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 *referant = 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 referant = 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 (UNLIKELY(new_type == SVt_REGEXP))
1466 sv->sv_u.svu_rx = (regexp *)new_body;
1467 else if (old_type < SVt_PV) {
1468 /* referant will be NULL unless the old type was SVt_IV emulating
1470 sv->sv_u.svu_rv = referant;
1474 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1475 (unsigned long)new_type);
1478 /* if this is zero, this is a body-less SVt_NULL, SVt_IV/SVt_RV,
1479 and sometimes SVt_NV */
1480 if (old_type_details->body_size) {
1484 /* Note that there is an assumption that all bodies of types that
1485 can be upgraded came from arenas. Only the more complex non-
1486 upgradable types are allowed to be directly malloc()ed. */
1487 assert(old_type_details->arena);
1488 del_body((void*)((char*)old_body + old_type_details->offset),
1489 &PL_body_roots[old_type]);
1495 =for apidoc sv_backoff
1497 Remove any string offset. You should normally use the C<SvOOK_off> macro
1503 /* prior to 5.000 stable, this function returned the new OOK-less SvFLAGS
1504 prior to 5.23.4 this function always returned 0
1508 Perl_sv_backoff(SV *const sv)
1511 const char * const s = SvPVX_const(sv);
1513 PERL_ARGS_ASSERT_SV_BACKOFF;
1516 assert(SvTYPE(sv) != SVt_PVHV);
1517 assert(SvTYPE(sv) != SVt_PVAV);
1519 SvOOK_offset(sv, delta);
1521 SvLEN_set(sv, SvLEN(sv) + delta);
1522 SvPV_set(sv, SvPVX(sv) - delta);
1523 SvFLAGS(sv) &= ~SVf_OOK;
1524 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
1531 Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1532 upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1533 Use the C<SvGROW> wrapper instead.
1538 static void S_sv_uncow(pTHX_ SV * const sv, const U32 flags);
1541 Perl_sv_grow(pTHX_ SV *const sv, STRLEN newlen)
1545 PERL_ARGS_ASSERT_SV_GROW;
1549 if (SvTYPE(sv) < SVt_PV) {
1550 sv_upgrade(sv, SVt_PV);
1551 s = SvPVX_mutable(sv);
1553 else if (SvOOK(sv)) { /* pv is offset? */
1555 s = SvPVX_mutable(sv);
1556 if (newlen > SvLEN(sv))
1557 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
1561 if (SvIsCOW(sv)) S_sv_uncow(aTHX_ sv, 0);
1562 s = SvPVX_mutable(sv);
1565 #ifdef PERL_COPY_ON_WRITE
1566 /* the new COW scheme uses SvPVX(sv)[SvLEN(sv)-1] (if spare)
1567 * to store the COW count. So in general, allocate one more byte than
1568 * asked for, to make it likely this byte is always spare: and thus
1569 * make more strings COW-able.
1570 * If the new size is a big power of two, don't bother: we assume the
1571 * caller wanted a nice 2^N sized block and will be annoyed at getting
1573 * Only increment if the allocation isn't MEM_SIZE_MAX,
1574 * otherwise it will wrap to 0.
1576 if ( (newlen < 0x1000 || (newlen & (newlen - 1)))
1577 && newlen != MEM_SIZE_MAX
1582 #if defined(PERL_USE_MALLOC_SIZE) && defined(Perl_safesysmalloc_size)
1583 #define PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1586 if (newlen > SvLEN(sv)) { /* need more room? */
1587 STRLEN minlen = SvCUR(sv);
1588 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
1589 if (newlen < minlen)
1591 #ifndef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1593 /* Don't round up on the first allocation, as odds are pretty good that
1594 * the initial request is accurate as to what is really needed */
1596 STRLEN rounded = PERL_STRLEN_ROUNDUP(newlen);
1597 if (rounded > newlen)
1601 if (SvLEN(sv) && s) {
1602 s = (char*)saferealloc(s, newlen);
1605 s = (char*)safemalloc(newlen);
1606 if (SvPVX_const(sv) && SvCUR(sv)) {
1607 Move(SvPVX_const(sv), s, SvCUR(sv), char);
1611 #ifdef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1612 /* Do this here, do it once, do it right, and then we will never get
1613 called back into sv_grow() unless there really is some growing
1615 SvLEN_set(sv, Perl_safesysmalloc_size(s));
1617 SvLEN_set(sv, newlen);
1624 =for apidoc sv_setiv
1626 Copies an integer into the given SV, upgrading first if necessary.
1627 Does not handle 'set' magic. See also C<L</sv_setiv_mg>>.
1633 Perl_sv_setiv(pTHX_ SV *const sv, const IV i)
1635 PERL_ARGS_ASSERT_SV_SETIV;
1637 SV_CHECK_THINKFIRST_COW_DROP(sv);
1638 switch (SvTYPE(sv)) {
1641 sv_upgrade(sv, SVt_IV);
1644 sv_upgrade(sv, SVt_PVIV);
1648 if (!isGV_with_GP(sv))
1655 /* diag_listed_as: Can't coerce %s to %s in %s */
1656 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
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))
1767 /* diag_listed_as: Can't coerce %s to %s in %s */
1768 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
1774 (void)SvNOK_only(sv); /* validate number */
1779 =for apidoc sv_setnv_mg
1781 Like C<sv_setnv>, but also handles 'set' magic.
1787 Perl_sv_setnv_mg(pTHX_ SV *const sv, const NV num)
1789 PERL_ARGS_ASSERT_SV_SETNV_MG;
1795 /* Return a cleaned-up, printable version of sv, for non-numeric, or
1796 * not incrementable warning display.
1797 * Originally part of S_not_a_number().
1798 * The return value may be != tmpbuf.
1802 S_sv_display(pTHX_ SV *const sv, char *tmpbuf, STRLEN tmpbuf_size) {
1805 PERL_ARGS_ASSERT_SV_DISPLAY;
1808 SV *dsv = newSVpvs_flags("", SVs_TEMP);
1809 pv = sv_uni_display(dsv, sv, 32, UNI_DISPLAY_ISPRINT);
1812 const char * const limit = tmpbuf + tmpbuf_size - 8;
1813 /* each *s can expand to 4 chars + "...\0",
1814 i.e. need room for 8 chars */
1816 const char *s = SvPVX_const(sv);
1817 const char * const end = s + SvCUR(sv);
1818 for ( ; s < end && d < limit; s++ ) {
1820 if (! isASCII(ch) && !isPRINT_LC(ch)) {
1824 /* Map to ASCII "equivalent" of Latin1 */
1825 ch = LATIN1_TO_NATIVE(NATIVE_TO_LATIN1(ch) & 127);
1831 else if (ch == '\r') {
1835 else if (ch == '\f') {
1839 else if (ch == '\\') {
1843 else if (ch == '\0') {
1847 else if (isPRINT_LC(ch))
1866 /* Print an "isn't numeric" warning, using a cleaned-up,
1867 * printable version of the offending string
1871 S_not_a_number(pTHX_ SV *const sv)
1876 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1878 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1881 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1882 /* diag_listed_as: Argument "%s" isn't numeric%s */
1883 "Argument \"%s\" isn't numeric in %s", pv,
1886 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1887 /* diag_listed_as: Argument "%s" isn't numeric%s */
1888 "Argument \"%s\" isn't numeric", pv);
1892 S_not_incrementable(pTHX_ SV *const sv) {
1896 PERL_ARGS_ASSERT_NOT_INCREMENTABLE;
1898 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1900 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1901 "Argument \"%s\" treated as 0 in increment (++)", pv);
1905 =for apidoc looks_like_number
1907 Test if the content of an SV looks like a number (or is a number).
1908 C<Inf> and C<Infinity> are treated as numbers (so will not issue a
1909 non-numeric warning), even if your C<atof()> doesn't grok them. Get-magic is
1916 Perl_looks_like_number(pTHX_ SV *const sv)
1922 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1924 if (SvPOK(sv) || SvPOKp(sv)) {
1925 sbegin = SvPV_nomg_const(sv, len);
1928 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
1929 numtype = grok_number(sbegin, len, NULL);
1930 return ((numtype & IS_NUMBER_TRAILING)) ? 0 : numtype;
1934 S_glob_2number(pTHX_ GV * const gv)
1936 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1938 /* We know that all GVs stringify to something that is not-a-number,
1939 so no need to test that. */
1940 if (ckWARN(WARN_NUMERIC))
1942 SV *const buffer = sv_newmortal();
1943 gv_efullname3(buffer, gv, "*");
1944 not_a_number(buffer);
1946 /* We just want something true to return, so that S_sv_2iuv_common
1947 can tail call us and return true. */
1951 /* Actually, ISO C leaves conversion of UV to IV undefined, but
1952 until proven guilty, assume that things are not that bad... */
1957 As 64 bit platforms often have an NV that doesn't preserve all bits of
1958 an IV (an assumption perl has been based on to date) it becomes necessary
1959 to remove the assumption that the NV always carries enough precision to
1960 recreate the IV whenever needed, and that the NV is the canonical form.
1961 Instead, IV/UV and NV need to be given equal rights. So as to not lose
1962 precision as a side effect of conversion (which would lead to insanity
1963 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
1964 1) to distinguish between IV/UV/NV slots that have a valid conversion cached
1965 where precision was lost, and IV/UV/NV slots that have a valid conversion
1966 which has lost no precision
1967 2) to ensure that if a numeric conversion to one form is requested that
1968 would lose precision, the precise conversion (or differently
1969 imprecise conversion) is also performed and cached, to prevent
1970 requests for different numeric formats on the same SV causing
1971 lossy conversion chains. (lossless conversion chains are perfectly
1976 SvIOKp is true if the IV slot contains a valid value
1977 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1978 SvNOKp is true if the NV slot contains a valid value
1979 SvNOK is true only if the NV value is accurate
1982 while converting from PV to NV, check to see if converting that NV to an
1983 IV(or UV) would lose accuracy over a direct conversion from PV to
1984 IV(or UV). If it would, cache both conversions, return NV, but mark
1985 SV as IOK NOKp (ie not NOK).
1987 While converting from PV to IV, check to see if converting that IV to an
1988 NV would lose accuracy over a direct conversion from PV to NV. If it
1989 would, cache both conversions, flag similarly.
1991 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
1992 correctly because if IV & NV were set NV *always* overruled.
1993 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
1994 changes - now IV and NV together means that the two are interchangeable:
1995 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
1997 The benefit of this is that operations such as pp_add know that if
1998 SvIOK is true for both left and right operands, then integer addition
1999 can be used instead of floating point (for cases where the result won't
2000 overflow). Before, floating point was always used, which could lead to
2001 loss of precision compared with integer addition.
2003 * making IV and NV equal status should make maths accurate on 64 bit
2005 * may speed up maths somewhat if pp_add and friends start to use
2006 integers when possible instead of fp. (Hopefully the overhead in
2007 looking for SvIOK and checking for overflow will not outweigh the
2008 fp to integer speedup)
2009 * will slow down integer operations (callers of SvIV) on "inaccurate"
2010 values, as the change from SvIOK to SvIOKp will cause a call into
2011 sv_2iv each time rather than a macro access direct to the IV slot
2012 * should speed up number->string conversion on integers as IV is
2013 favoured when IV and NV are equally accurate
2015 ####################################################################
2016 You had better be using SvIOK_notUV if you want an IV for arithmetic:
2017 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
2018 On the other hand, SvUOK is true iff UV.
2019 ####################################################################
2021 Your mileage will vary depending your CPU's relative fp to integer
2025 #ifndef NV_PRESERVES_UV
2026 # define IS_NUMBER_UNDERFLOW_IV 1
2027 # define IS_NUMBER_UNDERFLOW_UV 2
2028 # define IS_NUMBER_IV_AND_UV 2
2029 # define IS_NUMBER_OVERFLOW_IV 4
2030 # define IS_NUMBER_OVERFLOW_UV 5
2032 /* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
2034 /* For sv_2nv these three cases are "SvNOK and don't bother casting" */
2036 S_sv_2iuv_non_preserve(pTHX_ SV *const sv
2042 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
2043 PERL_UNUSED_CONTEXT;
2045 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));
2046 if (SvNVX(sv) < (NV)IV_MIN) {
2047 (void)SvIOKp_on(sv);
2049 SvIV_set(sv, IV_MIN);
2050 return IS_NUMBER_UNDERFLOW_IV;
2052 if (SvNVX(sv) > (NV)UV_MAX) {
2053 (void)SvIOKp_on(sv);
2056 SvUV_set(sv, UV_MAX);
2057 return IS_NUMBER_OVERFLOW_UV;
2059 (void)SvIOKp_on(sv);
2061 /* Can't use strtol etc to convert this string. (See truth table in
2063 if (SvNVX(sv) <= (UV)IV_MAX) {
2064 SvIV_set(sv, I_V(SvNVX(sv)));
2065 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2066 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
2068 /* Integer is imprecise. NOK, IOKp */
2070 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
2073 SvUV_set(sv, U_V(SvNVX(sv)));
2074 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2075 if (SvUVX(sv) == UV_MAX) {
2076 /* As we know that NVs don't preserve UVs, UV_MAX cannot
2077 possibly be preserved by NV. Hence, it must be overflow.
2079 return IS_NUMBER_OVERFLOW_UV;
2081 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
2083 /* Integer is imprecise. NOK, IOKp */
2085 return IS_NUMBER_OVERFLOW_IV;
2087 #endif /* !NV_PRESERVES_UV*/
2089 /* If numtype is infnan, set the NV of the sv accordingly.
2090 * If numtype is anything else, try setting the NV using Atof(PV). */
2092 # pragma warning(push)
2093 # pragma warning(disable:4756;disable:4056)
2096 S_sv_setnv(pTHX_ SV* sv, int numtype)
2098 bool pok = cBOOL(SvPOK(sv));
2101 if ((numtype & IS_NUMBER_INFINITY)) {
2102 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -NV_INF : NV_INF);
2107 if ((numtype & IS_NUMBER_NAN)) {
2108 SvNV_set(sv, NV_NAN);
2113 SvNV_set(sv, Atof(SvPVX_const(sv)));
2114 /* Purposefully no true nok here, since we don't want to blow
2115 * away the possible IOK/UV of an existing sv. */
2118 SvNOK_only(sv); /* No IV or UV please, this is pure infnan. */
2120 SvPOK_on(sv); /* PV is okay, though. */
2124 # pragma warning(pop)
2128 S_sv_2iuv_common(pTHX_ SV *const sv)
2130 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
2133 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
2134 * without also getting a cached IV/UV from it at the same time
2135 * (ie PV->NV conversion should detect loss of accuracy and cache
2136 * IV or UV at same time to avoid this. */
2137 /* IV-over-UV optimisation - choose to cache IV if possible */
2139 if (SvTYPE(sv) == SVt_NV)
2140 sv_upgrade(sv, SVt_PVNV);
2142 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
2143 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
2144 certainly cast into the IV range at IV_MAX, whereas the correct
2145 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
2147 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2148 if (Perl_isnan(SvNVX(sv))) {
2154 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2155 SvIV_set(sv, I_V(SvNVX(sv)));
2156 if (SvNVX(sv) == (NV) SvIVX(sv)
2157 #ifndef NV_PRESERVES_UV
2158 && SvIVX(sv) != IV_MIN /* avoid negating IV_MIN below */
2159 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2160 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2161 /* Don't flag it as "accurately an integer" if the number
2162 came from a (by definition imprecise) NV operation, and
2163 we're outside the range of NV integer precision */
2167 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2169 /* scalar has trailing garbage, eg "42a" */
2171 DEBUG_c(PerlIO_printf(Perl_debug_log,
2172 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (precise)\n",
2178 /* IV not precise. No need to convert from PV, as NV
2179 conversion would already have cached IV if it detected
2180 that PV->IV would be better than PV->NV->IV
2181 flags already correct - don't set public IOK. */
2182 DEBUG_c(PerlIO_printf(Perl_debug_log,
2183 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (imprecise)\n",
2188 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2189 but the cast (NV)IV_MIN rounds to a the value less (more
2190 negative) than IV_MIN which happens to be equal to SvNVX ??
2191 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2192 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2193 (NV)UVX == NVX are both true, but the values differ. :-(
2194 Hopefully for 2s complement IV_MIN is something like
2195 0x8000000000000000 which will be exact. NWC */
2198 SvUV_set(sv, U_V(SvNVX(sv)));
2200 (SvNVX(sv) == (NV) SvUVX(sv))
2201 #ifndef NV_PRESERVES_UV
2202 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2203 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2204 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2205 /* Don't flag it as "accurately an integer" if the number
2206 came from a (by definition imprecise) NV operation, and
2207 we're outside the range of NV integer precision */
2213 DEBUG_c(PerlIO_printf(Perl_debug_log,
2214 "0x%"UVxf" 2iv(%"UVuf" => %"IVdf") (as unsigned)\n",
2220 else if (SvPOKp(sv)) {
2222 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2223 /* We want to avoid a possible problem when we cache an IV/ a UV which
2224 may be later translated to an NV, and the resulting NV is not
2225 the same as the direct translation of the initial string
2226 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2227 be careful to ensure that the value with the .456 is around if the
2228 NV value is requested in the future).
2230 This means that if we cache such an IV/a UV, we need to cache the
2231 NV as well. Moreover, we trade speed for space, and do not
2232 cache the NV if we are sure it's not needed.
2235 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2236 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2237 == IS_NUMBER_IN_UV) {
2238 /* It's definitely an integer, only upgrade to PVIV */
2239 if (SvTYPE(sv) < SVt_PVIV)
2240 sv_upgrade(sv, SVt_PVIV);
2242 } else if (SvTYPE(sv) < SVt_PVNV)
2243 sv_upgrade(sv, SVt_PVNV);
2245 if ((numtype & (IS_NUMBER_INFINITY | IS_NUMBER_NAN))) {
2246 if (ckWARN(WARN_NUMERIC) && ((numtype & IS_NUMBER_TRAILING)))
2248 S_sv_setnv(aTHX_ sv, numtype);
2252 /* If NVs preserve UVs then we only use the UV value if we know that
2253 we aren't going to call atof() below. If NVs don't preserve UVs
2254 then the value returned may have more precision than atof() will
2255 return, even though value isn't perfectly accurate. */
2256 if ((numtype & (IS_NUMBER_IN_UV
2257 #ifdef NV_PRESERVES_UV
2260 )) == IS_NUMBER_IN_UV) {
2261 /* This won't turn off the public IOK flag if it was set above */
2262 (void)SvIOKp_on(sv);
2264 if (!(numtype & IS_NUMBER_NEG)) {
2266 if (value <= (UV)IV_MAX) {
2267 SvIV_set(sv, (IV)value);
2269 /* it didn't overflow, and it was positive. */
2270 SvUV_set(sv, value);
2274 /* 2s complement assumption */
2275 if (value <= (UV)IV_MIN) {
2276 SvIV_set(sv, value == (UV)IV_MIN
2277 ? IV_MIN : -(IV)value);
2279 /* Too negative for an IV. This is a double upgrade, but
2280 I'm assuming it will be rare. */
2281 if (SvTYPE(sv) < SVt_PVNV)
2282 sv_upgrade(sv, SVt_PVNV);
2286 SvNV_set(sv, -(NV)value);
2287 SvIV_set(sv, IV_MIN);
2291 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2292 will be in the previous block to set the IV slot, and the next
2293 block to set the NV slot. So no else here. */
2295 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2296 != IS_NUMBER_IN_UV) {
2297 /* It wasn't an (integer that doesn't overflow the UV). */
2298 S_sv_setnv(aTHX_ sv, numtype);
2300 if (! numtype && ckWARN(WARN_NUMERIC))
2303 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%" NVgf ")\n",
2304 PTR2UV(sv), SvNVX(sv)));
2306 #ifdef NV_PRESERVES_UV
2307 (void)SvIOKp_on(sv);
2309 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2310 if (Perl_isnan(SvNVX(sv))) {
2316 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2317 SvIV_set(sv, I_V(SvNVX(sv)));
2318 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2321 NOOP; /* Integer is imprecise. NOK, IOKp */
2323 /* UV will not work better than IV */
2325 if (SvNVX(sv) > (NV)UV_MAX) {
2327 /* Integer is inaccurate. NOK, IOKp, is UV */
2328 SvUV_set(sv, UV_MAX);
2330 SvUV_set(sv, U_V(SvNVX(sv)));
2331 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2332 NV preservse UV so can do correct comparison. */
2333 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2336 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
2341 #else /* NV_PRESERVES_UV */
2342 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2343 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
2344 /* The IV/UV slot will have been set from value returned by
2345 grok_number above. The NV slot has just been set using
2348 assert (SvIOKp(sv));
2350 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2351 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2352 /* Small enough to preserve all bits. */
2353 (void)SvIOKp_on(sv);
2355 SvIV_set(sv, I_V(SvNVX(sv)));
2356 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2358 /* Assumption: first non-preserved integer is < IV_MAX,
2359 this NV is in the preserved range, therefore: */
2360 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2362 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);
2366 0 0 already failed to read UV.
2367 0 1 already failed to read UV.
2368 1 0 you won't get here in this case. IV/UV
2369 slot set, public IOK, Atof() unneeded.
2370 1 1 already read UV.
2371 so there's no point in sv_2iuv_non_preserve() attempting
2372 to use atol, strtol, strtoul etc. */
2374 sv_2iuv_non_preserve (sv, numtype);
2376 sv_2iuv_non_preserve (sv);
2380 #endif /* NV_PRESERVES_UV */
2381 /* It might be more code efficient to go through the entire logic above
2382 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2383 gets complex and potentially buggy, so more programmer efficient
2384 to do it this way, by turning off the public flags: */
2386 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2390 if (isGV_with_GP(sv))
2391 return glob_2number(MUTABLE_GV(sv));
2393 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2395 if (SvTYPE(sv) < SVt_IV)
2396 /* Typically the caller expects that sv_any is not NULL now. */
2397 sv_upgrade(sv, SVt_IV);
2398 /* Return 0 from the caller. */
2405 =for apidoc sv_2iv_flags
2407 Return the integer value of an SV, doing any necessary string
2408 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2409 Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2415 Perl_sv_2iv_flags(pTHX_ SV *const sv, const I32 flags)
2417 PERL_ARGS_ASSERT_SV_2IV_FLAGS;
2419 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2420 && SvTYPE(sv) != SVt_PVFM);
2422 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2428 if (flags & SV_SKIP_OVERLOAD)
2430 tmpstr = AMG_CALLunary(sv, numer_amg);
2431 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2432 return SvIV(tmpstr);
2435 return PTR2IV(SvRV(sv));
2438 if (SvVALID(sv) || isREGEXP(sv)) {
2439 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2440 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2441 In practice they are extremely unlikely to actually get anywhere
2442 accessible by user Perl code - the only way that I'm aware of is when
2443 a constant subroutine which is used as the second argument to index.
2445 Regexps have no SvIVX and SvNVX fields.
2447 assert(isREGEXP(sv) || SvPOKp(sv));
2450 const char * const ptr =
2451 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2453 = grok_number(ptr, SvCUR(sv), &value);
2455 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2456 == IS_NUMBER_IN_UV) {
2457 /* It's definitely an integer */
2458 if (numtype & IS_NUMBER_NEG) {
2459 if (value < (UV)IV_MIN)
2462 if (value < (UV)IV_MAX)
2467 /* Quite wrong but no good choices. */
2468 if ((numtype & IS_NUMBER_INFINITY)) {
2469 return (numtype & IS_NUMBER_NEG) ? IV_MIN : IV_MAX;
2470 } else if ((numtype & IS_NUMBER_NAN)) {
2471 return 0; /* So wrong. */
2475 if (ckWARN(WARN_NUMERIC))
2478 return I_V(Atof(ptr));
2482 if (SvTHINKFIRST(sv)) {
2483 if (SvREADONLY(sv) && !SvOK(sv)) {
2484 if (ckWARN(WARN_UNINITIALIZED))
2491 if (S_sv_2iuv_common(aTHX_ sv))
2495 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"IVdf")\n",
2496 PTR2UV(sv),SvIVX(sv)));
2497 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
2501 =for apidoc sv_2uv_flags
2503 Return the unsigned integer value of an SV, doing any necessary string
2504 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2505 Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
2511 Perl_sv_2uv_flags(pTHX_ SV *const sv, const I32 flags)
2513 PERL_ARGS_ASSERT_SV_2UV_FLAGS;
2515 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2521 if (flags & SV_SKIP_OVERLOAD)
2523 tmpstr = AMG_CALLunary(sv, numer_amg);
2524 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2525 return SvUV(tmpstr);
2528 return PTR2UV(SvRV(sv));
2531 if (SvVALID(sv) || isREGEXP(sv)) {
2532 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2533 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2534 Regexps have no SvIVX and SvNVX fields. */
2535 assert(isREGEXP(sv) || SvPOKp(sv));
2538 const char * const ptr =
2539 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2541 = grok_number(ptr, SvCUR(sv), &value);
2543 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2544 == IS_NUMBER_IN_UV) {
2545 /* It's definitely an integer */
2546 if (!(numtype & IS_NUMBER_NEG))
2550 /* Quite wrong but no good choices. */
2551 if ((numtype & IS_NUMBER_INFINITY)) {
2552 return UV_MAX; /* So wrong. */
2553 } else if ((numtype & IS_NUMBER_NAN)) {
2554 return 0; /* So wrong. */
2558 if (ckWARN(WARN_NUMERIC))
2561 return U_V(Atof(ptr));
2565 if (SvTHINKFIRST(sv)) {
2566 if (SvREADONLY(sv) && !SvOK(sv)) {
2567 if (ckWARN(WARN_UNINITIALIZED))
2574 if (S_sv_2iuv_common(aTHX_ sv))
2578 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2uv(%"UVuf")\n",
2579 PTR2UV(sv),SvUVX(sv)));
2580 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
2584 =for apidoc sv_2nv_flags
2586 Return the num value of an SV, doing any necessary string or integer
2587 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2588 Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
2594 Perl_sv_2nv_flags(pTHX_ SV *const sv, const I32 flags)
2596 PERL_ARGS_ASSERT_SV_2NV_FLAGS;
2598 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2599 && SvTYPE(sv) != SVt_PVFM);
2600 if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) {
2601 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2602 the same flag bit as SVf_IVisUV, so must not let them cache NVs.
2603 Regexps have no SvIVX and SvNVX fields. */
2605 if (flags & SV_GMAGIC)
2609 if (SvPOKp(sv) && !SvIOKp(sv)) {
2610 ptr = SvPVX_const(sv);
2612 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
2613 !grok_number(ptr, SvCUR(sv), NULL))
2619 return (NV)SvUVX(sv);
2621 return (NV)SvIVX(sv);
2627 ptr = RX_WRAPPED((REGEXP *)sv);
2630 assert(SvTYPE(sv) >= SVt_PVMG);
2631 /* This falls through to the report_uninit near the end of the
2633 } else if (SvTHINKFIRST(sv)) {
2638 if (flags & SV_SKIP_OVERLOAD)
2640 tmpstr = AMG_CALLunary(sv, numer_amg);
2641 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2642 return SvNV(tmpstr);
2645 return PTR2NV(SvRV(sv));
2647 if (SvREADONLY(sv) && !SvOK(sv)) {
2648 if (ckWARN(WARN_UNINITIALIZED))
2653 if (SvTYPE(sv) < SVt_NV) {
2654 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2655 sv_upgrade(sv, SVt_NV);
2657 STORE_NUMERIC_LOCAL_SET_STANDARD();
2658 PerlIO_printf(Perl_debug_log,
2659 "0x%"UVxf" num(%" NVgf ")\n",
2660 PTR2UV(sv), SvNVX(sv));
2661 RESTORE_NUMERIC_LOCAL();
2664 else if (SvTYPE(sv) < SVt_PVNV)
2665 sv_upgrade(sv, SVt_PVNV);
2670 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
2671 #ifdef NV_PRESERVES_UV
2677 /* Only set the public NV OK flag if this NV preserves the IV */
2678 /* Check it's not 0xFFFFFFFFFFFFFFFF */
2680 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
2681 : (SvIVX(sv) == I_V(SvNVX(sv))))
2687 else if (SvPOKp(sv)) {
2689 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2690 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
2692 #ifdef NV_PRESERVES_UV
2693 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2694 == IS_NUMBER_IN_UV) {
2695 /* It's definitely an integer */
2696 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
2698 S_sv_setnv(aTHX_ sv, numtype);
2705 SvNV_set(sv, Atof(SvPVX_const(sv)));
2706 /* Only set the public NV OK flag if this NV preserves the value in
2707 the PV at least as well as an IV/UV would.
2708 Not sure how to do this 100% reliably. */
2709 /* if that shift count is out of range then Configure's test is
2710 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2712 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2713 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2714 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
2715 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2716 /* Can't use strtol etc to convert this string, so don't try.
2717 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2720 /* value has been set. It may not be precise. */
2721 if ((numtype & IS_NUMBER_NEG) && (value >= (UV)IV_MIN)) {
2722 /* 2s complement assumption for (UV)IV_MIN */
2723 SvNOK_on(sv); /* Integer is too negative. */
2728 if (numtype & IS_NUMBER_NEG) {
2729 /* -IV_MIN is undefined, but we should never reach
2730 * this point with both IS_NUMBER_NEG and value ==
2732 assert(value != (UV)IV_MIN);
2733 SvIV_set(sv, -(IV)value);
2734 } else if (value <= (UV)IV_MAX) {
2735 SvIV_set(sv, (IV)value);
2737 SvUV_set(sv, value);
2741 if (numtype & IS_NUMBER_NOT_INT) {
2742 /* I believe that even if the original PV had decimals,
2743 they are lost beyond the limit of the FP precision.
2744 However, neither is canonical, so both only get p
2745 flags. NWC, 2000/11/25 */
2746 /* Both already have p flags, so do nothing */
2748 const NV nv = SvNVX(sv);
2749 /* XXX should this spot have NAN_COMPARE_BROKEN, too? */
2750 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2751 if (SvIVX(sv) == I_V(nv)) {
2754 /* It had no "." so it must be integer. */
2758 /* between IV_MAX and NV(UV_MAX).
2759 Could be slightly > UV_MAX */
2761 if (numtype & IS_NUMBER_NOT_INT) {
2762 /* UV and NV both imprecise. */
2764 const UV nv_as_uv = U_V(nv);
2766 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2775 /* It might be more code efficient to go through the entire logic above
2776 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2777 gets complex and potentially buggy, so more programmer efficient
2778 to do it this way, by turning off the public flags: */
2780 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2781 #endif /* NV_PRESERVES_UV */
2784 if (isGV_with_GP(sv)) {
2785 glob_2number(MUTABLE_GV(sv));
2789 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2791 assert (SvTYPE(sv) >= SVt_NV);
2792 /* Typically the caller expects that sv_any is not NULL now. */
2793 /* XXX Ilya implies that this is a bug in callers that assume this
2794 and ideally should be fixed. */
2798 STORE_NUMERIC_LOCAL_SET_STANDARD();
2799 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2nv(%" NVgf ")\n",
2800 PTR2UV(sv), SvNVX(sv));
2801 RESTORE_NUMERIC_LOCAL();
2809 Return an SV with the numeric value of the source SV, doing any necessary
2810 reference or overload conversion. The caller is expected to have handled
2817 Perl_sv_2num(pTHX_ SV *const sv)
2819 PERL_ARGS_ASSERT_SV_2NUM;
2824 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
2825 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
2826 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2827 return sv_2num(tmpsv);
2829 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2832 /* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2833 * UV as a string towards the end of buf, and return pointers to start and
2836 * We assume that buf is at least TYPE_CHARS(UV) long.
2840 S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
2842 char *ptr = buf + TYPE_CHARS(UV);
2843 char * const ebuf = ptr;
2846 PERL_ARGS_ASSERT_UIV_2BUF;
2854 uv = (iv == IV_MIN) ? (UV)iv : (UV)(-iv);
2858 *--ptr = '0' + (char)(uv % 10);
2866 /* Helper for sv_2pv_flags and sv_vcatpvfn_flags. If the NV is an
2867 * infinity or a not-a-number, writes the appropriate strings to the
2868 * buffer, including a zero byte. On success returns the written length,
2869 * excluding the zero byte, on failure (not an infinity, not a nan)
2870 * returns zero, assert-fails on maxlen being too short.
2872 * XXX for "Inf", "-Inf", and "NaN", we could have three read-only
2873 * shared string constants we point to, instead of generating a new
2874 * string for each instance. */
2876 S_infnan_2pv(NV nv, char* buffer, size_t maxlen, char plus) {
2878 assert(maxlen >= 4);
2879 if (Perl_isinf(nv)) {
2881 if (maxlen < 5) /* "-Inf\0" */
2891 else if (Perl_isnan(nv)) {
2895 /* XXX optionally output the payload mantissa bits as
2896 * "(unsigned)" (to match the nan("...") C99 function,
2897 * or maybe as "(0xhhh...)" would make more sense...
2898 * provide a format string so that the user can decide?
2899 * NOTE: would affect the maxlen and assert() logic.*/
2904 assert((s == buffer + 3) || (s == buffer + 4));
2906 return s - buffer - 1; /* -1: excluding the zero byte */
2910 =for apidoc sv_2pv_flags
2912 Returns a pointer to the string value of an SV, and sets C<*lp> to its length.
2913 If flags has the C<SV_GMAGIC> bit set, does an C<mg_get()> first. Coerces C<sv> to a
2914 string if necessary. Normally invoked via the C<SvPV_flags> macro.
2915 C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
2921 Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
2925 PERL_ARGS_ASSERT_SV_2PV_FLAGS;
2927 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2928 && SvTYPE(sv) != SVt_PVFM);
2929 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2934 if (flags & SV_SKIP_OVERLOAD)
2936 tmpstr = AMG_CALLunary(sv, string_amg);
2937 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
2938 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2940 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2944 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2945 if (flags & SV_CONST_RETURN) {
2946 pv = (char *) SvPVX_const(tmpstr);
2948 pv = (flags & SV_MUTABLE_RETURN)
2949 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
2952 *lp = SvCUR(tmpstr);
2954 pv = sv_2pv_flags(tmpstr, lp, flags);
2967 SV *const referent = SvRV(sv);
2971 retval = buffer = savepvn("NULLREF", len);
2972 } else if (SvTYPE(referent) == SVt_REGEXP &&
2973 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
2974 amagic_is_enabled(string_amg))) {
2975 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
2979 /* If the regex is UTF-8 we want the containing scalar to
2980 have an UTF-8 flag too */
2987 *lp = RX_WRAPLEN(re);
2989 return RX_WRAPPED(re);
2991 const char *const typestr = sv_reftype(referent, 0);
2992 const STRLEN typelen = strlen(typestr);
2993 UV addr = PTR2UV(referent);
2994 const char *stashname = NULL;
2995 STRLEN stashnamelen = 0; /* hush, gcc */
2996 const char *buffer_end;
2998 if (SvOBJECT(referent)) {
2999 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
3002 stashname = HEK_KEY(name);
3003 stashnamelen = HEK_LEN(name);
3005 if (HEK_UTF8(name)) {
3011 stashname = "__ANON__";
3014 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
3015 + 2 * sizeof(UV) + 2 /* )\0 */;
3017 len = typelen + 3 /* (0x */
3018 + 2 * sizeof(UV) + 2 /* )\0 */;
3021 Newx(buffer, len, char);
3022 buffer_end = retval = buffer + len;
3024 /* Working backwards */
3028 *--retval = PL_hexdigit[addr & 15];
3029 } while (addr >>= 4);
3035 memcpy(retval, typestr, typelen);
3039 retval -= stashnamelen;
3040 memcpy(retval, stashname, stashnamelen);
3042 /* retval may not necessarily have reached the start of the
3044 assert (retval >= buffer);
3046 len = buffer_end - retval - 1; /* -1 for that \0 */
3058 if (flags & SV_MUTABLE_RETURN)
3059 return SvPVX_mutable(sv);
3060 if (flags & SV_CONST_RETURN)
3061 return (char *)SvPVX_const(sv);
3066 /* I'm assuming that if both IV and NV are equally valid then
3067 converting the IV is going to be more efficient */
3068 const U32 isUIOK = SvIsUV(sv);
3069 char buf[TYPE_CHARS(UV)];
3073 if (SvTYPE(sv) < SVt_PVIV)
3074 sv_upgrade(sv, SVt_PVIV);
3075 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
3077 /* inlined from sv_setpvn */
3078 s = SvGROW_mutable(sv, len + 1);
3079 Move(ptr, s, len, char);
3084 else if (SvNOK(sv)) {
3085 if (SvTYPE(sv) < SVt_PVNV)
3086 sv_upgrade(sv, SVt_PVNV);
3087 if (SvNVX(sv) == 0.0
3088 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
3089 && !Perl_isnan(SvNVX(sv))
3092 s = SvGROW_mutable(sv, 2);
3097 STRLEN size = 5; /* "-Inf\0" */
3099 s = SvGROW_mutable(sv, size);
3100 len = S_infnan_2pv(SvNVX(sv), s, size, 0);
3106 /* some Xenix systems wipe out errno here */
3115 5 + /* exponent digits */
3119 s = SvGROW_mutable(sv, size);
3120 #ifndef USE_LOCALE_NUMERIC
3121 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3127 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
3128 STORE_LC_NUMERIC_SET_TO_NEEDED();
3130 local_radix = PL_numeric_local && PL_numeric_radix_sv;
3131 if (local_radix && SvLEN(PL_numeric_radix_sv) > 1) {
3132 size += SvLEN(PL_numeric_radix_sv) - 1;
3133 s = SvGROW_mutable(sv, size);
3136 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3138 /* If the radix character is UTF-8, and actually is in the
3139 * output, turn on the UTF-8 flag for the scalar */
3141 && SvUTF8(PL_numeric_radix_sv)
3142 && instr(s, SvPVX_const(PL_numeric_radix_sv)))
3147 RESTORE_LC_NUMERIC();
3150 /* We don't call SvPOK_on(), because it may come to
3151 * pass that the locale changes so that the
3152 * stringification we just did is no longer correct. We
3153 * will have to re-stringify every time it is needed */
3160 else if (isGV_with_GP(sv)) {
3161 GV *const gv = MUTABLE_GV(sv);
3162 SV *const buffer = sv_newmortal();
3164 gv_efullname3(buffer, gv, "*");
3166 assert(SvPOK(buffer));
3170 *lp = SvCUR(buffer);
3171 return SvPVX(buffer);
3173 else if (isREGEXP(sv)) {
3174 if (lp) *lp = RX_WRAPLEN((REGEXP *)sv);
3175 return RX_WRAPPED((REGEXP *)sv);
3180 if (flags & SV_UNDEF_RETURNS_NULL)
3182 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
3184 /* Typically the caller expects that sv_any is not NULL now. */
3185 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
3186 sv_upgrade(sv, SVt_PV);
3191 const STRLEN len = s - SvPVX_const(sv);
3196 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
3197 PTR2UV(sv),SvPVX_const(sv)));
3198 if (flags & SV_CONST_RETURN)
3199 return (char *)SvPVX_const(sv);
3200 if (flags & SV_MUTABLE_RETURN)
3201 return SvPVX_mutable(sv);
3206 =for apidoc sv_copypv
3208 Copies a stringified representation of the source SV into the
3209 destination SV. Automatically performs any necessary C<mg_get> and
3210 coercion of numeric values into strings. Guaranteed to preserve
3211 C<UTF8> flag even from overloaded objects. Similar in nature to
3212 C<sv_2pv[_flags]> but operates directly on an SV instead of just the
3213 string. Mostly uses C<sv_2pv_flags> to do its work, except when that
3214 would lose the UTF-8'ness of the PV.
3216 =for apidoc sv_copypv_nomg
3218 Like C<sv_copypv>, but doesn't invoke get magic first.
3220 =for apidoc sv_copypv_flags
3222 Implementation of C<sv_copypv> and C<sv_copypv_nomg>. Calls get magic iff flags
3223 has the C<SV_GMAGIC> bit set.
3229 Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
3234 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
3236 s = SvPV_flags_const(ssv,len,(flags & SV_GMAGIC));
3237 sv_setpvn(dsv,s,len);
3245 =for apidoc sv_2pvbyte
3247 Return a pointer to the byte-encoded representation of the SV, and set C<*lp>
3248 to its length. May cause the SV to be downgraded from UTF-8 as a
3251 Usually accessed via the C<SvPVbyte> macro.
3257 Perl_sv_2pvbyte(pTHX_ SV *sv, STRLEN *const lp)
3259 PERL_ARGS_ASSERT_SV_2PVBYTE;
3262 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3263 || isGV_with_GP(sv) || SvROK(sv)) {
3264 SV *sv2 = sv_newmortal();
3265 sv_copypv_nomg(sv2,sv);
3268 sv_utf8_downgrade(sv,0);
3269 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3273 =for apidoc sv_2pvutf8
3275 Return a pointer to the UTF-8-encoded representation of the SV, and set C<*lp>
3276 to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3278 Usually accessed via the C<SvPVutf8> macro.
3284 Perl_sv_2pvutf8(pTHX_ SV *sv, STRLEN *const lp)
3286 PERL_ARGS_ASSERT_SV_2PVUTF8;
3288 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3289 || isGV_with_GP(sv) || SvROK(sv))
3290 sv = sv_mortalcopy(sv);
3293 sv_utf8_upgrade_nomg(sv);
3294 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3299 =for apidoc sv_2bool
3301 This macro is only used by C<sv_true()> or its macro equivalent, and only if
3302 the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>.
3303 It calls C<sv_2bool_flags> with the C<SV_GMAGIC> flag.
3305 =for apidoc sv_2bool_flags
3307 This function is only used by C<sv_true()> and friends, and only if
3308 the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>. If the flags
3309 contain C<SV_GMAGIC>, then it does an C<mg_get()> first.
3316 Perl_sv_2bool_flags(pTHX_ SV *sv, I32 flags)
3318 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
3321 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
3327 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
3328 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv)))) {
3331 if(SvGMAGICAL(sv)) {
3333 goto restart; /* call sv_2bool */
3335 /* expanded SvTRUE_common(sv, (flags = 0, goto restart)) */
3336 else if(!SvOK(sv)) {
3339 else if(SvPOK(sv)) {
3340 svb = SvPVXtrue(sv);
3342 else if((SvFLAGS(sv) & (SVf_IOK|SVf_NOK))) {
3343 svb = (SvIOK(sv) && SvIVX(sv) != 0)
3344 || (SvNOK(sv) && SvNVX(sv) != 0.0);
3348 goto restart; /* call sv_2bool_nomg */
3353 return SvRV(sv) != 0;
3357 RX_WRAPLEN(sv) > 1 || (RX_WRAPLEN(sv) && *RX_WRAPPED(sv) != '0');
3358 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
3362 =for apidoc sv_utf8_upgrade
3364 Converts the PV of an SV to its UTF-8-encoded form.
3365 Forces the SV to string form if it is not already.
3366 Will C<mg_get> on C<sv> if appropriate.
3367 Always sets the C<SvUTF8> flag to avoid future validity checks even
3368 if the whole string is the same in UTF-8 as not.
3369 Returns the number of bytes in the converted string
3371 This is not a general purpose byte encoding to Unicode interface:
3372 use the Encode extension for that.
3374 =for apidoc sv_utf8_upgrade_nomg
3376 Like C<sv_utf8_upgrade>, but doesn't do magic on C<sv>.
3378 =for apidoc sv_utf8_upgrade_flags
3380 Converts the PV of an SV to its UTF-8-encoded form.
3381 Forces the SV to string form if it is not already.
3382 Always sets the SvUTF8 flag to avoid future validity checks even
3383 if all the bytes are invariant in UTF-8.
3384 If C<flags> has C<SV_GMAGIC> bit set,
3385 will C<mg_get> on C<sv> if appropriate, else not.
3387 If C<flags> has C<SV_FORCE_UTF8_UPGRADE> set, this function assumes that the PV
3388 will expand when converted to UTF-8, and skips the extra work of checking for
3389 that. Typically this flag is used by a routine that has already parsed the
3390 string and found such characters, and passes this information on so that the
3391 work doesn't have to be repeated.
3393 Returns the number of bytes in the converted string.
3395 This is not a general purpose byte encoding to Unicode interface:
3396 use the Encode extension for that.
3398 =for apidoc sv_utf8_upgrade_flags_grow
3400 Like C<sv_utf8_upgrade_flags>, but has an additional parameter C<extra>, which is
3401 the number of unused bytes the string of C<sv> is guaranteed to have free after
3402 it upon return. This allows the caller to reserve extra space that it intends
3403 to fill, to avoid extra grows.
3405 C<sv_utf8_upgrade>, C<sv_utf8_upgrade_nomg>, and C<sv_utf8_upgrade_flags>
3406 are implemented in terms of this function.
3408 Returns the number of bytes in the converted string (not including the spares).
3412 (One might think that the calling routine could pass in the position of the
3413 first variant character when it has set SV_FORCE_UTF8_UPGRADE, so it wouldn't
3414 have to be found again. But that is not the case, because typically when the
3415 caller is likely to use this flag, it won't be calling this routine unless it
3416 finds something that won't fit into a byte. Otherwise it tries to not upgrade
3417 and just use bytes. But some things that do fit into a byte are variants in
3418 utf8, and the caller may not have been keeping track of these.)
3420 If the routine itself changes the string, it adds a trailing C<NUL>. Such a
3421 C<NUL> isn't guaranteed due to having other routines do the work in some input
3422 cases, or if the input is already flagged as being in utf8.
3424 The speed of this could perhaps be improved for many cases if someone wanted to
3425 write a fast function that counts the number of variant characters in a string,
3426 especially if it could return the position of the first one.
3431 Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV *const sv, const I32 flags, STRLEN extra)
3433 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
3435 if (sv == &PL_sv_undef)
3437 if (!SvPOK_nog(sv)) {
3439 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3440 (void) sv_2pv_flags(sv,&len, flags);
3442 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3446 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
3451 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3456 S_sv_uncow(aTHX_ sv, 0);
3459 if (SvCUR(sv) == 0) {
3460 if (extra) SvGROW(sv, extra);
3461 } else { /* Assume Latin-1/EBCDIC */
3462 /* This function could be much more efficient if we
3463 * had a FLAG in SVs to signal if there are any variant
3464 * chars in the PV. Given that there isn't such a flag
3465 * make the loop as fast as possible (although there are certainly ways
3466 * to speed this up, eg. through vectorization) */
3467 U8 * s = (U8 *) SvPVX_const(sv);
3468 U8 * e = (U8 *) SvEND(sv);
3470 STRLEN two_byte_count = 0;
3472 if (flags & SV_FORCE_UTF8_UPGRADE) goto must_be_utf8;
3474 /* See if really will need to convert to utf8. We mustn't rely on our
3475 * incoming SV being well formed and having a trailing '\0', as certain
3476 * code in pp_formline can send us partially built SVs. */
3480 if (NATIVE_BYTE_IS_INVARIANT(ch)) continue;
3482 t--; /* t already incremented; re-point to first variant */
3487 /* utf8 conversion not needed because all are invariants. Mark as
3488 * UTF-8 even if no variant - saves scanning loop */
3490 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3495 /* Here, the string should be converted to utf8, either because of an
3496 * input flag (two_byte_count = 0), or because a character that
3497 * requires 2 bytes was found (two_byte_count = 1). t points either to
3498 * the beginning of the string (if we didn't examine anything), or to
3499 * the first variant. In either case, everything from s to t - 1 will
3500 * occupy only 1 byte each on output.
3502 * There are two main ways to convert. One is to create a new string
3503 * and go through the input starting from the beginning, appending each
3504 * converted value onto the new string as we go along. It's probably
3505 * best to allocate enough space in the string for the worst possible
3506 * case rather than possibly running out of space and having to
3507 * reallocate and then copy what we've done so far. Since everything
3508 * from s to t - 1 is invariant, the destination can be initialized
3509 * with these using a fast memory copy
3511 * The other way is to figure out exactly how big the string should be
3512 * by parsing the entire input. Then you don't have to make it big
3513 * enough to handle the worst possible case, and more importantly, if
3514 * the string you already have is large enough, you don't have to
3515 * allocate a new string, you can copy the last character in the input
3516 * string to the final position(s) that will be occupied by the
3517 * converted string and go backwards, stopping at t, since everything
3518 * before that is invariant.
3520 * There are advantages and disadvantages to each method.
3522 * In the first method, we can allocate a new string, do the memory
3523 * copy from the s to t - 1, and then proceed through the rest of the
3524 * string byte-by-byte.
3526 * In the second method, we proceed through the rest of the input
3527 * string just calculating how big the converted string will be. Then
3528 * there are two cases:
3529 * 1) if the string has enough extra space to handle the converted
3530 * value. We go backwards through the string, converting until we
3531 * get to the position we are at now, and then stop. If this
3532 * position is far enough along in the string, this method is
3533 * faster than the other method. If the memory copy were the same
3534 * speed as the byte-by-byte loop, that position would be about
3535 * half-way, as at the half-way mark, parsing to the end and back
3536 * is one complete string's parse, the same amount as starting
3537 * over and going all the way through. Actually, it would be
3538 * somewhat less than half-way, as it's faster to just count bytes
3539 * than to also copy, and we don't have the overhead of allocating
3540 * a new string, changing the scalar to use it, and freeing the
3541 * existing one. But if the memory copy is fast, the break-even
3542 * point is somewhere after half way. The counting loop could be
3543 * sped up by vectorization, etc, to move the break-even point
3544 * further towards the beginning.
3545 * 2) if the string doesn't have enough space to handle the converted
3546 * value. A new string will have to be allocated, and one might
3547 * as well, given that, start from the beginning doing the first
3548 * method. We've spent extra time parsing the string and in
3549 * exchange all we've gotten is that we know precisely how big to
3550 * make the new one. Perl is more optimized for time than space,
3551 * so this case is a loser.
3552 * So what I've decided to do is not use the 2nd method unless it is
3553 * guaranteed that a new string won't have to be allocated, assuming
3554 * the worst case. I also decided not to put any more conditions on it
3555 * than this, for now. It seems likely that, since the worst case is
3556 * twice as big as the unknown portion of the string (plus 1), we won't
3557 * be guaranteed enough space, causing us to go to the first method,
3558 * unless the string is short, or the first variant character is near
3559 * the end of it. In either of these cases, it seems best to use the
3560 * 2nd method. The only circumstance I can think of where this would
3561 * be really slower is if the string had once had much more data in it
3562 * than it does now, but there is still a substantial amount in it */
3565 STRLEN invariant_head = t - s;
3566 STRLEN size = invariant_head + (e - t) * 2 + 1 + extra;
3567 if (SvLEN(sv) < size) {
3569 /* Here, have decided to allocate a new string */
3574 Newx(dst, size, U8);
3576 /* If no known invariants at the beginning of the input string,
3577 * set so starts from there. Otherwise, can use memory copy to
3578 * get up to where we are now, and then start from here */
3580 if (invariant_head == 0) {
3583 Copy(s, dst, invariant_head, char);
3584 d = dst + invariant_head;
3588 append_utf8_from_native_byte(*t, &d);
3592 SvPV_free(sv); /* No longer using pre-existing string */
3593 SvPV_set(sv, (char*)dst);
3594 SvCUR_set(sv, d - dst);
3595 SvLEN_set(sv, size);
3598 /* Here, have decided to get the exact size of the string.
3599 * Currently this happens only when we know that there is
3600 * guaranteed enough space to fit the converted string, so
3601 * don't have to worry about growing. If two_byte_count is 0,
3602 * then t points to the first byte of the string which hasn't
3603 * been examined yet. Otherwise two_byte_count is 1, and t
3604 * points to the first byte in the string that will expand to
3605 * two. Depending on this, start examining at t or 1 after t.
3608 U8 *d = t + two_byte_count;
3611 /* Count up the remaining bytes that expand to two */
3614 const U8 chr = *d++;
3615 if (! NATIVE_BYTE_IS_INVARIANT(chr)) two_byte_count++;
3618 /* The string will expand by just the number of bytes that
3619 * occupy two positions. But we are one afterwards because of
3620 * the increment just above. This is the place to put the
3621 * trailing NUL, and to set the length before we decrement */
3623 d += two_byte_count;
3624 SvCUR_set(sv, d - s);
3628 /* Having decremented d, it points to the position to put the
3629 * very last byte of the expanded string. Go backwards through
3630 * the string, copying and expanding as we go, stopping when we
3631 * get to the part that is invariant the rest of the way down */
3635 if (NATIVE_BYTE_IS_INVARIANT(*e)) {
3638 *d-- = UTF8_EIGHT_BIT_LO(*e);
3639 *d-- = UTF8_EIGHT_BIT_HI(*e);
3645 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3646 /* Update pos. We do it at the end rather than during
3647 * the upgrade, to avoid slowing down the common case
3648 * (upgrade without pos).
3649 * pos can be stored as either bytes or characters. Since
3650 * this was previously a byte string we can just turn off
3651 * the bytes flag. */
3652 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3654 mg->mg_flags &= ~MGf_BYTES;
3656 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3657 magic_setutf8(sv,mg); /* clear UTF8 cache */
3662 /* Mark as UTF-8 even if no variant - saves scanning loop */
3668 =for apidoc sv_utf8_downgrade
3670 Attempts to convert the PV of an SV from characters to bytes.
3671 If the PV contains a character that cannot fit
3672 in a byte, this conversion will fail;
3673 in this case, either returns false or, if C<fail_ok> is not
3676 This is not a general purpose Unicode to byte encoding interface:
3677 use the C<Encode> extension for that.
3683 Perl_sv_utf8_downgrade(pTHX_ SV *const sv, const bool fail_ok)
3685 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3687 if (SvPOKp(sv) && SvUTF8(sv)) {
3691 int mg_flags = SV_GMAGIC;
3694 S_sv_uncow(aTHX_ sv, 0);
3696 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3698 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3699 if (mg && mg->mg_len > 0 && mg->mg_flags & MGf_BYTES) {
3700 mg->mg_len = sv_pos_b2u_flags(sv, mg->mg_len,
3701 SV_GMAGIC|SV_CONST_RETURN);
3702 mg_flags = 0; /* sv_pos_b2u does get magic */
3704 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3705 magic_setutf8(sv,mg); /* clear UTF8 cache */
3708 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3710 if (!utf8_to_bytes(s, &len)) {
3715 Perl_croak(aTHX_ "Wide character in %s",
3718 Perl_croak(aTHX_ "Wide character");
3729 =for apidoc sv_utf8_encode
3731 Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3732 flag off so that it looks like octets again.
3738 Perl_sv_utf8_encode(pTHX_ SV *const sv)
3740 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3742 if (SvREADONLY(sv)) {
3743 sv_force_normal_flags(sv, 0);
3745 (void) sv_utf8_upgrade(sv);
3750 =for apidoc sv_utf8_decode
3752 If the PV of the SV is an octet sequence in UTF-8
3753 and contains a multiple-byte character, the C<SvUTF8> flag is turned on
3754 so that it looks like a character. If the PV contains only single-byte
3755 characters, the C<SvUTF8> flag stays off.
3756 Scans PV for validity and returns false if the PV is invalid UTF-8.
3762 Perl_sv_utf8_decode(pTHX_ SV *const sv)
3764 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3767 const U8 *start, *c;
3769 /* The octets may have got themselves encoded - get them back as
3772 if (!sv_utf8_downgrade(sv, TRUE))
3775 /* it is actually just a matter of turning the utf8 flag on, but
3776 * we want to make sure everything inside is valid utf8 first.
3778 c = start = (const U8 *) SvPVX_const(sv);
3779 if (!is_utf8_string(c, SvCUR(sv)))
3781 if (! is_utf8_invariant_string(c, SvCUR(sv))) {
3784 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3785 /* XXX Is this dead code? XS_utf8_decode calls SvSETMAGIC
3786 after this, clearing pos. Does anything on CPAN
3788 /* adjust pos to the start of a UTF8 char sequence */
3789 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3791 I32 pos = mg->mg_len;
3793 for (c = start + pos; c > start; c--) {
3794 if (UTF8_IS_START(*c))
3797 mg->mg_len = c - start;
3800 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3801 magic_setutf8(sv,mg); /* clear UTF8 cache */
3808 =for apidoc sv_setsv
3810 Copies the contents of the source SV C<ssv> into the destination SV
3811 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3812 function if the source SV needs to be reused. Does not handle 'set' magic on
3813 destination SV. Calls 'get' magic on source SV. Loosely speaking, it
3814 performs a copy-by-value, obliterating any previous content of the
3817 You probably want to use one of the assortment of wrappers, such as
3818 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3819 C<SvSetMagicSV_nosteal>.
3821 =for apidoc sv_setsv_flags
3823 Copies the contents of the source SV C<ssv> into the destination SV
3824 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3825 function if the source SV needs to be reused. Does not handle 'set' magic.
3826 Loosely speaking, it performs a copy-by-value, obliterating any previous
3827 content of the destination.
3828 If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
3829 C<ssv> if appropriate, else not. If the C<flags>
3830 parameter has the C<SV_NOSTEAL> bit set then the
3831 buffers of temps will not be stolen. C<sv_setsv>
3832 and C<sv_setsv_nomg> are implemented in terms of this function.
3834 You probably want to use one of the assortment of wrappers, such as
3835 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3836 C<SvSetMagicSV_nosteal>.
3838 This is the primary function for copying scalars, and most other
3839 copy-ish functions and macros use this underneath.
3845 S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
3847 I32 mro_changes = 0; /* 1 = method, 2 = isa, 3 = recursive isa */
3848 HV *old_stash = NULL;
3850 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB;
3852 if (dtype != SVt_PVGV && !isGV_with_GP(dstr)) {
3853 const char * const name = GvNAME(sstr);
3854 const STRLEN len = GvNAMELEN(sstr);
3856 if (dtype >= SVt_PV) {
3862 SvUPGRADE(dstr, SVt_PVGV);
3863 (void)SvOK_off(dstr);
3864 isGV_with_GP_on(dstr);
3866 GvSTASH(dstr) = GvSTASH(sstr);
3868 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
3869 gv_name_set(MUTABLE_GV(dstr), name, len,
3870 GV_ADD | (GvNAMEUTF8(sstr) ? SVf_UTF8 : 0 ));
3871 SvFAKE_on(dstr); /* can coerce to non-glob */
3874 if(GvGP(MUTABLE_GV(sstr))) {
3875 /* If source has method cache entry, clear it */
3877 SvREFCNT_dec(GvCV(sstr));
3878 GvCV_set(sstr, NULL);
3881 /* If source has a real method, then a method is
3884 GvCV((const GV *)sstr) && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3890 /* If dest already had a real method, that's a change as well */
3892 !mro_changes && GvGP(MUTABLE_GV(dstr)) && GvCVu((const GV *)dstr)
3893 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3898 /* We don't need to check the name of the destination if it was not a
3899 glob to begin with. */
3900 if(dtype == SVt_PVGV) {
3901 const char * const name = GvNAME((const GV *)dstr);
3904 /* The stash may have been detached from the symbol table, so
3906 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3910 const STRLEN len = GvNAMELEN(dstr);
3911 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
3912 || (len == 1 && name[0] == ':')) {
3915 /* Set aside the old stash, so we can reset isa caches on
3917 if((old_stash = GvHV(dstr)))
3918 /* Make sure we do not lose it early. */
3919 SvREFCNT_inc_simple_void_NN(
3920 sv_2mortal((SV *)old_stash)
3925 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
3928 /* freeing dstr's GP might free sstr (e.g. *x = $x),
3929 * so temporarily protect it */
3931 SAVEFREESV(SvREFCNT_inc_simple_NN(sstr));
3932 gp_free(MUTABLE_GV(dstr));
3933 GvINTRO_off(dstr); /* one-shot flag */
3934 GvGP_set(dstr, gp_ref(GvGP(sstr)));
3937 if (SvTAINTED(sstr))
3939 if (GvIMPORTED(dstr) != GVf_IMPORTED
3940 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
3942 GvIMPORTED_on(dstr);
3945 if(mro_changes == 2) {
3946 if (GvAV((const GV *)sstr)) {
3948 SV * const sref = (SV *)GvAV((const GV *)dstr);
3949 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
3950 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
3951 AV * const ary = newAV();
3952 av_push(ary, mg->mg_obj); /* takes the refcount */
3953 mg->mg_obj = (SV *)ary;
3955 av_push((AV *)mg->mg_obj, SvREFCNT_inc_simple_NN(dstr));
3957 else sv_magic(sref, dstr, PERL_MAGIC_isa, NULL, 0);
3959 mro_isa_changed_in(GvSTASH(dstr));
3961 else if(mro_changes == 3) {
3962 HV * const stash = GvHV(dstr);
3963 if(old_stash ? (HV *)HvENAME_get(old_stash) : stash)
3969 else if(mro_changes) mro_method_changed_in(GvSTASH(dstr));
3970 if (GvIO(dstr) && dtype == SVt_PVGV) {
3971 DEBUG_o(Perl_deb(aTHX_
3972 "glob_assign_glob clearing PL_stashcache\n"));
3973 /* It's a cache. It will rebuild itself quite happily.
3974 It's a lot of effort to work out exactly which key (or keys)
3975 might be invalidated by the creation of the this file handle.
3977 hv_clear(PL_stashcache);
3983 Perl_gv_setref(pTHX_ SV *const dstr, SV *const sstr)
3985 SV * const sref = SvRV(sstr);
3987 const int intro = GvINTRO(dstr);
3990 const U32 stype = SvTYPE(sref);
3992 PERL_ARGS_ASSERT_GV_SETREF;
3995 GvINTRO_off(dstr); /* one-shot flag */
3996 GvLINE(dstr) = CopLINE(PL_curcop);
3997 GvEGV(dstr) = MUTABLE_GV(dstr);
4002 location = (SV **) &(GvGP(dstr)->gp_cv); /* XXX bypassing GvCV_set */
4003 import_flag = GVf_IMPORTED_CV;
4006 location = (SV **) &GvHV(dstr);
4007 import_flag = GVf_IMPORTED_HV;
4010 location = (SV **) &GvAV(dstr);
4011 import_flag = GVf_IMPORTED_AV;
4014 location = (SV **) &GvIOp(dstr);
4017 location = (SV **) &GvFORM(dstr);
4020 location = &GvSV(dstr);
4021 import_flag = GVf_IMPORTED_SV;
4024 if (stype == SVt_PVCV) {
4025 /*if (GvCVGEN(dstr) && (GvCV(dstr) != (const CV *)sref || GvCVGEN(dstr))) {*/
4026 if (GvCVGEN(dstr)) {
4027 SvREFCNT_dec(GvCV(dstr));
4028 GvCV_set(dstr, NULL);
4029 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4032 /* SAVEt_GVSLOT takes more room on the savestack and has more
4033 overhead in leave_scope than SAVEt_GENERIC_SV. But for CVs
4034 leave_scope needs access to the GV so it can reset method
4035 caches. We must use SAVEt_GVSLOT whenever the type is
4036 SVt_PVCV, even if the stash is anonymous, as the stash may
4037 gain a name somehow before leave_scope. */
4038 if (stype == SVt_PVCV) {
4039 /* There is no save_pushptrptrptr. Creating it for this
4040 one call site would be overkill. So inline the ss add
4044 SS_ADD_PTR(location);
4045 SS_ADD_PTR(SvREFCNT_inc(*location));
4046 SS_ADD_UV(SAVEt_GVSLOT);
4049 else SAVEGENERICSV(*location);
4052 if (stype == SVt_PVCV && (*location != sref || GvCVGEN(dstr))) {
4053 CV* const cv = MUTABLE_CV(*location);
4055 if (!GvCVGEN((const GV *)dstr) &&
4056 (CvROOT(cv) || CvXSUB(cv)) &&
4057 /* redundant check that avoids creating the extra SV
4058 most of the time: */
4059 (CvCONST(cv) || ckWARN(WARN_REDEFINE)))
4061 SV * const new_const_sv =
4062 CvCONST((const CV *)sref)
4063 ? cv_const_sv((const CV *)sref)
4065 HV * const stash = GvSTASH((const GV *)dstr);
4066 report_redefined_cv(
4069 ? Perl_newSVpvf(aTHX_
4071 HEKfARG(HvNAME_HEK(stash)),
4072 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr))))
4073 : Perl_newSVpvf(aTHX_
4075 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr))))
4078 CvCONST((const CV *)sref) ? &new_const_sv : NULL
4082 cv_ckproto_len_flags(cv, (const GV *)dstr,
4083 SvPOK(sref) ? CvPROTO(sref) : NULL,
4084 SvPOK(sref) ? CvPROTOLEN(sref) : 0,
4085 SvPOK(sref) ? SvUTF8(sref) : 0);
4087 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4088 GvASSUMECV_on(dstr);
4089 if(GvSTASH(dstr)) { /* sub foo { 1 } sub bar { 2 } *bar = \&foo */
4090 if (intro && GvREFCNT(dstr) > 1) {
4091 /* temporary remove extra savestack's ref */
4093 gv_method_changed(dstr);
4096 else gv_method_changed(dstr);
4099 *location = SvREFCNT_inc_simple_NN(sref);
4100 if (import_flag && !(GvFLAGS(dstr) & import_flag)
4101 && CopSTASH_ne(PL_curcop, GvSTASH(dstr))) {
4102 GvFLAGS(dstr) |= import_flag;
4105 if (stype == SVt_PVHV) {
4106 const char * const name = GvNAME((GV*)dstr);
4107 const STRLEN len = GvNAMELEN(dstr);
4110 (len > 1 && name[len-2] == ':' && name[len-1] == ':')
4111 || (len == 1 && name[0] == ':')
4113 && (!dref || HvENAME_get(dref))
4116 (HV *)sref, (HV *)dref,
4122 stype == SVt_PVAV && sref != dref
4123 && strEQ(GvNAME((GV*)dstr), "ISA")
4124 /* The stash may have been detached from the symbol table, so
4125 check its name before doing anything. */
4126 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
4129 MAGIC * const omg = dref && SvSMAGICAL(dref)
4130 ? mg_find(dref, PERL_MAGIC_isa)
4132 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
4133 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
4134 AV * const ary = newAV();
4135 av_push(ary, mg->mg_obj); /* takes the refcount */
4136 mg->mg_obj = (SV *)ary;
4139 if (SvTYPE(omg->mg_obj) == SVt_PVAV) {
4140 SV **svp = AvARRAY((AV *)omg->mg_obj);
4141 I32 items = AvFILLp((AV *)omg->mg_obj) + 1;
4145 SvREFCNT_inc_simple_NN(*svp++)
4151 SvREFCNT_inc_simple_NN(omg->mg_obj)
4155 av_push((AV *)mg->mg_obj,SvREFCNT_inc_simple_NN(dstr));
4161 sref, omg ? omg->mg_obj : dstr, PERL_MAGIC_isa, NULL, 0
4163 for (i = 0; i <= AvFILL(sref); ++i) {
4164 SV **elem = av_fetch ((AV*)sref, i, 0);
4167 *elem, sref, PERL_MAGIC_isaelem, NULL, i
4171 mg = mg_find(sref, PERL_MAGIC_isa);
4173 /* Since the *ISA assignment could have affected more than
4174 one stash, don't call mro_isa_changed_in directly, but let
4175 magic_clearisa do it for us, as it already has the logic for
4176 dealing with globs vs arrays of globs. */
4178 Perl_magic_clearisa(aTHX_ NULL, mg);
4180 else if (stype == SVt_PVIO) {
4181 DEBUG_o(Perl_deb(aTHX_ "gv_setref clearing PL_stashcache\n"));
4182 /* It's a cache. It will rebuild itself quite happily.
4183 It's a lot of effort to work out exactly which key (or keys)
4184 might be invalidated by the creation of the this file handle.
4186 hv_clear(PL_stashcache);
4190 if (!intro) SvREFCNT_dec(dref);
4191 if (SvTAINTED(sstr))
4199 #ifdef PERL_DEBUG_READONLY_COW
4200 # include <sys/mman.h>
4202 # ifndef PERL_MEMORY_DEBUG_HEADER_SIZE
4203 # define PERL_MEMORY_DEBUG_HEADER_SIZE 0
4207 Perl_sv_buf_to_ro(pTHX_ SV *sv)
4209 struct perl_memory_debug_header * const header =
4210 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4211 const MEM_SIZE len = header->size;
4212 PERL_ARGS_ASSERT_SV_BUF_TO_RO;
4213 # ifdef PERL_TRACK_MEMPOOL
4214 if (!header->readonly) header->readonly = 1;
4216 if (mprotect(header, len, PROT_READ))
4217 Perl_warn(aTHX_ "mprotect RW for COW string %p %lu failed with %d",
4218 header, len, errno);
4222 S_sv_buf_to_rw(pTHX_ SV *sv)
4224 struct perl_memory_debug_header * const header =
4225 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4226 const MEM_SIZE len = header->size;
4227 PERL_ARGS_ASSERT_SV_BUF_TO_RW;
4228 if (mprotect(header, len, PROT_READ|PROT_WRITE))
4229 Perl_warn(aTHX_ "mprotect for COW string %p %lu failed with %d",
4230 header, len, errno);
4231 # ifdef PERL_TRACK_MEMPOOL
4232 header->readonly = 0;
4237 # define sv_buf_to_ro(sv) NOOP
4238 # define sv_buf_to_rw(sv) NOOP
4242 Perl_sv_setsv_flags(pTHX_ SV *dstr, SV* sstr, const I32 flags)
4247 unsigned int both_type;
4249 PERL_ARGS_ASSERT_SV_SETSV_FLAGS;
4251 if (UNLIKELY( sstr == dstr ))
4254 if (UNLIKELY( !sstr ))
4255 sstr = &PL_sv_undef;
4257 stype = SvTYPE(sstr);
4258 dtype = SvTYPE(dstr);
4259 both_type = (stype | dtype);
4261 /* with these values, we can check that both SVs are NULL/IV (and not
4262 * freed) just by testing the or'ed types */
4263 STATIC_ASSERT_STMT(SVt_NULL == 0);
4264 STATIC_ASSERT_STMT(SVt_IV == 1);
4265 if (both_type <= 1) {
4266 /* both src and dst are UNDEF/IV/RV, so we can do a lot of
4271 /* minimal subset of SV_CHECK_THINKFIRST_COW_DROP(dstr) */
4272 if (SvREADONLY(dstr))
4273 Perl_croak_no_modify();
4275 sv_unref_flags(dstr, 0);
4277 assert(!SvGMAGICAL(sstr));
4278 assert(!SvGMAGICAL(dstr));
4280 sflags = SvFLAGS(sstr);
4281 if (sflags & (SVf_IOK|SVf_ROK)) {
4282 SET_SVANY_FOR_BODYLESS_IV(dstr);
4283 new_dflags = SVt_IV;
4285 if (sflags & SVf_ROK) {
4286 dstr->sv_u.svu_rv = SvREFCNT_inc(SvRV(sstr));
4287 new_dflags |= SVf_ROK;
4290 /* both src and dst are <= SVt_IV, so sv_any points to the
4291 * head; so access the head directly
4293 assert( &(sstr->sv_u.svu_iv)
4294 == &(((XPVIV*) SvANY(sstr))->xiv_iv));
4295 assert( &(dstr->sv_u.svu_iv)
4296 == &(((XPVIV*) SvANY(dstr))->xiv_iv));
4297 dstr->sv_u.svu_iv = sstr->sv_u.svu_iv;
4298 new_dflags |= (SVf_IOK|SVp_IOK|(sflags & SVf_IVisUV));
4302 new_dflags = dtype; /* turn off everything except the type */
4304 SvFLAGS(dstr) = new_dflags;
4309 if (UNLIKELY(both_type == SVTYPEMASK)) {
4310 if (SvIS_FREED(dstr)) {
4311 Perl_croak(aTHX_ "panic: attempt to copy value %" SVf
4312 " to a freed scalar %p", SVfARG(sstr), (void *)dstr);
4314 if (SvIS_FREED(sstr)) {
4315 Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p to %p",
4316 (void*)sstr, (void*)dstr);
4322 SV_CHECK_THINKFIRST_COW_DROP(dstr);
4323 dtype = SvTYPE(dstr); /* THINKFIRST may have changed type */
4325 /* There's a lot of redundancy below but we're going for speed here */
4330 if (LIKELY( dtype != SVt_PVGV && dtype != SVt_PVLV )) {
4331 (void)SvOK_off(dstr);
4339 /* For performance, we inline promoting to type SVt_IV. */
4340 /* We're starting from SVt_NULL, so provided that define is
4341 * actual 0, we don't have to unset any SV type flags
4342 * to promote to SVt_IV. */
4343 STATIC_ASSERT_STMT(SVt_NULL == 0);
4344 SET_SVANY_FOR_BODYLESS_IV(dstr);
4345 SvFLAGS(dstr) |= SVt_IV;
4349 sv_upgrade(dstr, SVt_PVIV);
4353 goto end_of_first_switch;
4355 (void)SvIOK_only(dstr);
4356 SvIV_set(dstr, SvIVX(sstr));
4359 /* SvTAINTED can only be true if the SV has taint magic, which in
4360 turn means that the SV type is PVMG (or greater). This is the
4361 case statement for SVt_IV, so this cannot be true (whatever gcov
4363 assert(!SvTAINTED(sstr));
4368 if (dtype < SVt_PV && dtype != SVt_IV)
4369 sv_upgrade(dstr, SVt_IV);
4373 if (LIKELY( SvNOK(sstr) )) {
4377 sv_upgrade(dstr, SVt_NV);
4381 sv_upgrade(dstr, SVt_PVNV);
4385 goto end_of_first_switch;
4387 SvNV_set(dstr, SvNVX(sstr));
4388 (void)SvNOK_only(dstr);
4389 /* SvTAINTED can only be true if the SV has taint magic, which in
4390 turn means that the SV type is PVMG (or greater). This is the
4391 case statement for SVt_NV, so this cannot be true (whatever gcov
4393 assert(!SvTAINTED(sstr));
4400 sv_upgrade(dstr, SVt_PV);
4403 if (dtype < SVt_PVIV)
4404 sv_upgrade(dstr, SVt_PVIV);
4407 if (dtype < SVt_PVNV)
4408 sv_upgrade(dstr, SVt_PVNV);
4412 const char * const type = sv_reftype(sstr,0);
4414 /* diag_listed_as: Bizarre copy of %s */
4415 Perl_croak(aTHX_ "Bizarre copy of %s in %s", type, OP_DESC(PL_op));
4417 Perl_croak(aTHX_ "Bizarre copy of %s", type);
4419 NOT_REACHED; /* NOTREACHED */
4423 if (dtype < SVt_REGEXP)
4425 if (dtype >= SVt_PV) {
4431 sv_upgrade(dstr, SVt_REGEXP);
4439 if (SvGMAGICAL(sstr) && (flags & SV_GMAGIC)) {
4441 if (SvTYPE(sstr) != stype)
4442 stype = SvTYPE(sstr);
4444 if (isGV_with_GP(sstr) && dtype <= SVt_PVLV) {
4445 glob_assign_glob(dstr, sstr, dtype);
4448 if (stype == SVt_PVLV)
4450 if (isREGEXP(sstr)) goto upgregexp;
4451 SvUPGRADE(dstr, SVt_PVNV);
4454 SvUPGRADE(dstr, (svtype)stype);
4456 end_of_first_switch:
4458 /* dstr may have been upgraded. */
4459 dtype = SvTYPE(dstr);
4460 sflags = SvFLAGS(sstr);
4462 if (UNLIKELY( dtype == SVt_PVCV )) {
4463 /* Assigning to a subroutine sets the prototype. */
4466 const char *const ptr = SvPV_const(sstr, len);
4468 SvGROW(dstr, len + 1);
4469 Copy(ptr, SvPVX(dstr), len + 1, char);
4470 SvCUR_set(dstr, len);
4472 SvFLAGS(dstr) |= sflags & SVf_UTF8;
4473 CvAUTOLOAD_off(dstr);
4478 else if (UNLIKELY(dtype == SVt_PVAV || dtype == SVt_PVHV
4479 || dtype == SVt_PVFM))
4481 const char * const type = sv_reftype(dstr,0);
4483 /* diag_listed_as: Cannot copy to %s */
4484 Perl_croak(aTHX_ "Cannot copy to %s in %s", type, OP_DESC(PL_op));
4486 Perl_croak(aTHX_ "Cannot copy to %s", type);
4487 } else if (sflags & SVf_ROK) {
4488 if (isGV_with_GP(dstr)
4489 && SvTYPE(SvRV(sstr)) == SVt_PVGV && isGV_with_GP(SvRV(sstr))) {
4492 if (GvIMPORTED(dstr) != GVf_IMPORTED
4493 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
4495 GvIMPORTED_on(dstr);
4500 glob_assign_glob(dstr, sstr, dtype);
4504 if (dtype >= SVt_PV) {
4505 if (isGV_with_GP(dstr)) {
4506 gv_setref(dstr, sstr);
4509 if (SvPVX_const(dstr)) {
4515 (void)SvOK_off(dstr);
4516 SvRV_set(dstr, SvREFCNT_inc(SvRV(sstr)));
4517 SvFLAGS(dstr) |= sflags & SVf_ROK;
4518 assert(!(sflags & SVp_NOK));
4519 assert(!(sflags & SVp_IOK));
4520 assert(!(sflags & SVf_NOK));
4521 assert(!(sflags & SVf_IOK));
4523 else if (isGV_with_GP(dstr)) {
4524 if (!(sflags & SVf_OK)) {
4525 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4526 "Undefined value assigned to typeglob");
4529 GV *gv = gv_fetchsv_nomg(sstr, GV_ADD, SVt_PVGV);
4530 if (dstr != (const SV *)gv) {
4531 const char * const name = GvNAME((const GV *)dstr);
4532 const STRLEN len = GvNAMELEN(dstr);
4533 HV *old_stash = NULL;
4534 bool reset_isa = FALSE;
4535 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
4536 || (len == 1 && name[0] == ':')) {
4537 /* Set aside the old stash, so we can reset isa caches
4538 on its subclasses. */
4539 if((old_stash = GvHV(dstr))) {
4540 /* Make sure we do not lose it early. */
4541 SvREFCNT_inc_simple_void_NN(
4542 sv_2mortal((SV *)old_stash)
4549 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
4550 gp_free(MUTABLE_GV(dstr));
4552 GvGP_set(dstr, gp_ref(GvGP(gv)));
4555 HV * const stash = GvHV(dstr);
4557 old_stash ? (HV *)HvENAME_get(old_stash) : stash
4567 else if ((dtype == SVt_REGEXP || dtype == SVt_PVLV)
4568 && (stype == SVt_REGEXP || isREGEXP(sstr))) {
4569 reg_temp_copy((REGEXP*)dstr, (REGEXP*)sstr);
4571 else if (sflags & SVp_POK) {
4572 const STRLEN cur = SvCUR(sstr);
4573 const STRLEN len = SvLEN(sstr);
4576 * We have three basic ways to copy the string:
4582 * Which we choose is based on various factors. The following
4583 * things are listed in order of speed, fastest to slowest:
4585 * - Copying a short string
4586 * - Copy-on-write bookkeeping
4588 * - Copying a long string
4590 * We swipe the string (steal the string buffer) if the SV on the
4591 * rhs is about to be freed anyway (TEMP and refcnt==1). This is a
4592 * big win on long strings. It should be a win on short strings if
4593 * SvPVX_const(dstr) has to be allocated. If not, it should not
4594 * slow things down, as SvPVX_const(sstr) would have been freed
4597 * We also steal the buffer from a PADTMP (operator target) if it
4598 * is ‘long enough’. For short strings, a swipe does not help
4599 * here, as it causes more malloc calls the next time the target
4600 * is used. Benchmarks show that even if SvPVX_const(dstr) has to
4601 * be allocated it is still not worth swiping PADTMPs for short
4602 * strings, as the savings here are small.
4604 * If swiping is not an option, then we see whether it is
4605 * worth using copy-on-write. If the lhs already has a buf-
4606 * fer big enough and the string is short, we skip it and fall back
4607 * to method 3, since memcpy is faster for short strings than the
4608 * later bookkeeping overhead that copy-on-write entails.
4610 * If the rhs is not a copy-on-write string yet, then we also
4611 * consider whether the buffer is too large relative to the string
4612 * it holds. Some operations such as readline allocate a large
4613 * buffer in the expectation of reusing it. But turning such into
4614 * a COW buffer is counter-productive because it increases memory
4615 * usage by making readline allocate a new large buffer the sec-
4616 * ond time round. So, if the buffer is too large, again, we use
4619 * Finally, if there is no buffer on the left, or the buffer is too
4620 * small, then we use copy-on-write and make both SVs share the
4625 /* Whichever path we take through the next code, we want this true,
4626 and doing it now facilitates the COW check. */
4627 (void)SvPOK_only(dstr);
4631 /* slated for free anyway (and not COW)? */
4632 (sflags & (SVs_TEMP|SVf_IsCOW)) == SVs_TEMP
4633 /* or a swipable TARG */
4635 (SVs_PADTMP|SVf_READONLY|SVf_PROTECT|SVf_IsCOW))
4637 /* whose buffer is worth stealing */
4638 && CHECK_COWBUF_THRESHOLD(cur,len)
4641 !(sflags & SVf_OOK) && /* and not involved in OOK hack? */
4642 (!(flags & SV_NOSTEAL)) &&
4643 /* and we're allowed to steal temps */
4644 SvREFCNT(sstr) == 1 && /* and no other references to it? */
4645 len) /* and really is a string */
4646 { /* Passes the swipe test. */
4647 if (SvPVX_const(dstr)) /* we know that dtype >= SVt_PV */
4649 SvPV_set(dstr, SvPVX_mutable(sstr));
4650 SvLEN_set(dstr, SvLEN(sstr));
4651 SvCUR_set(dstr, SvCUR(sstr));
4654 (void)SvOK_off(sstr); /* NOTE: nukes most SvFLAGS on sstr */
4655 SvPV_set(sstr, NULL);
4660 else if (flags & SV_COW_SHARED_HASH_KEYS
4662 #ifdef PERL_COPY_ON_WRITE
4665 ( (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4666 /* If this is a regular (non-hek) COW, only so
4667 many COW "copies" are possible. */
4668 && CowREFCNT(sstr) != SV_COW_REFCNT_MAX ))
4669 : ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4670 && !(SvFLAGS(dstr) & SVf_BREAK)
4671 && CHECK_COW_THRESHOLD(cur,len) && cur+1 < len
4672 && (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4676 && !(SvFLAGS(dstr) & SVf_BREAK)
4679 /* Either it's a shared hash key, or it's suitable for
4682 PerlIO_printf(Perl_debug_log, "Copy on write: sstr --> dstr\n");
4687 if (!(sflags & SVf_IsCOW)) {
4689 CowREFCNT(sstr) = 0;
4692 if (SvPVX_const(dstr)) { /* we know that dtype >= SVt_PV */
4698 if (sflags & SVf_IsCOW) {
4702 SvPV_set(dstr, SvPVX_mutable(sstr));
4707 /* SvIsCOW_shared_hash */
4708 DEBUG_C(PerlIO_printf(Perl_debug_log,
4709 "Copy on write: Sharing hash\n"));
4711 assert (SvTYPE(dstr) >= SVt_PV);
4713 HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)))));
4715 SvLEN_set(dstr, len);
4716 SvCUR_set(dstr, cur);
4719 /* Failed the swipe test, and we cannot do copy-on-write either.
4720 Have to copy the string. */
4721 SvGROW(dstr, cur + 1); /* inlined from sv_setpvn */
4722 Move(SvPVX_const(sstr),SvPVX(dstr),cur,char);
4723 SvCUR_set(dstr, cur);
4724 *SvEND(dstr) = '\0';
4726 if (sflags & SVp_NOK) {
4727 SvNV_set(dstr, SvNVX(sstr));
4729 if (sflags & SVp_IOK) {
4730 SvIV_set(dstr, SvIVX(sstr));
4731 /* Must do this otherwise some other overloaded use of 0x80000000
4732 gets confused. I guess SVpbm_VALID */
4733 if (sflags & SVf_IVisUV)
4736 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_NOK|SVp_NOK|SVf_UTF8);
4738 const MAGIC * const smg = SvVSTRING_mg(sstr);
4740 sv_magic(dstr, NULL, PERL_MAGIC_vstring,
4741 smg->mg_ptr, smg->mg_len);
4742 SvRMAGICAL_on(dstr);
4746 else if (sflags & (SVp_IOK|SVp_NOK)) {
4747 (void)SvOK_off(dstr);
4748 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_IVisUV|SVf_NOK|SVp_NOK);
4749 if (sflags & SVp_IOK) {
4750 /* XXXX Do we want to set IsUV for IV(ROK)? Be extra safe... */
4751 SvIV_set(dstr, SvIVX(sstr));
4753 if (sflags & SVp_NOK) {
4754 SvNV_set(dstr, SvNVX(sstr));
4758 if (isGV_with_GP(sstr)) {
4759 gv_efullname3(dstr, MUTABLE_GV(sstr), "*");
4762 (void)SvOK_off(dstr);
4764 if (SvTAINTED(sstr))
4769 =for apidoc sv_setsv_mg
4771 Like C<sv_setsv>, but also handles 'set' magic.
4777 Perl_sv_setsv_mg(pTHX_ SV *const dstr, SV *const sstr)
4779 PERL_ARGS_ASSERT_SV_SETSV_MG;
4781 sv_setsv(dstr,sstr);
4786 # define SVt_COW SVt_PV
4788 Perl_sv_setsv_cow(pTHX_ SV *dstr, SV *sstr)
4790 STRLEN cur = SvCUR(sstr);
4791 STRLEN len = SvLEN(sstr);
4793 #if defined(PERL_DEBUG_READONLY_COW) && defined(PERL_COPY_ON_WRITE)
4794 const bool already = cBOOL(SvIsCOW(sstr));
4797 PERL_ARGS_ASSERT_SV_SETSV_COW;
4800 PerlIO_printf(Perl_debug_log, "Fast copy on write: %p -> %p\n",
4801 (void*)sstr, (void*)dstr);
4808 if (SvTHINKFIRST(dstr))
4809 sv_force_normal_flags(dstr, SV_COW_DROP_PV);
4810 else if (SvPVX_const(dstr))
4811 Safefree(SvPVX_mutable(dstr));
4815 SvUPGRADE(dstr, SVt_COW);
4817 assert (SvPOK(sstr));
4818 assert (SvPOKp(sstr));
4820 if (SvIsCOW(sstr)) {
4822 if (SvLEN(sstr) == 0) {
4823 /* source is a COW shared hash key. */
4824 DEBUG_C(PerlIO_printf(Perl_debug_log,
4825 "Fast copy on write: Sharing hash\n"));
4826 new_pv = HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr))));
4829 assert(SvCUR(sstr)+1 < SvLEN(sstr));
4830 assert(CowREFCNT(sstr) < SV_COW_REFCNT_MAX);
4832 assert ((SvFLAGS(sstr) & CAN_COW_MASK) == CAN_COW_FLAGS);
4833 SvUPGRADE(sstr, SVt_COW);
4835 DEBUG_C(PerlIO_printf(Perl_debug_log,
4836 "Fast copy on write: Converting sstr to COW\n"));
4837 CowREFCNT(sstr) = 0;
4839 # ifdef PERL_DEBUG_READONLY_COW
4840 if (already) sv_buf_to_rw(sstr);
4843 new_pv = SvPVX_mutable(sstr);
4847 SvPV_set(dstr, new_pv);
4848 SvFLAGS(dstr) = (SVt_COW|SVf_POK|SVp_POK|SVf_IsCOW);
4851 SvLEN_set(dstr, len);
4852 SvCUR_set(dstr, cur);
4861 =for apidoc sv_setpvn
4863 Copies a string (possibly containing embedded C<NUL> characters) into an SV.
4864 The C<len> parameter indicates the number of
4865 bytes to be copied. If the C<ptr> argument is NULL the SV will become
4866 undefined. Does not handle 'set' magic. See C<L</sv_setpvn_mg>>.
4872 Perl_sv_setpvn(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4876 PERL_ARGS_ASSERT_SV_SETPVN;
4878 SV_CHECK_THINKFIRST_COW_DROP(sv);
4884 /* len is STRLEN which is unsigned, need to copy to signed */
4887 Perl_croak(aTHX_ "panic: sv_setpvn called with negative strlen %"
4890 SvUPGRADE(sv, SVt_PV);
4892 dptr = SvGROW(sv, len + 1);
4893 Move(ptr,dptr,len,char);
4896 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4898 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4902 =for apidoc sv_setpvn_mg
4904 Like C<sv_setpvn>, but also handles 'set' magic.
4910 Perl_sv_setpvn_mg(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4912 PERL_ARGS_ASSERT_SV_SETPVN_MG;
4914 sv_setpvn(sv,ptr,len);
4919 =for apidoc sv_setpv
4921 Copies a string into an SV. The string must be terminated with a C<NUL>
4922 character, and not contain embeded C<NUL>'s.
4923 Does not handle 'set' magic. See C<L</sv_setpv_mg>>.
4929 Perl_sv_setpv(pTHX_ SV *const sv, const char *const ptr)
4933 PERL_ARGS_ASSERT_SV_SETPV;
4935 SV_CHECK_THINKFIRST_COW_DROP(sv);
4941 SvUPGRADE(sv, SVt_PV);
4943 SvGROW(sv, len + 1);
4944 Move(ptr,SvPVX(sv),len+1,char);
4946 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4948 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4952 =for apidoc sv_setpv_mg
4954 Like C<sv_setpv>, but also handles 'set' magic.
4960 Perl_sv_setpv_mg(pTHX_ SV *const sv, const char *const ptr)
4962 PERL_ARGS_ASSERT_SV_SETPV_MG;
4969 Perl_sv_sethek(pTHX_ SV *const sv, const HEK *const hek)
4971 PERL_ARGS_ASSERT_SV_SETHEK;
4977 if (HEK_LEN(hek) == HEf_SVKEY) {
4978 sv_setsv(sv, *(SV**)HEK_KEY(hek));
4981 const int flags = HEK_FLAGS(hek);
4982 if (flags & HVhek_WASUTF8) {
4983 STRLEN utf8_len = HEK_LEN(hek);
4984 char *as_utf8 = (char *)bytes_to_utf8((U8*)HEK_KEY(hek), &utf8_len);
4985 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
4988 } else if (flags & HVhek_UNSHARED) {
4989 sv_setpvn(sv, HEK_KEY(hek), HEK_LEN(hek));
4992 else SvUTF8_off(sv);
4996 SV_CHECK_THINKFIRST_COW_DROP(sv);
4997 SvUPGRADE(sv, SVt_PV);
4999 SvPV_set(sv,(char *)HEK_KEY(share_hek_hek(hek)));
5000 SvCUR_set(sv, HEK_LEN(hek));
5006 else SvUTF8_off(sv);
5014 =for apidoc sv_usepvn_flags
5016 Tells an SV to use C<ptr> to find its string value. Normally the
5017 string is stored inside the SV, but sv_usepvn allows the SV to use an
5018 outside string. C<ptr> should point to memory that was allocated
5019 by L<C<Newx>|perlclib/Memory Management and String Handling>. It must be
5020 the start of a C<Newx>-ed block of memory, and not a pointer to the
5021 middle of it (beware of L<C<OOK>|perlguts/Offsets> and copy-on-write),
5022 and not be from a non-C<Newx> memory allocator like C<malloc>. The
5023 string length, C<len>, must be supplied. By default this function
5024 will C<Renew> (i.e. realloc, move) the memory pointed to by C<ptr>,
5025 so that pointer should not be freed or used by the programmer after
5026 giving it to C<sv_usepvn>, and neither should any pointers from "behind"
5027 that pointer (e.g. ptr + 1) be used.
5029 If S<C<flags & SV_SMAGIC>> is true, will call C<SvSETMAGIC>. If
5030 S<C<flags> & SV_HAS_TRAILING_NUL>> is true, then C<ptr[len]> must be C<NUL>,
5032 will be skipped (i.e. the buffer is actually at least 1 byte longer than
5033 C<len>, and already meets the requirements for storing in C<SvPVX>).
5039 Perl_sv_usepvn_flags(pTHX_ SV *const sv, char *ptr, const STRLEN len, const U32 flags)
5043 PERL_ARGS_ASSERT_SV_USEPVN_FLAGS;
5045 SV_CHECK_THINKFIRST_COW_DROP(sv);
5046 SvUPGRADE(sv, SVt_PV);
5049 if (flags & SV_SMAGIC)
5053 if (SvPVX_const(sv))
5057 if (flags & SV_HAS_TRAILING_NUL)
5058 assert(ptr[len] == '\0');
5061 allocate = (flags & SV_HAS_TRAILING_NUL)
5063 #ifdef Perl_safesysmalloc_size
5066 PERL_STRLEN_ROUNDUP(len + 1);
5068 if (flags & SV_HAS_TRAILING_NUL) {
5069 /* It's long enough - do nothing.
5070 Specifically Perl_newCONSTSUB is relying on this. */
5073 /* Force a move to shake out bugs in callers. */
5074 char *new_ptr = (char*)safemalloc(allocate);
5075 Copy(ptr, new_ptr, len, char);
5076 PoisonFree(ptr,len,char);
5080 ptr = (char*) saferealloc (ptr, allocate);
5083 #ifdef Perl_safesysmalloc_size
5084 SvLEN_set(sv, Perl_safesysmalloc_size(ptr));
5086 SvLEN_set(sv, allocate);
5090 if (!(flags & SV_HAS_TRAILING_NUL)) {
5093 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5095 if (flags & SV_SMAGIC)
5100 =for apidoc sv_force_normal_flags
5102 Undo various types of fakery on an SV, where fakery means
5103 "more than" a string: if the PV is a shared string, make
5104 a private copy; if we're a ref, stop refing; if we're a glob, downgrade to
5105 an C<xpvmg>; if we're a copy-on-write scalar, this is the on-write time when
5106 we do the copy, and is also used locally; if this is a
5107 vstring, drop the vstring magic. If C<SV_COW_DROP_PV> is set
5108 then a copy-on-write scalar drops its PV buffer (if any) and becomes
5109 C<SvPOK_off> rather than making a copy. (Used where this
5110 scalar is about to be set to some other value.) In addition,
5111 the C<flags> parameter gets passed to C<sv_unref_flags()>
5112 when unreffing. C<sv_force_normal> calls this function
5113 with flags set to 0.
5115 This function is expected to be used to signal to perl that this SV is
5116 about to be written to, and any extra book-keeping needs to be taken care
5117 of. Hence, it croaks on read-only values.
5123 S_sv_uncow(pTHX_ SV * const sv, const U32 flags)
5125 assert(SvIsCOW(sv));
5128 const char * const pvx = SvPVX_const(sv);
5129 const STRLEN len = SvLEN(sv);
5130 const STRLEN cur = SvCUR(sv);
5133 PerlIO_printf(Perl_debug_log,
5134 "Copy on write: Force normal %ld\n",
5139 # ifdef PERL_COPY_ON_WRITE
5141 /* Must do this first, since the CowREFCNT uses SvPVX and
5142 we need to write to CowREFCNT, or de-RO the whole buffer if we are
5143 the only owner left of the buffer. */
5144 sv_buf_to_rw(sv); /* NOOP if RO-ing not supported */
5146 U8 cowrefcnt = CowREFCNT(sv);
5147 if(cowrefcnt != 0) {
5149 CowREFCNT(sv) = cowrefcnt;
5154 /* Else we are the only owner of the buffer. */
5159 /* This SV doesn't own the buffer, so need to Newx() a new one: */
5164 if (flags & SV_COW_DROP_PV) {
5165 /* OK, so we don't need to copy our buffer. */
5168 SvGROW(sv, cur + 1);
5169 Move(pvx,SvPVX(sv),cur,char);
5175 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5182 const char * const pvx = SvPVX_const(sv);
5183 const STRLEN len = SvCUR(sv);
5187 if (flags & SV_COW_DROP_PV) {
5188 /* OK, so we don't need to copy our buffer. */
5191 SvGROW(sv, len + 1);
5192 Move(pvx,SvPVX(sv),len,char);
5195 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5201 Perl_sv_force_normal_flags(pTHX_ SV *const sv, const U32 flags)
5203 PERL_ARGS_ASSERT_SV_FORCE_NORMAL_FLAGS;
5206 Perl_croak_no_modify();
5207 else if (SvIsCOW(sv) && LIKELY(SvTYPE(sv) != SVt_PVHV))
5208 S_sv_uncow(aTHX_ sv, flags);
5210 sv_unref_flags(sv, flags);
5211 else if (SvFAKE(sv) && isGV_with_GP(sv))
5212 sv_unglob(sv, flags);
5213 else if (SvFAKE(sv) && isREGEXP(sv)) {
5214 /* Need to downgrade the REGEXP to a simple(r) scalar. This is analogous
5215 to sv_unglob. We only need it here, so inline it. */
5216 const bool islv = SvTYPE(sv) == SVt_PVLV;
5217 const svtype new_type =
5218 islv ? SVt_NULL : SvMAGIC(sv) || SvSTASH(sv) ? SVt_PVMG : SVt_PV;
5219 SV *const temp = newSV_type(new_type);
5220 regexp *const temp_p = ReANY((REGEXP *)sv);
5222 if (new_type == SVt_PVMG) {
5223 SvMAGIC_set(temp, SvMAGIC(sv));
5224 SvMAGIC_set(sv, NULL);
5225 SvSTASH_set(temp, SvSTASH(sv));
5226 SvSTASH_set(sv, NULL);
5228 if (!islv) SvCUR_set(temp, SvCUR(sv));
5229 /* Remember that SvPVX is in the head, not the body. But
5230 RX_WRAPPED is in the body. */
5231 assert(ReANY((REGEXP *)sv)->mother_re);
5232 /* Their buffer is already owned by someone else. */
5233 if (flags & SV_COW_DROP_PV) {
5234 /* SvLEN is already 0. For SVt_REGEXP, we have a brand new
5235 zeroed body. For SVt_PVLV, it should have been set to 0
5236 before turning into a regexp. */
5237 assert(!SvLEN(islv ? sv : temp));
5238 sv->sv_u.svu_pv = 0;
5241 sv->sv_u.svu_pv = savepvn(RX_WRAPPED((REGEXP *)sv), SvCUR(sv));
5242 SvLEN_set(islv ? sv : temp, SvCUR(sv)+1);
5246 /* Now swap the rest of the bodies. */
5250 SvFLAGS(sv) &= ~SVTYPEMASK;
5251 SvFLAGS(sv) |= new_type;
5252 SvANY(sv) = SvANY(temp);
5255 SvFLAGS(temp) &= ~(SVTYPEMASK);
5256 SvFLAGS(temp) |= SVt_REGEXP|SVf_FAKE;
5257 SvANY(temp) = temp_p;
5258 temp->sv_u.svu_rx = (regexp *)temp_p;
5260 SvREFCNT_dec_NN(temp);
5262 else if (SvVOK(sv)) sv_unmagic(sv, PERL_MAGIC_vstring);
5268 Efficient removal of characters from the beginning of the string buffer.
5269 C<SvPOK(sv)>, or at least C<SvPOKp(sv)>, must be true and C<ptr> must be a
5270 pointer to somewhere inside the string buffer. C<ptr> becomes the first
5271 character of the adjusted string. Uses the C<OOK> hack. On return, only
5272 C<SvPOK(sv)> and C<SvPOKp(sv)> among the C<OK> flags will be true.
5274 Beware: after this function returns, C<ptr> and SvPVX_const(sv) may no longer
5275 refer to the same chunk of data.
5277 The unfortunate similarity of this function's name to that of Perl's C<chop>
5278 operator is strictly coincidental. This function works from the left;
5279 C<chop> works from the right.
5285 Perl_sv_chop(pTHX_ SV *const sv, const char *const ptr)
5296 PERL_ARGS_ASSERT_SV_CHOP;
5298 if (!ptr || !SvPOKp(sv))
5300 delta = ptr - SvPVX_const(sv);
5302 /* Nothing to do. */
5305 max_delta = SvLEN(sv) ? SvLEN(sv) : SvCUR(sv);
5306 if (delta > max_delta)
5307 Perl_croak(aTHX_ "panic: sv_chop ptr=%p, start=%p, end=%p",
5308 ptr, SvPVX_const(sv), SvPVX_const(sv) + max_delta);
5309 /* SvPVX(sv) may move in SV_CHECK_THINKFIRST(sv), so don't use ptr any more */
5310 SV_CHECK_THINKFIRST(sv);
5311 SvPOK_only_UTF8(sv);
5314 if (!SvLEN(sv)) { /* make copy of shared string */
5315 const char *pvx = SvPVX_const(sv);
5316 const STRLEN len = SvCUR(sv);
5317 SvGROW(sv, len + 1);
5318 Move(pvx,SvPVX(sv),len,char);
5324 SvOOK_offset(sv, old_delta);
5326 SvLEN_set(sv, SvLEN(sv) - delta);
5327 SvCUR_set(sv, SvCUR(sv) - delta);
5328 SvPV_set(sv, SvPVX(sv) + delta);
5330 p = (U8 *)SvPVX_const(sv);
5333 /* how many bytes were evacuated? we will fill them with sentinel
5334 bytes, except for the part holding the new offset of course. */
5337 evacn += (old_delta < 0x100 ? 1 : 1 + sizeof(STRLEN));
5339 assert(evacn <= delta + old_delta);
5343 /* This sets 'delta' to the accumulated value of all deltas so far */
5347 /* If 'delta' fits in a byte, store it just prior to the new beginning of
5348 * the string; otherwise store a 0 byte there and store 'delta' just prior
5349 * to that, using as many bytes as a STRLEN occupies. Thus it overwrites a
5350 * portion of the chopped part of the string */
5351 if (delta < 0x100) {
5355 p -= sizeof(STRLEN);
5356 Copy((U8*)&delta, p, sizeof(STRLEN), U8);
5360 /* Fill the preceding buffer with sentinals to verify that no-one is
5370 =for apidoc sv_catpvn
5372 Concatenates the string onto the end of the string which is in the SV.
5373 C<len> indicates number of bytes to copy. If the SV has the UTF-8
5374 status set, then the bytes appended should be valid UTF-8.
5375 Handles 'get' magic, but not 'set' magic. See C<L</sv_catpvn_mg>>.
5377 =for apidoc sv_catpvn_flags
5379 Concatenates the string onto the end of the string which is in the SV. The
5380 C<len> indicates number of bytes to copy.
5382 By default, the string appended is assumed to be valid UTF-8 if the SV has
5383 the UTF-8 status set, and a string of bytes otherwise. One can force the
5384 appended string to be interpreted as UTF-8 by supplying the C<SV_CATUTF8>
5385 flag, and as bytes by supplying the C<SV_CATBYTES> flag; the SV or the
5386 string appended will be upgraded to UTF-8 if necessary.
5388 If C<flags> has the C<SV_SMAGIC> bit set, will
5389 C<mg_set> on C<dsv> afterwards if appropriate.
5390 C<sv_catpvn> and C<sv_catpvn_nomg> are implemented
5391 in terms of this function.
5397 Perl_sv_catpvn_flags(pTHX_ SV *const dsv, const char *sstr, const STRLEN slen, const I32 flags)
5400 const char * const dstr = SvPV_force_flags(dsv, dlen, flags);
5402 PERL_ARGS_ASSERT_SV_CATPVN_FLAGS;
5403 assert((flags & (SV_CATBYTES|SV_CATUTF8)) != (SV_CATBYTES|SV_CATUTF8));
5405 if (!(flags & SV_CATBYTES) || !SvUTF8(dsv)) {
5406 if (flags & SV_CATUTF8 && !SvUTF8(dsv)) {
5407 sv_utf8_upgrade_flags_grow(dsv, 0, slen + 1);
5410 else SvGROW(dsv, dlen + slen + 1);
5412 sstr = SvPVX_const(dsv);
5413 Move(sstr, SvPVX(dsv) + dlen, slen, char);
5414 SvCUR_set(dsv, SvCUR(dsv) + slen);
5417 /* We inline bytes_to_utf8, to avoid an extra malloc. */
5418 const char * const send = sstr + slen;
5421 /* Something this code does not account for, which I think is
5422 impossible; it would require the same pv to be treated as
5423 bytes *and* utf8, which would indicate a bug elsewhere. */
5424 assert(sstr != dstr);
5426 SvGROW(dsv, dlen + slen * 2 + 1);
5427 d = (U8 *)SvPVX(dsv) + dlen;
5429 while (sstr < send) {
5430 append_utf8_from_native_byte(*sstr, &d);
5433 SvCUR_set(dsv, d-(const U8 *)SvPVX(dsv));
5436 (void)SvPOK_only_UTF8(dsv); /* validate pointer */
5438 if (flags & SV_SMAGIC)
5443 =for apidoc sv_catsv
5445 Concatenates the string from SV C<ssv> onto the end of the string in SV
5446 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5447 Handles 'get' magic on both SVs, but no 'set' magic. See C<L</sv_catsv_mg>>
5448 and C<L</sv_catsv_nomg>>.
5450 =for apidoc sv_catsv_flags
5452 Concatenates the string from SV C<ssv> onto the end of the string in SV
5453 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5454 If C<flags> has the C<SV_GMAGIC> bit set, will call C<mg_get> on both SVs if
5455 appropriate. If C<flags> has the C<SV_SMAGIC> bit set, C<mg_set> will be called on
5456 the modified SV afterward, if appropriate. C<sv_catsv>, C<sv_catsv_nomg>,
5457 and C<sv_catsv_mg> are implemented in terms of this function.
5462 Perl_sv_catsv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
5464 PERL_ARGS_ASSERT_SV_CATSV_FLAGS;
5468 const char *spv = SvPV_flags_const(ssv, slen, flags);
5469 if (flags & SV_GMAGIC)
5471 sv_catpvn_flags(dsv, spv, slen,
5472 DO_UTF8(ssv) ? SV_CATUTF8 : SV_CATBYTES);
5473 if (flags & SV_SMAGIC)
5479 =for apidoc sv_catpv
5481 Concatenates the C<NUL>-terminated string onto the end of the string which is
5483 If the SV has the UTF-8 status set, then the bytes appended should be
5484 valid UTF-8. Handles 'get' magic, but not 'set' magic. See
5490 Perl_sv_catpv(pTHX_ SV *const sv, const char *ptr)
5496 PERL_ARGS_ASSERT_SV_CATPV;
5500 junk = SvPV_force(sv, tlen);
5502 SvGROW(sv, tlen + len + 1);
5504 ptr = SvPVX_const(sv);
5505 Move(ptr,SvPVX(sv)+tlen,len+1,char);
5506 SvCUR_set(sv, SvCUR(sv) + len);
5507 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5512 =for apidoc sv_catpv_flags
5514 Concatenates the C<NUL>-terminated string onto the end of the string which is
5516 If the SV has the UTF-8 status set, then the bytes appended should
5517 be valid UTF-8. If C<flags> has the C<SV_SMAGIC> bit set, will C<mg_set>
5518 on the modified SV if appropriate.
5524 Perl_sv_catpv_flags(pTHX_ SV *dstr, const char *sstr, const I32 flags)
5526 PERL_ARGS_ASSERT_SV_CATPV_FLAGS;
5527 sv_catpvn_flags(dstr, sstr, strlen(sstr), flags);
5531 =for apidoc sv_catpv_mg
5533 Like C<sv_catpv>, but also handles 'set' magic.
5539 Perl_sv_catpv_mg(pTHX_ SV *const sv, const char *const ptr)
5541 PERL_ARGS_ASSERT_SV_CATPV_MG;
5550 Creates a new SV. A non-zero C<len> parameter indicates the number of
5551 bytes of preallocated string space the SV should have. An extra byte for a
5552 trailing C<NUL> is also reserved. (C<SvPOK> is not set for the SV even if string
5553 space is allocated.) The reference count for the new SV is set to 1.
5555 In 5.9.3, C<newSV()> replaces the older C<NEWSV()> API, and drops the first
5556 parameter, I<x>, a debug aid which allowed callers to identify themselves.
5557 This aid has been superseded by a new build option, C<PERL_MEM_LOG> (see
5558 L<perlhacktips/PERL_MEM_LOG>). The older API is still there for use in XS
5559 modules supporting older perls.
5565 Perl_newSV(pTHX_ const STRLEN len)
5571 sv_grow(sv, len + 1);
5576 =for apidoc sv_magicext
5578 Adds magic to an SV, upgrading it if necessary. Applies the
5579 supplied C<vtable> and returns a pointer to the magic added.
5581 Note that C<sv_magicext> will allow things that C<sv_magic> will not.
5582 In particular, you can add magic to C<SvREADONLY> SVs, and add more than
5583 one instance of the same C<how>.
5585 If C<namlen> is greater than zero then a C<savepvn> I<copy> of C<name> is
5586 stored, if C<namlen> is zero then C<name> is stored as-is and - as another
5587 special case - if C<(name && namlen == HEf_SVKEY)> then C<name> is assumed
5588 to contain an SV* and is stored as-is with its C<REFCNT> incremented.
5590 (This is now used as a subroutine by C<sv_magic>.)
5595 Perl_sv_magicext(pTHX_ SV *const sv, SV *const obj, const int how,
5596 const MGVTBL *const vtable, const char *const name, const I32 namlen)
5600 PERL_ARGS_ASSERT_SV_MAGICEXT;
5602 SvUPGRADE(sv, SVt_PVMG);
5603 Newxz(mg, 1, MAGIC);
5604 mg->mg_moremagic = SvMAGIC(sv);
5605 SvMAGIC_set(sv, mg);
5607 /* Sometimes a magic contains a reference loop, where the sv and
5608 object refer to each other. To prevent a reference loop that
5609 would prevent such objects being freed, we look for such loops
5610 and if we find one we avoid incrementing the object refcount.
5612 Note we cannot do this to avoid self-tie loops as intervening RV must
5613 have its REFCNT incremented to keep it in existence.
5616 if (!obj || obj == sv ||
5617 how == PERL_MAGIC_arylen ||
5618 how == PERL_MAGIC_symtab ||
5619 (SvTYPE(obj) == SVt_PVGV &&
5620 (GvSV(obj) == sv || GvHV(obj) == (const HV *)sv
5621 || GvAV(obj) == (const AV *)sv || GvCV(obj) == (const CV *)sv
5622 || GvIOp(obj) == (const IO *)sv || GvFORM(obj) == (const CV *)sv)))
5627 mg->mg_obj = SvREFCNT_inc_simple(obj);
5628 mg->mg_flags |= MGf_REFCOUNTED;
5631 /* Normal self-ties simply pass a null object, and instead of
5632 using mg_obj directly, use the SvTIED_obj macro to produce a
5633 new RV as needed. For glob "self-ties", we are tieing the PVIO
5634 with an RV obj pointing to the glob containing the PVIO. In
5635 this case, to avoid a reference loop, we need to weaken the
5639 if (how == PERL_MAGIC_tiedscalar && SvTYPE(sv) == SVt_PVIO &&
5640 obj && SvROK(obj) && GvIO(SvRV(obj)) == (const IO *)sv)
5646 mg->mg_len = namlen;
5649 mg->mg_ptr = savepvn(name, namlen);
5650 else if (namlen == HEf_SVKEY) {
5651 /* Yes, this is casting away const. This is only for the case of
5652 HEf_SVKEY. I think we need to document this aberation of the
5653 constness of the API, rather than making name non-const, as
5654 that change propagating outwards a long way. */
5655 mg->mg_ptr = (char*)SvREFCNT_inc_simple_NN((SV *)name);
5657 mg->mg_ptr = (char *) name;
5659 mg->mg_virtual = (MGVTBL *) vtable;
5666 Perl_sv_magicext_mglob(pTHX_ SV *sv)
5668 PERL_ARGS_ASSERT_SV_MAGICEXT_MGLOB;
5669 if (SvTYPE(sv) == SVt_PVLV && LvTYPE(sv) == 'y') {
5670 /* This sv is only a delegate. //g magic must be attached to
5675 return sv_magicext(sv, NULL, PERL_MAGIC_regex_global,
5676 &PL_vtbl_mglob, 0, 0);
5680 =for apidoc sv_magic
5682 Adds magic to an SV. First upgrades C<sv> to type C<SVt_PVMG> if
5683 necessary, then adds a new magic item of type C<how> to the head of the
5686 See C<L</sv_magicext>> (which C<sv_magic> now calls) for a description of the
5687 handling of the C<name> and C<namlen> arguments.
5689 You need to use C<sv_magicext> to add magic to C<SvREADONLY> SVs and also
5690 to add more than one instance of the same C<how>.
5696 Perl_sv_magic(pTHX_ SV *const sv, SV *const obj, const int how,
5697 const char *const name, const I32 namlen)
5699 const MGVTBL *vtable;
5702 unsigned int vtable_index;
5704 PERL_ARGS_ASSERT_SV_MAGIC;
5706 if (how < 0 || (unsigned)how >= C_ARRAY_LENGTH(PL_magic_data)
5707 || ((flags = PL_magic_data[how]),
5708 (vtable_index = flags & PERL_MAGIC_VTABLE_MASK)
5709 > magic_vtable_max))
5710 Perl_croak(aTHX_ "Don't know how to handle magic of type \\%o", how);
5712 /* PERL_MAGIC_ext is reserved for use by extensions not perl internals.
5713 Useful for attaching extension internal data to perl vars.
5714 Note that multiple extensions may clash if magical scalars
5715 etc holding private data from one are passed to another. */
5717 vtable = (vtable_index == magic_vtable_max)
5718 ? NULL : PL_magic_vtables + vtable_index;
5720 if (SvREADONLY(sv)) {
5722 !PERL_MAGIC_TYPE_READONLY_ACCEPTABLE(how)
5725 Perl_croak_no_modify();
5728 if (SvMAGICAL(sv) || (how == PERL_MAGIC_taint && SvTYPE(sv) >= SVt_PVMG)) {
5729 if (SvMAGIC(sv) && (mg = mg_find(sv, how))) {
5730 /* sv_magic() refuses to add a magic of the same 'how' as an
5733 if (how == PERL_MAGIC_taint)
5739 /* Force pos to be stored as characters, not bytes. */
5740 if (SvMAGICAL(sv) && DO_UTF8(sv)
5741 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
5743 && mg->mg_flags & MGf_BYTES) {
5744 mg->mg_len = (SSize_t)sv_pos_b2u_flags(sv, (STRLEN)mg->mg_len,
5746 mg->mg_flags &= ~MGf_BYTES;
5749 /* Rest of work is done else where */
5750 mg = sv_magicext(sv,obj,how,vtable,name,namlen);
5753 case PERL_MAGIC_taint:
5756 case PERL_MAGIC_ext:
5757 case PERL_MAGIC_dbfile:
5764 S_sv_unmagicext_flags(pTHX_ SV *const sv, const int type, MGVTBL *vtbl, const U32 flags)
5771 if (SvTYPE(sv) < SVt_PVMG || !SvMAGIC(sv))
5773 mgp = &(((XPVMG*) SvANY(sv))->xmg_u.xmg_magic);
5774 for (mg = *mgp; mg; mg = *mgp) {
5775 const MGVTBL* const virt = mg->mg_virtual;
5776 if (mg->mg_type == type && (!flags || virt == vtbl)) {
5777 *mgp = mg->mg_moremagic;
5778 if (virt && virt->svt_free)
5779 virt->svt_free(aTHX_ sv, mg);
5780 if (mg->mg_ptr && mg->mg_type != PERL_MAGIC_regex_global) {
5782 Safefree(mg->mg_ptr);
5783 else if (mg->mg_len == HEf_SVKEY)
5784 SvREFCNT_dec(MUTABLE_SV(mg->mg_ptr));
5785 else if (mg->mg_type == PERL_MAGIC_utf8)
5786 Safefree(mg->mg_ptr);
5788 if (mg->mg_flags & MGf_REFCOUNTED)
5789 SvREFCNT_dec(mg->mg_obj);
5793 mgp = &mg->mg_moremagic;
5796 if (SvMAGICAL(sv)) /* if we're under save_magic, wait for restore_magic; */
5797 mg_magical(sv); /* else fix the flags now */
5806 =for apidoc sv_unmagic
5808 Removes all magic of type C<type> from an SV.
5814 Perl_sv_unmagic(pTHX_ SV *const sv, const int type)
5816 PERL_ARGS_ASSERT_SV_UNMAGIC;
5817 return S_sv_unmagicext_flags(aTHX_ sv, type, NULL, 0);
5821 =for apidoc sv_unmagicext
5823 Removes all magic of type C<type> with the specified C<vtbl> from an SV.
5829 Perl_sv_unmagicext(pTHX_ SV *const sv, const int type, MGVTBL *vtbl)
5831 PERL_ARGS_ASSERT_SV_UNMAGICEXT;
5832 return S_sv_unmagicext_flags(aTHX_ sv, type, vtbl, 1);
5836 =for apidoc sv_rvweaken
5838 Weaken a reference: set the C<SvWEAKREF> flag on this RV; give the
5839 referred-to SV C<PERL_MAGIC_backref> magic if it hasn't already; and
5840 push a back-reference to this RV onto the array of backreferences
5841 associated with that magic. If the RV is magical, set magic will be
5842 called after the RV is cleared.
5848 Perl_sv_rvweaken(pTHX_ SV *const sv)
5852 PERL_ARGS_ASSERT_SV_RVWEAKEN;
5854 if (!SvOK(sv)) /* let undefs pass */
5857 Perl_croak(aTHX_ "Can't weaken a nonreference");
5858 else if (SvWEAKREF(sv)) {
5859 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is already weak");
5862 else if (SvREADONLY(sv)) croak_no_modify();
5864 Perl_sv_add_backref(aTHX_ tsv, sv);
5866 SvREFCNT_dec_NN(tsv);
5871 =for apidoc sv_get_backrefs
5873 If C<sv> is the target of a weak reference then it returns the back
5874 references structure associated with the sv; otherwise return C<NULL>.
5876 When returning a non-null result the type of the return is relevant. If it
5877 is an AV then the elements of the AV are the weak reference RVs which
5878 point at this item. If it is any other type then the item itself is the
5881 See also C<Perl_sv_add_backref()>, C<Perl_sv_del_backref()>,
5882 C<Perl_sv_kill_backrefs()>
5888 Perl_sv_get_backrefs(SV *const sv)
5892 PERL_ARGS_ASSERT_SV_GET_BACKREFS;
5894 /* find slot to store array or singleton backref */
5896 if (SvTYPE(sv) == SVt_PVHV) {
5898 struct xpvhv_aux * const iter = HvAUX((HV *)sv);
5899 backrefs = (SV *)iter->xhv_backreferences;
5901 } else if (SvMAGICAL(sv)) {
5902 MAGIC *mg = mg_find(sv, PERL_MAGIC_backref);
5904 backrefs = mg->mg_obj;
5909 /* Give tsv backref magic if it hasn't already got it, then push a
5910 * back-reference to sv onto the array associated with the backref magic.
5912 * As an optimisation, if there's only one backref and it's not an AV,
5913 * store it directly in the HvAUX or mg_obj slot, avoiding the need to
5914 * allocate an AV. (Whether the slot holds an AV tells us whether this is
5918 /* A discussion about the backreferences array and its refcount:
5920 * The AV holding the backreferences is pointed to either as the mg_obj of
5921 * PERL_MAGIC_backref, or in the specific case of a HV, from the
5922 * xhv_backreferences field. The array is created with a refcount
5923 * of 2. This means that if during global destruction the array gets
5924 * picked on before its parent to have its refcount decremented by the
5925 * random zapper, it won't actually be freed, meaning it's still there for
5926 * when its parent gets freed.
5928 * When the parent SV is freed, the extra ref is killed by
5929 * Perl_sv_kill_backrefs. The other ref is killed, in the case of magic,
5930 * by mg_free() / MGf_REFCOUNTED, or for a hash, by Perl_hv_kill_backrefs.
5932 * When a single backref SV is stored directly, it is not reference
5937 Perl_sv_add_backref(pTHX_ SV *const tsv, SV *const sv)
5943 PERL_ARGS_ASSERT_SV_ADD_BACKREF;
5945 /* find slot to store array or singleton backref */
5947 if (SvTYPE(tsv) == SVt_PVHV) {
5948 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
5951 mg = mg_find(tsv, PERL_MAGIC_backref);
5953 mg = sv_magicext(tsv, NULL, PERL_MAGIC_backref, &PL_vtbl_backref, NULL, 0);
5954 svp = &(mg->mg_obj);
5957 /* create or retrieve the array */
5959 if ( (!*svp && SvTYPE(sv) == SVt_PVAV)
5960 || (*svp && SvTYPE(*svp) != SVt_PVAV)
5964 mg->mg_flags |= MGf_REFCOUNTED;
5967 SvREFCNT_inc_simple_void_NN(av);
5968 /* av now has a refcnt of 2; see discussion above */
5969 av_extend(av, *svp ? 2 : 1);
5971 /* move single existing backref to the array */
5972 AvARRAY(av)[++AvFILLp(av)] = *svp; /* av_push() */
5977 av = MUTABLE_AV(*svp);
5979 /* optimisation: store single backref directly in HvAUX or mg_obj */
5983 assert(SvTYPE(av) == SVt_PVAV);
5984 if (AvFILLp(av) >= AvMAX(av)) {
5985 av_extend(av, AvFILLp(av)+1);
5988 /* push new backref */
5989 AvARRAY(av)[++AvFILLp(av)] = sv; /* av_push() */
5992 /* delete a back-reference to ourselves from the backref magic associated
5993 * with the SV we point to.
5997 Perl_sv_del_backref(pTHX_ SV *const tsv, SV *const sv)
6001 PERL_ARGS_ASSERT_SV_DEL_BACKREF;
6003 if (SvTYPE(tsv) == SVt_PVHV) {
6005 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
6007 else if (SvIS_FREED(tsv) && PL_phase == PERL_PHASE_DESTRUCT) {
6008 /* It's possible for the the last (strong) reference to tsv to have
6009 become freed *before* the last thing holding a weak reference.
6010 If both survive longer than the backreferences array, then when
6011 the referent's reference count drops to 0 and it is freed, it's
6012 not able to chase the backreferences, so they aren't NULLed.
6014 For example, a CV holds a weak reference to its stash. If both the
6015 CV and the stash survive longer than the backreferences array,
6016 and the CV gets picked for the SvBREAK() treatment first,
6017 *and* it turns out that the stash is only being kept alive because
6018 of an our variable in the pad of the CV, then midway during CV
6019 destruction the stash gets freed, but CvSTASH() isn't set to NULL.
6020 It ends up pointing to the freed HV. Hence it's chased in here, and
6021 if this block wasn't here, it would hit the !svp panic just below.
6023 I don't believe that "better" destruction ordering is going to help
6024 here - during global destruction there's always going to be the
6025 chance that something goes out of order. We've tried to make it
6026 foolproof before, and it only resulted in evolutionary pressure on
6027 fools. Which made us look foolish for our hubris. :-(
6033 = SvMAGICAL(tsv) ? mg_find(tsv, PERL_MAGIC_backref) : NULL;
6034 svp = mg ? &(mg->mg_obj) : NULL;
6038 Perl_croak(aTHX_ "panic: del_backref, svp=0");
6040 /* It's possible that sv is being freed recursively part way through the
6041 freeing of tsv. If this happens, the backreferences array of tsv has
6042 already been freed, and so svp will be NULL. If this is the case,
6043 we should not panic. Instead, nothing needs doing, so return. */
6044 if (PL_phase == PERL_PHASE_DESTRUCT && SvREFCNT(tsv) == 0)
6046 Perl_croak(aTHX_ "panic: del_backref, *svp=%p phase=%s refcnt=%" UVuf,
6047 (void*)*svp, PL_phase_names[PL_phase], (UV)SvREFCNT(tsv));
6050 if (SvTYPE(*svp) == SVt_PVAV) {
6054 AV * const av = (AV*)*svp;
6056 assert(!SvIS_FREED(av));
6060 /* for an SV with N weak references to it, if all those
6061 * weak refs are deleted, then sv_del_backref will be called
6062 * N times and O(N^2) compares will be done within the backref
6063 * array. To ameliorate this potential slowness, we:
6064 * 1) make sure this code is as tight as possible;
6065 * 2) when looking for SV, look for it at both the head and tail of the
6066 * array first before searching the rest, since some create/destroy
6067 * patterns will cause the backrefs to be freed in order.
6074 SV **p = &svp[fill];
6075 SV *const topsv = *p;
6082 /* We weren't the last entry.
6083 An unordered list has this property that you
6084 can take the last element off the end to fill
6085 the hole, and it's still an unordered list :-)
6091 break; /* should only be one */
6098 AvFILLp(av) = fill-1;
6100 else if (SvIS_FREED(*svp) && PL_phase == PERL_PHASE_DESTRUCT) {
6101 /* freed AV; skip */
6104 /* optimisation: only a single backref, stored directly */
6106 Perl_croak(aTHX_ "panic: del_backref, *svp=%p, sv=%p",
6107 (void*)*svp, (void*)sv);
6114 Perl_sv_kill_backrefs(pTHX_ SV *const sv, AV *const av)
6120 PERL_ARGS_ASSERT_SV_KILL_BACKREFS;
6125 /* after multiple passes through Perl_sv_clean_all() for a thingy
6126 * that has badly leaked, the backref array may have gotten freed,
6127 * since we only protect it against 1 round of cleanup */
6128 if (SvIS_FREED(av)) {
6129 if (PL_in_clean_all) /* All is fair */
6132 "panic: magic_killbackrefs (freed backref AV/SV)");
6136 is_array = (SvTYPE(av) == SVt_PVAV);
6138 assert(!SvIS_FREED(av));
6141 last = svp + AvFILLp(av);
6144 /* optimisation: only a single backref, stored directly */
6150 while (svp <= last) {
6152 SV *const referrer = *svp;
6153 if (SvWEAKREF(referrer)) {
6154 /* XXX Should we check that it hasn't changed? */
6155 assert(SvROK(referrer));
6156 SvRV_set(referrer, 0);
6158 SvWEAKREF_off(referrer);
6159 SvSETMAGIC(referrer);
6160 } else if (SvTYPE(referrer) == SVt_PVGV ||
6161 SvTYPE(referrer) == SVt_PVLV) {
6162 assert(SvTYPE(sv) == SVt_PVHV); /* stash backref */
6163 /* You lookin' at me? */
6164 assert(GvSTASH(referrer));
6165 assert(GvSTASH(referrer) == (const HV *)sv);
6166 GvSTASH(referrer) = 0;
6167 } else if (SvTYPE(referrer) == SVt_PVCV ||
6168 SvTYPE(referrer) == SVt_PVFM) {
6169 if (SvTYPE(sv) == SVt_PVHV) { /* stash backref */
6170 /* You lookin' at me? */
6171 assert(CvSTASH(referrer));
6172 assert(CvSTASH(referrer) == (const HV *)sv);
6173 SvANY(MUTABLE_CV(referrer))->xcv_stash = 0;
6176 assert(SvTYPE(sv) == SVt_PVGV);
6177 /* You lookin' at me? */
6178 assert(CvGV(referrer));
6179 assert(CvGV(referrer) == (const GV *)sv);
6180 anonymise_cv_maybe(MUTABLE_GV(sv),
6181 MUTABLE_CV(referrer));
6186 "panic: magic_killbackrefs (flags=%"UVxf")",
6187 (UV)SvFLAGS(referrer));
6198 SvREFCNT_dec_NN(av); /* remove extra count added by sv_add_backref() */
6204 =for apidoc sv_insert
6206 Inserts a string at the specified offset/length within the SV. Similar to
6207 the Perl C<substr()> function. Handles get magic.
6209 =for apidoc sv_insert_flags
6211 Same as C<sv_insert>, but the extra C<flags> are passed to the
6212 C<SvPV_force_flags> that applies to C<bigstr>.
6218 Perl_sv_insert_flags(pTHX_ SV *const bigstr, const STRLEN offset, const STRLEN len, const char *const little, const STRLEN littlelen, const U32 flags)
6224 SSize_t i; /* better be sizeof(STRLEN) or bad things happen */
6227 PERL_ARGS_ASSERT_SV_INSERT_FLAGS;
6229 SvPV_force_flags(bigstr, curlen, flags);
6230 (void)SvPOK_only_UTF8(bigstr);
6231 if (offset + len > curlen) {
6232 SvGROW(bigstr, offset+len+1);
6233 Zero(SvPVX(bigstr)+curlen, offset+len-curlen, char);
6234 SvCUR_set(bigstr, offset+len);
6238 i = littlelen - len;
6239 if (i > 0) { /* string might grow */
6240 big = SvGROW(bigstr, SvCUR(bigstr) + i + 1);
6241 mid = big + offset + len;
6242 midend = bigend = big + SvCUR(bigstr);
6245 while (midend > mid) /* shove everything down */
6246 *--bigend = *--midend;
6247 Move(little,big+offset,littlelen,char);
6248 SvCUR_set(bigstr, SvCUR(bigstr) + i);
6253 Move(little,SvPVX(bigstr)+offset,len,char);
6258 big = SvPVX(bigstr);
6261 bigend = big + SvCUR(bigstr);
6263 if (midend > bigend)
6264 Perl_croak(aTHX_ "panic: sv_insert, midend=%p, bigend=%p",
6267 if (mid - big > bigend - midend) { /* faster to shorten from end */
6269 Move(little, mid, littlelen,char);
6272 i = bigend - midend;
6274 Move(midend, mid, i,char);
6278 SvCUR_set(bigstr, mid - big);
6280 else if ((i = mid - big)) { /* faster from front */
6281 midend -= littlelen;
6283 Move(big, midend - i, i, char);
6284 sv_chop(bigstr,midend-i);
6286 Move(little, mid, littlelen,char);
6288 else if (littlelen) {
6289 midend -= littlelen;
6290 sv_chop(bigstr,midend);
6291 Move(little,midend,littlelen,char);
6294 sv_chop(bigstr,midend);
6300 =for apidoc sv_replace
6302 Make the first argument a copy of the second, then delete the original.
6303 The target SV physically takes over ownership of the body of the source SV
6304 and inherits its flags; however, the target keeps any magic it owns,
6305 and any magic in the source is discarded.
6306 Note that this is a rather specialist SV copying operation; most of the
6307 time you'll want to use C<sv_setsv> or one of its many macro front-ends.
6313 Perl_sv_replace(pTHX_ SV *const sv, SV *const nsv)
6315 const U32 refcnt = SvREFCNT(sv);
6317 PERL_ARGS_ASSERT_SV_REPLACE;
6319 SV_CHECK_THINKFIRST_COW_DROP(sv);
6320 if (SvREFCNT(nsv) != 1) {
6321 Perl_croak(aTHX_ "panic: reference miscount on nsv in sv_replace()"
6322 " (%" UVuf " != 1)", (UV) SvREFCNT(nsv));
6324 if (SvMAGICAL(sv)) {
6328 sv_upgrade(nsv, SVt_PVMG);
6329 SvMAGIC_set(nsv, SvMAGIC(sv));
6330 SvFLAGS(nsv) |= SvMAGICAL(sv);
6332 SvMAGIC_set(sv, NULL);
6336 assert(!SvREFCNT(sv));
6337 #ifdef DEBUG_LEAKING_SCALARS
6338 sv->sv_flags = nsv->sv_flags;
6339 sv->sv_any = nsv->sv_any;
6340 sv->sv_refcnt = nsv->sv_refcnt;
6341 sv->sv_u = nsv->sv_u;
6343 StructCopy(nsv,sv,SV);
6345 if(SvTYPE(sv) == SVt_IV) {
6346 SET_SVANY_FOR_BODYLESS_IV(sv);
6350 SvREFCNT(sv) = refcnt;
6351 SvFLAGS(nsv) |= SVTYPEMASK; /* Mark as freed */
6356 /* We're about to free a GV which has a CV that refers back to us.
6357 * If that CV will outlive us, make it anonymous (i.e. fix up its CvGV
6361 S_anonymise_cv_maybe(pTHX_ GV *gv, CV* cv)
6366 PERL_ARGS_ASSERT_ANONYMISE_CV_MAYBE;
6369 assert(SvREFCNT(gv) == 0);
6370 assert(isGV(gv) && isGV_with_GP(gv));
6372 assert(!CvANON(cv));
6373 assert(CvGV(cv) == gv);
6374 assert(!CvNAMED(cv));
6376 /* will the CV shortly be freed by gp_free() ? */
6377 if (GvCV(gv) == cv && GvGP(gv)->gp_refcnt < 2 && SvREFCNT(cv) < 2) {
6378 SvANY(cv)->xcv_gv_u.xcv_gv = NULL;
6382 /* if not, anonymise: */
6383 gvname = (GvSTASH(gv) && HvNAME(GvSTASH(gv)) && HvENAME(GvSTASH(gv)))
6384 ? newSVhek(HvENAME_HEK(GvSTASH(gv)))
6385 : newSVpvn_flags( "__ANON__", 8, 0 );
6386 sv_catpvs(gvname, "::__ANON__");
6387 anongv = gv_fetchsv(gvname, GV_ADDMULTI, SVt_PVCV);
6388 SvREFCNT_dec_NN(gvname);
6392 SvANY(cv)->xcv_gv_u.xcv_gv = MUTABLE_GV(SvREFCNT_inc(anongv));
6397 =for apidoc sv_clear
6399 Clear an SV: call any destructors, free up any memory used by the body,
6400 and free the body itself. The SV's head is I<not> freed, although
6401 its type is set to all 1's so that it won't inadvertently be assumed
6402 to be live during global destruction etc.
6403 This function should only be called when C<REFCNT> is zero. Most of the time
6404 you'll want to call C<sv_free()> (or its macro wrapper C<SvREFCNT_dec>)
6411 Perl_sv_clear(pTHX_ SV *const orig_sv)
6416 const struct body_details *sv_type_details;
6420 STRLEN hash_index = 0; /* initialise to make Coverity et al happy.
6421 Not strictly necessary */
6423 PERL_ARGS_ASSERT_SV_CLEAR;
6425 /* within this loop, sv is the SV currently being freed, and
6426 * iter_sv is the most recent AV or whatever that's being iterated
6427 * over to provide more SVs */
6433 assert(SvREFCNT(sv) == 0);
6434 assert(SvTYPE(sv) != (svtype)SVTYPEMASK);
6436 if (type <= SVt_IV) {
6437 /* See the comment in sv.h about the collusion between this
6438 * early return and the overloading of the NULL slots in the
6442 SvFLAGS(sv) &= SVf_BREAK;
6443 SvFLAGS(sv) |= SVTYPEMASK;
6447 /* objs are always >= MG, but pad names use the SVs_OBJECT flag
6448 for another purpose */
6449 assert(!SvOBJECT(sv) || type >= SVt_PVMG);
6451 if (type >= SVt_PVMG) {
6453 if (!curse(sv, 1)) goto get_next_sv;
6454 type = SvTYPE(sv); /* destructor may have changed it */
6456 /* Free back-references before magic, in case the magic calls
6457 * Perl code that has weak references to sv. */
6458 if (type == SVt_PVHV) {
6459 Perl_hv_kill_backrefs(aTHX_ MUTABLE_HV(sv));
6463 else if (SvMAGIC(sv)) {
6464 /* Free back-references before other types of magic. */
6465 sv_unmagic(sv, PERL_MAGIC_backref);
6471 /* case SVt_INVLIST: */
6474 IoIFP(sv) != PerlIO_stdin() &&
6475 IoIFP(sv) != PerlIO_stdout() &&
6476 IoIFP(sv) != PerlIO_stderr() &&
6477 !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6479 io_close(MUTABLE_IO(sv), NULL, FALSE,
6480 (IoTYPE(sv) == IoTYPE_WRONLY ||
6481 IoTYPE(sv) == IoTYPE_RDWR ||
6482 IoTYPE(sv) == IoTYPE_APPEND));
6484 if (IoDIRP(sv) && !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6485 PerlDir_close(IoDIRP(sv));
6486 IoDIRP(sv) = (DIR*)NULL;
6487 Safefree(IoTOP_NAME(sv));
6488 Safefree(IoFMT_NAME(sv));
6489 Safefree(IoBOTTOM_NAME(sv));
6490 if ((const GV *)sv == PL_statgv)
6494 /* FIXME for plugins */
6496 pregfree2((REGEXP*) sv);
6500 cv_undef(MUTABLE_CV(sv));
6501 /* If we're in a stash, we don't own a reference to it.
6502 * However it does have a back reference to us, which needs to
6504 if ((stash = CvSTASH(sv)))
6505 sv_del_backref(MUTABLE_SV(stash), sv);
6508 if (PL_last_swash_hv == (const HV *)sv) {
6509 PL_last_swash_hv = NULL;
6511 if (HvTOTALKEYS((HV*)sv) > 0) {
6513 /* this statement should match the one at the beginning of
6514 * hv_undef_flags() */
6515 if ( PL_phase != PERL_PHASE_DESTRUCT
6516 && (hek = HvNAME_HEK((HV*)sv)))
6518 if (PL_stashcache) {
6519 DEBUG_o(Perl_deb(aTHX_
6520 "sv_clear clearing PL_stashcache for '%"HEKf
6523 (void)hv_deletehek(PL_stashcache,
6526 hv_name_set((HV*)sv, NULL, 0, 0);
6529 /* save old iter_sv in unused SvSTASH field */
6530 assert(!SvOBJECT(sv));
6531 SvSTASH(sv) = (HV*)iter_sv;
6534 /* save old hash_index in unused SvMAGIC field */
6535 assert(!SvMAGICAL(sv));
6536 assert(!SvMAGIC(sv));
6537 ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index = hash_index;
6540 next_sv = Perl_hfree_next_entry(aTHX_ (HV*)sv, &hash_index);
6541 goto get_next_sv; /* process this new sv */
6543 /* free empty hash */
6544 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6545 assert(!HvARRAY((HV*)sv));
6549 AV* av = MUTABLE_AV(sv);
6550 if (PL_comppad == av) {
6554 if (AvREAL(av) && AvFILLp(av) > -1) {
6555 next_sv = AvARRAY(av)[AvFILLp(av)--];
6556 /* save old iter_sv in top-most slot of AV,
6557 * and pray that it doesn't get wiped in the meantime */
6558 AvARRAY(av)[AvMAX(av)] = iter_sv;
6560 goto get_next_sv; /* process this new sv */
6562 Safefree(AvALLOC(av));
6567 if (LvTYPE(sv) == 'T') { /* for tie: return HE to pool */
6568 SvREFCNT_dec(HeKEY_sv((HE*)LvTARG(sv)));
6569 HeNEXT((HE*)LvTARG(sv)) = PL_hv_fetch_ent_mh;
6570 PL_hv_fetch_ent_mh = (HE*)LvTARG(sv);
6572 else if (LvTYPE(sv) != 't') /* unless tie: unrefcnted fake SV** */
6573 SvREFCNT_dec(LvTARG(sv));
6574 if (isREGEXP(sv)) goto freeregexp;
6577 if (isGV_with_GP(sv)) {
6578 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
6579 && HvENAME_get(stash))
6580 mro_method_changed_in(stash);
6581 gp_free(MUTABLE_GV(sv));
6583 unshare_hek(GvNAME_HEK(sv));
6584 /* If we're in a stash, we don't own a reference to it.
6585 * However it does have a back reference to us, which
6586 * needs to be cleared. */
6587 if (!SvVALID(sv) && (stash = GvSTASH(sv)))
6588 sv_del_backref(MUTABLE_SV(stash), sv);
6590 /* FIXME. There are probably more unreferenced pointers to SVs
6591 * in the interpreter struct that we should check and tidy in
6592 * a similar fashion to this: */
6593 /* See also S_sv_unglob, which does the same thing. */
6594 if ((const GV *)sv == PL_last_in_gv)
6595 PL_last_in_gv = NULL;
6596 else if ((const GV *)sv == PL_statgv)
6598 else if ((const GV *)sv == PL_stderrgv)
6607 /* Don't bother with SvOOK_off(sv); as we're only going to
6611 SvOOK_offset(sv, offset);
6612 SvPV_set(sv, SvPVX_mutable(sv) - offset);
6613 /* Don't even bother with turning off the OOK flag. */
6618 SV * const target = SvRV(sv);
6620 sv_del_backref(target, sv);
6626 else if (SvPVX_const(sv)
6627 && !(SvTYPE(sv) == SVt_PVIO
6628 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6632 PerlIO_printf(Perl_debug_log, "Copy on write: clear\n");
6636 if (CowREFCNT(sv)) {
6643 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6648 Safefree(SvPVX_mutable(sv));
6652 else if (SvPVX_const(sv) && SvLEN(sv)
6653 && !(SvTYPE(sv) == SVt_PVIO
6654 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6655 Safefree(SvPVX_mutable(sv));
6656 else if (SvPVX_const(sv) && SvIsCOW(sv)) {
6657 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6667 SvFLAGS(sv) &= SVf_BREAK;
6668 SvFLAGS(sv) |= SVTYPEMASK;
6670 sv_type_details = bodies_by_type + type;
6671 if (sv_type_details->arena) {
6672 del_body(((char *)SvANY(sv) + sv_type_details->offset),
6673 &PL_body_roots[type]);
6675 else if (sv_type_details->body_size) {
6676 safefree(SvANY(sv));
6680 /* caller is responsible for freeing the head of the original sv */
6681 if (sv != orig_sv && !SvREFCNT(sv))
6684 /* grab and free next sv, if any */
6692 else if (!iter_sv) {
6694 } else if (SvTYPE(iter_sv) == SVt_PVAV) {
6695 AV *const av = (AV*)iter_sv;
6696 if (AvFILLp(av) > -1) {
6697 sv = AvARRAY(av)[AvFILLp(av)--];
6699 else { /* no more elements of current AV to free */
6702 /* restore previous value, squirrelled away */
6703 iter_sv = AvARRAY(av)[AvMAX(av)];
6704 Safefree(AvALLOC(av));
6707 } else if (SvTYPE(iter_sv) == SVt_PVHV) {
6708 sv = Perl_hfree_next_entry(aTHX_ (HV*)iter_sv, &hash_index);
6709 if (!sv && !HvTOTALKEYS((HV *)iter_sv)) {
6710 /* no more elements of current HV to free */
6713 /* Restore previous values of iter_sv and hash_index,
6714 * squirrelled away */
6715 assert(!SvOBJECT(sv));
6716 iter_sv = (SV*)SvSTASH(sv);
6717 assert(!SvMAGICAL(sv));
6718 hash_index = ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index;
6720 /* perl -DA does not like rubbish in SvMAGIC. */
6724 /* free any remaining detritus from the hash struct */
6725 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6726 assert(!HvARRAY((HV*)sv));
6731 /* unrolled SvREFCNT_dec and sv_free2 follows: */
6735 if (!SvREFCNT(sv)) {
6739 if (--(SvREFCNT(sv)))
6743 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6744 "Attempt to free temp prematurely: SV 0x%"UVxf
6745 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6749 if (SvIMMORTAL(sv)) {
6750 /* make sure SvREFCNT(sv)==0 happens very seldom */
6751 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6760 /* This routine curses the sv itself, not the object referenced by sv. So
6761 sv does not have to be ROK. */
6764 S_curse(pTHX_ SV * const sv, const bool check_refcnt) {
6765 PERL_ARGS_ASSERT_CURSE;
6766 assert(SvOBJECT(sv));
6768 if (PL_defstash && /* Still have a symbol table? */
6774 stash = SvSTASH(sv);
6775 assert(SvTYPE(stash) == SVt_PVHV);
6776 if (HvNAME(stash)) {
6777 CV* destructor = NULL;
6778 struct mro_meta *meta;
6780 assert (SvOOK(stash));
6782 DEBUG_o( Perl_deb(aTHX_ "Looking for DESTROY method for %s\n",
6785 /* don't make this an initialization above the assert, since it needs
6787 meta = HvMROMETA(stash);
6788 if (meta->destroy_gen && meta->destroy_gen == PL_sub_generation) {
6789 destructor = meta->destroy;
6790 DEBUG_o( Perl_deb(aTHX_ "Using cached DESTROY method %p for %s\n",
6791 (void *)destructor, HvNAME(stash)) );
6794 bool autoload = FALSE;
6796 gv_fetchmeth_pvn(stash, S_destroy, S_destroy_len, -1, 0);
6798 destructor = GvCV(gv);
6800 gv = gv_autoload_pvn(stash, S_destroy, S_destroy_len,
6801 GV_AUTOLOAD_ISMETHOD);
6803 destructor = GvCV(gv);
6807 /* we don't cache AUTOLOAD for DESTROY, since this code
6808 would then need to set $__PACKAGE__::AUTOLOAD, or the
6809 equivalent for XS AUTOLOADs */
6811 meta->destroy_gen = PL_sub_generation;
6812 meta->destroy = destructor;
6814 DEBUG_o( Perl_deb(aTHX_ "Set cached DESTROY method %p for %s\n",
6815 (void *)destructor, HvNAME(stash)) );
6818 DEBUG_o( Perl_deb(aTHX_ "Not caching AUTOLOAD for DESTROY method for %s\n",
6822 assert(!destructor || SvTYPE(destructor) == SVt_PVCV);
6824 /* A constant subroutine can have no side effects, so
6825 don't bother calling it. */
6826 && !CvCONST(destructor)
6827 /* Don't bother calling an empty destructor or one that
6828 returns immediately. */
6829 && (CvISXSUB(destructor)
6830 || (CvSTART(destructor)
6831 && (CvSTART(destructor)->op_next->op_type
6833 && (CvSTART(destructor)->op_next->op_type
6835 || CvSTART(destructor)->op_next->op_next->op_type
6841 SV* const tmpref = newRV(sv);
6842 SvREADONLY_on(tmpref); /* DESTROY() could be naughty */
6844 PUSHSTACKi(PERLSI_DESTROY);
6849 call_sv(MUTABLE_SV(destructor),
6850 G_DISCARD|G_EVAL|G_KEEPERR|G_VOID);
6854 if(SvREFCNT(tmpref) < 2) {
6855 /* tmpref is not kept alive! */
6857 SvRV_set(tmpref, NULL);
6860 SvREFCNT_dec_NN(tmpref);
6863 } while (SvOBJECT(sv) && SvSTASH(sv) != stash);
6866 if (check_refcnt && SvREFCNT(sv)) {
6867 if (PL_in_clean_objs)
6869 "DESTROY created new reference to dead object '%"HEKf"'",
6870 HEKfARG(HvNAME_HEK(stash)));
6871 /* DESTROY gave object new lease on life */
6877 HV * const stash = SvSTASH(sv);
6878 /* Curse before freeing the stash, as freeing the stash could cause
6879 a recursive call into S_curse. */
6880 SvOBJECT_off(sv); /* Curse the object. */
6881 SvSTASH_set(sv,0); /* SvREFCNT_dec may try to read this */
6882 SvREFCNT_dec(stash); /* possibly of changed persuasion */
6888 =for apidoc sv_newref
6890 Increment an SV's reference count. Use the C<SvREFCNT_inc()> wrapper
6897 Perl_sv_newref(pTHX_ SV *const sv)
6899 PERL_UNUSED_CONTEXT;
6908 Decrement an SV's reference count, and if it drops to zero, call
6909 C<sv_clear> to invoke destructors and free up any memory used by
6910 the body; finally, deallocating the SV's head itself.
6911 Normally called via a wrapper macro C<SvREFCNT_dec>.
6917 Perl_sv_free(pTHX_ SV *const sv)
6923 /* Private helper function for SvREFCNT_dec().
6924 * Called with rc set to original SvREFCNT(sv), where rc == 0 or 1 */
6927 Perl_sv_free2(pTHX_ SV *const sv, const U32 rc)
6931 PERL_ARGS_ASSERT_SV_FREE2;
6933 if (LIKELY( rc == 1 )) {
6939 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6940 "Attempt to free temp prematurely: SV 0x%"UVxf
6941 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6945 if (SvIMMORTAL(sv)) {
6946 /* make sure SvREFCNT(sv)==0 happens very seldom */
6947 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6951 if (! SvREFCNT(sv)) /* may have have been resurrected */
6956 /* handle exceptional cases */
6960 if (SvFLAGS(sv) & SVf_BREAK)
6961 /* this SV's refcnt has been artificially decremented to
6962 * trigger cleanup */
6964 if (PL_in_clean_all) /* All is fair */
6966 if (SvIMMORTAL(sv)) {
6967 /* make sure SvREFCNT(sv)==0 happens very seldom */
6968 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6971 if (ckWARN_d(WARN_INTERNAL)) {
6972 #ifdef DEBUG_LEAKING_SCALARS_FORK_DUMP
6973 Perl_dump_sv_child(aTHX_ sv);
6975 #ifdef DEBUG_LEAKING_SCALARS
6978 #ifdef DEBUG_LEAKING_SCALARS_ABORT
6979 if (PL_warnhook == PERL_WARNHOOK_FATAL
6980 || ckDEAD(packWARN(WARN_INTERNAL))) {
6981 /* Don't let Perl_warner cause us to escape our fate: */
6985 /* This may not return: */
6986 Perl_warner(aTHX_ packWARN(WARN_INTERNAL),
6987 "Attempt to free unreferenced scalar: SV 0x%"UVxf
6988 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6991 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7001 Returns the length of the string in the SV. Handles magic and type
7002 coercion and sets the UTF8 flag appropriately. See also C<L</SvCUR>>, which
7003 gives raw access to the C<xpv_cur> slot.
7009 Perl_sv_len(pTHX_ SV *const sv)
7016 (void)SvPV_const(sv, len);
7021 =for apidoc sv_len_utf8
7023 Returns the number of characters in the string in an SV, counting wide
7024 UTF-8 bytes as a single character. Handles magic and type coercion.
7030 * The length is cached in PERL_MAGIC_utf8, in the mg_len field. Also the
7031 * mg_ptr is used, by sv_pos_u2b() and sv_pos_b2u() - see the comments below.
7032 * (Note that the mg_len is not the length of the mg_ptr field.
7033 * This allows the cache to store the character length of the string without
7034 * needing to malloc() extra storage to attach to the mg_ptr.)
7039 Perl_sv_len_utf8(pTHX_ SV *const sv)
7045 return sv_len_utf8_nomg(sv);
7049 Perl_sv_len_utf8_nomg(pTHX_ SV * const sv)
7052 const U8 *s = (U8*)SvPV_nomg_const(sv, len);
7054 PERL_ARGS_ASSERT_SV_LEN_UTF8_NOMG;
7056 if (PL_utf8cache && SvUTF8(sv)) {
7058 MAGIC *mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_utf8) : NULL;
7060 if (mg && (mg->mg_len != -1 || mg->mg_ptr)) {
7061 if (mg->mg_len != -1)
7064 /* We can use the offset cache for a headstart.
7065 The longer value is stored in the first pair. */
7066 STRLEN *cache = (STRLEN *) mg->mg_ptr;
7068 ulen = cache[0] + Perl_utf8_length(aTHX_ s + cache[1],
7072 if (PL_utf8cache < 0) {
7073 const STRLEN real = Perl_utf8_length(aTHX_ s, s + len);
7074 assert_uft8_cache_coherent("sv_len_utf8", ulen, real, sv);
7078 ulen = Perl_utf8_length(aTHX_ s, s + len);
7079 utf8_mg_len_cache_update(sv, &mg, ulen);
7083 return SvUTF8(sv) ? Perl_utf8_length(aTHX_ s, s + len) : len;
7086 /* Walk forwards to find the byte corresponding to the passed in UTF-8
7089 S_sv_pos_u2b_forwards(const U8 *const start, const U8 *const send,
7090 STRLEN *const uoffset_p, bool *const at_end)
7092 const U8 *s = start;
7093 STRLEN uoffset = *uoffset_p;
7095 PERL_ARGS_ASSERT_SV_POS_U2B_FORWARDS;
7097 while (s < send && uoffset) {
7104 else if (s > send) {
7106 /* This is the existing behaviour. Possibly it should be a croak, as
7107 it's actually a bounds error */
7110 *uoffset_p -= uoffset;
7114 /* Given the length of the string in both bytes and UTF-8 characters, decide
7115 whether to walk forwards or backwards to find the byte corresponding to
7116 the passed in UTF-8 offset. */
7118 S_sv_pos_u2b_midway(const U8 *const start, const U8 *send,
7119 STRLEN uoffset, const STRLEN uend)
7121 STRLEN backw = uend - uoffset;
7123 PERL_ARGS_ASSERT_SV_POS_U2B_MIDWAY;
7125 if (uoffset < 2 * backw) {
7126 /* The assumption is that going forwards is twice the speed of going
7127 forward (that's where the 2 * backw comes from).
7128 (The real figure of course depends on the UTF-8 data.) */
7129 const U8 *s = start;
7131 while (s < send && uoffset--)
7141 while (UTF8_IS_CONTINUATION(*send))
7144 return send - start;
7147 /* For the string representation of the given scalar, find the byte
7148 corresponding to the passed in UTF-8 offset. uoffset0 and boffset0
7149 give another position in the string, *before* the sought offset, which
7150 (which is always true, as 0, 0 is a valid pair of positions), which should
7151 help reduce the amount of linear searching.
7152 If *mgp is non-NULL, it should point to the UTF-8 cache magic, which
7153 will be used to reduce the amount of linear searching. The cache will be
7154 created if necessary, and the found value offered to it for update. */
7156 S_sv_pos_u2b_cached(pTHX_ SV *const sv, MAGIC **const mgp, const U8 *const start,
7157 const U8 *const send, STRLEN uoffset,
7158 STRLEN uoffset0, STRLEN boffset0)
7160 STRLEN boffset = 0; /* Actually always set, but let's keep gcc happy. */
7162 bool at_end = FALSE;
7164 PERL_ARGS_ASSERT_SV_POS_U2B_CACHED;
7166 assert (uoffset >= uoffset0);
7171 if (!SvREADONLY(sv) && !SvGMAGICAL(sv) && SvPOK(sv)
7173 && (*mgp || (SvTYPE(sv) >= SVt_PVMG &&
7174 (*mgp = mg_find(sv, PERL_MAGIC_utf8))))) {
7175 if ((*mgp)->mg_ptr) {
7176 STRLEN *cache = (STRLEN *) (*mgp)->mg_ptr;
7177 if (cache[0] == uoffset) {
7178 /* An exact match. */
7181 if (cache[2] == uoffset) {
7182 /* An exact match. */
7186 if (cache[0] < uoffset) {
7187 /* The cache already knows part of the way. */
7188 if (cache[0] > uoffset0) {
7189 /* The cache knows more than the passed in pair */
7190 uoffset0 = cache[0];
7191 boffset0 = cache[1];
7193 if ((*mgp)->mg_len != -1) {
7194 /* And we know the end too. */
7196 + sv_pos_u2b_midway(start + boffset0, send,
7198 (*mgp)->mg_len - uoffset0);
7200 uoffset -= uoffset0;
7202 + sv_pos_u2b_forwards(start + boffset0,
7203 send, &uoffset, &at_end);
7204 uoffset += uoffset0;
7207 else if (cache[2] < uoffset) {
7208 /* We're between the two cache entries. */
7209 if (cache[2] > uoffset0) {
7210 /* and the cache knows more than the passed in pair */
7211 uoffset0 = cache[2];
7212 boffset0 = cache[3];
7216 + sv_pos_u2b_midway(start + boffset0,
7219 cache[0] - uoffset0);
7222 + sv_pos_u2b_midway(start + boffset0,
7225 cache[2] - uoffset0);
7229 else if ((*mgp)->mg_len != -1) {
7230 /* If we can take advantage of a passed in offset, do so. */
7231 /* In fact, offset0 is either 0, or less than offset, so don't
7232 need to worry about the other possibility. */
7234 + sv_pos_u2b_midway(start + boffset0, send,
7236 (*mgp)->mg_len - uoffset0);
7241 if (!found || PL_utf8cache < 0) {
7242 STRLEN real_boffset;
7243 uoffset -= uoffset0;
7244 real_boffset = boffset0 + sv_pos_u2b_forwards(start + boffset0,
7245 send, &uoffset, &at_end);
7246 uoffset += uoffset0;
7248 if (found && PL_utf8cache < 0)
7249 assert_uft8_cache_coherent("sv_pos_u2b_cache", boffset,
7251 boffset = real_boffset;
7254 if (PL_utf8cache && !SvGMAGICAL(sv) && SvPOK(sv)) {
7256 utf8_mg_len_cache_update(sv, mgp, uoffset);
7258 utf8_mg_pos_cache_update(sv, mgp, boffset, uoffset, send - start);
7265 =for apidoc sv_pos_u2b_flags
7267 Converts the offset from a count of UTF-8 chars from
7268 the start of the string, to a count of the equivalent number of bytes; if
7269 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7270 C<offset>, rather than from the start
7271 of the string. Handles type coercion.
7272 C<flags> is passed to C<SvPV_flags>, and usually should be
7273 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7279 * sv_pos_u2b_flags() uses, like sv_pos_b2u(), the mg_ptr of the potential
7280 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7281 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7286 Perl_sv_pos_u2b_flags(pTHX_ SV *const sv, STRLEN uoffset, STRLEN *const lenp,
7293 PERL_ARGS_ASSERT_SV_POS_U2B_FLAGS;
7295 start = (U8*)SvPV_flags(sv, len, flags);
7297 const U8 * const send = start + len;
7299 boffset = sv_pos_u2b_cached(sv, &mg, start, send, uoffset, 0, 0);
7302 && *lenp /* don't bother doing work for 0, as its bytes equivalent
7303 is 0, and *lenp is already set to that. */) {
7304 /* Convert the relative offset to absolute. */
7305 const STRLEN uoffset2 = uoffset + *lenp;
7306 const STRLEN boffset2
7307 = sv_pos_u2b_cached(sv, &mg, start, send, uoffset2,
7308 uoffset, boffset) - boffset;
7322 =for apidoc sv_pos_u2b
7324 Converts the value pointed to by C<offsetp> from a count of UTF-8 chars from
7325 the start of the string, to a count of the equivalent number of bytes; if
7326 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7327 the offset, rather than from the start of the string. Handles magic and
7330 Use C<sv_pos_u2b_flags> in preference, which correctly handles strings longer
7337 * sv_pos_u2b() uses, like sv_pos_b2u(), the mg_ptr of the potential
7338 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7339 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7343 /* This function is subject to size and sign problems */
7346 Perl_sv_pos_u2b(pTHX_ SV *const sv, I32 *const offsetp, I32 *const lenp)
7348 PERL_ARGS_ASSERT_SV_POS_U2B;
7351 STRLEN ulen = (STRLEN)*lenp;
7352 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, &ulen,
7353 SV_GMAGIC|SV_CONST_RETURN);
7356 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, NULL,
7357 SV_GMAGIC|SV_CONST_RETURN);
7362 S_utf8_mg_len_cache_update(pTHX_ SV *const sv, MAGIC **const mgp,
7365 PERL_ARGS_ASSERT_UTF8_MG_LEN_CACHE_UPDATE;
7366 if (SvREADONLY(sv) || SvGMAGICAL(sv) || !SvPOK(sv))
7369 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7370 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7371 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, &PL_vtbl_utf8, 0, 0);
7375 (*mgp)->mg_len = ulen;
7378 /* Create and update the UTF8 magic offset cache, with the proffered utf8/
7379 byte length pairing. The (byte) length of the total SV is passed in too,
7380 as blen, because for some (more esoteric) SVs, the call to SvPV_const()
7381 may not have updated SvCUR, so we can't rely on reading it directly.
7383 The proffered utf8/byte length pairing isn't used if the cache already has
7384 two pairs, and swapping either for the proffered pair would increase the
7385 RMS of the intervals between known byte offsets.
7387 The cache itself consists of 4 STRLEN values
7388 0: larger UTF-8 offset
7389 1: corresponding byte offset
7390 2: smaller UTF-8 offset
7391 3: corresponding byte offset
7393 Unused cache pairs have the value 0, 0.
7394 Keeping the cache "backwards" means that the invariant of
7395 cache[0] >= cache[2] is maintained even with empty slots, which means that
7396 the code that uses it doesn't need to worry if only 1 entry has actually
7397 been set to non-zero. It also makes the "position beyond the end of the
7398 cache" logic much simpler, as the first slot is always the one to start
7402 S_utf8_mg_pos_cache_update(pTHX_ SV *const sv, MAGIC **const mgp, const STRLEN byte,
7403 const STRLEN utf8, const STRLEN blen)
7407 PERL_ARGS_ASSERT_UTF8_MG_POS_CACHE_UPDATE;
7412 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7413 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7414 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, (MGVTBL*)&PL_vtbl_utf8, 0,
7416 (*mgp)->mg_len = -1;
7420 if (!(cache = (STRLEN *)(*mgp)->mg_ptr)) {
7421 Newxz(cache, PERL_MAGIC_UTF8_CACHESIZE * 2, STRLEN);
7422 (*mgp)->mg_ptr = (char *) cache;
7426 if (PL_utf8cache < 0 && SvPOKp(sv)) {
7427 /* SvPOKp() because, if sv is a reference, then SvPVX() is actually
7428 a pointer. Note that we no longer cache utf8 offsets on refer-
7429 ences, but this check is still a good idea, for robustness. */
7430 const U8 *start = (const U8 *) SvPVX_const(sv);
7431 const STRLEN realutf8 = utf8_length(start, start + byte);
7433 assert_uft8_cache_coherent("utf8_mg_pos_cache_update", utf8, realutf8,
7437 /* Cache is held with the later position first, to simplify the code
7438 that deals with unbounded ends. */
7440 ASSERT_UTF8_CACHE(cache);
7441 if (cache[1] == 0) {
7442 /* Cache is totally empty */
7445 } else if (cache[3] == 0) {
7446 if (byte > cache[1]) {
7447 /* New one is larger, so goes first. */
7448 cache[2] = cache[0];
7449 cache[3] = cache[1];
7457 /* float casts necessary? XXX */
7458 #define THREEWAY_SQUARE(a,b,c,d) \
7459 ((float)((d) - (c))) * ((float)((d) - (c))) \
7460 + ((float)((c) - (b))) * ((float)((c) - (b))) \
7461 + ((float)((b) - (a))) * ((float)((b) - (a)))
7463 /* Cache has 2 slots in use, and we know three potential pairs.
7464 Keep the two that give the lowest RMS distance. Do the
7465 calculation in bytes simply because we always know the byte
7466 length. squareroot has the same ordering as the positive value,
7467 so don't bother with the actual square root. */
7468 if (byte > cache[1]) {
7469 /* New position is after the existing pair of pairs. */
7470 const float keep_earlier
7471 = THREEWAY_SQUARE(0, cache[3], byte, blen);
7472 const float keep_later
7473 = THREEWAY_SQUARE(0, cache[1], byte, blen);
7475 if (keep_later < keep_earlier) {
7476 cache[2] = cache[0];
7477 cache[3] = cache[1];
7483 const float keep_later = THREEWAY_SQUARE(0, byte, cache[1], blen);
7484 float b, c, keep_earlier;
7485 if (byte > cache[3]) {
7486 /* New position is between the existing pair of pairs. */
7487 b = (float)cache[3];
7490 /* New position is before the existing pair of pairs. */
7492 c = (float)cache[3];
7494 keep_earlier = THREEWAY_SQUARE(0, b, c, blen);
7495 if (byte > cache[3]) {
7496 if (keep_later < keep_earlier) {
7506 if (! (keep_later < keep_earlier)) {
7507 cache[0] = cache[2];
7508 cache[1] = cache[3];
7515 ASSERT_UTF8_CACHE(cache);
7518 /* We already know all of the way, now we may be able to walk back. The same
7519 assumption is made as in S_sv_pos_u2b_midway(), namely that walking
7520 backward is half the speed of walking forward. */
7522 S_sv_pos_b2u_midway(pTHX_ const U8 *const s, const U8 *const target,
7523 const U8 *end, STRLEN endu)
7525 const STRLEN forw = target - s;
7526 STRLEN backw = end - target;
7528 PERL_ARGS_ASSERT_SV_POS_B2U_MIDWAY;
7530 if (forw < 2 * backw) {
7531 return utf8_length(s, target);
7534 while (end > target) {
7536 while (UTF8_IS_CONTINUATION(*end)) {
7545 =for apidoc sv_pos_b2u_flags
7547 Converts C<offset> from a count of bytes from the start of the string, to
7548 a count of the equivalent number of UTF-8 chars. Handles type coercion.
7549 C<flags> is passed to C<SvPV_flags>, and usually should be
7550 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7556 * sv_pos_b2u_flags() uses, like sv_pos_u2b_flags(), the mg_ptr of the
7557 * potential PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8
7562 Perl_sv_pos_b2u_flags(pTHX_ SV *const sv, STRLEN const offset, U32 flags)
7565 STRLEN len = 0; /* Actually always set, but let's keep gcc happy. */
7571 PERL_ARGS_ASSERT_SV_POS_B2U_FLAGS;
7573 s = (const U8*)SvPV_flags(sv, blen, flags);
7576 Perl_croak(aTHX_ "panic: sv_pos_b2u: bad byte offset, blen=%"UVuf
7577 ", byte=%"UVuf, (UV)blen, (UV)offset);
7583 && SvTYPE(sv) >= SVt_PVMG
7584 && (mg = mg_find(sv, PERL_MAGIC_utf8)))
7587 STRLEN * const cache = (STRLEN *) mg->mg_ptr;
7588 if (cache[1] == offset) {
7589 /* An exact match. */
7592 if (cache[3] == offset) {
7593 /* An exact match. */
7597 if (cache[1] < offset) {
7598 /* We already know part of the way. */
7599 if (mg->mg_len != -1) {
7600 /* Actually, we know the end too. */
7602 + S_sv_pos_b2u_midway(aTHX_ s + cache[1], send,
7603 s + blen, mg->mg_len - cache[0]);
7605 len = cache[0] + utf8_length(s + cache[1], send);
7608 else if (cache[3] < offset) {
7609 /* We're between the two cached pairs, so we do the calculation
7610 offset by the byte/utf-8 positions for the earlier pair,
7611 then add the utf-8 characters from the string start to
7613 len = S_sv_pos_b2u_midway(aTHX_ s + cache[3], send,
7614 s + cache[1], cache[0] - cache[2])
7618 else { /* cache[3] > offset */
7619 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + cache[3],
7623 ASSERT_UTF8_CACHE(cache);
7625 } else if (mg->mg_len != -1) {
7626 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + blen, mg->mg_len);
7630 if (!found || PL_utf8cache < 0) {
7631 const STRLEN real_len = utf8_length(s, send);
7633 if (found && PL_utf8cache < 0)
7634 assert_uft8_cache_coherent("sv_pos_b2u", len, real_len, sv);
7640 utf8_mg_len_cache_update(sv, &mg, len);
7642 utf8_mg_pos_cache_update(sv, &mg, offset, len, blen);
7649 =for apidoc sv_pos_b2u
7651 Converts the value pointed to by C<offsetp> from a count of bytes from the
7652 start of the string, to a count of the equivalent number of UTF-8 chars.
7653 Handles magic and type coercion.
7655 Use C<sv_pos_b2u_flags> in preference, which correctly handles strings
7662 * sv_pos_b2u() uses, like sv_pos_u2b(), the mg_ptr of the potential
7663 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7668 Perl_sv_pos_b2u(pTHX_ SV *const sv, I32 *const offsetp)
7670 PERL_ARGS_ASSERT_SV_POS_B2U;
7675 *offsetp = (I32)sv_pos_b2u_flags(sv, (STRLEN)*offsetp,
7676 SV_GMAGIC|SV_CONST_RETURN);
7680 S_assert_uft8_cache_coherent(pTHX_ const char *const func, STRLEN from_cache,
7681 STRLEN real, SV *const sv)
7683 PERL_ARGS_ASSERT_ASSERT_UFT8_CACHE_COHERENT;
7685 /* As this is debugging only code, save space by keeping this test here,
7686 rather than inlining it in all the callers. */
7687 if (from_cache == real)
7690 /* Need to turn the assertions off otherwise we may recurse infinitely
7691 while printing error messages. */
7692 SAVEI8(PL_utf8cache);
7694 Perl_croak(aTHX_ "panic: %s cache %"UVuf" real %"UVuf" for %"SVf,
7695 func, (UV) from_cache, (UV) real, SVfARG(sv));
7701 Returns a boolean indicating whether the strings in the two SVs are
7702 identical. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7703 coerce its args to strings if necessary.
7705 =for apidoc sv_eq_flags
7707 Returns a boolean indicating whether the strings in the two SVs are
7708 identical. Is UTF-8 and S<C<'use bytes'>> aware and coerces its args to strings
7709 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get-magic, too.
7715 Perl_sv_eq_flags(pTHX_ SV *sv1, SV *sv2, const U32 flags)
7722 SV* svrecode = NULL;
7729 /* if pv1 and pv2 are the same, second SvPV_const call may
7730 * invalidate pv1 (if we are handling magic), so we may need to
7732 if (sv1 == sv2 && flags & SV_GMAGIC
7733 && (SvTHINKFIRST(sv1) || SvGMAGICAL(sv1))) {
7734 pv1 = SvPV_const(sv1, cur1);
7735 sv1 = newSVpvn_flags(pv1, cur1, SVs_TEMP | SvUTF8(sv2));
7737 pv1 = SvPV_flags_const(sv1, cur1, flags);
7745 pv2 = SvPV_flags_const(sv2, cur2, flags);
7747 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7748 /* Differing utf8ness. */
7750 /* sv1 is the UTF-8 one */
7751 return bytes_cmp_utf8((const U8*)pv2, cur2,
7752 (const U8*)pv1, cur1) == 0;
7755 /* sv2 is the UTF-8 one */
7756 return bytes_cmp_utf8((const U8*)pv1, cur1,
7757 (const U8*)pv2, cur2) == 0;
7762 eq = (pv1 == pv2) || memEQ(pv1, pv2, cur1);
7764 SvREFCNT_dec(svrecode);
7772 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7773 string in C<sv1> is less than, equal to, or greater than the string in
7774 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7775 coerce its args to strings if necessary. See also C<L</sv_cmp_locale>>.
7777 =for apidoc sv_cmp_flags
7779 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7780 string in C<sv1> is less than, equal to, or greater than the string in
7781 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware and will coerce its args to strings
7782 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get magic. See
7783 also C<L</sv_cmp_locale_flags>>.
7789 Perl_sv_cmp(pTHX_ SV *const sv1, SV *const sv2)
7791 return sv_cmp_flags(sv1, sv2, SV_GMAGIC);
7795 Perl_sv_cmp_flags(pTHX_ SV *const sv1, SV *const sv2,
7799 const char *pv1, *pv2;
7801 SV *svrecode = NULL;
7808 pv1 = SvPV_flags_const(sv1, cur1, flags);
7815 pv2 = SvPV_flags_const(sv2, cur2, flags);
7817 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7818 /* Differing utf8ness. */
7820 const int retval = -bytes_cmp_utf8((const U8*)pv2, cur2,
7821 (const U8*)pv1, cur1);
7822 return retval ? retval < 0 ? -1 : +1 : 0;
7825 const int retval = bytes_cmp_utf8((const U8*)pv1, cur1,
7826 (const U8*)pv2, cur2);
7827 return retval ? retval < 0 ? -1 : +1 : 0;
7831 /* Here, if both are non-NULL, then they have the same UTF8ness. */
7834 cmp = cur2 ? -1 : 0;
7838 STRLEN shortest_len = cur1 < cur2 ? cur1 : cur2;
7841 if (! DO_UTF8(sv1)) {
7843 const I32 retval = memcmp((const void*)pv1,
7847 cmp = retval < 0 ? -1 : 1;
7848 } else if (cur1 == cur2) {
7851 cmp = cur1 < cur2 ? -1 : 1;
7855 else { /* Both are to be treated as UTF-EBCDIC */
7857 /* EBCDIC UTF-8 is complicated by the fact that it is based on I8
7858 * which remaps code points 0-255. We therefore generally have to
7859 * unmap back to the original values to get an accurate comparison.
7860 * But we don't have to do that for UTF-8 invariants, as by
7861 * definition, they aren't remapped, nor do we have to do it for
7862 * above-latin1 code points, as they also aren't remapped. (This
7863 * code also works on ASCII platforms, but the memcmp() above is
7866 const char *e = pv1 + shortest_len;
7868 /* Find the first bytes that differ between the two strings */
7869 while (pv1 < e && *pv1 == *pv2) {
7875 if (pv1 == e) { /* Are the same all the way to the end */
7879 cmp = cur1 < cur2 ? -1 : 1;
7882 else /* Here *pv1 and *pv2 are not equal, but all bytes earlier
7883 * in the strings were. The current bytes may or may not be
7884 * at the beginning of a character. But neither or both are
7885 * (or else earlier bytes would have been different). And
7886 * if we are in the middle of a character, the two
7887 * characters are comprised of the same number of bytes
7888 * (because in this case the start bytes are the same, and
7889 * the start bytes encode the character's length). */
7890 if (UTF8_IS_INVARIANT(*pv1))
7892 /* If both are invariants; can just compare directly */
7893 if (UTF8_IS_INVARIANT(*pv2)) {
7894 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
7896 else /* Since *pv1 is invariant, it is the whole character,
7897 which means it is at the beginning of a character.
7898 That means pv2 is also at the beginning of a
7899 character (see earlier comment). Since it isn't
7900 invariant, it must be a start byte. If it starts a
7901 character whose code point is above 255, that
7902 character is greater than any single-byte char, which
7904 if (UTF8_IS_ABOVE_LATIN1_START(*pv2))
7909 /* Here, pv2 points to a character composed of 2 bytes
7910 * whose code point is < 256. Get its code point and
7911 * compare with *pv1 */
7912 cmp = ((U8) *pv1 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
7917 else /* The code point starting at pv1 isn't a single byte */
7918 if (UTF8_IS_INVARIANT(*pv2))
7920 /* But here, the code point starting at *pv2 is a single byte,
7921 * and so *pv1 must begin a character, hence is a start byte.
7922 * If that character is above 255, it is larger than any
7923 * single-byte char, which *pv2 is */
7924 if (UTF8_IS_ABOVE_LATIN1_START(*pv1)) {
7928 /* Here, pv1 points to a character composed of 2 bytes
7929 * whose code point is < 256. Get its code point and
7930 * compare with the single byte character *pv2 */
7931 cmp = (EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1)) < (U8) *pv2)
7936 else /* Here, we've ruled out either *pv1 and *pv2 being
7937 invariant. That means both are part of variants, but not
7938 necessarily at the start of a character */
7939 if ( UTF8_IS_ABOVE_LATIN1_START(*pv1)
7940 || UTF8_IS_ABOVE_LATIN1_START(*pv2))
7942 /* Here, at least one is the start of a character, which means
7943 * the other is also a start byte. And the code point of at
7944 * least one of the characters is above 255. It is a
7945 * characteristic of UTF-EBCDIC that all start bytes for
7946 * above-latin1 code points are well behaved as far as code
7947 * point comparisons go, and all are larger than all other
7948 * start bytes, so the comparison with those is also well
7950 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
7953 /* Here both *pv1 and *pv2 are part of variant characters.
7954 * They could be both continuations, or both start characters.
7955 * (One or both could even be an illegal start character (for
7956 * an overlong) which for the purposes of sorting we treat as
7958 if (UTF8_IS_CONTINUATION(*pv1)) {
7960 /* If they are continuations for code points above 255,
7961 * then comparing the current byte is sufficient, as there
7962 * is no remapping of these and so the comparison is
7963 * well-behaved. We determine if they are such
7964 * continuations by looking at the preceding byte. It
7965 * could be a start byte, from which we can tell if it is
7966 * for an above 255 code point. Or it could be a
7967 * continuation, which means the character occupies at
7968 * least 3 bytes, so must be above 255. */
7969 if ( UTF8_IS_CONTINUATION(*(pv2 - 1))
7970 || UTF8_IS_ABOVE_LATIN1_START(*(pv2 -1)))
7972 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
7976 /* Here, the continuations are for code points below 256;
7977 * back up one to get to the start byte */
7982 /* We need to get the actual native code point of each of these
7983 * variants in order to compare them */
7984 cmp = ( EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1))
7985 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
7994 SvREFCNT_dec(svrecode);
8000 =for apidoc sv_cmp_locale
8002 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
8003 S<C<'use bytes'>> aware, handles get magic, and will coerce its args to strings
8004 if necessary. See also C<L</sv_cmp>>.
8006 =for apidoc sv_cmp_locale_flags
8008 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
8009 S<C<'use bytes'>> aware and will coerce its args to strings if necessary. If
8010 the flags contain C<SV_GMAGIC>, it handles get magic. See also
8011 C<L</sv_cmp_flags>>.
8017 Perl_sv_cmp_locale(pTHX_ SV *const sv1, SV *const sv2)
8019 return sv_cmp_locale_flags(sv1, sv2, SV_GMAGIC);
8023 Perl_sv_cmp_locale_flags(pTHX_ SV *const sv1, SV *const sv2,
8026 #ifdef USE_LOCALE_COLLATE
8032 if (PL_collation_standard)
8037 /* Revert to using raw compare if both operands exist, but either one
8038 * doesn't transform properly for collation */
8040 pv1 = sv_collxfrm_flags(sv1, &len1, flags);
8044 pv2 = sv_collxfrm_flags(sv2, &len2, flags);
8050 pv1 = sv1 ? sv_collxfrm_flags(sv1, &len1, flags) : (char *) NULL;
8051 pv2 = sv2 ? sv_collxfrm_flags(sv2, &len2, flags) : (char *) NULL;
8054 if (!pv1 || !len1) {
8065 retval = memcmp((void*)pv1, (void*)pv2, len1 < len2 ? len1 : len2);
8068 return retval < 0 ? -1 : 1;
8071 * When the result of collation is equality, that doesn't mean
8072 * that there are no differences -- some locales exclude some
8073 * characters from consideration. So to avoid false equalities,
8074 * we use the raw string as a tiebreaker.
8081 PERL_UNUSED_ARG(flags);
8082 #endif /* USE_LOCALE_COLLATE */
8084 return sv_cmp(sv1, sv2);
8088 #ifdef USE_LOCALE_COLLATE
8091 =for apidoc sv_collxfrm
8093 This calls C<sv_collxfrm_flags> with the SV_GMAGIC flag. See
8094 C<L</sv_collxfrm_flags>>.
8096 =for apidoc sv_collxfrm_flags
8098 Add Collate Transform magic to an SV if it doesn't already have it. If the
8099 flags contain C<SV_GMAGIC>, it handles get-magic.
8101 Any scalar variable may carry C<PERL_MAGIC_collxfrm> magic that contains the
8102 scalar data of the variable, but transformed to such a format that a normal
8103 memory comparison can be used to compare the data according to the locale
8110 Perl_sv_collxfrm_flags(pTHX_ SV *const sv, STRLEN *const nxp, const I32 flags)
8114 PERL_ARGS_ASSERT_SV_COLLXFRM_FLAGS;
8116 mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_collxfrm) : (MAGIC *) NULL;
8118 /* If we don't have collation magic on 'sv', or the locale has changed
8119 * since the last time we calculated it, get it and save it now */
8120 if (!mg || !mg->mg_ptr || *(U32*)mg->mg_ptr != PL_collation_ix) {
8125 /* Free the old space */
8127 Safefree(mg->mg_ptr);
8129 s = SvPV_flags_const(sv, len, flags);
8130 if ((xf = _mem_collxfrm(s, len, &xlen, cBOOL(SvUTF8(sv))))) {
8132 mg = sv_magicext(sv, 0, PERL_MAGIC_collxfrm, &PL_vtbl_collxfrm,
8147 if (mg && mg->mg_ptr) {
8149 return mg->mg_ptr + sizeof(PL_collation_ix);
8157 #endif /* USE_LOCALE_COLLATE */
8160 S_sv_gets_append_to_utf8(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8162 SV * const tsv = newSV(0);
8165 sv_gets(tsv, fp, 0);
8166 sv_utf8_upgrade_nomg(tsv);
8167 SvCUR_set(sv,append);
8170 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8174 S_sv_gets_read_record(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8177 const STRLEN recsize = SvUV(SvRV(PL_rs)); /* RsRECORD() guarantees > 0. */
8178 /* Grab the size of the record we're getting */
8179 char *buffer = SvGROW(sv, (STRLEN)(recsize + append + 1)) + append;
8186 /* With a true, record-oriented file on VMS, we need to use read directly
8187 * to ensure that we respect RMS record boundaries. The user is responsible
8188 * for providing a PL_rs value that corresponds to the FAB$W_MRS (maximum
8189 * record size) field. N.B. This is likely to produce invalid results on
8190 * varying-width character data when a record ends mid-character.
8192 fd = PerlIO_fileno(fp);
8194 && PerlLIO_fstat(fd, &st) == 0
8195 && (st.st_fab_rfm == FAB$C_VAR
8196 || st.st_fab_rfm == FAB$C_VFC
8197 || st.st_fab_rfm == FAB$C_FIX)) {
8199 bytesread = PerlLIO_read(fd, buffer, recsize);
8201 else /* in-memory file from PerlIO::Scalar
8202 * or not a record-oriented file
8206 bytesread = PerlIO_read(fp, buffer, recsize);
8208 /* At this point, the logic in sv_get() means that sv will
8209 be treated as utf-8 if the handle is utf8.
8211 if (PerlIO_isutf8(fp) && bytesread > 0) {
8212 char *bend = buffer + bytesread;
8213 char *bufp = buffer;
8214 size_t charcount = 0;
8215 bool charstart = TRUE;
8218 while (charcount < recsize) {
8219 /* count accumulated characters */
8220 while (bufp < bend) {
8222 skip = UTF8SKIP(bufp);
8224 if (bufp + skip > bend) {
8225 /* partial at the end */
8236 if (charcount < recsize) {
8238 STRLEN bufp_offset = bufp - buffer;
8239 SSize_t morebytesread;
8241 /* originally I read enough to fill any incomplete
8242 character and the first byte of the next
8243 character if needed, but if there's many
8244 multi-byte encoded characters we're going to be
8245 making a read call for every character beyond
8246 the original read size.
8248 So instead, read the rest of the character if
8249 any, and enough bytes to match at least the
8250 start bytes for each character we're going to
8254 readsize = recsize - charcount;
8256 readsize = skip - (bend - bufp) + recsize - charcount - 1;
8257 buffer = SvGROW(sv, append + bytesread + readsize + 1) + append;
8258 bend = buffer + bytesread;
8259 morebytesread = PerlIO_read(fp, bend, readsize);
8260 if (morebytesread <= 0) {
8261 /* we're done, if we still have incomplete
8262 characters the check code in sv_gets() will
8265 I'd originally considered doing
8266 PerlIO_ungetc() on all but the lead
8267 character of the incomplete character, but
8268 read() doesn't do that, so I don't.
8273 /* prepare to scan some more */
8274 bytesread += morebytesread;
8275 bend = buffer + bytesread;
8276 bufp = buffer + bufp_offset;
8284 SvCUR_set(sv, bytesread + append);
8285 buffer[bytesread] = '\0';
8286 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8292 Get a line from the filehandle and store it into the SV, optionally
8293 appending to the currently-stored string. If C<append> is not 0, the
8294 line is appended to the SV instead of overwriting it. C<append> should
8295 be set to the byte offset that the appended string should start at
8296 in the SV (typically, C<SvCUR(sv)> is a suitable choice).
8302 Perl_sv_gets(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8312 PERL_ARGS_ASSERT_SV_GETS;
8314 if (SvTHINKFIRST(sv))
8315 sv_force_normal_flags(sv, append ? 0 : SV_COW_DROP_PV);
8316 /* XXX. If you make this PVIV, then copy on write can copy scalars read
8318 However, perlbench says it's slower, because the existing swipe code
8319 is faster than copy on write.
8320 Swings and roundabouts. */
8321 SvUPGRADE(sv, SVt_PV);
8324 /* line is going to be appended to the existing buffer in the sv */
8325 if (PerlIO_isutf8(fp)) {
8327 sv_utf8_upgrade_nomg(sv);
8328 sv_pos_u2b(sv,&append,0);
8330 } else if (SvUTF8(sv)) {
8331 return S_sv_gets_append_to_utf8(aTHX_ sv, fp, append);
8337 /* not appending - "clear" the string by setting SvCUR to 0,
8338 * the pv is still avaiable. */
8341 if (PerlIO_isutf8(fp))
8344 if (IN_PERL_COMPILETIME) {
8345 /* we always read code in line mode */
8349 else if (RsSNARF(PL_rs)) {
8350 /* If it is a regular disk file use size from stat() as estimate
8351 of amount we are going to read -- may result in mallocing
8352 more memory than we really need if the layers below reduce
8353 the size we read (e.g. CRLF or a gzip layer).
8356 int fd = PerlIO_fileno(fp);
8357 if (fd >= 0 && (PerlLIO_fstat(fd, &st) == 0) && S_ISREG(st.st_mode)) {
8358 const Off_t offset = PerlIO_tell(fp);
8359 if (offset != (Off_t) -1 && st.st_size + append > offset) {
8360 #ifdef PERL_COPY_ON_WRITE
8361 /* Add an extra byte for the sake of copy-on-write's
8362 * buffer reference count. */
8363 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 2));
8365 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 1));
8372 else if (RsRECORD(PL_rs)) {
8373 return S_sv_gets_read_record(aTHX_ sv, fp, append);
8375 else if (RsPARA(PL_rs)) {
8381 /* Get $/ i.e. PL_rs into same encoding as stream wants */
8382 if (PerlIO_isutf8(fp)) {
8383 rsptr = SvPVutf8(PL_rs, rslen);
8386 if (SvUTF8(PL_rs)) {
8387 if (!sv_utf8_downgrade(PL_rs, TRUE)) {
8388 Perl_croak(aTHX_ "Wide character in $/");
8391 /* extract the raw pointer to the record separator */
8392 rsptr = SvPV_const(PL_rs, rslen);
8396 /* rslast is the last character in the record separator
8397 * note we don't use rslast except when rslen is true, so the
8398 * null assign is a placeholder. */
8399 rslast = rslen ? rsptr[rslen - 1] : '\0';
8401 if (rspara) { /* have to do this both before and after */
8402 do { /* to make sure file boundaries work right */
8405 i = PerlIO_getc(fp);
8409 PerlIO_ungetc(fp,i);
8415 /* See if we know enough about I/O mechanism to cheat it ! */
8417 /* This used to be #ifdef test - it is made run-time test for ease
8418 of abstracting out stdio interface. One call should be cheap
8419 enough here - and may even be a macro allowing compile
8423 if (PerlIO_fast_gets(fp)) {
8425 * We can do buffer based IO operations on this filehandle.
8427 * This means we can bypass a lot of subcalls and process
8428 * the buffer directly, it also means we know the upper bound
8429 * on the amount of data we might read of the current buffer
8430 * into our sv. Knowing this allows us to preallocate the pv
8431 * to be able to hold that maximum, which allows us to simplify
8432 * a lot of logic. */
8435 * We're going to steal some values from the stdio struct
8436 * and put EVERYTHING in the innermost loop into registers.
8438 STDCHAR *ptr; /* pointer into fp's read-ahead buffer */
8439 STRLEN bpx; /* length of the data in the target sv
8440 used to fix pointers after a SvGROW */
8441 I32 shortbuffered; /* If the pv buffer is shorter than the amount
8442 of data left in the read-ahead buffer.
8443 If 0 then the pv buffer can hold the full
8444 amount left, otherwise this is the amount it
8447 /* Here is some breathtakingly efficient cheating */
8449 /* When you read the following logic resist the urge to think
8450 * of record separators that are 1 byte long. They are an
8451 * uninteresting special (simple) case.
8453 * Instead think of record separators which are at least 2 bytes
8454 * long, and keep in mind that we need to deal with such
8455 * separators when they cross a read-ahead buffer boundary.
8457 * Also consider that we need to gracefully deal with separators
8458 * that may be longer than a single read ahead buffer.
8460 * Lastly do not forget we want to copy the delimiter as well. We
8461 * are copying all data in the file _up_to_and_including_ the separator
8464 * Now that you have all that in mind here is what is happening below:
8466 * 1. When we first enter the loop we do some memory book keeping to see
8467 * how much free space there is in the target SV. (This sub assumes that
8468 * it is operating on the same SV most of the time via $_ and that it is
8469 * going to be able to reuse the same pv buffer each call.) If there is
8470 * "enough" room then we set "shortbuffered" to how much space there is
8471 * and start reading forward.
8473 * 2. When we scan forward we copy from the read-ahead buffer to the target
8474 * SV's pv buffer. While we go we watch for the end of the read-ahead buffer,
8475 * and the end of the of pv, as well as for the "rslast", which is the last
8476 * char of the separator.
8478 * 3. When scanning forward if we see rslast then we jump backwards in *pv*
8479 * (which has a "complete" record up to the point we saw rslast) and check
8480 * it to see if it matches the separator. If it does we are done. If it doesn't
8481 * we continue on with the scan/copy.
8483 * 4. If we run out of read-ahead buffer (cnt goes to 0) then we have to get
8484 * the IO system to read the next buffer. We do this by doing a getc(), which
8485 * returns a single char read (or EOF), and prefills the buffer, and also
8486 * allows us to find out how full the buffer is. We use this information to
8487 * SvGROW() the sv to the size remaining in the buffer, after which we copy
8488 * the returned single char into the target sv, and then go back into scan
8491 * 5. If we run out of write-buffer then we SvGROW() it by the size of the
8492 * remaining space in the read-buffer.
8494 * Note that this code despite its twisty-turny nature is pretty darn slick.
8495 * It manages single byte separators, multi-byte cross boundary separators,
8496 * and cross-read-buffer separators cleanly and efficiently at the cost
8497 * of potentially greatly overallocating the target SV.
8503 /* get the number of bytes remaining in the read-ahead buffer
8504 * on first call on a given fp this will return 0.*/
8505 cnt = PerlIO_get_cnt(fp);
8507 /* make sure we have the room */
8508 if ((I32)(SvLEN(sv) - append) <= cnt + 1) {
8509 /* Not room for all of it
8510 if we are looking for a separator and room for some
8512 if (rslen && cnt > 80 && (I32)SvLEN(sv) > append) {
8513 /* just process what we have room for */
8514 shortbuffered = cnt - SvLEN(sv) + append + 1;
8515 cnt -= shortbuffered;
8518 /* ensure that the target sv has enough room to hold
8519 * the rest of the read-ahead buffer */
8521 /* remember that cnt can be negative */
8522 SvGROW(sv, (STRLEN)(append + (cnt <= 0 ? 2 : (cnt + 1))));
8526 /* we have enough room to hold the full buffer, lets scream */
8530 /* extract the pointer to sv's string buffer, offset by append as necessary */
8531 bp = (STDCHAR*)SvPVX_const(sv) + append; /* move these two too to registers */
8532 /* extract the point to the read-ahead buffer */
8533 ptr = (STDCHAR*)PerlIO_get_ptr(fp);
8535 /* some trace debug output */
8536 DEBUG_P(PerlIO_printf(Perl_debug_log,
8537 "Screamer: entering, ptr=%"UVuf", cnt=%ld\n",PTR2UV(ptr),(long)cnt));
8538 DEBUG_P(PerlIO_printf(Perl_debug_log,
8539 "Screamer: entering: PerlIO * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"
8541 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8542 PTR2UV(PerlIO_has_base(fp) ? PerlIO_get_base(fp) : 0)));
8546 /* if there is stuff left in the read-ahead buffer */
8548 /* if there is a separator */
8550 /* loop until we hit the end of the read-ahead buffer */
8551 while (cnt > 0) { /* this | eat */
8552 /* scan forward copying and searching for rslast as we go */
8554 if ((*bp++ = *ptr++) == rslast) /* really | dust */
8555 goto thats_all_folks; /* screams | sed :-) */
8559 /* no separator, slurp the full buffer */
8560 Copy(ptr, bp, cnt, char); /* this | eat */
8561 bp += cnt; /* screams | dust */
8562 ptr += cnt; /* louder | sed :-) */
8564 assert (!shortbuffered);
8565 goto cannot_be_shortbuffered;
8569 if (shortbuffered) { /* oh well, must extend */
8570 /* we didnt have enough room to fit the line into the target buffer
8571 * so we must extend the target buffer and keep going */
8572 cnt = shortbuffered;
8574 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8576 /* extned the target sv's buffer so it can hold the full read-ahead buffer */
8577 SvGROW(sv, SvLEN(sv) + append + cnt + 2);
8578 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8582 cannot_be_shortbuffered:
8583 /* we need to refill the read-ahead buffer if possible */
8585 DEBUG_P(PerlIO_printf(Perl_debug_log,
8586 "Screamer: going to getc, ptr=%"UVuf", cnt=%"IVdf"\n",
8587 PTR2UV(ptr),(IV)cnt));
8588 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* deregisterize cnt and ptr */
8590 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8591 "Screamer: pre: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf"\n",
8592 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8593 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8596 call PerlIO_getc() to let it prefill the lookahead buffer
8598 This used to call 'filbuf' in stdio form, but as that behaves like
8599 getc when cnt <= 0 we use PerlIO_getc here to avoid introducing
8600 another abstraction.
8602 Note we have to deal with the char in 'i' if we are not at EOF
8604 i = PerlIO_getc(fp); /* get more characters */
8606 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8607 "Screamer: post: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf"\n",
8608 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8609 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8611 /* find out how much is left in the read-ahead buffer, and rextract its pointer */
8612 cnt = PerlIO_get_cnt(fp);
8613 ptr = (STDCHAR*)PerlIO_get_ptr(fp); /* reregisterize cnt and ptr */
8614 DEBUG_P(PerlIO_printf(Perl_debug_log,
8615 "Screamer: after getc, ptr=%"UVuf", cnt=%"IVdf"\n",
8616 PTR2UV(ptr),(IV)cnt));
8618 if (i == EOF) /* all done for ever? */
8619 goto thats_really_all_folks;
8621 /* make sure we have enough space in the target sv */
8622 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8624 SvGROW(sv, bpx + cnt + 2);
8625 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8627 /* copy of the char we got from getc() */
8628 *bp++ = (STDCHAR)i; /* store character from PerlIO_getc */
8630 /* make sure we deal with the i being the last character of a separator */
8631 if (rslen && (STDCHAR)i == rslast) /* all done for now? */
8632 goto thats_all_folks;
8636 /* check if we have actually found the separator - only really applies
8638 if ((rslen > 1 && (STRLEN)(bp - (STDCHAR*)SvPVX_const(sv)) < rslen) ||
8639 memNE((char*)bp - rslen, rsptr, rslen))
8640 goto screamer; /* go back to the fray */
8641 thats_really_all_folks:
8643 cnt += shortbuffered;
8644 DEBUG_P(PerlIO_printf(Perl_debug_log,
8645 "Screamer: quitting, ptr=%"UVuf", cnt=%"IVdf"\n",PTR2UV(ptr),(IV)cnt));
8646 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* put these back or we're in trouble */
8647 DEBUG_P(PerlIO_printf(Perl_debug_log,
8648 "Screamer: end: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf
8650 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8651 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8653 SvCUR_set(sv, bp - (STDCHAR*)SvPVX_const(sv)); /* set length */
8654 DEBUG_P(PerlIO_printf(Perl_debug_log,
8655 "Screamer: done, len=%ld, string=|%.*s|\n",
8656 (long)SvCUR(sv),(int)SvCUR(sv),SvPVX_const(sv)));
8660 /*The big, slow, and stupid way. */
8661 #ifdef USE_HEAP_INSTEAD_OF_STACK /* Even slower way. */
8662 STDCHAR *buf = NULL;
8663 Newx(buf, 8192, STDCHAR);
8671 const STDCHAR * const bpe = buf + sizeof(buf);
8673 while ((i = PerlIO_getc(fp)) != EOF && (*bp++ = (STDCHAR)i) != rslast && bp < bpe)
8674 ; /* keep reading */
8678 cnt = PerlIO_read(fp,(char*)buf, sizeof(buf));
8679 /* Accommodate broken VAXC compiler, which applies U8 cast to
8680 * both args of ?: operator, causing EOF to change into 255
8683 i = (U8)buf[cnt - 1];
8689 cnt = 0; /* we do need to re-set the sv even when cnt <= 0 */
8691 sv_catpvn_nomg(sv, (char *) buf, cnt);
8693 sv_setpvn(sv, (char *) buf, cnt); /* "nomg" is implied */
8695 if (i != EOF && /* joy */
8697 SvCUR(sv) < rslen ||
8698 memNE(SvPVX_const(sv) + SvCUR(sv) - rslen, rsptr, rslen)))
8702 * If we're reading from a TTY and we get a short read,
8703 * indicating that the user hit his EOF character, we need
8704 * to notice it now, because if we try to read from the TTY
8705 * again, the EOF condition will disappear.
8707 * The comparison of cnt to sizeof(buf) is an optimization
8708 * that prevents unnecessary calls to feof().
8712 if (!(cnt < (I32)sizeof(buf) && PerlIO_eof(fp)))
8716 #ifdef USE_HEAP_INSTEAD_OF_STACK
8721 if (rspara) { /* have to do this both before and after */
8722 while (i != EOF) { /* to make sure file boundaries work right */
8723 i = PerlIO_getc(fp);
8725 PerlIO_ungetc(fp,i);
8731 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8737 Auto-increment of the value in the SV, doing string to numeric conversion
8738 if necessary. Handles 'get' magic and operator overloading.
8744 Perl_sv_inc(pTHX_ SV *const sv)
8753 =for apidoc sv_inc_nomg
8755 Auto-increment of the value in the SV, doing string to numeric conversion
8756 if necessary. Handles operator overloading. Skips handling 'get' magic.
8762 Perl_sv_inc_nomg(pTHX_ SV *const sv)
8769 if (SvTHINKFIRST(sv)) {
8770 if (SvREADONLY(sv)) {
8771 Perl_croak_no_modify();
8775 if (SvAMAGIC(sv) && AMG_CALLunary(sv, inc_amg))
8777 i = PTR2IV(SvRV(sv));
8781 else sv_force_normal_flags(sv, 0);
8783 flags = SvFLAGS(sv);
8784 if ((flags & (SVp_NOK|SVp_IOK)) == SVp_NOK) {
8785 /* It's (privately or publicly) a float, but not tested as an
8786 integer, so test it to see. */
8788 flags = SvFLAGS(sv);
8790 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8791 /* It's publicly an integer, or privately an integer-not-float */
8792 #ifdef PERL_PRESERVE_IVUV
8796 if (SvUVX(sv) == UV_MAX)
8797 sv_setnv(sv, UV_MAX_P1);
8799 (void)SvIOK_only_UV(sv);
8800 SvUV_set(sv, SvUVX(sv) + 1);
8802 if (SvIVX(sv) == IV_MAX)
8803 sv_setuv(sv, (UV)IV_MAX + 1);
8805 (void)SvIOK_only(sv);
8806 SvIV_set(sv, SvIVX(sv) + 1);
8811 if (flags & SVp_NOK) {
8812 const NV was = SvNVX(sv);
8813 if (LIKELY(!Perl_isinfnan(was)) &&
8814 NV_OVERFLOWS_INTEGERS_AT &&
8815 was >= NV_OVERFLOWS_INTEGERS_AT) {
8816 /* diag_listed_as: Lost precision when %s %f by 1 */
8817 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8818 "Lost precision when incrementing %" NVff " by 1",
8821 (void)SvNOK_only(sv);
8822 SvNV_set(sv, was + 1.0);
8826 /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */
8827 if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv)))
8828 Perl_croak_no_modify();
8830 if (!(flags & SVp_POK) || !*SvPVX_const(sv)) {
8831 if ((flags & SVTYPEMASK) < SVt_PVIV)
8832 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV ? SVt_PVIV : SVt_IV));
8833 (void)SvIOK_only(sv);
8838 while (isALPHA(*d)) d++;
8839 while (isDIGIT(*d)) d++;
8840 if (d < SvEND(sv)) {
8841 const int numtype = grok_number_flags(SvPVX_const(sv), SvCUR(sv), NULL, PERL_SCAN_TRAILING);
8842 #ifdef PERL_PRESERVE_IVUV
8843 /* Got to punt this as an integer if needs be, but we don't issue
8844 warnings. Probably ought to make the sv_iv_please() that does
8845 the conversion if possible, and silently. */
8846 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
8847 /* Need to try really hard to see if it's an integer.
8848 9.22337203685478e+18 is an integer.
8849 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
8850 so $a="9.22337203685478e+18"; $a+0; $a++
8851 needs to be the same as $a="9.22337203685478e+18"; $a++
8858 /* sv_2iv *should* have made this an NV */
8859 if (flags & SVp_NOK) {
8860 (void)SvNOK_only(sv);
8861 SvNV_set(sv, SvNVX(sv) + 1.0);
8864 /* I don't think we can get here. Maybe I should assert this
8865 And if we do get here I suspect that sv_setnv will croak. NWC
8867 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_inc punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"NVgf"\n",
8868 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
8870 #endif /* PERL_PRESERVE_IVUV */
8871 if (!numtype && ckWARN(WARN_NUMERIC))
8872 not_incrementable(sv);
8873 sv_setnv(sv,Atof(SvPVX_const(sv)) + 1.0);
8877 while (d >= SvPVX_const(sv)) {
8885 /* MKS: The original code here died if letters weren't consecutive.
8886 * at least it didn't have to worry about non-C locales. The
8887 * new code assumes that ('z'-'a')==('Z'-'A'), letters are
8888 * arranged in order (although not consecutively) and that only
8889 * [A-Za-z] are accepted by isALPHA in the C locale.
8891 if (isALPHA_FOLD_NE(*d, 'z')) {
8892 do { ++*d; } while (!isALPHA(*d));
8895 *(d--) -= 'z' - 'a';
8900 *(d--) -= 'z' - 'a' + 1;
8904 /* oh,oh, the number grew */
8905 SvGROW(sv, SvCUR(sv) + 2);
8906 SvCUR_set(sv, SvCUR(sv) + 1);
8907 for (d = SvPVX(sv) + SvCUR(sv); d > SvPVX_const(sv); d--)
8918 Auto-decrement of the value in the SV, doing string to numeric conversion
8919 if necessary. Handles 'get' magic and operator overloading.
8925 Perl_sv_dec(pTHX_ SV *const sv)
8934 =for apidoc sv_dec_nomg
8936 Auto-decrement of the value in the SV, doing string to numeric conversion
8937 if necessary. Handles operator overloading. Skips handling 'get' magic.
8943 Perl_sv_dec_nomg(pTHX_ SV *const sv)
8949 if (SvTHINKFIRST(sv)) {
8950 if (SvREADONLY(sv)) {
8951 Perl_croak_no_modify();
8955 if (SvAMAGIC(sv) && AMG_CALLunary(sv, dec_amg))
8957 i = PTR2IV(SvRV(sv));
8961 else sv_force_normal_flags(sv, 0);
8963 /* Unlike sv_inc we don't have to worry about string-never-numbers
8964 and keeping them magic. But we mustn't warn on punting */
8965 flags = SvFLAGS(sv);
8966 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8967 /* It's publicly an integer, or privately an integer-not-float */
8968 #ifdef PERL_PRESERVE_IVUV
8972 if (SvUVX(sv) == 0) {
8973 (void)SvIOK_only(sv);
8977 (void)SvIOK_only_UV(sv);
8978 SvUV_set(sv, SvUVX(sv) - 1);
8981 if (SvIVX(sv) == IV_MIN) {
8982 sv_setnv(sv, (NV)IV_MIN);
8986 (void)SvIOK_only(sv);
8987 SvIV_set(sv, SvIVX(sv) - 1);
8992 if (flags & SVp_NOK) {
8995 const NV was = SvNVX(sv);
8996 if (LIKELY(!Perl_isinfnan(was)) &&
8997 NV_OVERFLOWS_INTEGERS_AT &&
8998 was <= -NV_OVERFLOWS_INTEGERS_AT) {
8999 /* diag_listed_as: Lost precision when %s %f by 1 */
9000 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
9001 "Lost precision when decrementing %" NVff " by 1",
9004 (void)SvNOK_only(sv);
9005 SvNV_set(sv, was - 1.0);
9010 /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */
9011 if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv)))
9012 Perl_croak_no_modify();
9014 if (!(flags & SVp_POK)) {
9015 if ((flags & SVTYPEMASK) < SVt_PVIV)
9016 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV) ? SVt_PVIV : SVt_IV);
9018 (void)SvIOK_only(sv);
9021 #ifdef PERL_PRESERVE_IVUV
9023 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), NULL);
9024 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
9025 /* Need to try really hard to see if it's an integer.
9026 9.22337203685478e+18 is an integer.
9027 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
9028 so $a="9.22337203685478e+18"; $a+0; $a--
9029 needs to be the same as $a="9.22337203685478e+18"; $a--
9036 /* sv_2iv *should* have made this an NV */
9037 if (flags & SVp_NOK) {
9038 (void)SvNOK_only(sv);
9039 SvNV_set(sv, SvNVX(sv) - 1.0);
9042 /* I don't think we can get here. Maybe I should assert this
9043 And if we do get here I suspect that sv_setnv will croak. NWC
9045 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_dec punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"NVgf"\n",
9046 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
9049 #endif /* PERL_PRESERVE_IVUV */
9050 sv_setnv(sv,Atof(SvPVX_const(sv)) - 1.0); /* punt */
9053 /* this define is used to eliminate a chunk of duplicated but shared logic
9054 * it has the suffix __SV_C to signal that it isnt API, and isnt meant to be
9055 * used anywhere but here - yves
9057 #define PUSH_EXTEND_MORTAL__SV_C(AnSv) \
9059 SSize_t ix = ++PL_tmps_ix; \
9060 if (UNLIKELY(ix >= PL_tmps_max)) \
9061 ix = tmps_grow_p(ix); \
9062 PL_tmps_stack[ix] = (AnSv); \
9066 =for apidoc sv_mortalcopy
9068 Creates a new SV which is a copy of the original SV (using C<sv_setsv>).
9069 The new SV is marked as mortal. It will be destroyed "soon", either by an
9070 explicit call to C<FREETMPS>, or by an implicit call at places such as
9071 statement boundaries. See also C<L</sv_newmortal>> and C<L</sv_2mortal>>.
9076 /* Make a string that will exist for the duration of the expression
9077 * evaluation. Actually, it may have to last longer than that, but
9078 * hopefully we won't free it until it has been assigned to a
9079 * permanent location. */
9082 Perl_sv_mortalcopy_flags(pTHX_ SV *const oldstr, U32 flags)
9086 if (flags & SV_GMAGIC)
9087 SvGETMAGIC(oldstr); /* before new_SV, in case it dies */
9089 sv_setsv_flags(sv,oldstr,flags & ~SV_GMAGIC);
9090 PUSH_EXTEND_MORTAL__SV_C(sv);
9096 =for apidoc sv_newmortal
9098 Creates a new null SV which is mortal. The reference count of the SV is
9099 set to 1. It will be destroyed "soon", either by an explicit call to
9100 C<FREETMPS>, or by an implicit call at places such as statement boundaries.
9101 See also C<L</sv_mortalcopy>> and C<L</sv_2mortal>>.
9107 Perl_sv_newmortal(pTHX)
9112 SvFLAGS(sv) = SVs_TEMP;
9113 PUSH_EXTEND_MORTAL__SV_C(sv);
9119 =for apidoc newSVpvn_flags
9121 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9122 characters) into it. The reference count for the
9123 SV is set to 1. Note that if C<len> is zero, Perl will create a zero length
9124 string. You are responsible for ensuring that the source string is at least
9125 C<len> bytes long. If the C<s> argument is NULL the new SV will be undefined.
9126 Currently the only flag bits accepted are C<SVf_UTF8> and C<SVs_TEMP>.
9127 If C<SVs_TEMP> is set, then C<sv_2mortal()> is called on the result before
9128 returning. If C<SVf_UTF8> is set, C<s>
9129 is considered to be in UTF-8 and the
9130 C<SVf_UTF8> flag will be set on the new SV.
9131 C<newSVpvn_utf8()> is a convenience wrapper for this function, defined as
9133 #define newSVpvn_utf8(s, len, u) \
9134 newSVpvn_flags((s), (len), (u) ? SVf_UTF8 : 0)
9140 Perl_newSVpvn_flags(pTHX_ const char *const s, const STRLEN len, const U32 flags)
9144 /* All the flags we don't support must be zero.
9145 And we're new code so I'm going to assert this from the start. */
9146 assert(!(flags & ~(SVf_UTF8|SVs_TEMP)));
9148 sv_setpvn(sv,s,len);
9150 /* This code used to do a sv_2mortal(), however we now unroll the call to
9151 * sv_2mortal() and do what it does ourselves here. Since we have asserted
9152 * that flags can only have the SVf_UTF8 and/or SVs_TEMP flags set above we
9153 * can use it to enable the sv flags directly (bypassing SvTEMP_on), which
9154 * in turn means we dont need to mask out the SVf_UTF8 flag below, which
9155 * means that we eliminate quite a few steps than it looks - Yves
9156 * (explaining patch by gfx) */
9158 SvFLAGS(sv) |= flags;
9160 if(flags & SVs_TEMP){
9161 PUSH_EXTEND_MORTAL__SV_C(sv);
9168 =for apidoc sv_2mortal
9170 Marks an existing SV as mortal. The SV will be destroyed "soon", either
9171 by an explicit call to C<FREETMPS>, or by an implicit call at places such as
9172 statement boundaries. C<SvTEMP()> is turned on which means that the SV's
9173 string buffer can be "stolen" if this SV is copied. See also
9174 C<L</sv_newmortal>> and C<L</sv_mortalcopy>>.
9180 Perl_sv_2mortal(pTHX_ SV *const sv)
9187 PUSH_EXTEND_MORTAL__SV_C(sv);
9195 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9196 characters) into it. The reference count for the
9197 SV is set to 1. If C<len> is zero, Perl will compute the length using
9198 C<strlen()>, (which means if you use this option, that C<s> can't have embedded
9199 C<NUL> characters and has to have a terminating C<NUL> byte).
9201 For efficiency, consider using C<newSVpvn> instead.
9207 Perl_newSVpv(pTHX_ const char *const s, const STRLEN len)
9212 sv_setpvn(sv, s, len || s == NULL ? len : strlen(s));
9217 =for apidoc newSVpvn
9219 Creates a new SV and copies a string into it, which may contain C<NUL> characters
9220 (C<\0>) and other binary data. The reference count for the SV is set to 1.
9221 Note that if C<len> is zero, Perl will create a zero length (Perl) string. You
9222 are responsible for ensuring that the source buffer is at least
9223 C<len> bytes long. If the C<buffer> argument is NULL the new SV will be
9230 Perl_newSVpvn(pTHX_ const char *const buffer, const STRLEN len)
9234 sv_setpvn(sv,buffer,len);
9239 =for apidoc newSVhek
9241 Creates a new SV from the hash key structure. It will generate scalars that
9242 point to the shared string table where possible. Returns a new (undefined)
9243 SV if C<hek> is NULL.
9249 Perl_newSVhek(pTHX_ const HEK *const hek)
9258 if (HEK_LEN(hek) == HEf_SVKEY) {
9259 return newSVsv(*(SV**)HEK_KEY(hek));
9261 const int flags = HEK_FLAGS(hek);
9262 if (flags & HVhek_WASUTF8) {
9264 Andreas would like keys he put in as utf8 to come back as utf8
9266 STRLEN utf8_len = HEK_LEN(hek);
9267 SV * const sv = newSV_type(SVt_PV);
9268 char *as_utf8 = (char *)bytes_to_utf8 ((U8*)HEK_KEY(hek), &utf8_len);
9269 /* bytes_to_utf8() allocates a new string, which we can repurpose: */
9270 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
9273 } else if (flags & HVhek_UNSHARED) {
9274 /* A hash that isn't using shared hash keys has to have
9275 the flag in every key so that we know not to try to call
9276 share_hek_hek on it. */
9278 SV * const sv = newSVpvn (HEK_KEY(hek), HEK_LEN(hek));
9283 /* This will be overwhelminly the most common case. */
9285 /* Inline most of newSVpvn_share(), because share_hek_hek() is far
9286 more efficient than sharepvn(). */
9290 sv_upgrade(sv, SVt_PV);
9291 SvPV_set(sv, (char *)HEK_KEY(share_hek_hek(hek)));
9292 SvCUR_set(sv, HEK_LEN(hek));
9304 =for apidoc newSVpvn_share
9306 Creates a new SV with its C<SvPVX_const> pointing to a shared string in the string
9307 table. If the string does not already exist in the table, it is
9308 created first. Turns on the C<SvIsCOW> flag (or C<READONLY>
9309 and C<FAKE> in 5.16 and earlier). If the C<hash> parameter
9310 is non-zero, that value is used; otherwise the hash is computed.
9311 The string's hash can later be retrieved from the SV
9312 with the C<SvSHARED_HASH()> macro. The idea here is
9313 that as the string table is used for shared hash keys these strings will have
9314 C<SvPVX_const == HeKEY> and hash lookup will avoid string compare.
9320 Perl_newSVpvn_share(pTHX_ const char *src, I32 len, U32 hash)
9324 bool is_utf8 = FALSE;
9325 const char *const orig_src = src;
9328 STRLEN tmplen = -len;
9330 /* See the note in hv.c:hv_fetch() --jhi */
9331 src = (char*)bytes_from_utf8((const U8*)src, &tmplen, &is_utf8);
9335 PERL_HASH(hash, src, len);
9337 /* The logic for this is inlined in S_mro_get_linear_isa_dfs(), so if it
9338 changes here, update it there too. */
9339 sv_upgrade(sv, SVt_PV);
9340 SvPV_set(sv, sharepvn(src, is_utf8?-len:len, hash));
9347 if (src != orig_src)
9353 =for apidoc newSVpv_share
9355 Like C<newSVpvn_share>, but takes a C<NUL>-terminated string instead of a
9362 Perl_newSVpv_share(pTHX_ const char *src, U32 hash)
9364 return newSVpvn_share(src, strlen(src), hash);
9367 #if defined(PERL_IMPLICIT_CONTEXT)
9369 /* pTHX_ magic can't cope with varargs, so this is a no-context
9370 * version of the main function, (which may itself be aliased to us).
9371 * Don't access this version directly.
9375 Perl_newSVpvf_nocontext(const char *const pat, ...)
9381 PERL_ARGS_ASSERT_NEWSVPVF_NOCONTEXT;
9383 va_start(args, pat);
9384 sv = vnewSVpvf(pat, &args);
9391 =for apidoc newSVpvf
9393 Creates a new SV and initializes it with the string formatted like
9400 Perl_newSVpvf(pTHX_ const char *const pat, ...)
9405 PERL_ARGS_ASSERT_NEWSVPVF;
9407 va_start(args, pat);
9408 sv = vnewSVpvf(pat, &args);
9413 /* backend for newSVpvf() and newSVpvf_nocontext() */
9416 Perl_vnewSVpvf(pTHX_ const char *const pat, va_list *const args)
9420 PERL_ARGS_ASSERT_VNEWSVPVF;
9423 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
9430 Creates a new SV and copies a floating point value into it.
9431 The reference count for the SV is set to 1.
9437 Perl_newSVnv(pTHX_ const NV n)
9449 Creates a new SV and copies an integer into it. The reference count for the
9456 Perl_newSViv(pTHX_ const IV i)
9462 /* Inlining ONLY the small relevant subset of sv_setiv here
9463 * for performance. Makes a significant difference. */
9465 /* We're starting from SVt_FIRST, so provided that's
9466 * actual 0, we don't have to unset any SV type flags
9467 * to promote to SVt_IV. */
9468 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9470 SET_SVANY_FOR_BODYLESS_IV(sv);
9471 SvFLAGS(sv) |= SVt_IV;
9483 Creates a new SV and copies an unsigned integer into it.
9484 The reference count for the SV is set to 1.
9490 Perl_newSVuv(pTHX_ const UV u)
9494 /* Inlining ONLY the small relevant subset of sv_setuv here
9495 * for performance. Makes a significant difference. */
9497 /* Using ivs is more efficient than using uvs - see sv_setuv */
9498 if (u <= (UV)IV_MAX) {
9499 return newSViv((IV)u);
9504 /* We're starting from SVt_FIRST, so provided that's
9505 * actual 0, we don't have to unset any SV type flags
9506 * to promote to SVt_IV. */
9507 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9509 SET_SVANY_FOR_BODYLESS_IV(sv);
9510 SvFLAGS(sv) |= SVt_IV;
9512 (void)SvIsUV_on(sv);
9521 =for apidoc newSV_type
9523 Creates a new SV, of the type specified. The reference count for the new SV
9530 Perl_newSV_type(pTHX_ const svtype type)
9535 ASSUME(SvTYPE(sv) == SVt_FIRST);
9536 if(type != SVt_FIRST)
9537 sv_upgrade(sv, type);
9542 =for apidoc newRV_noinc
9544 Creates an RV wrapper for an SV. The reference count for the original
9545 SV is B<not> incremented.
9551 Perl_newRV_noinc(pTHX_ SV *const tmpRef)
9555 PERL_ARGS_ASSERT_NEWRV_NOINC;
9559 /* We're starting from SVt_FIRST, so provided that's
9560 * actual 0, we don't have to unset any SV type flags
9561 * to promote to SVt_IV. */
9562 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9564 SET_SVANY_FOR_BODYLESS_IV(sv);
9565 SvFLAGS(sv) |= SVt_IV;
9570 SvRV_set(sv, tmpRef);
9575 /* newRV_inc is the official function name to use now.
9576 * newRV_inc is in fact #defined to newRV in sv.h
9580 Perl_newRV(pTHX_ SV *const sv)
9582 PERL_ARGS_ASSERT_NEWRV;
9584 return newRV_noinc(SvREFCNT_inc_simple_NN(sv));
9590 Creates a new SV which is an exact duplicate of the original SV.
9597 Perl_newSVsv(pTHX_ SV *const old)
9603 if (SvTYPE(old) == (svtype)SVTYPEMASK) {
9604 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), "semi-panic: attempt to dup freed string");
9607 /* Do this here, otherwise we leak the new SV if this croaks. */
9610 /* SV_NOSTEAL prevents TEMP buffers being, well, stolen, and saves games
9611 with SvTEMP_off and SvTEMP_on round a call to sv_setsv. */
9612 sv_setsv_flags(sv, old, SV_NOSTEAL);
9617 =for apidoc sv_reset
9619 Underlying implementation for the C<reset> Perl function.
9620 Note that the perl-level function is vaguely deprecated.
9626 Perl_sv_reset(pTHX_ const char *s, HV *const stash)
9628 PERL_ARGS_ASSERT_SV_RESET;
9630 sv_resetpvn(*s ? s : NULL, strlen(s), stash);
9634 Perl_sv_resetpvn(pTHX_ const char *s, STRLEN len, HV * const stash)
9636 char todo[PERL_UCHAR_MAX+1];
9639 if (!stash || SvTYPE(stash) != SVt_PVHV)
9642 if (!s) { /* reset ?? searches */
9643 MAGIC * const mg = mg_find((const SV *)stash, PERL_MAGIC_symtab);
9645 const U32 count = mg->mg_len / sizeof(PMOP**);
9646 PMOP **pmp = (PMOP**) mg->mg_ptr;
9647 PMOP *const *const end = pmp + count;
9651 SvREADONLY_off(PL_regex_pad[(*pmp)->op_pmoffset]);
9653 (*pmp)->op_pmflags &= ~PMf_USED;
9661 /* reset variables */
9663 if (!HvARRAY(stash))
9666 Zero(todo, 256, char);
9670 I32 i = (unsigned char)*s;
9674 max = (unsigned char)*s++;
9675 for ( ; i <= max; i++) {
9678 for (i = 0; i <= (I32) HvMAX(stash); i++) {
9680 for (entry = HvARRAY(stash)[i];
9682 entry = HeNEXT(entry))
9687 if (!todo[(U8)*HeKEY(entry)])
9689 gv = MUTABLE_GV(HeVAL(entry));
9693 if (sv && !SvREADONLY(sv)) {
9694 SV_CHECK_THINKFIRST_COW_DROP(sv);
9695 if (!isGV(sv)) SvOK_off(sv);
9700 if (GvHV(gv) && !HvNAME_get(GvHV(gv))) {
9711 Using various gambits, try to get an IO from an SV: the IO slot if its a
9712 GV; or the recursive result if we're an RV; or the IO slot of the symbol
9713 named after the PV if we're a string.
9715 'Get' magic is ignored on the C<sv> passed in, but will be called on
9716 C<SvRV(sv)> if C<sv> is an RV.
9722 Perl_sv_2io(pTHX_ SV *const sv)
9727 PERL_ARGS_ASSERT_SV_2IO;
9729 switch (SvTYPE(sv)) {
9731 io = MUTABLE_IO(sv);
9735 if (isGV_with_GP(sv)) {
9736 gv = MUTABLE_GV(sv);
9739 Perl_croak(aTHX_ "Bad filehandle: %"HEKf,
9740 HEKfARG(GvNAME_HEK(gv)));
9746 Perl_croak(aTHX_ PL_no_usym, "filehandle");
9748 SvGETMAGIC(SvRV(sv));
9749 return sv_2io(SvRV(sv));
9751 gv = gv_fetchsv_nomg(sv, 0, SVt_PVIO);
9758 if (SvGMAGICAL(sv)) {
9759 newsv = sv_newmortal();
9760 sv_setsv_nomg(newsv, sv);
9762 Perl_croak(aTHX_ "Bad filehandle: %"SVf, SVfARG(newsv));
9772 Using various gambits, try to get a CV from an SV; in addition, try if
9773 possible to set C<*st> and C<*gvp> to the stash and GV associated with it.
9774 The flags in C<lref> are passed to C<gv_fetchsv>.
9780 Perl_sv_2cv(pTHX_ SV *sv, HV **const st, GV **const gvp, const I32 lref)
9785 PERL_ARGS_ASSERT_SV_2CV;
9792 switch (SvTYPE(sv)) {
9796 return MUTABLE_CV(sv);
9806 sv = amagic_deref_call(sv, to_cv_amg);
9809 if (SvTYPE(sv) == SVt_PVCV) {
9810 cv = MUTABLE_CV(sv);
9815 else if(SvGETMAGIC(sv), isGV_with_GP(sv))
9816 gv = MUTABLE_GV(sv);
9818 Perl_croak(aTHX_ "Not a subroutine reference");
9820 else if (isGV_with_GP(sv)) {
9821 gv = MUTABLE_GV(sv);
9824 gv = gv_fetchsv_nomg(sv, lref, SVt_PVCV);
9831 /* Some flags to gv_fetchsv mean don't really create the GV */
9832 if (!isGV_with_GP(gv)) {
9837 if (lref & ~GV_ADDMG && !GvCVu(gv)) {
9838 /* XXX this is probably not what they think they're getting.
9839 * It has the same effect as "sub name;", i.e. just a forward
9850 Returns true if the SV has a true value by Perl's rules.
9851 Use the C<SvTRUE> macro instead, which may call C<sv_true()> or may
9852 instead use an in-line version.
9858 Perl_sv_true(pTHX_ SV *const sv)
9863 const XPV* const tXpv = (XPV*)SvANY(sv);
9865 (tXpv->xpv_cur > 1 ||
9866 (tXpv->xpv_cur && *sv->sv_u.svu_pv != '0')))
9873 return SvIVX(sv) != 0;
9876 return SvNVX(sv) != 0.0;
9878 return sv_2bool(sv);
9884 =for apidoc sv_pvn_force
9886 Get a sensible string out of the SV somehow.
9887 A private implementation of the C<SvPV_force> macro for compilers which
9888 can't cope with complex macro expressions. Always use the macro instead.
9890 =for apidoc sv_pvn_force_flags
9892 Get a sensible string out of the SV somehow.
9893 If C<flags> has the C<SV_GMAGIC> bit set, will C<mg_get> on C<sv> if
9894 appropriate, else not. C<sv_pvn_force> and C<sv_pvn_force_nomg> are
9895 implemented in terms of this function.
9896 You normally want to use the various wrapper macros instead: see
9897 C<L</SvPV_force>> and C<L</SvPV_force_nomg>>.
9903 Perl_sv_pvn_force_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
9905 PERL_ARGS_ASSERT_SV_PVN_FORCE_FLAGS;
9907 if (flags & SV_GMAGIC) SvGETMAGIC(sv);
9908 if (SvTHINKFIRST(sv) && (!SvROK(sv) || SvREADONLY(sv)))
9909 sv_force_normal_flags(sv, 0);
9919 if (SvTYPE(sv) > SVt_PVLV
9920 || isGV_with_GP(sv))
9921 /* diag_listed_as: Can't coerce %s to %s in %s */
9922 Perl_croak(aTHX_ "Can't coerce %s to string in %s", sv_reftype(sv,0),
9924 s = sv_2pv_flags(sv, &len, flags &~ SV_GMAGIC);
9931 if (SvTYPE(sv) < SVt_PV ||
9932 s != SvPVX_const(sv)) { /* Almost, but not quite, sv_setpvn() */
9935 SvUPGRADE(sv, SVt_PV); /* Never FALSE */
9936 SvGROW(sv, len + 1);
9937 Move(s,SvPVX(sv),len,char);
9939 SvPVX(sv)[len] = '\0';
9942 SvPOK_on(sv); /* validate pointer */
9944 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
9945 PTR2UV(sv),SvPVX_const(sv)));
9948 (void)SvPOK_only_UTF8(sv);
9949 return SvPVX_mutable(sv);
9953 =for apidoc sv_pvbyten_force
9955 The backend for the C<SvPVbytex_force> macro. Always use the macro
9962 Perl_sv_pvbyten_force(pTHX_ SV *const sv, STRLEN *const lp)
9964 PERL_ARGS_ASSERT_SV_PVBYTEN_FORCE;
9966 sv_pvn_force(sv,lp);
9967 sv_utf8_downgrade(sv,0);
9973 =for apidoc sv_pvutf8n_force
9975 The backend for the C<SvPVutf8x_force> macro. Always use the macro
9982 Perl_sv_pvutf8n_force(pTHX_ SV *const sv, STRLEN *const lp)
9984 PERL_ARGS_ASSERT_SV_PVUTF8N_FORCE;
9987 sv_utf8_upgrade_nomg(sv);
9993 =for apidoc sv_reftype
9995 Returns a string describing what the SV is a reference to.
9997 If ob is true and the SV is blessed, the string is the class name,
9998 otherwise it is the type of the SV, "SCALAR", "ARRAY" etc.
10004 Perl_sv_reftype(pTHX_ const SV *const sv, const int ob)
10006 PERL_ARGS_ASSERT_SV_REFTYPE;
10007 if (ob && SvOBJECT(sv)) {
10008 return SvPV_nolen_const(sv_ref(NULL, sv, ob));
10011 /* WARNING - There is code, for instance in mg.c, that assumes that
10012 * the only reason that sv_reftype(sv,0) would return a string starting
10013 * with 'L' or 'S' is that it is a LVALUE or a SCALAR.
10014 * Yes this a dodgy way to do type checking, but it saves practically reimplementing
10015 * this routine inside other subs, and it saves time.
10016 * Do not change this assumption without searching for "dodgy type check" in
10019 switch (SvTYPE(sv)) {
10034 case SVt_PVLV: return (char *) (SvROK(sv) ? "REF"
10035 /* tied lvalues should appear to be
10036 * scalars for backwards compatibility */
10037 : (isALPHA_FOLD_EQ(LvTYPE(sv), 't'))
10038 ? "SCALAR" : "LVALUE");
10039 case SVt_PVAV: return "ARRAY";
10040 case SVt_PVHV: return "HASH";
10041 case SVt_PVCV: return "CODE";
10042 case SVt_PVGV: return (char *) (isGV_with_GP(sv)
10043 ? "GLOB" : "SCALAR");
10044 case SVt_PVFM: return "FORMAT";
10045 case SVt_PVIO: return "IO";
10046 case SVt_INVLIST: return "INVLIST";
10047 case SVt_REGEXP: return "REGEXP";
10048 default: return "UNKNOWN";
10056 Returns a SV describing what the SV passed in is a reference to.
10058 dst can be a SV to be set to the description or NULL, in which case a
10059 mortal SV is returned.
10061 If ob is true and the SV is blessed, the description is the class
10062 name, otherwise it is the type of the SV, "SCALAR", "ARRAY" etc.
10068 Perl_sv_ref(pTHX_ SV *dst, const SV *const sv, const int ob)
10070 PERL_ARGS_ASSERT_SV_REF;
10073 dst = sv_newmortal();
10075 if (ob && SvOBJECT(sv)) {
10076 HvNAME_get(SvSTASH(sv))
10077 ? sv_sethek(dst, HvNAME_HEK(SvSTASH(sv)))
10078 : sv_setpvn(dst, "__ANON__", 8);
10081 const char * reftype = sv_reftype(sv, 0);
10082 sv_setpv(dst, reftype);
10088 =for apidoc sv_isobject
10090 Returns a boolean indicating whether the SV is an RV pointing to a blessed
10091 object. If the SV is not an RV, or if the object is not blessed, then this
10098 Perl_sv_isobject(pTHX_ SV *sv)
10114 Returns a boolean indicating whether the SV is blessed into the specified
10115 class. This does not check for subtypes; use C<sv_derived_from> to verify
10116 an inheritance relationship.
10122 Perl_sv_isa(pTHX_ SV *sv, const char *const name)
10124 const char *hvname;
10126 PERL_ARGS_ASSERT_SV_ISA;
10136 hvname = HvNAME_get(SvSTASH(sv));
10140 return strEQ(hvname, name);
10144 =for apidoc newSVrv
10146 Creates a new SV for the existing RV, C<rv>, to point to. If C<rv> is not an
10147 RV then it will be upgraded to one. If C<classname> is non-null then the new
10148 SV will be blessed in the specified package. The new SV is returned and its
10149 reference count is 1. The reference count 1 is owned by C<rv>.
10155 Perl_newSVrv(pTHX_ SV *const rv, const char *const classname)
10159 PERL_ARGS_ASSERT_NEWSVRV;
10163 SV_CHECK_THINKFIRST_COW_DROP(rv);
10165 if (UNLIKELY( SvTYPE(rv) >= SVt_PVMG )) {
10166 const U32 refcnt = SvREFCNT(rv);
10170 SvREFCNT(rv) = refcnt;
10172 sv_upgrade(rv, SVt_IV);
10173 } else if (SvROK(rv)) {
10174 SvREFCNT_dec(SvRV(rv));
10176 prepare_SV_for_RV(rv);
10184 HV* const stash = gv_stashpv(classname, GV_ADD);
10185 (void)sv_bless(rv, stash);
10191 Perl_newSVavdefelem(pTHX_ AV *av, SSize_t ix, bool extendible)
10193 SV * const lv = newSV_type(SVt_PVLV);
10194 PERL_ARGS_ASSERT_NEWSVAVDEFELEM;
10196 sv_magic(lv, NULL, PERL_MAGIC_defelem, NULL, 0);
10197 LvTARG(lv) = SvREFCNT_inc_simple_NN(av);
10198 LvSTARGOFF(lv) = ix;
10199 LvTARGLEN(lv) = extendible ? 1 : (STRLEN)UV_MAX;
10204 =for apidoc sv_setref_pv
10206 Copies a pointer into a new SV, optionally blessing the SV. The C<rv>
10207 argument will be upgraded to an RV. That RV will be modified to point to
10208 the new SV. If the C<pv> argument is C<NULL>, then C<PL_sv_undef> will be placed
10209 into the SV. The C<classname> argument indicates the package for the
10210 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10211 will have a reference count of 1, and the RV will be returned.
10213 Do not use with other Perl types such as HV, AV, SV, CV, because those
10214 objects will become corrupted by the pointer copy process.
10216 Note that C<sv_setref_pvn> copies the string while this copies the pointer.
10222 Perl_sv_setref_pv(pTHX_ SV *const rv, const char *const classname, void *const pv)
10224 PERL_ARGS_ASSERT_SV_SETREF_PV;
10227 sv_setsv(rv, &PL_sv_undef);
10231 sv_setiv(newSVrv(rv,classname), PTR2IV(pv));
10236 =for apidoc sv_setref_iv
10238 Copies an integer into a new SV, optionally blessing the SV. The C<rv>
10239 argument will be upgraded to an RV. That RV will be modified to point to
10240 the new SV. The C<classname> argument indicates the package for the
10241 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10242 will have a reference count of 1, and the RV will be returned.
10248 Perl_sv_setref_iv(pTHX_ SV *const rv, const char *const classname, const IV iv)
10250 PERL_ARGS_ASSERT_SV_SETREF_IV;
10252 sv_setiv(newSVrv(rv,classname), iv);
10257 =for apidoc sv_setref_uv
10259 Copies an unsigned integer into a new SV, optionally blessing the SV. The C<rv>
10260 argument will be upgraded to an RV. That RV will be modified to point to
10261 the new SV. The C<classname> argument indicates the package for the
10262 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10263 will have a reference count of 1, and the RV will be returned.
10269 Perl_sv_setref_uv(pTHX_ SV *const rv, const char *const classname, const UV uv)
10271 PERL_ARGS_ASSERT_SV_SETREF_UV;
10273 sv_setuv(newSVrv(rv,classname), uv);
10278 =for apidoc sv_setref_nv
10280 Copies a double into a new SV, optionally blessing the SV. The C<rv>
10281 argument will be upgraded to an RV. That RV will be modified to point to
10282 the new SV. The C<classname> argument indicates the package for the
10283 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10284 will have a reference count of 1, and the RV will be returned.
10290 Perl_sv_setref_nv(pTHX_ SV *const rv, const char *const classname, const NV nv)
10292 PERL_ARGS_ASSERT_SV_SETREF_NV;
10294 sv_setnv(newSVrv(rv,classname), nv);
10299 =for apidoc sv_setref_pvn
10301 Copies a string into a new SV, optionally blessing the SV. The length of the
10302 string must be specified with C<n>. The C<rv> argument will be upgraded to
10303 an RV. That RV will be modified to point to the new SV. The C<classname>
10304 argument indicates the package for the blessing. Set C<classname> to
10305 C<NULL> to avoid the blessing. The new SV will have a reference count
10306 of 1, and the RV will be returned.
10308 Note that C<sv_setref_pv> copies the pointer while this copies the string.
10314 Perl_sv_setref_pvn(pTHX_ SV *const rv, const char *const classname,
10315 const char *const pv, const STRLEN n)
10317 PERL_ARGS_ASSERT_SV_SETREF_PVN;
10319 sv_setpvn(newSVrv(rv,classname), pv, n);
10324 =for apidoc sv_bless
10326 Blesses an SV into a specified package. The SV must be an RV. The package
10327 must be designated by its stash (see C<L</gv_stashpv>>). The reference count
10328 of the SV is unaffected.
10334 Perl_sv_bless(pTHX_ SV *const sv, HV *const stash)
10337 HV *oldstash = NULL;
10339 PERL_ARGS_ASSERT_SV_BLESS;
10343 Perl_croak(aTHX_ "Can't bless non-reference value");
10345 if (SvFLAGS(tmpRef) & (SVs_OBJECT|SVf_READONLY|SVf_PROTECT)) {
10346 if (SvREADONLY(tmpRef))
10347 Perl_croak_no_modify();
10348 if (SvOBJECT(tmpRef)) {
10349 oldstash = SvSTASH(tmpRef);
10352 SvOBJECT_on(tmpRef);
10353 SvUPGRADE(tmpRef, SVt_PVMG);
10354 SvSTASH_set(tmpRef, MUTABLE_HV(SvREFCNT_inc_simple(stash)));
10355 SvREFCNT_dec(oldstash);
10357 if(SvSMAGICAL(tmpRef))
10358 if(mg_find(tmpRef, PERL_MAGIC_ext) || mg_find(tmpRef, PERL_MAGIC_uvar))
10366 /* Downgrades a PVGV to a PVMG. If it's actually a PVLV, we leave the type
10367 * as it is after unglobbing it.
10370 PERL_STATIC_INLINE void
10371 S_sv_unglob(pTHX_ SV *const sv, U32 flags)
10375 SV * const temp = flags & SV_COW_DROP_PV ? NULL : sv_newmortal();
10377 PERL_ARGS_ASSERT_SV_UNGLOB;
10379 assert(SvTYPE(sv) == SVt_PVGV || SvTYPE(sv) == SVt_PVLV);
10381 if (!(flags & SV_COW_DROP_PV))
10382 gv_efullname3(temp, MUTABLE_GV(sv), "*");
10384 SvREFCNT_inc_simple_void_NN(sv_2mortal(sv));
10386 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
10387 && HvNAME_get(stash))
10388 mro_method_changed_in(stash);
10389 gp_free(MUTABLE_GV(sv));
10392 sv_del_backref(MUTABLE_SV(GvSTASH(sv)), sv);
10393 GvSTASH(sv) = NULL;
10396 if (GvNAME_HEK(sv)) {
10397 unshare_hek(GvNAME_HEK(sv));
10399 isGV_with_GP_off(sv);
10401 if(SvTYPE(sv) == SVt_PVGV) {
10402 /* need to keep SvANY(sv) in the right arena */
10403 xpvmg = new_XPVMG();
10404 StructCopy(SvANY(sv), xpvmg, XPVMG);
10405 del_XPVGV(SvANY(sv));
10408 SvFLAGS(sv) &= ~SVTYPEMASK;
10409 SvFLAGS(sv) |= SVt_PVMG;
10412 /* Intentionally not calling any local SET magic, as this isn't so much a
10413 set operation as merely an internal storage change. */
10414 if (flags & SV_COW_DROP_PV) SvOK_off(sv);
10415 else sv_setsv_flags(sv, temp, 0);
10417 if ((const GV *)sv == PL_last_in_gv)
10418 PL_last_in_gv = NULL;
10419 else if ((const GV *)sv == PL_statgv)
10424 =for apidoc sv_unref_flags
10426 Unsets the RV status of the SV, and decrements the reference count of
10427 whatever was being referenced by the RV. This can almost be thought of
10428 as a reversal of C<newSVrv>. The C<cflags> argument can contain
10429 C<SV_IMMEDIATE_UNREF> to force the reference count to be decremented
10430 (otherwise the decrementing is conditional on the reference count being
10431 different from one or the reference being a readonly SV).
10432 See C<L</SvROK_off>>.
10438 Perl_sv_unref_flags(pTHX_ SV *const ref, const U32 flags)
10440 SV* const target = SvRV(ref);
10442 PERL_ARGS_ASSERT_SV_UNREF_FLAGS;
10444 if (SvWEAKREF(ref)) {
10445 sv_del_backref(target, ref);
10446 SvWEAKREF_off(ref);
10447 SvRV_set(ref, NULL);
10450 SvRV_set(ref, NULL);
10452 /* You can't have a || SvREADONLY(target) here, as $a = $$a, where $a was
10453 assigned to as BEGIN {$a = \"Foo"} will fail. */
10454 if (SvREFCNT(target) != 1 || (flags & SV_IMMEDIATE_UNREF))
10455 SvREFCNT_dec_NN(target);
10456 else /* XXX Hack, but hard to make $a=$a->[1] work otherwise */
10457 sv_2mortal(target); /* Schedule for freeing later */
10461 =for apidoc sv_untaint
10463 Untaint an SV. Use C<SvTAINTED_off> instead.
10469 Perl_sv_untaint(pTHX_ SV *const sv)
10471 PERL_ARGS_ASSERT_SV_UNTAINT;
10472 PERL_UNUSED_CONTEXT;
10474 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10475 MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10482 =for apidoc sv_tainted
10484 Test an SV for taintedness. Use C<SvTAINTED> instead.
10490 Perl_sv_tainted(pTHX_ SV *const sv)
10492 PERL_ARGS_ASSERT_SV_TAINTED;
10493 PERL_UNUSED_CONTEXT;
10495 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10496 const MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10497 if (mg && (mg->mg_len & 1) )
10503 #ifndef NO_MATHOMS /* Can't move these to mathoms.c because call uiv_2buf(),
10504 private to this file */
10507 =for apidoc sv_setpviv
10509 Copies an integer into the given SV, also updating its string value.
10510 Does not handle 'set' magic. See C<L</sv_setpviv_mg>>.
10516 Perl_sv_setpviv(pTHX_ SV *const sv, const IV iv)
10518 char buf[TYPE_CHARS(UV)];
10520 char * const ptr = uiv_2buf(buf, iv, 0, 0, &ebuf);
10522 PERL_ARGS_ASSERT_SV_SETPVIV;
10524 sv_setpvn(sv, ptr, ebuf - ptr);
10528 =for apidoc sv_setpviv_mg
10530 Like C<sv_setpviv>, but also handles 'set' magic.
10536 Perl_sv_setpviv_mg(pTHX_ SV *const sv, const IV iv)
10538 PERL_ARGS_ASSERT_SV_SETPVIV_MG;
10540 sv_setpviv(sv, iv);
10544 #endif /* NO_MATHOMS */
10546 #if defined(PERL_IMPLICIT_CONTEXT)
10548 /* pTHX_ magic can't cope with varargs, so this is a no-context
10549 * version of the main function, (which may itself be aliased to us).
10550 * Don't access this version directly.
10554 Perl_sv_setpvf_nocontext(SV *const sv, const char *const pat, ...)
10559 PERL_ARGS_ASSERT_SV_SETPVF_NOCONTEXT;
10561 va_start(args, pat);
10562 sv_vsetpvf(sv, pat, &args);
10566 /* pTHX_ magic can't cope with varargs, so this is a no-context
10567 * version of the main function, (which may itself be aliased to us).
10568 * Don't access this version directly.
10572 Perl_sv_setpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10577 PERL_ARGS_ASSERT_SV_SETPVF_MG_NOCONTEXT;
10579 va_start(args, pat);
10580 sv_vsetpvf_mg(sv, pat, &args);
10586 =for apidoc sv_setpvf
10588 Works like C<sv_catpvf> but copies the text into the SV instead of
10589 appending it. Does not handle 'set' magic. See C<L</sv_setpvf_mg>>.
10595 Perl_sv_setpvf(pTHX_ SV *const sv, const char *const pat, ...)
10599 PERL_ARGS_ASSERT_SV_SETPVF;
10601 va_start(args, pat);
10602 sv_vsetpvf(sv, pat, &args);
10607 =for apidoc sv_vsetpvf
10609 Works like C<sv_vcatpvf> but copies the text into the SV instead of
10610 appending it. Does not handle 'set' magic. See C<L</sv_vsetpvf_mg>>.
10612 Usually used via its frontend C<sv_setpvf>.
10618 Perl_sv_vsetpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10620 PERL_ARGS_ASSERT_SV_VSETPVF;
10622 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10626 =for apidoc sv_setpvf_mg
10628 Like C<sv_setpvf>, but also handles 'set' magic.
10634 Perl_sv_setpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10638 PERL_ARGS_ASSERT_SV_SETPVF_MG;
10640 va_start(args, pat);
10641 sv_vsetpvf_mg(sv, pat, &args);
10646 =for apidoc sv_vsetpvf_mg
10648 Like C<sv_vsetpvf>, but also handles 'set' magic.
10650 Usually used via its frontend C<sv_setpvf_mg>.
10656 Perl_sv_vsetpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10658 PERL_ARGS_ASSERT_SV_VSETPVF_MG;
10660 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10664 #if defined(PERL_IMPLICIT_CONTEXT)
10666 /* pTHX_ magic can't cope with varargs, so this is a no-context
10667 * version of the main function, (which may itself be aliased to us).
10668 * Don't access this version directly.
10672 Perl_sv_catpvf_nocontext(SV *const sv, const char *const pat, ...)
10677 PERL_ARGS_ASSERT_SV_CATPVF_NOCONTEXT;
10679 va_start(args, pat);
10680 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10684 /* pTHX_ magic can't cope with varargs, so this is a no-context
10685 * version of the main function, (which may itself be aliased to us).
10686 * Don't access this version directly.
10690 Perl_sv_catpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10695 PERL_ARGS_ASSERT_SV_CATPVF_MG_NOCONTEXT;
10697 va_start(args, pat);
10698 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10705 =for apidoc sv_catpvf
10707 Processes its arguments like C<sv_catpvfn>, and appends the formatted
10708 output to an SV. As with C<sv_catpvfn> called with a non-null C-style
10709 variable argument list, argument reordering is not supported.
10710 If the appended data contains "wide" characters
10711 (including, but not limited to, SVs with a UTF-8 PV formatted with C<%s>,
10712 and characters >255 formatted with C<%c>), the original SV might get
10713 upgraded to UTF-8. Handles 'get' magic, but not 'set' magic. See
10714 C<L</sv_catpvf_mg>>. If the original SV was UTF-8, the pattern should be
10715 valid UTF-8; if the original SV was bytes, the pattern should be too.
10720 Perl_sv_catpvf(pTHX_ SV *const sv, const char *const pat, ...)
10724 PERL_ARGS_ASSERT_SV_CATPVF;
10726 va_start(args, pat);
10727 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10732 =for apidoc sv_vcatpvf
10734 Processes its arguments like C<sv_catpvfn> called with a non-null C-style
10735 variable argument list, and appends the formatted output
10736 to an SV. Does not handle 'set' magic. See C<L</sv_vcatpvf_mg>>.
10738 Usually used via its frontend C<sv_catpvf>.
10744 Perl_sv_vcatpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10746 PERL_ARGS_ASSERT_SV_VCATPVF;
10748 sv_vcatpvfn_flags(sv, pat, strlen(pat), args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10752 =for apidoc sv_catpvf_mg
10754 Like C<sv_catpvf>, but also handles 'set' magic.
10760 Perl_sv_catpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10764 PERL_ARGS_ASSERT_SV_CATPVF_MG;
10766 va_start(args, pat);
10767 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10773 =for apidoc sv_vcatpvf_mg
10775 Like C<sv_vcatpvf>, but also handles 'set' magic.
10777 Usually used via its frontend C<sv_catpvf_mg>.
10783 Perl_sv_vcatpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10785 PERL_ARGS_ASSERT_SV_VCATPVF_MG;
10787 sv_vcatpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10792 =for apidoc sv_vsetpvfn
10794 Works like C<sv_vcatpvfn> but copies the text into the SV instead of
10797 Usually used via one of its frontends C<sv_vsetpvf> and C<sv_vsetpvf_mg>.
10803 Perl_sv_vsetpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10804 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
10806 PERL_ARGS_ASSERT_SV_VSETPVFN;
10809 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, 0);
10814 * Warn of missing argument to sprintf. The value used in place of such
10815 * arguments should be &PL_sv_no; an undefined value would yield
10816 * inappropriate "use of uninit" warnings [perl #71000].
10819 S_warn_vcatpvfn_missing_argument(pTHX) {
10820 if (ckWARN(WARN_MISSING)) {
10821 Perl_warner(aTHX_ packWARN(WARN_MISSING), "Missing argument in %s",
10822 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
10828 S_expect_number(pTHX_ char **const pattern)
10832 PERL_ARGS_ASSERT_EXPECT_NUMBER;
10834 switch (**pattern) {
10835 case '1': case '2': case '3':
10836 case '4': case '5': case '6':
10837 case '7': case '8': case '9':
10838 var = *(*pattern)++ - '0';
10839 while (isDIGIT(**pattern)) {
10840 const I32 tmp = var * 10 + (*(*pattern)++ - '0');
10842 Perl_croak(aTHX_ "Integer overflow in format string for %s", (PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn"));
10850 S_F0convert(NV nv, char *const endbuf, STRLEN *const len)
10852 const int neg = nv < 0;
10855 PERL_ARGS_ASSERT_F0CONVERT;
10857 if (UNLIKELY(Perl_isinfnan(nv))) {
10858 STRLEN n = S_infnan_2pv(nv, endbuf - *len, *len, 0);
10868 if (uv & 1 && uv == nv)
10869 uv--; /* Round to even */
10871 const unsigned dig = uv % 10;
10873 } while (uv /= 10);
10884 =for apidoc sv_vcatpvfn
10886 =for apidoc sv_vcatpvfn_flags
10888 Processes its arguments like C<vsprintf> and appends the formatted output
10889 to an SV. Uses an array of SVs if the C-style variable argument list is
10890 missing (C<NULL>). Argument reordering (using format specifiers like C<%2$d>
10891 or C<%*2$d>) is supported only when using an array of SVs; using a C-style
10892 C<va_list> argument list with a format string that uses argument reordering
10893 will yield an exception.
10895 When running with taint checks enabled, indicates via
10896 C<maybe_tainted> if results are untrustworthy (often due to the use of
10899 If called as C<sv_vcatpvfn> or flags has the C<SV_GMAGIC> bit set, calls get magic.
10901 Usually used via one of its frontends C<sv_vcatpvf> and C<sv_vcatpvf_mg>.
10906 #define VECTORIZE_ARGS vecsv = va_arg(*args, SV*);\
10907 vecstr = (U8*)SvPV_const(vecsv,veclen);\
10908 vec_utf8 = DO_UTF8(vecsv);
10910 /* XXX maybe_tainted is never assigned to, so the doc above is lying. */
10913 Perl_sv_vcatpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10914 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
10916 PERL_ARGS_ASSERT_SV_VCATPVFN;
10918 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, SV_GMAGIC|SV_SMAGIC);
10921 #ifdef LONGDOUBLE_DOUBLEDOUBLE
10922 /* The first double can be as large as 2**1023, or '1' x '0' x 1023.
10923 * The second double can be as small as 2**-1074, or '0' x 1073 . '1'.
10924 * The sum of them can be '1' . '0' x 2096 . '1', with implied radix point
10925 * after the first 1023 zero bits.
10927 * XXX The 2098 is quite large (262.25 bytes) and therefore some sort
10928 * of dynamically growing buffer might be better, start at just 16 bytes
10929 * (for example) and grow only when necessary. Or maybe just by looking
10930 * at the exponents of the two doubles? */
10931 # define DOUBLEDOUBLE_MAXBITS 2098
10934 /* vhex will contain the values (0..15) of the hex digits ("nybbles"
10935 * of 4 bits); 1 for the implicit 1, and the mantissa bits, four bits
10936 * per xdigit. For the double-double case, this can be rather many.
10937 * The non-double-double-long-double overshoots since all bits of NV
10938 * are not mantissa bits, there are also exponent bits. */
10939 #ifdef LONGDOUBLE_DOUBLEDOUBLE
10940 # define VHEX_SIZE (3+DOUBLEDOUBLE_MAXBITS/4)
10942 # define VHEX_SIZE (1+(NVSIZE * 8)/4)
10945 /* If we do not have a known long double format, (including not using
10946 * long doubles, or long doubles being equal to doubles) then we will
10947 * fall back to the ldexp/frexp route, with which we can retrieve at
10948 * most as many bits as our widest unsigned integer type is. We try
10949 * to get a 64-bit unsigned integer even if we are not using a 64-bit UV.
10951 * (If you want to test the case of UVSIZE == 4, NVSIZE == 8,
10952 * set the MANTISSATYPE to int and the MANTISSASIZE to 4.)
10954 #if defined(HAS_QUAD) && defined(Uquad_t)
10955 # define MANTISSATYPE Uquad_t
10956 # define MANTISSASIZE 8
10958 # define MANTISSATYPE UV
10959 # define MANTISSASIZE UVSIZE
10962 #if defined(DOUBLE_LITTLE_ENDIAN) || defined(LONGDOUBLE_LITTLE_ENDIAN)
10963 # define HEXTRACT_LITTLE_ENDIAN
10964 #elif defined(DOUBLE_BIG_ENDIAN) || defined(LONGDOUBLE_BIG_ENDIAN)
10965 # define HEXTRACT_BIG_ENDIAN
10967 # define HEXTRACT_MIX_ENDIAN
10970 /* S_hextract() is a helper for Perl_sv_vcatpvfn_flags, for extracting
10971 * the hexadecimal values (for %a/%A). The nv is the NV where the value
10972 * are being extracted from (either directly from the long double in-memory
10973 * presentation, or from the uquad computed via frexp+ldexp). frexp also
10974 * is used to update the exponent. The subnormal is set to true
10975 * for IEEE 754 subnormals/denormals (including the x86 80-bit format).
10976 * The vhex is the pointer to the beginning of the output buffer of VHEX_SIZE.
10978 * The tricky part is that S_hextract() needs to be called twice:
10979 * the first time with vend as NULL, and the second time with vend as
10980 * the pointer returned by the first call. What happens is that on
10981 * the first round the output size is computed, and the intended
10982 * extraction sanity checked. On the second round the actual output
10983 * (the extraction of the hexadecimal values) takes place.
10984 * Sanity failures cause fatal failures during both rounds. */
10986 S_hextract(pTHX_ const NV nv, int* exponent, bool *subnormal,
10987 U8* vhex, U8* vend)
10991 int ixmin = 0, ixmax = 0;
10993 /* XXX Inf/NaN are not handled here, since it is
10994 * assumed they are to be output as "Inf" and "NaN". */
10996 /* These macros are just to reduce typos, they have multiple
10997 * repetitions below, but usually only one (or sometimes two)
10998 * of them is really being used. */
10999 /* HEXTRACT_OUTPUT() extracts the high nybble first. */
11000 #define HEXTRACT_OUTPUT_HI(ix) (*v++ = nvp[ix] >> 4)
11001 #define HEXTRACT_OUTPUT_LO(ix) (*v++ = nvp[ix] & 0xF)
11002 #define HEXTRACT_OUTPUT(ix) \
11004 HEXTRACT_OUTPUT_HI(ix); HEXTRACT_OUTPUT_LO(ix); \
11006 #define HEXTRACT_COUNT(ix, c) \
11008 v += c; if (ix < ixmin) ixmin = ix; else if (ix > ixmax) ixmax = ix; \
11010 #define HEXTRACT_BYTE(ix) \
11012 if (vend) HEXTRACT_OUTPUT(ix); else HEXTRACT_COUNT(ix, 2); \
11014 #define HEXTRACT_LO_NYBBLE(ix) \
11016 if (vend) HEXTRACT_OUTPUT_LO(ix); else HEXTRACT_COUNT(ix, 1); \
11018 /* HEXTRACT_TOP_NYBBLE is just convenience disguise,
11019 * to make it look less odd when the top bits of a NV
11020 * are extracted using HEXTRACT_LO_NYBBLE: the highest
11021 * order bits can be in the "low nybble" of a byte. */
11022 #define HEXTRACT_TOP_NYBBLE(ix) HEXTRACT_LO_NYBBLE(ix)
11023 #define HEXTRACT_BYTES_LE(a, b) \
11024 for (ix = a; ix >= b; ix--) { HEXTRACT_BYTE(ix); }
11025 #define HEXTRACT_BYTES_BE(a, b) \
11026 for (ix = a; ix <= b; ix++) { HEXTRACT_BYTE(ix); }
11027 #define HEXTRACT_GET_SUBNORMAL(nv) *subnormal = Perl_fp_class_denorm(nv)
11028 #define HEXTRACT_IMPLICIT_BIT(nv) \
11030 if (!*subnormal) { \
11031 if (vend) *v++ = ((nv) == 0.0) ? 0 : 1; else v++; \
11035 /* Most formats do. Those which don't should undef this.
11037 * But also note that IEEE 754 subnormals do not have it, or,
11038 * expressed alternatively, their implicit bit is zero. */
11039 #define HEXTRACT_HAS_IMPLICIT_BIT
11041 /* Many formats do. Those which don't should undef this. */
11042 #define HEXTRACT_HAS_TOP_NYBBLE
11044 /* HEXTRACTSIZE is the maximum number of xdigits. */
11045 #if defined(USE_LONG_DOUBLE) && defined(LONGDOUBLE_DOUBLEDOUBLE)
11046 # define HEXTRACTSIZE (2+DOUBLEDOUBLE_MAXBITS/4)
11048 # define HEXTRACTSIZE 2 * NVSIZE
11051 const U8* vmaxend = vhex + HEXTRACTSIZE;
11052 PERL_UNUSED_VAR(ix); /* might happen */
11053 (void)Perl_frexp(PERL_ABS(nv), exponent);
11054 *subnormal = FALSE;
11055 if (vend && (vend <= vhex || vend > vmaxend)) {
11056 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11057 Perl_croak(aTHX_ "Hexadecimal float: internal error (entry)");
11060 /* First check if using long doubles. */
11061 #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE)
11062 # if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN
11063 /* Used in e.g. VMS and HP-UX IA-64, e.g. -0.1L:
11064 * 9a 99 99 99 99 99 99 99 99 99 99 99 99 99 fb bf */
11065 /* The bytes 13..0 are the mantissa/fraction,
11066 * the 15,14 are the sign+exponent. */
11067 const U8* nvp = (const U8*)(&nv);
11068 HEXTRACT_GET_SUBNORMAL(nv);
11069 HEXTRACT_IMPLICIT_BIT(nv);
11070 # undef HEXTRACT_HAS_TOP_NYBBLE
11071 HEXTRACT_BYTES_LE(13, 0);
11072 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_BIG_ENDIAN
11073 /* Used in e.g. Solaris Sparc and HP-UX PA-RISC, e.g. -0.1L:
11074 * bf fb 99 99 99 99 99 99 99 99 99 99 99 99 99 9a */
11075 /* The bytes 2..15 are the mantissa/fraction,
11076 * the 0,1 are the sign+exponent. */
11077 const U8* nvp = (const U8*)(&nv);
11078 HEXTRACT_GET_SUBNORMAL(nv);
11079 HEXTRACT_IMPLICIT_BIT(nv);
11080 # undef HEXTRACT_HAS_TOP_NYBBLE
11081 HEXTRACT_BYTES_BE(2, 15);
11082 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN
11083 /* x86 80-bit "extended precision", 64 bits of mantissa / fraction /
11084 * significand, 15 bits of exponent, 1 bit of sign. No implicit bit.
11085 * NVSIZE can be either 12 (ILP32, Solaris x86) or 16 (LP64, Linux
11086 * and OS X), meaning that 2 or 6 bytes are empty padding. */
11087 /* The bytes 0..1 are the sign+exponent,
11088 * the bytes 2..9 are the mantissa/fraction. */
11089 const U8* nvp = (const U8*)(&nv);
11090 # undef HEXTRACT_HAS_IMPLICIT_BIT
11091 # undef HEXTRACT_HAS_TOP_NYBBLE
11092 HEXTRACT_GET_SUBNORMAL(nv);
11093 HEXTRACT_BYTES_LE(7, 0);
11094 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN
11095 /* Does this format ever happen? (Wikipedia says the Motorola
11096 * 6888x math coprocessors used format _like_ this but padded
11097 * to 96 bits with 16 unused bits between the exponent and the
11099 const U8* nvp = (const U8*)(&nv);
11100 # undef HEXTRACT_HAS_IMPLICIT_BIT
11101 # undef HEXTRACT_HAS_TOP_NYBBLE
11102 HEXTRACT_GET_SUBNORMAL(nv);
11103 HEXTRACT_BYTES_BE(0, 7);
11105 # define HEXTRACT_FALLBACK
11106 /* Double-double format: two doubles next to each other.
11107 * The first double is the high-order one, exactly like
11108 * it would be for a "lone" double. The second double
11109 * is shifted down using the exponent so that that there
11110 * are no common bits. The tricky part is that the value
11111 * of the double-double is the SUM of the two doubles and
11112 * the second one can be also NEGATIVE.
11114 * Because of this tricky construction the bytewise extraction we
11115 * use for the other long double formats doesn't work, we must
11116 * extract the values bit by bit.
11118 * The little-endian double-double is used .. somewhere?
11120 * The big endian double-double is used in e.g. PPC/Power (AIX)
11123 * The mantissa bits are in two separate stretches, e.g. for -0.1L:
11124 * 9a 99 99 99 99 99 59 bc 9a 99 99 99 99 99 b9 3f (LE)
11125 * 3f b9 99 99 99 99 99 9a bc 59 99 99 99 99 99 9a (BE)
11128 #else /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) */
11129 /* Using normal doubles, not long doubles.
11131 * We generate 4-bit xdigits (nybble/nibble) instead of 8-bit
11132 * bytes, since we might need to handle printf precision, and
11133 * also need to insert the radix. */
11135 # ifdef HEXTRACT_LITTLE_ENDIAN
11136 /* 0 1 2 3 4 5 6 7 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11137 const U8* nvp = (const U8*)(&nv);
11138 HEXTRACT_GET_SUBNORMAL(nv);
11139 HEXTRACT_IMPLICIT_BIT(nv);
11140 HEXTRACT_TOP_NYBBLE(6);
11141 HEXTRACT_BYTES_LE(5, 0);
11142 # elif defined(HEXTRACT_BIG_ENDIAN)
11143 /* 7 6 5 4 3 2 1 0 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11144 const U8* nvp = (const U8*)(&nv);
11145 HEXTRACT_GET_SUBNORMAL(nv);
11146 HEXTRACT_IMPLICIT_BIT(nv);
11147 HEXTRACT_TOP_NYBBLE(1);
11148 HEXTRACT_BYTES_BE(2, 7);
11149 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_LE_BE
11150 /* 4 5 6 7 0 1 2 3 (MSB = 7, LSB = 0, 6:7 = nybble:exponent:sign) */
11151 const U8* nvp = (const U8*)(&nv);
11152 HEXTRACT_GET_SUBNORMAL(nv);
11153 HEXTRACT_IMPLICIT_BIT(nv);
11154 HEXTRACT_TOP_NYBBLE(2); /* 6 */
11155 HEXTRACT_BYTE(1); /* 5 */
11156 HEXTRACT_BYTE(0); /* 4 */
11157 HEXTRACT_BYTE(7); /* 3 */
11158 HEXTRACT_BYTE(6); /* 2 */
11159 HEXTRACT_BYTE(5); /* 1 */
11160 HEXTRACT_BYTE(4); /* 0 */
11161 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_BE_LE
11162 /* 3 2 1 0 7 6 5 4 (MSB = 7, LSB = 0, 7:6 = sign:exponent:nybble) */
11163 const U8* nvp = (const U8*)(&nv);
11164 HEXTRACT_GET_SUBNORMAL(nv);
11165 HEXTRACT_IMPLICIT_BIT(nv);
11166 HEXTRACT_TOP_NYBBLE(5); /* 6 */
11167 HEXTRACT_BYTE(6); /* 5 */
11168 HEXTRACT_BYTE(7); /* 4 */
11169 HEXTRACT_BYTE(0); /* 3 */
11170 HEXTRACT_BYTE(1); /* 2 */
11171 HEXTRACT_BYTE(2); /* 1 */
11172 HEXTRACT_BYTE(3); /* 0 */
11174 # define HEXTRACT_FALLBACK
11177 # define HEXTRACT_FALLBACK
11179 #endif /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) #else */
11180 # ifdef HEXTRACT_FALLBACK
11181 HEXTRACT_GET_SUBNORMAL(nv);
11182 # undef HEXTRACT_HAS_TOP_NYBBLE /* Meaningless, but consistent. */
11183 /* The fallback is used for the double-double format, and
11184 * for unknown long double formats, and for unknown double
11185 * formats, or in general unknown NV formats. */
11186 if (nv == (NV)0.0) {
11194 NV d = nv < 0 ? -nv : nv;
11196 U8 ha = 0x0; /* hexvalue accumulator */
11197 U8 hd = 0x8; /* hexvalue digit */
11199 /* Shift d and e (and update exponent) so that e <= d < 2*e,
11200 * this is essentially manual frexp(). Multiplying by 0.5 and
11201 * doubling should be lossless in binary floating point. */
11211 while (d >= e + e) {
11215 /* Now e <= d < 2*e */
11217 /* First extract the leading hexdigit (the implicit bit). */
11233 /* Then extract the remaining hexdigits. */
11234 while (d > (NV)0.0) {
11240 /* Output or count in groups of four bits,
11241 * that is, when the hexdigit is down to one. */
11246 /* Reset the hexvalue. */
11255 /* Flush possible pending hexvalue. */
11265 /* Croak for various reasons: if the output pointer escaped the
11266 * output buffer, if the extraction index escaped the extraction
11267 * buffer, or if the ending output pointer didn't match the
11268 * previously computed value. */
11269 if (v <= vhex || v - vhex >= VHEX_SIZE ||
11270 /* For double-double the ixmin and ixmax stay at zero,
11271 * which is convenient since the HEXTRACTSIZE is tricky
11272 * for double-double. */
11273 ixmin < 0 || ixmax >= NVSIZE ||
11274 (vend && v != vend)) {
11275 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11276 Perl_croak(aTHX_ "Hexadecimal float: internal error (overflow)");
11281 /* Helper for sv_vcatpvfn_flags(). */
11282 #define FETCH_VCATPVFN_ARGUMENT(var, in_range, expr) \
11287 (var) = &PL_sv_no; /* [perl #71000] */ \
11288 arg_missing = TRUE; \
11293 Perl_sv_vcatpvfn_flags(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11294 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted,
11299 const char *patend;
11302 static const char nullstr[] = "(null)";
11304 bool has_utf8 = DO_UTF8(sv); /* has the result utf8? */
11305 const bool pat_utf8 = has_utf8; /* the pattern is in utf8? */
11307 /* Times 4: a decimal digit takes more than 3 binary digits.
11308 * NV_DIG: mantissa takes than many decimal digits.
11309 * Plus 32: Playing safe. */
11310 char ebuf[IV_DIG * 4 + NV_DIG + 32];
11311 bool no_redundant_warning = FALSE; /* did we use any explicit format parameter index? */
11312 bool hexfp = FALSE; /* hexadecimal floating point? */
11314 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
11316 PERL_ARGS_ASSERT_SV_VCATPVFN_FLAGS;
11317 PERL_UNUSED_ARG(maybe_tainted);
11319 if (flags & SV_GMAGIC)
11322 /* no matter what, this is a string now */
11323 (void)SvPV_force_nomg(sv, origlen);
11325 /* special-case "", "%s", and "%-p" (SVf - see below) */
11327 if (svmax && ckWARN(WARN_REDUNDANT))
11328 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11329 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11332 if (patlen == 2 && pat[0] == '%' && pat[1] == 's') {
11333 if (svmax > 1 && ckWARN(WARN_REDUNDANT))
11334 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11335 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11338 const char * const s = va_arg(*args, char*);
11339 sv_catpv_nomg(sv, s ? s : nullstr);
11341 else if (svix < svmax) {
11342 /* we want get magic on the source but not the target. sv_catsv can't do that, though */
11343 SvGETMAGIC(*svargs);
11344 sv_catsv_nomg(sv, *svargs);
11347 S_warn_vcatpvfn_missing_argument(aTHX);
11350 if (args && patlen == 3 && pat[0] == '%' &&
11351 pat[1] == '-' && pat[2] == 'p') {
11352 if (svmax > 1 && ckWARN(WARN_REDUNDANT))
11353 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11354 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11355 argsv = MUTABLE_SV(va_arg(*args, void*));
11356 sv_catsv_nomg(sv, argsv);
11360 #if !defined(USE_LONG_DOUBLE) && !defined(USE_QUADMATH)
11361 /* special-case "%.<number>[gf]" */
11362 if ( !args && patlen <= 5 && pat[0] == '%' && pat[1] == '.'
11363 && (pat[patlen-1] == 'g' || pat[patlen-1] == 'f') ) {
11364 unsigned digits = 0;
11368 while (*pp >= '0' && *pp <= '9')
11369 digits = 10 * digits + (*pp++ - '0');
11371 /* XXX: Why do this `svix < svmax` test? Couldn't we just
11372 format the first argument and WARN_REDUNDANT if svmax > 1?
11373 Munged by Nicholas Clark in v5.13.0-209-g95ea86d */
11374 if (pp - pat == (int)patlen - 1 && svix < svmax) {
11375 const NV nv = SvNV(*svargs);
11376 if (LIKELY(!Perl_isinfnan(nv))) {
11378 /* Add check for digits != 0 because it seems that some
11379 gconverts are buggy in this case, and we don't yet have
11380 a Configure test for this. */
11381 if (digits && digits < sizeof(ebuf) - NV_DIG - 10) {
11382 /* 0, point, slack */
11383 STORE_LC_NUMERIC_SET_TO_NEEDED();
11384 SNPRINTF_G(nv, ebuf, size, digits);
11385 sv_catpv_nomg(sv, ebuf);
11386 if (*ebuf) /* May return an empty string for digits==0 */
11389 } else if (!digits) {
11392 if ((p = F0convert(nv, ebuf + sizeof ebuf, &l))) {
11393 sv_catpvn_nomg(sv, p, l);
11400 #endif /* !USE_LONG_DOUBLE */
11402 if (!args && svix < svmax && DO_UTF8(*svargs))
11405 patend = (char*)pat + patlen;
11406 for (p = (char*)pat; p < patend; p = q) {
11409 bool vectorize = FALSE;
11410 bool vectorarg = FALSE;
11411 bool vec_utf8 = FALSE;
11417 bool has_precis = FALSE;
11419 const I32 osvix = svix;
11420 bool is_utf8 = FALSE; /* is this item utf8? */
11421 bool used_explicit_ix = FALSE;
11422 bool arg_missing = FALSE;
11423 #ifdef HAS_LDBL_SPRINTF_BUG
11424 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
11425 with sfio - Allen <allens@cpan.org> */
11426 bool fix_ldbl_sprintf_bug = FALSE;
11430 U8 utf8buf[UTF8_MAXBYTES+1];
11431 STRLEN esignlen = 0;
11433 const char *eptr = NULL;
11434 const char *fmtstart;
11437 const U8 *vecstr = NULL;
11444 /* We need a long double target in case HAS_LONG_DOUBLE,
11445 * even without USE_LONG_DOUBLE, so that we can printf with
11446 * long double formats, even without NV being long double.
11447 * But we call the target 'fv' instead of 'nv', since most of
11448 * the time it is not (most compilers these days recognize
11449 * "long double", even if only as a synonym for "double").
11451 #if defined(HAS_LONG_DOUBLE) && LONG_DOUBLESIZE > DOUBLESIZE && \
11452 defined(PERL_PRIgldbl) && !defined(USE_QUADMATH)
11454 # ifdef Perl_isfinitel
11455 # define FV_ISFINITE(x) Perl_isfinitel(x)
11457 # define FV_GF PERL_PRIgldbl
11458 # if defined(__VMS) && defined(__ia64) && defined(__IEEE_FLOAT)
11459 /* Work around breakage in OTS$CVT_FLOAT_T_X */
11460 # define NV_TO_FV(nv,fv) STMT_START { \
11462 fv = Perl_isnan(_dv) ? LDBL_QNAN : _dv; \
11465 # define NV_TO_FV(nv,fv) (fv)=(nv)
11469 # define FV_GF NVgf
11470 # define NV_TO_FV(nv,fv) (fv)=(nv)
11472 #ifndef FV_ISFINITE
11473 # define FV_ISFINITE(x) Perl_isfinite((NV)(x))
11479 const char *dotstr = ".";
11480 STRLEN dotstrlen = 1;
11481 I32 efix = 0; /* explicit format parameter index */
11482 I32 ewix = 0; /* explicit width index */
11483 I32 epix = 0; /* explicit precision index */
11484 I32 evix = 0; /* explicit vector index */
11485 bool asterisk = FALSE;
11486 bool infnan = FALSE;
11488 /* echo everything up to the next format specification */
11489 for (q = p; q < patend && *q != '%'; ++q) ;
11491 if (has_utf8 && !pat_utf8)
11492 sv_catpvn_nomg_utf8_upgrade(sv, p, q - p, nsv);
11494 sv_catpvn_nomg(sv, p, q - p);
11503 We allow format specification elements in this order:
11504 \d+\$ explicit format parameter index
11506 v|\*(\d+\$)?v vector with optional (optionally specified) arg
11507 0 flag (as above): repeated to allow "v02"
11508 \d+|\*(\d+\$)? width using optional (optionally specified) arg
11509 \.(\d*|\*(\d+\$)?) precision using optional (optionally specified) arg
11511 [%bcdefginopsuxDFOUX] format (mandatory)
11516 As of perl5.9.3, printf format checking is on by default.
11517 Internally, perl uses %p formats to provide an escape to
11518 some extended formatting. This block deals with those
11519 extensions: if it does not match, (char*)q is reset and
11520 the normal format processing code is used.
11522 Currently defined extensions are:
11523 %p include pointer address (standard)
11524 %-p (SVf) include an SV (previously %_)
11525 %-<num>p include an SV with precision <num>
11527 %3p include a HEK with precision of 256
11528 %4p char* preceded by utf8 flag and length
11529 %<num>p (where num is 1 or > 4) reserved for future
11532 Robin Barker 2005-07-14 (but modified since)
11534 %1p (VDf) removed. RMB 2007-10-19
11541 else if (strnEQ(q, UTF8f, sizeof(UTF8f)-1)) { /* UTF8f */
11542 /* The argument has already gone through cBOOL, so the cast
11544 is_utf8 = (bool)va_arg(*args, int);
11545 elen = va_arg(*args, UV);
11546 /* if utf8 length is larger than 0x7ffff..., then it might
11547 * have been a signed value that wrapped */
11548 if (elen > ((~(STRLEN)0) >> 1)) {
11549 assert(0); /* in DEBUGGING build we want to crash */
11550 elen= 0; /* otherwise we want to treat this as an empty string */
11552 eptr = va_arg(*args, char *);
11553 q += sizeof(UTF8f)-1;
11556 n = expect_number(&q);
11558 if (sv) { /* SVf */
11563 argsv = MUTABLE_SV(va_arg(*args, void*));
11564 eptr = SvPV_const(argsv, elen);
11565 if (DO_UTF8(argsv))
11569 else if (n==2 || n==3) { /* HEKf */
11570 HEK * const hek = va_arg(*args, HEK *);
11571 eptr = HEK_KEY(hek);
11572 elen = HEK_LEN(hek);
11573 if (HEK_UTF8(hek)) is_utf8 = TRUE;
11574 if (n==3) precis = 256, has_precis = TRUE;
11578 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
11579 "internal %%<num>p might conflict with future printf extensions");
11585 if ( (width = expect_number(&q)) ) {
11588 Perl_croak_nocontext(
11589 "Cannot yet reorder sv_catpvfn() arguments from va_list");
11592 used_explicit_ix = TRUE;
11604 if (plus == '+' && *q == ' ') /* '+' over ' ' */
11633 if ( (ewix = expect_number(&q)) ) {
11636 Perl_croak_nocontext(
11637 "Cannot yet reorder sv_catpvfn() arguments from va_list");
11638 used_explicit_ix = TRUE;
11648 if ((vectorarg = asterisk)) {
11661 width = expect_number(&q);
11664 if (vectorize && vectorarg) {
11665 /* vectorizing, but not with the default "." */
11667 vecsv = va_arg(*args, SV*);
11669 FETCH_VCATPVFN_ARGUMENT(
11670 vecsv, evix > 0 && evix <= svmax, svargs[evix-1]);
11672 FETCH_VCATPVFN_ARGUMENT(
11673 vecsv, svix < svmax, svargs[svix++]);
11675 dotstr = SvPV_const(vecsv, dotstrlen);
11676 /* Keep the DO_UTF8 test *after* the SvPV call, else things go
11677 bad with tied or overloaded values that return UTF8. */
11678 if (DO_UTF8(vecsv))
11680 else if (has_utf8) {
11681 vecsv = sv_mortalcopy(vecsv);
11682 sv_utf8_upgrade(vecsv);
11683 dotstr = SvPV_const(vecsv, dotstrlen);
11690 i = va_arg(*args, int);
11692 i = (ewix ? ewix <= svmax : svix < svmax) ?
11693 SvIVx(svargs[ewix ? ewix-1 : svix++]) : 0;
11695 width = (i < 0) ? -i : i;
11705 if ( (epix = expect_number(&q)) ) {
11708 Perl_croak_nocontext(
11709 "Cannot yet reorder sv_catpvfn() arguments from va_list");
11710 used_explicit_ix = TRUE;
11715 i = va_arg(*args, int);
11719 FETCH_VCATPVFN_ARGUMENT(
11720 precsv, epix > 0 && epix <= svmax, svargs[epix-1]);
11722 FETCH_VCATPVFN_ARGUMENT(
11723 precsv, svix < svmax, svargs[svix++]);
11724 i = precsv == &PL_sv_no ? 0 : SvIVx(precsv);
11727 has_precis = !(i < 0);
11731 while (isDIGIT(*q))
11732 precis = precis * 10 + (*q++ - '0');
11741 else if (efix ? (efix > 0 && efix <= svmax) : svix < svmax) {
11742 vecsv = svargs[efix ? efix-1 : svix++];
11743 vecstr = (U8*)SvPV_const(vecsv,veclen);
11744 vec_utf8 = DO_UTF8(vecsv);
11746 /* if this is a version object, we need to convert
11747 * back into v-string notation and then let the
11748 * vectorize happen normally
11750 if (sv_isobject(vecsv) && sv_derived_from(vecsv, "version")) {
11751 if ( hv_exists(MUTABLE_HV(SvRV(vecsv)), "alpha", 5 ) ) {
11752 Perl_ck_warner_d(aTHX_ packWARN(WARN_PRINTF),
11753 "vector argument not supported with alpha versions");
11756 vecsv = sv_newmortal();
11757 scan_vstring((char *)vecstr, (char *)vecstr + veclen,
11759 vecstr = (U8*)SvPV_const(vecsv, veclen);
11760 vec_utf8 = DO_UTF8(vecsv);
11774 case 'I': /* Ix, I32x, and I64x */
11775 # ifdef USE_64_BIT_INT
11776 if (q[1] == '6' && q[2] == '4') {
11782 if (q[1] == '3' && q[2] == '2') {
11786 # ifdef USE_64_BIT_INT
11792 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
11793 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
11796 # ifdef USE_QUADMATH
11809 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
11810 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
11811 if (*q == 'l') { /* lld, llf */
11820 if (*++q == 'h') { /* hhd, hhu */
11849 if (!vectorize && !args) {
11851 const I32 i = efix-1;
11852 FETCH_VCATPVFN_ARGUMENT(argsv, i >= 0 && i < svmax, svargs[i]);
11854 FETCH_VCATPVFN_ARGUMENT(argsv, svix >= 0 && svix < svmax,
11859 if (argsv && strchr("BbcDdiOopuUXx",*q)) {
11860 /* XXX va_arg(*args) case? need peek, use va_copy? */
11862 if (UNLIKELY(SvAMAGIC(argsv)))
11863 argsv = sv_2num(argsv);
11864 infnan = UNLIKELY(isinfnansv(argsv));
11867 switch (c = *q++) {
11875 Perl_croak(aTHX_ "Cannot printf %"NVgf" with '%c'",
11876 /* no va_arg() case */
11877 SvNV_nomg(argsv), (int)c);
11878 uv = (args) ? va_arg(*args, int) : SvIV_nomg(argsv);
11880 (!UVCHR_IS_INVARIANT(uv) && SvUTF8(sv)))
11882 eptr = (char*)utf8buf;
11883 elen = uvchr_to_utf8((U8*)eptr, uv) - utf8buf;
11897 eptr = va_arg(*args, char*);
11899 elen = strlen(eptr);
11901 eptr = (char *)nullstr;
11902 elen = sizeof nullstr - 1;
11906 eptr = SvPV_const(argsv, elen);
11907 if (DO_UTF8(argsv)) {
11908 STRLEN old_precis = precis;
11909 if (has_precis && precis < elen) {
11910 STRLEN ulen = sv_or_pv_len_utf8(argsv, eptr, elen);
11911 STRLEN p = precis > ulen ? ulen : precis;
11912 precis = sv_or_pv_pos_u2b(argsv, eptr, p, 0);
11913 /* sticks at end */
11915 if (width) { /* fudge width (can't fudge elen) */
11916 if (has_precis && precis < elen)
11917 width += precis - old_precis;
11920 elen - sv_or_pv_len_utf8(argsv,eptr,elen);
11927 if (has_precis && precis < elen)
11935 goto floating_point;
11937 if (alt || vectorize)
11939 uv = PTR2UV(args ? va_arg(*args, void*) : argsv);
11953 goto floating_point;
11958 goto donevalidconversion;
11960 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
11969 esignbuf[esignlen++] = plus;
11973 case 'c': iv = (char)va_arg(*args, int); break;
11974 case 'h': iv = (short)va_arg(*args, int); break;
11975 case 'l': iv = va_arg(*args, long); break;
11976 case 'V': iv = va_arg(*args, IV); break;
11977 case 'z': iv = va_arg(*args, SSize_t); break;
11978 #ifdef HAS_PTRDIFF_T
11979 case 't': iv = va_arg(*args, ptrdiff_t); break;
11981 default: iv = va_arg(*args, int); break;
11983 case 'j': iv = va_arg(*args, intmax_t); break;
11987 iv = va_arg(*args, Quad_t); break;
11994 IV tiv = SvIV_nomg(argsv); /* work around GCC bug #13488 */
11996 case 'c': iv = (char)tiv; break;
11997 case 'h': iv = (short)tiv; break;
11998 case 'l': iv = (long)tiv; break;
12000 default: iv = tiv; break;
12003 iv = (Quad_t)tiv; break;
12009 if ( !vectorize ) /* we already set uv above */
12014 esignbuf[esignlen++] = plus;
12017 uv = (iv == IV_MIN) ? (UV)iv : (UV)(-iv);
12018 esignbuf[esignlen++] = '-';
12057 goto floating_point;
12063 goto donevalidconversion;
12065 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
12076 case 'c': uv = (unsigned char)va_arg(*args, unsigned); break;
12077 case 'h': uv = (unsigned short)va_arg(*args, unsigned); break;
12078 case 'l': uv = va_arg(*args, unsigned long); break;
12079 case 'V': uv = va_arg(*args, UV); break;
12080 case 'z': uv = va_arg(*args, Size_t); break;
12081 #ifdef HAS_PTRDIFF_T
12082 case 't': uv = va_arg(*args, ptrdiff_t); break; /* will sign extend, but there is no uptrdiff_t, so oh well */
12085 case 'j': uv = va_arg(*args, uintmax_t); break;
12087 default: uv = va_arg(*args, unsigned); break;
12090 uv = va_arg(*args, Uquad_t); break;
12097 UV tuv = SvUV_nomg(argsv); /* work around GCC bug #13488 */
12099 case 'c': uv = (unsigned char)tuv; break;
12100 case 'h': uv = (unsigned short)tuv; break;
12101 case 'l': uv = (unsigned long)tuv; break;
12103 default: uv = tuv; break;
12106 uv = (Uquad_t)tuv; break;
12115 char *ptr = ebuf + sizeof ebuf;
12116 bool tempalt = uv ? alt : FALSE; /* Vectors can't change alt */
12122 p = (char *)((c == 'X') ? PL_hexdigit + 16 : PL_hexdigit);
12126 } while (uv >>= 4);
12128 esignbuf[esignlen++] = '0';
12129 esignbuf[esignlen++] = c; /* 'x' or 'X' */
12135 *--ptr = '0' + dig;
12136 } while (uv >>= 3);
12137 if (alt && *ptr != '0')
12143 *--ptr = '0' + dig;
12144 } while (uv >>= 1);
12146 esignbuf[esignlen++] = '0';
12147 esignbuf[esignlen++] = c;
12150 default: /* it had better be ten or less */
12153 *--ptr = '0' + dig;
12154 } while (uv /= base);
12157 elen = (ebuf + sizeof ebuf) - ptr;
12161 zeros = precis - elen;
12162 else if (precis == 0 && elen == 1 && *eptr == '0'
12163 && !(base == 8 && alt)) /* "%#.0o" prints "0" */
12166 /* a precision nullifies the 0 flag. */
12173 /* FLOATING POINT */
12178 c = 'f'; /* maybe %F isn't supported here */
12180 case 'e': case 'E':
12182 case 'g': case 'G':
12183 case 'a': case 'A':
12187 /* This is evil, but floating point is even more evil */
12189 /* for SV-style calling, we can only get NV
12190 for C-style calling, we assume %f is double;
12191 for simplicity we allow any of %Lf, %llf, %qf for long double
12195 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12199 /* [perl #20339] - we should accept and ignore %lf rather than die */
12203 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12204 intsize = args ? 0 : 'q';
12208 #if defined(HAS_LONG_DOUBLE)
12221 /* Now we need (long double) if intsize == 'q', else (double). */
12223 /* Note: do not pull NVs off the va_list with va_arg()
12224 * (pull doubles instead) because if you have a build
12225 * with long doubles, you would always be pulling long
12226 * doubles, which would badly break anyone using only
12227 * doubles (i.e. the majority of builds). In other
12228 * words, you cannot mix doubles and long doubles.
12229 * The only case where you can pull off long doubles
12230 * is when the format specifier explicitly asks so with
12232 #ifdef USE_QUADMATH
12233 fv = intsize == 'q' ?
12234 va_arg(*args, NV) : va_arg(*args, double);
12236 #elif LONG_DOUBLESIZE > DOUBLESIZE
12237 if (intsize == 'q') {
12238 fv = va_arg(*args, long double);
12241 nv = va_arg(*args, double);
12245 nv = va_arg(*args, double);
12251 if (!infnan) SvGETMAGIC(argsv);
12252 nv = SvNV_nomg(argsv);
12257 /* frexp() (or frexpl) has some unspecified behaviour for
12258 * nan/inf/-inf, so let's avoid calling that on non-finites. */
12259 if (isALPHA_FOLD_NE(c, 'e') && FV_ISFINITE(fv)) {
12261 (void)Perl_frexp((NV)fv, &i);
12262 if (i == PERL_INT_MIN)
12263 Perl_die(aTHX_ "panic: frexp: %"FV_GF, fv);
12264 /* Do not set hexfp earlier since we want to printf
12265 * Inf/NaN for Inf/NaN, not their hexfp. */
12266 hexfp = isALPHA_FOLD_EQ(c, 'a');
12267 if (UNLIKELY(hexfp)) {
12268 /* This seriously overshoots in most cases, but
12269 * better the undershooting. Firstly, all bytes
12270 * of the NV are not mantissa, some of them are
12271 * exponent. Secondly, for the reasonably common
12272 * long doubles case, the "80-bit extended", two
12273 * or six bytes of the NV are unused. */
12275 (fv < 0) ? 1 : 0 + /* possible unary minus */
12277 1 + /* the very unlikely carry */
12280 2 * NVSIZE + /* 2 hexdigits for each byte */
12282 6 + /* exponent: sign, plus up to 16383 (quad fp) */
12284 #ifdef LONGDOUBLE_DOUBLEDOUBLE
12285 /* However, for the "double double", we need more.
12286 * Since each double has their own exponent, the
12287 * doubles may float (haha) rather far from each
12288 * other, and the number of required bits is much
12289 * larger, up to total of DOUBLEDOUBLE_MAXBITS bits.
12290 * See the definition of DOUBLEDOUBLE_MAXBITS.
12292 * Need 2 hexdigits for each byte. */
12293 need += (DOUBLEDOUBLE_MAXBITS/8 + 1) * 2;
12294 /* the size for the exponent already added */
12296 #ifdef USE_LOCALE_NUMERIC
12297 STORE_LC_NUMERIC_SET_TO_NEEDED();
12298 if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC))
12299 need += SvLEN(PL_numeric_radix_sv);
12300 RESTORE_LC_NUMERIC();
12304 need = BIT_DIGITS(i);
12305 } /* if i < 0, the number of digits is hard to predict. */
12307 need += has_precis ? precis : 6; /* known default */
12312 #ifdef HAS_LDBL_SPRINTF_BUG
12313 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
12314 with sfio - Allen <allens@cpan.org> */
12317 # define MY_DBL_MAX DBL_MAX
12318 # else /* XXX guessing! HUGE_VAL may be defined as infinity, so not using */
12319 # if DOUBLESIZE >= 8
12320 # define MY_DBL_MAX 1.7976931348623157E+308L
12322 # define MY_DBL_MAX 3.40282347E+38L
12326 # ifdef HAS_LDBL_SPRINTF_BUG_LESS1 /* only between -1L & 1L - Allen */
12327 # define MY_DBL_MAX_BUG 1L
12329 # define MY_DBL_MAX_BUG MY_DBL_MAX
12333 # define MY_DBL_MIN DBL_MIN
12334 # else /* XXX guessing! -Allen */
12335 # if DOUBLESIZE >= 8
12336 # define MY_DBL_MIN 2.2250738585072014E-308L
12338 # define MY_DBL_MIN 1.17549435E-38L
12342 if ((intsize == 'q') && (c == 'f') &&
12343 ((fv < MY_DBL_MAX_BUG) && (fv > -MY_DBL_MAX_BUG)) &&
12344 (need < DBL_DIG)) {
12345 /* it's going to be short enough that
12346 * long double precision is not needed */
12348 if ((fv <= 0L) && (fv >= -0L))
12349 fix_ldbl_sprintf_bug = TRUE; /* 0 is 0 - easiest */
12351 /* would use Perl_fp_class as a double-check but not
12352 * functional on IRIX - see perl.h comments */
12354 if ((fv >= MY_DBL_MIN) || (fv <= -MY_DBL_MIN)) {
12355 /* It's within the range that a double can represent */
12356 #if defined(DBL_MAX) && !defined(DBL_MIN)
12357 if ((fv >= ((long double)1/DBL_MAX)) ||
12358 (fv <= (-(long double)1/DBL_MAX)))
12360 fix_ldbl_sprintf_bug = TRUE;
12363 if (fix_ldbl_sprintf_bug == TRUE) {
12373 # undef MY_DBL_MAX_BUG
12376 #endif /* HAS_LDBL_SPRINTF_BUG */
12378 need += 20; /* fudge factor */
12379 if (PL_efloatsize < need) {
12380 Safefree(PL_efloatbuf);
12381 PL_efloatsize = need + 20; /* more fudge */
12382 Newx(PL_efloatbuf, PL_efloatsize, char);
12383 PL_efloatbuf[0] = '\0';
12386 if ( !(width || left || plus || alt) && fill != '0'
12387 && has_precis && intsize != 'q' /* Shortcuts */
12388 && LIKELY(!Perl_isinfnan((NV)fv)) ) {
12389 /* See earlier comment about buggy Gconvert when digits,
12391 if ( c == 'g' && precis ) {
12392 STORE_LC_NUMERIC_SET_TO_NEEDED();
12393 SNPRINTF_G(fv, PL_efloatbuf, PL_efloatsize, precis);
12394 /* May return an empty string for digits==0 */
12395 if (*PL_efloatbuf) {
12396 elen = strlen(PL_efloatbuf);
12397 goto float_converted;
12399 } else if ( c == 'f' && !precis ) {
12400 if ((eptr = F0convert(nv, ebuf + sizeof ebuf, &elen)))
12405 if (UNLIKELY(hexfp)) {
12406 /* Hexadecimal floating point. */
12407 char* p = PL_efloatbuf;
12408 U8 vhex[VHEX_SIZE];
12409 U8* v = vhex; /* working pointer to vhex */
12410 U8* vend; /* pointer to one beyond last digit of vhex */
12411 U8* vfnz = NULL; /* first non-zero */
12412 U8* vlnz = NULL; /* last non-zero */
12413 U8* v0 = NULL; /* first output */
12414 const bool lower = (c == 'a');
12415 /* At output the values of vhex (up to vend) will
12416 * be mapped through the xdig to get the actual
12417 * human-readable xdigits. */
12418 const char* xdig = PL_hexdigit;
12419 int zerotail = 0; /* how many extra zeros to append */
12420 int exponent = 0; /* exponent of the floating point input */
12421 bool hexradix = FALSE; /* should we output the radix */
12422 bool subnormal = FALSE; /* IEEE 754 subnormal/denormal */
12423 bool negative = FALSE;
12425 /* XXX: NaN, Inf -- though they are printed as "NaN" and "Inf".
12427 * For example with denormals, (assuming the vanilla
12428 * 64-bit double): the exponent is zero. 1xp-1074 is
12429 * the smallest denormal and the smallest double, it
12430 * could be output also as 0x0.0000000000001p-1022 to
12431 * match its internal structure. */
12433 vend = S_hextract(aTHX_ nv, &exponent, &subnormal, vhex, NULL);
12434 S_hextract(aTHX_ nv, &exponent, &subnormal, vhex, vend);
12436 #if NVSIZE > DOUBLESIZE
12437 # ifdef HEXTRACT_HAS_IMPLICIT_BIT
12438 /* In this case there is an implicit bit,
12439 * and therefore the exponent is shifted by one. */
12442 # ifdef NV_X86_80_BIT
12444 /* The subnormals of the x86-80 have a base exponent of -16382,
12445 * (while the physical exponent bits are zero) but the frexp()
12446 * returned the scientific-style floating exponent. We want
12447 * to map the last one as:
12448 * -16831..-16384 -> -16382 (the last normal is 0x1p-16382)
12449 * -16835..-16388 -> -16384
12450 * since we want to keep the first hexdigit
12451 * as one of the [8421]. */
12452 exponent = -4 * ( (exponent + 1) / -4) - 2;
12457 /* TBD: other non-implicit-bit platforms than the x86-80. */
12461 negative = fv < 0 || Perl_signbit(nv);
12472 xdig += 16; /* Use uppercase hex. */
12475 /* Find the first non-zero xdigit. */
12476 for (v = vhex; v < vend; v++) {
12484 /* Find the last non-zero xdigit. */
12485 for (v = vend - 1; v >= vhex; v--) {
12492 #if NVSIZE == DOUBLESIZE
12498 #ifndef NV_X86_80_BIT
12500 /* IEEE 754 subnormals (but not the x86 80-bit):
12501 * we want "normalize" the subnormal,
12502 * so we need to right shift the hex nybbles
12503 * so that the output of the subnormal starts
12504 * from the first true bit. (Another, equally
12505 * valid, policy would be to dump the subnormal
12506 * nybbles as-is, to display the "physical" layout.) */
12509 /* Find the ceil(log2(v[0])) of
12510 * the top non-zero nybble. */
12511 for (i = vfnz[0], n = 0; i > 1; i >>= 1, n++) { }
12514 for (vshr = vlnz; vshr >= vfnz; vshr--) {
12515 vshr[1] |= (vshr[0] & (0xF >> (4 - n))) << (4 - n);
12529 U8* ve = (subnormal ? vlnz + 1 : vend);
12530 SSize_t vn = ve - (subnormal ? vfnz : vhex);
12531 if ((SSize_t)(precis + 1) < vn) {
12532 bool overflow = FALSE;
12533 if (v0[precis + 1] < 0x8) {
12534 /* Round down, nothing to do. */
12535 } else if (v0[precis + 1] > 0x8) {
12538 overflow = v0[precis] > 0xF;
12540 } else { /* v0[precis] == 0x8 */
12541 /* Half-point: round towards the one
12542 * with the even least-significant digit:
12550 * 78 -> 8 f8 -> 10 */
12551 if ((v0[precis] & 0x1)) {
12554 overflow = v0[precis] > 0xF;
12559 for (v = v0 + precis - 1; v >= v0; v--) {
12561 overflow = *v > 0xF;
12567 if (v == v0 - 1 && overflow) {
12568 /* If the overflow goes all the
12569 * way to the front, we need to
12570 * insert 0x1 in front, and adjust
12572 Move(v0, v0 + 1, vn, char);
12578 /* The new effective "last non zero". */
12579 vlnz = v0 + precis;
12583 subnormal ? precis - vn + 1 :
12584 precis - (vlnz - vhex);
12591 /* If there are non-zero xdigits, the radix
12592 * is output after the first one. */
12603 /* The radix is always output if precis, or if alt. */
12604 if (precis > 0 || alt) {
12609 #ifndef USE_LOCALE_NUMERIC
12612 STORE_LC_NUMERIC_SET_TO_NEEDED();
12613 if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC)) {
12615 const char* r = SvPV(PL_numeric_radix_sv, n);
12616 Copy(r, p, n, char);
12622 RESTORE_LC_NUMERIC();
12631 if (zerotail > 0) {
12632 while (zerotail--) {
12637 elen = p - PL_efloatbuf;
12638 elen += my_snprintf(p, PL_efloatsize - elen,
12639 "%c%+d", lower ? 'p' : 'P',
12642 if (elen < width) {
12644 /* Pad the back with spaces. */
12645 memset(PL_efloatbuf + elen, ' ', width - elen);
12647 else if (fill == '0') {
12648 /* Insert the zeros after the "0x" and the
12649 * the potential sign, but before the digits,
12650 * otherwise we end up with "0000xH.HHH...",
12651 * when we want "0x000H.HHH..." */
12652 STRLEN nzero = width - elen;
12653 char* zerox = PL_efloatbuf + 2;
12654 STRLEN nmove = elen - 2;
12655 if (negative || plus) {
12659 Move(zerox, zerox + nzero, nmove, char);
12660 memset(zerox, fill, nzero);
12663 /* Move it to the right. */
12664 Move(PL_efloatbuf, PL_efloatbuf + width - elen,
12666 /* Pad the front with spaces. */
12667 memset(PL_efloatbuf, ' ', width - elen);
12673 elen = S_infnan_2pv(nv, PL_efloatbuf, PL_efloatsize, plus);
12675 /* Not affecting infnan output: precision, alt, fill. */
12676 if (elen < width) {
12678 /* Pack the back with spaces. */
12679 memset(PL_efloatbuf + elen, ' ', width - elen);
12681 /* Move it to the right. */
12682 Move(PL_efloatbuf, PL_efloatbuf + width - elen,
12684 /* Pad the front with spaces. */
12685 memset(PL_efloatbuf, ' ', width - elen);
12693 char *ptr = ebuf + sizeof ebuf;
12696 #if defined(USE_QUADMATH)
12697 if (intsize == 'q') {
12701 /* FIXME: what to do if HAS_LONG_DOUBLE but not PERL_PRIfldbl? */
12702 #elif defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
12703 /* Note that this is HAS_LONG_DOUBLE and PERL_PRIfldbl,
12704 * not USE_LONG_DOUBLE and NVff. In other words,
12705 * this needs to work without USE_LONG_DOUBLE. */
12706 if (intsize == 'q') {
12707 /* Copy the one or more characters in a long double
12708 * format before the 'base' ([efgEFG]) character to
12709 * the format string. */
12710 static char const ldblf[] = PERL_PRIfldbl;
12711 char const *p = ldblf + sizeof(ldblf) - 3;
12712 while (p >= ldblf) { *--ptr = *p--; }
12717 do { *--ptr = '0' + (base % 10); } while (base /= 10);
12722 do { *--ptr = '0' + (base % 10); } while (base /= 10);
12734 /* No taint. Otherwise we are in the strange situation
12735 * where printf() taints but print($float) doesn't.
12738 STORE_LC_NUMERIC_SET_TO_NEEDED();
12740 /* hopefully the above makes ptr a very constrained format
12741 * that is safe to use, even though it's not literal */
12742 GCC_DIAG_IGNORE(-Wformat-nonliteral);
12743 #ifdef USE_QUADMATH
12745 const char* qfmt = quadmath_format_single(ptr);
12747 Perl_croak_nocontext("panic: quadmath invalid format \"%s\"", ptr);
12748 elen = quadmath_snprintf(PL_efloatbuf, PL_efloatsize,
12750 if ((IV)elen == -1)
12751 Perl_croak_nocontext("panic: quadmath_snprintf failed, format \"%s\"", qfmt);
12755 #elif defined(HAS_LONG_DOUBLE)
12756 elen = ((intsize == 'q')
12757 ? my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv)
12758 : my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, (double)fv));
12760 elen = my_sprintf(PL_efloatbuf, ptr, fv);
12766 eptr = PL_efloatbuf;
12767 assert((IV)elen > 0); /* here zero elen is bad */
12769 #ifdef USE_LOCALE_NUMERIC
12770 /* If the decimal point character in the string is UTF-8, make the
12772 if (PL_numeric_radix_sv && SvUTF8(PL_numeric_radix_sv)
12773 && instr(eptr, SvPVX_const(PL_numeric_radix_sv)))
12786 i = SvCUR(sv) - origlen;
12789 case 'c': *(va_arg(*args, char*)) = i; break;
12790 case 'h': *(va_arg(*args, short*)) = i; break;
12791 default: *(va_arg(*args, int*)) = i; break;
12792 case 'l': *(va_arg(*args, long*)) = i; break;
12793 case 'V': *(va_arg(*args, IV*)) = i; break;
12794 case 'z': *(va_arg(*args, SSize_t*)) = i; break;
12795 #ifdef HAS_PTRDIFF_T
12796 case 't': *(va_arg(*args, ptrdiff_t*)) = i; break;
12799 case 'j': *(va_arg(*args, intmax_t*)) = i; break;
12803 *(va_arg(*args, Quad_t*)) = i; break;
12810 sv_setuv_mg(argsv, has_utf8 ? (UV)sv_len_utf8(sv) : (UV)i);
12811 goto donevalidconversion;
12818 && (PL_op->op_type == OP_PRTF || PL_op->op_type == OP_SPRINTF)
12819 && ckWARN(WARN_PRINTF))
12821 SV * const msg = sv_newmortal();
12822 Perl_sv_setpvf(aTHX_ msg, "Invalid conversion in %sprintf: ",
12823 (PL_op->op_type == OP_PRTF) ? "" : "s");
12824 if (fmtstart < patend) {
12825 const char * const fmtend = q < patend ? q : patend;
12827 sv_catpvs(msg, "\"%");
12828 for (f = fmtstart; f < fmtend; f++) {
12830 sv_catpvn_nomg(msg, f, 1);
12832 Perl_sv_catpvf(aTHX_ msg,
12833 "\\%03"UVof, (UV)*f & 0xFF);
12836 sv_catpvs(msg, "\"");
12838 sv_catpvs(msg, "end of string");
12840 Perl_warner(aTHX_ packWARN(WARN_PRINTF), "%"SVf, SVfARG(msg)); /* yes, this is reentrant */
12843 /* output mangled stuff ... */
12849 /* ... right here, because formatting flags should not apply */
12850 SvGROW(sv, SvCUR(sv) + elen + 1);
12852 Copy(eptr, p, elen, char);
12855 SvCUR_set(sv, p - SvPVX_const(sv));
12857 continue; /* not "break" */
12860 if (is_utf8 != has_utf8) {
12863 sv_utf8_upgrade(sv);
12866 const STRLEN old_elen = elen;
12867 SV * const nsv = newSVpvn_flags(eptr, elen, SVs_TEMP);
12868 sv_utf8_upgrade(nsv);
12869 eptr = SvPVX_const(nsv);
12872 if (width) { /* fudge width (can't fudge elen) */
12873 width += elen - old_elen;
12879 /* signed value that's wrapped? */
12880 assert(elen <= ((~(STRLEN)0) >> 1));
12881 have = esignlen + zeros + elen;
12883 croak_memory_wrap();
12885 need = (have > width ? have : width);
12888 if (need >= (((STRLEN)~0) - SvCUR(sv) - dotstrlen - 1))
12889 croak_memory_wrap();
12890 SvGROW(sv, SvCUR(sv) + need + dotstrlen + 1);
12892 if (esignlen && fill == '0') {
12894 for (i = 0; i < (int)esignlen; i++)
12895 *p++ = esignbuf[i];
12897 if (gap && !left) {
12898 memset(p, fill, gap);
12901 if (esignlen && fill != '0') {
12903 for (i = 0; i < (int)esignlen; i++)
12904 *p++ = esignbuf[i];
12908 for (i = zeros; i; i--)
12912 Copy(eptr, p, elen, char);
12916 memset(p, ' ', gap);
12921 Copy(dotstr, p, dotstrlen, char);
12925 vectorize = FALSE; /* done iterating over vecstr */
12932 SvCUR_set(sv, p - SvPVX_const(sv));
12938 donevalidconversion:
12939 if (used_explicit_ix)
12940 no_redundant_warning = TRUE;
12942 S_warn_vcatpvfn_missing_argument(aTHX);
12945 /* Now that we've consumed all our printf format arguments (svix)
12946 * do we have things left on the stack that we didn't use?
12948 if (!no_redundant_warning && svmax >= svix + 1 && ckWARN(WARN_REDUNDANT)) {
12949 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
12950 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
12955 RESTORE_LC_NUMERIC(); /* Done outside loop, so don't have to save/restore
12959 /* =========================================================================
12961 =head1 Cloning an interpreter
12965 All the macros and functions in this section are for the private use of
12966 the main function, perl_clone().
12968 The foo_dup() functions make an exact copy of an existing foo thingy.
12969 During the course of a cloning, a hash table is used to map old addresses
12970 to new addresses. The table is created and manipulated with the
12971 ptr_table_* functions.
12973 * =========================================================================*/
12976 #if defined(USE_ITHREADS)
12978 /* XXX Remove this so it doesn't have to go thru the macro and return for nothing */
12979 #ifndef GpREFCNT_inc
12980 # define GpREFCNT_inc(gp) ((gp) ? (++(gp)->gp_refcnt, (gp)) : (GP*)NULL)
12984 /* Certain cases in Perl_ss_dup have been merged, by relying on the fact
12985 that currently av_dup, gv_dup and hv_dup are the same as sv_dup.
12986 If this changes, please unmerge ss_dup.
12987 Likewise, sv_dup_inc_multiple() relies on this fact. */
12988 #define sv_dup_inc_NN(s,t) SvREFCNT_inc_NN(sv_dup_inc(s,t))
12989 #define av_dup(s,t) MUTABLE_AV(sv_dup((const SV *)s,t))
12990 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
12991 #define hv_dup(s,t) MUTABLE_HV(sv_dup((const SV *)s,t))
12992 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
12993 #define cv_dup(s,t) MUTABLE_CV(sv_dup((const SV *)s,t))
12994 #define cv_dup_inc(s,t) MUTABLE_CV(sv_dup_inc((const SV *)s,t))
12995 #define io_dup(s,t) MUTABLE_IO(sv_dup((const SV *)s,t))
12996 #define io_dup_inc(s,t) MUTABLE_IO(sv_dup_inc((const SV *)s,t))
12997 #define gv_dup(s,t) MUTABLE_GV(sv_dup((const SV *)s,t))
12998 #define gv_dup_inc(s,t) MUTABLE_GV(sv_dup_inc((const SV *)s,t))
12999 #define SAVEPV(p) ((p) ? savepv(p) : NULL)
13000 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
13002 /* clone a parser */
13005 Perl_parser_dup(pTHX_ const yy_parser *const proto, CLONE_PARAMS *const param)
13009 PERL_ARGS_ASSERT_PARSER_DUP;
13014 /* look for it in the table first */
13015 parser = (yy_parser *)ptr_table_fetch(PL_ptr_table, proto);
13019 /* create anew and remember what it is */
13020 Newxz(parser, 1, yy_parser);
13021 ptr_table_store(PL_ptr_table, proto, parser);
13023 /* XXX these not yet duped */
13024 parser->old_parser = NULL;
13025 parser->stack = NULL;
13027 parser->stack_size = 0;
13028 /* XXX parser->stack->state = 0; */
13030 /* XXX eventually, just Copy() most of the parser struct ? */
13032 parser->lex_brackets = proto->lex_brackets;
13033 parser->lex_casemods = proto->lex_casemods;
13034 parser->lex_brackstack = savepvn(proto->lex_brackstack,
13035 (proto->lex_brackets < 120 ? 120 : proto->lex_brackets));
13036 parser->lex_casestack = savepvn(proto->lex_casestack,
13037 (proto->lex_casemods < 12 ? 12 : proto->lex_casemods));
13038 parser->lex_defer = proto->lex_defer;
13039 parser->lex_dojoin = proto->lex_dojoin;
13040 parser->lex_formbrack = proto->lex_formbrack;
13041 parser->lex_inpat = proto->lex_inpat;
13042 parser->lex_inwhat = proto->lex_inwhat;
13043 parser->lex_op = proto->lex_op;
13044 parser->lex_repl = sv_dup_inc(proto->lex_repl, param);
13045 parser->lex_starts = proto->lex_starts;
13046 parser->lex_stuff = sv_dup_inc(proto->lex_stuff, param);
13047 parser->multi_close = proto->multi_close;
13048 parser->multi_open = proto->multi_open;
13049 parser->multi_start = proto->multi_start;
13050 parser->multi_end = proto->multi_end;
13051 parser->preambled = proto->preambled;
13052 parser->lex_super_state = proto->lex_super_state;
13053 parser->lex_sub_inwhat = proto->lex_sub_inwhat;
13054 parser->lex_sub_op = proto->lex_sub_op;
13055 parser->lex_sub_repl= sv_dup_inc(proto->lex_sub_repl, param);
13056 parser->linestr = sv_dup_inc(proto->linestr, param);
13057 parser->expect = proto->expect;
13058 parser->copline = proto->copline;
13059 parser->last_lop_op = proto->last_lop_op;
13060 parser->lex_state = proto->lex_state;
13061 parser->rsfp = fp_dup(proto->rsfp, '<', param);
13062 /* rsfp_filters entries have fake IoDIRP() */
13063 parser->rsfp_filters= av_dup_inc(proto->rsfp_filters, param);
13064 parser->in_my = proto->in_my;
13065 parser->in_my_stash = hv_dup(proto->in_my_stash, param);
13066 parser->error_count = proto->error_count;
13067 parser->sig_elems = proto->sig_elems;
13068 parser->sig_optelems= proto->sig_optelems;
13069 parser->sig_slurpy = proto->sig_slurpy;
13070 parser->linestr = sv_dup_inc(proto->linestr, param);
13073 char * const ols = SvPVX(proto->linestr);
13074 char * const ls = SvPVX(parser->linestr);
13076 parser->bufptr = ls + (proto->bufptr >= ols ?
13077 proto->bufptr - ols : 0);
13078 parser->oldbufptr = ls + (proto->oldbufptr >= ols ?
13079 proto->oldbufptr - ols : 0);
13080 parser->oldoldbufptr= ls + (proto->oldoldbufptr >= ols ?
13081 proto->oldoldbufptr - ols : 0);
13082 parser->linestart = ls + (proto->linestart >= ols ?
13083 proto->linestart - ols : 0);
13084 parser->last_uni = ls + (proto->last_uni >= ols ?
13085 proto->last_uni - ols : 0);
13086 parser->last_lop = ls + (proto->last_lop >= ols ?
13087 proto->last_lop - ols : 0);
13089 parser->bufend = ls + SvCUR(parser->linestr);
13092 Copy(proto->tokenbuf, parser->tokenbuf, 256, char);
13095 Copy(proto->nextval, parser->nextval, 5, YYSTYPE);
13096 Copy(proto->nexttype, parser->nexttype, 5, I32);
13097 parser->nexttoke = proto->nexttoke;
13099 /* XXX should clone saved_curcop here, but we aren't passed
13100 * proto_perl; so do it in perl_clone_using instead */
13106 /* duplicate a file handle */
13109 Perl_fp_dup(pTHX_ PerlIO *const fp, const char type, CLONE_PARAMS *const param)
13113 PERL_ARGS_ASSERT_FP_DUP;
13114 PERL_UNUSED_ARG(type);
13117 return (PerlIO*)NULL;
13119 /* look for it in the table first */
13120 ret = (PerlIO*)ptr_table_fetch(PL_ptr_table, fp);
13124 /* create anew and remember what it is */
13125 #ifdef __amigaos4__
13126 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE|PERLIO_DUP_FD);
13128 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE);
13130 ptr_table_store(PL_ptr_table, fp, ret);
13134 /* duplicate a directory handle */
13137 Perl_dirp_dup(pTHX_ DIR *const dp, CLONE_PARAMS *const param)
13141 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
13143 const Direntry_t *dirent;
13144 char smallbuf[256]; /* XXX MAXPATHLEN, surely? */
13150 PERL_UNUSED_CONTEXT;
13151 PERL_ARGS_ASSERT_DIRP_DUP;
13156 /* look for it in the table first */
13157 ret = (DIR*)ptr_table_fetch(PL_ptr_table, dp);
13161 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
13163 PERL_UNUSED_ARG(param);
13167 /* open the current directory (so we can switch back) */
13168 if (!(pwd = PerlDir_open("."))) return (DIR *)NULL;
13170 /* chdir to our dir handle and open the present working directory */
13171 if (fchdir(my_dirfd(dp)) < 0 || !(ret = PerlDir_open("."))) {
13172 PerlDir_close(pwd);
13173 return (DIR *)NULL;
13175 /* Now we should have two dir handles pointing to the same dir. */
13177 /* Be nice to the calling code and chdir back to where we were. */
13178 /* XXX If this fails, then what? */
13179 PERL_UNUSED_RESULT(fchdir(my_dirfd(pwd)));
13181 /* We have no need of the pwd handle any more. */
13182 PerlDir_close(pwd);
13185 # define d_namlen(d) (d)->d_namlen
13187 # define d_namlen(d) strlen((d)->d_name)
13189 /* Iterate once through dp, to get the file name at the current posi-
13190 tion. Then step back. */
13191 pos = PerlDir_tell(dp);
13192 if ((dirent = PerlDir_read(dp))) {
13193 len = d_namlen(dirent);
13194 if (len > sizeof(dirent->d_name) && sizeof(dirent->d_name) > PTRSIZE) {
13195 /* If the len is somehow magically longer than the
13196 * maximum length of the directory entry, even though
13197 * we could fit it in a buffer, we could not copy it
13198 * from the dirent. Bail out. */
13199 PerlDir_close(ret);
13202 if (len <= sizeof smallbuf) name = smallbuf;
13203 else Newx(name, len, char);
13204 Move(dirent->d_name, name, len, char);
13206 PerlDir_seek(dp, pos);
13208 /* Iterate through the new dir handle, till we find a file with the
13210 if (!dirent) /* just before the end */
13212 pos = PerlDir_tell(ret);
13213 if (PerlDir_read(ret)) continue; /* not there yet */
13214 PerlDir_seek(ret, pos); /* step back */
13218 const long pos0 = PerlDir_tell(ret);
13220 pos = PerlDir_tell(ret);
13221 if ((dirent = PerlDir_read(ret))) {
13222 if (len == (STRLEN)d_namlen(dirent)
13223 && memEQ(name, dirent->d_name, len)) {
13225 PerlDir_seek(ret, pos); /* step back */
13228 /* else we are not there yet; keep iterating */
13230 else { /* This is not meant to happen. The best we can do is
13231 reset the iterator to the beginning. */
13232 PerlDir_seek(ret, pos0);
13239 if (name && name != smallbuf)
13244 ret = win32_dirp_dup(dp, param);
13247 /* pop it in the pointer table */
13249 ptr_table_store(PL_ptr_table, dp, ret);
13254 /* duplicate a typeglob */
13257 Perl_gp_dup(pTHX_ GP *const gp, CLONE_PARAMS *const param)
13261 PERL_ARGS_ASSERT_GP_DUP;
13265 /* look for it in the table first */
13266 ret = (GP*)ptr_table_fetch(PL_ptr_table, gp);
13270 /* create anew and remember what it is */
13272 ptr_table_store(PL_ptr_table, gp, ret);
13275 /* ret->gp_refcnt must be 0 before any other dups are called. We're relying
13276 on Newxz() to do this for us. */
13277 ret->gp_sv = sv_dup_inc(gp->gp_sv, param);
13278 ret->gp_io = io_dup_inc(gp->gp_io, param);
13279 ret->gp_form = cv_dup_inc(gp->gp_form, param);
13280 ret->gp_av = av_dup_inc(gp->gp_av, param);
13281 ret->gp_hv = hv_dup_inc(gp->gp_hv, param);
13282 ret->gp_egv = gv_dup(gp->gp_egv, param);/* GvEGV is not refcounted */
13283 ret->gp_cv = cv_dup_inc(gp->gp_cv, param);
13284 ret->gp_cvgen = gp->gp_cvgen;
13285 ret->gp_line = gp->gp_line;
13286 ret->gp_file_hek = hek_dup(gp->gp_file_hek, param);
13290 /* duplicate a chain of magic */
13293 Perl_mg_dup(pTHX_ MAGIC *mg, CLONE_PARAMS *const param)
13295 MAGIC *mgret = NULL;
13296 MAGIC **mgprev_p = &mgret;
13298 PERL_ARGS_ASSERT_MG_DUP;
13300 for (; mg; mg = mg->mg_moremagic) {
13303 if ((param->flags & CLONEf_JOIN_IN)
13304 && mg->mg_type == PERL_MAGIC_backref)
13305 /* when joining, we let the individual SVs add themselves to
13306 * backref as needed. */
13309 Newx(nmg, 1, MAGIC);
13311 mgprev_p = &(nmg->mg_moremagic);
13313 /* There was a comment "XXX copy dynamic vtable?" but as we don't have
13314 dynamic vtables, I'm not sure why Sarathy wrote it. The comment dates
13315 from the original commit adding Perl_mg_dup() - revision 4538.
13316 Similarly there is the annotation "XXX random ptr?" next to the
13317 assignment to nmg->mg_ptr. */
13320 /* FIXME for plugins
13321 if (nmg->mg_type == PERL_MAGIC_qr) {
13322 nmg->mg_obj = MUTABLE_SV(CALLREGDUPE((REGEXP*)nmg->mg_obj, param));
13326 nmg->mg_obj = (nmg->mg_flags & MGf_REFCOUNTED)
13327 ? nmg->mg_type == PERL_MAGIC_backref
13328 /* The backref AV has its reference
13329 * count deliberately bumped by 1 */
13330 ? SvREFCNT_inc(av_dup_inc((const AV *)
13331 nmg->mg_obj, param))
13332 : sv_dup_inc(nmg->mg_obj, param)
13333 : sv_dup(nmg->mg_obj, param);
13335 if (nmg->mg_ptr && nmg->mg_type != PERL_MAGIC_regex_global) {
13336 if (nmg->mg_len > 0) {
13337 nmg->mg_ptr = SAVEPVN(nmg->mg_ptr, nmg->mg_len);
13338 if (nmg->mg_type == PERL_MAGIC_overload_table &&
13339 AMT_AMAGIC((AMT*)nmg->mg_ptr))
13341 AMT * const namtp = (AMT*)nmg->mg_ptr;
13342 sv_dup_inc_multiple((SV**)(namtp->table),
13343 (SV**)(namtp->table), NofAMmeth, param);
13346 else if (nmg->mg_len == HEf_SVKEY)
13347 nmg->mg_ptr = (char*)sv_dup_inc((const SV *)nmg->mg_ptr, param);
13349 if ((nmg->mg_flags & MGf_DUP) && nmg->mg_virtual && nmg->mg_virtual->svt_dup) {
13350 nmg->mg_virtual->svt_dup(aTHX_ nmg, param);
13356 #endif /* USE_ITHREADS */
13358 struct ptr_tbl_arena {
13359 struct ptr_tbl_arena *next;
13360 struct ptr_tbl_ent array[1023/3]; /* as ptr_tbl_ent has 3 pointers. */
13363 /* create a new pointer-mapping table */
13366 Perl_ptr_table_new(pTHX)
13369 PERL_UNUSED_CONTEXT;
13371 Newx(tbl, 1, PTR_TBL_t);
13372 tbl->tbl_max = 511;
13373 tbl->tbl_items = 0;
13374 tbl->tbl_arena = NULL;
13375 tbl->tbl_arena_next = NULL;
13376 tbl->tbl_arena_end = NULL;
13377 Newxz(tbl->tbl_ary, tbl->tbl_max + 1, PTR_TBL_ENT_t*);
13381 #define PTR_TABLE_HASH(ptr) \
13382 ((PTR2UV(ptr) >> 3) ^ (PTR2UV(ptr) >> (3 + 7)) ^ (PTR2UV(ptr) >> (3 + 17)))
13384 /* map an existing pointer using a table */
13386 STATIC PTR_TBL_ENT_t *
13387 S_ptr_table_find(PTR_TBL_t *const tbl, const void *const sv)
13389 PTR_TBL_ENT_t *tblent;
13390 const UV hash = PTR_TABLE_HASH(sv);
13392 PERL_ARGS_ASSERT_PTR_TABLE_FIND;
13394 tblent = tbl->tbl_ary[hash & tbl->tbl_max];
13395 for (; tblent; tblent = tblent->next) {
13396 if (tblent->oldval == sv)
13403 Perl_ptr_table_fetch(pTHX_ PTR_TBL_t *const tbl, const void *const sv)
13405 PTR_TBL_ENT_t const *const tblent = ptr_table_find(tbl, sv);
13407 PERL_ARGS_ASSERT_PTR_TABLE_FETCH;
13408 PERL_UNUSED_CONTEXT;
13410 return tblent ? tblent->newval : NULL;
13413 /* add a new entry to a pointer-mapping table 'tbl'. In hash terms, 'oldsv' is
13414 * the key; 'newsv' is the value. The names "old" and "new" are specific to
13415 * the core's typical use of ptr_tables in thread cloning. */
13418 Perl_ptr_table_store(pTHX_ PTR_TBL_t *const tbl, const void *const oldsv, void *const newsv)
13420 PTR_TBL_ENT_t *tblent = ptr_table_find(tbl, oldsv);
13422 PERL_ARGS_ASSERT_PTR_TABLE_STORE;
13423 PERL_UNUSED_CONTEXT;
13426 tblent->newval = newsv;
13428 const UV entry = PTR_TABLE_HASH(oldsv) & tbl->tbl_max;
13430 if (tbl->tbl_arena_next == tbl->tbl_arena_end) {
13431 struct ptr_tbl_arena *new_arena;
13433 Newx(new_arena, 1, struct ptr_tbl_arena);
13434 new_arena->next = tbl->tbl_arena;
13435 tbl->tbl_arena = new_arena;
13436 tbl->tbl_arena_next = new_arena->array;
13437 tbl->tbl_arena_end = C_ARRAY_END(new_arena->array);
13440 tblent = tbl->tbl_arena_next++;
13442 tblent->oldval = oldsv;
13443 tblent->newval = newsv;
13444 tblent->next = tbl->tbl_ary[entry];
13445 tbl->tbl_ary[entry] = tblent;
13447 if (tblent->next && tbl->tbl_items > tbl->tbl_max)
13448 ptr_table_split(tbl);
13452 /* double the hash bucket size of an existing ptr table */
13455 Perl_ptr_table_split(pTHX_ PTR_TBL_t *const tbl)
13457 PTR_TBL_ENT_t **ary = tbl->tbl_ary;
13458 const UV oldsize = tbl->tbl_max + 1;
13459 UV newsize = oldsize * 2;
13462 PERL_ARGS_ASSERT_PTR_TABLE_SPLIT;
13463 PERL_UNUSED_CONTEXT;
13465 Renew(ary, newsize, PTR_TBL_ENT_t*);
13466 Zero(&ary[oldsize], newsize-oldsize, PTR_TBL_ENT_t*);
13467 tbl->tbl_max = --newsize;
13468 tbl->tbl_ary = ary;
13469 for (i=0; i < oldsize; i++, ary++) {
13470 PTR_TBL_ENT_t **entp = ary;
13471 PTR_TBL_ENT_t *ent = *ary;
13472 PTR_TBL_ENT_t **curentp;
13475 curentp = ary + oldsize;
13477 if ((newsize & PTR_TABLE_HASH(ent->oldval)) != i) {
13479 ent->next = *curentp;
13489 /* remove all the entries from a ptr table */
13490 /* Deprecated - will be removed post 5.14 */
13493 Perl_ptr_table_clear(pTHX_ PTR_TBL_t *const tbl)
13495 PERL_UNUSED_CONTEXT;
13496 if (tbl && tbl->tbl_items) {
13497 struct ptr_tbl_arena *arena = tbl->tbl_arena;
13499 Zero(tbl->tbl_ary, tbl->tbl_max + 1, struct ptr_tbl_ent *);
13502 struct ptr_tbl_arena *next = arena->next;
13508 tbl->tbl_items = 0;
13509 tbl->tbl_arena = NULL;
13510 tbl->tbl_arena_next = NULL;
13511 tbl->tbl_arena_end = NULL;
13515 /* clear and free a ptr table */
13518 Perl_ptr_table_free(pTHX_ PTR_TBL_t *const tbl)
13520 struct ptr_tbl_arena *arena;
13522 PERL_UNUSED_CONTEXT;
13528 arena = tbl->tbl_arena;
13531 struct ptr_tbl_arena *next = arena->next;
13537 Safefree(tbl->tbl_ary);
13541 #if defined(USE_ITHREADS)
13544 Perl_rvpv_dup(pTHX_ SV *const dstr, const SV *const sstr, CLONE_PARAMS *const param)
13546 PERL_ARGS_ASSERT_RVPV_DUP;
13548 assert(!isREGEXP(sstr));
13550 if (SvWEAKREF(sstr)) {
13551 SvRV_set(dstr, sv_dup(SvRV_const(sstr), param));
13552 if (param->flags & CLONEf_JOIN_IN) {
13553 /* if joining, we add any back references individually rather
13554 * than copying the whole backref array */
13555 Perl_sv_add_backref(aTHX_ SvRV(dstr), dstr);
13559 SvRV_set(dstr, sv_dup_inc(SvRV_const(sstr), param));
13561 else if (SvPVX_const(sstr)) {
13562 /* Has something there */
13564 /* Normal PV - clone whole allocated space */
13565 SvPV_set(dstr, SAVEPVN(SvPVX_const(sstr), SvLEN(sstr)-1));
13566 /* sstr may not be that normal, but actually copy on write.
13567 But we are a true, independent SV, so: */
13571 /* Special case - not normally malloced for some reason */
13572 if (isGV_with_GP(sstr)) {
13573 /* Don't need to do anything here. */
13575 else if ((SvIsCOW(sstr))) {
13576 /* A "shared" PV - clone it as "shared" PV */
13578 HEK_KEY(hek_dup(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)),
13582 /* Some other special case - random pointer */
13583 SvPV_set(dstr, (char *) SvPVX_const(sstr));
13588 /* Copy the NULL */
13589 SvPV_set(dstr, NULL);
13593 /* duplicate a list of SVs. source and dest may point to the same memory. */
13595 S_sv_dup_inc_multiple(pTHX_ SV *const *source, SV **dest,
13596 SSize_t items, CLONE_PARAMS *const param)
13598 PERL_ARGS_ASSERT_SV_DUP_INC_MULTIPLE;
13600 while (items-- > 0) {
13601 *dest++ = sv_dup_inc(*source++, param);
13607 /* duplicate an SV of any type (including AV, HV etc) */
13610 S_sv_dup_common(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13615 PERL_ARGS_ASSERT_SV_DUP_COMMON;
13617 if (SvTYPE(sstr) == (svtype)SVTYPEMASK) {
13618 #ifdef DEBUG_LEAKING_SCALARS_ABORT
13623 /* look for it in the table first */
13624 dstr = MUTABLE_SV(ptr_table_fetch(PL_ptr_table, sstr));
13628 if(param->flags & CLONEf_JOIN_IN) {
13629 /** We are joining here so we don't want do clone
13630 something that is bad **/
13631 if (SvTYPE(sstr) == SVt_PVHV) {
13632 const HEK * const hvname = HvNAME_HEK(sstr);
13634 /** don't clone stashes if they already exist **/
13635 dstr = MUTABLE_SV(gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
13636 HEK_UTF8(hvname) ? SVf_UTF8 : 0));
13637 ptr_table_store(PL_ptr_table, sstr, dstr);
13641 else if (SvTYPE(sstr) == SVt_PVGV && !SvFAKE(sstr)) {
13642 HV *stash = GvSTASH(sstr);
13643 const HEK * hvname;
13644 if (stash && (hvname = HvNAME_HEK(stash))) {
13645 /** don't clone GVs if they already exist **/
13647 stash = gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
13648 HEK_UTF8(hvname) ? SVf_UTF8 : 0);
13650 stash, GvNAME(sstr),
13656 if (svp && *svp && SvTYPE(*svp) == SVt_PVGV) {
13657 ptr_table_store(PL_ptr_table, sstr, *svp);
13664 /* create anew and remember what it is */
13667 #ifdef DEBUG_LEAKING_SCALARS
13668 dstr->sv_debug_optype = sstr->sv_debug_optype;
13669 dstr->sv_debug_line = sstr->sv_debug_line;
13670 dstr->sv_debug_inpad = sstr->sv_debug_inpad;
13671 dstr->sv_debug_parent = (SV*)sstr;
13672 FREE_SV_DEBUG_FILE(dstr);
13673 dstr->sv_debug_file = savesharedpv(sstr->sv_debug_file);
13676 ptr_table_store(PL_ptr_table, sstr, dstr);
13679 SvFLAGS(dstr) = SvFLAGS(sstr);
13680 SvFLAGS(dstr) &= ~SVf_OOK; /* don't propagate OOK hack */
13681 SvREFCNT(dstr) = 0; /* must be before any other dups! */
13684 if (SvANY(sstr) && PL_watch_pvx && SvPVX_const(sstr) == PL_watch_pvx)
13685 PerlIO_printf(Perl_debug_log, "watch at %p hit, found string \"%s\"\n",
13686 (void*)PL_watch_pvx, SvPVX_const(sstr));
13689 /* don't clone objects whose class has asked us not to */
13691 && ! (SvFLAGS(SvSTASH(sstr)) & SVphv_CLONEABLE))
13697 switch (SvTYPE(sstr)) {
13699 SvANY(dstr) = NULL;
13702 SET_SVANY_FOR_BODYLESS_IV(dstr);
13704 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
13706 SvIV_set(dstr, SvIVX(sstr));
13710 #if NVSIZE <= IVSIZE
13711 SET_SVANY_FOR_BODYLESS_NV(dstr);
13713 SvANY(dstr) = new_XNV();
13715 SvNV_set(dstr, SvNVX(sstr));
13719 /* These are all the types that need complex bodies allocating. */
13721 const svtype sv_type = SvTYPE(sstr);
13722 const struct body_details *const sv_type_details
13723 = bodies_by_type + sv_type;
13727 Perl_croak(aTHX_ "Bizarre SvTYPE [%" IVdf "]", (IV)SvTYPE(sstr));
13743 assert(sv_type_details->body_size);
13744 if (sv_type_details->arena) {
13745 new_body_inline(new_body, sv_type);
13747 = (void*)((char*)new_body - sv_type_details->offset);
13749 new_body = new_NOARENA(sv_type_details);
13753 SvANY(dstr) = new_body;
13756 Copy(((char*)SvANY(sstr)) + sv_type_details->offset,
13757 ((char*)SvANY(dstr)) + sv_type_details->offset,
13758 sv_type_details->copy, char);
13760 Copy(((char*)SvANY(sstr)),
13761 ((char*)SvANY(dstr)),
13762 sv_type_details->body_size + sv_type_details->offset, char);
13765 if (sv_type != SVt_PVAV && sv_type != SVt_PVHV
13766 && !isGV_with_GP(dstr)
13768 && !(sv_type == SVt_PVIO && !(IoFLAGS(dstr) & IOf_FAKE_DIRP)))
13769 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
13771 /* The Copy above means that all the source (unduplicated) pointers
13772 are now in the destination. We can check the flags and the
13773 pointers in either, but it's possible that there's less cache
13774 missing by always going for the destination.
13775 FIXME - instrument and check that assumption */
13776 if (sv_type >= SVt_PVMG) {
13778 SvMAGIC_set(dstr, mg_dup(SvMAGIC(dstr), param));
13779 if (SvOBJECT(dstr) && SvSTASH(dstr))
13780 SvSTASH_set(dstr, hv_dup_inc(SvSTASH(dstr), param));
13781 else SvSTASH_set(dstr, 0); /* don't copy DESTROY cache */
13784 /* The cast silences a GCC warning about unhandled types. */
13785 switch ((int)sv_type) {
13796 /* FIXME for plugins */
13797 dstr->sv_u.svu_rx = ((REGEXP *)dstr)->sv_any;
13798 re_dup_guts((REGEXP*) sstr, (REGEXP*) dstr, param);
13801 /* XXX LvTARGOFF sometimes holds PMOP* when DEBUGGING */
13802 if (LvTYPE(dstr) == 't') /* for tie: unrefcnted fake (SV**) */
13803 LvTARG(dstr) = dstr;
13804 else if (LvTYPE(dstr) == 'T') /* for tie: fake HE */
13805 LvTARG(dstr) = MUTABLE_SV(he_dup((HE*)LvTARG(dstr), 0, param));
13807 LvTARG(dstr) = sv_dup_inc(LvTARG(dstr), param);
13808 if (isREGEXP(sstr)) goto duprex;
13810 /* non-GP case already handled above */
13811 if(isGV_with_GP(sstr)) {
13812 GvNAME_HEK(dstr) = hek_dup(GvNAME_HEK(dstr), param);
13813 /* Don't call sv_add_backref here as it's going to be
13814 created as part of the magic cloning of the symbol
13815 table--unless this is during a join and the stash
13816 is not actually being cloned. */
13817 /* Danger Will Robinson - GvGP(dstr) isn't initialised
13818 at the point of this comment. */
13819 GvSTASH(dstr) = hv_dup(GvSTASH(dstr), param);
13820 if (param->flags & CLONEf_JOIN_IN)
13821 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
13822 GvGP_set(dstr, gp_dup(GvGP(sstr), param));
13823 (void)GpREFCNT_inc(GvGP(dstr));
13827 /* PL_parser->rsfp_filters entries have fake IoDIRP() */
13828 if(IoFLAGS(dstr) & IOf_FAKE_DIRP) {
13829 /* I have no idea why fake dirp (rsfps)
13830 should be treated differently but otherwise
13831 we end up with leaks -- sky*/
13832 IoTOP_GV(dstr) = gv_dup_inc(IoTOP_GV(dstr), param);
13833 IoFMT_GV(dstr) = gv_dup_inc(IoFMT_GV(dstr), param);
13834 IoBOTTOM_GV(dstr) = gv_dup_inc(IoBOTTOM_GV(dstr), param);
13836 IoTOP_GV(dstr) = gv_dup(IoTOP_GV(dstr), param);
13837 IoFMT_GV(dstr) = gv_dup(IoFMT_GV(dstr), param);
13838 IoBOTTOM_GV(dstr) = gv_dup(IoBOTTOM_GV(dstr), param);
13839 if (IoDIRP(dstr)) {
13840 IoDIRP(dstr) = dirp_dup(IoDIRP(dstr), param);
13843 /* IoDIRP(dstr) is already a copy of IoDIRP(sstr) */
13845 IoIFP(dstr) = fp_dup(IoIFP(sstr), IoTYPE(dstr), param);
13847 if (IoOFP(dstr) == IoIFP(sstr))
13848 IoOFP(dstr) = IoIFP(dstr);
13850 IoOFP(dstr) = fp_dup(IoOFP(dstr), IoTYPE(dstr), param);
13851 IoTOP_NAME(dstr) = SAVEPV(IoTOP_NAME(dstr));
13852 IoFMT_NAME(dstr) = SAVEPV(IoFMT_NAME(dstr));
13853 IoBOTTOM_NAME(dstr) = SAVEPV(IoBOTTOM_NAME(dstr));
13856 /* avoid cloning an empty array */
13857 if (AvARRAY((const AV *)sstr) && AvFILLp((const AV *)sstr) >= 0) {
13858 SV **dst_ary, **src_ary;
13859 SSize_t items = AvFILLp((const AV *)sstr) + 1;
13861 src_ary = AvARRAY((const AV *)sstr);
13862 Newxz(dst_ary, AvMAX((const AV *)sstr)+1, SV*);
13863 ptr_table_store(PL_ptr_table, src_ary, dst_ary);
13864 AvARRAY(MUTABLE_AV(dstr)) = dst_ary;
13865 AvALLOC((const AV *)dstr) = dst_ary;
13866 if (AvREAL((const AV *)sstr)) {
13867 dst_ary = sv_dup_inc_multiple(src_ary, dst_ary, items,
13871 while (items-- > 0)
13872 *dst_ary++ = sv_dup(*src_ary++, param);
13874 items = AvMAX((const AV *)sstr) - AvFILLp((const AV *)sstr);
13875 while (items-- > 0) {
13880 AvARRAY(MUTABLE_AV(dstr)) = NULL;
13881 AvALLOC((const AV *)dstr) = (SV**)NULL;
13882 AvMAX( (const AV *)dstr) = -1;
13883 AvFILLp((const AV *)dstr) = -1;
13887 if (HvARRAY((const HV *)sstr)) {
13889 const bool sharekeys = !!HvSHAREKEYS(sstr);
13890 XPVHV * const dxhv = (XPVHV*)SvANY(dstr);
13891 XPVHV * const sxhv = (XPVHV*)SvANY(sstr);
13893 Newx(darray, PERL_HV_ARRAY_ALLOC_BYTES(dxhv->xhv_max+1)
13894 + (SvOOK(sstr) ? sizeof(struct xpvhv_aux) : 0),
13896 HvARRAY(dstr) = (HE**)darray;
13897 while (i <= sxhv->xhv_max) {
13898 const HE * const source = HvARRAY(sstr)[i];
13899 HvARRAY(dstr)[i] = source
13900 ? he_dup(source, sharekeys, param) : 0;
13904 const struct xpvhv_aux * const saux = HvAUX(sstr);
13905 struct xpvhv_aux * const daux = HvAUX(dstr);
13906 /* This flag isn't copied. */
13909 if (saux->xhv_name_count) {
13910 HEK ** const sname = saux->xhv_name_u.xhvnameu_names;
13912 = saux->xhv_name_count < 0
13913 ? -saux->xhv_name_count
13914 : saux->xhv_name_count;
13915 HEK **shekp = sname + count;
13917 Newx(daux->xhv_name_u.xhvnameu_names, count, HEK *);
13918 dhekp = daux->xhv_name_u.xhvnameu_names + count;
13919 while (shekp-- > sname) {
13921 *dhekp = hek_dup(*shekp, param);
13925 daux->xhv_name_u.xhvnameu_name
13926 = hek_dup(saux->xhv_name_u.xhvnameu_name,
13929 daux->xhv_name_count = saux->xhv_name_count;
13931 daux->xhv_aux_flags = saux->xhv_aux_flags;
13932 #ifdef PERL_HASH_RANDOMIZE_KEYS
13933 daux->xhv_rand = saux->xhv_rand;
13934 daux->xhv_last_rand = saux->xhv_last_rand;
13936 daux->xhv_riter = saux->xhv_riter;
13937 daux->xhv_eiter = saux->xhv_eiter
13938 ? he_dup(saux->xhv_eiter,
13939 cBOOL(HvSHAREKEYS(sstr)), param) : 0;
13940 /* backref array needs refcnt=2; see sv_add_backref */
13941 daux->xhv_backreferences =
13942 (param->flags & CLONEf_JOIN_IN)
13943 /* when joining, we let the individual GVs and
13944 * CVs add themselves to backref as
13945 * needed. This avoids pulling in stuff
13946 * that isn't required, and simplifies the
13947 * case where stashes aren't cloned back
13948 * if they already exist in the parent
13951 : saux->xhv_backreferences
13952 ? (SvTYPE(saux->xhv_backreferences) == SVt_PVAV)
13953 ? MUTABLE_AV(SvREFCNT_inc(
13954 sv_dup_inc((const SV *)
13955 saux->xhv_backreferences, param)))
13956 : MUTABLE_AV(sv_dup((const SV *)
13957 saux->xhv_backreferences, param))
13960 daux->xhv_mro_meta = saux->xhv_mro_meta
13961 ? mro_meta_dup(saux->xhv_mro_meta, param)
13964 /* Record stashes for possible cloning in Perl_clone(). */
13966 av_push(param->stashes, dstr);
13970 HvARRAY(MUTABLE_HV(dstr)) = NULL;
13973 if (!(param->flags & CLONEf_COPY_STACKS)) {
13978 /* NOTE: not refcounted */
13979 SvANY(MUTABLE_CV(dstr))->xcv_stash =
13980 hv_dup(CvSTASH(dstr), param);
13981 if ((param->flags & CLONEf_JOIN_IN) && CvSTASH(dstr))
13982 Perl_sv_add_backref(aTHX_ MUTABLE_SV(CvSTASH(dstr)), dstr);
13983 if (!CvISXSUB(dstr)) {
13985 CvROOT(dstr) = OpREFCNT_inc(CvROOT(dstr));
13987 CvSLABBED_off(dstr);
13988 } else if (CvCONST(dstr)) {
13989 CvXSUBANY(dstr).any_ptr =
13990 sv_dup_inc((const SV *)CvXSUBANY(dstr).any_ptr, param);
13992 assert(!CvSLABBED(dstr));
13993 if (CvDYNFILE(dstr)) CvFILE(dstr) = SAVEPV(CvFILE(dstr));
13995 SvANY((CV *)dstr)->xcv_gv_u.xcv_hek =
13996 hek_dup(CvNAME_HEK((CV *)sstr), param);
13997 /* don't dup if copying back - CvGV isn't refcounted, so the
13998 * duped GV may never be freed. A bit of a hack! DAPM */
14000 SvANY(MUTABLE_CV(dstr))->xcv_gv_u.xcv_gv =
14002 ? gv_dup_inc(CvGV(sstr), param)
14003 : (param->flags & CLONEf_JOIN_IN)
14005 : gv_dup(CvGV(sstr), param);
14007 if (!CvISXSUB(sstr)) {
14008 PADLIST * padlist = CvPADLIST(sstr);
14010 padlist = padlist_dup(padlist, param);
14011 CvPADLIST_set(dstr, padlist);
14013 /* unthreaded perl can't sv_dup so we dont support unthreaded's CvHSCXT */
14014 PoisonPADLIST(dstr);
14017 CvWEAKOUTSIDE(sstr)
14018 ? cv_dup( CvOUTSIDE(dstr), param)
14019 : cv_dup_inc(CvOUTSIDE(dstr), param);
14029 Perl_sv_dup_inc(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14031 PERL_ARGS_ASSERT_SV_DUP_INC;
14032 return sstr ? SvREFCNT_inc(sv_dup_common(sstr, param)) : NULL;
14036 Perl_sv_dup(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14038 SV *dstr = sstr ? sv_dup_common(sstr, param) : NULL;
14039 PERL_ARGS_ASSERT_SV_DUP;
14041 /* Track every SV that (at least initially) had a reference count of 0.
14042 We need to do this by holding an actual reference to it in this array.
14043 If we attempt to cheat, turn AvREAL_off(), and store only pointers
14044 (akin to the stashes hash, and the perl stack), we come unstuck if
14045 a weak reference (or other SV legitimately SvREFCNT() == 0 for this
14046 thread) is manipulated in a CLONE method, because CLONE runs before the
14047 unreferenced array is walked to find SVs still with SvREFCNT() == 0
14048 (and fix things up by giving each a reference via the temps stack).
14049 Instead, during CLONE, if the 0-referenced SV has SvREFCNT_inc() and
14050 then SvREFCNT_dec(), it will be cleaned up (and added to the free list)
14051 before the walk of unreferenced happens and a reference to that is SV
14052 added to the temps stack. At which point we have the same SV considered
14053 to be in use, and free to be re-used. Not good.
14055 if (dstr && !(param->flags & CLONEf_COPY_STACKS) && !SvREFCNT(dstr)) {
14056 assert(param->unreferenced);
14057 av_push(param->unreferenced, SvREFCNT_inc(dstr));
14063 /* duplicate a context */
14066 Perl_cx_dup(pTHX_ PERL_CONTEXT *cxs, I32 ix, I32 max, CLONE_PARAMS* param)
14068 PERL_CONTEXT *ncxs;
14070 PERL_ARGS_ASSERT_CX_DUP;
14073 return (PERL_CONTEXT*)NULL;
14075 /* look for it in the table first */
14076 ncxs = (PERL_CONTEXT*)ptr_table_fetch(PL_ptr_table, cxs);
14080 /* create anew and remember what it is */
14081 Newx(ncxs, max + 1, PERL_CONTEXT);
14082 ptr_table_store(PL_ptr_table, cxs, ncxs);
14083 Copy(cxs, ncxs, max + 1, PERL_CONTEXT);
14086 PERL_CONTEXT * const ncx = &ncxs[ix];
14087 if (CxTYPE(ncx) == CXt_SUBST) {
14088 Perl_croak(aTHX_ "Cloning substitution context is unimplemented");
14091 ncx->blk_oldcop = (COP*)any_dup(ncx->blk_oldcop, param->proto_perl);
14092 switch (CxTYPE(ncx)) {
14094 ncx->blk_sub.cv = cv_dup_inc(ncx->blk_sub.cv, param);
14095 if(CxHASARGS(ncx)){
14096 ncx->blk_sub.savearray = av_dup_inc(ncx->blk_sub.savearray,param);
14098 ncx->blk_sub.savearray = NULL;
14100 ncx->blk_sub.prevcomppad = (PAD*)ptr_table_fetch(PL_ptr_table,
14101 ncx->blk_sub.prevcomppad);
14104 ncx->blk_eval.old_namesv = sv_dup_inc(ncx->blk_eval.old_namesv,
14106 /* XXX should this sv_dup_inc? Or only if SvSCREAM ???? */
14107 ncx->blk_eval.cur_text = sv_dup(ncx->blk_eval.cur_text, param);
14108 ncx->blk_eval.cv = cv_dup(ncx->blk_eval.cv, param);
14109 /* XXX what do do with cur_top_env ???? */
14111 case CXt_LOOP_LAZYSV:
14112 ncx->blk_loop.state_u.lazysv.end
14113 = sv_dup_inc(ncx->blk_loop.state_u.lazysv.end, param);
14114 /* Fallthrough: duplicate lazysv.cur by using the ary.ary
14115 duplication code instead.
14116 We are taking advantage of (1) av_dup_inc and sv_dup_inc
14117 actually being the same function, and (2) order
14118 equivalence of the two unions.
14119 We can assert the later [but only at run time :-(] */
14120 assert ((void *) &ncx->blk_loop.state_u.ary.ary ==
14121 (void *) &ncx->blk_loop.state_u.lazysv.cur);
14124 ncx->blk_loop.state_u.ary.ary
14125 = av_dup_inc(ncx->blk_loop.state_u.ary.ary, param);
14127 case CXt_LOOP_LIST:
14128 case CXt_LOOP_LAZYIV:
14129 /* code common to all 'for' CXt_LOOP_* types */
14130 ncx->blk_loop.itersave =
14131 sv_dup_inc(ncx->blk_loop.itersave, param);
14132 if (CxPADLOOP(ncx)) {
14133 PADOFFSET off = ncx->blk_loop.itervar_u.svp
14134 - &CX_CURPAD_SV(ncx->blk_loop, 0);
14135 ncx->blk_loop.oldcomppad =
14136 (PAD*)ptr_table_fetch(PL_ptr_table,
14137 ncx->blk_loop.oldcomppad);
14138 ncx->blk_loop.itervar_u.svp =
14139 &CX_CURPAD_SV(ncx->blk_loop, off);
14142 /* this copies the GV if CXp_FOR_GV, or the SV for an
14143 * alias (for \$x (...)) - relies on gv_dup being the
14144 * same as sv_dup */
14145 ncx->blk_loop.itervar_u.gv
14146 = gv_dup((const GV *)ncx->blk_loop.itervar_u.gv,
14150 case CXt_LOOP_PLAIN:
14153 ncx->blk_format.prevcomppad =
14154 (PAD*)ptr_table_fetch(PL_ptr_table,
14155 ncx->blk_format.prevcomppad);
14156 ncx->blk_format.cv = cv_dup_inc(ncx->blk_format.cv, param);
14157 ncx->blk_format.gv = gv_dup(ncx->blk_format.gv, param);
14158 ncx->blk_format.dfoutgv = gv_dup_inc(ncx->blk_format.dfoutgv,
14162 ncx->blk_givwhen.defsv_save =
14163 sv_dup_inc(ncx->blk_givwhen.defsv_save, param);
14176 /* duplicate a stack info structure */
14179 Perl_si_dup(pTHX_ PERL_SI *si, CLONE_PARAMS* param)
14183 PERL_ARGS_ASSERT_SI_DUP;
14186 return (PERL_SI*)NULL;
14188 /* look for it in the table first */
14189 nsi = (PERL_SI*)ptr_table_fetch(PL_ptr_table, si);
14193 /* create anew and remember what it is */
14194 Newxz(nsi, 1, PERL_SI);
14195 ptr_table_store(PL_ptr_table, si, nsi);
14197 nsi->si_stack = av_dup_inc(si->si_stack, param);
14198 nsi->si_cxix = si->si_cxix;
14199 nsi->si_cxmax = si->si_cxmax;
14200 nsi->si_cxstack = cx_dup(si->si_cxstack, si->si_cxix, si->si_cxmax, param);
14201 nsi->si_type = si->si_type;
14202 nsi->si_prev = si_dup(si->si_prev, param);
14203 nsi->si_next = si_dup(si->si_next, param);
14204 nsi->si_markoff = si->si_markoff;
14209 #define POPINT(ss,ix) ((ss)[--(ix)].any_i32)
14210 #define TOPINT(ss,ix) ((ss)[ix].any_i32)
14211 #define POPLONG(ss,ix) ((ss)[--(ix)].any_long)
14212 #define TOPLONG(ss,ix) ((ss)[ix].any_long)
14213 #define POPIV(ss,ix) ((ss)[--(ix)].any_iv)
14214 #define TOPIV(ss,ix) ((ss)[ix].any_iv)
14215 #define POPUV(ss,ix) ((ss)[--(ix)].any_uv)
14216 #define TOPUV(ss,ix) ((ss)[ix].any_uv)
14217 #define POPBOOL(ss,ix) ((ss)[--(ix)].any_bool)
14218 #define TOPBOOL(ss,ix) ((ss)[ix].any_bool)
14219 #define POPPTR(ss,ix) ((ss)[--(ix)].any_ptr)
14220 #define TOPPTR(ss,ix) ((ss)[ix].any_ptr)
14221 #define POPDPTR(ss,ix) ((ss)[--(ix)].any_dptr)
14222 #define TOPDPTR(ss,ix) ((ss)[ix].any_dptr)
14223 #define POPDXPTR(ss,ix) ((ss)[--(ix)].any_dxptr)
14224 #define TOPDXPTR(ss,ix) ((ss)[ix].any_dxptr)
14227 #define pv_dup_inc(p) SAVEPV(p)
14228 #define pv_dup(p) SAVEPV(p)
14229 #define svp_dup_inc(p,pp) any_dup(p,pp)
14231 /* map any object to the new equivent - either something in the
14232 * ptr table, or something in the interpreter structure
14236 Perl_any_dup(pTHX_ void *v, const PerlInterpreter *proto_perl)
14240 PERL_ARGS_ASSERT_ANY_DUP;
14243 return (void*)NULL;
14245 /* look for it in the table first */
14246 ret = ptr_table_fetch(PL_ptr_table, v);
14250 /* see if it is part of the interpreter structure */
14251 if (v >= (void*)proto_perl && v < (void*)(proto_perl+1))
14252 ret = (void*)(((char*)aTHX) + (((char*)v) - (char*)proto_perl));
14260 /* duplicate the save stack */
14263 Perl_ss_dup(pTHX_ PerlInterpreter *proto_perl, CLONE_PARAMS* param)
14266 ANY * const ss = proto_perl->Isavestack;
14267 const I32 max = proto_perl->Isavestack_max + SS_MAXPUSH;
14268 I32 ix = proto_perl->Isavestack_ix;
14281 void (*dptr) (void*);
14282 void (*dxptr) (pTHX_ void*);
14284 PERL_ARGS_ASSERT_SS_DUP;
14286 Newxz(nss, max, ANY);
14289 const UV uv = POPUV(ss,ix);
14290 const U8 type = (U8)uv & SAVE_MASK;
14292 TOPUV(nss,ix) = uv;
14294 case SAVEt_CLEARSV:
14295 case SAVEt_CLEARPADRANGE:
14297 case SAVEt_HELEM: /* hash element */
14298 case SAVEt_SV: /* scalar reference */
14299 sv = (const SV *)POPPTR(ss,ix);
14300 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14302 case SAVEt_ITEM: /* normal string */
14303 case SAVEt_GVSV: /* scalar slot in GV */
14304 sv = (const SV *)POPPTR(ss,ix);
14305 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14306 if (type == SAVEt_SV)
14310 case SAVEt_MORTALIZESV:
14311 case SAVEt_READONLY_OFF:
14312 sv = (const SV *)POPPTR(ss,ix);
14313 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14315 case SAVEt_FREEPADNAME:
14316 ptr = POPPTR(ss,ix);
14317 TOPPTR(nss,ix) = padname_dup((PADNAME *)ptr, param);
14318 PadnameREFCNT((PADNAME *)TOPPTR(nss,ix))++;
14320 case SAVEt_SHARED_PVREF: /* char* in shared space */
14321 c = (char*)POPPTR(ss,ix);
14322 TOPPTR(nss,ix) = savesharedpv(c);
14323 ptr = POPPTR(ss,ix);
14324 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14326 case SAVEt_GENERIC_SVREF: /* generic sv */
14327 case SAVEt_SVREF: /* scalar reference */
14328 sv = (const SV *)POPPTR(ss,ix);
14329 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14330 if (type == SAVEt_SVREF)
14331 SvREFCNT_inc_simple_void((SV *)TOPPTR(nss,ix));
14332 ptr = POPPTR(ss,ix);
14333 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14335 case SAVEt_GVSLOT: /* any slot in GV */
14336 sv = (const SV *)POPPTR(ss,ix);
14337 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14338 ptr = POPPTR(ss,ix);
14339 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14340 sv = (const SV *)POPPTR(ss,ix);
14341 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14343 case SAVEt_HV: /* hash reference */
14344 case SAVEt_AV: /* array reference */
14345 sv = (const SV *) POPPTR(ss,ix);
14346 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14348 case SAVEt_COMPPAD:
14350 sv = (const SV *) POPPTR(ss,ix);
14351 TOPPTR(nss,ix) = sv_dup(sv, param);
14353 case SAVEt_INT: /* int reference */
14354 ptr = POPPTR(ss,ix);
14355 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14356 intval = (int)POPINT(ss,ix);
14357 TOPINT(nss,ix) = intval;
14359 case SAVEt_LONG: /* long reference */
14360 ptr = POPPTR(ss,ix);
14361 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14362 longval = (long)POPLONG(ss,ix);
14363 TOPLONG(nss,ix) = longval;
14365 case SAVEt_I32: /* I32 reference */
14366 ptr = POPPTR(ss,ix);
14367 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14369 TOPINT(nss,ix) = i;
14371 case SAVEt_IV: /* IV reference */
14372 case SAVEt_STRLEN: /* STRLEN/size_t ref */
14373 ptr = POPPTR(ss,ix);
14374 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14376 TOPIV(nss,ix) = iv;
14378 case SAVEt_TMPSFLOOR:
14380 TOPIV(nss,ix) = iv;
14382 case SAVEt_HPTR: /* HV* reference */
14383 case SAVEt_APTR: /* AV* reference */
14384 case SAVEt_SPTR: /* SV* reference */
14385 ptr = POPPTR(ss,ix);
14386 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14387 sv = (const SV *)POPPTR(ss,ix);
14388 TOPPTR(nss,ix) = sv_dup(sv, param);
14390 case SAVEt_VPTR: /* random* reference */
14391 ptr = POPPTR(ss,ix);
14392 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14394 case SAVEt_INT_SMALL:
14395 case SAVEt_I32_SMALL:
14396 case SAVEt_I16: /* I16 reference */
14397 case SAVEt_I8: /* I8 reference */
14399 ptr = POPPTR(ss,ix);
14400 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14402 case SAVEt_GENERIC_PVREF: /* generic char* */
14403 case SAVEt_PPTR: /* char* reference */
14404 ptr = POPPTR(ss,ix);
14405 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14406 c = (char*)POPPTR(ss,ix);
14407 TOPPTR(nss,ix) = pv_dup(c);
14409 case SAVEt_GP: /* scalar reference */
14410 gp = (GP*)POPPTR(ss,ix);
14411 TOPPTR(nss,ix) = gp = gp_dup(gp, param);
14412 (void)GpREFCNT_inc(gp);
14413 gv = (const GV *)POPPTR(ss,ix);
14414 TOPPTR(nss,ix) = gv_dup_inc(gv, param);
14417 ptr = POPPTR(ss,ix);
14418 if (ptr && (((OP*)ptr)->op_private & OPpREFCOUNTED)) {
14419 /* these are assumed to be refcounted properly */
14421 switch (((OP*)ptr)->op_type) {
14423 case OP_LEAVESUBLV:
14427 case OP_LEAVEWRITE:
14428 TOPPTR(nss,ix) = ptr;
14431 (void) OpREFCNT_inc(o);
14435 TOPPTR(nss,ix) = NULL;
14440 TOPPTR(nss,ix) = NULL;
14442 case SAVEt_FREECOPHH:
14443 ptr = POPPTR(ss,ix);
14444 TOPPTR(nss,ix) = cophh_copy((COPHH *)ptr);
14446 case SAVEt_ADELETE:
14447 av = (const AV *)POPPTR(ss,ix);
14448 TOPPTR(nss,ix) = av_dup_inc(av, param);
14450 TOPINT(nss,ix) = i;
14453 hv = (const HV *)POPPTR(ss,ix);
14454 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14456 TOPINT(nss,ix) = i;
14459 c = (char*)POPPTR(ss,ix);
14460 TOPPTR(nss,ix) = pv_dup_inc(c);
14462 case SAVEt_STACK_POS: /* Position on Perl stack */
14464 TOPINT(nss,ix) = i;
14466 case SAVEt_DESTRUCTOR:
14467 ptr = POPPTR(ss,ix);
14468 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14469 dptr = POPDPTR(ss,ix);
14470 TOPDPTR(nss,ix) = DPTR2FPTR(void (*)(void*),
14471 any_dup(FPTR2DPTR(void *, dptr),
14474 case SAVEt_DESTRUCTOR_X:
14475 ptr = POPPTR(ss,ix);
14476 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14477 dxptr = POPDXPTR(ss,ix);
14478 TOPDXPTR(nss,ix) = DPTR2FPTR(void (*)(pTHX_ void*),
14479 any_dup(FPTR2DPTR(void *, dxptr),
14482 case SAVEt_REGCONTEXT:
14484 ix -= uv >> SAVE_TIGHT_SHIFT;
14486 case SAVEt_AELEM: /* array element */
14487 sv = (const SV *)POPPTR(ss,ix);
14488 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14490 TOPINT(nss,ix) = i;
14491 av = (const AV *)POPPTR(ss,ix);
14492 TOPPTR(nss,ix) = av_dup_inc(av, param);
14495 ptr = POPPTR(ss,ix);
14496 TOPPTR(nss,ix) = ptr;
14499 ptr = POPPTR(ss,ix);
14500 ptr = cophh_copy((COPHH*)ptr);
14501 TOPPTR(nss,ix) = ptr;
14503 TOPINT(nss,ix) = i;
14504 if (i & HINT_LOCALIZE_HH) {
14505 hv = (const HV *)POPPTR(ss,ix);
14506 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14509 case SAVEt_PADSV_AND_MORTALIZE:
14510 longval = (long)POPLONG(ss,ix);
14511 TOPLONG(nss,ix) = longval;
14512 ptr = POPPTR(ss,ix);
14513 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14514 sv = (const SV *)POPPTR(ss,ix);
14515 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14517 case SAVEt_SET_SVFLAGS:
14519 TOPINT(nss,ix) = i;
14521 TOPINT(nss,ix) = i;
14522 sv = (const SV *)POPPTR(ss,ix);
14523 TOPPTR(nss,ix) = sv_dup(sv, param);
14525 case SAVEt_COMPILE_WARNINGS:
14526 ptr = POPPTR(ss,ix);
14527 TOPPTR(nss,ix) = DUP_WARNINGS((STRLEN*)ptr);
14530 ptr = POPPTR(ss,ix);
14531 TOPPTR(nss,ix) = parser_dup((const yy_parser*)ptr, param);
14535 "panic: ss_dup inconsistency (%"IVdf")", (IV) type);
14543 /* if sv is a stash, call $class->CLONE_SKIP(), and set the SVphv_CLONEABLE
14544 * flag to the result. This is done for each stash before cloning starts,
14545 * so we know which stashes want their objects cloned */
14548 do_mark_cloneable_stash(pTHX_ SV *const sv)
14550 const HEK * const hvname = HvNAME_HEK((const HV *)sv);
14552 GV* const cloner = gv_fetchmethod_autoload(MUTABLE_HV(sv), "CLONE_SKIP", 0);
14553 SvFLAGS(sv) |= SVphv_CLONEABLE; /* clone objects by default */
14554 if (cloner && GvCV(cloner)) {
14561 mXPUSHs(newSVhek(hvname));
14563 call_sv(MUTABLE_SV(GvCV(cloner)), G_SCALAR);
14570 SvFLAGS(sv) &= ~SVphv_CLONEABLE;
14578 =for apidoc perl_clone
14580 Create and return a new interpreter by cloning the current one.
14582 C<perl_clone> takes these flags as parameters:
14584 C<CLONEf_COPY_STACKS> - is used to, well, copy the stacks also,
14585 without it we only clone the data and zero the stacks,
14586 with it we copy the stacks and the new perl interpreter is
14587 ready to run at the exact same point as the previous one.
14588 The pseudo-fork code uses C<COPY_STACKS> while the
14589 threads->create doesn't.
14591 C<CLONEf_KEEP_PTR_TABLE> -
14592 C<perl_clone> keeps a ptr_table with the pointer of the old
14593 variable as a key and the new variable as a value,
14594 this allows it to check if something has been cloned and not
14595 clone it again but rather just use the value and increase the
14596 refcount. If C<KEEP_PTR_TABLE> is not set then C<perl_clone> will kill
14597 the ptr_table using the function
14598 C<ptr_table_free(PL_ptr_table); PL_ptr_table = NULL;>,
14599 reason to keep it around is if you want to dup some of your own
14600 variable who are outside the graph perl scans, an example of this
14601 code is in F<threads.xs> create.
14603 C<CLONEf_CLONE_HOST> -
14604 This is a win32 thing, it is ignored on unix, it tells perls
14605 win32host code (which is c++) to clone itself, this is needed on
14606 win32 if you want to run two threads at the same time,
14607 if you just want to do some stuff in a separate perl interpreter
14608 and then throw it away and return to the original one,
14609 you don't need to do anything.
14614 /* XXX the above needs expanding by someone who actually understands it ! */
14615 EXTERN_C PerlInterpreter *
14616 perl_clone_host(PerlInterpreter* proto_perl, UV flags);
14619 perl_clone(PerlInterpreter *proto_perl, UV flags)
14622 #ifdef PERL_IMPLICIT_SYS
14624 PERL_ARGS_ASSERT_PERL_CLONE;
14626 /* perlhost.h so we need to call into it
14627 to clone the host, CPerlHost should have a c interface, sky */
14629 #ifndef __amigaos4__
14630 if (flags & CLONEf_CLONE_HOST) {
14631 return perl_clone_host(proto_perl,flags);
14634 return perl_clone_using(proto_perl, flags,
14636 proto_perl->IMemShared,
14637 proto_perl->IMemParse,
14639 proto_perl->IStdIO,
14643 proto_perl->IProc);
14647 perl_clone_using(PerlInterpreter *proto_perl, UV flags,
14648 struct IPerlMem* ipM, struct IPerlMem* ipMS,
14649 struct IPerlMem* ipMP, struct IPerlEnv* ipE,
14650 struct IPerlStdIO* ipStd, struct IPerlLIO* ipLIO,
14651 struct IPerlDir* ipD, struct IPerlSock* ipS,
14652 struct IPerlProc* ipP)
14654 /* XXX many of the string copies here can be optimized if they're
14655 * constants; they need to be allocated as common memory and just
14656 * their pointers copied. */
14659 CLONE_PARAMS clone_params;
14660 CLONE_PARAMS* const param = &clone_params;
14662 PerlInterpreter * const my_perl = (PerlInterpreter*)(*ipM->pMalloc)(ipM, sizeof(PerlInterpreter));
14664 PERL_ARGS_ASSERT_PERL_CLONE_USING;
14665 #else /* !PERL_IMPLICIT_SYS */
14667 CLONE_PARAMS clone_params;
14668 CLONE_PARAMS* param = &clone_params;
14669 PerlInterpreter * const my_perl = (PerlInterpreter*)PerlMem_malloc(sizeof(PerlInterpreter));
14671 PERL_ARGS_ASSERT_PERL_CLONE;
14672 #endif /* PERL_IMPLICIT_SYS */
14674 /* for each stash, determine whether its objects should be cloned */
14675 S_visit(proto_perl, do_mark_cloneable_stash, SVt_PVHV, SVTYPEMASK);
14676 PERL_SET_THX(my_perl);
14679 PoisonNew(my_perl, 1, PerlInterpreter);
14682 PL_defstash = NULL; /* may be used by perl malloc() */
14685 PL_scopestack_name = 0;
14687 PL_savestack_ix = 0;
14688 PL_savestack_max = -1;
14689 PL_sig_pending = 0;
14691 Zero(&PL_debug_pad, 1, struct perl_debug_pad);
14692 Zero(&PL_padname_undef, 1, PADNAME);
14693 Zero(&PL_padname_const, 1, PADNAME);
14694 # ifdef DEBUG_LEAKING_SCALARS
14695 PL_sv_serial = (((UV)my_perl >> 2) & 0xfff) * 1000000;
14697 # ifdef PERL_TRACE_OPS
14698 Zero(PL_op_exec_cnt, OP_max+2, UV);
14700 #else /* !DEBUGGING */
14701 Zero(my_perl, 1, PerlInterpreter);
14702 #endif /* DEBUGGING */
14704 #ifdef PERL_IMPLICIT_SYS
14705 /* host pointers */
14707 PL_MemShared = ipMS;
14708 PL_MemParse = ipMP;
14715 #endif /* PERL_IMPLICIT_SYS */
14718 param->flags = flags;
14719 /* Nothing in the core code uses this, but we make it available to
14720 extensions (using mg_dup). */
14721 param->proto_perl = proto_perl;
14722 /* Likely nothing will use this, but it is initialised to be consistent
14723 with Perl_clone_params_new(). */
14724 param->new_perl = my_perl;
14725 param->unreferenced = NULL;
14728 INIT_TRACK_MEMPOOL(my_perl->Imemory_debug_header, my_perl);
14730 PL_body_arenas = NULL;
14731 Zero(&PL_body_roots, 1, PL_body_roots);
14735 PL_sv_arenaroot = NULL;
14737 PL_debug = proto_perl->Idebug;
14739 /* dbargs array probably holds garbage */
14742 PL_compiling = proto_perl->Icompiling;
14744 /* pseudo environmental stuff */
14745 PL_origargc = proto_perl->Iorigargc;
14746 PL_origargv = proto_perl->Iorigargv;
14748 #ifndef NO_TAINT_SUPPORT
14749 /* Set tainting stuff before PerlIO_debug can possibly get called */
14750 PL_tainting = proto_perl->Itainting;
14751 PL_taint_warn = proto_perl->Itaint_warn;
14753 PL_tainting = FALSE;
14754 PL_taint_warn = FALSE;
14757 PL_minus_c = proto_perl->Iminus_c;
14759 PL_localpatches = proto_perl->Ilocalpatches;
14760 PL_splitstr = proto_perl->Isplitstr;
14761 PL_minus_n = proto_perl->Iminus_n;
14762 PL_minus_p = proto_perl->Iminus_p;
14763 PL_minus_l = proto_perl->Iminus_l;
14764 PL_minus_a = proto_perl->Iminus_a;
14765 PL_minus_E = proto_perl->Iminus_E;
14766 PL_minus_F = proto_perl->Iminus_F;
14767 PL_doswitches = proto_perl->Idoswitches;
14768 PL_dowarn = proto_perl->Idowarn;
14769 #ifdef PERL_SAWAMPERSAND
14770 PL_sawampersand = proto_perl->Isawampersand;
14772 PL_unsafe = proto_perl->Iunsafe;
14773 PL_perldb = proto_perl->Iperldb;
14774 PL_perl_destruct_level = proto_perl->Iperl_destruct_level;
14775 PL_exit_flags = proto_perl->Iexit_flags;
14777 /* XXX time(&PL_basetime) when asked for? */
14778 PL_basetime = proto_perl->Ibasetime;
14780 PL_maxsysfd = proto_perl->Imaxsysfd;
14781 PL_statusvalue = proto_perl->Istatusvalue;
14783 PL_statusvalue_vms = proto_perl->Istatusvalue_vms;
14785 PL_statusvalue_posix = proto_perl->Istatusvalue_posix;
14788 /* RE engine related */
14789 PL_regmatch_slab = NULL;
14790 PL_reg_curpm = NULL;
14792 PL_sub_generation = proto_perl->Isub_generation;
14794 /* funky return mechanisms */
14795 PL_forkprocess = proto_perl->Iforkprocess;
14797 /* internal state */
14798 PL_main_start = proto_perl->Imain_start;
14799 PL_eval_root = proto_perl->Ieval_root;
14800 PL_eval_start = proto_perl->Ieval_start;
14802 PL_filemode = proto_perl->Ifilemode;
14803 PL_lastfd = proto_perl->Ilastfd;
14804 PL_oldname = proto_perl->Ioldname; /* XXX not quite right */
14807 PL_gensym = proto_perl->Igensym;
14809 PL_laststatval = proto_perl->Ilaststatval;
14810 PL_laststype = proto_perl->Ilaststype;
14813 PL_profiledata = NULL;
14815 PL_generation = proto_perl->Igeneration;
14817 PL_in_clean_objs = proto_perl->Iin_clean_objs;
14818 PL_in_clean_all = proto_perl->Iin_clean_all;
14820 PL_delaymagic_uid = proto_perl->Idelaymagic_uid;
14821 PL_delaymagic_euid = proto_perl->Idelaymagic_euid;
14822 PL_delaymagic_gid = proto_perl->Idelaymagic_gid;
14823 PL_delaymagic_egid = proto_perl->Idelaymagic_egid;
14824 PL_nomemok = proto_perl->Inomemok;
14825 PL_an = proto_perl->Ian;
14826 PL_evalseq = proto_perl->Ievalseq;
14827 PL_origenviron = proto_perl->Iorigenviron; /* XXX not quite right */
14828 PL_origalen = proto_perl->Iorigalen;
14830 PL_sighandlerp = proto_perl->Isighandlerp;
14832 PL_runops = proto_perl->Irunops;
14834 PL_subline = proto_perl->Isubline;
14836 PL_cv_has_eval = proto_perl->Icv_has_eval;
14839 PL_cryptseen = proto_perl->Icryptseen;
14842 #ifdef USE_LOCALE_COLLATE
14843 PL_collation_ix = proto_perl->Icollation_ix;
14844 PL_collation_standard = proto_perl->Icollation_standard;
14845 PL_collxfrm_base = proto_perl->Icollxfrm_base;
14846 PL_collxfrm_mult = proto_perl->Icollxfrm_mult;
14847 PL_strxfrm_max_cp = proto_perl->Istrxfrm_max_cp;
14848 #endif /* USE_LOCALE_COLLATE */
14850 #ifdef USE_LOCALE_NUMERIC
14851 PL_numeric_standard = proto_perl->Inumeric_standard;
14852 PL_numeric_local = proto_perl->Inumeric_local;
14853 #endif /* !USE_LOCALE_NUMERIC */
14855 /* Did the locale setup indicate UTF-8? */
14856 PL_utf8locale = proto_perl->Iutf8locale;
14857 PL_in_utf8_CTYPE_locale = proto_perl->Iin_utf8_CTYPE_locale;
14858 PL_in_utf8_COLLATE_locale = proto_perl->Iin_utf8_COLLATE_locale;
14859 /* Unicode features (see perlrun/-C) */
14860 PL_unicode = proto_perl->Iunicode;
14862 /* Pre-5.8 signals control */
14863 PL_signals = proto_perl->Isignals;
14865 /* times() ticks per second */
14866 PL_clocktick = proto_perl->Iclocktick;
14868 /* Recursion stopper for PerlIO_find_layer */
14869 PL_in_load_module = proto_perl->Iin_load_module;
14871 /* sort() routine */
14872 PL_sort_RealCmp = proto_perl->Isort_RealCmp;
14874 /* Not really needed/useful since the reenrant_retint is "volatile",
14875 * but do it for consistency's sake. */
14876 PL_reentrant_retint = proto_perl->Ireentrant_retint;
14878 /* Hooks to shared SVs and locks. */
14879 PL_sharehook = proto_perl->Isharehook;
14880 PL_lockhook = proto_perl->Ilockhook;
14881 PL_unlockhook = proto_perl->Iunlockhook;
14882 PL_threadhook = proto_perl->Ithreadhook;
14883 PL_destroyhook = proto_perl->Idestroyhook;
14884 PL_signalhook = proto_perl->Isignalhook;
14886 PL_globhook = proto_perl->Iglobhook;
14889 PL_last_swash_hv = NULL; /* reinits on demand */
14890 PL_last_swash_klen = 0;
14891 PL_last_swash_key[0]= '\0';
14892 PL_last_swash_tmps = (U8*)NULL;
14893 PL_last_swash_slen = 0;
14895 PL_srand_called = proto_perl->Isrand_called;
14896 Copy(&(proto_perl->Irandom_state), &PL_random_state, 1, PL_RANDOM_STATE_TYPE);
14898 if (flags & CLONEf_COPY_STACKS) {
14899 /* next allocation will be PL_tmps_stack[PL_tmps_ix+1] */
14900 PL_tmps_ix = proto_perl->Itmps_ix;
14901 PL_tmps_max = proto_perl->Itmps_max;
14902 PL_tmps_floor = proto_perl->Itmps_floor;
14904 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
14905 * NOTE: unlike the others! */
14906 PL_scopestack_ix = proto_perl->Iscopestack_ix;
14907 PL_scopestack_max = proto_perl->Iscopestack_max;
14909 /* next SSPUSHFOO() sets PL_savestack[PL_savestack_ix]
14910 * NOTE: unlike the others! */
14911 PL_savestack_ix = proto_perl->Isavestack_ix;
14912 PL_savestack_max = proto_perl->Isavestack_max;
14915 PL_start_env = proto_perl->Istart_env; /* XXXXXX */
14916 PL_top_env = &PL_start_env;
14918 PL_op = proto_perl->Iop;
14921 PL_Xpv = (XPV*)NULL;
14922 my_perl->Ina = proto_perl->Ina;
14924 PL_statbuf = proto_perl->Istatbuf;
14925 PL_statcache = proto_perl->Istatcache;
14927 #ifndef NO_TAINT_SUPPORT
14928 PL_tainted = proto_perl->Itainted;
14930 PL_tainted = FALSE;
14932 PL_curpm = proto_perl->Icurpm; /* XXX No PMOP ref count */
14934 PL_chopset = proto_perl->Ichopset; /* XXX never deallocated */
14936 PL_restartjmpenv = proto_perl->Irestartjmpenv;
14937 PL_restartop = proto_perl->Irestartop;
14938 PL_in_eval = proto_perl->Iin_eval;
14939 PL_delaymagic = proto_perl->Idelaymagic;
14940 PL_phase = proto_perl->Iphase;
14941 PL_localizing = proto_perl->Ilocalizing;
14943 PL_hv_fetch_ent_mh = NULL;
14944 PL_modcount = proto_perl->Imodcount;
14945 PL_lastgotoprobe = NULL;
14946 PL_dumpindent = proto_perl->Idumpindent;
14948 PL_efloatbuf = NULL; /* reinits on demand */
14949 PL_efloatsize = 0; /* reinits on demand */
14953 PL_colorset = 0; /* reinits PL_colors[] */
14954 /*PL_colors[6] = {0,0,0,0,0,0};*/
14956 /* Pluggable optimizer */
14957 PL_peepp = proto_perl->Ipeepp;
14958 PL_rpeepp = proto_perl->Irpeepp;
14959 /* op_free() hook */
14960 PL_opfreehook = proto_perl->Iopfreehook;
14962 #ifdef USE_REENTRANT_API
14963 /* XXX: things like -Dm will segfault here in perlio, but doing
14964 * PERL_SET_CONTEXT(proto_perl);
14965 * breaks too many other things
14967 Perl_reentrant_init(aTHX);
14970 /* create SV map for pointer relocation */
14971 PL_ptr_table = ptr_table_new();
14973 /* initialize these special pointers as early as possible */
14975 ptr_table_store(PL_ptr_table, &proto_perl->Isv_undef, &PL_sv_undef);
14976 ptr_table_store(PL_ptr_table, &proto_perl->Isv_no, &PL_sv_no);
14977 ptr_table_store(PL_ptr_table, &proto_perl->Isv_yes, &PL_sv_yes);
14978 ptr_table_store(PL_ptr_table, &proto_perl->Ipadname_const,
14979 &PL_padname_const);
14981 /* create (a non-shared!) shared string table */
14982 PL_strtab = newHV();
14983 HvSHAREKEYS_off(PL_strtab);
14984 hv_ksplit(PL_strtab, HvTOTALKEYS(proto_perl->Istrtab));
14985 ptr_table_store(PL_ptr_table, proto_perl->Istrtab, PL_strtab);
14987 Zero(PL_sv_consts, SV_CONSTS_COUNT, SV*);
14989 /* This PV will be free'd special way so must set it same way op.c does */
14990 PL_compiling.cop_file = savesharedpv(PL_compiling.cop_file);
14991 ptr_table_store(PL_ptr_table, proto_perl->Icompiling.cop_file, PL_compiling.cop_file);
14993 ptr_table_store(PL_ptr_table, &proto_perl->Icompiling, &PL_compiling);
14994 PL_compiling.cop_warnings = DUP_WARNINGS(PL_compiling.cop_warnings);
14995 CopHINTHASH_set(&PL_compiling, cophh_copy(CopHINTHASH_get(&PL_compiling)));
14996 PL_curcop = (COP*)any_dup(proto_perl->Icurcop, proto_perl);
14998 param->stashes = newAV(); /* Setup array of objects to call clone on */
14999 /* This makes no difference to the implementation, as it always pushes
15000 and shifts pointers to other SVs without changing their reference
15001 count, with the array becoming empty before it is freed. However, it
15002 makes it conceptually clear what is going on, and will avoid some
15003 work inside av.c, filling slots between AvFILL() and AvMAX() with
15004 &PL_sv_undef, and SvREFCNT_dec()ing those. */
15005 AvREAL_off(param->stashes);
15007 if (!(flags & CLONEf_COPY_STACKS)) {
15008 param->unreferenced = newAV();
15011 #ifdef PERLIO_LAYERS
15012 /* Clone PerlIO tables as soon as we can handle general xx_dup() */
15013 PerlIO_clone(aTHX_ proto_perl, param);
15016 PL_envgv = gv_dup_inc(proto_perl->Ienvgv, param);
15017 PL_incgv = gv_dup_inc(proto_perl->Iincgv, param);
15018 PL_hintgv = gv_dup_inc(proto_perl->Ihintgv, param);
15019 PL_origfilename = SAVEPV(proto_perl->Iorigfilename);
15020 PL_xsubfilename = proto_perl->Ixsubfilename;
15021 PL_diehook = sv_dup_inc(proto_perl->Idiehook, param);
15022 PL_warnhook = sv_dup_inc(proto_perl->Iwarnhook, param);
15025 PL_patchlevel = sv_dup_inc(proto_perl->Ipatchlevel, param);
15026 PL_inplace = SAVEPV(proto_perl->Iinplace);
15027 PL_e_script = sv_dup_inc(proto_perl->Ie_script, param);
15029 /* magical thingies */
15031 sv_setpvs(PERL_DEBUG_PAD(0), ""); /* For regex debugging. */
15032 sv_setpvs(PERL_DEBUG_PAD(1), ""); /* ext/re needs these */
15033 sv_setpvs(PERL_DEBUG_PAD(2), ""); /* even without DEBUGGING. */
15036 /* Clone the regex array */
15037 /* ORANGE FIXME for plugins, probably in the SV dup code.
15038 newSViv(PTR2IV(CALLREGDUPE(
15039 INT2PTR(REGEXP *, SvIVX(regex)), param))))
15041 PL_regex_padav = av_dup_inc(proto_perl->Iregex_padav, param);
15042 PL_regex_pad = AvARRAY(PL_regex_padav);
15044 PL_stashpadmax = proto_perl->Istashpadmax;
15045 PL_stashpadix = proto_perl->Istashpadix ;
15046 Newx(PL_stashpad, PL_stashpadmax, HV *);
15049 for (; o < PL_stashpadmax; ++o)
15050 PL_stashpad[o] = hv_dup(proto_perl->Istashpad[o], param);
15053 /* shortcuts to various I/O objects */
15054 PL_ofsgv = gv_dup_inc(proto_perl->Iofsgv, param);
15055 PL_stdingv = gv_dup(proto_perl->Istdingv, param);
15056 PL_stderrgv = gv_dup(proto_perl->Istderrgv, param);
15057 PL_defgv = gv_dup(proto_perl->Idefgv, param);
15058 PL_argvgv = gv_dup_inc(proto_perl->Iargvgv, param);
15059 PL_argvoutgv = gv_dup(proto_perl->Iargvoutgv, param);
15060 PL_argvout_stack = av_dup_inc(proto_perl->Iargvout_stack, param);
15062 /* shortcuts to regexp stuff */
15063 PL_replgv = gv_dup_inc(proto_perl->Ireplgv, param);
15065 /* shortcuts to misc objects */
15066 PL_errgv = gv_dup(proto_perl->Ierrgv, param);
15068 /* shortcuts to debugging objects */
15069 PL_DBgv = gv_dup_inc(proto_perl->IDBgv, param);
15070 PL_DBline = gv_dup_inc(proto_perl->IDBline, param);
15071 PL_DBsub = gv_dup_inc(proto_perl->IDBsub, param);
15072 PL_DBsingle = sv_dup(proto_perl->IDBsingle, param);
15073 PL_DBtrace = sv_dup(proto_perl->IDBtrace, param);
15074 PL_DBsignal = sv_dup(proto_perl->IDBsignal, param);
15075 Copy(proto_perl->IDBcontrol, PL_DBcontrol, DBVARMG_COUNT, IV);
15077 /* symbol tables */
15078 PL_defstash = hv_dup_inc(proto_perl->Idefstash, param);
15079 PL_curstash = hv_dup_inc(proto_perl->Icurstash, param);
15080 PL_debstash = hv_dup(proto_perl->Idebstash, param);
15081 PL_globalstash = hv_dup(proto_perl->Iglobalstash, param);
15082 PL_curstname = sv_dup_inc(proto_perl->Icurstname, param);
15084 PL_beginav = av_dup_inc(proto_perl->Ibeginav, param);
15085 PL_beginav_save = av_dup_inc(proto_perl->Ibeginav_save, param);
15086 PL_checkav_save = av_dup_inc(proto_perl->Icheckav_save, param);
15087 PL_unitcheckav = av_dup_inc(proto_perl->Iunitcheckav, param);
15088 PL_unitcheckav_save = av_dup_inc(proto_perl->Iunitcheckav_save, param);
15089 PL_endav = av_dup_inc(proto_perl->Iendav, param);
15090 PL_checkav = av_dup_inc(proto_perl->Icheckav, param);
15091 PL_initav = av_dup_inc(proto_perl->Iinitav, param);
15092 PL_savebegin = proto_perl->Isavebegin;
15094 PL_isarev = hv_dup_inc(proto_perl->Iisarev, param);
15096 /* subprocess state */
15097 PL_fdpid = av_dup_inc(proto_perl->Ifdpid, param);
15099 if (proto_perl->Iop_mask)
15100 PL_op_mask = SAVEPVN(proto_perl->Iop_mask, PL_maxo);
15103 /* PL_asserting = proto_perl->Iasserting; */
15105 /* current interpreter roots */
15106 PL_main_cv = cv_dup_inc(proto_perl->Imain_cv, param);
15108 PL_main_root = OpREFCNT_inc(proto_perl->Imain_root);
15111 /* runtime control stuff */
15112 PL_curcopdb = (COP*)any_dup(proto_perl->Icurcopdb, proto_perl);
15114 PL_preambleav = av_dup_inc(proto_perl->Ipreambleav, param);
15116 PL_ors_sv = sv_dup_inc(proto_perl->Iors_sv, param);
15118 /* interpreter atexit processing */
15119 PL_exitlistlen = proto_perl->Iexitlistlen;
15120 if (PL_exitlistlen) {
15121 Newx(PL_exitlist, PL_exitlistlen, PerlExitListEntry);
15122 Copy(proto_perl->Iexitlist, PL_exitlist, PL_exitlistlen, PerlExitListEntry);
15125 PL_exitlist = (PerlExitListEntry*)NULL;
15127 PL_my_cxt_size = proto_perl->Imy_cxt_size;
15128 if (PL_my_cxt_size) {
15129 Newx(PL_my_cxt_list, PL_my_cxt_size, void *);
15130 Copy(proto_perl->Imy_cxt_list, PL_my_cxt_list, PL_my_cxt_size, void *);
15131 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
15132 Newx(PL_my_cxt_keys, PL_my_cxt_size, const char *);
15133 Copy(proto_perl->Imy_cxt_keys, PL_my_cxt_keys, PL_my_cxt_size, char *);
15137 PL_my_cxt_list = (void**)NULL;
15138 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
15139 PL_my_cxt_keys = (const char**)NULL;
15142 PL_modglobal = hv_dup_inc(proto_perl->Imodglobal, param);
15143 PL_custom_op_names = hv_dup_inc(proto_perl->Icustom_op_names,param);
15144 PL_custom_op_descs = hv_dup_inc(proto_perl->Icustom_op_descs,param);
15145 PL_custom_ops = hv_dup_inc(proto_perl->Icustom_ops, param);
15147 PL_compcv = cv_dup(proto_perl->Icompcv, param);
15149 PAD_CLONE_VARS(proto_perl, param);
15151 #ifdef HAVE_INTERP_INTERN
15152 sys_intern_dup(&proto_perl->Isys_intern, &PL_sys_intern);
15155 PL_DBcv = cv_dup(proto_perl->IDBcv, param);
15157 #ifdef PERL_USES_PL_PIDSTATUS
15158 PL_pidstatus = newHV(); /* XXX flag for cloning? */
15160 PL_osname = SAVEPV(proto_perl->Iosname);
15161 PL_parser = parser_dup(proto_perl->Iparser, param);
15163 /* XXX this only works if the saved cop has already been cloned */
15164 if (proto_perl->Iparser) {
15165 PL_parser->saved_curcop = (COP*)any_dup(
15166 proto_perl->Iparser->saved_curcop,
15170 PL_subname = sv_dup_inc(proto_perl->Isubname, param);
15172 #ifdef USE_LOCALE_CTYPE
15173 /* Should we warn if uses locale? */
15174 PL_warn_locale = sv_dup_inc(proto_perl->Iwarn_locale, param);
15177 #ifdef USE_LOCALE_COLLATE
15178 PL_collation_name = SAVEPV(proto_perl->Icollation_name);
15179 #endif /* USE_LOCALE_COLLATE */
15181 #ifdef USE_LOCALE_NUMERIC
15182 PL_numeric_name = SAVEPV(proto_perl->Inumeric_name);
15183 PL_numeric_radix_sv = sv_dup_inc(proto_perl->Inumeric_radix_sv, param);
15184 #endif /* !USE_LOCALE_NUMERIC */
15186 /* Unicode inversion lists */
15187 PL_Latin1 = sv_dup_inc(proto_perl->ILatin1, param);
15188 PL_UpperLatin1 = sv_dup_inc(proto_perl->IUpperLatin1, param);
15189 PL_AboveLatin1 = sv_dup_inc(proto_perl->IAboveLatin1, param);
15190 PL_InBitmap = sv_dup_inc(proto_perl->IInBitmap, param);
15192 PL_NonL1NonFinalFold = sv_dup_inc(proto_perl->INonL1NonFinalFold, param);
15193 PL_HasMultiCharFold = sv_dup_inc(proto_perl->IHasMultiCharFold, param);
15195 /* utf8 character class swashes */
15196 for (i = 0; i < POSIX_SWASH_COUNT; i++) {
15197 PL_utf8_swash_ptrs[i] = sv_dup_inc(proto_perl->Iutf8_swash_ptrs[i], param);
15199 for (i = 0; i < POSIX_CC_COUNT; i++) {
15200 PL_XPosix_ptrs[i] = sv_dup_inc(proto_perl->IXPosix_ptrs[i], param);
15202 PL_GCB_invlist = sv_dup_inc(proto_perl->IGCB_invlist, param);
15203 PL_SB_invlist = sv_dup_inc(proto_perl->ISB_invlist, param);
15204 PL_WB_invlist = sv_dup_inc(proto_perl->IWB_invlist, param);
15205 PL_utf8_mark = sv_dup_inc(proto_perl->Iutf8_mark, param);
15206 PL_utf8_toupper = sv_dup_inc(proto_perl->Iutf8_toupper, param);
15207 PL_utf8_totitle = sv_dup_inc(proto_perl->Iutf8_totitle, param);
15208 PL_utf8_tolower = sv_dup_inc(proto_perl->Iutf8_tolower, param);
15209 PL_utf8_tofold = sv_dup_inc(proto_perl->Iutf8_tofold, param);
15210 PL_utf8_idstart = sv_dup_inc(proto_perl->Iutf8_idstart, param);
15211 PL_utf8_xidstart = sv_dup_inc(proto_perl->Iutf8_xidstart, param);
15212 PL_utf8_perl_idstart = sv_dup_inc(proto_perl->Iutf8_perl_idstart, param);
15213 PL_utf8_perl_idcont = sv_dup_inc(proto_perl->Iutf8_perl_idcont, param);
15214 PL_utf8_idcont = sv_dup_inc(proto_perl->Iutf8_idcont, param);
15215 PL_utf8_xidcont = sv_dup_inc(proto_perl->Iutf8_xidcont, param);
15216 PL_utf8_foldable = sv_dup_inc(proto_perl->Iutf8_foldable, param);
15217 PL_utf8_charname_begin = sv_dup_inc(proto_perl->Iutf8_charname_begin, param);
15218 PL_utf8_charname_continue = sv_dup_inc(proto_perl->Iutf8_charname_continue, param);
15220 if (proto_perl->Ipsig_pend) {
15221 Newxz(PL_psig_pend, SIG_SIZE, int);
15224 PL_psig_pend = (int*)NULL;
15227 if (proto_perl->Ipsig_name) {
15228 Newx(PL_psig_name, 2 * SIG_SIZE, SV*);
15229 sv_dup_inc_multiple(proto_perl->Ipsig_name, PL_psig_name, 2 * SIG_SIZE,
15231 PL_psig_ptr = PL_psig_name + SIG_SIZE;
15234 PL_psig_ptr = (SV**)NULL;
15235 PL_psig_name = (SV**)NULL;
15238 if (flags & CLONEf_COPY_STACKS) {
15239 Newx(PL_tmps_stack, PL_tmps_max, SV*);
15240 sv_dup_inc_multiple(proto_perl->Itmps_stack, PL_tmps_stack,
15241 PL_tmps_ix+1, param);
15243 /* next PUSHMARK() sets *(PL_markstack_ptr+1) */
15244 i = proto_perl->Imarkstack_max - proto_perl->Imarkstack;
15245 Newxz(PL_markstack, i, I32);
15246 PL_markstack_max = PL_markstack + (proto_perl->Imarkstack_max
15247 - proto_perl->Imarkstack);
15248 PL_markstack_ptr = PL_markstack + (proto_perl->Imarkstack_ptr
15249 - proto_perl->Imarkstack);
15250 Copy(proto_perl->Imarkstack, PL_markstack,
15251 PL_markstack_ptr - PL_markstack + 1, I32);
15253 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
15254 * NOTE: unlike the others! */
15255 Newxz(PL_scopestack, PL_scopestack_max, I32);
15256 Copy(proto_perl->Iscopestack, PL_scopestack, PL_scopestack_ix, I32);
15259 Newxz(PL_scopestack_name, PL_scopestack_max, const char *);
15260 Copy(proto_perl->Iscopestack_name, PL_scopestack_name, PL_scopestack_ix, const char *);
15262 /* reset stack AV to correct length before its duped via
15263 * PL_curstackinfo */
15264 AvFILLp(proto_perl->Icurstack) =
15265 proto_perl->Istack_sp - proto_perl->Istack_base;
15267 /* NOTE: si_dup() looks at PL_markstack */
15268 PL_curstackinfo = si_dup(proto_perl->Icurstackinfo, param);
15270 /* PL_curstack = PL_curstackinfo->si_stack; */
15271 PL_curstack = av_dup(proto_perl->Icurstack, param);
15272 PL_mainstack = av_dup(proto_perl->Imainstack, param);
15274 /* next PUSHs() etc. set *(PL_stack_sp+1) */
15275 PL_stack_base = AvARRAY(PL_curstack);
15276 PL_stack_sp = PL_stack_base + (proto_perl->Istack_sp
15277 - proto_perl->Istack_base);
15278 PL_stack_max = PL_stack_base + AvMAX(PL_curstack);
15280 /*Newxz(PL_savestack, PL_savestack_max, ANY);*/
15281 PL_savestack = ss_dup(proto_perl, param);
15285 ENTER; /* perl_destruct() wants to LEAVE; */
15288 PL_statgv = gv_dup(proto_perl->Istatgv, param);
15289 PL_statname = sv_dup_inc(proto_perl->Istatname, param);
15291 PL_rs = sv_dup_inc(proto_perl->Irs, param);
15292 PL_last_in_gv = gv_dup(proto_perl->Ilast_in_gv, param);
15293 PL_defoutgv = gv_dup_inc(proto_perl->Idefoutgv, param);
15294 PL_toptarget = sv_dup_inc(proto_perl->Itoptarget, param);
15295 PL_bodytarget = sv_dup_inc(proto_perl->Ibodytarget, param);
15296 PL_formtarget = sv_dup(proto_perl->Iformtarget, param);
15298 PL_errors = sv_dup_inc(proto_perl->Ierrors, param);
15300 PL_sortcop = (OP*)any_dup(proto_perl->Isortcop, proto_perl);
15301 PL_firstgv = gv_dup_inc(proto_perl->Ifirstgv, param);
15302 PL_secondgv = gv_dup_inc(proto_perl->Isecondgv, param);
15304 PL_stashcache = newHV();
15306 PL_watchaddr = (char **) ptr_table_fetch(PL_ptr_table,
15307 proto_perl->Iwatchaddr);
15308 PL_watchok = PL_watchaddr ? * PL_watchaddr : NULL;
15309 if (PL_debug && PL_watchaddr) {
15310 PerlIO_printf(Perl_debug_log,
15311 "WATCHING: %"UVxf" cloned as %"UVxf" with value %"UVxf"\n",
15312 PTR2UV(proto_perl->Iwatchaddr), PTR2UV(PL_watchaddr),
15313 PTR2UV(PL_watchok));
15316 PL_registered_mros = hv_dup_inc(proto_perl->Iregistered_mros, param);
15317 PL_blockhooks = av_dup_inc(proto_perl->Iblockhooks, param);
15318 PL_utf8_foldclosures = hv_dup_inc(proto_perl->Iutf8_foldclosures, param);
15320 /* Call the ->CLONE method, if it exists, for each of the stashes
15321 identified by sv_dup() above.
15323 while(av_tindex(param->stashes) != -1) {
15324 HV* const stash = MUTABLE_HV(av_shift(param->stashes));
15325 GV* const cloner = gv_fetchmethod_autoload(stash, "CLONE", 0);
15326 if (cloner && GvCV(cloner)) {
15331 mXPUSHs(newSVhek(HvNAME_HEK(stash)));
15333 call_sv(MUTABLE_SV(GvCV(cloner)), G_DISCARD);
15339 if (!(flags & CLONEf_KEEP_PTR_TABLE)) {
15340 ptr_table_free(PL_ptr_table);
15341 PL_ptr_table = NULL;
15344 if (!(flags & CLONEf_COPY_STACKS)) {
15345 unreferenced_to_tmp_stack(param->unreferenced);
15348 SvREFCNT_dec(param->stashes);
15350 /* orphaned? eg threads->new inside BEGIN or use */
15351 if (PL_compcv && ! SvREFCNT(PL_compcv)) {
15352 SvREFCNT_inc_simple_void(PL_compcv);
15353 SAVEFREESV(PL_compcv);
15360 S_unreferenced_to_tmp_stack(pTHX_ AV *const unreferenced)
15362 PERL_ARGS_ASSERT_UNREFERENCED_TO_TMP_STACK;
15364 if (AvFILLp(unreferenced) > -1) {
15365 SV **svp = AvARRAY(unreferenced);
15366 SV **const last = svp + AvFILLp(unreferenced);
15370 if (SvREFCNT(*svp) == 1)
15372 } while (++svp <= last);
15374 EXTEND_MORTAL(count);
15375 svp = AvARRAY(unreferenced);
15378 if (SvREFCNT(*svp) == 1) {
15379 /* Our reference is the only one to this SV. This means that
15380 in this thread, the scalar effectively has a 0 reference.
15381 That doesn't work (cleanup never happens), so donate our
15382 reference to it onto the save stack. */
15383 PL_tmps_stack[++PL_tmps_ix] = *svp;
15385 /* As an optimisation, because we are already walking the
15386 entire array, instead of above doing either
15387 SvREFCNT_inc(*svp) or *svp = &PL_sv_undef, we can instead
15388 release our reference to the scalar, so that at the end of
15389 the array owns zero references to the scalars it happens to
15390 point to. We are effectively converting the array from
15391 AvREAL() on to AvREAL() off. This saves the av_clear()
15392 (triggered by the SvREFCNT_dec(unreferenced) below) from
15393 walking the array a second time. */
15394 SvREFCNT_dec(*svp);
15397 } while (++svp <= last);
15398 AvREAL_off(unreferenced);
15400 SvREFCNT_dec_NN(unreferenced);
15404 Perl_clone_params_del(CLONE_PARAMS *param)
15406 /* This seemingly funky ordering keeps the build with PERL_GLOBAL_STRUCT
15408 PerlInterpreter *const to = param->new_perl;
15410 PerlInterpreter *const was = PERL_GET_THX;
15412 PERL_ARGS_ASSERT_CLONE_PARAMS_DEL;
15418 SvREFCNT_dec(param->stashes);
15419 if (param->unreferenced)
15420 unreferenced_to_tmp_stack(param->unreferenced);
15430 Perl_clone_params_new(PerlInterpreter *const from, PerlInterpreter *const to)
15433 /* Need to play this game, as newAV() can call safesysmalloc(), and that
15434 does a dTHX; to get the context from thread local storage.
15435 FIXME - under PERL_CORE Newx(), Safefree() and friends should expand to
15436 a version that passes in my_perl. */
15437 PerlInterpreter *const was = PERL_GET_THX;
15438 CLONE_PARAMS *param;
15440 PERL_ARGS_ASSERT_CLONE_PARAMS_NEW;
15446 /* Given that we've set the context, we can do this unshared. */
15447 Newx(param, 1, CLONE_PARAMS);
15450 param->proto_perl = from;
15451 param->new_perl = to;
15452 param->stashes = (AV *)Perl_newSV_type(to, SVt_PVAV);
15453 AvREAL_off(param->stashes);
15454 param->unreferenced = (AV *)Perl_newSV_type(to, SVt_PVAV);
15462 #endif /* USE_ITHREADS */
15465 Perl_init_constants(pTHX)
15467 SvREFCNT(&PL_sv_undef) = SvREFCNT_IMMORTAL;
15468 SvFLAGS(&PL_sv_undef) = SVf_READONLY|SVf_PROTECT|SVt_NULL;
15469 SvANY(&PL_sv_undef) = NULL;
15471 SvANY(&PL_sv_no) = new_XPVNV();
15472 SvREFCNT(&PL_sv_no) = SvREFCNT_IMMORTAL;
15473 SvFLAGS(&PL_sv_no) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15474 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15477 SvANY(&PL_sv_yes) = new_XPVNV();
15478 SvREFCNT(&PL_sv_yes) = SvREFCNT_IMMORTAL;
15479 SvFLAGS(&PL_sv_yes) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15480 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15483 SvPV_set(&PL_sv_no, (char*)PL_No);
15484 SvCUR_set(&PL_sv_no, 0);
15485 SvLEN_set(&PL_sv_no, 0);
15486 SvIV_set(&PL_sv_no, 0);
15487 SvNV_set(&PL_sv_no, 0);
15489 SvPV_set(&PL_sv_yes, (char*)PL_Yes);
15490 SvCUR_set(&PL_sv_yes, 1);
15491 SvLEN_set(&PL_sv_yes, 0);
15492 SvIV_set(&PL_sv_yes, 1);
15493 SvNV_set(&PL_sv_yes, 1);
15495 PadnamePV(&PL_padname_const) = (char *)PL_No;
15499 =head1 Unicode Support
15501 =for apidoc sv_recode_to_utf8
15503 C<encoding> is assumed to be an C<Encode> object, on entry the PV
15504 of C<sv> is assumed to be octets in that encoding, and C<sv>
15505 will be converted into Unicode (and UTF-8).
15507 If C<sv> already is UTF-8 (or if it is not C<POK>), or if C<encoding>
15508 is not a reference, nothing is done to C<sv>. If C<encoding> is not
15509 an C<Encode::XS> Encoding object, bad things will happen.
15510 (See F<cpan/Encode/encoding.pm> and L<Encode>.)
15512 The PV of C<sv> is returned.
15517 Perl_sv_recode_to_utf8(pTHX_ SV *sv, SV *encoding)
15519 PERL_ARGS_ASSERT_SV_RECODE_TO_UTF8;
15521 if (SvPOK(sv) && !SvUTF8(sv) && !IN_BYTES && SvROK(encoding)) {
15530 if (SvPADTMP(nsv)) {
15531 nsv = sv_newmortal();
15532 SvSetSV_nosteal(nsv, sv);
15541 Passing sv_yes is wrong - it needs to be or'ed set of constants
15542 for Encode::XS, while UTf-8 decode (currently) assumes a true value means
15543 remove converted chars from source.
15545 Both will default the value - let them.
15547 XPUSHs(&PL_sv_yes);
15550 call_method("decode", G_SCALAR);
15554 s = SvPV_const(uni, len);
15555 if (s != SvPVX_const(sv)) {
15556 SvGROW(sv, len + 1);
15557 Move(s, SvPVX(sv), len + 1, char);
15558 SvCUR_set(sv, len);
15563 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
15564 /* clear pos and any utf8 cache */
15565 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
15568 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
15569 magic_setutf8(sv,mg); /* clear UTF8 cache */
15574 return SvPOKp(sv) ? SvPVX(sv) : NULL;
15578 =for apidoc sv_cat_decode
15580 C<encoding> is assumed to be an C<Encode> object, the PV of C<ssv> is
15581 assumed to be octets in that encoding and decoding the input starts
15582 from the position which S<C<(PV + *offset)>> pointed to. C<dsv> will be
15583 concatenated with the decoded UTF-8 string from C<ssv>. Decoding will terminate
15584 when the string C<tstr> appears in decoding output or the input ends on
15585 the PV of C<ssv>. The value which C<offset> points will be modified
15586 to the last input position on C<ssv>.
15588 Returns TRUE if the terminator was found, else returns FALSE.
15593 Perl_sv_cat_decode(pTHX_ SV *dsv, SV *encoding,
15594 SV *ssv, int *offset, char *tstr, int tlen)
15598 PERL_ARGS_ASSERT_SV_CAT_DECODE;
15600 if (SvPOK(ssv) && SvPOK(dsv) && SvROK(encoding)) {
15611 offsv = newSViv(*offset);
15613 mPUSHp(tstr, tlen);
15615 call_method("cat_decode", G_SCALAR);
15617 ret = SvTRUE(TOPs);
15618 *offset = SvIV(offsv);
15624 Perl_croak(aTHX_ "Invalid argument to sv_cat_decode");
15629 /* ---------------------------------------------------------------------
15631 * support functions for report_uninit()
15634 /* the maxiumum size of array or hash where we will scan looking
15635 * for the undefined element that triggered the warning */
15637 #define FUV_MAX_SEARCH_SIZE 1000
15639 /* Look for an entry in the hash whose value has the same SV as val;
15640 * If so, return a mortal copy of the key. */
15643 S_find_hash_subscript(pTHX_ const HV *const hv, const SV *const val)
15649 PERL_ARGS_ASSERT_FIND_HASH_SUBSCRIPT;
15651 if (!hv || SvMAGICAL(hv) || !HvARRAY(hv) ||
15652 (HvTOTALKEYS(hv) > FUV_MAX_SEARCH_SIZE))
15655 array = HvARRAY(hv);
15657 for (i=HvMAX(hv); i>=0; i--) {
15659 for (entry = array[i]; entry; entry = HeNEXT(entry)) {
15660 if (HeVAL(entry) != val)
15662 if ( HeVAL(entry) == &PL_sv_undef ||
15663 HeVAL(entry) == &PL_sv_placeholder)
15667 if (HeKLEN(entry) == HEf_SVKEY)
15668 return sv_mortalcopy(HeKEY_sv(entry));
15669 return sv_2mortal(newSVhek(HeKEY_hek(entry)));
15675 /* Look for an entry in the array whose value has the same SV as val;
15676 * If so, return the index, otherwise return -1. */
15679 S_find_array_subscript(pTHX_ const AV *const av, const SV *const val)
15681 PERL_ARGS_ASSERT_FIND_ARRAY_SUBSCRIPT;
15683 if (!av || SvMAGICAL(av) || !AvARRAY(av) ||
15684 (AvFILLp(av) > FUV_MAX_SEARCH_SIZE))
15687 if (val != &PL_sv_undef) {
15688 SV ** const svp = AvARRAY(av);
15691 for (i=AvFILLp(av); i>=0; i--)
15698 /* varname(): return the name of a variable, optionally with a subscript.
15699 * If gv is non-zero, use the name of that global, along with gvtype (one
15700 * of "$", "@", "%"); otherwise use the name of the lexical at pad offset
15701 * targ. Depending on the value of the subscript_type flag, return:
15704 #define FUV_SUBSCRIPT_NONE 1 /* "@foo" */
15705 #define FUV_SUBSCRIPT_ARRAY 2 /* "$foo[aindex]" */
15706 #define FUV_SUBSCRIPT_HASH 3 /* "$foo{keyname}" */
15707 #define FUV_SUBSCRIPT_WITHIN 4 /* "within @foo" */
15710 Perl_varname(pTHX_ const GV *const gv, const char gvtype, PADOFFSET targ,
15711 const SV *const keyname, SSize_t aindex, int subscript_type)
15714 SV * const name = sv_newmortal();
15715 if (gv && isGV(gv)) {
15717 buffer[0] = gvtype;
15720 /* as gv_fullname4(), but add literal '^' for $^FOO names */
15722 gv_fullname4(name, gv, buffer, 0);
15724 if ((unsigned int)SvPVX(name)[1] <= 26) {
15726 buffer[1] = SvPVX(name)[1] + 'A' - 1;
15728 /* Swap the 1 unprintable control character for the 2 byte pretty
15729 version - ie substr($name, 1, 1) = $buffer; */
15730 sv_insert(name, 1, 1, buffer, 2);
15734 CV * const cv = gv ? ((CV *)gv) : find_runcv(NULL);
15737 assert(!cv || SvTYPE(cv) == SVt_PVCV || SvTYPE(cv) == SVt_PVFM);
15739 if (!cv || !CvPADLIST(cv))
15741 sv = padnamelist_fetch(PadlistNAMES(CvPADLIST(cv)), targ);
15742 sv_setpvn(name, PadnamePV(sv), PadnameLEN(sv));
15746 if (subscript_type == FUV_SUBSCRIPT_HASH) {
15747 SV * const sv = newSV(0);
15749 const char * const pv = SvPV_nomg_const((SV*)keyname, len);
15751 *SvPVX(name) = '$';
15752 Perl_sv_catpvf(aTHX_ name, "{%s}",
15753 pv_pretty(sv, pv, len, 32, NULL, NULL,
15754 PERL_PV_PRETTY_DUMP | PERL_PV_ESCAPE_UNI_DETECT ));
15755 SvREFCNT_dec_NN(sv);
15757 else if (subscript_type == FUV_SUBSCRIPT_ARRAY) {
15758 *SvPVX(name) = '$';
15759 Perl_sv_catpvf(aTHX_ name, "[%"IVdf"]", (IV)aindex);
15761 else if (subscript_type == FUV_SUBSCRIPT_WITHIN) {
15762 /* We know that name has no magic, so can use 0 instead of SV_GMAGIC */
15763 Perl_sv_insert_flags(aTHX_ name, 0, 0, STR_WITH_LEN("within "), 0);
15771 =for apidoc find_uninit_var
15773 Find the name of the undefined variable (if any) that caused the operator
15774 to issue a "Use of uninitialized value" warning.
15775 If match is true, only return a name if its value matches C<uninit_sv>.
15776 So roughly speaking, if a unary operator (such as C<OP_COS>) generates a
15777 warning, then following the direct child of the op may yield an
15778 C<OP_PADSV> or C<OP_GV> that gives the name of the undefined variable. On the
15779 other hand, with C<OP_ADD> there are two branches to follow, so we only print
15780 the variable name if we get an exact match.
15781 C<desc_p> points to a string pointer holding the description of the op.
15782 This may be updated if needed.
15784 The name is returned as a mortal SV.
15786 Assumes that C<PL_op> is the OP that originally triggered the error, and that
15787 C<PL_comppad>/C<PL_curpad> points to the currently executing pad.
15793 S_find_uninit_var(pTHX_ const OP *const obase, const SV *const uninit_sv,
15794 bool match, const char **desc_p)
15799 const OP *o, *o2, *kid;
15801 PERL_ARGS_ASSERT_FIND_UNINIT_VAR;
15803 if (!obase || (match && (!uninit_sv || uninit_sv == &PL_sv_undef ||
15804 uninit_sv == &PL_sv_placeholder)))
15807 switch (obase->op_type) {
15814 const bool pad = ( obase->op_type == OP_PADAV
15815 || obase->op_type == OP_PADHV
15816 || obase->op_type == OP_PADRANGE
15819 const bool hash = ( obase->op_type == OP_PADHV
15820 || obase->op_type == OP_RV2HV
15821 || (obase->op_type == OP_PADRANGE
15822 && SvTYPE(PAD_SVl(obase->op_targ)) == SVt_PVHV)
15826 int subscript_type = FUV_SUBSCRIPT_WITHIN;
15828 if (pad) { /* @lex, %lex */
15829 sv = PAD_SVl(obase->op_targ);
15833 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
15834 /* @global, %global */
15835 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
15838 sv = hash ? MUTABLE_SV(GvHV(gv)): MUTABLE_SV(GvAV(gv));
15840 else if (obase == PL_op) /* @{expr}, %{expr} */
15841 return find_uninit_var(cUNOPx(obase)->op_first,
15842 uninit_sv, match, desc_p);
15843 else /* @{expr}, %{expr} as a sub-expression */
15847 /* attempt to find a match within the aggregate */
15849 keysv = find_hash_subscript((const HV*)sv, uninit_sv);
15851 subscript_type = FUV_SUBSCRIPT_HASH;
15854 index = find_array_subscript((const AV *)sv, uninit_sv);
15856 subscript_type = FUV_SUBSCRIPT_ARRAY;
15859 if (match && subscript_type == FUV_SUBSCRIPT_WITHIN)
15862 return varname(gv, (char)(hash ? '%' : '@'), obase->op_targ,
15863 keysv, index, subscript_type);
15867 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
15869 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
15870 if (!gv || !GvSTASH(gv))
15872 if (match && (GvSV(gv) != uninit_sv))
15874 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
15877 return find_uninit_var(cUNOPx(obase)->op_first, uninit_sv, 1, desc_p);
15880 if (match && PAD_SVl(obase->op_targ) != uninit_sv)
15882 return varname(NULL, '$', obase->op_targ,
15883 NULL, 0, FUV_SUBSCRIPT_NONE);
15886 gv = cGVOPx_gv(obase);
15887 if (!gv || (match && GvSV(gv) != uninit_sv) || !GvSTASH(gv))
15889 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
15891 case OP_AELEMFAST_LEX:
15894 AV *av = MUTABLE_AV(PAD_SV(obase->op_targ));
15895 if (!av || SvRMAGICAL(av))
15897 svp = av_fetch(av, (I8)obase->op_private, FALSE);
15898 if (!svp || *svp != uninit_sv)
15901 return varname(NULL, '$', obase->op_targ,
15902 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
15905 gv = cGVOPx_gv(obase);
15910 AV *const av = GvAV(gv);
15911 if (!av || SvRMAGICAL(av))
15913 svp = av_fetch(av, (I8)obase->op_private, FALSE);
15914 if (!svp || *svp != uninit_sv)
15917 return varname(gv, '$', 0,
15918 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
15920 NOT_REACHED; /* NOTREACHED */
15923 o = cUNOPx(obase)->op_first;
15924 if (!o || o->op_type != OP_NULL ||
15925 ! (o->op_targ == OP_AELEM || o->op_targ == OP_HELEM))
15927 return find_uninit_var(cBINOPo->op_last, uninit_sv, match, desc_p);
15932 bool negate = FALSE;
15934 if (PL_op == obase)
15935 /* $a[uninit_expr] or $h{uninit_expr} */
15936 return find_uninit_var(cBINOPx(obase)->op_last,
15937 uninit_sv, match, desc_p);
15940 o = cBINOPx(obase)->op_first;
15941 kid = cBINOPx(obase)->op_last;
15943 /* get the av or hv, and optionally the gv */
15945 if (o->op_type == OP_PADAV || o->op_type == OP_PADHV) {
15946 sv = PAD_SV(o->op_targ);
15948 else if ((o->op_type == OP_RV2AV || o->op_type == OP_RV2HV)
15949 && cUNOPo->op_first->op_type == OP_GV)
15951 gv = cGVOPx_gv(cUNOPo->op_first);
15955 == OP_RV2HV ? MUTABLE_SV(GvHV(gv)) : MUTABLE_SV(GvAV(gv));
15960 if (kid && kid->op_type == OP_NEGATE) {
15962 kid = cUNOPx(kid)->op_first;
15965 if (kid && kid->op_type == OP_CONST && SvOK(cSVOPx_sv(kid))) {
15966 /* index is constant */
15969 kidsv = newSVpvs_flags("-", SVs_TEMP);
15970 sv_catsv(kidsv, cSVOPx_sv(kid));
15973 kidsv = cSVOPx_sv(kid);
15977 if (obase->op_type == OP_HELEM) {
15978 HE* he = hv_fetch_ent(MUTABLE_HV(sv), kidsv, 0, 0);
15979 if (!he || HeVAL(he) != uninit_sv)
15983 SV * const opsv = cSVOPx_sv(kid);
15984 const IV opsviv = SvIV(opsv);
15985 SV * const * const svp = av_fetch(MUTABLE_AV(sv),
15986 negate ? - opsviv : opsviv,
15988 if (!svp || *svp != uninit_sv)
15992 if (obase->op_type == OP_HELEM)
15993 return varname(gv, '%', o->op_targ,
15994 kidsv, 0, FUV_SUBSCRIPT_HASH);
15996 return varname(gv, '@', o->op_targ, NULL,
15997 negate ? - SvIV(cSVOPx_sv(kid)) : SvIV(cSVOPx_sv(kid)),
15998 FUV_SUBSCRIPT_ARRAY);
16001 /* index is an expression;
16002 * attempt to find a match within the aggregate */
16003 if (obase->op_type == OP_HELEM) {
16004 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16006 return varname(gv, '%', o->op_targ,
16007 keysv, 0, FUV_SUBSCRIPT_HASH);
16010 const SSize_t index
16011 = find_array_subscript((const AV *)sv, uninit_sv);
16013 return varname(gv, '@', o->op_targ,
16014 NULL, index, FUV_SUBSCRIPT_ARRAY);
16019 (char)((o->op_type == OP_PADAV || o->op_type == OP_RV2AV)
16021 o->op_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
16023 NOT_REACHED; /* NOTREACHED */
16026 case OP_MULTIDEREF: {
16027 /* If we were executing OP_MULTIDEREF when the undef warning
16028 * triggered, then it must be one of the index values within
16029 * that triggered it. If not, then the only possibility is that
16030 * the value retrieved by the last aggregate index might be the
16031 * culprit. For the former, we set PL_multideref_pc each time before
16032 * using an index, so work though the item list until we reach
16033 * that point. For the latter, just work through the entire item
16034 * list; the last aggregate retrieved will be the candidate.
16035 * There is a third rare possibility: something triggered
16036 * magic while fetching an array/hash element. Just display
16037 * nothing in this case.
16040 /* the named aggregate, if any */
16041 PADOFFSET agg_targ = 0;
16043 /* the last-seen index */
16045 PADOFFSET index_targ;
16047 IV index_const_iv = 0; /* init for spurious compiler warn */
16048 SV *index_const_sv;
16049 int depth = 0; /* how many array/hash lookups we've done */
16051 UNOP_AUX_item *items = cUNOP_AUXx(obase)->op_aux;
16052 UNOP_AUX_item *last = NULL;
16053 UV actions = items->uv;
16056 if (PL_op == obase) {
16057 last = PL_multideref_pc;
16058 assert(last >= items && last <= items + items[-1].uv);
16065 switch (actions & MDEREF_ACTION_MASK) {
16067 case MDEREF_reload:
16068 actions = (++items)->uv;
16071 case MDEREF_HV_padhv_helem: /* $lex{...} */
16074 case MDEREF_AV_padav_aelem: /* $lex[...] */
16075 agg_targ = (++items)->pad_offset;
16079 case MDEREF_HV_gvhv_helem: /* $pkg{...} */
16082 case MDEREF_AV_gvav_aelem: /* $pkg[...] */
16084 agg_gv = (GV*)UNOP_AUX_item_sv(++items);
16085 assert(isGV_with_GP(agg_gv));
16088 case MDEREF_HV_gvsv_vivify_rv2hv_helem: /* $pkg->{...} */
16089 case MDEREF_HV_padsv_vivify_rv2hv_helem: /* $lex->{...} */
16092 case MDEREF_HV_pop_rv2hv_helem: /* expr->{...} */
16093 case MDEREF_HV_vivify_rv2hv_helem: /* vivify, ->{...} */
16099 case MDEREF_AV_gvsv_vivify_rv2av_aelem: /* $pkg->[...] */
16100 case MDEREF_AV_padsv_vivify_rv2av_aelem: /* $lex->[...] */
16103 case MDEREF_AV_pop_rv2av_aelem: /* expr->[...] */
16104 case MDEREF_AV_vivify_rv2av_aelem: /* vivify, ->[...] */
16111 index_const_sv = NULL;
16113 index_type = (actions & MDEREF_INDEX_MASK);
16114 switch (index_type) {
16115 case MDEREF_INDEX_none:
16117 case MDEREF_INDEX_const:
16119 index_const_sv = UNOP_AUX_item_sv(++items)
16121 index_const_iv = (++items)->iv;
16123 case MDEREF_INDEX_padsv:
16124 index_targ = (++items)->pad_offset;
16126 case MDEREF_INDEX_gvsv:
16127 index_gv = (GV*)UNOP_AUX_item_sv(++items);
16128 assert(isGV_with_GP(index_gv));
16132 if (index_type != MDEREF_INDEX_none)
16135 if ( index_type == MDEREF_INDEX_none
16136 || (actions & MDEREF_FLAG_last)
16137 || (last && items >= last)
16141 actions >>= MDEREF_SHIFT;
16144 if (PL_op == obase) {
16145 /* most likely index was undef */
16147 *desc_p = ( (actions & MDEREF_FLAG_last)
16148 && (obase->op_private
16149 & (OPpMULTIDEREF_EXISTS|OPpMULTIDEREF_DELETE)))
16151 (obase->op_private & OPpMULTIDEREF_EXISTS)
16154 : is_hv ? "hash element" : "array element";
16155 assert(index_type != MDEREF_INDEX_none);
16157 if (GvSV(index_gv) == uninit_sv)
16158 return varname(index_gv, '$', 0, NULL, 0,
16159 FUV_SUBSCRIPT_NONE);
16164 if (PL_curpad[index_targ] == uninit_sv)
16165 return varname(NULL, '$', index_targ,
16166 NULL, 0, FUV_SUBSCRIPT_NONE);
16170 /* If we got to this point it was undef on a const subscript,
16171 * so magic probably involved, e.g. $ISA[0]. Give up. */
16175 /* the SV returned by pp_multideref() was undef, if anything was */
16181 sv = PAD_SV(agg_targ);
16183 sv = is_hv ? MUTABLE_SV(GvHV(agg_gv)) : MUTABLE_SV(GvAV(agg_gv));
16187 if (index_type == MDEREF_INDEX_const) {
16192 HE* he = hv_fetch_ent(MUTABLE_HV(sv), index_const_sv, 0, 0);
16193 if (!he || HeVAL(he) != uninit_sv)
16197 SV * const * const svp =
16198 av_fetch(MUTABLE_AV(sv), index_const_iv, FALSE);
16199 if (!svp || *svp != uninit_sv)
16204 ? varname(agg_gv, '%', agg_targ,
16205 index_const_sv, 0, FUV_SUBSCRIPT_HASH)
16206 : varname(agg_gv, '@', agg_targ,
16207 NULL, index_const_iv, FUV_SUBSCRIPT_ARRAY);
16210 /* index is an var */
16212 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16214 return varname(agg_gv, '%', agg_targ,
16215 keysv, 0, FUV_SUBSCRIPT_HASH);
16218 const SSize_t index
16219 = find_array_subscript((const AV *)sv, uninit_sv);
16221 return varname(agg_gv, '@', agg_targ,
16222 NULL, index, FUV_SUBSCRIPT_ARRAY);
16226 return varname(agg_gv,
16228 agg_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
16230 NOT_REACHED; /* NOTREACHED */
16234 /* only examine RHS */
16235 return find_uninit_var(cBINOPx(obase)->op_first, uninit_sv,
16239 o = cUNOPx(obase)->op_first;
16240 if ( o->op_type == OP_PUSHMARK
16241 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)
16245 if (!OpHAS_SIBLING(o)) {
16246 /* one-arg version of open is highly magical */
16248 if (o->op_type == OP_GV) { /* open FOO; */
16250 if (match && GvSV(gv) != uninit_sv)
16252 return varname(gv, '$', 0,
16253 NULL, 0, FUV_SUBSCRIPT_NONE);
16255 /* other possibilities not handled are:
16256 * open $x; or open my $x; should return '${*$x}'
16257 * open expr; should return '$'.expr ideally
16264 /* ops where $_ may be an implicit arg */
16269 if ( !(obase->op_flags & OPf_STACKED)) {
16270 if (uninit_sv == DEFSV)
16271 return newSVpvs_flags("$_", SVs_TEMP);
16272 else if (obase->op_targ
16273 && uninit_sv == PAD_SVl(obase->op_targ))
16274 return varname(NULL, '$', obase->op_targ, NULL, 0,
16275 FUV_SUBSCRIPT_NONE);
16282 match = 1; /* print etc can return undef on defined args */
16283 /* skip filehandle as it can't produce 'undef' warning */
16284 o = cUNOPx(obase)->op_first;
16285 if ((obase->op_flags & OPf_STACKED)
16287 ( o->op_type == OP_PUSHMARK
16288 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)))
16289 o = OpSIBLING(OpSIBLING(o));
16293 case OP_ENTEREVAL: /* could be eval $undef or $x='$undef'; eval $x */
16294 case OP_CUSTOM: /* XS or custom code could trigger random warnings */
16296 /* the following ops are capable of returning PL_sv_undef even for
16297 * defined arg(s) */
16316 case OP_GETPEERNAME:
16364 case OP_SMARTMATCH:
16373 /* XXX tmp hack: these two may call an XS sub, and currently
16374 XS subs don't have a SUB entry on the context stack, so CV and
16375 pad determination goes wrong, and BAD things happen. So, just
16376 don't try to determine the value under those circumstances.
16377 Need a better fix at dome point. DAPM 11/2007 */
16383 GV * const gv = gv_fetchpvs(".", GV_NOTQUAL, SVt_PV);
16384 if (gv && GvSV(gv) == uninit_sv)
16385 return newSVpvs_flags("$.", SVs_TEMP);
16390 /* def-ness of rval pos() is independent of the def-ness of its arg */
16391 if ( !(obase->op_flags & OPf_MOD))
16396 if (SvROK(PL_rs) && uninit_sv == SvRV(PL_rs))
16397 return newSVpvs_flags("${$/}", SVs_TEMP);
16402 if (!(obase->op_flags & OPf_KIDS))
16404 o = cUNOPx(obase)->op_first;
16410 /* This loop checks all the kid ops, skipping any that cannot pos-
16411 * sibly be responsible for the uninitialized value; i.e., defined
16412 * constants and ops that return nothing. If there is only one op
16413 * left that is not skipped, then we *know* it is responsible for
16414 * the uninitialized value. If there is more than one op left, we
16415 * have to look for an exact match in the while() loop below.
16416 * Note that we skip padrange, because the individual pad ops that
16417 * it replaced are still in the tree, so we work on them instead.
16420 for (kid=o; kid; kid = OpSIBLING(kid)) {
16421 const OPCODE type = kid->op_type;
16422 if ( (type == OP_CONST && SvOK(cSVOPx_sv(kid)))
16423 || (type == OP_NULL && ! (kid->op_flags & OPf_KIDS))
16424 || (type == OP_PUSHMARK)
16425 || (type == OP_PADRANGE)
16429 if (o2) { /* more than one found */
16436 return find_uninit_var(o2, uninit_sv, match, desc_p);
16438 /* scan all args */
16440 sv = find_uninit_var(o, uninit_sv, 1, desc_p);
16452 =for apidoc report_uninit
16454 Print appropriate "Use of uninitialized variable" warning.
16460 Perl_report_uninit(pTHX_ const SV *uninit_sv)
16462 const char *desc = NULL;
16463 SV* varname = NULL;
16466 desc = PL_op->op_type == OP_STRINGIFY && PL_op->op_folded
16469 if (uninit_sv && PL_curpad) {
16470 varname = find_uninit_var(PL_op, uninit_sv, 0, &desc);
16472 sv_insert(varname, 0, 0, " ", 1);
16475 else if (PL_curstackinfo->si_type == PERLSI_SORT && cxstack_ix == 0)
16476 /* we've reached the end of a sort block or sub,
16477 * and the uninit value is probably what that code returned */
16480 /* PL_warn_uninit_sv is constant */
16481 GCC_DIAG_IGNORE(-Wformat-nonliteral);
16483 /* diag_listed_as: Use of uninitialized value%s */
16484 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit_sv,
16485 SVfARG(varname ? varname : &PL_sv_no),
16488 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit,
16494 * ex: set ts=8 sts=4 sw=4 et: