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.
134 An SV (or AV, HV, etc.) is allocated in two parts: the head (struct
135 sv, av, hv...) contains type and reference count information, and for
136 many types, a pointer to the body (struct xrv, xpv, xpviv...), which
137 contains fields specific to each type. Some types store all they need
138 in the head, so don't have a body.
140 In all but the most memory-paranoid configurations (ex: PURIFY), heads
141 and bodies are allocated out of arenas, which by default are
142 approximately 4K chunks of memory parcelled up into N heads or bodies.
143 Sv-bodies are allocated by their sv-type, guaranteeing size
144 consistency needed to allocate safely from arrays.
146 For SV-heads, the first slot in each arena is reserved, and holds a
147 link to the next arena, some flags, and a note of the number of slots.
148 Snaked through each arena chain is a linked list of free items; when
149 this becomes empty, an extra arena is allocated and divided up into N
150 items which are threaded into the free list.
152 SV-bodies are similar, but they use arena-sets by default, which
153 separate the link and info from the arena itself, and reclaim the 1st
154 slot in the arena. SV-bodies are further described later.
156 The following global variables are associated with arenas:
158 PL_sv_arenaroot pointer to list of SV arenas
159 PL_sv_root pointer to list of free SV structures
161 PL_body_arenas head of linked-list of body arenas
162 PL_body_roots[] array of pointers to list of free bodies of svtype
163 arrays are indexed by the svtype needed
165 A few special SV heads are not allocated from an arena, but are
166 instead directly created in the interpreter structure, eg PL_sv_undef.
167 The size of arenas can be changed from the default by setting
168 PERL_ARENA_SIZE appropriately at compile time.
170 The SV arena serves the secondary purpose of allowing still-live SVs
171 to be located and destroyed during final cleanup.
173 At the lowest level, the macros new_SV() and del_SV() grab and free
174 an SV head. (If debugging with -DD, del_SV() calls the function S_del_sv()
175 to return the SV to the free list with error checking.) new_SV() calls
176 more_sv() / sv_add_arena() to add an extra arena if the free list is empty.
177 SVs in the free list have their SvTYPE field set to all ones.
179 At the time of very final cleanup, sv_free_arenas() is called from
180 perl_destruct() to physically free all the arenas allocated since the
181 start of the interpreter.
183 The function visit() scans the SV arenas list, and calls a specified
184 function for each SV it finds which is still live - ie which has an SvTYPE
185 other than all 1's, and a non-zero SvREFCNT. visit() is used by the
186 following functions (specified as [function that calls visit()] / [function
187 called by visit() for each SV]):
189 sv_report_used() / do_report_used()
190 dump all remaining SVs (debugging aid)
192 sv_clean_objs() / do_clean_objs(),do_clean_named_objs(),
193 do_clean_named_io_objs(),do_curse()
194 Attempt to free all objects pointed to by RVs,
195 try to do the same for all objects indir-
196 ectly referenced by typeglobs too, and
197 then do a final sweep, cursing any
198 objects that remain. Called once from
199 perl_destruct(), prior to calling sv_clean_all()
202 sv_clean_all() / do_clean_all()
203 SvREFCNT_dec(sv) each remaining SV, possibly
204 triggering an sv_free(). It also sets the
205 SVf_BREAK flag on the SV to indicate that the
206 refcnt has been artificially lowered, and thus
207 stopping sv_free() from giving spurious warnings
208 about SVs which unexpectedly have a refcnt
209 of zero. called repeatedly from perl_destruct()
210 until there are no SVs left.
212 =head2 Arena allocator API Summary
214 Private API to rest of sv.c
218 new_XPVNV(), del_XPVGV(),
223 sv_report_used(), sv_clean_objs(), sv_clean_all(), sv_free_arenas()
227 * ========================================================================= */
230 * "A time to plant, and a time to uproot what was planted..."
234 # define MEM_LOG_NEW_SV(sv, file, line, func) \
235 Perl_mem_log_new_sv(sv, file, line, func)
236 # define MEM_LOG_DEL_SV(sv, file, line, func) \
237 Perl_mem_log_del_sv(sv, file, line, func)
239 # define MEM_LOG_NEW_SV(sv, file, line, func) NOOP
240 # define MEM_LOG_DEL_SV(sv, file, line, func) NOOP
243 #ifdef DEBUG_LEAKING_SCALARS
244 # define FREE_SV_DEBUG_FILE(sv) STMT_START { \
245 if ((sv)->sv_debug_file) PerlMemShared_free((sv)->sv_debug_file); \
247 # define DEBUG_SV_SERIAL(sv) \
248 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%" UVxf ": (%05ld) del_SV\n", \
249 PTR2UV(sv), (long)(sv)->sv_debug_serial))
251 # define FREE_SV_DEBUG_FILE(sv)
252 # define DEBUG_SV_SERIAL(sv) NOOP
256 # define SvARENA_CHAIN(sv) ((sv)->sv_u.svu_rv)
257 # define SvARENA_CHAIN_SET(sv,val) (sv)->sv_u.svu_rv = MUTABLE_SV((val))
258 /* Whilst I'd love to do this, it seems that things like to check on
260 # define POISON_SV_HEAD(sv) PoisonNew(sv, 1, struct STRUCT_SV)
262 # define POISON_SV_HEAD(sv) PoisonNew(&SvANY(sv), 1, void *), \
263 PoisonNew(&SvREFCNT(sv), 1, U32)
265 # define SvARENA_CHAIN(sv) SvANY(sv)
266 # define SvARENA_CHAIN_SET(sv,val) SvANY(sv) = (void *)(val)
267 # define POISON_SV_HEAD(sv)
270 /* Mark an SV head as unused, and add to free list.
272 * If SVf_BREAK is set, skip adding it to the free list, as this SV had
273 * its refcount artificially decremented during global destruction, so
274 * there may be dangling pointers to it. The last thing we want in that
275 * case is for it to be reused. */
277 #define plant_SV(p) \
279 const U32 old_flags = SvFLAGS(p); \
280 MEM_LOG_DEL_SV(p, __FILE__, __LINE__, FUNCTION__); \
281 DEBUG_SV_SERIAL(p); \
282 FREE_SV_DEBUG_FILE(p); \
284 SvFLAGS(p) = SVTYPEMASK; \
285 if (!(old_flags & SVf_BREAK)) { \
286 SvARENA_CHAIN_SET(p, PL_sv_root); \
292 #define uproot_SV(p) \
295 PL_sv_root = MUTABLE_SV(SvARENA_CHAIN(p)); \
300 /* make some more SVs by adding another arena */
306 char *chunk; /* must use New here to match call to */
307 Newx(chunk,PERL_ARENA_SIZE,char); /* Safefree() in sv_free_arenas() */
308 sv_add_arena(chunk, PERL_ARENA_SIZE, 0);
313 /* new_SV(): return a new, empty SV head */
315 #ifdef DEBUG_LEAKING_SCALARS
316 /* provide a real function for a debugger to play with */
318 S_new_SV(pTHX_ const char *file, int line, const char *func)
325 sv = S_more_sv(aTHX);
329 sv->sv_debug_optype = PL_op ? PL_op->op_type : 0;
330 sv->sv_debug_line = (U16) (PL_parser && PL_parser->copline != NOLINE
336 sv->sv_debug_inpad = 0;
337 sv->sv_debug_parent = NULL;
338 sv->sv_debug_file = PL_curcop ? savesharedpv(CopFILE(PL_curcop)): NULL;
340 sv->sv_debug_serial = PL_sv_serial++;
342 MEM_LOG_NEW_SV(sv, file, line, func);
343 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%" UVxf ": (%05ld) new_SV (from %s:%d [%s])\n",
344 PTR2UV(sv), (long)sv->sv_debug_serial, file, line, func));
348 # define new_SV(p) (p)=S_new_SV(aTHX_ __FILE__, __LINE__, FUNCTION__)
356 (p) = S_more_sv(aTHX); \
360 MEM_LOG_NEW_SV(p, __FILE__, __LINE__, FUNCTION__); \
365 /* del_SV(): return an empty SV head to the free list */
378 S_del_sv(pTHX_ SV *p)
380 PERL_ARGS_ASSERT_DEL_SV;
385 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
386 const SV * const sv = sva + 1;
387 const SV * const svend = &sva[SvREFCNT(sva)];
388 if (p >= sv && p < svend) {
394 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
395 "Attempt to free non-arena SV: 0x%" UVxf
396 pTHX__FORMAT, PTR2UV(p) pTHX__VALUE);
403 #else /* ! DEBUGGING */
405 #define del_SV(p) plant_SV(p)
407 #endif /* DEBUGGING */
411 =head1 SV Manipulation Functions
413 =for apidoc sv_add_arena
415 Given a chunk of memory, link it to the head of the list of arenas,
416 and split it into a list of free SVs.
422 S_sv_add_arena(pTHX_ char *const ptr, const U32 size, const U32 flags)
424 SV *const sva = MUTABLE_SV(ptr);
428 PERL_ARGS_ASSERT_SV_ADD_ARENA;
430 /* The first SV in an arena isn't an SV. */
431 SvANY(sva) = (void *) PL_sv_arenaroot; /* ptr to next arena */
432 SvREFCNT(sva) = size / sizeof(SV); /* number of SV slots */
433 SvFLAGS(sva) = flags; /* FAKE if not to be freed */
435 PL_sv_arenaroot = sva;
436 PL_sv_root = sva + 1;
438 svend = &sva[SvREFCNT(sva) - 1];
441 SvARENA_CHAIN_SET(sv, (sv + 1));
445 /* Must always set typemask because it's always checked in on cleanup
446 when the arenas are walked looking for objects. */
447 SvFLAGS(sv) = SVTYPEMASK;
450 SvARENA_CHAIN_SET(sv, 0);
454 SvFLAGS(sv) = SVTYPEMASK;
457 /* visit(): call the named function for each non-free SV in the arenas
458 * whose flags field matches the flags/mask args. */
461 S_visit(pTHX_ SVFUNC_t f, const U32 flags, const U32 mask)
466 PERL_ARGS_ASSERT_VISIT;
468 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
469 const SV * const svend = &sva[SvREFCNT(sva)];
471 for (sv = sva + 1; sv < svend; ++sv) {
472 if (SvTYPE(sv) != (svtype)SVTYPEMASK
473 && (sv->sv_flags & mask) == flags
486 /* called by sv_report_used() for each live SV */
489 do_report_used(pTHX_ SV *const sv)
491 if (SvTYPE(sv) != (svtype)SVTYPEMASK) {
492 PerlIO_printf(Perl_debug_log, "****\n");
499 =for apidoc sv_report_used
501 Dump the contents of all SVs not yet freed (debugging aid).
507 Perl_sv_report_used(pTHX)
510 visit(do_report_used, 0, 0);
516 /* called by sv_clean_objs() for each live SV */
519 do_clean_objs(pTHX_ SV *const ref)
523 SV * const target = SvRV(ref);
524 if (SvOBJECT(target)) {
525 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning object ref:\n "), sv_dump(ref)));
526 if (SvWEAKREF(ref)) {
527 sv_del_backref(target, ref);
533 SvREFCNT_dec_NN(target);
540 /* clear any slots in a GV which hold objects - except IO;
541 * called by sv_clean_objs() for each live GV */
544 do_clean_named_objs(pTHX_ SV *const sv)
547 assert(SvTYPE(sv) == SVt_PVGV);
548 assert(isGV_with_GP(sv));
552 /* freeing GP entries may indirectly free the current GV;
553 * hold onto it while we mess with the GP slots */
556 if ( ((obj = GvSV(sv) )) && SvOBJECT(obj)) {
557 DEBUG_D((PerlIO_printf(Perl_debug_log,
558 "Cleaning named glob SV object:\n "), sv_dump(obj)));
560 SvREFCNT_dec_NN(obj);
562 if ( ((obj = MUTABLE_SV(GvAV(sv)) )) && SvOBJECT(obj)) {
563 DEBUG_D((PerlIO_printf(Perl_debug_log,
564 "Cleaning named glob AV object:\n "), sv_dump(obj)));
566 SvREFCNT_dec_NN(obj);
568 if ( ((obj = MUTABLE_SV(GvHV(sv)) )) && SvOBJECT(obj)) {
569 DEBUG_D((PerlIO_printf(Perl_debug_log,
570 "Cleaning named glob HV object:\n "), sv_dump(obj)));
572 SvREFCNT_dec_NN(obj);
574 if ( ((obj = MUTABLE_SV(GvCV(sv)) )) && SvOBJECT(obj)) {
575 DEBUG_D((PerlIO_printf(Perl_debug_log,
576 "Cleaning named glob CV object:\n "), sv_dump(obj)));
578 SvREFCNT_dec_NN(obj);
580 SvREFCNT_dec_NN(sv); /* undo the inc above */
583 /* clear any IO slots in a GV which hold objects (except stderr, defout);
584 * called by sv_clean_objs() for each live GV */
587 do_clean_named_io_objs(pTHX_ SV *const sv)
590 assert(SvTYPE(sv) == SVt_PVGV);
591 assert(isGV_with_GP(sv));
592 if (!GvGP(sv) || sv == (SV*)PL_stderrgv || sv == (SV*)PL_defoutgv)
596 if ( ((obj = MUTABLE_SV(GvIO(sv)) )) && SvOBJECT(obj)) {
597 DEBUG_D((PerlIO_printf(Perl_debug_log,
598 "Cleaning named glob IO object:\n "), sv_dump(obj)));
600 SvREFCNT_dec_NN(obj);
602 SvREFCNT_dec_NN(sv); /* undo the inc above */
605 /* Void wrapper to pass to visit() */
607 do_curse(pTHX_ SV * const sv) {
608 if ((PL_stderrgv && GvGP(PL_stderrgv) && (SV*)GvIO(PL_stderrgv) == sv)
609 || (PL_defoutgv && GvGP(PL_defoutgv) && (SV*)GvIO(PL_defoutgv) == sv))
615 =for apidoc sv_clean_objs
617 Attempt to destroy all objects not yet freed.
623 Perl_sv_clean_objs(pTHX)
626 PL_in_clean_objs = TRUE;
627 visit(do_clean_objs, SVf_ROK, SVf_ROK);
628 /* Some barnacles may yet remain, clinging to typeglobs.
629 * Run the non-IO destructors first: they may want to output
630 * error messages, close files etc */
631 visit(do_clean_named_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
632 visit(do_clean_named_io_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
633 /* And if there are some very tenacious barnacles clinging to arrays,
634 closures, or what have you.... */
635 visit(do_curse, SVs_OBJECT, SVs_OBJECT);
636 olddef = PL_defoutgv;
637 PL_defoutgv = NULL; /* disable skip of PL_defoutgv */
638 if (olddef && isGV_with_GP(olddef))
639 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olddef));
640 olderr = PL_stderrgv;
641 PL_stderrgv = NULL; /* disable skip of PL_stderrgv */
642 if (olderr && isGV_with_GP(olderr))
643 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olderr));
644 SvREFCNT_dec(olddef);
645 PL_in_clean_objs = FALSE;
648 /* called by sv_clean_all() for each live SV */
651 do_clean_all(pTHX_ SV *const sv)
653 if (sv == (const SV *) PL_fdpid || sv == (const SV *)PL_strtab) {
654 /* don't clean pid table and strtab */
657 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning loops: SV at 0x%" UVxf "\n", PTR2UV(sv)) ));
658 SvFLAGS(sv) |= SVf_BREAK;
663 =for apidoc sv_clean_all
665 Decrement the refcnt of each remaining SV, possibly triggering a
666 cleanup. This function may have to be called multiple times to free
667 SVs which are in complex self-referential hierarchies.
673 Perl_sv_clean_all(pTHX)
676 PL_in_clean_all = TRUE;
677 cleaned = visit(do_clean_all, 0,0);
682 ARENASETS: a meta-arena implementation which separates arena-info
683 into struct arena_set, which contains an array of struct
684 arena_descs, each holding info for a single arena. By separating
685 the meta-info from the arena, we recover the 1st slot, formerly
686 borrowed for list management. The arena_set is about the size of an
687 arena, avoiding the needless malloc overhead of a naive linked-list.
689 The cost is 1 arena-set malloc per ~320 arena-mallocs, + the unused
690 memory in the last arena-set (1/2 on average). In trade, we get
691 back the 1st slot in each arena (ie 1.7% of a CV-arena, less for
692 smaller types). The recovery of the wasted space allows use of
693 small arenas for large, rare body types, by changing array* fields
694 in body_details_by_type[] below.
697 char *arena; /* the raw storage, allocated aligned */
698 size_t size; /* its size ~4k typ */
699 svtype utype; /* bodytype stored in arena */
704 /* Get the maximum number of elements in set[] such that struct arena_set
705 will fit within PERL_ARENA_SIZE, which is probably just under 4K, and
706 therefore likely to be 1 aligned memory page. */
708 #define ARENAS_PER_SET ((PERL_ARENA_SIZE - sizeof(struct arena_set*) \
709 - 2 * sizeof(int)) / sizeof (struct arena_desc))
712 struct arena_set* next;
713 unsigned int set_size; /* ie ARENAS_PER_SET */
714 unsigned int curr; /* index of next available arena-desc */
715 struct arena_desc set[ARENAS_PER_SET];
719 =for apidoc sv_free_arenas
721 Deallocate the memory used by all arenas. Note that all the individual SV
722 heads and bodies within the arenas must already have been freed.
728 Perl_sv_free_arenas(pTHX)
734 /* Free arenas here, but be careful about fake ones. (We assume
735 contiguity of the fake ones with the corresponding real ones.) */
737 for (sva = PL_sv_arenaroot; sva; sva = svanext) {
738 svanext = MUTABLE_SV(SvANY(sva));
739 while (svanext && SvFAKE(svanext))
740 svanext = MUTABLE_SV(SvANY(svanext));
747 struct arena_set *aroot = (struct arena_set*) PL_body_arenas;
750 struct arena_set *current = aroot;
753 assert(aroot->set[i].arena);
754 Safefree(aroot->set[i].arena);
762 i = PERL_ARENA_ROOTS_SIZE;
764 PL_body_roots[i] = 0;
771 Here are mid-level routines that manage the allocation of bodies out
772 of the various arenas. There are 5 kinds of arenas:
774 1. SV-head arenas, which are discussed and handled above
775 2. regular body arenas
776 3. arenas for reduced-size bodies
779 Arena types 2 & 3 are chained by body-type off an array of
780 arena-root pointers, which is indexed by svtype. Some of the
781 larger/less used body types are malloced singly, since a large
782 unused block of them is wasteful. Also, several svtypes dont have
783 bodies; the data fits into the sv-head itself. The arena-root
784 pointer thus has a few unused root-pointers (which may be hijacked
785 later for arena types 4,5)
787 3 differs from 2 as an optimization; some body types have several
788 unused fields in the front of the structure (which are kept in-place
789 for consistency). These bodies can be allocated in smaller chunks,
790 because the leading fields arent accessed. Pointers to such bodies
791 are decremented to point at the unused 'ghost' memory, knowing that
792 the pointers are used with offsets to the real memory.
795 =head1 SV-Body Allocation
799 Allocation of SV-bodies is similar to SV-heads, differing as follows;
800 the allocation mechanism is used for many body types, so is somewhat
801 more complicated, it uses arena-sets, and has no need for still-live
804 At the outermost level, (new|del)_X*V macros return bodies of the
805 appropriate type. These macros call either (new|del)_body_type or
806 (new|del)_body_allocated macro pairs, depending on specifics of the
807 type. Most body types use the former pair, the latter pair is used to
808 allocate body types with "ghost fields".
810 "ghost fields" are fields that are unused in certain types, and
811 consequently don't need to actually exist. They are declared because
812 they're part of a "base type", which allows use of functions as
813 methods. The simplest examples are AVs and HVs, 2 aggregate types
814 which don't use the fields which support SCALAR semantics.
816 For these types, the arenas are carved up into appropriately sized
817 chunks, we thus avoid wasted memory for those unaccessed members.
818 When bodies are allocated, we adjust the pointer back in memory by the
819 size of the part not allocated, so it's as if we allocated the full
820 structure. (But things will all go boom if you write to the part that
821 is "not there", because you'll be overwriting the last members of the
822 preceding structure in memory.)
824 We calculate the correction using the STRUCT_OFFSET macro on the first
825 member present. If the allocated structure is smaller (no initial NV
826 actually allocated) then the net effect is to subtract the size of the NV
827 from the pointer, to return a new pointer as if an initial NV were actually
828 allocated. (We were using structures named *_allocated for this, but
829 this turned out to be a subtle bug, because a structure without an NV
830 could have a lower alignment constraint, but the compiler is allowed to
831 optimised accesses based on the alignment constraint of the actual pointer
832 to the full structure, for example, using a single 64 bit load instruction
833 because it "knows" that two adjacent 32 bit members will be 8-byte aligned.)
835 This is the same trick as was used for NV and IV bodies. Ironically it
836 doesn't need to be used for NV bodies any more, because NV is now at
837 the start of the structure. IV bodies, and also in some builds NV bodies,
838 don't need it either, because they are no longer allocated.
840 In turn, the new_body_* allocators call S_new_body(), which invokes
841 new_body_inline macro, which takes a lock, and takes a body off the
842 linked list at PL_body_roots[sv_type], calling Perl_more_bodies() if
843 necessary to refresh an empty list. Then the lock is released, and
844 the body is returned.
846 Perl_more_bodies allocates a new arena, and carves it up into an array of N
847 bodies, which it strings into a linked list. It looks up arena-size
848 and body-size from the body_details table described below, thus
849 supporting the multiple body-types.
851 If PURIFY is defined, or PERL_ARENA_SIZE=0, arenas are not used, and
852 the (new|del)_X*V macros are mapped directly to malloc/free.
854 For each sv-type, struct body_details bodies_by_type[] carries
855 parameters which control these aspects of SV handling:
857 Arena_size determines whether arenas are used for this body type, and if
858 so, how big they are. PURIFY or PERL_ARENA_SIZE=0 set this field to
859 zero, forcing individual mallocs and frees.
861 Body_size determines how big a body is, and therefore how many fit into
862 each arena. Offset carries the body-pointer adjustment needed for
863 "ghost fields", and is used in *_allocated macros.
865 But its main purpose is to parameterize info needed in
866 Perl_sv_upgrade(). The info here dramatically simplifies the function
867 vs the implementation in 5.8.8, making it table-driven. All fields
868 are used for this, except for arena_size.
870 For the sv-types that have no bodies, arenas are not used, so those
871 PL_body_roots[sv_type] are unused, and can be overloaded. In
872 something of a special case, SVt_NULL is borrowed for HE arenas;
873 PL_body_roots[HE_SVSLOT=SVt_NULL] is filled by S_more_he, but the
874 bodies_by_type[SVt_NULL] slot is not used, as the table is not
879 struct body_details {
880 U8 body_size; /* Size to allocate */
881 U8 copy; /* Size of structure to copy (may be shorter) */
882 U8 offset; /* Size of unalloced ghost fields to first alloced field*/
883 PERL_BITFIELD8 type : 4; /* We have space for a sanity check. */
884 PERL_BITFIELD8 cant_upgrade : 1;/* Cannot upgrade this type */
885 PERL_BITFIELD8 zero_nv : 1; /* zero the NV when upgrading from this */
886 PERL_BITFIELD8 arena : 1; /* Allocated from an arena */
887 U32 arena_size; /* Size of arena to allocate */
895 /* With -DPURFIY we allocate everything directly, and don't use arenas.
896 This seems a rather elegant way to simplify some of the code below. */
897 #define HASARENA FALSE
899 #define HASARENA TRUE
901 #define NOARENA FALSE
903 /* Size the arenas to exactly fit a given number of bodies. A count
904 of 0 fits the max number bodies into a PERL_ARENA_SIZE.block,
905 simplifying the default. If count > 0, the arena is sized to fit
906 only that many bodies, allowing arenas to be used for large, rare
907 bodies (XPVFM, XPVIO) without undue waste. The arena size is
908 limited by PERL_ARENA_SIZE, so we can safely oversize the
911 #define FIT_ARENA0(body_size) \
912 ((size_t)(PERL_ARENA_SIZE / body_size) * body_size)
913 #define FIT_ARENAn(count,body_size) \
914 ( count * body_size <= PERL_ARENA_SIZE) \
915 ? count * body_size \
916 : FIT_ARENA0 (body_size)
917 #define FIT_ARENA(count,body_size) \
919 ? FIT_ARENAn (count, body_size) \
920 : FIT_ARENA0 (body_size))
922 /* Calculate the length to copy. Specifically work out the length less any
923 final padding the compiler needed to add. See the comment in sv_upgrade
924 for why copying the padding proved to be a bug. */
926 #define copy_length(type, last_member) \
927 STRUCT_OFFSET(type, last_member) \
928 + sizeof (((type*)SvANY((const SV *)0))->last_member)
930 static const struct body_details bodies_by_type[] = {
931 /* HEs use this offset for their arena. */
932 { 0, 0, 0, SVt_NULL, FALSE, NONV, NOARENA, 0 },
934 /* IVs are in the head, so the allocation size is 0. */
936 sizeof(IV), /* This is used to copy out the IV body. */
937 STRUCT_OFFSET(XPVIV, xiv_iv), SVt_IV, FALSE, NONV,
938 NOARENA /* IVS don't need an arena */, 0
943 STRUCT_OFFSET(XPVNV, xnv_u),
944 SVt_NV, FALSE, HADNV, NOARENA, 0 },
946 { sizeof(NV), sizeof(NV),
947 STRUCT_OFFSET(XPVNV, xnv_u),
948 SVt_NV, FALSE, HADNV, HASARENA, FIT_ARENA(0, sizeof(NV)) },
951 { sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur),
952 copy_length(XPV, xpv_len) - STRUCT_OFFSET(XPV, xpv_cur),
953 + STRUCT_OFFSET(XPV, xpv_cur),
954 SVt_PV, FALSE, NONV, HASARENA,
955 FIT_ARENA(0, sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur)) },
957 { sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur),
958 copy_length(XINVLIST, is_offset) - STRUCT_OFFSET(XPV, xpv_cur),
959 + STRUCT_OFFSET(XPV, xpv_cur),
960 SVt_INVLIST, TRUE, NONV, HASARENA,
961 FIT_ARENA(0, sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur)) },
963 { sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur),
964 copy_length(XPVIV, xiv_u) - STRUCT_OFFSET(XPV, xpv_cur),
965 + STRUCT_OFFSET(XPV, xpv_cur),
966 SVt_PVIV, FALSE, NONV, HASARENA,
967 FIT_ARENA(0, sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur)) },
969 { sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur),
970 copy_length(XPVNV, xnv_u) - STRUCT_OFFSET(XPV, xpv_cur),
971 + STRUCT_OFFSET(XPV, xpv_cur),
972 SVt_PVNV, FALSE, HADNV, HASARENA,
973 FIT_ARENA(0, sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur)) },
975 { sizeof(XPVMG), copy_length(XPVMG, xnv_u), 0, SVt_PVMG, FALSE, HADNV,
976 HASARENA, FIT_ARENA(0, sizeof(XPVMG)) },
981 SVt_REGEXP, TRUE, NONV, HASARENA,
982 FIT_ARENA(0, sizeof(regexp))
985 { sizeof(XPVGV), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV,
986 HASARENA, FIT_ARENA(0, sizeof(XPVGV)) },
988 { sizeof(XPVLV), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV,
989 HASARENA, FIT_ARENA(0, sizeof(XPVLV)) },
992 copy_length(XPVAV, xav_alloc),
994 SVt_PVAV, TRUE, NONV, HASARENA,
995 FIT_ARENA(0, sizeof(XPVAV)) },
998 copy_length(XPVHV, xhv_max),
1000 SVt_PVHV, TRUE, NONV, HASARENA,
1001 FIT_ARENA(0, sizeof(XPVHV)) },
1006 SVt_PVCV, TRUE, NONV, HASARENA,
1007 FIT_ARENA(0, sizeof(XPVCV)) },
1012 SVt_PVFM, TRUE, NONV, NOARENA,
1013 FIT_ARENA(20, sizeof(XPVFM)) },
1018 SVt_PVIO, TRUE, NONV, HASARENA,
1019 FIT_ARENA(24, sizeof(XPVIO)) },
1022 #define new_body_allocated(sv_type) \
1023 (void *)((char *)S_new_body(aTHX_ sv_type) \
1024 - bodies_by_type[sv_type].offset)
1026 /* return a thing to the free list */
1028 #define del_body(thing, root) \
1030 void ** const thing_copy = (void **)thing; \
1031 *thing_copy = *root; \
1032 *root = (void*)thing_copy; \
1036 #if !(NVSIZE <= IVSIZE)
1037 # define new_XNV() safemalloc(sizeof(XPVNV))
1039 #define new_XPVNV() safemalloc(sizeof(XPVNV))
1040 #define new_XPVMG() safemalloc(sizeof(XPVMG))
1042 #define del_XPVGV(p) safefree(p)
1046 #if !(NVSIZE <= IVSIZE)
1047 # define new_XNV() new_body_allocated(SVt_NV)
1049 #define new_XPVNV() new_body_allocated(SVt_PVNV)
1050 #define new_XPVMG() new_body_allocated(SVt_PVMG)
1052 #define del_XPVGV(p) del_body(p + bodies_by_type[SVt_PVGV].offset, \
1053 &PL_body_roots[SVt_PVGV])
1057 /* no arena for you! */
1059 #define new_NOARENA(details) \
1060 safemalloc((details)->body_size + (details)->offset)
1061 #define new_NOARENAZ(details) \
1062 safecalloc((details)->body_size + (details)->offset, 1)
1065 Perl_more_bodies (pTHX_ const svtype sv_type, const size_t body_size,
1066 const size_t arena_size)
1068 void ** const root = &PL_body_roots[sv_type];
1069 struct arena_desc *adesc;
1070 struct arena_set *aroot = (struct arena_set *) PL_body_arenas;
1074 const size_t good_arena_size = Perl_malloc_good_size(arena_size);
1075 #if defined(DEBUGGING) && defined(PERL_GLOBAL_STRUCT)
1078 #if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT_PRIVATE)
1079 static bool done_sanity_check;
1081 /* PERL_GLOBAL_STRUCT_PRIVATE cannot coexist with global
1082 * variables like done_sanity_check. */
1083 if (!done_sanity_check) {
1084 unsigned int i = SVt_LAST;
1086 done_sanity_check = TRUE;
1089 assert (bodies_by_type[i].type == i);
1095 /* may need new arena-set to hold new arena */
1096 if (!aroot || aroot->curr >= aroot->set_size) {
1097 struct arena_set *newroot;
1098 Newxz(newroot, 1, struct arena_set);
1099 newroot->set_size = ARENAS_PER_SET;
1100 newroot->next = aroot;
1102 PL_body_arenas = (void *) newroot;
1103 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot));
1106 /* ok, now have arena-set with at least 1 empty/available arena-desc */
1107 curr = aroot->curr++;
1108 adesc = &(aroot->set[curr]);
1109 assert(!adesc->arena);
1111 Newx(adesc->arena, good_arena_size, char);
1112 adesc->size = good_arena_size;
1113 adesc->utype = sv_type;
1114 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %" UVuf "\n",
1115 curr, (void*)adesc->arena, (UV)good_arena_size));
1117 start = (char *) adesc->arena;
1119 /* Get the address of the byte after the end of the last body we can fit.
1120 Remember, this is integer division: */
1121 end = start + good_arena_size / body_size * body_size;
1123 /* computed count doesn't reflect the 1st slot reservation */
1124 #if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE)
1125 DEBUG_m(PerlIO_printf(Perl_debug_log,
1126 "arena %p end %p arena-size %d (from %d) type %d "
1128 (void*)start, (void*)end, (int)good_arena_size,
1129 (int)arena_size, sv_type, (int)body_size,
1130 (int)good_arena_size / (int)body_size));
1132 DEBUG_m(PerlIO_printf(Perl_debug_log,
1133 "arena %p end %p arena-size %d type %d size %d ct %d\n",
1134 (void*)start, (void*)end,
1135 (int)arena_size, sv_type, (int)body_size,
1136 (int)good_arena_size / (int)body_size));
1138 *root = (void *)start;
1141 /* Where the next body would start: */
1142 char * const next = start + body_size;
1145 /* This is the last body: */
1146 assert(next == end);
1148 *(void **)start = 0;
1152 *(void**) start = (void *)next;
1157 /* grab a new thing from the free list, allocating more if necessary.
1158 The inline version is used for speed in hot routines, and the
1159 function using it serves the rest (unless PURIFY).
1161 #define new_body_inline(xpv, sv_type) \
1163 void ** const r3wt = &PL_body_roots[sv_type]; \
1164 xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \
1165 ? *((void **)(r3wt)) : Perl_more_bodies(aTHX_ sv_type, \
1166 bodies_by_type[sv_type].body_size,\
1167 bodies_by_type[sv_type].arena_size)); \
1168 *(r3wt) = *(void**)(xpv); \
1174 S_new_body(pTHX_ const svtype sv_type)
1177 new_body_inline(xpv, sv_type);
1183 static const struct body_details fake_rv =
1184 { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 };
1187 =for apidoc sv_upgrade
1189 Upgrade an SV to a more complex form. Generally adds a new body type to the
1190 SV, then copies across as much information as possible from the old body.
1191 It croaks if the SV is already in a more complex form than requested. You
1192 generally want to use the C<SvUPGRADE> macro wrapper, which checks the type
1193 before calling C<sv_upgrade>, and hence does not croak. See also
1200 Perl_sv_upgrade(pTHX_ SV *const sv, svtype new_type)
1204 const svtype old_type = SvTYPE(sv);
1205 const struct body_details *new_type_details;
1206 const struct body_details *old_type_details
1207 = bodies_by_type + old_type;
1208 SV *referent = NULL;
1210 PERL_ARGS_ASSERT_SV_UPGRADE;
1212 if (old_type == new_type)
1215 /* This clause was purposefully added ahead of the early return above to
1216 the shared string hackery for (sort {$a <=> $b} keys %hash), with the
1217 inference by Nick I-S that it would fix other troublesome cases. See
1218 changes 7162, 7163 (f130fd4589cf5fbb24149cd4db4137c8326f49c1 and parent)
1220 Given that shared hash key scalars are no longer PVIV, but PV, there is
1221 no longer need to unshare so as to free up the IVX slot for its proper
1222 purpose. So it's safe to move the early return earlier. */
1224 if (new_type > SVt_PVMG && SvIsCOW(sv)) {
1225 sv_force_normal_flags(sv, 0);
1228 old_body = SvANY(sv);
1230 /* Copying structures onto other structures that have been neatly zeroed
1231 has a subtle gotcha. Consider XPVMG
1233 +------+------+------+------+------+-------+-------+
1234 | NV | CUR | LEN | IV | MAGIC | STASH |
1235 +------+------+------+------+------+-------+-------+
1236 0 4 8 12 16 20 24 28
1238 where NVs are aligned to 8 bytes, so that sizeof that structure is
1239 actually 32 bytes long, with 4 bytes of padding at the end:
1241 +------+------+------+------+------+-------+-------+------+
1242 | NV | CUR | LEN | IV | MAGIC | STASH | ??? |
1243 +------+------+------+------+------+-------+-------+------+
1244 0 4 8 12 16 20 24 28 32
1246 so what happens if you allocate memory for this structure:
1248 +------+------+------+------+------+-------+-------+------+------+...
1249 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME |
1250 +------+------+------+------+------+-------+-------+------+------+...
1251 0 4 8 12 16 20 24 28 32 36
1253 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you
1254 expect, because you copy the area marked ??? onto GP. Now, ??? may have
1255 started out as zero once, but it's quite possible that it isn't. So now,
1256 rather than a nicely zeroed GP, you have it pointing somewhere random.
1259 (In fact, GP ends up pointing at a previous GP structure, because the
1260 principle cause of the padding in XPVMG getting garbage is a copy of
1261 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now
1262 this happens to be moot because XPVGV has been re-ordered, with GP
1263 no longer after STASH)
1265 So we are careful and work out the size of used parts of all the
1273 referent = SvRV(sv);
1274 old_type_details = &fake_rv;
1275 if (new_type == SVt_NV)
1276 new_type = SVt_PVNV;
1278 if (new_type < SVt_PVIV) {
1279 new_type = (new_type == SVt_NV)
1280 ? SVt_PVNV : SVt_PVIV;
1285 if (new_type < SVt_PVNV) {
1286 new_type = SVt_PVNV;
1290 assert(new_type > SVt_PV);
1291 STATIC_ASSERT_STMT(SVt_IV < SVt_PV);
1292 STATIC_ASSERT_STMT(SVt_NV < SVt_PV);
1299 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena,
1300 there's no way that it can be safely upgraded, because perl.c
1301 expects to Safefree(SvANY(PL_mess_sv)) */
1302 assert(sv != PL_mess_sv);
1305 if (UNLIKELY(old_type_details->cant_upgrade))
1306 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf,
1307 sv_reftype(sv, 0), (UV) old_type, (UV) new_type);
1310 if (UNLIKELY(old_type > new_type))
1311 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d",
1312 (int)old_type, (int)new_type);
1314 new_type_details = bodies_by_type + new_type;
1316 SvFLAGS(sv) &= ~SVTYPEMASK;
1317 SvFLAGS(sv) |= new_type;
1319 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of
1320 the return statements above will have triggered. */
1321 assert (new_type != SVt_NULL);
1324 assert(old_type == SVt_NULL);
1325 SET_SVANY_FOR_BODYLESS_IV(sv);
1329 assert(old_type == SVt_NULL);
1330 #if NVSIZE <= IVSIZE
1331 SET_SVANY_FOR_BODYLESS_NV(sv);
1333 SvANY(sv) = new_XNV();
1339 assert(new_type_details->body_size);
1342 assert(new_type_details->arena);
1343 assert(new_type_details->arena_size);
1344 /* This points to the start of the allocated area. */
1345 new_body_inline(new_body, new_type);
1346 Zero(new_body, new_type_details->body_size, char);
1347 new_body = ((char *)new_body) - new_type_details->offset;
1349 /* We always allocated the full length item with PURIFY. To do this
1350 we fake things so that arena is false for all 16 types.. */
1351 new_body = new_NOARENAZ(new_type_details);
1353 SvANY(sv) = new_body;
1354 if (new_type == SVt_PVAV) {
1358 if (old_type_details->body_size) {
1361 /* It will have been zeroed when the new body was allocated.
1362 Lets not write to it, in case it confuses a write-back
1368 #ifndef NODEFAULT_SHAREKEYS
1369 HvSHAREKEYS_on(sv); /* key-sharing on by default */
1371 /* start with PERL_HASH_DEFAULT_HvMAX+1 buckets: */
1372 HvMAX(sv) = PERL_HASH_DEFAULT_HvMAX;
1375 /* SVt_NULL isn't the only thing upgraded to AV or HV.
1376 The target created by newSVrv also is, and it can have magic.
1377 However, it never has SvPVX set.
1379 if (old_type == SVt_IV) {
1381 } else if (old_type >= SVt_PV) {
1382 assert(SvPVX_const(sv) == 0);
1385 if (old_type >= SVt_PVMG) {
1386 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic);
1387 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash);
1389 sv->sv_u.svu_array = NULL; /* or svu_hash */
1394 /* XXX Is this still needed? Was it ever needed? Surely as there is
1395 no route from NV to PVIV, NOK can never be true */
1396 assert(!SvNOKp(sv));
1410 assert(new_type_details->body_size);
1411 /* We always allocated the full length item with PURIFY. To do this
1412 we fake things so that arena is false for all 16 types.. */
1413 if(new_type_details->arena) {
1414 /* This points to the start of the allocated area. */
1415 new_body_inline(new_body, new_type);
1416 Zero(new_body, new_type_details->body_size, char);
1417 new_body = ((char *)new_body) - new_type_details->offset;
1419 new_body = new_NOARENAZ(new_type_details);
1421 SvANY(sv) = new_body;
1423 if (old_type_details->copy) {
1424 /* There is now the potential for an upgrade from something without
1425 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */
1426 int offset = old_type_details->offset;
1427 int length = old_type_details->copy;
1429 if (new_type_details->offset > old_type_details->offset) {
1430 const int difference
1431 = new_type_details->offset - old_type_details->offset;
1432 offset += difference;
1433 length -= difference;
1435 assert (length >= 0);
1437 Copy((char *)old_body + offset, (char *)new_body + offset, length,
1441 #ifndef NV_ZERO_IS_ALLBITS_ZERO
1442 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a
1443 * correct 0.0 for us. Otherwise, if the old body didn't have an
1444 * NV slot, but the new one does, then we need to initialise the
1445 * freshly created NV slot with whatever the correct bit pattern is
1447 if (old_type_details->zero_nv && !new_type_details->zero_nv
1448 && !isGV_with_GP(sv))
1452 if (UNLIKELY(new_type == SVt_PVIO)) {
1453 IO * const io = MUTABLE_IO(sv);
1454 GV *iogv = gv_fetchpvs("IO::File::", GV_ADD, SVt_PVHV);
1457 /* Clear the stashcache because a new IO could overrule a package
1459 DEBUG_o(Perl_deb(aTHX_ "sv_upgrade clearing PL_stashcache\n"));
1460 hv_clear(PL_stashcache);
1462 SvSTASH_set(io, MUTABLE_HV(SvREFCNT_inc(GvHV(iogv))));
1463 IoPAGE_LEN(sv) = 60;
1465 if (old_type < SVt_PV) {
1466 /* referent will be NULL unless the old type was SVt_IV emulating
1468 sv->sv_u.svu_rv = referent;
1472 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1473 (unsigned long)new_type);
1476 /* if this is zero, this is a body-less SVt_NULL, SVt_IV/SVt_RV,
1477 and sometimes SVt_NV */
1478 if (old_type_details->body_size) {
1482 /* Note that there is an assumption that all bodies of types that
1483 can be upgraded came from arenas. Only the more complex non-
1484 upgradable types are allowed to be directly malloc()ed. */
1485 assert(old_type_details->arena);
1486 del_body((void*)((char*)old_body + old_type_details->offset),
1487 &PL_body_roots[old_type]);
1493 =for apidoc sv_backoff
1495 Remove any string offset. You should normally use the C<SvOOK_off> macro
1501 /* prior to 5.000 stable, this function returned the new OOK-less SvFLAGS
1502 prior to 5.23.4 this function always returned 0
1506 Perl_sv_backoff(SV *const sv)
1509 const char * const s = SvPVX_const(sv);
1511 PERL_ARGS_ASSERT_SV_BACKOFF;
1514 assert(SvTYPE(sv) != SVt_PVHV);
1515 assert(SvTYPE(sv) != SVt_PVAV);
1517 SvOOK_offset(sv, delta);
1519 SvLEN_set(sv, SvLEN(sv) + delta);
1520 SvPV_set(sv, SvPVX(sv) - delta);
1521 SvFLAGS(sv) &= ~SVf_OOK;
1522 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
1527 /* forward declaration */
1528 static void S_sv_uncow(pTHX_ SV * const sv, const U32 flags);
1534 Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1535 upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1536 Use the C<SvGROW> wrapper instead.
1543 Perl_sv_grow(pTHX_ SV *const sv, STRLEN newlen)
1547 PERL_ARGS_ASSERT_SV_GROW;
1551 if (SvTYPE(sv) < SVt_PV) {
1552 sv_upgrade(sv, SVt_PV);
1553 s = SvPVX_mutable(sv);
1555 else if (SvOOK(sv)) { /* pv is offset? */
1557 s = SvPVX_mutable(sv);
1558 if (newlen > SvLEN(sv))
1559 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
1563 if (SvIsCOW(sv)) S_sv_uncow(aTHX_ sv, 0);
1564 s = SvPVX_mutable(sv);
1567 #ifdef PERL_COPY_ON_WRITE
1568 /* the new COW scheme uses SvPVX(sv)[SvLEN(sv)-1] (if spare)
1569 * to store the COW count. So in general, allocate one more byte than
1570 * asked for, to make it likely this byte is always spare: and thus
1571 * make more strings COW-able.
1573 * Only increment if the allocation isn't MEM_SIZE_MAX,
1574 * otherwise it will wrap to 0.
1576 if ( newlen != MEM_SIZE_MAX )
1580 #if defined(PERL_USE_MALLOC_SIZE) && defined(Perl_safesysmalloc_size)
1581 #define PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1584 if (newlen > SvLEN(sv)) { /* need more room? */
1585 STRLEN minlen = SvCUR(sv);
1586 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
1587 if (newlen < minlen)
1589 #ifndef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1591 /* Don't round up on the first allocation, as odds are pretty good that
1592 * the initial request is accurate as to what is really needed */
1594 STRLEN rounded = PERL_STRLEN_ROUNDUP(newlen);
1595 if (rounded > newlen)
1599 if (SvLEN(sv) && s) {
1600 s = (char*)saferealloc(s, newlen);
1603 s = (char*)safemalloc(newlen);
1604 if (SvPVX_const(sv) && SvCUR(sv)) {
1605 Move(SvPVX_const(sv), s, SvCUR(sv), char);
1609 #ifdef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1610 /* Do this here, do it once, do it right, and then we will never get
1611 called back into sv_grow() unless there really is some growing
1613 SvLEN_set(sv, Perl_safesysmalloc_size(s));
1615 SvLEN_set(sv, newlen);
1622 =for apidoc sv_setiv
1624 Copies an integer into the given SV, upgrading first if necessary.
1625 Does not handle 'set' magic. See also C<L</sv_setiv_mg>>.
1631 Perl_sv_setiv(pTHX_ SV *const sv, const IV i)
1633 PERL_ARGS_ASSERT_SV_SETIV;
1635 SV_CHECK_THINKFIRST_COW_DROP(sv);
1636 switch (SvTYPE(sv)) {
1639 sv_upgrade(sv, SVt_IV);
1642 sv_upgrade(sv, SVt_PVIV);
1646 if (!isGV_with_GP(sv))
1654 /* diag_listed_as: Can't coerce %s to %s in %s */
1655 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1657 NOT_REACHED; /* NOTREACHED */
1661 (void)SvIOK_only(sv); /* validate number */
1667 =for apidoc sv_setiv_mg
1669 Like C<sv_setiv>, but also handles 'set' magic.
1675 Perl_sv_setiv_mg(pTHX_ SV *const sv, const IV i)
1677 PERL_ARGS_ASSERT_SV_SETIV_MG;
1684 =for apidoc sv_setuv
1686 Copies an unsigned integer into the given SV, upgrading first if necessary.
1687 Does not handle 'set' magic. See also C<L</sv_setuv_mg>>.
1693 Perl_sv_setuv(pTHX_ SV *const sv, const UV u)
1695 PERL_ARGS_ASSERT_SV_SETUV;
1697 /* With the if statement to ensure that integers are stored as IVs whenever
1699 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
1702 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1704 If you wish to remove the following if statement, so that this routine
1705 (and its callers) always return UVs, please benchmark to see what the
1706 effect is. Modern CPUs may be different. Or may not :-)
1708 if (u <= (UV)IV_MAX) {
1709 sv_setiv(sv, (IV)u);
1718 =for apidoc sv_setuv_mg
1720 Like C<sv_setuv>, but also handles 'set' magic.
1726 Perl_sv_setuv_mg(pTHX_ SV *const sv, const UV u)
1728 PERL_ARGS_ASSERT_SV_SETUV_MG;
1735 =for apidoc sv_setnv
1737 Copies a double into the given SV, upgrading first if necessary.
1738 Does not handle 'set' magic. See also C<L</sv_setnv_mg>>.
1744 Perl_sv_setnv(pTHX_ SV *const sv, const NV num)
1746 PERL_ARGS_ASSERT_SV_SETNV;
1748 SV_CHECK_THINKFIRST_COW_DROP(sv);
1749 switch (SvTYPE(sv)) {
1752 sv_upgrade(sv, SVt_NV);
1756 sv_upgrade(sv, SVt_PVNV);
1760 if (!isGV_with_GP(sv))
1768 /* diag_listed_as: Can't coerce %s to %s in %s */
1769 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
1771 NOT_REACHED; /* NOTREACHED */
1776 (void)SvNOK_only(sv); /* validate number */
1781 =for apidoc sv_setnv_mg
1783 Like C<sv_setnv>, but also handles 'set' magic.
1789 Perl_sv_setnv_mg(pTHX_ SV *const sv, const NV num)
1791 PERL_ARGS_ASSERT_SV_SETNV_MG;
1797 /* Return a cleaned-up, printable version of sv, for non-numeric, or
1798 * not incrementable warning display.
1799 * Originally part of S_not_a_number().
1800 * The return value may be != tmpbuf.
1804 S_sv_display(pTHX_ SV *const sv, char *tmpbuf, STRLEN tmpbuf_size) {
1807 PERL_ARGS_ASSERT_SV_DISPLAY;
1810 SV *dsv = newSVpvs_flags("", SVs_TEMP);
1811 pv = sv_uni_display(dsv, sv, 32, UNI_DISPLAY_ISPRINT);
1814 const char * const limit = tmpbuf + tmpbuf_size - 8;
1815 /* each *s can expand to 4 chars + "...\0",
1816 i.e. need room for 8 chars */
1818 const char *s = SvPVX_const(sv);
1819 const char * const end = s + SvCUR(sv);
1820 for ( ; s < end && d < limit; s++ ) {
1822 if (! isASCII(ch) && !isPRINT_LC(ch)) {
1826 /* Map to ASCII "equivalent" of Latin1 */
1827 ch = LATIN1_TO_NATIVE(NATIVE_TO_LATIN1(ch) & 127);
1833 else if (ch == '\r') {
1837 else if (ch == '\f') {
1841 else if (ch == '\\') {
1845 else if (ch == '\0') {
1849 else if (isPRINT_LC(ch))
1868 /* Print an "isn't numeric" warning, using a cleaned-up,
1869 * printable version of the offending string
1873 S_not_a_number(pTHX_ SV *const sv)
1878 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1880 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1883 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1884 /* diag_listed_as: Argument "%s" isn't numeric%s */
1885 "Argument \"%s\" isn't numeric in %s", pv,
1888 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1889 /* diag_listed_as: Argument "%s" isn't numeric%s */
1890 "Argument \"%s\" isn't numeric", pv);
1894 S_not_incrementable(pTHX_ SV *const sv) {
1898 PERL_ARGS_ASSERT_NOT_INCREMENTABLE;
1900 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1902 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1903 "Argument \"%s\" treated as 0 in increment (++)", pv);
1907 =for apidoc looks_like_number
1909 Test if the content of an SV looks like a number (or is a number).
1910 C<Inf> and C<Infinity> are treated as numbers (so will not issue a
1911 non-numeric warning), even if your C<atof()> doesn't grok them. Get-magic is
1918 Perl_looks_like_number(pTHX_ SV *const sv)
1924 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1926 if (SvPOK(sv) || SvPOKp(sv)) {
1927 sbegin = SvPV_nomg_const(sv, len);
1930 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
1931 numtype = grok_number(sbegin, len, NULL);
1932 return ((numtype & IS_NUMBER_TRAILING)) ? 0 : numtype;
1936 S_glob_2number(pTHX_ GV * const gv)
1938 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1940 /* We know that all GVs stringify to something that is not-a-number,
1941 so no need to test that. */
1942 if (ckWARN(WARN_NUMERIC))
1944 SV *const buffer = sv_newmortal();
1945 gv_efullname3(buffer, gv, "*");
1946 not_a_number(buffer);
1948 /* We just want something true to return, so that S_sv_2iuv_common
1949 can tail call us and return true. */
1953 /* Actually, ISO C leaves conversion of UV to IV undefined, but
1954 until proven guilty, assume that things are not that bad... */
1959 As 64 bit platforms often have an NV that doesn't preserve all bits of
1960 an IV (an assumption perl has been based on to date) it becomes necessary
1961 to remove the assumption that the NV always carries enough precision to
1962 recreate the IV whenever needed, and that the NV is the canonical form.
1963 Instead, IV/UV and NV need to be given equal rights. So as to not lose
1964 precision as a side effect of conversion (which would lead to insanity
1965 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
1966 1) to distinguish between IV/UV/NV slots that have a valid conversion cached
1967 where precision was lost, and IV/UV/NV slots that have a valid conversion
1968 which has lost no precision
1969 2) to ensure that if a numeric conversion to one form is requested that
1970 would lose precision, the precise conversion (or differently
1971 imprecise conversion) is also performed and cached, to prevent
1972 requests for different numeric formats on the same SV causing
1973 lossy conversion chains. (lossless conversion chains are perfectly
1978 SvIOKp is true if the IV slot contains a valid value
1979 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1980 SvNOKp is true if the NV slot contains a valid value
1981 SvNOK is true only if the NV value is accurate
1984 while converting from PV to NV, check to see if converting that NV to an
1985 IV(or UV) would lose accuracy over a direct conversion from PV to
1986 IV(or UV). If it would, cache both conversions, return NV, but mark
1987 SV as IOK NOKp (ie not NOK).
1989 While converting from PV to IV, check to see if converting that IV to an
1990 NV would lose accuracy over a direct conversion from PV to NV. If it
1991 would, cache both conversions, flag similarly.
1993 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
1994 correctly because if IV & NV were set NV *always* overruled.
1995 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
1996 changes - now IV and NV together means that the two are interchangeable:
1997 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
1999 The benefit of this is that operations such as pp_add know that if
2000 SvIOK is true for both left and right operands, then integer addition
2001 can be used instead of floating point (for cases where the result won't
2002 overflow). Before, floating point was always used, which could lead to
2003 loss of precision compared with integer addition.
2005 * making IV and NV equal status should make maths accurate on 64 bit
2007 * may speed up maths somewhat if pp_add and friends start to use
2008 integers when possible instead of fp. (Hopefully the overhead in
2009 looking for SvIOK and checking for overflow will not outweigh the
2010 fp to integer speedup)
2011 * will slow down integer operations (callers of SvIV) on "inaccurate"
2012 values, as the change from SvIOK to SvIOKp will cause a call into
2013 sv_2iv each time rather than a macro access direct to the IV slot
2014 * should speed up number->string conversion on integers as IV is
2015 favoured when IV and NV are equally accurate
2017 ####################################################################
2018 You had better be using SvIOK_notUV if you want an IV for arithmetic:
2019 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
2020 On the other hand, SvUOK is true iff UV.
2021 ####################################################################
2023 Your mileage will vary depending your CPU's relative fp to integer
2027 #ifndef NV_PRESERVES_UV
2028 # define IS_NUMBER_UNDERFLOW_IV 1
2029 # define IS_NUMBER_UNDERFLOW_UV 2
2030 # define IS_NUMBER_IV_AND_UV 2
2031 # define IS_NUMBER_OVERFLOW_IV 4
2032 # define IS_NUMBER_OVERFLOW_UV 5
2034 /* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
2036 /* For sv_2nv these three cases are "SvNOK and don't bother casting" */
2038 S_sv_2iuv_non_preserve(pTHX_ SV *const sv
2044 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
2045 PERL_UNUSED_CONTEXT;
2047 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));
2048 if (SvNVX(sv) < (NV)IV_MIN) {
2049 (void)SvIOKp_on(sv);
2051 SvIV_set(sv, IV_MIN);
2052 return IS_NUMBER_UNDERFLOW_IV;
2054 if (SvNVX(sv) > (NV)UV_MAX) {
2055 (void)SvIOKp_on(sv);
2058 SvUV_set(sv, UV_MAX);
2059 return IS_NUMBER_OVERFLOW_UV;
2061 (void)SvIOKp_on(sv);
2063 /* Can't use strtol etc to convert this string. (See truth table in
2065 if (SvNVX(sv) <= (UV)IV_MAX) {
2066 SvIV_set(sv, I_V(SvNVX(sv)));
2067 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2068 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
2070 /* Integer is imprecise. NOK, IOKp */
2072 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
2075 SvUV_set(sv, U_V(SvNVX(sv)));
2076 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2077 if (SvUVX(sv) == UV_MAX) {
2078 /* As we know that NVs don't preserve UVs, UV_MAX cannot
2079 possibly be preserved by NV. Hence, it must be overflow.
2081 return IS_NUMBER_OVERFLOW_UV;
2083 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
2085 /* Integer is imprecise. NOK, IOKp */
2087 return IS_NUMBER_OVERFLOW_IV;
2089 #endif /* !NV_PRESERVES_UV*/
2091 /* If numtype is infnan, set the NV of the sv accordingly.
2092 * If numtype is anything else, try setting the NV using Atof(PV). */
2094 # pragma warning(push)
2095 # pragma warning(disable:4756;disable:4056)
2098 S_sv_setnv(pTHX_ SV* sv, int numtype)
2100 bool pok = cBOOL(SvPOK(sv));
2103 if ((numtype & IS_NUMBER_INFINITY)) {
2104 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -NV_INF : NV_INF);
2109 if ((numtype & IS_NUMBER_NAN)) {
2110 SvNV_set(sv, NV_NAN);
2115 SvNV_set(sv, Atof(SvPVX_const(sv)));
2116 /* Purposefully no true nok here, since we don't want to blow
2117 * away the possible IOK/UV of an existing sv. */
2120 SvNOK_only(sv); /* No IV or UV please, this is pure infnan. */
2122 SvPOK_on(sv); /* PV is okay, though. */
2126 # pragma warning(pop)
2130 S_sv_2iuv_common(pTHX_ SV *const sv)
2132 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
2135 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
2136 * without also getting a cached IV/UV from it at the same time
2137 * (ie PV->NV conversion should detect loss of accuracy and cache
2138 * IV or UV at same time to avoid this. */
2139 /* IV-over-UV optimisation - choose to cache IV if possible */
2141 if (SvTYPE(sv) == SVt_NV)
2142 sv_upgrade(sv, SVt_PVNV);
2144 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
2145 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
2146 certainly cast into the IV range at IV_MAX, whereas the correct
2147 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
2149 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2150 if (Perl_isnan(SvNVX(sv))) {
2156 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2157 SvIV_set(sv, I_V(SvNVX(sv)));
2158 if (SvNVX(sv) == (NV) SvIVX(sv)
2159 #ifndef NV_PRESERVES_UV
2160 && SvIVX(sv) != IV_MIN /* avoid negating IV_MIN below */
2161 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2162 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2163 /* Don't flag it as "accurately an integer" if the number
2164 came from a (by definition imprecise) NV operation, and
2165 we're outside the range of NV integer precision */
2169 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2171 /* scalar has trailing garbage, eg "42a" */
2173 DEBUG_c(PerlIO_printf(Perl_debug_log,
2174 "0x%" UVxf " iv(%" NVgf " => %" IVdf ") (precise)\n",
2180 /* IV not precise. No need to convert from PV, as NV
2181 conversion would already have cached IV if it detected
2182 that PV->IV would be better than PV->NV->IV
2183 flags already correct - don't set public IOK. */
2184 DEBUG_c(PerlIO_printf(Perl_debug_log,
2185 "0x%" UVxf " iv(%" NVgf " => %" IVdf ") (imprecise)\n",
2190 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2191 but the cast (NV)IV_MIN rounds to a the value less (more
2192 negative) than IV_MIN which happens to be equal to SvNVX ??
2193 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2194 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2195 (NV)UVX == NVX are both true, but the values differ. :-(
2196 Hopefully for 2s complement IV_MIN is something like
2197 0x8000000000000000 which will be exact. NWC */
2200 SvUV_set(sv, U_V(SvNVX(sv)));
2202 (SvNVX(sv) == (NV) SvUVX(sv))
2203 #ifndef NV_PRESERVES_UV
2204 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2205 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2206 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2207 /* Don't flag it as "accurately an integer" if the number
2208 came from a (by definition imprecise) NV operation, and
2209 we're outside the range of NV integer precision */
2215 DEBUG_c(PerlIO_printf(Perl_debug_log,
2216 "0x%" UVxf " 2iv(%" UVuf " => %" IVdf ") (as unsigned)\n",
2222 else if (SvPOKp(sv)) {
2225 const char *s = SvPVX_const(sv);
2226 const STRLEN cur = SvCUR(sv);
2228 /* short-cut for a single digit string like "1" */
2233 if (SvTYPE(sv) < SVt_PVIV)
2234 sv_upgrade(sv, SVt_PVIV);
2236 SvIV_set(sv, (IV)(c - '0'));
2241 numtype = grok_number(s, cur, &value);
2242 /* We want to avoid a possible problem when we cache an IV/ a UV which
2243 may be later translated to an NV, and the resulting NV is not
2244 the same as the direct translation of the initial string
2245 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2246 be careful to ensure that the value with the .456 is around if the
2247 NV value is requested in the future).
2249 This means that if we cache such an IV/a UV, we need to cache the
2250 NV as well. Moreover, we trade speed for space, and do not
2251 cache the NV if we are sure it's not needed.
2254 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2255 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2256 == IS_NUMBER_IN_UV) {
2257 /* It's definitely an integer, only upgrade to PVIV */
2258 if (SvTYPE(sv) < SVt_PVIV)
2259 sv_upgrade(sv, SVt_PVIV);
2261 } else if (SvTYPE(sv) < SVt_PVNV)
2262 sv_upgrade(sv, SVt_PVNV);
2264 if ((numtype & (IS_NUMBER_INFINITY | IS_NUMBER_NAN))) {
2265 if (ckWARN(WARN_NUMERIC) && ((numtype & IS_NUMBER_TRAILING)))
2267 S_sv_setnv(aTHX_ sv, numtype);
2271 /* If NVs preserve UVs then we only use the UV value if we know that
2272 we aren't going to call atof() below. If NVs don't preserve UVs
2273 then the value returned may have more precision than atof() will
2274 return, even though value isn't perfectly accurate. */
2275 if ((numtype & (IS_NUMBER_IN_UV
2276 #ifdef NV_PRESERVES_UV
2279 )) == IS_NUMBER_IN_UV) {
2280 /* This won't turn off the public IOK flag if it was set above */
2281 (void)SvIOKp_on(sv);
2283 if (!(numtype & IS_NUMBER_NEG)) {
2285 if (value <= (UV)IV_MAX) {
2286 SvIV_set(sv, (IV)value);
2288 /* it didn't overflow, and it was positive. */
2289 SvUV_set(sv, value);
2293 /* 2s complement assumption */
2294 if (value <= (UV)IV_MIN) {
2295 SvIV_set(sv, value == (UV)IV_MIN
2296 ? IV_MIN : -(IV)value);
2298 /* Too negative for an IV. This is a double upgrade, but
2299 I'm assuming it will be rare. */
2300 if (SvTYPE(sv) < SVt_PVNV)
2301 sv_upgrade(sv, SVt_PVNV);
2305 SvNV_set(sv, -(NV)value);
2306 SvIV_set(sv, IV_MIN);
2310 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2311 will be in the previous block to set the IV slot, and the next
2312 block to set the NV slot. So no else here. */
2314 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2315 != IS_NUMBER_IN_UV) {
2316 /* It wasn't an (integer that doesn't overflow the UV). */
2317 S_sv_setnv(aTHX_ sv, numtype);
2319 if (! numtype && ckWARN(WARN_NUMERIC))
2322 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2iv(%" NVgf ")\n",
2323 PTR2UV(sv), SvNVX(sv)));
2325 #ifdef NV_PRESERVES_UV
2326 (void)SvIOKp_on(sv);
2328 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2329 if (Perl_isnan(SvNVX(sv))) {
2335 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2336 SvIV_set(sv, I_V(SvNVX(sv)));
2337 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2340 NOOP; /* Integer is imprecise. NOK, IOKp */
2342 /* UV will not work better than IV */
2344 if (SvNVX(sv) > (NV)UV_MAX) {
2346 /* Integer is inaccurate. NOK, IOKp, is UV */
2347 SvUV_set(sv, UV_MAX);
2349 SvUV_set(sv, U_V(SvNVX(sv)));
2350 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2351 NV preservse UV so can do correct comparison. */
2352 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2355 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
2360 #else /* NV_PRESERVES_UV */
2361 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2362 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
2363 /* The IV/UV slot will have been set from value returned by
2364 grok_number above. The NV slot has just been set using
2367 assert (SvIOKp(sv));
2369 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2370 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2371 /* Small enough to preserve all bits. */
2372 (void)SvIOKp_on(sv);
2374 SvIV_set(sv, I_V(SvNVX(sv)));
2375 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2377 /* Assumption: first non-preserved integer is < IV_MAX,
2378 this NV is in the preserved range, therefore: */
2379 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2381 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);
2385 0 0 already failed to read UV.
2386 0 1 already failed to read UV.
2387 1 0 you won't get here in this case. IV/UV
2388 slot set, public IOK, Atof() unneeded.
2389 1 1 already read UV.
2390 so there's no point in sv_2iuv_non_preserve() attempting
2391 to use atol, strtol, strtoul etc. */
2393 sv_2iuv_non_preserve (sv, numtype);
2395 sv_2iuv_non_preserve (sv);
2399 #endif /* NV_PRESERVES_UV */
2400 /* It might be more code efficient to go through the entire logic above
2401 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2402 gets complex and potentially buggy, so more programmer efficient
2403 to do it this way, by turning off the public flags: */
2405 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2409 if (isGV_with_GP(sv))
2410 return glob_2number(MUTABLE_GV(sv));
2412 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2414 if (SvTYPE(sv) < SVt_IV)
2415 /* Typically the caller expects that sv_any is not NULL now. */
2416 sv_upgrade(sv, SVt_IV);
2417 /* Return 0 from the caller. */
2424 =for apidoc sv_2iv_flags
2426 Return the integer value of an SV, doing any necessary string
2427 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2428 Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2434 Perl_sv_2iv_flags(pTHX_ SV *const sv, const I32 flags)
2436 PERL_ARGS_ASSERT_SV_2IV_FLAGS;
2438 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2439 && SvTYPE(sv) != SVt_PVFM);
2441 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2447 if (flags & SV_SKIP_OVERLOAD)
2449 tmpstr = AMG_CALLunary(sv, numer_amg);
2450 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2451 return SvIV(tmpstr);
2454 return PTR2IV(SvRV(sv));
2457 if (SvVALID(sv) || isREGEXP(sv)) {
2458 /* FBMs use the space for SvIVX and SvNVX for other purposes, so
2459 must not let them cache IVs.
2460 In practice they are extremely unlikely to actually get anywhere
2461 accessible by user Perl code - the only way that I'm aware of is when
2462 a constant subroutine which is used as the second argument to index.
2464 Regexps have no SvIVX and SvNVX fields.
2469 const char * const ptr =
2470 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2472 = grok_number(ptr, SvCUR(sv), &value);
2474 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2475 == IS_NUMBER_IN_UV) {
2476 /* It's definitely an integer */
2477 if (numtype & IS_NUMBER_NEG) {
2478 if (value < (UV)IV_MIN)
2481 if (value < (UV)IV_MAX)
2486 /* Quite wrong but no good choices. */
2487 if ((numtype & IS_NUMBER_INFINITY)) {
2488 return (numtype & IS_NUMBER_NEG) ? IV_MIN : IV_MAX;
2489 } else if ((numtype & IS_NUMBER_NAN)) {
2490 return 0; /* So wrong. */
2494 if (ckWARN(WARN_NUMERIC))
2497 return I_V(Atof(ptr));
2501 if (SvTHINKFIRST(sv)) {
2502 if (SvREADONLY(sv) && !SvOK(sv)) {
2503 if (ckWARN(WARN_UNINITIALIZED))
2510 if (S_sv_2iuv_common(aTHX_ sv))
2514 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2iv(%" IVdf ")\n",
2515 PTR2UV(sv),SvIVX(sv)));
2516 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
2520 =for apidoc sv_2uv_flags
2522 Return the unsigned integer value of an SV, doing any necessary string
2523 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2524 Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
2530 Perl_sv_2uv_flags(pTHX_ SV *const sv, const I32 flags)
2532 PERL_ARGS_ASSERT_SV_2UV_FLAGS;
2534 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2540 if (flags & SV_SKIP_OVERLOAD)
2542 tmpstr = AMG_CALLunary(sv, numer_amg);
2543 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2544 return SvUV(tmpstr);
2547 return PTR2UV(SvRV(sv));
2550 if (SvVALID(sv) || isREGEXP(sv)) {
2551 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2552 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2553 Regexps have no SvIVX and SvNVX fields. */
2557 const char * const ptr =
2558 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2560 = grok_number(ptr, SvCUR(sv), &value);
2562 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2563 == IS_NUMBER_IN_UV) {
2564 /* It's definitely an integer */
2565 if (!(numtype & IS_NUMBER_NEG))
2569 /* Quite wrong but no good choices. */
2570 if ((numtype & IS_NUMBER_INFINITY)) {
2571 return UV_MAX; /* So wrong. */
2572 } else if ((numtype & IS_NUMBER_NAN)) {
2573 return 0; /* So wrong. */
2577 if (ckWARN(WARN_NUMERIC))
2580 return U_V(Atof(ptr));
2584 if (SvTHINKFIRST(sv)) {
2585 if (SvREADONLY(sv) && !SvOK(sv)) {
2586 if (ckWARN(WARN_UNINITIALIZED))
2593 if (S_sv_2iuv_common(aTHX_ sv))
2597 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2uv(%" UVuf ")\n",
2598 PTR2UV(sv),SvUVX(sv)));
2599 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
2603 =for apidoc sv_2nv_flags
2605 Return the num value of an SV, doing any necessary string or integer
2606 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2607 Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
2613 Perl_sv_2nv_flags(pTHX_ SV *const sv, const I32 flags)
2615 PERL_ARGS_ASSERT_SV_2NV_FLAGS;
2617 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2618 && SvTYPE(sv) != SVt_PVFM);
2619 if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) {
2620 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2621 the same flag bit as SVf_IVisUV, so must not let them cache NVs.
2622 Regexps have no SvIVX and SvNVX fields. */
2624 if (flags & SV_GMAGIC)
2628 if (SvPOKp(sv) && !SvIOKp(sv)) {
2629 ptr = SvPVX_const(sv);
2630 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
2631 !grok_number(ptr, SvCUR(sv), NULL))
2637 return (NV)SvUVX(sv);
2639 return (NV)SvIVX(sv);
2644 assert(SvTYPE(sv) >= SVt_PVMG);
2645 /* This falls through to the report_uninit near the end of the
2647 } else if (SvTHINKFIRST(sv)) {
2652 if (flags & SV_SKIP_OVERLOAD)
2654 tmpstr = AMG_CALLunary(sv, numer_amg);
2655 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2656 return SvNV(tmpstr);
2659 return PTR2NV(SvRV(sv));
2661 if (SvREADONLY(sv) && !SvOK(sv)) {
2662 if (ckWARN(WARN_UNINITIALIZED))
2667 if (SvTYPE(sv) < SVt_NV) {
2668 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2669 sv_upgrade(sv, SVt_NV);
2671 STORE_LC_NUMERIC_UNDERLYING_SET_STANDARD();
2672 PerlIO_printf(Perl_debug_log,
2673 "0x%" UVxf " num(%" NVgf ")\n",
2674 PTR2UV(sv), SvNVX(sv));
2675 RESTORE_LC_NUMERIC_UNDERLYING();
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. */
2812 STORE_LC_NUMERIC_UNDERLYING_SET_STANDARD();
2813 PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2nv(%" NVgf ")\n",
2814 PTR2UV(sv), SvNVX(sv));
2815 RESTORE_LC_NUMERIC_UNDERLYING();
2823 Return an SV with the numeric value of the source SV, doing any necessary
2824 reference or overload conversion. The caller is expected to have handled
2831 Perl_sv_2num(pTHX_ SV *const sv)
2833 PERL_ARGS_ASSERT_SV_2NUM;
2838 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
2839 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
2840 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2841 return sv_2num(tmpsv);
2843 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2846 /* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2847 * UV as a string towards the end of buf, and return pointers to start and
2850 * We assume that buf is at least TYPE_CHARS(UV) long.
2854 S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
2856 char *ptr = buf + TYPE_CHARS(UV);
2857 char * const ebuf = ptr;
2860 PERL_ARGS_ASSERT_UIV_2BUF;
2868 uv = (iv == IV_MIN) ? (UV)iv : (UV)(-iv);
2872 *--ptr = '0' + (char)(uv % 10);
2880 /* Helper for sv_2pv_flags and sv_vcatpvfn_flags. If the NV is an
2881 * infinity or a not-a-number, writes the appropriate strings to the
2882 * buffer, including a zero byte. On success returns the written length,
2883 * excluding the zero byte, on failure (not an infinity, not a nan)
2884 * returns zero, assert-fails on maxlen being too short.
2886 * XXX for "Inf", "-Inf", and "NaN", we could have three read-only
2887 * shared string constants we point to, instead of generating a new
2888 * string for each instance. */
2890 S_infnan_2pv(NV nv, char* buffer, size_t maxlen, char plus) {
2892 assert(maxlen >= 4);
2893 if (Perl_isinf(nv)) {
2895 if (maxlen < 5) /* "-Inf\0" */
2905 else if (Perl_isnan(nv)) {
2909 /* XXX optionally output the payload mantissa bits as
2910 * "(unsigned)" (to match the nan("...") C99 function,
2911 * or maybe as "(0xhhh...)" would make more sense...
2912 * provide a format string so that the user can decide?
2913 * NOTE: would affect the maxlen and assert() logic.*/
2918 assert((s == buffer + 3) || (s == buffer + 4));
2924 =for apidoc sv_2pv_flags
2926 Returns a pointer to the string value of an SV, and sets C<*lp> to its length.
2927 If flags has the C<SV_GMAGIC> bit set, does an C<mg_get()> first. Coerces C<sv> to a
2928 string if necessary. Normally invoked via the C<SvPV_flags> macro.
2929 C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
2935 Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
2939 PERL_ARGS_ASSERT_SV_2PV_FLAGS;
2941 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2942 && SvTYPE(sv) != SVt_PVFM);
2943 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2948 if (flags & SV_SKIP_OVERLOAD)
2950 tmpstr = AMG_CALLunary(sv, string_amg);
2951 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
2952 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2954 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2958 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2959 if (flags & SV_CONST_RETURN) {
2960 pv = (char *) SvPVX_const(tmpstr);
2962 pv = (flags & SV_MUTABLE_RETURN)
2963 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
2966 *lp = SvCUR(tmpstr);
2968 pv = sv_2pv_flags(tmpstr, lp, flags);
2981 SV *const referent = SvRV(sv);
2985 retval = buffer = savepvn("NULLREF", len);
2986 } else if (SvTYPE(referent) == SVt_REGEXP &&
2987 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
2988 amagic_is_enabled(string_amg))) {
2989 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
2993 /* If the regex is UTF-8 we want the containing scalar to
2994 have an UTF-8 flag too */
3001 *lp = RX_WRAPLEN(re);
3003 return RX_WRAPPED(re);
3005 const char *const typestr = sv_reftype(referent, 0);
3006 const STRLEN typelen = strlen(typestr);
3007 UV addr = PTR2UV(referent);
3008 const char *stashname = NULL;
3009 STRLEN stashnamelen = 0; /* hush, gcc */
3010 const char *buffer_end;
3012 if (SvOBJECT(referent)) {
3013 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
3016 stashname = HEK_KEY(name);
3017 stashnamelen = HEK_LEN(name);
3019 if (HEK_UTF8(name)) {
3025 stashname = "__ANON__";
3028 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
3029 + 2 * sizeof(UV) + 2 /* )\0 */;
3031 len = typelen + 3 /* (0x */
3032 + 2 * sizeof(UV) + 2 /* )\0 */;
3035 Newx(buffer, len, char);
3036 buffer_end = retval = buffer + len;
3038 /* Working backwards */
3042 *--retval = PL_hexdigit[addr & 15];
3043 } while (addr >>= 4);
3049 memcpy(retval, typestr, typelen);
3053 retval -= stashnamelen;
3054 memcpy(retval, stashname, stashnamelen);
3056 /* retval may not necessarily have reached the start of the
3058 assert (retval >= buffer);
3060 len = buffer_end - retval - 1; /* -1 for that \0 */
3072 if (flags & SV_MUTABLE_RETURN)
3073 return SvPVX_mutable(sv);
3074 if (flags & SV_CONST_RETURN)
3075 return (char *)SvPVX_const(sv);
3080 /* I'm assuming that if both IV and NV are equally valid then
3081 converting the IV is going to be more efficient */
3082 const U32 isUIOK = SvIsUV(sv);
3083 char buf[TYPE_CHARS(UV)];
3087 if (SvTYPE(sv) < SVt_PVIV)
3088 sv_upgrade(sv, SVt_PVIV);
3089 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
3091 /* inlined from sv_setpvn */
3092 s = SvGROW_mutable(sv, len + 1);
3093 Move(ptr, s, len, char);
3098 else if (SvNOK(sv)) {
3099 if (SvTYPE(sv) < SVt_PVNV)
3100 sv_upgrade(sv, SVt_PVNV);
3101 if (SvNVX(sv) == 0.0
3102 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
3103 && !Perl_isnan(SvNVX(sv))
3106 s = SvGROW_mutable(sv, 2);
3111 STRLEN size = 5; /* "-Inf\0" */
3113 s = SvGROW_mutable(sv, size);
3114 len = S_infnan_2pv(SvNVX(sv), s, size, 0);
3120 /* some Xenix systems wipe out errno here */
3129 5 + /* exponent digits */
3133 s = SvGROW_mutable(sv, size);
3134 #ifndef USE_LOCALE_NUMERIC
3135 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3141 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
3142 STORE_LC_NUMERIC_SET_TO_NEEDED();
3144 local_radix = PL_numeric_local && PL_numeric_radix_sv;
3145 if (local_radix && SvCUR(PL_numeric_radix_sv) > 1) {
3146 size += SvCUR(PL_numeric_radix_sv) - 1;
3147 s = SvGROW_mutable(sv, size);
3150 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3152 /* If the radix character is UTF-8, and actually is in the
3153 * output, turn on the UTF-8 flag for the scalar */
3155 && SvUTF8(PL_numeric_radix_sv)
3156 && instr(s, SvPVX_const(PL_numeric_radix_sv)))
3161 RESTORE_LC_NUMERIC();
3164 /* We don't call SvPOK_on(), because it may come to
3165 * pass that the locale changes so that the
3166 * stringification we just did is no longer correct. We
3167 * will have to re-stringify every time it is needed */
3174 else if (isGV_with_GP(sv)) {
3175 GV *const gv = MUTABLE_GV(sv);
3176 SV *const buffer = sv_newmortal();
3178 gv_efullname3(buffer, gv, "*");
3180 assert(SvPOK(buffer));
3186 *lp = SvCUR(buffer);
3187 return SvPVX(buffer);
3192 if (flags & SV_UNDEF_RETURNS_NULL)
3194 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
3196 /* Typically the caller expects that sv_any is not NULL now. */
3197 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
3198 sv_upgrade(sv, SVt_PV);
3203 const STRLEN len = s - SvPVX_const(sv);
3208 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2pv(%s)\n",
3209 PTR2UV(sv),SvPVX_const(sv)));
3210 if (flags & SV_CONST_RETURN)
3211 return (char *)SvPVX_const(sv);
3212 if (flags & SV_MUTABLE_RETURN)
3213 return SvPVX_mutable(sv);
3218 =for apidoc sv_copypv
3220 Copies a stringified representation of the source SV into the
3221 destination SV. Automatically performs any necessary C<mg_get> and
3222 coercion of numeric values into strings. Guaranteed to preserve
3223 C<UTF8> flag even from overloaded objects. Similar in nature to
3224 C<sv_2pv[_flags]> but operates directly on an SV instead of just the
3225 string. Mostly uses C<sv_2pv_flags> to do its work, except when that
3226 would lose the UTF-8'ness of the PV.
3228 =for apidoc sv_copypv_nomg
3230 Like C<sv_copypv>, but doesn't invoke get magic first.
3232 =for apidoc sv_copypv_flags
3234 Implementation of C<sv_copypv> and C<sv_copypv_nomg>. Calls get magic iff flags
3235 has the C<SV_GMAGIC> bit set.
3241 Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
3246 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
3248 s = SvPV_flags_const(ssv,len,(flags & SV_GMAGIC));
3249 sv_setpvn(dsv,s,len);
3257 =for apidoc sv_2pvbyte
3259 Return a pointer to the byte-encoded representation of the SV, and set C<*lp>
3260 to its length. May cause the SV to be downgraded from UTF-8 as a
3263 Usually accessed via the C<SvPVbyte> macro.
3269 Perl_sv_2pvbyte(pTHX_ SV *sv, STRLEN *const lp)
3271 PERL_ARGS_ASSERT_SV_2PVBYTE;
3274 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3275 || isGV_with_GP(sv) || SvROK(sv)) {
3276 SV *sv2 = sv_newmortal();
3277 sv_copypv_nomg(sv2,sv);
3280 sv_utf8_downgrade(sv,0);
3281 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3285 =for apidoc sv_2pvutf8
3287 Return a pointer to the UTF-8-encoded representation of the SV, and set C<*lp>
3288 to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3290 Usually accessed via the C<SvPVutf8> macro.
3296 Perl_sv_2pvutf8(pTHX_ SV *sv, STRLEN *const lp)
3298 PERL_ARGS_ASSERT_SV_2PVUTF8;
3300 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3301 || isGV_with_GP(sv) || SvROK(sv))
3302 sv = sv_mortalcopy(sv);
3305 sv_utf8_upgrade_nomg(sv);
3306 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3311 =for apidoc sv_2bool
3313 This macro is only used by C<sv_true()> or its macro equivalent, and only if
3314 the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>.
3315 It calls C<sv_2bool_flags> with the C<SV_GMAGIC> flag.
3317 =for apidoc sv_2bool_flags
3319 This function is only used by C<sv_true()> and friends, and only if
3320 the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>. If the flags
3321 contain C<SV_GMAGIC>, then it does an C<mg_get()> first.
3328 Perl_sv_2bool_flags(pTHX_ SV *sv, I32 flags)
3330 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
3333 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
3339 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
3340 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv)))) {
3343 if(SvGMAGICAL(sv)) {
3345 goto restart; /* call sv_2bool */
3347 /* expanded SvTRUE_common(sv, (flags = 0, goto restart)) */
3348 else if(!SvOK(sv)) {
3351 else if(SvPOK(sv)) {
3352 svb = SvPVXtrue(sv);
3354 else if((SvFLAGS(sv) & (SVf_IOK|SVf_NOK))) {
3355 svb = (SvIOK(sv) && SvIVX(sv) != 0)
3356 || (SvNOK(sv) && SvNVX(sv) != 0.0);
3360 goto restart; /* call sv_2bool_nomg */
3370 RX_WRAPLEN(sv) > 1 || (RX_WRAPLEN(sv) && *RX_WRAPPED(sv) != '0');
3372 if (SvNOK(sv) && !SvPOK(sv))
3373 return SvNVX(sv) != 0.0;
3375 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
3379 =for apidoc sv_utf8_upgrade
3381 Converts the PV of an SV to its UTF-8-encoded form.
3382 Forces the SV to string form if it is not already.
3383 Will C<mg_get> on C<sv> if appropriate.
3384 Always sets the C<SvUTF8> flag to avoid future validity checks even
3385 if the whole string is the same in UTF-8 as not.
3386 Returns the number of bytes in the converted string
3388 This is not a general purpose byte encoding to Unicode interface:
3389 use the Encode extension for that.
3391 =for apidoc sv_utf8_upgrade_nomg
3393 Like C<sv_utf8_upgrade>, but doesn't do magic on C<sv>.
3395 =for apidoc sv_utf8_upgrade_flags
3397 Converts the PV of an SV to its UTF-8-encoded form.
3398 Forces the SV to string form if it is not already.
3399 Always sets the SvUTF8 flag to avoid future validity checks even
3400 if all the bytes are invariant in UTF-8.
3401 If C<flags> has C<SV_GMAGIC> bit set,
3402 will C<mg_get> on C<sv> if appropriate, else not.
3404 If C<flags> has C<SV_FORCE_UTF8_UPGRADE> set, this function assumes that the PV
3405 will expand when converted to UTF-8, and skips the extra work of checking for
3406 that. Typically this flag is used by a routine that has already parsed the
3407 string and found such characters, and passes this information on so that the
3408 work doesn't have to be repeated.
3410 Returns the number of bytes in the converted string.
3412 This is not a general purpose byte encoding to Unicode interface:
3413 use the Encode extension for that.
3415 =for apidoc sv_utf8_upgrade_flags_grow
3417 Like C<sv_utf8_upgrade_flags>, but has an additional parameter C<extra>, which is
3418 the number of unused bytes the string of C<sv> is guaranteed to have free after
3419 it upon return. This allows the caller to reserve extra space that it intends
3420 to fill, to avoid extra grows.
3422 C<sv_utf8_upgrade>, C<sv_utf8_upgrade_nomg>, and C<sv_utf8_upgrade_flags>
3423 are implemented in terms of this function.
3425 Returns the number of bytes in the converted string (not including the spares).
3429 (One might think that the calling routine could pass in the position of the
3430 first variant character when it has set SV_FORCE_UTF8_UPGRADE, so it wouldn't
3431 have to be found again. But that is not the case, because typically when the
3432 caller is likely to use this flag, it won't be calling this routine unless it
3433 finds something that won't fit into a byte. Otherwise it tries to not upgrade
3434 and just use bytes. But some things that do fit into a byte are variants in
3435 utf8, and the caller may not have been keeping track of these.)
3437 If the routine itself changes the string, it adds a trailing C<NUL>. Such a
3438 C<NUL> isn't guaranteed due to having other routines do the work in some input
3439 cases, or if the input is already flagged as being in utf8.
3441 The speed of this could perhaps be improved for many cases if someone wanted to
3442 write a fast function that counts the number of variant characters in a string,
3443 especially if it could return the position of the first one.
3448 Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV *const sv, const I32 flags, STRLEN extra)
3450 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
3452 if (sv == &PL_sv_undef)
3454 if (!SvPOK_nog(sv)) {
3456 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3457 (void) sv_2pv_flags(sv,&len, flags);
3459 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3463 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
3467 /* SVt_REGEXP's shouldn't be upgraded to UTF8 - they're already
3468 * compiled and individual nodes will remain non-utf8 even if the
3469 * stringified version of the pattern gets upgraded. Whether the
3470 * PVX of a REGEXP should be grown or we should just croak, I don't
3472 if (SvUTF8(sv) || isREGEXP(sv)) {
3473 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3478 S_sv_uncow(aTHX_ sv, 0);
3481 if (SvCUR(sv) == 0) {
3482 if (extra) SvGROW(sv, extra);
3483 } else { /* Assume Latin-1/EBCDIC */
3484 /* This function could be much more efficient if we
3485 * had a FLAG in SVs to signal if there are any variant
3486 * chars in the PV. Given that there isn't such a flag
3487 * make the loop as fast as possible (although there are certainly ways
3488 * to speed this up, eg. through vectorization) */
3489 U8 * s = (U8 *) SvPVX_const(sv);
3490 U8 * e = (U8 *) SvEND(sv);
3492 STRLEN two_byte_count;
3494 if (flags & SV_FORCE_UTF8_UPGRADE) {
3498 if (is_utf8_invariant_string_loc(s, SvCUR(sv), (const U8 **) &t)) {
3500 /* utf8 conversion not needed because all are invariants. Mark
3501 * as UTF-8 even if no variant - saves scanning loop */
3503 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3507 /* Here, there is at least one variant, and t points to the first
3512 /* Note that the incoming SV may not have a trailing '\0', as certain
3513 * code in pp_formline can send us partially built SVs.
3515 * Here, the string should be converted to utf8, either because of an
3516 * input flag (which causes two_byte_count to be set to 0), or because
3517 * a character that requires 2 bytes was found (two_byte_count = 1). t
3518 * points either to the beginning of the string (if we didn't examine
3519 * anything), or to the first variant. In either case, everything from
3520 * s to t - 1 will occupy only 1 byte each on output.
3522 * There are two main ways to convert. One is to create a new string
3523 * and go through the input starting from the beginning, appending each
3524 * converted value onto the new string as we go along. It's probably
3525 * best to allocate enough space in the string for the worst possible
3526 * case rather than possibly running out of space and having to
3527 * reallocate and then copy what we've done so far. Since everything
3528 * from s to t - 1 is invariant, the destination can be initialized
3529 * with these using a fast memory copy
3531 * The other way is to figure out exactly how big the string should be,
3532 * by parsing the entire input. Then you don't have to make it big
3533 * enough to handle the worst possible case, and more importantly, if
3534 * the string you already have is large enough, you don't have to
3535 * allocate a new string, you can copy the last character in the input
3536 * string to the final position(s) that will be occupied by the
3537 * converted string and go backwards, stopping at t, since everything
3538 * before that is invariant.
3540 * There are advantages and disadvantages to each method.
3542 * In the first method, we can allocate a new string, do the memory
3543 * copy from the s to t - 1, and then proceed through the rest of the
3544 * string byte-by-byte.
3546 * In the second method, we proceed through the rest of the input
3547 * string just calculating how big the converted string will be. Then
3548 * there are two cases:
3549 * 1) if the string has enough extra space to handle the converted
3550 * value. We go backwards through the string, converting until we
3551 * get to the position we are at now, and then stop. If this
3552 * position is far enough along in the string, this method is
3553 * faster than the first method above. If the memory copy were
3554 * the same speed as the byte-by-byte loop, that position would be
3555 * about half-way, as at the half-way mark, parsing to the end and
3556 * back is one complete string's parse, the same amount as
3557 * starting over and going all the way through. Actually, it
3558 * would be somewhat less than half-way, as it's faster to just
3559 * count bytes than to also copy, and we don't have the overhead
3560 * of allocating a new string, changing the scalar to use it, and
3561 * freeing the existing one. But if the memory copy is fast, the
3562 * break-even point is somewhere after half way. The counting
3563 * loop could be sped up by vectorization, etc, to move the
3564 * break-even point further towards the beginning.
3565 * 2) if the string doesn't have enough space to handle the converted
3566 * value. A new string will have to be allocated, and one might
3567 * as well, given that, start from the beginning doing the first
3568 * method. We've spent extra time parsing the string and in
3569 * exchange all we've gotten is that we know precisely how big to
3570 * make the new one. Perl is more optimized for time than space,
3571 * so this case is a loser.
3572 * So what I've decided to do is not use the 2nd method unless it is
3573 * guaranteed that a new string won't have to be allocated, assuming
3574 * the worst case. I also decided not to put any more conditions on it
3575 * than this, for now. It seems likely that, since the worst case is
3576 * twice as big as the unknown portion of the string (plus 1), we won't
3577 * be guaranteed enough space, causing us to go to the first method,
3578 * unless the string is short, or the first variant character is near
3579 * the end of it. In either of these cases, it seems best to use the
3580 * 2nd method. The only circumstance I can think of where this would
3581 * be really slower is if the string had once had much more data in it
3582 * than it does now, but there is still a substantial amount in it */
3585 STRLEN invariant_head = t - s;
3586 STRLEN size = invariant_head + (e - t) * 2 + 1 + extra;
3587 if (SvLEN(sv) < size) {
3589 /* Here, have decided to allocate a new string */
3594 Newx(dst, size, U8);
3596 /* If no known invariants at the beginning of the input string,
3597 * set so starts from there. Otherwise, can use memory copy to
3598 * get up to where we are now, and then start from here */
3600 if (invariant_head == 0) {
3603 Copy(s, dst, invariant_head, char);
3604 d = dst + invariant_head;
3608 append_utf8_from_native_byte(*t, &d);
3612 SvPV_free(sv); /* No longer using pre-existing string */
3613 SvPV_set(sv, (char*)dst);
3614 SvCUR_set(sv, d - dst);
3615 SvLEN_set(sv, size);
3618 /* Here, have decided to get the exact size of the string.
3619 * Currently this happens only when we know that there is
3620 * guaranteed enough space to fit the converted string, so
3621 * don't have to worry about growing. If two_byte_count is 0,
3622 * then t points to the first byte of the string which hasn't
3623 * been examined yet. Otherwise two_byte_count is 1, and t
3624 * points to the first byte in the string that will expand to
3625 * two. Depending on this, start examining at t or 1 after t.
3628 U8 *d = t + two_byte_count;
3631 /* Count up the remaining bytes that expand to two */
3634 const U8 chr = *d++;
3635 if (! NATIVE_BYTE_IS_INVARIANT(chr)) two_byte_count++;
3638 /* The string will expand by just the number of bytes that
3639 * occupy two positions. But we are one afterwards because of
3640 * the increment just above. This is the place to put the
3641 * trailing NUL, and to set the length before we decrement */
3643 d += two_byte_count;
3644 SvCUR_set(sv, d - s);
3648 /* Having decremented d, it points to the position to put the
3649 * very last byte of the expanded string. Go backwards through
3650 * the string, copying and expanding as we go, stopping when we
3651 * get to the part that is invariant the rest of the way down */
3655 if (NATIVE_BYTE_IS_INVARIANT(*e)) {
3658 *d-- = UTF8_EIGHT_BIT_LO(*e);
3659 *d-- = UTF8_EIGHT_BIT_HI(*e);
3665 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3666 /* Update pos. We do it at the end rather than during
3667 * the upgrade, to avoid slowing down the common case
3668 * (upgrade without pos).
3669 * pos can be stored as either bytes or characters. Since
3670 * this was previously a byte string we can just turn off
3671 * the bytes flag. */
3672 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3674 mg->mg_flags &= ~MGf_BYTES;
3676 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3677 magic_setutf8(sv,mg); /* clear UTF8 cache */
3682 /* Mark as UTF-8 even if no variant - saves scanning loop */
3688 =for apidoc sv_utf8_downgrade
3690 Attempts to convert the PV of an SV from characters to bytes.
3691 If the PV contains a character that cannot fit
3692 in a byte, this conversion will fail;
3693 in this case, either returns false or, if C<fail_ok> is not
3696 This is not a general purpose Unicode to byte encoding interface:
3697 use the C<Encode> extension for that.
3703 Perl_sv_utf8_downgrade(pTHX_ SV *const sv, const bool fail_ok)
3705 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3707 if (SvPOKp(sv) && SvUTF8(sv)) {
3711 int mg_flags = SV_GMAGIC;
3714 S_sv_uncow(aTHX_ sv, 0);
3716 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3718 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3719 if (mg && mg->mg_len > 0 && mg->mg_flags & MGf_BYTES) {
3720 mg->mg_len = sv_pos_b2u_flags(sv, mg->mg_len,
3721 SV_GMAGIC|SV_CONST_RETURN);
3722 mg_flags = 0; /* sv_pos_b2u does get magic */
3724 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3725 magic_setutf8(sv,mg); /* clear UTF8 cache */
3728 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3730 if (!utf8_to_bytes(s, &len)) {
3735 Perl_croak(aTHX_ "Wide character in %s",
3738 Perl_croak(aTHX_ "Wide character");
3749 =for apidoc sv_utf8_encode
3751 Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3752 flag off so that it looks like octets again.
3758 Perl_sv_utf8_encode(pTHX_ SV *const sv)
3760 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3762 if (SvREADONLY(sv)) {
3763 sv_force_normal_flags(sv, 0);
3765 (void) sv_utf8_upgrade(sv);
3770 =for apidoc sv_utf8_decode
3772 If the PV of the SV is an octet sequence in Perl's extended UTF-8
3773 and contains a multiple-byte character, the C<SvUTF8> flag is turned on
3774 so that it looks like a character. If the PV contains only single-byte
3775 characters, the C<SvUTF8> flag stays off.
3776 Scans PV for validity and returns FALSE if the PV is invalid UTF-8.
3782 Perl_sv_utf8_decode(pTHX_ SV *const sv)
3784 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3787 const U8 *start, *c;
3789 /* The octets may have got themselves encoded - get them back as
3792 if (!sv_utf8_downgrade(sv, TRUE))
3795 /* it is actually just a matter of turning the utf8 flag on, but
3796 * we want to make sure everything inside is valid utf8 first.
3798 c = start = (const U8 *) SvPVX_const(sv);
3799 if (!is_utf8_string(c, SvCUR(sv)))
3801 if (! is_utf8_invariant_string(c, SvCUR(sv))) {
3804 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3805 /* XXX Is this dead code? XS_utf8_decode calls SvSETMAGIC
3806 after this, clearing pos. Does anything on CPAN
3808 /* adjust pos to the start of a UTF8 char sequence */
3809 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3811 I32 pos = mg->mg_len;
3813 for (c = start + pos; c > start; c--) {
3814 if (UTF8_IS_START(*c))
3817 mg->mg_len = c - start;
3820 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3821 magic_setutf8(sv,mg); /* clear UTF8 cache */
3828 =for apidoc sv_setsv
3830 Copies the contents of the source SV C<ssv> into the destination SV
3831 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3832 function if the source SV needs to be reused. Does not handle 'set' magic on
3833 destination SV. Calls 'get' magic on source SV. Loosely speaking, it
3834 performs a copy-by-value, obliterating any previous content of the
3837 You probably want to use one of the assortment of wrappers, such as
3838 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3839 C<SvSetMagicSV_nosteal>.
3841 =for apidoc sv_setsv_flags
3843 Copies the contents of the source SV C<ssv> into the destination SV
3844 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3845 function if the source SV needs to be reused. Does not handle 'set' magic.
3846 Loosely speaking, it performs a copy-by-value, obliterating any previous
3847 content of the destination.
3848 If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
3849 C<ssv> if appropriate, else not. If the C<flags>
3850 parameter has the C<SV_NOSTEAL> bit set then the
3851 buffers of temps will not be stolen. C<sv_setsv>
3852 and C<sv_setsv_nomg> are implemented in terms of this function.
3854 You probably want to use one of the assortment of wrappers, such as
3855 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3856 C<SvSetMagicSV_nosteal>.
3858 This is the primary function for copying scalars, and most other
3859 copy-ish functions and macros use this underneath.
3865 S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
3867 I32 mro_changes = 0; /* 1 = method, 2 = isa, 3 = recursive isa */
3868 HV *old_stash = NULL;
3870 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB;
3872 if (dtype != SVt_PVGV && !isGV_with_GP(dstr)) {
3873 const char * const name = GvNAME(sstr);
3874 const STRLEN len = GvNAMELEN(sstr);
3876 if (dtype >= SVt_PV) {
3882 SvUPGRADE(dstr, SVt_PVGV);
3883 (void)SvOK_off(dstr);
3884 isGV_with_GP_on(dstr);
3886 GvSTASH(dstr) = GvSTASH(sstr);
3888 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
3889 gv_name_set(MUTABLE_GV(dstr), name, len,
3890 GV_ADD | (GvNAMEUTF8(sstr) ? SVf_UTF8 : 0 ));
3891 SvFAKE_on(dstr); /* can coerce to non-glob */
3894 if(GvGP(MUTABLE_GV(sstr))) {
3895 /* If source has method cache entry, clear it */
3897 SvREFCNT_dec(GvCV(sstr));
3898 GvCV_set(sstr, NULL);
3901 /* If source has a real method, then a method is
3904 GvCV((const GV *)sstr) && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3910 /* If dest already had a real method, that's a change as well */
3912 !mro_changes && GvGP(MUTABLE_GV(dstr)) && GvCVu((const GV *)dstr)
3913 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3918 /* We don't need to check the name of the destination if it was not a
3919 glob to begin with. */
3920 if(dtype == SVt_PVGV) {
3921 const char * const name = GvNAME((const GV *)dstr);
3924 /* The stash may have been detached from the symbol table, so
3926 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3930 const STRLEN len = GvNAMELEN(dstr);
3931 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
3932 || (len == 1 && name[0] == ':')) {
3935 /* Set aside the old stash, so we can reset isa caches on
3937 if((old_stash = GvHV(dstr)))
3938 /* Make sure we do not lose it early. */
3939 SvREFCNT_inc_simple_void_NN(
3940 sv_2mortal((SV *)old_stash)
3945 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
3948 /* freeing dstr's GP might free sstr (e.g. *x = $x),
3949 * so temporarily protect it */
3951 SAVEFREESV(SvREFCNT_inc_simple_NN(sstr));
3952 gp_free(MUTABLE_GV(dstr));
3953 GvINTRO_off(dstr); /* one-shot flag */
3954 GvGP_set(dstr, gp_ref(GvGP(sstr)));
3957 if (SvTAINTED(sstr))
3959 if (GvIMPORTED(dstr) != GVf_IMPORTED
3960 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
3962 GvIMPORTED_on(dstr);
3965 if(mro_changes == 2) {
3966 if (GvAV((const GV *)sstr)) {
3968 SV * const sref = (SV *)GvAV((const GV *)dstr);
3969 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
3970 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
3971 AV * const ary = newAV();
3972 av_push(ary, mg->mg_obj); /* takes the refcount */
3973 mg->mg_obj = (SV *)ary;
3975 av_push((AV *)mg->mg_obj, SvREFCNT_inc_simple_NN(dstr));
3977 else sv_magic(sref, dstr, PERL_MAGIC_isa, NULL, 0);
3979 mro_isa_changed_in(GvSTASH(dstr));
3981 else if(mro_changes == 3) {
3982 HV * const stash = GvHV(dstr);
3983 if(old_stash ? (HV *)HvENAME_get(old_stash) : stash)
3989 else if(mro_changes) mro_method_changed_in(GvSTASH(dstr));
3990 if (GvIO(dstr) && dtype == SVt_PVGV) {
3991 DEBUG_o(Perl_deb(aTHX_
3992 "glob_assign_glob clearing PL_stashcache\n"));
3993 /* It's a cache. It will rebuild itself quite happily.
3994 It's a lot of effort to work out exactly which key (or keys)
3995 might be invalidated by the creation of the this file handle.
3997 hv_clear(PL_stashcache);
4003 Perl_gv_setref(pTHX_ SV *const dstr, SV *const sstr)
4005 SV * const sref = SvRV(sstr);
4007 const int intro = GvINTRO(dstr);
4010 const U32 stype = SvTYPE(sref);
4012 PERL_ARGS_ASSERT_GV_SETREF;
4015 GvINTRO_off(dstr); /* one-shot flag */
4016 GvLINE(dstr) = CopLINE(PL_curcop);
4017 GvEGV(dstr) = MUTABLE_GV(dstr);
4022 location = (SV **) &(GvGP(dstr)->gp_cv); /* XXX bypassing GvCV_set */
4023 import_flag = GVf_IMPORTED_CV;
4026 location = (SV **) &GvHV(dstr);
4027 import_flag = GVf_IMPORTED_HV;
4030 location = (SV **) &GvAV(dstr);
4031 import_flag = GVf_IMPORTED_AV;
4034 location = (SV **) &GvIOp(dstr);
4037 location = (SV **) &GvFORM(dstr);
4040 location = &GvSV(dstr);
4041 import_flag = GVf_IMPORTED_SV;
4044 if (stype == SVt_PVCV) {
4045 /*if (GvCVGEN(dstr) && (GvCV(dstr) != (const CV *)sref || GvCVGEN(dstr))) {*/
4046 if (GvCVGEN(dstr)) {
4047 SvREFCNT_dec(GvCV(dstr));
4048 GvCV_set(dstr, NULL);
4049 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4052 /* SAVEt_GVSLOT takes more room on the savestack and has more
4053 overhead in leave_scope than SAVEt_GENERIC_SV. But for CVs
4054 leave_scope needs access to the GV so it can reset method
4055 caches. We must use SAVEt_GVSLOT whenever the type is
4056 SVt_PVCV, even if the stash is anonymous, as the stash may
4057 gain a name somehow before leave_scope. */
4058 if (stype == SVt_PVCV) {
4059 /* There is no save_pushptrptrptr. Creating it for this
4060 one call site would be overkill. So inline the ss add
4064 SS_ADD_PTR(location);
4065 SS_ADD_PTR(SvREFCNT_inc(*location));
4066 SS_ADD_UV(SAVEt_GVSLOT);
4069 else SAVEGENERICSV(*location);
4072 if (stype == SVt_PVCV && (*location != sref || GvCVGEN(dstr))) {
4073 CV* const cv = MUTABLE_CV(*location);
4075 if (!GvCVGEN((const GV *)dstr) &&
4076 (CvROOT(cv) || CvXSUB(cv)) &&
4077 /* redundant check that avoids creating the extra SV
4078 most of the time: */
4079 (CvCONST(cv) || ckWARN(WARN_REDEFINE)))
4081 SV * const new_const_sv =
4082 CvCONST((const CV *)sref)
4083 ? cv_const_sv((const CV *)sref)
4085 HV * const stash = GvSTASH((const GV *)dstr);
4086 report_redefined_cv(
4089 ? Perl_newSVpvf(aTHX_
4090 "%" HEKf "::%" HEKf,
4091 HEKfARG(HvNAME_HEK(stash)),
4092 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr))))
4093 : Perl_newSVpvf(aTHX_
4095 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr))))
4098 CvCONST((const CV *)sref) ? &new_const_sv : NULL
4102 cv_ckproto_len_flags(cv, (const GV *)dstr,
4103 SvPOK(sref) ? CvPROTO(sref) : NULL,
4104 SvPOK(sref) ? CvPROTOLEN(sref) : 0,
4105 SvPOK(sref) ? SvUTF8(sref) : 0);
4107 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4108 GvASSUMECV_on(dstr);
4109 if(GvSTASH(dstr)) { /* sub foo { 1 } sub bar { 2 } *bar = \&foo */
4110 if (intro && GvREFCNT(dstr) > 1) {
4111 /* temporary remove extra savestack's ref */
4113 gv_method_changed(dstr);
4116 else gv_method_changed(dstr);
4119 *location = SvREFCNT_inc_simple_NN(sref);
4120 if (import_flag && !(GvFLAGS(dstr) & import_flag)
4121 && CopSTASH_ne(PL_curcop, GvSTASH(dstr))) {
4122 GvFLAGS(dstr) |= import_flag;
4125 if (stype == SVt_PVHV) {
4126 const char * const name = GvNAME((GV*)dstr);
4127 const STRLEN len = GvNAMELEN(dstr);
4130 (len > 1 && name[len-2] == ':' && name[len-1] == ':')
4131 || (len == 1 && name[0] == ':')
4133 && (!dref || HvENAME_get(dref))
4136 (HV *)sref, (HV *)dref,
4142 stype == SVt_PVAV && sref != dref
4143 && strEQ(GvNAME((GV*)dstr), "ISA")
4144 /* The stash may have been detached from the symbol table, so
4145 check its name before doing anything. */
4146 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
4149 MAGIC * const omg = dref && SvSMAGICAL(dref)
4150 ? mg_find(dref, PERL_MAGIC_isa)
4152 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
4153 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
4154 AV * const ary = newAV();
4155 av_push(ary, mg->mg_obj); /* takes the refcount */
4156 mg->mg_obj = (SV *)ary;
4159 if (SvTYPE(omg->mg_obj) == SVt_PVAV) {
4160 SV **svp = AvARRAY((AV *)omg->mg_obj);
4161 I32 items = AvFILLp((AV *)omg->mg_obj) + 1;
4165 SvREFCNT_inc_simple_NN(*svp++)
4171 SvREFCNT_inc_simple_NN(omg->mg_obj)
4175 av_push((AV *)mg->mg_obj,SvREFCNT_inc_simple_NN(dstr));
4181 sref, omg ? omg->mg_obj : dstr, PERL_MAGIC_isa, NULL, 0
4183 for (i = 0; i <= AvFILL(sref); ++i) {
4184 SV **elem = av_fetch ((AV*)sref, i, 0);
4187 *elem, sref, PERL_MAGIC_isaelem, NULL, i
4191 mg = mg_find(sref, PERL_MAGIC_isa);
4193 /* Since the *ISA assignment could have affected more than
4194 one stash, don't call mro_isa_changed_in directly, but let
4195 magic_clearisa do it for us, as it already has the logic for
4196 dealing with globs vs arrays of globs. */
4198 Perl_magic_clearisa(aTHX_ NULL, mg);
4200 else if (stype == SVt_PVIO) {
4201 DEBUG_o(Perl_deb(aTHX_ "gv_setref clearing PL_stashcache\n"));
4202 /* It's a cache. It will rebuild itself quite happily.
4203 It's a lot of effort to work out exactly which key (or keys)
4204 might be invalidated by the creation of the this file handle.
4206 hv_clear(PL_stashcache);
4210 if (!intro) SvREFCNT_dec(dref);
4211 if (SvTAINTED(sstr))
4219 #ifdef PERL_DEBUG_READONLY_COW
4220 # include <sys/mman.h>
4222 # ifndef PERL_MEMORY_DEBUG_HEADER_SIZE
4223 # define PERL_MEMORY_DEBUG_HEADER_SIZE 0
4227 Perl_sv_buf_to_ro(pTHX_ SV *sv)
4229 struct perl_memory_debug_header * const header =
4230 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4231 const MEM_SIZE len = header->size;
4232 PERL_ARGS_ASSERT_SV_BUF_TO_RO;
4233 # ifdef PERL_TRACK_MEMPOOL
4234 if (!header->readonly) header->readonly = 1;
4236 if (mprotect(header, len, PROT_READ))
4237 Perl_warn(aTHX_ "mprotect RW for COW string %p %lu failed with %d",
4238 header, len, errno);
4242 S_sv_buf_to_rw(pTHX_ SV *sv)
4244 struct perl_memory_debug_header * const header =
4245 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4246 const MEM_SIZE len = header->size;
4247 PERL_ARGS_ASSERT_SV_BUF_TO_RW;
4248 if (mprotect(header, len, PROT_READ|PROT_WRITE))
4249 Perl_warn(aTHX_ "mprotect for COW string %p %lu failed with %d",
4250 header, len, errno);
4251 # ifdef PERL_TRACK_MEMPOOL
4252 header->readonly = 0;
4257 # define sv_buf_to_ro(sv) NOOP
4258 # define sv_buf_to_rw(sv) NOOP
4262 Perl_sv_setsv_flags(pTHX_ SV *dstr, SV* sstr, const I32 flags)
4267 unsigned int both_type;
4269 PERL_ARGS_ASSERT_SV_SETSV_FLAGS;
4271 if (UNLIKELY( sstr == dstr ))
4274 if (UNLIKELY( !sstr ))
4275 sstr = &PL_sv_undef;
4277 stype = SvTYPE(sstr);
4278 dtype = SvTYPE(dstr);
4279 both_type = (stype | dtype);
4281 /* with these values, we can check that both SVs are NULL/IV (and not
4282 * freed) just by testing the or'ed types */
4283 STATIC_ASSERT_STMT(SVt_NULL == 0);
4284 STATIC_ASSERT_STMT(SVt_IV == 1);
4285 if (both_type <= 1) {
4286 /* both src and dst are UNDEF/IV/RV, so we can do a lot of
4292 /* minimal subset of SV_CHECK_THINKFIRST_COW_DROP(dstr) */
4293 if (SvREADONLY(dstr))
4294 Perl_croak_no_modify();
4296 if (SvWEAKREF(dstr))
4297 sv_unref_flags(dstr, 0);
4299 old_rv = SvRV(dstr);
4302 assert(!SvGMAGICAL(sstr));
4303 assert(!SvGMAGICAL(dstr));
4305 sflags = SvFLAGS(sstr);
4306 if (sflags & (SVf_IOK|SVf_ROK)) {
4307 SET_SVANY_FOR_BODYLESS_IV(dstr);
4308 new_dflags = SVt_IV;
4310 if (sflags & SVf_ROK) {
4311 dstr->sv_u.svu_rv = SvREFCNT_inc(SvRV(sstr));
4312 new_dflags |= SVf_ROK;
4315 /* both src and dst are <= SVt_IV, so sv_any points to the
4316 * head; so access the head directly
4318 assert( &(sstr->sv_u.svu_iv)
4319 == &(((XPVIV*) SvANY(sstr))->xiv_iv));
4320 assert( &(dstr->sv_u.svu_iv)
4321 == &(((XPVIV*) SvANY(dstr))->xiv_iv));
4322 dstr->sv_u.svu_iv = sstr->sv_u.svu_iv;
4323 new_dflags |= (SVf_IOK|SVp_IOK|(sflags & SVf_IVisUV));
4327 new_dflags = dtype; /* turn off everything except the type */
4329 SvFLAGS(dstr) = new_dflags;
4330 SvREFCNT_dec(old_rv);
4335 if (UNLIKELY(both_type == SVTYPEMASK)) {
4336 if (SvIS_FREED(dstr)) {
4337 Perl_croak(aTHX_ "panic: attempt to copy value %" SVf
4338 " to a freed scalar %p", SVfARG(sstr), (void *)dstr);
4340 if (SvIS_FREED(sstr)) {
4341 Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p to %p",
4342 (void*)sstr, (void*)dstr);
4348 SV_CHECK_THINKFIRST_COW_DROP(dstr);
4349 dtype = SvTYPE(dstr); /* THINKFIRST may have changed type */
4351 /* There's a lot of redundancy below but we're going for speed here */
4356 if (LIKELY( dtype != SVt_PVGV && dtype != SVt_PVLV )) {
4357 (void)SvOK_off(dstr);
4365 /* For performance, we inline promoting to type SVt_IV. */
4366 /* We're starting from SVt_NULL, so provided that define is
4367 * actual 0, we don't have to unset any SV type flags
4368 * to promote to SVt_IV. */
4369 STATIC_ASSERT_STMT(SVt_NULL == 0);
4370 SET_SVANY_FOR_BODYLESS_IV(dstr);
4371 SvFLAGS(dstr) |= SVt_IV;
4375 sv_upgrade(dstr, SVt_PVIV);
4379 goto end_of_first_switch;
4381 (void)SvIOK_only(dstr);
4382 SvIV_set(dstr, SvIVX(sstr));
4385 /* SvTAINTED can only be true if the SV has taint magic, which in
4386 turn means that the SV type is PVMG (or greater). This is the
4387 case statement for SVt_IV, so this cannot be true (whatever gcov
4389 assert(!SvTAINTED(sstr));
4394 if (dtype < SVt_PV && dtype != SVt_IV)
4395 sv_upgrade(dstr, SVt_IV);
4399 if (LIKELY( SvNOK(sstr) )) {
4403 sv_upgrade(dstr, SVt_NV);
4407 sv_upgrade(dstr, SVt_PVNV);
4411 goto end_of_first_switch;
4413 SvNV_set(dstr, SvNVX(sstr));
4414 (void)SvNOK_only(dstr);
4415 /* SvTAINTED can only be true if the SV has taint magic, which in
4416 turn means that the SV type is PVMG (or greater). This is the
4417 case statement for SVt_NV, so this cannot be true (whatever gcov
4419 assert(!SvTAINTED(sstr));
4426 sv_upgrade(dstr, SVt_PV);
4429 if (dtype < SVt_PVIV)
4430 sv_upgrade(dstr, SVt_PVIV);
4433 if (dtype < SVt_PVNV)
4434 sv_upgrade(dstr, SVt_PVNV);
4438 const char * const type = sv_reftype(sstr,0);
4440 /* diag_listed_as: Bizarre copy of %s */
4441 Perl_croak(aTHX_ "Bizarre copy of %s in %s", type, OP_DESC(PL_op));
4443 Perl_croak(aTHX_ "Bizarre copy of %s", type);
4445 NOT_REACHED; /* NOTREACHED */
4449 if (dtype < SVt_REGEXP)
4450 sv_upgrade(dstr, SVt_REGEXP);
4457 if (SvGMAGICAL(sstr) && (flags & SV_GMAGIC)) {
4459 if (SvTYPE(sstr) != stype)
4460 stype = SvTYPE(sstr);
4462 if (isGV_with_GP(sstr) && dtype <= SVt_PVLV) {
4463 glob_assign_glob(dstr, sstr, dtype);
4466 if (stype == SVt_PVLV)
4468 if (isREGEXP(sstr)) goto upgregexp;
4469 SvUPGRADE(dstr, SVt_PVNV);
4472 SvUPGRADE(dstr, (svtype)stype);
4474 end_of_first_switch:
4476 /* dstr may have been upgraded. */
4477 dtype = SvTYPE(dstr);
4478 sflags = SvFLAGS(sstr);
4480 if (UNLIKELY( dtype == SVt_PVCV )) {
4481 /* Assigning to a subroutine sets the prototype. */
4484 const char *const ptr = SvPV_const(sstr, len);
4486 SvGROW(dstr, len + 1);
4487 Copy(ptr, SvPVX(dstr), len + 1, char);
4488 SvCUR_set(dstr, len);
4490 SvFLAGS(dstr) |= sflags & SVf_UTF8;
4491 CvAUTOLOAD_off(dstr);
4496 else if (UNLIKELY(dtype == SVt_PVAV || dtype == SVt_PVHV
4497 || dtype == SVt_PVFM))
4499 const char * const type = sv_reftype(dstr,0);
4501 /* diag_listed_as: Cannot copy to %s */
4502 Perl_croak(aTHX_ "Cannot copy to %s in %s", type, OP_DESC(PL_op));
4504 Perl_croak(aTHX_ "Cannot copy to %s", type);
4505 } else if (sflags & SVf_ROK) {
4506 if (isGV_with_GP(dstr)
4507 && SvTYPE(SvRV(sstr)) == SVt_PVGV && isGV_with_GP(SvRV(sstr))) {
4510 if (GvIMPORTED(dstr) != GVf_IMPORTED
4511 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
4513 GvIMPORTED_on(dstr);
4518 glob_assign_glob(dstr, sstr, dtype);
4522 if (dtype >= SVt_PV) {
4523 if (isGV_with_GP(dstr)) {
4524 gv_setref(dstr, sstr);
4527 if (SvPVX_const(dstr)) {
4533 (void)SvOK_off(dstr);
4534 SvRV_set(dstr, SvREFCNT_inc(SvRV(sstr)));
4535 SvFLAGS(dstr) |= sflags & SVf_ROK;
4536 assert(!(sflags & SVp_NOK));
4537 assert(!(sflags & SVp_IOK));
4538 assert(!(sflags & SVf_NOK));
4539 assert(!(sflags & SVf_IOK));
4541 else if (isGV_with_GP(dstr)) {
4542 if (!(sflags & SVf_OK)) {
4543 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4544 "Undefined value assigned to typeglob");
4547 GV *gv = gv_fetchsv_nomg(sstr, GV_ADD, SVt_PVGV);
4548 if (dstr != (const SV *)gv) {
4549 const char * const name = GvNAME((const GV *)dstr);
4550 const STRLEN len = GvNAMELEN(dstr);
4551 HV *old_stash = NULL;
4552 bool reset_isa = FALSE;
4553 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
4554 || (len == 1 && name[0] == ':')) {
4555 /* Set aside the old stash, so we can reset isa caches
4556 on its subclasses. */
4557 if((old_stash = GvHV(dstr))) {
4558 /* Make sure we do not lose it early. */
4559 SvREFCNT_inc_simple_void_NN(
4560 sv_2mortal((SV *)old_stash)
4567 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
4568 gp_free(MUTABLE_GV(dstr));
4570 GvGP_set(dstr, gp_ref(GvGP(gv)));
4573 HV * const stash = GvHV(dstr);
4575 old_stash ? (HV *)HvENAME_get(old_stash) : stash
4585 else if ((dtype == SVt_REGEXP || dtype == SVt_PVLV)
4586 && (stype == SVt_REGEXP || isREGEXP(sstr))) {
4587 reg_temp_copy((REGEXP*)dstr, (REGEXP*)sstr);
4589 else if (sflags & SVp_POK) {
4590 const STRLEN cur = SvCUR(sstr);
4591 const STRLEN len = SvLEN(sstr);
4594 * We have three basic ways to copy the string:
4600 * Which we choose is based on various factors. The following
4601 * things are listed in order of speed, fastest to slowest:
4603 * - Copying a short string
4604 * - Copy-on-write bookkeeping
4606 * - Copying a long string
4608 * We swipe the string (steal the string buffer) if the SV on the
4609 * rhs is about to be freed anyway (TEMP and refcnt==1). This is a
4610 * big win on long strings. It should be a win on short strings if
4611 * SvPVX_const(dstr) has to be allocated. If not, it should not
4612 * slow things down, as SvPVX_const(sstr) would have been freed
4615 * We also steal the buffer from a PADTMP (operator target) if it
4616 * is ‘long enough’. For short strings, a swipe does not help
4617 * here, as it causes more malloc calls the next time the target
4618 * is used. Benchmarks show that even if SvPVX_const(dstr) has to
4619 * be allocated it is still not worth swiping PADTMPs for short
4620 * strings, as the savings here are small.
4622 * If swiping is not an option, then we see whether it is
4623 * worth using copy-on-write. If the lhs already has a buf-
4624 * fer big enough and the string is short, we skip it and fall back
4625 * to method 3, since memcpy is faster for short strings than the
4626 * later bookkeeping overhead that copy-on-write entails.
4628 * If the rhs is not a copy-on-write string yet, then we also
4629 * consider whether the buffer is too large relative to the string
4630 * it holds. Some operations such as readline allocate a large
4631 * buffer in the expectation of reusing it. But turning such into
4632 * a COW buffer is counter-productive because it increases memory
4633 * usage by making readline allocate a new large buffer the sec-
4634 * ond time round. So, if the buffer is too large, again, we use
4637 * Finally, if there is no buffer on the left, or the buffer is too
4638 * small, then we use copy-on-write and make both SVs share the
4643 /* Whichever path we take through the next code, we want this true,
4644 and doing it now facilitates the COW check. */
4645 (void)SvPOK_only(dstr);
4649 /* slated for free anyway (and not COW)? */
4650 (sflags & (SVs_TEMP|SVf_IsCOW)) == SVs_TEMP
4651 /* or a swipable TARG */
4653 (SVs_PADTMP|SVf_READONLY|SVf_PROTECT|SVf_IsCOW))
4655 /* whose buffer is worth stealing */
4656 && CHECK_COWBUF_THRESHOLD(cur,len)
4659 !(sflags & SVf_OOK) && /* and not involved in OOK hack? */
4660 (!(flags & SV_NOSTEAL)) &&
4661 /* and we're allowed to steal temps */
4662 SvREFCNT(sstr) == 1 && /* and no other references to it? */
4663 len) /* and really is a string */
4664 { /* Passes the swipe test. */
4665 if (SvPVX_const(dstr)) /* we know that dtype >= SVt_PV */
4667 SvPV_set(dstr, SvPVX_mutable(sstr));
4668 SvLEN_set(dstr, SvLEN(sstr));
4669 SvCUR_set(dstr, SvCUR(sstr));
4672 (void)SvOK_off(sstr); /* NOTE: nukes most SvFLAGS on sstr */
4673 SvPV_set(sstr, NULL);
4678 else if (flags & SV_COW_SHARED_HASH_KEYS
4680 #ifdef PERL_COPY_ON_WRITE
4683 ( (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4684 /* If this is a regular (non-hek) COW, only so
4685 many COW "copies" are possible. */
4686 && CowREFCNT(sstr) != SV_COW_REFCNT_MAX ))
4687 : ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4688 && !(SvFLAGS(dstr) & SVf_BREAK)
4689 && CHECK_COW_THRESHOLD(cur,len) && cur+1 < len
4690 && (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4694 && !(SvFLAGS(dstr) & SVf_BREAK)
4697 /* Either it's a shared hash key, or it's suitable for
4701 PerlIO_printf(Perl_debug_log, "Copy on write: sstr --> dstr\n");
4707 if (!(sflags & SVf_IsCOW)) {
4709 CowREFCNT(sstr) = 0;
4712 if (SvPVX_const(dstr)) { /* we know that dtype >= SVt_PV */
4718 if (sflags & SVf_IsCOW) {
4722 SvPV_set(dstr, SvPVX_mutable(sstr));
4727 /* SvIsCOW_shared_hash */
4728 DEBUG_C(PerlIO_printf(Perl_debug_log,
4729 "Copy on write: Sharing hash\n"));
4731 assert (SvTYPE(dstr) >= SVt_PV);
4733 HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)))));
4735 SvLEN_set(dstr, len);
4736 SvCUR_set(dstr, cur);
4739 /* Failed the swipe test, and we cannot do copy-on-write either.
4740 Have to copy the string. */
4741 SvGROW(dstr, cur + 1); /* inlined from sv_setpvn */
4742 Move(SvPVX_const(sstr),SvPVX(dstr),cur,char);
4743 SvCUR_set(dstr, cur);
4744 *SvEND(dstr) = '\0';
4746 if (sflags & SVp_NOK) {
4747 SvNV_set(dstr, SvNVX(sstr));
4749 if (sflags & SVp_IOK) {
4750 SvIV_set(dstr, SvIVX(sstr));
4751 if (sflags & SVf_IVisUV)
4754 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_NOK|SVp_NOK|SVf_UTF8);
4756 const MAGIC * const smg = SvVSTRING_mg(sstr);
4758 sv_magic(dstr, NULL, PERL_MAGIC_vstring,
4759 smg->mg_ptr, smg->mg_len);
4760 SvRMAGICAL_on(dstr);
4764 else if (sflags & (SVp_IOK|SVp_NOK)) {
4765 (void)SvOK_off(dstr);
4766 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_IVisUV|SVf_NOK|SVp_NOK);
4767 if (sflags & SVp_IOK) {
4768 /* XXXX Do we want to set IsUV for IV(ROK)? Be extra safe... */
4769 SvIV_set(dstr, SvIVX(sstr));
4771 if (sflags & SVp_NOK) {
4772 SvNV_set(dstr, SvNVX(sstr));
4776 if (isGV_with_GP(sstr)) {
4777 gv_efullname3(dstr, MUTABLE_GV(sstr), "*");
4780 (void)SvOK_off(dstr);
4782 if (SvTAINTED(sstr))
4788 =for apidoc sv_set_undef
4790 Equivalent to C<sv_setsv(sv, &PL_sv_undef)>, but more efficient.
4791 Doesn't handle set magic.
4793 The perl equivalent is C<$sv = undef;>. Note that it doesn't free any string
4794 buffer, unlike C<undef $sv>.
4796 Introduced in perl 5.25.12.
4802 Perl_sv_set_undef(pTHX_ SV *sv)
4804 U32 type = SvTYPE(sv);
4806 PERL_ARGS_ASSERT_SV_SET_UNDEF;
4808 /* shortcut, NULL, IV, RV */
4810 if (type <= SVt_IV) {
4811 assert(!SvGMAGICAL(sv));
4812 if (SvREADONLY(sv)) {
4813 /* does undeffing PL_sv_undef count as modifying a read-only
4814 * variable? Some XS code does this */
4815 if (sv == &PL_sv_undef)
4817 Perl_croak_no_modify();
4822 sv_unref_flags(sv, 0);
4825 SvFLAGS(sv) = type; /* quickly turn off all flags */
4826 SvREFCNT_dec_NN(rv);
4830 SvFLAGS(sv) = type; /* quickly turn off all flags */
4835 Perl_croak(aTHX_ "panic: attempt to undefine a freed scalar %p",
4838 SV_CHECK_THINKFIRST_COW_DROP(sv);
4840 if (isGV_with_GP(sv))
4841 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4842 "Undefined value assigned to typeglob");
4850 =for apidoc sv_setsv_mg
4852 Like C<sv_setsv>, but also handles 'set' magic.
4858 Perl_sv_setsv_mg(pTHX_ SV *const dstr, SV *const sstr)
4860 PERL_ARGS_ASSERT_SV_SETSV_MG;
4862 sv_setsv(dstr,sstr);
4867 # define SVt_COW SVt_PV
4869 Perl_sv_setsv_cow(pTHX_ SV *dstr, SV *sstr)
4871 STRLEN cur = SvCUR(sstr);
4872 STRLEN len = SvLEN(sstr);
4874 #if defined(PERL_DEBUG_READONLY_COW) && defined(PERL_COPY_ON_WRITE)
4875 const bool already = cBOOL(SvIsCOW(sstr));
4878 PERL_ARGS_ASSERT_SV_SETSV_COW;
4881 PerlIO_printf(Perl_debug_log, "Fast copy on write: %p -> %p\n",
4882 (void*)sstr, (void*)dstr);
4889 if (SvTHINKFIRST(dstr))
4890 sv_force_normal_flags(dstr, SV_COW_DROP_PV);
4891 else if (SvPVX_const(dstr))
4892 Safefree(SvPVX_mutable(dstr));
4896 SvUPGRADE(dstr, SVt_COW);
4898 assert (SvPOK(sstr));
4899 assert (SvPOKp(sstr));
4901 if (SvIsCOW(sstr)) {
4903 if (SvLEN(sstr) == 0) {
4904 /* source is a COW shared hash key. */
4905 DEBUG_C(PerlIO_printf(Perl_debug_log,
4906 "Fast copy on write: Sharing hash\n"));
4907 new_pv = HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr))));
4910 assert(SvCUR(sstr)+1 < SvLEN(sstr));
4911 assert(CowREFCNT(sstr) < SV_COW_REFCNT_MAX);
4913 assert ((SvFLAGS(sstr) & CAN_COW_MASK) == CAN_COW_FLAGS);
4914 SvUPGRADE(sstr, SVt_COW);
4916 DEBUG_C(PerlIO_printf(Perl_debug_log,
4917 "Fast copy on write: Converting sstr to COW\n"));
4918 CowREFCNT(sstr) = 0;
4920 # ifdef PERL_DEBUG_READONLY_COW
4921 if (already) sv_buf_to_rw(sstr);
4924 new_pv = SvPVX_mutable(sstr);
4928 SvPV_set(dstr, new_pv);
4929 SvFLAGS(dstr) = (SVt_COW|SVf_POK|SVp_POK|SVf_IsCOW);
4932 SvLEN_set(dstr, len);
4933 SvCUR_set(dstr, cur);
4943 =for apidoc sv_setpv_bufsize
4945 Sets the SV to be a string of cur bytes length, with at least
4946 len bytes available. Ensures that there is a null byte at SvEND.
4947 Returns a char * pointer to the SvPV buffer.
4953 Perl_sv_setpv_bufsize(pTHX_ SV *const sv, const STRLEN cur, const STRLEN len)
4957 PERL_ARGS_ASSERT_SV_SETPV_BUFSIZE;
4959 SV_CHECK_THINKFIRST_COW_DROP(sv);
4960 SvUPGRADE(sv, SVt_PV);
4961 pv = SvGROW(sv, len + 1);
4964 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4967 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4972 =for apidoc sv_setpvn
4974 Copies a string (possibly containing embedded C<NUL> characters) into an SV.
4975 The C<len> parameter indicates the number of
4976 bytes to be copied. If the C<ptr> argument is NULL the SV will become
4977 undefined. Does not handle 'set' magic. See C<L</sv_setpvn_mg>>.
4983 Perl_sv_setpvn(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4987 PERL_ARGS_ASSERT_SV_SETPVN;
4989 SV_CHECK_THINKFIRST_COW_DROP(sv);
4990 if (isGV_with_GP(sv))
4991 Perl_croak_no_modify();
4997 /* len is STRLEN which is unsigned, need to copy to signed */
5000 Perl_croak(aTHX_ "panic: sv_setpvn called with negative strlen %"
5003 SvUPGRADE(sv, SVt_PV);
5005 dptr = SvGROW(sv, len + 1);
5006 Move(ptr,dptr,len,char);
5009 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5011 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
5015 =for apidoc sv_setpvn_mg
5017 Like C<sv_setpvn>, but also handles 'set' magic.
5023 Perl_sv_setpvn_mg(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
5025 PERL_ARGS_ASSERT_SV_SETPVN_MG;
5027 sv_setpvn(sv,ptr,len);
5032 =for apidoc sv_setpv
5034 Copies a string into an SV. The string must be terminated with a C<NUL>
5035 character, and not contain embeded C<NUL>'s.
5036 Does not handle 'set' magic. See C<L</sv_setpv_mg>>.
5042 Perl_sv_setpv(pTHX_ SV *const sv, const char *const ptr)
5046 PERL_ARGS_ASSERT_SV_SETPV;
5048 SV_CHECK_THINKFIRST_COW_DROP(sv);
5054 SvUPGRADE(sv, SVt_PV);
5056 SvGROW(sv, len + 1);
5057 Move(ptr,SvPVX(sv),len+1,char);
5059 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5061 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
5065 =for apidoc sv_setpv_mg
5067 Like C<sv_setpv>, but also handles 'set' magic.
5073 Perl_sv_setpv_mg(pTHX_ SV *const sv, const char *const ptr)
5075 PERL_ARGS_ASSERT_SV_SETPV_MG;
5082 Perl_sv_sethek(pTHX_ SV *const sv, const HEK *const hek)
5084 PERL_ARGS_ASSERT_SV_SETHEK;
5090 if (HEK_LEN(hek) == HEf_SVKEY) {
5091 sv_setsv(sv, *(SV**)HEK_KEY(hek));
5094 const int flags = HEK_FLAGS(hek);
5095 if (flags & HVhek_WASUTF8) {
5096 STRLEN utf8_len = HEK_LEN(hek);
5097 char *as_utf8 = (char *)bytes_to_utf8((U8*)HEK_KEY(hek), &utf8_len);
5098 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
5101 } else if (flags & HVhek_UNSHARED) {
5102 sv_setpvn(sv, HEK_KEY(hek), HEK_LEN(hek));
5105 else SvUTF8_off(sv);
5109 SV_CHECK_THINKFIRST_COW_DROP(sv);
5110 SvUPGRADE(sv, SVt_PV);
5112 SvPV_set(sv,(char *)HEK_KEY(share_hek_hek(hek)));
5113 SvCUR_set(sv, HEK_LEN(hek));
5119 else SvUTF8_off(sv);
5127 =for apidoc sv_usepvn_flags
5129 Tells an SV to use C<ptr> to find its string value. Normally the
5130 string is stored inside the SV, but sv_usepvn allows the SV to use an
5131 outside string. C<ptr> should point to memory that was allocated
5132 by L<C<Newx>|perlclib/Memory Management and String Handling>. It must be
5133 the start of a C<Newx>-ed block of memory, and not a pointer to the
5134 middle of it (beware of L<C<OOK>|perlguts/Offsets> and copy-on-write),
5135 and not be from a non-C<Newx> memory allocator like C<malloc>. The
5136 string length, C<len>, must be supplied. By default this function
5137 will C<Renew> (i.e. realloc, move) the memory pointed to by C<ptr>,
5138 so that pointer should not be freed or used by the programmer after
5139 giving it to C<sv_usepvn>, and neither should any pointers from "behind"
5140 that pointer (e.g. ptr + 1) be used.
5142 If S<C<flags & SV_SMAGIC>> is true, will call C<SvSETMAGIC>. If
5143 S<C<flags> & SV_HAS_TRAILING_NUL>> is true, then C<ptr[len]> must be C<NUL>,
5145 will be skipped (i.e. the buffer is actually at least 1 byte longer than
5146 C<len>, and already meets the requirements for storing in C<SvPVX>).
5152 Perl_sv_usepvn_flags(pTHX_ SV *const sv, char *ptr, const STRLEN len, const U32 flags)
5156 PERL_ARGS_ASSERT_SV_USEPVN_FLAGS;
5158 SV_CHECK_THINKFIRST_COW_DROP(sv);
5159 SvUPGRADE(sv, SVt_PV);
5162 if (flags & SV_SMAGIC)
5166 if (SvPVX_const(sv))
5170 if (flags & SV_HAS_TRAILING_NUL)
5171 assert(ptr[len] == '\0');
5174 allocate = (flags & SV_HAS_TRAILING_NUL)
5176 #ifdef Perl_safesysmalloc_size
5179 PERL_STRLEN_ROUNDUP(len + 1);
5181 if (flags & SV_HAS_TRAILING_NUL) {
5182 /* It's long enough - do nothing.
5183 Specifically Perl_newCONSTSUB is relying on this. */
5186 /* Force a move to shake out bugs in callers. */
5187 char *new_ptr = (char*)safemalloc(allocate);
5188 Copy(ptr, new_ptr, len, char);
5189 PoisonFree(ptr,len,char);
5193 ptr = (char*) saferealloc (ptr, allocate);
5196 #ifdef Perl_safesysmalloc_size
5197 SvLEN_set(sv, Perl_safesysmalloc_size(ptr));
5199 SvLEN_set(sv, allocate);
5203 if (!(flags & SV_HAS_TRAILING_NUL)) {
5206 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5208 if (flags & SV_SMAGIC)
5214 S_sv_uncow(pTHX_ SV * const sv, const U32 flags)
5216 assert(SvIsCOW(sv));
5219 const char * const pvx = SvPVX_const(sv);
5220 const STRLEN len = SvLEN(sv);
5221 const STRLEN cur = SvCUR(sv);
5225 PerlIO_printf(Perl_debug_log,
5226 "Copy on write: Force normal %ld\n",
5232 # ifdef PERL_COPY_ON_WRITE
5234 /* Must do this first, since the CowREFCNT uses SvPVX and
5235 we need to write to CowREFCNT, or de-RO the whole buffer if we are
5236 the only owner left of the buffer. */
5237 sv_buf_to_rw(sv); /* NOOP if RO-ing not supported */
5239 U8 cowrefcnt = CowREFCNT(sv);
5240 if(cowrefcnt != 0) {
5242 CowREFCNT(sv) = cowrefcnt;
5247 /* Else we are the only owner of the buffer. */
5252 /* This SV doesn't own the buffer, so need to Newx() a new one: */
5257 if (flags & SV_COW_DROP_PV) {
5258 /* OK, so we don't need to copy our buffer. */
5261 SvGROW(sv, cur + 1);
5262 Move(pvx,SvPVX(sv),cur,char);
5268 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5276 const char * const pvx = SvPVX_const(sv);
5277 const STRLEN len = SvCUR(sv);
5281 if (flags & SV_COW_DROP_PV) {
5282 /* OK, so we don't need to copy our buffer. */
5285 SvGROW(sv, len + 1);
5286 Move(pvx,SvPVX(sv),len,char);
5289 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5296 =for apidoc sv_force_normal_flags
5298 Undo various types of fakery on an SV, where fakery means
5299 "more than" a string: if the PV is a shared string, make
5300 a private copy; if we're a ref, stop refing; if we're a glob, downgrade to
5301 an C<xpvmg>; if we're a copy-on-write scalar, this is the on-write time when
5302 we do the copy, and is also used locally; if this is a
5303 vstring, drop the vstring magic. If C<SV_COW_DROP_PV> is set
5304 then a copy-on-write scalar drops its PV buffer (if any) and becomes
5305 C<SvPOK_off> rather than making a copy. (Used where this
5306 scalar is about to be set to some other value.) In addition,
5307 the C<flags> parameter gets passed to C<sv_unref_flags()>
5308 when unreffing. C<sv_force_normal> calls this function
5309 with flags set to 0.
5311 This function is expected to be used to signal to perl that this SV is
5312 about to be written to, and any extra book-keeping needs to be taken care
5313 of. Hence, it croaks on read-only values.
5319 Perl_sv_force_normal_flags(pTHX_ SV *const sv, const U32 flags)
5321 PERL_ARGS_ASSERT_SV_FORCE_NORMAL_FLAGS;
5324 Perl_croak_no_modify();
5325 else if (SvIsCOW(sv) && LIKELY(SvTYPE(sv) != SVt_PVHV))
5326 S_sv_uncow(aTHX_ sv, flags);
5328 sv_unref_flags(sv, flags);
5329 else if (SvFAKE(sv) && isGV_with_GP(sv))
5330 sv_unglob(sv, flags);
5331 else if (SvFAKE(sv) && isREGEXP(sv)) {
5332 /* Need to downgrade the REGEXP to a simple(r) scalar. This is analogous
5333 to sv_unglob. We only need it here, so inline it. */
5334 const bool islv = SvTYPE(sv) == SVt_PVLV;
5335 const svtype new_type =
5336 islv ? SVt_NULL : SvMAGIC(sv) || SvSTASH(sv) ? SVt_PVMG : SVt_PV;
5337 SV *const temp = newSV_type(new_type);
5338 regexp *old_rx_body;
5340 if (new_type == SVt_PVMG) {
5341 SvMAGIC_set(temp, SvMAGIC(sv));
5342 SvMAGIC_set(sv, NULL);
5343 SvSTASH_set(temp, SvSTASH(sv));
5344 SvSTASH_set(sv, NULL);
5347 SvCUR_set(temp, SvCUR(sv));
5348 /* Remember that SvPVX is in the head, not the body. */
5349 assert(ReANY((REGEXP *)sv)->mother_re);
5352 /* LV-as-regex has sv->sv_any pointing to an XPVLV body,
5353 * whose xpvlenu_rx field points to the regex body */
5354 XPV *xpv = (XPV*)(SvANY(sv));
5355 old_rx_body = xpv->xpv_len_u.xpvlenu_rx;
5356 xpv->xpv_len_u.xpvlenu_rx = NULL;
5359 old_rx_body = ReANY((REGEXP *)sv);
5361 /* Their buffer is already owned by someone else. */
5362 if (flags & SV_COW_DROP_PV) {
5363 /* SvLEN is already 0. For SVt_REGEXP, we have a brand new
5364 zeroed body. For SVt_PVLV, we zeroed it above (len field
5365 a union with xpvlenu_rx) */
5366 assert(!SvLEN(islv ? sv : temp));
5367 sv->sv_u.svu_pv = 0;
5370 sv->sv_u.svu_pv = savepvn(RX_WRAPPED((REGEXP *)sv), SvCUR(sv));
5371 SvLEN_set(islv ? sv : temp, SvCUR(sv)+1);
5375 /* Now swap the rest of the bodies. */
5379 SvFLAGS(sv) &= ~SVTYPEMASK;
5380 SvFLAGS(sv) |= new_type;
5381 SvANY(sv) = SvANY(temp);
5384 SvFLAGS(temp) &= ~(SVTYPEMASK);
5385 SvFLAGS(temp) |= SVt_REGEXP|SVf_FAKE;
5386 SvANY(temp) = old_rx_body;
5388 SvREFCNT_dec_NN(temp);
5390 else if (SvVOK(sv)) sv_unmagic(sv, PERL_MAGIC_vstring);
5396 Efficient removal of characters from the beginning of the string buffer.
5397 C<SvPOK(sv)>, or at least C<SvPOKp(sv)>, must be true and C<ptr> must be a
5398 pointer to somewhere inside the string buffer. C<ptr> becomes the first
5399 character of the adjusted string. Uses the C<OOK> hack. On return, only
5400 C<SvPOK(sv)> and C<SvPOKp(sv)> among the C<OK> flags will be true.
5402 Beware: after this function returns, C<ptr> and SvPVX_const(sv) may no longer
5403 refer to the same chunk of data.
5405 The unfortunate similarity of this function's name to that of Perl's C<chop>
5406 operator is strictly coincidental. This function works from the left;
5407 C<chop> works from the right.
5413 Perl_sv_chop(pTHX_ SV *const sv, const char *const ptr)
5424 PERL_ARGS_ASSERT_SV_CHOP;
5426 if (!ptr || !SvPOKp(sv))
5428 delta = ptr - SvPVX_const(sv);
5430 /* Nothing to do. */
5433 max_delta = SvLEN(sv) ? SvLEN(sv) : SvCUR(sv);
5434 if (delta > max_delta)
5435 Perl_croak(aTHX_ "panic: sv_chop ptr=%p, start=%p, end=%p",
5436 ptr, SvPVX_const(sv), SvPVX_const(sv) + max_delta);
5437 /* SvPVX(sv) may move in SV_CHECK_THINKFIRST(sv), so don't use ptr any more */
5438 SV_CHECK_THINKFIRST(sv);
5439 SvPOK_only_UTF8(sv);
5442 if (!SvLEN(sv)) { /* make copy of shared string */
5443 const char *pvx = SvPVX_const(sv);
5444 const STRLEN len = SvCUR(sv);
5445 SvGROW(sv, len + 1);
5446 Move(pvx,SvPVX(sv),len,char);
5452 SvOOK_offset(sv, old_delta);
5454 SvLEN_set(sv, SvLEN(sv) - delta);
5455 SvCUR_set(sv, SvCUR(sv) - delta);
5456 SvPV_set(sv, SvPVX(sv) + delta);
5458 p = (U8 *)SvPVX_const(sv);
5461 /* how many bytes were evacuated? we will fill them with sentinel
5462 bytes, except for the part holding the new offset of course. */
5465 evacn += (old_delta < 0x100 ? 1 : 1 + sizeof(STRLEN));
5467 assert(evacn <= delta + old_delta);
5471 /* This sets 'delta' to the accumulated value of all deltas so far */
5475 /* If 'delta' fits in a byte, store it just prior to the new beginning of
5476 * the string; otherwise store a 0 byte there and store 'delta' just prior
5477 * to that, using as many bytes as a STRLEN occupies. Thus it overwrites a
5478 * portion of the chopped part of the string */
5479 if (delta < 0x100) {
5483 p -= sizeof(STRLEN);
5484 Copy((U8*)&delta, p, sizeof(STRLEN), U8);
5488 /* Fill the preceding buffer with sentinals to verify that no-one is
5498 =for apidoc sv_catpvn
5500 Concatenates the string onto the end of the string which is in the SV.
5501 C<len> indicates number of bytes to copy. If the SV has the UTF-8
5502 status set, then the bytes appended should be valid UTF-8.
5503 Handles 'get' magic, but not 'set' magic. See C<L</sv_catpvn_mg>>.
5505 =for apidoc sv_catpvn_flags
5507 Concatenates the string onto the end of the string which is in the SV. The
5508 C<len> indicates number of bytes to copy.
5510 By default, the string appended is assumed to be valid UTF-8 if the SV has
5511 the UTF-8 status set, and a string of bytes otherwise. One can force the
5512 appended string to be interpreted as UTF-8 by supplying the C<SV_CATUTF8>
5513 flag, and as bytes by supplying the C<SV_CATBYTES> flag; the SV or the
5514 string appended will be upgraded to UTF-8 if necessary.
5516 If C<flags> has the C<SV_SMAGIC> bit set, will
5517 C<mg_set> on C<dsv> afterwards if appropriate.
5518 C<sv_catpvn> and C<sv_catpvn_nomg> are implemented
5519 in terms of this function.
5525 Perl_sv_catpvn_flags(pTHX_ SV *const dsv, const char *sstr, const STRLEN slen, const I32 flags)
5528 const char * const dstr = SvPV_force_flags(dsv, dlen, flags);
5530 PERL_ARGS_ASSERT_SV_CATPVN_FLAGS;
5531 assert((flags & (SV_CATBYTES|SV_CATUTF8)) != (SV_CATBYTES|SV_CATUTF8));
5533 if (!(flags & SV_CATBYTES) || !SvUTF8(dsv)) {
5534 if (flags & SV_CATUTF8 && !SvUTF8(dsv)) {
5535 sv_utf8_upgrade_flags_grow(dsv, 0, slen + 1);
5538 else SvGROW(dsv, dlen + slen + 3);
5540 sstr = SvPVX_const(dsv);
5541 Move(sstr, SvPVX(dsv) + dlen, slen, char);
5542 SvCUR_set(dsv, SvCUR(dsv) + slen);
5545 /* We inline bytes_to_utf8, to avoid an extra malloc. */
5546 const char * const send = sstr + slen;
5549 /* Something this code does not account for, which I think is
5550 impossible; it would require the same pv to be treated as
5551 bytes *and* utf8, which would indicate a bug elsewhere. */
5552 assert(sstr != dstr);
5554 SvGROW(dsv, dlen + slen * 2 + 3);
5555 d = (U8 *)SvPVX(dsv) + dlen;
5557 while (sstr < send) {
5558 append_utf8_from_native_byte(*sstr, &d);
5561 SvCUR_set(dsv, d-(const U8 *)SvPVX(dsv));
5564 (void)SvPOK_only_UTF8(dsv); /* validate pointer */
5566 if (flags & SV_SMAGIC)
5571 =for apidoc sv_catsv
5573 Concatenates the string from SV C<ssv> onto the end of the string in SV
5574 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5575 Handles 'get' magic on both SVs, but no 'set' magic. See C<L</sv_catsv_mg>>
5576 and C<L</sv_catsv_nomg>>.
5578 =for apidoc sv_catsv_flags
5580 Concatenates the string from SV C<ssv> onto the end of the string in SV
5581 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5582 If C<flags> has the C<SV_GMAGIC> bit set, will call C<mg_get> on both SVs if
5583 appropriate. If C<flags> has the C<SV_SMAGIC> bit set, C<mg_set> will be called on
5584 the modified SV afterward, if appropriate. C<sv_catsv>, C<sv_catsv_nomg>,
5585 and C<sv_catsv_mg> are implemented in terms of this function.
5590 Perl_sv_catsv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
5592 PERL_ARGS_ASSERT_SV_CATSV_FLAGS;
5596 const char *spv = SvPV_flags_const(ssv, slen, flags);
5597 if (flags & SV_GMAGIC)
5599 sv_catpvn_flags(dsv, spv, slen,
5600 DO_UTF8(ssv) ? SV_CATUTF8 : SV_CATBYTES);
5601 if (flags & SV_SMAGIC)
5607 =for apidoc sv_catpv
5609 Concatenates the C<NUL>-terminated string onto the end of the string which is
5611 If the SV has the UTF-8 status set, then the bytes appended should be
5612 valid UTF-8. Handles 'get' magic, but not 'set' magic. See
5618 Perl_sv_catpv(pTHX_ SV *const sv, const char *ptr)
5624 PERL_ARGS_ASSERT_SV_CATPV;
5628 junk = SvPV_force(sv, tlen);
5630 SvGROW(sv, tlen + len + 1);
5632 ptr = SvPVX_const(sv);
5633 Move(ptr,SvPVX(sv)+tlen,len+1,char);
5634 SvCUR_set(sv, SvCUR(sv) + len);
5635 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5640 =for apidoc sv_catpv_flags
5642 Concatenates the C<NUL>-terminated string onto the end of the string which is
5644 If the SV has the UTF-8 status set, then the bytes appended should
5645 be valid UTF-8. If C<flags> has the C<SV_SMAGIC> bit set, will C<mg_set>
5646 on the modified SV if appropriate.
5652 Perl_sv_catpv_flags(pTHX_ SV *dstr, const char *sstr, const I32 flags)
5654 PERL_ARGS_ASSERT_SV_CATPV_FLAGS;
5655 sv_catpvn_flags(dstr, sstr, strlen(sstr), flags);
5659 =for apidoc sv_catpv_mg
5661 Like C<sv_catpv>, but also handles 'set' magic.
5667 Perl_sv_catpv_mg(pTHX_ SV *const sv, const char *const ptr)
5669 PERL_ARGS_ASSERT_SV_CATPV_MG;
5678 Creates a new SV. A non-zero C<len> parameter indicates the number of
5679 bytes of preallocated string space the SV should have. An extra byte for a
5680 trailing C<NUL> is also reserved. (C<SvPOK> is not set for the SV even if string
5681 space is allocated.) The reference count for the new SV is set to 1.
5683 In 5.9.3, C<newSV()> replaces the older C<NEWSV()> API, and drops the first
5684 parameter, I<x>, a debug aid which allowed callers to identify themselves.
5685 This aid has been superseded by a new build option, C<PERL_MEM_LOG> (see
5686 L<perlhacktips/PERL_MEM_LOG>). The older API is still there for use in XS
5687 modules supporting older perls.
5693 Perl_newSV(pTHX_ const STRLEN len)
5699 sv_grow(sv, len + 1);
5704 =for apidoc sv_magicext
5706 Adds magic to an SV, upgrading it if necessary. Applies the
5707 supplied C<vtable> and returns a pointer to the magic added.
5709 Note that C<sv_magicext> will allow things that C<sv_magic> will not.
5710 In particular, you can add magic to C<SvREADONLY> SVs, and add more than
5711 one instance of the same C<how>.
5713 If C<namlen> is greater than zero then a C<savepvn> I<copy> of C<name> is
5714 stored, if C<namlen> is zero then C<name> is stored as-is and - as another
5715 special case - if C<(name && namlen == HEf_SVKEY)> then C<name> is assumed
5716 to contain an SV* and is stored as-is with its C<REFCNT> incremented.
5718 (This is now used as a subroutine by C<sv_magic>.)
5723 Perl_sv_magicext(pTHX_ SV *const sv, SV *const obj, const int how,
5724 const MGVTBL *const vtable, const char *const name, const I32 namlen)
5728 PERL_ARGS_ASSERT_SV_MAGICEXT;
5730 SvUPGRADE(sv, SVt_PVMG);
5731 Newxz(mg, 1, MAGIC);
5732 mg->mg_moremagic = SvMAGIC(sv);
5733 SvMAGIC_set(sv, mg);
5735 /* Sometimes a magic contains a reference loop, where the sv and
5736 object refer to each other. To prevent a reference loop that
5737 would prevent such objects being freed, we look for such loops
5738 and if we find one we avoid incrementing the object refcount.
5740 Note we cannot do this to avoid self-tie loops as intervening RV must
5741 have its REFCNT incremented to keep it in existence.
5744 if (!obj || obj == sv ||
5745 how == PERL_MAGIC_arylen ||
5746 how == PERL_MAGIC_regdata ||
5747 how == PERL_MAGIC_regdatum ||
5748 how == PERL_MAGIC_symtab ||
5749 (SvTYPE(obj) == SVt_PVGV &&
5750 (GvSV(obj) == sv || GvHV(obj) == (const HV *)sv
5751 || GvAV(obj) == (const AV *)sv || GvCV(obj) == (const CV *)sv
5752 || GvIOp(obj) == (const IO *)sv || GvFORM(obj) == (const CV *)sv)))
5757 mg->mg_obj = SvREFCNT_inc_simple(obj);
5758 mg->mg_flags |= MGf_REFCOUNTED;
5761 /* Normal self-ties simply pass a null object, and instead of
5762 using mg_obj directly, use the SvTIED_obj macro to produce a
5763 new RV as needed. For glob "self-ties", we are tieing the PVIO
5764 with an RV obj pointing to the glob containing the PVIO. In
5765 this case, to avoid a reference loop, we need to weaken the
5769 if (how == PERL_MAGIC_tiedscalar && SvTYPE(sv) == SVt_PVIO &&
5770 obj && SvROK(obj) && GvIO(SvRV(obj)) == (const IO *)sv)
5776 mg->mg_len = namlen;
5779 mg->mg_ptr = savepvn(name, namlen);
5780 else if (namlen == HEf_SVKEY) {
5781 /* Yes, this is casting away const. This is only for the case of
5782 HEf_SVKEY. I think we need to document this aberation of the
5783 constness of the API, rather than making name non-const, as
5784 that change propagating outwards a long way. */
5785 mg->mg_ptr = (char*)SvREFCNT_inc_simple_NN((SV *)name);
5787 mg->mg_ptr = (char *) name;
5789 mg->mg_virtual = (MGVTBL *) vtable;
5796 Perl_sv_magicext_mglob(pTHX_ SV *sv)
5798 PERL_ARGS_ASSERT_SV_MAGICEXT_MGLOB;
5799 if (SvTYPE(sv) == SVt_PVLV && LvTYPE(sv) == 'y') {
5800 /* This sv is only a delegate. //g magic must be attached to
5805 return sv_magicext(sv, NULL, PERL_MAGIC_regex_global,
5806 &PL_vtbl_mglob, 0, 0);
5810 =for apidoc sv_magic
5812 Adds magic to an SV. First upgrades C<sv> to type C<SVt_PVMG> if
5813 necessary, then adds a new magic item of type C<how> to the head of the
5816 See C<L</sv_magicext>> (which C<sv_magic> now calls) for a description of the
5817 handling of the C<name> and C<namlen> arguments.
5819 You need to use C<sv_magicext> to add magic to C<SvREADONLY> SVs and also
5820 to add more than one instance of the same C<how>.
5826 Perl_sv_magic(pTHX_ SV *const sv, SV *const obj, const int how,
5827 const char *const name, const I32 namlen)
5829 const MGVTBL *vtable;
5832 unsigned int vtable_index;
5834 PERL_ARGS_ASSERT_SV_MAGIC;
5836 if (how < 0 || (unsigned)how >= C_ARRAY_LENGTH(PL_magic_data)
5837 || ((flags = PL_magic_data[how]),
5838 (vtable_index = flags & PERL_MAGIC_VTABLE_MASK)
5839 > magic_vtable_max))
5840 Perl_croak(aTHX_ "Don't know how to handle magic of type \\%o", how);
5842 /* PERL_MAGIC_ext is reserved for use by extensions not perl internals.
5843 Useful for attaching extension internal data to perl vars.
5844 Note that multiple extensions may clash if magical scalars
5845 etc holding private data from one are passed to another. */
5847 vtable = (vtable_index == magic_vtable_max)
5848 ? NULL : PL_magic_vtables + vtable_index;
5850 if (SvREADONLY(sv)) {
5852 !PERL_MAGIC_TYPE_READONLY_ACCEPTABLE(how)
5855 Perl_croak_no_modify();
5858 if (SvMAGICAL(sv) || (how == PERL_MAGIC_taint && SvTYPE(sv) >= SVt_PVMG)) {
5859 if (SvMAGIC(sv) && (mg = mg_find(sv, how))) {
5860 /* sv_magic() refuses to add a magic of the same 'how' as an
5863 if (how == PERL_MAGIC_taint)
5869 /* Force pos to be stored as characters, not bytes. */
5870 if (SvMAGICAL(sv) && DO_UTF8(sv)
5871 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
5873 && mg->mg_flags & MGf_BYTES) {
5874 mg->mg_len = (SSize_t)sv_pos_b2u_flags(sv, (STRLEN)mg->mg_len,
5876 mg->mg_flags &= ~MGf_BYTES;
5879 /* Rest of work is done else where */
5880 mg = sv_magicext(sv,obj,how,vtable,name,namlen);
5883 case PERL_MAGIC_taint:
5886 case PERL_MAGIC_ext:
5887 case PERL_MAGIC_dbfile:
5894 S_sv_unmagicext_flags(pTHX_ SV *const sv, const int type, MGVTBL *vtbl, const U32 flags)
5901 if (SvTYPE(sv) < SVt_PVMG || !SvMAGIC(sv))
5903 mgp = &(((XPVMG*) SvANY(sv))->xmg_u.xmg_magic);
5904 for (mg = *mgp; mg; mg = *mgp) {
5905 const MGVTBL* const virt = mg->mg_virtual;
5906 if (mg->mg_type == type && (!flags || virt == vtbl)) {
5907 *mgp = mg->mg_moremagic;
5908 if (virt && virt->svt_free)
5909 virt->svt_free(aTHX_ sv, mg);
5910 if (mg->mg_ptr && mg->mg_type != PERL_MAGIC_regex_global) {
5912 Safefree(mg->mg_ptr);
5913 else if (mg->mg_len == HEf_SVKEY)
5914 SvREFCNT_dec(MUTABLE_SV(mg->mg_ptr));
5915 else if (mg->mg_type == PERL_MAGIC_utf8)
5916 Safefree(mg->mg_ptr);
5918 if (mg->mg_flags & MGf_REFCOUNTED)
5919 SvREFCNT_dec(mg->mg_obj);
5923 mgp = &mg->mg_moremagic;
5926 if (SvMAGICAL(sv)) /* if we're under save_magic, wait for restore_magic; */
5927 mg_magical(sv); /* else fix the flags now */
5936 =for apidoc sv_unmagic
5938 Removes all magic of type C<type> from an SV.
5944 Perl_sv_unmagic(pTHX_ SV *const sv, const int type)
5946 PERL_ARGS_ASSERT_SV_UNMAGIC;
5947 return S_sv_unmagicext_flags(aTHX_ sv, type, NULL, 0);
5951 =for apidoc sv_unmagicext
5953 Removes all magic of type C<type> with the specified C<vtbl> from an SV.
5959 Perl_sv_unmagicext(pTHX_ SV *const sv, const int type, MGVTBL *vtbl)
5961 PERL_ARGS_ASSERT_SV_UNMAGICEXT;
5962 return S_sv_unmagicext_flags(aTHX_ sv, type, vtbl, 1);
5966 =for apidoc sv_rvweaken
5968 Weaken a reference: set the C<SvWEAKREF> flag on this RV; give the
5969 referred-to SV C<PERL_MAGIC_backref> magic if it hasn't already; and
5970 push a back-reference to this RV onto the array of backreferences
5971 associated with that magic. If the RV is magical, set magic will be
5972 called after the RV is cleared. Silently ignores C<undef> and warns
5973 on already-weak references.
5979 Perl_sv_rvweaken(pTHX_ SV *const sv)
5983 PERL_ARGS_ASSERT_SV_RVWEAKEN;
5985 if (!SvOK(sv)) /* let undefs pass */
5988 Perl_croak(aTHX_ "Can't weaken a nonreference");
5989 else if (SvWEAKREF(sv)) {
5990 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is already weak");
5993 else if (SvREADONLY(sv)) croak_no_modify();
5995 Perl_sv_add_backref(aTHX_ tsv, sv);
5997 SvREFCNT_dec_NN(tsv);
6002 =for apidoc sv_rvunweaken
6004 Unweaken a reference: Clear the C<SvWEAKREF> flag on this RV; remove
6005 the backreference to this RV from the array of backreferences
6006 associated with the target SV, increment the refcount of the target.
6007 Silently ignores C<undef> and warns on non-weak references.
6013 Perl_sv_rvunweaken(pTHX_ SV *const sv)
6017 PERL_ARGS_ASSERT_SV_RVUNWEAKEN;
6019 if (!SvOK(sv)) /* let undefs pass */
6022 Perl_croak(aTHX_ "Can't unweaken a nonreference");
6023 else if (!SvWEAKREF(sv)) {
6024 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is not weak");
6027 else if (SvREADONLY(sv)) croak_no_modify();
6032 SvREFCNT_inc_NN(tsv);
6033 Perl_sv_del_backref(aTHX_ tsv, sv);
6038 =for apidoc sv_get_backrefs
6040 If C<sv> is the target of a weak reference then it returns the back
6041 references structure associated with the sv; otherwise return C<NULL>.
6043 When returning a non-null result the type of the return is relevant. If it
6044 is an AV then the elements of the AV are the weak reference RVs which
6045 point at this item. If it is any other type then the item itself is the
6048 See also C<Perl_sv_add_backref()>, C<Perl_sv_del_backref()>,
6049 C<Perl_sv_kill_backrefs()>
6055 Perl_sv_get_backrefs(SV *const sv)
6059 PERL_ARGS_ASSERT_SV_GET_BACKREFS;
6061 /* find slot to store array or singleton backref */
6063 if (SvTYPE(sv) == SVt_PVHV) {
6065 struct xpvhv_aux * const iter = HvAUX((HV *)sv);
6066 backrefs = (SV *)iter->xhv_backreferences;
6068 } else if (SvMAGICAL(sv)) {
6069 MAGIC *mg = mg_find(sv, PERL_MAGIC_backref);
6071 backrefs = mg->mg_obj;
6076 /* Give tsv backref magic if it hasn't already got it, then push a
6077 * back-reference to sv onto the array associated with the backref magic.
6079 * As an optimisation, if there's only one backref and it's not an AV,
6080 * store it directly in the HvAUX or mg_obj slot, avoiding the need to
6081 * allocate an AV. (Whether the slot holds an AV tells us whether this is
6085 /* A discussion about the backreferences array and its refcount:
6087 * The AV holding the backreferences is pointed to either as the mg_obj of
6088 * PERL_MAGIC_backref, or in the specific case of a HV, from the
6089 * xhv_backreferences field. The array is created with a refcount
6090 * of 2. This means that if during global destruction the array gets
6091 * picked on before its parent to have its refcount decremented by the
6092 * random zapper, it won't actually be freed, meaning it's still there for
6093 * when its parent gets freed.
6095 * When the parent SV is freed, the extra ref is killed by
6096 * Perl_sv_kill_backrefs. The other ref is killed, in the case of magic,
6097 * by mg_free() / MGf_REFCOUNTED, or for a hash, by Perl_hv_kill_backrefs.
6099 * When a single backref SV is stored directly, it is not reference
6104 Perl_sv_add_backref(pTHX_ SV *const tsv, SV *const sv)
6110 PERL_ARGS_ASSERT_SV_ADD_BACKREF;
6112 /* find slot to store array or singleton backref */
6114 if (SvTYPE(tsv) == SVt_PVHV) {
6115 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
6118 mg = mg_find(tsv, PERL_MAGIC_backref);
6120 mg = sv_magicext(tsv, NULL, PERL_MAGIC_backref, &PL_vtbl_backref, NULL, 0);
6121 svp = &(mg->mg_obj);
6124 /* create or retrieve the array */
6126 if ( (!*svp && SvTYPE(sv) == SVt_PVAV)
6127 || (*svp && SvTYPE(*svp) != SVt_PVAV)
6131 mg->mg_flags |= MGf_REFCOUNTED;
6134 SvREFCNT_inc_simple_void_NN(av);
6135 /* av now has a refcnt of 2; see discussion above */
6136 av_extend(av, *svp ? 2 : 1);
6138 /* move single existing backref to the array */
6139 AvARRAY(av)[++AvFILLp(av)] = *svp; /* av_push() */
6144 av = MUTABLE_AV(*svp);
6146 /* optimisation: store single backref directly in HvAUX or mg_obj */
6150 assert(SvTYPE(av) == SVt_PVAV);
6151 if (AvFILLp(av) >= AvMAX(av)) {
6152 av_extend(av, AvFILLp(av)+1);
6155 /* push new backref */
6156 AvARRAY(av)[++AvFILLp(av)] = sv; /* av_push() */
6159 /* delete a back-reference to ourselves from the backref magic associated
6160 * with the SV we point to.
6164 Perl_sv_del_backref(pTHX_ SV *const tsv, SV *const sv)
6168 PERL_ARGS_ASSERT_SV_DEL_BACKREF;
6170 if (SvTYPE(tsv) == SVt_PVHV) {
6172 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
6174 else if (SvIS_FREED(tsv) && PL_phase == PERL_PHASE_DESTRUCT) {
6175 /* It's possible for the the last (strong) reference to tsv to have
6176 become freed *before* the last thing holding a weak reference.
6177 If both survive longer than the backreferences array, then when
6178 the referent's reference count drops to 0 and it is freed, it's
6179 not able to chase the backreferences, so they aren't NULLed.
6181 For example, a CV holds a weak reference to its stash. If both the
6182 CV and the stash survive longer than the backreferences array,
6183 and the CV gets picked for the SvBREAK() treatment first,
6184 *and* it turns out that the stash is only being kept alive because
6185 of an our variable in the pad of the CV, then midway during CV
6186 destruction the stash gets freed, but CvSTASH() isn't set to NULL.
6187 It ends up pointing to the freed HV. Hence it's chased in here, and
6188 if this block wasn't here, it would hit the !svp panic just below.
6190 I don't believe that "better" destruction ordering is going to help
6191 here - during global destruction there's always going to be the
6192 chance that something goes out of order. We've tried to make it
6193 foolproof before, and it only resulted in evolutionary pressure on
6194 fools. Which made us look foolish for our hubris. :-(
6200 = SvMAGICAL(tsv) ? mg_find(tsv, PERL_MAGIC_backref) : NULL;
6201 svp = mg ? &(mg->mg_obj) : NULL;
6205 Perl_croak(aTHX_ "panic: del_backref, svp=0");
6207 /* It's possible that sv is being freed recursively part way through the
6208 freeing of tsv. If this happens, the backreferences array of tsv has
6209 already been freed, and so svp will be NULL. If this is the case,
6210 we should not panic. Instead, nothing needs doing, so return. */
6211 if (PL_phase == PERL_PHASE_DESTRUCT && SvREFCNT(tsv) == 0)
6213 Perl_croak(aTHX_ "panic: del_backref, *svp=%p phase=%s refcnt=%" UVuf,
6214 (void*)*svp, PL_phase_names[PL_phase], (UV)SvREFCNT(tsv));
6217 if (SvTYPE(*svp) == SVt_PVAV) {
6221 AV * const av = (AV*)*svp;
6223 assert(!SvIS_FREED(av));
6227 /* for an SV with N weak references to it, if all those
6228 * weak refs are deleted, then sv_del_backref will be called
6229 * N times and O(N^2) compares will be done within the backref
6230 * array. To ameliorate this potential slowness, we:
6231 * 1) make sure this code is as tight as possible;
6232 * 2) when looking for SV, look for it at both the head and tail of the
6233 * array first before searching the rest, since some create/destroy
6234 * patterns will cause the backrefs to be freed in order.
6241 SV **p = &svp[fill];
6242 SV *const topsv = *p;
6249 /* We weren't the last entry.
6250 An unordered list has this property that you
6251 can take the last element off the end to fill
6252 the hole, and it's still an unordered list :-)
6258 break; /* should only be one */
6265 AvFILLp(av) = fill-1;
6267 else if (SvIS_FREED(*svp) && PL_phase == PERL_PHASE_DESTRUCT) {
6268 /* freed AV; skip */
6271 /* optimisation: only a single backref, stored directly */
6273 Perl_croak(aTHX_ "panic: del_backref, *svp=%p, sv=%p",
6274 (void*)*svp, (void*)sv);
6281 Perl_sv_kill_backrefs(pTHX_ SV *const sv, AV *const av)
6287 PERL_ARGS_ASSERT_SV_KILL_BACKREFS;
6292 /* after multiple passes through Perl_sv_clean_all() for a thingy
6293 * that has badly leaked, the backref array may have gotten freed,
6294 * since we only protect it against 1 round of cleanup */
6295 if (SvIS_FREED(av)) {
6296 if (PL_in_clean_all) /* All is fair */
6299 "panic: magic_killbackrefs (freed backref AV/SV)");
6303 is_array = (SvTYPE(av) == SVt_PVAV);
6305 assert(!SvIS_FREED(av));
6308 last = svp + AvFILLp(av);
6311 /* optimisation: only a single backref, stored directly */
6317 while (svp <= last) {
6319 SV *const referrer = *svp;
6320 if (SvWEAKREF(referrer)) {
6321 /* XXX Should we check that it hasn't changed? */
6322 assert(SvROK(referrer));
6323 SvRV_set(referrer, 0);
6325 SvWEAKREF_off(referrer);
6326 SvSETMAGIC(referrer);
6327 } else if (SvTYPE(referrer) == SVt_PVGV ||
6328 SvTYPE(referrer) == SVt_PVLV) {
6329 assert(SvTYPE(sv) == SVt_PVHV); /* stash backref */
6330 /* You lookin' at me? */
6331 assert(GvSTASH(referrer));
6332 assert(GvSTASH(referrer) == (const HV *)sv);
6333 GvSTASH(referrer) = 0;
6334 } else if (SvTYPE(referrer) == SVt_PVCV ||
6335 SvTYPE(referrer) == SVt_PVFM) {
6336 if (SvTYPE(sv) == SVt_PVHV) { /* stash backref */
6337 /* You lookin' at me? */
6338 assert(CvSTASH(referrer));
6339 assert(CvSTASH(referrer) == (const HV *)sv);
6340 SvANY(MUTABLE_CV(referrer))->xcv_stash = 0;
6343 assert(SvTYPE(sv) == SVt_PVGV);
6344 /* You lookin' at me? */
6345 assert(CvGV(referrer));
6346 assert(CvGV(referrer) == (const GV *)sv);
6347 anonymise_cv_maybe(MUTABLE_GV(sv),
6348 MUTABLE_CV(referrer));
6353 "panic: magic_killbackrefs (flags=%" UVxf ")",
6354 (UV)SvFLAGS(referrer));
6365 SvREFCNT_dec_NN(av); /* remove extra count added by sv_add_backref() */
6371 =for apidoc sv_insert
6373 Inserts a string at the specified offset/length within the SV. Similar to
6374 the Perl C<substr()> function. Handles get magic.
6376 =for apidoc sv_insert_flags
6378 Same as C<sv_insert>, but the extra C<flags> are passed to the
6379 C<SvPV_force_flags> that applies to C<bigstr>.
6385 Perl_sv_insert_flags(pTHX_ SV *const bigstr, const STRLEN offset, const STRLEN len, const char *little, const STRLEN littlelen, const U32 flags)
6391 SSize_t i; /* better be sizeof(STRLEN) or bad things happen */
6394 PERL_ARGS_ASSERT_SV_INSERT_FLAGS;
6396 SvPV_force_flags(bigstr, curlen, flags);
6397 (void)SvPOK_only_UTF8(bigstr);
6399 if (little >= SvPVX(bigstr) &&
6400 little < SvPVX(bigstr) + (SvLEN(bigstr) ? SvLEN(bigstr) : SvCUR(bigstr))) {
6401 /* little is a pointer to within bigstr, since we can reallocate bigstr,
6402 or little...little+littlelen might overlap offset...offset+len we make a copy
6404 little = savepvn(little, littlelen);
6408 if (offset + len > curlen) {
6409 SvGROW(bigstr, offset+len+1);
6410 Zero(SvPVX(bigstr)+curlen, offset+len-curlen, char);
6411 SvCUR_set(bigstr, offset+len);
6415 i = littlelen - len;
6416 if (i > 0) { /* string might grow */
6417 big = SvGROW(bigstr, SvCUR(bigstr) + i + 1);
6418 mid = big + offset + len;
6419 midend = bigend = big + SvCUR(bigstr);
6422 while (midend > mid) /* shove everything down */
6423 *--bigend = *--midend;
6424 Move(little,big+offset,littlelen,char);
6425 SvCUR_set(bigstr, SvCUR(bigstr) + i);
6430 Move(little,SvPVX(bigstr)+offset,len,char);
6435 big = SvPVX(bigstr);
6438 bigend = big + SvCUR(bigstr);
6440 if (midend > bigend)
6441 Perl_croak(aTHX_ "panic: sv_insert, midend=%p, bigend=%p",
6444 if (mid - big > bigend - midend) { /* faster to shorten from end */
6446 Move(little, mid, littlelen,char);
6449 i = bigend - midend;
6451 Move(midend, mid, i,char);
6455 SvCUR_set(bigstr, mid - big);
6457 else if ((i = mid - big)) { /* faster from front */
6458 midend -= littlelen;
6460 Move(big, midend - i, i, char);
6461 sv_chop(bigstr,midend-i);
6463 Move(little, mid, littlelen,char);
6465 else if (littlelen) {
6466 midend -= littlelen;
6467 sv_chop(bigstr,midend);
6468 Move(little,midend,littlelen,char);
6471 sv_chop(bigstr,midend);
6477 =for apidoc sv_replace
6479 Make the first argument a copy of the second, then delete the original.
6480 The target SV physically takes over ownership of the body of the source SV
6481 and inherits its flags; however, the target keeps any magic it owns,
6482 and any magic in the source is discarded.
6483 Note that this is a rather specialist SV copying operation; most of the
6484 time you'll want to use C<sv_setsv> or one of its many macro front-ends.
6490 Perl_sv_replace(pTHX_ SV *const sv, SV *const nsv)
6492 const U32 refcnt = SvREFCNT(sv);
6494 PERL_ARGS_ASSERT_SV_REPLACE;
6496 SV_CHECK_THINKFIRST_COW_DROP(sv);
6497 if (SvREFCNT(nsv) != 1) {
6498 Perl_croak(aTHX_ "panic: reference miscount on nsv in sv_replace()"
6499 " (%" UVuf " != 1)", (UV) SvREFCNT(nsv));
6501 if (SvMAGICAL(sv)) {
6505 sv_upgrade(nsv, SVt_PVMG);
6506 SvMAGIC_set(nsv, SvMAGIC(sv));
6507 SvFLAGS(nsv) |= SvMAGICAL(sv);
6509 SvMAGIC_set(sv, NULL);
6513 assert(!SvREFCNT(sv));
6514 #ifdef DEBUG_LEAKING_SCALARS
6515 sv->sv_flags = nsv->sv_flags;
6516 sv->sv_any = nsv->sv_any;
6517 sv->sv_refcnt = nsv->sv_refcnt;
6518 sv->sv_u = nsv->sv_u;
6520 StructCopy(nsv,sv,SV);
6522 if(SvTYPE(sv) == SVt_IV) {
6523 SET_SVANY_FOR_BODYLESS_IV(sv);
6527 SvREFCNT(sv) = refcnt;
6528 SvFLAGS(nsv) |= SVTYPEMASK; /* Mark as freed */
6533 /* We're about to free a GV which has a CV that refers back to us.
6534 * If that CV will outlive us, make it anonymous (i.e. fix up its CvGV
6538 S_anonymise_cv_maybe(pTHX_ GV *gv, CV* cv)
6543 PERL_ARGS_ASSERT_ANONYMISE_CV_MAYBE;
6546 assert(SvREFCNT(gv) == 0);
6547 assert(isGV(gv) && isGV_with_GP(gv));
6549 assert(!CvANON(cv));
6550 assert(CvGV(cv) == gv);
6551 assert(!CvNAMED(cv));
6553 /* will the CV shortly be freed by gp_free() ? */
6554 if (GvCV(gv) == cv && GvGP(gv)->gp_refcnt < 2 && SvREFCNT(cv) < 2) {
6555 SvANY(cv)->xcv_gv_u.xcv_gv = NULL;
6559 /* if not, anonymise: */
6560 gvname = (GvSTASH(gv) && HvNAME(GvSTASH(gv)) && HvENAME(GvSTASH(gv)))
6561 ? newSVhek(HvENAME_HEK(GvSTASH(gv)))
6562 : newSVpvn_flags( "__ANON__", 8, 0 );
6563 sv_catpvs(gvname, "::__ANON__");
6564 anongv = gv_fetchsv(gvname, GV_ADDMULTI, SVt_PVCV);
6565 SvREFCNT_dec_NN(gvname);
6569 SvANY(cv)->xcv_gv_u.xcv_gv = MUTABLE_GV(SvREFCNT_inc(anongv));
6574 =for apidoc sv_clear
6576 Clear an SV: call any destructors, free up any memory used by the body,
6577 and free the body itself. The SV's head is I<not> freed, although
6578 its type is set to all 1's so that it won't inadvertently be assumed
6579 to be live during global destruction etc.
6580 This function should only be called when C<REFCNT> is zero. Most of the time
6581 you'll want to call C<sv_free()> (or its macro wrapper C<SvREFCNT_dec>)
6588 Perl_sv_clear(pTHX_ SV *const orig_sv)
6593 const struct body_details *sv_type_details;
6597 STRLEN hash_index = 0; /* initialise to make Coverity et al happy.
6598 Not strictly necessary */
6600 PERL_ARGS_ASSERT_SV_CLEAR;
6602 /* within this loop, sv is the SV currently being freed, and
6603 * iter_sv is the most recent AV or whatever that's being iterated
6604 * over to provide more SVs */
6610 assert(SvREFCNT(sv) == 0);
6611 assert(SvTYPE(sv) != (svtype)SVTYPEMASK);
6613 if (type <= SVt_IV) {
6614 /* See the comment in sv.h about the collusion between this
6615 * early return and the overloading of the NULL slots in the
6619 SvFLAGS(sv) &= SVf_BREAK;
6620 SvFLAGS(sv) |= SVTYPEMASK;
6624 /* objs are always >= MG, but pad names use the SVs_OBJECT flag
6625 for another purpose */
6626 assert(!SvOBJECT(sv) || type >= SVt_PVMG);
6628 if (type >= SVt_PVMG) {
6630 if (!curse(sv, 1)) goto get_next_sv;
6631 type = SvTYPE(sv); /* destructor may have changed it */
6633 /* Free back-references before magic, in case the magic calls
6634 * Perl code that has weak references to sv. */
6635 if (type == SVt_PVHV) {
6636 Perl_hv_kill_backrefs(aTHX_ MUTABLE_HV(sv));
6640 else if (SvMAGIC(sv)) {
6641 /* Free back-references before other types of magic. */
6642 sv_unmagic(sv, PERL_MAGIC_backref);
6648 /* case SVt_INVLIST: */
6651 IoIFP(sv) != PerlIO_stdin() &&
6652 IoIFP(sv) != PerlIO_stdout() &&
6653 IoIFP(sv) != PerlIO_stderr() &&
6654 !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6656 io_close(MUTABLE_IO(sv), NULL, FALSE,
6657 (IoTYPE(sv) == IoTYPE_WRONLY ||
6658 IoTYPE(sv) == IoTYPE_RDWR ||
6659 IoTYPE(sv) == IoTYPE_APPEND));
6661 if (IoDIRP(sv) && !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6662 PerlDir_close(IoDIRP(sv));
6663 IoDIRP(sv) = (DIR*)NULL;
6664 Safefree(IoTOP_NAME(sv));
6665 Safefree(IoFMT_NAME(sv));
6666 Safefree(IoBOTTOM_NAME(sv));
6667 if ((const GV *)sv == PL_statgv)
6671 /* FIXME for plugins */
6672 pregfree2((REGEXP*) sv);
6676 cv_undef(MUTABLE_CV(sv));
6677 /* If we're in a stash, we don't own a reference to it.
6678 * However it does have a back reference to us, which needs to
6680 if ((stash = CvSTASH(sv)))
6681 sv_del_backref(MUTABLE_SV(stash), sv);
6684 if (PL_last_swash_hv == (const HV *)sv) {
6685 PL_last_swash_hv = NULL;
6687 if (HvTOTALKEYS((HV*)sv) > 0) {
6689 /* this statement should match the one at the beginning of
6690 * hv_undef_flags() */
6691 if ( PL_phase != PERL_PHASE_DESTRUCT
6692 && (hek = HvNAME_HEK((HV*)sv)))
6694 if (PL_stashcache) {
6695 DEBUG_o(Perl_deb(aTHX_
6696 "sv_clear clearing PL_stashcache for '%" HEKf
6699 (void)hv_deletehek(PL_stashcache,
6702 hv_name_set((HV*)sv, NULL, 0, 0);
6705 /* save old iter_sv in unused SvSTASH field */
6706 assert(!SvOBJECT(sv));
6707 SvSTASH(sv) = (HV*)iter_sv;
6710 /* save old hash_index in unused SvMAGIC field */
6711 assert(!SvMAGICAL(sv));
6712 assert(!SvMAGIC(sv));
6713 ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index = hash_index;
6716 next_sv = Perl_hfree_next_entry(aTHX_ (HV*)sv, &hash_index);
6717 goto get_next_sv; /* process this new sv */
6719 /* free empty hash */
6720 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6721 assert(!HvARRAY((HV*)sv));
6725 AV* av = MUTABLE_AV(sv);
6726 if (PL_comppad == av) {
6730 if (AvREAL(av) && AvFILLp(av) > -1) {
6731 next_sv = AvARRAY(av)[AvFILLp(av)--];
6732 /* save old iter_sv in top-most slot of AV,
6733 * and pray that it doesn't get wiped in the meantime */
6734 AvARRAY(av)[AvMAX(av)] = iter_sv;
6736 goto get_next_sv; /* process this new sv */
6738 Safefree(AvALLOC(av));
6743 if (LvTYPE(sv) == 'T') { /* for tie: return HE to pool */
6744 SvREFCNT_dec(HeKEY_sv((HE*)LvTARG(sv)));
6745 HeNEXT((HE*)LvTARG(sv)) = PL_hv_fetch_ent_mh;
6746 PL_hv_fetch_ent_mh = (HE*)LvTARG(sv);
6748 else if (LvTYPE(sv) != 't') /* unless tie: unrefcnted fake SV** */
6749 SvREFCNT_dec(LvTARG(sv));
6751 /* SvLEN points to a regex body. Free the body, then
6752 * set SvLEN to whatever value was in the now-freed
6753 * regex body. The PVX buffer is shared by multiple re's
6754 * and only freed once, by the re whose len in non-null */
6755 STRLEN len = ReANY(sv)->xpv_len;
6756 pregfree2((REGEXP*) sv);
6757 SvLEN_set((sv), len);
6762 if (isGV_with_GP(sv)) {
6763 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
6764 && HvENAME_get(stash))
6765 mro_method_changed_in(stash);
6766 gp_free(MUTABLE_GV(sv));
6768 unshare_hek(GvNAME_HEK(sv));
6769 /* If we're in a stash, we don't own a reference to it.
6770 * However it does have a back reference to us, which
6771 * needs to be cleared. */
6772 if ((stash = GvSTASH(sv)))
6773 sv_del_backref(MUTABLE_SV(stash), sv);
6775 /* FIXME. There are probably more unreferenced pointers to SVs
6776 * in the interpreter struct that we should check and tidy in
6777 * a similar fashion to this: */
6778 /* See also S_sv_unglob, which does the same thing. */
6779 if ((const GV *)sv == PL_last_in_gv)
6780 PL_last_in_gv = NULL;
6781 else if ((const GV *)sv == PL_statgv)
6783 else if ((const GV *)sv == PL_stderrgv)
6792 /* Don't bother with SvOOK_off(sv); as we're only going to
6796 SvOOK_offset(sv, offset);
6797 SvPV_set(sv, SvPVX_mutable(sv) - offset);
6798 /* Don't even bother with turning off the OOK flag. */
6803 SV * const target = SvRV(sv);
6805 sv_del_backref(target, sv);
6811 else if (SvPVX_const(sv)
6812 && !(SvTYPE(sv) == SVt_PVIO
6813 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6818 PerlIO_printf(Perl_debug_log, "Copy on write: clear\n");
6823 if (CowREFCNT(sv)) {
6830 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6835 Safefree(SvPVX_mutable(sv));
6839 else if (SvPVX_const(sv) && SvLEN(sv)
6840 && !(SvTYPE(sv) == SVt_PVIO
6841 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6842 Safefree(SvPVX_mutable(sv));
6843 else if (SvPVX_const(sv) && SvIsCOW(sv)) {
6844 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6854 SvFLAGS(sv) &= SVf_BREAK;
6855 SvFLAGS(sv) |= SVTYPEMASK;
6857 sv_type_details = bodies_by_type + type;
6858 if (sv_type_details->arena) {
6859 del_body(((char *)SvANY(sv) + sv_type_details->offset),
6860 &PL_body_roots[type]);
6862 else if (sv_type_details->body_size) {
6863 safefree(SvANY(sv));
6867 /* caller is responsible for freeing the head of the original sv */
6868 if (sv != orig_sv && !SvREFCNT(sv))
6871 /* grab and free next sv, if any */
6879 else if (!iter_sv) {
6881 } else if (SvTYPE(iter_sv) == SVt_PVAV) {
6882 AV *const av = (AV*)iter_sv;
6883 if (AvFILLp(av) > -1) {
6884 sv = AvARRAY(av)[AvFILLp(av)--];
6886 else { /* no more elements of current AV to free */
6889 /* restore previous value, squirrelled away */
6890 iter_sv = AvARRAY(av)[AvMAX(av)];
6891 Safefree(AvALLOC(av));
6894 } else if (SvTYPE(iter_sv) == SVt_PVHV) {
6895 sv = Perl_hfree_next_entry(aTHX_ (HV*)iter_sv, &hash_index);
6896 if (!sv && !HvTOTALKEYS((HV *)iter_sv)) {
6897 /* no more elements of current HV to free */
6900 /* Restore previous values of iter_sv and hash_index,
6901 * squirrelled away */
6902 assert(!SvOBJECT(sv));
6903 iter_sv = (SV*)SvSTASH(sv);
6904 assert(!SvMAGICAL(sv));
6905 hash_index = ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index;
6907 /* perl -DA does not like rubbish in SvMAGIC. */
6911 /* free any remaining detritus from the hash struct */
6912 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6913 assert(!HvARRAY((HV*)sv));
6918 /* unrolled SvREFCNT_dec and sv_free2 follows: */
6922 if (!SvREFCNT(sv)) {
6926 if (--(SvREFCNT(sv)))
6930 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6931 "Attempt to free temp prematurely: SV 0x%" UVxf
6932 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6936 if (SvIMMORTAL(sv)) {
6937 /* make sure SvREFCNT(sv)==0 happens very seldom */
6938 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6947 /* This routine curses the sv itself, not the object referenced by sv. So
6948 sv does not have to be ROK. */
6951 S_curse(pTHX_ SV * const sv, const bool check_refcnt) {
6952 PERL_ARGS_ASSERT_CURSE;
6953 assert(SvOBJECT(sv));
6955 if (PL_defstash && /* Still have a symbol table? */
6961 stash = SvSTASH(sv);
6962 assert(SvTYPE(stash) == SVt_PVHV);
6963 if (HvNAME(stash)) {
6964 CV* destructor = NULL;
6965 struct mro_meta *meta;
6967 assert (SvOOK(stash));
6969 DEBUG_o( Perl_deb(aTHX_ "Looking for DESTROY method for %s\n",
6972 /* don't make this an initialization above the assert, since it needs
6974 meta = HvMROMETA(stash);
6975 if (meta->destroy_gen && meta->destroy_gen == PL_sub_generation) {
6976 destructor = meta->destroy;
6977 DEBUG_o( Perl_deb(aTHX_ "Using cached DESTROY method %p for %s\n",
6978 (void *)destructor, HvNAME(stash)) );
6981 bool autoload = FALSE;
6983 gv_fetchmeth_pvn(stash, S_destroy, S_destroy_len, -1, 0);
6985 destructor = GvCV(gv);
6987 gv = gv_autoload_pvn(stash, S_destroy, S_destroy_len,
6988 GV_AUTOLOAD_ISMETHOD);
6990 destructor = GvCV(gv);
6994 /* we don't cache AUTOLOAD for DESTROY, since this code
6995 would then need to set $__PACKAGE__::AUTOLOAD, or the
6996 equivalent for XS AUTOLOADs */
6998 meta->destroy_gen = PL_sub_generation;
6999 meta->destroy = destructor;
7001 DEBUG_o( Perl_deb(aTHX_ "Set cached DESTROY method %p for %s\n",
7002 (void *)destructor, HvNAME(stash)) );
7005 DEBUG_o( Perl_deb(aTHX_ "Not caching AUTOLOAD for DESTROY method for %s\n",
7009 assert(!destructor || SvTYPE(destructor) == SVt_PVCV);
7011 /* A constant subroutine can have no side effects, so
7012 don't bother calling it. */
7013 && !CvCONST(destructor)
7014 /* Don't bother calling an empty destructor or one that
7015 returns immediately. */
7016 && (CvISXSUB(destructor)
7017 || (CvSTART(destructor)
7018 && (CvSTART(destructor)->op_next->op_type
7020 && (CvSTART(destructor)->op_next->op_type
7022 || CvSTART(destructor)->op_next->op_next->op_type
7028 SV* const tmpref = newRV(sv);
7029 SvREADONLY_on(tmpref); /* DESTROY() could be naughty */
7031 PUSHSTACKi(PERLSI_DESTROY);
7036 call_sv(MUTABLE_SV(destructor),
7037 G_DISCARD|G_EVAL|G_KEEPERR|G_VOID);
7041 if(SvREFCNT(tmpref) < 2) {
7042 /* tmpref is not kept alive! */
7044 SvRV_set(tmpref, NULL);
7047 SvREFCNT_dec_NN(tmpref);
7050 } while (SvOBJECT(sv) && SvSTASH(sv) != stash);
7053 if (check_refcnt && SvREFCNT(sv)) {
7054 if (PL_in_clean_objs)
7056 "DESTROY created new reference to dead object '%" HEKf "'",
7057 HEKfARG(HvNAME_HEK(stash)));
7058 /* DESTROY gave object new lease on life */
7064 HV * const stash = SvSTASH(sv);
7065 /* Curse before freeing the stash, as freeing the stash could cause
7066 a recursive call into S_curse. */
7067 SvOBJECT_off(sv); /* Curse the object. */
7068 SvSTASH_set(sv,0); /* SvREFCNT_dec may try to read this */
7069 SvREFCNT_dec(stash); /* possibly of changed persuasion */
7075 =for apidoc sv_newref
7077 Increment an SV's reference count. Use the C<SvREFCNT_inc()> wrapper
7084 Perl_sv_newref(pTHX_ SV *const sv)
7086 PERL_UNUSED_CONTEXT;
7095 Decrement an SV's reference count, and if it drops to zero, call
7096 C<sv_clear> to invoke destructors and free up any memory used by
7097 the body; finally, deallocating the SV's head itself.
7098 Normally called via a wrapper macro C<SvREFCNT_dec>.
7104 Perl_sv_free(pTHX_ SV *const sv)
7110 /* Private helper function for SvREFCNT_dec().
7111 * Called with rc set to original SvREFCNT(sv), where rc == 0 or 1 */
7114 Perl_sv_free2(pTHX_ SV *const sv, const U32 rc)
7118 PERL_ARGS_ASSERT_SV_FREE2;
7120 if (LIKELY( rc == 1 )) {
7126 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
7127 "Attempt to free temp prematurely: SV 0x%" UVxf
7128 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
7132 if (SvIMMORTAL(sv)) {
7133 /* make sure SvREFCNT(sv)==0 happens very seldom */
7134 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
7138 if (! SvREFCNT(sv)) /* may have have been resurrected */
7143 /* handle exceptional cases */
7147 if (SvFLAGS(sv) & SVf_BREAK)
7148 /* this SV's refcnt has been artificially decremented to
7149 * trigger cleanup */
7151 if (PL_in_clean_all) /* All is fair */
7153 if (SvIMMORTAL(sv)) {
7154 /* make sure SvREFCNT(sv)==0 happens very seldom */
7155 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
7158 if (ckWARN_d(WARN_INTERNAL)) {
7159 #ifdef DEBUG_LEAKING_SCALARS_FORK_DUMP
7160 Perl_dump_sv_child(aTHX_ sv);
7162 #ifdef DEBUG_LEAKING_SCALARS
7165 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7166 if (PL_warnhook == PERL_WARNHOOK_FATAL
7167 || ckDEAD(packWARN(WARN_INTERNAL))) {
7168 /* Don't let Perl_warner cause us to escape our fate: */
7172 /* This may not return: */
7173 Perl_warner(aTHX_ packWARN(WARN_INTERNAL),
7174 "Attempt to free unreferenced scalar: SV 0x%" UVxf
7175 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
7178 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7188 Returns the length of the string in the SV. Handles magic and type
7189 coercion and sets the UTF8 flag appropriately. See also C<L</SvCUR>>, which
7190 gives raw access to the C<xpv_cur> slot.
7196 Perl_sv_len(pTHX_ SV *const sv)
7203 (void)SvPV_const(sv, len);
7208 =for apidoc sv_len_utf8
7210 Returns the number of characters in the string in an SV, counting wide
7211 UTF-8 bytes as a single character. Handles magic and type coercion.
7217 * The length is cached in PERL_MAGIC_utf8, in the mg_len field. Also the
7218 * mg_ptr is used, by sv_pos_u2b() and sv_pos_b2u() - see the comments below.
7219 * (Note that the mg_len is not the length of the mg_ptr field.
7220 * This allows the cache to store the character length of the string without
7221 * needing to malloc() extra storage to attach to the mg_ptr.)
7226 Perl_sv_len_utf8(pTHX_ SV *const sv)
7232 return sv_len_utf8_nomg(sv);
7236 Perl_sv_len_utf8_nomg(pTHX_ SV * const sv)
7239 const U8 *s = (U8*)SvPV_nomg_const(sv, len);
7241 PERL_ARGS_ASSERT_SV_LEN_UTF8_NOMG;
7243 if (PL_utf8cache && SvUTF8(sv)) {
7245 MAGIC *mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_utf8) : NULL;
7247 if (mg && (mg->mg_len != -1 || mg->mg_ptr)) {
7248 if (mg->mg_len != -1)
7251 /* We can use the offset cache for a headstart.
7252 The longer value is stored in the first pair. */
7253 STRLEN *cache = (STRLEN *) mg->mg_ptr;
7255 ulen = cache[0] + Perl_utf8_length(aTHX_ s + cache[1],
7259 if (PL_utf8cache < 0) {
7260 const STRLEN real = Perl_utf8_length(aTHX_ s, s + len);
7261 assert_uft8_cache_coherent("sv_len_utf8", ulen, real, sv);
7265 ulen = Perl_utf8_length(aTHX_ s, s + len);
7266 utf8_mg_len_cache_update(sv, &mg, ulen);
7270 return SvUTF8(sv) ? Perl_utf8_length(aTHX_ s, s + len) : len;
7273 /* Walk forwards to find the byte corresponding to the passed in UTF-8
7276 S_sv_pos_u2b_forwards(const U8 *const start, const U8 *const send,
7277 STRLEN *const uoffset_p, bool *const at_end)
7279 const U8 *s = start;
7280 STRLEN uoffset = *uoffset_p;
7282 PERL_ARGS_ASSERT_SV_POS_U2B_FORWARDS;
7284 while (s < send && uoffset) {
7291 else if (s > send) {
7293 /* This is the existing behaviour. Possibly it should be a croak, as
7294 it's actually a bounds error */
7297 *uoffset_p -= uoffset;
7301 /* Given the length of the string in both bytes and UTF-8 characters, decide
7302 whether to walk forwards or backwards to find the byte corresponding to
7303 the passed in UTF-8 offset. */
7305 S_sv_pos_u2b_midway(const U8 *const start, const U8 *send,
7306 STRLEN uoffset, const STRLEN uend)
7308 STRLEN backw = uend - uoffset;
7310 PERL_ARGS_ASSERT_SV_POS_U2B_MIDWAY;
7312 if (uoffset < 2 * backw) {
7313 /* The assumption is that going forwards is twice the speed of going
7314 forward (that's where the 2 * backw comes from).
7315 (The real figure of course depends on the UTF-8 data.) */
7316 const U8 *s = start;
7318 while (s < send && uoffset--)
7328 while (UTF8_IS_CONTINUATION(*send))
7331 return send - start;
7334 /* For the string representation of the given scalar, find the byte
7335 corresponding to the passed in UTF-8 offset. uoffset0 and boffset0
7336 give another position in the string, *before* the sought offset, which
7337 (which is always true, as 0, 0 is a valid pair of positions), which should
7338 help reduce the amount of linear searching.
7339 If *mgp is non-NULL, it should point to the UTF-8 cache magic, which
7340 will be used to reduce the amount of linear searching. The cache will be
7341 created if necessary, and the found value offered to it for update. */
7343 S_sv_pos_u2b_cached(pTHX_ SV *const sv, MAGIC **const mgp, const U8 *const start,
7344 const U8 *const send, STRLEN uoffset,
7345 STRLEN uoffset0, STRLEN boffset0)
7347 STRLEN boffset = 0; /* Actually always set, but let's keep gcc happy. */
7349 bool at_end = FALSE;
7351 PERL_ARGS_ASSERT_SV_POS_U2B_CACHED;
7353 assert (uoffset >= uoffset0);
7358 if (!SvREADONLY(sv) && !SvGMAGICAL(sv) && SvPOK(sv)
7360 && (*mgp || (SvTYPE(sv) >= SVt_PVMG &&
7361 (*mgp = mg_find(sv, PERL_MAGIC_utf8))))) {
7362 if ((*mgp)->mg_ptr) {
7363 STRLEN *cache = (STRLEN *) (*mgp)->mg_ptr;
7364 if (cache[0] == uoffset) {
7365 /* An exact match. */
7368 if (cache[2] == uoffset) {
7369 /* An exact match. */
7373 if (cache[0] < uoffset) {
7374 /* The cache already knows part of the way. */
7375 if (cache[0] > uoffset0) {
7376 /* The cache knows more than the passed in pair */
7377 uoffset0 = cache[0];
7378 boffset0 = cache[1];
7380 if ((*mgp)->mg_len != -1) {
7381 /* And we know the end too. */
7383 + sv_pos_u2b_midway(start + boffset0, send,
7385 (*mgp)->mg_len - uoffset0);
7387 uoffset -= uoffset0;
7389 + sv_pos_u2b_forwards(start + boffset0,
7390 send, &uoffset, &at_end);
7391 uoffset += uoffset0;
7394 else if (cache[2] < uoffset) {
7395 /* We're between the two cache entries. */
7396 if (cache[2] > uoffset0) {
7397 /* and the cache knows more than the passed in pair */
7398 uoffset0 = cache[2];
7399 boffset0 = cache[3];
7403 + sv_pos_u2b_midway(start + boffset0,
7406 cache[0] - uoffset0);
7409 + sv_pos_u2b_midway(start + boffset0,
7412 cache[2] - uoffset0);
7416 else if ((*mgp)->mg_len != -1) {
7417 /* If we can take advantage of a passed in offset, do so. */
7418 /* In fact, offset0 is either 0, or less than offset, so don't
7419 need to worry about the other possibility. */
7421 + sv_pos_u2b_midway(start + boffset0, send,
7423 (*mgp)->mg_len - uoffset0);
7428 if (!found || PL_utf8cache < 0) {
7429 STRLEN real_boffset;
7430 uoffset -= uoffset0;
7431 real_boffset = boffset0 + sv_pos_u2b_forwards(start + boffset0,
7432 send, &uoffset, &at_end);
7433 uoffset += uoffset0;
7435 if (found && PL_utf8cache < 0)
7436 assert_uft8_cache_coherent("sv_pos_u2b_cache", boffset,
7438 boffset = real_boffset;
7441 if (PL_utf8cache && !SvGMAGICAL(sv) && SvPOK(sv)) {
7443 utf8_mg_len_cache_update(sv, mgp, uoffset);
7445 utf8_mg_pos_cache_update(sv, mgp, boffset, uoffset, send - start);
7452 =for apidoc sv_pos_u2b_flags
7454 Converts the offset from a count of UTF-8 chars from
7455 the start of the string, to a count of the equivalent number of bytes; if
7456 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7457 C<offset>, rather than from the start
7458 of the string. Handles type coercion.
7459 C<flags> is passed to C<SvPV_flags>, and usually should be
7460 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7466 * sv_pos_u2b_flags() uses, like sv_pos_b2u(), the mg_ptr of the potential
7467 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7468 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7473 Perl_sv_pos_u2b_flags(pTHX_ SV *const sv, STRLEN uoffset, STRLEN *const lenp,
7480 PERL_ARGS_ASSERT_SV_POS_U2B_FLAGS;
7482 start = (U8*)SvPV_flags(sv, len, flags);
7484 const U8 * const send = start + len;
7486 boffset = sv_pos_u2b_cached(sv, &mg, start, send, uoffset, 0, 0);
7489 && *lenp /* don't bother doing work for 0, as its bytes equivalent
7490 is 0, and *lenp is already set to that. */) {
7491 /* Convert the relative offset to absolute. */
7492 const STRLEN uoffset2 = uoffset + *lenp;
7493 const STRLEN boffset2
7494 = sv_pos_u2b_cached(sv, &mg, start, send, uoffset2,
7495 uoffset, boffset) - boffset;
7509 =for apidoc sv_pos_u2b
7511 Converts the value pointed to by C<offsetp> from a count of UTF-8 chars from
7512 the start of the string, to a count of the equivalent number of bytes; if
7513 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7514 the offset, rather than from the start of the string. Handles magic and
7517 Use C<sv_pos_u2b_flags> in preference, which correctly handles strings longer
7524 * sv_pos_u2b() uses, like sv_pos_b2u(), the mg_ptr of the potential
7525 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7526 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7530 /* This function is subject to size and sign problems */
7533 Perl_sv_pos_u2b(pTHX_ SV *const sv, I32 *const offsetp, I32 *const lenp)
7535 PERL_ARGS_ASSERT_SV_POS_U2B;
7538 STRLEN ulen = (STRLEN)*lenp;
7539 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, &ulen,
7540 SV_GMAGIC|SV_CONST_RETURN);
7543 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, NULL,
7544 SV_GMAGIC|SV_CONST_RETURN);
7549 S_utf8_mg_len_cache_update(pTHX_ SV *const sv, MAGIC **const mgp,
7552 PERL_ARGS_ASSERT_UTF8_MG_LEN_CACHE_UPDATE;
7553 if (SvREADONLY(sv) || SvGMAGICAL(sv) || !SvPOK(sv))
7556 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7557 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7558 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, &PL_vtbl_utf8, 0, 0);
7562 (*mgp)->mg_len = ulen;
7565 /* Create and update the UTF8 magic offset cache, with the proffered utf8/
7566 byte length pairing. The (byte) length of the total SV is passed in too,
7567 as blen, because for some (more esoteric) SVs, the call to SvPV_const()
7568 may not have updated SvCUR, so we can't rely on reading it directly.
7570 The proffered utf8/byte length pairing isn't used if the cache already has
7571 two pairs, and swapping either for the proffered pair would increase the
7572 RMS of the intervals between known byte offsets.
7574 The cache itself consists of 4 STRLEN values
7575 0: larger UTF-8 offset
7576 1: corresponding byte offset
7577 2: smaller UTF-8 offset
7578 3: corresponding byte offset
7580 Unused cache pairs have the value 0, 0.
7581 Keeping the cache "backwards" means that the invariant of
7582 cache[0] >= cache[2] is maintained even with empty slots, which means that
7583 the code that uses it doesn't need to worry if only 1 entry has actually
7584 been set to non-zero. It also makes the "position beyond the end of the
7585 cache" logic much simpler, as the first slot is always the one to start
7589 S_utf8_mg_pos_cache_update(pTHX_ SV *const sv, MAGIC **const mgp, const STRLEN byte,
7590 const STRLEN utf8, const STRLEN blen)
7594 PERL_ARGS_ASSERT_UTF8_MG_POS_CACHE_UPDATE;
7599 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7600 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7601 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, (MGVTBL*)&PL_vtbl_utf8, 0,
7603 (*mgp)->mg_len = -1;
7607 if (!(cache = (STRLEN *)(*mgp)->mg_ptr)) {
7608 Newxz(cache, PERL_MAGIC_UTF8_CACHESIZE * 2, STRLEN);
7609 (*mgp)->mg_ptr = (char *) cache;
7613 if (PL_utf8cache < 0 && SvPOKp(sv)) {
7614 /* SvPOKp() because, if sv is a reference, then SvPVX() is actually
7615 a pointer. Note that we no longer cache utf8 offsets on refer-
7616 ences, but this check is still a good idea, for robustness. */
7617 const U8 *start = (const U8 *) SvPVX_const(sv);
7618 const STRLEN realutf8 = utf8_length(start, start + byte);
7620 assert_uft8_cache_coherent("utf8_mg_pos_cache_update", utf8, realutf8,
7624 /* Cache is held with the later position first, to simplify the code
7625 that deals with unbounded ends. */
7627 ASSERT_UTF8_CACHE(cache);
7628 if (cache[1] == 0) {
7629 /* Cache is totally empty */
7632 } else if (cache[3] == 0) {
7633 if (byte > cache[1]) {
7634 /* New one is larger, so goes first. */
7635 cache[2] = cache[0];
7636 cache[3] = cache[1];
7644 /* float casts necessary? XXX */
7645 #define THREEWAY_SQUARE(a,b,c,d) \
7646 ((float)((d) - (c))) * ((float)((d) - (c))) \
7647 + ((float)((c) - (b))) * ((float)((c) - (b))) \
7648 + ((float)((b) - (a))) * ((float)((b) - (a)))
7650 /* Cache has 2 slots in use, and we know three potential pairs.
7651 Keep the two that give the lowest RMS distance. Do the
7652 calculation in bytes simply because we always know the byte
7653 length. squareroot has the same ordering as the positive value,
7654 so don't bother with the actual square root. */
7655 if (byte > cache[1]) {
7656 /* New position is after the existing pair of pairs. */
7657 const float keep_earlier
7658 = THREEWAY_SQUARE(0, cache[3], byte, blen);
7659 const float keep_later
7660 = THREEWAY_SQUARE(0, cache[1], byte, blen);
7662 if (keep_later < keep_earlier) {
7663 cache[2] = cache[0];
7664 cache[3] = cache[1];
7670 const float keep_later = THREEWAY_SQUARE(0, byte, cache[1], blen);
7671 float b, c, keep_earlier;
7672 if (byte > cache[3]) {
7673 /* New position is between the existing pair of pairs. */
7674 b = (float)cache[3];
7677 /* New position is before the existing pair of pairs. */
7679 c = (float)cache[3];
7681 keep_earlier = THREEWAY_SQUARE(0, b, c, blen);
7682 if (byte > cache[3]) {
7683 if (keep_later < keep_earlier) {
7693 if (! (keep_later < keep_earlier)) {
7694 cache[0] = cache[2];
7695 cache[1] = cache[3];
7702 ASSERT_UTF8_CACHE(cache);
7705 /* We already know all of the way, now we may be able to walk back. The same
7706 assumption is made as in S_sv_pos_u2b_midway(), namely that walking
7707 backward is half the speed of walking forward. */
7709 S_sv_pos_b2u_midway(pTHX_ const U8 *const s, const U8 *const target,
7710 const U8 *end, STRLEN endu)
7712 const STRLEN forw = target - s;
7713 STRLEN backw = end - target;
7715 PERL_ARGS_ASSERT_SV_POS_B2U_MIDWAY;
7717 if (forw < 2 * backw) {
7718 return utf8_length(s, target);
7721 while (end > target) {
7723 while (UTF8_IS_CONTINUATION(*end)) {
7732 =for apidoc sv_pos_b2u_flags
7734 Converts C<offset> from a count of bytes from the start of the string, to
7735 a count of the equivalent number of UTF-8 chars. Handles type coercion.
7736 C<flags> is passed to C<SvPV_flags>, and usually should be
7737 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7743 * sv_pos_b2u_flags() uses, like sv_pos_u2b_flags(), the mg_ptr of the
7744 * potential PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8
7749 Perl_sv_pos_b2u_flags(pTHX_ SV *const sv, STRLEN const offset, U32 flags)
7752 STRLEN len = 0; /* Actually always set, but let's keep gcc happy. */
7758 PERL_ARGS_ASSERT_SV_POS_B2U_FLAGS;
7760 s = (const U8*)SvPV_flags(sv, blen, flags);
7763 Perl_croak(aTHX_ "panic: sv_pos_b2u: bad byte offset, blen=%" UVuf
7764 ", byte=%" UVuf, (UV)blen, (UV)offset);
7770 && SvTYPE(sv) >= SVt_PVMG
7771 && (mg = mg_find(sv, PERL_MAGIC_utf8)))
7774 STRLEN * const cache = (STRLEN *) mg->mg_ptr;
7775 if (cache[1] == offset) {
7776 /* An exact match. */
7779 if (cache[3] == offset) {
7780 /* An exact match. */
7784 if (cache[1] < offset) {
7785 /* We already know part of the way. */
7786 if (mg->mg_len != -1) {
7787 /* Actually, we know the end too. */
7789 + S_sv_pos_b2u_midway(aTHX_ s + cache[1], send,
7790 s + blen, mg->mg_len - cache[0]);
7792 len = cache[0] + utf8_length(s + cache[1], send);
7795 else if (cache[3] < offset) {
7796 /* We're between the two cached pairs, so we do the calculation
7797 offset by the byte/utf-8 positions for the earlier pair,
7798 then add the utf-8 characters from the string start to
7800 len = S_sv_pos_b2u_midway(aTHX_ s + cache[3], send,
7801 s + cache[1], cache[0] - cache[2])
7805 else { /* cache[3] > offset */
7806 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + cache[3],
7810 ASSERT_UTF8_CACHE(cache);
7812 } else if (mg->mg_len != -1) {
7813 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + blen, mg->mg_len);
7817 if (!found || PL_utf8cache < 0) {
7818 const STRLEN real_len = utf8_length(s, send);
7820 if (found && PL_utf8cache < 0)
7821 assert_uft8_cache_coherent("sv_pos_b2u", len, real_len, sv);
7827 utf8_mg_len_cache_update(sv, &mg, len);
7829 utf8_mg_pos_cache_update(sv, &mg, offset, len, blen);
7836 =for apidoc sv_pos_b2u
7838 Converts the value pointed to by C<offsetp> from a count of bytes from the
7839 start of the string, to a count of the equivalent number of UTF-8 chars.
7840 Handles magic and type coercion.
7842 Use C<sv_pos_b2u_flags> in preference, which correctly handles strings
7849 * sv_pos_b2u() uses, like sv_pos_u2b(), the mg_ptr of the potential
7850 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7855 Perl_sv_pos_b2u(pTHX_ SV *const sv, I32 *const offsetp)
7857 PERL_ARGS_ASSERT_SV_POS_B2U;
7862 *offsetp = (I32)sv_pos_b2u_flags(sv, (STRLEN)*offsetp,
7863 SV_GMAGIC|SV_CONST_RETURN);
7867 S_assert_uft8_cache_coherent(pTHX_ const char *const func, STRLEN from_cache,
7868 STRLEN real, SV *const sv)
7870 PERL_ARGS_ASSERT_ASSERT_UFT8_CACHE_COHERENT;
7872 /* As this is debugging only code, save space by keeping this test here,
7873 rather than inlining it in all the callers. */
7874 if (from_cache == real)
7877 /* Need to turn the assertions off otherwise we may recurse infinitely
7878 while printing error messages. */
7879 SAVEI8(PL_utf8cache);
7881 Perl_croak(aTHX_ "panic: %s cache %" UVuf " real %" UVuf " for %" SVf,
7882 func, (UV) from_cache, (UV) real, SVfARG(sv));
7888 Returns a boolean indicating whether the strings in the two SVs are
7889 identical. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7890 coerce its args to strings if necessary.
7892 =for apidoc sv_eq_flags
7894 Returns a boolean indicating whether the strings in the two SVs are
7895 identical. Is UTF-8 and S<C<'use bytes'>> aware and coerces its args to strings
7896 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get-magic, too.
7902 Perl_sv_eq_flags(pTHX_ SV *sv1, SV *sv2, const U32 flags)
7909 SV* svrecode = NULL;
7916 /* if pv1 and pv2 are the same, second SvPV_const call may
7917 * invalidate pv1 (if we are handling magic), so we may need to
7919 if (sv1 == sv2 && flags & SV_GMAGIC
7920 && (SvTHINKFIRST(sv1) || SvGMAGICAL(sv1))) {
7921 pv1 = SvPV_const(sv1, cur1);
7922 sv1 = newSVpvn_flags(pv1, cur1, SVs_TEMP | SvUTF8(sv2));
7924 pv1 = SvPV_flags_const(sv1, cur1, flags);
7932 pv2 = SvPV_flags_const(sv2, cur2, flags);
7934 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7935 /* Differing utf8ness. */
7937 /* sv1 is the UTF-8 one */
7938 return bytes_cmp_utf8((const U8*)pv2, cur2,
7939 (const U8*)pv1, cur1) == 0;
7942 /* sv2 is the UTF-8 one */
7943 return bytes_cmp_utf8((const U8*)pv1, cur1,
7944 (const U8*)pv2, cur2) == 0;
7949 eq = (pv1 == pv2) || memEQ(pv1, pv2, cur1);
7951 SvREFCNT_dec(svrecode);
7959 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7960 string in C<sv1> is less than, equal to, or greater than the string in
7961 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7962 coerce its args to strings if necessary. See also C<L</sv_cmp_locale>>.
7964 =for apidoc sv_cmp_flags
7966 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7967 string in C<sv1> is less than, equal to, or greater than the string in
7968 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware and will coerce its args to strings
7969 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get magic. See
7970 also C<L</sv_cmp_locale_flags>>.
7976 Perl_sv_cmp(pTHX_ SV *const sv1, SV *const sv2)
7978 return sv_cmp_flags(sv1, sv2, SV_GMAGIC);
7982 Perl_sv_cmp_flags(pTHX_ SV *const sv1, SV *const sv2,
7986 const char *pv1, *pv2;
7988 SV *svrecode = NULL;
7995 pv1 = SvPV_flags_const(sv1, cur1, flags);
8002 pv2 = SvPV_flags_const(sv2, cur2, flags);
8004 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
8005 /* Differing utf8ness. */
8007 const int retval = -bytes_cmp_utf8((const U8*)pv2, cur2,
8008 (const U8*)pv1, cur1);
8009 return retval ? retval < 0 ? -1 : +1 : 0;
8012 const int retval = bytes_cmp_utf8((const U8*)pv1, cur1,
8013 (const U8*)pv2, cur2);
8014 return retval ? retval < 0 ? -1 : +1 : 0;
8018 /* Here, if both are non-NULL, then they have the same UTF8ness. */
8021 cmp = cur2 ? -1 : 0;
8025 STRLEN shortest_len = cur1 < cur2 ? cur1 : cur2;
8028 if (! DO_UTF8(sv1)) {
8030 const I32 retval = memcmp((const void*)pv1,
8034 cmp = retval < 0 ? -1 : 1;
8035 } else if (cur1 == cur2) {
8038 cmp = cur1 < cur2 ? -1 : 1;
8042 else { /* Both are to be treated as UTF-EBCDIC */
8044 /* EBCDIC UTF-8 is complicated by the fact that it is based on I8
8045 * which remaps code points 0-255. We therefore generally have to
8046 * unmap back to the original values to get an accurate comparison.
8047 * But we don't have to do that for UTF-8 invariants, as by
8048 * definition, they aren't remapped, nor do we have to do it for
8049 * above-latin1 code points, as they also aren't remapped. (This
8050 * code also works on ASCII platforms, but the memcmp() above is
8053 const char *e = pv1 + shortest_len;
8055 /* Find the first bytes that differ between the two strings */
8056 while (pv1 < e && *pv1 == *pv2) {
8062 if (pv1 == e) { /* Are the same all the way to the end */
8066 cmp = cur1 < cur2 ? -1 : 1;
8069 else /* Here *pv1 and *pv2 are not equal, but all bytes earlier
8070 * in the strings were. The current bytes may or may not be
8071 * at the beginning of a character. But neither or both are
8072 * (or else earlier bytes would have been different). And
8073 * if we are in the middle of a character, the two
8074 * characters are comprised of the same number of bytes
8075 * (because in this case the start bytes are the same, and
8076 * the start bytes encode the character's length). */
8077 if (UTF8_IS_INVARIANT(*pv1))
8079 /* If both are invariants; can just compare directly */
8080 if (UTF8_IS_INVARIANT(*pv2)) {
8081 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8083 else /* Since *pv1 is invariant, it is the whole character,
8084 which means it is at the beginning of a character.
8085 That means pv2 is also at the beginning of a
8086 character (see earlier comment). Since it isn't
8087 invariant, it must be a start byte. If it starts a
8088 character whose code point is above 255, that
8089 character is greater than any single-byte char, which
8091 if (UTF8_IS_ABOVE_LATIN1_START(*pv2))
8096 /* Here, pv2 points to a character composed of 2 bytes
8097 * whose code point is < 256. Get its code point and
8098 * compare with *pv1 */
8099 cmp = ((U8) *pv1 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
8104 else /* The code point starting at pv1 isn't a single byte */
8105 if (UTF8_IS_INVARIANT(*pv2))
8107 /* But here, the code point starting at *pv2 is a single byte,
8108 * and so *pv1 must begin a character, hence is a start byte.
8109 * If that character is above 255, it is larger than any
8110 * single-byte char, which *pv2 is */
8111 if (UTF8_IS_ABOVE_LATIN1_START(*pv1)) {
8115 /* Here, pv1 points to a character composed of 2 bytes
8116 * whose code point is < 256. Get its code point and
8117 * compare with the single byte character *pv2 */
8118 cmp = (EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1)) < (U8) *pv2)
8123 else /* Here, we've ruled out either *pv1 and *pv2 being
8124 invariant. That means both are part of variants, but not
8125 necessarily at the start of a character */
8126 if ( UTF8_IS_ABOVE_LATIN1_START(*pv1)
8127 || UTF8_IS_ABOVE_LATIN1_START(*pv2))
8129 /* Here, at least one is the start of a character, which means
8130 * the other is also a start byte. And the code point of at
8131 * least one of the characters is above 255. It is a
8132 * characteristic of UTF-EBCDIC that all start bytes for
8133 * above-latin1 code points are well behaved as far as code
8134 * point comparisons go, and all are larger than all other
8135 * start bytes, so the comparison with those is also well
8137 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8140 /* Here both *pv1 and *pv2 are part of variant characters.
8141 * They could be both continuations, or both start characters.
8142 * (One or both could even be an illegal start character (for
8143 * an overlong) which for the purposes of sorting we treat as
8145 if (UTF8_IS_CONTINUATION(*pv1)) {
8147 /* If they are continuations for code points above 255,
8148 * then comparing the current byte is sufficient, as there
8149 * is no remapping of these and so the comparison is
8150 * well-behaved. We determine if they are such
8151 * continuations by looking at the preceding byte. It
8152 * could be a start byte, from which we can tell if it is
8153 * for an above 255 code point. Or it could be a
8154 * continuation, which means the character occupies at
8155 * least 3 bytes, so must be above 255. */
8156 if ( UTF8_IS_CONTINUATION(*(pv2 - 1))
8157 || UTF8_IS_ABOVE_LATIN1_START(*(pv2 -1)))
8159 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8163 /* Here, the continuations are for code points below 256;
8164 * back up one to get to the start byte */
8169 /* We need to get the actual native code point of each of these
8170 * variants in order to compare them */
8171 cmp = ( EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1))
8172 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
8181 SvREFCNT_dec(svrecode);
8187 =for apidoc sv_cmp_locale
8189 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
8190 S<C<'use bytes'>> aware, handles get magic, and will coerce its args to strings
8191 if necessary. See also C<L</sv_cmp>>.
8193 =for apidoc sv_cmp_locale_flags
8195 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
8196 S<C<'use bytes'>> aware and will coerce its args to strings if necessary. If
8197 the flags contain C<SV_GMAGIC>, it handles get magic. See also
8198 C<L</sv_cmp_flags>>.
8204 Perl_sv_cmp_locale(pTHX_ SV *const sv1, SV *const sv2)
8206 return sv_cmp_locale_flags(sv1, sv2, SV_GMAGIC);
8210 Perl_sv_cmp_locale_flags(pTHX_ SV *const sv1, SV *const sv2,
8213 #ifdef USE_LOCALE_COLLATE
8219 if (PL_collation_standard)
8224 /* Revert to using raw compare if both operands exist, but either one
8225 * doesn't transform properly for collation */
8227 pv1 = sv_collxfrm_flags(sv1, &len1, flags);
8231 pv2 = sv_collxfrm_flags(sv2, &len2, flags);
8237 pv1 = sv1 ? sv_collxfrm_flags(sv1, &len1, flags) : (char *) NULL;
8238 pv2 = sv2 ? sv_collxfrm_flags(sv2, &len2, flags) : (char *) NULL;
8241 if (!pv1 || !len1) {
8252 retval = memcmp((void*)pv1, (void*)pv2, len1 < len2 ? len1 : len2);
8255 return retval < 0 ? -1 : 1;
8258 * When the result of collation is equality, that doesn't mean
8259 * that there are no differences -- some locales exclude some
8260 * characters from consideration. So to avoid false equalities,
8261 * we use the raw string as a tiebreaker.
8268 PERL_UNUSED_ARG(flags);
8269 #endif /* USE_LOCALE_COLLATE */
8271 return sv_cmp(sv1, sv2);
8275 #ifdef USE_LOCALE_COLLATE
8278 =for apidoc sv_collxfrm
8280 This calls C<sv_collxfrm_flags> with the SV_GMAGIC flag. See
8281 C<L</sv_collxfrm_flags>>.
8283 =for apidoc sv_collxfrm_flags
8285 Add Collate Transform magic to an SV if it doesn't already have it. If the
8286 flags contain C<SV_GMAGIC>, it handles get-magic.
8288 Any scalar variable may carry C<PERL_MAGIC_collxfrm> magic that contains the
8289 scalar data of the variable, but transformed to such a format that a normal
8290 memory comparison can be used to compare the data according to the locale
8297 Perl_sv_collxfrm_flags(pTHX_ SV *const sv, STRLEN *const nxp, const I32 flags)
8301 PERL_ARGS_ASSERT_SV_COLLXFRM_FLAGS;
8303 mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_collxfrm) : (MAGIC *) NULL;
8305 /* If we don't have collation magic on 'sv', or the locale has changed
8306 * since the last time we calculated it, get it and save it now */
8307 if (!mg || !mg->mg_ptr || *(U32*)mg->mg_ptr != PL_collation_ix) {
8312 /* Free the old space */
8314 Safefree(mg->mg_ptr);
8316 s = SvPV_flags_const(sv, len, flags);
8317 if ((xf = _mem_collxfrm(s, len, &xlen, cBOOL(SvUTF8(sv))))) {
8319 mg = sv_magicext(sv, 0, PERL_MAGIC_collxfrm, &PL_vtbl_collxfrm,
8334 if (mg && mg->mg_ptr) {
8336 return mg->mg_ptr + sizeof(PL_collation_ix);
8344 #endif /* USE_LOCALE_COLLATE */
8347 S_sv_gets_append_to_utf8(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8349 SV * const tsv = newSV(0);
8352 sv_gets(tsv, fp, 0);
8353 sv_utf8_upgrade_nomg(tsv);
8354 SvCUR_set(sv,append);
8357 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8361 S_sv_gets_read_record(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8364 const STRLEN recsize = SvUV(SvRV(PL_rs)); /* RsRECORD() guarantees > 0. */
8365 /* Grab the size of the record we're getting */
8366 char *buffer = SvGROW(sv, (STRLEN)(recsize + append + 1)) + append;
8373 /* With a true, record-oriented file on VMS, we need to use read directly
8374 * to ensure that we respect RMS record boundaries. The user is responsible
8375 * for providing a PL_rs value that corresponds to the FAB$W_MRS (maximum
8376 * record size) field. N.B. This is likely to produce invalid results on
8377 * varying-width character data when a record ends mid-character.
8379 fd = PerlIO_fileno(fp);
8381 && PerlLIO_fstat(fd, &st) == 0
8382 && (st.st_fab_rfm == FAB$C_VAR
8383 || st.st_fab_rfm == FAB$C_VFC
8384 || st.st_fab_rfm == FAB$C_FIX)) {
8386 bytesread = PerlLIO_read(fd, buffer, recsize);
8388 else /* in-memory file from PerlIO::Scalar
8389 * or not a record-oriented file
8393 bytesread = PerlIO_read(fp, buffer, recsize);
8395 /* At this point, the logic in sv_get() means that sv will
8396 be treated as utf-8 if the handle is utf8.
8398 if (PerlIO_isutf8(fp) && bytesread > 0) {
8399 char *bend = buffer + bytesread;
8400 char *bufp = buffer;
8401 size_t charcount = 0;
8402 bool charstart = TRUE;
8405 while (charcount < recsize) {
8406 /* count accumulated characters */
8407 while (bufp < bend) {
8409 skip = UTF8SKIP(bufp);
8411 if (bufp + skip > bend) {
8412 /* partial at the end */
8423 if (charcount < recsize) {
8425 STRLEN bufp_offset = bufp - buffer;
8426 SSize_t morebytesread;
8428 /* originally I read enough to fill any incomplete
8429 character and the first byte of the next
8430 character if needed, but if there's many
8431 multi-byte encoded characters we're going to be
8432 making a read call for every character beyond
8433 the original read size.
8435 So instead, read the rest of the character if
8436 any, and enough bytes to match at least the
8437 start bytes for each character we're going to
8441 readsize = recsize - charcount;
8443 readsize = skip - (bend - bufp) + recsize - charcount - 1;
8444 buffer = SvGROW(sv, append + bytesread + readsize + 1) + append;
8445 bend = buffer + bytesread;
8446 morebytesread = PerlIO_read(fp, bend, readsize);
8447 if (morebytesread <= 0) {
8448 /* we're done, if we still have incomplete
8449 characters the check code in sv_gets() will
8452 I'd originally considered doing
8453 PerlIO_ungetc() on all but the lead
8454 character of the incomplete character, but
8455 read() doesn't do that, so I don't.
8460 /* prepare to scan some more */
8461 bytesread += morebytesread;
8462 bend = buffer + bytesread;
8463 bufp = buffer + bufp_offset;
8471 SvCUR_set(sv, bytesread + append);
8472 buffer[bytesread] = '\0';
8473 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8479 Get a line from the filehandle and store it into the SV, optionally
8480 appending to the currently-stored string. If C<append> is not 0, the
8481 line is appended to the SV instead of overwriting it. C<append> should
8482 be set to the byte offset that the appended string should start at
8483 in the SV (typically, C<SvCUR(sv)> is a suitable choice).
8489 Perl_sv_gets(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8499 PERL_ARGS_ASSERT_SV_GETS;
8501 if (SvTHINKFIRST(sv))
8502 sv_force_normal_flags(sv, append ? 0 : SV_COW_DROP_PV);
8503 /* XXX. If you make this PVIV, then copy on write can copy scalars read
8505 However, perlbench says it's slower, because the existing swipe code
8506 is faster than copy on write.
8507 Swings and roundabouts. */
8508 SvUPGRADE(sv, SVt_PV);
8511 /* line is going to be appended to the existing buffer in the sv */
8512 if (PerlIO_isutf8(fp)) {
8514 sv_utf8_upgrade_nomg(sv);
8515 sv_pos_u2b(sv,&append,0);
8517 } else if (SvUTF8(sv)) {
8518 return S_sv_gets_append_to_utf8(aTHX_ sv, fp, append);
8524 /* not appending - "clear" the string by setting SvCUR to 0,
8525 * the pv is still avaiable. */
8528 if (PerlIO_isutf8(fp))
8531 if (IN_PERL_COMPILETIME) {
8532 /* we always read code in line mode */
8536 else if (RsSNARF(PL_rs)) {
8537 /* If it is a regular disk file use size from stat() as estimate
8538 of amount we are going to read -- may result in mallocing
8539 more memory than we really need if the layers below reduce
8540 the size we read (e.g. CRLF or a gzip layer).
8543 int fd = PerlIO_fileno(fp);
8544 if (fd >= 0 && (PerlLIO_fstat(fd, &st) == 0) && S_ISREG(st.st_mode)) {
8545 const Off_t offset = PerlIO_tell(fp);
8546 if (offset != (Off_t) -1 && st.st_size + append > offset) {
8547 #ifdef PERL_COPY_ON_WRITE
8548 /* Add an extra byte for the sake of copy-on-write's
8549 * buffer reference count. */
8550 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 2));
8552 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 1));
8559 else if (RsRECORD(PL_rs)) {
8560 return S_sv_gets_read_record(aTHX_ sv, fp, append);
8562 else if (RsPARA(PL_rs)) {
8568 /* Get $/ i.e. PL_rs into same encoding as stream wants */
8569 if (PerlIO_isutf8(fp)) {
8570 rsptr = SvPVutf8(PL_rs, rslen);
8573 if (SvUTF8(PL_rs)) {
8574 if (!sv_utf8_downgrade(PL_rs, TRUE)) {
8575 Perl_croak(aTHX_ "Wide character in $/");
8578 /* extract the raw pointer to the record separator */
8579 rsptr = SvPV_const(PL_rs, rslen);
8583 /* rslast is the last character in the record separator
8584 * note we don't use rslast except when rslen is true, so the
8585 * null assign is a placeholder. */
8586 rslast = rslen ? rsptr[rslen - 1] : '\0';
8588 if (rspara) { /* have to do this both before and after */
8589 do { /* to make sure file boundaries work right */
8592 i = PerlIO_getc(fp);
8596 PerlIO_ungetc(fp,i);
8602 /* See if we know enough about I/O mechanism to cheat it ! */
8604 /* This used to be #ifdef test - it is made run-time test for ease
8605 of abstracting out stdio interface. One call should be cheap
8606 enough here - and may even be a macro allowing compile
8610 if (PerlIO_fast_gets(fp)) {
8612 * We can do buffer based IO operations on this filehandle.
8614 * This means we can bypass a lot of subcalls and process
8615 * the buffer directly, it also means we know the upper bound
8616 * on the amount of data we might read of the current buffer
8617 * into our sv. Knowing this allows us to preallocate the pv
8618 * to be able to hold that maximum, which allows us to simplify
8619 * a lot of logic. */
8622 * We're going to steal some values from the stdio struct
8623 * and put EVERYTHING in the innermost loop into registers.
8625 STDCHAR *ptr; /* pointer into fp's read-ahead buffer */
8626 STRLEN bpx; /* length of the data in the target sv
8627 used to fix pointers after a SvGROW */
8628 I32 shortbuffered; /* If the pv buffer is shorter than the amount
8629 of data left in the read-ahead buffer.
8630 If 0 then the pv buffer can hold the full
8631 amount left, otherwise this is the amount it
8634 /* Here is some breathtakingly efficient cheating */
8636 /* When you read the following logic resist the urge to think
8637 * of record separators that are 1 byte long. They are an
8638 * uninteresting special (simple) case.
8640 * Instead think of record separators which are at least 2 bytes
8641 * long, and keep in mind that we need to deal with such
8642 * separators when they cross a read-ahead buffer boundary.
8644 * Also consider that we need to gracefully deal with separators
8645 * that may be longer than a single read ahead buffer.
8647 * Lastly do not forget we want to copy the delimiter as well. We
8648 * are copying all data in the file _up_to_and_including_ the separator
8651 * Now that you have all that in mind here is what is happening below:
8653 * 1. When we first enter the loop we do some memory book keeping to see
8654 * how much free space there is in the target SV. (This sub assumes that
8655 * it is operating on the same SV most of the time via $_ and that it is
8656 * going to be able to reuse the same pv buffer each call.) If there is
8657 * "enough" room then we set "shortbuffered" to how much space there is
8658 * and start reading forward.
8660 * 2. When we scan forward we copy from the read-ahead buffer to the target
8661 * SV's pv buffer. While we go we watch for the end of the read-ahead buffer,
8662 * and the end of the of pv, as well as for the "rslast", which is the last
8663 * char of the separator.
8665 * 3. When scanning forward if we see rslast then we jump backwards in *pv*
8666 * (which has a "complete" record up to the point we saw rslast) and check
8667 * it to see if it matches the separator. If it does we are done. If it doesn't
8668 * we continue on with the scan/copy.
8670 * 4. If we run out of read-ahead buffer (cnt goes to 0) then we have to get
8671 * the IO system to read the next buffer. We do this by doing a getc(), which
8672 * returns a single char read (or EOF), and prefills the buffer, and also
8673 * allows us to find out how full the buffer is. We use this information to
8674 * SvGROW() the sv to the size remaining in the buffer, after which we copy
8675 * the returned single char into the target sv, and then go back into scan
8678 * 5. If we run out of write-buffer then we SvGROW() it by the size of the
8679 * remaining space in the read-buffer.
8681 * Note that this code despite its twisty-turny nature is pretty darn slick.
8682 * It manages single byte separators, multi-byte cross boundary separators,
8683 * and cross-read-buffer separators cleanly and efficiently at the cost
8684 * of potentially greatly overallocating the target SV.
8690 /* get the number of bytes remaining in the read-ahead buffer
8691 * on first call on a given fp this will return 0.*/
8692 cnt = PerlIO_get_cnt(fp);
8694 /* make sure we have the room */
8695 if ((I32)(SvLEN(sv) - append) <= cnt + 1) {
8696 /* Not room for all of it
8697 if we are looking for a separator and room for some
8699 if (rslen && cnt > 80 && (I32)SvLEN(sv) > append) {
8700 /* just process what we have room for */
8701 shortbuffered = cnt - SvLEN(sv) + append + 1;
8702 cnt -= shortbuffered;
8705 /* ensure that the target sv has enough room to hold
8706 * the rest of the read-ahead buffer */
8708 /* remember that cnt can be negative */
8709 SvGROW(sv, (STRLEN)(append + (cnt <= 0 ? 2 : (cnt + 1))));
8713 /* we have enough room to hold the full buffer, lets scream */
8717 /* extract the pointer to sv's string buffer, offset by append as necessary */
8718 bp = (STDCHAR*)SvPVX_const(sv) + append; /* move these two too to registers */
8719 /* extract the point to the read-ahead buffer */
8720 ptr = (STDCHAR*)PerlIO_get_ptr(fp);
8722 /* some trace debug output */
8723 DEBUG_P(PerlIO_printf(Perl_debug_log,
8724 "Screamer: entering, ptr=%" UVuf ", cnt=%ld\n",PTR2UV(ptr),(long)cnt));
8725 DEBUG_P(PerlIO_printf(Perl_debug_log,
8726 "Screamer: entering: PerlIO * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%"
8728 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8729 PTR2UV(PerlIO_has_base(fp) ? PerlIO_get_base(fp) : 0)));
8733 /* if there is stuff left in the read-ahead buffer */
8735 /* if there is a separator */
8737 /* find next rslast */
8740 /* shortcut common case of blank line */
8742 if ((*bp++ = *ptr++) == rslast)
8743 goto thats_all_folks;
8745 p = (STDCHAR *)memchr(ptr, rslast, cnt);
8747 SSize_t got = p - ptr + 1;
8748 Copy(ptr, bp, got, STDCHAR);
8752 goto thats_all_folks;
8754 Copy(ptr, bp, cnt, STDCHAR);
8760 /* no separator, slurp the full buffer */
8761 Copy(ptr, bp, cnt, char); /* this | eat */
8762 bp += cnt; /* screams | dust */
8763 ptr += cnt; /* louder | sed :-) */
8765 assert (!shortbuffered);
8766 goto cannot_be_shortbuffered;
8770 if (shortbuffered) { /* oh well, must extend */
8771 /* we didnt have enough room to fit the line into the target buffer
8772 * so we must extend the target buffer and keep going */
8773 cnt = shortbuffered;
8775 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8777 /* extned the target sv's buffer so it can hold the full read-ahead buffer */
8778 SvGROW(sv, SvLEN(sv) + append + cnt + 2);
8779 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8783 cannot_be_shortbuffered:
8784 /* we need to refill the read-ahead buffer if possible */
8786 DEBUG_P(PerlIO_printf(Perl_debug_log,
8787 "Screamer: going to getc, ptr=%" UVuf ", cnt=%" IVdf "\n",
8788 PTR2UV(ptr),(IV)cnt));
8789 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* deregisterize cnt and ptr */
8791 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8792 "Screamer: pre: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf "\n",
8793 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8794 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8797 call PerlIO_getc() to let it prefill the lookahead buffer
8799 This used to call 'filbuf' in stdio form, but as that behaves like
8800 getc when cnt <= 0 we use PerlIO_getc here to avoid introducing
8801 another abstraction.
8803 Note we have to deal with the char in 'i' if we are not at EOF
8805 i = PerlIO_getc(fp); /* get more characters */
8807 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8808 "Screamer: post: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf "\n",
8809 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8810 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8812 /* find out how much is left in the read-ahead buffer, and rextract its pointer */
8813 cnt = PerlIO_get_cnt(fp);
8814 ptr = (STDCHAR*)PerlIO_get_ptr(fp); /* reregisterize cnt and ptr */
8815 DEBUG_P(PerlIO_printf(Perl_debug_log,
8816 "Screamer: after getc, ptr=%" UVuf ", cnt=%" IVdf "\n",
8817 PTR2UV(ptr),(IV)cnt));
8819 if (i == EOF) /* all done for ever? */
8820 goto thats_really_all_folks;
8822 /* make sure we have enough space in the target sv */
8823 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8825 SvGROW(sv, bpx + cnt + 2);
8826 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8828 /* copy of the char we got from getc() */
8829 *bp++ = (STDCHAR)i; /* store character from PerlIO_getc */
8831 /* make sure we deal with the i being the last character of a separator */
8832 if (rslen && (STDCHAR)i == rslast) /* all done for now? */
8833 goto thats_all_folks;
8837 /* check if we have actually found the separator - only really applies
8839 if ((rslen > 1 && (STRLEN)(bp - (STDCHAR*)SvPVX_const(sv)) < rslen) ||
8840 memNE((char*)bp - rslen, rsptr, rslen))
8841 goto screamer; /* go back to the fray */
8842 thats_really_all_folks:
8844 cnt += shortbuffered;
8845 DEBUG_P(PerlIO_printf(Perl_debug_log,
8846 "Screamer: quitting, ptr=%" UVuf ", cnt=%" IVdf "\n",PTR2UV(ptr),(IV)cnt));
8847 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* put these back or we're in trouble */
8848 DEBUG_P(PerlIO_printf(Perl_debug_log,
8849 "Screamer: end: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf
8851 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8852 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8854 SvCUR_set(sv, bp - (STDCHAR*)SvPVX_const(sv)); /* set length */
8855 DEBUG_P(PerlIO_printf(Perl_debug_log,
8856 "Screamer: done, len=%ld, string=|%.*s|\n",
8857 (long)SvCUR(sv),(int)SvCUR(sv),SvPVX_const(sv)));
8861 /*The big, slow, and stupid way. */
8862 #ifdef USE_HEAP_INSTEAD_OF_STACK /* Even slower way. */
8863 STDCHAR *buf = NULL;
8864 Newx(buf, 8192, STDCHAR);
8872 const STDCHAR * const bpe = buf + sizeof(buf);
8874 while ((i = PerlIO_getc(fp)) != EOF && (*bp++ = (STDCHAR)i) != rslast && bp < bpe)
8875 ; /* keep reading */
8879 cnt = PerlIO_read(fp,(char*)buf, sizeof(buf));
8880 /* Accommodate broken VAXC compiler, which applies U8 cast to
8881 * both args of ?: operator, causing EOF to change into 255
8884 i = (U8)buf[cnt - 1];
8890 cnt = 0; /* we do need to re-set the sv even when cnt <= 0 */
8892 sv_catpvn_nomg(sv, (char *) buf, cnt);
8894 sv_setpvn(sv, (char *) buf, cnt); /* "nomg" is implied */
8896 if (i != EOF && /* joy */
8898 SvCUR(sv) < rslen ||
8899 memNE(SvPVX_const(sv) + SvCUR(sv) - rslen, rsptr, rslen)))
8903 * If we're reading from a TTY and we get a short read,
8904 * indicating that the user hit his EOF character, we need
8905 * to notice it now, because if we try to read from the TTY
8906 * again, the EOF condition will disappear.
8908 * The comparison of cnt to sizeof(buf) is an optimization
8909 * that prevents unnecessary calls to feof().
8913 if (!(cnt < (I32)sizeof(buf) && PerlIO_eof(fp)))
8917 #ifdef USE_HEAP_INSTEAD_OF_STACK
8922 if (rspara) { /* have to do this both before and after */
8923 while (i != EOF) { /* to make sure file boundaries work right */
8924 i = PerlIO_getc(fp);
8926 PerlIO_ungetc(fp,i);
8932 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8938 Auto-increment of the value in the SV, doing string to numeric conversion
8939 if necessary. Handles 'get' magic and operator overloading.
8945 Perl_sv_inc(pTHX_ SV *const sv)
8954 =for apidoc sv_inc_nomg
8956 Auto-increment of the value in the SV, doing string to numeric conversion
8957 if necessary. Handles operator overloading. Skips handling 'get' magic.
8963 Perl_sv_inc_nomg(pTHX_ SV *const sv)
8970 if (SvTHINKFIRST(sv)) {
8971 if (SvREADONLY(sv)) {
8972 Perl_croak_no_modify();
8976 if (SvAMAGIC(sv) && AMG_CALLunary(sv, inc_amg))
8978 i = PTR2IV(SvRV(sv));
8982 else sv_force_normal_flags(sv, 0);
8984 flags = SvFLAGS(sv);
8985 if ((flags & (SVp_NOK|SVp_IOK)) == SVp_NOK) {
8986 /* It's (privately or publicly) a float, but not tested as an
8987 integer, so test it to see. */
8989 flags = SvFLAGS(sv);
8991 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8992 /* It's publicly an integer, or privately an integer-not-float */
8993 #ifdef PERL_PRESERVE_IVUV
8997 if (SvUVX(sv) == UV_MAX)
8998 sv_setnv(sv, UV_MAX_P1);
9000 (void)SvIOK_only_UV(sv);
9001 SvUV_set(sv, SvUVX(sv) + 1);
9003 if (SvIVX(sv) == IV_MAX)
9004 sv_setuv(sv, (UV)IV_MAX + 1);
9006 (void)SvIOK_only(sv);
9007 SvIV_set(sv, SvIVX(sv) + 1);
9012 if (flags & SVp_NOK) {
9013 const NV was = SvNVX(sv);
9014 if (LIKELY(!Perl_isinfnan(was)) &&
9015 NV_OVERFLOWS_INTEGERS_AT != 0.0 &&
9016 was >= NV_OVERFLOWS_INTEGERS_AT) {
9017 /* diag_listed_as: Lost precision when %s %f by 1 */
9018 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
9019 "Lost precision when incrementing %" NVff " by 1",
9022 (void)SvNOK_only(sv);
9023 SvNV_set(sv, was + 1.0);
9027 /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */
9028 if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv)))
9029 Perl_croak_no_modify();
9031 if (!(flags & SVp_POK) || !*SvPVX_const(sv)) {
9032 if ((flags & SVTYPEMASK) < SVt_PVIV)
9033 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV ? SVt_PVIV : SVt_IV));
9034 (void)SvIOK_only(sv);
9039 while (isALPHA(*d)) d++;
9040 while (isDIGIT(*d)) d++;
9041 if (d < SvEND(sv)) {
9042 const int numtype = grok_number_flags(SvPVX_const(sv), SvCUR(sv), NULL, PERL_SCAN_TRAILING);
9043 #ifdef PERL_PRESERVE_IVUV
9044 /* Got to punt this as an integer if needs be, but we don't issue
9045 warnings. Probably ought to make the sv_iv_please() that does
9046 the conversion if possible, and silently. */
9047 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
9048 /* Need to try really hard to see if it's an integer.
9049 9.22337203685478e+18 is an integer.
9050 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
9051 so $a="9.22337203685478e+18"; $a+0; $a++
9052 needs to be the same as $a="9.22337203685478e+18"; $a++
9059 /* sv_2iv *should* have made this an NV */
9060 if (flags & SVp_NOK) {
9061 (void)SvNOK_only(sv);
9062 SvNV_set(sv, SvNVX(sv) + 1.0);
9065 /* I don't think we can get here. Maybe I should assert this
9066 And if we do get here I suspect that sv_setnv will croak. NWC
9068 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_inc punt failed to convert '%s' to IOK or NOKp, UV=0x%" UVxf " NV=%" NVgf "\n",
9069 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
9071 #endif /* PERL_PRESERVE_IVUV */
9072 if (!numtype && ckWARN(WARN_NUMERIC))
9073 not_incrementable(sv);
9074 sv_setnv(sv,Atof(SvPVX_const(sv)) + 1.0);
9078 while (d >= SvPVX_const(sv)) {
9086 /* MKS: The original code here died if letters weren't consecutive.
9087 * at least it didn't have to worry about non-C locales. The
9088 * new code assumes that ('z'-'a')==('Z'-'A'), letters are
9089 * arranged in order (although not consecutively) and that only
9090 * [A-Za-z] are accepted by isALPHA in the C locale.
9092 if (isALPHA_FOLD_NE(*d, 'z')) {
9093 do { ++*d; } while (!isALPHA(*d));
9096 *(d--) -= 'z' - 'a';
9101 *(d--) -= 'z' - 'a' + 1;
9105 /* oh,oh, the number grew */
9106 SvGROW(sv, SvCUR(sv) + 2);
9107 SvCUR_set(sv, SvCUR(sv) + 1);
9108 for (d = SvPVX(sv) + SvCUR(sv); d > SvPVX_const(sv); d--)
9119 Auto-decrement of the value in the SV, doing string to numeric conversion
9120 if necessary. Handles 'get' magic and operator overloading.
9126 Perl_sv_dec(pTHX_ SV *const sv)
9135 =for apidoc sv_dec_nomg
9137 Auto-decrement of the value in the SV, doing string to numeric conversion
9138 if necessary. Handles operator overloading. Skips handling 'get' magic.
9144 Perl_sv_dec_nomg(pTHX_ SV *const sv)
9150 if (SvTHINKFIRST(sv)) {
9151 if (SvREADONLY(sv)) {
9152 Perl_croak_no_modify();
9156 if (SvAMAGIC(sv) && AMG_CALLunary(sv, dec_amg))
9158 i = PTR2IV(SvRV(sv));
9162 else sv_force_normal_flags(sv, 0);
9164 /* Unlike sv_inc we don't have to worry about string-never-numbers
9165 and keeping them magic. But we mustn't warn on punting */
9166 flags = SvFLAGS(sv);
9167 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
9168 /* It's publicly an integer, or privately an integer-not-float */
9169 #ifdef PERL_PRESERVE_IVUV
9173 if (SvUVX(sv) == 0) {
9174 (void)SvIOK_only(sv);
9178 (void)SvIOK_only_UV(sv);
9179 SvUV_set(sv, SvUVX(sv) - 1);
9182 if (SvIVX(sv) == IV_MIN) {
9183 sv_setnv(sv, (NV)IV_MIN);
9187 (void)SvIOK_only(sv);
9188 SvIV_set(sv, SvIVX(sv) - 1);
9193 if (flags & SVp_NOK) {
9196 const NV was = SvNVX(sv);
9197 if (LIKELY(!Perl_isinfnan(was)) &&
9198 NV_OVERFLOWS_INTEGERS_AT != 0.0 &&
9199 was <= -NV_OVERFLOWS_INTEGERS_AT) {
9200 /* diag_listed_as: Lost precision when %s %f by 1 */
9201 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
9202 "Lost precision when decrementing %" NVff " by 1",
9205 (void)SvNOK_only(sv);
9206 SvNV_set(sv, was - 1.0);
9211 /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */
9212 if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv)))
9213 Perl_croak_no_modify();
9215 if (!(flags & SVp_POK)) {
9216 if ((flags & SVTYPEMASK) < SVt_PVIV)
9217 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV) ? SVt_PVIV : SVt_IV);
9219 (void)SvIOK_only(sv);
9222 #ifdef PERL_PRESERVE_IVUV
9224 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), NULL);
9225 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
9226 /* Need to try really hard to see if it's an integer.
9227 9.22337203685478e+18 is an integer.
9228 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
9229 so $a="9.22337203685478e+18"; $a+0; $a--
9230 needs to be the same as $a="9.22337203685478e+18"; $a--
9237 /* sv_2iv *should* have made this an NV */
9238 if (flags & SVp_NOK) {
9239 (void)SvNOK_only(sv);
9240 SvNV_set(sv, SvNVX(sv) - 1.0);
9243 /* I don't think we can get here. Maybe I should assert this
9244 And if we do get here I suspect that sv_setnv will croak. NWC
9246 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_dec punt failed to convert '%s' to IOK or NOKp, UV=0x%" UVxf " NV=%" NVgf "\n",
9247 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
9250 #endif /* PERL_PRESERVE_IVUV */
9251 sv_setnv(sv,Atof(SvPVX_const(sv)) - 1.0); /* punt */
9254 /* this define is used to eliminate a chunk of duplicated but shared logic
9255 * it has the suffix __SV_C to signal that it isnt API, and isnt meant to be
9256 * used anywhere but here - yves
9258 #define PUSH_EXTEND_MORTAL__SV_C(AnSv) \
9260 SSize_t ix = ++PL_tmps_ix; \
9261 if (UNLIKELY(ix >= PL_tmps_max)) \
9262 ix = tmps_grow_p(ix); \
9263 PL_tmps_stack[ix] = (AnSv); \
9267 =for apidoc sv_mortalcopy
9269 Creates a new SV which is a copy of the original SV (using C<sv_setsv>).
9270 The new SV is marked as mortal. It will be destroyed "soon", either by an
9271 explicit call to C<FREETMPS>, or by an implicit call at places such as
9272 statement boundaries. See also C<L</sv_newmortal>> and C<L</sv_2mortal>>.
9277 /* Make a string that will exist for the duration of the expression
9278 * evaluation. Actually, it may have to last longer than that, but
9279 * hopefully we won't free it until it has been assigned to a
9280 * permanent location. */
9283 Perl_sv_mortalcopy_flags(pTHX_ SV *const oldstr, U32 flags)
9287 if (flags & SV_GMAGIC)
9288 SvGETMAGIC(oldstr); /* before new_SV, in case it dies */
9290 sv_setsv_flags(sv,oldstr,flags & ~SV_GMAGIC);
9291 PUSH_EXTEND_MORTAL__SV_C(sv);
9297 =for apidoc sv_newmortal
9299 Creates a new null SV which is mortal. The reference count of the SV is
9300 set to 1. It will be destroyed "soon", either by an explicit call to
9301 C<FREETMPS>, or by an implicit call at places such as statement boundaries.
9302 See also C<L</sv_mortalcopy>> and C<L</sv_2mortal>>.
9308 Perl_sv_newmortal(pTHX)
9313 SvFLAGS(sv) = SVs_TEMP;
9314 PUSH_EXTEND_MORTAL__SV_C(sv);
9320 =for apidoc newSVpvn_flags
9322 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9323 characters) into it. The reference count for the
9324 SV is set to 1. Note that if C<len> is zero, Perl will create a zero length
9325 string. You are responsible for ensuring that the source string is at least
9326 C<len> bytes long. If the C<s> argument is NULL the new SV will be undefined.
9327 Currently the only flag bits accepted are C<SVf_UTF8> and C<SVs_TEMP>.
9328 If C<SVs_TEMP> is set, then C<sv_2mortal()> is called on the result before
9329 returning. If C<SVf_UTF8> is set, C<s>
9330 is considered to be in UTF-8 and the
9331 C<SVf_UTF8> flag will be set on the new SV.
9332 C<newSVpvn_utf8()> is a convenience wrapper for this function, defined as
9334 #define newSVpvn_utf8(s, len, u) \
9335 newSVpvn_flags((s), (len), (u) ? SVf_UTF8 : 0)
9341 Perl_newSVpvn_flags(pTHX_ const char *const s, const STRLEN len, const U32 flags)
9345 /* All the flags we don't support must be zero.
9346 And we're new code so I'm going to assert this from the start. */
9347 assert(!(flags & ~(SVf_UTF8|SVs_TEMP)));
9349 sv_setpvn(sv,s,len);
9351 /* This code used to do a sv_2mortal(), however we now unroll the call to
9352 * sv_2mortal() and do what it does ourselves here. Since we have asserted
9353 * that flags can only have the SVf_UTF8 and/or SVs_TEMP flags set above we
9354 * can use it to enable the sv flags directly (bypassing SvTEMP_on), which
9355 * in turn means we dont need to mask out the SVf_UTF8 flag below, which
9356 * means that we eliminate quite a few steps than it looks - Yves
9357 * (explaining patch by gfx) */
9359 SvFLAGS(sv) |= flags;
9361 if(flags & SVs_TEMP){
9362 PUSH_EXTEND_MORTAL__SV_C(sv);
9369 =for apidoc sv_2mortal
9371 Marks an existing SV as mortal. The SV will be destroyed "soon", either
9372 by an explicit call to C<FREETMPS>, or by an implicit call at places such as
9373 statement boundaries. C<SvTEMP()> is turned on which means that the SV's
9374 string buffer can be "stolen" if this SV is copied. See also
9375 C<L</sv_newmortal>> and C<L</sv_mortalcopy>>.
9381 Perl_sv_2mortal(pTHX_ SV *const sv)
9388 PUSH_EXTEND_MORTAL__SV_C(sv);
9396 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9397 characters) into it. The reference count for the
9398 SV is set to 1. If C<len> is zero, Perl will compute the length using
9399 C<strlen()>, (which means if you use this option, that C<s> can't have embedded
9400 C<NUL> characters and has to have a terminating C<NUL> byte).
9402 This function can cause reliability issues if you are likely to pass in
9403 empty strings that are not null terminated, because it will run
9404 strlen on the string and potentially run past valid memory.
9406 Using L</newSVpvn> is a safer alternative for non C<NUL> terminated strings.
9407 For string literals use L</newSVpvs> instead. This function will work fine for
9408 C<NUL> terminated strings, but if you want to avoid the if statement on whether
9409 to call C<strlen> use C<newSVpvn> instead (calling C<strlen> yourself).
9415 Perl_newSVpv(pTHX_ const char *const s, const STRLEN len)
9420 sv_setpvn(sv, s, len || s == NULL ? len : strlen(s));
9425 =for apidoc newSVpvn
9427 Creates a new SV and copies a string into it, which may contain C<NUL> characters
9428 (C<\0>) and other binary data. The reference count for the SV is set to 1.
9429 Note that if C<len> is zero, Perl will create a zero length (Perl) string. You
9430 are responsible for ensuring that the source buffer is at least
9431 C<len> bytes long. If the C<buffer> argument is NULL the new SV will be
9438 Perl_newSVpvn(pTHX_ const char *const buffer, const STRLEN len)
9442 sv_setpvn(sv,buffer,len);
9447 =for apidoc newSVhek
9449 Creates a new SV from the hash key structure. It will generate scalars that
9450 point to the shared string table where possible. Returns a new (undefined)
9451 SV if C<hek> is NULL.
9457 Perl_newSVhek(pTHX_ const HEK *const hek)
9466 if (HEK_LEN(hek) == HEf_SVKEY) {
9467 return newSVsv(*(SV**)HEK_KEY(hek));
9469 const int flags = HEK_FLAGS(hek);
9470 if (flags & HVhek_WASUTF8) {
9472 Andreas would like keys he put in as utf8 to come back as utf8
9474 STRLEN utf8_len = HEK_LEN(hek);
9475 SV * const sv = newSV_type(SVt_PV);
9476 char *as_utf8 = (char *)bytes_to_utf8 ((U8*)HEK_KEY(hek), &utf8_len);
9477 /* bytes_to_utf8() allocates a new string, which we can repurpose: */
9478 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
9481 } else if (flags & HVhek_UNSHARED) {
9482 /* A hash that isn't using shared hash keys has to have
9483 the flag in every key so that we know not to try to call
9484 share_hek_hek on it. */
9486 SV * const sv = newSVpvn (HEK_KEY(hek), HEK_LEN(hek));
9491 /* This will be overwhelminly the most common case. */
9493 /* Inline most of newSVpvn_share(), because share_hek_hek() is far
9494 more efficient than sharepvn(). */
9498 sv_upgrade(sv, SVt_PV);
9499 SvPV_set(sv, (char *)HEK_KEY(share_hek_hek(hek)));
9500 SvCUR_set(sv, HEK_LEN(hek));
9512 =for apidoc newSVpvn_share
9514 Creates a new SV with its C<SvPVX_const> pointing to a shared string in the string
9515 table. If the string does not already exist in the table, it is
9516 created first. Turns on the C<SvIsCOW> flag (or C<READONLY>
9517 and C<FAKE> in 5.16 and earlier). If the C<hash> parameter
9518 is non-zero, that value is used; otherwise the hash is computed.
9519 The string's hash can later be retrieved from the SV
9520 with the C<SvSHARED_HASH()> macro. The idea here is
9521 that as the string table is used for shared hash keys these strings will have
9522 C<SvPVX_const == HeKEY> and hash lookup will avoid string compare.
9528 Perl_newSVpvn_share(pTHX_ const char *src, I32 len, U32 hash)
9532 bool is_utf8 = FALSE;
9533 const char *const orig_src = src;
9536 STRLEN tmplen = -len;
9538 /* See the note in hv.c:hv_fetch() --jhi */
9539 src = (char*)bytes_from_utf8((const U8*)src, &tmplen, &is_utf8);
9543 PERL_HASH(hash, src, len);
9545 /* The logic for this is inlined in S_mro_get_linear_isa_dfs(), so if it
9546 changes here, update it there too. */
9547 sv_upgrade(sv, SVt_PV);
9548 SvPV_set(sv, sharepvn(src, is_utf8?-len:len, hash));
9555 if (src != orig_src)
9561 =for apidoc newSVpv_share
9563 Like C<newSVpvn_share>, but takes a C<NUL>-terminated string instead of a
9570 Perl_newSVpv_share(pTHX_ const char *src, U32 hash)
9572 return newSVpvn_share(src, strlen(src), hash);
9575 #if defined(PERL_IMPLICIT_CONTEXT)
9577 /* pTHX_ magic can't cope with varargs, so this is a no-context
9578 * version of the main function, (which may itself be aliased to us).
9579 * Don't access this version directly.
9583 Perl_newSVpvf_nocontext(const char *const pat, ...)
9589 PERL_ARGS_ASSERT_NEWSVPVF_NOCONTEXT;
9591 va_start(args, pat);
9592 sv = vnewSVpvf(pat, &args);
9599 =for apidoc newSVpvf
9601 Creates a new SV and initializes it with the string formatted like
9608 Perl_newSVpvf(pTHX_ const char *const pat, ...)
9613 PERL_ARGS_ASSERT_NEWSVPVF;
9615 va_start(args, pat);
9616 sv = vnewSVpvf(pat, &args);
9621 /* backend for newSVpvf() and newSVpvf_nocontext() */
9624 Perl_vnewSVpvf(pTHX_ const char *const pat, va_list *const args)
9628 PERL_ARGS_ASSERT_VNEWSVPVF;
9631 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
9638 Creates a new SV and copies a floating point value into it.
9639 The reference count for the SV is set to 1.
9645 Perl_newSVnv(pTHX_ const NV n)
9657 Creates a new SV and copies an integer into it. The reference count for the
9664 Perl_newSViv(pTHX_ const IV i)
9670 /* Inlining ONLY the small relevant subset of sv_setiv here
9671 * for performance. Makes a significant difference. */
9673 /* We're starting from SVt_FIRST, so provided that's
9674 * actual 0, we don't have to unset any SV type flags
9675 * to promote to SVt_IV. */
9676 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9678 SET_SVANY_FOR_BODYLESS_IV(sv);
9679 SvFLAGS(sv) |= SVt_IV;
9691 Creates a new SV and copies an unsigned integer into it.
9692 The reference count for the SV is set to 1.
9698 Perl_newSVuv(pTHX_ const UV u)
9702 /* Inlining ONLY the small relevant subset of sv_setuv here
9703 * for performance. Makes a significant difference. */
9705 /* Using ivs is more efficient than using uvs - see sv_setuv */
9706 if (u <= (UV)IV_MAX) {
9707 return newSViv((IV)u);
9712 /* We're starting from SVt_FIRST, so provided that's
9713 * actual 0, we don't have to unset any SV type flags
9714 * to promote to SVt_IV. */
9715 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9717 SET_SVANY_FOR_BODYLESS_IV(sv);
9718 SvFLAGS(sv) |= SVt_IV;
9720 (void)SvIsUV_on(sv);
9729 =for apidoc newSV_type
9731 Creates a new SV, of the type specified. The reference count for the new SV
9738 Perl_newSV_type(pTHX_ const svtype type)
9743 ASSUME(SvTYPE(sv) == SVt_FIRST);
9744 if(type != SVt_FIRST)
9745 sv_upgrade(sv, type);
9750 =for apidoc newRV_noinc
9752 Creates an RV wrapper for an SV. The reference count for the original
9753 SV is B<not> incremented.
9759 Perl_newRV_noinc(pTHX_ SV *const tmpRef)
9763 PERL_ARGS_ASSERT_NEWRV_NOINC;
9767 /* We're starting from SVt_FIRST, so provided that's
9768 * actual 0, we don't have to unset any SV type flags
9769 * to promote to SVt_IV. */
9770 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9772 SET_SVANY_FOR_BODYLESS_IV(sv);
9773 SvFLAGS(sv) |= SVt_IV;
9778 SvRV_set(sv, tmpRef);
9783 /* newRV_inc is the official function name to use now.
9784 * newRV_inc is in fact #defined to newRV in sv.h
9788 Perl_newRV(pTHX_ SV *const sv)
9790 PERL_ARGS_ASSERT_NEWRV;
9792 return newRV_noinc(SvREFCNT_inc_simple_NN(sv));
9798 Creates a new SV which is an exact duplicate of the original SV.
9805 Perl_newSVsv(pTHX_ SV *const old)
9811 if (SvTYPE(old) == (svtype)SVTYPEMASK) {
9812 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), "semi-panic: attempt to dup freed string");
9815 /* Do this here, otherwise we leak the new SV if this croaks. */
9818 /* SV_NOSTEAL prevents TEMP buffers being, well, stolen, and saves games
9819 with SvTEMP_off and SvTEMP_on round a call to sv_setsv. */
9820 sv_setsv_flags(sv, old, SV_NOSTEAL);
9825 =for apidoc sv_reset
9827 Underlying implementation for the C<reset> Perl function.
9828 Note that the perl-level function is vaguely deprecated.
9834 Perl_sv_reset(pTHX_ const char *s, HV *const stash)
9836 PERL_ARGS_ASSERT_SV_RESET;
9838 sv_resetpvn(*s ? s : NULL, strlen(s), stash);
9842 Perl_sv_resetpvn(pTHX_ const char *s, STRLEN len, HV * const stash)
9844 char todo[PERL_UCHAR_MAX+1];
9847 if (!stash || SvTYPE(stash) != SVt_PVHV)
9850 if (!s) { /* reset ?? searches */
9851 MAGIC * const mg = mg_find((const SV *)stash, PERL_MAGIC_symtab);
9853 const U32 count = mg->mg_len / sizeof(PMOP**);
9854 PMOP **pmp = (PMOP**) mg->mg_ptr;
9855 PMOP *const *const end = pmp + count;
9859 SvREADONLY_off(PL_regex_pad[(*pmp)->op_pmoffset]);
9861 (*pmp)->op_pmflags &= ~PMf_USED;
9869 /* reset variables */
9871 if (!HvARRAY(stash))
9874 Zero(todo, 256, char);
9878 I32 i = (unsigned char)*s;
9882 max = (unsigned char)*s++;
9883 for ( ; i <= max; i++) {
9886 for (i = 0; i <= (I32) HvMAX(stash); i++) {
9888 for (entry = HvARRAY(stash)[i];
9890 entry = HeNEXT(entry))
9895 if (!todo[(U8)*HeKEY(entry)])
9897 gv = MUTABLE_GV(HeVAL(entry));
9901 if (sv && !SvREADONLY(sv)) {
9902 SV_CHECK_THINKFIRST_COW_DROP(sv);
9903 if (!isGV(sv)) SvOK_off(sv);
9908 if (GvHV(gv) && !HvNAME_get(GvHV(gv))) {
9919 Using various gambits, try to get an IO from an SV: the IO slot if its a
9920 GV; or the recursive result if we're an RV; or the IO slot of the symbol
9921 named after the PV if we're a string.
9923 'Get' magic is ignored on the C<sv> passed in, but will be called on
9924 C<SvRV(sv)> if C<sv> is an RV.
9930 Perl_sv_2io(pTHX_ SV *const sv)
9935 PERL_ARGS_ASSERT_SV_2IO;
9937 switch (SvTYPE(sv)) {
9939 io = MUTABLE_IO(sv);
9943 if (isGV_with_GP(sv)) {
9944 gv = MUTABLE_GV(sv);
9947 Perl_croak(aTHX_ "Bad filehandle: %" HEKf,
9948 HEKfARG(GvNAME_HEK(gv)));
9954 Perl_croak(aTHX_ PL_no_usym, "filehandle");
9956 SvGETMAGIC(SvRV(sv));
9957 return sv_2io(SvRV(sv));
9959 gv = gv_fetchsv_nomg(sv, 0, SVt_PVIO);
9966 if (SvGMAGICAL(sv)) {
9967 newsv = sv_newmortal();
9968 sv_setsv_nomg(newsv, sv);
9970 Perl_croak(aTHX_ "Bad filehandle: %" SVf, SVfARG(newsv));
9980 Using various gambits, try to get a CV from an SV; in addition, try if
9981 possible to set C<*st> and C<*gvp> to the stash and GV associated with it.
9982 The flags in C<lref> are passed to C<gv_fetchsv>.
9988 Perl_sv_2cv(pTHX_ SV *sv, HV **const st, GV **const gvp, const I32 lref)
9993 PERL_ARGS_ASSERT_SV_2CV;
10000 switch (SvTYPE(sv)) {
10004 return MUTABLE_CV(sv);
10014 sv = amagic_deref_call(sv, to_cv_amg);
10017 if (SvTYPE(sv) == SVt_PVCV) {
10018 cv = MUTABLE_CV(sv);
10023 else if(SvGETMAGIC(sv), isGV_with_GP(sv))
10024 gv = MUTABLE_GV(sv);
10026 Perl_croak(aTHX_ "Not a subroutine reference");
10028 else if (isGV_with_GP(sv)) {
10029 gv = MUTABLE_GV(sv);
10032 gv = gv_fetchsv_nomg(sv, lref, SVt_PVCV);
10039 /* Some flags to gv_fetchsv mean don't really create the GV */
10040 if (!isGV_with_GP(gv)) {
10044 *st = GvESTASH(gv);
10045 if (lref & ~GV_ADDMG && !GvCVu(gv)) {
10046 /* XXX this is probably not what they think they're getting.
10047 * It has the same effect as "sub name;", i.e. just a forward
10056 =for apidoc sv_true
10058 Returns true if the SV has a true value by Perl's rules.
10059 Use the C<SvTRUE> macro instead, which may call C<sv_true()> or may
10060 instead use an in-line version.
10066 Perl_sv_true(pTHX_ SV *const sv)
10071 const XPV* const tXpv = (XPV*)SvANY(sv);
10073 (tXpv->xpv_cur > 1 ||
10074 (tXpv->xpv_cur && *sv->sv_u.svu_pv != '0')))
10081 return SvIVX(sv) != 0;
10084 return SvNVX(sv) != 0.0;
10086 return sv_2bool(sv);
10092 =for apidoc sv_pvn_force
10094 Get a sensible string out of the SV somehow.
10095 A private implementation of the C<SvPV_force> macro for compilers which
10096 can't cope with complex macro expressions. Always use the macro instead.
10098 =for apidoc sv_pvn_force_flags
10100 Get a sensible string out of the SV somehow.
10101 If C<flags> has the C<SV_GMAGIC> bit set, will C<mg_get> on C<sv> if
10102 appropriate, else not. C<sv_pvn_force> and C<sv_pvn_force_nomg> are
10103 implemented in terms of this function.
10104 You normally want to use the various wrapper macros instead: see
10105 C<L</SvPV_force>> and C<L</SvPV_force_nomg>>.
10111 Perl_sv_pvn_force_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
10113 PERL_ARGS_ASSERT_SV_PVN_FORCE_FLAGS;
10115 if (flags & SV_GMAGIC) SvGETMAGIC(sv);
10116 if (SvTHINKFIRST(sv) && (!SvROK(sv) || SvREADONLY(sv)))
10117 sv_force_normal_flags(sv, 0);
10127 if (SvTYPE(sv) > SVt_PVLV
10128 || isGV_with_GP(sv))
10129 /* diag_listed_as: Can't coerce %s to %s in %s */
10130 Perl_croak(aTHX_ "Can't coerce %s to string in %s", sv_reftype(sv,0),
10132 s = sv_2pv_flags(sv, &len, flags &~ SV_GMAGIC);
10139 if (SvTYPE(sv) < SVt_PV ||
10140 s != SvPVX_const(sv)) { /* Almost, but not quite, sv_setpvn() */
10143 SvUPGRADE(sv, SVt_PV); /* Never FALSE */
10144 SvGROW(sv, len + 1);
10145 Move(s,SvPVX(sv),len,char);
10146 SvCUR_set(sv, len);
10147 SvPVX(sv)[len] = '\0';
10150 SvPOK_on(sv); /* validate pointer */
10152 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2pv(%s)\n",
10153 PTR2UV(sv),SvPVX_const(sv)));
10156 (void)SvPOK_only_UTF8(sv);
10157 return SvPVX_mutable(sv);
10161 =for apidoc sv_pvbyten_force
10163 The backend for the C<SvPVbytex_force> macro. Always use the macro
10170 Perl_sv_pvbyten_force(pTHX_ SV *const sv, STRLEN *const lp)
10172 PERL_ARGS_ASSERT_SV_PVBYTEN_FORCE;
10174 sv_pvn_force(sv,lp);
10175 sv_utf8_downgrade(sv,0);
10181 =for apidoc sv_pvutf8n_force
10183 The backend for the C<SvPVutf8x_force> macro. Always use the macro
10190 Perl_sv_pvutf8n_force(pTHX_ SV *const sv, STRLEN *const lp)
10192 PERL_ARGS_ASSERT_SV_PVUTF8N_FORCE;
10194 sv_pvn_force(sv,0);
10195 sv_utf8_upgrade_nomg(sv);
10201 =for apidoc sv_reftype
10203 Returns a string describing what the SV is a reference to.
10205 If ob is true and the SV is blessed, the string is the class name,
10206 otherwise it is the type of the SV, "SCALAR", "ARRAY" etc.
10212 Perl_sv_reftype(pTHX_ const SV *const sv, const int ob)
10214 PERL_ARGS_ASSERT_SV_REFTYPE;
10215 if (ob && SvOBJECT(sv)) {
10216 return SvPV_nolen_const(sv_ref(NULL, sv, ob));
10219 /* WARNING - There is code, for instance in mg.c, that assumes that
10220 * the only reason that sv_reftype(sv,0) would return a string starting
10221 * with 'L' or 'S' is that it is a LVALUE or a SCALAR.
10222 * Yes this a dodgy way to do type checking, but it saves practically reimplementing
10223 * this routine inside other subs, and it saves time.
10224 * Do not change this assumption without searching for "dodgy type check" in
10227 switch (SvTYPE(sv)) {
10242 case SVt_PVLV: return (char *) (SvROK(sv) ? "REF"
10243 /* tied lvalues should appear to be
10244 * scalars for backwards compatibility */
10245 : (isALPHA_FOLD_EQ(LvTYPE(sv), 't'))
10246 ? "SCALAR" : "LVALUE");
10247 case SVt_PVAV: return "ARRAY";
10248 case SVt_PVHV: return "HASH";
10249 case SVt_PVCV: return "CODE";
10250 case SVt_PVGV: return (char *) (isGV_with_GP(sv)
10251 ? "GLOB" : "SCALAR");
10252 case SVt_PVFM: return "FORMAT";
10253 case SVt_PVIO: return "IO";
10254 case SVt_INVLIST: return "INVLIST";
10255 case SVt_REGEXP: return "REGEXP";
10256 default: return "UNKNOWN";
10264 Returns a SV describing what the SV passed in is a reference to.
10266 dst can be a SV to be set to the description or NULL, in which case a
10267 mortal SV is returned.
10269 If ob is true and the SV is blessed, the description is the class
10270 name, otherwise it is the type of the SV, "SCALAR", "ARRAY" etc.
10276 Perl_sv_ref(pTHX_ SV *dst, const SV *const sv, const int ob)
10278 PERL_ARGS_ASSERT_SV_REF;
10281 dst = sv_newmortal();
10283 if (ob && SvOBJECT(sv)) {
10284 HvNAME_get(SvSTASH(sv))
10285 ? sv_sethek(dst, HvNAME_HEK(SvSTASH(sv)))
10286 : sv_setpvs(dst, "__ANON__");
10289 const char * reftype = sv_reftype(sv, 0);
10290 sv_setpv(dst, reftype);
10296 =for apidoc sv_isobject
10298 Returns a boolean indicating whether the SV is an RV pointing to a blessed
10299 object. If the SV is not an RV, or if the object is not blessed, then this
10306 Perl_sv_isobject(pTHX_ SV *sv)
10322 Returns a boolean indicating whether the SV is blessed into the specified
10323 class. This does not check for subtypes; use C<sv_derived_from> to verify
10324 an inheritance relationship.
10330 Perl_sv_isa(pTHX_ SV *sv, const char *const name)
10332 const char *hvname;
10334 PERL_ARGS_ASSERT_SV_ISA;
10344 hvname = HvNAME_get(SvSTASH(sv));
10348 return strEQ(hvname, name);
10352 =for apidoc newSVrv
10354 Creates a new SV for the existing RV, C<rv>, to point to. If C<rv> is not an
10355 RV then it will be upgraded to one. If C<classname> is non-null then the new
10356 SV will be blessed in the specified package. The new SV is returned and its
10357 reference count is 1. The reference count 1 is owned by C<rv>.
10363 Perl_newSVrv(pTHX_ SV *const rv, const char *const classname)
10367 PERL_ARGS_ASSERT_NEWSVRV;
10371 SV_CHECK_THINKFIRST_COW_DROP(rv);
10373 if (UNLIKELY( SvTYPE(rv) >= SVt_PVMG )) {
10374 const U32 refcnt = SvREFCNT(rv);
10378 SvREFCNT(rv) = refcnt;
10380 sv_upgrade(rv, SVt_IV);
10381 } else if (SvROK(rv)) {
10382 SvREFCNT_dec(SvRV(rv));
10384 prepare_SV_for_RV(rv);
10392 HV* const stash = gv_stashpv(classname, GV_ADD);
10393 (void)sv_bless(rv, stash);
10399 Perl_newSVavdefelem(pTHX_ AV *av, SSize_t ix, bool extendible)
10401 SV * const lv = newSV_type(SVt_PVLV);
10402 PERL_ARGS_ASSERT_NEWSVAVDEFELEM;
10404 sv_magic(lv, NULL, PERL_MAGIC_defelem, NULL, 0);
10405 LvTARG(lv) = SvREFCNT_inc_simple_NN(av);
10406 LvSTARGOFF(lv) = ix;
10407 LvTARGLEN(lv) = extendible ? 1 : (STRLEN)UV_MAX;
10412 =for apidoc sv_setref_pv
10414 Copies a pointer into a new SV, optionally blessing the SV. The C<rv>
10415 argument will be upgraded to an RV. That RV will be modified to point to
10416 the new SV. If the C<pv> argument is C<NULL>, then C<PL_sv_undef> will be placed
10417 into the SV. The C<classname> argument indicates the package for the
10418 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10419 will have a reference count of 1, and the RV will be returned.
10421 Do not use with other Perl types such as HV, AV, SV, CV, because those
10422 objects will become corrupted by the pointer copy process.
10424 Note that C<sv_setref_pvn> copies the string while this copies the pointer.
10430 Perl_sv_setref_pv(pTHX_ SV *const rv, const char *const classname, void *const pv)
10432 PERL_ARGS_ASSERT_SV_SETREF_PV;
10439 sv_setiv(newSVrv(rv,classname), PTR2IV(pv));
10444 =for apidoc sv_setref_iv
10446 Copies an integer into a new SV, optionally blessing the SV. The C<rv>
10447 argument will be upgraded to an RV. That RV will be modified to point to
10448 the new SV. The C<classname> argument indicates the package for the
10449 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10450 will have a reference count of 1, and the RV will be returned.
10456 Perl_sv_setref_iv(pTHX_ SV *const rv, const char *const classname, const IV iv)
10458 PERL_ARGS_ASSERT_SV_SETREF_IV;
10460 sv_setiv(newSVrv(rv,classname), iv);
10465 =for apidoc sv_setref_uv
10467 Copies an unsigned integer into a new SV, optionally blessing the SV. The C<rv>
10468 argument will be upgraded to an RV. That RV will be modified to point to
10469 the new SV. The C<classname> argument indicates the package for the
10470 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10471 will have a reference count of 1, and the RV will be returned.
10477 Perl_sv_setref_uv(pTHX_ SV *const rv, const char *const classname, const UV uv)
10479 PERL_ARGS_ASSERT_SV_SETREF_UV;
10481 sv_setuv(newSVrv(rv,classname), uv);
10486 =for apidoc sv_setref_nv
10488 Copies a double into a new SV, optionally blessing the SV. The C<rv>
10489 argument will be upgraded to an RV. That RV will be modified to point to
10490 the new SV. The C<classname> argument indicates the package for the
10491 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10492 will have a reference count of 1, and the RV will be returned.
10498 Perl_sv_setref_nv(pTHX_ SV *const rv, const char *const classname, const NV nv)
10500 PERL_ARGS_ASSERT_SV_SETREF_NV;
10502 sv_setnv(newSVrv(rv,classname), nv);
10507 =for apidoc sv_setref_pvn
10509 Copies a string into a new SV, optionally blessing the SV. The length of the
10510 string must be specified with C<n>. The C<rv> argument will be upgraded to
10511 an RV. That RV will be modified to point to the new SV. The C<classname>
10512 argument indicates the package for the blessing. Set C<classname> to
10513 C<NULL> to avoid the blessing. The new SV will have a reference count
10514 of 1, and the RV will be returned.
10516 Note that C<sv_setref_pv> copies the pointer while this copies the string.
10522 Perl_sv_setref_pvn(pTHX_ SV *const rv, const char *const classname,
10523 const char *const pv, const STRLEN n)
10525 PERL_ARGS_ASSERT_SV_SETREF_PVN;
10527 sv_setpvn(newSVrv(rv,classname), pv, n);
10532 =for apidoc sv_bless
10534 Blesses an SV into a specified package. The SV must be an RV. The package
10535 must be designated by its stash (see C<L</gv_stashpv>>). The reference count
10536 of the SV is unaffected.
10542 Perl_sv_bless(pTHX_ SV *const sv, HV *const stash)
10545 HV *oldstash = NULL;
10547 PERL_ARGS_ASSERT_SV_BLESS;
10551 Perl_croak(aTHX_ "Can't bless non-reference value");
10553 if (SvFLAGS(tmpRef) & (SVs_OBJECT|SVf_READONLY|SVf_PROTECT)) {
10554 if (SvREADONLY(tmpRef))
10555 Perl_croak_no_modify();
10556 if (SvOBJECT(tmpRef)) {
10557 oldstash = SvSTASH(tmpRef);
10560 SvOBJECT_on(tmpRef);
10561 SvUPGRADE(tmpRef, SVt_PVMG);
10562 SvSTASH_set(tmpRef, MUTABLE_HV(SvREFCNT_inc_simple(stash)));
10563 SvREFCNT_dec(oldstash);
10565 if(SvSMAGICAL(tmpRef))
10566 if(mg_find(tmpRef, PERL_MAGIC_ext) || mg_find(tmpRef, PERL_MAGIC_uvar))
10574 /* Downgrades a PVGV to a PVMG. If it's actually a PVLV, we leave the type
10575 * as it is after unglobbing it.
10578 PERL_STATIC_INLINE void
10579 S_sv_unglob(pTHX_ SV *const sv, U32 flags)
10583 SV * const temp = flags & SV_COW_DROP_PV ? NULL : sv_newmortal();
10585 PERL_ARGS_ASSERT_SV_UNGLOB;
10587 assert(SvTYPE(sv) == SVt_PVGV || SvTYPE(sv) == SVt_PVLV);
10589 if (!(flags & SV_COW_DROP_PV))
10590 gv_efullname3(temp, MUTABLE_GV(sv), "*");
10592 SvREFCNT_inc_simple_void_NN(sv_2mortal(sv));
10594 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
10595 && HvNAME_get(stash))
10596 mro_method_changed_in(stash);
10597 gp_free(MUTABLE_GV(sv));
10600 sv_del_backref(MUTABLE_SV(GvSTASH(sv)), sv);
10601 GvSTASH(sv) = NULL;
10604 if (GvNAME_HEK(sv)) {
10605 unshare_hek(GvNAME_HEK(sv));
10607 isGV_with_GP_off(sv);
10609 if(SvTYPE(sv) == SVt_PVGV) {
10610 /* need to keep SvANY(sv) in the right arena */
10611 xpvmg = new_XPVMG();
10612 StructCopy(SvANY(sv), xpvmg, XPVMG);
10613 del_XPVGV(SvANY(sv));
10616 SvFLAGS(sv) &= ~SVTYPEMASK;
10617 SvFLAGS(sv) |= SVt_PVMG;
10620 /* Intentionally not calling any local SET magic, as this isn't so much a
10621 set operation as merely an internal storage change. */
10622 if (flags & SV_COW_DROP_PV) SvOK_off(sv);
10623 else sv_setsv_flags(sv, temp, 0);
10625 if ((const GV *)sv == PL_last_in_gv)
10626 PL_last_in_gv = NULL;
10627 else if ((const GV *)sv == PL_statgv)
10632 =for apidoc sv_unref_flags
10634 Unsets the RV status of the SV, and decrements the reference count of
10635 whatever was being referenced by the RV. This can almost be thought of
10636 as a reversal of C<newSVrv>. The C<cflags> argument can contain
10637 C<SV_IMMEDIATE_UNREF> to force the reference count to be decremented
10638 (otherwise the decrementing is conditional on the reference count being
10639 different from one or the reference being a readonly SV).
10640 See C<L</SvROK_off>>.
10646 Perl_sv_unref_flags(pTHX_ SV *const ref, const U32 flags)
10648 SV* const target = SvRV(ref);
10650 PERL_ARGS_ASSERT_SV_UNREF_FLAGS;
10652 if (SvWEAKREF(ref)) {
10653 sv_del_backref(target, ref);
10654 SvWEAKREF_off(ref);
10655 SvRV_set(ref, NULL);
10658 SvRV_set(ref, NULL);
10660 /* You can't have a || SvREADONLY(target) here, as $a = $$a, where $a was
10661 assigned to as BEGIN {$a = \"Foo"} will fail. */
10662 if (SvREFCNT(target) != 1 || (flags & SV_IMMEDIATE_UNREF))
10663 SvREFCNT_dec_NN(target);
10664 else /* XXX Hack, but hard to make $a=$a->[1] work otherwise */
10665 sv_2mortal(target); /* Schedule for freeing later */
10669 =for apidoc sv_untaint
10671 Untaint an SV. Use C<SvTAINTED_off> instead.
10677 Perl_sv_untaint(pTHX_ SV *const sv)
10679 PERL_ARGS_ASSERT_SV_UNTAINT;
10680 PERL_UNUSED_CONTEXT;
10682 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10683 MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10690 =for apidoc sv_tainted
10692 Test an SV for taintedness. Use C<SvTAINTED> instead.
10698 Perl_sv_tainted(pTHX_ SV *const sv)
10700 PERL_ARGS_ASSERT_SV_TAINTED;
10701 PERL_UNUSED_CONTEXT;
10703 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10704 const MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10705 if (mg && (mg->mg_len & 1) )
10711 #ifndef NO_MATHOMS /* Can't move these to mathoms.c because call uiv_2buf(),
10712 private to this file */
10715 =for apidoc sv_setpviv
10717 Copies an integer into the given SV, also updating its string value.
10718 Does not handle 'set' magic. See C<L</sv_setpviv_mg>>.
10724 Perl_sv_setpviv(pTHX_ SV *const sv, const IV iv)
10726 char buf[TYPE_CHARS(UV)];
10728 char * const ptr = uiv_2buf(buf, iv, 0, 0, &ebuf);
10730 PERL_ARGS_ASSERT_SV_SETPVIV;
10732 sv_setpvn(sv, ptr, ebuf - ptr);
10736 =for apidoc sv_setpviv_mg
10738 Like C<sv_setpviv>, but also handles 'set' magic.
10744 Perl_sv_setpviv_mg(pTHX_ SV *const sv, const IV iv)
10746 PERL_ARGS_ASSERT_SV_SETPVIV_MG;
10748 sv_setpviv(sv, iv);
10752 #endif /* NO_MATHOMS */
10754 #if defined(PERL_IMPLICIT_CONTEXT)
10756 /* pTHX_ magic can't cope with varargs, so this is a no-context
10757 * version of the main function, (which may itself be aliased to us).
10758 * Don't access this version directly.
10762 Perl_sv_setpvf_nocontext(SV *const sv, const char *const pat, ...)
10767 PERL_ARGS_ASSERT_SV_SETPVF_NOCONTEXT;
10769 va_start(args, pat);
10770 sv_vsetpvf(sv, pat, &args);
10774 /* pTHX_ magic can't cope with varargs, so this is a no-context
10775 * version of the main function, (which may itself be aliased to us).
10776 * Don't access this version directly.
10780 Perl_sv_setpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10785 PERL_ARGS_ASSERT_SV_SETPVF_MG_NOCONTEXT;
10787 va_start(args, pat);
10788 sv_vsetpvf_mg(sv, pat, &args);
10794 =for apidoc sv_setpvf
10796 Works like C<sv_catpvf> but copies the text into the SV instead of
10797 appending it. Does not handle 'set' magic. See C<L</sv_setpvf_mg>>.
10803 Perl_sv_setpvf(pTHX_ SV *const sv, const char *const pat, ...)
10807 PERL_ARGS_ASSERT_SV_SETPVF;
10809 va_start(args, pat);
10810 sv_vsetpvf(sv, pat, &args);
10815 =for apidoc sv_vsetpvf
10817 Works like C<sv_vcatpvf> but copies the text into the SV instead of
10818 appending it. Does not handle 'set' magic. See C<L</sv_vsetpvf_mg>>.
10820 Usually used via its frontend C<sv_setpvf>.
10826 Perl_sv_vsetpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10828 PERL_ARGS_ASSERT_SV_VSETPVF;
10830 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10834 =for apidoc sv_setpvf_mg
10836 Like C<sv_setpvf>, but also handles 'set' magic.
10842 Perl_sv_setpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10846 PERL_ARGS_ASSERT_SV_SETPVF_MG;
10848 va_start(args, pat);
10849 sv_vsetpvf_mg(sv, pat, &args);
10854 =for apidoc sv_vsetpvf_mg
10856 Like C<sv_vsetpvf>, but also handles 'set' magic.
10858 Usually used via its frontend C<sv_setpvf_mg>.
10864 Perl_sv_vsetpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10866 PERL_ARGS_ASSERT_SV_VSETPVF_MG;
10868 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10872 #if defined(PERL_IMPLICIT_CONTEXT)
10874 /* pTHX_ magic can't cope with varargs, so this is a no-context
10875 * version of the main function, (which may itself be aliased to us).
10876 * Don't access this version directly.
10880 Perl_sv_catpvf_nocontext(SV *const sv, const char *const pat, ...)
10885 PERL_ARGS_ASSERT_SV_CATPVF_NOCONTEXT;
10887 va_start(args, pat);
10888 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10892 /* pTHX_ magic can't cope with varargs, so this is a no-context
10893 * version of the main function, (which may itself be aliased to us).
10894 * Don't access this version directly.
10898 Perl_sv_catpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10903 PERL_ARGS_ASSERT_SV_CATPVF_MG_NOCONTEXT;
10905 va_start(args, pat);
10906 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10913 =for apidoc sv_catpvf
10915 Processes its arguments like C<sv_catpvfn>, and appends the formatted
10916 output to an SV. As with C<sv_catpvfn> called with a non-null C-style
10917 variable argument list, argument reordering is not supported.
10918 If the appended data contains "wide" characters
10919 (including, but not limited to, SVs with a UTF-8 PV formatted with C<%s>,
10920 and characters >255 formatted with C<%c>), the original SV might get
10921 upgraded to UTF-8. Handles 'get' magic, but not 'set' magic. See
10922 C<L</sv_catpvf_mg>>. If the original SV was UTF-8, the pattern should be
10923 valid UTF-8; if the original SV was bytes, the pattern should be too.
10928 Perl_sv_catpvf(pTHX_ SV *const sv, const char *const pat, ...)
10932 PERL_ARGS_ASSERT_SV_CATPVF;
10934 va_start(args, pat);
10935 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10940 =for apidoc sv_vcatpvf
10942 Processes its arguments like C<sv_catpvfn> called with a non-null C-style
10943 variable argument list, and appends the formatted output
10944 to an SV. Does not handle 'set' magic. See C<L</sv_vcatpvf_mg>>.
10946 Usually used via its frontend C<sv_catpvf>.
10952 Perl_sv_vcatpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10954 PERL_ARGS_ASSERT_SV_VCATPVF;
10956 sv_vcatpvfn_flags(sv, pat, strlen(pat), args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10960 =for apidoc sv_catpvf_mg
10962 Like C<sv_catpvf>, but also handles 'set' magic.
10968 Perl_sv_catpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10972 PERL_ARGS_ASSERT_SV_CATPVF_MG;
10974 va_start(args, pat);
10975 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10981 =for apidoc sv_vcatpvf_mg
10983 Like C<sv_vcatpvf>, but also handles 'set' magic.
10985 Usually used via its frontend C<sv_catpvf_mg>.
10991 Perl_sv_vcatpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10993 PERL_ARGS_ASSERT_SV_VCATPVF_MG;
10995 sv_vcatpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
11000 =for apidoc sv_vsetpvfn
11002 Works like C<sv_vcatpvfn> but copies the text into the SV instead of
11005 Usually used via one of its frontends C<sv_vsetpvf> and C<sv_vsetpvf_mg>.
11011 Perl_sv_vsetpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11012 va_list *const args, SV **const svargs, const Size_t sv_count, bool *const maybe_tainted)
11014 PERL_ARGS_ASSERT_SV_VSETPVFN;
11017 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, sv_count, maybe_tainted, 0);
11021 /* simplified inline Perl_sv_catpvn_nomg() when you know the SV's SvPOK */
11023 PERL_STATIC_INLINE void
11024 S_sv_catpvn_simple(pTHX_ SV *const sv, const char* const buf, const STRLEN len)
11026 STRLEN const need = len + SvCUR(sv) + 1;
11029 /* can't wrap as both len and SvCUR() are allocated in
11030 * memory and together can't consume all the address space
11032 assert(need > len);
11037 Copy(buf, end, len, char);
11040 SvCUR_set(sv, need - 1);
11045 * Warn of missing argument to sprintf. The value used in place of such
11046 * arguments should be &PL_sv_no; an undefined value would yield
11047 * inappropriate "use of uninit" warnings [perl #71000].
11050 S_warn_vcatpvfn_missing_argument(pTHX) {
11051 if (ckWARN(WARN_MISSING)) {
11052 Perl_warner(aTHX_ packWARN(WARN_MISSING), "Missing argument in %s",
11053 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11062 Perl_croak(aTHX_ "Integer overflow in format string for %s",
11063 (PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn"));
11067 /* Given an int i from the next arg (if args is true) or an sv from an arg
11068 * (if args is false), try to extract a STRLEN-ranged value from the arg,
11069 * with overflow checking.
11070 * Sets *neg to true if the value was negative (untouched otherwise.
11071 * Returns the absolute value.
11072 * As an extra margin of safety, it croaks if the returned value would
11073 * exceed the maximum value of a STRLEN / 4.
11077 S_sprintf_arg_num_val(pTHX_ va_list *const args, int i, SV *sv, bool *neg)
11091 if (UNLIKELY(SvIsUV(sv))) {
11092 UV uv = SvUV_nomg(sv);
11094 S_croak_overflow();
11098 iv = SvIV_nomg(sv);
11102 S_croak_overflow();
11108 if (iv > (IV)(((STRLEN)~0) / 4))
11109 S_croak_overflow();
11115 /* Returns true if c is in the range '1'..'9'
11116 * Written with the cast so it only needs one conditional test
11118 #define IS_1_TO_9(c) ((U8)(c - '1') <= 8)
11120 /* Read in and return a number. Updates *pattern to point to the char
11121 * following the number. Expects the first char to 1..9.
11122 * Croaks if the number exceeds 1/4 of the maximum value of STRLEN.
11123 * This is a belt-and-braces safety measure to complement any
11124 * overflow/wrap checks done in the main body of sv_vcatpvfn_flags.
11125 * It means that e.g. on a 32-bit system the width/precision can't be more
11126 * than 1G, which seems reasonable.
11130 S_expect_number(pTHX_ const char **const pattern)
11134 PERL_ARGS_ASSERT_EXPECT_NUMBER;
11136 assert(IS_1_TO_9(**pattern));
11138 var = *(*pattern)++ - '0';
11139 while (isDIGIT(**pattern)) {
11140 /* if var * 10 + 9 would exceed 1/4 max strlen, croak */
11141 if (var > ((((STRLEN)~0) / 4 - 9) / 10))
11142 S_croak_overflow();
11143 var = var * 10 + (*(*pattern)++ - '0');
11148 /* Implement a fast "%.0f": given a pointer to the end of a buffer (caller
11149 * ensures it's big enough), back fill it with the rounded integer part of
11150 * nv. Returns ptr to start of string, and sets *len to its length.
11151 * Returns NULL if not convertible.
11155 S_F0convert(NV nv, char *const endbuf, STRLEN *const len)
11157 const int neg = nv < 0;
11160 PERL_ARGS_ASSERT_F0CONVERT;
11162 assert(!Perl_isinfnan(nv));
11169 if (uv & 1 && uv == nv)
11170 uv--; /* Round to even */
11172 const unsigned dig = uv % 10;
11174 } while (uv /= 10);
11184 /* XXX maybe_tainted is never assigned to, so the doc above is lying. */
11187 Perl_sv_vcatpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11188 va_list *const args, SV **const svargs, const Size_t sv_count, bool *const maybe_tainted)
11190 PERL_ARGS_ASSERT_SV_VCATPVFN;
11192 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, sv_count, maybe_tainted, SV_GMAGIC|SV_SMAGIC);
11196 /* For the vcatpvfn code, we need a long double target in case
11197 * HAS_LONG_DOUBLE, even without USE_LONG_DOUBLE, so that we can printf
11198 * with long double formats, even without NV being long double. But we
11199 * call the target 'fv' instead of 'nv', since most of the time it is not
11200 * (most compilers these days recognize "long double", even if only as a
11201 * synonym for "double").
11203 #if defined(HAS_LONG_DOUBLE) && LONG_DOUBLESIZE > DOUBLESIZE && \
11204 defined(PERL_PRIgldbl) && !defined(USE_QUADMATH)
11205 # define VCATPVFN_FV_GF PERL_PRIgldbl
11206 # if defined(__VMS) && defined(__ia64) && defined(__IEEE_FLOAT)
11207 /* Work around breakage in OTS$CVT_FLOAT_T_X */
11208 # define VCATPVFN_NV_TO_FV(nv,fv) \
11211 fv = Perl_isnan(_dv) ? LDBL_QNAN : _dv; \
11214 # define VCATPVFN_NV_TO_FV(nv,fv) (fv)=(nv)
11216 typedef long double vcatpvfn_long_double_t;
11218 # define VCATPVFN_FV_GF NVgf
11219 # define VCATPVFN_NV_TO_FV(nv,fv) (fv)=(nv)
11220 typedef NV vcatpvfn_long_double_t;
11223 #ifdef LONGDOUBLE_DOUBLEDOUBLE
11224 /* The first double can be as large as 2**1023, or '1' x '0' x 1023.
11225 * The second double can be as small as 2**-1074, or '0' x 1073 . '1'.
11226 * The sum of them can be '1' . '0' x 2096 . '1', with implied radix point
11227 * after the first 1023 zero bits.
11229 * XXX The 2098 is quite large (262.25 bytes) and therefore some sort
11230 * of dynamically growing buffer might be better, start at just 16 bytes
11231 * (for example) and grow only when necessary. Or maybe just by looking
11232 * at the exponents of the two doubles? */
11233 # define DOUBLEDOUBLE_MAXBITS 2098
11236 /* vhex will contain the values (0..15) of the hex digits ("nybbles"
11237 * of 4 bits); 1 for the implicit 1, and the mantissa bits, four bits
11238 * per xdigit. For the double-double case, this can be rather many.
11239 * The non-double-double-long-double overshoots since all bits of NV
11240 * are not mantissa bits, there are also exponent bits. */
11241 #ifdef LONGDOUBLE_DOUBLEDOUBLE
11242 # define VHEX_SIZE (3+DOUBLEDOUBLE_MAXBITS/4)
11244 # define VHEX_SIZE (1+(NVSIZE * 8)/4)
11247 /* If we do not have a known long double format, (including not using
11248 * long doubles, or long doubles being equal to doubles) then we will
11249 * fall back to the ldexp/frexp route, with which we can retrieve at
11250 * most as many bits as our widest unsigned integer type is. We try
11251 * to get a 64-bit unsigned integer even if we are not using a 64-bit UV.
11253 * (If you want to test the case of UVSIZE == 4, NVSIZE == 8,
11254 * set the MANTISSATYPE to int and the MANTISSASIZE to 4.)
11256 #if defined(HAS_QUAD) && defined(Uquad_t)
11257 # define MANTISSATYPE Uquad_t
11258 # define MANTISSASIZE 8
11260 # define MANTISSATYPE UV
11261 # define MANTISSASIZE UVSIZE
11264 #if defined(DOUBLE_LITTLE_ENDIAN) || defined(LONGDOUBLE_LITTLE_ENDIAN)
11265 # define HEXTRACT_LITTLE_ENDIAN
11266 #elif defined(DOUBLE_BIG_ENDIAN) || defined(LONGDOUBLE_BIG_ENDIAN)
11267 # define HEXTRACT_BIG_ENDIAN
11269 # define HEXTRACT_MIX_ENDIAN
11272 /* S_hextract() is a helper for S_format_hexfp, for extracting
11273 * the hexadecimal values (for %a/%A). The nv is the NV where the value
11274 * are being extracted from (either directly from the long double in-memory
11275 * presentation, or from the uquad computed via frexp+ldexp). frexp also
11276 * is used to update the exponent. The subnormal is set to true
11277 * for IEEE 754 subnormals/denormals (including the x86 80-bit format).
11278 * The vhex is the pointer to the beginning of the output buffer of VHEX_SIZE.
11280 * The tricky part is that S_hextract() needs to be called twice:
11281 * the first time with vend as NULL, and the second time with vend as
11282 * the pointer returned by the first call. What happens is that on
11283 * the first round the output size is computed, and the intended
11284 * extraction sanity checked. On the second round the actual output
11285 * (the extraction of the hexadecimal values) takes place.
11286 * Sanity failures cause fatal failures during both rounds. */
11288 S_hextract(pTHX_ const NV nv, int* exponent, bool *subnormal,
11289 U8* vhex, U8* vend)
11293 int ixmin = 0, ixmax = 0;
11295 /* XXX Inf/NaN are not handled here, since it is
11296 * assumed they are to be output as "Inf" and "NaN". */
11298 /* These macros are just to reduce typos, they have multiple
11299 * repetitions below, but usually only one (or sometimes two)
11300 * of them is really being used. */
11301 /* HEXTRACT_OUTPUT() extracts the high nybble first. */
11302 #define HEXTRACT_OUTPUT_HI(ix) (*v++ = nvp[ix] >> 4)
11303 #define HEXTRACT_OUTPUT_LO(ix) (*v++ = nvp[ix] & 0xF)
11304 #define HEXTRACT_OUTPUT(ix) \
11306 HEXTRACT_OUTPUT_HI(ix); HEXTRACT_OUTPUT_LO(ix); \
11308 #define HEXTRACT_COUNT(ix, c) \
11310 v += c; if (ix < ixmin) ixmin = ix; else if (ix > ixmax) ixmax = ix; \
11312 #define HEXTRACT_BYTE(ix) \
11314 if (vend) HEXTRACT_OUTPUT(ix); else HEXTRACT_COUNT(ix, 2); \
11316 #define HEXTRACT_LO_NYBBLE(ix) \
11318 if (vend) HEXTRACT_OUTPUT_LO(ix); else HEXTRACT_COUNT(ix, 1); \
11320 /* HEXTRACT_TOP_NYBBLE is just convenience disguise,
11321 * to make it look less odd when the top bits of a NV
11322 * are extracted using HEXTRACT_LO_NYBBLE: the highest
11323 * order bits can be in the "low nybble" of a byte. */
11324 #define HEXTRACT_TOP_NYBBLE(ix) HEXTRACT_LO_NYBBLE(ix)
11325 #define HEXTRACT_BYTES_LE(a, b) \
11326 for (ix = a; ix >= b; ix--) { HEXTRACT_BYTE(ix); }
11327 #define HEXTRACT_BYTES_BE(a, b) \
11328 for (ix = a; ix <= b; ix++) { HEXTRACT_BYTE(ix); }
11329 #define HEXTRACT_GET_SUBNORMAL(nv) *subnormal = Perl_fp_class_denorm(nv)
11330 #define HEXTRACT_IMPLICIT_BIT(nv) \
11332 if (!*subnormal) { \
11333 if (vend) *v++ = ((nv) == 0.0) ? 0 : 1; else v++; \
11337 /* Most formats do. Those which don't should undef this.
11339 * But also note that IEEE 754 subnormals do not have it, or,
11340 * expressed alternatively, their implicit bit is zero. */
11341 #define HEXTRACT_HAS_IMPLICIT_BIT
11343 /* Many formats do. Those which don't should undef this. */
11344 #define HEXTRACT_HAS_TOP_NYBBLE
11346 /* HEXTRACTSIZE is the maximum number of xdigits. */
11347 #if defined(USE_LONG_DOUBLE) && defined(LONGDOUBLE_DOUBLEDOUBLE)
11348 # define HEXTRACTSIZE (2+DOUBLEDOUBLE_MAXBITS/4)
11350 # define HEXTRACTSIZE 2 * NVSIZE
11353 const U8* vmaxend = vhex + HEXTRACTSIZE;
11355 assert(HEXTRACTSIZE <= VHEX_SIZE);
11357 PERL_UNUSED_VAR(ix); /* might happen */
11358 (void)Perl_frexp(PERL_ABS(nv), exponent);
11359 *subnormal = FALSE;
11360 if (vend && (vend <= vhex || vend > vmaxend)) {
11361 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11362 Perl_croak(aTHX_ "Hexadecimal float: internal error (entry)");
11365 /* First check if using long doubles. */
11366 #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE)
11367 # if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN
11368 /* Used in e.g. VMS and HP-UX IA-64, e.g. -0.1L:
11369 * 9a 99 99 99 99 99 99 99 99 99 99 99 99 99 fb bf */
11370 /* The bytes 13..0 are the mantissa/fraction,
11371 * the 15,14 are the sign+exponent. */
11372 const U8* nvp = (const U8*)(&nv);
11373 HEXTRACT_GET_SUBNORMAL(nv);
11374 HEXTRACT_IMPLICIT_BIT(nv);
11375 # undef HEXTRACT_HAS_TOP_NYBBLE
11376 HEXTRACT_BYTES_LE(13, 0);
11377 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_BIG_ENDIAN
11378 /* Used in e.g. Solaris Sparc and HP-UX PA-RISC, e.g. -0.1L:
11379 * bf fb 99 99 99 99 99 99 99 99 99 99 99 99 99 9a */
11380 /* The bytes 2..15 are the mantissa/fraction,
11381 * the 0,1 are the sign+exponent. */
11382 const U8* nvp = (const U8*)(&nv);
11383 HEXTRACT_GET_SUBNORMAL(nv);
11384 HEXTRACT_IMPLICIT_BIT(nv);
11385 # undef HEXTRACT_HAS_TOP_NYBBLE
11386 HEXTRACT_BYTES_BE(2, 15);
11387 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN
11388 /* x86 80-bit "extended precision", 64 bits of mantissa / fraction /
11389 * significand, 15 bits of exponent, 1 bit of sign. No implicit bit.
11390 * NVSIZE can be either 12 (ILP32, Solaris x86) or 16 (LP64, Linux
11391 * and OS X), meaning that 2 or 6 bytes are empty padding. */
11392 /* The bytes 0..1 are the sign+exponent,
11393 * the bytes 2..9 are the mantissa/fraction. */
11394 const U8* nvp = (const U8*)(&nv);
11395 # undef HEXTRACT_HAS_IMPLICIT_BIT
11396 # undef HEXTRACT_HAS_TOP_NYBBLE
11397 HEXTRACT_GET_SUBNORMAL(nv);
11398 HEXTRACT_BYTES_LE(7, 0);
11399 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN
11400 /* Does this format ever happen? (Wikipedia says the Motorola
11401 * 6888x math coprocessors used format _like_ this but padded
11402 * to 96 bits with 16 unused bits between the exponent and the
11404 const U8* nvp = (const U8*)(&nv);
11405 # undef HEXTRACT_HAS_IMPLICIT_BIT
11406 # undef HEXTRACT_HAS_TOP_NYBBLE
11407 HEXTRACT_GET_SUBNORMAL(nv);
11408 HEXTRACT_BYTES_BE(0, 7);
11410 # define HEXTRACT_FALLBACK
11411 /* Double-double format: two doubles next to each other.
11412 * The first double is the high-order one, exactly like
11413 * it would be for a "lone" double. The second double
11414 * is shifted down using the exponent so that that there
11415 * are no common bits. The tricky part is that the value
11416 * of the double-double is the SUM of the two doubles and
11417 * the second one can be also NEGATIVE.
11419 * Because of this tricky construction the bytewise extraction we
11420 * use for the other long double formats doesn't work, we must
11421 * extract the values bit by bit.
11423 * The little-endian double-double is used .. somewhere?
11425 * The big endian double-double is used in e.g. PPC/Power (AIX)
11428 * The mantissa bits are in two separate stretches, e.g. for -0.1L:
11429 * 9a 99 99 99 99 99 59 bc 9a 99 99 99 99 99 b9 3f (LE)
11430 * 3f b9 99 99 99 99 99 9a bc 59 99 99 99 99 99 9a (BE)
11433 #else /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) */
11434 /* Using normal doubles, not long doubles.
11436 * We generate 4-bit xdigits (nybble/nibble) instead of 8-bit
11437 * bytes, since we might need to handle printf precision, and
11438 * also need to insert the radix. */
11440 # ifdef HEXTRACT_LITTLE_ENDIAN
11441 /* 0 1 2 3 4 5 6 7 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11442 const U8* nvp = (const U8*)(&nv);
11443 HEXTRACT_GET_SUBNORMAL(nv);
11444 HEXTRACT_IMPLICIT_BIT(nv);
11445 HEXTRACT_TOP_NYBBLE(6);
11446 HEXTRACT_BYTES_LE(5, 0);
11447 # elif defined(HEXTRACT_BIG_ENDIAN)
11448 /* 7 6 5 4 3 2 1 0 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11449 const U8* nvp = (const U8*)(&nv);
11450 HEXTRACT_GET_SUBNORMAL(nv);
11451 HEXTRACT_IMPLICIT_BIT(nv);
11452 HEXTRACT_TOP_NYBBLE(1);
11453 HEXTRACT_BYTES_BE(2, 7);
11454 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_LE_BE
11455 /* 4 5 6 7 0 1 2 3 (MSB = 7, LSB = 0, 6:7 = nybble:exponent:sign) */
11456 const U8* nvp = (const U8*)(&nv);
11457 HEXTRACT_GET_SUBNORMAL(nv);
11458 HEXTRACT_IMPLICIT_BIT(nv);
11459 HEXTRACT_TOP_NYBBLE(2); /* 6 */
11460 HEXTRACT_BYTE(1); /* 5 */
11461 HEXTRACT_BYTE(0); /* 4 */
11462 HEXTRACT_BYTE(7); /* 3 */
11463 HEXTRACT_BYTE(6); /* 2 */
11464 HEXTRACT_BYTE(5); /* 1 */
11465 HEXTRACT_BYTE(4); /* 0 */
11466 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_BE_LE
11467 /* 3 2 1 0 7 6 5 4 (MSB = 7, LSB = 0, 7:6 = sign:exponent:nybble) */
11468 const U8* nvp = (const U8*)(&nv);
11469 HEXTRACT_GET_SUBNORMAL(nv);
11470 HEXTRACT_IMPLICIT_BIT(nv);
11471 HEXTRACT_TOP_NYBBLE(5); /* 6 */
11472 HEXTRACT_BYTE(6); /* 5 */
11473 HEXTRACT_BYTE(7); /* 4 */
11474 HEXTRACT_BYTE(0); /* 3 */
11475 HEXTRACT_BYTE(1); /* 2 */
11476 HEXTRACT_BYTE(2); /* 1 */
11477 HEXTRACT_BYTE(3); /* 0 */
11479 # define HEXTRACT_FALLBACK
11482 # define HEXTRACT_FALLBACK
11484 #endif /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) #else */
11486 #ifdef HEXTRACT_FALLBACK
11487 HEXTRACT_GET_SUBNORMAL(nv);
11488 # undef HEXTRACT_HAS_TOP_NYBBLE /* Meaningless, but consistent. */
11489 /* The fallback is used for the double-double format, and
11490 * for unknown long double formats, and for unknown double
11491 * formats, or in general unknown NV formats. */
11492 if (nv == (NV)0.0) {
11500 NV d = nv < 0 ? -nv : nv;
11502 U8 ha = 0x0; /* hexvalue accumulator */
11503 U8 hd = 0x8; /* hexvalue digit */
11505 /* Shift d and e (and update exponent) so that e <= d < 2*e,
11506 * this is essentially manual frexp(). Multiplying by 0.5 and
11507 * doubling should be lossless in binary floating point. */
11517 while (d >= e + e) {
11521 /* Now e <= d < 2*e */
11523 /* First extract the leading hexdigit (the implicit bit). */
11539 /* Then extract the remaining hexdigits. */
11540 while (d > (NV)0.0) {
11546 /* Output or count in groups of four bits,
11547 * that is, when the hexdigit is down to one. */
11552 /* Reset the hexvalue. */
11561 /* Flush possible pending hexvalue. */
11571 /* Croak for various reasons: if the output pointer escaped the
11572 * output buffer, if the extraction index escaped the extraction
11573 * buffer, or if the ending output pointer didn't match the
11574 * previously computed value. */
11575 if (v <= vhex || v - vhex >= VHEX_SIZE ||
11576 /* For double-double the ixmin and ixmax stay at zero,
11577 * which is convenient since the HEXTRACTSIZE is tricky
11578 * for double-double. */
11579 ixmin < 0 || ixmax >= NVSIZE ||
11580 (vend && v != vend)) {
11581 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11582 Perl_croak(aTHX_ "Hexadecimal float: internal error (overflow)");
11588 /* S_format_hexfp(): helper function for Perl_sv_vcatpvfn_flags().
11590 * Processes the %a/%A hexadecimal floating-point format, since the
11591 * built-in snprintf()s which are used for most of the f/p formats, don't
11592 * universally handle %a/%A.
11593 * Populates buf of length bufsize, and returns the length of the created
11595 * The rest of the args have the same meaning as the local vars of the
11596 * same name within Perl_sv_vcatpvfn_flags().
11598 * It assumes the caller has already done STORE_LC_NUMERIC_SET_TO_NEEDED();
11600 * It requires the caller to make buf large enough.
11604 S_format_hexfp(pTHX_ char * const buf, const STRLEN bufsize, const char c,
11605 const NV nv, const vcatpvfn_long_double_t fv,
11606 bool has_precis, STRLEN precis, STRLEN width,
11607 bool alt, char plus, bool left, bool fill)
11609 /* Hexadecimal floating point. */
11611 U8 vhex[VHEX_SIZE];
11612 U8* v = vhex; /* working pointer to vhex */
11613 U8* vend; /* pointer to one beyond last digit of vhex */
11614 U8* vfnz = NULL; /* first non-zero */
11615 U8* vlnz = NULL; /* last non-zero */
11616 U8* v0 = NULL; /* first output */
11617 const bool lower = (c == 'a');
11618 /* At output the values of vhex (up to vend) will
11619 * be mapped through the xdig to get the actual
11620 * human-readable xdigits. */
11621 const char* xdig = PL_hexdigit;
11622 STRLEN zerotail = 0; /* how many extra zeros to append */
11623 int exponent = 0; /* exponent of the floating point input */
11624 bool hexradix = FALSE; /* should we output the radix */
11625 bool subnormal = FALSE; /* IEEE 754 subnormal/denormal */
11626 bool negative = FALSE;
11629 /* XXX: NaN, Inf -- though they are printed as "NaN" and "Inf".
11631 * For example with denormals, (assuming the vanilla
11632 * 64-bit double): the exponent is zero. 1xp-1074 is
11633 * the smallest denormal and the smallest double, it
11634 * could be output also as 0x0.0000000000001p-1022 to
11635 * match its internal structure. */
11637 vend = S_hextract(aTHX_ nv, &exponent, &subnormal, vhex, NULL);
11638 S_hextract(aTHX_ nv, &exponent, &subnormal, vhex, vend);
11640 #if NVSIZE > DOUBLESIZE
11641 # ifdef HEXTRACT_HAS_IMPLICIT_BIT
11642 /* In this case there is an implicit bit,
11643 * and therefore the exponent is shifted by one. */
11645 # elif defined(NV_X86_80_BIT)
11647 /* The subnormals of the x86-80 have a base exponent of -16382,
11648 * (while the physical exponent bits are zero) but the frexp()
11649 * returned the scientific-style floating exponent. We want
11650 * to map the last one as:
11651 * -16831..-16384 -> -16382 (the last normal is 0x1p-16382)
11652 * -16835..-16388 -> -16384
11653 * since we want to keep the first hexdigit
11654 * as one of the [8421]. */
11655 exponent = -4 * ( (exponent + 1) / -4) - 2;
11659 /* TBD: other non-implicit-bit platforms than the x86-80. */
11663 negative = fv < 0 || Perl_signbit(nv);
11674 xdig += 16; /* Use uppercase hex. */
11677 /* Find the first non-zero xdigit. */
11678 for (v = vhex; v < vend; v++) {
11686 /* Find the last non-zero xdigit. */
11687 for (v = vend - 1; v >= vhex; v--) {
11694 #if NVSIZE == DOUBLESIZE
11700 #ifndef NV_X86_80_BIT
11702 /* IEEE 754 subnormals (but not the x86 80-bit):
11703 * we want "normalize" the subnormal,
11704 * so we need to right shift the hex nybbles
11705 * so that the output of the subnormal starts
11706 * from the first true bit. (Another, equally
11707 * valid, policy would be to dump the subnormal
11708 * nybbles as-is, to display the "physical" layout.) */
11711 /* Find the ceil(log2(v[0])) of
11712 * the top non-zero nybble. */
11713 for (i = vfnz[0], n = 0; i > 1; i >>= 1, n++) { }
11716 for (vshr = vlnz; vshr >= vfnz; vshr--) {
11717 vshr[1] |= (vshr[0] & (0xF >> (4 - n))) << (4 - n);
11731 U8* ve = (subnormal ? vlnz + 1 : vend);
11732 SSize_t vn = ve - v0;
11734 if (precis < (Size_t)(vn - 1)) {
11735 bool overflow = FALSE;
11736 if (v0[precis + 1] < 0x8) {
11737 /* Round down, nothing to do. */
11738 } else if (v0[precis + 1] > 0x8) {
11741 overflow = v0[precis] > 0xF;
11743 } else { /* v0[precis] == 0x8 */
11744 /* Half-point: round towards the one
11745 * with the even least-significant digit:
11753 * 78 -> 8 f8 -> 10 */
11754 if ((v0[precis] & 0x1)) {
11757 overflow = v0[precis] > 0xF;
11762 for (v = v0 + precis - 1; v >= v0; v--) {
11764 overflow = *v > 0xF;
11770 if (v == v0 - 1 && overflow) {
11771 /* If the overflow goes all the
11772 * way to the front, we need to
11773 * insert 0x1 in front, and adjust
11775 Move(v0, v0 + 1, vn - 1, char);
11781 /* The new effective "last non zero". */
11782 vlnz = v0 + precis;
11786 subnormal ? precis - vn + 1 :
11787 precis - (vlnz - vhex);
11794 /* If there are non-zero xdigits, the radix
11795 * is output after the first one. */
11806 /* The radix is always output if precis, or if alt. */
11807 if (precis > 0 || alt) {
11812 #ifndef USE_LOCALE_NUMERIC
11815 if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC)) {
11817 const char* r = SvPV(PL_numeric_radix_sv, n);
11818 Copy(r, p, n, char);
11832 if (zerotail > 0) {
11833 while (zerotail--) {
11840 /* sanity checks */
11841 if (elen >= bufsize || width >= bufsize)
11842 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11843 Perl_croak(aTHX_ "Hexadecimal float: internal error (overflow)");
11845 elen += my_snprintf(p, bufsize - elen,
11846 "%c%+d", lower ? 'p' : 'P',
11849 if (elen < width) {
11850 STRLEN gap = (STRLEN)(width - elen);
11852 /* Pad the back with spaces. */
11853 memset(buf + elen, ' ', gap);
11856 /* Insert the zeros after the "0x" and the
11857 * the potential sign, but before the digits,
11858 * otherwise we end up with "0000xH.HHH...",
11859 * when we want "0x000H.HHH..." */
11860 STRLEN nzero = gap;
11861 char* zerox = buf + 2;
11862 STRLEN nmove = elen - 2;
11863 if (negative || plus) {
11867 Move(zerox, zerox + nzero, nmove, char);
11868 memset(zerox, fill ? '0' : ' ', nzero);
11871 /* Move it to the right. */
11872 Move(buf, buf + gap,
11874 /* Pad the front with spaces. */
11875 memset(buf, ' ', gap);
11884 =for apidoc sv_vcatpvfn
11886 =for apidoc sv_vcatpvfn_flags
11888 Processes its arguments like C<vsprintf> and appends the formatted output
11889 to an SV. Uses an array of SVs if the C-style variable argument list is
11890 missing (C<NULL>). Argument reordering (using format specifiers like C<%2$d>
11891 or C<%*2$d>) is supported only when using an array of SVs; using a C-style
11892 C<va_list> argument list with a format string that uses argument reordering
11893 will yield an exception.
11895 When running with taint checks enabled, indicates via
11896 C<maybe_tainted> if results are untrustworthy (often due to the use of
11899 If called as C<sv_vcatpvfn> or flags has the C<SV_GMAGIC> bit set, calls get magic.
11901 It assumes that pat has the same utf8-ness as sv. It's the caller's
11902 responsibility to ensure that this is so.
11904 Usually used via one of its frontends C<sv_vcatpvf> and C<sv_vcatpvf_mg>.
11911 Perl_sv_vcatpvfn_flags(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11912 va_list *const args, SV **const svargs, const Size_t sv_count, bool *const maybe_tainted,
11915 const char *fmtstart; /* character following the current '%' */
11916 const char *q; /* current position within format */
11917 const char *patend;
11920 static const char nullstr[] = "(null)";
11922 bool has_utf8 = DO_UTF8(sv); /* has the result utf8? */
11923 const bool pat_utf8 = has_utf8; /* the pattern is in utf8? */
11924 /* Times 4: a decimal digit takes more than 3 binary digits.
11925 * NV_DIG: mantissa takes than many decimal digits.
11926 * Plus 32: Playing safe. */
11927 char ebuf[IV_DIG * 4 + NV_DIG + 32];
11928 bool no_redundant_warning = FALSE; /* did we use any explicit format parameter index? */
11929 #ifdef USE_LOCALE_NUMERIC
11930 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
11931 bool lc_numeric_set = FALSE; /* called STORE_LC_NUMERIC_SET_TO_NEEDED? */
11934 PERL_ARGS_ASSERT_SV_VCATPVFN_FLAGS;
11935 PERL_UNUSED_ARG(maybe_tainted);
11937 if (flags & SV_GMAGIC)
11940 /* no matter what, this is a string now */
11941 (void)SvPV_force_nomg(sv, origlen);
11943 /* the code that scans for flags etc following a % relies on
11944 * a '\0' being present to avoid falling off the end. Ideally that
11945 * should be fixed */
11946 assert(pat[patlen] == '\0');
11949 /* Special-case "", "%s", "%-p" (SVf - see below) and "%.0f".
11950 * In each case, if there isn't the correct number of args, instead
11951 * fall through to the main code to handle the issuing of any
11955 if (patlen == 0 && (args || sv_count == 0))
11958 if (patlen <= 4 && pat[0] == '%' && (args || sv_count == 1)) {
11961 if (patlen == 2 && pat[1] == 's') {
11963 const char * const s = va_arg(*args, char*);
11964 sv_catpv_nomg(sv, s ? s : nullstr);
11967 /* we want get magic on the source but not the target.
11968 * sv_catsv can't do that, though */
11969 SvGETMAGIC(*svargs);
11970 sv_catsv_nomg(sv, *svargs);
11977 if (patlen == 3 && pat[1] == '-' && pat[2] == 'p') {
11978 SV *asv = MUTABLE_SV(va_arg(*args, void*));
11979 sv_catsv_nomg(sv, asv);
11983 #if !defined(USE_LONG_DOUBLE) && !defined(USE_QUADMATH)
11984 /* special-case "%.0f" */
11985 else if ( patlen == 4
11986 && pat[1] == '.' && pat[2] == '0' && pat[3] == 'f')
11988 const NV nv = SvNV(*svargs);
11989 if (LIKELY(!Perl_isinfnan(nv))) {
11993 if ((p = F0convert(nv, ebuf + sizeof ebuf, &l))) {
11994 sv_catpvn_nomg(sv, p, l);
11999 #endif /* !USE_LONG_DOUBLE */
12003 patend = (char*)pat + patlen;
12004 for (fmtstart = pat; fmtstart < patend; fmtstart = q) {
12005 char intsize = 0; /* size qualifier in "%hi..." etc */
12006 bool alt = FALSE; /* has "%#..." */
12007 bool left = FALSE; /* has "%-..." */
12008 bool fill = FALSE; /* has "%0..." */
12009 char plus = 0; /* has "%+..." */
12010 STRLEN width = 0; /* value of "%NNN..." */
12011 bool has_precis = FALSE; /* has "%.NNN..." */
12012 STRLEN precis = 0; /* value of "%.NNN..." */
12013 int base = 0; /* base to print in, e.g. 8 for %o */
12014 UV uv = 0; /* the value to print of int-ish args */
12016 bool vectorize = FALSE; /* has "%v..." */
12017 bool vec_utf8 = FALSE; /* SvUTF8(vec arg) */
12018 const U8 *vecstr = NULL; /* SvPVX(vec arg) */
12019 STRLEN veclen = 0; /* SvCUR(vec arg) */
12020 const char *dotstr = NULL; /* separator string for %v */
12021 STRLEN dotstrlen; /* length of separator string for %v */
12023 Size_t efix = 0; /* explicit format parameter index */
12024 const Size_t osvix = svix; /* original index in case of bad fmt */
12026 bool is_utf8 = FALSE; /* is this item utf8? */
12027 bool arg_missing = FALSE; /* give "Missing argument" warning */
12028 char esignbuf[4]; /* holds sign prefix, e.g. "-0x" */
12029 STRLEN esignlen = 0; /* length of e.g. "-0x" */
12030 STRLEN zeros = 0; /* how many '0' to prepend */
12032 const char *eptr = NULL; /* the address of the element string */
12033 STRLEN elen = 0; /* the length of the element string */
12035 char c; /* the actual format ('d', s' etc) */
12038 /* echo everything up to the next format specification */
12039 for (q = fmtstart; q < patend && *q != '%'; ++q)
12042 if (q > fmtstart) {
12043 if (has_utf8 && !pat_utf8) {
12044 /* upgrade and copy the bytes of fmtstart..q-1 to utf8 on
12048 STRLEN need = SvCUR(sv) + (q - fmtstart) + 1;
12050 for (p = fmtstart; p < q; p++)
12051 if (!NATIVE_BYTE_IS_INVARIANT(*p))
12056 for (p = fmtstart; p < q; p++)
12057 append_utf8_from_native_byte((U8)*p, (U8**)&dst);
12059 SvCUR_set(sv, need - 1);
12062 S_sv_catpvn_simple(aTHX_ sv, fmtstart, q - fmtstart);
12067 fmtstart = q; /* fmtstart is char following the '%' */
12070 We allow format specification elements in this order:
12071 \d+\$ explicit format parameter index
12073 v|\*(\d+\$)?v vector with optional (optionally specified) arg
12074 0 flag (as above): repeated to allow "v02"
12075 \d+|\*(\d+\$)? width using optional (optionally specified) arg
12076 \.(\d*|\*(\d+\$)?) precision using optional (optionally specified) arg
12078 [%bcdefginopsuxDFOUX] format (mandatory)
12081 if (IS_1_TO_9(*q)) {
12082 width = expect_number(&q);
12085 Perl_croak_nocontext(
12086 "Cannot yet reorder sv_catpvfn() arguments from va_list");
12088 efix = (Size_t)width;
12090 no_redundant_warning = TRUE;
12102 if (plus == '+' && *q == ' ') /* '+' over ' ' */
12129 /* at this point we can expect one of:
12131 * 123 an explicit width
12132 * * width taken from next arg
12133 * *12$ width taken from 12th arg
12136 * But any width specification may be preceded by a v, in one of its
12141 * So an asterisk may be either a width specifier or a vector
12142 * separator arg specifier, and we don't know which initially
12147 STRLEN ix; /* explicit width/vector separator index */
12149 if (IS_1_TO_9(*q)) {
12150 ix = expect_number(&q);
12153 Perl_croak_nocontext(
12154 "Cannot yet reorder sv_catpvfn() arguments from va_list");
12155 no_redundant_warning = TRUE;
12164 /* The asterisk was for *v, *NNN$v: vectorizing, but not
12165 * with the default "." */
12170 vecsv = va_arg(*args, SV*);
12172 ix = ix ? ix - 1 : svix++;
12173 vecsv = ix < sv_count ? svargs[ix]
12174 : (arg_missing = TRUE, &PL_sv_no);
12176 dotstr = SvPV_const(vecsv, dotstrlen);
12177 /* Keep the DO_UTF8 test *after* the SvPV call, else things go
12178 bad with tied or overloaded values that return UTF8. */
12179 if (DO_UTF8(vecsv))
12181 else if (has_utf8) {
12182 vecsv = sv_mortalcopy(vecsv);
12183 sv_utf8_upgrade(vecsv);
12184 dotstr = SvPV_const(vecsv, dotstrlen);
12191 /* the asterisk specified a width */
12196 i = va_arg(*args, int);
12198 ix = ix ? ix - 1 : svix++;
12199 sv = (ix < sv_count) ? svargs[ix]
12200 : (arg_missing = TRUE, (SV*)NULL);
12202 width = S_sprintf_arg_num_val(aTHX_ args, i, sv, &left);
12205 else if (*q == 'v') {
12216 /* explicit width? */
12222 width = expect_number(&q);
12232 STRLEN ix; /* explicit precision index */
12234 if (IS_1_TO_9(*q)) {
12235 ix = expect_number(&q);
12238 Perl_croak_nocontext(
12239 "Cannot yet reorder sv_catpvfn() arguments from va_list");
12240 no_redundant_warning = TRUE;
12253 i = va_arg(*args, int);
12255 ix = ix ? ix - 1 : svix++;
12256 sv = (ix < sv_count) ? svargs[ix]
12257 : (arg_missing = TRUE, (SV*)NULL);
12259 precis = S_sprintf_arg_num_val(aTHX_ args, i, sv, &neg);
12264 /* although it doesn't seem documented, this code has long
12266 * no digits following the '.' is treated like '.0'
12267 * the number may be preceded by any number of zeroes,
12268 * e.g. "%.0001f", which is the same as "%.1f"
12269 * so I've kept that behaviour. DAPM May 2017
12273 precis = IS_1_TO_9(*q) ? expect_number(&q) : 0;
12282 case 'I': /* Ix, I32x, and I64x */
12283 # ifdef USE_64_BIT_INT
12284 if (q[1] == '6' && q[2] == '4') {
12290 if (q[1] == '3' && q[2] == '2') {
12294 # ifdef USE_64_BIT_INT
12300 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
12301 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
12304 # ifdef USE_QUADMATH
12317 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
12318 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
12319 if (*q == 'l') { /* lld, llf */
12328 if (*++q == 'h') { /* hhd, hhu */
12347 c = *q++; /* c now holds the conversion type */
12349 /* '%' doesn't have an arg, so skip arg processing */
12358 if (vectorize && !strchr("BbDdiOouUXx", c))
12361 /* get next arg (individual branches do their own va_arg()
12362 * handling for the args case) */
12365 efix = efix ? efix - 1 : svix++;
12366 argsv = efix < sv_count ? svargs[efix]
12367 : (arg_missing = TRUE, &PL_sv_no);
12377 eptr = va_arg(*args, char*);
12379 elen = strlen(eptr);
12381 eptr = (char *)nullstr;
12382 elen = sizeof nullstr - 1;
12386 eptr = SvPV_const(argsv, elen);
12387 if (DO_UTF8(argsv)) {
12388 STRLEN old_precis = precis;
12389 if (has_precis && precis < elen) {
12390 STRLEN ulen = sv_or_pv_len_utf8(argsv, eptr, elen);
12391 STRLEN p = precis > ulen ? ulen : precis;
12392 precis = sv_or_pv_pos_u2b(argsv, eptr, p, 0);
12393 /* sticks at end */
12395 if (width) { /* fudge width (can't fudge elen) */
12396 if (has_precis && precis < elen)
12397 width += precis - old_precis;
12400 elen - sv_or_pv_len_utf8(argsv,eptr,elen);
12407 if (has_precis && precis < elen)
12419 * "%...p" is normally treated like "%...x", except that the
12420 * number to print is the SV's address (or a pointer address
12421 * for C-ish sprintf).
12423 * However, the C-ish sprintf variant allows a few special
12424 * extensions. These are currently:
12426 * %-p (SVf) Like %s, but gets the string from an SV*
12427 * arg rather than a char* arg.
12428 * (This was previously %_).
12430 * %-<num>p Ditto but like %.<num>s (i.e. num is max width)
12432 * %2p (HEKf) Like %s, but using the key string in a HEK
12434 * %3p (HEKf256) Ditto but like %.256s
12436 * %d%lu%4p (UTF8f) A utf8 string. Consumes 3 args:
12437 * (cBOOL(utf8), len, string_buf).
12438 * It's handled by the "case 'd'" branch
12439 * rather than here.
12441 * %<num>p where num is 1 or > 4: reserved for future
12442 * extensions. Warns, but then is treated as a
12443 * general %p (print hex address) format.
12451 /* not %*p or %*1$p - any width was explicit */
12455 if (left) { /* %-p (SVf), %-NNNp */
12460 argsv = MUTABLE_SV(va_arg(*args, void*));
12461 eptr = SvPV_const(argsv, elen);
12462 if (DO_UTF8(argsv))
12467 else if (width == 2 || width == 3) { /* HEKf, HEKf256 */
12468 HEK * const hek = va_arg(*args, HEK *);
12469 eptr = HEK_KEY(hek);
12470 elen = HEK_LEN(hek);
12481 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
12482 "internal %%<num>p might conflict with future printf extensions");
12486 /* treat as normal %...p */
12488 uv = PTR2UV(args ? va_arg(*args, void*) : argsv);
12493 /* Ignore any size specifiers, since they're not documented as
12494 * being allowed for %c (ideally we should warn on e.g. '%hc').
12495 * Setting a default intsize, along with a positive
12496 * (which signals unsigned) base, causes, for C-ish use, the
12497 * va_arg to be interpreted as as unsigned int, when it's
12498 * actually signed, which will convert -ve values to high +ve
12499 * values. Note that unlike the libc %c, values > 255 will
12500 * convert to high unicode points rather than being truncated
12501 * to 8 bits. For perlish use, it will do SvUV(argsv), which
12502 * will again convert -ve args to high -ve values.
12505 base = 1; /* special value that indicates we're doing a 'c' */
12506 goto get_int_arg_val;
12515 goto get_int_arg_val;
12518 /* probably just a plain %d, but it might be the start of the
12519 * special UTF8f format, which usually looks something like
12520 * "%d%lu%4p" (the lu may vary by platform)
12522 assert((UTF8f)[0] == 'd');
12523 assert((UTF8f)[1] == '%');
12525 if ( args /* UTF8f only valid for C-ish sprintf */
12526 && q == fmtstart + 1 /* plain %d, not %....d */
12527 && patend >= fmtstart + sizeof(UTF8f) - 1 /* long enough */
12529 && strnEQ(q + 1, UTF8f + 2, sizeof(UTF8f) - 3))
12531 /* The argument has already gone through cBOOL, so the cast
12533 is_utf8 = (bool)va_arg(*args, int);
12534 elen = va_arg(*args, UV);
12535 /* if utf8 length is larger than 0x7ffff..., then it might
12536 * have been a signed value that wrapped */
12537 if (elen > ((~(STRLEN)0) >> 1)) {
12538 assert(0); /* in DEBUGGING build we want to crash */
12539 elen = 0; /* otherwise we want to treat this as an empty string */
12541 eptr = va_arg(*args, char *);
12542 q += sizeof(UTF8f) - 2;
12549 goto get_int_arg_val;
12560 goto get_int_arg_val;
12565 goto get_int_arg_val;
12576 goto get_int_arg_val;
12591 esignbuf[esignlen++] = plus;
12594 /* initialise the vector string to iterate over */
12596 vecsv = args ? va_arg(*args, SV*) : argsv;
12598 /* if this is a version object, we need to convert
12599 * back into v-string notation and then let the
12600 * vectorize happen normally
12602 if (sv_isobject(vecsv) && sv_derived_from(vecsv, "version")) {
12603 if ( hv_existss(MUTABLE_HV(SvRV(vecsv)), "alpha") ) {
12604 Perl_ck_warner_d(aTHX_ packWARN(WARN_PRINTF),
12605 "vector argument not supported with alpha versions");
12609 vecstr = (U8*)SvPV_const(vecsv,veclen);
12610 vecsv = sv_newmortal();
12611 scan_vstring((char *)vecstr, (char *)vecstr + veclen,
12615 vecstr = (U8*)SvPV_const(vecsv, veclen);
12616 vec_utf8 = DO_UTF8(vecsv);
12618 /* This is the re-entry point for when we're iterating
12619 * over the individual characters of a vector arg */
12622 goto done_valid_conversion;
12624 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
12634 /* test arg for inf/nan. This can trigger an unwanted
12635 * 'str' overload, so manually force 'num' overload first
12639 if (UNLIKELY(SvAMAGIC(argsv)))
12640 argsv = sv_2num(argsv);
12641 if (UNLIKELY(isinfnansv(argsv)))
12642 goto handle_infnan_argsv;
12646 /* signed int type */
12651 case 'c': iv = (char)va_arg(*args, int); break;
12652 case 'h': iv = (short)va_arg(*args, int); break;
12653 case 'l': iv = va_arg(*args, long); break;
12654 case 'V': iv = va_arg(*args, IV); break;
12655 case 'z': iv = va_arg(*args, SSize_t); break;
12656 #ifdef HAS_PTRDIFF_T
12657 case 't': iv = va_arg(*args, ptrdiff_t); break;
12659 default: iv = va_arg(*args, int); break;
12661 case 'j': iv = va_arg(*args, intmax_t); break;
12665 iv = va_arg(*args, Quad_t); break;
12672 /* assign to tiv then cast to iv to work around
12673 * 2003 GCC cast bug (gnu.org bugzilla #13488) */
12674 IV tiv = SvIV_nomg(argsv);
12676 case 'c': iv = (char)tiv; break;
12677 case 'h': iv = (short)tiv; break;
12678 case 'l': iv = (long)tiv; break;
12680 default: iv = tiv; break;
12683 iv = (Quad_t)tiv; break;
12690 /* now convert iv to uv */
12694 esignbuf[esignlen++] = plus;
12697 uv = (iv == IV_MIN) ? (UV)iv : (UV)(-iv);
12698 esignbuf[esignlen++] = '-';
12702 /* unsigned int type */
12705 case 'c': uv = (unsigned char)va_arg(*args, unsigned);
12707 case 'h': uv = (unsigned short)va_arg(*args, unsigned);
12709 case 'l': uv = va_arg(*args, unsigned long); break;
12710 case 'V': uv = va_arg(*args, UV); break;
12711 case 'z': uv = va_arg(*args, Size_t); break;
12712 #ifdef HAS_PTRDIFF_T
12713 /* will sign extend, but there is no
12714 * uptrdiff_t, so oh well */
12715 case 't': uv = va_arg(*args, ptrdiff_t); break;
12718 case 'j': uv = va_arg(*args, uintmax_t); break;
12720 default: uv = va_arg(*args, unsigned); break;
12723 uv = va_arg(*args, Uquad_t); break;
12730 /* assign to tiv then cast to iv to work around
12731 * 2003 GCC cast bug (gnu.org bugzilla #13488) */
12732 UV tuv = SvUV_nomg(argsv);
12734 case 'c': uv = (unsigned char)tuv; break;
12735 case 'h': uv = (unsigned short)tuv; break;
12736 case 'l': uv = (unsigned long)tuv; break;
12738 default: uv = tuv; break;
12741 uv = (Uquad_t)tuv; break;
12752 char *ptr = ebuf + sizeof ebuf;
12759 const char * const p =
12760 (c == 'X') ? PL_hexdigit + 16 : PL_hexdigit;
12765 } while (uv >>= 4);
12766 if (alt && *ptr != '0') {
12767 esignbuf[esignlen++] = '0';
12768 esignbuf[esignlen++] = c; /* 'x' or 'X' */
12775 *--ptr = '0' + dig;
12776 } while (uv >>= 3);
12777 if (alt && *ptr != '0')
12783 *--ptr = '0' + dig;
12784 } while (uv >>= 1);
12785 if (alt && *ptr != '0') {
12786 esignbuf[esignlen++] = '0';
12787 esignbuf[esignlen++] = c; /* 'b' or 'B' */
12792 /* special-case: base 1 indicates a 'c' format:
12793 * we use the common code for extracting a uv,
12794 * but handle that value differently here than
12795 * all the other int types */
12797 (!UVCHR_IS_INVARIANT(uv) && SvUTF8(sv)))
12800 assert(sizeof(ebuf) >= UTF8_MAXBYTES + 1);
12802 elen = uvchr_to_utf8((U8*)eptr, uv) - (U8*)ebuf;
12807 ebuf[0] = (char)uv;
12812 default: /* it had better be ten or less */
12815 *--ptr = '0' + dig;
12816 } while (uv /= base);
12819 elen = (ebuf + sizeof ebuf) - ptr;
12823 zeros = precis - elen;
12824 else if (precis == 0 && elen == 1 && *eptr == '0'
12825 && !(base == 8 && alt)) /* "%#.0o" prints "0" */
12828 /* a precision nullifies the 0 flag. */
12834 /* FLOATING POINT */
12837 c = 'f'; /* maybe %F isn't supported here */
12839 case 'e': case 'E':
12841 case 'g': case 'G':
12842 case 'a': case 'A':
12845 STRLEN float_need; /* what PL_efloatsize needs to become */
12846 bool hexfp; /* hexadecimal floating point? */
12848 vcatpvfn_long_double_t fv;
12851 /* This is evil, but floating point is even more evil */
12853 /* for SV-style calling, we can only get NV
12854 for C-style calling, we assume %f is double;
12855 for simplicity we allow any of %Lf, %llf, %qf for long double
12859 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12863 /* [perl #20339] - we should accept and ignore %lf rather than die */
12867 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12868 intsize = args ? 0 : 'q';
12872 #if defined(HAS_LONG_DOUBLE)
12885 /* Now we need (long double) if intsize == 'q', else (double). */
12887 /* Note: do not pull NVs off the va_list with va_arg()
12888 * (pull doubles instead) because if you have a build
12889 * with long doubles, you would always be pulling long
12890 * doubles, which would badly break anyone using only
12891 * doubles (i.e. the majority of builds). In other
12892 * words, you cannot mix doubles and long doubles.
12893 * The only case where you can pull off long doubles
12894 * is when the format specifier explicitly asks so with
12896 #ifdef USE_QUADMATH
12897 fv = intsize == 'q' ?
12898 va_arg(*args, NV) : va_arg(*args, double);
12900 #elif LONG_DOUBLESIZE > DOUBLESIZE
12901 if (intsize == 'q') {
12902 fv = va_arg(*args, long double);
12905 nv = va_arg(*args, double);
12906 VCATPVFN_NV_TO_FV(nv, fv);
12909 nv = va_arg(*args, double);
12916 /* we jump here if an int-ish format encountered an
12917 * infinite/Nan argsv. After setting nv/fv, it falls
12918 * into the isinfnan block which follows */
12919 handle_infnan_argsv:
12920 nv = SvNV_nomg(argsv);
12921 VCATPVFN_NV_TO_FV(nv, fv);
12924 if (Perl_isinfnan(nv)) {
12926 Perl_croak(aTHX_ "Cannot printf %" NVgf " with '%c'",
12927 SvNV_nomg(argsv), (int)c);
12929 elen = S_infnan_2pv(nv, ebuf, sizeof(ebuf), plus);
12938 /* special-case "%.0f" */
12942 && !(width || left || plus || alt)
12945 && ((eptr = F0convert(nv, ebuf + sizeof ebuf, &elen)))
12949 /* Determine the buffer size needed for the various
12950 * floating-point formats.
12952 * The basic possibilities are:
12955 * %f 1111111.123456789
12956 * %e 1.111111123e+06
12957 * %a 0x1.0f4471f9bp+20
12959 * %g 1.11111112e+15
12961 * where P is the value of the precision in the format, or 6
12962 * if not specified. Note the two possible output formats of
12963 * %g; in both cases the number of significant digits is <=
12966 * For most of the format types the maximum buffer size needed
12967 * is precision, plus: any leading 1 or 0x1, the radix
12968 * point, and an exponent. The difficult one is %f: for a
12969 * large positive exponent it can have many leading digits,
12970 * which needs to be calculated specially. Also %a is slightly
12971 * different in that in the absence of a specified precision,
12972 * it uses as many digits as necessary to distinguish
12973 * different values.
12975 * First, here are the constant bits. For ease of calculation
12976 * we over-estimate the needed buffer size, for example by
12977 * assuming all formats have an exponent and a leading 0x1.
12979 * Also for production use, add a little extra overhead for
12980 * safety's sake. Under debugging don't, as it means we're
12981 * more likely to quickly spot issues during development.
12984 float_need = 1 /* possible unary minus */
12985 + 4 /* "0x1" plus very unlikely carry */
12986 + 1 /* default radix point '.' */
12987 + 2 /* "e-", "p+" etc */
12988 + 6 /* exponent: up to 16383 (quad fp) */
12990 + 20 /* safety net */
12995 /* determine the radix point len, e.g. length(".") in "1.2" */
12996 #ifdef USE_LOCALE_NUMERIC
12997 /* note that we may either explicitly use PL_numeric_radix_sv
12998 * below, or implicitly, via an snprintf() variant.
12999 * Note also things like ps_AF.utf8 which has
13000 * "\N{ARABIC DECIMAL SEPARATOR} as a radix point */
13001 if (!lc_numeric_set) {
13002 /* only set once and reuse in-locale value on subsequent
13004 * XXX what happens if we die in an eval?
13006 STORE_LC_NUMERIC_SET_TO_NEEDED();
13007 lc_numeric_set = TRUE;
13010 if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC)) {
13011 /* this can't wrap unless PL_numeric_radix_sv is a string
13012 * consuming virtually all the 32-bit or 64-bit address
13015 float_need += (SvCUR(PL_numeric_radix_sv) - 1);
13017 /* floating-point formats only get utf8 if the radix point
13018 * is utf8. All other characters in the string are < 128
13019 * and so can be safely appended to both a non-utf8 and utf8
13021 * Note that this will convert the output to utf8 even if
13022 * the radix point didn't get output.
13024 if (SvUTF8(PL_numeric_radix_sv) && !has_utf8) {
13025 sv_utf8_upgrade(sv);
13033 if (isALPHA_FOLD_EQ(c, 'f')) {
13034 /* Determine how many digits before the radix point
13035 * might be emitted. frexp() (or frexpl) has some
13036 * unspecified behaviour for nan/inf/-inf, so lucky we've
13037 * already handled them above */
13039 int i = PERL_INT_MIN;
13040 (void)Perl_frexp((NV)fv, &i);
13041 if (i == PERL_INT_MIN)
13042 Perl_die(aTHX_ "panic: frexp: %" VCATPVFN_FV_GF, fv);
13045 digits = BIT_DIGITS(i);
13046 /* this can't overflow. 'digits' will only be a few
13047 * thousand even for the largest floating-point types.
13048 * And up until now float_need is just some small
13049 * constants plus radix len, which can't be in
13050 * overflow territory unless the radix SV is consuming
13051 * over 1/2 the address space */
13052 assert(float_need < ((STRLEN)~0) - digits);
13053 float_need += digits;
13056 else if (UNLIKELY(isALPHA_FOLD_EQ(c, 'a'))) {
13059 /* %a in the absence of precision may print as many
13060 * digits as needed to represent the entire mantissa
13062 * This estimate seriously overshoots in most cases,
13063 * but better the undershooting. Firstly, all bytes
13064 * of the NV are not mantissa, some of them are
13065 * exponent. Secondly, for the reasonably common
13066 * long doubles case, the "80-bit extended", two
13067 * or six bytes of the NV are unused. Also, we'll
13068 * still pick up an extra +6 from the default
13069 * precision calculation below. */
13071 #ifdef LONGDOUBLE_DOUBLEDOUBLE
13072 /* For the "double double", we need more.
13073 * Since each double has their own exponent, the
13074 * doubles may float (haha) rather far from each
13075 * other, and the number of required bits is much
13076 * larger, up to total of DOUBLEDOUBLE_MAXBITS bits.
13077 * See the definition of DOUBLEDOUBLE_MAXBITS.
13079 * Need 2 hexdigits for each byte. */
13080 (DOUBLEDOUBLE_MAXBITS/8 + 1) * 2;
13082 NVSIZE * 2; /* 2 hexdigits for each byte */
13084 /* see "this can't overflow" comment above */
13085 assert(float_need < ((STRLEN)~0) - digits);
13086 float_need += digits;
13089 /* special-case "%.<number>g" if it will fit in ebuf */
13091 && precis /* See earlier comment about buggy Gconvert
13092 when digits, aka precis, is 0 */
13094 /* check, in manner not involving wrapping, that it will
13096 && float_need < sizeof(ebuf)
13097 && sizeof(ebuf) - float_need > precis
13098 && !(width || left || plus || alt)
13102 SNPRINTF_G(fv, ebuf, sizeof(ebuf), precis);
13103 elen = strlen(ebuf);
13110 STRLEN pr = has_precis ? precis : 6; /* known default */
13111 /* this probably can't wrap, since precis is limited
13112 * to 1/4 address space size, but better safe than sorry
13114 if (float_need >= ((STRLEN)~0) - pr)
13115 croak_memory_wrap();
13119 if (float_need < width)
13120 float_need = width;
13122 if (PL_efloatsize <= float_need) {
13123 /* PL_efloatbuf should be at least 1 greater than
13124 * float_need to allow a trailing \0 to be returned by
13125 * snprintf(). If we need to grow, overgrow for the
13126 * benefit of future generations */
13127 const STRLEN extra = 0x20;
13128 if (float_need >= ((STRLEN)~0) - extra)
13129 croak_memory_wrap();
13130 float_need += extra;
13131 Safefree(PL_efloatbuf);
13132 PL_efloatsize = float_need;
13133 Newx(PL_efloatbuf, PL_efloatsize, char);
13134 PL_efloatbuf[0] = '\0';
13137 if (UNLIKELY(hexfp)) {
13138 elen = S_format_hexfp(aTHX_ PL_efloatbuf, PL_efloatsize, c,
13139 nv, fv, has_precis, precis, width,
13140 alt, plus, left, fill);
13143 char *ptr = ebuf + sizeof ebuf;
13146 #if defined(USE_QUADMATH)
13147 if (intsize == 'q') {
13151 /* FIXME: what to do if HAS_LONG_DOUBLE but not PERL_PRIfldbl? */
13152 #elif defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
13153 /* Note that this is HAS_LONG_DOUBLE and PERL_PRIfldbl,
13154 * not USE_LONG_DOUBLE and NVff. In other words,
13155 * this needs to work without USE_LONG_DOUBLE. */
13156 if (intsize == 'q') {
13157 /* Copy the one or more characters in a long double
13158 * format before the 'base' ([efgEFG]) character to
13159 * the format string. */
13160 static char const ldblf[] = PERL_PRIfldbl;
13161 char const *p = ldblf + sizeof(ldblf) - 3;
13162 while (p >= ldblf) { *--ptr = *p--; }
13167 do { *--ptr = '0' + (base % 10); } while (base /= 10);
13172 do { *--ptr = '0' + (base % 10); } while (base /= 10);
13184 /* No taint. Otherwise we are in the strange situation
13185 * where printf() taints but print($float) doesn't.
13188 /* hopefully the above makes ptr a very constrained format
13189 * that is safe to use, even though it's not literal */
13190 GCC_DIAG_IGNORE(-Wformat-nonliteral);
13191 #ifdef USE_QUADMATH
13193 const char* qfmt = quadmath_format_single(ptr);
13195 Perl_croak_nocontext("panic: quadmath invalid format \"%s\"", ptr);
13196 elen = quadmath_snprintf(PL_efloatbuf, PL_efloatsize,
13198 if ((IV)elen == -1) {
13201 Perl_croak_nocontext("panic: quadmath_snprintf failed, format \"%s\"", qfmt);
13206 #elif defined(HAS_LONG_DOUBLE)
13207 elen = ((intsize == 'q')
13208 ? my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv)
13209 : my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, (double)fv));
13211 elen = my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv);
13216 eptr = PL_efloatbuf;
13220 /* Since floating-point formats do their own formatting and
13221 * padding, we skip the main block of code at the end of this
13222 * loop which handles appending eptr to sv, and do our own
13223 * stripped-down version */
13228 assert(elen >= width);
13230 S_sv_catpvn_simple(aTHX_ sv, eptr, elen);
13232 goto done_valid_conversion;
13240 /* XXX ideally we should warn if any flags etc have been
13241 * set, e.g. "%-4.5n" */
13242 /* XXX if sv was originally non-utf8 with a char in the
13243 * range 0x80-0xff, then if it got upgraded, we should
13244 * calculate char len rather than byte len here */
13245 len = SvCUR(sv) - origlen;
13247 int i = (len > PERL_INT_MAX) ? PERL_INT_MAX : (int)len;
13250 case 'c': *(va_arg(*args, char*)) = i; break;
13251 case 'h': *(va_arg(*args, short*)) = i; break;
13252 default: *(va_arg(*args, int*)) = i; break;
13253 case 'l': *(va_arg(*args, long*)) = i; break;
13254 case 'V': *(va_arg(*args, IV*)) = i; break;
13255 case 'z': *(va_arg(*args, SSize_t*)) = i; break;
13256 #ifdef HAS_PTRDIFF_T
13257 case 't': *(va_arg(*args, ptrdiff_t*)) = i; break;
13260 case 'j': *(va_arg(*args, intmax_t*)) = i; break;
13264 *(va_arg(*args, Quad_t*)) = i; break;
13272 Perl_croak_nocontext(
13273 "Missing argument for %%n in %s",
13274 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
13275 sv_setuv_mg(argsv, has_utf8 ? (UV)sv_len_utf8(sv) : (UV)len);
13277 goto done_valid_conversion;
13285 && (PL_op->op_type == OP_PRTF || PL_op->op_type == OP_SPRINTF)
13286 && ckWARN(WARN_PRINTF))
13288 SV * const msg = sv_newmortal();
13289 Perl_sv_setpvf(aTHX_ msg, "Invalid conversion in %sprintf: ",
13290 (PL_op->op_type == OP_PRTF) ? "" : "s");
13291 if (fmtstart < patend) {
13292 const char * const fmtend = q < patend ? q : patend;
13294 sv_catpvs(msg, "\"%");
13295 for (f = fmtstart; f < fmtend; f++) {
13297 sv_catpvn_nomg(msg, f, 1);
13299 Perl_sv_catpvf(aTHX_ msg,
13300 "\\%03" UVof, (UV)*f & 0xFF);
13303 sv_catpvs(msg, "\"");
13305 sv_catpvs(msg, "end of string");
13307 Perl_warner(aTHX_ packWARN(WARN_PRINTF), "%" SVf, SVfARG(msg)); /* yes, this is reentrant */
13310 /* mangled format: output the '%', then continue from the
13311 * character following that */
13312 sv_catpvn_nomg(sv, fmtstart-1, 1);
13315 /* Any "redundant arg" warning from now onwards will probably
13316 * just be misleading, so don't bother. */
13317 no_redundant_warning = TRUE;
13318 continue; /* not "break" */
13321 if (is_utf8 != has_utf8) {
13324 sv_utf8_upgrade(sv);
13327 const STRLEN old_elen = elen;
13328 SV * const nsv = newSVpvn_flags(eptr, elen, SVs_TEMP);
13329 sv_utf8_upgrade(nsv);
13330 eptr = SvPVX_const(nsv);
13333 if (width) { /* fudge width (can't fudge elen) */
13334 width += elen - old_elen;
13341 /* append esignbuf, filler, zeros, eptr and dotstr to sv */
13344 STRLEN need, have, gap;
13348 /* signed value that's wrapped? */
13349 assert(elen <= ((~(STRLEN)0) >> 1));
13351 /* if zeros is non-zero, then it represents filler between
13352 * elen and precis. So adding elen and zeros together will
13353 * always be <= precis, and the addition can never wrap */
13354 assert(!zeros || (precis > elen && precis - elen == zeros));
13355 have = elen + zeros;
13357 if (have >= (((STRLEN)~0) - esignlen))
13358 croak_memory_wrap();
13361 need = (have > width ? have : width);
13364 if (need >= (((STRLEN)~0) - (SvCUR(sv) + 1)))
13365 croak_memory_wrap();
13366 need += (SvCUR(sv) + 1);
13373 for (i = 0; i < esignlen; i++)
13374 *s++ = esignbuf[i];
13375 for (i = zeros; i; i--)
13377 Copy(eptr, s, elen, char);
13379 for (i = gap; i; i--)
13384 for (i = 0; i < esignlen; i++)
13385 *s++ = esignbuf[i];
13390 for (i = gap; i; i--)
13392 for (i = 0; i < esignlen; i++)
13393 *s++ = esignbuf[i];
13396 for (i = zeros; i; i--)
13398 Copy(eptr, s, elen, char);
13403 SvCUR_set(sv, s - SvPVX_const(sv));
13411 if (vectorize && veclen) {
13412 /* we append the vector separator separately since %v isn't
13413 * very common: don't slow down the general case by adding
13414 * dotstrlen to need etc */
13415 sv_catpvn_nomg(sv, dotstr, dotstrlen);
13417 goto vector; /* do next iteration */
13420 done_valid_conversion:
13423 S_warn_vcatpvfn_missing_argument(aTHX);
13426 /* Now that we've consumed all our printf format arguments (svix)
13427 * do we have things left on the stack that we didn't use?
13429 if (!no_redundant_warning && sv_count >= svix + 1 && ckWARN(WARN_REDUNDANT)) {
13430 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
13431 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
13436 RESTORE_LC_NUMERIC(); /* Done outside loop, so don't have to save/restore
13440 /* =========================================================================
13442 =head1 Cloning an interpreter
13446 All the macros and functions in this section are for the private use of
13447 the main function, perl_clone().
13449 The foo_dup() functions make an exact copy of an existing foo thingy.
13450 During the course of a cloning, a hash table is used to map old addresses
13451 to new addresses. The table is created and manipulated with the
13452 ptr_table_* functions.
13454 * =========================================================================*/
13457 #if defined(USE_ITHREADS)
13459 /* XXX Remove this so it doesn't have to go thru the macro and return for nothing */
13460 #ifndef GpREFCNT_inc
13461 # define GpREFCNT_inc(gp) ((gp) ? (++(gp)->gp_refcnt, (gp)) : (GP*)NULL)
13465 /* Certain cases in Perl_ss_dup have been merged, by relying on the fact
13466 that currently av_dup, gv_dup and hv_dup are the same as sv_dup.
13467 If this changes, please unmerge ss_dup.
13468 Likewise, sv_dup_inc_multiple() relies on this fact. */
13469 #define sv_dup_inc_NN(s,t) SvREFCNT_inc_NN(sv_dup_inc(s,t))
13470 #define av_dup(s,t) MUTABLE_AV(sv_dup((const SV *)s,t))
13471 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
13472 #define hv_dup(s,t) MUTABLE_HV(sv_dup((const SV *)s,t))
13473 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
13474 #define cv_dup(s,t) MUTABLE_CV(sv_dup((const SV *)s,t))
13475 #define cv_dup_inc(s,t) MUTABLE_CV(sv_dup_inc((const SV *)s,t))
13476 #define io_dup(s,t) MUTABLE_IO(sv_dup((const SV *)s,t))
13477 #define io_dup_inc(s,t) MUTABLE_IO(sv_dup_inc((const SV *)s,t))
13478 #define gv_dup(s,t) MUTABLE_GV(sv_dup((const SV *)s,t))
13479 #define gv_dup_inc(s,t) MUTABLE_GV(sv_dup_inc((const SV *)s,t))
13480 #define SAVEPV(p) ((p) ? savepv(p) : NULL)
13481 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
13483 /* clone a parser */
13486 Perl_parser_dup(pTHX_ const yy_parser *const proto, CLONE_PARAMS *const param)
13490 PERL_ARGS_ASSERT_PARSER_DUP;
13495 /* look for it in the table first */
13496 parser = (yy_parser *)ptr_table_fetch(PL_ptr_table, proto);
13500 /* create anew and remember what it is */
13501 Newxz(parser, 1, yy_parser);
13502 ptr_table_store(PL_ptr_table, proto, parser);
13504 /* XXX these not yet duped */
13505 parser->old_parser = NULL;
13506 parser->stack = NULL;
13508 parser->stack_max1 = 0;
13509 /* XXX parser->stack->state = 0; */
13511 /* XXX eventually, just Copy() most of the parser struct ? */
13513 parser->lex_brackets = proto->lex_brackets;
13514 parser->lex_casemods = proto->lex_casemods;
13515 parser->lex_brackstack = savepvn(proto->lex_brackstack,
13516 (proto->lex_brackets < 120 ? 120 : proto->lex_brackets));
13517 parser->lex_casestack = savepvn(proto->lex_casestack,
13518 (proto->lex_casemods < 12 ? 12 : proto->lex_casemods));
13519 parser->lex_defer = proto->lex_defer;
13520 parser->lex_dojoin = proto->lex_dojoin;
13521 parser->lex_formbrack = proto->lex_formbrack;
13522 parser->lex_inpat = proto->lex_inpat;
13523 parser->lex_inwhat = proto->lex_inwhat;
13524 parser->lex_op = proto->lex_op;
13525 parser->lex_repl = sv_dup_inc(proto->lex_repl, param);
13526 parser->lex_starts = proto->lex_starts;
13527 parser->lex_stuff = sv_dup_inc(proto->lex_stuff, param);
13528 parser->multi_close = proto->multi_close;
13529 parser->multi_open = proto->multi_open;
13530 parser->multi_start = proto->multi_start;
13531 parser->multi_end = proto->multi_end;
13532 parser->preambled = proto->preambled;
13533 parser->lex_super_state = proto->lex_super_state;
13534 parser->lex_sub_inwhat = proto->lex_sub_inwhat;
13535 parser->lex_sub_op = proto->lex_sub_op;
13536 parser->lex_sub_repl= sv_dup_inc(proto->lex_sub_repl, param);
13537 parser->linestr = sv_dup_inc(proto->linestr, param);
13538 parser->expect = proto->expect;
13539 parser->copline = proto->copline;
13540 parser->last_lop_op = proto->last_lop_op;
13541 parser->lex_state = proto->lex_state;
13542 parser->rsfp = fp_dup(proto->rsfp, '<', param);
13543 /* rsfp_filters entries have fake IoDIRP() */
13544 parser->rsfp_filters= av_dup_inc(proto->rsfp_filters, param);
13545 parser->in_my = proto->in_my;
13546 parser->in_my_stash = hv_dup(proto->in_my_stash, param);
13547 parser->error_count = proto->error_count;
13548 parser->sig_elems = proto->sig_elems;
13549 parser->sig_optelems= proto->sig_optelems;
13550 parser->sig_slurpy = proto->sig_slurpy;
13551 parser->recheck_utf8_validity = proto->recheck_utf8_validity;
13552 parser->linestr = sv_dup_inc(proto->linestr, param);
13555 char * const ols = SvPVX(proto->linestr);
13556 char * const ls = SvPVX(parser->linestr);
13558 parser->bufptr = ls + (proto->bufptr >= ols ?
13559 proto->bufptr - ols : 0);
13560 parser->oldbufptr = ls + (proto->oldbufptr >= ols ?
13561 proto->oldbufptr - ols : 0);
13562 parser->oldoldbufptr= ls + (proto->oldoldbufptr >= ols ?
13563 proto->oldoldbufptr - ols : 0);
13564 parser->linestart = ls + (proto->linestart >= ols ?
13565 proto->linestart - ols : 0);
13566 parser->last_uni = ls + (proto->last_uni >= ols ?
13567 proto->last_uni - ols : 0);
13568 parser->last_lop = ls + (proto->last_lop >= ols ?
13569 proto->last_lop - ols : 0);
13571 parser->bufend = ls + SvCUR(parser->linestr);
13574 Copy(proto->tokenbuf, parser->tokenbuf, 256, char);
13577 Copy(proto->nextval, parser->nextval, 5, YYSTYPE);
13578 Copy(proto->nexttype, parser->nexttype, 5, I32);
13579 parser->nexttoke = proto->nexttoke;
13581 /* XXX should clone saved_curcop here, but we aren't passed
13582 * proto_perl; so do it in perl_clone_using instead */
13588 /* duplicate a file handle */
13591 Perl_fp_dup(pTHX_ PerlIO *const fp, const char type, CLONE_PARAMS *const param)
13595 PERL_ARGS_ASSERT_FP_DUP;
13596 PERL_UNUSED_ARG(type);
13599 return (PerlIO*)NULL;
13601 /* look for it in the table first */
13602 ret = (PerlIO*)ptr_table_fetch(PL_ptr_table, fp);
13606 /* create anew and remember what it is */
13607 #ifdef __amigaos4__
13608 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE|PERLIO_DUP_FD);
13610 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE);
13612 ptr_table_store(PL_ptr_table, fp, ret);
13616 /* duplicate a directory handle */
13619 Perl_dirp_dup(pTHX_ DIR *const dp, CLONE_PARAMS *const param)
13623 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
13625 const Direntry_t *dirent;
13626 char smallbuf[256]; /* XXX MAXPATHLEN, surely? */
13632 PERL_UNUSED_CONTEXT;
13633 PERL_ARGS_ASSERT_DIRP_DUP;
13638 /* look for it in the table first */
13639 ret = (DIR*)ptr_table_fetch(PL_ptr_table, dp);
13643 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
13645 PERL_UNUSED_ARG(param);
13649 /* open the current directory (so we can switch back) */
13650 if (!(pwd = PerlDir_open("."))) return (DIR *)NULL;
13652 /* chdir to our dir handle and open the present working directory */
13653 if (fchdir(my_dirfd(dp)) < 0 || !(ret = PerlDir_open("."))) {
13654 PerlDir_close(pwd);
13655 return (DIR *)NULL;
13657 /* Now we should have two dir handles pointing to the same dir. */
13659 /* Be nice to the calling code and chdir back to where we were. */
13660 /* XXX If this fails, then what? */
13661 PERL_UNUSED_RESULT(fchdir(my_dirfd(pwd)));
13663 /* We have no need of the pwd handle any more. */
13664 PerlDir_close(pwd);
13667 # define d_namlen(d) (d)->d_namlen
13669 # define d_namlen(d) strlen((d)->d_name)
13671 /* Iterate once through dp, to get the file name at the current posi-
13672 tion. Then step back. */
13673 pos = PerlDir_tell(dp);
13674 if ((dirent = PerlDir_read(dp))) {
13675 len = d_namlen(dirent);
13676 if (len > sizeof(dirent->d_name) && sizeof(dirent->d_name) > PTRSIZE) {
13677 /* If the len is somehow magically longer than the
13678 * maximum length of the directory entry, even though
13679 * we could fit it in a buffer, we could not copy it
13680 * from the dirent. Bail out. */
13681 PerlDir_close(ret);
13684 if (len <= sizeof smallbuf) name = smallbuf;
13685 else Newx(name, len, char);
13686 Move(dirent->d_name, name, len, char);
13688 PerlDir_seek(dp, pos);
13690 /* Iterate through the new dir handle, till we find a file with the
13692 if (!dirent) /* just before the end */
13694 pos = PerlDir_tell(ret);
13695 if (PerlDir_read(ret)) continue; /* not there yet */
13696 PerlDir_seek(ret, pos); /* step back */
13700 const long pos0 = PerlDir_tell(ret);
13702 pos = PerlDir_tell(ret);
13703 if ((dirent = PerlDir_read(ret))) {
13704 if (len == (STRLEN)d_namlen(dirent)
13705 && memEQ(name, dirent->d_name, len)) {
13707 PerlDir_seek(ret, pos); /* step back */
13710 /* else we are not there yet; keep iterating */
13712 else { /* This is not meant to happen. The best we can do is
13713 reset the iterator to the beginning. */
13714 PerlDir_seek(ret, pos0);
13721 if (name && name != smallbuf)
13726 ret = win32_dirp_dup(dp, param);
13729 /* pop it in the pointer table */
13731 ptr_table_store(PL_ptr_table, dp, ret);
13736 /* duplicate a typeglob */
13739 Perl_gp_dup(pTHX_ GP *const gp, CLONE_PARAMS *const param)
13743 PERL_ARGS_ASSERT_GP_DUP;
13747 /* look for it in the table first */
13748 ret = (GP*)ptr_table_fetch(PL_ptr_table, gp);
13752 /* create anew and remember what it is */
13754 ptr_table_store(PL_ptr_table, gp, ret);
13757 /* ret->gp_refcnt must be 0 before any other dups are called. We're relying
13758 on Newxz() to do this for us. */
13759 ret->gp_sv = sv_dup_inc(gp->gp_sv, param);
13760 ret->gp_io = io_dup_inc(gp->gp_io, param);
13761 ret->gp_form = cv_dup_inc(gp->gp_form, param);
13762 ret->gp_av = av_dup_inc(gp->gp_av, param);
13763 ret->gp_hv = hv_dup_inc(gp->gp_hv, param);
13764 ret->gp_egv = gv_dup(gp->gp_egv, param);/* GvEGV is not refcounted */
13765 ret->gp_cv = cv_dup_inc(gp->gp_cv, param);
13766 ret->gp_cvgen = gp->gp_cvgen;
13767 ret->gp_line = gp->gp_line;
13768 ret->gp_file_hek = hek_dup(gp->gp_file_hek, param);
13772 /* duplicate a chain of magic */
13775 Perl_mg_dup(pTHX_ MAGIC *mg, CLONE_PARAMS *const param)
13777 MAGIC *mgret = NULL;
13778 MAGIC **mgprev_p = &mgret;
13780 PERL_ARGS_ASSERT_MG_DUP;
13782 for (; mg; mg = mg->mg_moremagic) {
13785 if ((param->flags & CLONEf_JOIN_IN)
13786 && mg->mg_type == PERL_MAGIC_backref)
13787 /* when joining, we let the individual SVs add themselves to
13788 * backref as needed. */
13791 Newx(nmg, 1, MAGIC);
13793 mgprev_p = &(nmg->mg_moremagic);
13795 /* There was a comment "XXX copy dynamic vtable?" but as we don't have
13796 dynamic vtables, I'm not sure why Sarathy wrote it. The comment dates
13797 from the original commit adding Perl_mg_dup() - revision 4538.
13798 Similarly there is the annotation "XXX random ptr?" next to the
13799 assignment to nmg->mg_ptr. */
13802 /* FIXME for plugins
13803 if (nmg->mg_type == PERL_MAGIC_qr) {
13804 nmg->mg_obj = MUTABLE_SV(CALLREGDUPE((REGEXP*)nmg->mg_obj, param));
13808 nmg->mg_obj = (nmg->mg_flags & MGf_REFCOUNTED)
13809 ? nmg->mg_type == PERL_MAGIC_backref
13810 /* The backref AV has its reference
13811 * count deliberately bumped by 1 */
13812 ? SvREFCNT_inc(av_dup_inc((const AV *)
13813 nmg->mg_obj, param))
13814 : sv_dup_inc(nmg->mg_obj, param)
13815 : (nmg->mg_type == PERL_MAGIC_regdatum ||
13816 nmg->mg_type == PERL_MAGIC_regdata)
13818 : sv_dup(nmg->mg_obj, param);
13820 if (nmg->mg_ptr && nmg->mg_type != PERL_MAGIC_regex_global) {
13821 if (nmg->mg_len > 0) {
13822 nmg->mg_ptr = SAVEPVN(nmg->mg_ptr, nmg->mg_len);
13823 if (nmg->mg_type == PERL_MAGIC_overload_table &&
13824 AMT_AMAGIC((AMT*)nmg->mg_ptr))
13826 AMT * const namtp = (AMT*)nmg->mg_ptr;
13827 sv_dup_inc_multiple((SV**)(namtp->table),
13828 (SV**)(namtp->table), NofAMmeth, param);
13831 else if (nmg->mg_len == HEf_SVKEY)
13832 nmg->mg_ptr = (char*)sv_dup_inc((const SV *)nmg->mg_ptr, param);
13834 if ((nmg->mg_flags & MGf_DUP) && nmg->mg_virtual && nmg->mg_virtual->svt_dup) {
13835 nmg->mg_virtual->svt_dup(aTHX_ nmg, param);
13841 #endif /* USE_ITHREADS */
13843 struct ptr_tbl_arena {
13844 struct ptr_tbl_arena *next;
13845 struct ptr_tbl_ent array[1023/3]; /* as ptr_tbl_ent has 3 pointers. */
13848 /* create a new pointer-mapping table */
13851 Perl_ptr_table_new(pTHX)
13854 PERL_UNUSED_CONTEXT;
13856 Newx(tbl, 1, PTR_TBL_t);
13857 tbl->tbl_max = 511;
13858 tbl->tbl_items = 0;
13859 tbl->tbl_arena = NULL;
13860 tbl->tbl_arena_next = NULL;
13861 tbl->tbl_arena_end = NULL;
13862 Newxz(tbl->tbl_ary, tbl->tbl_max + 1, PTR_TBL_ENT_t*);
13866 #define PTR_TABLE_HASH(ptr) \
13867 ((PTR2UV(ptr) >> 3) ^ (PTR2UV(ptr) >> (3 + 7)) ^ (PTR2UV(ptr) >> (3 + 17)))
13869 /* map an existing pointer using a table */
13871 STATIC PTR_TBL_ENT_t *
13872 S_ptr_table_find(PTR_TBL_t *const tbl, const void *const sv)
13874 PTR_TBL_ENT_t *tblent;
13875 const UV hash = PTR_TABLE_HASH(sv);
13877 PERL_ARGS_ASSERT_PTR_TABLE_FIND;
13879 tblent = tbl->tbl_ary[hash & tbl->tbl_max];
13880 for (; tblent; tblent = tblent->next) {
13881 if (tblent->oldval == sv)
13888 Perl_ptr_table_fetch(pTHX_ PTR_TBL_t *const tbl, const void *const sv)
13890 PTR_TBL_ENT_t const *const tblent = ptr_table_find(tbl, sv);
13892 PERL_ARGS_ASSERT_PTR_TABLE_FETCH;
13893 PERL_UNUSED_CONTEXT;
13895 return tblent ? tblent->newval : NULL;
13898 /* add a new entry to a pointer-mapping table 'tbl'. In hash terms, 'oldsv' is
13899 * the key; 'newsv' is the value. The names "old" and "new" are specific to
13900 * the core's typical use of ptr_tables in thread cloning. */
13903 Perl_ptr_table_store(pTHX_ PTR_TBL_t *const tbl, const void *const oldsv, void *const newsv)
13905 PTR_TBL_ENT_t *tblent = ptr_table_find(tbl, oldsv);
13907 PERL_ARGS_ASSERT_PTR_TABLE_STORE;
13908 PERL_UNUSED_CONTEXT;
13911 tblent->newval = newsv;
13913 const UV entry = PTR_TABLE_HASH(oldsv) & tbl->tbl_max;
13915 if (tbl->tbl_arena_next == tbl->tbl_arena_end) {
13916 struct ptr_tbl_arena *new_arena;
13918 Newx(new_arena, 1, struct ptr_tbl_arena);
13919 new_arena->next = tbl->tbl_arena;
13920 tbl->tbl_arena = new_arena;
13921 tbl->tbl_arena_next = new_arena->array;
13922 tbl->tbl_arena_end = C_ARRAY_END(new_arena->array);
13925 tblent = tbl->tbl_arena_next++;
13927 tblent->oldval = oldsv;
13928 tblent->newval = newsv;
13929 tblent->next = tbl->tbl_ary[entry];
13930 tbl->tbl_ary[entry] = tblent;
13932 if (tblent->next && tbl->tbl_items > tbl->tbl_max)
13933 ptr_table_split(tbl);
13937 /* double the hash bucket size of an existing ptr table */
13940 Perl_ptr_table_split(pTHX_ PTR_TBL_t *const tbl)
13942 PTR_TBL_ENT_t **ary = tbl->tbl_ary;
13943 const UV oldsize = tbl->tbl_max + 1;
13944 UV newsize = oldsize * 2;
13947 PERL_ARGS_ASSERT_PTR_TABLE_SPLIT;
13948 PERL_UNUSED_CONTEXT;
13950 Renew(ary, newsize, PTR_TBL_ENT_t*);
13951 Zero(&ary[oldsize], newsize-oldsize, PTR_TBL_ENT_t*);
13952 tbl->tbl_max = --newsize;
13953 tbl->tbl_ary = ary;
13954 for (i=0; i < oldsize; i++, ary++) {
13955 PTR_TBL_ENT_t **entp = ary;
13956 PTR_TBL_ENT_t *ent = *ary;
13957 PTR_TBL_ENT_t **curentp;
13960 curentp = ary + oldsize;
13962 if ((newsize & PTR_TABLE_HASH(ent->oldval)) != i) {
13964 ent->next = *curentp;
13974 /* remove all the entries from a ptr table */
13975 /* Deprecated - will be removed post 5.14 */
13978 Perl_ptr_table_clear(pTHX_ PTR_TBL_t *const tbl)
13980 PERL_UNUSED_CONTEXT;
13981 if (tbl && tbl->tbl_items) {
13982 struct ptr_tbl_arena *arena = tbl->tbl_arena;
13984 Zero(tbl->tbl_ary, tbl->tbl_max + 1, struct ptr_tbl_ent *);
13987 struct ptr_tbl_arena *next = arena->next;
13993 tbl->tbl_items = 0;
13994 tbl->tbl_arena = NULL;
13995 tbl->tbl_arena_next = NULL;
13996 tbl->tbl_arena_end = NULL;
14000 /* clear and free a ptr table */
14003 Perl_ptr_table_free(pTHX_ PTR_TBL_t *const tbl)
14005 struct ptr_tbl_arena *arena;
14007 PERL_UNUSED_CONTEXT;
14013 arena = tbl->tbl_arena;
14016 struct ptr_tbl_arena *next = arena->next;
14022 Safefree(tbl->tbl_ary);
14026 #if defined(USE_ITHREADS)
14029 Perl_rvpv_dup(pTHX_ SV *const dstr, const SV *const sstr, CLONE_PARAMS *const param)
14031 PERL_ARGS_ASSERT_RVPV_DUP;
14033 assert(!isREGEXP(sstr));
14035 if (SvWEAKREF(sstr)) {
14036 SvRV_set(dstr, sv_dup(SvRV_const(sstr), param));
14037 if (param->flags & CLONEf_JOIN_IN) {
14038 /* if joining, we add any back references individually rather
14039 * than copying the whole backref array */
14040 Perl_sv_add_backref(aTHX_ SvRV(dstr), dstr);
14044 SvRV_set(dstr, sv_dup_inc(SvRV_const(sstr), param));
14046 else if (SvPVX_const(sstr)) {
14047 /* Has something there */
14049 /* Normal PV - clone whole allocated space */
14050 SvPV_set(dstr, SAVEPVN(SvPVX_const(sstr), SvLEN(sstr)-1));
14051 /* sstr may not be that normal, but actually copy on write.
14052 But we are a true, independent SV, so: */
14056 /* Special case - not normally malloced for some reason */
14057 if (isGV_with_GP(sstr)) {
14058 /* Don't need to do anything here. */
14060 else if ((SvIsCOW(sstr))) {
14061 /* A "shared" PV - clone it as "shared" PV */
14063 HEK_KEY(hek_dup(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)),
14067 /* Some other special case - random pointer */
14068 SvPV_set(dstr, (char *) SvPVX_const(sstr));
14073 /* Copy the NULL */
14074 SvPV_set(dstr, NULL);
14078 /* duplicate a list of SVs. source and dest may point to the same memory. */
14080 S_sv_dup_inc_multiple(pTHX_ SV *const *source, SV **dest,
14081 SSize_t items, CLONE_PARAMS *const param)
14083 PERL_ARGS_ASSERT_SV_DUP_INC_MULTIPLE;
14085 while (items-- > 0) {
14086 *dest++ = sv_dup_inc(*source++, param);
14092 /* duplicate an SV of any type (including AV, HV etc) */
14095 S_sv_dup_common(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14100 PERL_ARGS_ASSERT_SV_DUP_COMMON;
14102 if (SvTYPE(sstr) == (svtype)SVTYPEMASK) {
14103 #ifdef DEBUG_LEAKING_SCALARS_ABORT
14108 /* look for it in the table first */
14109 dstr = MUTABLE_SV(ptr_table_fetch(PL_ptr_table, sstr));
14113 if(param->flags & CLONEf_JOIN_IN) {
14114 /** We are joining here so we don't want do clone
14115 something that is bad **/
14116 if (SvTYPE(sstr) == SVt_PVHV) {
14117 const HEK * const hvname = HvNAME_HEK(sstr);
14119 /** don't clone stashes if they already exist **/
14120 dstr = MUTABLE_SV(gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
14121 HEK_UTF8(hvname) ? SVf_UTF8 : 0));
14122 ptr_table_store(PL_ptr_table, sstr, dstr);
14126 else if (SvTYPE(sstr) == SVt_PVGV && !SvFAKE(sstr)) {
14127 HV *stash = GvSTASH(sstr);
14128 const HEK * hvname;
14129 if (stash && (hvname = HvNAME_HEK(stash))) {
14130 /** don't clone GVs if they already exist **/
14132 stash = gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
14133 HEK_UTF8(hvname) ? SVf_UTF8 : 0);
14135 stash, GvNAME(sstr),
14141 if (svp && *svp && SvTYPE(*svp) == SVt_PVGV) {
14142 ptr_table_store(PL_ptr_table, sstr, *svp);
14149 /* create anew and remember what it is */
14152 #ifdef DEBUG_LEAKING_SCALARS
14153 dstr->sv_debug_optype = sstr->sv_debug_optype;
14154 dstr->sv_debug_line = sstr->sv_debug_line;
14155 dstr->sv_debug_inpad = sstr->sv_debug_inpad;
14156 dstr->sv_debug_parent = (SV*)sstr;
14157 FREE_SV_DEBUG_FILE(dstr);
14158 dstr->sv_debug_file = savesharedpv(sstr->sv_debug_file);
14161 ptr_table_store(PL_ptr_table, sstr, dstr);
14164 SvFLAGS(dstr) = SvFLAGS(sstr);
14165 SvFLAGS(dstr) &= ~SVf_OOK; /* don't propagate OOK hack */
14166 SvREFCNT(dstr) = 0; /* must be before any other dups! */
14169 if (SvANY(sstr) && PL_watch_pvx && SvPVX_const(sstr) == PL_watch_pvx)
14170 PerlIO_printf(Perl_debug_log, "watch at %p hit, found string \"%s\"\n",
14171 (void*)PL_watch_pvx, SvPVX_const(sstr));
14174 /* don't clone objects whose class has asked us not to */
14176 && ! (SvFLAGS(SvSTASH(sstr)) & SVphv_CLONEABLE))
14182 switch (SvTYPE(sstr)) {
14184 SvANY(dstr) = NULL;
14187 SET_SVANY_FOR_BODYLESS_IV(dstr);
14189 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
14191 SvIV_set(dstr, SvIVX(sstr));
14195 #if NVSIZE <= IVSIZE
14196 SET_SVANY_FOR_BODYLESS_NV(dstr);
14198 SvANY(dstr) = new_XNV();
14200 SvNV_set(dstr, SvNVX(sstr));
14204 /* These are all the types that need complex bodies allocating. */
14206 const svtype sv_type = SvTYPE(sstr);
14207 const struct body_details *const sv_type_details
14208 = bodies_by_type + sv_type;
14212 Perl_croak(aTHX_ "Bizarre SvTYPE [%" IVdf "]", (IV)SvTYPE(sstr));
14213 NOT_REACHED; /* NOTREACHED */
14229 assert(sv_type_details->body_size);
14230 if (sv_type_details->arena) {
14231 new_body_inline(new_body, sv_type);
14233 = (void*)((char*)new_body - sv_type_details->offset);
14235 new_body = new_NOARENA(sv_type_details);
14239 SvANY(dstr) = new_body;
14242 Copy(((char*)SvANY(sstr)) + sv_type_details->offset,
14243 ((char*)SvANY(dstr)) + sv_type_details->offset,
14244 sv_type_details->copy, char);
14246 Copy(((char*)SvANY(sstr)),
14247 ((char*)SvANY(dstr)),
14248 sv_type_details->body_size + sv_type_details->offset, char);
14251 if (sv_type != SVt_PVAV && sv_type != SVt_PVHV
14252 && !isGV_with_GP(dstr)
14254 && !(sv_type == SVt_PVIO && !(IoFLAGS(dstr) & IOf_FAKE_DIRP)))
14255 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
14257 /* The Copy above means that all the source (unduplicated) pointers
14258 are now in the destination. We can check the flags and the
14259 pointers in either, but it's possible that there's less cache
14260 missing by always going for the destination.
14261 FIXME - instrument and check that assumption */
14262 if (sv_type >= SVt_PVMG) {
14264 SvMAGIC_set(dstr, mg_dup(SvMAGIC(dstr), param));
14265 if (SvOBJECT(dstr) && SvSTASH(dstr))
14266 SvSTASH_set(dstr, hv_dup_inc(SvSTASH(dstr), param));
14267 else SvSTASH_set(dstr, 0); /* don't copy DESTROY cache */
14270 /* The cast silences a GCC warning about unhandled types. */
14271 switch ((int)sv_type) {
14282 /* FIXME for plugins */
14283 re_dup_guts((REGEXP*) sstr, (REGEXP*) dstr, param);
14286 /* XXX LvTARGOFF sometimes holds PMOP* when DEBUGGING */
14287 if (LvTYPE(dstr) == 't') /* for tie: unrefcnted fake (SV**) */
14288 LvTARG(dstr) = dstr;
14289 else if (LvTYPE(dstr) == 'T') /* for tie: fake HE */
14290 LvTARG(dstr) = MUTABLE_SV(he_dup((HE*)LvTARG(dstr), 0, param));
14292 LvTARG(dstr) = sv_dup_inc(LvTARG(dstr), param);
14293 if (isREGEXP(sstr)) goto duprex;
14295 /* non-GP case already handled above */
14296 if(isGV_with_GP(sstr)) {
14297 GvNAME_HEK(dstr) = hek_dup(GvNAME_HEK(dstr), param);
14298 /* Don't call sv_add_backref here as it's going to be
14299 created as part of the magic cloning of the symbol
14300 table--unless this is during a join and the stash
14301 is not actually being cloned. */
14302 /* Danger Will Robinson - GvGP(dstr) isn't initialised
14303 at the point of this comment. */
14304 GvSTASH(dstr) = hv_dup(GvSTASH(dstr), param);
14305 if (param->flags & CLONEf_JOIN_IN)
14306 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
14307 GvGP_set(dstr, gp_dup(GvGP(sstr), param));
14308 (void)GpREFCNT_inc(GvGP(dstr));
14312 /* PL_parser->rsfp_filters entries have fake IoDIRP() */
14313 if(IoFLAGS(dstr) & IOf_FAKE_DIRP) {
14314 /* I have no idea why fake dirp (rsfps)
14315 should be treated differently but otherwise
14316 we end up with leaks -- sky*/
14317 IoTOP_GV(dstr) = gv_dup_inc(IoTOP_GV(dstr), param);
14318 IoFMT_GV(dstr) = gv_dup_inc(IoFMT_GV(dstr), param);
14319 IoBOTTOM_GV(dstr) = gv_dup_inc(IoBOTTOM_GV(dstr), param);
14321 IoTOP_GV(dstr) = gv_dup(IoTOP_GV(dstr), param);
14322 IoFMT_GV(dstr) = gv_dup(IoFMT_GV(dstr), param);
14323 IoBOTTOM_GV(dstr) = gv_dup(IoBOTTOM_GV(dstr), param);
14324 if (IoDIRP(dstr)) {
14325 IoDIRP(dstr) = dirp_dup(IoDIRP(dstr), param);
14328 /* IoDIRP(dstr) is already a copy of IoDIRP(sstr) */
14330 IoIFP(dstr) = fp_dup(IoIFP(sstr), IoTYPE(dstr), param);
14332 if (IoOFP(dstr) == IoIFP(sstr))
14333 IoOFP(dstr) = IoIFP(dstr);
14335 IoOFP(dstr) = fp_dup(IoOFP(dstr), IoTYPE(dstr), param);
14336 IoTOP_NAME(dstr) = SAVEPV(IoTOP_NAME(dstr));
14337 IoFMT_NAME(dstr) = SAVEPV(IoFMT_NAME(dstr));
14338 IoBOTTOM_NAME(dstr) = SAVEPV(IoBOTTOM_NAME(dstr));
14341 /* avoid cloning an empty array */
14342 if (AvARRAY((const AV *)sstr) && AvFILLp((const AV *)sstr) >= 0) {
14343 SV **dst_ary, **src_ary;
14344 SSize_t items = AvFILLp((const AV *)sstr) + 1;
14346 src_ary = AvARRAY((const AV *)sstr);
14347 Newxz(dst_ary, AvMAX((const AV *)sstr)+1, SV*);
14348 ptr_table_store(PL_ptr_table, src_ary, dst_ary);
14349 AvARRAY(MUTABLE_AV(dstr)) = dst_ary;
14350 AvALLOC((const AV *)dstr) = dst_ary;
14351 if (AvREAL((const AV *)sstr)) {
14352 dst_ary = sv_dup_inc_multiple(src_ary, dst_ary, items,
14356 while (items-- > 0)
14357 *dst_ary++ = sv_dup(*src_ary++, param);
14359 items = AvMAX((const AV *)sstr) - AvFILLp((const AV *)sstr);
14360 while (items-- > 0) {
14365 AvARRAY(MUTABLE_AV(dstr)) = NULL;
14366 AvALLOC((const AV *)dstr) = (SV**)NULL;
14367 AvMAX( (const AV *)dstr) = -1;
14368 AvFILLp((const AV *)dstr) = -1;
14372 if (HvARRAY((const HV *)sstr)) {
14374 const bool sharekeys = !!HvSHAREKEYS(sstr);
14375 XPVHV * const dxhv = (XPVHV*)SvANY(dstr);
14376 XPVHV * const sxhv = (XPVHV*)SvANY(sstr);
14378 Newx(darray, PERL_HV_ARRAY_ALLOC_BYTES(dxhv->xhv_max+1)
14379 + (SvOOK(sstr) ? sizeof(struct xpvhv_aux) : 0),
14381 HvARRAY(dstr) = (HE**)darray;
14382 while (i <= sxhv->xhv_max) {
14383 const HE * const source = HvARRAY(sstr)[i];
14384 HvARRAY(dstr)[i] = source
14385 ? he_dup(source, sharekeys, param) : 0;
14389 const struct xpvhv_aux * const saux = HvAUX(sstr);
14390 struct xpvhv_aux * const daux = HvAUX(dstr);
14391 /* This flag isn't copied. */
14394 if (saux->xhv_name_count) {
14395 HEK ** const sname = saux->xhv_name_u.xhvnameu_names;
14397 = saux->xhv_name_count < 0
14398 ? -saux->xhv_name_count
14399 : saux->xhv_name_count;
14400 HEK **shekp = sname + count;
14402 Newx(daux->xhv_name_u.xhvnameu_names, count, HEK *);
14403 dhekp = daux->xhv_name_u.xhvnameu_names + count;
14404 while (shekp-- > sname) {
14406 *dhekp = hek_dup(*shekp, param);
14410 daux->xhv_name_u.xhvnameu_name
14411 = hek_dup(saux->xhv_name_u.xhvnameu_name,
14414 daux->xhv_name_count = saux->xhv_name_count;
14416 daux->xhv_aux_flags = saux->xhv_aux_flags;
14417 #ifdef PERL_HASH_RANDOMIZE_KEYS
14418 daux->xhv_rand = saux->xhv_rand;
14419 daux->xhv_last_rand = saux->xhv_last_rand;
14421 daux->xhv_riter = saux->xhv_riter;
14422 daux->xhv_eiter = saux->xhv_eiter
14423 ? he_dup(saux->xhv_eiter,
14424 cBOOL(HvSHAREKEYS(sstr)), param) : 0;
14425 /* backref array needs refcnt=2; see sv_add_backref */
14426 daux->xhv_backreferences =
14427 (param->flags & CLONEf_JOIN_IN)
14428 /* when joining, we let the individual GVs and
14429 * CVs add themselves to backref as
14430 * needed. This avoids pulling in stuff
14431 * that isn't required, and simplifies the
14432 * case where stashes aren't cloned back
14433 * if they already exist in the parent
14436 : saux->xhv_backreferences
14437 ? (SvTYPE(saux->xhv_backreferences) == SVt_PVAV)
14438 ? MUTABLE_AV(SvREFCNT_inc(
14439 sv_dup_inc((const SV *)
14440 saux->xhv_backreferences, param)))
14441 : MUTABLE_AV(sv_dup((const SV *)
14442 saux->xhv_backreferences, param))
14445 daux->xhv_mro_meta = saux->xhv_mro_meta
14446 ? mro_meta_dup(saux->xhv_mro_meta, param)
14449 /* Record stashes for possible cloning in Perl_clone(). */
14451 av_push(param->stashes, dstr);
14455 HvARRAY(MUTABLE_HV(dstr)) = NULL;
14458 if (!(param->flags & CLONEf_COPY_STACKS)) {
14463 /* NOTE: not refcounted */
14464 SvANY(MUTABLE_CV(dstr))->xcv_stash =
14465 hv_dup(CvSTASH(dstr), param);
14466 if ((param->flags & CLONEf_JOIN_IN) && CvSTASH(dstr))
14467 Perl_sv_add_backref(aTHX_ MUTABLE_SV(CvSTASH(dstr)), dstr);
14468 if (!CvISXSUB(dstr)) {
14470 CvROOT(dstr) = OpREFCNT_inc(CvROOT(dstr));
14472 CvSLABBED_off(dstr);
14473 } else if (CvCONST(dstr)) {
14474 CvXSUBANY(dstr).any_ptr =
14475 sv_dup_inc((const SV *)CvXSUBANY(dstr).any_ptr, param);
14477 assert(!CvSLABBED(dstr));
14478 if (CvDYNFILE(dstr)) CvFILE(dstr) = SAVEPV(CvFILE(dstr));
14480 SvANY((CV *)dstr)->xcv_gv_u.xcv_hek =
14481 hek_dup(CvNAME_HEK((CV *)sstr), param);
14482 /* don't dup if copying back - CvGV isn't refcounted, so the
14483 * duped GV may never be freed. A bit of a hack! DAPM */
14485 SvANY(MUTABLE_CV(dstr))->xcv_gv_u.xcv_gv =
14487 ? gv_dup_inc(CvGV(sstr), param)
14488 : (param->flags & CLONEf_JOIN_IN)
14490 : gv_dup(CvGV(sstr), param);
14492 if (!CvISXSUB(sstr)) {
14493 PADLIST * padlist = CvPADLIST(sstr);
14495 padlist = padlist_dup(padlist, param);
14496 CvPADLIST_set(dstr, padlist);
14498 /* unthreaded perl can't sv_dup so we dont support unthreaded's CvHSCXT */
14499 PoisonPADLIST(dstr);
14502 CvWEAKOUTSIDE(sstr)
14503 ? cv_dup( CvOUTSIDE(dstr), param)
14504 : cv_dup_inc(CvOUTSIDE(dstr), param);
14514 Perl_sv_dup_inc(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14516 PERL_ARGS_ASSERT_SV_DUP_INC;
14517 return sstr ? SvREFCNT_inc(sv_dup_common(sstr, param)) : NULL;
14521 Perl_sv_dup(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14523 SV *dstr = sstr ? sv_dup_common(sstr, param) : NULL;
14524 PERL_ARGS_ASSERT_SV_DUP;
14526 /* Track every SV that (at least initially) had a reference count of 0.
14527 We need to do this by holding an actual reference to it in this array.
14528 If we attempt to cheat, turn AvREAL_off(), and store only pointers
14529 (akin to the stashes hash, and the perl stack), we come unstuck if
14530 a weak reference (or other SV legitimately SvREFCNT() == 0 for this
14531 thread) is manipulated in a CLONE method, because CLONE runs before the
14532 unreferenced array is walked to find SVs still with SvREFCNT() == 0
14533 (and fix things up by giving each a reference via the temps stack).
14534 Instead, during CLONE, if the 0-referenced SV has SvREFCNT_inc() and
14535 then SvREFCNT_dec(), it will be cleaned up (and added to the free list)
14536 before the walk of unreferenced happens and a reference to that is SV
14537 added to the temps stack. At which point we have the same SV considered
14538 to be in use, and free to be re-used. Not good.
14540 if (dstr && !(param->flags & CLONEf_COPY_STACKS) && !SvREFCNT(dstr)) {
14541 assert(param->unreferenced);
14542 av_push(param->unreferenced, SvREFCNT_inc(dstr));
14548 /* duplicate a context */
14551 Perl_cx_dup(pTHX_ PERL_CONTEXT *cxs, I32 ix, I32 max, CLONE_PARAMS* param)
14553 PERL_CONTEXT *ncxs;
14555 PERL_ARGS_ASSERT_CX_DUP;
14558 return (PERL_CONTEXT*)NULL;
14560 /* look for it in the table first */
14561 ncxs = (PERL_CONTEXT*)ptr_table_fetch(PL_ptr_table, cxs);
14565 /* create anew and remember what it is */
14566 Newx(ncxs, max + 1, PERL_CONTEXT);
14567 ptr_table_store(PL_ptr_table, cxs, ncxs);
14568 Copy(cxs, ncxs, max + 1, PERL_CONTEXT);
14571 PERL_CONTEXT * const ncx = &ncxs[ix];
14572 if (CxTYPE(ncx) == CXt_SUBST) {
14573 Perl_croak(aTHX_ "Cloning substitution context is unimplemented");
14576 ncx->blk_oldcop = (COP*)any_dup(ncx->blk_oldcop, param->proto_perl);
14577 switch (CxTYPE(ncx)) {
14579 ncx->blk_sub.cv = cv_dup_inc(ncx->blk_sub.cv, param);
14580 if(CxHASARGS(ncx)){
14581 ncx->blk_sub.savearray = av_dup_inc(ncx->blk_sub.savearray,param);
14583 ncx->blk_sub.savearray = NULL;
14585 ncx->blk_sub.prevcomppad = (PAD*)ptr_table_fetch(PL_ptr_table,
14586 ncx->blk_sub.prevcomppad);
14589 ncx->blk_eval.old_namesv = sv_dup_inc(ncx->blk_eval.old_namesv,
14591 /* XXX should this sv_dup_inc? Or only if CxEVAL_TXT_REFCNTED ???? */
14592 ncx->blk_eval.cur_text = sv_dup(ncx->blk_eval.cur_text, param);
14593 ncx->blk_eval.cv = cv_dup(ncx->blk_eval.cv, param);
14594 /* XXX what do do with cur_top_env ???? */
14596 case CXt_LOOP_LAZYSV:
14597 ncx->blk_loop.state_u.lazysv.end
14598 = sv_dup_inc(ncx->blk_loop.state_u.lazysv.end, param);
14599 /* Fallthrough: duplicate lazysv.cur by using the ary.ary
14600 duplication code instead.
14601 We are taking advantage of (1) av_dup_inc and sv_dup_inc
14602 actually being the same function, and (2) order
14603 equivalence of the two unions.
14604 We can assert the later [but only at run time :-(] */
14605 assert ((void *) &ncx->blk_loop.state_u.ary.ary ==
14606 (void *) &ncx->blk_loop.state_u.lazysv.cur);
14609 ncx->blk_loop.state_u.ary.ary
14610 = av_dup_inc(ncx->blk_loop.state_u.ary.ary, param);
14612 case CXt_LOOP_LIST:
14613 case CXt_LOOP_LAZYIV:
14614 /* code common to all 'for' CXt_LOOP_* types */
14615 ncx->blk_loop.itersave =
14616 sv_dup_inc(ncx->blk_loop.itersave, param);
14617 if (CxPADLOOP(ncx)) {
14618 PADOFFSET off = ncx->blk_loop.itervar_u.svp
14619 - &CX_CURPAD_SV(ncx->blk_loop, 0);
14620 ncx->blk_loop.oldcomppad =
14621 (PAD*)ptr_table_fetch(PL_ptr_table,
14622 ncx->blk_loop.oldcomppad);
14623 ncx->blk_loop.itervar_u.svp =
14624 &CX_CURPAD_SV(ncx->blk_loop, off);
14627 /* this copies the GV if CXp_FOR_GV, or the SV for an
14628 * alias (for \$x (...)) - relies on gv_dup being the
14629 * same as sv_dup */
14630 ncx->blk_loop.itervar_u.gv
14631 = gv_dup((const GV *)ncx->blk_loop.itervar_u.gv,
14635 case CXt_LOOP_PLAIN:
14638 ncx->blk_format.prevcomppad =
14639 (PAD*)ptr_table_fetch(PL_ptr_table,
14640 ncx->blk_format.prevcomppad);
14641 ncx->blk_format.cv = cv_dup_inc(ncx->blk_format.cv, param);
14642 ncx->blk_format.gv = gv_dup(ncx->blk_format.gv, param);
14643 ncx->blk_format.dfoutgv = gv_dup_inc(ncx->blk_format.dfoutgv,
14647 ncx->blk_givwhen.defsv_save =
14648 sv_dup_inc(ncx->blk_givwhen.defsv_save, param);
14661 /* duplicate a stack info structure */
14664 Perl_si_dup(pTHX_ PERL_SI *si, CLONE_PARAMS* param)
14668 PERL_ARGS_ASSERT_SI_DUP;
14671 return (PERL_SI*)NULL;
14673 /* look for it in the table first */
14674 nsi = (PERL_SI*)ptr_table_fetch(PL_ptr_table, si);
14678 /* create anew and remember what it is */
14679 Newxz(nsi, 1, PERL_SI);
14680 ptr_table_store(PL_ptr_table, si, nsi);
14682 nsi->si_stack = av_dup_inc(si->si_stack, param);
14683 nsi->si_cxix = si->si_cxix;
14684 nsi->si_cxmax = si->si_cxmax;
14685 nsi->si_cxstack = cx_dup(si->si_cxstack, si->si_cxix, si->si_cxmax, param);
14686 nsi->si_type = si->si_type;
14687 nsi->si_prev = si_dup(si->si_prev, param);
14688 nsi->si_next = si_dup(si->si_next, param);
14689 nsi->si_markoff = si->si_markoff;
14694 #define POPINT(ss,ix) ((ss)[--(ix)].any_i32)
14695 #define TOPINT(ss,ix) ((ss)[ix].any_i32)
14696 #define POPLONG(ss,ix) ((ss)[--(ix)].any_long)
14697 #define TOPLONG(ss,ix) ((ss)[ix].any_long)
14698 #define POPIV(ss,ix) ((ss)[--(ix)].any_iv)
14699 #define TOPIV(ss,ix) ((ss)[ix].any_iv)
14700 #define POPUV(ss,ix) ((ss)[--(ix)].any_uv)
14701 #define TOPUV(ss,ix) ((ss)[ix].any_uv)
14702 #define POPBOOL(ss,ix) ((ss)[--(ix)].any_bool)
14703 #define TOPBOOL(ss,ix) ((ss)[ix].any_bool)
14704 #define POPPTR(ss,ix) ((ss)[--(ix)].any_ptr)
14705 #define TOPPTR(ss,ix) ((ss)[ix].any_ptr)
14706 #define POPDPTR(ss,ix) ((ss)[--(ix)].any_dptr)
14707 #define TOPDPTR(ss,ix) ((ss)[ix].any_dptr)
14708 #define POPDXPTR(ss,ix) ((ss)[--(ix)].any_dxptr)
14709 #define TOPDXPTR(ss,ix) ((ss)[ix].any_dxptr)
14712 #define pv_dup_inc(p) SAVEPV(p)
14713 #define pv_dup(p) SAVEPV(p)
14714 #define svp_dup_inc(p,pp) any_dup(p,pp)
14716 /* map any object to the new equivent - either something in the
14717 * ptr table, or something in the interpreter structure
14721 Perl_any_dup(pTHX_ void *v, const PerlInterpreter *proto_perl)
14725 PERL_ARGS_ASSERT_ANY_DUP;
14728 return (void*)NULL;
14730 /* look for it in the table first */
14731 ret = ptr_table_fetch(PL_ptr_table, v);
14735 /* see if it is part of the interpreter structure */
14736 if (v >= (void*)proto_perl && v < (void*)(proto_perl+1))
14737 ret = (void*)(((char*)aTHX) + (((char*)v) - (char*)proto_perl));
14745 /* duplicate the save stack */
14748 Perl_ss_dup(pTHX_ PerlInterpreter *proto_perl, CLONE_PARAMS* param)
14751 ANY * const ss = proto_perl->Isavestack;
14752 const I32 max = proto_perl->Isavestack_max + SS_MAXPUSH;
14753 I32 ix = proto_perl->Isavestack_ix;
14766 void (*dptr) (void*);
14767 void (*dxptr) (pTHX_ void*);
14769 PERL_ARGS_ASSERT_SS_DUP;
14771 Newxz(nss, max, ANY);
14774 const UV uv = POPUV(ss,ix);
14775 const U8 type = (U8)uv & SAVE_MASK;
14777 TOPUV(nss,ix) = uv;
14779 case SAVEt_CLEARSV:
14780 case SAVEt_CLEARPADRANGE:
14782 case SAVEt_HELEM: /* hash element */
14783 case SAVEt_SV: /* scalar reference */
14784 sv = (const SV *)POPPTR(ss,ix);
14785 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14787 case SAVEt_ITEM: /* normal string */
14788 case SAVEt_GVSV: /* scalar slot in GV */
14789 sv = (const SV *)POPPTR(ss,ix);
14790 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14791 if (type == SAVEt_SV)
14795 case SAVEt_MORTALIZESV:
14796 case SAVEt_READONLY_OFF:
14797 sv = (const SV *)POPPTR(ss,ix);
14798 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14800 case SAVEt_FREEPADNAME:
14801 ptr = POPPTR(ss,ix);
14802 TOPPTR(nss,ix) = padname_dup((PADNAME *)ptr, param);
14803 PadnameREFCNT((PADNAME *)TOPPTR(nss,ix))++;
14805 case SAVEt_SHARED_PVREF: /* char* in shared space */
14806 c = (char*)POPPTR(ss,ix);
14807 TOPPTR(nss,ix) = savesharedpv(c);
14808 ptr = POPPTR(ss,ix);
14809 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14811 case SAVEt_GENERIC_SVREF: /* generic sv */
14812 case SAVEt_SVREF: /* scalar reference */
14813 sv = (const SV *)POPPTR(ss,ix);
14814 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14815 if (type == SAVEt_SVREF)
14816 SvREFCNT_inc_simple_void((SV *)TOPPTR(nss,ix));
14817 ptr = POPPTR(ss,ix);
14818 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14820 case SAVEt_GVSLOT: /* any slot in GV */
14821 sv = (const SV *)POPPTR(ss,ix);
14822 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14823 ptr = POPPTR(ss,ix);
14824 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14825 sv = (const SV *)POPPTR(ss,ix);
14826 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14828 case SAVEt_HV: /* hash reference */
14829 case SAVEt_AV: /* array reference */
14830 sv = (const SV *) POPPTR(ss,ix);
14831 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14833 case SAVEt_COMPPAD:
14835 sv = (const SV *) POPPTR(ss,ix);
14836 TOPPTR(nss,ix) = sv_dup(sv, param);
14838 case SAVEt_INT: /* int reference */
14839 ptr = POPPTR(ss,ix);
14840 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14841 intval = (int)POPINT(ss,ix);
14842 TOPINT(nss,ix) = intval;
14844 case SAVEt_LONG: /* long reference */
14845 ptr = POPPTR(ss,ix);
14846 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14847 longval = (long)POPLONG(ss,ix);
14848 TOPLONG(nss,ix) = longval;
14850 case SAVEt_I32: /* I32 reference */
14851 ptr = POPPTR(ss,ix);
14852 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14854 TOPINT(nss,ix) = i;
14856 case SAVEt_IV: /* IV reference */
14857 case SAVEt_STRLEN: /* STRLEN/size_t ref */
14858 ptr = POPPTR(ss,ix);
14859 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14861 TOPIV(nss,ix) = iv;
14863 case SAVEt_TMPSFLOOR:
14865 TOPIV(nss,ix) = iv;
14867 case SAVEt_HPTR: /* HV* reference */
14868 case SAVEt_APTR: /* AV* reference */
14869 case SAVEt_SPTR: /* SV* reference */
14870 ptr = POPPTR(ss,ix);
14871 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14872 sv = (const SV *)POPPTR(ss,ix);
14873 TOPPTR(nss,ix) = sv_dup(sv, param);
14875 case SAVEt_VPTR: /* random* reference */
14876 ptr = POPPTR(ss,ix);
14877 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14879 case SAVEt_INT_SMALL:
14880 case SAVEt_I32_SMALL:
14881 case SAVEt_I16: /* I16 reference */
14882 case SAVEt_I8: /* I8 reference */
14884 ptr = POPPTR(ss,ix);
14885 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14887 case SAVEt_GENERIC_PVREF: /* generic char* */
14888 case SAVEt_PPTR: /* char* reference */
14889 ptr = POPPTR(ss,ix);
14890 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14891 c = (char*)POPPTR(ss,ix);
14892 TOPPTR(nss,ix) = pv_dup(c);
14894 case SAVEt_GP: /* scalar reference */
14895 gp = (GP*)POPPTR(ss,ix);
14896 TOPPTR(nss,ix) = gp = gp_dup(gp, param);
14897 (void)GpREFCNT_inc(gp);
14898 gv = (const GV *)POPPTR(ss,ix);
14899 TOPPTR(nss,ix) = gv_dup_inc(gv, param);
14902 ptr = POPPTR(ss,ix);
14903 if (ptr && (((OP*)ptr)->op_private & OPpREFCOUNTED)) {
14904 /* these are assumed to be refcounted properly */
14906 switch (((OP*)ptr)->op_type) {
14908 case OP_LEAVESUBLV:
14912 case OP_LEAVEWRITE:
14913 TOPPTR(nss,ix) = ptr;
14916 (void) OpREFCNT_inc(o);
14920 TOPPTR(nss,ix) = NULL;
14925 TOPPTR(nss,ix) = NULL;
14927 case SAVEt_FREECOPHH:
14928 ptr = POPPTR(ss,ix);
14929 TOPPTR(nss,ix) = cophh_copy((COPHH *)ptr);
14931 case SAVEt_ADELETE:
14932 av = (const AV *)POPPTR(ss,ix);
14933 TOPPTR(nss,ix) = av_dup_inc(av, param);
14935 TOPINT(nss,ix) = i;
14938 hv = (const HV *)POPPTR(ss,ix);
14939 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14941 TOPINT(nss,ix) = i;
14944 c = (char*)POPPTR(ss,ix);
14945 TOPPTR(nss,ix) = pv_dup_inc(c);
14947 case SAVEt_STACK_POS: /* Position on Perl stack */
14949 TOPINT(nss,ix) = i;
14951 case SAVEt_DESTRUCTOR:
14952 ptr = POPPTR(ss,ix);
14953 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14954 dptr = POPDPTR(ss,ix);
14955 TOPDPTR(nss,ix) = DPTR2FPTR(void (*)(void*),
14956 any_dup(FPTR2DPTR(void *, dptr),
14959 case SAVEt_DESTRUCTOR_X:
14960 ptr = POPPTR(ss,ix);
14961 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14962 dxptr = POPDXPTR(ss,ix);
14963 TOPDXPTR(nss,ix) = DPTR2FPTR(void (*)(pTHX_ void*),
14964 any_dup(FPTR2DPTR(void *, dxptr),
14967 case SAVEt_REGCONTEXT:
14969 ix -= uv >> SAVE_TIGHT_SHIFT;
14971 case SAVEt_AELEM: /* array element */
14972 sv = (const SV *)POPPTR(ss,ix);
14973 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14975 TOPINT(nss,ix) = i;
14976 av = (const AV *)POPPTR(ss,ix);
14977 TOPPTR(nss,ix) = av_dup_inc(av, param);
14980 ptr = POPPTR(ss,ix);
14981 TOPPTR(nss,ix) = ptr;
14984 ptr = POPPTR(ss,ix);
14985 ptr = cophh_copy((COPHH*)ptr);
14986 TOPPTR(nss,ix) = ptr;
14988 TOPINT(nss,ix) = i;
14989 if (i & HINT_LOCALIZE_HH) {
14990 hv = (const HV *)POPPTR(ss,ix);
14991 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14994 case SAVEt_PADSV_AND_MORTALIZE:
14995 longval = (long)POPLONG(ss,ix);
14996 TOPLONG(nss,ix) = longval;
14997 ptr = POPPTR(ss,ix);
14998 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14999 sv = (const SV *)POPPTR(ss,ix);
15000 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
15002 case SAVEt_SET_SVFLAGS:
15004 TOPINT(nss,ix) = i;
15006 TOPINT(nss,ix) = i;
15007 sv = (const SV *)POPPTR(ss,ix);
15008 TOPPTR(nss,ix) = sv_dup(sv, param);
15010 case SAVEt_COMPILE_WARNINGS:
15011 ptr = POPPTR(ss,ix);
15012 TOPPTR(nss,ix) = DUP_WARNINGS((STRLEN*)ptr);
15015 ptr = POPPTR(ss,ix);
15016 TOPPTR(nss,ix) = parser_dup((const yy_parser*)ptr, param);
15020 "panic: ss_dup inconsistency (%" IVdf ")", (IV) type);
15028 /* if sv is a stash, call $class->CLONE_SKIP(), and set the SVphv_CLONEABLE
15029 * flag to the result. This is done for each stash before cloning starts,
15030 * so we know which stashes want their objects cloned */
15033 do_mark_cloneable_stash(pTHX_ SV *const sv)
15035 const HEK * const hvname = HvNAME_HEK((const HV *)sv);
15037 GV* const cloner = gv_fetchmethod_autoload(MUTABLE_HV(sv), "CLONE_SKIP", 0);
15038 SvFLAGS(sv) |= SVphv_CLONEABLE; /* clone objects by default */
15039 if (cloner && GvCV(cloner)) {
15046 mXPUSHs(newSVhek(hvname));
15048 call_sv(MUTABLE_SV(GvCV(cloner)), G_SCALAR);
15055 SvFLAGS(sv) &= ~SVphv_CLONEABLE;
15063 =for apidoc perl_clone
15065 Create and return a new interpreter by cloning the current one.
15067 C<perl_clone> takes these flags as parameters:
15069 C<CLONEf_COPY_STACKS> - is used to, well, copy the stacks also,
15070 without it we only clone the data and zero the stacks,
15071 with it we copy the stacks and the new perl interpreter is
15072 ready to run at the exact same point as the previous one.
15073 The pseudo-fork code uses C<COPY_STACKS> while the
15074 threads->create doesn't.
15076 C<CLONEf_KEEP_PTR_TABLE> -
15077 C<perl_clone> keeps a ptr_table with the pointer of the old
15078 variable as a key and the new variable as a value,
15079 this allows it to check if something has been cloned and not
15080 clone it again but rather just use the value and increase the
15081 refcount. If C<KEEP_PTR_TABLE> is not set then C<perl_clone> will kill
15082 the ptr_table using the function
15083 C<ptr_table_free(PL_ptr_table); PL_ptr_table = NULL;>,
15084 reason to keep it around is if you want to dup some of your own
15085 variable who are outside the graph perl scans, an example of this
15086 code is in F<threads.xs> create.
15088 C<CLONEf_CLONE_HOST> -
15089 This is a win32 thing, it is ignored on unix, it tells perls
15090 win32host code (which is c++) to clone itself, this is needed on
15091 win32 if you want to run two threads at the same time,
15092 if you just want to do some stuff in a separate perl interpreter
15093 and then throw it away and return to the original one,
15094 you don't need to do anything.
15099 /* XXX the above needs expanding by someone who actually understands it ! */
15100 EXTERN_C PerlInterpreter *
15101 perl_clone_host(PerlInterpreter* proto_perl, UV flags);
15104 perl_clone(PerlInterpreter *proto_perl, UV flags)
15107 #ifdef PERL_IMPLICIT_SYS
15109 PERL_ARGS_ASSERT_PERL_CLONE;
15111 /* perlhost.h so we need to call into it
15112 to clone the host, CPerlHost should have a c interface, sky */
15114 #ifndef __amigaos4__
15115 if (flags & CLONEf_CLONE_HOST) {
15116 return perl_clone_host(proto_perl,flags);
15119 return perl_clone_using(proto_perl, flags,
15121 proto_perl->IMemShared,
15122 proto_perl->IMemParse,
15124 proto_perl->IStdIO,
15128 proto_perl->IProc);
15132 perl_clone_using(PerlInterpreter *proto_perl, UV flags,
15133 struct IPerlMem* ipM, struct IPerlMem* ipMS,
15134 struct IPerlMem* ipMP, struct IPerlEnv* ipE,
15135 struct IPerlStdIO* ipStd, struct IPerlLIO* ipLIO,
15136 struct IPerlDir* ipD, struct IPerlSock* ipS,
15137 struct IPerlProc* ipP)
15139 /* XXX many of the string copies here can be optimized if they're
15140 * constants; they need to be allocated as common memory and just
15141 * their pointers copied. */
15144 CLONE_PARAMS clone_params;
15145 CLONE_PARAMS* const param = &clone_params;
15147 PerlInterpreter * const my_perl = (PerlInterpreter*)(*ipM->pMalloc)(ipM, sizeof(PerlInterpreter));
15149 PERL_ARGS_ASSERT_PERL_CLONE_USING;
15150 #else /* !PERL_IMPLICIT_SYS */
15152 CLONE_PARAMS clone_params;
15153 CLONE_PARAMS* param = &clone_params;
15154 PerlInterpreter * const my_perl = (PerlInterpreter*)PerlMem_malloc(sizeof(PerlInterpreter));
15156 PERL_ARGS_ASSERT_PERL_CLONE;
15157 #endif /* PERL_IMPLICIT_SYS */
15159 /* for each stash, determine whether its objects should be cloned */
15160 S_visit(proto_perl, do_mark_cloneable_stash, SVt_PVHV, SVTYPEMASK);
15161 PERL_SET_THX(my_perl);
15164 PoisonNew(my_perl, 1, PerlInterpreter);
15167 PL_defstash = NULL; /* may be used by perl malloc() */
15170 PL_scopestack_name = 0;
15172 PL_savestack_ix = 0;
15173 PL_savestack_max = -1;
15174 PL_sig_pending = 0;
15176 Zero(&PL_debug_pad, 1, struct perl_debug_pad);
15177 Zero(&PL_padname_undef, 1, PADNAME);
15178 Zero(&PL_padname_const, 1, PADNAME);
15179 # ifdef DEBUG_LEAKING_SCALARS
15180 PL_sv_serial = (((UV)my_perl >> 2) & 0xfff) * 1000000;
15182 # ifdef PERL_TRACE_OPS
15183 Zero(PL_op_exec_cnt, OP_max+2, UV);
15185 #else /* !DEBUGGING */
15186 Zero(my_perl, 1, PerlInterpreter);
15187 #endif /* DEBUGGING */
15189 #ifdef PERL_IMPLICIT_SYS
15190 /* host pointers */
15192 PL_MemShared = ipMS;
15193 PL_MemParse = ipMP;
15200 #endif /* PERL_IMPLICIT_SYS */
15203 param->flags = flags;
15204 /* Nothing in the core code uses this, but we make it available to
15205 extensions (using mg_dup). */
15206 param->proto_perl = proto_perl;
15207 /* Likely nothing will use this, but it is initialised to be consistent
15208 with Perl_clone_params_new(). */
15209 param->new_perl = my_perl;
15210 param->unreferenced = NULL;
15213 INIT_TRACK_MEMPOOL(my_perl->Imemory_debug_header, my_perl);
15215 PL_body_arenas = NULL;
15216 Zero(&PL_body_roots, 1, PL_body_roots);
15220 PL_sv_arenaroot = NULL;
15222 PL_debug = proto_perl->Idebug;
15224 /* dbargs array probably holds garbage */
15227 PL_compiling = proto_perl->Icompiling;
15229 /* pseudo environmental stuff */
15230 PL_origargc = proto_perl->Iorigargc;
15231 PL_origargv = proto_perl->Iorigargv;
15233 #ifndef NO_TAINT_SUPPORT
15234 /* Set tainting stuff before PerlIO_debug can possibly get called */
15235 PL_tainting = proto_perl->Itainting;
15236 PL_taint_warn = proto_perl->Itaint_warn;
15238 PL_tainting = FALSE;
15239 PL_taint_warn = FALSE;
15242 PL_minus_c = proto_perl->Iminus_c;
15244 PL_localpatches = proto_perl->Ilocalpatches;
15245 PL_splitstr = proto_perl->Isplitstr;
15246 PL_minus_n = proto_perl->Iminus_n;
15247 PL_minus_p = proto_perl->Iminus_p;
15248 PL_minus_l = proto_perl->Iminus_l;
15249 PL_minus_a = proto_perl->Iminus_a;
15250 PL_minus_E = proto_perl->Iminus_E;
15251 PL_minus_F = proto_perl->Iminus_F;
15252 PL_doswitches = proto_perl->Idoswitches;
15253 PL_dowarn = proto_perl->Idowarn;
15254 #ifdef PERL_SAWAMPERSAND
15255 PL_sawampersand = proto_perl->Isawampersand;
15257 PL_unsafe = proto_perl->Iunsafe;
15258 PL_perldb = proto_perl->Iperldb;
15259 PL_perl_destruct_level = proto_perl->Iperl_destruct_level;
15260 PL_exit_flags = proto_perl->Iexit_flags;
15262 /* XXX time(&PL_basetime) when asked for? */
15263 PL_basetime = proto_perl->Ibasetime;
15265 PL_maxsysfd = proto_perl->Imaxsysfd;
15266 PL_statusvalue = proto_perl->Istatusvalue;
15268 PL_statusvalue_vms = proto_perl->Istatusvalue_vms;
15270 PL_statusvalue_posix = proto_perl->Istatusvalue_posix;
15273 /* RE engine related */
15274 PL_regmatch_slab = NULL;
15275 PL_reg_curpm = NULL;
15277 PL_sub_generation = proto_perl->Isub_generation;
15279 /* funky return mechanisms */
15280 PL_forkprocess = proto_perl->Iforkprocess;
15282 /* internal state */
15283 PL_main_start = proto_perl->Imain_start;
15284 PL_eval_root = proto_perl->Ieval_root;
15285 PL_eval_start = proto_perl->Ieval_start;
15287 PL_filemode = proto_perl->Ifilemode;
15288 PL_lastfd = proto_perl->Ilastfd;
15289 PL_oldname = proto_perl->Ioldname; /* XXX not quite right */
15292 PL_gensym = proto_perl->Igensym;
15294 PL_laststatval = proto_perl->Ilaststatval;
15295 PL_laststype = proto_perl->Ilaststype;
15298 PL_profiledata = NULL;
15300 PL_generation = proto_perl->Igeneration;
15302 PL_in_clean_objs = proto_perl->Iin_clean_objs;
15303 PL_in_clean_all = proto_perl->Iin_clean_all;
15305 PL_delaymagic_uid = proto_perl->Idelaymagic_uid;
15306 PL_delaymagic_euid = proto_perl->Idelaymagic_euid;
15307 PL_delaymagic_gid = proto_perl->Idelaymagic_gid;
15308 PL_delaymagic_egid = proto_perl->Idelaymagic_egid;
15309 PL_nomemok = proto_perl->Inomemok;
15310 PL_an = proto_perl->Ian;
15311 PL_evalseq = proto_perl->Ievalseq;
15312 PL_origenviron = proto_perl->Iorigenviron; /* XXX not quite right */
15313 PL_origalen = proto_perl->Iorigalen;
15315 PL_sighandlerp = proto_perl->Isighandlerp;
15317 PL_runops = proto_perl->Irunops;
15319 PL_subline = proto_perl->Isubline;
15321 PL_cv_has_eval = proto_perl->Icv_has_eval;
15324 PL_cryptseen = proto_perl->Icryptseen;
15327 #ifdef USE_LOCALE_COLLATE
15328 PL_collation_ix = proto_perl->Icollation_ix;
15329 PL_collation_standard = proto_perl->Icollation_standard;
15330 PL_collxfrm_base = proto_perl->Icollxfrm_base;
15331 PL_collxfrm_mult = proto_perl->Icollxfrm_mult;
15332 PL_strxfrm_max_cp = proto_perl->Istrxfrm_max_cp;
15333 #endif /* USE_LOCALE_COLLATE */
15335 #ifdef USE_LOCALE_NUMERIC
15336 PL_numeric_standard = proto_perl->Inumeric_standard;
15337 PL_numeric_local = proto_perl->Inumeric_local;
15338 #endif /* !USE_LOCALE_NUMERIC */
15340 /* Did the locale setup indicate UTF-8? */
15341 PL_utf8locale = proto_perl->Iutf8locale;
15342 PL_in_utf8_CTYPE_locale = proto_perl->Iin_utf8_CTYPE_locale;
15343 PL_in_utf8_COLLATE_locale = proto_perl->Iin_utf8_COLLATE_locale;
15344 /* Unicode features (see perlrun/-C) */
15345 PL_unicode = proto_perl->Iunicode;
15347 /* Pre-5.8 signals control */
15348 PL_signals = proto_perl->Isignals;
15350 /* times() ticks per second */
15351 PL_clocktick = proto_perl->Iclocktick;
15353 /* Recursion stopper for PerlIO_find_layer */
15354 PL_in_load_module = proto_perl->Iin_load_module;
15356 /* sort() routine */
15357 PL_sort_RealCmp = proto_perl->Isort_RealCmp;
15359 /* Not really needed/useful since the reenrant_retint is "volatile",
15360 * but do it for consistency's sake. */
15361 PL_reentrant_retint = proto_perl->Ireentrant_retint;
15363 /* Hooks to shared SVs and locks. */
15364 PL_sharehook = proto_perl->Isharehook;
15365 PL_lockhook = proto_perl->Ilockhook;
15366 PL_unlockhook = proto_perl->Iunlockhook;
15367 PL_threadhook = proto_perl->Ithreadhook;
15368 PL_destroyhook = proto_perl->Idestroyhook;
15369 PL_signalhook = proto_perl->Isignalhook;
15371 PL_globhook = proto_perl->Iglobhook;
15374 PL_last_swash_hv = NULL; /* reinits on demand */
15375 PL_last_swash_klen = 0;
15376 PL_last_swash_key[0]= '\0';
15377 PL_last_swash_tmps = (U8*)NULL;
15378 PL_last_swash_slen = 0;
15380 PL_srand_called = proto_perl->Isrand_called;
15381 Copy(&(proto_perl->Irandom_state), &PL_random_state, 1, PL_RANDOM_STATE_TYPE);
15383 if (flags & CLONEf_COPY_STACKS) {
15384 /* next allocation will be PL_tmps_stack[PL_tmps_ix+1] */
15385 PL_tmps_ix = proto_perl->Itmps_ix;
15386 PL_tmps_max = proto_perl->Itmps_max;
15387 PL_tmps_floor = proto_perl->Itmps_floor;
15389 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
15390 * NOTE: unlike the others! */
15391 PL_scopestack_ix = proto_perl->Iscopestack_ix;
15392 PL_scopestack_max = proto_perl->Iscopestack_max;
15394 /* next SSPUSHFOO() sets PL_savestack[PL_savestack_ix]
15395 * NOTE: unlike the others! */
15396 PL_savestack_ix = proto_perl->Isavestack_ix;
15397 PL_savestack_max = proto_perl->Isavestack_max;
15400 PL_start_env = proto_perl->Istart_env; /* XXXXXX */
15401 PL_top_env = &PL_start_env;
15403 PL_op = proto_perl->Iop;
15406 PL_Xpv = (XPV*)NULL;
15407 my_perl->Ina = proto_perl->Ina;
15409 PL_statcache = proto_perl->Istatcache;
15411 #ifndef NO_TAINT_SUPPORT
15412 PL_tainted = proto_perl->Itainted;
15414 PL_tainted = FALSE;
15416 PL_curpm = proto_perl->Icurpm; /* XXX No PMOP ref count */
15418 PL_chopset = proto_perl->Ichopset; /* XXX never deallocated */
15420 PL_restartjmpenv = proto_perl->Irestartjmpenv;
15421 PL_restartop = proto_perl->Irestartop;
15422 PL_in_eval = proto_perl->Iin_eval;
15423 PL_delaymagic = proto_perl->Idelaymagic;
15424 PL_phase = proto_perl->Iphase;
15425 PL_localizing = proto_perl->Ilocalizing;
15427 PL_hv_fetch_ent_mh = NULL;
15428 PL_modcount = proto_perl->Imodcount;
15429 PL_lastgotoprobe = NULL;
15430 PL_dumpindent = proto_perl->Idumpindent;
15432 PL_efloatbuf = NULL; /* reinits on demand */
15433 PL_efloatsize = 0; /* reinits on demand */
15437 PL_colorset = 0; /* reinits PL_colors[] */
15438 /*PL_colors[6] = {0,0,0,0,0,0};*/
15440 /* Pluggable optimizer */
15441 PL_peepp = proto_perl->Ipeepp;
15442 PL_rpeepp = proto_perl->Irpeepp;
15443 /* op_free() hook */
15444 PL_opfreehook = proto_perl->Iopfreehook;
15446 #ifdef USE_REENTRANT_API
15447 /* XXX: things like -Dm will segfault here in perlio, but doing
15448 * PERL_SET_CONTEXT(proto_perl);
15449 * breaks too many other things
15451 Perl_reentrant_init(aTHX);
15454 /* create SV map for pointer relocation */
15455 PL_ptr_table = ptr_table_new();
15457 /* initialize these special pointers as early as possible */
15459 ptr_table_store(PL_ptr_table, &proto_perl->Isv_undef, &PL_sv_undef);
15460 ptr_table_store(PL_ptr_table, &proto_perl->Isv_no, &PL_sv_no);
15461 ptr_table_store(PL_ptr_table, &proto_perl->Isv_zero, &PL_sv_zero);
15462 ptr_table_store(PL_ptr_table, &proto_perl->Isv_yes, &PL_sv_yes);
15463 ptr_table_store(PL_ptr_table, &proto_perl->Ipadname_const,
15464 &PL_padname_const);
15466 /* create (a non-shared!) shared string table */
15467 PL_strtab = newHV();
15468 HvSHAREKEYS_off(PL_strtab);
15469 hv_ksplit(PL_strtab, HvTOTALKEYS(proto_perl->Istrtab));
15470 ptr_table_store(PL_ptr_table, proto_perl->Istrtab, PL_strtab);
15472 Zero(PL_sv_consts, SV_CONSTS_COUNT, SV*);
15474 /* This PV will be free'd special way so must set it same way op.c does */
15475 PL_compiling.cop_file = savesharedpv(PL_compiling.cop_file);
15476 ptr_table_store(PL_ptr_table, proto_perl->Icompiling.cop_file, PL_compiling.cop_file);
15478 ptr_table_store(PL_ptr_table, &proto_perl->Icompiling, &PL_compiling);
15479 PL_compiling.cop_warnings = DUP_WARNINGS(PL_compiling.cop_warnings);
15480 CopHINTHASH_set(&PL_compiling, cophh_copy(CopHINTHASH_get(&PL_compiling)));
15481 PL_curcop = (COP*)any_dup(proto_perl->Icurcop, proto_perl);
15483 param->stashes = newAV(); /* Setup array of objects to call clone on */
15484 /* This makes no difference to the implementation, as it always pushes
15485 and shifts pointers to other SVs without changing their reference
15486 count, with the array becoming empty before it is freed. However, it
15487 makes it conceptually clear what is going on, and will avoid some
15488 work inside av.c, filling slots between AvFILL() and AvMAX() with
15489 &PL_sv_undef, and SvREFCNT_dec()ing those. */
15490 AvREAL_off(param->stashes);
15492 if (!(flags & CLONEf_COPY_STACKS)) {
15493 param->unreferenced = newAV();
15496 #ifdef PERLIO_LAYERS
15497 /* Clone PerlIO tables as soon as we can handle general xx_dup() */
15498 PerlIO_clone(aTHX_ proto_perl, param);
15501 PL_envgv = gv_dup_inc(proto_perl->Ienvgv, param);
15502 PL_incgv = gv_dup_inc(proto_perl->Iincgv, param);
15503 PL_hintgv = gv_dup_inc(proto_perl->Ihintgv, param);
15504 PL_origfilename = SAVEPV(proto_perl->Iorigfilename);
15505 PL_xsubfilename = proto_perl->Ixsubfilename;
15506 PL_diehook = sv_dup_inc(proto_perl->Idiehook, param);
15507 PL_warnhook = sv_dup_inc(proto_perl->Iwarnhook, param);
15510 PL_patchlevel = sv_dup_inc(proto_perl->Ipatchlevel, param);
15511 PL_inplace = SAVEPV(proto_perl->Iinplace);
15512 PL_e_script = sv_dup_inc(proto_perl->Ie_script, param);
15514 /* magical thingies */
15516 SvPVCLEAR(PERL_DEBUG_PAD(0)); /* For regex debugging. */
15517 SvPVCLEAR(PERL_DEBUG_PAD(1)); /* ext/re needs these */
15518 SvPVCLEAR(PERL_DEBUG_PAD(2)); /* even without DEBUGGING. */
15521 /* Clone the regex array */
15522 /* ORANGE FIXME for plugins, probably in the SV dup code.
15523 newSViv(PTR2IV(CALLREGDUPE(
15524 INT2PTR(REGEXP *, SvIVX(regex)), param))))
15526 PL_regex_padav = av_dup_inc(proto_perl->Iregex_padav, param);
15527 PL_regex_pad = AvARRAY(PL_regex_padav);
15529 PL_stashpadmax = proto_perl->Istashpadmax;
15530 PL_stashpadix = proto_perl->Istashpadix ;
15531 Newx(PL_stashpad, PL_stashpadmax, HV *);
15534 for (; o < PL_stashpadmax; ++o)
15535 PL_stashpad[o] = hv_dup(proto_perl->Istashpad[o], param);
15538 /* shortcuts to various I/O objects */
15539 PL_ofsgv = gv_dup_inc(proto_perl->Iofsgv, param);
15540 PL_stdingv = gv_dup(proto_perl->Istdingv, param);
15541 PL_stderrgv = gv_dup(proto_perl->Istderrgv, param);
15542 PL_defgv = gv_dup(proto_perl->Idefgv, param);
15543 PL_argvgv = gv_dup_inc(proto_perl->Iargvgv, param);
15544 PL_argvoutgv = gv_dup(proto_perl->Iargvoutgv, param);
15545 PL_argvout_stack = av_dup_inc(proto_perl->Iargvout_stack, param);
15547 /* shortcuts to regexp stuff */
15548 PL_replgv = gv_dup_inc(proto_perl->Ireplgv, param);
15550 /* shortcuts to misc objects */
15551 PL_errgv = gv_dup(proto_perl->Ierrgv, param);
15553 /* shortcuts to debugging objects */
15554 PL_DBgv = gv_dup_inc(proto_perl->IDBgv, param);
15555 PL_DBline = gv_dup_inc(proto_perl->IDBline, param);
15556 PL_DBsub = gv_dup_inc(proto_perl->IDBsub, param);
15557 PL_DBsingle = sv_dup(proto_perl->IDBsingle, param);
15558 PL_DBtrace = sv_dup(proto_perl->IDBtrace, param);
15559 PL_DBsignal = sv_dup(proto_perl->IDBsignal, param);
15560 Copy(proto_perl->IDBcontrol, PL_DBcontrol, DBVARMG_COUNT, IV);
15562 /* symbol tables */
15563 PL_defstash = hv_dup_inc(proto_perl->Idefstash, param);
15564 PL_curstash = hv_dup_inc(proto_perl->Icurstash, param);
15565 PL_debstash = hv_dup(proto_perl->Idebstash, param);
15566 PL_globalstash = hv_dup(proto_perl->Iglobalstash, param);
15567 PL_curstname = sv_dup_inc(proto_perl->Icurstname, param);
15569 PL_beginav = av_dup_inc(proto_perl->Ibeginav, param);
15570 PL_beginav_save = av_dup_inc(proto_perl->Ibeginav_save, param);
15571 PL_checkav_save = av_dup_inc(proto_perl->Icheckav_save, param);
15572 PL_unitcheckav = av_dup_inc(proto_perl->Iunitcheckav, param);
15573 PL_unitcheckav_save = av_dup_inc(proto_perl->Iunitcheckav_save, param);
15574 PL_endav = av_dup_inc(proto_perl->Iendav, param);
15575 PL_checkav = av_dup_inc(proto_perl->Icheckav, param);
15576 PL_initav = av_dup_inc(proto_perl->Iinitav, param);
15577 PL_savebegin = proto_perl->Isavebegin;
15579 PL_isarev = hv_dup_inc(proto_perl->Iisarev, param);
15581 /* subprocess state */
15582 PL_fdpid = av_dup_inc(proto_perl->Ifdpid, param);
15584 if (proto_perl->Iop_mask)
15585 PL_op_mask = SAVEPVN(proto_perl->Iop_mask, PL_maxo);
15588 /* PL_asserting = proto_perl->Iasserting; */
15590 /* current interpreter roots */
15591 PL_main_cv = cv_dup_inc(proto_perl->Imain_cv, param);
15593 PL_main_root = OpREFCNT_inc(proto_perl->Imain_root);
15596 /* runtime control stuff */
15597 PL_curcopdb = (COP*)any_dup(proto_perl->Icurcopdb, proto_perl);
15599 PL_preambleav = av_dup_inc(proto_perl->Ipreambleav, param);
15601 PL_ors_sv = sv_dup_inc(proto_perl->Iors_sv, param);
15603 /* interpreter atexit processing */
15604 PL_exitlistlen = proto_perl->Iexitlistlen;
15605 if (PL_exitlistlen) {
15606 Newx(PL_exitlist, PL_exitlistlen, PerlExitListEntry);
15607 Copy(proto_perl->Iexitlist, PL_exitlist, PL_exitlistlen, PerlExitListEntry);
15610 PL_exitlist = (PerlExitListEntry*)NULL;
15612 PL_my_cxt_size = proto_perl->Imy_cxt_size;
15613 if (PL_my_cxt_size) {
15614 Newx(PL_my_cxt_list, PL_my_cxt_size, void *);
15615 Copy(proto_perl->Imy_cxt_list, PL_my_cxt_list, PL_my_cxt_size, void *);
15616 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
15617 Newx(PL_my_cxt_keys, PL_my_cxt_size, const char *);
15618 Copy(proto_perl->Imy_cxt_keys, PL_my_cxt_keys, PL_my_cxt_size, char *);
15622 PL_my_cxt_list = (void**)NULL;
15623 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
15624 PL_my_cxt_keys = (const char**)NULL;
15627 PL_modglobal = hv_dup_inc(proto_perl->Imodglobal, param);
15628 PL_custom_op_names = hv_dup_inc(proto_perl->Icustom_op_names,param);
15629 PL_custom_op_descs = hv_dup_inc(proto_perl->Icustom_op_descs,param);
15630 PL_custom_ops = hv_dup_inc(proto_perl->Icustom_ops, param);
15632 PL_compcv = cv_dup(proto_perl->Icompcv, param);
15634 PAD_CLONE_VARS(proto_perl, param);
15636 #ifdef HAVE_INTERP_INTERN
15637 sys_intern_dup(&proto_perl->Isys_intern, &PL_sys_intern);
15640 PL_DBcv = cv_dup(proto_perl->IDBcv, param);
15642 #ifdef PERL_USES_PL_PIDSTATUS
15643 PL_pidstatus = newHV(); /* XXX flag for cloning? */
15645 PL_osname = SAVEPV(proto_perl->Iosname);
15646 PL_parser = parser_dup(proto_perl->Iparser, param);
15648 /* XXX this only works if the saved cop has already been cloned */
15649 if (proto_perl->Iparser) {
15650 PL_parser->saved_curcop = (COP*)any_dup(
15651 proto_perl->Iparser->saved_curcop,
15655 PL_subname = sv_dup_inc(proto_perl->Isubname, param);
15657 #ifdef USE_LOCALE_CTYPE
15658 /* Should we warn if uses locale? */
15659 PL_warn_locale = sv_dup_inc(proto_perl->Iwarn_locale, param);
15662 #ifdef USE_LOCALE_COLLATE
15663 PL_collation_name = SAVEPV(proto_perl->Icollation_name);
15664 #endif /* USE_LOCALE_COLLATE */
15666 #ifdef USE_LOCALE_NUMERIC
15667 PL_numeric_name = SAVEPV(proto_perl->Inumeric_name);
15668 PL_numeric_radix_sv = sv_dup_inc(proto_perl->Inumeric_radix_sv, param);
15669 #endif /* !USE_LOCALE_NUMERIC */
15671 PL_langinfo_buf = NULL;
15672 PL_langinfo_bufsize = 0;
15674 /* Unicode inversion lists */
15675 PL_Latin1 = sv_dup_inc(proto_perl->ILatin1, param);
15676 PL_UpperLatin1 = sv_dup_inc(proto_perl->IUpperLatin1, param);
15677 PL_AboveLatin1 = sv_dup_inc(proto_perl->IAboveLatin1, param);
15678 PL_InBitmap = sv_dup_inc(proto_perl->IInBitmap, param);
15680 PL_NonL1NonFinalFold = sv_dup_inc(proto_perl->INonL1NonFinalFold, param);
15681 PL_HasMultiCharFold = sv_dup_inc(proto_perl->IHasMultiCharFold, param);
15683 /* utf8 character class swashes */
15684 for (i = 0; i < POSIX_SWASH_COUNT; i++) {
15685 PL_utf8_swash_ptrs[i] = sv_dup_inc(proto_perl->Iutf8_swash_ptrs[i], param);
15687 for (i = 0; i < POSIX_CC_COUNT; i++) {
15688 PL_XPosix_ptrs[i] = sv_dup_inc(proto_perl->IXPosix_ptrs[i], param);
15690 PL_GCB_invlist = sv_dup_inc(proto_perl->IGCB_invlist, param);
15691 PL_SB_invlist = sv_dup_inc(proto_perl->ISB_invlist, param);
15692 PL_WB_invlist = sv_dup_inc(proto_perl->IWB_invlist, param);
15693 PL_seen_deprecated_macro = hv_dup_inc(proto_perl->Iseen_deprecated_macro, param);
15694 PL_utf8_mark = sv_dup_inc(proto_perl->Iutf8_mark, param);
15695 PL_utf8_toupper = sv_dup_inc(proto_perl->Iutf8_toupper, param);
15696 PL_utf8_totitle = sv_dup_inc(proto_perl->Iutf8_totitle, param);
15697 PL_utf8_tolower = sv_dup_inc(proto_perl->Iutf8_tolower, param);
15698 PL_utf8_tofold = sv_dup_inc(proto_perl->Iutf8_tofold, param);
15699 PL_utf8_idstart = sv_dup_inc(proto_perl->Iutf8_idstart, param);
15700 PL_utf8_xidstart = sv_dup_inc(proto_perl->Iutf8_xidstart, param);
15701 PL_utf8_perl_idstart = sv_dup_inc(proto_perl->Iutf8_perl_idstart, param);
15702 PL_utf8_perl_idcont = sv_dup_inc(proto_perl->Iutf8_perl_idcont, param);
15703 PL_utf8_idcont = sv_dup_inc(proto_perl->Iutf8_idcont, param);
15704 PL_utf8_xidcont = sv_dup_inc(proto_perl->Iutf8_xidcont, param);
15705 PL_utf8_foldable = sv_dup_inc(proto_perl->Iutf8_foldable, param);
15706 PL_utf8_charname_begin = sv_dup_inc(proto_perl->Iutf8_charname_begin, param);
15707 PL_utf8_charname_continue = sv_dup_inc(proto_perl->Iutf8_charname_continue, param);
15709 if (proto_perl->Ipsig_pend) {
15710 Newxz(PL_psig_pend, SIG_SIZE, int);
15713 PL_psig_pend = (int*)NULL;
15716 if (proto_perl->Ipsig_name) {
15717 Newx(PL_psig_name, 2 * SIG_SIZE, SV*);
15718 sv_dup_inc_multiple(proto_perl->Ipsig_name, PL_psig_name, 2 * SIG_SIZE,
15720 PL_psig_ptr = PL_psig_name + SIG_SIZE;
15723 PL_psig_ptr = (SV**)NULL;
15724 PL_psig_name = (SV**)NULL;
15727 if (flags & CLONEf_COPY_STACKS) {
15728 Newx(PL_tmps_stack, PL_tmps_max, SV*);
15729 sv_dup_inc_multiple(proto_perl->Itmps_stack, PL_tmps_stack,
15730 PL_tmps_ix+1, param);
15732 /* next PUSHMARK() sets *(PL_markstack_ptr+1) */
15733 i = proto_perl->Imarkstack_max - proto_perl->Imarkstack;
15734 Newxz(PL_markstack, i, I32);
15735 PL_markstack_max = PL_markstack + (proto_perl->Imarkstack_max
15736 - proto_perl->Imarkstack);
15737 PL_markstack_ptr = PL_markstack + (proto_perl->Imarkstack_ptr
15738 - proto_perl->Imarkstack);
15739 Copy(proto_perl->Imarkstack, PL_markstack,
15740 PL_markstack_ptr - PL_markstack + 1, I32);
15742 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
15743 * NOTE: unlike the others! */
15744 Newxz(PL_scopestack, PL_scopestack_max, I32);
15745 Copy(proto_perl->Iscopestack, PL_scopestack, PL_scopestack_ix, I32);
15748 Newxz(PL_scopestack_name, PL_scopestack_max, const char *);
15749 Copy(proto_perl->Iscopestack_name, PL_scopestack_name, PL_scopestack_ix, const char *);
15751 /* reset stack AV to correct length before its duped via
15752 * PL_curstackinfo */
15753 AvFILLp(proto_perl->Icurstack) =
15754 proto_perl->Istack_sp - proto_perl->Istack_base;
15756 /* NOTE: si_dup() looks at PL_markstack */
15757 PL_curstackinfo = si_dup(proto_perl->Icurstackinfo, param);
15759 /* PL_curstack = PL_curstackinfo->si_stack; */
15760 PL_curstack = av_dup(proto_perl->Icurstack, param);
15761 PL_mainstack = av_dup(proto_perl->Imainstack, param);
15763 /* next PUSHs() etc. set *(PL_stack_sp+1) */
15764 PL_stack_base = AvARRAY(PL_curstack);
15765 PL_stack_sp = PL_stack_base + (proto_perl->Istack_sp
15766 - proto_perl->Istack_base);
15767 PL_stack_max = PL_stack_base + AvMAX(PL_curstack);
15769 /*Newxz(PL_savestack, PL_savestack_max, ANY);*/
15770 PL_savestack = ss_dup(proto_perl, param);
15774 ENTER; /* perl_destruct() wants to LEAVE; */
15777 PL_statgv = gv_dup(proto_perl->Istatgv, param);
15778 PL_statname = sv_dup_inc(proto_perl->Istatname, param);
15780 PL_rs = sv_dup_inc(proto_perl->Irs, param);
15781 PL_last_in_gv = gv_dup(proto_perl->Ilast_in_gv, param);
15782 PL_defoutgv = gv_dup_inc(proto_perl->Idefoutgv, param);
15783 PL_toptarget = sv_dup_inc(proto_perl->Itoptarget, param);
15784 PL_bodytarget = sv_dup_inc(proto_perl->Ibodytarget, param);
15785 PL_formtarget = sv_dup(proto_perl->Iformtarget, param);
15787 PL_errors = sv_dup_inc(proto_perl->Ierrors, param);
15789 PL_sortcop = (OP*)any_dup(proto_perl->Isortcop, proto_perl);
15790 PL_firstgv = gv_dup_inc(proto_perl->Ifirstgv, param);
15791 PL_secondgv = gv_dup_inc(proto_perl->Isecondgv, param);
15793 PL_stashcache = newHV();
15795 PL_watchaddr = (char **) ptr_table_fetch(PL_ptr_table,
15796 proto_perl->Iwatchaddr);
15797 PL_watchok = PL_watchaddr ? * PL_watchaddr : NULL;
15798 if (PL_debug && PL_watchaddr) {
15799 PerlIO_printf(Perl_debug_log,
15800 "WATCHING: %" UVxf " cloned as %" UVxf " with value %" UVxf "\n",
15801 PTR2UV(proto_perl->Iwatchaddr), PTR2UV(PL_watchaddr),
15802 PTR2UV(PL_watchok));
15805 PL_registered_mros = hv_dup_inc(proto_perl->Iregistered_mros, param);
15806 PL_blockhooks = av_dup_inc(proto_perl->Iblockhooks, param);
15807 PL_utf8_foldclosures = hv_dup_inc(proto_perl->Iutf8_foldclosures, param);
15809 /* Call the ->CLONE method, if it exists, for each of the stashes
15810 identified by sv_dup() above.
15812 while(av_tindex(param->stashes) != -1) {
15813 HV* const stash = MUTABLE_HV(av_shift(param->stashes));
15814 GV* const cloner = gv_fetchmethod_autoload(stash, "CLONE", 0);
15815 if (cloner && GvCV(cloner)) {
15820 mXPUSHs(newSVhek(HvNAME_HEK(stash)));
15822 call_sv(MUTABLE_SV(GvCV(cloner)), G_DISCARD);
15828 if (!(flags & CLONEf_KEEP_PTR_TABLE)) {
15829 ptr_table_free(PL_ptr_table);
15830 PL_ptr_table = NULL;
15833 if (!(flags & CLONEf_COPY_STACKS)) {
15834 unreferenced_to_tmp_stack(param->unreferenced);
15837 SvREFCNT_dec(param->stashes);
15839 /* orphaned? eg threads->new inside BEGIN or use */
15840 if (PL_compcv && ! SvREFCNT(PL_compcv)) {
15841 SvREFCNT_inc_simple_void(PL_compcv);
15842 SAVEFREESV(PL_compcv);
15849 S_unreferenced_to_tmp_stack(pTHX_ AV *const unreferenced)
15851 PERL_ARGS_ASSERT_UNREFERENCED_TO_TMP_STACK;
15853 if (AvFILLp(unreferenced) > -1) {
15854 SV **svp = AvARRAY(unreferenced);
15855 SV **const last = svp + AvFILLp(unreferenced);
15859 if (SvREFCNT(*svp) == 1)
15861 } while (++svp <= last);
15863 EXTEND_MORTAL(count);
15864 svp = AvARRAY(unreferenced);
15867 if (SvREFCNT(*svp) == 1) {
15868 /* Our reference is the only one to this SV. This means that
15869 in this thread, the scalar effectively has a 0 reference.
15870 That doesn't work (cleanup never happens), so donate our
15871 reference to it onto the save stack. */
15872 PL_tmps_stack[++PL_tmps_ix] = *svp;
15874 /* As an optimisation, because we are already walking the
15875 entire array, instead of above doing either
15876 SvREFCNT_inc(*svp) or *svp = &PL_sv_undef, we can instead
15877 release our reference to the scalar, so that at the end of
15878 the array owns zero references to the scalars it happens to
15879 point to. We are effectively converting the array from
15880 AvREAL() on to AvREAL() off. This saves the av_clear()
15881 (triggered by the SvREFCNT_dec(unreferenced) below) from
15882 walking the array a second time. */
15883 SvREFCNT_dec(*svp);
15886 } while (++svp <= last);
15887 AvREAL_off(unreferenced);
15889 SvREFCNT_dec_NN(unreferenced);
15893 Perl_clone_params_del(CLONE_PARAMS *param)
15895 /* This seemingly funky ordering keeps the build with PERL_GLOBAL_STRUCT
15897 PerlInterpreter *const to = param->new_perl;
15899 PerlInterpreter *const was = PERL_GET_THX;
15901 PERL_ARGS_ASSERT_CLONE_PARAMS_DEL;
15907 SvREFCNT_dec(param->stashes);
15908 if (param->unreferenced)
15909 unreferenced_to_tmp_stack(param->unreferenced);
15919 Perl_clone_params_new(PerlInterpreter *const from, PerlInterpreter *const to)
15922 /* Need to play this game, as newAV() can call safesysmalloc(), and that
15923 does a dTHX; to get the context from thread local storage.
15924 FIXME - under PERL_CORE Newx(), Safefree() and friends should expand to
15925 a version that passes in my_perl. */
15926 PerlInterpreter *const was = PERL_GET_THX;
15927 CLONE_PARAMS *param;
15929 PERL_ARGS_ASSERT_CLONE_PARAMS_NEW;
15935 /* Given that we've set the context, we can do this unshared. */
15936 Newx(param, 1, CLONE_PARAMS);
15939 param->proto_perl = from;
15940 param->new_perl = to;
15941 param->stashes = (AV *)Perl_newSV_type(to, SVt_PVAV);
15942 AvREAL_off(param->stashes);
15943 param->unreferenced = (AV *)Perl_newSV_type(to, SVt_PVAV);
15951 #endif /* USE_ITHREADS */
15954 Perl_init_constants(pTHX)
15956 SvREFCNT(&PL_sv_undef) = SvREFCNT_IMMORTAL;
15957 SvFLAGS(&PL_sv_undef) = SVf_READONLY|SVf_PROTECT|SVt_NULL;
15958 SvANY(&PL_sv_undef) = NULL;
15960 SvANY(&PL_sv_no) = new_XPVNV();
15961 SvREFCNT(&PL_sv_no) = SvREFCNT_IMMORTAL;
15962 SvFLAGS(&PL_sv_no) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15963 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15966 SvANY(&PL_sv_yes) = new_XPVNV();
15967 SvREFCNT(&PL_sv_yes) = SvREFCNT_IMMORTAL;
15968 SvFLAGS(&PL_sv_yes) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15969 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15972 SvANY(&PL_sv_zero) = new_XPVNV();
15973 SvREFCNT(&PL_sv_zero) = SvREFCNT_IMMORTAL;
15974 SvFLAGS(&PL_sv_zero) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15975 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15979 SvPV_set(&PL_sv_no, (char*)PL_No);
15980 SvCUR_set(&PL_sv_no, 0);
15981 SvLEN_set(&PL_sv_no, 0);
15982 SvIV_set(&PL_sv_no, 0);
15983 SvNV_set(&PL_sv_no, 0);
15985 SvPV_set(&PL_sv_yes, (char*)PL_Yes);
15986 SvCUR_set(&PL_sv_yes, 1);
15987 SvLEN_set(&PL_sv_yes, 0);
15988 SvIV_set(&PL_sv_yes, 1);
15989 SvNV_set(&PL_sv_yes, 1);
15991 SvPV_set(&PL_sv_zero, (char*)PL_Zero);
15992 SvCUR_set(&PL_sv_zero, 1);
15993 SvLEN_set(&PL_sv_zero, 0);
15994 SvIV_set(&PL_sv_zero, 0);
15995 SvNV_set(&PL_sv_zero, 0);
15997 PadnamePV(&PL_padname_const) = (char *)PL_No;
15999 assert(SvIMMORTAL_INTERP(&PL_sv_yes));
16000 assert(SvIMMORTAL_INTERP(&PL_sv_undef));
16001 assert(SvIMMORTAL_INTERP(&PL_sv_no));
16002 assert(SvIMMORTAL_INTERP(&PL_sv_zero));
16004 assert(SvIMMORTAL(&PL_sv_yes));
16005 assert(SvIMMORTAL(&PL_sv_undef));
16006 assert(SvIMMORTAL(&PL_sv_no));
16007 assert(SvIMMORTAL(&PL_sv_zero));
16009 assert( SvIMMORTAL_TRUE(&PL_sv_yes));
16010 assert(!SvIMMORTAL_TRUE(&PL_sv_undef));
16011 assert(!SvIMMORTAL_TRUE(&PL_sv_no));
16012 assert(!SvIMMORTAL_TRUE(&PL_sv_zero));
16014 assert( SvTRUE_nomg_NN(&PL_sv_yes));
16015 assert(!SvTRUE_nomg_NN(&PL_sv_undef));
16016 assert(!SvTRUE_nomg_NN(&PL_sv_no));
16017 assert(!SvTRUE_nomg_NN(&PL_sv_zero));
16021 =head1 Unicode Support
16023 =for apidoc sv_recode_to_utf8
16025 C<encoding> is assumed to be an C<Encode> object, on entry the PV
16026 of C<sv> is assumed to be octets in that encoding, and C<sv>
16027 will be converted into Unicode (and UTF-8).
16029 If C<sv> already is UTF-8 (or if it is not C<POK>), or if C<encoding>
16030 is not a reference, nothing is done to C<sv>. If C<encoding> is not
16031 an C<Encode::XS> Encoding object, bad things will happen.
16032 (See F<cpan/Encode/encoding.pm> and L<Encode>.)
16034 The PV of C<sv> is returned.
16039 Perl_sv_recode_to_utf8(pTHX_ SV *sv, SV *encoding)
16041 PERL_ARGS_ASSERT_SV_RECODE_TO_UTF8;
16043 if (SvPOK(sv) && !SvUTF8(sv) && !IN_BYTES && SvROK(encoding)) {
16052 if (SvPADTMP(nsv)) {
16053 nsv = sv_newmortal();
16054 SvSetSV_nosteal(nsv, sv);
16063 Passing sv_yes is wrong - it needs to be or'ed set of constants
16064 for Encode::XS, while UTf-8 decode (currently) assumes a true value means
16065 remove converted chars from source.
16067 Both will default the value - let them.
16069 XPUSHs(&PL_sv_yes);
16072 call_method("decode", G_SCALAR);
16076 s = SvPV_const(uni, len);
16077 if (s != SvPVX_const(sv)) {
16078 SvGROW(sv, len + 1);
16079 Move(s, SvPVX(sv), len + 1, char);
16080 SvCUR_set(sv, len);
16085 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
16086 /* clear pos and any utf8 cache */
16087 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
16090 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
16091 magic_setutf8(sv,mg); /* clear UTF8 cache */
16096 return SvPOKp(sv) ? SvPVX(sv) : NULL;
16100 =for apidoc sv_cat_decode
16102 C<encoding> is assumed to be an C<Encode> object, the PV of C<ssv> is
16103 assumed to be octets in that encoding and decoding the input starts
16104 from the position which S<C<(PV + *offset)>> pointed to. C<dsv> will be
16105 concatenated with the decoded UTF-8 string from C<ssv>. Decoding will terminate
16106 when the string C<tstr> appears in decoding output or the input ends on
16107 the PV of C<ssv>. The value which C<offset> points will be modified
16108 to the last input position on C<ssv>.
16110 Returns TRUE if the terminator was found, else returns FALSE.
16115 Perl_sv_cat_decode(pTHX_ SV *dsv, SV *encoding,
16116 SV *ssv, int *offset, char *tstr, int tlen)
16120 PERL_ARGS_ASSERT_SV_CAT_DECODE;
16122 if (SvPOK(ssv) && SvPOK(dsv) && SvROK(encoding)) {
16133 offsv = newSViv(*offset);
16135 mPUSHp(tstr, tlen);
16137 call_method("cat_decode", G_SCALAR);
16139 ret = SvTRUE(TOPs);
16140 *offset = SvIV(offsv);
16146 Perl_croak(aTHX_ "Invalid argument to sv_cat_decode");
16151 /* ---------------------------------------------------------------------
16153 * support functions for report_uninit()
16156 /* the maxiumum size of array or hash where we will scan looking
16157 * for the undefined element that triggered the warning */
16159 #define FUV_MAX_SEARCH_SIZE 1000
16161 /* Look for an entry in the hash whose value has the same SV as val;
16162 * If so, return a mortal copy of the key. */
16165 S_find_hash_subscript(pTHX_ const HV *const hv, const SV *const val)
16171 PERL_ARGS_ASSERT_FIND_HASH_SUBSCRIPT;
16173 if (!hv || SvMAGICAL(hv) || !HvARRAY(hv) ||
16174 (HvTOTALKEYS(hv) > FUV_MAX_SEARCH_SIZE))
16177 array = HvARRAY(hv);
16179 for (i=HvMAX(hv); i>=0; i--) {
16181 for (entry = array[i]; entry; entry = HeNEXT(entry)) {
16182 if (HeVAL(entry) != val)
16184 if ( HeVAL(entry) == &PL_sv_undef ||
16185 HeVAL(entry) == &PL_sv_placeholder)
16189 if (HeKLEN(entry) == HEf_SVKEY)
16190 return sv_mortalcopy(HeKEY_sv(entry));
16191 return sv_2mortal(newSVhek(HeKEY_hek(entry)));
16197 /* Look for an entry in the array whose value has the same SV as val;
16198 * If so, return the index, otherwise return -1. */
16201 S_find_array_subscript(pTHX_ const AV *const av, const SV *const val)
16203 PERL_ARGS_ASSERT_FIND_ARRAY_SUBSCRIPT;
16205 if (!av || SvMAGICAL(av) || !AvARRAY(av) ||
16206 (AvFILLp(av) > FUV_MAX_SEARCH_SIZE))
16209 if (val != &PL_sv_undef) {
16210 SV ** const svp = AvARRAY(av);
16213 for (i=AvFILLp(av); i>=0; i--)
16220 /* varname(): return the name of a variable, optionally with a subscript.
16221 * If gv is non-zero, use the name of that global, along with gvtype (one
16222 * of "$", "@", "%"); otherwise use the name of the lexical at pad offset
16223 * targ. Depending on the value of the subscript_type flag, return:
16226 #define FUV_SUBSCRIPT_NONE 1 /* "@foo" */
16227 #define FUV_SUBSCRIPT_ARRAY 2 /* "$foo[aindex]" */
16228 #define FUV_SUBSCRIPT_HASH 3 /* "$foo{keyname}" */
16229 #define FUV_SUBSCRIPT_WITHIN 4 /* "within @foo" */
16232 Perl_varname(pTHX_ const GV *const gv, const char gvtype, PADOFFSET targ,
16233 const SV *const keyname, SSize_t aindex, int subscript_type)
16236 SV * const name = sv_newmortal();
16237 if (gv && isGV(gv)) {
16239 buffer[0] = gvtype;
16242 /* as gv_fullname4(), but add literal '^' for $^FOO names */
16244 gv_fullname4(name, gv, buffer, 0);
16246 if ((unsigned int)SvPVX(name)[1] <= 26) {
16248 buffer[1] = SvPVX(name)[1] + 'A' - 1;
16250 /* Swap the 1 unprintable control character for the 2 byte pretty
16251 version - ie substr($name, 1, 1) = $buffer; */
16252 sv_insert(name, 1, 1, buffer, 2);
16256 CV * const cv = gv ? ((CV *)gv) : find_runcv(NULL);
16259 assert(!cv || SvTYPE(cv) == SVt_PVCV || SvTYPE(cv) == SVt_PVFM);
16261 if (!cv || !CvPADLIST(cv))
16263 sv = padnamelist_fetch(PadlistNAMES(CvPADLIST(cv)), targ);
16264 sv_setpvn(name, PadnamePV(sv), PadnameLEN(sv));
16268 if (subscript_type == FUV_SUBSCRIPT_HASH) {
16269 SV * const sv = newSV(0);
16271 const char * const pv = SvPV_nomg_const((SV*)keyname, len);
16273 *SvPVX(name) = '$';
16274 Perl_sv_catpvf(aTHX_ name, "{%s}",
16275 pv_pretty(sv, pv, len, 32, NULL, NULL,
16276 PERL_PV_PRETTY_DUMP | PERL_PV_ESCAPE_UNI_DETECT ));
16277 SvREFCNT_dec_NN(sv);
16279 else if (subscript_type == FUV_SUBSCRIPT_ARRAY) {
16280 *SvPVX(name) = '$';
16281 Perl_sv_catpvf(aTHX_ name, "[%" IVdf "]", (IV)aindex);
16283 else if (subscript_type == FUV_SUBSCRIPT_WITHIN) {
16284 /* We know that name has no magic, so can use 0 instead of SV_GMAGIC */
16285 Perl_sv_insert_flags(aTHX_ name, 0, 0, STR_WITH_LEN("within "), 0);
16293 =for apidoc find_uninit_var
16295 Find the name of the undefined variable (if any) that caused the operator
16296 to issue a "Use of uninitialized value" warning.
16297 If match is true, only return a name if its value matches C<uninit_sv>.
16298 So roughly speaking, if a unary operator (such as C<OP_COS>) generates a
16299 warning, then following the direct child of the op may yield an
16300 C<OP_PADSV> or C<OP_GV> that gives the name of the undefined variable. On the
16301 other hand, with C<OP_ADD> there are two branches to follow, so we only print
16302 the variable name if we get an exact match.
16303 C<desc_p> points to a string pointer holding the description of the op.
16304 This may be updated if needed.
16306 The name is returned as a mortal SV.
16308 Assumes that C<PL_op> is the OP that originally triggered the error, and that
16309 C<PL_comppad>/C<PL_curpad> points to the currently executing pad.
16315 S_find_uninit_var(pTHX_ const OP *const obase, const SV *const uninit_sv,
16316 bool match, const char **desc_p)
16321 const OP *o, *o2, *kid;
16323 PERL_ARGS_ASSERT_FIND_UNINIT_VAR;
16325 if (!obase || (match && (!uninit_sv || uninit_sv == &PL_sv_undef ||
16326 uninit_sv == &PL_sv_placeholder)))
16329 switch (obase->op_type) {
16332 /* undef should care if its args are undef - any warnings
16333 * will be from tied/magic vars */
16341 const bool pad = ( obase->op_type == OP_PADAV
16342 || obase->op_type == OP_PADHV
16343 || obase->op_type == OP_PADRANGE
16346 const bool hash = ( obase->op_type == OP_PADHV
16347 || obase->op_type == OP_RV2HV
16348 || (obase->op_type == OP_PADRANGE
16349 && SvTYPE(PAD_SVl(obase->op_targ)) == SVt_PVHV)
16353 int subscript_type = FUV_SUBSCRIPT_WITHIN;
16355 if (pad) { /* @lex, %lex */
16356 sv = PAD_SVl(obase->op_targ);
16360 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
16361 /* @global, %global */
16362 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
16365 sv = hash ? MUTABLE_SV(GvHV(gv)): MUTABLE_SV(GvAV(gv));
16367 else if (obase == PL_op) /* @{expr}, %{expr} */
16368 return find_uninit_var(cUNOPx(obase)->op_first,
16369 uninit_sv, match, desc_p);
16370 else /* @{expr}, %{expr} as a sub-expression */
16374 /* attempt to find a match within the aggregate */
16376 keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16378 subscript_type = FUV_SUBSCRIPT_HASH;
16381 index = find_array_subscript((const AV *)sv, uninit_sv);
16383 subscript_type = FUV_SUBSCRIPT_ARRAY;
16386 if (match && subscript_type == FUV_SUBSCRIPT_WITHIN)
16389 return varname(gv, (char)(hash ? '%' : '@'), obase->op_targ,
16390 keysv, index, subscript_type);
16394 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
16396 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
16397 if (!gv || !GvSTASH(gv))
16399 if (match && (GvSV(gv) != uninit_sv))
16401 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
16404 return find_uninit_var(cUNOPx(obase)->op_first, uninit_sv, 1, desc_p);
16407 if (match && PAD_SVl(obase->op_targ) != uninit_sv)
16409 return varname(NULL, '$', obase->op_targ,
16410 NULL, 0, FUV_SUBSCRIPT_NONE);
16413 gv = cGVOPx_gv(obase);
16414 if (!gv || (match && GvSV(gv) != uninit_sv) || !GvSTASH(gv))
16416 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
16418 case OP_AELEMFAST_LEX:
16421 AV *av = MUTABLE_AV(PAD_SV(obase->op_targ));
16422 if (!av || SvRMAGICAL(av))
16424 svp = av_fetch(av, (I8)obase->op_private, FALSE);
16425 if (!svp || *svp != uninit_sv)
16428 return varname(NULL, '$', obase->op_targ,
16429 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
16432 gv = cGVOPx_gv(obase);
16437 AV *const av = GvAV(gv);
16438 if (!av || SvRMAGICAL(av))
16440 svp = av_fetch(av, (I8)obase->op_private, FALSE);
16441 if (!svp || *svp != uninit_sv)
16444 return varname(gv, '$', 0,
16445 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
16447 NOT_REACHED; /* NOTREACHED */
16450 o = cUNOPx(obase)->op_first;
16451 if (!o || o->op_type != OP_NULL ||
16452 ! (o->op_targ == OP_AELEM || o->op_targ == OP_HELEM))
16454 return find_uninit_var(cBINOPo->op_last, uninit_sv, match, desc_p);
16459 bool negate = FALSE;
16461 if (PL_op == obase)
16462 /* $a[uninit_expr] or $h{uninit_expr} */
16463 return find_uninit_var(cBINOPx(obase)->op_last,
16464 uninit_sv, match, desc_p);
16467 o = cBINOPx(obase)->op_first;
16468 kid = cBINOPx(obase)->op_last;
16470 /* get the av or hv, and optionally the gv */
16472 if (o->op_type == OP_PADAV || o->op_type == OP_PADHV) {
16473 sv = PAD_SV(o->op_targ);
16475 else if ((o->op_type == OP_RV2AV || o->op_type == OP_RV2HV)
16476 && cUNOPo->op_first->op_type == OP_GV)
16478 gv = cGVOPx_gv(cUNOPo->op_first);
16482 == OP_RV2HV ? MUTABLE_SV(GvHV(gv)) : MUTABLE_SV(GvAV(gv));
16487 if (kid && kid->op_type == OP_NEGATE) {
16489 kid = cUNOPx(kid)->op_first;
16492 if (kid && kid->op_type == OP_CONST && SvOK(cSVOPx_sv(kid))) {
16493 /* index is constant */
16496 kidsv = newSVpvs_flags("-", SVs_TEMP);
16497 sv_catsv(kidsv, cSVOPx_sv(kid));
16500 kidsv = cSVOPx_sv(kid);
16504 if (obase->op_type == OP_HELEM) {
16505 HE* he = hv_fetch_ent(MUTABLE_HV(sv), kidsv, 0, 0);
16506 if (!he || HeVAL(he) != uninit_sv)
16510 SV * const opsv = cSVOPx_sv(kid);
16511 const IV opsviv = SvIV(opsv);
16512 SV * const * const svp = av_fetch(MUTABLE_AV(sv),
16513 negate ? - opsviv : opsviv,
16515 if (!svp || *svp != uninit_sv)
16519 if (obase->op_type == OP_HELEM)
16520 return varname(gv, '%', o->op_targ,
16521 kidsv, 0, FUV_SUBSCRIPT_HASH);
16523 return varname(gv, '@', o->op_targ, NULL,
16524 negate ? - SvIV(cSVOPx_sv(kid)) : SvIV(cSVOPx_sv(kid)),
16525 FUV_SUBSCRIPT_ARRAY);
16528 /* index is an expression;
16529 * attempt to find a match within the aggregate */
16530 if (obase->op_type == OP_HELEM) {
16531 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16533 return varname(gv, '%', o->op_targ,
16534 keysv, 0, FUV_SUBSCRIPT_HASH);
16537 const SSize_t index
16538 = find_array_subscript((const AV *)sv, uninit_sv);
16540 return varname(gv, '@', o->op_targ,
16541 NULL, index, FUV_SUBSCRIPT_ARRAY);
16546 (char)((o->op_type == OP_PADAV || o->op_type == OP_RV2AV)
16548 o->op_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
16550 NOT_REACHED; /* NOTREACHED */
16553 case OP_MULTIDEREF: {
16554 /* If we were executing OP_MULTIDEREF when the undef warning
16555 * triggered, then it must be one of the index values within
16556 * that triggered it. If not, then the only possibility is that
16557 * the value retrieved by the last aggregate index might be the
16558 * culprit. For the former, we set PL_multideref_pc each time before
16559 * using an index, so work though the item list until we reach
16560 * that point. For the latter, just work through the entire item
16561 * list; the last aggregate retrieved will be the candidate.
16562 * There is a third rare possibility: something triggered
16563 * magic while fetching an array/hash element. Just display
16564 * nothing in this case.
16567 /* the named aggregate, if any */
16568 PADOFFSET agg_targ = 0;
16570 /* the last-seen index */
16572 PADOFFSET index_targ;
16574 IV index_const_iv = 0; /* init for spurious compiler warn */
16575 SV *index_const_sv;
16576 int depth = 0; /* how many array/hash lookups we've done */
16578 UNOP_AUX_item *items = cUNOP_AUXx(obase)->op_aux;
16579 UNOP_AUX_item *last = NULL;
16580 UV actions = items->uv;
16583 if (PL_op == obase) {
16584 last = PL_multideref_pc;
16585 assert(last >= items && last <= items + items[-1].uv);
16592 switch (actions & MDEREF_ACTION_MASK) {
16594 case MDEREF_reload:
16595 actions = (++items)->uv;
16598 case MDEREF_HV_padhv_helem: /* $lex{...} */
16601 case MDEREF_AV_padav_aelem: /* $lex[...] */
16602 agg_targ = (++items)->pad_offset;
16606 case MDEREF_HV_gvhv_helem: /* $pkg{...} */
16609 case MDEREF_AV_gvav_aelem: /* $pkg[...] */
16611 agg_gv = (GV*)UNOP_AUX_item_sv(++items);
16612 assert(isGV_with_GP(agg_gv));
16615 case MDEREF_HV_gvsv_vivify_rv2hv_helem: /* $pkg->{...} */
16616 case MDEREF_HV_padsv_vivify_rv2hv_helem: /* $lex->{...} */
16619 case MDEREF_HV_pop_rv2hv_helem: /* expr->{...} */
16620 case MDEREF_HV_vivify_rv2hv_helem: /* vivify, ->{...} */
16626 case MDEREF_AV_gvsv_vivify_rv2av_aelem: /* $pkg->[...] */
16627 case MDEREF_AV_padsv_vivify_rv2av_aelem: /* $lex->[...] */
16630 case MDEREF_AV_pop_rv2av_aelem: /* expr->[...] */
16631 case MDEREF_AV_vivify_rv2av_aelem: /* vivify, ->[...] */
16638 index_const_sv = NULL;
16640 index_type = (actions & MDEREF_INDEX_MASK);
16641 switch (index_type) {
16642 case MDEREF_INDEX_none:
16644 case MDEREF_INDEX_const:
16646 index_const_sv = UNOP_AUX_item_sv(++items)
16648 index_const_iv = (++items)->iv;
16650 case MDEREF_INDEX_padsv:
16651 index_targ = (++items)->pad_offset;
16653 case MDEREF_INDEX_gvsv:
16654 index_gv = (GV*)UNOP_AUX_item_sv(++items);
16655 assert(isGV_with_GP(index_gv));
16659 if (index_type != MDEREF_INDEX_none)
16662 if ( index_type == MDEREF_INDEX_none
16663 || (actions & MDEREF_FLAG_last)
16664 || (last && items >= last)
16668 actions >>= MDEREF_SHIFT;
16671 if (PL_op == obase) {
16672 /* most likely index was undef */
16674 *desc_p = ( (actions & MDEREF_FLAG_last)
16675 && (obase->op_private
16676 & (OPpMULTIDEREF_EXISTS|OPpMULTIDEREF_DELETE)))
16678 (obase->op_private & OPpMULTIDEREF_EXISTS)
16681 : is_hv ? "hash element" : "array element";
16682 assert(index_type != MDEREF_INDEX_none);
16684 if (GvSV(index_gv) == uninit_sv)
16685 return varname(index_gv, '$', 0, NULL, 0,
16686 FUV_SUBSCRIPT_NONE);
16691 if (PL_curpad[index_targ] == uninit_sv)
16692 return varname(NULL, '$', index_targ,
16693 NULL, 0, FUV_SUBSCRIPT_NONE);
16697 /* If we got to this point it was undef on a const subscript,
16698 * so magic probably involved, e.g. $ISA[0]. Give up. */
16702 /* the SV returned by pp_multideref() was undef, if anything was */
16708 sv = PAD_SV(agg_targ);
16710 sv = is_hv ? MUTABLE_SV(GvHV(agg_gv)) : MUTABLE_SV(GvAV(agg_gv));
16714 if (index_type == MDEREF_INDEX_const) {
16719 HE* he = hv_fetch_ent(MUTABLE_HV(sv), index_const_sv, 0, 0);
16720 if (!he || HeVAL(he) != uninit_sv)
16724 SV * const * const svp =
16725 av_fetch(MUTABLE_AV(sv), index_const_iv, FALSE);
16726 if (!svp || *svp != uninit_sv)
16731 ? varname(agg_gv, '%', agg_targ,
16732 index_const_sv, 0, FUV_SUBSCRIPT_HASH)
16733 : varname(agg_gv, '@', agg_targ,
16734 NULL, index_const_iv, FUV_SUBSCRIPT_ARRAY);
16737 /* index is an var */
16739 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16741 return varname(agg_gv, '%', agg_targ,
16742 keysv, 0, FUV_SUBSCRIPT_HASH);
16745 const SSize_t index
16746 = find_array_subscript((const AV *)sv, uninit_sv);
16748 return varname(agg_gv, '@', agg_targ,
16749 NULL, index, FUV_SUBSCRIPT_ARRAY);
16753 return varname(agg_gv,
16755 agg_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
16757 NOT_REACHED; /* NOTREACHED */
16761 /* only examine RHS */
16762 return find_uninit_var(cBINOPx(obase)->op_first, uninit_sv,
16766 o = cUNOPx(obase)->op_first;
16767 if ( o->op_type == OP_PUSHMARK
16768 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)
16772 if (!OpHAS_SIBLING(o)) {
16773 /* one-arg version of open is highly magical */
16775 if (o->op_type == OP_GV) { /* open FOO; */
16777 if (match && GvSV(gv) != uninit_sv)
16779 return varname(gv, '$', 0,
16780 NULL, 0, FUV_SUBSCRIPT_NONE);
16782 /* other possibilities not handled are:
16783 * open $x; or open my $x; should return '${*$x}'
16784 * open expr; should return '$'.expr ideally
16791 /* ops where $_ may be an implicit arg */
16796 if ( !(obase->op_flags & OPf_STACKED)) {
16797 if (uninit_sv == DEFSV)
16798 return newSVpvs_flags("$_", SVs_TEMP);
16799 else if (obase->op_targ
16800 && uninit_sv == PAD_SVl(obase->op_targ))
16801 return varname(NULL, '$', obase->op_targ, NULL, 0,
16802 FUV_SUBSCRIPT_NONE);
16809 match = 1; /* print etc can return undef on defined args */
16810 /* skip filehandle as it can't produce 'undef' warning */
16811 o = cUNOPx(obase)->op_first;
16812 if ((obase->op_flags & OPf_STACKED)
16814 ( o->op_type == OP_PUSHMARK
16815 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)))
16816 o = OpSIBLING(OpSIBLING(o));
16820 case OP_ENTEREVAL: /* could be eval $undef or $x='$undef'; eval $x */
16821 case OP_CUSTOM: /* XS or custom code could trigger random warnings */
16823 /* the following ops are capable of returning PL_sv_undef even for
16824 * defined arg(s) */
16843 case OP_GETPEERNAME:
16890 case OP_SMARTMATCH:
16899 /* XXX tmp hack: these two may call an XS sub, and currently
16900 XS subs don't have a SUB entry on the context stack, so CV and
16901 pad determination goes wrong, and BAD things happen. So, just
16902 don't try to determine the value under those circumstances.
16903 Need a better fix at dome point. DAPM 11/2007 */
16909 GV * const gv = gv_fetchpvs(".", GV_NOTQUAL, SVt_PV);
16910 if (gv && GvSV(gv) == uninit_sv)
16911 return newSVpvs_flags("$.", SVs_TEMP);
16916 /* def-ness of rval pos() is independent of the def-ness of its arg */
16917 if ( !(obase->op_flags & OPf_MOD))
16923 if (SvROK(PL_rs) && uninit_sv == SvRV(PL_rs))
16924 return newSVpvs_flags("${$/}", SVs_TEMP);
16929 if (!(obase->op_flags & OPf_KIDS))
16931 o = cUNOPx(obase)->op_first;
16937 /* This loop checks all the kid ops, skipping any that cannot pos-
16938 * sibly be responsible for the uninitialized value; i.e., defined
16939 * constants and ops that return nothing. If there is only one op
16940 * left that is not skipped, then we *know* it is responsible for
16941 * the uninitialized value. If there is more than one op left, we
16942 * have to look for an exact match in the while() loop below.
16943 * Note that we skip padrange, because the individual pad ops that
16944 * it replaced are still in the tree, so we work on them instead.
16947 for (kid=o; kid; kid = OpSIBLING(kid)) {
16948 const OPCODE type = kid->op_type;
16949 if ( (type == OP_CONST && SvOK(cSVOPx_sv(kid)))
16950 || (type == OP_NULL && ! (kid->op_flags & OPf_KIDS))
16951 || (type == OP_PUSHMARK)
16952 || (type == OP_PADRANGE)
16956 if (o2) { /* more than one found */
16963 return find_uninit_var(o2, uninit_sv, match, desc_p);
16965 /* scan all args */
16967 sv = find_uninit_var(o, uninit_sv, 1, desc_p);
16979 =for apidoc report_uninit
16981 Print appropriate "Use of uninitialized variable" warning.
16987 Perl_report_uninit(pTHX_ const SV *uninit_sv)
16989 const char *desc = NULL;
16990 SV* varname = NULL;
16993 desc = PL_op->op_type == OP_STRINGIFY && PL_op->op_folded
16996 if (uninit_sv && PL_curpad) {
16997 varname = find_uninit_var(PL_op, uninit_sv, 0, &desc);
16999 sv_insert(varname, 0, 0, " ", 1);
17002 else if (PL_curstackinfo->si_type == PERLSI_SORT && cxstack_ix == 0)
17003 /* we've reached the end of a sort block or sub,
17004 * and the uninit value is probably what that code returned */
17007 /* PL_warn_uninit_sv is constant */
17008 GCC_DIAG_IGNORE(-Wformat-nonliteral);
17010 /* diag_listed_as: Use of uninitialized value%s */
17011 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit_sv,
17012 SVfARG(varname ? varname : &PL_sv_no),
17015 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit,
17021 * ex: set ts=8 sts=4 sw=4 et: