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 #ifdef PERL_NEW_COPY_ON_WRITE
52 # ifndef SV_COW_THRESHOLD
53 # define SV_COW_THRESHOLD 0 /* COW iff len > K */
55 # ifndef SV_COWBUF_THRESHOLD
56 # define SV_COWBUF_THRESHOLD 1250 /* COW iff len > K */
58 # ifndef SV_COW_MAX_WASTE_THRESHOLD
59 # define SV_COW_MAX_WASTE_THRESHOLD 80 /* COW iff (len - cur) < K */
61 # ifndef SV_COWBUF_WASTE_THRESHOLD
62 # define SV_COWBUF_WASTE_THRESHOLD 80 /* COW iff (len - cur) < K */
64 # ifndef SV_COW_MAX_WASTE_FACTOR_THRESHOLD
65 # define SV_COW_MAX_WASTE_FACTOR_THRESHOLD 2 /* COW iff len < (cur * K) */
67 # ifndef SV_COWBUF_WASTE_FACTOR_THRESHOLD
68 # define SV_COWBUF_WASTE_FACTOR_THRESHOLD 2 /* COW iff len < (cur * K) */
71 /* Work around compiler warnings about unsigned >= THRESHOLD when thres-
74 # define GE_COW_THRESHOLD(cur) ((cur) >= SV_COW_THRESHOLD)
76 # define GE_COW_THRESHOLD(cur) 1
78 #if SV_COWBUF_THRESHOLD
79 # define GE_COWBUF_THRESHOLD(cur) ((cur) >= SV_COWBUF_THRESHOLD)
81 # define GE_COWBUF_THRESHOLD(cur) 1
83 #if SV_COW_MAX_WASTE_THRESHOLD
84 # define GE_COW_MAX_WASTE_THRESHOLD(cur,len) (((len)-(cur)) < SV_COW_MAX_WASTE_THRESHOLD)
86 # define GE_COW_MAX_WASTE_THRESHOLD(cur,len) 1
88 #if SV_COWBUF_WASTE_THRESHOLD
89 # define GE_COWBUF_WASTE_THRESHOLD(cur,len) (((len)-(cur)) < SV_COWBUF_WASTE_THRESHOLD)
91 # define GE_COWBUF_WASTE_THRESHOLD(cur,len) 1
93 #if SV_COW_MAX_WASTE_FACTOR_THRESHOLD
94 # define GE_COW_MAX_WASTE_FACTOR_THRESHOLD(cur,len) ((len) < SV_COW_MAX_WASTE_FACTOR_THRESHOLD * (cur))
96 # define GE_COW_MAX_WASTE_FACTOR_THRESHOLD(cur,len) 1
98 #if SV_COWBUF_WASTE_FACTOR_THRESHOLD
99 # define GE_COWBUF_WASTE_FACTOR_THRESHOLD(cur,len) ((len) < SV_COWBUF_WASTE_FACTOR_THRESHOLD * (cur))
101 # define GE_COWBUF_WASTE_FACTOR_THRESHOLD(cur,len) 1
104 #define CHECK_COW_THRESHOLD(cur,len) (\
105 GE_COW_THRESHOLD((cur)) && \
106 GE_COW_MAX_WASTE_THRESHOLD((cur),(len)) && \
107 GE_COW_MAX_WASTE_FACTOR_THRESHOLD((cur),(len)) \
109 #define CHECK_COWBUF_THRESHOLD(cur,len) (\
110 GE_COWBUF_THRESHOLD((cur)) && \
111 GE_COWBUF_WASTE_THRESHOLD((cur),(len)) && \
112 GE_COWBUF_WASTE_FACTOR_THRESHOLD((cur),(len)) \
115 #ifdef PERL_UTF8_CACHE_ASSERT
116 /* if adding more checks watch out for the following tests:
117 * t/op/index.t t/op/length.t t/op/pat.t t/op/substr.t
118 * lib/utf8.t lib/Unicode/Collate/t/index.t
121 # define ASSERT_UTF8_CACHE(cache) \
122 STMT_START { if (cache) { assert((cache)[0] <= (cache)[1]); \
123 assert((cache)[2] <= (cache)[3]); \
124 assert((cache)[3] <= (cache)[1]);} \
127 # define ASSERT_UTF8_CACHE(cache) NOOP
130 #ifdef PERL_OLD_COPY_ON_WRITE
131 #define SV_COW_NEXT_SV(sv) INT2PTR(SV *,SvUVX(sv))
132 #define SV_COW_NEXT_SV_SET(current,next) SvUV_set(current, PTR2UV(next))
135 /* ============================================================================
137 =head1 Allocation and deallocation of SVs.
138 An SV (or AV, HV, etc.) is allocated in two parts: the head (struct
139 sv, av, hv...) contains type and reference count information, and for
140 many types, a pointer to the body (struct xrv, xpv, xpviv...), which
141 contains fields specific to each type. Some types store all they need
142 in the head, so don't have a body.
144 In all but the most memory-paranoid configurations (ex: PURIFY), heads
145 and bodies are allocated out of arenas, which by default are
146 approximately 4K chunks of memory parcelled up into N heads or bodies.
147 Sv-bodies are allocated by their sv-type, guaranteeing size
148 consistency needed to allocate safely from arrays.
150 For SV-heads, the first slot in each arena is reserved, and holds a
151 link to the next arena, some flags, and a note of the number of slots.
152 Snaked through each arena chain is a linked list of free items; when
153 this becomes empty, an extra arena is allocated and divided up into N
154 items which are threaded into the free list.
156 SV-bodies are similar, but they use arena-sets by default, which
157 separate the link and info from the arena itself, and reclaim the 1st
158 slot in the arena. SV-bodies are further described later.
160 The following global variables are associated with arenas:
162 PL_sv_arenaroot pointer to list of SV arenas
163 PL_sv_root pointer to list of free SV structures
165 PL_body_arenas head of linked-list of body arenas
166 PL_body_roots[] array of pointers to list of free bodies of svtype
167 arrays are indexed by the svtype needed
169 A few special SV heads are not allocated from an arena, but are
170 instead directly created in the interpreter structure, eg PL_sv_undef.
171 The size of arenas can be changed from the default by setting
172 PERL_ARENA_SIZE appropriately at compile time.
174 The SV arena serves the secondary purpose of allowing still-live SVs
175 to be located and destroyed during final cleanup.
177 At the lowest level, the macros new_SV() and del_SV() grab and free
178 an SV head. (If debugging with -DD, del_SV() calls the function S_del_sv()
179 to return the SV to the free list with error checking.) new_SV() calls
180 more_sv() / sv_add_arena() to add an extra arena if the free list is empty.
181 SVs in the free list have their SvTYPE field set to all ones.
183 At the time of very final cleanup, sv_free_arenas() is called from
184 perl_destruct() to physically free all the arenas allocated since the
185 start of the interpreter.
187 The function visit() scans the SV arenas list, and calls a specified
188 function for each SV it finds which is still live - ie which has an SvTYPE
189 other than all 1's, and a non-zero SvREFCNT. visit() is used by the
190 following functions (specified as [function that calls visit()] / [function
191 called by visit() for each SV]):
193 sv_report_used() / do_report_used()
194 dump all remaining SVs (debugging aid)
196 sv_clean_objs() / do_clean_objs(),do_clean_named_objs(),
197 do_clean_named_io_objs(),do_curse()
198 Attempt to free all objects pointed to by RVs,
199 try to do the same for all objects indir-
200 ectly referenced by typeglobs too, and
201 then do a final sweep, cursing any
202 objects that remain. Called once from
203 perl_destruct(), prior to calling sv_clean_all()
206 sv_clean_all() / do_clean_all()
207 SvREFCNT_dec(sv) each remaining SV, possibly
208 triggering an sv_free(). It also sets the
209 SVf_BREAK flag on the SV to indicate that the
210 refcnt has been artificially lowered, and thus
211 stopping sv_free() from giving spurious warnings
212 about SVs which unexpectedly have a refcnt
213 of zero. called repeatedly from perl_destruct()
214 until there are no SVs left.
216 =head2 Arena allocator API Summary
218 Private API to rest of sv.c
222 new_XPVNV(), del_XPVGV(),
227 sv_report_used(), sv_clean_objs(), sv_clean_all(), sv_free_arenas()
231 * ========================================================================= */
234 * "A time to plant, and a time to uproot what was planted..."
238 # define MEM_LOG_NEW_SV(sv, file, line, func) \
239 Perl_mem_log_new_sv(sv, file, line, func)
240 # define MEM_LOG_DEL_SV(sv, file, line, func) \
241 Perl_mem_log_del_sv(sv, file, line, func)
243 # define MEM_LOG_NEW_SV(sv, file, line, func) NOOP
244 # define MEM_LOG_DEL_SV(sv, file, line, func) NOOP
247 #ifdef DEBUG_LEAKING_SCALARS
248 # define FREE_SV_DEBUG_FILE(sv) STMT_START { \
249 if ((sv)->sv_debug_file) PerlMemShared_free((sv)->sv_debug_file); \
251 # define DEBUG_SV_SERIAL(sv) \
252 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) del_SV\n", \
253 PTR2UV(sv), (long)(sv)->sv_debug_serial))
255 # define FREE_SV_DEBUG_FILE(sv)
256 # define DEBUG_SV_SERIAL(sv) NOOP
260 # define SvARENA_CHAIN(sv) ((sv)->sv_u.svu_rv)
261 # define SvARENA_CHAIN_SET(sv,val) (sv)->sv_u.svu_rv = MUTABLE_SV((val))
262 /* Whilst I'd love to do this, it seems that things like to check on
264 # define POSION_SV_HEAD(sv) PoisonNew(sv, 1, struct STRUCT_SV)
266 # define POSION_SV_HEAD(sv) PoisonNew(&SvANY(sv), 1, void *), \
267 PoisonNew(&SvREFCNT(sv), 1, U32)
269 # define SvARENA_CHAIN(sv) SvANY(sv)
270 # define SvARENA_CHAIN_SET(sv,val) SvANY(sv) = (void *)(val)
271 # define POSION_SV_HEAD(sv)
274 /* Mark an SV head as unused, and add to free list.
276 * If SVf_BREAK is set, skip adding it to the free list, as this SV had
277 * its refcount artificially decremented during global destruction, so
278 * there may be dangling pointers to it. The last thing we want in that
279 * case is for it to be reused. */
281 #define plant_SV(p) \
283 const U32 old_flags = SvFLAGS(p); \
284 MEM_LOG_DEL_SV(p, __FILE__, __LINE__, FUNCTION__); \
285 DEBUG_SV_SERIAL(p); \
286 FREE_SV_DEBUG_FILE(p); \
288 SvFLAGS(p) = SVTYPEMASK; \
289 if (!(old_flags & SVf_BREAK)) { \
290 SvARENA_CHAIN_SET(p, PL_sv_root); \
296 #define uproot_SV(p) \
299 PL_sv_root = MUTABLE_SV(SvARENA_CHAIN(p)); \
304 /* make some more SVs by adding another arena */
310 char *chunk; /* must use New here to match call to */
311 Newx(chunk,PERL_ARENA_SIZE,char); /* Safefree() in sv_free_arenas() */
312 sv_add_arena(chunk, PERL_ARENA_SIZE, 0);
317 /* new_SV(): return a new, empty SV head */
319 #ifdef DEBUG_LEAKING_SCALARS
320 /* provide a real function for a debugger to play with */
322 S_new_SV(pTHX_ const char *file, int line, const char *func)
329 sv = S_more_sv(aTHX);
333 sv->sv_debug_optype = PL_op ? PL_op->op_type : 0;
334 sv->sv_debug_line = (U16) (PL_parser && PL_parser->copline != NOLINE
340 sv->sv_debug_inpad = 0;
341 sv->sv_debug_parent = NULL;
342 sv->sv_debug_file = PL_curcop ? savesharedpv(CopFILE(PL_curcop)): NULL;
344 sv->sv_debug_serial = PL_sv_serial++;
346 MEM_LOG_NEW_SV(sv, file, line, func);
347 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) new_SV (from %s:%d [%s])\n",
348 PTR2UV(sv), (long)sv->sv_debug_serial, file, line, func));
352 # define new_SV(p) (p)=S_new_SV(aTHX_ __FILE__, __LINE__, FUNCTION__)
360 (p) = S_more_sv(aTHX); \
364 MEM_LOG_NEW_SV(p, __FILE__, __LINE__, FUNCTION__); \
369 /* del_SV(): return an empty SV head to the free list */
382 S_del_sv(pTHX_ SV *p)
384 PERL_ARGS_ASSERT_DEL_SV;
389 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
390 const SV * const sv = sva + 1;
391 const SV * const svend = &sva[SvREFCNT(sva)];
392 if (p >= sv && p < svend) {
398 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
399 "Attempt to free non-arena SV: 0x%"UVxf
400 pTHX__FORMAT, PTR2UV(p) pTHX__VALUE);
407 #else /* ! DEBUGGING */
409 #define del_SV(p) plant_SV(p)
411 #endif /* DEBUGGING */
415 =head1 SV Manipulation Functions
417 =for apidoc sv_add_arena
419 Given a chunk of memory, link it to the head of the list of arenas,
420 and split it into a list of free SVs.
426 S_sv_add_arena(pTHX_ char *const ptr, const U32 size, const U32 flags)
428 SV *const sva = MUTABLE_SV(ptr);
432 PERL_ARGS_ASSERT_SV_ADD_ARENA;
434 /* The first SV in an arena isn't an SV. */
435 SvANY(sva) = (void *) PL_sv_arenaroot; /* ptr to next arena */
436 SvREFCNT(sva) = size / sizeof(SV); /* number of SV slots */
437 SvFLAGS(sva) = flags; /* FAKE if not to be freed */
439 PL_sv_arenaroot = sva;
440 PL_sv_root = sva + 1;
442 svend = &sva[SvREFCNT(sva) - 1];
445 SvARENA_CHAIN_SET(sv, (sv + 1));
449 /* Must always set typemask because it's always checked in on cleanup
450 when the arenas are walked looking for objects. */
451 SvFLAGS(sv) = SVTYPEMASK;
454 SvARENA_CHAIN_SET(sv, 0);
458 SvFLAGS(sv) = SVTYPEMASK;
461 /* visit(): call the named function for each non-free SV in the arenas
462 * whose flags field matches the flags/mask args. */
465 S_visit(pTHX_ SVFUNC_t f, const U32 flags, const U32 mask)
470 PERL_ARGS_ASSERT_VISIT;
472 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
473 const SV * const svend = &sva[SvREFCNT(sva)];
475 for (sv = sva + 1; sv < svend; ++sv) {
476 if (SvTYPE(sv) != (svtype)SVTYPEMASK
477 && (sv->sv_flags & mask) == flags
490 /* called by sv_report_used() for each live SV */
493 do_report_used(pTHX_ SV *const sv)
495 if (SvTYPE(sv) != (svtype)SVTYPEMASK) {
496 PerlIO_printf(Perl_debug_log, "****\n");
503 =for apidoc sv_report_used
505 Dump the contents of all SVs not yet freed (debugging aid).
511 Perl_sv_report_used(pTHX)
514 visit(do_report_used, 0, 0);
520 /* called by sv_clean_objs() for each live SV */
523 do_clean_objs(pTHX_ SV *const ref)
527 SV * const target = SvRV(ref);
528 if (SvOBJECT(target)) {
529 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning object ref:\n "), sv_dump(ref)));
530 if (SvWEAKREF(ref)) {
531 sv_del_backref(target, ref);
537 SvREFCNT_dec_NN(target);
544 /* clear any slots in a GV which hold objects - except IO;
545 * called by sv_clean_objs() for each live GV */
548 do_clean_named_objs(pTHX_ SV *const sv)
551 assert(SvTYPE(sv) == SVt_PVGV);
552 assert(isGV_with_GP(sv));
556 /* freeing GP entries may indirectly free the current GV;
557 * hold onto it while we mess with the GP slots */
560 if ( ((obj = GvSV(sv) )) && SvOBJECT(obj)) {
561 DEBUG_D((PerlIO_printf(Perl_debug_log,
562 "Cleaning named glob SV object:\n "), sv_dump(obj)));
564 SvREFCNT_dec_NN(obj);
566 if ( ((obj = MUTABLE_SV(GvAV(sv)) )) && SvOBJECT(obj)) {
567 DEBUG_D((PerlIO_printf(Perl_debug_log,
568 "Cleaning named glob AV object:\n "), sv_dump(obj)));
570 SvREFCNT_dec_NN(obj);
572 if ( ((obj = MUTABLE_SV(GvHV(sv)) )) && SvOBJECT(obj)) {
573 DEBUG_D((PerlIO_printf(Perl_debug_log,
574 "Cleaning named glob HV object:\n "), sv_dump(obj)));
576 SvREFCNT_dec_NN(obj);
578 if ( ((obj = MUTABLE_SV(GvCV(sv)) )) && SvOBJECT(obj)) {
579 DEBUG_D((PerlIO_printf(Perl_debug_log,
580 "Cleaning named glob CV object:\n "), sv_dump(obj)));
582 SvREFCNT_dec_NN(obj);
584 SvREFCNT_dec_NN(sv); /* undo the inc above */
587 /* clear any IO slots in a GV which hold objects (except stderr, defout);
588 * called by sv_clean_objs() for each live GV */
591 do_clean_named_io_objs(pTHX_ SV *const sv)
594 assert(SvTYPE(sv) == SVt_PVGV);
595 assert(isGV_with_GP(sv));
596 if (!GvGP(sv) || sv == (SV*)PL_stderrgv || sv == (SV*)PL_defoutgv)
600 if ( ((obj = MUTABLE_SV(GvIO(sv)) )) && SvOBJECT(obj)) {
601 DEBUG_D((PerlIO_printf(Perl_debug_log,
602 "Cleaning named glob IO object:\n "), sv_dump(obj)));
604 SvREFCNT_dec_NN(obj);
606 SvREFCNT_dec_NN(sv); /* undo the inc above */
609 /* Void wrapper to pass to visit() */
611 do_curse(pTHX_ SV * const sv) {
612 if ((PL_stderrgv && GvGP(PL_stderrgv) && (SV*)GvIO(PL_stderrgv) == sv)
613 || (PL_defoutgv && GvGP(PL_defoutgv) && (SV*)GvIO(PL_defoutgv) == sv))
619 =for apidoc sv_clean_objs
621 Attempt to destroy all objects not yet freed.
627 Perl_sv_clean_objs(pTHX)
630 PL_in_clean_objs = TRUE;
631 visit(do_clean_objs, SVf_ROK, SVf_ROK);
632 /* Some barnacles may yet remain, clinging to typeglobs.
633 * Run the non-IO destructors first: they may want to output
634 * error messages, close files etc */
635 visit(do_clean_named_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
636 visit(do_clean_named_io_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
637 /* And if there are some very tenacious barnacles clinging to arrays,
638 closures, or what have you.... */
639 visit(do_curse, SVs_OBJECT, SVs_OBJECT);
640 olddef = PL_defoutgv;
641 PL_defoutgv = NULL; /* disable skip of PL_defoutgv */
642 if (olddef && isGV_with_GP(olddef))
643 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olddef));
644 olderr = PL_stderrgv;
645 PL_stderrgv = NULL; /* disable skip of PL_stderrgv */
646 if (olderr && isGV_with_GP(olderr))
647 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olderr));
648 SvREFCNT_dec(olddef);
649 PL_in_clean_objs = FALSE;
652 /* called by sv_clean_all() for each live SV */
655 do_clean_all(pTHX_ SV *const sv)
657 if (sv == (const SV *) PL_fdpid || sv == (const SV *)PL_strtab) {
658 /* don't clean pid table and strtab */
661 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning loops: SV at 0x%"UVxf"\n", PTR2UV(sv)) ));
662 SvFLAGS(sv) |= SVf_BREAK;
667 =for apidoc sv_clean_all
669 Decrement the refcnt of each remaining SV, possibly triggering a
670 cleanup. This function may have to be called multiple times to free
671 SVs which are in complex self-referential hierarchies.
677 Perl_sv_clean_all(pTHX)
680 PL_in_clean_all = TRUE;
681 cleaned = visit(do_clean_all, 0,0);
686 ARENASETS: a meta-arena implementation which separates arena-info
687 into struct arena_set, which contains an array of struct
688 arena_descs, each holding info for a single arena. By separating
689 the meta-info from the arena, we recover the 1st slot, formerly
690 borrowed for list management. The arena_set is about the size of an
691 arena, avoiding the needless malloc overhead of a naive linked-list.
693 The cost is 1 arena-set malloc per ~320 arena-mallocs, + the unused
694 memory in the last arena-set (1/2 on average). In trade, we get
695 back the 1st slot in each arena (ie 1.7% of a CV-arena, less for
696 smaller types). The recovery of the wasted space allows use of
697 small arenas for large, rare body types, by changing array* fields
698 in body_details_by_type[] below.
701 char *arena; /* the raw storage, allocated aligned */
702 size_t size; /* its size ~4k typ */
703 svtype utype; /* bodytype stored in arena */
708 /* Get the maximum number of elements in set[] such that struct arena_set
709 will fit within PERL_ARENA_SIZE, which is probably just under 4K, and
710 therefore likely to be 1 aligned memory page. */
712 #define ARENAS_PER_SET ((PERL_ARENA_SIZE - sizeof(struct arena_set*) \
713 - 2 * sizeof(int)) / sizeof (struct arena_desc))
716 struct arena_set* next;
717 unsigned int set_size; /* ie ARENAS_PER_SET */
718 unsigned int curr; /* index of next available arena-desc */
719 struct arena_desc set[ARENAS_PER_SET];
723 =for apidoc sv_free_arenas
725 Deallocate the memory used by all arenas. Note that all the individual SV
726 heads and bodies within the arenas must already have been freed.
732 Perl_sv_free_arenas(pTHX)
738 /* Free arenas here, but be careful about fake ones. (We assume
739 contiguity of the fake ones with the corresponding real ones.) */
741 for (sva = PL_sv_arenaroot; sva; sva = svanext) {
742 svanext = MUTABLE_SV(SvANY(sva));
743 while (svanext && SvFAKE(svanext))
744 svanext = MUTABLE_SV(SvANY(svanext));
751 struct arena_set *aroot = (struct arena_set*) PL_body_arenas;
754 struct arena_set *current = aroot;
757 assert(aroot->set[i].arena);
758 Safefree(aroot->set[i].arena);
766 i = PERL_ARENA_ROOTS_SIZE;
768 PL_body_roots[i] = 0;
775 Here are mid-level routines that manage the allocation of bodies out
776 of the various arenas. There are 5 kinds of arenas:
778 1. SV-head arenas, which are discussed and handled above
779 2. regular body arenas
780 3. arenas for reduced-size bodies
783 Arena types 2 & 3 are chained by body-type off an array of
784 arena-root pointers, which is indexed by svtype. Some of the
785 larger/less used body types are malloced singly, since a large
786 unused block of them is wasteful. Also, several svtypes dont have
787 bodies; the data fits into the sv-head itself. The arena-root
788 pointer thus has a few unused root-pointers (which may be hijacked
789 later for arena types 4,5)
791 3 differs from 2 as an optimization; some body types have several
792 unused fields in the front of the structure (which are kept in-place
793 for consistency). These bodies can be allocated in smaller chunks,
794 because the leading fields arent accessed. Pointers to such bodies
795 are decremented to point at the unused 'ghost' memory, knowing that
796 the pointers are used with offsets to the real memory.
799 =head1 SV-Body Allocation
803 Allocation of SV-bodies is similar to SV-heads, differing as follows;
804 the allocation mechanism is used for many body types, so is somewhat
805 more complicated, it uses arena-sets, and has no need for still-live
808 At the outermost level, (new|del)_X*V macros return bodies of the
809 appropriate type. These macros call either (new|del)_body_type or
810 (new|del)_body_allocated macro pairs, depending on specifics of the
811 type. Most body types use the former pair, the latter pair is used to
812 allocate body types with "ghost fields".
814 "ghost fields" are fields that are unused in certain types, and
815 consequently don't need to actually exist. They are declared because
816 they're part of a "base type", which allows use of functions as
817 methods. The simplest examples are AVs and HVs, 2 aggregate types
818 which don't use the fields which support SCALAR semantics.
820 For these types, the arenas are carved up into appropriately sized
821 chunks, we thus avoid wasted memory for those unaccessed members.
822 When bodies are allocated, we adjust the pointer back in memory by the
823 size of the part not allocated, so it's as if we allocated the full
824 structure. (But things will all go boom if you write to the part that
825 is "not there", because you'll be overwriting the last members of the
826 preceding structure in memory.)
828 We calculate the correction using the STRUCT_OFFSET macro on the first
829 member present. If the allocated structure is smaller (no initial NV
830 actually allocated) then the net effect is to subtract the size of the NV
831 from the pointer, to return a new pointer as if an initial NV were actually
832 allocated. (We were using structures named *_allocated for this, but
833 this turned out to be a subtle bug, because a structure without an NV
834 could have a lower alignment constraint, but the compiler is allowed to
835 optimised accesses based on the alignment constraint of the actual pointer
836 to the full structure, for example, using a single 64 bit load instruction
837 because it "knows" that two adjacent 32 bit members will be 8-byte aligned.)
839 This is the same trick as was used for NV and IV bodies. Ironically it
840 doesn't need to be used for NV bodies any more, because NV is now at
841 the start of the structure. IV bodies, and also in some builds NV bodies,
842 don't need it either, because they are no longer allocated.
844 In turn, the new_body_* allocators call S_new_body(), which invokes
845 new_body_inline macro, which takes a lock, and takes a body off the
846 linked list at PL_body_roots[sv_type], calling Perl_more_bodies() if
847 necessary to refresh an empty list. Then the lock is released, and
848 the body is returned.
850 Perl_more_bodies allocates a new arena, and carves it up into an array of N
851 bodies, which it strings into a linked list. It looks up arena-size
852 and body-size from the body_details table described below, thus
853 supporting the multiple body-types.
855 If PURIFY is defined, or PERL_ARENA_SIZE=0, arenas are not used, and
856 the (new|del)_X*V macros are mapped directly to malloc/free.
858 For each sv-type, struct body_details bodies_by_type[] carries
859 parameters which control these aspects of SV handling:
861 Arena_size determines whether arenas are used for this body type, and if
862 so, how big they are. PURIFY or PERL_ARENA_SIZE=0 set this field to
863 zero, forcing individual mallocs and frees.
865 Body_size determines how big a body is, and therefore how many fit into
866 each arena. Offset carries the body-pointer adjustment needed for
867 "ghost fields", and is used in *_allocated macros.
869 But its main purpose is to parameterize info needed in
870 Perl_sv_upgrade(). The info here dramatically simplifies the function
871 vs the implementation in 5.8.8, making it table-driven. All fields
872 are used for this, except for arena_size.
874 For the sv-types that have no bodies, arenas are not used, so those
875 PL_body_roots[sv_type] are unused, and can be overloaded. In
876 something of a special case, SVt_NULL is borrowed for HE arenas;
877 PL_body_roots[HE_SVSLOT=SVt_NULL] is filled by S_more_he, but the
878 bodies_by_type[SVt_NULL] slot is not used, as the table is not
883 struct body_details {
884 U8 body_size; /* Size to allocate */
885 U8 copy; /* Size of structure to copy (may be shorter) */
886 U8 offset; /* Size of unalloced ghost fields to first alloced field*/
887 PERL_BITFIELD8 type : 4; /* We have space for a sanity check. */
888 PERL_BITFIELD8 cant_upgrade : 1;/* Cannot upgrade this type */
889 PERL_BITFIELD8 zero_nv : 1; /* zero the NV when upgrading from this */
890 PERL_BITFIELD8 arena : 1; /* Allocated from an arena */
891 U32 arena_size; /* Size of arena to allocate */
899 /* With -DPURFIY we allocate everything directly, and don't use arenas.
900 This seems a rather elegant way to simplify some of the code below. */
901 #define HASARENA FALSE
903 #define HASARENA TRUE
905 #define NOARENA FALSE
907 /* Size the arenas to exactly fit a given number of bodies. A count
908 of 0 fits the max number bodies into a PERL_ARENA_SIZE.block,
909 simplifying the default. If count > 0, the arena is sized to fit
910 only that many bodies, allowing arenas to be used for large, rare
911 bodies (XPVFM, XPVIO) without undue waste. The arena size is
912 limited by PERL_ARENA_SIZE, so we can safely oversize the
915 #define FIT_ARENA0(body_size) \
916 ((size_t)(PERL_ARENA_SIZE / body_size) * body_size)
917 #define FIT_ARENAn(count,body_size) \
918 ( count * body_size <= PERL_ARENA_SIZE) \
919 ? count * body_size \
920 : FIT_ARENA0 (body_size)
921 #define FIT_ARENA(count,body_size) \
923 ? FIT_ARENAn (count, body_size) \
924 : FIT_ARENA0 (body_size))
926 /* Calculate the length to copy. Specifically work out the length less any
927 final padding the compiler needed to add. See the comment in sv_upgrade
928 for why copying the padding proved to be a bug. */
930 #define copy_length(type, last_member) \
931 STRUCT_OFFSET(type, last_member) \
932 + sizeof (((type*)SvANY((const SV *)0))->last_member)
934 static const struct body_details bodies_by_type[] = {
935 /* HEs use this offset for their arena. */
936 { 0, 0, 0, SVt_NULL, FALSE, NONV, NOARENA, 0 },
938 /* IVs are in the head, so the allocation size is 0. */
940 sizeof(IV), /* This is used to copy out the IV body. */
941 STRUCT_OFFSET(XPVIV, xiv_iv), SVt_IV, FALSE, NONV,
942 NOARENA /* IVS don't need an arena */, 0
947 STRUCT_OFFSET(XPVNV, xnv_u),
948 SVt_NV, FALSE, HADNV, NOARENA, 0 },
950 { sizeof(NV), sizeof(NV),
951 STRUCT_OFFSET(XPVNV, xnv_u),
952 SVt_NV, FALSE, HADNV, HASARENA, FIT_ARENA(0, sizeof(NV)) },
955 { sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur),
956 copy_length(XPV, xpv_len) - STRUCT_OFFSET(XPV, xpv_cur),
957 + STRUCT_OFFSET(XPV, xpv_cur),
958 SVt_PV, FALSE, NONV, HASARENA,
959 FIT_ARENA(0, sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur)) },
961 { sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur),
962 copy_length(XINVLIST, is_offset) - STRUCT_OFFSET(XPV, xpv_cur),
963 + STRUCT_OFFSET(XPV, xpv_cur),
964 SVt_INVLIST, TRUE, NONV, HASARENA,
965 FIT_ARENA(0, sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur)) },
967 { sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur),
968 copy_length(XPVIV, xiv_u) - STRUCT_OFFSET(XPV, xpv_cur),
969 + STRUCT_OFFSET(XPV, xpv_cur),
970 SVt_PVIV, FALSE, NONV, HASARENA,
971 FIT_ARENA(0, sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur)) },
973 { sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur),
974 copy_length(XPVNV, xnv_u) - STRUCT_OFFSET(XPV, xpv_cur),
975 + STRUCT_OFFSET(XPV, xpv_cur),
976 SVt_PVNV, FALSE, HADNV, HASARENA,
977 FIT_ARENA(0, sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur)) },
979 { sizeof(XPVMG), copy_length(XPVMG, xnv_u), 0, SVt_PVMG, FALSE, HADNV,
980 HASARENA, FIT_ARENA(0, sizeof(XPVMG)) },
985 SVt_REGEXP, TRUE, NONV, HASARENA,
986 FIT_ARENA(0, sizeof(regexp))
989 { sizeof(XPVGV), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV,
990 HASARENA, FIT_ARENA(0, sizeof(XPVGV)) },
992 { sizeof(XPVLV), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV,
993 HASARENA, FIT_ARENA(0, sizeof(XPVLV)) },
996 copy_length(XPVAV, xav_alloc),
998 SVt_PVAV, TRUE, NONV, HASARENA,
999 FIT_ARENA(0, sizeof(XPVAV)) },
1002 copy_length(XPVHV, xhv_max),
1004 SVt_PVHV, TRUE, NONV, HASARENA,
1005 FIT_ARENA(0, sizeof(XPVHV)) },
1010 SVt_PVCV, TRUE, NONV, HASARENA,
1011 FIT_ARENA(0, sizeof(XPVCV)) },
1016 SVt_PVFM, TRUE, NONV, NOARENA,
1017 FIT_ARENA(20, sizeof(XPVFM)) },
1022 SVt_PVIO, TRUE, NONV, HASARENA,
1023 FIT_ARENA(24, sizeof(XPVIO)) },
1026 #define new_body_allocated(sv_type) \
1027 (void *)((char *)S_new_body(aTHX_ sv_type) \
1028 - bodies_by_type[sv_type].offset)
1030 /* return a thing to the free list */
1032 #define del_body(thing, root) \
1034 void ** const thing_copy = (void **)thing; \
1035 *thing_copy = *root; \
1036 *root = (void*)thing_copy; \
1040 #if !(NVSIZE <= IVSIZE)
1041 # define new_XNV() safemalloc(sizeof(XPVNV))
1043 #define new_XPVNV() safemalloc(sizeof(XPVNV))
1044 #define new_XPVMG() safemalloc(sizeof(XPVMG))
1046 #define del_XPVGV(p) safefree(p)
1050 #if !(NVSIZE <= IVSIZE)
1051 # define new_XNV() new_body_allocated(SVt_NV)
1053 #define new_XPVNV() new_body_allocated(SVt_PVNV)
1054 #define new_XPVMG() new_body_allocated(SVt_PVMG)
1056 #define del_XPVGV(p) del_body(p + bodies_by_type[SVt_PVGV].offset, \
1057 &PL_body_roots[SVt_PVGV])
1061 /* no arena for you! */
1063 #define new_NOARENA(details) \
1064 safemalloc((details)->body_size + (details)->offset)
1065 #define new_NOARENAZ(details) \
1066 safecalloc((details)->body_size + (details)->offset, 1)
1069 Perl_more_bodies (pTHX_ const svtype sv_type, const size_t body_size,
1070 const size_t arena_size)
1072 void ** const root = &PL_body_roots[sv_type];
1073 struct arena_desc *adesc;
1074 struct arena_set *aroot = (struct arena_set *) PL_body_arenas;
1078 const size_t good_arena_size = Perl_malloc_good_size(arena_size);
1079 #if defined(DEBUGGING) && defined(PERL_GLOBAL_STRUCT)
1082 #if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT_PRIVATE)
1083 static bool done_sanity_check;
1085 /* PERL_GLOBAL_STRUCT_PRIVATE cannot coexist with global
1086 * variables like done_sanity_check. */
1087 if (!done_sanity_check) {
1088 unsigned int i = SVt_LAST;
1090 done_sanity_check = TRUE;
1093 assert (bodies_by_type[i].type == i);
1099 /* may need new arena-set to hold new arena */
1100 if (!aroot || aroot->curr >= aroot->set_size) {
1101 struct arena_set *newroot;
1102 Newxz(newroot, 1, struct arena_set);
1103 newroot->set_size = ARENAS_PER_SET;
1104 newroot->next = aroot;
1106 PL_body_arenas = (void *) newroot;
1107 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot));
1110 /* ok, now have arena-set with at least 1 empty/available arena-desc */
1111 curr = aroot->curr++;
1112 adesc = &(aroot->set[curr]);
1113 assert(!adesc->arena);
1115 Newx(adesc->arena, good_arena_size, char);
1116 adesc->size = good_arena_size;
1117 adesc->utype = sv_type;
1118 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %"UVuf"\n",
1119 curr, (void*)adesc->arena, (UV)good_arena_size));
1121 start = (char *) adesc->arena;
1123 /* Get the address of the byte after the end of the last body we can fit.
1124 Remember, this is integer division: */
1125 end = start + good_arena_size / body_size * body_size;
1127 /* computed count doesn't reflect the 1st slot reservation */
1128 #if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE)
1129 DEBUG_m(PerlIO_printf(Perl_debug_log,
1130 "arena %p end %p arena-size %d (from %d) type %d "
1132 (void*)start, (void*)end, (int)good_arena_size,
1133 (int)arena_size, sv_type, (int)body_size,
1134 (int)good_arena_size / (int)body_size));
1136 DEBUG_m(PerlIO_printf(Perl_debug_log,
1137 "arena %p end %p arena-size %d type %d size %d ct %d\n",
1138 (void*)start, (void*)end,
1139 (int)arena_size, sv_type, (int)body_size,
1140 (int)good_arena_size / (int)body_size));
1142 *root = (void *)start;
1145 /* Where the next body would start: */
1146 char * const next = start + body_size;
1149 /* This is the last body: */
1150 assert(next == end);
1152 *(void **)start = 0;
1156 *(void**) start = (void *)next;
1161 /* grab a new thing from the free list, allocating more if necessary.
1162 The inline version is used for speed in hot routines, and the
1163 function using it serves the rest (unless PURIFY).
1165 #define new_body_inline(xpv, sv_type) \
1167 void ** const r3wt = &PL_body_roots[sv_type]; \
1168 xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \
1169 ? *((void **)(r3wt)) : Perl_more_bodies(aTHX_ sv_type, \
1170 bodies_by_type[sv_type].body_size,\
1171 bodies_by_type[sv_type].arena_size)); \
1172 *(r3wt) = *(void**)(xpv); \
1178 S_new_body(pTHX_ const svtype sv_type)
1181 new_body_inline(xpv, sv_type);
1187 static const struct body_details fake_rv =
1188 { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 };
1191 =for apidoc sv_upgrade
1193 Upgrade an SV to a more complex form. Generally adds a new body type to the
1194 SV, then copies across as much information as possible from the old body.
1195 It croaks if the SV is already in a more complex form than requested. You
1196 generally want to use the C<SvUPGRADE> macro wrapper, which checks the type
1197 before calling C<sv_upgrade>, and hence does not croak. See also
1204 Perl_sv_upgrade(pTHX_ SV *const sv, svtype new_type)
1208 const svtype old_type = SvTYPE(sv);
1209 const struct body_details *new_type_details;
1210 const struct body_details *old_type_details
1211 = bodies_by_type + old_type;
1212 SV *referant = NULL;
1214 PERL_ARGS_ASSERT_SV_UPGRADE;
1216 if (old_type == new_type)
1219 /* This clause was purposefully added ahead of the early return above to
1220 the shared string hackery for (sort {$a <=> $b} keys %hash), with the
1221 inference by Nick I-S that it would fix other troublesome cases. See
1222 changes 7162, 7163 (f130fd4589cf5fbb24149cd4db4137c8326f49c1 and parent)
1224 Given that shared hash key scalars are no longer PVIV, but PV, there is
1225 no longer need to unshare so as to free up the IVX slot for its proper
1226 purpose. So it's safe to move the early return earlier. */
1228 if (new_type > SVt_PVMG && SvIsCOW(sv)) {
1229 sv_force_normal_flags(sv, 0);
1232 old_body = SvANY(sv);
1234 /* Copying structures onto other structures that have been neatly zeroed
1235 has a subtle gotcha. Consider XPVMG
1237 +------+------+------+------+------+-------+-------+
1238 | NV | CUR | LEN | IV | MAGIC | STASH |
1239 +------+------+------+------+------+-------+-------+
1240 0 4 8 12 16 20 24 28
1242 where NVs are aligned to 8 bytes, so that sizeof that structure is
1243 actually 32 bytes long, with 4 bytes of padding at the end:
1245 +------+------+------+------+------+-------+-------+------+
1246 | NV | CUR | LEN | IV | MAGIC | STASH | ??? |
1247 +------+------+------+------+------+-------+-------+------+
1248 0 4 8 12 16 20 24 28 32
1250 so what happens if you allocate memory for this structure:
1252 +------+------+------+------+------+-------+-------+------+------+...
1253 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME |
1254 +------+------+------+------+------+-------+-------+------+------+...
1255 0 4 8 12 16 20 24 28 32 36
1257 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you
1258 expect, because you copy the area marked ??? onto GP. Now, ??? may have
1259 started out as zero once, but it's quite possible that it isn't. So now,
1260 rather than a nicely zeroed GP, you have it pointing somewhere random.
1263 (In fact, GP ends up pointing at a previous GP structure, because the
1264 principle cause of the padding in XPVMG getting garbage is a copy of
1265 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now
1266 this happens to be moot because XPVGV has been re-ordered, with GP
1267 no longer after STASH)
1269 So we are careful and work out the size of used parts of all the
1277 referant = SvRV(sv);
1278 old_type_details = &fake_rv;
1279 if (new_type == SVt_NV)
1280 new_type = SVt_PVNV;
1282 if (new_type < SVt_PVIV) {
1283 new_type = (new_type == SVt_NV)
1284 ? SVt_PVNV : SVt_PVIV;
1289 if (new_type < SVt_PVNV) {
1290 new_type = SVt_PVNV;
1294 assert(new_type > SVt_PV);
1295 assert(SVt_IV < SVt_PV);
1296 assert(SVt_NV < SVt_PV);
1303 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena,
1304 there's no way that it can be safely upgraded, because perl.c
1305 expects to Safefree(SvANY(PL_mess_sv)) */
1306 assert(sv != PL_mess_sv);
1307 /* This flag bit is used to mean other things in other scalar types.
1308 Given that it only has meaning inside the pad, it shouldn't be set
1309 on anything that can get upgraded. */
1310 assert(!SvPAD_TYPED(sv));
1313 if (UNLIKELY(old_type_details->cant_upgrade))
1314 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf,
1315 sv_reftype(sv, 0), (UV) old_type, (UV) new_type);
1318 if (UNLIKELY(old_type > new_type))
1319 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d",
1320 (int)old_type, (int)new_type);
1322 new_type_details = bodies_by_type + new_type;
1324 SvFLAGS(sv) &= ~SVTYPEMASK;
1325 SvFLAGS(sv) |= new_type;
1327 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of
1328 the return statements above will have triggered. */
1329 assert (new_type != SVt_NULL);
1332 assert(old_type == SVt_NULL);
1333 SvANY(sv) = (XPVIV*)((char*)&(sv->sv_u.svu_iv) - STRUCT_OFFSET(XPVIV, xiv_iv));
1337 assert(old_type == SVt_NULL);
1338 #if NVSIZE <= IVSIZE
1339 SvANY(sv) = (XPVNV*)((char*)&(sv->sv_u.svu_nv) - STRUCT_OFFSET(XPVNV, xnv_u.xnv_nv));
1341 SvANY(sv) = new_XNV();
1347 assert(new_type_details->body_size);
1350 assert(new_type_details->arena);
1351 assert(new_type_details->arena_size);
1352 /* This points to the start of the allocated area. */
1353 new_body_inline(new_body, new_type);
1354 Zero(new_body, new_type_details->body_size, char);
1355 new_body = ((char *)new_body) - new_type_details->offset;
1357 /* We always allocated the full length item with PURIFY. To do this
1358 we fake things so that arena is false for all 16 types.. */
1359 new_body = new_NOARENAZ(new_type_details);
1361 SvANY(sv) = new_body;
1362 if (new_type == SVt_PVAV) {
1366 if (old_type_details->body_size) {
1369 /* It will have been zeroed when the new body was allocated.
1370 Lets not write to it, in case it confuses a write-back
1376 #ifndef NODEFAULT_SHAREKEYS
1377 HvSHAREKEYS_on(sv); /* key-sharing on by default */
1379 /* start with PERL_HASH_DEFAULT_HvMAX+1 buckets: */
1380 HvMAX(sv) = PERL_HASH_DEFAULT_HvMAX;
1383 /* SVt_NULL isn't the only thing upgraded to AV or HV.
1384 The target created by newSVrv also is, and it can have magic.
1385 However, it never has SvPVX set.
1387 if (old_type == SVt_IV) {
1389 } else if (old_type >= SVt_PV) {
1390 assert(SvPVX_const(sv) == 0);
1393 if (old_type >= SVt_PVMG) {
1394 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic);
1395 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash);
1397 sv->sv_u.svu_array = NULL; /* or svu_hash */
1402 /* XXX Is this still needed? Was it ever needed? Surely as there is
1403 no route from NV to PVIV, NOK can never be true */
1404 assert(!SvNOKp(sv));
1417 assert(new_type_details->body_size);
1418 /* We always allocated the full length item with PURIFY. To do this
1419 we fake things so that arena is false for all 16 types.. */
1420 if(new_type_details->arena) {
1421 /* This points to the start of the allocated area. */
1422 new_body_inline(new_body, new_type);
1423 Zero(new_body, new_type_details->body_size, char);
1424 new_body = ((char *)new_body) - new_type_details->offset;
1426 new_body = new_NOARENAZ(new_type_details);
1428 SvANY(sv) = new_body;
1430 if (old_type_details->copy) {
1431 /* There is now the potential for an upgrade from something without
1432 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */
1433 int offset = old_type_details->offset;
1434 int length = old_type_details->copy;
1436 if (new_type_details->offset > old_type_details->offset) {
1437 const int difference
1438 = new_type_details->offset - old_type_details->offset;
1439 offset += difference;
1440 length -= difference;
1442 assert (length >= 0);
1444 Copy((char *)old_body + offset, (char *)new_body + offset, length,
1448 #ifndef NV_ZERO_IS_ALLBITS_ZERO
1449 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a
1450 * correct 0.0 for us. Otherwise, if the old body didn't have an
1451 * NV slot, but the new one does, then we need to initialise the
1452 * freshly created NV slot with whatever the correct bit pattern is
1454 if (old_type_details->zero_nv && !new_type_details->zero_nv
1455 && !isGV_with_GP(sv))
1459 if (UNLIKELY(new_type == SVt_PVIO)) {
1460 IO * const io = MUTABLE_IO(sv);
1461 GV *iogv = gv_fetchpvs("IO::File::", GV_ADD, SVt_PVHV);
1464 /* Clear the stashcache because a new IO could overrule a package
1466 DEBUG_o(Perl_deb(aTHX_ "sv_upgrade clearing PL_stashcache\n"));
1467 hv_clear(PL_stashcache);
1469 SvSTASH_set(io, MUTABLE_HV(SvREFCNT_inc(GvHV(iogv))));
1470 IoPAGE_LEN(sv) = 60;
1472 if (UNLIKELY(new_type == SVt_REGEXP))
1473 sv->sv_u.svu_rx = (regexp *)new_body;
1474 else if (old_type < SVt_PV) {
1475 /* referant will be NULL unless the old type was SVt_IV emulating
1477 sv->sv_u.svu_rv = referant;
1481 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1482 (unsigned long)new_type);
1485 /* if this is zero, this is a body-less SVt_NULL, SVt_IV/SVt_RV,
1486 and sometimes SVt_NV */
1487 if (old_type_details->body_size) {
1491 /* Note that there is an assumption that all bodies of types that
1492 can be upgraded came from arenas. Only the more complex non-
1493 upgradable types are allowed to be directly malloc()ed. */
1494 assert(old_type_details->arena);
1495 del_body((void*)((char*)old_body + old_type_details->offset),
1496 &PL_body_roots[old_type]);
1502 =for apidoc sv_backoff
1504 Remove any string offset. You should normally use the C<SvOOK_off> macro
1511 Perl_sv_backoff(SV *const sv)
1514 const char * const s = SvPVX_const(sv);
1516 PERL_ARGS_ASSERT_SV_BACKOFF;
1519 assert(SvTYPE(sv) != SVt_PVHV);
1520 assert(SvTYPE(sv) != SVt_PVAV);
1522 SvOOK_offset(sv, delta);
1524 SvLEN_set(sv, SvLEN(sv) + delta);
1525 SvPV_set(sv, SvPVX(sv) - delta);
1526 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
1527 SvFLAGS(sv) &= ~SVf_OOK;
1534 Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1535 upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1536 Use the C<SvGROW> wrapper instead.
1541 static void S_sv_uncow(pTHX_ SV * const sv, const U32 flags);
1544 Perl_sv_grow(pTHX_ SV *const sv, STRLEN newlen)
1548 PERL_ARGS_ASSERT_SV_GROW;
1552 if (SvTYPE(sv) < SVt_PV) {
1553 sv_upgrade(sv, SVt_PV);
1554 s = SvPVX_mutable(sv);
1556 else if (SvOOK(sv)) { /* pv is offset? */
1558 s = SvPVX_mutable(sv);
1559 if (newlen > SvLEN(sv))
1560 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
1564 if (SvIsCOW(sv)) S_sv_uncow(aTHX_ sv, 0);
1565 s = SvPVX_mutable(sv);
1568 #ifdef PERL_NEW_COPY_ON_WRITE
1569 /* the new COW scheme uses SvPVX(sv)[SvLEN(sv)-1] (if spare)
1570 * to store the COW count. So in general, allocate one more byte than
1571 * asked for, to make it likely this byte is always spare: and thus
1572 * make more strings COW-able.
1573 * If the new size is a big power of two, don't bother: we assume the
1574 * caller wanted a nice 2^N sized block and will be annoyed at getting
1580 #if defined(PERL_USE_MALLOC_SIZE) && defined(Perl_safesysmalloc_size)
1581 #define PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1584 if (newlen > SvLEN(sv)) { /* need more room? */
1585 STRLEN minlen = SvCUR(sv);
1586 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
1587 if (newlen < minlen)
1589 #ifndef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1591 /* Don't round up on the first allocation, as odds are pretty good that
1592 * the initial request is accurate as to what is really needed */
1594 newlen = PERL_STRLEN_ROUNDUP(newlen);
1597 if (SvLEN(sv) && s) {
1598 s = (char*)saferealloc(s, newlen);
1601 s = (char*)safemalloc(newlen);
1602 if (SvPVX_const(sv) && SvCUR(sv)) {
1603 Move(SvPVX_const(sv), s, (newlen < SvCUR(sv)) ? newlen : SvCUR(sv), char);
1607 #ifdef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1608 /* Do this here, do it once, do it right, and then we will never get
1609 called back into sv_grow() unless there really is some growing
1611 SvLEN_set(sv, Perl_safesysmalloc_size(s));
1613 SvLEN_set(sv, newlen);
1620 =for apidoc sv_setiv
1622 Copies an integer into the given SV, upgrading first if necessary.
1623 Does not handle 'set' magic. See also C<sv_setiv_mg>.
1629 Perl_sv_setiv(pTHX_ SV *const sv, const IV i)
1631 PERL_ARGS_ASSERT_SV_SETIV;
1633 SV_CHECK_THINKFIRST_COW_DROP(sv);
1634 switch (SvTYPE(sv)) {
1637 sv_upgrade(sv, SVt_IV);
1640 sv_upgrade(sv, SVt_PVIV);
1644 if (!isGV_with_GP(sv))
1651 /* diag_listed_as: Can't coerce %s to %s in %s */
1652 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1656 (void)SvIOK_only(sv); /* validate number */
1662 =for apidoc sv_setiv_mg
1664 Like C<sv_setiv>, but also handles 'set' magic.
1670 Perl_sv_setiv_mg(pTHX_ SV *const sv, const IV i)
1672 PERL_ARGS_ASSERT_SV_SETIV_MG;
1679 =for apidoc sv_setuv
1681 Copies an unsigned integer into the given SV, upgrading first if necessary.
1682 Does not handle 'set' magic. See also C<sv_setuv_mg>.
1688 Perl_sv_setuv(pTHX_ SV *const sv, const UV u)
1690 PERL_ARGS_ASSERT_SV_SETUV;
1692 /* With the if statement to ensure that integers are stored as IVs whenever
1694 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
1697 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1699 If you wish to remove the following if statement, so that this routine
1700 (and its callers) always return UVs, please benchmark to see what the
1701 effect is. Modern CPUs may be different. Or may not :-)
1703 if (u <= (UV)IV_MAX) {
1704 sv_setiv(sv, (IV)u);
1713 =for apidoc sv_setuv_mg
1715 Like C<sv_setuv>, but also handles 'set' magic.
1721 Perl_sv_setuv_mg(pTHX_ SV *const sv, const UV u)
1723 PERL_ARGS_ASSERT_SV_SETUV_MG;
1730 =for apidoc sv_setnv
1732 Copies a double into the given SV, upgrading first if necessary.
1733 Does not handle 'set' magic. See also C<sv_setnv_mg>.
1739 Perl_sv_setnv(pTHX_ SV *const sv, const NV num)
1741 PERL_ARGS_ASSERT_SV_SETNV;
1743 SV_CHECK_THINKFIRST_COW_DROP(sv);
1744 switch (SvTYPE(sv)) {
1747 sv_upgrade(sv, SVt_NV);
1751 sv_upgrade(sv, SVt_PVNV);
1755 if (!isGV_with_GP(sv))
1762 /* diag_listed_as: Can't coerce %s to %s in %s */
1763 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
1768 (void)SvNOK_only(sv); /* validate number */
1773 =for apidoc sv_setnv_mg
1775 Like C<sv_setnv>, but also handles 'set' magic.
1781 Perl_sv_setnv_mg(pTHX_ SV *const sv, const NV num)
1783 PERL_ARGS_ASSERT_SV_SETNV_MG;
1789 /* Return a cleaned-up, printable version of sv, for non-numeric, or
1790 * not incrementable warning display.
1791 * Originally part of S_not_a_number().
1792 * The return value may be != tmpbuf.
1796 S_sv_display(pTHX_ SV *const sv, char *tmpbuf, STRLEN tmpbuf_size) {
1799 PERL_ARGS_ASSERT_SV_DISPLAY;
1802 SV *dsv = newSVpvs_flags("", SVs_TEMP);
1803 pv = sv_uni_display(dsv, sv, 10, UNI_DISPLAY_ISPRINT);
1806 const char * const limit = tmpbuf + tmpbuf_size - 8;
1807 /* each *s can expand to 4 chars + "...\0",
1808 i.e. need room for 8 chars */
1810 const char *s = SvPVX_const(sv);
1811 const char * const end = s + SvCUR(sv);
1812 for ( ; s < end && d < limit; s++ ) {
1814 if (! isASCII(ch) && !isPRINT_LC(ch)) {
1818 /* Map to ASCII "equivalent" of Latin1 */
1819 ch = LATIN1_TO_NATIVE(NATIVE_TO_LATIN1(ch) & 127);
1825 else if (ch == '\r') {
1829 else if (ch == '\f') {
1833 else if (ch == '\\') {
1837 else if (ch == '\0') {
1841 else if (isPRINT_LC(ch))
1860 /* Print an "isn't numeric" warning, using a cleaned-up,
1861 * printable version of the offending string
1865 S_not_a_number(pTHX_ SV *const sv)
1870 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1872 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1875 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1876 /* diag_listed_as: Argument "%s" isn't numeric%s */
1877 "Argument \"%s\" isn't numeric in %s", pv,
1880 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1881 /* diag_listed_as: Argument "%s" isn't numeric%s */
1882 "Argument \"%s\" isn't numeric", pv);
1886 S_not_incrementable(pTHX_ SV *const sv) {
1890 PERL_ARGS_ASSERT_NOT_INCREMENTABLE;
1892 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1894 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1895 "Argument \"%s\" treated as 0 in increment (++)", pv);
1899 =for apidoc looks_like_number
1901 Test if the content of an SV looks like a number (or is a number).
1902 C<Inf> and C<Infinity> are treated as numbers (so will not issue a
1903 non-numeric warning), even if your atof() doesn't grok them. Get-magic is
1910 Perl_looks_like_number(pTHX_ SV *const sv)
1915 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1917 if (SvPOK(sv) || SvPOKp(sv)) {
1918 sbegin = SvPV_nomg_const(sv, len);
1921 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
1922 return grok_number(sbegin, len, NULL);
1926 S_glob_2number(pTHX_ GV * const gv)
1928 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1930 /* We know that all GVs stringify to something that is not-a-number,
1931 so no need to test that. */
1932 if (ckWARN(WARN_NUMERIC))
1934 SV *const buffer = sv_newmortal();
1935 gv_efullname3(buffer, gv, "*");
1936 not_a_number(buffer);
1938 /* We just want something true to return, so that S_sv_2iuv_common
1939 can tail call us and return true. */
1943 /* Actually, ISO C leaves conversion of UV to IV undefined, but
1944 until proven guilty, assume that things are not that bad... */
1949 As 64 bit platforms often have an NV that doesn't preserve all bits of
1950 an IV (an assumption perl has been based on to date) it becomes necessary
1951 to remove the assumption that the NV always carries enough precision to
1952 recreate the IV whenever needed, and that the NV is the canonical form.
1953 Instead, IV/UV and NV need to be given equal rights. So as to not lose
1954 precision as a side effect of conversion (which would lead to insanity
1955 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
1956 1) to distinguish between IV/UV/NV slots that have a valid conversion cached
1957 where precision was lost, and IV/UV/NV slots that have a valid conversion
1958 which has lost no precision
1959 2) to ensure that if a numeric conversion to one form is requested that
1960 would lose precision, the precise conversion (or differently
1961 imprecise conversion) is also performed and cached, to prevent
1962 requests for different numeric formats on the same SV causing
1963 lossy conversion chains. (lossless conversion chains are perfectly
1968 SvIOKp is true if the IV slot contains a valid value
1969 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1970 SvNOKp is true if the NV slot contains a valid value
1971 SvNOK is true only if the NV value is accurate
1974 while converting from PV to NV, check to see if converting that NV to an
1975 IV(or UV) would lose accuracy over a direct conversion from PV to
1976 IV(or UV). If it would, cache both conversions, return NV, but mark
1977 SV as IOK NOKp (ie not NOK).
1979 While converting from PV to IV, check to see if converting that IV to an
1980 NV would lose accuracy over a direct conversion from PV to NV. If it
1981 would, cache both conversions, flag similarly.
1983 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
1984 correctly because if IV & NV were set NV *always* overruled.
1985 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
1986 changes - now IV and NV together means that the two are interchangeable:
1987 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
1989 The benefit of this is that operations such as pp_add know that if
1990 SvIOK is true for both left and right operands, then integer addition
1991 can be used instead of floating point (for cases where the result won't
1992 overflow). Before, floating point was always used, which could lead to
1993 loss of precision compared with integer addition.
1995 * making IV and NV equal status should make maths accurate on 64 bit
1997 * may speed up maths somewhat if pp_add and friends start to use
1998 integers when possible instead of fp. (Hopefully the overhead in
1999 looking for SvIOK and checking for overflow will not outweigh the
2000 fp to integer speedup)
2001 * will slow down integer operations (callers of SvIV) on "inaccurate"
2002 values, as the change from SvIOK to SvIOKp will cause a call into
2003 sv_2iv each time rather than a macro access direct to the IV slot
2004 * should speed up number->string conversion on integers as IV is
2005 favoured when IV and NV are equally accurate
2007 ####################################################################
2008 You had better be using SvIOK_notUV if you want an IV for arithmetic:
2009 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
2010 On the other hand, SvUOK is true iff UV.
2011 ####################################################################
2013 Your mileage will vary depending your CPU's relative fp to integer
2017 #ifndef NV_PRESERVES_UV
2018 # define IS_NUMBER_UNDERFLOW_IV 1
2019 # define IS_NUMBER_UNDERFLOW_UV 2
2020 # define IS_NUMBER_IV_AND_UV 2
2021 # define IS_NUMBER_OVERFLOW_IV 4
2022 # define IS_NUMBER_OVERFLOW_UV 5
2024 /* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
2026 /* For sv_2nv these three cases are "SvNOK and don't bother casting" */
2028 S_sv_2iuv_non_preserve(pTHX_ SV *const sv
2034 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
2035 PERL_UNUSED_CONTEXT;
2037 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));
2038 if (SvNVX(sv) < (NV)IV_MIN) {
2039 (void)SvIOKp_on(sv);
2041 SvIV_set(sv, IV_MIN);
2042 return IS_NUMBER_UNDERFLOW_IV;
2044 if (SvNVX(sv) > (NV)UV_MAX) {
2045 (void)SvIOKp_on(sv);
2048 SvUV_set(sv, UV_MAX);
2049 return IS_NUMBER_OVERFLOW_UV;
2051 (void)SvIOKp_on(sv);
2053 /* Can't use strtol etc to convert this string. (See truth table in
2055 if (SvNVX(sv) <= (UV)IV_MAX) {
2056 SvIV_set(sv, I_V(SvNVX(sv)));
2057 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2058 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
2060 /* Integer is imprecise. NOK, IOKp */
2062 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
2065 SvUV_set(sv, U_V(SvNVX(sv)));
2066 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2067 if (SvUVX(sv) == UV_MAX) {
2068 /* As we know that NVs don't preserve UVs, UV_MAX cannot
2069 possibly be preserved by NV. Hence, it must be overflow.
2071 return IS_NUMBER_OVERFLOW_UV;
2073 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
2075 /* Integer is imprecise. NOK, IOKp */
2077 return IS_NUMBER_OVERFLOW_IV;
2079 #endif /* !NV_PRESERVES_UV*/
2081 /* If numtype is infnan, set the NV of the sv accordingly.
2082 * If numtype is anything else, try setting the NV using Atof(PV). */
2084 S_sv_setnv(pTHX_ SV* sv, int numtype)
2086 bool pok = cBOOL(SvPOK(sv));
2088 if ((numtype & IS_NUMBER_INFINITY)) {
2089 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -NV_INF : NV_INF);
2092 else if ((numtype & IS_NUMBER_NAN)) {
2093 SvNV_set(sv, NV_NAN);
2097 SvNV_set(sv, Atof(SvPVX_const(sv)));
2098 /* Purposefully no true nok here, since we don't want to blow
2099 * away the possible IOK/UV of an existing sv. */
2102 SvNOK_only(sv); /* No IV or UV please, this is pure infnan. */
2104 SvPOK_on(sv); /* PV is okay, though. */
2109 S_sv_2iuv_common(pTHX_ SV *const sv)
2111 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
2114 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
2115 * without also getting a cached IV/UV from it at the same time
2116 * (ie PV->NV conversion should detect loss of accuracy and cache
2117 * IV or UV at same time to avoid this. */
2118 /* IV-over-UV optimisation - choose to cache IV if possible */
2120 if (UNLIKELY(Perl_isinfnan(SvNVX(sv))))
2123 if (SvTYPE(sv) == SVt_NV)
2124 sv_upgrade(sv, SVt_PVNV);
2126 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
2127 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
2128 certainly cast into the IV range at IV_MAX, whereas the correct
2129 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
2131 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2132 SvIV_set(sv, I_V(SvNVX(sv)));
2133 if (SvNVX(sv) == (NV) SvIVX(sv)
2134 #ifndef NV_PRESERVES_UV
2135 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2136 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2137 /* Don't flag it as "accurately an integer" if the number
2138 came from a (by definition imprecise) NV operation, and
2139 we're outside the range of NV integer precision */
2143 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2145 /* scalar has trailing garbage, eg "42a" */
2147 DEBUG_c(PerlIO_printf(Perl_debug_log,
2148 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (precise)\n",
2154 /* IV not precise. No need to convert from PV, as NV
2155 conversion would already have cached IV if it detected
2156 that PV->IV would be better than PV->NV->IV
2157 flags already correct - don't set public IOK. */
2158 DEBUG_c(PerlIO_printf(Perl_debug_log,
2159 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (imprecise)\n",
2164 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2165 but the cast (NV)IV_MIN rounds to a the value less (more
2166 negative) than IV_MIN which happens to be equal to SvNVX ??
2167 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2168 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2169 (NV)UVX == NVX are both true, but the values differ. :-(
2170 Hopefully for 2s complement IV_MIN is something like
2171 0x8000000000000000 which will be exact. NWC */
2174 SvUV_set(sv, U_V(SvNVX(sv)));
2176 (SvNVX(sv) == (NV) SvUVX(sv))
2177 #ifndef NV_PRESERVES_UV
2178 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2179 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2180 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2181 /* Don't flag it as "accurately an integer" if the number
2182 came from a (by definition imprecise) NV operation, and
2183 we're outside the range of NV integer precision */
2189 DEBUG_c(PerlIO_printf(Perl_debug_log,
2190 "0x%"UVxf" 2iv(%"UVuf" => %"IVdf") (as unsigned)\n",
2196 else if (SvPOKp(sv)) {
2198 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2199 /* We want to avoid a possible problem when we cache an IV/ a UV which
2200 may be later translated to an NV, and the resulting NV is not
2201 the same as the direct translation of the initial string
2202 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2203 be careful to ensure that the value with the .456 is around if the
2204 NV value is requested in the future).
2206 This means that if we cache such an IV/a UV, we need to cache the
2207 NV as well. Moreover, we trade speed for space, and do not
2208 cache the NV if we are sure it's not needed.
2211 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2212 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2213 == IS_NUMBER_IN_UV) {
2214 /* It's definitely an integer, only upgrade to PVIV */
2215 if (SvTYPE(sv) < SVt_PVIV)
2216 sv_upgrade(sv, SVt_PVIV);
2218 } else if (SvTYPE(sv) < SVt_PVNV)
2219 sv_upgrade(sv, SVt_PVNV);
2221 if ((numtype & (IS_NUMBER_INFINITY | IS_NUMBER_NAN))) {
2222 S_sv_setnv(aTHX_ sv, numtype);
2226 /* If NVs preserve UVs then we only use the UV value if we know that
2227 we aren't going to call atof() below. If NVs don't preserve UVs
2228 then the value returned may have more precision than atof() will
2229 return, even though value isn't perfectly accurate. */
2230 if ((numtype & (IS_NUMBER_IN_UV
2231 #ifdef NV_PRESERVES_UV
2234 )) == IS_NUMBER_IN_UV) {
2235 /* This won't turn off the public IOK flag if it was set above */
2236 (void)SvIOKp_on(sv);
2238 if (!(numtype & IS_NUMBER_NEG)) {
2240 if (value <= (UV)IV_MAX) {
2241 SvIV_set(sv, (IV)value);
2243 /* it didn't overflow, and it was positive. */
2244 SvUV_set(sv, value);
2248 /* 2s complement assumption */
2249 if (value <= (UV)IV_MIN) {
2250 SvIV_set(sv, -(IV)value);
2252 /* Too negative for an IV. This is a double upgrade, but
2253 I'm assuming it will be rare. */
2254 if (SvTYPE(sv) < SVt_PVNV)
2255 sv_upgrade(sv, SVt_PVNV);
2259 SvNV_set(sv, -(NV)value);
2260 SvIV_set(sv, IV_MIN);
2264 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2265 will be in the previous block to set the IV slot, and the next
2266 block to set the NV slot. So no else here. */
2268 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2269 != IS_NUMBER_IN_UV) {
2270 /* It wasn't an (integer that doesn't overflow the UV). */
2271 S_sv_setnv(aTHX_ sv, numtype);
2273 if (! numtype && ckWARN(WARN_NUMERIC))
2276 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%" NVgf ")\n",
2277 PTR2UV(sv), SvNVX(sv)));
2279 #ifdef NV_PRESERVES_UV
2280 (void)SvIOKp_on(sv);
2282 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2283 SvIV_set(sv, I_V(SvNVX(sv)));
2284 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2287 NOOP; /* Integer is imprecise. NOK, IOKp */
2289 /* UV will not work better than IV */
2291 if (SvNVX(sv) > (NV)UV_MAX) {
2293 /* Integer is inaccurate. NOK, IOKp, is UV */
2294 SvUV_set(sv, UV_MAX);
2296 SvUV_set(sv, U_V(SvNVX(sv)));
2297 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2298 NV preservse UV so can do correct comparison. */
2299 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2302 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
2307 #else /* NV_PRESERVES_UV */
2308 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2309 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
2310 /* The IV/UV slot will have been set from value returned by
2311 grok_number above. The NV slot has just been set using
2314 assert (SvIOKp(sv));
2316 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2317 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2318 /* Small enough to preserve all bits. */
2319 (void)SvIOKp_on(sv);
2321 SvIV_set(sv, I_V(SvNVX(sv)));
2322 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2324 /* Assumption: first non-preserved integer is < IV_MAX,
2325 this NV is in the preserved range, therefore: */
2326 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2328 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);
2332 0 0 already failed to read UV.
2333 0 1 already failed to read UV.
2334 1 0 you won't get here in this case. IV/UV
2335 slot set, public IOK, Atof() unneeded.
2336 1 1 already read UV.
2337 so there's no point in sv_2iuv_non_preserve() attempting
2338 to use atol, strtol, strtoul etc. */
2340 sv_2iuv_non_preserve (sv, numtype);
2342 sv_2iuv_non_preserve (sv);
2346 #endif /* NV_PRESERVES_UV */
2347 /* It might be more code efficient to go through the entire logic above
2348 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2349 gets complex and potentially buggy, so more programmer efficient
2350 to do it this way, by turning off the public flags: */
2352 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2356 if (isGV_with_GP(sv))
2357 return glob_2number(MUTABLE_GV(sv));
2359 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2361 if (SvTYPE(sv) < SVt_IV)
2362 /* Typically the caller expects that sv_any is not NULL now. */
2363 sv_upgrade(sv, SVt_IV);
2364 /* Return 0 from the caller. */
2371 =for apidoc sv_2iv_flags
2373 Return the integer value of an SV, doing any necessary string
2374 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2375 Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2381 Perl_sv_2iv_flags(pTHX_ SV *const sv, const I32 flags)
2383 PERL_ARGS_ASSERT_SV_2IV_FLAGS;
2385 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2386 && SvTYPE(sv) != SVt_PVFM);
2388 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2391 if (SvNOK(sv) && UNLIKELY(Perl_isinfnan(SvNVX(sv))))
2392 return 0; /* So wrong but what can we do. */
2397 if (flags & SV_SKIP_OVERLOAD)
2399 tmpstr = AMG_CALLunary(sv, numer_amg);
2400 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2401 return SvIV(tmpstr);
2404 return PTR2IV(SvRV(sv));
2407 if (SvVALID(sv) || isREGEXP(sv)) {
2408 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2409 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2410 In practice they are extremely unlikely to actually get anywhere
2411 accessible by user Perl code - the only way that I'm aware of is when
2412 a constant subroutine which is used as the second argument to index.
2414 Regexps have no SvIVX and SvNVX fields.
2416 assert(isREGEXP(sv) || SvPOKp(sv));
2419 const char * const ptr =
2420 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2421 const int numtype = grok_number(ptr, SvCUR(sv), &value);
2423 assert((numtype & (IS_NUMBER_INFINITY | IS_NUMBER_NAN)) == 0);
2425 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2426 == IS_NUMBER_IN_UV) {
2427 /* It's definitely an integer */
2428 if (numtype & IS_NUMBER_NEG) {
2429 if (value < (UV)IV_MIN)
2432 if (value < (UV)IV_MAX)
2438 if (ckWARN(WARN_NUMERIC))
2441 return I_V(Atof(ptr));
2445 if (SvTHINKFIRST(sv)) {
2446 #ifdef PERL_OLD_COPY_ON_WRITE
2448 sv_force_normal_flags(sv, 0);
2451 if (SvREADONLY(sv) && !SvOK(sv)) {
2452 if (ckWARN(WARN_UNINITIALIZED))
2459 if (S_sv_2iuv_common(aTHX_ sv))
2463 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"IVdf")\n",
2464 PTR2UV(sv),SvIVX(sv)));
2465 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
2469 =for apidoc sv_2uv_flags
2471 Return the unsigned integer value of an SV, doing any necessary string
2472 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2473 Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
2479 Perl_sv_2uv_flags(pTHX_ SV *const sv, const I32 flags)
2481 PERL_ARGS_ASSERT_SV_2UV_FLAGS;
2483 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2486 if (SvNOK(sv) && UNLIKELY(Perl_isinfnan(SvNVX(sv))))
2487 return 0; /* So wrong but what can we do. */
2492 if (flags & SV_SKIP_OVERLOAD)
2494 tmpstr = AMG_CALLunary(sv, numer_amg);
2495 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2496 return SvUV(tmpstr);
2499 return PTR2UV(SvRV(sv));
2502 if (SvVALID(sv) || isREGEXP(sv)) {
2503 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2504 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2505 Regexps have no SvIVX and SvNVX fields. */
2506 assert(isREGEXP(sv) || SvPOKp(sv));
2509 const char * const ptr =
2510 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2511 const int numtype = grok_number(ptr, SvCUR(sv), &value);
2513 assert((numtype & (IS_NUMBER_INFINITY | IS_NUMBER_NAN)) == 0);
2515 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2516 == IS_NUMBER_IN_UV) {
2517 /* It's definitely an integer */
2518 if (!(numtype & IS_NUMBER_NEG))
2523 if (ckWARN(WARN_NUMERIC))
2526 return U_V(Atof(ptr));
2530 if (SvTHINKFIRST(sv)) {
2531 #ifdef PERL_OLD_COPY_ON_WRITE
2533 sv_force_normal_flags(sv, 0);
2536 if (SvREADONLY(sv) && !SvOK(sv)) {
2537 if (ckWARN(WARN_UNINITIALIZED))
2544 if (S_sv_2iuv_common(aTHX_ sv))
2548 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2uv(%"UVuf")\n",
2549 PTR2UV(sv),SvUVX(sv)));
2550 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
2554 =for apidoc sv_2nv_flags
2556 Return the num value of an SV, doing any necessary string or integer
2557 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2558 Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
2564 Perl_sv_2nv_flags(pTHX_ SV *const sv, const I32 flags)
2566 PERL_ARGS_ASSERT_SV_2NV_FLAGS;
2568 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2569 && SvTYPE(sv) != SVt_PVFM);
2570 if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) {
2571 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2572 the same flag bit as SVf_IVisUV, so must not let them cache NVs.
2573 Regexps have no SvIVX and SvNVX fields. */
2575 if (flags & SV_GMAGIC)
2579 if (SvPOKp(sv) && !SvIOKp(sv)) {
2580 ptr = SvPVX_const(sv);
2582 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
2583 !grok_number(ptr, SvCUR(sv), NULL))
2589 return (NV)SvUVX(sv);
2591 return (NV)SvIVX(sv);
2597 ptr = RX_WRAPPED((REGEXP *)sv);
2600 assert(SvTYPE(sv) >= SVt_PVMG);
2601 /* This falls through to the report_uninit near the end of the
2603 } else if (SvTHINKFIRST(sv)) {
2608 if (flags & SV_SKIP_OVERLOAD)
2610 tmpstr = AMG_CALLunary(sv, numer_amg);
2611 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2612 return SvNV(tmpstr);
2615 return PTR2NV(SvRV(sv));
2617 #ifdef PERL_OLD_COPY_ON_WRITE
2619 sv_force_normal_flags(sv, 0);
2622 if (SvREADONLY(sv) && !SvOK(sv)) {
2623 if (ckWARN(WARN_UNINITIALIZED))
2628 if (SvTYPE(sv) < SVt_NV) {
2629 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2630 sv_upgrade(sv, SVt_NV);
2632 STORE_NUMERIC_LOCAL_SET_STANDARD();
2633 PerlIO_printf(Perl_debug_log,
2634 "0x%"UVxf" num(%" NVgf ")\n",
2635 PTR2UV(sv), SvNVX(sv));
2636 RESTORE_NUMERIC_LOCAL();
2639 else if (SvTYPE(sv) < SVt_PVNV)
2640 sv_upgrade(sv, SVt_PVNV);
2645 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
2646 #ifdef NV_PRESERVES_UV
2652 /* Only set the public NV OK flag if this NV preserves the IV */
2653 /* Check it's not 0xFFFFFFFFFFFFFFFF */
2655 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
2656 : (SvIVX(sv) == I_V(SvNVX(sv))))
2662 else if (SvPOKp(sv)) {
2664 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2665 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
2667 #ifdef NV_PRESERVES_UV
2668 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2669 == IS_NUMBER_IN_UV) {
2670 /* It's definitely an integer */
2671 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
2673 S_sv_setnv(aTHX_ sv, numtype);
2680 if ((numtype & IS_NUMBER_INFINITY)) {
2681 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -NV_INF : NV_INF);
2683 } else if ((numtype & IS_NUMBER_NAN)) {
2684 SvNV_set(sv, NV_NAN);
2687 SvNV_set(sv, Atof(SvPVX_const(sv)));
2688 /* Only set the public NV OK flag if this NV preserves the value in
2689 the PV at least as well as an IV/UV would.
2690 Not sure how to do this 100% reliably. */
2691 /* if that shift count is out of range then Configure's test is
2692 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2694 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2695 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2696 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
2697 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2698 /* Can't use strtol etc to convert this string, so don't try.
2699 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2702 /* value has been set. It may not be precise. */
2703 if ((numtype & IS_NUMBER_NEG) && (value > (UV)IV_MIN)) {
2704 /* 2s complement assumption for (UV)IV_MIN */
2705 SvNOK_on(sv); /* Integer is too negative. */
2710 if (numtype & IS_NUMBER_NEG) {
2711 SvIV_set(sv, -(IV)value);
2712 } else if (value <= (UV)IV_MAX) {
2713 SvIV_set(sv, (IV)value);
2715 SvUV_set(sv, value);
2719 if (numtype & IS_NUMBER_NOT_INT) {
2720 /* I believe that even if the original PV had decimals,
2721 they are lost beyond the limit of the FP precision.
2722 However, neither is canonical, so both only get p
2723 flags. NWC, 2000/11/25 */
2724 /* Both already have p flags, so do nothing */
2726 const NV nv = SvNVX(sv);
2727 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2728 if (SvIVX(sv) == I_V(nv)) {
2731 /* It had no "." so it must be integer. */
2735 /* between IV_MAX and NV(UV_MAX).
2736 Could be slightly > UV_MAX */
2738 if (numtype & IS_NUMBER_NOT_INT) {
2739 /* UV and NV both imprecise. */
2741 const UV nv_as_uv = U_V(nv);
2743 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2752 /* It might be more code efficient to go through the entire logic above
2753 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2754 gets complex and potentially buggy, so more programmer efficient
2755 to do it this way, by turning off the public flags: */
2757 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2759 #endif /* NV_PRESERVES_UV */
2762 if (isGV_with_GP(sv)) {
2763 glob_2number(MUTABLE_GV(sv));
2767 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2769 assert (SvTYPE(sv) >= SVt_NV);
2770 /* Typically the caller expects that sv_any is not NULL now. */
2771 /* XXX Ilya implies that this is a bug in callers that assume this
2772 and ideally should be fixed. */
2776 STORE_NUMERIC_LOCAL_SET_STANDARD();
2777 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2nv(%" NVgf ")\n",
2778 PTR2UV(sv), SvNVX(sv));
2779 RESTORE_NUMERIC_LOCAL();
2787 Return an SV with the numeric value of the source SV, doing any necessary
2788 reference or overload conversion. You must use the C<SvNUM(sv)> macro to
2789 access this function.
2795 Perl_sv_2num(pTHX_ SV *const sv)
2797 PERL_ARGS_ASSERT_SV_2NUM;
2802 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
2803 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
2804 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2805 return sv_2num(tmpsv);
2807 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2810 /* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2811 * UV as a string towards the end of buf, and return pointers to start and
2814 * We assume that buf is at least TYPE_CHARS(UV) long.
2818 S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
2820 char *ptr = buf + TYPE_CHARS(UV);
2821 char * const ebuf = ptr;
2824 PERL_ARGS_ASSERT_UIV_2BUF;
2836 *--ptr = '0' + (char)(uv % 10);
2844 /* Helper for sv_2pv_flags and sv_vcatpvfn_flags. If the NV is an
2845 * infinity or a not-a-number, writes the appropriate strings to the
2846 * buffer, including a zero byte. On success returns the written length,
2847 * excluding the zero byte, on failure (not an infinity, not a nan, or the
2848 * maxlen too small) returns zero.
2850 * XXX for "Inf", "-Inf", and "NaN", we could have three read-only
2851 * shared string constants we point to, instead of generating a new
2852 * string for each instance. */
2854 S_infnan_2pv(NV nv, char* buffer, size_t maxlen) {
2855 assert(maxlen >= 4);
2856 if (maxlen < 4) /* "Inf\0", "NaN\0" */
2860 if (Perl_isinf(nv)) {
2862 if (maxlen < 5) /* "-Inf\0" */
2869 } else if (Perl_isnan(nv)) {
2873 /* XXX optionally output the payload mantissa bits as
2874 * "(unsigned)" (to match the nan("...") C99 function,
2875 * or maybe as "(0xhhh...)" would make more sense...
2876 * provide a format string so that the user can decide?
2877 * NOTE: would affect the maxlen and assert() logic.*/
2882 assert((s == buffer + 3) || (s == buffer + 4));
2884 return s - buffer - 1; /* -1: excluding the zero byte */
2889 =for apidoc sv_2pv_flags
2891 Returns a pointer to the string value of an SV, and sets *lp to its length.
2892 If flags includes SV_GMAGIC, does an mg_get() first. Coerces sv to a
2893 string if necessary. Normally invoked via the C<SvPV_flags> macro.
2894 C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
2900 Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
2904 PERL_ARGS_ASSERT_SV_2PV_FLAGS;
2906 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2907 && SvTYPE(sv) != SVt_PVFM);
2908 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2913 if (flags & SV_SKIP_OVERLOAD)
2915 tmpstr = AMG_CALLunary(sv, string_amg);
2916 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
2917 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2919 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2923 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2924 if (flags & SV_CONST_RETURN) {
2925 pv = (char *) SvPVX_const(tmpstr);
2927 pv = (flags & SV_MUTABLE_RETURN)
2928 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
2931 *lp = SvCUR(tmpstr);
2933 pv = sv_2pv_flags(tmpstr, lp, flags);
2946 SV *const referent = SvRV(sv);
2950 retval = buffer = savepvn("NULLREF", len);
2951 } else if (SvTYPE(referent) == SVt_REGEXP &&
2952 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
2953 amagic_is_enabled(string_amg))) {
2954 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
2958 /* If the regex is UTF-8 we want the containing scalar to
2959 have an UTF-8 flag too */
2966 *lp = RX_WRAPLEN(re);
2968 return RX_WRAPPED(re);
2970 const char *const typestr = sv_reftype(referent, 0);
2971 const STRLEN typelen = strlen(typestr);
2972 UV addr = PTR2UV(referent);
2973 const char *stashname = NULL;
2974 STRLEN stashnamelen = 0; /* hush, gcc */
2975 const char *buffer_end;
2977 if (SvOBJECT(referent)) {
2978 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
2981 stashname = HEK_KEY(name);
2982 stashnamelen = HEK_LEN(name);
2984 if (HEK_UTF8(name)) {
2990 stashname = "__ANON__";
2993 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
2994 + 2 * sizeof(UV) + 2 /* )\0 */;
2996 len = typelen + 3 /* (0x */
2997 + 2 * sizeof(UV) + 2 /* )\0 */;
3000 Newx(buffer, len, char);
3001 buffer_end = retval = buffer + len;
3003 /* Working backwards */
3007 *--retval = PL_hexdigit[addr & 15];
3008 } while (addr >>= 4);
3014 memcpy(retval, typestr, typelen);
3018 retval -= stashnamelen;
3019 memcpy(retval, stashname, stashnamelen);
3021 /* retval may not necessarily have reached the start of the
3023 assert (retval >= buffer);
3025 len = buffer_end - retval - 1; /* -1 for that \0 */
3037 if (flags & SV_MUTABLE_RETURN)
3038 return SvPVX_mutable(sv);
3039 if (flags & SV_CONST_RETURN)
3040 return (char *)SvPVX_const(sv);
3045 /* I'm assuming that if both IV and NV are equally valid then
3046 converting the IV is going to be more efficient */
3047 const U32 isUIOK = SvIsUV(sv);
3048 char buf[TYPE_CHARS(UV)];
3052 if (SvTYPE(sv) < SVt_PVIV)
3053 sv_upgrade(sv, SVt_PVIV);
3054 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
3056 /* inlined from sv_setpvn */
3057 s = SvGROW_mutable(sv, len + 1);
3058 Move(ptr, s, len, char);
3063 else if (SvNOK(sv)) {
3064 if (SvTYPE(sv) < SVt_PVNV)
3065 sv_upgrade(sv, SVt_PVNV);
3066 if (SvNVX(sv) == 0.0
3067 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
3068 /* XXX Create SvNVXeq(sv, x)? Or just SvNVXzero(sv)? */
3069 && !Perl_isnan(SvNVX(sv))
3072 s = SvGROW_mutable(sv, 2);
3077 STRLEN size = 5; /* "-Inf\0" */
3079 s = SvGROW_mutable(sv, size);
3080 len = S_infnan_2pv(SvNVX(sv), s, size);
3086 /* some Xenix systems wipe out errno here */
3095 5 + /* exponent digits */
3099 s = SvGROW_mutable(sv, size);
3100 #ifndef USE_LOCALE_NUMERIC
3101 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3107 DECLARE_STORE_LC_NUMERIC_SET_TO_NEEDED();
3111 PL_numeric_radix_sv &&
3112 SvUTF8(PL_numeric_radix_sv);
3113 if (local_radix && SvLEN(PL_numeric_radix_sv) > 1) {
3114 size += SvLEN(PL_numeric_radix_sv) - 1;
3115 s = SvGROW_mutable(sv, size);
3118 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3120 /* If the radix character is UTF-8, and actually is in the
3121 * output, turn on the UTF-8 flag for the scalar */
3123 instr(s, SvPVX_const(PL_numeric_radix_sv))) {
3127 RESTORE_LC_NUMERIC();
3130 /* We don't call SvPOK_on(), because it may come to
3131 * pass that the locale changes so that the
3132 * stringification we just did is no longer correct. We
3133 * will have to re-stringify every time it is needed */
3140 else if (isGV_with_GP(sv)) {
3141 GV *const gv = MUTABLE_GV(sv);
3142 SV *const buffer = sv_newmortal();
3144 gv_efullname3(buffer, gv, "*");
3146 assert(SvPOK(buffer));
3150 *lp = SvCUR(buffer);
3151 return SvPVX(buffer);
3153 else if (isREGEXP(sv)) {
3154 if (lp) *lp = RX_WRAPLEN((REGEXP *)sv);
3155 return RX_WRAPPED((REGEXP *)sv);
3160 if (flags & SV_UNDEF_RETURNS_NULL)
3162 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
3164 /* Typically the caller expects that sv_any is not NULL now. */
3165 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
3166 sv_upgrade(sv, SVt_PV);
3171 const STRLEN len = s - SvPVX_const(sv);
3176 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
3177 PTR2UV(sv),SvPVX_const(sv)));
3178 if (flags & SV_CONST_RETURN)
3179 return (char *)SvPVX_const(sv);
3180 if (flags & SV_MUTABLE_RETURN)
3181 return SvPVX_mutable(sv);
3186 =for apidoc sv_copypv
3188 Copies a stringified representation of the source SV into the
3189 destination SV. Automatically performs any necessary mg_get and
3190 coercion of numeric values into strings. Guaranteed to preserve
3191 UTF8 flag even from overloaded objects. Similar in nature to
3192 sv_2pv[_flags] but operates directly on an SV instead of just the
3193 string. Mostly uses sv_2pv_flags to do its work, except when that
3194 would lose the UTF-8'ness of the PV.
3196 =for apidoc sv_copypv_nomg
3198 Like sv_copypv, but doesn't invoke get magic first.
3200 =for apidoc sv_copypv_flags
3202 Implementation of sv_copypv and sv_copypv_nomg. Calls get magic iff flags
3209 Perl_sv_copypv(pTHX_ SV *const dsv, SV *const ssv)
3211 PERL_ARGS_ASSERT_SV_COPYPV;
3213 sv_copypv_flags(dsv, ssv, 0);
3217 Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
3222 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
3224 s = SvPV_flags_const(ssv,len,(flags & SV_GMAGIC));
3225 sv_setpvn(dsv,s,len);
3233 =for apidoc sv_2pvbyte
3235 Return a pointer to the byte-encoded representation of the SV, and set *lp
3236 to its length. May cause the SV to be downgraded from UTF-8 as a
3239 Usually accessed via the C<SvPVbyte> macro.
3245 Perl_sv_2pvbyte(pTHX_ SV *sv, STRLEN *const lp)
3247 PERL_ARGS_ASSERT_SV_2PVBYTE;
3250 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3251 || isGV_with_GP(sv) || SvROK(sv)) {
3252 SV *sv2 = sv_newmortal();
3253 sv_copypv_nomg(sv2,sv);
3256 sv_utf8_downgrade(sv,0);
3257 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3261 =for apidoc sv_2pvutf8
3263 Return a pointer to the UTF-8-encoded representation of the SV, and set *lp
3264 to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3266 Usually accessed via the C<SvPVutf8> macro.
3272 Perl_sv_2pvutf8(pTHX_ SV *sv, STRLEN *const lp)
3274 PERL_ARGS_ASSERT_SV_2PVUTF8;
3276 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3277 || isGV_with_GP(sv) || SvROK(sv))
3278 sv = sv_mortalcopy(sv);
3281 sv_utf8_upgrade_nomg(sv);
3282 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3287 =for apidoc sv_2bool
3289 This macro is only used by sv_true() or its macro equivalent, and only if
3290 the latter's argument is neither SvPOK, SvIOK nor SvNOK.
3291 It calls sv_2bool_flags with the SV_GMAGIC flag.
3293 =for apidoc sv_2bool_flags
3295 This function is only used by sv_true() and friends, and only if
3296 the latter's argument is neither SvPOK, SvIOK nor SvNOK. If the flags
3297 contain SV_GMAGIC, then it does an mg_get() first.
3304 Perl_sv_2bool_flags(pTHX_ SV *sv, I32 flags)
3306 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
3309 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
3315 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
3316 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv)))) {
3319 if(SvGMAGICAL(sv)) {
3321 goto restart; /* call sv_2bool */
3323 /* expanded SvTRUE_common(sv, (flags = 0, goto restart)) */
3324 else if(!SvOK(sv)) {
3327 else if(SvPOK(sv)) {
3328 svb = SvPVXtrue(sv);
3330 else if((SvFLAGS(sv) & (SVf_IOK|SVf_NOK))) {
3331 svb = (SvIOK(sv) && SvIVX(sv) != 0)
3332 || (SvNOK(sv) && SvNVX(sv) != 0.0);
3336 goto restart; /* call sv_2bool_nomg */
3341 return SvRV(sv) != 0;
3345 RX_WRAPLEN(sv) > 1 || (RX_WRAPLEN(sv) && *RX_WRAPPED(sv) != '0');
3346 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
3350 =for apidoc sv_utf8_upgrade
3352 Converts the PV of an SV to its UTF-8-encoded form.
3353 Forces the SV to string form if it is not already.
3354 Will C<mg_get> on C<sv> if appropriate.
3355 Always sets the SvUTF8 flag to avoid future validity checks even
3356 if the whole string is the same in UTF-8 as not.
3357 Returns the number of bytes in the converted string
3359 This is not a general purpose byte encoding to Unicode interface:
3360 use the Encode extension for that.
3362 =for apidoc sv_utf8_upgrade_nomg
3364 Like sv_utf8_upgrade, but doesn't do magic on C<sv>.
3366 =for apidoc sv_utf8_upgrade_flags
3368 Converts the PV of an SV to its UTF-8-encoded form.
3369 Forces the SV to string form if it is not already.
3370 Always sets the SvUTF8 flag to avoid future validity checks even
3371 if all the bytes are invariant in UTF-8.
3372 If C<flags> has C<SV_GMAGIC> bit set,
3373 will C<mg_get> on C<sv> if appropriate, else not.
3375 If C<flags> has SV_FORCE_UTF8_UPGRADE set, this function assumes that the PV
3376 will expand when converted to UTF-8, and skips the extra work of checking for
3377 that. Typically this flag is used by a routine that has already parsed the
3378 string and found such characters, and passes this information on so that the
3379 work doesn't have to be repeated.
3381 Returns the number of bytes in the converted string.
3383 This is not a general purpose byte encoding to Unicode interface:
3384 use the Encode extension for that.
3386 =for apidoc sv_utf8_upgrade_flags_grow
3388 Like sv_utf8_upgrade_flags, but has an additional parameter C<extra>, which is
3389 the number of unused bytes the string of 'sv' is guaranteed to have free after
3390 it upon return. This allows the caller to reserve extra space that it intends
3391 to fill, to avoid extra grows.
3393 C<sv_utf8_upgrade>, C<sv_utf8_upgrade_nomg>, and C<sv_utf8_upgrade_flags>
3394 are implemented in terms of this function.
3396 Returns the number of bytes in the converted string (not including the spares).
3400 (One might think that the calling routine could pass in the position of the
3401 first variant character when it has set SV_FORCE_UTF8_UPGRADE, so it wouldn't
3402 have to be found again. But that is not the case, because typically when the
3403 caller is likely to use this flag, it won't be calling this routine unless it
3404 finds something that won't fit into a byte. Otherwise it tries to not upgrade
3405 and just use bytes. But some things that do fit into a byte are variants in
3406 utf8, and the caller may not have been keeping track of these.)
3408 If the routine itself changes the string, it adds a trailing C<NUL>. Such a
3409 C<NUL> isn't guaranteed due to having other routines do the work in some input
3410 cases, or if the input is already flagged as being in utf8.
3412 The speed of this could perhaps be improved for many cases if someone wanted to
3413 write a fast function that counts the number of variant characters in a string,
3414 especially if it could return the position of the first one.
3419 Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV *const sv, const I32 flags, STRLEN extra)
3421 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
3423 if (sv == &PL_sv_undef)
3425 if (!SvPOK_nog(sv)) {
3427 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3428 (void) sv_2pv_flags(sv,&len, flags);
3430 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3434 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
3439 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3444 S_sv_uncow(aTHX_ sv, 0);
3447 if (PL_encoding && !(flags & SV_UTF8_NO_ENCODING)) {
3448 sv_recode_to_utf8(sv, PL_encoding);
3449 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3453 if (SvCUR(sv) == 0) {
3454 if (extra) SvGROW(sv, extra);
3455 } else { /* Assume Latin-1/EBCDIC */
3456 /* This function could be much more efficient if we
3457 * had a FLAG in SVs to signal if there are any variant
3458 * chars in the PV. Given that there isn't such a flag
3459 * make the loop as fast as possible (although there are certainly ways
3460 * to speed this up, eg. through vectorization) */
3461 U8 * s = (U8 *) SvPVX_const(sv);
3462 U8 * e = (U8 *) SvEND(sv);
3464 STRLEN two_byte_count = 0;
3466 if (flags & SV_FORCE_UTF8_UPGRADE) goto must_be_utf8;
3468 /* See if really will need to convert to utf8. We mustn't rely on our
3469 * incoming SV being well formed and having a trailing '\0', as certain
3470 * code in pp_formline can send us partially built SVs. */
3474 if (NATIVE_BYTE_IS_INVARIANT(ch)) continue;
3476 t--; /* t already incremented; re-point to first variant */
3481 /* utf8 conversion not needed because all are invariants. Mark as
3482 * UTF-8 even if no variant - saves scanning loop */
3484 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3489 /* Here, the string should be converted to utf8, either because of an
3490 * input flag (two_byte_count = 0), or because a character that
3491 * requires 2 bytes was found (two_byte_count = 1). t points either to
3492 * the beginning of the string (if we didn't examine anything), or to
3493 * the first variant. In either case, everything from s to t - 1 will
3494 * occupy only 1 byte each on output.
3496 * There are two main ways to convert. One is to create a new string
3497 * and go through the input starting from the beginning, appending each
3498 * converted value onto the new string as we go along. It's probably
3499 * best to allocate enough space in the string for the worst possible
3500 * case rather than possibly running out of space and having to
3501 * reallocate and then copy what we've done so far. Since everything
3502 * from s to t - 1 is invariant, the destination can be initialized
3503 * with these using a fast memory copy
3505 * The other way is to figure out exactly how big the string should be
3506 * by parsing the entire input. Then you don't have to make it big
3507 * enough to handle the worst possible case, and more importantly, if
3508 * the string you already have is large enough, you don't have to
3509 * allocate a new string, you can copy the last character in the input
3510 * string to the final position(s) that will be occupied by the
3511 * converted string and go backwards, stopping at t, since everything
3512 * before that is invariant.
3514 * There are advantages and disadvantages to each method.
3516 * In the first method, we can allocate a new string, do the memory
3517 * copy from the s to t - 1, and then proceed through the rest of the
3518 * string byte-by-byte.
3520 * In the second method, we proceed through the rest of the input
3521 * string just calculating how big the converted string will be. Then
3522 * there are two cases:
3523 * 1) if the string has enough extra space to handle the converted
3524 * value. We go backwards through the string, converting until we
3525 * get to the position we are at now, and then stop. If this
3526 * position is far enough along in the string, this method is
3527 * faster than the other method. If the memory copy were the same
3528 * speed as the byte-by-byte loop, that position would be about
3529 * half-way, as at the half-way mark, parsing to the end and back
3530 * is one complete string's parse, the same amount as starting
3531 * over and going all the way through. Actually, it would be
3532 * somewhat less than half-way, as it's faster to just count bytes
3533 * than to also copy, and we don't have the overhead of allocating
3534 * a new string, changing the scalar to use it, and freeing the
3535 * existing one. But if the memory copy is fast, the break-even
3536 * point is somewhere after half way. The counting loop could be
3537 * sped up by vectorization, etc, to move the break-even point
3538 * further towards the beginning.
3539 * 2) if the string doesn't have enough space to handle the converted
3540 * value. A new string will have to be allocated, and one might
3541 * as well, given that, start from the beginning doing the first
3542 * method. We've spent extra time parsing the string and in
3543 * exchange all we've gotten is that we know precisely how big to
3544 * make the new one. Perl is more optimized for time than space,
3545 * so this case is a loser.
3546 * So what I've decided to do is not use the 2nd method unless it is
3547 * guaranteed that a new string won't have to be allocated, assuming
3548 * the worst case. I also decided not to put any more conditions on it
3549 * than this, for now. It seems likely that, since the worst case is
3550 * twice as big as the unknown portion of the string (plus 1), we won't
3551 * be guaranteed enough space, causing us to go to the first method,
3552 * unless the string is short, or the first variant character is near
3553 * the end of it. In either of these cases, it seems best to use the
3554 * 2nd method. The only circumstance I can think of where this would
3555 * be really slower is if the string had once had much more data in it
3556 * than it does now, but there is still a substantial amount in it */
3559 STRLEN invariant_head = t - s;
3560 STRLEN size = invariant_head + (e - t) * 2 + 1 + extra;
3561 if (SvLEN(sv) < size) {
3563 /* Here, have decided to allocate a new string */
3568 Newx(dst, size, U8);
3570 /* If no known invariants at the beginning of the input string,
3571 * set so starts from there. Otherwise, can use memory copy to
3572 * get up to where we are now, and then start from here */
3574 if (invariant_head == 0) {
3577 Copy(s, dst, invariant_head, char);
3578 d = dst + invariant_head;
3582 append_utf8_from_native_byte(*t, &d);
3586 SvPV_free(sv); /* No longer using pre-existing string */
3587 SvPV_set(sv, (char*)dst);
3588 SvCUR_set(sv, d - dst);
3589 SvLEN_set(sv, size);
3592 /* Here, have decided to get the exact size of the string.
3593 * Currently this happens only when we know that there is
3594 * guaranteed enough space to fit the converted string, so
3595 * don't have to worry about growing. If two_byte_count is 0,
3596 * then t points to the first byte of the string which hasn't
3597 * been examined yet. Otherwise two_byte_count is 1, and t
3598 * points to the first byte in the string that will expand to
3599 * two. Depending on this, start examining at t or 1 after t.
3602 U8 *d = t + two_byte_count;
3605 /* Count up the remaining bytes that expand to two */
3608 const U8 chr = *d++;
3609 if (! NATIVE_BYTE_IS_INVARIANT(chr)) two_byte_count++;
3612 /* The string will expand by just the number of bytes that
3613 * occupy two positions. But we are one afterwards because of
3614 * the increment just above. This is the place to put the
3615 * trailing NUL, and to set the length before we decrement */
3617 d += two_byte_count;
3618 SvCUR_set(sv, d - s);
3622 /* Having decremented d, it points to the position to put the
3623 * very last byte of the expanded string. Go backwards through
3624 * the string, copying and expanding as we go, stopping when we
3625 * get to the part that is invariant the rest of the way down */
3629 if (NATIVE_BYTE_IS_INVARIANT(*e)) {
3632 *d-- = UTF8_EIGHT_BIT_LO(*e);
3633 *d-- = UTF8_EIGHT_BIT_HI(*e);
3639 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3640 /* Update pos. We do it at the end rather than during
3641 * the upgrade, to avoid slowing down the common case
3642 * (upgrade without pos).
3643 * pos can be stored as either bytes or characters. Since
3644 * this was previously a byte string we can just turn off
3645 * the bytes flag. */
3646 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3648 mg->mg_flags &= ~MGf_BYTES;
3650 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3651 magic_setutf8(sv,mg); /* clear UTF8 cache */
3656 /* Mark as UTF-8 even if no variant - saves scanning loop */
3662 =for apidoc sv_utf8_downgrade
3664 Attempts to convert the PV of an SV from characters to bytes.
3665 If the PV contains a character that cannot fit
3666 in a byte, this conversion will fail;
3667 in this case, either returns false or, if C<fail_ok> is not
3670 This is not a general purpose Unicode to byte encoding interface:
3671 use the Encode extension for that.
3677 Perl_sv_utf8_downgrade(pTHX_ SV *const sv, const bool fail_ok)
3679 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3681 if (SvPOKp(sv) && SvUTF8(sv)) {
3685 int mg_flags = SV_GMAGIC;
3688 S_sv_uncow(aTHX_ sv, 0);
3690 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3692 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3693 if (mg && mg->mg_len > 0 && mg->mg_flags & MGf_BYTES) {
3694 mg->mg_len = sv_pos_b2u_flags(sv, mg->mg_len,
3695 SV_GMAGIC|SV_CONST_RETURN);
3696 mg_flags = 0; /* sv_pos_b2u does get magic */
3698 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3699 magic_setutf8(sv,mg); /* clear UTF8 cache */
3702 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3704 if (!utf8_to_bytes(s, &len)) {
3709 Perl_croak(aTHX_ "Wide character in %s",
3712 Perl_croak(aTHX_ "Wide character");
3723 =for apidoc sv_utf8_encode
3725 Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3726 flag off so that it looks like octets again.
3732 Perl_sv_utf8_encode(pTHX_ SV *const sv)
3734 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3736 if (SvREADONLY(sv)) {
3737 sv_force_normal_flags(sv, 0);
3739 (void) sv_utf8_upgrade(sv);
3744 =for apidoc sv_utf8_decode
3746 If the PV of the SV is an octet sequence in UTF-8
3747 and contains a multiple-byte character, the C<SvUTF8> flag is turned on
3748 so that it looks like a character. If the PV contains only single-byte
3749 characters, the C<SvUTF8> flag stays off.
3750 Scans PV for validity and returns false if the PV is invalid UTF-8.
3756 Perl_sv_utf8_decode(pTHX_ SV *const sv)
3758 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3761 const U8 *start, *c;
3764 /* The octets may have got themselves encoded - get them back as
3767 if (!sv_utf8_downgrade(sv, TRUE))
3770 /* it is actually just a matter of turning the utf8 flag on, but
3771 * we want to make sure everything inside is valid utf8 first.
3773 c = start = (const U8 *) SvPVX_const(sv);
3774 if (!is_utf8_string(c, SvCUR(sv)))
3776 e = (const U8 *) SvEND(sv);
3779 if (!UTF8_IS_INVARIANT(ch)) {
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. <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 gp_free(MUTABLE_GV(dstr));
3929 GvINTRO_off(dstr); /* one-shot flag */
3930 GvGP_set(dstr, gp_ref(GvGP(sstr)));
3931 if (SvTAINTED(sstr))
3933 if (GvIMPORTED(dstr) != GVf_IMPORTED
3934 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
3936 GvIMPORTED_on(dstr);
3939 if(mro_changes == 2) {
3940 if (GvAV((const GV *)sstr)) {
3942 SV * const sref = (SV *)GvAV((const GV *)dstr);
3943 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
3944 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
3945 AV * const ary = newAV();
3946 av_push(ary, mg->mg_obj); /* takes the refcount */
3947 mg->mg_obj = (SV *)ary;
3949 av_push((AV *)mg->mg_obj, SvREFCNT_inc_simple_NN(dstr));
3951 else sv_magic(sref, dstr, PERL_MAGIC_isa, NULL, 0);
3953 mro_isa_changed_in(GvSTASH(dstr));
3955 else if(mro_changes == 3) {
3956 HV * const stash = GvHV(dstr);
3957 if(old_stash ? (HV *)HvENAME_get(old_stash) : stash)
3963 else if(mro_changes) mro_method_changed_in(GvSTASH(dstr));
3964 if (GvIO(dstr) && dtype == SVt_PVGV) {
3965 DEBUG_o(Perl_deb(aTHX_
3966 "glob_assign_glob clearing PL_stashcache\n"));
3967 /* It's a cache. It will rebuild itself quite happily.
3968 It's a lot of effort to work out exactly which key (or keys)
3969 might be invalidated by the creation of the this file handle.
3971 hv_clear(PL_stashcache);
3977 S_glob_assign_ref(pTHX_ SV *const dstr, SV *const sstr)
3979 SV * const sref = SvRV(sstr);
3981 const int intro = GvINTRO(dstr);
3984 const U32 stype = SvTYPE(sref);
3986 PERL_ARGS_ASSERT_GLOB_ASSIGN_REF;
3989 GvINTRO_off(dstr); /* one-shot flag */
3990 GvLINE(dstr) = CopLINE(PL_curcop);
3991 GvEGV(dstr) = MUTABLE_GV(dstr);
3996 location = (SV **) &(GvGP(dstr)->gp_cv); /* XXX bypassing GvCV_set */
3997 import_flag = GVf_IMPORTED_CV;
4000 location = (SV **) &GvHV(dstr);
4001 import_flag = GVf_IMPORTED_HV;
4004 location = (SV **) &GvAV(dstr);
4005 import_flag = GVf_IMPORTED_AV;
4008 location = (SV **) &GvIOp(dstr);
4011 location = (SV **) &GvFORM(dstr);
4014 location = &GvSV(dstr);
4015 import_flag = GVf_IMPORTED_SV;
4018 if (stype == SVt_PVCV) {
4019 /*if (GvCVGEN(dstr) && (GvCV(dstr) != (const CV *)sref || GvCVGEN(dstr))) {*/
4020 if (GvCVGEN(dstr)) {
4021 SvREFCNT_dec(GvCV(dstr));
4022 GvCV_set(dstr, NULL);
4023 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4026 /* SAVEt_GVSLOT takes more room on the savestack and has more
4027 overhead in leave_scope than SAVEt_GENERIC_SV. But for CVs
4028 leave_scope needs access to the GV so it can reset method
4029 caches. We must use SAVEt_GVSLOT whenever the type is
4030 SVt_PVCV, even if the stash is anonymous, as the stash may
4031 gain a name somehow before leave_scope. */
4032 if (stype == SVt_PVCV) {
4033 /* There is no save_pushptrptrptr. Creating it for this
4034 one call site would be overkill. So inline the ss add
4038 SS_ADD_PTR(location);
4039 SS_ADD_PTR(SvREFCNT_inc(*location));
4040 SS_ADD_UV(SAVEt_GVSLOT);
4043 else SAVEGENERICSV(*location);
4046 if (stype == SVt_PVCV && (*location != sref || GvCVGEN(dstr))) {
4047 CV* const cv = MUTABLE_CV(*location);
4049 if (!GvCVGEN((const GV *)dstr) &&
4050 (CvROOT(cv) || CvXSUB(cv)) &&
4051 /* redundant check that avoids creating the extra SV
4052 most of the time: */
4053 (CvCONST(cv) || ckWARN(WARN_REDEFINE)))
4055 SV * const new_const_sv =
4056 CvCONST((const CV *)sref)
4057 ? cv_const_sv((const CV *)sref)
4059 report_redefined_cv(
4060 sv_2mortal(Perl_newSVpvf(aTHX_
4063 HvNAME_HEK(GvSTASH((const GV *)dstr))
4065 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr)))
4068 CvCONST((const CV *)sref) ? &new_const_sv : NULL
4072 cv_ckproto_len_flags(cv, (const GV *)dstr,
4073 SvPOK(sref) ? CvPROTO(sref) : NULL,
4074 SvPOK(sref) ? CvPROTOLEN(sref) : 0,
4075 SvPOK(sref) ? SvUTF8(sref) : 0);
4077 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4078 GvASSUMECV_on(dstr);
4079 if(GvSTASH(dstr)) { /* sub foo { 1 } sub bar { 2 } *bar = \&foo */
4080 if (intro && GvREFCNT(dstr) > 1) {
4081 /* temporary remove extra savestack's ref */
4083 gv_method_changed(dstr);
4086 else gv_method_changed(dstr);
4089 *location = SvREFCNT_inc_simple_NN(sref);
4090 if (import_flag && !(GvFLAGS(dstr) & import_flag)
4091 && CopSTASH_ne(PL_curcop, GvSTASH(dstr))) {
4092 GvFLAGS(dstr) |= import_flag;
4094 if (import_flag == GVf_IMPORTED_SV) {
4097 SS_ADD_PTR(gp_ref(GvGP(dstr)));
4098 SS_ADD_UV(SAVEt_GP_ALIASED_SV
4099 | cBOOL(GvALIASED_SV(dstr)) << 8);
4102 /* Turn off the flag if sref is not referenced elsewhere,
4103 even by weak refs. (SvRMAGICAL is a pessimistic check for
4105 if (SvREFCNT(sref) <= 2 && !SvRMAGICAL(sref))
4106 GvALIASED_SV_off(dstr);
4108 GvALIASED_SV_on(dstr);
4110 if (stype == SVt_PVHV) {
4111 const char * const name = GvNAME((GV*)dstr);
4112 const STRLEN len = GvNAMELEN(dstr);
4115 (len > 1 && name[len-2] == ':' && name[len-1] == ':')
4116 || (len == 1 && name[0] == ':')
4118 && (!dref || HvENAME_get(dref))
4121 (HV *)sref, (HV *)dref,
4127 stype == SVt_PVAV && sref != dref
4128 && strEQ(GvNAME((GV*)dstr), "ISA")
4129 /* The stash may have been detached from the symbol table, so
4130 check its name before doing anything. */
4131 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
4134 MAGIC * const omg = dref && SvSMAGICAL(dref)
4135 ? mg_find(dref, PERL_MAGIC_isa)
4137 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
4138 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
4139 AV * const ary = newAV();
4140 av_push(ary, mg->mg_obj); /* takes the refcount */
4141 mg->mg_obj = (SV *)ary;
4144 if (SvTYPE(omg->mg_obj) == SVt_PVAV) {
4145 SV **svp = AvARRAY((AV *)omg->mg_obj);
4146 I32 items = AvFILLp((AV *)omg->mg_obj) + 1;
4150 SvREFCNT_inc_simple_NN(*svp++)
4156 SvREFCNT_inc_simple_NN(omg->mg_obj)
4160 av_push((AV *)mg->mg_obj,SvREFCNT_inc_simple_NN(dstr));
4165 sref, omg ? omg->mg_obj : dstr, PERL_MAGIC_isa, NULL, 0
4167 mg = mg_find(sref, PERL_MAGIC_isa);
4169 /* Since the *ISA assignment could have affected more than
4170 one stash, don't call mro_isa_changed_in directly, but let
4171 magic_clearisa do it for us, as it already has the logic for
4172 dealing with globs vs arrays of globs. */
4174 Perl_magic_clearisa(aTHX_ NULL, mg);
4176 else if (stype == SVt_PVIO) {
4177 DEBUG_o(Perl_deb(aTHX_ "glob_assign_ref clearing PL_stashcache\n"));
4178 /* It's a cache. It will rebuild itself quite happily.
4179 It's a lot of effort to work out exactly which key (or keys)
4180 might be invalidated by the creation of the this file handle.
4182 hv_clear(PL_stashcache);
4186 if (!intro) SvREFCNT_dec(dref);
4187 if (SvTAINTED(sstr))
4195 #ifdef PERL_DEBUG_READONLY_COW
4196 # include <sys/mman.h>
4198 # ifndef PERL_MEMORY_DEBUG_HEADER_SIZE
4199 # define PERL_MEMORY_DEBUG_HEADER_SIZE 0
4203 Perl_sv_buf_to_ro(pTHX_ SV *sv)
4205 struct perl_memory_debug_header * const header =
4206 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4207 const MEM_SIZE len = header->size;
4208 PERL_ARGS_ASSERT_SV_BUF_TO_RO;
4209 # ifdef PERL_TRACK_MEMPOOL
4210 if (!header->readonly) header->readonly = 1;
4212 if (mprotect(header, len, PROT_READ))
4213 Perl_warn(aTHX_ "mprotect RW for COW string %p %lu failed with %d",
4214 header, len, errno);
4218 S_sv_buf_to_rw(pTHX_ SV *sv)
4220 struct perl_memory_debug_header * const header =
4221 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4222 const MEM_SIZE len = header->size;
4223 PERL_ARGS_ASSERT_SV_BUF_TO_RW;
4224 if (mprotect(header, len, PROT_READ|PROT_WRITE))
4225 Perl_warn(aTHX_ "mprotect for COW string %p %lu failed with %d",
4226 header, len, errno);
4227 # ifdef PERL_TRACK_MEMPOOL
4228 header->readonly = 0;
4233 # define sv_buf_to_ro(sv) NOOP
4234 # define sv_buf_to_rw(sv) NOOP
4238 Perl_sv_setsv_flags(pTHX_ SV *dstr, SV* sstr, const I32 flags)
4244 PERL_ARGS_ASSERT_SV_SETSV_FLAGS;
4249 if (SvIS_FREED(dstr)) {
4250 Perl_croak(aTHX_ "panic: attempt to copy value %" SVf
4251 " to a freed scalar %p", SVfARG(sstr), (void *)dstr);
4253 SV_CHECK_THINKFIRST_COW_DROP(dstr);
4255 sstr = &PL_sv_undef;
4256 if (SvIS_FREED(sstr)) {
4257 Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p to %p",
4258 (void*)sstr, (void*)dstr);
4260 stype = SvTYPE(sstr);
4261 dtype = SvTYPE(dstr);
4263 /* There's a lot of redundancy below but we're going for speed here */
4268 if (dtype != SVt_PVGV && dtype != SVt_PVLV) {
4269 (void)SvOK_off(dstr);
4277 sv_upgrade(dstr, SVt_IV);
4281 sv_upgrade(dstr, SVt_PVIV);
4285 goto end_of_first_switch;
4287 (void)SvIOK_only(dstr);
4288 SvIV_set(dstr, SvIVX(sstr));
4291 /* SvTAINTED can only be true if the SV has taint magic, which in
4292 turn means that the SV type is PVMG (or greater). This is the
4293 case statement for SVt_IV, so this cannot be true (whatever gcov
4295 assert(!SvTAINTED(sstr));
4300 if (dtype < SVt_PV && dtype != SVt_IV)
4301 sv_upgrade(dstr, SVt_IV);
4309 sv_upgrade(dstr, SVt_NV);
4313 sv_upgrade(dstr, SVt_PVNV);
4317 goto end_of_first_switch;
4319 SvNV_set(dstr, SvNVX(sstr));
4320 (void)SvNOK_only(dstr);
4321 /* SvTAINTED can only be true if the SV has taint magic, which in
4322 turn means that the SV type is PVMG (or greater). This is the
4323 case statement for SVt_NV, so this cannot be true (whatever gcov
4325 assert(!SvTAINTED(sstr));
4332 sv_upgrade(dstr, SVt_PV);
4335 if (dtype < SVt_PVIV)
4336 sv_upgrade(dstr, SVt_PVIV);
4339 if (dtype < SVt_PVNV)
4340 sv_upgrade(dstr, SVt_PVNV);
4344 const char * const type = sv_reftype(sstr,0);
4346 /* diag_listed_as: Bizarre copy of %s */
4347 Perl_croak(aTHX_ "Bizarre copy of %s in %s", type, OP_DESC(PL_op));
4349 Perl_croak(aTHX_ "Bizarre copy of %s", type);
4351 NOT_REACHED; /* NOTREACHED */
4355 if (dtype < SVt_REGEXP)
4357 if (dtype >= SVt_PV) {
4363 sv_upgrade(dstr, SVt_REGEXP);
4371 if (SvGMAGICAL(sstr) && (flags & SV_GMAGIC)) {
4373 if (SvTYPE(sstr) != stype)
4374 stype = SvTYPE(sstr);
4376 if (isGV_with_GP(sstr) && dtype <= SVt_PVLV) {
4377 glob_assign_glob(dstr, sstr, dtype);
4380 if (stype == SVt_PVLV)
4382 if (isREGEXP(sstr)) goto upgregexp;
4383 SvUPGRADE(dstr, SVt_PVNV);
4386 SvUPGRADE(dstr, (svtype)stype);
4388 end_of_first_switch:
4390 /* dstr may have been upgraded. */
4391 dtype = SvTYPE(dstr);
4392 sflags = SvFLAGS(sstr);
4394 if (dtype == SVt_PVCV) {
4395 /* Assigning to a subroutine sets the prototype. */
4398 const char *const ptr = SvPV_const(sstr, len);
4400 SvGROW(dstr, len + 1);
4401 Copy(ptr, SvPVX(dstr), len + 1, char);
4402 SvCUR_set(dstr, len);
4404 SvFLAGS(dstr) |= sflags & SVf_UTF8;
4405 CvAUTOLOAD_off(dstr);
4410 else if (dtype == SVt_PVAV || dtype == SVt_PVHV || dtype == SVt_PVFM) {
4411 const char * const type = sv_reftype(dstr,0);
4413 /* diag_listed_as: Cannot copy to %s */
4414 Perl_croak(aTHX_ "Cannot copy to %s in %s", type, OP_DESC(PL_op));
4416 Perl_croak(aTHX_ "Cannot copy to %s", type);
4417 } else if (sflags & SVf_ROK) {
4418 if (isGV_with_GP(dstr)
4419 && SvTYPE(SvRV(sstr)) == SVt_PVGV && isGV_with_GP(SvRV(sstr))) {
4422 if (GvIMPORTED(dstr) != GVf_IMPORTED
4423 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
4425 GvIMPORTED_on(dstr);
4430 glob_assign_glob(dstr, sstr, dtype);
4434 if (dtype >= SVt_PV) {
4435 if (isGV_with_GP(dstr)) {
4436 glob_assign_ref(dstr, sstr);
4439 if (SvPVX_const(dstr)) {
4445 (void)SvOK_off(dstr);
4446 SvRV_set(dstr, SvREFCNT_inc(SvRV(sstr)));
4447 SvFLAGS(dstr) |= sflags & SVf_ROK;
4448 assert(!(sflags & SVp_NOK));
4449 assert(!(sflags & SVp_IOK));
4450 assert(!(sflags & SVf_NOK));
4451 assert(!(sflags & SVf_IOK));
4453 else if (isGV_with_GP(dstr)) {
4454 if (!(sflags & SVf_OK)) {
4455 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4456 "Undefined value assigned to typeglob");
4459 GV *gv = gv_fetchsv_nomg(sstr, GV_ADD, SVt_PVGV);
4460 if (dstr != (const SV *)gv) {
4461 const char * const name = GvNAME((const GV *)dstr);
4462 const STRLEN len = GvNAMELEN(dstr);
4463 HV *old_stash = NULL;
4464 bool reset_isa = FALSE;
4465 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
4466 || (len == 1 && name[0] == ':')) {
4467 /* Set aside the old stash, so we can reset isa caches
4468 on its subclasses. */
4469 if((old_stash = GvHV(dstr))) {
4470 /* Make sure we do not lose it early. */
4471 SvREFCNT_inc_simple_void_NN(
4472 sv_2mortal((SV *)old_stash)
4479 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
4480 gp_free(MUTABLE_GV(dstr));
4482 GvGP_set(dstr, gp_ref(GvGP(gv)));
4485 HV * const stash = GvHV(dstr);
4487 old_stash ? (HV *)HvENAME_get(old_stash) : stash
4497 else if ((dtype == SVt_REGEXP || dtype == SVt_PVLV)
4498 && (stype == SVt_REGEXP || isREGEXP(sstr))) {
4499 reg_temp_copy((REGEXP*)dstr, (REGEXP*)sstr);
4501 else if (sflags & SVp_POK) {
4502 const STRLEN cur = SvCUR(sstr);
4503 const STRLEN len = SvLEN(sstr);
4506 * We have three basic ways to copy the string:
4512 * Which we choose is based on various factors. The following
4513 * things are listed in order of speed, fastest to slowest:
4515 * - Copying a short string
4516 * - Copy-on-write bookkeeping
4518 * - Copying a long string
4520 * We swipe the string (steal the string buffer) if the SV on the
4521 * rhs is about to be freed anyway (TEMP and refcnt==1). This is a
4522 * big win on long strings. It should be a win on short strings if
4523 * SvPVX_const(dstr) has to be allocated. If not, it should not
4524 * slow things down, as SvPVX_const(sstr) would have been freed
4527 * We also steal the buffer from a PADTMP (operator target) if it
4528 * is ‘long enough’. For short strings, a swipe does not help
4529 * here, as it causes more malloc calls the next time the target
4530 * is used. Benchmarks show that even if SvPVX_const(dstr) has to
4531 * be allocated it is still not worth swiping PADTMPs for short
4532 * strings, as the savings here are small.
4534 * If swiping is not an option, then we see whether it is
4535 * worth using copy-on-write. If the lhs already has a buf-
4536 * fer big enough and the string is short, we skip it and fall back
4537 * to method 3, since memcpy is faster for short strings than the
4538 * later bookkeeping overhead that copy-on-write entails.
4540 * If the rhs is not a copy-on-write string yet, then we also
4541 * consider whether the buffer is too large relative to the string
4542 * it holds. Some operations such as readline allocate a large
4543 * buffer in the expectation of reusing it. But turning such into
4544 * a COW buffer is counter-productive because it increases memory
4545 * usage by making readline allocate a new large buffer the sec-
4546 * ond time round. So, if the buffer is too large, again, we use
4549 * Finally, if there is no buffer on the left, or the buffer is too
4550 * small, then we use copy-on-write and make both SVs share the
4555 /* Whichever path we take through the next code, we want this true,
4556 and doing it now facilitates the COW check. */
4557 (void)SvPOK_only(dstr);
4561 /* slated for free anyway (and not COW)? */
4562 (sflags & (SVs_TEMP|SVf_IsCOW)) == SVs_TEMP
4563 /* or a swipable TARG */
4565 (SVs_PADTMP|SVf_READONLY|SVf_PROTECT|SVf_IsCOW))
4567 /* whose buffer is worth stealing */
4568 && CHECK_COWBUF_THRESHOLD(cur,len)
4571 !(sflags & SVf_OOK) && /* and not involved in OOK hack? */
4572 (!(flags & SV_NOSTEAL)) &&
4573 /* and we're allowed to steal temps */
4574 SvREFCNT(sstr) == 1 && /* and no other references to it? */
4575 len) /* and really is a string */
4576 { /* Passes the swipe test. */
4577 if (SvPVX_const(dstr)) /* we know that dtype >= SVt_PV */
4579 SvPV_set(dstr, SvPVX_mutable(sstr));
4580 SvLEN_set(dstr, SvLEN(sstr));
4581 SvCUR_set(dstr, SvCUR(sstr));
4584 (void)SvOK_off(sstr); /* NOTE: nukes most SvFLAGS on sstr */
4585 SvPV_set(sstr, NULL);
4590 else if (flags & SV_COW_SHARED_HASH_KEYS
4592 #ifdef PERL_OLD_COPY_ON_WRITE
4593 ( sflags & SVf_IsCOW
4594 || ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4595 && (SvFLAGS(dstr) & CAN_COW_MASK) == CAN_COW_FLAGS
4596 && SvTYPE(sstr) >= SVt_PVIV && len
4599 #elif defined(PERL_NEW_COPY_ON_WRITE)
4602 ( (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4603 /* If this is a regular (non-hek) COW, only so
4604 many COW "copies" are possible. */
4605 && CowREFCNT(sstr) != SV_COW_REFCNT_MAX ))
4606 : ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4607 && !(SvFLAGS(dstr) & SVf_BREAK)
4608 && CHECK_COW_THRESHOLD(cur,len) && cur+1 < len
4609 && (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4613 && !(SvFLAGS(dstr) & SVf_BREAK)
4616 /* Either it's a shared hash key, or it's suitable for
4619 PerlIO_printf(Perl_debug_log, "Copy on write: sstr --> dstr\n");
4624 if (!(sflags & SVf_IsCOW)) {
4626 # ifdef PERL_OLD_COPY_ON_WRITE
4627 /* Make the source SV into a loop of 1.
4628 (about to become 2) */
4629 SV_COW_NEXT_SV_SET(sstr, sstr);
4631 CowREFCNT(sstr) = 0;
4635 if (SvPVX_const(dstr)) { /* we know that dtype >= SVt_PV */
4641 # ifdef PERL_OLD_COPY_ON_WRITE
4642 assert (SvTYPE(dstr) >= SVt_PVIV);
4643 /* SvIsCOW_normal */
4644 /* splice us in between source and next-after-source. */
4645 SV_COW_NEXT_SV_SET(dstr, SV_COW_NEXT_SV(sstr));
4646 SV_COW_NEXT_SV_SET(sstr, dstr);
4648 if (sflags & SVf_IsCOW) {
4653 SvPV_set(dstr, SvPVX_mutable(sstr));
4658 /* SvIsCOW_shared_hash */
4659 DEBUG_C(PerlIO_printf(Perl_debug_log,
4660 "Copy on write: Sharing hash\n"));
4662 assert (SvTYPE(dstr) >= SVt_PV);
4664 HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)))));
4666 SvLEN_set(dstr, len);
4667 SvCUR_set(dstr, cur);
4670 /* Failed the swipe test, and we cannot do copy-on-write either.
4671 Have to copy the string. */
4672 SvGROW(dstr, cur + 1); /* inlined from sv_setpvn */
4673 Move(SvPVX_const(sstr),SvPVX(dstr),cur,char);
4674 SvCUR_set(dstr, cur);
4675 *SvEND(dstr) = '\0';
4677 if (sflags & SVp_NOK) {
4678 SvNV_set(dstr, SvNVX(sstr));
4680 if (sflags & SVp_IOK) {
4681 SvIV_set(dstr, SvIVX(sstr));
4682 /* Must do this otherwise some other overloaded use of 0x80000000
4683 gets confused. I guess SVpbm_VALID */
4684 if (sflags & SVf_IVisUV)
4687 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_NOK|SVp_NOK|SVf_UTF8);
4689 const MAGIC * const smg = SvVSTRING_mg(sstr);
4691 sv_magic(dstr, NULL, PERL_MAGIC_vstring,
4692 smg->mg_ptr, smg->mg_len);
4693 SvRMAGICAL_on(dstr);
4697 else if (sflags & (SVp_IOK|SVp_NOK)) {
4698 (void)SvOK_off(dstr);
4699 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_IVisUV|SVf_NOK|SVp_NOK);
4700 if (sflags & SVp_IOK) {
4701 /* XXXX Do we want to set IsUV for IV(ROK)? Be extra safe... */
4702 SvIV_set(dstr, SvIVX(sstr));
4704 if (sflags & SVp_NOK) {
4705 SvNV_set(dstr, SvNVX(sstr));
4709 if (isGV_with_GP(sstr)) {
4710 gv_efullname3(dstr, MUTABLE_GV(sstr), "*");
4713 (void)SvOK_off(dstr);
4715 if (SvTAINTED(sstr))
4720 =for apidoc sv_setsv_mg
4722 Like C<sv_setsv>, but also handles 'set' magic.
4728 Perl_sv_setsv_mg(pTHX_ SV *const dstr, SV *const sstr)
4730 PERL_ARGS_ASSERT_SV_SETSV_MG;
4732 sv_setsv(dstr,sstr);
4737 # ifdef PERL_OLD_COPY_ON_WRITE
4738 # define SVt_COW SVt_PVIV
4740 # define SVt_COW SVt_PV
4743 Perl_sv_setsv_cow(pTHX_ SV *dstr, SV *sstr)
4745 STRLEN cur = SvCUR(sstr);
4746 STRLEN len = SvLEN(sstr);
4748 #if defined(PERL_DEBUG_READONLY_COW) && defined(PERL_NEW_COPY_ON_WRITE)
4749 const bool already = cBOOL(SvIsCOW(sstr));
4752 PERL_ARGS_ASSERT_SV_SETSV_COW;
4755 PerlIO_printf(Perl_debug_log, "Fast copy on write: %p -> %p\n",
4756 (void*)sstr, (void*)dstr);
4763 if (SvTHINKFIRST(dstr))
4764 sv_force_normal_flags(dstr, SV_COW_DROP_PV);
4765 else if (SvPVX_const(dstr))
4766 Safefree(SvPVX_mutable(dstr));
4770 SvUPGRADE(dstr, SVt_COW);
4772 assert (SvPOK(sstr));
4773 assert (SvPOKp(sstr));
4774 # ifdef PERL_OLD_COPY_ON_WRITE
4775 assert (!SvIOK(sstr));
4776 assert (!SvIOKp(sstr));
4777 assert (!SvNOK(sstr));
4778 assert (!SvNOKp(sstr));
4781 if (SvIsCOW(sstr)) {
4783 if (SvLEN(sstr) == 0) {
4784 /* source is a COW shared hash key. */
4785 DEBUG_C(PerlIO_printf(Perl_debug_log,
4786 "Fast copy on write: Sharing hash\n"));
4787 new_pv = HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr))));
4790 # ifdef PERL_OLD_COPY_ON_WRITE
4791 SV_COW_NEXT_SV_SET(dstr, SV_COW_NEXT_SV(sstr));
4793 assert(SvCUR(sstr)+1 < SvLEN(sstr));
4794 assert(CowREFCNT(sstr) < SV_COW_REFCNT_MAX);
4797 assert ((SvFLAGS(sstr) & CAN_COW_MASK) == CAN_COW_FLAGS);
4798 SvUPGRADE(sstr, SVt_COW);
4800 DEBUG_C(PerlIO_printf(Perl_debug_log,
4801 "Fast copy on write: Converting sstr to COW\n"));
4802 # ifdef PERL_OLD_COPY_ON_WRITE
4803 SV_COW_NEXT_SV_SET(dstr, sstr);
4805 CowREFCNT(sstr) = 0;
4808 # ifdef PERL_OLD_COPY_ON_WRITE
4809 SV_COW_NEXT_SV_SET(sstr, dstr);
4811 # ifdef PERL_DEBUG_READONLY_COW
4812 if (already) sv_buf_to_rw(sstr);
4816 new_pv = SvPVX_mutable(sstr);
4820 SvPV_set(dstr, new_pv);
4821 SvFLAGS(dstr) = (SVt_COW|SVf_POK|SVp_POK|SVf_IsCOW);
4824 SvLEN_set(dstr, len);
4825 SvCUR_set(dstr, cur);
4834 =for apidoc sv_setpvn
4836 Copies a string (possibly containing embedded C<NUL> characters) into an SV.
4837 The C<len> parameter indicates the number of
4838 bytes to be copied. If the C<ptr> argument is NULL the SV will become
4839 undefined. Does not handle 'set' magic. See C<sv_setpvn_mg>.
4845 Perl_sv_setpvn(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4849 PERL_ARGS_ASSERT_SV_SETPVN;
4851 SV_CHECK_THINKFIRST_COW_DROP(sv);
4857 /* len is STRLEN which is unsigned, need to copy to signed */
4860 Perl_croak(aTHX_ "panic: sv_setpvn called with negative strlen %"
4863 SvUPGRADE(sv, SVt_PV);
4865 dptr = SvGROW(sv, len + 1);
4866 Move(ptr,dptr,len,char);
4869 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4871 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4875 =for apidoc sv_setpvn_mg
4877 Like C<sv_setpvn>, but also handles 'set' magic.
4883 Perl_sv_setpvn_mg(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4885 PERL_ARGS_ASSERT_SV_SETPVN_MG;
4887 sv_setpvn(sv,ptr,len);
4892 =for apidoc sv_setpv
4894 Copies a string into an SV. The string must be terminated with a C<NUL>
4896 Does not handle 'set' magic. See C<sv_setpv_mg>.
4902 Perl_sv_setpv(pTHX_ SV *const sv, const char *const ptr)
4906 PERL_ARGS_ASSERT_SV_SETPV;
4908 SV_CHECK_THINKFIRST_COW_DROP(sv);
4914 SvUPGRADE(sv, SVt_PV);
4916 SvGROW(sv, len + 1);
4917 Move(ptr,SvPVX(sv),len+1,char);
4919 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4921 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4925 =for apidoc sv_setpv_mg
4927 Like C<sv_setpv>, but also handles 'set' magic.
4933 Perl_sv_setpv_mg(pTHX_ SV *const sv, const char *const ptr)
4935 PERL_ARGS_ASSERT_SV_SETPV_MG;
4942 Perl_sv_sethek(pTHX_ SV *const sv, const HEK *const hek)
4944 PERL_ARGS_ASSERT_SV_SETHEK;
4950 if (HEK_LEN(hek) == HEf_SVKEY) {
4951 sv_setsv(sv, *(SV**)HEK_KEY(hek));
4954 const int flags = HEK_FLAGS(hek);
4955 if (flags & HVhek_WASUTF8) {
4956 STRLEN utf8_len = HEK_LEN(hek);
4957 char *as_utf8 = (char *)bytes_to_utf8((U8*)HEK_KEY(hek), &utf8_len);
4958 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
4961 } else if (flags & HVhek_UNSHARED) {
4962 sv_setpvn(sv, HEK_KEY(hek), HEK_LEN(hek));
4965 else SvUTF8_off(sv);
4969 SV_CHECK_THINKFIRST_COW_DROP(sv);
4970 SvUPGRADE(sv, SVt_PV);
4972 SvPV_set(sv,(char *)HEK_KEY(share_hek_hek(hek)));
4973 SvCUR_set(sv, HEK_LEN(hek));
4979 else SvUTF8_off(sv);
4987 =for apidoc sv_usepvn_flags
4989 Tells an SV to use C<ptr> to find its string value. Normally the
4990 string is stored inside the SV, but sv_usepvn allows the SV to use an
4991 outside string. The C<ptr> should point to memory that was allocated
4992 by L<Newx|perlclib/Memory Management and String Handling>. It must be
4993 the start of a Newx-ed block of memory, and not a pointer to the
4994 middle of it (beware of L<OOK|perlguts/Offsets> and copy-on-write),
4995 and not be from a non-Newx memory allocator like C<malloc>. The
4996 string length, C<len>, must be supplied. By default this function
4997 will C<Renew> (i.e. realloc, move) the memory pointed to by C<ptr>,
4998 so that pointer should not be freed or used by the programmer after
4999 giving it to sv_usepvn, and neither should any pointers from "behind"
5000 that pointer (e.g. ptr + 1) be used.
5002 If C<flags> & SV_SMAGIC is true, will call SvSETMAGIC. If C<flags> &
5003 SV_HAS_TRAILING_NUL is true, then C<ptr[len]> must be C<NUL>, and the realloc
5004 will be skipped (i.e. the buffer is actually at least 1 byte longer than
5005 C<len>, and already meets the requirements for storing in C<SvPVX>).
5011 Perl_sv_usepvn_flags(pTHX_ SV *const sv, char *ptr, const STRLEN len, const U32 flags)
5015 PERL_ARGS_ASSERT_SV_USEPVN_FLAGS;
5017 SV_CHECK_THINKFIRST_COW_DROP(sv);
5018 SvUPGRADE(sv, SVt_PV);
5021 if (flags & SV_SMAGIC)
5025 if (SvPVX_const(sv))
5029 if (flags & SV_HAS_TRAILING_NUL)
5030 assert(ptr[len] == '\0');
5033 allocate = (flags & SV_HAS_TRAILING_NUL)
5035 #ifdef Perl_safesysmalloc_size
5038 PERL_STRLEN_ROUNDUP(len + 1);
5040 if (flags & SV_HAS_TRAILING_NUL) {
5041 /* It's long enough - do nothing.
5042 Specifically Perl_newCONSTSUB is relying on this. */
5045 /* Force a move to shake out bugs in callers. */
5046 char *new_ptr = (char*)safemalloc(allocate);
5047 Copy(ptr, new_ptr, len, char);
5048 PoisonFree(ptr,len,char);
5052 ptr = (char*) saferealloc (ptr, allocate);
5055 #ifdef Perl_safesysmalloc_size
5056 SvLEN_set(sv, Perl_safesysmalloc_size(ptr));
5058 SvLEN_set(sv, allocate);
5062 if (!(flags & SV_HAS_TRAILING_NUL)) {
5065 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5067 if (flags & SV_SMAGIC)
5071 #ifdef PERL_OLD_COPY_ON_WRITE
5072 /* Need to do this *after* making the SV normal, as we need the buffer
5073 pointer to remain valid until after we've copied it. If we let go too early,
5074 another thread could invalidate it by unsharing last of the same hash key
5075 (which it can do by means other than releasing copy-on-write Svs)
5076 or by changing the other copy-on-write SVs in the loop. */
5078 S_sv_release_COW(pTHX_ SV *sv, const char *pvx, SV *after)
5080 PERL_ARGS_ASSERT_SV_RELEASE_COW;
5082 { /* this SV was SvIsCOW_normal(sv) */
5083 /* we need to find the SV pointing to us. */
5084 SV *current = SV_COW_NEXT_SV(after);
5086 if (current == sv) {
5087 /* The SV we point to points back to us (there were only two of us
5089 Hence other SV is no longer copy on write either. */
5091 sv_buf_to_rw(after);
5093 /* We need to follow the pointers around the loop. */
5095 while ((next = SV_COW_NEXT_SV(current)) != sv) {
5098 /* don't loop forever if the structure is bust, and we have
5099 a pointer into a closed loop. */
5100 assert (current != after);
5101 assert (SvPVX_const(current) == pvx);
5103 /* Make the SV before us point to the SV after us. */
5104 SV_COW_NEXT_SV_SET(current, after);
5110 =for apidoc sv_force_normal_flags
5112 Undo various types of fakery on an SV, where fakery means
5113 "more than" a string: if the PV is a shared string, make
5114 a private copy; if we're a ref, stop refing; if we're a glob, downgrade to
5115 an xpvmg; if we're a copy-on-write scalar, this is the on-write time when
5116 we do the copy, and is also used locally; if this is a
5117 vstring, drop the vstring magic. If C<SV_COW_DROP_PV> is set
5118 then a copy-on-write scalar drops its PV buffer (if any) and becomes
5119 SvPOK_off rather than making a copy. (Used where this
5120 scalar is about to be set to some other value.) In addition,
5121 the C<flags> parameter gets passed to C<sv_unref_flags()>
5122 when unreffing. C<sv_force_normal> calls this function
5123 with flags set to 0.
5125 This function is expected to be used to signal to perl that this SV is
5126 about to be written to, and any extra book-keeping needs to be taken care
5127 of. Hence, it croaks on read-only values.
5133 S_sv_uncow(pTHX_ SV * const sv, const U32 flags)
5135 assert(SvIsCOW(sv));
5138 const char * const pvx = SvPVX_const(sv);
5139 const STRLEN len = SvLEN(sv);
5140 const STRLEN cur = SvCUR(sv);
5141 # ifdef PERL_OLD_COPY_ON_WRITE
5142 /* next COW sv in the loop. If len is 0 then this is a shared-hash
5143 key scalar, so we mustn't attempt to call SV_COW_NEXT_SV(), as
5144 we'll fail an assertion. */
5145 SV * const next = len ? SV_COW_NEXT_SV(sv) : 0;
5149 PerlIO_printf(Perl_debug_log,
5150 "Copy on write: Force normal %ld\n",
5155 # ifdef PERL_NEW_COPY_ON_WRITE
5156 if (len && CowREFCNT(sv) == 0)
5157 /* We own the buffer ourselves. */
5163 /* This SV doesn't own the buffer, so need to Newx() a new one: */
5164 # ifdef PERL_NEW_COPY_ON_WRITE
5165 /* Must do this first, since the macro uses SvPVX. */
5175 if (flags & SV_COW_DROP_PV) {
5176 /* OK, so we don't need to copy our buffer. */
5179 SvGROW(sv, cur + 1);
5180 Move(pvx,SvPVX(sv),cur,char);
5185 # ifdef PERL_OLD_COPY_ON_WRITE
5186 sv_release_COW(sv, pvx, next);
5189 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5196 const char * const pvx = SvPVX_const(sv);
5197 const STRLEN len = SvCUR(sv);
5201 if (flags & SV_COW_DROP_PV) {
5202 /* OK, so we don't need to copy our buffer. */
5205 SvGROW(sv, len + 1);
5206 Move(pvx,SvPVX(sv),len,char);
5209 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5215 Perl_sv_force_normal_flags(pTHX_ SV *const sv, const U32 flags)
5217 PERL_ARGS_ASSERT_SV_FORCE_NORMAL_FLAGS;
5220 Perl_croak_no_modify();
5221 else if (SvIsCOW(sv) && LIKELY(SvTYPE(sv) != SVt_PVHV))
5222 S_sv_uncow(aTHX_ sv, flags);
5224 sv_unref_flags(sv, flags);
5225 else if (SvFAKE(sv) && isGV_with_GP(sv))
5226 sv_unglob(sv, flags);
5227 else if (SvFAKE(sv) && isREGEXP(sv)) {
5228 /* Need to downgrade the REGEXP to a simple(r) scalar. This is analogous
5229 to sv_unglob. We only need it here, so inline it. */
5230 const bool islv = SvTYPE(sv) == SVt_PVLV;
5231 const svtype new_type =
5232 islv ? SVt_NULL : SvMAGIC(sv) || SvSTASH(sv) ? SVt_PVMG : SVt_PV;
5233 SV *const temp = newSV_type(new_type);
5234 regexp *const temp_p = ReANY((REGEXP *)sv);
5236 if (new_type == SVt_PVMG) {
5237 SvMAGIC_set(temp, SvMAGIC(sv));
5238 SvMAGIC_set(sv, NULL);
5239 SvSTASH_set(temp, SvSTASH(sv));
5240 SvSTASH_set(sv, NULL);
5242 if (!islv) SvCUR_set(temp, SvCUR(sv));
5243 /* Remember that SvPVX is in the head, not the body. But
5244 RX_WRAPPED is in the body. */
5245 assert(ReANY((REGEXP *)sv)->mother_re);
5246 /* Their buffer is already owned by someone else. */
5247 if (flags & SV_COW_DROP_PV) {
5248 /* SvLEN is already 0. For SVt_REGEXP, we have a brand new
5249 zeroed body. For SVt_PVLV, it should have been set to 0
5250 before turning into a regexp. */
5251 assert(!SvLEN(islv ? sv : temp));
5252 sv->sv_u.svu_pv = 0;
5255 sv->sv_u.svu_pv = savepvn(RX_WRAPPED((REGEXP *)sv), SvCUR(sv));
5256 SvLEN_set(islv ? sv : temp, SvCUR(sv)+1);
5260 /* Now swap the rest of the bodies. */
5264 SvFLAGS(sv) &= ~SVTYPEMASK;
5265 SvFLAGS(sv) |= new_type;
5266 SvANY(sv) = SvANY(temp);
5269 SvFLAGS(temp) &= ~(SVTYPEMASK);
5270 SvFLAGS(temp) |= SVt_REGEXP|SVf_FAKE;
5271 SvANY(temp) = temp_p;
5272 temp->sv_u.svu_rx = (regexp *)temp_p;
5274 SvREFCNT_dec_NN(temp);
5276 else if (SvVOK(sv)) sv_unmagic(sv, PERL_MAGIC_vstring);
5282 Efficient removal of characters from the beginning of the string buffer.
5283 SvPOK(sv), or at least SvPOKp(sv), must be true and the C<ptr> must be a
5284 pointer to somewhere inside the string buffer. The C<ptr> becomes the first
5285 character of the adjusted string. Uses the "OOK hack". On return, only
5286 SvPOK(sv) and SvPOKp(sv) among the OK flags will be true.
5288 Beware: after this function returns, C<ptr> and SvPVX_const(sv) may no longer
5289 refer to the same chunk of data.
5291 The unfortunate similarity of this function's name to that of Perl's C<chop>
5292 operator is strictly coincidental. This function works from the left;
5293 C<chop> works from the right.
5299 Perl_sv_chop(pTHX_ SV *const sv, const char *const ptr)
5310 PERL_ARGS_ASSERT_SV_CHOP;
5312 if (!ptr || !SvPOKp(sv))
5314 delta = ptr - SvPVX_const(sv);
5316 /* Nothing to do. */
5319 max_delta = SvLEN(sv) ? SvLEN(sv) : SvCUR(sv);
5320 if (delta > max_delta)
5321 Perl_croak(aTHX_ "panic: sv_chop ptr=%p, start=%p, end=%p",
5322 ptr, SvPVX_const(sv), SvPVX_const(sv) + max_delta);
5323 /* SvPVX(sv) may move in SV_CHECK_THINKFIRST(sv), so don't use ptr any more */
5324 SV_CHECK_THINKFIRST(sv);
5325 SvPOK_only_UTF8(sv);
5328 if (!SvLEN(sv)) { /* make copy of shared string */
5329 const char *pvx = SvPVX_const(sv);
5330 const STRLEN len = SvCUR(sv);
5331 SvGROW(sv, len + 1);
5332 Move(pvx,SvPVX(sv),len,char);
5338 SvOOK_offset(sv, old_delta);
5340 SvLEN_set(sv, SvLEN(sv) - delta);
5341 SvCUR_set(sv, SvCUR(sv) - delta);
5342 SvPV_set(sv, SvPVX(sv) + delta);
5344 p = (U8 *)SvPVX_const(sv);
5347 /* how many bytes were evacuated? we will fill them with sentinel
5348 bytes, except for the part holding the new offset of course. */
5351 evacn += (old_delta < 0x100 ? 1 : 1 + sizeof(STRLEN));
5353 assert(evacn <= delta + old_delta);
5357 /* This sets 'delta' to the accumulated value of all deltas so far */
5361 /* If 'delta' fits in a byte, store it just prior to the new beginning of
5362 * the string; otherwise store a 0 byte there and store 'delta' just prior
5363 * to that, using as many bytes as a STRLEN occupies. Thus it overwrites a
5364 * portion of the chopped part of the string */
5365 if (delta < 0x100) {
5369 p -= sizeof(STRLEN);
5370 Copy((U8*)&delta, p, sizeof(STRLEN), U8);
5374 /* Fill the preceding buffer with sentinals to verify that no-one is
5384 =for apidoc sv_catpvn
5386 Concatenates the string onto the end of the string which is in the SV. The
5387 C<len> indicates number of bytes to copy. If the SV has the UTF-8
5388 status set, then the bytes appended should be valid UTF-8.
5389 Handles 'get' magic, but not 'set' magic. See C<sv_catpvn_mg>.
5391 =for apidoc sv_catpvn_flags
5393 Concatenates the string onto the end of the string which is in the SV. The
5394 C<len> indicates number of bytes to copy. If the SV has the UTF-8
5395 status set, then the bytes appended should be valid UTF-8.
5396 If C<flags> has the C<SV_SMAGIC> bit set, will
5397 C<mg_set> on C<dsv> afterwards if appropriate.
5398 C<sv_catpvn> and C<sv_catpvn_nomg> are implemented
5399 in terms of this function.
5405 Perl_sv_catpvn_flags(pTHX_ SV *const dsv, const char *sstr, const STRLEN slen, const I32 flags)
5408 const char * const dstr = SvPV_force_flags(dsv, dlen, flags);
5410 PERL_ARGS_ASSERT_SV_CATPVN_FLAGS;
5411 assert((flags & (SV_CATBYTES|SV_CATUTF8)) != (SV_CATBYTES|SV_CATUTF8));
5413 if (!(flags & SV_CATBYTES) || !SvUTF8(dsv)) {
5414 if (flags & SV_CATUTF8 && !SvUTF8(dsv)) {
5415 sv_utf8_upgrade_flags_grow(dsv, 0, slen + 1);
5418 else SvGROW(dsv, dlen + slen + 1);
5420 sstr = SvPVX_const(dsv);
5421 Move(sstr, SvPVX(dsv) + dlen, slen, char);
5422 SvCUR_set(dsv, SvCUR(dsv) + slen);
5425 /* We inline bytes_to_utf8, to avoid an extra malloc. */
5426 const char * const send = sstr + slen;
5429 /* Something this code does not account for, which I think is
5430 impossible; it would require the same pv to be treated as
5431 bytes *and* utf8, which would indicate a bug elsewhere. */
5432 assert(sstr != dstr);
5434 SvGROW(dsv, dlen + slen * 2 + 1);
5435 d = (U8 *)SvPVX(dsv) + dlen;
5437 while (sstr < send) {
5438 append_utf8_from_native_byte(*sstr, &d);
5441 SvCUR_set(dsv, d-(const U8 *)SvPVX(dsv));
5444 (void)SvPOK_only_UTF8(dsv); /* validate pointer */
5446 if (flags & SV_SMAGIC)
5451 =for apidoc sv_catsv
5453 Concatenates the string from SV C<ssv> onto the end of the string in SV
5454 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5455 Handles 'get' magic on both SVs, but no 'set' magic. See C<sv_catsv_mg> and
5458 =for apidoc sv_catsv_flags
5460 Concatenates the string from SV C<ssv> onto the end of the string in SV
5461 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5462 If C<flags> include C<SV_GMAGIC> bit set, will call C<mg_get> on both SVs if
5463 appropriate. If C<flags> include C<SV_SMAGIC>, C<mg_set> will be called on
5464 the modified SV afterward, if appropriate. C<sv_catsv>, C<sv_catsv_nomg>,
5465 and C<sv_catsv_mg> are implemented in terms of this function.
5470 Perl_sv_catsv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
5472 PERL_ARGS_ASSERT_SV_CATSV_FLAGS;
5476 const char *spv = SvPV_flags_const(ssv, slen, flags);
5478 if (flags & SV_GMAGIC)
5480 sv_catpvn_flags(dsv, spv, slen,
5481 DO_UTF8(ssv) ? SV_CATUTF8 : SV_CATBYTES);
5482 if (flags & SV_SMAGIC)
5489 =for apidoc sv_catpv
5491 Concatenates the C<NUL>-terminated string onto the end of the string which is
5493 If the SV has the UTF-8 status set, then the bytes appended should be
5494 valid UTF-8. Handles 'get' magic, but not 'set' magic. See C<sv_catpv_mg>.
5499 Perl_sv_catpv(pTHX_ SV *const sv, const char *ptr)
5505 PERL_ARGS_ASSERT_SV_CATPV;
5509 junk = SvPV_force(sv, tlen);
5511 SvGROW(sv, tlen + len + 1);
5513 ptr = SvPVX_const(sv);
5514 Move(ptr,SvPVX(sv)+tlen,len+1,char);
5515 SvCUR_set(sv, SvCUR(sv) + len);
5516 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5521 =for apidoc sv_catpv_flags
5523 Concatenates the C<NUL>-terminated string onto the end of the string which is
5525 If the SV has the UTF-8 status set, then the bytes appended should
5526 be valid UTF-8. If C<flags> has the C<SV_SMAGIC> bit set, will C<mg_set>
5527 on the modified SV if appropriate.
5533 Perl_sv_catpv_flags(pTHX_ SV *dstr, const char *sstr, const I32 flags)
5535 PERL_ARGS_ASSERT_SV_CATPV_FLAGS;
5536 sv_catpvn_flags(dstr, sstr, strlen(sstr), flags);
5540 =for apidoc sv_catpv_mg
5542 Like C<sv_catpv>, but also handles 'set' magic.
5548 Perl_sv_catpv_mg(pTHX_ SV *const sv, const char *const ptr)
5550 PERL_ARGS_ASSERT_SV_CATPV_MG;
5559 Creates a new SV. A non-zero C<len> parameter indicates the number of
5560 bytes of preallocated string space the SV should have. An extra byte for a
5561 trailing C<NUL> is also reserved. (SvPOK is not set for the SV even if string
5562 space is allocated.) The reference count for the new SV is set to 1.
5564 In 5.9.3, newSV() replaces the older NEWSV() API, and drops the first
5565 parameter, I<x>, a debug aid which allowed callers to identify themselves.
5566 This aid has been superseded by a new build option, PERL_MEM_LOG (see
5567 L<perlhacktips/PERL_MEM_LOG>). The older API is still there for use in XS
5568 modules supporting older perls.
5574 Perl_newSV(pTHX_ const STRLEN len)
5580 sv_upgrade(sv, SVt_PV);
5581 SvGROW(sv, len + 1);
5586 =for apidoc sv_magicext
5588 Adds magic to an SV, upgrading it if necessary. Applies the
5589 supplied vtable and returns a pointer to the magic added.
5591 Note that C<sv_magicext> will allow things that C<sv_magic> will not.
5592 In particular, you can add magic to SvREADONLY SVs, and add more than
5593 one instance of the same 'how'.
5595 If C<namlen> is greater than zero then a C<savepvn> I<copy> of C<name> is
5596 stored, if C<namlen> is zero then C<name> is stored as-is and - as another
5597 special case - if C<(name && namlen == HEf_SVKEY)> then C<name> is assumed
5598 to contain an C<SV*> and is stored as-is with its REFCNT incremented.
5600 (This is now used as a subroutine by C<sv_magic>.)
5605 Perl_sv_magicext(pTHX_ SV *const sv, SV *const obj, const int how,
5606 const MGVTBL *const vtable, const char *const name, const I32 namlen)
5610 PERL_ARGS_ASSERT_SV_MAGICEXT;
5612 if (SvTYPE(sv)==SVt_PVAV) { assert (!AvPAD_NAMELIST(sv)); }
5614 SvUPGRADE(sv, SVt_PVMG);
5615 Newxz(mg, 1, MAGIC);
5616 mg->mg_moremagic = SvMAGIC(sv);
5617 SvMAGIC_set(sv, mg);
5619 /* Sometimes a magic contains a reference loop, where the sv and
5620 object refer to each other. To prevent a reference loop that
5621 would prevent such objects being freed, we look for such loops
5622 and if we find one we avoid incrementing the object refcount.
5624 Note we cannot do this to avoid self-tie loops as intervening RV must
5625 have its REFCNT incremented to keep it in existence.
5628 if (!obj || obj == sv ||
5629 how == PERL_MAGIC_arylen ||
5630 how == PERL_MAGIC_symtab ||
5631 (SvTYPE(obj) == SVt_PVGV &&
5632 (GvSV(obj) == sv || GvHV(obj) == (const HV *)sv
5633 || GvAV(obj) == (const AV *)sv || GvCV(obj) == (const CV *)sv
5634 || GvIOp(obj) == (const IO *)sv || GvFORM(obj) == (const CV *)sv)))
5639 mg->mg_obj = SvREFCNT_inc_simple(obj);
5640 mg->mg_flags |= MGf_REFCOUNTED;
5643 /* Normal self-ties simply pass a null object, and instead of
5644 using mg_obj directly, use the SvTIED_obj macro to produce a
5645 new RV as needed. For glob "self-ties", we are tieing the PVIO
5646 with an RV obj pointing to the glob containing the PVIO. In
5647 this case, to avoid a reference loop, we need to weaken the
5651 if (how == PERL_MAGIC_tiedscalar && SvTYPE(sv) == SVt_PVIO &&
5652 obj && SvROK(obj) && GvIO(SvRV(obj)) == (const IO *)sv)
5658 mg->mg_len = namlen;
5661 mg->mg_ptr = savepvn(name, namlen);
5662 else if (namlen == HEf_SVKEY) {
5663 /* Yes, this is casting away const. This is only for the case of
5664 HEf_SVKEY. I think we need to document this aberation of the
5665 constness of the API, rather than making name non-const, as
5666 that change propagating outwards a long way. */
5667 mg->mg_ptr = (char*)SvREFCNT_inc_simple_NN((SV *)name);
5669 mg->mg_ptr = (char *) name;
5671 mg->mg_virtual = (MGVTBL *) vtable;
5678 Perl_sv_magicext_mglob(pTHX_ SV *sv)
5680 PERL_ARGS_ASSERT_SV_MAGICEXT_MGLOB;
5681 if (SvTYPE(sv) == SVt_PVLV && LvTYPE(sv) == 'y') {
5682 /* This sv is only a delegate. //g magic must be attached to
5687 #ifdef PERL_OLD_COPY_ON_WRITE
5689 sv_force_normal_flags(sv, 0);
5691 return sv_magicext(sv, NULL, PERL_MAGIC_regex_global,
5692 &PL_vtbl_mglob, 0, 0);
5696 =for apidoc sv_magic
5698 Adds magic to an SV. First upgrades C<sv> to type C<SVt_PVMG> if
5699 necessary, then adds a new magic item of type C<how> to the head of the
5702 See C<sv_magicext> (which C<sv_magic> now calls) for a description of the
5703 handling of the C<name> and C<namlen> arguments.
5705 You need to use C<sv_magicext> to add magic to SvREADONLY SVs and also
5706 to add more than one instance of the same 'how'.
5712 Perl_sv_magic(pTHX_ SV *const sv, SV *const obj, const int how,
5713 const char *const name, const I32 namlen)
5715 const MGVTBL *vtable;
5718 unsigned int vtable_index;
5720 PERL_ARGS_ASSERT_SV_MAGIC;
5722 if (how < 0 || (unsigned)how >= C_ARRAY_LENGTH(PL_magic_data)
5723 || ((flags = PL_magic_data[how]),
5724 (vtable_index = flags & PERL_MAGIC_VTABLE_MASK)
5725 > magic_vtable_max))
5726 Perl_croak(aTHX_ "Don't know how to handle magic of type \\%o", how);
5728 /* PERL_MAGIC_ext is reserved for use by extensions not perl internals.
5729 Useful for attaching extension internal data to perl vars.
5730 Note that multiple extensions may clash if magical scalars
5731 etc holding private data from one are passed to another. */
5733 vtable = (vtable_index == magic_vtable_max)
5734 ? NULL : PL_magic_vtables + vtable_index;
5736 #ifdef PERL_OLD_COPY_ON_WRITE
5738 sv_force_normal_flags(sv, 0);
5740 if (SvREADONLY(sv)) {
5742 !PERL_MAGIC_TYPE_READONLY_ACCEPTABLE(how)
5745 Perl_croak_no_modify();
5748 if (SvMAGICAL(sv) || (how == PERL_MAGIC_taint && SvTYPE(sv) >= SVt_PVMG)) {
5749 if (SvMAGIC(sv) && (mg = mg_find(sv, how))) {
5750 /* sv_magic() refuses to add a magic of the same 'how' as an
5753 if (how == PERL_MAGIC_taint)
5759 /* Force pos to be stored as characters, not bytes. */
5760 if (SvMAGICAL(sv) && DO_UTF8(sv)
5761 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
5763 && mg->mg_flags & MGf_BYTES) {
5764 mg->mg_len = (SSize_t)sv_pos_b2u_flags(sv, (STRLEN)mg->mg_len,
5766 mg->mg_flags &= ~MGf_BYTES;
5769 /* Rest of work is done else where */
5770 mg = sv_magicext(sv,obj,how,vtable,name,namlen);
5773 case PERL_MAGIC_taint:
5776 case PERL_MAGIC_ext:
5777 case PERL_MAGIC_dbfile:
5784 S_sv_unmagicext_flags(pTHX_ SV *const sv, const int type, MGVTBL *vtbl, const U32 flags)
5791 if (SvTYPE(sv) < SVt_PVMG || !SvMAGIC(sv))
5793 mgp = &(((XPVMG*) SvANY(sv))->xmg_u.xmg_magic);
5794 for (mg = *mgp; mg; mg = *mgp) {
5795 const MGVTBL* const virt = mg->mg_virtual;
5796 if (mg->mg_type == type && (!flags || virt == vtbl)) {
5797 *mgp = mg->mg_moremagic;
5798 if (virt && virt->svt_free)
5799 virt->svt_free(aTHX_ sv, mg);
5800 if (mg->mg_ptr && mg->mg_type != PERL_MAGIC_regex_global) {
5802 Safefree(mg->mg_ptr);
5803 else if (mg->mg_len == HEf_SVKEY)
5804 SvREFCNT_dec(MUTABLE_SV(mg->mg_ptr));
5805 else if (mg->mg_type == PERL_MAGIC_utf8)
5806 Safefree(mg->mg_ptr);
5808 if (mg->mg_flags & MGf_REFCOUNTED)
5809 SvREFCNT_dec(mg->mg_obj);
5813 mgp = &mg->mg_moremagic;
5816 if (SvMAGICAL(sv)) /* if we're under save_magic, wait for restore_magic; */
5817 mg_magical(sv); /* else fix the flags now */
5821 SvFLAGS(sv) |= (SvFLAGS(sv) & (SVp_IOK|SVp_NOK|SVp_POK)) >> PRIVSHIFT;
5827 =for apidoc sv_unmagic
5829 Removes all magic of type C<type> from an SV.
5835 Perl_sv_unmagic(pTHX_ SV *const sv, const int type)
5837 PERL_ARGS_ASSERT_SV_UNMAGIC;
5838 return S_sv_unmagicext_flags(aTHX_ sv, type, NULL, 0);
5842 =for apidoc sv_unmagicext
5844 Removes all magic of type C<type> with the specified C<vtbl> from an SV.
5850 Perl_sv_unmagicext(pTHX_ SV *const sv, const int type, MGVTBL *vtbl)
5852 PERL_ARGS_ASSERT_SV_UNMAGICEXT;
5853 return S_sv_unmagicext_flags(aTHX_ sv, type, vtbl, 1);
5857 =for apidoc sv_rvweaken
5859 Weaken a reference: set the C<SvWEAKREF> flag on this RV; give the
5860 referred-to SV C<PERL_MAGIC_backref> magic if it hasn't already; and
5861 push a back-reference to this RV onto the array of backreferences
5862 associated with that magic. If the RV is magical, set magic will be
5863 called after the RV is cleared.
5869 Perl_sv_rvweaken(pTHX_ SV *const sv)
5873 PERL_ARGS_ASSERT_SV_RVWEAKEN;
5875 if (!SvOK(sv)) /* let undefs pass */
5878 Perl_croak(aTHX_ "Can't weaken a nonreference");
5879 else if (SvWEAKREF(sv)) {
5880 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is already weak");
5883 else if (SvREADONLY(sv)) croak_no_modify();
5885 Perl_sv_add_backref(aTHX_ tsv, sv);
5887 SvREFCNT_dec_NN(tsv);
5891 /* Give tsv backref magic if it hasn't already got it, then push a
5892 * back-reference to sv onto the array associated with the backref magic.
5894 * As an optimisation, if there's only one backref and it's not an AV,
5895 * store it directly in the HvAUX or mg_obj slot, avoiding the need to
5896 * allocate an AV. (Whether the slot holds an AV tells us whether this is
5900 /* A discussion about the backreferences array and its refcount:
5902 * The AV holding the backreferences is pointed to either as the mg_obj of
5903 * PERL_MAGIC_backref, or in the specific case of a HV, from the
5904 * xhv_backreferences field. The array is created with a refcount
5905 * of 2. This means that if during global destruction the array gets
5906 * picked on before its parent to have its refcount decremented by the
5907 * random zapper, it won't actually be freed, meaning it's still there for
5908 * when its parent gets freed.
5910 * When the parent SV is freed, the extra ref is killed by
5911 * Perl_sv_kill_backrefs. The other ref is killed, in the case of magic,
5912 * by mg_free() / MGf_REFCOUNTED, or for a hash, by Perl_hv_kill_backrefs.
5914 * When a single backref SV is stored directly, it is not reference
5919 Perl_sv_add_backref(pTHX_ SV *const tsv, SV *const sv)
5925 PERL_ARGS_ASSERT_SV_ADD_BACKREF;
5927 /* find slot to store array or singleton backref */
5929 if (SvTYPE(tsv) == SVt_PVHV) {
5930 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
5933 mg = mg_find(tsv, PERL_MAGIC_backref);
5935 mg = sv_magicext(tsv, NULL, PERL_MAGIC_backref, &PL_vtbl_backref, NULL, 0);
5936 svp = &(mg->mg_obj);
5939 /* create or retrieve the array */
5941 if ( (!*svp && SvTYPE(sv) == SVt_PVAV)
5942 || (*svp && SvTYPE(*svp) != SVt_PVAV)
5946 mg->mg_flags |= MGf_REFCOUNTED;
5949 SvREFCNT_inc_simple_void_NN(av);
5950 /* av now has a refcnt of 2; see discussion above */
5951 av_extend(av, *svp ? 2 : 1);
5953 /* move single existing backref to the array */
5954 AvARRAY(av)[++AvFILLp(av)] = *svp; /* av_push() */
5959 av = MUTABLE_AV(*svp);
5961 /* optimisation: store single backref directly in HvAUX or mg_obj */
5965 assert(SvTYPE(av) == SVt_PVAV);
5966 if (AvFILLp(av) >= AvMAX(av)) {
5967 av_extend(av, AvFILLp(av)+1);
5970 /* push new backref */
5971 AvARRAY(av)[++AvFILLp(av)] = sv; /* av_push() */
5974 /* delete a back-reference to ourselves from the backref magic associated
5975 * with the SV we point to.
5979 Perl_sv_del_backref(pTHX_ SV *const tsv, SV *const sv)
5983 PERL_ARGS_ASSERT_SV_DEL_BACKREF;
5985 if (SvTYPE(tsv) == SVt_PVHV) {
5987 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
5989 else if (SvIS_FREED(tsv) && PL_phase == PERL_PHASE_DESTRUCT) {
5990 /* It's possible for the the last (strong) reference to tsv to have
5991 become freed *before* the last thing holding a weak reference.
5992 If both survive longer than the backreferences array, then when
5993 the referent's reference count drops to 0 and it is freed, it's
5994 not able to chase the backreferences, so they aren't NULLed.
5996 For example, a CV holds a weak reference to its stash. If both the
5997 CV and the stash survive longer than the backreferences array,
5998 and the CV gets picked for the SvBREAK() treatment first,
5999 *and* it turns out that the stash is only being kept alive because
6000 of an our variable in the pad of the CV, then midway during CV
6001 destruction the stash gets freed, but CvSTASH() isn't set to NULL.
6002 It ends up pointing to the freed HV. Hence it's chased in here, and
6003 if this block wasn't here, it would hit the !svp panic just below.
6005 I don't believe that "better" destruction ordering is going to help
6006 here - during global destruction there's always going to be the
6007 chance that something goes out of order. We've tried to make it
6008 foolproof before, and it only resulted in evolutionary pressure on
6009 fools. Which made us look foolish for our hubris. :-(
6015 = SvMAGICAL(tsv) ? mg_find(tsv, PERL_MAGIC_backref) : NULL;
6016 svp = mg ? &(mg->mg_obj) : NULL;
6020 Perl_croak(aTHX_ "panic: del_backref, svp=0");
6022 /* It's possible that sv is being freed recursively part way through the
6023 freeing of tsv. If this happens, the backreferences array of tsv has
6024 already been freed, and so svp will be NULL. If this is the case,
6025 we should not panic. Instead, nothing needs doing, so return. */
6026 if (PL_phase == PERL_PHASE_DESTRUCT && SvREFCNT(tsv) == 0)
6028 Perl_croak(aTHX_ "panic: del_backref, *svp=%p phase=%s refcnt=%" UVuf,
6029 (void*)*svp, PL_phase_names[PL_phase], (UV)SvREFCNT(tsv));
6032 if (SvTYPE(*svp) == SVt_PVAV) {
6036 AV * const av = (AV*)*svp;
6038 assert(!SvIS_FREED(av));
6042 /* for an SV with N weak references to it, if all those
6043 * weak refs are deleted, then sv_del_backref will be called
6044 * N times and O(N^2) compares will be done within the backref
6045 * array. To ameliorate this potential slowness, we:
6046 * 1) make sure this code is as tight as possible;
6047 * 2) when looking for SV, look for it at both the head and tail of the
6048 * array first before searching the rest, since some create/destroy
6049 * patterns will cause the backrefs to be freed in order.
6056 SV **p = &svp[fill];
6057 SV *const topsv = *p;
6064 /* We weren't the last entry.
6065 An unordered list has this property that you
6066 can take the last element off the end to fill
6067 the hole, and it's still an unordered list :-)
6073 break; /* should only be one */
6080 AvFILLp(av) = fill-1;
6082 else if (SvIS_FREED(*svp) && PL_phase == PERL_PHASE_DESTRUCT) {
6083 /* freed AV; skip */
6086 /* optimisation: only a single backref, stored directly */
6088 Perl_croak(aTHX_ "panic: del_backref, *svp=%p, sv=%p",
6089 (void*)*svp, (void*)sv);
6096 Perl_sv_kill_backrefs(pTHX_ SV *const sv, AV *const av)
6102 PERL_ARGS_ASSERT_SV_KILL_BACKREFS;
6107 /* after multiple passes through Perl_sv_clean_all() for a thingy
6108 * that has badly leaked, the backref array may have gotten freed,
6109 * since we only protect it against 1 round of cleanup */
6110 if (SvIS_FREED(av)) {
6111 if (PL_in_clean_all) /* All is fair */
6114 "panic: magic_killbackrefs (freed backref AV/SV)");
6118 is_array = (SvTYPE(av) == SVt_PVAV);
6120 assert(!SvIS_FREED(av));
6123 last = svp + AvFILLp(av);
6126 /* optimisation: only a single backref, stored directly */
6132 while (svp <= last) {
6134 SV *const referrer = *svp;
6135 if (SvWEAKREF(referrer)) {
6136 /* XXX Should we check that it hasn't changed? */
6137 assert(SvROK(referrer));
6138 SvRV_set(referrer, 0);
6140 SvWEAKREF_off(referrer);
6141 SvSETMAGIC(referrer);
6142 } else if (SvTYPE(referrer) == SVt_PVGV ||
6143 SvTYPE(referrer) == SVt_PVLV) {
6144 assert(SvTYPE(sv) == SVt_PVHV); /* stash backref */
6145 /* You lookin' at me? */
6146 assert(GvSTASH(referrer));
6147 assert(GvSTASH(referrer) == (const HV *)sv);
6148 GvSTASH(referrer) = 0;
6149 } else if (SvTYPE(referrer) == SVt_PVCV ||
6150 SvTYPE(referrer) == SVt_PVFM) {
6151 if (SvTYPE(sv) == SVt_PVHV) { /* stash backref */
6152 /* You lookin' at me? */
6153 assert(CvSTASH(referrer));
6154 assert(CvSTASH(referrer) == (const HV *)sv);
6155 SvANY(MUTABLE_CV(referrer))->xcv_stash = 0;
6158 assert(SvTYPE(sv) == SVt_PVGV);
6159 /* You lookin' at me? */
6160 assert(CvGV(referrer));
6161 assert(CvGV(referrer) == (const GV *)sv);
6162 anonymise_cv_maybe(MUTABLE_GV(sv),
6163 MUTABLE_CV(referrer));
6168 "panic: magic_killbackrefs (flags=%"UVxf")",
6169 (UV)SvFLAGS(referrer));
6180 SvREFCNT_dec_NN(av); /* remove extra count added by sv_add_backref() */
6186 =for apidoc sv_insert
6188 Inserts a string at the specified offset/length within the SV. Similar to
6189 the Perl substr() function. Handles get magic.
6191 =for apidoc sv_insert_flags
6193 Same as C<sv_insert>, but the extra C<flags> are passed to the
6194 C<SvPV_force_flags> that applies to C<bigstr>.
6200 Perl_sv_insert_flags(pTHX_ SV *const bigstr, const STRLEN offset, const STRLEN len, const char *const little, const STRLEN littlelen, const U32 flags)
6206 SSize_t i; /* better be sizeof(STRLEN) or bad things happen */
6209 PERL_ARGS_ASSERT_SV_INSERT_FLAGS;
6212 Perl_croak(aTHX_ "Can't modify nonexistent substring");
6213 SvPV_force_flags(bigstr, curlen, flags);
6214 (void)SvPOK_only_UTF8(bigstr);
6215 if (offset + len > curlen) {
6216 SvGROW(bigstr, offset+len+1);
6217 Zero(SvPVX(bigstr)+curlen, offset+len-curlen, char);
6218 SvCUR_set(bigstr, offset+len);
6222 i = littlelen - len;
6223 if (i > 0) { /* string might grow */
6224 big = SvGROW(bigstr, SvCUR(bigstr) + i + 1);
6225 mid = big + offset + len;
6226 midend = bigend = big + SvCUR(bigstr);
6229 while (midend > mid) /* shove everything down */
6230 *--bigend = *--midend;
6231 Move(little,big+offset,littlelen,char);
6232 SvCUR_set(bigstr, SvCUR(bigstr) + i);
6237 Move(little,SvPVX(bigstr)+offset,len,char);
6242 big = SvPVX(bigstr);
6245 bigend = big + SvCUR(bigstr);
6247 if (midend > bigend)
6248 Perl_croak(aTHX_ "panic: sv_insert, midend=%p, bigend=%p",
6251 if (mid - big > bigend - midend) { /* faster to shorten from end */
6253 Move(little, mid, littlelen,char);
6256 i = bigend - midend;
6258 Move(midend, mid, i,char);
6262 SvCUR_set(bigstr, mid - big);
6264 else if ((i = mid - big)) { /* faster from front */
6265 midend -= littlelen;
6267 Move(big, midend - i, i, char);
6268 sv_chop(bigstr,midend-i);
6270 Move(little, mid, littlelen,char);
6272 else if (littlelen) {
6273 midend -= littlelen;
6274 sv_chop(bigstr,midend);
6275 Move(little,midend,littlelen,char);
6278 sv_chop(bigstr,midend);
6284 =for apidoc sv_replace
6286 Make the first argument a copy of the second, then delete the original.
6287 The target SV physically takes over ownership of the body of the source SV
6288 and inherits its flags; however, the target keeps any magic it owns,
6289 and any magic in the source is discarded.
6290 Note that this is a rather specialist SV copying operation; most of the
6291 time you'll want to use C<sv_setsv> or one of its many macro front-ends.
6297 Perl_sv_replace(pTHX_ SV *const sv, SV *const nsv)
6299 const U32 refcnt = SvREFCNT(sv);
6301 PERL_ARGS_ASSERT_SV_REPLACE;
6303 SV_CHECK_THINKFIRST_COW_DROP(sv);
6304 if (SvREFCNT(nsv) != 1) {
6305 Perl_croak(aTHX_ "panic: reference miscount on nsv in sv_replace()"
6306 " (%" UVuf " != 1)", (UV) SvREFCNT(nsv));
6308 if (SvMAGICAL(sv)) {
6312 sv_upgrade(nsv, SVt_PVMG);
6313 SvMAGIC_set(nsv, SvMAGIC(sv));
6314 SvFLAGS(nsv) |= SvMAGICAL(sv);
6316 SvMAGIC_set(sv, NULL);
6320 assert(!SvREFCNT(sv));
6321 #ifdef DEBUG_LEAKING_SCALARS
6322 sv->sv_flags = nsv->sv_flags;
6323 sv->sv_any = nsv->sv_any;
6324 sv->sv_refcnt = nsv->sv_refcnt;
6325 sv->sv_u = nsv->sv_u;
6327 StructCopy(nsv,sv,SV);
6329 if(SvTYPE(sv) == SVt_IV) {
6331 = (XPVIV*)((char*)&(sv->sv_u.svu_iv) - STRUCT_OFFSET(XPVIV, xiv_iv));
6335 #ifdef PERL_OLD_COPY_ON_WRITE
6336 if (SvIsCOW_normal(nsv)) {
6337 /* We need to follow the pointers around the loop to make the
6338 previous SV point to sv, rather than nsv. */
6341 while ((next = SV_COW_NEXT_SV(current)) != nsv) {
6344 assert(SvPVX_const(current) == SvPVX_const(nsv));
6346 /* Make the SV before us point to the SV after us. */
6348 PerlIO_printf(Perl_debug_log, "previous is\n");
6350 PerlIO_printf(Perl_debug_log,
6351 "move it from 0x%"UVxf" to 0x%"UVxf"\n",
6352 (UV) SV_COW_NEXT_SV(current), (UV) sv);
6354 SV_COW_NEXT_SV_SET(current, sv);
6357 SvREFCNT(sv) = refcnt;
6358 SvFLAGS(nsv) |= SVTYPEMASK; /* Mark as freed */
6363 /* We're about to free a GV which has a CV that refers back to us.
6364 * If that CV will outlive us, make it anonymous (i.e. fix up its CvGV
6368 S_anonymise_cv_maybe(pTHX_ GV *gv, CV* cv)
6373 PERL_ARGS_ASSERT_ANONYMISE_CV_MAYBE;
6376 assert(SvREFCNT(gv) == 0);
6377 assert(isGV(gv) && isGV_with_GP(gv));
6379 assert(!CvANON(cv));
6380 assert(CvGV(cv) == gv);
6381 assert(!CvNAMED(cv));
6383 /* will the CV shortly be freed by gp_free() ? */
6384 if (GvCV(gv) == cv && GvGP(gv)->gp_refcnt < 2 && SvREFCNT(cv) < 2) {
6385 SvANY(cv)->xcv_gv_u.xcv_gv = NULL;
6389 /* if not, anonymise: */
6390 gvname = (GvSTASH(gv) && HvNAME(GvSTASH(gv)) && HvENAME(GvSTASH(gv)))
6391 ? newSVhek(HvENAME_HEK(GvSTASH(gv)))
6392 : newSVpvn_flags( "__ANON__", 8, 0 );
6393 sv_catpvs(gvname, "::__ANON__");
6394 anongv = gv_fetchsv(gvname, GV_ADDMULTI, SVt_PVCV);
6395 SvREFCNT_dec_NN(gvname);
6399 SvANY(cv)->xcv_gv_u.xcv_gv = MUTABLE_GV(SvREFCNT_inc(anongv));
6404 =for apidoc sv_clear
6406 Clear an SV: call any destructors, free up any memory used by the body,
6407 and free the body itself. The SV's head is I<not> freed, although
6408 its type is set to all 1's so that it won't inadvertently be assumed
6409 to be live during global destruction etc.
6410 This function should only be called when REFCNT is zero. Most of the time
6411 you'll want to call C<sv_free()> (or its macro wrapper C<SvREFCNT_dec>)
6418 Perl_sv_clear(pTHX_ SV *const orig_sv)
6423 const struct body_details *sv_type_details;
6429 PERL_ARGS_ASSERT_SV_CLEAR;
6431 /* within this loop, sv is the SV currently being freed, and
6432 * iter_sv is the most recent AV or whatever that's being iterated
6433 * over to provide more SVs */
6439 assert(SvREFCNT(sv) == 0);
6440 assert(SvTYPE(sv) != (svtype)SVTYPEMASK);
6442 if (type <= SVt_IV) {
6443 /* See the comment in sv.h about the collusion between this
6444 * early return and the overloading of the NULL slots in the
6448 SvFLAGS(sv) &= SVf_BREAK;
6449 SvFLAGS(sv) |= SVTYPEMASK;
6453 assert(!SvOBJECT(sv) || type >= SVt_PVMG); /* objs are always >= MG */
6455 if (type >= SVt_PVMG) {
6457 if (!curse(sv, 1)) goto get_next_sv;
6458 type = SvTYPE(sv); /* destructor may have changed it */
6460 /* Free back-references before magic, in case the magic calls
6461 * Perl code that has weak references to sv. */
6462 if (type == SVt_PVHV) {
6463 Perl_hv_kill_backrefs(aTHX_ MUTABLE_HV(sv));
6467 else if (type == SVt_PVMG && SvPAD_OUR(sv)) {
6468 SvREFCNT_dec(SvOURSTASH(sv));
6470 else if (type == SVt_PVAV && AvPAD_NAMELIST(sv)) {
6471 assert(!SvMAGICAL(sv));
6472 } else if (SvMAGIC(sv)) {
6473 /* Free back-references before other types of magic. */
6474 sv_unmagic(sv, PERL_MAGIC_backref);
6478 if (type == SVt_PVMG && SvPAD_TYPED(sv))
6479 SvREFCNT_dec(SvSTASH(sv));
6482 /* case SVt_INVLIST: */
6485 IoIFP(sv) != PerlIO_stdin() &&
6486 IoIFP(sv) != PerlIO_stdout() &&
6487 IoIFP(sv) != PerlIO_stderr() &&
6488 !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6490 io_close(MUTABLE_IO(sv), FALSE);
6492 if (IoDIRP(sv) && !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6493 PerlDir_close(IoDIRP(sv));
6494 IoDIRP(sv) = (DIR*)NULL;
6495 Safefree(IoTOP_NAME(sv));
6496 Safefree(IoFMT_NAME(sv));
6497 Safefree(IoBOTTOM_NAME(sv));
6498 if ((const GV *)sv == PL_statgv)
6502 /* FIXME for plugins */
6504 pregfree2((REGEXP*) sv);
6508 cv_undef(MUTABLE_CV(sv));
6509 /* If we're in a stash, we don't own a reference to it.
6510 * However it does have a back reference to us, which needs to
6512 if ((stash = CvSTASH(sv)))
6513 sv_del_backref(MUTABLE_SV(stash), sv);
6516 if (PL_last_swash_hv == (const HV *)sv) {
6517 PL_last_swash_hv = NULL;
6519 if (HvTOTALKEYS((HV*)sv) > 0) {
6521 /* this statement should match the one at the beginning of
6522 * hv_undef_flags() */
6523 if ( PL_phase != PERL_PHASE_DESTRUCT
6524 && (name = HvNAME((HV*)sv)))
6526 if (PL_stashcache) {
6527 DEBUG_o(Perl_deb(aTHX_ "sv_clear clearing PL_stashcache for '%"SVf"'\n",
6529 (void)hv_deletehek(PL_stashcache,
6530 HvNAME_HEK((HV*)sv), G_DISCARD);
6532 hv_name_set((HV*)sv, NULL, 0, 0);
6535 /* save old iter_sv in unused SvSTASH field */
6536 assert(!SvOBJECT(sv));
6537 SvSTASH(sv) = (HV*)iter_sv;
6540 /* save old hash_index in unused SvMAGIC field */
6541 assert(!SvMAGICAL(sv));
6542 assert(!SvMAGIC(sv));
6543 ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index = hash_index;
6546 next_sv = Perl_hfree_next_entry(aTHX_ (HV*)sv, &hash_index);
6547 goto get_next_sv; /* process this new sv */
6549 /* free empty hash */
6550 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6551 assert(!HvARRAY((HV*)sv));
6555 AV* av = MUTABLE_AV(sv);
6556 if (PL_comppad == av) {
6560 if (AvREAL(av) && AvFILLp(av) > -1) {
6561 next_sv = AvARRAY(av)[AvFILLp(av)--];
6562 /* save old iter_sv in top-most slot of AV,
6563 * and pray that it doesn't get wiped in the meantime */
6564 AvARRAY(av)[AvMAX(av)] = iter_sv;
6566 goto get_next_sv; /* process this new sv */
6568 Safefree(AvALLOC(av));
6573 if (LvTYPE(sv) == 'T') { /* for tie: return HE to pool */
6574 SvREFCNT_dec(HeKEY_sv((HE*)LvTARG(sv)));
6575 HeNEXT((HE*)LvTARG(sv)) = PL_hv_fetch_ent_mh;
6576 PL_hv_fetch_ent_mh = (HE*)LvTARG(sv);
6578 else if (LvTYPE(sv) != 't') /* unless tie: unrefcnted fake SV** */
6579 SvREFCNT_dec(LvTARG(sv));
6580 if (isREGEXP(sv)) goto freeregexp;
6582 if (isGV_with_GP(sv)) {
6583 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
6584 && HvENAME_get(stash))
6585 mro_method_changed_in(stash);
6586 gp_free(MUTABLE_GV(sv));
6588 unshare_hek(GvNAME_HEK(sv));
6589 /* If we're in a stash, we don't own a reference to it.
6590 * However it does have a back reference to us, which
6591 * needs to be cleared. */
6592 if (!SvVALID(sv) && (stash = GvSTASH(sv)))
6593 sv_del_backref(MUTABLE_SV(stash), sv);
6595 /* FIXME. There are probably more unreferenced pointers to SVs
6596 * in the interpreter struct that we should check and tidy in
6597 * a similar fashion to this: */
6598 /* See also S_sv_unglob, which does the same thing. */
6599 if ((const GV *)sv == PL_last_in_gv)
6600 PL_last_in_gv = NULL;
6601 else if ((const GV *)sv == PL_statgv)
6603 else if ((const GV *)sv == PL_stderrgv)
6611 /* Don't bother with SvOOK_off(sv); as we're only going to
6615 SvOOK_offset(sv, offset);
6616 SvPV_set(sv, SvPVX_mutable(sv) - offset);
6617 /* Don't even bother with turning off the OOK flag. */
6622 SV * const target = SvRV(sv);
6624 sv_del_backref(target, sv);
6630 else if (SvPVX_const(sv)
6631 && !(SvTYPE(sv) == SVt_PVIO
6632 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6636 PerlIO_printf(Perl_debug_log, "Copy on write: clear\n");
6640 # ifdef PERL_OLD_COPY_ON_WRITE
6641 sv_release_COW(sv, SvPVX_const(sv), SV_COW_NEXT_SV(sv));
6643 if (CowREFCNT(sv)) {
6651 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6655 # ifdef PERL_OLD_COPY_ON_WRITE
6659 Safefree(SvPVX_mutable(sv));
6663 else if (SvPVX_const(sv) && SvLEN(sv)
6664 && !(SvTYPE(sv) == SVt_PVIO
6665 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6666 Safefree(SvPVX_mutable(sv));
6667 else if (SvPVX_const(sv) && SvIsCOW(sv)) {
6668 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6678 SvFLAGS(sv) &= SVf_BREAK;
6679 SvFLAGS(sv) |= SVTYPEMASK;
6681 sv_type_details = bodies_by_type + type;
6682 if (sv_type_details->arena) {
6683 del_body(((char *)SvANY(sv) + sv_type_details->offset),
6684 &PL_body_roots[type]);
6686 else if (sv_type_details->body_size) {
6687 safefree(SvANY(sv));
6691 /* caller is responsible for freeing the head of the original sv */
6692 if (sv != orig_sv && !SvREFCNT(sv))
6695 /* grab and free next sv, if any */
6703 else if (!iter_sv) {
6705 } else if (SvTYPE(iter_sv) == SVt_PVAV) {
6706 AV *const av = (AV*)iter_sv;
6707 if (AvFILLp(av) > -1) {
6708 sv = AvARRAY(av)[AvFILLp(av)--];
6710 else { /* no more elements of current AV to free */
6713 /* restore previous value, squirrelled away */
6714 iter_sv = AvARRAY(av)[AvMAX(av)];
6715 Safefree(AvALLOC(av));
6718 } else if (SvTYPE(iter_sv) == SVt_PVHV) {
6719 sv = Perl_hfree_next_entry(aTHX_ (HV*)iter_sv, &hash_index);
6720 if (!sv && !HvTOTALKEYS((HV *)iter_sv)) {
6721 /* no more elements of current HV to free */
6724 /* Restore previous values of iter_sv and hash_index,
6725 * squirrelled away */
6726 assert(!SvOBJECT(sv));
6727 iter_sv = (SV*)SvSTASH(sv);
6728 assert(!SvMAGICAL(sv));
6729 hash_index = ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index;
6731 /* perl -DA does not like rubbish in SvMAGIC. */
6735 /* free any remaining detritus from the hash struct */
6736 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6737 assert(!HvARRAY((HV*)sv));
6742 /* unrolled SvREFCNT_dec and sv_free2 follows: */
6746 if (!SvREFCNT(sv)) {
6750 if (--(SvREFCNT(sv)))
6754 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6755 "Attempt to free temp prematurely: SV 0x%"UVxf
6756 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6760 if (SvIMMORTAL(sv)) {
6761 /* make sure SvREFCNT(sv)==0 happens very seldom */
6762 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6771 /* This routine curses the sv itself, not the object referenced by sv. So
6772 sv does not have to be ROK. */
6775 S_curse(pTHX_ SV * const sv, const bool check_refcnt) {
6776 PERL_ARGS_ASSERT_CURSE;
6777 assert(SvOBJECT(sv));
6779 if (PL_defstash && /* Still have a symbol table? */
6785 stash = SvSTASH(sv);
6786 assert(SvTYPE(stash) == SVt_PVHV);
6787 if (HvNAME(stash)) {
6788 CV* destructor = NULL;
6789 assert (SvOOK(stash));
6790 if (!SvOBJECT(stash)) destructor = (CV *)SvSTASH(stash);
6791 if (!destructor || HvMROMETA(stash)->destroy_gen
6792 != PL_sub_generation)
6795 gv_fetchmeth_autoload(stash, "DESTROY", 7, 0);
6796 if (gv) destructor = GvCV(gv);
6797 if (!SvOBJECT(stash))
6800 destructor ? (HV *)destructor : ((HV *)0)+1;
6801 HvAUX(stash)->xhv_mro_meta->destroy_gen =
6805 assert(!destructor || destructor == ((CV *)0)+1
6806 || SvTYPE(destructor) == SVt_PVCV);
6807 if (destructor && destructor != ((CV *)0)+1
6808 /* A constant subroutine can have no side effects, so
6809 don't bother calling it. */
6810 && !CvCONST(destructor)
6811 /* Don't bother calling an empty destructor or one that
6812 returns immediately. */
6813 && (CvISXSUB(destructor)
6814 || (CvSTART(destructor)
6815 && (CvSTART(destructor)->op_next->op_type
6817 && (CvSTART(destructor)->op_next->op_type
6819 || CvSTART(destructor)->op_next->op_next->op_type
6825 SV* const tmpref = newRV(sv);
6826 SvREADONLY_on(tmpref); /* DESTROY() could be naughty */
6828 PUSHSTACKi(PERLSI_DESTROY);
6833 call_sv(MUTABLE_SV(destructor),
6834 G_DISCARD|G_EVAL|G_KEEPERR|G_VOID);
6838 if(SvREFCNT(tmpref) < 2) {
6839 /* tmpref is not kept alive! */
6841 SvRV_set(tmpref, NULL);
6844 SvREFCNT_dec_NN(tmpref);
6847 } while (SvOBJECT(sv) && SvSTASH(sv) != stash);
6850 if (check_refcnt && SvREFCNT(sv)) {
6851 if (PL_in_clean_objs)
6853 "DESTROY created new reference to dead object '%"HEKf"'",
6854 HEKfARG(HvNAME_HEK(stash)));
6855 /* DESTROY gave object new lease on life */
6861 HV * const stash = SvSTASH(sv);
6862 /* Curse before freeing the stash, as freeing the stash could cause
6863 a recursive call into S_curse. */
6864 SvOBJECT_off(sv); /* Curse the object. */
6865 SvSTASH_set(sv,0); /* SvREFCNT_dec may try to read this */
6866 SvREFCNT_dec(stash); /* possibly of changed persuasion */
6872 =for apidoc sv_newref
6874 Increment an SV's reference count. Use the C<SvREFCNT_inc()> wrapper
6881 Perl_sv_newref(pTHX_ SV *const sv)
6883 PERL_UNUSED_CONTEXT;
6892 Decrement an SV's reference count, and if it drops to zero, call
6893 C<sv_clear> to invoke destructors and free up any memory used by
6894 the body; finally, deallocate the SV's head itself.
6895 Normally called via a wrapper macro C<SvREFCNT_dec>.
6901 Perl_sv_free(pTHX_ SV *const sv)
6907 /* Private helper function for SvREFCNT_dec().
6908 * Called with rc set to original SvREFCNT(sv), where rc == 0 or 1 */
6911 Perl_sv_free2(pTHX_ SV *const sv, const U32 rc)
6915 PERL_ARGS_ASSERT_SV_FREE2;
6917 if (LIKELY( rc == 1 )) {
6923 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6924 "Attempt to free temp prematurely: SV 0x%"UVxf
6925 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6929 if (SvIMMORTAL(sv)) {
6930 /* make sure SvREFCNT(sv)==0 happens very seldom */
6931 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6935 if (! SvREFCNT(sv)) /* may have have been resurrected */
6940 /* handle exceptional cases */
6944 if (SvFLAGS(sv) & SVf_BREAK)
6945 /* this SV's refcnt has been artificially decremented to
6946 * trigger cleanup */
6948 if (PL_in_clean_all) /* All is fair */
6950 if (SvIMMORTAL(sv)) {
6951 /* make sure SvREFCNT(sv)==0 happens very seldom */
6952 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6955 if (ckWARN_d(WARN_INTERNAL)) {
6956 #ifdef DEBUG_LEAKING_SCALARS_FORK_DUMP
6957 Perl_dump_sv_child(aTHX_ sv);
6959 #ifdef DEBUG_LEAKING_SCALARS
6962 #ifdef DEBUG_LEAKING_SCALARS_ABORT
6963 if (PL_warnhook == PERL_WARNHOOK_FATAL
6964 || ckDEAD(packWARN(WARN_INTERNAL))) {
6965 /* Don't let Perl_warner cause us to escape our fate: */
6969 /* This may not return: */
6970 Perl_warner(aTHX_ packWARN(WARN_INTERNAL),
6971 "Attempt to free unreferenced scalar: SV 0x%"UVxf
6972 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6975 #ifdef DEBUG_LEAKING_SCALARS_ABORT
6985 Returns the length of the string in the SV. Handles magic and type
6986 coercion and sets the UTF8 flag appropriately. See also C<SvCUR>, which
6987 gives raw access to the xpv_cur slot.
6993 Perl_sv_len(pTHX_ SV *const sv)
7000 (void)SvPV_const(sv, len);
7005 =for apidoc sv_len_utf8
7007 Returns the number of characters in the string in an SV, counting wide
7008 UTF-8 bytes as a single character. Handles magic and type coercion.
7014 * The length is cached in PERL_MAGIC_utf8, in the mg_len field. Also the
7015 * mg_ptr is used, by sv_pos_u2b() and sv_pos_b2u() - see the comments below.
7016 * (Note that the mg_len is not the length of the mg_ptr field.
7017 * This allows the cache to store the character length of the string without
7018 * needing to malloc() extra storage to attach to the mg_ptr.)
7023 Perl_sv_len_utf8(pTHX_ SV *const sv)
7029 return sv_len_utf8_nomg(sv);
7033 Perl_sv_len_utf8_nomg(pTHX_ SV * const sv)
7036 const U8 *s = (U8*)SvPV_nomg_const(sv, len);
7038 PERL_ARGS_ASSERT_SV_LEN_UTF8_NOMG;
7040 if (PL_utf8cache && SvUTF8(sv)) {
7042 MAGIC *mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_utf8) : NULL;
7044 if (mg && (mg->mg_len != -1 || mg->mg_ptr)) {
7045 if (mg->mg_len != -1)
7048 /* We can use the offset cache for a headstart.
7049 The longer value is stored in the first pair. */
7050 STRLEN *cache = (STRLEN *) mg->mg_ptr;
7052 ulen = cache[0] + Perl_utf8_length(aTHX_ s + cache[1],
7056 if (PL_utf8cache < 0) {
7057 const STRLEN real = Perl_utf8_length(aTHX_ s, s + len);
7058 assert_uft8_cache_coherent("sv_len_utf8", ulen, real, sv);
7062 ulen = Perl_utf8_length(aTHX_ s, s + len);
7063 utf8_mg_len_cache_update(sv, &mg, ulen);
7067 return SvUTF8(sv) ? Perl_utf8_length(aTHX_ s, s + len) : len;
7070 /* Walk forwards to find the byte corresponding to the passed in UTF-8
7073 S_sv_pos_u2b_forwards(const U8 *const start, const U8 *const send,
7074 STRLEN *const uoffset_p, bool *const at_end)
7076 const U8 *s = start;
7077 STRLEN uoffset = *uoffset_p;
7079 PERL_ARGS_ASSERT_SV_POS_U2B_FORWARDS;
7081 while (s < send && uoffset) {
7088 else if (s > send) {
7090 /* This is the existing behaviour. Possibly it should be a croak, as
7091 it's actually a bounds error */
7094 *uoffset_p -= uoffset;
7098 /* Given the length of the string in both bytes and UTF-8 characters, decide
7099 whether to walk forwards or backwards to find the byte corresponding to
7100 the passed in UTF-8 offset. */
7102 S_sv_pos_u2b_midway(const U8 *const start, const U8 *send,
7103 STRLEN uoffset, const STRLEN uend)
7105 STRLEN backw = uend - uoffset;
7107 PERL_ARGS_ASSERT_SV_POS_U2B_MIDWAY;
7109 if (uoffset < 2 * backw) {
7110 /* The assumption is that going forwards is twice the speed of going
7111 forward (that's where the 2 * backw comes from).
7112 (The real figure of course depends on the UTF-8 data.) */
7113 const U8 *s = start;
7115 while (s < send && uoffset--)
7125 while (UTF8_IS_CONTINUATION(*send))
7128 return send - start;
7131 /* For the string representation of the given scalar, find the byte
7132 corresponding to the passed in UTF-8 offset. uoffset0 and boffset0
7133 give another position in the string, *before* the sought offset, which
7134 (which is always true, as 0, 0 is a valid pair of positions), which should
7135 help reduce the amount of linear searching.
7136 If *mgp is non-NULL, it should point to the UTF-8 cache magic, which
7137 will be used to reduce the amount of linear searching. The cache will be
7138 created if necessary, and the found value offered to it for update. */
7140 S_sv_pos_u2b_cached(pTHX_ SV *const sv, MAGIC **const mgp, const U8 *const start,
7141 const U8 *const send, STRLEN uoffset,
7142 STRLEN uoffset0, STRLEN boffset0)
7144 STRLEN boffset = 0; /* Actually always set, but let's keep gcc happy. */
7146 bool at_end = FALSE;
7148 PERL_ARGS_ASSERT_SV_POS_U2B_CACHED;
7150 assert (uoffset >= uoffset0);
7155 if (!SvREADONLY(sv) && !SvGMAGICAL(sv) && SvPOK(sv)
7157 && (*mgp || (SvTYPE(sv) >= SVt_PVMG &&
7158 (*mgp = mg_find(sv, PERL_MAGIC_utf8))))) {
7159 if ((*mgp)->mg_ptr) {
7160 STRLEN *cache = (STRLEN *) (*mgp)->mg_ptr;
7161 if (cache[0] == uoffset) {
7162 /* An exact match. */
7165 if (cache[2] == uoffset) {
7166 /* An exact match. */
7170 if (cache[0] < uoffset) {
7171 /* The cache already knows part of the way. */
7172 if (cache[0] > uoffset0) {
7173 /* The cache knows more than the passed in pair */
7174 uoffset0 = cache[0];
7175 boffset0 = cache[1];
7177 if ((*mgp)->mg_len != -1) {
7178 /* And we know the end too. */
7180 + sv_pos_u2b_midway(start + boffset0, send,
7182 (*mgp)->mg_len - uoffset0);
7184 uoffset -= uoffset0;
7186 + sv_pos_u2b_forwards(start + boffset0,
7187 send, &uoffset, &at_end);
7188 uoffset += uoffset0;
7191 else if (cache[2] < uoffset) {
7192 /* We're between the two cache entries. */
7193 if (cache[2] > uoffset0) {
7194 /* and the cache knows more than the passed in pair */
7195 uoffset0 = cache[2];
7196 boffset0 = cache[3];
7200 + sv_pos_u2b_midway(start + boffset0,
7203 cache[0] - uoffset0);
7206 + sv_pos_u2b_midway(start + boffset0,
7209 cache[2] - uoffset0);
7213 else if ((*mgp)->mg_len != -1) {
7214 /* If we can take advantage of a passed in offset, do so. */
7215 /* In fact, offset0 is either 0, or less than offset, so don't
7216 need to worry about the other possibility. */
7218 + sv_pos_u2b_midway(start + boffset0, send,
7220 (*mgp)->mg_len - uoffset0);
7225 if (!found || PL_utf8cache < 0) {
7226 STRLEN real_boffset;
7227 uoffset -= uoffset0;
7228 real_boffset = boffset0 + sv_pos_u2b_forwards(start + boffset0,
7229 send, &uoffset, &at_end);
7230 uoffset += uoffset0;
7232 if (found && PL_utf8cache < 0)
7233 assert_uft8_cache_coherent("sv_pos_u2b_cache", boffset,
7235 boffset = real_boffset;
7238 if (PL_utf8cache && !SvGMAGICAL(sv) && SvPOK(sv)) {
7240 utf8_mg_len_cache_update(sv, mgp, uoffset);
7242 utf8_mg_pos_cache_update(sv, mgp, boffset, uoffset, send - start);
7249 =for apidoc sv_pos_u2b_flags
7251 Converts the offset from a count of UTF-8 chars from
7252 the start of the string, to a count of the equivalent number of bytes; if
7253 lenp is non-zero, it does the same to lenp, but this time starting from
7254 the offset, rather than from the start
7255 of the string. Handles type coercion.
7256 I<flags> is passed to C<SvPV_flags>, and usually should be
7257 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7263 * sv_pos_u2b_flags() uses, like sv_pos_b2u(), the mg_ptr of the potential
7264 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7265 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7270 Perl_sv_pos_u2b_flags(pTHX_ SV *const sv, STRLEN uoffset, STRLEN *const lenp,
7277 PERL_ARGS_ASSERT_SV_POS_U2B_FLAGS;
7279 start = (U8*)SvPV_flags(sv, len, flags);
7281 const U8 * const send = start + len;
7283 boffset = sv_pos_u2b_cached(sv, &mg, start, send, uoffset, 0, 0);
7286 && *lenp /* don't bother doing work for 0, as its bytes equivalent
7287 is 0, and *lenp is already set to that. */) {
7288 /* Convert the relative offset to absolute. */
7289 const STRLEN uoffset2 = uoffset + *lenp;
7290 const STRLEN boffset2
7291 = sv_pos_u2b_cached(sv, &mg, start, send, uoffset2,
7292 uoffset, boffset) - boffset;
7306 =for apidoc sv_pos_u2b
7308 Converts the value pointed to by offsetp from a count of UTF-8 chars from
7309 the start of the string, to a count of the equivalent number of bytes; if
7310 lenp is non-zero, it does the same to lenp, but this time starting from
7311 the offset, rather than from the start of the string. Handles magic and
7314 Use C<sv_pos_u2b_flags> in preference, which correctly handles strings longer
7321 * sv_pos_u2b() uses, like sv_pos_b2u(), the mg_ptr of the potential
7322 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7323 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7327 /* This function is subject to size and sign problems */
7330 Perl_sv_pos_u2b(pTHX_ SV *const sv, I32 *const offsetp, I32 *const lenp)
7332 PERL_ARGS_ASSERT_SV_POS_U2B;
7335 STRLEN ulen = (STRLEN)*lenp;
7336 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, &ulen,
7337 SV_GMAGIC|SV_CONST_RETURN);
7340 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, NULL,
7341 SV_GMAGIC|SV_CONST_RETURN);
7346 S_utf8_mg_len_cache_update(pTHX_ SV *const sv, MAGIC **const mgp,
7349 PERL_ARGS_ASSERT_UTF8_MG_LEN_CACHE_UPDATE;
7350 if (SvREADONLY(sv) || SvGMAGICAL(sv) || !SvPOK(sv))
7353 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7354 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7355 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, &PL_vtbl_utf8, 0, 0);
7359 (*mgp)->mg_len = ulen;
7362 /* Create and update the UTF8 magic offset cache, with the proffered utf8/
7363 byte length pairing. The (byte) length of the total SV is passed in too,
7364 as blen, because for some (more esoteric) SVs, the call to SvPV_const()
7365 may not have updated SvCUR, so we can't rely on reading it directly.
7367 The proffered utf8/byte length pairing isn't used if the cache already has
7368 two pairs, and swapping either for the proffered pair would increase the
7369 RMS of the intervals between known byte offsets.
7371 The cache itself consists of 4 STRLEN values
7372 0: larger UTF-8 offset
7373 1: corresponding byte offset
7374 2: smaller UTF-8 offset
7375 3: corresponding byte offset
7377 Unused cache pairs have the value 0, 0.
7378 Keeping the cache "backwards" means that the invariant of
7379 cache[0] >= cache[2] is maintained even with empty slots, which means that
7380 the code that uses it doesn't need to worry if only 1 entry has actually
7381 been set to non-zero. It also makes the "position beyond the end of the
7382 cache" logic much simpler, as the first slot is always the one to start
7386 S_utf8_mg_pos_cache_update(pTHX_ SV *const sv, MAGIC **const mgp, const STRLEN byte,
7387 const STRLEN utf8, const STRLEN blen)
7391 PERL_ARGS_ASSERT_UTF8_MG_POS_CACHE_UPDATE;
7396 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7397 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7398 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, (MGVTBL*)&PL_vtbl_utf8, 0,
7400 (*mgp)->mg_len = -1;
7404 if (!(cache = (STRLEN *)(*mgp)->mg_ptr)) {
7405 Newxz(cache, PERL_MAGIC_UTF8_CACHESIZE * 2, STRLEN);
7406 (*mgp)->mg_ptr = (char *) cache;
7410 if (PL_utf8cache < 0 && SvPOKp(sv)) {
7411 /* SvPOKp() because, if sv is a reference, then SvPVX() is actually
7412 a pointer. Note that we no longer cache utf8 offsets on refer-
7413 ences, but this check is still a good idea, for robustness. */
7414 const U8 *start = (const U8 *) SvPVX_const(sv);
7415 const STRLEN realutf8 = utf8_length(start, start + byte);
7417 assert_uft8_cache_coherent("utf8_mg_pos_cache_update", utf8, realutf8,
7421 /* Cache is held with the later position first, to simplify the code
7422 that deals with unbounded ends. */
7424 ASSERT_UTF8_CACHE(cache);
7425 if (cache[1] == 0) {
7426 /* Cache is totally empty */
7429 } else if (cache[3] == 0) {
7430 if (byte > cache[1]) {
7431 /* New one is larger, so goes first. */
7432 cache[2] = cache[0];
7433 cache[3] = cache[1];
7441 /* float casts necessary? XXX */
7442 #define THREEWAY_SQUARE(a,b,c,d) \
7443 ((float)((d) - (c))) * ((float)((d) - (c))) \
7444 + ((float)((c) - (b))) * ((float)((c) - (b))) \
7445 + ((float)((b) - (a))) * ((float)((b) - (a)))
7447 /* Cache has 2 slots in use, and we know three potential pairs.
7448 Keep the two that give the lowest RMS distance. Do the
7449 calculation in bytes simply because we always know the byte
7450 length. squareroot has the same ordering as the positive value,
7451 so don't bother with the actual square root. */
7452 if (byte > cache[1]) {
7453 /* New position is after the existing pair of pairs. */
7454 const float keep_earlier
7455 = THREEWAY_SQUARE(0, cache[3], byte, blen);
7456 const float keep_later
7457 = THREEWAY_SQUARE(0, cache[1], byte, blen);
7459 if (keep_later < keep_earlier) {
7460 cache[2] = cache[0];
7461 cache[3] = cache[1];
7467 const float keep_later = THREEWAY_SQUARE(0, byte, cache[1], blen);
7468 float b, c, keep_earlier;
7469 if (byte > cache[3]) {
7470 /* New position is between the existing pair of pairs. */
7474 /* New position is before the existing pair of pairs. */
7478 keep_earlier = THREEWAY_SQUARE(0, b, c, blen);
7479 if (byte > cache[3]) {
7480 if (keep_later < keep_earlier) {
7490 if (! (keep_later < keep_earlier)) {
7491 cache[0] = cache[2];
7492 cache[1] = cache[3];
7499 ASSERT_UTF8_CACHE(cache);
7502 /* We already know all of the way, now we may be able to walk back. The same
7503 assumption is made as in S_sv_pos_u2b_midway(), namely that walking
7504 backward is half the speed of walking forward. */
7506 S_sv_pos_b2u_midway(pTHX_ const U8 *const s, const U8 *const target,
7507 const U8 *end, STRLEN endu)
7509 const STRLEN forw = target - s;
7510 STRLEN backw = end - target;
7512 PERL_ARGS_ASSERT_SV_POS_B2U_MIDWAY;
7514 if (forw < 2 * backw) {
7515 return utf8_length(s, target);
7518 while (end > target) {
7520 while (UTF8_IS_CONTINUATION(*end)) {
7529 =for apidoc sv_pos_b2u_flags
7531 Converts the offset from a count of bytes from the start of the string, to
7532 a count of the equivalent number of UTF-8 chars. Handles type coercion.
7533 I<flags> is passed to C<SvPV_flags>, and usually should be
7534 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7540 * sv_pos_b2u_flags() uses, like sv_pos_u2b_flags(), the mg_ptr of the
7541 * potential PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8
7546 Perl_sv_pos_b2u_flags(pTHX_ SV *const sv, STRLEN const offset, U32 flags)
7549 STRLEN len = 0; /* Actually always set, but let's keep gcc happy. */
7555 PERL_ARGS_ASSERT_SV_POS_B2U_FLAGS;
7557 s = (const U8*)SvPV_flags(sv, blen, flags);
7560 Perl_croak(aTHX_ "panic: sv_pos_b2u: bad byte offset, blen=%"UVuf
7561 ", byte=%"UVuf, (UV)blen, (UV)offset);
7567 && SvTYPE(sv) >= SVt_PVMG
7568 && (mg = mg_find(sv, PERL_MAGIC_utf8)))
7571 STRLEN * const cache = (STRLEN *) mg->mg_ptr;
7572 if (cache[1] == offset) {
7573 /* An exact match. */
7576 if (cache[3] == offset) {
7577 /* An exact match. */
7581 if (cache[1] < offset) {
7582 /* We already know part of the way. */
7583 if (mg->mg_len != -1) {
7584 /* Actually, we know the end too. */
7586 + S_sv_pos_b2u_midway(aTHX_ s + cache[1], send,
7587 s + blen, mg->mg_len - cache[0]);
7589 len = cache[0] + utf8_length(s + cache[1], send);
7592 else if (cache[3] < offset) {
7593 /* We're between the two cached pairs, so we do the calculation
7594 offset by the byte/utf-8 positions for the earlier pair,
7595 then add the utf-8 characters from the string start to
7597 len = S_sv_pos_b2u_midway(aTHX_ s + cache[3], send,
7598 s + cache[1], cache[0] - cache[2])
7602 else { /* cache[3] > offset */
7603 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + cache[3],
7607 ASSERT_UTF8_CACHE(cache);
7609 } else if (mg->mg_len != -1) {
7610 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + blen, mg->mg_len);
7614 if (!found || PL_utf8cache < 0) {
7615 const STRLEN real_len = utf8_length(s, send);
7617 if (found && PL_utf8cache < 0)
7618 assert_uft8_cache_coherent("sv_pos_b2u", len, real_len, sv);
7624 utf8_mg_len_cache_update(sv, &mg, len);
7626 utf8_mg_pos_cache_update(sv, &mg, offset, len, blen);
7633 =for apidoc sv_pos_b2u
7635 Converts the value pointed to by offsetp from a count of bytes from the
7636 start of the string, to a count of the equivalent number of UTF-8 chars.
7637 Handles magic and type coercion.
7639 Use C<sv_pos_b2u_flags> in preference, which correctly handles strings
7646 * sv_pos_b2u() uses, like sv_pos_u2b(), the mg_ptr of the potential
7647 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7652 Perl_sv_pos_b2u(pTHX_ SV *const sv, I32 *const offsetp)
7654 PERL_ARGS_ASSERT_SV_POS_B2U;
7659 *offsetp = (I32)sv_pos_b2u_flags(sv, (STRLEN)*offsetp,
7660 SV_GMAGIC|SV_CONST_RETURN);
7664 S_assert_uft8_cache_coherent(pTHX_ const char *const func, STRLEN from_cache,
7665 STRLEN real, SV *const sv)
7667 PERL_ARGS_ASSERT_ASSERT_UFT8_CACHE_COHERENT;
7669 /* As this is debugging only code, save space by keeping this test here,
7670 rather than inlining it in all the callers. */
7671 if (from_cache == real)
7674 /* Need to turn the assertions off otherwise we may recurse infinitely
7675 while printing error messages. */
7676 SAVEI8(PL_utf8cache);
7678 Perl_croak(aTHX_ "panic: %s cache %"UVuf" real %"UVuf" for %"SVf,
7679 func, (UV) from_cache, (UV) real, SVfARG(sv));
7685 Returns a boolean indicating whether the strings in the two SVs are
7686 identical. Is UTF-8 and 'use bytes' aware, handles get magic, and will
7687 coerce its args to strings if necessary.
7689 =for apidoc sv_eq_flags
7691 Returns a boolean indicating whether the strings in the two SVs are
7692 identical. Is UTF-8 and 'use bytes' aware and coerces its args to strings
7693 if necessary. If the flags include SV_GMAGIC, it handles get-magic, too.
7699 Perl_sv_eq_flags(pTHX_ SV *sv1, SV *sv2, const U32 flags)
7706 SV* svrecode = NULL;
7713 /* if pv1 and pv2 are the same, second SvPV_const call may
7714 * invalidate pv1 (if we are handling magic), so we may need to
7716 if (sv1 == sv2 && flags & SV_GMAGIC
7717 && (SvTHINKFIRST(sv1) || SvGMAGICAL(sv1))) {
7718 pv1 = SvPV_const(sv1, cur1);
7719 sv1 = newSVpvn_flags(pv1, cur1, SVs_TEMP | SvUTF8(sv2));
7721 pv1 = SvPV_flags_const(sv1, cur1, flags);
7729 pv2 = SvPV_flags_const(sv2, cur2, flags);
7731 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7732 /* Differing utf8ness.
7733 * Do not UTF8size the comparands as a side-effect. */
7736 svrecode = newSVpvn(pv2, cur2);
7737 sv_recode_to_utf8(svrecode, PL_encoding);
7738 pv2 = SvPV_const(svrecode, cur2);
7741 svrecode = newSVpvn(pv1, cur1);
7742 sv_recode_to_utf8(svrecode, PL_encoding);
7743 pv1 = SvPV_const(svrecode, cur1);
7745 /* Now both are in UTF-8. */
7747 SvREFCNT_dec_NN(svrecode);
7753 /* sv1 is the UTF-8 one */
7754 return bytes_cmp_utf8((const U8*)pv2, cur2,
7755 (const U8*)pv1, cur1) == 0;
7758 /* sv2 is the UTF-8 one */
7759 return bytes_cmp_utf8((const U8*)pv1, cur1,
7760 (const U8*)pv2, cur2) == 0;
7766 eq = (pv1 == pv2) || memEQ(pv1, pv2, cur1);
7768 SvREFCNT_dec(svrecode);
7776 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7777 string in C<sv1> is less than, equal to, or greater than the string in
7778 C<sv2>. Is UTF-8 and 'use bytes' aware, handles get magic, and will
7779 coerce its args to strings if necessary. See also C<sv_cmp_locale>.
7781 =for apidoc sv_cmp_flags
7783 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7784 string in C<sv1> is less than, equal to, or greater than the string in
7785 C<sv2>. Is UTF-8 and 'use bytes' aware and will coerce its args to strings
7786 if necessary. If the flags include SV_GMAGIC, it handles get magic. See
7787 also C<sv_cmp_locale_flags>.
7793 Perl_sv_cmp(pTHX_ SV *const sv1, SV *const sv2)
7795 return sv_cmp_flags(sv1, sv2, SV_GMAGIC);
7799 Perl_sv_cmp_flags(pTHX_ SV *const sv1, SV *const sv2,
7803 const char *pv1, *pv2;
7805 SV *svrecode = NULL;
7812 pv1 = SvPV_flags_const(sv1, cur1, flags);
7819 pv2 = SvPV_flags_const(sv2, cur2, flags);
7821 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7822 /* Differing utf8ness.
7823 * Do not UTF8size the comparands as a side-effect. */
7826 svrecode = newSVpvn(pv2, cur2);
7827 sv_recode_to_utf8(svrecode, PL_encoding);
7828 pv2 = SvPV_const(svrecode, cur2);
7831 const int retval = -bytes_cmp_utf8((const U8*)pv2, cur2,
7832 (const U8*)pv1, cur1);
7833 return retval ? retval < 0 ? -1 : +1 : 0;
7838 svrecode = newSVpvn(pv1, cur1);
7839 sv_recode_to_utf8(svrecode, PL_encoding);
7840 pv1 = SvPV_const(svrecode, cur1);
7843 const int retval = bytes_cmp_utf8((const U8*)pv1, cur1,
7844 (const U8*)pv2, cur2);
7845 return retval ? retval < 0 ? -1 : +1 : 0;
7851 cmp = cur2 ? -1 : 0;
7855 const I32 retval = memcmp((const void*)pv1, (const void*)pv2, cur1 < cur2 ? cur1 : cur2);
7858 cmp = retval < 0 ? -1 : 1;
7859 } else if (cur1 == cur2) {
7862 cmp = cur1 < cur2 ? -1 : 1;
7866 SvREFCNT_dec(svrecode);
7872 =for apidoc sv_cmp_locale
7874 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
7875 'use bytes' aware, handles get magic, and will coerce its args to strings
7876 if necessary. See also C<sv_cmp>.
7878 =for apidoc sv_cmp_locale_flags
7880 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
7881 'use bytes' aware and will coerce its args to strings if necessary. If the
7882 flags contain SV_GMAGIC, it handles get magic. See also C<sv_cmp_flags>.
7888 Perl_sv_cmp_locale(pTHX_ SV *const sv1, SV *const sv2)
7890 return sv_cmp_locale_flags(sv1, sv2, SV_GMAGIC);
7894 Perl_sv_cmp_locale_flags(pTHX_ SV *const sv1, SV *const sv2,
7897 #ifdef USE_LOCALE_COLLATE
7903 if (PL_collation_standard)
7907 pv1 = sv1 ? sv_collxfrm_flags(sv1, &len1, flags) : (char *) NULL;
7909 pv2 = sv2 ? sv_collxfrm_flags(sv2, &len2, flags) : (char *) NULL;
7911 if (!pv1 || !len1) {
7922 retval = memcmp((void*)pv1, (void*)pv2, len1 < len2 ? len1 : len2);
7925 return retval < 0 ? -1 : 1;
7928 * When the result of collation is equality, that doesn't mean
7929 * that there are no differences -- some locales exclude some
7930 * characters from consideration. So to avoid false equalities,
7931 * we use the raw string as a tiebreaker.
7938 PERL_UNUSED_ARG(flags);
7939 #endif /* USE_LOCALE_COLLATE */
7941 return sv_cmp(sv1, sv2);
7945 #ifdef USE_LOCALE_COLLATE
7948 =for apidoc sv_collxfrm
7950 This calls C<sv_collxfrm_flags> with the SV_GMAGIC flag. See
7951 C<sv_collxfrm_flags>.
7953 =for apidoc sv_collxfrm_flags
7955 Add Collate Transform magic to an SV if it doesn't already have it. If the
7956 flags contain SV_GMAGIC, it handles get-magic.
7958 Any scalar variable may carry PERL_MAGIC_collxfrm magic that contains the
7959 scalar data of the variable, but transformed to such a format that a normal
7960 memory comparison can be used to compare the data according to the locale
7967 Perl_sv_collxfrm_flags(pTHX_ SV *const sv, STRLEN *const nxp, const I32 flags)
7971 PERL_ARGS_ASSERT_SV_COLLXFRM_FLAGS;
7973 mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_collxfrm) : (MAGIC *) NULL;
7974 if (!mg || !mg->mg_ptr || *(U32*)mg->mg_ptr != PL_collation_ix) {
7980 Safefree(mg->mg_ptr);
7981 s = SvPV_flags_const(sv, len, flags);
7982 if ((xf = mem_collxfrm(s, len, &xlen))) {
7984 #ifdef PERL_OLD_COPY_ON_WRITE
7986 sv_force_normal_flags(sv, 0);
7988 mg = sv_magicext(sv, 0, PERL_MAGIC_collxfrm, &PL_vtbl_collxfrm,
8002 if (mg && mg->mg_ptr) {
8004 return mg->mg_ptr + sizeof(PL_collation_ix);
8012 #endif /* USE_LOCALE_COLLATE */
8015 S_sv_gets_append_to_utf8(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8017 SV * const tsv = newSV(0);
8020 sv_gets(tsv, fp, 0);
8021 sv_utf8_upgrade_nomg(tsv);
8022 SvCUR_set(sv,append);
8025 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8029 S_sv_gets_read_record(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8032 const STRLEN recsize = SvUV(SvRV(PL_rs)); /* RsRECORD() guarantees > 0. */
8033 /* Grab the size of the record we're getting */
8034 char *buffer = SvGROW(sv, (STRLEN)(recsize + append + 1)) + append;
8041 /* With a true, record-oriented file on VMS, we need to use read directly
8042 * to ensure that we respect RMS record boundaries. The user is responsible
8043 * for providing a PL_rs value that corresponds to the FAB$W_MRS (maximum
8044 * record size) field. N.B. This is likely to produce invalid results on
8045 * varying-width character data when a record ends mid-character.
8047 fd = PerlIO_fileno(fp);
8049 && PerlLIO_fstat(fd, &st) == 0
8050 && (st.st_fab_rfm == FAB$C_VAR
8051 || st.st_fab_rfm == FAB$C_VFC
8052 || st.st_fab_rfm == FAB$C_FIX)) {
8054 bytesread = PerlLIO_read(fd, buffer, recsize);
8056 else /* in-memory file from PerlIO::Scalar
8057 * or not a record-oriented file
8061 bytesread = PerlIO_read(fp, buffer, recsize);
8063 /* At this point, the logic in sv_get() means that sv will
8064 be treated as utf-8 if the handle is utf8.
8066 if (PerlIO_isutf8(fp) && bytesread > 0) {
8067 char *bend = buffer + bytesread;
8068 char *bufp = buffer;
8069 size_t charcount = 0;
8070 bool charstart = TRUE;
8073 while (charcount < recsize) {
8074 /* count accumulated characters */
8075 while (bufp < bend) {
8077 skip = UTF8SKIP(bufp);
8079 if (bufp + skip > bend) {
8080 /* partial at the end */
8091 if (charcount < recsize) {
8093 STRLEN bufp_offset = bufp - buffer;
8094 SSize_t morebytesread;
8096 /* originally I read enough to fill any incomplete
8097 character and the first byte of the next
8098 character if needed, but if there's many
8099 multi-byte encoded characters we're going to be
8100 making a read call for every character beyond
8101 the original read size.
8103 So instead, read the rest of the character if
8104 any, and enough bytes to match at least the
8105 start bytes for each character we're going to
8109 readsize = recsize - charcount;
8111 readsize = skip - (bend - bufp) + recsize - charcount - 1;
8112 buffer = SvGROW(sv, append + bytesread + readsize + 1) + append;
8113 bend = buffer + bytesread;
8114 morebytesread = PerlIO_read(fp, bend, readsize);
8115 if (morebytesread <= 0) {
8116 /* we're done, if we still have incomplete
8117 characters the check code in sv_gets() will
8120 I'd originally considered doing
8121 PerlIO_ungetc() on all but the lead
8122 character of the incomplete character, but
8123 read() doesn't do that, so I don't.
8128 /* prepare to scan some more */
8129 bytesread += morebytesread;
8130 bend = buffer + bytesread;
8131 bufp = buffer + bufp_offset;
8139 SvCUR_set(sv, bytesread + append);
8140 buffer[bytesread] = '\0';
8141 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8147 Get a line from the filehandle and store it into the SV, optionally
8148 appending to the currently-stored string. If C<append> is not 0, the
8149 line is appended to the SV instead of overwriting it. C<append> should
8150 be set to the byte offset that the appended string should start at
8151 in the SV (typically, C<SvCUR(sv)> is a suitable choice).
8157 Perl_sv_gets(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8167 PERL_ARGS_ASSERT_SV_GETS;
8169 if (SvTHINKFIRST(sv))
8170 sv_force_normal_flags(sv, append ? 0 : SV_COW_DROP_PV);
8171 /* XXX. If you make this PVIV, then copy on write can copy scalars read
8173 However, perlbench says it's slower, because the existing swipe code
8174 is faster than copy on write.
8175 Swings and roundabouts. */
8176 SvUPGRADE(sv, SVt_PV);
8179 /* line is going to be appended to the existing buffer in the sv */
8180 if (PerlIO_isutf8(fp)) {
8182 sv_utf8_upgrade_nomg(sv);
8183 sv_pos_u2b(sv,&append,0);
8185 } else if (SvUTF8(sv)) {
8186 return S_sv_gets_append_to_utf8(aTHX_ sv, fp, append);
8192 /* not appending - "clear" the string by setting SvCUR to 0,
8193 * the pv is still avaiable. */
8196 if (PerlIO_isutf8(fp))
8199 if (IN_PERL_COMPILETIME) {
8200 /* we always read code in line mode */
8204 else if (RsSNARF(PL_rs)) {
8205 /* If it is a regular disk file use size from stat() as estimate
8206 of amount we are going to read -- may result in mallocing
8207 more memory than we really need if the layers below reduce
8208 the size we read (e.g. CRLF or a gzip layer).
8211 if (!PerlLIO_fstat(PerlIO_fileno(fp), &st) && S_ISREG(st.st_mode)) {
8212 const Off_t offset = PerlIO_tell(fp);
8213 if (offset != (Off_t) -1 && st.st_size + append > offset) {
8214 #ifdef PERL_NEW_COPY_ON_WRITE
8215 /* Add an extra byte for the sake of copy-on-write's
8216 * buffer reference count. */
8217 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 2));
8219 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 1));
8226 else if (RsRECORD(PL_rs)) {
8227 return S_sv_gets_read_record(aTHX_ sv, fp, append);
8229 else if (RsPARA(PL_rs)) {
8235 /* Get $/ i.e. PL_rs into same encoding as stream wants */
8236 if (PerlIO_isutf8(fp)) {
8237 rsptr = SvPVutf8(PL_rs, rslen);
8240 if (SvUTF8(PL_rs)) {
8241 if (!sv_utf8_downgrade(PL_rs, TRUE)) {
8242 Perl_croak(aTHX_ "Wide character in $/");
8245 /* extract the raw pointer to the record separator */
8246 rsptr = SvPV_const(PL_rs, rslen);
8250 /* rslast is the last character in the record separator
8251 * note we don't use rslast except when rslen is true, so the
8252 * null assign is a placeholder. */
8253 rslast = rslen ? rsptr[rslen - 1] : '\0';
8255 if (rspara) { /* have to do this both before and after */
8256 do { /* to make sure file boundaries work right */
8259 i = PerlIO_getc(fp);
8263 PerlIO_ungetc(fp,i);
8269 /* See if we know enough about I/O mechanism to cheat it ! */
8271 /* This used to be #ifdef test - it is made run-time test for ease
8272 of abstracting out stdio interface. One call should be cheap
8273 enough here - and may even be a macro allowing compile
8277 if (PerlIO_fast_gets(fp)) {
8279 * We can do buffer based IO operations on this filehandle.
8281 * This means we can bypass a lot of subcalls and process
8282 * the buffer directly, it also means we know the upper bound
8283 * on the amount of data we might read of the current buffer
8284 * into our sv. Knowing this allows us to preallocate the pv
8285 * to be able to hold that maximum, which allows us to simplify
8286 * a lot of logic. */
8289 * We're going to steal some values from the stdio struct
8290 * and put EVERYTHING in the innermost loop into registers.
8292 STDCHAR *ptr; /* pointer into fp's read-ahead buffer */
8293 STRLEN bpx; /* length of the data in the target sv
8294 used to fix pointers after a SvGROW */
8295 I32 shortbuffered; /* If the pv buffer is shorter than the amount
8296 of data left in the read-ahead buffer.
8297 If 0 then the pv buffer can hold the full
8298 amount left, otherwise this is the amount it
8301 #if defined(__VMS) && defined(PERLIO_IS_STDIO)
8302 /* An ungetc()d char is handled separately from the regular
8303 * buffer, so we getc() it back out and stuff it in the buffer.
8305 i = PerlIO_getc(fp);
8306 if (i == EOF) return 0;
8307 *(--((*fp)->_ptr)) = (unsigned char) i;
8311 /* Here is some breathtakingly efficient cheating */
8313 /* When you read the following logic resist the urge to think
8314 * of record separators that are 1 byte long. They are an
8315 * uninteresting special (simple) case.
8317 * Instead think of record separators which are at least 2 bytes
8318 * long, and keep in mind that we need to deal with such
8319 * separators when they cross a read-ahead buffer boundary.
8321 * Also consider that we need to gracefully deal with separators
8322 * that may be longer than a single read ahead buffer.
8324 * Lastly do not forget we want to copy the delimiter as well. We
8325 * are copying all data in the file _up_to_and_including_ the separator
8328 * Now that you have all that in mind here is what is happening below:
8330 * 1. When we first enter the loop we do some memory book keeping to see
8331 * how much free space there is in the target SV. (This sub assumes that
8332 * it is operating on the same SV most of the time via $_ and that it is
8333 * going to be able to reuse the same pv buffer each call.) If there is
8334 * "enough" room then we set "shortbuffered" to how much space there is
8335 * and start reading forward.
8337 * 2. When we scan forward we copy from the read-ahead buffer to the target
8338 * SV's pv buffer. While we go we watch for the end of the read-ahead buffer,
8339 * and the end of the of pv, as well as for the "rslast", which is the last
8340 * char of the separator.
8342 * 3. When scanning forward if we see rslast then we jump backwards in *pv*
8343 * (which has a "complete" record up to the point we saw rslast) and check
8344 * it to see if it matches the separator. If it does we are done. If it doesn't
8345 * we continue on with the scan/copy.
8347 * 4. If we run out of read-ahead buffer (cnt goes to 0) then we have to get
8348 * the IO system to read the next buffer. We do this by doing a getc(), which
8349 * returns a single char read (or EOF), and prefills the buffer, and also
8350 * allows us to find out how full the buffer is. We use this information to
8351 * SvGROW() the sv to the size remaining in the buffer, after which we copy
8352 * the returned single char into the target sv, and then go back into scan
8355 * 5. If we run out of write-buffer then we SvGROW() it by the size of the
8356 * remaining space in the read-buffer.
8358 * Note that this code despite its twisty-turny nature is pretty darn slick.
8359 * It manages single byte separators, multi-byte cross boundary separators,
8360 * and cross-read-buffer separators cleanly and efficiently at the cost
8361 * of potentially greatly overallocating the target SV.
8367 /* get the number of bytes remaining in the read-ahead buffer
8368 * on first call on a given fp this will return 0.*/
8369 cnt = PerlIO_get_cnt(fp);
8371 /* make sure we have the room */
8372 if ((I32)(SvLEN(sv) - append) <= cnt + 1) {
8373 /* Not room for all of it
8374 if we are looking for a separator and room for some
8376 if (rslen && cnt > 80 && (I32)SvLEN(sv) > append) {
8377 /* just process what we have room for */
8378 shortbuffered = cnt - SvLEN(sv) + append + 1;
8379 cnt -= shortbuffered;
8382 /* ensure that the target sv has enough room to hold
8383 * the rest of the read-ahead buffer */
8385 /* remember that cnt can be negative */
8386 SvGROW(sv, (STRLEN)(append + (cnt <= 0 ? 2 : (cnt + 1))));
8390 /* we have enough room to hold the full buffer, lets scream */
8394 /* extract the pointer to sv's string buffer, offset by append as necessary */
8395 bp = (STDCHAR*)SvPVX_const(sv) + append; /* move these two too to registers */
8396 /* extract the point to the read-ahead buffer */
8397 ptr = (STDCHAR*)PerlIO_get_ptr(fp);
8399 /* some trace debug output */
8400 DEBUG_P(PerlIO_printf(Perl_debug_log,
8401 "Screamer: entering, ptr=%"UVuf", cnt=%ld\n",PTR2UV(ptr),(long)cnt));
8402 DEBUG_P(PerlIO_printf(Perl_debug_log,
8403 "Screamer: entering: PerlIO * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"
8405 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8406 PTR2UV(PerlIO_has_base(fp) ? PerlIO_get_base(fp) : 0)));
8410 /* if there is stuff left in the read-ahead buffer */
8412 /* if there is a separator */
8414 /* loop until we hit the end of the read-ahead buffer */
8415 while (cnt > 0) { /* this | eat */
8416 /* scan forward copying and searching for rslast as we go */
8418 if ((*bp++ = *ptr++) == rslast) /* really | dust */
8419 goto thats_all_folks; /* screams | sed :-) */
8423 /* no separator, slurp the full buffer */
8424 Copy(ptr, bp, cnt, char); /* this | eat */
8425 bp += cnt; /* screams | dust */
8426 ptr += cnt; /* louder | sed :-) */
8428 assert (!shortbuffered);
8429 goto cannot_be_shortbuffered;
8433 if (shortbuffered) { /* oh well, must extend */
8434 /* we didnt have enough room to fit the line into the target buffer
8435 * so we must extend the target buffer and keep going */
8436 cnt = shortbuffered;
8438 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8440 /* extned the target sv's buffer so it can hold the full read-ahead buffer */
8441 SvGROW(sv, SvLEN(sv) + append + cnt + 2);
8442 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8446 cannot_be_shortbuffered:
8447 /* we need to refill the read-ahead buffer if possible */
8449 DEBUG_P(PerlIO_printf(Perl_debug_log,
8450 "Screamer: going to getc, ptr=%"UVuf", cnt=%"IVdf"\n",
8451 PTR2UV(ptr),(IV)cnt));
8452 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* deregisterize cnt and ptr */
8454 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8455 "Screamer: pre: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf"\n",
8456 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8457 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8460 call PerlIO_getc() to let it prefill the lookahead buffer
8462 This used to call 'filbuf' in stdio form, but as that behaves like
8463 getc when cnt <= 0 we use PerlIO_getc here to avoid introducing
8464 another abstraction.
8466 Note we have to deal with the char in 'i' if we are not at EOF
8468 i = PerlIO_getc(fp); /* get more characters */
8470 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8471 "Screamer: post: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf"\n",
8472 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8473 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8475 /* find out how much is left in the read-ahead buffer, and rextract its pointer */
8476 cnt = PerlIO_get_cnt(fp);
8477 ptr = (STDCHAR*)PerlIO_get_ptr(fp); /* reregisterize cnt and ptr */
8478 DEBUG_P(PerlIO_printf(Perl_debug_log,
8479 "Screamer: after getc, ptr=%"UVuf", cnt=%"IVdf"\n",
8480 PTR2UV(ptr),(IV)cnt));
8482 if (i == EOF) /* all done for ever? */
8483 goto thats_really_all_folks;
8485 /* make sure we have enough space in the target sv */
8486 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8488 SvGROW(sv, bpx + cnt + 2);
8489 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8491 /* copy of the char we got from getc() */
8492 *bp++ = (STDCHAR)i; /* store character from PerlIO_getc */
8494 /* make sure we deal with the i being the last character of a separator */
8495 if (rslen && (STDCHAR)i == rslast) /* all done for now? */
8496 goto thats_all_folks;
8500 /* check if we have actually found the separator - only really applies
8502 if ((rslen > 1 && (STRLEN)(bp - (STDCHAR*)SvPVX_const(sv)) < rslen) ||
8503 memNE((char*)bp - rslen, rsptr, rslen))
8504 goto screamer; /* go back to the fray */
8505 thats_really_all_folks:
8507 cnt += shortbuffered;
8508 DEBUG_P(PerlIO_printf(Perl_debug_log,
8509 "Screamer: quitting, ptr=%"UVuf", cnt=%"IVdf"\n",PTR2UV(ptr),(IV)cnt));
8510 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* put these back or we're in trouble */
8511 DEBUG_P(PerlIO_printf(Perl_debug_log,
8512 "Screamer: end: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf
8514 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8515 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8517 SvCUR_set(sv, bp - (STDCHAR*)SvPVX_const(sv)); /* set length */
8518 DEBUG_P(PerlIO_printf(Perl_debug_log,
8519 "Screamer: done, len=%ld, string=|%.*s|\n",
8520 (long)SvCUR(sv),(int)SvCUR(sv),SvPVX_const(sv)));
8524 /*The big, slow, and stupid way. */
8525 #ifdef USE_HEAP_INSTEAD_OF_STACK /* Even slower way. */
8526 STDCHAR *buf = NULL;
8527 Newx(buf, 8192, STDCHAR);
8535 const STDCHAR * const bpe = buf + sizeof(buf);
8537 while ((i = PerlIO_getc(fp)) != EOF && (*bp++ = (STDCHAR)i) != rslast && bp < bpe)
8538 ; /* keep reading */
8542 cnt = PerlIO_read(fp,(char*)buf, sizeof(buf));
8543 /* Accommodate broken VAXC compiler, which applies U8 cast to
8544 * both args of ?: operator, causing EOF to change into 255
8547 i = (U8)buf[cnt - 1];
8553 cnt = 0; /* we do need to re-set the sv even when cnt <= 0 */
8555 sv_catpvn_nomg(sv, (char *) buf, cnt);
8557 sv_setpvn(sv, (char *) buf, cnt); /* "nomg" is implied */
8559 if (i != EOF && /* joy */
8561 SvCUR(sv) < rslen ||
8562 memNE(SvPVX_const(sv) + SvCUR(sv) - rslen, rsptr, rslen)))
8566 * If we're reading from a TTY and we get a short read,
8567 * indicating that the user hit his EOF character, we need
8568 * to notice it now, because if we try to read from the TTY
8569 * again, the EOF condition will disappear.
8571 * The comparison of cnt to sizeof(buf) is an optimization
8572 * that prevents unnecessary calls to feof().
8576 if (!(cnt < (I32)sizeof(buf) && PerlIO_eof(fp)))
8580 #ifdef USE_HEAP_INSTEAD_OF_STACK
8585 if (rspara) { /* have to do this both before and after */
8586 while (i != EOF) { /* to make sure file boundaries work right */
8587 i = PerlIO_getc(fp);
8589 PerlIO_ungetc(fp,i);
8595 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8601 Auto-increment of the value in the SV, doing string to numeric conversion
8602 if necessary. Handles 'get' magic and operator overloading.
8608 Perl_sv_inc(pTHX_ SV *const sv)
8617 =for apidoc sv_inc_nomg
8619 Auto-increment of the value in the SV, doing string to numeric conversion
8620 if necessary. Handles operator overloading. Skips handling 'get' magic.
8626 Perl_sv_inc_nomg(pTHX_ SV *const sv)
8633 if (SvTHINKFIRST(sv)) {
8634 if (SvREADONLY(sv)) {
8635 Perl_croak_no_modify();
8639 if (SvAMAGIC(sv) && AMG_CALLunary(sv, inc_amg))
8641 i = PTR2IV(SvRV(sv));
8645 else sv_force_normal_flags(sv, 0);
8647 flags = SvFLAGS(sv);
8648 if ((flags & (SVp_NOK|SVp_IOK)) == SVp_NOK) {
8649 /* It's (privately or publicly) a float, but not tested as an
8650 integer, so test it to see. */
8652 flags = SvFLAGS(sv);
8654 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8655 /* It's publicly an integer, or privately an integer-not-float */
8656 #ifdef PERL_PRESERVE_IVUV
8660 if (SvUVX(sv) == UV_MAX)
8661 sv_setnv(sv, UV_MAX_P1);
8663 (void)SvIOK_only_UV(sv);
8664 SvUV_set(sv, SvUVX(sv) + 1);
8666 if (SvIVX(sv) == IV_MAX)
8667 sv_setuv(sv, (UV)IV_MAX + 1);
8669 (void)SvIOK_only(sv);
8670 SvIV_set(sv, SvIVX(sv) + 1);
8675 if (flags & SVp_NOK) {
8676 const NV was = SvNVX(sv);
8677 if (LIKELY(!Perl_isinfnan(was)) &&
8678 NV_OVERFLOWS_INTEGERS_AT &&
8679 was >= NV_OVERFLOWS_INTEGERS_AT) {
8680 /* diag_listed_as: Lost precision when %s %f by 1 */
8681 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8682 "Lost precision when incrementing %" NVff " by 1",
8685 (void)SvNOK_only(sv);
8686 SvNV_set(sv, was + 1.0);
8690 if (!(flags & SVp_POK) || !*SvPVX_const(sv)) {
8691 if ((flags & SVTYPEMASK) < SVt_PVIV)
8692 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV ? SVt_PVIV : SVt_IV));
8693 (void)SvIOK_only(sv);
8698 while (isALPHA(*d)) d++;
8699 while (isDIGIT(*d)) d++;
8700 if (d < SvEND(sv)) {
8701 const int numtype = grok_number_flags(SvPVX_const(sv), SvCUR(sv), NULL, PERL_SCAN_TRAILING);
8702 #ifdef PERL_PRESERVE_IVUV
8703 /* Got to punt this as an integer if needs be, but we don't issue
8704 warnings. Probably ought to make the sv_iv_please() that does
8705 the conversion if possible, and silently. */
8706 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
8707 /* Need to try really hard to see if it's an integer.
8708 9.22337203685478e+18 is an integer.
8709 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
8710 so $a="9.22337203685478e+18"; $a+0; $a++
8711 needs to be the same as $a="9.22337203685478e+18"; $a++
8718 /* sv_2iv *should* have made this an NV */
8719 if (flags & SVp_NOK) {
8720 (void)SvNOK_only(sv);
8721 SvNV_set(sv, SvNVX(sv) + 1.0);
8724 /* I don't think we can get here. Maybe I should assert this
8725 And if we do get here I suspect that sv_setnv will croak. NWC
8727 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_inc punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"NVgf"\n",
8728 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
8730 #endif /* PERL_PRESERVE_IVUV */
8731 if (!numtype && ckWARN(WARN_NUMERIC))
8732 not_incrementable(sv);
8733 sv_setnv(sv,Atof(SvPVX_const(sv)) + 1.0);
8737 while (d >= SvPVX_const(sv)) {
8745 /* MKS: The original code here died if letters weren't consecutive.
8746 * at least it didn't have to worry about non-C locales. The
8747 * new code assumes that ('z'-'a')==('Z'-'A'), letters are
8748 * arranged in order (although not consecutively) and that only
8749 * [A-Za-z] are accepted by isALPHA in the C locale.
8751 if (isALPHA_FOLD_NE(*d, 'z')) {
8752 do { ++*d; } while (!isALPHA(*d));
8755 *(d--) -= 'z' - 'a';
8760 *(d--) -= 'z' - 'a' + 1;
8764 /* oh,oh, the number grew */
8765 SvGROW(sv, SvCUR(sv) + 2);
8766 SvCUR_set(sv, SvCUR(sv) + 1);
8767 for (d = SvPVX(sv) + SvCUR(sv); d > SvPVX_const(sv); d--)
8778 Auto-decrement of the value in the SV, doing string to numeric conversion
8779 if necessary. Handles 'get' magic and operator overloading.
8785 Perl_sv_dec(pTHX_ SV *const sv)
8794 =for apidoc sv_dec_nomg
8796 Auto-decrement of the value in the SV, doing string to numeric conversion
8797 if necessary. Handles operator overloading. Skips handling 'get' magic.
8803 Perl_sv_dec_nomg(pTHX_ SV *const sv)
8809 if (SvTHINKFIRST(sv)) {
8810 if (SvREADONLY(sv)) {
8811 Perl_croak_no_modify();
8815 if (SvAMAGIC(sv) && AMG_CALLunary(sv, dec_amg))
8817 i = PTR2IV(SvRV(sv));
8821 else sv_force_normal_flags(sv, 0);
8823 /* Unlike sv_inc we don't have to worry about string-never-numbers
8824 and keeping them magic. But we mustn't warn on punting */
8825 flags = SvFLAGS(sv);
8826 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8827 /* It's publicly an integer, or privately an integer-not-float */
8828 #ifdef PERL_PRESERVE_IVUV
8832 if (SvUVX(sv) == 0) {
8833 (void)SvIOK_only(sv);
8837 (void)SvIOK_only_UV(sv);
8838 SvUV_set(sv, SvUVX(sv) - 1);
8841 if (SvIVX(sv) == IV_MIN) {
8842 sv_setnv(sv, (NV)IV_MIN);
8846 (void)SvIOK_only(sv);
8847 SvIV_set(sv, SvIVX(sv) - 1);
8852 if (flags & SVp_NOK) {
8855 const NV was = SvNVX(sv);
8856 if (LIKELY(!Perl_isinfnan(was)) &&
8857 NV_OVERFLOWS_INTEGERS_AT &&
8858 was <= -NV_OVERFLOWS_INTEGERS_AT) {
8859 /* diag_listed_as: Lost precision when %s %f by 1 */
8860 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8861 "Lost precision when decrementing %" NVff " by 1",
8864 (void)SvNOK_only(sv);
8865 SvNV_set(sv, was - 1.0);
8869 if (!(flags & SVp_POK)) {
8870 if ((flags & SVTYPEMASK) < SVt_PVIV)
8871 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV) ? SVt_PVIV : SVt_IV);
8873 (void)SvIOK_only(sv);
8876 #ifdef PERL_PRESERVE_IVUV
8878 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), NULL);
8879 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
8880 /* Need to try really hard to see if it's an integer.
8881 9.22337203685478e+18 is an integer.
8882 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
8883 so $a="9.22337203685478e+18"; $a+0; $a--
8884 needs to be the same as $a="9.22337203685478e+18"; $a--
8891 /* sv_2iv *should* have made this an NV */
8892 if (flags & SVp_NOK) {
8893 (void)SvNOK_only(sv);
8894 SvNV_set(sv, SvNVX(sv) - 1.0);
8897 /* I don't think we can get here. Maybe I should assert this
8898 And if we do get here I suspect that sv_setnv will croak. NWC
8900 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_dec punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"NVgf"\n",
8901 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
8904 #endif /* PERL_PRESERVE_IVUV */
8905 sv_setnv(sv,Atof(SvPVX_const(sv)) - 1.0); /* punt */
8908 /* this define is used to eliminate a chunk of duplicated but shared logic
8909 * it has the suffix __SV_C to signal that it isnt API, and isnt meant to be
8910 * used anywhere but here - yves
8912 #define PUSH_EXTEND_MORTAL__SV_C(AnSv) \
8915 PL_tmps_stack[++PL_tmps_ix] = (AnSv); \
8919 =for apidoc sv_mortalcopy
8921 Creates a new SV which is a copy of the original SV (using C<sv_setsv>).
8922 The new SV is marked as mortal. It will be destroyed "soon", either by an
8923 explicit call to FREETMPS, or by an implicit call at places such as
8924 statement boundaries. See also C<sv_newmortal> and C<sv_2mortal>.
8929 /* Make a string that will exist for the duration of the expression
8930 * evaluation. Actually, it may have to last longer than that, but
8931 * hopefully we won't free it until it has been assigned to a
8932 * permanent location. */
8935 Perl_sv_mortalcopy_flags(pTHX_ SV *const oldstr, U32 flags)
8939 if (flags & SV_GMAGIC)
8940 SvGETMAGIC(oldstr); /* before new_SV, in case it dies */
8942 sv_setsv_flags(sv,oldstr,flags & ~SV_GMAGIC);
8943 PUSH_EXTEND_MORTAL__SV_C(sv);
8949 =for apidoc sv_newmortal
8951 Creates a new null SV which is mortal. The reference count of the SV is
8952 set to 1. It will be destroyed "soon", either by an explicit call to
8953 FREETMPS, or by an implicit call at places such as statement boundaries.
8954 See also C<sv_mortalcopy> and C<sv_2mortal>.
8960 Perl_sv_newmortal(pTHX)
8965 SvFLAGS(sv) = SVs_TEMP;
8966 PUSH_EXTEND_MORTAL__SV_C(sv);
8972 =for apidoc newSVpvn_flags
8974 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
8975 characters) into it. The reference count for the
8976 SV is set to 1. Note that if C<len> is zero, Perl will create a zero length
8977 string. You are responsible for ensuring that the source string is at least
8978 C<len> bytes long. If the C<s> argument is NULL the new SV will be undefined.
8979 Currently the only flag bits accepted are C<SVf_UTF8> and C<SVs_TEMP>.
8980 If C<SVs_TEMP> is set, then C<sv_2mortal()> is called on the result before
8981 returning. If C<SVf_UTF8> is set, C<s>
8982 is considered to be in UTF-8 and the
8983 C<SVf_UTF8> flag will be set on the new SV.
8984 C<newSVpvn_utf8()> is a convenience wrapper for this function, defined as
8986 #define newSVpvn_utf8(s, len, u) \
8987 newSVpvn_flags((s), (len), (u) ? SVf_UTF8 : 0)
8993 Perl_newSVpvn_flags(pTHX_ const char *const s, const STRLEN len, const U32 flags)
8997 /* All the flags we don't support must be zero.
8998 And we're new code so I'm going to assert this from the start. */
8999 assert(!(flags & ~(SVf_UTF8|SVs_TEMP)));
9001 sv_setpvn(sv,s,len);
9003 /* This code used to do a sv_2mortal(), however we now unroll the call to
9004 * sv_2mortal() and do what it does ourselves here. Since we have asserted
9005 * that flags can only have the SVf_UTF8 and/or SVs_TEMP flags set above we
9006 * can use it to enable the sv flags directly (bypassing SvTEMP_on), which
9007 * in turn means we dont need to mask out the SVf_UTF8 flag below, which
9008 * means that we eliminate quite a few steps than it looks - Yves
9009 * (explaining patch by gfx) */
9011 SvFLAGS(sv) |= flags;
9013 if(flags & SVs_TEMP){
9014 PUSH_EXTEND_MORTAL__SV_C(sv);
9021 =for apidoc sv_2mortal
9023 Marks an existing SV as mortal. The SV will be destroyed "soon", either
9024 by an explicit call to FREETMPS, or by an implicit call at places such as
9025 statement boundaries. SvTEMP() is turned on which means that the SV's
9026 string buffer can be "stolen" if this SV is copied. See also C<sv_newmortal>
9027 and C<sv_mortalcopy>.
9033 Perl_sv_2mortal(pTHX_ SV *const sv)
9040 PUSH_EXTEND_MORTAL__SV_C(sv);
9048 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9049 characters) into it. The reference count for the
9050 SV is set to 1. If C<len> is zero, Perl will compute the length using
9051 strlen(), (which means if you use this option, that C<s> can't have embedded
9052 C<NUL> characters and has to have a terminating C<NUL> byte).
9054 For efficiency, consider using C<newSVpvn> instead.
9060 Perl_newSVpv(pTHX_ const char *const s, const STRLEN len)
9065 sv_setpvn(sv, s, len || s == NULL ? len : strlen(s));
9070 =for apidoc newSVpvn
9072 Creates a new SV and copies a string into it, which may contain C<NUL> characters
9073 (C<\0>) and other binary data. The reference count for the SV is set to 1.
9074 Note that if C<len> is zero, Perl will create a zero length (Perl) string. You
9075 are responsible for ensuring that the source buffer is at least
9076 C<len> bytes long. If the C<buffer> argument is NULL the new SV will be
9083 Perl_newSVpvn(pTHX_ const char *const buffer, const STRLEN len)
9087 sv_setpvn(sv,buffer,len);
9092 =for apidoc newSVhek
9094 Creates a new SV from the hash key structure. It will generate scalars that
9095 point to the shared string table where possible. Returns a new (undefined)
9096 SV if the hek is NULL.
9102 Perl_newSVhek(pTHX_ const HEK *const hek)
9111 if (HEK_LEN(hek) == HEf_SVKEY) {
9112 return newSVsv(*(SV**)HEK_KEY(hek));
9114 const int flags = HEK_FLAGS(hek);
9115 if (flags & HVhek_WASUTF8) {
9117 Andreas would like keys he put in as utf8 to come back as utf8
9119 STRLEN utf8_len = HEK_LEN(hek);
9120 SV * const sv = newSV_type(SVt_PV);
9121 char *as_utf8 = (char *)bytes_to_utf8 ((U8*)HEK_KEY(hek), &utf8_len);
9122 /* bytes_to_utf8() allocates a new string, which we can repurpose: */
9123 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
9126 } else if (flags & HVhek_UNSHARED) {
9127 /* A hash that isn't using shared hash keys has to have
9128 the flag in every key so that we know not to try to call
9129 share_hek_hek on it. */
9131 SV * const sv = newSVpvn (HEK_KEY(hek), HEK_LEN(hek));
9136 /* This will be overwhelminly the most common case. */
9138 /* Inline most of newSVpvn_share(), because share_hek_hek() is far
9139 more efficient than sharepvn(). */
9143 sv_upgrade(sv, SVt_PV);
9144 SvPV_set(sv, (char *)HEK_KEY(share_hek_hek(hek)));
9145 SvCUR_set(sv, HEK_LEN(hek));
9157 =for apidoc newSVpvn_share
9159 Creates a new SV with its SvPVX_const pointing to a shared string in the string
9160 table. If the string does not already exist in the table, it is
9161 created first. Turns on the SvIsCOW flag (or READONLY
9162 and FAKE in 5.16 and earlier). If the C<hash> parameter
9163 is non-zero, that value is used; otherwise the hash is computed.
9164 The string's hash can later be retrieved from the SV
9165 with the C<SvSHARED_HASH()> macro. The idea here is
9166 that as the string table is used for shared hash keys these strings will have
9167 SvPVX_const == HeKEY and hash lookup will avoid string compare.
9173 Perl_newSVpvn_share(pTHX_ const char *src, I32 len, U32 hash)
9177 bool is_utf8 = FALSE;
9178 const char *const orig_src = src;
9181 STRLEN tmplen = -len;
9183 /* See the note in hv.c:hv_fetch() --jhi */
9184 src = (char*)bytes_from_utf8((const U8*)src, &tmplen, &is_utf8);
9188 PERL_HASH(hash, src, len);
9190 /* The logic for this is inlined in S_mro_get_linear_isa_dfs(), so if it
9191 changes here, update it there too. */
9192 sv_upgrade(sv, SVt_PV);
9193 SvPV_set(sv, sharepvn(src, is_utf8?-len:len, hash));
9200 if (src != orig_src)
9206 =for apidoc newSVpv_share
9208 Like C<newSVpvn_share>, but takes a C<NUL>-terminated string instead of a
9215 Perl_newSVpv_share(pTHX_ const char *src, U32 hash)
9217 return newSVpvn_share(src, strlen(src), hash);
9220 #if defined(PERL_IMPLICIT_CONTEXT)
9222 /* pTHX_ magic can't cope with varargs, so this is a no-context
9223 * version of the main function, (which may itself be aliased to us).
9224 * Don't access this version directly.
9228 Perl_newSVpvf_nocontext(const char *const pat, ...)
9234 PERL_ARGS_ASSERT_NEWSVPVF_NOCONTEXT;
9236 va_start(args, pat);
9237 sv = vnewSVpvf(pat, &args);
9244 =for apidoc newSVpvf
9246 Creates a new SV and initializes it with the string formatted like
9253 Perl_newSVpvf(pTHX_ const char *const pat, ...)
9258 PERL_ARGS_ASSERT_NEWSVPVF;
9260 va_start(args, pat);
9261 sv = vnewSVpvf(pat, &args);
9266 /* backend for newSVpvf() and newSVpvf_nocontext() */
9269 Perl_vnewSVpvf(pTHX_ const char *const pat, va_list *const args)
9273 PERL_ARGS_ASSERT_VNEWSVPVF;
9276 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
9283 Creates a new SV and copies a floating point value into it.
9284 The reference count for the SV is set to 1.
9290 Perl_newSVnv(pTHX_ const NV n)
9302 Creates a new SV and copies an integer into it. The reference count for the
9309 Perl_newSViv(pTHX_ const IV i)
9321 Creates a new SV and copies an unsigned integer into it.
9322 The reference count for the SV is set to 1.
9328 Perl_newSVuv(pTHX_ const UV u)
9338 =for apidoc newSV_type
9340 Creates a new SV, of the type specified. The reference count for the new SV
9347 Perl_newSV_type(pTHX_ const svtype type)
9352 sv_upgrade(sv, type);
9357 =for apidoc newRV_noinc
9359 Creates an RV wrapper for an SV. The reference count for the original
9360 SV is B<not> incremented.
9366 Perl_newRV_noinc(pTHX_ SV *const tmpRef)
9368 SV *sv = newSV_type(SVt_IV);
9370 PERL_ARGS_ASSERT_NEWRV_NOINC;
9373 SvRV_set(sv, tmpRef);
9378 /* newRV_inc is the official function name to use now.
9379 * newRV_inc is in fact #defined to newRV in sv.h
9383 Perl_newRV(pTHX_ SV *const sv)
9385 PERL_ARGS_ASSERT_NEWRV;
9387 return newRV_noinc(SvREFCNT_inc_simple_NN(sv));
9393 Creates a new SV which is an exact duplicate of the original SV.
9400 Perl_newSVsv(pTHX_ SV *const old)
9406 if (SvTYPE(old) == (svtype)SVTYPEMASK) {
9407 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), "semi-panic: attempt to dup freed string");
9410 /* Do this here, otherwise we leak the new SV if this croaks. */
9413 /* SV_NOSTEAL prevents TEMP buffers being, well, stolen, and saves games
9414 with SvTEMP_off and SvTEMP_on round a call to sv_setsv. */
9415 sv_setsv_flags(sv, old, SV_NOSTEAL);
9420 =for apidoc sv_reset
9422 Underlying implementation for the C<reset> Perl function.
9423 Note that the perl-level function is vaguely deprecated.
9429 Perl_sv_reset(pTHX_ const char *s, HV *const stash)
9431 PERL_ARGS_ASSERT_SV_RESET;
9433 sv_resetpvn(*s ? s : NULL, strlen(s), stash);
9437 Perl_sv_resetpvn(pTHX_ const char *s, STRLEN len, HV * const stash)
9439 char todo[PERL_UCHAR_MAX+1];
9442 if (!stash || SvTYPE(stash) != SVt_PVHV)
9445 if (!s) { /* reset ?? searches */
9446 MAGIC * const mg = mg_find((const SV *)stash, PERL_MAGIC_symtab);
9448 const U32 count = mg->mg_len / sizeof(PMOP**);
9449 PMOP **pmp = (PMOP**) mg->mg_ptr;
9450 PMOP *const *const end = pmp + count;
9454 SvREADONLY_off(PL_regex_pad[(*pmp)->op_pmoffset]);
9456 (*pmp)->op_pmflags &= ~PMf_USED;
9464 /* reset variables */
9466 if (!HvARRAY(stash))
9469 Zero(todo, 256, char);
9473 I32 i = (unsigned char)*s;
9477 max = (unsigned char)*s++;
9478 for ( ; i <= max; i++) {
9481 for (i = 0; i <= (I32) HvMAX(stash); i++) {
9483 for (entry = HvARRAY(stash)[i];
9485 entry = HeNEXT(entry))
9490 if (!todo[(U8)*HeKEY(entry)])
9492 gv = MUTABLE_GV(HeVAL(entry));
9494 if (sv && !SvREADONLY(sv)) {
9495 SV_CHECK_THINKFIRST_COW_DROP(sv);
9496 if (!isGV(sv)) SvOK_off(sv);
9501 if (GvHV(gv) && !HvNAME_get(GvHV(gv))) {
9512 Using various gambits, try to get an IO from an SV: the IO slot if its a
9513 GV; or the recursive result if we're an RV; or the IO slot of the symbol
9514 named after the PV if we're a string.
9516 'Get' magic is ignored on the sv passed in, but will be called on
9517 C<SvRV(sv)> if sv is an RV.
9523 Perl_sv_2io(pTHX_ SV *const sv)
9528 PERL_ARGS_ASSERT_SV_2IO;
9530 switch (SvTYPE(sv)) {
9532 io = MUTABLE_IO(sv);
9536 if (isGV_with_GP(sv)) {
9537 gv = MUTABLE_GV(sv);
9540 Perl_croak(aTHX_ "Bad filehandle: %"HEKf,
9541 HEKfARG(GvNAME_HEK(gv)));
9547 Perl_croak(aTHX_ PL_no_usym, "filehandle");
9549 SvGETMAGIC(SvRV(sv));
9550 return sv_2io(SvRV(sv));
9552 gv = gv_fetchsv_nomg(sv, 0, SVt_PVIO);
9559 if (SvGMAGICAL(sv)) {
9560 newsv = sv_newmortal();
9561 sv_setsv_nomg(newsv, sv);
9563 Perl_croak(aTHX_ "Bad filehandle: %"SVf, SVfARG(newsv));
9573 Using various gambits, try to get a CV from an SV; in addition, try if
9574 possible to set C<*st> and C<*gvp> to the stash and GV associated with it.
9575 The flags in C<lref> are passed to gv_fetchsv.
9581 Perl_sv_2cv(pTHX_ SV *sv, HV **const st, GV **const gvp, const I32 lref)
9586 PERL_ARGS_ASSERT_SV_2CV;
9593 switch (SvTYPE(sv)) {
9597 return MUTABLE_CV(sv);
9607 sv = amagic_deref_call(sv, to_cv_amg);
9610 if (SvTYPE(sv) == SVt_PVCV) {
9611 cv = MUTABLE_CV(sv);
9616 else if(SvGETMAGIC(sv), isGV_with_GP(sv))
9617 gv = MUTABLE_GV(sv);
9619 Perl_croak(aTHX_ "Not a subroutine reference");
9621 else if (isGV_with_GP(sv)) {
9622 gv = MUTABLE_GV(sv);
9625 gv = gv_fetchsv_nomg(sv, lref, SVt_PVCV);
9632 /* Some flags to gv_fetchsv mean don't really create the GV */
9633 if (!isGV_with_GP(gv)) {
9638 if (lref & ~GV_ADDMG && !GvCVu(gv)) {
9639 /* XXX this is probably not what they think they're getting.
9640 * It has the same effect as "sub name;", i.e. just a forward
9651 Returns true if the SV has a true value by Perl's rules.
9652 Use the C<SvTRUE> macro instead, which may call C<sv_true()> or may
9653 instead use an in-line version.
9659 Perl_sv_true(pTHX_ SV *const sv)
9664 const XPV* const tXpv = (XPV*)SvANY(sv);
9666 (tXpv->xpv_cur > 1 ||
9667 (tXpv->xpv_cur && *sv->sv_u.svu_pv != '0')))
9674 return SvIVX(sv) != 0;
9677 return SvNVX(sv) != 0.0;
9679 return sv_2bool(sv);
9685 =for apidoc sv_pvn_force
9687 Get a sensible string out of the SV somehow.
9688 A private implementation of the C<SvPV_force> macro for compilers which
9689 can't cope with complex macro expressions. Always use the macro instead.
9691 =for apidoc sv_pvn_force_flags
9693 Get a sensible string out of the SV somehow.
9694 If C<flags> has C<SV_GMAGIC> bit set, will C<mg_get> on C<sv> if
9695 appropriate, else not. C<sv_pvn_force> and C<sv_pvn_force_nomg> are
9696 implemented in terms of this function.
9697 You normally want to use the various wrapper macros instead: see
9698 C<SvPV_force> and C<SvPV_force_nomg>
9704 Perl_sv_pvn_force_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
9706 PERL_ARGS_ASSERT_SV_PVN_FORCE_FLAGS;
9708 if (flags & SV_GMAGIC) SvGETMAGIC(sv);
9709 if (SvTHINKFIRST(sv) && (!SvROK(sv) || SvREADONLY(sv)))
9710 sv_force_normal_flags(sv, 0);
9720 if (SvTYPE(sv) > SVt_PVLV
9721 || isGV_with_GP(sv))
9722 /* diag_listed_as: Can't coerce %s to %s in %s */
9723 Perl_croak(aTHX_ "Can't coerce %s to string in %s", sv_reftype(sv,0),
9725 s = sv_2pv_flags(sv, &len, flags &~ SV_GMAGIC);
9732 if (SvTYPE(sv) < SVt_PV ||
9733 s != SvPVX_const(sv)) { /* Almost, but not quite, sv_setpvn() */
9736 SvUPGRADE(sv, SVt_PV); /* Never FALSE */
9737 SvGROW(sv, len + 1);
9738 Move(s,SvPVX(sv),len,char);
9740 SvPVX(sv)[len] = '\0';
9743 SvPOK_on(sv); /* validate pointer */
9745 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
9746 PTR2UV(sv),SvPVX_const(sv)));
9749 (void)SvPOK_only_UTF8(sv);
9750 return SvPVX_mutable(sv);
9754 =for apidoc sv_pvbyten_force
9756 The backend for the C<SvPVbytex_force> macro. Always use the macro
9763 Perl_sv_pvbyten_force(pTHX_ SV *const sv, STRLEN *const lp)
9765 PERL_ARGS_ASSERT_SV_PVBYTEN_FORCE;
9767 sv_pvn_force(sv,lp);
9768 sv_utf8_downgrade(sv,0);
9774 =for apidoc sv_pvutf8n_force
9776 The backend for the C<SvPVutf8x_force> macro. Always use the macro
9783 Perl_sv_pvutf8n_force(pTHX_ SV *const sv, STRLEN *const lp)
9785 PERL_ARGS_ASSERT_SV_PVUTF8N_FORCE;
9788 sv_utf8_upgrade_nomg(sv);
9794 =for apidoc sv_reftype
9796 Returns a string describing what the SV is a reference to.
9802 Perl_sv_reftype(pTHX_ const SV *const sv, const int ob)
9804 PERL_ARGS_ASSERT_SV_REFTYPE;
9805 if (ob && SvOBJECT(sv)) {
9806 return SvPV_nolen_const(sv_ref(NULL, sv, ob));
9809 /* WARNING - There is code, for instance in mg.c, that assumes that
9810 * the only reason that sv_reftype(sv,0) would return a string starting
9811 * with 'L' or 'S' is that it is a LVALUE or a SCALAR.
9812 * Yes this a dodgy way to do type checking, but it saves practically reimplementing
9813 * this routine inside other subs, and it saves time.
9814 * Do not change this assumption without searching for "dodgy type check" in
9817 switch (SvTYPE(sv)) {
9832 case SVt_PVLV: return (char *) (SvROK(sv) ? "REF"
9833 /* tied lvalues should appear to be
9834 * scalars for backwards compatibility */
9835 : (isALPHA_FOLD_EQ(LvTYPE(sv), 't'))
9836 ? "SCALAR" : "LVALUE");
9837 case SVt_PVAV: return "ARRAY";
9838 case SVt_PVHV: return "HASH";
9839 case SVt_PVCV: return "CODE";
9840 case SVt_PVGV: return (char *) (isGV_with_GP(sv)
9841 ? "GLOB" : "SCALAR");
9842 case SVt_PVFM: return "FORMAT";
9843 case SVt_PVIO: return "IO";
9844 case SVt_INVLIST: return "INVLIST";
9845 case SVt_REGEXP: return "REGEXP";
9846 default: return "UNKNOWN";
9854 Returns a SV describing what the SV passed in is a reference to.
9860 Perl_sv_ref(pTHX_ SV *dst, const SV *const sv, const int ob)
9862 PERL_ARGS_ASSERT_SV_REF;
9865 dst = sv_newmortal();
9867 if (ob && SvOBJECT(sv)) {
9868 HvNAME_get(SvSTASH(sv))
9869 ? sv_sethek(dst, HvNAME_HEK(SvSTASH(sv)))
9870 : sv_setpvn(dst, "__ANON__", 8);
9873 const char * reftype = sv_reftype(sv, 0);
9874 sv_setpv(dst, reftype);
9880 =for apidoc sv_isobject
9882 Returns a boolean indicating whether the SV is an RV pointing to a blessed
9883 object. If the SV is not an RV, or if the object is not blessed, then this
9890 Perl_sv_isobject(pTHX_ SV *sv)
9906 Returns a boolean indicating whether the SV is blessed into the specified
9907 class. This does not check for subtypes; use C<sv_derived_from> to verify
9908 an inheritance relationship.
9914 Perl_sv_isa(pTHX_ SV *sv, const char *const name)
9918 PERL_ARGS_ASSERT_SV_ISA;
9928 hvname = HvNAME_get(SvSTASH(sv));
9932 return strEQ(hvname, name);
9938 Creates a new SV for the existing RV, C<rv>, to point to. If C<rv> is not an
9939 RV then it will be upgraded to one. If C<classname> is non-null then the new
9940 SV will be blessed in the specified package. The new SV is returned and its
9941 reference count is 1. The reference count 1 is owned by C<rv>.
9947 Perl_newSVrv(pTHX_ SV *const rv, const char *const classname)
9951 PERL_ARGS_ASSERT_NEWSVRV;
9955 SV_CHECK_THINKFIRST_COW_DROP(rv);
9957 if (SvTYPE(rv) >= SVt_PVMG) {
9958 const U32 refcnt = SvREFCNT(rv);
9962 SvREFCNT(rv) = refcnt;
9964 sv_upgrade(rv, SVt_IV);
9965 } else if (SvROK(rv)) {
9966 SvREFCNT_dec(SvRV(rv));
9968 prepare_SV_for_RV(rv);
9976 HV* const stash = gv_stashpv(classname, GV_ADD);
9977 (void)sv_bless(rv, stash);
9983 Perl_newSVavdefelem(pTHX_ AV *av, SSize_t ix, bool extendible)
9985 SV * const lv = newSV_type(SVt_PVLV);
9986 PERL_ARGS_ASSERT_NEWSVAVDEFELEM;
9988 sv_magic(lv, NULL, PERL_MAGIC_defelem, NULL, 0);
9989 LvTARG(lv) = SvREFCNT_inc_simple_NN(av);
9990 LvSTARGOFF(lv) = ix;
9991 LvTARGLEN(lv) = extendible ? 1 : (STRLEN)UV_MAX;
9996 =for apidoc sv_setref_pv
9998 Copies a pointer into a new SV, optionally blessing the SV. The C<rv>
9999 argument will be upgraded to an RV. That RV will be modified to point to
10000 the new SV. If the C<pv> argument is NULL then C<PL_sv_undef> will be placed
10001 into the SV. The C<classname> argument indicates the package for the
10002 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10003 will have a reference count of 1, and the RV will be returned.
10005 Do not use with other Perl types such as HV, AV, SV, CV, because those
10006 objects will become corrupted by the pointer copy process.
10008 Note that C<sv_setref_pvn> copies the string while this copies the pointer.
10014 Perl_sv_setref_pv(pTHX_ SV *const rv, const char *const classname, void *const pv)
10016 PERL_ARGS_ASSERT_SV_SETREF_PV;
10019 sv_setsv(rv, &PL_sv_undef);
10023 sv_setiv(newSVrv(rv,classname), PTR2IV(pv));
10028 =for apidoc sv_setref_iv
10030 Copies an integer into a new SV, optionally blessing the SV. The C<rv>
10031 argument will be upgraded to an RV. That RV will be modified to point to
10032 the new SV. The C<classname> argument indicates the package for the
10033 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10034 will have a reference count of 1, and the RV will be returned.
10040 Perl_sv_setref_iv(pTHX_ SV *const rv, const char *const classname, const IV iv)
10042 PERL_ARGS_ASSERT_SV_SETREF_IV;
10044 sv_setiv(newSVrv(rv,classname), iv);
10049 =for apidoc sv_setref_uv
10051 Copies an unsigned integer into a new SV, optionally blessing the SV. The C<rv>
10052 argument will be upgraded to an RV. That RV will be modified to point to
10053 the new SV. The C<classname> argument indicates the package for the
10054 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10055 will have a reference count of 1, and the RV will be returned.
10061 Perl_sv_setref_uv(pTHX_ SV *const rv, const char *const classname, const UV uv)
10063 PERL_ARGS_ASSERT_SV_SETREF_UV;
10065 sv_setuv(newSVrv(rv,classname), uv);
10070 =for apidoc sv_setref_nv
10072 Copies a double into a new SV, optionally blessing the SV. The C<rv>
10073 argument will be upgraded to an RV. That RV will be modified to point to
10074 the new SV. The C<classname> argument indicates the package for the
10075 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10076 will have a reference count of 1, and the RV will be returned.
10082 Perl_sv_setref_nv(pTHX_ SV *const rv, const char *const classname, const NV nv)
10084 PERL_ARGS_ASSERT_SV_SETREF_NV;
10086 sv_setnv(newSVrv(rv,classname), nv);
10091 =for apidoc sv_setref_pvn
10093 Copies a string into a new SV, optionally blessing the SV. The length of the
10094 string must be specified with C<n>. The C<rv> argument will be upgraded to
10095 an RV. That RV will be modified to point to the new SV. The C<classname>
10096 argument indicates the package for the blessing. Set C<classname> to
10097 C<NULL> to avoid the blessing. The new SV will have a reference count
10098 of 1, and the RV will be returned.
10100 Note that C<sv_setref_pv> copies the pointer while this copies the string.
10106 Perl_sv_setref_pvn(pTHX_ SV *const rv, const char *const classname,
10107 const char *const pv, const STRLEN n)
10109 PERL_ARGS_ASSERT_SV_SETREF_PVN;
10111 sv_setpvn(newSVrv(rv,classname), pv, n);
10116 =for apidoc sv_bless
10118 Blesses an SV into a specified package. The SV must be an RV. The package
10119 must be designated by its stash (see C<gv_stashpv()>). The reference count
10120 of the SV is unaffected.
10126 Perl_sv_bless(pTHX_ SV *const sv, HV *const stash)
10129 HV *oldstash = NULL;
10131 PERL_ARGS_ASSERT_SV_BLESS;
10135 Perl_croak(aTHX_ "Can't bless non-reference value");
10137 if (SvFLAGS(tmpRef) & (SVs_OBJECT|SVf_READONLY|SVf_PROTECT)) {
10138 if (SvREADONLY(tmpRef))
10139 Perl_croak_no_modify();
10140 if (SvOBJECT(tmpRef)) {
10141 oldstash = SvSTASH(tmpRef);
10144 SvOBJECT_on(tmpRef);
10145 SvUPGRADE(tmpRef, SVt_PVMG);
10146 SvSTASH_set(tmpRef, MUTABLE_HV(SvREFCNT_inc_simple(stash)));
10147 SvREFCNT_dec(oldstash);
10149 if(SvSMAGICAL(tmpRef))
10150 if(mg_find(tmpRef, PERL_MAGIC_ext) || mg_find(tmpRef, PERL_MAGIC_uvar))
10158 /* Downgrades a PVGV to a PVMG. If it's actually a PVLV, we leave the type
10159 * as it is after unglobbing it.
10162 PERL_STATIC_INLINE void
10163 S_sv_unglob(pTHX_ SV *const sv, U32 flags)
10167 SV * const temp = flags & SV_COW_DROP_PV ? NULL : sv_newmortal();
10169 PERL_ARGS_ASSERT_SV_UNGLOB;
10171 assert(SvTYPE(sv) == SVt_PVGV || SvTYPE(sv) == SVt_PVLV);
10173 if (!(flags & SV_COW_DROP_PV))
10174 gv_efullname3(temp, MUTABLE_GV(sv), "*");
10176 SvREFCNT_inc_simple_void_NN(sv_2mortal(sv));
10178 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
10179 && HvNAME_get(stash))
10180 mro_method_changed_in(stash);
10181 gp_free(MUTABLE_GV(sv));
10184 sv_del_backref(MUTABLE_SV(GvSTASH(sv)), sv);
10185 GvSTASH(sv) = NULL;
10188 if (GvNAME_HEK(sv)) {
10189 unshare_hek(GvNAME_HEK(sv));
10191 isGV_with_GP_off(sv);
10193 if(SvTYPE(sv) == SVt_PVGV) {
10194 /* need to keep SvANY(sv) in the right arena */
10195 xpvmg = new_XPVMG();
10196 StructCopy(SvANY(sv), xpvmg, XPVMG);
10197 del_XPVGV(SvANY(sv));
10200 SvFLAGS(sv) &= ~SVTYPEMASK;
10201 SvFLAGS(sv) |= SVt_PVMG;
10204 /* Intentionally not calling any local SET magic, as this isn't so much a
10205 set operation as merely an internal storage change. */
10206 if (flags & SV_COW_DROP_PV) SvOK_off(sv);
10207 else sv_setsv_flags(sv, temp, 0);
10209 if ((const GV *)sv == PL_last_in_gv)
10210 PL_last_in_gv = NULL;
10211 else if ((const GV *)sv == PL_statgv)
10216 =for apidoc sv_unref_flags
10218 Unsets the RV status of the SV, and decrements the reference count of
10219 whatever was being referenced by the RV. This can almost be thought of
10220 as a reversal of C<newSVrv>. The C<cflags> argument can contain
10221 C<SV_IMMEDIATE_UNREF> to force the reference count to be decremented
10222 (otherwise the decrementing is conditional on the reference count being
10223 different from one or the reference being a readonly SV).
10230 Perl_sv_unref_flags(pTHX_ SV *const ref, const U32 flags)
10232 SV* const target = SvRV(ref);
10234 PERL_ARGS_ASSERT_SV_UNREF_FLAGS;
10236 if (SvWEAKREF(ref)) {
10237 sv_del_backref(target, ref);
10238 SvWEAKREF_off(ref);
10239 SvRV_set(ref, NULL);
10242 SvRV_set(ref, NULL);
10244 /* You can't have a || SvREADONLY(target) here, as $a = $$a, where $a was
10245 assigned to as BEGIN {$a = \"Foo"} will fail. */
10246 if (SvREFCNT(target) != 1 || (flags & SV_IMMEDIATE_UNREF))
10247 SvREFCNT_dec_NN(target);
10248 else /* XXX Hack, but hard to make $a=$a->[1] work otherwise */
10249 sv_2mortal(target); /* Schedule for freeing later */
10253 =for apidoc sv_untaint
10255 Untaint an SV. Use C<SvTAINTED_off> instead.
10261 Perl_sv_untaint(pTHX_ SV *const sv)
10263 PERL_ARGS_ASSERT_SV_UNTAINT;
10264 PERL_UNUSED_CONTEXT;
10266 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10267 MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10274 =for apidoc sv_tainted
10276 Test an SV for taintedness. Use C<SvTAINTED> instead.
10282 Perl_sv_tainted(pTHX_ SV *const sv)
10284 PERL_ARGS_ASSERT_SV_TAINTED;
10285 PERL_UNUSED_CONTEXT;
10287 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10288 const MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10289 if (mg && (mg->mg_len & 1) )
10296 =for apidoc sv_setpviv
10298 Copies an integer into the given SV, also updating its string value.
10299 Does not handle 'set' magic. See C<sv_setpviv_mg>.
10305 Perl_sv_setpviv(pTHX_ SV *const sv, const IV iv)
10307 char buf[TYPE_CHARS(UV)];
10309 char * const ptr = uiv_2buf(buf, iv, 0, 0, &ebuf);
10311 PERL_ARGS_ASSERT_SV_SETPVIV;
10313 sv_setpvn(sv, ptr, ebuf - ptr);
10317 =for apidoc sv_setpviv_mg
10319 Like C<sv_setpviv>, but also handles 'set' magic.
10325 Perl_sv_setpviv_mg(pTHX_ SV *const sv, const IV iv)
10327 PERL_ARGS_ASSERT_SV_SETPVIV_MG;
10329 sv_setpviv(sv, iv);
10333 #if defined(PERL_IMPLICIT_CONTEXT)
10335 /* pTHX_ magic can't cope with varargs, so this is a no-context
10336 * version of the main function, (which may itself be aliased to us).
10337 * Don't access this version directly.
10341 Perl_sv_setpvf_nocontext(SV *const sv, const char *const pat, ...)
10346 PERL_ARGS_ASSERT_SV_SETPVF_NOCONTEXT;
10348 va_start(args, pat);
10349 sv_vsetpvf(sv, pat, &args);
10353 /* pTHX_ magic can't cope with varargs, so this is a no-context
10354 * version of the main function, (which may itself be aliased to us).
10355 * Don't access this version directly.
10359 Perl_sv_setpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10364 PERL_ARGS_ASSERT_SV_SETPVF_MG_NOCONTEXT;
10366 va_start(args, pat);
10367 sv_vsetpvf_mg(sv, pat, &args);
10373 =for apidoc sv_setpvf
10375 Works like C<sv_catpvf> but copies the text into the SV instead of
10376 appending it. Does not handle 'set' magic. See C<sv_setpvf_mg>.
10382 Perl_sv_setpvf(pTHX_ SV *const sv, const char *const pat, ...)
10386 PERL_ARGS_ASSERT_SV_SETPVF;
10388 va_start(args, pat);
10389 sv_vsetpvf(sv, pat, &args);
10394 =for apidoc sv_vsetpvf
10396 Works like C<sv_vcatpvf> but copies the text into the SV instead of
10397 appending it. Does not handle 'set' magic. See C<sv_vsetpvf_mg>.
10399 Usually used via its frontend C<sv_setpvf>.
10405 Perl_sv_vsetpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10407 PERL_ARGS_ASSERT_SV_VSETPVF;
10409 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10413 =for apidoc sv_setpvf_mg
10415 Like C<sv_setpvf>, but also handles 'set' magic.
10421 Perl_sv_setpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10425 PERL_ARGS_ASSERT_SV_SETPVF_MG;
10427 va_start(args, pat);
10428 sv_vsetpvf_mg(sv, pat, &args);
10433 =for apidoc sv_vsetpvf_mg
10435 Like C<sv_vsetpvf>, but also handles 'set' magic.
10437 Usually used via its frontend C<sv_setpvf_mg>.
10443 Perl_sv_vsetpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10445 PERL_ARGS_ASSERT_SV_VSETPVF_MG;
10447 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10451 #if defined(PERL_IMPLICIT_CONTEXT)
10453 /* pTHX_ magic can't cope with varargs, so this is a no-context
10454 * version of the main function, (which may itself be aliased to us).
10455 * Don't access this version directly.
10459 Perl_sv_catpvf_nocontext(SV *const sv, const char *const pat, ...)
10464 PERL_ARGS_ASSERT_SV_CATPVF_NOCONTEXT;
10466 va_start(args, pat);
10467 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10471 /* pTHX_ magic can't cope with varargs, so this is a no-context
10472 * version of the main function, (which may itself be aliased to us).
10473 * Don't access this version directly.
10477 Perl_sv_catpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10482 PERL_ARGS_ASSERT_SV_CATPVF_MG_NOCONTEXT;
10484 va_start(args, pat);
10485 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10492 =for apidoc sv_catpvf
10494 Processes its arguments like C<sprintf> and appends the formatted
10495 output to an SV. If the appended data contains "wide" characters
10496 (including, but not limited to, SVs with a UTF-8 PV formatted with %s,
10497 and characters >255 formatted with %c), the original SV might get
10498 upgraded to UTF-8. Handles 'get' magic, but not 'set' magic. See
10499 C<sv_catpvf_mg>. If the original SV was UTF-8, the pattern should be
10500 valid UTF-8; if the original SV was bytes, the pattern should be too.
10505 Perl_sv_catpvf(pTHX_ SV *const sv, const char *const pat, ...)
10509 PERL_ARGS_ASSERT_SV_CATPVF;
10511 va_start(args, pat);
10512 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10517 =for apidoc sv_vcatpvf
10519 Processes its arguments like C<vsprintf> and appends the formatted output
10520 to an SV. Does not handle 'set' magic. See C<sv_vcatpvf_mg>.
10522 Usually used via its frontend C<sv_catpvf>.
10528 Perl_sv_vcatpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10530 PERL_ARGS_ASSERT_SV_VCATPVF;
10532 sv_vcatpvfn_flags(sv, pat, strlen(pat), args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10536 =for apidoc sv_catpvf_mg
10538 Like C<sv_catpvf>, but also handles 'set' magic.
10544 Perl_sv_catpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10548 PERL_ARGS_ASSERT_SV_CATPVF_MG;
10550 va_start(args, pat);
10551 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10557 =for apidoc sv_vcatpvf_mg
10559 Like C<sv_vcatpvf>, but also handles 'set' magic.
10561 Usually used via its frontend C<sv_catpvf_mg>.
10567 Perl_sv_vcatpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10569 PERL_ARGS_ASSERT_SV_VCATPVF_MG;
10571 sv_vcatpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10576 =for apidoc sv_vsetpvfn
10578 Works like C<sv_vcatpvfn> but copies the text into the SV instead of
10581 Usually used via one of its frontends C<sv_vsetpvf> and C<sv_vsetpvf_mg>.
10587 Perl_sv_vsetpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10588 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
10590 PERL_ARGS_ASSERT_SV_VSETPVFN;
10593 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, 0);
10598 * Warn of missing argument to sprintf, and then return a defined value
10599 * to avoid inappropriate "use of uninit" warnings [perl #71000].
10602 S_vcatpvfn_missing_argument(pTHX) {
10603 if (ckWARN(WARN_MISSING)) {
10604 Perl_warner(aTHX_ packWARN(WARN_MISSING), "Missing argument in %s",
10605 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
10612 S_expect_number(pTHX_ char **const pattern)
10616 PERL_ARGS_ASSERT_EXPECT_NUMBER;
10618 switch (**pattern) {
10619 case '1': case '2': case '3':
10620 case '4': case '5': case '6':
10621 case '7': case '8': case '9':
10622 var = *(*pattern)++ - '0';
10623 while (isDIGIT(**pattern)) {
10624 const I32 tmp = var * 10 + (*(*pattern)++ - '0');
10626 Perl_croak(aTHX_ "Integer overflow in format string for %s", (PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn"));
10634 S_F0convert(NV nv, char *const endbuf, STRLEN *const len)
10636 const int neg = nv < 0;
10639 PERL_ARGS_ASSERT_F0CONVERT;
10641 if (UNLIKELY(Perl_isinfnan(nv))) {
10642 STRLEN n = S_infnan_2pv(nv, endbuf - *len, *len);
10652 if (uv & 1 && uv == nv)
10653 uv--; /* Round to even */
10655 const unsigned dig = uv % 10;
10657 } while (uv /= 10);
10668 =for apidoc sv_vcatpvfn
10670 =for apidoc sv_vcatpvfn_flags
10672 Processes its arguments like C<vsprintf> and appends the formatted output
10673 to an SV. Uses an array of SVs if the C style variable argument list is
10674 missing (NULL). When running with taint checks enabled, indicates via
10675 C<maybe_tainted> if results are untrustworthy (often due to the use of
10678 If called as C<sv_vcatpvfn> or flags include C<SV_GMAGIC>, calls get magic.
10680 Usually used via one of its frontends C<sv_vcatpvf> and C<sv_vcatpvf_mg>.
10685 #define VECTORIZE_ARGS vecsv = va_arg(*args, SV*);\
10686 vecstr = (U8*)SvPV_const(vecsv,veclen);\
10687 vec_utf8 = DO_UTF8(vecsv);
10689 /* XXX maybe_tainted is never assigned to, so the doc above is lying. */
10692 Perl_sv_vcatpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10693 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
10695 PERL_ARGS_ASSERT_SV_VCATPVFN;
10697 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, SV_GMAGIC|SV_SMAGIC);
10700 #if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN || \
10701 LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN || \
10702 LONG_DOUBLEKIND == LONG_DOUBLE_IS_DOUBLEDOUBLE_128_BIT_LITTLE_ENDIAN
10703 # define LONGDOUBLE_LITTLE_ENDIAN
10706 #if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_BIG_ENDIAN || \
10707 LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN || \
10708 LONG_DOUBLEKIND == LONG_DOUBLE_IS_DOUBLEDOUBLE_128_BIT_BIG_ENDIAN
10709 # define LONGDOUBLE_BIG_ENDIAN
10712 #if LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN || \
10713 LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN
10714 # define LONGDOUBLE_X86_80_BIT
10717 #if LONG_DOUBLEKIND == LONG_DOUBLE_IS_DOUBLEDOUBLE_128_BIT_LITTLE_ENDIAN || \
10718 LONG_DOUBLEKIND == LONG_DOUBLE_IS_DOUBLEDOUBLE_128_BIT_BIG_ENDIAN
10719 # define LONGDOUBLE_DOUBLEDOUBLE
10720 # define DOUBLEDOUBLE_MAXBITS 1028
10723 /* vhex will contain the values (0..15) of the hex digits ("nybbles"
10724 * of 4 bits); 1 for the implicit 1, and the mantissa bits, four bits
10726 #ifdef LONGDOUBLE_DOUBLEDOUBLE
10727 # define VHEX_SIZE (1+DOUBLEDOUBLE_MAXBITS/4)
10729 # define VHEX_SIZE (1+128/4)
10732 /* If we do not have a known long double format, (including not using
10733 * long doubles, or long doubles being equal to doubles) then we will
10734 * fall back to the ldexp/frexp route, with which we can retrieve at
10735 * most as many bits as our widest unsigned integer type is. We try
10736 * to get a 64-bit unsigned integer even if we are not using a 64-bit UV.
10738 * (If you want to test the case of UVSIZE == 4, NVSIZE == 8,
10739 * set the MANTISSATYPE to int and the MANTISSASIZE to 4.)
10741 #if defined(HAS_QUAD) && defined(Uquad_t)
10742 # define MANTISSATYPE Uquad_t
10743 # define MANTISSASIZE 8
10745 # define MANTISSATYPE UV
10746 # define MANTISSASIZE UVSIZE
10749 /* We make here the wild assumption that the endianness of doubles
10750 * is similar to the endianness of integers, and that there is no
10751 * middle-endianness. This may come back to haunt us (the rumor
10752 * has it that ARM can be quite haunted). */
10753 #if BYTEORDER == 0x12345678 || BYTEORDER == 0x1234 || \
10754 defined(DOUBLEKIND_LITTLE_ENDIAN)
10755 # define HEXTRACT_LITTLE_ENDIAN
10757 # define HEXTRACT_BIG_ENDIAN
10760 /* S_hextract() is a helper for Perl_sv_vcatpvfn_flags, for extracting
10761 * the hexadecimal values (for %a/%A). The nv is the NV where the value
10762 * are being extracted from (either directly from the long double in-memory
10763 * presentation, or from the uquad computed via frexp+ldexp). frexp also
10764 * is used to update the exponent. vhex is the pointer to the beginning
10765 * of the output buffer (of VHEX_SIZE).
10767 * The tricky part is that S_hextract() needs to be called twice:
10768 * the first time with vend as NULL, and the second time with vend as
10769 * the pointer returned by the first call. What happens is that on
10770 * the first round the output size is computed, and the intended
10771 * extraction sanity checked. On the second round the actual output
10772 * (the extraction of the hexadecimal values) takes place.
10773 * Sanity failures cause fatal failures during both rounds. */
10775 S_hextract(pTHX_ const NV nv, int* exponent, U8* vhex, U8* vend)
10779 int ixmin = 0, ixmax = 0;
10781 /* XXX Inf/NaN/denormal handling in the HEXTRACT_IMPLICIT_BIT,
10782 * and elsewhere. */
10784 /* These macros are just to reduce typos, they have multiple
10785 * repetitions below, but usually only one (or sometimes two)
10786 * of them is really being used. */
10787 /* HEXTRACT_OUTPUT() extracts the high nybble first. */
10788 #define HEXTRACT_OUTPUT_HI(ix) (*v++ = nvp[ix] >> 4)
10789 #define HEXTRACT_OUTPUT_LO(ix) (*v++ = nvp[ix] & 0xF)
10790 #define HEXTRACT_OUTPUT(ix) \
10792 HEXTRACT_OUTPUT_HI(ix); HEXTRACT_OUTPUT_LO(ix); \
10794 #define HEXTRACT_COUNT(ix, c) \
10796 v += c; if (ix < ixmin) ixmin = ix; else if (ix > ixmax) ixmax = ix; \
10798 #define HEXTRACT_BYTE(ix) \
10800 if (vend) HEXTRACT_OUTPUT(ix); else HEXTRACT_COUNT(ix, 2); \
10802 #define HEXTRACT_LO_NYBBLE(ix) \
10804 if (vend) HEXTRACT_OUTPUT_LO(ix); else HEXTRACT_COUNT(ix, 1); \
10806 # define HEXTRACT_IMPLICIT_BIT(nv) \
10808 if (vend) *v++ = ((nv) == 0.0) ? 0 : 1; else v++; \
10811 #ifdef LONGDOUBLE_DOUBLEDOUBLE
10812 # define HEXTRACTSIZE (DOUBLEDOUBLE_MAXBITS/8)
10814 # define HEXTRACTSIZE NVSIZE
10817 const U8* nvp = (const U8*)(&nv);
10818 const U8* vmaxend = vhex + 2 * HEXTRACTSIZE + 1;
10819 (void)Perl_frexp(PERL_ABS(nv), exponent);
10820 if (vend && (vend <= vhex || vend > vmaxend))
10821 Perl_croak(aTHX_ "Hexadecimal float: internal error");
10823 /* First check if using long doubles. */
10824 #if NVSIZE > DOUBLESIZE
10825 # if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN
10826 /* Used in e.g. VMS and HP-UX IA-64, e.g. -0.1L:
10827 * 9a 99 99 99 99 99 99 99 99 99 99 99 99 99 fb 3f */
10828 /* The bytes 13..0 are the mantissa/fraction,
10829 * the 15,14 are the sign+exponent. */
10830 HEXTRACT_IMPLICIT_BIT(nv);
10831 for (ix = 13; ix >= 0; ix--) {
10834 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_BIG_ENDIAN
10835 /* Used in e.g. Solaris Sparc and HP-UX PA-RISC, e.g. -0.1L:
10836 * bf fb 99 99 99 99 99 99 99 99 99 99 99 99 99 9a */
10837 /* The bytes 2..15 are the mantissa/fraction,
10838 * the 0,1 are the sign+exponent. */
10839 HEXTRACT_IMPLICIT_BIT(nv);
10840 for (ix = 2; ix <= 15; ix++) {
10843 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN
10844 /* x86 80-bit "extended precision", 64 bits of mantissa / fraction /
10845 * significand, 15 bits of exponent, 1 bit of sign. NVSIZE can
10846 * be either 12 (ILP32, Solaris x86) or 16 (LP64, Linux and OS X),
10847 * meaning that 2 or 6 bytes are empty padding. */
10848 /* The bytes 7..0 are the mantissa/fraction */
10850 /* Intentionally NO HEXTRACT_IMPLICIT_BIT here. */
10851 for (ix = 7; ix >= 0; ix--) {
10854 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN
10855 /* Does this format ever happen? (Wikipedia says the Motorola
10856 * 6888x math coprocessors used format _like_ this but padded
10857 * to 96 bits with 16 unused bits between the exponent and the
10860 /* Intentionally NO HEXTRACT_IMPLICIT_BIT here. */
10861 for (ix = 0; ix < 8; ix++) {
10864 # elif defined(LONGDOUBLE_DOUBLEDOUBLE)
10865 /* Double-double format: two doubles next to each other.
10866 * The first double is the high-order one, exactly like
10867 * it would be for a "lone" double. The second double
10868 * is shifted down using the exponent so that that there
10869 * are no common bits. The tricky part is that the value
10870 * of the double-double is the SUM of the two doubles and
10871 * the second one can be also NEGATIVE.
10873 * Because of this tricky construction the bytewise extraction we
10874 * use for the other long double formats doesn't work, we must
10875 * extract the values bit by bit.
10877 * The little-endian double-double is used .. somewhere?
10879 * The big endian double-double is used in e.g. PPC/Power (AIX)
10882 * The mantissa bits are in two separate stretches, e.g. for -0.1L:
10883 * 9a 99 99 99 99 99 59 bc 9a 99 99 99 99 99 b9 3f (LE)
10884 * 3f b9 99 99 99 99 99 9a bc 59 99 99 99 99 99 9a (BE)
10887 if (nv == (NV)0.0) {
10895 NV d = nv < 0 ? -nv : nv;
10897 U8 ha = 0x0; /* hexvalue accumulator */
10898 U8 hd = 0x8; /* hexvalue digit */
10900 /* Shift d and e (and update exponent) so that e <= d < 2*e,
10901 * this is essentially manual frexp(). Multiplying by 0.5 and
10902 * doubling should be lossless in binary floating point. */
10912 while (d >= e + e) {
10916 /* Now e <= d < 2*e */
10918 /* First extract the leading hexdigit (the implicit bit). */
10934 /* Then extract the remaining hexdigits. */
10935 while (d > (NV)0.0) {
10941 /* Output or count in groups of four bits,
10942 * that is, when the hexdigit is down to one. */
10947 /* Reset the hexvalue. */
10956 /* Flush possible pending hexvalue. */
10966 "Hexadecimal float: unsupported long double format");
10969 /* Using normal doubles, not long doubles.
10971 * We generate 4-bit xdigits (nybble/nibble) instead of 8-bit
10972 * bytes, since we might need to handle printf precision, and
10973 * also need to insert the radix. */
10974 HEXTRACT_IMPLICIT_BIT(nv);
10975 # ifdef HEXTRACT_LITTLE_ENDIAN
10976 HEXTRACT_LO_NYBBLE(6);
10977 for (ix = 5; ix >= 0; ix--) {
10981 HEXTRACT_LO_NYBBLE(1);
10982 for (ix = 2; ix < HEXTRACTSIZE; ix++) {
10987 /* Croak for various reasons: if the output pointer escaped the
10988 * output buffer, if the extraction index escaped the extraction
10989 * buffer, or if the ending output pointer didn't match the
10990 * previously computed value. */
10991 if (v <= vhex || v - vhex >= VHEX_SIZE ||
10992 /* For double-double the ixmin and ixmax stay at zero,
10993 * which is convenient since the HEXTRACTSIZE is tricky
10994 * for double-double. */
10995 ixmin < 0 || ixmax >= HEXTRACTSIZE ||
10996 (vend && v != vend))
10997 Perl_croak(aTHX_ "Hexadecimal float: internal error");
11002 Perl_sv_vcatpvfn_flags(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11003 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted,
11008 const char *patend;
11011 static const char nullstr[] = "(null)";
11013 bool has_utf8 = DO_UTF8(sv); /* has the result utf8? */
11014 const bool pat_utf8 = has_utf8; /* the pattern is in utf8? */
11016 /* Times 4: a decimal digit takes more than 3 binary digits.
11017 * NV_DIG: mantissa takes than many decimal digits.
11018 * Plus 32: Playing safe. */
11019 char ebuf[IV_DIG * 4 + NV_DIG + 32];
11020 bool no_redundant_warning = FALSE; /* did we use any explicit format parameter index? */
11021 bool hexfp = FALSE; /* hexadecimal floating point? */
11023 DECLARATION_FOR_STORE_LC_NUMERIC_SET_TO_NEEDED;
11025 PERL_ARGS_ASSERT_SV_VCATPVFN_FLAGS;
11026 PERL_UNUSED_ARG(maybe_tainted);
11028 if (flags & SV_GMAGIC)
11031 /* no matter what, this is a string now */
11032 (void)SvPV_force_nomg(sv, origlen);
11034 /* special-case "", "%s", and "%-p" (SVf - see below) */
11036 if (svmax && ckWARN(WARN_REDUNDANT))
11037 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11038 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11041 if (patlen == 2 && pat[0] == '%' && pat[1] == 's') {
11042 if (svmax > 1 && ckWARN(WARN_REDUNDANT))
11043 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11044 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11047 const char * const s = va_arg(*args, char*);
11048 sv_catpv_nomg(sv, s ? s : nullstr);
11050 else if (svix < svmax) {
11051 /* we want get magic on the source but not the target. sv_catsv can't do that, though */
11052 SvGETMAGIC(*svargs);
11053 sv_catsv_nomg(sv, *svargs);
11056 S_vcatpvfn_missing_argument(aTHX);
11059 if (args && patlen == 3 && pat[0] == '%' &&
11060 pat[1] == '-' && pat[2] == 'p') {
11061 if (svmax > 1 && ckWARN(WARN_REDUNDANT))
11062 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11063 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11064 argsv = MUTABLE_SV(va_arg(*args, void*));
11065 sv_catsv_nomg(sv, argsv);
11069 #ifndef USE_LONG_DOUBLE
11070 /* special-case "%.<number>[gf]" */
11071 if ( !args && patlen <= 5 && pat[0] == '%' && pat[1] == '.'
11072 && (pat[patlen-1] == 'g' || pat[patlen-1] == 'f') ) {
11073 unsigned digits = 0;
11077 while (*pp >= '0' && *pp <= '9')
11078 digits = 10 * digits + (*pp++ - '0');
11080 /* XXX: Why do this `svix < svmax` test? Couldn't we just
11081 format the first argument and WARN_REDUNDANT if svmax > 1?
11082 Munged by Nicholas Clark in v5.13.0-209-g95ea86d */
11083 if (pp - pat == (int)patlen - 1 && svix < svmax) {
11084 const NV nv = SvNV(*svargs);
11085 if (LIKELY(!Perl_isinfnan(nv))) {
11087 /* Add check for digits != 0 because it seems that some
11088 gconverts are buggy in this case, and we don't yet have
11089 a Configure test for this. */
11090 if (digits && digits < sizeof(ebuf) - NV_DIG - 10) {
11091 /* 0, point, slack */
11092 STORE_LC_NUMERIC_SET_TO_NEEDED();
11093 SNPRINTF_G(nv, ebuf, size, digits);
11094 sv_catpv_nomg(sv, ebuf);
11095 if (*ebuf) /* May return an empty string for digits==0 */
11098 } else if (!digits) {
11101 if ((p = F0convert(nv, ebuf + sizeof ebuf, &l))) {
11102 sv_catpvn_nomg(sv, p, l);
11109 #endif /* !USE_LONG_DOUBLE */
11111 if (!args && svix < svmax && DO_UTF8(*svargs))
11114 patend = (char*)pat + patlen;
11115 for (p = (char*)pat; p < patend; p = q) {
11118 bool vectorize = FALSE;
11119 bool vectorarg = FALSE;
11120 bool vec_utf8 = FALSE;
11126 bool has_precis = FALSE;
11128 const I32 osvix = svix;
11129 bool is_utf8 = FALSE; /* is this item utf8? */
11130 #ifdef HAS_LDBL_SPRINTF_BUG
11131 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
11132 with sfio - Allen <allens@cpan.org> */
11133 bool fix_ldbl_sprintf_bug = FALSE;
11137 U8 utf8buf[UTF8_MAXBYTES+1];
11138 STRLEN esignlen = 0;
11140 const char *eptr = NULL;
11141 const char *fmtstart;
11144 const U8 *vecstr = NULL;
11151 /* We need a long double target in case HAS_LONG_DOUBLE,
11152 * even without USE_LONG_DOUBLE, so that we can printf with
11153 * long double formats, even without NV being long double.
11154 * But we call the target 'fv' instead of 'nv', since most of
11155 * the time it is not (most compilers these days recognize
11156 * "long double", even if only as a synonym for "double").
11158 #if defined(HAS_LONG_DOUBLE) && LONG_DOUBLESIZE > DOUBLESIZE && \
11159 defined(PERL_PRIgldbl) && !defined(USE_QUADMATH)
11161 # define FV_ISFINITE(x) Perl_isfinitel(x)
11162 # define FV_GF PERL_PRIgldbl
11163 # if defined(__VMS) && defined(__ia64) && defined(__IEEE_FLOAT)
11164 /* Work around breakage in OTS$CVT_FLOAT_T_X */
11165 # define NV_TO_FV(nv,fv) STMT_START { \
11167 fv = Perl_isnan(_dv) ? LDBL_QNAN : _dv; \
11170 # define NV_TO_FV(nv,fv) (fv)=(nv)
11174 # define FV_ISFINITE(x) Perl_isfinite((NV)(x))
11175 # define FV_GF NVgf
11176 # define NV_TO_FV(nv,fv) (fv)=(nv)
11181 const char *dotstr = ".";
11182 STRLEN dotstrlen = 1;
11183 I32 efix = 0; /* explicit format parameter index */
11184 I32 ewix = 0; /* explicit width index */
11185 I32 epix = 0; /* explicit precision index */
11186 I32 evix = 0; /* explicit vector index */
11187 bool asterisk = FALSE;
11188 bool infnan = FALSE;
11190 /* echo everything up to the next format specification */
11191 for (q = p; q < patend && *q != '%'; ++q) ;
11193 if (has_utf8 && !pat_utf8)
11194 sv_catpvn_nomg_utf8_upgrade(sv, p, q - p, nsv);
11196 sv_catpvn_nomg(sv, p, q - p);
11205 We allow format specification elements in this order:
11206 \d+\$ explicit format parameter index
11208 v|\*(\d+\$)?v vector with optional (optionally specified) arg
11209 0 flag (as above): repeated to allow "v02"
11210 \d+|\*(\d+\$)? width using optional (optionally specified) arg
11211 \.(\d*|\*(\d+\$)?) precision using optional (optionally specified) arg
11213 [%bcdefginopsuxDFOUX] format (mandatory)
11218 As of perl5.9.3, printf format checking is on by default.
11219 Internally, perl uses %p formats to provide an escape to
11220 some extended formatting. This block deals with those
11221 extensions: if it does not match, (char*)q is reset and
11222 the normal format processing code is used.
11224 Currently defined extensions are:
11225 %p include pointer address (standard)
11226 %-p (SVf) include an SV (previously %_)
11227 %-<num>p include an SV with precision <num>
11229 %3p include a HEK with precision of 256
11230 %4p char* preceded by utf8 flag and length
11231 %<num>p (where num is 1 or > 4) reserved for future
11234 Robin Barker 2005-07-14 (but modified since)
11236 %1p (VDf) removed. RMB 2007-10-19
11243 else if (strnEQ(q, UTF8f, sizeof(UTF8f)-1)) { /* UTF8f */
11244 /* The argument has already gone through cBOOL, so the cast
11246 is_utf8 = (bool)va_arg(*args, int);
11247 elen = va_arg(*args, UV);
11248 if ((IV)elen < 0) {
11249 /* check if utf8 length is larger than 0 when cast to IV */
11250 assert( (IV)elen >= 0 ); /* in DEBUGGING build we want to crash */
11251 elen= 0; /* otherwise we want to treat this as an empty string */
11253 eptr = va_arg(*args, char *);
11254 q += sizeof(UTF8f)-1;
11257 n = expect_number(&q);
11259 if (sv) { /* SVf */
11264 argsv = MUTABLE_SV(va_arg(*args, void*));
11265 eptr = SvPV_const(argsv, elen);
11266 if (DO_UTF8(argsv))
11270 else if (n==2 || n==3) { /* HEKf */
11271 HEK * const hek = va_arg(*args, HEK *);
11272 eptr = HEK_KEY(hek);
11273 elen = HEK_LEN(hek);
11274 if (HEK_UTF8(hek)) is_utf8 = TRUE;
11275 if (n==3) precis = 256, has_precis = TRUE;
11279 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
11280 "internal %%<num>p might conflict with future printf extensions");
11286 if ( (width = expect_number(&q)) ) {
11290 if (!no_redundant_warning)
11291 /* I've forgotten if it's a better
11292 micro-optimization to always set this or to
11293 only set it if it's unset */
11294 no_redundant_warning = TRUE;
11306 if (plus == '+' && *q == ' ') /* '+' over ' ' */
11335 if ( (ewix = expect_number(&q)) )
11344 if ((vectorarg = asterisk)) {
11357 width = expect_number(&q);
11360 if (vectorize && vectorarg) {
11361 /* vectorizing, but not with the default "." */
11363 vecsv = va_arg(*args, SV*);
11365 vecsv = (evix > 0 && evix <= svmax)
11366 ? svargs[evix-1] : S_vcatpvfn_missing_argument(aTHX);
11368 vecsv = svix < svmax
11369 ? svargs[svix++] : S_vcatpvfn_missing_argument(aTHX);
11371 dotstr = SvPV_const(vecsv, dotstrlen);
11372 /* Keep the DO_UTF8 test *after* the SvPV call, else things go
11373 bad with tied or overloaded values that return UTF8. */
11374 if (DO_UTF8(vecsv))
11376 else if (has_utf8) {
11377 vecsv = sv_mortalcopy(vecsv);
11378 sv_utf8_upgrade(vecsv);
11379 dotstr = SvPV_const(vecsv, dotstrlen);
11386 i = va_arg(*args, int);
11388 i = (ewix ? ewix <= svmax : svix < svmax) ?
11389 SvIVx(svargs[ewix ? ewix-1 : svix++]) : 0;
11391 width = (i < 0) ? -i : i;
11401 if ( ((epix = expect_number(&q))) && (*q++ != '$') )
11403 /* XXX: todo, support specified precision parameter */
11407 i = va_arg(*args, int);
11409 i = (ewix ? ewix <= svmax : svix < svmax)
11410 ? SvIVx(svargs[ewix ? ewix-1 : svix++]) : 0;
11412 has_precis = !(i < 0);
11416 while (isDIGIT(*q))
11417 precis = precis * 10 + (*q++ - '0');
11426 else if (efix ? (efix > 0 && efix <= svmax) : svix < svmax) {
11427 vecsv = svargs[efix ? efix-1 : svix++];
11428 vecstr = (U8*)SvPV_const(vecsv,veclen);
11429 vec_utf8 = DO_UTF8(vecsv);
11431 /* if this is a version object, we need to convert
11432 * back into v-string notation and then let the
11433 * vectorize happen normally
11435 if (sv_isobject(vecsv) && sv_derived_from(vecsv, "version")) {
11436 if ( hv_exists(MUTABLE_HV(SvRV(vecsv)), "alpha", 5 ) ) {
11437 Perl_ck_warner_d(aTHX_ packWARN(WARN_PRINTF),
11438 "vector argument not supported with alpha versions");
11441 vecsv = sv_newmortal();
11442 scan_vstring((char *)vecstr, (char *)vecstr + veclen,
11444 vecstr = (U8*)SvPV_const(vecsv, veclen);
11445 vec_utf8 = DO_UTF8(vecsv);
11459 case 'I': /* Ix, I32x, and I64x */
11460 # ifdef USE_64_BIT_INT
11461 if (q[1] == '6' && q[2] == '4') {
11467 if (q[1] == '3' && q[2] == '2') {
11471 # ifdef USE_64_BIT_INT
11477 #if IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)
11480 #ifdef USE_QUADMATH
11493 #if IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)
11494 if (*q == 'l') { /* lld, llf */
11503 if (*++q == 'h') { /* hhd, hhu */
11532 if (!vectorize && !args) {
11534 const I32 i = efix-1;
11535 argsv = (i >= 0 && i < svmax)
11536 ? svargs[i] : S_vcatpvfn_missing_argument(aTHX);
11538 argsv = (svix >= 0 && svix < svmax)
11539 ? svargs[svix++] : S_vcatpvfn_missing_argument(aTHX);
11543 if (argsv && strchr("BbcDdiOopuUXx",*q)) {
11544 /* XXX va_arg(*args) case? need peek, use va_copy? */
11546 infnan = UNLIKELY(isinfnansv(argsv));
11549 switch (c = *q++) {
11557 Perl_croak(aTHX_ "Cannot printf %"NVgf" with '%c'",
11558 /* no va_arg() case */
11559 SvNV_nomg(argsv), (int)c);
11560 uv = (args) ? va_arg(*args, int) : SvIV_nomg(argsv);
11562 (!UVCHR_IS_INVARIANT(uv) && SvUTF8(sv)))
11564 eptr = (char*)utf8buf;
11565 elen = uvchr_to_utf8((U8*)eptr, uv) - utf8buf;
11579 eptr = va_arg(*args, char*);
11581 elen = strlen(eptr);
11583 eptr = (char *)nullstr;
11584 elen = sizeof nullstr - 1;
11588 eptr = SvPV_const(argsv, elen);
11589 if (DO_UTF8(argsv)) {
11590 STRLEN old_precis = precis;
11591 if (has_precis && precis < elen) {
11592 STRLEN ulen = sv_or_pv_len_utf8(argsv, eptr, elen);
11593 STRLEN p = precis > ulen ? ulen : precis;
11594 precis = sv_or_pv_pos_u2b(argsv, eptr, p, 0);
11595 /* sticks at end */
11597 if (width) { /* fudge width (can't fudge elen) */
11598 if (has_precis && precis < elen)
11599 width += precis - old_precis;
11602 elen - sv_or_pv_len_utf8(argsv,eptr,elen);
11609 if (has_precis && precis < elen)
11617 goto floating_point;
11619 if (alt || vectorize)
11621 uv = PTR2UV(args ? va_arg(*args, void*) : argsv);
11635 goto floating_point;
11642 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
11651 esignbuf[esignlen++] = plus;
11655 case 'c': iv = (char)va_arg(*args, int); break;
11656 case 'h': iv = (short)va_arg(*args, int); break;
11657 case 'l': iv = va_arg(*args, long); break;
11658 case 'V': iv = va_arg(*args, IV); break;
11659 case 'z': iv = va_arg(*args, SSize_t); break;
11660 #ifdef HAS_PTRDIFF_T
11661 case 't': iv = va_arg(*args, ptrdiff_t); break;
11663 default: iv = va_arg(*args, int); break;
11665 case 'j': iv = va_arg(*args, intmax_t); break;
11669 iv = va_arg(*args, Quad_t); break;
11676 IV tiv = SvIV_nomg(argsv); /* work around GCC bug #13488 */
11678 case 'c': iv = (char)tiv; break;
11679 case 'h': iv = (short)tiv; break;
11680 case 'l': iv = (long)tiv; break;
11682 default: iv = tiv; break;
11685 iv = (Quad_t)tiv; break;
11691 if ( !vectorize ) /* we already set uv above */
11696 esignbuf[esignlen++] = plus;
11700 esignbuf[esignlen++] = '-';
11739 goto floating_point;
11747 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
11758 case 'c': uv = (unsigned char)va_arg(*args, unsigned); break;
11759 case 'h': uv = (unsigned short)va_arg(*args, unsigned); break;
11760 case 'l': uv = va_arg(*args, unsigned long); break;
11761 case 'V': uv = va_arg(*args, UV); break;
11762 case 'z': uv = va_arg(*args, Size_t); break;
11763 #ifdef HAS_PTRDIFF_T
11764 case 't': uv = va_arg(*args, ptrdiff_t); break; /* will sign extend, but there is no uptrdiff_t, so oh well */
11767 case 'j': uv = va_arg(*args, uintmax_t); break;
11769 default: uv = va_arg(*args, unsigned); break;
11772 uv = va_arg(*args, Uquad_t); break;
11779 UV tuv = SvUV_nomg(argsv); /* work around GCC bug #13488 */
11781 case 'c': uv = (unsigned char)tuv; break;
11782 case 'h': uv = (unsigned short)tuv; break;
11783 case 'l': uv = (unsigned long)tuv; break;
11785 default: uv = tuv; break;
11788 uv = (Uquad_t)tuv; break;
11797 char *ptr = ebuf + sizeof ebuf;
11798 bool tempalt = uv ? alt : FALSE; /* Vectors can't change alt */
11804 p = (char *)((c == 'X') ? PL_hexdigit + 16 : PL_hexdigit);
11808 } while (uv >>= 4);
11810 esignbuf[esignlen++] = '0';
11811 esignbuf[esignlen++] = c; /* 'x' or 'X' */
11817 *--ptr = '0' + dig;
11818 } while (uv >>= 3);
11819 if (alt && *ptr != '0')
11825 *--ptr = '0' + dig;
11826 } while (uv >>= 1);
11828 esignbuf[esignlen++] = '0';
11829 esignbuf[esignlen++] = c;
11832 default: /* it had better be ten or less */
11835 *--ptr = '0' + dig;
11836 } while (uv /= base);
11839 elen = (ebuf + sizeof ebuf) - ptr;
11843 zeros = precis - elen;
11844 else if (precis == 0 && elen == 1 && *eptr == '0'
11845 && !(base == 8 && alt)) /* "%#.0o" prints "0" */
11848 /* a precision nullifies the 0 flag. */
11855 /* FLOATING POINT */
11860 c = 'f'; /* maybe %F isn't supported here */
11862 case 'e': case 'E':
11864 case 'g': case 'G':
11865 case 'a': case 'A':
11869 /* This is evil, but floating point is even more evil */
11871 /* for SV-style calling, we can only get NV
11872 for C-style calling, we assume %f is double;
11873 for simplicity we allow any of %Lf, %llf, %qf for long double
11877 #if defined(USE_LONG_DOUBLE)
11881 /* [perl #20339] - we should accept and ignore %lf rather than die */
11885 #if defined(USE_LONG_DOUBLE)
11886 intsize = args ? 0 : 'q';
11890 #if defined(HAS_LONG_DOUBLE)
11903 /* Now we need (long double) if intsize == 'q', else (double). */
11905 /* Note: do not pull NVs off the va_list with va_arg()
11906 * (pull doubles instead) because if you have a build
11907 * with long doubles, you would always be pulling long
11908 * doubles, which would badly break anyone using only
11909 * doubles (i.e. the majority of builds). In other
11910 * words, you cannot mix doubles and long doubles.
11911 * The only case where you can pull off long doubles
11912 * is when the format specifier explicitly asks so with
11914 #ifdef USE_QUADMATH
11915 fv = intsize == 'q' ?
11916 va_arg(*args, NV) : va_arg(*args, double);
11917 #elif LONG_DOUBLESIZE > DOUBLESIZE
11918 if (intsize == 'q')
11919 fv = va_arg(*args, long double);
11921 NV_TO_FV(va_arg(*args, double), fv);
11923 fv = va_arg(*args, double);
11928 if (!infnan) SvGETMAGIC(argsv);
11929 NV_TO_FV(SvNV_nomg(argsv), fv);
11933 /* frexp() (or frexpl) has some unspecified behaviour for
11934 * nan/inf/-inf, so let's avoid calling that on non-finites. */
11935 if (isALPHA_FOLD_NE(c, 'e') && FV_ISFINITE(fv)) {
11937 (void)Perl_frexp((NV)fv, &i);
11938 if (i == PERL_INT_MIN)
11939 Perl_die(aTHX_ "panic: frexp: %"FV_GF, fv);
11940 /* Do not set hexfp earlier since we want to printf
11941 * Inf/NaN for Inf/NaN, not their hexfp. */
11942 hexfp = isALPHA_FOLD_EQ(c, 'a');
11943 if (UNLIKELY(hexfp)) {
11944 /* This seriously overshoots in most cases, but
11945 * better the undershooting. Firstly, all bytes
11946 * of the NV are not mantissa, some of them are
11947 * exponent. Secondly, for the reasonably common
11948 * long doubles case, the "80-bit extended", two
11949 * or six bytes of the NV are unused. */
11951 (fv < 0) ? 1 : 0 + /* possible unary minus */
11953 1 + /* the very unlikely carry */
11956 2 * NVSIZE + /* 2 hexdigits for each byte */
11958 6 + /* exponent: sign, plus up to 16383 (quad fp) */
11960 #ifdef LONGDOUBLE_DOUBLEDOUBLE
11961 /* However, for the "double double", we need more.
11962 * Since each double has their own exponent, the
11963 * doubles may float (haha) rather far from each
11964 * other, and the number of required bits is much
11965 * larger, up to total of 1028 bits. (NOTE: this
11966 * is not actually implemented properly yet,
11967 * we are using just the first double, see
11968 * S_hextract() for details. But let's prepare
11969 * for the future.) */
11971 /* 2 hexdigits for each byte. */
11972 need += (DOUBLEDOUBLE_MAXBITS/8 + 1) * 2;
11973 /* the size for the exponent already added */
11975 #ifdef USE_LOCALE_NUMERIC
11976 STORE_LC_NUMERIC_SET_TO_NEEDED();
11977 if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC))
11978 need += SvLEN(PL_numeric_radix_sv);
11979 RESTORE_LC_NUMERIC();
11983 need = BIT_DIGITS(i);
11984 } /* if i < 0, the number of digits is hard to predict. */
11986 need += has_precis ? precis : 6; /* known default */
11991 #ifdef HAS_LDBL_SPRINTF_BUG
11992 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
11993 with sfio - Allen <allens@cpan.org> */
11996 # define MY_DBL_MAX DBL_MAX
11997 # else /* XXX guessing! HUGE_VAL may be defined as infinity, so not using */
11998 # if DOUBLESIZE >= 8
11999 # define MY_DBL_MAX 1.7976931348623157E+308L
12001 # define MY_DBL_MAX 3.40282347E+38L
12005 # ifdef HAS_LDBL_SPRINTF_BUG_LESS1 /* only between -1L & 1L - Allen */
12006 # define MY_DBL_MAX_BUG 1L
12008 # define MY_DBL_MAX_BUG MY_DBL_MAX
12012 # define MY_DBL_MIN DBL_MIN
12013 # else /* XXX guessing! -Allen */
12014 # if DOUBLESIZE >= 8
12015 # define MY_DBL_MIN 2.2250738585072014E-308L
12017 # define MY_DBL_MIN 1.17549435E-38L
12021 if ((intsize == 'q') && (c == 'f') &&
12022 ((fv < MY_DBL_MAX_BUG) && (fv > -MY_DBL_MAX_BUG)) &&
12023 (need < DBL_DIG)) {
12024 /* it's going to be short enough that
12025 * long double precision is not needed */
12027 if ((fv <= 0L) && (fv >= -0L))
12028 fix_ldbl_sprintf_bug = TRUE; /* 0 is 0 - easiest */
12030 /* would use Perl_fp_class as a double-check but not
12031 * functional on IRIX - see perl.h comments */
12033 if ((fv >= MY_DBL_MIN) || (fv <= -MY_DBL_MIN)) {
12034 /* It's within the range that a double can represent */
12035 #if defined(DBL_MAX) && !defined(DBL_MIN)
12036 if ((fv >= ((long double)1/DBL_MAX)) ||
12037 (fv <= (-(long double)1/DBL_MAX)))
12039 fix_ldbl_sprintf_bug = TRUE;
12042 if (fix_ldbl_sprintf_bug == TRUE) {
12052 # undef MY_DBL_MAX_BUG
12055 #endif /* HAS_LDBL_SPRINTF_BUG */
12057 need += 20; /* fudge factor */
12058 if (PL_efloatsize < need) {
12059 Safefree(PL_efloatbuf);
12060 PL_efloatsize = need + 20; /* more fudge */
12061 Newx(PL_efloatbuf, PL_efloatsize, char);
12062 PL_efloatbuf[0] = '\0';
12065 if ( !(width || left || plus || alt) && fill != '0'
12066 && has_precis && intsize != 'q' /* Shortcuts */
12067 && LIKELY(!Perl_isinfnan((NV)fv)) ) {
12068 /* See earlier comment about buggy Gconvert when digits,
12070 if ( c == 'g' && precis ) {
12071 STORE_LC_NUMERIC_SET_TO_NEEDED();
12072 SNPRINTF_G(fv, PL_efloatbuf, PL_efloatsize, precis);
12073 /* May return an empty string for digits==0 */
12074 if (*PL_efloatbuf) {
12075 elen = strlen(PL_efloatbuf);
12076 goto float_converted;
12078 } else if ( c == 'f' && !precis ) {
12079 if ((eptr = F0convert(fv, ebuf + sizeof ebuf, &elen)))
12084 if (UNLIKELY(hexfp)) {
12085 /* Hexadecimal floating point. */
12086 char* p = PL_efloatbuf;
12087 U8 vhex[VHEX_SIZE];
12088 U8* v = vhex; /* working pointer to vhex */
12089 U8* vend; /* pointer to one beyond last digit of vhex */
12090 U8* vfnz = NULL; /* first non-zero */
12091 const bool lower = (c == 'a');
12092 /* At output the values of vhex (up to vend) will
12093 * be mapped through the xdig to get the actual
12094 * human-readable xdigits. */
12095 const char* xdig = PL_hexdigit;
12096 int zerotail = 0; /* how many extra zeros to append */
12097 int exponent = 0; /* exponent of the floating point input */
12099 /* XXX: denormals, NaN, Inf.
12101 * For example with denormals, (assuming the vanilla
12102 * 64-bit double): the exponent is zero. 1xp-1074 is
12103 * the smallest denormal and the smallest double, it
12104 * should be output as 0x0.0000000000001p-1022 to
12105 * match its internal structure. */
12107 /* Note: fv can be (and often is) long double.
12108 * Here it is explicitly cast to NV. */
12109 vend = S_hextract(aTHX_ (NV)fv, &exponent, vhex, NULL);
12110 S_hextract(aTHX_ (NV)fv, &exponent, vhex, vend);
12112 #if NVSIZE > DOUBLESIZE
12113 # ifdef LONGDOUBLE_X86_80_BIT
12130 xdig += 16; /* Use uppercase hex. */
12133 /* Find the first non-zero xdigit. */
12134 for (v = vhex; v < vend; v++) {
12142 U8* vlnz = NULL; /* The last non-zero. */
12144 /* Find the last non-zero xdigit. */
12145 for (v = vend - 1; v >= vhex; v--) {
12152 #if NVSIZE == DOUBLESIZE
12158 v = vhex + precis + 1;
12160 /* Round away from zero: if the tail
12161 * beyond the precis xdigits is equal to
12162 * or greater than 0x8000... */
12163 bool round = *v > 0x8;
12164 if (!round && *v == 0x8) {
12165 for (v++; v < vend; v++) {
12173 for (v = vhex + precis; v >= vhex; v--) {
12180 /* If the carry goes all the way to
12181 * the front, we need to output
12182 * a single '1'. This goes against
12183 * the "xdigit and then radix"
12184 * but since this is "cannot happen"
12185 * category, that is probably good. */
12190 /* The new effective "last non zero". */
12191 vlnz = vhex + precis;
12194 zerotail = precis - (vlnz - vhex);
12201 /* The radix is always output after the first
12202 * non-zero xdigit, or if alt. */
12203 if (vfnz < vlnz || alt) {
12204 #ifndef USE_LOCALE_NUMERIC
12207 STORE_LC_NUMERIC_SET_TO_NEEDED();
12208 if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC)) {
12210 const char* r = SvPV(PL_numeric_radix_sv, n);
12211 Copy(r, p, n, char);
12217 RESTORE_LC_NUMERIC();
12232 elen = p - PL_efloatbuf;
12233 elen += my_snprintf(p, PL_efloatsize - elen,
12234 "%c%+d", lower ? 'p' : 'P',
12237 if (elen < width) {
12239 /* Pad the back with spaces. */
12240 memset(PL_efloatbuf + elen, ' ', width - elen);
12242 else if (fill == '0') {
12243 /* Insert the zeros between the "0x" and
12244 * the digits, otherwise we end up with
12246 STRLEN nzero = width - elen;
12247 char* zerox = PL_efloatbuf + 2;
12248 Move(zerox, zerox + nzero, elen - 2, char);
12249 memset(zerox, fill, nzero);
12252 /* Move it to the right. */
12253 Move(PL_efloatbuf, PL_efloatbuf + width - elen,
12255 /* Pad the front with spaces. */
12256 memset(PL_efloatbuf, ' ', width - elen);
12262 elen = S_infnan_2pv(fv, PL_efloatbuf, PL_efloatsize);
12265 char *ptr = ebuf + sizeof ebuf;
12268 /* FIXME: what to do if HAS_LONG_DOUBLE but not PERL_PRIfldbl? */
12269 #if defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
12270 /* Note that this is HAS_LONG_DOUBLE and PERL_PRIfldbl,
12271 * not USE_LONG_DOUBLE and NVff. In other words,
12272 * this needs to work without USE_LONG_DOUBLE. */
12273 if (intsize == 'q') {
12274 /* Copy the one or more characters in a long double
12275 * format before the 'base' ([efgEFG]) character to
12276 * the format string. */
12277 #ifdef USE_QUADMATH
12280 static char const ldblf[] = PERL_PRIfldbl;
12281 char const *p = ldblf + sizeof(ldblf) - 3;
12282 while (p >= ldblf) { *--ptr = *p--; }
12288 do { *--ptr = '0' + (base % 10); } while (base /= 10);
12293 do { *--ptr = '0' + (base % 10); } while (base /= 10);
12305 /* No taint. Otherwise we are in the strange situation
12306 * where printf() taints but print($float) doesn't.
12309 STORE_LC_NUMERIC_SET_TO_NEEDED();
12311 /* hopefully the above makes ptr a very constrained format
12312 * that is safe to use, even though it's not literal */
12313 GCC_DIAG_IGNORE(-Wformat-nonliteral);
12314 #ifdef USE_QUADMATH
12316 const char* qfmt = quadmath_format_single(ptr);
12318 Perl_croak_nocontext("panic: quadmath invalid format \"%s\"", ptr);
12319 elen = quadmath_snprintf(PL_efloatbuf, PL_efloatsize,
12321 if ((IV)elen == -1)
12322 Perl_croak_nocontext("panic: quadmath_snprintf failed, format \"%s|'", qfmt);
12326 #elif defined(HAS_LONG_DOUBLE)
12327 elen = ((intsize == 'q')
12328 ? my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv)
12329 : my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, (double)fv));
12331 elen = my_sprintf(PL_efloatbuf, ptr, fv);
12337 eptr = PL_efloatbuf;
12338 assert((IV)elen > 0); /* here zero elen is bad */
12340 #ifdef USE_LOCALE_NUMERIC
12341 /* If the decimal point character in the string is UTF-8, make the
12343 if (PL_numeric_radix_sv && SvUTF8(PL_numeric_radix_sv)
12344 && instr(eptr, SvPVX_const(PL_numeric_radix_sv)))
12357 i = SvCUR(sv) - origlen;
12360 case 'c': *(va_arg(*args, char*)) = i; break;
12361 case 'h': *(va_arg(*args, short*)) = i; break;
12362 default: *(va_arg(*args, int*)) = i; break;
12363 case 'l': *(va_arg(*args, long*)) = i; break;
12364 case 'V': *(va_arg(*args, IV*)) = i; break;
12365 case 'z': *(va_arg(*args, SSize_t*)) = i; break;
12366 #ifdef HAS_PTRDIFF_T
12367 case 't': *(va_arg(*args, ptrdiff_t*)) = i; break;
12370 case 'j': *(va_arg(*args, intmax_t*)) = i; break;
12374 *(va_arg(*args, Quad_t*)) = i; break;
12381 sv_setuv_mg(argsv, has_utf8 ? (UV)sv_len_utf8(sv) : (UV)i);
12382 continue; /* not "break" */
12389 && (PL_op->op_type == OP_PRTF || PL_op->op_type == OP_SPRINTF)
12390 && ckWARN(WARN_PRINTF))
12392 SV * const msg = sv_newmortal();
12393 Perl_sv_setpvf(aTHX_ msg, "Invalid conversion in %sprintf: ",
12394 (PL_op->op_type == OP_PRTF) ? "" : "s");
12395 if (fmtstart < patend) {
12396 const char * const fmtend = q < patend ? q : patend;
12398 sv_catpvs(msg, "\"%");
12399 for (f = fmtstart; f < fmtend; f++) {
12401 sv_catpvn_nomg(msg, f, 1);
12403 Perl_sv_catpvf(aTHX_ msg,
12404 "\\%03"UVof, (UV)*f & 0xFF);
12407 sv_catpvs(msg, "\"");
12409 sv_catpvs(msg, "end of string");
12411 Perl_warner(aTHX_ packWARN(WARN_PRINTF), "%"SVf, SVfARG(msg)); /* yes, this is reentrant */
12414 /* output mangled stuff ... */
12420 /* ... right here, because formatting flags should not apply */
12421 SvGROW(sv, SvCUR(sv) + elen + 1);
12423 Copy(eptr, p, elen, char);
12426 SvCUR_set(sv, p - SvPVX_const(sv));
12428 continue; /* not "break" */
12431 if (is_utf8 != has_utf8) {
12434 sv_utf8_upgrade(sv);
12437 const STRLEN old_elen = elen;
12438 SV * const nsv = newSVpvn_flags(eptr, elen, SVs_TEMP);
12439 sv_utf8_upgrade(nsv);
12440 eptr = SvPVX_const(nsv);
12443 if (width) { /* fudge width (can't fudge elen) */
12444 width += elen - old_elen;
12450 assert((IV)elen >= 0); /* here zero elen is fine */
12451 have = esignlen + zeros + elen;
12453 croak_memory_wrap();
12455 need = (have > width ? have : width);
12458 if (need >= (((STRLEN)~0) - SvCUR(sv) - dotstrlen - 1))
12459 croak_memory_wrap();
12460 SvGROW(sv, SvCUR(sv) + need + dotstrlen + 1);
12462 if (esignlen && fill == '0') {
12464 for (i = 0; i < (int)esignlen; i++)
12465 *p++ = esignbuf[i];
12467 if (gap && !left) {
12468 memset(p, fill, gap);
12471 if (esignlen && fill != '0') {
12473 for (i = 0; i < (int)esignlen; i++)
12474 *p++ = esignbuf[i];
12478 for (i = zeros; i; i--)
12482 Copy(eptr, p, elen, char);
12486 memset(p, ' ', gap);
12491 Copy(dotstr, p, dotstrlen, char);
12495 vectorize = FALSE; /* done iterating over vecstr */
12502 SvCUR_set(sv, p - SvPVX_const(sv));
12509 /* Now that we've consumed all our printf format arguments (svix)
12510 * do we have things left on the stack that we didn't use?
12512 if (!no_redundant_warning && svmax >= svix + 1 && ckWARN(WARN_REDUNDANT)) {
12513 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
12514 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
12519 RESTORE_LC_NUMERIC(); /* Done outside loop, so don't have to save/restore
12523 /* =========================================================================
12525 =head1 Cloning an interpreter
12529 All the macros and functions in this section are for the private use of
12530 the main function, perl_clone().
12532 The foo_dup() functions make an exact copy of an existing foo thingy.
12533 During the course of a cloning, a hash table is used to map old addresses
12534 to new addresses. The table is created and manipulated with the
12535 ptr_table_* functions.
12537 * =========================================================================*/
12540 #if defined(USE_ITHREADS)
12542 /* XXX Remove this so it doesn't have to go thru the macro and return for nothing */
12543 #ifndef GpREFCNT_inc
12544 # define GpREFCNT_inc(gp) ((gp) ? (++(gp)->gp_refcnt, (gp)) : (GP*)NULL)
12548 /* Certain cases in Perl_ss_dup have been merged, by relying on the fact
12549 that currently av_dup, gv_dup and hv_dup are the same as sv_dup.
12550 If this changes, please unmerge ss_dup.
12551 Likewise, sv_dup_inc_multiple() relies on this fact. */
12552 #define sv_dup_inc_NN(s,t) SvREFCNT_inc_NN(sv_dup_inc(s,t))
12553 #define av_dup(s,t) MUTABLE_AV(sv_dup((const SV *)s,t))
12554 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
12555 #define hv_dup(s,t) MUTABLE_HV(sv_dup((const SV *)s,t))
12556 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
12557 #define cv_dup(s,t) MUTABLE_CV(sv_dup((const SV *)s,t))
12558 #define cv_dup_inc(s,t) MUTABLE_CV(sv_dup_inc((const SV *)s,t))
12559 #define io_dup(s,t) MUTABLE_IO(sv_dup((const SV *)s,t))
12560 #define io_dup_inc(s,t) MUTABLE_IO(sv_dup_inc((const SV *)s,t))
12561 #define gv_dup(s,t) MUTABLE_GV(sv_dup((const SV *)s,t))
12562 #define gv_dup_inc(s,t) MUTABLE_GV(sv_dup_inc((const SV *)s,t))
12563 #define SAVEPV(p) ((p) ? savepv(p) : NULL)
12564 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
12566 /* clone a parser */
12569 Perl_parser_dup(pTHX_ const yy_parser *const proto, CLONE_PARAMS *const param)
12573 PERL_ARGS_ASSERT_PARSER_DUP;
12578 /* look for it in the table first */
12579 parser = (yy_parser *)ptr_table_fetch(PL_ptr_table, proto);
12583 /* create anew and remember what it is */
12584 Newxz(parser, 1, yy_parser);
12585 ptr_table_store(PL_ptr_table, proto, parser);
12587 /* XXX these not yet duped */
12588 parser->old_parser = NULL;
12589 parser->stack = NULL;
12591 parser->stack_size = 0;
12592 /* XXX parser->stack->state = 0; */
12594 /* XXX eventually, just Copy() most of the parser struct ? */
12596 parser->lex_brackets = proto->lex_brackets;
12597 parser->lex_casemods = proto->lex_casemods;
12598 parser->lex_brackstack = savepvn(proto->lex_brackstack,
12599 (proto->lex_brackets < 120 ? 120 : proto->lex_brackets));
12600 parser->lex_casestack = savepvn(proto->lex_casestack,
12601 (proto->lex_casemods < 12 ? 12 : proto->lex_casemods));
12602 parser->lex_defer = proto->lex_defer;
12603 parser->lex_dojoin = proto->lex_dojoin;
12604 parser->lex_formbrack = proto->lex_formbrack;
12605 parser->lex_inpat = proto->lex_inpat;
12606 parser->lex_inwhat = proto->lex_inwhat;
12607 parser->lex_op = proto->lex_op;
12608 parser->lex_repl = sv_dup_inc(proto->lex_repl, param);
12609 parser->lex_starts = proto->lex_starts;
12610 parser->lex_stuff = sv_dup_inc(proto->lex_stuff, param);
12611 parser->multi_close = proto->multi_close;
12612 parser->multi_open = proto->multi_open;
12613 parser->multi_start = proto->multi_start;
12614 parser->multi_end = proto->multi_end;
12615 parser->preambled = proto->preambled;
12616 parser->sublex_info = proto->sublex_info; /* XXX not quite right */
12617 parser->linestr = sv_dup_inc(proto->linestr, param);
12618 parser->expect = proto->expect;
12619 parser->copline = proto->copline;
12620 parser->last_lop_op = proto->last_lop_op;
12621 parser->lex_state = proto->lex_state;
12622 parser->rsfp = fp_dup(proto->rsfp, '<', param);
12623 /* rsfp_filters entries have fake IoDIRP() */
12624 parser->rsfp_filters= av_dup_inc(proto->rsfp_filters, param);
12625 parser->in_my = proto->in_my;
12626 parser->in_my_stash = hv_dup(proto->in_my_stash, param);
12627 parser->error_count = proto->error_count;
12630 parser->linestr = sv_dup_inc(proto->linestr, param);
12633 char * const ols = SvPVX(proto->linestr);
12634 char * const ls = SvPVX(parser->linestr);
12636 parser->bufptr = ls + (proto->bufptr >= ols ?
12637 proto->bufptr - ols : 0);
12638 parser->oldbufptr = ls + (proto->oldbufptr >= ols ?
12639 proto->oldbufptr - ols : 0);
12640 parser->oldoldbufptr= ls + (proto->oldoldbufptr >= ols ?
12641 proto->oldoldbufptr - ols : 0);
12642 parser->linestart = ls + (proto->linestart >= ols ?
12643 proto->linestart - ols : 0);
12644 parser->last_uni = ls + (proto->last_uni >= ols ?
12645 proto->last_uni - ols : 0);
12646 parser->last_lop = ls + (proto->last_lop >= ols ?
12647 proto->last_lop - ols : 0);
12649 parser->bufend = ls + SvCUR(parser->linestr);
12652 Copy(proto->tokenbuf, parser->tokenbuf, 256, char);
12655 Copy(proto->nextval, parser->nextval, 5, YYSTYPE);
12656 Copy(proto->nexttype, parser->nexttype, 5, I32);
12657 parser->nexttoke = proto->nexttoke;
12659 /* XXX should clone saved_curcop here, but we aren't passed
12660 * proto_perl; so do it in perl_clone_using instead */
12666 /* duplicate a file handle */
12669 Perl_fp_dup(pTHX_ PerlIO *const fp, const char type, CLONE_PARAMS *const param)
12673 PERL_ARGS_ASSERT_FP_DUP;
12674 PERL_UNUSED_ARG(type);
12677 return (PerlIO*)NULL;
12679 /* look for it in the table first */
12680 ret = (PerlIO*)ptr_table_fetch(PL_ptr_table, fp);
12684 /* create anew and remember what it is */
12685 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE);
12686 ptr_table_store(PL_ptr_table, fp, ret);
12690 /* duplicate a directory handle */
12693 Perl_dirp_dup(pTHX_ DIR *const dp, CLONE_PARAMS *const param)
12697 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
12699 const Direntry_t *dirent;
12700 char smallbuf[256];
12706 PERL_UNUSED_CONTEXT;
12707 PERL_ARGS_ASSERT_DIRP_DUP;
12712 /* look for it in the table first */
12713 ret = (DIR*)ptr_table_fetch(PL_ptr_table, dp);
12717 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
12719 PERL_UNUSED_ARG(param);
12723 /* open the current directory (so we can switch back) */
12724 if (!(pwd = PerlDir_open("."))) return (DIR *)NULL;
12726 /* chdir to our dir handle and open the present working directory */
12727 if (fchdir(my_dirfd(dp)) < 0 || !(ret = PerlDir_open("."))) {
12728 PerlDir_close(pwd);
12729 return (DIR *)NULL;
12731 /* Now we should have two dir handles pointing to the same dir. */
12733 /* Be nice to the calling code and chdir back to where we were. */
12734 /* XXX If this fails, then what? */
12735 PERL_UNUSED_RESULT(fchdir(my_dirfd(pwd)));
12737 /* We have no need of the pwd handle any more. */
12738 PerlDir_close(pwd);
12741 # define d_namlen(d) (d)->d_namlen
12743 # define d_namlen(d) strlen((d)->d_name)
12745 /* Iterate once through dp, to get the file name at the current posi-
12746 tion. Then step back. */
12747 pos = PerlDir_tell(dp);
12748 if ((dirent = PerlDir_read(dp))) {
12749 len = d_namlen(dirent);
12750 if (len <= sizeof smallbuf) name = smallbuf;
12751 else Newx(name, len, char);
12752 Move(dirent->d_name, name, len, char);
12754 PerlDir_seek(dp, pos);
12756 /* Iterate through the new dir handle, till we find a file with the
12758 if (!dirent) /* just before the end */
12760 pos = PerlDir_tell(ret);
12761 if (PerlDir_read(ret)) continue; /* not there yet */
12762 PerlDir_seek(ret, pos); /* step back */
12766 const long pos0 = PerlDir_tell(ret);
12768 pos = PerlDir_tell(ret);
12769 if ((dirent = PerlDir_read(ret))) {
12770 if (len == (STRLEN)d_namlen(dirent)
12771 && memEQ(name, dirent->d_name, len)) {
12773 PerlDir_seek(ret, pos); /* step back */
12776 /* else we are not there yet; keep iterating */
12778 else { /* This is not meant to happen. The best we can do is
12779 reset the iterator to the beginning. */
12780 PerlDir_seek(ret, pos0);
12787 if (name && name != smallbuf)
12792 ret = win32_dirp_dup(dp, param);
12795 /* pop it in the pointer table */
12797 ptr_table_store(PL_ptr_table, dp, ret);
12802 /* duplicate a typeglob */
12805 Perl_gp_dup(pTHX_ GP *const gp, CLONE_PARAMS *const param)
12809 PERL_ARGS_ASSERT_GP_DUP;
12813 /* look for it in the table first */
12814 ret = (GP*)ptr_table_fetch(PL_ptr_table, gp);
12818 /* create anew and remember what it is */
12820 ptr_table_store(PL_ptr_table, gp, ret);
12823 /* ret->gp_refcnt must be 0 before any other dups are called. We're relying
12824 on Newxz() to do this for us. */
12825 ret->gp_sv = sv_dup_inc(gp->gp_sv, param);
12826 ret->gp_io = io_dup_inc(gp->gp_io, param);
12827 ret->gp_form = cv_dup_inc(gp->gp_form, param);
12828 ret->gp_av = av_dup_inc(gp->gp_av, param);
12829 ret->gp_hv = hv_dup_inc(gp->gp_hv, param);
12830 ret->gp_egv = gv_dup(gp->gp_egv, param);/* GvEGV is not refcounted */
12831 ret->gp_cv = cv_dup_inc(gp->gp_cv, param);
12832 ret->gp_cvgen = gp->gp_cvgen;
12833 ret->gp_line = gp->gp_line;
12834 ret->gp_file_hek = hek_dup(gp->gp_file_hek, param);
12838 /* duplicate a chain of magic */
12841 Perl_mg_dup(pTHX_ MAGIC *mg, CLONE_PARAMS *const param)
12843 MAGIC *mgret = NULL;
12844 MAGIC **mgprev_p = &mgret;
12846 PERL_ARGS_ASSERT_MG_DUP;
12848 for (; mg; mg = mg->mg_moremagic) {
12851 if ((param->flags & CLONEf_JOIN_IN)
12852 && mg->mg_type == PERL_MAGIC_backref)
12853 /* when joining, we let the individual SVs add themselves to
12854 * backref as needed. */
12857 Newx(nmg, 1, MAGIC);
12859 mgprev_p = &(nmg->mg_moremagic);
12861 /* There was a comment "XXX copy dynamic vtable?" but as we don't have
12862 dynamic vtables, I'm not sure why Sarathy wrote it. The comment dates
12863 from the original commit adding Perl_mg_dup() - revision 4538.
12864 Similarly there is the annotation "XXX random ptr?" next to the
12865 assignment to nmg->mg_ptr. */
12868 /* FIXME for plugins
12869 if (nmg->mg_type == PERL_MAGIC_qr) {
12870 nmg->mg_obj = MUTABLE_SV(CALLREGDUPE((REGEXP*)nmg->mg_obj, param));
12874 nmg->mg_obj = (nmg->mg_flags & MGf_REFCOUNTED)
12875 ? nmg->mg_type == PERL_MAGIC_backref
12876 /* The backref AV has its reference
12877 * count deliberately bumped by 1 */
12878 ? SvREFCNT_inc(av_dup_inc((const AV *)
12879 nmg->mg_obj, param))
12880 : sv_dup_inc(nmg->mg_obj, param)
12881 : sv_dup(nmg->mg_obj, param);
12883 if (nmg->mg_ptr && nmg->mg_type != PERL_MAGIC_regex_global) {
12884 if (nmg->mg_len > 0) {
12885 nmg->mg_ptr = SAVEPVN(nmg->mg_ptr, nmg->mg_len);
12886 if (nmg->mg_type == PERL_MAGIC_overload_table &&
12887 AMT_AMAGIC((AMT*)nmg->mg_ptr))
12889 AMT * const namtp = (AMT*)nmg->mg_ptr;
12890 sv_dup_inc_multiple((SV**)(namtp->table),
12891 (SV**)(namtp->table), NofAMmeth, param);
12894 else if (nmg->mg_len == HEf_SVKEY)
12895 nmg->mg_ptr = (char*)sv_dup_inc((const SV *)nmg->mg_ptr, param);
12897 if ((nmg->mg_flags & MGf_DUP) && nmg->mg_virtual && nmg->mg_virtual->svt_dup) {
12898 nmg->mg_virtual->svt_dup(aTHX_ nmg, param);
12904 #endif /* USE_ITHREADS */
12906 struct ptr_tbl_arena {
12907 struct ptr_tbl_arena *next;
12908 struct ptr_tbl_ent array[1023/3]; /* as ptr_tbl_ent has 3 pointers. */
12911 /* create a new pointer-mapping table */
12914 Perl_ptr_table_new(pTHX)
12917 PERL_UNUSED_CONTEXT;
12919 Newx(tbl, 1, PTR_TBL_t);
12920 tbl->tbl_max = 511;
12921 tbl->tbl_items = 0;
12922 tbl->tbl_arena = NULL;
12923 tbl->tbl_arena_next = NULL;
12924 tbl->tbl_arena_end = NULL;
12925 Newxz(tbl->tbl_ary, tbl->tbl_max + 1, PTR_TBL_ENT_t*);
12929 #define PTR_TABLE_HASH(ptr) \
12930 ((PTR2UV(ptr) >> 3) ^ (PTR2UV(ptr) >> (3 + 7)) ^ (PTR2UV(ptr) >> (3 + 17)))
12932 /* map an existing pointer using a table */
12934 STATIC PTR_TBL_ENT_t *
12935 S_ptr_table_find(PTR_TBL_t *const tbl, const void *const sv)
12937 PTR_TBL_ENT_t *tblent;
12938 const UV hash = PTR_TABLE_HASH(sv);
12940 PERL_ARGS_ASSERT_PTR_TABLE_FIND;
12942 tblent = tbl->tbl_ary[hash & tbl->tbl_max];
12943 for (; tblent; tblent = tblent->next) {
12944 if (tblent->oldval == sv)
12951 Perl_ptr_table_fetch(pTHX_ PTR_TBL_t *const tbl, const void *const sv)
12953 PTR_TBL_ENT_t const *const tblent = ptr_table_find(tbl, sv);
12955 PERL_ARGS_ASSERT_PTR_TABLE_FETCH;
12956 PERL_UNUSED_CONTEXT;
12958 return tblent ? tblent->newval : NULL;
12961 /* add a new entry to a pointer-mapping table 'tbl'. In hash terms, 'oldsv' is
12962 * the key; 'newsv' is the value. The names "old" and "new" are specific to
12963 * the core's typical use of ptr_tables in thread cloning. */
12966 Perl_ptr_table_store(pTHX_ PTR_TBL_t *const tbl, const void *const oldsv, void *const newsv)
12968 PTR_TBL_ENT_t *tblent = ptr_table_find(tbl, oldsv);
12970 PERL_ARGS_ASSERT_PTR_TABLE_STORE;
12971 PERL_UNUSED_CONTEXT;
12974 tblent->newval = newsv;
12976 const UV entry = PTR_TABLE_HASH(oldsv) & tbl->tbl_max;
12978 if (tbl->tbl_arena_next == tbl->tbl_arena_end) {
12979 struct ptr_tbl_arena *new_arena;
12981 Newx(new_arena, 1, struct ptr_tbl_arena);
12982 new_arena->next = tbl->tbl_arena;
12983 tbl->tbl_arena = new_arena;
12984 tbl->tbl_arena_next = new_arena->array;
12985 tbl->tbl_arena_end = C_ARRAY_END(new_arena->array);
12988 tblent = tbl->tbl_arena_next++;
12990 tblent->oldval = oldsv;
12991 tblent->newval = newsv;
12992 tblent->next = tbl->tbl_ary[entry];
12993 tbl->tbl_ary[entry] = tblent;
12995 if (tblent->next && tbl->tbl_items > tbl->tbl_max)
12996 ptr_table_split(tbl);
13000 /* double the hash bucket size of an existing ptr table */
13003 Perl_ptr_table_split(pTHX_ PTR_TBL_t *const tbl)
13005 PTR_TBL_ENT_t **ary = tbl->tbl_ary;
13006 const UV oldsize = tbl->tbl_max + 1;
13007 UV newsize = oldsize * 2;
13010 PERL_ARGS_ASSERT_PTR_TABLE_SPLIT;
13011 PERL_UNUSED_CONTEXT;
13013 Renew(ary, newsize, PTR_TBL_ENT_t*);
13014 Zero(&ary[oldsize], newsize-oldsize, PTR_TBL_ENT_t*);
13015 tbl->tbl_max = --newsize;
13016 tbl->tbl_ary = ary;
13017 for (i=0; i < oldsize; i++, ary++) {
13018 PTR_TBL_ENT_t **entp = ary;
13019 PTR_TBL_ENT_t *ent = *ary;
13020 PTR_TBL_ENT_t **curentp;
13023 curentp = ary + oldsize;
13025 if ((newsize & PTR_TABLE_HASH(ent->oldval)) != i) {
13027 ent->next = *curentp;
13037 /* remove all the entries from a ptr table */
13038 /* Deprecated - will be removed post 5.14 */
13041 Perl_ptr_table_clear(pTHX_ PTR_TBL_t *const tbl)
13043 PERL_UNUSED_CONTEXT;
13044 if (tbl && tbl->tbl_items) {
13045 struct ptr_tbl_arena *arena = tbl->tbl_arena;
13047 Zero(tbl->tbl_ary, tbl->tbl_max + 1, struct ptr_tbl_ent **);
13050 struct ptr_tbl_arena *next = arena->next;
13056 tbl->tbl_items = 0;
13057 tbl->tbl_arena = NULL;
13058 tbl->tbl_arena_next = NULL;
13059 tbl->tbl_arena_end = NULL;
13063 /* clear and free a ptr table */
13066 Perl_ptr_table_free(pTHX_ PTR_TBL_t *const tbl)
13068 struct ptr_tbl_arena *arena;
13070 PERL_UNUSED_CONTEXT;
13076 arena = tbl->tbl_arena;
13079 struct ptr_tbl_arena *next = arena->next;
13085 Safefree(tbl->tbl_ary);
13089 #if defined(USE_ITHREADS)
13092 Perl_rvpv_dup(pTHX_ SV *const dstr, const SV *const sstr, CLONE_PARAMS *const param)
13094 PERL_ARGS_ASSERT_RVPV_DUP;
13096 assert(!isREGEXP(sstr));
13098 if (SvWEAKREF(sstr)) {
13099 SvRV_set(dstr, sv_dup(SvRV_const(sstr), param));
13100 if (param->flags & CLONEf_JOIN_IN) {
13101 /* if joining, we add any back references individually rather
13102 * than copying the whole backref array */
13103 Perl_sv_add_backref(aTHX_ SvRV(dstr), dstr);
13107 SvRV_set(dstr, sv_dup_inc(SvRV_const(sstr), param));
13109 else if (SvPVX_const(sstr)) {
13110 /* Has something there */
13112 /* Normal PV - clone whole allocated space */
13113 SvPV_set(dstr, SAVEPVN(SvPVX_const(sstr), SvLEN(sstr)-1));
13114 /* sstr may not be that normal, but actually copy on write.
13115 But we are a true, independent SV, so: */
13119 /* Special case - not normally malloced for some reason */
13120 if (isGV_with_GP(sstr)) {
13121 /* Don't need to do anything here. */
13123 else if ((SvIsCOW(sstr))) {
13124 /* A "shared" PV - clone it as "shared" PV */
13126 HEK_KEY(hek_dup(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)),
13130 /* Some other special case - random pointer */
13131 SvPV_set(dstr, (char *) SvPVX_const(sstr));
13136 /* Copy the NULL */
13137 SvPV_set(dstr, NULL);
13141 /* duplicate a list of SVs. source and dest may point to the same memory. */
13143 S_sv_dup_inc_multiple(pTHX_ SV *const *source, SV **dest,
13144 SSize_t items, CLONE_PARAMS *const param)
13146 PERL_ARGS_ASSERT_SV_DUP_INC_MULTIPLE;
13148 while (items-- > 0) {
13149 *dest++ = sv_dup_inc(*source++, param);
13155 /* duplicate an SV of any type (including AV, HV etc) */
13158 S_sv_dup_common(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13163 PERL_ARGS_ASSERT_SV_DUP_COMMON;
13165 if (SvTYPE(sstr) == (svtype)SVTYPEMASK) {
13166 #ifdef DEBUG_LEAKING_SCALARS_ABORT
13171 /* look for it in the table first */
13172 dstr = MUTABLE_SV(ptr_table_fetch(PL_ptr_table, sstr));
13176 if(param->flags & CLONEf_JOIN_IN) {
13177 /** We are joining here so we don't want do clone
13178 something that is bad **/
13179 if (SvTYPE(sstr) == SVt_PVHV) {
13180 const HEK * const hvname = HvNAME_HEK(sstr);
13182 /** don't clone stashes if they already exist **/
13183 dstr = MUTABLE_SV(gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
13184 HEK_UTF8(hvname) ? SVf_UTF8 : 0));
13185 ptr_table_store(PL_ptr_table, sstr, dstr);
13189 else if (SvTYPE(sstr) == SVt_PVGV && !SvFAKE(sstr)) {
13190 HV *stash = GvSTASH(sstr);
13191 const HEK * hvname;
13192 if (stash && (hvname = HvNAME_HEK(stash))) {
13193 /** don't clone GVs if they already exist **/
13195 stash = gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
13196 HEK_UTF8(hvname) ? SVf_UTF8 : 0);
13198 stash, GvNAME(sstr),
13204 if (svp && *svp && SvTYPE(*svp) == SVt_PVGV) {
13205 ptr_table_store(PL_ptr_table, sstr, *svp);
13212 /* create anew and remember what it is */
13215 #ifdef DEBUG_LEAKING_SCALARS
13216 dstr->sv_debug_optype = sstr->sv_debug_optype;
13217 dstr->sv_debug_line = sstr->sv_debug_line;
13218 dstr->sv_debug_inpad = sstr->sv_debug_inpad;
13219 dstr->sv_debug_parent = (SV*)sstr;
13220 FREE_SV_DEBUG_FILE(dstr);
13221 dstr->sv_debug_file = savesharedpv(sstr->sv_debug_file);
13224 ptr_table_store(PL_ptr_table, sstr, dstr);
13227 SvFLAGS(dstr) = SvFLAGS(sstr);
13228 SvFLAGS(dstr) &= ~SVf_OOK; /* don't propagate OOK hack */
13229 SvREFCNT(dstr) = 0; /* must be before any other dups! */
13232 if (SvANY(sstr) && PL_watch_pvx && SvPVX_const(sstr) == PL_watch_pvx)
13233 PerlIO_printf(Perl_debug_log, "watch at %p hit, found string \"%s\"\n",
13234 (void*)PL_watch_pvx, SvPVX_const(sstr));
13237 /* don't clone objects whose class has asked us not to */
13238 if (SvOBJECT(sstr) && ! (SvFLAGS(SvSTASH(sstr)) & SVphv_CLONEABLE)) {
13243 switch (SvTYPE(sstr)) {
13245 SvANY(dstr) = NULL;
13248 SvANY(dstr) = (XPVIV*)((char*)&(dstr->sv_u.svu_iv) - STRUCT_OFFSET(XPVIV, xiv_iv));
13250 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
13252 SvIV_set(dstr, SvIVX(sstr));
13256 #if NVSIZE <= IVSIZE
13257 SvANY(dstr) = (XPVNV*)((char*)&(dstr->sv_u.svu_nv) - STRUCT_OFFSET(XPVNV, xnv_u.xnv_nv));
13259 SvANY(dstr) = new_XNV();
13261 SvNV_set(dstr, SvNVX(sstr));
13265 /* These are all the types that need complex bodies allocating. */
13267 const svtype sv_type = SvTYPE(sstr);
13268 const struct body_details *const sv_type_details
13269 = bodies_by_type + sv_type;
13273 Perl_croak(aTHX_ "Bizarre SvTYPE [%" IVdf "]", (IV)SvTYPE(sstr));
13289 assert(sv_type_details->body_size);
13290 if (sv_type_details->arena) {
13291 new_body_inline(new_body, sv_type);
13293 = (void*)((char*)new_body - sv_type_details->offset);
13295 new_body = new_NOARENA(sv_type_details);
13299 SvANY(dstr) = new_body;
13302 Copy(((char*)SvANY(sstr)) + sv_type_details->offset,
13303 ((char*)SvANY(dstr)) + sv_type_details->offset,
13304 sv_type_details->copy, char);
13306 Copy(((char*)SvANY(sstr)),
13307 ((char*)SvANY(dstr)),
13308 sv_type_details->body_size + sv_type_details->offset, char);
13311 if (sv_type != SVt_PVAV && sv_type != SVt_PVHV
13312 && !isGV_with_GP(dstr)
13314 && !(sv_type == SVt_PVIO && !(IoFLAGS(dstr) & IOf_FAKE_DIRP)))
13315 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
13317 /* The Copy above means that all the source (unduplicated) pointers
13318 are now in the destination. We can check the flags and the
13319 pointers in either, but it's possible that there's less cache
13320 missing by always going for the destination.
13321 FIXME - instrument and check that assumption */
13322 if (sv_type >= SVt_PVMG) {
13323 if ((sv_type == SVt_PVMG) && SvPAD_OUR(dstr)) {
13324 SvOURSTASH_set(dstr, hv_dup_inc(SvOURSTASH(dstr), param));
13325 } else if (sv_type == SVt_PVAV && AvPAD_NAMELIST(dstr)) {
13327 } else if (SvMAGIC(dstr))
13328 SvMAGIC_set(dstr, mg_dup(SvMAGIC(dstr), param));
13329 if (SvOBJECT(dstr) && SvSTASH(dstr))
13330 SvSTASH_set(dstr, hv_dup_inc(SvSTASH(dstr), param));
13331 else SvSTASH_set(dstr, 0); /* don't copy DESTROY cache */
13334 /* The cast silences a GCC warning about unhandled types. */
13335 switch ((int)sv_type) {
13346 /* FIXME for plugins */
13347 dstr->sv_u.svu_rx = ((REGEXP *)dstr)->sv_any;
13348 re_dup_guts((REGEXP*) sstr, (REGEXP*) dstr, param);
13351 /* XXX LvTARGOFF sometimes holds PMOP* when DEBUGGING */
13352 if (LvTYPE(dstr) == 't') /* for tie: unrefcnted fake (SV**) */
13353 LvTARG(dstr) = dstr;
13354 else if (LvTYPE(dstr) == 'T') /* for tie: fake HE */
13355 LvTARG(dstr) = MUTABLE_SV(he_dup((HE*)LvTARG(dstr), 0, param));
13357 LvTARG(dstr) = sv_dup_inc(LvTARG(dstr), param);
13358 if (isREGEXP(sstr)) goto duprex;
13360 /* non-GP case already handled above */
13361 if(isGV_with_GP(sstr)) {
13362 GvNAME_HEK(dstr) = hek_dup(GvNAME_HEK(dstr), param);
13363 /* Don't call sv_add_backref here as it's going to be
13364 created as part of the magic cloning of the symbol
13365 table--unless this is during a join and the stash
13366 is not actually being cloned. */
13367 /* Danger Will Robinson - GvGP(dstr) isn't initialised
13368 at the point of this comment. */
13369 GvSTASH(dstr) = hv_dup(GvSTASH(dstr), param);
13370 if (param->flags & CLONEf_JOIN_IN)
13371 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
13372 GvGP_set(dstr, gp_dup(GvGP(sstr), param));
13373 (void)GpREFCNT_inc(GvGP(dstr));
13377 /* PL_parser->rsfp_filters entries have fake IoDIRP() */
13378 if(IoFLAGS(dstr) & IOf_FAKE_DIRP) {
13379 /* I have no idea why fake dirp (rsfps)
13380 should be treated differently but otherwise
13381 we end up with leaks -- sky*/
13382 IoTOP_GV(dstr) = gv_dup_inc(IoTOP_GV(dstr), param);
13383 IoFMT_GV(dstr) = gv_dup_inc(IoFMT_GV(dstr), param);
13384 IoBOTTOM_GV(dstr) = gv_dup_inc(IoBOTTOM_GV(dstr), param);
13386 IoTOP_GV(dstr) = gv_dup(IoTOP_GV(dstr), param);
13387 IoFMT_GV(dstr) = gv_dup(IoFMT_GV(dstr), param);
13388 IoBOTTOM_GV(dstr) = gv_dup(IoBOTTOM_GV(dstr), param);
13389 if (IoDIRP(dstr)) {
13390 IoDIRP(dstr) = dirp_dup(IoDIRP(dstr), param);
13393 /* IoDIRP(dstr) is already a copy of IoDIRP(sstr) */
13395 IoIFP(dstr) = fp_dup(IoIFP(sstr), IoTYPE(dstr), param);
13397 if (IoOFP(dstr) == IoIFP(sstr))
13398 IoOFP(dstr) = IoIFP(dstr);
13400 IoOFP(dstr) = fp_dup(IoOFP(dstr), IoTYPE(dstr), param);
13401 IoTOP_NAME(dstr) = SAVEPV(IoTOP_NAME(dstr));
13402 IoFMT_NAME(dstr) = SAVEPV(IoFMT_NAME(dstr));
13403 IoBOTTOM_NAME(dstr) = SAVEPV(IoBOTTOM_NAME(dstr));
13406 /* avoid cloning an empty array */
13407 if (AvARRAY((const AV *)sstr) && AvFILLp((const AV *)sstr) >= 0) {
13408 SV **dst_ary, **src_ary;
13409 SSize_t items = AvFILLp((const AV *)sstr) + 1;
13411 src_ary = AvARRAY((const AV *)sstr);
13412 Newxz(dst_ary, AvMAX((const AV *)sstr)+1, SV*);
13413 ptr_table_store(PL_ptr_table, src_ary, dst_ary);
13414 AvARRAY(MUTABLE_AV(dstr)) = dst_ary;
13415 AvALLOC((const AV *)dstr) = dst_ary;
13416 if (AvREAL((const AV *)sstr)) {
13417 dst_ary = sv_dup_inc_multiple(src_ary, dst_ary, items,
13421 while (items-- > 0)
13422 *dst_ary++ = sv_dup(*src_ary++, param);
13424 items = AvMAX((const AV *)sstr) - AvFILLp((const AV *)sstr);
13425 while (items-- > 0) {
13426 *dst_ary++ = &PL_sv_undef;
13430 AvARRAY(MUTABLE_AV(dstr)) = NULL;
13431 AvALLOC((const AV *)dstr) = (SV**)NULL;
13432 AvMAX( (const AV *)dstr) = -1;
13433 AvFILLp((const AV *)dstr) = -1;
13437 if (HvARRAY((const HV *)sstr)) {
13439 const bool sharekeys = !!HvSHAREKEYS(sstr);
13440 XPVHV * const dxhv = (XPVHV*)SvANY(dstr);
13441 XPVHV * const sxhv = (XPVHV*)SvANY(sstr);
13443 Newx(darray, PERL_HV_ARRAY_ALLOC_BYTES(dxhv->xhv_max+1)
13444 + (SvOOK(sstr) ? sizeof(struct xpvhv_aux) : 0),
13446 HvARRAY(dstr) = (HE**)darray;
13447 while (i <= sxhv->xhv_max) {
13448 const HE * const source = HvARRAY(sstr)[i];
13449 HvARRAY(dstr)[i] = source
13450 ? he_dup(source, sharekeys, param) : 0;
13454 const struct xpvhv_aux * const saux = HvAUX(sstr);
13455 struct xpvhv_aux * const daux = HvAUX(dstr);
13456 /* This flag isn't copied. */
13459 if (saux->xhv_name_count) {
13460 HEK ** const sname = saux->xhv_name_u.xhvnameu_names;
13462 = saux->xhv_name_count < 0
13463 ? -saux->xhv_name_count
13464 : saux->xhv_name_count;
13465 HEK **shekp = sname + count;
13467 Newx(daux->xhv_name_u.xhvnameu_names, count, HEK *);
13468 dhekp = daux->xhv_name_u.xhvnameu_names + count;
13469 while (shekp-- > sname) {
13471 *dhekp = hek_dup(*shekp, param);
13475 daux->xhv_name_u.xhvnameu_name
13476 = hek_dup(saux->xhv_name_u.xhvnameu_name,
13479 daux->xhv_name_count = saux->xhv_name_count;
13481 daux->xhv_fill_lazy = saux->xhv_fill_lazy;
13482 daux->xhv_aux_flags = saux->xhv_aux_flags;
13483 #ifdef PERL_HASH_RANDOMIZE_KEYS
13484 daux->xhv_rand = saux->xhv_rand;
13485 daux->xhv_last_rand = saux->xhv_last_rand;
13487 daux->xhv_riter = saux->xhv_riter;
13488 daux->xhv_eiter = saux->xhv_eiter
13489 ? he_dup(saux->xhv_eiter,
13490 cBOOL(HvSHAREKEYS(sstr)), param) : 0;
13491 /* backref array needs refcnt=2; see sv_add_backref */
13492 daux->xhv_backreferences =
13493 (param->flags & CLONEf_JOIN_IN)
13494 /* when joining, we let the individual GVs and
13495 * CVs add themselves to backref as
13496 * needed. This avoids pulling in stuff
13497 * that isn't required, and simplifies the
13498 * case where stashes aren't cloned back
13499 * if they already exist in the parent
13502 : saux->xhv_backreferences
13503 ? (SvTYPE(saux->xhv_backreferences) == SVt_PVAV)
13504 ? MUTABLE_AV(SvREFCNT_inc(
13505 sv_dup_inc((const SV *)
13506 saux->xhv_backreferences, param)))
13507 : MUTABLE_AV(sv_dup((const SV *)
13508 saux->xhv_backreferences, param))
13511 daux->xhv_mro_meta = saux->xhv_mro_meta
13512 ? mro_meta_dup(saux->xhv_mro_meta, param)
13515 /* Record stashes for possible cloning in Perl_clone(). */
13517 av_push(param->stashes, dstr);
13521 HvARRAY(MUTABLE_HV(dstr)) = NULL;
13524 if (!(param->flags & CLONEf_COPY_STACKS)) {
13529 /* NOTE: not refcounted */
13530 SvANY(MUTABLE_CV(dstr))->xcv_stash =
13531 hv_dup(CvSTASH(dstr), param);
13532 if ((param->flags & CLONEf_JOIN_IN) && CvSTASH(dstr))
13533 Perl_sv_add_backref(aTHX_ MUTABLE_SV(CvSTASH(dstr)), dstr);
13534 if (!CvISXSUB(dstr)) {
13536 CvROOT(dstr) = OpREFCNT_inc(CvROOT(dstr));
13538 CvSLABBED_off(dstr);
13539 } else if (CvCONST(dstr)) {
13540 CvXSUBANY(dstr).any_ptr =
13541 sv_dup_inc((const SV *)CvXSUBANY(dstr).any_ptr, param);
13543 assert(!CvSLABBED(dstr));
13544 if (CvDYNFILE(dstr)) CvFILE(dstr) = SAVEPV(CvFILE(dstr));
13546 SvANY((CV *)dstr)->xcv_gv_u.xcv_hek =
13547 hek_dup(CvNAME_HEK((CV *)sstr), param);
13548 /* don't dup if copying back - CvGV isn't refcounted, so the
13549 * duped GV may never be freed. A bit of a hack! DAPM */
13551 SvANY(MUTABLE_CV(dstr))->xcv_gv_u.xcv_gv =
13553 ? gv_dup_inc(CvGV(sstr), param)
13554 : (param->flags & CLONEf_JOIN_IN)
13556 : gv_dup(CvGV(sstr), param);
13558 CvPADLIST(dstr) = padlist_dup(CvPADLIST(sstr), param);
13560 CvWEAKOUTSIDE(sstr)
13561 ? cv_dup( CvOUTSIDE(dstr), param)
13562 : cv_dup_inc(CvOUTSIDE(dstr), param);
13572 Perl_sv_dup_inc(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13574 PERL_ARGS_ASSERT_SV_DUP_INC;
13575 return sstr ? SvREFCNT_inc(sv_dup_common(sstr, param)) : NULL;
13579 Perl_sv_dup(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13581 SV *dstr = sstr ? sv_dup_common(sstr, param) : NULL;
13582 PERL_ARGS_ASSERT_SV_DUP;
13584 /* Track every SV that (at least initially) had a reference count of 0.
13585 We need to do this by holding an actual reference to it in this array.
13586 If we attempt to cheat, turn AvREAL_off(), and store only pointers
13587 (akin to the stashes hash, and the perl stack), we come unstuck if
13588 a weak reference (or other SV legitimately SvREFCNT() == 0 for this
13589 thread) is manipulated in a CLONE method, because CLONE runs before the
13590 unreferenced array is walked to find SVs still with SvREFCNT() == 0
13591 (and fix things up by giving each a reference via the temps stack).
13592 Instead, during CLONE, if the 0-referenced SV has SvREFCNT_inc() and
13593 then SvREFCNT_dec(), it will be cleaned up (and added to the free list)
13594 before the walk of unreferenced happens and a reference to that is SV
13595 added to the temps stack. At which point we have the same SV considered
13596 to be in use, and free to be re-used. Not good.
13598 if (dstr && !(param->flags & CLONEf_COPY_STACKS) && !SvREFCNT(dstr)) {
13599 assert(param->unreferenced);
13600 av_push(param->unreferenced, SvREFCNT_inc(dstr));
13606 /* duplicate a context */
13609 Perl_cx_dup(pTHX_ PERL_CONTEXT *cxs, I32 ix, I32 max, CLONE_PARAMS* param)
13611 PERL_CONTEXT *ncxs;
13613 PERL_ARGS_ASSERT_CX_DUP;
13616 return (PERL_CONTEXT*)NULL;
13618 /* look for it in the table first */
13619 ncxs = (PERL_CONTEXT*)ptr_table_fetch(PL_ptr_table, cxs);
13623 /* create anew and remember what it is */
13624 Newx(ncxs, max + 1, PERL_CONTEXT);
13625 ptr_table_store(PL_ptr_table, cxs, ncxs);
13626 Copy(cxs, ncxs, max + 1, PERL_CONTEXT);
13629 PERL_CONTEXT * const ncx = &ncxs[ix];
13630 if (CxTYPE(ncx) == CXt_SUBST) {
13631 Perl_croak(aTHX_ "Cloning substitution context is unimplemented");
13634 ncx->blk_oldcop = (COP*)any_dup(ncx->blk_oldcop, param->proto_perl);
13635 switch (CxTYPE(ncx)) {
13637 ncx->blk_sub.cv = (ncx->blk_sub.olddepth == 0
13638 ? cv_dup_inc(ncx->blk_sub.cv, param)
13639 : cv_dup(ncx->blk_sub.cv,param));
13640 if(CxHASARGS(ncx)){
13641 ncx->blk_sub.argarray = av_dup_inc(ncx->blk_sub.argarray,param);
13642 ncx->blk_sub.savearray = av_dup_inc(ncx->blk_sub.savearray,param);
13644 ncx->blk_sub.argarray = NULL;
13645 ncx->blk_sub.savearray = NULL;
13647 ncx->blk_sub.oldcomppad = (PAD*)ptr_table_fetch(PL_ptr_table,
13648 ncx->blk_sub.oldcomppad);
13651 ncx->blk_eval.old_namesv = sv_dup_inc(ncx->blk_eval.old_namesv,
13653 ncx->blk_eval.cur_text = sv_dup(ncx->blk_eval.cur_text, param);
13654 ncx->blk_eval.cv = cv_dup(ncx->blk_eval.cv, param);
13656 case CXt_LOOP_LAZYSV:
13657 ncx->blk_loop.state_u.lazysv.end
13658 = sv_dup_inc(ncx->blk_loop.state_u.lazysv.end, param);
13659 /* We are taking advantage of av_dup_inc and sv_dup_inc
13660 actually being the same function, and order equivalence of
13662 We can assert the later [but only at run time :-(] */
13663 assert ((void *) &ncx->blk_loop.state_u.ary.ary ==
13664 (void *) &ncx->blk_loop.state_u.lazysv.cur);
13666 ncx->blk_loop.state_u.ary.ary
13667 = av_dup_inc(ncx->blk_loop.state_u.ary.ary, param);
13668 case CXt_LOOP_LAZYIV:
13669 case CXt_LOOP_PLAIN:
13670 if (CxPADLOOP(ncx)) {
13671 ncx->blk_loop.itervar_u.oldcomppad
13672 = (PAD*)ptr_table_fetch(PL_ptr_table,
13673 ncx->blk_loop.itervar_u.oldcomppad);
13675 ncx->blk_loop.itervar_u.gv
13676 = gv_dup((const GV *)ncx->blk_loop.itervar_u.gv,
13681 ncx->blk_format.cv = cv_dup(ncx->blk_format.cv, param);
13682 ncx->blk_format.gv = gv_dup(ncx->blk_format.gv, param);
13683 ncx->blk_format.dfoutgv = gv_dup_inc(ncx->blk_format.dfoutgv,
13698 /* duplicate a stack info structure */
13701 Perl_si_dup(pTHX_ PERL_SI *si, CLONE_PARAMS* param)
13705 PERL_ARGS_ASSERT_SI_DUP;
13708 return (PERL_SI*)NULL;
13710 /* look for it in the table first */
13711 nsi = (PERL_SI*)ptr_table_fetch(PL_ptr_table, si);
13715 /* create anew and remember what it is */
13716 Newxz(nsi, 1, PERL_SI);
13717 ptr_table_store(PL_ptr_table, si, nsi);
13719 nsi->si_stack = av_dup_inc(si->si_stack, param);
13720 nsi->si_cxix = si->si_cxix;
13721 nsi->si_cxmax = si->si_cxmax;
13722 nsi->si_cxstack = cx_dup(si->si_cxstack, si->si_cxix, si->si_cxmax, param);
13723 nsi->si_type = si->si_type;
13724 nsi->si_prev = si_dup(si->si_prev, param);
13725 nsi->si_next = si_dup(si->si_next, param);
13726 nsi->si_markoff = si->si_markoff;
13731 #define POPINT(ss,ix) ((ss)[--(ix)].any_i32)
13732 #define TOPINT(ss,ix) ((ss)[ix].any_i32)
13733 #define POPLONG(ss,ix) ((ss)[--(ix)].any_long)
13734 #define TOPLONG(ss,ix) ((ss)[ix].any_long)
13735 #define POPIV(ss,ix) ((ss)[--(ix)].any_iv)
13736 #define TOPIV(ss,ix) ((ss)[ix].any_iv)
13737 #define POPUV(ss,ix) ((ss)[--(ix)].any_uv)
13738 #define TOPUV(ss,ix) ((ss)[ix].any_uv)
13739 #define POPBOOL(ss,ix) ((ss)[--(ix)].any_bool)
13740 #define TOPBOOL(ss,ix) ((ss)[ix].any_bool)
13741 #define POPPTR(ss,ix) ((ss)[--(ix)].any_ptr)
13742 #define TOPPTR(ss,ix) ((ss)[ix].any_ptr)
13743 #define POPDPTR(ss,ix) ((ss)[--(ix)].any_dptr)
13744 #define TOPDPTR(ss,ix) ((ss)[ix].any_dptr)
13745 #define POPDXPTR(ss,ix) ((ss)[--(ix)].any_dxptr)
13746 #define TOPDXPTR(ss,ix) ((ss)[ix].any_dxptr)
13749 #define pv_dup_inc(p) SAVEPV(p)
13750 #define pv_dup(p) SAVEPV(p)
13751 #define svp_dup_inc(p,pp) any_dup(p,pp)
13753 /* map any object to the new equivent - either something in the
13754 * ptr table, or something in the interpreter structure
13758 Perl_any_dup(pTHX_ void *v, const PerlInterpreter *proto_perl)
13762 PERL_ARGS_ASSERT_ANY_DUP;
13765 return (void*)NULL;
13767 /* look for it in the table first */
13768 ret = ptr_table_fetch(PL_ptr_table, v);
13772 /* see if it is part of the interpreter structure */
13773 if (v >= (void*)proto_perl && v < (void*)(proto_perl+1))
13774 ret = (void*)(((char*)aTHX) + (((char*)v) - (char*)proto_perl));
13782 /* duplicate the save stack */
13785 Perl_ss_dup(pTHX_ PerlInterpreter *proto_perl, CLONE_PARAMS* param)
13788 ANY * const ss = proto_perl->Isavestack;
13789 const I32 max = proto_perl->Isavestack_max;
13790 I32 ix = proto_perl->Isavestack_ix;
13803 void (*dptr) (void*);
13804 void (*dxptr) (pTHX_ void*);
13806 PERL_ARGS_ASSERT_SS_DUP;
13808 Newxz(nss, max, ANY);
13811 const UV uv = POPUV(ss,ix);
13812 const U8 type = (U8)uv & SAVE_MASK;
13814 TOPUV(nss,ix) = uv;
13816 case SAVEt_CLEARSV:
13817 case SAVEt_CLEARPADRANGE:
13819 case SAVEt_HELEM: /* hash element */
13820 sv = (const SV *)POPPTR(ss,ix);
13821 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13823 case SAVEt_ITEM: /* normal string */
13824 case SAVEt_GVSV: /* scalar slot in GV */
13825 case SAVEt_SV: /* scalar reference */
13826 sv = (const SV *)POPPTR(ss,ix);
13827 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13830 case SAVEt_MORTALIZESV:
13831 case SAVEt_READONLY_OFF:
13832 sv = (const SV *)POPPTR(ss,ix);
13833 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13835 case SAVEt_SHARED_PVREF: /* char* in shared space */
13836 c = (char*)POPPTR(ss,ix);
13837 TOPPTR(nss,ix) = savesharedpv(c);
13838 ptr = POPPTR(ss,ix);
13839 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13841 case SAVEt_GENERIC_SVREF: /* generic sv */
13842 case SAVEt_SVREF: /* scalar reference */
13843 sv = (const SV *)POPPTR(ss,ix);
13844 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13845 ptr = POPPTR(ss,ix);
13846 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
13848 case SAVEt_GVSLOT: /* any slot in GV */
13849 sv = (const SV *)POPPTR(ss,ix);
13850 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13851 ptr = POPPTR(ss,ix);
13852 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
13853 sv = (const SV *)POPPTR(ss,ix);
13854 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13856 case SAVEt_HV: /* hash reference */
13857 case SAVEt_AV: /* array reference */
13858 sv = (const SV *) POPPTR(ss,ix);
13859 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13861 case SAVEt_COMPPAD:
13863 sv = (const SV *) POPPTR(ss,ix);
13864 TOPPTR(nss,ix) = sv_dup(sv, param);
13866 case SAVEt_INT: /* int reference */
13867 ptr = POPPTR(ss,ix);
13868 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13869 intval = (int)POPINT(ss,ix);
13870 TOPINT(nss,ix) = intval;
13872 case SAVEt_LONG: /* long reference */
13873 ptr = POPPTR(ss,ix);
13874 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13875 longval = (long)POPLONG(ss,ix);
13876 TOPLONG(nss,ix) = longval;
13878 case SAVEt_I32: /* I32 reference */
13879 ptr = POPPTR(ss,ix);
13880 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13882 TOPINT(nss,ix) = i;
13884 case SAVEt_IV: /* IV reference */
13885 case SAVEt_STRLEN: /* STRLEN/size_t ref */
13886 ptr = POPPTR(ss,ix);
13887 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13889 TOPIV(nss,ix) = iv;
13891 case SAVEt_HPTR: /* HV* reference */
13892 case SAVEt_APTR: /* AV* reference */
13893 case SAVEt_SPTR: /* SV* reference */
13894 ptr = POPPTR(ss,ix);
13895 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13896 sv = (const SV *)POPPTR(ss,ix);
13897 TOPPTR(nss,ix) = sv_dup(sv, param);
13899 case SAVEt_VPTR: /* random* reference */
13900 ptr = POPPTR(ss,ix);
13901 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13903 case SAVEt_INT_SMALL:
13904 case SAVEt_I32_SMALL:
13905 case SAVEt_I16: /* I16 reference */
13906 case SAVEt_I8: /* I8 reference */
13908 ptr = POPPTR(ss,ix);
13909 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13911 case SAVEt_GENERIC_PVREF: /* generic char* */
13912 case SAVEt_PPTR: /* char* reference */
13913 ptr = POPPTR(ss,ix);
13914 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13915 c = (char*)POPPTR(ss,ix);
13916 TOPPTR(nss,ix) = pv_dup(c);
13918 case SAVEt_GP: /* scalar reference */
13919 gp = (GP*)POPPTR(ss,ix);
13920 TOPPTR(nss,ix) = gp = gp_dup(gp, param);
13921 (void)GpREFCNT_inc(gp);
13922 gv = (const GV *)POPPTR(ss,ix);
13923 TOPPTR(nss,ix) = gv_dup_inc(gv, param);
13926 ptr = POPPTR(ss,ix);
13927 if (ptr && (((OP*)ptr)->op_private & OPpREFCOUNTED)) {
13928 /* these are assumed to be refcounted properly */
13930 switch (((OP*)ptr)->op_type) {
13932 case OP_LEAVESUBLV:
13936 case OP_LEAVEWRITE:
13937 TOPPTR(nss,ix) = ptr;
13940 (void) OpREFCNT_inc(o);
13944 TOPPTR(nss,ix) = NULL;
13949 TOPPTR(nss,ix) = NULL;
13951 case SAVEt_FREECOPHH:
13952 ptr = POPPTR(ss,ix);
13953 TOPPTR(nss,ix) = cophh_copy((COPHH *)ptr);
13955 case SAVEt_ADELETE:
13956 av = (const AV *)POPPTR(ss,ix);
13957 TOPPTR(nss,ix) = av_dup_inc(av, param);
13959 TOPINT(nss,ix) = i;
13962 hv = (const HV *)POPPTR(ss,ix);
13963 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
13965 TOPINT(nss,ix) = i;
13968 c = (char*)POPPTR(ss,ix);
13969 TOPPTR(nss,ix) = pv_dup_inc(c);
13971 case SAVEt_STACK_POS: /* Position on Perl stack */
13973 TOPINT(nss,ix) = i;
13975 case SAVEt_DESTRUCTOR:
13976 ptr = POPPTR(ss,ix);
13977 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
13978 dptr = POPDPTR(ss,ix);
13979 TOPDPTR(nss,ix) = DPTR2FPTR(void (*)(void*),
13980 any_dup(FPTR2DPTR(void *, dptr),
13983 case SAVEt_DESTRUCTOR_X:
13984 ptr = POPPTR(ss,ix);
13985 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
13986 dxptr = POPDXPTR(ss,ix);
13987 TOPDXPTR(nss,ix) = DPTR2FPTR(void (*)(pTHX_ void*),
13988 any_dup(FPTR2DPTR(void *, dxptr),
13991 case SAVEt_REGCONTEXT:
13993 ix -= uv >> SAVE_TIGHT_SHIFT;
13995 case SAVEt_AELEM: /* array element */
13996 sv = (const SV *)POPPTR(ss,ix);
13997 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13999 TOPINT(nss,ix) = i;
14000 av = (const AV *)POPPTR(ss,ix);
14001 TOPPTR(nss,ix) = av_dup_inc(av, param);
14004 ptr = POPPTR(ss,ix);
14005 TOPPTR(nss,ix) = ptr;
14008 ptr = POPPTR(ss,ix);
14009 ptr = cophh_copy((COPHH*)ptr);
14010 TOPPTR(nss,ix) = ptr;
14012 TOPINT(nss,ix) = i;
14013 if (i & HINT_LOCALIZE_HH) {
14014 hv = (const HV *)POPPTR(ss,ix);
14015 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14018 case SAVEt_PADSV_AND_MORTALIZE:
14019 longval = (long)POPLONG(ss,ix);
14020 TOPLONG(nss,ix) = longval;
14021 ptr = POPPTR(ss,ix);
14022 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14023 sv = (const SV *)POPPTR(ss,ix);
14024 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14026 case SAVEt_SET_SVFLAGS:
14028 TOPINT(nss,ix) = i;
14030 TOPINT(nss,ix) = i;
14031 sv = (const SV *)POPPTR(ss,ix);
14032 TOPPTR(nss,ix) = sv_dup(sv, param);
14034 case SAVEt_COMPILE_WARNINGS:
14035 ptr = POPPTR(ss,ix);
14036 TOPPTR(nss,ix) = DUP_WARNINGS((STRLEN*)ptr);
14039 ptr = POPPTR(ss,ix);
14040 TOPPTR(nss,ix) = parser_dup((const yy_parser*)ptr, param);
14042 case SAVEt_GP_ALIASED_SV:
14043 ptr = POPPTR(ss,ix);
14044 TOPPTR(nss,ix) = gp_dup((GP *)ptr, param);
14045 ((GP *)ptr)->gp_refcnt++;
14049 "panic: ss_dup inconsistency (%"IVdf")", (IV) type);
14057 /* if sv is a stash, call $class->CLONE_SKIP(), and set the SVphv_CLONEABLE
14058 * flag to the result. This is done for each stash before cloning starts,
14059 * so we know which stashes want their objects cloned */
14062 do_mark_cloneable_stash(pTHX_ SV *const sv)
14064 const HEK * const hvname = HvNAME_HEK((const HV *)sv);
14066 GV* const cloner = gv_fetchmethod_autoload(MUTABLE_HV(sv), "CLONE_SKIP", 0);
14067 SvFLAGS(sv) |= SVphv_CLONEABLE; /* clone objects by default */
14068 if (cloner && GvCV(cloner)) {
14075 mXPUSHs(newSVhek(hvname));
14077 call_sv(MUTABLE_SV(GvCV(cloner)), G_SCALAR);
14084 SvFLAGS(sv) &= ~SVphv_CLONEABLE;
14092 =for apidoc perl_clone
14094 Create and return a new interpreter by cloning the current one.
14096 perl_clone takes these flags as parameters:
14098 CLONEf_COPY_STACKS - is used to, well, copy the stacks also,
14099 without it we only clone the data and zero the stacks,
14100 with it we copy the stacks and the new perl interpreter is
14101 ready to run at the exact same point as the previous one.
14102 The pseudo-fork code uses COPY_STACKS while the
14103 threads->create doesn't.
14105 CLONEf_KEEP_PTR_TABLE -
14106 perl_clone keeps a ptr_table with the pointer of the old
14107 variable as a key and the new variable as a value,
14108 this allows it to check if something has been cloned and not
14109 clone it again but rather just use the value and increase the
14110 refcount. If KEEP_PTR_TABLE is not set then perl_clone will kill
14111 the ptr_table using the function
14112 C<ptr_table_free(PL_ptr_table); PL_ptr_table = NULL;>,
14113 reason to keep it around is if you want to dup some of your own
14114 variable who are outside the graph perl scans, example of this
14115 code is in threads.xs create.
14117 CLONEf_CLONE_HOST -
14118 This is a win32 thing, it is ignored on unix, it tells perls
14119 win32host code (which is c++) to clone itself, this is needed on
14120 win32 if you want to run two threads at the same time,
14121 if you just want to do some stuff in a separate perl interpreter
14122 and then throw it away and return to the original one,
14123 you don't need to do anything.
14128 /* XXX the above needs expanding by someone who actually understands it ! */
14129 EXTERN_C PerlInterpreter *
14130 perl_clone_host(PerlInterpreter* proto_perl, UV flags);
14133 perl_clone(PerlInterpreter *proto_perl, UV flags)
14136 #ifdef PERL_IMPLICIT_SYS
14138 PERL_ARGS_ASSERT_PERL_CLONE;
14140 /* perlhost.h so we need to call into it
14141 to clone the host, CPerlHost should have a c interface, sky */
14143 if (flags & CLONEf_CLONE_HOST) {
14144 return perl_clone_host(proto_perl,flags);
14146 return perl_clone_using(proto_perl, flags,
14148 proto_perl->IMemShared,
14149 proto_perl->IMemParse,
14151 proto_perl->IStdIO,
14155 proto_perl->IProc);
14159 perl_clone_using(PerlInterpreter *proto_perl, UV flags,
14160 struct IPerlMem* ipM, struct IPerlMem* ipMS,
14161 struct IPerlMem* ipMP, struct IPerlEnv* ipE,
14162 struct IPerlStdIO* ipStd, struct IPerlLIO* ipLIO,
14163 struct IPerlDir* ipD, struct IPerlSock* ipS,
14164 struct IPerlProc* ipP)
14166 /* XXX many of the string copies here can be optimized if they're
14167 * constants; they need to be allocated as common memory and just
14168 * their pointers copied. */
14171 CLONE_PARAMS clone_params;
14172 CLONE_PARAMS* const param = &clone_params;
14174 PerlInterpreter * const my_perl = (PerlInterpreter*)(*ipM->pMalloc)(ipM, sizeof(PerlInterpreter));
14176 PERL_ARGS_ASSERT_PERL_CLONE_USING;
14177 #else /* !PERL_IMPLICIT_SYS */
14179 CLONE_PARAMS clone_params;
14180 CLONE_PARAMS* param = &clone_params;
14181 PerlInterpreter * const my_perl = (PerlInterpreter*)PerlMem_malloc(sizeof(PerlInterpreter));
14183 PERL_ARGS_ASSERT_PERL_CLONE;
14184 #endif /* PERL_IMPLICIT_SYS */
14186 /* for each stash, determine whether its objects should be cloned */
14187 S_visit(proto_perl, do_mark_cloneable_stash, SVt_PVHV, SVTYPEMASK);
14188 PERL_SET_THX(my_perl);
14191 PoisonNew(my_perl, 1, PerlInterpreter);
14194 PL_defstash = NULL; /* may be used by perl malloc() */
14197 PL_scopestack_name = 0;
14199 PL_savestack_ix = 0;
14200 PL_savestack_max = -1;
14201 PL_sig_pending = 0;
14203 Zero(&PL_debug_pad, 1, struct perl_debug_pad);
14204 # ifdef DEBUG_LEAKING_SCALARS
14205 PL_sv_serial = (((UV)my_perl >> 2) & 0xfff) * 1000000;
14207 #else /* !DEBUGGING */
14208 Zero(my_perl, 1, PerlInterpreter);
14209 #endif /* DEBUGGING */
14211 #ifdef PERL_IMPLICIT_SYS
14212 /* host pointers */
14214 PL_MemShared = ipMS;
14215 PL_MemParse = ipMP;
14222 #endif /* PERL_IMPLICIT_SYS */
14225 param->flags = flags;
14226 /* Nothing in the core code uses this, but we make it available to
14227 extensions (using mg_dup). */
14228 param->proto_perl = proto_perl;
14229 /* Likely nothing will use this, but it is initialised to be consistent
14230 with Perl_clone_params_new(). */
14231 param->new_perl = my_perl;
14232 param->unreferenced = NULL;
14235 INIT_TRACK_MEMPOOL(my_perl->Imemory_debug_header, my_perl);
14237 PL_body_arenas = NULL;
14238 Zero(&PL_body_roots, 1, PL_body_roots);
14242 PL_sv_arenaroot = NULL;
14244 PL_debug = proto_perl->Idebug;
14246 /* dbargs array probably holds garbage */
14249 PL_compiling = proto_perl->Icompiling;
14251 /* pseudo environmental stuff */
14252 PL_origargc = proto_perl->Iorigargc;
14253 PL_origargv = proto_perl->Iorigargv;
14255 #ifndef NO_TAINT_SUPPORT
14256 /* Set tainting stuff before PerlIO_debug can possibly get called */
14257 PL_tainting = proto_perl->Itainting;
14258 PL_taint_warn = proto_perl->Itaint_warn;
14260 PL_tainting = FALSE;
14261 PL_taint_warn = FALSE;
14264 PL_minus_c = proto_perl->Iminus_c;
14266 PL_localpatches = proto_perl->Ilocalpatches;
14267 PL_splitstr = proto_perl->Isplitstr;
14268 PL_minus_n = proto_perl->Iminus_n;
14269 PL_minus_p = proto_perl->Iminus_p;
14270 PL_minus_l = proto_perl->Iminus_l;
14271 PL_minus_a = proto_perl->Iminus_a;
14272 PL_minus_E = proto_perl->Iminus_E;
14273 PL_minus_F = proto_perl->Iminus_F;
14274 PL_doswitches = proto_perl->Idoswitches;
14275 PL_dowarn = proto_perl->Idowarn;
14276 PL_sawalias = proto_perl->Isawalias;
14277 #ifdef PERL_SAWAMPERSAND
14278 PL_sawampersand = proto_perl->Isawampersand;
14280 PL_unsafe = proto_perl->Iunsafe;
14281 PL_perldb = proto_perl->Iperldb;
14282 PL_perl_destruct_level = proto_perl->Iperl_destruct_level;
14283 PL_exit_flags = proto_perl->Iexit_flags;
14285 /* XXX time(&PL_basetime) when asked for? */
14286 PL_basetime = proto_perl->Ibasetime;
14288 PL_maxsysfd = proto_perl->Imaxsysfd;
14289 PL_statusvalue = proto_perl->Istatusvalue;
14291 PL_statusvalue_vms = proto_perl->Istatusvalue_vms;
14293 PL_statusvalue_posix = proto_perl->Istatusvalue_posix;
14296 /* RE engine related */
14297 PL_regmatch_slab = NULL;
14298 PL_reg_curpm = NULL;
14300 PL_sub_generation = proto_perl->Isub_generation;
14302 /* funky return mechanisms */
14303 PL_forkprocess = proto_perl->Iforkprocess;
14305 /* internal state */
14306 PL_maxo = proto_perl->Imaxo;
14308 PL_main_start = proto_perl->Imain_start;
14309 PL_eval_root = proto_perl->Ieval_root;
14310 PL_eval_start = proto_perl->Ieval_start;
14312 PL_filemode = proto_perl->Ifilemode;
14313 PL_lastfd = proto_perl->Ilastfd;
14314 PL_oldname = proto_perl->Ioldname; /* XXX not quite right */
14317 PL_gensym = proto_perl->Igensym;
14319 PL_laststatval = proto_perl->Ilaststatval;
14320 PL_laststype = proto_perl->Ilaststype;
14323 PL_profiledata = NULL;
14325 PL_generation = proto_perl->Igeneration;
14327 PL_in_clean_objs = proto_perl->Iin_clean_objs;
14328 PL_in_clean_all = proto_perl->Iin_clean_all;
14330 PL_delaymagic_uid = proto_perl->Idelaymagic_uid;
14331 PL_delaymagic_euid = proto_perl->Idelaymagic_euid;
14332 PL_delaymagic_gid = proto_perl->Idelaymagic_gid;
14333 PL_delaymagic_egid = proto_perl->Idelaymagic_egid;
14334 PL_nomemok = proto_perl->Inomemok;
14335 PL_an = proto_perl->Ian;
14336 PL_evalseq = proto_perl->Ievalseq;
14337 PL_origenviron = proto_perl->Iorigenviron; /* XXX not quite right */
14338 PL_origalen = proto_perl->Iorigalen;
14340 PL_sighandlerp = proto_perl->Isighandlerp;
14342 PL_runops = proto_perl->Irunops;
14344 PL_subline = proto_perl->Isubline;
14347 PL_cryptseen = proto_perl->Icryptseen;
14350 #ifdef USE_LOCALE_COLLATE
14351 PL_collation_ix = proto_perl->Icollation_ix;
14352 PL_collation_standard = proto_perl->Icollation_standard;
14353 PL_collxfrm_base = proto_perl->Icollxfrm_base;
14354 PL_collxfrm_mult = proto_perl->Icollxfrm_mult;
14355 #endif /* USE_LOCALE_COLLATE */
14357 #ifdef USE_LOCALE_NUMERIC
14358 PL_numeric_standard = proto_perl->Inumeric_standard;
14359 PL_numeric_local = proto_perl->Inumeric_local;
14360 #endif /* !USE_LOCALE_NUMERIC */
14362 /* Did the locale setup indicate UTF-8? */
14363 PL_utf8locale = proto_perl->Iutf8locale;
14364 PL_in_utf8_CTYPE_locale = proto_perl->Iin_utf8_CTYPE_locale;
14365 /* Unicode features (see perlrun/-C) */
14366 PL_unicode = proto_perl->Iunicode;
14368 /* Pre-5.8 signals control */
14369 PL_signals = proto_perl->Isignals;
14371 /* times() ticks per second */
14372 PL_clocktick = proto_perl->Iclocktick;
14374 /* Recursion stopper for PerlIO_find_layer */
14375 PL_in_load_module = proto_perl->Iin_load_module;
14377 /* sort() routine */
14378 PL_sort_RealCmp = proto_perl->Isort_RealCmp;
14380 /* Not really needed/useful since the reenrant_retint is "volatile",
14381 * but do it for consistency's sake. */
14382 PL_reentrant_retint = proto_perl->Ireentrant_retint;
14384 /* Hooks to shared SVs and locks. */
14385 PL_sharehook = proto_perl->Isharehook;
14386 PL_lockhook = proto_perl->Ilockhook;
14387 PL_unlockhook = proto_perl->Iunlockhook;
14388 PL_threadhook = proto_perl->Ithreadhook;
14389 PL_destroyhook = proto_perl->Idestroyhook;
14390 PL_signalhook = proto_perl->Isignalhook;
14392 PL_globhook = proto_perl->Iglobhook;
14395 PL_last_swash_hv = NULL; /* reinits on demand */
14396 PL_last_swash_klen = 0;
14397 PL_last_swash_key[0]= '\0';
14398 PL_last_swash_tmps = (U8*)NULL;
14399 PL_last_swash_slen = 0;
14401 PL_srand_called = proto_perl->Isrand_called;
14402 Copy(&(proto_perl->Irandom_state), &PL_random_state, 1, PL_RANDOM_STATE_TYPE);
14404 if (flags & CLONEf_COPY_STACKS) {
14405 /* next allocation will be PL_tmps_stack[PL_tmps_ix+1] */
14406 PL_tmps_ix = proto_perl->Itmps_ix;
14407 PL_tmps_max = proto_perl->Itmps_max;
14408 PL_tmps_floor = proto_perl->Itmps_floor;
14410 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
14411 * NOTE: unlike the others! */
14412 PL_scopestack_ix = proto_perl->Iscopestack_ix;
14413 PL_scopestack_max = proto_perl->Iscopestack_max;
14415 /* next SSPUSHFOO() sets PL_savestack[PL_savestack_ix]
14416 * NOTE: unlike the others! */
14417 PL_savestack_ix = proto_perl->Isavestack_ix;
14418 PL_savestack_max = proto_perl->Isavestack_max;
14421 PL_start_env = proto_perl->Istart_env; /* XXXXXX */
14422 PL_top_env = &PL_start_env;
14424 PL_op = proto_perl->Iop;
14427 PL_Xpv = (XPV*)NULL;
14428 my_perl->Ina = proto_perl->Ina;
14430 PL_statbuf = proto_perl->Istatbuf;
14431 PL_statcache = proto_perl->Istatcache;
14433 #ifndef NO_TAINT_SUPPORT
14434 PL_tainted = proto_perl->Itainted;
14436 PL_tainted = FALSE;
14438 PL_curpm = proto_perl->Icurpm; /* XXX No PMOP ref count */
14440 PL_chopset = proto_perl->Ichopset; /* XXX never deallocated */
14442 PL_restartjmpenv = proto_perl->Irestartjmpenv;
14443 PL_restartop = proto_perl->Irestartop;
14444 PL_in_eval = proto_perl->Iin_eval;
14445 PL_delaymagic = proto_perl->Idelaymagic;
14446 PL_phase = proto_perl->Iphase;
14447 PL_localizing = proto_perl->Ilocalizing;
14449 PL_hv_fetch_ent_mh = NULL;
14450 PL_modcount = proto_perl->Imodcount;
14451 PL_lastgotoprobe = NULL;
14452 PL_dumpindent = proto_perl->Idumpindent;
14454 PL_efloatbuf = NULL; /* reinits on demand */
14455 PL_efloatsize = 0; /* reinits on demand */
14459 PL_colorset = 0; /* reinits PL_colors[] */
14460 /*PL_colors[6] = {0,0,0,0,0,0};*/
14462 /* Pluggable optimizer */
14463 PL_peepp = proto_perl->Ipeepp;
14464 PL_rpeepp = proto_perl->Irpeepp;
14465 /* op_free() hook */
14466 PL_opfreehook = proto_perl->Iopfreehook;
14468 #ifdef USE_REENTRANT_API
14469 /* XXX: things like -Dm will segfault here in perlio, but doing
14470 * PERL_SET_CONTEXT(proto_perl);
14471 * breaks too many other things
14473 Perl_reentrant_init(aTHX);
14476 /* create SV map for pointer relocation */
14477 PL_ptr_table = ptr_table_new();
14479 /* initialize these special pointers as early as possible */
14481 ptr_table_store(PL_ptr_table, &proto_perl->Isv_undef, &PL_sv_undef);
14482 ptr_table_store(PL_ptr_table, &proto_perl->Isv_no, &PL_sv_no);
14483 ptr_table_store(PL_ptr_table, &proto_perl->Isv_yes, &PL_sv_yes);
14485 /* create (a non-shared!) shared string table */
14486 PL_strtab = newHV();
14487 HvSHAREKEYS_off(PL_strtab);
14488 hv_ksplit(PL_strtab, HvTOTALKEYS(proto_perl->Istrtab));
14489 ptr_table_store(PL_ptr_table, proto_perl->Istrtab, PL_strtab);
14491 Zero(PL_sv_consts, SV_CONSTS_COUNT, SV*);
14493 /* This PV will be free'd special way so must set it same way op.c does */
14494 PL_compiling.cop_file = savesharedpv(PL_compiling.cop_file);
14495 ptr_table_store(PL_ptr_table, proto_perl->Icompiling.cop_file, PL_compiling.cop_file);
14497 ptr_table_store(PL_ptr_table, &proto_perl->Icompiling, &PL_compiling);
14498 PL_compiling.cop_warnings = DUP_WARNINGS(PL_compiling.cop_warnings);
14499 CopHINTHASH_set(&PL_compiling, cophh_copy(CopHINTHASH_get(&PL_compiling)));
14500 PL_curcop = (COP*)any_dup(proto_perl->Icurcop, proto_perl);
14502 param->stashes = newAV(); /* Setup array of objects to call clone on */
14503 /* This makes no difference to the implementation, as it always pushes
14504 and shifts pointers to other SVs without changing their reference
14505 count, with the array becoming empty before it is freed. However, it
14506 makes it conceptually clear what is going on, and will avoid some
14507 work inside av.c, filling slots between AvFILL() and AvMAX() with
14508 &PL_sv_undef, and SvREFCNT_dec()ing those. */
14509 AvREAL_off(param->stashes);
14511 if (!(flags & CLONEf_COPY_STACKS)) {
14512 param->unreferenced = newAV();
14515 #ifdef PERLIO_LAYERS
14516 /* Clone PerlIO tables as soon as we can handle general xx_dup() */
14517 PerlIO_clone(aTHX_ proto_perl, param);
14520 PL_envgv = gv_dup_inc(proto_perl->Ienvgv, param);
14521 PL_incgv = gv_dup_inc(proto_perl->Iincgv, param);
14522 PL_hintgv = gv_dup_inc(proto_perl->Ihintgv, param);
14523 PL_origfilename = SAVEPV(proto_perl->Iorigfilename);
14524 PL_diehook = sv_dup_inc(proto_perl->Idiehook, param);
14525 PL_warnhook = sv_dup_inc(proto_perl->Iwarnhook, param);
14528 PL_patchlevel = sv_dup_inc(proto_perl->Ipatchlevel, param);
14529 PL_apiversion = sv_dup_inc(proto_perl->Iapiversion, param);
14530 PL_inplace = SAVEPV(proto_perl->Iinplace);
14531 PL_e_script = sv_dup_inc(proto_perl->Ie_script, param);
14533 /* magical thingies */
14535 PL_encoding = sv_dup(proto_perl->Iencoding, param);
14537 sv_setpvs(PERL_DEBUG_PAD(0), ""); /* For regex debugging. */
14538 sv_setpvs(PERL_DEBUG_PAD(1), ""); /* ext/re needs these */
14539 sv_setpvs(PERL_DEBUG_PAD(2), ""); /* even without DEBUGGING. */
14542 /* Clone the regex array */
14543 /* ORANGE FIXME for plugins, probably in the SV dup code.
14544 newSViv(PTR2IV(CALLREGDUPE(
14545 INT2PTR(REGEXP *, SvIVX(regex)), param))))
14547 PL_regex_padav = av_dup_inc(proto_perl->Iregex_padav, param);
14548 PL_regex_pad = AvARRAY(PL_regex_padav);
14550 PL_stashpadmax = proto_perl->Istashpadmax;
14551 PL_stashpadix = proto_perl->Istashpadix ;
14552 Newx(PL_stashpad, PL_stashpadmax, HV *);
14555 for (; o < PL_stashpadmax; ++o)
14556 PL_stashpad[o] = hv_dup(proto_perl->Istashpad[o], param);
14559 /* shortcuts to various I/O objects */
14560 PL_ofsgv = gv_dup_inc(proto_perl->Iofsgv, param);
14561 PL_stdingv = gv_dup(proto_perl->Istdingv, param);
14562 PL_stderrgv = gv_dup(proto_perl->Istderrgv, param);
14563 PL_defgv = gv_dup(proto_perl->Idefgv, param);
14564 PL_argvgv = gv_dup_inc(proto_perl->Iargvgv, param);
14565 PL_argvoutgv = gv_dup(proto_perl->Iargvoutgv, param);
14566 PL_argvout_stack = av_dup_inc(proto_perl->Iargvout_stack, param);
14568 /* shortcuts to regexp stuff */
14569 PL_replgv = gv_dup_inc(proto_perl->Ireplgv, param);
14571 /* shortcuts to misc objects */
14572 PL_errgv = gv_dup(proto_perl->Ierrgv, param);
14574 /* shortcuts to debugging objects */
14575 PL_DBgv = gv_dup_inc(proto_perl->IDBgv, param);
14576 PL_DBline = gv_dup_inc(proto_perl->IDBline, param);
14577 PL_DBsub = gv_dup_inc(proto_perl->IDBsub, param);
14578 PL_DBsingle = sv_dup(proto_perl->IDBsingle, param);
14579 PL_DBtrace = sv_dup(proto_perl->IDBtrace, param);
14580 PL_DBsignal = sv_dup(proto_perl->IDBsignal, param);
14582 /* symbol tables */
14583 PL_defstash = hv_dup_inc(proto_perl->Idefstash, param);
14584 PL_curstash = hv_dup_inc(proto_perl->Icurstash, param);
14585 PL_debstash = hv_dup(proto_perl->Idebstash, param);
14586 PL_globalstash = hv_dup(proto_perl->Iglobalstash, param);
14587 PL_curstname = sv_dup_inc(proto_perl->Icurstname, param);
14589 PL_beginav = av_dup_inc(proto_perl->Ibeginav, param);
14590 PL_beginav_save = av_dup_inc(proto_perl->Ibeginav_save, param);
14591 PL_checkav_save = av_dup_inc(proto_perl->Icheckav_save, param);
14592 PL_unitcheckav = av_dup_inc(proto_perl->Iunitcheckav, param);
14593 PL_unitcheckav_save = av_dup_inc(proto_perl->Iunitcheckav_save, param);
14594 PL_endav = av_dup_inc(proto_perl->Iendav, param);
14595 PL_checkav = av_dup_inc(proto_perl->Icheckav, param);
14596 PL_initav = av_dup_inc(proto_perl->Iinitav, param);
14598 PL_isarev = hv_dup_inc(proto_perl->Iisarev, param);
14600 /* subprocess state */
14601 PL_fdpid = av_dup_inc(proto_perl->Ifdpid, param);
14603 if (proto_perl->Iop_mask)
14604 PL_op_mask = SAVEPVN(proto_perl->Iop_mask, PL_maxo);
14607 /* PL_asserting = proto_perl->Iasserting; */
14609 /* current interpreter roots */
14610 PL_main_cv = cv_dup_inc(proto_perl->Imain_cv, param);
14612 PL_main_root = OpREFCNT_inc(proto_perl->Imain_root);
14615 /* runtime control stuff */
14616 PL_curcopdb = (COP*)any_dup(proto_perl->Icurcopdb, proto_perl);
14618 PL_preambleav = av_dup_inc(proto_perl->Ipreambleav, param);
14620 PL_ors_sv = sv_dup_inc(proto_perl->Iors_sv, param);
14622 /* interpreter atexit processing */
14623 PL_exitlistlen = proto_perl->Iexitlistlen;
14624 if (PL_exitlistlen) {
14625 Newx(PL_exitlist, PL_exitlistlen, PerlExitListEntry);
14626 Copy(proto_perl->Iexitlist, PL_exitlist, PL_exitlistlen, PerlExitListEntry);
14629 PL_exitlist = (PerlExitListEntry*)NULL;
14631 PL_my_cxt_size = proto_perl->Imy_cxt_size;
14632 if (PL_my_cxt_size) {
14633 Newx(PL_my_cxt_list, PL_my_cxt_size, void *);
14634 Copy(proto_perl->Imy_cxt_list, PL_my_cxt_list, PL_my_cxt_size, void *);
14635 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
14636 Newx(PL_my_cxt_keys, PL_my_cxt_size, const char *);
14637 Copy(proto_perl->Imy_cxt_keys, PL_my_cxt_keys, PL_my_cxt_size, char *);
14641 PL_my_cxt_list = (void**)NULL;
14642 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
14643 PL_my_cxt_keys = (const char**)NULL;
14646 PL_modglobal = hv_dup_inc(proto_perl->Imodglobal, param);
14647 PL_custom_op_names = hv_dup_inc(proto_perl->Icustom_op_names,param);
14648 PL_custom_op_descs = hv_dup_inc(proto_perl->Icustom_op_descs,param);
14649 PL_custom_ops = hv_dup_inc(proto_perl->Icustom_ops, param);
14651 PL_compcv = cv_dup(proto_perl->Icompcv, param);
14653 PAD_CLONE_VARS(proto_perl, param);
14655 #ifdef HAVE_INTERP_INTERN
14656 sys_intern_dup(&proto_perl->Isys_intern, &PL_sys_intern);
14659 PL_DBcv = cv_dup(proto_perl->IDBcv, param);
14661 #ifdef PERL_USES_PL_PIDSTATUS
14662 PL_pidstatus = newHV(); /* XXX flag for cloning? */
14664 PL_osname = SAVEPV(proto_perl->Iosname);
14665 PL_parser = parser_dup(proto_perl->Iparser, param);
14667 /* XXX this only works if the saved cop has already been cloned */
14668 if (proto_perl->Iparser) {
14669 PL_parser->saved_curcop = (COP*)any_dup(
14670 proto_perl->Iparser->saved_curcop,
14674 PL_subname = sv_dup_inc(proto_perl->Isubname, param);
14676 #ifdef USE_LOCALE_COLLATE
14677 PL_collation_name = SAVEPV(proto_perl->Icollation_name);
14678 #endif /* USE_LOCALE_COLLATE */
14680 #ifdef USE_LOCALE_NUMERIC
14681 PL_numeric_name = SAVEPV(proto_perl->Inumeric_name);
14682 PL_numeric_radix_sv = sv_dup_inc(proto_perl->Inumeric_radix_sv, param);
14683 #endif /* !USE_LOCALE_NUMERIC */
14685 /* Unicode inversion lists */
14686 PL_Latin1 = sv_dup_inc(proto_perl->ILatin1, param);
14687 PL_UpperLatin1 = sv_dup_inc(proto_perl->IUpperLatin1, param);
14688 PL_AboveLatin1 = sv_dup_inc(proto_perl->IAboveLatin1, param);
14689 PL_InBitmap = sv_dup_inc(proto_perl->IInBitmap, param);
14691 PL_NonL1NonFinalFold = sv_dup_inc(proto_perl->INonL1NonFinalFold, param);
14692 PL_HasMultiCharFold = sv_dup_inc(proto_perl->IHasMultiCharFold, param);
14694 /* utf8 character class swashes */
14695 for (i = 0; i < POSIX_SWASH_COUNT; i++) {
14696 PL_utf8_swash_ptrs[i] = sv_dup_inc(proto_perl->Iutf8_swash_ptrs[i], param);
14698 for (i = 0; i < POSIX_CC_COUNT; i++) {
14699 PL_XPosix_ptrs[i] = sv_dup_inc(proto_perl->IXPosix_ptrs[i], param);
14701 PL_utf8_mark = sv_dup_inc(proto_perl->Iutf8_mark, param);
14702 PL_utf8_X_regular_begin = sv_dup_inc(proto_perl->Iutf8_X_regular_begin, param);
14703 PL_utf8_X_extend = sv_dup_inc(proto_perl->Iutf8_X_extend, param);
14704 PL_utf8_toupper = sv_dup_inc(proto_perl->Iutf8_toupper, param);
14705 PL_utf8_totitle = sv_dup_inc(proto_perl->Iutf8_totitle, param);
14706 PL_utf8_tolower = sv_dup_inc(proto_perl->Iutf8_tolower, param);
14707 PL_utf8_tofold = sv_dup_inc(proto_perl->Iutf8_tofold, param);
14708 PL_utf8_idstart = sv_dup_inc(proto_perl->Iutf8_idstart, param);
14709 PL_utf8_xidstart = sv_dup_inc(proto_perl->Iutf8_xidstart, param);
14710 PL_utf8_perl_idstart = sv_dup_inc(proto_perl->Iutf8_perl_idstart, param);
14711 PL_utf8_perl_idcont = sv_dup_inc(proto_perl->Iutf8_perl_idcont, param);
14712 PL_utf8_idcont = sv_dup_inc(proto_perl->Iutf8_idcont, param);
14713 PL_utf8_xidcont = sv_dup_inc(proto_perl->Iutf8_xidcont, param);
14714 PL_utf8_foldable = sv_dup_inc(proto_perl->Iutf8_foldable, param);
14715 PL_utf8_charname_begin = sv_dup_inc(proto_perl->Iutf8_charname_begin, param);
14716 PL_utf8_charname_continue = sv_dup_inc(proto_perl->Iutf8_charname_continue, param);
14718 if (proto_perl->Ipsig_pend) {
14719 Newxz(PL_psig_pend, SIG_SIZE, int);
14722 PL_psig_pend = (int*)NULL;
14725 if (proto_perl->Ipsig_name) {
14726 Newx(PL_psig_name, 2 * SIG_SIZE, SV*);
14727 sv_dup_inc_multiple(proto_perl->Ipsig_name, PL_psig_name, 2 * SIG_SIZE,
14729 PL_psig_ptr = PL_psig_name + SIG_SIZE;
14732 PL_psig_ptr = (SV**)NULL;
14733 PL_psig_name = (SV**)NULL;
14736 if (flags & CLONEf_COPY_STACKS) {
14737 Newx(PL_tmps_stack, PL_tmps_max, SV*);
14738 sv_dup_inc_multiple(proto_perl->Itmps_stack, PL_tmps_stack,
14739 PL_tmps_ix+1, param);
14741 /* next PUSHMARK() sets *(PL_markstack_ptr+1) */
14742 i = proto_perl->Imarkstack_max - proto_perl->Imarkstack;
14743 Newxz(PL_markstack, i, I32);
14744 PL_markstack_max = PL_markstack + (proto_perl->Imarkstack_max
14745 - proto_perl->Imarkstack);
14746 PL_markstack_ptr = PL_markstack + (proto_perl->Imarkstack_ptr
14747 - proto_perl->Imarkstack);
14748 Copy(proto_perl->Imarkstack, PL_markstack,
14749 PL_markstack_ptr - PL_markstack + 1, I32);
14751 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
14752 * NOTE: unlike the others! */
14753 Newxz(PL_scopestack, PL_scopestack_max, I32);
14754 Copy(proto_perl->Iscopestack, PL_scopestack, PL_scopestack_ix, I32);
14757 Newxz(PL_scopestack_name, PL_scopestack_max, const char *);
14758 Copy(proto_perl->Iscopestack_name, PL_scopestack_name, PL_scopestack_ix, const char *);
14760 /* reset stack AV to correct length before its duped via
14761 * PL_curstackinfo */
14762 AvFILLp(proto_perl->Icurstack) =
14763 proto_perl->Istack_sp - proto_perl->Istack_base;
14765 /* NOTE: si_dup() looks at PL_markstack */
14766 PL_curstackinfo = si_dup(proto_perl->Icurstackinfo, param);
14768 /* PL_curstack = PL_curstackinfo->si_stack; */
14769 PL_curstack = av_dup(proto_perl->Icurstack, param);
14770 PL_mainstack = av_dup(proto_perl->Imainstack, param);
14772 /* next PUSHs() etc. set *(PL_stack_sp+1) */
14773 PL_stack_base = AvARRAY(PL_curstack);
14774 PL_stack_sp = PL_stack_base + (proto_perl->Istack_sp
14775 - proto_perl->Istack_base);
14776 PL_stack_max = PL_stack_base + AvMAX(PL_curstack);
14778 /*Newxz(PL_savestack, PL_savestack_max, ANY);*/
14779 PL_savestack = ss_dup(proto_perl, param);
14783 ENTER; /* perl_destruct() wants to LEAVE; */
14786 PL_statgv = gv_dup(proto_perl->Istatgv, param);
14787 PL_statname = sv_dup_inc(proto_perl->Istatname, param);
14789 PL_rs = sv_dup_inc(proto_perl->Irs, param);
14790 PL_last_in_gv = gv_dup(proto_perl->Ilast_in_gv, param);
14791 PL_defoutgv = gv_dup_inc(proto_perl->Idefoutgv, param);
14792 PL_toptarget = sv_dup_inc(proto_perl->Itoptarget, param);
14793 PL_bodytarget = sv_dup_inc(proto_perl->Ibodytarget, param);
14794 PL_formtarget = sv_dup(proto_perl->Iformtarget, param);
14796 PL_errors = sv_dup_inc(proto_perl->Ierrors, param);
14798 PL_sortcop = (OP*)any_dup(proto_perl->Isortcop, proto_perl);
14799 PL_firstgv = gv_dup_inc(proto_perl->Ifirstgv, param);
14800 PL_secondgv = gv_dup_inc(proto_perl->Isecondgv, param);
14802 PL_stashcache = newHV();
14804 PL_watchaddr = (char **) ptr_table_fetch(PL_ptr_table,
14805 proto_perl->Iwatchaddr);
14806 PL_watchok = PL_watchaddr ? * PL_watchaddr : NULL;
14807 if (PL_debug && PL_watchaddr) {
14808 PerlIO_printf(Perl_debug_log,
14809 "WATCHING: %"UVxf" cloned as %"UVxf" with value %"UVxf"\n",
14810 PTR2UV(proto_perl->Iwatchaddr), PTR2UV(PL_watchaddr),
14811 PTR2UV(PL_watchok));
14814 PL_registered_mros = hv_dup_inc(proto_perl->Iregistered_mros, param);
14815 PL_blockhooks = av_dup_inc(proto_perl->Iblockhooks, param);
14816 PL_utf8_foldclosures = hv_dup_inc(proto_perl->Iutf8_foldclosures, param);
14818 /* Call the ->CLONE method, if it exists, for each of the stashes
14819 identified by sv_dup() above.
14821 while(av_tindex(param->stashes) != -1) {
14822 HV* const stash = MUTABLE_HV(av_shift(param->stashes));
14823 GV* const cloner = gv_fetchmethod_autoload(stash, "CLONE", 0);
14824 if (cloner && GvCV(cloner)) {
14829 mXPUSHs(newSVhek(HvNAME_HEK(stash)));
14831 call_sv(MUTABLE_SV(GvCV(cloner)), G_DISCARD);
14837 if (!(flags & CLONEf_KEEP_PTR_TABLE)) {
14838 ptr_table_free(PL_ptr_table);
14839 PL_ptr_table = NULL;
14842 if (!(flags & CLONEf_COPY_STACKS)) {
14843 unreferenced_to_tmp_stack(param->unreferenced);
14846 SvREFCNT_dec(param->stashes);
14848 /* orphaned? eg threads->new inside BEGIN or use */
14849 if (PL_compcv && ! SvREFCNT(PL_compcv)) {
14850 SvREFCNT_inc_simple_void(PL_compcv);
14851 SAVEFREESV(PL_compcv);
14858 S_unreferenced_to_tmp_stack(pTHX_ AV *const unreferenced)
14860 PERL_ARGS_ASSERT_UNREFERENCED_TO_TMP_STACK;
14862 if (AvFILLp(unreferenced) > -1) {
14863 SV **svp = AvARRAY(unreferenced);
14864 SV **const last = svp + AvFILLp(unreferenced);
14868 if (SvREFCNT(*svp) == 1)
14870 } while (++svp <= last);
14872 EXTEND_MORTAL(count);
14873 svp = AvARRAY(unreferenced);
14876 if (SvREFCNT(*svp) == 1) {
14877 /* Our reference is the only one to this SV. This means that
14878 in this thread, the scalar effectively has a 0 reference.
14879 That doesn't work (cleanup never happens), so donate our
14880 reference to it onto the save stack. */
14881 PL_tmps_stack[++PL_tmps_ix] = *svp;
14883 /* As an optimisation, because we are already walking the
14884 entire array, instead of above doing either
14885 SvREFCNT_inc(*svp) or *svp = &PL_sv_undef, we can instead
14886 release our reference to the scalar, so that at the end of
14887 the array owns zero references to the scalars it happens to
14888 point to. We are effectively converting the array from
14889 AvREAL() on to AvREAL() off. This saves the av_clear()
14890 (triggered by the SvREFCNT_dec(unreferenced) below) from
14891 walking the array a second time. */
14892 SvREFCNT_dec(*svp);
14895 } while (++svp <= last);
14896 AvREAL_off(unreferenced);
14898 SvREFCNT_dec_NN(unreferenced);
14902 Perl_clone_params_del(CLONE_PARAMS *param)
14904 /* This seemingly funky ordering keeps the build with PERL_GLOBAL_STRUCT
14906 PerlInterpreter *const to = param->new_perl;
14908 PerlInterpreter *const was = PERL_GET_THX;
14910 PERL_ARGS_ASSERT_CLONE_PARAMS_DEL;
14916 SvREFCNT_dec(param->stashes);
14917 if (param->unreferenced)
14918 unreferenced_to_tmp_stack(param->unreferenced);
14928 Perl_clone_params_new(PerlInterpreter *const from, PerlInterpreter *const to)
14931 /* Need to play this game, as newAV() can call safesysmalloc(), and that
14932 does a dTHX; to get the context from thread local storage.
14933 FIXME - under PERL_CORE Newx(), Safefree() and friends should expand to
14934 a version that passes in my_perl. */
14935 PerlInterpreter *const was = PERL_GET_THX;
14936 CLONE_PARAMS *param;
14938 PERL_ARGS_ASSERT_CLONE_PARAMS_NEW;
14944 /* Given that we've set the context, we can do this unshared. */
14945 Newx(param, 1, CLONE_PARAMS);
14948 param->proto_perl = from;
14949 param->new_perl = to;
14950 param->stashes = (AV *)Perl_newSV_type(to, SVt_PVAV);
14951 AvREAL_off(param->stashes);
14952 param->unreferenced = (AV *)Perl_newSV_type(to, SVt_PVAV);
14960 #endif /* USE_ITHREADS */
14963 Perl_init_constants(pTHX)
14965 SvREFCNT(&PL_sv_undef) = SvREFCNT_IMMORTAL;
14966 SvFLAGS(&PL_sv_undef) = SVf_READONLY|SVf_PROTECT|SVt_NULL;
14967 SvANY(&PL_sv_undef) = NULL;
14969 SvANY(&PL_sv_no) = new_XPVNV();
14970 SvREFCNT(&PL_sv_no) = SvREFCNT_IMMORTAL;
14971 SvFLAGS(&PL_sv_no) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
14972 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
14975 SvANY(&PL_sv_yes) = new_XPVNV();
14976 SvREFCNT(&PL_sv_yes) = SvREFCNT_IMMORTAL;
14977 SvFLAGS(&PL_sv_yes) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
14978 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
14981 SvPV_set(&PL_sv_no, (char*)PL_No);
14982 SvCUR_set(&PL_sv_no, 0);
14983 SvLEN_set(&PL_sv_no, 0);
14984 SvIV_set(&PL_sv_no, 0);
14985 SvNV_set(&PL_sv_no, 0);
14987 SvPV_set(&PL_sv_yes, (char*)PL_Yes);
14988 SvCUR_set(&PL_sv_yes, 1);
14989 SvLEN_set(&PL_sv_yes, 0);
14990 SvIV_set(&PL_sv_yes, 1);
14991 SvNV_set(&PL_sv_yes, 1);
14995 =head1 Unicode Support
14997 =for apidoc sv_recode_to_utf8
14999 The encoding is assumed to be an Encode object, on entry the PV
15000 of the sv is assumed to be octets in that encoding, and the sv
15001 will be converted into Unicode (and UTF-8).
15003 If the sv already is UTF-8 (or if it is not POK), or if the encoding
15004 is not a reference, nothing is done to the sv. If the encoding is not
15005 an C<Encode::XS> Encoding object, bad things will happen.
15006 (See F<lib/encoding.pm> and L<Encode>.)
15008 The PV of the sv is returned.
15013 Perl_sv_recode_to_utf8(pTHX_ SV *sv, SV *encoding)
15015 PERL_ARGS_ASSERT_SV_RECODE_TO_UTF8;
15017 if (SvPOK(sv) && !SvUTF8(sv) && !IN_BYTES && SvROK(encoding)) {
15026 if (SvPADTMP(nsv)) {
15027 nsv = sv_newmortal();
15028 SvSetSV_nosteal(nsv, sv);
15036 Passing sv_yes is wrong - it needs to be or'ed set of constants
15037 for Encode::XS, while UTf-8 decode (currently) assumes a true value means
15038 remove converted chars from source.
15040 Both will default the value - let them.
15042 XPUSHs(&PL_sv_yes);
15045 call_method("decode", G_SCALAR);
15049 s = SvPV_const(uni, len);
15050 if (s != SvPVX_const(sv)) {
15051 SvGROW(sv, len + 1);
15052 Move(s, SvPVX(sv), len + 1, char);
15053 SvCUR_set(sv, len);
15058 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
15059 /* clear pos and any utf8 cache */
15060 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
15063 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
15064 magic_setutf8(sv,mg); /* clear UTF8 cache */
15069 return SvPOKp(sv) ? SvPVX(sv) : NULL;
15073 =for apidoc sv_cat_decode
15075 The encoding is assumed to be an Encode object, the PV of the ssv is
15076 assumed to be octets in that encoding and decoding the input starts
15077 from the position which (PV + *offset) pointed to. The dsv will be
15078 concatenated the decoded UTF-8 string from ssv. Decoding will terminate
15079 when the string tstr appears in decoding output or the input ends on
15080 the PV of the ssv. The value which the offset points will be modified
15081 to the last input position on the ssv.
15083 Returns TRUE if the terminator was found, else returns FALSE.
15088 Perl_sv_cat_decode(pTHX_ SV *dsv, SV *encoding,
15089 SV *ssv, int *offset, char *tstr, int tlen)
15093 PERL_ARGS_ASSERT_SV_CAT_DECODE;
15095 if (SvPOK(ssv) && SvPOK(dsv) && SvROK(encoding) && offset) {
15105 offsv = newSViv(*offset);
15107 mPUSHp(tstr, tlen);
15109 call_method("cat_decode", G_SCALAR);
15111 ret = SvTRUE(TOPs);
15112 *offset = SvIV(offsv);
15118 Perl_croak(aTHX_ "Invalid argument to sv_cat_decode");
15123 /* ---------------------------------------------------------------------
15125 * support functions for report_uninit()
15128 /* the maxiumum size of array or hash where we will scan looking
15129 * for the undefined element that triggered the warning */
15131 #define FUV_MAX_SEARCH_SIZE 1000
15133 /* Look for an entry in the hash whose value has the same SV as val;
15134 * If so, return a mortal copy of the key. */
15137 S_find_hash_subscript(pTHX_ const HV *const hv, const SV *const val)
15143 PERL_ARGS_ASSERT_FIND_HASH_SUBSCRIPT;
15145 if (!hv || SvMAGICAL(hv) || !HvARRAY(hv) ||
15146 (HvTOTALKEYS(hv) > FUV_MAX_SEARCH_SIZE))
15149 array = HvARRAY(hv);
15151 for (i=HvMAX(hv); i>=0; i--) {
15153 for (entry = array[i]; entry; entry = HeNEXT(entry)) {
15154 if (HeVAL(entry) != val)
15156 if ( HeVAL(entry) == &PL_sv_undef ||
15157 HeVAL(entry) == &PL_sv_placeholder)
15161 if (HeKLEN(entry) == HEf_SVKEY)
15162 return sv_mortalcopy(HeKEY_sv(entry));
15163 return sv_2mortal(newSVhek(HeKEY_hek(entry)));
15169 /* Look for an entry in the array whose value has the same SV as val;
15170 * If so, return the index, otherwise return -1. */
15173 S_find_array_subscript(pTHX_ const AV *const av, const SV *const val)
15175 PERL_ARGS_ASSERT_FIND_ARRAY_SUBSCRIPT;
15177 if (!av || SvMAGICAL(av) || !AvARRAY(av) ||
15178 (AvFILLp(av) > FUV_MAX_SEARCH_SIZE))
15181 if (val != &PL_sv_undef) {
15182 SV ** const svp = AvARRAY(av);
15185 for (i=AvFILLp(av); i>=0; i--)
15192 /* varname(): return the name of a variable, optionally with a subscript.
15193 * If gv is non-zero, use the name of that global, along with gvtype (one
15194 * of "$", "@", "%"); otherwise use the name of the lexical at pad offset
15195 * targ. Depending on the value of the subscript_type flag, return:
15198 #define FUV_SUBSCRIPT_NONE 1 /* "@foo" */
15199 #define FUV_SUBSCRIPT_ARRAY 2 /* "$foo[aindex]" */
15200 #define FUV_SUBSCRIPT_HASH 3 /* "$foo{keyname}" */
15201 #define FUV_SUBSCRIPT_WITHIN 4 /* "within @foo" */
15204 Perl_varname(pTHX_ const GV *const gv, const char gvtype, PADOFFSET targ,
15205 const SV *const keyname, I32 aindex, int subscript_type)
15208 SV * const name = sv_newmortal();
15209 if (gv && isGV(gv)) {
15211 buffer[0] = gvtype;
15214 /* as gv_fullname4(), but add literal '^' for $^FOO names */
15216 gv_fullname4(name, gv, buffer, 0);
15218 if ((unsigned int)SvPVX(name)[1] <= 26) {
15220 buffer[1] = SvPVX(name)[1] + 'A' - 1;
15222 /* Swap the 1 unprintable control character for the 2 byte pretty
15223 version - ie substr($name, 1, 1) = $buffer; */
15224 sv_insert(name, 1, 1, buffer, 2);
15228 CV * const cv = gv ? ((CV *)gv) : find_runcv(NULL);
15232 assert(!cv || SvTYPE(cv) == SVt_PVCV || SvTYPE(cv) == SVt_PVFM);
15234 if (!cv || !CvPADLIST(cv))
15236 av = *PadlistARRAY(CvPADLIST(cv));
15237 sv = *av_fetch(av, targ, FALSE);
15238 sv_setsv_flags(name, sv, 0);
15241 if (subscript_type == FUV_SUBSCRIPT_HASH) {
15242 SV * const sv = newSV(0);
15243 *SvPVX(name) = '$';
15244 Perl_sv_catpvf(aTHX_ name, "{%s}",
15245 pv_pretty(sv, SvPVX_const(keyname), SvCUR(keyname), 32, NULL, NULL,
15246 PERL_PV_PRETTY_DUMP | PERL_PV_ESCAPE_UNI_DETECT ));
15247 SvREFCNT_dec_NN(sv);
15249 else if (subscript_type == FUV_SUBSCRIPT_ARRAY) {
15250 *SvPVX(name) = '$';
15251 Perl_sv_catpvf(aTHX_ name, "[%"IVdf"]", (IV)aindex);
15253 else if (subscript_type == FUV_SUBSCRIPT_WITHIN) {
15254 /* We know that name has no magic, so can use 0 instead of SV_GMAGIC */
15255 Perl_sv_insert_flags(aTHX_ name, 0, 0, STR_WITH_LEN("within "), 0);
15263 =for apidoc find_uninit_var
15265 Find the name of the undefined variable (if any) that caused the operator
15266 to issue a "Use of uninitialized value" warning.
15267 If match is true, only return a name if its value matches uninit_sv.
15268 So roughly speaking, if a unary operator (such as OP_COS) generates a
15269 warning, then following the direct child of the op may yield an
15270 OP_PADSV or OP_GV that gives the name of the undefined variable. On the
15271 other hand, with OP_ADD there are two branches to follow, so we only print
15272 the variable name if we get an exact match.
15274 The name is returned as a mortal SV.
15276 Assumes that PL_op is the op that originally triggered the error, and that
15277 PL_comppad/PL_curpad points to the currently executing pad.
15283 S_find_uninit_var(pTHX_ const OP *const obase, const SV *const uninit_sv,
15289 const OP *o, *o2, *kid;
15291 if (!obase || (match && (!uninit_sv || uninit_sv == &PL_sv_undef ||
15292 uninit_sv == &PL_sv_placeholder)))
15295 switch (obase->op_type) {
15302 const bool pad = ( obase->op_type == OP_PADAV
15303 || obase->op_type == OP_PADHV
15304 || obase->op_type == OP_PADRANGE
15307 const bool hash = ( obase->op_type == OP_PADHV
15308 || obase->op_type == OP_RV2HV
15309 || (obase->op_type == OP_PADRANGE
15310 && SvTYPE(PAD_SVl(obase->op_targ)) == SVt_PVHV)
15314 int subscript_type = FUV_SUBSCRIPT_WITHIN;
15316 if (pad) { /* @lex, %lex */
15317 sv = PAD_SVl(obase->op_targ);
15321 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
15322 /* @global, %global */
15323 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
15326 sv = hash ? MUTABLE_SV(GvHV(gv)): MUTABLE_SV(GvAV(gv));
15328 else if (obase == PL_op) /* @{expr}, %{expr} */
15329 return find_uninit_var(cUNOPx(obase)->op_first,
15331 else /* @{expr}, %{expr} as a sub-expression */
15335 /* attempt to find a match within the aggregate */
15337 keysv = find_hash_subscript((const HV*)sv, uninit_sv);
15339 subscript_type = FUV_SUBSCRIPT_HASH;
15342 index = find_array_subscript((const AV *)sv, uninit_sv);
15344 subscript_type = FUV_SUBSCRIPT_ARRAY;
15347 if (match && subscript_type == FUV_SUBSCRIPT_WITHIN)
15350 return varname(gv, hash ? '%' : '@', obase->op_targ,
15351 keysv, index, subscript_type);
15355 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
15357 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
15358 if (!gv || !GvSTASH(gv))
15360 if (match && (GvSV(gv) != uninit_sv))
15362 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
15365 return find_uninit_var(cUNOPx(obase)->op_first, uninit_sv, 1);
15368 if (match && PAD_SVl(obase->op_targ) != uninit_sv)
15370 return varname(NULL, '$', obase->op_targ,
15371 NULL, 0, FUV_SUBSCRIPT_NONE);
15374 gv = cGVOPx_gv(obase);
15375 if (!gv || (match && GvSV(gv) != uninit_sv) || !GvSTASH(gv))
15377 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
15379 case OP_AELEMFAST_LEX:
15382 AV *av = MUTABLE_AV(PAD_SV(obase->op_targ));
15383 if (!av || SvRMAGICAL(av))
15385 svp = av_fetch(av, (I8)obase->op_private, FALSE);
15386 if (!svp || *svp != uninit_sv)
15389 return varname(NULL, '$', obase->op_targ,
15390 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
15393 gv = cGVOPx_gv(obase);
15398 AV *const av = GvAV(gv);
15399 if (!av || SvRMAGICAL(av))
15401 svp = av_fetch(av, (I8)obase->op_private, FALSE);
15402 if (!svp || *svp != uninit_sv)
15405 return varname(gv, '$', 0,
15406 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
15408 NOT_REACHED; /* NOTREACHED */
15411 o = cUNOPx(obase)->op_first;
15412 if (!o || o->op_type != OP_NULL ||
15413 ! (o->op_targ == OP_AELEM || o->op_targ == OP_HELEM))
15415 return find_uninit_var(cBINOPo->op_last, uninit_sv, match);
15420 bool negate = FALSE;
15422 if (PL_op == obase)
15423 /* $a[uninit_expr] or $h{uninit_expr} */
15424 return find_uninit_var(cBINOPx(obase)->op_last, uninit_sv, match);
15427 o = cBINOPx(obase)->op_first;
15428 kid = cBINOPx(obase)->op_last;
15430 /* get the av or hv, and optionally the gv */
15432 if (o->op_type == OP_PADAV || o->op_type == OP_PADHV) {
15433 sv = PAD_SV(o->op_targ);
15435 else if ((o->op_type == OP_RV2AV || o->op_type == OP_RV2HV)
15436 && cUNOPo->op_first->op_type == OP_GV)
15438 gv = cGVOPx_gv(cUNOPo->op_first);
15442 == OP_RV2HV ? MUTABLE_SV(GvHV(gv)) : MUTABLE_SV(GvAV(gv));
15447 if (kid && kid->op_type == OP_NEGATE) {
15449 kid = cUNOPx(kid)->op_first;
15452 if (kid && kid->op_type == OP_CONST && SvOK(cSVOPx_sv(kid))) {
15453 /* index is constant */
15456 kidsv = sv_2mortal(newSVpvs("-"));
15457 sv_catsv(kidsv, cSVOPx_sv(kid));
15460 kidsv = cSVOPx_sv(kid);
15464 if (obase->op_type == OP_HELEM) {
15465 HE* he = hv_fetch_ent(MUTABLE_HV(sv), kidsv, 0, 0);
15466 if (!he || HeVAL(he) != uninit_sv)
15470 SV * const opsv = cSVOPx_sv(kid);
15471 const IV opsviv = SvIV(opsv);
15472 SV * const * const svp = av_fetch(MUTABLE_AV(sv),
15473 negate ? - opsviv : opsviv,
15475 if (!svp || *svp != uninit_sv)
15479 if (obase->op_type == OP_HELEM)
15480 return varname(gv, '%', o->op_targ,
15481 kidsv, 0, FUV_SUBSCRIPT_HASH);
15483 return varname(gv, '@', o->op_targ, NULL,
15484 negate ? - SvIV(cSVOPx_sv(kid)) : SvIV(cSVOPx_sv(kid)),
15485 FUV_SUBSCRIPT_ARRAY);
15488 /* index is an expression;
15489 * attempt to find a match within the aggregate */
15490 if (obase->op_type == OP_HELEM) {
15491 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
15493 return varname(gv, '%', o->op_targ,
15494 keysv, 0, FUV_SUBSCRIPT_HASH);
15498 = find_array_subscript((const AV *)sv, uninit_sv);
15500 return varname(gv, '@', o->op_targ,
15501 NULL, index, FUV_SUBSCRIPT_ARRAY);
15506 (o->op_type == OP_PADAV || o->op_type == OP_RV2AV)
15508 o->op_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
15510 NOT_REACHED; /* NOTREACHED */
15514 /* only examine RHS */
15515 return find_uninit_var(cBINOPx(obase)->op_first, uninit_sv, match);
15518 o = cUNOPx(obase)->op_first;
15519 if ( o->op_type == OP_PUSHMARK
15520 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)
15524 if (!OP_HAS_SIBLING(o)) {
15525 /* one-arg version of open is highly magical */
15527 if (o->op_type == OP_GV) { /* open FOO; */
15529 if (match && GvSV(gv) != uninit_sv)
15531 return varname(gv, '$', 0,
15532 NULL, 0, FUV_SUBSCRIPT_NONE);
15534 /* other possibilities not handled are:
15535 * open $x; or open my $x; should return '${*$x}'
15536 * open expr; should return '$'.expr ideally
15542 /* ops where $_ may be an implicit arg */
15547 if ( !(obase->op_flags & OPf_STACKED)) {
15548 if (uninit_sv == ((obase->op_private & OPpTARGET_MY)
15549 ? PAD_SVl(obase->op_targ)
15552 sv = sv_newmortal();
15553 sv_setpvs(sv, "$_");
15562 match = 1; /* print etc can return undef on defined args */
15563 /* skip filehandle as it can't produce 'undef' warning */
15564 o = cUNOPx(obase)->op_first;
15565 if ((obase->op_flags & OPf_STACKED)
15567 ( o->op_type == OP_PUSHMARK
15568 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)))
15569 o = OP_SIBLING(OP_SIBLING(o));
15573 case OP_ENTEREVAL: /* could be eval $undef or $x='$undef'; eval $x */
15574 case OP_CUSTOM: /* XS or custom code could trigger random warnings */
15576 /* the following ops are capable of returning PL_sv_undef even for
15577 * defined arg(s) */
15596 case OP_GETPEERNAME:
15644 case OP_SMARTMATCH:
15653 /* XXX tmp hack: these two may call an XS sub, and currently
15654 XS subs don't have a SUB entry on the context stack, so CV and
15655 pad determination goes wrong, and BAD things happen. So, just
15656 don't try to determine the value under those circumstances.
15657 Need a better fix at dome point. DAPM 11/2007 */
15663 GV * const gv = gv_fetchpvs(".", GV_NOTQUAL, SVt_PV);
15664 if (gv && GvSV(gv) == uninit_sv)
15665 return newSVpvs_flags("$.", SVs_TEMP);
15670 /* def-ness of rval pos() is independent of the def-ness of its arg */
15671 if ( !(obase->op_flags & OPf_MOD))
15676 if (SvROK(PL_rs) && uninit_sv == SvRV(PL_rs))
15677 return newSVpvs_flags("${$/}", SVs_TEMP);
15682 if (!(obase->op_flags & OPf_KIDS))
15684 o = cUNOPx(obase)->op_first;
15690 /* This loop checks all the kid ops, skipping any that cannot pos-
15691 * sibly be responsible for the uninitialized value; i.e., defined
15692 * constants and ops that return nothing. If there is only one op
15693 * left that is not skipped, then we *know* it is responsible for
15694 * the uninitialized value. If there is more than one op left, we
15695 * have to look for an exact match in the while() loop below.
15696 * Note that we skip padrange, because the individual pad ops that
15697 * it replaced are still in the tree, so we work on them instead.
15700 for (kid=o; kid; kid = OP_SIBLING(kid)) {
15701 const OPCODE type = kid->op_type;
15702 if ( (type == OP_CONST && SvOK(cSVOPx_sv(kid)))
15703 || (type == OP_NULL && ! (kid->op_flags & OPf_KIDS))
15704 || (type == OP_PUSHMARK)
15705 || (type == OP_PADRANGE)
15709 if (o2) { /* more than one found */
15716 return find_uninit_var(o2, uninit_sv, match);
15718 /* scan all args */
15720 sv = find_uninit_var(o, uninit_sv, 1);
15732 =for apidoc report_uninit
15734 Print appropriate "Use of uninitialized variable" warning.
15740 Perl_report_uninit(pTHX_ const SV *uninit_sv)
15743 SV* varname = NULL;
15744 if (uninit_sv && PL_curpad) {
15745 varname = find_uninit_var(PL_op, uninit_sv,0);
15747 sv_insert(varname, 0, 0, " ", 1);
15749 /* PL_warn_uninit_sv is constant */
15750 GCC_DIAG_IGNORE(-Wformat-nonliteral);
15751 /* diag_listed_as: Use of uninitialized value%s */
15752 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit_sv,
15753 SVfARG(varname ? varname : &PL_sv_no),
15754 " in ", OP_DESC(PL_op));
15758 /* PL_warn_uninit is constant */
15759 GCC_DIAG_IGNORE(-Wformat-nonliteral);
15760 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit,
15768 * c-indentation-style: bsd
15769 * c-basic-offset: 4
15770 * indent-tabs-mode: nil
15773 * ex: set ts=8 sts=4 sw=4 et: