e,
11469 * this is essentially manual frexp(). Multiplying by 0.5 and
11470 * doubling should be lossless in binary floating point. */
11480 while (d >= e + e) {
11484 /* Now e <= d < 2*e */
11486 /* First extract the leading hexdigit (the implicit bit). */
11502 /* Then extract the remaining hexdigits. */
11503 while (d > (NV)0.0) {
11509 /* Output or count in groups of four bits,
11510 * that is, when the hexdigit is down to one. */
11515 /* Reset the hexvalue. */
11524 /* Flush possible pending hexvalue. */
11534 /* Croak for various reasons: if the output pointer escaped the
11535 * output buffer, if the extraction index escaped the extraction
11536 * buffer, or if the ending output pointer didn't match the
11537 * previously computed value. */
11538 if (v <= vhex || v - vhex >= VHEX_SIZE ||
11539 /* For double-double the ixmin and ixmax stay at zero,
11540 * which is convenient since the HEXTRACTSIZE is tricky
11541 * for double-double. */
11542 ixmin < 0 || ixmax >= NVSIZE ||
11543 (vend && v != vend)) {
11544 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11545 Perl_croak(aTHX_ "Hexadecimal float: internal error (overflow)");
11551 /* S_format_hexfp(): helper function for Perl_sv_vcatpvfn_flags().
11553 * Processes the %a/%A hexadecimal floating-point format, since the
11554 * built-in snprintf()s which are used for most of the f/p formats, don't
11555 * universally handle %a/%A.
11556 * Populates buf of length bufsize, and returns the length of the created
11558 * The rest of the args have the same meaning as the local vars of the
11559 * same name within Perl_sv_vcatpvfn_flags().
11561 * It assumes the caller has already done STORE_LC_NUMERIC_SET_TO_NEEDED();
11563 * It requires the caller to make buf large enough.
11567 S_format_hexfp(pTHX_ char * const buf, const STRLEN bufsize, const char c,
11568 const NV nv, const vcatpvfn_long_double_t fv,
11569 bool has_precis, STRLEN precis, STRLEN width,
11570 bool alt, char plus, bool left, bool fill)
11572 /* Hexadecimal floating point. */
11574 U8 vhex[VHEX_SIZE];
11575 U8* v = vhex; /* working pointer to vhex */
11576 U8* vend; /* pointer to one beyond last digit of vhex */
11577 U8* vfnz = NULL; /* first non-zero */
11578 U8* vlnz = NULL; /* last non-zero */
11579 U8* v0 = NULL; /* first output */
11580 const bool lower = (c == 'a');
11581 /* At output the values of vhex (up to vend) will
11582 * be mapped through the xdig to get the actual
11583 * human-readable xdigits. */
11584 const char* xdig = PL_hexdigit;
11585 STRLEN zerotail = 0; /* how many extra zeros to append */
11586 int exponent = 0; /* exponent of the floating point input */
11587 bool hexradix = FALSE; /* should we output the radix */
11588 bool subnormal = FALSE; /* IEEE 754 subnormal/denormal */
11589 bool negative = FALSE;
11592 /* XXX: NaN, Inf -- though they are printed as "NaN" and "Inf".
11594 * For example with denormals, (assuming the vanilla
11595 * 64-bit double): the exponent is zero. 1xp-1074 is
11596 * the smallest denormal and the smallest double, it
11597 * could be output also as 0x0.0000000000001p-1022 to
11598 * match its internal structure. */
11600 vend = S_hextract(aTHX_ nv, &exponent, &subnormal, vhex, NULL);
11601 S_hextract(aTHX_ nv, &exponent, &subnormal, vhex, vend);
11603 #if NVSIZE > DOUBLESIZE
11604 # ifdef HEXTRACT_HAS_IMPLICIT_BIT
11605 /* In this case there is an implicit bit,
11606 * and therefore the exponent is shifted by one. */
11608 # elif defined(NV_X86_80_BIT)
11610 /* The subnormals of the x86-80 have a base exponent of -16382,
11611 * (while the physical exponent bits are zero) but the frexp()
11612 * returned the scientific-style floating exponent. We want
11613 * to map the last one as:
11614 * -16831..-16384 -> -16382 (the last normal is 0x1p-16382)
11615 * -16835..-16388 -> -16384
11616 * since we want to keep the first hexdigit
11617 * as one of the [8421]. */
11618 exponent = -4 * ( (exponent + 1) / -4) - 2;
11622 /* TBD: other non-implicit-bit platforms than the x86-80. */
11626 negative = fv < 0 || Perl_signbit(nv);
11637 xdig += 16; /* Use uppercase hex. */
11640 /* Find the first non-zero xdigit. */
11641 for (v = vhex; v < vend; v++) {
11649 /* Find the last non-zero xdigit. */
11650 for (v = vend - 1; v >= vhex; v--) {
11657 #if NVSIZE == DOUBLESIZE
11663 #ifndef NV_X86_80_BIT
11665 /* IEEE 754 subnormals (but not the x86 80-bit):
11666 * we want "normalize" the subnormal,
11667 * so we need to right shift the hex nybbles
11668 * so that the output of the subnormal starts
11669 * from the first true bit. (Another, equally
11670 * valid, policy would be to dump the subnormal
11671 * nybbles as-is, to display the "physical" layout.) */
11674 /* Find the ceil(log2(v[0])) of
11675 * the top non-zero nybble. */
11676 for (i = vfnz[0], n = 0; i > 1; i >>= 1, n++) { }
11680 for (vshr = vlnz; vshr >= vfnz; vshr--) {
11681 vshr[1] |= (vshr[0] & (0xF >> (4 - n))) << (4 - n);
11695 U8* ve = (subnormal ? vlnz + 1 : vend);
11696 SSize_t vn = ve - v0;
11698 if (precis < (Size_t)(vn - 1)) {
11699 bool overflow = FALSE;
11700 if (v0[precis + 1] < 0x8) {
11701 /* Round down, nothing to do. */
11702 } else if (v0[precis + 1] > 0x8) {
11705 overflow = v0[precis] > 0xF;
11707 } else { /* v0[precis] == 0x8 */
11708 /* Half-point: round towards the one
11709 * with the even least-significant digit:
11717 * 78 -> 8 f8 -> 10 */
11718 if ((v0[precis] & 0x1)) {
11721 overflow = v0[precis] > 0xF;
11726 for (v = v0 + precis - 1; v >= v0; v--) {
11728 overflow = *v > 0xF;
11734 if (v == v0 - 1 && overflow) {
11735 /* If the overflow goes all the
11736 * way to the front, we need to
11737 * insert 0x1 in front, and adjust
11739 Move(v0, v0 + 1, vn - 1, char);
11745 /* The new effective "last non zero". */
11746 vlnz = v0 + precis;
11750 subnormal ? precis - vn + 1 :
11751 precis - (vlnz - vhex);
11758 /* If there are non-zero xdigits, the radix
11759 * is output after the first one. */
11770 /* The radix is always output if precis, or if alt. */
11771 if (precis > 0 || alt) {
11776 #ifndef USE_LOCALE_NUMERIC
11779 if (IN_LC(LC_NUMERIC)) {
11781 const char* r = SvPV(PL_numeric_radix_sv, n);
11782 Copy(r, p, n, char);
11796 if (zerotail > 0) {
11797 while (zerotail--) {
11804 /* sanity checks */
11805 if (elen >= bufsize || width >= bufsize)
11806 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11807 Perl_croak(aTHX_ "Hexadecimal float: internal error (overflow)");
11809 elen += my_snprintf(p, bufsize - elen,
11810 "%c%+d", lower ? 'p' : 'P',
11813 if (elen < width) {
11814 STRLEN gap = (STRLEN)(width - elen);
11816 /* Pad the back with spaces. */
11817 memset(buf + elen, ' ', gap);
11820 /* Insert the zeros after the "0x" and the
11821 * the potential sign, but before the digits,
11822 * otherwise we end up with "0000xH.HHH...",
11823 * when we want "0x000H.HHH..." */
11824 STRLEN nzero = gap;
11825 char* zerox = buf + 2;
11826 STRLEN nmove = elen - 2;
11827 if (negative || plus) {
11831 Move(zerox, zerox + nzero, nmove, char);
11832 memset(zerox, fill ? '0' : ' ', nzero);
11835 /* Move it to the right. */
11836 Move(buf, buf + gap,
11838 /* Pad the front with spaces. */
11839 memset(buf, ' ', gap);
11848 =for apidoc sv_vcatpvfn
11850 =for apidoc sv_vcatpvfn_flags
11852 Processes its arguments like C<vsprintf> and appends the formatted output
11853 to an SV. Uses an array of SVs if the C-style variable argument list is
11854 missing (C<NULL>). Argument reordering (using format specifiers like C<%2$d>
11855 or C<%*2$d>) is supported only when using an array of SVs; using a C-style
11856 C<va_list> argument list with a format string that uses argument reordering
11857 will yield an exception.
11859 When running with taint checks enabled, indicates via
11860 C<maybe_tainted> if results are untrustworthy (often due to the use of
11863 If called as C<sv_vcatpvfn> or flags has the C<SV_GMAGIC> bit set, calls get magic.
11865 It assumes that pat has the same utf8-ness as sv. It's the caller's
11866 responsibility to ensure that this is so.
11868 Usually used via one of its frontends C<sv_vcatpvf> and C<sv_vcatpvf_mg>.
11875 Perl_sv_vcatpvfn_flags(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11876 va_list *const args, SV **const svargs, const Size_t sv_count, bool *const maybe_tainted,
11879 const char *fmtstart; /* character following the current '%' */
11880 const char *q; /* current position within format */
11881 const char *patend;
11884 static const char nullstr[] = "(null)";
11885 bool has_utf8 = DO_UTF8(sv); /* has the result utf8? */
11886 const bool pat_utf8 = has_utf8; /* the pattern is in utf8? */
11887 /* Times 4: a decimal digit takes more than 3 binary digits.
11888 * NV_DIG: mantissa takes that many decimal digits.
11889 * Plus 32: Playing safe. */
11890 char ebuf[IV_DIG * 4 + NV_DIG + 32];
11891 bool no_redundant_warning = FALSE; /* did we use any explicit format parameter index? */
11892 #ifdef USE_LOCALE_NUMERIC
11893 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
11894 bool lc_numeric_set = FALSE; /* called STORE_LC_NUMERIC_SET_TO_NEEDED? */
11897 PERL_ARGS_ASSERT_SV_VCATPVFN_FLAGS;
11898 PERL_UNUSED_ARG(maybe_tainted);
11900 if (flags & SV_GMAGIC)
11903 /* no matter what, this is a string now */
11904 (void)SvPV_force_nomg(sv, origlen);
11906 /* the code that scans for flags etc following a % relies on
11907 * a '\0' being present to avoid falling off the end. Ideally that
11908 * should be fixed */
11909 assert(pat[patlen] == '\0');
11912 /* Special-case "", "%s", "%-p" (SVf - see below) and "%.0f".
11913 * In each case, if there isn't the correct number of args, instead
11914 * fall through to the main code to handle the issuing of any
11918 if (patlen == 0 && (args || sv_count == 0))
11921 if (patlen <= 4 && pat[0] == '%' && (args || sv_count == 1)) {
11924 if (patlen == 2 && pat[1] == 's') {
11926 const char * const s = va_arg(*args, char*);
11927 sv_catpv_nomg(sv, s ? s : nullstr);
11930 /* we want get magic on the source but not the target.
11931 * sv_catsv can't do that, though */
11932 SvGETMAGIC(*svargs);
11933 sv_catsv_nomg(sv, *svargs);
11940 if (patlen == 3 && pat[1] == '-' && pat[2] == 'p') {
11941 SV *asv = MUTABLE_SV(va_arg(*args, void*));
11942 sv_catsv_nomg(sv, asv);
11946 #if !defined(USE_LONG_DOUBLE) && !defined(USE_QUADMATH)
11947 /* special-case "%.0f" */
11948 else if ( patlen == 4
11949 && pat[1] == '.' && pat[2] == '0' && pat[3] == 'f')
11951 const NV nv = SvNV(*svargs);
11952 if (LIKELY(!Perl_isinfnan(nv))) {
11956 if ((p = F0convert(nv, ebuf + sizeof ebuf, &l))) {
11957 sv_catpvn_nomg(sv, p, l);
11962 #endif /* !USE_LONG_DOUBLE */
11966 patend = (char*)pat + patlen;
11967 for (fmtstart = pat; fmtstart < patend; fmtstart = q) {
11968 char intsize = 0; /* size qualifier in "%hi..." etc */
11969 bool alt = FALSE; /* has "%#..." */
11970 bool left = FALSE; /* has "%-..." */
11971 bool fill = FALSE; /* has "%0..." */
11972 char plus = 0; /* has "%+..." */
11973 STRLEN width = 0; /* value of "%NNN..." */
11974 bool has_precis = FALSE; /* has "%.NNN..." */
11975 STRLEN precis = 0; /* value of "%.NNN..." */
11976 int base = 0; /* base to print in, e.g. 8 for %o */
11977 UV uv = 0; /* the value to print of int-ish args */
11979 bool vectorize = FALSE; /* has "%v..." */
11980 bool vec_utf8 = FALSE; /* SvUTF8(vec arg) */
11981 const U8 *vecstr = NULL; /* SvPVX(vec arg) */
11982 STRLEN veclen = 0; /* SvCUR(vec arg) */
11983 const char *dotstr = NULL; /* separator string for %v */
11984 STRLEN dotstrlen; /* length of separator string for %v */
11986 Size_t efix = 0; /* explicit format parameter index */
11987 const Size_t osvix = svix; /* original index in case of bad fmt */
11990 bool is_utf8 = FALSE; /* is this item utf8? */
11991 bool arg_missing = FALSE; /* give "Missing argument" warning */
11992 char esignbuf[4]; /* holds sign prefix, e.g. "-0x" */
11993 STRLEN esignlen = 0; /* length of e.g. "-0x" */
11994 STRLEN zeros = 0; /* how many '0' to prepend */
11996 const char *eptr = NULL; /* the address of the element string */
11997 STRLEN elen = 0; /* the length of the element string */
11999 char c; /* the actual format ('d', s' etc) */
12002 /* echo everything up to the next format specification */
12003 for (q = fmtstart; q < patend && *q != '%'; ++q)
12006 if (q > fmtstart) {
12007 if (has_utf8 && !pat_utf8) {
12008 /* upgrade and copy the bytes of fmtstart..q-1 to utf8 on
12012 STRLEN need = SvCUR(sv) + (q - fmtstart) + 1;
12014 for (p = fmtstart; p < q; p++)
12015 if (!NATIVE_BYTE_IS_INVARIANT(*p))
12020 for (p = fmtstart; p < q; p++)
12021 append_utf8_from_native_byte((U8)*p, (U8**)&dst);
12023 SvCUR_set(sv, need - 1);
12026 S_sv_catpvn_simple(aTHX_ sv, fmtstart, q - fmtstart);
12031 fmtstart = q; /* fmtstart is char following the '%' */
12034 We allow format specification elements in this order:
12035 \d+\$ explicit format parameter index
12037 v|\*(\d+\$)?v vector with optional (optionally specified) arg
12038 0 flag (as above): repeated to allow "v02"
12039 \d+|\*(\d+\$)? width using optional (optionally specified) arg
12040 \.(\d*|\*(\d+\$)?) precision using optional (optionally specified) arg
12042 [%bcdefginopsuxDFOUX] format (mandatory)
12045 if (IS_1_TO_9(*q)) {
12046 width = expect_number(&q);
12049 Perl_croak_nocontext(
12050 "Cannot yet reorder sv_catpvfn() arguments from va_list");
12052 efix = (Size_t)width;
12054 no_redundant_warning = TRUE;
12066 if (plus == '+' && *q == ' ') /* '+' over ' ' */
12093 /* at this point we can expect one of:
12095 * 123 an explicit width
12096 * * width taken from next arg
12097 * *12$ width taken from 12th arg
12100 * But any width specification may be preceded by a v, in one of its
12105 * So an asterisk may be either a width specifier or a vector
12106 * separator arg specifier, and we don't know which initially
12111 STRLEN ix; /* explicit width/vector separator index */
12113 if (IS_1_TO_9(*q)) {
12114 ix = expect_number(&q);
12117 Perl_croak_nocontext(
12118 "Cannot yet reorder sv_catpvfn() arguments from va_list");
12119 no_redundant_warning = TRUE;
12128 /* The asterisk was for *v, *NNN$v: vectorizing, but not
12129 * with the default "." */
12134 vecsv = va_arg(*args, SV*);
12136 ix = ix ? ix - 1 : svix++;
12137 vecsv = ix < sv_count ? svargs[ix]
12138 : (arg_missing = TRUE, &PL_sv_no);
12140 dotstr = SvPV_const(vecsv, dotstrlen);
12141 /* Keep the DO_UTF8 test *after* the SvPV call, else things go
12142 bad with tied or overloaded values that return UTF8. */
12143 if (DO_UTF8(vecsv))
12145 else if (has_utf8) {
12146 vecsv = sv_mortalcopy(vecsv);
12147 sv_utf8_upgrade(vecsv);
12148 dotstr = SvPV_const(vecsv, dotstrlen);
12155 /* the asterisk specified a width */
12160 i = va_arg(*args, int);
12162 ix = ix ? ix - 1 : svix++;
12163 sv = (ix < sv_count) ? svargs[ix]
12164 : (arg_missing = TRUE, (SV*)NULL);
12166 width = S_sprintf_arg_num_val(aTHX_ args, i, sv, &left);
12169 else if (*q == 'v') {
12180 /* explicit width? */
12186 width = expect_number(&q);
12196 STRLEN ix; /* explicit precision index */
12198 if (IS_1_TO_9(*q)) {
12199 ix = expect_number(&q);
12202 Perl_croak_nocontext(
12203 "Cannot yet reorder sv_catpvfn() arguments from va_list");
12204 no_redundant_warning = TRUE;
12217 i = va_arg(*args, int);
12219 ix = ix ? ix - 1 : svix++;
12220 sv = (ix < sv_count) ? svargs[ix]
12221 : (arg_missing = TRUE, (SV*)NULL);
12223 precis = S_sprintf_arg_num_val(aTHX_ args, i, sv, &neg);
12228 /* although it doesn't seem documented, this code has long
12230 * no digits following the '.' is treated like '.0'
12231 * the number may be preceded by any number of zeroes,
12232 * e.g. "%.0001f", which is the same as "%.1f"
12233 * so I've kept that behaviour. DAPM May 2017
12237 precis = IS_1_TO_9(*q) ? expect_number(&q) : 0;
12246 case 'I': /* Ix, I32x, and I64x */
12247 # ifdef USE_64_BIT_INT
12248 if (q[1] == '6' && q[2] == '4') {
12254 if (q[1] == '3' && q[2] == '2') {
12258 # ifdef USE_64_BIT_INT
12264 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
12265 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
12268 # ifdef USE_QUADMATH
12281 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
12282 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
12283 if (*q == 'l') { /* lld, llf */
12292 if (*++q == 'h') { /* hhd, hhu */
12309 c = *q++; /* c now holds the conversion type */
12311 /* '%' doesn't have an arg, so skip arg processing */
12320 if (vectorize && !strchr("BbDdiOouUXx", c))
12323 /* get next arg (individual branches do their own va_arg()
12324 * handling for the args case) */
12327 efix = efix ? efix - 1 : svix++;
12328 argsv = efix < sv_count ? svargs[efix]
12329 : (arg_missing = TRUE, &PL_sv_no);
12339 eptr = va_arg(*args, char*);
12342 elen = my_strnlen(eptr, precis);
12344 elen = strlen(eptr);
12346 eptr = (char *)nullstr;
12347 elen = sizeof nullstr - 1;
12351 eptr = SvPV_const(argsv, elen);
12352 if (DO_UTF8(argsv)) {
12353 STRLEN old_precis = precis;
12354 if (has_precis && precis < elen) {
12355 STRLEN ulen = sv_or_pv_len_utf8(argsv, eptr, elen);
12356 STRLEN p = precis > ulen ? ulen : precis;
12357 precis = sv_or_pv_pos_u2b(argsv, eptr, p, 0);
12358 /* sticks at end */
12360 if (width) { /* fudge width (can't fudge elen) */
12361 if (has_precis && precis < elen)
12362 width += precis - old_precis;
12365 elen - sv_or_pv_len_utf8(argsv,eptr,elen);
12372 if (has_precis && precis < elen)
12384 * "%...p" is normally treated like "%...x", except that the
12385 * number to print is the SV's address (or a pointer address
12386 * for C-ish sprintf).
12388 * However, the C-ish sprintf variant allows a few special
12389 * extensions. These are currently:
12391 * %-p (SVf) Like %s, but gets the string from an SV*
12392 * arg rather than a char* arg.
12393 * (This was previously %_).
12395 * %-<num>p Ditto but like %.<num>s (i.e. num is max width)
12397 * %2p (HEKf) Like %s, but using the key string in a HEK
12399 * %3p (HEKf256) Ditto but like %.256s
12401 * %d%lu%4p (UTF8f) A utf8 string. Consumes 3 args:
12402 * (cBOOL(utf8), len, string_buf).
12403 * It's handled by the "case 'd'" branch
12404 * rather than here.
12406 * %<num>p where num is 1 or > 4: reserved for future
12407 * extensions. Warns, but then is treated as a
12408 * general %p (print hex address) format.
12416 /* not %*p or %*1$p - any width was explicit */
12420 if (left) { /* %-p (SVf), %-NNNp */
12425 argsv = MUTABLE_SV(va_arg(*args, void*));
12426 eptr = SvPV_const(argsv, elen);
12427 if (DO_UTF8(argsv))
12432 else if (width == 2 || width == 3) { /* HEKf, HEKf256 */
12433 HEK * const hek = va_arg(*args, HEK *);
12434 eptr = HEK_KEY(hek);
12435 elen = HEK_LEN(hek);
12446 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
12447 "internal %%<num>p might conflict with future printf extensions");
12451 /* treat as normal %...p */
12453 uv = PTR2UV(args ? va_arg(*args, void*) : argsv);
12458 /* Ignore any size specifiers, since they're not documented as
12459 * being allowed for %c (ideally we should warn on e.g. '%hc').
12460 * Setting a default intsize, along with a positive
12461 * (which signals unsigned) base, causes, for C-ish use, the
12462 * va_arg to be interpreted as as unsigned int, when it's
12463 * actually signed, which will convert -ve values to high +ve
12464 * values. Note that unlike the libc %c, values > 255 will
12465 * convert to high unicode points rather than being truncated
12466 * to 8 bits. For perlish use, it will do SvUV(argsv), which
12467 * will again convert -ve args to high -ve values.
12470 base = 1; /* special value that indicates we're doing a 'c' */
12471 goto get_int_arg_val;
12480 goto get_int_arg_val;
12483 /* probably just a plain %d, but it might be the start of the
12484 * special UTF8f format, which usually looks something like
12485 * "%d%lu%4p" (the lu may vary by platform)
12487 assert((UTF8f)[0] == 'd');
12488 assert((UTF8f)[1] == '%');
12490 if ( args /* UTF8f only valid for C-ish sprintf */
12491 && q == fmtstart + 1 /* plain %d, not %....d */
12492 && patend >= fmtstart + sizeof(UTF8f) - 1 /* long enough */
12494 && strnEQ(q + 1, UTF8f + 2, sizeof(UTF8f) - 3))
12496 /* The argument has already gone through cBOOL, so the cast
12498 is_utf8 = (bool)va_arg(*args, int);
12499 elen = va_arg(*args, UV);
12500 /* if utf8 length is larger than 0x7ffff..., then it might
12501 * have been a signed value that wrapped */
12502 if (elen > ((~(STRLEN)0) >> 1)) {
12503 assert(0); /* in DEBUGGING build we want to crash */
12504 elen = 0; /* otherwise we want to treat this as an empty string */
12506 eptr = va_arg(*args, char *);
12507 q += sizeof(UTF8f) - 2;
12514 goto get_int_arg_val;
12525 goto get_int_arg_val;
12530 goto get_int_arg_val;
12541 goto get_int_arg_val;
12556 esignbuf[esignlen++] = plus;
12559 /* initialise the vector string to iterate over */
12561 vecsv = args ? va_arg(*args, SV*) : argsv;
12563 /* if this is a version object, we need to convert
12564 * back into v-string notation and then let the
12565 * vectorize happen normally
12567 if (sv_isobject(vecsv) && sv_derived_from(vecsv, "version")) {
12568 if ( hv_existss(MUTABLE_HV(SvRV(vecsv)), "alpha") ) {
12569 Perl_ck_warner_d(aTHX_ packWARN(WARN_PRINTF),
12570 "vector argument not supported with alpha versions");
12574 vecstr = (U8*)SvPV_const(vecsv,veclen);
12575 vecsv = sv_newmortal();
12576 scan_vstring((char *)vecstr, (char *)vecstr + veclen,
12580 vecstr = (U8*)SvPV_const(vecsv, veclen);
12581 vec_utf8 = DO_UTF8(vecsv);
12583 /* This is the re-entry point for when we're iterating
12584 * over the individual characters of a vector arg */
12587 goto done_valid_conversion;
12589 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
12599 /* test arg for inf/nan. This can trigger an unwanted
12600 * 'str' overload, so manually force 'num' overload first
12604 if (UNLIKELY(SvAMAGIC(argsv)))
12605 argsv = sv_2num(argsv);
12606 if (UNLIKELY(isinfnansv(argsv)))
12607 goto handle_infnan_argsv;
12611 /* signed int type */
12616 case 'c': iv = (char)va_arg(*args, int); break;
12617 case 'h': iv = (short)va_arg(*args, int); break;
12618 case 'l': iv = va_arg(*args, long); break;
12619 case 'V': iv = va_arg(*args, IV); break;
12620 case 'z': iv = va_arg(*args, SSize_t); break;
12621 #ifdef HAS_PTRDIFF_T
12622 case 't': iv = va_arg(*args, ptrdiff_t); break;
12624 default: iv = va_arg(*args, int); break;
12625 case 'j': iv = va_arg(*args, PERL_INTMAX_T); break;
12628 iv = va_arg(*args, Quad_t); break;
12635 /* assign to tiv then cast to iv to work around
12636 * 2003 GCC cast bug (gnu.org bugzilla #13488) */
12637 IV tiv = SvIV_nomg(argsv);
12639 case 'c': iv = (char)tiv; break;
12640 case 'h': iv = (short)tiv; break;
12641 case 'l': iv = (long)tiv; break;
12643 default: iv = tiv; break;
12646 iv = (Quad_t)tiv; break;
12653 /* now convert iv to uv */
12657 esignbuf[esignlen++] = plus;
12661 esignbuf[esignlen++] = '-';
12665 /* unsigned int type */
12668 case 'c': uv = (unsigned char)va_arg(*args, unsigned);
12670 case 'h': uv = (unsigned short)va_arg(*args, unsigned);
12672 case 'l': uv = va_arg(*args, unsigned long); break;
12673 case 'V': uv = va_arg(*args, UV); break;
12674 case 'z': uv = va_arg(*args, Size_t); break;
12675 #ifdef HAS_PTRDIFF_T
12676 /* will sign extend, but there is no
12677 * uptrdiff_t, so oh well */
12678 case 't': uv = va_arg(*args, ptrdiff_t); break;
12680 case 'j': uv = va_arg(*args, PERL_UINTMAX_T); break;
12681 default: uv = va_arg(*args, unsigned); break;
12684 uv = va_arg(*args, Uquad_t); break;
12691 /* assign to tiv then cast to iv to work around
12692 * 2003 GCC cast bug (gnu.org bugzilla #13488) */
12693 UV tuv = SvUV_nomg(argsv);
12695 case 'c': uv = (unsigned char)tuv; break;
12696 case 'h': uv = (unsigned short)tuv; break;
12697 case 'l': uv = (unsigned long)tuv; break;
12699 default: uv = tuv; break;
12702 uv = (Uquad_t)tuv; break;
12713 char *ptr = ebuf + sizeof ebuf;
12720 const char * const p =
12721 (c == 'X') ? PL_hexdigit + 16 : PL_hexdigit;
12726 } while (uv >>= 4);
12727 if (alt && *ptr != '0') {
12728 esignbuf[esignlen++] = '0';
12729 esignbuf[esignlen++] = c; /* 'x' or 'X' */
12736 *--ptr = '0' + dig;
12737 } while (uv >>= 3);
12738 if (alt && *ptr != '0')
12744 *--ptr = '0' + dig;
12745 } while (uv >>= 1);
12746 if (alt && *ptr != '0') {
12747 esignbuf[esignlen++] = '0';
12748 esignbuf[esignlen++] = c; /* 'b' or 'B' */
12753 /* special-case: base 1 indicates a 'c' format:
12754 * we use the common code for extracting a uv,
12755 * but handle that value differently here than
12756 * all the other int types */
12758 (!UVCHR_IS_INVARIANT(uv) && SvUTF8(sv)))
12761 assert(sizeof(ebuf) >= UTF8_MAXBYTES + 1);
12763 elen = uvchr_to_utf8((U8*)eptr, uv) - (U8*)ebuf;
12768 ebuf[0] = (char)uv;
12773 default: /* it had better be ten or less */
12776 *--ptr = '0' + dig;
12777 } while (uv /= base);
12780 elen = (ebuf + sizeof ebuf) - ptr;
12784 zeros = precis - elen;
12785 else if (precis == 0 && elen == 1 && *eptr == '0'
12786 && !(base == 8 && alt)) /* "%#.0o" prints "0" */
12789 /* a precision nullifies the 0 flag. */
12795 /* FLOATING POINT */
12798 c = 'f'; /* maybe %F isn't supported here */
12800 case 'e': case 'E':
12802 case 'g': case 'G':
12803 case 'a': case 'A':
12806 STRLEN float_need; /* what PL_efloatsize needs to become */
12807 bool hexfp; /* hexadecimal floating point? */
12809 vcatpvfn_long_double_t fv;
12812 /* This is evil, but floating point is even more evil */
12814 /* for SV-style calling, we can only get NV
12815 for C-style calling, we assume %f is double;
12816 for simplicity we allow any of %Lf, %llf, %qf for long double
12820 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12824 /* [perl #20339] - we should accept and ignore %lf rather than die */
12828 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12829 intsize = args ? 0 : 'q';
12833 #if defined(HAS_LONG_DOUBLE)
12846 /* Now we need (long double) if intsize == 'q', else (double). */
12848 /* Note: do not pull NVs off the va_list with va_arg()
12849 * (pull doubles instead) because if you have a build
12850 * with long doubles, you would always be pulling long
12851 * doubles, which would badly break anyone using only
12852 * doubles (i.e. the majority of builds). In other
12853 * words, you cannot mix doubles and long doubles.
12854 * The only case where you can pull off long doubles
12855 * is when the format specifier explicitly asks so with
12857 #ifdef USE_QUADMATH
12858 fv = intsize == 'q' ?
12859 va_arg(*args, NV) : va_arg(*args, double);
12861 #elif LONG_DOUBLESIZE > DOUBLESIZE
12862 if (intsize == 'q') {
12863 fv = va_arg(*args, long double);
12866 nv = va_arg(*args, double);
12867 VCATPVFN_NV_TO_FV(nv, fv);
12870 nv = va_arg(*args, double);
12877 /* we jump here if an int-ish format encountered an
12878 * infinite/Nan argsv. After setting nv/fv, it falls
12879 * into the isinfnan block which follows */
12880 handle_infnan_argsv:
12881 nv = SvNV_nomg(argsv);
12882 VCATPVFN_NV_TO_FV(nv, fv);
12885 if (Perl_isinfnan(nv)) {
12887 Perl_croak(aTHX_ "Cannot printf %" NVgf " with '%c'",
12888 SvNV_nomg(argsv), (int)c);
12890 elen = S_infnan_2pv(nv, ebuf, sizeof(ebuf), plus);
12899 /* special-case "%.0f" */
12903 && !(width || left || plus || alt)
12906 && ((eptr = F0convert(nv, ebuf + sizeof ebuf, &elen)))
12910 /* Determine the buffer size needed for the various
12911 * floating-point formats.
12913 * The basic possibilities are:
12916 * %f 1111111.123456789
12917 * %e 1.111111123e+06
12918 * %a 0x1.0f4471f9bp+20
12920 * %g 1.11111112e+15
12922 * where P is the value of the precision in the format, or 6
12923 * if not specified. Note the two possible output formats of
12924 * %g; in both cases the number of significant digits is <=
12927 * For most of the format types the maximum buffer size needed
12928 * is precision, plus: any leading 1 or 0x1, the radix
12929 * point, and an exponent. The difficult one is %f: for a
12930 * large positive exponent it can have many leading digits,
12931 * which needs to be calculated specially. Also %a is slightly
12932 * different in that in the absence of a specified precision,
12933 * it uses as many digits as necessary to distinguish
12934 * different values.
12936 * First, here are the constant bits. For ease of calculation
12937 * we over-estimate the needed buffer size, for example by
12938 * assuming all formats have an exponent and a leading 0x1.
12940 * Also for production use, add a little extra overhead for
12941 * safety's sake. Under debugging don't, as it means we're
12942 * more likely to quickly spot issues during development.
12945 float_need = 1 /* possible unary minus */
12946 + 4 /* "0x1" plus very unlikely carry */
12947 + 1 /* default radix point '.' */
12948 + 2 /* "e-", "p+" etc */
12949 + 6 /* exponent: up to 16383 (quad fp) */
12951 + 20 /* safety net */
12956 /* determine the radix point len, e.g. length(".") in "1.2" */
12957 #ifdef USE_LOCALE_NUMERIC
12958 /* note that we may either explicitly use PL_numeric_radix_sv
12959 * below, or implicitly, via an snprintf() variant.
12960 * Note also things like ps_AF.utf8 which has
12961 * "\N{ARABIC DECIMAL SEPARATOR} as a radix point */
12962 if (!lc_numeric_set) {
12963 /* only set once and reuse in-locale value on subsequent
12965 * XXX what happens if we die in an eval?
12967 STORE_LC_NUMERIC_SET_TO_NEEDED();
12968 lc_numeric_set = TRUE;
12971 if (IN_LC(LC_NUMERIC)) {
12972 /* this can't wrap unless PL_numeric_radix_sv is a string
12973 * consuming virtually all the 32-bit or 64-bit address
12976 float_need += (SvCUR(PL_numeric_radix_sv) - 1);
12978 /* floating-point formats only get utf8 if the radix point
12979 * is utf8. All other characters in the string are < 128
12980 * and so can be safely appended to both a non-utf8 and utf8
12982 * Note that this will convert the output to utf8 even if
12983 * the radix point didn't get output.
12985 if (SvUTF8(PL_numeric_radix_sv) && !has_utf8) {
12986 sv_utf8_upgrade(sv);
12994 if (isALPHA_FOLD_EQ(c, 'f')) {
12995 /* Determine how many digits before the radix point
12996 * might be emitted. frexp() (or frexpl) has some
12997 * unspecified behaviour for nan/inf/-inf, so lucky we've
12998 * already handled them above */
13000 int i = PERL_INT_MIN;
13001 (void)Perl_frexp((NV)fv, &i);
13002 if (i == PERL_INT_MIN)
13003 Perl_die(aTHX_ "panic: frexp: %" VCATPVFN_FV_GF, fv);
13006 digits = BIT_DIGITS(i);
13007 /* this can't overflow. 'digits' will only be a few
13008 * thousand even for the largest floating-point types.
13009 * And up until now float_need is just some small
13010 * constants plus radix len, which can't be in
13011 * overflow territory unless the radix SV is consuming
13012 * over 1/2 the address space */
13013 assert(float_need < ((STRLEN)~0) - digits);
13014 float_need += digits;
13017 else if (UNLIKELY(isALPHA_FOLD_EQ(c, 'a'))) {
13020 /* %a in the absence of precision may print as many
13021 * digits as needed to represent the entire mantissa
13023 * This estimate seriously overshoots in most cases,
13024 * but better the undershooting. Firstly, all bytes
13025 * of the NV are not mantissa, some of them are
13026 * exponent. Secondly, for the reasonably common
13027 * long doubles case, the "80-bit extended", two
13028 * or six bytes of the NV are unused. Also, we'll
13029 * still pick up an extra +6 from the default
13030 * precision calculation below. */
13032 #ifdef LONGDOUBLE_DOUBLEDOUBLE
13033 /* For the "double double", we need more.
13034 * Since each double has their own exponent, the
13035 * doubles may float (haha) rather far from each
13036 * other, and the number of required bits is much
13037 * larger, up to total of DOUBLEDOUBLE_MAXBITS bits.
13038 * See the definition of DOUBLEDOUBLE_MAXBITS.
13040 * Need 2 hexdigits for each byte. */
13041 (DOUBLEDOUBLE_MAXBITS/8 + 1) * 2;
13043 NVSIZE * 2; /* 2 hexdigits for each byte */
13045 /* see "this can't overflow" comment above */
13046 assert(float_need < ((STRLEN)~0) - digits);
13047 float_need += digits;
13050 /* special-case "%.<number>g" if it will fit in ebuf */
13052 && precis /* See earlier comment about buggy Gconvert
13053 when digits, aka precis, is 0 */
13055 /* check, in manner not involving wrapping, that it will
13057 && float_need < sizeof(ebuf)
13058 && sizeof(ebuf) - float_need > precis
13059 && !(width || left || plus || alt)
13063 SNPRINTF_G(fv, ebuf, sizeof(ebuf), precis);
13064 elen = strlen(ebuf);
13071 STRLEN pr = has_precis ? precis : 6; /* known default */
13072 /* this probably can't wrap, since precis is limited
13073 * to 1/4 address space size, but better safe than sorry
13075 if (float_need >= ((STRLEN)~0) - pr)
13076 croak_memory_wrap();
13080 if (float_need < width)
13081 float_need = width;
13083 if (PL_efloatsize <= float_need) {
13084 /* PL_efloatbuf should be at least 1 greater than
13085 * float_need to allow a trailing \0 to be returned by
13086 * snprintf(). If we need to grow, overgrow for the
13087 * benefit of future generations */
13088 const STRLEN extra = 0x20;
13089 if (float_need >= ((STRLEN)~0) - extra)
13090 croak_memory_wrap();
13091 float_need += extra;
13092 Safefree(PL_efloatbuf);
13093 PL_efloatsize = float_need;
13094 Newx(PL_efloatbuf, PL_efloatsize, char);
13095 PL_efloatbuf[0] = '\0';
13098 if (UNLIKELY(hexfp)) {
13099 elen = S_format_hexfp(aTHX_ PL_efloatbuf, PL_efloatsize, c,
13100 nv, fv, has_precis, precis, width,
13101 alt, plus, left, fill);
13104 char *ptr = ebuf + sizeof ebuf;
13107 #if defined(USE_QUADMATH)
13108 if (intsize == 'q') {
13112 /* FIXME: what to do if HAS_LONG_DOUBLE but not PERL_PRIfldbl? */
13113 #elif defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
13114 /* Note that this is HAS_LONG_DOUBLE and PERL_PRIfldbl,
13115 * not USE_LONG_DOUBLE and NVff. In other words,
13116 * this needs to work without USE_LONG_DOUBLE. */
13117 if (intsize == 'q') {
13118 /* Copy the one or more characters in a long double
13119 * format before the 'base' ([efgEFG]) character to
13120 * the format string. */
13121 static char const ldblf[] = PERL_PRIfldbl;
13122 char const *p = ldblf + sizeof(ldblf) - 3;
13123 while (p >= ldblf) { *--ptr = *p--; }
13128 do { *--ptr = '0' + (base % 10); } while (base /= 10);
13133 do { *--ptr = '0' + (base % 10); } while (base /= 10);
13145 /* No taint. Otherwise we are in the strange situation
13146 * where printf() taints but print($float) doesn't.
13149 /* hopefully the above makes ptr a very constrained format
13150 * that is safe to use, even though it's not literal */
13151 GCC_DIAG_IGNORE_STMT(-Wformat-nonliteral);
13152 #ifdef USE_QUADMATH
13154 const char* qfmt = quadmath_format_single(ptr);
13156 Perl_croak_nocontext("panic: quadmath invalid format \"%s\"", ptr);
13157 elen = quadmath_snprintf(PL_efloatbuf, PL_efloatsize,
13159 if ((IV)elen == -1) {
13162 Perl_croak_nocontext("panic: quadmath_snprintf failed, format \"%s\"", qfmt);
13167 #elif defined(HAS_LONG_DOUBLE)
13168 elen = ((intsize == 'q')
13169 ? my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv)
13170 : my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, (double)fv));
13172 elen = my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv);
13174 GCC_DIAG_RESTORE_STMT;
13177 eptr = PL_efloatbuf;
13181 /* Since floating-point formats do their own formatting and
13182 * padding, we skip the main block of code at the end of this
13183 * loop which handles appending eptr to sv, and do our own
13184 * stripped-down version */
13189 assert(elen >= width);
13191 S_sv_catpvn_simple(aTHX_ sv, eptr, elen);
13193 goto done_valid_conversion;
13201 /* XXX ideally we should warn if any flags etc have been
13202 * set, e.g. "%-4.5n" */
13203 /* XXX if sv was originally non-utf8 with a char in the
13204 * range 0x80-0xff, then if it got upgraded, we should
13205 * calculate char len rather than byte len here */
13206 len = SvCUR(sv) - origlen;
13208 int i = (len > PERL_INT_MAX) ? PERL_INT_MAX : (int)len;
13211 case 'c': *(va_arg(*args, char*)) = i; break;
13212 case 'h': *(va_arg(*args, short*)) = i; break;
13213 default: *(va_arg(*args, int*)) = i; break;
13214 case 'l': *(va_arg(*args, long*)) = i; break;
13215 case 'V': *(va_arg(*args, IV*)) = i; break;
13216 case 'z': *(va_arg(*args, SSize_t*)) = i; break;
13217 #ifdef HAS_PTRDIFF_T
13218 case 't': *(va_arg(*args, ptrdiff_t*)) = i; break;
13220 case 'j': *(va_arg(*args, PERL_INTMAX_T*)) = i; break;
13223 *(va_arg(*args, Quad_t*)) = i; break;
13231 Perl_croak_nocontext(
13232 "Missing argument for %%n in %s",
13233 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
13234 sv_setuv_mg(argsv, has_utf8 ? (UV)sv_len_utf8(sv) : (UV)len);
13236 goto done_valid_conversion;
13244 && (PL_op->op_type == OP_PRTF || PL_op->op_type == OP_SPRINTF)
13245 && ckWARN(WARN_PRINTF))
13247 SV * const msg = sv_newmortal();
13248 Perl_sv_setpvf(aTHX_ msg, "Invalid conversion in %sprintf: ",
13249 (PL_op->op_type == OP_PRTF) ? "" : "s");
13250 if (fmtstart < patend) {
13251 const char * const fmtend = q < patend ? q : patend;
13253 sv_catpvs(msg, "\"%");
13254 for (f = fmtstart; f < fmtend; f++) {
13256 sv_catpvn_nomg(msg, f, 1);
13258 Perl_sv_catpvf(aTHX_ msg,
13259 "\\%03" UVof, (UV)*f & 0xFF);
13262 sv_catpvs(msg, "\"");
13264 sv_catpvs(msg, "end of string");
13266 Perl_warner(aTHX_ packWARN(WARN_PRINTF), "%" SVf, SVfARG(msg)); /* yes, this is reentrant */
13269 /* mangled format: output the '%', then continue from the
13270 * character following that */
13271 sv_catpvn_nomg(sv, fmtstart-1, 1);
13274 /* Any "redundant arg" warning from now onwards will probably
13275 * just be misleading, so don't bother. */
13276 no_redundant_warning = TRUE;
13277 continue; /* not "break" */
13280 if (is_utf8 != has_utf8) {
13283 sv_utf8_upgrade(sv);
13286 const STRLEN old_elen = elen;
13287 SV * const nsv = newSVpvn_flags(eptr, elen, SVs_TEMP);
13288 sv_utf8_upgrade(nsv);
13289 eptr = SvPVX_const(nsv);
13292 if (width) { /* fudge width (can't fudge elen) */
13293 width += elen - old_elen;
13300 /* append esignbuf, filler, zeros, eptr and dotstr to sv */
13303 STRLEN need, have, gap;
13307 /* signed value that's wrapped? */
13308 assert(elen <= ((~(STRLEN)0) >> 1));
13310 /* if zeros is non-zero, then it represents filler between
13311 * elen and precis. So adding elen and zeros together will
13312 * always be <= precis, and the addition can never wrap */
13313 assert(!zeros || (precis > elen && precis - elen == zeros));
13314 have = elen + zeros;
13316 if (have >= (((STRLEN)~0) - esignlen))
13317 croak_memory_wrap();
13320 need = (have > width ? have : width);
13323 if (need >= (((STRLEN)~0) - (SvCUR(sv) + 1)))
13324 croak_memory_wrap();
13325 need += (SvCUR(sv) + 1);
13332 for (i = 0; i < esignlen; i++)
13333 *s++ = esignbuf[i];
13334 for (i = zeros; i; i--)
13336 Copy(eptr, s, elen, char);
13338 for (i = gap; i; i--)
13343 for (i = 0; i < esignlen; i++)
13344 *s++ = esignbuf[i];
13349 for (i = gap; i; i--)
13351 for (i = 0; i < esignlen; i++)
13352 *s++ = esignbuf[i];
13355 for (i = zeros; i; i--)
13357 Copy(eptr, s, elen, char);
13362 SvCUR_set(sv, s - SvPVX_const(sv));
13370 if (vectorize && veclen) {
13371 /* we append the vector separator separately since %v isn't
13372 * very common: don't slow down the general case by adding
13373 * dotstrlen to need etc */
13374 sv_catpvn_nomg(sv, dotstr, dotstrlen);
13376 goto vector; /* do next iteration */
13379 done_valid_conversion:
13382 S_warn_vcatpvfn_missing_argument(aTHX);
13385 /* Now that we've consumed all our printf format arguments (svix)
13386 * do we have things left on the stack that we didn't use?
13388 if (!no_redundant_warning && sv_count >= svix + 1 && ckWARN(WARN_REDUNDANT)) {
13389 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
13390 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
13395 #ifdef USE_LOCALE_NUMERIC
13397 if (lc_numeric_set) {
13398 RESTORE_LC_NUMERIC(); /* Done outside loop, so don't have to
13399 save/restore each iteration. */
13406 /* =========================================================================
13408 =head1 Cloning an interpreter
13412 All the macros and functions in this section are for the private use of
13413 the main function, perl_clone().
13415 The foo_dup() functions make an exact copy of an existing foo thingy.
13416 During the course of a cloning, a hash table is used to map old addresses
13417 to new addresses. The table is created and manipulated with the
13418 ptr_table_* functions.
13420 * =========================================================================*/
13423 #if defined(USE_ITHREADS)
13425 /* XXX Remove this so it doesn't have to go thru the macro and return for nothing */
13426 #ifndef GpREFCNT_inc
13427 # define GpREFCNT_inc(gp) ((gp) ? (++(gp)->gp_refcnt, (gp)) : (GP*)NULL)
13431 /* Certain cases in Perl_ss_dup have been merged, by relying on the fact
13432 that currently av_dup, gv_dup and hv_dup are the same as sv_dup.
13433 If this changes, please unmerge ss_dup.
13434 Likewise, sv_dup_inc_multiple() relies on this fact. */
13435 #define sv_dup_inc_NN(s,t) SvREFCNT_inc_NN(sv_dup_inc(s,t))
13436 #define av_dup(s,t) MUTABLE_AV(sv_dup((const SV *)s,t))
13437 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
13438 #define hv_dup(s,t) MUTABLE_HV(sv_dup((const SV *)s,t))
13439 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
13440 #define cv_dup(s,t) MUTABLE_CV(sv_dup((const SV *)s,t))
13441 #define cv_dup_inc(s,t) MUTABLE_CV(sv_dup_inc((const SV *)s,t))
13442 #define io_dup(s,t) MUTABLE_IO(sv_dup((const SV *)s,t))
13443 #define io_dup_inc(s,t) MUTABLE_IO(sv_dup_inc((const SV *)s,t))
13444 #define gv_dup(s,t) MUTABLE_GV(sv_dup((const SV *)s,t))
13445 #define gv_dup_inc(s,t) MUTABLE_GV(sv_dup_inc((const SV *)s,t))
13446 #define SAVEPV(p) ((p) ? savepv(p) : NULL)
13447 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
13449 /* clone a parser */
13452 Perl_parser_dup(pTHX_ const yy_parser *const proto, CLONE_PARAMS *const param)
13456 PERL_ARGS_ASSERT_PARSER_DUP;
13461 /* look for it in the table first */
13462 parser = (yy_parser *)ptr_table_fetch(PL_ptr_table, proto);
13466 /* create anew and remember what it is */
13467 Newxz(parser, 1, yy_parser);
13468 ptr_table_store(PL_ptr_table, proto, parser);
13470 /* XXX eventually, just Copy() most of the parser struct ? */
13472 parser->lex_brackets = proto->lex_brackets;
13473 parser->lex_casemods = proto->lex_casemods;
13474 parser->lex_brackstack = savepvn(proto->lex_brackstack,
13475 (proto->lex_brackets < 120 ? 120 : proto->lex_brackets));
13476 parser->lex_casestack = savepvn(proto->lex_casestack,
13477 (proto->lex_casemods < 12 ? 12 : proto->lex_casemods));
13478 parser->lex_defer = proto->lex_defer;
13479 parser->lex_dojoin = proto->lex_dojoin;
13480 parser->lex_formbrack = proto->lex_formbrack;
13481 parser->lex_inpat = proto->lex_inpat;
13482 parser->lex_inwhat = proto->lex_inwhat;
13483 parser->lex_op = proto->lex_op;
13484 parser->lex_repl = sv_dup_inc(proto->lex_repl, param);
13485 parser->lex_starts = proto->lex_starts;
13486 parser->lex_stuff = sv_dup_inc(proto->lex_stuff, param);
13487 parser->multi_close = proto->multi_close;
13488 parser->multi_open = proto->multi_open;
13489 parser->multi_start = proto->multi_start;
13490 parser->multi_end = proto->multi_end;
13491 parser->preambled = proto->preambled;
13492 parser->lex_super_state = proto->lex_super_state;
13493 parser->lex_sub_inwhat = proto->lex_sub_inwhat;
13494 parser->lex_sub_op = proto->lex_sub_op;
13495 parser->lex_sub_repl= sv_dup_inc(proto->lex_sub_repl, param);
13496 parser->linestr = sv_dup_inc(proto->linestr, param);
13497 parser->expect = proto->expect;
13498 parser->copline = proto->copline;
13499 parser->last_lop_op = proto->last_lop_op;
13500 parser->lex_state = proto->lex_state;
13501 parser->rsfp = fp_dup(proto->rsfp, '<', param);
13502 /* rsfp_filters entries have fake IoDIRP() */
13503 parser->rsfp_filters= av_dup_inc(proto->rsfp_filters, param);
13504 parser->in_my = proto->in_my;
13505 parser->in_my_stash = hv_dup(proto->in_my_stash, param);
13506 parser->error_count = proto->error_count;
13507 parser->sig_elems = proto->sig_elems;
13508 parser->sig_optelems= proto->sig_optelems;
13509 parser->sig_slurpy = proto->sig_slurpy;
13510 parser->recheck_utf8_validity = proto->recheck_utf8_validity;
13513 char * const ols = SvPVX(proto->linestr);
13514 char * const ls = SvPVX(parser->linestr);
13516 parser->bufptr = ls + (proto->bufptr >= ols ?
13517 proto->bufptr - ols : 0);
13518 parser->oldbufptr = ls + (proto->oldbufptr >= ols ?
13519 proto->oldbufptr - ols : 0);
13520 parser->oldoldbufptr= ls + (proto->oldoldbufptr >= ols ?
13521 proto->oldoldbufptr - ols : 0);
13522 parser->linestart = ls + (proto->linestart >= ols ?
13523 proto->linestart - ols : 0);
13524 parser->last_uni = ls + (proto->last_uni >= ols ?
13525 proto->last_uni - ols : 0);
13526 parser->last_lop = ls + (proto->last_lop >= ols ?
13527 proto->last_lop - ols : 0);
13529 parser->bufend = ls + SvCUR(parser->linestr);
13532 Copy(proto->tokenbuf, parser->tokenbuf, 256, char);
13535 Copy(proto->nextval, parser->nextval, 5, YYSTYPE);
13536 Copy(proto->nexttype, parser->nexttype, 5, I32);
13537 parser->nexttoke = proto->nexttoke;
13539 /* XXX should clone saved_curcop here, but we aren't passed
13540 * proto_perl; so do it in perl_clone_using instead */
13546 /* duplicate a file handle */
13549 Perl_fp_dup(pTHX_ PerlIO *const fp, const char type, CLONE_PARAMS *const param)
13553 PERL_ARGS_ASSERT_FP_DUP;
13554 PERL_UNUSED_ARG(type);
13557 return (PerlIO*)NULL;
13559 /* look for it in the table first */
13560 ret = (PerlIO*)ptr_table_fetch(PL_ptr_table, fp);
13564 /* create anew and remember what it is */
13565 #ifdef __amigaos4__
13566 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE|PERLIO_DUP_FD);
13568 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE);
13570 ptr_table_store(PL_ptr_table, fp, ret);
13574 /* duplicate a directory handle */
13577 Perl_dirp_dup(pTHX_ DIR *const dp, CLONE_PARAMS *const param)
13581 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
13583 const Direntry_t *dirent;
13584 char smallbuf[256]; /* XXX MAXPATHLEN, surely? */
13590 PERL_UNUSED_CONTEXT;
13591 PERL_ARGS_ASSERT_DIRP_DUP;
13596 /* look for it in the table first */
13597 ret = (DIR*)ptr_table_fetch(PL_ptr_table, dp);
13601 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
13603 PERL_UNUSED_ARG(param);
13607 /* open the current directory (so we can switch back) */
13608 if (!(pwd = PerlDir_open("."))) return (DIR *)NULL;
13610 /* chdir to our dir handle and open the present working directory */
13611 if (fchdir(my_dirfd(dp)) < 0 || !(ret = PerlDir_open("."))) {
13612 PerlDir_close(pwd);
13613 return (DIR *)NULL;
13615 /* Now we should have two dir handles pointing to the same dir. */
13617 /* Be nice to the calling code and chdir back to where we were. */
13618 /* XXX If this fails, then what? */
13619 PERL_UNUSED_RESULT(fchdir(my_dirfd(pwd)));
13621 /* We have no need of the pwd handle any more. */
13622 PerlDir_close(pwd);
13625 # define d_namlen(d) (d)->d_namlen
13627 # define d_namlen(d) strlen((d)->d_name)
13629 /* Iterate once through dp, to get the file name at the current posi-
13630 tion. Then step back. */
13631 pos = PerlDir_tell(dp);
13632 if ((dirent = PerlDir_read(dp))) {
13633 len = d_namlen(dirent);
13634 if (len > sizeof(dirent->d_name) && sizeof(dirent->d_name) > PTRSIZE) {
13635 /* If the len is somehow magically longer than the
13636 * maximum length of the directory entry, even though
13637 * we could fit it in a buffer, we could not copy it
13638 * from the dirent. Bail out. */
13639 PerlDir_close(ret);
13642 if (len <= sizeof smallbuf) name = smallbuf;
13643 else Newx(name, len, char);
13644 Move(dirent->d_name, name, len, char);
13646 PerlDir_seek(dp, pos);
13648 /* Iterate through the new dir handle, till we find a file with the
13650 if (!dirent) /* just before the end */
13652 pos = PerlDir_tell(ret);
13653 if (PerlDir_read(ret)) continue; /* not there yet */
13654 PerlDir_seek(ret, pos); /* step back */
13658 const long pos0 = PerlDir_tell(ret);
13660 pos = PerlDir_tell(ret);
13661 if ((dirent = PerlDir_read(ret))) {
13662 if (len == (STRLEN)d_namlen(dirent)
13663 && memEQ(name, dirent->d_name, len)) {
13665 PerlDir_seek(ret, pos); /* step back */
13668 /* else we are not there yet; keep iterating */
13670 else { /* This is not meant to happen. The best we can do is
13671 reset the iterator to the beginning. */
13672 PerlDir_seek(ret, pos0);
13679 if (name && name != smallbuf)
13684 ret = win32_dirp_dup(dp, param);
13687 /* pop it in the pointer table */
13689 ptr_table_store(PL_ptr_table, dp, ret);
13694 /* duplicate a typeglob */
13697 Perl_gp_dup(pTHX_ GP *const gp, CLONE_PARAMS *const param)
13701 PERL_ARGS_ASSERT_GP_DUP;
13705 /* look for it in the table first */
13706 ret = (GP*)ptr_table_fetch(PL_ptr_table, gp);
13710 /* create anew and remember what it is */
13712 ptr_table_store(PL_ptr_table, gp, ret);
13715 /* ret->gp_refcnt must be 0 before any other dups are called. We're relying
13716 on Newxz() to do this for us. */
13717 ret->gp_sv = sv_dup_inc(gp->gp_sv, param);
13718 ret->gp_io = io_dup_inc(gp->gp_io, param);
13719 ret->gp_form = cv_dup_inc(gp->gp_form, param);
13720 ret->gp_av = av_dup_inc(gp->gp_av, param);
13721 ret->gp_hv = hv_dup_inc(gp->gp_hv, param);
13722 ret->gp_egv = gv_dup(gp->gp_egv, param);/* GvEGV is not refcounted */
13723 ret->gp_cv = cv_dup_inc(gp->gp_cv, param);
13724 ret->gp_cvgen = gp->gp_cvgen;
13725 ret->gp_line = gp->gp_line;
13726 ret->gp_file_hek = hek_dup(gp->gp_file_hek, param);
13730 /* duplicate a chain of magic */
13733 Perl_mg_dup(pTHX_ MAGIC *mg, CLONE_PARAMS *const param)
13735 MAGIC *mgret = NULL;
13736 MAGIC **mgprev_p = &mgret;
13738 PERL_ARGS_ASSERT_MG_DUP;
13740 for (; mg; mg = mg->mg_moremagic) {
13743 if ((param->flags & CLONEf_JOIN_IN)
13744 && mg->mg_type == PERL_MAGIC_backref)
13745 /* when joining, we let the individual SVs add themselves to
13746 * backref as needed. */
13749 Newx(nmg, 1, MAGIC);
13751 mgprev_p = &(nmg->mg_moremagic);
13753 /* There was a comment "XXX copy dynamic vtable?" but as we don't have
13754 dynamic vtables, I'm not sure why Sarathy wrote it. The comment dates
13755 from the original commit adding Perl_mg_dup() - revision 4538.
13756 Similarly there is the annotation "XXX random ptr?" next to the
13757 assignment to nmg->mg_ptr. */
13760 /* FIXME for plugins
13761 if (nmg->mg_type == PERL_MAGIC_qr) {
13762 nmg->mg_obj = MUTABLE_SV(CALLREGDUPE((REGEXP*)nmg->mg_obj, param));
13766 nmg->mg_obj = (nmg->mg_flags & MGf_REFCOUNTED)
13767 ? nmg->mg_type == PERL_MAGIC_backref
13768 /* The backref AV has its reference
13769 * count deliberately bumped by 1 */
13770 ? SvREFCNT_inc(av_dup_inc((const AV *)
13771 nmg->mg_obj, param))
13772 : sv_dup_inc(nmg->mg_obj, param)
13773 : (nmg->mg_type == PERL_MAGIC_regdatum ||
13774 nmg->mg_type == PERL_MAGIC_regdata)
13776 : sv_dup(nmg->mg_obj, param);
13778 if (nmg->mg_ptr && nmg->mg_type != PERL_MAGIC_regex_global) {
13779 if (nmg->mg_len > 0) {
13780 nmg->mg_ptr = SAVEPVN(nmg->mg_ptr, nmg->mg_len);
13781 if (nmg->mg_type == PERL_MAGIC_overload_table &&
13782 AMT_AMAGIC((AMT*)nmg->mg_ptr))
13784 AMT * const namtp = (AMT*)nmg->mg_ptr;
13785 sv_dup_inc_multiple((SV**)(namtp->table),
13786 (SV**)(namtp->table), NofAMmeth, param);
13789 else if (nmg->mg_len == HEf_SVKEY)
13790 nmg->mg_ptr = (char*)sv_dup_inc((const SV *)nmg->mg_ptr, param);
13792 if ((nmg->mg_flags & MGf_DUP) && nmg->mg_virtual && nmg->mg_virtual->svt_dup) {
13793 nmg->mg_virtual->svt_dup(aTHX_ nmg, param);
13799 #endif /* USE_ITHREADS */
13801 struct ptr_tbl_arena {
13802 struct ptr_tbl_arena *next;
13803 struct ptr_tbl_ent array[1023/3]; /* as ptr_tbl_ent has 3 pointers. */
13806 /* create a new pointer-mapping table */
13809 Perl_ptr_table_new(pTHX)
13812 PERL_UNUSED_CONTEXT;
13814 Newx(tbl, 1, PTR_TBL_t);
13815 tbl->tbl_max = 511;
13816 tbl->tbl_items = 0;
13817 tbl->tbl_arena = NULL;
13818 tbl->tbl_arena_next = NULL;
13819 tbl->tbl_arena_end = NULL;
13820 Newxz(tbl->tbl_ary, tbl->tbl_max + 1, PTR_TBL_ENT_t*);
13824 #define PTR_TABLE_HASH(ptr) \
13825 ((PTR2UV(ptr) >> 3) ^ (PTR2UV(ptr) >> (3 + 7)) ^ (PTR2UV(ptr) >> (3 + 17)))
13827 /* map an existing pointer using a table */
13829 STATIC PTR_TBL_ENT_t *
13830 S_ptr_table_find(PTR_TBL_t *const tbl, const void *const sv)
13832 PTR_TBL_ENT_t *tblent;
13833 const UV hash = PTR_TABLE_HASH(sv);
13835 PERL_ARGS_ASSERT_PTR_TABLE_FIND;
13837 tblent = tbl->tbl_ary[hash & tbl->tbl_max];
13838 for (; tblent; tblent = tblent->next) {
13839 if (tblent->oldval == sv)
13846 Perl_ptr_table_fetch(pTHX_ PTR_TBL_t *const tbl, const void *const sv)
13848 PTR_TBL_ENT_t const *const tblent = ptr_table_find(tbl, sv);
13850 PERL_ARGS_ASSERT_PTR_TABLE_FETCH;
13851 PERL_UNUSED_CONTEXT;
13853 return tblent ? tblent->newval : NULL;
13856 /* add a new entry to a pointer-mapping table 'tbl'. In hash terms, 'oldsv' is
13857 * the key; 'newsv' is the value. The names "old" and "new" are specific to
13858 * the core's typical use of ptr_tables in thread cloning. */
13861 Perl_ptr_table_store(pTHX_ PTR_TBL_t *const tbl, const void *const oldsv, void *const newsv)
13863 PTR_TBL_ENT_t *tblent = ptr_table_find(tbl, oldsv);
13865 PERL_ARGS_ASSERT_PTR_TABLE_STORE;
13866 PERL_UNUSED_CONTEXT;
13869 tblent->newval = newsv;
13871 const UV entry = PTR_TABLE_HASH(oldsv) & tbl->tbl_max;
13873 if (tbl->tbl_arena_next == tbl->tbl_arena_end) {
13874 struct ptr_tbl_arena *new_arena;
13876 Newx(new_arena, 1, struct ptr_tbl_arena);
13877 new_arena->next = tbl->tbl_arena;
13878 tbl->tbl_arena = new_arena;
13879 tbl->tbl_arena_next = new_arena->array;
13880 tbl->tbl_arena_end = C_ARRAY_END(new_arena->array);
13883 tblent = tbl->tbl_arena_next++;
13885 tblent->oldval = oldsv;
13886 tblent->newval = newsv;
13887 tblent->next = tbl->tbl_ary[entry];
13888 tbl->tbl_ary[entry] = tblent;
13890 if (tblent->next && tbl->tbl_items > tbl->tbl_max)
13891 ptr_table_split(tbl);
13895 /* double the hash bucket size of an existing ptr table */
13898 Perl_ptr_table_split(pTHX_ PTR_TBL_t *const tbl)
13900 PTR_TBL_ENT_t **ary = tbl->tbl_ary;
13901 const UV oldsize = tbl->tbl_max + 1;
13902 UV newsize = oldsize * 2;
13905 PERL_ARGS_ASSERT_PTR_TABLE_SPLIT;
13906 PERL_UNUSED_CONTEXT;
13908 Renew(ary, newsize, PTR_TBL_ENT_t*);
13909 Zero(&ary[oldsize], newsize-oldsize, PTR_TBL_ENT_t*);
13910 tbl->tbl_max = --newsize;
13911 tbl->tbl_ary = ary;
13912 for (i=0; i < oldsize; i++, ary++) {
13913 PTR_TBL_ENT_t **entp = ary;
13914 PTR_TBL_ENT_t *ent = *ary;
13915 PTR_TBL_ENT_t **curentp;
13918 curentp = ary + oldsize;
13920 if ((newsize & PTR_TABLE_HASH(ent->oldval)) != i) {
13922 ent->next = *curentp;
13932 /* remove all the entries from a ptr table */
13933 /* Deprecated - will be removed post 5.14 */
13936 Perl_ptr_table_clear(pTHX_ PTR_TBL_t *const tbl)
13938 PERL_UNUSED_CONTEXT;
13939 if (tbl && tbl->tbl_items) {
13940 struct ptr_tbl_arena *arena = tbl->tbl_arena;
13942 Zero(tbl->tbl_ary, tbl->tbl_max + 1, struct ptr_tbl_ent *);
13945 struct ptr_tbl_arena *next = arena->next;
13951 tbl->tbl_items = 0;
13952 tbl->tbl_arena = NULL;
13953 tbl->tbl_arena_next = NULL;
13954 tbl->tbl_arena_end = NULL;
13958 /* clear and free a ptr table */
13961 Perl_ptr_table_free(pTHX_ PTR_TBL_t *const tbl)
13963 struct ptr_tbl_arena *arena;
13965 PERL_UNUSED_CONTEXT;
13971 arena = tbl->tbl_arena;
13974 struct ptr_tbl_arena *next = arena->next;
13980 Safefree(tbl->tbl_ary);
13984 #if defined(USE_ITHREADS)
13987 Perl_rvpv_dup(pTHX_ SV *const dstr, const SV *const sstr, CLONE_PARAMS *const param)
13989 PERL_ARGS_ASSERT_RVPV_DUP;
13991 assert(!isREGEXP(sstr));
13993 if (SvWEAKREF(sstr)) {
13994 SvRV_set(dstr, sv_dup(SvRV_const(sstr), param));
13995 if (param->flags & CLONEf_JOIN_IN) {
13996 /* if joining, we add any back references individually rather
13997 * than copying the whole backref array */
13998 Perl_sv_add_backref(aTHX_ SvRV(dstr), dstr);
14002 SvRV_set(dstr, sv_dup_inc(SvRV_const(sstr), param));
14004 else if (SvPVX_const(sstr)) {
14005 /* Has something there */
14007 /* Normal PV - clone whole allocated space */
14008 SvPV_set(dstr, SAVEPVN(SvPVX_const(sstr), SvLEN(sstr)-1));
14009 /* sstr may not be that normal, but actually copy on write.
14010 But we are a true, independent SV, so: */
14014 /* Special case - not normally malloced for some reason */
14015 if (isGV_with_GP(sstr)) {
14016 /* Don't need to do anything here. */
14018 else if ((SvIsCOW(sstr))) {
14019 /* A "shared" PV - clone it as "shared" PV */
14021 HEK_KEY(hek_dup(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)),
14025 /* Some other special case - random pointer */
14026 SvPV_set(dstr, (char *) SvPVX_const(sstr));
14031 /* Copy the NULL */
14032 SvPV_set(dstr, NULL);
14036 /* duplicate a list of SVs. source and dest may point to the same memory. */
14038 S_sv_dup_inc_multiple(pTHX_ SV *const *source, SV **dest,
14039 SSize_t items, CLONE_PARAMS *const param)
14041 PERL_ARGS_ASSERT_SV_DUP_INC_MULTIPLE;
14043 while (items-- > 0) {
14044 *dest++ = sv_dup_inc(*source++, param);
14050 /* duplicate an SV of any type (including AV, HV etc) */
14053 S_sv_dup_common(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14058 PERL_ARGS_ASSERT_SV_DUP_COMMON;
14060 if (SvTYPE(sstr) == (svtype)SVTYPEMASK) {
14061 #ifdef DEBUG_LEAKING_SCALARS_ABORT
14066 /* look for it in the table first */
14067 dstr = MUTABLE_SV(ptr_table_fetch(PL_ptr_table, sstr));
14071 if(param->flags & CLONEf_JOIN_IN) {
14072 /** We are joining here so we don't want do clone
14073 something that is bad **/
14074 if (SvTYPE(sstr) == SVt_PVHV) {
14075 const HEK * const hvname = HvNAME_HEK(sstr);
14077 /** don't clone stashes if they already exist **/
14078 dstr = MUTABLE_SV(gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
14079 HEK_UTF8(hvname) ? SVf_UTF8 : 0));
14080 ptr_table_store(PL_ptr_table, sstr, dstr);
14084 else if (SvTYPE(sstr) == SVt_PVGV && !SvFAKE(sstr)) {
14085 HV *stash = GvSTASH(sstr);
14086 const HEK * hvname;
14087 if (stash && (hvname = HvNAME_HEK(stash))) {
14088 /** don't clone GVs if they already exist **/
14090 stash = gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
14091 HEK_UTF8(hvname) ? SVf_UTF8 : 0);
14093 stash, GvNAME(sstr),
14099 if (svp && *svp && SvTYPE(*svp) == SVt_PVGV) {
14100 ptr_table_store(PL_ptr_table, sstr, *svp);
14107 /* create anew and remember what it is */
14110 #ifdef DEBUG_LEAKING_SCALARS
14111 dstr->sv_debug_optype = sstr->sv_debug_optype;
14112 dstr->sv_debug_line = sstr->sv_debug_line;
14113 dstr->sv_debug_inpad = sstr->sv_debug_inpad;
14114 dstr->sv_debug_parent = (SV*)sstr;
14115 FREE_SV_DEBUG_FILE(dstr);
14116 dstr->sv_debug_file = savesharedpv(sstr->sv_debug_file);
14119 ptr_table_store(PL_ptr_table, sstr, dstr);
14122 SvFLAGS(dstr) = SvFLAGS(sstr);
14123 SvFLAGS(dstr) &= ~SVf_OOK; /* don't propagate OOK hack */
14124 SvREFCNT(dstr) = 0; /* must be before any other dups! */
14127 if (SvANY(sstr) && PL_watch_pvx && SvPVX_const(sstr) == PL_watch_pvx)
14128 PerlIO_printf(Perl_debug_log, "watch at %p hit, found string \"%s\"\n",
14129 (void*)PL_watch_pvx, SvPVX_const(sstr));
14132 /* don't clone objects whose class has asked us not to */
14134 && ! (SvFLAGS(SvSTASH(sstr)) & SVphv_CLONEABLE))
14140 switch (SvTYPE(sstr)) {
14142 SvANY(dstr) = NULL;
14145 SET_SVANY_FOR_BODYLESS_IV(dstr);
14147 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
14149 SvIV_set(dstr, SvIVX(sstr));
14153 #if NVSIZE <= IVSIZE
14154 SET_SVANY_FOR_BODYLESS_NV(dstr);
14156 SvANY(dstr) = new_XNV();
14158 SvNV_set(dstr, SvNVX(sstr));
14162 /* These are all the types that need complex bodies allocating. */
14164 const svtype sv_type = SvTYPE(sstr);
14165 const struct body_details *const sv_type_details
14166 = bodies_by_type + sv_type;
14170 Perl_croak(aTHX_ "Bizarre SvTYPE [%" IVdf "]", (IV)SvTYPE(sstr));
14171 NOT_REACHED; /* NOTREACHED */
14187 assert(sv_type_details->body_size);
14188 if (sv_type_details->arena) {
14189 new_body_inline(new_body, sv_type);
14191 = (void*)((char*)new_body - sv_type_details->offset);
14193 new_body = new_NOARENA(sv_type_details);
14197 SvANY(dstr) = new_body;
14200 Copy(((char*)SvANY(sstr)) + sv_type_details->offset,
14201 ((char*)SvANY(dstr)) + sv_type_details->offset,
14202 sv_type_details->copy, char);
14204 Copy(((char*)SvANY(sstr)),
14205 ((char*)SvANY(dstr)),
14206 sv_type_details->body_size + sv_type_details->offset, char);
14209 if (sv_type != SVt_PVAV && sv_type != SVt_PVHV
14210 && !isGV_with_GP(dstr)
14212 && !(sv_type == SVt_PVIO && !(IoFLAGS(dstr) & IOf_FAKE_DIRP)))
14213 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
14215 /* The Copy above means that all the source (unduplicated) pointers
14216 are now in the destination. We can check the flags and the
14217 pointers in either, but it's possible that there's less cache
14218 missing by always going for the destination.
14219 FIXME - instrument and check that assumption */
14220 if (sv_type >= SVt_PVMG) {
14222 SvMAGIC_set(dstr, mg_dup(SvMAGIC(dstr), param));
14223 if (SvOBJECT(dstr) && SvSTASH(dstr))
14224 SvSTASH_set(dstr, hv_dup_inc(SvSTASH(dstr), param));
14225 else SvSTASH_set(dstr, 0); /* don't copy DESTROY cache */
14228 /* The cast silences a GCC warning about unhandled types. */
14229 switch ((int)sv_type) {
14240 /* FIXME for plugins */
14241 re_dup_guts((REGEXP*) sstr, (REGEXP*) dstr, param);
14244 /* XXX LvTARGOFF sometimes holds PMOP* when DEBUGGING */
14245 if (LvTYPE(dstr) == 't') /* for tie: unrefcnted fake (SV**) */
14246 LvTARG(dstr) = dstr;
14247 else if (LvTYPE(dstr) == 'T') /* for tie: fake HE */
14248 LvTARG(dstr) = MUTABLE_SV(he_dup((HE*)LvTARG(dstr), 0, param));
14250 LvTARG(dstr) = sv_dup_inc(LvTARG(dstr), param);
14251 if (isREGEXP(sstr)) goto duprex;
14254 /* non-GP case already handled above */
14255 if(isGV_with_GP(sstr)) {
14256 GvNAME_HEK(dstr) = hek_dup(GvNAME_HEK(dstr), param);
14257 /* Don't call sv_add_backref here as it's going to be
14258 created as part of the magic cloning of the symbol
14259 table--unless this is during a join and the stash
14260 is not actually being cloned. */
14261 /* Danger Will Robinson - GvGP(dstr) isn't initialised
14262 at the point of this comment. */
14263 GvSTASH(dstr) = hv_dup(GvSTASH(dstr), param);
14264 if (param->flags & CLONEf_JOIN_IN)
14265 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
14266 GvGP_set(dstr, gp_dup(GvGP(sstr), param));
14267 (void)GpREFCNT_inc(GvGP(dstr));
14271 /* PL_parser->rsfp_filters entries have fake IoDIRP() */
14272 if(IoFLAGS(dstr) & IOf_FAKE_DIRP) {
14273 /* I have no idea why fake dirp (rsfps)
14274 should be treated differently but otherwise
14275 we end up with leaks -- sky*/
14276 IoTOP_GV(dstr) = gv_dup_inc(IoTOP_GV(dstr), param);
14277 IoFMT_GV(dstr) = gv_dup_inc(IoFMT_GV(dstr), param);
14278 IoBOTTOM_GV(dstr) = gv_dup_inc(IoBOTTOM_GV(dstr), param);
14280 IoTOP_GV(dstr) = gv_dup(IoTOP_GV(dstr), param);
14281 IoFMT_GV(dstr) = gv_dup(IoFMT_GV(dstr), param);
14282 IoBOTTOM_GV(dstr) = gv_dup(IoBOTTOM_GV(dstr), param);
14283 if (IoDIRP(dstr)) {
14284 IoDIRP(dstr) = dirp_dup(IoDIRP(dstr), param);
14287 /* IoDIRP(dstr) is already a copy of IoDIRP(sstr) */
14289 IoIFP(dstr) = fp_dup(IoIFP(sstr), IoTYPE(dstr), param);
14291 if (IoOFP(dstr) == IoIFP(sstr))
14292 IoOFP(dstr) = IoIFP(dstr);
14294 IoOFP(dstr) = fp_dup(IoOFP(dstr), IoTYPE(dstr), param);
14295 IoTOP_NAME(dstr) = SAVEPV(IoTOP_NAME(dstr));
14296 IoFMT_NAME(dstr) = SAVEPV(IoFMT_NAME(dstr));
14297 IoBOTTOM_NAME(dstr) = SAVEPV(IoBOTTOM_NAME(dstr));
14300 /* avoid cloning an empty array */
14301 if (AvARRAY((const AV *)sstr) && AvFILLp((const AV *)sstr) >= 0) {
14302 SV **dst_ary, **src_ary;
14303 SSize_t items = AvFILLp((const AV *)sstr) + 1;
14305 src_ary = AvARRAY((const AV *)sstr);
14306 Newx(dst_ary, AvMAX((const AV *)sstr)+1, SV*);
14307 ptr_table_store(PL_ptr_table, src_ary, dst_ary);
14308 AvARRAY(MUTABLE_AV(dstr)) = dst_ary;
14309 AvALLOC((const AV *)dstr) = dst_ary;
14310 if (AvREAL((const AV *)sstr)) {
14311 dst_ary = sv_dup_inc_multiple(src_ary, dst_ary, items,
14315 while (items-- > 0)
14316 *dst_ary++ = sv_dup(*src_ary++, param);
14318 items = AvMAX((const AV *)sstr) - AvFILLp((const AV *)sstr);
14319 while (items-- > 0) {
14324 AvARRAY(MUTABLE_AV(dstr)) = NULL;
14325 AvALLOC((const AV *)dstr) = (SV**)NULL;
14326 AvMAX( (const AV *)dstr) = -1;
14327 AvFILLp((const AV *)dstr) = -1;
14331 if (HvARRAY((const HV *)sstr)) {
14333 const bool sharekeys = !!HvSHAREKEYS(sstr);
14334 XPVHV * const dxhv = (XPVHV*)SvANY(dstr);
14335 XPVHV * const sxhv = (XPVHV*)SvANY(sstr);
14337 Newx(darray, PERL_HV_ARRAY_ALLOC_BYTES(dxhv->xhv_max+1)
14338 + (SvOOK(sstr) ? sizeof(struct xpvhv_aux) : 0),
14340 HvARRAY(dstr) = (HE**)darray;
14341 while (i <= sxhv->xhv_max) {
14342 const HE * const source = HvARRAY(sstr)[i];
14343 HvARRAY(dstr)[i] = source
14344 ? he_dup(source, sharekeys, param) : 0;
14348 const struct xpvhv_aux * const saux = HvAUX(sstr);
14349 struct xpvhv_aux * const daux = HvAUX(dstr);
14350 /* This flag isn't copied. */
14353 if (saux->xhv_name_count) {
14354 HEK ** const sname = saux->xhv_name_u.xhvnameu_names;
14356 = saux->xhv_name_count < 0
14357 ? -saux->xhv_name_count
14358 : saux->xhv_name_count;
14359 HEK **shekp = sname + count;
14361 Newx(daux->xhv_name_u.xhvnameu_names, count, HEK *);
14362 dhekp = daux->xhv_name_u.xhvnameu_names + count;
14363 while (shekp-- > sname) {
14365 *dhekp = hek_dup(*shekp, param);
14369 daux->xhv_name_u.xhvnameu_name
14370 = hek_dup(saux->xhv_name_u.xhvnameu_name,
14373 daux->xhv_name_count = saux->xhv_name_count;
14375 daux->xhv_aux_flags = saux->xhv_aux_flags;
14376 #ifdef PERL_HASH_RANDOMIZE_KEYS
14377 daux->xhv_rand = saux->xhv_rand;
14378 daux->xhv_last_rand = saux->xhv_last_rand;
14380 daux->xhv_riter = saux->xhv_riter;
14381 daux->xhv_eiter = saux->xhv_eiter
14382 ? he_dup(saux->xhv_eiter,
14383 cBOOL(HvSHAREKEYS(sstr)), param) : 0;
14384 /* backref array needs refcnt=2; see sv_add_backref */
14385 daux->xhv_backreferences =
14386 (param->flags & CLONEf_JOIN_IN)
14387 /* when joining, we let the individual GVs and
14388 * CVs add themselves to backref as
14389 * needed. This avoids pulling in stuff
14390 * that isn't required, and simplifies the
14391 * case where stashes aren't cloned back
14392 * if they already exist in the parent
14395 : saux->xhv_backreferences
14396 ? (SvTYPE(saux->xhv_backreferences) == SVt_PVAV)
14397 ? MUTABLE_AV(SvREFCNT_inc(
14398 sv_dup_inc((const SV *)
14399 saux->xhv_backreferences, param)))
14400 : MUTABLE_AV(sv_dup((const SV *)
14401 saux->xhv_backreferences, param))
14404 daux->xhv_mro_meta = saux->xhv_mro_meta
14405 ? mro_meta_dup(saux->xhv_mro_meta, param)
14408 /* Record stashes for possible cloning in Perl_clone(). */
14410 av_push(param->stashes, dstr);
14414 HvARRAY(MUTABLE_HV(dstr)) = NULL;
14417 if (!(param->flags & CLONEf_COPY_STACKS)) {
14422 /* NOTE: not refcounted */
14423 SvANY(MUTABLE_CV(dstr))->xcv_stash =
14424 hv_dup(CvSTASH(dstr), param);
14425 if ((param->flags & CLONEf_JOIN_IN) && CvSTASH(dstr))
14426 Perl_sv_add_backref(aTHX_ MUTABLE_SV(CvSTASH(dstr)), dstr);
14427 if (!CvISXSUB(dstr)) {
14429 CvROOT(dstr) = OpREFCNT_inc(CvROOT(dstr));
14431 CvSLABBED_off(dstr);
14432 } else if (CvCONST(dstr)) {
14433 CvXSUBANY(dstr).any_ptr =
14434 sv_dup_inc((const SV *)CvXSUBANY(dstr).any_ptr, param);
14436 assert(!CvSLABBED(dstr));
14437 if (CvDYNFILE(dstr)) CvFILE(dstr) = SAVEPV(CvFILE(dstr));
14439 SvANY((CV *)dstr)->xcv_gv_u.xcv_hek =
14440 hek_dup(CvNAME_HEK((CV *)sstr), param);
14441 /* don't dup if copying back - CvGV isn't refcounted, so the
14442 * duped GV may never be freed. A bit of a hack! DAPM */
14444 SvANY(MUTABLE_CV(dstr))->xcv_gv_u.xcv_gv =
14446 ? gv_dup_inc(CvGV(sstr), param)
14447 : (param->flags & CLONEf_JOIN_IN)
14449 : gv_dup(CvGV(sstr), param);
14451 if (!CvISXSUB(sstr)) {
14452 PADLIST * padlist = CvPADLIST(sstr);
14454 padlist = padlist_dup(padlist, param);
14455 CvPADLIST_set(dstr, padlist);
14457 /* unthreaded perl can't sv_dup so we dont support unthreaded's CvHSCXT */
14458 PoisonPADLIST(dstr);
14461 CvWEAKOUTSIDE(sstr)
14462 ? cv_dup( CvOUTSIDE(dstr), param)
14463 : cv_dup_inc(CvOUTSIDE(dstr), param);
14473 Perl_sv_dup_inc(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14475 PERL_ARGS_ASSERT_SV_DUP_INC;
14476 return sstr ? SvREFCNT_inc(sv_dup_common(sstr, param)) : NULL;
14480 Perl_sv_dup(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14482 SV *dstr = sstr ? sv_dup_common(sstr, param) : NULL;
14483 PERL_ARGS_ASSERT_SV_DUP;
14485 /* Track every SV that (at least initially) had a reference count of 0.
14486 We need to do this by holding an actual reference to it in this array.
14487 If we attempt to cheat, turn AvREAL_off(), and store only pointers
14488 (akin to the stashes hash, and the perl stack), we come unstuck if
14489 a weak reference (or other SV legitimately SvREFCNT() == 0 for this
14490 thread) is manipulated in a CLONE method, because CLONE runs before the
14491 unreferenced array is walked to find SVs still with SvREFCNT() == 0
14492 (and fix things up by giving each a reference via the temps stack).
14493 Instead, during CLONE, if the 0-referenced SV has SvREFCNT_inc() and
14494 then SvREFCNT_dec(), it will be cleaned up (and added to the free list)
14495 before the walk of unreferenced happens and a reference to that is SV
14496 added to the temps stack. At which point we have the same SV considered
14497 to be in use, and free to be re-used. Not good.
14499 if (dstr && !(param->flags & CLONEf_COPY_STACKS) && !SvREFCNT(dstr)) {
14500 assert(param->unreferenced);
14501 av_push(param->unreferenced, SvREFCNT_inc(dstr));
14507 /* duplicate a context */
14510 Perl_cx_dup(pTHX_ PERL_CONTEXT *cxs, I32 ix, I32 max, CLONE_PARAMS* param)
14512 PERL_CONTEXT *ncxs;
14514 PERL_ARGS_ASSERT_CX_DUP;
14517 return (PERL_CONTEXT*)NULL;
14519 /* look for it in the table first */
14520 ncxs = (PERL_CONTEXT*)ptr_table_fetch(PL_ptr_table, cxs);
14524 /* create anew and remember what it is */
14525 Newx(ncxs, max + 1, PERL_CONTEXT);
14526 ptr_table_store(PL_ptr_table, cxs, ncxs);
14527 Copy(cxs, ncxs, max + 1, PERL_CONTEXT);
14530 PERL_CONTEXT * const ncx = &ncxs[ix];
14531 if (CxTYPE(ncx) == CXt_SUBST) {
14532 Perl_croak(aTHX_ "Cloning substitution context is unimplemented");
14535 ncx->blk_oldcop = (COP*)any_dup(ncx->blk_oldcop, param->proto_perl);
14536 switch (CxTYPE(ncx)) {
14538 ncx->blk_sub.cv = cv_dup_inc(ncx->blk_sub.cv, param);
14539 if(CxHASARGS(ncx)){
14540 ncx->blk_sub.savearray = av_dup_inc(ncx->blk_sub.savearray,param);
14542 ncx->blk_sub.savearray = NULL;
14544 ncx->blk_sub.prevcomppad = (PAD*)ptr_table_fetch(PL_ptr_table,
14545 ncx->blk_sub.prevcomppad);
14548 ncx->blk_eval.old_namesv = sv_dup_inc(ncx->blk_eval.old_namesv,
14550 /* XXX should this sv_dup_inc? Or only if CxEVAL_TXT_REFCNTED ???? */
14551 ncx->blk_eval.cur_text = sv_dup(ncx->blk_eval.cur_text, param);
14552 ncx->blk_eval.cv = cv_dup(ncx->blk_eval.cv, param);
14553 /* XXX what do do with cur_top_env ???? */
14555 case CXt_LOOP_LAZYSV:
14556 ncx->blk_loop.state_u.lazysv.end
14557 = sv_dup_inc(ncx->blk_loop.state_u.lazysv.end, param);
14558 /* Fallthrough: duplicate lazysv.cur by using the ary.ary
14559 duplication code instead.
14560 We are taking advantage of (1) av_dup_inc and sv_dup_inc
14561 actually being the same function, and (2) order
14562 equivalence of the two unions.
14563 We can assert the later [but only at run time :-(] */
14564 assert ((void *) &ncx->blk_loop.state_u.ary.ary ==
14565 (void *) &ncx->blk_loop.state_u.lazysv.cur);
14568 ncx->blk_loop.state_u.ary.ary
14569 = av_dup_inc(ncx->blk_loop.state_u.ary.ary, param);
14571 case CXt_LOOP_LIST:
14572 case CXt_LOOP_LAZYIV:
14573 /* code common to all 'for' CXt_LOOP_* types */
14574 ncx->blk_loop.itersave =
14575 sv_dup_inc(ncx->blk_loop.itersave, param);
14576 if (CxPADLOOP(ncx)) {
14577 PADOFFSET off = ncx->blk_loop.itervar_u.svp
14578 - &CX_CURPAD_SV(ncx->blk_loop, 0);
14579 ncx->blk_loop.oldcomppad =
14580 (PAD*)ptr_table_fetch(PL_ptr_table,
14581 ncx->blk_loop.oldcomppad);
14582 ncx->blk_loop.itervar_u.svp =
14583 &CX_CURPAD_SV(ncx->blk_loop, off);
14586 /* this copies the GV if CXp_FOR_GV, or the SV for an
14587 * alias (for \$x (...)) - relies on gv_dup being the
14588 * same as sv_dup */
14589 ncx->blk_loop.itervar_u.gv
14590 = gv_dup((const GV *)ncx->blk_loop.itervar_u.gv,
14594 case CXt_LOOP_PLAIN:
14597 ncx->blk_format.prevcomppad =
14598 (PAD*)ptr_table_fetch(PL_ptr_table,
14599 ncx->blk_format.prevcomppad);
14600 ncx->blk_format.cv = cv_dup_inc(ncx->blk_format.cv, param);
14601 ncx->blk_format.gv = gv_dup(ncx->blk_format.gv, param);
14602 ncx->blk_format.dfoutgv = gv_dup_inc(ncx->blk_format.dfoutgv,
14606 ncx->blk_givwhen.defsv_save =
14607 sv_dup_inc(ncx->blk_givwhen.defsv_save, param);
14620 /* duplicate a stack info structure */
14623 Perl_si_dup(pTHX_ PERL_SI *si, CLONE_PARAMS* param)
14627 PERL_ARGS_ASSERT_SI_DUP;
14630 return (PERL_SI*)NULL;
14632 /* look for it in the table first */
14633 nsi = (PERL_SI*)ptr_table_fetch(PL_ptr_table, si);
14637 /* create anew and remember what it is */
14638 Newx(nsi, 1, PERL_SI);
14639 ptr_table_store(PL_ptr_table, si, nsi);
14641 nsi->si_stack = av_dup_inc(si->si_stack, param);
14642 nsi->si_cxix = si->si_cxix;
14643 nsi->si_cxmax = si->si_cxmax;
14644 nsi->si_cxstack = cx_dup(si->si_cxstack, si->si_cxix, si->si_cxmax, param);
14645 nsi->si_type = si->si_type;
14646 nsi->si_prev = si_dup(si->si_prev, param);
14647 nsi->si_next = si_dup(si->si_next, param);
14648 nsi->si_markoff = si->si_markoff;
14649 #if defined DEBUGGING && !defined DEBUGGING_RE_ONLY
14650 nsi->si_stack_hwm = 0;
14656 #define POPINT(ss,ix) ((ss)[--(ix)].any_i32)
14657 #define TOPINT(ss,ix) ((ss)[ix].any_i32)
14658 #define POPLONG(ss,ix) ((ss)[--(ix)].any_long)
14659 #define TOPLONG(ss,ix) ((ss)[ix].any_long)
14660 #define POPIV(ss,ix) ((ss)[--(ix)].any_iv)
14661 #define TOPIV(ss,ix) ((ss)[ix].any_iv)
14662 #define POPUV(ss,ix) ((ss)[--(ix)].any_uv)
14663 #define TOPUV(ss,ix) ((ss)[ix].any_uv)
14664 #define POPBOOL(ss,ix) ((ss)[--(ix)].any_bool)
14665 #define TOPBOOL(ss,ix) ((ss)[ix].any_bool)
14666 #define POPPTR(ss,ix) ((ss)[--(ix)].any_ptr)
14667 #define TOPPTR(ss,ix) ((ss)[ix].any_ptr)
14668 #define POPDPTR(ss,ix) ((ss)[--(ix)].any_dptr)
14669 #define TOPDPTR(ss,ix) ((ss)[ix].any_dptr)
14670 #define POPDXPTR(ss,ix) ((ss)[--(ix)].any_dxptr)
14671 #define TOPDXPTR(ss,ix) ((ss)[ix].any_dxptr)
14674 #define pv_dup_inc(p) SAVEPV(p)
14675 #define pv_dup(p) SAVEPV(p)
14676 #define svp_dup_inc(p,pp) any_dup(p,pp)
14678 /* map any object to the new equivent - either something in the
14679 * ptr table, or something in the interpreter structure
14683 Perl_any_dup(pTHX_ void *v, const PerlInterpreter *proto_perl)
14687 PERL_ARGS_ASSERT_ANY_DUP;
14690 return (void*)NULL;
14692 /* look for it in the table first */
14693 ret = ptr_table_fetch(PL_ptr_table, v);
14697 /* see if it is part of the interpreter structure */
14698 if (v >= (void*)proto_perl && v < (void*)(proto_perl+1))
14699 ret = (void*)(((char*)aTHX) + (((char*)v) - (char*)proto_perl));
14707 /* duplicate the save stack */
14710 Perl_ss_dup(pTHX_ PerlInterpreter *proto_perl, CLONE_PARAMS* param)
14713 ANY * const ss = proto_perl->Isavestack;
14714 const I32 max = proto_perl->Isavestack_max + SS_MAXPUSH;
14715 I32 ix = proto_perl->Isavestack_ix;
14728 void (*dptr) (void*);
14729 void (*dxptr) (pTHX_ void*);
14731 PERL_ARGS_ASSERT_SS_DUP;
14733 Newx(nss, max, ANY);
14736 const UV uv = POPUV(ss,ix);
14737 const U8 type = (U8)uv & SAVE_MASK;
14739 TOPUV(nss,ix) = uv;
14741 case SAVEt_CLEARSV:
14742 case SAVEt_CLEARPADRANGE:
14744 case SAVEt_HELEM: /* hash element */
14745 case SAVEt_SV: /* scalar reference */
14746 sv = (const SV *)POPPTR(ss,ix);
14747 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14749 case SAVEt_ITEM: /* normal string */
14750 case SAVEt_GVSV: /* scalar slot in GV */
14751 sv = (const SV *)POPPTR(ss,ix);
14752 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14753 if (type == SAVEt_SV)
14757 case SAVEt_MORTALIZESV:
14758 case SAVEt_READONLY_OFF:
14759 sv = (const SV *)POPPTR(ss,ix);
14760 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14762 case SAVEt_FREEPADNAME:
14763 ptr = POPPTR(ss,ix);
14764 TOPPTR(nss,ix) = padname_dup((PADNAME *)ptr, param);
14765 PadnameREFCNT((PADNAME *)TOPPTR(nss,ix))++;
14767 case SAVEt_SHARED_PVREF: /* char* in shared space */
14768 c = (char*)POPPTR(ss,ix);
14769 TOPPTR(nss,ix) = savesharedpv(c);
14770 ptr = POPPTR(ss,ix);
14771 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14773 case SAVEt_GENERIC_SVREF: /* generic sv */
14774 case SAVEt_SVREF: /* scalar reference */
14775 sv = (const SV *)POPPTR(ss,ix);
14776 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14777 if (type == SAVEt_SVREF)
14778 SvREFCNT_inc_simple_void((SV *)TOPPTR(nss,ix));
14779 ptr = POPPTR(ss,ix);
14780 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14782 case SAVEt_GVSLOT: /* any slot in GV */
14783 sv = (const SV *)POPPTR(ss,ix);
14784 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14785 ptr = POPPTR(ss,ix);
14786 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14787 sv = (const SV *)POPPTR(ss,ix);
14788 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14790 case SAVEt_HV: /* hash reference */
14791 case SAVEt_AV: /* array reference */
14792 sv = (const SV *) POPPTR(ss,ix);
14793 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14795 case SAVEt_COMPPAD:
14797 sv = (const SV *) POPPTR(ss,ix);
14798 TOPPTR(nss,ix) = sv_dup(sv, param);
14800 case SAVEt_INT: /* int reference */
14801 ptr = POPPTR(ss,ix);
14802 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14803 intval = (int)POPINT(ss,ix);
14804 TOPINT(nss,ix) = intval;
14806 case SAVEt_LONG: /* long reference */
14807 ptr = POPPTR(ss,ix);
14808 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14809 longval = (long)POPLONG(ss,ix);
14810 TOPLONG(nss,ix) = longval;
14812 case SAVEt_I32: /* I32 reference */
14813 ptr = POPPTR(ss,ix);
14814 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14816 TOPINT(nss,ix) = i;
14818 case SAVEt_IV: /* IV reference */
14819 case SAVEt_STRLEN: /* STRLEN/size_t ref */
14820 ptr = POPPTR(ss,ix);
14821 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14823 TOPIV(nss,ix) = iv;
14825 case SAVEt_TMPSFLOOR:
14827 TOPIV(nss,ix) = iv;
14829 case SAVEt_HPTR: /* HV* reference */
14830 case SAVEt_APTR: /* AV* reference */
14831 case SAVEt_SPTR: /* SV* reference */
14832 ptr = POPPTR(ss,ix);
14833 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14834 sv = (const SV *)POPPTR(ss,ix);
14835 TOPPTR(nss,ix) = sv_dup(sv, param);
14837 case SAVEt_VPTR: /* random* reference */
14838 ptr = POPPTR(ss,ix);
14839 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14841 case SAVEt_INT_SMALL:
14842 case SAVEt_I32_SMALL:
14843 case SAVEt_I16: /* I16 reference */
14844 case SAVEt_I8: /* I8 reference */
14846 ptr = POPPTR(ss,ix);
14847 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14849 case SAVEt_GENERIC_PVREF: /* generic char* */
14850 case SAVEt_PPTR: /* char* reference */
14851 ptr = POPPTR(ss,ix);
14852 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14853 c = (char*)POPPTR(ss,ix);
14854 TOPPTR(nss,ix) = pv_dup(c);
14856 case SAVEt_GP: /* scalar reference */
14857 gp = (GP*)POPPTR(ss,ix);
14858 TOPPTR(nss,ix) = gp = gp_dup(gp, param);
14859 (void)GpREFCNT_inc(gp);
14860 gv = (const GV *)POPPTR(ss,ix);
14861 TOPPTR(nss,ix) = gv_dup_inc(gv, param);
14864 ptr = POPPTR(ss,ix);
14865 if (ptr && (((OP*)ptr)->op_private & OPpREFCOUNTED)) {
14866 /* these are assumed to be refcounted properly */
14868 switch (((OP*)ptr)->op_type) {
14870 case OP_LEAVESUBLV:
14874 case OP_LEAVEWRITE:
14875 TOPPTR(nss,ix) = ptr;
14878 (void) OpREFCNT_inc(o);
14882 TOPPTR(nss,ix) = NULL;
14887 TOPPTR(nss,ix) = NULL;
14889 case SAVEt_FREECOPHH:
14890 ptr = POPPTR(ss,ix);
14891 TOPPTR(nss,ix) = cophh_copy((COPHH *)ptr);
14893 case SAVEt_ADELETE:
14894 av = (const AV *)POPPTR(ss,ix);
14895 TOPPTR(nss,ix) = av_dup_inc(av, param);
14897 TOPINT(nss,ix) = i;
14900 hv = (const HV *)POPPTR(ss,ix);
14901 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14903 TOPINT(nss,ix) = i;
14906 c = (char*)POPPTR(ss,ix);
14907 TOPPTR(nss,ix) = pv_dup_inc(c);
14909 case SAVEt_STACK_POS: /* Position on Perl stack */
14911 TOPINT(nss,ix) = i;
14913 case SAVEt_DESTRUCTOR:
14914 ptr = POPPTR(ss,ix);
14915 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14916 dptr = POPDPTR(ss,ix);
14917 TOPDPTR(nss,ix) = DPTR2FPTR(void (*)(void*),
14918 any_dup(FPTR2DPTR(void *, dptr),
14921 case SAVEt_DESTRUCTOR_X:
14922 ptr = POPPTR(ss,ix);
14923 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14924 dxptr = POPDXPTR(ss,ix);
14925 TOPDXPTR(nss,ix) = DPTR2FPTR(void (*)(pTHX_ void*),
14926 any_dup(FPTR2DPTR(void *, dxptr),
14929 case SAVEt_REGCONTEXT:
14931 ix -= uv >> SAVE_TIGHT_SHIFT;
14933 case SAVEt_AELEM: /* array element */
14934 sv = (const SV *)POPPTR(ss,ix);
14935 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14937 TOPIV(nss,ix) = iv;
14938 av = (const AV *)POPPTR(ss,ix);
14939 TOPPTR(nss,ix) = av_dup_inc(av, param);
14942 ptr = POPPTR(ss,ix);
14943 TOPPTR(nss,ix) = ptr;
14946 ptr = POPPTR(ss,ix);
14947 ptr = cophh_copy((COPHH*)ptr);
14948 TOPPTR(nss,ix) = ptr;
14950 TOPINT(nss,ix) = i;
14951 if (i & HINT_LOCALIZE_HH) {
14952 hv = (const HV *)POPPTR(ss,ix);
14953 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14956 case SAVEt_PADSV_AND_MORTALIZE:
14957 longval = (long)POPLONG(ss,ix);
14958 TOPLONG(nss,ix) = longval;
14959 ptr = POPPTR(ss,ix);
14960 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14961 sv = (const SV *)POPPTR(ss,ix);
14962 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14964 case SAVEt_SET_SVFLAGS:
14966 TOPINT(nss,ix) = i;
14968 TOPINT(nss,ix) = i;
14969 sv = (const SV *)POPPTR(ss,ix);
14970 TOPPTR(nss,ix) = sv_dup(sv, param);
14972 case SAVEt_COMPILE_WARNINGS:
14973 ptr = POPPTR(ss,ix);
14974 TOPPTR(nss,ix) = DUP_WARNINGS((STRLEN*)ptr);
14977 ptr = POPPTR(ss,ix);
14978 TOPPTR(nss,ix) = parser_dup((const yy_parser*)ptr, param);
14982 "panic: ss_dup inconsistency (%" IVdf ")", (IV) type);
14990 /* if sv is a stash, call $class->CLONE_SKIP(), and set the SVphv_CLONEABLE
14991 * flag to the result. This is done for each stash before cloning starts,
14992 * so we know which stashes want their objects cloned */
14995 do_mark_cloneable_stash(pTHX_ SV *const sv)
14997 const HEK * const hvname = HvNAME_HEK((const HV *)sv);
14999 GV* const cloner = gv_fetchmethod_autoload(MUTABLE_HV(sv), "CLONE_SKIP", 0);
15000 SvFLAGS(sv) |= SVphv_CLONEABLE; /* clone objects by default */
15001 if (cloner && GvCV(cloner)) {
15008 mXPUSHs(newSVhek(hvname));
15010 call_sv(MUTABLE_SV(GvCV(cloner)), G_SCALAR);
15017 SvFLAGS(sv) &= ~SVphv_CLONEABLE;
15025 =for apidoc perl_clone
15027 Create and return a new interpreter by cloning the current one.
15029 C<perl_clone> takes these flags as parameters:
15031 C<CLONEf_COPY_STACKS> - is used to, well, copy the stacks also,
15032 without it we only clone the data and zero the stacks,
15033 with it we copy the stacks and the new perl interpreter is
15034 ready to run at the exact same point as the previous one.
15035 The pseudo-fork code uses C<COPY_STACKS> while the
15036 threads->create doesn't.
15038 C<CLONEf_KEEP_PTR_TABLE> -
15039 C<perl_clone> keeps a ptr_table with the pointer of the old
15040 variable as a key and the new variable as a value,
15041 this allows it to check if something has been cloned and not
15042 clone it again but rather just use the value and increase the
15043 refcount. If C<KEEP_PTR_TABLE> is not set then C<perl_clone> will kill
15044 the ptr_table using the function
15045 C<ptr_table_free(PL_ptr_table); PL_ptr_table = NULL;>,
15046 reason to keep it around is if you want to dup some of your own
15047 variable who are outside the graph perl scans, an example of this
15048 code is in F<threads.xs> create.
15050 C<CLONEf_CLONE_HOST> -
15051 This is a win32 thing, it is ignored on unix, it tells perls
15052 win32host code (which is c++) to clone itself, this is needed on
15053 win32 if you want to run two threads at the same time,
15054 if you just want to do some stuff in a separate perl interpreter
15055 and then throw it away and return to the original one,
15056 you don't need to do anything.
15061 /* XXX the above needs expanding by someone who actually understands it ! */
15062 EXTERN_C PerlInterpreter *
15063 perl_clone_host(PerlInterpreter* proto_perl, UV flags);
15066 perl_clone(PerlInterpreter *proto_perl, UV flags)
15069 #ifdef PERL_IMPLICIT_SYS
15071 PERL_ARGS_ASSERT_PERL_CLONE;
15073 /* perlhost.h so we need to call into it
15074 to clone the host, CPerlHost should have a c interface, sky */
15076 #ifndef __amigaos4__
15077 if (flags & CLONEf_CLONE_HOST) {
15078 return perl_clone_host(proto_perl,flags);
15081 return perl_clone_using(proto_perl, flags,
15083 proto_perl->IMemShared,
15084 proto_perl->IMemParse,
15086 proto_perl->IStdIO,
15090 proto_perl->IProc);
15094 perl_clone_using(PerlInterpreter *proto_perl, UV flags,
15095 struct IPerlMem* ipM, struct IPerlMem* ipMS,
15096 struct IPerlMem* ipMP, struct IPerlEnv* ipE,
15097 struct IPerlStdIO* ipStd, struct IPerlLIO* ipLIO,
15098 struct IPerlDir* ipD, struct IPerlSock* ipS,
15099 struct IPerlProc* ipP)
15101 /* XXX many of the string copies here can be optimized if they're
15102 * constants; they need to be allocated as common memory and just
15103 * their pointers copied. */
15106 CLONE_PARAMS clone_params;
15107 CLONE_PARAMS* const param = &clone_params;
15109 PerlInterpreter * const my_perl = (PerlInterpreter*)(*ipM->pMalloc)(ipM, sizeof(PerlInterpreter));
15111 PERL_ARGS_ASSERT_PERL_CLONE_USING;
15112 #else /* !PERL_IMPLICIT_SYS */
15114 CLONE_PARAMS clone_params;
15115 CLONE_PARAMS* param = &clone_params;
15116 PerlInterpreter * const my_perl = (PerlInterpreter*)PerlMem_malloc(sizeof(PerlInterpreter));
15118 PERL_ARGS_ASSERT_PERL_CLONE;
15119 #endif /* PERL_IMPLICIT_SYS */
15121 /* for each stash, determine whether its objects should be cloned */
15122 S_visit(proto_perl, do_mark_cloneable_stash, SVt_PVHV, SVTYPEMASK);
15123 PERL_SET_THX(my_perl);
15126 PoisonNew(my_perl, 1, PerlInterpreter);
15129 PL_defstash = NULL; /* may be used by perl malloc() */
15132 PL_scopestack_name = 0;
15134 PL_savestack_ix = 0;
15135 PL_savestack_max = -1;
15136 PL_sig_pending = 0;
15138 Zero(&PL_debug_pad, 1, struct perl_debug_pad);
15139 Zero(&PL_padname_undef, 1, PADNAME);
15140 Zero(&PL_padname_const, 1, PADNAME);
15141 # ifdef DEBUG_LEAKING_SCALARS
15142 PL_sv_serial = (((UV)my_perl >> 2) & 0xfff) * 1000000;
15144 # ifdef PERL_TRACE_OPS
15145 Zero(PL_op_exec_cnt, OP_max+2, UV);
15147 #else /* !DEBUGGING */
15148 Zero(my_perl, 1, PerlInterpreter);
15149 #endif /* DEBUGGING */
15151 #ifdef PERL_IMPLICIT_SYS
15152 /* host pointers */
15154 PL_MemShared = ipMS;
15155 PL_MemParse = ipMP;
15162 #endif /* PERL_IMPLICIT_SYS */
15165 param->flags = flags;
15166 /* Nothing in the core code uses this, but we make it available to
15167 extensions (using mg_dup). */
15168 param->proto_perl = proto_perl;
15169 /* Likely nothing will use this, but it is initialised to be consistent
15170 with Perl_clone_params_new(). */
15171 param->new_perl = my_perl;
15172 param->unreferenced = NULL;
15175 INIT_TRACK_MEMPOOL(my_perl->Imemory_debug_header, my_perl);
15177 PL_body_arenas = NULL;
15178 Zero(&PL_body_roots, 1, PL_body_roots);
15182 PL_sv_arenaroot = NULL;
15184 PL_debug = proto_perl->Idebug;
15186 /* dbargs array probably holds garbage */
15189 PL_compiling = proto_perl->Icompiling;
15191 /* pseudo environmental stuff */
15192 PL_origargc = proto_perl->Iorigargc;
15193 PL_origargv = proto_perl->Iorigargv;
15195 #ifndef NO_TAINT_SUPPORT
15196 /* Set tainting stuff before PerlIO_debug can possibly get called */
15197 PL_tainting = proto_perl->Itainting;
15198 PL_taint_warn = proto_perl->Itaint_warn;
15200 PL_tainting = FALSE;
15201 PL_taint_warn = FALSE;
15204 PL_minus_c = proto_perl->Iminus_c;
15206 PL_localpatches = proto_perl->Ilocalpatches;
15207 PL_splitstr = proto_perl->Isplitstr;
15208 PL_minus_n = proto_perl->Iminus_n;
15209 PL_minus_p = proto_perl->Iminus_p;
15210 PL_minus_l = proto_perl->Iminus_l;
15211 PL_minus_a = proto_perl->Iminus_a;
15212 PL_minus_E = proto_perl->Iminus_E;
15213 PL_minus_F = proto_perl->Iminus_F;
15214 PL_doswitches = proto_perl->Idoswitches;
15215 PL_dowarn = proto_perl->Idowarn;
15216 #ifdef PERL_SAWAMPERSAND
15217 PL_sawampersand = proto_perl->Isawampersand;
15219 PL_unsafe = proto_perl->Iunsafe;
15220 PL_perldb = proto_perl->Iperldb;
15221 PL_perl_destruct_level = proto_perl->Iperl_destruct_level;
15222 PL_exit_flags = proto_perl->Iexit_flags;
15224 /* XXX time(&PL_basetime) when asked for? */
15225 PL_basetime = proto_perl->Ibasetime;
15227 PL_maxsysfd = proto_perl->Imaxsysfd;
15228 PL_statusvalue = proto_perl->Istatusvalue;
15230 PL_statusvalue_vms = proto_perl->Istatusvalue_vms;
15232 PL_statusvalue_posix = proto_perl->Istatusvalue_posix;
15235 /* RE engine related */
15236 PL_regmatch_slab = NULL;
15237 PL_reg_curpm = NULL;
15239 PL_sub_generation = proto_perl->Isub_generation;
15241 /* funky return mechanisms */
15242 PL_forkprocess = proto_perl->Iforkprocess;
15244 /* internal state */
15245 PL_main_start = proto_perl->Imain_start;
15246 PL_eval_root = proto_perl->Ieval_root;
15247 PL_eval_start = proto_perl->Ieval_start;
15249 PL_filemode = proto_perl->Ifilemode;
15250 PL_lastfd = proto_perl->Ilastfd;
15251 PL_oldname = proto_perl->Ioldname; /* XXX not quite right */
15252 PL_gensym = proto_perl->Igensym;
15254 PL_laststatval = proto_perl->Ilaststatval;
15255 PL_laststype = proto_perl->Ilaststype;
15258 PL_profiledata = NULL;
15260 PL_generation = proto_perl->Igeneration;
15262 PL_in_clean_objs = proto_perl->Iin_clean_objs;
15263 PL_in_clean_all = proto_perl->Iin_clean_all;
15265 PL_delaymagic_uid = proto_perl->Idelaymagic_uid;
15266 PL_delaymagic_euid = proto_perl->Idelaymagic_euid;
15267 PL_delaymagic_gid = proto_perl->Idelaymagic_gid;
15268 PL_delaymagic_egid = proto_perl->Idelaymagic_egid;
15269 PL_nomemok = proto_perl->Inomemok;
15270 PL_an = proto_perl->Ian;
15271 PL_evalseq = proto_perl->Ievalseq;
15272 PL_origenviron = proto_perl->Iorigenviron; /* XXX not quite right */
15273 PL_origalen = proto_perl->Iorigalen;
15275 PL_sighandlerp = proto_perl->Isighandlerp;
15277 PL_runops = proto_perl->Irunops;
15279 PL_subline = proto_perl->Isubline;
15281 PL_cv_has_eval = proto_perl->Icv_has_eval;
15284 PL_cryptseen = proto_perl->Icryptseen;
15287 #ifdef USE_LOCALE_COLLATE
15288 PL_collation_ix = proto_perl->Icollation_ix;
15289 PL_collation_standard = proto_perl->Icollation_standard;
15290 PL_collxfrm_base = proto_perl->Icollxfrm_base;
15291 PL_collxfrm_mult = proto_perl->Icollxfrm_mult;
15292 PL_strxfrm_max_cp = proto_perl->Istrxfrm_max_cp;
15293 #endif /* USE_LOCALE_COLLATE */
15295 #ifdef USE_LOCALE_NUMERIC
15296 PL_numeric_standard = proto_perl->Inumeric_standard;
15297 PL_numeric_underlying = proto_perl->Inumeric_underlying;
15298 PL_numeric_underlying_is_standard = proto_perl->Inumeric_underlying_is_standard;
15299 #endif /* !USE_LOCALE_NUMERIC */
15301 /* Did the locale setup indicate UTF-8? */
15302 PL_utf8locale = proto_perl->Iutf8locale;
15303 PL_in_utf8_CTYPE_locale = proto_perl->Iin_utf8_CTYPE_locale;
15304 PL_in_utf8_COLLATE_locale = proto_perl->Iin_utf8_COLLATE_locale;
15305 my_strlcpy(PL_locale_utf8ness, proto_perl->Ilocale_utf8ness, sizeof(PL_locale_utf8ness));
15306 #if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
15307 PL_lc_numeric_mutex_depth = 0;
15309 /* Unicode features (see perlrun/-C) */
15310 PL_unicode = proto_perl->Iunicode;
15312 /* Pre-5.8 signals control */
15313 PL_signals = proto_perl->Isignals;
15315 /* times() ticks per second */
15316 PL_clocktick = proto_perl->Iclocktick;
15318 /* Recursion stopper for PerlIO_find_layer */
15319 PL_in_load_module = proto_perl->Iin_load_module;
15321 /* sort() routine */
15322 PL_sort_RealCmp = proto_perl->Isort_RealCmp;
15324 /* Not really needed/useful since the reenrant_retint is "volatile",
15325 * but do it for consistency's sake. */
15326 PL_reentrant_retint = proto_perl->Ireentrant_retint;
15328 /* Hooks to shared SVs and locks. */
15329 PL_sharehook = proto_perl->Isharehook;
15330 PL_lockhook = proto_perl->Ilockhook;
15331 PL_unlockhook = proto_perl->Iunlockhook;
15332 PL_threadhook = proto_perl->Ithreadhook;
15333 PL_destroyhook = proto_perl->Idestroyhook;
15334 PL_signalhook = proto_perl->Isignalhook;
15336 PL_globhook = proto_perl->Iglobhook;
15339 PL_last_swash_hv = NULL; /* reinits on demand */
15340 PL_last_swash_klen = 0;
15341 PL_last_swash_key[0]= '\0';
15342 PL_last_swash_tmps = (U8*)NULL;
15343 PL_last_swash_slen = 0;
15345 PL_srand_called = proto_perl->Isrand_called;
15346 Copy(&(proto_perl->Irandom_state), &PL_random_state, 1, PL_RANDOM_STATE_TYPE);
15348 if (flags & CLONEf_COPY_STACKS) {
15349 /* next allocation will be PL_tmps_stack[PL_tmps_ix+1] */
15350 PL_tmps_ix = proto_perl->Itmps_ix;
15351 PL_tmps_max = proto_perl->Itmps_max;
15352 PL_tmps_floor = proto_perl->Itmps_floor;
15354 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
15355 * NOTE: unlike the others! */
15356 PL_scopestack_ix = proto_perl->Iscopestack_ix;
15357 PL_scopestack_max = proto_perl->Iscopestack_max;
15359 /* next SSPUSHFOO() sets PL_savestack[PL_savestack_ix]
15360 * NOTE: unlike the others! */
15361 PL_savestack_ix = proto_perl->Isavestack_ix;
15362 PL_savestack_max = proto_perl->Isavestack_max;
15365 PL_start_env = proto_perl->Istart_env; /* XXXXXX */
15366 PL_top_env = &PL_start_env;
15368 PL_op = proto_perl->Iop;
15371 PL_Xpv = (XPV*)NULL;
15372 my_perl->Ina = proto_perl->Ina;
15374 PL_statcache = proto_perl->Istatcache;
15376 #ifndef NO_TAINT_SUPPORT
15377 PL_tainted = proto_perl->Itainted;
15379 PL_tainted = FALSE;
15381 PL_curpm = proto_perl->Icurpm; /* XXX No PMOP ref count */
15383 PL_chopset = proto_perl->Ichopset; /* XXX never deallocated */
15385 PL_restartjmpenv = proto_perl->Irestartjmpenv;
15386 PL_restartop = proto_perl->Irestartop;
15387 PL_in_eval = proto_perl->Iin_eval;
15388 PL_delaymagic = proto_perl->Idelaymagic;
15389 PL_phase = proto_perl->Iphase;
15390 PL_localizing = proto_perl->Ilocalizing;
15392 PL_hv_fetch_ent_mh = NULL;
15393 PL_modcount = proto_perl->Imodcount;
15394 PL_lastgotoprobe = NULL;
15395 PL_dumpindent = proto_perl->Idumpindent;
15397 PL_efloatbuf = NULL; /* reinits on demand */
15398 PL_efloatsize = 0; /* reinits on demand */
15402 PL_colorset = 0; /* reinits PL_colors[] */
15403 /*PL_colors[6] = {0,0,0,0,0,0};*/
15405 /* Pluggable optimizer */
15406 PL_peepp = proto_perl->Ipeepp;
15407 PL_rpeepp = proto_perl->Irpeepp;
15408 /* op_free() hook */
15409 PL_opfreehook = proto_perl->Iopfreehook;
15411 #ifdef USE_REENTRANT_API
15412 /* XXX: things like -Dm will segfault here in perlio, but doing
15413 * PERL_SET_CONTEXT(proto_perl);
15414 * breaks too many other things
15416 Perl_reentrant_init(aTHX);
15419 /* create SV map for pointer relocation */
15420 PL_ptr_table = ptr_table_new();
15422 /* initialize these special pointers as early as possible */
15424 ptr_table_store(PL_ptr_table, &proto_perl->Isv_undef, &PL_sv_undef);
15425 ptr_table_store(PL_ptr_table, &proto_perl->Isv_no, &PL_sv_no);
15426 ptr_table_store(PL_ptr_table, &proto_perl->Isv_zero, &PL_sv_zero);
15427 ptr_table_store(PL_ptr_table, &proto_perl->Isv_yes, &PL_sv_yes);
15428 ptr_table_store(PL_ptr_table, &proto_perl->Ipadname_const,
15429 &PL_padname_const);
15431 /* create (a non-shared!) shared string table */
15432 PL_strtab = newHV();
15433 HvSHAREKEYS_off(PL_strtab);
15434 hv_ksplit(PL_strtab, HvTOTALKEYS(proto_perl->Istrtab));
15435 ptr_table_store(PL_ptr_table, proto_perl->Istrtab, PL_strtab);
15437 Zero(PL_sv_consts, SV_CONSTS_COUNT, SV*);
15439 /* This PV will be free'd special way so must set it same way op.c does */
15440 PL_compiling.cop_file = savesharedpv(PL_compiling.cop_file);
15441 ptr_table_store(PL_ptr_table, proto_perl->Icompiling.cop_file, PL_compiling.cop_file);
15443 ptr_table_store(PL_ptr_table, &proto_perl->Icompiling, &PL_compiling);
15444 PL_compiling.cop_warnings = DUP_WARNINGS(PL_compiling.cop_warnings);
15445 CopHINTHASH_set(&PL_compiling, cophh_copy(CopHINTHASH_get(&PL_compiling)));
15446 PL_curcop = (COP*)any_dup(proto_perl->Icurcop, proto_perl);
15448 param->stashes = newAV(); /* Setup array of objects to call clone on */
15449 /* This makes no difference to the implementation, as it always pushes
15450 and shifts pointers to other SVs without changing their reference
15451 count, with the array becoming empty before it is freed. However, it
15452 makes it conceptually clear what is going on, and will avoid some
15453 work inside av.c, filling slots between AvFILL() and AvMAX() with
15454 &PL_sv_undef, and SvREFCNT_dec()ing those. */
15455 AvREAL_off(param->stashes);
15457 if (!(flags & CLONEf_COPY_STACKS)) {
15458 param->unreferenced = newAV();
15461 #ifdef PERLIO_LAYERS
15462 /* Clone PerlIO tables as soon as we can handle general xx_dup() */
15463 PerlIO_clone(aTHX_ proto_perl, param);
15466 PL_envgv = gv_dup_inc(proto_perl->Ienvgv, param);
15467 PL_incgv = gv_dup_inc(proto_perl->Iincgv, param);
15468 PL_hintgv = gv_dup_inc(proto_perl->Ihintgv, param);
15469 PL_origfilename = SAVEPV(proto_perl->Iorigfilename);
15470 PL_xsubfilename = proto_perl->Ixsubfilename;
15471 PL_diehook = sv_dup_inc(proto_perl->Idiehook, param);
15472 PL_warnhook = sv_dup_inc(proto_perl->Iwarnhook, param);
15475 PL_patchlevel = sv_dup_inc(proto_perl->Ipatchlevel, param);
15476 PL_inplace = SAVEPV(proto_perl->Iinplace);
15477 PL_e_script = sv_dup_inc(proto_perl->Ie_script, param);
15479 /* magical thingies */
15481 SvPVCLEAR(PERL_DEBUG_PAD(0)); /* For regex debugging. */
15482 SvPVCLEAR(PERL_DEBUG_PAD(1)); /* ext/re needs these */
15483 SvPVCLEAR(PERL_DEBUG_PAD(2)); /* even without DEBUGGING. */
15486 /* Clone the regex array */
15487 /* ORANGE FIXME for plugins, probably in the SV dup code.
15488 newSViv(PTR2IV(CALLREGDUPE(
15489 INT2PTR(REGEXP *, SvIVX(regex)), param))))
15491 PL_regex_padav = av_dup_inc(proto_perl->Iregex_padav, param);
15492 PL_regex_pad = AvARRAY(PL_regex_padav);
15494 PL_stashpadmax = proto_perl->Istashpadmax;
15495 PL_stashpadix = proto_perl->Istashpadix ;
15496 Newx(PL_stashpad, PL_stashpadmax, HV *);
15499 for (; o < PL_stashpadmax; ++o)
15500 PL_stashpad[o] = hv_dup(proto_perl->Istashpad[o], param);
15503 /* shortcuts to various I/O objects */
15504 PL_ofsgv = gv_dup_inc(proto_perl->Iofsgv, param);
15505 PL_stdingv = gv_dup(proto_perl->Istdingv, param);
15506 PL_stderrgv = gv_dup(proto_perl->Istderrgv, param);
15507 PL_defgv = gv_dup(proto_perl->Idefgv, param);
15508 PL_argvgv = gv_dup_inc(proto_perl->Iargvgv, param);
15509 PL_argvoutgv = gv_dup(proto_perl->Iargvoutgv, param);
15510 PL_argvout_stack = av_dup_inc(proto_perl->Iargvout_stack, param);
15512 /* shortcuts to regexp stuff */
15513 PL_replgv = gv_dup_inc(proto_perl->Ireplgv, param);
15515 /* shortcuts to misc objects */
15516 PL_errgv = gv_dup(proto_perl->Ierrgv, param);
15518 /* shortcuts to debugging objects */
15519 PL_DBgv = gv_dup_inc(proto_perl->IDBgv, param);
15520 PL_DBline = gv_dup_inc(proto_perl->IDBline, param);
15521 PL_DBsub = gv_dup_inc(proto_perl->IDBsub, param);
15522 PL_DBsingle = sv_dup(proto_perl->IDBsingle, param);
15523 PL_DBtrace = sv_dup(proto_perl->IDBtrace, param);
15524 PL_DBsignal = sv_dup(proto_perl->IDBsignal, param);
15525 Copy(proto_perl->IDBcontrol, PL_DBcontrol, DBVARMG_COUNT, IV);
15527 /* symbol tables */
15528 PL_defstash = hv_dup_inc(proto_perl->Idefstash, param);
15529 PL_curstash = hv_dup_inc(proto_perl->Icurstash, param);
15530 PL_debstash = hv_dup(proto_perl->Idebstash, param);
15531 PL_globalstash = hv_dup(proto_perl->Iglobalstash, param);
15532 PL_curstname = sv_dup_inc(proto_perl->Icurstname, param);
15534 PL_beginav = av_dup_inc(proto_perl->Ibeginav, param);
15535 PL_beginav_save = av_dup_inc(proto_perl->Ibeginav_save, param);
15536 PL_checkav_save = av_dup_inc(proto_perl->Icheckav_save, param);
15537 PL_unitcheckav = av_dup_inc(proto_perl->Iunitcheckav, param);
15538 PL_unitcheckav_save = av_dup_inc(proto_perl->Iunitcheckav_save, param);
15539 PL_endav = av_dup_inc(proto_perl->Iendav, param);
15540 PL_checkav = av_dup_inc(proto_perl->Icheckav, param);
15541 PL_initav = av_dup_inc(proto_perl->Iinitav, param);
15542 PL_savebegin = proto_perl->Isavebegin;
15544 PL_isarev = hv_dup_inc(proto_perl->Iisarev, param);
15546 /* subprocess state */
15547 PL_fdpid = av_dup_inc(proto_perl->Ifdpid, param);
15549 if (proto_perl->Iop_mask)
15550 PL_op_mask = SAVEPVN(proto_perl->Iop_mask, PL_maxo);
15553 /* PL_asserting = proto_perl->Iasserting; */
15555 /* current interpreter roots */
15556 PL_main_cv = cv_dup_inc(proto_perl->Imain_cv, param);
15558 PL_main_root = OpREFCNT_inc(proto_perl->Imain_root);
15561 /* runtime control stuff */
15562 PL_curcopdb = (COP*)any_dup(proto_perl->Icurcopdb, proto_perl);
15564 PL_preambleav = av_dup_inc(proto_perl->Ipreambleav, param);
15566 PL_ors_sv = sv_dup_inc(proto_perl->Iors_sv, param);
15568 /* interpreter atexit processing */
15569 PL_exitlistlen = proto_perl->Iexitlistlen;
15570 if (PL_exitlistlen) {
15571 Newx(PL_exitlist, PL_exitlistlen, PerlExitListEntry);
15572 Copy(proto_perl->Iexitlist, PL_exitlist, PL_exitlistlen, PerlExitListEntry);
15575 PL_exitlist = (PerlExitListEntry*)NULL;
15577 PL_my_cxt_size = proto_perl->Imy_cxt_size;
15578 if (PL_my_cxt_size) {
15579 Newx(PL_my_cxt_list, PL_my_cxt_size, void *);
15580 Copy(proto_perl->Imy_cxt_list, PL_my_cxt_list, PL_my_cxt_size, void *);
15583 PL_my_cxt_list = (void**)NULL;
15585 PL_modglobal = hv_dup_inc(proto_perl->Imodglobal, param);
15586 PL_custom_op_names = hv_dup_inc(proto_perl->Icustom_op_names,param);
15587 PL_custom_op_descs = hv_dup_inc(proto_perl->Icustom_op_descs,param);
15588 PL_custom_ops = hv_dup_inc(proto_perl->Icustom_ops, param);
15590 PL_compcv = cv_dup(proto_perl->Icompcv, param);
15592 PAD_CLONE_VARS(proto_perl, param);
15594 #ifdef HAVE_INTERP_INTERN
15595 sys_intern_dup(&proto_perl->Isys_intern, &PL_sys_intern);
15598 PL_DBcv = cv_dup(proto_perl->IDBcv, param);
15600 #ifdef PERL_USES_PL_PIDSTATUS
15601 PL_pidstatus = newHV(); /* XXX flag for cloning? */
15603 PL_osname = SAVEPV(proto_perl->Iosname);
15604 PL_parser = parser_dup(proto_perl->Iparser, param);
15606 /* XXX this only works if the saved cop has already been cloned */
15607 if (proto_perl->Iparser) {
15608 PL_parser->saved_curcop = (COP*)any_dup(
15609 proto_perl->Iparser->saved_curcop,
15613 PL_subname = sv_dup_inc(proto_perl->Isubname, param);
15615 #if defined(USE_POSIX_2008_LOCALE) \
15616 && defined(USE_THREAD_SAFE_LOCALE) \
15617 && ! defined(HAS_QUERYLOCALE)
15618 for (i = 0; i < (int) C_ARRAY_LENGTH(PL_curlocales); i++) {
15619 PL_curlocales[i] = savepv("."); /* An illegal value */
15622 #ifdef USE_LOCALE_CTYPE
15623 /* Should we warn if uses locale? */
15624 PL_warn_locale = sv_dup_inc(proto_perl->Iwarn_locale, param);
15627 #ifdef USE_LOCALE_COLLATE
15628 PL_collation_name = SAVEPV(proto_perl->Icollation_name);
15629 #endif /* USE_LOCALE_COLLATE */
15631 #ifdef USE_LOCALE_NUMERIC
15632 PL_numeric_name = SAVEPV(proto_perl->Inumeric_name);
15633 PL_numeric_radix_sv = sv_dup_inc(proto_perl->Inumeric_radix_sv, param);
15635 # if defined(HAS_POSIX_2008_LOCALE)
15636 PL_underlying_numeric_obj = NULL;
15638 #endif /* !USE_LOCALE_NUMERIC */
15640 PL_langinfo_buf = NULL;
15641 PL_langinfo_bufsize = 0;
15643 PL_setlocale_buf = NULL;
15644 PL_setlocale_bufsize = 0;
15646 /* utf8 character class swashes */
15647 PL_seen_deprecated_macro = hv_dup_inc(proto_perl->Iseen_deprecated_macro, param);
15649 if (proto_perl->Ipsig_pend) {
15650 Newxz(PL_psig_pend, SIG_SIZE, int);
15653 PL_psig_pend = (int*)NULL;
15656 if (proto_perl->Ipsig_name) {
15657 Newx(PL_psig_name, 2 * SIG_SIZE, SV*);
15658 sv_dup_inc_multiple(proto_perl->Ipsig_name, PL_psig_name, 2 * SIG_SIZE,
15660 PL_psig_ptr = PL_psig_name + SIG_SIZE;
15663 PL_psig_ptr = (SV**)NULL;
15664 PL_psig_name = (SV**)NULL;
15667 if (flags & CLONEf_COPY_STACKS) {
15668 Newx(PL_tmps_stack, PL_tmps_max, SV*);
15669 sv_dup_inc_multiple(proto_perl->Itmps_stack, PL_tmps_stack,
15670 PL_tmps_ix+1, param);
15672 /* next PUSHMARK() sets *(PL_markstack_ptr+1) */
15673 i = proto_perl->Imarkstack_max - proto_perl->Imarkstack;
15674 Newx(PL_markstack, i, I32);
15675 PL_markstack_max = PL_markstack + (proto_perl->Imarkstack_max
15676 - proto_perl->Imarkstack);
15677 PL_markstack_ptr = PL_markstack + (proto_perl->Imarkstack_ptr
15678 - proto_perl->Imarkstack);
15679 Copy(proto_perl->Imarkstack, PL_markstack,
15680 PL_markstack_ptr - PL_markstack + 1, I32);
15682 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
15683 * NOTE: unlike the others! */
15684 Newx(PL_scopestack, PL_scopestack_max, I32);
15685 Copy(proto_perl->Iscopestack, PL_scopestack, PL_scopestack_ix, I32);
15688 Newx(PL_scopestack_name, PL_scopestack_max, const char *);
15689 Copy(proto_perl->Iscopestack_name, PL_scopestack_name, PL_scopestack_ix, const char *);
15691 /* reset stack AV to correct length before its duped via
15692 * PL_curstackinfo */
15693 AvFILLp(proto_perl->Icurstack) =
15694 proto_perl->Istack_sp - proto_perl->Istack_base;
15696 /* NOTE: si_dup() looks at PL_markstack */
15697 PL_curstackinfo = si_dup(proto_perl->Icurstackinfo, param);
15699 /* PL_curstack = PL_curstackinfo->si_stack; */
15700 PL_curstack = av_dup(proto_perl->Icurstack, param);
15701 PL_mainstack = av_dup(proto_perl->Imainstack, param);
15703 /* next PUSHs() etc. set *(PL_stack_sp+1) */
15704 PL_stack_base = AvARRAY(PL_curstack);
15705 PL_stack_sp = PL_stack_base + (proto_perl->Istack_sp
15706 - proto_perl->Istack_base);
15707 PL_stack_max = PL_stack_base + AvMAX(PL_curstack);
15709 /*Newxz(PL_savestack, PL_savestack_max, ANY);*/
15710 PL_savestack = ss_dup(proto_perl, param);
15714 ENTER; /* perl_destruct() wants to LEAVE; */
15717 PL_statgv = gv_dup(proto_perl->Istatgv, param);
15718 PL_statname = sv_dup_inc(proto_perl->Istatname, param);
15720 PL_rs = sv_dup_inc(proto_perl->Irs, param);
15721 PL_last_in_gv = gv_dup(proto_perl->Ilast_in_gv, param);
15722 PL_defoutgv = gv_dup_inc(proto_perl->Idefoutgv, param);
15723 PL_toptarget = sv_dup_inc(proto_perl->Itoptarget, param);
15724 PL_bodytarget = sv_dup_inc(proto_perl->Ibodytarget, param);
15725 PL_formtarget = sv_dup(proto_perl->Iformtarget, param);
15727 PL_errors = sv_dup_inc(proto_perl->Ierrors, param);
15729 PL_sortcop = (OP*)any_dup(proto_perl->Isortcop, proto_perl);
15730 PL_firstgv = gv_dup_inc(proto_perl->Ifirstgv, param);
15731 PL_secondgv = gv_dup_inc(proto_perl->Isecondgv, param);
15733 PL_stashcache = newHV();
15735 PL_watchaddr = (char **) ptr_table_fetch(PL_ptr_table,
15736 proto_perl->Iwatchaddr);
15737 PL_watchok = PL_watchaddr ? * PL_watchaddr : NULL;
15738 if (PL_debug && PL_watchaddr) {
15739 PerlIO_printf(Perl_debug_log,
15740 "WATCHING: %" UVxf " cloned as %" UVxf " with value %" UVxf "\n",
15741 PTR2UV(proto_perl->Iwatchaddr), PTR2UV(PL_watchaddr),
15742 PTR2UV(PL_watchok));
15745 PL_registered_mros = hv_dup_inc(proto_perl->Iregistered_mros, param);
15746 PL_blockhooks = av_dup_inc(proto_perl->Iblockhooks, param);
15748 /* Call the ->CLONE method, if it exists, for each of the stashes
15749 identified by sv_dup() above.
15751 while(av_tindex(param->stashes) != -1) {
15752 HV* const stash = MUTABLE_HV(av_shift(param->stashes));
15753 GV* const cloner = gv_fetchmethod_autoload(stash, "CLONE", 0);
15754 if (cloner && GvCV(cloner)) {
15759 mXPUSHs(newSVhek(HvNAME_HEK(stash)));
15761 call_sv(MUTABLE_SV(GvCV(cloner)), G_DISCARD);
15767 if (!(flags & CLONEf_KEEP_PTR_TABLE)) {
15768 ptr_table_free(PL_ptr_table);
15769 PL_ptr_table = NULL;
15772 if (!(flags & CLONEf_COPY_STACKS)) {
15773 unreferenced_to_tmp_stack(param->unreferenced);
15776 SvREFCNT_dec(param->stashes);
15778 /* orphaned? eg threads->new inside BEGIN or use */
15779 if (PL_compcv && ! SvREFCNT(PL_compcv)) {
15780 SvREFCNT_inc_simple_void(PL_compcv);
15781 SAVEFREESV(PL_compcv);
15788 S_unreferenced_to_tmp_stack(pTHX_ AV *const unreferenced)
15790 PERL_ARGS_ASSERT_UNREFERENCED_TO_TMP_STACK;
15792 if (AvFILLp(unreferenced) > -1) {
15793 SV **svp = AvARRAY(unreferenced);
15794 SV **const last = svp + AvFILLp(unreferenced);
15798 if (SvREFCNT(*svp) == 1)
15800 } while (++svp <= last);
15802 EXTEND_MORTAL(count);
15803 svp = AvARRAY(unreferenced);
15806 if (SvREFCNT(*svp) == 1) {
15807 /* Our reference is the only one to this SV. This means that
15808 in this thread, the scalar effectively has a 0 reference.
15809 That doesn't work (cleanup never happens), so donate our
15810 reference to it onto the save stack. */
15811 PL_tmps_stack[++PL_tmps_ix] = *svp;
15813 /* As an optimisation, because we are already walking the
15814 entire array, instead of above doing either
15815 SvREFCNT_inc(*svp) or *svp = &PL_sv_undef, we can instead
15816 release our reference to the scalar, so that at the end of
15817 the array owns zero references to the scalars it happens to
15818 point to. We are effectively converting the array from
15819 AvREAL() on to AvREAL() off. This saves the av_clear()
15820 (triggered by the SvREFCNT_dec(unreferenced) below) from
15821 walking the array a second time. */
15822 SvREFCNT_dec(*svp);
15825 } while (++svp <= last);
15826 AvREAL_off(unreferenced);
15828 SvREFCNT_dec_NN(unreferenced);
15832 Perl_clone_params_del(CLONE_PARAMS *param)
15834 /* This seemingly funky ordering keeps the build with PERL_GLOBAL_STRUCT
15836 PerlInterpreter *const to = param->new_perl;
15838 PerlInterpreter *const was = PERL_GET_THX;
15840 PERL_ARGS_ASSERT_CLONE_PARAMS_DEL;
15846 SvREFCNT_dec(param->stashes);
15847 if (param->unreferenced)
15848 unreferenced_to_tmp_stack(param->unreferenced);
15858 Perl_clone_params_new(PerlInterpreter *const from, PerlInterpreter *const to)
15861 /* Need to play this game, as newAV() can call safesysmalloc(), and that
15862 does a dTHX; to get the context from thread local storage.
15863 FIXME - under PERL_CORE Newx(), Safefree() and friends should expand to
15864 a version that passes in my_perl. */
15865 PerlInterpreter *const was = PERL_GET_THX;
15866 CLONE_PARAMS *param;
15868 PERL_ARGS_ASSERT_CLONE_PARAMS_NEW;
15874 /* Given that we've set the context, we can do this unshared. */
15875 Newx(param, 1, CLONE_PARAMS);
15878 param->proto_perl = from;
15879 param->new_perl = to;
15880 param->stashes = (AV *)Perl_newSV_type(to, SVt_PVAV);
15881 AvREAL_off(param->stashes);
15882 param->unreferenced = (AV *)Perl_newSV_type(to, SVt_PVAV);
15890 #endif /* USE_ITHREADS */
15893 Perl_init_constants(pTHX)
15897 SvREFCNT(&PL_sv_undef) = SvREFCNT_IMMORTAL;
15898 SvFLAGS(&PL_sv_undef) = SVf_READONLY|SVf_PROTECT|SVt_NULL;
15899 SvANY(&PL_sv_undef) = NULL;
15901 SvANY(&PL_sv_no) = new_XPVNV();
15902 SvREFCNT(&PL_sv_no) = SvREFCNT_IMMORTAL;
15903 SvFLAGS(&PL_sv_no) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15904 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15907 SvANY(&PL_sv_yes) = new_XPVNV();
15908 SvREFCNT(&PL_sv_yes) = SvREFCNT_IMMORTAL;
15909 SvFLAGS(&PL_sv_yes) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15910 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15913 SvANY(&PL_sv_zero) = new_XPVNV();
15914 SvREFCNT(&PL_sv_zero) = SvREFCNT_IMMORTAL;
15915 SvFLAGS(&PL_sv_zero) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15916 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15920 SvPV_set(&PL_sv_no, (char*)PL_No);
15921 SvCUR_set(&PL_sv_no, 0);
15922 SvLEN_set(&PL_sv_no, 0);
15923 SvIV_set(&PL_sv_no, 0);
15924 SvNV_set(&PL_sv_no, 0);
15926 SvPV_set(&PL_sv_yes, (char*)PL_Yes);
15927 SvCUR_set(&PL_sv_yes, 1);
15928 SvLEN_set(&PL_sv_yes, 0);
15929 SvIV_set(&PL_sv_yes, 1);
15930 SvNV_set(&PL_sv_yes, 1);
15932 SvPV_set(&PL_sv_zero, (char*)PL_Zero);
15933 SvCUR_set(&PL_sv_zero, 1);
15934 SvLEN_set(&PL_sv_zero, 0);
15935 SvIV_set(&PL_sv_zero, 0);
15936 SvNV_set(&PL_sv_zero, 0);
15938 PadnamePV(&PL_padname_const) = (char *)PL_No;
15940 assert(SvIMMORTAL_INTERP(&PL_sv_yes));
15941 assert(SvIMMORTAL_INTERP(&PL_sv_undef));
15942 assert(SvIMMORTAL_INTERP(&PL_sv_no));
15943 assert(SvIMMORTAL_INTERP(&PL_sv_zero));
15945 assert(SvIMMORTAL(&PL_sv_yes));
15946 assert(SvIMMORTAL(&PL_sv_undef));
15947 assert(SvIMMORTAL(&PL_sv_no));
15948 assert(SvIMMORTAL(&PL_sv_zero));
15950 assert( SvIMMORTAL_TRUE(&PL_sv_yes));
15951 assert(!SvIMMORTAL_TRUE(&PL_sv_undef));
15952 assert(!SvIMMORTAL_TRUE(&PL_sv_no));
15953 assert(!SvIMMORTAL_TRUE(&PL_sv_zero));
15955 assert( SvTRUE_nomg_NN(&PL_sv_yes));
15956 assert(!SvTRUE_nomg_NN(&PL_sv_undef));
15957 assert(!SvTRUE_nomg_NN(&PL_sv_no));
15958 assert(!SvTRUE_nomg_NN(&PL_sv_zero));
15962 =head1 Unicode Support
15964 =for apidoc sv_recode_to_utf8
15966 C<encoding> is assumed to be an C<Encode> object, on entry the PV
15967 of C<sv> is assumed to be octets in that encoding, and C<sv>
15968 will be converted into Unicode (and UTF-8).
15970 If C<sv> already is UTF-8 (or if it is not C<POK>), or if C<encoding>
15971 is not a reference, nothing is done to C<sv>. If C<encoding> is not
15972 an C<Encode::XS> Encoding object, bad things will happen.
15973 (See F<cpan/Encode/encoding.pm> and L<Encode>.)
15975 The PV of C<sv> is returned.
15980 Perl_sv_recode_to_utf8(pTHX_ SV *sv, SV *encoding)
15982 PERL_ARGS_ASSERT_SV_RECODE_TO_UTF8;
15984 if (SvPOK(sv) && !SvUTF8(sv) && !IN_BYTES && SvROK(encoding)) {
15993 if (SvPADTMP(nsv)) {
15994 nsv = sv_newmortal();
15995 SvSetSV_nosteal(nsv, sv);
16004 Passing sv_yes is wrong - it needs to be or'ed set of constants
16005 for Encode::XS, while UTf-8 decode (currently) assumes a true value means
16006 remove converted chars from source.
16008 Both will default the value - let them.
16010 XPUSHs(&PL_sv_yes);
16013 call_method("decode", G_SCALAR);
16017 s = SvPV_const(uni, len);
16018 if (s != SvPVX_const(sv)) {
16019 SvGROW(sv, len + 1);
16020 Move(s, SvPVX(sv), len + 1, char);
16021 SvCUR_set(sv, len);
16026 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
16027 /* clear pos and any utf8 cache */
16028 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
16031 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
16032 magic_setutf8(sv,mg); /* clear UTF8 cache */
16037 return SvPOKp(sv) ? SvPVX(sv) : NULL;
16041 =for apidoc sv_cat_decode
16043 C<encoding> is assumed to be an C<Encode> object, the PV of C<ssv> is
16044 assumed to be octets in that encoding and decoding the input starts
16045 from the position which S<C<(PV + *offset)>> pointed to. C<dsv> will be
16046 concatenated with the decoded UTF-8 string from C<ssv>. Decoding will terminate
16047 when the string C<tstr> appears in decoding output or the input ends on
16048 the PV of C<ssv>. The value which C<offset> points will be modified
16049 to the last input position on C<ssv>.
16051 Returns TRUE if the terminator was found, else returns FALSE.
16056 Perl_sv_cat_decode(pTHX_ SV *dsv, SV *encoding,
16057 SV *ssv, int *offset, char *tstr, int tlen)
16061 PERL_ARGS_ASSERT_SV_CAT_DECODE;
16063 if (SvPOK(ssv) && SvPOK(dsv) && SvROK(encoding)) {
16074 offsv = newSViv(*offset);
16076 mPUSHp(tstr, tlen);
16078 call_method("cat_decode", G_SCALAR);
16080 ret = SvTRUE(TOPs);
16081 *offset = SvIV(offsv);
16087 Perl_croak(aTHX_ "Invalid argument to sv_cat_decode");
16092 /* ---------------------------------------------------------------------
16094 * support functions for report_uninit()
16097 /* the maxiumum size of array or hash where we will scan looking
16098 * for the undefined element that triggered the warning */
16100 #define FUV_MAX_SEARCH_SIZE 1000
16102 /* Look for an entry in the hash whose value has the same SV as val;
16103 * If so, return a mortal copy of the key. */
16106 S_find_hash_subscript(pTHX_ const HV *const hv, const SV *const val)
16112 PERL_ARGS_ASSERT_FIND_HASH_SUBSCRIPT;
16114 if (!hv || SvMAGICAL(hv) || !HvARRAY(hv) ||
16115 (HvTOTALKEYS(hv) > FUV_MAX_SEARCH_SIZE))
16118 array = HvARRAY(hv);
16120 for (i=HvMAX(hv); i>=0; i--) {
16122 for (entry = array[i]; entry; entry = HeNEXT(entry)) {
16123 if (HeVAL(entry) != val)
16125 if ( HeVAL(entry) == &PL_sv_undef ||
16126 HeVAL(entry) == &PL_sv_placeholder)
16130 if (HeKLEN(entry) == HEf_SVKEY)
16131 return sv_mortalcopy(HeKEY_sv(entry));
16132 return sv_2mortal(newSVhek(HeKEY_hek(entry)));
16138 /* Look for an entry in the array whose value has the same SV as val;
16139 * If so, return the index, otherwise return -1. */
16142 S_find_array_subscript(pTHX_ const AV *const av, const SV *const val)
16144 PERL_ARGS_ASSERT_FIND_ARRAY_SUBSCRIPT;
16146 if (!av || SvMAGICAL(av) || !AvARRAY(av) ||
16147 (AvFILLp(av) > FUV_MAX_SEARCH_SIZE))
16150 if (val != &PL_sv_undef) {
16151 SV ** const svp = AvARRAY(av);
16154 for (i=AvFILLp(av); i>=0; i--)
16161 /* varname(): return the name of a variable, optionally with a subscript.
16162 * If gv is non-zero, use the name of that global, along with gvtype (one
16163 * of "$", "@", "%"); otherwise use the name of the lexical at pad offset
16164 * targ. Depending on the value of the subscript_type flag, return:
16167 #define FUV_SUBSCRIPT_NONE 1 /* "@foo" */
16168 #define FUV_SUBSCRIPT_ARRAY 2 /* "$foo[aindex]" */
16169 #define FUV_SUBSCRIPT_HASH 3 /* "$foo{keyname}" */
16170 #define FUV_SUBSCRIPT_WITHIN 4 /* "within @foo" */
16173 Perl_varname(pTHX_ const GV *const gv, const char gvtype, PADOFFSET targ,
16174 const SV *const keyname, SSize_t aindex, int subscript_type)
16177 SV * const name = sv_newmortal();
16178 if (gv && isGV(gv)) {
16180 buffer[0] = gvtype;
16183 /* as gv_fullname4(), but add literal '^' for $^FOO names */
16185 gv_fullname4(name, gv, buffer, 0);
16187 if ((unsigned int)SvPVX(name)[1] <= 26) {
16189 buffer[1] = SvPVX(name)[1] + 'A' - 1;
16191 /* Swap the 1 unprintable control character for the 2 byte pretty
16192 version - ie substr($name, 1, 1) = $buffer; */
16193 sv_insert(name, 1, 1, buffer, 2);
16197 CV * const cv = gv ? ((CV *)gv) : find_runcv(NULL);
16200 assert(!cv || SvTYPE(cv) == SVt_PVCV || SvTYPE(cv) == SVt_PVFM);
16202 if (!cv || !CvPADLIST(cv))
16204 sv = padnamelist_fetch(PadlistNAMES(CvPADLIST(cv)), targ);
16205 sv_setpvn(name, PadnamePV(sv), PadnameLEN(sv));
16209 if (subscript_type == FUV_SUBSCRIPT_HASH) {
16210 SV * const sv = newSV(0);
16212 const char * const pv = SvPV_nomg_const((SV*)keyname, len);
16214 *SvPVX(name) = '$';
16215 Perl_sv_catpvf(aTHX_ name, "{%s}",
16216 pv_pretty(sv, pv, len, 32, NULL, NULL,
16217 PERL_PV_PRETTY_DUMP | PERL_PV_ESCAPE_UNI_DETECT ));
16218 SvREFCNT_dec_NN(sv);
16220 else if (subscript_type == FUV_SUBSCRIPT_ARRAY) {
16221 *SvPVX(name) = '$';
16222 Perl_sv_catpvf(aTHX_ name, "[%" IVdf "]", (IV)aindex);
16224 else if (subscript_type == FUV_SUBSCRIPT_WITHIN) {
16225 /* We know that name has no magic, so can use 0 instead of SV_GMAGIC */
16226 Perl_sv_insert_flags(aTHX_ name, 0, 0, STR_WITH_LEN("within "), 0);
16234 =for apidoc find_uninit_var
16236 Find the name of the undefined variable (if any) that caused the operator
16237 to issue a "Use of uninitialized value" warning.
16238 If match is true, only return a name if its value matches C<uninit_sv>.
16239 So roughly speaking, if a unary operator (such as C<OP_COS>) generates a
16240 warning, then following the direct child of the op may yield an
16241 C<OP_PADSV> or C<OP_GV> that gives the name of the undefined variable. On the
16242 other hand, with C<OP_ADD> there are two branches to follow, so we only print
16243 the variable name if we get an exact match.
16244 C<desc_p> points to a string pointer holding the description of the op.
16245 This may be updated if needed.
16247 The name is returned as a mortal SV.
16249 Assumes that C<PL_op> is the OP that originally triggered the error, and that
16250 C<PL_comppad>/C<PL_curpad> points to the currently executing pad.
16256 S_find_uninit_var(pTHX_ const OP *const obase, const SV *const uninit_sv,
16257 bool match, const char **desc_p)
16262 const OP *o, *o2, *kid;
16264 PERL_ARGS_ASSERT_FIND_UNINIT_VAR;
16266 if (!obase || (match && (!uninit_sv || uninit_sv == &PL_sv_undef ||
16267 uninit_sv == &PL_sv_placeholder)))
16270 switch (obase->op_type) {
16273 /* undef should care if its args are undef - any warnings
16274 * will be from tied/magic vars */
16282 const bool pad = ( obase->op_type == OP_PADAV
16283 || obase->op_type == OP_PADHV
16284 || obase->op_type == OP_PADRANGE
16287 const bool hash = ( obase->op_type == OP_PADHV
16288 || obase->op_type == OP_RV2HV
16289 || (obase->op_type == OP_PADRANGE
16290 && SvTYPE(PAD_SVl(obase->op_targ)) == SVt_PVHV)
16294 int subscript_type = FUV_SUBSCRIPT_WITHIN;
16296 if (pad) { /* @lex, %lex */
16297 sv = PAD_SVl(obase->op_targ);
16301 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
16302 /* @global, %global */
16303 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
16306 sv = hash ? MUTABLE_SV(GvHV(gv)): MUTABLE_SV(GvAV(gv));
16308 else if (obase == PL_op) /* @{expr}, %{expr} */
16309 return find_uninit_var(cUNOPx(obase)->op_first,
16310 uninit_sv, match, desc_p);
16311 else /* @{expr}, %{expr} as a sub-expression */
16315 /* attempt to find a match within the aggregate */
16317 keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16319 subscript_type = FUV_SUBSCRIPT_HASH;
16322 index = find_array_subscript((const AV *)sv, uninit_sv);
16324 subscript_type = FUV_SUBSCRIPT_ARRAY;
16327 if (match && subscript_type == FUV_SUBSCRIPT_WITHIN)
16330 return varname(gv, (char)(hash ? '%' : '@'), obase->op_targ,
16331 keysv, index, subscript_type);
16335 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
16337 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
16338 if (!gv || !GvSTASH(gv))
16340 if (match && (GvSV(gv) != uninit_sv))
16342 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
16345 return find_uninit_var(cUNOPx(obase)->op_first, uninit_sv, 1, desc_p);
16348 if (match && PAD_SVl(obase->op_targ) != uninit_sv)
16350 return varname(NULL, '$', obase->op_targ,
16351 NULL, 0, FUV_SUBSCRIPT_NONE);
16354 gv = cGVOPx_gv(obase);
16355 if (!gv || (match && GvSV(gv) != uninit_sv) || !GvSTASH(gv))
16357 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
16359 case OP_AELEMFAST_LEX:
16362 AV *av = MUTABLE_AV(PAD_SV(obase->op_targ));
16363 if (!av || SvRMAGICAL(av))
16365 svp = av_fetch(av, (I8)obase->op_private, FALSE);
16366 if (!svp || *svp != uninit_sv)
16369 return varname(NULL, '$', obase->op_targ,
16370 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
16373 gv = cGVOPx_gv(obase);
16378 AV *const av = GvAV(gv);
16379 if (!av || SvRMAGICAL(av))
16381 svp = av_fetch(av, (I8)obase->op_private, FALSE);
16382 if (!svp || *svp != uninit_sv)
16385 return varname(gv, '$', 0,
16386 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
16388 NOT_REACHED; /* NOTREACHED */
16391 o = cUNOPx(obase)->op_first;
16392 if (!o || o->op_type != OP_NULL ||
16393 ! (o->op_targ == OP_AELEM || o->op_targ == OP_HELEM))
16395 return find_uninit_var(cBINOPo->op_last, uninit_sv, match, desc_p);
16400 bool negate = FALSE;
16402 if (PL_op == obase)
16403 /* $a[uninit_expr] or $h{uninit_expr} */
16404 return find_uninit_var(cBINOPx(obase)->op_last,
16405 uninit_sv, match, desc_p);
16408 o = cBINOPx(obase)->op_first;
16409 kid = cBINOPx(obase)->op_last;
16411 /* get the av or hv, and optionally the gv */
16413 if (o->op_type == OP_PADAV || o->op_type == OP_PADHV) {
16414 sv = PAD_SV(o->op_targ);
16416 else if ((o->op_type == OP_RV2AV || o->op_type == OP_RV2HV)
16417 && cUNOPo->op_first->op_type == OP_GV)
16419 gv = cGVOPx_gv(cUNOPo->op_first);
16423 == OP_RV2HV ? MUTABLE_SV(GvHV(gv)) : MUTABLE_SV(GvAV(gv));
16428 if (kid && kid->op_type == OP_NEGATE) {
16430 kid = cUNOPx(kid)->op_first;
16433 if (kid && kid->op_type == OP_CONST && SvOK(cSVOPx_sv(kid))) {
16434 /* index is constant */
16437 kidsv = newSVpvs_flags("-", SVs_TEMP);
16438 sv_catsv(kidsv, cSVOPx_sv(kid));
16441 kidsv = cSVOPx_sv(kid);
16445 if (obase->op_type == OP_HELEM) {
16446 HE* he = hv_fetch_ent(MUTABLE_HV(sv), kidsv, 0, 0);
16447 if (!he || HeVAL(he) != uninit_sv)
16451 SV * const opsv = cSVOPx_sv(kid);
16452 const IV opsviv = SvIV(opsv);
16453 SV * const * const svp = av_fetch(MUTABLE_AV(sv),
16454 negate ? - opsviv : opsviv,
16456 if (!svp || *svp != uninit_sv)
16460 if (obase->op_type == OP_HELEM)
16461 return varname(gv, '%', o->op_targ,
16462 kidsv, 0, FUV_SUBSCRIPT_HASH);
16464 return varname(gv, '@', o->op_targ, NULL,
16465 negate ? - SvIV(cSVOPx_sv(kid)) : SvIV(cSVOPx_sv(kid)),
16466 FUV_SUBSCRIPT_ARRAY);
16469 /* index is an expression;
16470 * attempt to find a match within the aggregate */
16471 if (obase->op_type == OP_HELEM) {
16472 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16474 return varname(gv, '%', o->op_targ,
16475 keysv, 0, FUV_SUBSCRIPT_HASH);
16478 const SSize_t index
16479 = find_array_subscript((const AV *)sv, uninit_sv);
16481 return varname(gv, '@', o->op_targ,
16482 NULL, index, FUV_SUBSCRIPT_ARRAY);
16487 (char)((o->op_type == OP_PADAV || o->op_type == OP_RV2AV)
16489 o->op_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
16491 NOT_REACHED; /* NOTREACHED */
16494 case OP_MULTIDEREF: {
16495 /* If we were executing OP_MULTIDEREF when the undef warning
16496 * triggered, then it must be one of the index values within
16497 * that triggered it. If not, then the only possibility is that
16498 * the value retrieved by the last aggregate index might be the
16499 * culprit. For the former, we set PL_multideref_pc each time before
16500 * using an index, so work though the item list until we reach
16501 * that point. For the latter, just work through the entire item
16502 * list; the last aggregate retrieved will be the candidate.
16503 * There is a third rare possibility: something triggered
16504 * magic while fetching an array/hash element. Just display
16505 * nothing in this case.
16508 /* the named aggregate, if any */
16509 PADOFFSET agg_targ = 0;
16511 /* the last-seen index */
16513 PADOFFSET index_targ;
16515 IV index_const_iv = 0; /* init for spurious compiler warn */
16516 SV *index_const_sv;
16517 int depth = 0; /* how many array/hash lookups we've done */
16519 UNOP_AUX_item *items = cUNOP_AUXx(obase)->op_aux;
16520 UNOP_AUX_item *last = NULL;
16521 UV actions = items->uv;
16524 if (PL_op == obase) {
16525 last = PL_multideref_pc;
16526 assert(last >= items && last <= items + items[-1].uv);
16533 switch (actions & MDEREF_ACTION_MASK) {
16535 case MDEREF_reload:
16536 actions = (++items)->uv;
16539 case MDEREF_HV_padhv_helem: /* $lex{...} */
16542 case MDEREF_AV_padav_aelem: /* $lex[...] */
16543 agg_targ = (++items)->pad_offset;
16547 case MDEREF_HV_gvhv_helem: /* $pkg{...} */
16550 case MDEREF_AV_gvav_aelem: /* $pkg[...] */
16552 agg_gv = (GV*)UNOP_AUX_item_sv(++items);
16553 assert(isGV_with_GP(agg_gv));
16556 case MDEREF_HV_gvsv_vivify_rv2hv_helem: /* $pkg->{...} */
16557 case MDEREF_HV_padsv_vivify_rv2hv_helem: /* $lex->{...} */
16560 case MDEREF_HV_pop_rv2hv_helem: /* expr->{...} */
16561 case MDEREF_HV_vivify_rv2hv_helem: /* vivify, ->{...} */
16567 case MDEREF_AV_gvsv_vivify_rv2av_aelem: /* $pkg->[...] */
16568 case MDEREF_AV_padsv_vivify_rv2av_aelem: /* $lex->[...] */
16571 case MDEREF_AV_pop_rv2av_aelem: /* expr->[...] */
16572 case MDEREF_AV_vivify_rv2av_aelem: /* vivify, ->[...] */
16579 index_const_sv = NULL;
16581 index_type = (actions & MDEREF_INDEX_MASK);
16582 switch (index_type) {
16583 case MDEREF_INDEX_none:
16585 case MDEREF_INDEX_const:
16587 index_const_sv = UNOP_AUX_item_sv(++items)
16589 index_const_iv = (++items)->iv;
16591 case MDEREF_INDEX_padsv:
16592 index_targ = (++items)->pad_offset;
16594 case MDEREF_INDEX_gvsv:
16595 index_gv = (GV*)UNOP_AUX_item_sv(++items);
16596 assert(isGV_with_GP(index_gv));
16600 if (index_type != MDEREF_INDEX_none)
16603 if ( index_type == MDEREF_INDEX_none
16604 || (actions & MDEREF_FLAG_last)
16605 || (last && items >= last)
16609 actions >>= MDEREF_SHIFT;
16612 if (PL_op == obase) {
16613 /* most likely index was undef */
16615 *desc_p = ( (actions & MDEREF_FLAG_last)
16616 && (obase->op_private
16617 & (OPpMULTIDEREF_EXISTS|OPpMULTIDEREF_DELETE)))
16619 (obase->op_private & OPpMULTIDEREF_EXISTS)
16622 : is_hv ? "hash element" : "array element";
16623 assert(index_type != MDEREF_INDEX_none);
16625 if (GvSV(index_gv) == uninit_sv)
16626 return varname(index_gv, '$', 0, NULL, 0,
16627 FUV_SUBSCRIPT_NONE);
16632 if (PL_curpad[index_targ] == uninit_sv)
16633 return varname(NULL, '$', index_targ,
16634 NULL, 0, FUV_SUBSCRIPT_NONE);
16638 /* If we got to this point it was undef on a const subscript,
16639 * so magic probably involved, e.g. $ISA[0]. Give up. */
16643 /* the SV returned by pp_multideref() was undef, if anything was */
16649 sv = PAD_SV(agg_targ);
16651 sv = is_hv ? MUTABLE_SV(GvHV(agg_gv)) : MUTABLE_SV(GvAV(agg_gv));
16655 if (index_type == MDEREF_INDEX_const) {
16660 HE* he = hv_fetch_ent(MUTABLE_HV(sv), index_const_sv, 0, 0);
16661 if (!he || HeVAL(he) != uninit_sv)
16665 SV * const * const svp =
16666 av_fetch(MUTABLE_AV(sv), index_const_iv, FALSE);
16667 if (!svp || *svp != uninit_sv)
16672 ? varname(agg_gv, '%', agg_targ,
16673 index_const_sv, 0, FUV_SUBSCRIPT_HASH)
16674 : varname(agg_gv, '@', agg_targ,
16675 NULL, index_const_iv, FUV_SUBSCRIPT_ARRAY);
16678 /* index is an var */
16680 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16682 return varname(agg_gv, '%', agg_targ,
16683 keysv, 0, FUV_SUBSCRIPT_HASH);
16686 const SSize_t index
16687 = find_array_subscript((const AV *)sv, uninit_sv);
16689 return varname(agg_gv, '@', agg_targ,
16690 NULL, index, FUV_SUBSCRIPT_ARRAY);
16694 return varname(agg_gv,
16696 agg_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
16698 NOT_REACHED; /* NOTREACHED */
16702 /* only examine RHS */
16703 return find_uninit_var(cBINOPx(obase)->op_first, uninit_sv,
16707 o = cUNOPx(obase)->op_first;
16708 if ( o->op_type == OP_PUSHMARK
16709 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)
16713 if (!OpHAS_SIBLING(o)) {
16714 /* one-arg version of open is highly magical */
16716 if (o->op_type == OP_GV) { /* open FOO; */
16718 if (match && GvSV(gv) != uninit_sv)
16720 return varname(gv, '$', 0,
16721 NULL, 0, FUV_SUBSCRIPT_NONE);
16723 /* other possibilities not handled are:
16724 * open $x; or open my $x; should return '${*$x}'
16725 * open expr; should return '$'.expr ideally
16732 /* ops where $_ may be an implicit arg */
16737 if ( !(obase->op_flags & OPf_STACKED)) {
16738 if (uninit_sv == DEFSV)
16739 return newSVpvs_flags("$_", SVs_TEMP);
16740 else if (obase->op_targ
16741 && uninit_sv == PAD_SVl(obase->op_targ))
16742 return varname(NULL, '$', obase->op_targ, NULL, 0,
16743 FUV_SUBSCRIPT_NONE);
16750 match = 1; /* print etc can return undef on defined args */
16751 /* skip filehandle as it can't produce 'undef' warning */
16752 o = cUNOPx(obase)->op_first;
16753 if ((obase->op_flags & OPf_STACKED)
16755 ( o->op_type == OP_PUSHMARK
16756 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)))
16757 o = OpSIBLING(OpSIBLING(o));
16761 case OP_ENTEREVAL: /* could be eval $undef or $x='$undef'; eval $x */
16762 case OP_CUSTOM: /* XS or custom code could trigger random warnings */
16764 /* the following ops are capable of returning PL_sv_undef even for
16765 * defined arg(s) */
16784 case OP_GETPEERNAME:
16831 case OP_SMARTMATCH:
16840 /* XXX tmp hack: these two may call an XS sub, and currently
16841 XS subs don't have a SUB entry on the context stack, so CV and
16842 pad determination goes wrong, and BAD things happen. So, just
16843 don't try to determine the value under those circumstances.
16844 Need a better fix at dome point. DAPM 11/2007 */
16850 GV * const gv = gv_fetchpvs(".", GV_NOTQUAL, SVt_PV);
16851 if (gv && GvSV(gv) == uninit_sv)
16852 return newSVpvs_flags("$.", SVs_TEMP);
16857 /* def-ness of rval pos() is independent of the def-ness of its arg */
16858 if ( !(obase->op_flags & OPf_MOD))
16864 if (SvROK(PL_rs) && uninit_sv == SvRV(PL_rs))
16865 return newSVpvs_flags("${$/}", SVs_TEMP);
16870 if (!(obase->op_flags & OPf_KIDS))
16872 o = cUNOPx(obase)->op_first;
16878 /* This loop checks all the kid ops, skipping any that cannot pos-
16879 * sibly be responsible for the uninitialized value; i.e., defined
16880 * constants and ops that return nothing. If there is only one op
16881 * left that is not skipped, then we *know* it is responsible for
16882 * the uninitialized value. If there is more than one op left, we
16883 * have to look for an exact match in the while() loop below.
16884 * Note that we skip padrange, because the individual pad ops that
16885 * it replaced are still in the tree, so we work on them instead.
16888 for (kid=o; kid; kid = OpSIBLING(kid)) {
16889 const OPCODE type = kid->op_type;
16890 if ( (type == OP_CONST && SvOK(cSVOPx_sv(kid)))
16891 || (type == OP_NULL && ! (kid->op_flags & OPf_KIDS))
16892 || (type == OP_PUSHMARK)
16893 || (type == OP_PADRANGE)
16897 if (o2) { /* more than one found */
16904 return find_uninit_var(o2, uninit_sv, match, desc_p);
16906 /* scan all args */
16908 sv = find_uninit_var(o, uninit_sv, 1, desc_p);
16920 =for apidoc report_uninit
16922 Print appropriate "Use of uninitialized variable" warning.
16928 Perl_report_uninit(pTHX_ const SV *uninit_sv)
16930 const char *desc = NULL;
16931 SV* varname = NULL;
16934 desc = PL_op->op_type == OP_STRINGIFY && PL_op->op_folded
16936 : PL_op->op_type == OP_MULTICONCAT
16937 && (PL_op->op_private & OPpMULTICONCAT_FAKE)
16940 if (uninit_sv && PL_curpad) {
16941 varname = find_uninit_var(PL_op, uninit_sv, 0, &desc);
16943 sv_insert(varname, 0, 0, " ", 1);
16946 else if (PL_curstackinfo->si_type == PERLSI_SORT && cxstack_ix == 0)
16947 /* we've reached the end of a sort block or sub,
16948 * and the uninit value is probably what that code returned */
16951 /* PL_warn_uninit_sv is constant */
16952 GCC_DIAG_IGNORE_STMT(-Wformat-nonliteral);
16954 /* diag_listed_as: Use of uninitialized value%s */
16955 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit_sv,
16956 SVfARG(varname ? varname : &PL_sv_no),
16959 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit,
16961 GCC_DIAG_RESTORE_STMT;
16965 * ex: set ts=8 sts=4 sw=4 et: