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_underlying && 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);
3922 const STRLEN len = GvNAMELEN(dstr);
3923 if(memEQs(name, len, "ISA")
3924 /* The stash may have been detached from the symbol table, so
3926 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3930 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
3931 || (len == 1 && name[0] == ':')) {
3934 /* Set aside the old stash, so we can reset isa caches on
3936 if((old_stash = GvHV(dstr)))
3937 /* Make sure we do not lose it early. */
3938 SvREFCNT_inc_simple_void_NN(
3939 sv_2mortal((SV *)old_stash)
3944 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
3947 /* freeing dstr's GP might free sstr (e.g. *x = $x),
3948 * so temporarily protect it */
3950 SAVEFREESV(SvREFCNT_inc_simple_NN(sstr));
3951 gp_free(MUTABLE_GV(dstr));
3952 GvINTRO_off(dstr); /* one-shot flag */
3953 GvGP_set(dstr, gp_ref(GvGP(sstr)));
3956 if (SvTAINTED(sstr))
3958 if (GvIMPORTED(dstr) != GVf_IMPORTED
3959 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
3961 GvIMPORTED_on(dstr);
3964 if(mro_changes == 2) {
3965 if (GvAV((const GV *)sstr)) {
3967 SV * const sref = (SV *)GvAV((const GV *)dstr);
3968 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
3969 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
3970 AV * const ary = newAV();
3971 av_push(ary, mg->mg_obj); /* takes the refcount */
3972 mg->mg_obj = (SV *)ary;
3974 av_push((AV *)mg->mg_obj, SvREFCNT_inc_simple_NN(dstr));
3976 else sv_magic(sref, dstr, PERL_MAGIC_isa, NULL, 0);
3978 mro_isa_changed_in(GvSTASH(dstr));
3980 else if(mro_changes == 3) {
3981 HV * const stash = GvHV(dstr);
3982 if(old_stash ? (HV *)HvENAME_get(old_stash) : stash)
3988 else if(mro_changes) mro_method_changed_in(GvSTASH(dstr));
3989 if (GvIO(dstr) && dtype == SVt_PVGV) {
3990 DEBUG_o(Perl_deb(aTHX_
3991 "glob_assign_glob clearing PL_stashcache\n"));
3992 /* It's a cache. It will rebuild itself quite happily.
3993 It's a lot of effort to work out exactly which key (or keys)
3994 might be invalidated by the creation of the this file handle.
3996 hv_clear(PL_stashcache);
4002 Perl_gv_setref(pTHX_ SV *const dstr, SV *const sstr)
4004 SV * const sref = SvRV(sstr);
4006 const int intro = GvINTRO(dstr);
4009 const U32 stype = SvTYPE(sref);
4011 PERL_ARGS_ASSERT_GV_SETREF;
4014 GvINTRO_off(dstr); /* one-shot flag */
4015 GvLINE(dstr) = CopLINE(PL_curcop);
4016 GvEGV(dstr) = MUTABLE_GV(dstr);
4021 location = (SV **) &(GvGP(dstr)->gp_cv); /* XXX bypassing GvCV_set */
4022 import_flag = GVf_IMPORTED_CV;
4025 location = (SV **) &GvHV(dstr);
4026 import_flag = GVf_IMPORTED_HV;
4029 location = (SV **) &GvAV(dstr);
4030 import_flag = GVf_IMPORTED_AV;
4033 location = (SV **) &GvIOp(dstr);
4036 location = (SV **) &GvFORM(dstr);
4039 location = &GvSV(dstr);
4040 import_flag = GVf_IMPORTED_SV;
4043 if (stype == SVt_PVCV) {
4044 /*if (GvCVGEN(dstr) && (GvCV(dstr) != (const CV *)sref || GvCVGEN(dstr))) {*/
4045 if (GvCVGEN(dstr)) {
4046 SvREFCNT_dec(GvCV(dstr));
4047 GvCV_set(dstr, NULL);
4048 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4051 /* SAVEt_GVSLOT takes more room on the savestack and has more
4052 overhead in leave_scope than SAVEt_GENERIC_SV. But for CVs
4053 leave_scope needs access to the GV so it can reset method
4054 caches. We must use SAVEt_GVSLOT whenever the type is
4055 SVt_PVCV, even if the stash is anonymous, as the stash may
4056 gain a name somehow before leave_scope. */
4057 if (stype == SVt_PVCV) {
4058 /* There is no save_pushptrptrptr. Creating it for this
4059 one call site would be overkill. So inline the ss add
4063 SS_ADD_PTR(location);
4064 SS_ADD_PTR(SvREFCNT_inc(*location));
4065 SS_ADD_UV(SAVEt_GVSLOT);
4068 else SAVEGENERICSV(*location);
4071 if (stype == SVt_PVCV && (*location != sref || GvCVGEN(dstr))) {
4072 CV* const cv = MUTABLE_CV(*location);
4074 if (!GvCVGEN((const GV *)dstr) &&
4075 (CvROOT(cv) || CvXSUB(cv)) &&
4076 /* redundant check that avoids creating the extra SV
4077 most of the time: */
4078 (CvCONST(cv) || ckWARN(WARN_REDEFINE)))
4080 SV * const new_const_sv =
4081 CvCONST((const CV *)sref)
4082 ? cv_const_sv((const CV *)sref)
4084 HV * const stash = GvSTASH((const GV *)dstr);
4085 report_redefined_cv(
4088 ? Perl_newSVpvf(aTHX_
4089 "%" HEKf "::%" HEKf,
4090 HEKfARG(HvNAME_HEK(stash)),
4091 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr))))
4092 : Perl_newSVpvf(aTHX_
4094 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr))))
4097 CvCONST((const CV *)sref) ? &new_const_sv : NULL
4101 cv_ckproto_len_flags(cv, (const GV *)dstr,
4102 SvPOK(sref) ? CvPROTO(sref) : NULL,
4103 SvPOK(sref) ? CvPROTOLEN(sref) : 0,
4104 SvPOK(sref) ? SvUTF8(sref) : 0);
4106 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4107 GvASSUMECV_on(dstr);
4108 if(GvSTASH(dstr)) { /* sub foo { 1 } sub bar { 2 } *bar = \&foo */
4109 if (intro && GvREFCNT(dstr) > 1) {
4110 /* temporary remove extra savestack's ref */
4112 gv_method_changed(dstr);
4115 else gv_method_changed(dstr);
4118 *location = SvREFCNT_inc_simple_NN(sref);
4119 if (import_flag && !(GvFLAGS(dstr) & import_flag)
4120 && CopSTASH_ne(PL_curcop, GvSTASH(dstr))) {
4121 GvFLAGS(dstr) |= import_flag;
4124 if (stype == SVt_PVHV) {
4125 const char * const name = GvNAME((GV*)dstr);
4126 const STRLEN len = GvNAMELEN(dstr);
4129 (len > 1 && name[len-2] == ':' && name[len-1] == ':')
4130 || (len == 1 && name[0] == ':')
4132 && (!dref || HvENAME_get(dref))
4135 (HV *)sref, (HV *)dref,
4141 stype == SVt_PVAV && sref != dref
4142 && memEQs(GvNAME((GV*)dstr), GvNAMELEN((GV*)dstr), "ISA")
4143 /* The stash may have been detached from the symbol table, so
4144 check its name before doing anything. */
4145 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
4148 MAGIC * const omg = dref && SvSMAGICAL(dref)
4149 ? mg_find(dref, PERL_MAGIC_isa)
4151 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
4152 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
4153 AV * const ary = newAV();
4154 av_push(ary, mg->mg_obj); /* takes the refcount */
4155 mg->mg_obj = (SV *)ary;
4158 if (SvTYPE(omg->mg_obj) == SVt_PVAV) {
4159 SV **svp = AvARRAY((AV *)omg->mg_obj);
4160 I32 items = AvFILLp((AV *)omg->mg_obj) + 1;
4164 SvREFCNT_inc_simple_NN(*svp++)
4170 SvREFCNT_inc_simple_NN(omg->mg_obj)
4174 av_push((AV *)mg->mg_obj,SvREFCNT_inc_simple_NN(dstr));
4180 sref, omg ? omg->mg_obj : dstr, PERL_MAGIC_isa, NULL, 0
4182 for (i = 0; i <= AvFILL(sref); ++i) {
4183 SV **elem = av_fetch ((AV*)sref, i, 0);
4186 *elem, sref, PERL_MAGIC_isaelem, NULL, i
4190 mg = mg_find(sref, PERL_MAGIC_isa);
4192 /* Since the *ISA assignment could have affected more than
4193 one stash, don't call mro_isa_changed_in directly, but let
4194 magic_clearisa do it for us, as it already has the logic for
4195 dealing with globs vs arrays of globs. */
4197 Perl_magic_clearisa(aTHX_ NULL, mg);
4199 else if (stype == SVt_PVIO) {
4200 DEBUG_o(Perl_deb(aTHX_ "gv_setref clearing PL_stashcache\n"));
4201 /* It's a cache. It will rebuild itself quite happily.
4202 It's a lot of effort to work out exactly which key (or keys)
4203 might be invalidated by the creation of the this file handle.
4205 hv_clear(PL_stashcache);
4209 if (!intro) SvREFCNT_dec(dref);
4210 if (SvTAINTED(sstr))
4218 #ifdef PERL_DEBUG_READONLY_COW
4219 # include <sys/mman.h>
4221 # ifndef PERL_MEMORY_DEBUG_HEADER_SIZE
4222 # define PERL_MEMORY_DEBUG_HEADER_SIZE 0
4226 Perl_sv_buf_to_ro(pTHX_ SV *sv)
4228 struct perl_memory_debug_header * const header =
4229 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4230 const MEM_SIZE len = header->size;
4231 PERL_ARGS_ASSERT_SV_BUF_TO_RO;
4232 # ifdef PERL_TRACK_MEMPOOL
4233 if (!header->readonly) header->readonly = 1;
4235 if (mprotect(header, len, PROT_READ))
4236 Perl_warn(aTHX_ "mprotect RW for COW string %p %lu failed with %d",
4237 header, len, errno);
4241 S_sv_buf_to_rw(pTHX_ SV *sv)
4243 struct perl_memory_debug_header * const header =
4244 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4245 const MEM_SIZE len = header->size;
4246 PERL_ARGS_ASSERT_SV_BUF_TO_RW;
4247 if (mprotect(header, len, PROT_READ|PROT_WRITE))
4248 Perl_warn(aTHX_ "mprotect for COW string %p %lu failed with %d",
4249 header, len, errno);
4250 # ifdef PERL_TRACK_MEMPOOL
4251 header->readonly = 0;
4256 # define sv_buf_to_ro(sv) NOOP
4257 # define sv_buf_to_rw(sv) NOOP
4261 Perl_sv_setsv_flags(pTHX_ SV *dstr, SV* sstr, const I32 flags)
4266 unsigned int both_type;
4268 PERL_ARGS_ASSERT_SV_SETSV_FLAGS;
4270 if (UNLIKELY( sstr == dstr ))
4273 if (UNLIKELY( !sstr ))
4274 sstr = &PL_sv_undef;
4276 stype = SvTYPE(sstr);
4277 dtype = SvTYPE(dstr);
4278 both_type = (stype | dtype);
4280 /* with these values, we can check that both SVs are NULL/IV (and not
4281 * freed) just by testing the or'ed types */
4282 STATIC_ASSERT_STMT(SVt_NULL == 0);
4283 STATIC_ASSERT_STMT(SVt_IV == 1);
4284 if (both_type <= 1) {
4285 /* both src and dst are UNDEF/IV/RV, so we can do a lot of
4291 /* minimal subset of SV_CHECK_THINKFIRST_COW_DROP(dstr) */
4292 if (SvREADONLY(dstr))
4293 Perl_croak_no_modify();
4295 if (SvWEAKREF(dstr))
4296 sv_unref_flags(dstr, 0);
4298 old_rv = SvRV(dstr);
4301 assert(!SvGMAGICAL(sstr));
4302 assert(!SvGMAGICAL(dstr));
4304 sflags = SvFLAGS(sstr);
4305 if (sflags & (SVf_IOK|SVf_ROK)) {
4306 SET_SVANY_FOR_BODYLESS_IV(dstr);
4307 new_dflags = SVt_IV;
4309 if (sflags & SVf_ROK) {
4310 dstr->sv_u.svu_rv = SvREFCNT_inc(SvRV(sstr));
4311 new_dflags |= SVf_ROK;
4314 /* both src and dst are <= SVt_IV, so sv_any points to the
4315 * head; so access the head directly
4317 assert( &(sstr->sv_u.svu_iv)
4318 == &(((XPVIV*) SvANY(sstr))->xiv_iv));
4319 assert( &(dstr->sv_u.svu_iv)
4320 == &(((XPVIV*) SvANY(dstr))->xiv_iv));
4321 dstr->sv_u.svu_iv = sstr->sv_u.svu_iv;
4322 new_dflags |= (SVf_IOK|SVp_IOK|(sflags & SVf_IVisUV));
4326 new_dflags = dtype; /* turn off everything except the type */
4328 SvFLAGS(dstr) = new_dflags;
4329 SvREFCNT_dec(old_rv);
4334 if (UNLIKELY(both_type == SVTYPEMASK)) {
4335 if (SvIS_FREED(dstr)) {
4336 Perl_croak(aTHX_ "panic: attempt to copy value %" SVf
4337 " to a freed scalar %p", SVfARG(sstr), (void *)dstr);
4339 if (SvIS_FREED(sstr)) {
4340 Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p to %p",
4341 (void*)sstr, (void*)dstr);
4347 SV_CHECK_THINKFIRST_COW_DROP(dstr);
4348 dtype = SvTYPE(dstr); /* THINKFIRST may have changed type */
4350 /* There's a lot of redundancy below but we're going for speed here */
4355 if (LIKELY( dtype != SVt_PVGV && dtype != SVt_PVLV )) {
4356 (void)SvOK_off(dstr);
4364 /* For performance, we inline promoting to type SVt_IV. */
4365 /* We're starting from SVt_NULL, so provided that define is
4366 * actual 0, we don't have to unset any SV type flags
4367 * to promote to SVt_IV. */
4368 STATIC_ASSERT_STMT(SVt_NULL == 0);
4369 SET_SVANY_FOR_BODYLESS_IV(dstr);
4370 SvFLAGS(dstr) |= SVt_IV;
4374 sv_upgrade(dstr, SVt_PVIV);
4378 goto end_of_first_switch;
4380 (void)SvIOK_only(dstr);
4381 SvIV_set(dstr, SvIVX(sstr));
4384 /* SvTAINTED can only be true if the SV has taint magic, which in
4385 turn means that the SV type is PVMG (or greater). This is the
4386 case statement for SVt_IV, so this cannot be true (whatever gcov
4388 assert(!SvTAINTED(sstr));
4393 if (dtype < SVt_PV && dtype != SVt_IV)
4394 sv_upgrade(dstr, SVt_IV);
4398 if (LIKELY( SvNOK(sstr) )) {
4402 sv_upgrade(dstr, SVt_NV);
4406 sv_upgrade(dstr, SVt_PVNV);
4410 goto end_of_first_switch;
4412 SvNV_set(dstr, SvNVX(sstr));
4413 (void)SvNOK_only(dstr);
4414 /* SvTAINTED can only be true if the SV has taint magic, which in
4415 turn means that the SV type is PVMG (or greater). This is the
4416 case statement for SVt_NV, so this cannot be true (whatever gcov
4418 assert(!SvTAINTED(sstr));
4425 sv_upgrade(dstr, SVt_PV);
4428 if (dtype < SVt_PVIV)
4429 sv_upgrade(dstr, SVt_PVIV);
4432 if (dtype < SVt_PVNV)
4433 sv_upgrade(dstr, SVt_PVNV);
4437 const char * const type = sv_reftype(sstr,0);
4439 /* diag_listed_as: Bizarre copy of %s */
4440 Perl_croak(aTHX_ "Bizarre copy of %s in %s", type, OP_DESC(PL_op));
4442 Perl_croak(aTHX_ "Bizarre copy of %s", type);
4444 NOT_REACHED; /* NOTREACHED */
4448 if (dtype < SVt_REGEXP)
4449 sv_upgrade(dstr, SVt_REGEXP);
4456 if (SvGMAGICAL(sstr) && (flags & SV_GMAGIC)) {
4458 if (SvTYPE(sstr) != stype)
4459 stype = SvTYPE(sstr);
4461 if (isGV_with_GP(sstr) && dtype <= SVt_PVLV) {
4462 glob_assign_glob(dstr, sstr, dtype);
4465 if (stype == SVt_PVLV)
4467 if (isREGEXP(sstr)) goto upgregexp;
4468 SvUPGRADE(dstr, SVt_PVNV);
4471 SvUPGRADE(dstr, (svtype)stype);
4473 end_of_first_switch:
4475 /* dstr may have been upgraded. */
4476 dtype = SvTYPE(dstr);
4477 sflags = SvFLAGS(sstr);
4479 if (UNLIKELY( dtype == SVt_PVCV )) {
4480 /* Assigning to a subroutine sets the prototype. */
4483 const char *const ptr = SvPV_const(sstr, len);
4485 SvGROW(dstr, len + 1);
4486 Copy(ptr, SvPVX(dstr), len + 1, char);
4487 SvCUR_set(dstr, len);
4489 SvFLAGS(dstr) |= sflags & SVf_UTF8;
4490 CvAUTOLOAD_off(dstr);
4495 else if (UNLIKELY(dtype == SVt_PVAV || dtype == SVt_PVHV
4496 || dtype == SVt_PVFM))
4498 const char * const type = sv_reftype(dstr,0);
4500 /* diag_listed_as: Cannot copy to %s */
4501 Perl_croak(aTHX_ "Cannot copy to %s in %s", type, OP_DESC(PL_op));
4503 Perl_croak(aTHX_ "Cannot copy to %s", type);
4504 } else if (sflags & SVf_ROK) {
4505 if (isGV_with_GP(dstr)
4506 && SvTYPE(SvRV(sstr)) == SVt_PVGV && isGV_with_GP(SvRV(sstr))) {
4509 if (GvIMPORTED(dstr) != GVf_IMPORTED
4510 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
4512 GvIMPORTED_on(dstr);
4517 glob_assign_glob(dstr, sstr, dtype);
4521 if (dtype >= SVt_PV) {
4522 if (isGV_with_GP(dstr)) {
4523 gv_setref(dstr, sstr);
4526 if (SvPVX_const(dstr)) {
4532 (void)SvOK_off(dstr);
4533 SvRV_set(dstr, SvREFCNT_inc(SvRV(sstr)));
4534 SvFLAGS(dstr) |= sflags & SVf_ROK;
4535 assert(!(sflags & SVp_NOK));
4536 assert(!(sflags & SVp_IOK));
4537 assert(!(sflags & SVf_NOK));
4538 assert(!(sflags & SVf_IOK));
4540 else if (isGV_with_GP(dstr)) {
4541 if (!(sflags & SVf_OK)) {
4542 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4543 "Undefined value assigned to typeglob");
4546 GV *gv = gv_fetchsv_nomg(sstr, GV_ADD, SVt_PVGV);
4547 if (dstr != (const SV *)gv) {
4548 const char * const name = GvNAME((const GV *)dstr);
4549 const STRLEN len = GvNAMELEN(dstr);
4550 HV *old_stash = NULL;
4551 bool reset_isa = FALSE;
4552 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
4553 || (len == 1 && name[0] == ':')) {
4554 /* Set aside the old stash, so we can reset isa caches
4555 on its subclasses. */
4556 if((old_stash = GvHV(dstr))) {
4557 /* Make sure we do not lose it early. */
4558 SvREFCNT_inc_simple_void_NN(
4559 sv_2mortal((SV *)old_stash)
4566 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
4567 gp_free(MUTABLE_GV(dstr));
4569 GvGP_set(dstr, gp_ref(GvGP(gv)));
4572 HV * const stash = GvHV(dstr);
4574 old_stash ? (HV *)HvENAME_get(old_stash) : stash
4584 else if ((dtype == SVt_REGEXP || dtype == SVt_PVLV)
4585 && (stype == SVt_REGEXP || isREGEXP(sstr))) {
4586 reg_temp_copy((REGEXP*)dstr, (REGEXP*)sstr);
4588 else if (sflags & SVp_POK) {
4589 const STRLEN cur = SvCUR(sstr);
4590 const STRLEN len = SvLEN(sstr);
4593 * We have three basic ways to copy the string:
4599 * Which we choose is based on various factors. The following
4600 * things are listed in order of speed, fastest to slowest:
4602 * - Copying a short string
4603 * - Copy-on-write bookkeeping
4605 * - Copying a long string
4607 * We swipe the string (steal the string buffer) if the SV on the
4608 * rhs is about to be freed anyway (TEMP and refcnt==1). This is a
4609 * big win on long strings. It should be a win on short strings if
4610 * SvPVX_const(dstr) has to be allocated. If not, it should not
4611 * slow things down, as SvPVX_const(sstr) would have been freed
4614 * We also steal the buffer from a PADTMP (operator target) if it
4615 * is ‘long enough’. For short strings, a swipe does not help
4616 * here, as it causes more malloc calls the next time the target
4617 * is used. Benchmarks show that even if SvPVX_const(dstr) has to
4618 * be allocated it is still not worth swiping PADTMPs for short
4619 * strings, as the savings here are small.
4621 * If swiping is not an option, then we see whether it is
4622 * worth using copy-on-write. If the lhs already has a buf-
4623 * fer big enough and the string is short, we skip it and fall back
4624 * to method 3, since memcpy is faster for short strings than the
4625 * later bookkeeping overhead that copy-on-write entails.
4627 * If the rhs is not a copy-on-write string yet, then we also
4628 * consider whether the buffer is too large relative to the string
4629 * it holds. Some operations such as readline allocate a large
4630 * buffer in the expectation of reusing it. But turning such into
4631 * a COW buffer is counter-productive because it increases memory
4632 * usage by making readline allocate a new large buffer the sec-
4633 * ond time round. So, if the buffer is too large, again, we use
4636 * Finally, if there is no buffer on the left, or the buffer is too
4637 * small, then we use copy-on-write and make both SVs share the
4642 /* Whichever path we take through the next code, we want this true,
4643 and doing it now facilitates the COW check. */
4644 (void)SvPOK_only(dstr);
4648 /* slated for free anyway (and not COW)? */
4649 (sflags & (SVs_TEMP|SVf_IsCOW)) == SVs_TEMP
4650 /* or a swipable TARG */
4652 (SVs_PADTMP|SVf_READONLY|SVf_PROTECT|SVf_IsCOW))
4654 /* whose buffer is worth stealing */
4655 && CHECK_COWBUF_THRESHOLD(cur,len)
4658 !(sflags & SVf_OOK) && /* and not involved in OOK hack? */
4659 (!(flags & SV_NOSTEAL)) &&
4660 /* and we're allowed to steal temps */
4661 SvREFCNT(sstr) == 1 && /* and no other references to it? */
4662 len) /* and really is a string */
4663 { /* Passes the swipe test. */
4664 if (SvPVX_const(dstr)) /* we know that dtype >= SVt_PV */
4666 SvPV_set(dstr, SvPVX_mutable(sstr));
4667 SvLEN_set(dstr, SvLEN(sstr));
4668 SvCUR_set(dstr, SvCUR(sstr));
4671 (void)SvOK_off(sstr); /* NOTE: nukes most SvFLAGS on sstr */
4672 SvPV_set(sstr, NULL);
4677 else if (flags & SV_COW_SHARED_HASH_KEYS
4679 #ifdef PERL_COPY_ON_WRITE
4682 ( (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4683 /* If this is a regular (non-hek) COW, only so
4684 many COW "copies" are possible. */
4685 && CowREFCNT(sstr) != SV_COW_REFCNT_MAX ))
4686 : ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4687 && !(SvFLAGS(dstr) & SVf_BREAK)
4688 && CHECK_COW_THRESHOLD(cur,len) && cur+1 < len
4689 && (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4693 && !(SvFLAGS(dstr) & SVf_BREAK)
4696 /* Either it's a shared hash key, or it's suitable for
4700 PerlIO_printf(Perl_debug_log, "Copy on write: sstr --> dstr\n");
4706 if (!(sflags & SVf_IsCOW)) {
4708 CowREFCNT(sstr) = 0;
4711 if (SvPVX_const(dstr)) { /* we know that dtype >= SVt_PV */
4717 if (sflags & SVf_IsCOW) {
4721 SvPV_set(dstr, SvPVX_mutable(sstr));
4726 /* SvIsCOW_shared_hash */
4727 DEBUG_C(PerlIO_printf(Perl_debug_log,
4728 "Copy on write: Sharing hash\n"));
4730 assert (SvTYPE(dstr) >= SVt_PV);
4732 HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)))));
4734 SvLEN_set(dstr, len);
4735 SvCUR_set(dstr, cur);
4738 /* Failed the swipe test, and we cannot do copy-on-write either.
4739 Have to copy the string. */
4740 SvGROW(dstr, cur + 1); /* inlined from sv_setpvn */
4741 Move(SvPVX_const(sstr),SvPVX(dstr),cur,char);
4742 SvCUR_set(dstr, cur);
4743 *SvEND(dstr) = '\0';
4745 if (sflags & SVp_NOK) {
4746 SvNV_set(dstr, SvNVX(sstr));
4748 if (sflags & SVp_IOK) {
4749 SvIV_set(dstr, SvIVX(sstr));
4750 if (sflags & SVf_IVisUV)
4753 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_NOK|SVp_NOK|SVf_UTF8);
4755 const MAGIC * const smg = SvVSTRING_mg(sstr);
4757 sv_magic(dstr, NULL, PERL_MAGIC_vstring,
4758 smg->mg_ptr, smg->mg_len);
4759 SvRMAGICAL_on(dstr);
4763 else if (sflags & (SVp_IOK|SVp_NOK)) {
4764 (void)SvOK_off(dstr);
4765 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_IVisUV|SVf_NOK|SVp_NOK);
4766 if (sflags & SVp_IOK) {
4767 /* XXXX Do we want to set IsUV for IV(ROK)? Be extra safe... */
4768 SvIV_set(dstr, SvIVX(sstr));
4770 if (sflags & SVp_NOK) {
4771 SvNV_set(dstr, SvNVX(sstr));
4775 if (isGV_with_GP(sstr)) {
4776 gv_efullname3(dstr, MUTABLE_GV(sstr), "*");
4779 (void)SvOK_off(dstr);
4781 if (SvTAINTED(sstr))
4787 =for apidoc sv_set_undef
4789 Equivalent to C<sv_setsv(sv, &PL_sv_undef)>, but more efficient.
4790 Doesn't handle set magic.
4792 The perl equivalent is C<$sv = undef;>. Note that it doesn't free any string
4793 buffer, unlike C<undef $sv>.
4795 Introduced in perl 5.25.12.
4801 Perl_sv_set_undef(pTHX_ SV *sv)
4803 U32 type = SvTYPE(sv);
4805 PERL_ARGS_ASSERT_SV_SET_UNDEF;
4807 /* shortcut, NULL, IV, RV */
4809 if (type <= SVt_IV) {
4810 assert(!SvGMAGICAL(sv));
4811 if (SvREADONLY(sv)) {
4812 /* does undeffing PL_sv_undef count as modifying a read-only
4813 * variable? Some XS code does this */
4814 if (sv == &PL_sv_undef)
4816 Perl_croak_no_modify();
4821 sv_unref_flags(sv, 0);
4824 SvFLAGS(sv) = type; /* quickly turn off all flags */
4825 SvREFCNT_dec_NN(rv);
4829 SvFLAGS(sv) = type; /* quickly turn off all flags */
4834 Perl_croak(aTHX_ "panic: attempt to undefine a freed scalar %p",
4837 SV_CHECK_THINKFIRST_COW_DROP(sv);
4839 if (isGV_with_GP(sv))
4840 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4841 "Undefined value assigned to typeglob");
4849 =for apidoc sv_setsv_mg
4851 Like C<sv_setsv>, but also handles 'set' magic.
4857 Perl_sv_setsv_mg(pTHX_ SV *const dstr, SV *const sstr)
4859 PERL_ARGS_ASSERT_SV_SETSV_MG;
4861 sv_setsv(dstr,sstr);
4866 # define SVt_COW SVt_PV
4868 Perl_sv_setsv_cow(pTHX_ SV *dstr, SV *sstr)
4870 STRLEN cur = SvCUR(sstr);
4871 STRLEN len = SvLEN(sstr);
4873 #if defined(PERL_DEBUG_READONLY_COW) && defined(PERL_COPY_ON_WRITE)
4874 const bool already = cBOOL(SvIsCOW(sstr));
4877 PERL_ARGS_ASSERT_SV_SETSV_COW;
4880 PerlIO_printf(Perl_debug_log, "Fast copy on write: %p -> %p\n",
4881 (void*)sstr, (void*)dstr);
4888 if (SvTHINKFIRST(dstr))
4889 sv_force_normal_flags(dstr, SV_COW_DROP_PV);
4890 else if (SvPVX_const(dstr))
4891 Safefree(SvPVX_mutable(dstr));
4895 SvUPGRADE(dstr, SVt_COW);
4897 assert (SvPOK(sstr));
4898 assert (SvPOKp(sstr));
4900 if (SvIsCOW(sstr)) {
4902 if (SvLEN(sstr) == 0) {
4903 /* source is a COW shared hash key. */
4904 DEBUG_C(PerlIO_printf(Perl_debug_log,
4905 "Fast copy on write: Sharing hash\n"));
4906 new_pv = HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr))));
4909 assert(SvCUR(sstr)+1 < SvLEN(sstr));
4910 assert(CowREFCNT(sstr) < SV_COW_REFCNT_MAX);
4912 assert ((SvFLAGS(sstr) & CAN_COW_MASK) == CAN_COW_FLAGS);
4913 SvUPGRADE(sstr, SVt_COW);
4915 DEBUG_C(PerlIO_printf(Perl_debug_log,
4916 "Fast copy on write: Converting sstr to COW\n"));
4917 CowREFCNT(sstr) = 0;
4919 # ifdef PERL_DEBUG_READONLY_COW
4920 if (already) sv_buf_to_rw(sstr);
4923 new_pv = SvPVX_mutable(sstr);
4927 SvPV_set(dstr, new_pv);
4928 SvFLAGS(dstr) = (SVt_COW|SVf_POK|SVp_POK|SVf_IsCOW);
4931 SvLEN_set(dstr, len);
4932 SvCUR_set(dstr, cur);
4942 =for apidoc sv_setpv_bufsize
4944 Sets the SV to be a string of cur bytes length, with at least
4945 len bytes available. Ensures that there is a null byte at SvEND.
4946 Returns a char * pointer to the SvPV buffer.
4952 Perl_sv_setpv_bufsize(pTHX_ SV *const sv, const STRLEN cur, const STRLEN len)
4956 PERL_ARGS_ASSERT_SV_SETPV_BUFSIZE;
4958 SV_CHECK_THINKFIRST_COW_DROP(sv);
4959 SvUPGRADE(sv, SVt_PV);
4960 pv = SvGROW(sv, len + 1);
4963 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4966 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4971 =for apidoc sv_setpvn
4973 Copies a string (possibly containing embedded C<NUL> characters) into an SV.
4974 The C<len> parameter indicates the number of
4975 bytes to be copied. If the C<ptr> argument is NULL the SV will become
4976 undefined. Does not handle 'set' magic. See C<L</sv_setpvn_mg>>.
4982 Perl_sv_setpvn(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4986 PERL_ARGS_ASSERT_SV_SETPVN;
4988 SV_CHECK_THINKFIRST_COW_DROP(sv);
4989 if (isGV_with_GP(sv))
4990 Perl_croak_no_modify();
4996 /* len is STRLEN which is unsigned, need to copy to signed */
4999 Perl_croak(aTHX_ "panic: sv_setpvn called with negative strlen %"
5002 SvUPGRADE(sv, SVt_PV);
5004 dptr = SvGROW(sv, len + 1);
5005 Move(ptr,dptr,len,char);
5008 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5010 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
5014 =for apidoc sv_setpvn_mg
5016 Like C<sv_setpvn>, but also handles 'set' magic.
5022 Perl_sv_setpvn_mg(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
5024 PERL_ARGS_ASSERT_SV_SETPVN_MG;
5026 sv_setpvn(sv,ptr,len);
5031 =for apidoc sv_setpv
5033 Copies a string into an SV. The string must be terminated with a C<NUL>
5034 character, and not contain embeded C<NUL>'s.
5035 Does not handle 'set' magic. See C<L</sv_setpv_mg>>.
5041 Perl_sv_setpv(pTHX_ SV *const sv, const char *const ptr)
5045 PERL_ARGS_ASSERT_SV_SETPV;
5047 SV_CHECK_THINKFIRST_COW_DROP(sv);
5053 SvUPGRADE(sv, SVt_PV);
5055 SvGROW(sv, len + 1);
5056 Move(ptr,SvPVX(sv),len+1,char);
5058 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5060 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
5064 =for apidoc sv_setpv_mg
5066 Like C<sv_setpv>, but also handles 'set' magic.
5072 Perl_sv_setpv_mg(pTHX_ SV *const sv, const char *const ptr)
5074 PERL_ARGS_ASSERT_SV_SETPV_MG;
5081 Perl_sv_sethek(pTHX_ SV *const sv, const HEK *const hek)
5083 PERL_ARGS_ASSERT_SV_SETHEK;
5089 if (HEK_LEN(hek) == HEf_SVKEY) {
5090 sv_setsv(sv, *(SV**)HEK_KEY(hek));
5093 const int flags = HEK_FLAGS(hek);
5094 if (flags & HVhek_WASUTF8) {
5095 STRLEN utf8_len = HEK_LEN(hek);
5096 char *as_utf8 = (char *)bytes_to_utf8((U8*)HEK_KEY(hek), &utf8_len);
5097 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
5100 } else if (flags & HVhek_UNSHARED) {
5101 sv_setpvn(sv, HEK_KEY(hek), HEK_LEN(hek));
5104 else SvUTF8_off(sv);
5108 SV_CHECK_THINKFIRST_COW_DROP(sv);
5109 SvUPGRADE(sv, SVt_PV);
5111 SvPV_set(sv,(char *)HEK_KEY(share_hek_hek(hek)));
5112 SvCUR_set(sv, HEK_LEN(hek));
5118 else SvUTF8_off(sv);
5126 =for apidoc sv_usepvn_flags
5128 Tells an SV to use C<ptr> to find its string value. Normally the
5129 string is stored inside the SV, but sv_usepvn allows the SV to use an
5130 outside string. C<ptr> should point to memory that was allocated
5131 by L<C<Newx>|perlclib/Memory Management and String Handling>. It must be
5132 the start of a C<Newx>-ed block of memory, and not a pointer to the
5133 middle of it (beware of L<C<OOK>|perlguts/Offsets> and copy-on-write),
5134 and not be from a non-C<Newx> memory allocator like C<malloc>. The
5135 string length, C<len>, must be supplied. By default this function
5136 will C<Renew> (i.e. realloc, move) the memory pointed to by C<ptr>,
5137 so that pointer should not be freed or used by the programmer after
5138 giving it to C<sv_usepvn>, and neither should any pointers from "behind"
5139 that pointer (e.g. ptr + 1) be used.
5141 If S<C<flags & SV_SMAGIC>> is true, will call C<SvSETMAGIC>. If
5142 S<C<flags> & SV_HAS_TRAILING_NUL>> is true, then C<ptr[len]> must be C<NUL>,
5144 will be skipped (i.e. the buffer is actually at least 1 byte longer than
5145 C<len>, and already meets the requirements for storing in C<SvPVX>).
5151 Perl_sv_usepvn_flags(pTHX_ SV *const sv, char *ptr, const STRLEN len, const U32 flags)
5155 PERL_ARGS_ASSERT_SV_USEPVN_FLAGS;
5157 SV_CHECK_THINKFIRST_COW_DROP(sv);
5158 SvUPGRADE(sv, SVt_PV);
5161 if (flags & SV_SMAGIC)
5165 if (SvPVX_const(sv))
5169 if (flags & SV_HAS_TRAILING_NUL)
5170 assert(ptr[len] == '\0');
5173 allocate = (flags & SV_HAS_TRAILING_NUL)
5175 #ifdef Perl_safesysmalloc_size
5178 PERL_STRLEN_ROUNDUP(len + 1);
5180 if (flags & SV_HAS_TRAILING_NUL) {
5181 /* It's long enough - do nothing.
5182 Specifically Perl_newCONSTSUB is relying on this. */
5185 /* Force a move to shake out bugs in callers. */
5186 char *new_ptr = (char*)safemalloc(allocate);
5187 Copy(ptr, new_ptr, len, char);
5188 PoisonFree(ptr,len,char);
5192 ptr = (char*) saferealloc (ptr, allocate);
5195 #ifdef Perl_safesysmalloc_size
5196 SvLEN_set(sv, Perl_safesysmalloc_size(ptr));
5198 SvLEN_set(sv, allocate);
5202 if (!(flags & SV_HAS_TRAILING_NUL)) {
5205 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5207 if (flags & SV_SMAGIC)
5213 S_sv_uncow(pTHX_ SV * const sv, const U32 flags)
5215 assert(SvIsCOW(sv));
5218 const char * const pvx = SvPVX_const(sv);
5219 const STRLEN len = SvLEN(sv);
5220 const STRLEN cur = SvCUR(sv);
5224 PerlIO_printf(Perl_debug_log,
5225 "Copy on write: Force normal %ld\n",
5231 # ifdef PERL_COPY_ON_WRITE
5233 /* Must do this first, since the CowREFCNT uses SvPVX and
5234 we need to write to CowREFCNT, or de-RO the whole buffer if we are
5235 the only owner left of the buffer. */
5236 sv_buf_to_rw(sv); /* NOOP if RO-ing not supported */
5238 U8 cowrefcnt = CowREFCNT(sv);
5239 if(cowrefcnt != 0) {
5241 CowREFCNT(sv) = cowrefcnt;
5246 /* Else we are the only owner of the buffer. */
5251 /* This SV doesn't own the buffer, so need to Newx() a new one: */
5256 if (flags & SV_COW_DROP_PV) {
5257 /* OK, so we don't need to copy our buffer. */
5260 SvGROW(sv, cur + 1);
5261 Move(pvx,SvPVX(sv),cur,char);
5267 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5275 const char * const pvx = SvPVX_const(sv);
5276 const STRLEN len = SvCUR(sv);
5280 if (flags & SV_COW_DROP_PV) {
5281 /* OK, so we don't need to copy our buffer. */
5284 SvGROW(sv, len + 1);
5285 Move(pvx,SvPVX(sv),len,char);
5288 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5295 =for apidoc sv_force_normal_flags
5297 Undo various types of fakery on an SV, where fakery means
5298 "more than" a string: if the PV is a shared string, make
5299 a private copy; if we're a ref, stop refing; if we're a glob, downgrade to
5300 an C<xpvmg>; if we're a copy-on-write scalar, this is the on-write time when
5301 we do the copy, and is also used locally; if this is a
5302 vstring, drop the vstring magic. If C<SV_COW_DROP_PV> is set
5303 then a copy-on-write scalar drops its PV buffer (if any) and becomes
5304 C<SvPOK_off> rather than making a copy. (Used where this
5305 scalar is about to be set to some other value.) In addition,
5306 the C<flags> parameter gets passed to C<sv_unref_flags()>
5307 when unreffing. C<sv_force_normal> calls this function
5308 with flags set to 0.
5310 This function is expected to be used to signal to perl that this SV is
5311 about to be written to, and any extra book-keeping needs to be taken care
5312 of. Hence, it croaks on read-only values.
5318 Perl_sv_force_normal_flags(pTHX_ SV *const sv, const U32 flags)
5320 PERL_ARGS_ASSERT_SV_FORCE_NORMAL_FLAGS;
5323 Perl_croak_no_modify();
5324 else if (SvIsCOW(sv) && LIKELY(SvTYPE(sv) != SVt_PVHV))
5325 S_sv_uncow(aTHX_ sv, flags);
5327 sv_unref_flags(sv, flags);
5328 else if (SvFAKE(sv) && isGV_with_GP(sv))
5329 sv_unglob(sv, flags);
5330 else if (SvFAKE(sv) && isREGEXP(sv)) {
5331 /* Need to downgrade the REGEXP to a simple(r) scalar. This is analogous
5332 to sv_unglob. We only need it here, so inline it. */
5333 const bool islv = SvTYPE(sv) == SVt_PVLV;
5334 const svtype new_type =
5335 islv ? SVt_NULL : SvMAGIC(sv) || SvSTASH(sv) ? SVt_PVMG : SVt_PV;
5336 SV *const temp = newSV_type(new_type);
5337 regexp *old_rx_body;
5339 if (new_type == SVt_PVMG) {
5340 SvMAGIC_set(temp, SvMAGIC(sv));
5341 SvMAGIC_set(sv, NULL);
5342 SvSTASH_set(temp, SvSTASH(sv));
5343 SvSTASH_set(sv, NULL);
5346 SvCUR_set(temp, SvCUR(sv));
5347 /* Remember that SvPVX is in the head, not the body. */
5348 assert(ReANY((REGEXP *)sv)->mother_re);
5351 /* LV-as-regex has sv->sv_any pointing to an XPVLV body,
5352 * whose xpvlenu_rx field points to the regex body */
5353 XPV *xpv = (XPV*)(SvANY(sv));
5354 old_rx_body = xpv->xpv_len_u.xpvlenu_rx;
5355 xpv->xpv_len_u.xpvlenu_rx = NULL;
5358 old_rx_body = ReANY((REGEXP *)sv);
5360 /* Their buffer is already owned by someone else. */
5361 if (flags & SV_COW_DROP_PV) {
5362 /* SvLEN is already 0. For SVt_REGEXP, we have a brand new
5363 zeroed body. For SVt_PVLV, we zeroed it above (len field
5364 a union with xpvlenu_rx) */
5365 assert(!SvLEN(islv ? sv : temp));
5366 sv->sv_u.svu_pv = 0;
5369 sv->sv_u.svu_pv = savepvn(RX_WRAPPED((REGEXP *)sv), SvCUR(sv));
5370 SvLEN_set(islv ? sv : temp, SvCUR(sv)+1);
5374 /* Now swap the rest of the bodies. */
5378 SvFLAGS(sv) &= ~SVTYPEMASK;
5379 SvFLAGS(sv) |= new_type;
5380 SvANY(sv) = SvANY(temp);
5383 SvFLAGS(temp) &= ~(SVTYPEMASK);
5384 SvFLAGS(temp) |= SVt_REGEXP|SVf_FAKE;
5385 SvANY(temp) = old_rx_body;
5387 SvREFCNT_dec_NN(temp);
5389 else if (SvVOK(sv)) sv_unmagic(sv, PERL_MAGIC_vstring);
5395 Efficient removal of characters from the beginning of the string buffer.
5396 C<SvPOK(sv)>, or at least C<SvPOKp(sv)>, must be true and C<ptr> must be a
5397 pointer to somewhere inside the string buffer. C<ptr> becomes the first
5398 character of the adjusted string. Uses the C<OOK> hack. On return, only
5399 C<SvPOK(sv)> and C<SvPOKp(sv)> among the C<OK> flags will be true.
5401 Beware: after this function returns, C<ptr> and SvPVX_const(sv) may no longer
5402 refer to the same chunk of data.
5404 The unfortunate similarity of this function's name to that of Perl's C<chop>
5405 operator is strictly coincidental. This function works from the left;
5406 C<chop> works from the right.
5412 Perl_sv_chop(pTHX_ SV *const sv, const char *const ptr)
5423 PERL_ARGS_ASSERT_SV_CHOP;
5425 if (!ptr || !SvPOKp(sv))
5427 delta = ptr - SvPVX_const(sv);
5429 /* Nothing to do. */
5432 max_delta = SvLEN(sv) ? SvLEN(sv) : SvCUR(sv);
5433 if (delta > max_delta)
5434 Perl_croak(aTHX_ "panic: sv_chop ptr=%p, start=%p, end=%p",
5435 ptr, SvPVX_const(sv), SvPVX_const(sv) + max_delta);
5436 /* SvPVX(sv) may move in SV_CHECK_THINKFIRST(sv), so don't use ptr any more */
5437 SV_CHECK_THINKFIRST(sv);
5438 SvPOK_only_UTF8(sv);
5441 if (!SvLEN(sv)) { /* make copy of shared string */
5442 const char *pvx = SvPVX_const(sv);
5443 const STRLEN len = SvCUR(sv);
5444 SvGROW(sv, len + 1);
5445 Move(pvx,SvPVX(sv),len,char);
5451 SvOOK_offset(sv, old_delta);
5453 SvLEN_set(sv, SvLEN(sv) - delta);
5454 SvCUR_set(sv, SvCUR(sv) - delta);
5455 SvPV_set(sv, SvPVX(sv) + delta);
5457 p = (U8 *)SvPVX_const(sv);
5460 /* how many bytes were evacuated? we will fill them with sentinel
5461 bytes, except for the part holding the new offset of course. */
5464 evacn += (old_delta < 0x100 ? 1 : 1 + sizeof(STRLEN));
5466 assert(evacn <= delta + old_delta);
5470 /* This sets 'delta' to the accumulated value of all deltas so far */
5474 /* If 'delta' fits in a byte, store it just prior to the new beginning of
5475 * the string; otherwise store a 0 byte there and store 'delta' just prior
5476 * to that, using as many bytes as a STRLEN occupies. Thus it overwrites a
5477 * portion of the chopped part of the string */
5478 if (delta < 0x100) {
5482 p -= sizeof(STRLEN);
5483 Copy((U8*)&delta, p, sizeof(STRLEN), U8);
5487 /* Fill the preceding buffer with sentinals to verify that no-one is
5497 =for apidoc sv_catpvn
5499 Concatenates the string onto the end of the string which is in the SV.
5500 C<len> indicates number of bytes to copy. If the SV has the UTF-8
5501 status set, then the bytes appended should be valid UTF-8.
5502 Handles 'get' magic, but not 'set' magic. See C<L</sv_catpvn_mg>>.
5504 =for apidoc sv_catpvn_flags
5506 Concatenates the string onto the end of the string which is in the SV. The
5507 C<len> indicates number of bytes to copy.
5509 By default, the string appended is assumed to be valid UTF-8 if the SV has
5510 the UTF-8 status set, and a string of bytes otherwise. One can force the
5511 appended string to be interpreted as UTF-8 by supplying the C<SV_CATUTF8>
5512 flag, and as bytes by supplying the C<SV_CATBYTES> flag; the SV or the
5513 string appended will be upgraded to UTF-8 if necessary.
5515 If C<flags> has the C<SV_SMAGIC> bit set, will
5516 C<mg_set> on C<dsv> afterwards if appropriate.
5517 C<sv_catpvn> and C<sv_catpvn_nomg> are implemented
5518 in terms of this function.
5524 Perl_sv_catpvn_flags(pTHX_ SV *const dsv, const char *sstr, const STRLEN slen, const I32 flags)
5527 const char * const dstr = SvPV_force_flags(dsv, dlen, flags);
5529 PERL_ARGS_ASSERT_SV_CATPVN_FLAGS;
5530 assert((flags & (SV_CATBYTES|SV_CATUTF8)) != (SV_CATBYTES|SV_CATUTF8));
5532 if (!(flags & SV_CATBYTES) || !SvUTF8(dsv)) {
5533 if (flags & SV_CATUTF8 && !SvUTF8(dsv)) {
5534 sv_utf8_upgrade_flags_grow(dsv, 0, slen + 1);
5537 else SvGROW(dsv, dlen + slen + 3);
5539 sstr = SvPVX_const(dsv);
5540 Move(sstr, SvPVX(dsv) + dlen, slen, char);
5541 SvCUR_set(dsv, SvCUR(dsv) + slen);
5544 /* We inline bytes_to_utf8, to avoid an extra malloc. */
5545 const char * const send = sstr + slen;
5548 /* Something this code does not account for, which I think is
5549 impossible; it would require the same pv to be treated as
5550 bytes *and* utf8, which would indicate a bug elsewhere. */
5551 assert(sstr != dstr);
5553 SvGROW(dsv, dlen + slen * 2 + 3);
5554 d = (U8 *)SvPVX(dsv) + dlen;
5556 while (sstr < send) {
5557 append_utf8_from_native_byte(*sstr, &d);
5560 SvCUR_set(dsv, d-(const U8 *)SvPVX(dsv));
5563 (void)SvPOK_only_UTF8(dsv); /* validate pointer */
5565 if (flags & SV_SMAGIC)
5570 =for apidoc sv_catsv
5572 Concatenates the string from SV C<ssv> onto the end of the string in SV
5573 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5574 Handles 'get' magic on both SVs, but no 'set' magic. See C<L</sv_catsv_mg>>
5575 and C<L</sv_catsv_nomg>>.
5577 =for apidoc sv_catsv_flags
5579 Concatenates the string from SV C<ssv> onto the end of the string in SV
5580 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5581 If C<flags> has the C<SV_GMAGIC> bit set, will call C<mg_get> on both SVs if
5582 appropriate. If C<flags> has the C<SV_SMAGIC> bit set, C<mg_set> will be called on
5583 the modified SV afterward, if appropriate. C<sv_catsv>, C<sv_catsv_nomg>,
5584 and C<sv_catsv_mg> are implemented in terms of this function.
5589 Perl_sv_catsv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
5591 PERL_ARGS_ASSERT_SV_CATSV_FLAGS;
5595 const char *spv = SvPV_flags_const(ssv, slen, flags);
5596 if (flags & SV_GMAGIC)
5598 sv_catpvn_flags(dsv, spv, slen,
5599 DO_UTF8(ssv) ? SV_CATUTF8 : SV_CATBYTES);
5600 if (flags & SV_SMAGIC)
5606 =for apidoc sv_catpv
5608 Concatenates the C<NUL>-terminated string onto the end of the string which is
5610 If the SV has the UTF-8 status set, then the bytes appended should be
5611 valid UTF-8. Handles 'get' magic, but not 'set' magic. See
5617 Perl_sv_catpv(pTHX_ SV *const sv, const char *ptr)
5623 PERL_ARGS_ASSERT_SV_CATPV;
5627 junk = SvPV_force(sv, tlen);
5629 SvGROW(sv, tlen + len + 1);
5631 ptr = SvPVX_const(sv);
5632 Move(ptr,SvPVX(sv)+tlen,len+1,char);
5633 SvCUR_set(sv, SvCUR(sv) + len);
5634 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5639 =for apidoc sv_catpv_flags
5641 Concatenates the C<NUL>-terminated string onto the end of the string which is
5643 If the SV has the UTF-8 status set, then the bytes appended should
5644 be valid UTF-8. If C<flags> has the C<SV_SMAGIC> bit set, will C<mg_set>
5645 on the modified SV if appropriate.
5651 Perl_sv_catpv_flags(pTHX_ SV *dstr, const char *sstr, const I32 flags)
5653 PERL_ARGS_ASSERT_SV_CATPV_FLAGS;
5654 sv_catpvn_flags(dstr, sstr, strlen(sstr), flags);
5658 =for apidoc sv_catpv_mg
5660 Like C<sv_catpv>, but also handles 'set' magic.
5666 Perl_sv_catpv_mg(pTHX_ SV *const sv, const char *const ptr)
5668 PERL_ARGS_ASSERT_SV_CATPV_MG;
5677 Creates a new SV. A non-zero C<len> parameter indicates the number of
5678 bytes of preallocated string space the SV should have. An extra byte for a
5679 trailing C<NUL> is also reserved. (C<SvPOK> is not set for the SV even if string
5680 space is allocated.) The reference count for the new SV is set to 1.
5682 In 5.9.3, C<newSV()> replaces the older C<NEWSV()> API, and drops the first
5683 parameter, I<x>, a debug aid which allowed callers to identify themselves.
5684 This aid has been superseded by a new build option, C<PERL_MEM_LOG> (see
5685 L<perlhacktips/PERL_MEM_LOG>). The older API is still there for use in XS
5686 modules supporting older perls.
5692 Perl_newSV(pTHX_ const STRLEN len)
5698 sv_grow(sv, len + 1);
5703 =for apidoc sv_magicext
5705 Adds magic to an SV, upgrading it if necessary. Applies the
5706 supplied C<vtable> and returns a pointer to the magic added.
5708 Note that C<sv_magicext> will allow things that C<sv_magic> will not.
5709 In particular, you can add magic to C<SvREADONLY> SVs, and add more than
5710 one instance of the same C<how>.
5712 If C<namlen> is greater than zero then a C<savepvn> I<copy> of C<name> is
5713 stored, if C<namlen> is zero then C<name> is stored as-is and - as another
5714 special case - if C<(name && namlen == HEf_SVKEY)> then C<name> is assumed
5715 to contain an SV* and is stored as-is with its C<REFCNT> incremented.
5717 (This is now used as a subroutine by C<sv_magic>.)
5722 Perl_sv_magicext(pTHX_ SV *const sv, SV *const obj, const int how,
5723 const MGVTBL *const vtable, const char *const name, const I32 namlen)
5727 PERL_ARGS_ASSERT_SV_MAGICEXT;
5729 SvUPGRADE(sv, SVt_PVMG);
5730 Newxz(mg, 1, MAGIC);
5731 mg->mg_moremagic = SvMAGIC(sv);
5732 SvMAGIC_set(sv, mg);
5734 /* Sometimes a magic contains a reference loop, where the sv and
5735 object refer to each other. To prevent a reference loop that
5736 would prevent such objects being freed, we look for such loops
5737 and if we find one we avoid incrementing the object refcount.
5739 Note we cannot do this to avoid self-tie loops as intervening RV must
5740 have its REFCNT incremented to keep it in existence.
5743 if (!obj || obj == sv ||
5744 how == PERL_MAGIC_arylen ||
5745 how == PERL_MAGIC_regdata ||
5746 how == PERL_MAGIC_regdatum ||
5747 how == PERL_MAGIC_symtab ||
5748 (SvTYPE(obj) == SVt_PVGV &&
5749 (GvSV(obj) == sv || GvHV(obj) == (const HV *)sv
5750 || GvAV(obj) == (const AV *)sv || GvCV(obj) == (const CV *)sv
5751 || GvIOp(obj) == (const IO *)sv || GvFORM(obj) == (const CV *)sv)))
5756 mg->mg_obj = SvREFCNT_inc_simple(obj);
5757 mg->mg_flags |= MGf_REFCOUNTED;
5760 /* Normal self-ties simply pass a null object, and instead of
5761 using mg_obj directly, use the SvTIED_obj macro to produce a
5762 new RV as needed. For glob "self-ties", we are tieing the PVIO
5763 with an RV obj pointing to the glob containing the PVIO. In
5764 this case, to avoid a reference loop, we need to weaken the
5768 if (how == PERL_MAGIC_tiedscalar && SvTYPE(sv) == SVt_PVIO &&
5769 obj && SvROK(obj) && GvIO(SvRV(obj)) == (const IO *)sv)
5775 mg->mg_len = namlen;
5778 mg->mg_ptr = savepvn(name, namlen);
5779 else if (namlen == HEf_SVKEY) {
5780 /* Yes, this is casting away const. This is only for the case of
5781 HEf_SVKEY. I think we need to document this aberation of the
5782 constness of the API, rather than making name non-const, as
5783 that change propagating outwards a long way. */
5784 mg->mg_ptr = (char*)SvREFCNT_inc_simple_NN((SV *)name);
5786 mg->mg_ptr = (char *) name;
5788 mg->mg_virtual = (MGVTBL *) vtable;
5795 Perl_sv_magicext_mglob(pTHX_ SV *sv)
5797 PERL_ARGS_ASSERT_SV_MAGICEXT_MGLOB;
5798 if (SvTYPE(sv) == SVt_PVLV && LvTYPE(sv) == 'y') {
5799 /* This sv is only a delegate. //g magic must be attached to
5804 return sv_magicext(sv, NULL, PERL_MAGIC_regex_global,
5805 &PL_vtbl_mglob, 0, 0);
5809 =for apidoc sv_magic
5811 Adds magic to an SV. First upgrades C<sv> to type C<SVt_PVMG> if
5812 necessary, then adds a new magic item of type C<how> to the head of the
5815 See C<L</sv_magicext>> (which C<sv_magic> now calls) for a description of the
5816 handling of the C<name> and C<namlen> arguments.
5818 You need to use C<sv_magicext> to add magic to C<SvREADONLY> SVs and also
5819 to add more than one instance of the same C<how>.
5825 Perl_sv_magic(pTHX_ SV *const sv, SV *const obj, const int how,
5826 const char *const name, const I32 namlen)
5828 const MGVTBL *vtable;
5831 unsigned int vtable_index;
5833 PERL_ARGS_ASSERT_SV_MAGIC;
5835 if (how < 0 || (unsigned)how >= C_ARRAY_LENGTH(PL_magic_data)
5836 || ((flags = PL_magic_data[how]),
5837 (vtable_index = flags & PERL_MAGIC_VTABLE_MASK)
5838 > magic_vtable_max))
5839 Perl_croak(aTHX_ "Don't know how to handle magic of type \\%o", how);
5841 /* PERL_MAGIC_ext is reserved for use by extensions not perl internals.
5842 Useful for attaching extension internal data to perl vars.
5843 Note that multiple extensions may clash if magical scalars
5844 etc holding private data from one are passed to another. */
5846 vtable = (vtable_index == magic_vtable_max)
5847 ? NULL : PL_magic_vtables + vtable_index;
5849 if (SvREADONLY(sv)) {
5851 !PERL_MAGIC_TYPE_READONLY_ACCEPTABLE(how)
5854 Perl_croak_no_modify();
5857 if (SvMAGICAL(sv) || (how == PERL_MAGIC_taint && SvTYPE(sv) >= SVt_PVMG)) {
5858 if (SvMAGIC(sv) && (mg = mg_find(sv, how))) {
5859 /* sv_magic() refuses to add a magic of the same 'how' as an
5862 if (how == PERL_MAGIC_taint)
5868 /* Force pos to be stored as characters, not bytes. */
5869 if (SvMAGICAL(sv) && DO_UTF8(sv)
5870 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
5872 && mg->mg_flags & MGf_BYTES) {
5873 mg->mg_len = (SSize_t)sv_pos_b2u_flags(sv, (STRLEN)mg->mg_len,
5875 mg->mg_flags &= ~MGf_BYTES;
5878 /* Rest of work is done else where */
5879 mg = sv_magicext(sv,obj,how,vtable,name,namlen);
5882 case PERL_MAGIC_taint:
5885 case PERL_MAGIC_ext:
5886 case PERL_MAGIC_dbfile:
5893 S_sv_unmagicext_flags(pTHX_ SV *const sv, const int type, MGVTBL *vtbl, const U32 flags)
5900 if (SvTYPE(sv) < SVt_PVMG || !SvMAGIC(sv))
5902 mgp = &(((XPVMG*) SvANY(sv))->xmg_u.xmg_magic);
5903 for (mg = *mgp; mg; mg = *mgp) {
5904 const MGVTBL* const virt = mg->mg_virtual;
5905 if (mg->mg_type == type && (!flags || virt == vtbl)) {
5906 *mgp = mg->mg_moremagic;
5907 if (virt && virt->svt_free)
5908 virt->svt_free(aTHX_ sv, mg);
5909 if (mg->mg_ptr && mg->mg_type != PERL_MAGIC_regex_global) {
5911 Safefree(mg->mg_ptr);
5912 else if (mg->mg_len == HEf_SVKEY)
5913 SvREFCNT_dec(MUTABLE_SV(mg->mg_ptr));
5914 else if (mg->mg_type == PERL_MAGIC_utf8)
5915 Safefree(mg->mg_ptr);
5917 if (mg->mg_flags & MGf_REFCOUNTED)
5918 SvREFCNT_dec(mg->mg_obj);
5922 mgp = &mg->mg_moremagic;
5925 if (SvMAGICAL(sv)) /* if we're under save_magic, wait for restore_magic; */
5926 mg_magical(sv); /* else fix the flags now */
5935 =for apidoc sv_unmagic
5937 Removes all magic of type C<type> from an SV.
5943 Perl_sv_unmagic(pTHX_ SV *const sv, const int type)
5945 PERL_ARGS_ASSERT_SV_UNMAGIC;
5946 return S_sv_unmagicext_flags(aTHX_ sv, type, NULL, 0);
5950 =for apidoc sv_unmagicext
5952 Removes all magic of type C<type> with the specified C<vtbl> from an SV.
5958 Perl_sv_unmagicext(pTHX_ SV *const sv, const int type, MGVTBL *vtbl)
5960 PERL_ARGS_ASSERT_SV_UNMAGICEXT;
5961 return S_sv_unmagicext_flags(aTHX_ sv, type, vtbl, 1);
5965 =for apidoc sv_rvweaken
5967 Weaken a reference: set the C<SvWEAKREF> flag on this RV; give the
5968 referred-to SV C<PERL_MAGIC_backref> magic if it hasn't already; and
5969 push a back-reference to this RV onto the array of backreferences
5970 associated with that magic. If the RV is magical, set magic will be
5971 called after the RV is cleared. Silently ignores C<undef> and warns
5972 on already-weak references.
5978 Perl_sv_rvweaken(pTHX_ SV *const sv)
5982 PERL_ARGS_ASSERT_SV_RVWEAKEN;
5984 if (!SvOK(sv)) /* let undefs pass */
5987 Perl_croak(aTHX_ "Can't weaken a nonreference");
5988 else if (SvWEAKREF(sv)) {
5989 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is already weak");
5992 else if (SvREADONLY(sv)) croak_no_modify();
5994 Perl_sv_add_backref(aTHX_ tsv, sv);
5996 SvREFCNT_dec_NN(tsv);
6001 =for apidoc sv_rvunweaken
6003 Unweaken a reference: Clear the C<SvWEAKREF> flag on this RV; remove
6004 the backreference to this RV from the array of backreferences
6005 associated with the target SV, increment the refcount of the target.
6006 Silently ignores C<undef> and warns on non-weak references.
6012 Perl_sv_rvunweaken(pTHX_ SV *const sv)
6016 PERL_ARGS_ASSERT_SV_RVUNWEAKEN;
6018 if (!SvOK(sv)) /* let undefs pass */
6021 Perl_croak(aTHX_ "Can't unweaken a nonreference");
6022 else if (!SvWEAKREF(sv)) {
6023 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is not weak");
6026 else if (SvREADONLY(sv)) croak_no_modify();
6031 SvREFCNT_inc_NN(tsv);
6032 Perl_sv_del_backref(aTHX_ tsv, sv);
6037 =for apidoc sv_get_backrefs
6039 If C<sv> is the target of a weak reference then it returns the back
6040 references structure associated with the sv; otherwise return C<NULL>.
6042 When returning a non-null result the type of the return is relevant. If it
6043 is an AV then the elements of the AV are the weak reference RVs which
6044 point at this item. If it is any other type then the item itself is the
6047 See also C<Perl_sv_add_backref()>, C<Perl_sv_del_backref()>,
6048 C<Perl_sv_kill_backrefs()>
6054 Perl_sv_get_backrefs(SV *const sv)
6058 PERL_ARGS_ASSERT_SV_GET_BACKREFS;
6060 /* find slot to store array or singleton backref */
6062 if (SvTYPE(sv) == SVt_PVHV) {
6064 struct xpvhv_aux * const iter = HvAUX((HV *)sv);
6065 backrefs = (SV *)iter->xhv_backreferences;
6067 } else if (SvMAGICAL(sv)) {
6068 MAGIC *mg = mg_find(sv, PERL_MAGIC_backref);
6070 backrefs = mg->mg_obj;
6075 /* Give tsv backref magic if it hasn't already got it, then push a
6076 * back-reference to sv onto the array associated with the backref magic.
6078 * As an optimisation, if there's only one backref and it's not an AV,
6079 * store it directly in the HvAUX or mg_obj slot, avoiding the need to
6080 * allocate an AV. (Whether the slot holds an AV tells us whether this is
6084 /* A discussion about the backreferences array and its refcount:
6086 * The AV holding the backreferences is pointed to either as the mg_obj of
6087 * PERL_MAGIC_backref, or in the specific case of a HV, from the
6088 * xhv_backreferences field. The array is created with a refcount
6089 * of 2. This means that if during global destruction the array gets
6090 * picked on before its parent to have its refcount decremented by the
6091 * random zapper, it won't actually be freed, meaning it's still there for
6092 * when its parent gets freed.
6094 * When the parent SV is freed, the extra ref is killed by
6095 * Perl_sv_kill_backrefs. The other ref is killed, in the case of magic,
6096 * by mg_free() / MGf_REFCOUNTED, or for a hash, by Perl_hv_kill_backrefs.
6098 * When a single backref SV is stored directly, it is not reference
6103 Perl_sv_add_backref(pTHX_ SV *const tsv, SV *const sv)
6109 PERL_ARGS_ASSERT_SV_ADD_BACKREF;
6111 /* find slot to store array or singleton backref */
6113 if (SvTYPE(tsv) == SVt_PVHV) {
6114 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
6117 mg = mg_find(tsv, PERL_MAGIC_backref);
6119 mg = sv_magicext(tsv, NULL, PERL_MAGIC_backref, &PL_vtbl_backref, NULL, 0);
6120 svp = &(mg->mg_obj);
6123 /* create or retrieve the array */
6125 if ( (!*svp && SvTYPE(sv) == SVt_PVAV)
6126 || (*svp && SvTYPE(*svp) != SVt_PVAV)
6130 mg->mg_flags |= MGf_REFCOUNTED;
6133 SvREFCNT_inc_simple_void_NN(av);
6134 /* av now has a refcnt of 2; see discussion above */
6135 av_extend(av, *svp ? 2 : 1);
6137 /* move single existing backref to the array */
6138 AvARRAY(av)[++AvFILLp(av)] = *svp; /* av_push() */
6143 av = MUTABLE_AV(*svp);
6145 /* optimisation: store single backref directly in HvAUX or mg_obj */
6149 assert(SvTYPE(av) == SVt_PVAV);
6150 if (AvFILLp(av) >= AvMAX(av)) {
6151 av_extend(av, AvFILLp(av)+1);
6154 /* push new backref */
6155 AvARRAY(av)[++AvFILLp(av)] = sv; /* av_push() */
6158 /* delete a back-reference to ourselves from the backref magic associated
6159 * with the SV we point to.
6163 Perl_sv_del_backref(pTHX_ SV *const tsv, SV *const sv)
6167 PERL_ARGS_ASSERT_SV_DEL_BACKREF;
6169 if (SvTYPE(tsv) == SVt_PVHV) {
6171 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
6173 else if (SvIS_FREED(tsv) && PL_phase == PERL_PHASE_DESTRUCT) {
6174 /* It's possible for the the last (strong) reference to tsv to have
6175 become freed *before* the last thing holding a weak reference.
6176 If both survive longer than the backreferences array, then when
6177 the referent's reference count drops to 0 and it is freed, it's
6178 not able to chase the backreferences, so they aren't NULLed.
6180 For example, a CV holds a weak reference to its stash. If both the
6181 CV and the stash survive longer than the backreferences array,
6182 and the CV gets picked for the SvBREAK() treatment first,
6183 *and* it turns out that the stash is only being kept alive because
6184 of an our variable in the pad of the CV, then midway during CV
6185 destruction the stash gets freed, but CvSTASH() isn't set to NULL.
6186 It ends up pointing to the freed HV. Hence it's chased in here, and
6187 if this block wasn't here, it would hit the !svp panic just below.
6189 I don't believe that "better" destruction ordering is going to help
6190 here - during global destruction there's always going to be the
6191 chance that something goes out of order. We've tried to make it
6192 foolproof before, and it only resulted in evolutionary pressure on
6193 fools. Which made us look foolish for our hubris. :-(
6199 = SvMAGICAL(tsv) ? mg_find(tsv, PERL_MAGIC_backref) : NULL;
6200 svp = mg ? &(mg->mg_obj) : NULL;
6204 Perl_croak(aTHX_ "panic: del_backref, svp=0");
6206 /* It's possible that sv is being freed recursively part way through the
6207 freeing of tsv. If this happens, the backreferences array of tsv has
6208 already been freed, and so svp will be NULL. If this is the case,
6209 we should not panic. Instead, nothing needs doing, so return. */
6210 if (PL_phase == PERL_PHASE_DESTRUCT && SvREFCNT(tsv) == 0)
6212 Perl_croak(aTHX_ "panic: del_backref, *svp=%p phase=%s refcnt=%" UVuf,
6213 (void*)*svp, PL_phase_names[PL_phase], (UV)SvREFCNT(tsv));
6216 if (SvTYPE(*svp) == SVt_PVAV) {
6220 AV * const av = (AV*)*svp;
6222 assert(!SvIS_FREED(av));
6226 /* for an SV with N weak references to it, if all those
6227 * weak refs are deleted, then sv_del_backref will be called
6228 * N times and O(N^2) compares will be done within the backref
6229 * array. To ameliorate this potential slowness, we:
6230 * 1) make sure this code is as tight as possible;
6231 * 2) when looking for SV, look for it at both the head and tail of the
6232 * array first before searching the rest, since some create/destroy
6233 * patterns will cause the backrefs to be freed in order.
6240 SV **p = &svp[fill];
6241 SV *const topsv = *p;
6248 /* We weren't the last entry.
6249 An unordered list has this property that you
6250 can take the last element off the end to fill
6251 the hole, and it's still an unordered list :-)
6257 break; /* should only be one */
6264 AvFILLp(av) = fill-1;
6266 else if (SvIS_FREED(*svp) && PL_phase == PERL_PHASE_DESTRUCT) {
6267 /* freed AV; skip */
6270 /* optimisation: only a single backref, stored directly */
6272 Perl_croak(aTHX_ "panic: del_backref, *svp=%p, sv=%p",
6273 (void*)*svp, (void*)sv);
6280 Perl_sv_kill_backrefs(pTHX_ SV *const sv, AV *const av)
6286 PERL_ARGS_ASSERT_SV_KILL_BACKREFS;
6291 /* after multiple passes through Perl_sv_clean_all() for a thingy
6292 * that has badly leaked, the backref array may have gotten freed,
6293 * since we only protect it against 1 round of cleanup */
6294 if (SvIS_FREED(av)) {
6295 if (PL_in_clean_all) /* All is fair */
6298 "panic: magic_killbackrefs (freed backref AV/SV)");
6302 is_array = (SvTYPE(av) == SVt_PVAV);
6304 assert(!SvIS_FREED(av));
6307 last = svp + AvFILLp(av);
6310 /* optimisation: only a single backref, stored directly */
6316 while (svp <= last) {
6318 SV *const referrer = *svp;
6319 if (SvWEAKREF(referrer)) {
6320 /* XXX Should we check that it hasn't changed? */
6321 assert(SvROK(referrer));
6322 SvRV_set(referrer, 0);
6324 SvWEAKREF_off(referrer);
6325 SvSETMAGIC(referrer);
6326 } else if (SvTYPE(referrer) == SVt_PVGV ||
6327 SvTYPE(referrer) == SVt_PVLV) {
6328 assert(SvTYPE(sv) == SVt_PVHV); /* stash backref */
6329 /* You lookin' at me? */
6330 assert(GvSTASH(referrer));
6331 assert(GvSTASH(referrer) == (const HV *)sv);
6332 GvSTASH(referrer) = 0;
6333 } else if (SvTYPE(referrer) == SVt_PVCV ||
6334 SvTYPE(referrer) == SVt_PVFM) {
6335 if (SvTYPE(sv) == SVt_PVHV) { /* stash backref */
6336 /* You lookin' at me? */
6337 assert(CvSTASH(referrer));
6338 assert(CvSTASH(referrer) == (const HV *)sv);
6339 SvANY(MUTABLE_CV(referrer))->xcv_stash = 0;
6342 assert(SvTYPE(sv) == SVt_PVGV);
6343 /* You lookin' at me? */
6344 assert(CvGV(referrer));
6345 assert(CvGV(referrer) == (const GV *)sv);
6346 anonymise_cv_maybe(MUTABLE_GV(sv),
6347 MUTABLE_CV(referrer));
6352 "panic: magic_killbackrefs (flags=%" UVxf ")",
6353 (UV)SvFLAGS(referrer));
6364 SvREFCNT_dec_NN(av); /* remove extra count added by sv_add_backref() */
6370 =for apidoc sv_insert
6372 Inserts a string at the specified offset/length within the SV. Similar to
6373 the Perl C<substr()> function. Handles get magic.
6375 =for apidoc sv_insert_flags
6377 Same as C<sv_insert>, but the extra C<flags> are passed to the
6378 C<SvPV_force_flags> that applies to C<bigstr>.
6384 Perl_sv_insert_flags(pTHX_ SV *const bigstr, const STRLEN offset, const STRLEN len, const char *little, const STRLEN littlelen, const U32 flags)
6390 SSize_t i; /* better be sizeof(STRLEN) or bad things happen */
6393 PERL_ARGS_ASSERT_SV_INSERT_FLAGS;
6395 SvPV_force_flags(bigstr, curlen, flags);
6396 (void)SvPOK_only_UTF8(bigstr);
6398 if (little >= SvPVX(bigstr) &&
6399 little < SvPVX(bigstr) + (SvLEN(bigstr) ? SvLEN(bigstr) : SvCUR(bigstr))) {
6400 /* little is a pointer to within bigstr, since we can reallocate bigstr,
6401 or little...little+littlelen might overlap offset...offset+len we make a copy
6403 little = savepvn(little, littlelen);
6407 if (offset + len > curlen) {
6408 SvGROW(bigstr, offset+len+1);
6409 Zero(SvPVX(bigstr)+curlen, offset+len-curlen, char);
6410 SvCUR_set(bigstr, offset+len);
6414 i = littlelen - len;
6415 if (i > 0) { /* string might grow */
6416 big = SvGROW(bigstr, SvCUR(bigstr) + i + 1);
6417 mid = big + offset + len;
6418 midend = bigend = big + SvCUR(bigstr);
6421 while (midend > mid) /* shove everything down */
6422 *--bigend = *--midend;
6423 Move(little,big+offset,littlelen,char);
6424 SvCUR_set(bigstr, SvCUR(bigstr) + i);
6429 Move(little,SvPVX(bigstr)+offset,len,char);
6434 big = SvPVX(bigstr);
6437 bigend = big + SvCUR(bigstr);
6439 if (midend > bigend)
6440 Perl_croak(aTHX_ "panic: sv_insert, midend=%p, bigend=%p",
6443 if (mid - big > bigend - midend) { /* faster to shorten from end */
6445 Move(little, mid, littlelen,char);
6448 i = bigend - midend;
6450 Move(midend, mid, i,char);
6454 SvCUR_set(bigstr, mid - big);
6456 else if ((i = mid - big)) { /* faster from front */
6457 midend -= littlelen;
6459 Move(big, midend - i, i, char);
6460 sv_chop(bigstr,midend-i);
6462 Move(little, mid, littlelen,char);
6464 else if (littlelen) {
6465 midend -= littlelen;
6466 sv_chop(bigstr,midend);
6467 Move(little,midend,littlelen,char);
6470 sv_chop(bigstr,midend);
6476 =for apidoc sv_replace
6478 Make the first argument a copy of the second, then delete the original.
6479 The target SV physically takes over ownership of the body of the source SV
6480 and inherits its flags; however, the target keeps any magic it owns,
6481 and any magic in the source is discarded.
6482 Note that this is a rather specialist SV copying operation; most of the
6483 time you'll want to use C<sv_setsv> or one of its many macro front-ends.
6489 Perl_sv_replace(pTHX_ SV *const sv, SV *const nsv)
6491 const U32 refcnt = SvREFCNT(sv);
6493 PERL_ARGS_ASSERT_SV_REPLACE;
6495 SV_CHECK_THINKFIRST_COW_DROP(sv);
6496 if (SvREFCNT(nsv) != 1) {
6497 Perl_croak(aTHX_ "panic: reference miscount on nsv in sv_replace()"
6498 " (%" UVuf " != 1)", (UV) SvREFCNT(nsv));
6500 if (SvMAGICAL(sv)) {
6504 sv_upgrade(nsv, SVt_PVMG);
6505 SvMAGIC_set(nsv, SvMAGIC(sv));
6506 SvFLAGS(nsv) |= SvMAGICAL(sv);
6508 SvMAGIC_set(sv, NULL);
6512 assert(!SvREFCNT(sv));
6513 #ifdef DEBUG_LEAKING_SCALARS
6514 sv->sv_flags = nsv->sv_flags;
6515 sv->sv_any = nsv->sv_any;
6516 sv->sv_refcnt = nsv->sv_refcnt;
6517 sv->sv_u = nsv->sv_u;
6519 StructCopy(nsv,sv,SV);
6521 if(SvTYPE(sv) == SVt_IV) {
6522 SET_SVANY_FOR_BODYLESS_IV(sv);
6526 SvREFCNT(sv) = refcnt;
6527 SvFLAGS(nsv) |= SVTYPEMASK; /* Mark as freed */
6532 /* We're about to free a GV which has a CV that refers back to us.
6533 * If that CV will outlive us, make it anonymous (i.e. fix up its CvGV
6537 S_anonymise_cv_maybe(pTHX_ GV *gv, CV* cv)
6542 PERL_ARGS_ASSERT_ANONYMISE_CV_MAYBE;
6545 assert(SvREFCNT(gv) == 0);
6546 assert(isGV(gv) && isGV_with_GP(gv));
6548 assert(!CvANON(cv));
6549 assert(CvGV(cv) == gv);
6550 assert(!CvNAMED(cv));
6552 /* will the CV shortly be freed by gp_free() ? */
6553 if (GvCV(gv) == cv && GvGP(gv)->gp_refcnt < 2 && SvREFCNT(cv) < 2) {
6554 SvANY(cv)->xcv_gv_u.xcv_gv = NULL;
6558 /* if not, anonymise: */
6559 gvname = (GvSTASH(gv) && HvNAME(GvSTASH(gv)) && HvENAME(GvSTASH(gv)))
6560 ? newSVhek(HvENAME_HEK(GvSTASH(gv)))
6561 : newSVpvn_flags( "__ANON__", 8, 0 );
6562 sv_catpvs(gvname, "::__ANON__");
6563 anongv = gv_fetchsv(gvname, GV_ADDMULTI, SVt_PVCV);
6564 SvREFCNT_dec_NN(gvname);
6568 SvANY(cv)->xcv_gv_u.xcv_gv = MUTABLE_GV(SvREFCNT_inc(anongv));
6573 =for apidoc sv_clear
6575 Clear an SV: call any destructors, free up any memory used by the body,
6576 and free the body itself. The SV's head is I<not> freed, although
6577 its type is set to all 1's so that it won't inadvertently be assumed
6578 to be live during global destruction etc.
6579 This function should only be called when C<REFCNT> is zero. Most of the time
6580 you'll want to call C<sv_free()> (or its macro wrapper C<SvREFCNT_dec>)
6587 Perl_sv_clear(pTHX_ SV *const orig_sv)
6592 const struct body_details *sv_type_details;
6596 STRLEN hash_index = 0; /* initialise to make Coverity et al happy.
6597 Not strictly necessary */
6599 PERL_ARGS_ASSERT_SV_CLEAR;
6601 /* within this loop, sv is the SV currently being freed, and
6602 * iter_sv is the most recent AV or whatever that's being iterated
6603 * over to provide more SVs */
6609 assert(SvREFCNT(sv) == 0);
6610 assert(SvTYPE(sv) != (svtype)SVTYPEMASK);
6612 if (type <= SVt_IV) {
6613 /* See the comment in sv.h about the collusion between this
6614 * early return and the overloading of the NULL slots in the
6618 SvFLAGS(sv) &= SVf_BREAK;
6619 SvFLAGS(sv) |= SVTYPEMASK;
6623 /* objs are always >= MG, but pad names use the SVs_OBJECT flag
6624 for another purpose */
6625 assert(!SvOBJECT(sv) || type >= SVt_PVMG);
6627 if (type >= SVt_PVMG) {
6629 if (!curse(sv, 1)) goto get_next_sv;
6630 type = SvTYPE(sv); /* destructor may have changed it */
6632 /* Free back-references before magic, in case the magic calls
6633 * Perl code that has weak references to sv. */
6634 if (type == SVt_PVHV) {
6635 Perl_hv_kill_backrefs(aTHX_ MUTABLE_HV(sv));
6639 else if (SvMAGIC(sv)) {
6640 /* Free back-references before other types of magic. */
6641 sv_unmagic(sv, PERL_MAGIC_backref);
6647 /* case SVt_INVLIST: */
6650 IoIFP(sv) != PerlIO_stdin() &&
6651 IoIFP(sv) != PerlIO_stdout() &&
6652 IoIFP(sv) != PerlIO_stderr() &&
6653 !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6655 io_close(MUTABLE_IO(sv), NULL, FALSE,
6656 (IoTYPE(sv) == IoTYPE_WRONLY ||
6657 IoTYPE(sv) == IoTYPE_RDWR ||
6658 IoTYPE(sv) == IoTYPE_APPEND));
6660 if (IoDIRP(sv) && !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6661 PerlDir_close(IoDIRP(sv));
6662 IoDIRP(sv) = (DIR*)NULL;
6663 Safefree(IoTOP_NAME(sv));
6664 Safefree(IoFMT_NAME(sv));
6665 Safefree(IoBOTTOM_NAME(sv));
6666 if ((const GV *)sv == PL_statgv)
6670 /* FIXME for plugins */
6671 pregfree2((REGEXP*) sv);
6675 cv_undef(MUTABLE_CV(sv));
6676 /* If we're in a stash, we don't own a reference to it.
6677 * However it does have a back reference to us, which needs to
6679 if ((stash = CvSTASH(sv)))
6680 sv_del_backref(MUTABLE_SV(stash), sv);
6683 if (PL_last_swash_hv == (const HV *)sv) {
6684 PL_last_swash_hv = NULL;
6686 if (HvTOTALKEYS((HV*)sv) > 0) {
6688 /* this statement should match the one at the beginning of
6689 * hv_undef_flags() */
6690 if ( PL_phase != PERL_PHASE_DESTRUCT
6691 && (hek = HvNAME_HEK((HV*)sv)))
6693 if (PL_stashcache) {
6694 DEBUG_o(Perl_deb(aTHX_
6695 "sv_clear clearing PL_stashcache for '%" HEKf
6698 (void)hv_deletehek(PL_stashcache,
6701 hv_name_set((HV*)sv, NULL, 0, 0);
6704 /* save old iter_sv in unused SvSTASH field */
6705 assert(!SvOBJECT(sv));
6706 SvSTASH(sv) = (HV*)iter_sv;
6709 /* save old hash_index in unused SvMAGIC field */
6710 assert(!SvMAGICAL(sv));
6711 assert(!SvMAGIC(sv));
6712 ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index = hash_index;
6715 next_sv = Perl_hfree_next_entry(aTHX_ (HV*)sv, &hash_index);
6716 goto get_next_sv; /* process this new sv */
6718 /* free empty hash */
6719 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6720 assert(!HvARRAY((HV*)sv));
6724 AV* av = MUTABLE_AV(sv);
6725 if (PL_comppad == av) {
6729 if (AvREAL(av) && AvFILLp(av) > -1) {
6730 next_sv = AvARRAY(av)[AvFILLp(av)--];
6731 /* save old iter_sv in top-most slot of AV,
6732 * and pray that it doesn't get wiped in the meantime */
6733 AvARRAY(av)[AvMAX(av)] = iter_sv;
6735 goto get_next_sv; /* process this new sv */
6737 Safefree(AvALLOC(av));
6742 if (LvTYPE(sv) == 'T') { /* for tie: return HE to pool */
6743 SvREFCNT_dec(HeKEY_sv((HE*)LvTARG(sv)));
6744 HeNEXT((HE*)LvTARG(sv)) = PL_hv_fetch_ent_mh;
6745 PL_hv_fetch_ent_mh = (HE*)LvTARG(sv);
6747 else if (LvTYPE(sv) != 't') /* unless tie: unrefcnted fake SV** */
6748 SvREFCNT_dec(LvTARG(sv));
6750 /* SvLEN points to a regex body. Free the body, then
6751 * set SvLEN to whatever value was in the now-freed
6752 * regex body. The PVX buffer is shared by multiple re's
6753 * and only freed once, by the re whose len in non-null */
6754 STRLEN len = ReANY(sv)->xpv_len;
6755 pregfree2((REGEXP*) sv);
6756 SvLEN_set((sv), len);
6761 if (isGV_with_GP(sv)) {
6762 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
6763 && HvENAME_get(stash))
6764 mro_method_changed_in(stash);
6765 gp_free(MUTABLE_GV(sv));
6767 unshare_hek(GvNAME_HEK(sv));
6768 /* If we're in a stash, we don't own a reference to it.
6769 * However it does have a back reference to us, which
6770 * needs to be cleared. */
6771 if ((stash = GvSTASH(sv)))
6772 sv_del_backref(MUTABLE_SV(stash), sv);
6774 /* FIXME. There are probably more unreferenced pointers to SVs
6775 * in the interpreter struct that we should check and tidy in
6776 * a similar fashion to this: */
6777 /* See also S_sv_unglob, which does the same thing. */
6778 if ((const GV *)sv == PL_last_in_gv)
6779 PL_last_in_gv = NULL;
6780 else if ((const GV *)sv == PL_statgv)
6782 else if ((const GV *)sv == PL_stderrgv)
6791 /* Don't bother with SvOOK_off(sv); as we're only going to
6795 SvOOK_offset(sv, offset);
6796 SvPV_set(sv, SvPVX_mutable(sv) - offset);
6797 /* Don't even bother with turning off the OOK flag. */
6802 SV * const target = SvRV(sv);
6804 sv_del_backref(target, sv);
6810 else if (SvPVX_const(sv)
6811 && !(SvTYPE(sv) == SVt_PVIO
6812 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6817 PerlIO_printf(Perl_debug_log, "Copy on write: clear\n");
6822 if (CowREFCNT(sv)) {
6829 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6834 Safefree(SvPVX_mutable(sv));
6838 else if (SvPVX_const(sv) && SvLEN(sv)
6839 && !(SvTYPE(sv) == SVt_PVIO
6840 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6841 Safefree(SvPVX_mutable(sv));
6842 else if (SvPVX_const(sv) && SvIsCOW(sv)) {
6843 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6853 SvFLAGS(sv) &= SVf_BREAK;
6854 SvFLAGS(sv) |= SVTYPEMASK;
6856 sv_type_details = bodies_by_type + type;
6857 if (sv_type_details->arena) {
6858 del_body(((char *)SvANY(sv) + sv_type_details->offset),
6859 &PL_body_roots[type]);
6861 else if (sv_type_details->body_size) {
6862 safefree(SvANY(sv));
6866 /* caller is responsible for freeing the head of the original sv */
6867 if (sv != orig_sv && !SvREFCNT(sv))
6870 /* grab and free next sv, if any */
6878 else if (!iter_sv) {
6880 } else if (SvTYPE(iter_sv) == SVt_PVAV) {
6881 AV *const av = (AV*)iter_sv;
6882 if (AvFILLp(av) > -1) {
6883 sv = AvARRAY(av)[AvFILLp(av)--];
6885 else { /* no more elements of current AV to free */
6888 /* restore previous value, squirrelled away */
6889 iter_sv = AvARRAY(av)[AvMAX(av)];
6890 Safefree(AvALLOC(av));
6893 } else if (SvTYPE(iter_sv) == SVt_PVHV) {
6894 sv = Perl_hfree_next_entry(aTHX_ (HV*)iter_sv, &hash_index);
6895 if (!sv && !HvTOTALKEYS((HV *)iter_sv)) {
6896 /* no more elements of current HV to free */
6899 /* Restore previous values of iter_sv and hash_index,
6900 * squirrelled away */
6901 assert(!SvOBJECT(sv));
6902 iter_sv = (SV*)SvSTASH(sv);
6903 assert(!SvMAGICAL(sv));
6904 hash_index = ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index;
6906 /* perl -DA does not like rubbish in SvMAGIC. */
6910 /* free any remaining detritus from the hash struct */
6911 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6912 assert(!HvARRAY((HV*)sv));
6917 /* unrolled SvREFCNT_dec and sv_free2 follows: */
6921 if (!SvREFCNT(sv)) {
6925 if (--(SvREFCNT(sv)))
6929 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6930 "Attempt to free temp prematurely: SV 0x%" UVxf
6931 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6935 if (SvIMMORTAL(sv)) {
6936 /* make sure SvREFCNT(sv)==0 happens very seldom */
6937 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6946 /* This routine curses the sv itself, not the object referenced by sv. So
6947 sv does not have to be ROK. */
6950 S_curse(pTHX_ SV * const sv, const bool check_refcnt) {
6951 PERL_ARGS_ASSERT_CURSE;
6952 assert(SvOBJECT(sv));
6954 if (PL_defstash && /* Still have a symbol table? */
6960 stash = SvSTASH(sv);
6961 assert(SvTYPE(stash) == SVt_PVHV);
6962 if (HvNAME(stash)) {
6963 CV* destructor = NULL;
6964 struct mro_meta *meta;
6966 assert (SvOOK(stash));
6968 DEBUG_o( Perl_deb(aTHX_ "Looking for DESTROY method for %s\n",
6971 /* don't make this an initialization above the assert, since it needs
6973 meta = HvMROMETA(stash);
6974 if (meta->destroy_gen && meta->destroy_gen == PL_sub_generation) {
6975 destructor = meta->destroy;
6976 DEBUG_o( Perl_deb(aTHX_ "Using cached DESTROY method %p for %s\n",
6977 (void *)destructor, HvNAME(stash)) );
6980 bool autoload = FALSE;
6982 gv_fetchmeth_pvn(stash, S_destroy, S_destroy_len, -1, 0);
6984 destructor = GvCV(gv);
6986 gv = gv_autoload_pvn(stash, S_destroy, S_destroy_len,
6987 GV_AUTOLOAD_ISMETHOD);
6989 destructor = GvCV(gv);
6993 /* we don't cache AUTOLOAD for DESTROY, since this code
6994 would then need to set $__PACKAGE__::AUTOLOAD, or the
6995 equivalent for XS AUTOLOADs */
6997 meta->destroy_gen = PL_sub_generation;
6998 meta->destroy = destructor;
7000 DEBUG_o( Perl_deb(aTHX_ "Set cached DESTROY method %p for %s\n",
7001 (void *)destructor, HvNAME(stash)) );
7004 DEBUG_o( Perl_deb(aTHX_ "Not caching AUTOLOAD for DESTROY method for %s\n",
7008 assert(!destructor || SvTYPE(destructor) == SVt_PVCV);
7010 /* A constant subroutine can have no side effects, so
7011 don't bother calling it. */
7012 && !CvCONST(destructor)
7013 /* Don't bother calling an empty destructor or one that
7014 returns immediately. */
7015 && (CvISXSUB(destructor)
7016 || (CvSTART(destructor)
7017 && (CvSTART(destructor)->op_next->op_type
7019 && (CvSTART(destructor)->op_next->op_type
7021 || CvSTART(destructor)->op_next->op_next->op_type
7027 SV* const tmpref = newRV(sv);
7028 SvREADONLY_on(tmpref); /* DESTROY() could be naughty */
7030 PUSHSTACKi(PERLSI_DESTROY);
7035 call_sv(MUTABLE_SV(destructor),
7036 G_DISCARD|G_EVAL|G_KEEPERR|G_VOID);
7040 if(SvREFCNT(tmpref) < 2) {
7041 /* tmpref is not kept alive! */
7043 SvRV_set(tmpref, NULL);
7046 SvREFCNT_dec_NN(tmpref);
7049 } while (SvOBJECT(sv) && SvSTASH(sv) != stash);
7052 if (check_refcnt && SvREFCNT(sv)) {
7053 if (PL_in_clean_objs)
7055 "DESTROY created new reference to dead object '%" HEKf "'",
7056 HEKfARG(HvNAME_HEK(stash)));
7057 /* DESTROY gave object new lease on life */
7063 HV * const stash = SvSTASH(sv);
7064 /* Curse before freeing the stash, as freeing the stash could cause
7065 a recursive call into S_curse. */
7066 SvOBJECT_off(sv); /* Curse the object. */
7067 SvSTASH_set(sv,0); /* SvREFCNT_dec may try to read this */
7068 SvREFCNT_dec(stash); /* possibly of changed persuasion */
7074 =for apidoc sv_newref
7076 Increment an SV's reference count. Use the C<SvREFCNT_inc()> wrapper
7083 Perl_sv_newref(pTHX_ SV *const sv)
7085 PERL_UNUSED_CONTEXT;
7094 Decrement an SV's reference count, and if it drops to zero, call
7095 C<sv_clear> to invoke destructors and free up any memory used by
7096 the body; finally, deallocating the SV's head itself.
7097 Normally called via a wrapper macro C<SvREFCNT_dec>.
7103 Perl_sv_free(pTHX_ SV *const sv)
7109 /* Private helper function for SvREFCNT_dec().
7110 * Called with rc set to original SvREFCNT(sv), where rc == 0 or 1 */
7113 Perl_sv_free2(pTHX_ SV *const sv, const U32 rc)
7117 PERL_ARGS_ASSERT_SV_FREE2;
7119 if (LIKELY( rc == 1 )) {
7125 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
7126 "Attempt to free temp prematurely: SV 0x%" UVxf
7127 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
7131 if (SvIMMORTAL(sv)) {
7132 /* make sure SvREFCNT(sv)==0 happens very seldom */
7133 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
7137 if (! SvREFCNT(sv)) /* may have have been resurrected */
7142 /* handle exceptional cases */
7146 if (SvFLAGS(sv) & SVf_BREAK)
7147 /* this SV's refcnt has been artificially decremented to
7148 * trigger cleanup */
7150 if (PL_in_clean_all) /* All is fair */
7152 if (SvIMMORTAL(sv)) {
7153 /* make sure SvREFCNT(sv)==0 happens very seldom */
7154 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
7157 if (ckWARN_d(WARN_INTERNAL)) {
7158 #ifdef DEBUG_LEAKING_SCALARS_FORK_DUMP
7159 Perl_dump_sv_child(aTHX_ sv);
7161 #ifdef DEBUG_LEAKING_SCALARS
7164 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7165 if (PL_warnhook == PERL_WARNHOOK_FATAL
7166 || ckDEAD(packWARN(WARN_INTERNAL))) {
7167 /* Don't let Perl_warner cause us to escape our fate: */
7171 /* This may not return: */
7172 Perl_warner(aTHX_ packWARN(WARN_INTERNAL),
7173 "Attempt to free unreferenced scalar: SV 0x%" UVxf
7174 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
7177 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7187 Returns the length of the string in the SV. Handles magic and type
7188 coercion and sets the UTF8 flag appropriately. See also C<L</SvCUR>>, which
7189 gives raw access to the C<xpv_cur> slot.
7195 Perl_sv_len(pTHX_ SV *const sv)
7202 (void)SvPV_const(sv, len);
7207 =for apidoc sv_len_utf8
7209 Returns the number of characters in the string in an SV, counting wide
7210 UTF-8 bytes as a single character. Handles magic and type coercion.
7216 * The length is cached in PERL_MAGIC_utf8, in the mg_len field. Also the
7217 * mg_ptr is used, by sv_pos_u2b() and sv_pos_b2u() - see the comments below.
7218 * (Note that the mg_len is not the length of the mg_ptr field.
7219 * This allows the cache to store the character length of the string without
7220 * needing to malloc() extra storage to attach to the mg_ptr.)
7225 Perl_sv_len_utf8(pTHX_ SV *const sv)
7231 return sv_len_utf8_nomg(sv);
7235 Perl_sv_len_utf8_nomg(pTHX_ SV * const sv)
7238 const U8 *s = (U8*)SvPV_nomg_const(sv, len);
7240 PERL_ARGS_ASSERT_SV_LEN_UTF8_NOMG;
7242 if (PL_utf8cache && SvUTF8(sv)) {
7244 MAGIC *mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_utf8) : NULL;
7246 if (mg && (mg->mg_len != -1 || mg->mg_ptr)) {
7247 if (mg->mg_len != -1)
7250 /* We can use the offset cache for a headstart.
7251 The longer value is stored in the first pair. */
7252 STRLEN *cache = (STRLEN *) mg->mg_ptr;
7254 ulen = cache[0] + Perl_utf8_length(aTHX_ s + cache[1],
7258 if (PL_utf8cache < 0) {
7259 const STRLEN real = Perl_utf8_length(aTHX_ s, s + len);
7260 assert_uft8_cache_coherent("sv_len_utf8", ulen, real, sv);
7264 ulen = Perl_utf8_length(aTHX_ s, s + len);
7265 utf8_mg_len_cache_update(sv, &mg, ulen);
7269 return SvUTF8(sv) ? Perl_utf8_length(aTHX_ s, s + len) : len;
7272 /* Walk forwards to find the byte corresponding to the passed in UTF-8
7275 S_sv_pos_u2b_forwards(const U8 *const start, const U8 *const send,
7276 STRLEN *const uoffset_p, bool *const at_end)
7278 const U8 *s = start;
7279 STRLEN uoffset = *uoffset_p;
7281 PERL_ARGS_ASSERT_SV_POS_U2B_FORWARDS;
7283 while (s < send && uoffset) {
7290 else if (s > send) {
7292 /* This is the existing behaviour. Possibly it should be a croak, as
7293 it's actually a bounds error */
7296 *uoffset_p -= uoffset;
7300 /* Given the length of the string in both bytes and UTF-8 characters, decide
7301 whether to walk forwards or backwards to find the byte corresponding to
7302 the passed in UTF-8 offset. */
7304 S_sv_pos_u2b_midway(const U8 *const start, const U8 *send,
7305 STRLEN uoffset, const STRLEN uend)
7307 STRLEN backw = uend - uoffset;
7309 PERL_ARGS_ASSERT_SV_POS_U2B_MIDWAY;
7311 if (uoffset < 2 * backw) {
7312 /* The assumption is that going forwards is twice the speed of going
7313 forward (that's where the 2 * backw comes from).
7314 (The real figure of course depends on the UTF-8 data.) */
7315 const U8 *s = start;
7317 while (s < send && uoffset--)
7327 while (UTF8_IS_CONTINUATION(*send))
7330 return send - start;
7333 /* For the string representation of the given scalar, find the byte
7334 corresponding to the passed in UTF-8 offset. uoffset0 and boffset0
7335 give another position in the string, *before* the sought offset, which
7336 (which is always true, as 0, 0 is a valid pair of positions), which should
7337 help reduce the amount of linear searching.
7338 If *mgp is non-NULL, it should point to the UTF-8 cache magic, which
7339 will be used to reduce the amount of linear searching. The cache will be
7340 created if necessary, and the found value offered to it for update. */
7342 S_sv_pos_u2b_cached(pTHX_ SV *const sv, MAGIC **const mgp, const U8 *const start,
7343 const U8 *const send, STRLEN uoffset,
7344 STRLEN uoffset0, STRLEN boffset0)
7346 STRLEN boffset = 0; /* Actually always set, but let's keep gcc happy. */
7348 bool at_end = FALSE;
7350 PERL_ARGS_ASSERT_SV_POS_U2B_CACHED;
7352 assert (uoffset >= uoffset0);
7357 if (!SvREADONLY(sv) && !SvGMAGICAL(sv) && SvPOK(sv)
7359 && (*mgp || (SvTYPE(sv) >= SVt_PVMG &&
7360 (*mgp = mg_find(sv, PERL_MAGIC_utf8))))) {
7361 if ((*mgp)->mg_ptr) {
7362 STRLEN *cache = (STRLEN *) (*mgp)->mg_ptr;
7363 if (cache[0] == uoffset) {
7364 /* An exact match. */
7367 if (cache[2] == uoffset) {
7368 /* An exact match. */
7372 if (cache[0] < uoffset) {
7373 /* The cache already knows part of the way. */
7374 if (cache[0] > uoffset0) {
7375 /* The cache knows more than the passed in pair */
7376 uoffset0 = cache[0];
7377 boffset0 = cache[1];
7379 if ((*mgp)->mg_len != -1) {
7380 /* And we know the end too. */
7382 + sv_pos_u2b_midway(start + boffset0, send,
7384 (*mgp)->mg_len - uoffset0);
7386 uoffset -= uoffset0;
7388 + sv_pos_u2b_forwards(start + boffset0,
7389 send, &uoffset, &at_end);
7390 uoffset += uoffset0;
7393 else if (cache[2] < uoffset) {
7394 /* We're between the two cache entries. */
7395 if (cache[2] > uoffset0) {
7396 /* and the cache knows more than the passed in pair */
7397 uoffset0 = cache[2];
7398 boffset0 = cache[3];
7402 + sv_pos_u2b_midway(start + boffset0,
7405 cache[0] - uoffset0);
7408 + sv_pos_u2b_midway(start + boffset0,
7411 cache[2] - uoffset0);
7415 else if ((*mgp)->mg_len != -1) {
7416 /* If we can take advantage of a passed in offset, do so. */
7417 /* In fact, offset0 is either 0, or less than offset, so don't
7418 need to worry about the other possibility. */
7420 + sv_pos_u2b_midway(start + boffset0, send,
7422 (*mgp)->mg_len - uoffset0);
7427 if (!found || PL_utf8cache < 0) {
7428 STRLEN real_boffset;
7429 uoffset -= uoffset0;
7430 real_boffset = boffset0 + sv_pos_u2b_forwards(start + boffset0,
7431 send, &uoffset, &at_end);
7432 uoffset += uoffset0;
7434 if (found && PL_utf8cache < 0)
7435 assert_uft8_cache_coherent("sv_pos_u2b_cache", boffset,
7437 boffset = real_boffset;
7440 if (PL_utf8cache && !SvGMAGICAL(sv) && SvPOK(sv)) {
7442 utf8_mg_len_cache_update(sv, mgp, uoffset);
7444 utf8_mg_pos_cache_update(sv, mgp, boffset, uoffset, send - start);
7451 =for apidoc sv_pos_u2b_flags
7453 Converts the offset from a count of UTF-8 chars from
7454 the start of the string, to a count of the equivalent number of bytes; if
7455 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7456 C<offset>, rather than from the start
7457 of the string. Handles type coercion.
7458 C<flags> is passed to C<SvPV_flags>, and usually should be
7459 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7465 * sv_pos_u2b_flags() uses, like sv_pos_b2u(), the mg_ptr of the potential
7466 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7467 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7472 Perl_sv_pos_u2b_flags(pTHX_ SV *const sv, STRLEN uoffset, STRLEN *const lenp,
7479 PERL_ARGS_ASSERT_SV_POS_U2B_FLAGS;
7481 start = (U8*)SvPV_flags(sv, len, flags);
7483 const U8 * const send = start + len;
7485 boffset = sv_pos_u2b_cached(sv, &mg, start, send, uoffset, 0, 0);
7488 && *lenp /* don't bother doing work for 0, as its bytes equivalent
7489 is 0, and *lenp is already set to that. */) {
7490 /* Convert the relative offset to absolute. */
7491 const STRLEN uoffset2 = uoffset + *lenp;
7492 const STRLEN boffset2
7493 = sv_pos_u2b_cached(sv, &mg, start, send, uoffset2,
7494 uoffset, boffset) - boffset;
7508 =for apidoc sv_pos_u2b
7510 Converts the value pointed to by C<offsetp> from a count of UTF-8 chars from
7511 the start of the string, to a count of the equivalent number of bytes; if
7512 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7513 the offset, rather than from the start of the string. Handles magic and
7516 Use C<sv_pos_u2b_flags> in preference, which correctly handles strings longer
7523 * sv_pos_u2b() uses, like sv_pos_b2u(), the mg_ptr of the potential
7524 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7525 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7529 /* This function is subject to size and sign problems */
7532 Perl_sv_pos_u2b(pTHX_ SV *const sv, I32 *const offsetp, I32 *const lenp)
7534 PERL_ARGS_ASSERT_SV_POS_U2B;
7537 STRLEN ulen = (STRLEN)*lenp;
7538 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, &ulen,
7539 SV_GMAGIC|SV_CONST_RETURN);
7542 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, NULL,
7543 SV_GMAGIC|SV_CONST_RETURN);
7548 S_utf8_mg_len_cache_update(pTHX_ SV *const sv, MAGIC **const mgp,
7551 PERL_ARGS_ASSERT_UTF8_MG_LEN_CACHE_UPDATE;
7552 if (SvREADONLY(sv) || SvGMAGICAL(sv) || !SvPOK(sv))
7555 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7556 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7557 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, &PL_vtbl_utf8, 0, 0);
7561 (*mgp)->mg_len = ulen;
7564 /* Create and update the UTF8 magic offset cache, with the proffered utf8/
7565 byte length pairing. The (byte) length of the total SV is passed in too,
7566 as blen, because for some (more esoteric) SVs, the call to SvPV_const()
7567 may not have updated SvCUR, so we can't rely on reading it directly.
7569 The proffered utf8/byte length pairing isn't used if the cache already has
7570 two pairs, and swapping either for the proffered pair would increase the
7571 RMS of the intervals between known byte offsets.
7573 The cache itself consists of 4 STRLEN values
7574 0: larger UTF-8 offset
7575 1: corresponding byte offset
7576 2: smaller UTF-8 offset
7577 3: corresponding byte offset
7579 Unused cache pairs have the value 0, 0.
7580 Keeping the cache "backwards" means that the invariant of
7581 cache[0] >= cache[2] is maintained even with empty slots, which means that
7582 the code that uses it doesn't need to worry if only 1 entry has actually
7583 been set to non-zero. It also makes the "position beyond the end of the
7584 cache" logic much simpler, as the first slot is always the one to start
7588 S_utf8_mg_pos_cache_update(pTHX_ SV *const sv, MAGIC **const mgp, const STRLEN byte,
7589 const STRLEN utf8, const STRLEN blen)
7593 PERL_ARGS_ASSERT_UTF8_MG_POS_CACHE_UPDATE;
7598 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7599 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7600 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, (MGVTBL*)&PL_vtbl_utf8, 0,
7602 (*mgp)->mg_len = -1;
7606 if (!(cache = (STRLEN *)(*mgp)->mg_ptr)) {
7607 Newxz(cache, PERL_MAGIC_UTF8_CACHESIZE * 2, STRLEN);
7608 (*mgp)->mg_ptr = (char *) cache;
7612 if (PL_utf8cache < 0 && SvPOKp(sv)) {
7613 /* SvPOKp() because, if sv is a reference, then SvPVX() is actually
7614 a pointer. Note that we no longer cache utf8 offsets on refer-
7615 ences, but this check is still a good idea, for robustness. */
7616 const U8 *start = (const U8 *) SvPVX_const(sv);
7617 const STRLEN realutf8 = utf8_length(start, start + byte);
7619 assert_uft8_cache_coherent("utf8_mg_pos_cache_update", utf8, realutf8,
7623 /* Cache is held with the later position first, to simplify the code
7624 that deals with unbounded ends. */
7626 ASSERT_UTF8_CACHE(cache);
7627 if (cache[1] == 0) {
7628 /* Cache is totally empty */
7631 } else if (cache[3] == 0) {
7632 if (byte > cache[1]) {
7633 /* New one is larger, so goes first. */
7634 cache[2] = cache[0];
7635 cache[3] = cache[1];
7643 /* float casts necessary? XXX */
7644 #define THREEWAY_SQUARE(a,b,c,d) \
7645 ((float)((d) - (c))) * ((float)((d) - (c))) \
7646 + ((float)((c) - (b))) * ((float)((c) - (b))) \
7647 + ((float)((b) - (a))) * ((float)((b) - (a)))
7649 /* Cache has 2 slots in use, and we know three potential pairs.
7650 Keep the two that give the lowest RMS distance. Do the
7651 calculation in bytes simply because we always know the byte
7652 length. squareroot has the same ordering as the positive value,
7653 so don't bother with the actual square root. */
7654 if (byte > cache[1]) {
7655 /* New position is after the existing pair of pairs. */
7656 const float keep_earlier
7657 = THREEWAY_SQUARE(0, cache[3], byte, blen);
7658 const float keep_later
7659 = THREEWAY_SQUARE(0, cache[1], byte, blen);
7661 if (keep_later < keep_earlier) {
7662 cache[2] = cache[0];
7663 cache[3] = cache[1];
7669 const float keep_later = THREEWAY_SQUARE(0, byte, cache[1], blen);
7670 float b, c, keep_earlier;
7671 if (byte > cache[3]) {
7672 /* New position is between the existing pair of pairs. */
7673 b = (float)cache[3];
7676 /* New position is before the existing pair of pairs. */
7678 c = (float)cache[3];
7680 keep_earlier = THREEWAY_SQUARE(0, b, c, blen);
7681 if (byte > cache[3]) {
7682 if (keep_later < keep_earlier) {
7692 if (! (keep_later < keep_earlier)) {
7693 cache[0] = cache[2];
7694 cache[1] = cache[3];
7701 ASSERT_UTF8_CACHE(cache);
7704 /* We already know all of the way, now we may be able to walk back. The same
7705 assumption is made as in S_sv_pos_u2b_midway(), namely that walking
7706 backward is half the speed of walking forward. */
7708 S_sv_pos_b2u_midway(pTHX_ const U8 *const s, const U8 *const target,
7709 const U8 *end, STRLEN endu)
7711 const STRLEN forw = target - s;
7712 STRLEN backw = end - target;
7714 PERL_ARGS_ASSERT_SV_POS_B2U_MIDWAY;
7716 if (forw < 2 * backw) {
7717 return utf8_length(s, target);
7720 while (end > target) {
7722 while (UTF8_IS_CONTINUATION(*end)) {
7731 =for apidoc sv_pos_b2u_flags
7733 Converts C<offset> from a count of bytes from the start of the string, to
7734 a count of the equivalent number of UTF-8 chars. Handles type coercion.
7735 C<flags> is passed to C<SvPV_flags>, and usually should be
7736 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7742 * sv_pos_b2u_flags() uses, like sv_pos_u2b_flags(), the mg_ptr of the
7743 * potential PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8
7748 Perl_sv_pos_b2u_flags(pTHX_ SV *const sv, STRLEN const offset, U32 flags)
7751 STRLEN len = 0; /* Actually always set, but let's keep gcc happy. */
7757 PERL_ARGS_ASSERT_SV_POS_B2U_FLAGS;
7759 s = (const U8*)SvPV_flags(sv, blen, flags);
7762 Perl_croak(aTHX_ "panic: sv_pos_b2u: bad byte offset, blen=%" UVuf
7763 ", byte=%" UVuf, (UV)blen, (UV)offset);
7769 && SvTYPE(sv) >= SVt_PVMG
7770 && (mg = mg_find(sv, PERL_MAGIC_utf8)))
7773 STRLEN * const cache = (STRLEN *) mg->mg_ptr;
7774 if (cache[1] == offset) {
7775 /* An exact match. */
7778 if (cache[3] == offset) {
7779 /* An exact match. */
7783 if (cache[1] < offset) {
7784 /* We already know part of the way. */
7785 if (mg->mg_len != -1) {
7786 /* Actually, we know the end too. */
7788 + S_sv_pos_b2u_midway(aTHX_ s + cache[1], send,
7789 s + blen, mg->mg_len - cache[0]);
7791 len = cache[0] + utf8_length(s + cache[1], send);
7794 else if (cache[3] < offset) {
7795 /* We're between the two cached pairs, so we do the calculation
7796 offset by the byte/utf-8 positions for the earlier pair,
7797 then add the utf-8 characters from the string start to
7799 len = S_sv_pos_b2u_midway(aTHX_ s + cache[3], send,
7800 s + cache[1], cache[0] - cache[2])
7804 else { /* cache[3] > offset */
7805 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + cache[3],
7809 ASSERT_UTF8_CACHE(cache);
7811 } else if (mg->mg_len != -1) {
7812 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + blen, mg->mg_len);
7816 if (!found || PL_utf8cache < 0) {
7817 const STRLEN real_len = utf8_length(s, send);
7819 if (found && PL_utf8cache < 0)
7820 assert_uft8_cache_coherent("sv_pos_b2u", len, real_len, sv);
7826 utf8_mg_len_cache_update(sv, &mg, len);
7828 utf8_mg_pos_cache_update(sv, &mg, offset, len, blen);
7835 =for apidoc sv_pos_b2u
7837 Converts the value pointed to by C<offsetp> from a count of bytes from the
7838 start of the string, to a count of the equivalent number of UTF-8 chars.
7839 Handles magic and type coercion.
7841 Use C<sv_pos_b2u_flags> in preference, which correctly handles strings
7848 * sv_pos_b2u() uses, like sv_pos_u2b(), the mg_ptr of the potential
7849 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7854 Perl_sv_pos_b2u(pTHX_ SV *const sv, I32 *const offsetp)
7856 PERL_ARGS_ASSERT_SV_POS_B2U;
7861 *offsetp = (I32)sv_pos_b2u_flags(sv, (STRLEN)*offsetp,
7862 SV_GMAGIC|SV_CONST_RETURN);
7866 S_assert_uft8_cache_coherent(pTHX_ const char *const func, STRLEN from_cache,
7867 STRLEN real, SV *const sv)
7869 PERL_ARGS_ASSERT_ASSERT_UFT8_CACHE_COHERENT;
7871 /* As this is debugging only code, save space by keeping this test here,
7872 rather than inlining it in all the callers. */
7873 if (from_cache == real)
7876 /* Need to turn the assertions off otherwise we may recurse infinitely
7877 while printing error messages. */
7878 SAVEI8(PL_utf8cache);
7880 Perl_croak(aTHX_ "panic: %s cache %" UVuf " real %" UVuf " for %" SVf,
7881 func, (UV) from_cache, (UV) real, SVfARG(sv));
7887 Returns a boolean indicating whether the strings in the two SVs are
7888 identical. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7889 coerce its args to strings if necessary.
7891 =for apidoc sv_eq_flags
7893 Returns a boolean indicating whether the strings in the two SVs are
7894 identical. Is UTF-8 and S<C<'use bytes'>> aware and coerces its args to strings
7895 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get-magic, too.
7901 Perl_sv_eq_flags(pTHX_ SV *sv1, SV *sv2, const U32 flags)
7908 SV* svrecode = NULL;
7915 /* if pv1 and pv2 are the same, second SvPV_const call may
7916 * invalidate pv1 (if we are handling magic), so we may need to
7918 if (sv1 == sv2 && flags & SV_GMAGIC
7919 && (SvTHINKFIRST(sv1) || SvGMAGICAL(sv1))) {
7920 pv1 = SvPV_const(sv1, cur1);
7921 sv1 = newSVpvn_flags(pv1, cur1, SVs_TEMP | SvUTF8(sv2));
7923 pv1 = SvPV_flags_const(sv1, cur1, flags);
7931 pv2 = SvPV_flags_const(sv2, cur2, flags);
7933 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7934 /* Differing utf8ness. */
7936 /* sv1 is the UTF-8 one */
7937 return bytes_cmp_utf8((const U8*)pv2, cur2,
7938 (const U8*)pv1, cur1) == 0;
7941 /* sv2 is the UTF-8 one */
7942 return bytes_cmp_utf8((const U8*)pv1, cur1,
7943 (const U8*)pv2, cur2) == 0;
7948 eq = (pv1 == pv2) || memEQ(pv1, pv2, cur1);
7950 SvREFCNT_dec(svrecode);
7958 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7959 string in C<sv1> is less than, equal to, or greater than the string in
7960 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7961 coerce its args to strings if necessary. See also C<L</sv_cmp_locale>>.
7963 =for apidoc sv_cmp_flags
7965 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7966 string in C<sv1> is less than, equal to, or greater than the string in
7967 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware and will coerce its args to strings
7968 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get magic. See
7969 also C<L</sv_cmp_locale_flags>>.
7975 Perl_sv_cmp(pTHX_ SV *const sv1, SV *const sv2)
7977 return sv_cmp_flags(sv1, sv2, SV_GMAGIC);
7981 Perl_sv_cmp_flags(pTHX_ SV *const sv1, SV *const sv2,
7985 const char *pv1, *pv2;
7987 SV *svrecode = NULL;
7994 pv1 = SvPV_flags_const(sv1, cur1, flags);
8001 pv2 = SvPV_flags_const(sv2, cur2, flags);
8003 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
8004 /* Differing utf8ness. */
8006 const int retval = -bytes_cmp_utf8((const U8*)pv2, cur2,
8007 (const U8*)pv1, cur1);
8008 return retval ? retval < 0 ? -1 : +1 : 0;
8011 const int retval = bytes_cmp_utf8((const U8*)pv1, cur1,
8012 (const U8*)pv2, cur2);
8013 return retval ? retval < 0 ? -1 : +1 : 0;
8017 /* Here, if both are non-NULL, then they have the same UTF8ness. */
8020 cmp = cur2 ? -1 : 0;
8024 STRLEN shortest_len = cur1 < cur2 ? cur1 : cur2;
8027 if (! DO_UTF8(sv1)) {
8029 const I32 retval = memcmp((const void*)pv1,
8033 cmp = retval < 0 ? -1 : 1;
8034 } else if (cur1 == cur2) {
8037 cmp = cur1 < cur2 ? -1 : 1;
8041 else { /* Both are to be treated as UTF-EBCDIC */
8043 /* EBCDIC UTF-8 is complicated by the fact that it is based on I8
8044 * which remaps code points 0-255. We therefore generally have to
8045 * unmap back to the original values to get an accurate comparison.
8046 * But we don't have to do that for UTF-8 invariants, as by
8047 * definition, they aren't remapped, nor do we have to do it for
8048 * above-latin1 code points, as they also aren't remapped. (This
8049 * code also works on ASCII platforms, but the memcmp() above is
8052 const char *e = pv1 + shortest_len;
8054 /* Find the first bytes that differ between the two strings */
8055 while (pv1 < e && *pv1 == *pv2) {
8061 if (pv1 == e) { /* Are the same all the way to the end */
8065 cmp = cur1 < cur2 ? -1 : 1;
8068 else /* Here *pv1 and *pv2 are not equal, but all bytes earlier
8069 * in the strings were. The current bytes may or may not be
8070 * at the beginning of a character. But neither or both are
8071 * (or else earlier bytes would have been different). And
8072 * if we are in the middle of a character, the two
8073 * characters are comprised of the same number of bytes
8074 * (because in this case the start bytes are the same, and
8075 * the start bytes encode the character's length). */
8076 if (UTF8_IS_INVARIANT(*pv1))
8078 /* If both are invariants; can just compare directly */
8079 if (UTF8_IS_INVARIANT(*pv2)) {
8080 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8082 else /* Since *pv1 is invariant, it is the whole character,
8083 which means it is at the beginning of a character.
8084 That means pv2 is also at the beginning of a
8085 character (see earlier comment). Since it isn't
8086 invariant, it must be a start byte. If it starts a
8087 character whose code point is above 255, that
8088 character is greater than any single-byte char, which
8090 if (UTF8_IS_ABOVE_LATIN1_START(*pv2))
8095 /* Here, pv2 points to a character composed of 2 bytes
8096 * whose code point is < 256. Get its code point and
8097 * compare with *pv1 */
8098 cmp = ((U8) *pv1 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
8103 else /* The code point starting at pv1 isn't a single byte */
8104 if (UTF8_IS_INVARIANT(*pv2))
8106 /* But here, the code point starting at *pv2 is a single byte,
8107 * and so *pv1 must begin a character, hence is a start byte.
8108 * If that character is above 255, it is larger than any
8109 * single-byte char, which *pv2 is */
8110 if (UTF8_IS_ABOVE_LATIN1_START(*pv1)) {
8114 /* Here, pv1 points to a character composed of 2 bytes
8115 * whose code point is < 256. Get its code point and
8116 * compare with the single byte character *pv2 */
8117 cmp = (EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1)) < (U8) *pv2)
8122 else /* Here, we've ruled out either *pv1 and *pv2 being
8123 invariant. That means both are part of variants, but not
8124 necessarily at the start of a character */
8125 if ( UTF8_IS_ABOVE_LATIN1_START(*pv1)
8126 || UTF8_IS_ABOVE_LATIN1_START(*pv2))
8128 /* Here, at least one is the start of a character, which means
8129 * the other is also a start byte. And the code point of at
8130 * least one of the characters is above 255. It is a
8131 * characteristic of UTF-EBCDIC that all start bytes for
8132 * above-latin1 code points are well behaved as far as code
8133 * point comparisons go, and all are larger than all other
8134 * start bytes, so the comparison with those is also well
8136 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8139 /* Here both *pv1 and *pv2 are part of variant characters.
8140 * They could be both continuations, or both start characters.
8141 * (One or both could even be an illegal start character (for
8142 * an overlong) which for the purposes of sorting we treat as
8144 if (UTF8_IS_CONTINUATION(*pv1)) {
8146 /* If they are continuations for code points above 255,
8147 * then comparing the current byte is sufficient, as there
8148 * is no remapping of these and so the comparison is
8149 * well-behaved. We determine if they are such
8150 * continuations by looking at the preceding byte. It
8151 * could be a start byte, from which we can tell if it is
8152 * for an above 255 code point. Or it could be a
8153 * continuation, which means the character occupies at
8154 * least 3 bytes, so must be above 255. */
8155 if ( UTF8_IS_CONTINUATION(*(pv2 - 1))
8156 || UTF8_IS_ABOVE_LATIN1_START(*(pv2 -1)))
8158 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8162 /* Here, the continuations are for code points below 256;
8163 * back up one to get to the start byte */
8168 /* We need to get the actual native code point of each of these
8169 * variants in order to compare them */
8170 cmp = ( EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1))
8171 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
8180 SvREFCNT_dec(svrecode);
8186 =for apidoc sv_cmp_locale
8188 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
8189 S<C<'use bytes'>> aware, handles get magic, and will coerce its args to strings
8190 if necessary. See also C<L</sv_cmp>>.
8192 =for apidoc sv_cmp_locale_flags
8194 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
8195 S<C<'use bytes'>> aware and will coerce its args to strings if necessary. If
8196 the flags contain C<SV_GMAGIC>, it handles get magic. See also
8197 C<L</sv_cmp_flags>>.
8203 Perl_sv_cmp_locale(pTHX_ SV *const sv1, SV *const sv2)
8205 return sv_cmp_locale_flags(sv1, sv2, SV_GMAGIC);
8209 Perl_sv_cmp_locale_flags(pTHX_ SV *const sv1, SV *const sv2,
8212 #ifdef USE_LOCALE_COLLATE
8218 if (PL_collation_standard)
8223 /* Revert to using raw compare if both operands exist, but either one
8224 * doesn't transform properly for collation */
8226 pv1 = sv_collxfrm_flags(sv1, &len1, flags);
8230 pv2 = sv_collxfrm_flags(sv2, &len2, flags);
8236 pv1 = sv1 ? sv_collxfrm_flags(sv1, &len1, flags) : (char *) NULL;
8237 pv2 = sv2 ? sv_collxfrm_flags(sv2, &len2, flags) : (char *) NULL;
8240 if (!pv1 || !len1) {
8251 retval = memcmp((void*)pv1, (void*)pv2, len1 < len2 ? len1 : len2);
8254 return retval < 0 ? -1 : 1;
8257 * When the result of collation is equality, that doesn't mean
8258 * that there are no differences -- some locales exclude some
8259 * characters from consideration. So to avoid false equalities,
8260 * we use the raw string as a tiebreaker.
8267 PERL_UNUSED_ARG(flags);
8268 #endif /* USE_LOCALE_COLLATE */
8270 return sv_cmp(sv1, sv2);
8274 #ifdef USE_LOCALE_COLLATE
8277 =for apidoc sv_collxfrm
8279 This calls C<sv_collxfrm_flags> with the SV_GMAGIC flag. See
8280 C<L</sv_collxfrm_flags>>.
8282 =for apidoc sv_collxfrm_flags
8284 Add Collate Transform magic to an SV if it doesn't already have it. If the
8285 flags contain C<SV_GMAGIC>, it handles get-magic.
8287 Any scalar variable may carry C<PERL_MAGIC_collxfrm> magic that contains the
8288 scalar data of the variable, but transformed to such a format that a normal
8289 memory comparison can be used to compare the data according to the locale
8296 Perl_sv_collxfrm_flags(pTHX_ SV *const sv, STRLEN *const nxp, const I32 flags)
8300 PERL_ARGS_ASSERT_SV_COLLXFRM_FLAGS;
8302 mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_collxfrm) : (MAGIC *) NULL;
8304 /* If we don't have collation magic on 'sv', or the locale has changed
8305 * since the last time we calculated it, get it and save it now */
8306 if (!mg || !mg->mg_ptr || *(U32*)mg->mg_ptr != PL_collation_ix) {
8311 /* Free the old space */
8313 Safefree(mg->mg_ptr);
8315 s = SvPV_flags_const(sv, len, flags);
8316 if ((xf = _mem_collxfrm(s, len, &xlen, cBOOL(SvUTF8(sv))))) {
8318 mg = sv_magicext(sv, 0, PERL_MAGIC_collxfrm, &PL_vtbl_collxfrm,
8333 if (mg && mg->mg_ptr) {
8335 return mg->mg_ptr + sizeof(PL_collation_ix);
8343 #endif /* USE_LOCALE_COLLATE */
8346 S_sv_gets_append_to_utf8(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8348 SV * const tsv = newSV(0);
8351 sv_gets(tsv, fp, 0);
8352 sv_utf8_upgrade_nomg(tsv);
8353 SvCUR_set(sv,append);
8356 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8360 S_sv_gets_read_record(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8363 const STRLEN recsize = SvUV(SvRV(PL_rs)); /* RsRECORD() guarantees > 0. */
8364 /* Grab the size of the record we're getting */
8365 char *buffer = SvGROW(sv, (STRLEN)(recsize + append + 1)) + append;
8372 /* With a true, record-oriented file on VMS, we need to use read directly
8373 * to ensure that we respect RMS record boundaries. The user is responsible
8374 * for providing a PL_rs value that corresponds to the FAB$W_MRS (maximum
8375 * record size) field. N.B. This is likely to produce invalid results on
8376 * varying-width character data when a record ends mid-character.
8378 fd = PerlIO_fileno(fp);
8380 && PerlLIO_fstat(fd, &st) == 0
8381 && (st.st_fab_rfm == FAB$C_VAR
8382 || st.st_fab_rfm == FAB$C_VFC
8383 || st.st_fab_rfm == FAB$C_FIX)) {
8385 bytesread = PerlLIO_read(fd, buffer, recsize);
8387 else /* in-memory file from PerlIO::Scalar
8388 * or not a record-oriented file
8392 bytesread = PerlIO_read(fp, buffer, recsize);
8394 /* At this point, the logic in sv_get() means that sv will
8395 be treated as utf-8 if the handle is utf8.
8397 if (PerlIO_isutf8(fp) && bytesread > 0) {
8398 char *bend = buffer + bytesread;
8399 char *bufp = buffer;
8400 size_t charcount = 0;
8401 bool charstart = TRUE;
8404 while (charcount < recsize) {
8405 /* count accumulated characters */
8406 while (bufp < bend) {
8408 skip = UTF8SKIP(bufp);
8410 if (bufp + skip > bend) {
8411 /* partial at the end */
8422 if (charcount < recsize) {
8424 STRLEN bufp_offset = bufp - buffer;
8425 SSize_t morebytesread;
8427 /* originally I read enough to fill any incomplete
8428 character and the first byte of the next
8429 character if needed, but if there's many
8430 multi-byte encoded characters we're going to be
8431 making a read call for every character beyond
8432 the original read size.
8434 So instead, read the rest of the character if
8435 any, and enough bytes to match at least the
8436 start bytes for each character we're going to
8440 readsize = recsize - charcount;
8442 readsize = skip - (bend - bufp) + recsize - charcount - 1;
8443 buffer = SvGROW(sv, append + bytesread + readsize + 1) + append;
8444 bend = buffer + bytesread;
8445 morebytesread = PerlIO_read(fp, bend, readsize);
8446 if (morebytesread <= 0) {
8447 /* we're done, if we still have incomplete
8448 characters the check code in sv_gets() will
8451 I'd originally considered doing
8452 PerlIO_ungetc() on all but the lead
8453 character of the incomplete character, but
8454 read() doesn't do that, so I don't.
8459 /* prepare to scan some more */
8460 bytesread += morebytesread;
8461 bend = buffer + bytesread;
8462 bufp = buffer + bufp_offset;
8470 SvCUR_set(sv, bytesread + append);
8471 buffer[bytesread] = '\0';
8472 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8478 Get a line from the filehandle and store it into the SV, optionally
8479 appending to the currently-stored string. If C<append> is not 0, the
8480 line is appended to the SV instead of overwriting it. C<append> should
8481 be set to the byte offset that the appended string should start at
8482 in the SV (typically, C<SvCUR(sv)> is a suitable choice).
8488 Perl_sv_gets(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8498 PERL_ARGS_ASSERT_SV_GETS;
8500 if (SvTHINKFIRST(sv))
8501 sv_force_normal_flags(sv, append ? 0 : SV_COW_DROP_PV);
8502 /* XXX. If you make this PVIV, then copy on write can copy scalars read
8504 However, perlbench says it's slower, because the existing swipe code
8505 is faster than copy on write.
8506 Swings and roundabouts. */
8507 SvUPGRADE(sv, SVt_PV);
8510 /* line is going to be appended to the existing buffer in the sv */
8511 if (PerlIO_isutf8(fp)) {
8513 sv_utf8_upgrade_nomg(sv);
8514 sv_pos_u2b(sv,&append,0);
8516 } else if (SvUTF8(sv)) {
8517 return S_sv_gets_append_to_utf8(aTHX_ sv, fp, append);
8523 /* not appending - "clear" the string by setting SvCUR to 0,
8524 * the pv is still avaiable. */
8527 if (PerlIO_isutf8(fp))
8530 if (IN_PERL_COMPILETIME) {
8531 /* we always read code in line mode */
8535 else if (RsSNARF(PL_rs)) {
8536 /* If it is a regular disk file use size from stat() as estimate
8537 of amount we are going to read -- may result in mallocing
8538 more memory than we really need if the layers below reduce
8539 the size we read (e.g. CRLF or a gzip layer).
8542 int fd = PerlIO_fileno(fp);
8543 if (fd >= 0 && (PerlLIO_fstat(fd, &st) == 0) && S_ISREG(st.st_mode)) {
8544 const Off_t offset = PerlIO_tell(fp);
8545 if (offset != (Off_t) -1 && st.st_size + append > offset) {
8546 #ifdef PERL_COPY_ON_WRITE
8547 /* Add an extra byte for the sake of copy-on-write's
8548 * buffer reference count. */
8549 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 2));
8551 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 1));
8558 else if (RsRECORD(PL_rs)) {
8559 return S_sv_gets_read_record(aTHX_ sv, fp, append);
8561 else if (RsPARA(PL_rs)) {
8567 /* Get $/ i.e. PL_rs into same encoding as stream wants */
8568 if (PerlIO_isutf8(fp)) {
8569 rsptr = SvPVutf8(PL_rs, rslen);
8572 if (SvUTF8(PL_rs)) {
8573 if (!sv_utf8_downgrade(PL_rs, TRUE)) {
8574 Perl_croak(aTHX_ "Wide character in $/");
8577 /* extract the raw pointer to the record separator */
8578 rsptr = SvPV_const(PL_rs, rslen);
8582 /* rslast is the last character in the record separator
8583 * note we don't use rslast except when rslen is true, so the
8584 * null assign is a placeholder. */
8585 rslast = rslen ? rsptr[rslen - 1] : '\0';
8587 if (rspara) { /* have to do this both before and after */
8588 do { /* to make sure file boundaries work right */
8591 i = PerlIO_getc(fp);
8595 PerlIO_ungetc(fp,i);
8601 /* See if we know enough about I/O mechanism to cheat it ! */
8603 /* This used to be #ifdef test - it is made run-time test for ease
8604 of abstracting out stdio interface. One call should be cheap
8605 enough here - and may even be a macro allowing compile
8609 if (PerlIO_fast_gets(fp)) {
8611 * We can do buffer based IO operations on this filehandle.
8613 * This means we can bypass a lot of subcalls and process
8614 * the buffer directly, it also means we know the upper bound
8615 * on the amount of data we might read of the current buffer
8616 * into our sv. Knowing this allows us to preallocate the pv
8617 * to be able to hold that maximum, which allows us to simplify
8618 * a lot of logic. */
8621 * We're going to steal some values from the stdio struct
8622 * and put EVERYTHING in the innermost loop into registers.
8624 STDCHAR *ptr; /* pointer into fp's read-ahead buffer */
8625 STRLEN bpx; /* length of the data in the target sv
8626 used to fix pointers after a SvGROW */
8627 I32 shortbuffered; /* If the pv buffer is shorter than the amount
8628 of data left in the read-ahead buffer.
8629 If 0 then the pv buffer can hold the full
8630 amount left, otherwise this is the amount it
8633 /* Here is some breathtakingly efficient cheating */
8635 /* When you read the following logic resist the urge to think
8636 * of record separators that are 1 byte long. They are an
8637 * uninteresting special (simple) case.
8639 * Instead think of record separators which are at least 2 bytes
8640 * long, and keep in mind that we need to deal with such
8641 * separators when they cross a read-ahead buffer boundary.
8643 * Also consider that we need to gracefully deal with separators
8644 * that may be longer than a single read ahead buffer.
8646 * Lastly do not forget we want to copy the delimiter as well. We
8647 * are copying all data in the file _up_to_and_including_ the separator
8650 * Now that you have all that in mind here is what is happening below:
8652 * 1. When we first enter the loop we do some memory book keeping to see
8653 * how much free space there is in the target SV. (This sub assumes that
8654 * it is operating on the same SV most of the time via $_ and that it is
8655 * going to be able to reuse the same pv buffer each call.) If there is
8656 * "enough" room then we set "shortbuffered" to how much space there is
8657 * and start reading forward.
8659 * 2. When we scan forward we copy from the read-ahead buffer to the target
8660 * SV's pv buffer. While we go we watch for the end of the read-ahead buffer,
8661 * and the end of the of pv, as well as for the "rslast", which is the last
8662 * char of the separator.
8664 * 3. When scanning forward if we see rslast then we jump backwards in *pv*
8665 * (which has a "complete" record up to the point we saw rslast) and check
8666 * it to see if it matches the separator. If it does we are done. If it doesn't
8667 * we continue on with the scan/copy.
8669 * 4. If we run out of read-ahead buffer (cnt goes to 0) then we have to get
8670 * the IO system to read the next buffer. We do this by doing a getc(), which
8671 * returns a single char read (or EOF), and prefills the buffer, and also
8672 * allows us to find out how full the buffer is. We use this information to
8673 * SvGROW() the sv to the size remaining in the buffer, after which we copy
8674 * the returned single char into the target sv, and then go back into scan
8677 * 5. If we run out of write-buffer then we SvGROW() it by the size of the
8678 * remaining space in the read-buffer.
8680 * Note that this code despite its twisty-turny nature is pretty darn slick.
8681 * It manages single byte separators, multi-byte cross boundary separators,
8682 * and cross-read-buffer separators cleanly and efficiently at the cost
8683 * of potentially greatly overallocating the target SV.
8689 /* get the number of bytes remaining in the read-ahead buffer
8690 * on first call on a given fp this will return 0.*/
8691 cnt = PerlIO_get_cnt(fp);
8693 /* make sure we have the room */
8694 if ((I32)(SvLEN(sv) - append) <= cnt + 1) {
8695 /* Not room for all of it
8696 if we are looking for a separator and room for some
8698 if (rslen && cnt > 80 && (I32)SvLEN(sv) > append) {
8699 /* just process what we have room for */
8700 shortbuffered = cnt - SvLEN(sv) + append + 1;
8701 cnt -= shortbuffered;
8704 /* ensure that the target sv has enough room to hold
8705 * the rest of the read-ahead buffer */
8707 /* remember that cnt can be negative */
8708 SvGROW(sv, (STRLEN)(append + (cnt <= 0 ? 2 : (cnt + 1))));
8712 /* we have enough room to hold the full buffer, lets scream */
8716 /* extract the pointer to sv's string buffer, offset by append as necessary */
8717 bp = (STDCHAR*)SvPVX_const(sv) + append; /* move these two too to registers */
8718 /* extract the point to the read-ahead buffer */
8719 ptr = (STDCHAR*)PerlIO_get_ptr(fp);
8721 /* some trace debug output */
8722 DEBUG_P(PerlIO_printf(Perl_debug_log,
8723 "Screamer: entering, ptr=%" UVuf ", cnt=%ld\n",PTR2UV(ptr),(long)cnt));
8724 DEBUG_P(PerlIO_printf(Perl_debug_log,
8725 "Screamer: entering: PerlIO * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%"
8727 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8728 PTR2UV(PerlIO_has_base(fp) ? PerlIO_get_base(fp) : 0)));
8732 /* if there is stuff left in the read-ahead buffer */
8734 /* if there is a separator */
8736 /* find next rslast */
8739 /* shortcut common case of blank line */
8741 if ((*bp++ = *ptr++) == rslast)
8742 goto thats_all_folks;
8744 p = (STDCHAR *)memchr(ptr, rslast, cnt);
8746 SSize_t got = p - ptr + 1;
8747 Copy(ptr, bp, got, STDCHAR);
8751 goto thats_all_folks;
8753 Copy(ptr, bp, cnt, STDCHAR);
8759 /* no separator, slurp the full buffer */
8760 Copy(ptr, bp, cnt, char); /* this | eat */
8761 bp += cnt; /* screams | dust */
8762 ptr += cnt; /* louder | sed :-) */
8764 assert (!shortbuffered);
8765 goto cannot_be_shortbuffered;
8769 if (shortbuffered) { /* oh well, must extend */
8770 /* we didnt have enough room to fit the line into the target buffer
8771 * so we must extend the target buffer and keep going */
8772 cnt = shortbuffered;
8774 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8776 /* extned the target sv's buffer so it can hold the full read-ahead buffer */
8777 SvGROW(sv, SvLEN(sv) + append + cnt + 2);
8778 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8782 cannot_be_shortbuffered:
8783 /* we need to refill the read-ahead buffer if possible */
8785 DEBUG_P(PerlIO_printf(Perl_debug_log,
8786 "Screamer: going to getc, ptr=%" UVuf ", cnt=%" IVdf "\n",
8787 PTR2UV(ptr),(IV)cnt));
8788 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* deregisterize cnt and ptr */
8790 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8791 "Screamer: pre: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf "\n",
8792 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8793 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8796 call PerlIO_getc() to let it prefill the lookahead buffer
8798 This used to call 'filbuf' in stdio form, but as that behaves like
8799 getc when cnt <= 0 we use PerlIO_getc here to avoid introducing
8800 another abstraction.
8802 Note we have to deal with the char in 'i' if we are not at EOF
8804 i = PerlIO_getc(fp); /* get more characters */
8806 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8807 "Screamer: post: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf "\n",
8808 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8809 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8811 /* find out how much is left in the read-ahead buffer, and rextract its pointer */
8812 cnt = PerlIO_get_cnt(fp);
8813 ptr = (STDCHAR*)PerlIO_get_ptr(fp); /* reregisterize cnt and ptr */
8814 DEBUG_P(PerlIO_printf(Perl_debug_log,
8815 "Screamer: after getc, ptr=%" UVuf ", cnt=%" IVdf "\n",
8816 PTR2UV(ptr),(IV)cnt));
8818 if (i == EOF) /* all done for ever? */
8819 goto thats_really_all_folks;
8821 /* make sure we have enough space in the target sv */
8822 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8824 SvGROW(sv, bpx + cnt + 2);
8825 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8827 /* copy of the char we got from getc() */
8828 *bp++ = (STDCHAR)i; /* store character from PerlIO_getc */
8830 /* make sure we deal with the i being the last character of a separator */
8831 if (rslen && (STDCHAR)i == rslast) /* all done for now? */
8832 goto thats_all_folks;
8836 /* check if we have actually found the separator - only really applies
8838 if ((rslen > 1 && (STRLEN)(bp - (STDCHAR*)SvPVX_const(sv)) < rslen) ||
8839 memNE((char*)bp - rslen, rsptr, rslen))
8840 goto screamer; /* go back to the fray */
8841 thats_really_all_folks:
8843 cnt += shortbuffered;
8844 DEBUG_P(PerlIO_printf(Perl_debug_log,
8845 "Screamer: quitting, ptr=%" UVuf ", cnt=%" IVdf "\n",PTR2UV(ptr),(IV)cnt));
8846 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* put these back or we're in trouble */
8847 DEBUG_P(PerlIO_printf(Perl_debug_log,
8848 "Screamer: end: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf
8850 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8851 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8853 SvCUR_set(sv, bp - (STDCHAR*)SvPVX_const(sv)); /* set length */
8854 DEBUG_P(PerlIO_printf(Perl_debug_log,
8855 "Screamer: done, len=%ld, string=|%.*s|\n",
8856 (long)SvCUR(sv),(int)SvCUR(sv),SvPVX_const(sv)));
8860 /*The big, slow, and stupid way. */
8861 #ifdef USE_HEAP_INSTEAD_OF_STACK /* Even slower way. */
8862 STDCHAR *buf = NULL;
8863 Newx(buf, 8192, STDCHAR);
8871 const STDCHAR * const bpe = buf + sizeof(buf);
8873 while ((i = PerlIO_getc(fp)) != EOF && (*bp++ = (STDCHAR)i) != rslast && bp < bpe)
8874 ; /* keep reading */
8878 cnt = PerlIO_read(fp,(char*)buf, sizeof(buf));
8879 /* Accommodate broken VAXC compiler, which applies U8 cast to
8880 * both args of ?: operator, causing EOF to change into 255
8883 i = (U8)buf[cnt - 1];
8889 cnt = 0; /* we do need to re-set the sv even when cnt <= 0 */
8891 sv_catpvn_nomg(sv, (char *) buf, cnt);
8893 sv_setpvn(sv, (char *) buf, cnt); /* "nomg" is implied */
8895 if (i != EOF && /* joy */
8897 SvCUR(sv) < rslen ||
8898 memNE(SvPVX_const(sv) + SvCUR(sv) - rslen, rsptr, rslen)))
8902 * If we're reading from a TTY and we get a short read,
8903 * indicating that the user hit his EOF character, we need
8904 * to notice it now, because if we try to read from the TTY
8905 * again, the EOF condition will disappear.
8907 * The comparison of cnt to sizeof(buf) is an optimization
8908 * that prevents unnecessary calls to feof().
8912 if (!(cnt < (I32)sizeof(buf) && PerlIO_eof(fp)))
8916 #ifdef USE_HEAP_INSTEAD_OF_STACK
8921 if (rspara) { /* have to do this both before and after */
8922 while (i != EOF) { /* to make sure file boundaries work right */
8923 i = PerlIO_getc(fp);
8925 PerlIO_ungetc(fp,i);
8931 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8937 Auto-increment of the value in the SV, doing string to numeric conversion
8938 if necessary. Handles 'get' magic and operator overloading.
8944 Perl_sv_inc(pTHX_ SV *const sv)
8953 =for apidoc sv_inc_nomg
8955 Auto-increment of the value in the SV, doing string to numeric conversion
8956 if necessary. Handles operator overloading. Skips handling 'get' magic.
8962 Perl_sv_inc_nomg(pTHX_ SV *const sv)
8969 if (SvTHINKFIRST(sv)) {
8970 if (SvREADONLY(sv)) {
8971 Perl_croak_no_modify();
8975 if (SvAMAGIC(sv) && AMG_CALLunary(sv, inc_amg))
8977 i = PTR2IV(SvRV(sv));
8981 else sv_force_normal_flags(sv, 0);
8983 flags = SvFLAGS(sv);
8984 if ((flags & (SVp_NOK|SVp_IOK)) == SVp_NOK) {
8985 /* It's (privately or publicly) a float, but not tested as an
8986 integer, so test it to see. */
8988 flags = SvFLAGS(sv);
8990 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8991 /* It's publicly an integer, or privately an integer-not-float */
8992 #ifdef PERL_PRESERVE_IVUV
8996 if (SvUVX(sv) == UV_MAX)
8997 sv_setnv(sv, UV_MAX_P1);
8999 (void)SvIOK_only_UV(sv);
9000 SvUV_set(sv, SvUVX(sv) + 1);
9002 if (SvIVX(sv) == IV_MAX)
9003 sv_setuv(sv, (UV)IV_MAX + 1);
9005 (void)SvIOK_only(sv);
9006 SvIV_set(sv, SvIVX(sv) + 1);
9011 if (flags & SVp_NOK) {
9012 const NV was = SvNVX(sv);
9013 if (LIKELY(!Perl_isinfnan(was)) &&
9014 NV_OVERFLOWS_INTEGERS_AT != 0.0 &&
9015 was >= NV_OVERFLOWS_INTEGERS_AT) {
9016 /* diag_listed_as: Lost precision when %s %f by 1 */
9017 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
9018 "Lost precision when incrementing %" NVff " by 1",
9021 (void)SvNOK_only(sv);
9022 SvNV_set(sv, was + 1.0);
9026 /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */
9027 if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv)))
9028 Perl_croak_no_modify();
9030 if (!(flags & SVp_POK) || !*SvPVX_const(sv)) {
9031 if ((flags & SVTYPEMASK) < SVt_PVIV)
9032 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV ? SVt_PVIV : SVt_IV));
9033 (void)SvIOK_only(sv);
9038 while (isALPHA(*d)) d++;
9039 while (isDIGIT(*d)) d++;
9040 if (d < SvEND(sv)) {
9041 const int numtype = grok_number_flags(SvPVX_const(sv), SvCUR(sv), NULL, PERL_SCAN_TRAILING);
9042 #ifdef PERL_PRESERVE_IVUV
9043 /* Got to punt this as an integer if needs be, but we don't issue
9044 warnings. Probably ought to make the sv_iv_please() that does
9045 the conversion if possible, and silently. */
9046 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
9047 /* Need to try really hard to see if it's an integer.
9048 9.22337203685478e+18 is an integer.
9049 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
9050 so $a="9.22337203685478e+18"; $a+0; $a++
9051 needs to be the same as $a="9.22337203685478e+18"; $a++
9058 /* sv_2iv *should* have made this an NV */
9059 if (flags & SVp_NOK) {
9060 (void)SvNOK_only(sv);
9061 SvNV_set(sv, SvNVX(sv) + 1.0);
9064 /* I don't think we can get here. Maybe I should assert this
9065 And if we do get here I suspect that sv_setnv will croak. NWC
9067 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_inc punt failed to convert '%s' to IOK or NOKp, UV=0x%" UVxf " NV=%" NVgf "\n",
9068 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
9070 #endif /* PERL_PRESERVE_IVUV */
9071 if (!numtype && ckWARN(WARN_NUMERIC))
9072 not_incrementable(sv);
9073 sv_setnv(sv,Atof(SvPVX_const(sv)) + 1.0);
9077 while (d >= SvPVX_const(sv)) {
9085 /* MKS: The original code here died if letters weren't consecutive.
9086 * at least it didn't have to worry about non-C locales. The
9087 * new code assumes that ('z'-'a')==('Z'-'A'), letters are
9088 * arranged in order (although not consecutively) and that only
9089 * [A-Za-z] are accepted by isALPHA in the C locale.
9091 if (isALPHA_FOLD_NE(*d, 'z')) {
9092 do { ++*d; } while (!isALPHA(*d));
9095 *(d--) -= 'z' - 'a';
9100 *(d--) -= 'z' - 'a' + 1;
9104 /* oh,oh, the number grew */
9105 SvGROW(sv, SvCUR(sv) + 2);
9106 SvCUR_set(sv, SvCUR(sv) + 1);
9107 for (d = SvPVX(sv) + SvCUR(sv); d > SvPVX_const(sv); d--)
9118 Auto-decrement of the value in the SV, doing string to numeric conversion
9119 if necessary. Handles 'get' magic and operator overloading.
9125 Perl_sv_dec(pTHX_ SV *const sv)
9134 =for apidoc sv_dec_nomg
9136 Auto-decrement of the value in the SV, doing string to numeric conversion
9137 if necessary. Handles operator overloading. Skips handling 'get' magic.
9143 Perl_sv_dec_nomg(pTHX_ SV *const sv)
9149 if (SvTHINKFIRST(sv)) {
9150 if (SvREADONLY(sv)) {
9151 Perl_croak_no_modify();
9155 if (SvAMAGIC(sv) && AMG_CALLunary(sv, dec_amg))
9157 i = PTR2IV(SvRV(sv));
9161 else sv_force_normal_flags(sv, 0);
9163 /* Unlike sv_inc we don't have to worry about string-never-numbers
9164 and keeping them magic. But we mustn't warn on punting */
9165 flags = SvFLAGS(sv);
9166 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
9167 /* It's publicly an integer, or privately an integer-not-float */
9168 #ifdef PERL_PRESERVE_IVUV
9172 if (SvUVX(sv) == 0) {
9173 (void)SvIOK_only(sv);
9177 (void)SvIOK_only_UV(sv);
9178 SvUV_set(sv, SvUVX(sv) - 1);
9181 if (SvIVX(sv) == IV_MIN) {
9182 sv_setnv(sv, (NV)IV_MIN);
9186 (void)SvIOK_only(sv);
9187 SvIV_set(sv, SvIVX(sv) - 1);
9192 if (flags & SVp_NOK) {
9195 const NV was = SvNVX(sv);
9196 if (LIKELY(!Perl_isinfnan(was)) &&
9197 NV_OVERFLOWS_INTEGERS_AT != 0.0 &&
9198 was <= -NV_OVERFLOWS_INTEGERS_AT) {
9199 /* diag_listed_as: Lost precision when %s %f by 1 */
9200 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
9201 "Lost precision when decrementing %" NVff " by 1",
9204 (void)SvNOK_only(sv);
9205 SvNV_set(sv, was - 1.0);
9210 /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */
9211 if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv)))
9212 Perl_croak_no_modify();
9214 if (!(flags & SVp_POK)) {
9215 if ((flags & SVTYPEMASK) < SVt_PVIV)
9216 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV) ? SVt_PVIV : SVt_IV);
9218 (void)SvIOK_only(sv);
9221 #ifdef PERL_PRESERVE_IVUV
9223 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), NULL);
9224 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
9225 /* Need to try really hard to see if it's an integer.
9226 9.22337203685478e+18 is an integer.
9227 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
9228 so $a="9.22337203685478e+18"; $a+0; $a--
9229 needs to be the same as $a="9.22337203685478e+18"; $a--
9236 /* sv_2iv *should* have made this an NV */
9237 if (flags & SVp_NOK) {
9238 (void)SvNOK_only(sv);
9239 SvNV_set(sv, SvNVX(sv) - 1.0);
9242 /* I don't think we can get here. Maybe I should assert this
9243 And if we do get here I suspect that sv_setnv will croak. NWC
9245 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_dec punt failed to convert '%s' to IOK or NOKp, UV=0x%" UVxf " NV=%" NVgf "\n",
9246 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
9249 #endif /* PERL_PRESERVE_IVUV */
9250 sv_setnv(sv,Atof(SvPVX_const(sv)) - 1.0); /* punt */
9253 /* this define is used to eliminate a chunk of duplicated but shared logic
9254 * it has the suffix __SV_C to signal that it isnt API, and isnt meant to be
9255 * used anywhere but here - yves
9257 #define PUSH_EXTEND_MORTAL__SV_C(AnSv) \
9259 SSize_t ix = ++PL_tmps_ix; \
9260 if (UNLIKELY(ix >= PL_tmps_max)) \
9261 ix = tmps_grow_p(ix); \
9262 PL_tmps_stack[ix] = (AnSv); \
9266 =for apidoc sv_mortalcopy
9268 Creates a new SV which is a copy of the original SV (using C<sv_setsv>).
9269 The new SV is marked as mortal. It will be destroyed "soon", either by an
9270 explicit call to C<FREETMPS>, or by an implicit call at places such as
9271 statement boundaries. See also C<L</sv_newmortal>> and C<L</sv_2mortal>>.
9276 /* Make a string that will exist for the duration of the expression
9277 * evaluation. Actually, it may have to last longer than that, but
9278 * hopefully we won't free it until it has been assigned to a
9279 * permanent location. */
9282 Perl_sv_mortalcopy_flags(pTHX_ SV *const oldstr, U32 flags)
9286 if (flags & SV_GMAGIC)
9287 SvGETMAGIC(oldstr); /* before new_SV, in case it dies */
9289 sv_setsv_flags(sv,oldstr,flags & ~SV_GMAGIC);
9290 PUSH_EXTEND_MORTAL__SV_C(sv);
9296 =for apidoc sv_newmortal
9298 Creates a new null SV which is mortal. The reference count of the SV is
9299 set to 1. It will be destroyed "soon", either by an explicit call to
9300 C<FREETMPS>, or by an implicit call at places such as statement boundaries.
9301 See also C<L</sv_mortalcopy>> and C<L</sv_2mortal>>.
9307 Perl_sv_newmortal(pTHX)
9312 SvFLAGS(sv) = SVs_TEMP;
9313 PUSH_EXTEND_MORTAL__SV_C(sv);
9319 =for apidoc newSVpvn_flags
9321 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9322 characters) into it. The reference count for the
9323 SV is set to 1. Note that if C<len> is zero, Perl will create a zero length
9324 string. You are responsible for ensuring that the source string is at least
9325 C<len> bytes long. If the C<s> argument is NULL the new SV will be undefined.
9326 Currently the only flag bits accepted are C<SVf_UTF8> and C<SVs_TEMP>.
9327 If C<SVs_TEMP> is set, then C<sv_2mortal()> is called on the result before
9328 returning. If C<SVf_UTF8> is set, C<s>
9329 is considered to be in UTF-8 and the
9330 C<SVf_UTF8> flag will be set on the new SV.
9331 C<newSVpvn_utf8()> is a convenience wrapper for this function, defined as
9333 #define newSVpvn_utf8(s, len, u) \
9334 newSVpvn_flags((s), (len), (u) ? SVf_UTF8 : 0)
9340 Perl_newSVpvn_flags(pTHX_ const char *const s, const STRLEN len, const U32 flags)
9344 /* All the flags we don't support must be zero.
9345 And we're new code so I'm going to assert this from the start. */
9346 assert(!(flags & ~(SVf_UTF8|SVs_TEMP)));
9348 sv_setpvn(sv,s,len);
9350 /* This code used to do a sv_2mortal(), however we now unroll the call to
9351 * sv_2mortal() and do what it does ourselves here. Since we have asserted
9352 * that flags can only have the SVf_UTF8 and/or SVs_TEMP flags set above we
9353 * can use it to enable the sv flags directly (bypassing SvTEMP_on), which
9354 * in turn means we dont need to mask out the SVf_UTF8 flag below, which
9355 * means that we eliminate quite a few steps than it looks - Yves
9356 * (explaining patch by gfx) */
9358 SvFLAGS(sv) |= flags;
9360 if(flags & SVs_TEMP){
9361 PUSH_EXTEND_MORTAL__SV_C(sv);
9368 =for apidoc sv_2mortal
9370 Marks an existing SV as mortal. The SV will be destroyed "soon", either
9371 by an explicit call to C<FREETMPS>, or by an implicit call at places such as
9372 statement boundaries. C<SvTEMP()> is turned on which means that the SV's
9373 string buffer can be "stolen" if this SV is copied. See also
9374 C<L</sv_newmortal>> and C<L</sv_mortalcopy>>.
9380 Perl_sv_2mortal(pTHX_ SV *const sv)
9387 PUSH_EXTEND_MORTAL__SV_C(sv);
9395 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9396 characters) into it. The reference count for the
9397 SV is set to 1. If C<len> is zero, Perl will compute the length using
9398 C<strlen()>, (which means if you use this option, that C<s> can't have embedded
9399 C<NUL> characters and has to have a terminating C<NUL> byte).
9401 This function can cause reliability issues if you are likely to pass in
9402 empty strings that are not null terminated, because it will run
9403 strlen on the string and potentially run past valid memory.
9405 Using L</newSVpvn> is a safer alternative for non C<NUL> terminated strings.
9406 For string literals use L</newSVpvs> instead. This function will work fine for
9407 C<NUL> terminated strings, but if you want to avoid the if statement on whether
9408 to call C<strlen> use C<newSVpvn> instead (calling C<strlen> yourself).
9414 Perl_newSVpv(pTHX_ const char *const s, const STRLEN len)
9419 sv_setpvn(sv, s, len || s == NULL ? len : strlen(s));
9424 =for apidoc newSVpvn
9426 Creates a new SV and copies a string into it, which may contain C<NUL> characters
9427 (C<\0>) and other binary data. The reference count for the SV is set to 1.
9428 Note that if C<len> is zero, Perl will create a zero length (Perl) string. You
9429 are responsible for ensuring that the source buffer is at least
9430 C<len> bytes long. If the C<buffer> argument is NULL the new SV will be
9437 Perl_newSVpvn(pTHX_ const char *const buffer, const STRLEN len)
9441 sv_setpvn(sv,buffer,len);
9446 =for apidoc newSVhek
9448 Creates a new SV from the hash key structure. It will generate scalars that
9449 point to the shared string table where possible. Returns a new (undefined)
9450 SV if C<hek> is NULL.
9456 Perl_newSVhek(pTHX_ const HEK *const hek)
9465 if (HEK_LEN(hek) == HEf_SVKEY) {
9466 return newSVsv(*(SV**)HEK_KEY(hek));
9468 const int flags = HEK_FLAGS(hek);
9469 if (flags & HVhek_WASUTF8) {
9471 Andreas would like keys he put in as utf8 to come back as utf8
9473 STRLEN utf8_len = HEK_LEN(hek);
9474 SV * const sv = newSV_type(SVt_PV);
9475 char *as_utf8 = (char *)bytes_to_utf8 ((U8*)HEK_KEY(hek), &utf8_len);
9476 /* bytes_to_utf8() allocates a new string, which we can repurpose: */
9477 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
9480 } else if (flags & HVhek_UNSHARED) {
9481 /* A hash that isn't using shared hash keys has to have
9482 the flag in every key so that we know not to try to call
9483 share_hek_hek on it. */
9485 SV * const sv = newSVpvn (HEK_KEY(hek), HEK_LEN(hek));
9490 /* This will be overwhelminly the most common case. */
9492 /* Inline most of newSVpvn_share(), because share_hek_hek() is far
9493 more efficient than sharepvn(). */
9497 sv_upgrade(sv, SVt_PV);
9498 SvPV_set(sv, (char *)HEK_KEY(share_hek_hek(hek)));
9499 SvCUR_set(sv, HEK_LEN(hek));
9511 =for apidoc newSVpvn_share
9513 Creates a new SV with its C<SvPVX_const> pointing to a shared string in the string
9514 table. If the string does not already exist in the table, it is
9515 created first. Turns on the C<SvIsCOW> flag (or C<READONLY>
9516 and C<FAKE> in 5.16 and earlier). If the C<hash> parameter
9517 is non-zero, that value is used; otherwise the hash is computed.
9518 The string's hash can later be retrieved from the SV
9519 with the C<SvSHARED_HASH()> macro. The idea here is
9520 that as the string table is used for shared hash keys these strings will have
9521 C<SvPVX_const == HeKEY> and hash lookup will avoid string compare.
9527 Perl_newSVpvn_share(pTHX_ const char *src, I32 len, U32 hash)
9531 bool is_utf8 = FALSE;
9532 const char *const orig_src = src;
9535 STRLEN tmplen = -len;
9537 /* See the note in hv.c:hv_fetch() --jhi */
9538 src = (char*)bytes_from_utf8((const U8*)src, &tmplen, &is_utf8);
9542 PERL_HASH(hash, src, len);
9544 /* The logic for this is inlined in S_mro_get_linear_isa_dfs(), so if it
9545 changes here, update it there too. */
9546 sv_upgrade(sv, SVt_PV);
9547 SvPV_set(sv, sharepvn(src, is_utf8?-len:len, hash));
9554 if (src != orig_src)
9560 =for apidoc newSVpv_share
9562 Like C<newSVpvn_share>, but takes a C<NUL>-terminated string instead of a
9569 Perl_newSVpv_share(pTHX_ const char *src, U32 hash)
9571 return newSVpvn_share(src, strlen(src), hash);
9574 #if defined(PERL_IMPLICIT_CONTEXT)
9576 /* pTHX_ magic can't cope with varargs, so this is a no-context
9577 * version of the main function, (which may itself be aliased to us).
9578 * Don't access this version directly.
9582 Perl_newSVpvf_nocontext(const char *const pat, ...)
9588 PERL_ARGS_ASSERT_NEWSVPVF_NOCONTEXT;
9590 va_start(args, pat);
9591 sv = vnewSVpvf(pat, &args);
9598 =for apidoc newSVpvf
9600 Creates a new SV and initializes it with the string formatted like
9607 Perl_newSVpvf(pTHX_ const char *const pat, ...)
9612 PERL_ARGS_ASSERT_NEWSVPVF;
9614 va_start(args, pat);
9615 sv = vnewSVpvf(pat, &args);
9620 /* backend for newSVpvf() and newSVpvf_nocontext() */
9623 Perl_vnewSVpvf(pTHX_ const char *const pat, va_list *const args)
9627 PERL_ARGS_ASSERT_VNEWSVPVF;
9630 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
9637 Creates a new SV and copies a floating point value into it.
9638 The reference count for the SV is set to 1.
9644 Perl_newSVnv(pTHX_ const NV n)
9656 Creates a new SV and copies an integer into it. The reference count for the
9663 Perl_newSViv(pTHX_ const IV i)
9669 /* Inlining ONLY the small relevant subset of sv_setiv here
9670 * for performance. Makes a significant difference. */
9672 /* We're starting from SVt_FIRST, so provided that's
9673 * actual 0, we don't have to unset any SV type flags
9674 * to promote to SVt_IV. */
9675 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9677 SET_SVANY_FOR_BODYLESS_IV(sv);
9678 SvFLAGS(sv) |= SVt_IV;
9690 Creates a new SV and copies an unsigned integer into it.
9691 The reference count for the SV is set to 1.
9697 Perl_newSVuv(pTHX_ const UV u)
9701 /* Inlining ONLY the small relevant subset of sv_setuv here
9702 * for performance. Makes a significant difference. */
9704 /* Using ivs is more efficient than using uvs - see sv_setuv */
9705 if (u <= (UV)IV_MAX) {
9706 return newSViv((IV)u);
9711 /* We're starting from SVt_FIRST, so provided that's
9712 * actual 0, we don't have to unset any SV type flags
9713 * to promote to SVt_IV. */
9714 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9716 SET_SVANY_FOR_BODYLESS_IV(sv);
9717 SvFLAGS(sv) |= SVt_IV;
9719 (void)SvIsUV_on(sv);
9728 =for apidoc newSV_type
9730 Creates a new SV, of the type specified. The reference count for the new SV
9737 Perl_newSV_type(pTHX_ const svtype type)
9742 ASSUME(SvTYPE(sv) == SVt_FIRST);
9743 if(type != SVt_FIRST)
9744 sv_upgrade(sv, type);
9749 =for apidoc newRV_noinc
9751 Creates an RV wrapper for an SV. The reference count for the original
9752 SV is B<not> incremented.
9758 Perl_newRV_noinc(pTHX_ SV *const tmpRef)
9762 PERL_ARGS_ASSERT_NEWRV_NOINC;
9766 /* We're starting from SVt_FIRST, so provided that's
9767 * actual 0, we don't have to unset any SV type flags
9768 * to promote to SVt_IV. */
9769 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9771 SET_SVANY_FOR_BODYLESS_IV(sv);
9772 SvFLAGS(sv) |= SVt_IV;
9777 SvRV_set(sv, tmpRef);
9782 /* newRV_inc is the official function name to use now.
9783 * newRV_inc is in fact #defined to newRV in sv.h
9787 Perl_newRV(pTHX_ SV *const sv)
9789 PERL_ARGS_ASSERT_NEWRV;
9791 return newRV_noinc(SvREFCNT_inc_simple_NN(sv));
9797 Creates a new SV which is an exact duplicate of the original SV.
9804 Perl_newSVsv(pTHX_ SV *const old)
9810 if (SvTYPE(old) == (svtype)SVTYPEMASK) {
9811 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), "semi-panic: attempt to dup freed string");
9814 /* Do this here, otherwise we leak the new SV if this croaks. */
9817 /* SV_NOSTEAL prevents TEMP buffers being, well, stolen, and saves games
9818 with SvTEMP_off and SvTEMP_on round a call to sv_setsv. */
9819 sv_setsv_flags(sv, old, SV_NOSTEAL);
9824 =for apidoc sv_reset
9826 Underlying implementation for the C<reset> Perl function.
9827 Note that the perl-level function is vaguely deprecated.
9833 Perl_sv_reset(pTHX_ const char *s, HV *const stash)
9835 PERL_ARGS_ASSERT_SV_RESET;
9837 sv_resetpvn(*s ? s : NULL, strlen(s), stash);
9841 Perl_sv_resetpvn(pTHX_ const char *s, STRLEN len, HV * const stash)
9843 char todo[PERL_UCHAR_MAX+1];
9846 if (!stash || SvTYPE(stash) != SVt_PVHV)
9849 if (!s) { /* reset ?? searches */
9850 MAGIC * const mg = mg_find((const SV *)stash, PERL_MAGIC_symtab);
9852 const U32 count = mg->mg_len / sizeof(PMOP**);
9853 PMOP **pmp = (PMOP**) mg->mg_ptr;
9854 PMOP *const *const end = pmp + count;
9858 SvREADONLY_off(PL_regex_pad[(*pmp)->op_pmoffset]);
9860 (*pmp)->op_pmflags &= ~PMf_USED;
9868 /* reset variables */
9870 if (!HvARRAY(stash))
9873 Zero(todo, 256, char);
9877 I32 i = (unsigned char)*s;
9881 max = (unsigned char)*s++;
9882 for ( ; i <= max; i++) {
9885 for (i = 0; i <= (I32) HvMAX(stash); i++) {
9887 for (entry = HvARRAY(stash)[i];
9889 entry = HeNEXT(entry))
9894 if (!todo[(U8)*HeKEY(entry)])
9896 gv = MUTABLE_GV(HeVAL(entry));
9900 if (sv && !SvREADONLY(sv)) {
9901 SV_CHECK_THINKFIRST_COW_DROP(sv);
9902 if (!isGV(sv)) SvOK_off(sv);
9907 if (GvHV(gv) && !HvNAME_get(GvHV(gv))) {
9918 Using various gambits, try to get an IO from an SV: the IO slot if its a
9919 GV; or the recursive result if we're an RV; or the IO slot of the symbol
9920 named after the PV if we're a string.
9922 'Get' magic is ignored on the C<sv> passed in, but will be called on
9923 C<SvRV(sv)> if C<sv> is an RV.
9929 Perl_sv_2io(pTHX_ SV *const sv)
9934 PERL_ARGS_ASSERT_SV_2IO;
9936 switch (SvTYPE(sv)) {
9938 io = MUTABLE_IO(sv);
9942 if (isGV_with_GP(sv)) {
9943 gv = MUTABLE_GV(sv);
9946 Perl_croak(aTHX_ "Bad filehandle: %" HEKf,
9947 HEKfARG(GvNAME_HEK(gv)));
9953 Perl_croak(aTHX_ PL_no_usym, "filehandle");
9955 SvGETMAGIC(SvRV(sv));
9956 return sv_2io(SvRV(sv));
9958 gv = gv_fetchsv_nomg(sv, 0, SVt_PVIO);
9965 if (SvGMAGICAL(sv)) {
9966 newsv = sv_newmortal();
9967 sv_setsv_nomg(newsv, sv);
9969 Perl_croak(aTHX_ "Bad filehandle: %" SVf, SVfARG(newsv));
9979 Using various gambits, try to get a CV from an SV; in addition, try if
9980 possible to set C<*st> and C<*gvp> to the stash and GV associated with it.
9981 The flags in C<lref> are passed to C<gv_fetchsv>.
9987 Perl_sv_2cv(pTHX_ SV *sv, HV **const st, GV **const gvp, const I32 lref)
9992 PERL_ARGS_ASSERT_SV_2CV;
9999 switch (SvTYPE(sv)) {
10003 return MUTABLE_CV(sv);
10013 sv = amagic_deref_call(sv, to_cv_amg);
10016 if (SvTYPE(sv) == SVt_PVCV) {
10017 cv = MUTABLE_CV(sv);
10022 else if(SvGETMAGIC(sv), isGV_with_GP(sv))
10023 gv = MUTABLE_GV(sv);
10025 Perl_croak(aTHX_ "Not a subroutine reference");
10027 else if (isGV_with_GP(sv)) {
10028 gv = MUTABLE_GV(sv);
10031 gv = gv_fetchsv_nomg(sv, lref, SVt_PVCV);
10038 /* Some flags to gv_fetchsv mean don't really create the GV */
10039 if (!isGV_with_GP(gv)) {
10043 *st = GvESTASH(gv);
10044 if (lref & ~GV_ADDMG && !GvCVu(gv)) {
10045 /* XXX this is probably not what they think they're getting.
10046 * It has the same effect as "sub name;", i.e. just a forward
10055 =for apidoc sv_true
10057 Returns true if the SV has a true value by Perl's rules.
10058 Use the C<SvTRUE> macro instead, which may call C<sv_true()> or may
10059 instead use an in-line version.
10065 Perl_sv_true(pTHX_ SV *const sv)
10070 const XPV* const tXpv = (XPV*)SvANY(sv);
10072 (tXpv->xpv_cur > 1 ||
10073 (tXpv->xpv_cur && *sv->sv_u.svu_pv != '0')))
10080 return SvIVX(sv) != 0;
10083 return SvNVX(sv) != 0.0;
10085 return sv_2bool(sv);
10091 =for apidoc sv_pvn_force
10093 Get a sensible string out of the SV somehow.
10094 A private implementation of the C<SvPV_force> macro for compilers which
10095 can't cope with complex macro expressions. Always use the macro instead.
10097 =for apidoc sv_pvn_force_flags
10099 Get a sensible string out of the SV somehow.
10100 If C<flags> has the C<SV_GMAGIC> bit set, will C<mg_get> on C<sv> if
10101 appropriate, else not. C<sv_pvn_force> and C<sv_pvn_force_nomg> are
10102 implemented in terms of this function.
10103 You normally want to use the various wrapper macros instead: see
10104 C<L</SvPV_force>> and C<L</SvPV_force_nomg>>.
10110 Perl_sv_pvn_force_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
10112 PERL_ARGS_ASSERT_SV_PVN_FORCE_FLAGS;
10114 if (flags & SV_GMAGIC) SvGETMAGIC(sv);
10115 if (SvTHINKFIRST(sv) && (!SvROK(sv) || SvREADONLY(sv)))
10116 sv_force_normal_flags(sv, 0);
10126 if (SvTYPE(sv) > SVt_PVLV
10127 || isGV_with_GP(sv))
10128 /* diag_listed_as: Can't coerce %s to %s in %s */
10129 Perl_croak(aTHX_ "Can't coerce %s to string in %s", sv_reftype(sv,0),
10131 s = sv_2pv_flags(sv, &len, flags &~ SV_GMAGIC);
10138 if (SvTYPE(sv) < SVt_PV ||
10139 s != SvPVX_const(sv)) { /* Almost, but not quite, sv_setpvn() */
10142 SvUPGRADE(sv, SVt_PV); /* Never FALSE */
10143 SvGROW(sv, len + 1);
10144 Move(s,SvPVX(sv),len,char);
10145 SvCUR_set(sv, len);
10146 SvPVX(sv)[len] = '\0';
10149 SvPOK_on(sv); /* validate pointer */
10151 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2pv(%s)\n",
10152 PTR2UV(sv),SvPVX_const(sv)));
10155 (void)SvPOK_only_UTF8(sv);
10156 return SvPVX_mutable(sv);
10160 =for apidoc sv_pvbyten_force
10162 The backend for the C<SvPVbytex_force> macro. Always use the macro
10169 Perl_sv_pvbyten_force(pTHX_ SV *const sv, STRLEN *const lp)
10171 PERL_ARGS_ASSERT_SV_PVBYTEN_FORCE;
10173 sv_pvn_force(sv,lp);
10174 sv_utf8_downgrade(sv,0);
10180 =for apidoc sv_pvutf8n_force
10182 The backend for the C<SvPVutf8x_force> macro. Always use the macro
10189 Perl_sv_pvutf8n_force(pTHX_ SV *const sv, STRLEN *const lp)
10191 PERL_ARGS_ASSERT_SV_PVUTF8N_FORCE;
10193 sv_pvn_force(sv,0);
10194 sv_utf8_upgrade_nomg(sv);
10200 =for apidoc sv_reftype
10202 Returns a string describing what the SV is a reference to.
10204 If ob is true and the SV is blessed, the string is the class name,
10205 otherwise it is the type of the SV, "SCALAR", "ARRAY" etc.
10211 Perl_sv_reftype(pTHX_ const SV *const sv, const int ob)
10213 PERL_ARGS_ASSERT_SV_REFTYPE;
10214 if (ob && SvOBJECT(sv)) {
10215 return SvPV_nolen_const(sv_ref(NULL, sv, ob));
10218 /* WARNING - There is code, for instance in mg.c, that assumes that
10219 * the only reason that sv_reftype(sv,0) would return a string starting
10220 * with 'L' or 'S' is that it is a LVALUE or a SCALAR.
10221 * Yes this a dodgy way to do type checking, but it saves practically reimplementing
10222 * this routine inside other subs, and it saves time.
10223 * Do not change this assumption without searching for "dodgy type check" in
10226 switch (SvTYPE(sv)) {
10241 case SVt_PVLV: return (char *) (SvROK(sv) ? "REF"
10242 /* tied lvalues should appear to be
10243 * scalars for backwards compatibility */
10244 : (isALPHA_FOLD_EQ(LvTYPE(sv), 't'))
10245 ? "SCALAR" : "LVALUE");
10246 case SVt_PVAV: return "ARRAY";
10247 case SVt_PVHV: return "HASH";
10248 case SVt_PVCV: return "CODE";
10249 case SVt_PVGV: return (char *) (isGV_with_GP(sv)
10250 ? "GLOB" : "SCALAR");
10251 case SVt_PVFM: return "FORMAT";
10252 case SVt_PVIO: return "IO";
10253 case SVt_INVLIST: return "INVLIST";
10254 case SVt_REGEXP: return "REGEXP";
10255 default: return "UNKNOWN";
10263 Returns a SV describing what the SV passed in is a reference to.
10265 dst can be a SV to be set to the description or NULL, in which case a
10266 mortal SV is returned.
10268 If ob is true and the SV is blessed, the description is the class
10269 name, otherwise it is the type of the SV, "SCALAR", "ARRAY" etc.
10275 Perl_sv_ref(pTHX_ SV *dst, const SV *const sv, const int ob)
10277 PERL_ARGS_ASSERT_SV_REF;
10280 dst = sv_newmortal();
10282 if (ob && SvOBJECT(sv)) {
10283 HvNAME_get(SvSTASH(sv))
10284 ? sv_sethek(dst, HvNAME_HEK(SvSTASH(sv)))
10285 : sv_setpvs(dst, "__ANON__");
10288 const char * reftype = sv_reftype(sv, 0);
10289 sv_setpv(dst, reftype);
10295 =for apidoc sv_isobject
10297 Returns a boolean indicating whether the SV is an RV pointing to a blessed
10298 object. If the SV is not an RV, or if the object is not blessed, then this
10305 Perl_sv_isobject(pTHX_ SV *sv)
10321 Returns a boolean indicating whether the SV is blessed into the specified
10322 class. This does not check for subtypes; use C<sv_derived_from> to verify
10323 an inheritance relationship.
10329 Perl_sv_isa(pTHX_ SV *sv, const char *const name)
10331 const char *hvname;
10333 PERL_ARGS_ASSERT_SV_ISA;
10343 hvname = HvNAME_get(SvSTASH(sv));
10347 return strEQ(hvname, name);
10351 =for apidoc newSVrv
10353 Creates a new SV for the existing RV, C<rv>, to point to. If C<rv> is not an
10354 RV then it will be upgraded to one. If C<classname> is non-null then the new
10355 SV will be blessed in the specified package. The new SV is returned and its
10356 reference count is 1. The reference count 1 is owned by C<rv>.
10362 Perl_newSVrv(pTHX_ SV *const rv, const char *const classname)
10366 PERL_ARGS_ASSERT_NEWSVRV;
10370 SV_CHECK_THINKFIRST_COW_DROP(rv);
10372 if (UNLIKELY( SvTYPE(rv) >= SVt_PVMG )) {
10373 const U32 refcnt = SvREFCNT(rv);
10377 SvREFCNT(rv) = refcnt;
10379 sv_upgrade(rv, SVt_IV);
10380 } else if (SvROK(rv)) {
10381 SvREFCNT_dec(SvRV(rv));
10383 prepare_SV_for_RV(rv);
10391 HV* const stash = gv_stashpv(classname, GV_ADD);
10392 (void)sv_bless(rv, stash);
10398 Perl_newSVavdefelem(pTHX_ AV *av, SSize_t ix, bool extendible)
10400 SV * const lv = newSV_type(SVt_PVLV);
10401 PERL_ARGS_ASSERT_NEWSVAVDEFELEM;
10403 sv_magic(lv, NULL, PERL_MAGIC_defelem, NULL, 0);
10404 LvTARG(lv) = SvREFCNT_inc_simple_NN(av);
10405 LvSTARGOFF(lv) = ix;
10406 LvTARGLEN(lv) = extendible ? 1 : (STRLEN)UV_MAX;
10411 =for apidoc sv_setref_pv
10413 Copies a pointer into a new SV, optionally blessing the SV. The C<rv>
10414 argument will be upgraded to an RV. That RV will be modified to point to
10415 the new SV. If the C<pv> argument is C<NULL>, then C<PL_sv_undef> will be placed
10416 into the SV. The C<classname> argument indicates the package for the
10417 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10418 will have a reference count of 1, and the RV will be returned.
10420 Do not use with other Perl types such as HV, AV, SV, CV, because those
10421 objects will become corrupted by the pointer copy process.
10423 Note that C<sv_setref_pvn> copies the string while this copies the pointer.
10429 Perl_sv_setref_pv(pTHX_ SV *const rv, const char *const classname, void *const pv)
10431 PERL_ARGS_ASSERT_SV_SETREF_PV;
10438 sv_setiv(newSVrv(rv,classname), PTR2IV(pv));
10443 =for apidoc sv_setref_iv
10445 Copies an integer into a new SV, optionally blessing the SV. The C<rv>
10446 argument will be upgraded to an RV. That RV will be modified to point to
10447 the new SV. The C<classname> argument indicates the package for the
10448 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10449 will have a reference count of 1, and the RV will be returned.
10455 Perl_sv_setref_iv(pTHX_ SV *const rv, const char *const classname, const IV iv)
10457 PERL_ARGS_ASSERT_SV_SETREF_IV;
10459 sv_setiv(newSVrv(rv,classname), iv);
10464 =for apidoc sv_setref_uv
10466 Copies an unsigned integer into a new SV, optionally blessing the SV. The C<rv>
10467 argument will be upgraded to an RV. That RV will be modified to point to
10468 the new SV. The C<classname> argument indicates the package for the
10469 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10470 will have a reference count of 1, and the RV will be returned.
10476 Perl_sv_setref_uv(pTHX_ SV *const rv, const char *const classname, const UV uv)
10478 PERL_ARGS_ASSERT_SV_SETREF_UV;
10480 sv_setuv(newSVrv(rv,classname), uv);
10485 =for apidoc sv_setref_nv
10487 Copies a double into a new SV, optionally blessing the SV. The C<rv>
10488 argument will be upgraded to an RV. That RV will be modified to point to
10489 the new SV. The C<classname> argument indicates the package for the
10490 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10491 will have a reference count of 1, and the RV will be returned.
10497 Perl_sv_setref_nv(pTHX_ SV *const rv, const char *const classname, const NV nv)
10499 PERL_ARGS_ASSERT_SV_SETREF_NV;
10501 sv_setnv(newSVrv(rv,classname), nv);
10506 =for apidoc sv_setref_pvn
10508 Copies a string into a new SV, optionally blessing the SV. The length of the
10509 string must be specified with C<n>. The C<rv> argument will be upgraded to
10510 an RV. That RV will be modified to point to the new SV. The C<classname>
10511 argument indicates the package for the blessing. Set C<classname> to
10512 C<NULL> to avoid the blessing. The new SV will have a reference count
10513 of 1, and the RV will be returned.
10515 Note that C<sv_setref_pv> copies the pointer while this copies the string.
10521 Perl_sv_setref_pvn(pTHX_ SV *const rv, const char *const classname,
10522 const char *const pv, const STRLEN n)
10524 PERL_ARGS_ASSERT_SV_SETREF_PVN;
10526 sv_setpvn(newSVrv(rv,classname), pv, n);
10531 =for apidoc sv_bless
10533 Blesses an SV into a specified package. The SV must be an RV. The package
10534 must be designated by its stash (see C<L</gv_stashpv>>). The reference count
10535 of the SV is unaffected.
10541 Perl_sv_bless(pTHX_ SV *const sv, HV *const stash)
10544 HV *oldstash = NULL;
10546 PERL_ARGS_ASSERT_SV_BLESS;
10550 Perl_croak(aTHX_ "Can't bless non-reference value");
10552 if (SvFLAGS(tmpRef) & (SVs_OBJECT|SVf_READONLY|SVf_PROTECT)) {
10553 if (SvREADONLY(tmpRef))
10554 Perl_croak_no_modify();
10555 if (SvOBJECT(tmpRef)) {
10556 oldstash = SvSTASH(tmpRef);
10559 SvOBJECT_on(tmpRef);
10560 SvUPGRADE(tmpRef, SVt_PVMG);
10561 SvSTASH_set(tmpRef, MUTABLE_HV(SvREFCNT_inc_simple(stash)));
10562 SvREFCNT_dec(oldstash);
10564 if(SvSMAGICAL(tmpRef))
10565 if(mg_find(tmpRef, PERL_MAGIC_ext) || mg_find(tmpRef, PERL_MAGIC_uvar))
10573 /* Downgrades a PVGV to a PVMG. If it's actually a PVLV, we leave the type
10574 * as it is after unglobbing it.
10577 PERL_STATIC_INLINE void
10578 S_sv_unglob(pTHX_ SV *const sv, U32 flags)
10582 SV * const temp = flags & SV_COW_DROP_PV ? NULL : sv_newmortal();
10584 PERL_ARGS_ASSERT_SV_UNGLOB;
10586 assert(SvTYPE(sv) == SVt_PVGV || SvTYPE(sv) == SVt_PVLV);
10588 if (!(flags & SV_COW_DROP_PV))
10589 gv_efullname3(temp, MUTABLE_GV(sv), "*");
10591 SvREFCNT_inc_simple_void_NN(sv_2mortal(sv));
10593 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
10594 && HvNAME_get(stash))
10595 mro_method_changed_in(stash);
10596 gp_free(MUTABLE_GV(sv));
10599 sv_del_backref(MUTABLE_SV(GvSTASH(sv)), sv);
10600 GvSTASH(sv) = NULL;
10603 if (GvNAME_HEK(sv)) {
10604 unshare_hek(GvNAME_HEK(sv));
10606 isGV_with_GP_off(sv);
10608 if(SvTYPE(sv) == SVt_PVGV) {
10609 /* need to keep SvANY(sv) in the right arena */
10610 xpvmg = new_XPVMG();
10611 StructCopy(SvANY(sv), xpvmg, XPVMG);
10612 del_XPVGV(SvANY(sv));
10615 SvFLAGS(sv) &= ~SVTYPEMASK;
10616 SvFLAGS(sv) |= SVt_PVMG;
10619 /* Intentionally not calling any local SET magic, as this isn't so much a
10620 set operation as merely an internal storage change. */
10621 if (flags & SV_COW_DROP_PV) SvOK_off(sv);
10622 else sv_setsv_flags(sv, temp, 0);
10624 if ((const GV *)sv == PL_last_in_gv)
10625 PL_last_in_gv = NULL;
10626 else if ((const GV *)sv == PL_statgv)
10631 =for apidoc sv_unref_flags
10633 Unsets the RV status of the SV, and decrements the reference count of
10634 whatever was being referenced by the RV. This can almost be thought of
10635 as a reversal of C<newSVrv>. The C<cflags> argument can contain
10636 C<SV_IMMEDIATE_UNREF> to force the reference count to be decremented
10637 (otherwise the decrementing is conditional on the reference count being
10638 different from one or the reference being a readonly SV).
10639 See C<L</SvROK_off>>.
10645 Perl_sv_unref_flags(pTHX_ SV *const ref, const U32 flags)
10647 SV* const target = SvRV(ref);
10649 PERL_ARGS_ASSERT_SV_UNREF_FLAGS;
10651 if (SvWEAKREF(ref)) {
10652 sv_del_backref(target, ref);
10653 SvWEAKREF_off(ref);
10654 SvRV_set(ref, NULL);
10657 SvRV_set(ref, NULL);
10659 /* You can't have a || SvREADONLY(target) here, as $a = $$a, where $a was
10660 assigned to as BEGIN {$a = \"Foo"} will fail. */
10661 if (SvREFCNT(target) != 1 || (flags & SV_IMMEDIATE_UNREF))
10662 SvREFCNT_dec_NN(target);
10663 else /* XXX Hack, but hard to make $a=$a->[1] work otherwise */
10664 sv_2mortal(target); /* Schedule for freeing later */
10668 =for apidoc sv_untaint
10670 Untaint an SV. Use C<SvTAINTED_off> instead.
10676 Perl_sv_untaint(pTHX_ SV *const sv)
10678 PERL_ARGS_ASSERT_SV_UNTAINT;
10679 PERL_UNUSED_CONTEXT;
10681 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10682 MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10689 =for apidoc sv_tainted
10691 Test an SV for taintedness. Use C<SvTAINTED> instead.
10697 Perl_sv_tainted(pTHX_ SV *const sv)
10699 PERL_ARGS_ASSERT_SV_TAINTED;
10700 PERL_UNUSED_CONTEXT;
10702 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10703 const MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10704 if (mg && (mg->mg_len & 1) )
10710 #ifndef NO_MATHOMS /* Can't move these to mathoms.c because call uiv_2buf(),
10711 private to this file */
10714 =for apidoc sv_setpviv
10716 Copies an integer into the given SV, also updating its string value.
10717 Does not handle 'set' magic. See C<L</sv_setpviv_mg>>.
10723 Perl_sv_setpviv(pTHX_ SV *const sv, const IV iv)
10725 char buf[TYPE_CHARS(UV)];
10727 char * const ptr = uiv_2buf(buf, iv, 0, 0, &ebuf);
10729 PERL_ARGS_ASSERT_SV_SETPVIV;
10731 sv_setpvn(sv, ptr, ebuf - ptr);
10735 =for apidoc sv_setpviv_mg
10737 Like C<sv_setpviv>, but also handles 'set' magic.
10743 Perl_sv_setpviv_mg(pTHX_ SV *const sv, const IV iv)
10745 PERL_ARGS_ASSERT_SV_SETPVIV_MG;
10747 sv_setpviv(sv, iv);
10751 #endif /* NO_MATHOMS */
10753 #if defined(PERL_IMPLICIT_CONTEXT)
10755 /* pTHX_ magic can't cope with varargs, so this is a no-context
10756 * version of the main function, (which may itself be aliased to us).
10757 * Don't access this version directly.
10761 Perl_sv_setpvf_nocontext(SV *const sv, const char *const pat, ...)
10766 PERL_ARGS_ASSERT_SV_SETPVF_NOCONTEXT;
10768 va_start(args, pat);
10769 sv_vsetpvf(sv, pat, &args);
10773 /* pTHX_ magic can't cope with varargs, so this is a no-context
10774 * version of the main function, (which may itself be aliased to us).
10775 * Don't access this version directly.
10779 Perl_sv_setpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10784 PERL_ARGS_ASSERT_SV_SETPVF_MG_NOCONTEXT;
10786 va_start(args, pat);
10787 sv_vsetpvf_mg(sv, pat, &args);
10793 =for apidoc sv_setpvf
10795 Works like C<sv_catpvf> but copies the text into the SV instead of
10796 appending it. Does not handle 'set' magic. See C<L</sv_setpvf_mg>>.
10802 Perl_sv_setpvf(pTHX_ SV *const sv, const char *const pat, ...)
10806 PERL_ARGS_ASSERT_SV_SETPVF;
10808 va_start(args, pat);
10809 sv_vsetpvf(sv, pat, &args);
10814 =for apidoc sv_vsetpvf
10816 Works like C<sv_vcatpvf> but copies the text into the SV instead of
10817 appending it. Does not handle 'set' magic. See C<L</sv_vsetpvf_mg>>.
10819 Usually used via its frontend C<sv_setpvf>.
10825 Perl_sv_vsetpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10827 PERL_ARGS_ASSERT_SV_VSETPVF;
10829 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10833 =for apidoc sv_setpvf_mg
10835 Like C<sv_setpvf>, but also handles 'set' magic.
10841 Perl_sv_setpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10845 PERL_ARGS_ASSERT_SV_SETPVF_MG;
10847 va_start(args, pat);
10848 sv_vsetpvf_mg(sv, pat, &args);
10853 =for apidoc sv_vsetpvf_mg
10855 Like C<sv_vsetpvf>, but also handles 'set' magic.
10857 Usually used via its frontend C<sv_setpvf_mg>.
10863 Perl_sv_vsetpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10865 PERL_ARGS_ASSERT_SV_VSETPVF_MG;
10867 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10871 #if defined(PERL_IMPLICIT_CONTEXT)
10873 /* pTHX_ magic can't cope with varargs, so this is a no-context
10874 * version of the main function, (which may itself be aliased to us).
10875 * Don't access this version directly.
10879 Perl_sv_catpvf_nocontext(SV *const sv, const char *const pat, ...)
10884 PERL_ARGS_ASSERT_SV_CATPVF_NOCONTEXT;
10886 va_start(args, pat);
10887 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10891 /* pTHX_ magic can't cope with varargs, so this is a no-context
10892 * version of the main function, (which may itself be aliased to us).
10893 * Don't access this version directly.
10897 Perl_sv_catpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10902 PERL_ARGS_ASSERT_SV_CATPVF_MG_NOCONTEXT;
10904 va_start(args, pat);
10905 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10912 =for apidoc sv_catpvf
10914 Processes its arguments like C<sv_catpvfn>, and appends the formatted
10915 output to an SV. As with C<sv_catpvfn> called with a non-null C-style
10916 variable argument list, argument reordering is not supported.
10917 If the appended data contains "wide" characters
10918 (including, but not limited to, SVs with a UTF-8 PV formatted with C<%s>,
10919 and characters >255 formatted with C<%c>), the original SV might get
10920 upgraded to UTF-8. Handles 'get' magic, but not 'set' magic. See
10921 C<L</sv_catpvf_mg>>. If the original SV was UTF-8, the pattern should be
10922 valid UTF-8; if the original SV was bytes, the pattern should be too.
10927 Perl_sv_catpvf(pTHX_ SV *const sv, const char *const pat, ...)
10931 PERL_ARGS_ASSERT_SV_CATPVF;
10933 va_start(args, pat);
10934 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10939 =for apidoc sv_vcatpvf
10941 Processes its arguments like C<sv_catpvfn> called with a non-null C-style
10942 variable argument list, and appends the formatted output
10943 to an SV. Does not handle 'set' magic. See C<L</sv_vcatpvf_mg>>.
10945 Usually used via its frontend C<sv_catpvf>.
10951 Perl_sv_vcatpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10953 PERL_ARGS_ASSERT_SV_VCATPVF;
10955 sv_vcatpvfn_flags(sv, pat, strlen(pat), args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10959 =for apidoc sv_catpvf_mg
10961 Like C<sv_catpvf>, but also handles 'set' magic.
10967 Perl_sv_catpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10971 PERL_ARGS_ASSERT_SV_CATPVF_MG;
10973 va_start(args, pat);
10974 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10980 =for apidoc sv_vcatpvf_mg
10982 Like C<sv_vcatpvf>, but also handles 'set' magic.
10984 Usually used via its frontend C<sv_catpvf_mg>.
10990 Perl_sv_vcatpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10992 PERL_ARGS_ASSERT_SV_VCATPVF_MG;
10994 sv_vcatpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10999 =for apidoc sv_vsetpvfn
11001 Works like C<sv_vcatpvfn> but copies the text into the SV instead of
11004 Usually used via one of its frontends C<sv_vsetpvf> and C<sv_vsetpvf_mg>.
11010 Perl_sv_vsetpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11011 va_list *const args, SV **const svargs, const Size_t sv_count, bool *const maybe_tainted)
11013 PERL_ARGS_ASSERT_SV_VSETPVFN;
11016 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, sv_count, maybe_tainted, 0);
11020 /* simplified inline Perl_sv_catpvn_nomg() when you know the SV's SvPOK */
11022 PERL_STATIC_INLINE void
11023 S_sv_catpvn_simple(pTHX_ SV *const sv, const char* const buf, const STRLEN len)
11025 STRLEN const need = len + SvCUR(sv) + 1;
11028 /* can't wrap as both len and SvCUR() are allocated in
11029 * memory and together can't consume all the address space
11031 assert(need > len);
11036 Copy(buf, end, len, char);
11039 SvCUR_set(sv, need - 1);
11044 * Warn of missing argument to sprintf. The value used in place of such
11045 * arguments should be &PL_sv_no; an undefined value would yield
11046 * inappropriate "use of uninit" warnings [perl #71000].
11049 S_warn_vcatpvfn_missing_argument(pTHX) {
11050 if (ckWARN(WARN_MISSING)) {
11051 Perl_warner(aTHX_ packWARN(WARN_MISSING), "Missing argument in %s",
11052 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11061 Perl_croak(aTHX_ "Integer overflow in format string for %s",
11062 (PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn"));
11066 /* Given an int i from the next arg (if args is true) or an sv from an arg
11067 * (if args is false), try to extract a STRLEN-ranged value from the arg,
11068 * with overflow checking.
11069 * Sets *neg to true if the value was negative (untouched otherwise.
11070 * Returns the absolute value.
11071 * As an extra margin of safety, it croaks if the returned value would
11072 * exceed the maximum value of a STRLEN / 4.
11076 S_sprintf_arg_num_val(pTHX_ va_list *const args, int i, SV *sv, bool *neg)
11090 if (UNLIKELY(SvIsUV(sv))) {
11091 UV uv = SvUV_nomg(sv);
11093 S_croak_overflow();
11097 iv = SvIV_nomg(sv);
11101 S_croak_overflow();
11107 if (iv > (IV)(((STRLEN)~0) / 4))
11108 S_croak_overflow();
11114 /* Returns true if c is in the range '1'..'9'
11115 * Written with the cast so it only needs one conditional test
11117 #define IS_1_TO_9(c) ((U8)(c - '1') <= 8)
11119 /* Read in and return a number. Updates *pattern to point to the char
11120 * following the number. Expects the first char to 1..9.
11121 * Croaks if the number exceeds 1/4 of the maximum value of STRLEN.
11122 * This is a belt-and-braces safety measure to complement any
11123 * overflow/wrap checks done in the main body of sv_vcatpvfn_flags.
11124 * It means that e.g. on a 32-bit system the width/precision can't be more
11125 * than 1G, which seems reasonable.
11129 S_expect_number(pTHX_ const char **const pattern)
11133 PERL_ARGS_ASSERT_EXPECT_NUMBER;
11135 assert(IS_1_TO_9(**pattern));
11137 var = *(*pattern)++ - '0';
11138 while (isDIGIT(**pattern)) {
11139 /* if var * 10 + 9 would exceed 1/4 max strlen, croak */
11140 if (var > ((((STRLEN)~0) / 4 - 9) / 10))
11141 S_croak_overflow();
11142 var = var * 10 + (*(*pattern)++ - '0');
11147 /* Implement a fast "%.0f": given a pointer to the end of a buffer (caller
11148 * ensures it's big enough), back fill it with the rounded integer part of
11149 * nv. Returns ptr to start of string, and sets *len to its length.
11150 * Returns NULL if not convertible.
11154 S_F0convert(NV nv, char *const endbuf, STRLEN *const len)
11156 const int neg = nv < 0;
11159 PERL_ARGS_ASSERT_F0CONVERT;
11161 assert(!Perl_isinfnan(nv));
11168 if (uv & 1 && uv == nv)
11169 uv--; /* Round to even */
11171 const unsigned dig = uv % 10;
11173 } while (uv /= 10);
11183 /* XXX maybe_tainted is never assigned to, so the doc above is lying. */
11186 Perl_sv_vcatpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11187 va_list *const args, SV **const svargs, const Size_t sv_count, bool *const maybe_tainted)
11189 PERL_ARGS_ASSERT_SV_VCATPVFN;
11191 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, sv_count, maybe_tainted, SV_GMAGIC|SV_SMAGIC);
11195 /* For the vcatpvfn code, we need a long double target in case
11196 * HAS_LONG_DOUBLE, even without USE_LONG_DOUBLE, so that we can printf
11197 * with long double formats, even without NV being long double. But we
11198 * call the target 'fv' instead of 'nv', since most of the time it is not
11199 * (most compilers these days recognize "long double", even if only as a
11200 * synonym for "double").
11202 #if defined(HAS_LONG_DOUBLE) && LONG_DOUBLESIZE > DOUBLESIZE && \
11203 defined(PERL_PRIgldbl) && !defined(USE_QUADMATH)
11204 # define VCATPVFN_FV_GF PERL_PRIgldbl
11205 # if defined(__VMS) && defined(__ia64) && defined(__IEEE_FLOAT)
11206 /* Work around breakage in OTS$CVT_FLOAT_T_X */
11207 # define VCATPVFN_NV_TO_FV(nv,fv) \
11210 fv = Perl_isnan(_dv) ? LDBL_QNAN : _dv; \
11213 # define VCATPVFN_NV_TO_FV(nv,fv) (fv)=(nv)
11215 typedef long double vcatpvfn_long_double_t;
11217 # define VCATPVFN_FV_GF NVgf
11218 # define VCATPVFN_NV_TO_FV(nv,fv) (fv)=(nv)
11219 typedef NV vcatpvfn_long_double_t;
11222 #ifdef LONGDOUBLE_DOUBLEDOUBLE
11223 /* The first double can be as large as 2**1023, or '1' x '0' x 1023.
11224 * The second double can be as small as 2**-1074, or '0' x 1073 . '1'.
11225 * The sum of them can be '1' . '0' x 2096 . '1', with implied radix point
11226 * after the first 1023 zero bits.
11228 * XXX The 2098 is quite large (262.25 bytes) and therefore some sort
11229 * of dynamically growing buffer might be better, start at just 16 bytes
11230 * (for example) and grow only when necessary. Or maybe just by looking
11231 * at the exponents of the two doubles? */
11232 # define DOUBLEDOUBLE_MAXBITS 2098
11235 /* vhex will contain the values (0..15) of the hex digits ("nybbles"
11236 * of 4 bits); 1 for the implicit 1, and the mantissa bits, four bits
11237 * per xdigit. For the double-double case, this can be rather many.
11238 * The non-double-double-long-double overshoots since all bits of NV
11239 * are not mantissa bits, there are also exponent bits. */
11240 #ifdef LONGDOUBLE_DOUBLEDOUBLE
11241 # define VHEX_SIZE (3+DOUBLEDOUBLE_MAXBITS/4)
11243 # define VHEX_SIZE (1+(NVSIZE * 8)/4)
11246 /* If we do not have a known long double format, (including not using
11247 * long doubles, or long doubles being equal to doubles) then we will
11248 * fall back to the ldexp/frexp route, with which we can retrieve at
11249 * most as many bits as our widest unsigned integer type is. We try
11250 * to get a 64-bit unsigned integer even if we are not using a 64-bit UV.
11252 * (If you want to test the case of UVSIZE == 4, NVSIZE == 8,
11253 * set the MANTISSATYPE to int and the MANTISSASIZE to 4.)
11255 #if defined(HAS_QUAD) && defined(Uquad_t)
11256 # define MANTISSATYPE Uquad_t
11257 # define MANTISSASIZE 8
11259 # define MANTISSATYPE UV
11260 # define MANTISSASIZE UVSIZE
11263 #if defined(DOUBLE_LITTLE_ENDIAN) || defined(LONGDOUBLE_LITTLE_ENDIAN)
11264 # define HEXTRACT_LITTLE_ENDIAN
11265 #elif defined(DOUBLE_BIG_ENDIAN) || defined(LONGDOUBLE_BIG_ENDIAN)
11266 # define HEXTRACT_BIG_ENDIAN
11268 # define HEXTRACT_MIX_ENDIAN
11271 /* S_hextract() is a helper for S_format_hexfp, for extracting
11272 * the hexadecimal values (for %a/%A). The nv is the NV where the value
11273 * are being extracted from (either directly from the long double in-memory
11274 * presentation, or from the uquad computed via frexp+ldexp). frexp also
11275 * is used to update the exponent. The subnormal is set to true
11276 * for IEEE 754 subnormals/denormals (including the x86 80-bit format).
11277 * The vhex is the pointer to the beginning of the output buffer of VHEX_SIZE.
11279 * The tricky part is that S_hextract() needs to be called twice:
11280 * the first time with vend as NULL, and the second time with vend as
11281 * the pointer returned by the first call. What happens is that on
11282 * the first round the output size is computed, and the intended
11283 * extraction sanity checked. On the second round the actual output
11284 * (the extraction of the hexadecimal values) takes place.
11285 * Sanity failures cause fatal failures during both rounds. */
11287 S_hextract(pTHX_ const NV nv, int* exponent, bool *subnormal,
11288 U8* vhex, U8* vend)
11292 int ixmin = 0, ixmax = 0;
11294 /* XXX Inf/NaN are not handled here, since it is
11295 * assumed they are to be output as "Inf" and "NaN". */
11297 /* These macros are just to reduce typos, they have multiple
11298 * repetitions below, but usually only one (or sometimes two)
11299 * of them is really being used. */
11300 /* HEXTRACT_OUTPUT() extracts the high nybble first. */
11301 #define HEXTRACT_OUTPUT_HI(ix) (*v++ = nvp[ix] >> 4)
11302 #define HEXTRACT_OUTPUT_LO(ix) (*v++ = nvp[ix] & 0xF)
11303 #define HEXTRACT_OUTPUT(ix) \
11305 HEXTRACT_OUTPUT_HI(ix); HEXTRACT_OUTPUT_LO(ix); \
11307 #define HEXTRACT_COUNT(ix, c) \
11309 v += c; if (ix < ixmin) ixmin = ix; else if (ix > ixmax) ixmax = ix; \
11311 #define HEXTRACT_BYTE(ix) \
11313 if (vend) HEXTRACT_OUTPUT(ix); else HEXTRACT_COUNT(ix, 2); \
11315 #define HEXTRACT_LO_NYBBLE(ix) \
11317 if (vend) HEXTRACT_OUTPUT_LO(ix); else HEXTRACT_COUNT(ix, 1); \
11319 /* HEXTRACT_TOP_NYBBLE is just convenience disguise,
11320 * to make it look less odd when the top bits of a NV
11321 * are extracted using HEXTRACT_LO_NYBBLE: the highest
11322 * order bits can be in the "low nybble" of a byte. */
11323 #define HEXTRACT_TOP_NYBBLE(ix) HEXTRACT_LO_NYBBLE(ix)
11324 #define HEXTRACT_BYTES_LE(a, b) \
11325 for (ix = a; ix >= b; ix--) { HEXTRACT_BYTE(ix); }
11326 #define HEXTRACT_BYTES_BE(a, b) \
11327 for (ix = a; ix <= b; ix++) { HEXTRACT_BYTE(ix); }
11328 #define HEXTRACT_GET_SUBNORMAL(nv) *subnormal = Perl_fp_class_denorm(nv)
11329 #define HEXTRACT_IMPLICIT_BIT(nv) \
11331 if (!*subnormal) { \
11332 if (vend) *v++ = ((nv) == 0.0) ? 0 : 1; else v++; \
11336 /* Most formats do. Those which don't should undef this.
11338 * But also note that IEEE 754 subnormals do not have it, or,
11339 * expressed alternatively, their implicit bit is zero. */
11340 #define HEXTRACT_HAS_IMPLICIT_BIT
11342 /* Many formats do. Those which don't should undef this. */
11343 #define HEXTRACT_HAS_TOP_NYBBLE
11345 /* HEXTRACTSIZE is the maximum number of xdigits. */
11346 #if defined(USE_LONG_DOUBLE) && defined(LONGDOUBLE_DOUBLEDOUBLE)
11347 # define HEXTRACTSIZE (2+DOUBLEDOUBLE_MAXBITS/4)
11349 # define HEXTRACTSIZE 2 * NVSIZE
11352 const U8* vmaxend = vhex + HEXTRACTSIZE;
11354 assert(HEXTRACTSIZE <= VHEX_SIZE);
11356 PERL_UNUSED_VAR(ix); /* might happen */
11357 (void)Perl_frexp(PERL_ABS(nv), exponent);
11358 *subnormal = FALSE;
11359 if (vend && (vend <= vhex || vend > vmaxend)) {
11360 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11361 Perl_croak(aTHX_ "Hexadecimal float: internal error (entry)");
11364 /* First check if using long doubles. */
11365 #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE)
11366 # if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN
11367 /* Used in e.g. VMS and HP-UX IA-64, e.g. -0.1L:
11368 * 9a 99 99 99 99 99 99 99 99 99 99 99 99 99 fb bf */
11369 /* The bytes 13..0 are the mantissa/fraction,
11370 * the 15,14 are the sign+exponent. */
11371 const U8* nvp = (const U8*)(&nv);
11372 HEXTRACT_GET_SUBNORMAL(nv);
11373 HEXTRACT_IMPLICIT_BIT(nv);
11374 # undef HEXTRACT_HAS_TOP_NYBBLE
11375 HEXTRACT_BYTES_LE(13, 0);
11376 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_BIG_ENDIAN
11377 /* Used in e.g. Solaris Sparc and HP-UX PA-RISC, e.g. -0.1L:
11378 * bf fb 99 99 99 99 99 99 99 99 99 99 99 99 99 9a */
11379 /* The bytes 2..15 are the mantissa/fraction,
11380 * the 0,1 are the sign+exponent. */
11381 const U8* nvp = (const U8*)(&nv);
11382 HEXTRACT_GET_SUBNORMAL(nv);
11383 HEXTRACT_IMPLICIT_BIT(nv);
11384 # undef HEXTRACT_HAS_TOP_NYBBLE
11385 HEXTRACT_BYTES_BE(2, 15);
11386 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN
11387 /* x86 80-bit "extended precision", 64 bits of mantissa / fraction /
11388 * significand, 15 bits of exponent, 1 bit of sign. No implicit bit.
11389 * NVSIZE can be either 12 (ILP32, Solaris x86) or 16 (LP64, Linux
11390 * and OS X), meaning that 2 or 6 bytes are empty padding. */
11391 /* The bytes 0..1 are the sign+exponent,
11392 * the bytes 2..9 are the mantissa/fraction. */
11393 const U8* nvp = (const U8*)(&nv);
11394 # undef HEXTRACT_HAS_IMPLICIT_BIT
11395 # undef HEXTRACT_HAS_TOP_NYBBLE
11396 HEXTRACT_GET_SUBNORMAL(nv);
11397 HEXTRACT_BYTES_LE(7, 0);
11398 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN
11399 /* Does this format ever happen? (Wikipedia says the Motorola
11400 * 6888x math coprocessors used format _like_ this but padded
11401 * to 96 bits with 16 unused bits between the exponent and the
11403 const U8* nvp = (const U8*)(&nv);
11404 # undef HEXTRACT_HAS_IMPLICIT_BIT
11405 # undef HEXTRACT_HAS_TOP_NYBBLE
11406 HEXTRACT_GET_SUBNORMAL(nv);
11407 HEXTRACT_BYTES_BE(0, 7);
11409 # define HEXTRACT_FALLBACK
11410 /* Double-double format: two doubles next to each other.
11411 * The first double is the high-order one, exactly like
11412 * it would be for a "lone" double. The second double
11413 * is shifted down using the exponent so that that there
11414 * are no common bits. The tricky part is that the value
11415 * of the double-double is the SUM of the two doubles and
11416 * the second one can be also NEGATIVE.
11418 * Because of this tricky construction the bytewise extraction we
11419 * use for the other long double formats doesn't work, we must
11420 * extract the values bit by bit.
11422 * The little-endian double-double is used .. somewhere?
11424 * The big endian double-double is used in e.g. PPC/Power (AIX)
11427 * The mantissa bits are in two separate stretches, e.g. for -0.1L:
11428 * 9a 99 99 99 99 99 59 bc 9a 99 99 99 99 99 b9 3f (LE)
11429 * 3f b9 99 99 99 99 99 9a bc 59 99 99 99 99 99 9a (BE)
11432 #else /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) */
11433 /* Using normal doubles, not long doubles.
11435 * We generate 4-bit xdigits (nybble/nibble) instead of 8-bit
11436 * bytes, since we might need to handle printf precision, and
11437 * also need to insert the radix. */
11439 # ifdef HEXTRACT_LITTLE_ENDIAN
11440 /* 0 1 2 3 4 5 6 7 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11441 const U8* nvp = (const U8*)(&nv);
11442 HEXTRACT_GET_SUBNORMAL(nv);
11443 HEXTRACT_IMPLICIT_BIT(nv);
11444 HEXTRACT_TOP_NYBBLE(6);
11445 HEXTRACT_BYTES_LE(5, 0);
11446 # elif defined(HEXTRACT_BIG_ENDIAN)
11447 /* 7 6 5 4 3 2 1 0 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11448 const U8* nvp = (const U8*)(&nv);
11449 HEXTRACT_GET_SUBNORMAL(nv);
11450 HEXTRACT_IMPLICIT_BIT(nv);
11451 HEXTRACT_TOP_NYBBLE(1);
11452 HEXTRACT_BYTES_BE(2, 7);
11453 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_LE_BE
11454 /* 4 5 6 7 0 1 2 3 (MSB = 7, LSB = 0, 6:7 = nybble:exponent:sign) */
11455 const U8* nvp = (const U8*)(&nv);
11456 HEXTRACT_GET_SUBNORMAL(nv);
11457 HEXTRACT_IMPLICIT_BIT(nv);
11458 HEXTRACT_TOP_NYBBLE(2); /* 6 */
11459 HEXTRACT_BYTE(1); /* 5 */
11460 HEXTRACT_BYTE(0); /* 4 */
11461 HEXTRACT_BYTE(7); /* 3 */
11462 HEXTRACT_BYTE(6); /* 2 */
11463 HEXTRACT_BYTE(5); /* 1 */
11464 HEXTRACT_BYTE(4); /* 0 */
11465 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_BE_LE
11466 /* 3 2 1 0 7 6 5 4 (MSB = 7, LSB = 0, 7:6 = sign:exponent:nybble) */
11467 const U8* nvp = (const U8*)(&nv);
11468 HEXTRACT_GET_SUBNORMAL(nv);
11469 HEXTRACT_IMPLICIT_BIT(nv);
11470 HEXTRACT_TOP_NYBBLE(5); /* 6 */
11471 HEXTRACT_BYTE(6); /* 5 */
11472 HEXTRACT_BYTE(7); /* 4 */
11473 HEXTRACT_BYTE(0); /* 3 */
11474 HEXTRACT_BYTE(1); /* 2 */
11475 HEXTRACT_BYTE(2); /* 1 */
11476 HEXTRACT_BYTE(3); /* 0 */
11478 # define HEXTRACT_FALLBACK
11481 # define HEXTRACT_FALLBACK
11483 #endif /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) #else */
11485 #ifdef HEXTRACT_FALLBACK
11486 HEXTRACT_GET_SUBNORMAL(nv);
11487 # undef HEXTRACT_HAS_TOP_NYBBLE /* Meaningless, but consistent. */
11488 /* The fallback is used for the double-double format, and
11489 * for unknown long double formats, and for unknown double
11490 * formats, or in general unknown NV formats. */
11491 if (nv == (NV)0.0) {
11499 NV d = nv < 0 ? -nv : nv;
11501 U8 ha = 0x0; /* hexvalue accumulator */
11502 U8 hd = 0x8; /* hexvalue digit */
11504 /* Shift d and e (and update exponent) so that e <= d < 2*e,
11505 * this is essentially manual frexp(). Multiplying by 0.5 and
11506 * doubling should be lossless in binary floating point. */
11516 while (d >= e + e) {
11520 /* Now e <= d < 2*e */
11522 /* First extract the leading hexdigit (the implicit bit). */
11538 /* Then extract the remaining hexdigits. */
11539 while (d > (NV)0.0) {
11545 /* Output or count in groups of four bits,
11546 * that is, when the hexdigit is down to one. */
11551 /* Reset the hexvalue. */
11560 /* Flush possible pending hexvalue. */
11570 /* Croak for various reasons: if the output pointer escaped the
11571 * output buffer, if the extraction index escaped the extraction
11572 * buffer, or if the ending output pointer didn't match the
11573 * previously computed value. */
11574 if (v <= vhex || v - vhex >= VHEX_SIZE ||
11575 /* For double-double the ixmin and ixmax stay at zero,
11576 * which is convenient since the HEXTRACTSIZE is tricky
11577 * for double-double. */
11578 ixmin < 0 || ixmax >= NVSIZE ||
11579 (vend && v != vend)) {
11580 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11581 Perl_croak(aTHX_ "Hexadecimal float: internal error (overflow)");
11587 /* S_format_hexfp(): helper function for Perl_sv_vcatpvfn_flags().
11589 * Processes the %a/%A hexadecimal floating-point format, since the
11590 * built-in snprintf()s which are used for most of the f/p formats, don't
11591 * universally handle %a/%A.
11592 * Populates buf of length bufsize, and returns the length of the created
11594 * The rest of the args have the same meaning as the local vars of the
11595 * same name within Perl_sv_vcatpvfn_flags().
11597 * It assumes the caller has already done STORE_LC_NUMERIC_SET_TO_NEEDED();
11599 * It requires the caller to make buf large enough.
11603 S_format_hexfp(pTHX_ char * const buf, const STRLEN bufsize, const char c,
11604 const NV nv, const vcatpvfn_long_double_t fv,
11605 bool has_precis, STRLEN precis, STRLEN width,
11606 bool alt, char plus, bool left, bool fill)
11608 /* Hexadecimal floating point. */
11610 U8 vhex[VHEX_SIZE];
11611 U8* v = vhex; /* working pointer to vhex */
11612 U8* vend; /* pointer to one beyond last digit of vhex */
11613 U8* vfnz = NULL; /* first non-zero */
11614 U8* vlnz = NULL; /* last non-zero */
11615 U8* v0 = NULL; /* first output */
11616 const bool lower = (c == 'a');
11617 /* At output the values of vhex (up to vend) will
11618 * be mapped through the xdig to get the actual
11619 * human-readable xdigits. */
11620 const char* xdig = PL_hexdigit;
11621 STRLEN zerotail = 0; /* how many extra zeros to append */
11622 int exponent = 0; /* exponent of the floating point input */
11623 bool hexradix = FALSE; /* should we output the radix */
11624 bool subnormal = FALSE; /* IEEE 754 subnormal/denormal */
11625 bool negative = FALSE;
11628 /* XXX: NaN, Inf -- though they are printed as "NaN" and "Inf".
11630 * For example with denormals, (assuming the vanilla
11631 * 64-bit double): the exponent is zero. 1xp-1074 is
11632 * the smallest denormal and the smallest double, it
11633 * could be output also as 0x0.0000000000001p-1022 to
11634 * match its internal structure. */
11636 vend = S_hextract(aTHX_ nv, &exponent, &subnormal, vhex, NULL);
11637 S_hextract(aTHX_ nv, &exponent, &subnormal, vhex, vend);
11639 #if NVSIZE > DOUBLESIZE
11640 # ifdef HEXTRACT_HAS_IMPLICIT_BIT
11641 /* In this case there is an implicit bit,
11642 * and therefore the exponent is shifted by one. */
11644 # elif defined(NV_X86_80_BIT)
11646 /* The subnormals of the x86-80 have a base exponent of -16382,
11647 * (while the physical exponent bits are zero) but the frexp()
11648 * returned the scientific-style floating exponent. We want
11649 * to map the last one as:
11650 * -16831..-16384 -> -16382 (the last normal is 0x1p-16382)
11651 * -16835..-16388 -> -16384
11652 * since we want to keep the first hexdigit
11653 * as one of the [8421]. */
11654 exponent = -4 * ( (exponent + 1) / -4) - 2;
11658 /* TBD: other non-implicit-bit platforms than the x86-80. */
11662 negative = fv < 0 || Perl_signbit(nv);
11673 xdig += 16; /* Use uppercase hex. */
11676 /* Find the first non-zero xdigit. */
11677 for (v = vhex; v < vend; v++) {
11685 /* Find the last non-zero xdigit. */
11686 for (v = vend - 1; v >= vhex; v--) {
11693 #if NVSIZE == DOUBLESIZE
11699 #ifndef NV_X86_80_BIT
11701 /* IEEE 754 subnormals (but not the x86 80-bit):
11702 * we want "normalize" the subnormal,
11703 * so we need to right shift the hex nybbles
11704 * so that the output of the subnormal starts
11705 * from the first true bit. (Another, equally
11706 * valid, policy would be to dump the subnormal
11707 * nybbles as-is, to display the "physical" layout.) */
11710 /* Find the ceil(log2(v[0])) of
11711 * the top non-zero nybble. */
11712 for (i = vfnz[0], n = 0; i > 1; i >>= 1, n++) { }
11715 for (vshr = vlnz; vshr >= vfnz; vshr--) {
11716 vshr[1] |= (vshr[0] & (0xF >> (4 - n))) << (4 - n);
11730 U8* ve = (subnormal ? vlnz + 1 : vend);
11731 SSize_t vn = ve - v0;
11733 if (precis < (Size_t)(vn - 1)) {
11734 bool overflow = FALSE;
11735 if (v0[precis + 1] < 0x8) {
11736 /* Round down, nothing to do. */
11737 } else if (v0[precis + 1] > 0x8) {
11740 overflow = v0[precis] > 0xF;
11742 } else { /* v0[precis] == 0x8 */
11743 /* Half-point: round towards the one
11744 * with the even least-significant digit:
11752 * 78 -> 8 f8 -> 10 */
11753 if ((v0[precis] & 0x1)) {
11756 overflow = v0[precis] > 0xF;
11761 for (v = v0 + precis - 1; v >= v0; v--) {
11763 overflow = *v > 0xF;
11769 if (v == v0 - 1 && overflow) {
11770 /* If the overflow goes all the
11771 * way to the front, we need to
11772 * insert 0x1 in front, and adjust
11774 Move(v0, v0 + 1, vn - 1, char);
11780 /* The new effective "last non zero". */
11781 vlnz = v0 + precis;
11785 subnormal ? precis - vn + 1 :
11786 precis - (vlnz - vhex);
11793 /* If there are non-zero xdigits, the radix
11794 * is output after the first one. */
11805 /* The radix is always output if precis, or if alt. */
11806 if (precis > 0 || alt) {
11811 #ifndef USE_LOCALE_NUMERIC
11814 if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC)) {
11816 const char* r = SvPV(PL_numeric_radix_sv, n);
11817 Copy(r, p, n, char);
11831 if (zerotail > 0) {
11832 while (zerotail--) {
11839 /* sanity checks */
11840 if (elen >= bufsize || width >= bufsize)
11841 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11842 Perl_croak(aTHX_ "Hexadecimal float: internal error (overflow)");
11844 elen += my_snprintf(p, bufsize - elen,
11845 "%c%+d", lower ? 'p' : 'P',
11848 if (elen < width) {
11849 STRLEN gap = (STRLEN)(width - elen);
11851 /* Pad the back with spaces. */
11852 memset(buf + elen, ' ', gap);
11855 /* Insert the zeros after the "0x" and the
11856 * the potential sign, but before the digits,
11857 * otherwise we end up with "0000xH.HHH...",
11858 * when we want "0x000H.HHH..." */
11859 STRLEN nzero = gap;
11860 char* zerox = buf + 2;
11861 STRLEN nmove = elen - 2;
11862 if (negative || plus) {
11866 Move(zerox, zerox + nzero, nmove, char);
11867 memset(zerox, fill ? '0' : ' ', nzero);
11870 /* Move it to the right. */
11871 Move(buf, buf + gap,
11873 /* Pad the front with spaces. */
11874 memset(buf, ' ', gap);
11883 =for apidoc sv_vcatpvfn
11885 =for apidoc sv_vcatpvfn_flags
11887 Processes its arguments like C<vsprintf> and appends the formatted output
11888 to an SV. Uses an array of SVs if the C-style variable argument list is
11889 missing (C<NULL>). Argument reordering (using format specifiers like C<%2$d>
11890 or C<%*2$d>) is supported only when using an array of SVs; using a C-style
11891 C<va_list> argument list with a format string that uses argument reordering
11892 will yield an exception.
11894 When running with taint checks enabled, indicates via
11895 C<maybe_tainted> if results are untrustworthy (often due to the use of
11898 If called as C<sv_vcatpvfn> or flags has the C<SV_GMAGIC> bit set, calls get magic.
11900 It assumes that pat has the same utf8-ness as sv. It's the caller's
11901 responsibility to ensure that this is so.
11903 Usually used via one of its frontends C<sv_vcatpvf> and C<sv_vcatpvf_mg>.
11910 Perl_sv_vcatpvfn_flags(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11911 va_list *const args, SV **const svargs, const Size_t sv_count, bool *const maybe_tainted,
11914 const char *fmtstart; /* character following the current '%' */
11915 const char *q; /* current position within format */
11916 const char *patend;
11919 static const char nullstr[] = "(null)";
11921 bool has_utf8 = DO_UTF8(sv); /* has the result utf8? */
11922 const bool pat_utf8 = has_utf8; /* the pattern is in utf8? */
11923 /* Times 4: a decimal digit takes more than 3 binary digits.
11924 * NV_DIG: mantissa takes that many decimal digits.
11925 * Plus 32: Playing safe. */
11926 char ebuf[IV_DIG * 4 + NV_DIG + 32];
11927 bool no_redundant_warning = FALSE; /* did we use any explicit format parameter index? */
11928 #ifdef USE_LOCALE_NUMERIC
11929 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
11930 bool lc_numeric_set = FALSE; /* called STORE_LC_NUMERIC_SET_TO_NEEDED? */
11933 PERL_ARGS_ASSERT_SV_VCATPVFN_FLAGS;
11934 PERL_UNUSED_ARG(maybe_tainted);
11936 if (flags & SV_GMAGIC)
11939 /* no matter what, this is a string now */
11940 (void)SvPV_force_nomg(sv, origlen);
11942 /* the code that scans for flags etc following a % relies on
11943 * a '\0' being present to avoid falling off the end. Ideally that
11944 * should be fixed */
11945 assert(pat[patlen] == '\0');
11948 /* Special-case "", "%s", "%-p" (SVf - see below) and "%.0f".
11949 * In each case, if there isn't the correct number of args, instead
11950 * fall through to the main code to handle the issuing of any
11954 if (patlen == 0 && (args || sv_count == 0))
11957 if (patlen <= 4 && pat[0] == '%' && (args || sv_count == 1)) {
11960 if (patlen == 2 && pat[1] == 's') {
11962 const char * const s = va_arg(*args, char*);
11963 sv_catpv_nomg(sv, s ? s : nullstr);
11966 /* we want get magic on the source but not the target.
11967 * sv_catsv can't do that, though */
11968 SvGETMAGIC(*svargs);
11969 sv_catsv_nomg(sv, *svargs);
11976 if (patlen == 3 && pat[1] == '-' && pat[2] == 'p') {
11977 SV *asv = MUTABLE_SV(va_arg(*args, void*));
11978 sv_catsv_nomg(sv, asv);
11982 #if !defined(USE_LONG_DOUBLE) && !defined(USE_QUADMATH)
11983 /* special-case "%.0f" */
11984 else if ( patlen == 4
11985 && pat[1] == '.' && pat[2] == '0' && pat[3] == 'f')
11987 const NV nv = SvNV(*svargs);
11988 if (LIKELY(!Perl_isinfnan(nv))) {
11992 if ((p = F0convert(nv, ebuf + sizeof ebuf, &l))) {
11993 sv_catpvn_nomg(sv, p, l);
11998 #endif /* !USE_LONG_DOUBLE */
12002 patend = (char*)pat + patlen;
12003 for (fmtstart = pat; fmtstart < patend; fmtstart = q) {
12004 char intsize = 0; /* size qualifier in "%hi..." etc */
12005 bool alt = FALSE; /* has "%#..." */
12006 bool left = FALSE; /* has "%-..." */
12007 bool fill = FALSE; /* has "%0..." */
12008 char plus = 0; /* has "%+..." */
12009 STRLEN width = 0; /* value of "%NNN..." */
12010 bool has_precis = FALSE; /* has "%.NNN..." */
12011 STRLEN precis = 0; /* value of "%.NNN..." */
12012 int base = 0; /* base to print in, e.g. 8 for %o */
12013 UV uv = 0; /* the value to print of int-ish args */
12015 bool vectorize = FALSE; /* has "%v..." */
12016 bool vec_utf8 = FALSE; /* SvUTF8(vec arg) */
12017 const U8 *vecstr = NULL; /* SvPVX(vec arg) */
12018 STRLEN veclen = 0; /* SvCUR(vec arg) */
12019 const char *dotstr = NULL; /* separator string for %v */
12020 STRLEN dotstrlen; /* length of separator string for %v */
12022 Size_t efix = 0; /* explicit format parameter index */
12023 const Size_t osvix = svix; /* original index in case of bad fmt */
12025 bool is_utf8 = FALSE; /* is this item utf8? */
12026 bool arg_missing = FALSE; /* give "Missing argument" warning */
12027 char esignbuf[4]; /* holds sign prefix, e.g. "-0x" */
12028 STRLEN esignlen = 0; /* length of e.g. "-0x" */
12029 STRLEN zeros = 0; /* how many '0' to prepend */
12031 const char *eptr = NULL; /* the address of the element string */
12032 STRLEN elen = 0; /* the length of the element string */
12034 char c; /* the actual format ('d', s' etc) */
12037 /* echo everything up to the next format specification */
12038 for (q = fmtstart; q < patend && *q != '%'; ++q)
12041 if (q > fmtstart) {
12042 if (has_utf8 && !pat_utf8) {
12043 /* upgrade and copy the bytes of fmtstart..q-1 to utf8 on
12047 STRLEN need = SvCUR(sv) + (q - fmtstart) + 1;
12049 for (p = fmtstart; p < q; p++)
12050 if (!NATIVE_BYTE_IS_INVARIANT(*p))
12055 for (p = fmtstart; p < q; p++)
12056 append_utf8_from_native_byte((U8)*p, (U8**)&dst);
12058 SvCUR_set(sv, need - 1);
12061 S_sv_catpvn_simple(aTHX_ sv, fmtstart, q - fmtstart);
12066 fmtstart = q; /* fmtstart is char following the '%' */
12069 We allow format specification elements in this order:
12070 \d+\$ explicit format parameter index
12072 v|\*(\d+\$)?v vector with optional (optionally specified) arg
12073 0 flag (as above): repeated to allow "v02"
12074 \d+|\*(\d+\$)? width using optional (optionally specified) arg
12075 \.(\d*|\*(\d+\$)?) precision using optional (optionally specified) arg
12077 [%bcdefginopsuxDFOUX] format (mandatory)
12080 if (IS_1_TO_9(*q)) {
12081 width = expect_number(&q);
12084 Perl_croak_nocontext(
12085 "Cannot yet reorder sv_catpvfn() arguments from va_list");
12087 efix = (Size_t)width;
12089 no_redundant_warning = TRUE;
12101 if (plus == '+' && *q == ' ') /* '+' over ' ' */
12128 /* at this point we can expect one of:
12130 * 123 an explicit width
12131 * * width taken from next arg
12132 * *12$ width taken from 12th arg
12135 * But any width specification may be preceded by a v, in one of its
12140 * So an asterisk may be either a width specifier or a vector
12141 * separator arg specifier, and we don't know which initially
12146 STRLEN ix; /* explicit width/vector separator index */
12148 if (IS_1_TO_9(*q)) {
12149 ix = expect_number(&q);
12152 Perl_croak_nocontext(
12153 "Cannot yet reorder sv_catpvfn() arguments from va_list");
12154 no_redundant_warning = TRUE;
12163 /* The asterisk was for *v, *NNN$v: vectorizing, but not
12164 * with the default "." */
12169 vecsv = va_arg(*args, SV*);
12171 ix = ix ? ix - 1 : svix++;
12172 vecsv = ix < sv_count ? svargs[ix]
12173 : (arg_missing = TRUE, &PL_sv_no);
12175 dotstr = SvPV_const(vecsv, dotstrlen);
12176 /* Keep the DO_UTF8 test *after* the SvPV call, else things go
12177 bad with tied or overloaded values that return UTF8. */
12178 if (DO_UTF8(vecsv))
12180 else if (has_utf8) {
12181 vecsv = sv_mortalcopy(vecsv);
12182 sv_utf8_upgrade(vecsv);
12183 dotstr = SvPV_const(vecsv, dotstrlen);
12190 /* the asterisk specified a width */
12195 i = va_arg(*args, int);
12197 ix = ix ? ix - 1 : svix++;
12198 sv = (ix < sv_count) ? svargs[ix]
12199 : (arg_missing = TRUE, (SV*)NULL);
12201 width = S_sprintf_arg_num_val(aTHX_ args, i, sv, &left);
12204 else if (*q == 'v') {
12215 /* explicit width? */
12221 width = expect_number(&q);
12231 STRLEN ix; /* explicit precision index */
12233 if (IS_1_TO_9(*q)) {
12234 ix = expect_number(&q);
12237 Perl_croak_nocontext(
12238 "Cannot yet reorder sv_catpvfn() arguments from va_list");
12239 no_redundant_warning = TRUE;
12252 i = va_arg(*args, int);
12254 ix = ix ? ix - 1 : svix++;
12255 sv = (ix < sv_count) ? svargs[ix]
12256 : (arg_missing = TRUE, (SV*)NULL);
12258 precis = S_sprintf_arg_num_val(aTHX_ args, i, sv, &neg);
12263 /* although it doesn't seem documented, this code has long
12265 * no digits following the '.' is treated like '.0'
12266 * the number may be preceded by any number of zeroes,
12267 * e.g. "%.0001f", which is the same as "%.1f"
12268 * so I've kept that behaviour. DAPM May 2017
12272 precis = IS_1_TO_9(*q) ? expect_number(&q) : 0;
12281 case 'I': /* Ix, I32x, and I64x */
12282 # ifdef USE_64_BIT_INT
12283 if (q[1] == '6' && q[2] == '4') {
12289 if (q[1] == '3' && q[2] == '2') {
12293 # ifdef USE_64_BIT_INT
12299 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
12300 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
12303 # ifdef USE_QUADMATH
12316 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
12317 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
12318 if (*q == 'l') { /* lld, llf */
12327 if (*++q == 'h') { /* hhd, hhu */
12346 c = *q++; /* c now holds the conversion type */
12348 /* '%' doesn't have an arg, so skip arg processing */
12357 if (vectorize && !strchr("BbDdiOouUXx", c))
12360 /* get next arg (individual branches do their own va_arg()
12361 * handling for the args case) */
12364 efix = efix ? efix - 1 : svix++;
12365 argsv = efix < sv_count ? svargs[efix]
12366 : (arg_missing = TRUE, &PL_sv_no);
12376 eptr = va_arg(*args, char*);
12379 elen = my_strnlen(eptr, precis);
12381 elen = strlen(eptr);
12383 eptr = (char *)nullstr;
12384 elen = sizeof nullstr - 1;
12388 eptr = SvPV_const(argsv, elen);
12389 if (DO_UTF8(argsv)) {
12390 STRLEN old_precis = precis;
12391 if (has_precis && precis < elen) {
12392 STRLEN ulen = sv_or_pv_len_utf8(argsv, eptr, elen);
12393 STRLEN p = precis > ulen ? ulen : precis;
12394 precis = sv_or_pv_pos_u2b(argsv, eptr, p, 0);
12395 /* sticks at end */
12397 if (width) { /* fudge width (can't fudge elen) */
12398 if (has_precis && precis < elen)
12399 width += precis - old_precis;
12402 elen - sv_or_pv_len_utf8(argsv,eptr,elen);
12409 if (has_precis && precis < elen)
12421 * "%...p" is normally treated like "%...x", except that the
12422 * number to print is the SV's address (or a pointer address
12423 * for C-ish sprintf).
12425 * However, the C-ish sprintf variant allows a few special
12426 * extensions. These are currently:
12428 * %-p (SVf) Like %s, but gets the string from an SV*
12429 * arg rather than a char* arg.
12430 * (This was previously %_).
12432 * %-<num>p Ditto but like %.<num>s (i.e. num is max width)
12434 * %2p (HEKf) Like %s, but using the key string in a HEK
12436 * %3p (HEKf256) Ditto but like %.256s
12438 * %d%lu%4p (UTF8f) A utf8 string. Consumes 3 args:
12439 * (cBOOL(utf8), len, string_buf).
12440 * It's handled by the "case 'd'" branch
12441 * rather than here.
12443 * %<num>p where num is 1 or > 4: reserved for future
12444 * extensions. Warns, but then is treated as a
12445 * general %p (print hex address) format.
12453 /* not %*p or %*1$p - any width was explicit */
12457 if (left) { /* %-p (SVf), %-NNNp */
12462 argsv = MUTABLE_SV(va_arg(*args, void*));
12463 eptr = SvPV_const(argsv, elen);
12464 if (DO_UTF8(argsv))
12469 else if (width == 2 || width == 3) { /* HEKf, HEKf256 */
12470 HEK * const hek = va_arg(*args, HEK *);
12471 eptr = HEK_KEY(hek);
12472 elen = HEK_LEN(hek);
12483 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
12484 "internal %%<num>p might conflict with future printf extensions");
12488 /* treat as normal %...p */
12490 uv = PTR2UV(args ? va_arg(*args, void*) : argsv);
12495 /* Ignore any size specifiers, since they're not documented as
12496 * being allowed for %c (ideally we should warn on e.g. '%hc').
12497 * Setting a default intsize, along with a positive
12498 * (which signals unsigned) base, causes, for C-ish use, the
12499 * va_arg to be interpreted as as unsigned int, when it's
12500 * actually signed, which will convert -ve values to high +ve
12501 * values. Note that unlike the libc %c, values > 255 will
12502 * convert to high unicode points rather than being truncated
12503 * to 8 bits. For perlish use, it will do SvUV(argsv), which
12504 * will again convert -ve args to high -ve values.
12507 base = 1; /* special value that indicates we're doing a 'c' */
12508 goto get_int_arg_val;
12517 goto get_int_arg_val;
12520 /* probably just a plain %d, but it might be the start of the
12521 * special UTF8f format, which usually looks something like
12522 * "%d%lu%4p" (the lu may vary by platform)
12524 assert((UTF8f)[0] == 'd');
12525 assert((UTF8f)[1] == '%');
12527 if ( args /* UTF8f only valid for C-ish sprintf */
12528 && q == fmtstart + 1 /* plain %d, not %....d */
12529 && patend >= fmtstart + sizeof(UTF8f) - 1 /* long enough */
12531 && strnEQ(q + 1, UTF8f + 2, sizeof(UTF8f) - 3))
12533 /* The argument has already gone through cBOOL, so the cast
12535 is_utf8 = (bool)va_arg(*args, int);
12536 elen = va_arg(*args, UV);
12537 /* if utf8 length is larger than 0x7ffff..., then it might
12538 * have been a signed value that wrapped */
12539 if (elen > ((~(STRLEN)0) >> 1)) {
12540 assert(0); /* in DEBUGGING build we want to crash */
12541 elen = 0; /* otherwise we want to treat this as an empty string */
12543 eptr = va_arg(*args, char *);
12544 q += sizeof(UTF8f) - 2;
12551 goto get_int_arg_val;
12562 goto get_int_arg_val;
12567 goto get_int_arg_val;
12578 goto get_int_arg_val;
12593 esignbuf[esignlen++] = plus;
12596 /* initialise the vector string to iterate over */
12598 vecsv = args ? va_arg(*args, SV*) : argsv;
12600 /* if this is a version object, we need to convert
12601 * back into v-string notation and then let the
12602 * vectorize happen normally
12604 if (sv_isobject(vecsv) && sv_derived_from(vecsv, "version")) {
12605 if ( hv_existss(MUTABLE_HV(SvRV(vecsv)), "alpha") ) {
12606 Perl_ck_warner_d(aTHX_ packWARN(WARN_PRINTF),
12607 "vector argument not supported with alpha versions");
12611 vecstr = (U8*)SvPV_const(vecsv,veclen);
12612 vecsv = sv_newmortal();
12613 scan_vstring((char *)vecstr, (char *)vecstr + veclen,
12617 vecstr = (U8*)SvPV_const(vecsv, veclen);
12618 vec_utf8 = DO_UTF8(vecsv);
12620 /* This is the re-entry point for when we're iterating
12621 * over the individual characters of a vector arg */
12624 goto done_valid_conversion;
12626 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
12636 /* test arg for inf/nan. This can trigger an unwanted
12637 * 'str' overload, so manually force 'num' overload first
12641 if (UNLIKELY(SvAMAGIC(argsv)))
12642 argsv = sv_2num(argsv);
12643 if (UNLIKELY(isinfnansv(argsv)))
12644 goto handle_infnan_argsv;
12648 /* signed int type */
12653 case 'c': iv = (char)va_arg(*args, int); break;
12654 case 'h': iv = (short)va_arg(*args, int); break;
12655 case 'l': iv = va_arg(*args, long); break;
12656 case 'V': iv = va_arg(*args, IV); break;
12657 case 'z': iv = va_arg(*args, SSize_t); break;
12658 #ifdef HAS_PTRDIFF_T
12659 case 't': iv = va_arg(*args, ptrdiff_t); break;
12661 default: iv = va_arg(*args, int); break;
12663 case 'j': iv = va_arg(*args, intmax_t); break;
12667 iv = va_arg(*args, Quad_t); break;
12674 /* assign to tiv then cast to iv to work around
12675 * 2003 GCC cast bug (gnu.org bugzilla #13488) */
12676 IV tiv = SvIV_nomg(argsv);
12678 case 'c': iv = (char)tiv; break;
12679 case 'h': iv = (short)tiv; break;
12680 case 'l': iv = (long)tiv; break;
12682 default: iv = tiv; break;
12685 iv = (Quad_t)tiv; break;
12692 /* now convert iv to uv */
12696 esignbuf[esignlen++] = plus;
12699 uv = (iv == IV_MIN) ? (UV)iv : (UV)(-iv);
12700 esignbuf[esignlen++] = '-';
12704 /* unsigned int type */
12707 case 'c': uv = (unsigned char)va_arg(*args, unsigned);
12709 case 'h': uv = (unsigned short)va_arg(*args, unsigned);
12711 case 'l': uv = va_arg(*args, unsigned long); break;
12712 case 'V': uv = va_arg(*args, UV); break;
12713 case 'z': uv = va_arg(*args, Size_t); break;
12714 #ifdef HAS_PTRDIFF_T
12715 /* will sign extend, but there is no
12716 * uptrdiff_t, so oh well */
12717 case 't': uv = va_arg(*args, ptrdiff_t); break;
12720 case 'j': uv = va_arg(*args, uintmax_t); break;
12722 default: uv = va_arg(*args, unsigned); break;
12725 uv = va_arg(*args, Uquad_t); break;
12732 /* assign to tiv then cast to iv to work around
12733 * 2003 GCC cast bug (gnu.org bugzilla #13488) */
12734 UV tuv = SvUV_nomg(argsv);
12736 case 'c': uv = (unsigned char)tuv; break;
12737 case 'h': uv = (unsigned short)tuv; break;
12738 case 'l': uv = (unsigned long)tuv; break;
12740 default: uv = tuv; break;
12743 uv = (Uquad_t)tuv; break;
12754 char *ptr = ebuf + sizeof ebuf;
12761 const char * const p =
12762 (c == 'X') ? PL_hexdigit + 16 : PL_hexdigit;
12767 } while (uv >>= 4);
12768 if (alt && *ptr != '0') {
12769 esignbuf[esignlen++] = '0';
12770 esignbuf[esignlen++] = c; /* 'x' or 'X' */
12777 *--ptr = '0' + dig;
12778 } while (uv >>= 3);
12779 if (alt && *ptr != '0')
12785 *--ptr = '0' + dig;
12786 } while (uv >>= 1);
12787 if (alt && *ptr != '0') {
12788 esignbuf[esignlen++] = '0';
12789 esignbuf[esignlen++] = c; /* 'b' or 'B' */
12794 /* special-case: base 1 indicates a 'c' format:
12795 * we use the common code for extracting a uv,
12796 * but handle that value differently here than
12797 * all the other int types */
12799 (!UVCHR_IS_INVARIANT(uv) && SvUTF8(sv)))
12802 assert(sizeof(ebuf) >= UTF8_MAXBYTES + 1);
12804 elen = uvchr_to_utf8((U8*)eptr, uv) - (U8*)ebuf;
12809 ebuf[0] = (char)uv;
12814 default: /* it had better be ten or less */
12817 *--ptr = '0' + dig;
12818 } while (uv /= base);
12821 elen = (ebuf + sizeof ebuf) - ptr;
12825 zeros = precis - elen;
12826 else if (precis == 0 && elen == 1 && *eptr == '0'
12827 && !(base == 8 && alt)) /* "%#.0o" prints "0" */
12830 /* a precision nullifies the 0 flag. */
12836 /* FLOATING POINT */
12839 c = 'f'; /* maybe %F isn't supported here */
12841 case 'e': case 'E':
12843 case 'g': case 'G':
12844 case 'a': case 'A':
12847 STRLEN float_need; /* what PL_efloatsize needs to become */
12848 bool hexfp; /* hexadecimal floating point? */
12850 vcatpvfn_long_double_t fv;
12853 /* This is evil, but floating point is even more evil */
12855 /* for SV-style calling, we can only get NV
12856 for C-style calling, we assume %f is double;
12857 for simplicity we allow any of %Lf, %llf, %qf for long double
12861 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12865 /* [perl #20339] - we should accept and ignore %lf rather than die */
12869 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12870 intsize = args ? 0 : 'q';
12874 #if defined(HAS_LONG_DOUBLE)
12887 /* Now we need (long double) if intsize == 'q', else (double). */
12889 /* Note: do not pull NVs off the va_list with va_arg()
12890 * (pull doubles instead) because if you have a build
12891 * with long doubles, you would always be pulling long
12892 * doubles, which would badly break anyone using only
12893 * doubles (i.e. the majority of builds). In other
12894 * words, you cannot mix doubles and long doubles.
12895 * The only case where you can pull off long doubles
12896 * is when the format specifier explicitly asks so with
12898 #ifdef USE_QUADMATH
12899 fv = intsize == 'q' ?
12900 va_arg(*args, NV) : va_arg(*args, double);
12902 #elif LONG_DOUBLESIZE > DOUBLESIZE
12903 if (intsize == 'q') {
12904 fv = va_arg(*args, long double);
12907 nv = va_arg(*args, double);
12908 VCATPVFN_NV_TO_FV(nv, fv);
12911 nv = va_arg(*args, double);
12918 /* we jump here if an int-ish format encountered an
12919 * infinite/Nan argsv. After setting nv/fv, it falls
12920 * into the isinfnan block which follows */
12921 handle_infnan_argsv:
12922 nv = SvNV_nomg(argsv);
12923 VCATPVFN_NV_TO_FV(nv, fv);
12926 if (Perl_isinfnan(nv)) {
12928 Perl_croak(aTHX_ "Cannot printf %" NVgf " with '%c'",
12929 SvNV_nomg(argsv), (int)c);
12931 elen = S_infnan_2pv(nv, ebuf, sizeof(ebuf), plus);
12940 /* special-case "%.0f" */
12944 && !(width || left || plus || alt)
12947 && ((eptr = F0convert(nv, ebuf + sizeof ebuf, &elen)))
12951 /* Determine the buffer size needed for the various
12952 * floating-point formats.
12954 * The basic possibilities are:
12957 * %f 1111111.123456789
12958 * %e 1.111111123e+06
12959 * %a 0x1.0f4471f9bp+20
12961 * %g 1.11111112e+15
12963 * where P is the value of the precision in the format, or 6
12964 * if not specified. Note the two possible output formats of
12965 * %g; in both cases the number of significant digits is <=
12968 * For most of the format types the maximum buffer size needed
12969 * is precision, plus: any leading 1 or 0x1, the radix
12970 * point, and an exponent. The difficult one is %f: for a
12971 * large positive exponent it can have many leading digits,
12972 * which needs to be calculated specially. Also %a is slightly
12973 * different in that in the absence of a specified precision,
12974 * it uses as many digits as necessary to distinguish
12975 * different values.
12977 * First, here are the constant bits. For ease of calculation
12978 * we over-estimate the needed buffer size, for example by
12979 * assuming all formats have an exponent and a leading 0x1.
12981 * Also for production use, add a little extra overhead for
12982 * safety's sake. Under debugging don't, as it means we're
12983 * more likely to quickly spot issues during development.
12986 float_need = 1 /* possible unary minus */
12987 + 4 /* "0x1" plus very unlikely carry */
12988 + 1 /* default radix point '.' */
12989 + 2 /* "e-", "p+" etc */
12990 + 6 /* exponent: up to 16383 (quad fp) */
12992 + 20 /* safety net */
12997 /* determine the radix point len, e.g. length(".") in "1.2" */
12998 #ifdef USE_LOCALE_NUMERIC
12999 /* note that we may either explicitly use PL_numeric_radix_sv
13000 * below, or implicitly, via an snprintf() variant.
13001 * Note also things like ps_AF.utf8 which has
13002 * "\N{ARABIC DECIMAL SEPARATOR} as a radix point */
13003 if (!lc_numeric_set) {
13004 /* only set once and reuse in-locale value on subsequent
13006 * XXX what happens if we die in an eval?
13008 STORE_LC_NUMERIC_SET_TO_NEEDED();
13009 lc_numeric_set = TRUE;
13012 if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC)) {
13013 /* this can't wrap unless PL_numeric_radix_sv is a string
13014 * consuming virtually all the 32-bit or 64-bit address
13017 float_need += (SvCUR(PL_numeric_radix_sv) - 1);
13019 /* floating-point formats only get utf8 if the radix point
13020 * is utf8. All other characters in the string are < 128
13021 * and so can be safely appended to both a non-utf8 and utf8
13023 * Note that this will convert the output to utf8 even if
13024 * the radix point didn't get output.
13026 if (SvUTF8(PL_numeric_radix_sv) && !has_utf8) {
13027 sv_utf8_upgrade(sv);
13035 if (isALPHA_FOLD_EQ(c, 'f')) {
13036 /* Determine how many digits before the radix point
13037 * might be emitted. frexp() (or frexpl) has some
13038 * unspecified behaviour for nan/inf/-inf, so lucky we've
13039 * already handled them above */
13041 int i = PERL_INT_MIN;
13042 (void)Perl_frexp((NV)fv, &i);
13043 if (i == PERL_INT_MIN)
13044 Perl_die(aTHX_ "panic: frexp: %" VCATPVFN_FV_GF, fv);
13047 digits = BIT_DIGITS(i);
13048 /* this can't overflow. 'digits' will only be a few
13049 * thousand even for the largest floating-point types.
13050 * And up until now float_need is just some small
13051 * constants plus radix len, which can't be in
13052 * overflow territory unless the radix SV is consuming
13053 * over 1/2 the address space */
13054 assert(float_need < ((STRLEN)~0) - digits);
13055 float_need += digits;
13058 else if (UNLIKELY(isALPHA_FOLD_EQ(c, 'a'))) {
13061 /* %a in the absence of precision may print as many
13062 * digits as needed to represent the entire mantissa
13064 * This estimate seriously overshoots in most cases,
13065 * but better the undershooting. Firstly, all bytes
13066 * of the NV are not mantissa, some of them are
13067 * exponent. Secondly, for the reasonably common
13068 * long doubles case, the "80-bit extended", two
13069 * or six bytes of the NV are unused. Also, we'll
13070 * still pick up an extra +6 from the default
13071 * precision calculation below. */
13073 #ifdef LONGDOUBLE_DOUBLEDOUBLE
13074 /* For the "double double", we need more.
13075 * Since each double has their own exponent, the
13076 * doubles may float (haha) rather far from each
13077 * other, and the number of required bits is much
13078 * larger, up to total of DOUBLEDOUBLE_MAXBITS bits.
13079 * See the definition of DOUBLEDOUBLE_MAXBITS.
13081 * Need 2 hexdigits for each byte. */
13082 (DOUBLEDOUBLE_MAXBITS/8 + 1) * 2;
13084 NVSIZE * 2; /* 2 hexdigits for each byte */
13086 /* see "this can't overflow" comment above */
13087 assert(float_need < ((STRLEN)~0) - digits);
13088 float_need += digits;
13091 /* special-case "%.<number>g" if it will fit in ebuf */
13093 && precis /* See earlier comment about buggy Gconvert
13094 when digits, aka precis, is 0 */
13096 /* check, in manner not involving wrapping, that it will
13098 && float_need < sizeof(ebuf)
13099 && sizeof(ebuf) - float_need > precis
13100 && !(width || left || plus || alt)
13104 SNPRINTF_G(fv, ebuf, sizeof(ebuf), precis);
13105 elen = strlen(ebuf);
13112 STRLEN pr = has_precis ? precis : 6; /* known default */
13113 /* this probably can't wrap, since precis is limited
13114 * to 1/4 address space size, but better safe than sorry
13116 if (float_need >= ((STRLEN)~0) - pr)
13117 croak_memory_wrap();
13121 if (float_need < width)
13122 float_need = width;
13124 if (PL_efloatsize <= float_need) {
13125 /* PL_efloatbuf should be at least 1 greater than
13126 * float_need to allow a trailing \0 to be returned by
13127 * snprintf(). If we need to grow, overgrow for the
13128 * benefit of future generations */
13129 const STRLEN extra = 0x20;
13130 if (float_need >= ((STRLEN)~0) - extra)
13131 croak_memory_wrap();
13132 float_need += extra;
13133 Safefree(PL_efloatbuf);
13134 PL_efloatsize = float_need;
13135 Newx(PL_efloatbuf, PL_efloatsize, char);
13136 PL_efloatbuf[0] = '\0';
13139 if (UNLIKELY(hexfp)) {
13140 elen = S_format_hexfp(aTHX_ PL_efloatbuf, PL_efloatsize, c,
13141 nv, fv, has_precis, precis, width,
13142 alt, plus, left, fill);
13145 char *ptr = ebuf + sizeof ebuf;
13148 #if defined(USE_QUADMATH)
13149 if (intsize == 'q') {
13153 /* FIXME: what to do if HAS_LONG_DOUBLE but not PERL_PRIfldbl? */
13154 #elif defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
13155 /* Note that this is HAS_LONG_DOUBLE and PERL_PRIfldbl,
13156 * not USE_LONG_DOUBLE and NVff. In other words,
13157 * this needs to work without USE_LONG_DOUBLE. */
13158 if (intsize == 'q') {
13159 /* Copy the one or more characters in a long double
13160 * format before the 'base' ([efgEFG]) character to
13161 * the format string. */
13162 static char const ldblf[] = PERL_PRIfldbl;
13163 char const *p = ldblf + sizeof(ldblf) - 3;
13164 while (p >= ldblf) { *--ptr = *p--; }
13169 do { *--ptr = '0' + (base % 10); } while (base /= 10);
13174 do { *--ptr = '0' + (base % 10); } while (base /= 10);
13186 /* No taint. Otherwise we are in the strange situation
13187 * where printf() taints but print($float) doesn't.
13190 /* hopefully the above makes ptr a very constrained format
13191 * that is safe to use, even though it's not literal */
13192 GCC_DIAG_IGNORE(-Wformat-nonliteral);
13193 #ifdef USE_QUADMATH
13195 const char* qfmt = quadmath_format_single(ptr);
13197 Perl_croak_nocontext("panic: quadmath invalid format \"%s\"", ptr);
13198 elen = quadmath_snprintf(PL_efloatbuf, PL_efloatsize,
13200 if ((IV)elen == -1) {
13203 Perl_croak_nocontext("panic: quadmath_snprintf failed, format \"%s\"", qfmt);
13208 #elif defined(HAS_LONG_DOUBLE)
13209 elen = ((intsize == 'q')
13210 ? my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv)
13211 : my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, (double)fv));
13213 elen = my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv);
13218 eptr = PL_efloatbuf;
13222 /* Since floating-point formats do their own formatting and
13223 * padding, we skip the main block of code at the end of this
13224 * loop which handles appending eptr to sv, and do our own
13225 * stripped-down version */
13230 assert(elen >= width);
13232 S_sv_catpvn_simple(aTHX_ sv, eptr, elen);
13234 goto done_valid_conversion;
13242 /* XXX ideally we should warn if any flags etc have been
13243 * set, e.g. "%-4.5n" */
13244 /* XXX if sv was originally non-utf8 with a char in the
13245 * range 0x80-0xff, then if it got upgraded, we should
13246 * calculate char len rather than byte len here */
13247 len = SvCUR(sv) - origlen;
13249 int i = (len > PERL_INT_MAX) ? PERL_INT_MAX : (int)len;
13252 case 'c': *(va_arg(*args, char*)) = i; break;
13253 case 'h': *(va_arg(*args, short*)) = i; break;
13254 default: *(va_arg(*args, int*)) = i; break;
13255 case 'l': *(va_arg(*args, long*)) = i; break;
13256 case 'V': *(va_arg(*args, IV*)) = i; break;
13257 case 'z': *(va_arg(*args, SSize_t*)) = i; break;
13258 #ifdef HAS_PTRDIFF_T
13259 case 't': *(va_arg(*args, ptrdiff_t*)) = i; break;
13262 case 'j': *(va_arg(*args, intmax_t*)) = i; break;
13266 *(va_arg(*args, Quad_t*)) = i; break;
13274 Perl_croak_nocontext(
13275 "Missing argument for %%n in %s",
13276 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
13277 sv_setuv_mg(argsv, has_utf8 ? (UV)sv_len_utf8(sv) : (UV)len);
13279 goto done_valid_conversion;
13287 && (PL_op->op_type == OP_PRTF || PL_op->op_type == OP_SPRINTF)
13288 && ckWARN(WARN_PRINTF))
13290 SV * const msg = sv_newmortal();
13291 Perl_sv_setpvf(aTHX_ msg, "Invalid conversion in %sprintf: ",
13292 (PL_op->op_type == OP_PRTF) ? "" : "s");
13293 if (fmtstart < patend) {
13294 const char * const fmtend = q < patend ? q : patend;
13296 sv_catpvs(msg, "\"%");
13297 for (f = fmtstart; f < fmtend; f++) {
13299 sv_catpvn_nomg(msg, f, 1);
13301 Perl_sv_catpvf(aTHX_ msg,
13302 "\\%03" UVof, (UV)*f & 0xFF);
13305 sv_catpvs(msg, "\"");
13307 sv_catpvs(msg, "end of string");
13309 Perl_warner(aTHX_ packWARN(WARN_PRINTF), "%" SVf, SVfARG(msg)); /* yes, this is reentrant */
13312 /* mangled format: output the '%', then continue from the
13313 * character following that */
13314 sv_catpvn_nomg(sv, fmtstart-1, 1);
13317 /* Any "redundant arg" warning from now onwards will probably
13318 * just be misleading, so don't bother. */
13319 no_redundant_warning = TRUE;
13320 continue; /* not "break" */
13323 if (is_utf8 != has_utf8) {
13326 sv_utf8_upgrade(sv);
13329 const STRLEN old_elen = elen;
13330 SV * const nsv = newSVpvn_flags(eptr, elen, SVs_TEMP);
13331 sv_utf8_upgrade(nsv);
13332 eptr = SvPVX_const(nsv);
13335 if (width) { /* fudge width (can't fudge elen) */
13336 width += elen - old_elen;
13343 /* append esignbuf, filler, zeros, eptr and dotstr to sv */
13346 STRLEN need, have, gap;
13350 /* signed value that's wrapped? */
13351 assert(elen <= ((~(STRLEN)0) >> 1));
13353 /* if zeros is non-zero, then it represents filler between
13354 * elen and precis. So adding elen and zeros together will
13355 * always be <= precis, and the addition can never wrap */
13356 assert(!zeros || (precis > elen && precis - elen == zeros));
13357 have = elen + zeros;
13359 if (have >= (((STRLEN)~0) - esignlen))
13360 croak_memory_wrap();
13363 need = (have > width ? have : width);
13366 if (need >= (((STRLEN)~0) - (SvCUR(sv) + 1)))
13367 croak_memory_wrap();
13368 need += (SvCUR(sv) + 1);
13375 for (i = 0; i < esignlen; i++)
13376 *s++ = esignbuf[i];
13377 for (i = zeros; i; i--)
13379 Copy(eptr, s, elen, char);
13381 for (i = gap; i; i--)
13386 for (i = 0; i < esignlen; i++)
13387 *s++ = esignbuf[i];
13392 for (i = gap; i; i--)
13394 for (i = 0; i < esignlen; i++)
13395 *s++ = esignbuf[i];
13398 for (i = zeros; i; i--)
13400 Copy(eptr, s, elen, char);
13405 SvCUR_set(sv, s - SvPVX_const(sv));
13413 if (vectorize && veclen) {
13414 /* we append the vector separator separately since %v isn't
13415 * very common: don't slow down the general case by adding
13416 * dotstrlen to need etc */
13417 sv_catpvn_nomg(sv, dotstr, dotstrlen);
13419 goto vector; /* do next iteration */
13422 done_valid_conversion:
13425 S_warn_vcatpvfn_missing_argument(aTHX);
13428 /* Now that we've consumed all our printf format arguments (svix)
13429 * do we have things left on the stack that we didn't use?
13431 if (!no_redundant_warning && sv_count >= svix + 1 && ckWARN(WARN_REDUNDANT)) {
13432 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
13433 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
13438 RESTORE_LC_NUMERIC(); /* Done outside loop, so don't have to save/restore
13442 /* =========================================================================
13444 =head1 Cloning an interpreter
13448 All the macros and functions in this section are for the private use of
13449 the main function, perl_clone().
13451 The foo_dup() functions make an exact copy of an existing foo thingy.
13452 During the course of a cloning, a hash table is used to map old addresses
13453 to new addresses. The table is created and manipulated with the
13454 ptr_table_* functions.
13456 * =========================================================================*/
13459 #if defined(USE_ITHREADS)
13461 /* XXX Remove this so it doesn't have to go thru the macro and return for nothing */
13462 #ifndef GpREFCNT_inc
13463 # define GpREFCNT_inc(gp) ((gp) ? (++(gp)->gp_refcnt, (gp)) : (GP*)NULL)
13467 /* Certain cases in Perl_ss_dup have been merged, by relying on the fact
13468 that currently av_dup, gv_dup and hv_dup are the same as sv_dup.
13469 If this changes, please unmerge ss_dup.
13470 Likewise, sv_dup_inc_multiple() relies on this fact. */
13471 #define sv_dup_inc_NN(s,t) SvREFCNT_inc_NN(sv_dup_inc(s,t))
13472 #define av_dup(s,t) MUTABLE_AV(sv_dup((const SV *)s,t))
13473 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
13474 #define hv_dup(s,t) MUTABLE_HV(sv_dup((const SV *)s,t))
13475 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
13476 #define cv_dup(s,t) MUTABLE_CV(sv_dup((const SV *)s,t))
13477 #define cv_dup_inc(s,t) MUTABLE_CV(sv_dup_inc((const SV *)s,t))
13478 #define io_dup(s,t) MUTABLE_IO(sv_dup((const SV *)s,t))
13479 #define io_dup_inc(s,t) MUTABLE_IO(sv_dup_inc((const SV *)s,t))
13480 #define gv_dup(s,t) MUTABLE_GV(sv_dup((const SV *)s,t))
13481 #define gv_dup_inc(s,t) MUTABLE_GV(sv_dup_inc((const SV *)s,t))
13482 #define SAVEPV(p) ((p) ? savepv(p) : NULL)
13483 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
13485 /* clone a parser */
13488 Perl_parser_dup(pTHX_ const yy_parser *const proto, CLONE_PARAMS *const param)
13492 PERL_ARGS_ASSERT_PARSER_DUP;
13497 /* look for it in the table first */
13498 parser = (yy_parser *)ptr_table_fetch(PL_ptr_table, proto);
13502 /* create anew and remember what it is */
13503 Newxz(parser, 1, yy_parser);
13504 ptr_table_store(PL_ptr_table, proto, parser);
13506 /* XXX eventually, just Copy() most of the parser struct ? */
13508 parser->lex_brackets = proto->lex_brackets;
13509 parser->lex_casemods = proto->lex_casemods;
13510 parser->lex_brackstack = savepvn(proto->lex_brackstack,
13511 (proto->lex_brackets < 120 ? 120 : proto->lex_brackets));
13512 parser->lex_casestack = savepvn(proto->lex_casestack,
13513 (proto->lex_casemods < 12 ? 12 : proto->lex_casemods));
13514 parser->lex_defer = proto->lex_defer;
13515 parser->lex_dojoin = proto->lex_dojoin;
13516 parser->lex_formbrack = proto->lex_formbrack;
13517 parser->lex_inpat = proto->lex_inpat;
13518 parser->lex_inwhat = proto->lex_inwhat;
13519 parser->lex_op = proto->lex_op;
13520 parser->lex_repl = sv_dup_inc(proto->lex_repl, param);
13521 parser->lex_starts = proto->lex_starts;
13522 parser->lex_stuff = sv_dup_inc(proto->lex_stuff, param);
13523 parser->multi_close = proto->multi_close;
13524 parser->multi_open = proto->multi_open;
13525 parser->multi_start = proto->multi_start;
13526 parser->multi_end = proto->multi_end;
13527 parser->preambled = proto->preambled;
13528 parser->lex_super_state = proto->lex_super_state;
13529 parser->lex_sub_inwhat = proto->lex_sub_inwhat;
13530 parser->lex_sub_op = proto->lex_sub_op;
13531 parser->lex_sub_repl= sv_dup_inc(proto->lex_sub_repl, param);
13532 parser->linestr = sv_dup_inc(proto->linestr, param);
13533 parser->expect = proto->expect;
13534 parser->copline = proto->copline;
13535 parser->last_lop_op = proto->last_lop_op;
13536 parser->lex_state = proto->lex_state;
13537 parser->rsfp = fp_dup(proto->rsfp, '<', param);
13538 /* rsfp_filters entries have fake IoDIRP() */
13539 parser->rsfp_filters= av_dup_inc(proto->rsfp_filters, param);
13540 parser->in_my = proto->in_my;
13541 parser->in_my_stash = hv_dup(proto->in_my_stash, param);
13542 parser->error_count = proto->error_count;
13543 parser->sig_elems = proto->sig_elems;
13544 parser->sig_optelems= proto->sig_optelems;
13545 parser->sig_slurpy = proto->sig_slurpy;
13546 parser->recheck_utf8_validity = proto->recheck_utf8_validity;
13549 char * const ols = SvPVX(proto->linestr);
13550 char * const ls = SvPVX(parser->linestr);
13552 parser->bufptr = ls + (proto->bufptr >= ols ?
13553 proto->bufptr - ols : 0);
13554 parser->oldbufptr = ls + (proto->oldbufptr >= ols ?
13555 proto->oldbufptr - ols : 0);
13556 parser->oldoldbufptr= ls + (proto->oldoldbufptr >= ols ?
13557 proto->oldoldbufptr - ols : 0);
13558 parser->linestart = ls + (proto->linestart >= ols ?
13559 proto->linestart - ols : 0);
13560 parser->last_uni = ls + (proto->last_uni >= ols ?
13561 proto->last_uni - ols : 0);
13562 parser->last_lop = ls + (proto->last_lop >= ols ?
13563 proto->last_lop - ols : 0);
13565 parser->bufend = ls + SvCUR(parser->linestr);
13568 Copy(proto->tokenbuf, parser->tokenbuf, 256, char);
13571 Copy(proto->nextval, parser->nextval, 5, YYSTYPE);
13572 Copy(proto->nexttype, parser->nexttype, 5, I32);
13573 parser->nexttoke = proto->nexttoke;
13575 /* XXX should clone saved_curcop here, but we aren't passed
13576 * proto_perl; so do it in perl_clone_using instead */
13582 /* duplicate a file handle */
13585 Perl_fp_dup(pTHX_ PerlIO *const fp, const char type, CLONE_PARAMS *const param)
13589 PERL_ARGS_ASSERT_FP_DUP;
13590 PERL_UNUSED_ARG(type);
13593 return (PerlIO*)NULL;
13595 /* look for it in the table first */
13596 ret = (PerlIO*)ptr_table_fetch(PL_ptr_table, fp);
13600 /* create anew and remember what it is */
13601 #ifdef __amigaos4__
13602 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE|PERLIO_DUP_FD);
13604 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE);
13606 ptr_table_store(PL_ptr_table, fp, ret);
13610 /* duplicate a directory handle */
13613 Perl_dirp_dup(pTHX_ DIR *const dp, CLONE_PARAMS *const param)
13617 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
13619 const Direntry_t *dirent;
13620 char smallbuf[256]; /* XXX MAXPATHLEN, surely? */
13626 PERL_UNUSED_CONTEXT;
13627 PERL_ARGS_ASSERT_DIRP_DUP;
13632 /* look for it in the table first */
13633 ret = (DIR*)ptr_table_fetch(PL_ptr_table, dp);
13637 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
13639 PERL_UNUSED_ARG(param);
13643 /* open the current directory (so we can switch back) */
13644 if (!(pwd = PerlDir_open("."))) return (DIR *)NULL;
13646 /* chdir to our dir handle and open the present working directory */
13647 if (fchdir(my_dirfd(dp)) < 0 || !(ret = PerlDir_open("."))) {
13648 PerlDir_close(pwd);
13649 return (DIR *)NULL;
13651 /* Now we should have two dir handles pointing to the same dir. */
13653 /* Be nice to the calling code and chdir back to where we were. */
13654 /* XXX If this fails, then what? */
13655 PERL_UNUSED_RESULT(fchdir(my_dirfd(pwd)));
13657 /* We have no need of the pwd handle any more. */
13658 PerlDir_close(pwd);
13661 # define d_namlen(d) (d)->d_namlen
13663 # define d_namlen(d) strlen((d)->d_name)
13665 /* Iterate once through dp, to get the file name at the current posi-
13666 tion. Then step back. */
13667 pos = PerlDir_tell(dp);
13668 if ((dirent = PerlDir_read(dp))) {
13669 len = d_namlen(dirent);
13670 if (len > sizeof(dirent->d_name) && sizeof(dirent->d_name) > PTRSIZE) {
13671 /* If the len is somehow magically longer than the
13672 * maximum length of the directory entry, even though
13673 * we could fit it in a buffer, we could not copy it
13674 * from the dirent. Bail out. */
13675 PerlDir_close(ret);
13678 if (len <= sizeof smallbuf) name = smallbuf;
13679 else Newx(name, len, char);
13680 Move(dirent->d_name, name, len, char);
13682 PerlDir_seek(dp, pos);
13684 /* Iterate through the new dir handle, till we find a file with the
13686 if (!dirent) /* just before the end */
13688 pos = PerlDir_tell(ret);
13689 if (PerlDir_read(ret)) continue; /* not there yet */
13690 PerlDir_seek(ret, pos); /* step back */
13694 const long pos0 = PerlDir_tell(ret);
13696 pos = PerlDir_tell(ret);
13697 if ((dirent = PerlDir_read(ret))) {
13698 if (len == (STRLEN)d_namlen(dirent)
13699 && memEQ(name, dirent->d_name, len)) {
13701 PerlDir_seek(ret, pos); /* step back */
13704 /* else we are not there yet; keep iterating */
13706 else { /* This is not meant to happen. The best we can do is
13707 reset the iterator to the beginning. */
13708 PerlDir_seek(ret, pos0);
13715 if (name && name != smallbuf)
13720 ret = win32_dirp_dup(dp, param);
13723 /* pop it in the pointer table */
13725 ptr_table_store(PL_ptr_table, dp, ret);
13730 /* duplicate a typeglob */
13733 Perl_gp_dup(pTHX_ GP *const gp, CLONE_PARAMS *const param)
13737 PERL_ARGS_ASSERT_GP_DUP;
13741 /* look for it in the table first */
13742 ret = (GP*)ptr_table_fetch(PL_ptr_table, gp);
13746 /* create anew and remember what it is */
13748 ptr_table_store(PL_ptr_table, gp, ret);
13751 /* ret->gp_refcnt must be 0 before any other dups are called. We're relying
13752 on Newxz() to do this for us. */
13753 ret->gp_sv = sv_dup_inc(gp->gp_sv, param);
13754 ret->gp_io = io_dup_inc(gp->gp_io, param);
13755 ret->gp_form = cv_dup_inc(gp->gp_form, param);
13756 ret->gp_av = av_dup_inc(gp->gp_av, param);
13757 ret->gp_hv = hv_dup_inc(gp->gp_hv, param);
13758 ret->gp_egv = gv_dup(gp->gp_egv, param);/* GvEGV is not refcounted */
13759 ret->gp_cv = cv_dup_inc(gp->gp_cv, param);
13760 ret->gp_cvgen = gp->gp_cvgen;
13761 ret->gp_line = gp->gp_line;
13762 ret->gp_file_hek = hek_dup(gp->gp_file_hek, param);
13766 /* duplicate a chain of magic */
13769 Perl_mg_dup(pTHX_ MAGIC *mg, CLONE_PARAMS *const param)
13771 MAGIC *mgret = NULL;
13772 MAGIC **mgprev_p = &mgret;
13774 PERL_ARGS_ASSERT_MG_DUP;
13776 for (; mg; mg = mg->mg_moremagic) {
13779 if ((param->flags & CLONEf_JOIN_IN)
13780 && mg->mg_type == PERL_MAGIC_backref)
13781 /* when joining, we let the individual SVs add themselves to
13782 * backref as needed. */
13785 Newx(nmg, 1, MAGIC);
13787 mgprev_p = &(nmg->mg_moremagic);
13789 /* There was a comment "XXX copy dynamic vtable?" but as we don't have
13790 dynamic vtables, I'm not sure why Sarathy wrote it. The comment dates
13791 from the original commit adding Perl_mg_dup() - revision 4538.
13792 Similarly there is the annotation "XXX random ptr?" next to the
13793 assignment to nmg->mg_ptr. */
13796 /* FIXME for plugins
13797 if (nmg->mg_type == PERL_MAGIC_qr) {
13798 nmg->mg_obj = MUTABLE_SV(CALLREGDUPE((REGEXP*)nmg->mg_obj, param));
13802 nmg->mg_obj = (nmg->mg_flags & MGf_REFCOUNTED)
13803 ? nmg->mg_type == PERL_MAGIC_backref
13804 /* The backref AV has its reference
13805 * count deliberately bumped by 1 */
13806 ? SvREFCNT_inc(av_dup_inc((const AV *)
13807 nmg->mg_obj, param))
13808 : sv_dup_inc(nmg->mg_obj, param)
13809 : (nmg->mg_type == PERL_MAGIC_regdatum ||
13810 nmg->mg_type == PERL_MAGIC_regdata)
13812 : sv_dup(nmg->mg_obj, param);
13814 if (nmg->mg_ptr && nmg->mg_type != PERL_MAGIC_regex_global) {
13815 if (nmg->mg_len > 0) {
13816 nmg->mg_ptr = SAVEPVN(nmg->mg_ptr, nmg->mg_len);
13817 if (nmg->mg_type == PERL_MAGIC_overload_table &&
13818 AMT_AMAGIC((AMT*)nmg->mg_ptr))
13820 AMT * const namtp = (AMT*)nmg->mg_ptr;
13821 sv_dup_inc_multiple((SV**)(namtp->table),
13822 (SV**)(namtp->table), NofAMmeth, param);
13825 else if (nmg->mg_len == HEf_SVKEY)
13826 nmg->mg_ptr = (char*)sv_dup_inc((const SV *)nmg->mg_ptr, param);
13828 if ((nmg->mg_flags & MGf_DUP) && nmg->mg_virtual && nmg->mg_virtual->svt_dup) {
13829 nmg->mg_virtual->svt_dup(aTHX_ nmg, param);
13835 #endif /* USE_ITHREADS */
13837 struct ptr_tbl_arena {
13838 struct ptr_tbl_arena *next;
13839 struct ptr_tbl_ent array[1023/3]; /* as ptr_tbl_ent has 3 pointers. */
13842 /* create a new pointer-mapping table */
13845 Perl_ptr_table_new(pTHX)
13848 PERL_UNUSED_CONTEXT;
13850 Newx(tbl, 1, PTR_TBL_t);
13851 tbl->tbl_max = 511;
13852 tbl->tbl_items = 0;
13853 tbl->tbl_arena = NULL;
13854 tbl->tbl_arena_next = NULL;
13855 tbl->tbl_arena_end = NULL;
13856 Newxz(tbl->tbl_ary, tbl->tbl_max + 1, PTR_TBL_ENT_t*);
13860 #define PTR_TABLE_HASH(ptr) \
13861 ((PTR2UV(ptr) >> 3) ^ (PTR2UV(ptr) >> (3 + 7)) ^ (PTR2UV(ptr) >> (3 + 17)))
13863 /* map an existing pointer using a table */
13865 STATIC PTR_TBL_ENT_t *
13866 S_ptr_table_find(PTR_TBL_t *const tbl, const void *const sv)
13868 PTR_TBL_ENT_t *tblent;
13869 const UV hash = PTR_TABLE_HASH(sv);
13871 PERL_ARGS_ASSERT_PTR_TABLE_FIND;
13873 tblent = tbl->tbl_ary[hash & tbl->tbl_max];
13874 for (; tblent; tblent = tblent->next) {
13875 if (tblent->oldval == sv)
13882 Perl_ptr_table_fetch(pTHX_ PTR_TBL_t *const tbl, const void *const sv)
13884 PTR_TBL_ENT_t const *const tblent = ptr_table_find(tbl, sv);
13886 PERL_ARGS_ASSERT_PTR_TABLE_FETCH;
13887 PERL_UNUSED_CONTEXT;
13889 return tblent ? tblent->newval : NULL;
13892 /* add a new entry to a pointer-mapping table 'tbl'. In hash terms, 'oldsv' is
13893 * the key; 'newsv' is the value. The names "old" and "new" are specific to
13894 * the core's typical use of ptr_tables in thread cloning. */
13897 Perl_ptr_table_store(pTHX_ PTR_TBL_t *const tbl, const void *const oldsv, void *const newsv)
13899 PTR_TBL_ENT_t *tblent = ptr_table_find(tbl, oldsv);
13901 PERL_ARGS_ASSERT_PTR_TABLE_STORE;
13902 PERL_UNUSED_CONTEXT;
13905 tblent->newval = newsv;
13907 const UV entry = PTR_TABLE_HASH(oldsv) & tbl->tbl_max;
13909 if (tbl->tbl_arena_next == tbl->tbl_arena_end) {
13910 struct ptr_tbl_arena *new_arena;
13912 Newx(new_arena, 1, struct ptr_tbl_arena);
13913 new_arena->next = tbl->tbl_arena;
13914 tbl->tbl_arena = new_arena;
13915 tbl->tbl_arena_next = new_arena->array;
13916 tbl->tbl_arena_end = C_ARRAY_END(new_arena->array);
13919 tblent = tbl->tbl_arena_next++;
13921 tblent->oldval = oldsv;
13922 tblent->newval = newsv;
13923 tblent->next = tbl->tbl_ary[entry];
13924 tbl->tbl_ary[entry] = tblent;
13926 if (tblent->next && tbl->tbl_items > tbl->tbl_max)
13927 ptr_table_split(tbl);
13931 /* double the hash bucket size of an existing ptr table */
13934 Perl_ptr_table_split(pTHX_ PTR_TBL_t *const tbl)
13936 PTR_TBL_ENT_t **ary = tbl->tbl_ary;
13937 const UV oldsize = tbl->tbl_max + 1;
13938 UV newsize = oldsize * 2;
13941 PERL_ARGS_ASSERT_PTR_TABLE_SPLIT;
13942 PERL_UNUSED_CONTEXT;
13944 Renew(ary, newsize, PTR_TBL_ENT_t*);
13945 Zero(&ary[oldsize], newsize-oldsize, PTR_TBL_ENT_t*);
13946 tbl->tbl_max = --newsize;
13947 tbl->tbl_ary = ary;
13948 for (i=0; i < oldsize; i++, ary++) {
13949 PTR_TBL_ENT_t **entp = ary;
13950 PTR_TBL_ENT_t *ent = *ary;
13951 PTR_TBL_ENT_t **curentp;
13954 curentp = ary + oldsize;
13956 if ((newsize & PTR_TABLE_HASH(ent->oldval)) != i) {
13958 ent->next = *curentp;
13968 /* remove all the entries from a ptr table */
13969 /* Deprecated - will be removed post 5.14 */
13972 Perl_ptr_table_clear(pTHX_ PTR_TBL_t *const tbl)
13974 PERL_UNUSED_CONTEXT;
13975 if (tbl && tbl->tbl_items) {
13976 struct ptr_tbl_arena *arena = tbl->tbl_arena;
13978 Zero(tbl->tbl_ary, tbl->tbl_max + 1, struct ptr_tbl_ent *);
13981 struct ptr_tbl_arena *next = arena->next;
13987 tbl->tbl_items = 0;
13988 tbl->tbl_arena = NULL;
13989 tbl->tbl_arena_next = NULL;
13990 tbl->tbl_arena_end = NULL;
13994 /* clear and free a ptr table */
13997 Perl_ptr_table_free(pTHX_ PTR_TBL_t *const tbl)
13999 struct ptr_tbl_arena *arena;
14001 PERL_UNUSED_CONTEXT;
14007 arena = tbl->tbl_arena;
14010 struct ptr_tbl_arena *next = arena->next;
14016 Safefree(tbl->tbl_ary);
14020 #if defined(USE_ITHREADS)
14023 Perl_rvpv_dup(pTHX_ SV *const dstr, const SV *const sstr, CLONE_PARAMS *const param)
14025 PERL_ARGS_ASSERT_RVPV_DUP;
14027 assert(!isREGEXP(sstr));
14029 if (SvWEAKREF(sstr)) {
14030 SvRV_set(dstr, sv_dup(SvRV_const(sstr), param));
14031 if (param->flags & CLONEf_JOIN_IN) {
14032 /* if joining, we add any back references individually rather
14033 * than copying the whole backref array */
14034 Perl_sv_add_backref(aTHX_ SvRV(dstr), dstr);
14038 SvRV_set(dstr, sv_dup_inc(SvRV_const(sstr), param));
14040 else if (SvPVX_const(sstr)) {
14041 /* Has something there */
14043 /* Normal PV - clone whole allocated space */
14044 SvPV_set(dstr, SAVEPVN(SvPVX_const(sstr), SvLEN(sstr)-1));
14045 /* sstr may not be that normal, but actually copy on write.
14046 But we are a true, independent SV, so: */
14050 /* Special case - not normally malloced for some reason */
14051 if (isGV_with_GP(sstr)) {
14052 /* Don't need to do anything here. */
14054 else if ((SvIsCOW(sstr))) {
14055 /* A "shared" PV - clone it as "shared" PV */
14057 HEK_KEY(hek_dup(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)),
14061 /* Some other special case - random pointer */
14062 SvPV_set(dstr, (char *) SvPVX_const(sstr));
14067 /* Copy the NULL */
14068 SvPV_set(dstr, NULL);
14072 /* duplicate a list of SVs. source and dest may point to the same memory. */
14074 S_sv_dup_inc_multiple(pTHX_ SV *const *source, SV **dest,
14075 SSize_t items, CLONE_PARAMS *const param)
14077 PERL_ARGS_ASSERT_SV_DUP_INC_MULTIPLE;
14079 while (items-- > 0) {
14080 *dest++ = sv_dup_inc(*source++, param);
14086 /* duplicate an SV of any type (including AV, HV etc) */
14089 S_sv_dup_common(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14094 PERL_ARGS_ASSERT_SV_DUP_COMMON;
14096 if (SvTYPE(sstr) == (svtype)SVTYPEMASK) {
14097 #ifdef DEBUG_LEAKING_SCALARS_ABORT
14102 /* look for it in the table first */
14103 dstr = MUTABLE_SV(ptr_table_fetch(PL_ptr_table, sstr));
14107 if(param->flags & CLONEf_JOIN_IN) {
14108 /** We are joining here so we don't want do clone
14109 something that is bad **/
14110 if (SvTYPE(sstr) == SVt_PVHV) {
14111 const HEK * const hvname = HvNAME_HEK(sstr);
14113 /** don't clone stashes if they already exist **/
14114 dstr = MUTABLE_SV(gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
14115 HEK_UTF8(hvname) ? SVf_UTF8 : 0));
14116 ptr_table_store(PL_ptr_table, sstr, dstr);
14120 else if (SvTYPE(sstr) == SVt_PVGV && !SvFAKE(sstr)) {
14121 HV *stash = GvSTASH(sstr);
14122 const HEK * hvname;
14123 if (stash && (hvname = HvNAME_HEK(stash))) {
14124 /** don't clone GVs if they already exist **/
14126 stash = gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
14127 HEK_UTF8(hvname) ? SVf_UTF8 : 0);
14129 stash, GvNAME(sstr),
14135 if (svp && *svp && SvTYPE(*svp) == SVt_PVGV) {
14136 ptr_table_store(PL_ptr_table, sstr, *svp);
14143 /* create anew and remember what it is */
14146 #ifdef DEBUG_LEAKING_SCALARS
14147 dstr->sv_debug_optype = sstr->sv_debug_optype;
14148 dstr->sv_debug_line = sstr->sv_debug_line;
14149 dstr->sv_debug_inpad = sstr->sv_debug_inpad;
14150 dstr->sv_debug_parent = (SV*)sstr;
14151 FREE_SV_DEBUG_FILE(dstr);
14152 dstr->sv_debug_file = savesharedpv(sstr->sv_debug_file);
14155 ptr_table_store(PL_ptr_table, sstr, dstr);
14158 SvFLAGS(dstr) = SvFLAGS(sstr);
14159 SvFLAGS(dstr) &= ~SVf_OOK; /* don't propagate OOK hack */
14160 SvREFCNT(dstr) = 0; /* must be before any other dups! */
14163 if (SvANY(sstr) && PL_watch_pvx && SvPVX_const(sstr) == PL_watch_pvx)
14164 PerlIO_printf(Perl_debug_log, "watch at %p hit, found string \"%s\"\n",
14165 (void*)PL_watch_pvx, SvPVX_const(sstr));
14168 /* don't clone objects whose class has asked us not to */
14170 && ! (SvFLAGS(SvSTASH(sstr)) & SVphv_CLONEABLE))
14176 switch (SvTYPE(sstr)) {
14178 SvANY(dstr) = NULL;
14181 SET_SVANY_FOR_BODYLESS_IV(dstr);
14183 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
14185 SvIV_set(dstr, SvIVX(sstr));
14189 #if NVSIZE <= IVSIZE
14190 SET_SVANY_FOR_BODYLESS_NV(dstr);
14192 SvANY(dstr) = new_XNV();
14194 SvNV_set(dstr, SvNVX(sstr));
14198 /* These are all the types that need complex bodies allocating. */
14200 const svtype sv_type = SvTYPE(sstr);
14201 const struct body_details *const sv_type_details
14202 = bodies_by_type + sv_type;
14206 Perl_croak(aTHX_ "Bizarre SvTYPE [%" IVdf "]", (IV)SvTYPE(sstr));
14207 NOT_REACHED; /* NOTREACHED */
14223 assert(sv_type_details->body_size);
14224 if (sv_type_details->arena) {
14225 new_body_inline(new_body, sv_type);
14227 = (void*)((char*)new_body - sv_type_details->offset);
14229 new_body = new_NOARENA(sv_type_details);
14233 SvANY(dstr) = new_body;
14236 Copy(((char*)SvANY(sstr)) + sv_type_details->offset,
14237 ((char*)SvANY(dstr)) + sv_type_details->offset,
14238 sv_type_details->copy, char);
14240 Copy(((char*)SvANY(sstr)),
14241 ((char*)SvANY(dstr)),
14242 sv_type_details->body_size + sv_type_details->offset, char);
14245 if (sv_type != SVt_PVAV && sv_type != SVt_PVHV
14246 && !isGV_with_GP(dstr)
14248 && !(sv_type == SVt_PVIO && !(IoFLAGS(dstr) & IOf_FAKE_DIRP)))
14249 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
14251 /* The Copy above means that all the source (unduplicated) pointers
14252 are now in the destination. We can check the flags and the
14253 pointers in either, but it's possible that there's less cache
14254 missing by always going for the destination.
14255 FIXME - instrument and check that assumption */
14256 if (sv_type >= SVt_PVMG) {
14258 SvMAGIC_set(dstr, mg_dup(SvMAGIC(dstr), param));
14259 if (SvOBJECT(dstr) && SvSTASH(dstr))
14260 SvSTASH_set(dstr, hv_dup_inc(SvSTASH(dstr), param));
14261 else SvSTASH_set(dstr, 0); /* don't copy DESTROY cache */
14264 /* The cast silences a GCC warning about unhandled types. */
14265 switch ((int)sv_type) {
14276 /* FIXME for plugins */
14277 re_dup_guts((REGEXP*) sstr, (REGEXP*) dstr, param);
14280 /* XXX LvTARGOFF sometimes holds PMOP* when DEBUGGING */
14281 if (LvTYPE(dstr) == 't') /* for tie: unrefcnted fake (SV**) */
14282 LvTARG(dstr) = dstr;
14283 else if (LvTYPE(dstr) == 'T') /* for tie: fake HE */
14284 LvTARG(dstr) = MUTABLE_SV(he_dup((HE*)LvTARG(dstr), 0, param));
14286 LvTARG(dstr) = sv_dup_inc(LvTARG(dstr), param);
14287 if (isREGEXP(sstr)) goto duprex;
14290 /* non-GP case already handled above */
14291 if(isGV_with_GP(sstr)) {
14292 GvNAME_HEK(dstr) = hek_dup(GvNAME_HEK(dstr), param);
14293 /* Don't call sv_add_backref here as it's going to be
14294 created as part of the magic cloning of the symbol
14295 table--unless this is during a join and the stash
14296 is not actually being cloned. */
14297 /* Danger Will Robinson - GvGP(dstr) isn't initialised
14298 at the point of this comment. */
14299 GvSTASH(dstr) = hv_dup(GvSTASH(dstr), param);
14300 if (param->flags & CLONEf_JOIN_IN)
14301 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
14302 GvGP_set(dstr, gp_dup(GvGP(sstr), param));
14303 (void)GpREFCNT_inc(GvGP(dstr));
14307 /* PL_parser->rsfp_filters entries have fake IoDIRP() */
14308 if(IoFLAGS(dstr) & IOf_FAKE_DIRP) {
14309 /* I have no idea why fake dirp (rsfps)
14310 should be treated differently but otherwise
14311 we end up with leaks -- sky*/
14312 IoTOP_GV(dstr) = gv_dup_inc(IoTOP_GV(dstr), param);
14313 IoFMT_GV(dstr) = gv_dup_inc(IoFMT_GV(dstr), param);
14314 IoBOTTOM_GV(dstr) = gv_dup_inc(IoBOTTOM_GV(dstr), param);
14316 IoTOP_GV(dstr) = gv_dup(IoTOP_GV(dstr), param);
14317 IoFMT_GV(dstr) = gv_dup(IoFMT_GV(dstr), param);
14318 IoBOTTOM_GV(dstr) = gv_dup(IoBOTTOM_GV(dstr), param);
14319 if (IoDIRP(dstr)) {
14320 IoDIRP(dstr) = dirp_dup(IoDIRP(dstr), param);
14323 /* IoDIRP(dstr) is already a copy of IoDIRP(sstr) */
14325 IoIFP(dstr) = fp_dup(IoIFP(sstr), IoTYPE(dstr), param);
14327 if (IoOFP(dstr) == IoIFP(sstr))
14328 IoOFP(dstr) = IoIFP(dstr);
14330 IoOFP(dstr) = fp_dup(IoOFP(dstr), IoTYPE(dstr), param);
14331 IoTOP_NAME(dstr) = SAVEPV(IoTOP_NAME(dstr));
14332 IoFMT_NAME(dstr) = SAVEPV(IoFMT_NAME(dstr));
14333 IoBOTTOM_NAME(dstr) = SAVEPV(IoBOTTOM_NAME(dstr));
14336 /* avoid cloning an empty array */
14337 if (AvARRAY((const AV *)sstr) && AvFILLp((const AV *)sstr) >= 0) {
14338 SV **dst_ary, **src_ary;
14339 SSize_t items = AvFILLp((const AV *)sstr) + 1;
14341 src_ary = AvARRAY((const AV *)sstr);
14342 Newx(dst_ary, AvMAX((const AV *)sstr)+1, SV*);
14343 ptr_table_store(PL_ptr_table, src_ary, dst_ary);
14344 AvARRAY(MUTABLE_AV(dstr)) = dst_ary;
14345 AvALLOC((const AV *)dstr) = dst_ary;
14346 if (AvREAL((const AV *)sstr)) {
14347 dst_ary = sv_dup_inc_multiple(src_ary, dst_ary, items,
14351 while (items-- > 0)
14352 *dst_ary++ = sv_dup(*src_ary++, param);
14354 items = AvMAX((const AV *)sstr) - AvFILLp((const AV *)sstr);
14355 while (items-- > 0) {
14360 AvARRAY(MUTABLE_AV(dstr)) = NULL;
14361 AvALLOC((const AV *)dstr) = (SV**)NULL;
14362 AvMAX( (const AV *)dstr) = -1;
14363 AvFILLp((const AV *)dstr) = -1;
14367 if (HvARRAY((const HV *)sstr)) {
14369 const bool sharekeys = !!HvSHAREKEYS(sstr);
14370 XPVHV * const dxhv = (XPVHV*)SvANY(dstr);
14371 XPVHV * const sxhv = (XPVHV*)SvANY(sstr);
14373 Newx(darray, PERL_HV_ARRAY_ALLOC_BYTES(dxhv->xhv_max+1)
14374 + (SvOOK(sstr) ? sizeof(struct xpvhv_aux) : 0),
14376 HvARRAY(dstr) = (HE**)darray;
14377 while (i <= sxhv->xhv_max) {
14378 const HE * const source = HvARRAY(sstr)[i];
14379 HvARRAY(dstr)[i] = source
14380 ? he_dup(source, sharekeys, param) : 0;
14384 const struct xpvhv_aux * const saux = HvAUX(sstr);
14385 struct xpvhv_aux * const daux = HvAUX(dstr);
14386 /* This flag isn't copied. */
14389 if (saux->xhv_name_count) {
14390 HEK ** const sname = saux->xhv_name_u.xhvnameu_names;
14392 = saux->xhv_name_count < 0
14393 ? -saux->xhv_name_count
14394 : saux->xhv_name_count;
14395 HEK **shekp = sname + count;
14397 Newx(daux->xhv_name_u.xhvnameu_names, count, HEK *);
14398 dhekp = daux->xhv_name_u.xhvnameu_names + count;
14399 while (shekp-- > sname) {
14401 *dhekp = hek_dup(*shekp, param);
14405 daux->xhv_name_u.xhvnameu_name
14406 = hek_dup(saux->xhv_name_u.xhvnameu_name,
14409 daux->xhv_name_count = saux->xhv_name_count;
14411 daux->xhv_aux_flags = saux->xhv_aux_flags;
14412 #ifdef PERL_HASH_RANDOMIZE_KEYS
14413 daux->xhv_rand = saux->xhv_rand;
14414 daux->xhv_last_rand = saux->xhv_last_rand;
14416 daux->xhv_riter = saux->xhv_riter;
14417 daux->xhv_eiter = saux->xhv_eiter
14418 ? he_dup(saux->xhv_eiter,
14419 cBOOL(HvSHAREKEYS(sstr)), param) : 0;
14420 /* backref array needs refcnt=2; see sv_add_backref */
14421 daux->xhv_backreferences =
14422 (param->flags & CLONEf_JOIN_IN)
14423 /* when joining, we let the individual GVs and
14424 * CVs add themselves to backref as
14425 * needed. This avoids pulling in stuff
14426 * that isn't required, and simplifies the
14427 * case where stashes aren't cloned back
14428 * if they already exist in the parent
14431 : saux->xhv_backreferences
14432 ? (SvTYPE(saux->xhv_backreferences) == SVt_PVAV)
14433 ? MUTABLE_AV(SvREFCNT_inc(
14434 sv_dup_inc((const SV *)
14435 saux->xhv_backreferences, param)))
14436 : MUTABLE_AV(sv_dup((const SV *)
14437 saux->xhv_backreferences, param))
14440 daux->xhv_mro_meta = saux->xhv_mro_meta
14441 ? mro_meta_dup(saux->xhv_mro_meta, param)
14444 /* Record stashes for possible cloning in Perl_clone(). */
14446 av_push(param->stashes, dstr);
14450 HvARRAY(MUTABLE_HV(dstr)) = NULL;
14453 if (!(param->flags & CLONEf_COPY_STACKS)) {
14458 /* NOTE: not refcounted */
14459 SvANY(MUTABLE_CV(dstr))->xcv_stash =
14460 hv_dup(CvSTASH(dstr), param);
14461 if ((param->flags & CLONEf_JOIN_IN) && CvSTASH(dstr))
14462 Perl_sv_add_backref(aTHX_ MUTABLE_SV(CvSTASH(dstr)), dstr);
14463 if (!CvISXSUB(dstr)) {
14465 CvROOT(dstr) = OpREFCNT_inc(CvROOT(dstr));
14467 CvSLABBED_off(dstr);
14468 } else if (CvCONST(dstr)) {
14469 CvXSUBANY(dstr).any_ptr =
14470 sv_dup_inc((const SV *)CvXSUBANY(dstr).any_ptr, param);
14472 assert(!CvSLABBED(dstr));
14473 if (CvDYNFILE(dstr)) CvFILE(dstr) = SAVEPV(CvFILE(dstr));
14475 SvANY((CV *)dstr)->xcv_gv_u.xcv_hek =
14476 hek_dup(CvNAME_HEK((CV *)sstr), param);
14477 /* don't dup if copying back - CvGV isn't refcounted, so the
14478 * duped GV may never be freed. A bit of a hack! DAPM */
14480 SvANY(MUTABLE_CV(dstr))->xcv_gv_u.xcv_gv =
14482 ? gv_dup_inc(CvGV(sstr), param)
14483 : (param->flags & CLONEf_JOIN_IN)
14485 : gv_dup(CvGV(sstr), param);
14487 if (!CvISXSUB(sstr)) {
14488 PADLIST * padlist = CvPADLIST(sstr);
14490 padlist = padlist_dup(padlist, param);
14491 CvPADLIST_set(dstr, padlist);
14493 /* unthreaded perl can't sv_dup so we dont support unthreaded's CvHSCXT */
14494 PoisonPADLIST(dstr);
14497 CvWEAKOUTSIDE(sstr)
14498 ? cv_dup( CvOUTSIDE(dstr), param)
14499 : cv_dup_inc(CvOUTSIDE(dstr), param);
14509 Perl_sv_dup_inc(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14511 PERL_ARGS_ASSERT_SV_DUP_INC;
14512 return sstr ? SvREFCNT_inc(sv_dup_common(sstr, param)) : NULL;
14516 Perl_sv_dup(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14518 SV *dstr = sstr ? sv_dup_common(sstr, param) : NULL;
14519 PERL_ARGS_ASSERT_SV_DUP;
14521 /* Track every SV that (at least initially) had a reference count of 0.
14522 We need to do this by holding an actual reference to it in this array.
14523 If we attempt to cheat, turn AvREAL_off(), and store only pointers
14524 (akin to the stashes hash, and the perl stack), we come unstuck if
14525 a weak reference (or other SV legitimately SvREFCNT() == 0 for this
14526 thread) is manipulated in a CLONE method, because CLONE runs before the
14527 unreferenced array is walked to find SVs still with SvREFCNT() == 0
14528 (and fix things up by giving each a reference via the temps stack).
14529 Instead, during CLONE, if the 0-referenced SV has SvREFCNT_inc() and
14530 then SvREFCNT_dec(), it will be cleaned up (and added to the free list)
14531 before the walk of unreferenced happens and a reference to that is SV
14532 added to the temps stack. At which point we have the same SV considered
14533 to be in use, and free to be re-used. Not good.
14535 if (dstr && !(param->flags & CLONEf_COPY_STACKS) && !SvREFCNT(dstr)) {
14536 assert(param->unreferenced);
14537 av_push(param->unreferenced, SvREFCNT_inc(dstr));
14543 /* duplicate a context */
14546 Perl_cx_dup(pTHX_ PERL_CONTEXT *cxs, I32 ix, I32 max, CLONE_PARAMS* param)
14548 PERL_CONTEXT *ncxs;
14550 PERL_ARGS_ASSERT_CX_DUP;
14553 return (PERL_CONTEXT*)NULL;
14555 /* look for it in the table first */
14556 ncxs = (PERL_CONTEXT*)ptr_table_fetch(PL_ptr_table, cxs);
14560 /* create anew and remember what it is */
14561 Newx(ncxs, max + 1, PERL_CONTEXT);
14562 ptr_table_store(PL_ptr_table, cxs, ncxs);
14563 Copy(cxs, ncxs, max + 1, PERL_CONTEXT);
14566 PERL_CONTEXT * const ncx = &ncxs[ix];
14567 if (CxTYPE(ncx) == CXt_SUBST) {
14568 Perl_croak(aTHX_ "Cloning substitution context is unimplemented");
14571 ncx->blk_oldcop = (COP*)any_dup(ncx->blk_oldcop, param->proto_perl);
14572 switch (CxTYPE(ncx)) {
14574 ncx->blk_sub.cv = cv_dup_inc(ncx->blk_sub.cv, param);
14575 if(CxHASARGS(ncx)){
14576 ncx->blk_sub.savearray = av_dup_inc(ncx->blk_sub.savearray,param);
14578 ncx->blk_sub.savearray = NULL;
14580 ncx->blk_sub.prevcomppad = (PAD*)ptr_table_fetch(PL_ptr_table,
14581 ncx->blk_sub.prevcomppad);
14584 ncx->blk_eval.old_namesv = sv_dup_inc(ncx->blk_eval.old_namesv,
14586 /* XXX should this sv_dup_inc? Or only if CxEVAL_TXT_REFCNTED ???? */
14587 ncx->blk_eval.cur_text = sv_dup(ncx->blk_eval.cur_text, param);
14588 ncx->blk_eval.cv = cv_dup(ncx->blk_eval.cv, param);
14589 /* XXX what do do with cur_top_env ???? */
14591 case CXt_LOOP_LAZYSV:
14592 ncx->blk_loop.state_u.lazysv.end
14593 = sv_dup_inc(ncx->blk_loop.state_u.lazysv.end, param);
14594 /* Fallthrough: duplicate lazysv.cur by using the ary.ary
14595 duplication code instead.
14596 We are taking advantage of (1) av_dup_inc and sv_dup_inc
14597 actually being the same function, and (2) order
14598 equivalence of the two unions.
14599 We can assert the later [but only at run time :-(] */
14600 assert ((void *) &ncx->blk_loop.state_u.ary.ary ==
14601 (void *) &ncx->blk_loop.state_u.lazysv.cur);
14604 ncx->blk_loop.state_u.ary.ary
14605 = av_dup_inc(ncx->blk_loop.state_u.ary.ary, param);
14607 case CXt_LOOP_LIST:
14608 case CXt_LOOP_LAZYIV:
14609 /* code common to all 'for' CXt_LOOP_* types */
14610 ncx->blk_loop.itersave =
14611 sv_dup_inc(ncx->blk_loop.itersave, param);
14612 if (CxPADLOOP(ncx)) {
14613 PADOFFSET off = ncx->blk_loop.itervar_u.svp
14614 - &CX_CURPAD_SV(ncx->blk_loop, 0);
14615 ncx->blk_loop.oldcomppad =
14616 (PAD*)ptr_table_fetch(PL_ptr_table,
14617 ncx->blk_loop.oldcomppad);
14618 ncx->blk_loop.itervar_u.svp =
14619 &CX_CURPAD_SV(ncx->blk_loop, off);
14622 /* this copies the GV if CXp_FOR_GV, or the SV for an
14623 * alias (for \$x (...)) - relies on gv_dup being the
14624 * same as sv_dup */
14625 ncx->blk_loop.itervar_u.gv
14626 = gv_dup((const GV *)ncx->blk_loop.itervar_u.gv,
14630 case CXt_LOOP_PLAIN:
14633 ncx->blk_format.prevcomppad =
14634 (PAD*)ptr_table_fetch(PL_ptr_table,
14635 ncx->blk_format.prevcomppad);
14636 ncx->blk_format.cv = cv_dup_inc(ncx->blk_format.cv, param);
14637 ncx->blk_format.gv = gv_dup(ncx->blk_format.gv, param);
14638 ncx->blk_format.dfoutgv = gv_dup_inc(ncx->blk_format.dfoutgv,
14642 ncx->blk_givwhen.defsv_save =
14643 sv_dup_inc(ncx->blk_givwhen.defsv_save, param);
14656 /* duplicate a stack info structure */
14659 Perl_si_dup(pTHX_ PERL_SI *si, CLONE_PARAMS* param)
14663 PERL_ARGS_ASSERT_SI_DUP;
14666 return (PERL_SI*)NULL;
14668 /* look for it in the table first */
14669 nsi = (PERL_SI*)ptr_table_fetch(PL_ptr_table, si);
14673 /* create anew and remember what it is */
14674 Newx(nsi, 1, PERL_SI);
14675 ptr_table_store(PL_ptr_table, si, nsi);
14677 nsi->si_stack = av_dup_inc(si->si_stack, param);
14678 nsi->si_cxix = si->si_cxix;
14679 nsi->si_cxmax = si->si_cxmax;
14680 nsi->si_cxstack = cx_dup(si->si_cxstack, si->si_cxix, si->si_cxmax, param);
14681 nsi->si_type = si->si_type;
14682 nsi->si_prev = si_dup(si->si_prev, param);
14683 nsi->si_next = si_dup(si->si_next, param);
14684 nsi->si_markoff = si->si_markoff;
14685 #if defined DEBUGGING && !defined DEBUGGING_RE_ONLY
14686 nsi->si_stack_hwm = 0;
14692 #define POPINT(ss,ix) ((ss)[--(ix)].any_i32)
14693 #define TOPINT(ss,ix) ((ss)[ix].any_i32)
14694 #define POPLONG(ss,ix) ((ss)[--(ix)].any_long)
14695 #define TOPLONG(ss,ix) ((ss)[ix].any_long)
14696 #define POPIV(ss,ix) ((ss)[--(ix)].any_iv)
14697 #define TOPIV(ss,ix) ((ss)[ix].any_iv)
14698 #define POPUV(ss,ix) ((ss)[--(ix)].any_uv)
14699 #define TOPUV(ss,ix) ((ss)[ix].any_uv)
14700 #define POPBOOL(ss,ix) ((ss)[--(ix)].any_bool)
14701 #define TOPBOOL(ss,ix) ((ss)[ix].any_bool)
14702 #define POPPTR(ss,ix) ((ss)[--(ix)].any_ptr)
14703 #define TOPPTR(ss,ix) ((ss)[ix].any_ptr)
14704 #define POPDPTR(ss,ix) ((ss)[--(ix)].any_dptr)
14705 #define TOPDPTR(ss,ix) ((ss)[ix].any_dptr)
14706 #define POPDXPTR(ss,ix) ((ss)[--(ix)].any_dxptr)
14707 #define TOPDXPTR(ss,ix) ((ss)[ix].any_dxptr)
14710 #define pv_dup_inc(p) SAVEPV(p)
14711 #define pv_dup(p) SAVEPV(p)
14712 #define svp_dup_inc(p,pp) any_dup(p,pp)
14714 /* map any object to the new equivent - either something in the
14715 * ptr table, or something in the interpreter structure
14719 Perl_any_dup(pTHX_ void *v, const PerlInterpreter *proto_perl)
14723 PERL_ARGS_ASSERT_ANY_DUP;
14726 return (void*)NULL;
14728 /* look for it in the table first */
14729 ret = ptr_table_fetch(PL_ptr_table, v);
14733 /* see if it is part of the interpreter structure */
14734 if (v >= (void*)proto_perl && v < (void*)(proto_perl+1))
14735 ret = (void*)(((char*)aTHX) + (((char*)v) - (char*)proto_perl));
14743 /* duplicate the save stack */
14746 Perl_ss_dup(pTHX_ PerlInterpreter *proto_perl, CLONE_PARAMS* param)
14749 ANY * const ss = proto_perl->Isavestack;
14750 const I32 max = proto_perl->Isavestack_max + SS_MAXPUSH;
14751 I32 ix = proto_perl->Isavestack_ix;
14764 void (*dptr) (void*);
14765 void (*dxptr) (pTHX_ void*);
14767 PERL_ARGS_ASSERT_SS_DUP;
14769 Newx(nss, max, ANY);
14772 const UV uv = POPUV(ss,ix);
14773 const U8 type = (U8)uv & SAVE_MASK;
14775 TOPUV(nss,ix) = uv;
14777 case SAVEt_CLEARSV:
14778 case SAVEt_CLEARPADRANGE:
14780 case SAVEt_HELEM: /* hash element */
14781 case SAVEt_SV: /* scalar reference */
14782 sv = (const SV *)POPPTR(ss,ix);
14783 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14785 case SAVEt_ITEM: /* normal string */
14786 case SAVEt_GVSV: /* scalar slot in GV */
14787 sv = (const SV *)POPPTR(ss,ix);
14788 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14789 if (type == SAVEt_SV)
14793 case SAVEt_MORTALIZESV:
14794 case SAVEt_READONLY_OFF:
14795 sv = (const SV *)POPPTR(ss,ix);
14796 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14798 case SAVEt_FREEPADNAME:
14799 ptr = POPPTR(ss,ix);
14800 TOPPTR(nss,ix) = padname_dup((PADNAME *)ptr, param);
14801 PadnameREFCNT((PADNAME *)TOPPTR(nss,ix))++;
14803 case SAVEt_SHARED_PVREF: /* char* in shared space */
14804 c = (char*)POPPTR(ss,ix);
14805 TOPPTR(nss,ix) = savesharedpv(c);
14806 ptr = POPPTR(ss,ix);
14807 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14809 case SAVEt_GENERIC_SVREF: /* generic sv */
14810 case SAVEt_SVREF: /* scalar reference */
14811 sv = (const SV *)POPPTR(ss,ix);
14812 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14813 if (type == SAVEt_SVREF)
14814 SvREFCNT_inc_simple_void((SV *)TOPPTR(nss,ix));
14815 ptr = POPPTR(ss,ix);
14816 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14818 case SAVEt_GVSLOT: /* any slot in GV */
14819 sv = (const SV *)POPPTR(ss,ix);
14820 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14821 ptr = POPPTR(ss,ix);
14822 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14823 sv = (const SV *)POPPTR(ss,ix);
14824 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14826 case SAVEt_HV: /* hash reference */
14827 case SAVEt_AV: /* array reference */
14828 sv = (const SV *) POPPTR(ss,ix);
14829 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14831 case SAVEt_COMPPAD:
14833 sv = (const SV *) POPPTR(ss,ix);
14834 TOPPTR(nss,ix) = sv_dup(sv, param);
14836 case SAVEt_INT: /* int reference */
14837 ptr = POPPTR(ss,ix);
14838 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14839 intval = (int)POPINT(ss,ix);
14840 TOPINT(nss,ix) = intval;
14842 case SAVEt_LONG: /* long reference */
14843 ptr = POPPTR(ss,ix);
14844 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14845 longval = (long)POPLONG(ss,ix);
14846 TOPLONG(nss,ix) = longval;
14848 case SAVEt_I32: /* I32 reference */
14849 ptr = POPPTR(ss,ix);
14850 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14852 TOPINT(nss,ix) = i;
14854 case SAVEt_IV: /* IV reference */
14855 case SAVEt_STRLEN: /* STRLEN/size_t ref */
14856 ptr = POPPTR(ss,ix);
14857 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14859 TOPIV(nss,ix) = iv;
14861 case SAVEt_TMPSFLOOR:
14863 TOPIV(nss,ix) = iv;
14865 case SAVEt_HPTR: /* HV* reference */
14866 case SAVEt_APTR: /* AV* reference */
14867 case SAVEt_SPTR: /* SV* reference */
14868 ptr = POPPTR(ss,ix);
14869 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14870 sv = (const SV *)POPPTR(ss,ix);
14871 TOPPTR(nss,ix) = sv_dup(sv, param);
14873 case SAVEt_VPTR: /* random* reference */
14874 ptr = POPPTR(ss,ix);
14875 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14877 case SAVEt_INT_SMALL:
14878 case SAVEt_I32_SMALL:
14879 case SAVEt_I16: /* I16 reference */
14880 case SAVEt_I8: /* I8 reference */
14882 ptr = POPPTR(ss,ix);
14883 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14885 case SAVEt_GENERIC_PVREF: /* generic char* */
14886 case SAVEt_PPTR: /* char* reference */
14887 ptr = POPPTR(ss,ix);
14888 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14889 c = (char*)POPPTR(ss,ix);
14890 TOPPTR(nss,ix) = pv_dup(c);
14892 case SAVEt_GP: /* scalar reference */
14893 gp = (GP*)POPPTR(ss,ix);
14894 TOPPTR(nss,ix) = gp = gp_dup(gp, param);
14895 (void)GpREFCNT_inc(gp);
14896 gv = (const GV *)POPPTR(ss,ix);
14897 TOPPTR(nss,ix) = gv_dup_inc(gv, param);
14900 ptr = POPPTR(ss,ix);
14901 if (ptr && (((OP*)ptr)->op_private & OPpREFCOUNTED)) {
14902 /* these are assumed to be refcounted properly */
14904 switch (((OP*)ptr)->op_type) {
14906 case OP_LEAVESUBLV:
14910 case OP_LEAVEWRITE:
14911 TOPPTR(nss,ix) = ptr;
14914 (void) OpREFCNT_inc(o);
14918 TOPPTR(nss,ix) = NULL;
14923 TOPPTR(nss,ix) = NULL;
14925 case SAVEt_FREECOPHH:
14926 ptr = POPPTR(ss,ix);
14927 TOPPTR(nss,ix) = cophh_copy((COPHH *)ptr);
14929 case SAVEt_ADELETE:
14930 av = (const AV *)POPPTR(ss,ix);
14931 TOPPTR(nss,ix) = av_dup_inc(av, param);
14933 TOPINT(nss,ix) = i;
14936 hv = (const HV *)POPPTR(ss,ix);
14937 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14939 TOPINT(nss,ix) = i;
14942 c = (char*)POPPTR(ss,ix);
14943 TOPPTR(nss,ix) = pv_dup_inc(c);
14945 case SAVEt_STACK_POS: /* Position on Perl stack */
14947 TOPINT(nss,ix) = i;
14949 case SAVEt_DESTRUCTOR:
14950 ptr = POPPTR(ss,ix);
14951 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14952 dptr = POPDPTR(ss,ix);
14953 TOPDPTR(nss,ix) = DPTR2FPTR(void (*)(void*),
14954 any_dup(FPTR2DPTR(void *, dptr),
14957 case SAVEt_DESTRUCTOR_X:
14958 ptr = POPPTR(ss,ix);
14959 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14960 dxptr = POPDXPTR(ss,ix);
14961 TOPDXPTR(nss,ix) = DPTR2FPTR(void (*)(pTHX_ void*),
14962 any_dup(FPTR2DPTR(void *, dxptr),
14965 case SAVEt_REGCONTEXT:
14967 ix -= uv >> SAVE_TIGHT_SHIFT;
14969 case SAVEt_AELEM: /* array element */
14970 sv = (const SV *)POPPTR(ss,ix);
14971 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14973 TOPIV(nss,ix) = iv;
14974 av = (const AV *)POPPTR(ss,ix);
14975 TOPPTR(nss,ix) = av_dup_inc(av, param);
14978 ptr = POPPTR(ss,ix);
14979 TOPPTR(nss,ix) = ptr;
14982 ptr = POPPTR(ss,ix);
14983 ptr = cophh_copy((COPHH*)ptr);
14984 TOPPTR(nss,ix) = ptr;
14986 TOPINT(nss,ix) = i;
14987 if (i & HINT_LOCALIZE_HH) {
14988 hv = (const HV *)POPPTR(ss,ix);
14989 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14992 case SAVEt_PADSV_AND_MORTALIZE:
14993 longval = (long)POPLONG(ss,ix);
14994 TOPLONG(nss,ix) = longval;
14995 ptr = POPPTR(ss,ix);
14996 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14997 sv = (const SV *)POPPTR(ss,ix);
14998 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
15000 case SAVEt_SET_SVFLAGS:
15002 TOPINT(nss,ix) = i;
15004 TOPINT(nss,ix) = i;
15005 sv = (const SV *)POPPTR(ss,ix);
15006 TOPPTR(nss,ix) = sv_dup(sv, param);
15008 case SAVEt_COMPILE_WARNINGS:
15009 ptr = POPPTR(ss,ix);
15010 TOPPTR(nss,ix) = DUP_WARNINGS((STRLEN*)ptr);
15013 ptr = POPPTR(ss,ix);
15014 TOPPTR(nss,ix) = parser_dup((const yy_parser*)ptr, param);
15018 "panic: ss_dup inconsistency (%" IVdf ")", (IV) type);
15026 /* if sv is a stash, call $class->CLONE_SKIP(), and set the SVphv_CLONEABLE
15027 * flag to the result. This is done for each stash before cloning starts,
15028 * so we know which stashes want their objects cloned */
15031 do_mark_cloneable_stash(pTHX_ SV *const sv)
15033 const HEK * const hvname = HvNAME_HEK((const HV *)sv);
15035 GV* const cloner = gv_fetchmethod_autoload(MUTABLE_HV(sv), "CLONE_SKIP", 0);
15036 SvFLAGS(sv) |= SVphv_CLONEABLE; /* clone objects by default */
15037 if (cloner && GvCV(cloner)) {
15044 mXPUSHs(newSVhek(hvname));
15046 call_sv(MUTABLE_SV(GvCV(cloner)), G_SCALAR);
15053 SvFLAGS(sv) &= ~SVphv_CLONEABLE;
15061 =for apidoc perl_clone
15063 Create and return a new interpreter by cloning the current one.
15065 C<perl_clone> takes these flags as parameters:
15067 C<CLONEf_COPY_STACKS> - is used to, well, copy the stacks also,
15068 without it we only clone the data and zero the stacks,
15069 with it we copy the stacks and the new perl interpreter is
15070 ready to run at the exact same point as the previous one.
15071 The pseudo-fork code uses C<COPY_STACKS> while the
15072 threads->create doesn't.
15074 C<CLONEf_KEEP_PTR_TABLE> -
15075 C<perl_clone> keeps a ptr_table with the pointer of the old
15076 variable as a key and the new variable as a value,
15077 this allows it to check if something has been cloned and not
15078 clone it again but rather just use the value and increase the
15079 refcount. If C<KEEP_PTR_TABLE> is not set then C<perl_clone> will kill
15080 the ptr_table using the function
15081 C<ptr_table_free(PL_ptr_table); PL_ptr_table = NULL;>,
15082 reason to keep it around is if you want to dup some of your own
15083 variable who are outside the graph perl scans, an example of this
15084 code is in F<threads.xs> create.
15086 C<CLONEf_CLONE_HOST> -
15087 This is a win32 thing, it is ignored on unix, it tells perls
15088 win32host code (which is c++) to clone itself, this is needed on
15089 win32 if you want to run two threads at the same time,
15090 if you just want to do some stuff in a separate perl interpreter
15091 and then throw it away and return to the original one,
15092 you don't need to do anything.
15097 /* XXX the above needs expanding by someone who actually understands it ! */
15098 EXTERN_C PerlInterpreter *
15099 perl_clone_host(PerlInterpreter* proto_perl, UV flags);
15102 perl_clone(PerlInterpreter *proto_perl, UV flags)
15105 #ifdef PERL_IMPLICIT_SYS
15107 PERL_ARGS_ASSERT_PERL_CLONE;
15109 /* perlhost.h so we need to call into it
15110 to clone the host, CPerlHost should have a c interface, sky */
15112 #ifndef __amigaos4__
15113 if (flags & CLONEf_CLONE_HOST) {
15114 return perl_clone_host(proto_perl,flags);
15117 return perl_clone_using(proto_perl, flags,
15119 proto_perl->IMemShared,
15120 proto_perl->IMemParse,
15122 proto_perl->IStdIO,
15126 proto_perl->IProc);
15130 perl_clone_using(PerlInterpreter *proto_perl, UV flags,
15131 struct IPerlMem* ipM, struct IPerlMem* ipMS,
15132 struct IPerlMem* ipMP, struct IPerlEnv* ipE,
15133 struct IPerlStdIO* ipStd, struct IPerlLIO* ipLIO,
15134 struct IPerlDir* ipD, struct IPerlSock* ipS,
15135 struct IPerlProc* ipP)
15137 /* XXX many of the string copies here can be optimized if they're
15138 * constants; they need to be allocated as common memory and just
15139 * their pointers copied. */
15142 CLONE_PARAMS clone_params;
15143 CLONE_PARAMS* const param = &clone_params;
15145 PerlInterpreter * const my_perl = (PerlInterpreter*)(*ipM->pMalloc)(ipM, sizeof(PerlInterpreter));
15147 PERL_ARGS_ASSERT_PERL_CLONE_USING;
15148 #else /* !PERL_IMPLICIT_SYS */
15150 CLONE_PARAMS clone_params;
15151 CLONE_PARAMS* param = &clone_params;
15152 PerlInterpreter * const my_perl = (PerlInterpreter*)PerlMem_malloc(sizeof(PerlInterpreter));
15154 PERL_ARGS_ASSERT_PERL_CLONE;
15155 #endif /* PERL_IMPLICIT_SYS */
15157 /* for each stash, determine whether its objects should be cloned */
15158 S_visit(proto_perl, do_mark_cloneable_stash, SVt_PVHV, SVTYPEMASK);
15159 PERL_SET_THX(my_perl);
15162 PoisonNew(my_perl, 1, PerlInterpreter);
15165 PL_defstash = NULL; /* may be used by perl malloc() */
15168 PL_scopestack_name = 0;
15170 PL_savestack_ix = 0;
15171 PL_savestack_max = -1;
15172 PL_sig_pending = 0;
15174 Zero(&PL_debug_pad, 1, struct perl_debug_pad);
15175 Zero(&PL_padname_undef, 1, PADNAME);
15176 Zero(&PL_padname_const, 1, PADNAME);
15177 # ifdef DEBUG_LEAKING_SCALARS
15178 PL_sv_serial = (((UV)my_perl >> 2) & 0xfff) * 1000000;
15180 # ifdef PERL_TRACE_OPS
15181 Zero(PL_op_exec_cnt, OP_max+2, UV);
15183 #else /* !DEBUGGING */
15184 Zero(my_perl, 1, PerlInterpreter);
15185 #endif /* DEBUGGING */
15187 #ifdef PERL_IMPLICIT_SYS
15188 /* host pointers */
15190 PL_MemShared = ipMS;
15191 PL_MemParse = ipMP;
15198 #endif /* PERL_IMPLICIT_SYS */
15201 param->flags = flags;
15202 /* Nothing in the core code uses this, but we make it available to
15203 extensions (using mg_dup). */
15204 param->proto_perl = proto_perl;
15205 /* Likely nothing will use this, but it is initialised to be consistent
15206 with Perl_clone_params_new(). */
15207 param->new_perl = my_perl;
15208 param->unreferenced = NULL;
15211 INIT_TRACK_MEMPOOL(my_perl->Imemory_debug_header, my_perl);
15213 PL_body_arenas = NULL;
15214 Zero(&PL_body_roots, 1, PL_body_roots);
15218 PL_sv_arenaroot = NULL;
15220 PL_debug = proto_perl->Idebug;
15222 /* dbargs array probably holds garbage */
15225 PL_compiling = proto_perl->Icompiling;
15227 /* pseudo environmental stuff */
15228 PL_origargc = proto_perl->Iorigargc;
15229 PL_origargv = proto_perl->Iorigargv;
15231 #ifndef NO_TAINT_SUPPORT
15232 /* Set tainting stuff before PerlIO_debug can possibly get called */
15233 PL_tainting = proto_perl->Itainting;
15234 PL_taint_warn = proto_perl->Itaint_warn;
15236 PL_tainting = FALSE;
15237 PL_taint_warn = FALSE;
15240 PL_minus_c = proto_perl->Iminus_c;
15242 PL_localpatches = proto_perl->Ilocalpatches;
15243 PL_splitstr = proto_perl->Isplitstr;
15244 PL_minus_n = proto_perl->Iminus_n;
15245 PL_minus_p = proto_perl->Iminus_p;
15246 PL_minus_l = proto_perl->Iminus_l;
15247 PL_minus_a = proto_perl->Iminus_a;
15248 PL_minus_E = proto_perl->Iminus_E;
15249 PL_minus_F = proto_perl->Iminus_F;
15250 PL_doswitches = proto_perl->Idoswitches;
15251 PL_dowarn = proto_perl->Idowarn;
15252 #ifdef PERL_SAWAMPERSAND
15253 PL_sawampersand = proto_perl->Isawampersand;
15255 PL_unsafe = proto_perl->Iunsafe;
15256 PL_perldb = proto_perl->Iperldb;
15257 PL_perl_destruct_level = proto_perl->Iperl_destruct_level;
15258 PL_exit_flags = proto_perl->Iexit_flags;
15260 /* XXX time(&PL_basetime) when asked for? */
15261 PL_basetime = proto_perl->Ibasetime;
15263 PL_maxsysfd = proto_perl->Imaxsysfd;
15264 PL_statusvalue = proto_perl->Istatusvalue;
15266 PL_statusvalue_vms = proto_perl->Istatusvalue_vms;
15268 PL_statusvalue_posix = proto_perl->Istatusvalue_posix;
15271 /* RE engine related */
15272 PL_regmatch_slab = NULL;
15273 PL_reg_curpm = NULL;
15275 PL_sub_generation = proto_perl->Isub_generation;
15277 /* funky return mechanisms */
15278 PL_forkprocess = proto_perl->Iforkprocess;
15280 /* internal state */
15281 PL_main_start = proto_perl->Imain_start;
15282 PL_eval_root = proto_perl->Ieval_root;
15283 PL_eval_start = proto_perl->Ieval_start;
15285 PL_filemode = proto_perl->Ifilemode;
15286 PL_lastfd = proto_perl->Ilastfd;
15287 PL_oldname = proto_perl->Ioldname; /* XXX not quite right */
15290 PL_gensym = proto_perl->Igensym;
15292 PL_laststatval = proto_perl->Ilaststatval;
15293 PL_laststype = proto_perl->Ilaststype;
15296 PL_profiledata = NULL;
15298 PL_generation = proto_perl->Igeneration;
15300 PL_in_clean_objs = proto_perl->Iin_clean_objs;
15301 PL_in_clean_all = proto_perl->Iin_clean_all;
15303 PL_delaymagic_uid = proto_perl->Idelaymagic_uid;
15304 PL_delaymagic_euid = proto_perl->Idelaymagic_euid;
15305 PL_delaymagic_gid = proto_perl->Idelaymagic_gid;
15306 PL_delaymagic_egid = proto_perl->Idelaymagic_egid;
15307 PL_nomemok = proto_perl->Inomemok;
15308 PL_an = proto_perl->Ian;
15309 PL_evalseq = proto_perl->Ievalseq;
15310 PL_origenviron = proto_perl->Iorigenviron; /* XXX not quite right */
15311 PL_origalen = proto_perl->Iorigalen;
15313 PL_sighandlerp = proto_perl->Isighandlerp;
15315 PL_runops = proto_perl->Irunops;
15317 PL_subline = proto_perl->Isubline;
15319 PL_cv_has_eval = proto_perl->Icv_has_eval;
15322 PL_cryptseen = proto_perl->Icryptseen;
15325 #ifdef USE_LOCALE_COLLATE
15326 PL_collation_ix = proto_perl->Icollation_ix;
15327 PL_collation_standard = proto_perl->Icollation_standard;
15328 PL_collxfrm_base = proto_perl->Icollxfrm_base;
15329 PL_collxfrm_mult = proto_perl->Icollxfrm_mult;
15330 PL_strxfrm_max_cp = proto_perl->Istrxfrm_max_cp;
15331 #endif /* USE_LOCALE_COLLATE */
15333 #ifdef USE_LOCALE_NUMERIC
15334 PL_numeric_standard = proto_perl->Inumeric_standard;
15335 PL_numeric_underlying = proto_perl->Inumeric_underlying;
15336 #endif /* !USE_LOCALE_NUMERIC */
15338 /* Did the locale setup indicate UTF-8? */
15339 PL_utf8locale = proto_perl->Iutf8locale;
15340 PL_in_utf8_CTYPE_locale = proto_perl->Iin_utf8_CTYPE_locale;
15341 PL_in_utf8_COLLATE_locale = proto_perl->Iin_utf8_COLLATE_locale;
15342 /* Unicode features (see perlrun/-C) */
15343 PL_unicode = proto_perl->Iunicode;
15345 /* Pre-5.8 signals control */
15346 PL_signals = proto_perl->Isignals;
15348 /* times() ticks per second */
15349 PL_clocktick = proto_perl->Iclocktick;
15351 /* Recursion stopper for PerlIO_find_layer */
15352 PL_in_load_module = proto_perl->Iin_load_module;
15354 /* sort() routine */
15355 PL_sort_RealCmp = proto_perl->Isort_RealCmp;
15357 /* Not really needed/useful since the reenrant_retint is "volatile",
15358 * but do it for consistency's sake. */
15359 PL_reentrant_retint = proto_perl->Ireentrant_retint;
15361 /* Hooks to shared SVs and locks. */
15362 PL_sharehook = proto_perl->Isharehook;
15363 PL_lockhook = proto_perl->Ilockhook;
15364 PL_unlockhook = proto_perl->Iunlockhook;
15365 PL_threadhook = proto_perl->Ithreadhook;
15366 PL_destroyhook = proto_perl->Idestroyhook;
15367 PL_signalhook = proto_perl->Isignalhook;
15369 PL_globhook = proto_perl->Iglobhook;
15372 PL_last_swash_hv = NULL; /* reinits on demand */
15373 PL_last_swash_klen = 0;
15374 PL_last_swash_key[0]= '\0';
15375 PL_last_swash_tmps = (U8*)NULL;
15376 PL_last_swash_slen = 0;
15378 PL_srand_called = proto_perl->Isrand_called;
15379 Copy(&(proto_perl->Irandom_state), &PL_random_state, 1, PL_RANDOM_STATE_TYPE);
15381 if (flags & CLONEf_COPY_STACKS) {
15382 /* next allocation will be PL_tmps_stack[PL_tmps_ix+1] */
15383 PL_tmps_ix = proto_perl->Itmps_ix;
15384 PL_tmps_max = proto_perl->Itmps_max;
15385 PL_tmps_floor = proto_perl->Itmps_floor;
15387 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
15388 * NOTE: unlike the others! */
15389 PL_scopestack_ix = proto_perl->Iscopestack_ix;
15390 PL_scopestack_max = proto_perl->Iscopestack_max;
15392 /* next SSPUSHFOO() sets PL_savestack[PL_savestack_ix]
15393 * NOTE: unlike the others! */
15394 PL_savestack_ix = proto_perl->Isavestack_ix;
15395 PL_savestack_max = proto_perl->Isavestack_max;
15398 PL_start_env = proto_perl->Istart_env; /* XXXXXX */
15399 PL_top_env = &PL_start_env;
15401 PL_op = proto_perl->Iop;
15404 PL_Xpv = (XPV*)NULL;
15405 my_perl->Ina = proto_perl->Ina;
15407 PL_statcache = proto_perl->Istatcache;
15409 #ifndef NO_TAINT_SUPPORT
15410 PL_tainted = proto_perl->Itainted;
15412 PL_tainted = FALSE;
15414 PL_curpm = proto_perl->Icurpm; /* XXX No PMOP ref count */
15416 PL_chopset = proto_perl->Ichopset; /* XXX never deallocated */
15418 PL_restartjmpenv = proto_perl->Irestartjmpenv;
15419 PL_restartop = proto_perl->Irestartop;
15420 PL_in_eval = proto_perl->Iin_eval;
15421 PL_delaymagic = proto_perl->Idelaymagic;
15422 PL_phase = proto_perl->Iphase;
15423 PL_localizing = proto_perl->Ilocalizing;
15425 PL_hv_fetch_ent_mh = NULL;
15426 PL_modcount = proto_perl->Imodcount;
15427 PL_lastgotoprobe = NULL;
15428 PL_dumpindent = proto_perl->Idumpindent;
15430 PL_efloatbuf = NULL; /* reinits on demand */
15431 PL_efloatsize = 0; /* reinits on demand */
15435 PL_colorset = 0; /* reinits PL_colors[] */
15436 /*PL_colors[6] = {0,0,0,0,0,0};*/
15438 /* Pluggable optimizer */
15439 PL_peepp = proto_perl->Ipeepp;
15440 PL_rpeepp = proto_perl->Irpeepp;
15441 /* op_free() hook */
15442 PL_opfreehook = proto_perl->Iopfreehook;
15444 #ifdef USE_REENTRANT_API
15445 /* XXX: things like -Dm will segfault here in perlio, but doing
15446 * PERL_SET_CONTEXT(proto_perl);
15447 * breaks too many other things
15449 Perl_reentrant_init(aTHX);
15452 /* create SV map for pointer relocation */
15453 PL_ptr_table = ptr_table_new();
15455 /* initialize these special pointers as early as possible */
15457 ptr_table_store(PL_ptr_table, &proto_perl->Isv_undef, &PL_sv_undef);
15458 ptr_table_store(PL_ptr_table, &proto_perl->Isv_no, &PL_sv_no);
15459 ptr_table_store(PL_ptr_table, &proto_perl->Isv_zero, &PL_sv_zero);
15460 ptr_table_store(PL_ptr_table, &proto_perl->Isv_yes, &PL_sv_yes);
15461 ptr_table_store(PL_ptr_table, &proto_perl->Ipadname_const,
15462 &PL_padname_const);
15464 /* create (a non-shared!) shared string table */
15465 PL_strtab = newHV();
15466 HvSHAREKEYS_off(PL_strtab);
15467 hv_ksplit(PL_strtab, HvTOTALKEYS(proto_perl->Istrtab));
15468 ptr_table_store(PL_ptr_table, proto_perl->Istrtab, PL_strtab);
15470 Zero(PL_sv_consts, SV_CONSTS_COUNT, SV*);
15472 /* This PV will be free'd special way so must set it same way op.c does */
15473 PL_compiling.cop_file = savesharedpv(PL_compiling.cop_file);
15474 ptr_table_store(PL_ptr_table, proto_perl->Icompiling.cop_file, PL_compiling.cop_file);
15476 ptr_table_store(PL_ptr_table, &proto_perl->Icompiling, &PL_compiling);
15477 PL_compiling.cop_warnings = DUP_WARNINGS(PL_compiling.cop_warnings);
15478 CopHINTHASH_set(&PL_compiling, cophh_copy(CopHINTHASH_get(&PL_compiling)));
15479 PL_curcop = (COP*)any_dup(proto_perl->Icurcop, proto_perl);
15481 param->stashes = newAV(); /* Setup array of objects to call clone on */
15482 /* This makes no difference to the implementation, as it always pushes
15483 and shifts pointers to other SVs without changing their reference
15484 count, with the array becoming empty before it is freed. However, it
15485 makes it conceptually clear what is going on, and will avoid some
15486 work inside av.c, filling slots between AvFILL() and AvMAX() with
15487 &PL_sv_undef, and SvREFCNT_dec()ing those. */
15488 AvREAL_off(param->stashes);
15490 if (!(flags & CLONEf_COPY_STACKS)) {
15491 param->unreferenced = newAV();
15494 #ifdef PERLIO_LAYERS
15495 /* Clone PerlIO tables as soon as we can handle general xx_dup() */
15496 PerlIO_clone(aTHX_ proto_perl, param);
15499 PL_envgv = gv_dup_inc(proto_perl->Ienvgv, param);
15500 PL_incgv = gv_dup_inc(proto_perl->Iincgv, param);
15501 PL_hintgv = gv_dup_inc(proto_perl->Ihintgv, param);
15502 PL_origfilename = SAVEPV(proto_perl->Iorigfilename);
15503 PL_xsubfilename = proto_perl->Ixsubfilename;
15504 PL_diehook = sv_dup_inc(proto_perl->Idiehook, param);
15505 PL_warnhook = sv_dup_inc(proto_perl->Iwarnhook, param);
15508 PL_patchlevel = sv_dup_inc(proto_perl->Ipatchlevel, param);
15509 PL_inplace = SAVEPV(proto_perl->Iinplace);
15510 PL_e_script = sv_dup_inc(proto_perl->Ie_script, param);
15512 /* magical thingies */
15514 SvPVCLEAR(PERL_DEBUG_PAD(0)); /* For regex debugging. */
15515 SvPVCLEAR(PERL_DEBUG_PAD(1)); /* ext/re needs these */
15516 SvPVCLEAR(PERL_DEBUG_PAD(2)); /* even without DEBUGGING. */
15519 /* Clone the regex array */
15520 /* ORANGE FIXME for plugins, probably in the SV dup code.
15521 newSViv(PTR2IV(CALLREGDUPE(
15522 INT2PTR(REGEXP *, SvIVX(regex)), param))))
15524 PL_regex_padav = av_dup_inc(proto_perl->Iregex_padav, param);
15525 PL_regex_pad = AvARRAY(PL_regex_padav);
15527 PL_stashpadmax = proto_perl->Istashpadmax;
15528 PL_stashpadix = proto_perl->Istashpadix ;
15529 Newx(PL_stashpad, PL_stashpadmax, HV *);
15532 for (; o < PL_stashpadmax; ++o)
15533 PL_stashpad[o] = hv_dup(proto_perl->Istashpad[o], param);
15536 /* shortcuts to various I/O objects */
15537 PL_ofsgv = gv_dup_inc(proto_perl->Iofsgv, param);
15538 PL_stdingv = gv_dup(proto_perl->Istdingv, param);
15539 PL_stderrgv = gv_dup(proto_perl->Istderrgv, param);
15540 PL_defgv = gv_dup(proto_perl->Idefgv, param);
15541 PL_argvgv = gv_dup_inc(proto_perl->Iargvgv, param);
15542 PL_argvoutgv = gv_dup(proto_perl->Iargvoutgv, param);
15543 PL_argvout_stack = av_dup_inc(proto_perl->Iargvout_stack, param);
15545 /* shortcuts to regexp stuff */
15546 PL_replgv = gv_dup_inc(proto_perl->Ireplgv, param);
15548 /* shortcuts to misc objects */
15549 PL_errgv = gv_dup(proto_perl->Ierrgv, param);
15551 /* shortcuts to debugging objects */
15552 PL_DBgv = gv_dup_inc(proto_perl->IDBgv, param);
15553 PL_DBline = gv_dup_inc(proto_perl->IDBline, param);
15554 PL_DBsub = gv_dup_inc(proto_perl->IDBsub, param);
15555 PL_DBsingle = sv_dup(proto_perl->IDBsingle, param);
15556 PL_DBtrace = sv_dup(proto_perl->IDBtrace, param);
15557 PL_DBsignal = sv_dup(proto_perl->IDBsignal, param);
15558 Copy(proto_perl->IDBcontrol, PL_DBcontrol, DBVARMG_COUNT, IV);
15560 /* symbol tables */
15561 PL_defstash = hv_dup_inc(proto_perl->Idefstash, param);
15562 PL_curstash = hv_dup_inc(proto_perl->Icurstash, param);
15563 PL_debstash = hv_dup(proto_perl->Idebstash, param);
15564 PL_globalstash = hv_dup(proto_perl->Iglobalstash, param);
15565 PL_curstname = sv_dup_inc(proto_perl->Icurstname, param);
15567 PL_beginav = av_dup_inc(proto_perl->Ibeginav, param);
15568 PL_beginav_save = av_dup_inc(proto_perl->Ibeginav_save, param);
15569 PL_checkav_save = av_dup_inc(proto_perl->Icheckav_save, param);
15570 PL_unitcheckav = av_dup_inc(proto_perl->Iunitcheckav, param);
15571 PL_unitcheckav_save = av_dup_inc(proto_perl->Iunitcheckav_save, param);
15572 PL_endav = av_dup_inc(proto_perl->Iendav, param);
15573 PL_checkav = av_dup_inc(proto_perl->Icheckav, param);
15574 PL_initav = av_dup_inc(proto_perl->Iinitav, param);
15575 PL_savebegin = proto_perl->Isavebegin;
15577 PL_isarev = hv_dup_inc(proto_perl->Iisarev, param);
15579 /* subprocess state */
15580 PL_fdpid = av_dup_inc(proto_perl->Ifdpid, param);
15582 if (proto_perl->Iop_mask)
15583 PL_op_mask = SAVEPVN(proto_perl->Iop_mask, PL_maxo);
15586 /* PL_asserting = proto_perl->Iasserting; */
15588 /* current interpreter roots */
15589 PL_main_cv = cv_dup_inc(proto_perl->Imain_cv, param);
15591 PL_main_root = OpREFCNT_inc(proto_perl->Imain_root);
15594 /* runtime control stuff */
15595 PL_curcopdb = (COP*)any_dup(proto_perl->Icurcopdb, proto_perl);
15597 PL_preambleav = av_dup_inc(proto_perl->Ipreambleav, param);
15599 PL_ors_sv = sv_dup_inc(proto_perl->Iors_sv, param);
15601 /* interpreter atexit processing */
15602 PL_exitlistlen = proto_perl->Iexitlistlen;
15603 if (PL_exitlistlen) {
15604 Newx(PL_exitlist, PL_exitlistlen, PerlExitListEntry);
15605 Copy(proto_perl->Iexitlist, PL_exitlist, PL_exitlistlen, PerlExitListEntry);
15608 PL_exitlist = (PerlExitListEntry*)NULL;
15610 PL_my_cxt_size = proto_perl->Imy_cxt_size;
15611 if (PL_my_cxt_size) {
15612 Newx(PL_my_cxt_list, PL_my_cxt_size, void *);
15613 Copy(proto_perl->Imy_cxt_list, PL_my_cxt_list, PL_my_cxt_size, void *);
15614 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
15615 Newx(PL_my_cxt_keys, PL_my_cxt_size, const char *);
15616 Copy(proto_perl->Imy_cxt_keys, PL_my_cxt_keys, PL_my_cxt_size, char *);
15620 PL_my_cxt_list = (void**)NULL;
15621 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
15622 PL_my_cxt_keys = (const char**)NULL;
15625 PL_modglobal = hv_dup_inc(proto_perl->Imodglobal, param);
15626 PL_custom_op_names = hv_dup_inc(proto_perl->Icustom_op_names,param);
15627 PL_custom_op_descs = hv_dup_inc(proto_perl->Icustom_op_descs,param);
15628 PL_custom_ops = hv_dup_inc(proto_perl->Icustom_ops, param);
15630 PL_compcv = cv_dup(proto_perl->Icompcv, param);
15632 PAD_CLONE_VARS(proto_perl, param);
15634 #ifdef HAVE_INTERP_INTERN
15635 sys_intern_dup(&proto_perl->Isys_intern, &PL_sys_intern);
15638 PL_DBcv = cv_dup(proto_perl->IDBcv, param);
15640 #ifdef PERL_USES_PL_PIDSTATUS
15641 PL_pidstatus = newHV(); /* XXX flag for cloning? */
15643 PL_osname = SAVEPV(proto_perl->Iosname);
15644 PL_parser = parser_dup(proto_perl->Iparser, param);
15646 /* XXX this only works if the saved cop has already been cloned */
15647 if (proto_perl->Iparser) {
15648 PL_parser->saved_curcop = (COP*)any_dup(
15649 proto_perl->Iparser->saved_curcop,
15653 PL_subname = sv_dup_inc(proto_perl->Isubname, param);
15655 #ifdef USE_LOCALE_CTYPE
15656 /* Should we warn if uses locale? */
15657 PL_warn_locale = sv_dup_inc(proto_perl->Iwarn_locale, param);
15660 #ifdef USE_LOCALE_COLLATE
15661 PL_collation_name = SAVEPV(proto_perl->Icollation_name);
15662 #endif /* USE_LOCALE_COLLATE */
15664 #ifdef USE_LOCALE_NUMERIC
15665 PL_numeric_name = SAVEPV(proto_perl->Inumeric_name);
15666 PL_numeric_radix_sv = sv_dup_inc(proto_perl->Inumeric_radix_sv, param);
15667 #endif /* !USE_LOCALE_NUMERIC */
15669 PL_langinfo_buf = NULL;
15670 PL_langinfo_bufsize = 0;
15672 /* Unicode inversion lists */
15673 PL_Latin1 = sv_dup_inc(proto_perl->ILatin1, param);
15674 PL_UpperLatin1 = sv_dup_inc(proto_perl->IUpperLatin1, param);
15675 PL_AboveLatin1 = sv_dup_inc(proto_perl->IAboveLatin1, param);
15676 PL_InBitmap = sv_dup_inc(proto_perl->IInBitmap, param);
15678 PL_NonL1NonFinalFold = sv_dup_inc(proto_perl->INonL1NonFinalFold, param);
15679 PL_HasMultiCharFold = sv_dup_inc(proto_perl->IHasMultiCharFold, param);
15681 /* utf8 character class swashes */
15682 for (i = 0; i < POSIX_SWASH_COUNT; i++) {
15683 PL_utf8_swash_ptrs[i] = sv_dup_inc(proto_perl->Iutf8_swash_ptrs[i], param);
15685 for (i = 0; i < POSIX_CC_COUNT; i++) {
15686 PL_XPosix_ptrs[i] = sv_dup_inc(proto_perl->IXPosix_ptrs[i], param);
15688 PL_GCB_invlist = sv_dup_inc(proto_perl->IGCB_invlist, param);
15689 PL_SB_invlist = sv_dup_inc(proto_perl->ISB_invlist, param);
15690 PL_WB_invlist = sv_dup_inc(proto_perl->IWB_invlist, param);
15691 PL_seen_deprecated_macro = hv_dup_inc(proto_perl->Iseen_deprecated_macro, param);
15692 PL_utf8_mark = sv_dup_inc(proto_perl->Iutf8_mark, param);
15693 PL_utf8_toupper = sv_dup_inc(proto_perl->Iutf8_toupper, param);
15694 PL_utf8_totitle = sv_dup_inc(proto_perl->Iutf8_totitle, param);
15695 PL_utf8_tolower = sv_dup_inc(proto_perl->Iutf8_tolower, param);
15696 PL_utf8_tofold = sv_dup_inc(proto_perl->Iutf8_tofold, param);
15697 PL_utf8_idstart = sv_dup_inc(proto_perl->Iutf8_idstart, param);
15698 PL_utf8_xidstart = sv_dup_inc(proto_perl->Iutf8_xidstart, param);
15699 PL_utf8_perl_idstart = sv_dup_inc(proto_perl->Iutf8_perl_idstart, param);
15700 PL_utf8_perl_idcont = sv_dup_inc(proto_perl->Iutf8_perl_idcont, param);
15701 PL_utf8_idcont = sv_dup_inc(proto_perl->Iutf8_idcont, param);
15702 PL_utf8_xidcont = sv_dup_inc(proto_perl->Iutf8_xidcont, param);
15703 PL_utf8_foldable = sv_dup_inc(proto_perl->Iutf8_foldable, param);
15704 PL_utf8_charname_begin = sv_dup_inc(proto_perl->Iutf8_charname_begin, param);
15705 PL_utf8_charname_continue = sv_dup_inc(proto_perl->Iutf8_charname_continue, param);
15707 if (proto_perl->Ipsig_pend) {
15708 Newxz(PL_psig_pend, SIG_SIZE, int);
15711 PL_psig_pend = (int*)NULL;
15714 if (proto_perl->Ipsig_name) {
15715 Newx(PL_psig_name, 2 * SIG_SIZE, SV*);
15716 sv_dup_inc_multiple(proto_perl->Ipsig_name, PL_psig_name, 2 * SIG_SIZE,
15718 PL_psig_ptr = PL_psig_name + SIG_SIZE;
15721 PL_psig_ptr = (SV**)NULL;
15722 PL_psig_name = (SV**)NULL;
15725 if (flags & CLONEf_COPY_STACKS) {
15726 Newx(PL_tmps_stack, PL_tmps_max, SV*);
15727 sv_dup_inc_multiple(proto_perl->Itmps_stack, PL_tmps_stack,
15728 PL_tmps_ix+1, param);
15730 /* next PUSHMARK() sets *(PL_markstack_ptr+1) */
15731 i = proto_perl->Imarkstack_max - proto_perl->Imarkstack;
15732 Newx(PL_markstack, i, I32);
15733 PL_markstack_max = PL_markstack + (proto_perl->Imarkstack_max
15734 - proto_perl->Imarkstack);
15735 PL_markstack_ptr = PL_markstack + (proto_perl->Imarkstack_ptr
15736 - proto_perl->Imarkstack);
15737 Copy(proto_perl->Imarkstack, PL_markstack,
15738 PL_markstack_ptr - PL_markstack + 1, I32);
15740 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
15741 * NOTE: unlike the others! */
15742 Newx(PL_scopestack, PL_scopestack_max, I32);
15743 Copy(proto_perl->Iscopestack, PL_scopestack, PL_scopestack_ix, I32);
15746 Newx(PL_scopestack_name, PL_scopestack_max, const char *);
15747 Copy(proto_perl->Iscopestack_name, PL_scopestack_name, PL_scopestack_ix, const char *);
15749 /* reset stack AV to correct length before its duped via
15750 * PL_curstackinfo */
15751 AvFILLp(proto_perl->Icurstack) =
15752 proto_perl->Istack_sp - proto_perl->Istack_base;
15754 /* NOTE: si_dup() looks at PL_markstack */
15755 PL_curstackinfo = si_dup(proto_perl->Icurstackinfo, param);
15757 /* PL_curstack = PL_curstackinfo->si_stack; */
15758 PL_curstack = av_dup(proto_perl->Icurstack, param);
15759 PL_mainstack = av_dup(proto_perl->Imainstack, param);
15761 /* next PUSHs() etc. set *(PL_stack_sp+1) */
15762 PL_stack_base = AvARRAY(PL_curstack);
15763 PL_stack_sp = PL_stack_base + (proto_perl->Istack_sp
15764 - proto_perl->Istack_base);
15765 PL_stack_max = PL_stack_base + AvMAX(PL_curstack);
15767 /*Newxz(PL_savestack, PL_savestack_max, ANY);*/
15768 PL_savestack = ss_dup(proto_perl, param);
15772 ENTER; /* perl_destruct() wants to LEAVE; */
15775 PL_statgv = gv_dup(proto_perl->Istatgv, param);
15776 PL_statname = sv_dup_inc(proto_perl->Istatname, param);
15778 PL_rs = sv_dup_inc(proto_perl->Irs, param);
15779 PL_last_in_gv = gv_dup(proto_perl->Ilast_in_gv, param);
15780 PL_defoutgv = gv_dup_inc(proto_perl->Idefoutgv, param);
15781 PL_toptarget = sv_dup_inc(proto_perl->Itoptarget, param);
15782 PL_bodytarget = sv_dup_inc(proto_perl->Ibodytarget, param);
15783 PL_formtarget = sv_dup(proto_perl->Iformtarget, param);
15785 PL_errors = sv_dup_inc(proto_perl->Ierrors, param);
15787 PL_sortcop = (OP*)any_dup(proto_perl->Isortcop, proto_perl);
15788 PL_firstgv = gv_dup_inc(proto_perl->Ifirstgv, param);
15789 PL_secondgv = gv_dup_inc(proto_perl->Isecondgv, param);
15791 PL_stashcache = newHV();
15793 PL_watchaddr = (char **) ptr_table_fetch(PL_ptr_table,
15794 proto_perl->Iwatchaddr);
15795 PL_watchok = PL_watchaddr ? * PL_watchaddr : NULL;
15796 if (PL_debug && PL_watchaddr) {
15797 PerlIO_printf(Perl_debug_log,
15798 "WATCHING: %" UVxf " cloned as %" UVxf " with value %" UVxf "\n",
15799 PTR2UV(proto_perl->Iwatchaddr), PTR2UV(PL_watchaddr),
15800 PTR2UV(PL_watchok));
15803 PL_registered_mros = hv_dup_inc(proto_perl->Iregistered_mros, param);
15804 PL_blockhooks = av_dup_inc(proto_perl->Iblockhooks, param);
15805 PL_utf8_foldclosures = hv_dup_inc(proto_perl->Iutf8_foldclosures, param);
15807 /* Call the ->CLONE method, if it exists, for each of the stashes
15808 identified by sv_dup() above.
15810 while(av_tindex(param->stashes) != -1) {
15811 HV* const stash = MUTABLE_HV(av_shift(param->stashes));
15812 GV* const cloner = gv_fetchmethod_autoload(stash, "CLONE", 0);
15813 if (cloner && GvCV(cloner)) {
15818 mXPUSHs(newSVhek(HvNAME_HEK(stash)));
15820 call_sv(MUTABLE_SV(GvCV(cloner)), G_DISCARD);
15826 if (!(flags & CLONEf_KEEP_PTR_TABLE)) {
15827 ptr_table_free(PL_ptr_table);
15828 PL_ptr_table = NULL;
15831 if (!(flags & CLONEf_COPY_STACKS)) {
15832 unreferenced_to_tmp_stack(param->unreferenced);
15835 SvREFCNT_dec(param->stashes);
15837 /* orphaned? eg threads->new inside BEGIN or use */
15838 if (PL_compcv && ! SvREFCNT(PL_compcv)) {
15839 SvREFCNT_inc_simple_void(PL_compcv);
15840 SAVEFREESV(PL_compcv);
15847 S_unreferenced_to_tmp_stack(pTHX_ AV *const unreferenced)
15849 PERL_ARGS_ASSERT_UNREFERENCED_TO_TMP_STACK;
15851 if (AvFILLp(unreferenced) > -1) {
15852 SV **svp = AvARRAY(unreferenced);
15853 SV **const last = svp + AvFILLp(unreferenced);
15857 if (SvREFCNT(*svp) == 1)
15859 } while (++svp <= last);
15861 EXTEND_MORTAL(count);
15862 svp = AvARRAY(unreferenced);
15865 if (SvREFCNT(*svp) == 1) {
15866 /* Our reference is the only one to this SV. This means that
15867 in this thread, the scalar effectively has a 0 reference.
15868 That doesn't work (cleanup never happens), so donate our
15869 reference to it onto the save stack. */
15870 PL_tmps_stack[++PL_tmps_ix] = *svp;
15872 /* As an optimisation, because we are already walking the
15873 entire array, instead of above doing either
15874 SvREFCNT_inc(*svp) or *svp = &PL_sv_undef, we can instead
15875 release our reference to the scalar, so that at the end of
15876 the array owns zero references to the scalars it happens to
15877 point to. We are effectively converting the array from
15878 AvREAL() on to AvREAL() off. This saves the av_clear()
15879 (triggered by the SvREFCNT_dec(unreferenced) below) from
15880 walking the array a second time. */
15881 SvREFCNT_dec(*svp);
15884 } while (++svp <= last);
15885 AvREAL_off(unreferenced);
15887 SvREFCNT_dec_NN(unreferenced);
15891 Perl_clone_params_del(CLONE_PARAMS *param)
15893 /* This seemingly funky ordering keeps the build with PERL_GLOBAL_STRUCT
15895 PerlInterpreter *const to = param->new_perl;
15897 PerlInterpreter *const was = PERL_GET_THX;
15899 PERL_ARGS_ASSERT_CLONE_PARAMS_DEL;
15905 SvREFCNT_dec(param->stashes);
15906 if (param->unreferenced)
15907 unreferenced_to_tmp_stack(param->unreferenced);
15917 Perl_clone_params_new(PerlInterpreter *const from, PerlInterpreter *const to)
15920 /* Need to play this game, as newAV() can call safesysmalloc(), and that
15921 does a dTHX; to get the context from thread local storage.
15922 FIXME - under PERL_CORE Newx(), Safefree() and friends should expand to
15923 a version that passes in my_perl. */
15924 PerlInterpreter *const was = PERL_GET_THX;
15925 CLONE_PARAMS *param;
15927 PERL_ARGS_ASSERT_CLONE_PARAMS_NEW;
15933 /* Given that we've set the context, we can do this unshared. */
15934 Newx(param, 1, CLONE_PARAMS);
15937 param->proto_perl = from;
15938 param->new_perl = to;
15939 param->stashes = (AV *)Perl_newSV_type(to, SVt_PVAV);
15940 AvREAL_off(param->stashes);
15941 param->unreferenced = (AV *)Perl_newSV_type(to, SVt_PVAV);
15949 #endif /* USE_ITHREADS */
15952 Perl_init_constants(pTHX)
15954 SvREFCNT(&PL_sv_undef) = SvREFCNT_IMMORTAL;
15955 SvFLAGS(&PL_sv_undef) = SVf_READONLY|SVf_PROTECT|SVt_NULL;
15956 SvANY(&PL_sv_undef) = NULL;
15958 SvANY(&PL_sv_no) = new_XPVNV();
15959 SvREFCNT(&PL_sv_no) = SvREFCNT_IMMORTAL;
15960 SvFLAGS(&PL_sv_no) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15961 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15964 SvANY(&PL_sv_yes) = new_XPVNV();
15965 SvREFCNT(&PL_sv_yes) = SvREFCNT_IMMORTAL;
15966 SvFLAGS(&PL_sv_yes) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15967 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15970 SvANY(&PL_sv_zero) = new_XPVNV();
15971 SvREFCNT(&PL_sv_zero) = SvREFCNT_IMMORTAL;
15972 SvFLAGS(&PL_sv_zero) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15973 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15977 SvPV_set(&PL_sv_no, (char*)PL_No);
15978 SvCUR_set(&PL_sv_no, 0);
15979 SvLEN_set(&PL_sv_no, 0);
15980 SvIV_set(&PL_sv_no, 0);
15981 SvNV_set(&PL_sv_no, 0);
15983 SvPV_set(&PL_sv_yes, (char*)PL_Yes);
15984 SvCUR_set(&PL_sv_yes, 1);
15985 SvLEN_set(&PL_sv_yes, 0);
15986 SvIV_set(&PL_sv_yes, 1);
15987 SvNV_set(&PL_sv_yes, 1);
15989 SvPV_set(&PL_sv_zero, (char*)PL_Zero);
15990 SvCUR_set(&PL_sv_zero, 1);
15991 SvLEN_set(&PL_sv_zero, 0);
15992 SvIV_set(&PL_sv_zero, 0);
15993 SvNV_set(&PL_sv_zero, 0);
15995 PadnamePV(&PL_padname_const) = (char *)PL_No;
15997 assert(SvIMMORTAL_INTERP(&PL_sv_yes));
15998 assert(SvIMMORTAL_INTERP(&PL_sv_undef));
15999 assert(SvIMMORTAL_INTERP(&PL_sv_no));
16000 assert(SvIMMORTAL_INTERP(&PL_sv_zero));
16002 assert(SvIMMORTAL(&PL_sv_yes));
16003 assert(SvIMMORTAL(&PL_sv_undef));
16004 assert(SvIMMORTAL(&PL_sv_no));
16005 assert(SvIMMORTAL(&PL_sv_zero));
16007 assert( SvIMMORTAL_TRUE(&PL_sv_yes));
16008 assert(!SvIMMORTAL_TRUE(&PL_sv_undef));
16009 assert(!SvIMMORTAL_TRUE(&PL_sv_no));
16010 assert(!SvIMMORTAL_TRUE(&PL_sv_zero));
16012 assert( SvTRUE_nomg_NN(&PL_sv_yes));
16013 assert(!SvTRUE_nomg_NN(&PL_sv_undef));
16014 assert(!SvTRUE_nomg_NN(&PL_sv_no));
16015 assert(!SvTRUE_nomg_NN(&PL_sv_zero));
16019 =head1 Unicode Support
16021 =for apidoc sv_recode_to_utf8
16023 C<encoding> is assumed to be an C<Encode> object, on entry the PV
16024 of C<sv> is assumed to be octets in that encoding, and C<sv>
16025 will be converted into Unicode (and UTF-8).
16027 If C<sv> already is UTF-8 (or if it is not C<POK>), or if C<encoding>
16028 is not a reference, nothing is done to C<sv>. If C<encoding> is not
16029 an C<Encode::XS> Encoding object, bad things will happen.
16030 (See F<cpan/Encode/encoding.pm> and L<Encode>.)
16032 The PV of C<sv> is returned.
16037 Perl_sv_recode_to_utf8(pTHX_ SV *sv, SV *encoding)
16039 PERL_ARGS_ASSERT_SV_RECODE_TO_UTF8;
16041 if (SvPOK(sv) && !SvUTF8(sv) && !IN_BYTES && SvROK(encoding)) {
16050 if (SvPADTMP(nsv)) {
16051 nsv = sv_newmortal();
16052 SvSetSV_nosteal(nsv, sv);
16061 Passing sv_yes is wrong - it needs to be or'ed set of constants
16062 for Encode::XS, while UTf-8 decode (currently) assumes a true value means
16063 remove converted chars from source.
16065 Both will default the value - let them.
16067 XPUSHs(&PL_sv_yes);
16070 call_method("decode", G_SCALAR);
16074 s = SvPV_const(uni, len);
16075 if (s != SvPVX_const(sv)) {
16076 SvGROW(sv, len + 1);
16077 Move(s, SvPVX(sv), len + 1, char);
16078 SvCUR_set(sv, len);
16083 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
16084 /* clear pos and any utf8 cache */
16085 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
16088 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
16089 magic_setutf8(sv,mg); /* clear UTF8 cache */
16094 return SvPOKp(sv) ? SvPVX(sv) : NULL;
16098 =for apidoc sv_cat_decode
16100 C<encoding> is assumed to be an C<Encode> object, the PV of C<ssv> is
16101 assumed to be octets in that encoding and decoding the input starts
16102 from the position which S<C<(PV + *offset)>> pointed to. C<dsv> will be
16103 concatenated with the decoded UTF-8 string from C<ssv>. Decoding will terminate
16104 when the string C<tstr> appears in decoding output or the input ends on
16105 the PV of C<ssv>. The value which C<offset> points will be modified
16106 to the last input position on C<ssv>.
16108 Returns TRUE if the terminator was found, else returns FALSE.
16113 Perl_sv_cat_decode(pTHX_ SV *dsv, SV *encoding,
16114 SV *ssv, int *offset, char *tstr, int tlen)
16118 PERL_ARGS_ASSERT_SV_CAT_DECODE;
16120 if (SvPOK(ssv) && SvPOK(dsv) && SvROK(encoding)) {
16131 offsv = newSViv(*offset);
16133 mPUSHp(tstr, tlen);
16135 call_method("cat_decode", G_SCALAR);
16137 ret = SvTRUE(TOPs);
16138 *offset = SvIV(offsv);
16144 Perl_croak(aTHX_ "Invalid argument to sv_cat_decode");
16149 /* ---------------------------------------------------------------------
16151 * support functions for report_uninit()
16154 /* the maxiumum size of array or hash where we will scan looking
16155 * for the undefined element that triggered the warning */
16157 #define FUV_MAX_SEARCH_SIZE 1000
16159 /* Look for an entry in the hash whose value has the same SV as val;
16160 * If so, return a mortal copy of the key. */
16163 S_find_hash_subscript(pTHX_ const HV *const hv, const SV *const val)
16169 PERL_ARGS_ASSERT_FIND_HASH_SUBSCRIPT;
16171 if (!hv || SvMAGICAL(hv) || !HvARRAY(hv) ||
16172 (HvTOTALKEYS(hv) > FUV_MAX_SEARCH_SIZE))
16175 array = HvARRAY(hv);
16177 for (i=HvMAX(hv); i>=0; i--) {
16179 for (entry = array[i]; entry; entry = HeNEXT(entry)) {
16180 if (HeVAL(entry) != val)
16182 if ( HeVAL(entry) == &PL_sv_undef ||
16183 HeVAL(entry) == &PL_sv_placeholder)
16187 if (HeKLEN(entry) == HEf_SVKEY)
16188 return sv_mortalcopy(HeKEY_sv(entry));
16189 return sv_2mortal(newSVhek(HeKEY_hek(entry)));
16195 /* Look for an entry in the array whose value has the same SV as val;
16196 * If so, return the index, otherwise return -1. */
16199 S_find_array_subscript(pTHX_ const AV *const av, const SV *const val)
16201 PERL_ARGS_ASSERT_FIND_ARRAY_SUBSCRIPT;
16203 if (!av || SvMAGICAL(av) || !AvARRAY(av) ||
16204 (AvFILLp(av) > FUV_MAX_SEARCH_SIZE))
16207 if (val != &PL_sv_undef) {
16208 SV ** const svp = AvARRAY(av);
16211 for (i=AvFILLp(av); i>=0; i--)
16218 /* varname(): return the name of a variable, optionally with a subscript.
16219 * If gv is non-zero, use the name of that global, along with gvtype (one
16220 * of "$", "@", "%"); otherwise use the name of the lexical at pad offset
16221 * targ. Depending on the value of the subscript_type flag, return:
16224 #define FUV_SUBSCRIPT_NONE 1 /* "@foo" */
16225 #define FUV_SUBSCRIPT_ARRAY 2 /* "$foo[aindex]" */
16226 #define FUV_SUBSCRIPT_HASH 3 /* "$foo{keyname}" */
16227 #define FUV_SUBSCRIPT_WITHIN 4 /* "within @foo" */
16230 Perl_varname(pTHX_ const GV *const gv, const char gvtype, PADOFFSET targ,
16231 const SV *const keyname, SSize_t aindex, int subscript_type)
16234 SV * const name = sv_newmortal();
16235 if (gv && isGV(gv)) {
16237 buffer[0] = gvtype;
16240 /* as gv_fullname4(), but add literal '^' for $^FOO names */
16242 gv_fullname4(name, gv, buffer, 0);
16244 if ((unsigned int)SvPVX(name)[1] <= 26) {
16246 buffer[1] = SvPVX(name)[1] + 'A' - 1;
16248 /* Swap the 1 unprintable control character for the 2 byte pretty
16249 version - ie substr($name, 1, 1) = $buffer; */
16250 sv_insert(name, 1, 1, buffer, 2);
16254 CV * const cv = gv ? ((CV *)gv) : find_runcv(NULL);
16257 assert(!cv || SvTYPE(cv) == SVt_PVCV || SvTYPE(cv) == SVt_PVFM);
16259 if (!cv || !CvPADLIST(cv))
16261 sv = padnamelist_fetch(PadlistNAMES(CvPADLIST(cv)), targ);
16262 sv_setpvn(name, PadnamePV(sv), PadnameLEN(sv));
16266 if (subscript_type == FUV_SUBSCRIPT_HASH) {
16267 SV * const sv = newSV(0);
16269 const char * const pv = SvPV_nomg_const((SV*)keyname, len);
16271 *SvPVX(name) = '$';
16272 Perl_sv_catpvf(aTHX_ name, "{%s}",
16273 pv_pretty(sv, pv, len, 32, NULL, NULL,
16274 PERL_PV_PRETTY_DUMP | PERL_PV_ESCAPE_UNI_DETECT ));
16275 SvREFCNT_dec_NN(sv);
16277 else if (subscript_type == FUV_SUBSCRIPT_ARRAY) {
16278 *SvPVX(name) = '$';
16279 Perl_sv_catpvf(aTHX_ name, "[%" IVdf "]", (IV)aindex);
16281 else if (subscript_type == FUV_SUBSCRIPT_WITHIN) {
16282 /* We know that name has no magic, so can use 0 instead of SV_GMAGIC */
16283 Perl_sv_insert_flags(aTHX_ name, 0, 0, STR_WITH_LEN("within "), 0);
16291 =for apidoc find_uninit_var
16293 Find the name of the undefined variable (if any) that caused the operator
16294 to issue a "Use of uninitialized value" warning.
16295 If match is true, only return a name if its value matches C<uninit_sv>.
16296 So roughly speaking, if a unary operator (such as C<OP_COS>) generates a
16297 warning, then following the direct child of the op may yield an
16298 C<OP_PADSV> or C<OP_GV> that gives the name of the undefined variable. On the
16299 other hand, with C<OP_ADD> there are two branches to follow, so we only print
16300 the variable name if we get an exact match.
16301 C<desc_p> points to a string pointer holding the description of the op.
16302 This may be updated if needed.
16304 The name is returned as a mortal SV.
16306 Assumes that C<PL_op> is the OP that originally triggered the error, and that
16307 C<PL_comppad>/C<PL_curpad> points to the currently executing pad.
16313 S_find_uninit_var(pTHX_ const OP *const obase, const SV *const uninit_sv,
16314 bool match, const char **desc_p)
16319 const OP *o, *o2, *kid;
16321 PERL_ARGS_ASSERT_FIND_UNINIT_VAR;
16323 if (!obase || (match && (!uninit_sv || uninit_sv == &PL_sv_undef ||
16324 uninit_sv == &PL_sv_placeholder)))
16327 switch (obase->op_type) {
16330 /* undef should care if its args are undef - any warnings
16331 * will be from tied/magic vars */
16339 const bool pad = ( obase->op_type == OP_PADAV
16340 || obase->op_type == OP_PADHV
16341 || obase->op_type == OP_PADRANGE
16344 const bool hash = ( obase->op_type == OP_PADHV
16345 || obase->op_type == OP_RV2HV
16346 || (obase->op_type == OP_PADRANGE
16347 && SvTYPE(PAD_SVl(obase->op_targ)) == SVt_PVHV)
16351 int subscript_type = FUV_SUBSCRIPT_WITHIN;
16353 if (pad) { /* @lex, %lex */
16354 sv = PAD_SVl(obase->op_targ);
16358 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
16359 /* @global, %global */
16360 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
16363 sv = hash ? MUTABLE_SV(GvHV(gv)): MUTABLE_SV(GvAV(gv));
16365 else if (obase == PL_op) /* @{expr}, %{expr} */
16366 return find_uninit_var(cUNOPx(obase)->op_first,
16367 uninit_sv, match, desc_p);
16368 else /* @{expr}, %{expr} as a sub-expression */
16372 /* attempt to find a match within the aggregate */
16374 keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16376 subscript_type = FUV_SUBSCRIPT_HASH;
16379 index = find_array_subscript((const AV *)sv, uninit_sv);
16381 subscript_type = FUV_SUBSCRIPT_ARRAY;
16384 if (match && subscript_type == FUV_SUBSCRIPT_WITHIN)
16387 return varname(gv, (char)(hash ? '%' : '@'), obase->op_targ,
16388 keysv, index, subscript_type);
16392 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
16394 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
16395 if (!gv || !GvSTASH(gv))
16397 if (match && (GvSV(gv) != uninit_sv))
16399 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
16402 return find_uninit_var(cUNOPx(obase)->op_first, uninit_sv, 1, desc_p);
16405 if (match && PAD_SVl(obase->op_targ) != uninit_sv)
16407 return varname(NULL, '$', obase->op_targ,
16408 NULL, 0, FUV_SUBSCRIPT_NONE);
16411 gv = cGVOPx_gv(obase);
16412 if (!gv || (match && GvSV(gv) != uninit_sv) || !GvSTASH(gv))
16414 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
16416 case OP_AELEMFAST_LEX:
16419 AV *av = MUTABLE_AV(PAD_SV(obase->op_targ));
16420 if (!av || SvRMAGICAL(av))
16422 svp = av_fetch(av, (I8)obase->op_private, FALSE);
16423 if (!svp || *svp != uninit_sv)
16426 return varname(NULL, '$', obase->op_targ,
16427 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
16430 gv = cGVOPx_gv(obase);
16435 AV *const av = GvAV(gv);
16436 if (!av || SvRMAGICAL(av))
16438 svp = av_fetch(av, (I8)obase->op_private, FALSE);
16439 if (!svp || *svp != uninit_sv)
16442 return varname(gv, '$', 0,
16443 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
16445 NOT_REACHED; /* NOTREACHED */
16448 o = cUNOPx(obase)->op_first;
16449 if (!o || o->op_type != OP_NULL ||
16450 ! (o->op_targ == OP_AELEM || o->op_targ == OP_HELEM))
16452 return find_uninit_var(cBINOPo->op_last, uninit_sv, match, desc_p);
16457 bool negate = FALSE;
16459 if (PL_op == obase)
16460 /* $a[uninit_expr] or $h{uninit_expr} */
16461 return find_uninit_var(cBINOPx(obase)->op_last,
16462 uninit_sv, match, desc_p);
16465 o = cBINOPx(obase)->op_first;
16466 kid = cBINOPx(obase)->op_last;
16468 /* get the av or hv, and optionally the gv */
16470 if (o->op_type == OP_PADAV || o->op_type == OP_PADHV) {
16471 sv = PAD_SV(o->op_targ);
16473 else if ((o->op_type == OP_RV2AV || o->op_type == OP_RV2HV)
16474 && cUNOPo->op_first->op_type == OP_GV)
16476 gv = cGVOPx_gv(cUNOPo->op_first);
16480 == OP_RV2HV ? MUTABLE_SV(GvHV(gv)) : MUTABLE_SV(GvAV(gv));
16485 if (kid && kid->op_type == OP_NEGATE) {
16487 kid = cUNOPx(kid)->op_first;
16490 if (kid && kid->op_type == OP_CONST && SvOK(cSVOPx_sv(kid))) {
16491 /* index is constant */
16494 kidsv = newSVpvs_flags("-", SVs_TEMP);
16495 sv_catsv(kidsv, cSVOPx_sv(kid));
16498 kidsv = cSVOPx_sv(kid);
16502 if (obase->op_type == OP_HELEM) {
16503 HE* he = hv_fetch_ent(MUTABLE_HV(sv), kidsv, 0, 0);
16504 if (!he || HeVAL(he) != uninit_sv)
16508 SV * const opsv = cSVOPx_sv(kid);
16509 const IV opsviv = SvIV(opsv);
16510 SV * const * const svp = av_fetch(MUTABLE_AV(sv),
16511 negate ? - opsviv : opsviv,
16513 if (!svp || *svp != uninit_sv)
16517 if (obase->op_type == OP_HELEM)
16518 return varname(gv, '%', o->op_targ,
16519 kidsv, 0, FUV_SUBSCRIPT_HASH);
16521 return varname(gv, '@', o->op_targ, NULL,
16522 negate ? - SvIV(cSVOPx_sv(kid)) : SvIV(cSVOPx_sv(kid)),
16523 FUV_SUBSCRIPT_ARRAY);
16526 /* index is an expression;
16527 * attempt to find a match within the aggregate */
16528 if (obase->op_type == OP_HELEM) {
16529 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16531 return varname(gv, '%', o->op_targ,
16532 keysv, 0, FUV_SUBSCRIPT_HASH);
16535 const SSize_t index
16536 = find_array_subscript((const AV *)sv, uninit_sv);
16538 return varname(gv, '@', o->op_targ,
16539 NULL, index, FUV_SUBSCRIPT_ARRAY);
16544 (char)((o->op_type == OP_PADAV || o->op_type == OP_RV2AV)
16546 o->op_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
16548 NOT_REACHED; /* NOTREACHED */
16551 case OP_MULTIDEREF: {
16552 /* If we were executing OP_MULTIDEREF when the undef warning
16553 * triggered, then it must be one of the index values within
16554 * that triggered it. If not, then the only possibility is that
16555 * the value retrieved by the last aggregate index might be the
16556 * culprit. For the former, we set PL_multideref_pc each time before
16557 * using an index, so work though the item list until we reach
16558 * that point. For the latter, just work through the entire item
16559 * list; the last aggregate retrieved will be the candidate.
16560 * There is a third rare possibility: something triggered
16561 * magic while fetching an array/hash element. Just display
16562 * nothing in this case.
16565 /* the named aggregate, if any */
16566 PADOFFSET agg_targ = 0;
16568 /* the last-seen index */
16570 PADOFFSET index_targ;
16572 IV index_const_iv = 0; /* init for spurious compiler warn */
16573 SV *index_const_sv;
16574 int depth = 0; /* how many array/hash lookups we've done */
16576 UNOP_AUX_item *items = cUNOP_AUXx(obase)->op_aux;
16577 UNOP_AUX_item *last = NULL;
16578 UV actions = items->uv;
16581 if (PL_op == obase) {
16582 last = PL_multideref_pc;
16583 assert(last >= items && last <= items + items[-1].uv);
16590 switch (actions & MDEREF_ACTION_MASK) {
16592 case MDEREF_reload:
16593 actions = (++items)->uv;
16596 case MDEREF_HV_padhv_helem: /* $lex{...} */
16599 case MDEREF_AV_padav_aelem: /* $lex[...] */
16600 agg_targ = (++items)->pad_offset;
16604 case MDEREF_HV_gvhv_helem: /* $pkg{...} */
16607 case MDEREF_AV_gvav_aelem: /* $pkg[...] */
16609 agg_gv = (GV*)UNOP_AUX_item_sv(++items);
16610 assert(isGV_with_GP(agg_gv));
16613 case MDEREF_HV_gvsv_vivify_rv2hv_helem: /* $pkg->{...} */
16614 case MDEREF_HV_padsv_vivify_rv2hv_helem: /* $lex->{...} */
16617 case MDEREF_HV_pop_rv2hv_helem: /* expr->{...} */
16618 case MDEREF_HV_vivify_rv2hv_helem: /* vivify, ->{...} */
16624 case MDEREF_AV_gvsv_vivify_rv2av_aelem: /* $pkg->[...] */
16625 case MDEREF_AV_padsv_vivify_rv2av_aelem: /* $lex->[...] */
16628 case MDEREF_AV_pop_rv2av_aelem: /* expr->[...] */
16629 case MDEREF_AV_vivify_rv2av_aelem: /* vivify, ->[...] */
16636 index_const_sv = NULL;
16638 index_type = (actions & MDEREF_INDEX_MASK);
16639 switch (index_type) {
16640 case MDEREF_INDEX_none:
16642 case MDEREF_INDEX_const:
16644 index_const_sv = UNOP_AUX_item_sv(++items)
16646 index_const_iv = (++items)->iv;
16648 case MDEREF_INDEX_padsv:
16649 index_targ = (++items)->pad_offset;
16651 case MDEREF_INDEX_gvsv:
16652 index_gv = (GV*)UNOP_AUX_item_sv(++items);
16653 assert(isGV_with_GP(index_gv));
16657 if (index_type != MDEREF_INDEX_none)
16660 if ( index_type == MDEREF_INDEX_none
16661 || (actions & MDEREF_FLAG_last)
16662 || (last && items >= last)
16666 actions >>= MDEREF_SHIFT;
16669 if (PL_op == obase) {
16670 /* most likely index was undef */
16672 *desc_p = ( (actions & MDEREF_FLAG_last)
16673 && (obase->op_private
16674 & (OPpMULTIDEREF_EXISTS|OPpMULTIDEREF_DELETE)))
16676 (obase->op_private & OPpMULTIDEREF_EXISTS)
16679 : is_hv ? "hash element" : "array element";
16680 assert(index_type != MDEREF_INDEX_none);
16682 if (GvSV(index_gv) == uninit_sv)
16683 return varname(index_gv, '$', 0, NULL, 0,
16684 FUV_SUBSCRIPT_NONE);
16689 if (PL_curpad[index_targ] == uninit_sv)
16690 return varname(NULL, '$', index_targ,
16691 NULL, 0, FUV_SUBSCRIPT_NONE);
16695 /* If we got to this point it was undef on a const subscript,
16696 * so magic probably involved, e.g. $ISA[0]. Give up. */
16700 /* the SV returned by pp_multideref() was undef, if anything was */
16706 sv = PAD_SV(agg_targ);
16708 sv = is_hv ? MUTABLE_SV(GvHV(agg_gv)) : MUTABLE_SV(GvAV(agg_gv));
16712 if (index_type == MDEREF_INDEX_const) {
16717 HE* he = hv_fetch_ent(MUTABLE_HV(sv), index_const_sv, 0, 0);
16718 if (!he || HeVAL(he) != uninit_sv)
16722 SV * const * const svp =
16723 av_fetch(MUTABLE_AV(sv), index_const_iv, FALSE);
16724 if (!svp || *svp != uninit_sv)
16729 ? varname(agg_gv, '%', agg_targ,
16730 index_const_sv, 0, FUV_SUBSCRIPT_HASH)
16731 : varname(agg_gv, '@', agg_targ,
16732 NULL, index_const_iv, FUV_SUBSCRIPT_ARRAY);
16735 /* index is an var */
16737 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16739 return varname(agg_gv, '%', agg_targ,
16740 keysv, 0, FUV_SUBSCRIPT_HASH);
16743 const SSize_t index
16744 = find_array_subscript((const AV *)sv, uninit_sv);
16746 return varname(agg_gv, '@', agg_targ,
16747 NULL, index, FUV_SUBSCRIPT_ARRAY);
16751 return varname(agg_gv,
16753 agg_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
16755 NOT_REACHED; /* NOTREACHED */
16759 /* only examine RHS */
16760 return find_uninit_var(cBINOPx(obase)->op_first, uninit_sv,
16764 o = cUNOPx(obase)->op_first;
16765 if ( o->op_type == OP_PUSHMARK
16766 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)
16770 if (!OpHAS_SIBLING(o)) {
16771 /* one-arg version of open is highly magical */
16773 if (o->op_type == OP_GV) { /* open FOO; */
16775 if (match && GvSV(gv) != uninit_sv)
16777 return varname(gv, '$', 0,
16778 NULL, 0, FUV_SUBSCRIPT_NONE);
16780 /* other possibilities not handled are:
16781 * open $x; or open my $x; should return '${*$x}'
16782 * open expr; should return '$'.expr ideally
16789 /* ops where $_ may be an implicit arg */
16794 if ( !(obase->op_flags & OPf_STACKED)) {
16795 if (uninit_sv == DEFSV)
16796 return newSVpvs_flags("$_", SVs_TEMP);
16797 else if (obase->op_targ
16798 && uninit_sv == PAD_SVl(obase->op_targ))
16799 return varname(NULL, '$', obase->op_targ, NULL, 0,
16800 FUV_SUBSCRIPT_NONE);
16807 match = 1; /* print etc can return undef on defined args */
16808 /* skip filehandle as it can't produce 'undef' warning */
16809 o = cUNOPx(obase)->op_first;
16810 if ((obase->op_flags & OPf_STACKED)
16812 ( o->op_type == OP_PUSHMARK
16813 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)))
16814 o = OpSIBLING(OpSIBLING(o));
16818 case OP_ENTEREVAL: /* could be eval $undef or $x='$undef'; eval $x */
16819 case OP_CUSTOM: /* XS or custom code could trigger random warnings */
16821 /* the following ops are capable of returning PL_sv_undef even for
16822 * defined arg(s) */
16841 case OP_GETPEERNAME:
16888 case OP_SMARTMATCH:
16897 /* XXX tmp hack: these two may call an XS sub, and currently
16898 XS subs don't have a SUB entry on the context stack, so CV and
16899 pad determination goes wrong, and BAD things happen. So, just
16900 don't try to determine the value under those circumstances.
16901 Need a better fix at dome point. DAPM 11/2007 */
16907 GV * const gv = gv_fetchpvs(".", GV_NOTQUAL, SVt_PV);
16908 if (gv && GvSV(gv) == uninit_sv)
16909 return newSVpvs_flags("$.", SVs_TEMP);
16914 /* def-ness of rval pos() is independent of the def-ness of its arg */
16915 if ( !(obase->op_flags & OPf_MOD))
16921 if (SvROK(PL_rs) && uninit_sv == SvRV(PL_rs))
16922 return newSVpvs_flags("${$/}", SVs_TEMP);
16927 if (!(obase->op_flags & OPf_KIDS))
16929 o = cUNOPx(obase)->op_first;
16935 /* This loop checks all the kid ops, skipping any that cannot pos-
16936 * sibly be responsible for the uninitialized value; i.e., defined
16937 * constants and ops that return nothing. If there is only one op
16938 * left that is not skipped, then we *know* it is responsible for
16939 * the uninitialized value. If there is more than one op left, we
16940 * have to look for an exact match in the while() loop below.
16941 * Note that we skip padrange, because the individual pad ops that
16942 * it replaced are still in the tree, so we work on them instead.
16945 for (kid=o; kid; kid = OpSIBLING(kid)) {
16946 const OPCODE type = kid->op_type;
16947 if ( (type == OP_CONST && SvOK(cSVOPx_sv(kid)))
16948 || (type == OP_NULL && ! (kid->op_flags & OPf_KIDS))
16949 || (type == OP_PUSHMARK)
16950 || (type == OP_PADRANGE)
16954 if (o2) { /* more than one found */
16961 return find_uninit_var(o2, uninit_sv, match, desc_p);
16963 /* scan all args */
16965 sv = find_uninit_var(o, uninit_sv, 1, desc_p);
16977 =for apidoc report_uninit
16979 Print appropriate "Use of uninitialized variable" warning.
16985 Perl_report_uninit(pTHX_ const SV *uninit_sv)
16987 const char *desc = NULL;
16988 SV* varname = NULL;
16991 desc = PL_op->op_type == OP_STRINGIFY && PL_op->op_folded
16993 : PL_op->op_type == OP_MULTICONCAT
16994 && (PL_op->op_private & OPpMULTICONCAT_FAKE)
16997 if (uninit_sv && PL_curpad) {
16998 varname = find_uninit_var(PL_op, uninit_sv, 0, &desc);
17000 sv_insert(varname, 0, 0, " ", 1);
17003 else if (PL_curstackinfo->si_type == PERLSI_SORT && cxstack_ix == 0)
17004 /* we've reached the end of a sort block or sub,
17005 * and the uninit value is probably what that code returned */
17008 /* PL_warn_uninit_sv is constant */
17009 GCC_DIAG_IGNORE(-Wformat-nonliteral);
17011 /* diag_listed_as: Use of uninitialized value%s */
17012 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit_sv,
17013 SVfARG(varname ? varname : &PL_sv_no),
17016 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit,
17022 * ex: set ts=8 sts=4 sw=4 et: