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
1528 /* prior to 5.000 stable, this function returned the new OOK-less SvFLAGS
1529 prior to 5.23.4 this function always returned 0
1533 Perl_sv_backoff(SV *const sv)
1536 const char * const s = SvPVX_const(sv);
1538 PERL_ARGS_ASSERT_SV_BACKOFF;
1541 assert(SvTYPE(sv) != SVt_PVHV);
1542 assert(SvTYPE(sv) != SVt_PVAV);
1544 SvOOK_offset(sv, delta);
1546 SvLEN_set(sv, SvLEN(sv) + delta);
1547 SvPV_set(sv, SvPVX(sv) - delta);
1548 SvFLAGS(sv) &= ~SVf_OOK;
1549 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
1556 Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1557 upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1558 Use the C<SvGROW> wrapper instead.
1563 static void S_sv_uncow(pTHX_ SV * const sv, const U32 flags);
1566 Perl_sv_grow(pTHX_ SV *const sv, STRLEN newlen)
1570 PERL_ARGS_ASSERT_SV_GROW;
1574 if (SvTYPE(sv) < SVt_PV) {
1575 sv_upgrade(sv, SVt_PV);
1576 s = SvPVX_mutable(sv);
1578 else if (SvOOK(sv)) { /* pv is offset? */
1580 s = SvPVX_mutable(sv);
1581 if (newlen > SvLEN(sv))
1582 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
1586 if (SvIsCOW(sv)) S_sv_uncow(aTHX_ sv, 0);
1587 s = SvPVX_mutable(sv);
1590 #ifdef PERL_COPY_ON_WRITE
1591 /* the new COW scheme uses SvPVX(sv)[SvLEN(sv)-1] (if spare)
1592 * to store the COW count. So in general, allocate one more byte than
1593 * asked for, to make it likely this byte is always spare: and thus
1594 * make more strings COW-able.
1595 * If the new size is a big power of two, don't bother: we assume the
1596 * caller wanted a nice 2^N sized block and will be annoyed at getting
1598 * Only increment if the allocation isn't MEM_SIZE_MAX,
1599 * otherwise it will wrap to 0.
1601 if (newlen & 0xff && newlen != MEM_SIZE_MAX)
1605 #if defined(PERL_USE_MALLOC_SIZE) && defined(Perl_safesysmalloc_size)
1606 #define PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1609 if (newlen > SvLEN(sv)) { /* need more room? */
1610 STRLEN minlen = SvCUR(sv);
1611 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
1612 if (newlen < minlen)
1614 #ifndef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1616 /* Don't round up on the first allocation, as odds are pretty good that
1617 * the initial request is accurate as to what is really needed */
1619 STRLEN rounded = PERL_STRLEN_ROUNDUP(newlen);
1620 if (rounded > newlen)
1624 if (SvLEN(sv) && s) {
1625 s = (char*)saferealloc(s, newlen);
1628 s = (char*)safemalloc(newlen);
1629 if (SvPVX_const(sv) && SvCUR(sv)) {
1630 Move(SvPVX_const(sv), s, (newlen < SvCUR(sv)) ? newlen : SvCUR(sv), char);
1634 #ifdef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1635 /* Do this here, do it once, do it right, and then we will never get
1636 called back into sv_grow() unless there really is some growing
1638 SvLEN_set(sv, Perl_safesysmalloc_size(s));
1640 SvLEN_set(sv, newlen);
1647 =for apidoc sv_setiv
1649 Copies an integer into the given SV, upgrading first if necessary.
1650 Does not handle 'set' magic. See also C<L</sv_setiv_mg>>.
1656 Perl_sv_setiv(pTHX_ SV *const sv, const IV i)
1658 PERL_ARGS_ASSERT_SV_SETIV;
1660 SV_CHECK_THINKFIRST_COW_DROP(sv);
1661 switch (SvTYPE(sv)) {
1664 sv_upgrade(sv, SVt_IV);
1667 sv_upgrade(sv, SVt_PVIV);
1671 if (!isGV_with_GP(sv))
1678 /* diag_listed_as: Can't coerce %s to %s in %s */
1679 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1684 (void)SvIOK_only(sv); /* validate number */
1690 =for apidoc sv_setiv_mg
1692 Like C<sv_setiv>, but also handles 'set' magic.
1698 Perl_sv_setiv_mg(pTHX_ SV *const sv, const IV i)
1700 PERL_ARGS_ASSERT_SV_SETIV_MG;
1707 =for apidoc sv_setuv
1709 Copies an unsigned integer into the given SV, upgrading first if necessary.
1710 Does not handle 'set' magic. See also C<L</sv_setuv_mg>>.
1716 Perl_sv_setuv(pTHX_ SV *const sv, const UV u)
1718 PERL_ARGS_ASSERT_SV_SETUV;
1720 /* With the if statement to ensure that integers are stored as IVs whenever
1722 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
1725 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1727 If you wish to remove the following if statement, so that this routine
1728 (and its callers) always return UVs, please benchmark to see what the
1729 effect is. Modern CPUs may be different. Or may not :-)
1731 if (u <= (UV)IV_MAX) {
1732 sv_setiv(sv, (IV)u);
1741 =for apidoc sv_setuv_mg
1743 Like C<sv_setuv>, but also handles 'set' magic.
1749 Perl_sv_setuv_mg(pTHX_ SV *const sv, const UV u)
1751 PERL_ARGS_ASSERT_SV_SETUV_MG;
1758 =for apidoc sv_setnv
1760 Copies a double into the given SV, upgrading first if necessary.
1761 Does not handle 'set' magic. See also C<L</sv_setnv_mg>>.
1767 Perl_sv_setnv(pTHX_ SV *const sv, const NV num)
1769 PERL_ARGS_ASSERT_SV_SETNV;
1771 SV_CHECK_THINKFIRST_COW_DROP(sv);
1772 switch (SvTYPE(sv)) {
1775 sv_upgrade(sv, SVt_NV);
1779 sv_upgrade(sv, SVt_PVNV);
1783 if (!isGV_with_GP(sv))
1790 /* diag_listed_as: Can't coerce %s to %s in %s */
1791 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
1797 (void)SvNOK_only(sv); /* validate number */
1802 =for apidoc sv_setnv_mg
1804 Like C<sv_setnv>, but also handles 'set' magic.
1810 Perl_sv_setnv_mg(pTHX_ SV *const sv, const NV num)
1812 PERL_ARGS_ASSERT_SV_SETNV_MG;
1818 /* Return a cleaned-up, printable version of sv, for non-numeric, or
1819 * not incrementable warning display.
1820 * Originally part of S_not_a_number().
1821 * The return value may be != tmpbuf.
1825 S_sv_display(pTHX_ SV *const sv, char *tmpbuf, STRLEN tmpbuf_size) {
1828 PERL_ARGS_ASSERT_SV_DISPLAY;
1831 SV *dsv = newSVpvs_flags("", SVs_TEMP);
1832 pv = sv_uni_display(dsv, sv, 32, UNI_DISPLAY_ISPRINT);
1835 const char * const limit = tmpbuf + tmpbuf_size - 8;
1836 /* each *s can expand to 4 chars + "...\0",
1837 i.e. need room for 8 chars */
1839 const char *s = SvPVX_const(sv);
1840 const char * const end = s + SvCUR(sv);
1841 for ( ; s < end && d < limit; s++ ) {
1843 if (! isASCII(ch) && !isPRINT_LC(ch)) {
1847 /* Map to ASCII "equivalent" of Latin1 */
1848 ch = LATIN1_TO_NATIVE(NATIVE_TO_LATIN1(ch) & 127);
1854 else if (ch == '\r') {
1858 else if (ch == '\f') {
1862 else if (ch == '\\') {
1866 else if (ch == '\0') {
1870 else if (isPRINT_LC(ch))
1889 /* Print an "isn't numeric" warning, using a cleaned-up,
1890 * printable version of the offending string
1894 S_not_a_number(pTHX_ SV *const sv)
1899 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1901 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1904 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1905 /* diag_listed_as: Argument "%s" isn't numeric%s */
1906 "Argument \"%s\" isn't numeric in %s", pv,
1909 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1910 /* diag_listed_as: Argument "%s" isn't numeric%s */
1911 "Argument \"%s\" isn't numeric", pv);
1915 S_not_incrementable(pTHX_ SV *const sv) {
1919 PERL_ARGS_ASSERT_NOT_INCREMENTABLE;
1921 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1923 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1924 "Argument \"%s\" treated as 0 in increment (++)", pv);
1928 =for apidoc looks_like_number
1930 Test if the content of an SV looks like a number (or is a number).
1931 C<Inf> and C<Infinity> are treated as numbers (so will not issue a
1932 non-numeric warning), even if your C<atof()> doesn't grok them. Get-magic is
1939 Perl_looks_like_number(pTHX_ SV *const sv)
1945 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1947 if (SvPOK(sv) || SvPOKp(sv)) {
1948 sbegin = SvPV_nomg_const(sv, len);
1951 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
1952 numtype = grok_number(sbegin, len, NULL);
1953 return ((numtype & IS_NUMBER_TRAILING)) ? 0 : numtype;
1957 S_glob_2number(pTHX_ GV * const gv)
1959 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1961 /* We know that all GVs stringify to something that is not-a-number,
1962 so no need to test that. */
1963 if (ckWARN(WARN_NUMERIC))
1965 SV *const buffer = sv_newmortal();
1966 gv_efullname3(buffer, gv, "*");
1967 not_a_number(buffer);
1969 /* We just want something true to return, so that S_sv_2iuv_common
1970 can tail call us and return true. */
1974 /* Actually, ISO C leaves conversion of UV to IV undefined, but
1975 until proven guilty, assume that things are not that bad... */
1980 As 64 bit platforms often have an NV that doesn't preserve all bits of
1981 an IV (an assumption perl has been based on to date) it becomes necessary
1982 to remove the assumption that the NV always carries enough precision to
1983 recreate the IV whenever needed, and that the NV is the canonical form.
1984 Instead, IV/UV and NV need to be given equal rights. So as to not lose
1985 precision as a side effect of conversion (which would lead to insanity
1986 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
1987 1) to distinguish between IV/UV/NV slots that have a valid conversion cached
1988 where precision was lost, and IV/UV/NV slots that have a valid conversion
1989 which has lost no precision
1990 2) to ensure that if a numeric conversion to one form is requested that
1991 would lose precision, the precise conversion (or differently
1992 imprecise conversion) is also performed and cached, to prevent
1993 requests for different numeric formats on the same SV causing
1994 lossy conversion chains. (lossless conversion chains are perfectly
1999 SvIOKp is true if the IV slot contains a valid value
2000 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
2001 SvNOKp is true if the NV slot contains a valid value
2002 SvNOK is true only if the NV value is accurate
2005 while converting from PV to NV, check to see if converting that NV to an
2006 IV(or UV) would lose accuracy over a direct conversion from PV to
2007 IV(or UV). If it would, cache both conversions, return NV, but mark
2008 SV as IOK NOKp (ie not NOK).
2010 While converting from PV to IV, check to see if converting that IV to an
2011 NV would lose accuracy over a direct conversion from PV to NV. If it
2012 would, cache both conversions, flag similarly.
2014 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
2015 correctly because if IV & NV were set NV *always* overruled.
2016 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
2017 changes - now IV and NV together means that the two are interchangeable:
2018 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
2020 The benefit of this is that operations such as pp_add know that if
2021 SvIOK is true for both left and right operands, then integer addition
2022 can be used instead of floating point (for cases where the result won't
2023 overflow). Before, floating point was always used, which could lead to
2024 loss of precision compared with integer addition.
2026 * making IV and NV equal status should make maths accurate on 64 bit
2028 * may speed up maths somewhat if pp_add and friends start to use
2029 integers when possible instead of fp. (Hopefully the overhead in
2030 looking for SvIOK and checking for overflow will not outweigh the
2031 fp to integer speedup)
2032 * will slow down integer operations (callers of SvIV) on "inaccurate"
2033 values, as the change from SvIOK to SvIOKp will cause a call into
2034 sv_2iv each time rather than a macro access direct to the IV slot
2035 * should speed up number->string conversion on integers as IV is
2036 favoured when IV and NV are equally accurate
2038 ####################################################################
2039 You had better be using SvIOK_notUV if you want an IV for arithmetic:
2040 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
2041 On the other hand, SvUOK is true iff UV.
2042 ####################################################################
2044 Your mileage will vary depending your CPU's relative fp to integer
2048 #ifndef NV_PRESERVES_UV
2049 # define IS_NUMBER_UNDERFLOW_IV 1
2050 # define IS_NUMBER_UNDERFLOW_UV 2
2051 # define IS_NUMBER_IV_AND_UV 2
2052 # define IS_NUMBER_OVERFLOW_IV 4
2053 # define IS_NUMBER_OVERFLOW_UV 5
2055 /* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
2057 /* For sv_2nv these three cases are "SvNOK and don't bother casting" */
2059 S_sv_2iuv_non_preserve(pTHX_ SV *const sv
2065 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
2066 PERL_UNUSED_CONTEXT;
2068 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));
2069 if (SvNVX(sv) < (NV)IV_MIN) {
2070 (void)SvIOKp_on(sv);
2072 SvIV_set(sv, IV_MIN);
2073 return IS_NUMBER_UNDERFLOW_IV;
2075 if (SvNVX(sv) > (NV)UV_MAX) {
2076 (void)SvIOKp_on(sv);
2079 SvUV_set(sv, UV_MAX);
2080 return IS_NUMBER_OVERFLOW_UV;
2082 (void)SvIOKp_on(sv);
2084 /* Can't use strtol etc to convert this string. (See truth table in
2086 if (SvNVX(sv) <= (UV)IV_MAX) {
2087 SvIV_set(sv, I_V(SvNVX(sv)));
2088 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2089 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
2091 /* Integer is imprecise. NOK, IOKp */
2093 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
2096 SvUV_set(sv, U_V(SvNVX(sv)));
2097 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2098 if (SvUVX(sv) == UV_MAX) {
2099 /* As we know that NVs don't preserve UVs, UV_MAX cannot
2100 possibly be preserved by NV. Hence, it must be overflow.
2102 return IS_NUMBER_OVERFLOW_UV;
2104 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
2106 /* Integer is imprecise. NOK, IOKp */
2108 return IS_NUMBER_OVERFLOW_IV;
2110 #endif /* !NV_PRESERVES_UV*/
2112 /* If numtype is infnan, set the NV of the sv accordingly.
2113 * If numtype is anything else, try setting the NV using Atof(PV). */
2115 # pragma warning(push)
2116 # pragma warning(disable:4756;disable:4056)
2119 S_sv_setnv(pTHX_ SV* sv, int numtype)
2121 bool pok = cBOOL(SvPOK(sv));
2123 if ((numtype & IS_NUMBER_INFINITY)) {
2124 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -NV_INF : NV_INF);
2127 else if ((numtype & IS_NUMBER_NAN)) {
2128 SvNV_set(sv, NV_NAN);
2132 SvNV_set(sv, Atof(SvPVX_const(sv)));
2133 /* Purposefully no true nok here, since we don't want to blow
2134 * away the possible IOK/UV of an existing sv. */
2137 SvNOK_only(sv); /* No IV or UV please, this is pure infnan. */
2139 SvPOK_on(sv); /* PV is okay, though. */
2143 # pragma warning(pop)
2147 S_sv_2iuv_common(pTHX_ SV *const sv)
2149 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
2152 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
2153 * without also getting a cached IV/UV from it at the same time
2154 * (ie PV->NV conversion should detect loss of accuracy and cache
2155 * IV or UV at same time to avoid this. */
2156 /* IV-over-UV optimisation - choose to cache IV if possible */
2158 if (SvTYPE(sv) == SVt_NV)
2159 sv_upgrade(sv, SVt_PVNV);
2161 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
2162 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
2163 certainly cast into the IV range at IV_MAX, whereas the correct
2164 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
2166 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2167 if (Perl_isnan(SvNVX(sv))) {
2173 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2174 SvIV_set(sv, I_V(SvNVX(sv)));
2175 if (SvNVX(sv) == (NV) SvIVX(sv)
2176 #ifndef NV_PRESERVES_UV
2177 && SvIVX(sv) != IV_MIN /* avoid negating IV_MIN below */
2178 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2179 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2180 /* Don't flag it as "accurately an integer" if the number
2181 came from a (by definition imprecise) NV operation, and
2182 we're outside the range of NV integer precision */
2186 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2188 /* scalar has trailing garbage, eg "42a" */
2190 DEBUG_c(PerlIO_printf(Perl_debug_log,
2191 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (precise)\n",
2197 /* IV not precise. No need to convert from PV, as NV
2198 conversion would already have cached IV if it detected
2199 that PV->IV would be better than PV->NV->IV
2200 flags already correct - don't set public IOK. */
2201 DEBUG_c(PerlIO_printf(Perl_debug_log,
2202 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (imprecise)\n",
2207 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2208 but the cast (NV)IV_MIN rounds to a the value less (more
2209 negative) than IV_MIN which happens to be equal to SvNVX ??
2210 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2211 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2212 (NV)UVX == NVX are both true, but the values differ. :-(
2213 Hopefully for 2s complement IV_MIN is something like
2214 0x8000000000000000 which will be exact. NWC */
2217 SvUV_set(sv, U_V(SvNVX(sv)));
2219 (SvNVX(sv) == (NV) SvUVX(sv))
2220 #ifndef NV_PRESERVES_UV
2221 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2222 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2223 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2224 /* Don't flag it as "accurately an integer" if the number
2225 came from a (by definition imprecise) NV operation, and
2226 we're outside the range of NV integer precision */
2232 DEBUG_c(PerlIO_printf(Perl_debug_log,
2233 "0x%"UVxf" 2iv(%"UVuf" => %"IVdf") (as unsigned)\n",
2239 else if (SvPOKp(sv)) {
2241 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2242 /* We want to avoid a possible problem when we cache an IV/ a UV which
2243 may be later translated to an NV, and the resulting NV is not
2244 the same as the direct translation of the initial string
2245 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2246 be careful to ensure that the value with the .456 is around if the
2247 NV value is requested in the future).
2249 This means that if we cache such an IV/a UV, we need to cache the
2250 NV as well. Moreover, we trade speed for space, and do not
2251 cache the NV if we are sure it's not needed.
2254 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2255 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2256 == IS_NUMBER_IN_UV) {
2257 /* It's definitely an integer, only upgrade to PVIV */
2258 if (SvTYPE(sv) < SVt_PVIV)
2259 sv_upgrade(sv, SVt_PVIV);
2261 } else if (SvTYPE(sv) < SVt_PVNV)
2262 sv_upgrade(sv, SVt_PVNV);
2264 if ((numtype & (IS_NUMBER_INFINITY | IS_NUMBER_NAN))) {
2265 if (ckWARN(WARN_NUMERIC) && ((numtype & IS_NUMBER_TRAILING)))
2267 S_sv_setnv(aTHX_ sv, numtype);
2271 /* If NVs preserve UVs then we only use the UV value if we know that
2272 we aren't going to call atof() below. If NVs don't preserve UVs
2273 then the value returned may have more precision than atof() will
2274 return, even though value isn't perfectly accurate. */
2275 if ((numtype & (IS_NUMBER_IN_UV
2276 #ifdef NV_PRESERVES_UV
2279 )) == IS_NUMBER_IN_UV) {
2280 /* This won't turn off the public IOK flag if it was set above */
2281 (void)SvIOKp_on(sv);
2283 if (!(numtype & IS_NUMBER_NEG)) {
2285 if (value <= (UV)IV_MAX) {
2286 SvIV_set(sv, (IV)value);
2288 /* it didn't overflow, and it was positive. */
2289 SvUV_set(sv, value);
2293 /* 2s complement assumption */
2294 if (value <= (UV)IV_MIN) {
2295 SvIV_set(sv, value == (UV)IV_MIN
2296 ? IV_MIN : -(IV)value);
2298 /* Too negative for an IV. This is a double upgrade, but
2299 I'm assuming it will be rare. */
2300 if (SvTYPE(sv) < SVt_PVNV)
2301 sv_upgrade(sv, SVt_PVNV);
2305 SvNV_set(sv, -(NV)value);
2306 SvIV_set(sv, IV_MIN);
2310 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2311 will be in the previous block to set the IV slot, and the next
2312 block to set the NV slot. So no else here. */
2314 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2315 != IS_NUMBER_IN_UV) {
2316 /* It wasn't an (integer that doesn't overflow the UV). */
2317 S_sv_setnv(aTHX_ sv, numtype);
2319 if (! numtype && ckWARN(WARN_NUMERIC))
2322 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%" NVgf ")\n",
2323 PTR2UV(sv), SvNVX(sv)));
2325 #ifdef NV_PRESERVES_UV
2326 (void)SvIOKp_on(sv);
2328 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2329 if (Perl_isnan(SvNVX(sv))) {
2335 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2336 SvIV_set(sv, I_V(SvNVX(sv)));
2337 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2340 NOOP; /* Integer is imprecise. NOK, IOKp */
2342 /* UV will not work better than IV */
2344 if (SvNVX(sv) > (NV)UV_MAX) {
2346 /* Integer is inaccurate. NOK, IOKp, is UV */
2347 SvUV_set(sv, UV_MAX);
2349 SvUV_set(sv, U_V(SvNVX(sv)));
2350 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2351 NV preservse UV so can do correct comparison. */
2352 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2355 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
2360 #else /* NV_PRESERVES_UV */
2361 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2362 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
2363 /* The IV/UV slot will have been set from value returned by
2364 grok_number above. The NV slot has just been set using
2367 assert (SvIOKp(sv));
2369 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2370 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2371 /* Small enough to preserve all bits. */
2372 (void)SvIOKp_on(sv);
2374 SvIV_set(sv, I_V(SvNVX(sv)));
2375 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2377 /* Assumption: first non-preserved integer is < IV_MAX,
2378 this NV is in the preserved range, therefore: */
2379 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2381 Perl_croak(aTHX_ "sv_2iv assumed (U_V(fabs((double)SvNVX(sv))) < (UV)IV_MAX) but SvNVX(sv)=%"NVgf" U_V is 0x%"UVxf", IV_MAX is 0x%"UVxf"\n", SvNVX(sv), U_V(SvNVX(sv)), (UV)IV_MAX);
2385 0 0 already failed to read UV.
2386 0 1 already failed to read UV.
2387 1 0 you won't get here in this case. IV/UV
2388 slot set, public IOK, Atof() unneeded.
2389 1 1 already read UV.
2390 so there's no point in sv_2iuv_non_preserve() attempting
2391 to use atol, strtol, strtoul etc. */
2393 sv_2iuv_non_preserve (sv, numtype);
2395 sv_2iuv_non_preserve (sv);
2399 #endif /* NV_PRESERVES_UV */
2400 /* It might be more code efficient to go through the entire logic above
2401 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2402 gets complex and potentially buggy, so more programmer efficient
2403 to do it this way, by turning off the public flags: */
2405 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2409 if (isGV_with_GP(sv))
2410 return glob_2number(MUTABLE_GV(sv));
2412 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2414 if (SvTYPE(sv) < SVt_IV)
2415 /* Typically the caller expects that sv_any is not NULL now. */
2416 sv_upgrade(sv, SVt_IV);
2417 /* Return 0 from the caller. */
2424 =for apidoc sv_2iv_flags
2426 Return the integer value of an SV, doing any necessary string
2427 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2428 Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2434 Perl_sv_2iv_flags(pTHX_ SV *const sv, const I32 flags)
2436 PERL_ARGS_ASSERT_SV_2IV_FLAGS;
2438 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2439 && SvTYPE(sv) != SVt_PVFM);
2441 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2447 if (flags & SV_SKIP_OVERLOAD)
2449 tmpstr = AMG_CALLunary(sv, numer_amg);
2450 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2451 return SvIV(tmpstr);
2454 return PTR2IV(SvRV(sv));
2457 if (SvVALID(sv) || isREGEXP(sv)) {
2458 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2459 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2460 In practice they are extremely unlikely to actually get anywhere
2461 accessible by user Perl code - the only way that I'm aware of is when
2462 a constant subroutine which is used as the second argument to index.
2464 Regexps have no SvIVX and SvNVX fields.
2466 assert(isREGEXP(sv) || SvPOKp(sv));
2469 const char * const ptr =
2470 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2472 = grok_number(ptr, SvCUR(sv), &value);
2474 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2475 == IS_NUMBER_IN_UV) {
2476 /* It's definitely an integer */
2477 if (numtype & IS_NUMBER_NEG) {
2478 if (value < (UV)IV_MIN)
2481 if (value < (UV)IV_MAX)
2486 /* Quite wrong but no good choices. */
2487 if ((numtype & IS_NUMBER_INFINITY)) {
2488 return (numtype & IS_NUMBER_NEG) ? IV_MIN : IV_MAX;
2489 } else if ((numtype & IS_NUMBER_NAN)) {
2490 return 0; /* So wrong. */
2494 if (ckWARN(WARN_NUMERIC))
2497 return I_V(Atof(ptr));
2501 if (SvTHINKFIRST(sv)) {
2502 if (SvREADONLY(sv) && !SvOK(sv)) {
2503 if (ckWARN(WARN_UNINITIALIZED))
2510 if (S_sv_2iuv_common(aTHX_ sv))
2514 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"IVdf")\n",
2515 PTR2UV(sv),SvIVX(sv)));
2516 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
2520 =for apidoc sv_2uv_flags
2522 Return the unsigned integer value of an SV, doing any necessary string
2523 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2524 Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
2530 Perl_sv_2uv_flags(pTHX_ SV *const sv, const I32 flags)
2532 PERL_ARGS_ASSERT_SV_2UV_FLAGS;
2534 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2540 if (flags & SV_SKIP_OVERLOAD)
2542 tmpstr = AMG_CALLunary(sv, numer_amg);
2543 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2544 return SvUV(tmpstr);
2547 return PTR2UV(SvRV(sv));
2550 if (SvVALID(sv) || isREGEXP(sv)) {
2551 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2552 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2553 Regexps have no SvIVX and SvNVX fields. */
2554 assert(isREGEXP(sv) || SvPOKp(sv));
2557 const char * const ptr =
2558 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2560 = grok_number(ptr, SvCUR(sv), &value);
2562 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2563 == IS_NUMBER_IN_UV) {
2564 /* It's definitely an integer */
2565 if (!(numtype & IS_NUMBER_NEG))
2569 /* Quite wrong but no good choices. */
2570 if ((numtype & IS_NUMBER_INFINITY)) {
2571 return UV_MAX; /* So wrong. */
2572 } else if ((numtype & IS_NUMBER_NAN)) {
2573 return 0; /* So wrong. */
2577 if (ckWARN(WARN_NUMERIC))
2580 return U_V(Atof(ptr));
2584 if (SvTHINKFIRST(sv)) {
2585 if (SvREADONLY(sv) && !SvOK(sv)) {
2586 if (ckWARN(WARN_UNINITIALIZED))
2593 if (S_sv_2iuv_common(aTHX_ sv))
2597 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2uv(%"UVuf")\n",
2598 PTR2UV(sv),SvUVX(sv)));
2599 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
2603 =for apidoc sv_2nv_flags
2605 Return the num value of an SV, doing any necessary string or integer
2606 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2607 Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
2613 Perl_sv_2nv_flags(pTHX_ SV *const sv, const I32 flags)
2615 PERL_ARGS_ASSERT_SV_2NV_FLAGS;
2617 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2618 && SvTYPE(sv) != SVt_PVFM);
2619 if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) {
2620 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2621 the same flag bit as SVf_IVisUV, so must not let them cache NVs.
2622 Regexps have no SvIVX and SvNVX fields. */
2624 if (flags & SV_GMAGIC)
2628 if (SvPOKp(sv) && !SvIOKp(sv)) {
2629 ptr = SvPVX_const(sv);
2631 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
2632 !grok_number(ptr, SvCUR(sv), NULL))
2638 return (NV)SvUVX(sv);
2640 return (NV)SvIVX(sv);
2646 ptr = RX_WRAPPED((REGEXP *)sv);
2649 assert(SvTYPE(sv) >= SVt_PVMG);
2650 /* This falls through to the report_uninit near the end of the
2652 } else if (SvTHINKFIRST(sv)) {
2657 if (flags & SV_SKIP_OVERLOAD)
2659 tmpstr = AMG_CALLunary(sv, numer_amg);
2660 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2661 return SvNV(tmpstr);
2664 return PTR2NV(SvRV(sv));
2666 if (SvREADONLY(sv) && !SvOK(sv)) {
2667 if (ckWARN(WARN_UNINITIALIZED))
2672 if (SvTYPE(sv) < SVt_NV) {
2673 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2674 sv_upgrade(sv, SVt_NV);
2676 STORE_NUMERIC_LOCAL_SET_STANDARD();
2677 PerlIO_printf(Perl_debug_log,
2678 "0x%"UVxf" num(%" NVgf ")\n",
2679 PTR2UV(sv), SvNVX(sv));
2680 RESTORE_NUMERIC_LOCAL();
2683 else if (SvTYPE(sv) < SVt_PVNV)
2684 sv_upgrade(sv, SVt_PVNV);
2689 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
2690 #ifdef NV_PRESERVES_UV
2696 /* Only set the public NV OK flag if this NV preserves the IV */
2697 /* Check it's not 0xFFFFFFFFFFFFFFFF */
2699 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
2700 : (SvIVX(sv) == I_V(SvNVX(sv))))
2706 else if (SvPOKp(sv)) {
2708 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2709 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
2711 #ifdef NV_PRESERVES_UV
2712 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2713 == IS_NUMBER_IN_UV) {
2714 /* It's definitely an integer */
2715 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
2717 S_sv_setnv(aTHX_ sv, numtype);
2724 SvNV_set(sv, Atof(SvPVX_const(sv)));
2725 /* Only set the public NV OK flag if this NV preserves the value in
2726 the PV at least as well as an IV/UV would.
2727 Not sure how to do this 100% reliably. */
2728 /* if that shift count is out of range then Configure's test is
2729 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2731 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2732 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2733 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
2734 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2735 /* Can't use strtol etc to convert this string, so don't try.
2736 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2739 /* value has been set. It may not be precise. */
2740 if ((numtype & IS_NUMBER_NEG) && (value >= (UV)IV_MIN)) {
2741 /* 2s complement assumption for (UV)IV_MIN */
2742 SvNOK_on(sv); /* Integer is too negative. */
2747 if (numtype & IS_NUMBER_NEG) {
2748 /* -IV_MIN is undefined, but we should never reach
2749 * this point with both IS_NUMBER_NEG and value ==
2751 assert(value != (UV)IV_MIN);
2752 SvIV_set(sv, -(IV)value);
2753 } else if (value <= (UV)IV_MAX) {
2754 SvIV_set(sv, (IV)value);
2756 SvUV_set(sv, value);
2760 if (numtype & IS_NUMBER_NOT_INT) {
2761 /* I believe that even if the original PV had decimals,
2762 they are lost beyond the limit of the FP precision.
2763 However, neither is canonical, so both only get p
2764 flags. NWC, 2000/11/25 */
2765 /* Both already have p flags, so do nothing */
2767 const NV nv = SvNVX(sv);
2768 /* XXX should this spot have NAN_COMPARE_BROKEN, too? */
2769 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2770 if (SvIVX(sv) == I_V(nv)) {
2773 /* It had no "." so it must be integer. */
2777 /* between IV_MAX and NV(UV_MAX).
2778 Could be slightly > UV_MAX */
2780 if (numtype & IS_NUMBER_NOT_INT) {
2781 /* UV and NV both imprecise. */
2783 const UV nv_as_uv = U_V(nv);
2785 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2794 /* It might be more code efficient to go through the entire logic above
2795 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2796 gets complex and potentially buggy, so more programmer efficient
2797 to do it this way, by turning off the public flags: */
2799 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2800 #endif /* NV_PRESERVES_UV */
2803 if (isGV_with_GP(sv)) {
2804 glob_2number(MUTABLE_GV(sv));
2808 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2810 assert (SvTYPE(sv) >= SVt_NV);
2811 /* Typically the caller expects that sv_any is not NULL now. */
2812 /* XXX Ilya implies that this is a bug in callers that assume this
2813 and ideally should be fixed. */
2817 STORE_NUMERIC_LOCAL_SET_STANDARD();
2818 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2nv(%" NVgf ")\n",
2819 PTR2UV(sv), SvNVX(sv));
2820 RESTORE_NUMERIC_LOCAL();
2828 Return an SV with the numeric value of the source SV, doing any necessary
2829 reference or overload conversion. The caller is expected to have handled
2836 Perl_sv_2num(pTHX_ SV *const sv)
2838 PERL_ARGS_ASSERT_SV_2NUM;
2843 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
2844 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
2845 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2846 return sv_2num(tmpsv);
2848 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2851 /* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2852 * UV as a string towards the end of buf, and return pointers to start and
2855 * We assume that buf is at least TYPE_CHARS(UV) long.
2859 S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
2861 char *ptr = buf + TYPE_CHARS(UV);
2862 char * const ebuf = ptr;
2865 PERL_ARGS_ASSERT_UIV_2BUF;
2873 uv = (iv == IV_MIN) ? (UV)iv : (UV)(-iv);
2877 *--ptr = '0' + (char)(uv % 10);
2885 /* Helper for sv_2pv_flags and sv_vcatpvfn_flags. If the NV is an
2886 * infinity or a not-a-number, writes the appropriate strings to the
2887 * buffer, including a zero byte. On success returns the written length,
2888 * excluding the zero byte, on failure (not an infinity, not a nan)
2889 * returns zero, assert-fails on maxlen being too short.
2891 * XXX for "Inf", "-Inf", and "NaN", we could have three read-only
2892 * shared string constants we point to, instead of generating a new
2893 * string for each instance. */
2895 S_infnan_2pv(NV nv, char* buffer, size_t maxlen, char plus) {
2897 assert(maxlen >= 4);
2898 if (Perl_isinf(nv)) {
2900 if (maxlen < 5) /* "-Inf\0" */
2910 else if (Perl_isnan(nv)) {
2914 /* XXX optionally output the payload mantissa bits as
2915 * "(unsigned)" (to match the nan("...") C99 function,
2916 * or maybe as "(0xhhh...)" would make more sense...
2917 * provide a format string so that the user can decide?
2918 * NOTE: would affect the maxlen and assert() logic.*/
2923 assert((s == buffer + 3) || (s == buffer + 4));
2925 return s - buffer - 1; /* -1: excluding the zero byte */
2929 =for apidoc sv_2pv_flags
2931 Returns a pointer to the string value of an SV, and sets C<*lp> to its length.
2932 If flags has the C<SV_GMAGIC> bit set, does an C<mg_get()> first. Coerces C<sv> to a
2933 string if necessary. Normally invoked via the C<SvPV_flags> macro.
2934 C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
2940 Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
2944 PERL_ARGS_ASSERT_SV_2PV_FLAGS;
2946 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2947 && SvTYPE(sv) != SVt_PVFM);
2948 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2953 if (flags & SV_SKIP_OVERLOAD)
2955 tmpstr = AMG_CALLunary(sv, string_amg);
2956 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
2957 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2959 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2963 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2964 if (flags & SV_CONST_RETURN) {
2965 pv = (char *) SvPVX_const(tmpstr);
2967 pv = (flags & SV_MUTABLE_RETURN)
2968 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
2971 *lp = SvCUR(tmpstr);
2973 pv = sv_2pv_flags(tmpstr, lp, flags);
2986 SV *const referent = SvRV(sv);
2990 retval = buffer = savepvn("NULLREF", len);
2991 } else if (SvTYPE(referent) == SVt_REGEXP &&
2992 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
2993 amagic_is_enabled(string_amg))) {
2994 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
2998 /* If the regex is UTF-8 we want the containing scalar to
2999 have an UTF-8 flag too */
3006 *lp = RX_WRAPLEN(re);
3008 return RX_WRAPPED(re);
3010 const char *const typestr = sv_reftype(referent, 0);
3011 const STRLEN typelen = strlen(typestr);
3012 UV addr = PTR2UV(referent);
3013 const char *stashname = NULL;
3014 STRLEN stashnamelen = 0; /* hush, gcc */
3015 const char *buffer_end;
3017 if (SvOBJECT(referent)) {
3018 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
3021 stashname = HEK_KEY(name);
3022 stashnamelen = HEK_LEN(name);
3024 if (HEK_UTF8(name)) {
3030 stashname = "__ANON__";
3033 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
3034 + 2 * sizeof(UV) + 2 /* )\0 */;
3036 len = typelen + 3 /* (0x */
3037 + 2 * sizeof(UV) + 2 /* )\0 */;
3040 Newx(buffer, len, char);
3041 buffer_end = retval = buffer + len;
3043 /* Working backwards */
3047 *--retval = PL_hexdigit[addr & 15];
3048 } while (addr >>= 4);
3054 memcpy(retval, typestr, typelen);
3058 retval -= stashnamelen;
3059 memcpy(retval, stashname, stashnamelen);
3061 /* retval may not necessarily have reached the start of the
3063 assert (retval >= buffer);
3065 len = buffer_end - retval - 1; /* -1 for that \0 */
3077 if (flags & SV_MUTABLE_RETURN)
3078 return SvPVX_mutable(sv);
3079 if (flags & SV_CONST_RETURN)
3080 return (char *)SvPVX_const(sv);
3085 /* I'm assuming that if both IV and NV are equally valid then
3086 converting the IV is going to be more efficient */
3087 const U32 isUIOK = SvIsUV(sv);
3088 char buf[TYPE_CHARS(UV)];
3092 if (SvTYPE(sv) < SVt_PVIV)
3093 sv_upgrade(sv, SVt_PVIV);
3094 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
3096 /* inlined from sv_setpvn */
3097 s = SvGROW_mutable(sv, len + 1);
3098 Move(ptr, s, len, char);
3103 else if (SvNOK(sv)) {
3104 if (SvTYPE(sv) < SVt_PVNV)
3105 sv_upgrade(sv, SVt_PVNV);
3106 if (SvNVX(sv) == 0.0
3107 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
3108 && !Perl_isnan(SvNVX(sv))
3111 s = SvGROW_mutable(sv, 2);
3116 STRLEN size = 5; /* "-Inf\0" */
3118 s = SvGROW_mutable(sv, size);
3119 len = S_infnan_2pv(SvNVX(sv), s, size, 0);
3125 /* some Xenix systems wipe out errno here */
3134 5 + /* exponent digits */
3138 s = SvGROW_mutable(sv, size);
3139 #ifndef USE_LOCALE_NUMERIC
3140 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3146 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
3147 STORE_LC_NUMERIC_SET_TO_NEEDED();
3151 PL_numeric_radix_sv &&
3152 SvUTF8(PL_numeric_radix_sv);
3153 if (local_radix && SvLEN(PL_numeric_radix_sv) > 1) {
3154 size += SvLEN(PL_numeric_radix_sv) - 1;
3155 s = SvGROW_mutable(sv, size);
3158 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3160 /* If the radix character is UTF-8, and actually is in the
3161 * output, turn on the UTF-8 flag for the scalar */
3163 instr(s, SvPVX_const(PL_numeric_radix_sv))) {
3167 RESTORE_LC_NUMERIC();
3170 /* We don't call SvPOK_on(), because it may come to
3171 * pass that the locale changes so that the
3172 * stringification we just did is no longer correct. We
3173 * will have to re-stringify every time it is needed */
3180 else if (isGV_with_GP(sv)) {
3181 GV *const gv = MUTABLE_GV(sv);
3182 SV *const buffer = sv_newmortal();
3184 gv_efullname3(buffer, gv, "*");
3186 assert(SvPOK(buffer));
3190 *lp = SvCUR(buffer);
3191 return SvPVX(buffer);
3193 else if (isREGEXP(sv)) {
3194 if (lp) *lp = RX_WRAPLEN((REGEXP *)sv);
3195 return RX_WRAPPED((REGEXP *)sv);
3200 if (flags & SV_UNDEF_RETURNS_NULL)
3202 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
3204 /* Typically the caller expects that sv_any is not NULL now. */
3205 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
3206 sv_upgrade(sv, SVt_PV);
3211 const STRLEN len = s - SvPVX_const(sv);
3216 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
3217 PTR2UV(sv),SvPVX_const(sv)));
3218 if (flags & SV_CONST_RETURN)
3219 return (char *)SvPVX_const(sv);
3220 if (flags & SV_MUTABLE_RETURN)
3221 return SvPVX_mutable(sv);
3226 =for apidoc sv_copypv
3228 Copies a stringified representation of the source SV into the
3229 destination SV. Automatically performs any necessary C<mg_get> and
3230 coercion of numeric values into strings. Guaranteed to preserve
3231 C<UTF8> flag even from overloaded objects. Similar in nature to
3232 C<sv_2pv[_flags]> but operates directly on an SV instead of just the
3233 string. Mostly uses C<sv_2pv_flags> to do its work, except when that
3234 would lose the UTF-8'ness of the PV.
3236 =for apidoc sv_copypv_nomg
3238 Like C<sv_copypv>, but doesn't invoke get magic first.
3240 =for apidoc sv_copypv_flags
3242 Implementation of C<sv_copypv> and C<sv_copypv_nomg>. Calls get magic iff flags
3243 has the C<SV_GMAGIC> bit set.
3249 Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
3254 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
3256 s = SvPV_flags_const(ssv,len,(flags & SV_GMAGIC));
3257 sv_setpvn(dsv,s,len);
3265 =for apidoc sv_2pvbyte
3267 Return a pointer to the byte-encoded representation of the SV, and set C<*lp>
3268 to its length. May cause the SV to be downgraded from UTF-8 as a
3271 Usually accessed via the C<SvPVbyte> macro.
3277 Perl_sv_2pvbyte(pTHX_ SV *sv, STRLEN *const lp)
3279 PERL_ARGS_ASSERT_SV_2PVBYTE;
3282 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3283 || isGV_with_GP(sv) || SvROK(sv)) {
3284 SV *sv2 = sv_newmortal();
3285 sv_copypv_nomg(sv2,sv);
3288 sv_utf8_downgrade(sv,0);
3289 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3293 =for apidoc sv_2pvutf8
3295 Return a pointer to the UTF-8-encoded representation of the SV, and set C<*lp>
3296 to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3298 Usually accessed via the C<SvPVutf8> macro.
3304 Perl_sv_2pvutf8(pTHX_ SV *sv, STRLEN *const lp)
3306 PERL_ARGS_ASSERT_SV_2PVUTF8;
3308 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3309 || isGV_with_GP(sv) || SvROK(sv))
3310 sv = sv_mortalcopy(sv);
3313 sv_utf8_upgrade_nomg(sv);
3314 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3319 =for apidoc sv_2bool
3321 This macro is only used by C<sv_true()> or its macro equivalent, and only if
3322 the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>.
3323 It calls C<sv_2bool_flags> with the C<SV_GMAGIC> flag.
3325 =for apidoc sv_2bool_flags
3327 This function is only used by C<sv_true()> and friends, and only if
3328 the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>. If the flags
3329 contain C<SV_GMAGIC>, then it does an C<mg_get()> first.
3336 Perl_sv_2bool_flags(pTHX_ SV *sv, I32 flags)
3338 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
3341 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
3347 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
3348 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv)))) {
3351 if(SvGMAGICAL(sv)) {
3353 goto restart; /* call sv_2bool */
3355 /* expanded SvTRUE_common(sv, (flags = 0, goto restart)) */
3356 else if(!SvOK(sv)) {
3359 else if(SvPOK(sv)) {
3360 svb = SvPVXtrue(sv);
3362 else if((SvFLAGS(sv) & (SVf_IOK|SVf_NOK))) {
3363 svb = (SvIOK(sv) && SvIVX(sv) != 0)
3364 || (SvNOK(sv) && SvNVX(sv) != 0.0);
3368 goto restart; /* call sv_2bool_nomg */
3373 return SvRV(sv) != 0;
3377 RX_WRAPLEN(sv) > 1 || (RX_WRAPLEN(sv) && *RX_WRAPPED(sv) != '0');
3378 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
3382 =for apidoc sv_utf8_upgrade
3384 Converts the PV of an SV to its UTF-8-encoded form.
3385 Forces the SV to string form if it is not already.
3386 Will C<mg_get> on C<sv> if appropriate.
3387 Always sets the C<SvUTF8> flag to avoid future validity checks even
3388 if the whole string is the same in UTF-8 as not.
3389 Returns the number of bytes in the converted string
3391 This is not a general purpose byte encoding to Unicode interface:
3392 use the Encode extension for that.
3394 =for apidoc sv_utf8_upgrade_nomg
3396 Like C<sv_utf8_upgrade>, but doesn't do magic on C<sv>.
3398 =for apidoc sv_utf8_upgrade_flags
3400 Converts the PV of an SV to its UTF-8-encoded form.
3401 Forces the SV to string form if it is not already.
3402 Always sets the SvUTF8 flag to avoid future validity checks even
3403 if all the bytes are invariant in UTF-8.
3404 If C<flags> has C<SV_GMAGIC> bit set,
3405 will C<mg_get> on C<sv> if appropriate, else not.
3407 If C<flags> has C<SV_FORCE_UTF8_UPGRADE> set, this function assumes that the PV
3408 will expand when converted to UTF-8, and skips the extra work of checking for
3409 that. Typically this flag is used by a routine that has already parsed the
3410 string and found such characters, and passes this information on so that the
3411 work doesn't have to be repeated.
3413 Returns the number of bytes in the converted string.
3415 This is not a general purpose byte encoding to Unicode interface:
3416 use the Encode extension for that.
3418 =for apidoc sv_utf8_upgrade_flags_grow
3420 Like C<sv_utf8_upgrade_flags>, but has an additional parameter C<extra>, which is
3421 the number of unused bytes the string of C<sv> is guaranteed to have free after
3422 it upon return. This allows the caller to reserve extra space that it intends
3423 to fill, to avoid extra grows.
3425 C<sv_utf8_upgrade>, C<sv_utf8_upgrade_nomg>, and C<sv_utf8_upgrade_flags>
3426 are implemented in terms of this function.
3428 Returns the number of bytes in the converted string (not including the spares).
3432 (One might think that the calling routine could pass in the position of the
3433 first variant character when it has set SV_FORCE_UTF8_UPGRADE, so it wouldn't
3434 have to be found again. But that is not the case, because typically when the
3435 caller is likely to use this flag, it won't be calling this routine unless it
3436 finds something that won't fit into a byte. Otherwise it tries to not upgrade
3437 and just use bytes. But some things that do fit into a byte are variants in
3438 utf8, and the caller may not have been keeping track of these.)
3440 If the routine itself changes the string, it adds a trailing C<NUL>. Such a
3441 C<NUL> isn't guaranteed due to having other routines do the work in some input
3442 cases, or if the input is already flagged as being in utf8.
3444 The speed of this could perhaps be improved for many cases if someone wanted to
3445 write a fast function that counts the number of variant characters in a string,
3446 especially if it could return the position of the first one.
3451 Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV *const sv, const I32 flags, STRLEN extra)
3453 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
3455 if (sv == &PL_sv_undef)
3457 if (!SvPOK_nog(sv)) {
3459 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3460 (void) sv_2pv_flags(sv,&len, flags);
3462 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3466 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
3471 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3476 S_sv_uncow(aTHX_ sv, 0);
3479 if (IN_ENCODING && !(flags & SV_UTF8_NO_ENCODING)) {
3480 sv_recode_to_utf8(sv, _get_encoding());
3481 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3485 if (SvCUR(sv) == 0) {
3486 if (extra) SvGROW(sv, extra);
3487 } else { /* Assume Latin-1/EBCDIC */
3488 /* This function could be much more efficient if we
3489 * had a FLAG in SVs to signal if there are any variant
3490 * chars in the PV. Given that there isn't such a flag
3491 * make the loop as fast as possible (although there are certainly ways
3492 * to speed this up, eg. through vectorization) */
3493 U8 * s = (U8 *) SvPVX_const(sv);
3494 U8 * e = (U8 *) SvEND(sv);
3496 STRLEN two_byte_count = 0;
3498 if (flags & SV_FORCE_UTF8_UPGRADE) goto must_be_utf8;
3500 /* See if really will need to convert to utf8. We mustn't rely on our
3501 * incoming SV being well formed and having a trailing '\0', as certain
3502 * code in pp_formline can send us partially built SVs. */
3506 if (NATIVE_BYTE_IS_INVARIANT(ch)) continue;
3508 t--; /* t already incremented; re-point to first variant */
3513 /* utf8 conversion not needed because all are invariants. Mark as
3514 * UTF-8 even if no variant - saves scanning loop */
3516 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3521 /* Here, the string should be converted to utf8, either because of an
3522 * input flag (two_byte_count = 0), or because a character that
3523 * requires 2 bytes was found (two_byte_count = 1). t points either to
3524 * the beginning of the string (if we didn't examine anything), or to
3525 * the first variant. In either case, everything from s to t - 1 will
3526 * occupy only 1 byte each on output.
3528 * There are two main ways to convert. One is to create a new string
3529 * and go through the input starting from the beginning, appending each
3530 * converted value onto the new string as we go along. It's probably
3531 * best to allocate enough space in the string for the worst possible
3532 * case rather than possibly running out of space and having to
3533 * reallocate and then copy what we've done so far. Since everything
3534 * from s to t - 1 is invariant, the destination can be initialized
3535 * with these using a fast memory copy
3537 * The other way is to figure out exactly how big the string should be
3538 * by parsing the entire input. Then you don't have to make it big
3539 * enough to handle the worst possible case, and more importantly, if
3540 * the string you already have is large enough, you don't have to
3541 * allocate a new string, you can copy the last character in the input
3542 * string to the final position(s) that will be occupied by the
3543 * converted string and go backwards, stopping at t, since everything
3544 * before that is invariant.
3546 * There are advantages and disadvantages to each method.
3548 * In the first method, we can allocate a new string, do the memory
3549 * copy from the s to t - 1, and then proceed through the rest of the
3550 * string byte-by-byte.
3552 * In the second method, we proceed through the rest of the input
3553 * string just calculating how big the converted string will be. Then
3554 * there are two cases:
3555 * 1) if the string has enough extra space to handle the converted
3556 * value. We go backwards through the string, converting until we
3557 * get to the position we are at now, and then stop. If this
3558 * position is far enough along in the string, this method is
3559 * faster than the other method. If the memory copy were the same
3560 * speed as the byte-by-byte loop, that position would be about
3561 * half-way, as at the half-way mark, parsing to the end and back
3562 * is one complete string's parse, the same amount as starting
3563 * over and going all the way through. Actually, it would be
3564 * somewhat less than half-way, as it's faster to just count bytes
3565 * than to also copy, and we don't have the overhead of allocating
3566 * a new string, changing the scalar to use it, and freeing the
3567 * existing one. But if the memory copy is fast, the break-even
3568 * point is somewhere after half way. The counting loop could be
3569 * sped up by vectorization, etc, to move the break-even point
3570 * further towards the beginning.
3571 * 2) if the string doesn't have enough space to handle the converted
3572 * value. A new string will have to be allocated, and one might
3573 * as well, given that, start from the beginning doing the first
3574 * method. We've spent extra time parsing the string and in
3575 * exchange all we've gotten is that we know precisely how big to
3576 * make the new one. Perl is more optimized for time than space,
3577 * so this case is a loser.
3578 * So what I've decided to do is not use the 2nd method unless it is
3579 * guaranteed that a new string won't have to be allocated, assuming
3580 * the worst case. I also decided not to put any more conditions on it
3581 * than this, for now. It seems likely that, since the worst case is
3582 * twice as big as the unknown portion of the string (plus 1), we won't
3583 * be guaranteed enough space, causing us to go to the first method,
3584 * unless the string is short, or the first variant character is near
3585 * the end of it. In either of these cases, it seems best to use the
3586 * 2nd method. The only circumstance I can think of where this would
3587 * be really slower is if the string had once had much more data in it
3588 * than it does now, but there is still a substantial amount in it */
3591 STRLEN invariant_head = t - s;
3592 STRLEN size = invariant_head + (e - t) * 2 + 1 + extra;
3593 if (SvLEN(sv) < size) {
3595 /* Here, have decided to allocate a new string */
3600 Newx(dst, size, U8);
3602 /* If no known invariants at the beginning of the input string,
3603 * set so starts from there. Otherwise, can use memory copy to
3604 * get up to where we are now, and then start from here */
3606 if (invariant_head == 0) {
3609 Copy(s, dst, invariant_head, char);
3610 d = dst + invariant_head;
3614 append_utf8_from_native_byte(*t, &d);
3618 SvPV_free(sv); /* No longer using pre-existing string */
3619 SvPV_set(sv, (char*)dst);
3620 SvCUR_set(sv, d - dst);
3621 SvLEN_set(sv, size);
3624 /* Here, have decided to get the exact size of the string.
3625 * Currently this happens only when we know that there is
3626 * guaranteed enough space to fit the converted string, so
3627 * don't have to worry about growing. If two_byte_count is 0,
3628 * then t points to the first byte of the string which hasn't
3629 * been examined yet. Otherwise two_byte_count is 1, and t
3630 * points to the first byte in the string that will expand to
3631 * two. Depending on this, start examining at t or 1 after t.
3634 U8 *d = t + two_byte_count;
3637 /* Count up the remaining bytes that expand to two */
3640 const U8 chr = *d++;
3641 if (! NATIVE_BYTE_IS_INVARIANT(chr)) two_byte_count++;
3644 /* The string will expand by just the number of bytes that
3645 * occupy two positions. But we are one afterwards because of
3646 * the increment just above. This is the place to put the
3647 * trailing NUL, and to set the length before we decrement */
3649 d += two_byte_count;
3650 SvCUR_set(sv, d - s);
3654 /* Having decremented d, it points to the position to put the
3655 * very last byte of the expanded string. Go backwards through
3656 * the string, copying and expanding as we go, stopping when we
3657 * get to the part that is invariant the rest of the way down */
3661 if (NATIVE_BYTE_IS_INVARIANT(*e)) {
3664 *d-- = UTF8_EIGHT_BIT_LO(*e);
3665 *d-- = UTF8_EIGHT_BIT_HI(*e);
3671 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3672 /* Update pos. We do it at the end rather than during
3673 * the upgrade, to avoid slowing down the common case
3674 * (upgrade without pos).
3675 * pos can be stored as either bytes or characters. Since
3676 * this was previously a byte string we can just turn off
3677 * the bytes flag. */
3678 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3680 mg->mg_flags &= ~MGf_BYTES;
3682 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3683 magic_setutf8(sv,mg); /* clear UTF8 cache */
3688 /* Mark as UTF-8 even if no variant - saves scanning loop */
3694 =for apidoc sv_utf8_downgrade
3696 Attempts to convert the PV of an SV from characters to bytes.
3697 If the PV contains a character that cannot fit
3698 in a byte, this conversion will fail;
3699 in this case, either returns false or, if C<fail_ok> is not
3702 This is not a general purpose Unicode to byte encoding interface:
3703 use the C<Encode> extension for that.
3709 Perl_sv_utf8_downgrade(pTHX_ SV *const sv, const bool fail_ok)
3711 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3713 if (SvPOKp(sv) && SvUTF8(sv)) {
3717 int mg_flags = SV_GMAGIC;
3720 S_sv_uncow(aTHX_ sv, 0);
3722 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3724 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3725 if (mg && mg->mg_len > 0 && mg->mg_flags & MGf_BYTES) {
3726 mg->mg_len = sv_pos_b2u_flags(sv, mg->mg_len,
3727 SV_GMAGIC|SV_CONST_RETURN);
3728 mg_flags = 0; /* sv_pos_b2u does get magic */
3730 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3731 magic_setutf8(sv,mg); /* clear UTF8 cache */
3734 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3736 if (!utf8_to_bytes(s, &len)) {
3741 Perl_croak(aTHX_ "Wide character in %s",
3744 Perl_croak(aTHX_ "Wide character");
3755 =for apidoc sv_utf8_encode
3757 Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3758 flag off so that it looks like octets again.
3764 Perl_sv_utf8_encode(pTHX_ SV *const sv)
3766 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3768 if (SvREADONLY(sv)) {
3769 sv_force_normal_flags(sv, 0);
3771 (void) sv_utf8_upgrade(sv);
3776 =for apidoc sv_utf8_decode
3778 If the PV of the SV is an octet sequence in UTF-8
3779 and contains a multiple-byte character, the C<SvUTF8> flag is turned on
3780 so that it looks like a character. If the PV contains only single-byte
3781 characters, the C<SvUTF8> flag stays off.
3782 Scans PV for validity and returns false if the PV is invalid UTF-8.
3788 Perl_sv_utf8_decode(pTHX_ SV *const sv)
3790 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3793 const U8 *start, *c;
3796 /* The octets may have got themselves encoded - get them back as
3799 if (!sv_utf8_downgrade(sv, TRUE))
3802 /* it is actually just a matter of turning the utf8 flag on, but
3803 * we want to make sure everything inside is valid utf8 first.
3805 c = start = (const U8 *) SvPVX_const(sv);
3806 if (!is_utf8_string(c, SvCUR(sv)))
3808 e = (const U8 *) SvEND(sv);
3811 if (!UTF8_IS_INVARIANT(ch)) {
3816 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3817 /* XXX Is this dead code? XS_utf8_decode calls SvSETMAGIC
3818 after this, clearing pos. Does anything on CPAN
3820 /* adjust pos to the start of a UTF8 char sequence */
3821 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3823 I32 pos = mg->mg_len;
3825 for (c = start + pos; c > start; c--) {
3826 if (UTF8_IS_START(*c))
3829 mg->mg_len = c - start;
3832 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3833 magic_setutf8(sv,mg); /* clear UTF8 cache */
3840 =for apidoc sv_setsv
3842 Copies the contents of the source SV C<ssv> into the destination SV
3843 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3844 function if the source SV needs to be reused. Does not handle 'set' magic on
3845 destination SV. Calls 'get' magic on source SV. Loosely speaking, it
3846 performs a copy-by-value, obliterating any previous content of the
3849 You probably want to use one of the assortment of wrappers, such as
3850 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3851 C<SvSetMagicSV_nosteal>.
3853 =for apidoc sv_setsv_flags
3855 Copies the contents of the source SV C<ssv> into the destination SV
3856 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3857 function if the source SV needs to be reused. Does not handle 'set' magic.
3858 Loosely speaking, it performs a copy-by-value, obliterating any previous
3859 content of the destination.
3860 If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
3861 C<ssv> if appropriate, else not. If the C<flags>
3862 parameter has the C<SV_NOSTEAL> bit set then the
3863 buffers of temps will not be stolen. C<sv_setsv>
3864 and C<sv_setsv_nomg> are implemented in terms of this function.
3866 You probably want to use one of the assortment of wrappers, such as
3867 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3868 C<SvSetMagicSV_nosteal>.
3870 This is the primary function for copying scalars, and most other
3871 copy-ish functions and macros use this underneath.
3877 S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
3879 I32 mro_changes = 0; /* 1 = method, 2 = isa, 3 = recursive isa */
3880 HV *old_stash = NULL;
3882 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB;
3884 if (dtype != SVt_PVGV && !isGV_with_GP(dstr)) {
3885 const char * const name = GvNAME(sstr);
3886 const STRLEN len = GvNAMELEN(sstr);
3888 if (dtype >= SVt_PV) {
3894 SvUPGRADE(dstr, SVt_PVGV);
3895 (void)SvOK_off(dstr);
3896 isGV_with_GP_on(dstr);
3898 GvSTASH(dstr) = GvSTASH(sstr);
3900 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
3901 gv_name_set(MUTABLE_GV(dstr), name, len,
3902 GV_ADD | (GvNAMEUTF8(sstr) ? SVf_UTF8 : 0 ));
3903 SvFAKE_on(dstr); /* can coerce to non-glob */
3906 if(GvGP(MUTABLE_GV(sstr))) {
3907 /* If source has method cache entry, clear it */
3909 SvREFCNT_dec(GvCV(sstr));
3910 GvCV_set(sstr, NULL);
3913 /* If source has a real method, then a method is
3916 GvCV((const GV *)sstr) && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3922 /* If dest already had a real method, that's a change as well */
3924 !mro_changes && GvGP(MUTABLE_GV(dstr)) && GvCVu((const GV *)dstr)
3925 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3930 /* We don't need to check the name of the destination if it was not a
3931 glob to begin with. */
3932 if(dtype == SVt_PVGV) {
3933 const char * const name = GvNAME((const GV *)dstr);
3936 /* The stash may have been detached from the symbol table, so
3938 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3942 const STRLEN len = GvNAMELEN(dstr);
3943 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
3944 || (len == 1 && name[0] == ':')) {
3947 /* Set aside the old stash, so we can reset isa caches on
3949 if((old_stash = GvHV(dstr)))
3950 /* Make sure we do not lose it early. */
3951 SvREFCNT_inc_simple_void_NN(
3952 sv_2mortal((SV *)old_stash)
3957 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
3960 /* freeing dstr's GP might free sstr (e.g. *x = $x),
3961 * so temporarily protect it */
3963 SAVEFREESV(SvREFCNT_inc_simple_NN(sstr));
3964 gp_free(MUTABLE_GV(dstr));
3965 GvINTRO_off(dstr); /* one-shot flag */
3966 GvGP_set(dstr, gp_ref(GvGP(sstr)));
3969 if (SvTAINTED(sstr))
3971 if (GvIMPORTED(dstr) != GVf_IMPORTED
3972 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
3974 GvIMPORTED_on(dstr);
3977 if(mro_changes == 2) {
3978 if (GvAV((const GV *)sstr)) {
3980 SV * const sref = (SV *)GvAV((const GV *)dstr);
3981 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
3982 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
3983 AV * const ary = newAV();
3984 av_push(ary, mg->mg_obj); /* takes the refcount */
3985 mg->mg_obj = (SV *)ary;
3987 av_push((AV *)mg->mg_obj, SvREFCNT_inc_simple_NN(dstr));
3989 else sv_magic(sref, dstr, PERL_MAGIC_isa, NULL, 0);
3991 mro_isa_changed_in(GvSTASH(dstr));
3993 else if(mro_changes == 3) {
3994 HV * const stash = GvHV(dstr);
3995 if(old_stash ? (HV *)HvENAME_get(old_stash) : stash)
4001 else if(mro_changes) mro_method_changed_in(GvSTASH(dstr));
4002 if (GvIO(dstr) && dtype == SVt_PVGV) {
4003 DEBUG_o(Perl_deb(aTHX_
4004 "glob_assign_glob clearing PL_stashcache\n"));
4005 /* It's a cache. It will rebuild itself quite happily.
4006 It's a lot of effort to work out exactly which key (or keys)
4007 might be invalidated by the creation of the this file handle.
4009 hv_clear(PL_stashcache);
4015 Perl_gv_setref(pTHX_ SV *const dstr, SV *const sstr)
4017 SV * const sref = SvRV(sstr);
4019 const int intro = GvINTRO(dstr);
4022 const U32 stype = SvTYPE(sref);
4024 PERL_ARGS_ASSERT_GV_SETREF;
4027 GvINTRO_off(dstr); /* one-shot flag */
4028 GvLINE(dstr) = CopLINE(PL_curcop);
4029 GvEGV(dstr) = MUTABLE_GV(dstr);
4034 location = (SV **) &(GvGP(dstr)->gp_cv); /* XXX bypassing GvCV_set */
4035 import_flag = GVf_IMPORTED_CV;
4038 location = (SV **) &GvHV(dstr);
4039 import_flag = GVf_IMPORTED_HV;
4042 location = (SV **) &GvAV(dstr);
4043 import_flag = GVf_IMPORTED_AV;
4046 location = (SV **) &GvIOp(dstr);
4049 location = (SV **) &GvFORM(dstr);
4052 location = &GvSV(dstr);
4053 import_flag = GVf_IMPORTED_SV;
4056 if (stype == SVt_PVCV) {
4057 /*if (GvCVGEN(dstr) && (GvCV(dstr) != (const CV *)sref || GvCVGEN(dstr))) {*/
4058 if (GvCVGEN(dstr)) {
4059 SvREFCNT_dec(GvCV(dstr));
4060 GvCV_set(dstr, NULL);
4061 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4064 /* SAVEt_GVSLOT takes more room on the savestack and has more
4065 overhead in leave_scope than SAVEt_GENERIC_SV. But for CVs
4066 leave_scope needs access to the GV so it can reset method
4067 caches. We must use SAVEt_GVSLOT whenever the type is
4068 SVt_PVCV, even if the stash is anonymous, as the stash may
4069 gain a name somehow before leave_scope. */
4070 if (stype == SVt_PVCV) {
4071 /* There is no save_pushptrptrptr. Creating it for this
4072 one call site would be overkill. So inline the ss add
4076 SS_ADD_PTR(location);
4077 SS_ADD_PTR(SvREFCNT_inc(*location));
4078 SS_ADD_UV(SAVEt_GVSLOT);
4081 else SAVEGENERICSV(*location);
4084 if (stype == SVt_PVCV && (*location != sref || GvCVGEN(dstr))) {
4085 CV* const cv = MUTABLE_CV(*location);
4087 if (!GvCVGEN((const GV *)dstr) &&
4088 (CvROOT(cv) || CvXSUB(cv)) &&
4089 /* redundant check that avoids creating the extra SV
4090 most of the time: */
4091 (CvCONST(cv) || ckWARN(WARN_REDEFINE)))
4093 SV * const new_const_sv =
4094 CvCONST((const CV *)sref)
4095 ? cv_const_sv((const CV *)sref)
4097 report_redefined_cv(
4098 sv_2mortal(Perl_newSVpvf(aTHX_
4101 HvNAME_HEK(GvSTASH((const GV *)dstr))
4103 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr)))
4106 CvCONST((const CV *)sref) ? &new_const_sv : NULL
4110 cv_ckproto_len_flags(cv, (const GV *)dstr,
4111 SvPOK(sref) ? CvPROTO(sref) : NULL,
4112 SvPOK(sref) ? CvPROTOLEN(sref) : 0,
4113 SvPOK(sref) ? SvUTF8(sref) : 0);
4115 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4116 GvASSUMECV_on(dstr);
4117 if(GvSTASH(dstr)) { /* sub foo { 1 } sub bar { 2 } *bar = \&foo */
4118 if (intro && GvREFCNT(dstr) > 1) {
4119 /* temporary remove extra savestack's ref */
4121 gv_method_changed(dstr);
4124 else gv_method_changed(dstr);
4127 *location = SvREFCNT_inc_simple_NN(sref);
4128 if (import_flag && !(GvFLAGS(dstr) & import_flag)
4129 && CopSTASH_ne(PL_curcop, GvSTASH(dstr))) {
4130 GvFLAGS(dstr) |= import_flag;
4133 if (stype == SVt_PVHV) {
4134 const char * const name = GvNAME((GV*)dstr);
4135 const STRLEN len = GvNAMELEN(dstr);
4138 (len > 1 && name[len-2] == ':' && name[len-1] == ':')
4139 || (len == 1 && name[0] == ':')
4141 && (!dref || HvENAME_get(dref))
4144 (HV *)sref, (HV *)dref,
4150 stype == SVt_PVAV && sref != dref
4151 && strEQ(GvNAME((GV*)dstr), "ISA")
4152 /* The stash may have been detached from the symbol table, so
4153 check its name before doing anything. */
4154 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
4157 MAGIC * const omg = dref && SvSMAGICAL(dref)
4158 ? mg_find(dref, PERL_MAGIC_isa)
4160 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
4161 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
4162 AV * const ary = newAV();
4163 av_push(ary, mg->mg_obj); /* takes the refcount */
4164 mg->mg_obj = (SV *)ary;
4167 if (SvTYPE(omg->mg_obj) == SVt_PVAV) {
4168 SV **svp = AvARRAY((AV *)omg->mg_obj);
4169 I32 items = AvFILLp((AV *)omg->mg_obj) + 1;
4173 SvREFCNT_inc_simple_NN(*svp++)
4179 SvREFCNT_inc_simple_NN(omg->mg_obj)
4183 av_push((AV *)mg->mg_obj,SvREFCNT_inc_simple_NN(dstr));
4188 sref, omg ? omg->mg_obj : dstr, PERL_MAGIC_isa, NULL, 0
4190 mg = mg_find(sref, PERL_MAGIC_isa);
4192 /* Since the *ISA assignment could have affected more than
4193 one stash, don't call mro_isa_changed_in directly, but let
4194 magic_clearisa do it for us, as it already has the logic for
4195 dealing with globs vs arrays of globs. */
4197 Perl_magic_clearisa(aTHX_ NULL, mg);
4199 else if (stype == SVt_PVIO) {
4200 DEBUG_o(Perl_deb(aTHX_ "gv_setref clearing PL_stashcache\n"));
4201 /* It's a cache. It will rebuild itself quite happily.
4202 It's a lot of effort to work out exactly which key (or keys)
4203 might be invalidated by the creation of the this file handle.
4205 hv_clear(PL_stashcache);
4209 if (!intro) SvREFCNT_dec(dref);
4210 if (SvTAINTED(sstr))
4218 #ifdef PERL_DEBUG_READONLY_COW
4219 # include <sys/mman.h>
4221 # ifndef PERL_MEMORY_DEBUG_HEADER_SIZE
4222 # define PERL_MEMORY_DEBUG_HEADER_SIZE 0
4226 Perl_sv_buf_to_ro(pTHX_ SV *sv)
4228 struct perl_memory_debug_header * const header =
4229 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4230 const MEM_SIZE len = header->size;
4231 PERL_ARGS_ASSERT_SV_BUF_TO_RO;
4232 # ifdef PERL_TRACK_MEMPOOL
4233 if (!header->readonly) header->readonly = 1;
4235 if (mprotect(header, len, PROT_READ))
4236 Perl_warn(aTHX_ "mprotect RW for COW string %p %lu failed with %d",
4237 header, len, errno);
4241 S_sv_buf_to_rw(pTHX_ SV *sv)
4243 struct perl_memory_debug_header * const header =
4244 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4245 const MEM_SIZE len = header->size;
4246 PERL_ARGS_ASSERT_SV_BUF_TO_RW;
4247 if (mprotect(header, len, PROT_READ|PROT_WRITE))
4248 Perl_warn(aTHX_ "mprotect for COW string %p %lu failed with %d",
4249 header, len, errno);
4250 # ifdef PERL_TRACK_MEMPOOL
4251 header->readonly = 0;
4256 # define sv_buf_to_ro(sv) NOOP
4257 # define sv_buf_to_rw(sv) NOOP
4261 Perl_sv_setsv_flags(pTHX_ SV *dstr, SV* sstr, const I32 flags)
4267 PERL_ARGS_ASSERT_SV_SETSV_FLAGS;
4269 if (UNLIKELY( sstr == dstr ))
4272 if (SvIS_FREED(dstr)) {
4273 Perl_croak(aTHX_ "panic: attempt to copy value %" SVf
4274 " to a freed scalar %p", SVfARG(sstr), (void *)dstr);
4276 SV_CHECK_THINKFIRST_COW_DROP(dstr);
4277 if (UNLIKELY( !sstr ))
4278 sstr = &PL_sv_undef;
4279 if (SvIS_FREED(sstr)) {
4280 Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p to %p",
4281 (void*)sstr, (void*)dstr);
4283 stype = SvTYPE(sstr);
4284 dtype = SvTYPE(dstr);
4286 /* There's a lot of redundancy below but we're going for speed here */
4291 if (LIKELY( dtype != SVt_PVGV && dtype != SVt_PVLV )) {
4292 (void)SvOK_off(dstr);
4300 /* For performance, we inline promoting to type SVt_IV. */
4301 /* We're starting from SVt_NULL, so provided that define is
4302 * actual 0, we don't have to unset any SV type flags
4303 * to promote to SVt_IV. */
4304 STATIC_ASSERT_STMT(SVt_NULL == 0);
4305 SET_SVANY_FOR_BODYLESS_IV(dstr);
4306 SvFLAGS(dstr) |= SVt_IV;
4310 sv_upgrade(dstr, SVt_PVIV);
4314 goto end_of_first_switch;
4316 (void)SvIOK_only(dstr);
4317 SvIV_set(dstr, SvIVX(sstr));
4320 /* SvTAINTED can only be true if the SV has taint magic, which in
4321 turn means that the SV type is PVMG (or greater). This is the
4322 case statement for SVt_IV, so this cannot be true (whatever gcov
4324 assert(!SvTAINTED(sstr));
4329 if (dtype < SVt_PV && dtype != SVt_IV)
4330 sv_upgrade(dstr, SVt_IV);
4334 if (LIKELY( SvNOK(sstr) )) {
4338 sv_upgrade(dstr, SVt_NV);
4342 sv_upgrade(dstr, SVt_PVNV);
4346 goto end_of_first_switch;
4348 SvNV_set(dstr, SvNVX(sstr));
4349 (void)SvNOK_only(dstr);
4350 /* SvTAINTED can only be true if the SV has taint magic, which in
4351 turn means that the SV type is PVMG (or greater). This is the
4352 case statement for SVt_NV, so this cannot be true (whatever gcov
4354 assert(!SvTAINTED(sstr));
4361 sv_upgrade(dstr, SVt_PV);
4364 if (dtype < SVt_PVIV)
4365 sv_upgrade(dstr, SVt_PVIV);
4368 if (dtype < SVt_PVNV)
4369 sv_upgrade(dstr, SVt_PVNV);
4373 const char * const type = sv_reftype(sstr,0);
4375 /* diag_listed_as: Bizarre copy of %s */
4376 Perl_croak(aTHX_ "Bizarre copy of %s in %s", type, OP_DESC(PL_op));
4378 Perl_croak(aTHX_ "Bizarre copy of %s", type);
4380 NOT_REACHED; /* NOTREACHED */
4384 if (dtype < SVt_REGEXP)
4386 if (dtype >= SVt_PV) {
4392 sv_upgrade(dstr, SVt_REGEXP);
4400 if (SvGMAGICAL(sstr) && (flags & SV_GMAGIC)) {
4402 if (SvTYPE(sstr) != stype)
4403 stype = SvTYPE(sstr);
4405 if (isGV_with_GP(sstr) && dtype <= SVt_PVLV) {
4406 glob_assign_glob(dstr, sstr, dtype);
4409 if (stype == SVt_PVLV)
4411 if (isREGEXP(sstr)) goto upgregexp;
4412 SvUPGRADE(dstr, SVt_PVNV);
4415 SvUPGRADE(dstr, (svtype)stype);
4417 end_of_first_switch:
4419 /* dstr may have been upgraded. */
4420 dtype = SvTYPE(dstr);
4421 sflags = SvFLAGS(sstr);
4423 if (UNLIKELY( dtype == SVt_PVCV )) {
4424 /* Assigning to a subroutine sets the prototype. */
4427 const char *const ptr = SvPV_const(sstr, len);
4429 SvGROW(dstr, len + 1);
4430 Copy(ptr, SvPVX(dstr), len + 1, char);
4431 SvCUR_set(dstr, len);
4433 SvFLAGS(dstr) |= sflags & SVf_UTF8;
4434 CvAUTOLOAD_off(dstr);
4439 else if (UNLIKELY(dtype == SVt_PVAV || dtype == SVt_PVHV
4440 || dtype == SVt_PVFM))
4442 const char * const type = sv_reftype(dstr,0);
4444 /* diag_listed_as: Cannot copy to %s */
4445 Perl_croak(aTHX_ "Cannot copy to %s in %s", type, OP_DESC(PL_op));
4447 Perl_croak(aTHX_ "Cannot copy to %s", type);
4448 } else if (sflags & SVf_ROK) {
4449 if (isGV_with_GP(dstr)
4450 && SvTYPE(SvRV(sstr)) == SVt_PVGV && isGV_with_GP(SvRV(sstr))) {
4453 if (GvIMPORTED(dstr) != GVf_IMPORTED
4454 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
4456 GvIMPORTED_on(dstr);
4461 glob_assign_glob(dstr, sstr, dtype);
4465 if (dtype >= SVt_PV) {
4466 if (isGV_with_GP(dstr)) {
4467 gv_setref(dstr, sstr);
4470 if (SvPVX_const(dstr)) {
4476 (void)SvOK_off(dstr);
4477 SvRV_set(dstr, SvREFCNT_inc(SvRV(sstr)));
4478 SvFLAGS(dstr) |= sflags & SVf_ROK;
4479 assert(!(sflags & SVp_NOK));
4480 assert(!(sflags & SVp_IOK));
4481 assert(!(sflags & SVf_NOK));
4482 assert(!(sflags & SVf_IOK));
4484 else if (isGV_with_GP(dstr)) {
4485 if (!(sflags & SVf_OK)) {
4486 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4487 "Undefined value assigned to typeglob");
4490 GV *gv = gv_fetchsv_nomg(sstr, GV_ADD, SVt_PVGV);
4491 if (dstr != (const SV *)gv) {
4492 const char * const name = GvNAME((const GV *)dstr);
4493 const STRLEN len = GvNAMELEN(dstr);
4494 HV *old_stash = NULL;
4495 bool reset_isa = FALSE;
4496 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
4497 || (len == 1 && name[0] == ':')) {
4498 /* Set aside the old stash, so we can reset isa caches
4499 on its subclasses. */
4500 if((old_stash = GvHV(dstr))) {
4501 /* Make sure we do not lose it early. */
4502 SvREFCNT_inc_simple_void_NN(
4503 sv_2mortal((SV *)old_stash)
4510 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
4511 gp_free(MUTABLE_GV(dstr));
4513 GvGP_set(dstr, gp_ref(GvGP(gv)));
4516 HV * const stash = GvHV(dstr);
4518 old_stash ? (HV *)HvENAME_get(old_stash) : stash
4528 else if ((dtype == SVt_REGEXP || dtype == SVt_PVLV)
4529 && (stype == SVt_REGEXP || isREGEXP(sstr))) {
4530 reg_temp_copy((REGEXP*)dstr, (REGEXP*)sstr);
4532 else if (sflags & SVp_POK) {
4533 const STRLEN cur = SvCUR(sstr);
4534 const STRLEN len = SvLEN(sstr);
4537 * We have three basic ways to copy the string:
4543 * Which we choose is based on various factors. The following
4544 * things are listed in order of speed, fastest to slowest:
4546 * - Copying a short string
4547 * - Copy-on-write bookkeeping
4549 * - Copying a long string
4551 * We swipe the string (steal the string buffer) if the SV on the
4552 * rhs is about to be freed anyway (TEMP and refcnt==1). This is a
4553 * big win on long strings. It should be a win on short strings if
4554 * SvPVX_const(dstr) has to be allocated. If not, it should not
4555 * slow things down, as SvPVX_const(sstr) would have been freed
4558 * We also steal the buffer from a PADTMP (operator target) if it
4559 * is ‘long enough’. For short strings, a swipe does not help
4560 * here, as it causes more malloc calls the next time the target
4561 * is used. Benchmarks show that even if SvPVX_const(dstr) has to
4562 * be allocated it is still not worth swiping PADTMPs for short
4563 * strings, as the savings here are small.
4565 * If swiping is not an option, then we see whether it is
4566 * worth using copy-on-write. If the lhs already has a buf-
4567 * fer big enough and the string is short, we skip it and fall back
4568 * to method 3, since memcpy is faster for short strings than the
4569 * later bookkeeping overhead that copy-on-write entails.
4571 * If the rhs is not a copy-on-write string yet, then we also
4572 * consider whether the buffer is too large relative to the string
4573 * it holds. Some operations such as readline allocate a large
4574 * buffer in the expectation of reusing it. But turning such into
4575 * a COW buffer is counter-productive because it increases memory
4576 * usage by making readline allocate a new large buffer the sec-
4577 * ond time round. So, if the buffer is too large, again, we use
4580 * Finally, if there is no buffer on the left, or the buffer is too
4581 * small, then we use copy-on-write and make both SVs share the
4586 /* Whichever path we take through the next code, we want this true,
4587 and doing it now facilitates the COW check. */
4588 (void)SvPOK_only(dstr);
4592 /* slated for free anyway (and not COW)? */
4593 (sflags & (SVs_TEMP|SVf_IsCOW)) == SVs_TEMP
4594 /* or a swipable TARG */
4596 (SVs_PADTMP|SVf_READONLY|SVf_PROTECT|SVf_IsCOW))
4598 /* whose buffer is worth stealing */
4599 && CHECK_COWBUF_THRESHOLD(cur,len)
4602 !(sflags & SVf_OOK) && /* and not involved in OOK hack? */
4603 (!(flags & SV_NOSTEAL)) &&
4604 /* and we're allowed to steal temps */
4605 SvREFCNT(sstr) == 1 && /* and no other references to it? */
4606 len) /* and really is a string */
4607 { /* Passes the swipe test. */
4608 if (SvPVX_const(dstr)) /* we know that dtype >= SVt_PV */
4610 SvPV_set(dstr, SvPVX_mutable(sstr));
4611 SvLEN_set(dstr, SvLEN(sstr));
4612 SvCUR_set(dstr, SvCUR(sstr));
4615 (void)SvOK_off(sstr); /* NOTE: nukes most SvFLAGS on sstr */
4616 SvPV_set(sstr, NULL);
4621 else if (flags & SV_COW_SHARED_HASH_KEYS
4623 #ifdef PERL_COPY_ON_WRITE
4626 ( (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4627 /* If this is a regular (non-hek) COW, only so
4628 many COW "copies" are possible. */
4629 && CowREFCNT(sstr) != SV_COW_REFCNT_MAX ))
4630 : ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4631 && !(SvFLAGS(dstr) & SVf_BREAK)
4632 && CHECK_COW_THRESHOLD(cur,len) && cur+1 < len
4633 && (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4637 && !(SvFLAGS(dstr) & SVf_BREAK)
4640 /* Either it's a shared hash key, or it's suitable for
4643 PerlIO_printf(Perl_debug_log, "Copy on write: sstr --> dstr\n");
4648 if (!(sflags & SVf_IsCOW)) {
4650 CowREFCNT(sstr) = 0;
4653 if (SvPVX_const(dstr)) { /* we know that dtype >= SVt_PV */
4659 if (sflags & SVf_IsCOW) {
4663 SvPV_set(dstr, SvPVX_mutable(sstr));
4668 /* SvIsCOW_shared_hash */
4669 DEBUG_C(PerlIO_printf(Perl_debug_log,
4670 "Copy on write: Sharing hash\n"));
4672 assert (SvTYPE(dstr) >= SVt_PV);
4674 HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)))));
4676 SvLEN_set(dstr, len);
4677 SvCUR_set(dstr, cur);
4680 /* Failed the swipe test, and we cannot do copy-on-write either.
4681 Have to copy the string. */
4682 SvGROW(dstr, cur + 1); /* inlined from sv_setpvn */
4683 Move(SvPVX_const(sstr),SvPVX(dstr),cur,char);
4684 SvCUR_set(dstr, cur);
4685 *SvEND(dstr) = '\0';
4687 if (sflags & SVp_NOK) {
4688 SvNV_set(dstr, SvNVX(sstr));
4690 if (sflags & SVp_IOK) {
4691 SvIV_set(dstr, SvIVX(sstr));
4692 /* Must do this otherwise some other overloaded use of 0x80000000
4693 gets confused. I guess SVpbm_VALID */
4694 if (sflags & SVf_IVisUV)
4697 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_NOK|SVp_NOK|SVf_UTF8);
4699 const MAGIC * const smg = SvVSTRING_mg(sstr);
4701 sv_magic(dstr, NULL, PERL_MAGIC_vstring,
4702 smg->mg_ptr, smg->mg_len);
4703 SvRMAGICAL_on(dstr);
4707 else if (sflags & (SVp_IOK|SVp_NOK)) {
4708 (void)SvOK_off(dstr);
4709 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_IVisUV|SVf_NOK|SVp_NOK);
4710 if (sflags & SVp_IOK) {
4711 /* XXXX Do we want to set IsUV for IV(ROK)? Be extra safe... */
4712 SvIV_set(dstr, SvIVX(sstr));
4714 if (sflags & SVp_NOK) {
4715 SvNV_set(dstr, SvNVX(sstr));
4719 if (isGV_with_GP(sstr)) {
4720 gv_efullname3(dstr, MUTABLE_GV(sstr), "*");
4723 (void)SvOK_off(dstr);
4725 if (SvTAINTED(sstr))
4730 =for apidoc sv_setsv_mg
4732 Like C<sv_setsv>, but also handles 'set' magic.
4738 Perl_sv_setsv_mg(pTHX_ SV *const dstr, SV *const sstr)
4740 PERL_ARGS_ASSERT_SV_SETSV_MG;
4742 sv_setsv(dstr,sstr);
4747 # define SVt_COW SVt_PV
4749 Perl_sv_setsv_cow(pTHX_ SV *dstr, SV *sstr)
4751 STRLEN cur = SvCUR(sstr);
4752 STRLEN len = SvLEN(sstr);
4754 #if defined(PERL_DEBUG_READONLY_COW) && defined(PERL_COPY_ON_WRITE)
4755 const bool already = cBOOL(SvIsCOW(sstr));
4758 PERL_ARGS_ASSERT_SV_SETSV_COW;
4761 PerlIO_printf(Perl_debug_log, "Fast copy on write: %p -> %p\n",
4762 (void*)sstr, (void*)dstr);
4769 if (SvTHINKFIRST(dstr))
4770 sv_force_normal_flags(dstr, SV_COW_DROP_PV);
4771 else if (SvPVX_const(dstr))
4772 Safefree(SvPVX_mutable(dstr));
4776 SvUPGRADE(dstr, SVt_COW);
4778 assert (SvPOK(sstr));
4779 assert (SvPOKp(sstr));
4781 if (SvIsCOW(sstr)) {
4783 if (SvLEN(sstr) == 0) {
4784 /* source is a COW shared hash key. */
4785 DEBUG_C(PerlIO_printf(Perl_debug_log,
4786 "Fast copy on write: Sharing hash\n"));
4787 new_pv = HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr))));
4790 assert(SvCUR(sstr)+1 < SvLEN(sstr));
4791 assert(CowREFCNT(sstr) < SV_COW_REFCNT_MAX);
4793 assert ((SvFLAGS(sstr) & CAN_COW_MASK) == CAN_COW_FLAGS);
4794 SvUPGRADE(sstr, SVt_COW);
4796 DEBUG_C(PerlIO_printf(Perl_debug_log,
4797 "Fast copy on write: Converting sstr to COW\n"));
4798 CowREFCNT(sstr) = 0;
4800 # ifdef PERL_DEBUG_READONLY_COW
4801 if (already) sv_buf_to_rw(sstr);
4804 new_pv = SvPVX_mutable(sstr);
4808 SvPV_set(dstr, new_pv);
4809 SvFLAGS(dstr) = (SVt_COW|SVf_POK|SVp_POK|SVf_IsCOW);
4812 SvLEN_set(dstr, len);
4813 SvCUR_set(dstr, cur);
4822 =for apidoc sv_setpvn
4824 Copies a string (possibly containing embedded C<NUL> characters) into an SV.
4825 The C<len> parameter indicates the number of
4826 bytes to be copied. If the C<ptr> argument is NULL the SV will become
4827 undefined. Does not handle 'set' magic. See C<L</sv_setpvn_mg>>.
4833 Perl_sv_setpvn(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4837 PERL_ARGS_ASSERT_SV_SETPVN;
4839 SV_CHECK_THINKFIRST_COW_DROP(sv);
4845 /* len is STRLEN which is unsigned, need to copy to signed */
4848 Perl_croak(aTHX_ "panic: sv_setpvn called with negative strlen %"
4851 SvUPGRADE(sv, SVt_PV);
4853 dptr = SvGROW(sv, len + 1);
4854 Move(ptr,dptr,len,char);
4857 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4859 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4863 =for apidoc sv_setpvn_mg
4865 Like C<sv_setpvn>, but also handles 'set' magic.
4871 Perl_sv_setpvn_mg(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4873 PERL_ARGS_ASSERT_SV_SETPVN_MG;
4875 sv_setpvn(sv,ptr,len);
4880 =for apidoc sv_setpv
4882 Copies a string into an SV. The string must be terminated with a C<NUL>
4884 Does not handle 'set' magic. See C<L</sv_setpv_mg>>.
4890 Perl_sv_setpv(pTHX_ SV *const sv, const char *const ptr)
4894 PERL_ARGS_ASSERT_SV_SETPV;
4896 SV_CHECK_THINKFIRST_COW_DROP(sv);
4902 SvUPGRADE(sv, SVt_PV);
4904 SvGROW(sv, len + 1);
4905 Move(ptr,SvPVX(sv),len+1,char);
4907 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4909 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4913 =for apidoc sv_setpv_mg
4915 Like C<sv_setpv>, but also handles 'set' magic.
4921 Perl_sv_setpv_mg(pTHX_ SV *const sv, const char *const ptr)
4923 PERL_ARGS_ASSERT_SV_SETPV_MG;
4930 Perl_sv_sethek(pTHX_ SV *const sv, const HEK *const hek)
4932 PERL_ARGS_ASSERT_SV_SETHEK;
4938 if (HEK_LEN(hek) == HEf_SVKEY) {
4939 sv_setsv(sv, *(SV**)HEK_KEY(hek));
4942 const int flags = HEK_FLAGS(hek);
4943 if (flags & HVhek_WASUTF8) {
4944 STRLEN utf8_len = HEK_LEN(hek);
4945 char *as_utf8 = (char *)bytes_to_utf8((U8*)HEK_KEY(hek), &utf8_len);
4946 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
4949 } else if (flags & HVhek_UNSHARED) {
4950 sv_setpvn(sv, HEK_KEY(hek), HEK_LEN(hek));
4953 else SvUTF8_off(sv);
4957 SV_CHECK_THINKFIRST_COW_DROP(sv);
4958 SvUPGRADE(sv, SVt_PV);
4960 SvPV_set(sv,(char *)HEK_KEY(share_hek_hek(hek)));
4961 SvCUR_set(sv, HEK_LEN(hek));
4967 else SvUTF8_off(sv);
4975 =for apidoc sv_usepvn_flags
4977 Tells an SV to use C<ptr> to find its string value. Normally the
4978 string is stored inside the SV, but sv_usepvn allows the SV to use an
4979 outside string. C<ptr> should point to memory that was allocated
4980 by L<C<Newx>|perlclib/Memory Management and String Handling>. It must be
4981 the start of a C<Newx>-ed block of memory, and not a pointer to the
4982 middle of it (beware of L<C<OOK>|perlguts/Offsets> and copy-on-write),
4983 and not be from a non-C<Newx> memory allocator like C<malloc>. The
4984 string length, C<len>, must be supplied. By default this function
4985 will C<Renew> (i.e. realloc, move) the memory pointed to by C<ptr>,
4986 so that pointer should not be freed or used by the programmer after
4987 giving it to C<sv_usepvn>, and neither should any pointers from "behind"
4988 that pointer (e.g. ptr + 1) be used.
4990 If S<C<flags & SV_SMAGIC>> is true, will call C<SvSETMAGIC>. If
4991 S<C<flags> & SV_HAS_TRAILING_NUL>> is true, then C<ptr[len]> must be C<NUL>,
4993 will be skipped (i.e. the buffer is actually at least 1 byte longer than
4994 C<len>, and already meets the requirements for storing in C<SvPVX>).
5000 Perl_sv_usepvn_flags(pTHX_ SV *const sv, char *ptr, const STRLEN len, const U32 flags)
5004 PERL_ARGS_ASSERT_SV_USEPVN_FLAGS;
5006 SV_CHECK_THINKFIRST_COW_DROP(sv);
5007 SvUPGRADE(sv, SVt_PV);
5010 if (flags & SV_SMAGIC)
5014 if (SvPVX_const(sv))
5018 if (flags & SV_HAS_TRAILING_NUL)
5019 assert(ptr[len] == '\0');
5022 allocate = (flags & SV_HAS_TRAILING_NUL)
5024 #ifdef Perl_safesysmalloc_size
5027 PERL_STRLEN_ROUNDUP(len + 1);
5029 if (flags & SV_HAS_TRAILING_NUL) {
5030 /* It's long enough - do nothing.
5031 Specifically Perl_newCONSTSUB is relying on this. */
5034 /* Force a move to shake out bugs in callers. */
5035 char *new_ptr = (char*)safemalloc(allocate);
5036 Copy(ptr, new_ptr, len, char);
5037 PoisonFree(ptr,len,char);
5041 ptr = (char*) saferealloc (ptr, allocate);
5044 #ifdef Perl_safesysmalloc_size
5045 SvLEN_set(sv, Perl_safesysmalloc_size(ptr));
5047 SvLEN_set(sv, allocate);
5051 if (!(flags & SV_HAS_TRAILING_NUL)) {
5054 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5056 if (flags & SV_SMAGIC)
5061 =for apidoc sv_force_normal_flags
5063 Undo various types of fakery on an SV, where fakery means
5064 "more than" a string: if the PV is a shared string, make
5065 a private copy; if we're a ref, stop refing; if we're a glob, downgrade to
5066 an C<xpvmg>; if we're a copy-on-write scalar, this is the on-write time when
5067 we do the copy, and is also used locally; if this is a
5068 vstring, drop the vstring magic. If C<SV_COW_DROP_PV> is set
5069 then a copy-on-write scalar drops its PV buffer (if any) and becomes
5070 C<SvPOK_off> rather than making a copy. (Used where this
5071 scalar is about to be set to some other value.) In addition,
5072 the C<flags> parameter gets passed to C<sv_unref_flags()>
5073 when unreffing. C<sv_force_normal> calls this function
5074 with flags set to 0.
5076 This function is expected to be used to signal to perl that this SV is
5077 about to be written to, and any extra book-keeping needs to be taken care
5078 of. Hence, it croaks on read-only values.
5084 S_sv_uncow(pTHX_ SV * const sv, const U32 flags)
5086 assert(SvIsCOW(sv));
5089 const char * const pvx = SvPVX_const(sv);
5090 const STRLEN len = SvLEN(sv);
5091 const STRLEN cur = SvCUR(sv);
5094 PerlIO_printf(Perl_debug_log,
5095 "Copy on write: Force normal %ld\n",
5100 # ifdef PERL_COPY_ON_WRITE
5102 /* Must do this first, since the CowREFCNT uses SvPVX and
5103 we need to write to CowREFCNT, or de-RO the whole buffer if we are
5104 the only owner left of the buffer. */
5105 sv_buf_to_rw(sv); /* NOOP if RO-ing not supported */
5107 U8 cowrefcnt = CowREFCNT(sv);
5108 if(cowrefcnt != 0) {
5110 CowREFCNT(sv) = cowrefcnt;
5115 /* Else we are the only owner of the buffer. */
5120 /* This SV doesn't own the buffer, so need to Newx() a new one: */
5125 if (flags & SV_COW_DROP_PV) {
5126 /* OK, so we don't need to copy our buffer. */
5129 SvGROW(sv, cur + 1);
5130 Move(pvx,SvPVX(sv),cur,char);
5136 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5143 const char * const pvx = SvPVX_const(sv);
5144 const STRLEN len = SvCUR(sv);
5148 if (flags & SV_COW_DROP_PV) {
5149 /* OK, so we don't need to copy our buffer. */
5152 SvGROW(sv, len + 1);
5153 Move(pvx,SvPVX(sv),len,char);
5156 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5162 Perl_sv_force_normal_flags(pTHX_ SV *const sv, const U32 flags)
5164 PERL_ARGS_ASSERT_SV_FORCE_NORMAL_FLAGS;
5167 Perl_croak_no_modify();
5168 else if (SvIsCOW(sv) && LIKELY(SvTYPE(sv) != SVt_PVHV))
5169 S_sv_uncow(aTHX_ sv, flags);
5171 sv_unref_flags(sv, flags);
5172 else if (SvFAKE(sv) && isGV_with_GP(sv))
5173 sv_unglob(sv, flags);
5174 else if (SvFAKE(sv) && isREGEXP(sv)) {
5175 /* Need to downgrade the REGEXP to a simple(r) scalar. This is analogous
5176 to sv_unglob. We only need it here, so inline it. */
5177 const bool islv = SvTYPE(sv) == SVt_PVLV;
5178 const svtype new_type =
5179 islv ? SVt_NULL : SvMAGIC(sv) || SvSTASH(sv) ? SVt_PVMG : SVt_PV;
5180 SV *const temp = newSV_type(new_type);
5181 regexp *const temp_p = ReANY((REGEXP *)sv);
5183 if (new_type == SVt_PVMG) {
5184 SvMAGIC_set(temp, SvMAGIC(sv));
5185 SvMAGIC_set(sv, NULL);
5186 SvSTASH_set(temp, SvSTASH(sv));
5187 SvSTASH_set(sv, NULL);
5189 if (!islv) SvCUR_set(temp, SvCUR(sv));
5190 /* Remember that SvPVX is in the head, not the body. But
5191 RX_WRAPPED is in the body. */
5192 assert(ReANY((REGEXP *)sv)->mother_re);
5193 /* Their buffer is already owned by someone else. */
5194 if (flags & SV_COW_DROP_PV) {
5195 /* SvLEN is already 0. For SVt_REGEXP, we have a brand new
5196 zeroed body. For SVt_PVLV, it should have been set to 0
5197 before turning into a regexp. */
5198 assert(!SvLEN(islv ? sv : temp));
5199 sv->sv_u.svu_pv = 0;
5202 sv->sv_u.svu_pv = savepvn(RX_WRAPPED((REGEXP *)sv), SvCUR(sv));
5203 SvLEN_set(islv ? sv : temp, SvCUR(sv)+1);
5207 /* Now swap the rest of the bodies. */
5211 SvFLAGS(sv) &= ~SVTYPEMASK;
5212 SvFLAGS(sv) |= new_type;
5213 SvANY(sv) = SvANY(temp);
5216 SvFLAGS(temp) &= ~(SVTYPEMASK);
5217 SvFLAGS(temp) |= SVt_REGEXP|SVf_FAKE;
5218 SvANY(temp) = temp_p;
5219 temp->sv_u.svu_rx = (regexp *)temp_p;
5221 SvREFCNT_dec_NN(temp);
5223 else if (SvVOK(sv)) sv_unmagic(sv, PERL_MAGIC_vstring);
5229 Efficient removal of characters from the beginning of the string buffer.
5230 C<SvPOK(sv)>, or at least C<SvPOKp(sv)>, must be true and C<ptr> must be a
5231 pointer to somewhere inside the string buffer. C<ptr> becomes the first
5232 character of the adjusted string. Uses the C<OOK> hack. On return, only
5233 C<SvPOK(sv)> and C<SvPOKp(sv)> among the C<OK> flags will be true.
5235 Beware: after this function returns, C<ptr> and SvPVX_const(sv) may no longer
5236 refer to the same chunk of data.
5238 The unfortunate similarity of this function's name to that of Perl's C<chop>
5239 operator is strictly coincidental. This function works from the left;
5240 C<chop> works from the right.
5246 Perl_sv_chop(pTHX_ SV *const sv, const char *const ptr)
5257 PERL_ARGS_ASSERT_SV_CHOP;
5259 if (!ptr || !SvPOKp(sv))
5261 delta = ptr - SvPVX_const(sv);
5263 /* Nothing to do. */
5266 max_delta = SvLEN(sv) ? SvLEN(sv) : SvCUR(sv);
5267 if (delta > max_delta)
5268 Perl_croak(aTHX_ "panic: sv_chop ptr=%p, start=%p, end=%p",
5269 ptr, SvPVX_const(sv), SvPVX_const(sv) + max_delta);
5270 /* SvPVX(sv) may move in SV_CHECK_THINKFIRST(sv), so don't use ptr any more */
5271 SV_CHECK_THINKFIRST(sv);
5272 SvPOK_only_UTF8(sv);
5275 if (!SvLEN(sv)) { /* make copy of shared string */
5276 const char *pvx = SvPVX_const(sv);
5277 const STRLEN len = SvCUR(sv);
5278 SvGROW(sv, len + 1);
5279 Move(pvx,SvPVX(sv),len,char);
5285 SvOOK_offset(sv, old_delta);
5287 SvLEN_set(sv, SvLEN(sv) - delta);
5288 SvCUR_set(sv, SvCUR(sv) - delta);
5289 SvPV_set(sv, SvPVX(sv) + delta);
5291 p = (U8 *)SvPVX_const(sv);
5294 /* how many bytes were evacuated? we will fill them with sentinel
5295 bytes, except for the part holding the new offset of course. */
5298 evacn += (old_delta < 0x100 ? 1 : 1 + sizeof(STRLEN));
5300 assert(evacn <= delta + old_delta);
5304 /* This sets 'delta' to the accumulated value of all deltas so far */
5308 /* If 'delta' fits in a byte, store it just prior to the new beginning of
5309 * the string; otherwise store a 0 byte there and store 'delta' just prior
5310 * to that, using as many bytes as a STRLEN occupies. Thus it overwrites a
5311 * portion of the chopped part of the string */
5312 if (delta < 0x100) {
5316 p -= sizeof(STRLEN);
5317 Copy((U8*)&delta, p, sizeof(STRLEN), U8);
5321 /* Fill the preceding buffer with sentinals to verify that no-one is
5331 =for apidoc sv_catpvn
5333 Concatenates the string onto the end of the string which is in the SV.
5334 C<len> indicates number of bytes to copy. If the SV has the UTF-8
5335 status set, then the bytes appended should be valid UTF-8.
5336 Handles 'get' magic, but not 'set' magic. See C<L</sv_catpvn_mg>>.
5338 =for apidoc sv_catpvn_flags
5340 Concatenates the string onto the end of the string which is in the SV. The
5341 C<len> indicates number of bytes to copy.
5343 By default, the string appended is assumed to be valid UTF-8 if the SV has
5344 the UTF-8 status set, and a string of bytes otherwise. One can force the
5345 appended string to be interpreted as UTF-8 by supplying the C<SV_CATUTF8>
5346 flag, and as bytes by supplying the C<SV_CATBYTES> flag; the SV or the
5347 string appended will be upgraded to UTF-8 if necessary.
5349 If C<flags> has the C<SV_SMAGIC> bit set, will
5350 C<mg_set> on C<dsv> afterwards if appropriate.
5351 C<sv_catpvn> and C<sv_catpvn_nomg> are implemented
5352 in terms of this function.
5358 Perl_sv_catpvn_flags(pTHX_ SV *const dsv, const char *sstr, const STRLEN slen, const I32 flags)
5361 const char * const dstr = SvPV_force_flags(dsv, dlen, flags);
5363 PERL_ARGS_ASSERT_SV_CATPVN_FLAGS;
5364 assert((flags & (SV_CATBYTES|SV_CATUTF8)) != (SV_CATBYTES|SV_CATUTF8));
5366 if (!(flags & SV_CATBYTES) || !SvUTF8(dsv)) {
5367 if (flags & SV_CATUTF8 && !SvUTF8(dsv)) {
5368 sv_utf8_upgrade_flags_grow(dsv, 0, slen + 1);
5371 else SvGROW(dsv, dlen + slen + 1);
5373 sstr = SvPVX_const(dsv);
5374 Move(sstr, SvPVX(dsv) + dlen, slen, char);
5375 SvCUR_set(dsv, SvCUR(dsv) + slen);
5378 /* We inline bytes_to_utf8, to avoid an extra malloc. */
5379 const char * const send = sstr + slen;
5382 /* Something this code does not account for, which I think is
5383 impossible; it would require the same pv to be treated as
5384 bytes *and* utf8, which would indicate a bug elsewhere. */
5385 assert(sstr != dstr);
5387 SvGROW(dsv, dlen + slen * 2 + 1);
5388 d = (U8 *)SvPVX(dsv) + dlen;
5390 while (sstr < send) {
5391 append_utf8_from_native_byte(*sstr, &d);
5394 SvCUR_set(dsv, d-(const U8 *)SvPVX(dsv));
5397 (void)SvPOK_only_UTF8(dsv); /* validate pointer */
5399 if (flags & SV_SMAGIC)
5404 =for apidoc sv_catsv
5406 Concatenates the string from SV C<ssv> onto the end of the string in SV
5407 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5408 Handles 'get' magic on both SVs, but no 'set' magic. See C<L</sv_catsv_mg>>
5409 and C<L</sv_catsv_nomg>>.
5411 =for apidoc sv_catsv_flags
5413 Concatenates the string from SV C<ssv> onto the end of the string in SV
5414 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5415 If C<flags> has the C<SV_GMAGIC> bit set, will call C<mg_get> on both SVs if
5416 appropriate. If C<flags> has the C<SV_SMAGIC> bit set, C<mg_set> will be called on
5417 the modified SV afterward, if appropriate. C<sv_catsv>, C<sv_catsv_nomg>,
5418 and C<sv_catsv_mg> are implemented in terms of this function.
5423 Perl_sv_catsv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
5425 PERL_ARGS_ASSERT_SV_CATSV_FLAGS;
5429 const char *spv = SvPV_flags_const(ssv, slen, flags);
5430 if (flags & SV_GMAGIC)
5432 sv_catpvn_flags(dsv, spv, slen,
5433 DO_UTF8(ssv) ? SV_CATUTF8 : SV_CATBYTES);
5434 if (flags & SV_SMAGIC)
5440 =for apidoc sv_catpv
5442 Concatenates the C<NUL>-terminated string onto the end of the string which is
5444 If the SV has the UTF-8 status set, then the bytes appended should be
5445 valid UTF-8. Handles 'get' magic, but not 'set' magic. See
5451 Perl_sv_catpv(pTHX_ SV *const sv, const char *ptr)
5457 PERL_ARGS_ASSERT_SV_CATPV;
5461 junk = SvPV_force(sv, tlen);
5463 SvGROW(sv, tlen + len + 1);
5465 ptr = SvPVX_const(sv);
5466 Move(ptr,SvPVX(sv)+tlen,len+1,char);
5467 SvCUR_set(sv, SvCUR(sv) + len);
5468 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5473 =for apidoc sv_catpv_flags
5475 Concatenates the C<NUL>-terminated string onto the end of the string which is
5477 If the SV has the UTF-8 status set, then the bytes appended should
5478 be valid UTF-8. If C<flags> has the C<SV_SMAGIC> bit set, will C<mg_set>
5479 on the modified SV if appropriate.
5485 Perl_sv_catpv_flags(pTHX_ SV *dstr, const char *sstr, const I32 flags)
5487 PERL_ARGS_ASSERT_SV_CATPV_FLAGS;
5488 sv_catpvn_flags(dstr, sstr, strlen(sstr), flags);
5492 =for apidoc sv_catpv_mg
5494 Like C<sv_catpv>, but also handles 'set' magic.
5500 Perl_sv_catpv_mg(pTHX_ SV *const sv, const char *const ptr)
5502 PERL_ARGS_ASSERT_SV_CATPV_MG;
5511 Creates a new SV. A non-zero C<len> parameter indicates the number of
5512 bytes of preallocated string space the SV should have. An extra byte for a
5513 trailing C<NUL> is also reserved. (C<SvPOK> is not set for the SV even if string
5514 space is allocated.) The reference count for the new SV is set to 1.
5516 In 5.9.3, C<newSV()> replaces the older C<NEWSV()> API, and drops the first
5517 parameter, I<x>, a debug aid which allowed callers to identify themselves.
5518 This aid has been superseded by a new build option, C<PERL_MEM_LOG> (see
5519 L<perlhacktips/PERL_MEM_LOG>). The older API is still there for use in XS
5520 modules supporting older perls.
5526 Perl_newSV(pTHX_ const STRLEN len)
5532 sv_grow(sv, len + 1);
5537 =for apidoc sv_magicext
5539 Adds magic to an SV, upgrading it if necessary. Applies the
5540 supplied C<vtable> and returns a pointer to the magic added.
5542 Note that C<sv_magicext> will allow things that C<sv_magic> will not.
5543 In particular, you can add magic to C<SvREADONLY> SVs, and add more than
5544 one instance of the same C<how>.
5546 If C<namlen> is greater than zero then a C<savepvn> I<copy> of C<name> is
5547 stored, if C<namlen> is zero then C<name> is stored as-is and - as another
5548 special case - if C<(name && namlen == HEf_SVKEY)> then C<name> is assumed
5549 to contain an SV* and is stored as-is with its C<REFCNT> incremented.
5551 (This is now used as a subroutine by C<sv_magic>.)
5556 Perl_sv_magicext(pTHX_ SV *const sv, SV *const obj, const int how,
5557 const MGVTBL *const vtable, const char *const name, const I32 namlen)
5561 PERL_ARGS_ASSERT_SV_MAGICEXT;
5563 SvUPGRADE(sv, SVt_PVMG);
5564 Newxz(mg, 1, MAGIC);
5565 mg->mg_moremagic = SvMAGIC(sv);
5566 SvMAGIC_set(sv, mg);
5568 /* Sometimes a magic contains a reference loop, where the sv and
5569 object refer to each other. To prevent a reference loop that
5570 would prevent such objects being freed, we look for such loops
5571 and if we find one we avoid incrementing the object refcount.
5573 Note we cannot do this to avoid self-tie loops as intervening RV must
5574 have its REFCNT incremented to keep it in existence.
5577 if (!obj || obj == sv ||
5578 how == PERL_MAGIC_arylen ||
5579 how == PERL_MAGIC_symtab ||
5580 (SvTYPE(obj) == SVt_PVGV &&
5581 (GvSV(obj) == sv || GvHV(obj) == (const HV *)sv
5582 || GvAV(obj) == (const AV *)sv || GvCV(obj) == (const CV *)sv
5583 || GvIOp(obj) == (const IO *)sv || GvFORM(obj) == (const CV *)sv)))
5588 mg->mg_obj = SvREFCNT_inc_simple(obj);
5589 mg->mg_flags |= MGf_REFCOUNTED;
5592 /* Normal self-ties simply pass a null object, and instead of
5593 using mg_obj directly, use the SvTIED_obj macro to produce a
5594 new RV as needed. For glob "self-ties", we are tieing the PVIO
5595 with an RV obj pointing to the glob containing the PVIO. In
5596 this case, to avoid a reference loop, we need to weaken the
5600 if (how == PERL_MAGIC_tiedscalar && SvTYPE(sv) == SVt_PVIO &&
5601 obj && SvROK(obj) && GvIO(SvRV(obj)) == (const IO *)sv)
5607 mg->mg_len = namlen;
5610 mg->mg_ptr = savepvn(name, namlen);
5611 else if (namlen == HEf_SVKEY) {
5612 /* Yes, this is casting away const. This is only for the case of
5613 HEf_SVKEY. I think we need to document this aberation of the
5614 constness of the API, rather than making name non-const, as
5615 that change propagating outwards a long way. */
5616 mg->mg_ptr = (char*)SvREFCNT_inc_simple_NN((SV *)name);
5618 mg->mg_ptr = (char *) name;
5620 mg->mg_virtual = (MGVTBL *) vtable;
5627 Perl_sv_magicext_mglob(pTHX_ SV *sv)
5629 PERL_ARGS_ASSERT_SV_MAGICEXT_MGLOB;
5630 if (SvTYPE(sv) == SVt_PVLV && LvTYPE(sv) == 'y') {
5631 /* This sv is only a delegate. //g magic must be attached to
5636 return sv_magicext(sv, NULL, PERL_MAGIC_regex_global,
5637 &PL_vtbl_mglob, 0, 0);
5641 =for apidoc sv_magic
5643 Adds magic to an SV. First upgrades C<sv> to type C<SVt_PVMG> if
5644 necessary, then adds a new magic item of type C<how> to the head of the
5647 See C<L</sv_magicext>> (which C<sv_magic> now calls) for a description of the
5648 handling of the C<name> and C<namlen> arguments.
5650 You need to use C<sv_magicext> to add magic to C<SvREADONLY> SVs and also
5651 to add more than one instance of the same C<how>.
5657 Perl_sv_magic(pTHX_ SV *const sv, SV *const obj, const int how,
5658 const char *const name, const I32 namlen)
5660 const MGVTBL *vtable;
5663 unsigned int vtable_index;
5665 PERL_ARGS_ASSERT_SV_MAGIC;
5667 if (how < 0 || (unsigned)how >= C_ARRAY_LENGTH(PL_magic_data)
5668 || ((flags = PL_magic_data[how]),
5669 (vtable_index = flags & PERL_MAGIC_VTABLE_MASK)
5670 > magic_vtable_max))
5671 Perl_croak(aTHX_ "Don't know how to handle magic of type \\%o", how);
5673 /* PERL_MAGIC_ext is reserved for use by extensions not perl internals.
5674 Useful for attaching extension internal data to perl vars.
5675 Note that multiple extensions may clash if magical scalars
5676 etc holding private data from one are passed to another. */
5678 vtable = (vtable_index == magic_vtable_max)
5679 ? NULL : PL_magic_vtables + vtable_index;
5681 if (SvREADONLY(sv)) {
5683 !PERL_MAGIC_TYPE_READONLY_ACCEPTABLE(how)
5686 Perl_croak_no_modify();
5689 if (SvMAGICAL(sv) || (how == PERL_MAGIC_taint && SvTYPE(sv) >= SVt_PVMG)) {
5690 if (SvMAGIC(sv) && (mg = mg_find(sv, how))) {
5691 /* sv_magic() refuses to add a magic of the same 'how' as an
5694 if (how == PERL_MAGIC_taint)
5700 /* Force pos to be stored as characters, not bytes. */
5701 if (SvMAGICAL(sv) && DO_UTF8(sv)
5702 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
5704 && mg->mg_flags & MGf_BYTES) {
5705 mg->mg_len = (SSize_t)sv_pos_b2u_flags(sv, (STRLEN)mg->mg_len,
5707 mg->mg_flags &= ~MGf_BYTES;
5710 /* Rest of work is done else where */
5711 mg = sv_magicext(sv,obj,how,vtable,name,namlen);
5714 case PERL_MAGIC_taint:
5717 case PERL_MAGIC_ext:
5718 case PERL_MAGIC_dbfile:
5725 S_sv_unmagicext_flags(pTHX_ SV *const sv, const int type, MGVTBL *vtbl, const U32 flags)
5732 if (SvTYPE(sv) < SVt_PVMG || !SvMAGIC(sv))
5734 mgp = &(((XPVMG*) SvANY(sv))->xmg_u.xmg_magic);
5735 for (mg = *mgp; mg; mg = *mgp) {
5736 const MGVTBL* const virt = mg->mg_virtual;
5737 if (mg->mg_type == type && (!flags || virt == vtbl)) {
5738 *mgp = mg->mg_moremagic;
5739 if (virt && virt->svt_free)
5740 virt->svt_free(aTHX_ sv, mg);
5741 if (mg->mg_ptr && mg->mg_type != PERL_MAGIC_regex_global) {
5743 Safefree(mg->mg_ptr);
5744 else if (mg->mg_len == HEf_SVKEY)
5745 SvREFCNT_dec(MUTABLE_SV(mg->mg_ptr));
5746 else if (mg->mg_type == PERL_MAGIC_utf8)
5747 Safefree(mg->mg_ptr);
5749 if (mg->mg_flags & MGf_REFCOUNTED)
5750 SvREFCNT_dec(mg->mg_obj);
5754 mgp = &mg->mg_moremagic;
5757 if (SvMAGICAL(sv)) /* if we're under save_magic, wait for restore_magic; */
5758 mg_magical(sv); /* else fix the flags now */
5762 SvFLAGS(sv) |= (SvFLAGS(sv) & (SVp_IOK|SVp_NOK|SVp_POK)) >> PRIVSHIFT;
5768 =for apidoc sv_unmagic
5770 Removes all magic of type C<type> from an SV.
5776 Perl_sv_unmagic(pTHX_ SV *const sv, const int type)
5778 PERL_ARGS_ASSERT_SV_UNMAGIC;
5779 return S_sv_unmagicext_flags(aTHX_ sv, type, NULL, 0);
5783 =for apidoc sv_unmagicext
5785 Removes all magic of type C<type> with the specified C<vtbl> from an SV.
5791 Perl_sv_unmagicext(pTHX_ SV *const sv, const int type, MGVTBL *vtbl)
5793 PERL_ARGS_ASSERT_SV_UNMAGICEXT;
5794 return S_sv_unmagicext_flags(aTHX_ sv, type, vtbl, 1);
5798 =for apidoc sv_rvweaken
5800 Weaken a reference: set the C<SvWEAKREF> flag on this RV; give the
5801 referred-to SV C<PERL_MAGIC_backref> magic if it hasn't already; and
5802 push a back-reference to this RV onto the array of backreferences
5803 associated with that magic. If the RV is magical, set magic will be
5804 called after the RV is cleared.
5810 Perl_sv_rvweaken(pTHX_ SV *const sv)
5814 PERL_ARGS_ASSERT_SV_RVWEAKEN;
5816 if (!SvOK(sv)) /* let undefs pass */
5819 Perl_croak(aTHX_ "Can't weaken a nonreference");
5820 else if (SvWEAKREF(sv)) {
5821 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is already weak");
5824 else if (SvREADONLY(sv)) croak_no_modify();
5826 Perl_sv_add_backref(aTHX_ tsv, sv);
5828 SvREFCNT_dec_NN(tsv);
5833 =for apidoc sv_get_backrefs
5835 If C<sv> is the target of a weak reference then it returns the back
5836 references structure associated with the sv; otherwise return C<NULL>.
5838 When returning a non-null result the type of the return is relevant. If it
5839 is an AV then the elements of the AV are the weak reference RVs which
5840 point at this item. If it is any other type then the item itself is the
5843 See also C<Perl_sv_add_backref()>, C<Perl_sv_del_backref()>,
5844 C<Perl_sv_kill_backrefs()>
5850 Perl_sv_get_backrefs(SV *const sv)
5854 PERL_ARGS_ASSERT_SV_GET_BACKREFS;
5856 /* find slot to store array or singleton backref */
5858 if (SvTYPE(sv) == SVt_PVHV) {
5860 struct xpvhv_aux * const iter = HvAUX((HV *)sv);
5861 backrefs = (SV *)iter->xhv_backreferences;
5863 } else if (SvMAGICAL(sv)) {
5864 MAGIC *mg = mg_find(sv, PERL_MAGIC_backref);
5866 backrefs = mg->mg_obj;
5871 /* Give tsv backref magic if it hasn't already got it, then push a
5872 * back-reference to sv onto the array associated with the backref magic.
5874 * As an optimisation, if there's only one backref and it's not an AV,
5875 * store it directly in the HvAUX or mg_obj slot, avoiding the need to
5876 * allocate an AV. (Whether the slot holds an AV tells us whether this is
5880 /* A discussion about the backreferences array and its refcount:
5882 * The AV holding the backreferences is pointed to either as the mg_obj of
5883 * PERL_MAGIC_backref, or in the specific case of a HV, from the
5884 * xhv_backreferences field. The array is created with a refcount
5885 * of 2. This means that if during global destruction the array gets
5886 * picked on before its parent to have its refcount decremented by the
5887 * random zapper, it won't actually be freed, meaning it's still there for
5888 * when its parent gets freed.
5890 * When the parent SV is freed, the extra ref is killed by
5891 * Perl_sv_kill_backrefs. The other ref is killed, in the case of magic,
5892 * by mg_free() / MGf_REFCOUNTED, or for a hash, by Perl_hv_kill_backrefs.
5894 * When a single backref SV is stored directly, it is not reference
5899 Perl_sv_add_backref(pTHX_ SV *const tsv, SV *const sv)
5905 PERL_ARGS_ASSERT_SV_ADD_BACKREF;
5907 /* find slot to store array or singleton backref */
5909 if (SvTYPE(tsv) == SVt_PVHV) {
5910 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
5913 mg = mg_find(tsv, PERL_MAGIC_backref);
5915 mg = sv_magicext(tsv, NULL, PERL_MAGIC_backref, &PL_vtbl_backref, NULL, 0);
5916 svp = &(mg->mg_obj);
5919 /* create or retrieve the array */
5921 if ( (!*svp && SvTYPE(sv) == SVt_PVAV)
5922 || (*svp && SvTYPE(*svp) != SVt_PVAV)
5926 mg->mg_flags |= MGf_REFCOUNTED;
5929 SvREFCNT_inc_simple_void_NN(av);
5930 /* av now has a refcnt of 2; see discussion above */
5931 av_extend(av, *svp ? 2 : 1);
5933 /* move single existing backref to the array */
5934 AvARRAY(av)[++AvFILLp(av)] = *svp; /* av_push() */
5939 av = MUTABLE_AV(*svp);
5941 /* optimisation: store single backref directly in HvAUX or mg_obj */
5945 assert(SvTYPE(av) == SVt_PVAV);
5946 if (AvFILLp(av) >= AvMAX(av)) {
5947 av_extend(av, AvFILLp(av)+1);
5950 /* push new backref */
5951 AvARRAY(av)[++AvFILLp(av)] = sv; /* av_push() */
5954 /* delete a back-reference to ourselves from the backref magic associated
5955 * with the SV we point to.
5959 Perl_sv_del_backref(pTHX_ SV *const tsv, SV *const sv)
5963 PERL_ARGS_ASSERT_SV_DEL_BACKREF;
5965 if (SvTYPE(tsv) == SVt_PVHV) {
5967 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
5969 else if (SvIS_FREED(tsv) && PL_phase == PERL_PHASE_DESTRUCT) {
5970 /* It's possible for the the last (strong) reference to tsv to have
5971 become freed *before* the last thing holding a weak reference.
5972 If both survive longer than the backreferences array, then when
5973 the referent's reference count drops to 0 and it is freed, it's
5974 not able to chase the backreferences, so they aren't NULLed.
5976 For example, a CV holds a weak reference to its stash. If both the
5977 CV and the stash survive longer than the backreferences array,
5978 and the CV gets picked for the SvBREAK() treatment first,
5979 *and* it turns out that the stash is only being kept alive because
5980 of an our variable in the pad of the CV, then midway during CV
5981 destruction the stash gets freed, but CvSTASH() isn't set to NULL.
5982 It ends up pointing to the freed HV. Hence it's chased in here, and
5983 if this block wasn't here, it would hit the !svp panic just below.
5985 I don't believe that "better" destruction ordering is going to help
5986 here - during global destruction there's always going to be the
5987 chance that something goes out of order. We've tried to make it
5988 foolproof before, and it only resulted in evolutionary pressure on
5989 fools. Which made us look foolish for our hubris. :-(
5995 = SvMAGICAL(tsv) ? mg_find(tsv, PERL_MAGIC_backref) : NULL;
5996 svp = mg ? &(mg->mg_obj) : NULL;
6000 Perl_croak(aTHX_ "panic: del_backref, svp=0");
6002 /* It's possible that sv is being freed recursively part way through the
6003 freeing of tsv. If this happens, the backreferences array of tsv has
6004 already been freed, and so svp will be NULL. If this is the case,
6005 we should not panic. Instead, nothing needs doing, so return. */
6006 if (PL_phase == PERL_PHASE_DESTRUCT && SvREFCNT(tsv) == 0)
6008 Perl_croak(aTHX_ "panic: del_backref, *svp=%p phase=%s refcnt=%" UVuf,
6009 (void*)*svp, PL_phase_names[PL_phase], (UV)SvREFCNT(tsv));
6012 if (SvTYPE(*svp) == SVt_PVAV) {
6016 AV * const av = (AV*)*svp;
6018 assert(!SvIS_FREED(av));
6022 /* for an SV with N weak references to it, if all those
6023 * weak refs are deleted, then sv_del_backref will be called
6024 * N times and O(N^2) compares will be done within the backref
6025 * array. To ameliorate this potential slowness, we:
6026 * 1) make sure this code is as tight as possible;
6027 * 2) when looking for SV, look for it at both the head and tail of the
6028 * array first before searching the rest, since some create/destroy
6029 * patterns will cause the backrefs to be freed in order.
6036 SV **p = &svp[fill];
6037 SV *const topsv = *p;
6044 /* We weren't the last entry.
6045 An unordered list has this property that you
6046 can take the last element off the end to fill
6047 the hole, and it's still an unordered list :-)
6053 break; /* should only be one */
6060 AvFILLp(av) = fill-1;
6062 else if (SvIS_FREED(*svp) && PL_phase == PERL_PHASE_DESTRUCT) {
6063 /* freed AV; skip */
6066 /* optimisation: only a single backref, stored directly */
6068 Perl_croak(aTHX_ "panic: del_backref, *svp=%p, sv=%p",
6069 (void*)*svp, (void*)sv);
6076 Perl_sv_kill_backrefs(pTHX_ SV *const sv, AV *const av)
6082 PERL_ARGS_ASSERT_SV_KILL_BACKREFS;
6087 /* after multiple passes through Perl_sv_clean_all() for a thingy
6088 * that has badly leaked, the backref array may have gotten freed,
6089 * since we only protect it against 1 round of cleanup */
6090 if (SvIS_FREED(av)) {
6091 if (PL_in_clean_all) /* All is fair */
6094 "panic: magic_killbackrefs (freed backref AV/SV)");
6098 is_array = (SvTYPE(av) == SVt_PVAV);
6100 assert(!SvIS_FREED(av));
6103 last = svp + AvFILLp(av);
6106 /* optimisation: only a single backref, stored directly */
6112 while (svp <= last) {
6114 SV *const referrer = *svp;
6115 if (SvWEAKREF(referrer)) {
6116 /* XXX Should we check that it hasn't changed? */
6117 assert(SvROK(referrer));
6118 SvRV_set(referrer, 0);
6120 SvWEAKREF_off(referrer);
6121 SvSETMAGIC(referrer);
6122 } else if (SvTYPE(referrer) == SVt_PVGV ||
6123 SvTYPE(referrer) == SVt_PVLV) {
6124 assert(SvTYPE(sv) == SVt_PVHV); /* stash backref */
6125 /* You lookin' at me? */
6126 assert(GvSTASH(referrer));
6127 assert(GvSTASH(referrer) == (const HV *)sv);
6128 GvSTASH(referrer) = 0;
6129 } else if (SvTYPE(referrer) == SVt_PVCV ||
6130 SvTYPE(referrer) == SVt_PVFM) {
6131 if (SvTYPE(sv) == SVt_PVHV) { /* stash backref */
6132 /* You lookin' at me? */
6133 assert(CvSTASH(referrer));
6134 assert(CvSTASH(referrer) == (const HV *)sv);
6135 SvANY(MUTABLE_CV(referrer))->xcv_stash = 0;
6138 assert(SvTYPE(sv) == SVt_PVGV);
6139 /* You lookin' at me? */
6140 assert(CvGV(referrer));
6141 assert(CvGV(referrer) == (const GV *)sv);
6142 anonymise_cv_maybe(MUTABLE_GV(sv),
6143 MUTABLE_CV(referrer));
6148 "panic: magic_killbackrefs (flags=%"UVxf")",
6149 (UV)SvFLAGS(referrer));
6160 SvREFCNT_dec_NN(av); /* remove extra count added by sv_add_backref() */
6166 =for apidoc sv_insert
6168 Inserts a string at the specified offset/length within the SV. Similar to
6169 the Perl C<substr()> function. Handles get magic.
6171 =for apidoc sv_insert_flags
6173 Same as C<sv_insert>, but the extra C<flags> are passed to the
6174 C<SvPV_force_flags> that applies to C<bigstr>.
6180 Perl_sv_insert_flags(pTHX_ SV *const bigstr, const STRLEN offset, const STRLEN len, const char *const little, const STRLEN littlelen, const U32 flags)
6186 SSize_t i; /* better be sizeof(STRLEN) or bad things happen */
6189 PERL_ARGS_ASSERT_SV_INSERT_FLAGS;
6191 SvPV_force_flags(bigstr, curlen, flags);
6192 (void)SvPOK_only_UTF8(bigstr);
6193 if (offset + len > curlen) {
6194 SvGROW(bigstr, offset+len+1);
6195 Zero(SvPVX(bigstr)+curlen, offset+len-curlen, char);
6196 SvCUR_set(bigstr, offset+len);
6200 i = littlelen - len;
6201 if (i > 0) { /* string might grow */
6202 big = SvGROW(bigstr, SvCUR(bigstr) + i + 1);
6203 mid = big + offset + len;
6204 midend = bigend = big + SvCUR(bigstr);
6207 while (midend > mid) /* shove everything down */
6208 *--bigend = *--midend;
6209 Move(little,big+offset,littlelen,char);
6210 SvCUR_set(bigstr, SvCUR(bigstr) + i);
6215 Move(little,SvPVX(bigstr)+offset,len,char);
6220 big = SvPVX(bigstr);
6223 bigend = big + SvCUR(bigstr);
6225 if (midend > bigend)
6226 Perl_croak(aTHX_ "panic: sv_insert, midend=%p, bigend=%p",
6229 if (mid - big > bigend - midend) { /* faster to shorten from end */
6231 Move(little, mid, littlelen,char);
6234 i = bigend - midend;
6236 Move(midend, mid, i,char);
6240 SvCUR_set(bigstr, mid - big);
6242 else if ((i = mid - big)) { /* faster from front */
6243 midend -= littlelen;
6245 Move(big, midend - i, i, char);
6246 sv_chop(bigstr,midend-i);
6248 Move(little, mid, littlelen,char);
6250 else if (littlelen) {
6251 midend -= littlelen;
6252 sv_chop(bigstr,midend);
6253 Move(little,midend,littlelen,char);
6256 sv_chop(bigstr,midend);
6262 =for apidoc sv_replace
6264 Make the first argument a copy of the second, then delete the original.
6265 The target SV physically takes over ownership of the body of the source SV
6266 and inherits its flags; however, the target keeps any magic it owns,
6267 and any magic in the source is discarded.
6268 Note that this is a rather specialist SV copying operation; most of the
6269 time you'll want to use C<sv_setsv> or one of its many macro front-ends.
6275 Perl_sv_replace(pTHX_ SV *const sv, SV *const nsv)
6277 const U32 refcnt = SvREFCNT(sv);
6279 PERL_ARGS_ASSERT_SV_REPLACE;
6281 SV_CHECK_THINKFIRST_COW_DROP(sv);
6282 if (SvREFCNT(nsv) != 1) {
6283 Perl_croak(aTHX_ "panic: reference miscount on nsv in sv_replace()"
6284 " (%" UVuf " != 1)", (UV) SvREFCNT(nsv));
6286 if (SvMAGICAL(sv)) {
6290 sv_upgrade(nsv, SVt_PVMG);
6291 SvMAGIC_set(nsv, SvMAGIC(sv));
6292 SvFLAGS(nsv) |= SvMAGICAL(sv);
6294 SvMAGIC_set(sv, NULL);
6298 assert(!SvREFCNT(sv));
6299 #ifdef DEBUG_LEAKING_SCALARS
6300 sv->sv_flags = nsv->sv_flags;
6301 sv->sv_any = nsv->sv_any;
6302 sv->sv_refcnt = nsv->sv_refcnt;
6303 sv->sv_u = nsv->sv_u;
6305 StructCopy(nsv,sv,SV);
6307 if(SvTYPE(sv) == SVt_IV) {
6308 SET_SVANY_FOR_BODYLESS_IV(sv);
6312 SvREFCNT(sv) = refcnt;
6313 SvFLAGS(nsv) |= SVTYPEMASK; /* Mark as freed */
6318 /* We're about to free a GV which has a CV that refers back to us.
6319 * If that CV will outlive us, make it anonymous (i.e. fix up its CvGV
6323 S_anonymise_cv_maybe(pTHX_ GV *gv, CV* cv)
6328 PERL_ARGS_ASSERT_ANONYMISE_CV_MAYBE;
6331 assert(SvREFCNT(gv) == 0);
6332 assert(isGV(gv) && isGV_with_GP(gv));
6334 assert(!CvANON(cv));
6335 assert(CvGV(cv) == gv);
6336 assert(!CvNAMED(cv));
6338 /* will the CV shortly be freed by gp_free() ? */
6339 if (GvCV(gv) == cv && GvGP(gv)->gp_refcnt < 2 && SvREFCNT(cv) < 2) {
6340 SvANY(cv)->xcv_gv_u.xcv_gv = NULL;
6344 /* if not, anonymise: */
6345 gvname = (GvSTASH(gv) && HvNAME(GvSTASH(gv)) && HvENAME(GvSTASH(gv)))
6346 ? newSVhek(HvENAME_HEK(GvSTASH(gv)))
6347 : newSVpvn_flags( "__ANON__", 8, 0 );
6348 sv_catpvs(gvname, "::__ANON__");
6349 anongv = gv_fetchsv(gvname, GV_ADDMULTI, SVt_PVCV);
6350 SvREFCNT_dec_NN(gvname);
6354 SvANY(cv)->xcv_gv_u.xcv_gv = MUTABLE_GV(SvREFCNT_inc(anongv));
6359 =for apidoc sv_clear
6361 Clear an SV: call any destructors, free up any memory used by the body,
6362 and free the body itself. The SV's head is I<not> freed, although
6363 its type is set to all 1's so that it won't inadvertently be assumed
6364 to be live during global destruction etc.
6365 This function should only be called when C<REFCNT> is zero. Most of the time
6366 you'll want to call C<sv_free()> (or its macro wrapper C<SvREFCNT_dec>)
6373 Perl_sv_clear(pTHX_ SV *const orig_sv)
6378 const struct body_details *sv_type_details;
6382 STRLEN hash_index = 0; /* initialise to make Coverity et al happy.
6383 Not strictly necessary */
6385 PERL_ARGS_ASSERT_SV_CLEAR;
6387 /* within this loop, sv is the SV currently being freed, and
6388 * iter_sv is the most recent AV or whatever that's being iterated
6389 * over to provide more SVs */
6395 assert(SvREFCNT(sv) == 0);
6396 assert(SvTYPE(sv) != (svtype)SVTYPEMASK);
6398 if (type <= SVt_IV) {
6399 /* See the comment in sv.h about the collusion between this
6400 * early return and the overloading of the NULL slots in the
6404 SvFLAGS(sv) &= SVf_BREAK;
6405 SvFLAGS(sv) |= SVTYPEMASK;
6409 /* objs are always >= MG, but pad names use the SVs_OBJECT flag
6410 for another purpose */
6411 assert(!SvOBJECT(sv) || type >= SVt_PVMG);
6413 if (type >= SVt_PVMG) {
6415 if (!curse(sv, 1)) goto get_next_sv;
6416 type = SvTYPE(sv); /* destructor may have changed it */
6418 /* Free back-references before magic, in case the magic calls
6419 * Perl code that has weak references to sv. */
6420 if (type == SVt_PVHV) {
6421 Perl_hv_kill_backrefs(aTHX_ MUTABLE_HV(sv));
6425 else if (SvMAGIC(sv)) {
6426 /* Free back-references before other types of magic. */
6427 sv_unmagic(sv, PERL_MAGIC_backref);
6433 /* case SVt_INVLIST: */
6436 IoIFP(sv) != PerlIO_stdin() &&
6437 IoIFP(sv) != PerlIO_stdout() &&
6438 IoIFP(sv) != PerlIO_stderr() &&
6439 !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6441 io_close(MUTABLE_IO(sv), NULL, FALSE,
6442 (IoTYPE(sv) == IoTYPE_WRONLY ||
6443 IoTYPE(sv) == IoTYPE_RDWR ||
6444 IoTYPE(sv) == IoTYPE_APPEND));
6446 if (IoDIRP(sv) && !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6447 PerlDir_close(IoDIRP(sv));
6448 IoDIRP(sv) = (DIR*)NULL;
6449 Safefree(IoTOP_NAME(sv));
6450 Safefree(IoFMT_NAME(sv));
6451 Safefree(IoBOTTOM_NAME(sv));
6452 if ((const GV *)sv == PL_statgv)
6456 /* FIXME for plugins */
6458 pregfree2((REGEXP*) sv);
6462 cv_undef(MUTABLE_CV(sv));
6463 /* If we're in a stash, we don't own a reference to it.
6464 * However it does have a back reference to us, which needs to
6466 if ((stash = CvSTASH(sv)))
6467 sv_del_backref(MUTABLE_SV(stash), sv);
6470 if (PL_last_swash_hv == (const HV *)sv) {
6471 PL_last_swash_hv = NULL;
6473 if (HvTOTALKEYS((HV*)sv) > 0) {
6475 /* this statement should match the one at the beginning of
6476 * hv_undef_flags() */
6477 if ( PL_phase != PERL_PHASE_DESTRUCT
6478 && (hek = HvNAME_HEK((HV*)sv)))
6480 if (PL_stashcache) {
6481 DEBUG_o(Perl_deb(aTHX_
6482 "sv_clear clearing PL_stashcache for '%"HEKf
6485 (void)hv_deletehek(PL_stashcache,
6488 hv_name_set((HV*)sv, NULL, 0, 0);
6491 /* save old iter_sv in unused SvSTASH field */
6492 assert(!SvOBJECT(sv));
6493 SvSTASH(sv) = (HV*)iter_sv;
6496 /* save old hash_index in unused SvMAGIC field */
6497 assert(!SvMAGICAL(sv));
6498 assert(!SvMAGIC(sv));
6499 ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index = hash_index;
6502 next_sv = Perl_hfree_next_entry(aTHX_ (HV*)sv, &hash_index);
6503 goto get_next_sv; /* process this new sv */
6505 /* free empty hash */
6506 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6507 assert(!HvARRAY((HV*)sv));
6511 AV* av = MUTABLE_AV(sv);
6512 if (PL_comppad == av) {
6516 if (AvREAL(av) && AvFILLp(av) > -1) {
6517 next_sv = AvARRAY(av)[AvFILLp(av)--];
6518 /* save old iter_sv in top-most slot of AV,
6519 * and pray that it doesn't get wiped in the meantime */
6520 AvARRAY(av)[AvMAX(av)] = iter_sv;
6522 goto get_next_sv; /* process this new sv */
6524 Safefree(AvALLOC(av));
6529 if (LvTYPE(sv) == 'T') { /* for tie: return HE to pool */
6530 SvREFCNT_dec(HeKEY_sv((HE*)LvTARG(sv)));
6531 HeNEXT((HE*)LvTARG(sv)) = PL_hv_fetch_ent_mh;
6532 PL_hv_fetch_ent_mh = (HE*)LvTARG(sv);
6534 else if (LvTYPE(sv) != 't') /* unless tie: unrefcnted fake SV** */
6535 SvREFCNT_dec(LvTARG(sv));
6536 if (isREGEXP(sv)) goto freeregexp;
6539 if (isGV_with_GP(sv)) {
6540 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
6541 && HvENAME_get(stash))
6542 mro_method_changed_in(stash);
6543 gp_free(MUTABLE_GV(sv));
6545 unshare_hek(GvNAME_HEK(sv));
6546 /* If we're in a stash, we don't own a reference to it.
6547 * However it does have a back reference to us, which
6548 * needs to be cleared. */
6549 if (!SvVALID(sv) && (stash = GvSTASH(sv)))
6550 sv_del_backref(MUTABLE_SV(stash), sv);
6552 /* FIXME. There are probably more unreferenced pointers to SVs
6553 * in the interpreter struct that we should check and tidy in
6554 * a similar fashion to this: */
6555 /* See also S_sv_unglob, which does the same thing. */
6556 if ((const GV *)sv == PL_last_in_gv)
6557 PL_last_in_gv = NULL;
6558 else if ((const GV *)sv == PL_statgv)
6560 else if ((const GV *)sv == PL_stderrgv)
6569 /* Don't bother with SvOOK_off(sv); as we're only going to
6573 SvOOK_offset(sv, offset);
6574 SvPV_set(sv, SvPVX_mutable(sv) - offset);
6575 /* Don't even bother with turning off the OOK flag. */
6580 SV * const target = SvRV(sv);
6582 sv_del_backref(target, sv);
6588 else if (SvPVX_const(sv)
6589 && !(SvTYPE(sv) == SVt_PVIO
6590 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6594 PerlIO_printf(Perl_debug_log, "Copy on write: clear\n");
6598 if (CowREFCNT(sv)) {
6605 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6610 Safefree(SvPVX_mutable(sv));
6614 else if (SvPVX_const(sv) && SvLEN(sv)
6615 && !(SvTYPE(sv) == SVt_PVIO
6616 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6617 Safefree(SvPVX_mutable(sv));
6618 else if (SvPVX_const(sv) && SvIsCOW(sv)) {
6619 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6629 SvFLAGS(sv) &= SVf_BREAK;
6630 SvFLAGS(sv) |= SVTYPEMASK;
6632 sv_type_details = bodies_by_type + type;
6633 if (sv_type_details->arena) {
6634 del_body(((char *)SvANY(sv) + sv_type_details->offset),
6635 &PL_body_roots[type]);
6637 else if (sv_type_details->body_size) {
6638 safefree(SvANY(sv));
6642 /* caller is responsible for freeing the head of the original sv */
6643 if (sv != orig_sv && !SvREFCNT(sv))
6646 /* grab and free next sv, if any */
6654 else if (!iter_sv) {
6656 } else if (SvTYPE(iter_sv) == SVt_PVAV) {
6657 AV *const av = (AV*)iter_sv;
6658 if (AvFILLp(av) > -1) {
6659 sv = AvARRAY(av)[AvFILLp(av)--];
6661 else { /* no more elements of current AV to free */
6664 /* restore previous value, squirrelled away */
6665 iter_sv = AvARRAY(av)[AvMAX(av)];
6666 Safefree(AvALLOC(av));
6669 } else if (SvTYPE(iter_sv) == SVt_PVHV) {
6670 sv = Perl_hfree_next_entry(aTHX_ (HV*)iter_sv, &hash_index);
6671 if (!sv && !HvTOTALKEYS((HV *)iter_sv)) {
6672 /* no more elements of current HV to free */
6675 /* Restore previous values of iter_sv and hash_index,
6676 * squirrelled away */
6677 assert(!SvOBJECT(sv));
6678 iter_sv = (SV*)SvSTASH(sv);
6679 assert(!SvMAGICAL(sv));
6680 hash_index = ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index;
6682 /* perl -DA does not like rubbish in SvMAGIC. */
6686 /* free any remaining detritus from the hash struct */
6687 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6688 assert(!HvARRAY((HV*)sv));
6693 /* unrolled SvREFCNT_dec and sv_free2 follows: */
6697 if (!SvREFCNT(sv)) {
6701 if (--(SvREFCNT(sv)))
6705 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6706 "Attempt to free temp prematurely: SV 0x%"UVxf
6707 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6711 if (SvIMMORTAL(sv)) {
6712 /* make sure SvREFCNT(sv)==0 happens very seldom */
6713 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6722 /* This routine curses the sv itself, not the object referenced by sv. So
6723 sv does not have to be ROK. */
6726 S_curse(pTHX_ SV * const sv, const bool check_refcnt) {
6727 PERL_ARGS_ASSERT_CURSE;
6728 assert(SvOBJECT(sv));
6730 if (PL_defstash && /* Still have a symbol table? */
6736 stash = SvSTASH(sv);
6737 assert(SvTYPE(stash) == SVt_PVHV);
6738 if (HvNAME(stash)) {
6739 CV* destructor = NULL;
6740 assert (SvOOK(stash));
6741 if (!SvOBJECT(stash)) destructor = (CV *)SvSTASH(stash);
6742 if (!destructor || HvMROMETA(stash)->destroy_gen
6743 != PL_sub_generation)
6746 gv_fetchmeth_autoload(stash, "DESTROY", 7, 0);
6747 if (gv) destructor = GvCV(gv);
6748 if (!SvOBJECT(stash))
6751 destructor ? (HV *)destructor : ((HV *)0)+1;
6752 HvAUX(stash)->xhv_mro_meta->destroy_gen =
6756 assert(!destructor || destructor == ((CV *)0)+1
6757 || SvTYPE(destructor) == SVt_PVCV);
6758 if (destructor && destructor != ((CV *)0)+1
6759 /* A constant subroutine can have no side effects, so
6760 don't bother calling it. */
6761 && !CvCONST(destructor)
6762 /* Don't bother calling an empty destructor or one that
6763 returns immediately. */
6764 && (CvISXSUB(destructor)
6765 || (CvSTART(destructor)
6766 && (CvSTART(destructor)->op_next->op_type
6768 && (CvSTART(destructor)->op_next->op_type
6770 || CvSTART(destructor)->op_next->op_next->op_type
6776 SV* const tmpref = newRV(sv);
6777 SvREADONLY_on(tmpref); /* DESTROY() could be naughty */
6779 PUSHSTACKi(PERLSI_DESTROY);
6784 call_sv(MUTABLE_SV(destructor),
6785 G_DISCARD|G_EVAL|G_KEEPERR|G_VOID);
6789 if(SvREFCNT(tmpref) < 2) {
6790 /* tmpref is not kept alive! */
6792 SvRV_set(tmpref, NULL);
6795 SvREFCNT_dec_NN(tmpref);
6798 } while (SvOBJECT(sv) && SvSTASH(sv) != stash);
6801 if (check_refcnt && SvREFCNT(sv)) {
6802 if (PL_in_clean_objs)
6804 "DESTROY created new reference to dead object '%"HEKf"'",
6805 HEKfARG(HvNAME_HEK(stash)));
6806 /* DESTROY gave object new lease on life */
6812 HV * const stash = SvSTASH(sv);
6813 /* Curse before freeing the stash, as freeing the stash could cause
6814 a recursive call into S_curse. */
6815 SvOBJECT_off(sv); /* Curse the object. */
6816 SvSTASH_set(sv,0); /* SvREFCNT_dec may try to read this */
6817 SvREFCNT_dec(stash); /* possibly of changed persuasion */
6823 =for apidoc sv_newref
6825 Increment an SV's reference count. Use the C<SvREFCNT_inc()> wrapper
6832 Perl_sv_newref(pTHX_ SV *const sv)
6834 PERL_UNUSED_CONTEXT;
6843 Decrement an SV's reference count, and if it drops to zero, call
6844 C<sv_clear> to invoke destructors and free up any memory used by
6845 the body; finally, deallocating the SV's head itself.
6846 Normally called via a wrapper macro C<SvREFCNT_dec>.
6852 Perl_sv_free(pTHX_ SV *const sv)
6858 /* Private helper function for SvREFCNT_dec().
6859 * Called with rc set to original SvREFCNT(sv), where rc == 0 or 1 */
6862 Perl_sv_free2(pTHX_ SV *const sv, const U32 rc)
6866 PERL_ARGS_ASSERT_SV_FREE2;
6868 if (LIKELY( rc == 1 )) {
6874 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6875 "Attempt to free temp prematurely: SV 0x%"UVxf
6876 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6880 if (SvIMMORTAL(sv)) {
6881 /* make sure SvREFCNT(sv)==0 happens very seldom */
6882 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6886 if (! SvREFCNT(sv)) /* may have have been resurrected */
6891 /* handle exceptional cases */
6895 if (SvFLAGS(sv) & SVf_BREAK)
6896 /* this SV's refcnt has been artificially decremented to
6897 * trigger cleanup */
6899 if (PL_in_clean_all) /* All is fair */
6901 if (SvIMMORTAL(sv)) {
6902 /* make sure SvREFCNT(sv)==0 happens very seldom */
6903 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6906 if (ckWARN_d(WARN_INTERNAL)) {
6907 #ifdef DEBUG_LEAKING_SCALARS_FORK_DUMP
6908 Perl_dump_sv_child(aTHX_ sv);
6910 #ifdef DEBUG_LEAKING_SCALARS
6913 #ifdef DEBUG_LEAKING_SCALARS_ABORT
6914 if (PL_warnhook == PERL_WARNHOOK_FATAL
6915 || ckDEAD(packWARN(WARN_INTERNAL))) {
6916 /* Don't let Perl_warner cause us to escape our fate: */
6920 /* This may not return: */
6921 Perl_warner(aTHX_ packWARN(WARN_INTERNAL),
6922 "Attempt to free unreferenced scalar: SV 0x%"UVxf
6923 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6926 #ifdef DEBUG_LEAKING_SCALARS_ABORT
6936 Returns the length of the string in the SV. Handles magic and type
6937 coercion and sets the UTF8 flag appropriately. See also C<L</SvCUR>>, which
6938 gives raw access to the C<xpv_cur> slot.
6944 Perl_sv_len(pTHX_ SV *const sv)
6951 (void)SvPV_const(sv, len);
6956 =for apidoc sv_len_utf8
6958 Returns the number of characters in the string in an SV, counting wide
6959 UTF-8 bytes as a single character. Handles magic and type coercion.
6965 * The length is cached in PERL_MAGIC_utf8, in the mg_len field. Also the
6966 * mg_ptr is used, by sv_pos_u2b() and sv_pos_b2u() - see the comments below.
6967 * (Note that the mg_len is not the length of the mg_ptr field.
6968 * This allows the cache to store the character length of the string without
6969 * needing to malloc() extra storage to attach to the mg_ptr.)
6974 Perl_sv_len_utf8(pTHX_ SV *const sv)
6980 return sv_len_utf8_nomg(sv);
6984 Perl_sv_len_utf8_nomg(pTHX_ SV * const sv)
6987 const U8 *s = (U8*)SvPV_nomg_const(sv, len);
6989 PERL_ARGS_ASSERT_SV_LEN_UTF8_NOMG;
6991 if (PL_utf8cache && SvUTF8(sv)) {
6993 MAGIC *mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_utf8) : NULL;
6995 if (mg && (mg->mg_len != -1 || mg->mg_ptr)) {
6996 if (mg->mg_len != -1)
6999 /* We can use the offset cache for a headstart.
7000 The longer value is stored in the first pair. */
7001 STRLEN *cache = (STRLEN *) mg->mg_ptr;
7003 ulen = cache[0] + Perl_utf8_length(aTHX_ s + cache[1],
7007 if (PL_utf8cache < 0) {
7008 const STRLEN real = Perl_utf8_length(aTHX_ s, s + len);
7009 assert_uft8_cache_coherent("sv_len_utf8", ulen, real, sv);
7013 ulen = Perl_utf8_length(aTHX_ s, s + len);
7014 utf8_mg_len_cache_update(sv, &mg, ulen);
7018 return SvUTF8(sv) ? Perl_utf8_length(aTHX_ s, s + len) : len;
7021 /* Walk forwards to find the byte corresponding to the passed in UTF-8
7024 S_sv_pos_u2b_forwards(const U8 *const start, const U8 *const send,
7025 STRLEN *const uoffset_p, bool *const at_end)
7027 const U8 *s = start;
7028 STRLEN uoffset = *uoffset_p;
7030 PERL_ARGS_ASSERT_SV_POS_U2B_FORWARDS;
7032 while (s < send && uoffset) {
7039 else if (s > send) {
7041 /* This is the existing behaviour. Possibly it should be a croak, as
7042 it's actually a bounds error */
7045 *uoffset_p -= uoffset;
7049 /* Given the length of the string in both bytes and UTF-8 characters, decide
7050 whether to walk forwards or backwards to find the byte corresponding to
7051 the passed in UTF-8 offset. */
7053 S_sv_pos_u2b_midway(const U8 *const start, const U8 *send,
7054 STRLEN uoffset, const STRLEN uend)
7056 STRLEN backw = uend - uoffset;
7058 PERL_ARGS_ASSERT_SV_POS_U2B_MIDWAY;
7060 if (uoffset < 2 * backw) {
7061 /* The assumption is that going forwards is twice the speed of going
7062 forward (that's where the 2 * backw comes from).
7063 (The real figure of course depends on the UTF-8 data.) */
7064 const U8 *s = start;
7066 while (s < send && uoffset--)
7076 while (UTF8_IS_CONTINUATION(*send))
7079 return send - start;
7082 /* For the string representation of the given scalar, find the byte
7083 corresponding to the passed in UTF-8 offset. uoffset0 and boffset0
7084 give another position in the string, *before* the sought offset, which
7085 (which is always true, as 0, 0 is a valid pair of positions), which should
7086 help reduce the amount of linear searching.
7087 If *mgp is non-NULL, it should point to the UTF-8 cache magic, which
7088 will be used to reduce the amount of linear searching. The cache will be
7089 created if necessary, and the found value offered to it for update. */
7091 S_sv_pos_u2b_cached(pTHX_ SV *const sv, MAGIC **const mgp, const U8 *const start,
7092 const U8 *const send, STRLEN uoffset,
7093 STRLEN uoffset0, STRLEN boffset0)
7095 STRLEN boffset = 0; /* Actually always set, but let's keep gcc happy. */
7097 bool at_end = FALSE;
7099 PERL_ARGS_ASSERT_SV_POS_U2B_CACHED;
7101 assert (uoffset >= uoffset0);
7106 if (!SvREADONLY(sv) && !SvGMAGICAL(sv) && SvPOK(sv)
7108 && (*mgp || (SvTYPE(sv) >= SVt_PVMG &&
7109 (*mgp = mg_find(sv, PERL_MAGIC_utf8))))) {
7110 if ((*mgp)->mg_ptr) {
7111 STRLEN *cache = (STRLEN *) (*mgp)->mg_ptr;
7112 if (cache[0] == uoffset) {
7113 /* An exact match. */
7116 if (cache[2] == uoffset) {
7117 /* An exact match. */
7121 if (cache[0] < uoffset) {
7122 /* The cache already knows part of the way. */
7123 if (cache[0] > uoffset0) {
7124 /* The cache knows more than the passed in pair */
7125 uoffset0 = cache[0];
7126 boffset0 = cache[1];
7128 if ((*mgp)->mg_len != -1) {
7129 /* And we know the end too. */
7131 + sv_pos_u2b_midway(start + boffset0, send,
7133 (*mgp)->mg_len - uoffset0);
7135 uoffset -= uoffset0;
7137 + sv_pos_u2b_forwards(start + boffset0,
7138 send, &uoffset, &at_end);
7139 uoffset += uoffset0;
7142 else if (cache[2] < uoffset) {
7143 /* We're between the two cache entries. */
7144 if (cache[2] > uoffset0) {
7145 /* and the cache knows more than the passed in pair */
7146 uoffset0 = cache[2];
7147 boffset0 = cache[3];
7151 + sv_pos_u2b_midway(start + boffset0,
7154 cache[0] - uoffset0);
7157 + sv_pos_u2b_midway(start + boffset0,
7160 cache[2] - uoffset0);
7164 else if ((*mgp)->mg_len != -1) {
7165 /* If we can take advantage of a passed in offset, do so. */
7166 /* In fact, offset0 is either 0, or less than offset, so don't
7167 need to worry about the other possibility. */
7169 + sv_pos_u2b_midway(start + boffset0, send,
7171 (*mgp)->mg_len - uoffset0);
7176 if (!found || PL_utf8cache < 0) {
7177 STRLEN real_boffset;
7178 uoffset -= uoffset0;
7179 real_boffset = boffset0 + sv_pos_u2b_forwards(start + boffset0,
7180 send, &uoffset, &at_end);
7181 uoffset += uoffset0;
7183 if (found && PL_utf8cache < 0)
7184 assert_uft8_cache_coherent("sv_pos_u2b_cache", boffset,
7186 boffset = real_boffset;
7189 if (PL_utf8cache && !SvGMAGICAL(sv) && SvPOK(sv)) {
7191 utf8_mg_len_cache_update(sv, mgp, uoffset);
7193 utf8_mg_pos_cache_update(sv, mgp, boffset, uoffset, send - start);
7200 =for apidoc sv_pos_u2b_flags
7202 Converts the offset from a count of UTF-8 chars from
7203 the start of the string, to a count of the equivalent number of bytes; if
7204 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7205 C<offset>, rather than from the start
7206 of the string. Handles type coercion.
7207 C<flags> is passed to C<SvPV_flags>, and usually should be
7208 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7214 * sv_pos_u2b_flags() uses, like sv_pos_b2u(), the mg_ptr of the potential
7215 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7216 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7221 Perl_sv_pos_u2b_flags(pTHX_ SV *const sv, STRLEN uoffset, STRLEN *const lenp,
7228 PERL_ARGS_ASSERT_SV_POS_U2B_FLAGS;
7230 start = (U8*)SvPV_flags(sv, len, flags);
7232 const U8 * const send = start + len;
7234 boffset = sv_pos_u2b_cached(sv, &mg, start, send, uoffset, 0, 0);
7237 && *lenp /* don't bother doing work for 0, as its bytes equivalent
7238 is 0, and *lenp is already set to that. */) {
7239 /* Convert the relative offset to absolute. */
7240 const STRLEN uoffset2 = uoffset + *lenp;
7241 const STRLEN boffset2
7242 = sv_pos_u2b_cached(sv, &mg, start, send, uoffset2,
7243 uoffset, boffset) - boffset;
7257 =for apidoc sv_pos_u2b
7259 Converts the value pointed to by C<offsetp> from a count of UTF-8 chars from
7260 the start of the string, to a count of the equivalent number of bytes; if
7261 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7262 the offset, rather than from the start of the string. Handles magic and
7265 Use C<sv_pos_u2b_flags> in preference, which correctly handles strings longer
7272 * sv_pos_u2b() uses, like sv_pos_b2u(), the mg_ptr of the potential
7273 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7274 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7278 /* This function is subject to size and sign problems */
7281 Perl_sv_pos_u2b(pTHX_ SV *const sv, I32 *const offsetp, I32 *const lenp)
7283 PERL_ARGS_ASSERT_SV_POS_U2B;
7286 STRLEN ulen = (STRLEN)*lenp;
7287 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, &ulen,
7288 SV_GMAGIC|SV_CONST_RETURN);
7291 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, NULL,
7292 SV_GMAGIC|SV_CONST_RETURN);
7297 S_utf8_mg_len_cache_update(pTHX_ SV *const sv, MAGIC **const mgp,
7300 PERL_ARGS_ASSERT_UTF8_MG_LEN_CACHE_UPDATE;
7301 if (SvREADONLY(sv) || SvGMAGICAL(sv) || !SvPOK(sv))
7304 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7305 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7306 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, &PL_vtbl_utf8, 0, 0);
7310 (*mgp)->mg_len = ulen;
7313 /* Create and update the UTF8 magic offset cache, with the proffered utf8/
7314 byte length pairing. The (byte) length of the total SV is passed in too,
7315 as blen, because for some (more esoteric) SVs, the call to SvPV_const()
7316 may not have updated SvCUR, so we can't rely on reading it directly.
7318 The proffered utf8/byte length pairing isn't used if the cache already has
7319 two pairs, and swapping either for the proffered pair would increase the
7320 RMS of the intervals between known byte offsets.
7322 The cache itself consists of 4 STRLEN values
7323 0: larger UTF-8 offset
7324 1: corresponding byte offset
7325 2: smaller UTF-8 offset
7326 3: corresponding byte offset
7328 Unused cache pairs have the value 0, 0.
7329 Keeping the cache "backwards" means that the invariant of
7330 cache[0] >= cache[2] is maintained even with empty slots, which means that
7331 the code that uses it doesn't need to worry if only 1 entry has actually
7332 been set to non-zero. It also makes the "position beyond the end of the
7333 cache" logic much simpler, as the first slot is always the one to start
7337 S_utf8_mg_pos_cache_update(pTHX_ SV *const sv, MAGIC **const mgp, const STRLEN byte,
7338 const STRLEN utf8, const STRLEN blen)
7342 PERL_ARGS_ASSERT_UTF8_MG_POS_CACHE_UPDATE;
7347 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7348 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7349 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, (MGVTBL*)&PL_vtbl_utf8, 0,
7351 (*mgp)->mg_len = -1;
7355 if (!(cache = (STRLEN *)(*mgp)->mg_ptr)) {
7356 Newxz(cache, PERL_MAGIC_UTF8_CACHESIZE * 2, STRLEN);
7357 (*mgp)->mg_ptr = (char *) cache;
7361 if (PL_utf8cache < 0 && SvPOKp(sv)) {
7362 /* SvPOKp() because, if sv is a reference, then SvPVX() is actually
7363 a pointer. Note that we no longer cache utf8 offsets on refer-
7364 ences, but this check is still a good idea, for robustness. */
7365 const U8 *start = (const U8 *) SvPVX_const(sv);
7366 const STRLEN realutf8 = utf8_length(start, start + byte);
7368 assert_uft8_cache_coherent("utf8_mg_pos_cache_update", utf8, realutf8,
7372 /* Cache is held with the later position first, to simplify the code
7373 that deals with unbounded ends. */
7375 ASSERT_UTF8_CACHE(cache);
7376 if (cache[1] == 0) {
7377 /* Cache is totally empty */
7380 } else if (cache[3] == 0) {
7381 if (byte > cache[1]) {
7382 /* New one is larger, so goes first. */
7383 cache[2] = cache[0];
7384 cache[3] = cache[1];
7392 /* float casts necessary? XXX */
7393 #define THREEWAY_SQUARE(a,b,c,d) \
7394 ((float)((d) - (c))) * ((float)((d) - (c))) \
7395 + ((float)((c) - (b))) * ((float)((c) - (b))) \
7396 + ((float)((b) - (a))) * ((float)((b) - (a)))
7398 /* Cache has 2 slots in use, and we know three potential pairs.
7399 Keep the two that give the lowest RMS distance. Do the
7400 calculation in bytes simply because we always know the byte
7401 length. squareroot has the same ordering as the positive value,
7402 so don't bother with the actual square root. */
7403 if (byte > cache[1]) {
7404 /* New position is after the existing pair of pairs. */
7405 const float keep_earlier
7406 = THREEWAY_SQUARE(0, cache[3], byte, blen);
7407 const float keep_later
7408 = THREEWAY_SQUARE(0, cache[1], byte, blen);
7410 if (keep_later < keep_earlier) {
7411 cache[2] = cache[0];
7412 cache[3] = cache[1];
7418 const float keep_later = THREEWAY_SQUARE(0, byte, cache[1], blen);
7419 float b, c, keep_earlier;
7420 if (byte > cache[3]) {
7421 /* New position is between the existing pair of pairs. */
7422 b = (float)cache[3];
7425 /* New position is before the existing pair of pairs. */
7427 c = (float)cache[3];
7429 keep_earlier = THREEWAY_SQUARE(0, b, c, blen);
7430 if (byte > cache[3]) {
7431 if (keep_later < keep_earlier) {
7441 if (! (keep_later < keep_earlier)) {
7442 cache[0] = cache[2];
7443 cache[1] = cache[3];
7450 ASSERT_UTF8_CACHE(cache);
7453 /* We already know all of the way, now we may be able to walk back. The same
7454 assumption is made as in S_sv_pos_u2b_midway(), namely that walking
7455 backward is half the speed of walking forward. */
7457 S_sv_pos_b2u_midway(pTHX_ const U8 *const s, const U8 *const target,
7458 const U8 *end, STRLEN endu)
7460 const STRLEN forw = target - s;
7461 STRLEN backw = end - target;
7463 PERL_ARGS_ASSERT_SV_POS_B2U_MIDWAY;
7465 if (forw < 2 * backw) {
7466 return utf8_length(s, target);
7469 while (end > target) {
7471 while (UTF8_IS_CONTINUATION(*end)) {
7480 =for apidoc sv_pos_b2u_flags
7482 Converts C<offset> from a count of bytes from the start of the string, to
7483 a count of the equivalent number of UTF-8 chars. Handles type coercion.
7484 C<flags> is passed to C<SvPV_flags>, and usually should be
7485 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7491 * sv_pos_b2u_flags() uses, like sv_pos_u2b_flags(), the mg_ptr of the
7492 * potential PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8
7497 Perl_sv_pos_b2u_flags(pTHX_ SV *const sv, STRLEN const offset, U32 flags)
7500 STRLEN len = 0; /* Actually always set, but let's keep gcc happy. */
7506 PERL_ARGS_ASSERT_SV_POS_B2U_FLAGS;
7508 s = (const U8*)SvPV_flags(sv, blen, flags);
7511 Perl_croak(aTHX_ "panic: sv_pos_b2u: bad byte offset, blen=%"UVuf
7512 ", byte=%"UVuf, (UV)blen, (UV)offset);
7518 && SvTYPE(sv) >= SVt_PVMG
7519 && (mg = mg_find(sv, PERL_MAGIC_utf8)))
7522 STRLEN * const cache = (STRLEN *) mg->mg_ptr;
7523 if (cache[1] == offset) {
7524 /* An exact match. */
7527 if (cache[3] == offset) {
7528 /* An exact match. */
7532 if (cache[1] < offset) {
7533 /* We already know part of the way. */
7534 if (mg->mg_len != -1) {
7535 /* Actually, we know the end too. */
7537 + S_sv_pos_b2u_midway(aTHX_ s + cache[1], send,
7538 s + blen, mg->mg_len - cache[0]);
7540 len = cache[0] + utf8_length(s + cache[1], send);
7543 else if (cache[3] < offset) {
7544 /* We're between the two cached pairs, so we do the calculation
7545 offset by the byte/utf-8 positions for the earlier pair,
7546 then add the utf-8 characters from the string start to
7548 len = S_sv_pos_b2u_midway(aTHX_ s + cache[3], send,
7549 s + cache[1], cache[0] - cache[2])
7553 else { /* cache[3] > offset */
7554 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + cache[3],
7558 ASSERT_UTF8_CACHE(cache);
7560 } else if (mg->mg_len != -1) {
7561 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + blen, mg->mg_len);
7565 if (!found || PL_utf8cache < 0) {
7566 const STRLEN real_len = utf8_length(s, send);
7568 if (found && PL_utf8cache < 0)
7569 assert_uft8_cache_coherent("sv_pos_b2u", len, real_len, sv);
7575 utf8_mg_len_cache_update(sv, &mg, len);
7577 utf8_mg_pos_cache_update(sv, &mg, offset, len, blen);
7584 =for apidoc sv_pos_b2u
7586 Converts the value pointed to by C<offsetp> from a count of bytes from the
7587 start of the string, to a count of the equivalent number of UTF-8 chars.
7588 Handles magic and type coercion.
7590 Use C<sv_pos_b2u_flags> in preference, which correctly handles strings
7597 * sv_pos_b2u() uses, like sv_pos_u2b(), the mg_ptr of the potential
7598 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7603 Perl_sv_pos_b2u(pTHX_ SV *const sv, I32 *const offsetp)
7605 PERL_ARGS_ASSERT_SV_POS_B2U;
7610 *offsetp = (I32)sv_pos_b2u_flags(sv, (STRLEN)*offsetp,
7611 SV_GMAGIC|SV_CONST_RETURN);
7615 S_assert_uft8_cache_coherent(pTHX_ const char *const func, STRLEN from_cache,
7616 STRLEN real, SV *const sv)
7618 PERL_ARGS_ASSERT_ASSERT_UFT8_CACHE_COHERENT;
7620 /* As this is debugging only code, save space by keeping this test here,
7621 rather than inlining it in all the callers. */
7622 if (from_cache == real)
7625 /* Need to turn the assertions off otherwise we may recurse infinitely
7626 while printing error messages. */
7627 SAVEI8(PL_utf8cache);
7629 Perl_croak(aTHX_ "panic: %s cache %"UVuf" real %"UVuf" for %"SVf,
7630 func, (UV) from_cache, (UV) real, SVfARG(sv));
7636 Returns a boolean indicating whether the strings in the two SVs are
7637 identical. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7638 coerce its args to strings if necessary.
7640 =for apidoc sv_eq_flags
7642 Returns a boolean indicating whether the strings in the two SVs are
7643 identical. Is UTF-8 and S<C<'use bytes'>> aware and coerces its args to strings
7644 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get-magic, too.
7650 Perl_sv_eq_flags(pTHX_ SV *sv1, SV *sv2, const U32 flags)
7657 SV* svrecode = NULL;
7664 /* if pv1 and pv2 are the same, second SvPV_const call may
7665 * invalidate pv1 (if we are handling magic), so we may need to
7667 if (sv1 == sv2 && flags & SV_GMAGIC
7668 && (SvTHINKFIRST(sv1) || SvGMAGICAL(sv1))) {
7669 pv1 = SvPV_const(sv1, cur1);
7670 sv1 = newSVpvn_flags(pv1, cur1, SVs_TEMP | SvUTF8(sv2));
7672 pv1 = SvPV_flags_const(sv1, cur1, flags);
7680 pv2 = SvPV_flags_const(sv2, cur2, flags);
7682 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7683 /* Differing utf8ness.
7684 * Do not UTF8size the comparands as a side-effect. */
7687 svrecode = newSVpvn(pv2, cur2);
7688 sv_recode_to_utf8(svrecode, _get_encoding());
7689 pv2 = SvPV_const(svrecode, cur2);
7692 svrecode = newSVpvn(pv1, cur1);
7693 sv_recode_to_utf8(svrecode, _get_encoding());
7694 pv1 = SvPV_const(svrecode, cur1);
7696 /* Now both are in UTF-8. */
7698 SvREFCNT_dec_NN(svrecode);
7704 /* sv1 is the UTF-8 one */
7705 return bytes_cmp_utf8((const U8*)pv2, cur2,
7706 (const U8*)pv1, cur1) == 0;
7709 /* sv2 is the UTF-8 one */
7710 return bytes_cmp_utf8((const U8*)pv1, cur1,
7711 (const U8*)pv2, cur2) == 0;
7717 eq = (pv1 == pv2) || memEQ(pv1, pv2, cur1);
7719 SvREFCNT_dec(svrecode);
7727 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7728 string in C<sv1> is less than, equal to, or greater than the string in
7729 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7730 coerce its args to strings if necessary. See also C<L</sv_cmp_locale>>.
7732 =for apidoc sv_cmp_flags
7734 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7735 string in C<sv1> is less than, equal to, or greater than the string in
7736 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware and will coerce its args to strings
7737 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get magic. See
7738 also C<L</sv_cmp_locale_flags>>.
7744 Perl_sv_cmp(pTHX_ SV *const sv1, SV *const sv2)
7746 return sv_cmp_flags(sv1, sv2, SV_GMAGIC);
7750 Perl_sv_cmp_flags(pTHX_ SV *const sv1, SV *const sv2,
7754 const char *pv1, *pv2;
7756 SV *svrecode = NULL;
7763 pv1 = SvPV_flags_const(sv1, cur1, flags);
7770 pv2 = SvPV_flags_const(sv2, cur2, flags);
7772 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7773 /* Differing utf8ness.
7774 * Do not UTF8size the comparands as a side-effect. */
7777 svrecode = newSVpvn(pv2, cur2);
7778 sv_recode_to_utf8(svrecode, _get_encoding());
7779 pv2 = SvPV_const(svrecode, cur2);
7782 const int retval = -bytes_cmp_utf8((const U8*)pv2, cur2,
7783 (const U8*)pv1, cur1);
7784 return retval ? retval < 0 ? -1 : +1 : 0;
7789 svrecode = newSVpvn(pv1, cur1);
7790 sv_recode_to_utf8(svrecode, _get_encoding());
7791 pv1 = SvPV_const(svrecode, cur1);
7794 const int retval = bytes_cmp_utf8((const U8*)pv1, cur1,
7795 (const U8*)pv2, cur2);
7796 return retval ? retval < 0 ? -1 : +1 : 0;
7801 /* Here, if both are non-NULL, then they have the same UTF8ness. */
7804 cmp = cur2 ? -1 : 0;
7808 STRLEN shortest_len = cur1 < cur2 ? cur1 : cur2;
7811 if (! DO_UTF8(sv1)) {
7813 const I32 retval = memcmp((const void*)pv1,
7817 cmp = retval < 0 ? -1 : 1;
7818 } else if (cur1 == cur2) {
7821 cmp = cur1 < cur2 ? -1 : 1;
7825 else { /* Both are to be treated as UTF-EBCDIC */
7827 /* EBCDIC UTF-8 is complicated by the fact that it is based on I8
7828 * which remaps code points 0-255. We therefore generally have to
7829 * unmap back to the original values to get an accurate comparison.
7830 * But we don't have to do that for UTF-8 invariants, as by
7831 * definition, they aren't remapped, nor do we have to do it for
7832 * above-latin1 code points, as they also aren't remapped. (This
7833 * code also works on ASCII platforms, but the memcmp() above is
7836 const char *e = pv1 + shortest_len;
7838 /* Find the first bytes that differ between the two strings */
7839 while (pv1 < e && *pv1 == *pv2) {
7845 if (pv1 == e) { /* Are the same all the way to the end */
7849 cmp = cur1 < cur2 ? -1 : 1;
7852 else /* Here *pv1 and *pv2 are not equal, but all bytes earlier
7853 * in the strings were. The current bytes may or may not be
7854 * at the beginning of a character. But neither or both are
7855 * (or else earlier bytes would have been different). And
7856 * if we are in the middle of a character, the two
7857 * characters are comprised of the same number of bytes
7858 * (because in this case the start bytes are the same, and
7859 * the start bytes encode the character's length). */
7860 if (UTF8_IS_INVARIANT(*pv1))
7862 /* If both are invariants; can just compare directly */
7863 if (UTF8_IS_INVARIANT(*pv2)) {
7864 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
7866 else /* Since *pv1 is invariant, it is the whole character,
7867 which means it is at the beginning of a character.
7868 That means pv2 is also at the beginning of a
7869 character (see earlier comment). Since it isn't
7870 invariant, it must be a start byte. If it starts a
7871 character whose code point is above 255, that
7872 character is greater than any single-byte char, which
7874 if (UTF8_IS_ABOVE_LATIN1_START(*pv2))
7879 /* Here, pv2 points to a character composed of 2 bytes
7880 * whose code point is < 256. Get its code point and
7881 * compare with *pv1 */
7882 cmp = ((U8) *pv1 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
7887 else /* The code point starting at pv1 isn't a single byte */
7888 if (UTF8_IS_INVARIANT(*pv2))
7890 /* But here, the code point starting at *pv2 is a single byte,
7891 * and so *pv1 must begin a character, hence is a start byte.
7892 * If that character is above 255, it is larger than any
7893 * single-byte char, which *pv2 is */
7894 if (UTF8_IS_ABOVE_LATIN1_START(*pv1)) {
7898 /* Here, pv1 points to a character composed of 2 bytes
7899 * whose code point is < 256. Get its code point and
7900 * compare with the single byte character *pv2 */
7901 cmp = (EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1)) < (U8) *pv2)
7906 else /* Here, we've ruled out either *pv1 and *pv2 being
7907 invariant. That means both are part of variants, but not
7908 necessarily at the start of a character */
7909 if ( UTF8_IS_ABOVE_LATIN1_START(*pv1)
7910 || UTF8_IS_ABOVE_LATIN1_START(*pv2))
7912 /* Here, at least one is the start of a character, which means
7913 * the other is also a start byte. And the code point of at
7914 * least one of the characters is above 255. It is a
7915 * characteristic of UTF-EBCDIC that all start bytes for
7916 * above-latin1 code points are well behaved as far as code
7917 * point comparisons go, and all are larger than all other
7918 * start bytes, so the comparison with those is also well
7920 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
7923 /* Here both *pv1 and *pv2 are part of variant characters.
7924 * They could be both continuations, or both start characters.
7925 * (One or both could even be an illegal start character (for
7926 * an overlong) which for the purposes of sorting we treat as
7928 if (UTF8_IS_CONTINUATION(*pv1)) {
7930 /* If they are continuations for code points above 255,
7931 * then comparing the current byte is sufficient, as there
7932 * is no remapping of these and so the comparison is
7933 * well-behaved. We determine if they are such
7934 * continuations by looking at the preceding byte. It
7935 * could be a start byte, from which we can tell if it is
7936 * for an above 255 code point. Or it could be a
7937 * continuation, which means the character occupies at
7938 * least 3 bytes, so must be above 255. */
7939 if ( UTF8_IS_CONTINUATION(*(pv2 - 1))
7940 || UTF8_IS_ABOVE_LATIN1_START(*(pv2 -1)))
7942 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
7946 /* Here, the continuations are for code points below 256;
7947 * back up one to get to the start byte */
7952 /* We need to get the actual native code point of each of these
7953 * variants in order to compare them */
7954 cmp = ( EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1))
7955 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
7964 SvREFCNT_dec(svrecode);
7970 =for apidoc sv_cmp_locale
7972 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
7973 S<C<'use bytes'>> aware, handles get magic, and will coerce its args to strings
7974 if necessary. See also C<L</sv_cmp>>.
7976 =for apidoc sv_cmp_locale_flags
7978 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
7979 S<C<'use bytes'>> aware and will coerce its args to strings if necessary. If
7980 the flags contain C<SV_GMAGIC>, it handles get magic. See also
7981 C<L</sv_cmp_flags>>.
7987 Perl_sv_cmp_locale(pTHX_ SV *const sv1, SV *const sv2)
7989 return sv_cmp_locale_flags(sv1, sv2, SV_GMAGIC);
7993 Perl_sv_cmp_locale_flags(pTHX_ SV *const sv1, SV *const sv2,
7996 #ifdef USE_LOCALE_COLLATE
8002 if (PL_collation_standard)
8006 pv1 = sv1 ? sv_collxfrm_flags(sv1, &len1, flags) : (char *) NULL;
8008 pv2 = sv2 ? sv_collxfrm_flags(sv2, &len2, flags) : (char *) NULL;
8010 if (!pv1 || !len1) {
8021 retval = memcmp((void*)pv1, (void*)pv2, len1 < len2 ? len1 : len2);
8024 return retval < 0 ? -1 : 1;
8027 * When the result of collation is equality, that doesn't mean
8028 * that there are no differences -- some locales exclude some
8029 * characters from consideration. So to avoid false equalities,
8030 * we use the raw string as a tiebreaker.
8037 PERL_UNUSED_ARG(flags);
8038 #endif /* USE_LOCALE_COLLATE */
8040 return sv_cmp(sv1, sv2);
8044 #ifdef USE_LOCALE_COLLATE
8047 =for apidoc sv_collxfrm
8049 This calls C<sv_collxfrm_flags> with the SV_GMAGIC flag. See
8050 C<L</sv_collxfrm_flags>>.
8052 =for apidoc sv_collxfrm_flags
8054 Add Collate Transform magic to an SV if it doesn't already have it. If the
8055 flags contain C<SV_GMAGIC>, it handles get-magic.
8057 Any scalar variable may carry C<PERL_MAGIC_collxfrm> magic that contains the
8058 scalar data of the variable, but transformed to such a format that a normal
8059 memory comparison can be used to compare the data according to the locale
8066 Perl_sv_collxfrm_flags(pTHX_ SV *const sv, STRLEN *const nxp, const I32 flags)
8070 PERL_ARGS_ASSERT_SV_COLLXFRM_FLAGS;
8072 mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_collxfrm) : (MAGIC *) NULL;
8073 if (!mg || !mg->mg_ptr || *(U32*)mg->mg_ptr != PL_collation_ix) {
8079 Safefree(mg->mg_ptr);
8080 s = SvPV_flags_const(sv, len, flags);
8081 if ((xf = mem_collxfrm(s, len, &xlen))) {
8083 mg = sv_magicext(sv, 0, PERL_MAGIC_collxfrm, &PL_vtbl_collxfrm,
8097 if (mg && mg->mg_ptr) {
8099 return mg->mg_ptr + sizeof(PL_collation_ix);
8107 #endif /* USE_LOCALE_COLLATE */
8110 S_sv_gets_append_to_utf8(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8112 SV * const tsv = newSV(0);
8115 sv_gets(tsv, fp, 0);
8116 sv_utf8_upgrade_nomg(tsv);
8117 SvCUR_set(sv,append);
8120 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8124 S_sv_gets_read_record(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8127 const STRLEN recsize = SvUV(SvRV(PL_rs)); /* RsRECORD() guarantees > 0. */
8128 /* Grab the size of the record we're getting */
8129 char *buffer = SvGROW(sv, (STRLEN)(recsize + append + 1)) + append;
8136 /* With a true, record-oriented file on VMS, we need to use read directly
8137 * to ensure that we respect RMS record boundaries. The user is responsible
8138 * for providing a PL_rs value that corresponds to the FAB$W_MRS (maximum
8139 * record size) field. N.B. This is likely to produce invalid results on
8140 * varying-width character data when a record ends mid-character.
8142 fd = PerlIO_fileno(fp);
8144 && PerlLIO_fstat(fd, &st) == 0
8145 && (st.st_fab_rfm == FAB$C_VAR
8146 || st.st_fab_rfm == FAB$C_VFC
8147 || st.st_fab_rfm == FAB$C_FIX)) {
8149 bytesread = PerlLIO_read(fd, buffer, recsize);
8151 else /* in-memory file from PerlIO::Scalar
8152 * or not a record-oriented file
8156 bytesread = PerlIO_read(fp, buffer, recsize);
8158 /* At this point, the logic in sv_get() means that sv will
8159 be treated as utf-8 if the handle is utf8.
8161 if (PerlIO_isutf8(fp) && bytesread > 0) {
8162 char *bend = buffer + bytesread;
8163 char *bufp = buffer;
8164 size_t charcount = 0;
8165 bool charstart = TRUE;
8168 while (charcount < recsize) {
8169 /* count accumulated characters */
8170 while (bufp < bend) {
8172 skip = UTF8SKIP(bufp);
8174 if (bufp + skip > bend) {
8175 /* partial at the end */
8186 if (charcount < recsize) {
8188 STRLEN bufp_offset = bufp - buffer;
8189 SSize_t morebytesread;
8191 /* originally I read enough to fill any incomplete
8192 character and the first byte of the next
8193 character if needed, but if there's many
8194 multi-byte encoded characters we're going to be
8195 making a read call for every character beyond
8196 the original read size.
8198 So instead, read the rest of the character if
8199 any, and enough bytes to match at least the
8200 start bytes for each character we're going to
8204 readsize = recsize - charcount;
8206 readsize = skip - (bend - bufp) + recsize - charcount - 1;
8207 buffer = SvGROW(sv, append + bytesread + readsize + 1) + append;
8208 bend = buffer + bytesread;
8209 morebytesread = PerlIO_read(fp, bend, readsize);
8210 if (morebytesread <= 0) {
8211 /* we're done, if we still have incomplete
8212 characters the check code in sv_gets() will
8215 I'd originally considered doing
8216 PerlIO_ungetc() on all but the lead
8217 character of the incomplete character, but
8218 read() doesn't do that, so I don't.
8223 /* prepare to scan some more */
8224 bytesread += morebytesread;
8225 bend = buffer + bytesread;
8226 bufp = buffer + bufp_offset;
8234 SvCUR_set(sv, bytesread + append);
8235 buffer[bytesread] = '\0';
8236 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8242 Get a line from the filehandle and store it into the SV, optionally
8243 appending to the currently-stored string. If C<append> is not 0, the
8244 line is appended to the SV instead of overwriting it. C<append> should
8245 be set to the byte offset that the appended string should start at
8246 in the SV (typically, C<SvCUR(sv)> is a suitable choice).
8252 Perl_sv_gets(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8262 PERL_ARGS_ASSERT_SV_GETS;
8264 if (SvTHINKFIRST(sv))
8265 sv_force_normal_flags(sv, append ? 0 : SV_COW_DROP_PV);
8266 /* XXX. If you make this PVIV, then copy on write can copy scalars read
8268 However, perlbench says it's slower, because the existing swipe code
8269 is faster than copy on write.
8270 Swings and roundabouts. */
8271 SvUPGRADE(sv, SVt_PV);
8274 /* line is going to be appended to the existing buffer in the sv */
8275 if (PerlIO_isutf8(fp)) {
8277 sv_utf8_upgrade_nomg(sv);
8278 sv_pos_u2b(sv,&append,0);
8280 } else if (SvUTF8(sv)) {
8281 return S_sv_gets_append_to_utf8(aTHX_ sv, fp, append);
8287 /* not appending - "clear" the string by setting SvCUR to 0,
8288 * the pv is still avaiable. */
8291 if (PerlIO_isutf8(fp))
8294 if (IN_PERL_COMPILETIME) {
8295 /* we always read code in line mode */
8299 else if (RsSNARF(PL_rs)) {
8300 /* If it is a regular disk file use size from stat() as estimate
8301 of amount we are going to read -- may result in mallocing
8302 more memory than we really need if the layers below reduce
8303 the size we read (e.g. CRLF or a gzip layer).
8306 int fd = PerlIO_fileno(fp);
8307 if (fd >= 0 && (PerlLIO_fstat(fd, &st) == 0) && S_ISREG(st.st_mode)) {
8308 const Off_t offset = PerlIO_tell(fp);
8309 if (offset != (Off_t) -1 && st.st_size + append > offset) {
8310 #ifdef PERL_COPY_ON_WRITE
8311 /* Add an extra byte for the sake of copy-on-write's
8312 * buffer reference count. */
8313 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 2));
8315 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 1));
8322 else if (RsRECORD(PL_rs)) {
8323 return S_sv_gets_read_record(aTHX_ sv, fp, append);
8325 else if (RsPARA(PL_rs)) {
8331 /* Get $/ i.e. PL_rs into same encoding as stream wants */
8332 if (PerlIO_isutf8(fp)) {
8333 rsptr = SvPVutf8(PL_rs, rslen);
8336 if (SvUTF8(PL_rs)) {
8337 if (!sv_utf8_downgrade(PL_rs, TRUE)) {
8338 Perl_croak(aTHX_ "Wide character in $/");
8341 /* extract the raw pointer to the record separator */
8342 rsptr = SvPV_const(PL_rs, rslen);
8346 /* rslast is the last character in the record separator
8347 * note we don't use rslast except when rslen is true, so the
8348 * null assign is a placeholder. */
8349 rslast = rslen ? rsptr[rslen - 1] : '\0';
8351 if (rspara) { /* have to do this both before and after */
8352 do { /* to make sure file boundaries work right */
8355 i = PerlIO_getc(fp);
8359 PerlIO_ungetc(fp,i);
8365 /* See if we know enough about I/O mechanism to cheat it ! */
8367 /* This used to be #ifdef test - it is made run-time test for ease
8368 of abstracting out stdio interface. One call should be cheap
8369 enough here - and may even be a macro allowing compile
8373 if (PerlIO_fast_gets(fp)) {
8375 * We can do buffer based IO operations on this filehandle.
8377 * This means we can bypass a lot of subcalls and process
8378 * the buffer directly, it also means we know the upper bound
8379 * on the amount of data we might read of the current buffer
8380 * into our sv. Knowing this allows us to preallocate the pv
8381 * to be able to hold that maximum, which allows us to simplify
8382 * a lot of logic. */
8385 * We're going to steal some values from the stdio struct
8386 * and put EVERYTHING in the innermost loop into registers.
8388 STDCHAR *ptr; /* pointer into fp's read-ahead buffer */
8389 STRLEN bpx; /* length of the data in the target sv
8390 used to fix pointers after a SvGROW */
8391 I32 shortbuffered; /* If the pv buffer is shorter than the amount
8392 of data left in the read-ahead buffer.
8393 If 0 then the pv buffer can hold the full
8394 amount left, otherwise this is the amount it
8397 /* Here is some breathtakingly efficient cheating */
8399 /* When you read the following logic resist the urge to think
8400 * of record separators that are 1 byte long. They are an
8401 * uninteresting special (simple) case.
8403 * Instead think of record separators which are at least 2 bytes
8404 * long, and keep in mind that we need to deal with such
8405 * separators when they cross a read-ahead buffer boundary.
8407 * Also consider that we need to gracefully deal with separators
8408 * that may be longer than a single read ahead buffer.
8410 * Lastly do not forget we want to copy the delimiter as well. We
8411 * are copying all data in the file _up_to_and_including_ the separator
8414 * Now that you have all that in mind here is what is happening below:
8416 * 1. When we first enter the loop we do some memory book keeping to see
8417 * how much free space there is in the target SV. (This sub assumes that
8418 * it is operating on the same SV most of the time via $_ and that it is
8419 * going to be able to reuse the same pv buffer each call.) If there is
8420 * "enough" room then we set "shortbuffered" to how much space there is
8421 * and start reading forward.
8423 * 2. When we scan forward we copy from the read-ahead buffer to the target
8424 * SV's pv buffer. While we go we watch for the end of the read-ahead buffer,
8425 * and the end of the of pv, as well as for the "rslast", which is the last
8426 * char of the separator.
8428 * 3. When scanning forward if we see rslast then we jump backwards in *pv*
8429 * (which has a "complete" record up to the point we saw rslast) and check
8430 * it to see if it matches the separator. If it does we are done. If it doesn't
8431 * we continue on with the scan/copy.
8433 * 4. If we run out of read-ahead buffer (cnt goes to 0) then we have to get
8434 * the IO system to read the next buffer. We do this by doing a getc(), which
8435 * returns a single char read (or EOF), and prefills the buffer, and also
8436 * allows us to find out how full the buffer is. We use this information to
8437 * SvGROW() the sv to the size remaining in the buffer, after which we copy
8438 * the returned single char into the target sv, and then go back into scan
8441 * 5. If we run out of write-buffer then we SvGROW() it by the size of the
8442 * remaining space in the read-buffer.
8444 * Note that this code despite its twisty-turny nature is pretty darn slick.
8445 * It manages single byte separators, multi-byte cross boundary separators,
8446 * and cross-read-buffer separators cleanly and efficiently at the cost
8447 * of potentially greatly overallocating the target SV.
8453 /* get the number of bytes remaining in the read-ahead buffer
8454 * on first call on a given fp this will return 0.*/
8455 cnt = PerlIO_get_cnt(fp);
8457 /* make sure we have the room */
8458 if ((I32)(SvLEN(sv) - append) <= cnt + 1) {
8459 /* Not room for all of it
8460 if we are looking for a separator and room for some
8462 if (rslen && cnt > 80 && (I32)SvLEN(sv) > append) {
8463 /* just process what we have room for */
8464 shortbuffered = cnt - SvLEN(sv) + append + 1;
8465 cnt -= shortbuffered;
8468 /* ensure that the target sv has enough room to hold
8469 * the rest of the read-ahead buffer */
8471 /* remember that cnt can be negative */
8472 SvGROW(sv, (STRLEN)(append + (cnt <= 0 ? 2 : (cnt + 1))));
8476 /* we have enough room to hold the full buffer, lets scream */
8480 /* extract the pointer to sv's string buffer, offset by append as necessary */
8481 bp = (STDCHAR*)SvPVX_const(sv) + append; /* move these two too to registers */
8482 /* extract the point to the read-ahead buffer */
8483 ptr = (STDCHAR*)PerlIO_get_ptr(fp);
8485 /* some trace debug output */
8486 DEBUG_P(PerlIO_printf(Perl_debug_log,
8487 "Screamer: entering, ptr=%"UVuf", cnt=%ld\n",PTR2UV(ptr),(long)cnt));
8488 DEBUG_P(PerlIO_printf(Perl_debug_log,
8489 "Screamer: entering: PerlIO * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"
8491 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8492 PTR2UV(PerlIO_has_base(fp) ? PerlIO_get_base(fp) : 0)));
8496 /* if there is stuff left in the read-ahead buffer */
8498 /* if there is a separator */
8500 /* loop until we hit the end of the read-ahead buffer */
8501 while (cnt > 0) { /* this | eat */
8502 /* scan forward copying and searching for rslast as we go */
8504 if ((*bp++ = *ptr++) == rslast) /* really | dust */
8505 goto thats_all_folks; /* screams | sed :-) */
8509 /* no separator, slurp the full buffer */
8510 Copy(ptr, bp, cnt, char); /* this | eat */
8511 bp += cnt; /* screams | dust */
8512 ptr += cnt; /* louder | sed :-) */
8514 assert (!shortbuffered);
8515 goto cannot_be_shortbuffered;
8519 if (shortbuffered) { /* oh well, must extend */
8520 /* we didnt have enough room to fit the line into the target buffer
8521 * so we must extend the target buffer and keep going */
8522 cnt = shortbuffered;
8524 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8526 /* extned the target sv's buffer so it can hold the full read-ahead buffer */
8527 SvGROW(sv, SvLEN(sv) + append + cnt + 2);
8528 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8532 cannot_be_shortbuffered:
8533 /* we need to refill the read-ahead buffer if possible */
8535 DEBUG_P(PerlIO_printf(Perl_debug_log,
8536 "Screamer: going to getc, ptr=%"UVuf", cnt=%"IVdf"\n",
8537 PTR2UV(ptr),(IV)cnt));
8538 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* deregisterize cnt and ptr */
8540 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8541 "Screamer: pre: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf"\n",
8542 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8543 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8546 call PerlIO_getc() to let it prefill the lookahead buffer
8548 This used to call 'filbuf' in stdio form, but as that behaves like
8549 getc when cnt <= 0 we use PerlIO_getc here to avoid introducing
8550 another abstraction.
8552 Note we have to deal with the char in 'i' if we are not at EOF
8554 i = PerlIO_getc(fp); /* get more characters */
8556 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8557 "Screamer: post: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf"\n",
8558 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8559 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8561 /* find out how much is left in the read-ahead buffer, and rextract its pointer */
8562 cnt = PerlIO_get_cnt(fp);
8563 ptr = (STDCHAR*)PerlIO_get_ptr(fp); /* reregisterize cnt and ptr */
8564 DEBUG_P(PerlIO_printf(Perl_debug_log,
8565 "Screamer: after getc, ptr=%"UVuf", cnt=%"IVdf"\n",
8566 PTR2UV(ptr),(IV)cnt));
8568 if (i == EOF) /* all done for ever? */
8569 goto thats_really_all_folks;
8571 /* make sure we have enough space in the target sv */
8572 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8574 SvGROW(sv, bpx + cnt + 2);
8575 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8577 /* copy of the char we got from getc() */
8578 *bp++ = (STDCHAR)i; /* store character from PerlIO_getc */
8580 /* make sure we deal with the i being the last character of a separator */
8581 if (rslen && (STDCHAR)i == rslast) /* all done for now? */
8582 goto thats_all_folks;
8586 /* check if we have actually found the separator - only really applies
8588 if ((rslen > 1 && (STRLEN)(bp - (STDCHAR*)SvPVX_const(sv)) < rslen) ||
8589 memNE((char*)bp - rslen, rsptr, rslen))
8590 goto screamer; /* go back to the fray */
8591 thats_really_all_folks:
8593 cnt += shortbuffered;
8594 DEBUG_P(PerlIO_printf(Perl_debug_log,
8595 "Screamer: quitting, ptr=%"UVuf", cnt=%"IVdf"\n",PTR2UV(ptr),(IV)cnt));
8596 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* put these back or we're in trouble */
8597 DEBUG_P(PerlIO_printf(Perl_debug_log,
8598 "Screamer: end: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf
8600 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8601 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8603 SvCUR_set(sv, bp - (STDCHAR*)SvPVX_const(sv)); /* set length */
8604 DEBUG_P(PerlIO_printf(Perl_debug_log,
8605 "Screamer: done, len=%ld, string=|%.*s|\n",
8606 (long)SvCUR(sv),(int)SvCUR(sv),SvPVX_const(sv)));
8610 /*The big, slow, and stupid way. */
8611 #ifdef USE_HEAP_INSTEAD_OF_STACK /* Even slower way. */
8612 STDCHAR *buf = NULL;
8613 Newx(buf, 8192, STDCHAR);
8621 const STDCHAR * const bpe = buf + sizeof(buf);
8623 while ((i = PerlIO_getc(fp)) != EOF && (*bp++ = (STDCHAR)i) != rslast && bp < bpe)
8624 ; /* keep reading */
8628 cnt = PerlIO_read(fp,(char*)buf, sizeof(buf));
8629 /* Accommodate broken VAXC compiler, which applies U8 cast to
8630 * both args of ?: operator, causing EOF to change into 255
8633 i = (U8)buf[cnt - 1];
8639 cnt = 0; /* we do need to re-set the sv even when cnt <= 0 */
8641 sv_catpvn_nomg(sv, (char *) buf, cnt);
8643 sv_setpvn(sv, (char *) buf, cnt); /* "nomg" is implied */
8645 if (i != EOF && /* joy */
8647 SvCUR(sv) < rslen ||
8648 memNE(SvPVX_const(sv) + SvCUR(sv) - rslen, rsptr, rslen)))
8652 * If we're reading from a TTY and we get a short read,
8653 * indicating that the user hit his EOF character, we need
8654 * to notice it now, because if we try to read from the TTY
8655 * again, the EOF condition will disappear.
8657 * The comparison of cnt to sizeof(buf) is an optimization
8658 * that prevents unnecessary calls to feof().
8662 if (!(cnt < (I32)sizeof(buf) && PerlIO_eof(fp)))
8666 #ifdef USE_HEAP_INSTEAD_OF_STACK
8671 if (rspara) { /* have to do this both before and after */
8672 while (i != EOF) { /* to make sure file boundaries work right */
8673 i = PerlIO_getc(fp);
8675 PerlIO_ungetc(fp,i);
8681 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8687 Auto-increment of the value in the SV, doing string to numeric conversion
8688 if necessary. Handles 'get' magic and operator overloading.
8694 Perl_sv_inc(pTHX_ SV *const sv)
8703 =for apidoc sv_inc_nomg
8705 Auto-increment of the value in the SV, doing string to numeric conversion
8706 if necessary. Handles operator overloading. Skips handling 'get' magic.
8712 Perl_sv_inc_nomg(pTHX_ SV *const sv)
8719 if (SvTHINKFIRST(sv)) {
8720 if (SvREADONLY(sv)) {
8721 Perl_croak_no_modify();
8725 if (SvAMAGIC(sv) && AMG_CALLunary(sv, inc_amg))
8727 i = PTR2IV(SvRV(sv));
8731 else sv_force_normal_flags(sv, 0);
8733 flags = SvFLAGS(sv);
8734 if ((flags & (SVp_NOK|SVp_IOK)) == SVp_NOK) {
8735 /* It's (privately or publicly) a float, but not tested as an
8736 integer, so test it to see. */
8738 flags = SvFLAGS(sv);
8740 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8741 /* It's publicly an integer, or privately an integer-not-float */
8742 #ifdef PERL_PRESERVE_IVUV
8746 if (SvUVX(sv) == UV_MAX)
8747 sv_setnv(sv, UV_MAX_P1);
8749 (void)SvIOK_only_UV(sv);
8750 SvUV_set(sv, SvUVX(sv) + 1);
8752 if (SvIVX(sv) == IV_MAX)
8753 sv_setuv(sv, (UV)IV_MAX + 1);
8755 (void)SvIOK_only(sv);
8756 SvIV_set(sv, SvIVX(sv) + 1);
8761 if (flags & SVp_NOK) {
8762 const NV was = SvNVX(sv);
8763 if (LIKELY(!Perl_isinfnan(was)) &&
8764 NV_OVERFLOWS_INTEGERS_AT &&
8765 was >= NV_OVERFLOWS_INTEGERS_AT) {
8766 /* diag_listed_as: Lost precision when %s %f by 1 */
8767 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8768 "Lost precision when incrementing %" NVff " by 1",
8771 (void)SvNOK_only(sv);
8772 SvNV_set(sv, was + 1.0);
8776 if (!(flags & SVp_POK) || !*SvPVX_const(sv)) {
8777 if ((flags & SVTYPEMASK) < SVt_PVIV)
8778 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV ? SVt_PVIV : SVt_IV));
8779 (void)SvIOK_only(sv);
8784 while (isALPHA(*d)) d++;
8785 while (isDIGIT(*d)) d++;
8786 if (d < SvEND(sv)) {
8787 const int numtype = grok_number_flags(SvPVX_const(sv), SvCUR(sv), NULL, PERL_SCAN_TRAILING);
8788 #ifdef PERL_PRESERVE_IVUV
8789 /* Got to punt this as an integer if needs be, but we don't issue
8790 warnings. Probably ought to make the sv_iv_please() that does
8791 the conversion if possible, and silently. */
8792 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
8793 /* Need to try really hard to see if it's an integer.
8794 9.22337203685478e+18 is an integer.
8795 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
8796 so $a="9.22337203685478e+18"; $a+0; $a++
8797 needs to be the same as $a="9.22337203685478e+18"; $a++
8804 /* sv_2iv *should* have made this an NV */
8805 if (flags & SVp_NOK) {
8806 (void)SvNOK_only(sv);
8807 SvNV_set(sv, SvNVX(sv) + 1.0);
8810 /* I don't think we can get here. Maybe I should assert this
8811 And if we do get here I suspect that sv_setnv will croak. NWC
8813 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_inc punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"NVgf"\n",
8814 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
8816 #endif /* PERL_PRESERVE_IVUV */
8817 if (!numtype && ckWARN(WARN_NUMERIC))
8818 not_incrementable(sv);
8819 sv_setnv(sv,Atof(SvPVX_const(sv)) + 1.0);
8823 while (d >= SvPVX_const(sv)) {
8831 /* MKS: The original code here died if letters weren't consecutive.
8832 * at least it didn't have to worry about non-C locales. The
8833 * new code assumes that ('z'-'a')==('Z'-'A'), letters are
8834 * arranged in order (although not consecutively) and that only
8835 * [A-Za-z] are accepted by isALPHA in the C locale.
8837 if (isALPHA_FOLD_NE(*d, 'z')) {
8838 do { ++*d; } while (!isALPHA(*d));
8841 *(d--) -= 'z' - 'a';
8846 *(d--) -= 'z' - 'a' + 1;
8850 /* oh,oh, the number grew */
8851 SvGROW(sv, SvCUR(sv) + 2);
8852 SvCUR_set(sv, SvCUR(sv) + 1);
8853 for (d = SvPVX(sv) + SvCUR(sv); d > SvPVX_const(sv); d--)
8864 Auto-decrement of the value in the SV, doing string to numeric conversion
8865 if necessary. Handles 'get' magic and operator overloading.
8871 Perl_sv_dec(pTHX_ SV *const sv)
8880 =for apidoc sv_dec_nomg
8882 Auto-decrement of the value in the SV, doing string to numeric conversion
8883 if necessary. Handles operator overloading. Skips handling 'get' magic.
8889 Perl_sv_dec_nomg(pTHX_ SV *const sv)
8895 if (SvTHINKFIRST(sv)) {
8896 if (SvREADONLY(sv)) {
8897 Perl_croak_no_modify();
8901 if (SvAMAGIC(sv) && AMG_CALLunary(sv, dec_amg))
8903 i = PTR2IV(SvRV(sv));
8907 else sv_force_normal_flags(sv, 0);
8909 /* Unlike sv_inc we don't have to worry about string-never-numbers
8910 and keeping them magic. But we mustn't warn on punting */
8911 flags = SvFLAGS(sv);
8912 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8913 /* It's publicly an integer, or privately an integer-not-float */
8914 #ifdef PERL_PRESERVE_IVUV
8918 if (SvUVX(sv) == 0) {
8919 (void)SvIOK_only(sv);
8923 (void)SvIOK_only_UV(sv);
8924 SvUV_set(sv, SvUVX(sv) - 1);
8927 if (SvIVX(sv) == IV_MIN) {
8928 sv_setnv(sv, (NV)IV_MIN);
8932 (void)SvIOK_only(sv);
8933 SvIV_set(sv, SvIVX(sv) - 1);
8938 if (flags & SVp_NOK) {
8941 const NV was = SvNVX(sv);
8942 if (LIKELY(!Perl_isinfnan(was)) &&
8943 NV_OVERFLOWS_INTEGERS_AT &&
8944 was <= -NV_OVERFLOWS_INTEGERS_AT) {
8945 /* diag_listed_as: Lost precision when %s %f by 1 */
8946 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8947 "Lost precision when decrementing %" NVff " by 1",
8950 (void)SvNOK_only(sv);
8951 SvNV_set(sv, was - 1.0);
8955 if (!(flags & SVp_POK)) {
8956 if ((flags & SVTYPEMASK) < SVt_PVIV)
8957 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV) ? SVt_PVIV : SVt_IV);
8959 (void)SvIOK_only(sv);
8962 #ifdef PERL_PRESERVE_IVUV
8964 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), NULL);
8965 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
8966 /* Need to try really hard to see if it's an integer.
8967 9.22337203685478e+18 is an integer.
8968 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
8969 so $a="9.22337203685478e+18"; $a+0; $a--
8970 needs to be the same as $a="9.22337203685478e+18"; $a--
8977 /* sv_2iv *should* have made this an NV */
8978 if (flags & SVp_NOK) {
8979 (void)SvNOK_only(sv);
8980 SvNV_set(sv, SvNVX(sv) - 1.0);
8983 /* I don't think we can get here. Maybe I should assert this
8984 And if we do get here I suspect that sv_setnv will croak. NWC
8986 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_dec punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"NVgf"\n",
8987 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
8990 #endif /* PERL_PRESERVE_IVUV */
8991 sv_setnv(sv,Atof(SvPVX_const(sv)) - 1.0); /* punt */
8994 /* this define is used to eliminate a chunk of duplicated but shared logic
8995 * it has the suffix __SV_C to signal that it isnt API, and isnt meant to be
8996 * used anywhere but here - yves
8998 #define PUSH_EXTEND_MORTAL__SV_C(AnSv) \
9000 SSize_t ix = ++PL_tmps_ix; \
9001 if (UNLIKELY(ix >= PL_tmps_max)) \
9002 ix = tmps_grow_p(ix); \
9003 PL_tmps_stack[ix] = (AnSv); \
9007 =for apidoc sv_mortalcopy
9009 Creates a new SV which is a copy of the original SV (using C<sv_setsv>).
9010 The new SV is marked as mortal. It will be destroyed "soon", either by an
9011 explicit call to C<FREETMPS>, or by an implicit call at places such as
9012 statement boundaries. See also C<L</sv_newmortal>> and C<L</sv_2mortal>>.
9017 /* Make a string that will exist for the duration of the expression
9018 * evaluation. Actually, it may have to last longer than that, but
9019 * hopefully we won't free it until it has been assigned to a
9020 * permanent location. */
9023 Perl_sv_mortalcopy_flags(pTHX_ SV *const oldstr, U32 flags)
9027 if (flags & SV_GMAGIC)
9028 SvGETMAGIC(oldstr); /* before new_SV, in case it dies */
9030 sv_setsv_flags(sv,oldstr,flags & ~SV_GMAGIC);
9031 PUSH_EXTEND_MORTAL__SV_C(sv);
9037 =for apidoc sv_newmortal
9039 Creates a new null SV which is mortal. The reference count of the SV is
9040 set to 1. It will be destroyed "soon", either by an explicit call to
9041 C<FREETMPS>, or by an implicit call at places such as statement boundaries.
9042 See also C<L</sv_mortalcopy>> and C<L</sv_2mortal>>.
9048 Perl_sv_newmortal(pTHX)
9053 SvFLAGS(sv) = SVs_TEMP;
9054 PUSH_EXTEND_MORTAL__SV_C(sv);
9060 =for apidoc newSVpvn_flags
9062 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9063 characters) into it. The reference count for the
9064 SV is set to 1. Note that if C<len> is zero, Perl will create a zero length
9065 string. You are responsible for ensuring that the source string is at least
9066 C<len> bytes long. If the C<s> argument is NULL the new SV will be undefined.
9067 Currently the only flag bits accepted are C<SVf_UTF8> and C<SVs_TEMP>.
9068 If C<SVs_TEMP> is set, then C<sv_2mortal()> is called on the result before
9069 returning. If C<SVf_UTF8> is set, C<s>
9070 is considered to be in UTF-8 and the
9071 C<SVf_UTF8> flag will be set on the new SV.
9072 C<newSVpvn_utf8()> is a convenience wrapper for this function, defined as
9074 #define newSVpvn_utf8(s, len, u) \
9075 newSVpvn_flags((s), (len), (u) ? SVf_UTF8 : 0)
9081 Perl_newSVpvn_flags(pTHX_ const char *const s, const STRLEN len, const U32 flags)
9085 /* All the flags we don't support must be zero.
9086 And we're new code so I'm going to assert this from the start. */
9087 assert(!(flags & ~(SVf_UTF8|SVs_TEMP)));
9089 sv_setpvn(sv,s,len);
9091 /* This code used to do a sv_2mortal(), however we now unroll the call to
9092 * sv_2mortal() and do what it does ourselves here. Since we have asserted
9093 * that flags can only have the SVf_UTF8 and/or SVs_TEMP flags set above we
9094 * can use it to enable the sv flags directly (bypassing SvTEMP_on), which
9095 * in turn means we dont need to mask out the SVf_UTF8 flag below, which
9096 * means that we eliminate quite a few steps than it looks - Yves
9097 * (explaining patch by gfx) */
9099 SvFLAGS(sv) |= flags;
9101 if(flags & SVs_TEMP){
9102 PUSH_EXTEND_MORTAL__SV_C(sv);
9109 =for apidoc sv_2mortal
9111 Marks an existing SV as mortal. The SV will be destroyed "soon", either
9112 by an explicit call to C<FREETMPS>, or by an implicit call at places such as
9113 statement boundaries. C<SvTEMP()> is turned on which means that the SV's
9114 string buffer can be "stolen" if this SV is copied. See also
9115 C<L</sv_newmortal>> and C<L</sv_mortalcopy>>.
9121 Perl_sv_2mortal(pTHX_ SV *const sv)
9128 PUSH_EXTEND_MORTAL__SV_C(sv);
9136 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9137 characters) into it. The reference count for the
9138 SV is set to 1. If C<len> is zero, Perl will compute the length using
9139 C<strlen()>, (which means if you use this option, that C<s> can't have embedded
9140 C<NUL> characters and has to have a terminating C<NUL> byte).
9142 For efficiency, consider using C<newSVpvn> instead.
9148 Perl_newSVpv(pTHX_ const char *const s, const STRLEN len)
9153 sv_setpvn(sv, s, len || s == NULL ? len : strlen(s));
9158 =for apidoc newSVpvn
9160 Creates a new SV and copies a string into it, which may contain C<NUL> characters
9161 (C<\0>) and other binary data. The reference count for the SV is set to 1.
9162 Note that if C<len> is zero, Perl will create a zero length (Perl) string. You
9163 are responsible for ensuring that the source buffer is at least
9164 C<len> bytes long. If the C<buffer> argument is NULL the new SV will be
9171 Perl_newSVpvn(pTHX_ const char *const buffer, const STRLEN len)
9175 sv_setpvn(sv,buffer,len);
9180 =for apidoc newSVhek
9182 Creates a new SV from the hash key structure. It will generate scalars that
9183 point to the shared string table where possible. Returns a new (undefined)
9184 SV if C<hek> is NULL.
9190 Perl_newSVhek(pTHX_ const HEK *const hek)
9199 if (HEK_LEN(hek) == HEf_SVKEY) {
9200 return newSVsv(*(SV**)HEK_KEY(hek));
9202 const int flags = HEK_FLAGS(hek);
9203 if (flags & HVhek_WASUTF8) {
9205 Andreas would like keys he put in as utf8 to come back as utf8
9207 STRLEN utf8_len = HEK_LEN(hek);
9208 SV * const sv = newSV_type(SVt_PV);
9209 char *as_utf8 = (char *)bytes_to_utf8 ((U8*)HEK_KEY(hek), &utf8_len);
9210 /* bytes_to_utf8() allocates a new string, which we can repurpose: */
9211 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
9214 } else if (flags & HVhek_UNSHARED) {
9215 /* A hash that isn't using shared hash keys has to have
9216 the flag in every key so that we know not to try to call
9217 share_hek_hek on it. */
9219 SV * const sv = newSVpvn (HEK_KEY(hek), HEK_LEN(hek));
9224 /* This will be overwhelminly the most common case. */
9226 /* Inline most of newSVpvn_share(), because share_hek_hek() is far
9227 more efficient than sharepvn(). */
9231 sv_upgrade(sv, SVt_PV);
9232 SvPV_set(sv, (char *)HEK_KEY(share_hek_hek(hek)));
9233 SvCUR_set(sv, HEK_LEN(hek));
9245 =for apidoc newSVpvn_share
9247 Creates a new SV with its C<SvPVX_const> pointing to a shared string in the string
9248 table. If the string does not already exist in the table, it is
9249 created first. Turns on the C<SvIsCOW> flag (or C<READONLY>
9250 and C<FAKE> in 5.16 and earlier). If the C<hash> parameter
9251 is non-zero, that value is used; otherwise the hash is computed.
9252 The string's hash can later be retrieved from the SV
9253 with the C<SvSHARED_HASH()> macro. The idea here is
9254 that as the string table is used for shared hash keys these strings will have
9255 C<SvPVX_const == HeKEY> and hash lookup will avoid string compare.
9261 Perl_newSVpvn_share(pTHX_ const char *src, I32 len, U32 hash)
9265 bool is_utf8 = FALSE;
9266 const char *const orig_src = src;
9269 STRLEN tmplen = -len;
9271 /* See the note in hv.c:hv_fetch() --jhi */
9272 src = (char*)bytes_from_utf8((const U8*)src, &tmplen, &is_utf8);
9276 PERL_HASH(hash, src, len);
9278 /* The logic for this is inlined in S_mro_get_linear_isa_dfs(), so if it
9279 changes here, update it there too. */
9280 sv_upgrade(sv, SVt_PV);
9281 SvPV_set(sv, sharepvn(src, is_utf8?-len:len, hash));
9288 if (src != orig_src)
9294 =for apidoc newSVpv_share
9296 Like C<newSVpvn_share>, but takes a C<NUL>-terminated string instead of a
9303 Perl_newSVpv_share(pTHX_ const char *src, U32 hash)
9305 return newSVpvn_share(src, strlen(src), hash);
9308 #if defined(PERL_IMPLICIT_CONTEXT)
9310 /* pTHX_ magic can't cope with varargs, so this is a no-context
9311 * version of the main function, (which may itself be aliased to us).
9312 * Don't access this version directly.
9316 Perl_newSVpvf_nocontext(const char *const pat, ...)
9322 PERL_ARGS_ASSERT_NEWSVPVF_NOCONTEXT;
9324 va_start(args, pat);
9325 sv = vnewSVpvf(pat, &args);
9332 =for apidoc newSVpvf
9334 Creates a new SV and initializes it with the string formatted like
9341 Perl_newSVpvf(pTHX_ const char *const pat, ...)
9346 PERL_ARGS_ASSERT_NEWSVPVF;
9348 va_start(args, pat);
9349 sv = vnewSVpvf(pat, &args);
9354 /* backend for newSVpvf() and newSVpvf_nocontext() */
9357 Perl_vnewSVpvf(pTHX_ const char *const pat, va_list *const args)
9361 PERL_ARGS_ASSERT_VNEWSVPVF;
9364 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
9371 Creates a new SV and copies a floating point value into it.
9372 The reference count for the SV is set to 1.
9378 Perl_newSVnv(pTHX_ const NV n)
9390 Creates a new SV and copies an integer into it. The reference count for the
9397 Perl_newSViv(pTHX_ const IV i)
9403 /* Inlining ONLY the small relevant subset of sv_setiv here
9404 * for performance. Makes a significant difference. */
9406 /* We're starting from SVt_FIRST, so provided that's
9407 * actual 0, we don't have to unset any SV type flags
9408 * to promote to SVt_IV. */
9409 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9411 SET_SVANY_FOR_BODYLESS_IV(sv);
9412 SvFLAGS(sv) |= SVt_IV;
9424 Creates a new SV and copies an unsigned integer into it.
9425 The reference count for the SV is set to 1.
9431 Perl_newSVuv(pTHX_ const UV u)
9435 /* Inlining ONLY the small relevant subset of sv_setuv here
9436 * for performance. Makes a significant difference. */
9438 /* Using ivs is more efficient than using uvs - see sv_setuv */
9439 if (u <= (UV)IV_MAX) {
9440 return newSViv((IV)u);
9445 /* We're starting from SVt_FIRST, so provided that's
9446 * actual 0, we don't have to unset any SV type flags
9447 * to promote to SVt_IV. */
9448 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9450 SET_SVANY_FOR_BODYLESS_IV(sv);
9451 SvFLAGS(sv) |= SVt_IV;
9453 (void)SvIsUV_on(sv);
9462 =for apidoc newSV_type
9464 Creates a new SV, of the type specified. The reference count for the new SV
9471 Perl_newSV_type(pTHX_ const svtype type)
9476 ASSUME(SvTYPE(sv) == SVt_FIRST);
9477 if(type != SVt_FIRST)
9478 sv_upgrade(sv, type);
9483 =for apidoc newRV_noinc
9485 Creates an RV wrapper for an SV. The reference count for the original
9486 SV is B<not> incremented.
9492 Perl_newRV_noinc(pTHX_ SV *const tmpRef)
9496 PERL_ARGS_ASSERT_NEWRV_NOINC;
9500 /* We're starting from SVt_FIRST, so provided that's
9501 * actual 0, we don't have to unset any SV type flags
9502 * to promote to SVt_IV. */
9503 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9505 SET_SVANY_FOR_BODYLESS_IV(sv);
9506 SvFLAGS(sv) |= SVt_IV;
9511 SvRV_set(sv, tmpRef);
9516 /* newRV_inc is the official function name to use now.
9517 * newRV_inc is in fact #defined to newRV in sv.h
9521 Perl_newRV(pTHX_ SV *const sv)
9523 PERL_ARGS_ASSERT_NEWRV;
9525 return newRV_noinc(SvREFCNT_inc_simple_NN(sv));
9531 Creates a new SV which is an exact duplicate of the original SV.
9538 Perl_newSVsv(pTHX_ SV *const old)
9544 if (SvTYPE(old) == (svtype)SVTYPEMASK) {
9545 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), "semi-panic: attempt to dup freed string");
9548 /* Do this here, otherwise we leak the new SV if this croaks. */
9551 /* SV_NOSTEAL prevents TEMP buffers being, well, stolen, and saves games
9552 with SvTEMP_off and SvTEMP_on round a call to sv_setsv. */
9553 sv_setsv_flags(sv, old, SV_NOSTEAL);
9558 =for apidoc sv_reset
9560 Underlying implementation for the C<reset> Perl function.
9561 Note that the perl-level function is vaguely deprecated.
9567 Perl_sv_reset(pTHX_ const char *s, HV *const stash)
9569 PERL_ARGS_ASSERT_SV_RESET;
9571 sv_resetpvn(*s ? s : NULL, strlen(s), stash);
9575 Perl_sv_resetpvn(pTHX_ const char *s, STRLEN len, HV * const stash)
9577 char todo[PERL_UCHAR_MAX+1];
9580 if (!stash || SvTYPE(stash) != SVt_PVHV)
9583 if (!s) { /* reset ?? searches */
9584 MAGIC * const mg = mg_find((const SV *)stash, PERL_MAGIC_symtab);
9586 const U32 count = mg->mg_len / sizeof(PMOP**);
9587 PMOP **pmp = (PMOP**) mg->mg_ptr;
9588 PMOP *const *const end = pmp + count;
9592 SvREADONLY_off(PL_regex_pad[(*pmp)->op_pmoffset]);
9594 (*pmp)->op_pmflags &= ~PMf_USED;
9602 /* reset variables */
9604 if (!HvARRAY(stash))
9607 Zero(todo, 256, char);
9611 I32 i = (unsigned char)*s;
9615 max = (unsigned char)*s++;
9616 for ( ; i <= max; i++) {
9619 for (i = 0; i <= (I32) HvMAX(stash); i++) {
9621 for (entry = HvARRAY(stash)[i];
9623 entry = HeNEXT(entry))
9628 if (!todo[(U8)*HeKEY(entry)])
9630 gv = MUTABLE_GV(HeVAL(entry));
9632 if (sv && !SvREADONLY(sv)) {
9633 SV_CHECK_THINKFIRST_COW_DROP(sv);
9634 if (!isGV(sv)) SvOK_off(sv);
9639 if (GvHV(gv) && !HvNAME_get(GvHV(gv))) {
9650 Using various gambits, try to get an IO from an SV: the IO slot if its a
9651 GV; or the recursive result if we're an RV; or the IO slot of the symbol
9652 named after the PV if we're a string.
9654 'Get' magic is ignored on the C<sv> passed in, but will be called on
9655 C<SvRV(sv)> if C<sv> is an RV.
9661 Perl_sv_2io(pTHX_ SV *const sv)
9666 PERL_ARGS_ASSERT_SV_2IO;
9668 switch (SvTYPE(sv)) {
9670 io = MUTABLE_IO(sv);
9674 if (isGV_with_GP(sv)) {
9675 gv = MUTABLE_GV(sv);
9678 Perl_croak(aTHX_ "Bad filehandle: %"HEKf,
9679 HEKfARG(GvNAME_HEK(gv)));
9685 Perl_croak(aTHX_ PL_no_usym, "filehandle");
9687 SvGETMAGIC(SvRV(sv));
9688 return sv_2io(SvRV(sv));
9690 gv = gv_fetchsv_nomg(sv, 0, SVt_PVIO);
9697 if (SvGMAGICAL(sv)) {
9698 newsv = sv_newmortal();
9699 sv_setsv_nomg(newsv, sv);
9701 Perl_croak(aTHX_ "Bad filehandle: %"SVf, SVfARG(newsv));
9711 Using various gambits, try to get a CV from an SV; in addition, try if
9712 possible to set C<*st> and C<*gvp> to the stash and GV associated with it.
9713 The flags in C<lref> are passed to C<gv_fetchsv>.
9719 Perl_sv_2cv(pTHX_ SV *sv, HV **const st, GV **const gvp, const I32 lref)
9724 PERL_ARGS_ASSERT_SV_2CV;
9731 switch (SvTYPE(sv)) {
9735 return MUTABLE_CV(sv);
9745 sv = amagic_deref_call(sv, to_cv_amg);
9748 if (SvTYPE(sv) == SVt_PVCV) {
9749 cv = MUTABLE_CV(sv);
9754 else if(SvGETMAGIC(sv), isGV_with_GP(sv))
9755 gv = MUTABLE_GV(sv);
9757 Perl_croak(aTHX_ "Not a subroutine reference");
9759 else if (isGV_with_GP(sv)) {
9760 gv = MUTABLE_GV(sv);
9763 gv = gv_fetchsv_nomg(sv, lref, SVt_PVCV);
9770 /* Some flags to gv_fetchsv mean don't really create the GV */
9771 if (!isGV_with_GP(gv)) {
9776 if (lref & ~GV_ADDMG && !GvCVu(gv)) {
9777 /* XXX this is probably not what they think they're getting.
9778 * It has the same effect as "sub name;", i.e. just a forward
9789 Returns true if the SV has a true value by Perl's rules.
9790 Use the C<SvTRUE> macro instead, which may call C<sv_true()> or may
9791 instead use an in-line version.
9797 Perl_sv_true(pTHX_ SV *const sv)
9802 const XPV* const tXpv = (XPV*)SvANY(sv);
9804 (tXpv->xpv_cur > 1 ||
9805 (tXpv->xpv_cur && *sv->sv_u.svu_pv != '0')))
9812 return SvIVX(sv) != 0;
9815 return SvNVX(sv) != 0.0;
9817 return sv_2bool(sv);
9823 =for apidoc sv_pvn_force
9825 Get a sensible string out of the SV somehow.
9826 A private implementation of the C<SvPV_force> macro for compilers which
9827 can't cope with complex macro expressions. Always use the macro instead.
9829 =for apidoc sv_pvn_force_flags
9831 Get a sensible string out of the SV somehow.
9832 If C<flags> has the C<SV_GMAGIC> bit set, will C<mg_get> on C<sv> if
9833 appropriate, else not. C<sv_pvn_force> and C<sv_pvn_force_nomg> are
9834 implemented in terms of this function.
9835 You normally want to use the various wrapper macros instead: see
9836 C<L</SvPV_force>> and C<L</SvPV_force_nomg>>.
9842 Perl_sv_pvn_force_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
9844 PERL_ARGS_ASSERT_SV_PVN_FORCE_FLAGS;
9846 if (flags & SV_GMAGIC) SvGETMAGIC(sv);
9847 if (SvTHINKFIRST(sv) && (!SvROK(sv) || SvREADONLY(sv)))
9848 sv_force_normal_flags(sv, 0);
9858 if (SvTYPE(sv) > SVt_PVLV
9859 || isGV_with_GP(sv))
9860 /* diag_listed_as: Can't coerce %s to %s in %s */
9861 Perl_croak(aTHX_ "Can't coerce %s to string in %s", sv_reftype(sv,0),
9863 s = sv_2pv_flags(sv, &len, flags &~ SV_GMAGIC);
9870 if (SvTYPE(sv) < SVt_PV ||
9871 s != SvPVX_const(sv)) { /* Almost, but not quite, sv_setpvn() */
9874 SvUPGRADE(sv, SVt_PV); /* Never FALSE */
9875 SvGROW(sv, len + 1);
9876 Move(s,SvPVX(sv),len,char);
9878 SvPVX(sv)[len] = '\0';
9881 SvPOK_on(sv); /* validate pointer */
9883 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
9884 PTR2UV(sv),SvPVX_const(sv)));
9887 (void)SvPOK_only_UTF8(sv);
9888 return SvPVX_mutable(sv);
9892 =for apidoc sv_pvbyten_force
9894 The backend for the C<SvPVbytex_force> macro. Always use the macro
9901 Perl_sv_pvbyten_force(pTHX_ SV *const sv, STRLEN *const lp)
9903 PERL_ARGS_ASSERT_SV_PVBYTEN_FORCE;
9905 sv_pvn_force(sv,lp);
9906 sv_utf8_downgrade(sv,0);
9912 =for apidoc sv_pvutf8n_force
9914 The backend for the C<SvPVutf8x_force> macro. Always use the macro
9921 Perl_sv_pvutf8n_force(pTHX_ SV *const sv, STRLEN *const lp)
9923 PERL_ARGS_ASSERT_SV_PVUTF8N_FORCE;
9926 sv_utf8_upgrade_nomg(sv);
9932 =for apidoc sv_reftype
9934 Returns a string describing what the SV is a reference to.
9940 Perl_sv_reftype(pTHX_ const SV *const sv, const int ob)
9942 PERL_ARGS_ASSERT_SV_REFTYPE;
9943 if (ob && SvOBJECT(sv)) {
9944 return SvPV_nolen_const(sv_ref(NULL, sv, ob));
9947 /* WARNING - There is code, for instance in mg.c, that assumes that
9948 * the only reason that sv_reftype(sv,0) would return a string starting
9949 * with 'L' or 'S' is that it is a LVALUE or a SCALAR.
9950 * Yes this a dodgy way to do type checking, but it saves practically reimplementing
9951 * this routine inside other subs, and it saves time.
9952 * Do not change this assumption without searching for "dodgy type check" in
9955 switch (SvTYPE(sv)) {
9970 case SVt_PVLV: return (char *) (SvROK(sv) ? "REF"
9971 /* tied lvalues should appear to be
9972 * scalars for backwards compatibility */
9973 : (isALPHA_FOLD_EQ(LvTYPE(sv), 't'))
9974 ? "SCALAR" : "LVALUE");
9975 case SVt_PVAV: return "ARRAY";
9976 case SVt_PVHV: return "HASH";
9977 case SVt_PVCV: return "CODE";
9978 case SVt_PVGV: return (char *) (isGV_with_GP(sv)
9979 ? "GLOB" : "SCALAR");
9980 case SVt_PVFM: return "FORMAT";
9981 case SVt_PVIO: return "IO";
9982 case SVt_INVLIST: return "INVLIST";
9983 case SVt_REGEXP: return "REGEXP";
9984 default: return "UNKNOWN";
9992 Returns a SV describing what the SV passed in is a reference to.
9998 Perl_sv_ref(pTHX_ SV *dst, const SV *const sv, const int ob)
10000 PERL_ARGS_ASSERT_SV_REF;
10003 dst = sv_newmortal();
10005 if (ob && SvOBJECT(sv)) {
10006 HvNAME_get(SvSTASH(sv))
10007 ? sv_sethek(dst, HvNAME_HEK(SvSTASH(sv)))
10008 : sv_setpvn(dst, "__ANON__", 8);
10011 const char * reftype = sv_reftype(sv, 0);
10012 sv_setpv(dst, reftype);
10018 =for apidoc sv_isobject
10020 Returns a boolean indicating whether the SV is an RV pointing to a blessed
10021 object. If the SV is not an RV, or if the object is not blessed, then this
10028 Perl_sv_isobject(pTHX_ SV *sv)
10044 Returns a boolean indicating whether the SV is blessed into the specified
10045 class. This does not check for subtypes; use C<sv_derived_from> to verify
10046 an inheritance relationship.
10052 Perl_sv_isa(pTHX_ SV *sv, const char *const name)
10054 const char *hvname;
10056 PERL_ARGS_ASSERT_SV_ISA;
10066 hvname = HvNAME_get(SvSTASH(sv));
10070 return strEQ(hvname, name);
10074 =for apidoc newSVrv
10076 Creates a new SV for the existing RV, C<rv>, to point to. If C<rv> is not an
10077 RV then it will be upgraded to one. If C<classname> is non-null then the new
10078 SV will be blessed in the specified package. The new SV is returned and its
10079 reference count is 1. The reference count 1 is owned by C<rv>.
10085 Perl_newSVrv(pTHX_ SV *const rv, const char *const classname)
10089 PERL_ARGS_ASSERT_NEWSVRV;
10093 SV_CHECK_THINKFIRST_COW_DROP(rv);
10095 if (UNLIKELY( SvTYPE(rv) >= SVt_PVMG )) {
10096 const U32 refcnt = SvREFCNT(rv);
10100 SvREFCNT(rv) = refcnt;
10102 sv_upgrade(rv, SVt_IV);
10103 } else if (SvROK(rv)) {
10104 SvREFCNT_dec(SvRV(rv));
10106 prepare_SV_for_RV(rv);
10114 HV* const stash = gv_stashpv(classname, GV_ADD);
10115 (void)sv_bless(rv, stash);
10121 Perl_newSVavdefelem(pTHX_ AV *av, SSize_t ix, bool extendible)
10123 SV * const lv = newSV_type(SVt_PVLV);
10124 PERL_ARGS_ASSERT_NEWSVAVDEFELEM;
10126 sv_magic(lv, NULL, PERL_MAGIC_defelem, NULL, 0);
10127 LvTARG(lv) = SvREFCNT_inc_simple_NN(av);
10128 LvSTARGOFF(lv) = ix;
10129 LvTARGLEN(lv) = extendible ? 1 : (STRLEN)UV_MAX;
10134 =for apidoc sv_setref_pv
10136 Copies a pointer into a new SV, optionally blessing the SV. The C<rv>
10137 argument will be upgraded to an RV. That RV will be modified to point to
10138 the new SV. If the C<pv> argument is C<NULL>, then C<PL_sv_undef> will be placed
10139 into the SV. The C<classname> argument indicates the package for the
10140 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10141 will have a reference count of 1, and the RV will be returned.
10143 Do not use with other Perl types such as HV, AV, SV, CV, because those
10144 objects will become corrupted by the pointer copy process.
10146 Note that C<sv_setref_pvn> copies the string while this copies the pointer.
10152 Perl_sv_setref_pv(pTHX_ SV *const rv, const char *const classname, void *const pv)
10154 PERL_ARGS_ASSERT_SV_SETREF_PV;
10157 sv_setsv(rv, &PL_sv_undef);
10161 sv_setiv(newSVrv(rv,classname), PTR2IV(pv));
10166 =for apidoc sv_setref_iv
10168 Copies an integer into a new SV, optionally blessing the SV. The C<rv>
10169 argument will be upgraded to an RV. That RV will be modified to point to
10170 the new SV. The C<classname> argument indicates the package for the
10171 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10172 will have a reference count of 1, and the RV will be returned.
10178 Perl_sv_setref_iv(pTHX_ SV *const rv, const char *const classname, const IV iv)
10180 PERL_ARGS_ASSERT_SV_SETREF_IV;
10182 sv_setiv(newSVrv(rv,classname), iv);
10187 =for apidoc sv_setref_uv
10189 Copies an unsigned integer into a new SV, optionally blessing the SV. The C<rv>
10190 argument will be upgraded to an RV. That RV will be modified to point to
10191 the new SV. The C<classname> argument indicates the package for the
10192 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10193 will have a reference count of 1, and the RV will be returned.
10199 Perl_sv_setref_uv(pTHX_ SV *const rv, const char *const classname, const UV uv)
10201 PERL_ARGS_ASSERT_SV_SETREF_UV;
10203 sv_setuv(newSVrv(rv,classname), uv);
10208 =for apidoc sv_setref_nv
10210 Copies a double into a new SV, optionally blessing the SV. The C<rv>
10211 argument will be upgraded to an RV. That RV will be modified to point to
10212 the new SV. The C<classname> argument indicates the package for the
10213 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10214 will have a reference count of 1, and the RV will be returned.
10220 Perl_sv_setref_nv(pTHX_ SV *const rv, const char *const classname, const NV nv)
10222 PERL_ARGS_ASSERT_SV_SETREF_NV;
10224 sv_setnv(newSVrv(rv,classname), nv);
10229 =for apidoc sv_setref_pvn
10231 Copies a string into a new SV, optionally blessing the SV. The length of the
10232 string must be specified with C<n>. The C<rv> argument will be upgraded to
10233 an RV. That RV will be modified to point to the new SV. The C<classname>
10234 argument indicates the package for the blessing. Set C<classname> to
10235 C<NULL> to avoid the blessing. The new SV will have a reference count
10236 of 1, and the RV will be returned.
10238 Note that C<sv_setref_pv> copies the pointer while this copies the string.
10244 Perl_sv_setref_pvn(pTHX_ SV *const rv, const char *const classname,
10245 const char *const pv, const STRLEN n)
10247 PERL_ARGS_ASSERT_SV_SETREF_PVN;
10249 sv_setpvn(newSVrv(rv,classname), pv, n);
10254 =for apidoc sv_bless
10256 Blesses an SV into a specified package. The SV must be an RV. The package
10257 must be designated by its stash (see C<L</gv_stashpv>>). The reference count
10258 of the SV is unaffected.
10264 Perl_sv_bless(pTHX_ SV *const sv, HV *const stash)
10267 HV *oldstash = NULL;
10269 PERL_ARGS_ASSERT_SV_BLESS;
10273 Perl_croak(aTHX_ "Can't bless non-reference value");
10275 if (SvFLAGS(tmpRef) & (SVs_OBJECT|SVf_READONLY|SVf_PROTECT)) {
10276 if (SvREADONLY(tmpRef))
10277 Perl_croak_no_modify();
10278 if (SvOBJECT(tmpRef)) {
10279 oldstash = SvSTASH(tmpRef);
10282 SvOBJECT_on(tmpRef);
10283 SvUPGRADE(tmpRef, SVt_PVMG);
10284 SvSTASH_set(tmpRef, MUTABLE_HV(SvREFCNT_inc_simple(stash)));
10285 SvREFCNT_dec(oldstash);
10287 if(SvSMAGICAL(tmpRef))
10288 if(mg_find(tmpRef, PERL_MAGIC_ext) || mg_find(tmpRef, PERL_MAGIC_uvar))
10296 /* Downgrades a PVGV to a PVMG. If it's actually a PVLV, we leave the type
10297 * as it is after unglobbing it.
10300 PERL_STATIC_INLINE void
10301 S_sv_unglob(pTHX_ SV *const sv, U32 flags)
10305 SV * const temp = flags & SV_COW_DROP_PV ? NULL : sv_newmortal();
10307 PERL_ARGS_ASSERT_SV_UNGLOB;
10309 assert(SvTYPE(sv) == SVt_PVGV || SvTYPE(sv) == SVt_PVLV);
10311 if (!(flags & SV_COW_DROP_PV))
10312 gv_efullname3(temp, MUTABLE_GV(sv), "*");
10314 SvREFCNT_inc_simple_void_NN(sv_2mortal(sv));
10316 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
10317 && HvNAME_get(stash))
10318 mro_method_changed_in(stash);
10319 gp_free(MUTABLE_GV(sv));
10322 sv_del_backref(MUTABLE_SV(GvSTASH(sv)), sv);
10323 GvSTASH(sv) = NULL;
10326 if (GvNAME_HEK(sv)) {
10327 unshare_hek(GvNAME_HEK(sv));
10329 isGV_with_GP_off(sv);
10331 if(SvTYPE(sv) == SVt_PVGV) {
10332 /* need to keep SvANY(sv) in the right arena */
10333 xpvmg = new_XPVMG();
10334 StructCopy(SvANY(sv), xpvmg, XPVMG);
10335 del_XPVGV(SvANY(sv));
10338 SvFLAGS(sv) &= ~SVTYPEMASK;
10339 SvFLAGS(sv) |= SVt_PVMG;
10342 /* Intentionally not calling any local SET magic, as this isn't so much a
10343 set operation as merely an internal storage change. */
10344 if (flags & SV_COW_DROP_PV) SvOK_off(sv);
10345 else sv_setsv_flags(sv, temp, 0);
10347 if ((const GV *)sv == PL_last_in_gv)
10348 PL_last_in_gv = NULL;
10349 else if ((const GV *)sv == PL_statgv)
10354 =for apidoc sv_unref_flags
10356 Unsets the RV status of the SV, and decrements the reference count of
10357 whatever was being referenced by the RV. This can almost be thought of
10358 as a reversal of C<newSVrv>. The C<cflags> argument can contain
10359 C<SV_IMMEDIATE_UNREF> to force the reference count to be decremented
10360 (otherwise the decrementing is conditional on the reference count being
10361 different from one or the reference being a readonly SV).
10362 See C<L</SvROK_off>>.
10368 Perl_sv_unref_flags(pTHX_ SV *const ref, const U32 flags)
10370 SV* const target = SvRV(ref);
10372 PERL_ARGS_ASSERT_SV_UNREF_FLAGS;
10374 if (SvWEAKREF(ref)) {
10375 sv_del_backref(target, ref);
10376 SvWEAKREF_off(ref);
10377 SvRV_set(ref, NULL);
10380 SvRV_set(ref, NULL);
10382 /* You can't have a || SvREADONLY(target) here, as $a = $$a, where $a was
10383 assigned to as BEGIN {$a = \"Foo"} will fail. */
10384 if (SvREFCNT(target) != 1 || (flags & SV_IMMEDIATE_UNREF))
10385 SvREFCNT_dec_NN(target);
10386 else /* XXX Hack, but hard to make $a=$a->[1] work otherwise */
10387 sv_2mortal(target); /* Schedule for freeing later */
10391 =for apidoc sv_untaint
10393 Untaint an SV. Use C<SvTAINTED_off> instead.
10399 Perl_sv_untaint(pTHX_ SV *const sv)
10401 PERL_ARGS_ASSERT_SV_UNTAINT;
10402 PERL_UNUSED_CONTEXT;
10404 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10405 MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10412 =for apidoc sv_tainted
10414 Test an SV for taintedness. Use C<SvTAINTED> instead.
10420 Perl_sv_tainted(pTHX_ SV *const sv)
10422 PERL_ARGS_ASSERT_SV_TAINTED;
10423 PERL_UNUSED_CONTEXT;
10425 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10426 const MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10427 if (mg && (mg->mg_len & 1) )
10434 =for apidoc sv_setpviv
10436 Copies an integer into the given SV, also updating its string value.
10437 Does not handle 'set' magic. See C<L</sv_setpviv_mg>>.
10443 Perl_sv_setpviv(pTHX_ SV *const sv, const IV iv)
10445 char buf[TYPE_CHARS(UV)];
10447 char * const ptr = uiv_2buf(buf, iv, 0, 0, &ebuf);
10449 PERL_ARGS_ASSERT_SV_SETPVIV;
10451 sv_setpvn(sv, ptr, ebuf - ptr);
10455 =for apidoc sv_setpviv_mg
10457 Like C<sv_setpviv>, but also handles 'set' magic.
10463 Perl_sv_setpviv_mg(pTHX_ SV *const sv, const IV iv)
10465 PERL_ARGS_ASSERT_SV_SETPVIV_MG;
10467 sv_setpviv(sv, iv);
10471 #if defined(PERL_IMPLICIT_CONTEXT)
10473 /* pTHX_ magic can't cope with varargs, so this is a no-context
10474 * version of the main function, (which may itself be aliased to us).
10475 * Don't access this version directly.
10479 Perl_sv_setpvf_nocontext(SV *const sv, const char *const pat, ...)
10484 PERL_ARGS_ASSERT_SV_SETPVF_NOCONTEXT;
10486 va_start(args, pat);
10487 sv_vsetpvf(sv, pat, &args);
10491 /* pTHX_ magic can't cope with varargs, so this is a no-context
10492 * version of the main function, (which may itself be aliased to us).
10493 * Don't access this version directly.
10497 Perl_sv_setpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10502 PERL_ARGS_ASSERT_SV_SETPVF_MG_NOCONTEXT;
10504 va_start(args, pat);
10505 sv_vsetpvf_mg(sv, pat, &args);
10511 =for apidoc sv_setpvf
10513 Works like C<sv_catpvf> but copies the text into the SV instead of
10514 appending it. Does not handle 'set' magic. See C<L</sv_setpvf_mg>>.
10520 Perl_sv_setpvf(pTHX_ SV *const sv, const char *const pat, ...)
10524 PERL_ARGS_ASSERT_SV_SETPVF;
10526 va_start(args, pat);
10527 sv_vsetpvf(sv, pat, &args);
10532 =for apidoc sv_vsetpvf
10534 Works like C<sv_vcatpvf> but copies the text into the SV instead of
10535 appending it. Does not handle 'set' magic. See C<L</sv_vsetpvf_mg>>.
10537 Usually used via its frontend C<sv_setpvf>.
10543 Perl_sv_vsetpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10545 PERL_ARGS_ASSERT_SV_VSETPVF;
10547 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10551 =for apidoc sv_setpvf_mg
10553 Like C<sv_setpvf>, but also handles 'set' magic.
10559 Perl_sv_setpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10563 PERL_ARGS_ASSERT_SV_SETPVF_MG;
10565 va_start(args, pat);
10566 sv_vsetpvf_mg(sv, pat, &args);
10571 =for apidoc sv_vsetpvf_mg
10573 Like C<sv_vsetpvf>, but also handles 'set' magic.
10575 Usually used via its frontend C<sv_setpvf_mg>.
10581 Perl_sv_vsetpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10583 PERL_ARGS_ASSERT_SV_VSETPVF_MG;
10585 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10589 #if defined(PERL_IMPLICIT_CONTEXT)
10591 /* pTHX_ magic can't cope with varargs, so this is a no-context
10592 * version of the main function, (which may itself be aliased to us).
10593 * Don't access this version directly.
10597 Perl_sv_catpvf_nocontext(SV *const sv, const char *const pat, ...)
10602 PERL_ARGS_ASSERT_SV_CATPVF_NOCONTEXT;
10604 va_start(args, pat);
10605 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10609 /* pTHX_ magic can't cope with varargs, so this is a no-context
10610 * version of the main function, (which may itself be aliased to us).
10611 * Don't access this version directly.
10615 Perl_sv_catpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10620 PERL_ARGS_ASSERT_SV_CATPVF_MG_NOCONTEXT;
10622 va_start(args, pat);
10623 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10630 =for apidoc sv_catpvf
10632 Processes its arguments like C<sv_catpvfn>, and appends the formatted
10633 output to an SV. As with C<sv_catpvfn> called with a non-null C-style
10634 variable argument list, argument reordering is not supported.
10635 If the appended data contains "wide" characters
10636 (including, but not limited to, SVs with a UTF-8 PV formatted with C<%s>,
10637 and characters >255 formatted with C<%c>), the original SV might get
10638 upgraded to UTF-8. Handles 'get' magic, but not 'set' magic. See
10639 C<L</sv_catpvf_mg>>. If the original SV was UTF-8, the pattern should be
10640 valid UTF-8; if the original SV was bytes, the pattern should be too.
10645 Perl_sv_catpvf(pTHX_ SV *const sv, const char *const pat, ...)
10649 PERL_ARGS_ASSERT_SV_CATPVF;
10651 va_start(args, pat);
10652 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10657 =for apidoc sv_vcatpvf
10659 Processes its arguments like C<sv_catpvfn> called with a non-null C-style
10660 variable argument list, and appends the formatted
10661 to an SV. Does not handle 'set' magic. See C<L</sv_vcatpvf_mg>>.
10663 Usually used via its frontend C<sv_catpvf>.
10669 Perl_sv_vcatpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10671 PERL_ARGS_ASSERT_SV_VCATPVF;
10673 sv_vcatpvfn_flags(sv, pat, strlen(pat), args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10677 =for apidoc sv_catpvf_mg
10679 Like C<sv_catpvf>, but also handles 'set' magic.
10685 Perl_sv_catpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10689 PERL_ARGS_ASSERT_SV_CATPVF_MG;
10691 va_start(args, pat);
10692 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10698 =for apidoc sv_vcatpvf_mg
10700 Like C<sv_vcatpvf>, but also handles 'set' magic.
10702 Usually used via its frontend C<sv_catpvf_mg>.
10708 Perl_sv_vcatpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10710 PERL_ARGS_ASSERT_SV_VCATPVF_MG;
10712 sv_vcatpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10717 =for apidoc sv_vsetpvfn
10719 Works like C<sv_vcatpvfn> but copies the text into the SV instead of
10722 Usually used via one of its frontends C<sv_vsetpvf> and C<sv_vsetpvf_mg>.
10728 Perl_sv_vsetpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10729 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
10731 PERL_ARGS_ASSERT_SV_VSETPVFN;
10734 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, 0);
10739 * Warn of missing argument to sprintf. The value used in place of such
10740 * arguments should be &PL_sv_no; an undefined value would yield
10741 * inappropriate "use of uninit" warnings [perl #71000].
10744 S_warn_vcatpvfn_missing_argument(pTHX) {
10745 if (ckWARN(WARN_MISSING)) {
10746 Perl_warner(aTHX_ packWARN(WARN_MISSING), "Missing argument in %s",
10747 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
10753 S_expect_number(pTHX_ char **const pattern)
10757 PERL_ARGS_ASSERT_EXPECT_NUMBER;
10759 switch (**pattern) {
10760 case '1': case '2': case '3':
10761 case '4': case '5': case '6':
10762 case '7': case '8': case '9':
10763 var = *(*pattern)++ - '0';
10764 while (isDIGIT(**pattern)) {
10765 const I32 tmp = var * 10 + (*(*pattern)++ - '0');
10767 Perl_croak(aTHX_ "Integer overflow in format string for %s", (PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn"));
10775 S_F0convert(NV nv, char *const endbuf, STRLEN *const len)
10777 const int neg = nv < 0;
10780 PERL_ARGS_ASSERT_F0CONVERT;
10782 if (UNLIKELY(Perl_isinfnan(nv))) {
10783 STRLEN n = S_infnan_2pv(nv, endbuf - *len, *len, 0);
10793 if (uv & 1 && uv == nv)
10794 uv--; /* Round to even */
10796 const unsigned dig = uv % 10;
10798 } while (uv /= 10);
10809 =for apidoc sv_vcatpvfn
10811 =for apidoc sv_vcatpvfn_flags
10813 Processes its arguments like C<vsprintf> and appends the formatted output
10814 to an SV. Uses an array of SVs if the C-style variable argument list is
10815 missing (C<NULL>). Argument reordering (using format specifiers like C<%2$d>
10816 or C<%*2$d>) is supported only when using an array of SVs; using a C-style
10817 C<va_list> argument list with a format string that uses argument reordering
10818 will yield an exception.
10820 When running with taint checks enabled, indicates via
10821 C<maybe_tainted> if results are untrustworthy (often due to the use of
10824 If called as C<sv_vcatpvfn> or flags has the C<SV_GMAGIC> bit set, calls get magic.
10826 Usually used via one of its frontends C<sv_vcatpvf> and C<sv_vcatpvf_mg>.
10831 #define VECTORIZE_ARGS vecsv = va_arg(*args, SV*);\
10832 vecstr = (U8*)SvPV_const(vecsv,veclen);\
10833 vec_utf8 = DO_UTF8(vecsv);
10835 /* XXX maybe_tainted is never assigned to, so the doc above is lying. */
10838 Perl_sv_vcatpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10839 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
10841 PERL_ARGS_ASSERT_SV_VCATPVFN;
10843 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, SV_GMAGIC|SV_SMAGIC);
10846 #ifdef LONGDOUBLE_DOUBLEDOUBLE
10847 /* The first double can be as large as 2**1023, or '1' x '0' x 1023.
10848 * The second double can be as small as 2**-1074, or '0' x 1073 . '1'.
10849 * The sum of them can be '1' . '0' x 2096 . '1', with implied radix point
10850 * after the first 1023 zero bits.
10852 * XXX The 2098 is quite large (262.25 bytes) and therefore some sort
10853 * of dynamically growing buffer might be better, start at just 16 bytes
10854 * (for example) and grow only when necessary. Or maybe just by looking
10855 * at the exponents of the two doubles? */
10856 # define DOUBLEDOUBLE_MAXBITS 2098
10859 /* vhex will contain the values (0..15) of the hex digits ("nybbles"
10860 * of 4 bits); 1 for the implicit 1, and the mantissa bits, four bits
10861 * per xdigit. For the double-double case, this can be rather many.
10862 * The non-double-double-long-double overshoots since all bits of NV
10863 * are not mantissa bits, there are also exponent bits. */
10864 #ifdef LONGDOUBLE_DOUBLEDOUBLE
10865 # define VHEX_SIZE (3+DOUBLEDOUBLE_MAXBITS/4)
10867 # define VHEX_SIZE (1+(NVSIZE * 8)/4)
10870 /* If we do not have a known long double format, (including not using
10871 * long doubles, or long doubles being equal to doubles) then we will
10872 * fall back to the ldexp/frexp route, with which we can retrieve at
10873 * most as many bits as our widest unsigned integer type is. We try
10874 * to get a 64-bit unsigned integer even if we are not using a 64-bit UV.
10876 * (If you want to test the case of UVSIZE == 4, NVSIZE == 8,
10877 * set the MANTISSATYPE to int and the MANTISSASIZE to 4.)
10879 #if defined(HAS_QUAD) && defined(Uquad_t)
10880 # define MANTISSATYPE Uquad_t
10881 # define MANTISSASIZE 8
10883 # define MANTISSATYPE UV
10884 # define MANTISSASIZE UVSIZE
10887 #if defined(DOUBLE_LITTLE_ENDIAN) || defined(LONGDOUBLE_LITTLE_ENDIAN)
10888 # define HEXTRACT_LITTLE_ENDIAN
10889 #elif defined(DOUBLE_BIG_ENDIAN) || defined(LONGDOUBLE_BIG_ENDIAN)
10890 # define HEXTRACT_BIG_ENDIAN
10892 # define HEXTRACT_MIX_ENDIAN
10895 /* S_hextract() is a helper for Perl_sv_vcatpvfn_flags, for extracting
10896 * the hexadecimal values (for %a/%A). The nv is the NV where the value
10897 * are being extracted from (either directly from the long double in-memory
10898 * presentation, or from the uquad computed via frexp+ldexp). frexp also
10899 * is used to update the exponent. vhex is the pointer to the beginning
10900 * of the output buffer (of VHEX_SIZE).
10902 * The tricky part is that S_hextract() needs to be called twice:
10903 * the first time with vend as NULL, and the second time with vend as
10904 * the pointer returned by the first call. What happens is that on
10905 * the first round the output size is computed, and the intended
10906 * extraction sanity checked. On the second round the actual output
10907 * (the extraction of the hexadecimal values) takes place.
10908 * Sanity failures cause fatal failures during both rounds. */
10910 S_hextract(pTHX_ const NV nv, int* exponent, U8* vhex, U8* vend)
10914 int ixmin = 0, ixmax = 0;
10916 /* XXX Inf/NaN/denormal handling in the HEXTRACT_IMPLICIT_BIT,
10917 * and elsewhere. */
10919 /* These macros are just to reduce typos, they have multiple
10920 * repetitions below, but usually only one (or sometimes two)
10921 * of them is really being used. */
10922 /* HEXTRACT_OUTPUT() extracts the high nybble first. */
10923 #define HEXTRACT_OUTPUT_HI(ix) (*v++ = nvp[ix] >> 4)
10924 #define HEXTRACT_OUTPUT_LO(ix) (*v++ = nvp[ix] & 0xF)
10925 #define HEXTRACT_OUTPUT(ix) \
10927 HEXTRACT_OUTPUT_HI(ix); HEXTRACT_OUTPUT_LO(ix); \
10929 #define HEXTRACT_COUNT(ix, c) \
10931 v += c; if (ix < ixmin) ixmin = ix; else if (ix > ixmax) ixmax = ix; \
10933 #define HEXTRACT_BYTE(ix) \
10935 if (vend) HEXTRACT_OUTPUT(ix); else HEXTRACT_COUNT(ix, 2); \
10937 #define HEXTRACT_LO_NYBBLE(ix) \
10939 if (vend) HEXTRACT_OUTPUT_LO(ix); else HEXTRACT_COUNT(ix, 1); \
10941 /* HEXTRACT_TOP_NYBBLE is just convenience disguise,
10942 * to make it look less odd when the top bits of a NV
10943 * are extracted using HEXTRACT_LO_NYBBLE: the highest
10944 * order bits can be in the "low nybble" of a byte. */
10945 #define HEXTRACT_TOP_NYBBLE(ix) HEXTRACT_LO_NYBBLE(ix)
10946 #define HEXTRACT_BYTES_LE(a, b) \
10947 for (ix = a; ix >= b; ix--) { HEXTRACT_BYTE(ix); }
10948 #define HEXTRACT_BYTES_BE(a, b) \
10949 for (ix = a; ix <= b; ix++) { HEXTRACT_BYTE(ix); }
10950 #define HEXTRACT_IMPLICIT_BIT(nv) \
10952 if (vend) *v++ = ((nv) == 0.0) ? 0 : 1; else v++; \
10955 /* Most formats do. Those which don't should undef this. */
10956 #define HEXTRACT_HAS_IMPLICIT_BIT
10957 /* Many formats do. Those which don't should undef this. */
10958 #define HEXTRACT_HAS_TOP_NYBBLE
10960 /* HEXTRACTSIZE is the maximum number of xdigits. */
10961 #if defined(USE_LONG_DOUBLE) && defined(LONGDOUBLE_DOUBLEDOUBLE)
10962 # define HEXTRACTSIZE (2+DOUBLEDOUBLE_MAXBITS/4)
10964 # define HEXTRACTSIZE 2 * NVSIZE
10967 const U8* vmaxend = vhex + HEXTRACTSIZE;
10968 PERL_UNUSED_VAR(ix); /* might happen */
10969 (void)Perl_frexp(PERL_ABS(nv), exponent);
10970 if (vend && (vend <= vhex || vend > vmaxend)) {
10971 /* diag_listed_as: Hexadecimal float: internal error (%s) */
10972 Perl_croak(aTHX_ "Hexadecimal float: internal error (entry)");
10975 /* First check if using long doubles. */
10976 #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE)
10977 # if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN
10978 /* Used in e.g. VMS and HP-UX IA-64, e.g. -0.1L:
10979 * 9a 99 99 99 99 99 99 99 99 99 99 99 99 99 fb 3f */
10980 /* The bytes 13..0 are the mantissa/fraction,
10981 * the 15,14 are the sign+exponent. */
10982 const U8* nvp = (const U8*)(&nv);
10983 HEXTRACT_IMPLICIT_BIT(nv);
10984 # undef HEXTRACT_HAS_TOP_NYBBLE
10985 HEXTRACT_BYTES_LE(13, 0);
10986 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_BIG_ENDIAN
10987 /* Used in e.g. Solaris Sparc and HP-UX PA-RISC, e.g. -0.1L:
10988 * bf fb 99 99 99 99 99 99 99 99 99 99 99 99 99 9a */
10989 /* The bytes 2..15 are the mantissa/fraction,
10990 * the 0,1 are the sign+exponent. */
10991 const U8* nvp = (const U8*)(&nv);
10992 HEXTRACT_IMPLICIT_BIT(nv);
10993 # undef HEXTRACT_HAS_TOP_NYBBLE
10994 HEXTRACT_BYTES_BE(2, 15);
10995 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN
10996 /* x86 80-bit "extended precision", 64 bits of mantissa / fraction /
10997 * significand, 15 bits of exponent, 1 bit of sign. NVSIZE can
10998 * be either 12 (ILP32, Solaris x86) or 16 (LP64, Linux and OS X),
10999 * meaning that 2 or 6 bytes are empty padding. */
11000 /* The bytes 7..0 are the mantissa/fraction */
11001 const U8* nvp = (const U8*)(&nv);
11002 # undef HEXTRACT_HAS_IMPLICIT_BIT
11003 # undef HEXTRACT_HAS_TOP_NYBBLE
11004 HEXTRACT_BYTES_LE(7, 0);
11005 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN
11006 /* Does this format ever happen? (Wikipedia says the Motorola
11007 * 6888x math coprocessors used format _like_ this but padded
11008 * to 96 bits with 16 unused bits between the exponent and the
11010 const U8* nvp = (const U8*)(&nv);
11011 # undef HEXTRACT_HAS_IMPLICIT_BIT
11012 # undef HEXTRACT_HAS_TOP_NYBBLE
11013 HEXTRACT_BYTES_BE(0, 7);
11015 # define HEXTRACT_FALLBACK
11016 /* Double-double format: two doubles next to each other.
11017 * The first double is the high-order one, exactly like
11018 * it would be for a "lone" double. The second double
11019 * is shifted down using the exponent so that that there
11020 * are no common bits. The tricky part is that the value
11021 * of the double-double is the SUM of the two doubles and
11022 * the second one can be also NEGATIVE.
11024 * Because of this tricky construction the bytewise extraction we
11025 * use for the other long double formats doesn't work, we must
11026 * extract the values bit by bit.
11028 * The little-endian double-double is used .. somewhere?
11030 * The big endian double-double is used in e.g. PPC/Power (AIX)
11033 * The mantissa bits are in two separate stretches, e.g. for -0.1L:
11034 * 9a 99 99 99 99 99 59 bc 9a 99 99 99 99 99 b9 3f (LE)
11035 * 3f b9 99 99 99 99 99 9a bc 59 99 99 99 99 99 9a (BE)
11038 #else /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) */
11039 /* Using normal doubles, not long doubles.
11041 * We generate 4-bit xdigits (nybble/nibble) instead of 8-bit
11042 * bytes, since we might need to handle printf precision, and
11043 * also need to insert the radix. */
11045 # ifdef HEXTRACT_LITTLE_ENDIAN
11046 /* 0 1 2 3 4 5 6 7 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11047 const U8* nvp = (const U8*)(&nv);
11048 HEXTRACT_IMPLICIT_BIT(nv);
11049 HEXTRACT_TOP_NYBBLE(6);
11050 HEXTRACT_BYTES_LE(5, 0);
11051 # elif defined(HEXTRACT_BIG_ENDIAN)
11052 /* 7 6 5 4 3 2 1 0 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11053 const U8* nvp = (const U8*)(&nv);
11054 HEXTRACT_IMPLICIT_BIT(nv);
11055 HEXTRACT_TOP_NYBBLE(1);
11056 HEXTRACT_BYTES_BE(2, 7);
11057 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_LE_BE
11058 /* 4 5 6 7 0 1 2 3 (MSB = 7, LSB = 0, 6:7 = nybble:exponent:sign) */
11059 const U8* nvp = (const U8*)(&nv);
11060 HEXTRACT_IMPLICIT_BIT(nv);
11061 HEXTRACT_TOP_NYBBLE(2); /* 6 */
11062 HEXTRACT_BYTE(1); /* 5 */
11063 HEXTRACT_BYTE(0); /* 4 */
11064 HEXTRACT_BYTE(7); /* 3 */
11065 HEXTRACT_BYTE(6); /* 2 */
11066 HEXTRACT_BYTE(5); /* 1 */
11067 HEXTRACT_BYTE(4); /* 0 */
11068 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_BE_LE
11069 /* 3 2 1 0 7 6 5 4 (MSB = 7, LSB = 0, 7:6 = sign:exponent:nybble) */
11070 const U8* nvp = (const U8*)(&nv);
11071 HEXTRACT_IMPLICIT_BIT(nv);
11072 HEXTRACT_TOP_NYBBLE(5); /* 6 */
11073 HEXTRACT_BYTE(6); /* 5 */
11074 HEXTRACT_BYTE(7); /* 4 */
11075 HEXTRACT_BYTE(0); /* 3 */
11076 HEXTRACT_BYTE(1); /* 2 */
11077 HEXTRACT_BYTE(2); /* 1 */
11078 HEXTRACT_BYTE(3); /* 0 */
11080 # define HEXTRACT_FALLBACK
11083 # define HEXTRACT_FALLBACK
11085 #endif /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) #else */
11086 # ifdef HEXTRACT_FALLBACK
11087 # undef HEXTRACT_HAS_TOP_NYBBLE /* Meaningless, but consistent. */
11088 /* The fallback is used for the double-double format, and
11089 * for unknown long double formats, and for unknown double
11090 * formats, or in general unknown NV formats. */
11091 if (nv == (NV)0.0) {
11099 NV d = nv < 0 ? -nv : nv;
11101 U8 ha = 0x0; /* hexvalue accumulator */
11102 U8 hd = 0x8; /* hexvalue digit */
11104 /* Shift d and e (and update exponent) so that e <= d < 2*e,
11105 * this is essentially manual frexp(). Multiplying by 0.5 and
11106 * doubling should be lossless in binary floating point. */
11116 while (d >= e + e) {
11120 /* Now e <= d < 2*e */
11122 /* First extract the leading hexdigit (the implicit bit). */
11138 /* Then extract the remaining hexdigits. */
11139 while (d > (NV)0.0) {
11145 /* Output or count in groups of four bits,
11146 * that is, when the hexdigit is down to one. */
11151 /* Reset the hexvalue. */
11160 /* Flush possible pending hexvalue. */
11170 /* Croak for various reasons: if the output pointer escaped the
11171 * output buffer, if the extraction index escaped the extraction
11172 * buffer, or if the ending output pointer didn't match the
11173 * previously computed value. */
11174 if (v <= vhex || v - vhex >= VHEX_SIZE ||
11175 /* For double-double the ixmin and ixmax stay at zero,
11176 * which is convenient since the HEXTRACTSIZE is tricky
11177 * for double-double. */
11178 ixmin < 0 || ixmax >= NVSIZE ||
11179 (vend && v != vend)) {
11180 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11181 Perl_croak(aTHX_ "Hexadecimal float: internal error (overflow)");
11186 /* Helper for sv_vcatpvfn_flags(). */
11187 #define FETCH_VCATPVFN_ARGUMENT(var, in_range, expr) \
11192 (var) = &PL_sv_no; /* [perl #71000] */ \
11193 arg_missing = TRUE; \
11198 Perl_sv_vcatpvfn_flags(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11199 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted,
11204 const char *patend;
11207 static const char nullstr[] = "(null)";
11209 bool has_utf8 = DO_UTF8(sv); /* has the result utf8? */
11210 const bool pat_utf8 = has_utf8; /* the pattern is in utf8? */
11212 /* Times 4: a decimal digit takes more than 3 binary digits.
11213 * NV_DIG: mantissa takes than many decimal digits.
11214 * Plus 32: Playing safe. */
11215 char ebuf[IV_DIG * 4 + NV_DIG + 32];
11216 bool no_redundant_warning = FALSE; /* did we use any explicit format parameter index? */
11217 bool hexfp = FALSE; /* hexadecimal floating point? */
11219 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
11221 PERL_ARGS_ASSERT_SV_VCATPVFN_FLAGS;
11222 PERL_UNUSED_ARG(maybe_tainted);
11224 if (flags & SV_GMAGIC)
11227 /* no matter what, this is a string now */
11228 (void)SvPV_force_nomg(sv, origlen);
11230 /* special-case "", "%s", and "%-p" (SVf - see below) */
11232 if (svmax && ckWARN(WARN_REDUNDANT))
11233 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11234 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11237 if (patlen == 2 && pat[0] == '%' && pat[1] == 's') {
11238 if (svmax > 1 && ckWARN(WARN_REDUNDANT))
11239 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11240 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11243 const char * const s = va_arg(*args, char*);
11244 sv_catpv_nomg(sv, s ? s : nullstr);
11246 else if (svix < svmax) {
11247 /* we want get magic on the source but not the target. sv_catsv can't do that, though */
11248 SvGETMAGIC(*svargs);
11249 sv_catsv_nomg(sv, *svargs);
11252 S_warn_vcatpvfn_missing_argument(aTHX);
11255 if (args && patlen == 3 && pat[0] == '%' &&
11256 pat[1] == '-' && pat[2] == 'p') {
11257 if (svmax > 1 && ckWARN(WARN_REDUNDANT))
11258 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11259 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11260 argsv = MUTABLE_SV(va_arg(*args, void*));
11261 sv_catsv_nomg(sv, argsv);
11265 #if !defined(USE_LONG_DOUBLE) && !defined(USE_QUADMATH)
11266 /* special-case "%.<number>[gf]" */
11267 if ( !args && patlen <= 5 && pat[0] == '%' && pat[1] == '.'
11268 && (pat[patlen-1] == 'g' || pat[patlen-1] == 'f') ) {
11269 unsigned digits = 0;
11273 while (*pp >= '0' && *pp <= '9')
11274 digits = 10 * digits + (*pp++ - '0');
11276 /* XXX: Why do this `svix < svmax` test? Couldn't we just
11277 format the first argument and WARN_REDUNDANT if svmax > 1?
11278 Munged by Nicholas Clark in v5.13.0-209-g95ea86d */
11279 if (pp - pat == (int)patlen - 1 && svix < svmax) {
11280 const NV nv = SvNV(*svargs);
11281 if (LIKELY(!Perl_isinfnan(nv))) {
11283 /* Add check for digits != 0 because it seems that some
11284 gconverts are buggy in this case, and we don't yet have
11285 a Configure test for this. */
11286 if (digits && digits < sizeof(ebuf) - NV_DIG - 10) {
11287 /* 0, point, slack */
11288 STORE_LC_NUMERIC_SET_TO_NEEDED();
11289 SNPRINTF_G(nv, ebuf, size, digits);
11290 sv_catpv_nomg(sv, ebuf);
11291 if (*ebuf) /* May return an empty string for digits==0 */
11294 } else if (!digits) {
11297 if ((p = F0convert(nv, ebuf + sizeof ebuf, &l))) {
11298 sv_catpvn_nomg(sv, p, l);
11305 #endif /* !USE_LONG_DOUBLE */
11307 if (!args && svix < svmax && DO_UTF8(*svargs))
11310 patend = (char*)pat + patlen;
11311 for (p = (char*)pat; p < patend; p = q) {
11314 bool vectorize = FALSE;
11315 bool vectorarg = FALSE;
11316 bool vec_utf8 = FALSE;
11322 bool has_precis = FALSE;
11324 const I32 osvix = svix;
11325 bool is_utf8 = FALSE; /* is this item utf8? */
11326 bool used_explicit_ix = FALSE;
11327 bool arg_missing = FALSE;
11328 #ifdef HAS_LDBL_SPRINTF_BUG
11329 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
11330 with sfio - Allen <allens@cpan.org> */
11331 bool fix_ldbl_sprintf_bug = FALSE;
11335 U8 utf8buf[UTF8_MAXBYTES+1];
11336 STRLEN esignlen = 0;
11338 const char *eptr = NULL;
11339 const char *fmtstart;
11342 const U8 *vecstr = NULL;
11349 /* We need a long double target in case HAS_LONG_DOUBLE,
11350 * even without USE_LONG_DOUBLE, so that we can printf with
11351 * long double formats, even without NV being long double.
11352 * But we call the target 'fv' instead of 'nv', since most of
11353 * the time it is not (most compilers these days recognize
11354 * "long double", even if only as a synonym for "double").
11356 #if defined(HAS_LONG_DOUBLE) && LONG_DOUBLESIZE > DOUBLESIZE && \
11357 defined(PERL_PRIgldbl) && !defined(USE_QUADMATH)
11359 # ifdef Perl_isfinitel
11360 # define FV_ISFINITE(x) Perl_isfinitel(x)
11362 # define FV_GF PERL_PRIgldbl
11363 # if defined(__VMS) && defined(__ia64) && defined(__IEEE_FLOAT)
11364 /* Work around breakage in OTS$CVT_FLOAT_T_X */
11365 # define NV_TO_FV(nv,fv) STMT_START { \
11367 fv = Perl_isnan(_dv) ? LDBL_QNAN : _dv; \
11370 # define NV_TO_FV(nv,fv) (fv)=(nv)
11374 # define FV_GF NVgf
11375 # define NV_TO_FV(nv,fv) (fv)=(nv)
11377 #ifndef FV_ISFINITE
11378 # define FV_ISFINITE(x) Perl_isfinite((NV)(x))
11384 const char *dotstr = ".";
11385 STRLEN dotstrlen = 1;
11386 I32 efix = 0; /* explicit format parameter index */
11387 I32 ewix = 0; /* explicit width index */
11388 I32 epix = 0; /* explicit precision index */
11389 I32 evix = 0; /* explicit vector index */
11390 bool asterisk = FALSE;
11391 bool infnan = FALSE;
11393 /* echo everything up to the next format specification */
11394 for (q = p; q < patend && *q != '%'; ++q) ;
11396 if (has_utf8 && !pat_utf8)
11397 sv_catpvn_nomg_utf8_upgrade(sv, p, q - p, nsv);
11399 sv_catpvn_nomg(sv, p, q - p);
11408 We allow format specification elements in this order:
11409 \d+\$ explicit format parameter index
11411 v|\*(\d+\$)?v vector with optional (optionally specified) arg
11412 0 flag (as above): repeated to allow "v02"
11413 \d+|\*(\d+\$)? width using optional (optionally specified) arg
11414 \.(\d*|\*(\d+\$)?) precision using optional (optionally specified) arg
11416 [%bcdefginopsuxDFOUX] format (mandatory)
11421 As of perl5.9.3, printf format checking is on by default.
11422 Internally, perl uses %p formats to provide an escape to
11423 some extended formatting. This block deals with those
11424 extensions: if it does not match, (char*)q is reset and
11425 the normal format processing code is used.
11427 Currently defined extensions are:
11428 %p include pointer address (standard)
11429 %-p (SVf) include an SV (previously %_)
11430 %-<num>p include an SV with precision <num>
11432 %3p include a HEK with precision of 256
11433 %4p char* preceded by utf8 flag and length
11434 %<num>p (where num is 1 or > 4) reserved for future
11437 Robin Barker 2005-07-14 (but modified since)
11439 %1p (VDf) removed. RMB 2007-10-19
11446 else if (strnEQ(q, UTF8f, sizeof(UTF8f)-1)) { /* UTF8f */
11447 /* The argument has already gone through cBOOL, so the cast
11449 is_utf8 = (bool)va_arg(*args, int);
11450 elen = va_arg(*args, UV);
11451 /* if utf8 length is larger than 0x7ffff..., then it might
11452 * have been a signed value that wrapped */
11453 if (elen > ((~(STRLEN)0) >> 1)) {
11454 assert(0); /* in DEBUGGING build we want to crash */
11455 elen= 0; /* otherwise we want to treat this as an empty string */
11457 eptr = va_arg(*args, char *);
11458 q += sizeof(UTF8f)-1;
11461 n = expect_number(&q);
11463 if (sv) { /* SVf */
11468 argsv = MUTABLE_SV(va_arg(*args, void*));
11469 eptr = SvPV_const(argsv, elen);
11470 if (DO_UTF8(argsv))
11474 else if (n==2 || n==3) { /* HEKf */
11475 HEK * const hek = va_arg(*args, HEK *);
11476 eptr = HEK_KEY(hek);
11477 elen = HEK_LEN(hek);
11478 if (HEK_UTF8(hek)) is_utf8 = TRUE;
11479 if (n==3) precis = 256, has_precis = TRUE;
11483 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
11484 "internal %%<num>p might conflict with future printf extensions");
11490 if ( (width = expect_number(&q)) ) {
11493 Perl_croak_nocontext(
11494 "Cannot yet reorder sv_catpvfn() arguments from va_list");
11497 used_explicit_ix = TRUE;
11509 if (plus == '+' && *q == ' ') /* '+' over ' ' */
11538 if ( (ewix = expect_number(&q)) ) {
11541 Perl_croak_nocontext(
11542 "Cannot yet reorder sv_catpvfn() arguments from va_list");
11543 used_explicit_ix = TRUE;
11553 if ((vectorarg = asterisk)) {
11566 width = expect_number(&q);
11569 if (vectorize && vectorarg) {
11570 /* vectorizing, but not with the default "." */
11572 vecsv = va_arg(*args, SV*);
11574 FETCH_VCATPVFN_ARGUMENT(
11575 vecsv, evix > 0 && evix <= svmax, svargs[evix-1]);
11577 FETCH_VCATPVFN_ARGUMENT(
11578 vecsv, svix < svmax, svargs[svix++]);
11580 dotstr = SvPV_const(vecsv, dotstrlen);
11581 /* Keep the DO_UTF8 test *after* the SvPV call, else things go
11582 bad with tied or overloaded values that return UTF8. */
11583 if (DO_UTF8(vecsv))
11585 else if (has_utf8) {
11586 vecsv = sv_mortalcopy(vecsv);
11587 sv_utf8_upgrade(vecsv);
11588 dotstr = SvPV_const(vecsv, dotstrlen);
11595 i = va_arg(*args, int);
11597 i = (ewix ? ewix <= svmax : svix < svmax) ?
11598 SvIVx(svargs[ewix ? ewix-1 : svix++]) : 0;
11600 width = (i < 0) ? -i : i;
11610 if ( (epix = expect_number(&q)) ) {
11613 Perl_croak_nocontext(
11614 "Cannot yet reorder sv_catpvfn() arguments from va_list");
11615 used_explicit_ix = TRUE;
11620 i = va_arg(*args, int);
11624 FETCH_VCATPVFN_ARGUMENT(
11625 precsv, epix > 0 && epix <= svmax, svargs[epix-1]);
11627 FETCH_VCATPVFN_ARGUMENT(
11628 precsv, svix < svmax, svargs[svix++]);
11629 i = precsv == &PL_sv_no ? 0 : SvIVx(precsv);
11632 has_precis = !(i < 0);
11636 while (isDIGIT(*q))
11637 precis = precis * 10 + (*q++ - '0');
11646 else if (efix ? (efix > 0 && efix <= svmax) : svix < svmax) {
11647 vecsv = svargs[efix ? efix-1 : svix++];
11648 vecstr = (U8*)SvPV_const(vecsv,veclen);
11649 vec_utf8 = DO_UTF8(vecsv);
11651 /* if this is a version object, we need to convert
11652 * back into v-string notation and then let the
11653 * vectorize happen normally
11655 if (sv_isobject(vecsv) && sv_derived_from(vecsv, "version")) {
11656 if ( hv_exists(MUTABLE_HV(SvRV(vecsv)), "alpha", 5 ) ) {
11657 Perl_ck_warner_d(aTHX_ packWARN(WARN_PRINTF),
11658 "vector argument not supported with alpha versions");
11661 vecsv = sv_newmortal();
11662 scan_vstring((char *)vecstr, (char *)vecstr + veclen,
11664 vecstr = (U8*)SvPV_const(vecsv, veclen);
11665 vec_utf8 = DO_UTF8(vecsv);
11679 case 'I': /* Ix, I32x, and I64x */
11680 # ifdef USE_64_BIT_INT
11681 if (q[1] == '6' && q[2] == '4') {
11687 if (q[1] == '3' && q[2] == '2') {
11691 # ifdef USE_64_BIT_INT
11697 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
11698 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
11701 # ifdef USE_QUADMATH
11714 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
11715 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
11716 if (*q == 'l') { /* lld, llf */
11725 if (*++q == 'h') { /* hhd, hhu */
11754 if (!vectorize && !args) {
11756 const I32 i = efix-1;
11757 FETCH_VCATPVFN_ARGUMENT(argsv, i >= 0 && i < svmax, svargs[i]);
11759 FETCH_VCATPVFN_ARGUMENT(argsv, svix >= 0 && svix < svmax,
11764 if (argsv && strchr("BbcDdiOopuUXx",*q)) {
11765 /* XXX va_arg(*args) case? need peek, use va_copy? */
11767 if (UNLIKELY(SvAMAGIC(argsv)))
11768 argsv = sv_2num(argsv);
11769 infnan = UNLIKELY(isinfnansv(argsv));
11772 switch (c = *q++) {
11780 Perl_croak(aTHX_ "Cannot printf %"NVgf" with '%c'",
11781 /* no va_arg() case */
11782 SvNV_nomg(argsv), (int)c);
11783 uv = (args) ? va_arg(*args, int) : SvIV_nomg(argsv);
11785 (!UVCHR_IS_INVARIANT(uv) && SvUTF8(sv)))
11787 eptr = (char*)utf8buf;
11788 elen = uvchr_to_utf8((U8*)eptr, uv) - utf8buf;
11802 eptr = va_arg(*args, char*);
11804 elen = strlen(eptr);
11806 eptr = (char *)nullstr;
11807 elen = sizeof nullstr - 1;
11811 eptr = SvPV_const(argsv, elen);
11812 if (DO_UTF8(argsv)) {
11813 STRLEN old_precis = precis;
11814 if (has_precis && precis < elen) {
11815 STRLEN ulen = sv_or_pv_len_utf8(argsv, eptr, elen);
11816 STRLEN p = precis > ulen ? ulen : precis;
11817 precis = sv_or_pv_pos_u2b(argsv, eptr, p, 0);
11818 /* sticks at end */
11820 if (width) { /* fudge width (can't fudge elen) */
11821 if (has_precis && precis < elen)
11822 width += precis - old_precis;
11825 elen - sv_or_pv_len_utf8(argsv,eptr,elen);
11832 if (has_precis && precis < elen)
11840 goto floating_point;
11842 if (alt || vectorize)
11844 uv = PTR2UV(args ? va_arg(*args, void*) : argsv);
11858 goto floating_point;
11863 goto donevalidconversion;
11865 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
11874 esignbuf[esignlen++] = plus;
11878 case 'c': iv = (char)va_arg(*args, int); break;
11879 case 'h': iv = (short)va_arg(*args, int); break;
11880 case 'l': iv = va_arg(*args, long); break;
11881 case 'V': iv = va_arg(*args, IV); break;
11882 case 'z': iv = va_arg(*args, SSize_t); break;
11883 #ifdef HAS_PTRDIFF_T
11884 case 't': iv = va_arg(*args, ptrdiff_t); break;
11886 default: iv = va_arg(*args, int); break;
11888 case 'j': iv = va_arg(*args, intmax_t); break;
11892 iv = va_arg(*args, Quad_t); break;
11899 IV tiv = SvIV_nomg(argsv); /* work around GCC bug #13488 */
11901 case 'c': iv = (char)tiv; break;
11902 case 'h': iv = (short)tiv; break;
11903 case 'l': iv = (long)tiv; break;
11905 default: iv = tiv; break;
11908 iv = (Quad_t)tiv; break;
11914 if ( !vectorize ) /* we already set uv above */
11919 esignbuf[esignlen++] = plus;
11922 uv = (iv == IV_MIN) ? (UV)iv : (UV)(-iv);
11923 esignbuf[esignlen++] = '-';
11962 goto floating_point;
11968 goto donevalidconversion;
11970 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
11981 case 'c': uv = (unsigned char)va_arg(*args, unsigned); break;
11982 case 'h': uv = (unsigned short)va_arg(*args, unsigned); break;
11983 case 'l': uv = va_arg(*args, unsigned long); break;
11984 case 'V': uv = va_arg(*args, UV); break;
11985 case 'z': uv = va_arg(*args, Size_t); break;
11986 #ifdef HAS_PTRDIFF_T
11987 case 't': uv = va_arg(*args, ptrdiff_t); break; /* will sign extend, but there is no uptrdiff_t, so oh well */
11990 case 'j': uv = va_arg(*args, uintmax_t); break;
11992 default: uv = va_arg(*args, unsigned); break;
11995 uv = va_arg(*args, Uquad_t); break;
12002 UV tuv = SvUV_nomg(argsv); /* work around GCC bug #13488 */
12004 case 'c': uv = (unsigned char)tuv; break;
12005 case 'h': uv = (unsigned short)tuv; break;
12006 case 'l': uv = (unsigned long)tuv; break;
12008 default: uv = tuv; break;
12011 uv = (Uquad_t)tuv; break;
12020 char *ptr = ebuf + sizeof ebuf;
12021 bool tempalt = uv ? alt : FALSE; /* Vectors can't change alt */
12027 p = (char *)((c == 'X') ? PL_hexdigit + 16 : PL_hexdigit);
12031 } while (uv >>= 4);
12033 esignbuf[esignlen++] = '0';
12034 esignbuf[esignlen++] = c; /* 'x' or 'X' */
12040 *--ptr = '0' + dig;
12041 } while (uv >>= 3);
12042 if (alt && *ptr != '0')
12048 *--ptr = '0' + dig;
12049 } while (uv >>= 1);
12051 esignbuf[esignlen++] = '0';
12052 esignbuf[esignlen++] = c;
12055 default: /* it had better be ten or less */
12058 *--ptr = '0' + dig;
12059 } while (uv /= base);
12062 elen = (ebuf + sizeof ebuf) - ptr;
12066 zeros = precis - elen;
12067 else if (precis == 0 && elen == 1 && *eptr == '0'
12068 && !(base == 8 && alt)) /* "%#.0o" prints "0" */
12071 /* a precision nullifies the 0 flag. */
12078 /* FLOATING POINT */
12083 c = 'f'; /* maybe %F isn't supported here */
12085 case 'e': case 'E':
12087 case 'g': case 'G':
12088 case 'a': case 'A':
12092 /* This is evil, but floating point is even more evil */
12094 /* for SV-style calling, we can only get NV
12095 for C-style calling, we assume %f is double;
12096 for simplicity we allow any of %Lf, %llf, %qf for long double
12100 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12104 /* [perl #20339] - we should accept and ignore %lf rather than die */
12108 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12109 intsize = args ? 0 : 'q';
12113 #if defined(HAS_LONG_DOUBLE)
12126 /* Now we need (long double) if intsize == 'q', else (double). */
12128 /* Note: do not pull NVs off the va_list with va_arg()
12129 * (pull doubles instead) because if you have a build
12130 * with long doubles, you would always be pulling long
12131 * doubles, which would badly break anyone using only
12132 * doubles (i.e. the majority of builds). In other
12133 * words, you cannot mix doubles and long doubles.
12134 * The only case where you can pull off long doubles
12135 * is when the format specifier explicitly asks so with
12137 #ifdef USE_QUADMATH
12138 fv = intsize == 'q' ?
12139 va_arg(*args, NV) : va_arg(*args, double);
12141 #elif LONG_DOUBLESIZE > DOUBLESIZE
12142 if (intsize == 'q') {
12143 fv = va_arg(*args, long double);
12146 nv = va_arg(*args, double);
12150 nv = va_arg(*args, double);
12156 if (!infnan) SvGETMAGIC(argsv);
12157 nv = SvNV_nomg(argsv);
12162 /* frexp() (or frexpl) has some unspecified behaviour for
12163 * nan/inf/-inf, so let's avoid calling that on non-finites. */
12164 if (isALPHA_FOLD_NE(c, 'e') && FV_ISFINITE(fv)) {
12166 (void)Perl_frexp((NV)fv, &i);
12167 if (i == PERL_INT_MIN)
12168 Perl_die(aTHX_ "panic: frexp: %"FV_GF, fv);
12169 /* Do not set hexfp earlier since we want to printf
12170 * Inf/NaN for Inf/NaN, not their hexfp. */
12171 hexfp = isALPHA_FOLD_EQ(c, 'a');
12172 if (UNLIKELY(hexfp)) {
12173 /* This seriously overshoots in most cases, but
12174 * better the undershooting. Firstly, all bytes
12175 * of the NV are not mantissa, some of them are
12176 * exponent. Secondly, for the reasonably common
12177 * long doubles case, the "80-bit extended", two
12178 * or six bytes of the NV are unused. */
12180 (fv < 0) ? 1 : 0 + /* possible unary minus */
12182 1 + /* the very unlikely carry */
12185 2 * NVSIZE + /* 2 hexdigits for each byte */
12187 6 + /* exponent: sign, plus up to 16383 (quad fp) */
12189 #ifdef LONGDOUBLE_DOUBLEDOUBLE
12190 /* However, for the "double double", we need more.
12191 * Since each double has their own exponent, the
12192 * doubles may float (haha) rather far from each
12193 * other, and the number of required bits is much
12194 * larger, up to total of DOUBLEDOUBLE_MAXBITS bits.
12195 * See the definition of DOUBLEDOUBLE_MAXBITS.
12197 * Need 2 hexdigits for each byte. */
12198 need += (DOUBLEDOUBLE_MAXBITS/8 + 1) * 2;
12199 /* the size for the exponent already added */
12201 #ifdef USE_LOCALE_NUMERIC
12202 STORE_LC_NUMERIC_SET_TO_NEEDED();
12203 if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC))
12204 need += SvLEN(PL_numeric_radix_sv);
12205 RESTORE_LC_NUMERIC();
12209 need = BIT_DIGITS(i);
12210 } /* if i < 0, the number of digits is hard to predict. */
12212 need += has_precis ? precis : 6; /* known default */
12217 #ifdef HAS_LDBL_SPRINTF_BUG
12218 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
12219 with sfio - Allen <allens@cpan.org> */
12222 # define MY_DBL_MAX DBL_MAX
12223 # else /* XXX guessing! HUGE_VAL may be defined as infinity, so not using */
12224 # if DOUBLESIZE >= 8
12225 # define MY_DBL_MAX 1.7976931348623157E+308L
12227 # define MY_DBL_MAX 3.40282347E+38L
12231 # ifdef HAS_LDBL_SPRINTF_BUG_LESS1 /* only between -1L & 1L - Allen */
12232 # define MY_DBL_MAX_BUG 1L
12234 # define MY_DBL_MAX_BUG MY_DBL_MAX
12238 # define MY_DBL_MIN DBL_MIN
12239 # else /* XXX guessing! -Allen */
12240 # if DOUBLESIZE >= 8
12241 # define MY_DBL_MIN 2.2250738585072014E-308L
12243 # define MY_DBL_MIN 1.17549435E-38L
12247 if ((intsize == 'q') && (c == 'f') &&
12248 ((fv < MY_DBL_MAX_BUG) && (fv > -MY_DBL_MAX_BUG)) &&
12249 (need < DBL_DIG)) {
12250 /* it's going to be short enough that
12251 * long double precision is not needed */
12253 if ((fv <= 0L) && (fv >= -0L))
12254 fix_ldbl_sprintf_bug = TRUE; /* 0 is 0 - easiest */
12256 /* would use Perl_fp_class as a double-check but not
12257 * functional on IRIX - see perl.h comments */
12259 if ((fv >= MY_DBL_MIN) || (fv <= -MY_DBL_MIN)) {
12260 /* It's within the range that a double can represent */
12261 #if defined(DBL_MAX) && !defined(DBL_MIN)
12262 if ((fv >= ((long double)1/DBL_MAX)) ||
12263 (fv <= (-(long double)1/DBL_MAX)))
12265 fix_ldbl_sprintf_bug = TRUE;
12268 if (fix_ldbl_sprintf_bug == TRUE) {
12278 # undef MY_DBL_MAX_BUG
12281 #endif /* HAS_LDBL_SPRINTF_BUG */
12283 need += 20; /* fudge factor */
12284 if (PL_efloatsize < need) {
12285 Safefree(PL_efloatbuf);
12286 PL_efloatsize = need + 20; /* more fudge */
12287 Newx(PL_efloatbuf, PL_efloatsize, char);
12288 PL_efloatbuf[0] = '\0';
12291 if ( !(width || left || plus || alt) && fill != '0'
12292 && has_precis && intsize != 'q' /* Shortcuts */
12293 && LIKELY(!Perl_isinfnan((NV)fv)) ) {
12294 /* See earlier comment about buggy Gconvert when digits,
12296 if ( c == 'g' && precis ) {
12297 STORE_LC_NUMERIC_SET_TO_NEEDED();
12298 SNPRINTF_G(fv, PL_efloatbuf, PL_efloatsize, precis);
12299 /* May return an empty string for digits==0 */
12300 if (*PL_efloatbuf) {
12301 elen = strlen(PL_efloatbuf);
12302 goto float_converted;
12304 } else if ( c == 'f' && !precis ) {
12305 if ((eptr = F0convert(nv, ebuf + sizeof ebuf, &elen)))
12310 if (UNLIKELY(hexfp)) {
12311 /* Hexadecimal floating point. */
12312 char* p = PL_efloatbuf;
12313 U8 vhex[VHEX_SIZE];
12314 U8* v = vhex; /* working pointer to vhex */
12315 U8* vend; /* pointer to one beyond last digit of vhex */
12316 U8* vfnz = NULL; /* first non-zero */
12317 const bool lower = (c == 'a');
12318 /* At output the values of vhex (up to vend) will
12319 * be mapped through the xdig to get the actual
12320 * human-readable xdigits. */
12321 const char* xdig = PL_hexdigit;
12322 int zerotail = 0; /* how many extra zeros to append */
12323 int exponent = 0; /* exponent of the floating point input */
12325 /* XXX: denormals, NaN, Inf.
12327 * For example with denormals, (assuming the vanilla
12328 * 64-bit double): the exponent is zero. 1xp-1074 is
12329 * the smallest denormal and the smallest double, it
12330 * should be output as 0x0.0000000000001p-1022 to
12331 * match its internal structure. */
12333 vend = S_hextract(aTHX_ nv, &exponent, vhex, NULL);
12334 S_hextract(aTHX_ nv, &exponent, vhex, vend);
12336 #if NVSIZE > DOUBLESIZE
12337 # ifdef HEXTRACT_HAS_IMPLICIT_BIT
12338 /* In this case there is an implicit bit,
12339 * and therefore the exponent is shifted shift by one. */
12342 /* In this case there is no implicit bit,
12343 * and the exponent is shifted by the first xdigit. */
12358 xdig += 16; /* Use uppercase hex. */
12361 /* Find the first non-zero xdigit. */
12362 for (v = vhex; v < vend; v++) {
12370 U8* vlnz = NULL; /* The last non-zero. */
12372 /* Find the last non-zero xdigit. */
12373 for (v = vend - 1; v >= vhex; v--) {
12380 #if NVSIZE == DOUBLESIZE
12386 if ((SSize_t)(precis + 1) < vend - vhex) {
12389 v = vhex + precis + 1;
12390 /* Round away from zero: if the tail
12391 * beyond the precis xdigits is equal to
12392 * or greater than 0x8000... */
12394 if (!round && *v == 0x8) {
12395 for (v++; v < vend; v++) {
12403 for (v = vhex + precis; v >= vhex; v--) {
12410 /* If the carry goes all the way to
12411 * the front, we need to output
12412 * a single '1'. This goes against
12413 * the "xdigit and then radix"
12414 * but since this is "cannot happen"
12415 * category, that is probably good. */
12420 /* The new effective "last non zero". */
12421 vlnz = vhex + precis;
12424 zerotail = precis - (vlnz - vhex);
12431 /* The radix is always output after the first
12432 * non-zero xdigit, or if alt. */
12433 if (vfnz < vlnz || alt) {
12434 #ifndef USE_LOCALE_NUMERIC
12437 STORE_LC_NUMERIC_SET_TO_NEEDED();
12438 if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC)) {
12440 const char* r = SvPV(PL_numeric_radix_sv, n);
12441 Copy(r, p, n, char);
12447 RESTORE_LC_NUMERIC();
12462 elen = p - PL_efloatbuf;
12463 elen += my_snprintf(p, PL_efloatsize - elen,
12464 "%c%+d", lower ? 'p' : 'P',
12467 if (elen < width) {
12469 /* Pad the back with spaces. */
12470 memset(PL_efloatbuf + elen, ' ', width - elen);
12472 else if (fill == '0') {
12473 /* Insert the zeros between the "0x" and
12474 * the digits, otherwise we end up with
12476 STRLEN nzero = width - elen;
12477 char* zerox = PL_efloatbuf + 2;
12478 Move(zerox, zerox + nzero, elen - 2, char);
12479 memset(zerox, fill, nzero);
12482 /* Move it to the right. */
12483 Move(PL_efloatbuf, PL_efloatbuf + width - elen,
12485 /* Pad the front with spaces. */
12486 memset(PL_efloatbuf, ' ', width - elen);
12492 elen = S_infnan_2pv(nv, PL_efloatbuf, PL_efloatsize, plus);
12494 /* Not affecting infnan output: precision, alt, fill. */
12495 if (elen < width) {
12497 /* Pack the back with spaces. */
12498 memset(PL_efloatbuf + elen, ' ', width - elen);
12500 /* Move it to the right. */
12501 Move(PL_efloatbuf, PL_efloatbuf + width - elen,
12503 /* Pad the front with spaces. */
12504 memset(PL_efloatbuf, ' ', width - elen);
12512 char *ptr = ebuf + sizeof ebuf;
12515 #if defined(USE_QUADMATH)
12516 if (intsize == 'q') {
12520 /* FIXME: what to do if HAS_LONG_DOUBLE but not PERL_PRIfldbl? */
12521 #elif defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
12522 /* Note that this is HAS_LONG_DOUBLE and PERL_PRIfldbl,
12523 * not USE_LONG_DOUBLE and NVff. In other words,
12524 * this needs to work without USE_LONG_DOUBLE. */
12525 if (intsize == 'q') {
12526 /* Copy the one or more characters in a long double
12527 * format before the 'base' ([efgEFG]) character to
12528 * the format string. */
12529 static char const ldblf[] = PERL_PRIfldbl;
12530 char const *p = ldblf + sizeof(ldblf) - 3;
12531 while (p >= ldblf) { *--ptr = *p--; }
12536 do { *--ptr = '0' + (base % 10); } while (base /= 10);
12541 do { *--ptr = '0' + (base % 10); } while (base /= 10);
12553 /* No taint. Otherwise we are in the strange situation
12554 * where printf() taints but print($float) doesn't.
12557 STORE_LC_NUMERIC_SET_TO_NEEDED();
12559 /* hopefully the above makes ptr a very constrained format
12560 * that is safe to use, even though it's not literal */
12561 GCC_DIAG_IGNORE(-Wformat-nonliteral);
12562 #ifdef USE_QUADMATH
12564 const char* qfmt = quadmath_format_single(ptr);
12566 Perl_croak_nocontext("panic: quadmath invalid format \"%s\"", ptr);
12567 elen = quadmath_snprintf(PL_efloatbuf, PL_efloatsize,
12569 if ((IV)elen == -1)
12570 Perl_croak_nocontext("panic: quadmath_snprintf failed, format \"%s|'", qfmt);
12574 #elif defined(HAS_LONG_DOUBLE)
12575 elen = ((intsize == 'q')
12576 ? my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv)
12577 : my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, (double)fv));
12579 elen = my_sprintf(PL_efloatbuf, ptr, fv);
12585 eptr = PL_efloatbuf;
12586 assert((IV)elen > 0); /* here zero elen is bad */
12588 #ifdef USE_LOCALE_NUMERIC
12589 /* If the decimal point character in the string is UTF-8, make the
12591 if (PL_numeric_radix_sv && SvUTF8(PL_numeric_radix_sv)
12592 && instr(eptr, SvPVX_const(PL_numeric_radix_sv)))
12605 i = SvCUR(sv) - origlen;
12608 case 'c': *(va_arg(*args, char*)) = i; break;
12609 case 'h': *(va_arg(*args, short*)) = i; break;
12610 default: *(va_arg(*args, int*)) = i; break;
12611 case 'l': *(va_arg(*args, long*)) = i; break;
12612 case 'V': *(va_arg(*args, IV*)) = i; break;
12613 case 'z': *(va_arg(*args, SSize_t*)) = i; break;
12614 #ifdef HAS_PTRDIFF_T
12615 case 't': *(va_arg(*args, ptrdiff_t*)) = i; break;
12618 case 'j': *(va_arg(*args, intmax_t*)) = i; break;
12622 *(va_arg(*args, Quad_t*)) = i; break;
12629 sv_setuv_mg(argsv, has_utf8 ? (UV)sv_len_utf8(sv) : (UV)i);
12630 goto donevalidconversion;
12637 && (PL_op->op_type == OP_PRTF || PL_op->op_type == OP_SPRINTF)
12638 && ckWARN(WARN_PRINTF))
12640 SV * const msg = sv_newmortal();
12641 Perl_sv_setpvf(aTHX_ msg, "Invalid conversion in %sprintf: ",
12642 (PL_op->op_type == OP_PRTF) ? "" : "s");
12643 if (fmtstart < patend) {
12644 const char * const fmtend = q < patend ? q : patend;
12646 sv_catpvs(msg, "\"%");
12647 for (f = fmtstart; f < fmtend; f++) {
12649 sv_catpvn_nomg(msg, f, 1);
12651 Perl_sv_catpvf(aTHX_ msg,
12652 "\\%03"UVof, (UV)*f & 0xFF);
12655 sv_catpvs(msg, "\"");
12657 sv_catpvs(msg, "end of string");
12659 Perl_warner(aTHX_ packWARN(WARN_PRINTF), "%"SVf, SVfARG(msg)); /* yes, this is reentrant */
12662 /* output mangled stuff ... */
12668 /* ... right here, because formatting flags should not apply */
12669 SvGROW(sv, SvCUR(sv) + elen + 1);
12671 Copy(eptr, p, elen, char);
12674 SvCUR_set(sv, p - SvPVX_const(sv));
12676 continue; /* not "break" */
12679 if (is_utf8 != has_utf8) {
12682 sv_utf8_upgrade(sv);
12685 const STRLEN old_elen = elen;
12686 SV * const nsv = newSVpvn_flags(eptr, elen, SVs_TEMP);
12687 sv_utf8_upgrade(nsv);
12688 eptr = SvPVX_const(nsv);
12691 if (width) { /* fudge width (can't fudge elen) */
12692 width += elen - old_elen;
12698 /* signed value that's wrapped? */
12699 assert(elen <= ((~(STRLEN)0) >> 1));
12700 have = esignlen + zeros + elen;
12702 croak_memory_wrap();
12704 need = (have > width ? have : width);
12707 if (need >= (((STRLEN)~0) - SvCUR(sv) - dotstrlen - 1))
12708 croak_memory_wrap();
12709 SvGROW(sv, SvCUR(sv) + need + dotstrlen + 1);
12711 if (esignlen && fill == '0') {
12713 for (i = 0; i < (int)esignlen; i++)
12714 *p++ = esignbuf[i];
12716 if (gap && !left) {
12717 memset(p, fill, gap);
12720 if (esignlen && fill != '0') {
12722 for (i = 0; i < (int)esignlen; i++)
12723 *p++ = esignbuf[i];
12727 for (i = zeros; i; i--)
12731 Copy(eptr, p, elen, char);
12735 memset(p, ' ', gap);
12740 Copy(dotstr, p, dotstrlen, char);
12744 vectorize = FALSE; /* done iterating over vecstr */
12751 SvCUR_set(sv, p - SvPVX_const(sv));
12757 donevalidconversion:
12758 if (used_explicit_ix)
12759 no_redundant_warning = TRUE;
12761 S_warn_vcatpvfn_missing_argument(aTHX);
12764 /* Now that we've consumed all our printf format arguments (svix)
12765 * do we have things left on the stack that we didn't use?
12767 if (!no_redundant_warning && svmax >= svix + 1 && ckWARN(WARN_REDUNDANT)) {
12768 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
12769 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
12774 RESTORE_LC_NUMERIC(); /* Done outside loop, so don't have to save/restore
12778 /* =========================================================================
12780 =head1 Cloning an interpreter
12784 All the macros and functions in this section are for the private use of
12785 the main function, perl_clone().
12787 The foo_dup() functions make an exact copy of an existing foo thingy.
12788 During the course of a cloning, a hash table is used to map old addresses
12789 to new addresses. The table is created and manipulated with the
12790 ptr_table_* functions.
12792 * =========================================================================*/
12795 #if defined(USE_ITHREADS)
12797 /* XXX Remove this so it doesn't have to go thru the macro and return for nothing */
12798 #ifndef GpREFCNT_inc
12799 # define GpREFCNT_inc(gp) ((gp) ? (++(gp)->gp_refcnt, (gp)) : (GP*)NULL)
12803 /* Certain cases in Perl_ss_dup have been merged, by relying on the fact
12804 that currently av_dup, gv_dup and hv_dup are the same as sv_dup.
12805 If this changes, please unmerge ss_dup.
12806 Likewise, sv_dup_inc_multiple() relies on this fact. */
12807 #define sv_dup_inc_NN(s,t) SvREFCNT_inc_NN(sv_dup_inc(s,t))
12808 #define av_dup(s,t) MUTABLE_AV(sv_dup((const SV *)s,t))
12809 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
12810 #define hv_dup(s,t) MUTABLE_HV(sv_dup((const SV *)s,t))
12811 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
12812 #define cv_dup(s,t) MUTABLE_CV(sv_dup((const SV *)s,t))
12813 #define cv_dup_inc(s,t) MUTABLE_CV(sv_dup_inc((const SV *)s,t))
12814 #define io_dup(s,t) MUTABLE_IO(sv_dup((const SV *)s,t))
12815 #define io_dup_inc(s,t) MUTABLE_IO(sv_dup_inc((const SV *)s,t))
12816 #define gv_dup(s,t) MUTABLE_GV(sv_dup((const SV *)s,t))
12817 #define gv_dup_inc(s,t) MUTABLE_GV(sv_dup_inc((const SV *)s,t))
12818 #define SAVEPV(p) ((p) ? savepv(p) : NULL)
12819 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
12821 /* clone a parser */
12824 Perl_parser_dup(pTHX_ const yy_parser *const proto, CLONE_PARAMS *const param)
12828 PERL_ARGS_ASSERT_PARSER_DUP;
12833 /* look for it in the table first */
12834 parser = (yy_parser *)ptr_table_fetch(PL_ptr_table, proto);
12838 /* create anew and remember what it is */
12839 Newxz(parser, 1, yy_parser);
12840 ptr_table_store(PL_ptr_table, proto, parser);
12842 /* XXX these not yet duped */
12843 parser->old_parser = NULL;
12844 parser->stack = NULL;
12846 parser->stack_size = 0;
12847 /* XXX parser->stack->state = 0; */
12849 /* XXX eventually, just Copy() most of the parser struct ? */
12851 parser->lex_brackets = proto->lex_brackets;
12852 parser->lex_casemods = proto->lex_casemods;
12853 parser->lex_brackstack = savepvn(proto->lex_brackstack,
12854 (proto->lex_brackets < 120 ? 120 : proto->lex_brackets));
12855 parser->lex_casestack = savepvn(proto->lex_casestack,
12856 (proto->lex_casemods < 12 ? 12 : proto->lex_casemods));
12857 parser->lex_defer = proto->lex_defer;
12858 parser->lex_dojoin = proto->lex_dojoin;
12859 parser->lex_formbrack = proto->lex_formbrack;
12860 parser->lex_inpat = proto->lex_inpat;
12861 parser->lex_inwhat = proto->lex_inwhat;
12862 parser->lex_op = proto->lex_op;
12863 parser->lex_repl = sv_dup_inc(proto->lex_repl, param);
12864 parser->lex_starts = proto->lex_starts;
12865 parser->lex_stuff = sv_dup_inc(proto->lex_stuff, param);
12866 parser->multi_close = proto->multi_close;
12867 parser->multi_open = proto->multi_open;
12868 parser->multi_start = proto->multi_start;
12869 parser->multi_end = proto->multi_end;
12870 parser->preambled = proto->preambled;
12871 parser->sublex_info = proto->sublex_info; /* XXX not quite right */
12872 parser->linestr = sv_dup_inc(proto->linestr, param);
12873 parser->expect = proto->expect;
12874 parser->copline = proto->copline;
12875 parser->last_lop_op = proto->last_lop_op;
12876 parser->lex_state = proto->lex_state;
12877 parser->rsfp = fp_dup(proto->rsfp, '<', param);
12878 /* rsfp_filters entries have fake IoDIRP() */
12879 parser->rsfp_filters= av_dup_inc(proto->rsfp_filters, param);
12880 parser->in_my = proto->in_my;
12881 parser->in_my_stash = hv_dup(proto->in_my_stash, param);
12882 parser->error_count = proto->error_count;
12885 parser->linestr = sv_dup_inc(proto->linestr, param);
12888 char * const ols = SvPVX(proto->linestr);
12889 char * const ls = SvPVX(parser->linestr);
12891 parser->bufptr = ls + (proto->bufptr >= ols ?
12892 proto->bufptr - ols : 0);
12893 parser->oldbufptr = ls + (proto->oldbufptr >= ols ?
12894 proto->oldbufptr - ols : 0);
12895 parser->oldoldbufptr= ls + (proto->oldoldbufptr >= ols ?
12896 proto->oldoldbufptr - ols : 0);
12897 parser->linestart = ls + (proto->linestart >= ols ?
12898 proto->linestart - ols : 0);
12899 parser->last_uni = ls + (proto->last_uni >= ols ?
12900 proto->last_uni - ols : 0);
12901 parser->last_lop = ls + (proto->last_lop >= ols ?
12902 proto->last_lop - ols : 0);
12904 parser->bufend = ls + SvCUR(parser->linestr);
12907 Copy(proto->tokenbuf, parser->tokenbuf, 256, char);
12910 Copy(proto->nextval, parser->nextval, 5, YYSTYPE);
12911 Copy(proto->nexttype, parser->nexttype, 5, I32);
12912 parser->nexttoke = proto->nexttoke;
12914 /* XXX should clone saved_curcop here, but we aren't passed
12915 * proto_perl; so do it in perl_clone_using instead */
12921 /* duplicate a file handle */
12924 Perl_fp_dup(pTHX_ PerlIO *const fp, const char type, CLONE_PARAMS *const param)
12928 PERL_ARGS_ASSERT_FP_DUP;
12929 PERL_UNUSED_ARG(type);
12932 return (PerlIO*)NULL;
12934 /* look for it in the table first */
12935 ret = (PerlIO*)ptr_table_fetch(PL_ptr_table, fp);
12939 /* create anew and remember what it is */
12940 #ifdef __amigaos4__
12941 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE|PERLIO_DUP_FD);
12943 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE);
12945 ptr_table_store(PL_ptr_table, fp, ret);
12949 /* duplicate a directory handle */
12952 Perl_dirp_dup(pTHX_ DIR *const dp, CLONE_PARAMS *const param)
12956 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
12958 const Direntry_t *dirent;
12959 char smallbuf[256];
12965 PERL_UNUSED_CONTEXT;
12966 PERL_ARGS_ASSERT_DIRP_DUP;
12971 /* look for it in the table first */
12972 ret = (DIR*)ptr_table_fetch(PL_ptr_table, dp);
12976 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
12978 PERL_UNUSED_ARG(param);
12982 /* open the current directory (so we can switch back) */
12983 if (!(pwd = PerlDir_open("."))) return (DIR *)NULL;
12985 /* chdir to our dir handle and open the present working directory */
12986 if (fchdir(my_dirfd(dp)) < 0 || !(ret = PerlDir_open("."))) {
12987 PerlDir_close(pwd);
12988 return (DIR *)NULL;
12990 /* Now we should have two dir handles pointing to the same dir. */
12992 /* Be nice to the calling code and chdir back to where we were. */
12993 /* XXX If this fails, then what? */
12994 PERL_UNUSED_RESULT(fchdir(my_dirfd(pwd)));
12996 /* We have no need of the pwd handle any more. */
12997 PerlDir_close(pwd);
13000 # define d_namlen(d) (d)->d_namlen
13002 # define d_namlen(d) strlen((d)->d_name)
13004 /* Iterate once through dp, to get the file name at the current posi-
13005 tion. Then step back. */
13006 pos = PerlDir_tell(dp);
13007 if ((dirent = PerlDir_read(dp))) {
13008 len = d_namlen(dirent);
13009 if (len <= sizeof smallbuf) name = smallbuf;
13010 else Newx(name, len, char);
13011 Move(dirent->d_name, name, len, char);
13013 PerlDir_seek(dp, pos);
13015 /* Iterate through the new dir handle, till we find a file with the
13017 if (!dirent) /* just before the end */
13019 pos = PerlDir_tell(ret);
13020 if (PerlDir_read(ret)) continue; /* not there yet */
13021 PerlDir_seek(ret, pos); /* step back */
13025 const long pos0 = PerlDir_tell(ret);
13027 pos = PerlDir_tell(ret);
13028 if ((dirent = PerlDir_read(ret))) {
13029 if (len == (STRLEN)d_namlen(dirent)
13030 && memEQ(name, dirent->d_name, len)) {
13032 PerlDir_seek(ret, pos); /* step back */
13035 /* else we are not there yet; keep iterating */
13037 else { /* This is not meant to happen. The best we can do is
13038 reset the iterator to the beginning. */
13039 PerlDir_seek(ret, pos0);
13046 if (name && name != smallbuf)
13051 ret = win32_dirp_dup(dp, param);
13054 /* pop it in the pointer table */
13056 ptr_table_store(PL_ptr_table, dp, ret);
13061 /* duplicate a typeglob */
13064 Perl_gp_dup(pTHX_ GP *const gp, CLONE_PARAMS *const param)
13068 PERL_ARGS_ASSERT_GP_DUP;
13072 /* look for it in the table first */
13073 ret = (GP*)ptr_table_fetch(PL_ptr_table, gp);
13077 /* create anew and remember what it is */
13079 ptr_table_store(PL_ptr_table, gp, ret);
13082 /* ret->gp_refcnt must be 0 before any other dups are called. We're relying
13083 on Newxz() to do this for us. */
13084 ret->gp_sv = sv_dup_inc(gp->gp_sv, param);
13085 ret->gp_io = io_dup_inc(gp->gp_io, param);
13086 ret->gp_form = cv_dup_inc(gp->gp_form, param);
13087 ret->gp_av = av_dup_inc(gp->gp_av, param);
13088 ret->gp_hv = hv_dup_inc(gp->gp_hv, param);
13089 ret->gp_egv = gv_dup(gp->gp_egv, param);/* GvEGV is not refcounted */
13090 ret->gp_cv = cv_dup_inc(gp->gp_cv, param);
13091 ret->gp_cvgen = gp->gp_cvgen;
13092 ret->gp_line = gp->gp_line;
13093 ret->gp_file_hek = hek_dup(gp->gp_file_hek, param);
13097 /* duplicate a chain of magic */
13100 Perl_mg_dup(pTHX_ MAGIC *mg, CLONE_PARAMS *const param)
13102 MAGIC *mgret = NULL;
13103 MAGIC **mgprev_p = &mgret;
13105 PERL_ARGS_ASSERT_MG_DUP;
13107 for (; mg; mg = mg->mg_moremagic) {
13110 if ((param->flags & CLONEf_JOIN_IN)
13111 && mg->mg_type == PERL_MAGIC_backref)
13112 /* when joining, we let the individual SVs add themselves to
13113 * backref as needed. */
13116 Newx(nmg, 1, MAGIC);
13118 mgprev_p = &(nmg->mg_moremagic);
13120 /* There was a comment "XXX copy dynamic vtable?" but as we don't have
13121 dynamic vtables, I'm not sure why Sarathy wrote it. The comment dates
13122 from the original commit adding Perl_mg_dup() - revision 4538.
13123 Similarly there is the annotation "XXX random ptr?" next to the
13124 assignment to nmg->mg_ptr. */
13127 /* FIXME for plugins
13128 if (nmg->mg_type == PERL_MAGIC_qr) {
13129 nmg->mg_obj = MUTABLE_SV(CALLREGDUPE((REGEXP*)nmg->mg_obj, param));
13133 nmg->mg_obj = (nmg->mg_flags & MGf_REFCOUNTED)
13134 ? nmg->mg_type == PERL_MAGIC_backref
13135 /* The backref AV has its reference
13136 * count deliberately bumped by 1 */
13137 ? SvREFCNT_inc(av_dup_inc((const AV *)
13138 nmg->mg_obj, param))
13139 : sv_dup_inc(nmg->mg_obj, param)
13140 : sv_dup(nmg->mg_obj, param);
13142 if (nmg->mg_ptr && nmg->mg_type != PERL_MAGIC_regex_global) {
13143 if (nmg->mg_len > 0) {
13144 nmg->mg_ptr = SAVEPVN(nmg->mg_ptr, nmg->mg_len);
13145 if (nmg->mg_type == PERL_MAGIC_overload_table &&
13146 AMT_AMAGIC((AMT*)nmg->mg_ptr))
13148 AMT * const namtp = (AMT*)nmg->mg_ptr;
13149 sv_dup_inc_multiple((SV**)(namtp->table),
13150 (SV**)(namtp->table), NofAMmeth, param);
13153 else if (nmg->mg_len == HEf_SVKEY)
13154 nmg->mg_ptr = (char*)sv_dup_inc((const SV *)nmg->mg_ptr, param);
13156 if ((nmg->mg_flags & MGf_DUP) && nmg->mg_virtual && nmg->mg_virtual->svt_dup) {
13157 nmg->mg_virtual->svt_dup(aTHX_ nmg, param);
13163 #endif /* USE_ITHREADS */
13165 struct ptr_tbl_arena {
13166 struct ptr_tbl_arena *next;
13167 struct ptr_tbl_ent array[1023/3]; /* as ptr_tbl_ent has 3 pointers. */
13170 /* create a new pointer-mapping table */
13173 Perl_ptr_table_new(pTHX)
13176 PERL_UNUSED_CONTEXT;
13178 Newx(tbl, 1, PTR_TBL_t);
13179 tbl->tbl_max = 511;
13180 tbl->tbl_items = 0;
13181 tbl->tbl_arena = NULL;
13182 tbl->tbl_arena_next = NULL;
13183 tbl->tbl_arena_end = NULL;
13184 Newxz(tbl->tbl_ary, tbl->tbl_max + 1, PTR_TBL_ENT_t*);
13188 #define PTR_TABLE_HASH(ptr) \
13189 ((PTR2UV(ptr) >> 3) ^ (PTR2UV(ptr) >> (3 + 7)) ^ (PTR2UV(ptr) >> (3 + 17)))
13191 /* map an existing pointer using a table */
13193 STATIC PTR_TBL_ENT_t *
13194 S_ptr_table_find(PTR_TBL_t *const tbl, const void *const sv)
13196 PTR_TBL_ENT_t *tblent;
13197 const UV hash = PTR_TABLE_HASH(sv);
13199 PERL_ARGS_ASSERT_PTR_TABLE_FIND;
13201 tblent = tbl->tbl_ary[hash & tbl->tbl_max];
13202 for (; tblent; tblent = tblent->next) {
13203 if (tblent->oldval == sv)
13210 Perl_ptr_table_fetch(pTHX_ PTR_TBL_t *const tbl, const void *const sv)
13212 PTR_TBL_ENT_t const *const tblent = ptr_table_find(tbl, sv);
13214 PERL_ARGS_ASSERT_PTR_TABLE_FETCH;
13215 PERL_UNUSED_CONTEXT;
13217 return tblent ? tblent->newval : NULL;
13220 /* add a new entry to a pointer-mapping table 'tbl'. In hash terms, 'oldsv' is
13221 * the key; 'newsv' is the value. The names "old" and "new" are specific to
13222 * the core's typical use of ptr_tables in thread cloning. */
13225 Perl_ptr_table_store(pTHX_ PTR_TBL_t *const tbl, const void *const oldsv, void *const newsv)
13227 PTR_TBL_ENT_t *tblent = ptr_table_find(tbl, oldsv);
13229 PERL_ARGS_ASSERT_PTR_TABLE_STORE;
13230 PERL_UNUSED_CONTEXT;
13233 tblent->newval = newsv;
13235 const UV entry = PTR_TABLE_HASH(oldsv) & tbl->tbl_max;
13237 if (tbl->tbl_arena_next == tbl->tbl_arena_end) {
13238 struct ptr_tbl_arena *new_arena;
13240 Newx(new_arena, 1, struct ptr_tbl_arena);
13241 new_arena->next = tbl->tbl_arena;
13242 tbl->tbl_arena = new_arena;
13243 tbl->tbl_arena_next = new_arena->array;
13244 tbl->tbl_arena_end = C_ARRAY_END(new_arena->array);
13247 tblent = tbl->tbl_arena_next++;
13249 tblent->oldval = oldsv;
13250 tblent->newval = newsv;
13251 tblent->next = tbl->tbl_ary[entry];
13252 tbl->tbl_ary[entry] = tblent;
13254 if (tblent->next && tbl->tbl_items > tbl->tbl_max)
13255 ptr_table_split(tbl);
13259 /* double the hash bucket size of an existing ptr table */
13262 Perl_ptr_table_split(pTHX_ PTR_TBL_t *const tbl)
13264 PTR_TBL_ENT_t **ary = tbl->tbl_ary;
13265 const UV oldsize = tbl->tbl_max + 1;
13266 UV newsize = oldsize * 2;
13269 PERL_ARGS_ASSERT_PTR_TABLE_SPLIT;
13270 PERL_UNUSED_CONTEXT;
13272 Renew(ary, newsize, PTR_TBL_ENT_t*);
13273 Zero(&ary[oldsize], newsize-oldsize, PTR_TBL_ENT_t*);
13274 tbl->tbl_max = --newsize;
13275 tbl->tbl_ary = ary;
13276 for (i=0; i < oldsize; i++, ary++) {
13277 PTR_TBL_ENT_t **entp = ary;
13278 PTR_TBL_ENT_t *ent = *ary;
13279 PTR_TBL_ENT_t **curentp;
13282 curentp = ary + oldsize;
13284 if ((newsize & PTR_TABLE_HASH(ent->oldval)) != i) {
13286 ent->next = *curentp;
13296 /* remove all the entries from a ptr table */
13297 /* Deprecated - will be removed post 5.14 */
13300 Perl_ptr_table_clear(pTHX_ PTR_TBL_t *const tbl)
13302 PERL_UNUSED_CONTEXT;
13303 if (tbl && tbl->tbl_items) {
13304 struct ptr_tbl_arena *arena = tbl->tbl_arena;
13306 Zero(tbl->tbl_ary, tbl->tbl_max + 1, struct ptr_tbl_ent *);
13309 struct ptr_tbl_arena *next = arena->next;
13315 tbl->tbl_items = 0;
13316 tbl->tbl_arena = NULL;
13317 tbl->tbl_arena_next = NULL;
13318 tbl->tbl_arena_end = NULL;
13322 /* clear and free a ptr table */
13325 Perl_ptr_table_free(pTHX_ PTR_TBL_t *const tbl)
13327 struct ptr_tbl_arena *arena;
13329 PERL_UNUSED_CONTEXT;
13335 arena = tbl->tbl_arena;
13338 struct ptr_tbl_arena *next = arena->next;
13344 Safefree(tbl->tbl_ary);
13348 #if defined(USE_ITHREADS)
13351 Perl_rvpv_dup(pTHX_ SV *const dstr, const SV *const sstr, CLONE_PARAMS *const param)
13353 PERL_ARGS_ASSERT_RVPV_DUP;
13355 assert(!isREGEXP(sstr));
13357 if (SvWEAKREF(sstr)) {
13358 SvRV_set(dstr, sv_dup(SvRV_const(sstr), param));
13359 if (param->flags & CLONEf_JOIN_IN) {
13360 /* if joining, we add any back references individually rather
13361 * than copying the whole backref array */
13362 Perl_sv_add_backref(aTHX_ SvRV(dstr), dstr);
13366 SvRV_set(dstr, sv_dup_inc(SvRV_const(sstr), param));
13368 else if (SvPVX_const(sstr)) {
13369 /* Has something there */
13371 /* Normal PV - clone whole allocated space */
13372 SvPV_set(dstr, SAVEPVN(SvPVX_const(sstr), SvLEN(sstr)-1));
13373 /* sstr may not be that normal, but actually copy on write.
13374 But we are a true, independent SV, so: */
13378 /* Special case - not normally malloced for some reason */
13379 if (isGV_with_GP(sstr)) {
13380 /* Don't need to do anything here. */
13382 else if ((SvIsCOW(sstr))) {
13383 /* A "shared" PV - clone it as "shared" PV */
13385 HEK_KEY(hek_dup(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)),
13389 /* Some other special case - random pointer */
13390 SvPV_set(dstr, (char *) SvPVX_const(sstr));
13395 /* Copy the NULL */
13396 SvPV_set(dstr, NULL);
13400 /* duplicate a list of SVs. source and dest may point to the same memory. */
13402 S_sv_dup_inc_multiple(pTHX_ SV *const *source, SV **dest,
13403 SSize_t items, CLONE_PARAMS *const param)
13405 PERL_ARGS_ASSERT_SV_DUP_INC_MULTIPLE;
13407 while (items-- > 0) {
13408 *dest++ = sv_dup_inc(*source++, param);
13414 /* duplicate an SV of any type (including AV, HV etc) */
13417 S_sv_dup_common(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13422 PERL_ARGS_ASSERT_SV_DUP_COMMON;
13424 if (SvTYPE(sstr) == (svtype)SVTYPEMASK) {
13425 #ifdef DEBUG_LEAKING_SCALARS_ABORT
13430 /* look for it in the table first */
13431 dstr = MUTABLE_SV(ptr_table_fetch(PL_ptr_table, sstr));
13435 if(param->flags & CLONEf_JOIN_IN) {
13436 /** We are joining here so we don't want do clone
13437 something that is bad **/
13438 if (SvTYPE(sstr) == SVt_PVHV) {
13439 const HEK * const hvname = HvNAME_HEK(sstr);
13441 /** don't clone stashes if they already exist **/
13442 dstr = MUTABLE_SV(gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
13443 HEK_UTF8(hvname) ? SVf_UTF8 : 0));
13444 ptr_table_store(PL_ptr_table, sstr, dstr);
13448 else if (SvTYPE(sstr) == SVt_PVGV && !SvFAKE(sstr)) {
13449 HV *stash = GvSTASH(sstr);
13450 const HEK * hvname;
13451 if (stash && (hvname = HvNAME_HEK(stash))) {
13452 /** don't clone GVs if they already exist **/
13454 stash = gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
13455 HEK_UTF8(hvname) ? SVf_UTF8 : 0);
13457 stash, GvNAME(sstr),
13463 if (svp && *svp && SvTYPE(*svp) == SVt_PVGV) {
13464 ptr_table_store(PL_ptr_table, sstr, *svp);
13471 /* create anew and remember what it is */
13474 #ifdef DEBUG_LEAKING_SCALARS
13475 dstr->sv_debug_optype = sstr->sv_debug_optype;
13476 dstr->sv_debug_line = sstr->sv_debug_line;
13477 dstr->sv_debug_inpad = sstr->sv_debug_inpad;
13478 dstr->sv_debug_parent = (SV*)sstr;
13479 FREE_SV_DEBUG_FILE(dstr);
13480 dstr->sv_debug_file = savesharedpv(sstr->sv_debug_file);
13483 ptr_table_store(PL_ptr_table, sstr, dstr);
13486 SvFLAGS(dstr) = SvFLAGS(sstr);
13487 SvFLAGS(dstr) &= ~SVf_OOK; /* don't propagate OOK hack */
13488 SvREFCNT(dstr) = 0; /* must be before any other dups! */
13491 if (SvANY(sstr) && PL_watch_pvx && SvPVX_const(sstr) == PL_watch_pvx)
13492 PerlIO_printf(Perl_debug_log, "watch at %p hit, found string \"%s\"\n",
13493 (void*)PL_watch_pvx, SvPVX_const(sstr));
13496 /* don't clone objects whose class has asked us not to */
13498 && ! (SvFLAGS(SvSTASH(sstr)) & SVphv_CLONEABLE))
13504 switch (SvTYPE(sstr)) {
13506 SvANY(dstr) = NULL;
13509 SET_SVANY_FOR_BODYLESS_IV(dstr);
13511 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
13513 SvIV_set(dstr, SvIVX(sstr));
13517 #if NVSIZE <= IVSIZE
13518 SET_SVANY_FOR_BODYLESS_NV(dstr);
13520 SvANY(dstr) = new_XNV();
13522 SvNV_set(dstr, SvNVX(sstr));
13526 /* These are all the types that need complex bodies allocating. */
13528 const svtype sv_type = SvTYPE(sstr);
13529 const struct body_details *const sv_type_details
13530 = bodies_by_type + sv_type;
13534 Perl_croak(aTHX_ "Bizarre SvTYPE [%" IVdf "]", (IV)SvTYPE(sstr));
13550 assert(sv_type_details->body_size);
13551 if (sv_type_details->arena) {
13552 new_body_inline(new_body, sv_type);
13554 = (void*)((char*)new_body - sv_type_details->offset);
13556 new_body = new_NOARENA(sv_type_details);
13560 SvANY(dstr) = new_body;
13563 Copy(((char*)SvANY(sstr)) + sv_type_details->offset,
13564 ((char*)SvANY(dstr)) + sv_type_details->offset,
13565 sv_type_details->copy, char);
13567 Copy(((char*)SvANY(sstr)),
13568 ((char*)SvANY(dstr)),
13569 sv_type_details->body_size + sv_type_details->offset, char);
13572 if (sv_type != SVt_PVAV && sv_type != SVt_PVHV
13573 && !isGV_with_GP(dstr)
13575 && !(sv_type == SVt_PVIO && !(IoFLAGS(dstr) & IOf_FAKE_DIRP)))
13576 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
13578 /* The Copy above means that all the source (unduplicated) pointers
13579 are now in the destination. We can check the flags and the
13580 pointers in either, but it's possible that there's less cache
13581 missing by always going for the destination.
13582 FIXME - instrument and check that assumption */
13583 if (sv_type >= SVt_PVMG) {
13585 SvMAGIC_set(dstr, mg_dup(SvMAGIC(dstr), param));
13586 if (SvOBJECT(dstr) && SvSTASH(dstr))
13587 SvSTASH_set(dstr, hv_dup_inc(SvSTASH(dstr), param));
13588 else SvSTASH_set(dstr, 0); /* don't copy DESTROY cache */
13591 /* The cast silences a GCC warning about unhandled types. */
13592 switch ((int)sv_type) {
13603 /* FIXME for plugins */
13604 dstr->sv_u.svu_rx = ((REGEXP *)dstr)->sv_any;
13605 re_dup_guts((REGEXP*) sstr, (REGEXP*) dstr, param);
13608 /* XXX LvTARGOFF sometimes holds PMOP* when DEBUGGING */
13609 if (LvTYPE(dstr) == 't') /* for tie: unrefcnted fake (SV**) */
13610 LvTARG(dstr) = dstr;
13611 else if (LvTYPE(dstr) == 'T') /* for tie: fake HE */
13612 LvTARG(dstr) = MUTABLE_SV(he_dup((HE*)LvTARG(dstr), 0, param));
13614 LvTARG(dstr) = sv_dup_inc(LvTARG(dstr), param);
13615 if (isREGEXP(sstr)) goto duprex;
13617 /* non-GP case already handled above */
13618 if(isGV_with_GP(sstr)) {
13619 GvNAME_HEK(dstr) = hek_dup(GvNAME_HEK(dstr), param);
13620 /* Don't call sv_add_backref here as it's going to be
13621 created as part of the magic cloning of the symbol
13622 table--unless this is during a join and the stash
13623 is not actually being cloned. */
13624 /* Danger Will Robinson - GvGP(dstr) isn't initialised
13625 at the point of this comment. */
13626 GvSTASH(dstr) = hv_dup(GvSTASH(dstr), param);
13627 if (param->flags & CLONEf_JOIN_IN)
13628 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
13629 GvGP_set(dstr, gp_dup(GvGP(sstr), param));
13630 (void)GpREFCNT_inc(GvGP(dstr));
13634 /* PL_parser->rsfp_filters entries have fake IoDIRP() */
13635 if(IoFLAGS(dstr) & IOf_FAKE_DIRP) {
13636 /* I have no idea why fake dirp (rsfps)
13637 should be treated differently but otherwise
13638 we end up with leaks -- sky*/
13639 IoTOP_GV(dstr) = gv_dup_inc(IoTOP_GV(dstr), param);
13640 IoFMT_GV(dstr) = gv_dup_inc(IoFMT_GV(dstr), param);
13641 IoBOTTOM_GV(dstr) = gv_dup_inc(IoBOTTOM_GV(dstr), param);
13643 IoTOP_GV(dstr) = gv_dup(IoTOP_GV(dstr), param);
13644 IoFMT_GV(dstr) = gv_dup(IoFMT_GV(dstr), param);
13645 IoBOTTOM_GV(dstr) = gv_dup(IoBOTTOM_GV(dstr), param);
13646 if (IoDIRP(dstr)) {
13647 IoDIRP(dstr) = dirp_dup(IoDIRP(dstr), param);
13650 /* IoDIRP(dstr) is already a copy of IoDIRP(sstr) */
13652 IoIFP(dstr) = fp_dup(IoIFP(sstr), IoTYPE(dstr), param);
13654 if (IoOFP(dstr) == IoIFP(sstr))
13655 IoOFP(dstr) = IoIFP(dstr);
13657 IoOFP(dstr) = fp_dup(IoOFP(dstr), IoTYPE(dstr), param);
13658 IoTOP_NAME(dstr) = SAVEPV(IoTOP_NAME(dstr));
13659 IoFMT_NAME(dstr) = SAVEPV(IoFMT_NAME(dstr));
13660 IoBOTTOM_NAME(dstr) = SAVEPV(IoBOTTOM_NAME(dstr));
13663 /* avoid cloning an empty array */
13664 if (AvARRAY((const AV *)sstr) && AvFILLp((const AV *)sstr) >= 0) {
13665 SV **dst_ary, **src_ary;
13666 SSize_t items = AvFILLp((const AV *)sstr) + 1;
13668 src_ary = AvARRAY((const AV *)sstr);
13669 Newxz(dst_ary, AvMAX((const AV *)sstr)+1, SV*);
13670 ptr_table_store(PL_ptr_table, src_ary, dst_ary);
13671 AvARRAY(MUTABLE_AV(dstr)) = dst_ary;
13672 AvALLOC((const AV *)dstr) = dst_ary;
13673 if (AvREAL((const AV *)sstr)) {
13674 dst_ary = sv_dup_inc_multiple(src_ary, dst_ary, items,
13678 while (items-- > 0)
13679 *dst_ary++ = sv_dup(*src_ary++, param);
13681 items = AvMAX((const AV *)sstr) - AvFILLp((const AV *)sstr);
13682 while (items-- > 0) {
13687 AvARRAY(MUTABLE_AV(dstr)) = NULL;
13688 AvALLOC((const AV *)dstr) = (SV**)NULL;
13689 AvMAX( (const AV *)dstr) = -1;
13690 AvFILLp((const AV *)dstr) = -1;
13694 if (HvARRAY((const HV *)sstr)) {
13696 const bool sharekeys = !!HvSHAREKEYS(sstr);
13697 XPVHV * const dxhv = (XPVHV*)SvANY(dstr);
13698 XPVHV * const sxhv = (XPVHV*)SvANY(sstr);
13700 Newx(darray, PERL_HV_ARRAY_ALLOC_BYTES(dxhv->xhv_max+1)
13701 + (SvOOK(sstr) ? sizeof(struct xpvhv_aux) : 0),
13703 HvARRAY(dstr) = (HE**)darray;
13704 while (i <= sxhv->xhv_max) {
13705 const HE * const source = HvARRAY(sstr)[i];
13706 HvARRAY(dstr)[i] = source
13707 ? he_dup(source, sharekeys, param) : 0;
13711 const struct xpvhv_aux * const saux = HvAUX(sstr);
13712 struct xpvhv_aux * const daux = HvAUX(dstr);
13713 /* This flag isn't copied. */
13716 if (saux->xhv_name_count) {
13717 HEK ** const sname = saux->xhv_name_u.xhvnameu_names;
13719 = saux->xhv_name_count < 0
13720 ? -saux->xhv_name_count
13721 : saux->xhv_name_count;
13722 HEK **shekp = sname + count;
13724 Newx(daux->xhv_name_u.xhvnameu_names, count, HEK *);
13725 dhekp = daux->xhv_name_u.xhvnameu_names + count;
13726 while (shekp-- > sname) {
13728 *dhekp = hek_dup(*shekp, param);
13732 daux->xhv_name_u.xhvnameu_name
13733 = hek_dup(saux->xhv_name_u.xhvnameu_name,
13736 daux->xhv_name_count = saux->xhv_name_count;
13738 daux->xhv_fill_lazy = saux->xhv_fill_lazy;
13739 daux->xhv_aux_flags = saux->xhv_aux_flags;
13740 #ifdef PERL_HASH_RANDOMIZE_KEYS
13741 daux->xhv_rand = saux->xhv_rand;
13742 daux->xhv_last_rand = saux->xhv_last_rand;
13744 daux->xhv_riter = saux->xhv_riter;
13745 daux->xhv_eiter = saux->xhv_eiter
13746 ? he_dup(saux->xhv_eiter,
13747 cBOOL(HvSHAREKEYS(sstr)), param) : 0;
13748 /* backref array needs refcnt=2; see sv_add_backref */
13749 daux->xhv_backreferences =
13750 (param->flags & CLONEf_JOIN_IN)
13751 /* when joining, we let the individual GVs and
13752 * CVs add themselves to backref as
13753 * needed. This avoids pulling in stuff
13754 * that isn't required, and simplifies the
13755 * case where stashes aren't cloned back
13756 * if they already exist in the parent
13759 : saux->xhv_backreferences
13760 ? (SvTYPE(saux->xhv_backreferences) == SVt_PVAV)
13761 ? MUTABLE_AV(SvREFCNT_inc(
13762 sv_dup_inc((const SV *)
13763 saux->xhv_backreferences, param)))
13764 : MUTABLE_AV(sv_dup((const SV *)
13765 saux->xhv_backreferences, param))
13768 daux->xhv_mro_meta = saux->xhv_mro_meta
13769 ? mro_meta_dup(saux->xhv_mro_meta, param)
13772 /* Record stashes for possible cloning in Perl_clone(). */
13774 av_push(param->stashes, dstr);
13778 HvARRAY(MUTABLE_HV(dstr)) = NULL;
13781 if (!(param->flags & CLONEf_COPY_STACKS)) {
13786 /* NOTE: not refcounted */
13787 SvANY(MUTABLE_CV(dstr))->xcv_stash =
13788 hv_dup(CvSTASH(dstr), param);
13789 if ((param->flags & CLONEf_JOIN_IN) && CvSTASH(dstr))
13790 Perl_sv_add_backref(aTHX_ MUTABLE_SV(CvSTASH(dstr)), dstr);
13791 if (!CvISXSUB(dstr)) {
13793 CvROOT(dstr) = OpREFCNT_inc(CvROOT(dstr));
13795 CvSLABBED_off(dstr);
13796 } else if (CvCONST(dstr)) {
13797 CvXSUBANY(dstr).any_ptr =
13798 sv_dup_inc((const SV *)CvXSUBANY(dstr).any_ptr, param);
13800 assert(!CvSLABBED(dstr));
13801 if (CvDYNFILE(dstr)) CvFILE(dstr) = SAVEPV(CvFILE(dstr));
13803 SvANY((CV *)dstr)->xcv_gv_u.xcv_hek =
13804 hek_dup(CvNAME_HEK((CV *)sstr), param);
13805 /* don't dup if copying back - CvGV isn't refcounted, so the
13806 * duped GV may never be freed. A bit of a hack! DAPM */
13808 SvANY(MUTABLE_CV(dstr))->xcv_gv_u.xcv_gv =
13810 ? gv_dup_inc(CvGV(sstr), param)
13811 : (param->flags & CLONEf_JOIN_IN)
13813 : gv_dup(CvGV(sstr), param);
13815 if (!CvISXSUB(sstr)) {
13816 PADLIST * padlist = CvPADLIST(sstr);
13818 padlist = padlist_dup(padlist, param);
13819 CvPADLIST_set(dstr, padlist);
13821 /* unthreaded perl can't sv_dup so we dont support unthreaded's CvHSCXT */
13822 PoisonPADLIST(dstr);
13825 CvWEAKOUTSIDE(sstr)
13826 ? cv_dup( CvOUTSIDE(dstr), param)
13827 : cv_dup_inc(CvOUTSIDE(dstr), param);
13837 Perl_sv_dup_inc(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13839 PERL_ARGS_ASSERT_SV_DUP_INC;
13840 return sstr ? SvREFCNT_inc(sv_dup_common(sstr, param)) : NULL;
13844 Perl_sv_dup(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13846 SV *dstr = sstr ? sv_dup_common(sstr, param) : NULL;
13847 PERL_ARGS_ASSERT_SV_DUP;
13849 /* Track every SV that (at least initially) had a reference count of 0.
13850 We need to do this by holding an actual reference to it in this array.
13851 If we attempt to cheat, turn AvREAL_off(), and store only pointers
13852 (akin to the stashes hash, and the perl stack), we come unstuck if
13853 a weak reference (or other SV legitimately SvREFCNT() == 0 for this
13854 thread) is manipulated in a CLONE method, because CLONE runs before the
13855 unreferenced array is walked to find SVs still with SvREFCNT() == 0
13856 (and fix things up by giving each a reference via the temps stack).
13857 Instead, during CLONE, if the 0-referenced SV has SvREFCNT_inc() and
13858 then SvREFCNT_dec(), it will be cleaned up (and added to the free list)
13859 before the walk of unreferenced happens and a reference to that is SV
13860 added to the temps stack. At which point we have the same SV considered
13861 to be in use, and free to be re-used. Not good.
13863 if (dstr && !(param->flags & CLONEf_COPY_STACKS) && !SvREFCNT(dstr)) {
13864 assert(param->unreferenced);
13865 av_push(param->unreferenced, SvREFCNT_inc(dstr));
13871 /* duplicate a context */
13874 Perl_cx_dup(pTHX_ PERL_CONTEXT *cxs, I32 ix, I32 max, CLONE_PARAMS* param)
13876 PERL_CONTEXT *ncxs;
13878 PERL_ARGS_ASSERT_CX_DUP;
13881 return (PERL_CONTEXT*)NULL;
13883 /* look for it in the table first */
13884 ncxs = (PERL_CONTEXT*)ptr_table_fetch(PL_ptr_table, cxs);
13888 /* create anew and remember what it is */
13889 Newx(ncxs, max + 1, PERL_CONTEXT);
13890 ptr_table_store(PL_ptr_table, cxs, ncxs);
13891 Copy(cxs, ncxs, max + 1, PERL_CONTEXT);
13894 PERL_CONTEXT * const ncx = &ncxs[ix];
13895 if (CxTYPE(ncx) == CXt_SUBST) {
13896 Perl_croak(aTHX_ "Cloning substitution context is unimplemented");
13899 ncx->blk_oldcop = (COP*)any_dup(ncx->blk_oldcop, param->proto_perl);
13900 switch (CxTYPE(ncx)) {
13902 ncx->blk_sub.cv = (ncx->blk_sub.olddepth == 0
13903 ? cv_dup_inc(ncx->blk_sub.cv, param)
13904 : cv_dup(ncx->blk_sub.cv,param));
13905 if(CxHASARGS(ncx)){
13906 ncx->blk_sub.argarray = av_dup_inc(ncx->blk_sub.argarray,param);
13907 ncx->blk_sub.savearray = av_dup_inc(ncx->blk_sub.savearray,param);
13909 ncx->blk_sub.argarray = NULL;
13910 ncx->blk_sub.savearray = NULL;
13912 ncx->blk_sub.oldcomppad = (PAD*)ptr_table_fetch(PL_ptr_table,
13913 ncx->blk_sub.oldcomppad);
13916 ncx->blk_eval.old_namesv = sv_dup_inc(ncx->blk_eval.old_namesv,
13918 ncx->blk_eval.cur_text = sv_dup(ncx->blk_eval.cur_text, param);
13919 ncx->blk_eval.cv = cv_dup(ncx->blk_eval.cv, param);
13921 case CXt_LOOP_LAZYSV:
13922 ncx->blk_loop.state_u.lazysv.end
13923 = sv_dup_inc(ncx->blk_loop.state_u.lazysv.end, param);
13924 /* Fallthrough: duplicate lazysv.cur by using the ary.ary
13925 duplication code instead.
13926 We are taking advantage of (1) av_dup_inc and sv_dup_inc
13927 actually being the same function, and (2) order
13928 equivalence of the two unions.
13929 We can assert the later [but only at run time :-(] */
13930 assert ((void *) &ncx->blk_loop.state_u.ary.ary ==
13931 (void *) &ncx->blk_loop.state_u.lazysv.cur);
13934 ncx->blk_loop.state_u.ary.ary
13935 = av_dup_inc(ncx->blk_loop.state_u.ary.ary, param);
13937 case CXt_LOOP_LAZYIV:
13938 case CXt_LOOP_PLAIN:
13939 /* code common to all CXt_LOOP_* types */
13940 if (CxPADLOOP(ncx)) {
13941 ncx->blk_loop.itervar_u.oldcomppad
13942 = (PAD*)ptr_table_fetch(PL_ptr_table,
13943 ncx->blk_loop.itervar_u.oldcomppad);
13945 ncx->blk_loop.itervar_u.gv
13946 = gv_dup((const GV *)ncx->blk_loop.itervar_u.gv,
13951 ncx->blk_format.cv = cv_dup(ncx->blk_format.cv, param);
13952 ncx->blk_format.gv = gv_dup(ncx->blk_format.gv, param);
13953 ncx->blk_format.dfoutgv = gv_dup_inc(ncx->blk_format.dfoutgv,
13968 /* duplicate a stack info structure */
13971 Perl_si_dup(pTHX_ PERL_SI *si, CLONE_PARAMS* param)
13975 PERL_ARGS_ASSERT_SI_DUP;
13978 return (PERL_SI*)NULL;
13980 /* look for it in the table first */
13981 nsi = (PERL_SI*)ptr_table_fetch(PL_ptr_table, si);
13985 /* create anew and remember what it is */
13986 Newxz(nsi, 1, PERL_SI);
13987 ptr_table_store(PL_ptr_table, si, nsi);
13989 nsi->si_stack = av_dup_inc(si->si_stack, param);
13990 nsi->si_cxix = si->si_cxix;
13991 nsi->si_cxmax = si->si_cxmax;
13992 nsi->si_cxstack = cx_dup(si->si_cxstack, si->si_cxix, si->si_cxmax, param);
13993 nsi->si_type = si->si_type;
13994 nsi->si_prev = si_dup(si->si_prev, param);
13995 nsi->si_next = si_dup(si->si_next, param);
13996 nsi->si_markoff = si->si_markoff;
14001 #define POPINT(ss,ix) ((ss)[--(ix)].any_i32)
14002 #define TOPINT(ss,ix) ((ss)[ix].any_i32)
14003 #define POPLONG(ss,ix) ((ss)[--(ix)].any_long)
14004 #define TOPLONG(ss,ix) ((ss)[ix].any_long)
14005 #define POPIV(ss,ix) ((ss)[--(ix)].any_iv)
14006 #define TOPIV(ss,ix) ((ss)[ix].any_iv)
14007 #define POPUV(ss,ix) ((ss)[--(ix)].any_uv)
14008 #define TOPUV(ss,ix) ((ss)[ix].any_uv)
14009 #define POPBOOL(ss,ix) ((ss)[--(ix)].any_bool)
14010 #define TOPBOOL(ss,ix) ((ss)[ix].any_bool)
14011 #define POPPTR(ss,ix) ((ss)[--(ix)].any_ptr)
14012 #define TOPPTR(ss,ix) ((ss)[ix].any_ptr)
14013 #define POPDPTR(ss,ix) ((ss)[--(ix)].any_dptr)
14014 #define TOPDPTR(ss,ix) ((ss)[ix].any_dptr)
14015 #define POPDXPTR(ss,ix) ((ss)[--(ix)].any_dxptr)
14016 #define TOPDXPTR(ss,ix) ((ss)[ix].any_dxptr)
14019 #define pv_dup_inc(p) SAVEPV(p)
14020 #define pv_dup(p) SAVEPV(p)
14021 #define svp_dup_inc(p,pp) any_dup(p,pp)
14023 /* map any object to the new equivent - either something in the
14024 * ptr table, or something in the interpreter structure
14028 Perl_any_dup(pTHX_ void *v, const PerlInterpreter *proto_perl)
14032 PERL_ARGS_ASSERT_ANY_DUP;
14035 return (void*)NULL;
14037 /* look for it in the table first */
14038 ret = ptr_table_fetch(PL_ptr_table, v);
14042 /* see if it is part of the interpreter structure */
14043 if (v >= (void*)proto_perl && v < (void*)(proto_perl+1))
14044 ret = (void*)(((char*)aTHX) + (((char*)v) - (char*)proto_perl));
14052 /* duplicate the save stack */
14055 Perl_ss_dup(pTHX_ PerlInterpreter *proto_perl, CLONE_PARAMS* param)
14058 ANY * const ss = proto_perl->Isavestack;
14059 const I32 max = proto_perl->Isavestack_max;
14060 I32 ix = proto_perl->Isavestack_ix;
14073 void (*dptr) (void*);
14074 void (*dxptr) (pTHX_ void*);
14076 PERL_ARGS_ASSERT_SS_DUP;
14078 Newxz(nss, max, ANY);
14081 const UV uv = POPUV(ss,ix);
14082 const U8 type = (U8)uv & SAVE_MASK;
14084 TOPUV(nss,ix) = uv;
14086 case SAVEt_CLEARSV:
14087 case SAVEt_CLEARPADRANGE:
14089 case SAVEt_HELEM: /* hash element */
14090 case SAVEt_SV: /* scalar reference */
14091 sv = (const SV *)POPPTR(ss,ix);
14092 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14094 case SAVEt_ITEM: /* normal string */
14095 case SAVEt_GVSV: /* scalar slot in GV */
14096 sv = (const SV *)POPPTR(ss,ix);
14097 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14098 if (type == SAVEt_SV)
14102 case SAVEt_MORTALIZESV:
14103 case SAVEt_READONLY_OFF:
14104 sv = (const SV *)POPPTR(ss,ix);
14105 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14107 case SAVEt_FREEPADNAME:
14108 ptr = POPPTR(ss,ix);
14109 TOPPTR(nss,ix) = padname_dup((PADNAME *)ptr, param);
14110 PadnameREFCNT((PADNAME *)TOPPTR(nss,ix))++;
14112 case SAVEt_SHARED_PVREF: /* char* in shared space */
14113 c = (char*)POPPTR(ss,ix);
14114 TOPPTR(nss,ix) = savesharedpv(c);
14115 ptr = POPPTR(ss,ix);
14116 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14118 case SAVEt_GENERIC_SVREF: /* generic sv */
14119 case SAVEt_SVREF: /* scalar reference */
14120 sv = (const SV *)POPPTR(ss,ix);
14121 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14122 if (type == SAVEt_SVREF)
14123 SvREFCNT_inc_simple_void((SV *)TOPPTR(nss,ix));
14124 ptr = POPPTR(ss,ix);
14125 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14127 case SAVEt_GVSLOT: /* any slot in GV */
14128 sv = (const SV *)POPPTR(ss,ix);
14129 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14130 ptr = POPPTR(ss,ix);
14131 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14132 sv = (const SV *)POPPTR(ss,ix);
14133 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14135 case SAVEt_HV: /* hash reference */
14136 case SAVEt_AV: /* array reference */
14137 sv = (const SV *) POPPTR(ss,ix);
14138 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14140 case SAVEt_COMPPAD:
14142 sv = (const SV *) POPPTR(ss,ix);
14143 TOPPTR(nss,ix) = sv_dup(sv, param);
14145 case SAVEt_INT: /* int reference */
14146 ptr = POPPTR(ss,ix);
14147 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14148 intval = (int)POPINT(ss,ix);
14149 TOPINT(nss,ix) = intval;
14151 case SAVEt_LONG: /* long reference */
14152 ptr = POPPTR(ss,ix);
14153 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14154 longval = (long)POPLONG(ss,ix);
14155 TOPLONG(nss,ix) = longval;
14157 case SAVEt_I32: /* I32 reference */
14158 ptr = POPPTR(ss,ix);
14159 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14161 TOPINT(nss,ix) = i;
14163 case SAVEt_IV: /* IV reference */
14164 case SAVEt_STRLEN: /* STRLEN/size_t ref */
14165 ptr = POPPTR(ss,ix);
14166 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14168 TOPIV(nss,ix) = iv;
14170 case SAVEt_HPTR: /* HV* reference */
14171 case SAVEt_APTR: /* AV* reference */
14172 case SAVEt_SPTR: /* SV* reference */
14173 ptr = POPPTR(ss,ix);
14174 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14175 sv = (const SV *)POPPTR(ss,ix);
14176 TOPPTR(nss,ix) = sv_dup(sv, param);
14178 case SAVEt_VPTR: /* random* reference */
14179 ptr = POPPTR(ss,ix);
14180 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14182 case SAVEt_INT_SMALL:
14183 case SAVEt_I32_SMALL:
14184 case SAVEt_I16: /* I16 reference */
14185 case SAVEt_I8: /* I8 reference */
14187 ptr = POPPTR(ss,ix);
14188 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14190 case SAVEt_GENERIC_PVREF: /* generic char* */
14191 case SAVEt_PPTR: /* char* reference */
14192 ptr = POPPTR(ss,ix);
14193 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14194 c = (char*)POPPTR(ss,ix);
14195 TOPPTR(nss,ix) = pv_dup(c);
14197 case SAVEt_GP: /* scalar reference */
14198 gp = (GP*)POPPTR(ss,ix);
14199 TOPPTR(nss,ix) = gp = gp_dup(gp, param);
14200 (void)GpREFCNT_inc(gp);
14201 gv = (const GV *)POPPTR(ss,ix);
14202 TOPPTR(nss,ix) = gv_dup_inc(gv, param);
14205 ptr = POPPTR(ss,ix);
14206 if (ptr && (((OP*)ptr)->op_private & OPpREFCOUNTED)) {
14207 /* these are assumed to be refcounted properly */
14209 switch (((OP*)ptr)->op_type) {
14211 case OP_LEAVESUBLV:
14215 case OP_LEAVEWRITE:
14216 TOPPTR(nss,ix) = ptr;
14219 (void) OpREFCNT_inc(o);
14223 TOPPTR(nss,ix) = NULL;
14228 TOPPTR(nss,ix) = NULL;
14230 case SAVEt_FREECOPHH:
14231 ptr = POPPTR(ss,ix);
14232 TOPPTR(nss,ix) = cophh_copy((COPHH *)ptr);
14234 case SAVEt_ADELETE:
14235 av = (const AV *)POPPTR(ss,ix);
14236 TOPPTR(nss,ix) = av_dup_inc(av, param);
14238 TOPINT(nss,ix) = i;
14241 hv = (const HV *)POPPTR(ss,ix);
14242 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14244 TOPINT(nss,ix) = i;
14247 c = (char*)POPPTR(ss,ix);
14248 TOPPTR(nss,ix) = pv_dup_inc(c);
14250 case SAVEt_STACK_POS: /* Position on Perl stack */
14252 TOPINT(nss,ix) = i;
14254 case SAVEt_DESTRUCTOR:
14255 ptr = POPPTR(ss,ix);
14256 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14257 dptr = POPDPTR(ss,ix);
14258 TOPDPTR(nss,ix) = DPTR2FPTR(void (*)(void*),
14259 any_dup(FPTR2DPTR(void *, dptr),
14262 case SAVEt_DESTRUCTOR_X:
14263 ptr = POPPTR(ss,ix);
14264 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14265 dxptr = POPDXPTR(ss,ix);
14266 TOPDXPTR(nss,ix) = DPTR2FPTR(void (*)(pTHX_ void*),
14267 any_dup(FPTR2DPTR(void *, dxptr),
14270 case SAVEt_REGCONTEXT:
14272 ix -= uv >> SAVE_TIGHT_SHIFT;
14274 case SAVEt_AELEM: /* array element */
14275 sv = (const SV *)POPPTR(ss,ix);
14276 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14278 TOPINT(nss,ix) = i;
14279 av = (const AV *)POPPTR(ss,ix);
14280 TOPPTR(nss,ix) = av_dup_inc(av, param);
14283 ptr = POPPTR(ss,ix);
14284 TOPPTR(nss,ix) = ptr;
14287 ptr = POPPTR(ss,ix);
14288 ptr = cophh_copy((COPHH*)ptr);
14289 TOPPTR(nss,ix) = ptr;
14291 TOPINT(nss,ix) = i;
14292 if (i & HINT_LOCALIZE_HH) {
14293 hv = (const HV *)POPPTR(ss,ix);
14294 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14297 case SAVEt_PADSV_AND_MORTALIZE:
14298 longval = (long)POPLONG(ss,ix);
14299 TOPLONG(nss,ix) = longval;
14300 ptr = POPPTR(ss,ix);
14301 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14302 sv = (const SV *)POPPTR(ss,ix);
14303 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14305 case SAVEt_SET_SVFLAGS:
14307 TOPINT(nss,ix) = i;
14309 TOPINT(nss,ix) = i;
14310 sv = (const SV *)POPPTR(ss,ix);
14311 TOPPTR(nss,ix) = sv_dup(sv, param);
14313 case SAVEt_COMPILE_WARNINGS:
14314 ptr = POPPTR(ss,ix);
14315 TOPPTR(nss,ix) = DUP_WARNINGS((STRLEN*)ptr);
14318 ptr = POPPTR(ss,ix);
14319 TOPPTR(nss,ix) = parser_dup((const yy_parser*)ptr, param);
14323 "panic: ss_dup inconsistency (%"IVdf")", (IV) type);
14331 /* if sv is a stash, call $class->CLONE_SKIP(), and set the SVphv_CLONEABLE
14332 * flag to the result. This is done for each stash before cloning starts,
14333 * so we know which stashes want their objects cloned */
14336 do_mark_cloneable_stash(pTHX_ SV *const sv)
14338 const HEK * const hvname = HvNAME_HEK((const HV *)sv);
14340 GV* const cloner = gv_fetchmethod_autoload(MUTABLE_HV(sv), "CLONE_SKIP", 0);
14341 SvFLAGS(sv) |= SVphv_CLONEABLE; /* clone objects by default */
14342 if (cloner && GvCV(cloner)) {
14349 mXPUSHs(newSVhek(hvname));
14351 call_sv(MUTABLE_SV(GvCV(cloner)), G_SCALAR);
14358 SvFLAGS(sv) &= ~SVphv_CLONEABLE;
14366 =for apidoc perl_clone
14368 Create and return a new interpreter by cloning the current one.
14370 C<perl_clone> takes these flags as parameters:
14372 C<CLONEf_COPY_STACKS> - is used to, well, copy the stacks also,
14373 without it we only clone the data and zero the stacks,
14374 with it we copy the stacks and the new perl interpreter is
14375 ready to run at the exact same point as the previous one.
14376 The pseudo-fork code uses C<COPY_STACKS> while the
14377 threads->create doesn't.
14379 C<CLONEf_KEEP_PTR_TABLE> -
14380 C<perl_clone> keeps a ptr_table with the pointer of the old
14381 variable as a key and the new variable as a value,
14382 this allows it to check if something has been cloned and not
14383 clone it again but rather just use the value and increase the
14384 refcount. If C<KEEP_PTR_TABLE> is not set then C<perl_clone> will kill
14385 the ptr_table using the function
14386 C<ptr_table_free(PL_ptr_table); PL_ptr_table = NULL;>,
14387 reason to keep it around is if you want to dup some of your own
14388 variable who are outside the graph perl scans, an example of this
14389 code is in F<threads.xs> create.
14391 C<CLONEf_CLONE_HOST> -
14392 This is a win32 thing, it is ignored on unix, it tells perls
14393 win32host code (which is c++) to clone itself, this is needed on
14394 win32 if you want to run two threads at the same time,
14395 if you just want to do some stuff in a separate perl interpreter
14396 and then throw it away and return to the original one,
14397 you don't need to do anything.
14402 /* XXX the above needs expanding by someone who actually understands it ! */
14403 EXTERN_C PerlInterpreter *
14404 perl_clone_host(PerlInterpreter* proto_perl, UV flags);
14407 perl_clone(PerlInterpreter *proto_perl, UV flags)
14410 #ifdef PERL_IMPLICIT_SYS
14412 PERL_ARGS_ASSERT_PERL_CLONE;
14414 /* perlhost.h so we need to call into it
14415 to clone the host, CPerlHost should have a c interface, sky */
14417 #ifndef __amigaos4__
14418 if (flags & CLONEf_CLONE_HOST) {
14419 return perl_clone_host(proto_perl,flags);
14422 return perl_clone_using(proto_perl, flags,
14424 proto_perl->IMemShared,
14425 proto_perl->IMemParse,
14427 proto_perl->IStdIO,
14431 proto_perl->IProc);
14435 perl_clone_using(PerlInterpreter *proto_perl, UV flags,
14436 struct IPerlMem* ipM, struct IPerlMem* ipMS,
14437 struct IPerlMem* ipMP, struct IPerlEnv* ipE,
14438 struct IPerlStdIO* ipStd, struct IPerlLIO* ipLIO,
14439 struct IPerlDir* ipD, struct IPerlSock* ipS,
14440 struct IPerlProc* ipP)
14442 /* XXX many of the string copies here can be optimized if they're
14443 * constants; they need to be allocated as common memory and just
14444 * their pointers copied. */
14447 CLONE_PARAMS clone_params;
14448 CLONE_PARAMS* const param = &clone_params;
14450 PerlInterpreter * const my_perl = (PerlInterpreter*)(*ipM->pMalloc)(ipM, sizeof(PerlInterpreter));
14452 PERL_ARGS_ASSERT_PERL_CLONE_USING;
14453 #else /* !PERL_IMPLICIT_SYS */
14455 CLONE_PARAMS clone_params;
14456 CLONE_PARAMS* param = &clone_params;
14457 PerlInterpreter * const my_perl = (PerlInterpreter*)PerlMem_malloc(sizeof(PerlInterpreter));
14459 PERL_ARGS_ASSERT_PERL_CLONE;
14460 #endif /* PERL_IMPLICIT_SYS */
14462 /* for each stash, determine whether its objects should be cloned */
14463 S_visit(proto_perl, do_mark_cloneable_stash, SVt_PVHV, SVTYPEMASK);
14464 PERL_SET_THX(my_perl);
14467 PoisonNew(my_perl, 1, PerlInterpreter);
14470 PL_defstash = NULL; /* may be used by perl malloc() */
14473 PL_scopestack_name = 0;
14475 PL_savestack_ix = 0;
14476 PL_savestack_max = -1;
14477 PL_sig_pending = 0;
14479 Zero(&PL_debug_pad, 1, struct perl_debug_pad);
14480 Zero(&PL_padname_undef, 1, PADNAME);
14481 Zero(&PL_padname_const, 1, PADNAME);
14482 # ifdef DEBUG_LEAKING_SCALARS
14483 PL_sv_serial = (((UV)my_perl >> 2) & 0xfff) * 1000000;
14485 # ifdef PERL_TRACE_OPS
14486 Zero(PL_op_exec_cnt, OP_max+2, UV);
14488 #else /* !DEBUGGING */
14489 Zero(my_perl, 1, PerlInterpreter);
14490 #endif /* DEBUGGING */
14492 #ifdef PERL_IMPLICIT_SYS
14493 /* host pointers */
14495 PL_MemShared = ipMS;
14496 PL_MemParse = ipMP;
14503 #endif /* PERL_IMPLICIT_SYS */
14506 param->flags = flags;
14507 /* Nothing in the core code uses this, but we make it available to
14508 extensions (using mg_dup). */
14509 param->proto_perl = proto_perl;
14510 /* Likely nothing will use this, but it is initialised to be consistent
14511 with Perl_clone_params_new(). */
14512 param->new_perl = my_perl;
14513 param->unreferenced = NULL;
14516 INIT_TRACK_MEMPOOL(my_perl->Imemory_debug_header, my_perl);
14518 PL_body_arenas = NULL;
14519 Zero(&PL_body_roots, 1, PL_body_roots);
14523 PL_sv_arenaroot = NULL;
14525 PL_debug = proto_perl->Idebug;
14527 /* dbargs array probably holds garbage */
14530 PL_compiling = proto_perl->Icompiling;
14532 /* pseudo environmental stuff */
14533 PL_origargc = proto_perl->Iorigargc;
14534 PL_origargv = proto_perl->Iorigargv;
14536 #ifndef NO_TAINT_SUPPORT
14537 /* Set tainting stuff before PerlIO_debug can possibly get called */
14538 PL_tainting = proto_perl->Itainting;
14539 PL_taint_warn = proto_perl->Itaint_warn;
14541 PL_tainting = FALSE;
14542 PL_taint_warn = FALSE;
14545 PL_minus_c = proto_perl->Iminus_c;
14547 PL_localpatches = proto_perl->Ilocalpatches;
14548 PL_splitstr = proto_perl->Isplitstr;
14549 PL_minus_n = proto_perl->Iminus_n;
14550 PL_minus_p = proto_perl->Iminus_p;
14551 PL_minus_l = proto_perl->Iminus_l;
14552 PL_minus_a = proto_perl->Iminus_a;
14553 PL_minus_E = proto_perl->Iminus_E;
14554 PL_minus_F = proto_perl->Iminus_F;
14555 PL_doswitches = proto_perl->Idoswitches;
14556 PL_dowarn = proto_perl->Idowarn;
14557 #ifdef PERL_SAWAMPERSAND
14558 PL_sawampersand = proto_perl->Isawampersand;
14560 PL_unsafe = proto_perl->Iunsafe;
14561 PL_perldb = proto_perl->Iperldb;
14562 PL_perl_destruct_level = proto_perl->Iperl_destruct_level;
14563 PL_exit_flags = proto_perl->Iexit_flags;
14565 /* XXX time(&PL_basetime) when asked for? */
14566 PL_basetime = proto_perl->Ibasetime;
14568 PL_maxsysfd = proto_perl->Imaxsysfd;
14569 PL_statusvalue = proto_perl->Istatusvalue;
14571 PL_statusvalue_vms = proto_perl->Istatusvalue_vms;
14573 PL_statusvalue_posix = proto_perl->Istatusvalue_posix;
14576 /* RE engine related */
14577 PL_regmatch_slab = NULL;
14578 PL_reg_curpm = NULL;
14580 PL_sub_generation = proto_perl->Isub_generation;
14582 /* funky return mechanisms */
14583 PL_forkprocess = proto_perl->Iforkprocess;
14585 /* internal state */
14586 PL_maxo = proto_perl->Imaxo;
14588 PL_main_start = proto_perl->Imain_start;
14589 PL_eval_root = proto_perl->Ieval_root;
14590 PL_eval_start = proto_perl->Ieval_start;
14592 PL_filemode = proto_perl->Ifilemode;
14593 PL_lastfd = proto_perl->Ilastfd;
14594 PL_oldname = proto_perl->Ioldname; /* XXX not quite right */
14597 PL_gensym = proto_perl->Igensym;
14599 PL_laststatval = proto_perl->Ilaststatval;
14600 PL_laststype = proto_perl->Ilaststype;
14603 PL_profiledata = NULL;
14605 PL_generation = proto_perl->Igeneration;
14607 PL_in_clean_objs = proto_perl->Iin_clean_objs;
14608 PL_in_clean_all = proto_perl->Iin_clean_all;
14610 PL_delaymagic_uid = proto_perl->Idelaymagic_uid;
14611 PL_delaymagic_euid = proto_perl->Idelaymagic_euid;
14612 PL_delaymagic_gid = proto_perl->Idelaymagic_gid;
14613 PL_delaymagic_egid = proto_perl->Idelaymagic_egid;
14614 PL_nomemok = proto_perl->Inomemok;
14615 PL_an = proto_perl->Ian;
14616 PL_evalseq = proto_perl->Ievalseq;
14617 PL_origenviron = proto_perl->Iorigenviron; /* XXX not quite right */
14618 PL_origalen = proto_perl->Iorigalen;
14620 PL_sighandlerp = proto_perl->Isighandlerp;
14622 PL_runops = proto_perl->Irunops;
14624 PL_subline = proto_perl->Isubline;
14626 PL_cv_has_eval = proto_perl->Icv_has_eval;
14629 PL_cryptseen = proto_perl->Icryptseen;
14632 #ifdef USE_LOCALE_COLLATE
14633 PL_collation_ix = proto_perl->Icollation_ix;
14634 PL_collation_standard = proto_perl->Icollation_standard;
14635 PL_collxfrm_base = proto_perl->Icollxfrm_base;
14636 PL_collxfrm_mult = proto_perl->Icollxfrm_mult;
14637 #endif /* USE_LOCALE_COLLATE */
14639 #ifdef USE_LOCALE_NUMERIC
14640 PL_numeric_standard = proto_perl->Inumeric_standard;
14641 PL_numeric_local = proto_perl->Inumeric_local;
14642 #endif /* !USE_LOCALE_NUMERIC */
14644 /* Did the locale setup indicate UTF-8? */
14645 PL_utf8locale = proto_perl->Iutf8locale;
14646 PL_in_utf8_CTYPE_locale = proto_perl->Iin_utf8_CTYPE_locale;
14647 /* Unicode features (see perlrun/-C) */
14648 PL_unicode = proto_perl->Iunicode;
14650 /* Pre-5.8 signals control */
14651 PL_signals = proto_perl->Isignals;
14653 /* times() ticks per second */
14654 PL_clocktick = proto_perl->Iclocktick;
14656 /* Recursion stopper for PerlIO_find_layer */
14657 PL_in_load_module = proto_perl->Iin_load_module;
14659 /* sort() routine */
14660 PL_sort_RealCmp = proto_perl->Isort_RealCmp;
14662 /* Not really needed/useful since the reenrant_retint is "volatile",
14663 * but do it for consistency's sake. */
14664 PL_reentrant_retint = proto_perl->Ireentrant_retint;
14666 /* Hooks to shared SVs and locks. */
14667 PL_sharehook = proto_perl->Isharehook;
14668 PL_lockhook = proto_perl->Ilockhook;
14669 PL_unlockhook = proto_perl->Iunlockhook;
14670 PL_threadhook = proto_perl->Ithreadhook;
14671 PL_destroyhook = proto_perl->Idestroyhook;
14672 PL_signalhook = proto_perl->Isignalhook;
14674 PL_globhook = proto_perl->Iglobhook;
14677 PL_last_swash_hv = NULL; /* reinits on demand */
14678 PL_last_swash_klen = 0;
14679 PL_last_swash_key[0]= '\0';
14680 PL_last_swash_tmps = (U8*)NULL;
14681 PL_last_swash_slen = 0;
14683 PL_srand_called = proto_perl->Isrand_called;
14684 Copy(&(proto_perl->Irandom_state), &PL_random_state, 1, PL_RANDOM_STATE_TYPE);
14686 if (flags & CLONEf_COPY_STACKS) {
14687 /* next allocation will be PL_tmps_stack[PL_tmps_ix+1] */
14688 PL_tmps_ix = proto_perl->Itmps_ix;
14689 PL_tmps_max = proto_perl->Itmps_max;
14690 PL_tmps_floor = proto_perl->Itmps_floor;
14692 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
14693 * NOTE: unlike the others! */
14694 PL_scopestack_ix = proto_perl->Iscopestack_ix;
14695 PL_scopestack_max = proto_perl->Iscopestack_max;
14697 /* next SSPUSHFOO() sets PL_savestack[PL_savestack_ix]
14698 * NOTE: unlike the others! */
14699 PL_savestack_ix = proto_perl->Isavestack_ix;
14700 PL_savestack_max = proto_perl->Isavestack_max;
14703 PL_start_env = proto_perl->Istart_env; /* XXXXXX */
14704 PL_top_env = &PL_start_env;
14706 PL_op = proto_perl->Iop;
14709 PL_Xpv = (XPV*)NULL;
14710 my_perl->Ina = proto_perl->Ina;
14712 PL_statbuf = proto_perl->Istatbuf;
14713 PL_statcache = proto_perl->Istatcache;
14715 #ifndef NO_TAINT_SUPPORT
14716 PL_tainted = proto_perl->Itainted;
14718 PL_tainted = FALSE;
14720 PL_curpm = proto_perl->Icurpm; /* XXX No PMOP ref count */
14722 PL_chopset = proto_perl->Ichopset; /* XXX never deallocated */
14724 PL_restartjmpenv = proto_perl->Irestartjmpenv;
14725 PL_restartop = proto_perl->Irestartop;
14726 PL_in_eval = proto_perl->Iin_eval;
14727 PL_delaymagic = proto_perl->Idelaymagic;
14728 PL_phase = proto_perl->Iphase;
14729 PL_localizing = proto_perl->Ilocalizing;
14731 PL_hv_fetch_ent_mh = NULL;
14732 PL_modcount = proto_perl->Imodcount;
14733 PL_lastgotoprobe = NULL;
14734 PL_dumpindent = proto_perl->Idumpindent;
14736 PL_efloatbuf = NULL; /* reinits on demand */
14737 PL_efloatsize = 0; /* reinits on demand */
14741 PL_colorset = 0; /* reinits PL_colors[] */
14742 /*PL_colors[6] = {0,0,0,0,0,0};*/
14744 /* Pluggable optimizer */
14745 PL_peepp = proto_perl->Ipeepp;
14746 PL_rpeepp = proto_perl->Irpeepp;
14747 /* op_free() hook */
14748 PL_opfreehook = proto_perl->Iopfreehook;
14750 #ifdef USE_REENTRANT_API
14751 /* XXX: things like -Dm will segfault here in perlio, but doing
14752 * PERL_SET_CONTEXT(proto_perl);
14753 * breaks too many other things
14755 Perl_reentrant_init(aTHX);
14758 /* create SV map for pointer relocation */
14759 PL_ptr_table = ptr_table_new();
14761 /* initialize these special pointers as early as possible */
14763 ptr_table_store(PL_ptr_table, &proto_perl->Isv_undef, &PL_sv_undef);
14764 ptr_table_store(PL_ptr_table, &proto_perl->Isv_no, &PL_sv_no);
14765 ptr_table_store(PL_ptr_table, &proto_perl->Isv_yes, &PL_sv_yes);
14766 ptr_table_store(PL_ptr_table, &proto_perl->Ipadname_const,
14767 &PL_padname_const);
14769 /* create (a non-shared!) shared string table */
14770 PL_strtab = newHV();
14771 HvSHAREKEYS_off(PL_strtab);
14772 hv_ksplit(PL_strtab, HvTOTALKEYS(proto_perl->Istrtab));
14773 ptr_table_store(PL_ptr_table, proto_perl->Istrtab, PL_strtab);
14775 Zero(PL_sv_consts, SV_CONSTS_COUNT, SV*);
14777 /* This PV will be free'd special way so must set it same way op.c does */
14778 PL_compiling.cop_file = savesharedpv(PL_compiling.cop_file);
14779 ptr_table_store(PL_ptr_table, proto_perl->Icompiling.cop_file, PL_compiling.cop_file);
14781 ptr_table_store(PL_ptr_table, &proto_perl->Icompiling, &PL_compiling);
14782 PL_compiling.cop_warnings = DUP_WARNINGS(PL_compiling.cop_warnings);
14783 CopHINTHASH_set(&PL_compiling, cophh_copy(CopHINTHASH_get(&PL_compiling)));
14784 PL_curcop = (COP*)any_dup(proto_perl->Icurcop, proto_perl);
14786 param->stashes = newAV(); /* Setup array of objects to call clone on */
14787 /* This makes no difference to the implementation, as it always pushes
14788 and shifts pointers to other SVs without changing their reference
14789 count, with the array becoming empty before it is freed. However, it
14790 makes it conceptually clear what is going on, and will avoid some
14791 work inside av.c, filling slots between AvFILL() and AvMAX() with
14792 &PL_sv_undef, and SvREFCNT_dec()ing those. */
14793 AvREAL_off(param->stashes);
14795 if (!(flags & CLONEf_COPY_STACKS)) {
14796 param->unreferenced = newAV();
14799 #ifdef PERLIO_LAYERS
14800 /* Clone PerlIO tables as soon as we can handle general xx_dup() */
14801 PerlIO_clone(aTHX_ proto_perl, param);
14804 PL_envgv = gv_dup_inc(proto_perl->Ienvgv, param);
14805 PL_incgv = gv_dup_inc(proto_perl->Iincgv, param);
14806 PL_hintgv = gv_dup_inc(proto_perl->Ihintgv, param);
14807 PL_origfilename = SAVEPV(proto_perl->Iorigfilename);
14808 PL_xsubfilename = proto_perl->Ixsubfilename;
14809 PL_diehook = sv_dup_inc(proto_perl->Idiehook, param);
14810 PL_warnhook = sv_dup_inc(proto_perl->Iwarnhook, param);
14813 PL_patchlevel = sv_dup_inc(proto_perl->Ipatchlevel, param);
14814 PL_inplace = SAVEPV(proto_perl->Iinplace);
14815 PL_e_script = sv_dup_inc(proto_perl->Ie_script, param);
14817 /* magical thingies */
14819 PL_encoding = sv_dup(proto_perl->Iencoding, param);
14820 PL_lex_encoding = sv_dup(proto_perl->Ilex_encoding, param);
14822 sv_setpvs(PERL_DEBUG_PAD(0), ""); /* For regex debugging. */
14823 sv_setpvs(PERL_DEBUG_PAD(1), ""); /* ext/re needs these */
14824 sv_setpvs(PERL_DEBUG_PAD(2), ""); /* even without DEBUGGING. */
14827 /* Clone the regex array */
14828 /* ORANGE FIXME for plugins, probably in the SV dup code.
14829 newSViv(PTR2IV(CALLREGDUPE(
14830 INT2PTR(REGEXP *, SvIVX(regex)), param))))
14832 PL_regex_padav = av_dup_inc(proto_perl->Iregex_padav, param);
14833 PL_regex_pad = AvARRAY(PL_regex_padav);
14835 PL_stashpadmax = proto_perl->Istashpadmax;
14836 PL_stashpadix = proto_perl->Istashpadix ;
14837 Newx(PL_stashpad, PL_stashpadmax, HV *);
14840 for (; o < PL_stashpadmax; ++o)
14841 PL_stashpad[o] = hv_dup(proto_perl->Istashpad[o], param);
14844 /* shortcuts to various I/O objects */
14845 PL_ofsgv = gv_dup_inc(proto_perl->Iofsgv, param);
14846 PL_stdingv = gv_dup(proto_perl->Istdingv, param);
14847 PL_stderrgv = gv_dup(proto_perl->Istderrgv, param);
14848 PL_defgv = gv_dup(proto_perl->Idefgv, param);
14849 PL_argvgv = gv_dup_inc(proto_perl->Iargvgv, param);
14850 PL_argvoutgv = gv_dup(proto_perl->Iargvoutgv, param);
14851 PL_argvout_stack = av_dup_inc(proto_perl->Iargvout_stack, param);
14853 /* shortcuts to regexp stuff */
14854 PL_replgv = gv_dup_inc(proto_perl->Ireplgv, param);
14856 /* shortcuts to misc objects */
14857 PL_errgv = gv_dup(proto_perl->Ierrgv, param);
14859 /* shortcuts to debugging objects */
14860 PL_DBgv = gv_dup_inc(proto_perl->IDBgv, param);
14861 PL_DBline = gv_dup_inc(proto_perl->IDBline, param);
14862 PL_DBsub = gv_dup_inc(proto_perl->IDBsub, param);
14863 PL_DBsingle = sv_dup(proto_perl->IDBsingle, param);
14864 PL_DBtrace = sv_dup(proto_perl->IDBtrace, param);
14865 PL_DBsignal = sv_dup(proto_perl->IDBsignal, param);
14866 Copy(proto_perl->IDBcontrol, PL_DBcontrol, DBVARMG_COUNT, IV);
14868 /* symbol tables */
14869 PL_defstash = hv_dup_inc(proto_perl->Idefstash, param);
14870 PL_curstash = hv_dup_inc(proto_perl->Icurstash, param);
14871 PL_debstash = hv_dup(proto_perl->Idebstash, param);
14872 PL_globalstash = hv_dup(proto_perl->Iglobalstash, param);
14873 PL_curstname = sv_dup_inc(proto_perl->Icurstname, param);
14875 PL_beginav = av_dup_inc(proto_perl->Ibeginav, param);
14876 PL_beginav_save = av_dup_inc(proto_perl->Ibeginav_save, param);
14877 PL_checkav_save = av_dup_inc(proto_perl->Icheckav_save, param);
14878 PL_unitcheckav = av_dup_inc(proto_perl->Iunitcheckav, param);
14879 PL_unitcheckav_save = av_dup_inc(proto_perl->Iunitcheckav_save, param);
14880 PL_endav = av_dup_inc(proto_perl->Iendav, param);
14881 PL_checkav = av_dup_inc(proto_perl->Icheckav, param);
14882 PL_initav = av_dup_inc(proto_perl->Iinitav, param);
14883 PL_savebegin = proto_perl->Isavebegin;
14885 PL_isarev = hv_dup_inc(proto_perl->Iisarev, param);
14887 /* subprocess state */
14888 PL_fdpid = av_dup_inc(proto_perl->Ifdpid, param);
14890 if (proto_perl->Iop_mask)
14891 PL_op_mask = SAVEPVN(proto_perl->Iop_mask, PL_maxo);
14894 /* PL_asserting = proto_perl->Iasserting; */
14896 /* current interpreter roots */
14897 PL_main_cv = cv_dup_inc(proto_perl->Imain_cv, param);
14899 PL_main_root = OpREFCNT_inc(proto_perl->Imain_root);
14902 /* runtime control stuff */
14903 PL_curcopdb = (COP*)any_dup(proto_perl->Icurcopdb, proto_perl);
14905 PL_preambleav = av_dup_inc(proto_perl->Ipreambleav, param);
14907 PL_ors_sv = sv_dup_inc(proto_perl->Iors_sv, param);
14909 /* interpreter atexit processing */
14910 PL_exitlistlen = proto_perl->Iexitlistlen;
14911 if (PL_exitlistlen) {
14912 Newx(PL_exitlist, PL_exitlistlen, PerlExitListEntry);
14913 Copy(proto_perl->Iexitlist, PL_exitlist, PL_exitlistlen, PerlExitListEntry);
14916 PL_exitlist = (PerlExitListEntry*)NULL;
14918 PL_my_cxt_size = proto_perl->Imy_cxt_size;
14919 if (PL_my_cxt_size) {
14920 Newx(PL_my_cxt_list, PL_my_cxt_size, void *);
14921 Copy(proto_perl->Imy_cxt_list, PL_my_cxt_list, PL_my_cxt_size, void *);
14922 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
14923 Newx(PL_my_cxt_keys, PL_my_cxt_size, const char *);
14924 Copy(proto_perl->Imy_cxt_keys, PL_my_cxt_keys, PL_my_cxt_size, char *);
14928 PL_my_cxt_list = (void**)NULL;
14929 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
14930 PL_my_cxt_keys = (const char**)NULL;
14933 PL_modglobal = hv_dup_inc(proto_perl->Imodglobal, param);
14934 PL_custom_op_names = hv_dup_inc(proto_perl->Icustom_op_names,param);
14935 PL_custom_op_descs = hv_dup_inc(proto_perl->Icustom_op_descs,param);
14936 PL_custom_ops = hv_dup_inc(proto_perl->Icustom_ops, param);
14938 PL_compcv = cv_dup(proto_perl->Icompcv, param);
14940 PAD_CLONE_VARS(proto_perl, param);
14942 #ifdef HAVE_INTERP_INTERN
14943 sys_intern_dup(&proto_perl->Isys_intern, &PL_sys_intern);
14946 PL_DBcv = cv_dup(proto_perl->IDBcv, param);
14948 #ifdef PERL_USES_PL_PIDSTATUS
14949 PL_pidstatus = newHV(); /* XXX flag for cloning? */
14951 PL_osname = SAVEPV(proto_perl->Iosname);
14952 PL_parser = parser_dup(proto_perl->Iparser, param);
14954 /* XXX this only works if the saved cop has already been cloned */
14955 if (proto_perl->Iparser) {
14956 PL_parser->saved_curcop = (COP*)any_dup(
14957 proto_perl->Iparser->saved_curcop,
14961 PL_subname = sv_dup_inc(proto_perl->Isubname, param);
14963 #ifdef USE_LOCALE_CTYPE
14964 /* Should we warn if uses locale? */
14965 PL_warn_locale = sv_dup_inc(proto_perl->Iwarn_locale, param);
14968 #ifdef USE_LOCALE_COLLATE
14969 PL_collation_name = SAVEPV(proto_perl->Icollation_name);
14970 #endif /* USE_LOCALE_COLLATE */
14972 #ifdef USE_LOCALE_NUMERIC
14973 PL_numeric_name = SAVEPV(proto_perl->Inumeric_name);
14974 PL_numeric_radix_sv = sv_dup_inc(proto_perl->Inumeric_radix_sv, param);
14975 #endif /* !USE_LOCALE_NUMERIC */
14977 /* Unicode inversion lists */
14978 PL_Latin1 = sv_dup_inc(proto_perl->ILatin1, param);
14979 PL_UpperLatin1 = sv_dup_inc(proto_perl->IUpperLatin1, param);
14980 PL_AboveLatin1 = sv_dup_inc(proto_perl->IAboveLatin1, param);
14981 PL_InBitmap = sv_dup_inc(proto_perl->IInBitmap, param);
14983 PL_NonL1NonFinalFold = sv_dup_inc(proto_perl->INonL1NonFinalFold, param);
14984 PL_HasMultiCharFold = sv_dup_inc(proto_perl->IHasMultiCharFold, param);
14986 /* utf8 character class swashes */
14987 for (i = 0; i < POSIX_SWASH_COUNT; i++) {
14988 PL_utf8_swash_ptrs[i] = sv_dup_inc(proto_perl->Iutf8_swash_ptrs[i], param);
14990 for (i = 0; i < POSIX_CC_COUNT; i++) {
14991 PL_XPosix_ptrs[i] = sv_dup_inc(proto_perl->IXPosix_ptrs[i], param);
14993 PL_GCB_invlist = sv_dup_inc(proto_perl->IGCB_invlist, param);
14994 PL_SB_invlist = sv_dup_inc(proto_perl->ISB_invlist, param);
14995 PL_WB_invlist = sv_dup_inc(proto_perl->IWB_invlist, param);
14996 PL_utf8_mark = sv_dup_inc(proto_perl->Iutf8_mark, param);
14997 PL_utf8_toupper = sv_dup_inc(proto_perl->Iutf8_toupper, param);
14998 PL_utf8_totitle = sv_dup_inc(proto_perl->Iutf8_totitle, param);
14999 PL_utf8_tolower = sv_dup_inc(proto_perl->Iutf8_tolower, param);
15000 PL_utf8_tofold = sv_dup_inc(proto_perl->Iutf8_tofold, param);
15001 PL_utf8_idstart = sv_dup_inc(proto_perl->Iutf8_idstart, param);
15002 PL_utf8_xidstart = sv_dup_inc(proto_perl->Iutf8_xidstart, param);
15003 PL_utf8_perl_idstart = sv_dup_inc(proto_perl->Iutf8_perl_idstart, param);
15004 PL_utf8_perl_idcont = sv_dup_inc(proto_perl->Iutf8_perl_idcont, param);
15005 PL_utf8_idcont = sv_dup_inc(proto_perl->Iutf8_idcont, param);
15006 PL_utf8_xidcont = sv_dup_inc(proto_perl->Iutf8_xidcont, param);
15007 PL_utf8_foldable = sv_dup_inc(proto_perl->Iutf8_foldable, param);
15008 PL_utf8_charname_begin = sv_dup_inc(proto_perl->Iutf8_charname_begin, param);
15009 PL_utf8_charname_continue = sv_dup_inc(proto_perl->Iutf8_charname_continue, param);
15011 if (proto_perl->Ipsig_pend) {
15012 Newxz(PL_psig_pend, SIG_SIZE, int);
15015 PL_psig_pend = (int*)NULL;
15018 if (proto_perl->Ipsig_name) {
15019 Newx(PL_psig_name, 2 * SIG_SIZE, SV*);
15020 sv_dup_inc_multiple(proto_perl->Ipsig_name, PL_psig_name, 2 * SIG_SIZE,
15022 PL_psig_ptr = PL_psig_name + SIG_SIZE;
15025 PL_psig_ptr = (SV**)NULL;
15026 PL_psig_name = (SV**)NULL;
15029 if (flags & CLONEf_COPY_STACKS) {
15030 Newx(PL_tmps_stack, PL_tmps_max, SV*);
15031 sv_dup_inc_multiple(proto_perl->Itmps_stack, PL_tmps_stack,
15032 PL_tmps_ix+1, param);
15034 /* next PUSHMARK() sets *(PL_markstack_ptr+1) */
15035 i = proto_perl->Imarkstack_max - proto_perl->Imarkstack;
15036 Newxz(PL_markstack, i, I32);
15037 PL_markstack_max = PL_markstack + (proto_perl->Imarkstack_max
15038 - proto_perl->Imarkstack);
15039 PL_markstack_ptr = PL_markstack + (proto_perl->Imarkstack_ptr
15040 - proto_perl->Imarkstack);
15041 Copy(proto_perl->Imarkstack, PL_markstack,
15042 PL_markstack_ptr - PL_markstack + 1, I32);
15044 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
15045 * NOTE: unlike the others! */
15046 Newxz(PL_scopestack, PL_scopestack_max, I32);
15047 Copy(proto_perl->Iscopestack, PL_scopestack, PL_scopestack_ix, I32);
15050 Newxz(PL_scopestack_name, PL_scopestack_max, const char *);
15051 Copy(proto_perl->Iscopestack_name, PL_scopestack_name, PL_scopestack_ix, const char *);
15053 /* reset stack AV to correct length before its duped via
15054 * PL_curstackinfo */
15055 AvFILLp(proto_perl->Icurstack) =
15056 proto_perl->Istack_sp - proto_perl->Istack_base;
15058 /* NOTE: si_dup() looks at PL_markstack */
15059 PL_curstackinfo = si_dup(proto_perl->Icurstackinfo, param);
15061 /* PL_curstack = PL_curstackinfo->si_stack; */
15062 PL_curstack = av_dup(proto_perl->Icurstack, param);
15063 PL_mainstack = av_dup(proto_perl->Imainstack, param);
15065 /* next PUSHs() etc. set *(PL_stack_sp+1) */
15066 PL_stack_base = AvARRAY(PL_curstack);
15067 PL_stack_sp = PL_stack_base + (proto_perl->Istack_sp
15068 - proto_perl->Istack_base);
15069 PL_stack_max = PL_stack_base + AvMAX(PL_curstack);
15071 /*Newxz(PL_savestack, PL_savestack_max, ANY);*/
15072 PL_savestack = ss_dup(proto_perl, param);
15076 ENTER; /* perl_destruct() wants to LEAVE; */
15079 PL_statgv = gv_dup(proto_perl->Istatgv, param);
15080 PL_statname = sv_dup_inc(proto_perl->Istatname, param);
15082 PL_rs = sv_dup_inc(proto_perl->Irs, param);
15083 PL_last_in_gv = gv_dup(proto_perl->Ilast_in_gv, param);
15084 PL_defoutgv = gv_dup_inc(proto_perl->Idefoutgv, param);
15085 PL_toptarget = sv_dup_inc(proto_perl->Itoptarget, param);
15086 PL_bodytarget = sv_dup_inc(proto_perl->Ibodytarget, param);
15087 PL_formtarget = sv_dup(proto_perl->Iformtarget, param);
15089 PL_errors = sv_dup_inc(proto_perl->Ierrors, param);
15091 PL_sortcop = (OP*)any_dup(proto_perl->Isortcop, proto_perl);
15092 PL_firstgv = gv_dup_inc(proto_perl->Ifirstgv, param);
15093 PL_secondgv = gv_dup_inc(proto_perl->Isecondgv, param);
15095 PL_stashcache = newHV();
15097 PL_watchaddr = (char **) ptr_table_fetch(PL_ptr_table,
15098 proto_perl->Iwatchaddr);
15099 PL_watchok = PL_watchaddr ? * PL_watchaddr : NULL;
15100 if (PL_debug && PL_watchaddr) {
15101 PerlIO_printf(Perl_debug_log,
15102 "WATCHING: %"UVxf" cloned as %"UVxf" with value %"UVxf"\n",
15103 PTR2UV(proto_perl->Iwatchaddr), PTR2UV(PL_watchaddr),
15104 PTR2UV(PL_watchok));
15107 PL_registered_mros = hv_dup_inc(proto_perl->Iregistered_mros, param);
15108 PL_blockhooks = av_dup_inc(proto_perl->Iblockhooks, param);
15109 PL_utf8_foldclosures = hv_dup_inc(proto_perl->Iutf8_foldclosures, param);
15111 /* Call the ->CLONE method, if it exists, for each of the stashes
15112 identified by sv_dup() above.
15114 while(av_tindex(param->stashes) != -1) {
15115 HV* const stash = MUTABLE_HV(av_shift(param->stashes));
15116 GV* const cloner = gv_fetchmethod_autoload(stash, "CLONE", 0);
15117 if (cloner && GvCV(cloner)) {
15122 mXPUSHs(newSVhek(HvNAME_HEK(stash)));
15124 call_sv(MUTABLE_SV(GvCV(cloner)), G_DISCARD);
15130 if (!(flags & CLONEf_KEEP_PTR_TABLE)) {
15131 ptr_table_free(PL_ptr_table);
15132 PL_ptr_table = NULL;
15135 if (!(flags & CLONEf_COPY_STACKS)) {
15136 unreferenced_to_tmp_stack(param->unreferenced);
15139 SvREFCNT_dec(param->stashes);
15141 /* orphaned? eg threads->new inside BEGIN or use */
15142 if (PL_compcv && ! SvREFCNT(PL_compcv)) {
15143 SvREFCNT_inc_simple_void(PL_compcv);
15144 SAVEFREESV(PL_compcv);
15151 S_unreferenced_to_tmp_stack(pTHX_ AV *const unreferenced)
15153 PERL_ARGS_ASSERT_UNREFERENCED_TO_TMP_STACK;
15155 if (AvFILLp(unreferenced) > -1) {
15156 SV **svp = AvARRAY(unreferenced);
15157 SV **const last = svp + AvFILLp(unreferenced);
15161 if (SvREFCNT(*svp) == 1)
15163 } while (++svp <= last);
15165 EXTEND_MORTAL(count);
15166 svp = AvARRAY(unreferenced);
15169 if (SvREFCNT(*svp) == 1) {
15170 /* Our reference is the only one to this SV. This means that
15171 in this thread, the scalar effectively has a 0 reference.
15172 That doesn't work (cleanup never happens), so donate our
15173 reference to it onto the save stack. */
15174 PL_tmps_stack[++PL_tmps_ix] = *svp;
15176 /* As an optimisation, because we are already walking the
15177 entire array, instead of above doing either
15178 SvREFCNT_inc(*svp) or *svp = &PL_sv_undef, we can instead
15179 release our reference to the scalar, so that at the end of
15180 the array owns zero references to the scalars it happens to
15181 point to. We are effectively converting the array from
15182 AvREAL() on to AvREAL() off. This saves the av_clear()
15183 (triggered by the SvREFCNT_dec(unreferenced) below) from
15184 walking the array a second time. */
15185 SvREFCNT_dec(*svp);
15188 } while (++svp <= last);
15189 AvREAL_off(unreferenced);
15191 SvREFCNT_dec_NN(unreferenced);
15195 Perl_clone_params_del(CLONE_PARAMS *param)
15197 /* This seemingly funky ordering keeps the build with PERL_GLOBAL_STRUCT
15199 PerlInterpreter *const to = param->new_perl;
15201 PerlInterpreter *const was = PERL_GET_THX;
15203 PERL_ARGS_ASSERT_CLONE_PARAMS_DEL;
15209 SvREFCNT_dec(param->stashes);
15210 if (param->unreferenced)
15211 unreferenced_to_tmp_stack(param->unreferenced);
15221 Perl_clone_params_new(PerlInterpreter *const from, PerlInterpreter *const to)
15224 /* Need to play this game, as newAV() can call safesysmalloc(), and that
15225 does a dTHX; to get the context from thread local storage.
15226 FIXME - under PERL_CORE Newx(), Safefree() and friends should expand to
15227 a version that passes in my_perl. */
15228 PerlInterpreter *const was = PERL_GET_THX;
15229 CLONE_PARAMS *param;
15231 PERL_ARGS_ASSERT_CLONE_PARAMS_NEW;
15237 /* Given that we've set the context, we can do this unshared. */
15238 Newx(param, 1, CLONE_PARAMS);
15241 param->proto_perl = from;
15242 param->new_perl = to;
15243 param->stashes = (AV *)Perl_newSV_type(to, SVt_PVAV);
15244 AvREAL_off(param->stashes);
15245 param->unreferenced = (AV *)Perl_newSV_type(to, SVt_PVAV);
15253 #endif /* USE_ITHREADS */
15256 Perl_init_constants(pTHX)
15258 SvREFCNT(&PL_sv_undef) = SvREFCNT_IMMORTAL;
15259 SvFLAGS(&PL_sv_undef) = SVf_READONLY|SVf_PROTECT|SVt_NULL;
15260 SvANY(&PL_sv_undef) = NULL;
15262 SvANY(&PL_sv_no) = new_XPVNV();
15263 SvREFCNT(&PL_sv_no) = SvREFCNT_IMMORTAL;
15264 SvFLAGS(&PL_sv_no) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15265 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15268 SvANY(&PL_sv_yes) = new_XPVNV();
15269 SvREFCNT(&PL_sv_yes) = SvREFCNT_IMMORTAL;
15270 SvFLAGS(&PL_sv_yes) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15271 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15274 SvPV_set(&PL_sv_no, (char*)PL_No);
15275 SvCUR_set(&PL_sv_no, 0);
15276 SvLEN_set(&PL_sv_no, 0);
15277 SvIV_set(&PL_sv_no, 0);
15278 SvNV_set(&PL_sv_no, 0);
15280 SvPV_set(&PL_sv_yes, (char*)PL_Yes);
15281 SvCUR_set(&PL_sv_yes, 1);
15282 SvLEN_set(&PL_sv_yes, 0);
15283 SvIV_set(&PL_sv_yes, 1);
15284 SvNV_set(&PL_sv_yes, 1);
15286 PadnamePV(&PL_padname_const) = (char *)PL_No;
15290 =head1 Unicode Support
15292 =for apidoc sv_recode_to_utf8
15294 C<encoding> is assumed to be an C<Encode> object, on entry the PV
15295 of C<sv> is assumed to be octets in that encoding, and C<sv>
15296 will be converted into Unicode (and UTF-8).
15298 If C<sv> already is UTF-8 (or if it is not C<POK>), or if C<encoding>
15299 is not a reference, nothing is done to C<sv>. If C<encoding> is not
15300 an C<Encode::XS> Encoding object, bad things will happen.
15301 (See F<lib/encoding.pm> and L<Encode>.)
15303 The PV of C<sv> is returned.
15308 Perl_sv_recode_to_utf8(pTHX_ SV *sv, SV *encoding)
15310 PERL_ARGS_ASSERT_SV_RECODE_TO_UTF8;
15312 if (SvPOK(sv) && !SvUTF8(sv) && !IN_BYTES && SvROK(encoding)) {
15321 if (SvPADTMP(nsv)) {
15322 nsv = sv_newmortal();
15323 SvSetSV_nosteal(nsv, sv);
15332 Passing sv_yes is wrong - it needs to be or'ed set of constants
15333 for Encode::XS, while UTf-8 decode (currently) assumes a true value means
15334 remove converted chars from source.
15336 Both will default the value - let them.
15338 XPUSHs(&PL_sv_yes);
15341 call_method("decode", G_SCALAR);
15345 s = SvPV_const(uni, len);
15346 if (s != SvPVX_const(sv)) {
15347 SvGROW(sv, len + 1);
15348 Move(s, SvPVX(sv), len + 1, char);
15349 SvCUR_set(sv, len);
15354 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
15355 /* clear pos and any utf8 cache */
15356 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
15359 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
15360 magic_setutf8(sv,mg); /* clear UTF8 cache */
15365 return SvPOKp(sv) ? SvPVX(sv) : NULL;
15369 =for apidoc sv_cat_decode
15371 C<encoding> is assumed to be an C<Encode> object, the PV of C<ssv> is
15372 assumed to be octets in that encoding and decoding the input starts
15373 from the position which S<C<(PV + *offset)>> pointed to. C<dsv> will be
15374 concatenated with the decoded UTF-8 string from C<ssv>. Decoding will terminate
15375 when the string C<tstr> appears in decoding output or the input ends on
15376 the PV of C<ssv>. The value which C<offset> points will be modified
15377 to the last input position on C<ssv>.
15379 Returns TRUE if the terminator was found, else returns FALSE.
15384 Perl_sv_cat_decode(pTHX_ SV *dsv, SV *encoding,
15385 SV *ssv, int *offset, char *tstr, int tlen)
15389 PERL_ARGS_ASSERT_SV_CAT_DECODE;
15391 if (SvPOK(ssv) && SvPOK(dsv) && SvROK(encoding)) {
15402 offsv = newSViv(*offset);
15404 mPUSHp(tstr, tlen);
15406 call_method("cat_decode", G_SCALAR);
15408 ret = SvTRUE(TOPs);
15409 *offset = SvIV(offsv);
15415 Perl_croak(aTHX_ "Invalid argument to sv_cat_decode");
15420 /* ---------------------------------------------------------------------
15422 * support functions for report_uninit()
15425 /* the maxiumum size of array or hash where we will scan looking
15426 * for the undefined element that triggered the warning */
15428 #define FUV_MAX_SEARCH_SIZE 1000
15430 /* Look for an entry in the hash whose value has the same SV as val;
15431 * If so, return a mortal copy of the key. */
15434 S_find_hash_subscript(pTHX_ const HV *const hv, const SV *const val)
15440 PERL_ARGS_ASSERT_FIND_HASH_SUBSCRIPT;
15442 if (!hv || SvMAGICAL(hv) || !HvARRAY(hv) ||
15443 (HvTOTALKEYS(hv) > FUV_MAX_SEARCH_SIZE))
15446 array = HvARRAY(hv);
15448 for (i=HvMAX(hv); i>=0; i--) {
15450 for (entry = array[i]; entry; entry = HeNEXT(entry)) {
15451 if (HeVAL(entry) != val)
15453 if ( HeVAL(entry) == &PL_sv_undef ||
15454 HeVAL(entry) == &PL_sv_placeholder)
15458 if (HeKLEN(entry) == HEf_SVKEY)
15459 return sv_mortalcopy(HeKEY_sv(entry));
15460 return sv_2mortal(newSVhek(HeKEY_hek(entry)));
15466 /* Look for an entry in the array whose value has the same SV as val;
15467 * If so, return the index, otherwise return -1. */
15470 S_find_array_subscript(pTHX_ const AV *const av, const SV *const val)
15472 PERL_ARGS_ASSERT_FIND_ARRAY_SUBSCRIPT;
15474 if (!av || SvMAGICAL(av) || !AvARRAY(av) ||
15475 (AvFILLp(av) > FUV_MAX_SEARCH_SIZE))
15478 if (val != &PL_sv_undef) {
15479 SV ** const svp = AvARRAY(av);
15482 for (i=AvFILLp(av); i>=0; i--)
15489 /* varname(): return the name of a variable, optionally with a subscript.
15490 * If gv is non-zero, use the name of that global, along with gvtype (one
15491 * of "$", "@", "%"); otherwise use the name of the lexical at pad offset
15492 * targ. Depending on the value of the subscript_type flag, return:
15495 #define FUV_SUBSCRIPT_NONE 1 /* "@foo" */
15496 #define FUV_SUBSCRIPT_ARRAY 2 /* "$foo[aindex]" */
15497 #define FUV_SUBSCRIPT_HASH 3 /* "$foo{keyname}" */
15498 #define FUV_SUBSCRIPT_WITHIN 4 /* "within @foo" */
15501 Perl_varname(pTHX_ const GV *const gv, const char gvtype, PADOFFSET targ,
15502 const SV *const keyname, I32 aindex, int subscript_type)
15505 SV * const name = sv_newmortal();
15506 if (gv && isGV(gv)) {
15508 buffer[0] = gvtype;
15511 /* as gv_fullname4(), but add literal '^' for $^FOO names */
15513 gv_fullname4(name, gv, buffer, 0);
15515 if ((unsigned int)SvPVX(name)[1] <= 26) {
15517 buffer[1] = SvPVX(name)[1] + 'A' - 1;
15519 /* Swap the 1 unprintable control character for the 2 byte pretty
15520 version - ie substr($name, 1, 1) = $buffer; */
15521 sv_insert(name, 1, 1, buffer, 2);
15525 CV * const cv = gv ? ((CV *)gv) : find_runcv(NULL);
15528 assert(!cv || SvTYPE(cv) == SVt_PVCV || SvTYPE(cv) == SVt_PVFM);
15530 if (!cv || !CvPADLIST(cv))
15532 sv = padnamelist_fetch(PadlistNAMES(CvPADLIST(cv)), targ);
15533 sv_setpvn(name, PadnamePV(sv), PadnameLEN(sv));
15537 if (subscript_type == FUV_SUBSCRIPT_HASH) {
15538 SV * const sv = newSV(0);
15539 *SvPVX(name) = '$';
15540 Perl_sv_catpvf(aTHX_ name, "{%s}",
15541 pv_pretty(sv, SvPVX_const(keyname), SvCUR(keyname), 32, NULL, NULL,
15542 PERL_PV_PRETTY_DUMP | PERL_PV_ESCAPE_UNI_DETECT ));
15543 SvREFCNT_dec_NN(sv);
15545 else if (subscript_type == FUV_SUBSCRIPT_ARRAY) {
15546 *SvPVX(name) = '$';
15547 Perl_sv_catpvf(aTHX_ name, "[%"IVdf"]", (IV)aindex);
15549 else if (subscript_type == FUV_SUBSCRIPT_WITHIN) {
15550 /* We know that name has no magic, so can use 0 instead of SV_GMAGIC */
15551 Perl_sv_insert_flags(aTHX_ name, 0, 0, STR_WITH_LEN("within "), 0);
15559 =for apidoc find_uninit_var
15561 Find the name of the undefined variable (if any) that caused the operator
15562 to issue a "Use of uninitialized value" warning.
15563 If match is true, only return a name if its value matches C<uninit_sv>.
15564 So roughly speaking, if a unary operator (such as C<OP_COS>) generates a
15565 warning, then following the direct child of the op may yield an
15566 C<OP_PADSV> or C<OP_GV> that gives the name of the undefined variable. On the
15567 other hand, with C<OP_ADD> there are two branches to follow, so we only print
15568 the variable name if we get an exact match.
15569 C<desc_p> points to a string pointer holding the description of the op.
15570 This may be updated if needed.
15572 The name is returned as a mortal SV.
15574 Assumes that C<PL_op> is the OP that originally triggered the error, and that
15575 C<PL_comppad>/C<PL_curpad> points to the currently executing pad.
15581 S_find_uninit_var(pTHX_ const OP *const obase, const SV *const uninit_sv,
15582 bool match, const char **desc_p)
15587 const OP *o, *o2, *kid;
15589 PERL_ARGS_ASSERT_FIND_UNINIT_VAR;
15591 if (!obase || (match && (!uninit_sv || uninit_sv == &PL_sv_undef ||
15592 uninit_sv == &PL_sv_placeholder)))
15595 switch (obase->op_type) {
15602 const bool pad = ( obase->op_type == OP_PADAV
15603 || obase->op_type == OP_PADHV
15604 || obase->op_type == OP_PADRANGE
15607 const bool hash = ( obase->op_type == OP_PADHV
15608 || obase->op_type == OP_RV2HV
15609 || (obase->op_type == OP_PADRANGE
15610 && SvTYPE(PAD_SVl(obase->op_targ)) == SVt_PVHV)
15614 int subscript_type = FUV_SUBSCRIPT_WITHIN;
15616 if (pad) { /* @lex, %lex */
15617 sv = PAD_SVl(obase->op_targ);
15621 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
15622 /* @global, %global */
15623 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
15626 sv = hash ? MUTABLE_SV(GvHV(gv)): MUTABLE_SV(GvAV(gv));
15628 else if (obase == PL_op) /* @{expr}, %{expr} */
15629 return find_uninit_var(cUNOPx(obase)->op_first,
15630 uninit_sv, match, desc_p);
15631 else /* @{expr}, %{expr} as a sub-expression */
15635 /* attempt to find a match within the aggregate */
15637 keysv = find_hash_subscript((const HV*)sv, uninit_sv);
15639 subscript_type = FUV_SUBSCRIPT_HASH;
15642 index = find_array_subscript((const AV *)sv, uninit_sv);
15644 subscript_type = FUV_SUBSCRIPT_ARRAY;
15647 if (match && subscript_type == FUV_SUBSCRIPT_WITHIN)
15650 return varname(gv, (char)(hash ? '%' : '@'), obase->op_targ,
15651 keysv, index, subscript_type);
15655 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
15657 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
15658 if (!gv || !GvSTASH(gv))
15660 if (match && (GvSV(gv) != uninit_sv))
15662 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
15665 return find_uninit_var(cUNOPx(obase)->op_first, uninit_sv, 1, desc_p);
15668 if (match && PAD_SVl(obase->op_targ) != uninit_sv)
15670 return varname(NULL, '$', obase->op_targ,
15671 NULL, 0, FUV_SUBSCRIPT_NONE);
15674 gv = cGVOPx_gv(obase);
15675 if (!gv || (match && GvSV(gv) != uninit_sv) || !GvSTASH(gv))
15677 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
15679 case OP_AELEMFAST_LEX:
15682 AV *av = MUTABLE_AV(PAD_SV(obase->op_targ));
15683 if (!av || SvRMAGICAL(av))
15685 svp = av_fetch(av, (I8)obase->op_private, FALSE);
15686 if (!svp || *svp != uninit_sv)
15689 return varname(NULL, '$', obase->op_targ,
15690 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
15693 gv = cGVOPx_gv(obase);
15698 AV *const av = GvAV(gv);
15699 if (!av || SvRMAGICAL(av))
15701 svp = av_fetch(av, (I8)obase->op_private, FALSE);
15702 if (!svp || *svp != uninit_sv)
15705 return varname(gv, '$', 0,
15706 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
15708 NOT_REACHED; /* NOTREACHED */
15711 o = cUNOPx(obase)->op_first;
15712 if (!o || o->op_type != OP_NULL ||
15713 ! (o->op_targ == OP_AELEM || o->op_targ == OP_HELEM))
15715 return find_uninit_var(cBINOPo->op_last, uninit_sv, match, desc_p);
15720 bool negate = FALSE;
15722 if (PL_op == obase)
15723 /* $a[uninit_expr] or $h{uninit_expr} */
15724 return find_uninit_var(cBINOPx(obase)->op_last,
15725 uninit_sv, match, desc_p);
15728 o = cBINOPx(obase)->op_first;
15729 kid = cBINOPx(obase)->op_last;
15731 /* get the av or hv, and optionally the gv */
15733 if (o->op_type == OP_PADAV || o->op_type == OP_PADHV) {
15734 sv = PAD_SV(o->op_targ);
15736 else if ((o->op_type == OP_RV2AV || o->op_type == OP_RV2HV)
15737 && cUNOPo->op_first->op_type == OP_GV)
15739 gv = cGVOPx_gv(cUNOPo->op_first);
15743 == OP_RV2HV ? MUTABLE_SV(GvHV(gv)) : MUTABLE_SV(GvAV(gv));
15748 if (kid && kid->op_type == OP_NEGATE) {
15750 kid = cUNOPx(kid)->op_first;
15753 if (kid && kid->op_type == OP_CONST && SvOK(cSVOPx_sv(kid))) {
15754 /* index is constant */
15757 kidsv = newSVpvs_flags("-", SVs_TEMP);
15758 sv_catsv(kidsv, cSVOPx_sv(kid));
15761 kidsv = cSVOPx_sv(kid);
15765 if (obase->op_type == OP_HELEM) {
15766 HE* he = hv_fetch_ent(MUTABLE_HV(sv), kidsv, 0, 0);
15767 if (!he || HeVAL(he) != uninit_sv)
15771 SV * const opsv = cSVOPx_sv(kid);
15772 const IV opsviv = SvIV(opsv);
15773 SV * const * const svp = av_fetch(MUTABLE_AV(sv),
15774 negate ? - opsviv : opsviv,
15776 if (!svp || *svp != uninit_sv)
15780 if (obase->op_type == OP_HELEM)
15781 return varname(gv, '%', o->op_targ,
15782 kidsv, 0, FUV_SUBSCRIPT_HASH);
15784 return varname(gv, '@', o->op_targ, NULL,
15785 negate ? - SvIV(cSVOPx_sv(kid)) : SvIV(cSVOPx_sv(kid)),
15786 FUV_SUBSCRIPT_ARRAY);
15789 /* index is an expression;
15790 * attempt to find a match within the aggregate */
15791 if (obase->op_type == OP_HELEM) {
15792 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
15794 return varname(gv, '%', o->op_targ,
15795 keysv, 0, FUV_SUBSCRIPT_HASH);
15799 = find_array_subscript((const AV *)sv, uninit_sv);
15801 return varname(gv, '@', o->op_targ,
15802 NULL, index, FUV_SUBSCRIPT_ARRAY);
15807 (char)((o->op_type == OP_PADAV || o->op_type == OP_RV2AV)
15809 o->op_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
15811 NOT_REACHED; /* NOTREACHED */
15814 case OP_MULTIDEREF: {
15815 /* If we were executing OP_MULTIDEREF when the undef warning
15816 * triggered, then it must be one of the index values within
15817 * that triggered it. If not, then the only possibility is that
15818 * the value retrieved by the last aggregate lookup might be the
15819 * culprit. For the former, we set PL_multideref_pc each time before
15820 * using an index, so work though the item list until we reach
15821 * that point. For the latter, just work through the entire item
15822 * list; the last aggregate retrieved will be the candidate.
15825 /* the named aggregate, if any */
15826 PADOFFSET agg_targ = 0;
15828 /* the last-seen index */
15830 PADOFFSET index_targ;
15832 IV index_const_iv = 0; /* init for spurious compiler warn */
15833 SV *index_const_sv;
15834 int depth = 0; /* how many array/hash lookups we've done */
15836 UNOP_AUX_item *items = cUNOP_AUXx(obase)->op_aux;
15837 UNOP_AUX_item *last = NULL;
15838 UV actions = items->uv;
15841 if (PL_op == obase) {
15842 last = PL_multideref_pc;
15843 assert(last >= items && last <= items + items[-1].uv);
15850 switch (actions & MDEREF_ACTION_MASK) {
15852 case MDEREF_reload:
15853 actions = (++items)->uv;
15856 case MDEREF_HV_padhv_helem: /* $lex{...} */
15859 case MDEREF_AV_padav_aelem: /* $lex[...] */
15860 agg_targ = (++items)->pad_offset;
15864 case MDEREF_HV_gvhv_helem: /* $pkg{...} */
15867 case MDEREF_AV_gvav_aelem: /* $pkg[...] */
15869 agg_gv = (GV*)UNOP_AUX_item_sv(++items);
15870 assert(isGV_with_GP(agg_gv));
15873 case MDEREF_HV_gvsv_vivify_rv2hv_helem: /* $pkg->{...} */
15874 case MDEREF_HV_padsv_vivify_rv2hv_helem: /* $lex->{...} */
15877 case MDEREF_HV_pop_rv2hv_helem: /* expr->{...} */
15878 case MDEREF_HV_vivify_rv2hv_helem: /* vivify, ->{...} */
15884 case MDEREF_AV_gvsv_vivify_rv2av_aelem: /* $pkg->[...] */
15885 case MDEREF_AV_padsv_vivify_rv2av_aelem: /* $lex->[...] */
15888 case MDEREF_AV_pop_rv2av_aelem: /* expr->[...] */
15889 case MDEREF_AV_vivify_rv2av_aelem: /* vivify, ->[...] */
15896 index_const_sv = NULL;
15898 index_type = (actions & MDEREF_INDEX_MASK);
15899 switch (index_type) {
15900 case MDEREF_INDEX_none:
15902 case MDEREF_INDEX_const:
15904 index_const_sv = UNOP_AUX_item_sv(++items)
15906 index_const_iv = (++items)->iv;
15908 case MDEREF_INDEX_padsv:
15909 index_targ = (++items)->pad_offset;
15911 case MDEREF_INDEX_gvsv:
15912 index_gv = (GV*)UNOP_AUX_item_sv(++items);
15913 assert(isGV_with_GP(index_gv));
15917 if (index_type != MDEREF_INDEX_none)
15920 if ( index_type == MDEREF_INDEX_none
15921 || (actions & MDEREF_FLAG_last)
15922 || (last && items == last)
15926 actions >>= MDEREF_SHIFT;
15929 if (PL_op == obase) {
15930 /* index was undef */
15932 *desc_p = ( (actions & MDEREF_FLAG_last)
15933 && (obase->op_private
15934 & (OPpMULTIDEREF_EXISTS|OPpMULTIDEREF_DELETE)))
15936 (obase->op_private & OPpMULTIDEREF_EXISTS)
15939 : is_hv ? "hash element" : "array element";
15940 assert(index_type != MDEREF_INDEX_none);
15942 return varname(index_gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
15944 return varname(NULL, '$', index_targ,
15945 NULL, 0, FUV_SUBSCRIPT_NONE);
15946 assert(is_hv); /* AV index is an IV and can't be undef */
15947 /* can a const HV index ever be undef? */
15951 /* the SV returned by pp_multideref() was undef, if anything was */
15957 sv = PAD_SV(agg_targ);
15959 sv = is_hv ? MUTABLE_SV(GvHV(agg_gv)) : MUTABLE_SV(GvAV(agg_gv));
15963 if (index_type == MDEREF_INDEX_const) {
15968 HE* he = hv_fetch_ent(MUTABLE_HV(sv), index_const_sv, 0, 0);
15969 if (!he || HeVAL(he) != uninit_sv)
15973 SV * const * const svp =
15974 av_fetch(MUTABLE_AV(sv), index_const_iv, FALSE);
15975 if (!svp || *svp != uninit_sv)
15980 ? varname(agg_gv, '%', agg_targ,
15981 index_const_sv, 0, FUV_SUBSCRIPT_HASH)
15982 : varname(agg_gv, '@', agg_targ,
15983 NULL, index_const_iv, FUV_SUBSCRIPT_ARRAY);
15986 /* index is an var */
15988 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
15990 return varname(agg_gv, '%', agg_targ,
15991 keysv, 0, FUV_SUBSCRIPT_HASH);
15995 = find_array_subscript((const AV *)sv, uninit_sv);
15997 return varname(agg_gv, '@', agg_targ,
15998 NULL, index, FUV_SUBSCRIPT_ARRAY);
16002 return varname(agg_gv,
16004 agg_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
16006 NOT_REACHED; /* NOTREACHED */
16010 /* only examine RHS */
16011 return find_uninit_var(cBINOPx(obase)->op_first, uninit_sv,
16015 o = cUNOPx(obase)->op_first;
16016 if ( o->op_type == OP_PUSHMARK
16017 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)
16021 if (!OpHAS_SIBLING(o)) {
16022 /* one-arg version of open is highly magical */
16024 if (o->op_type == OP_GV) { /* open FOO; */
16026 if (match && GvSV(gv) != uninit_sv)
16028 return varname(gv, '$', 0,
16029 NULL, 0, FUV_SUBSCRIPT_NONE);
16031 /* other possibilities not handled are:
16032 * open $x; or open my $x; should return '${*$x}'
16033 * open expr; should return '$'.expr ideally
16039 /* ops where $_ may be an implicit arg */
16044 if ( !(obase->op_flags & OPf_STACKED)) {
16045 if (uninit_sv == DEFSV)
16046 return newSVpvs_flags("$_", SVs_TEMP);
16047 else if (obase->op_targ
16048 && uninit_sv == PAD_SVl(obase->op_targ))
16049 return varname(NULL, '$', obase->op_targ, NULL, 0,
16050 FUV_SUBSCRIPT_NONE);
16057 match = 1; /* print etc can return undef on defined args */
16058 /* skip filehandle as it can't produce 'undef' warning */
16059 o = cUNOPx(obase)->op_first;
16060 if ((obase->op_flags & OPf_STACKED)
16062 ( o->op_type == OP_PUSHMARK
16063 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)))
16064 o = OpSIBLING(OpSIBLING(o));
16068 case OP_ENTEREVAL: /* could be eval $undef or $x='$undef'; eval $x */
16069 case OP_CUSTOM: /* XS or custom code could trigger random warnings */
16071 /* the following ops are capable of returning PL_sv_undef even for
16072 * defined arg(s) */
16091 case OP_GETPEERNAME:
16139 case OP_SMARTMATCH:
16148 /* XXX tmp hack: these two may call an XS sub, and currently
16149 XS subs don't have a SUB entry on the context stack, so CV and
16150 pad determination goes wrong, and BAD things happen. So, just
16151 don't try to determine the value under those circumstances.
16152 Need a better fix at dome point. DAPM 11/2007 */
16158 GV * const gv = gv_fetchpvs(".", GV_NOTQUAL, SVt_PV);
16159 if (gv && GvSV(gv) == uninit_sv)
16160 return newSVpvs_flags("$.", SVs_TEMP);
16165 /* def-ness of rval pos() is independent of the def-ness of its arg */
16166 if ( !(obase->op_flags & OPf_MOD))
16171 if (SvROK(PL_rs) && uninit_sv == SvRV(PL_rs))
16172 return newSVpvs_flags("${$/}", SVs_TEMP);
16177 if (!(obase->op_flags & OPf_KIDS))
16179 o = cUNOPx(obase)->op_first;
16185 /* This loop checks all the kid ops, skipping any that cannot pos-
16186 * sibly be responsible for the uninitialized value; i.e., defined
16187 * constants and ops that return nothing. If there is only one op
16188 * left that is not skipped, then we *know* it is responsible for
16189 * the uninitialized value. If there is more than one op left, we
16190 * have to look for an exact match in the while() loop below.
16191 * Note that we skip padrange, because the individual pad ops that
16192 * it replaced are still in the tree, so we work on them instead.
16195 for (kid=o; kid; kid = OpSIBLING(kid)) {
16196 const OPCODE type = kid->op_type;
16197 if ( (type == OP_CONST && SvOK(cSVOPx_sv(kid)))
16198 || (type == OP_NULL && ! (kid->op_flags & OPf_KIDS))
16199 || (type == OP_PUSHMARK)
16200 || (type == OP_PADRANGE)
16204 if (o2) { /* more than one found */
16211 return find_uninit_var(o2, uninit_sv, match, desc_p);
16213 /* scan all args */
16215 sv = find_uninit_var(o, uninit_sv, 1, desc_p);
16227 =for apidoc report_uninit
16229 Print appropriate "Use of uninitialized variable" warning.
16235 Perl_report_uninit(pTHX_ const SV *uninit_sv)
16237 const char *desc = NULL;
16238 SV* varname = NULL;
16241 desc = PL_op->op_type == OP_STRINGIFY && PL_op->op_folded
16244 if (uninit_sv && PL_curpad) {
16245 varname = find_uninit_var(PL_op, uninit_sv, 0, &desc);
16247 sv_insert(varname, 0, 0, " ", 1);
16250 else if (PL_curstackinfo->si_type == PERLSI_SORT
16251 && CxMULTICALL(&cxstack[cxstack_ix]))
16253 /* we've reached the end of a sort block or sub,
16254 * and the uninit value is probably what that code returned */
16258 /* PL_warn_uninit_sv is constant */
16259 GCC_DIAG_IGNORE(-Wformat-nonliteral);
16261 /* diag_listed_as: Use of uninitialized value%s */
16262 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit_sv,
16263 SVfARG(varname ? varname : &PL_sv_no),
16266 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit,
16272 * ex: set ts=8 sts=4 sw=4 et: