3 * Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000,
4 * 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009 by Larry Wall
7 * You may distribute under the terms of either the GNU General Public
8 * License or the Artistic License, as specified in the README file.
13 * 'I wonder what the Entish is for "yes" and "no",' he thought.
16 * [p.480 of _The Lord of the Rings_, III/iv: "Treebeard"]
22 * This file contains the code that creates, manipulates and destroys
23 * scalar values (SVs). The other types (AV, HV, GV, etc.) reuse the
24 * structure of an SV, so their creation and destruction is handled
25 * here; higher-level functions are in av.c, hv.c, and so on. Opcode
26 * level functions (eg. substr, split, join) for each of the types are
39 /* Missing proto on LynxOS */
40 char *gconvert(double, int, int, char *);
44 # define SNPRINTF_G(nv, buffer, size, ndig) \
45 quadmath_snprintf(buffer, size, "%.*Qg", (int)ndig, (NV)(nv))
47 # define SNPRINTF_G(nv, buffer, size, ndig) \
48 PERL_UNUSED_RESULT(Gconvert((NV)(nv), (int)ndig, 0, buffer))
51 #ifndef SV_COW_THRESHOLD
52 # define SV_COW_THRESHOLD 0 /* COW iff len > K */
54 #ifndef SV_COWBUF_THRESHOLD
55 # define SV_COWBUF_THRESHOLD 1250 /* COW iff len > K */
57 #ifndef SV_COW_MAX_WASTE_THRESHOLD
58 # define SV_COW_MAX_WASTE_THRESHOLD 80 /* COW iff (len - cur) < K */
60 #ifndef SV_COWBUF_WASTE_THRESHOLD
61 # define SV_COWBUF_WASTE_THRESHOLD 80 /* COW iff (len - cur) < K */
63 #ifndef SV_COW_MAX_WASTE_FACTOR_THRESHOLD
64 # define SV_COW_MAX_WASTE_FACTOR_THRESHOLD 2 /* COW iff len < (cur * K) */
66 #ifndef SV_COWBUF_WASTE_FACTOR_THRESHOLD
67 # define SV_COWBUF_WASTE_FACTOR_THRESHOLD 2 /* COW iff len < (cur * K) */
69 /* Work around compiler warnings about unsigned >= THRESHOLD when thres-
72 # define GE_COW_THRESHOLD(cur) ((cur) >= SV_COW_THRESHOLD)
74 # define GE_COW_THRESHOLD(cur) 1
76 #if SV_COWBUF_THRESHOLD
77 # define GE_COWBUF_THRESHOLD(cur) ((cur) >= SV_COWBUF_THRESHOLD)
79 # define GE_COWBUF_THRESHOLD(cur) 1
81 #if SV_COW_MAX_WASTE_THRESHOLD
82 # define GE_COW_MAX_WASTE_THRESHOLD(cur,len) (((len)-(cur)) < SV_COW_MAX_WASTE_THRESHOLD)
84 # define GE_COW_MAX_WASTE_THRESHOLD(cur,len) 1
86 #if SV_COWBUF_WASTE_THRESHOLD
87 # define GE_COWBUF_WASTE_THRESHOLD(cur,len) (((len)-(cur)) < SV_COWBUF_WASTE_THRESHOLD)
89 # define GE_COWBUF_WASTE_THRESHOLD(cur,len) 1
91 #if SV_COW_MAX_WASTE_FACTOR_THRESHOLD
92 # define GE_COW_MAX_WASTE_FACTOR_THRESHOLD(cur,len) ((len) < SV_COW_MAX_WASTE_FACTOR_THRESHOLD * (cur))
94 # define GE_COW_MAX_WASTE_FACTOR_THRESHOLD(cur,len) 1
96 #if SV_COWBUF_WASTE_FACTOR_THRESHOLD
97 # define GE_COWBUF_WASTE_FACTOR_THRESHOLD(cur,len) ((len) < SV_COWBUF_WASTE_FACTOR_THRESHOLD * (cur))
99 # define GE_COWBUF_WASTE_FACTOR_THRESHOLD(cur,len) 1
102 #define CHECK_COW_THRESHOLD(cur,len) (\
103 GE_COW_THRESHOLD((cur)) && \
104 GE_COW_MAX_WASTE_THRESHOLD((cur),(len)) && \
105 GE_COW_MAX_WASTE_FACTOR_THRESHOLD((cur),(len)) \
107 #define CHECK_COWBUF_THRESHOLD(cur,len) (\
108 GE_COWBUF_THRESHOLD((cur)) && \
109 GE_COWBUF_WASTE_THRESHOLD((cur),(len)) && \
110 GE_COWBUF_WASTE_FACTOR_THRESHOLD((cur),(len)) \
113 #ifdef PERL_UTF8_CACHE_ASSERT
114 /* if adding more checks watch out for the following tests:
115 * t/op/index.t t/op/length.t t/op/pat.t t/op/substr.t
116 * lib/utf8.t lib/Unicode/Collate/t/index.t
119 # define ASSERT_UTF8_CACHE(cache) \
120 STMT_START { if (cache) { assert((cache)[0] <= (cache)[1]); \
121 assert((cache)[2] <= (cache)[3]); \
122 assert((cache)[3] <= (cache)[1]);} \
125 # define ASSERT_UTF8_CACHE(cache) NOOP
128 /* ============================================================================
130 =head1 Allocation and deallocation of SVs.
131 An SV (or AV, HV, etc.) is allocated in two parts: the head (struct
132 sv, av, hv...) contains type and reference count information, and for
133 many types, a pointer to the body (struct xrv, xpv, xpviv...), which
134 contains fields specific to each type. Some types store all they need
135 in the head, so don't have a body.
137 In all but the most memory-paranoid configurations (ex: PURIFY), heads
138 and bodies are allocated out of arenas, which by default are
139 approximately 4K chunks of memory parcelled up into N heads or bodies.
140 Sv-bodies are allocated by their sv-type, guaranteeing size
141 consistency needed to allocate safely from arrays.
143 For SV-heads, the first slot in each arena is reserved, and holds a
144 link to the next arena, some flags, and a note of the number of slots.
145 Snaked through each arena chain is a linked list of free items; when
146 this becomes empty, an extra arena is allocated and divided up into N
147 items which are threaded into the free list.
149 SV-bodies are similar, but they use arena-sets by default, which
150 separate the link and info from the arena itself, and reclaim the 1st
151 slot in the arena. SV-bodies are further described later.
153 The following global variables are associated with arenas:
155 PL_sv_arenaroot pointer to list of SV arenas
156 PL_sv_root pointer to list of free SV structures
158 PL_body_arenas head of linked-list of body arenas
159 PL_body_roots[] array of pointers to list of free bodies of svtype
160 arrays are indexed by the svtype needed
162 A few special SV heads are not allocated from an arena, but are
163 instead directly created in the interpreter structure, eg PL_sv_undef.
164 The size of arenas can be changed from the default by setting
165 PERL_ARENA_SIZE appropriately at compile time.
167 The SV arena serves the secondary purpose of allowing still-live SVs
168 to be located and destroyed during final cleanup.
170 At the lowest level, the macros new_SV() and del_SV() grab and free
171 an SV head. (If debugging with -DD, del_SV() calls the function S_del_sv()
172 to return the SV to the free list with error checking.) new_SV() calls
173 more_sv() / sv_add_arena() to add an extra arena if the free list is empty.
174 SVs in the free list have their SvTYPE field set to all ones.
176 At the time of very final cleanup, sv_free_arenas() is called from
177 perl_destruct() to physically free all the arenas allocated since the
178 start of the interpreter.
180 The function visit() scans the SV arenas list, and calls a specified
181 function for each SV it finds which is still live - ie which has an SvTYPE
182 other than all 1's, and a non-zero SvREFCNT. visit() is used by the
183 following functions (specified as [function that calls visit()] / [function
184 called by visit() for each SV]):
186 sv_report_used() / do_report_used()
187 dump all remaining SVs (debugging aid)
189 sv_clean_objs() / do_clean_objs(),do_clean_named_objs(),
190 do_clean_named_io_objs(),do_curse()
191 Attempt to free all objects pointed to by RVs,
192 try to do the same for all objects indir-
193 ectly referenced by typeglobs too, and
194 then do a final sweep, cursing any
195 objects that remain. Called once from
196 perl_destruct(), prior to calling sv_clean_all()
199 sv_clean_all() / do_clean_all()
200 SvREFCNT_dec(sv) each remaining SV, possibly
201 triggering an sv_free(). It also sets the
202 SVf_BREAK flag on the SV to indicate that the
203 refcnt has been artificially lowered, and thus
204 stopping sv_free() from giving spurious warnings
205 about SVs which unexpectedly have a refcnt
206 of zero. called repeatedly from perl_destruct()
207 until there are no SVs left.
209 =head2 Arena allocator API Summary
211 Private API to rest of sv.c
215 new_XPVNV(), del_XPVGV(),
220 sv_report_used(), sv_clean_objs(), sv_clean_all(), sv_free_arenas()
224 * ========================================================================= */
227 * "A time to plant, and a time to uproot what was planted..."
231 # define MEM_LOG_NEW_SV(sv, file, line, func) \
232 Perl_mem_log_new_sv(sv, file, line, func)
233 # define MEM_LOG_DEL_SV(sv, file, line, func) \
234 Perl_mem_log_del_sv(sv, file, line, func)
236 # define MEM_LOG_NEW_SV(sv, file, line, func) NOOP
237 # define MEM_LOG_DEL_SV(sv, file, line, func) NOOP
240 #ifdef DEBUG_LEAKING_SCALARS
241 # define FREE_SV_DEBUG_FILE(sv) STMT_START { \
242 if ((sv)->sv_debug_file) PerlMemShared_free((sv)->sv_debug_file); \
244 # define DEBUG_SV_SERIAL(sv) \
245 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) del_SV\n", \
246 PTR2UV(sv), (long)(sv)->sv_debug_serial))
248 # define FREE_SV_DEBUG_FILE(sv)
249 # define DEBUG_SV_SERIAL(sv) NOOP
253 # define SvARENA_CHAIN(sv) ((sv)->sv_u.svu_rv)
254 # define SvARENA_CHAIN_SET(sv,val) (sv)->sv_u.svu_rv = MUTABLE_SV((val))
255 /* Whilst I'd love to do this, it seems that things like to check on
257 # define POISON_SV_HEAD(sv) PoisonNew(sv, 1, struct STRUCT_SV)
259 # define POISON_SV_HEAD(sv) PoisonNew(&SvANY(sv), 1, void *), \
260 PoisonNew(&SvREFCNT(sv), 1, U32)
262 # define SvARENA_CHAIN(sv) SvANY(sv)
263 # define SvARENA_CHAIN_SET(sv,val) SvANY(sv) = (void *)(val)
264 # define POISON_SV_HEAD(sv)
267 /* Mark an SV head as unused, and add to free list.
269 * If SVf_BREAK is set, skip adding it to the free list, as this SV had
270 * its refcount artificially decremented during global destruction, so
271 * there may be dangling pointers to it. The last thing we want in that
272 * case is for it to be reused. */
274 #define plant_SV(p) \
276 const U32 old_flags = SvFLAGS(p); \
277 MEM_LOG_DEL_SV(p, __FILE__, __LINE__, FUNCTION__); \
278 DEBUG_SV_SERIAL(p); \
279 FREE_SV_DEBUG_FILE(p); \
281 SvFLAGS(p) = SVTYPEMASK; \
282 if (!(old_flags & SVf_BREAK)) { \
283 SvARENA_CHAIN_SET(p, PL_sv_root); \
289 #define uproot_SV(p) \
292 PL_sv_root = MUTABLE_SV(SvARENA_CHAIN(p)); \
297 /* make some more SVs by adding another arena */
303 char *chunk; /* must use New here to match call to */
304 Newx(chunk,PERL_ARENA_SIZE,char); /* Safefree() in sv_free_arenas() */
305 sv_add_arena(chunk, PERL_ARENA_SIZE, 0);
310 /* new_SV(): return a new, empty SV head */
312 #ifdef DEBUG_LEAKING_SCALARS
313 /* provide a real function for a debugger to play with */
315 S_new_SV(pTHX_ const char *file, int line, const char *func)
322 sv = S_more_sv(aTHX);
326 sv->sv_debug_optype = PL_op ? PL_op->op_type : 0;
327 sv->sv_debug_line = (U16) (PL_parser && PL_parser->copline != NOLINE
333 sv->sv_debug_inpad = 0;
334 sv->sv_debug_parent = NULL;
335 sv->sv_debug_file = PL_curcop ? savesharedpv(CopFILE(PL_curcop)): NULL;
337 sv->sv_debug_serial = PL_sv_serial++;
339 MEM_LOG_NEW_SV(sv, file, line, func);
340 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) new_SV (from %s:%d [%s])\n",
341 PTR2UV(sv), (long)sv->sv_debug_serial, file, line, func));
345 # define new_SV(p) (p)=S_new_SV(aTHX_ __FILE__, __LINE__, FUNCTION__)
353 (p) = S_more_sv(aTHX); \
357 MEM_LOG_NEW_SV(p, __FILE__, __LINE__, FUNCTION__); \
362 /* del_SV(): return an empty SV head to the free list */
375 S_del_sv(pTHX_ SV *p)
377 PERL_ARGS_ASSERT_DEL_SV;
382 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
383 const SV * const sv = sva + 1;
384 const SV * const svend = &sva[SvREFCNT(sva)];
385 if (p >= sv && p < svend) {
391 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
392 "Attempt to free non-arena SV: 0x%"UVxf
393 pTHX__FORMAT, PTR2UV(p) pTHX__VALUE);
400 #else /* ! DEBUGGING */
402 #define del_SV(p) plant_SV(p)
404 #endif /* DEBUGGING */
407 * Bodyless IVs and NVs!
409 * Since 5.9.2, we can avoid allocating a body for SVt_IV-type SVs.
410 * Since the larger IV-holding variants of SVs store their integer
411 * values in their respective bodies, the family of SvIV() accessor
412 * macros would naively have to branch on the SV type to find the
413 * integer value either in the HEAD or BODY. In order to avoid this
414 * expensive branch, a clever soul has deployed a great hack:
415 * We set up the SvANY pointer such that instead of pointing to a
416 * real body, it points into the memory before the location of the
417 * head. We compute this pointer such that the location of
418 * the integer member of the hypothetical body struct happens to
419 * be the same as the location of the integer member of the bodyless
420 * SV head. This now means that the SvIV() family of accessors can
421 * always read from the (hypothetical or real) body via SvANY.
423 * Since the 5.21 dev series, we employ the same trick for NVs
424 * if the architecture can support it (NVSIZE <= IVSIZE).
427 /* The following two macros compute the necessary offsets for the above
428 * trick and store them in SvANY for SvIV() (and friends) to use. */
429 #define SET_SVANY_FOR_BODYLESS_IV(sv) \
430 SvANY(sv) = (XPVIV*)((char*)&(sv->sv_u.svu_iv) - STRUCT_OFFSET(XPVIV, xiv_iv))
432 #define SET_SVANY_FOR_BODYLESS_NV(sv) \
433 SvANY(sv) = (XPVNV*)((char*)&(sv->sv_u.svu_nv) - STRUCT_OFFSET(XPVNV, xnv_u.xnv_nv))
436 =head1 SV Manipulation Functions
438 =for apidoc sv_add_arena
440 Given a chunk of memory, link it to the head of the list of arenas,
441 and split it into a list of free SVs.
447 S_sv_add_arena(pTHX_ char *const ptr, const U32 size, const U32 flags)
449 SV *const sva = MUTABLE_SV(ptr);
453 PERL_ARGS_ASSERT_SV_ADD_ARENA;
455 /* The first SV in an arena isn't an SV. */
456 SvANY(sva) = (void *) PL_sv_arenaroot; /* ptr to next arena */
457 SvREFCNT(sva) = size / sizeof(SV); /* number of SV slots */
458 SvFLAGS(sva) = flags; /* FAKE if not to be freed */
460 PL_sv_arenaroot = sva;
461 PL_sv_root = sva + 1;
463 svend = &sva[SvREFCNT(sva) - 1];
466 SvARENA_CHAIN_SET(sv, (sv + 1));
470 /* Must always set typemask because it's always checked in on cleanup
471 when the arenas are walked looking for objects. */
472 SvFLAGS(sv) = SVTYPEMASK;
475 SvARENA_CHAIN_SET(sv, 0);
479 SvFLAGS(sv) = SVTYPEMASK;
482 /* visit(): call the named function for each non-free SV in the arenas
483 * whose flags field matches the flags/mask args. */
486 S_visit(pTHX_ SVFUNC_t f, const U32 flags, const U32 mask)
491 PERL_ARGS_ASSERT_VISIT;
493 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
494 const SV * const svend = &sva[SvREFCNT(sva)];
496 for (sv = sva + 1; sv < svend; ++sv) {
497 if (SvTYPE(sv) != (svtype)SVTYPEMASK
498 && (sv->sv_flags & mask) == flags
511 /* called by sv_report_used() for each live SV */
514 do_report_used(pTHX_ SV *const sv)
516 if (SvTYPE(sv) != (svtype)SVTYPEMASK) {
517 PerlIO_printf(Perl_debug_log, "****\n");
524 =for apidoc sv_report_used
526 Dump the contents of all SVs not yet freed (debugging aid).
532 Perl_sv_report_used(pTHX)
535 visit(do_report_used, 0, 0);
541 /* called by sv_clean_objs() for each live SV */
544 do_clean_objs(pTHX_ SV *const ref)
548 SV * const target = SvRV(ref);
549 if (SvOBJECT(target)) {
550 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning object ref:\n "), sv_dump(ref)));
551 if (SvWEAKREF(ref)) {
552 sv_del_backref(target, ref);
558 SvREFCNT_dec_NN(target);
565 /* clear any slots in a GV which hold objects - except IO;
566 * called by sv_clean_objs() for each live GV */
569 do_clean_named_objs(pTHX_ SV *const sv)
572 assert(SvTYPE(sv) == SVt_PVGV);
573 assert(isGV_with_GP(sv));
577 /* freeing GP entries may indirectly free the current GV;
578 * hold onto it while we mess with the GP slots */
581 if ( ((obj = GvSV(sv) )) && SvOBJECT(obj)) {
582 DEBUG_D((PerlIO_printf(Perl_debug_log,
583 "Cleaning named glob SV object:\n "), sv_dump(obj)));
585 SvREFCNT_dec_NN(obj);
587 if ( ((obj = MUTABLE_SV(GvAV(sv)) )) && SvOBJECT(obj)) {
588 DEBUG_D((PerlIO_printf(Perl_debug_log,
589 "Cleaning named glob AV object:\n "), sv_dump(obj)));
591 SvREFCNT_dec_NN(obj);
593 if ( ((obj = MUTABLE_SV(GvHV(sv)) )) && SvOBJECT(obj)) {
594 DEBUG_D((PerlIO_printf(Perl_debug_log,
595 "Cleaning named glob HV object:\n "), sv_dump(obj)));
597 SvREFCNT_dec_NN(obj);
599 if ( ((obj = MUTABLE_SV(GvCV(sv)) )) && SvOBJECT(obj)) {
600 DEBUG_D((PerlIO_printf(Perl_debug_log,
601 "Cleaning named glob CV object:\n "), sv_dump(obj)));
603 SvREFCNT_dec_NN(obj);
605 SvREFCNT_dec_NN(sv); /* undo the inc above */
608 /* clear any IO slots in a GV which hold objects (except stderr, defout);
609 * called by sv_clean_objs() for each live GV */
612 do_clean_named_io_objs(pTHX_ SV *const sv)
615 assert(SvTYPE(sv) == SVt_PVGV);
616 assert(isGV_with_GP(sv));
617 if (!GvGP(sv) || sv == (SV*)PL_stderrgv || sv == (SV*)PL_defoutgv)
621 if ( ((obj = MUTABLE_SV(GvIO(sv)) )) && SvOBJECT(obj)) {
622 DEBUG_D((PerlIO_printf(Perl_debug_log,
623 "Cleaning named glob IO object:\n "), sv_dump(obj)));
625 SvREFCNT_dec_NN(obj);
627 SvREFCNT_dec_NN(sv); /* undo the inc above */
630 /* Void wrapper to pass to visit() */
632 do_curse(pTHX_ SV * const sv) {
633 if ((PL_stderrgv && GvGP(PL_stderrgv) && (SV*)GvIO(PL_stderrgv) == sv)
634 || (PL_defoutgv && GvGP(PL_defoutgv) && (SV*)GvIO(PL_defoutgv) == sv))
640 =for apidoc sv_clean_objs
642 Attempt to destroy all objects not yet freed.
648 Perl_sv_clean_objs(pTHX)
651 PL_in_clean_objs = TRUE;
652 visit(do_clean_objs, SVf_ROK, SVf_ROK);
653 /* Some barnacles may yet remain, clinging to typeglobs.
654 * Run the non-IO destructors first: they may want to output
655 * error messages, close files etc */
656 visit(do_clean_named_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
657 visit(do_clean_named_io_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
658 /* And if there are some very tenacious barnacles clinging to arrays,
659 closures, or what have you.... */
660 visit(do_curse, SVs_OBJECT, SVs_OBJECT);
661 olddef = PL_defoutgv;
662 PL_defoutgv = NULL; /* disable skip of PL_defoutgv */
663 if (olddef && isGV_with_GP(olddef))
664 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olddef));
665 olderr = PL_stderrgv;
666 PL_stderrgv = NULL; /* disable skip of PL_stderrgv */
667 if (olderr && isGV_with_GP(olderr))
668 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olderr));
669 SvREFCNT_dec(olddef);
670 PL_in_clean_objs = FALSE;
673 /* called by sv_clean_all() for each live SV */
676 do_clean_all(pTHX_ SV *const sv)
678 if (sv == (const SV *) PL_fdpid || sv == (const SV *)PL_strtab) {
679 /* don't clean pid table and strtab */
682 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning loops: SV at 0x%"UVxf"\n", PTR2UV(sv)) ));
683 SvFLAGS(sv) |= SVf_BREAK;
688 =for apidoc sv_clean_all
690 Decrement the refcnt of each remaining SV, possibly triggering a
691 cleanup. This function may have to be called multiple times to free
692 SVs which are in complex self-referential hierarchies.
698 Perl_sv_clean_all(pTHX)
701 PL_in_clean_all = TRUE;
702 cleaned = visit(do_clean_all, 0,0);
707 ARENASETS: a meta-arena implementation which separates arena-info
708 into struct arena_set, which contains an array of struct
709 arena_descs, each holding info for a single arena. By separating
710 the meta-info from the arena, we recover the 1st slot, formerly
711 borrowed for list management. The arena_set is about the size of an
712 arena, avoiding the needless malloc overhead of a naive linked-list.
714 The cost is 1 arena-set malloc per ~320 arena-mallocs, + the unused
715 memory in the last arena-set (1/2 on average). In trade, we get
716 back the 1st slot in each arena (ie 1.7% of a CV-arena, less for
717 smaller types). The recovery of the wasted space allows use of
718 small arenas for large, rare body types, by changing array* fields
719 in body_details_by_type[] below.
722 char *arena; /* the raw storage, allocated aligned */
723 size_t size; /* its size ~4k typ */
724 svtype utype; /* bodytype stored in arena */
729 /* Get the maximum number of elements in set[] such that struct arena_set
730 will fit within PERL_ARENA_SIZE, which is probably just under 4K, and
731 therefore likely to be 1 aligned memory page. */
733 #define ARENAS_PER_SET ((PERL_ARENA_SIZE - sizeof(struct arena_set*) \
734 - 2 * sizeof(int)) / sizeof (struct arena_desc))
737 struct arena_set* next;
738 unsigned int set_size; /* ie ARENAS_PER_SET */
739 unsigned int curr; /* index of next available arena-desc */
740 struct arena_desc set[ARENAS_PER_SET];
744 =for apidoc sv_free_arenas
746 Deallocate the memory used by all arenas. Note that all the individual SV
747 heads and bodies within the arenas must already have been freed.
753 Perl_sv_free_arenas(pTHX)
759 /* Free arenas here, but be careful about fake ones. (We assume
760 contiguity of the fake ones with the corresponding real ones.) */
762 for (sva = PL_sv_arenaroot; sva; sva = svanext) {
763 svanext = MUTABLE_SV(SvANY(sva));
764 while (svanext && SvFAKE(svanext))
765 svanext = MUTABLE_SV(SvANY(svanext));
772 struct arena_set *aroot = (struct arena_set*) PL_body_arenas;
775 struct arena_set *current = aroot;
778 assert(aroot->set[i].arena);
779 Safefree(aroot->set[i].arena);
787 i = PERL_ARENA_ROOTS_SIZE;
789 PL_body_roots[i] = 0;
796 Here are mid-level routines that manage the allocation of bodies out
797 of the various arenas. There are 5 kinds of arenas:
799 1. SV-head arenas, which are discussed and handled above
800 2. regular body arenas
801 3. arenas for reduced-size bodies
804 Arena types 2 & 3 are chained by body-type off an array of
805 arena-root pointers, which is indexed by svtype. Some of the
806 larger/less used body types are malloced singly, since a large
807 unused block of them is wasteful. Also, several svtypes dont have
808 bodies; the data fits into the sv-head itself. The arena-root
809 pointer thus has a few unused root-pointers (which may be hijacked
810 later for arena types 4,5)
812 3 differs from 2 as an optimization; some body types have several
813 unused fields in the front of the structure (which are kept in-place
814 for consistency). These bodies can be allocated in smaller chunks,
815 because the leading fields arent accessed. Pointers to such bodies
816 are decremented to point at the unused 'ghost' memory, knowing that
817 the pointers are used with offsets to the real memory.
820 =head1 SV-Body Allocation
824 Allocation of SV-bodies is similar to SV-heads, differing as follows;
825 the allocation mechanism is used for many body types, so is somewhat
826 more complicated, it uses arena-sets, and has no need for still-live
829 At the outermost level, (new|del)_X*V macros return bodies of the
830 appropriate type. These macros call either (new|del)_body_type or
831 (new|del)_body_allocated macro pairs, depending on specifics of the
832 type. Most body types use the former pair, the latter pair is used to
833 allocate body types with "ghost fields".
835 "ghost fields" are fields that are unused in certain types, and
836 consequently don't need to actually exist. They are declared because
837 they're part of a "base type", which allows use of functions as
838 methods. The simplest examples are AVs and HVs, 2 aggregate types
839 which don't use the fields which support SCALAR semantics.
841 For these types, the arenas are carved up into appropriately sized
842 chunks, we thus avoid wasted memory for those unaccessed members.
843 When bodies are allocated, we adjust the pointer back in memory by the
844 size of the part not allocated, so it's as if we allocated the full
845 structure. (But things will all go boom if you write to the part that
846 is "not there", because you'll be overwriting the last members of the
847 preceding structure in memory.)
849 We calculate the correction using the STRUCT_OFFSET macro on the first
850 member present. If the allocated structure is smaller (no initial NV
851 actually allocated) then the net effect is to subtract the size of the NV
852 from the pointer, to return a new pointer as if an initial NV were actually
853 allocated. (We were using structures named *_allocated for this, but
854 this turned out to be a subtle bug, because a structure without an NV
855 could have a lower alignment constraint, but the compiler is allowed to
856 optimised accesses based on the alignment constraint of the actual pointer
857 to the full structure, for example, using a single 64 bit load instruction
858 because it "knows" that two adjacent 32 bit members will be 8-byte aligned.)
860 This is the same trick as was used for NV and IV bodies. Ironically it
861 doesn't need to be used for NV bodies any more, because NV is now at
862 the start of the structure. IV bodies, and also in some builds NV bodies,
863 don't need it either, because they are no longer allocated.
865 In turn, the new_body_* allocators call S_new_body(), which invokes
866 new_body_inline macro, which takes a lock, and takes a body off the
867 linked list at PL_body_roots[sv_type], calling Perl_more_bodies() if
868 necessary to refresh an empty list. Then the lock is released, and
869 the body is returned.
871 Perl_more_bodies allocates a new arena, and carves it up into an array of N
872 bodies, which it strings into a linked list. It looks up arena-size
873 and body-size from the body_details table described below, thus
874 supporting the multiple body-types.
876 If PURIFY is defined, or PERL_ARENA_SIZE=0, arenas are not used, and
877 the (new|del)_X*V macros are mapped directly to malloc/free.
879 For each sv-type, struct body_details bodies_by_type[] carries
880 parameters which control these aspects of SV handling:
882 Arena_size determines whether arenas are used for this body type, and if
883 so, how big they are. PURIFY or PERL_ARENA_SIZE=0 set this field to
884 zero, forcing individual mallocs and frees.
886 Body_size determines how big a body is, and therefore how many fit into
887 each arena. Offset carries the body-pointer adjustment needed for
888 "ghost fields", and is used in *_allocated macros.
890 But its main purpose is to parameterize info needed in
891 Perl_sv_upgrade(). The info here dramatically simplifies the function
892 vs the implementation in 5.8.8, making it table-driven. All fields
893 are used for this, except for arena_size.
895 For the sv-types that have no bodies, arenas are not used, so those
896 PL_body_roots[sv_type] are unused, and can be overloaded. In
897 something of a special case, SVt_NULL is borrowed for HE arenas;
898 PL_body_roots[HE_SVSLOT=SVt_NULL] is filled by S_more_he, but the
899 bodies_by_type[SVt_NULL] slot is not used, as the table is not
904 struct body_details {
905 U8 body_size; /* Size to allocate */
906 U8 copy; /* Size of structure to copy (may be shorter) */
907 U8 offset; /* Size of unalloced ghost fields to first alloced field*/
908 PERL_BITFIELD8 type : 4; /* We have space for a sanity check. */
909 PERL_BITFIELD8 cant_upgrade : 1;/* Cannot upgrade this type */
910 PERL_BITFIELD8 zero_nv : 1; /* zero the NV when upgrading from this */
911 PERL_BITFIELD8 arena : 1; /* Allocated from an arena */
912 U32 arena_size; /* Size of arena to allocate */
920 /* With -DPURFIY we allocate everything directly, and don't use arenas.
921 This seems a rather elegant way to simplify some of the code below. */
922 #define HASARENA FALSE
924 #define HASARENA TRUE
926 #define NOARENA FALSE
928 /* Size the arenas to exactly fit a given number of bodies. A count
929 of 0 fits the max number bodies into a PERL_ARENA_SIZE.block,
930 simplifying the default. If count > 0, the arena is sized to fit
931 only that many bodies, allowing arenas to be used for large, rare
932 bodies (XPVFM, XPVIO) without undue waste. The arena size is
933 limited by PERL_ARENA_SIZE, so we can safely oversize the
936 #define FIT_ARENA0(body_size) \
937 ((size_t)(PERL_ARENA_SIZE / body_size) * body_size)
938 #define FIT_ARENAn(count,body_size) \
939 ( count * body_size <= PERL_ARENA_SIZE) \
940 ? count * body_size \
941 : FIT_ARENA0 (body_size)
942 #define FIT_ARENA(count,body_size) \
944 ? FIT_ARENAn (count, body_size) \
945 : FIT_ARENA0 (body_size))
947 /* Calculate the length to copy. Specifically work out the length less any
948 final padding the compiler needed to add. See the comment in sv_upgrade
949 for why copying the padding proved to be a bug. */
951 #define copy_length(type, last_member) \
952 STRUCT_OFFSET(type, last_member) \
953 + sizeof (((type*)SvANY((const SV *)0))->last_member)
955 static const struct body_details bodies_by_type[] = {
956 /* HEs use this offset for their arena. */
957 { 0, 0, 0, SVt_NULL, FALSE, NONV, NOARENA, 0 },
959 /* IVs are in the head, so the allocation size is 0. */
961 sizeof(IV), /* This is used to copy out the IV body. */
962 STRUCT_OFFSET(XPVIV, xiv_iv), SVt_IV, FALSE, NONV,
963 NOARENA /* IVS don't need an arena */, 0
968 STRUCT_OFFSET(XPVNV, xnv_u),
969 SVt_NV, FALSE, HADNV, NOARENA, 0 },
971 { sizeof(NV), sizeof(NV),
972 STRUCT_OFFSET(XPVNV, xnv_u),
973 SVt_NV, FALSE, HADNV, HASARENA, FIT_ARENA(0, sizeof(NV)) },
976 { sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur),
977 copy_length(XPV, xpv_len) - STRUCT_OFFSET(XPV, xpv_cur),
978 + STRUCT_OFFSET(XPV, xpv_cur),
979 SVt_PV, FALSE, NONV, HASARENA,
980 FIT_ARENA(0, sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur)) },
982 { sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur),
983 copy_length(XINVLIST, is_offset) - STRUCT_OFFSET(XPV, xpv_cur),
984 + STRUCT_OFFSET(XPV, xpv_cur),
985 SVt_INVLIST, TRUE, NONV, HASARENA,
986 FIT_ARENA(0, sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur)) },
988 { sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur),
989 copy_length(XPVIV, xiv_u) - STRUCT_OFFSET(XPV, xpv_cur),
990 + STRUCT_OFFSET(XPV, xpv_cur),
991 SVt_PVIV, FALSE, NONV, HASARENA,
992 FIT_ARENA(0, sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur)) },
994 { sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur),
995 copy_length(XPVNV, xnv_u) - STRUCT_OFFSET(XPV, xpv_cur),
996 + STRUCT_OFFSET(XPV, xpv_cur),
997 SVt_PVNV, FALSE, HADNV, HASARENA,
998 FIT_ARENA(0, sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur)) },
1000 { sizeof(XPVMG), copy_length(XPVMG, xnv_u), 0, SVt_PVMG, FALSE, HADNV,
1001 HASARENA, FIT_ARENA(0, sizeof(XPVMG)) },
1006 SVt_REGEXP, TRUE, NONV, HASARENA,
1007 FIT_ARENA(0, sizeof(regexp))
1010 { sizeof(XPVGV), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV,
1011 HASARENA, FIT_ARENA(0, sizeof(XPVGV)) },
1013 { sizeof(XPVLV), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV,
1014 HASARENA, FIT_ARENA(0, sizeof(XPVLV)) },
1017 copy_length(XPVAV, xav_alloc),
1019 SVt_PVAV, TRUE, NONV, HASARENA,
1020 FIT_ARENA(0, sizeof(XPVAV)) },
1023 copy_length(XPVHV, xhv_max),
1025 SVt_PVHV, TRUE, NONV, HASARENA,
1026 FIT_ARENA(0, sizeof(XPVHV)) },
1031 SVt_PVCV, TRUE, NONV, HASARENA,
1032 FIT_ARENA(0, sizeof(XPVCV)) },
1037 SVt_PVFM, TRUE, NONV, NOARENA,
1038 FIT_ARENA(20, sizeof(XPVFM)) },
1043 SVt_PVIO, TRUE, NONV, HASARENA,
1044 FIT_ARENA(24, sizeof(XPVIO)) },
1047 #define new_body_allocated(sv_type) \
1048 (void *)((char *)S_new_body(aTHX_ sv_type) \
1049 - bodies_by_type[sv_type].offset)
1051 /* return a thing to the free list */
1053 #define del_body(thing, root) \
1055 void ** const thing_copy = (void **)thing; \
1056 *thing_copy = *root; \
1057 *root = (void*)thing_copy; \
1061 #if !(NVSIZE <= IVSIZE)
1062 # define new_XNV() safemalloc(sizeof(XPVNV))
1064 #define new_XPVNV() safemalloc(sizeof(XPVNV))
1065 #define new_XPVMG() safemalloc(sizeof(XPVMG))
1067 #define del_XPVGV(p) safefree(p)
1071 #if !(NVSIZE <= IVSIZE)
1072 # define new_XNV() new_body_allocated(SVt_NV)
1074 #define new_XPVNV() new_body_allocated(SVt_PVNV)
1075 #define new_XPVMG() new_body_allocated(SVt_PVMG)
1077 #define del_XPVGV(p) del_body(p + bodies_by_type[SVt_PVGV].offset, \
1078 &PL_body_roots[SVt_PVGV])
1082 /* no arena for you! */
1084 #define new_NOARENA(details) \
1085 safemalloc((details)->body_size + (details)->offset)
1086 #define new_NOARENAZ(details) \
1087 safecalloc((details)->body_size + (details)->offset, 1)
1090 Perl_more_bodies (pTHX_ const svtype sv_type, const size_t body_size,
1091 const size_t arena_size)
1093 void ** const root = &PL_body_roots[sv_type];
1094 struct arena_desc *adesc;
1095 struct arena_set *aroot = (struct arena_set *) PL_body_arenas;
1099 const size_t good_arena_size = Perl_malloc_good_size(arena_size);
1100 #if defined(DEBUGGING) && defined(PERL_GLOBAL_STRUCT)
1103 #if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT_PRIVATE)
1104 static bool done_sanity_check;
1106 /* PERL_GLOBAL_STRUCT_PRIVATE cannot coexist with global
1107 * variables like done_sanity_check. */
1108 if (!done_sanity_check) {
1109 unsigned int i = SVt_LAST;
1111 done_sanity_check = TRUE;
1114 assert (bodies_by_type[i].type == i);
1120 /* may need new arena-set to hold new arena */
1121 if (!aroot || aroot->curr >= aroot->set_size) {
1122 struct arena_set *newroot;
1123 Newxz(newroot, 1, struct arena_set);
1124 newroot->set_size = ARENAS_PER_SET;
1125 newroot->next = aroot;
1127 PL_body_arenas = (void *) newroot;
1128 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot));
1131 /* ok, now have arena-set with at least 1 empty/available arena-desc */
1132 curr = aroot->curr++;
1133 adesc = &(aroot->set[curr]);
1134 assert(!adesc->arena);
1136 Newx(adesc->arena, good_arena_size, char);
1137 adesc->size = good_arena_size;
1138 adesc->utype = sv_type;
1139 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %"UVuf"\n",
1140 curr, (void*)adesc->arena, (UV)good_arena_size));
1142 start = (char *) adesc->arena;
1144 /* Get the address of the byte after the end of the last body we can fit.
1145 Remember, this is integer division: */
1146 end = start + good_arena_size / body_size * body_size;
1148 /* computed count doesn't reflect the 1st slot reservation */
1149 #if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE)
1150 DEBUG_m(PerlIO_printf(Perl_debug_log,
1151 "arena %p end %p arena-size %d (from %d) type %d "
1153 (void*)start, (void*)end, (int)good_arena_size,
1154 (int)arena_size, sv_type, (int)body_size,
1155 (int)good_arena_size / (int)body_size));
1157 DEBUG_m(PerlIO_printf(Perl_debug_log,
1158 "arena %p end %p arena-size %d type %d size %d ct %d\n",
1159 (void*)start, (void*)end,
1160 (int)arena_size, sv_type, (int)body_size,
1161 (int)good_arena_size / (int)body_size));
1163 *root = (void *)start;
1166 /* Where the next body would start: */
1167 char * const next = start + body_size;
1170 /* This is the last body: */
1171 assert(next == end);
1173 *(void **)start = 0;
1177 *(void**) start = (void *)next;
1182 /* grab a new thing from the free list, allocating more if necessary.
1183 The inline version is used for speed in hot routines, and the
1184 function using it serves the rest (unless PURIFY).
1186 #define new_body_inline(xpv, sv_type) \
1188 void ** const r3wt = &PL_body_roots[sv_type]; \
1189 xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \
1190 ? *((void **)(r3wt)) : Perl_more_bodies(aTHX_ sv_type, \
1191 bodies_by_type[sv_type].body_size,\
1192 bodies_by_type[sv_type].arena_size)); \
1193 *(r3wt) = *(void**)(xpv); \
1199 S_new_body(pTHX_ const svtype sv_type)
1202 new_body_inline(xpv, sv_type);
1208 static const struct body_details fake_rv =
1209 { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 };
1212 =for apidoc sv_upgrade
1214 Upgrade an SV to a more complex form. Generally adds a new body type to the
1215 SV, then copies across as much information as possible from the old body.
1216 It croaks if the SV is already in a more complex form than requested. You
1217 generally want to use the C<SvUPGRADE> macro wrapper, which checks the type
1218 before calling C<sv_upgrade>, and hence does not croak. See also
1225 Perl_sv_upgrade(pTHX_ SV *const sv, svtype new_type)
1229 const svtype old_type = SvTYPE(sv);
1230 const struct body_details *new_type_details;
1231 const struct body_details *old_type_details
1232 = bodies_by_type + old_type;
1233 SV *referant = NULL;
1235 PERL_ARGS_ASSERT_SV_UPGRADE;
1237 if (old_type == new_type)
1240 /* This clause was purposefully added ahead of the early return above to
1241 the shared string hackery for (sort {$a <=> $b} keys %hash), with the
1242 inference by Nick I-S that it would fix other troublesome cases. See
1243 changes 7162, 7163 (f130fd4589cf5fbb24149cd4db4137c8326f49c1 and parent)
1245 Given that shared hash key scalars are no longer PVIV, but PV, there is
1246 no longer need to unshare so as to free up the IVX slot for its proper
1247 purpose. So it's safe to move the early return earlier. */
1249 if (new_type > SVt_PVMG && SvIsCOW(sv)) {
1250 sv_force_normal_flags(sv, 0);
1253 old_body = SvANY(sv);
1255 /* Copying structures onto other structures that have been neatly zeroed
1256 has a subtle gotcha. Consider XPVMG
1258 +------+------+------+------+------+-------+-------+
1259 | NV | CUR | LEN | IV | MAGIC | STASH |
1260 +------+------+------+------+------+-------+-------+
1261 0 4 8 12 16 20 24 28
1263 where NVs are aligned to 8 bytes, so that sizeof that structure is
1264 actually 32 bytes long, with 4 bytes of padding at the end:
1266 +------+------+------+------+------+-------+-------+------+
1267 | NV | CUR | LEN | IV | MAGIC | STASH | ??? |
1268 +------+------+------+------+------+-------+-------+------+
1269 0 4 8 12 16 20 24 28 32
1271 so what happens if you allocate memory for this structure:
1273 +------+------+------+------+------+-------+-------+------+------+...
1274 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME |
1275 +------+------+------+------+------+-------+-------+------+------+...
1276 0 4 8 12 16 20 24 28 32 36
1278 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you
1279 expect, because you copy the area marked ??? onto GP. Now, ??? may have
1280 started out as zero once, but it's quite possible that it isn't. So now,
1281 rather than a nicely zeroed GP, you have it pointing somewhere random.
1284 (In fact, GP ends up pointing at a previous GP structure, because the
1285 principle cause of the padding in XPVMG getting garbage is a copy of
1286 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now
1287 this happens to be moot because XPVGV has been re-ordered, with GP
1288 no longer after STASH)
1290 So we are careful and work out the size of used parts of all the
1298 referant = SvRV(sv);
1299 old_type_details = &fake_rv;
1300 if (new_type == SVt_NV)
1301 new_type = SVt_PVNV;
1303 if (new_type < SVt_PVIV) {
1304 new_type = (new_type == SVt_NV)
1305 ? SVt_PVNV : SVt_PVIV;
1310 if (new_type < SVt_PVNV) {
1311 new_type = SVt_PVNV;
1315 assert(new_type > SVt_PV);
1316 STATIC_ASSERT_STMT(SVt_IV < SVt_PV);
1317 STATIC_ASSERT_STMT(SVt_NV < SVt_PV);
1324 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena,
1325 there's no way that it can be safely upgraded, because perl.c
1326 expects to Safefree(SvANY(PL_mess_sv)) */
1327 assert(sv != PL_mess_sv);
1330 if (UNLIKELY(old_type_details->cant_upgrade))
1331 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf,
1332 sv_reftype(sv, 0), (UV) old_type, (UV) new_type);
1335 if (UNLIKELY(old_type > new_type))
1336 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d",
1337 (int)old_type, (int)new_type);
1339 new_type_details = bodies_by_type + new_type;
1341 SvFLAGS(sv) &= ~SVTYPEMASK;
1342 SvFLAGS(sv) |= new_type;
1344 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of
1345 the return statements above will have triggered. */
1346 assert (new_type != SVt_NULL);
1349 assert(old_type == SVt_NULL);
1350 SET_SVANY_FOR_BODYLESS_IV(sv);
1354 assert(old_type == SVt_NULL);
1355 #if NVSIZE <= IVSIZE
1356 SET_SVANY_FOR_BODYLESS_NV(sv);
1358 SvANY(sv) = new_XNV();
1364 assert(new_type_details->body_size);
1367 assert(new_type_details->arena);
1368 assert(new_type_details->arena_size);
1369 /* This points to the start of the allocated area. */
1370 new_body_inline(new_body, new_type);
1371 Zero(new_body, new_type_details->body_size, char);
1372 new_body = ((char *)new_body) - new_type_details->offset;
1374 /* We always allocated the full length item with PURIFY. To do this
1375 we fake things so that arena is false for all 16 types.. */
1376 new_body = new_NOARENAZ(new_type_details);
1378 SvANY(sv) = new_body;
1379 if (new_type == SVt_PVAV) {
1383 if (old_type_details->body_size) {
1386 /* It will have been zeroed when the new body was allocated.
1387 Lets not write to it, in case it confuses a write-back
1393 #ifndef NODEFAULT_SHAREKEYS
1394 HvSHAREKEYS_on(sv); /* key-sharing on by default */
1396 /* start with PERL_HASH_DEFAULT_HvMAX+1 buckets: */
1397 HvMAX(sv) = PERL_HASH_DEFAULT_HvMAX;
1400 /* SVt_NULL isn't the only thing upgraded to AV or HV.
1401 The target created by newSVrv also is, and it can have magic.
1402 However, it never has SvPVX set.
1404 if (old_type == SVt_IV) {
1406 } else if (old_type >= SVt_PV) {
1407 assert(SvPVX_const(sv) == 0);
1410 if (old_type >= SVt_PVMG) {
1411 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic);
1412 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash);
1414 sv->sv_u.svu_array = NULL; /* or svu_hash */
1419 /* XXX Is this still needed? Was it ever needed? Surely as there is
1420 no route from NV to PVIV, NOK can never be true */
1421 assert(!SvNOKp(sv));
1435 assert(new_type_details->body_size);
1436 /* We always allocated the full length item with PURIFY. To do this
1437 we fake things so that arena is false for all 16 types.. */
1438 if(new_type_details->arena) {
1439 /* This points to the start of the allocated area. */
1440 new_body_inline(new_body, new_type);
1441 Zero(new_body, new_type_details->body_size, char);
1442 new_body = ((char *)new_body) - new_type_details->offset;
1444 new_body = new_NOARENAZ(new_type_details);
1446 SvANY(sv) = new_body;
1448 if (old_type_details->copy) {
1449 /* There is now the potential for an upgrade from something without
1450 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */
1451 int offset = old_type_details->offset;
1452 int length = old_type_details->copy;
1454 if (new_type_details->offset > old_type_details->offset) {
1455 const int difference
1456 = new_type_details->offset - old_type_details->offset;
1457 offset += difference;
1458 length -= difference;
1460 assert (length >= 0);
1462 Copy((char *)old_body + offset, (char *)new_body + offset, length,
1466 #ifndef NV_ZERO_IS_ALLBITS_ZERO
1467 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a
1468 * correct 0.0 for us. Otherwise, if the old body didn't have an
1469 * NV slot, but the new one does, then we need to initialise the
1470 * freshly created NV slot with whatever the correct bit pattern is
1472 if (old_type_details->zero_nv && !new_type_details->zero_nv
1473 && !isGV_with_GP(sv))
1477 if (UNLIKELY(new_type == SVt_PVIO)) {
1478 IO * const io = MUTABLE_IO(sv);
1479 GV *iogv = gv_fetchpvs("IO::File::", GV_ADD, SVt_PVHV);
1482 /* Clear the stashcache because a new IO could overrule a package
1484 DEBUG_o(Perl_deb(aTHX_ "sv_upgrade clearing PL_stashcache\n"));
1485 hv_clear(PL_stashcache);
1487 SvSTASH_set(io, MUTABLE_HV(SvREFCNT_inc(GvHV(iogv))));
1488 IoPAGE_LEN(sv) = 60;
1490 if (UNLIKELY(new_type == SVt_REGEXP))
1491 sv->sv_u.svu_rx = (regexp *)new_body;
1492 else if (old_type < SVt_PV) {
1493 /* referant will be NULL unless the old type was SVt_IV emulating
1495 sv->sv_u.svu_rv = referant;
1499 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1500 (unsigned long)new_type);
1503 /* if this is zero, this is a body-less SVt_NULL, SVt_IV/SVt_RV,
1504 and sometimes SVt_NV */
1505 if (old_type_details->body_size) {
1509 /* Note that there is an assumption that all bodies of types that
1510 can be upgraded came from arenas. Only the more complex non-
1511 upgradable types are allowed to be directly malloc()ed. */
1512 assert(old_type_details->arena);
1513 del_body((void*)((char*)old_body + old_type_details->offset),
1514 &PL_body_roots[old_type]);
1520 =for apidoc sv_backoff
1522 Remove any string offset. You should normally use the C<SvOOK_off> macro
1529 Perl_sv_backoff(SV *const sv)
1532 const char * const s = SvPVX_const(sv);
1534 PERL_ARGS_ASSERT_SV_BACKOFF;
1537 assert(SvTYPE(sv) != SVt_PVHV);
1538 assert(SvTYPE(sv) != SVt_PVAV);
1540 SvOOK_offset(sv, delta);
1542 SvLEN_set(sv, SvLEN(sv) + delta);
1543 SvPV_set(sv, SvPVX(sv) - delta);
1544 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
1545 SvFLAGS(sv) &= ~SVf_OOK;
1552 Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1553 upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1554 Use the C<SvGROW> wrapper instead.
1559 static void S_sv_uncow(pTHX_ SV * const sv, const U32 flags);
1562 Perl_sv_grow(pTHX_ SV *const sv, STRLEN newlen)
1566 PERL_ARGS_ASSERT_SV_GROW;
1570 if (SvTYPE(sv) < SVt_PV) {
1571 sv_upgrade(sv, SVt_PV);
1572 s = SvPVX_mutable(sv);
1574 else if (SvOOK(sv)) { /* pv is offset? */
1576 s = SvPVX_mutable(sv);
1577 if (newlen > SvLEN(sv))
1578 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
1582 if (SvIsCOW(sv)) S_sv_uncow(aTHX_ sv, 0);
1583 s = SvPVX_mutable(sv);
1586 #ifdef PERL_COPY_ON_WRITE
1587 /* the new COW scheme uses SvPVX(sv)[SvLEN(sv)-1] (if spare)
1588 * to store the COW count. So in general, allocate one more byte than
1589 * asked for, to make it likely this byte is always spare: and thus
1590 * make more strings COW-able.
1591 * If the new size is a big power of two, don't bother: we assume the
1592 * caller wanted a nice 2^N sized block and will be annoyed at getting
1594 * Only increment if the allocation isn't MEM_SIZE_MAX,
1595 * otherwise it will wrap to 0.
1597 if (newlen & 0xff && newlen != MEM_SIZE_MAX)
1601 #if defined(PERL_USE_MALLOC_SIZE) && defined(Perl_safesysmalloc_size)
1602 #define PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1605 if (newlen > SvLEN(sv)) { /* need more room? */
1606 STRLEN minlen = SvCUR(sv);
1607 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
1608 if (newlen < minlen)
1610 #ifndef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1612 /* Don't round up on the first allocation, as odds are pretty good that
1613 * the initial request is accurate as to what is really needed */
1615 STRLEN rounded = PERL_STRLEN_ROUNDUP(newlen);
1616 if (rounded > newlen)
1620 if (SvLEN(sv) && s) {
1621 s = (char*)saferealloc(s, newlen);
1624 s = (char*)safemalloc(newlen);
1625 if (SvPVX_const(sv) && SvCUR(sv)) {
1626 Move(SvPVX_const(sv), s, (newlen < SvCUR(sv)) ? newlen : SvCUR(sv), char);
1630 #ifdef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1631 /* Do this here, do it once, do it right, and then we will never get
1632 called back into sv_grow() unless there really is some growing
1634 SvLEN_set(sv, Perl_safesysmalloc_size(s));
1636 SvLEN_set(sv, newlen);
1643 =for apidoc sv_setiv
1645 Copies an integer into the given SV, upgrading first if necessary.
1646 Does not handle 'set' magic. See also C<L</sv_setiv_mg>>.
1652 Perl_sv_setiv(pTHX_ SV *const sv, const IV i)
1654 PERL_ARGS_ASSERT_SV_SETIV;
1656 SV_CHECK_THINKFIRST_COW_DROP(sv);
1657 switch (SvTYPE(sv)) {
1660 sv_upgrade(sv, SVt_IV);
1663 sv_upgrade(sv, SVt_PVIV);
1667 if (!isGV_with_GP(sv))
1674 /* diag_listed_as: Can't coerce %s to %s in %s */
1675 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1680 (void)SvIOK_only(sv); /* validate number */
1686 =for apidoc sv_setiv_mg
1688 Like C<sv_setiv>, but also handles 'set' magic.
1694 Perl_sv_setiv_mg(pTHX_ SV *const sv, const IV i)
1696 PERL_ARGS_ASSERT_SV_SETIV_MG;
1703 =for apidoc sv_setuv
1705 Copies an unsigned integer into the given SV, upgrading first if necessary.
1706 Does not handle 'set' magic. See also C<L</sv_setuv_mg>>.
1712 Perl_sv_setuv(pTHX_ SV *const sv, const UV u)
1714 PERL_ARGS_ASSERT_SV_SETUV;
1716 /* With the if statement to ensure that integers are stored as IVs whenever
1718 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
1721 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1723 If you wish to remove the following if statement, so that this routine
1724 (and its callers) always return UVs, please benchmark to see what the
1725 effect is. Modern CPUs may be different. Or may not :-)
1727 if (u <= (UV)IV_MAX) {
1728 sv_setiv(sv, (IV)u);
1737 =for apidoc sv_setuv_mg
1739 Like C<sv_setuv>, but also handles 'set' magic.
1745 Perl_sv_setuv_mg(pTHX_ SV *const sv, const UV u)
1747 PERL_ARGS_ASSERT_SV_SETUV_MG;
1754 =for apidoc sv_setnv
1756 Copies a double into the given SV, upgrading first if necessary.
1757 Does not handle 'set' magic. See also C<L</sv_setnv_mg>>.
1763 Perl_sv_setnv(pTHX_ SV *const sv, const NV num)
1765 PERL_ARGS_ASSERT_SV_SETNV;
1767 SV_CHECK_THINKFIRST_COW_DROP(sv);
1768 switch (SvTYPE(sv)) {
1771 sv_upgrade(sv, SVt_NV);
1775 sv_upgrade(sv, SVt_PVNV);
1779 if (!isGV_with_GP(sv))
1786 /* diag_listed_as: Can't coerce %s to %s in %s */
1787 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
1793 (void)SvNOK_only(sv); /* validate number */
1798 =for apidoc sv_setnv_mg
1800 Like C<sv_setnv>, but also handles 'set' magic.
1806 Perl_sv_setnv_mg(pTHX_ SV *const sv, const NV num)
1808 PERL_ARGS_ASSERT_SV_SETNV_MG;
1814 /* Return a cleaned-up, printable version of sv, for non-numeric, or
1815 * not incrementable warning display.
1816 * Originally part of S_not_a_number().
1817 * The return value may be != tmpbuf.
1821 S_sv_display(pTHX_ SV *const sv, char *tmpbuf, STRLEN tmpbuf_size) {
1824 PERL_ARGS_ASSERT_SV_DISPLAY;
1827 SV *dsv = newSVpvs_flags("", SVs_TEMP);
1828 pv = sv_uni_display(dsv, sv, 32, UNI_DISPLAY_ISPRINT);
1831 const char * const limit = tmpbuf + tmpbuf_size - 8;
1832 /* each *s can expand to 4 chars + "...\0",
1833 i.e. need room for 8 chars */
1835 const char *s = SvPVX_const(sv);
1836 const char * const end = s + SvCUR(sv);
1837 for ( ; s < end && d < limit; s++ ) {
1839 if (! isASCII(ch) && !isPRINT_LC(ch)) {
1843 /* Map to ASCII "equivalent" of Latin1 */
1844 ch = LATIN1_TO_NATIVE(NATIVE_TO_LATIN1(ch) & 127);
1850 else if (ch == '\r') {
1854 else if (ch == '\f') {
1858 else if (ch == '\\') {
1862 else if (ch == '\0') {
1866 else if (isPRINT_LC(ch))
1885 /* Print an "isn't numeric" warning, using a cleaned-up,
1886 * printable version of the offending string
1890 S_not_a_number(pTHX_ SV *const sv)
1895 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1897 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1900 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1901 /* diag_listed_as: Argument "%s" isn't numeric%s */
1902 "Argument \"%s\" isn't numeric in %s", pv,
1905 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1906 /* diag_listed_as: Argument "%s" isn't numeric%s */
1907 "Argument \"%s\" isn't numeric", pv);
1911 S_not_incrementable(pTHX_ SV *const sv) {
1915 PERL_ARGS_ASSERT_NOT_INCREMENTABLE;
1917 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1919 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1920 "Argument \"%s\" treated as 0 in increment (++)", pv);
1924 =for apidoc looks_like_number
1926 Test if the content of an SV looks like a number (or is a number).
1927 C<Inf> and C<Infinity> are treated as numbers (so will not issue a
1928 non-numeric warning), even if your C<atof()> doesn't grok them. Get-magic is
1935 Perl_looks_like_number(pTHX_ SV *const sv)
1941 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1943 if (SvPOK(sv) || SvPOKp(sv)) {
1944 sbegin = SvPV_nomg_const(sv, len);
1947 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
1948 numtype = grok_number(sbegin, len, NULL);
1949 return ((numtype & IS_NUMBER_TRAILING)) ? 0 : numtype;
1953 S_glob_2number(pTHX_ GV * const gv)
1955 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1957 /* We know that all GVs stringify to something that is not-a-number,
1958 so no need to test that. */
1959 if (ckWARN(WARN_NUMERIC))
1961 SV *const buffer = sv_newmortal();
1962 gv_efullname3(buffer, gv, "*");
1963 not_a_number(buffer);
1965 /* We just want something true to return, so that S_sv_2iuv_common
1966 can tail call us and return true. */
1970 /* Actually, ISO C leaves conversion of UV to IV undefined, but
1971 until proven guilty, assume that things are not that bad... */
1976 As 64 bit platforms often have an NV that doesn't preserve all bits of
1977 an IV (an assumption perl has been based on to date) it becomes necessary
1978 to remove the assumption that the NV always carries enough precision to
1979 recreate the IV whenever needed, and that the NV is the canonical form.
1980 Instead, IV/UV and NV need to be given equal rights. So as to not lose
1981 precision as a side effect of conversion (which would lead to insanity
1982 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
1983 1) to distinguish between IV/UV/NV slots that have a valid conversion cached
1984 where precision was lost, and IV/UV/NV slots that have a valid conversion
1985 which has lost no precision
1986 2) to ensure that if a numeric conversion to one form is requested that
1987 would lose precision, the precise conversion (or differently
1988 imprecise conversion) is also performed and cached, to prevent
1989 requests for different numeric formats on the same SV causing
1990 lossy conversion chains. (lossless conversion chains are perfectly
1995 SvIOKp is true if the IV slot contains a valid value
1996 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1997 SvNOKp is true if the NV slot contains a valid value
1998 SvNOK is true only if the NV value is accurate
2001 while converting from PV to NV, check to see if converting that NV to an
2002 IV(or UV) would lose accuracy over a direct conversion from PV to
2003 IV(or UV). If it would, cache both conversions, return NV, but mark
2004 SV as IOK NOKp (ie not NOK).
2006 While converting from PV to IV, check to see if converting that IV to an
2007 NV would lose accuracy over a direct conversion from PV to NV. If it
2008 would, cache both conversions, flag similarly.
2010 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
2011 correctly because if IV & NV were set NV *always* overruled.
2012 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
2013 changes - now IV and NV together means that the two are interchangeable:
2014 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
2016 The benefit of this is that operations such as pp_add know that if
2017 SvIOK is true for both left and right operands, then integer addition
2018 can be used instead of floating point (for cases where the result won't
2019 overflow). Before, floating point was always used, which could lead to
2020 loss of precision compared with integer addition.
2022 * making IV and NV equal status should make maths accurate on 64 bit
2024 * may speed up maths somewhat if pp_add and friends start to use
2025 integers when possible instead of fp. (Hopefully the overhead in
2026 looking for SvIOK and checking for overflow will not outweigh the
2027 fp to integer speedup)
2028 * will slow down integer operations (callers of SvIV) on "inaccurate"
2029 values, as the change from SvIOK to SvIOKp will cause a call into
2030 sv_2iv each time rather than a macro access direct to the IV slot
2031 * should speed up number->string conversion on integers as IV is
2032 favoured when IV and NV are equally accurate
2034 ####################################################################
2035 You had better be using SvIOK_notUV if you want an IV for arithmetic:
2036 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
2037 On the other hand, SvUOK is true iff UV.
2038 ####################################################################
2040 Your mileage will vary depending your CPU's relative fp to integer
2044 #ifndef NV_PRESERVES_UV
2045 # define IS_NUMBER_UNDERFLOW_IV 1
2046 # define IS_NUMBER_UNDERFLOW_UV 2
2047 # define IS_NUMBER_IV_AND_UV 2
2048 # define IS_NUMBER_OVERFLOW_IV 4
2049 # define IS_NUMBER_OVERFLOW_UV 5
2051 /* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
2053 /* For sv_2nv these three cases are "SvNOK and don't bother casting" */
2055 S_sv_2iuv_non_preserve(pTHX_ SV *const sv
2061 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
2062 PERL_UNUSED_CONTEXT;
2064 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));
2065 if (SvNVX(sv) < (NV)IV_MIN) {
2066 (void)SvIOKp_on(sv);
2068 SvIV_set(sv, IV_MIN);
2069 return IS_NUMBER_UNDERFLOW_IV;
2071 if (SvNVX(sv) > (NV)UV_MAX) {
2072 (void)SvIOKp_on(sv);
2075 SvUV_set(sv, UV_MAX);
2076 return IS_NUMBER_OVERFLOW_UV;
2078 (void)SvIOKp_on(sv);
2080 /* Can't use strtol etc to convert this string. (See truth table in
2082 if (SvNVX(sv) <= (UV)IV_MAX) {
2083 SvIV_set(sv, I_V(SvNVX(sv)));
2084 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2085 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
2087 /* Integer is imprecise. NOK, IOKp */
2089 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
2092 SvUV_set(sv, U_V(SvNVX(sv)));
2093 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2094 if (SvUVX(sv) == UV_MAX) {
2095 /* As we know that NVs don't preserve UVs, UV_MAX cannot
2096 possibly be preserved by NV. Hence, it must be overflow.
2098 return IS_NUMBER_OVERFLOW_UV;
2100 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
2102 /* Integer is imprecise. NOK, IOKp */
2104 return IS_NUMBER_OVERFLOW_IV;
2106 #endif /* !NV_PRESERVES_UV*/
2108 /* If numtype is infnan, set the NV of the sv accordingly.
2109 * If numtype is anything else, try setting the NV using Atof(PV). */
2111 # pragma warning(push)
2112 # pragma warning(disable:4756;disable:4056)
2115 S_sv_setnv(pTHX_ SV* sv, int numtype)
2117 bool pok = cBOOL(SvPOK(sv));
2119 if ((numtype & IS_NUMBER_INFINITY)) {
2120 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -NV_INF : NV_INF);
2123 else if ((numtype & IS_NUMBER_NAN)) {
2124 SvNV_set(sv, NV_NAN);
2128 SvNV_set(sv, Atof(SvPVX_const(sv)));
2129 /* Purposefully no true nok here, since we don't want to blow
2130 * away the possible IOK/UV of an existing sv. */
2133 SvNOK_only(sv); /* No IV or UV please, this is pure infnan. */
2135 SvPOK_on(sv); /* PV is okay, though. */
2139 # pragma warning(pop)
2143 S_sv_2iuv_common(pTHX_ SV *const sv)
2145 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
2148 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
2149 * without also getting a cached IV/UV from it at the same time
2150 * (ie PV->NV conversion should detect loss of accuracy and cache
2151 * IV or UV at same time to avoid this. */
2152 /* IV-over-UV optimisation - choose to cache IV if possible */
2154 if (SvTYPE(sv) == SVt_NV)
2155 sv_upgrade(sv, SVt_PVNV);
2157 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
2158 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
2159 certainly cast into the IV range at IV_MAX, whereas the correct
2160 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
2162 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2163 if (Perl_isnan(SvNVX(sv))) {
2169 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2170 SvIV_set(sv, I_V(SvNVX(sv)));
2171 if (SvNVX(sv) == (NV) SvIVX(sv)
2172 #ifndef NV_PRESERVES_UV
2173 && SvIVX(sv) != IV_MIN /* avoid negating IV_MIN below */
2174 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2175 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2176 /* Don't flag it as "accurately an integer" if the number
2177 came from a (by definition imprecise) NV operation, and
2178 we're outside the range of NV integer precision */
2182 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2184 /* scalar has trailing garbage, eg "42a" */
2186 DEBUG_c(PerlIO_printf(Perl_debug_log,
2187 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (precise)\n",
2193 /* IV not precise. No need to convert from PV, as NV
2194 conversion would already have cached IV if it detected
2195 that PV->IV would be better than PV->NV->IV
2196 flags already correct - don't set public IOK. */
2197 DEBUG_c(PerlIO_printf(Perl_debug_log,
2198 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (imprecise)\n",
2203 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2204 but the cast (NV)IV_MIN rounds to a the value less (more
2205 negative) than IV_MIN which happens to be equal to SvNVX ??
2206 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2207 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2208 (NV)UVX == NVX are both true, but the values differ. :-(
2209 Hopefully for 2s complement IV_MIN is something like
2210 0x8000000000000000 which will be exact. NWC */
2213 SvUV_set(sv, U_V(SvNVX(sv)));
2215 (SvNVX(sv) == (NV) SvUVX(sv))
2216 #ifndef NV_PRESERVES_UV
2217 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2218 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2219 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2220 /* Don't flag it as "accurately an integer" if the number
2221 came from a (by definition imprecise) NV operation, and
2222 we're outside the range of NV integer precision */
2228 DEBUG_c(PerlIO_printf(Perl_debug_log,
2229 "0x%"UVxf" 2iv(%"UVuf" => %"IVdf") (as unsigned)\n",
2235 else if (SvPOKp(sv)) {
2237 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2238 /* We want to avoid a possible problem when we cache an IV/ a UV which
2239 may be later translated to an NV, and the resulting NV is not
2240 the same as the direct translation of the initial string
2241 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2242 be careful to ensure that the value with the .456 is around if the
2243 NV value is requested in the future).
2245 This means that if we cache such an IV/a UV, we need to cache the
2246 NV as well. Moreover, we trade speed for space, and do not
2247 cache the NV if we are sure it's not needed.
2250 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2251 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2252 == IS_NUMBER_IN_UV) {
2253 /* It's definitely an integer, only upgrade to PVIV */
2254 if (SvTYPE(sv) < SVt_PVIV)
2255 sv_upgrade(sv, SVt_PVIV);
2257 } else if (SvTYPE(sv) < SVt_PVNV)
2258 sv_upgrade(sv, SVt_PVNV);
2260 if ((numtype & (IS_NUMBER_INFINITY | IS_NUMBER_NAN))) {
2261 if (ckWARN(WARN_NUMERIC) && ((numtype & IS_NUMBER_TRAILING)))
2263 S_sv_setnv(aTHX_ sv, numtype);
2267 /* If NVs preserve UVs then we only use the UV value if we know that
2268 we aren't going to call atof() below. If NVs don't preserve UVs
2269 then the value returned may have more precision than atof() will
2270 return, even though value isn't perfectly accurate. */
2271 if ((numtype & (IS_NUMBER_IN_UV
2272 #ifdef NV_PRESERVES_UV
2275 )) == IS_NUMBER_IN_UV) {
2276 /* This won't turn off the public IOK flag if it was set above */
2277 (void)SvIOKp_on(sv);
2279 if (!(numtype & IS_NUMBER_NEG)) {
2281 if (value <= (UV)IV_MAX) {
2282 SvIV_set(sv, (IV)value);
2284 /* it didn't overflow, and it was positive. */
2285 SvUV_set(sv, value);
2289 /* 2s complement assumption */
2290 if (value <= (UV)IV_MIN) {
2291 SvIV_set(sv, value == (UV)IV_MIN
2292 ? IV_MIN : -(IV)value);
2294 /* Too negative for an IV. This is a double upgrade, but
2295 I'm assuming it will be rare. */
2296 if (SvTYPE(sv) < SVt_PVNV)
2297 sv_upgrade(sv, SVt_PVNV);
2301 SvNV_set(sv, -(NV)value);
2302 SvIV_set(sv, IV_MIN);
2306 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2307 will be in the previous block to set the IV slot, and the next
2308 block to set the NV slot. So no else here. */
2310 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2311 != IS_NUMBER_IN_UV) {
2312 /* It wasn't an (integer that doesn't overflow the UV). */
2313 S_sv_setnv(aTHX_ sv, numtype);
2315 if (! numtype && ckWARN(WARN_NUMERIC))
2318 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%" NVgf ")\n",
2319 PTR2UV(sv), SvNVX(sv)));
2321 #ifdef NV_PRESERVES_UV
2322 (void)SvIOKp_on(sv);
2324 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2325 if (Perl_isnan(SvNVX(sv))) {
2331 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2332 SvIV_set(sv, I_V(SvNVX(sv)));
2333 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2336 NOOP; /* Integer is imprecise. NOK, IOKp */
2338 /* UV will not work better than IV */
2340 if (SvNVX(sv) > (NV)UV_MAX) {
2342 /* Integer is inaccurate. NOK, IOKp, is UV */
2343 SvUV_set(sv, UV_MAX);
2345 SvUV_set(sv, U_V(SvNVX(sv)));
2346 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2347 NV preservse UV so can do correct comparison. */
2348 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2351 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
2356 #else /* NV_PRESERVES_UV */
2357 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2358 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
2359 /* The IV/UV slot will have been set from value returned by
2360 grok_number above. The NV slot has just been set using
2363 assert (SvIOKp(sv));
2365 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2366 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2367 /* Small enough to preserve all bits. */
2368 (void)SvIOKp_on(sv);
2370 SvIV_set(sv, I_V(SvNVX(sv)));
2371 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2373 /* Assumption: first non-preserved integer is < IV_MAX,
2374 this NV is in the preserved range, therefore: */
2375 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2377 Perl_croak(aTHX_ "sv_2iv assumed (U_V(fabs((double)SvNVX(sv))) < (UV)IV_MAX) but SvNVX(sv)=%"NVgf" U_V is 0x%"UVxf", IV_MAX is 0x%"UVxf"\n", SvNVX(sv), U_V(SvNVX(sv)), (UV)IV_MAX);
2381 0 0 already failed to read UV.
2382 0 1 already failed to read UV.
2383 1 0 you won't get here in this case. IV/UV
2384 slot set, public IOK, Atof() unneeded.
2385 1 1 already read UV.
2386 so there's no point in sv_2iuv_non_preserve() attempting
2387 to use atol, strtol, strtoul etc. */
2389 sv_2iuv_non_preserve (sv, numtype);
2391 sv_2iuv_non_preserve (sv);
2395 #endif /* NV_PRESERVES_UV */
2396 /* It might be more code efficient to go through the entire logic above
2397 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2398 gets complex and potentially buggy, so more programmer efficient
2399 to do it this way, by turning off the public flags: */
2401 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2405 if (isGV_with_GP(sv))
2406 return glob_2number(MUTABLE_GV(sv));
2408 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2410 if (SvTYPE(sv) < SVt_IV)
2411 /* Typically the caller expects that sv_any is not NULL now. */
2412 sv_upgrade(sv, SVt_IV);
2413 /* Return 0 from the caller. */
2420 =for apidoc sv_2iv_flags
2422 Return the integer value of an SV, doing any necessary string
2423 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2424 Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2430 Perl_sv_2iv_flags(pTHX_ SV *const sv, const I32 flags)
2432 PERL_ARGS_ASSERT_SV_2IV_FLAGS;
2434 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2435 && SvTYPE(sv) != SVt_PVFM);
2437 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2443 if (flags & SV_SKIP_OVERLOAD)
2445 tmpstr = AMG_CALLunary(sv, numer_amg);
2446 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2447 return SvIV(tmpstr);
2450 return PTR2IV(SvRV(sv));
2453 if (SvVALID(sv) || isREGEXP(sv)) {
2454 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2455 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2456 In practice they are extremely unlikely to actually get anywhere
2457 accessible by user Perl code - the only way that I'm aware of is when
2458 a constant subroutine which is used as the second argument to index.
2460 Regexps have no SvIVX and SvNVX fields.
2462 assert(isREGEXP(sv) || SvPOKp(sv));
2465 const char * const ptr =
2466 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2468 = grok_number(ptr, SvCUR(sv), &value);
2470 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2471 == IS_NUMBER_IN_UV) {
2472 /* It's definitely an integer */
2473 if (numtype & IS_NUMBER_NEG) {
2474 if (value < (UV)IV_MIN)
2477 if (value < (UV)IV_MAX)
2482 /* Quite wrong but no good choices. */
2483 if ((numtype & IS_NUMBER_INFINITY)) {
2484 return (numtype & IS_NUMBER_NEG) ? IV_MIN : IV_MAX;
2485 } else if ((numtype & IS_NUMBER_NAN)) {
2486 return 0; /* So wrong. */
2490 if (ckWARN(WARN_NUMERIC))
2493 return I_V(Atof(ptr));
2497 if (SvTHINKFIRST(sv)) {
2498 if (SvREADONLY(sv) && !SvOK(sv)) {
2499 if (ckWARN(WARN_UNINITIALIZED))
2506 if (S_sv_2iuv_common(aTHX_ sv))
2510 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"IVdf")\n",
2511 PTR2UV(sv),SvIVX(sv)));
2512 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
2516 =for apidoc sv_2uv_flags
2518 Return the unsigned integer value of an SV, doing any necessary string
2519 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2520 Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
2526 Perl_sv_2uv_flags(pTHX_ SV *const sv, const I32 flags)
2528 PERL_ARGS_ASSERT_SV_2UV_FLAGS;
2530 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2536 if (flags & SV_SKIP_OVERLOAD)
2538 tmpstr = AMG_CALLunary(sv, numer_amg);
2539 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2540 return SvUV(tmpstr);
2543 return PTR2UV(SvRV(sv));
2546 if (SvVALID(sv) || isREGEXP(sv)) {
2547 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2548 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2549 Regexps have no SvIVX and SvNVX fields. */
2550 assert(isREGEXP(sv) || SvPOKp(sv));
2553 const char * const ptr =
2554 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2556 = grok_number(ptr, SvCUR(sv), &value);
2558 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2559 == IS_NUMBER_IN_UV) {
2560 /* It's definitely an integer */
2561 if (!(numtype & IS_NUMBER_NEG))
2565 /* Quite wrong but no good choices. */
2566 if ((numtype & IS_NUMBER_INFINITY)) {
2567 return UV_MAX; /* So wrong. */
2568 } else if ((numtype & IS_NUMBER_NAN)) {
2569 return 0; /* So wrong. */
2573 if (ckWARN(WARN_NUMERIC))
2576 return U_V(Atof(ptr));
2580 if (SvTHINKFIRST(sv)) {
2581 if (SvREADONLY(sv) && !SvOK(sv)) {
2582 if (ckWARN(WARN_UNINITIALIZED))
2589 if (S_sv_2iuv_common(aTHX_ sv))
2593 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2uv(%"UVuf")\n",
2594 PTR2UV(sv),SvUVX(sv)));
2595 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
2599 =for apidoc sv_2nv_flags
2601 Return the num value of an SV, doing any necessary string or integer
2602 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2603 Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
2609 Perl_sv_2nv_flags(pTHX_ SV *const sv, const I32 flags)
2611 PERL_ARGS_ASSERT_SV_2NV_FLAGS;
2613 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2614 && SvTYPE(sv) != SVt_PVFM);
2615 if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) {
2616 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2617 the same flag bit as SVf_IVisUV, so must not let them cache NVs.
2618 Regexps have no SvIVX and SvNVX fields. */
2620 if (flags & SV_GMAGIC)
2624 if (SvPOKp(sv) && !SvIOKp(sv)) {
2625 ptr = SvPVX_const(sv);
2627 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
2628 !grok_number(ptr, SvCUR(sv), NULL))
2634 return (NV)SvUVX(sv);
2636 return (NV)SvIVX(sv);
2642 ptr = RX_WRAPPED((REGEXP *)sv);
2645 assert(SvTYPE(sv) >= SVt_PVMG);
2646 /* This falls through to the report_uninit near the end of the
2648 } else if (SvTHINKFIRST(sv)) {
2653 if (flags & SV_SKIP_OVERLOAD)
2655 tmpstr = AMG_CALLunary(sv, numer_amg);
2656 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2657 return SvNV(tmpstr);
2660 return PTR2NV(SvRV(sv));
2662 if (SvREADONLY(sv) && !SvOK(sv)) {
2663 if (ckWARN(WARN_UNINITIALIZED))
2668 if (SvTYPE(sv) < SVt_NV) {
2669 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2670 sv_upgrade(sv, SVt_NV);
2672 STORE_NUMERIC_LOCAL_SET_STANDARD();
2673 PerlIO_printf(Perl_debug_log,
2674 "0x%"UVxf" num(%" NVgf ")\n",
2675 PTR2UV(sv), SvNVX(sv));
2676 RESTORE_NUMERIC_LOCAL();
2679 else if (SvTYPE(sv) < SVt_PVNV)
2680 sv_upgrade(sv, SVt_PVNV);
2685 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
2686 #ifdef NV_PRESERVES_UV
2692 /* Only set the public NV OK flag if this NV preserves the IV */
2693 /* Check it's not 0xFFFFFFFFFFFFFFFF */
2695 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
2696 : (SvIVX(sv) == I_V(SvNVX(sv))))
2702 else if (SvPOKp(sv)) {
2704 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2705 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
2707 #ifdef NV_PRESERVES_UV
2708 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2709 == IS_NUMBER_IN_UV) {
2710 /* It's definitely an integer */
2711 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
2713 S_sv_setnv(aTHX_ sv, numtype);
2720 SvNV_set(sv, Atof(SvPVX_const(sv)));
2721 /* Only set the public NV OK flag if this NV preserves the value in
2722 the PV at least as well as an IV/UV would.
2723 Not sure how to do this 100% reliably. */
2724 /* if that shift count is out of range then Configure's test is
2725 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2727 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2728 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2729 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
2730 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2731 /* Can't use strtol etc to convert this string, so don't try.
2732 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2735 /* value has been set. It may not be precise. */
2736 if ((numtype & IS_NUMBER_NEG) && (value >= (UV)IV_MIN)) {
2737 /* 2s complement assumption for (UV)IV_MIN */
2738 SvNOK_on(sv); /* Integer is too negative. */
2743 if (numtype & IS_NUMBER_NEG) {
2744 /* -IV_MIN is undefined, but we should never reach
2745 * this point with both IS_NUMBER_NEG and value ==
2747 assert(value != (UV)IV_MIN);
2748 SvIV_set(sv, -(IV)value);
2749 } else if (value <= (UV)IV_MAX) {
2750 SvIV_set(sv, (IV)value);
2752 SvUV_set(sv, value);
2756 if (numtype & IS_NUMBER_NOT_INT) {
2757 /* I believe that even if the original PV had decimals,
2758 they are lost beyond the limit of the FP precision.
2759 However, neither is canonical, so both only get p
2760 flags. NWC, 2000/11/25 */
2761 /* Both already have p flags, so do nothing */
2763 const NV nv = SvNVX(sv);
2764 /* XXX should this spot have NAN_COMPARE_BROKEN, too? */
2765 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2766 if (SvIVX(sv) == I_V(nv)) {
2769 /* It had no "." so it must be integer. */
2773 /* between IV_MAX and NV(UV_MAX).
2774 Could be slightly > UV_MAX */
2776 if (numtype & IS_NUMBER_NOT_INT) {
2777 /* UV and NV both imprecise. */
2779 const UV nv_as_uv = U_V(nv);
2781 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2790 /* It might be more code efficient to go through the entire logic above
2791 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2792 gets complex and potentially buggy, so more programmer efficient
2793 to do it this way, by turning off the public flags: */
2795 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2796 #endif /* NV_PRESERVES_UV */
2799 if (isGV_with_GP(sv)) {
2800 glob_2number(MUTABLE_GV(sv));
2804 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2806 assert (SvTYPE(sv) >= SVt_NV);
2807 /* Typically the caller expects that sv_any is not NULL now. */
2808 /* XXX Ilya implies that this is a bug in callers that assume this
2809 and ideally should be fixed. */
2813 STORE_NUMERIC_LOCAL_SET_STANDARD();
2814 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2nv(%" NVgf ")\n",
2815 PTR2UV(sv), SvNVX(sv));
2816 RESTORE_NUMERIC_LOCAL();
2824 Return an SV with the numeric value of the source SV, doing any necessary
2825 reference or overload conversion. The caller is expected to have handled
2832 Perl_sv_2num(pTHX_ SV *const sv)
2834 PERL_ARGS_ASSERT_SV_2NUM;
2839 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
2840 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
2841 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2842 return sv_2num(tmpsv);
2844 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2847 /* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2848 * UV as a string towards the end of buf, and return pointers to start and
2851 * We assume that buf is at least TYPE_CHARS(UV) long.
2855 S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
2857 char *ptr = buf + TYPE_CHARS(UV);
2858 char * const ebuf = ptr;
2861 PERL_ARGS_ASSERT_UIV_2BUF;
2869 uv = (iv == IV_MIN) ? (UV)iv : (UV)(-iv);
2873 *--ptr = '0' + (char)(uv % 10);
2881 /* Helper for sv_2pv_flags and sv_vcatpvfn_flags. If the NV is an
2882 * infinity or a not-a-number, writes the appropriate strings to the
2883 * buffer, including a zero byte. On success returns the written length,
2884 * excluding the zero byte, on failure (not an infinity, not a nan)
2885 * returns zero, assert-fails on maxlen being too short.
2887 * XXX for "Inf", "-Inf", and "NaN", we could have three read-only
2888 * shared string constants we point to, instead of generating a new
2889 * string for each instance. */
2891 S_infnan_2pv(NV nv, char* buffer, size_t maxlen, char plus) {
2893 assert(maxlen >= 4);
2894 if (Perl_isinf(nv)) {
2896 if (maxlen < 5) /* "-Inf\0" */
2906 else if (Perl_isnan(nv)) {
2910 /* XXX optionally output the payload mantissa bits as
2911 * "(unsigned)" (to match the nan("...") C99 function,
2912 * or maybe as "(0xhhh...)" would make more sense...
2913 * provide a format string so that the user can decide?
2914 * NOTE: would affect the maxlen and assert() logic.*/
2919 assert((s == buffer + 3) || (s == buffer + 4));
2921 return s - buffer - 1; /* -1: excluding the zero byte */
2925 =for apidoc sv_2pv_flags
2927 Returns a pointer to the string value of an SV, and sets C<*lp> to its length.
2928 If flags has the C<SV_GMAGIC> bit set, does an C<mg_get()> first. Coerces C<sv> to a
2929 string if necessary. Normally invoked via the C<SvPV_flags> macro.
2930 C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
2936 Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
2940 PERL_ARGS_ASSERT_SV_2PV_FLAGS;
2942 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2943 && SvTYPE(sv) != SVt_PVFM);
2944 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2949 if (flags & SV_SKIP_OVERLOAD)
2951 tmpstr = AMG_CALLunary(sv, string_amg);
2952 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
2953 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2955 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2959 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2960 if (flags & SV_CONST_RETURN) {
2961 pv = (char *) SvPVX_const(tmpstr);
2963 pv = (flags & SV_MUTABLE_RETURN)
2964 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
2967 *lp = SvCUR(tmpstr);
2969 pv = sv_2pv_flags(tmpstr, lp, flags);
2982 SV *const referent = SvRV(sv);
2986 retval = buffer = savepvn("NULLREF", len);
2987 } else if (SvTYPE(referent) == SVt_REGEXP &&
2988 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
2989 amagic_is_enabled(string_amg))) {
2990 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
2994 /* If the regex is UTF-8 we want the containing scalar to
2995 have an UTF-8 flag too */
3002 *lp = RX_WRAPLEN(re);
3004 return RX_WRAPPED(re);
3006 const char *const typestr = sv_reftype(referent, 0);
3007 const STRLEN typelen = strlen(typestr);
3008 UV addr = PTR2UV(referent);
3009 const char *stashname = NULL;
3010 STRLEN stashnamelen = 0; /* hush, gcc */
3011 const char *buffer_end;
3013 if (SvOBJECT(referent)) {
3014 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
3017 stashname = HEK_KEY(name);
3018 stashnamelen = HEK_LEN(name);
3020 if (HEK_UTF8(name)) {
3026 stashname = "__ANON__";
3029 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
3030 + 2 * sizeof(UV) + 2 /* )\0 */;
3032 len = typelen + 3 /* (0x */
3033 + 2 * sizeof(UV) + 2 /* )\0 */;
3036 Newx(buffer, len, char);
3037 buffer_end = retval = buffer + len;
3039 /* Working backwards */
3043 *--retval = PL_hexdigit[addr & 15];
3044 } while (addr >>= 4);
3050 memcpy(retval, typestr, typelen);
3054 retval -= stashnamelen;
3055 memcpy(retval, stashname, stashnamelen);
3057 /* retval may not necessarily have reached the start of the
3059 assert (retval >= buffer);
3061 len = buffer_end - retval - 1; /* -1 for that \0 */
3073 if (flags & SV_MUTABLE_RETURN)
3074 return SvPVX_mutable(sv);
3075 if (flags & SV_CONST_RETURN)
3076 return (char *)SvPVX_const(sv);
3081 /* I'm assuming that if both IV and NV are equally valid then
3082 converting the IV is going to be more efficient */
3083 const U32 isUIOK = SvIsUV(sv);
3084 char buf[TYPE_CHARS(UV)];
3088 if (SvTYPE(sv) < SVt_PVIV)
3089 sv_upgrade(sv, SVt_PVIV);
3090 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
3092 /* inlined from sv_setpvn */
3093 s = SvGROW_mutable(sv, len + 1);
3094 Move(ptr, s, len, char);
3099 else if (SvNOK(sv)) {
3100 if (SvTYPE(sv) < SVt_PVNV)
3101 sv_upgrade(sv, SVt_PVNV);
3102 if (SvNVX(sv) == 0.0
3103 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
3104 && !Perl_isnan(SvNVX(sv))
3107 s = SvGROW_mutable(sv, 2);
3112 STRLEN size = 5; /* "-Inf\0" */
3114 s = SvGROW_mutable(sv, size);
3115 len = S_infnan_2pv(SvNVX(sv), s, size, 0);
3121 /* some Xenix systems wipe out errno here */
3130 5 + /* exponent digits */
3134 s = SvGROW_mutable(sv, size);
3135 #ifndef USE_LOCALE_NUMERIC
3136 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3142 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
3143 STORE_LC_NUMERIC_SET_TO_NEEDED();
3147 PL_numeric_radix_sv &&
3148 SvUTF8(PL_numeric_radix_sv);
3149 if (local_radix && SvLEN(PL_numeric_radix_sv) > 1) {
3150 size += SvLEN(PL_numeric_radix_sv) - 1;
3151 s = SvGROW_mutable(sv, size);
3154 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3156 /* If the radix character is UTF-8, and actually is in the
3157 * output, turn on the UTF-8 flag for the scalar */
3159 instr(s, SvPVX_const(PL_numeric_radix_sv))) {
3163 RESTORE_LC_NUMERIC();
3166 /* We don't call SvPOK_on(), because it may come to
3167 * pass that the locale changes so that the
3168 * stringification we just did is no longer correct. We
3169 * will have to re-stringify every time it is needed */
3176 else if (isGV_with_GP(sv)) {
3177 GV *const gv = MUTABLE_GV(sv);
3178 SV *const buffer = sv_newmortal();
3180 gv_efullname3(buffer, gv, "*");
3182 assert(SvPOK(buffer));
3186 *lp = SvCUR(buffer);
3187 return SvPVX(buffer);
3189 else if (isREGEXP(sv)) {
3190 if (lp) *lp = RX_WRAPLEN((REGEXP *)sv);
3191 return RX_WRAPPED((REGEXP *)sv);
3196 if (flags & SV_UNDEF_RETURNS_NULL)
3198 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
3200 /* Typically the caller expects that sv_any is not NULL now. */
3201 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
3202 sv_upgrade(sv, SVt_PV);
3207 const STRLEN len = s - SvPVX_const(sv);
3212 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
3213 PTR2UV(sv),SvPVX_const(sv)));
3214 if (flags & SV_CONST_RETURN)
3215 return (char *)SvPVX_const(sv);
3216 if (flags & SV_MUTABLE_RETURN)
3217 return SvPVX_mutable(sv);
3222 =for apidoc sv_copypv
3224 Copies a stringified representation of the source SV into the
3225 destination SV. Automatically performs any necessary C<mg_get> and
3226 coercion of numeric values into strings. Guaranteed to preserve
3227 C<UTF8> flag even from overloaded objects. Similar in nature to
3228 C<sv_2pv[_flags]> but operates directly on an SV instead of just the
3229 string. Mostly uses C<sv_2pv_flags> to do its work, except when that
3230 would lose the UTF-8'ness of the PV.
3232 =for apidoc sv_copypv_nomg
3234 Like C<sv_copypv>, but doesn't invoke get magic first.
3236 =for apidoc sv_copypv_flags
3238 Implementation of C<sv_copypv> and C<sv_copypv_nomg>. Calls get magic iff flags
3239 has the C<SV_GMAGIC> bit set.
3245 Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
3250 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
3252 s = SvPV_flags_const(ssv,len,(flags & SV_GMAGIC));
3253 sv_setpvn(dsv,s,len);
3261 =for apidoc sv_2pvbyte
3263 Return a pointer to the byte-encoded representation of the SV, and set C<*lp>
3264 to its length. May cause the SV to be downgraded from UTF-8 as a
3267 Usually accessed via the C<SvPVbyte> macro.
3273 Perl_sv_2pvbyte(pTHX_ SV *sv, STRLEN *const lp)
3275 PERL_ARGS_ASSERT_SV_2PVBYTE;
3278 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3279 || isGV_with_GP(sv) || SvROK(sv)) {
3280 SV *sv2 = sv_newmortal();
3281 sv_copypv_nomg(sv2,sv);
3284 sv_utf8_downgrade(sv,0);
3285 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3289 =for apidoc sv_2pvutf8
3291 Return a pointer to the UTF-8-encoded representation of the SV, and set C<*lp>
3292 to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3294 Usually accessed via the C<SvPVutf8> macro.
3300 Perl_sv_2pvutf8(pTHX_ SV *sv, STRLEN *const lp)
3302 PERL_ARGS_ASSERT_SV_2PVUTF8;
3304 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3305 || isGV_with_GP(sv) || SvROK(sv))
3306 sv = sv_mortalcopy(sv);
3309 sv_utf8_upgrade_nomg(sv);
3310 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3315 =for apidoc sv_2bool
3317 This macro is only used by C<sv_true()> or its macro equivalent, and only if
3318 the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>.
3319 It calls C<sv_2bool_flags> with the C<SV_GMAGIC> flag.
3321 =for apidoc sv_2bool_flags
3323 This function is only used by C<sv_true()> and friends, and only if
3324 the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>. If the flags
3325 contain C<SV_GMAGIC>, then it does an C<mg_get()> first.
3332 Perl_sv_2bool_flags(pTHX_ SV *sv, I32 flags)
3334 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
3337 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
3343 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
3344 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv)))) {
3347 if(SvGMAGICAL(sv)) {
3349 goto restart; /* call sv_2bool */
3351 /* expanded SvTRUE_common(sv, (flags = 0, goto restart)) */
3352 else if(!SvOK(sv)) {
3355 else if(SvPOK(sv)) {
3356 svb = SvPVXtrue(sv);
3358 else if((SvFLAGS(sv) & (SVf_IOK|SVf_NOK))) {
3359 svb = (SvIOK(sv) && SvIVX(sv) != 0)
3360 || (SvNOK(sv) && SvNVX(sv) != 0.0);
3364 goto restart; /* call sv_2bool_nomg */
3369 return SvRV(sv) != 0;
3373 RX_WRAPLEN(sv) > 1 || (RX_WRAPLEN(sv) && *RX_WRAPPED(sv) != '0');
3374 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
3378 =for apidoc sv_utf8_upgrade
3380 Converts the PV of an SV to its UTF-8-encoded form.
3381 Forces the SV to string form if it is not already.
3382 Will C<mg_get> on C<sv> if appropriate.
3383 Always sets the C<SvUTF8> flag to avoid future validity checks even
3384 if the whole string is the same in UTF-8 as not.
3385 Returns the number of bytes in the converted string
3387 This is not a general purpose byte encoding to Unicode interface:
3388 use the Encode extension for that.
3390 =for apidoc sv_utf8_upgrade_nomg
3392 Like C<sv_utf8_upgrade>, but doesn't do magic on C<sv>.
3394 =for apidoc sv_utf8_upgrade_flags
3396 Converts the PV of an SV to its UTF-8-encoded form.
3397 Forces the SV to string form if it is not already.
3398 Always sets the SvUTF8 flag to avoid future validity checks even
3399 if all the bytes are invariant in UTF-8.
3400 If C<flags> has C<SV_GMAGIC> bit set,
3401 will C<mg_get> on C<sv> if appropriate, else not.
3403 If C<flags> has C<SV_FORCE_UTF8_UPGRADE> set, this function assumes that the PV
3404 will expand when converted to UTF-8, and skips the extra work of checking for
3405 that. Typically this flag is used by a routine that has already parsed the
3406 string and found such characters, and passes this information on so that the
3407 work doesn't have to be repeated.
3409 Returns the number of bytes in the converted string.
3411 This is not a general purpose byte encoding to Unicode interface:
3412 use the Encode extension for that.
3414 =for apidoc sv_utf8_upgrade_flags_grow
3416 Like C<sv_utf8_upgrade_flags>, but has an additional parameter C<extra>, which is
3417 the number of unused bytes the string of C<sv> is guaranteed to have free after
3418 it upon return. This allows the caller to reserve extra space that it intends
3419 to fill, to avoid extra grows.
3421 C<sv_utf8_upgrade>, C<sv_utf8_upgrade_nomg>, and C<sv_utf8_upgrade_flags>
3422 are implemented in terms of this function.
3424 Returns the number of bytes in the converted string (not including the spares).
3428 (One might think that the calling routine could pass in the position of the
3429 first variant character when it has set SV_FORCE_UTF8_UPGRADE, so it wouldn't
3430 have to be found again. But that is not the case, because typically when the
3431 caller is likely to use this flag, it won't be calling this routine unless it
3432 finds something that won't fit into a byte. Otherwise it tries to not upgrade
3433 and just use bytes. But some things that do fit into a byte are variants in
3434 utf8, and the caller may not have been keeping track of these.)
3436 If the routine itself changes the string, it adds a trailing C<NUL>. Such a
3437 C<NUL> isn't guaranteed due to having other routines do the work in some input
3438 cases, or if the input is already flagged as being in utf8.
3440 The speed of this could perhaps be improved for many cases if someone wanted to
3441 write a fast function that counts the number of variant characters in a string,
3442 especially if it could return the position of the first one.
3447 Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV *const sv, const I32 flags, STRLEN extra)
3449 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
3451 if (sv == &PL_sv_undef)
3453 if (!SvPOK_nog(sv)) {
3455 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3456 (void) sv_2pv_flags(sv,&len, flags);
3458 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3462 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
3467 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3472 S_sv_uncow(aTHX_ sv, 0);
3475 if (IN_ENCODING && !(flags & SV_UTF8_NO_ENCODING)) {
3476 sv_recode_to_utf8(sv, _get_encoding());
3477 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3481 if (SvCUR(sv) == 0) {
3482 if (extra) SvGROW(sv, extra);
3483 } else { /* Assume Latin-1/EBCDIC */
3484 /* This function could be much more efficient if we
3485 * had a FLAG in SVs to signal if there are any variant
3486 * chars in the PV. Given that there isn't such a flag
3487 * make the loop as fast as possible (although there are certainly ways
3488 * to speed this up, eg. through vectorization) */
3489 U8 * s = (U8 *) SvPVX_const(sv);
3490 U8 * e = (U8 *) SvEND(sv);
3492 STRLEN two_byte_count = 0;
3494 if (flags & SV_FORCE_UTF8_UPGRADE) goto must_be_utf8;
3496 /* See if really will need to convert to utf8. We mustn't rely on our
3497 * incoming SV being well formed and having a trailing '\0', as certain
3498 * code in pp_formline can send us partially built SVs. */
3502 if (NATIVE_BYTE_IS_INVARIANT(ch)) continue;
3504 t--; /* t already incremented; re-point to first variant */
3509 /* utf8 conversion not needed because all are invariants. Mark as
3510 * UTF-8 even if no variant - saves scanning loop */
3512 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3517 /* Here, the string should be converted to utf8, either because of an
3518 * input flag (two_byte_count = 0), or because a character that
3519 * requires 2 bytes was found (two_byte_count = 1). t points either to
3520 * the beginning of the string (if we didn't examine anything), or to
3521 * the first variant. In either case, everything from s to t - 1 will
3522 * occupy only 1 byte each on output.
3524 * There are two main ways to convert. One is to create a new string
3525 * and go through the input starting from the beginning, appending each
3526 * converted value onto the new string as we go along. It's probably
3527 * best to allocate enough space in the string for the worst possible
3528 * case rather than possibly running out of space and having to
3529 * reallocate and then copy what we've done so far. Since everything
3530 * from s to t - 1 is invariant, the destination can be initialized
3531 * with these using a fast memory copy
3533 * The other way is to figure out exactly how big the string should be
3534 * by parsing the entire input. Then you don't have to make it big
3535 * enough to handle the worst possible case, and more importantly, if
3536 * the string you already have is large enough, you don't have to
3537 * allocate a new string, you can copy the last character in the input
3538 * string to the final position(s) that will be occupied by the
3539 * converted string and go backwards, stopping at t, since everything
3540 * before that is invariant.
3542 * There are advantages and disadvantages to each method.
3544 * In the first method, we can allocate a new string, do the memory
3545 * copy from the s to t - 1, and then proceed through the rest of the
3546 * string byte-by-byte.
3548 * In the second method, we proceed through the rest of the input
3549 * string just calculating how big the converted string will be. Then
3550 * there are two cases:
3551 * 1) if the string has enough extra space to handle the converted
3552 * value. We go backwards through the string, converting until we
3553 * get to the position we are at now, and then stop. If this
3554 * position is far enough along in the string, this method is
3555 * faster than the other method. If the memory copy were the same
3556 * speed as the byte-by-byte loop, that position would be about
3557 * half-way, as at the half-way mark, parsing to the end and back
3558 * is one complete string's parse, the same amount as starting
3559 * over and going all the way through. Actually, it would be
3560 * somewhat less than half-way, as it's faster to just count bytes
3561 * than to also copy, and we don't have the overhead of allocating
3562 * a new string, changing the scalar to use it, and freeing the
3563 * existing one. But if the memory copy is fast, the break-even
3564 * point is somewhere after half way. The counting loop could be
3565 * sped up by vectorization, etc, to move the break-even point
3566 * further towards the beginning.
3567 * 2) if the string doesn't have enough space to handle the converted
3568 * value. A new string will have to be allocated, and one might
3569 * as well, given that, start from the beginning doing the first
3570 * method. We've spent extra time parsing the string and in
3571 * exchange all we've gotten is that we know precisely how big to
3572 * make the new one. Perl is more optimized for time than space,
3573 * so this case is a loser.
3574 * So what I've decided to do is not use the 2nd method unless it is
3575 * guaranteed that a new string won't have to be allocated, assuming
3576 * the worst case. I also decided not to put any more conditions on it
3577 * than this, for now. It seems likely that, since the worst case is
3578 * twice as big as the unknown portion of the string (plus 1), we won't
3579 * be guaranteed enough space, causing us to go to the first method,
3580 * unless the string is short, or the first variant character is near
3581 * the end of it. In either of these cases, it seems best to use the
3582 * 2nd method. The only circumstance I can think of where this would
3583 * be really slower is if the string had once had much more data in it
3584 * than it does now, but there is still a substantial amount in it */
3587 STRLEN invariant_head = t - s;
3588 STRLEN size = invariant_head + (e - t) * 2 + 1 + extra;
3589 if (SvLEN(sv) < size) {
3591 /* Here, have decided to allocate a new string */
3596 Newx(dst, size, U8);
3598 /* If no known invariants at the beginning of the input string,
3599 * set so starts from there. Otherwise, can use memory copy to
3600 * get up to where we are now, and then start from here */
3602 if (invariant_head == 0) {
3605 Copy(s, dst, invariant_head, char);
3606 d = dst + invariant_head;
3610 append_utf8_from_native_byte(*t, &d);
3614 SvPV_free(sv); /* No longer using pre-existing string */
3615 SvPV_set(sv, (char*)dst);
3616 SvCUR_set(sv, d - dst);
3617 SvLEN_set(sv, size);
3620 /* Here, have decided to get the exact size of the string.
3621 * Currently this happens only when we know that there is
3622 * guaranteed enough space to fit the converted string, so
3623 * don't have to worry about growing. If two_byte_count is 0,
3624 * then t points to the first byte of the string which hasn't
3625 * been examined yet. Otherwise two_byte_count is 1, and t
3626 * points to the first byte in the string that will expand to
3627 * two. Depending on this, start examining at t or 1 after t.
3630 U8 *d = t + two_byte_count;
3633 /* Count up the remaining bytes that expand to two */
3636 const U8 chr = *d++;
3637 if (! NATIVE_BYTE_IS_INVARIANT(chr)) two_byte_count++;
3640 /* The string will expand by just the number of bytes that
3641 * occupy two positions. But we are one afterwards because of
3642 * the increment just above. This is the place to put the
3643 * trailing NUL, and to set the length before we decrement */
3645 d += two_byte_count;
3646 SvCUR_set(sv, d - s);
3650 /* Having decremented d, it points to the position to put the
3651 * very last byte of the expanded string. Go backwards through
3652 * the string, copying and expanding as we go, stopping when we
3653 * get to the part that is invariant the rest of the way down */
3657 if (NATIVE_BYTE_IS_INVARIANT(*e)) {
3660 *d-- = UTF8_EIGHT_BIT_LO(*e);
3661 *d-- = UTF8_EIGHT_BIT_HI(*e);
3667 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3668 /* Update pos. We do it at the end rather than during
3669 * the upgrade, to avoid slowing down the common case
3670 * (upgrade without pos).
3671 * pos can be stored as either bytes or characters. Since
3672 * this was previously a byte string we can just turn off
3673 * the bytes flag. */
3674 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3676 mg->mg_flags &= ~MGf_BYTES;
3678 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3679 magic_setutf8(sv,mg); /* clear UTF8 cache */
3684 /* Mark as UTF-8 even if no variant - saves scanning loop */
3690 =for apidoc sv_utf8_downgrade
3692 Attempts to convert the PV of an SV from characters to bytes.
3693 If the PV contains a character that cannot fit
3694 in a byte, this conversion will fail;
3695 in this case, either returns false or, if C<fail_ok> is not
3698 This is not a general purpose Unicode to byte encoding interface:
3699 use the C<Encode> extension for that.
3705 Perl_sv_utf8_downgrade(pTHX_ SV *const sv, const bool fail_ok)
3707 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3709 if (SvPOKp(sv) && SvUTF8(sv)) {
3713 int mg_flags = SV_GMAGIC;
3716 S_sv_uncow(aTHX_ sv, 0);
3718 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3720 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3721 if (mg && mg->mg_len > 0 && mg->mg_flags & MGf_BYTES) {
3722 mg->mg_len = sv_pos_b2u_flags(sv, mg->mg_len,
3723 SV_GMAGIC|SV_CONST_RETURN);
3724 mg_flags = 0; /* sv_pos_b2u does get magic */
3726 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3727 magic_setutf8(sv,mg); /* clear UTF8 cache */
3730 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3732 if (!utf8_to_bytes(s, &len)) {
3737 Perl_croak(aTHX_ "Wide character in %s",
3740 Perl_croak(aTHX_ "Wide character");
3751 =for apidoc sv_utf8_encode
3753 Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3754 flag off so that it looks like octets again.
3760 Perl_sv_utf8_encode(pTHX_ SV *const sv)
3762 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3764 if (SvREADONLY(sv)) {
3765 sv_force_normal_flags(sv, 0);
3767 (void) sv_utf8_upgrade(sv);
3772 =for apidoc sv_utf8_decode
3774 If the PV of the SV is an octet sequence in UTF-8
3775 and contains a multiple-byte character, the C<SvUTF8> flag is turned on
3776 so that it looks like a character. If the PV contains only single-byte
3777 characters, the C<SvUTF8> flag stays off.
3778 Scans PV for validity and returns false if the PV is invalid UTF-8.
3784 Perl_sv_utf8_decode(pTHX_ SV *const sv)
3786 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3789 const U8 *start, *c;
3792 /* The octets may have got themselves encoded - get them back as
3795 if (!sv_utf8_downgrade(sv, TRUE))
3798 /* it is actually just a matter of turning the utf8 flag on, but
3799 * we want to make sure everything inside is valid utf8 first.
3801 c = start = (const U8 *) SvPVX_const(sv);
3802 if (!is_utf8_string(c, SvCUR(sv)))
3804 e = (const U8 *) SvEND(sv);
3807 if (!UTF8_IS_INVARIANT(ch)) {
3812 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3813 /* XXX Is this dead code? XS_utf8_decode calls SvSETMAGIC
3814 after this, clearing pos. Does anything on CPAN
3816 /* adjust pos to the start of a UTF8 char sequence */
3817 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3819 I32 pos = mg->mg_len;
3821 for (c = start + pos; c > start; c--) {
3822 if (UTF8_IS_START(*c))
3825 mg->mg_len = c - start;
3828 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3829 magic_setutf8(sv,mg); /* clear UTF8 cache */
3836 =for apidoc sv_setsv
3838 Copies the contents of the source SV C<ssv> into the destination SV
3839 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3840 function if the source SV needs to be reused. Does not handle 'set' magic on
3841 destination SV. Calls 'get' magic on source SV. Loosely speaking, it
3842 performs a copy-by-value, obliterating any previous content of the
3845 You probably want to use one of the assortment of wrappers, such as
3846 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3847 C<SvSetMagicSV_nosteal>.
3849 =for apidoc sv_setsv_flags
3851 Copies the contents of the source SV C<ssv> into the destination SV
3852 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3853 function if the source SV needs to be reused. Does not handle 'set' magic.
3854 Loosely speaking, it performs a copy-by-value, obliterating any previous
3855 content of the destination.
3856 If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
3857 C<ssv> if appropriate, else not. If the C<flags>
3858 parameter has the C<SV_NOSTEAL> bit set then the
3859 buffers of temps will not be stolen. C<sv_setsv>
3860 and C<sv_setsv_nomg> are implemented in terms of this function.
3862 You probably want to use one of the assortment of wrappers, such as
3863 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3864 C<SvSetMagicSV_nosteal>.
3866 This is the primary function for copying scalars, and most other
3867 copy-ish functions and macros use this underneath.
3873 S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
3875 I32 mro_changes = 0; /* 1 = method, 2 = isa, 3 = recursive isa */
3876 HV *old_stash = NULL;
3878 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB;
3880 if (dtype != SVt_PVGV && !isGV_with_GP(dstr)) {
3881 const char * const name = GvNAME(sstr);
3882 const STRLEN len = GvNAMELEN(sstr);
3884 if (dtype >= SVt_PV) {
3890 SvUPGRADE(dstr, SVt_PVGV);
3891 (void)SvOK_off(dstr);
3892 isGV_with_GP_on(dstr);
3894 GvSTASH(dstr) = GvSTASH(sstr);
3896 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
3897 gv_name_set(MUTABLE_GV(dstr), name, len,
3898 GV_ADD | (GvNAMEUTF8(sstr) ? SVf_UTF8 : 0 ));
3899 SvFAKE_on(dstr); /* can coerce to non-glob */
3902 if(GvGP(MUTABLE_GV(sstr))) {
3903 /* If source has method cache entry, clear it */
3905 SvREFCNT_dec(GvCV(sstr));
3906 GvCV_set(sstr, NULL);
3909 /* If source has a real method, then a method is
3912 GvCV((const GV *)sstr) && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3918 /* If dest already had a real method, that's a change as well */
3920 !mro_changes && GvGP(MUTABLE_GV(dstr)) && GvCVu((const GV *)dstr)
3921 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3926 /* We don't need to check the name of the destination if it was not a
3927 glob to begin with. */
3928 if(dtype == SVt_PVGV) {
3929 const char * const name = GvNAME((const GV *)dstr);
3932 /* The stash may have been detached from the symbol table, so
3934 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3938 const STRLEN len = GvNAMELEN(dstr);
3939 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
3940 || (len == 1 && name[0] == ':')) {
3943 /* Set aside the old stash, so we can reset isa caches on
3945 if((old_stash = GvHV(dstr)))
3946 /* Make sure we do not lose it early. */
3947 SvREFCNT_inc_simple_void_NN(
3948 sv_2mortal((SV *)old_stash)
3953 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
3956 /* freeing dstr's GP might free sstr (e.g. *x = $x),
3957 * so temporarily protect it */
3959 SAVEFREESV(SvREFCNT_inc_simple_NN(sstr));
3960 gp_free(MUTABLE_GV(dstr));
3961 GvINTRO_off(dstr); /* one-shot flag */
3962 GvGP_set(dstr, gp_ref(GvGP(sstr)));
3965 if (SvTAINTED(sstr))
3967 if (GvIMPORTED(dstr) != GVf_IMPORTED
3968 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
3970 GvIMPORTED_on(dstr);
3973 if(mro_changes == 2) {
3974 if (GvAV((const GV *)sstr)) {
3976 SV * const sref = (SV *)GvAV((const GV *)dstr);
3977 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
3978 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
3979 AV * const ary = newAV();
3980 av_push(ary, mg->mg_obj); /* takes the refcount */
3981 mg->mg_obj = (SV *)ary;
3983 av_push((AV *)mg->mg_obj, SvREFCNT_inc_simple_NN(dstr));
3985 else sv_magic(sref, dstr, PERL_MAGIC_isa, NULL, 0);
3987 mro_isa_changed_in(GvSTASH(dstr));
3989 else if(mro_changes == 3) {
3990 HV * const stash = GvHV(dstr);
3991 if(old_stash ? (HV *)HvENAME_get(old_stash) : stash)
3997 else if(mro_changes) mro_method_changed_in(GvSTASH(dstr));
3998 if (GvIO(dstr) && dtype == SVt_PVGV) {
3999 DEBUG_o(Perl_deb(aTHX_
4000 "glob_assign_glob clearing PL_stashcache\n"));
4001 /* It's a cache. It will rebuild itself quite happily.
4002 It's a lot of effort to work out exactly which key (or keys)
4003 might be invalidated by the creation of the this file handle.
4005 hv_clear(PL_stashcache);
4011 Perl_gv_setref(pTHX_ SV *const dstr, SV *const sstr)
4013 SV * const sref = SvRV(sstr);
4015 const int intro = GvINTRO(dstr);
4018 const U32 stype = SvTYPE(sref);
4020 PERL_ARGS_ASSERT_GV_SETREF;
4023 GvINTRO_off(dstr); /* one-shot flag */
4024 GvLINE(dstr) = CopLINE(PL_curcop);
4025 GvEGV(dstr) = MUTABLE_GV(dstr);
4030 location = (SV **) &(GvGP(dstr)->gp_cv); /* XXX bypassing GvCV_set */
4031 import_flag = GVf_IMPORTED_CV;
4034 location = (SV **) &GvHV(dstr);
4035 import_flag = GVf_IMPORTED_HV;
4038 location = (SV **) &GvAV(dstr);
4039 import_flag = GVf_IMPORTED_AV;
4042 location = (SV **) &GvIOp(dstr);
4045 location = (SV **) &GvFORM(dstr);
4048 location = &GvSV(dstr);
4049 import_flag = GVf_IMPORTED_SV;
4052 if (stype == SVt_PVCV) {
4053 /*if (GvCVGEN(dstr) && (GvCV(dstr) != (const CV *)sref || GvCVGEN(dstr))) {*/
4054 if (GvCVGEN(dstr)) {
4055 SvREFCNT_dec(GvCV(dstr));
4056 GvCV_set(dstr, NULL);
4057 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4060 /* SAVEt_GVSLOT takes more room on the savestack and has more
4061 overhead in leave_scope than SAVEt_GENERIC_SV. But for CVs
4062 leave_scope needs access to the GV so it can reset method
4063 caches. We must use SAVEt_GVSLOT whenever the type is
4064 SVt_PVCV, even if the stash is anonymous, as the stash may
4065 gain a name somehow before leave_scope. */
4066 if (stype == SVt_PVCV) {
4067 /* There is no save_pushptrptrptr. Creating it for this
4068 one call site would be overkill. So inline the ss add
4072 SS_ADD_PTR(location);
4073 SS_ADD_PTR(SvREFCNT_inc(*location));
4074 SS_ADD_UV(SAVEt_GVSLOT);
4077 else SAVEGENERICSV(*location);
4080 if (stype == SVt_PVCV && (*location != sref || GvCVGEN(dstr))) {
4081 CV* const cv = MUTABLE_CV(*location);
4083 if (!GvCVGEN((const GV *)dstr) &&
4084 (CvROOT(cv) || CvXSUB(cv)) &&
4085 /* redundant check that avoids creating the extra SV
4086 most of the time: */
4087 (CvCONST(cv) || ckWARN(WARN_REDEFINE)))
4089 SV * const new_const_sv =
4090 CvCONST((const CV *)sref)
4091 ? cv_const_sv((const CV *)sref)
4093 report_redefined_cv(
4094 sv_2mortal(Perl_newSVpvf(aTHX_
4097 HvNAME_HEK(GvSTASH((const GV *)dstr))
4099 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr)))
4102 CvCONST((const CV *)sref) ? &new_const_sv : NULL
4106 cv_ckproto_len_flags(cv, (const GV *)dstr,
4107 SvPOK(sref) ? CvPROTO(sref) : NULL,
4108 SvPOK(sref) ? CvPROTOLEN(sref) : 0,
4109 SvPOK(sref) ? SvUTF8(sref) : 0);
4111 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4112 GvASSUMECV_on(dstr);
4113 if(GvSTASH(dstr)) { /* sub foo { 1 } sub bar { 2 } *bar = \&foo */
4114 if (intro && GvREFCNT(dstr) > 1) {
4115 /* temporary remove extra savestack's ref */
4117 gv_method_changed(dstr);
4120 else gv_method_changed(dstr);
4123 *location = SvREFCNT_inc_simple_NN(sref);
4124 if (import_flag && !(GvFLAGS(dstr) & import_flag)
4125 && CopSTASH_ne(PL_curcop, GvSTASH(dstr))) {
4126 GvFLAGS(dstr) |= import_flag;
4129 if (stype == SVt_PVHV) {
4130 const char * const name = GvNAME((GV*)dstr);
4131 const STRLEN len = GvNAMELEN(dstr);
4134 (len > 1 && name[len-2] == ':' && name[len-1] == ':')
4135 || (len == 1 && name[0] == ':')
4137 && (!dref || HvENAME_get(dref))
4140 (HV *)sref, (HV *)dref,
4146 stype == SVt_PVAV && sref != dref
4147 && strEQ(GvNAME((GV*)dstr), "ISA")
4148 /* The stash may have been detached from the symbol table, so
4149 check its name before doing anything. */
4150 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
4153 MAGIC * const omg = dref && SvSMAGICAL(dref)
4154 ? mg_find(dref, PERL_MAGIC_isa)
4156 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
4157 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
4158 AV * const ary = newAV();
4159 av_push(ary, mg->mg_obj); /* takes the refcount */
4160 mg->mg_obj = (SV *)ary;
4163 if (SvTYPE(omg->mg_obj) == SVt_PVAV) {
4164 SV **svp = AvARRAY((AV *)omg->mg_obj);
4165 I32 items = AvFILLp((AV *)omg->mg_obj) + 1;
4169 SvREFCNT_inc_simple_NN(*svp++)
4175 SvREFCNT_inc_simple_NN(omg->mg_obj)
4179 av_push((AV *)mg->mg_obj,SvREFCNT_inc_simple_NN(dstr));
4184 sref, omg ? omg->mg_obj : dstr, PERL_MAGIC_isa, NULL, 0
4186 mg = mg_find(sref, PERL_MAGIC_isa);
4188 /* Since the *ISA assignment could have affected more than
4189 one stash, don't call mro_isa_changed_in directly, but let
4190 magic_clearisa do it for us, as it already has the logic for
4191 dealing with globs vs arrays of globs. */
4193 Perl_magic_clearisa(aTHX_ NULL, mg);
4195 else if (stype == SVt_PVIO) {
4196 DEBUG_o(Perl_deb(aTHX_ "gv_setref clearing PL_stashcache\n"));
4197 /* It's a cache. It will rebuild itself quite happily.
4198 It's a lot of effort to work out exactly which key (or keys)
4199 might be invalidated by the creation of the this file handle.
4201 hv_clear(PL_stashcache);
4205 if (!intro) SvREFCNT_dec(dref);
4206 if (SvTAINTED(sstr))
4214 #ifdef PERL_DEBUG_READONLY_COW
4215 # include <sys/mman.h>
4217 # ifndef PERL_MEMORY_DEBUG_HEADER_SIZE
4218 # define PERL_MEMORY_DEBUG_HEADER_SIZE 0
4222 Perl_sv_buf_to_ro(pTHX_ SV *sv)
4224 struct perl_memory_debug_header * const header =
4225 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4226 const MEM_SIZE len = header->size;
4227 PERL_ARGS_ASSERT_SV_BUF_TO_RO;
4228 # ifdef PERL_TRACK_MEMPOOL
4229 if (!header->readonly) header->readonly = 1;
4231 if (mprotect(header, len, PROT_READ))
4232 Perl_warn(aTHX_ "mprotect RW for COW string %p %lu failed with %d",
4233 header, len, errno);
4237 S_sv_buf_to_rw(pTHX_ SV *sv)
4239 struct perl_memory_debug_header * const header =
4240 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4241 const MEM_SIZE len = header->size;
4242 PERL_ARGS_ASSERT_SV_BUF_TO_RW;
4243 if (mprotect(header, len, PROT_READ|PROT_WRITE))
4244 Perl_warn(aTHX_ "mprotect for COW string %p %lu failed with %d",
4245 header, len, errno);
4246 # ifdef PERL_TRACK_MEMPOOL
4247 header->readonly = 0;
4252 # define sv_buf_to_ro(sv) NOOP
4253 # define sv_buf_to_rw(sv) NOOP
4257 Perl_sv_setsv_flags(pTHX_ SV *dstr, SV* sstr, const I32 flags)
4263 PERL_ARGS_ASSERT_SV_SETSV_FLAGS;
4265 if (UNLIKELY( sstr == dstr ))
4268 if (SvIS_FREED(dstr)) {
4269 Perl_croak(aTHX_ "panic: attempt to copy value %" SVf
4270 " to a freed scalar %p", SVfARG(sstr), (void *)dstr);
4272 SV_CHECK_THINKFIRST_COW_DROP(dstr);
4273 if (UNLIKELY( !sstr ))
4274 sstr = &PL_sv_undef;
4275 if (SvIS_FREED(sstr)) {
4276 Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p to %p",
4277 (void*)sstr, (void*)dstr);
4279 stype = SvTYPE(sstr);
4280 dtype = SvTYPE(dstr);
4282 /* There's a lot of redundancy below but we're going for speed here */
4287 if (LIKELY( dtype != SVt_PVGV && dtype != SVt_PVLV )) {
4288 (void)SvOK_off(dstr);
4296 /* For performance, we inline promoting to type SVt_IV. */
4297 /* We're starting from SVt_NULL, so provided that define is
4298 * actual 0, we don't have to unset any SV type flags
4299 * to promote to SVt_IV. */
4300 STATIC_ASSERT_STMT(SVt_NULL == 0);
4301 SET_SVANY_FOR_BODYLESS_IV(dstr);
4302 SvFLAGS(dstr) |= SVt_IV;
4306 sv_upgrade(dstr, SVt_PVIV);
4310 goto end_of_first_switch;
4312 (void)SvIOK_only(dstr);
4313 SvIV_set(dstr, SvIVX(sstr));
4316 /* SvTAINTED can only be true if the SV has taint magic, which in
4317 turn means that the SV type is PVMG (or greater). This is the
4318 case statement for SVt_IV, so this cannot be true (whatever gcov
4320 assert(!SvTAINTED(sstr));
4325 if (dtype < SVt_PV && dtype != SVt_IV)
4326 sv_upgrade(dstr, SVt_IV);
4330 if (LIKELY( SvNOK(sstr) )) {
4334 sv_upgrade(dstr, SVt_NV);
4338 sv_upgrade(dstr, SVt_PVNV);
4342 goto end_of_first_switch;
4344 SvNV_set(dstr, SvNVX(sstr));
4345 (void)SvNOK_only(dstr);
4346 /* SvTAINTED can only be true if the SV has taint magic, which in
4347 turn means that the SV type is PVMG (or greater). This is the
4348 case statement for SVt_NV, so this cannot be true (whatever gcov
4350 assert(!SvTAINTED(sstr));
4357 sv_upgrade(dstr, SVt_PV);
4360 if (dtype < SVt_PVIV)
4361 sv_upgrade(dstr, SVt_PVIV);
4364 if (dtype < SVt_PVNV)
4365 sv_upgrade(dstr, SVt_PVNV);
4369 const char * const type = sv_reftype(sstr,0);
4371 /* diag_listed_as: Bizarre copy of %s */
4372 Perl_croak(aTHX_ "Bizarre copy of %s in %s", type, OP_DESC(PL_op));
4374 Perl_croak(aTHX_ "Bizarre copy of %s", type);
4376 NOT_REACHED; /* NOTREACHED */
4380 if (dtype < SVt_REGEXP)
4382 if (dtype >= SVt_PV) {
4388 sv_upgrade(dstr, SVt_REGEXP);
4396 if (SvGMAGICAL(sstr) && (flags & SV_GMAGIC)) {
4398 if (SvTYPE(sstr) != stype)
4399 stype = SvTYPE(sstr);
4401 if (isGV_with_GP(sstr) && dtype <= SVt_PVLV) {
4402 glob_assign_glob(dstr, sstr, dtype);
4405 if (stype == SVt_PVLV)
4407 if (isREGEXP(sstr)) goto upgregexp;
4408 SvUPGRADE(dstr, SVt_PVNV);
4411 SvUPGRADE(dstr, (svtype)stype);
4413 end_of_first_switch:
4415 /* dstr may have been upgraded. */
4416 dtype = SvTYPE(dstr);
4417 sflags = SvFLAGS(sstr);
4419 if (UNLIKELY( dtype == SVt_PVCV )) {
4420 /* Assigning to a subroutine sets the prototype. */
4423 const char *const ptr = SvPV_const(sstr, len);
4425 SvGROW(dstr, len + 1);
4426 Copy(ptr, SvPVX(dstr), len + 1, char);
4427 SvCUR_set(dstr, len);
4429 SvFLAGS(dstr) |= sflags & SVf_UTF8;
4430 CvAUTOLOAD_off(dstr);
4435 else if (UNLIKELY(dtype == SVt_PVAV || dtype == SVt_PVHV
4436 || dtype == SVt_PVFM))
4438 const char * const type = sv_reftype(dstr,0);
4440 /* diag_listed_as: Cannot copy to %s */
4441 Perl_croak(aTHX_ "Cannot copy to %s in %s", type, OP_DESC(PL_op));
4443 Perl_croak(aTHX_ "Cannot copy to %s", type);
4444 } else if (sflags & SVf_ROK) {
4445 if (isGV_with_GP(dstr)
4446 && SvTYPE(SvRV(sstr)) == SVt_PVGV && isGV_with_GP(SvRV(sstr))) {
4449 if (GvIMPORTED(dstr) != GVf_IMPORTED
4450 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
4452 GvIMPORTED_on(dstr);
4457 glob_assign_glob(dstr, sstr, dtype);
4461 if (dtype >= SVt_PV) {
4462 if (isGV_with_GP(dstr)) {
4463 gv_setref(dstr, sstr);
4466 if (SvPVX_const(dstr)) {
4472 (void)SvOK_off(dstr);
4473 SvRV_set(dstr, SvREFCNT_inc(SvRV(sstr)));
4474 SvFLAGS(dstr) |= sflags & SVf_ROK;
4475 assert(!(sflags & SVp_NOK));
4476 assert(!(sflags & SVp_IOK));
4477 assert(!(sflags & SVf_NOK));
4478 assert(!(sflags & SVf_IOK));
4480 else if (isGV_with_GP(dstr)) {
4481 if (!(sflags & SVf_OK)) {
4482 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4483 "Undefined value assigned to typeglob");
4486 GV *gv = gv_fetchsv_nomg(sstr, GV_ADD, SVt_PVGV);
4487 if (dstr != (const SV *)gv) {
4488 const char * const name = GvNAME((const GV *)dstr);
4489 const STRLEN len = GvNAMELEN(dstr);
4490 HV *old_stash = NULL;
4491 bool reset_isa = FALSE;
4492 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
4493 || (len == 1 && name[0] == ':')) {
4494 /* Set aside the old stash, so we can reset isa caches
4495 on its subclasses. */
4496 if((old_stash = GvHV(dstr))) {
4497 /* Make sure we do not lose it early. */
4498 SvREFCNT_inc_simple_void_NN(
4499 sv_2mortal((SV *)old_stash)
4506 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
4507 gp_free(MUTABLE_GV(dstr));
4509 GvGP_set(dstr, gp_ref(GvGP(gv)));
4512 HV * const stash = GvHV(dstr);
4514 old_stash ? (HV *)HvENAME_get(old_stash) : stash
4524 else if ((dtype == SVt_REGEXP || dtype == SVt_PVLV)
4525 && (stype == SVt_REGEXP || isREGEXP(sstr))) {
4526 reg_temp_copy((REGEXP*)dstr, (REGEXP*)sstr);
4528 else if (sflags & SVp_POK) {
4529 const STRLEN cur = SvCUR(sstr);
4530 const STRLEN len = SvLEN(sstr);
4533 * We have three basic ways to copy the string:
4539 * Which we choose is based on various factors. The following
4540 * things are listed in order of speed, fastest to slowest:
4542 * - Copying a short string
4543 * - Copy-on-write bookkeeping
4545 * - Copying a long string
4547 * We swipe the string (steal the string buffer) if the SV on the
4548 * rhs is about to be freed anyway (TEMP and refcnt==1). This is a
4549 * big win on long strings. It should be a win on short strings if
4550 * SvPVX_const(dstr) has to be allocated. If not, it should not
4551 * slow things down, as SvPVX_const(sstr) would have been freed
4554 * We also steal the buffer from a PADTMP (operator target) if it
4555 * is ‘long enough’. For short strings, a swipe does not help
4556 * here, as it causes more malloc calls the next time the target
4557 * is used. Benchmarks show that even if SvPVX_const(dstr) has to
4558 * be allocated it is still not worth swiping PADTMPs for short
4559 * strings, as the savings here are small.
4561 * If swiping is not an option, then we see whether it is
4562 * worth using copy-on-write. If the lhs already has a buf-
4563 * fer big enough and the string is short, we skip it and fall back
4564 * to method 3, since memcpy is faster for short strings than the
4565 * later bookkeeping overhead that copy-on-write entails.
4567 * If the rhs is not a copy-on-write string yet, then we also
4568 * consider whether the buffer is too large relative to the string
4569 * it holds. Some operations such as readline allocate a large
4570 * buffer in the expectation of reusing it. But turning such into
4571 * a COW buffer is counter-productive because it increases memory
4572 * usage by making readline allocate a new large buffer the sec-
4573 * ond time round. So, if the buffer is too large, again, we use
4576 * Finally, if there is no buffer on the left, or the buffer is too
4577 * small, then we use copy-on-write and make both SVs share the
4582 /* Whichever path we take through the next code, we want this true,
4583 and doing it now facilitates the COW check. */
4584 (void)SvPOK_only(dstr);
4588 /* slated for free anyway (and not COW)? */
4589 (sflags & (SVs_TEMP|SVf_IsCOW)) == SVs_TEMP
4590 /* or a swipable TARG */
4592 (SVs_PADTMP|SVf_READONLY|SVf_PROTECT|SVf_IsCOW))
4594 /* whose buffer is worth stealing */
4595 && CHECK_COWBUF_THRESHOLD(cur,len)
4598 !(sflags & SVf_OOK) && /* and not involved in OOK hack? */
4599 (!(flags & SV_NOSTEAL)) &&
4600 /* and we're allowed to steal temps */
4601 SvREFCNT(sstr) == 1 && /* and no other references to it? */
4602 len) /* and really is a string */
4603 { /* Passes the swipe test. */
4604 if (SvPVX_const(dstr)) /* we know that dtype >= SVt_PV */
4606 SvPV_set(dstr, SvPVX_mutable(sstr));
4607 SvLEN_set(dstr, SvLEN(sstr));
4608 SvCUR_set(dstr, SvCUR(sstr));
4611 (void)SvOK_off(sstr); /* NOTE: nukes most SvFLAGS on sstr */
4612 SvPV_set(sstr, NULL);
4617 else if (flags & SV_COW_SHARED_HASH_KEYS
4619 #ifdef PERL_COPY_ON_WRITE
4622 ( (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4623 /* If this is a regular (non-hek) COW, only so
4624 many COW "copies" are possible. */
4625 && CowREFCNT(sstr) != SV_COW_REFCNT_MAX ))
4626 : ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4627 && !(SvFLAGS(dstr) & SVf_BREAK)
4628 && CHECK_COW_THRESHOLD(cur,len) && cur+1 < len
4629 && (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4633 && !(SvFLAGS(dstr) & SVf_BREAK)
4636 /* Either it's a shared hash key, or it's suitable for
4639 PerlIO_printf(Perl_debug_log, "Copy on write: sstr --> dstr\n");
4644 if (!(sflags & SVf_IsCOW)) {
4646 CowREFCNT(sstr) = 0;
4649 if (SvPVX_const(dstr)) { /* we know that dtype >= SVt_PV */
4655 if (sflags & SVf_IsCOW) {
4659 SvPV_set(dstr, SvPVX_mutable(sstr));
4664 /* SvIsCOW_shared_hash */
4665 DEBUG_C(PerlIO_printf(Perl_debug_log,
4666 "Copy on write: Sharing hash\n"));
4668 assert (SvTYPE(dstr) >= SVt_PV);
4670 HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)))));
4672 SvLEN_set(dstr, len);
4673 SvCUR_set(dstr, cur);
4676 /* Failed the swipe test, and we cannot do copy-on-write either.
4677 Have to copy the string. */
4678 SvGROW(dstr, cur + 1); /* inlined from sv_setpvn */
4679 Move(SvPVX_const(sstr),SvPVX(dstr),cur,char);
4680 SvCUR_set(dstr, cur);
4681 *SvEND(dstr) = '\0';
4683 if (sflags & SVp_NOK) {
4684 SvNV_set(dstr, SvNVX(sstr));
4686 if (sflags & SVp_IOK) {
4687 SvIV_set(dstr, SvIVX(sstr));
4688 /* Must do this otherwise some other overloaded use of 0x80000000
4689 gets confused. I guess SVpbm_VALID */
4690 if (sflags & SVf_IVisUV)
4693 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_NOK|SVp_NOK|SVf_UTF8);
4695 const MAGIC * const smg = SvVSTRING_mg(sstr);
4697 sv_magic(dstr, NULL, PERL_MAGIC_vstring,
4698 smg->mg_ptr, smg->mg_len);
4699 SvRMAGICAL_on(dstr);
4703 else if (sflags & (SVp_IOK|SVp_NOK)) {
4704 (void)SvOK_off(dstr);
4705 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_IVisUV|SVf_NOK|SVp_NOK);
4706 if (sflags & SVp_IOK) {
4707 /* XXXX Do we want to set IsUV for IV(ROK)? Be extra safe... */
4708 SvIV_set(dstr, SvIVX(sstr));
4710 if (sflags & SVp_NOK) {
4711 SvNV_set(dstr, SvNVX(sstr));
4715 if (isGV_with_GP(sstr)) {
4716 gv_efullname3(dstr, MUTABLE_GV(sstr), "*");
4719 (void)SvOK_off(dstr);
4721 if (SvTAINTED(sstr))
4726 =for apidoc sv_setsv_mg
4728 Like C<sv_setsv>, but also handles 'set' magic.
4734 Perl_sv_setsv_mg(pTHX_ SV *const dstr, SV *const sstr)
4736 PERL_ARGS_ASSERT_SV_SETSV_MG;
4738 sv_setsv(dstr,sstr);
4743 # define SVt_COW SVt_PV
4745 Perl_sv_setsv_cow(pTHX_ SV *dstr, SV *sstr)
4747 STRLEN cur = SvCUR(sstr);
4748 STRLEN len = SvLEN(sstr);
4750 #if defined(PERL_DEBUG_READONLY_COW) && defined(PERL_COPY_ON_WRITE)
4751 const bool already = cBOOL(SvIsCOW(sstr));
4754 PERL_ARGS_ASSERT_SV_SETSV_COW;
4757 PerlIO_printf(Perl_debug_log, "Fast copy on write: %p -> %p\n",
4758 (void*)sstr, (void*)dstr);
4765 if (SvTHINKFIRST(dstr))
4766 sv_force_normal_flags(dstr, SV_COW_DROP_PV);
4767 else if (SvPVX_const(dstr))
4768 Safefree(SvPVX_mutable(dstr));
4772 SvUPGRADE(dstr, SVt_COW);
4774 assert (SvPOK(sstr));
4775 assert (SvPOKp(sstr));
4777 if (SvIsCOW(sstr)) {
4779 if (SvLEN(sstr) == 0) {
4780 /* source is a COW shared hash key. */
4781 DEBUG_C(PerlIO_printf(Perl_debug_log,
4782 "Fast copy on write: Sharing hash\n"));
4783 new_pv = HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr))));
4786 assert(SvCUR(sstr)+1 < SvLEN(sstr));
4787 assert(CowREFCNT(sstr) < SV_COW_REFCNT_MAX);
4789 assert ((SvFLAGS(sstr) & CAN_COW_MASK) == CAN_COW_FLAGS);
4790 SvUPGRADE(sstr, SVt_COW);
4792 DEBUG_C(PerlIO_printf(Perl_debug_log,
4793 "Fast copy on write: Converting sstr to COW\n"));
4794 CowREFCNT(sstr) = 0;
4796 # ifdef PERL_DEBUG_READONLY_COW
4797 if (already) sv_buf_to_rw(sstr);
4800 new_pv = SvPVX_mutable(sstr);
4804 SvPV_set(dstr, new_pv);
4805 SvFLAGS(dstr) = (SVt_COW|SVf_POK|SVp_POK|SVf_IsCOW);
4808 SvLEN_set(dstr, len);
4809 SvCUR_set(dstr, cur);
4818 =for apidoc sv_setpvn
4820 Copies a string (possibly containing embedded C<NUL> characters) into an SV.
4821 The C<len> parameter indicates the number of
4822 bytes to be copied. If the C<ptr> argument is NULL the SV will become
4823 undefined. Does not handle 'set' magic. See C<L</sv_setpvn_mg>>.
4829 Perl_sv_setpvn(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4833 PERL_ARGS_ASSERT_SV_SETPVN;
4835 SV_CHECK_THINKFIRST_COW_DROP(sv);
4841 /* len is STRLEN which is unsigned, need to copy to signed */
4844 Perl_croak(aTHX_ "panic: sv_setpvn called with negative strlen %"
4847 SvUPGRADE(sv, SVt_PV);
4849 dptr = SvGROW(sv, len + 1);
4850 Move(ptr,dptr,len,char);
4853 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4855 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4859 =for apidoc sv_setpvn_mg
4861 Like C<sv_setpvn>, but also handles 'set' magic.
4867 Perl_sv_setpvn_mg(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4869 PERL_ARGS_ASSERT_SV_SETPVN_MG;
4871 sv_setpvn(sv,ptr,len);
4876 =for apidoc sv_setpv
4878 Copies a string into an SV. The string must be terminated with a C<NUL>
4880 Does not handle 'set' magic. See C<L</sv_setpv_mg>>.
4886 Perl_sv_setpv(pTHX_ SV *const sv, const char *const ptr)
4890 PERL_ARGS_ASSERT_SV_SETPV;
4892 SV_CHECK_THINKFIRST_COW_DROP(sv);
4898 SvUPGRADE(sv, SVt_PV);
4900 SvGROW(sv, len + 1);
4901 Move(ptr,SvPVX(sv),len+1,char);
4903 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4905 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4909 =for apidoc sv_setpv_mg
4911 Like C<sv_setpv>, but also handles 'set' magic.
4917 Perl_sv_setpv_mg(pTHX_ SV *const sv, const char *const ptr)
4919 PERL_ARGS_ASSERT_SV_SETPV_MG;
4926 Perl_sv_sethek(pTHX_ SV *const sv, const HEK *const hek)
4928 PERL_ARGS_ASSERT_SV_SETHEK;
4934 if (HEK_LEN(hek) == HEf_SVKEY) {
4935 sv_setsv(sv, *(SV**)HEK_KEY(hek));
4938 const int flags = HEK_FLAGS(hek);
4939 if (flags & HVhek_WASUTF8) {
4940 STRLEN utf8_len = HEK_LEN(hek);
4941 char *as_utf8 = (char *)bytes_to_utf8((U8*)HEK_KEY(hek), &utf8_len);
4942 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
4945 } else if (flags & HVhek_UNSHARED) {
4946 sv_setpvn(sv, HEK_KEY(hek), HEK_LEN(hek));
4949 else SvUTF8_off(sv);
4953 SV_CHECK_THINKFIRST_COW_DROP(sv);
4954 SvUPGRADE(sv, SVt_PV);
4956 SvPV_set(sv,(char *)HEK_KEY(share_hek_hek(hek)));
4957 SvCUR_set(sv, HEK_LEN(hek));
4963 else SvUTF8_off(sv);
4971 =for apidoc sv_usepvn_flags
4973 Tells an SV to use C<ptr> to find its string value. Normally the
4974 string is stored inside the SV, but sv_usepvn allows the SV to use an
4975 outside string. C<ptr> should point to memory that was allocated
4976 by L<C<Newx>|perlclib/Memory Management and String Handling>. It must be
4977 the start of a C<Newx>-ed block of memory, and not a pointer to the
4978 middle of it (beware of L<C<OOK>|perlguts/Offsets> and copy-on-write),
4979 and not be from a non-C<Newx> memory allocator like C<malloc>. The
4980 string length, C<len>, must be supplied. By default this function
4981 will C<Renew> (i.e. realloc, move) the memory pointed to by C<ptr>,
4982 so that pointer should not be freed or used by the programmer after
4983 giving it to C<sv_usepvn>, and neither should any pointers from "behind"
4984 that pointer (e.g. ptr + 1) be used.
4986 If S<C<flags & SV_SMAGIC>> is true, will call C<SvSETMAGIC>. If
4987 S<C<flags> & SV_HAS_TRAILING_NUL>> is true, then C<ptr[len]> must be C<NUL>,
4989 will be skipped (i.e. the buffer is actually at least 1 byte longer than
4990 C<len>, and already meets the requirements for storing in C<SvPVX>).
4996 Perl_sv_usepvn_flags(pTHX_ SV *const sv, char *ptr, const STRLEN len, const U32 flags)
5000 PERL_ARGS_ASSERT_SV_USEPVN_FLAGS;
5002 SV_CHECK_THINKFIRST_COW_DROP(sv);
5003 SvUPGRADE(sv, SVt_PV);
5006 if (flags & SV_SMAGIC)
5010 if (SvPVX_const(sv))
5014 if (flags & SV_HAS_TRAILING_NUL)
5015 assert(ptr[len] == '\0');
5018 allocate = (flags & SV_HAS_TRAILING_NUL)
5020 #ifdef Perl_safesysmalloc_size
5023 PERL_STRLEN_ROUNDUP(len + 1);
5025 if (flags & SV_HAS_TRAILING_NUL) {
5026 /* It's long enough - do nothing.
5027 Specifically Perl_newCONSTSUB is relying on this. */
5030 /* Force a move to shake out bugs in callers. */
5031 char *new_ptr = (char*)safemalloc(allocate);
5032 Copy(ptr, new_ptr, len, char);
5033 PoisonFree(ptr,len,char);
5037 ptr = (char*) saferealloc (ptr, allocate);
5040 #ifdef Perl_safesysmalloc_size
5041 SvLEN_set(sv, Perl_safesysmalloc_size(ptr));
5043 SvLEN_set(sv, allocate);
5047 if (!(flags & SV_HAS_TRAILING_NUL)) {
5050 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5052 if (flags & SV_SMAGIC)
5057 =for apidoc sv_force_normal_flags
5059 Undo various types of fakery on an SV, where fakery means
5060 "more than" a string: if the PV is a shared string, make
5061 a private copy; if we're a ref, stop refing; if we're a glob, downgrade to
5062 an C<xpvmg>; if we're a copy-on-write scalar, this is the on-write time when
5063 we do the copy, and is also used locally; if this is a
5064 vstring, drop the vstring magic. If C<SV_COW_DROP_PV> is set
5065 then a copy-on-write scalar drops its PV buffer (if any) and becomes
5066 C<SvPOK_off> rather than making a copy. (Used where this
5067 scalar is about to be set to some other value.) In addition,
5068 the C<flags> parameter gets passed to C<sv_unref_flags()>
5069 when unreffing. C<sv_force_normal> calls this function
5070 with flags set to 0.
5072 This function is expected to be used to signal to perl that this SV is
5073 about to be written to, and any extra book-keeping needs to be taken care
5074 of. Hence, it croaks on read-only values.
5080 S_sv_uncow(pTHX_ SV * const sv, const U32 flags)
5082 assert(SvIsCOW(sv));
5085 const char * const pvx = SvPVX_const(sv);
5086 const STRLEN len = SvLEN(sv);
5087 const STRLEN cur = SvCUR(sv);
5090 PerlIO_printf(Perl_debug_log,
5091 "Copy on write: Force normal %ld\n",
5096 # ifdef PERL_COPY_ON_WRITE
5098 /* Must do this first, since the CowREFCNT uses SvPVX and
5099 we need to write to CowREFCNT, or de-RO the whole buffer if we are
5100 the only owner left of the buffer. */
5101 sv_buf_to_rw(sv); /* NOOP if RO-ing not supported */
5103 U8 cowrefcnt = CowREFCNT(sv);
5104 if(cowrefcnt != 0) {
5106 CowREFCNT(sv) = cowrefcnt;
5111 /* Else we are the only owner of the buffer. */
5116 /* This SV doesn't own the buffer, so need to Newx() a new one: */
5121 if (flags & SV_COW_DROP_PV) {
5122 /* OK, so we don't need to copy our buffer. */
5125 SvGROW(sv, cur + 1);
5126 Move(pvx,SvPVX(sv),cur,char);
5132 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5139 const char * const pvx = SvPVX_const(sv);
5140 const STRLEN len = SvCUR(sv);
5144 if (flags & SV_COW_DROP_PV) {
5145 /* OK, so we don't need to copy our buffer. */
5148 SvGROW(sv, len + 1);
5149 Move(pvx,SvPVX(sv),len,char);
5152 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5158 Perl_sv_force_normal_flags(pTHX_ SV *const sv, const U32 flags)
5160 PERL_ARGS_ASSERT_SV_FORCE_NORMAL_FLAGS;
5163 Perl_croak_no_modify();
5164 else if (SvIsCOW(sv) && LIKELY(SvTYPE(sv) != SVt_PVHV))
5165 S_sv_uncow(aTHX_ sv, flags);
5167 sv_unref_flags(sv, flags);
5168 else if (SvFAKE(sv) && isGV_with_GP(sv))
5169 sv_unglob(sv, flags);
5170 else if (SvFAKE(sv) && isREGEXP(sv)) {
5171 /* Need to downgrade the REGEXP to a simple(r) scalar. This is analogous
5172 to sv_unglob. We only need it here, so inline it. */
5173 const bool islv = SvTYPE(sv) == SVt_PVLV;
5174 const svtype new_type =
5175 islv ? SVt_NULL : SvMAGIC(sv) || SvSTASH(sv) ? SVt_PVMG : SVt_PV;
5176 SV *const temp = newSV_type(new_type);
5177 regexp *const temp_p = ReANY((REGEXP *)sv);
5179 if (new_type == SVt_PVMG) {
5180 SvMAGIC_set(temp, SvMAGIC(sv));
5181 SvMAGIC_set(sv, NULL);
5182 SvSTASH_set(temp, SvSTASH(sv));
5183 SvSTASH_set(sv, NULL);
5185 if (!islv) SvCUR_set(temp, SvCUR(sv));
5186 /* Remember that SvPVX is in the head, not the body. But
5187 RX_WRAPPED is in the body. */
5188 assert(ReANY((REGEXP *)sv)->mother_re);
5189 /* Their buffer is already owned by someone else. */
5190 if (flags & SV_COW_DROP_PV) {
5191 /* SvLEN is already 0. For SVt_REGEXP, we have a brand new
5192 zeroed body. For SVt_PVLV, it should have been set to 0
5193 before turning into a regexp. */
5194 assert(!SvLEN(islv ? sv : temp));
5195 sv->sv_u.svu_pv = 0;
5198 sv->sv_u.svu_pv = savepvn(RX_WRAPPED((REGEXP *)sv), SvCUR(sv));
5199 SvLEN_set(islv ? sv : temp, SvCUR(sv)+1);
5203 /* Now swap the rest of the bodies. */
5207 SvFLAGS(sv) &= ~SVTYPEMASK;
5208 SvFLAGS(sv) |= new_type;
5209 SvANY(sv) = SvANY(temp);
5212 SvFLAGS(temp) &= ~(SVTYPEMASK);
5213 SvFLAGS(temp) |= SVt_REGEXP|SVf_FAKE;
5214 SvANY(temp) = temp_p;
5215 temp->sv_u.svu_rx = (regexp *)temp_p;
5217 SvREFCNT_dec_NN(temp);
5219 else if (SvVOK(sv)) sv_unmagic(sv, PERL_MAGIC_vstring);
5225 Efficient removal of characters from the beginning of the string buffer.
5226 C<SvPOK(sv)>, or at least C<SvPOKp(sv)>, must be true and C<ptr> must be a
5227 pointer to somewhere inside the string buffer. C<ptr> becomes the first
5228 character of the adjusted string. Uses the C<OOK> hack. On return, only
5229 C<SvPOK(sv)> and C<SvPOKp(sv)> among the C<OK> flags will be true.
5231 Beware: after this function returns, C<ptr> and SvPVX_const(sv) may no longer
5232 refer to the same chunk of data.
5234 The unfortunate similarity of this function's name to that of Perl's C<chop>
5235 operator is strictly coincidental. This function works from the left;
5236 C<chop> works from the right.
5242 Perl_sv_chop(pTHX_ SV *const sv, const char *const ptr)
5253 PERL_ARGS_ASSERT_SV_CHOP;
5255 if (!ptr || !SvPOKp(sv))
5257 delta = ptr - SvPVX_const(sv);
5259 /* Nothing to do. */
5262 max_delta = SvLEN(sv) ? SvLEN(sv) : SvCUR(sv);
5263 if (delta > max_delta)
5264 Perl_croak(aTHX_ "panic: sv_chop ptr=%p, start=%p, end=%p",
5265 ptr, SvPVX_const(sv), SvPVX_const(sv) + max_delta);
5266 /* SvPVX(sv) may move in SV_CHECK_THINKFIRST(sv), so don't use ptr any more */
5267 SV_CHECK_THINKFIRST(sv);
5268 SvPOK_only_UTF8(sv);
5271 if (!SvLEN(sv)) { /* make copy of shared string */
5272 const char *pvx = SvPVX_const(sv);
5273 const STRLEN len = SvCUR(sv);
5274 SvGROW(sv, len + 1);
5275 Move(pvx,SvPVX(sv),len,char);
5281 SvOOK_offset(sv, old_delta);
5283 SvLEN_set(sv, SvLEN(sv) - delta);
5284 SvCUR_set(sv, SvCUR(sv) - delta);
5285 SvPV_set(sv, SvPVX(sv) + delta);
5287 p = (U8 *)SvPVX_const(sv);
5290 /* how many bytes were evacuated? we will fill them with sentinel
5291 bytes, except for the part holding the new offset of course. */
5294 evacn += (old_delta < 0x100 ? 1 : 1 + sizeof(STRLEN));
5296 assert(evacn <= delta + old_delta);
5300 /* This sets 'delta' to the accumulated value of all deltas so far */
5304 /* If 'delta' fits in a byte, store it just prior to the new beginning of
5305 * the string; otherwise store a 0 byte there and store 'delta' just prior
5306 * to that, using as many bytes as a STRLEN occupies. Thus it overwrites a
5307 * portion of the chopped part of the string */
5308 if (delta < 0x100) {
5312 p -= sizeof(STRLEN);
5313 Copy((U8*)&delta, p, sizeof(STRLEN), U8);
5317 /* Fill the preceding buffer with sentinals to verify that no-one is
5327 =for apidoc sv_catpvn
5329 Concatenates the string onto the end of the string which is in the SV.
5330 C<len> indicates number of bytes to copy. If the SV has the UTF-8
5331 status set, then the bytes appended should be valid UTF-8.
5332 Handles 'get' magic, but not 'set' magic. See C<L</sv_catpvn_mg>>.
5334 =for apidoc sv_catpvn_flags
5336 Concatenates the string onto the end of the string which is in the SV. The
5337 C<len> indicates number of bytes to copy.
5339 By default, the string appended is assumed to be valid UTF-8 if the SV has
5340 the UTF-8 status set, and a string of bytes otherwise. One can force the
5341 appended string to be interpreted as UTF-8 by supplying the C<SV_CATUTF8>
5342 flag, and as bytes by supplying the C<SV_CATBYTES> flag; the SV or the
5343 string appended will be upgraded to UTF-8 if necessary.
5345 If C<flags> has the C<SV_SMAGIC> bit set, will
5346 C<mg_set> on C<dsv> afterwards if appropriate.
5347 C<sv_catpvn> and C<sv_catpvn_nomg> are implemented
5348 in terms of this function.
5354 Perl_sv_catpvn_flags(pTHX_ SV *const dsv, const char *sstr, const STRLEN slen, const I32 flags)
5357 const char * const dstr = SvPV_force_flags(dsv, dlen, flags);
5359 PERL_ARGS_ASSERT_SV_CATPVN_FLAGS;
5360 assert((flags & (SV_CATBYTES|SV_CATUTF8)) != (SV_CATBYTES|SV_CATUTF8));
5362 if (!(flags & SV_CATBYTES) || !SvUTF8(dsv)) {
5363 if (flags & SV_CATUTF8 && !SvUTF8(dsv)) {
5364 sv_utf8_upgrade_flags_grow(dsv, 0, slen + 1);
5367 else SvGROW(dsv, dlen + slen + 1);
5369 sstr = SvPVX_const(dsv);
5370 Move(sstr, SvPVX(dsv) + dlen, slen, char);
5371 SvCUR_set(dsv, SvCUR(dsv) + slen);
5374 /* We inline bytes_to_utf8, to avoid an extra malloc. */
5375 const char * const send = sstr + slen;
5378 /* Something this code does not account for, which I think is
5379 impossible; it would require the same pv to be treated as
5380 bytes *and* utf8, which would indicate a bug elsewhere. */
5381 assert(sstr != dstr);
5383 SvGROW(dsv, dlen + slen * 2 + 1);
5384 d = (U8 *)SvPVX(dsv) + dlen;
5386 while (sstr < send) {
5387 append_utf8_from_native_byte(*sstr, &d);
5390 SvCUR_set(dsv, d-(const U8 *)SvPVX(dsv));
5393 (void)SvPOK_only_UTF8(dsv); /* validate pointer */
5395 if (flags & SV_SMAGIC)
5400 =for apidoc sv_catsv
5402 Concatenates the string from SV C<ssv> onto the end of the string in SV
5403 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5404 Handles 'get' magic on both SVs, but no 'set' magic. See C<L</sv_catsv_mg>>
5405 and C<L</sv_catsv_nomg>>.
5407 =for apidoc sv_catsv_flags
5409 Concatenates the string from SV C<ssv> onto the end of the string in SV
5410 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5411 If C<flags> has the C<SV_GMAGIC> bit set, will call C<mg_get> on both SVs if
5412 appropriate. If C<flags> has the C<SV_SMAGIC> bit set, C<mg_set> will be called on
5413 the modified SV afterward, if appropriate. C<sv_catsv>, C<sv_catsv_nomg>,
5414 and C<sv_catsv_mg> are implemented in terms of this function.
5419 Perl_sv_catsv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
5421 PERL_ARGS_ASSERT_SV_CATSV_FLAGS;
5425 const char *spv = SvPV_flags_const(ssv, slen, flags);
5426 if (flags & SV_GMAGIC)
5428 sv_catpvn_flags(dsv, spv, slen,
5429 DO_UTF8(ssv) ? SV_CATUTF8 : SV_CATBYTES);
5430 if (flags & SV_SMAGIC)
5436 =for apidoc sv_catpv
5438 Concatenates the C<NUL>-terminated string onto the end of the string which is
5440 If the SV has the UTF-8 status set, then the bytes appended should be
5441 valid UTF-8. Handles 'get' magic, but not 'set' magic. See
5447 Perl_sv_catpv(pTHX_ SV *const sv, const char *ptr)
5453 PERL_ARGS_ASSERT_SV_CATPV;
5457 junk = SvPV_force(sv, tlen);
5459 SvGROW(sv, tlen + len + 1);
5461 ptr = SvPVX_const(sv);
5462 Move(ptr,SvPVX(sv)+tlen,len+1,char);
5463 SvCUR_set(sv, SvCUR(sv) + len);
5464 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5469 =for apidoc sv_catpv_flags
5471 Concatenates the C<NUL>-terminated string onto the end of the string which is
5473 If the SV has the UTF-8 status set, then the bytes appended should
5474 be valid UTF-8. If C<flags> has the C<SV_SMAGIC> bit set, will C<mg_set>
5475 on the modified SV if appropriate.
5481 Perl_sv_catpv_flags(pTHX_ SV *dstr, const char *sstr, const I32 flags)
5483 PERL_ARGS_ASSERT_SV_CATPV_FLAGS;
5484 sv_catpvn_flags(dstr, sstr, strlen(sstr), flags);
5488 =for apidoc sv_catpv_mg
5490 Like C<sv_catpv>, but also handles 'set' magic.
5496 Perl_sv_catpv_mg(pTHX_ SV *const sv, const char *const ptr)
5498 PERL_ARGS_ASSERT_SV_CATPV_MG;
5507 Creates a new SV. A non-zero C<len> parameter indicates the number of
5508 bytes of preallocated string space the SV should have. An extra byte for a
5509 trailing C<NUL> is also reserved. (C<SvPOK> is not set for the SV even if string
5510 space is allocated.) The reference count for the new SV is set to 1.
5512 In 5.9.3, C<newSV()> replaces the older C<NEWSV()> API, and drops the first
5513 parameter, I<x>, a debug aid which allowed callers to identify themselves.
5514 This aid has been superseded by a new build option, C<PERL_MEM_LOG> (see
5515 L<perlhacktips/PERL_MEM_LOG>). The older API is still there for use in XS
5516 modules supporting older perls.
5522 Perl_newSV(pTHX_ const STRLEN len)
5528 sv_grow(sv, len + 1);
5533 =for apidoc sv_magicext
5535 Adds magic to an SV, upgrading it if necessary. Applies the
5536 supplied C<vtable> and returns a pointer to the magic added.
5538 Note that C<sv_magicext> will allow things that C<sv_magic> will not.
5539 In particular, you can add magic to C<SvREADONLY> SVs, and add more than
5540 one instance of the same C<how>.
5542 If C<namlen> is greater than zero then a C<savepvn> I<copy> of C<name> is
5543 stored, if C<namlen> is zero then C<name> is stored as-is and - as another
5544 special case - if C<(name && namlen == HEf_SVKEY)> then C<name> is assumed
5545 to contain an SV* and is stored as-is with its C<REFCNT> incremented.
5547 (This is now used as a subroutine by C<sv_magic>.)
5552 Perl_sv_magicext(pTHX_ SV *const sv, SV *const obj, const int how,
5553 const MGVTBL *const vtable, const char *const name, const I32 namlen)
5557 PERL_ARGS_ASSERT_SV_MAGICEXT;
5559 SvUPGRADE(sv, SVt_PVMG);
5560 Newxz(mg, 1, MAGIC);
5561 mg->mg_moremagic = SvMAGIC(sv);
5562 SvMAGIC_set(sv, mg);
5564 /* Sometimes a magic contains a reference loop, where the sv and
5565 object refer to each other. To prevent a reference loop that
5566 would prevent such objects being freed, we look for such loops
5567 and if we find one we avoid incrementing the object refcount.
5569 Note we cannot do this to avoid self-tie loops as intervening RV must
5570 have its REFCNT incremented to keep it in existence.
5573 if (!obj || obj == sv ||
5574 how == PERL_MAGIC_arylen ||
5575 how == PERL_MAGIC_symtab ||
5576 (SvTYPE(obj) == SVt_PVGV &&
5577 (GvSV(obj) == sv || GvHV(obj) == (const HV *)sv
5578 || GvAV(obj) == (const AV *)sv || GvCV(obj) == (const CV *)sv
5579 || GvIOp(obj) == (const IO *)sv || GvFORM(obj) == (const CV *)sv)))
5584 mg->mg_obj = SvREFCNT_inc_simple(obj);
5585 mg->mg_flags |= MGf_REFCOUNTED;
5588 /* Normal self-ties simply pass a null object, and instead of
5589 using mg_obj directly, use the SvTIED_obj macro to produce a
5590 new RV as needed. For glob "self-ties", we are tieing the PVIO
5591 with an RV obj pointing to the glob containing the PVIO. In
5592 this case, to avoid a reference loop, we need to weaken the
5596 if (how == PERL_MAGIC_tiedscalar && SvTYPE(sv) == SVt_PVIO &&
5597 obj && SvROK(obj) && GvIO(SvRV(obj)) == (const IO *)sv)
5603 mg->mg_len = namlen;
5606 mg->mg_ptr = savepvn(name, namlen);
5607 else if (namlen == HEf_SVKEY) {
5608 /* Yes, this is casting away const. This is only for the case of
5609 HEf_SVKEY. I think we need to document this aberation of the
5610 constness of the API, rather than making name non-const, as
5611 that change propagating outwards a long way. */
5612 mg->mg_ptr = (char*)SvREFCNT_inc_simple_NN((SV *)name);
5614 mg->mg_ptr = (char *) name;
5616 mg->mg_virtual = (MGVTBL *) vtable;
5623 Perl_sv_magicext_mglob(pTHX_ SV *sv)
5625 PERL_ARGS_ASSERT_SV_MAGICEXT_MGLOB;
5626 if (SvTYPE(sv) == SVt_PVLV && LvTYPE(sv) == 'y') {
5627 /* This sv is only a delegate. //g magic must be attached to
5632 return sv_magicext(sv, NULL, PERL_MAGIC_regex_global,
5633 &PL_vtbl_mglob, 0, 0);
5637 =for apidoc sv_magic
5639 Adds magic to an SV. First upgrades C<sv> to type C<SVt_PVMG> if
5640 necessary, then adds a new magic item of type C<how> to the head of the
5643 See C<L</sv_magicext>> (which C<sv_magic> now calls) for a description of the
5644 handling of the C<name> and C<namlen> arguments.
5646 You need to use C<sv_magicext> to add magic to C<SvREADONLY> SVs and also
5647 to add more than one instance of the same C<how>.
5653 Perl_sv_magic(pTHX_ SV *const sv, SV *const obj, const int how,
5654 const char *const name, const I32 namlen)
5656 const MGVTBL *vtable;
5659 unsigned int vtable_index;
5661 PERL_ARGS_ASSERT_SV_MAGIC;
5663 if (how < 0 || (unsigned)how >= C_ARRAY_LENGTH(PL_magic_data)
5664 || ((flags = PL_magic_data[how]),
5665 (vtable_index = flags & PERL_MAGIC_VTABLE_MASK)
5666 > magic_vtable_max))
5667 Perl_croak(aTHX_ "Don't know how to handle magic of type \\%o", how);
5669 /* PERL_MAGIC_ext is reserved for use by extensions not perl internals.
5670 Useful for attaching extension internal data to perl vars.
5671 Note that multiple extensions may clash if magical scalars
5672 etc holding private data from one are passed to another. */
5674 vtable = (vtable_index == magic_vtable_max)
5675 ? NULL : PL_magic_vtables + vtable_index;
5677 if (SvREADONLY(sv)) {
5679 !PERL_MAGIC_TYPE_READONLY_ACCEPTABLE(how)
5682 Perl_croak_no_modify();
5685 if (SvMAGICAL(sv) || (how == PERL_MAGIC_taint && SvTYPE(sv) >= SVt_PVMG)) {
5686 if (SvMAGIC(sv) && (mg = mg_find(sv, how))) {
5687 /* sv_magic() refuses to add a magic of the same 'how' as an
5690 if (how == PERL_MAGIC_taint)
5696 /* Force pos to be stored as characters, not bytes. */
5697 if (SvMAGICAL(sv) && DO_UTF8(sv)
5698 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
5700 && mg->mg_flags & MGf_BYTES) {
5701 mg->mg_len = (SSize_t)sv_pos_b2u_flags(sv, (STRLEN)mg->mg_len,
5703 mg->mg_flags &= ~MGf_BYTES;
5706 /* Rest of work is done else where */
5707 mg = sv_magicext(sv,obj,how,vtable,name,namlen);
5710 case PERL_MAGIC_taint:
5713 case PERL_MAGIC_ext:
5714 case PERL_MAGIC_dbfile:
5721 S_sv_unmagicext_flags(pTHX_ SV *const sv, const int type, MGVTBL *vtbl, const U32 flags)
5728 if (SvTYPE(sv) < SVt_PVMG || !SvMAGIC(sv))
5730 mgp = &(((XPVMG*) SvANY(sv))->xmg_u.xmg_magic);
5731 for (mg = *mgp; mg; mg = *mgp) {
5732 const MGVTBL* const virt = mg->mg_virtual;
5733 if (mg->mg_type == type && (!flags || virt == vtbl)) {
5734 *mgp = mg->mg_moremagic;
5735 if (virt && virt->svt_free)
5736 virt->svt_free(aTHX_ sv, mg);
5737 if (mg->mg_ptr && mg->mg_type != PERL_MAGIC_regex_global) {
5739 Safefree(mg->mg_ptr);
5740 else if (mg->mg_len == HEf_SVKEY)
5741 SvREFCNT_dec(MUTABLE_SV(mg->mg_ptr));
5742 else if (mg->mg_type == PERL_MAGIC_utf8)
5743 Safefree(mg->mg_ptr);
5745 if (mg->mg_flags & MGf_REFCOUNTED)
5746 SvREFCNT_dec(mg->mg_obj);
5750 mgp = &mg->mg_moremagic;
5753 if (SvMAGICAL(sv)) /* if we're under save_magic, wait for restore_magic; */
5754 mg_magical(sv); /* else fix the flags now */
5758 SvFLAGS(sv) |= (SvFLAGS(sv) & (SVp_IOK|SVp_NOK|SVp_POK)) >> PRIVSHIFT;
5764 =for apidoc sv_unmagic
5766 Removes all magic of type C<type> from an SV.
5772 Perl_sv_unmagic(pTHX_ SV *const sv, const int type)
5774 PERL_ARGS_ASSERT_SV_UNMAGIC;
5775 return S_sv_unmagicext_flags(aTHX_ sv, type, NULL, 0);
5779 =for apidoc sv_unmagicext
5781 Removes all magic of type C<type> with the specified C<vtbl> from an SV.
5787 Perl_sv_unmagicext(pTHX_ SV *const sv, const int type, MGVTBL *vtbl)
5789 PERL_ARGS_ASSERT_SV_UNMAGICEXT;
5790 return S_sv_unmagicext_flags(aTHX_ sv, type, vtbl, 1);
5794 =for apidoc sv_rvweaken
5796 Weaken a reference: set the C<SvWEAKREF> flag on this RV; give the
5797 referred-to SV C<PERL_MAGIC_backref> magic if it hasn't already; and
5798 push a back-reference to this RV onto the array of backreferences
5799 associated with that magic. If the RV is magical, set magic will be
5800 called after the RV is cleared.
5806 Perl_sv_rvweaken(pTHX_ SV *const sv)
5810 PERL_ARGS_ASSERT_SV_RVWEAKEN;
5812 if (!SvOK(sv)) /* let undefs pass */
5815 Perl_croak(aTHX_ "Can't weaken a nonreference");
5816 else if (SvWEAKREF(sv)) {
5817 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is already weak");
5820 else if (SvREADONLY(sv)) croak_no_modify();
5822 Perl_sv_add_backref(aTHX_ tsv, sv);
5824 SvREFCNT_dec_NN(tsv);
5829 =for apidoc sv_get_backrefs
5831 If C<sv> is the target of a weak reference then it returns the back
5832 references structure associated with the sv; otherwise return C<NULL>.
5834 When returning a non-null result the type of the return is relevant. If it
5835 is an AV then the elements of the AV are the weak reference RVs which
5836 point at this item. If it is any other type then the item itself is the
5839 See also C<Perl_sv_add_backref()>, C<Perl_sv_del_backref()>,
5840 C<Perl_sv_kill_backrefs()>
5846 Perl_sv_get_backrefs(SV *const sv)
5850 PERL_ARGS_ASSERT_SV_GET_BACKREFS;
5852 /* find slot to store array or singleton backref */
5854 if (SvTYPE(sv) == SVt_PVHV) {
5856 struct xpvhv_aux * const iter = HvAUX((HV *)sv);
5857 backrefs = (SV *)iter->xhv_backreferences;
5859 } else if (SvMAGICAL(sv)) {
5860 MAGIC *mg = mg_find(sv, PERL_MAGIC_backref);
5862 backrefs = mg->mg_obj;
5867 /* Give tsv backref magic if it hasn't already got it, then push a
5868 * back-reference to sv onto the array associated with the backref magic.
5870 * As an optimisation, if there's only one backref and it's not an AV,
5871 * store it directly in the HvAUX or mg_obj slot, avoiding the need to
5872 * allocate an AV. (Whether the slot holds an AV tells us whether this is
5876 /* A discussion about the backreferences array and its refcount:
5878 * The AV holding the backreferences is pointed to either as the mg_obj of
5879 * PERL_MAGIC_backref, or in the specific case of a HV, from the
5880 * xhv_backreferences field. The array is created with a refcount
5881 * of 2. This means that if during global destruction the array gets
5882 * picked on before its parent to have its refcount decremented by the
5883 * random zapper, it won't actually be freed, meaning it's still there for
5884 * when its parent gets freed.
5886 * When the parent SV is freed, the extra ref is killed by
5887 * Perl_sv_kill_backrefs. The other ref is killed, in the case of magic,
5888 * by mg_free() / MGf_REFCOUNTED, or for a hash, by Perl_hv_kill_backrefs.
5890 * When a single backref SV is stored directly, it is not reference
5895 Perl_sv_add_backref(pTHX_ SV *const tsv, SV *const sv)
5901 PERL_ARGS_ASSERT_SV_ADD_BACKREF;
5903 /* find slot to store array or singleton backref */
5905 if (SvTYPE(tsv) == SVt_PVHV) {
5906 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
5909 mg = mg_find(tsv, PERL_MAGIC_backref);
5911 mg = sv_magicext(tsv, NULL, PERL_MAGIC_backref, &PL_vtbl_backref, NULL, 0);
5912 svp = &(mg->mg_obj);
5915 /* create or retrieve the array */
5917 if ( (!*svp && SvTYPE(sv) == SVt_PVAV)
5918 || (*svp && SvTYPE(*svp) != SVt_PVAV)
5922 mg->mg_flags |= MGf_REFCOUNTED;
5925 SvREFCNT_inc_simple_void_NN(av);
5926 /* av now has a refcnt of 2; see discussion above */
5927 av_extend(av, *svp ? 2 : 1);
5929 /* move single existing backref to the array */
5930 AvARRAY(av)[++AvFILLp(av)] = *svp; /* av_push() */
5935 av = MUTABLE_AV(*svp);
5937 /* optimisation: store single backref directly in HvAUX or mg_obj */
5941 assert(SvTYPE(av) == SVt_PVAV);
5942 if (AvFILLp(av) >= AvMAX(av)) {
5943 av_extend(av, AvFILLp(av)+1);
5946 /* push new backref */
5947 AvARRAY(av)[++AvFILLp(av)] = sv; /* av_push() */
5950 /* delete a back-reference to ourselves from the backref magic associated
5951 * with the SV we point to.
5955 Perl_sv_del_backref(pTHX_ SV *const tsv, SV *const sv)
5959 PERL_ARGS_ASSERT_SV_DEL_BACKREF;
5961 if (SvTYPE(tsv) == SVt_PVHV) {
5963 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
5965 else if (SvIS_FREED(tsv) && PL_phase == PERL_PHASE_DESTRUCT) {
5966 /* It's possible for the the last (strong) reference to tsv to have
5967 become freed *before* the last thing holding a weak reference.
5968 If both survive longer than the backreferences array, then when
5969 the referent's reference count drops to 0 and it is freed, it's
5970 not able to chase the backreferences, so they aren't NULLed.
5972 For example, a CV holds a weak reference to its stash. If both the
5973 CV and the stash survive longer than the backreferences array,
5974 and the CV gets picked for the SvBREAK() treatment first,
5975 *and* it turns out that the stash is only being kept alive because
5976 of an our variable in the pad of the CV, then midway during CV
5977 destruction the stash gets freed, but CvSTASH() isn't set to NULL.
5978 It ends up pointing to the freed HV. Hence it's chased in here, and
5979 if this block wasn't here, it would hit the !svp panic just below.
5981 I don't believe that "better" destruction ordering is going to help
5982 here - during global destruction there's always going to be the
5983 chance that something goes out of order. We've tried to make it
5984 foolproof before, and it only resulted in evolutionary pressure on
5985 fools. Which made us look foolish for our hubris. :-(
5991 = SvMAGICAL(tsv) ? mg_find(tsv, PERL_MAGIC_backref) : NULL;
5992 svp = mg ? &(mg->mg_obj) : NULL;
5996 Perl_croak(aTHX_ "panic: del_backref, svp=0");
5998 /* It's possible that sv is being freed recursively part way through the
5999 freeing of tsv. If this happens, the backreferences array of tsv has
6000 already been freed, and so svp will be NULL. If this is the case,
6001 we should not panic. Instead, nothing needs doing, so return. */
6002 if (PL_phase == PERL_PHASE_DESTRUCT && SvREFCNT(tsv) == 0)
6004 Perl_croak(aTHX_ "panic: del_backref, *svp=%p phase=%s refcnt=%" UVuf,
6005 (void*)*svp, PL_phase_names[PL_phase], (UV)SvREFCNT(tsv));
6008 if (SvTYPE(*svp) == SVt_PVAV) {
6012 AV * const av = (AV*)*svp;
6014 assert(!SvIS_FREED(av));
6018 /* for an SV with N weak references to it, if all those
6019 * weak refs are deleted, then sv_del_backref will be called
6020 * N times and O(N^2) compares will be done within the backref
6021 * array. To ameliorate this potential slowness, we:
6022 * 1) make sure this code is as tight as possible;
6023 * 2) when looking for SV, look for it at both the head and tail of the
6024 * array first before searching the rest, since some create/destroy
6025 * patterns will cause the backrefs to be freed in order.
6032 SV **p = &svp[fill];
6033 SV *const topsv = *p;
6040 /* We weren't the last entry.
6041 An unordered list has this property that you
6042 can take the last element off the end to fill
6043 the hole, and it's still an unordered list :-)
6049 break; /* should only be one */
6056 AvFILLp(av) = fill-1;
6058 else if (SvIS_FREED(*svp) && PL_phase == PERL_PHASE_DESTRUCT) {
6059 /* freed AV; skip */
6062 /* optimisation: only a single backref, stored directly */
6064 Perl_croak(aTHX_ "panic: del_backref, *svp=%p, sv=%p",
6065 (void*)*svp, (void*)sv);
6072 Perl_sv_kill_backrefs(pTHX_ SV *const sv, AV *const av)
6078 PERL_ARGS_ASSERT_SV_KILL_BACKREFS;
6083 /* after multiple passes through Perl_sv_clean_all() for a thingy
6084 * that has badly leaked, the backref array may have gotten freed,
6085 * since we only protect it against 1 round of cleanup */
6086 if (SvIS_FREED(av)) {
6087 if (PL_in_clean_all) /* All is fair */
6090 "panic: magic_killbackrefs (freed backref AV/SV)");
6094 is_array = (SvTYPE(av) == SVt_PVAV);
6096 assert(!SvIS_FREED(av));
6099 last = svp + AvFILLp(av);
6102 /* optimisation: only a single backref, stored directly */
6108 while (svp <= last) {
6110 SV *const referrer = *svp;
6111 if (SvWEAKREF(referrer)) {
6112 /* XXX Should we check that it hasn't changed? */
6113 assert(SvROK(referrer));
6114 SvRV_set(referrer, 0);
6116 SvWEAKREF_off(referrer);
6117 SvSETMAGIC(referrer);
6118 } else if (SvTYPE(referrer) == SVt_PVGV ||
6119 SvTYPE(referrer) == SVt_PVLV) {
6120 assert(SvTYPE(sv) == SVt_PVHV); /* stash backref */
6121 /* You lookin' at me? */
6122 assert(GvSTASH(referrer));
6123 assert(GvSTASH(referrer) == (const HV *)sv);
6124 GvSTASH(referrer) = 0;
6125 } else if (SvTYPE(referrer) == SVt_PVCV ||
6126 SvTYPE(referrer) == SVt_PVFM) {
6127 if (SvTYPE(sv) == SVt_PVHV) { /* stash backref */
6128 /* You lookin' at me? */
6129 assert(CvSTASH(referrer));
6130 assert(CvSTASH(referrer) == (const HV *)sv);
6131 SvANY(MUTABLE_CV(referrer))->xcv_stash = 0;
6134 assert(SvTYPE(sv) == SVt_PVGV);
6135 /* You lookin' at me? */
6136 assert(CvGV(referrer));
6137 assert(CvGV(referrer) == (const GV *)sv);
6138 anonymise_cv_maybe(MUTABLE_GV(sv),
6139 MUTABLE_CV(referrer));
6144 "panic: magic_killbackrefs (flags=%"UVxf")",
6145 (UV)SvFLAGS(referrer));
6156 SvREFCNT_dec_NN(av); /* remove extra count added by sv_add_backref() */
6162 =for apidoc sv_insert
6164 Inserts a string at the specified offset/length within the SV. Similar to
6165 the Perl C<substr()> function. Handles get magic.
6167 =for apidoc sv_insert_flags
6169 Same as C<sv_insert>, but the extra C<flags> are passed to the
6170 C<SvPV_force_flags> that applies to C<bigstr>.
6176 Perl_sv_insert_flags(pTHX_ SV *const bigstr, const STRLEN offset, const STRLEN len, const char *const little, const STRLEN littlelen, const U32 flags)
6182 SSize_t i; /* better be sizeof(STRLEN) or bad things happen */
6185 PERL_ARGS_ASSERT_SV_INSERT_FLAGS;
6187 SvPV_force_flags(bigstr, curlen, flags);
6188 (void)SvPOK_only_UTF8(bigstr);
6189 if (offset + len > curlen) {
6190 SvGROW(bigstr, offset+len+1);
6191 Zero(SvPVX(bigstr)+curlen, offset+len-curlen, char);
6192 SvCUR_set(bigstr, offset+len);
6196 i = littlelen - len;
6197 if (i > 0) { /* string might grow */
6198 big = SvGROW(bigstr, SvCUR(bigstr) + i + 1);
6199 mid = big + offset + len;
6200 midend = bigend = big + SvCUR(bigstr);
6203 while (midend > mid) /* shove everything down */
6204 *--bigend = *--midend;
6205 Move(little,big+offset,littlelen,char);
6206 SvCUR_set(bigstr, SvCUR(bigstr) + i);
6211 Move(little,SvPVX(bigstr)+offset,len,char);
6216 big = SvPVX(bigstr);
6219 bigend = big + SvCUR(bigstr);
6221 if (midend > bigend)
6222 Perl_croak(aTHX_ "panic: sv_insert, midend=%p, bigend=%p",
6225 if (mid - big > bigend - midend) { /* faster to shorten from end */
6227 Move(little, mid, littlelen,char);
6230 i = bigend - midend;
6232 Move(midend, mid, i,char);
6236 SvCUR_set(bigstr, mid - big);
6238 else if ((i = mid - big)) { /* faster from front */
6239 midend -= littlelen;
6241 Move(big, midend - i, i, char);
6242 sv_chop(bigstr,midend-i);
6244 Move(little, mid, littlelen,char);
6246 else if (littlelen) {
6247 midend -= littlelen;
6248 sv_chop(bigstr,midend);
6249 Move(little,midend,littlelen,char);
6252 sv_chop(bigstr,midend);
6258 =for apidoc sv_replace
6260 Make the first argument a copy of the second, then delete the original.
6261 The target SV physically takes over ownership of the body of the source SV
6262 and inherits its flags; however, the target keeps any magic it owns,
6263 and any magic in the source is discarded.
6264 Note that this is a rather specialist SV copying operation; most of the
6265 time you'll want to use C<sv_setsv> or one of its many macro front-ends.
6271 Perl_sv_replace(pTHX_ SV *const sv, SV *const nsv)
6273 const U32 refcnt = SvREFCNT(sv);
6275 PERL_ARGS_ASSERT_SV_REPLACE;
6277 SV_CHECK_THINKFIRST_COW_DROP(sv);
6278 if (SvREFCNT(nsv) != 1) {
6279 Perl_croak(aTHX_ "panic: reference miscount on nsv in sv_replace()"
6280 " (%" UVuf " != 1)", (UV) SvREFCNT(nsv));
6282 if (SvMAGICAL(sv)) {
6286 sv_upgrade(nsv, SVt_PVMG);
6287 SvMAGIC_set(nsv, SvMAGIC(sv));
6288 SvFLAGS(nsv) |= SvMAGICAL(sv);
6290 SvMAGIC_set(sv, NULL);
6294 assert(!SvREFCNT(sv));
6295 #ifdef DEBUG_LEAKING_SCALARS
6296 sv->sv_flags = nsv->sv_flags;
6297 sv->sv_any = nsv->sv_any;
6298 sv->sv_refcnt = nsv->sv_refcnt;
6299 sv->sv_u = nsv->sv_u;
6301 StructCopy(nsv,sv,SV);
6303 if(SvTYPE(sv) == SVt_IV) {
6304 SET_SVANY_FOR_BODYLESS_IV(sv);
6308 SvREFCNT(sv) = refcnt;
6309 SvFLAGS(nsv) |= SVTYPEMASK; /* Mark as freed */
6314 /* We're about to free a GV which has a CV that refers back to us.
6315 * If that CV will outlive us, make it anonymous (i.e. fix up its CvGV
6319 S_anonymise_cv_maybe(pTHX_ GV *gv, CV* cv)
6324 PERL_ARGS_ASSERT_ANONYMISE_CV_MAYBE;
6327 assert(SvREFCNT(gv) == 0);
6328 assert(isGV(gv) && isGV_with_GP(gv));
6330 assert(!CvANON(cv));
6331 assert(CvGV(cv) == gv);
6332 assert(!CvNAMED(cv));
6334 /* will the CV shortly be freed by gp_free() ? */
6335 if (GvCV(gv) == cv && GvGP(gv)->gp_refcnt < 2 && SvREFCNT(cv) < 2) {
6336 SvANY(cv)->xcv_gv_u.xcv_gv = NULL;
6340 /* if not, anonymise: */
6341 gvname = (GvSTASH(gv) && HvNAME(GvSTASH(gv)) && HvENAME(GvSTASH(gv)))
6342 ? newSVhek(HvENAME_HEK(GvSTASH(gv)))
6343 : newSVpvn_flags( "__ANON__", 8, 0 );
6344 sv_catpvs(gvname, "::__ANON__");
6345 anongv = gv_fetchsv(gvname, GV_ADDMULTI, SVt_PVCV);
6346 SvREFCNT_dec_NN(gvname);
6350 SvANY(cv)->xcv_gv_u.xcv_gv = MUTABLE_GV(SvREFCNT_inc(anongv));
6355 =for apidoc sv_clear
6357 Clear an SV: call any destructors, free up any memory used by the body,
6358 and free the body itself. The SV's head is I<not> freed, although
6359 its type is set to all 1's so that it won't inadvertently be assumed
6360 to be live during global destruction etc.
6361 This function should only be called when C<REFCNT> is zero. Most of the time
6362 you'll want to call C<sv_free()> (or its macro wrapper C<SvREFCNT_dec>)
6369 Perl_sv_clear(pTHX_ SV *const orig_sv)
6374 const struct body_details *sv_type_details;
6378 STRLEN hash_index = 0; /* initialise to make Coverity et al happy.
6379 Not strictly necessary */
6381 PERL_ARGS_ASSERT_SV_CLEAR;
6383 /* within this loop, sv is the SV currently being freed, and
6384 * iter_sv is the most recent AV or whatever that's being iterated
6385 * over to provide more SVs */
6391 assert(SvREFCNT(sv) == 0);
6392 assert(SvTYPE(sv) != (svtype)SVTYPEMASK);
6394 if (type <= SVt_IV) {
6395 /* See the comment in sv.h about the collusion between this
6396 * early return and the overloading of the NULL slots in the
6400 SvFLAGS(sv) &= SVf_BREAK;
6401 SvFLAGS(sv) |= SVTYPEMASK;
6405 /* objs are always >= MG, but pad names use the SVs_OBJECT flag
6406 for another purpose */
6407 assert(!SvOBJECT(sv) || type >= SVt_PVMG);
6409 if (type >= SVt_PVMG) {
6411 if (!curse(sv, 1)) goto get_next_sv;
6412 type = SvTYPE(sv); /* destructor may have changed it */
6414 /* Free back-references before magic, in case the magic calls
6415 * Perl code that has weak references to sv. */
6416 if (type == SVt_PVHV) {
6417 Perl_hv_kill_backrefs(aTHX_ MUTABLE_HV(sv));
6421 else if (SvMAGIC(sv)) {
6422 /* Free back-references before other types of magic. */
6423 sv_unmagic(sv, PERL_MAGIC_backref);
6429 /* case SVt_INVLIST: */
6432 IoIFP(sv) != PerlIO_stdin() &&
6433 IoIFP(sv) != PerlIO_stdout() &&
6434 IoIFP(sv) != PerlIO_stderr() &&
6435 !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6437 io_close(MUTABLE_IO(sv), NULL, FALSE,
6438 (IoTYPE(sv) == IoTYPE_WRONLY ||
6439 IoTYPE(sv) == IoTYPE_RDWR ||
6440 IoTYPE(sv) == IoTYPE_APPEND));
6442 if (IoDIRP(sv) && !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6443 PerlDir_close(IoDIRP(sv));
6444 IoDIRP(sv) = (DIR*)NULL;
6445 Safefree(IoTOP_NAME(sv));
6446 Safefree(IoFMT_NAME(sv));
6447 Safefree(IoBOTTOM_NAME(sv));
6448 if ((const GV *)sv == PL_statgv)
6452 /* FIXME for plugins */
6454 pregfree2((REGEXP*) sv);
6458 cv_undef(MUTABLE_CV(sv));
6459 /* If we're in a stash, we don't own a reference to it.
6460 * However it does have a back reference to us, which needs to
6462 if ((stash = CvSTASH(sv)))
6463 sv_del_backref(MUTABLE_SV(stash), sv);
6466 if (PL_last_swash_hv == (const HV *)sv) {
6467 PL_last_swash_hv = NULL;
6469 if (HvTOTALKEYS((HV*)sv) > 0) {
6471 /* this statement should match the one at the beginning of
6472 * hv_undef_flags() */
6473 if ( PL_phase != PERL_PHASE_DESTRUCT
6474 && (hek = HvNAME_HEK((HV*)sv)))
6476 if (PL_stashcache) {
6477 DEBUG_o(Perl_deb(aTHX_
6478 "sv_clear clearing PL_stashcache for '%"HEKf
6481 (void)hv_deletehek(PL_stashcache,
6484 hv_name_set((HV*)sv, NULL, 0, 0);
6487 /* save old iter_sv in unused SvSTASH field */
6488 assert(!SvOBJECT(sv));
6489 SvSTASH(sv) = (HV*)iter_sv;
6492 /* save old hash_index in unused SvMAGIC field */
6493 assert(!SvMAGICAL(sv));
6494 assert(!SvMAGIC(sv));
6495 ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index = hash_index;
6498 next_sv = Perl_hfree_next_entry(aTHX_ (HV*)sv, &hash_index);
6499 goto get_next_sv; /* process this new sv */
6501 /* free empty hash */
6502 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6503 assert(!HvARRAY((HV*)sv));
6507 AV* av = MUTABLE_AV(sv);
6508 if (PL_comppad == av) {
6512 if (AvREAL(av) && AvFILLp(av) > -1) {
6513 next_sv = AvARRAY(av)[AvFILLp(av)--];
6514 /* save old iter_sv in top-most slot of AV,
6515 * and pray that it doesn't get wiped in the meantime */
6516 AvARRAY(av)[AvMAX(av)] = iter_sv;
6518 goto get_next_sv; /* process this new sv */
6520 Safefree(AvALLOC(av));
6525 if (LvTYPE(sv) == 'T') { /* for tie: return HE to pool */
6526 SvREFCNT_dec(HeKEY_sv((HE*)LvTARG(sv)));
6527 HeNEXT((HE*)LvTARG(sv)) = PL_hv_fetch_ent_mh;
6528 PL_hv_fetch_ent_mh = (HE*)LvTARG(sv);
6530 else if (LvTYPE(sv) != 't') /* unless tie: unrefcnted fake SV** */
6531 SvREFCNT_dec(LvTARG(sv));
6532 if (isREGEXP(sv)) goto freeregexp;
6535 if (isGV_with_GP(sv)) {
6536 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
6537 && HvENAME_get(stash))
6538 mro_method_changed_in(stash);
6539 gp_free(MUTABLE_GV(sv));
6541 unshare_hek(GvNAME_HEK(sv));
6542 /* If we're in a stash, we don't own a reference to it.
6543 * However it does have a back reference to us, which
6544 * needs to be cleared. */
6545 if (!SvVALID(sv) && (stash = GvSTASH(sv)))
6546 sv_del_backref(MUTABLE_SV(stash), sv);
6548 /* FIXME. There are probably more unreferenced pointers to SVs
6549 * in the interpreter struct that we should check and tidy in
6550 * a similar fashion to this: */
6551 /* See also S_sv_unglob, which does the same thing. */
6552 if ((const GV *)sv == PL_last_in_gv)
6553 PL_last_in_gv = NULL;
6554 else if ((const GV *)sv == PL_statgv)
6556 else if ((const GV *)sv == PL_stderrgv)
6565 /* Don't bother with SvOOK_off(sv); as we're only going to
6569 SvOOK_offset(sv, offset);
6570 SvPV_set(sv, SvPVX_mutable(sv) - offset);
6571 /* Don't even bother with turning off the OOK flag. */
6576 SV * const target = SvRV(sv);
6578 sv_del_backref(target, sv);
6584 else if (SvPVX_const(sv)
6585 && !(SvTYPE(sv) == SVt_PVIO
6586 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6590 PerlIO_printf(Perl_debug_log, "Copy on write: clear\n");
6594 if (CowREFCNT(sv)) {
6601 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6606 Safefree(SvPVX_mutable(sv));
6610 else if (SvPVX_const(sv) && SvLEN(sv)
6611 && !(SvTYPE(sv) == SVt_PVIO
6612 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6613 Safefree(SvPVX_mutable(sv));
6614 else if (SvPVX_const(sv) && SvIsCOW(sv)) {
6615 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6625 SvFLAGS(sv) &= SVf_BREAK;
6626 SvFLAGS(sv) |= SVTYPEMASK;
6628 sv_type_details = bodies_by_type + type;
6629 if (sv_type_details->arena) {
6630 del_body(((char *)SvANY(sv) + sv_type_details->offset),
6631 &PL_body_roots[type]);
6633 else if (sv_type_details->body_size) {
6634 safefree(SvANY(sv));
6638 /* caller is responsible for freeing the head of the original sv */
6639 if (sv != orig_sv && !SvREFCNT(sv))
6642 /* grab and free next sv, if any */
6650 else if (!iter_sv) {
6652 } else if (SvTYPE(iter_sv) == SVt_PVAV) {
6653 AV *const av = (AV*)iter_sv;
6654 if (AvFILLp(av) > -1) {
6655 sv = AvARRAY(av)[AvFILLp(av)--];
6657 else { /* no more elements of current AV to free */
6660 /* restore previous value, squirrelled away */
6661 iter_sv = AvARRAY(av)[AvMAX(av)];
6662 Safefree(AvALLOC(av));
6665 } else if (SvTYPE(iter_sv) == SVt_PVHV) {
6666 sv = Perl_hfree_next_entry(aTHX_ (HV*)iter_sv, &hash_index);
6667 if (!sv && !HvTOTALKEYS((HV *)iter_sv)) {
6668 /* no more elements of current HV to free */
6671 /* Restore previous values of iter_sv and hash_index,
6672 * squirrelled away */
6673 assert(!SvOBJECT(sv));
6674 iter_sv = (SV*)SvSTASH(sv);
6675 assert(!SvMAGICAL(sv));
6676 hash_index = ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index;
6678 /* perl -DA does not like rubbish in SvMAGIC. */
6682 /* free any remaining detritus from the hash struct */
6683 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6684 assert(!HvARRAY((HV*)sv));
6689 /* unrolled SvREFCNT_dec and sv_free2 follows: */
6693 if (!SvREFCNT(sv)) {
6697 if (--(SvREFCNT(sv)))
6701 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6702 "Attempt to free temp prematurely: SV 0x%"UVxf
6703 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6707 if (SvIMMORTAL(sv)) {
6708 /* make sure SvREFCNT(sv)==0 happens very seldom */
6709 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6718 /* This routine curses the sv itself, not the object referenced by sv. So
6719 sv does not have to be ROK. */
6722 S_curse(pTHX_ SV * const sv, const bool check_refcnt) {
6723 PERL_ARGS_ASSERT_CURSE;
6724 assert(SvOBJECT(sv));
6726 if (PL_defstash && /* Still have a symbol table? */
6732 stash = SvSTASH(sv);
6733 assert(SvTYPE(stash) == SVt_PVHV);
6734 if (HvNAME(stash)) {
6735 CV* destructor = NULL;
6736 assert (SvOOK(stash));
6737 if (!SvOBJECT(stash)) destructor = (CV *)SvSTASH(stash);
6738 if (!destructor || HvMROMETA(stash)->destroy_gen
6739 != PL_sub_generation)
6742 gv_fetchmeth_autoload(stash, "DESTROY", 7, 0);
6743 if (gv) destructor = GvCV(gv);
6744 if (!SvOBJECT(stash))
6747 destructor ? (HV *)destructor : ((HV *)0)+1;
6748 HvAUX(stash)->xhv_mro_meta->destroy_gen =
6752 assert(!destructor || destructor == ((CV *)0)+1
6753 || SvTYPE(destructor) == SVt_PVCV);
6754 if (destructor && destructor != ((CV *)0)+1
6755 /* A constant subroutine can have no side effects, so
6756 don't bother calling it. */
6757 && !CvCONST(destructor)
6758 /* Don't bother calling an empty destructor or one that
6759 returns immediately. */
6760 && (CvISXSUB(destructor)
6761 || (CvSTART(destructor)
6762 && (CvSTART(destructor)->op_next->op_type
6764 && (CvSTART(destructor)->op_next->op_type
6766 || CvSTART(destructor)->op_next->op_next->op_type
6772 SV* const tmpref = newRV(sv);
6773 SvREADONLY_on(tmpref); /* DESTROY() could be naughty */
6775 PUSHSTACKi(PERLSI_DESTROY);
6780 call_sv(MUTABLE_SV(destructor),
6781 G_DISCARD|G_EVAL|G_KEEPERR|G_VOID);
6785 if(SvREFCNT(tmpref) < 2) {
6786 /* tmpref is not kept alive! */
6788 SvRV_set(tmpref, NULL);
6791 SvREFCNT_dec_NN(tmpref);
6794 } while (SvOBJECT(sv) && SvSTASH(sv) != stash);
6797 if (check_refcnt && SvREFCNT(sv)) {
6798 if (PL_in_clean_objs)
6800 "DESTROY created new reference to dead object '%"HEKf"'",
6801 HEKfARG(HvNAME_HEK(stash)));
6802 /* DESTROY gave object new lease on life */
6808 HV * const stash = SvSTASH(sv);
6809 /* Curse before freeing the stash, as freeing the stash could cause
6810 a recursive call into S_curse. */
6811 SvOBJECT_off(sv); /* Curse the object. */
6812 SvSTASH_set(sv,0); /* SvREFCNT_dec may try to read this */
6813 SvREFCNT_dec(stash); /* possibly of changed persuasion */
6819 =for apidoc sv_newref
6821 Increment an SV's reference count. Use the C<SvREFCNT_inc()> wrapper
6828 Perl_sv_newref(pTHX_ SV *const sv)
6830 PERL_UNUSED_CONTEXT;
6839 Decrement an SV's reference count, and if it drops to zero, call
6840 C<sv_clear> to invoke destructors and free up any memory used by
6841 the body; finally, deallocating the SV's head itself.
6842 Normally called via a wrapper macro C<SvREFCNT_dec>.
6848 Perl_sv_free(pTHX_ SV *const sv)
6854 /* Private helper function for SvREFCNT_dec().
6855 * Called with rc set to original SvREFCNT(sv), where rc == 0 or 1 */
6858 Perl_sv_free2(pTHX_ SV *const sv, const U32 rc)
6862 PERL_ARGS_ASSERT_SV_FREE2;
6864 if (LIKELY( rc == 1 )) {
6870 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6871 "Attempt to free temp prematurely: SV 0x%"UVxf
6872 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6876 if (SvIMMORTAL(sv)) {
6877 /* make sure SvREFCNT(sv)==0 happens very seldom */
6878 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6882 if (! SvREFCNT(sv)) /* may have have been resurrected */
6887 /* handle exceptional cases */
6891 if (SvFLAGS(sv) & SVf_BREAK)
6892 /* this SV's refcnt has been artificially decremented to
6893 * trigger cleanup */
6895 if (PL_in_clean_all) /* All is fair */
6897 if (SvIMMORTAL(sv)) {
6898 /* make sure SvREFCNT(sv)==0 happens very seldom */
6899 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6902 if (ckWARN_d(WARN_INTERNAL)) {
6903 #ifdef DEBUG_LEAKING_SCALARS_FORK_DUMP
6904 Perl_dump_sv_child(aTHX_ sv);
6906 #ifdef DEBUG_LEAKING_SCALARS
6909 #ifdef DEBUG_LEAKING_SCALARS_ABORT
6910 if (PL_warnhook == PERL_WARNHOOK_FATAL
6911 || ckDEAD(packWARN(WARN_INTERNAL))) {
6912 /* Don't let Perl_warner cause us to escape our fate: */
6916 /* This may not return: */
6917 Perl_warner(aTHX_ packWARN(WARN_INTERNAL),
6918 "Attempt to free unreferenced scalar: SV 0x%"UVxf
6919 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6922 #ifdef DEBUG_LEAKING_SCALARS_ABORT
6932 Returns the length of the string in the SV. Handles magic and type
6933 coercion and sets the UTF8 flag appropriately. See also C<L</SvCUR>>, which
6934 gives raw access to the C<xpv_cur> slot.
6940 Perl_sv_len(pTHX_ SV *const sv)
6947 (void)SvPV_const(sv, len);
6952 =for apidoc sv_len_utf8
6954 Returns the number of characters in the string in an SV, counting wide
6955 UTF-8 bytes as a single character. Handles magic and type coercion.
6961 * The length is cached in PERL_MAGIC_utf8, in the mg_len field. Also the
6962 * mg_ptr is used, by sv_pos_u2b() and sv_pos_b2u() - see the comments below.
6963 * (Note that the mg_len is not the length of the mg_ptr field.
6964 * This allows the cache to store the character length of the string without
6965 * needing to malloc() extra storage to attach to the mg_ptr.)
6970 Perl_sv_len_utf8(pTHX_ SV *const sv)
6976 return sv_len_utf8_nomg(sv);
6980 Perl_sv_len_utf8_nomg(pTHX_ SV * const sv)
6983 const U8 *s = (U8*)SvPV_nomg_const(sv, len);
6985 PERL_ARGS_ASSERT_SV_LEN_UTF8_NOMG;
6987 if (PL_utf8cache && SvUTF8(sv)) {
6989 MAGIC *mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_utf8) : NULL;
6991 if (mg && (mg->mg_len != -1 || mg->mg_ptr)) {
6992 if (mg->mg_len != -1)
6995 /* We can use the offset cache for a headstart.
6996 The longer value is stored in the first pair. */
6997 STRLEN *cache = (STRLEN *) mg->mg_ptr;
6999 ulen = cache[0] + Perl_utf8_length(aTHX_ s + cache[1],
7003 if (PL_utf8cache < 0) {
7004 const STRLEN real = Perl_utf8_length(aTHX_ s, s + len);
7005 assert_uft8_cache_coherent("sv_len_utf8", ulen, real, sv);
7009 ulen = Perl_utf8_length(aTHX_ s, s + len);
7010 utf8_mg_len_cache_update(sv, &mg, ulen);
7014 return SvUTF8(sv) ? Perl_utf8_length(aTHX_ s, s + len) : len;
7017 /* Walk forwards to find the byte corresponding to the passed in UTF-8
7020 S_sv_pos_u2b_forwards(const U8 *const start, const U8 *const send,
7021 STRLEN *const uoffset_p, bool *const at_end)
7023 const U8 *s = start;
7024 STRLEN uoffset = *uoffset_p;
7026 PERL_ARGS_ASSERT_SV_POS_U2B_FORWARDS;
7028 while (s < send && uoffset) {
7035 else if (s > send) {
7037 /* This is the existing behaviour. Possibly it should be a croak, as
7038 it's actually a bounds error */
7041 *uoffset_p -= uoffset;
7045 /* Given the length of the string in both bytes and UTF-8 characters, decide
7046 whether to walk forwards or backwards to find the byte corresponding to
7047 the passed in UTF-8 offset. */
7049 S_sv_pos_u2b_midway(const U8 *const start, const U8 *send,
7050 STRLEN uoffset, const STRLEN uend)
7052 STRLEN backw = uend - uoffset;
7054 PERL_ARGS_ASSERT_SV_POS_U2B_MIDWAY;
7056 if (uoffset < 2 * backw) {
7057 /* The assumption is that going forwards is twice the speed of going
7058 forward (that's where the 2 * backw comes from).
7059 (The real figure of course depends on the UTF-8 data.) */
7060 const U8 *s = start;
7062 while (s < send && uoffset--)
7072 while (UTF8_IS_CONTINUATION(*send))
7075 return send - start;
7078 /* For the string representation of the given scalar, find the byte
7079 corresponding to the passed in UTF-8 offset. uoffset0 and boffset0
7080 give another position in the string, *before* the sought offset, which
7081 (which is always true, as 0, 0 is a valid pair of positions), which should
7082 help reduce the amount of linear searching.
7083 If *mgp is non-NULL, it should point to the UTF-8 cache magic, which
7084 will be used to reduce the amount of linear searching. The cache will be
7085 created if necessary, and the found value offered to it for update. */
7087 S_sv_pos_u2b_cached(pTHX_ SV *const sv, MAGIC **const mgp, const U8 *const start,
7088 const U8 *const send, STRLEN uoffset,
7089 STRLEN uoffset0, STRLEN boffset0)
7091 STRLEN boffset = 0; /* Actually always set, but let's keep gcc happy. */
7093 bool at_end = FALSE;
7095 PERL_ARGS_ASSERT_SV_POS_U2B_CACHED;
7097 assert (uoffset >= uoffset0);
7102 if (!SvREADONLY(sv) && !SvGMAGICAL(sv) && SvPOK(sv)
7104 && (*mgp || (SvTYPE(sv) >= SVt_PVMG &&
7105 (*mgp = mg_find(sv, PERL_MAGIC_utf8))))) {
7106 if ((*mgp)->mg_ptr) {
7107 STRLEN *cache = (STRLEN *) (*mgp)->mg_ptr;
7108 if (cache[0] == uoffset) {
7109 /* An exact match. */
7112 if (cache[2] == uoffset) {
7113 /* An exact match. */
7117 if (cache[0] < uoffset) {
7118 /* The cache already knows part of the way. */
7119 if (cache[0] > uoffset0) {
7120 /* The cache knows more than the passed in pair */
7121 uoffset0 = cache[0];
7122 boffset0 = cache[1];
7124 if ((*mgp)->mg_len != -1) {
7125 /* And we know the end too. */
7127 + sv_pos_u2b_midway(start + boffset0, send,
7129 (*mgp)->mg_len - uoffset0);
7131 uoffset -= uoffset0;
7133 + sv_pos_u2b_forwards(start + boffset0,
7134 send, &uoffset, &at_end);
7135 uoffset += uoffset0;
7138 else if (cache[2] < uoffset) {
7139 /* We're between the two cache entries. */
7140 if (cache[2] > uoffset0) {
7141 /* and the cache knows more than the passed in pair */
7142 uoffset0 = cache[2];
7143 boffset0 = cache[3];
7147 + sv_pos_u2b_midway(start + boffset0,
7150 cache[0] - uoffset0);
7153 + sv_pos_u2b_midway(start + boffset0,
7156 cache[2] - uoffset0);
7160 else if ((*mgp)->mg_len != -1) {
7161 /* If we can take advantage of a passed in offset, do so. */
7162 /* In fact, offset0 is either 0, or less than offset, so don't
7163 need to worry about the other possibility. */
7165 + sv_pos_u2b_midway(start + boffset0, send,
7167 (*mgp)->mg_len - uoffset0);
7172 if (!found || PL_utf8cache < 0) {
7173 STRLEN real_boffset;
7174 uoffset -= uoffset0;
7175 real_boffset = boffset0 + sv_pos_u2b_forwards(start + boffset0,
7176 send, &uoffset, &at_end);
7177 uoffset += uoffset0;
7179 if (found && PL_utf8cache < 0)
7180 assert_uft8_cache_coherent("sv_pos_u2b_cache", boffset,
7182 boffset = real_boffset;
7185 if (PL_utf8cache && !SvGMAGICAL(sv) && SvPOK(sv)) {
7187 utf8_mg_len_cache_update(sv, mgp, uoffset);
7189 utf8_mg_pos_cache_update(sv, mgp, boffset, uoffset, send - start);
7196 =for apidoc sv_pos_u2b_flags
7198 Converts the offset from a count of UTF-8 chars from
7199 the start of the string, to a count of the equivalent number of bytes; if
7200 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7201 C<offset>, rather than from the start
7202 of the string. Handles type coercion.
7203 C<flags> is passed to C<SvPV_flags>, and usually should be
7204 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7210 * sv_pos_u2b_flags() uses, like sv_pos_b2u(), the mg_ptr of the potential
7211 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7212 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7217 Perl_sv_pos_u2b_flags(pTHX_ SV *const sv, STRLEN uoffset, STRLEN *const lenp,
7224 PERL_ARGS_ASSERT_SV_POS_U2B_FLAGS;
7226 start = (U8*)SvPV_flags(sv, len, flags);
7228 const U8 * const send = start + len;
7230 boffset = sv_pos_u2b_cached(sv, &mg, start, send, uoffset, 0, 0);
7233 && *lenp /* don't bother doing work for 0, as its bytes equivalent
7234 is 0, and *lenp is already set to that. */) {
7235 /* Convert the relative offset to absolute. */
7236 const STRLEN uoffset2 = uoffset + *lenp;
7237 const STRLEN boffset2
7238 = sv_pos_u2b_cached(sv, &mg, start, send, uoffset2,
7239 uoffset, boffset) - boffset;
7253 =for apidoc sv_pos_u2b
7255 Converts the value pointed to by C<offsetp> from a count of UTF-8 chars from
7256 the start of the string, to a count of the equivalent number of bytes; if
7257 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7258 the offset, rather than from the start of the string. Handles magic and
7261 Use C<sv_pos_u2b_flags> in preference, which correctly handles strings longer
7268 * sv_pos_u2b() uses, like sv_pos_b2u(), the mg_ptr of the potential
7269 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7270 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7274 /* This function is subject to size and sign problems */
7277 Perl_sv_pos_u2b(pTHX_ SV *const sv, I32 *const offsetp, I32 *const lenp)
7279 PERL_ARGS_ASSERT_SV_POS_U2B;
7282 STRLEN ulen = (STRLEN)*lenp;
7283 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, &ulen,
7284 SV_GMAGIC|SV_CONST_RETURN);
7287 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, NULL,
7288 SV_GMAGIC|SV_CONST_RETURN);
7293 S_utf8_mg_len_cache_update(pTHX_ SV *const sv, MAGIC **const mgp,
7296 PERL_ARGS_ASSERT_UTF8_MG_LEN_CACHE_UPDATE;
7297 if (SvREADONLY(sv) || SvGMAGICAL(sv) || !SvPOK(sv))
7300 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7301 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7302 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, &PL_vtbl_utf8, 0, 0);
7306 (*mgp)->mg_len = ulen;
7309 /* Create and update the UTF8 magic offset cache, with the proffered utf8/
7310 byte length pairing. The (byte) length of the total SV is passed in too,
7311 as blen, because for some (more esoteric) SVs, the call to SvPV_const()
7312 may not have updated SvCUR, so we can't rely on reading it directly.
7314 The proffered utf8/byte length pairing isn't used if the cache already has
7315 two pairs, and swapping either for the proffered pair would increase the
7316 RMS of the intervals between known byte offsets.
7318 The cache itself consists of 4 STRLEN values
7319 0: larger UTF-8 offset
7320 1: corresponding byte offset
7321 2: smaller UTF-8 offset
7322 3: corresponding byte offset
7324 Unused cache pairs have the value 0, 0.
7325 Keeping the cache "backwards" means that the invariant of
7326 cache[0] >= cache[2] is maintained even with empty slots, which means that
7327 the code that uses it doesn't need to worry if only 1 entry has actually
7328 been set to non-zero. It also makes the "position beyond the end of the
7329 cache" logic much simpler, as the first slot is always the one to start
7333 S_utf8_mg_pos_cache_update(pTHX_ SV *const sv, MAGIC **const mgp, const STRLEN byte,
7334 const STRLEN utf8, const STRLEN blen)
7338 PERL_ARGS_ASSERT_UTF8_MG_POS_CACHE_UPDATE;
7343 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7344 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7345 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, (MGVTBL*)&PL_vtbl_utf8, 0,
7347 (*mgp)->mg_len = -1;
7351 if (!(cache = (STRLEN *)(*mgp)->mg_ptr)) {
7352 Newxz(cache, PERL_MAGIC_UTF8_CACHESIZE * 2, STRLEN);
7353 (*mgp)->mg_ptr = (char *) cache;
7357 if (PL_utf8cache < 0 && SvPOKp(sv)) {
7358 /* SvPOKp() because, if sv is a reference, then SvPVX() is actually
7359 a pointer. Note that we no longer cache utf8 offsets on refer-
7360 ences, but this check is still a good idea, for robustness. */
7361 const U8 *start = (const U8 *) SvPVX_const(sv);
7362 const STRLEN realutf8 = utf8_length(start, start + byte);
7364 assert_uft8_cache_coherent("utf8_mg_pos_cache_update", utf8, realutf8,
7368 /* Cache is held with the later position first, to simplify the code
7369 that deals with unbounded ends. */
7371 ASSERT_UTF8_CACHE(cache);
7372 if (cache[1] == 0) {
7373 /* Cache is totally empty */
7376 } else if (cache[3] == 0) {
7377 if (byte > cache[1]) {
7378 /* New one is larger, so goes first. */
7379 cache[2] = cache[0];
7380 cache[3] = cache[1];
7388 /* float casts necessary? XXX */
7389 #define THREEWAY_SQUARE(a,b,c,d) \
7390 ((float)((d) - (c))) * ((float)((d) - (c))) \
7391 + ((float)((c) - (b))) * ((float)((c) - (b))) \
7392 + ((float)((b) - (a))) * ((float)((b) - (a)))
7394 /* Cache has 2 slots in use, and we know three potential pairs.
7395 Keep the two that give the lowest RMS distance. Do the
7396 calculation in bytes simply because we always know the byte
7397 length. squareroot has the same ordering as the positive value,
7398 so don't bother with the actual square root. */
7399 if (byte > cache[1]) {
7400 /* New position is after the existing pair of pairs. */
7401 const float keep_earlier
7402 = THREEWAY_SQUARE(0, cache[3], byte, blen);
7403 const float keep_later
7404 = THREEWAY_SQUARE(0, cache[1], byte, blen);
7406 if (keep_later < keep_earlier) {
7407 cache[2] = cache[0];
7408 cache[3] = cache[1];
7414 const float keep_later = THREEWAY_SQUARE(0, byte, cache[1], blen);
7415 float b, c, keep_earlier;
7416 if (byte > cache[3]) {
7417 /* New position is between the existing pair of pairs. */
7418 b = (float)cache[3];
7421 /* New position is before the existing pair of pairs. */
7423 c = (float)cache[3];
7425 keep_earlier = THREEWAY_SQUARE(0, b, c, blen);
7426 if (byte > cache[3]) {
7427 if (keep_later < keep_earlier) {
7437 if (! (keep_later < keep_earlier)) {
7438 cache[0] = cache[2];
7439 cache[1] = cache[3];
7446 ASSERT_UTF8_CACHE(cache);
7449 /* We already know all of the way, now we may be able to walk back. The same
7450 assumption is made as in S_sv_pos_u2b_midway(), namely that walking
7451 backward is half the speed of walking forward. */
7453 S_sv_pos_b2u_midway(pTHX_ const U8 *const s, const U8 *const target,
7454 const U8 *end, STRLEN endu)
7456 const STRLEN forw = target - s;
7457 STRLEN backw = end - target;
7459 PERL_ARGS_ASSERT_SV_POS_B2U_MIDWAY;
7461 if (forw < 2 * backw) {
7462 return utf8_length(s, target);
7465 while (end > target) {
7467 while (UTF8_IS_CONTINUATION(*end)) {
7476 =for apidoc sv_pos_b2u_flags
7478 Converts C<offset> from a count of bytes from the start of the string, to
7479 a count of the equivalent number of UTF-8 chars. Handles type coercion.
7480 C<flags> is passed to C<SvPV_flags>, and usually should be
7481 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7487 * sv_pos_b2u_flags() uses, like sv_pos_u2b_flags(), the mg_ptr of the
7488 * potential PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8
7493 Perl_sv_pos_b2u_flags(pTHX_ SV *const sv, STRLEN const offset, U32 flags)
7496 STRLEN len = 0; /* Actually always set, but let's keep gcc happy. */
7502 PERL_ARGS_ASSERT_SV_POS_B2U_FLAGS;
7504 s = (const U8*)SvPV_flags(sv, blen, flags);
7507 Perl_croak(aTHX_ "panic: sv_pos_b2u: bad byte offset, blen=%"UVuf
7508 ", byte=%"UVuf, (UV)blen, (UV)offset);
7514 && SvTYPE(sv) >= SVt_PVMG
7515 && (mg = mg_find(sv, PERL_MAGIC_utf8)))
7518 STRLEN * const cache = (STRLEN *) mg->mg_ptr;
7519 if (cache[1] == offset) {
7520 /* An exact match. */
7523 if (cache[3] == offset) {
7524 /* An exact match. */
7528 if (cache[1] < offset) {
7529 /* We already know part of the way. */
7530 if (mg->mg_len != -1) {
7531 /* Actually, we know the end too. */
7533 + S_sv_pos_b2u_midway(aTHX_ s + cache[1], send,
7534 s + blen, mg->mg_len - cache[0]);
7536 len = cache[0] + utf8_length(s + cache[1], send);
7539 else if (cache[3] < offset) {
7540 /* We're between the two cached pairs, so we do the calculation
7541 offset by the byte/utf-8 positions for the earlier pair,
7542 then add the utf-8 characters from the string start to
7544 len = S_sv_pos_b2u_midway(aTHX_ s + cache[3], send,
7545 s + cache[1], cache[0] - cache[2])
7549 else { /* cache[3] > offset */
7550 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + cache[3],
7554 ASSERT_UTF8_CACHE(cache);
7556 } else if (mg->mg_len != -1) {
7557 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + blen, mg->mg_len);
7561 if (!found || PL_utf8cache < 0) {
7562 const STRLEN real_len = utf8_length(s, send);
7564 if (found && PL_utf8cache < 0)
7565 assert_uft8_cache_coherent("sv_pos_b2u", len, real_len, sv);
7571 utf8_mg_len_cache_update(sv, &mg, len);
7573 utf8_mg_pos_cache_update(sv, &mg, offset, len, blen);
7580 =for apidoc sv_pos_b2u
7582 Converts the value pointed to by C<offsetp> from a count of bytes from the
7583 start of the string, to a count of the equivalent number of UTF-8 chars.
7584 Handles magic and type coercion.
7586 Use C<sv_pos_b2u_flags> in preference, which correctly handles strings
7593 * sv_pos_b2u() uses, like sv_pos_u2b(), the mg_ptr of the potential
7594 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7599 Perl_sv_pos_b2u(pTHX_ SV *const sv, I32 *const offsetp)
7601 PERL_ARGS_ASSERT_SV_POS_B2U;
7606 *offsetp = (I32)sv_pos_b2u_flags(sv, (STRLEN)*offsetp,
7607 SV_GMAGIC|SV_CONST_RETURN);
7611 S_assert_uft8_cache_coherent(pTHX_ const char *const func, STRLEN from_cache,
7612 STRLEN real, SV *const sv)
7614 PERL_ARGS_ASSERT_ASSERT_UFT8_CACHE_COHERENT;
7616 /* As this is debugging only code, save space by keeping this test here,
7617 rather than inlining it in all the callers. */
7618 if (from_cache == real)
7621 /* Need to turn the assertions off otherwise we may recurse infinitely
7622 while printing error messages. */
7623 SAVEI8(PL_utf8cache);
7625 Perl_croak(aTHX_ "panic: %s cache %"UVuf" real %"UVuf" for %"SVf,
7626 func, (UV) from_cache, (UV) real, SVfARG(sv));
7632 Returns a boolean indicating whether the strings in the two SVs are
7633 identical. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7634 coerce its args to strings if necessary.
7636 =for apidoc sv_eq_flags
7638 Returns a boolean indicating whether the strings in the two SVs are
7639 identical. Is UTF-8 and S<C<'use bytes'>> aware and coerces its args to strings
7640 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get-magic, too.
7646 Perl_sv_eq_flags(pTHX_ SV *sv1, SV *sv2, const U32 flags)
7653 SV* svrecode = NULL;
7660 /* if pv1 and pv2 are the same, second SvPV_const call may
7661 * invalidate pv1 (if we are handling magic), so we may need to
7663 if (sv1 == sv2 && flags & SV_GMAGIC
7664 && (SvTHINKFIRST(sv1) || SvGMAGICAL(sv1))) {
7665 pv1 = SvPV_const(sv1, cur1);
7666 sv1 = newSVpvn_flags(pv1, cur1, SVs_TEMP | SvUTF8(sv2));
7668 pv1 = SvPV_flags_const(sv1, cur1, flags);
7676 pv2 = SvPV_flags_const(sv2, cur2, flags);
7678 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7679 /* Differing utf8ness.
7680 * Do not UTF8size the comparands as a side-effect. */
7683 svrecode = newSVpvn(pv2, cur2);
7684 sv_recode_to_utf8(svrecode, _get_encoding());
7685 pv2 = SvPV_const(svrecode, cur2);
7688 svrecode = newSVpvn(pv1, cur1);
7689 sv_recode_to_utf8(svrecode, _get_encoding());
7690 pv1 = SvPV_const(svrecode, cur1);
7692 /* Now both are in UTF-8. */
7694 SvREFCNT_dec_NN(svrecode);
7700 /* sv1 is the UTF-8 one */
7701 return bytes_cmp_utf8((const U8*)pv2, cur2,
7702 (const U8*)pv1, cur1) == 0;
7705 /* sv2 is the UTF-8 one */
7706 return bytes_cmp_utf8((const U8*)pv1, cur1,
7707 (const U8*)pv2, cur2) == 0;
7713 eq = (pv1 == pv2) || memEQ(pv1, pv2, cur1);
7715 SvREFCNT_dec(svrecode);
7723 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7724 string in C<sv1> is less than, equal to, or greater than the string in
7725 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7726 coerce its args to strings if necessary. See also C<L</sv_cmp_locale>>.
7728 =for apidoc sv_cmp_flags
7730 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7731 string in C<sv1> is less than, equal to, or greater than the string in
7732 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware and will coerce its args to strings
7733 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get magic. See
7734 also C<L</sv_cmp_locale_flags>>.
7740 Perl_sv_cmp(pTHX_ SV *const sv1, SV *const sv2)
7742 return sv_cmp_flags(sv1, sv2, SV_GMAGIC);
7746 Perl_sv_cmp_flags(pTHX_ SV *const sv1, SV *const sv2,
7750 const char *pv1, *pv2;
7752 SV *svrecode = NULL;
7759 pv1 = SvPV_flags_const(sv1, cur1, flags);
7766 pv2 = SvPV_flags_const(sv2, cur2, flags);
7768 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7769 /* Differing utf8ness.
7770 * Do not UTF8size the comparands as a side-effect. */
7773 svrecode = newSVpvn(pv2, cur2);
7774 sv_recode_to_utf8(svrecode, _get_encoding());
7775 pv2 = SvPV_const(svrecode, cur2);
7778 const int retval = -bytes_cmp_utf8((const U8*)pv2, cur2,
7779 (const U8*)pv1, cur1);
7780 return retval ? retval < 0 ? -1 : +1 : 0;
7785 svrecode = newSVpvn(pv1, cur1);
7786 sv_recode_to_utf8(svrecode, _get_encoding());
7787 pv1 = SvPV_const(svrecode, cur1);
7790 const int retval = bytes_cmp_utf8((const U8*)pv1, cur1,
7791 (const U8*)pv2, cur2);
7792 return retval ? retval < 0 ? -1 : +1 : 0;
7797 /* Here, if both are non-NULL, then they have the same UTF8ness. */
7800 cmp = cur2 ? -1 : 0;
7804 STRLEN shortest_len = cur1 < cur2 ? cur1 : cur2;
7807 if (! DO_UTF8(sv1)) {
7809 const I32 retval = memcmp((const void*)pv1,
7813 cmp = retval < 0 ? -1 : 1;
7814 } else if (cur1 == cur2) {
7817 cmp = cur1 < cur2 ? -1 : 1;
7821 else { /* Both are to be treated as UTF-EBCDIC */
7823 /* EBCDIC UTF-8 is complicated by the fact that it is based on I8
7824 * which remaps code points 0-255. We therefore generally have to
7825 * unmap back to the original values to get an accurate comparison.
7826 * But we don't have to do that for UTF-8 invariants, as by
7827 * definition, they aren't remapped, nor do we have to do it for
7828 * above-latin1 code points, as they also aren't remapped. (This
7829 * code also works on ASCII platforms, but the memcmp() above is
7832 const char *e = pv1 + shortest_len;
7834 /* Find the first bytes that differ between the two strings */
7835 while (pv1 < e && *pv1 == *pv2) {
7841 if (pv1 == e) { /* Are the same all the way to the end */
7845 cmp = cur1 < cur2 ? -1 : 1;
7848 else /* Here *pv1 and *pv2 are not equal, but all bytes earlier
7849 * in the strings were. The current bytes may or may not be
7850 * at the beginning of a character. But neither or both are
7851 * (or else earlier bytes would have been different). And
7852 * if we are in the middle of a character, the two
7853 * characters are comprised of the same number of bytes
7854 * (because in this case the start bytes are the same, and
7855 * the start bytes encode the character's length). */
7856 if (UTF8_IS_INVARIANT(*pv1))
7858 /* If both are invariants; can just compare directly */
7859 if (UTF8_IS_INVARIANT(*pv2)) {
7860 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
7862 else /* Since *pv1 is invariant, it is the whole character,
7863 which means it is at the beginning of a character.
7864 That means pv2 is also at the beginning of a
7865 character (see earlier comment). Since it isn't
7866 invariant, it must be a start byte. If it starts a
7867 character whose code point is above 255, that
7868 character is greater than any single-byte char, which
7870 if (UTF8_IS_ABOVE_LATIN1_START(*pv2))
7875 /* Here, pv2 points to a character composed of 2 bytes
7876 * whose code point is < 256. Get its code point and
7877 * compare with *pv1 */
7878 cmp = ((U8) *pv1 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
7883 else /* The code point starting at pv1 isn't a single byte */
7884 if (UTF8_IS_INVARIANT(*pv2))
7886 /* But here, the code point starting at *pv2 is a single byte,
7887 * and so *pv1 must begin a character, hence is a start byte.
7888 * If that character is above 255, it is larger than any
7889 * single-byte char, which *pv2 is */
7890 if (UTF8_IS_ABOVE_LATIN1_START(*pv1)) {
7894 /* Here, pv1 points to a character composed of 2 bytes
7895 * whose code point is < 256. Get its code point and
7896 * compare with the single byte character *pv2 */
7897 cmp = (EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1)) < (U8) *pv2)
7902 else /* Here, we've ruled out either *pv1 and *pv2 being
7903 invariant. That means both are part of variants, but not
7904 necessarily at the start of a character */
7905 if ( UTF8_IS_ABOVE_LATIN1_START(*pv1)
7906 || UTF8_IS_ABOVE_LATIN1_START(*pv2))
7908 /* Here, at least one is the start of a character, which means
7909 * the other is also a start byte. And the code point of at
7910 * least one of the characters is above 255. It is a
7911 * characteristic of UTF-EBCDIC that all start bytes for
7912 * above-latin1 code points are well behaved as far as code
7913 * point comparisons go, and all are larger than all other
7914 * start bytes, so the comparison with those is also well
7916 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
7919 /* Here both *pv1 and *pv2 are part of variant characters.
7920 * They could be both continuations, or both start characters.
7921 * (One or both could even be an illegal start character (for
7922 * an overlong) which for the purposes of sorting we treat as
7924 if (UTF8_IS_CONTINUATION(*pv1)) {
7926 /* If they are continuations for code points above 255,
7927 * then comparing the current byte is sufficient, as there
7928 * is no remapping of these and so the comparison is
7929 * well-behaved. We determine if they are such
7930 * continuations by looking at the preceding byte. It
7931 * could be a start byte, from which we can tell if it is
7932 * for an above 255 code point. Or it could be a
7933 * continuation, which means the character occupies at
7934 * least 3 bytes, so must be above 255. */
7935 if ( UTF8_IS_CONTINUATION(*(pv2 - 1))
7936 || UTF8_IS_ABOVE_LATIN1_START(*(pv2 -1)))
7938 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
7942 /* Here, the continuations are for code points below 256;
7943 * back up one to get to the start byte */
7948 /* We need to get the actual native code point of each of these
7949 * variants in order to compare them */
7950 cmp = ( EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1))
7951 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
7960 SvREFCNT_dec(svrecode);
7966 =for apidoc sv_cmp_locale
7968 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
7969 S<C<'use bytes'>> aware, handles get magic, and will coerce its args to strings
7970 if necessary. See also C<L</sv_cmp>>.
7972 =for apidoc sv_cmp_locale_flags
7974 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
7975 S<C<'use bytes'>> aware and will coerce its args to strings if necessary. If
7976 the flags contain C<SV_GMAGIC>, it handles get magic. See also
7977 C<L</sv_cmp_flags>>.
7983 Perl_sv_cmp_locale(pTHX_ SV *const sv1, SV *const sv2)
7985 return sv_cmp_locale_flags(sv1, sv2, SV_GMAGIC);
7989 Perl_sv_cmp_locale_flags(pTHX_ SV *const sv1, SV *const sv2,
7992 #ifdef USE_LOCALE_COLLATE
7998 if (PL_collation_standard)
8002 pv1 = sv1 ? sv_collxfrm_flags(sv1, &len1, flags) : (char *) NULL;
8004 pv2 = sv2 ? sv_collxfrm_flags(sv2, &len2, flags) : (char *) NULL;
8006 if (!pv1 || !len1) {
8017 retval = memcmp((void*)pv1, (void*)pv2, len1 < len2 ? len1 : len2);
8020 return retval < 0 ? -1 : 1;
8023 * When the result of collation is equality, that doesn't mean
8024 * that there are no differences -- some locales exclude some
8025 * characters from consideration. So to avoid false equalities,
8026 * we use the raw string as a tiebreaker.
8033 PERL_UNUSED_ARG(flags);
8034 #endif /* USE_LOCALE_COLLATE */
8036 return sv_cmp(sv1, sv2);
8040 #ifdef USE_LOCALE_COLLATE
8043 =for apidoc sv_collxfrm
8045 This calls C<sv_collxfrm_flags> with the SV_GMAGIC flag. See
8046 C<L</sv_collxfrm_flags>>.
8048 =for apidoc sv_collxfrm_flags
8050 Add Collate Transform magic to an SV if it doesn't already have it. If the
8051 flags contain C<SV_GMAGIC>, it handles get-magic.
8053 Any scalar variable may carry C<PERL_MAGIC_collxfrm> magic that contains the
8054 scalar data of the variable, but transformed to such a format that a normal
8055 memory comparison can be used to compare the data according to the locale
8062 Perl_sv_collxfrm_flags(pTHX_ SV *const sv, STRLEN *const nxp, const I32 flags)
8066 PERL_ARGS_ASSERT_SV_COLLXFRM_FLAGS;
8068 mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_collxfrm) : (MAGIC *) NULL;
8069 if (!mg || !mg->mg_ptr || *(U32*)mg->mg_ptr != PL_collation_ix) {
8075 Safefree(mg->mg_ptr);
8076 s = SvPV_flags_const(sv, len, flags);
8077 if ((xf = mem_collxfrm(s, len, &xlen))) {
8079 mg = sv_magicext(sv, 0, PERL_MAGIC_collxfrm, &PL_vtbl_collxfrm,
8093 if (mg && mg->mg_ptr) {
8095 return mg->mg_ptr + sizeof(PL_collation_ix);
8103 #endif /* USE_LOCALE_COLLATE */
8106 S_sv_gets_append_to_utf8(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8108 SV * const tsv = newSV(0);
8111 sv_gets(tsv, fp, 0);
8112 sv_utf8_upgrade_nomg(tsv);
8113 SvCUR_set(sv,append);
8116 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8120 S_sv_gets_read_record(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8123 const STRLEN recsize = SvUV(SvRV(PL_rs)); /* RsRECORD() guarantees > 0. */
8124 /* Grab the size of the record we're getting */
8125 char *buffer = SvGROW(sv, (STRLEN)(recsize + append + 1)) + append;
8132 /* With a true, record-oriented file on VMS, we need to use read directly
8133 * to ensure that we respect RMS record boundaries. The user is responsible
8134 * for providing a PL_rs value that corresponds to the FAB$W_MRS (maximum
8135 * record size) field. N.B. This is likely to produce invalid results on
8136 * varying-width character data when a record ends mid-character.
8138 fd = PerlIO_fileno(fp);
8140 && PerlLIO_fstat(fd, &st) == 0
8141 && (st.st_fab_rfm == FAB$C_VAR
8142 || st.st_fab_rfm == FAB$C_VFC
8143 || st.st_fab_rfm == FAB$C_FIX)) {
8145 bytesread = PerlLIO_read(fd, buffer, recsize);
8147 else /* in-memory file from PerlIO::Scalar
8148 * or not a record-oriented file
8152 bytesread = PerlIO_read(fp, buffer, recsize);
8154 /* At this point, the logic in sv_get() means that sv will
8155 be treated as utf-8 if the handle is utf8.
8157 if (PerlIO_isutf8(fp) && bytesread > 0) {
8158 char *bend = buffer + bytesread;
8159 char *bufp = buffer;
8160 size_t charcount = 0;
8161 bool charstart = TRUE;
8164 while (charcount < recsize) {
8165 /* count accumulated characters */
8166 while (bufp < bend) {
8168 skip = UTF8SKIP(bufp);
8170 if (bufp + skip > bend) {
8171 /* partial at the end */
8182 if (charcount < recsize) {
8184 STRLEN bufp_offset = bufp - buffer;
8185 SSize_t morebytesread;
8187 /* originally I read enough to fill any incomplete
8188 character and the first byte of the next
8189 character if needed, but if there's many
8190 multi-byte encoded characters we're going to be
8191 making a read call for every character beyond
8192 the original read size.
8194 So instead, read the rest of the character if
8195 any, and enough bytes to match at least the
8196 start bytes for each character we're going to
8200 readsize = recsize - charcount;
8202 readsize = skip - (bend - bufp) + recsize - charcount - 1;
8203 buffer = SvGROW(sv, append + bytesread + readsize + 1) + append;
8204 bend = buffer + bytesread;
8205 morebytesread = PerlIO_read(fp, bend, readsize);
8206 if (morebytesread <= 0) {
8207 /* we're done, if we still have incomplete
8208 characters the check code in sv_gets() will
8211 I'd originally considered doing
8212 PerlIO_ungetc() on all but the lead
8213 character of the incomplete character, but
8214 read() doesn't do that, so I don't.
8219 /* prepare to scan some more */
8220 bytesread += morebytesread;
8221 bend = buffer + bytesread;
8222 bufp = buffer + bufp_offset;
8230 SvCUR_set(sv, bytesread + append);
8231 buffer[bytesread] = '\0';
8232 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8238 Get a line from the filehandle and store it into the SV, optionally
8239 appending to the currently-stored string. If C<append> is not 0, the
8240 line is appended to the SV instead of overwriting it. C<append> should
8241 be set to the byte offset that the appended string should start at
8242 in the SV (typically, C<SvCUR(sv)> is a suitable choice).
8248 Perl_sv_gets(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8258 PERL_ARGS_ASSERT_SV_GETS;
8260 if (SvTHINKFIRST(sv))
8261 sv_force_normal_flags(sv, append ? 0 : SV_COW_DROP_PV);
8262 /* XXX. If you make this PVIV, then copy on write can copy scalars read
8264 However, perlbench says it's slower, because the existing swipe code
8265 is faster than copy on write.
8266 Swings and roundabouts. */
8267 SvUPGRADE(sv, SVt_PV);
8270 /* line is going to be appended to the existing buffer in the sv */
8271 if (PerlIO_isutf8(fp)) {
8273 sv_utf8_upgrade_nomg(sv);
8274 sv_pos_u2b(sv,&append,0);
8276 } else if (SvUTF8(sv)) {
8277 return S_sv_gets_append_to_utf8(aTHX_ sv, fp, append);
8283 /* not appending - "clear" the string by setting SvCUR to 0,
8284 * the pv is still avaiable. */
8287 if (PerlIO_isutf8(fp))
8290 if (IN_PERL_COMPILETIME) {
8291 /* we always read code in line mode */
8295 else if (RsSNARF(PL_rs)) {
8296 /* If it is a regular disk file use size from stat() as estimate
8297 of amount we are going to read -- may result in mallocing
8298 more memory than we really need if the layers below reduce
8299 the size we read (e.g. CRLF or a gzip layer).
8302 int fd = PerlIO_fileno(fp);
8303 if (fd >= 0 && (PerlLIO_fstat(fd, &st) == 0) && S_ISREG(st.st_mode)) {
8304 const Off_t offset = PerlIO_tell(fp);
8305 if (offset != (Off_t) -1 && st.st_size + append > offset) {
8306 #ifdef PERL_COPY_ON_WRITE
8307 /* Add an extra byte for the sake of copy-on-write's
8308 * buffer reference count. */
8309 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 2));
8311 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 1));
8318 else if (RsRECORD(PL_rs)) {
8319 return S_sv_gets_read_record(aTHX_ sv, fp, append);
8321 else if (RsPARA(PL_rs)) {
8327 /* Get $/ i.e. PL_rs into same encoding as stream wants */
8328 if (PerlIO_isutf8(fp)) {
8329 rsptr = SvPVutf8(PL_rs, rslen);
8332 if (SvUTF8(PL_rs)) {
8333 if (!sv_utf8_downgrade(PL_rs, TRUE)) {
8334 Perl_croak(aTHX_ "Wide character in $/");
8337 /* extract the raw pointer to the record separator */
8338 rsptr = SvPV_const(PL_rs, rslen);
8342 /* rslast is the last character in the record separator
8343 * note we don't use rslast except when rslen is true, so the
8344 * null assign is a placeholder. */
8345 rslast = rslen ? rsptr[rslen - 1] : '\0';
8347 if (rspara) { /* have to do this both before and after */
8348 do { /* to make sure file boundaries work right */
8351 i = PerlIO_getc(fp);
8355 PerlIO_ungetc(fp,i);
8361 /* See if we know enough about I/O mechanism to cheat it ! */
8363 /* This used to be #ifdef test - it is made run-time test for ease
8364 of abstracting out stdio interface. One call should be cheap
8365 enough here - and may even be a macro allowing compile
8369 if (PerlIO_fast_gets(fp)) {
8371 * We can do buffer based IO operations on this filehandle.
8373 * This means we can bypass a lot of subcalls and process
8374 * the buffer directly, it also means we know the upper bound
8375 * on the amount of data we might read of the current buffer
8376 * into our sv. Knowing this allows us to preallocate the pv
8377 * to be able to hold that maximum, which allows us to simplify
8378 * a lot of logic. */
8381 * We're going to steal some values from the stdio struct
8382 * and put EVERYTHING in the innermost loop into registers.
8384 STDCHAR *ptr; /* pointer into fp's read-ahead buffer */
8385 STRLEN bpx; /* length of the data in the target sv
8386 used to fix pointers after a SvGROW */
8387 I32 shortbuffered; /* If the pv buffer is shorter than the amount
8388 of data left in the read-ahead buffer.
8389 If 0 then the pv buffer can hold the full
8390 amount left, otherwise this is the amount it
8393 /* Here is some breathtakingly efficient cheating */
8395 /* When you read the following logic resist the urge to think
8396 * of record separators that are 1 byte long. They are an
8397 * uninteresting special (simple) case.
8399 * Instead think of record separators which are at least 2 bytes
8400 * long, and keep in mind that we need to deal with such
8401 * separators when they cross a read-ahead buffer boundary.
8403 * Also consider that we need to gracefully deal with separators
8404 * that may be longer than a single read ahead buffer.
8406 * Lastly do not forget we want to copy the delimiter as well. We
8407 * are copying all data in the file _up_to_and_including_ the separator
8410 * Now that you have all that in mind here is what is happening below:
8412 * 1. When we first enter the loop we do some memory book keeping to see
8413 * how much free space there is in the target SV. (This sub assumes that
8414 * it is operating on the same SV most of the time via $_ and that it is
8415 * going to be able to reuse the same pv buffer each call.) If there is
8416 * "enough" room then we set "shortbuffered" to how much space there is
8417 * and start reading forward.
8419 * 2. When we scan forward we copy from the read-ahead buffer to the target
8420 * SV's pv buffer. While we go we watch for the end of the read-ahead buffer,
8421 * and the end of the of pv, as well as for the "rslast", which is the last
8422 * char of the separator.
8424 * 3. When scanning forward if we see rslast then we jump backwards in *pv*
8425 * (which has a "complete" record up to the point we saw rslast) and check
8426 * it to see if it matches the separator. If it does we are done. If it doesn't
8427 * we continue on with the scan/copy.
8429 * 4. If we run out of read-ahead buffer (cnt goes to 0) then we have to get
8430 * the IO system to read the next buffer. We do this by doing a getc(), which
8431 * returns a single char read (or EOF), and prefills the buffer, and also
8432 * allows us to find out how full the buffer is. We use this information to
8433 * SvGROW() the sv to the size remaining in the buffer, after which we copy
8434 * the returned single char into the target sv, and then go back into scan
8437 * 5. If we run out of write-buffer then we SvGROW() it by the size of the
8438 * remaining space in the read-buffer.
8440 * Note that this code despite its twisty-turny nature is pretty darn slick.
8441 * It manages single byte separators, multi-byte cross boundary separators,
8442 * and cross-read-buffer separators cleanly and efficiently at the cost
8443 * of potentially greatly overallocating the target SV.
8449 /* get the number of bytes remaining in the read-ahead buffer
8450 * on first call on a given fp this will return 0.*/
8451 cnt = PerlIO_get_cnt(fp);
8453 /* make sure we have the room */
8454 if ((I32)(SvLEN(sv) - append) <= cnt + 1) {
8455 /* Not room for all of it
8456 if we are looking for a separator and room for some
8458 if (rslen && cnt > 80 && (I32)SvLEN(sv) > append) {
8459 /* just process what we have room for */
8460 shortbuffered = cnt - SvLEN(sv) + append + 1;
8461 cnt -= shortbuffered;
8464 /* ensure that the target sv has enough room to hold
8465 * the rest of the read-ahead buffer */
8467 /* remember that cnt can be negative */
8468 SvGROW(sv, (STRLEN)(append + (cnt <= 0 ? 2 : (cnt + 1))));
8472 /* we have enough room to hold the full buffer, lets scream */
8476 /* extract the pointer to sv's string buffer, offset by append as necessary */
8477 bp = (STDCHAR*)SvPVX_const(sv) + append; /* move these two too to registers */
8478 /* extract the point to the read-ahead buffer */
8479 ptr = (STDCHAR*)PerlIO_get_ptr(fp);
8481 /* some trace debug output */
8482 DEBUG_P(PerlIO_printf(Perl_debug_log,
8483 "Screamer: entering, ptr=%"UVuf", cnt=%ld\n",PTR2UV(ptr),(long)cnt));
8484 DEBUG_P(PerlIO_printf(Perl_debug_log,
8485 "Screamer: entering: PerlIO * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"
8487 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8488 PTR2UV(PerlIO_has_base(fp) ? PerlIO_get_base(fp) : 0)));
8492 /* if there is stuff left in the read-ahead buffer */
8494 /* if there is a separator */
8496 /* loop until we hit the end of the read-ahead buffer */
8497 while (cnt > 0) { /* this | eat */
8498 /* scan forward copying and searching for rslast as we go */
8500 if ((*bp++ = *ptr++) == rslast) /* really | dust */
8501 goto thats_all_folks; /* screams | sed :-) */
8505 /* no separator, slurp the full buffer */
8506 Copy(ptr, bp, cnt, char); /* this | eat */
8507 bp += cnt; /* screams | dust */
8508 ptr += cnt; /* louder | sed :-) */
8510 assert (!shortbuffered);
8511 goto cannot_be_shortbuffered;
8515 if (shortbuffered) { /* oh well, must extend */
8516 /* we didnt have enough room to fit the line into the target buffer
8517 * so we must extend the target buffer and keep going */
8518 cnt = shortbuffered;
8520 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8522 /* extned the target sv's buffer so it can hold the full read-ahead buffer */
8523 SvGROW(sv, SvLEN(sv) + append + cnt + 2);
8524 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8528 cannot_be_shortbuffered:
8529 /* we need to refill the read-ahead buffer if possible */
8531 DEBUG_P(PerlIO_printf(Perl_debug_log,
8532 "Screamer: going to getc, ptr=%"UVuf", cnt=%"IVdf"\n",
8533 PTR2UV(ptr),(IV)cnt));
8534 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* deregisterize cnt and ptr */
8536 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8537 "Screamer: pre: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf"\n",
8538 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8539 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8542 call PerlIO_getc() to let it prefill the lookahead buffer
8544 This used to call 'filbuf' in stdio form, but as that behaves like
8545 getc when cnt <= 0 we use PerlIO_getc here to avoid introducing
8546 another abstraction.
8548 Note we have to deal with the char in 'i' if we are not at EOF
8550 i = PerlIO_getc(fp); /* get more characters */
8552 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8553 "Screamer: post: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf"\n",
8554 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8555 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8557 /* find out how much is left in the read-ahead buffer, and rextract its pointer */
8558 cnt = PerlIO_get_cnt(fp);
8559 ptr = (STDCHAR*)PerlIO_get_ptr(fp); /* reregisterize cnt and ptr */
8560 DEBUG_P(PerlIO_printf(Perl_debug_log,
8561 "Screamer: after getc, ptr=%"UVuf", cnt=%"IVdf"\n",
8562 PTR2UV(ptr),(IV)cnt));
8564 if (i == EOF) /* all done for ever? */
8565 goto thats_really_all_folks;
8567 /* make sure we have enough space in the target sv */
8568 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8570 SvGROW(sv, bpx + cnt + 2);
8571 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8573 /* copy of the char we got from getc() */
8574 *bp++ = (STDCHAR)i; /* store character from PerlIO_getc */
8576 /* make sure we deal with the i being the last character of a separator */
8577 if (rslen && (STDCHAR)i == rslast) /* all done for now? */
8578 goto thats_all_folks;
8582 /* check if we have actually found the separator - only really applies
8584 if ((rslen > 1 && (STRLEN)(bp - (STDCHAR*)SvPVX_const(sv)) < rslen) ||
8585 memNE((char*)bp - rslen, rsptr, rslen))
8586 goto screamer; /* go back to the fray */
8587 thats_really_all_folks:
8589 cnt += shortbuffered;
8590 DEBUG_P(PerlIO_printf(Perl_debug_log,
8591 "Screamer: quitting, ptr=%"UVuf", cnt=%"IVdf"\n",PTR2UV(ptr),(IV)cnt));
8592 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* put these back or we're in trouble */
8593 DEBUG_P(PerlIO_printf(Perl_debug_log,
8594 "Screamer: end: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf
8596 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8597 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8599 SvCUR_set(sv, bp - (STDCHAR*)SvPVX_const(sv)); /* set length */
8600 DEBUG_P(PerlIO_printf(Perl_debug_log,
8601 "Screamer: done, len=%ld, string=|%.*s|\n",
8602 (long)SvCUR(sv),(int)SvCUR(sv),SvPVX_const(sv)));
8606 /*The big, slow, and stupid way. */
8607 #ifdef USE_HEAP_INSTEAD_OF_STACK /* Even slower way. */
8608 STDCHAR *buf = NULL;
8609 Newx(buf, 8192, STDCHAR);
8617 const STDCHAR * const bpe = buf + sizeof(buf);
8619 while ((i = PerlIO_getc(fp)) != EOF && (*bp++ = (STDCHAR)i) != rslast && bp < bpe)
8620 ; /* keep reading */
8624 cnt = PerlIO_read(fp,(char*)buf, sizeof(buf));
8625 /* Accommodate broken VAXC compiler, which applies U8 cast to
8626 * both args of ?: operator, causing EOF to change into 255
8629 i = (U8)buf[cnt - 1];
8635 cnt = 0; /* we do need to re-set the sv even when cnt <= 0 */
8637 sv_catpvn_nomg(sv, (char *) buf, cnt);
8639 sv_setpvn(sv, (char *) buf, cnt); /* "nomg" is implied */
8641 if (i != EOF && /* joy */
8643 SvCUR(sv) < rslen ||
8644 memNE(SvPVX_const(sv) + SvCUR(sv) - rslen, rsptr, rslen)))
8648 * If we're reading from a TTY and we get a short read,
8649 * indicating that the user hit his EOF character, we need
8650 * to notice it now, because if we try to read from the TTY
8651 * again, the EOF condition will disappear.
8653 * The comparison of cnt to sizeof(buf) is an optimization
8654 * that prevents unnecessary calls to feof().
8658 if (!(cnt < (I32)sizeof(buf) && PerlIO_eof(fp)))
8662 #ifdef USE_HEAP_INSTEAD_OF_STACK
8667 if (rspara) { /* have to do this both before and after */
8668 while (i != EOF) { /* to make sure file boundaries work right */
8669 i = PerlIO_getc(fp);
8671 PerlIO_ungetc(fp,i);
8677 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8683 Auto-increment of the value in the SV, doing string to numeric conversion
8684 if necessary. Handles 'get' magic and operator overloading.
8690 Perl_sv_inc(pTHX_ SV *const sv)
8699 =for apidoc sv_inc_nomg
8701 Auto-increment of the value in the SV, doing string to numeric conversion
8702 if necessary. Handles operator overloading. Skips handling 'get' magic.
8708 Perl_sv_inc_nomg(pTHX_ SV *const sv)
8715 if (SvTHINKFIRST(sv)) {
8716 if (SvREADONLY(sv)) {
8717 Perl_croak_no_modify();
8721 if (SvAMAGIC(sv) && AMG_CALLunary(sv, inc_amg))
8723 i = PTR2IV(SvRV(sv));
8727 else sv_force_normal_flags(sv, 0);
8729 flags = SvFLAGS(sv);
8730 if ((flags & (SVp_NOK|SVp_IOK)) == SVp_NOK) {
8731 /* It's (privately or publicly) a float, but not tested as an
8732 integer, so test it to see. */
8734 flags = SvFLAGS(sv);
8736 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8737 /* It's publicly an integer, or privately an integer-not-float */
8738 #ifdef PERL_PRESERVE_IVUV
8742 if (SvUVX(sv) == UV_MAX)
8743 sv_setnv(sv, UV_MAX_P1);
8745 (void)SvIOK_only_UV(sv);
8746 SvUV_set(sv, SvUVX(sv) + 1);
8748 if (SvIVX(sv) == IV_MAX)
8749 sv_setuv(sv, (UV)IV_MAX + 1);
8751 (void)SvIOK_only(sv);
8752 SvIV_set(sv, SvIVX(sv) + 1);
8757 if (flags & SVp_NOK) {
8758 const NV was = SvNVX(sv);
8759 if (LIKELY(!Perl_isinfnan(was)) &&
8760 NV_OVERFLOWS_INTEGERS_AT &&
8761 was >= NV_OVERFLOWS_INTEGERS_AT) {
8762 /* diag_listed_as: Lost precision when %s %f by 1 */
8763 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8764 "Lost precision when incrementing %" NVff " by 1",
8767 (void)SvNOK_only(sv);
8768 SvNV_set(sv, was + 1.0);
8772 if (!(flags & SVp_POK) || !*SvPVX_const(sv)) {
8773 if ((flags & SVTYPEMASK) < SVt_PVIV)
8774 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV ? SVt_PVIV : SVt_IV));
8775 (void)SvIOK_only(sv);
8780 while (isALPHA(*d)) d++;
8781 while (isDIGIT(*d)) d++;
8782 if (d < SvEND(sv)) {
8783 const int numtype = grok_number_flags(SvPVX_const(sv), SvCUR(sv), NULL, PERL_SCAN_TRAILING);
8784 #ifdef PERL_PRESERVE_IVUV
8785 /* Got to punt this as an integer if needs be, but we don't issue
8786 warnings. Probably ought to make the sv_iv_please() that does
8787 the conversion if possible, and silently. */
8788 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
8789 /* Need to try really hard to see if it's an integer.
8790 9.22337203685478e+18 is an integer.
8791 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
8792 so $a="9.22337203685478e+18"; $a+0; $a++
8793 needs to be the same as $a="9.22337203685478e+18"; $a++
8800 /* sv_2iv *should* have made this an NV */
8801 if (flags & SVp_NOK) {
8802 (void)SvNOK_only(sv);
8803 SvNV_set(sv, SvNVX(sv) + 1.0);
8806 /* I don't think we can get here. Maybe I should assert this
8807 And if we do get here I suspect that sv_setnv will croak. NWC
8809 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_inc punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"NVgf"\n",
8810 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
8812 #endif /* PERL_PRESERVE_IVUV */
8813 if (!numtype && ckWARN(WARN_NUMERIC))
8814 not_incrementable(sv);
8815 sv_setnv(sv,Atof(SvPVX_const(sv)) + 1.0);
8819 while (d >= SvPVX_const(sv)) {
8827 /* MKS: The original code here died if letters weren't consecutive.
8828 * at least it didn't have to worry about non-C locales. The
8829 * new code assumes that ('z'-'a')==('Z'-'A'), letters are
8830 * arranged in order (although not consecutively) and that only
8831 * [A-Za-z] are accepted by isALPHA in the C locale.
8833 if (isALPHA_FOLD_NE(*d, 'z')) {
8834 do { ++*d; } while (!isALPHA(*d));
8837 *(d--) -= 'z' - 'a';
8842 *(d--) -= 'z' - 'a' + 1;
8846 /* oh,oh, the number grew */
8847 SvGROW(sv, SvCUR(sv) + 2);
8848 SvCUR_set(sv, SvCUR(sv) + 1);
8849 for (d = SvPVX(sv) + SvCUR(sv); d > SvPVX_const(sv); d--)
8860 Auto-decrement of the value in the SV, doing string to numeric conversion
8861 if necessary. Handles 'get' magic and operator overloading.
8867 Perl_sv_dec(pTHX_ SV *const sv)
8876 =for apidoc sv_dec_nomg
8878 Auto-decrement of the value in the SV, doing string to numeric conversion
8879 if necessary. Handles operator overloading. Skips handling 'get' magic.
8885 Perl_sv_dec_nomg(pTHX_ SV *const sv)
8891 if (SvTHINKFIRST(sv)) {
8892 if (SvREADONLY(sv)) {
8893 Perl_croak_no_modify();
8897 if (SvAMAGIC(sv) && AMG_CALLunary(sv, dec_amg))
8899 i = PTR2IV(SvRV(sv));
8903 else sv_force_normal_flags(sv, 0);
8905 /* Unlike sv_inc we don't have to worry about string-never-numbers
8906 and keeping them magic. But we mustn't warn on punting */
8907 flags = SvFLAGS(sv);
8908 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8909 /* It's publicly an integer, or privately an integer-not-float */
8910 #ifdef PERL_PRESERVE_IVUV
8914 if (SvUVX(sv) == 0) {
8915 (void)SvIOK_only(sv);
8919 (void)SvIOK_only_UV(sv);
8920 SvUV_set(sv, SvUVX(sv) - 1);
8923 if (SvIVX(sv) == IV_MIN) {
8924 sv_setnv(sv, (NV)IV_MIN);
8928 (void)SvIOK_only(sv);
8929 SvIV_set(sv, SvIVX(sv) - 1);
8934 if (flags & SVp_NOK) {
8937 const NV was = SvNVX(sv);
8938 if (LIKELY(!Perl_isinfnan(was)) &&
8939 NV_OVERFLOWS_INTEGERS_AT &&
8940 was <= -NV_OVERFLOWS_INTEGERS_AT) {
8941 /* diag_listed_as: Lost precision when %s %f by 1 */
8942 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8943 "Lost precision when decrementing %" NVff " by 1",
8946 (void)SvNOK_only(sv);
8947 SvNV_set(sv, was - 1.0);
8951 if (!(flags & SVp_POK)) {
8952 if ((flags & SVTYPEMASK) < SVt_PVIV)
8953 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV) ? SVt_PVIV : SVt_IV);
8955 (void)SvIOK_only(sv);
8958 #ifdef PERL_PRESERVE_IVUV
8960 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), NULL);
8961 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
8962 /* Need to try really hard to see if it's an integer.
8963 9.22337203685478e+18 is an integer.
8964 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
8965 so $a="9.22337203685478e+18"; $a+0; $a--
8966 needs to be the same as $a="9.22337203685478e+18"; $a--
8973 /* sv_2iv *should* have made this an NV */
8974 if (flags & SVp_NOK) {
8975 (void)SvNOK_only(sv);
8976 SvNV_set(sv, SvNVX(sv) - 1.0);
8979 /* I don't think we can get here. Maybe I should assert this
8980 And if we do get here I suspect that sv_setnv will croak. NWC
8982 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_dec punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"NVgf"\n",
8983 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
8986 #endif /* PERL_PRESERVE_IVUV */
8987 sv_setnv(sv,Atof(SvPVX_const(sv)) - 1.0); /* punt */
8990 /* this define is used to eliminate a chunk of duplicated but shared logic
8991 * it has the suffix __SV_C to signal that it isnt API, and isnt meant to be
8992 * used anywhere but here - yves
8994 #define PUSH_EXTEND_MORTAL__SV_C(AnSv) \
8996 SSize_t ix = ++PL_tmps_ix; \
8997 if (UNLIKELY(ix >= PL_tmps_max)) \
8998 ix = tmps_grow_p(ix); \
8999 PL_tmps_stack[ix] = (AnSv); \
9003 =for apidoc sv_mortalcopy
9005 Creates a new SV which is a copy of the original SV (using C<sv_setsv>).
9006 The new SV is marked as mortal. It will be destroyed "soon", either by an
9007 explicit call to C<FREETMPS>, or by an implicit call at places such as
9008 statement boundaries. See also C<L</sv_newmortal>> and C<L</sv_2mortal>>.
9013 /* Make a string that will exist for the duration of the expression
9014 * evaluation. Actually, it may have to last longer than that, but
9015 * hopefully we won't free it until it has been assigned to a
9016 * permanent location. */
9019 Perl_sv_mortalcopy_flags(pTHX_ SV *const oldstr, U32 flags)
9023 if (flags & SV_GMAGIC)
9024 SvGETMAGIC(oldstr); /* before new_SV, in case it dies */
9026 sv_setsv_flags(sv,oldstr,flags & ~SV_GMAGIC);
9027 PUSH_EXTEND_MORTAL__SV_C(sv);
9033 =for apidoc sv_newmortal
9035 Creates a new null SV which is mortal. The reference count of the SV is
9036 set to 1. It will be destroyed "soon", either by an explicit call to
9037 C<FREETMPS>, or by an implicit call at places such as statement boundaries.
9038 See also C<L</sv_mortalcopy>> and C<L</sv_2mortal>>.
9044 Perl_sv_newmortal(pTHX)
9049 SvFLAGS(sv) = SVs_TEMP;
9050 PUSH_EXTEND_MORTAL__SV_C(sv);
9056 =for apidoc newSVpvn_flags
9058 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9059 characters) into it. The reference count for the
9060 SV is set to 1. Note that if C<len> is zero, Perl will create a zero length
9061 string. You are responsible for ensuring that the source string is at least
9062 C<len> bytes long. If the C<s> argument is NULL the new SV will be undefined.
9063 Currently the only flag bits accepted are C<SVf_UTF8> and C<SVs_TEMP>.
9064 If C<SVs_TEMP> is set, then C<sv_2mortal()> is called on the result before
9065 returning. If C<SVf_UTF8> is set, C<s>
9066 is considered to be in UTF-8 and the
9067 C<SVf_UTF8> flag will be set on the new SV.
9068 C<newSVpvn_utf8()> is a convenience wrapper for this function, defined as
9070 #define newSVpvn_utf8(s, len, u) \
9071 newSVpvn_flags((s), (len), (u) ? SVf_UTF8 : 0)
9077 Perl_newSVpvn_flags(pTHX_ const char *const s, const STRLEN len, const U32 flags)
9081 /* All the flags we don't support must be zero.
9082 And we're new code so I'm going to assert this from the start. */
9083 assert(!(flags & ~(SVf_UTF8|SVs_TEMP)));
9085 sv_setpvn(sv,s,len);
9087 /* This code used to do a sv_2mortal(), however we now unroll the call to
9088 * sv_2mortal() and do what it does ourselves here. Since we have asserted
9089 * that flags can only have the SVf_UTF8 and/or SVs_TEMP flags set above we
9090 * can use it to enable the sv flags directly (bypassing SvTEMP_on), which
9091 * in turn means we dont need to mask out the SVf_UTF8 flag below, which
9092 * means that we eliminate quite a few steps than it looks - Yves
9093 * (explaining patch by gfx) */
9095 SvFLAGS(sv) |= flags;
9097 if(flags & SVs_TEMP){
9098 PUSH_EXTEND_MORTAL__SV_C(sv);
9105 =for apidoc sv_2mortal
9107 Marks an existing SV as mortal. The SV will be destroyed "soon", either
9108 by an explicit call to C<FREETMPS>, or by an implicit call at places such as
9109 statement boundaries. C<SvTEMP()> is turned on which means that the SV's
9110 string buffer can be "stolen" if this SV is copied. See also
9111 C<L</sv_newmortal>> and C<L</sv_mortalcopy>>.
9117 Perl_sv_2mortal(pTHX_ SV *const sv)
9124 PUSH_EXTEND_MORTAL__SV_C(sv);
9132 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9133 characters) into it. The reference count for the
9134 SV is set to 1. If C<len> is zero, Perl will compute the length using
9135 C<strlen()>, (which means if you use this option, that C<s> can't have embedded
9136 C<NUL> characters and has to have a terminating C<NUL> byte).
9138 For efficiency, consider using C<newSVpvn> instead.
9144 Perl_newSVpv(pTHX_ const char *const s, const STRLEN len)
9149 sv_setpvn(sv, s, len || s == NULL ? len : strlen(s));
9154 =for apidoc newSVpvn
9156 Creates a new SV and copies a string into it, which may contain C<NUL> characters
9157 (C<\0>) and other binary data. The reference count for the SV is set to 1.
9158 Note that if C<len> is zero, Perl will create a zero length (Perl) string. You
9159 are responsible for ensuring that the source buffer is at least
9160 C<len> bytes long. If the C<buffer> argument is NULL the new SV will be
9167 Perl_newSVpvn(pTHX_ const char *const buffer, const STRLEN len)
9171 sv_setpvn(sv,buffer,len);
9176 =for apidoc newSVhek
9178 Creates a new SV from the hash key structure. It will generate scalars that
9179 point to the shared string table where possible. Returns a new (undefined)
9180 SV if C<hek> is NULL.
9186 Perl_newSVhek(pTHX_ const HEK *const hek)
9195 if (HEK_LEN(hek) == HEf_SVKEY) {
9196 return newSVsv(*(SV**)HEK_KEY(hek));
9198 const int flags = HEK_FLAGS(hek);
9199 if (flags & HVhek_WASUTF8) {
9201 Andreas would like keys he put in as utf8 to come back as utf8
9203 STRLEN utf8_len = HEK_LEN(hek);
9204 SV * const sv = newSV_type(SVt_PV);
9205 char *as_utf8 = (char *)bytes_to_utf8 ((U8*)HEK_KEY(hek), &utf8_len);
9206 /* bytes_to_utf8() allocates a new string, which we can repurpose: */
9207 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
9210 } else if (flags & HVhek_UNSHARED) {
9211 /* A hash that isn't using shared hash keys has to have
9212 the flag in every key so that we know not to try to call
9213 share_hek_hek on it. */
9215 SV * const sv = newSVpvn (HEK_KEY(hek), HEK_LEN(hek));
9220 /* This will be overwhelminly the most common case. */
9222 /* Inline most of newSVpvn_share(), because share_hek_hek() is far
9223 more efficient than sharepvn(). */
9227 sv_upgrade(sv, SVt_PV);
9228 SvPV_set(sv, (char *)HEK_KEY(share_hek_hek(hek)));
9229 SvCUR_set(sv, HEK_LEN(hek));
9241 =for apidoc newSVpvn_share
9243 Creates a new SV with its C<SvPVX_const> pointing to a shared string in the string
9244 table. If the string does not already exist in the table, it is
9245 created first. Turns on the C<SvIsCOW> flag (or C<READONLY>
9246 and C<FAKE> in 5.16 and earlier). If the C<hash> parameter
9247 is non-zero, that value is used; otherwise the hash is computed.
9248 The string's hash can later be retrieved from the SV
9249 with the C<SvSHARED_HASH()> macro. The idea here is
9250 that as the string table is used for shared hash keys these strings will have
9251 C<SvPVX_const == HeKEY> and hash lookup will avoid string compare.
9257 Perl_newSVpvn_share(pTHX_ const char *src, I32 len, U32 hash)
9261 bool is_utf8 = FALSE;
9262 const char *const orig_src = src;
9265 STRLEN tmplen = -len;
9267 /* See the note in hv.c:hv_fetch() --jhi */
9268 src = (char*)bytes_from_utf8((const U8*)src, &tmplen, &is_utf8);
9272 PERL_HASH(hash, src, len);
9274 /* The logic for this is inlined in S_mro_get_linear_isa_dfs(), so if it
9275 changes here, update it there too. */
9276 sv_upgrade(sv, SVt_PV);
9277 SvPV_set(sv, sharepvn(src, is_utf8?-len:len, hash));
9284 if (src != orig_src)
9290 =for apidoc newSVpv_share
9292 Like C<newSVpvn_share>, but takes a C<NUL>-terminated string instead of a
9299 Perl_newSVpv_share(pTHX_ const char *src, U32 hash)
9301 return newSVpvn_share(src, strlen(src), hash);
9304 #if defined(PERL_IMPLICIT_CONTEXT)
9306 /* pTHX_ magic can't cope with varargs, so this is a no-context
9307 * version of the main function, (which may itself be aliased to us).
9308 * Don't access this version directly.
9312 Perl_newSVpvf_nocontext(const char *const pat, ...)
9318 PERL_ARGS_ASSERT_NEWSVPVF_NOCONTEXT;
9320 va_start(args, pat);
9321 sv = vnewSVpvf(pat, &args);
9328 =for apidoc newSVpvf
9330 Creates a new SV and initializes it with the string formatted like
9337 Perl_newSVpvf(pTHX_ const char *const pat, ...)
9342 PERL_ARGS_ASSERT_NEWSVPVF;
9344 va_start(args, pat);
9345 sv = vnewSVpvf(pat, &args);
9350 /* backend for newSVpvf() and newSVpvf_nocontext() */
9353 Perl_vnewSVpvf(pTHX_ const char *const pat, va_list *const args)
9357 PERL_ARGS_ASSERT_VNEWSVPVF;
9360 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
9367 Creates a new SV and copies a floating point value into it.
9368 The reference count for the SV is set to 1.
9374 Perl_newSVnv(pTHX_ const NV n)
9386 Creates a new SV and copies an integer into it. The reference count for the
9393 Perl_newSViv(pTHX_ const IV i)
9399 /* Inlining ONLY the small relevant subset of sv_setiv here
9400 * for performance. Makes a significant difference. */
9402 /* We're starting from SVt_FIRST, so provided that's
9403 * actual 0, we don't have to unset any SV type flags
9404 * to promote to SVt_IV. */
9405 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9407 SET_SVANY_FOR_BODYLESS_IV(sv);
9408 SvFLAGS(sv) |= SVt_IV;
9420 Creates a new SV and copies an unsigned integer into it.
9421 The reference count for the SV is set to 1.
9427 Perl_newSVuv(pTHX_ const UV u)
9431 /* Inlining ONLY the small relevant subset of sv_setuv here
9432 * for performance. Makes a significant difference. */
9434 /* Using ivs is more efficient than using uvs - see sv_setuv */
9435 if (u <= (UV)IV_MAX) {
9436 return newSViv((IV)u);
9441 /* We're starting from SVt_FIRST, so provided that's
9442 * actual 0, we don't have to unset any SV type flags
9443 * to promote to SVt_IV. */
9444 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9446 SET_SVANY_FOR_BODYLESS_IV(sv);
9447 SvFLAGS(sv) |= SVt_IV;
9449 (void)SvIsUV_on(sv);
9458 =for apidoc newSV_type
9460 Creates a new SV, of the type specified. The reference count for the new SV
9467 Perl_newSV_type(pTHX_ const svtype type)
9472 ASSUME(SvTYPE(sv) == SVt_FIRST);
9473 if(type != SVt_FIRST)
9474 sv_upgrade(sv, type);
9479 =for apidoc newRV_noinc
9481 Creates an RV wrapper for an SV. The reference count for the original
9482 SV is B<not> incremented.
9488 Perl_newRV_noinc(pTHX_ SV *const tmpRef)
9492 PERL_ARGS_ASSERT_NEWRV_NOINC;
9496 /* We're starting from SVt_FIRST, so provided that's
9497 * actual 0, we don't have to unset any SV type flags
9498 * to promote to SVt_IV. */
9499 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9501 SET_SVANY_FOR_BODYLESS_IV(sv);
9502 SvFLAGS(sv) |= SVt_IV;
9507 SvRV_set(sv, tmpRef);
9512 /* newRV_inc is the official function name to use now.
9513 * newRV_inc is in fact #defined to newRV in sv.h
9517 Perl_newRV(pTHX_ SV *const sv)
9519 PERL_ARGS_ASSERT_NEWRV;
9521 return newRV_noinc(SvREFCNT_inc_simple_NN(sv));
9527 Creates a new SV which is an exact duplicate of the original SV.
9534 Perl_newSVsv(pTHX_ SV *const old)
9540 if (SvTYPE(old) == (svtype)SVTYPEMASK) {
9541 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), "semi-panic: attempt to dup freed string");
9544 /* Do this here, otherwise we leak the new SV if this croaks. */
9547 /* SV_NOSTEAL prevents TEMP buffers being, well, stolen, and saves games
9548 with SvTEMP_off and SvTEMP_on round a call to sv_setsv. */
9549 sv_setsv_flags(sv, old, SV_NOSTEAL);
9554 =for apidoc sv_reset
9556 Underlying implementation for the C<reset> Perl function.
9557 Note that the perl-level function is vaguely deprecated.
9563 Perl_sv_reset(pTHX_ const char *s, HV *const stash)
9565 PERL_ARGS_ASSERT_SV_RESET;
9567 sv_resetpvn(*s ? s : NULL, strlen(s), stash);
9571 Perl_sv_resetpvn(pTHX_ const char *s, STRLEN len, HV * const stash)
9573 char todo[PERL_UCHAR_MAX+1];
9576 if (!stash || SvTYPE(stash) != SVt_PVHV)
9579 if (!s) { /* reset ?? searches */
9580 MAGIC * const mg = mg_find((const SV *)stash, PERL_MAGIC_symtab);
9582 const U32 count = mg->mg_len / sizeof(PMOP**);
9583 PMOP **pmp = (PMOP**) mg->mg_ptr;
9584 PMOP *const *const end = pmp + count;
9588 SvREADONLY_off(PL_regex_pad[(*pmp)->op_pmoffset]);
9590 (*pmp)->op_pmflags &= ~PMf_USED;
9598 /* reset variables */
9600 if (!HvARRAY(stash))
9603 Zero(todo, 256, char);
9607 I32 i = (unsigned char)*s;
9611 max = (unsigned char)*s++;
9612 for ( ; i <= max; i++) {
9615 for (i = 0; i <= (I32) HvMAX(stash); i++) {
9617 for (entry = HvARRAY(stash)[i];
9619 entry = HeNEXT(entry))
9624 if (!todo[(U8)*HeKEY(entry)])
9626 gv = MUTABLE_GV(HeVAL(entry));
9628 if (sv && !SvREADONLY(sv)) {
9629 SV_CHECK_THINKFIRST_COW_DROP(sv);
9630 if (!isGV(sv)) SvOK_off(sv);
9635 if (GvHV(gv) && !HvNAME_get(GvHV(gv))) {
9646 Using various gambits, try to get an IO from an SV: the IO slot if its a
9647 GV; or the recursive result if we're an RV; or the IO slot of the symbol
9648 named after the PV if we're a string.
9650 'Get' magic is ignored on the C<sv> passed in, but will be called on
9651 C<SvRV(sv)> if C<sv> is an RV.
9657 Perl_sv_2io(pTHX_ SV *const sv)
9662 PERL_ARGS_ASSERT_SV_2IO;
9664 switch (SvTYPE(sv)) {
9666 io = MUTABLE_IO(sv);
9670 if (isGV_with_GP(sv)) {
9671 gv = MUTABLE_GV(sv);
9674 Perl_croak(aTHX_ "Bad filehandle: %"HEKf,
9675 HEKfARG(GvNAME_HEK(gv)));
9681 Perl_croak(aTHX_ PL_no_usym, "filehandle");
9683 SvGETMAGIC(SvRV(sv));
9684 return sv_2io(SvRV(sv));
9686 gv = gv_fetchsv_nomg(sv, 0, SVt_PVIO);
9693 if (SvGMAGICAL(sv)) {
9694 newsv = sv_newmortal();
9695 sv_setsv_nomg(newsv, sv);
9697 Perl_croak(aTHX_ "Bad filehandle: %"SVf, SVfARG(newsv));
9707 Using various gambits, try to get a CV from an SV; in addition, try if
9708 possible to set C<*st> and C<*gvp> to the stash and GV associated with it.
9709 The flags in C<lref> are passed to C<gv_fetchsv>.
9715 Perl_sv_2cv(pTHX_ SV *sv, HV **const st, GV **const gvp, const I32 lref)
9720 PERL_ARGS_ASSERT_SV_2CV;
9727 switch (SvTYPE(sv)) {
9731 return MUTABLE_CV(sv);
9741 sv = amagic_deref_call(sv, to_cv_amg);
9744 if (SvTYPE(sv) == SVt_PVCV) {
9745 cv = MUTABLE_CV(sv);
9750 else if(SvGETMAGIC(sv), isGV_with_GP(sv))
9751 gv = MUTABLE_GV(sv);
9753 Perl_croak(aTHX_ "Not a subroutine reference");
9755 else if (isGV_with_GP(sv)) {
9756 gv = MUTABLE_GV(sv);
9759 gv = gv_fetchsv_nomg(sv, lref, SVt_PVCV);
9766 /* Some flags to gv_fetchsv mean don't really create the GV */
9767 if (!isGV_with_GP(gv)) {
9772 if (lref & ~GV_ADDMG && !GvCVu(gv)) {
9773 /* XXX this is probably not what they think they're getting.
9774 * It has the same effect as "sub name;", i.e. just a forward
9785 Returns true if the SV has a true value by Perl's rules.
9786 Use the C<SvTRUE> macro instead, which may call C<sv_true()> or may
9787 instead use an in-line version.
9793 Perl_sv_true(pTHX_ SV *const sv)
9798 const XPV* const tXpv = (XPV*)SvANY(sv);
9800 (tXpv->xpv_cur > 1 ||
9801 (tXpv->xpv_cur && *sv->sv_u.svu_pv != '0')))
9808 return SvIVX(sv) != 0;
9811 return SvNVX(sv) != 0.0;
9813 return sv_2bool(sv);
9819 =for apidoc sv_pvn_force
9821 Get a sensible string out of the SV somehow.
9822 A private implementation of the C<SvPV_force> macro for compilers which
9823 can't cope with complex macro expressions. Always use the macro instead.
9825 =for apidoc sv_pvn_force_flags
9827 Get a sensible string out of the SV somehow.
9828 If C<flags> has the C<SV_GMAGIC> bit set, will C<mg_get> on C<sv> if
9829 appropriate, else not. C<sv_pvn_force> and C<sv_pvn_force_nomg> are
9830 implemented in terms of this function.
9831 You normally want to use the various wrapper macros instead: see
9832 C<L</SvPV_force>> and C<L</SvPV_force_nomg>>.
9838 Perl_sv_pvn_force_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
9840 PERL_ARGS_ASSERT_SV_PVN_FORCE_FLAGS;
9842 if (flags & SV_GMAGIC) SvGETMAGIC(sv);
9843 if (SvTHINKFIRST(sv) && (!SvROK(sv) || SvREADONLY(sv)))
9844 sv_force_normal_flags(sv, 0);
9854 if (SvTYPE(sv) > SVt_PVLV
9855 || isGV_with_GP(sv))
9856 /* diag_listed_as: Can't coerce %s to %s in %s */
9857 Perl_croak(aTHX_ "Can't coerce %s to string in %s", sv_reftype(sv,0),
9859 s = sv_2pv_flags(sv, &len, flags &~ SV_GMAGIC);
9866 if (SvTYPE(sv) < SVt_PV ||
9867 s != SvPVX_const(sv)) { /* Almost, but not quite, sv_setpvn() */
9870 SvUPGRADE(sv, SVt_PV); /* Never FALSE */
9871 SvGROW(sv, len + 1);
9872 Move(s,SvPVX(sv),len,char);
9874 SvPVX(sv)[len] = '\0';
9877 SvPOK_on(sv); /* validate pointer */
9879 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
9880 PTR2UV(sv),SvPVX_const(sv)));
9883 (void)SvPOK_only_UTF8(sv);
9884 return SvPVX_mutable(sv);
9888 =for apidoc sv_pvbyten_force
9890 The backend for the C<SvPVbytex_force> macro. Always use the macro
9897 Perl_sv_pvbyten_force(pTHX_ SV *const sv, STRLEN *const lp)
9899 PERL_ARGS_ASSERT_SV_PVBYTEN_FORCE;
9901 sv_pvn_force(sv,lp);
9902 sv_utf8_downgrade(sv,0);
9908 =for apidoc sv_pvutf8n_force
9910 The backend for the C<SvPVutf8x_force> macro. Always use the macro
9917 Perl_sv_pvutf8n_force(pTHX_ SV *const sv, STRLEN *const lp)
9919 PERL_ARGS_ASSERT_SV_PVUTF8N_FORCE;
9922 sv_utf8_upgrade_nomg(sv);
9928 =for apidoc sv_reftype
9930 Returns a string describing what the SV is a reference to.
9936 Perl_sv_reftype(pTHX_ const SV *const sv, const int ob)
9938 PERL_ARGS_ASSERT_SV_REFTYPE;
9939 if (ob && SvOBJECT(sv)) {
9940 return SvPV_nolen_const(sv_ref(NULL, sv, ob));
9943 /* WARNING - There is code, for instance in mg.c, that assumes that
9944 * the only reason that sv_reftype(sv,0) would return a string starting
9945 * with 'L' or 'S' is that it is a LVALUE or a SCALAR.
9946 * Yes this a dodgy way to do type checking, but it saves practically reimplementing
9947 * this routine inside other subs, and it saves time.
9948 * Do not change this assumption without searching for "dodgy type check" in
9951 switch (SvTYPE(sv)) {
9966 case SVt_PVLV: return (char *) (SvROK(sv) ? "REF"
9967 /* tied lvalues should appear to be
9968 * scalars for backwards compatibility */
9969 : (isALPHA_FOLD_EQ(LvTYPE(sv), 't'))
9970 ? "SCALAR" : "LVALUE");
9971 case SVt_PVAV: return "ARRAY";
9972 case SVt_PVHV: return "HASH";
9973 case SVt_PVCV: return "CODE";
9974 case SVt_PVGV: return (char *) (isGV_with_GP(sv)
9975 ? "GLOB" : "SCALAR");
9976 case SVt_PVFM: return "FORMAT";
9977 case SVt_PVIO: return "IO";
9978 case SVt_INVLIST: return "INVLIST";
9979 case SVt_REGEXP: return "REGEXP";
9980 default: return "UNKNOWN";
9988 Returns a SV describing what the SV passed in is a reference to.
9994 Perl_sv_ref(pTHX_ SV *dst, const SV *const sv, const int ob)
9996 PERL_ARGS_ASSERT_SV_REF;
9999 dst = sv_newmortal();
10001 if (ob && SvOBJECT(sv)) {
10002 HvNAME_get(SvSTASH(sv))
10003 ? sv_sethek(dst, HvNAME_HEK(SvSTASH(sv)))
10004 : sv_setpvn(dst, "__ANON__", 8);
10007 const char * reftype = sv_reftype(sv, 0);
10008 sv_setpv(dst, reftype);
10014 =for apidoc sv_isobject
10016 Returns a boolean indicating whether the SV is an RV pointing to a blessed
10017 object. If the SV is not an RV, or if the object is not blessed, then this
10024 Perl_sv_isobject(pTHX_ SV *sv)
10040 Returns a boolean indicating whether the SV is blessed into the specified
10041 class. This does not check for subtypes; use C<sv_derived_from> to verify
10042 an inheritance relationship.
10048 Perl_sv_isa(pTHX_ SV *sv, const char *const name)
10050 const char *hvname;
10052 PERL_ARGS_ASSERT_SV_ISA;
10062 hvname = HvNAME_get(SvSTASH(sv));
10066 return strEQ(hvname, name);
10070 =for apidoc newSVrv
10072 Creates a new SV for the existing RV, C<rv>, to point to. If C<rv> is not an
10073 RV then it will be upgraded to one. If C<classname> is non-null then the new
10074 SV will be blessed in the specified package. The new SV is returned and its
10075 reference count is 1. The reference count 1 is owned by C<rv>.
10081 Perl_newSVrv(pTHX_ SV *const rv, const char *const classname)
10085 PERL_ARGS_ASSERT_NEWSVRV;
10089 SV_CHECK_THINKFIRST_COW_DROP(rv);
10091 if (UNLIKELY( SvTYPE(rv) >= SVt_PVMG )) {
10092 const U32 refcnt = SvREFCNT(rv);
10096 SvREFCNT(rv) = refcnt;
10098 sv_upgrade(rv, SVt_IV);
10099 } else if (SvROK(rv)) {
10100 SvREFCNT_dec(SvRV(rv));
10102 prepare_SV_for_RV(rv);
10110 HV* const stash = gv_stashpv(classname, GV_ADD);
10111 (void)sv_bless(rv, stash);
10117 Perl_newSVavdefelem(pTHX_ AV *av, SSize_t ix, bool extendible)
10119 SV * const lv = newSV_type(SVt_PVLV);
10120 PERL_ARGS_ASSERT_NEWSVAVDEFELEM;
10122 sv_magic(lv, NULL, PERL_MAGIC_defelem, NULL, 0);
10123 LvTARG(lv) = SvREFCNT_inc_simple_NN(av);
10124 LvSTARGOFF(lv) = ix;
10125 LvTARGLEN(lv) = extendible ? 1 : (STRLEN)UV_MAX;
10130 =for apidoc sv_setref_pv
10132 Copies a pointer into a new SV, optionally blessing the SV. The C<rv>
10133 argument will be upgraded to an RV. That RV will be modified to point to
10134 the new SV. If the C<pv> argument is C<NULL>, then C<PL_sv_undef> will be placed
10135 into the SV. The C<classname> argument indicates the package for the
10136 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10137 will have a reference count of 1, and the RV will be returned.
10139 Do not use with other Perl types such as HV, AV, SV, CV, because those
10140 objects will become corrupted by the pointer copy process.
10142 Note that C<sv_setref_pvn> copies the string while this copies the pointer.
10148 Perl_sv_setref_pv(pTHX_ SV *const rv, const char *const classname, void *const pv)
10150 PERL_ARGS_ASSERT_SV_SETREF_PV;
10153 sv_setsv(rv, &PL_sv_undef);
10157 sv_setiv(newSVrv(rv,classname), PTR2IV(pv));
10162 =for apidoc sv_setref_iv
10164 Copies an integer into a new SV, optionally blessing the SV. The C<rv>
10165 argument will be upgraded to an RV. That RV will be modified to point to
10166 the new SV. The C<classname> argument indicates the package for the
10167 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10168 will have a reference count of 1, and the RV will be returned.
10174 Perl_sv_setref_iv(pTHX_ SV *const rv, const char *const classname, const IV iv)
10176 PERL_ARGS_ASSERT_SV_SETREF_IV;
10178 sv_setiv(newSVrv(rv,classname), iv);
10183 =for apidoc sv_setref_uv
10185 Copies an unsigned integer into a new SV, optionally blessing the SV. The C<rv>
10186 argument will be upgraded to an RV. That RV will be modified to point to
10187 the new SV. The C<classname> argument indicates the package for the
10188 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10189 will have a reference count of 1, and the RV will be returned.
10195 Perl_sv_setref_uv(pTHX_ SV *const rv, const char *const classname, const UV uv)
10197 PERL_ARGS_ASSERT_SV_SETREF_UV;
10199 sv_setuv(newSVrv(rv,classname), uv);
10204 =for apidoc sv_setref_nv
10206 Copies a double into a new SV, optionally blessing the SV. The C<rv>
10207 argument will be upgraded to an RV. That RV will be modified to point to
10208 the new SV. The C<classname> argument indicates the package for the
10209 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10210 will have a reference count of 1, and the RV will be returned.
10216 Perl_sv_setref_nv(pTHX_ SV *const rv, const char *const classname, const NV nv)
10218 PERL_ARGS_ASSERT_SV_SETREF_NV;
10220 sv_setnv(newSVrv(rv,classname), nv);
10225 =for apidoc sv_setref_pvn
10227 Copies a string into a new SV, optionally blessing the SV. The length of the
10228 string must be specified with C<n>. The C<rv> argument will be upgraded to
10229 an RV. That RV will be modified to point to the new SV. The C<classname>
10230 argument indicates the package for the blessing. Set C<classname> to
10231 C<NULL> to avoid the blessing. The new SV will have a reference count
10232 of 1, and the RV will be returned.
10234 Note that C<sv_setref_pv> copies the pointer while this copies the string.
10240 Perl_sv_setref_pvn(pTHX_ SV *const rv, const char *const classname,
10241 const char *const pv, const STRLEN n)
10243 PERL_ARGS_ASSERT_SV_SETREF_PVN;
10245 sv_setpvn(newSVrv(rv,classname), pv, n);
10250 =for apidoc sv_bless
10252 Blesses an SV into a specified package. The SV must be an RV. The package
10253 must be designated by its stash (see C<L</gv_stashpv>>). The reference count
10254 of the SV is unaffected.
10260 Perl_sv_bless(pTHX_ SV *const sv, HV *const stash)
10263 HV *oldstash = NULL;
10265 PERL_ARGS_ASSERT_SV_BLESS;
10269 Perl_croak(aTHX_ "Can't bless non-reference value");
10271 if (SvFLAGS(tmpRef) & (SVs_OBJECT|SVf_READONLY|SVf_PROTECT)) {
10272 if (SvREADONLY(tmpRef))
10273 Perl_croak_no_modify();
10274 if (SvOBJECT(tmpRef)) {
10275 oldstash = SvSTASH(tmpRef);
10278 SvOBJECT_on(tmpRef);
10279 SvUPGRADE(tmpRef, SVt_PVMG);
10280 SvSTASH_set(tmpRef, MUTABLE_HV(SvREFCNT_inc_simple(stash)));
10281 SvREFCNT_dec(oldstash);
10283 if(SvSMAGICAL(tmpRef))
10284 if(mg_find(tmpRef, PERL_MAGIC_ext) || mg_find(tmpRef, PERL_MAGIC_uvar))
10292 /* Downgrades a PVGV to a PVMG. If it's actually a PVLV, we leave the type
10293 * as it is after unglobbing it.
10296 PERL_STATIC_INLINE void
10297 S_sv_unglob(pTHX_ SV *const sv, U32 flags)
10301 SV * const temp = flags & SV_COW_DROP_PV ? NULL : sv_newmortal();
10303 PERL_ARGS_ASSERT_SV_UNGLOB;
10305 assert(SvTYPE(sv) == SVt_PVGV || SvTYPE(sv) == SVt_PVLV);
10307 if (!(flags & SV_COW_DROP_PV))
10308 gv_efullname3(temp, MUTABLE_GV(sv), "*");
10310 SvREFCNT_inc_simple_void_NN(sv_2mortal(sv));
10312 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
10313 && HvNAME_get(stash))
10314 mro_method_changed_in(stash);
10315 gp_free(MUTABLE_GV(sv));
10318 sv_del_backref(MUTABLE_SV(GvSTASH(sv)), sv);
10319 GvSTASH(sv) = NULL;
10322 if (GvNAME_HEK(sv)) {
10323 unshare_hek(GvNAME_HEK(sv));
10325 isGV_with_GP_off(sv);
10327 if(SvTYPE(sv) == SVt_PVGV) {
10328 /* need to keep SvANY(sv) in the right arena */
10329 xpvmg = new_XPVMG();
10330 StructCopy(SvANY(sv), xpvmg, XPVMG);
10331 del_XPVGV(SvANY(sv));
10334 SvFLAGS(sv) &= ~SVTYPEMASK;
10335 SvFLAGS(sv) |= SVt_PVMG;
10338 /* Intentionally not calling any local SET magic, as this isn't so much a
10339 set operation as merely an internal storage change. */
10340 if (flags & SV_COW_DROP_PV) SvOK_off(sv);
10341 else sv_setsv_flags(sv, temp, 0);
10343 if ((const GV *)sv == PL_last_in_gv)
10344 PL_last_in_gv = NULL;
10345 else if ((const GV *)sv == PL_statgv)
10350 =for apidoc sv_unref_flags
10352 Unsets the RV status of the SV, and decrements the reference count of
10353 whatever was being referenced by the RV. This can almost be thought of
10354 as a reversal of C<newSVrv>. The C<cflags> argument can contain
10355 C<SV_IMMEDIATE_UNREF> to force the reference count to be decremented
10356 (otherwise the decrementing is conditional on the reference count being
10357 different from one or the reference being a readonly SV).
10358 See C<L</SvROK_off>>.
10364 Perl_sv_unref_flags(pTHX_ SV *const ref, const U32 flags)
10366 SV* const target = SvRV(ref);
10368 PERL_ARGS_ASSERT_SV_UNREF_FLAGS;
10370 if (SvWEAKREF(ref)) {
10371 sv_del_backref(target, ref);
10372 SvWEAKREF_off(ref);
10373 SvRV_set(ref, NULL);
10376 SvRV_set(ref, NULL);
10378 /* You can't have a || SvREADONLY(target) here, as $a = $$a, where $a was
10379 assigned to as BEGIN {$a = \"Foo"} will fail. */
10380 if (SvREFCNT(target) != 1 || (flags & SV_IMMEDIATE_UNREF))
10381 SvREFCNT_dec_NN(target);
10382 else /* XXX Hack, but hard to make $a=$a->[1] work otherwise */
10383 sv_2mortal(target); /* Schedule for freeing later */
10387 =for apidoc sv_untaint
10389 Untaint an SV. Use C<SvTAINTED_off> instead.
10395 Perl_sv_untaint(pTHX_ SV *const sv)
10397 PERL_ARGS_ASSERT_SV_UNTAINT;
10398 PERL_UNUSED_CONTEXT;
10400 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10401 MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10408 =for apidoc sv_tainted
10410 Test an SV for taintedness. Use C<SvTAINTED> instead.
10416 Perl_sv_tainted(pTHX_ SV *const sv)
10418 PERL_ARGS_ASSERT_SV_TAINTED;
10419 PERL_UNUSED_CONTEXT;
10421 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10422 const MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10423 if (mg && (mg->mg_len & 1) )
10430 =for apidoc sv_setpviv
10432 Copies an integer into the given SV, also updating its string value.
10433 Does not handle 'set' magic. See C<L</sv_setpviv_mg>>.
10439 Perl_sv_setpviv(pTHX_ SV *const sv, const IV iv)
10441 char buf[TYPE_CHARS(UV)];
10443 char * const ptr = uiv_2buf(buf, iv, 0, 0, &ebuf);
10445 PERL_ARGS_ASSERT_SV_SETPVIV;
10447 sv_setpvn(sv, ptr, ebuf - ptr);
10451 =for apidoc sv_setpviv_mg
10453 Like C<sv_setpviv>, but also handles 'set' magic.
10459 Perl_sv_setpviv_mg(pTHX_ SV *const sv, const IV iv)
10461 PERL_ARGS_ASSERT_SV_SETPVIV_MG;
10463 sv_setpviv(sv, iv);
10467 #if defined(PERL_IMPLICIT_CONTEXT)
10469 /* pTHX_ magic can't cope with varargs, so this is a no-context
10470 * version of the main function, (which may itself be aliased to us).
10471 * Don't access this version directly.
10475 Perl_sv_setpvf_nocontext(SV *const sv, const char *const pat, ...)
10480 PERL_ARGS_ASSERT_SV_SETPVF_NOCONTEXT;
10482 va_start(args, pat);
10483 sv_vsetpvf(sv, pat, &args);
10487 /* pTHX_ magic can't cope with varargs, so this is a no-context
10488 * version of the main function, (which may itself be aliased to us).
10489 * Don't access this version directly.
10493 Perl_sv_setpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10498 PERL_ARGS_ASSERT_SV_SETPVF_MG_NOCONTEXT;
10500 va_start(args, pat);
10501 sv_vsetpvf_mg(sv, pat, &args);
10507 =for apidoc sv_setpvf
10509 Works like C<sv_catpvf> but copies the text into the SV instead of
10510 appending it. Does not handle 'set' magic. See C<L</sv_setpvf_mg>>.
10516 Perl_sv_setpvf(pTHX_ SV *const sv, const char *const pat, ...)
10520 PERL_ARGS_ASSERT_SV_SETPVF;
10522 va_start(args, pat);
10523 sv_vsetpvf(sv, pat, &args);
10528 =for apidoc sv_vsetpvf
10530 Works like C<sv_vcatpvf> but copies the text into the SV instead of
10531 appending it. Does not handle 'set' magic. See C<L</sv_vsetpvf_mg>>.
10533 Usually used via its frontend C<sv_setpvf>.
10539 Perl_sv_vsetpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10541 PERL_ARGS_ASSERT_SV_VSETPVF;
10543 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10547 =for apidoc sv_setpvf_mg
10549 Like C<sv_setpvf>, but also handles 'set' magic.
10555 Perl_sv_setpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10559 PERL_ARGS_ASSERT_SV_SETPVF_MG;
10561 va_start(args, pat);
10562 sv_vsetpvf_mg(sv, pat, &args);
10567 =for apidoc sv_vsetpvf_mg
10569 Like C<sv_vsetpvf>, but also handles 'set' magic.
10571 Usually used via its frontend C<sv_setpvf_mg>.
10577 Perl_sv_vsetpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10579 PERL_ARGS_ASSERT_SV_VSETPVF_MG;
10581 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10585 #if defined(PERL_IMPLICIT_CONTEXT)
10587 /* pTHX_ magic can't cope with varargs, so this is a no-context
10588 * version of the main function, (which may itself be aliased to us).
10589 * Don't access this version directly.
10593 Perl_sv_catpvf_nocontext(SV *const sv, const char *const pat, ...)
10598 PERL_ARGS_ASSERT_SV_CATPVF_NOCONTEXT;
10600 va_start(args, pat);
10601 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10605 /* pTHX_ magic can't cope with varargs, so this is a no-context
10606 * version of the main function, (which may itself be aliased to us).
10607 * Don't access this version directly.
10611 Perl_sv_catpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10616 PERL_ARGS_ASSERT_SV_CATPVF_MG_NOCONTEXT;
10618 va_start(args, pat);
10619 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10626 =for apidoc sv_catpvf
10628 Processes its arguments like C<sv_catpvfn>, and appends the formatted
10629 output to an SV. As with C<sv_catpvfn> called with a non-null C-style
10630 variable argument list, argument reordering is not supported.
10631 If the appended data contains "wide" characters
10632 (including, but not limited to, SVs with a UTF-8 PV formatted with C<%s>,
10633 and characters >255 formatted with C<%c>), the original SV might get
10634 upgraded to UTF-8. Handles 'get' magic, but not 'set' magic. See
10635 C<L</sv_catpvf_mg>>. If the original SV was UTF-8, the pattern should be
10636 valid UTF-8; if the original SV was bytes, the pattern should be too.
10641 Perl_sv_catpvf(pTHX_ SV *const sv, const char *const pat, ...)
10645 PERL_ARGS_ASSERT_SV_CATPVF;
10647 va_start(args, pat);
10648 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10653 =for apidoc sv_vcatpvf
10655 Processes its arguments like C<sv_catpvfn> called with a non-null C-style
10656 variable argument list, and appends the formatted
10657 to an SV. Does not handle 'set' magic. See C<L</sv_vcatpvf_mg>>.
10659 Usually used via its frontend C<sv_catpvf>.
10665 Perl_sv_vcatpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10667 PERL_ARGS_ASSERT_SV_VCATPVF;
10669 sv_vcatpvfn_flags(sv, pat, strlen(pat), args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10673 =for apidoc sv_catpvf_mg
10675 Like C<sv_catpvf>, but also handles 'set' magic.
10681 Perl_sv_catpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10685 PERL_ARGS_ASSERT_SV_CATPVF_MG;
10687 va_start(args, pat);
10688 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10694 =for apidoc sv_vcatpvf_mg
10696 Like C<sv_vcatpvf>, but also handles 'set' magic.
10698 Usually used via its frontend C<sv_catpvf_mg>.
10704 Perl_sv_vcatpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10706 PERL_ARGS_ASSERT_SV_VCATPVF_MG;
10708 sv_vcatpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10713 =for apidoc sv_vsetpvfn
10715 Works like C<sv_vcatpvfn> but copies the text into the SV instead of
10718 Usually used via one of its frontends C<sv_vsetpvf> and C<sv_vsetpvf_mg>.
10724 Perl_sv_vsetpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10725 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
10727 PERL_ARGS_ASSERT_SV_VSETPVFN;
10730 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, 0);
10735 * Warn of missing argument to sprintf. The value used in place of such
10736 * arguments should be &PL_sv_no; an undefined value would yield
10737 * inappropriate "use of uninit" warnings [perl #71000].
10740 S_warn_vcatpvfn_missing_argument(pTHX) {
10741 if (ckWARN(WARN_MISSING)) {
10742 Perl_warner(aTHX_ packWARN(WARN_MISSING), "Missing argument in %s",
10743 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
10749 S_expect_number(pTHX_ char **const pattern)
10753 PERL_ARGS_ASSERT_EXPECT_NUMBER;
10755 switch (**pattern) {
10756 case '1': case '2': case '3':
10757 case '4': case '5': case '6':
10758 case '7': case '8': case '9':
10759 var = *(*pattern)++ - '0';
10760 while (isDIGIT(**pattern)) {
10761 const I32 tmp = var * 10 + (*(*pattern)++ - '0');
10763 Perl_croak(aTHX_ "Integer overflow in format string for %s", (PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn"));
10771 S_F0convert(NV nv, char *const endbuf, STRLEN *const len)
10773 const int neg = nv < 0;
10776 PERL_ARGS_ASSERT_F0CONVERT;
10778 if (UNLIKELY(Perl_isinfnan(nv))) {
10779 STRLEN n = S_infnan_2pv(nv, endbuf - *len, *len, 0);
10789 if (uv & 1 && uv == nv)
10790 uv--; /* Round to even */
10792 const unsigned dig = uv % 10;
10794 } while (uv /= 10);
10805 =for apidoc sv_vcatpvfn
10807 =for apidoc sv_vcatpvfn_flags
10809 Processes its arguments like C<vsprintf> and appends the formatted output
10810 to an SV. Uses an array of SVs if the C-style variable argument list is
10811 missing (C<NULL>). Argument reordering (using format specifiers like C<%2$d>
10812 or C<%*2$d>) is supported only when using an array of SVs; using a C-style
10813 C<va_list> argument list with a format string that uses argument reordering
10814 will yield an exception.
10816 When running with taint checks enabled, indicates via
10817 C<maybe_tainted> if results are untrustworthy (often due to the use of
10820 If called as C<sv_vcatpvfn> or flags has the C<SV_GMAGIC> bit set, calls get magic.
10822 Usually used via one of its frontends C<sv_vcatpvf> and C<sv_vcatpvf_mg>.
10827 #define VECTORIZE_ARGS vecsv = va_arg(*args, SV*);\
10828 vecstr = (U8*)SvPV_const(vecsv,veclen);\
10829 vec_utf8 = DO_UTF8(vecsv);
10831 /* XXX maybe_tainted is never assigned to, so the doc above is lying. */
10834 Perl_sv_vcatpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10835 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
10837 PERL_ARGS_ASSERT_SV_VCATPVFN;
10839 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, SV_GMAGIC|SV_SMAGIC);
10842 #ifdef LONGDOUBLE_DOUBLEDOUBLE
10843 /* The first double can be as large as 2**1023, or '1' x '0' x 1023.
10844 * The second double can be as small as 2**-1074, or '0' x 1073 . '1'.
10845 * The sum of them can be '1' . '0' x 2096 . '1', with implied radix point
10846 * after the first 1023 zero bits.
10848 * XXX The 2098 is quite large (262.25 bytes) and therefore some sort
10849 * of dynamically growing buffer might be better, start at just 16 bytes
10850 * (for example) and grow only when necessary. Or maybe just by looking
10851 * at the exponents of the two doubles? */
10852 # define DOUBLEDOUBLE_MAXBITS 2098
10855 /* vhex will contain the values (0..15) of the hex digits ("nybbles"
10856 * of 4 bits); 1 for the implicit 1, and the mantissa bits, four bits
10857 * per xdigit. For the double-double case, this can be rather many.
10858 * The non-double-double-long-double overshoots since all bits of NV
10859 * are not mantissa bits, there are also exponent bits. */
10860 #ifdef LONGDOUBLE_DOUBLEDOUBLE
10861 # define VHEX_SIZE (3+DOUBLEDOUBLE_MAXBITS/4)
10863 # define VHEX_SIZE (1+(NVSIZE * 8)/4)
10866 /* If we do not have a known long double format, (including not using
10867 * long doubles, or long doubles being equal to doubles) then we will
10868 * fall back to the ldexp/frexp route, with which we can retrieve at
10869 * most as many bits as our widest unsigned integer type is. We try
10870 * to get a 64-bit unsigned integer even if we are not using a 64-bit UV.
10872 * (If you want to test the case of UVSIZE == 4, NVSIZE == 8,
10873 * set the MANTISSATYPE to int and the MANTISSASIZE to 4.)
10875 #if defined(HAS_QUAD) && defined(Uquad_t)
10876 # define MANTISSATYPE Uquad_t
10877 # define MANTISSASIZE 8
10879 # define MANTISSATYPE UV
10880 # define MANTISSASIZE UVSIZE
10883 #if defined(DOUBLE_LITTLE_ENDIAN) || defined(LONGDOUBLE_LITTLE_ENDIAN)
10884 # define HEXTRACT_LITTLE_ENDIAN
10885 #elif defined(DOUBLE_BIG_ENDIAN) || defined(LONGDOUBLE_BIG_ENDIAN)
10886 # define HEXTRACT_BIG_ENDIAN
10888 # define HEXTRACT_MIX_ENDIAN
10891 /* S_hextract() is a helper for Perl_sv_vcatpvfn_flags, for extracting
10892 * the hexadecimal values (for %a/%A). The nv is the NV where the value
10893 * are being extracted from (either directly from the long double in-memory
10894 * presentation, or from the uquad computed via frexp+ldexp). frexp also
10895 * is used to update the exponent. vhex is the pointer to the beginning
10896 * of the output buffer (of VHEX_SIZE).
10898 * The tricky part is that S_hextract() needs to be called twice:
10899 * the first time with vend as NULL, and the second time with vend as
10900 * the pointer returned by the first call. What happens is that on
10901 * the first round the output size is computed, and the intended
10902 * extraction sanity checked. On the second round the actual output
10903 * (the extraction of the hexadecimal values) takes place.
10904 * Sanity failures cause fatal failures during both rounds. */
10906 S_hextract(pTHX_ const NV nv, int* exponent, U8* vhex, U8* vend)
10910 int ixmin = 0, ixmax = 0;
10912 /* XXX Inf/NaN/denormal handling in the HEXTRACT_IMPLICIT_BIT,
10913 * and elsewhere. */
10915 /* These macros are just to reduce typos, they have multiple
10916 * repetitions below, but usually only one (or sometimes two)
10917 * of them is really being used. */
10918 /* HEXTRACT_OUTPUT() extracts the high nybble first. */
10919 #define HEXTRACT_OUTPUT_HI(ix) (*v++ = nvp[ix] >> 4)
10920 #define HEXTRACT_OUTPUT_LO(ix) (*v++ = nvp[ix] & 0xF)
10921 #define HEXTRACT_OUTPUT(ix) \
10923 HEXTRACT_OUTPUT_HI(ix); HEXTRACT_OUTPUT_LO(ix); \
10925 #define HEXTRACT_COUNT(ix, c) \
10927 v += c; if (ix < ixmin) ixmin = ix; else if (ix > ixmax) ixmax = ix; \
10929 #define HEXTRACT_BYTE(ix) \
10931 if (vend) HEXTRACT_OUTPUT(ix); else HEXTRACT_COUNT(ix, 2); \
10933 #define HEXTRACT_LO_NYBBLE(ix) \
10935 if (vend) HEXTRACT_OUTPUT_LO(ix); else HEXTRACT_COUNT(ix, 1); \
10937 /* HEXTRACT_TOP_NYBBLE is just convenience disguise,
10938 * to make it look less odd when the top bits of a NV
10939 * are extracted using HEXTRACT_LO_NYBBLE: the highest
10940 * order bits can be in the "low nybble" of a byte. */
10941 #define HEXTRACT_TOP_NYBBLE(ix) HEXTRACT_LO_NYBBLE(ix)
10942 #define HEXTRACT_BYTES_LE(a, b) \
10943 for (ix = a; ix >= b; ix--) { HEXTRACT_BYTE(ix); }
10944 #define HEXTRACT_BYTES_BE(a, b) \
10945 for (ix = a; ix <= b; ix++) { HEXTRACT_BYTE(ix); }
10946 #define HEXTRACT_IMPLICIT_BIT(nv) \
10948 if (vend) *v++ = ((nv) == 0.0) ? 0 : 1; else v++; \
10951 /* Most formats do. Those which don't should undef this. */
10952 #define HEXTRACT_HAS_IMPLICIT_BIT
10953 /* Many formats do. Those which don't should undef this. */
10954 #define HEXTRACT_HAS_TOP_NYBBLE
10956 /* HEXTRACTSIZE is the maximum number of xdigits. */
10957 #if defined(USE_LONG_DOUBLE) && defined(LONGDOUBLE_DOUBLEDOUBLE)
10958 # define HEXTRACTSIZE (2+DOUBLEDOUBLE_MAXBITS/4)
10960 # define HEXTRACTSIZE 2 * NVSIZE
10963 const U8* vmaxend = vhex + HEXTRACTSIZE;
10964 PERL_UNUSED_VAR(ix); /* might happen */
10965 (void)Perl_frexp(PERL_ABS(nv), exponent);
10966 if (vend && (vend <= vhex || vend > vmaxend)) {
10967 /* diag_listed_as: Hexadecimal float: internal error (%s) */
10968 Perl_croak(aTHX_ "Hexadecimal float: internal error (entry)");
10971 /* First check if using long doubles. */
10972 #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE)
10973 # if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN
10974 /* Used in e.g. VMS and HP-UX IA-64, e.g. -0.1L:
10975 * 9a 99 99 99 99 99 99 99 99 99 99 99 99 99 fb 3f */
10976 /* The bytes 13..0 are the mantissa/fraction,
10977 * the 15,14 are the sign+exponent. */
10978 const U8* nvp = (const U8*)(&nv);
10979 HEXTRACT_IMPLICIT_BIT(nv);
10980 # undef HEXTRACT_HAS_TOP_NYBBLE
10981 HEXTRACT_BYTES_LE(13, 0);
10982 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_BIG_ENDIAN
10983 /* Used in e.g. Solaris Sparc and HP-UX PA-RISC, e.g. -0.1L:
10984 * bf fb 99 99 99 99 99 99 99 99 99 99 99 99 99 9a */
10985 /* The bytes 2..15 are the mantissa/fraction,
10986 * the 0,1 are the sign+exponent. */
10987 const U8* nvp = (const U8*)(&nv);
10988 HEXTRACT_IMPLICIT_BIT(nv);
10989 # undef HEXTRACT_HAS_TOP_NYBBLE
10990 HEXTRACT_BYTES_BE(2, 15);
10991 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN
10992 /* x86 80-bit "extended precision", 64 bits of mantissa / fraction /
10993 * significand, 15 bits of exponent, 1 bit of sign. NVSIZE can
10994 * be either 12 (ILP32, Solaris x86) or 16 (LP64, Linux and OS X),
10995 * meaning that 2 or 6 bytes are empty padding. */
10996 /* The bytes 7..0 are the mantissa/fraction */
10997 const U8* nvp = (const U8*)(&nv);
10998 # undef HEXTRACT_HAS_IMPLICIT_BIT
10999 # undef HEXTRACT_HAS_TOP_NYBBLE
11000 HEXTRACT_BYTES_LE(7, 0);
11001 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN
11002 /* Does this format ever happen? (Wikipedia says the Motorola
11003 * 6888x math coprocessors used format _like_ this but padded
11004 * to 96 bits with 16 unused bits between the exponent and the
11006 const U8* nvp = (const U8*)(&nv);
11007 # undef HEXTRACT_HAS_IMPLICIT_BIT
11008 # undef HEXTRACT_HAS_TOP_NYBBLE
11009 HEXTRACT_BYTES_BE(0, 7);
11011 # define HEXTRACT_FALLBACK
11012 /* Double-double format: two doubles next to each other.
11013 * The first double is the high-order one, exactly like
11014 * it would be for a "lone" double. The second double
11015 * is shifted down using the exponent so that that there
11016 * are no common bits. The tricky part is that the value
11017 * of the double-double is the SUM of the two doubles and
11018 * the second one can be also NEGATIVE.
11020 * Because of this tricky construction the bytewise extraction we
11021 * use for the other long double formats doesn't work, we must
11022 * extract the values bit by bit.
11024 * The little-endian double-double is used .. somewhere?
11026 * The big endian double-double is used in e.g. PPC/Power (AIX)
11029 * The mantissa bits are in two separate stretches, e.g. for -0.1L:
11030 * 9a 99 99 99 99 99 59 bc 9a 99 99 99 99 99 b9 3f (LE)
11031 * 3f b9 99 99 99 99 99 9a bc 59 99 99 99 99 99 9a (BE)
11034 #else /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) */
11035 /* Using normal doubles, not long doubles.
11037 * We generate 4-bit xdigits (nybble/nibble) instead of 8-bit
11038 * bytes, since we might need to handle printf precision, and
11039 * also need to insert the radix. */
11041 # ifdef HEXTRACT_LITTLE_ENDIAN
11042 /* 0 1 2 3 4 5 6 7 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11043 const U8* nvp = (const U8*)(&nv);
11044 HEXTRACT_IMPLICIT_BIT(nv);
11045 HEXTRACT_TOP_NYBBLE(6);
11046 HEXTRACT_BYTES_LE(5, 0);
11047 # elif defined(HEXTRACT_BIG_ENDIAN)
11048 /* 7 6 5 4 3 2 1 0 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11049 const U8* nvp = (const U8*)(&nv);
11050 HEXTRACT_IMPLICIT_BIT(nv);
11051 HEXTRACT_TOP_NYBBLE(1);
11052 HEXTRACT_BYTES_BE(2, 7);
11053 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_LE_BE
11054 /* 4 5 6 7 0 1 2 3 (MSB = 7, LSB = 0, 6:7 = nybble:exponent:sign) */
11055 const U8* nvp = (const U8*)(&nv);
11056 HEXTRACT_IMPLICIT_BIT(nv);
11057 HEXTRACT_TOP_NYBBLE(2); /* 6 */
11058 HEXTRACT_BYTE(1); /* 5 */
11059 HEXTRACT_BYTE(0); /* 4 */
11060 HEXTRACT_BYTE(7); /* 3 */
11061 HEXTRACT_BYTE(6); /* 2 */
11062 HEXTRACT_BYTE(5); /* 1 */
11063 HEXTRACT_BYTE(4); /* 0 */
11064 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_BE_LE
11065 /* 3 2 1 0 7 6 5 4 (MSB = 7, LSB = 0, 7:6 = sign:exponent:nybble) */
11066 const U8* nvp = (const U8*)(&nv);
11067 HEXTRACT_IMPLICIT_BIT(nv);
11068 HEXTRACT_TOP_NYBBLE(5); /* 6 */
11069 HEXTRACT_BYTE(6); /* 5 */
11070 HEXTRACT_BYTE(7); /* 4 */
11071 HEXTRACT_BYTE(0); /* 3 */
11072 HEXTRACT_BYTE(1); /* 2 */
11073 HEXTRACT_BYTE(2); /* 1 */
11074 HEXTRACT_BYTE(3); /* 0 */
11076 # define HEXTRACT_FALLBACK
11079 # define HEXTRACT_FALLBACK
11081 #endif /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) #else */
11082 # ifdef HEXTRACT_FALLBACK
11083 # undef HEXTRACT_HAS_TOP_NYBBLE /* Meaningless, but consistent. */
11084 /* The fallback is used for the double-double format, and
11085 * for unknown long double formats, and for unknown double
11086 * formats, or in general unknown NV formats. */
11087 if (nv == (NV)0.0) {
11095 NV d = nv < 0 ? -nv : nv;
11097 U8 ha = 0x0; /* hexvalue accumulator */
11098 U8 hd = 0x8; /* hexvalue digit */
11100 /* Shift d and e (and update exponent) so that e <= d < 2*e,
11101 * this is essentially manual frexp(). Multiplying by 0.5 and
11102 * doubling should be lossless in binary floating point. */
11112 while (d >= e + e) {
11116 /* Now e <= d < 2*e */
11118 /* First extract the leading hexdigit (the implicit bit). */
11134 /* Then extract the remaining hexdigits. */
11135 while (d > (NV)0.0) {
11141 /* Output or count in groups of four bits,
11142 * that is, when the hexdigit is down to one. */
11147 /* Reset the hexvalue. */
11156 /* Flush possible pending hexvalue. */
11166 /* Croak for various reasons: if the output pointer escaped the
11167 * output buffer, if the extraction index escaped the extraction
11168 * buffer, or if the ending output pointer didn't match the
11169 * previously computed value. */
11170 if (v <= vhex || v - vhex >= VHEX_SIZE ||
11171 /* For double-double the ixmin and ixmax stay at zero,
11172 * which is convenient since the HEXTRACTSIZE is tricky
11173 * for double-double. */
11174 ixmin < 0 || ixmax >= NVSIZE ||
11175 (vend && v != vend)) {
11176 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11177 Perl_croak(aTHX_ "Hexadecimal float: internal error (overflow)");
11182 /* Helper for sv_vcatpvfn_flags(). */
11183 #define FETCH_VCATPVFN_ARGUMENT(var, in_range, expr) \
11188 (var) = &PL_sv_no; /* [perl #71000] */ \
11189 arg_missing = TRUE; \
11194 Perl_sv_vcatpvfn_flags(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11195 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted,
11200 const char *patend;
11203 static const char nullstr[] = "(null)";
11205 bool has_utf8 = DO_UTF8(sv); /* has the result utf8? */
11206 const bool pat_utf8 = has_utf8; /* the pattern is in utf8? */
11208 /* Times 4: a decimal digit takes more than 3 binary digits.
11209 * NV_DIG: mantissa takes than many decimal digits.
11210 * Plus 32: Playing safe. */
11211 char ebuf[IV_DIG * 4 + NV_DIG + 32];
11212 bool no_redundant_warning = FALSE; /* did we use any explicit format parameter index? */
11213 bool hexfp = FALSE; /* hexadecimal floating point? */
11215 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
11217 PERL_ARGS_ASSERT_SV_VCATPVFN_FLAGS;
11218 PERL_UNUSED_ARG(maybe_tainted);
11220 if (flags & SV_GMAGIC)
11223 /* no matter what, this is a string now */
11224 (void)SvPV_force_nomg(sv, origlen);
11226 /* special-case "", "%s", and "%-p" (SVf - see below) */
11228 if (svmax && ckWARN(WARN_REDUNDANT))
11229 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11230 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11233 if (patlen == 2 && pat[0] == '%' && pat[1] == 's') {
11234 if (svmax > 1 && ckWARN(WARN_REDUNDANT))
11235 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11236 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11239 const char * const s = va_arg(*args, char*);
11240 sv_catpv_nomg(sv, s ? s : nullstr);
11242 else if (svix < svmax) {
11243 /* we want get magic on the source but not the target. sv_catsv can't do that, though */
11244 SvGETMAGIC(*svargs);
11245 sv_catsv_nomg(sv, *svargs);
11248 S_warn_vcatpvfn_missing_argument(aTHX);
11251 if (args && patlen == 3 && pat[0] == '%' &&
11252 pat[1] == '-' && pat[2] == 'p') {
11253 if (svmax > 1 && ckWARN(WARN_REDUNDANT))
11254 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11255 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11256 argsv = MUTABLE_SV(va_arg(*args, void*));
11257 sv_catsv_nomg(sv, argsv);
11261 #if !defined(USE_LONG_DOUBLE) && !defined(USE_QUADMATH)
11262 /* special-case "%.<number>[gf]" */
11263 if ( !args && patlen <= 5 && pat[0] == '%' && pat[1] == '.'
11264 && (pat[patlen-1] == 'g' || pat[patlen-1] == 'f') ) {
11265 unsigned digits = 0;
11269 while (*pp >= '0' && *pp <= '9')
11270 digits = 10 * digits + (*pp++ - '0');
11272 /* XXX: Why do this `svix < svmax` test? Couldn't we just
11273 format the first argument and WARN_REDUNDANT if svmax > 1?
11274 Munged by Nicholas Clark in v5.13.0-209-g95ea86d */
11275 if (pp - pat == (int)patlen - 1 && svix < svmax) {
11276 const NV nv = SvNV(*svargs);
11277 if (LIKELY(!Perl_isinfnan(nv))) {
11279 /* Add check for digits != 0 because it seems that some
11280 gconverts are buggy in this case, and we don't yet have
11281 a Configure test for this. */
11282 if (digits && digits < sizeof(ebuf) - NV_DIG - 10) {
11283 /* 0, point, slack */
11284 STORE_LC_NUMERIC_SET_TO_NEEDED();
11285 SNPRINTF_G(nv, ebuf, size, digits);
11286 sv_catpv_nomg(sv, ebuf);
11287 if (*ebuf) /* May return an empty string for digits==0 */
11290 } else if (!digits) {
11293 if ((p = F0convert(nv, ebuf + sizeof ebuf, &l))) {
11294 sv_catpvn_nomg(sv, p, l);
11301 #endif /* !USE_LONG_DOUBLE */
11303 if (!args && svix < svmax && DO_UTF8(*svargs))
11306 patend = (char*)pat + patlen;
11307 for (p = (char*)pat; p < patend; p = q) {
11310 bool vectorize = FALSE;
11311 bool vectorarg = FALSE;
11312 bool vec_utf8 = FALSE;
11318 bool has_precis = FALSE;
11320 const I32 osvix = svix;
11321 bool is_utf8 = FALSE; /* is this item utf8? */
11322 bool used_explicit_ix = FALSE;
11323 bool arg_missing = FALSE;
11324 #ifdef HAS_LDBL_SPRINTF_BUG
11325 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
11326 with sfio - Allen <allens@cpan.org> */
11327 bool fix_ldbl_sprintf_bug = FALSE;
11331 U8 utf8buf[UTF8_MAXBYTES+1];
11332 STRLEN esignlen = 0;
11334 const char *eptr = NULL;
11335 const char *fmtstart;
11338 const U8 *vecstr = NULL;
11345 /* We need a long double target in case HAS_LONG_DOUBLE,
11346 * even without USE_LONG_DOUBLE, so that we can printf with
11347 * long double formats, even without NV being long double.
11348 * But we call the target 'fv' instead of 'nv', since most of
11349 * the time it is not (most compilers these days recognize
11350 * "long double", even if only as a synonym for "double").
11352 #if defined(HAS_LONG_DOUBLE) && LONG_DOUBLESIZE > DOUBLESIZE && \
11353 defined(PERL_PRIgldbl) && !defined(USE_QUADMATH)
11355 # ifdef Perl_isfinitel
11356 # define FV_ISFINITE(x) Perl_isfinitel(x)
11358 # define FV_GF PERL_PRIgldbl
11359 # if defined(__VMS) && defined(__ia64) && defined(__IEEE_FLOAT)
11360 /* Work around breakage in OTS$CVT_FLOAT_T_X */
11361 # define NV_TO_FV(nv,fv) STMT_START { \
11363 fv = Perl_isnan(_dv) ? LDBL_QNAN : _dv; \
11366 # define NV_TO_FV(nv,fv) (fv)=(nv)
11370 # define FV_GF NVgf
11371 # define NV_TO_FV(nv,fv) (fv)=(nv)
11373 #ifndef FV_ISFINITE
11374 # define FV_ISFINITE(x) Perl_isfinite((NV)(x))
11380 const char *dotstr = ".";
11381 STRLEN dotstrlen = 1;
11382 I32 efix = 0; /* explicit format parameter index */
11383 I32 ewix = 0; /* explicit width index */
11384 I32 epix = 0; /* explicit precision index */
11385 I32 evix = 0; /* explicit vector index */
11386 bool asterisk = FALSE;
11387 bool infnan = FALSE;
11389 /* echo everything up to the next format specification */
11390 for (q = p; q < patend && *q != '%'; ++q) ;
11392 if (has_utf8 && !pat_utf8)
11393 sv_catpvn_nomg_utf8_upgrade(sv, p, q - p, nsv);
11395 sv_catpvn_nomg(sv, p, q - p);
11404 We allow format specification elements in this order:
11405 \d+\$ explicit format parameter index
11407 v|\*(\d+\$)?v vector with optional (optionally specified) arg
11408 0 flag (as above): repeated to allow "v02"
11409 \d+|\*(\d+\$)? width using optional (optionally specified) arg
11410 \.(\d*|\*(\d+\$)?) precision using optional (optionally specified) arg
11412 [%bcdefginopsuxDFOUX] format (mandatory)
11417 As of perl5.9.3, printf format checking is on by default.
11418 Internally, perl uses %p formats to provide an escape to
11419 some extended formatting. This block deals with those
11420 extensions: if it does not match, (char*)q is reset and
11421 the normal format processing code is used.
11423 Currently defined extensions are:
11424 %p include pointer address (standard)
11425 %-p (SVf) include an SV (previously %_)
11426 %-<num>p include an SV with precision <num>
11428 %3p include a HEK with precision of 256
11429 %4p char* preceded by utf8 flag and length
11430 %<num>p (where num is 1 or > 4) reserved for future
11433 Robin Barker 2005-07-14 (but modified since)
11435 %1p (VDf) removed. RMB 2007-10-19
11442 else if (strnEQ(q, UTF8f, sizeof(UTF8f)-1)) { /* UTF8f */
11443 /* The argument has already gone through cBOOL, so the cast
11445 is_utf8 = (bool)va_arg(*args, int);
11446 elen = va_arg(*args, UV);
11447 if ((IV)elen < 0) {
11448 /* check if utf8 length is larger than 0 when cast to IV */
11449 assert( (IV)elen >= 0 ); /* in DEBUGGING build we want to crash */
11450 elen= 0; /* otherwise we want to treat this as an empty string */
11452 eptr = va_arg(*args, char *);
11453 q += sizeof(UTF8f)-1;
11456 n = expect_number(&q);
11458 if (sv) { /* SVf */
11463 argsv = MUTABLE_SV(va_arg(*args, void*));
11464 eptr = SvPV_const(argsv, elen);
11465 if (DO_UTF8(argsv))
11469 else if (n==2 || n==3) { /* HEKf */
11470 HEK * const hek = va_arg(*args, HEK *);
11471 eptr = HEK_KEY(hek);
11472 elen = HEK_LEN(hek);
11473 if (HEK_UTF8(hek)) is_utf8 = TRUE;
11474 if (n==3) precis = 256, has_precis = TRUE;
11478 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
11479 "internal %%<num>p might conflict with future printf extensions");
11485 if ( (width = expect_number(&q)) ) {
11488 Perl_croak_nocontext(
11489 "Cannot yet reorder sv_catpvfn() arguments from va_list");
11492 used_explicit_ix = TRUE;
11504 if (plus == '+' && *q == ' ') /* '+' over ' ' */
11533 if ( (ewix = expect_number(&q)) ) {
11536 Perl_croak_nocontext(
11537 "Cannot yet reorder sv_catpvfn() arguments from va_list");
11538 used_explicit_ix = TRUE;
11548 if ((vectorarg = asterisk)) {
11561 width = expect_number(&q);
11564 if (vectorize && vectorarg) {
11565 /* vectorizing, but not with the default "." */
11567 vecsv = va_arg(*args, SV*);
11569 FETCH_VCATPVFN_ARGUMENT(
11570 vecsv, evix > 0 && evix <= svmax, svargs[evix-1]);
11572 FETCH_VCATPVFN_ARGUMENT(
11573 vecsv, svix < svmax, svargs[svix++]);
11575 dotstr = SvPV_const(vecsv, dotstrlen);
11576 /* Keep the DO_UTF8 test *after* the SvPV call, else things go
11577 bad with tied or overloaded values that return UTF8. */
11578 if (DO_UTF8(vecsv))
11580 else if (has_utf8) {
11581 vecsv = sv_mortalcopy(vecsv);
11582 sv_utf8_upgrade(vecsv);
11583 dotstr = SvPV_const(vecsv, dotstrlen);
11590 i = va_arg(*args, int);
11592 i = (ewix ? ewix <= svmax : svix < svmax) ?
11593 SvIVx(svargs[ewix ? ewix-1 : svix++]) : 0;
11595 width = (i < 0) ? -i : i;
11605 if ( (epix = expect_number(&q)) ) {
11608 Perl_croak_nocontext(
11609 "Cannot yet reorder sv_catpvfn() arguments from va_list");
11610 used_explicit_ix = TRUE;
11615 i = va_arg(*args, int);
11619 FETCH_VCATPVFN_ARGUMENT(
11620 precsv, epix > 0 && epix <= svmax, svargs[epix-1]);
11622 FETCH_VCATPVFN_ARGUMENT(
11623 precsv, svix < svmax, svargs[svix++]);
11624 i = precsv == &PL_sv_no ? 0 : SvIVx(precsv);
11627 has_precis = !(i < 0);
11631 while (isDIGIT(*q))
11632 precis = precis * 10 + (*q++ - '0');
11641 else if (efix ? (efix > 0 && efix <= svmax) : svix < svmax) {
11642 vecsv = svargs[efix ? efix-1 : svix++];
11643 vecstr = (U8*)SvPV_const(vecsv,veclen);
11644 vec_utf8 = DO_UTF8(vecsv);
11646 /* if this is a version object, we need to convert
11647 * back into v-string notation and then let the
11648 * vectorize happen normally
11650 if (sv_isobject(vecsv) && sv_derived_from(vecsv, "version")) {
11651 if ( hv_exists(MUTABLE_HV(SvRV(vecsv)), "alpha", 5 ) ) {
11652 Perl_ck_warner_d(aTHX_ packWARN(WARN_PRINTF),
11653 "vector argument not supported with alpha versions");
11656 vecsv = sv_newmortal();
11657 scan_vstring((char *)vecstr, (char *)vecstr + veclen,
11659 vecstr = (U8*)SvPV_const(vecsv, veclen);
11660 vec_utf8 = DO_UTF8(vecsv);
11674 case 'I': /* Ix, I32x, and I64x */
11675 # ifdef USE_64_BIT_INT
11676 if (q[1] == '6' && q[2] == '4') {
11682 if (q[1] == '3' && q[2] == '2') {
11686 # ifdef USE_64_BIT_INT
11692 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
11693 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
11696 # ifdef USE_QUADMATH
11709 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
11710 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
11711 if (*q == 'l') { /* lld, llf */
11720 if (*++q == 'h') { /* hhd, hhu */
11749 if (!vectorize && !args) {
11751 const I32 i = efix-1;
11752 FETCH_VCATPVFN_ARGUMENT(argsv, i >= 0 && i < svmax, svargs[i]);
11754 FETCH_VCATPVFN_ARGUMENT(argsv, svix >= 0 && svix < svmax,
11759 if (argsv && strchr("BbcDdiOopuUXx",*q)) {
11760 /* XXX va_arg(*args) case? need peek, use va_copy? */
11762 if (UNLIKELY(SvAMAGIC(argsv)))
11763 argsv = sv_2num(argsv);
11764 infnan = UNLIKELY(isinfnansv(argsv));
11767 switch (c = *q++) {
11775 Perl_croak(aTHX_ "Cannot printf %"NVgf" with '%c'",
11776 /* no va_arg() case */
11777 SvNV_nomg(argsv), (int)c);
11778 uv = (args) ? va_arg(*args, int) : SvIV_nomg(argsv);
11780 (!UVCHR_IS_INVARIANT(uv) && SvUTF8(sv)))
11782 eptr = (char*)utf8buf;
11783 elen = uvchr_to_utf8((U8*)eptr, uv) - utf8buf;
11797 eptr = va_arg(*args, char*);
11799 elen = strlen(eptr);
11801 eptr = (char *)nullstr;
11802 elen = sizeof nullstr - 1;
11806 eptr = SvPV_const(argsv, elen);
11807 if (DO_UTF8(argsv)) {
11808 STRLEN old_precis = precis;
11809 if (has_precis && precis < elen) {
11810 STRLEN ulen = sv_or_pv_len_utf8(argsv, eptr, elen);
11811 STRLEN p = precis > ulen ? ulen : precis;
11812 precis = sv_or_pv_pos_u2b(argsv, eptr, p, 0);
11813 /* sticks at end */
11815 if (width) { /* fudge width (can't fudge elen) */
11816 if (has_precis && precis < elen)
11817 width += precis - old_precis;
11820 elen - sv_or_pv_len_utf8(argsv,eptr,elen);
11827 if (has_precis && precis < elen)
11835 goto floating_point;
11837 if (alt || vectorize)
11839 uv = PTR2UV(args ? va_arg(*args, void*) : argsv);
11853 goto floating_point;
11858 goto donevalidconversion;
11860 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
11869 esignbuf[esignlen++] = plus;
11873 case 'c': iv = (char)va_arg(*args, int); break;
11874 case 'h': iv = (short)va_arg(*args, int); break;
11875 case 'l': iv = va_arg(*args, long); break;
11876 case 'V': iv = va_arg(*args, IV); break;
11877 case 'z': iv = va_arg(*args, SSize_t); break;
11878 #ifdef HAS_PTRDIFF_T
11879 case 't': iv = va_arg(*args, ptrdiff_t); break;
11881 default: iv = va_arg(*args, int); break;
11883 case 'j': iv = va_arg(*args, intmax_t); break;
11887 iv = va_arg(*args, Quad_t); break;
11894 IV tiv = SvIV_nomg(argsv); /* work around GCC bug #13488 */
11896 case 'c': iv = (char)tiv; break;
11897 case 'h': iv = (short)tiv; break;
11898 case 'l': iv = (long)tiv; break;
11900 default: iv = tiv; break;
11903 iv = (Quad_t)tiv; break;
11909 if ( !vectorize ) /* we already set uv above */
11914 esignbuf[esignlen++] = plus;
11917 uv = (iv == IV_MIN) ? (UV)iv : (UV)(-iv);
11918 esignbuf[esignlen++] = '-';
11957 goto floating_point;
11963 goto donevalidconversion;
11965 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
11976 case 'c': uv = (unsigned char)va_arg(*args, unsigned); break;
11977 case 'h': uv = (unsigned short)va_arg(*args, unsigned); break;
11978 case 'l': uv = va_arg(*args, unsigned long); break;
11979 case 'V': uv = va_arg(*args, UV); break;
11980 case 'z': uv = va_arg(*args, Size_t); break;
11981 #ifdef HAS_PTRDIFF_T
11982 case 't': uv = va_arg(*args, ptrdiff_t); break; /* will sign extend, but there is no uptrdiff_t, so oh well */
11985 case 'j': uv = va_arg(*args, uintmax_t); break;
11987 default: uv = va_arg(*args, unsigned); break;
11990 uv = va_arg(*args, Uquad_t); break;
11997 UV tuv = SvUV_nomg(argsv); /* work around GCC bug #13488 */
11999 case 'c': uv = (unsigned char)tuv; break;
12000 case 'h': uv = (unsigned short)tuv; break;
12001 case 'l': uv = (unsigned long)tuv; break;
12003 default: uv = tuv; break;
12006 uv = (Uquad_t)tuv; break;
12015 char *ptr = ebuf + sizeof ebuf;
12016 bool tempalt = uv ? alt : FALSE; /* Vectors can't change alt */
12022 p = (char *)((c == 'X') ? PL_hexdigit + 16 : PL_hexdigit);
12026 } while (uv >>= 4);
12028 esignbuf[esignlen++] = '0';
12029 esignbuf[esignlen++] = c; /* 'x' or 'X' */
12035 *--ptr = '0' + dig;
12036 } while (uv >>= 3);
12037 if (alt && *ptr != '0')
12043 *--ptr = '0' + dig;
12044 } while (uv >>= 1);
12046 esignbuf[esignlen++] = '0';
12047 esignbuf[esignlen++] = c;
12050 default: /* it had better be ten or less */
12053 *--ptr = '0' + dig;
12054 } while (uv /= base);
12057 elen = (ebuf + sizeof ebuf) - ptr;
12061 zeros = precis - elen;
12062 else if (precis == 0 && elen == 1 && *eptr == '0'
12063 && !(base == 8 && alt)) /* "%#.0o" prints "0" */
12066 /* a precision nullifies the 0 flag. */
12073 /* FLOATING POINT */
12078 c = 'f'; /* maybe %F isn't supported here */
12080 case 'e': case 'E':
12082 case 'g': case 'G':
12083 case 'a': case 'A':
12087 /* This is evil, but floating point is even more evil */
12089 /* for SV-style calling, we can only get NV
12090 for C-style calling, we assume %f is double;
12091 for simplicity we allow any of %Lf, %llf, %qf for long double
12095 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12099 /* [perl #20339] - we should accept and ignore %lf rather than die */
12103 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12104 intsize = args ? 0 : 'q';
12108 #if defined(HAS_LONG_DOUBLE)
12121 /* Now we need (long double) if intsize == 'q', else (double). */
12123 /* Note: do not pull NVs off the va_list with va_arg()
12124 * (pull doubles instead) because if you have a build
12125 * with long doubles, you would always be pulling long
12126 * doubles, which would badly break anyone using only
12127 * doubles (i.e. the majority of builds). In other
12128 * words, you cannot mix doubles and long doubles.
12129 * The only case where you can pull off long doubles
12130 * is when the format specifier explicitly asks so with
12132 #ifdef USE_QUADMATH
12133 fv = intsize == 'q' ?
12134 va_arg(*args, NV) : va_arg(*args, double);
12136 #elif LONG_DOUBLESIZE > DOUBLESIZE
12137 if (intsize == 'q') {
12138 fv = va_arg(*args, long double);
12141 nv = va_arg(*args, double);
12145 nv = va_arg(*args, double);
12151 if (!infnan) SvGETMAGIC(argsv);
12152 nv = SvNV_nomg(argsv);
12157 /* frexp() (or frexpl) has some unspecified behaviour for
12158 * nan/inf/-inf, so let's avoid calling that on non-finites. */
12159 if (isALPHA_FOLD_NE(c, 'e') && FV_ISFINITE(fv)) {
12161 (void)Perl_frexp((NV)fv, &i);
12162 if (i == PERL_INT_MIN)
12163 Perl_die(aTHX_ "panic: frexp: %"FV_GF, fv);
12164 /* Do not set hexfp earlier since we want to printf
12165 * Inf/NaN for Inf/NaN, not their hexfp. */
12166 hexfp = isALPHA_FOLD_EQ(c, 'a');
12167 if (UNLIKELY(hexfp)) {
12168 /* This seriously overshoots in most cases, but
12169 * better the undershooting. Firstly, all bytes
12170 * of the NV are not mantissa, some of them are
12171 * exponent. Secondly, for the reasonably common
12172 * long doubles case, the "80-bit extended", two
12173 * or six bytes of the NV are unused. */
12175 (fv < 0) ? 1 : 0 + /* possible unary minus */
12177 1 + /* the very unlikely carry */
12180 2 * NVSIZE + /* 2 hexdigits for each byte */
12182 6 + /* exponent: sign, plus up to 16383 (quad fp) */
12184 #ifdef LONGDOUBLE_DOUBLEDOUBLE
12185 /* However, for the "double double", we need more.
12186 * Since each double has their own exponent, the
12187 * doubles may float (haha) rather far from each
12188 * other, and the number of required bits is much
12189 * larger, up to total of DOUBLEDOUBLE_MAXBITS bits.
12190 * See the definition of DOUBLEDOUBLE_MAXBITS.
12192 * Need 2 hexdigits for each byte. */
12193 need += (DOUBLEDOUBLE_MAXBITS/8 + 1) * 2;
12194 /* the size for the exponent already added */
12196 #ifdef USE_LOCALE_NUMERIC
12197 STORE_LC_NUMERIC_SET_TO_NEEDED();
12198 if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC))
12199 need += SvLEN(PL_numeric_radix_sv);
12200 RESTORE_LC_NUMERIC();
12204 need = BIT_DIGITS(i);
12205 } /* if i < 0, the number of digits is hard to predict. */
12207 need += has_precis ? precis : 6; /* known default */
12212 #ifdef HAS_LDBL_SPRINTF_BUG
12213 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
12214 with sfio - Allen <allens@cpan.org> */
12217 # define MY_DBL_MAX DBL_MAX
12218 # else /* XXX guessing! HUGE_VAL may be defined as infinity, so not using */
12219 # if DOUBLESIZE >= 8
12220 # define MY_DBL_MAX 1.7976931348623157E+308L
12222 # define MY_DBL_MAX 3.40282347E+38L
12226 # ifdef HAS_LDBL_SPRINTF_BUG_LESS1 /* only between -1L & 1L - Allen */
12227 # define MY_DBL_MAX_BUG 1L
12229 # define MY_DBL_MAX_BUG MY_DBL_MAX
12233 # define MY_DBL_MIN DBL_MIN
12234 # else /* XXX guessing! -Allen */
12235 # if DOUBLESIZE >= 8
12236 # define MY_DBL_MIN 2.2250738585072014E-308L
12238 # define MY_DBL_MIN 1.17549435E-38L
12242 if ((intsize == 'q') && (c == 'f') &&
12243 ((fv < MY_DBL_MAX_BUG) && (fv > -MY_DBL_MAX_BUG)) &&
12244 (need < DBL_DIG)) {
12245 /* it's going to be short enough that
12246 * long double precision is not needed */
12248 if ((fv <= 0L) && (fv >= -0L))
12249 fix_ldbl_sprintf_bug = TRUE; /* 0 is 0 - easiest */
12251 /* would use Perl_fp_class as a double-check but not
12252 * functional on IRIX - see perl.h comments */
12254 if ((fv >= MY_DBL_MIN) || (fv <= -MY_DBL_MIN)) {
12255 /* It's within the range that a double can represent */
12256 #if defined(DBL_MAX) && !defined(DBL_MIN)
12257 if ((fv >= ((long double)1/DBL_MAX)) ||
12258 (fv <= (-(long double)1/DBL_MAX)))
12260 fix_ldbl_sprintf_bug = TRUE;
12263 if (fix_ldbl_sprintf_bug == TRUE) {
12273 # undef MY_DBL_MAX_BUG
12276 #endif /* HAS_LDBL_SPRINTF_BUG */
12278 need += 20; /* fudge factor */
12279 if (PL_efloatsize < need) {
12280 Safefree(PL_efloatbuf);
12281 PL_efloatsize = need + 20; /* more fudge */
12282 Newx(PL_efloatbuf, PL_efloatsize, char);
12283 PL_efloatbuf[0] = '\0';
12286 if ( !(width || left || plus || alt) && fill != '0'
12287 && has_precis && intsize != 'q' /* Shortcuts */
12288 && LIKELY(!Perl_isinfnan((NV)fv)) ) {
12289 /* See earlier comment about buggy Gconvert when digits,
12291 if ( c == 'g' && precis ) {
12292 STORE_LC_NUMERIC_SET_TO_NEEDED();
12293 SNPRINTF_G(fv, PL_efloatbuf, PL_efloatsize, precis);
12294 /* May return an empty string for digits==0 */
12295 if (*PL_efloatbuf) {
12296 elen = strlen(PL_efloatbuf);
12297 goto float_converted;
12299 } else if ( c == 'f' && !precis ) {
12300 if ((eptr = F0convert(nv, ebuf + sizeof ebuf, &elen)))
12305 if (UNLIKELY(hexfp)) {
12306 /* Hexadecimal floating point. */
12307 char* p = PL_efloatbuf;
12308 U8 vhex[VHEX_SIZE];
12309 U8* v = vhex; /* working pointer to vhex */
12310 U8* vend; /* pointer to one beyond last digit of vhex */
12311 U8* vfnz = NULL; /* first non-zero */
12312 const bool lower = (c == 'a');
12313 /* At output the values of vhex (up to vend) will
12314 * be mapped through the xdig to get the actual
12315 * human-readable xdigits. */
12316 const char* xdig = PL_hexdigit;
12317 int zerotail = 0; /* how many extra zeros to append */
12318 int exponent = 0; /* exponent of the floating point input */
12320 /* XXX: denormals, NaN, Inf.
12322 * For example with denormals, (assuming the vanilla
12323 * 64-bit double): the exponent is zero. 1xp-1074 is
12324 * the smallest denormal and the smallest double, it
12325 * should be output as 0x0.0000000000001p-1022 to
12326 * match its internal structure. */
12328 vend = S_hextract(aTHX_ nv, &exponent, vhex, NULL);
12329 S_hextract(aTHX_ nv, &exponent, vhex, vend);
12331 #if NVSIZE > DOUBLESIZE
12332 # ifdef HEXTRACT_HAS_IMPLICIT_BIT
12333 /* In this case there is an implicit bit,
12334 * and therefore the exponent is shifted shift by one. */
12337 /* In this case there is no implicit bit,
12338 * and the exponent is shifted by the first xdigit. */
12353 xdig += 16; /* Use uppercase hex. */
12356 /* Find the first non-zero xdigit. */
12357 for (v = vhex; v < vend; v++) {
12365 U8* vlnz = NULL; /* The last non-zero. */
12367 /* Find the last non-zero xdigit. */
12368 for (v = vend - 1; v >= vhex; v--) {
12375 #if NVSIZE == DOUBLESIZE
12381 if ((SSize_t)(precis + 1) < vend - vhex) {
12384 v = vhex + precis + 1;
12385 /* Round away from zero: if the tail
12386 * beyond the precis xdigits is equal to
12387 * or greater than 0x8000... */
12389 if (!round && *v == 0x8) {
12390 for (v++; v < vend; v++) {
12398 for (v = vhex + precis; v >= vhex; v--) {
12405 /* If the carry goes all the way to
12406 * the front, we need to output
12407 * a single '1'. This goes against
12408 * the "xdigit and then radix"
12409 * but since this is "cannot happen"
12410 * category, that is probably good. */
12415 /* The new effective "last non zero". */
12416 vlnz = vhex + precis;
12419 zerotail = precis - (vlnz - vhex);
12426 /* The radix is always output after the first
12427 * non-zero xdigit, or if alt. */
12428 if (vfnz < vlnz || alt) {
12429 #ifndef USE_LOCALE_NUMERIC
12432 STORE_LC_NUMERIC_SET_TO_NEEDED();
12433 if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC)) {
12435 const char* r = SvPV(PL_numeric_radix_sv, n);
12436 Copy(r, p, n, char);
12442 RESTORE_LC_NUMERIC();
12457 elen = p - PL_efloatbuf;
12458 elen += my_snprintf(p, PL_efloatsize - elen,
12459 "%c%+d", lower ? 'p' : 'P',
12462 if (elen < width) {
12464 /* Pad the back with spaces. */
12465 memset(PL_efloatbuf + elen, ' ', width - elen);
12467 else if (fill == '0') {
12468 /* Insert the zeros between the "0x" and
12469 * the digits, otherwise we end up with
12471 STRLEN nzero = width - elen;
12472 char* zerox = PL_efloatbuf + 2;
12473 Move(zerox, zerox + nzero, elen - 2, char);
12474 memset(zerox, fill, nzero);
12477 /* Move it to the right. */
12478 Move(PL_efloatbuf, PL_efloatbuf + width - elen,
12480 /* Pad the front with spaces. */
12481 memset(PL_efloatbuf, ' ', width - elen);
12487 elen = S_infnan_2pv(nv, PL_efloatbuf, PL_efloatsize, plus);
12489 /* Not affecting infnan output: precision, alt, fill. */
12490 if (elen < width) {
12492 /* Pack the back with spaces. */
12493 memset(PL_efloatbuf + elen, ' ', width - elen);
12495 /* Move it to the right. */
12496 Move(PL_efloatbuf, PL_efloatbuf + width - elen,
12498 /* Pad the front with spaces. */
12499 memset(PL_efloatbuf, ' ', width - elen);
12507 char *ptr = ebuf + sizeof ebuf;
12510 #if defined(USE_QUADMATH)
12511 if (intsize == 'q') {
12515 /* FIXME: what to do if HAS_LONG_DOUBLE but not PERL_PRIfldbl? */
12516 #elif defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
12517 /* Note that this is HAS_LONG_DOUBLE and PERL_PRIfldbl,
12518 * not USE_LONG_DOUBLE and NVff. In other words,
12519 * this needs to work without USE_LONG_DOUBLE. */
12520 if (intsize == 'q') {
12521 /* Copy the one or more characters in a long double
12522 * format before the 'base' ([efgEFG]) character to
12523 * the format string. */
12524 static char const ldblf[] = PERL_PRIfldbl;
12525 char const *p = ldblf + sizeof(ldblf) - 3;
12526 while (p >= ldblf) { *--ptr = *p--; }
12531 do { *--ptr = '0' + (base % 10); } while (base /= 10);
12536 do { *--ptr = '0' + (base % 10); } while (base /= 10);
12548 /* No taint. Otherwise we are in the strange situation
12549 * where printf() taints but print($float) doesn't.
12552 STORE_LC_NUMERIC_SET_TO_NEEDED();
12554 /* hopefully the above makes ptr a very constrained format
12555 * that is safe to use, even though it's not literal */
12556 GCC_DIAG_IGNORE(-Wformat-nonliteral);
12557 #ifdef USE_QUADMATH
12559 const char* qfmt = quadmath_format_single(ptr);
12561 Perl_croak_nocontext("panic: quadmath invalid format \"%s\"", ptr);
12562 elen = quadmath_snprintf(PL_efloatbuf, PL_efloatsize,
12564 if ((IV)elen == -1)
12565 Perl_croak_nocontext("panic: quadmath_snprintf failed, format \"%s|'", qfmt);
12569 #elif defined(HAS_LONG_DOUBLE)
12570 elen = ((intsize == 'q')
12571 ? my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv)
12572 : my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, (double)fv));
12574 elen = my_sprintf(PL_efloatbuf, ptr, fv);
12580 eptr = PL_efloatbuf;
12581 assert((IV)elen > 0); /* here zero elen is bad */
12583 #ifdef USE_LOCALE_NUMERIC
12584 /* If the decimal point character in the string is UTF-8, make the
12586 if (PL_numeric_radix_sv && SvUTF8(PL_numeric_radix_sv)
12587 && instr(eptr, SvPVX_const(PL_numeric_radix_sv)))
12600 i = SvCUR(sv) - origlen;
12603 case 'c': *(va_arg(*args, char*)) = i; break;
12604 case 'h': *(va_arg(*args, short*)) = i; break;
12605 default: *(va_arg(*args, int*)) = i; break;
12606 case 'l': *(va_arg(*args, long*)) = i; break;
12607 case 'V': *(va_arg(*args, IV*)) = i; break;
12608 case 'z': *(va_arg(*args, SSize_t*)) = i; break;
12609 #ifdef HAS_PTRDIFF_T
12610 case 't': *(va_arg(*args, ptrdiff_t*)) = i; break;
12613 case 'j': *(va_arg(*args, intmax_t*)) = i; break;
12617 *(va_arg(*args, Quad_t*)) = i; break;
12624 sv_setuv_mg(argsv, has_utf8 ? (UV)sv_len_utf8(sv) : (UV)i);
12625 goto donevalidconversion;
12632 && (PL_op->op_type == OP_PRTF || PL_op->op_type == OP_SPRINTF)
12633 && ckWARN(WARN_PRINTF))
12635 SV * const msg = sv_newmortal();
12636 Perl_sv_setpvf(aTHX_ msg, "Invalid conversion in %sprintf: ",
12637 (PL_op->op_type == OP_PRTF) ? "" : "s");
12638 if (fmtstart < patend) {
12639 const char * const fmtend = q < patend ? q : patend;
12641 sv_catpvs(msg, "\"%");
12642 for (f = fmtstart; f < fmtend; f++) {
12644 sv_catpvn_nomg(msg, f, 1);
12646 Perl_sv_catpvf(aTHX_ msg,
12647 "\\%03"UVof, (UV)*f & 0xFF);
12650 sv_catpvs(msg, "\"");
12652 sv_catpvs(msg, "end of string");
12654 Perl_warner(aTHX_ packWARN(WARN_PRINTF), "%"SVf, SVfARG(msg)); /* yes, this is reentrant */
12657 /* output mangled stuff ... */
12663 /* ... right here, because formatting flags should not apply */
12664 SvGROW(sv, SvCUR(sv) + elen + 1);
12666 Copy(eptr, p, elen, char);
12669 SvCUR_set(sv, p - SvPVX_const(sv));
12671 continue; /* not "break" */
12674 if (is_utf8 != has_utf8) {
12677 sv_utf8_upgrade(sv);
12680 const STRLEN old_elen = elen;
12681 SV * const nsv = newSVpvn_flags(eptr, elen, SVs_TEMP);
12682 sv_utf8_upgrade(nsv);
12683 eptr = SvPVX_const(nsv);
12686 if (width) { /* fudge width (can't fudge elen) */
12687 width += elen - old_elen;
12693 assert((IV)elen >= 0); /* here zero elen is fine */
12694 have = esignlen + zeros + elen;
12696 croak_memory_wrap();
12698 need = (have > width ? have : width);
12701 if (need >= (((STRLEN)~0) - SvCUR(sv) - dotstrlen - 1))
12702 croak_memory_wrap();
12703 SvGROW(sv, SvCUR(sv) + need + dotstrlen + 1);
12705 if (esignlen && fill == '0') {
12707 for (i = 0; i < (int)esignlen; i++)
12708 *p++ = esignbuf[i];
12710 if (gap && !left) {
12711 memset(p, fill, gap);
12714 if (esignlen && fill != '0') {
12716 for (i = 0; i < (int)esignlen; i++)
12717 *p++ = esignbuf[i];
12721 for (i = zeros; i; i--)
12725 Copy(eptr, p, elen, char);
12729 memset(p, ' ', gap);
12734 Copy(dotstr, p, dotstrlen, char);
12738 vectorize = FALSE; /* done iterating over vecstr */
12745 SvCUR_set(sv, p - SvPVX_const(sv));
12751 donevalidconversion:
12752 if (used_explicit_ix)
12753 no_redundant_warning = TRUE;
12755 S_warn_vcatpvfn_missing_argument(aTHX);
12758 /* Now that we've consumed all our printf format arguments (svix)
12759 * do we have things left on the stack that we didn't use?
12761 if (!no_redundant_warning && svmax >= svix + 1 && ckWARN(WARN_REDUNDANT)) {
12762 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
12763 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
12768 RESTORE_LC_NUMERIC(); /* Done outside loop, so don't have to save/restore
12772 /* =========================================================================
12774 =head1 Cloning an interpreter
12778 All the macros and functions in this section are for the private use of
12779 the main function, perl_clone().
12781 The foo_dup() functions make an exact copy of an existing foo thingy.
12782 During the course of a cloning, a hash table is used to map old addresses
12783 to new addresses. The table is created and manipulated with the
12784 ptr_table_* functions.
12786 * =========================================================================*/
12789 #if defined(USE_ITHREADS)
12791 /* XXX Remove this so it doesn't have to go thru the macro and return for nothing */
12792 #ifndef GpREFCNT_inc
12793 # define GpREFCNT_inc(gp) ((gp) ? (++(gp)->gp_refcnt, (gp)) : (GP*)NULL)
12797 /* Certain cases in Perl_ss_dup have been merged, by relying on the fact
12798 that currently av_dup, gv_dup and hv_dup are the same as sv_dup.
12799 If this changes, please unmerge ss_dup.
12800 Likewise, sv_dup_inc_multiple() relies on this fact. */
12801 #define sv_dup_inc_NN(s,t) SvREFCNT_inc_NN(sv_dup_inc(s,t))
12802 #define av_dup(s,t) MUTABLE_AV(sv_dup((const SV *)s,t))
12803 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
12804 #define hv_dup(s,t) MUTABLE_HV(sv_dup((const SV *)s,t))
12805 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
12806 #define cv_dup(s,t) MUTABLE_CV(sv_dup((const SV *)s,t))
12807 #define cv_dup_inc(s,t) MUTABLE_CV(sv_dup_inc((const SV *)s,t))
12808 #define io_dup(s,t) MUTABLE_IO(sv_dup((const SV *)s,t))
12809 #define io_dup_inc(s,t) MUTABLE_IO(sv_dup_inc((const SV *)s,t))
12810 #define gv_dup(s,t) MUTABLE_GV(sv_dup((const SV *)s,t))
12811 #define gv_dup_inc(s,t) MUTABLE_GV(sv_dup_inc((const SV *)s,t))
12812 #define SAVEPV(p) ((p) ? savepv(p) : NULL)
12813 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
12815 /* clone a parser */
12818 Perl_parser_dup(pTHX_ const yy_parser *const proto, CLONE_PARAMS *const param)
12822 PERL_ARGS_ASSERT_PARSER_DUP;
12827 /* look for it in the table first */
12828 parser = (yy_parser *)ptr_table_fetch(PL_ptr_table, proto);
12832 /* create anew and remember what it is */
12833 Newxz(parser, 1, yy_parser);
12834 ptr_table_store(PL_ptr_table, proto, parser);
12836 /* XXX these not yet duped */
12837 parser->old_parser = NULL;
12838 parser->stack = NULL;
12840 parser->stack_size = 0;
12841 /* XXX parser->stack->state = 0; */
12843 /* XXX eventually, just Copy() most of the parser struct ? */
12845 parser->lex_brackets = proto->lex_brackets;
12846 parser->lex_casemods = proto->lex_casemods;
12847 parser->lex_brackstack = savepvn(proto->lex_brackstack,
12848 (proto->lex_brackets < 120 ? 120 : proto->lex_brackets));
12849 parser->lex_casestack = savepvn(proto->lex_casestack,
12850 (proto->lex_casemods < 12 ? 12 : proto->lex_casemods));
12851 parser->lex_defer = proto->lex_defer;
12852 parser->lex_dojoin = proto->lex_dojoin;
12853 parser->lex_formbrack = proto->lex_formbrack;
12854 parser->lex_inpat = proto->lex_inpat;
12855 parser->lex_inwhat = proto->lex_inwhat;
12856 parser->lex_op = proto->lex_op;
12857 parser->lex_repl = sv_dup_inc(proto->lex_repl, param);
12858 parser->lex_starts = proto->lex_starts;
12859 parser->lex_stuff = sv_dup_inc(proto->lex_stuff, param);
12860 parser->multi_close = proto->multi_close;
12861 parser->multi_open = proto->multi_open;
12862 parser->multi_start = proto->multi_start;
12863 parser->multi_end = proto->multi_end;
12864 parser->preambled = proto->preambled;
12865 parser->sublex_info = proto->sublex_info; /* XXX not quite right */
12866 parser->linestr = sv_dup_inc(proto->linestr, param);
12867 parser->expect = proto->expect;
12868 parser->copline = proto->copline;
12869 parser->last_lop_op = proto->last_lop_op;
12870 parser->lex_state = proto->lex_state;
12871 parser->rsfp = fp_dup(proto->rsfp, '<', param);
12872 /* rsfp_filters entries have fake IoDIRP() */
12873 parser->rsfp_filters= av_dup_inc(proto->rsfp_filters, param);
12874 parser->in_my = proto->in_my;
12875 parser->in_my_stash = hv_dup(proto->in_my_stash, param);
12876 parser->error_count = proto->error_count;
12879 parser->linestr = sv_dup_inc(proto->linestr, param);
12882 char * const ols = SvPVX(proto->linestr);
12883 char * const ls = SvPVX(parser->linestr);
12885 parser->bufptr = ls + (proto->bufptr >= ols ?
12886 proto->bufptr - ols : 0);
12887 parser->oldbufptr = ls + (proto->oldbufptr >= ols ?
12888 proto->oldbufptr - ols : 0);
12889 parser->oldoldbufptr= ls + (proto->oldoldbufptr >= ols ?
12890 proto->oldoldbufptr - ols : 0);
12891 parser->linestart = ls + (proto->linestart >= ols ?
12892 proto->linestart - ols : 0);
12893 parser->last_uni = ls + (proto->last_uni >= ols ?
12894 proto->last_uni - ols : 0);
12895 parser->last_lop = ls + (proto->last_lop >= ols ?
12896 proto->last_lop - ols : 0);
12898 parser->bufend = ls + SvCUR(parser->linestr);
12901 Copy(proto->tokenbuf, parser->tokenbuf, 256, char);
12904 Copy(proto->nextval, parser->nextval, 5, YYSTYPE);
12905 Copy(proto->nexttype, parser->nexttype, 5, I32);
12906 parser->nexttoke = proto->nexttoke;
12908 /* XXX should clone saved_curcop here, but we aren't passed
12909 * proto_perl; so do it in perl_clone_using instead */
12915 /* duplicate a file handle */
12918 Perl_fp_dup(pTHX_ PerlIO *const fp, const char type, CLONE_PARAMS *const param)
12922 PERL_ARGS_ASSERT_FP_DUP;
12923 PERL_UNUSED_ARG(type);
12926 return (PerlIO*)NULL;
12928 /* look for it in the table first */
12929 ret = (PerlIO*)ptr_table_fetch(PL_ptr_table, fp);
12933 /* create anew and remember what it is */
12934 #ifdef __amigaos4__
12935 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE|PERLIO_DUP_FD);
12937 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE);
12939 ptr_table_store(PL_ptr_table, fp, ret);
12943 /* duplicate a directory handle */
12946 Perl_dirp_dup(pTHX_ DIR *const dp, CLONE_PARAMS *const param)
12950 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
12952 const Direntry_t *dirent;
12953 char smallbuf[256];
12959 PERL_UNUSED_CONTEXT;
12960 PERL_ARGS_ASSERT_DIRP_DUP;
12965 /* look for it in the table first */
12966 ret = (DIR*)ptr_table_fetch(PL_ptr_table, dp);
12970 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
12972 PERL_UNUSED_ARG(param);
12976 /* open the current directory (so we can switch back) */
12977 if (!(pwd = PerlDir_open("."))) return (DIR *)NULL;
12979 /* chdir to our dir handle and open the present working directory */
12980 if (fchdir(my_dirfd(dp)) < 0 || !(ret = PerlDir_open("."))) {
12981 PerlDir_close(pwd);
12982 return (DIR *)NULL;
12984 /* Now we should have two dir handles pointing to the same dir. */
12986 /* Be nice to the calling code and chdir back to where we were. */
12987 /* XXX If this fails, then what? */
12988 PERL_UNUSED_RESULT(fchdir(my_dirfd(pwd)));
12990 /* We have no need of the pwd handle any more. */
12991 PerlDir_close(pwd);
12994 # define d_namlen(d) (d)->d_namlen
12996 # define d_namlen(d) strlen((d)->d_name)
12998 /* Iterate once through dp, to get the file name at the current posi-
12999 tion. Then step back. */
13000 pos = PerlDir_tell(dp);
13001 if ((dirent = PerlDir_read(dp))) {
13002 len = d_namlen(dirent);
13003 if (len <= sizeof smallbuf) name = smallbuf;
13004 else Newx(name, len, char);
13005 Move(dirent->d_name, name, len, char);
13007 PerlDir_seek(dp, pos);
13009 /* Iterate through the new dir handle, till we find a file with the
13011 if (!dirent) /* just before the end */
13013 pos = PerlDir_tell(ret);
13014 if (PerlDir_read(ret)) continue; /* not there yet */
13015 PerlDir_seek(ret, pos); /* step back */
13019 const long pos0 = PerlDir_tell(ret);
13021 pos = PerlDir_tell(ret);
13022 if ((dirent = PerlDir_read(ret))) {
13023 if (len == (STRLEN)d_namlen(dirent)
13024 && memEQ(name, dirent->d_name, len)) {
13026 PerlDir_seek(ret, pos); /* step back */
13029 /* else we are not there yet; keep iterating */
13031 else { /* This is not meant to happen. The best we can do is
13032 reset the iterator to the beginning. */
13033 PerlDir_seek(ret, pos0);
13040 if (name && name != smallbuf)
13045 ret = win32_dirp_dup(dp, param);
13048 /* pop it in the pointer table */
13050 ptr_table_store(PL_ptr_table, dp, ret);
13055 /* duplicate a typeglob */
13058 Perl_gp_dup(pTHX_ GP *const gp, CLONE_PARAMS *const param)
13062 PERL_ARGS_ASSERT_GP_DUP;
13066 /* look for it in the table first */
13067 ret = (GP*)ptr_table_fetch(PL_ptr_table, gp);
13071 /* create anew and remember what it is */
13073 ptr_table_store(PL_ptr_table, gp, ret);
13076 /* ret->gp_refcnt must be 0 before any other dups are called. We're relying
13077 on Newxz() to do this for us. */
13078 ret->gp_sv = sv_dup_inc(gp->gp_sv, param);
13079 ret->gp_io = io_dup_inc(gp->gp_io, param);
13080 ret->gp_form = cv_dup_inc(gp->gp_form, param);
13081 ret->gp_av = av_dup_inc(gp->gp_av, param);
13082 ret->gp_hv = hv_dup_inc(gp->gp_hv, param);
13083 ret->gp_egv = gv_dup(gp->gp_egv, param);/* GvEGV is not refcounted */
13084 ret->gp_cv = cv_dup_inc(gp->gp_cv, param);
13085 ret->gp_cvgen = gp->gp_cvgen;
13086 ret->gp_line = gp->gp_line;
13087 ret->gp_file_hek = hek_dup(gp->gp_file_hek, param);
13091 /* duplicate a chain of magic */
13094 Perl_mg_dup(pTHX_ MAGIC *mg, CLONE_PARAMS *const param)
13096 MAGIC *mgret = NULL;
13097 MAGIC **mgprev_p = &mgret;
13099 PERL_ARGS_ASSERT_MG_DUP;
13101 for (; mg; mg = mg->mg_moremagic) {
13104 if ((param->flags & CLONEf_JOIN_IN)
13105 && mg->mg_type == PERL_MAGIC_backref)
13106 /* when joining, we let the individual SVs add themselves to
13107 * backref as needed. */
13110 Newx(nmg, 1, MAGIC);
13112 mgprev_p = &(nmg->mg_moremagic);
13114 /* There was a comment "XXX copy dynamic vtable?" but as we don't have
13115 dynamic vtables, I'm not sure why Sarathy wrote it. The comment dates
13116 from the original commit adding Perl_mg_dup() - revision 4538.
13117 Similarly there is the annotation "XXX random ptr?" next to the
13118 assignment to nmg->mg_ptr. */
13121 /* FIXME for plugins
13122 if (nmg->mg_type == PERL_MAGIC_qr) {
13123 nmg->mg_obj = MUTABLE_SV(CALLREGDUPE((REGEXP*)nmg->mg_obj, param));
13127 nmg->mg_obj = (nmg->mg_flags & MGf_REFCOUNTED)
13128 ? nmg->mg_type == PERL_MAGIC_backref
13129 /* The backref AV has its reference
13130 * count deliberately bumped by 1 */
13131 ? SvREFCNT_inc(av_dup_inc((const AV *)
13132 nmg->mg_obj, param))
13133 : sv_dup_inc(nmg->mg_obj, param)
13134 : sv_dup(nmg->mg_obj, param);
13136 if (nmg->mg_ptr && nmg->mg_type != PERL_MAGIC_regex_global) {
13137 if (nmg->mg_len > 0) {
13138 nmg->mg_ptr = SAVEPVN(nmg->mg_ptr, nmg->mg_len);
13139 if (nmg->mg_type == PERL_MAGIC_overload_table &&
13140 AMT_AMAGIC((AMT*)nmg->mg_ptr))
13142 AMT * const namtp = (AMT*)nmg->mg_ptr;
13143 sv_dup_inc_multiple((SV**)(namtp->table),
13144 (SV**)(namtp->table), NofAMmeth, param);
13147 else if (nmg->mg_len == HEf_SVKEY)
13148 nmg->mg_ptr = (char*)sv_dup_inc((const SV *)nmg->mg_ptr, param);
13150 if ((nmg->mg_flags & MGf_DUP) && nmg->mg_virtual && nmg->mg_virtual->svt_dup) {
13151 nmg->mg_virtual->svt_dup(aTHX_ nmg, param);
13157 #endif /* USE_ITHREADS */
13159 struct ptr_tbl_arena {
13160 struct ptr_tbl_arena *next;
13161 struct ptr_tbl_ent array[1023/3]; /* as ptr_tbl_ent has 3 pointers. */
13164 /* create a new pointer-mapping table */
13167 Perl_ptr_table_new(pTHX)
13170 PERL_UNUSED_CONTEXT;
13172 Newx(tbl, 1, PTR_TBL_t);
13173 tbl->tbl_max = 511;
13174 tbl->tbl_items = 0;
13175 tbl->tbl_arena = NULL;
13176 tbl->tbl_arena_next = NULL;
13177 tbl->tbl_arena_end = NULL;
13178 Newxz(tbl->tbl_ary, tbl->tbl_max + 1, PTR_TBL_ENT_t*);
13182 #define PTR_TABLE_HASH(ptr) \
13183 ((PTR2UV(ptr) >> 3) ^ (PTR2UV(ptr) >> (3 + 7)) ^ (PTR2UV(ptr) >> (3 + 17)))
13185 /* map an existing pointer using a table */
13187 STATIC PTR_TBL_ENT_t *
13188 S_ptr_table_find(PTR_TBL_t *const tbl, const void *const sv)
13190 PTR_TBL_ENT_t *tblent;
13191 const UV hash = PTR_TABLE_HASH(sv);
13193 PERL_ARGS_ASSERT_PTR_TABLE_FIND;
13195 tblent = tbl->tbl_ary[hash & tbl->tbl_max];
13196 for (; tblent; tblent = tblent->next) {
13197 if (tblent->oldval == sv)
13204 Perl_ptr_table_fetch(pTHX_ PTR_TBL_t *const tbl, const void *const sv)
13206 PTR_TBL_ENT_t const *const tblent = ptr_table_find(tbl, sv);
13208 PERL_ARGS_ASSERT_PTR_TABLE_FETCH;
13209 PERL_UNUSED_CONTEXT;
13211 return tblent ? tblent->newval : NULL;
13214 /* add a new entry to a pointer-mapping table 'tbl'. In hash terms, 'oldsv' is
13215 * the key; 'newsv' is the value. The names "old" and "new" are specific to
13216 * the core's typical use of ptr_tables in thread cloning. */
13219 Perl_ptr_table_store(pTHX_ PTR_TBL_t *const tbl, const void *const oldsv, void *const newsv)
13221 PTR_TBL_ENT_t *tblent = ptr_table_find(tbl, oldsv);
13223 PERL_ARGS_ASSERT_PTR_TABLE_STORE;
13224 PERL_UNUSED_CONTEXT;
13227 tblent->newval = newsv;
13229 const UV entry = PTR_TABLE_HASH(oldsv) & tbl->tbl_max;
13231 if (tbl->tbl_arena_next == tbl->tbl_arena_end) {
13232 struct ptr_tbl_arena *new_arena;
13234 Newx(new_arena, 1, struct ptr_tbl_arena);
13235 new_arena->next = tbl->tbl_arena;
13236 tbl->tbl_arena = new_arena;
13237 tbl->tbl_arena_next = new_arena->array;
13238 tbl->tbl_arena_end = C_ARRAY_END(new_arena->array);
13241 tblent = tbl->tbl_arena_next++;
13243 tblent->oldval = oldsv;
13244 tblent->newval = newsv;
13245 tblent->next = tbl->tbl_ary[entry];
13246 tbl->tbl_ary[entry] = tblent;
13248 if (tblent->next && tbl->tbl_items > tbl->tbl_max)
13249 ptr_table_split(tbl);
13253 /* double the hash bucket size of an existing ptr table */
13256 Perl_ptr_table_split(pTHX_ PTR_TBL_t *const tbl)
13258 PTR_TBL_ENT_t **ary = tbl->tbl_ary;
13259 const UV oldsize = tbl->tbl_max + 1;
13260 UV newsize = oldsize * 2;
13263 PERL_ARGS_ASSERT_PTR_TABLE_SPLIT;
13264 PERL_UNUSED_CONTEXT;
13266 Renew(ary, newsize, PTR_TBL_ENT_t*);
13267 Zero(&ary[oldsize], newsize-oldsize, PTR_TBL_ENT_t*);
13268 tbl->tbl_max = --newsize;
13269 tbl->tbl_ary = ary;
13270 for (i=0; i < oldsize; i++, ary++) {
13271 PTR_TBL_ENT_t **entp = ary;
13272 PTR_TBL_ENT_t *ent = *ary;
13273 PTR_TBL_ENT_t **curentp;
13276 curentp = ary + oldsize;
13278 if ((newsize & PTR_TABLE_HASH(ent->oldval)) != i) {
13280 ent->next = *curentp;
13290 /* remove all the entries from a ptr table */
13291 /* Deprecated - will be removed post 5.14 */
13294 Perl_ptr_table_clear(pTHX_ PTR_TBL_t *const tbl)
13296 PERL_UNUSED_CONTEXT;
13297 if (tbl && tbl->tbl_items) {
13298 struct ptr_tbl_arena *arena = tbl->tbl_arena;
13300 Zero(tbl->tbl_ary, tbl->tbl_max + 1, struct ptr_tbl_ent *);
13303 struct ptr_tbl_arena *next = arena->next;
13309 tbl->tbl_items = 0;
13310 tbl->tbl_arena = NULL;
13311 tbl->tbl_arena_next = NULL;
13312 tbl->tbl_arena_end = NULL;
13316 /* clear and free a ptr table */
13319 Perl_ptr_table_free(pTHX_ PTR_TBL_t *const tbl)
13321 struct ptr_tbl_arena *arena;
13323 PERL_UNUSED_CONTEXT;
13329 arena = tbl->tbl_arena;
13332 struct ptr_tbl_arena *next = arena->next;
13338 Safefree(tbl->tbl_ary);
13342 #if defined(USE_ITHREADS)
13345 Perl_rvpv_dup(pTHX_ SV *const dstr, const SV *const sstr, CLONE_PARAMS *const param)
13347 PERL_ARGS_ASSERT_RVPV_DUP;
13349 assert(!isREGEXP(sstr));
13351 if (SvWEAKREF(sstr)) {
13352 SvRV_set(dstr, sv_dup(SvRV_const(sstr), param));
13353 if (param->flags & CLONEf_JOIN_IN) {
13354 /* if joining, we add any back references individually rather
13355 * than copying the whole backref array */
13356 Perl_sv_add_backref(aTHX_ SvRV(dstr), dstr);
13360 SvRV_set(dstr, sv_dup_inc(SvRV_const(sstr), param));
13362 else if (SvPVX_const(sstr)) {
13363 /* Has something there */
13365 /* Normal PV - clone whole allocated space */
13366 SvPV_set(dstr, SAVEPVN(SvPVX_const(sstr), SvLEN(sstr)-1));
13367 /* sstr may not be that normal, but actually copy on write.
13368 But we are a true, independent SV, so: */
13372 /* Special case - not normally malloced for some reason */
13373 if (isGV_with_GP(sstr)) {
13374 /* Don't need to do anything here. */
13376 else if ((SvIsCOW(sstr))) {
13377 /* A "shared" PV - clone it as "shared" PV */
13379 HEK_KEY(hek_dup(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)),
13383 /* Some other special case - random pointer */
13384 SvPV_set(dstr, (char *) SvPVX_const(sstr));
13389 /* Copy the NULL */
13390 SvPV_set(dstr, NULL);
13394 /* duplicate a list of SVs. source and dest may point to the same memory. */
13396 S_sv_dup_inc_multiple(pTHX_ SV *const *source, SV **dest,
13397 SSize_t items, CLONE_PARAMS *const param)
13399 PERL_ARGS_ASSERT_SV_DUP_INC_MULTIPLE;
13401 while (items-- > 0) {
13402 *dest++ = sv_dup_inc(*source++, param);
13408 /* duplicate an SV of any type (including AV, HV etc) */
13411 S_sv_dup_common(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13416 PERL_ARGS_ASSERT_SV_DUP_COMMON;
13418 if (SvTYPE(sstr) == (svtype)SVTYPEMASK) {
13419 #ifdef DEBUG_LEAKING_SCALARS_ABORT
13424 /* look for it in the table first */
13425 dstr = MUTABLE_SV(ptr_table_fetch(PL_ptr_table, sstr));
13429 if(param->flags & CLONEf_JOIN_IN) {
13430 /** We are joining here so we don't want do clone
13431 something that is bad **/
13432 if (SvTYPE(sstr) == SVt_PVHV) {
13433 const HEK * const hvname = HvNAME_HEK(sstr);
13435 /** don't clone stashes if they already exist **/
13436 dstr = MUTABLE_SV(gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
13437 HEK_UTF8(hvname) ? SVf_UTF8 : 0));
13438 ptr_table_store(PL_ptr_table, sstr, dstr);
13442 else if (SvTYPE(sstr) == SVt_PVGV && !SvFAKE(sstr)) {
13443 HV *stash = GvSTASH(sstr);
13444 const HEK * hvname;
13445 if (stash && (hvname = HvNAME_HEK(stash))) {
13446 /** don't clone GVs if they already exist **/
13448 stash = gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
13449 HEK_UTF8(hvname) ? SVf_UTF8 : 0);
13451 stash, GvNAME(sstr),
13457 if (svp && *svp && SvTYPE(*svp) == SVt_PVGV) {
13458 ptr_table_store(PL_ptr_table, sstr, *svp);
13465 /* create anew and remember what it is */
13468 #ifdef DEBUG_LEAKING_SCALARS
13469 dstr->sv_debug_optype = sstr->sv_debug_optype;
13470 dstr->sv_debug_line = sstr->sv_debug_line;
13471 dstr->sv_debug_inpad = sstr->sv_debug_inpad;
13472 dstr->sv_debug_parent = (SV*)sstr;
13473 FREE_SV_DEBUG_FILE(dstr);
13474 dstr->sv_debug_file = savesharedpv(sstr->sv_debug_file);
13477 ptr_table_store(PL_ptr_table, sstr, dstr);
13480 SvFLAGS(dstr) = SvFLAGS(sstr);
13481 SvFLAGS(dstr) &= ~SVf_OOK; /* don't propagate OOK hack */
13482 SvREFCNT(dstr) = 0; /* must be before any other dups! */
13485 if (SvANY(sstr) && PL_watch_pvx && SvPVX_const(sstr) == PL_watch_pvx)
13486 PerlIO_printf(Perl_debug_log, "watch at %p hit, found string \"%s\"\n",
13487 (void*)PL_watch_pvx, SvPVX_const(sstr));
13490 /* don't clone objects whose class has asked us not to */
13492 && ! (SvFLAGS(SvSTASH(sstr)) & SVphv_CLONEABLE))
13498 switch (SvTYPE(sstr)) {
13500 SvANY(dstr) = NULL;
13503 SET_SVANY_FOR_BODYLESS_IV(dstr);
13505 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
13507 SvIV_set(dstr, SvIVX(sstr));
13511 #if NVSIZE <= IVSIZE
13512 SET_SVANY_FOR_BODYLESS_NV(dstr);
13514 SvANY(dstr) = new_XNV();
13516 SvNV_set(dstr, SvNVX(sstr));
13520 /* These are all the types that need complex bodies allocating. */
13522 const svtype sv_type = SvTYPE(sstr);
13523 const struct body_details *const sv_type_details
13524 = bodies_by_type + sv_type;
13528 Perl_croak(aTHX_ "Bizarre SvTYPE [%" IVdf "]", (IV)SvTYPE(sstr));
13544 assert(sv_type_details->body_size);
13545 if (sv_type_details->arena) {
13546 new_body_inline(new_body, sv_type);
13548 = (void*)((char*)new_body - sv_type_details->offset);
13550 new_body = new_NOARENA(sv_type_details);
13554 SvANY(dstr) = new_body;
13557 Copy(((char*)SvANY(sstr)) + sv_type_details->offset,
13558 ((char*)SvANY(dstr)) + sv_type_details->offset,
13559 sv_type_details->copy, char);
13561 Copy(((char*)SvANY(sstr)),
13562 ((char*)SvANY(dstr)),
13563 sv_type_details->body_size + sv_type_details->offset, char);
13566 if (sv_type != SVt_PVAV && sv_type != SVt_PVHV
13567 && !isGV_with_GP(dstr)
13569 && !(sv_type == SVt_PVIO && !(IoFLAGS(dstr) & IOf_FAKE_DIRP)))
13570 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
13572 /* The Copy above means that all the source (unduplicated) pointers
13573 are now in the destination. We can check the flags and the
13574 pointers in either, but it's possible that there's less cache
13575 missing by always going for the destination.
13576 FIXME - instrument and check that assumption */
13577 if (sv_type >= SVt_PVMG) {
13579 SvMAGIC_set(dstr, mg_dup(SvMAGIC(dstr), param));
13580 if (SvOBJECT(dstr) && SvSTASH(dstr))
13581 SvSTASH_set(dstr, hv_dup_inc(SvSTASH(dstr), param));
13582 else SvSTASH_set(dstr, 0); /* don't copy DESTROY cache */
13585 /* The cast silences a GCC warning about unhandled types. */
13586 switch ((int)sv_type) {
13597 /* FIXME for plugins */
13598 dstr->sv_u.svu_rx = ((REGEXP *)dstr)->sv_any;
13599 re_dup_guts((REGEXP*) sstr, (REGEXP*) dstr, param);
13602 /* XXX LvTARGOFF sometimes holds PMOP* when DEBUGGING */
13603 if (LvTYPE(dstr) == 't') /* for tie: unrefcnted fake (SV**) */
13604 LvTARG(dstr) = dstr;
13605 else if (LvTYPE(dstr) == 'T') /* for tie: fake HE */
13606 LvTARG(dstr) = MUTABLE_SV(he_dup((HE*)LvTARG(dstr), 0, param));
13608 LvTARG(dstr) = sv_dup_inc(LvTARG(dstr), param);
13609 if (isREGEXP(sstr)) goto duprex;
13611 /* non-GP case already handled above */
13612 if(isGV_with_GP(sstr)) {
13613 GvNAME_HEK(dstr) = hek_dup(GvNAME_HEK(dstr), param);
13614 /* Don't call sv_add_backref here as it's going to be
13615 created as part of the magic cloning of the symbol
13616 table--unless this is during a join and the stash
13617 is not actually being cloned. */
13618 /* Danger Will Robinson - GvGP(dstr) isn't initialised
13619 at the point of this comment. */
13620 GvSTASH(dstr) = hv_dup(GvSTASH(dstr), param);
13621 if (param->flags & CLONEf_JOIN_IN)
13622 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
13623 GvGP_set(dstr, gp_dup(GvGP(sstr), param));
13624 (void)GpREFCNT_inc(GvGP(dstr));
13628 /* PL_parser->rsfp_filters entries have fake IoDIRP() */
13629 if(IoFLAGS(dstr) & IOf_FAKE_DIRP) {
13630 /* I have no idea why fake dirp (rsfps)
13631 should be treated differently but otherwise
13632 we end up with leaks -- sky*/
13633 IoTOP_GV(dstr) = gv_dup_inc(IoTOP_GV(dstr), param);
13634 IoFMT_GV(dstr) = gv_dup_inc(IoFMT_GV(dstr), param);
13635 IoBOTTOM_GV(dstr) = gv_dup_inc(IoBOTTOM_GV(dstr), param);
13637 IoTOP_GV(dstr) = gv_dup(IoTOP_GV(dstr), param);
13638 IoFMT_GV(dstr) = gv_dup(IoFMT_GV(dstr), param);
13639 IoBOTTOM_GV(dstr) = gv_dup(IoBOTTOM_GV(dstr), param);
13640 if (IoDIRP(dstr)) {
13641 IoDIRP(dstr) = dirp_dup(IoDIRP(dstr), param);
13644 /* IoDIRP(dstr) is already a copy of IoDIRP(sstr) */
13646 IoIFP(dstr) = fp_dup(IoIFP(sstr), IoTYPE(dstr), param);
13648 if (IoOFP(dstr) == IoIFP(sstr))
13649 IoOFP(dstr) = IoIFP(dstr);
13651 IoOFP(dstr) = fp_dup(IoOFP(dstr), IoTYPE(dstr), param);
13652 IoTOP_NAME(dstr) = SAVEPV(IoTOP_NAME(dstr));
13653 IoFMT_NAME(dstr) = SAVEPV(IoFMT_NAME(dstr));
13654 IoBOTTOM_NAME(dstr) = SAVEPV(IoBOTTOM_NAME(dstr));
13657 /* avoid cloning an empty array */
13658 if (AvARRAY((const AV *)sstr) && AvFILLp((const AV *)sstr) >= 0) {
13659 SV **dst_ary, **src_ary;
13660 SSize_t items = AvFILLp((const AV *)sstr) + 1;
13662 src_ary = AvARRAY((const AV *)sstr);
13663 Newxz(dst_ary, AvMAX((const AV *)sstr)+1, SV*);
13664 ptr_table_store(PL_ptr_table, src_ary, dst_ary);
13665 AvARRAY(MUTABLE_AV(dstr)) = dst_ary;
13666 AvALLOC((const AV *)dstr) = dst_ary;
13667 if (AvREAL((const AV *)sstr)) {
13668 dst_ary = sv_dup_inc_multiple(src_ary, dst_ary, items,
13672 while (items-- > 0)
13673 *dst_ary++ = sv_dup(*src_ary++, param);
13675 items = AvMAX((const AV *)sstr) - AvFILLp((const AV *)sstr);
13676 while (items-- > 0) {
13681 AvARRAY(MUTABLE_AV(dstr)) = NULL;
13682 AvALLOC((const AV *)dstr) = (SV**)NULL;
13683 AvMAX( (const AV *)dstr) = -1;
13684 AvFILLp((const AV *)dstr) = -1;
13688 if (HvARRAY((const HV *)sstr)) {
13690 const bool sharekeys = !!HvSHAREKEYS(sstr);
13691 XPVHV * const dxhv = (XPVHV*)SvANY(dstr);
13692 XPVHV * const sxhv = (XPVHV*)SvANY(sstr);
13694 Newx(darray, PERL_HV_ARRAY_ALLOC_BYTES(dxhv->xhv_max+1)
13695 + (SvOOK(sstr) ? sizeof(struct xpvhv_aux) : 0),
13697 HvARRAY(dstr) = (HE**)darray;
13698 while (i <= sxhv->xhv_max) {
13699 const HE * const source = HvARRAY(sstr)[i];
13700 HvARRAY(dstr)[i] = source
13701 ? he_dup(source, sharekeys, param) : 0;
13705 const struct xpvhv_aux * const saux = HvAUX(sstr);
13706 struct xpvhv_aux * const daux = HvAUX(dstr);
13707 /* This flag isn't copied. */
13710 if (saux->xhv_name_count) {
13711 HEK ** const sname = saux->xhv_name_u.xhvnameu_names;
13713 = saux->xhv_name_count < 0
13714 ? -saux->xhv_name_count
13715 : saux->xhv_name_count;
13716 HEK **shekp = sname + count;
13718 Newx(daux->xhv_name_u.xhvnameu_names, count, HEK *);
13719 dhekp = daux->xhv_name_u.xhvnameu_names + count;
13720 while (shekp-- > sname) {
13722 *dhekp = hek_dup(*shekp, param);
13726 daux->xhv_name_u.xhvnameu_name
13727 = hek_dup(saux->xhv_name_u.xhvnameu_name,
13730 daux->xhv_name_count = saux->xhv_name_count;
13732 daux->xhv_fill_lazy = saux->xhv_fill_lazy;
13733 daux->xhv_aux_flags = saux->xhv_aux_flags;
13734 #ifdef PERL_HASH_RANDOMIZE_KEYS
13735 daux->xhv_rand = saux->xhv_rand;
13736 daux->xhv_last_rand = saux->xhv_last_rand;
13738 daux->xhv_riter = saux->xhv_riter;
13739 daux->xhv_eiter = saux->xhv_eiter
13740 ? he_dup(saux->xhv_eiter,
13741 cBOOL(HvSHAREKEYS(sstr)), param) : 0;
13742 /* backref array needs refcnt=2; see sv_add_backref */
13743 daux->xhv_backreferences =
13744 (param->flags & CLONEf_JOIN_IN)
13745 /* when joining, we let the individual GVs and
13746 * CVs add themselves to backref as
13747 * needed. This avoids pulling in stuff
13748 * that isn't required, and simplifies the
13749 * case where stashes aren't cloned back
13750 * if they already exist in the parent
13753 : saux->xhv_backreferences
13754 ? (SvTYPE(saux->xhv_backreferences) == SVt_PVAV)
13755 ? MUTABLE_AV(SvREFCNT_inc(
13756 sv_dup_inc((const SV *)
13757 saux->xhv_backreferences, param)))
13758 : MUTABLE_AV(sv_dup((const SV *)
13759 saux->xhv_backreferences, param))
13762 daux->xhv_mro_meta = saux->xhv_mro_meta
13763 ? mro_meta_dup(saux->xhv_mro_meta, param)
13766 /* Record stashes for possible cloning in Perl_clone(). */
13768 av_push(param->stashes, dstr);
13772 HvARRAY(MUTABLE_HV(dstr)) = NULL;
13775 if (!(param->flags & CLONEf_COPY_STACKS)) {
13780 /* NOTE: not refcounted */
13781 SvANY(MUTABLE_CV(dstr))->xcv_stash =
13782 hv_dup(CvSTASH(dstr), param);
13783 if ((param->flags & CLONEf_JOIN_IN) && CvSTASH(dstr))
13784 Perl_sv_add_backref(aTHX_ MUTABLE_SV(CvSTASH(dstr)), dstr);
13785 if (!CvISXSUB(dstr)) {
13787 CvROOT(dstr) = OpREFCNT_inc(CvROOT(dstr));
13789 CvSLABBED_off(dstr);
13790 } else if (CvCONST(dstr)) {
13791 CvXSUBANY(dstr).any_ptr =
13792 sv_dup_inc((const SV *)CvXSUBANY(dstr).any_ptr, param);
13794 assert(!CvSLABBED(dstr));
13795 if (CvDYNFILE(dstr)) CvFILE(dstr) = SAVEPV(CvFILE(dstr));
13797 SvANY((CV *)dstr)->xcv_gv_u.xcv_hek =
13798 hek_dup(CvNAME_HEK((CV *)sstr), param);
13799 /* don't dup if copying back - CvGV isn't refcounted, so the
13800 * duped GV may never be freed. A bit of a hack! DAPM */
13802 SvANY(MUTABLE_CV(dstr))->xcv_gv_u.xcv_gv =
13804 ? gv_dup_inc(CvGV(sstr), param)
13805 : (param->flags & CLONEf_JOIN_IN)
13807 : gv_dup(CvGV(sstr), param);
13809 if (!CvISXSUB(sstr)) {
13810 PADLIST * padlist = CvPADLIST(sstr);
13812 padlist = padlist_dup(padlist, param);
13813 CvPADLIST_set(dstr, padlist);
13815 /* unthreaded perl can't sv_dup so we dont support unthreaded's CvHSCXT */
13816 PoisonPADLIST(dstr);
13819 CvWEAKOUTSIDE(sstr)
13820 ? cv_dup( CvOUTSIDE(dstr), param)
13821 : cv_dup_inc(CvOUTSIDE(dstr), param);
13831 Perl_sv_dup_inc(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13833 PERL_ARGS_ASSERT_SV_DUP_INC;
13834 return sstr ? SvREFCNT_inc(sv_dup_common(sstr, param)) : NULL;
13838 Perl_sv_dup(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13840 SV *dstr = sstr ? sv_dup_common(sstr, param) : NULL;
13841 PERL_ARGS_ASSERT_SV_DUP;
13843 /* Track every SV that (at least initially) had a reference count of 0.
13844 We need to do this by holding an actual reference to it in this array.
13845 If we attempt to cheat, turn AvREAL_off(), and store only pointers
13846 (akin to the stashes hash, and the perl stack), we come unstuck if
13847 a weak reference (or other SV legitimately SvREFCNT() == 0 for this
13848 thread) is manipulated in a CLONE method, because CLONE runs before the
13849 unreferenced array is walked to find SVs still with SvREFCNT() == 0
13850 (and fix things up by giving each a reference via the temps stack).
13851 Instead, during CLONE, if the 0-referenced SV has SvREFCNT_inc() and
13852 then SvREFCNT_dec(), it will be cleaned up (and added to the free list)
13853 before the walk of unreferenced happens and a reference to that is SV
13854 added to the temps stack. At which point we have the same SV considered
13855 to be in use, and free to be re-used. Not good.
13857 if (dstr && !(param->flags & CLONEf_COPY_STACKS) && !SvREFCNT(dstr)) {
13858 assert(param->unreferenced);
13859 av_push(param->unreferenced, SvREFCNT_inc(dstr));
13865 /* duplicate a context */
13868 Perl_cx_dup(pTHX_ PERL_CONTEXT *cxs, I32 ix, I32 max, CLONE_PARAMS* param)
13870 PERL_CONTEXT *ncxs;
13872 PERL_ARGS_ASSERT_CX_DUP;
13875 return (PERL_CONTEXT*)NULL;
13877 /* look for it in the table first */
13878 ncxs = (PERL_CONTEXT*)ptr_table_fetch(PL_ptr_table, cxs);
13882 /* create anew and remember what it is */
13883 Newx(ncxs, max + 1, PERL_CONTEXT);
13884 ptr_table_store(PL_ptr_table, cxs, ncxs);
13885 Copy(cxs, ncxs, max + 1, PERL_CONTEXT);
13888 PERL_CONTEXT * const ncx = &ncxs[ix];
13889 if (CxTYPE(ncx) == CXt_SUBST) {
13890 Perl_croak(aTHX_ "Cloning substitution context is unimplemented");
13893 ncx->blk_oldcop = (COP*)any_dup(ncx->blk_oldcop, param->proto_perl);
13894 switch (CxTYPE(ncx)) {
13896 ncx->blk_sub.cv = (ncx->blk_sub.olddepth == 0
13897 ? cv_dup_inc(ncx->blk_sub.cv, param)
13898 : cv_dup(ncx->blk_sub.cv,param));
13899 if(CxHASARGS(ncx)){
13900 ncx->blk_sub.argarray = av_dup_inc(ncx->blk_sub.argarray,param);
13901 ncx->blk_sub.savearray = av_dup_inc(ncx->blk_sub.savearray,param);
13903 ncx->blk_sub.argarray = NULL;
13904 ncx->blk_sub.savearray = NULL;
13906 ncx->blk_sub.oldcomppad = (PAD*)ptr_table_fetch(PL_ptr_table,
13907 ncx->blk_sub.oldcomppad);
13910 ncx->blk_eval.old_namesv = sv_dup_inc(ncx->blk_eval.old_namesv,
13912 ncx->blk_eval.cur_text = sv_dup(ncx->blk_eval.cur_text, param);
13913 ncx->blk_eval.cv = cv_dup(ncx->blk_eval.cv, param);
13915 case CXt_LOOP_LAZYSV:
13916 ncx->blk_loop.state_u.lazysv.end
13917 = sv_dup_inc(ncx->blk_loop.state_u.lazysv.end, param);
13918 /* Fallthrough: duplicate lazysv.cur by using the ary.ary
13919 duplication code instead.
13920 We are taking advantage of (1) av_dup_inc and sv_dup_inc
13921 actually being the same function, and (2) order
13922 equivalence of the two unions.
13923 We can assert the later [but only at run time :-(] */
13924 assert ((void *) &ncx->blk_loop.state_u.ary.ary ==
13925 (void *) &ncx->blk_loop.state_u.lazysv.cur);
13928 ncx->blk_loop.state_u.ary.ary
13929 = av_dup_inc(ncx->blk_loop.state_u.ary.ary, param);
13931 case CXt_LOOP_LAZYIV:
13932 case CXt_LOOP_PLAIN:
13933 /* code common to all CXt_LOOP_* types */
13934 if (CxPADLOOP(ncx)) {
13935 ncx->blk_loop.itervar_u.oldcomppad
13936 = (PAD*)ptr_table_fetch(PL_ptr_table,
13937 ncx->blk_loop.itervar_u.oldcomppad);
13939 ncx->blk_loop.itervar_u.gv
13940 = gv_dup((const GV *)ncx->blk_loop.itervar_u.gv,
13945 ncx->blk_format.cv = cv_dup(ncx->blk_format.cv, param);
13946 ncx->blk_format.gv = gv_dup(ncx->blk_format.gv, param);
13947 ncx->blk_format.dfoutgv = gv_dup_inc(ncx->blk_format.dfoutgv,
13962 /* duplicate a stack info structure */
13965 Perl_si_dup(pTHX_ PERL_SI *si, CLONE_PARAMS* param)
13969 PERL_ARGS_ASSERT_SI_DUP;
13972 return (PERL_SI*)NULL;
13974 /* look for it in the table first */
13975 nsi = (PERL_SI*)ptr_table_fetch(PL_ptr_table, si);
13979 /* create anew and remember what it is */
13980 Newxz(nsi, 1, PERL_SI);
13981 ptr_table_store(PL_ptr_table, si, nsi);
13983 nsi->si_stack = av_dup_inc(si->si_stack, param);
13984 nsi->si_cxix = si->si_cxix;
13985 nsi->si_cxmax = si->si_cxmax;
13986 nsi->si_cxstack = cx_dup(si->si_cxstack, si->si_cxix, si->si_cxmax, param);
13987 nsi->si_type = si->si_type;
13988 nsi->si_prev = si_dup(si->si_prev, param);
13989 nsi->si_next = si_dup(si->si_next, param);
13990 nsi->si_markoff = si->si_markoff;
13995 #define POPINT(ss,ix) ((ss)[--(ix)].any_i32)
13996 #define TOPINT(ss,ix) ((ss)[ix].any_i32)
13997 #define POPLONG(ss,ix) ((ss)[--(ix)].any_long)
13998 #define TOPLONG(ss,ix) ((ss)[ix].any_long)
13999 #define POPIV(ss,ix) ((ss)[--(ix)].any_iv)
14000 #define TOPIV(ss,ix) ((ss)[ix].any_iv)
14001 #define POPUV(ss,ix) ((ss)[--(ix)].any_uv)
14002 #define TOPUV(ss,ix) ((ss)[ix].any_uv)
14003 #define POPBOOL(ss,ix) ((ss)[--(ix)].any_bool)
14004 #define TOPBOOL(ss,ix) ((ss)[ix].any_bool)
14005 #define POPPTR(ss,ix) ((ss)[--(ix)].any_ptr)
14006 #define TOPPTR(ss,ix) ((ss)[ix].any_ptr)
14007 #define POPDPTR(ss,ix) ((ss)[--(ix)].any_dptr)
14008 #define TOPDPTR(ss,ix) ((ss)[ix].any_dptr)
14009 #define POPDXPTR(ss,ix) ((ss)[--(ix)].any_dxptr)
14010 #define TOPDXPTR(ss,ix) ((ss)[ix].any_dxptr)
14013 #define pv_dup_inc(p) SAVEPV(p)
14014 #define pv_dup(p) SAVEPV(p)
14015 #define svp_dup_inc(p,pp) any_dup(p,pp)
14017 /* map any object to the new equivent - either something in the
14018 * ptr table, or something in the interpreter structure
14022 Perl_any_dup(pTHX_ void *v, const PerlInterpreter *proto_perl)
14026 PERL_ARGS_ASSERT_ANY_DUP;
14029 return (void*)NULL;
14031 /* look for it in the table first */
14032 ret = ptr_table_fetch(PL_ptr_table, v);
14036 /* see if it is part of the interpreter structure */
14037 if (v >= (void*)proto_perl && v < (void*)(proto_perl+1))
14038 ret = (void*)(((char*)aTHX) + (((char*)v) - (char*)proto_perl));
14046 /* duplicate the save stack */
14049 Perl_ss_dup(pTHX_ PerlInterpreter *proto_perl, CLONE_PARAMS* param)
14052 ANY * const ss = proto_perl->Isavestack;
14053 const I32 max = proto_perl->Isavestack_max;
14054 I32 ix = proto_perl->Isavestack_ix;
14067 void (*dptr) (void*);
14068 void (*dxptr) (pTHX_ void*);
14070 PERL_ARGS_ASSERT_SS_DUP;
14072 Newxz(nss, max, ANY);
14075 const UV uv = POPUV(ss,ix);
14076 const U8 type = (U8)uv & SAVE_MASK;
14078 TOPUV(nss,ix) = uv;
14080 case SAVEt_CLEARSV:
14081 case SAVEt_CLEARPADRANGE:
14083 case SAVEt_HELEM: /* hash element */
14084 case SAVEt_SV: /* scalar reference */
14085 sv = (const SV *)POPPTR(ss,ix);
14086 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14088 case SAVEt_ITEM: /* normal string */
14089 case SAVEt_GVSV: /* scalar slot in GV */
14090 sv = (const SV *)POPPTR(ss,ix);
14091 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14092 if (type == SAVEt_SV)
14096 case SAVEt_MORTALIZESV:
14097 case SAVEt_READONLY_OFF:
14098 sv = (const SV *)POPPTR(ss,ix);
14099 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14101 case SAVEt_FREEPADNAME:
14102 ptr = POPPTR(ss,ix);
14103 TOPPTR(nss,ix) = padname_dup((PADNAME *)ptr, param);
14104 PadnameREFCNT((PADNAME *)TOPPTR(nss,ix))++;
14106 case SAVEt_SHARED_PVREF: /* char* in shared space */
14107 c = (char*)POPPTR(ss,ix);
14108 TOPPTR(nss,ix) = savesharedpv(c);
14109 ptr = POPPTR(ss,ix);
14110 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14112 case SAVEt_GENERIC_SVREF: /* generic sv */
14113 case SAVEt_SVREF: /* scalar reference */
14114 sv = (const SV *)POPPTR(ss,ix);
14115 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14116 if (type == SAVEt_SVREF)
14117 SvREFCNT_inc_simple_void((SV *)TOPPTR(nss,ix));
14118 ptr = POPPTR(ss,ix);
14119 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14121 case SAVEt_GVSLOT: /* any slot in GV */
14122 sv = (const SV *)POPPTR(ss,ix);
14123 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14124 ptr = POPPTR(ss,ix);
14125 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14126 sv = (const SV *)POPPTR(ss,ix);
14127 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14129 case SAVEt_HV: /* hash reference */
14130 case SAVEt_AV: /* array reference */
14131 sv = (const SV *) POPPTR(ss,ix);
14132 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14134 case SAVEt_COMPPAD:
14136 sv = (const SV *) POPPTR(ss,ix);
14137 TOPPTR(nss,ix) = sv_dup(sv, param);
14139 case SAVEt_INT: /* int reference */
14140 ptr = POPPTR(ss,ix);
14141 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14142 intval = (int)POPINT(ss,ix);
14143 TOPINT(nss,ix) = intval;
14145 case SAVEt_LONG: /* long reference */
14146 ptr = POPPTR(ss,ix);
14147 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14148 longval = (long)POPLONG(ss,ix);
14149 TOPLONG(nss,ix) = longval;
14151 case SAVEt_I32: /* I32 reference */
14152 ptr = POPPTR(ss,ix);
14153 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14155 TOPINT(nss,ix) = i;
14157 case SAVEt_IV: /* IV reference */
14158 case SAVEt_STRLEN: /* STRLEN/size_t ref */
14159 ptr = POPPTR(ss,ix);
14160 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14162 TOPIV(nss,ix) = iv;
14164 case SAVEt_HPTR: /* HV* reference */
14165 case SAVEt_APTR: /* AV* reference */
14166 case SAVEt_SPTR: /* SV* reference */
14167 ptr = POPPTR(ss,ix);
14168 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14169 sv = (const SV *)POPPTR(ss,ix);
14170 TOPPTR(nss,ix) = sv_dup(sv, param);
14172 case SAVEt_VPTR: /* random* reference */
14173 ptr = POPPTR(ss,ix);
14174 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14176 case SAVEt_INT_SMALL:
14177 case SAVEt_I32_SMALL:
14178 case SAVEt_I16: /* I16 reference */
14179 case SAVEt_I8: /* I8 reference */
14181 ptr = POPPTR(ss,ix);
14182 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14184 case SAVEt_GENERIC_PVREF: /* generic char* */
14185 case SAVEt_PPTR: /* char* reference */
14186 ptr = POPPTR(ss,ix);
14187 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14188 c = (char*)POPPTR(ss,ix);
14189 TOPPTR(nss,ix) = pv_dup(c);
14191 case SAVEt_GP: /* scalar reference */
14192 gp = (GP*)POPPTR(ss,ix);
14193 TOPPTR(nss,ix) = gp = gp_dup(gp, param);
14194 (void)GpREFCNT_inc(gp);
14195 gv = (const GV *)POPPTR(ss,ix);
14196 TOPPTR(nss,ix) = gv_dup_inc(gv, param);
14199 ptr = POPPTR(ss,ix);
14200 if (ptr && (((OP*)ptr)->op_private & OPpREFCOUNTED)) {
14201 /* these are assumed to be refcounted properly */
14203 switch (((OP*)ptr)->op_type) {
14205 case OP_LEAVESUBLV:
14209 case OP_LEAVEWRITE:
14210 TOPPTR(nss,ix) = ptr;
14213 (void) OpREFCNT_inc(o);
14217 TOPPTR(nss,ix) = NULL;
14222 TOPPTR(nss,ix) = NULL;
14224 case SAVEt_FREECOPHH:
14225 ptr = POPPTR(ss,ix);
14226 TOPPTR(nss,ix) = cophh_copy((COPHH *)ptr);
14228 case SAVEt_ADELETE:
14229 av = (const AV *)POPPTR(ss,ix);
14230 TOPPTR(nss,ix) = av_dup_inc(av, param);
14232 TOPINT(nss,ix) = i;
14235 hv = (const HV *)POPPTR(ss,ix);
14236 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14238 TOPINT(nss,ix) = i;
14241 c = (char*)POPPTR(ss,ix);
14242 TOPPTR(nss,ix) = pv_dup_inc(c);
14244 case SAVEt_STACK_POS: /* Position on Perl stack */
14246 TOPINT(nss,ix) = i;
14248 case SAVEt_DESTRUCTOR:
14249 ptr = POPPTR(ss,ix);
14250 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14251 dptr = POPDPTR(ss,ix);
14252 TOPDPTR(nss,ix) = DPTR2FPTR(void (*)(void*),
14253 any_dup(FPTR2DPTR(void *, dptr),
14256 case SAVEt_DESTRUCTOR_X:
14257 ptr = POPPTR(ss,ix);
14258 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14259 dxptr = POPDXPTR(ss,ix);
14260 TOPDXPTR(nss,ix) = DPTR2FPTR(void (*)(pTHX_ void*),
14261 any_dup(FPTR2DPTR(void *, dxptr),
14264 case SAVEt_REGCONTEXT:
14266 ix -= uv >> SAVE_TIGHT_SHIFT;
14268 case SAVEt_AELEM: /* array element */
14269 sv = (const SV *)POPPTR(ss,ix);
14270 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14272 TOPINT(nss,ix) = i;
14273 av = (const AV *)POPPTR(ss,ix);
14274 TOPPTR(nss,ix) = av_dup_inc(av, param);
14277 ptr = POPPTR(ss,ix);
14278 TOPPTR(nss,ix) = ptr;
14281 ptr = POPPTR(ss,ix);
14282 ptr = cophh_copy((COPHH*)ptr);
14283 TOPPTR(nss,ix) = ptr;
14285 TOPINT(nss,ix) = i;
14286 if (i & HINT_LOCALIZE_HH) {
14287 hv = (const HV *)POPPTR(ss,ix);
14288 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14291 case SAVEt_PADSV_AND_MORTALIZE:
14292 longval = (long)POPLONG(ss,ix);
14293 TOPLONG(nss,ix) = longval;
14294 ptr = POPPTR(ss,ix);
14295 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14296 sv = (const SV *)POPPTR(ss,ix);
14297 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14299 case SAVEt_SET_SVFLAGS:
14301 TOPINT(nss,ix) = i;
14303 TOPINT(nss,ix) = i;
14304 sv = (const SV *)POPPTR(ss,ix);
14305 TOPPTR(nss,ix) = sv_dup(sv, param);
14307 case SAVEt_COMPILE_WARNINGS:
14308 ptr = POPPTR(ss,ix);
14309 TOPPTR(nss,ix) = DUP_WARNINGS((STRLEN*)ptr);
14312 ptr = POPPTR(ss,ix);
14313 TOPPTR(nss,ix) = parser_dup((const yy_parser*)ptr, param);
14317 "panic: ss_dup inconsistency (%"IVdf")", (IV) type);
14325 /* if sv is a stash, call $class->CLONE_SKIP(), and set the SVphv_CLONEABLE
14326 * flag to the result. This is done for each stash before cloning starts,
14327 * so we know which stashes want their objects cloned */
14330 do_mark_cloneable_stash(pTHX_ SV *const sv)
14332 const HEK * const hvname = HvNAME_HEK((const HV *)sv);
14334 GV* const cloner = gv_fetchmethod_autoload(MUTABLE_HV(sv), "CLONE_SKIP", 0);
14335 SvFLAGS(sv) |= SVphv_CLONEABLE; /* clone objects by default */
14336 if (cloner && GvCV(cloner)) {
14343 mXPUSHs(newSVhek(hvname));
14345 call_sv(MUTABLE_SV(GvCV(cloner)), G_SCALAR);
14352 SvFLAGS(sv) &= ~SVphv_CLONEABLE;
14360 =for apidoc perl_clone
14362 Create and return a new interpreter by cloning the current one.
14364 C<perl_clone> takes these flags as parameters:
14366 C<CLONEf_COPY_STACKS> - is used to, well, copy the stacks also,
14367 without it we only clone the data and zero the stacks,
14368 with it we copy the stacks and the new perl interpreter is
14369 ready to run at the exact same point as the previous one.
14370 The pseudo-fork code uses C<COPY_STACKS> while the
14371 threads->create doesn't.
14373 C<CLONEf_KEEP_PTR_TABLE> -
14374 C<perl_clone> keeps a ptr_table with the pointer of the old
14375 variable as a key and the new variable as a value,
14376 this allows it to check if something has been cloned and not
14377 clone it again but rather just use the value and increase the
14378 refcount. If C<KEEP_PTR_TABLE> is not set then C<perl_clone> will kill
14379 the ptr_table using the function
14380 C<ptr_table_free(PL_ptr_table); PL_ptr_table = NULL;>,
14381 reason to keep it around is if you want to dup some of your own
14382 variable who are outside the graph perl scans, an example of this
14383 code is in F<threads.xs> create.
14385 C<CLONEf_CLONE_HOST> -
14386 This is a win32 thing, it is ignored on unix, it tells perls
14387 win32host code (which is c++) to clone itself, this is needed on
14388 win32 if you want to run two threads at the same time,
14389 if you just want to do some stuff in a separate perl interpreter
14390 and then throw it away and return to the original one,
14391 you don't need to do anything.
14396 /* XXX the above needs expanding by someone who actually understands it ! */
14397 EXTERN_C PerlInterpreter *
14398 perl_clone_host(PerlInterpreter* proto_perl, UV flags);
14401 perl_clone(PerlInterpreter *proto_perl, UV flags)
14404 #ifdef PERL_IMPLICIT_SYS
14406 PERL_ARGS_ASSERT_PERL_CLONE;
14408 /* perlhost.h so we need to call into it
14409 to clone the host, CPerlHost should have a c interface, sky */
14411 #ifndef __amigaos4__
14412 if (flags & CLONEf_CLONE_HOST) {
14413 return perl_clone_host(proto_perl,flags);
14416 return perl_clone_using(proto_perl, flags,
14418 proto_perl->IMemShared,
14419 proto_perl->IMemParse,
14421 proto_perl->IStdIO,
14425 proto_perl->IProc);
14429 perl_clone_using(PerlInterpreter *proto_perl, UV flags,
14430 struct IPerlMem* ipM, struct IPerlMem* ipMS,
14431 struct IPerlMem* ipMP, struct IPerlEnv* ipE,
14432 struct IPerlStdIO* ipStd, struct IPerlLIO* ipLIO,
14433 struct IPerlDir* ipD, struct IPerlSock* ipS,
14434 struct IPerlProc* ipP)
14436 /* XXX many of the string copies here can be optimized if they're
14437 * constants; they need to be allocated as common memory and just
14438 * their pointers copied. */
14441 CLONE_PARAMS clone_params;
14442 CLONE_PARAMS* const param = &clone_params;
14444 PerlInterpreter * const my_perl = (PerlInterpreter*)(*ipM->pMalloc)(ipM, sizeof(PerlInterpreter));
14446 PERL_ARGS_ASSERT_PERL_CLONE_USING;
14447 #else /* !PERL_IMPLICIT_SYS */
14449 CLONE_PARAMS clone_params;
14450 CLONE_PARAMS* param = &clone_params;
14451 PerlInterpreter * const my_perl = (PerlInterpreter*)PerlMem_malloc(sizeof(PerlInterpreter));
14453 PERL_ARGS_ASSERT_PERL_CLONE;
14454 #endif /* PERL_IMPLICIT_SYS */
14456 /* for each stash, determine whether its objects should be cloned */
14457 S_visit(proto_perl, do_mark_cloneable_stash, SVt_PVHV, SVTYPEMASK);
14458 PERL_SET_THX(my_perl);
14461 PoisonNew(my_perl, 1, PerlInterpreter);
14464 PL_defstash = NULL; /* may be used by perl malloc() */
14467 PL_scopestack_name = 0;
14469 PL_savestack_ix = 0;
14470 PL_savestack_max = -1;
14471 PL_sig_pending = 0;
14473 Zero(&PL_debug_pad, 1, struct perl_debug_pad);
14474 Zero(&PL_padname_undef, 1, PADNAME);
14475 Zero(&PL_padname_const, 1, PADNAME);
14476 # ifdef DEBUG_LEAKING_SCALARS
14477 PL_sv_serial = (((UV)my_perl >> 2) & 0xfff) * 1000000;
14479 # ifdef PERL_TRACE_OPS
14480 Zero(PL_op_exec_cnt, OP_max+2, UV);
14482 #else /* !DEBUGGING */
14483 Zero(my_perl, 1, PerlInterpreter);
14484 #endif /* DEBUGGING */
14486 #ifdef PERL_IMPLICIT_SYS
14487 /* host pointers */
14489 PL_MemShared = ipMS;
14490 PL_MemParse = ipMP;
14497 #endif /* PERL_IMPLICIT_SYS */
14500 param->flags = flags;
14501 /* Nothing in the core code uses this, but we make it available to
14502 extensions (using mg_dup). */
14503 param->proto_perl = proto_perl;
14504 /* Likely nothing will use this, but it is initialised to be consistent
14505 with Perl_clone_params_new(). */
14506 param->new_perl = my_perl;
14507 param->unreferenced = NULL;
14510 INIT_TRACK_MEMPOOL(my_perl->Imemory_debug_header, my_perl);
14512 PL_body_arenas = NULL;
14513 Zero(&PL_body_roots, 1, PL_body_roots);
14517 PL_sv_arenaroot = NULL;
14519 PL_debug = proto_perl->Idebug;
14521 /* dbargs array probably holds garbage */
14524 PL_compiling = proto_perl->Icompiling;
14526 /* pseudo environmental stuff */
14527 PL_origargc = proto_perl->Iorigargc;
14528 PL_origargv = proto_perl->Iorigargv;
14530 #ifndef NO_TAINT_SUPPORT
14531 /* Set tainting stuff before PerlIO_debug can possibly get called */
14532 PL_tainting = proto_perl->Itainting;
14533 PL_taint_warn = proto_perl->Itaint_warn;
14535 PL_tainting = FALSE;
14536 PL_taint_warn = FALSE;
14539 PL_minus_c = proto_perl->Iminus_c;
14541 PL_localpatches = proto_perl->Ilocalpatches;
14542 PL_splitstr = proto_perl->Isplitstr;
14543 PL_minus_n = proto_perl->Iminus_n;
14544 PL_minus_p = proto_perl->Iminus_p;
14545 PL_minus_l = proto_perl->Iminus_l;
14546 PL_minus_a = proto_perl->Iminus_a;
14547 PL_minus_E = proto_perl->Iminus_E;
14548 PL_minus_F = proto_perl->Iminus_F;
14549 PL_doswitches = proto_perl->Idoswitches;
14550 PL_dowarn = proto_perl->Idowarn;
14551 #ifdef PERL_SAWAMPERSAND
14552 PL_sawampersand = proto_perl->Isawampersand;
14554 PL_unsafe = proto_perl->Iunsafe;
14555 PL_perldb = proto_perl->Iperldb;
14556 PL_perl_destruct_level = proto_perl->Iperl_destruct_level;
14557 PL_exit_flags = proto_perl->Iexit_flags;
14559 /* XXX time(&PL_basetime) when asked for? */
14560 PL_basetime = proto_perl->Ibasetime;
14562 PL_maxsysfd = proto_perl->Imaxsysfd;
14563 PL_statusvalue = proto_perl->Istatusvalue;
14565 PL_statusvalue_vms = proto_perl->Istatusvalue_vms;
14567 PL_statusvalue_posix = proto_perl->Istatusvalue_posix;
14570 /* RE engine related */
14571 PL_regmatch_slab = NULL;
14572 PL_reg_curpm = NULL;
14574 PL_sub_generation = proto_perl->Isub_generation;
14576 /* funky return mechanisms */
14577 PL_forkprocess = proto_perl->Iforkprocess;
14579 /* internal state */
14580 PL_maxo = proto_perl->Imaxo;
14582 PL_main_start = proto_perl->Imain_start;
14583 PL_eval_root = proto_perl->Ieval_root;
14584 PL_eval_start = proto_perl->Ieval_start;
14586 PL_filemode = proto_perl->Ifilemode;
14587 PL_lastfd = proto_perl->Ilastfd;
14588 PL_oldname = proto_perl->Ioldname; /* XXX not quite right */
14591 PL_gensym = proto_perl->Igensym;
14593 PL_laststatval = proto_perl->Ilaststatval;
14594 PL_laststype = proto_perl->Ilaststype;
14597 PL_profiledata = NULL;
14599 PL_generation = proto_perl->Igeneration;
14601 PL_in_clean_objs = proto_perl->Iin_clean_objs;
14602 PL_in_clean_all = proto_perl->Iin_clean_all;
14604 PL_delaymagic_uid = proto_perl->Idelaymagic_uid;
14605 PL_delaymagic_euid = proto_perl->Idelaymagic_euid;
14606 PL_delaymagic_gid = proto_perl->Idelaymagic_gid;
14607 PL_delaymagic_egid = proto_perl->Idelaymagic_egid;
14608 PL_nomemok = proto_perl->Inomemok;
14609 PL_an = proto_perl->Ian;
14610 PL_evalseq = proto_perl->Ievalseq;
14611 PL_origenviron = proto_perl->Iorigenviron; /* XXX not quite right */
14612 PL_origalen = proto_perl->Iorigalen;
14614 PL_sighandlerp = proto_perl->Isighandlerp;
14616 PL_runops = proto_perl->Irunops;
14618 PL_subline = proto_perl->Isubline;
14620 PL_cv_has_eval = proto_perl->Icv_has_eval;
14623 PL_cryptseen = proto_perl->Icryptseen;
14626 #ifdef USE_LOCALE_COLLATE
14627 PL_collation_ix = proto_perl->Icollation_ix;
14628 PL_collation_standard = proto_perl->Icollation_standard;
14629 PL_collxfrm_base = proto_perl->Icollxfrm_base;
14630 PL_collxfrm_mult = proto_perl->Icollxfrm_mult;
14631 #endif /* USE_LOCALE_COLLATE */
14633 #ifdef USE_LOCALE_NUMERIC
14634 PL_numeric_standard = proto_perl->Inumeric_standard;
14635 PL_numeric_local = proto_perl->Inumeric_local;
14636 #endif /* !USE_LOCALE_NUMERIC */
14638 /* Did the locale setup indicate UTF-8? */
14639 PL_utf8locale = proto_perl->Iutf8locale;
14640 PL_in_utf8_CTYPE_locale = proto_perl->Iin_utf8_CTYPE_locale;
14641 /* Unicode features (see perlrun/-C) */
14642 PL_unicode = proto_perl->Iunicode;
14644 /* Pre-5.8 signals control */
14645 PL_signals = proto_perl->Isignals;
14647 /* times() ticks per second */
14648 PL_clocktick = proto_perl->Iclocktick;
14650 /* Recursion stopper for PerlIO_find_layer */
14651 PL_in_load_module = proto_perl->Iin_load_module;
14653 /* sort() routine */
14654 PL_sort_RealCmp = proto_perl->Isort_RealCmp;
14656 /* Not really needed/useful since the reenrant_retint is "volatile",
14657 * but do it for consistency's sake. */
14658 PL_reentrant_retint = proto_perl->Ireentrant_retint;
14660 /* Hooks to shared SVs and locks. */
14661 PL_sharehook = proto_perl->Isharehook;
14662 PL_lockhook = proto_perl->Ilockhook;
14663 PL_unlockhook = proto_perl->Iunlockhook;
14664 PL_threadhook = proto_perl->Ithreadhook;
14665 PL_destroyhook = proto_perl->Idestroyhook;
14666 PL_signalhook = proto_perl->Isignalhook;
14668 PL_globhook = proto_perl->Iglobhook;
14671 PL_last_swash_hv = NULL; /* reinits on demand */
14672 PL_last_swash_klen = 0;
14673 PL_last_swash_key[0]= '\0';
14674 PL_last_swash_tmps = (U8*)NULL;
14675 PL_last_swash_slen = 0;
14677 PL_srand_called = proto_perl->Isrand_called;
14678 Copy(&(proto_perl->Irandom_state), &PL_random_state, 1, PL_RANDOM_STATE_TYPE);
14680 if (flags & CLONEf_COPY_STACKS) {
14681 /* next allocation will be PL_tmps_stack[PL_tmps_ix+1] */
14682 PL_tmps_ix = proto_perl->Itmps_ix;
14683 PL_tmps_max = proto_perl->Itmps_max;
14684 PL_tmps_floor = proto_perl->Itmps_floor;
14686 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
14687 * NOTE: unlike the others! */
14688 PL_scopestack_ix = proto_perl->Iscopestack_ix;
14689 PL_scopestack_max = proto_perl->Iscopestack_max;
14691 /* next SSPUSHFOO() sets PL_savestack[PL_savestack_ix]
14692 * NOTE: unlike the others! */
14693 PL_savestack_ix = proto_perl->Isavestack_ix;
14694 PL_savestack_max = proto_perl->Isavestack_max;
14697 PL_start_env = proto_perl->Istart_env; /* XXXXXX */
14698 PL_top_env = &PL_start_env;
14700 PL_op = proto_perl->Iop;
14703 PL_Xpv = (XPV*)NULL;
14704 my_perl->Ina = proto_perl->Ina;
14706 PL_statbuf = proto_perl->Istatbuf;
14707 PL_statcache = proto_perl->Istatcache;
14709 #ifndef NO_TAINT_SUPPORT
14710 PL_tainted = proto_perl->Itainted;
14712 PL_tainted = FALSE;
14714 PL_curpm = proto_perl->Icurpm; /* XXX No PMOP ref count */
14716 PL_chopset = proto_perl->Ichopset; /* XXX never deallocated */
14718 PL_restartjmpenv = proto_perl->Irestartjmpenv;
14719 PL_restartop = proto_perl->Irestartop;
14720 PL_in_eval = proto_perl->Iin_eval;
14721 PL_delaymagic = proto_perl->Idelaymagic;
14722 PL_phase = proto_perl->Iphase;
14723 PL_localizing = proto_perl->Ilocalizing;
14725 PL_hv_fetch_ent_mh = NULL;
14726 PL_modcount = proto_perl->Imodcount;
14727 PL_lastgotoprobe = NULL;
14728 PL_dumpindent = proto_perl->Idumpindent;
14730 PL_efloatbuf = NULL; /* reinits on demand */
14731 PL_efloatsize = 0; /* reinits on demand */
14735 PL_colorset = 0; /* reinits PL_colors[] */
14736 /*PL_colors[6] = {0,0,0,0,0,0};*/
14738 /* Pluggable optimizer */
14739 PL_peepp = proto_perl->Ipeepp;
14740 PL_rpeepp = proto_perl->Irpeepp;
14741 /* op_free() hook */
14742 PL_opfreehook = proto_perl->Iopfreehook;
14744 #ifdef USE_REENTRANT_API
14745 /* XXX: things like -Dm will segfault here in perlio, but doing
14746 * PERL_SET_CONTEXT(proto_perl);
14747 * breaks too many other things
14749 Perl_reentrant_init(aTHX);
14752 /* create SV map for pointer relocation */
14753 PL_ptr_table = ptr_table_new();
14755 /* initialize these special pointers as early as possible */
14757 ptr_table_store(PL_ptr_table, &proto_perl->Isv_undef, &PL_sv_undef);
14758 ptr_table_store(PL_ptr_table, &proto_perl->Isv_no, &PL_sv_no);
14759 ptr_table_store(PL_ptr_table, &proto_perl->Isv_yes, &PL_sv_yes);
14760 ptr_table_store(PL_ptr_table, &proto_perl->Ipadname_const,
14761 &PL_padname_const);
14763 /* create (a non-shared!) shared string table */
14764 PL_strtab = newHV();
14765 HvSHAREKEYS_off(PL_strtab);
14766 hv_ksplit(PL_strtab, HvTOTALKEYS(proto_perl->Istrtab));
14767 ptr_table_store(PL_ptr_table, proto_perl->Istrtab, PL_strtab);
14769 Zero(PL_sv_consts, SV_CONSTS_COUNT, SV*);
14771 /* This PV will be free'd special way so must set it same way op.c does */
14772 PL_compiling.cop_file = savesharedpv(PL_compiling.cop_file);
14773 ptr_table_store(PL_ptr_table, proto_perl->Icompiling.cop_file, PL_compiling.cop_file);
14775 ptr_table_store(PL_ptr_table, &proto_perl->Icompiling, &PL_compiling);
14776 PL_compiling.cop_warnings = DUP_WARNINGS(PL_compiling.cop_warnings);
14777 CopHINTHASH_set(&PL_compiling, cophh_copy(CopHINTHASH_get(&PL_compiling)));
14778 PL_curcop = (COP*)any_dup(proto_perl->Icurcop, proto_perl);
14780 param->stashes = newAV(); /* Setup array of objects to call clone on */
14781 /* This makes no difference to the implementation, as it always pushes
14782 and shifts pointers to other SVs without changing their reference
14783 count, with the array becoming empty before it is freed. However, it
14784 makes it conceptually clear what is going on, and will avoid some
14785 work inside av.c, filling slots between AvFILL() and AvMAX() with
14786 &PL_sv_undef, and SvREFCNT_dec()ing those. */
14787 AvREAL_off(param->stashes);
14789 if (!(flags & CLONEf_COPY_STACKS)) {
14790 param->unreferenced = newAV();
14793 #ifdef PERLIO_LAYERS
14794 /* Clone PerlIO tables as soon as we can handle general xx_dup() */
14795 PerlIO_clone(aTHX_ proto_perl, param);
14798 PL_envgv = gv_dup_inc(proto_perl->Ienvgv, param);
14799 PL_incgv = gv_dup_inc(proto_perl->Iincgv, param);
14800 PL_hintgv = gv_dup_inc(proto_perl->Ihintgv, param);
14801 PL_origfilename = SAVEPV(proto_perl->Iorigfilename);
14802 PL_xsubfilename = proto_perl->Ixsubfilename;
14803 PL_diehook = sv_dup_inc(proto_perl->Idiehook, param);
14804 PL_warnhook = sv_dup_inc(proto_perl->Iwarnhook, param);
14807 PL_patchlevel = sv_dup_inc(proto_perl->Ipatchlevel, param);
14808 PL_inplace = SAVEPV(proto_perl->Iinplace);
14809 PL_e_script = sv_dup_inc(proto_perl->Ie_script, param);
14811 /* magical thingies */
14813 PL_encoding = sv_dup(proto_perl->Iencoding, param);
14814 PL_lex_encoding = sv_dup(proto_perl->Ilex_encoding, param);
14816 sv_setpvs(PERL_DEBUG_PAD(0), ""); /* For regex debugging. */
14817 sv_setpvs(PERL_DEBUG_PAD(1), ""); /* ext/re needs these */
14818 sv_setpvs(PERL_DEBUG_PAD(2), ""); /* even without DEBUGGING. */
14821 /* Clone the regex array */
14822 /* ORANGE FIXME for plugins, probably in the SV dup code.
14823 newSViv(PTR2IV(CALLREGDUPE(
14824 INT2PTR(REGEXP *, SvIVX(regex)), param))))
14826 PL_regex_padav = av_dup_inc(proto_perl->Iregex_padav, param);
14827 PL_regex_pad = AvARRAY(PL_regex_padav);
14829 PL_stashpadmax = proto_perl->Istashpadmax;
14830 PL_stashpadix = proto_perl->Istashpadix ;
14831 Newx(PL_stashpad, PL_stashpadmax, HV *);
14834 for (; o < PL_stashpadmax; ++o)
14835 PL_stashpad[o] = hv_dup(proto_perl->Istashpad[o], param);
14838 /* shortcuts to various I/O objects */
14839 PL_ofsgv = gv_dup_inc(proto_perl->Iofsgv, param);
14840 PL_stdingv = gv_dup(proto_perl->Istdingv, param);
14841 PL_stderrgv = gv_dup(proto_perl->Istderrgv, param);
14842 PL_defgv = gv_dup(proto_perl->Idefgv, param);
14843 PL_argvgv = gv_dup_inc(proto_perl->Iargvgv, param);
14844 PL_argvoutgv = gv_dup(proto_perl->Iargvoutgv, param);
14845 PL_argvout_stack = av_dup_inc(proto_perl->Iargvout_stack, param);
14847 /* shortcuts to regexp stuff */
14848 PL_replgv = gv_dup_inc(proto_perl->Ireplgv, param);
14850 /* shortcuts to misc objects */
14851 PL_errgv = gv_dup(proto_perl->Ierrgv, param);
14853 /* shortcuts to debugging objects */
14854 PL_DBgv = gv_dup_inc(proto_perl->IDBgv, param);
14855 PL_DBline = gv_dup_inc(proto_perl->IDBline, param);
14856 PL_DBsub = gv_dup_inc(proto_perl->IDBsub, param);
14857 PL_DBsingle = sv_dup(proto_perl->IDBsingle, param);
14858 PL_DBtrace = sv_dup(proto_perl->IDBtrace, param);
14859 PL_DBsignal = sv_dup(proto_perl->IDBsignal, param);
14860 Copy(proto_perl->IDBcontrol, PL_DBcontrol, DBVARMG_COUNT, IV);
14862 /* symbol tables */
14863 PL_defstash = hv_dup_inc(proto_perl->Idefstash, param);
14864 PL_curstash = hv_dup_inc(proto_perl->Icurstash, param);
14865 PL_debstash = hv_dup(proto_perl->Idebstash, param);
14866 PL_globalstash = hv_dup(proto_perl->Iglobalstash, param);
14867 PL_curstname = sv_dup_inc(proto_perl->Icurstname, param);
14869 PL_beginav = av_dup_inc(proto_perl->Ibeginav, param);
14870 PL_beginav_save = av_dup_inc(proto_perl->Ibeginav_save, param);
14871 PL_checkav_save = av_dup_inc(proto_perl->Icheckav_save, param);
14872 PL_unitcheckav = av_dup_inc(proto_perl->Iunitcheckav, param);
14873 PL_unitcheckav_save = av_dup_inc(proto_perl->Iunitcheckav_save, param);
14874 PL_endav = av_dup_inc(proto_perl->Iendav, param);
14875 PL_checkav = av_dup_inc(proto_perl->Icheckav, param);
14876 PL_initav = av_dup_inc(proto_perl->Iinitav, param);
14877 PL_savebegin = proto_perl->Isavebegin;
14879 PL_isarev = hv_dup_inc(proto_perl->Iisarev, param);
14881 /* subprocess state */
14882 PL_fdpid = av_dup_inc(proto_perl->Ifdpid, param);
14884 if (proto_perl->Iop_mask)
14885 PL_op_mask = SAVEPVN(proto_perl->Iop_mask, PL_maxo);
14888 /* PL_asserting = proto_perl->Iasserting; */
14890 /* current interpreter roots */
14891 PL_main_cv = cv_dup_inc(proto_perl->Imain_cv, param);
14893 PL_main_root = OpREFCNT_inc(proto_perl->Imain_root);
14896 /* runtime control stuff */
14897 PL_curcopdb = (COP*)any_dup(proto_perl->Icurcopdb, proto_perl);
14899 PL_preambleav = av_dup_inc(proto_perl->Ipreambleav, param);
14901 PL_ors_sv = sv_dup_inc(proto_perl->Iors_sv, param);
14903 /* interpreter atexit processing */
14904 PL_exitlistlen = proto_perl->Iexitlistlen;
14905 if (PL_exitlistlen) {
14906 Newx(PL_exitlist, PL_exitlistlen, PerlExitListEntry);
14907 Copy(proto_perl->Iexitlist, PL_exitlist, PL_exitlistlen, PerlExitListEntry);
14910 PL_exitlist = (PerlExitListEntry*)NULL;
14912 PL_my_cxt_size = proto_perl->Imy_cxt_size;
14913 if (PL_my_cxt_size) {
14914 Newx(PL_my_cxt_list, PL_my_cxt_size, void *);
14915 Copy(proto_perl->Imy_cxt_list, PL_my_cxt_list, PL_my_cxt_size, void *);
14916 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
14917 Newx(PL_my_cxt_keys, PL_my_cxt_size, const char *);
14918 Copy(proto_perl->Imy_cxt_keys, PL_my_cxt_keys, PL_my_cxt_size, char *);
14922 PL_my_cxt_list = (void**)NULL;
14923 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
14924 PL_my_cxt_keys = (const char**)NULL;
14927 PL_modglobal = hv_dup_inc(proto_perl->Imodglobal, param);
14928 PL_custom_op_names = hv_dup_inc(proto_perl->Icustom_op_names,param);
14929 PL_custom_op_descs = hv_dup_inc(proto_perl->Icustom_op_descs,param);
14930 PL_custom_ops = hv_dup_inc(proto_perl->Icustom_ops, param);
14932 PL_compcv = cv_dup(proto_perl->Icompcv, param);
14934 PAD_CLONE_VARS(proto_perl, param);
14936 #ifdef HAVE_INTERP_INTERN
14937 sys_intern_dup(&proto_perl->Isys_intern, &PL_sys_intern);
14940 PL_DBcv = cv_dup(proto_perl->IDBcv, param);
14942 #ifdef PERL_USES_PL_PIDSTATUS
14943 PL_pidstatus = newHV(); /* XXX flag for cloning? */
14945 PL_osname = SAVEPV(proto_perl->Iosname);
14946 PL_parser = parser_dup(proto_perl->Iparser, param);
14948 /* XXX this only works if the saved cop has already been cloned */
14949 if (proto_perl->Iparser) {
14950 PL_parser->saved_curcop = (COP*)any_dup(
14951 proto_perl->Iparser->saved_curcop,
14955 PL_subname = sv_dup_inc(proto_perl->Isubname, param);
14957 #ifdef USE_LOCALE_CTYPE
14958 /* Should we warn if uses locale? */
14959 PL_warn_locale = sv_dup_inc(proto_perl->Iwarn_locale, param);
14962 #ifdef USE_LOCALE_COLLATE
14963 PL_collation_name = SAVEPV(proto_perl->Icollation_name);
14964 #endif /* USE_LOCALE_COLLATE */
14966 #ifdef USE_LOCALE_NUMERIC
14967 PL_numeric_name = SAVEPV(proto_perl->Inumeric_name);
14968 PL_numeric_radix_sv = sv_dup_inc(proto_perl->Inumeric_radix_sv, param);
14969 #endif /* !USE_LOCALE_NUMERIC */
14971 /* Unicode inversion lists */
14972 PL_Latin1 = sv_dup_inc(proto_perl->ILatin1, param);
14973 PL_UpperLatin1 = sv_dup_inc(proto_perl->IUpperLatin1, param);
14974 PL_AboveLatin1 = sv_dup_inc(proto_perl->IAboveLatin1, param);
14975 PL_InBitmap = sv_dup_inc(proto_perl->IInBitmap, param);
14977 PL_NonL1NonFinalFold = sv_dup_inc(proto_perl->INonL1NonFinalFold, param);
14978 PL_HasMultiCharFold = sv_dup_inc(proto_perl->IHasMultiCharFold, param);
14980 /* utf8 character class swashes */
14981 for (i = 0; i < POSIX_SWASH_COUNT; i++) {
14982 PL_utf8_swash_ptrs[i] = sv_dup_inc(proto_perl->Iutf8_swash_ptrs[i], param);
14984 for (i = 0; i < POSIX_CC_COUNT; i++) {
14985 PL_XPosix_ptrs[i] = sv_dup_inc(proto_perl->IXPosix_ptrs[i], param);
14987 PL_GCB_invlist = sv_dup_inc(proto_perl->IGCB_invlist, param);
14988 PL_SB_invlist = sv_dup_inc(proto_perl->ISB_invlist, param);
14989 PL_WB_invlist = sv_dup_inc(proto_perl->IWB_invlist, param);
14990 PL_utf8_mark = sv_dup_inc(proto_perl->Iutf8_mark, param);
14991 PL_utf8_toupper = sv_dup_inc(proto_perl->Iutf8_toupper, param);
14992 PL_utf8_totitle = sv_dup_inc(proto_perl->Iutf8_totitle, param);
14993 PL_utf8_tolower = sv_dup_inc(proto_perl->Iutf8_tolower, param);
14994 PL_utf8_tofold = sv_dup_inc(proto_perl->Iutf8_tofold, param);
14995 PL_utf8_idstart = sv_dup_inc(proto_perl->Iutf8_idstart, param);
14996 PL_utf8_xidstart = sv_dup_inc(proto_perl->Iutf8_xidstart, param);
14997 PL_utf8_perl_idstart = sv_dup_inc(proto_perl->Iutf8_perl_idstart, param);
14998 PL_utf8_perl_idcont = sv_dup_inc(proto_perl->Iutf8_perl_idcont, param);
14999 PL_utf8_idcont = sv_dup_inc(proto_perl->Iutf8_idcont, param);
15000 PL_utf8_xidcont = sv_dup_inc(proto_perl->Iutf8_xidcont, param);
15001 PL_utf8_foldable = sv_dup_inc(proto_perl->Iutf8_foldable, param);
15002 PL_utf8_charname_begin = sv_dup_inc(proto_perl->Iutf8_charname_begin, param);
15003 PL_utf8_charname_continue = sv_dup_inc(proto_perl->Iutf8_charname_continue, param);
15005 if (proto_perl->Ipsig_pend) {
15006 Newxz(PL_psig_pend, SIG_SIZE, int);
15009 PL_psig_pend = (int*)NULL;
15012 if (proto_perl->Ipsig_name) {
15013 Newx(PL_psig_name, 2 * SIG_SIZE, SV*);
15014 sv_dup_inc_multiple(proto_perl->Ipsig_name, PL_psig_name, 2 * SIG_SIZE,
15016 PL_psig_ptr = PL_psig_name + SIG_SIZE;
15019 PL_psig_ptr = (SV**)NULL;
15020 PL_psig_name = (SV**)NULL;
15023 if (flags & CLONEf_COPY_STACKS) {
15024 Newx(PL_tmps_stack, PL_tmps_max, SV*);
15025 sv_dup_inc_multiple(proto_perl->Itmps_stack, PL_tmps_stack,
15026 PL_tmps_ix+1, param);
15028 /* next PUSHMARK() sets *(PL_markstack_ptr+1) */
15029 i = proto_perl->Imarkstack_max - proto_perl->Imarkstack;
15030 Newxz(PL_markstack, i, I32);
15031 PL_markstack_max = PL_markstack + (proto_perl->Imarkstack_max
15032 - proto_perl->Imarkstack);
15033 PL_markstack_ptr = PL_markstack + (proto_perl->Imarkstack_ptr
15034 - proto_perl->Imarkstack);
15035 Copy(proto_perl->Imarkstack, PL_markstack,
15036 PL_markstack_ptr - PL_markstack + 1, I32);
15038 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
15039 * NOTE: unlike the others! */
15040 Newxz(PL_scopestack, PL_scopestack_max, I32);
15041 Copy(proto_perl->Iscopestack, PL_scopestack, PL_scopestack_ix, I32);
15044 Newxz(PL_scopestack_name, PL_scopestack_max, const char *);
15045 Copy(proto_perl->Iscopestack_name, PL_scopestack_name, PL_scopestack_ix, const char *);
15047 /* reset stack AV to correct length before its duped via
15048 * PL_curstackinfo */
15049 AvFILLp(proto_perl->Icurstack) =
15050 proto_perl->Istack_sp - proto_perl->Istack_base;
15052 /* NOTE: si_dup() looks at PL_markstack */
15053 PL_curstackinfo = si_dup(proto_perl->Icurstackinfo, param);
15055 /* PL_curstack = PL_curstackinfo->si_stack; */
15056 PL_curstack = av_dup(proto_perl->Icurstack, param);
15057 PL_mainstack = av_dup(proto_perl->Imainstack, param);
15059 /* next PUSHs() etc. set *(PL_stack_sp+1) */
15060 PL_stack_base = AvARRAY(PL_curstack);
15061 PL_stack_sp = PL_stack_base + (proto_perl->Istack_sp
15062 - proto_perl->Istack_base);
15063 PL_stack_max = PL_stack_base + AvMAX(PL_curstack);
15065 /*Newxz(PL_savestack, PL_savestack_max, ANY);*/
15066 PL_savestack = ss_dup(proto_perl, param);
15070 ENTER; /* perl_destruct() wants to LEAVE; */
15073 PL_statgv = gv_dup(proto_perl->Istatgv, param);
15074 PL_statname = sv_dup_inc(proto_perl->Istatname, param);
15076 PL_rs = sv_dup_inc(proto_perl->Irs, param);
15077 PL_last_in_gv = gv_dup(proto_perl->Ilast_in_gv, param);
15078 PL_defoutgv = gv_dup_inc(proto_perl->Idefoutgv, param);
15079 PL_toptarget = sv_dup_inc(proto_perl->Itoptarget, param);
15080 PL_bodytarget = sv_dup_inc(proto_perl->Ibodytarget, param);
15081 PL_formtarget = sv_dup(proto_perl->Iformtarget, param);
15083 PL_errors = sv_dup_inc(proto_perl->Ierrors, param);
15085 PL_sortcop = (OP*)any_dup(proto_perl->Isortcop, proto_perl);
15086 PL_firstgv = gv_dup_inc(proto_perl->Ifirstgv, param);
15087 PL_secondgv = gv_dup_inc(proto_perl->Isecondgv, param);
15089 PL_stashcache = newHV();
15091 PL_watchaddr = (char **) ptr_table_fetch(PL_ptr_table,
15092 proto_perl->Iwatchaddr);
15093 PL_watchok = PL_watchaddr ? * PL_watchaddr : NULL;
15094 if (PL_debug && PL_watchaddr) {
15095 PerlIO_printf(Perl_debug_log,
15096 "WATCHING: %"UVxf" cloned as %"UVxf" with value %"UVxf"\n",
15097 PTR2UV(proto_perl->Iwatchaddr), PTR2UV(PL_watchaddr),
15098 PTR2UV(PL_watchok));
15101 PL_registered_mros = hv_dup_inc(proto_perl->Iregistered_mros, param);
15102 PL_blockhooks = av_dup_inc(proto_perl->Iblockhooks, param);
15103 PL_utf8_foldclosures = hv_dup_inc(proto_perl->Iutf8_foldclosures, param);
15105 /* Call the ->CLONE method, if it exists, for each of the stashes
15106 identified by sv_dup() above.
15108 while(av_tindex(param->stashes) != -1) {
15109 HV* const stash = MUTABLE_HV(av_shift(param->stashes));
15110 GV* const cloner = gv_fetchmethod_autoload(stash, "CLONE", 0);
15111 if (cloner && GvCV(cloner)) {
15116 mXPUSHs(newSVhek(HvNAME_HEK(stash)));
15118 call_sv(MUTABLE_SV(GvCV(cloner)), G_DISCARD);
15124 if (!(flags & CLONEf_KEEP_PTR_TABLE)) {
15125 ptr_table_free(PL_ptr_table);
15126 PL_ptr_table = NULL;
15129 if (!(flags & CLONEf_COPY_STACKS)) {
15130 unreferenced_to_tmp_stack(param->unreferenced);
15133 SvREFCNT_dec(param->stashes);
15135 /* orphaned? eg threads->new inside BEGIN or use */
15136 if (PL_compcv && ! SvREFCNT(PL_compcv)) {
15137 SvREFCNT_inc_simple_void(PL_compcv);
15138 SAVEFREESV(PL_compcv);
15145 S_unreferenced_to_tmp_stack(pTHX_ AV *const unreferenced)
15147 PERL_ARGS_ASSERT_UNREFERENCED_TO_TMP_STACK;
15149 if (AvFILLp(unreferenced) > -1) {
15150 SV **svp = AvARRAY(unreferenced);
15151 SV **const last = svp + AvFILLp(unreferenced);
15155 if (SvREFCNT(*svp) == 1)
15157 } while (++svp <= last);
15159 EXTEND_MORTAL(count);
15160 svp = AvARRAY(unreferenced);
15163 if (SvREFCNT(*svp) == 1) {
15164 /* Our reference is the only one to this SV. This means that
15165 in this thread, the scalar effectively has a 0 reference.
15166 That doesn't work (cleanup never happens), so donate our
15167 reference to it onto the save stack. */
15168 PL_tmps_stack[++PL_tmps_ix] = *svp;
15170 /* As an optimisation, because we are already walking the
15171 entire array, instead of above doing either
15172 SvREFCNT_inc(*svp) or *svp = &PL_sv_undef, we can instead
15173 release our reference to the scalar, so that at the end of
15174 the array owns zero references to the scalars it happens to
15175 point to. We are effectively converting the array from
15176 AvREAL() on to AvREAL() off. This saves the av_clear()
15177 (triggered by the SvREFCNT_dec(unreferenced) below) from
15178 walking the array a second time. */
15179 SvREFCNT_dec(*svp);
15182 } while (++svp <= last);
15183 AvREAL_off(unreferenced);
15185 SvREFCNT_dec_NN(unreferenced);
15189 Perl_clone_params_del(CLONE_PARAMS *param)
15191 /* This seemingly funky ordering keeps the build with PERL_GLOBAL_STRUCT
15193 PerlInterpreter *const to = param->new_perl;
15195 PerlInterpreter *const was = PERL_GET_THX;
15197 PERL_ARGS_ASSERT_CLONE_PARAMS_DEL;
15203 SvREFCNT_dec(param->stashes);
15204 if (param->unreferenced)
15205 unreferenced_to_tmp_stack(param->unreferenced);
15215 Perl_clone_params_new(PerlInterpreter *const from, PerlInterpreter *const to)
15218 /* Need to play this game, as newAV() can call safesysmalloc(), and that
15219 does a dTHX; to get the context from thread local storage.
15220 FIXME - under PERL_CORE Newx(), Safefree() and friends should expand to
15221 a version that passes in my_perl. */
15222 PerlInterpreter *const was = PERL_GET_THX;
15223 CLONE_PARAMS *param;
15225 PERL_ARGS_ASSERT_CLONE_PARAMS_NEW;
15231 /* Given that we've set the context, we can do this unshared. */
15232 Newx(param, 1, CLONE_PARAMS);
15235 param->proto_perl = from;
15236 param->new_perl = to;
15237 param->stashes = (AV *)Perl_newSV_type(to, SVt_PVAV);
15238 AvREAL_off(param->stashes);
15239 param->unreferenced = (AV *)Perl_newSV_type(to, SVt_PVAV);
15247 #endif /* USE_ITHREADS */
15250 Perl_init_constants(pTHX)
15252 SvREFCNT(&PL_sv_undef) = SvREFCNT_IMMORTAL;
15253 SvFLAGS(&PL_sv_undef) = SVf_READONLY|SVf_PROTECT|SVt_NULL;
15254 SvANY(&PL_sv_undef) = NULL;
15256 SvANY(&PL_sv_no) = new_XPVNV();
15257 SvREFCNT(&PL_sv_no) = SvREFCNT_IMMORTAL;
15258 SvFLAGS(&PL_sv_no) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15259 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15262 SvANY(&PL_sv_yes) = new_XPVNV();
15263 SvREFCNT(&PL_sv_yes) = SvREFCNT_IMMORTAL;
15264 SvFLAGS(&PL_sv_yes) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15265 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15268 SvPV_set(&PL_sv_no, (char*)PL_No);
15269 SvCUR_set(&PL_sv_no, 0);
15270 SvLEN_set(&PL_sv_no, 0);
15271 SvIV_set(&PL_sv_no, 0);
15272 SvNV_set(&PL_sv_no, 0);
15274 SvPV_set(&PL_sv_yes, (char*)PL_Yes);
15275 SvCUR_set(&PL_sv_yes, 1);
15276 SvLEN_set(&PL_sv_yes, 0);
15277 SvIV_set(&PL_sv_yes, 1);
15278 SvNV_set(&PL_sv_yes, 1);
15280 PadnamePV(&PL_padname_const) = (char *)PL_No;
15284 =head1 Unicode Support
15286 =for apidoc sv_recode_to_utf8
15288 C<encoding> is assumed to be an C<Encode> object, on entry the PV
15289 of C<sv> is assumed to be octets in that encoding, and C<sv>
15290 will be converted into Unicode (and UTF-8).
15292 If C<sv> already is UTF-8 (or if it is not C<POK>), or if C<encoding>
15293 is not a reference, nothing is done to C<sv>. If C<encoding> is not
15294 an C<Encode::XS> Encoding object, bad things will happen.
15295 (See F<lib/encoding.pm> and L<Encode>.)
15297 The PV of C<sv> is returned.
15302 Perl_sv_recode_to_utf8(pTHX_ SV *sv, SV *encoding)
15304 PERL_ARGS_ASSERT_SV_RECODE_TO_UTF8;
15306 if (SvPOK(sv) && !SvUTF8(sv) && !IN_BYTES && SvROK(encoding)) {
15315 if (SvPADTMP(nsv)) {
15316 nsv = sv_newmortal();
15317 SvSetSV_nosteal(nsv, sv);
15326 Passing sv_yes is wrong - it needs to be or'ed set of constants
15327 for Encode::XS, while UTf-8 decode (currently) assumes a true value means
15328 remove converted chars from source.
15330 Both will default the value - let them.
15332 XPUSHs(&PL_sv_yes);
15335 call_method("decode", G_SCALAR);
15339 s = SvPV_const(uni, len);
15340 if (s != SvPVX_const(sv)) {
15341 SvGROW(sv, len + 1);
15342 Move(s, SvPVX(sv), len + 1, char);
15343 SvCUR_set(sv, len);
15348 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
15349 /* clear pos and any utf8 cache */
15350 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
15353 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
15354 magic_setutf8(sv,mg); /* clear UTF8 cache */
15359 return SvPOKp(sv) ? SvPVX(sv) : NULL;
15363 =for apidoc sv_cat_decode
15365 C<encoding> is assumed to be an C<Encode> object, the PV of C<ssv> is
15366 assumed to be octets in that encoding and decoding the input starts
15367 from the position which S<C<(PV + *offset)>> pointed to. C<dsv> will be
15368 concatenated with the decoded UTF-8 string from C<ssv>. Decoding will terminate
15369 when the string C<tstr> appears in decoding output or the input ends on
15370 the PV of C<ssv>. The value which C<offset> points will be modified
15371 to the last input position on C<ssv>.
15373 Returns TRUE if the terminator was found, else returns FALSE.
15378 Perl_sv_cat_decode(pTHX_ SV *dsv, SV *encoding,
15379 SV *ssv, int *offset, char *tstr, int tlen)
15383 PERL_ARGS_ASSERT_SV_CAT_DECODE;
15385 if (SvPOK(ssv) && SvPOK(dsv) && SvROK(encoding)) {
15396 offsv = newSViv(*offset);
15398 mPUSHp(tstr, tlen);
15400 call_method("cat_decode", G_SCALAR);
15402 ret = SvTRUE(TOPs);
15403 *offset = SvIV(offsv);
15409 Perl_croak(aTHX_ "Invalid argument to sv_cat_decode");
15414 /* ---------------------------------------------------------------------
15416 * support functions for report_uninit()
15419 /* the maxiumum size of array or hash where we will scan looking
15420 * for the undefined element that triggered the warning */
15422 #define FUV_MAX_SEARCH_SIZE 1000
15424 /* Look for an entry in the hash whose value has the same SV as val;
15425 * If so, return a mortal copy of the key. */
15428 S_find_hash_subscript(pTHX_ const HV *const hv, const SV *const val)
15434 PERL_ARGS_ASSERT_FIND_HASH_SUBSCRIPT;
15436 if (!hv || SvMAGICAL(hv) || !HvARRAY(hv) ||
15437 (HvTOTALKEYS(hv) > FUV_MAX_SEARCH_SIZE))
15440 array = HvARRAY(hv);
15442 for (i=HvMAX(hv); i>=0; i--) {
15444 for (entry = array[i]; entry; entry = HeNEXT(entry)) {
15445 if (HeVAL(entry) != val)
15447 if ( HeVAL(entry) == &PL_sv_undef ||
15448 HeVAL(entry) == &PL_sv_placeholder)
15452 if (HeKLEN(entry) == HEf_SVKEY)
15453 return sv_mortalcopy(HeKEY_sv(entry));
15454 return sv_2mortal(newSVhek(HeKEY_hek(entry)));
15460 /* Look for an entry in the array whose value has the same SV as val;
15461 * If so, return the index, otherwise return -1. */
15464 S_find_array_subscript(pTHX_ const AV *const av, const SV *const val)
15466 PERL_ARGS_ASSERT_FIND_ARRAY_SUBSCRIPT;
15468 if (!av || SvMAGICAL(av) || !AvARRAY(av) ||
15469 (AvFILLp(av) > FUV_MAX_SEARCH_SIZE))
15472 if (val != &PL_sv_undef) {
15473 SV ** const svp = AvARRAY(av);
15476 for (i=AvFILLp(av); i>=0; i--)
15483 /* varname(): return the name of a variable, optionally with a subscript.
15484 * If gv is non-zero, use the name of that global, along with gvtype (one
15485 * of "$", "@", "%"); otherwise use the name of the lexical at pad offset
15486 * targ. Depending on the value of the subscript_type flag, return:
15489 #define FUV_SUBSCRIPT_NONE 1 /* "@foo" */
15490 #define FUV_SUBSCRIPT_ARRAY 2 /* "$foo[aindex]" */
15491 #define FUV_SUBSCRIPT_HASH 3 /* "$foo{keyname}" */
15492 #define FUV_SUBSCRIPT_WITHIN 4 /* "within @foo" */
15495 Perl_varname(pTHX_ const GV *const gv, const char gvtype, PADOFFSET targ,
15496 const SV *const keyname, I32 aindex, int subscript_type)
15499 SV * const name = sv_newmortal();
15500 if (gv && isGV(gv)) {
15502 buffer[0] = gvtype;
15505 /* as gv_fullname4(), but add literal '^' for $^FOO names */
15507 gv_fullname4(name, gv, buffer, 0);
15509 if ((unsigned int)SvPVX(name)[1] <= 26) {
15511 buffer[1] = SvPVX(name)[1] + 'A' - 1;
15513 /* Swap the 1 unprintable control character for the 2 byte pretty
15514 version - ie substr($name, 1, 1) = $buffer; */
15515 sv_insert(name, 1, 1, buffer, 2);
15519 CV * const cv = gv ? ((CV *)gv) : find_runcv(NULL);
15522 assert(!cv || SvTYPE(cv) == SVt_PVCV || SvTYPE(cv) == SVt_PVFM);
15524 if (!cv || !CvPADLIST(cv))
15526 sv = padnamelist_fetch(PadlistNAMES(CvPADLIST(cv)), targ);
15527 sv_setpvn(name, PadnamePV(sv), PadnameLEN(sv));
15531 if (subscript_type == FUV_SUBSCRIPT_HASH) {
15532 SV * const sv = newSV(0);
15533 *SvPVX(name) = '$';
15534 Perl_sv_catpvf(aTHX_ name, "{%s}",
15535 pv_pretty(sv, SvPVX_const(keyname), SvCUR(keyname), 32, NULL, NULL,
15536 PERL_PV_PRETTY_DUMP | PERL_PV_ESCAPE_UNI_DETECT ));
15537 SvREFCNT_dec_NN(sv);
15539 else if (subscript_type == FUV_SUBSCRIPT_ARRAY) {
15540 *SvPVX(name) = '$';
15541 Perl_sv_catpvf(aTHX_ name, "[%"IVdf"]", (IV)aindex);
15543 else if (subscript_type == FUV_SUBSCRIPT_WITHIN) {
15544 /* We know that name has no magic, so can use 0 instead of SV_GMAGIC */
15545 Perl_sv_insert_flags(aTHX_ name, 0, 0, STR_WITH_LEN("within "), 0);
15553 =for apidoc find_uninit_var
15555 Find the name of the undefined variable (if any) that caused the operator
15556 to issue a "Use of uninitialized value" warning.
15557 If match is true, only return a name if its value matches C<uninit_sv>.
15558 So roughly speaking, if a unary operator (such as C<OP_COS>) generates a
15559 warning, then following the direct child of the op may yield an
15560 C<OP_PADSV> or C<OP_GV> that gives the name of the undefined variable. On the
15561 other hand, with C<OP_ADD> there are two branches to follow, so we only print
15562 the variable name if we get an exact match.
15563 C<desc_p> points to a string pointer holding the description of the op.
15564 This may be updated if needed.
15566 The name is returned as a mortal SV.
15568 Assumes that C<PL_op> is the OP that originally triggered the error, and that
15569 C<PL_comppad>/C<PL_curpad> points to the currently executing pad.
15575 S_find_uninit_var(pTHX_ const OP *const obase, const SV *const uninit_sv,
15576 bool match, const char **desc_p)
15581 const OP *o, *o2, *kid;
15583 PERL_ARGS_ASSERT_FIND_UNINIT_VAR;
15585 if (!obase || (match && (!uninit_sv || uninit_sv == &PL_sv_undef ||
15586 uninit_sv == &PL_sv_placeholder)))
15589 switch (obase->op_type) {
15596 const bool pad = ( obase->op_type == OP_PADAV
15597 || obase->op_type == OP_PADHV
15598 || obase->op_type == OP_PADRANGE
15601 const bool hash = ( obase->op_type == OP_PADHV
15602 || obase->op_type == OP_RV2HV
15603 || (obase->op_type == OP_PADRANGE
15604 && SvTYPE(PAD_SVl(obase->op_targ)) == SVt_PVHV)
15608 int subscript_type = FUV_SUBSCRIPT_WITHIN;
15610 if (pad) { /* @lex, %lex */
15611 sv = PAD_SVl(obase->op_targ);
15615 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
15616 /* @global, %global */
15617 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
15620 sv = hash ? MUTABLE_SV(GvHV(gv)): MUTABLE_SV(GvAV(gv));
15622 else if (obase == PL_op) /* @{expr}, %{expr} */
15623 return find_uninit_var(cUNOPx(obase)->op_first,
15624 uninit_sv, match, desc_p);
15625 else /* @{expr}, %{expr} as a sub-expression */
15629 /* attempt to find a match within the aggregate */
15631 keysv = find_hash_subscript((const HV*)sv, uninit_sv);
15633 subscript_type = FUV_SUBSCRIPT_HASH;
15636 index = find_array_subscript((const AV *)sv, uninit_sv);
15638 subscript_type = FUV_SUBSCRIPT_ARRAY;
15641 if (match && subscript_type == FUV_SUBSCRIPT_WITHIN)
15644 return varname(gv, (char)(hash ? '%' : '@'), obase->op_targ,
15645 keysv, index, subscript_type);
15649 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
15651 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
15652 if (!gv || !GvSTASH(gv))
15654 if (match && (GvSV(gv) != uninit_sv))
15656 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
15659 return find_uninit_var(cUNOPx(obase)->op_first, uninit_sv, 1, desc_p);
15662 if (match && PAD_SVl(obase->op_targ) != uninit_sv)
15664 return varname(NULL, '$', obase->op_targ,
15665 NULL, 0, FUV_SUBSCRIPT_NONE);
15668 gv = cGVOPx_gv(obase);
15669 if (!gv || (match && GvSV(gv) != uninit_sv) || !GvSTASH(gv))
15671 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
15673 case OP_AELEMFAST_LEX:
15676 AV *av = MUTABLE_AV(PAD_SV(obase->op_targ));
15677 if (!av || SvRMAGICAL(av))
15679 svp = av_fetch(av, (I8)obase->op_private, FALSE);
15680 if (!svp || *svp != uninit_sv)
15683 return varname(NULL, '$', obase->op_targ,
15684 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
15687 gv = cGVOPx_gv(obase);
15692 AV *const av = GvAV(gv);
15693 if (!av || SvRMAGICAL(av))
15695 svp = av_fetch(av, (I8)obase->op_private, FALSE);
15696 if (!svp || *svp != uninit_sv)
15699 return varname(gv, '$', 0,
15700 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
15702 NOT_REACHED; /* NOTREACHED */
15705 o = cUNOPx(obase)->op_first;
15706 if (!o || o->op_type != OP_NULL ||
15707 ! (o->op_targ == OP_AELEM || o->op_targ == OP_HELEM))
15709 return find_uninit_var(cBINOPo->op_last, uninit_sv, match, desc_p);
15714 bool negate = FALSE;
15716 if (PL_op == obase)
15717 /* $a[uninit_expr] or $h{uninit_expr} */
15718 return find_uninit_var(cBINOPx(obase)->op_last,
15719 uninit_sv, match, desc_p);
15722 o = cBINOPx(obase)->op_first;
15723 kid = cBINOPx(obase)->op_last;
15725 /* get the av or hv, and optionally the gv */
15727 if (o->op_type == OP_PADAV || o->op_type == OP_PADHV) {
15728 sv = PAD_SV(o->op_targ);
15730 else if ((o->op_type == OP_RV2AV || o->op_type == OP_RV2HV)
15731 && cUNOPo->op_first->op_type == OP_GV)
15733 gv = cGVOPx_gv(cUNOPo->op_first);
15737 == OP_RV2HV ? MUTABLE_SV(GvHV(gv)) : MUTABLE_SV(GvAV(gv));
15742 if (kid && kid->op_type == OP_NEGATE) {
15744 kid = cUNOPx(kid)->op_first;
15747 if (kid && kid->op_type == OP_CONST && SvOK(cSVOPx_sv(kid))) {
15748 /* index is constant */
15751 kidsv = newSVpvs_flags("-", SVs_TEMP);
15752 sv_catsv(kidsv, cSVOPx_sv(kid));
15755 kidsv = cSVOPx_sv(kid);
15759 if (obase->op_type == OP_HELEM) {
15760 HE* he = hv_fetch_ent(MUTABLE_HV(sv), kidsv, 0, 0);
15761 if (!he || HeVAL(he) != uninit_sv)
15765 SV * const opsv = cSVOPx_sv(kid);
15766 const IV opsviv = SvIV(opsv);
15767 SV * const * const svp = av_fetch(MUTABLE_AV(sv),
15768 negate ? - opsviv : opsviv,
15770 if (!svp || *svp != uninit_sv)
15774 if (obase->op_type == OP_HELEM)
15775 return varname(gv, '%', o->op_targ,
15776 kidsv, 0, FUV_SUBSCRIPT_HASH);
15778 return varname(gv, '@', o->op_targ, NULL,
15779 negate ? - SvIV(cSVOPx_sv(kid)) : SvIV(cSVOPx_sv(kid)),
15780 FUV_SUBSCRIPT_ARRAY);
15783 /* index is an expression;
15784 * attempt to find a match within the aggregate */
15785 if (obase->op_type == OP_HELEM) {
15786 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
15788 return varname(gv, '%', o->op_targ,
15789 keysv, 0, FUV_SUBSCRIPT_HASH);
15793 = find_array_subscript((const AV *)sv, uninit_sv);
15795 return varname(gv, '@', o->op_targ,
15796 NULL, index, FUV_SUBSCRIPT_ARRAY);
15801 (char)((o->op_type == OP_PADAV || o->op_type == OP_RV2AV)
15803 o->op_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
15805 NOT_REACHED; /* NOTREACHED */
15808 case OP_MULTIDEREF: {
15809 /* If we were executing OP_MULTIDEREF when the undef warning
15810 * triggered, then it must be one of the index values within
15811 * that triggered it. If not, then the only possibility is that
15812 * the value retrieved by the last aggregate lookup might be the
15813 * culprit. For the former, we set PL_multideref_pc each time before
15814 * using an index, so work though the item list until we reach
15815 * that point. For the latter, just work through the entire item
15816 * list; the last aggregate retrieved will be the candidate.
15819 /* the named aggregate, if any */
15820 PADOFFSET agg_targ = 0;
15822 /* the last-seen index */
15824 PADOFFSET index_targ;
15826 IV index_const_iv = 0; /* init for spurious compiler warn */
15827 SV *index_const_sv;
15828 int depth = 0; /* how many array/hash lookups we've done */
15830 UNOP_AUX_item *items = cUNOP_AUXx(obase)->op_aux;
15831 UNOP_AUX_item *last = NULL;
15832 UV actions = items->uv;
15835 if (PL_op == obase) {
15836 last = PL_multideref_pc;
15837 assert(last >= items && last <= items + items[-1].uv);
15844 switch (actions & MDEREF_ACTION_MASK) {
15846 case MDEREF_reload:
15847 actions = (++items)->uv;
15850 case MDEREF_HV_padhv_helem: /* $lex{...} */
15853 case MDEREF_AV_padav_aelem: /* $lex[...] */
15854 agg_targ = (++items)->pad_offset;
15858 case MDEREF_HV_gvhv_helem: /* $pkg{...} */
15861 case MDEREF_AV_gvav_aelem: /* $pkg[...] */
15863 agg_gv = (GV*)UNOP_AUX_item_sv(++items);
15864 assert(isGV_with_GP(agg_gv));
15867 case MDEREF_HV_gvsv_vivify_rv2hv_helem: /* $pkg->{...} */
15868 case MDEREF_HV_padsv_vivify_rv2hv_helem: /* $lex->{...} */
15871 case MDEREF_HV_pop_rv2hv_helem: /* expr->{...} */
15872 case MDEREF_HV_vivify_rv2hv_helem: /* vivify, ->{...} */
15878 case MDEREF_AV_gvsv_vivify_rv2av_aelem: /* $pkg->[...] */
15879 case MDEREF_AV_padsv_vivify_rv2av_aelem: /* $lex->[...] */
15882 case MDEREF_AV_pop_rv2av_aelem: /* expr->[...] */
15883 case MDEREF_AV_vivify_rv2av_aelem: /* vivify, ->[...] */
15890 index_const_sv = NULL;
15892 index_type = (actions & MDEREF_INDEX_MASK);
15893 switch (index_type) {
15894 case MDEREF_INDEX_none:
15896 case MDEREF_INDEX_const:
15898 index_const_sv = UNOP_AUX_item_sv(++items)
15900 index_const_iv = (++items)->iv;
15902 case MDEREF_INDEX_padsv:
15903 index_targ = (++items)->pad_offset;
15905 case MDEREF_INDEX_gvsv:
15906 index_gv = (GV*)UNOP_AUX_item_sv(++items);
15907 assert(isGV_with_GP(index_gv));
15911 if (index_type != MDEREF_INDEX_none)
15914 if ( index_type == MDEREF_INDEX_none
15915 || (actions & MDEREF_FLAG_last)
15916 || (last && items == last)
15920 actions >>= MDEREF_SHIFT;
15923 if (PL_op == obase) {
15924 /* index was undef */
15926 *desc_p = ( (actions & MDEREF_FLAG_last)
15927 && (obase->op_private
15928 & (OPpMULTIDEREF_EXISTS|OPpMULTIDEREF_DELETE)))
15930 (obase->op_private & OPpMULTIDEREF_EXISTS)
15933 : is_hv ? "hash element" : "array element";
15934 assert(index_type != MDEREF_INDEX_none);
15936 return varname(index_gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
15938 return varname(NULL, '$', index_targ,
15939 NULL, 0, FUV_SUBSCRIPT_NONE);
15940 assert(is_hv); /* AV index is an IV and can't be undef */
15941 /* can a const HV index ever be undef? */
15945 /* the SV returned by pp_multideref() was undef, if anything was */
15951 sv = PAD_SV(agg_targ);
15953 sv = is_hv ? MUTABLE_SV(GvHV(agg_gv)) : MUTABLE_SV(GvAV(agg_gv));
15957 if (index_type == MDEREF_INDEX_const) {
15962 HE* he = hv_fetch_ent(MUTABLE_HV(sv), index_const_sv, 0, 0);
15963 if (!he || HeVAL(he) != uninit_sv)
15967 SV * const * const svp =
15968 av_fetch(MUTABLE_AV(sv), index_const_iv, FALSE);
15969 if (!svp || *svp != uninit_sv)
15974 ? varname(agg_gv, '%', agg_targ,
15975 index_const_sv, 0, FUV_SUBSCRIPT_HASH)
15976 : varname(agg_gv, '@', agg_targ,
15977 NULL, index_const_iv, FUV_SUBSCRIPT_ARRAY);
15980 /* index is an var */
15982 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
15984 return varname(agg_gv, '%', agg_targ,
15985 keysv, 0, FUV_SUBSCRIPT_HASH);
15989 = find_array_subscript((const AV *)sv, uninit_sv);
15991 return varname(agg_gv, '@', agg_targ,
15992 NULL, index, FUV_SUBSCRIPT_ARRAY);
15996 return varname(agg_gv,
15998 agg_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
16000 NOT_REACHED; /* NOTREACHED */
16004 /* only examine RHS */
16005 return find_uninit_var(cBINOPx(obase)->op_first, uninit_sv,
16009 o = cUNOPx(obase)->op_first;
16010 if ( o->op_type == OP_PUSHMARK
16011 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)
16015 if (!OpHAS_SIBLING(o)) {
16016 /* one-arg version of open is highly magical */
16018 if (o->op_type == OP_GV) { /* open FOO; */
16020 if (match && GvSV(gv) != uninit_sv)
16022 return varname(gv, '$', 0,
16023 NULL, 0, FUV_SUBSCRIPT_NONE);
16025 /* other possibilities not handled are:
16026 * open $x; or open my $x; should return '${*$x}'
16027 * open expr; should return '$'.expr ideally
16033 /* ops where $_ may be an implicit arg */
16038 if ( !(obase->op_flags & OPf_STACKED)) {
16039 if (uninit_sv == DEFSV)
16040 return newSVpvs_flags("$_", SVs_TEMP);
16041 else if (obase->op_targ
16042 && uninit_sv == PAD_SVl(obase->op_targ))
16043 return varname(NULL, '$', obase->op_targ, NULL, 0,
16044 FUV_SUBSCRIPT_NONE);
16051 match = 1; /* print etc can return undef on defined args */
16052 /* skip filehandle as it can't produce 'undef' warning */
16053 o = cUNOPx(obase)->op_first;
16054 if ((obase->op_flags & OPf_STACKED)
16056 ( o->op_type == OP_PUSHMARK
16057 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)))
16058 o = OpSIBLING(OpSIBLING(o));
16062 case OP_ENTEREVAL: /* could be eval $undef or $x='$undef'; eval $x */
16063 case OP_CUSTOM: /* XS or custom code could trigger random warnings */
16065 /* the following ops are capable of returning PL_sv_undef even for
16066 * defined arg(s) */
16085 case OP_GETPEERNAME:
16133 case OP_SMARTMATCH:
16142 /* XXX tmp hack: these two may call an XS sub, and currently
16143 XS subs don't have a SUB entry on the context stack, so CV and
16144 pad determination goes wrong, and BAD things happen. So, just
16145 don't try to determine the value under those circumstances.
16146 Need a better fix at dome point. DAPM 11/2007 */
16152 GV * const gv = gv_fetchpvs(".", GV_NOTQUAL, SVt_PV);
16153 if (gv && GvSV(gv) == uninit_sv)
16154 return newSVpvs_flags("$.", SVs_TEMP);
16159 /* def-ness of rval pos() is independent of the def-ness of its arg */
16160 if ( !(obase->op_flags & OPf_MOD))
16165 if (SvROK(PL_rs) && uninit_sv == SvRV(PL_rs))
16166 return newSVpvs_flags("${$/}", SVs_TEMP);
16171 if (!(obase->op_flags & OPf_KIDS))
16173 o = cUNOPx(obase)->op_first;
16179 /* This loop checks all the kid ops, skipping any that cannot pos-
16180 * sibly be responsible for the uninitialized value; i.e., defined
16181 * constants and ops that return nothing. If there is only one op
16182 * left that is not skipped, then we *know* it is responsible for
16183 * the uninitialized value. If there is more than one op left, we
16184 * have to look for an exact match in the while() loop below.
16185 * Note that we skip padrange, because the individual pad ops that
16186 * it replaced are still in the tree, so we work on them instead.
16189 for (kid=o; kid; kid = OpSIBLING(kid)) {
16190 const OPCODE type = kid->op_type;
16191 if ( (type == OP_CONST && SvOK(cSVOPx_sv(kid)))
16192 || (type == OP_NULL && ! (kid->op_flags & OPf_KIDS))
16193 || (type == OP_PUSHMARK)
16194 || (type == OP_PADRANGE)
16198 if (o2) { /* more than one found */
16205 return find_uninit_var(o2, uninit_sv, match, desc_p);
16207 /* scan all args */
16209 sv = find_uninit_var(o, uninit_sv, 1, desc_p);
16221 =for apidoc report_uninit
16223 Print appropriate "Use of uninitialized variable" warning.
16229 Perl_report_uninit(pTHX_ const SV *uninit_sv)
16231 const char *desc = NULL;
16232 SV* varname = NULL;
16235 desc = PL_op->op_type == OP_STRINGIFY && PL_op->op_folded
16238 if (uninit_sv && PL_curpad) {
16239 varname = find_uninit_var(PL_op, uninit_sv, 0, &desc);
16241 sv_insert(varname, 0, 0, " ", 1);
16244 else if (PL_curstackinfo->si_type == PERLSI_SORT
16245 && CxMULTICALL(&cxstack[cxstack_ix]))
16247 /* we've reached the end of a sort block or sub,
16248 * and the uninit value is probably what that code returned */
16252 /* PL_warn_uninit_sv is constant */
16253 GCC_DIAG_IGNORE(-Wformat-nonliteral);
16255 /* diag_listed_as: Use of uninitialized value%s */
16256 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit_sv,
16257 SVfARG(varname ? varname : &PL_sv_no),
16260 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit,
16266 * ex: set ts=8 sts=4 sw=4 et: