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 #ifdef PERL_OLD_COPY_ON_WRITE
129 #define SV_COW_NEXT_SV(sv) INT2PTR(SV *,SvUVX(sv))
130 #define SV_COW_NEXT_SV_SET(current,next) SvUV_set(current, PTR2UV(next))
133 /* ============================================================================
135 =head1 Allocation and deallocation of SVs.
136 An SV (or AV, HV, etc.) is allocated in two parts: the head (struct
137 sv, av, hv...) contains type and reference count information, and for
138 many types, a pointer to the body (struct xrv, xpv, xpviv...), which
139 contains fields specific to each type. Some types store all they need
140 in the head, so don't have a body.
142 In all but the most memory-paranoid configurations (ex: PURIFY), heads
143 and bodies are allocated out of arenas, which by default are
144 approximately 4K chunks of memory parcelled up into N heads or bodies.
145 Sv-bodies are allocated by their sv-type, guaranteeing size
146 consistency needed to allocate safely from arrays.
148 For SV-heads, the first slot in each arena is reserved, and holds a
149 link to the next arena, some flags, and a note of the number of slots.
150 Snaked through each arena chain is a linked list of free items; when
151 this becomes empty, an extra arena is allocated and divided up into N
152 items which are threaded into the free list.
154 SV-bodies are similar, but they use arena-sets by default, which
155 separate the link and info from the arena itself, and reclaim the 1st
156 slot in the arena. SV-bodies are further described later.
158 The following global variables are associated with arenas:
160 PL_sv_arenaroot pointer to list of SV arenas
161 PL_sv_root pointer to list of free SV structures
163 PL_body_arenas head of linked-list of body arenas
164 PL_body_roots[] array of pointers to list of free bodies of svtype
165 arrays are indexed by the svtype needed
167 A few special SV heads are not allocated from an arena, but are
168 instead directly created in the interpreter structure, eg PL_sv_undef.
169 The size of arenas can be changed from the default by setting
170 PERL_ARENA_SIZE appropriately at compile time.
172 The SV arena serves the secondary purpose of allowing still-live SVs
173 to be located and destroyed during final cleanup.
175 At the lowest level, the macros new_SV() and del_SV() grab and free
176 an SV head. (If debugging with -DD, del_SV() calls the function S_del_sv()
177 to return the SV to the free list with error checking.) new_SV() calls
178 more_sv() / sv_add_arena() to add an extra arena if the free list is empty.
179 SVs in the free list have their SvTYPE field set to all ones.
181 At the time of very final cleanup, sv_free_arenas() is called from
182 perl_destruct() to physically free all the arenas allocated since the
183 start of the interpreter.
185 The function visit() scans the SV arenas list, and calls a specified
186 function for each SV it finds which is still live - ie which has an SvTYPE
187 other than all 1's, and a non-zero SvREFCNT. visit() is used by the
188 following functions (specified as [function that calls visit()] / [function
189 called by visit() for each SV]):
191 sv_report_used() / do_report_used()
192 dump all remaining SVs (debugging aid)
194 sv_clean_objs() / do_clean_objs(),do_clean_named_objs(),
195 do_clean_named_io_objs(),do_curse()
196 Attempt to free all objects pointed to by RVs,
197 try to do the same for all objects indir-
198 ectly referenced by typeglobs too, and
199 then do a final sweep, cursing any
200 objects that remain. Called once from
201 perl_destruct(), prior to calling sv_clean_all()
204 sv_clean_all() / do_clean_all()
205 SvREFCNT_dec(sv) each remaining SV, possibly
206 triggering an sv_free(). It also sets the
207 SVf_BREAK flag on the SV to indicate that the
208 refcnt has been artificially lowered, and thus
209 stopping sv_free() from giving spurious warnings
210 about SVs which unexpectedly have a refcnt
211 of zero. called repeatedly from perl_destruct()
212 until there are no SVs left.
214 =head2 Arena allocator API Summary
216 Private API to rest of sv.c
220 new_XPVNV(), del_XPVGV(),
225 sv_report_used(), sv_clean_objs(), sv_clean_all(), sv_free_arenas()
229 * ========================================================================= */
232 * "A time to plant, and a time to uproot what was planted..."
236 # define MEM_LOG_NEW_SV(sv, file, line, func) \
237 Perl_mem_log_new_sv(sv, file, line, func)
238 # define MEM_LOG_DEL_SV(sv, file, line, func) \
239 Perl_mem_log_del_sv(sv, file, line, func)
241 # define MEM_LOG_NEW_SV(sv, file, line, func) NOOP
242 # define MEM_LOG_DEL_SV(sv, file, line, func) NOOP
245 #ifdef DEBUG_LEAKING_SCALARS
246 # define FREE_SV_DEBUG_FILE(sv) STMT_START { \
247 if ((sv)->sv_debug_file) PerlMemShared_free((sv)->sv_debug_file); \
249 # define DEBUG_SV_SERIAL(sv) \
250 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) del_SV\n", \
251 PTR2UV(sv), (long)(sv)->sv_debug_serial))
253 # define FREE_SV_DEBUG_FILE(sv)
254 # define DEBUG_SV_SERIAL(sv) NOOP
258 # define SvARENA_CHAIN(sv) ((sv)->sv_u.svu_rv)
259 # define SvARENA_CHAIN_SET(sv,val) (sv)->sv_u.svu_rv = MUTABLE_SV((val))
260 /* Whilst I'd love to do this, it seems that things like to check on
262 # define POSION_SV_HEAD(sv) PoisonNew(sv, 1, struct STRUCT_SV)
264 # define POSION_SV_HEAD(sv) PoisonNew(&SvANY(sv), 1, void *), \
265 PoisonNew(&SvREFCNT(sv), 1, U32)
267 # define SvARENA_CHAIN(sv) SvANY(sv)
268 # define SvARENA_CHAIN_SET(sv,val) SvANY(sv) = (void *)(val)
269 # define POSION_SV_HEAD(sv)
272 /* Mark an SV head as unused, and add to free list.
274 * If SVf_BREAK is set, skip adding it to the free list, as this SV had
275 * its refcount artificially decremented during global destruction, so
276 * there may be dangling pointers to it. The last thing we want in that
277 * case is for it to be reused. */
279 #define plant_SV(p) \
281 const U32 old_flags = SvFLAGS(p); \
282 MEM_LOG_DEL_SV(p, __FILE__, __LINE__, FUNCTION__); \
283 DEBUG_SV_SERIAL(p); \
284 FREE_SV_DEBUG_FILE(p); \
286 SvFLAGS(p) = SVTYPEMASK; \
287 if (!(old_flags & SVf_BREAK)) { \
288 SvARENA_CHAIN_SET(p, PL_sv_root); \
294 #define uproot_SV(p) \
297 PL_sv_root = MUTABLE_SV(SvARENA_CHAIN(p)); \
302 /* make some more SVs by adding another arena */
308 char *chunk; /* must use New here to match call to */
309 Newx(chunk,PERL_ARENA_SIZE,char); /* Safefree() in sv_free_arenas() */
310 sv_add_arena(chunk, PERL_ARENA_SIZE, 0);
315 /* new_SV(): return a new, empty SV head */
317 #ifdef DEBUG_LEAKING_SCALARS
318 /* provide a real function for a debugger to play with */
320 S_new_SV(pTHX_ const char *file, int line, const char *func)
327 sv = S_more_sv(aTHX);
331 sv->sv_debug_optype = PL_op ? PL_op->op_type : 0;
332 sv->sv_debug_line = (U16) (PL_parser && PL_parser->copline != NOLINE
338 sv->sv_debug_inpad = 0;
339 sv->sv_debug_parent = NULL;
340 sv->sv_debug_file = PL_curcop ? savesharedpv(CopFILE(PL_curcop)): NULL;
342 sv->sv_debug_serial = PL_sv_serial++;
344 MEM_LOG_NEW_SV(sv, file, line, func);
345 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) new_SV (from %s:%d [%s])\n",
346 PTR2UV(sv), (long)sv->sv_debug_serial, file, line, func));
350 # define new_SV(p) (p)=S_new_SV(aTHX_ __FILE__, __LINE__, FUNCTION__)
358 (p) = S_more_sv(aTHX); \
362 MEM_LOG_NEW_SV(p, __FILE__, __LINE__, FUNCTION__); \
367 /* del_SV(): return an empty SV head to the free list */
380 S_del_sv(pTHX_ SV *p)
382 PERL_ARGS_ASSERT_DEL_SV;
387 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
388 const SV * const sv = sva + 1;
389 const SV * const svend = &sva[SvREFCNT(sva)];
390 if (p >= sv && p < svend) {
396 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
397 "Attempt to free non-arena SV: 0x%"UVxf
398 pTHX__FORMAT, PTR2UV(p) pTHX__VALUE);
405 #else /* ! DEBUGGING */
407 #define del_SV(p) plant_SV(p)
409 #endif /* DEBUGGING */
413 =head1 SV Manipulation Functions
415 =for apidoc sv_add_arena
417 Given a chunk of memory, link it to the head of the list of arenas,
418 and split it into a list of free SVs.
424 S_sv_add_arena(pTHX_ char *const ptr, const U32 size, const U32 flags)
426 SV *const sva = MUTABLE_SV(ptr);
430 PERL_ARGS_ASSERT_SV_ADD_ARENA;
432 /* The first SV in an arena isn't an SV. */
433 SvANY(sva) = (void *) PL_sv_arenaroot; /* ptr to next arena */
434 SvREFCNT(sva) = size / sizeof(SV); /* number of SV slots */
435 SvFLAGS(sva) = flags; /* FAKE if not to be freed */
437 PL_sv_arenaroot = sva;
438 PL_sv_root = sva + 1;
440 svend = &sva[SvREFCNT(sva) - 1];
443 SvARENA_CHAIN_SET(sv, (sv + 1));
447 /* Must always set typemask because it's always checked in on cleanup
448 when the arenas are walked looking for objects. */
449 SvFLAGS(sv) = SVTYPEMASK;
452 SvARENA_CHAIN_SET(sv, 0);
456 SvFLAGS(sv) = SVTYPEMASK;
459 /* visit(): call the named function for each non-free SV in the arenas
460 * whose flags field matches the flags/mask args. */
463 S_visit(pTHX_ SVFUNC_t f, const U32 flags, const U32 mask)
468 PERL_ARGS_ASSERT_VISIT;
470 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
471 const SV * const svend = &sva[SvREFCNT(sva)];
473 for (sv = sva + 1; sv < svend; ++sv) {
474 if (SvTYPE(sv) != (svtype)SVTYPEMASK
475 && (sv->sv_flags & mask) == flags
488 /* called by sv_report_used() for each live SV */
491 do_report_used(pTHX_ SV *const sv)
493 if (SvTYPE(sv) != (svtype)SVTYPEMASK) {
494 PerlIO_printf(Perl_debug_log, "****\n");
501 =for apidoc sv_report_used
503 Dump the contents of all SVs not yet freed (debugging aid).
509 Perl_sv_report_used(pTHX)
512 visit(do_report_used, 0, 0);
518 /* called by sv_clean_objs() for each live SV */
521 do_clean_objs(pTHX_ SV *const ref)
525 SV * const target = SvRV(ref);
526 if (SvOBJECT(target)) {
527 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning object ref:\n "), sv_dump(ref)));
528 if (SvWEAKREF(ref)) {
529 sv_del_backref(target, ref);
535 SvREFCNT_dec_NN(target);
542 /* clear any slots in a GV which hold objects - except IO;
543 * called by sv_clean_objs() for each live GV */
546 do_clean_named_objs(pTHX_ SV *const sv)
549 assert(SvTYPE(sv) == SVt_PVGV);
550 assert(isGV_with_GP(sv));
554 /* freeing GP entries may indirectly free the current GV;
555 * hold onto it while we mess with the GP slots */
558 if ( ((obj = GvSV(sv) )) && SvOBJECT(obj)) {
559 DEBUG_D((PerlIO_printf(Perl_debug_log,
560 "Cleaning named glob SV object:\n "), sv_dump(obj)));
562 SvREFCNT_dec_NN(obj);
564 if ( ((obj = MUTABLE_SV(GvAV(sv)) )) && SvOBJECT(obj)) {
565 DEBUG_D((PerlIO_printf(Perl_debug_log,
566 "Cleaning named glob AV object:\n "), sv_dump(obj)));
568 SvREFCNT_dec_NN(obj);
570 if ( ((obj = MUTABLE_SV(GvHV(sv)) )) && SvOBJECT(obj)) {
571 DEBUG_D((PerlIO_printf(Perl_debug_log,
572 "Cleaning named glob HV object:\n "), sv_dump(obj)));
574 SvREFCNT_dec_NN(obj);
576 if ( ((obj = MUTABLE_SV(GvCV(sv)) )) && SvOBJECT(obj)) {
577 DEBUG_D((PerlIO_printf(Perl_debug_log,
578 "Cleaning named glob CV object:\n "), sv_dump(obj)));
580 SvREFCNT_dec_NN(obj);
582 SvREFCNT_dec_NN(sv); /* undo the inc above */
585 /* clear any IO slots in a GV which hold objects (except stderr, defout);
586 * called by sv_clean_objs() for each live GV */
589 do_clean_named_io_objs(pTHX_ SV *const sv)
592 assert(SvTYPE(sv) == SVt_PVGV);
593 assert(isGV_with_GP(sv));
594 if (!GvGP(sv) || sv == (SV*)PL_stderrgv || sv == (SV*)PL_defoutgv)
598 if ( ((obj = MUTABLE_SV(GvIO(sv)) )) && SvOBJECT(obj)) {
599 DEBUG_D((PerlIO_printf(Perl_debug_log,
600 "Cleaning named glob IO object:\n "), sv_dump(obj)));
602 SvREFCNT_dec_NN(obj);
604 SvREFCNT_dec_NN(sv); /* undo the inc above */
607 /* Void wrapper to pass to visit() */
609 do_curse(pTHX_ SV * const sv) {
610 if ((PL_stderrgv && GvGP(PL_stderrgv) && (SV*)GvIO(PL_stderrgv) == sv)
611 || (PL_defoutgv && GvGP(PL_defoutgv) && (SV*)GvIO(PL_defoutgv) == sv))
617 =for apidoc sv_clean_objs
619 Attempt to destroy all objects not yet freed.
625 Perl_sv_clean_objs(pTHX)
628 PL_in_clean_objs = TRUE;
629 visit(do_clean_objs, SVf_ROK, SVf_ROK);
630 /* Some barnacles may yet remain, clinging to typeglobs.
631 * Run the non-IO destructors first: they may want to output
632 * error messages, close files etc */
633 visit(do_clean_named_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
634 visit(do_clean_named_io_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
635 /* And if there are some very tenacious barnacles clinging to arrays,
636 closures, or what have you.... */
637 visit(do_curse, SVs_OBJECT, SVs_OBJECT);
638 olddef = PL_defoutgv;
639 PL_defoutgv = NULL; /* disable skip of PL_defoutgv */
640 if (olddef && isGV_with_GP(olddef))
641 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olddef));
642 olderr = PL_stderrgv;
643 PL_stderrgv = NULL; /* disable skip of PL_stderrgv */
644 if (olderr && isGV_with_GP(olderr))
645 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olderr));
646 SvREFCNT_dec(olddef);
647 PL_in_clean_objs = FALSE;
650 /* called by sv_clean_all() for each live SV */
653 do_clean_all(pTHX_ SV *const sv)
655 if (sv == (const SV *) PL_fdpid || sv == (const SV *)PL_strtab) {
656 /* don't clean pid table and strtab */
659 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning loops: SV at 0x%"UVxf"\n", PTR2UV(sv)) ));
660 SvFLAGS(sv) |= SVf_BREAK;
665 =for apidoc sv_clean_all
667 Decrement the refcnt of each remaining SV, possibly triggering a
668 cleanup. This function may have to be called multiple times to free
669 SVs which are in complex self-referential hierarchies.
675 Perl_sv_clean_all(pTHX)
678 PL_in_clean_all = TRUE;
679 cleaned = visit(do_clean_all, 0,0);
684 ARENASETS: a meta-arena implementation which separates arena-info
685 into struct arena_set, which contains an array of struct
686 arena_descs, each holding info for a single arena. By separating
687 the meta-info from the arena, we recover the 1st slot, formerly
688 borrowed for list management. The arena_set is about the size of an
689 arena, avoiding the needless malloc overhead of a naive linked-list.
691 The cost is 1 arena-set malloc per ~320 arena-mallocs, + the unused
692 memory in the last arena-set (1/2 on average). In trade, we get
693 back the 1st slot in each arena (ie 1.7% of a CV-arena, less for
694 smaller types). The recovery of the wasted space allows use of
695 small arenas for large, rare body types, by changing array* fields
696 in body_details_by_type[] below.
699 char *arena; /* the raw storage, allocated aligned */
700 size_t size; /* its size ~4k typ */
701 svtype utype; /* bodytype stored in arena */
706 /* Get the maximum number of elements in set[] such that struct arena_set
707 will fit within PERL_ARENA_SIZE, which is probably just under 4K, and
708 therefore likely to be 1 aligned memory page. */
710 #define ARENAS_PER_SET ((PERL_ARENA_SIZE - sizeof(struct arena_set*) \
711 - 2 * sizeof(int)) / sizeof (struct arena_desc))
714 struct arena_set* next;
715 unsigned int set_size; /* ie ARENAS_PER_SET */
716 unsigned int curr; /* index of next available arena-desc */
717 struct arena_desc set[ARENAS_PER_SET];
721 =for apidoc sv_free_arenas
723 Deallocate the memory used by all arenas. Note that all the individual SV
724 heads and bodies within the arenas must already have been freed.
730 Perl_sv_free_arenas(pTHX)
736 /* Free arenas here, but be careful about fake ones. (We assume
737 contiguity of the fake ones with the corresponding real ones.) */
739 for (sva = PL_sv_arenaroot; sva; sva = svanext) {
740 svanext = MUTABLE_SV(SvANY(sva));
741 while (svanext && SvFAKE(svanext))
742 svanext = MUTABLE_SV(SvANY(svanext));
749 struct arena_set *aroot = (struct arena_set*) PL_body_arenas;
752 struct arena_set *current = aroot;
755 assert(aroot->set[i].arena);
756 Safefree(aroot->set[i].arena);
764 i = PERL_ARENA_ROOTS_SIZE;
766 PL_body_roots[i] = 0;
773 Here are mid-level routines that manage the allocation of bodies out
774 of the various arenas. There are 5 kinds of arenas:
776 1. SV-head arenas, which are discussed and handled above
777 2. regular body arenas
778 3. arenas for reduced-size bodies
781 Arena types 2 & 3 are chained by body-type off an array of
782 arena-root pointers, which is indexed by svtype. Some of the
783 larger/less used body types are malloced singly, since a large
784 unused block of them is wasteful. Also, several svtypes dont have
785 bodies; the data fits into the sv-head itself. The arena-root
786 pointer thus has a few unused root-pointers (which may be hijacked
787 later for arena types 4,5)
789 3 differs from 2 as an optimization; some body types have several
790 unused fields in the front of the structure (which are kept in-place
791 for consistency). These bodies can be allocated in smaller chunks,
792 because the leading fields arent accessed. Pointers to such bodies
793 are decremented to point at the unused 'ghost' memory, knowing that
794 the pointers are used with offsets to the real memory.
797 =head1 SV-Body Allocation
801 Allocation of SV-bodies is similar to SV-heads, differing as follows;
802 the allocation mechanism is used for many body types, so is somewhat
803 more complicated, it uses arena-sets, and has no need for still-live
806 At the outermost level, (new|del)_X*V macros return bodies of the
807 appropriate type. These macros call either (new|del)_body_type or
808 (new|del)_body_allocated macro pairs, depending on specifics of the
809 type. Most body types use the former pair, the latter pair is used to
810 allocate body types with "ghost fields".
812 "ghost fields" are fields that are unused in certain types, and
813 consequently don't need to actually exist. They are declared because
814 they're part of a "base type", which allows use of functions as
815 methods. The simplest examples are AVs and HVs, 2 aggregate types
816 which don't use the fields which support SCALAR semantics.
818 For these types, the arenas are carved up into appropriately sized
819 chunks, we thus avoid wasted memory for those unaccessed members.
820 When bodies are allocated, we adjust the pointer back in memory by the
821 size of the part not allocated, so it's as if we allocated the full
822 structure. (But things will all go boom if you write to the part that
823 is "not there", because you'll be overwriting the last members of the
824 preceding structure in memory.)
826 We calculate the correction using the STRUCT_OFFSET macro on the first
827 member present. If the allocated structure is smaller (no initial NV
828 actually allocated) then the net effect is to subtract the size of the NV
829 from the pointer, to return a new pointer as if an initial NV were actually
830 allocated. (We were using structures named *_allocated for this, but
831 this turned out to be a subtle bug, because a structure without an NV
832 could have a lower alignment constraint, but the compiler is allowed to
833 optimised accesses based on the alignment constraint of the actual pointer
834 to the full structure, for example, using a single 64 bit load instruction
835 because it "knows" that two adjacent 32 bit members will be 8-byte aligned.)
837 This is the same trick as was used for NV and IV bodies. Ironically it
838 doesn't need to be used for NV bodies any more, because NV is now at
839 the start of the structure. IV bodies, and also in some builds NV bodies,
840 don't need it either, because they are no longer allocated.
842 In turn, the new_body_* allocators call S_new_body(), which invokes
843 new_body_inline macro, which takes a lock, and takes a body off the
844 linked list at PL_body_roots[sv_type], calling Perl_more_bodies() if
845 necessary to refresh an empty list. Then the lock is released, and
846 the body is returned.
848 Perl_more_bodies allocates a new arena, and carves it up into an array of N
849 bodies, which it strings into a linked list. It looks up arena-size
850 and body-size from the body_details table described below, thus
851 supporting the multiple body-types.
853 If PURIFY is defined, or PERL_ARENA_SIZE=0, arenas are not used, and
854 the (new|del)_X*V macros are mapped directly to malloc/free.
856 For each sv-type, struct body_details bodies_by_type[] carries
857 parameters which control these aspects of SV handling:
859 Arena_size determines whether arenas are used for this body type, and if
860 so, how big they are. PURIFY or PERL_ARENA_SIZE=0 set this field to
861 zero, forcing individual mallocs and frees.
863 Body_size determines how big a body is, and therefore how many fit into
864 each arena. Offset carries the body-pointer adjustment needed for
865 "ghost fields", and is used in *_allocated macros.
867 But its main purpose is to parameterize info needed in
868 Perl_sv_upgrade(). The info here dramatically simplifies the function
869 vs the implementation in 5.8.8, making it table-driven. All fields
870 are used for this, except for arena_size.
872 For the sv-types that have no bodies, arenas are not used, so those
873 PL_body_roots[sv_type] are unused, and can be overloaded. In
874 something of a special case, SVt_NULL is borrowed for HE arenas;
875 PL_body_roots[HE_SVSLOT=SVt_NULL] is filled by S_more_he, but the
876 bodies_by_type[SVt_NULL] slot is not used, as the table is not
881 struct body_details {
882 U8 body_size; /* Size to allocate */
883 U8 copy; /* Size of structure to copy (may be shorter) */
884 U8 offset; /* Size of unalloced ghost fields to first alloced field*/
885 PERL_BITFIELD8 type : 4; /* We have space for a sanity check. */
886 PERL_BITFIELD8 cant_upgrade : 1;/* Cannot upgrade this type */
887 PERL_BITFIELD8 zero_nv : 1; /* zero the NV when upgrading from this */
888 PERL_BITFIELD8 arena : 1; /* Allocated from an arena */
889 U32 arena_size; /* Size of arena to allocate */
897 /* With -DPURFIY we allocate everything directly, and don't use arenas.
898 This seems a rather elegant way to simplify some of the code below. */
899 #define HASARENA FALSE
901 #define HASARENA TRUE
903 #define NOARENA FALSE
905 /* Size the arenas to exactly fit a given number of bodies. A count
906 of 0 fits the max number bodies into a PERL_ARENA_SIZE.block,
907 simplifying the default. If count > 0, the arena is sized to fit
908 only that many bodies, allowing arenas to be used for large, rare
909 bodies (XPVFM, XPVIO) without undue waste. The arena size is
910 limited by PERL_ARENA_SIZE, so we can safely oversize the
913 #define FIT_ARENA0(body_size) \
914 ((size_t)(PERL_ARENA_SIZE / body_size) * body_size)
915 #define FIT_ARENAn(count,body_size) \
916 ( count * body_size <= PERL_ARENA_SIZE) \
917 ? count * body_size \
918 : FIT_ARENA0 (body_size)
919 #define FIT_ARENA(count,body_size) \
921 ? FIT_ARENAn (count, body_size) \
922 : FIT_ARENA0 (body_size))
924 /* Calculate the length to copy. Specifically work out the length less any
925 final padding the compiler needed to add. See the comment in sv_upgrade
926 for why copying the padding proved to be a bug. */
928 #define copy_length(type, last_member) \
929 STRUCT_OFFSET(type, last_member) \
930 + sizeof (((type*)SvANY((const SV *)0))->last_member)
932 static const struct body_details bodies_by_type[] = {
933 /* HEs use this offset for their arena. */
934 { 0, 0, 0, SVt_NULL, FALSE, NONV, NOARENA, 0 },
936 /* IVs are in the head, so the allocation size is 0. */
938 sizeof(IV), /* This is used to copy out the IV body. */
939 STRUCT_OFFSET(XPVIV, xiv_iv), SVt_IV, FALSE, NONV,
940 NOARENA /* IVS don't need an arena */, 0
945 STRUCT_OFFSET(XPVNV, xnv_u),
946 SVt_NV, FALSE, HADNV, NOARENA, 0 },
948 { sizeof(NV), sizeof(NV),
949 STRUCT_OFFSET(XPVNV, xnv_u),
950 SVt_NV, FALSE, HADNV, HASARENA, FIT_ARENA(0, sizeof(NV)) },
953 { sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur),
954 copy_length(XPV, xpv_len) - STRUCT_OFFSET(XPV, xpv_cur),
955 + STRUCT_OFFSET(XPV, xpv_cur),
956 SVt_PV, FALSE, NONV, HASARENA,
957 FIT_ARENA(0, sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur)) },
959 { sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur),
960 copy_length(XINVLIST, is_offset) - STRUCT_OFFSET(XPV, xpv_cur),
961 + STRUCT_OFFSET(XPV, xpv_cur),
962 SVt_INVLIST, TRUE, NONV, HASARENA,
963 FIT_ARENA(0, sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur)) },
965 { sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur),
966 copy_length(XPVIV, xiv_u) - STRUCT_OFFSET(XPV, xpv_cur),
967 + STRUCT_OFFSET(XPV, xpv_cur),
968 SVt_PVIV, FALSE, NONV, HASARENA,
969 FIT_ARENA(0, sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur)) },
971 { sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur),
972 copy_length(XPVNV, xnv_u) - STRUCT_OFFSET(XPV, xpv_cur),
973 + STRUCT_OFFSET(XPV, xpv_cur),
974 SVt_PVNV, FALSE, HADNV, HASARENA,
975 FIT_ARENA(0, sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur)) },
977 { sizeof(XPVMG), copy_length(XPVMG, xnv_u), 0, SVt_PVMG, FALSE, HADNV,
978 HASARENA, FIT_ARENA(0, sizeof(XPVMG)) },
983 SVt_REGEXP, TRUE, NONV, HASARENA,
984 FIT_ARENA(0, sizeof(regexp))
987 { sizeof(XPVGV), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV,
988 HASARENA, FIT_ARENA(0, sizeof(XPVGV)) },
990 { sizeof(XPVLV), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV,
991 HASARENA, FIT_ARENA(0, sizeof(XPVLV)) },
994 copy_length(XPVAV, xav_alloc),
996 SVt_PVAV, TRUE, NONV, HASARENA,
997 FIT_ARENA(0, sizeof(XPVAV)) },
1000 copy_length(XPVHV, xhv_max),
1002 SVt_PVHV, TRUE, NONV, HASARENA,
1003 FIT_ARENA(0, sizeof(XPVHV)) },
1008 SVt_PVCV, TRUE, NONV, HASARENA,
1009 FIT_ARENA(0, sizeof(XPVCV)) },
1014 SVt_PVFM, TRUE, NONV, NOARENA,
1015 FIT_ARENA(20, sizeof(XPVFM)) },
1020 SVt_PVIO, TRUE, NONV, HASARENA,
1021 FIT_ARENA(24, sizeof(XPVIO)) },
1024 #define new_body_allocated(sv_type) \
1025 (void *)((char *)S_new_body(aTHX_ sv_type) \
1026 - bodies_by_type[sv_type].offset)
1028 /* return a thing to the free list */
1030 #define del_body(thing, root) \
1032 void ** const thing_copy = (void **)thing; \
1033 *thing_copy = *root; \
1034 *root = (void*)thing_copy; \
1038 #if !(NVSIZE <= IVSIZE)
1039 # define new_XNV() safemalloc(sizeof(XPVNV))
1041 #define new_XPVNV() safemalloc(sizeof(XPVNV))
1042 #define new_XPVMG() safemalloc(sizeof(XPVMG))
1044 #define del_XPVGV(p) safefree(p)
1048 #if !(NVSIZE <= IVSIZE)
1049 # define new_XNV() new_body_allocated(SVt_NV)
1051 #define new_XPVNV() new_body_allocated(SVt_PVNV)
1052 #define new_XPVMG() new_body_allocated(SVt_PVMG)
1054 #define del_XPVGV(p) del_body(p + bodies_by_type[SVt_PVGV].offset, \
1055 &PL_body_roots[SVt_PVGV])
1059 /* no arena for you! */
1061 #define new_NOARENA(details) \
1062 safemalloc((details)->body_size + (details)->offset)
1063 #define new_NOARENAZ(details) \
1064 safecalloc((details)->body_size + (details)->offset, 1)
1067 Perl_more_bodies (pTHX_ const svtype sv_type, const size_t body_size,
1068 const size_t arena_size)
1070 void ** const root = &PL_body_roots[sv_type];
1071 struct arena_desc *adesc;
1072 struct arena_set *aroot = (struct arena_set *) PL_body_arenas;
1076 const size_t good_arena_size = Perl_malloc_good_size(arena_size);
1077 #if defined(DEBUGGING) && defined(PERL_GLOBAL_STRUCT)
1080 #if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT_PRIVATE)
1081 static bool done_sanity_check;
1083 /* PERL_GLOBAL_STRUCT_PRIVATE cannot coexist with global
1084 * variables like done_sanity_check. */
1085 if (!done_sanity_check) {
1086 unsigned int i = SVt_LAST;
1088 done_sanity_check = TRUE;
1091 assert (bodies_by_type[i].type == i);
1097 /* may need new arena-set to hold new arena */
1098 if (!aroot || aroot->curr >= aroot->set_size) {
1099 struct arena_set *newroot;
1100 Newxz(newroot, 1, struct arena_set);
1101 newroot->set_size = ARENAS_PER_SET;
1102 newroot->next = aroot;
1104 PL_body_arenas = (void *) newroot;
1105 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot));
1108 /* ok, now have arena-set with at least 1 empty/available arena-desc */
1109 curr = aroot->curr++;
1110 adesc = &(aroot->set[curr]);
1111 assert(!adesc->arena);
1113 Newx(adesc->arena, good_arena_size, char);
1114 adesc->size = good_arena_size;
1115 adesc->utype = sv_type;
1116 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %"UVuf"\n",
1117 curr, (void*)adesc->arena, (UV)good_arena_size));
1119 start = (char *) adesc->arena;
1121 /* Get the address of the byte after the end of the last body we can fit.
1122 Remember, this is integer division: */
1123 end = start + good_arena_size / body_size * body_size;
1125 /* computed count doesn't reflect the 1st slot reservation */
1126 #if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE)
1127 DEBUG_m(PerlIO_printf(Perl_debug_log,
1128 "arena %p end %p arena-size %d (from %d) type %d "
1130 (void*)start, (void*)end, (int)good_arena_size,
1131 (int)arena_size, sv_type, (int)body_size,
1132 (int)good_arena_size / (int)body_size));
1134 DEBUG_m(PerlIO_printf(Perl_debug_log,
1135 "arena %p end %p arena-size %d type %d size %d ct %d\n",
1136 (void*)start, (void*)end,
1137 (int)arena_size, sv_type, (int)body_size,
1138 (int)good_arena_size / (int)body_size));
1140 *root = (void *)start;
1143 /* Where the next body would start: */
1144 char * const next = start + body_size;
1147 /* This is the last body: */
1148 assert(next == end);
1150 *(void **)start = 0;
1154 *(void**) start = (void *)next;
1159 /* grab a new thing from the free list, allocating more if necessary.
1160 The inline version is used for speed in hot routines, and the
1161 function using it serves the rest (unless PURIFY).
1163 #define new_body_inline(xpv, sv_type) \
1165 void ** const r3wt = &PL_body_roots[sv_type]; \
1166 xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \
1167 ? *((void **)(r3wt)) : Perl_more_bodies(aTHX_ sv_type, \
1168 bodies_by_type[sv_type].body_size,\
1169 bodies_by_type[sv_type].arena_size)); \
1170 *(r3wt) = *(void**)(xpv); \
1176 S_new_body(pTHX_ const svtype sv_type)
1179 new_body_inline(xpv, sv_type);
1185 static const struct body_details fake_rv =
1186 { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 };
1189 =for apidoc sv_upgrade
1191 Upgrade an SV to a more complex form. Generally adds a new body type to the
1192 SV, then copies across as much information as possible from the old body.
1193 It croaks if the SV is already in a more complex form than requested. You
1194 generally want to use the C<SvUPGRADE> macro wrapper, which checks the type
1195 before calling C<sv_upgrade>, and hence does not croak. See also
1202 Perl_sv_upgrade(pTHX_ SV *const sv, svtype new_type)
1206 const svtype old_type = SvTYPE(sv);
1207 const struct body_details *new_type_details;
1208 const struct body_details *old_type_details
1209 = bodies_by_type + old_type;
1210 SV *referant = NULL;
1212 PERL_ARGS_ASSERT_SV_UPGRADE;
1214 if (old_type == new_type)
1217 /* This clause was purposefully added ahead of the early return above to
1218 the shared string hackery for (sort {$a <=> $b} keys %hash), with the
1219 inference by Nick I-S that it would fix other troublesome cases. See
1220 changes 7162, 7163 (f130fd4589cf5fbb24149cd4db4137c8326f49c1 and parent)
1222 Given that shared hash key scalars are no longer PVIV, but PV, there is
1223 no longer need to unshare so as to free up the IVX slot for its proper
1224 purpose. So it's safe to move the early return earlier. */
1226 if (new_type > SVt_PVMG && SvIsCOW(sv)) {
1227 sv_force_normal_flags(sv, 0);
1230 old_body = SvANY(sv);
1232 /* Copying structures onto other structures that have been neatly zeroed
1233 has a subtle gotcha. Consider XPVMG
1235 +------+------+------+------+------+-------+-------+
1236 | NV | CUR | LEN | IV | MAGIC | STASH |
1237 +------+------+------+------+------+-------+-------+
1238 0 4 8 12 16 20 24 28
1240 where NVs are aligned to 8 bytes, so that sizeof that structure is
1241 actually 32 bytes long, with 4 bytes of padding at the end:
1243 +------+------+------+------+------+-------+-------+------+
1244 | NV | CUR | LEN | IV | MAGIC | STASH | ??? |
1245 +------+------+------+------+------+-------+-------+------+
1246 0 4 8 12 16 20 24 28 32
1248 so what happens if you allocate memory for this structure:
1250 +------+------+------+------+------+-------+-------+------+------+...
1251 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME |
1252 +------+------+------+------+------+-------+-------+------+------+...
1253 0 4 8 12 16 20 24 28 32 36
1255 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you
1256 expect, because you copy the area marked ??? onto GP. Now, ??? may have
1257 started out as zero once, but it's quite possible that it isn't. So now,
1258 rather than a nicely zeroed GP, you have it pointing somewhere random.
1261 (In fact, GP ends up pointing at a previous GP structure, because the
1262 principle cause of the padding in XPVMG getting garbage is a copy of
1263 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now
1264 this happens to be moot because XPVGV has been re-ordered, with GP
1265 no longer after STASH)
1267 So we are careful and work out the size of used parts of all the
1275 referant = SvRV(sv);
1276 old_type_details = &fake_rv;
1277 if (new_type == SVt_NV)
1278 new_type = SVt_PVNV;
1280 if (new_type < SVt_PVIV) {
1281 new_type = (new_type == SVt_NV)
1282 ? SVt_PVNV : SVt_PVIV;
1287 if (new_type < SVt_PVNV) {
1288 new_type = SVt_PVNV;
1292 assert(new_type > SVt_PV);
1293 assert(SVt_IV < SVt_PV);
1294 assert(SVt_NV < SVt_PV);
1301 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena,
1302 there's no way that it can be safely upgraded, because perl.c
1303 expects to Safefree(SvANY(PL_mess_sv)) */
1304 assert(sv != PL_mess_sv);
1305 /* This flag bit is used to mean other things in other scalar types.
1306 Given that it only has meaning inside the pad, it shouldn't be set
1307 on anything that can get upgraded. */
1308 assert(!SvPAD_TYPED(sv));
1311 if (UNLIKELY(old_type_details->cant_upgrade))
1312 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf,
1313 sv_reftype(sv, 0), (UV) old_type, (UV) new_type);
1316 if (UNLIKELY(old_type > new_type))
1317 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d",
1318 (int)old_type, (int)new_type);
1320 new_type_details = bodies_by_type + new_type;
1322 SvFLAGS(sv) &= ~SVTYPEMASK;
1323 SvFLAGS(sv) |= new_type;
1325 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of
1326 the return statements above will have triggered. */
1327 assert (new_type != SVt_NULL);
1330 assert(old_type == SVt_NULL);
1331 SvANY(sv) = (XPVIV*)((char*)&(sv->sv_u.svu_iv) - STRUCT_OFFSET(XPVIV, xiv_iv));
1335 assert(old_type == SVt_NULL);
1336 #if NVSIZE <= IVSIZE
1337 SvANY(sv) = (XPVNV*)((char*)&(sv->sv_u.svu_nv) - STRUCT_OFFSET(XPVNV, xnv_u.xnv_nv));
1339 SvANY(sv) = new_XNV();
1345 assert(new_type_details->body_size);
1348 assert(new_type_details->arena);
1349 assert(new_type_details->arena_size);
1350 /* This points to the start of the allocated area. */
1351 new_body_inline(new_body, new_type);
1352 Zero(new_body, new_type_details->body_size, char);
1353 new_body = ((char *)new_body) - new_type_details->offset;
1355 /* We always allocated the full length item with PURIFY. To do this
1356 we fake things so that arena is false for all 16 types.. */
1357 new_body = new_NOARENAZ(new_type_details);
1359 SvANY(sv) = new_body;
1360 if (new_type == SVt_PVAV) {
1364 if (old_type_details->body_size) {
1367 /* It will have been zeroed when the new body was allocated.
1368 Lets not write to it, in case it confuses a write-back
1374 #ifndef NODEFAULT_SHAREKEYS
1375 HvSHAREKEYS_on(sv); /* key-sharing on by default */
1377 /* start with PERL_HASH_DEFAULT_HvMAX+1 buckets: */
1378 HvMAX(sv) = PERL_HASH_DEFAULT_HvMAX;
1381 /* SVt_NULL isn't the only thing upgraded to AV or HV.
1382 The target created by newSVrv also is, and it can have magic.
1383 However, it never has SvPVX set.
1385 if (old_type == SVt_IV) {
1387 } else if (old_type >= SVt_PV) {
1388 assert(SvPVX_const(sv) == 0);
1391 if (old_type >= SVt_PVMG) {
1392 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic);
1393 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash);
1395 sv->sv_u.svu_array = NULL; /* or svu_hash */
1400 /* XXX Is this still needed? Was it ever needed? Surely as there is
1401 no route from NV to PVIV, NOK can never be true */
1402 assert(!SvNOKp(sv));
1415 assert(new_type_details->body_size);
1416 /* We always allocated the full length item with PURIFY. To do this
1417 we fake things so that arena is false for all 16 types.. */
1418 if(new_type_details->arena) {
1419 /* This points to the start of the allocated area. */
1420 new_body_inline(new_body, new_type);
1421 Zero(new_body, new_type_details->body_size, char);
1422 new_body = ((char *)new_body) - new_type_details->offset;
1424 new_body = new_NOARENAZ(new_type_details);
1426 SvANY(sv) = new_body;
1428 if (old_type_details->copy) {
1429 /* There is now the potential for an upgrade from something without
1430 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */
1431 int offset = old_type_details->offset;
1432 int length = old_type_details->copy;
1434 if (new_type_details->offset > old_type_details->offset) {
1435 const int difference
1436 = new_type_details->offset - old_type_details->offset;
1437 offset += difference;
1438 length -= difference;
1440 assert (length >= 0);
1442 Copy((char *)old_body + offset, (char *)new_body + offset, length,
1446 #ifndef NV_ZERO_IS_ALLBITS_ZERO
1447 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a
1448 * correct 0.0 for us. Otherwise, if the old body didn't have an
1449 * NV slot, but the new one does, then we need to initialise the
1450 * freshly created NV slot with whatever the correct bit pattern is
1452 if (old_type_details->zero_nv && !new_type_details->zero_nv
1453 && !isGV_with_GP(sv))
1457 if (UNLIKELY(new_type == SVt_PVIO)) {
1458 IO * const io = MUTABLE_IO(sv);
1459 GV *iogv = gv_fetchpvs("IO::File::", GV_ADD, SVt_PVHV);
1462 /* Clear the stashcache because a new IO could overrule a package
1464 DEBUG_o(Perl_deb(aTHX_ "sv_upgrade clearing PL_stashcache\n"));
1465 hv_clear(PL_stashcache);
1467 SvSTASH_set(io, MUTABLE_HV(SvREFCNT_inc(GvHV(iogv))));
1468 IoPAGE_LEN(sv) = 60;
1470 if (UNLIKELY(new_type == SVt_REGEXP))
1471 sv->sv_u.svu_rx = (regexp *)new_body;
1472 else if (old_type < SVt_PV) {
1473 /* referant will be NULL unless the old type was SVt_IV emulating
1475 sv->sv_u.svu_rv = referant;
1479 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1480 (unsigned long)new_type);
1483 /* if this is zero, this is a body-less SVt_NULL, SVt_IV/SVt_RV,
1484 and sometimes SVt_NV */
1485 if (old_type_details->body_size) {
1489 /* Note that there is an assumption that all bodies of types that
1490 can be upgraded came from arenas. Only the more complex non-
1491 upgradable types are allowed to be directly malloc()ed. */
1492 assert(old_type_details->arena);
1493 del_body((void*)((char*)old_body + old_type_details->offset),
1494 &PL_body_roots[old_type]);
1500 =for apidoc sv_backoff
1502 Remove any string offset. You should normally use the C<SvOOK_off> macro
1509 Perl_sv_backoff(SV *const sv)
1512 const char * const s = SvPVX_const(sv);
1514 PERL_ARGS_ASSERT_SV_BACKOFF;
1517 assert(SvTYPE(sv) != SVt_PVHV);
1518 assert(SvTYPE(sv) != SVt_PVAV);
1520 SvOOK_offset(sv, delta);
1522 SvLEN_set(sv, SvLEN(sv) + delta);
1523 SvPV_set(sv, SvPVX(sv) - delta);
1524 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
1525 SvFLAGS(sv) &= ~SVf_OOK;
1532 Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1533 upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1534 Use the C<SvGROW> wrapper instead.
1539 static void S_sv_uncow(pTHX_ SV * const sv, const U32 flags);
1542 Perl_sv_grow(pTHX_ SV *const sv, STRLEN newlen)
1546 PERL_ARGS_ASSERT_SV_GROW;
1550 if (SvTYPE(sv) < SVt_PV) {
1551 sv_upgrade(sv, SVt_PV);
1552 s = SvPVX_mutable(sv);
1554 else if (SvOOK(sv)) { /* pv is offset? */
1556 s = SvPVX_mutable(sv);
1557 if (newlen > SvLEN(sv))
1558 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
1562 if (SvIsCOW(sv)) S_sv_uncow(aTHX_ sv, 0);
1563 s = SvPVX_mutable(sv);
1566 #ifdef PERL_NEW_COPY_ON_WRITE
1567 /* the new COW scheme uses SvPVX(sv)[SvLEN(sv)-1] (if spare)
1568 * to store the COW count. So in general, allocate one more byte than
1569 * asked for, to make it likely this byte is always spare: and thus
1570 * make more strings COW-able.
1571 * If the new size is a big power of two, don't bother: we assume the
1572 * caller wanted a nice 2^N sized block and will be annoyed at getting
1578 #if defined(PERL_USE_MALLOC_SIZE) && defined(Perl_safesysmalloc_size)
1579 #define PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1582 if (newlen > SvLEN(sv)) { /* need more room? */
1583 STRLEN minlen = SvCUR(sv);
1584 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
1585 if (newlen < minlen)
1587 #ifndef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1589 /* Don't round up on the first allocation, as odds are pretty good that
1590 * the initial request is accurate as to what is really needed */
1592 newlen = PERL_STRLEN_ROUNDUP(newlen);
1595 if (SvLEN(sv) && s) {
1596 s = (char*)saferealloc(s, newlen);
1599 s = (char*)safemalloc(newlen);
1600 if (SvPVX_const(sv) && SvCUR(sv)) {
1601 Move(SvPVX_const(sv), s, (newlen < SvCUR(sv)) ? newlen : SvCUR(sv), char);
1605 #ifdef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1606 /* Do this here, do it once, do it right, and then we will never get
1607 called back into sv_grow() unless there really is some growing
1609 SvLEN_set(sv, Perl_safesysmalloc_size(s));
1611 SvLEN_set(sv, newlen);
1618 =for apidoc sv_setiv
1620 Copies an integer into the given SV, upgrading first if necessary.
1621 Does not handle 'set' magic. See also C<sv_setiv_mg>.
1627 Perl_sv_setiv(pTHX_ SV *const sv, const IV i)
1629 PERL_ARGS_ASSERT_SV_SETIV;
1631 SV_CHECK_THINKFIRST_COW_DROP(sv);
1632 switch (SvTYPE(sv)) {
1635 sv_upgrade(sv, SVt_IV);
1638 sv_upgrade(sv, SVt_PVIV);
1642 if (!isGV_with_GP(sv))
1649 /* diag_listed_as: Can't coerce %s to %s in %s */
1650 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1654 (void)SvIOK_only(sv); /* validate number */
1660 =for apidoc sv_setiv_mg
1662 Like C<sv_setiv>, but also handles 'set' magic.
1668 Perl_sv_setiv_mg(pTHX_ SV *const sv, const IV i)
1670 PERL_ARGS_ASSERT_SV_SETIV_MG;
1677 =for apidoc sv_setuv
1679 Copies an unsigned integer into the given SV, upgrading first if necessary.
1680 Does not handle 'set' magic. See also C<sv_setuv_mg>.
1686 Perl_sv_setuv(pTHX_ SV *const sv, const UV u)
1688 PERL_ARGS_ASSERT_SV_SETUV;
1690 /* With the if statement to ensure that integers are stored as IVs whenever
1692 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
1695 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1697 If you wish to remove the following if statement, so that this routine
1698 (and its callers) always return UVs, please benchmark to see what the
1699 effect is. Modern CPUs may be different. Or may not :-)
1701 if (u <= (UV)IV_MAX) {
1702 sv_setiv(sv, (IV)u);
1711 =for apidoc sv_setuv_mg
1713 Like C<sv_setuv>, but also handles 'set' magic.
1719 Perl_sv_setuv_mg(pTHX_ SV *const sv, const UV u)
1721 PERL_ARGS_ASSERT_SV_SETUV_MG;
1728 =for apidoc sv_setnv
1730 Copies a double into the given SV, upgrading first if necessary.
1731 Does not handle 'set' magic. See also C<sv_setnv_mg>.
1737 Perl_sv_setnv(pTHX_ SV *const sv, const NV num)
1739 PERL_ARGS_ASSERT_SV_SETNV;
1741 SV_CHECK_THINKFIRST_COW_DROP(sv);
1742 switch (SvTYPE(sv)) {
1745 sv_upgrade(sv, SVt_NV);
1749 sv_upgrade(sv, SVt_PVNV);
1753 if (!isGV_with_GP(sv))
1760 /* diag_listed_as: Can't coerce %s to %s in %s */
1761 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
1766 (void)SvNOK_only(sv); /* validate number */
1771 =for apidoc sv_setnv_mg
1773 Like C<sv_setnv>, but also handles 'set' magic.
1779 Perl_sv_setnv_mg(pTHX_ SV *const sv, const NV num)
1781 PERL_ARGS_ASSERT_SV_SETNV_MG;
1787 /* Return a cleaned-up, printable version of sv, for non-numeric, or
1788 * not incrementable warning display.
1789 * Originally part of S_not_a_number().
1790 * The return value may be != tmpbuf.
1794 S_sv_display(pTHX_ SV *const sv, char *tmpbuf, STRLEN tmpbuf_size) {
1797 PERL_ARGS_ASSERT_SV_DISPLAY;
1800 SV *dsv = newSVpvs_flags("", SVs_TEMP);
1801 pv = sv_uni_display(dsv, sv, 10, UNI_DISPLAY_ISPRINT);
1804 const char * const limit = tmpbuf + tmpbuf_size - 8;
1805 /* each *s can expand to 4 chars + "...\0",
1806 i.e. need room for 8 chars */
1808 const char *s = SvPVX_const(sv);
1809 const char * const end = s + SvCUR(sv);
1810 for ( ; s < end && d < limit; s++ ) {
1812 if (! isASCII(ch) && !isPRINT_LC(ch)) {
1816 /* Map to ASCII "equivalent" of Latin1 */
1817 ch = LATIN1_TO_NATIVE(NATIVE_TO_LATIN1(ch) & 127);
1823 else if (ch == '\r') {
1827 else if (ch == '\f') {
1831 else if (ch == '\\') {
1835 else if (ch == '\0') {
1839 else if (isPRINT_LC(ch))
1858 /* Print an "isn't numeric" warning, using a cleaned-up,
1859 * printable version of the offending string
1863 S_not_a_number(pTHX_ SV *const sv)
1868 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1870 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1873 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1874 /* diag_listed_as: Argument "%s" isn't numeric%s */
1875 "Argument \"%s\" isn't numeric in %s", pv,
1878 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1879 /* diag_listed_as: Argument "%s" isn't numeric%s */
1880 "Argument \"%s\" isn't numeric", pv);
1884 S_not_incrementable(pTHX_ SV *const sv) {
1888 PERL_ARGS_ASSERT_NOT_INCREMENTABLE;
1890 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1892 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1893 "Argument \"%s\" treated as 0 in increment (++)", pv);
1897 =for apidoc looks_like_number
1899 Test if the content of an SV looks like a number (or is a number).
1900 C<Inf> and C<Infinity> are treated as numbers (so will not issue a
1901 non-numeric warning), even if your atof() doesn't grok them. Get-magic is
1908 Perl_looks_like_number(pTHX_ SV *const sv)
1913 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1915 if (SvPOK(sv) || SvPOKp(sv)) {
1916 sbegin = SvPV_nomg_const(sv, len);
1919 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
1920 return grok_number(sbegin, len, NULL);
1924 S_glob_2number(pTHX_ GV * const gv)
1926 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1928 /* We know that all GVs stringify to something that is not-a-number,
1929 so no need to test that. */
1930 if (ckWARN(WARN_NUMERIC))
1932 SV *const buffer = sv_newmortal();
1933 gv_efullname3(buffer, gv, "*");
1934 not_a_number(buffer);
1936 /* We just want something true to return, so that S_sv_2iuv_common
1937 can tail call us and return true. */
1941 /* Actually, ISO C leaves conversion of UV to IV undefined, but
1942 until proven guilty, assume that things are not that bad... */
1947 As 64 bit platforms often have an NV that doesn't preserve all bits of
1948 an IV (an assumption perl has been based on to date) it becomes necessary
1949 to remove the assumption that the NV always carries enough precision to
1950 recreate the IV whenever needed, and that the NV is the canonical form.
1951 Instead, IV/UV and NV need to be given equal rights. So as to not lose
1952 precision as a side effect of conversion (which would lead to insanity
1953 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
1954 1) to distinguish between IV/UV/NV slots that have a valid conversion cached
1955 where precision was lost, and IV/UV/NV slots that have a valid conversion
1956 which has lost no precision
1957 2) to ensure that if a numeric conversion to one form is requested that
1958 would lose precision, the precise conversion (or differently
1959 imprecise conversion) is also performed and cached, to prevent
1960 requests for different numeric formats on the same SV causing
1961 lossy conversion chains. (lossless conversion chains are perfectly
1966 SvIOKp is true if the IV slot contains a valid value
1967 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1968 SvNOKp is true if the NV slot contains a valid value
1969 SvNOK is true only if the NV value is accurate
1972 while converting from PV to NV, check to see if converting that NV to an
1973 IV(or UV) would lose accuracy over a direct conversion from PV to
1974 IV(or UV). If it would, cache both conversions, return NV, but mark
1975 SV as IOK NOKp (ie not NOK).
1977 While converting from PV to IV, check to see if converting that IV to an
1978 NV would lose accuracy over a direct conversion from PV to NV. If it
1979 would, cache both conversions, flag similarly.
1981 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
1982 correctly because if IV & NV were set NV *always* overruled.
1983 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
1984 changes - now IV and NV together means that the two are interchangeable:
1985 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
1987 The benefit of this is that operations such as pp_add know that if
1988 SvIOK is true for both left and right operands, then integer addition
1989 can be used instead of floating point (for cases where the result won't
1990 overflow). Before, floating point was always used, which could lead to
1991 loss of precision compared with integer addition.
1993 * making IV and NV equal status should make maths accurate on 64 bit
1995 * may speed up maths somewhat if pp_add and friends start to use
1996 integers when possible instead of fp. (Hopefully the overhead in
1997 looking for SvIOK and checking for overflow will not outweigh the
1998 fp to integer speedup)
1999 * will slow down integer operations (callers of SvIV) on "inaccurate"
2000 values, as the change from SvIOK to SvIOKp will cause a call into
2001 sv_2iv each time rather than a macro access direct to the IV slot
2002 * should speed up number->string conversion on integers as IV is
2003 favoured when IV and NV are equally accurate
2005 ####################################################################
2006 You had better be using SvIOK_notUV if you want an IV for arithmetic:
2007 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
2008 On the other hand, SvUOK is true iff UV.
2009 ####################################################################
2011 Your mileage will vary depending your CPU's relative fp to integer
2015 #ifndef NV_PRESERVES_UV
2016 # define IS_NUMBER_UNDERFLOW_IV 1
2017 # define IS_NUMBER_UNDERFLOW_UV 2
2018 # define IS_NUMBER_IV_AND_UV 2
2019 # define IS_NUMBER_OVERFLOW_IV 4
2020 # define IS_NUMBER_OVERFLOW_UV 5
2022 /* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
2024 /* For sv_2nv these three cases are "SvNOK and don't bother casting" */
2026 S_sv_2iuv_non_preserve(pTHX_ SV *const sv
2032 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
2033 PERL_UNUSED_CONTEXT;
2035 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));
2036 if (SvNVX(sv) < (NV)IV_MIN) {
2037 (void)SvIOKp_on(sv);
2039 SvIV_set(sv, IV_MIN);
2040 return IS_NUMBER_UNDERFLOW_IV;
2042 if (SvNVX(sv) > (NV)UV_MAX) {
2043 (void)SvIOKp_on(sv);
2046 SvUV_set(sv, UV_MAX);
2047 return IS_NUMBER_OVERFLOW_UV;
2049 (void)SvIOKp_on(sv);
2051 /* Can't use strtol etc to convert this string. (See truth table in
2053 if (SvNVX(sv) <= (UV)IV_MAX) {
2054 SvIV_set(sv, I_V(SvNVX(sv)));
2055 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2056 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
2058 /* Integer is imprecise. NOK, IOKp */
2060 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
2063 SvUV_set(sv, U_V(SvNVX(sv)));
2064 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2065 if (SvUVX(sv) == UV_MAX) {
2066 /* As we know that NVs don't preserve UVs, UV_MAX cannot
2067 possibly be preserved by NV. Hence, it must be overflow.
2069 return IS_NUMBER_OVERFLOW_UV;
2071 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
2073 /* Integer is imprecise. NOK, IOKp */
2075 return IS_NUMBER_OVERFLOW_IV;
2077 #endif /* !NV_PRESERVES_UV*/
2079 /* If numtype is infnan, set the NV of the sv accordingly.
2080 * If numtype is anything else, try setting the NV using Atof(PV). */
2082 S_sv_setnv(pTHX_ SV* sv, int numtype)
2084 bool pok = cBOOL(SvPOK(sv));
2086 if ((numtype & IS_NUMBER_INFINITY)) {
2087 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -NV_INF : NV_INF);
2090 else if ((numtype & IS_NUMBER_NAN)) {
2091 SvNV_set(sv, NV_NAN);
2095 SvNV_set(sv, Atof(SvPVX_const(sv)));
2096 /* Purposefully no true nok here, since we don't want to blow
2097 * away the possible IOK/UV of an existing sv. */
2100 SvNOK_only(sv); /* No IV or UV please, this is pure infnan. */
2102 SvPOK_on(sv); /* PV is okay, though. */
2107 S_sv_2iuv_common(pTHX_ SV *const sv)
2109 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
2112 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
2113 * without also getting a cached IV/UV from it at the same time
2114 * (ie PV->NV conversion should detect loss of accuracy and cache
2115 * IV or UV at same time to avoid this. */
2116 /* IV-over-UV optimisation - choose to cache IV if possible */
2118 if (SvTYPE(sv) == SVt_NV)
2119 sv_upgrade(sv, SVt_PVNV);
2121 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
2122 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
2123 certainly cast into the IV range at IV_MAX, whereas the correct
2124 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
2126 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2127 if (Perl_isnan(SvNVX(sv))) {
2133 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2134 SvIV_set(sv, I_V(SvNVX(sv)));
2135 if (SvNVX(sv) == (NV) SvIVX(sv)
2136 #ifndef NV_PRESERVES_UV
2137 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2138 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2139 /* Don't flag it as "accurately an integer" if the number
2140 came from a (by definition imprecise) NV operation, and
2141 we're outside the range of NV integer precision */
2145 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2147 /* scalar has trailing garbage, eg "42a" */
2149 DEBUG_c(PerlIO_printf(Perl_debug_log,
2150 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (precise)\n",
2156 /* IV not precise. No need to convert from PV, as NV
2157 conversion would already have cached IV if it detected
2158 that PV->IV would be better than PV->NV->IV
2159 flags already correct - don't set public IOK. */
2160 DEBUG_c(PerlIO_printf(Perl_debug_log,
2161 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (imprecise)\n",
2166 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2167 but the cast (NV)IV_MIN rounds to a the value less (more
2168 negative) than IV_MIN which happens to be equal to SvNVX ??
2169 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2170 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2171 (NV)UVX == NVX are both true, but the values differ. :-(
2172 Hopefully for 2s complement IV_MIN is something like
2173 0x8000000000000000 which will be exact. NWC */
2176 SvUV_set(sv, U_V(SvNVX(sv)));
2178 (SvNVX(sv) == (NV) SvUVX(sv))
2179 #ifndef NV_PRESERVES_UV
2180 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2181 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2182 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2183 /* Don't flag it as "accurately an integer" if the number
2184 came from a (by definition imprecise) NV operation, and
2185 we're outside the range of NV integer precision */
2191 DEBUG_c(PerlIO_printf(Perl_debug_log,
2192 "0x%"UVxf" 2iv(%"UVuf" => %"IVdf") (as unsigned)\n",
2198 else if (SvPOKp(sv)) {
2200 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2201 /* We want to avoid a possible problem when we cache an IV/ a UV which
2202 may be later translated to an NV, and the resulting NV is not
2203 the same as the direct translation of the initial string
2204 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2205 be careful to ensure that the value with the .456 is around if the
2206 NV value is requested in the future).
2208 This means that if we cache such an IV/a UV, we need to cache the
2209 NV as well. Moreover, we trade speed for space, and do not
2210 cache the NV if we are sure it's not needed.
2213 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2214 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2215 == IS_NUMBER_IN_UV) {
2216 /* It's definitely an integer, only upgrade to PVIV */
2217 if (SvTYPE(sv) < SVt_PVIV)
2218 sv_upgrade(sv, SVt_PVIV);
2220 } else if (SvTYPE(sv) < SVt_PVNV)
2221 sv_upgrade(sv, SVt_PVNV);
2223 if ((numtype & (IS_NUMBER_INFINITY | IS_NUMBER_NAN))) {
2224 S_sv_setnv(aTHX_ sv, numtype);
2228 /* If NVs preserve UVs then we only use the UV value if we know that
2229 we aren't going to call atof() below. If NVs don't preserve UVs
2230 then the value returned may have more precision than atof() will
2231 return, even though value isn't perfectly accurate. */
2232 if ((numtype & (IS_NUMBER_IN_UV
2233 #ifdef NV_PRESERVES_UV
2236 )) == IS_NUMBER_IN_UV) {
2237 /* This won't turn off the public IOK flag if it was set above */
2238 (void)SvIOKp_on(sv);
2240 if (!(numtype & IS_NUMBER_NEG)) {
2242 if (value <= (UV)IV_MAX) {
2243 SvIV_set(sv, (IV)value);
2245 /* it didn't overflow, and it was positive. */
2246 SvUV_set(sv, value);
2250 /* 2s complement assumption */
2251 if (value <= (UV)IV_MIN) {
2252 SvIV_set(sv, -(IV)value);
2254 /* Too negative for an IV. This is a double upgrade, but
2255 I'm assuming it will be rare. */
2256 if (SvTYPE(sv) < SVt_PVNV)
2257 sv_upgrade(sv, SVt_PVNV);
2261 SvNV_set(sv, -(NV)value);
2262 SvIV_set(sv, IV_MIN);
2266 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2267 will be in the previous block to set the IV slot, and the next
2268 block to set the NV slot. So no else here. */
2270 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2271 != IS_NUMBER_IN_UV) {
2272 /* It wasn't an (integer that doesn't overflow the UV). */
2273 S_sv_setnv(aTHX_ sv, numtype);
2275 if (! numtype && ckWARN(WARN_NUMERIC))
2278 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%" NVgf ")\n",
2279 PTR2UV(sv), SvNVX(sv)));
2281 #ifdef NV_PRESERVES_UV
2282 (void)SvIOKp_on(sv);
2284 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2285 if (Perl_isnan(SvNVX(sv))) {
2291 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2292 SvIV_set(sv, I_V(SvNVX(sv)));
2293 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2296 NOOP; /* Integer is imprecise. NOK, IOKp */
2298 /* UV will not work better than IV */
2300 if (SvNVX(sv) > (NV)UV_MAX) {
2302 /* Integer is inaccurate. NOK, IOKp, is UV */
2303 SvUV_set(sv, UV_MAX);
2305 SvUV_set(sv, U_V(SvNVX(sv)));
2306 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2307 NV preservse UV so can do correct comparison. */
2308 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2311 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
2316 #else /* NV_PRESERVES_UV */
2317 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2318 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
2319 /* The IV/UV slot will have been set from value returned by
2320 grok_number above. The NV slot has just been set using
2323 assert (SvIOKp(sv));
2325 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2326 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2327 /* Small enough to preserve all bits. */
2328 (void)SvIOKp_on(sv);
2330 SvIV_set(sv, I_V(SvNVX(sv)));
2331 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2333 /* Assumption: first non-preserved integer is < IV_MAX,
2334 this NV is in the preserved range, therefore: */
2335 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2337 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);
2341 0 0 already failed to read UV.
2342 0 1 already failed to read UV.
2343 1 0 you won't get here in this case. IV/UV
2344 slot set, public IOK, Atof() unneeded.
2345 1 1 already read UV.
2346 so there's no point in sv_2iuv_non_preserve() attempting
2347 to use atol, strtol, strtoul etc. */
2349 sv_2iuv_non_preserve (sv, numtype);
2351 sv_2iuv_non_preserve (sv);
2355 #endif /* NV_PRESERVES_UV */
2356 /* It might be more code efficient to go through the entire logic above
2357 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2358 gets complex and potentially buggy, so more programmer efficient
2359 to do it this way, by turning off the public flags: */
2361 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2365 if (isGV_with_GP(sv))
2366 return glob_2number(MUTABLE_GV(sv));
2368 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2370 if (SvTYPE(sv) < SVt_IV)
2371 /* Typically the caller expects that sv_any is not NULL now. */
2372 sv_upgrade(sv, SVt_IV);
2373 /* Return 0 from the caller. */
2380 =for apidoc sv_2iv_flags
2382 Return the integer value of an SV, doing any necessary string
2383 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2384 Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2390 Perl_sv_2iv_flags(pTHX_ SV *const sv, const I32 flags)
2392 PERL_ARGS_ASSERT_SV_2IV_FLAGS;
2394 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2395 && SvTYPE(sv) != SVt_PVFM);
2397 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2403 if (flags & SV_SKIP_OVERLOAD)
2405 tmpstr = AMG_CALLunary(sv, numer_amg);
2406 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2407 return SvIV(tmpstr);
2410 return PTR2IV(SvRV(sv));
2413 if (SvVALID(sv) || isREGEXP(sv)) {
2414 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2415 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2416 In practice they are extremely unlikely to actually get anywhere
2417 accessible by user Perl code - the only way that I'm aware of is when
2418 a constant subroutine which is used as the second argument to index.
2420 Regexps have no SvIVX and SvNVX fields.
2422 assert(isREGEXP(sv) || SvPOKp(sv));
2425 const char * const ptr =
2426 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2428 = grok_number(ptr, SvCUR(sv), &value);
2430 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2431 == IS_NUMBER_IN_UV) {
2432 /* It's definitely an integer */
2433 if (numtype & IS_NUMBER_NEG) {
2434 if (value < (UV)IV_MIN)
2437 if (value < (UV)IV_MAX)
2442 /* Quite wrong but no good choices. */
2443 if ((numtype & IS_NUMBER_INFINITY)) {
2444 return (numtype & IS_NUMBER_NEG) ? IV_MIN : IV_MAX;
2445 } else if ((numtype & IS_NUMBER_NAN)) {
2446 return 0; /* So wrong. */
2450 if (ckWARN(WARN_NUMERIC))
2453 return I_V(Atof(ptr));
2457 if (SvTHINKFIRST(sv)) {
2458 #ifdef PERL_OLD_COPY_ON_WRITE
2460 sv_force_normal_flags(sv, 0);
2463 if (SvREADONLY(sv) && !SvOK(sv)) {
2464 if (ckWARN(WARN_UNINITIALIZED))
2471 if (S_sv_2iuv_common(aTHX_ sv))
2475 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"IVdf")\n",
2476 PTR2UV(sv),SvIVX(sv)));
2477 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
2481 =for apidoc sv_2uv_flags
2483 Return the unsigned integer value of an SV, doing any necessary string
2484 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2485 Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
2491 Perl_sv_2uv_flags(pTHX_ SV *const sv, const I32 flags)
2493 PERL_ARGS_ASSERT_SV_2UV_FLAGS;
2495 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2501 if (flags & SV_SKIP_OVERLOAD)
2503 tmpstr = AMG_CALLunary(sv, numer_amg);
2504 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2505 return SvUV(tmpstr);
2508 return PTR2UV(SvRV(sv));
2511 if (SvVALID(sv) || isREGEXP(sv)) {
2512 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2513 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2514 Regexps have no SvIVX and SvNVX fields. */
2515 assert(isREGEXP(sv) || SvPOKp(sv));
2518 const char * const ptr =
2519 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2521 = grok_number(ptr, SvCUR(sv), &value);
2523 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2524 == IS_NUMBER_IN_UV) {
2525 /* It's definitely an integer */
2526 if (!(numtype & IS_NUMBER_NEG))
2530 /* Quite wrong but no good choices. */
2531 if ((numtype & IS_NUMBER_INFINITY)) {
2532 return UV_MAX; /* So wrong. */
2533 } else if ((numtype & IS_NUMBER_NAN)) {
2534 return 0; /* So wrong. */
2538 if (ckWARN(WARN_NUMERIC))
2541 return U_V(Atof(ptr));
2545 if (SvTHINKFIRST(sv)) {
2546 #ifdef PERL_OLD_COPY_ON_WRITE
2548 sv_force_normal_flags(sv, 0);
2551 if (SvREADONLY(sv) && !SvOK(sv)) {
2552 if (ckWARN(WARN_UNINITIALIZED))
2559 if (S_sv_2iuv_common(aTHX_ sv))
2563 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2uv(%"UVuf")\n",
2564 PTR2UV(sv),SvUVX(sv)));
2565 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
2569 =for apidoc sv_2nv_flags
2571 Return the num value of an SV, doing any necessary string or integer
2572 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2573 Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
2579 Perl_sv_2nv_flags(pTHX_ SV *const sv, const I32 flags)
2581 PERL_ARGS_ASSERT_SV_2NV_FLAGS;
2583 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2584 && SvTYPE(sv) != SVt_PVFM);
2585 if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) {
2586 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2587 the same flag bit as SVf_IVisUV, so must not let them cache NVs.
2588 Regexps have no SvIVX and SvNVX fields. */
2590 if (flags & SV_GMAGIC)
2594 if (SvPOKp(sv) && !SvIOKp(sv)) {
2595 ptr = SvPVX_const(sv);
2597 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
2598 !grok_number(ptr, SvCUR(sv), NULL))
2604 return (NV)SvUVX(sv);
2606 return (NV)SvIVX(sv);
2612 ptr = RX_WRAPPED((REGEXP *)sv);
2615 assert(SvTYPE(sv) >= SVt_PVMG);
2616 /* This falls through to the report_uninit near the end of the
2618 } else if (SvTHINKFIRST(sv)) {
2623 if (flags & SV_SKIP_OVERLOAD)
2625 tmpstr = AMG_CALLunary(sv, numer_amg);
2626 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2627 return SvNV(tmpstr);
2630 return PTR2NV(SvRV(sv));
2632 #ifdef PERL_OLD_COPY_ON_WRITE
2634 sv_force_normal_flags(sv, 0);
2637 if (SvREADONLY(sv) && !SvOK(sv)) {
2638 if (ckWARN(WARN_UNINITIALIZED))
2643 if (SvTYPE(sv) < SVt_NV) {
2644 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2645 sv_upgrade(sv, SVt_NV);
2647 STORE_NUMERIC_LOCAL_SET_STANDARD();
2648 PerlIO_printf(Perl_debug_log,
2649 "0x%"UVxf" num(%" NVgf ")\n",
2650 PTR2UV(sv), SvNVX(sv));
2651 RESTORE_NUMERIC_LOCAL();
2654 else if (SvTYPE(sv) < SVt_PVNV)
2655 sv_upgrade(sv, SVt_PVNV);
2660 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
2661 #ifdef NV_PRESERVES_UV
2667 /* Only set the public NV OK flag if this NV preserves the IV */
2668 /* Check it's not 0xFFFFFFFFFFFFFFFF */
2670 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
2671 : (SvIVX(sv) == I_V(SvNVX(sv))))
2677 else if (SvPOKp(sv)) {
2679 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2680 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
2682 #ifdef NV_PRESERVES_UV
2683 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2684 == IS_NUMBER_IN_UV) {
2685 /* It's definitely an integer */
2686 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
2688 S_sv_setnv(aTHX_ sv, numtype);
2695 SvNV_set(sv, Atof(SvPVX_const(sv)));
2696 /* Only set the public NV OK flag if this NV preserves the value in
2697 the PV at least as well as an IV/UV would.
2698 Not sure how to do this 100% reliably. */
2699 /* if that shift count is out of range then Configure's test is
2700 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2702 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2703 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2704 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
2705 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2706 /* Can't use strtol etc to convert this string, so don't try.
2707 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2710 /* value has been set. It may not be precise. */
2711 if ((numtype & IS_NUMBER_NEG) && (value > (UV)IV_MIN)) {
2712 /* 2s complement assumption for (UV)IV_MIN */
2713 SvNOK_on(sv); /* Integer is too negative. */
2718 if (numtype & IS_NUMBER_NEG) {
2719 SvIV_set(sv, -(IV)value);
2720 } else if (value <= (UV)IV_MAX) {
2721 SvIV_set(sv, (IV)value);
2723 SvUV_set(sv, value);
2727 if (numtype & IS_NUMBER_NOT_INT) {
2728 /* I believe that even if the original PV had decimals,
2729 they are lost beyond the limit of the FP precision.
2730 However, neither is canonical, so both only get p
2731 flags. NWC, 2000/11/25 */
2732 /* Both already have p flags, so do nothing */
2734 const NV nv = SvNVX(sv);
2735 /* XXX should this spot have NAN_COMPARE_BROKEN, too? */
2736 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2737 if (SvIVX(sv) == I_V(nv)) {
2740 /* It had no "." so it must be integer. */
2744 /* between IV_MAX and NV(UV_MAX).
2745 Could be slightly > UV_MAX */
2747 if (numtype & IS_NUMBER_NOT_INT) {
2748 /* UV and NV both imprecise. */
2750 const UV nv_as_uv = U_V(nv);
2752 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2761 /* It might be more code efficient to go through the entire logic above
2762 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2763 gets complex and potentially buggy, so more programmer efficient
2764 to do it this way, by turning off the public flags: */
2766 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2767 #endif /* NV_PRESERVES_UV */
2770 if (isGV_with_GP(sv)) {
2771 glob_2number(MUTABLE_GV(sv));
2775 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2777 assert (SvTYPE(sv) >= SVt_NV);
2778 /* Typically the caller expects that sv_any is not NULL now. */
2779 /* XXX Ilya implies that this is a bug in callers that assume this
2780 and ideally should be fixed. */
2784 STORE_NUMERIC_LOCAL_SET_STANDARD();
2785 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2nv(%" NVgf ")\n",
2786 PTR2UV(sv), SvNVX(sv));
2787 RESTORE_NUMERIC_LOCAL();
2795 Return an SV with the numeric value of the source SV, doing any necessary
2796 reference or overload conversion. You must use the C<SvNUM(sv)> macro to
2797 access this function.
2803 Perl_sv_2num(pTHX_ SV *const sv)
2805 PERL_ARGS_ASSERT_SV_2NUM;
2810 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
2811 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
2812 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2813 return sv_2num(tmpsv);
2815 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2818 /* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2819 * UV as a string towards the end of buf, and return pointers to start and
2822 * We assume that buf is at least TYPE_CHARS(UV) long.
2826 S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
2828 char *ptr = buf + TYPE_CHARS(UV);
2829 char * const ebuf = ptr;
2832 PERL_ARGS_ASSERT_UIV_2BUF;
2844 *--ptr = '0' + (char)(uv % 10);
2852 /* Helper for sv_2pv_flags and sv_vcatpvfn_flags. If the NV is an
2853 * infinity or a not-a-number, writes the appropriate strings to the
2854 * buffer, including a zero byte. On success returns the written length,
2855 * excluding the zero byte, on failure (not an infinity, not a nan, or the
2856 * maxlen too small) returns zero.
2858 * XXX for "Inf", "-Inf", and "NaN", we could have three read-only
2859 * shared string constants we point to, instead of generating a new
2860 * string for each instance. */
2862 S_infnan_2pv(NV nv, char* buffer, size_t maxlen) {
2863 assert(maxlen >= 4);
2864 if (maxlen < 4) /* "Inf\0", "NaN\0" */
2868 if (Perl_isinf(nv)) {
2870 if (maxlen < 5) /* "-Inf\0" */
2877 } else if (Perl_isnan(nv)) {
2881 /* XXX optionally output the payload mantissa bits as
2882 * "(unsigned)" (to match the nan("...") C99 function,
2883 * or maybe as "(0xhhh...)" would make more sense...
2884 * provide a format string so that the user can decide?
2885 * NOTE: would affect the maxlen and assert() logic.*/
2890 assert((s == buffer + 3) || (s == buffer + 4));
2892 return s - buffer - 1; /* -1: excluding the zero byte */
2897 =for apidoc sv_2pv_flags
2899 Returns a pointer to the string value of an SV, and sets *lp to its length.
2900 If flags includes SV_GMAGIC, does an mg_get() first. Coerces sv to a
2901 string if necessary. Normally invoked via the C<SvPV_flags> macro.
2902 C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
2908 Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
2912 PERL_ARGS_ASSERT_SV_2PV_FLAGS;
2914 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2915 && SvTYPE(sv) != SVt_PVFM);
2916 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2921 if (flags & SV_SKIP_OVERLOAD)
2923 tmpstr = AMG_CALLunary(sv, string_amg);
2924 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
2925 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2927 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2931 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2932 if (flags & SV_CONST_RETURN) {
2933 pv = (char *) SvPVX_const(tmpstr);
2935 pv = (flags & SV_MUTABLE_RETURN)
2936 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
2939 *lp = SvCUR(tmpstr);
2941 pv = sv_2pv_flags(tmpstr, lp, flags);
2954 SV *const referent = SvRV(sv);
2958 retval = buffer = savepvn("NULLREF", len);
2959 } else if (SvTYPE(referent) == SVt_REGEXP &&
2960 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
2961 amagic_is_enabled(string_amg))) {
2962 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
2966 /* If the regex is UTF-8 we want the containing scalar to
2967 have an UTF-8 flag too */
2974 *lp = RX_WRAPLEN(re);
2976 return RX_WRAPPED(re);
2978 const char *const typestr = sv_reftype(referent, 0);
2979 const STRLEN typelen = strlen(typestr);
2980 UV addr = PTR2UV(referent);
2981 const char *stashname = NULL;
2982 STRLEN stashnamelen = 0; /* hush, gcc */
2983 const char *buffer_end;
2985 if (SvOBJECT(referent)) {
2986 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
2989 stashname = HEK_KEY(name);
2990 stashnamelen = HEK_LEN(name);
2992 if (HEK_UTF8(name)) {
2998 stashname = "__ANON__";
3001 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
3002 + 2 * sizeof(UV) + 2 /* )\0 */;
3004 len = typelen + 3 /* (0x */
3005 + 2 * sizeof(UV) + 2 /* )\0 */;
3008 Newx(buffer, len, char);
3009 buffer_end = retval = buffer + len;
3011 /* Working backwards */
3015 *--retval = PL_hexdigit[addr & 15];
3016 } while (addr >>= 4);
3022 memcpy(retval, typestr, typelen);
3026 retval -= stashnamelen;
3027 memcpy(retval, stashname, stashnamelen);
3029 /* retval may not necessarily have reached the start of the
3031 assert (retval >= buffer);
3033 len = buffer_end - retval - 1; /* -1 for that \0 */
3045 if (flags & SV_MUTABLE_RETURN)
3046 return SvPVX_mutable(sv);
3047 if (flags & SV_CONST_RETURN)
3048 return (char *)SvPVX_const(sv);
3053 /* I'm assuming that if both IV and NV are equally valid then
3054 converting the IV is going to be more efficient */
3055 const U32 isUIOK = SvIsUV(sv);
3056 char buf[TYPE_CHARS(UV)];
3060 if (SvTYPE(sv) < SVt_PVIV)
3061 sv_upgrade(sv, SVt_PVIV);
3062 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
3064 /* inlined from sv_setpvn */
3065 s = SvGROW_mutable(sv, len + 1);
3066 Move(ptr, s, len, char);
3071 else if (SvNOK(sv)) {
3072 if (SvTYPE(sv) < SVt_PVNV)
3073 sv_upgrade(sv, SVt_PVNV);
3074 if (SvNVX(sv) == 0.0
3075 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
3076 && !Perl_isnan(SvNVX(sv))
3079 s = SvGROW_mutable(sv, 2);
3084 STRLEN size = 5; /* "-Inf\0" */
3086 s = SvGROW_mutable(sv, size);
3087 len = S_infnan_2pv(SvNVX(sv), s, size);
3093 /* some Xenix systems wipe out errno here */
3102 5 + /* exponent digits */
3106 s = SvGROW_mutable(sv, size);
3107 #ifndef USE_LOCALE_NUMERIC
3108 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3114 DECLARE_STORE_LC_NUMERIC_SET_TO_NEEDED();
3118 PL_numeric_radix_sv &&
3119 SvUTF8(PL_numeric_radix_sv);
3120 if (local_radix && SvLEN(PL_numeric_radix_sv) > 1) {
3121 size += SvLEN(PL_numeric_radix_sv) - 1;
3122 s = SvGROW_mutable(sv, size);
3125 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3127 /* If the radix character is UTF-8, and actually is in the
3128 * output, turn on the UTF-8 flag for the scalar */
3130 instr(s, SvPVX_const(PL_numeric_radix_sv))) {
3134 RESTORE_LC_NUMERIC();
3137 /* We don't call SvPOK_on(), because it may come to
3138 * pass that the locale changes so that the
3139 * stringification we just did is no longer correct. We
3140 * will have to re-stringify every time it is needed */
3147 else if (isGV_with_GP(sv)) {
3148 GV *const gv = MUTABLE_GV(sv);
3149 SV *const buffer = sv_newmortal();
3151 gv_efullname3(buffer, gv, "*");
3153 assert(SvPOK(buffer));
3157 *lp = SvCUR(buffer);
3158 return SvPVX(buffer);
3160 else if (isREGEXP(sv)) {
3161 if (lp) *lp = RX_WRAPLEN((REGEXP *)sv);
3162 return RX_WRAPPED((REGEXP *)sv);
3167 if (flags & SV_UNDEF_RETURNS_NULL)
3169 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
3171 /* Typically the caller expects that sv_any is not NULL now. */
3172 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
3173 sv_upgrade(sv, SVt_PV);
3178 const STRLEN len = s - SvPVX_const(sv);
3183 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
3184 PTR2UV(sv),SvPVX_const(sv)));
3185 if (flags & SV_CONST_RETURN)
3186 return (char *)SvPVX_const(sv);
3187 if (flags & SV_MUTABLE_RETURN)
3188 return SvPVX_mutable(sv);
3193 =for apidoc sv_copypv
3195 Copies a stringified representation of the source SV into the
3196 destination SV. Automatically performs any necessary mg_get and
3197 coercion of numeric values into strings. Guaranteed to preserve
3198 UTF8 flag even from overloaded objects. Similar in nature to
3199 sv_2pv[_flags] but operates directly on an SV instead of just the
3200 string. Mostly uses sv_2pv_flags to do its work, except when that
3201 would lose the UTF-8'ness of the PV.
3203 =for apidoc sv_copypv_nomg
3205 Like sv_copypv, but doesn't invoke get magic first.
3207 =for apidoc sv_copypv_flags
3209 Implementation of sv_copypv and sv_copypv_nomg. Calls get magic iff flags
3216 Perl_sv_copypv(pTHX_ SV *const dsv, SV *const ssv)
3218 PERL_ARGS_ASSERT_SV_COPYPV;
3220 sv_copypv_flags(dsv, ssv, 0);
3224 Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
3229 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
3231 s = SvPV_flags_const(ssv,len,(flags & SV_GMAGIC));
3232 sv_setpvn(dsv,s,len);
3240 =for apidoc sv_2pvbyte
3242 Return a pointer to the byte-encoded representation of the SV, and set *lp
3243 to its length. May cause the SV to be downgraded from UTF-8 as a
3246 Usually accessed via the C<SvPVbyte> macro.
3252 Perl_sv_2pvbyte(pTHX_ SV *sv, STRLEN *const lp)
3254 PERL_ARGS_ASSERT_SV_2PVBYTE;
3257 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3258 || isGV_with_GP(sv) || SvROK(sv)) {
3259 SV *sv2 = sv_newmortal();
3260 sv_copypv_nomg(sv2,sv);
3263 sv_utf8_downgrade(sv,0);
3264 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3268 =for apidoc sv_2pvutf8
3270 Return a pointer to the UTF-8-encoded representation of the SV, and set *lp
3271 to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3273 Usually accessed via the C<SvPVutf8> macro.
3279 Perl_sv_2pvutf8(pTHX_ SV *sv, STRLEN *const lp)
3281 PERL_ARGS_ASSERT_SV_2PVUTF8;
3283 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3284 || isGV_with_GP(sv) || SvROK(sv))
3285 sv = sv_mortalcopy(sv);
3288 sv_utf8_upgrade_nomg(sv);
3289 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3294 =for apidoc sv_2bool
3296 This macro is only used by sv_true() or its macro equivalent, and only if
3297 the latter's argument is neither SvPOK, SvIOK nor SvNOK.
3298 It calls sv_2bool_flags with the SV_GMAGIC flag.
3300 =for apidoc sv_2bool_flags
3302 This function is only used by sv_true() and friends, and only if
3303 the latter's argument is neither SvPOK, SvIOK nor SvNOK. If the flags
3304 contain SV_GMAGIC, then it does an mg_get() first.
3311 Perl_sv_2bool_flags(pTHX_ SV *sv, I32 flags)
3313 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
3316 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
3322 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
3323 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv)))) {
3326 if(SvGMAGICAL(sv)) {
3328 goto restart; /* call sv_2bool */
3330 /* expanded SvTRUE_common(sv, (flags = 0, goto restart)) */
3331 else if(!SvOK(sv)) {
3334 else if(SvPOK(sv)) {
3335 svb = SvPVXtrue(sv);
3337 else if((SvFLAGS(sv) & (SVf_IOK|SVf_NOK))) {
3338 svb = (SvIOK(sv) && SvIVX(sv) != 0)
3339 || (SvNOK(sv) && SvNVX(sv) != 0.0);
3343 goto restart; /* call sv_2bool_nomg */
3348 return SvRV(sv) != 0;
3352 RX_WRAPLEN(sv) > 1 || (RX_WRAPLEN(sv) && *RX_WRAPPED(sv) != '0');
3353 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
3357 =for apidoc sv_utf8_upgrade
3359 Converts the PV of an SV to its UTF-8-encoded form.
3360 Forces the SV to string form if it is not already.
3361 Will C<mg_get> on C<sv> if appropriate.
3362 Always sets the SvUTF8 flag to avoid future validity checks even
3363 if the whole string is the same in UTF-8 as not.
3364 Returns the number of bytes in the converted string
3366 This is not a general purpose byte encoding to Unicode interface:
3367 use the Encode extension for that.
3369 =for apidoc sv_utf8_upgrade_nomg
3371 Like sv_utf8_upgrade, but doesn't do magic on C<sv>.
3373 =for apidoc sv_utf8_upgrade_flags
3375 Converts the PV of an SV to its UTF-8-encoded form.
3376 Forces the SV to string form if it is not already.
3377 Always sets the SvUTF8 flag to avoid future validity checks even
3378 if all the bytes are invariant in UTF-8.
3379 If C<flags> has C<SV_GMAGIC> bit set,
3380 will C<mg_get> on C<sv> if appropriate, else not.
3382 If C<flags> has SV_FORCE_UTF8_UPGRADE set, this function assumes that the PV
3383 will expand when converted to UTF-8, and skips the extra work of checking for
3384 that. Typically this flag is used by a routine that has already parsed the
3385 string and found such characters, and passes this information on so that the
3386 work doesn't have to be repeated.
3388 Returns the number of bytes in the converted string.
3390 This is not a general purpose byte encoding to Unicode interface:
3391 use the Encode extension for that.
3393 =for apidoc sv_utf8_upgrade_flags_grow
3395 Like sv_utf8_upgrade_flags, but has an additional parameter C<extra>, which is
3396 the number of unused bytes the string of 'sv' is guaranteed to have free after
3397 it upon return. This allows the caller to reserve extra space that it intends
3398 to fill, to avoid extra grows.
3400 C<sv_utf8_upgrade>, C<sv_utf8_upgrade_nomg>, and C<sv_utf8_upgrade_flags>
3401 are implemented in terms of this function.
3403 Returns the number of bytes in the converted string (not including the spares).
3407 (One might think that the calling routine could pass in the position of the
3408 first variant character when it has set SV_FORCE_UTF8_UPGRADE, so it wouldn't
3409 have to be found again. But that is not the case, because typically when the
3410 caller is likely to use this flag, it won't be calling this routine unless it
3411 finds something that won't fit into a byte. Otherwise it tries to not upgrade
3412 and just use bytes. But some things that do fit into a byte are variants in
3413 utf8, and the caller may not have been keeping track of these.)
3415 If the routine itself changes the string, it adds a trailing C<NUL>. Such a
3416 C<NUL> isn't guaranteed due to having other routines do the work in some input
3417 cases, or if the input is already flagged as being in utf8.
3419 The speed of this could perhaps be improved for many cases if someone wanted to
3420 write a fast function that counts the number of variant characters in a string,
3421 especially if it could return the position of the first one.
3426 Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV *const sv, const I32 flags, STRLEN extra)
3428 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
3430 if (sv == &PL_sv_undef)
3432 if (!SvPOK_nog(sv)) {
3434 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3435 (void) sv_2pv_flags(sv,&len, flags);
3437 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3441 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
3446 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3451 S_sv_uncow(aTHX_ sv, 0);
3454 if (PL_encoding && !(flags & SV_UTF8_NO_ENCODING)) {
3455 sv_recode_to_utf8(sv, PL_encoding);
3456 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3460 if (SvCUR(sv) == 0) {
3461 if (extra) SvGROW(sv, extra);
3462 } else { /* Assume Latin-1/EBCDIC */
3463 /* This function could be much more efficient if we
3464 * had a FLAG in SVs to signal if there are any variant
3465 * chars in the PV. Given that there isn't such a flag
3466 * make the loop as fast as possible (although there are certainly ways
3467 * to speed this up, eg. through vectorization) */
3468 U8 * s = (U8 *) SvPVX_const(sv);
3469 U8 * e = (U8 *) SvEND(sv);
3471 STRLEN two_byte_count = 0;
3473 if (flags & SV_FORCE_UTF8_UPGRADE) goto must_be_utf8;
3475 /* See if really will need to convert to utf8. We mustn't rely on our
3476 * incoming SV being well formed and having a trailing '\0', as certain
3477 * code in pp_formline can send us partially built SVs. */
3481 if (NATIVE_BYTE_IS_INVARIANT(ch)) continue;
3483 t--; /* t already incremented; re-point to first variant */
3488 /* utf8 conversion not needed because all are invariants. Mark as
3489 * UTF-8 even if no variant - saves scanning loop */
3491 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3496 /* Here, the string should be converted to utf8, either because of an
3497 * input flag (two_byte_count = 0), or because a character that
3498 * requires 2 bytes was found (two_byte_count = 1). t points either to
3499 * the beginning of the string (if we didn't examine anything), or to
3500 * the first variant. In either case, everything from s to t - 1 will
3501 * occupy only 1 byte each on output.
3503 * There are two main ways to convert. One is to create a new string
3504 * and go through the input starting from the beginning, appending each
3505 * converted value onto the new string as we go along. It's probably
3506 * best to allocate enough space in the string for the worst possible
3507 * case rather than possibly running out of space and having to
3508 * reallocate and then copy what we've done so far. Since everything
3509 * from s to t - 1 is invariant, the destination can be initialized
3510 * with these using a fast memory copy
3512 * The other way is to figure out exactly how big the string should be
3513 * by parsing the entire input. Then you don't have to make it big
3514 * enough to handle the worst possible case, and more importantly, if
3515 * the string you already have is large enough, you don't have to
3516 * allocate a new string, you can copy the last character in the input
3517 * string to the final position(s) that will be occupied by the
3518 * converted string and go backwards, stopping at t, since everything
3519 * before that is invariant.
3521 * There are advantages and disadvantages to each method.
3523 * In the first method, we can allocate a new string, do the memory
3524 * copy from the s to t - 1, and then proceed through the rest of the
3525 * string byte-by-byte.
3527 * In the second method, we proceed through the rest of the input
3528 * string just calculating how big the converted string will be. Then
3529 * there are two cases:
3530 * 1) if the string has enough extra space to handle the converted
3531 * value. We go backwards through the string, converting until we
3532 * get to the position we are at now, and then stop. If this
3533 * position is far enough along in the string, this method is
3534 * faster than the other method. If the memory copy were the same
3535 * speed as the byte-by-byte loop, that position would be about
3536 * half-way, as at the half-way mark, parsing to the end and back
3537 * is one complete string's parse, the same amount as starting
3538 * over and going all the way through. Actually, it would be
3539 * somewhat less than half-way, as it's faster to just count bytes
3540 * than to also copy, and we don't have the overhead of allocating
3541 * a new string, changing the scalar to use it, and freeing the
3542 * existing one. But if the memory copy is fast, the break-even
3543 * point is somewhere after half way. The counting loop could be
3544 * sped up by vectorization, etc, to move the break-even point
3545 * further towards the beginning.
3546 * 2) if the string doesn't have enough space to handle the converted
3547 * value. A new string will have to be allocated, and one might
3548 * as well, given that, start from the beginning doing the first
3549 * method. We've spent extra time parsing the string and in
3550 * exchange all we've gotten is that we know precisely how big to
3551 * make the new one. Perl is more optimized for time than space,
3552 * so this case is a loser.
3553 * So what I've decided to do is not use the 2nd method unless it is
3554 * guaranteed that a new string won't have to be allocated, assuming
3555 * the worst case. I also decided not to put any more conditions on it
3556 * than this, for now. It seems likely that, since the worst case is
3557 * twice as big as the unknown portion of the string (plus 1), we won't
3558 * be guaranteed enough space, causing us to go to the first method,
3559 * unless the string is short, or the first variant character is near
3560 * the end of it. In either of these cases, it seems best to use the
3561 * 2nd method. The only circumstance I can think of where this would
3562 * be really slower is if the string had once had much more data in it
3563 * than it does now, but there is still a substantial amount in it */
3566 STRLEN invariant_head = t - s;
3567 STRLEN size = invariant_head + (e - t) * 2 + 1 + extra;
3568 if (SvLEN(sv) < size) {
3570 /* Here, have decided to allocate a new string */
3575 Newx(dst, size, U8);
3577 /* If no known invariants at the beginning of the input string,
3578 * set so starts from there. Otherwise, can use memory copy to
3579 * get up to where we are now, and then start from here */
3581 if (invariant_head == 0) {
3584 Copy(s, dst, invariant_head, char);
3585 d = dst + invariant_head;
3589 append_utf8_from_native_byte(*t, &d);
3593 SvPV_free(sv); /* No longer using pre-existing string */
3594 SvPV_set(sv, (char*)dst);
3595 SvCUR_set(sv, d - dst);
3596 SvLEN_set(sv, size);
3599 /* Here, have decided to get the exact size of the string.
3600 * Currently this happens only when we know that there is
3601 * guaranteed enough space to fit the converted string, so
3602 * don't have to worry about growing. If two_byte_count is 0,
3603 * then t points to the first byte of the string which hasn't
3604 * been examined yet. Otherwise two_byte_count is 1, and t
3605 * points to the first byte in the string that will expand to
3606 * two. Depending on this, start examining at t or 1 after t.
3609 U8 *d = t + two_byte_count;
3612 /* Count up the remaining bytes that expand to two */
3615 const U8 chr = *d++;
3616 if (! NATIVE_BYTE_IS_INVARIANT(chr)) two_byte_count++;
3619 /* The string will expand by just the number of bytes that
3620 * occupy two positions. But we are one afterwards because of
3621 * the increment just above. This is the place to put the
3622 * trailing NUL, and to set the length before we decrement */
3624 d += two_byte_count;
3625 SvCUR_set(sv, d - s);
3629 /* Having decremented d, it points to the position to put the
3630 * very last byte of the expanded string. Go backwards through
3631 * the string, copying and expanding as we go, stopping when we
3632 * get to the part that is invariant the rest of the way down */
3636 if (NATIVE_BYTE_IS_INVARIANT(*e)) {
3639 *d-- = UTF8_EIGHT_BIT_LO(*e);
3640 *d-- = UTF8_EIGHT_BIT_HI(*e);
3646 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3647 /* Update pos. We do it at the end rather than during
3648 * the upgrade, to avoid slowing down the common case
3649 * (upgrade without pos).
3650 * pos can be stored as either bytes or characters. Since
3651 * this was previously a byte string we can just turn off
3652 * the bytes flag. */
3653 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3655 mg->mg_flags &= ~MGf_BYTES;
3657 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3658 magic_setutf8(sv,mg); /* clear UTF8 cache */
3663 /* Mark as UTF-8 even if no variant - saves scanning loop */
3669 =for apidoc sv_utf8_downgrade
3671 Attempts to convert the PV of an SV from characters to bytes.
3672 If the PV contains a character that cannot fit
3673 in a byte, this conversion will fail;
3674 in this case, either returns false or, if C<fail_ok> is not
3677 This is not a general purpose Unicode to byte encoding interface:
3678 use the Encode extension for that.
3684 Perl_sv_utf8_downgrade(pTHX_ SV *const sv, const bool fail_ok)
3686 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3688 if (SvPOKp(sv) && SvUTF8(sv)) {
3692 int mg_flags = SV_GMAGIC;
3695 S_sv_uncow(aTHX_ sv, 0);
3697 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3699 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3700 if (mg && mg->mg_len > 0 && mg->mg_flags & MGf_BYTES) {
3701 mg->mg_len = sv_pos_b2u_flags(sv, mg->mg_len,
3702 SV_GMAGIC|SV_CONST_RETURN);
3703 mg_flags = 0; /* sv_pos_b2u does get magic */
3705 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3706 magic_setutf8(sv,mg); /* clear UTF8 cache */
3709 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3711 if (!utf8_to_bytes(s, &len)) {
3716 Perl_croak(aTHX_ "Wide character in %s",
3719 Perl_croak(aTHX_ "Wide character");
3730 =for apidoc sv_utf8_encode
3732 Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3733 flag off so that it looks like octets again.
3739 Perl_sv_utf8_encode(pTHX_ SV *const sv)
3741 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3743 if (SvREADONLY(sv)) {
3744 sv_force_normal_flags(sv, 0);
3746 (void) sv_utf8_upgrade(sv);
3751 =for apidoc sv_utf8_decode
3753 If the PV of the SV is an octet sequence in UTF-8
3754 and contains a multiple-byte character, the C<SvUTF8> flag is turned on
3755 so that it looks like a character. If the PV contains only single-byte
3756 characters, the C<SvUTF8> flag stays off.
3757 Scans PV for validity and returns false if the PV is invalid UTF-8.
3763 Perl_sv_utf8_decode(pTHX_ SV *const sv)
3765 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3768 const U8 *start, *c;
3771 /* The octets may have got themselves encoded - get them back as
3774 if (!sv_utf8_downgrade(sv, TRUE))
3777 /* it is actually just a matter of turning the utf8 flag on, but
3778 * we want to make sure everything inside is valid utf8 first.
3780 c = start = (const U8 *) SvPVX_const(sv);
3781 if (!is_utf8_string(c, SvCUR(sv)))
3783 e = (const U8 *) SvEND(sv);
3786 if (!UTF8_IS_INVARIANT(ch)) {
3791 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3792 /* XXX Is this dead code? XS_utf8_decode calls SvSETMAGIC
3793 after this, clearing pos. Does anything on CPAN
3795 /* adjust pos to the start of a UTF8 char sequence */
3796 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3798 I32 pos = mg->mg_len;
3800 for (c = start + pos; c > start; c--) {
3801 if (UTF8_IS_START(*c))
3804 mg->mg_len = c - start;
3807 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3808 magic_setutf8(sv,mg); /* clear UTF8 cache */
3815 =for apidoc sv_setsv
3817 Copies the contents of the source SV C<ssv> into the destination SV
3818 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3819 function if the source SV needs to be reused. Does not handle 'set' magic on
3820 destination SV. Calls 'get' magic on source SV. Loosely speaking, it
3821 performs a copy-by-value, obliterating any previous content of the
3824 You probably want to use one of the assortment of wrappers, such as
3825 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3826 C<SvSetMagicSV_nosteal>.
3828 =for apidoc sv_setsv_flags
3830 Copies the contents of the source SV C<ssv> into the destination SV
3831 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3832 function if the source SV needs to be reused. Does not handle 'set' magic.
3833 Loosely speaking, it performs a copy-by-value, obliterating any previous
3834 content of the destination.
3835 If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
3836 C<ssv> if appropriate, else not. If the C<flags>
3837 parameter has the C<SV_NOSTEAL> bit set then the
3838 buffers of temps will not be stolen. <sv_setsv>
3839 and C<sv_setsv_nomg> are implemented in terms of this function.
3841 You probably want to use one of the assortment of wrappers, such as
3842 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3843 C<SvSetMagicSV_nosteal>.
3845 This is the primary function for copying scalars, and most other
3846 copy-ish functions and macros use this underneath.
3852 S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
3854 I32 mro_changes = 0; /* 1 = method, 2 = isa, 3 = recursive isa */
3855 HV *old_stash = NULL;
3857 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB;
3859 if (dtype != SVt_PVGV && !isGV_with_GP(dstr)) {
3860 const char * const name = GvNAME(sstr);
3861 const STRLEN len = GvNAMELEN(sstr);
3863 if (dtype >= SVt_PV) {
3869 SvUPGRADE(dstr, SVt_PVGV);
3870 (void)SvOK_off(dstr);
3871 isGV_with_GP_on(dstr);
3873 GvSTASH(dstr) = GvSTASH(sstr);
3875 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
3876 gv_name_set(MUTABLE_GV(dstr), name, len,
3877 GV_ADD | (GvNAMEUTF8(sstr) ? SVf_UTF8 : 0 ));
3878 SvFAKE_on(dstr); /* can coerce to non-glob */
3881 if(GvGP(MUTABLE_GV(sstr))) {
3882 /* If source has method cache entry, clear it */
3884 SvREFCNT_dec(GvCV(sstr));
3885 GvCV_set(sstr, NULL);
3888 /* If source has a real method, then a method is
3891 GvCV((const GV *)sstr) && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3897 /* If dest already had a real method, that's a change as well */
3899 !mro_changes && GvGP(MUTABLE_GV(dstr)) && GvCVu((const GV *)dstr)
3900 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3905 /* We don't need to check the name of the destination if it was not a
3906 glob to begin with. */
3907 if(dtype == SVt_PVGV) {
3908 const char * const name = GvNAME((const GV *)dstr);
3911 /* The stash may have been detached from the symbol table, so
3913 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3917 const STRLEN len = GvNAMELEN(dstr);
3918 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
3919 || (len == 1 && name[0] == ':')) {
3922 /* Set aside the old stash, so we can reset isa caches on
3924 if((old_stash = GvHV(dstr)))
3925 /* Make sure we do not lose it early. */
3926 SvREFCNT_inc_simple_void_NN(
3927 sv_2mortal((SV *)old_stash)
3932 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
3935 gp_free(MUTABLE_GV(dstr));
3936 GvINTRO_off(dstr); /* one-shot flag */
3937 GvGP_set(dstr, gp_ref(GvGP(sstr)));
3938 if (SvTAINTED(sstr))
3940 if (GvIMPORTED(dstr) != GVf_IMPORTED
3941 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
3943 GvIMPORTED_on(dstr);
3946 if(mro_changes == 2) {
3947 if (GvAV((const GV *)sstr)) {
3949 SV * const sref = (SV *)GvAV((const GV *)dstr);
3950 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
3951 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
3952 AV * const ary = newAV();
3953 av_push(ary, mg->mg_obj); /* takes the refcount */
3954 mg->mg_obj = (SV *)ary;
3956 av_push((AV *)mg->mg_obj, SvREFCNT_inc_simple_NN(dstr));
3958 else sv_magic(sref, dstr, PERL_MAGIC_isa, NULL, 0);
3960 mro_isa_changed_in(GvSTASH(dstr));
3962 else if(mro_changes == 3) {
3963 HV * const stash = GvHV(dstr);
3964 if(old_stash ? (HV *)HvENAME_get(old_stash) : stash)
3970 else if(mro_changes) mro_method_changed_in(GvSTASH(dstr));
3971 if (GvIO(dstr) && dtype == SVt_PVGV) {
3972 DEBUG_o(Perl_deb(aTHX_
3973 "glob_assign_glob clearing PL_stashcache\n"));
3974 /* It's a cache. It will rebuild itself quite happily.
3975 It's a lot of effort to work out exactly which key (or keys)
3976 might be invalidated by the creation of the this file handle.
3978 hv_clear(PL_stashcache);
3984 Perl_gv_setref(pTHX_ SV *const dstr, SV *const sstr)
3986 SV * const sref = SvRV(sstr);
3988 const int intro = GvINTRO(dstr);
3991 const U32 stype = SvTYPE(sref);
3993 PERL_ARGS_ASSERT_GV_SETREF;
3996 GvINTRO_off(dstr); /* one-shot flag */
3997 GvLINE(dstr) = CopLINE(PL_curcop);
3998 GvEGV(dstr) = MUTABLE_GV(dstr);
4003 location = (SV **) &(GvGP(dstr)->gp_cv); /* XXX bypassing GvCV_set */
4004 import_flag = GVf_IMPORTED_CV;
4007 location = (SV **) &GvHV(dstr);
4008 import_flag = GVf_IMPORTED_HV;
4011 location = (SV **) &GvAV(dstr);
4012 import_flag = GVf_IMPORTED_AV;
4015 location = (SV **) &GvIOp(dstr);
4018 location = (SV **) &GvFORM(dstr);
4021 location = &GvSV(dstr);
4022 import_flag = GVf_IMPORTED_SV;
4025 if (stype == SVt_PVCV) {
4026 /*if (GvCVGEN(dstr) && (GvCV(dstr) != (const CV *)sref || GvCVGEN(dstr))) {*/
4027 if (GvCVGEN(dstr)) {
4028 SvREFCNT_dec(GvCV(dstr));
4029 GvCV_set(dstr, NULL);
4030 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4033 /* SAVEt_GVSLOT takes more room on the savestack and has more
4034 overhead in leave_scope than SAVEt_GENERIC_SV. But for CVs
4035 leave_scope needs access to the GV so it can reset method
4036 caches. We must use SAVEt_GVSLOT whenever the type is
4037 SVt_PVCV, even if the stash is anonymous, as the stash may
4038 gain a name somehow before leave_scope. */
4039 if (stype == SVt_PVCV) {
4040 /* There is no save_pushptrptrptr. Creating it for this
4041 one call site would be overkill. So inline the ss add
4045 SS_ADD_PTR(location);
4046 SS_ADD_PTR(SvREFCNT_inc(*location));
4047 SS_ADD_UV(SAVEt_GVSLOT);
4050 else SAVEGENERICSV(*location);
4053 if (stype == SVt_PVCV && (*location != sref || GvCVGEN(dstr))) {
4054 CV* const cv = MUTABLE_CV(*location);
4056 if (!GvCVGEN((const GV *)dstr) &&
4057 (CvROOT(cv) || CvXSUB(cv)) &&
4058 /* redundant check that avoids creating the extra SV
4059 most of the time: */
4060 (CvCONST(cv) || ckWARN(WARN_REDEFINE)))
4062 SV * const new_const_sv =
4063 CvCONST((const CV *)sref)
4064 ? cv_const_sv((const CV *)sref)
4066 report_redefined_cv(
4067 sv_2mortal(Perl_newSVpvf(aTHX_
4070 HvNAME_HEK(GvSTASH((const GV *)dstr))
4072 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr)))
4075 CvCONST((const CV *)sref) ? &new_const_sv : NULL
4079 cv_ckproto_len_flags(cv, (const GV *)dstr,
4080 SvPOK(sref) ? CvPROTO(sref) : NULL,
4081 SvPOK(sref) ? CvPROTOLEN(sref) : 0,
4082 SvPOK(sref) ? SvUTF8(sref) : 0);
4084 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4085 GvASSUMECV_on(dstr);
4086 if(GvSTASH(dstr)) { /* sub foo { 1 } sub bar { 2 } *bar = \&foo */
4087 if (intro && GvREFCNT(dstr) > 1) {
4088 /* temporary remove extra savestack's ref */
4090 gv_method_changed(dstr);
4093 else gv_method_changed(dstr);
4096 *location = SvREFCNT_inc_simple_NN(sref);
4097 if (import_flag && !(GvFLAGS(dstr) & import_flag)
4098 && CopSTASH_ne(PL_curcop, GvSTASH(dstr))) {
4099 GvFLAGS(dstr) |= import_flag;
4101 if (import_flag == GVf_IMPORTED_SV) {
4103 save_aliased_sv((GV *)dstr);
4105 /* Turn off the flag if sref is not referenced elsewhere,
4106 even by weak refs. (SvRMAGICAL is a pessimistic check for
4108 if (SvREFCNT(sref) <= 2 && !SvRMAGICAL(sref))
4109 GvALIASED_SV_off(dstr);
4111 GvALIASED_SV_on(dstr);
4113 if (stype == SVt_PVHV) {
4114 const char * const name = GvNAME((GV*)dstr);
4115 const STRLEN len = GvNAMELEN(dstr);
4118 (len > 1 && name[len-2] == ':' && name[len-1] == ':')
4119 || (len == 1 && name[0] == ':')
4121 && (!dref || HvENAME_get(dref))
4124 (HV *)sref, (HV *)dref,
4130 stype == SVt_PVAV && sref != dref
4131 && strEQ(GvNAME((GV*)dstr), "ISA")
4132 /* The stash may have been detached from the symbol table, so
4133 check its name before doing anything. */
4134 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
4137 MAGIC * const omg = dref && SvSMAGICAL(dref)
4138 ? mg_find(dref, PERL_MAGIC_isa)
4140 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
4141 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
4142 AV * const ary = newAV();
4143 av_push(ary, mg->mg_obj); /* takes the refcount */
4144 mg->mg_obj = (SV *)ary;
4147 if (SvTYPE(omg->mg_obj) == SVt_PVAV) {
4148 SV **svp = AvARRAY((AV *)omg->mg_obj);
4149 I32 items = AvFILLp((AV *)omg->mg_obj) + 1;
4153 SvREFCNT_inc_simple_NN(*svp++)
4159 SvREFCNT_inc_simple_NN(omg->mg_obj)
4163 av_push((AV *)mg->mg_obj,SvREFCNT_inc_simple_NN(dstr));
4168 sref, omg ? omg->mg_obj : dstr, PERL_MAGIC_isa, NULL, 0
4170 mg = mg_find(sref, PERL_MAGIC_isa);
4172 /* Since the *ISA assignment could have affected more than
4173 one stash, don't call mro_isa_changed_in directly, but let
4174 magic_clearisa do it for us, as it already has the logic for
4175 dealing with globs vs arrays of globs. */
4177 Perl_magic_clearisa(aTHX_ NULL, mg);
4179 else if (stype == SVt_PVIO) {
4180 DEBUG_o(Perl_deb(aTHX_ "gv_setref clearing PL_stashcache\n"));
4181 /* It's a cache. It will rebuild itself quite happily.
4182 It's a lot of effort to work out exactly which key (or keys)
4183 might be invalidated by the creation of the this file handle.
4185 hv_clear(PL_stashcache);
4189 if (!intro) SvREFCNT_dec(dref);
4190 if (SvTAINTED(sstr))
4198 #ifdef PERL_DEBUG_READONLY_COW
4199 # include <sys/mman.h>
4201 # ifndef PERL_MEMORY_DEBUG_HEADER_SIZE
4202 # define PERL_MEMORY_DEBUG_HEADER_SIZE 0
4206 Perl_sv_buf_to_ro(pTHX_ SV *sv)
4208 struct perl_memory_debug_header * const header =
4209 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4210 const MEM_SIZE len = header->size;
4211 PERL_ARGS_ASSERT_SV_BUF_TO_RO;
4212 # ifdef PERL_TRACK_MEMPOOL
4213 if (!header->readonly) header->readonly = 1;
4215 if (mprotect(header, len, PROT_READ))
4216 Perl_warn(aTHX_ "mprotect RW for COW string %p %lu failed with %d",
4217 header, len, errno);
4221 S_sv_buf_to_rw(pTHX_ SV *sv)
4223 struct perl_memory_debug_header * const header =
4224 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4225 const MEM_SIZE len = header->size;
4226 PERL_ARGS_ASSERT_SV_BUF_TO_RW;
4227 if (mprotect(header, len, PROT_READ|PROT_WRITE))
4228 Perl_warn(aTHX_ "mprotect for COW string %p %lu failed with %d",
4229 header, len, errno);
4230 # ifdef PERL_TRACK_MEMPOOL
4231 header->readonly = 0;
4236 # define sv_buf_to_ro(sv) NOOP
4237 # define sv_buf_to_rw(sv) NOOP
4241 Perl_sv_setsv_flags(pTHX_ SV *dstr, SV* sstr, const I32 flags)
4247 PERL_ARGS_ASSERT_SV_SETSV_FLAGS;
4252 if (SvIS_FREED(dstr)) {
4253 Perl_croak(aTHX_ "panic: attempt to copy value %" SVf
4254 " to a freed scalar %p", SVfARG(sstr), (void *)dstr);
4256 SV_CHECK_THINKFIRST_COW_DROP(dstr);
4258 sstr = &PL_sv_undef;
4259 if (SvIS_FREED(sstr)) {
4260 Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p to %p",
4261 (void*)sstr, (void*)dstr);
4263 stype = SvTYPE(sstr);
4264 dtype = SvTYPE(dstr);
4266 /* There's a lot of redundancy below but we're going for speed here */
4271 if (dtype != SVt_PVGV && dtype != SVt_PVLV) {
4272 (void)SvOK_off(dstr);
4280 sv_upgrade(dstr, SVt_IV);
4284 sv_upgrade(dstr, SVt_PVIV);
4288 goto end_of_first_switch;
4290 (void)SvIOK_only(dstr);
4291 SvIV_set(dstr, SvIVX(sstr));
4294 /* SvTAINTED can only be true if the SV has taint magic, which in
4295 turn means that the SV type is PVMG (or greater). This is the
4296 case statement for SVt_IV, so this cannot be true (whatever gcov
4298 assert(!SvTAINTED(sstr));
4303 if (dtype < SVt_PV && dtype != SVt_IV)
4304 sv_upgrade(dstr, SVt_IV);
4312 sv_upgrade(dstr, SVt_NV);
4316 sv_upgrade(dstr, SVt_PVNV);
4320 goto end_of_first_switch;
4322 SvNV_set(dstr, SvNVX(sstr));
4323 (void)SvNOK_only(dstr);
4324 /* SvTAINTED can only be true if the SV has taint magic, which in
4325 turn means that the SV type is PVMG (or greater). This is the
4326 case statement for SVt_NV, so this cannot be true (whatever gcov
4328 assert(!SvTAINTED(sstr));
4335 sv_upgrade(dstr, SVt_PV);
4338 if (dtype < SVt_PVIV)
4339 sv_upgrade(dstr, SVt_PVIV);
4342 if (dtype < SVt_PVNV)
4343 sv_upgrade(dstr, SVt_PVNV);
4347 const char * const type = sv_reftype(sstr,0);
4349 /* diag_listed_as: Bizarre copy of %s */
4350 Perl_croak(aTHX_ "Bizarre copy of %s in %s", type, OP_DESC(PL_op));
4352 Perl_croak(aTHX_ "Bizarre copy of %s", type);
4354 NOT_REACHED; /* NOTREACHED */
4358 if (dtype < SVt_REGEXP)
4360 if (dtype >= SVt_PV) {
4366 sv_upgrade(dstr, SVt_REGEXP);
4374 if (SvGMAGICAL(sstr) && (flags & SV_GMAGIC)) {
4376 if (SvTYPE(sstr) != stype)
4377 stype = SvTYPE(sstr);
4379 if (isGV_with_GP(sstr) && dtype <= SVt_PVLV) {
4380 glob_assign_glob(dstr, sstr, dtype);
4383 if (stype == SVt_PVLV)
4385 if (isREGEXP(sstr)) goto upgregexp;
4386 SvUPGRADE(dstr, SVt_PVNV);
4389 SvUPGRADE(dstr, (svtype)stype);
4391 end_of_first_switch:
4393 /* dstr may have been upgraded. */
4394 dtype = SvTYPE(dstr);
4395 sflags = SvFLAGS(sstr);
4397 if (dtype == SVt_PVCV) {
4398 /* Assigning to a subroutine sets the prototype. */
4401 const char *const ptr = SvPV_const(sstr, len);
4403 SvGROW(dstr, len + 1);
4404 Copy(ptr, SvPVX(dstr), len + 1, char);
4405 SvCUR_set(dstr, len);
4407 SvFLAGS(dstr) |= sflags & SVf_UTF8;
4408 CvAUTOLOAD_off(dstr);
4413 else if (dtype == SVt_PVAV || dtype == SVt_PVHV || dtype == SVt_PVFM) {
4414 const char * const type = sv_reftype(dstr,0);
4416 /* diag_listed_as: Cannot copy to %s */
4417 Perl_croak(aTHX_ "Cannot copy to %s in %s", type, OP_DESC(PL_op));
4419 Perl_croak(aTHX_ "Cannot copy to %s", type);
4420 } else if (sflags & SVf_ROK) {
4421 if (isGV_with_GP(dstr)
4422 && SvTYPE(SvRV(sstr)) == SVt_PVGV && isGV_with_GP(SvRV(sstr))) {
4425 if (GvIMPORTED(dstr) != GVf_IMPORTED
4426 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
4428 GvIMPORTED_on(dstr);
4433 glob_assign_glob(dstr, sstr, dtype);
4437 if (dtype >= SVt_PV) {
4438 if (isGV_with_GP(dstr)) {
4439 gv_setref(dstr, sstr);
4442 if (SvPVX_const(dstr)) {
4448 (void)SvOK_off(dstr);
4449 SvRV_set(dstr, SvREFCNT_inc(SvRV(sstr)));
4450 SvFLAGS(dstr) |= sflags & SVf_ROK;
4451 assert(!(sflags & SVp_NOK));
4452 assert(!(sflags & SVp_IOK));
4453 assert(!(sflags & SVf_NOK));
4454 assert(!(sflags & SVf_IOK));
4456 else if (isGV_with_GP(dstr)) {
4457 if (!(sflags & SVf_OK)) {
4458 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4459 "Undefined value assigned to typeglob");
4462 GV *gv = gv_fetchsv_nomg(sstr, GV_ADD, SVt_PVGV);
4463 if (dstr != (const SV *)gv) {
4464 const char * const name = GvNAME((const GV *)dstr);
4465 const STRLEN len = GvNAMELEN(dstr);
4466 HV *old_stash = NULL;
4467 bool reset_isa = FALSE;
4468 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
4469 || (len == 1 && name[0] == ':')) {
4470 /* Set aside the old stash, so we can reset isa caches
4471 on its subclasses. */
4472 if((old_stash = GvHV(dstr))) {
4473 /* Make sure we do not lose it early. */
4474 SvREFCNT_inc_simple_void_NN(
4475 sv_2mortal((SV *)old_stash)
4482 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
4483 gp_free(MUTABLE_GV(dstr));
4485 GvGP_set(dstr, gp_ref(GvGP(gv)));
4488 HV * const stash = GvHV(dstr);
4490 old_stash ? (HV *)HvENAME_get(old_stash) : stash
4500 else if ((dtype == SVt_REGEXP || dtype == SVt_PVLV)
4501 && (stype == SVt_REGEXP || isREGEXP(sstr))) {
4502 reg_temp_copy((REGEXP*)dstr, (REGEXP*)sstr);
4504 else if (sflags & SVp_POK) {
4505 const STRLEN cur = SvCUR(sstr);
4506 const STRLEN len = SvLEN(sstr);
4509 * We have three basic ways to copy the string:
4515 * Which we choose is based on various factors. The following
4516 * things are listed in order of speed, fastest to slowest:
4518 * - Copying a short string
4519 * - Copy-on-write bookkeeping
4521 * - Copying a long string
4523 * We swipe the string (steal the string buffer) if the SV on the
4524 * rhs is about to be freed anyway (TEMP and refcnt==1). This is a
4525 * big win on long strings. It should be a win on short strings if
4526 * SvPVX_const(dstr) has to be allocated. If not, it should not
4527 * slow things down, as SvPVX_const(sstr) would have been freed
4530 * We also steal the buffer from a PADTMP (operator target) if it
4531 * is ‘long enough’. For short strings, a swipe does not help
4532 * here, as it causes more malloc calls the next time the target
4533 * is used. Benchmarks show that even if SvPVX_const(dstr) has to
4534 * be allocated it is still not worth swiping PADTMPs for short
4535 * strings, as the savings here are small.
4537 * If swiping is not an option, then we see whether it is
4538 * worth using copy-on-write. If the lhs already has a buf-
4539 * fer big enough and the string is short, we skip it and fall back
4540 * to method 3, since memcpy is faster for short strings than the
4541 * later bookkeeping overhead that copy-on-write entails.
4543 * If the rhs is not a copy-on-write string yet, then we also
4544 * consider whether the buffer is too large relative to the string
4545 * it holds. Some operations such as readline allocate a large
4546 * buffer in the expectation of reusing it. But turning such into
4547 * a COW buffer is counter-productive because it increases memory
4548 * usage by making readline allocate a new large buffer the sec-
4549 * ond time round. So, if the buffer is too large, again, we use
4552 * Finally, if there is no buffer on the left, or the buffer is too
4553 * small, then we use copy-on-write and make both SVs share the
4558 /* Whichever path we take through the next code, we want this true,
4559 and doing it now facilitates the COW check. */
4560 (void)SvPOK_only(dstr);
4564 /* slated for free anyway (and not COW)? */
4565 (sflags & (SVs_TEMP|SVf_IsCOW)) == SVs_TEMP
4566 /* or a swipable TARG */
4568 (SVs_PADTMP|SVf_READONLY|SVf_PROTECT|SVf_IsCOW))
4570 /* whose buffer is worth stealing */
4571 && CHECK_COWBUF_THRESHOLD(cur,len)
4574 !(sflags & SVf_OOK) && /* and not involved in OOK hack? */
4575 (!(flags & SV_NOSTEAL)) &&
4576 /* and we're allowed to steal temps */
4577 SvREFCNT(sstr) == 1 && /* and no other references to it? */
4578 len) /* and really is a string */
4579 { /* Passes the swipe test. */
4580 if (SvPVX_const(dstr)) /* we know that dtype >= SVt_PV */
4582 SvPV_set(dstr, SvPVX_mutable(sstr));
4583 SvLEN_set(dstr, SvLEN(sstr));
4584 SvCUR_set(dstr, SvCUR(sstr));
4587 (void)SvOK_off(sstr); /* NOTE: nukes most SvFLAGS on sstr */
4588 SvPV_set(sstr, NULL);
4593 else if (flags & SV_COW_SHARED_HASH_KEYS
4595 #ifdef PERL_OLD_COPY_ON_WRITE
4596 ( sflags & SVf_IsCOW
4597 || ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4598 && (SvFLAGS(dstr) & CAN_COW_MASK) == CAN_COW_FLAGS
4599 && SvTYPE(sstr) >= SVt_PVIV && len
4602 #elif defined(PERL_NEW_COPY_ON_WRITE)
4605 ( (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4606 /* If this is a regular (non-hek) COW, only so
4607 many COW "copies" are possible. */
4608 && CowREFCNT(sstr) != SV_COW_REFCNT_MAX ))
4609 : ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4610 && !(SvFLAGS(dstr) & SVf_BREAK)
4611 && CHECK_COW_THRESHOLD(cur,len) && cur+1 < len
4612 && (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4616 && !(SvFLAGS(dstr) & SVf_BREAK)
4619 /* Either it's a shared hash key, or it's suitable for
4622 PerlIO_printf(Perl_debug_log, "Copy on write: sstr --> dstr\n");
4627 if (!(sflags & SVf_IsCOW)) {
4629 # ifdef PERL_OLD_COPY_ON_WRITE
4630 /* Make the source SV into a loop of 1.
4631 (about to become 2) */
4632 SV_COW_NEXT_SV_SET(sstr, sstr);
4634 CowREFCNT(sstr) = 0;
4638 if (SvPVX_const(dstr)) { /* we know that dtype >= SVt_PV */
4644 # ifdef PERL_OLD_COPY_ON_WRITE
4645 assert (SvTYPE(dstr) >= SVt_PVIV);
4646 /* SvIsCOW_normal */
4647 /* splice us in between source and next-after-source. */
4648 SV_COW_NEXT_SV_SET(dstr, SV_COW_NEXT_SV(sstr));
4649 SV_COW_NEXT_SV_SET(sstr, dstr);
4651 if (sflags & SVf_IsCOW) {
4656 SvPV_set(dstr, SvPVX_mutable(sstr));
4661 /* SvIsCOW_shared_hash */
4662 DEBUG_C(PerlIO_printf(Perl_debug_log,
4663 "Copy on write: Sharing hash\n"));
4665 assert (SvTYPE(dstr) >= SVt_PV);
4667 HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)))));
4669 SvLEN_set(dstr, len);
4670 SvCUR_set(dstr, cur);
4673 /* Failed the swipe test, and we cannot do copy-on-write either.
4674 Have to copy the string. */
4675 SvGROW(dstr, cur + 1); /* inlined from sv_setpvn */
4676 Move(SvPVX_const(sstr),SvPVX(dstr),cur,char);
4677 SvCUR_set(dstr, cur);
4678 *SvEND(dstr) = '\0';
4680 if (sflags & SVp_NOK) {
4681 SvNV_set(dstr, SvNVX(sstr));
4683 if (sflags & SVp_IOK) {
4684 SvIV_set(dstr, SvIVX(sstr));
4685 /* Must do this otherwise some other overloaded use of 0x80000000
4686 gets confused. I guess SVpbm_VALID */
4687 if (sflags & SVf_IVisUV)
4690 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_NOK|SVp_NOK|SVf_UTF8);
4692 const MAGIC * const smg = SvVSTRING_mg(sstr);
4694 sv_magic(dstr, NULL, PERL_MAGIC_vstring,
4695 smg->mg_ptr, smg->mg_len);
4696 SvRMAGICAL_on(dstr);
4700 else if (sflags & (SVp_IOK|SVp_NOK)) {
4701 (void)SvOK_off(dstr);
4702 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_IVisUV|SVf_NOK|SVp_NOK);
4703 if (sflags & SVp_IOK) {
4704 /* XXXX Do we want to set IsUV for IV(ROK)? Be extra safe... */
4705 SvIV_set(dstr, SvIVX(sstr));
4707 if (sflags & SVp_NOK) {
4708 SvNV_set(dstr, SvNVX(sstr));
4712 if (isGV_with_GP(sstr)) {
4713 gv_efullname3(dstr, MUTABLE_GV(sstr), "*");
4716 (void)SvOK_off(dstr);
4718 if (SvTAINTED(sstr))
4723 =for apidoc sv_setsv_mg
4725 Like C<sv_setsv>, but also handles 'set' magic.
4731 Perl_sv_setsv_mg(pTHX_ SV *const dstr, SV *const sstr)
4733 PERL_ARGS_ASSERT_SV_SETSV_MG;
4735 sv_setsv(dstr,sstr);
4740 # ifdef PERL_OLD_COPY_ON_WRITE
4741 # define SVt_COW SVt_PVIV
4743 # define SVt_COW SVt_PV
4746 Perl_sv_setsv_cow(pTHX_ SV *dstr, SV *sstr)
4748 STRLEN cur = SvCUR(sstr);
4749 STRLEN len = SvLEN(sstr);
4751 #if defined(PERL_DEBUG_READONLY_COW) && defined(PERL_NEW_COPY_ON_WRITE)
4752 const bool already = cBOOL(SvIsCOW(sstr));
4755 PERL_ARGS_ASSERT_SV_SETSV_COW;
4758 PerlIO_printf(Perl_debug_log, "Fast copy on write: %p -> %p\n",
4759 (void*)sstr, (void*)dstr);
4766 if (SvTHINKFIRST(dstr))
4767 sv_force_normal_flags(dstr, SV_COW_DROP_PV);
4768 else if (SvPVX_const(dstr))
4769 Safefree(SvPVX_mutable(dstr));
4773 SvUPGRADE(dstr, SVt_COW);
4775 assert (SvPOK(sstr));
4776 assert (SvPOKp(sstr));
4777 # ifdef PERL_OLD_COPY_ON_WRITE
4778 assert (!SvIOK(sstr));
4779 assert (!SvIOKp(sstr));
4780 assert (!SvNOK(sstr));
4781 assert (!SvNOKp(sstr));
4784 if (SvIsCOW(sstr)) {
4786 if (SvLEN(sstr) == 0) {
4787 /* source is a COW shared hash key. */
4788 DEBUG_C(PerlIO_printf(Perl_debug_log,
4789 "Fast copy on write: Sharing hash\n"));
4790 new_pv = HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr))));
4793 # ifdef PERL_OLD_COPY_ON_WRITE
4794 SV_COW_NEXT_SV_SET(dstr, SV_COW_NEXT_SV(sstr));
4796 assert(SvCUR(sstr)+1 < SvLEN(sstr));
4797 assert(CowREFCNT(sstr) < SV_COW_REFCNT_MAX);
4800 assert ((SvFLAGS(sstr) & CAN_COW_MASK) == CAN_COW_FLAGS);
4801 SvUPGRADE(sstr, SVt_COW);
4803 DEBUG_C(PerlIO_printf(Perl_debug_log,
4804 "Fast copy on write: Converting sstr to COW\n"));
4805 # ifdef PERL_OLD_COPY_ON_WRITE
4806 SV_COW_NEXT_SV_SET(dstr, sstr);
4808 CowREFCNT(sstr) = 0;
4811 # ifdef PERL_OLD_COPY_ON_WRITE
4812 SV_COW_NEXT_SV_SET(sstr, dstr);
4814 # ifdef PERL_DEBUG_READONLY_COW
4815 if (already) sv_buf_to_rw(sstr);
4819 new_pv = SvPVX_mutable(sstr);
4823 SvPV_set(dstr, new_pv);
4824 SvFLAGS(dstr) = (SVt_COW|SVf_POK|SVp_POK|SVf_IsCOW);
4827 SvLEN_set(dstr, len);
4828 SvCUR_set(dstr, cur);
4837 =for apidoc sv_setpvn
4839 Copies a string (possibly containing embedded C<NUL> characters) into an SV.
4840 The C<len> parameter indicates the number of
4841 bytes to be copied. If the C<ptr> argument is NULL the SV will become
4842 undefined. Does not handle 'set' magic. See C<sv_setpvn_mg>.
4848 Perl_sv_setpvn(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4852 PERL_ARGS_ASSERT_SV_SETPVN;
4854 SV_CHECK_THINKFIRST_COW_DROP(sv);
4860 /* len is STRLEN which is unsigned, need to copy to signed */
4863 Perl_croak(aTHX_ "panic: sv_setpvn called with negative strlen %"
4866 SvUPGRADE(sv, SVt_PV);
4868 dptr = SvGROW(sv, len + 1);
4869 Move(ptr,dptr,len,char);
4872 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4874 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4878 =for apidoc sv_setpvn_mg
4880 Like C<sv_setpvn>, but also handles 'set' magic.
4886 Perl_sv_setpvn_mg(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4888 PERL_ARGS_ASSERT_SV_SETPVN_MG;
4890 sv_setpvn(sv,ptr,len);
4895 =for apidoc sv_setpv
4897 Copies a string into an SV. The string must be terminated with a C<NUL>
4899 Does not handle 'set' magic. See C<sv_setpv_mg>.
4905 Perl_sv_setpv(pTHX_ SV *const sv, const char *const ptr)
4909 PERL_ARGS_ASSERT_SV_SETPV;
4911 SV_CHECK_THINKFIRST_COW_DROP(sv);
4917 SvUPGRADE(sv, SVt_PV);
4919 SvGROW(sv, len + 1);
4920 Move(ptr,SvPVX(sv),len+1,char);
4922 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4924 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4928 =for apidoc sv_setpv_mg
4930 Like C<sv_setpv>, but also handles 'set' magic.
4936 Perl_sv_setpv_mg(pTHX_ SV *const sv, const char *const ptr)
4938 PERL_ARGS_ASSERT_SV_SETPV_MG;
4945 Perl_sv_sethek(pTHX_ SV *const sv, const HEK *const hek)
4947 PERL_ARGS_ASSERT_SV_SETHEK;
4953 if (HEK_LEN(hek) == HEf_SVKEY) {
4954 sv_setsv(sv, *(SV**)HEK_KEY(hek));
4957 const int flags = HEK_FLAGS(hek);
4958 if (flags & HVhek_WASUTF8) {
4959 STRLEN utf8_len = HEK_LEN(hek);
4960 char *as_utf8 = (char *)bytes_to_utf8((U8*)HEK_KEY(hek), &utf8_len);
4961 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
4964 } else if (flags & HVhek_UNSHARED) {
4965 sv_setpvn(sv, HEK_KEY(hek), HEK_LEN(hek));
4968 else SvUTF8_off(sv);
4972 SV_CHECK_THINKFIRST_COW_DROP(sv);
4973 SvUPGRADE(sv, SVt_PV);
4975 SvPV_set(sv,(char *)HEK_KEY(share_hek_hek(hek)));
4976 SvCUR_set(sv, HEK_LEN(hek));
4982 else SvUTF8_off(sv);
4990 =for apidoc sv_usepvn_flags
4992 Tells an SV to use C<ptr> to find its string value. Normally the
4993 string is stored inside the SV, but sv_usepvn allows the SV to use an
4994 outside string. The C<ptr> should point to memory that was allocated
4995 by L<Newx|perlclib/Memory Management and String Handling>. It must be
4996 the start of a Newx-ed block of memory, and not a pointer to the
4997 middle of it (beware of L<OOK|perlguts/Offsets> and copy-on-write),
4998 and not be from a non-Newx memory allocator like C<malloc>. The
4999 string length, C<len>, must be supplied. By default this function
5000 will C<Renew> (i.e. realloc, move) the memory pointed to by C<ptr>,
5001 so that pointer should not be freed or used by the programmer after
5002 giving it to sv_usepvn, and neither should any pointers from "behind"
5003 that pointer (e.g. ptr + 1) be used.
5005 If C<flags> & SV_SMAGIC is true, will call SvSETMAGIC. If C<flags> &
5006 SV_HAS_TRAILING_NUL is true, then C<ptr[len]> must be C<NUL>, and the realloc
5007 will be skipped (i.e. the buffer is actually at least 1 byte longer than
5008 C<len>, and already meets the requirements for storing in C<SvPVX>).
5014 Perl_sv_usepvn_flags(pTHX_ SV *const sv, char *ptr, const STRLEN len, const U32 flags)
5018 PERL_ARGS_ASSERT_SV_USEPVN_FLAGS;
5020 SV_CHECK_THINKFIRST_COW_DROP(sv);
5021 SvUPGRADE(sv, SVt_PV);
5024 if (flags & SV_SMAGIC)
5028 if (SvPVX_const(sv))
5032 if (flags & SV_HAS_TRAILING_NUL)
5033 assert(ptr[len] == '\0');
5036 allocate = (flags & SV_HAS_TRAILING_NUL)
5038 #ifdef Perl_safesysmalloc_size
5041 PERL_STRLEN_ROUNDUP(len + 1);
5043 if (flags & SV_HAS_TRAILING_NUL) {
5044 /* It's long enough - do nothing.
5045 Specifically Perl_newCONSTSUB is relying on this. */
5048 /* Force a move to shake out bugs in callers. */
5049 char *new_ptr = (char*)safemalloc(allocate);
5050 Copy(ptr, new_ptr, len, char);
5051 PoisonFree(ptr,len,char);
5055 ptr = (char*) saferealloc (ptr, allocate);
5058 #ifdef Perl_safesysmalloc_size
5059 SvLEN_set(sv, Perl_safesysmalloc_size(ptr));
5061 SvLEN_set(sv, allocate);
5065 if (!(flags & SV_HAS_TRAILING_NUL)) {
5068 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5070 if (flags & SV_SMAGIC)
5074 #ifdef PERL_OLD_COPY_ON_WRITE
5075 /* Need to do this *after* making the SV normal, as we need the buffer
5076 pointer to remain valid until after we've copied it. If we let go too early,
5077 another thread could invalidate it by unsharing last of the same hash key
5078 (which it can do by means other than releasing copy-on-write Svs)
5079 or by changing the other copy-on-write SVs in the loop. */
5081 S_sv_release_COW(pTHX_ SV *sv, const char *pvx, SV *after)
5083 PERL_ARGS_ASSERT_SV_RELEASE_COW;
5085 { /* this SV was SvIsCOW_normal(sv) */
5086 /* we need to find the SV pointing to us. */
5087 SV *current = SV_COW_NEXT_SV(after);
5089 if (current == sv) {
5090 /* The SV we point to points back to us (there were only two of us
5092 Hence other SV is no longer copy on write either. */
5094 sv_buf_to_rw(after);
5096 /* We need to follow the pointers around the loop. */
5098 while ((next = SV_COW_NEXT_SV(current)) != sv) {
5101 /* don't loop forever if the structure is bust, and we have
5102 a pointer into a closed loop. */
5103 assert (current != after);
5104 assert (SvPVX_const(current) == pvx);
5106 /* Make the SV before us point to the SV after us. */
5107 SV_COW_NEXT_SV_SET(current, after);
5113 =for apidoc sv_force_normal_flags
5115 Undo various types of fakery on an SV, where fakery means
5116 "more than" a string: if the PV is a shared string, make
5117 a private copy; if we're a ref, stop refing; if we're a glob, downgrade to
5118 an xpvmg; if we're a copy-on-write scalar, this is the on-write time when
5119 we do the copy, and is also used locally; if this is a
5120 vstring, drop the vstring magic. If C<SV_COW_DROP_PV> is set
5121 then a copy-on-write scalar drops its PV buffer (if any) and becomes
5122 SvPOK_off rather than making a copy. (Used where this
5123 scalar is about to be set to some other value.) In addition,
5124 the C<flags> parameter gets passed to C<sv_unref_flags()>
5125 when unreffing. C<sv_force_normal> calls this function
5126 with flags set to 0.
5128 This function is expected to be used to signal to perl that this SV is
5129 about to be written to, and any extra book-keeping needs to be taken care
5130 of. Hence, it croaks on read-only values.
5136 S_sv_uncow(pTHX_ SV * const sv, const U32 flags)
5138 assert(SvIsCOW(sv));
5141 const char * const pvx = SvPVX_const(sv);
5142 const STRLEN len = SvLEN(sv);
5143 const STRLEN cur = SvCUR(sv);
5144 # ifdef PERL_OLD_COPY_ON_WRITE
5145 /* next COW sv in the loop. If len is 0 then this is a shared-hash
5146 key scalar, so we mustn't attempt to call SV_COW_NEXT_SV(), as
5147 we'll fail an assertion. */
5148 SV * const next = len ? SV_COW_NEXT_SV(sv) : 0;
5152 PerlIO_printf(Perl_debug_log,
5153 "Copy on write: Force normal %ld\n",
5158 # ifdef PERL_NEW_COPY_ON_WRITE
5159 if (len && CowREFCNT(sv) == 0)
5160 /* We own the buffer ourselves. */
5166 /* This SV doesn't own the buffer, so need to Newx() a new one: */
5167 # ifdef PERL_NEW_COPY_ON_WRITE
5168 /* Must do this first, since the macro uses SvPVX. */
5178 if (flags & SV_COW_DROP_PV) {
5179 /* OK, so we don't need to copy our buffer. */
5182 SvGROW(sv, cur + 1);
5183 Move(pvx,SvPVX(sv),cur,char);
5188 # ifdef PERL_OLD_COPY_ON_WRITE
5189 sv_release_COW(sv, pvx, next);
5192 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5199 const char * const pvx = SvPVX_const(sv);
5200 const STRLEN len = SvCUR(sv);
5204 if (flags & SV_COW_DROP_PV) {
5205 /* OK, so we don't need to copy our buffer. */
5208 SvGROW(sv, len + 1);
5209 Move(pvx,SvPVX(sv),len,char);
5212 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5218 Perl_sv_force_normal_flags(pTHX_ SV *const sv, const U32 flags)
5220 PERL_ARGS_ASSERT_SV_FORCE_NORMAL_FLAGS;
5223 Perl_croak_no_modify();
5224 else if (SvIsCOW(sv) && LIKELY(SvTYPE(sv) != SVt_PVHV))
5225 S_sv_uncow(aTHX_ sv, flags);
5227 sv_unref_flags(sv, flags);
5228 else if (SvFAKE(sv) && isGV_with_GP(sv))
5229 sv_unglob(sv, flags);
5230 else if (SvFAKE(sv) && isREGEXP(sv)) {
5231 /* Need to downgrade the REGEXP to a simple(r) scalar. This is analogous
5232 to sv_unglob. We only need it here, so inline it. */
5233 const bool islv = SvTYPE(sv) == SVt_PVLV;
5234 const svtype new_type =
5235 islv ? SVt_NULL : SvMAGIC(sv) || SvSTASH(sv) ? SVt_PVMG : SVt_PV;
5236 SV *const temp = newSV_type(new_type);
5237 regexp *const temp_p = ReANY((REGEXP *)sv);
5239 if (new_type == SVt_PVMG) {
5240 SvMAGIC_set(temp, SvMAGIC(sv));
5241 SvMAGIC_set(sv, NULL);
5242 SvSTASH_set(temp, SvSTASH(sv));
5243 SvSTASH_set(sv, NULL);
5245 if (!islv) SvCUR_set(temp, SvCUR(sv));
5246 /* Remember that SvPVX is in the head, not the body. But
5247 RX_WRAPPED is in the body. */
5248 assert(ReANY((REGEXP *)sv)->mother_re);
5249 /* Their buffer is already owned by someone else. */
5250 if (flags & SV_COW_DROP_PV) {
5251 /* SvLEN is already 0. For SVt_REGEXP, we have a brand new
5252 zeroed body. For SVt_PVLV, it should have been set to 0
5253 before turning into a regexp. */
5254 assert(!SvLEN(islv ? sv : temp));
5255 sv->sv_u.svu_pv = 0;
5258 sv->sv_u.svu_pv = savepvn(RX_WRAPPED((REGEXP *)sv), SvCUR(sv));
5259 SvLEN_set(islv ? sv : temp, SvCUR(sv)+1);
5263 /* Now swap the rest of the bodies. */
5267 SvFLAGS(sv) &= ~SVTYPEMASK;
5268 SvFLAGS(sv) |= new_type;
5269 SvANY(sv) = SvANY(temp);
5272 SvFLAGS(temp) &= ~(SVTYPEMASK);
5273 SvFLAGS(temp) |= SVt_REGEXP|SVf_FAKE;
5274 SvANY(temp) = temp_p;
5275 temp->sv_u.svu_rx = (regexp *)temp_p;
5277 SvREFCNT_dec_NN(temp);
5279 else if (SvVOK(sv)) sv_unmagic(sv, PERL_MAGIC_vstring);
5285 Efficient removal of characters from the beginning of the string buffer.
5286 SvPOK(sv), or at least SvPOKp(sv), must be true and the C<ptr> must be a
5287 pointer to somewhere inside the string buffer. The C<ptr> becomes the first
5288 character of the adjusted string. Uses the "OOK hack". On return, only
5289 SvPOK(sv) and SvPOKp(sv) among the OK flags will be true.
5291 Beware: after this function returns, C<ptr> and SvPVX_const(sv) may no longer
5292 refer to the same chunk of data.
5294 The unfortunate similarity of this function's name to that of Perl's C<chop>
5295 operator is strictly coincidental. This function works from the left;
5296 C<chop> works from the right.
5302 Perl_sv_chop(pTHX_ SV *const sv, const char *const ptr)
5313 PERL_ARGS_ASSERT_SV_CHOP;
5315 if (!ptr || !SvPOKp(sv))
5317 delta = ptr - SvPVX_const(sv);
5319 /* Nothing to do. */
5322 max_delta = SvLEN(sv) ? SvLEN(sv) : SvCUR(sv);
5323 if (delta > max_delta)
5324 Perl_croak(aTHX_ "panic: sv_chop ptr=%p, start=%p, end=%p",
5325 ptr, SvPVX_const(sv), SvPVX_const(sv) + max_delta);
5326 /* SvPVX(sv) may move in SV_CHECK_THINKFIRST(sv), so don't use ptr any more */
5327 SV_CHECK_THINKFIRST(sv);
5328 SvPOK_only_UTF8(sv);
5331 if (!SvLEN(sv)) { /* make copy of shared string */
5332 const char *pvx = SvPVX_const(sv);
5333 const STRLEN len = SvCUR(sv);
5334 SvGROW(sv, len + 1);
5335 Move(pvx,SvPVX(sv),len,char);
5341 SvOOK_offset(sv, old_delta);
5343 SvLEN_set(sv, SvLEN(sv) - delta);
5344 SvCUR_set(sv, SvCUR(sv) - delta);
5345 SvPV_set(sv, SvPVX(sv) + delta);
5347 p = (U8 *)SvPVX_const(sv);
5350 /* how many bytes were evacuated? we will fill them with sentinel
5351 bytes, except for the part holding the new offset of course. */
5354 evacn += (old_delta < 0x100 ? 1 : 1 + sizeof(STRLEN));
5356 assert(evacn <= delta + old_delta);
5360 /* This sets 'delta' to the accumulated value of all deltas so far */
5364 /* If 'delta' fits in a byte, store it just prior to the new beginning of
5365 * the string; otherwise store a 0 byte there and store 'delta' just prior
5366 * to that, using as many bytes as a STRLEN occupies. Thus it overwrites a
5367 * portion of the chopped part of the string */
5368 if (delta < 0x100) {
5372 p -= sizeof(STRLEN);
5373 Copy((U8*)&delta, p, sizeof(STRLEN), U8);
5377 /* Fill the preceding buffer with sentinals to verify that no-one is
5387 =for apidoc sv_catpvn
5389 Concatenates the string onto the end of the string which is in the SV. The
5390 C<len> indicates number of bytes to copy. If the SV has the UTF-8
5391 status set, then the bytes appended should be valid UTF-8.
5392 Handles 'get' magic, but not 'set' magic. See C<sv_catpvn_mg>.
5394 =for apidoc sv_catpvn_flags
5396 Concatenates the string onto the end of the string which is in the SV. The
5397 C<len> indicates number of bytes to copy.
5399 By default, the string appended is assumed to be valid UTF-8 if the SV has
5400 the UTF-8 status set, and a string of bytes otherwise. One can force the
5401 appended string to be interpreted as UTF-8 by supplying the C<SV_CATUTF8>
5402 flag, and as bytes by supplying the C<SV_CATBYTES> flag; the SV or the
5403 string appended will be upgraded to UTF-8 if necessary.
5405 If C<flags> has the C<SV_SMAGIC> bit set, will
5406 C<mg_set> on C<dsv> afterwards if appropriate.
5407 C<sv_catpvn> and C<sv_catpvn_nomg> are implemented
5408 in terms of this function.
5414 Perl_sv_catpvn_flags(pTHX_ SV *const dsv, const char *sstr, const STRLEN slen, const I32 flags)
5417 const char * const dstr = SvPV_force_flags(dsv, dlen, flags);
5419 PERL_ARGS_ASSERT_SV_CATPVN_FLAGS;
5420 assert((flags & (SV_CATBYTES|SV_CATUTF8)) != (SV_CATBYTES|SV_CATUTF8));
5422 if (!(flags & SV_CATBYTES) || !SvUTF8(dsv)) {
5423 if (flags & SV_CATUTF8 && !SvUTF8(dsv)) {
5424 sv_utf8_upgrade_flags_grow(dsv, 0, slen + 1);
5427 else SvGROW(dsv, dlen + slen + 1);
5429 sstr = SvPVX_const(dsv);
5430 Move(sstr, SvPVX(dsv) + dlen, slen, char);
5431 SvCUR_set(dsv, SvCUR(dsv) + slen);
5434 /* We inline bytes_to_utf8, to avoid an extra malloc. */
5435 const char * const send = sstr + slen;
5438 /* Something this code does not account for, which I think is
5439 impossible; it would require the same pv to be treated as
5440 bytes *and* utf8, which would indicate a bug elsewhere. */
5441 assert(sstr != dstr);
5443 SvGROW(dsv, dlen + slen * 2 + 1);
5444 d = (U8 *)SvPVX(dsv) + dlen;
5446 while (sstr < send) {
5447 append_utf8_from_native_byte(*sstr, &d);
5450 SvCUR_set(dsv, d-(const U8 *)SvPVX(dsv));
5453 (void)SvPOK_only_UTF8(dsv); /* validate pointer */
5455 if (flags & SV_SMAGIC)
5460 =for apidoc sv_catsv
5462 Concatenates the string from SV C<ssv> onto the end of the string in SV
5463 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5464 Handles 'get' magic on both SVs, but no 'set' magic. See C<sv_catsv_mg> and
5467 =for apidoc sv_catsv_flags
5469 Concatenates the string from SV C<ssv> onto the end of the string in SV
5470 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5471 If C<flags> include C<SV_GMAGIC> bit set, will call C<mg_get> on both SVs if
5472 appropriate. If C<flags> include C<SV_SMAGIC>, C<mg_set> will be called on
5473 the modified SV afterward, if appropriate. C<sv_catsv>, C<sv_catsv_nomg>,
5474 and C<sv_catsv_mg> are implemented in terms of this function.
5479 Perl_sv_catsv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
5481 PERL_ARGS_ASSERT_SV_CATSV_FLAGS;
5485 const char *spv = SvPV_flags_const(ssv, slen, flags);
5486 if (flags & SV_GMAGIC)
5488 sv_catpvn_flags(dsv, spv, slen,
5489 DO_UTF8(ssv) ? SV_CATUTF8 : SV_CATBYTES);
5490 if (flags & SV_SMAGIC)
5496 =for apidoc sv_catpv
5498 Concatenates the C<NUL>-terminated string onto the end of the string which is
5500 If the SV has the UTF-8 status set, then the bytes appended should be
5501 valid UTF-8. Handles 'get' magic, but not 'set' magic. See C<sv_catpv_mg>.
5506 Perl_sv_catpv(pTHX_ SV *const sv, const char *ptr)
5512 PERL_ARGS_ASSERT_SV_CATPV;
5516 junk = SvPV_force(sv, tlen);
5518 SvGROW(sv, tlen + len + 1);
5520 ptr = SvPVX_const(sv);
5521 Move(ptr,SvPVX(sv)+tlen,len+1,char);
5522 SvCUR_set(sv, SvCUR(sv) + len);
5523 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5528 =for apidoc sv_catpv_flags
5530 Concatenates the C<NUL>-terminated string onto the end of the string which is
5532 If the SV has the UTF-8 status set, then the bytes appended should
5533 be valid UTF-8. If C<flags> has the C<SV_SMAGIC> bit set, will C<mg_set>
5534 on the modified SV if appropriate.
5540 Perl_sv_catpv_flags(pTHX_ SV *dstr, const char *sstr, const I32 flags)
5542 PERL_ARGS_ASSERT_SV_CATPV_FLAGS;
5543 sv_catpvn_flags(dstr, sstr, strlen(sstr), flags);
5547 =for apidoc sv_catpv_mg
5549 Like C<sv_catpv>, but also handles 'set' magic.
5555 Perl_sv_catpv_mg(pTHX_ SV *const sv, const char *const ptr)
5557 PERL_ARGS_ASSERT_SV_CATPV_MG;
5566 Creates a new SV. A non-zero C<len> parameter indicates the number of
5567 bytes of preallocated string space the SV should have. An extra byte for a
5568 trailing C<NUL> is also reserved. (SvPOK is not set for the SV even if string
5569 space is allocated.) The reference count for the new SV is set to 1.
5571 In 5.9.3, newSV() replaces the older NEWSV() API, and drops the first
5572 parameter, I<x>, a debug aid which allowed callers to identify themselves.
5573 This aid has been superseded by a new build option, PERL_MEM_LOG (see
5574 L<perlhacktips/PERL_MEM_LOG>). The older API is still there for use in XS
5575 modules supporting older perls.
5581 Perl_newSV(pTHX_ const STRLEN len)
5587 sv_grow(sv, len + 1);
5592 =for apidoc sv_magicext
5594 Adds magic to an SV, upgrading it if necessary. Applies the
5595 supplied vtable and returns a pointer to the magic added.
5597 Note that C<sv_magicext> will allow things that C<sv_magic> will not.
5598 In particular, you can add magic to SvREADONLY SVs, and add more than
5599 one instance of the same 'how'.
5601 If C<namlen> is greater than zero then a C<savepvn> I<copy> of C<name> is
5602 stored, if C<namlen> is zero then C<name> is stored as-is and - as another
5603 special case - if C<(name && namlen == HEf_SVKEY)> then C<name> is assumed
5604 to contain an C<SV*> and is stored as-is with its REFCNT incremented.
5606 (This is now used as a subroutine by C<sv_magic>.)
5611 Perl_sv_magicext(pTHX_ SV *const sv, SV *const obj, const int how,
5612 const MGVTBL *const vtable, const char *const name, const I32 namlen)
5616 PERL_ARGS_ASSERT_SV_MAGICEXT;
5618 if (SvTYPE(sv)==SVt_PVAV) { assert (!AvPAD_NAMELIST(sv)); }
5620 SvUPGRADE(sv, SVt_PVMG);
5621 Newxz(mg, 1, MAGIC);
5622 mg->mg_moremagic = SvMAGIC(sv);
5623 SvMAGIC_set(sv, mg);
5625 /* Sometimes a magic contains a reference loop, where the sv and
5626 object refer to each other. To prevent a reference loop that
5627 would prevent such objects being freed, we look for such loops
5628 and if we find one we avoid incrementing the object refcount.
5630 Note we cannot do this to avoid self-tie loops as intervening RV must
5631 have its REFCNT incremented to keep it in existence.
5634 if (!obj || obj == sv ||
5635 how == PERL_MAGIC_arylen ||
5636 how == PERL_MAGIC_symtab ||
5637 (SvTYPE(obj) == SVt_PVGV &&
5638 (GvSV(obj) == sv || GvHV(obj) == (const HV *)sv
5639 || GvAV(obj) == (const AV *)sv || GvCV(obj) == (const CV *)sv
5640 || GvIOp(obj) == (const IO *)sv || GvFORM(obj) == (const CV *)sv)))
5645 mg->mg_obj = SvREFCNT_inc_simple(obj);
5646 mg->mg_flags |= MGf_REFCOUNTED;
5649 /* Normal self-ties simply pass a null object, and instead of
5650 using mg_obj directly, use the SvTIED_obj macro to produce a
5651 new RV as needed. For glob "self-ties", we are tieing the PVIO
5652 with an RV obj pointing to the glob containing the PVIO. In
5653 this case, to avoid a reference loop, we need to weaken the
5657 if (how == PERL_MAGIC_tiedscalar && SvTYPE(sv) == SVt_PVIO &&
5658 obj && SvROK(obj) && GvIO(SvRV(obj)) == (const IO *)sv)
5664 mg->mg_len = namlen;
5667 mg->mg_ptr = savepvn(name, namlen);
5668 else if (namlen == HEf_SVKEY) {
5669 /* Yes, this is casting away const. This is only for the case of
5670 HEf_SVKEY. I think we need to document this aberation of the
5671 constness of the API, rather than making name non-const, as
5672 that change propagating outwards a long way. */
5673 mg->mg_ptr = (char*)SvREFCNT_inc_simple_NN((SV *)name);
5675 mg->mg_ptr = (char *) name;
5677 mg->mg_virtual = (MGVTBL *) vtable;
5684 Perl_sv_magicext_mglob(pTHX_ SV *sv)
5686 PERL_ARGS_ASSERT_SV_MAGICEXT_MGLOB;
5687 if (SvTYPE(sv) == SVt_PVLV && LvTYPE(sv) == 'y') {
5688 /* This sv is only a delegate. //g magic must be attached to
5693 #ifdef PERL_OLD_COPY_ON_WRITE
5695 sv_force_normal_flags(sv, 0);
5697 return sv_magicext(sv, NULL, PERL_MAGIC_regex_global,
5698 &PL_vtbl_mglob, 0, 0);
5702 =for apidoc sv_magic
5704 Adds magic to an SV. First upgrades C<sv> to type C<SVt_PVMG> if
5705 necessary, then adds a new magic item of type C<how> to the head of the
5708 See C<sv_magicext> (which C<sv_magic> now calls) for a description of the
5709 handling of the C<name> and C<namlen> arguments.
5711 You need to use C<sv_magicext> to add magic to SvREADONLY SVs and also
5712 to add more than one instance of the same 'how'.
5718 Perl_sv_magic(pTHX_ SV *const sv, SV *const obj, const int how,
5719 const char *const name, const I32 namlen)
5721 const MGVTBL *vtable;
5724 unsigned int vtable_index;
5726 PERL_ARGS_ASSERT_SV_MAGIC;
5728 if (how < 0 || (unsigned)how >= C_ARRAY_LENGTH(PL_magic_data)
5729 || ((flags = PL_magic_data[how]),
5730 (vtable_index = flags & PERL_MAGIC_VTABLE_MASK)
5731 > magic_vtable_max))
5732 Perl_croak(aTHX_ "Don't know how to handle magic of type \\%o", how);
5734 /* PERL_MAGIC_ext is reserved for use by extensions not perl internals.
5735 Useful for attaching extension internal data to perl vars.
5736 Note that multiple extensions may clash if magical scalars
5737 etc holding private data from one are passed to another. */
5739 vtable = (vtable_index == magic_vtable_max)
5740 ? NULL : PL_magic_vtables + vtable_index;
5742 #ifdef PERL_OLD_COPY_ON_WRITE
5744 sv_force_normal_flags(sv, 0);
5746 if (SvREADONLY(sv)) {
5748 !PERL_MAGIC_TYPE_READONLY_ACCEPTABLE(how)
5751 Perl_croak_no_modify();
5754 if (SvMAGICAL(sv) || (how == PERL_MAGIC_taint && SvTYPE(sv) >= SVt_PVMG)) {
5755 if (SvMAGIC(sv) && (mg = mg_find(sv, how))) {
5756 /* sv_magic() refuses to add a magic of the same 'how' as an
5759 if (how == PERL_MAGIC_taint)
5765 /* Force pos to be stored as characters, not bytes. */
5766 if (SvMAGICAL(sv) && DO_UTF8(sv)
5767 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
5769 && mg->mg_flags & MGf_BYTES) {
5770 mg->mg_len = (SSize_t)sv_pos_b2u_flags(sv, (STRLEN)mg->mg_len,
5772 mg->mg_flags &= ~MGf_BYTES;
5775 /* Rest of work is done else where */
5776 mg = sv_magicext(sv,obj,how,vtable,name,namlen);
5779 case PERL_MAGIC_taint:
5782 case PERL_MAGIC_ext:
5783 case PERL_MAGIC_dbfile:
5790 S_sv_unmagicext_flags(pTHX_ SV *const sv, const int type, MGVTBL *vtbl, const U32 flags)
5797 if (SvTYPE(sv) < SVt_PVMG || !SvMAGIC(sv))
5799 mgp = &(((XPVMG*) SvANY(sv))->xmg_u.xmg_magic);
5800 for (mg = *mgp; mg; mg = *mgp) {
5801 const MGVTBL* const virt = mg->mg_virtual;
5802 if (mg->mg_type == type && (!flags || virt == vtbl)) {
5803 *mgp = mg->mg_moremagic;
5804 if (virt && virt->svt_free)
5805 virt->svt_free(aTHX_ sv, mg);
5806 if (mg->mg_ptr && mg->mg_type != PERL_MAGIC_regex_global) {
5808 Safefree(mg->mg_ptr);
5809 else if (mg->mg_len == HEf_SVKEY)
5810 SvREFCNT_dec(MUTABLE_SV(mg->mg_ptr));
5811 else if (mg->mg_type == PERL_MAGIC_utf8)
5812 Safefree(mg->mg_ptr);
5814 if (mg->mg_flags & MGf_REFCOUNTED)
5815 SvREFCNT_dec(mg->mg_obj);
5819 mgp = &mg->mg_moremagic;
5822 if (SvMAGICAL(sv)) /* if we're under save_magic, wait for restore_magic; */
5823 mg_magical(sv); /* else fix the flags now */
5827 SvFLAGS(sv) |= (SvFLAGS(sv) & (SVp_IOK|SVp_NOK|SVp_POK)) >> PRIVSHIFT;
5833 =for apidoc sv_unmagic
5835 Removes all magic of type C<type> from an SV.
5841 Perl_sv_unmagic(pTHX_ SV *const sv, const int type)
5843 PERL_ARGS_ASSERT_SV_UNMAGIC;
5844 return S_sv_unmagicext_flags(aTHX_ sv, type, NULL, 0);
5848 =for apidoc sv_unmagicext
5850 Removes all magic of type C<type> with the specified C<vtbl> from an SV.
5856 Perl_sv_unmagicext(pTHX_ SV *const sv, const int type, MGVTBL *vtbl)
5858 PERL_ARGS_ASSERT_SV_UNMAGICEXT;
5859 return S_sv_unmagicext_flags(aTHX_ sv, type, vtbl, 1);
5863 =for apidoc sv_rvweaken
5865 Weaken a reference: set the C<SvWEAKREF> flag on this RV; give the
5866 referred-to SV C<PERL_MAGIC_backref> magic if it hasn't already; and
5867 push a back-reference to this RV onto the array of backreferences
5868 associated with that magic. If the RV is magical, set magic will be
5869 called after the RV is cleared.
5875 Perl_sv_rvweaken(pTHX_ SV *const sv)
5879 PERL_ARGS_ASSERT_SV_RVWEAKEN;
5881 if (!SvOK(sv)) /* let undefs pass */
5884 Perl_croak(aTHX_ "Can't weaken a nonreference");
5885 else if (SvWEAKREF(sv)) {
5886 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is already weak");
5889 else if (SvREADONLY(sv)) croak_no_modify();
5891 Perl_sv_add_backref(aTHX_ tsv, sv);
5893 SvREFCNT_dec_NN(tsv);
5897 /* Give tsv backref magic if it hasn't already got it, then push a
5898 * back-reference to sv onto the array associated with the backref magic.
5900 * As an optimisation, if there's only one backref and it's not an AV,
5901 * store it directly in the HvAUX or mg_obj slot, avoiding the need to
5902 * allocate an AV. (Whether the slot holds an AV tells us whether this is
5906 /* A discussion about the backreferences array and its refcount:
5908 * The AV holding the backreferences is pointed to either as the mg_obj of
5909 * PERL_MAGIC_backref, or in the specific case of a HV, from the
5910 * xhv_backreferences field. The array is created with a refcount
5911 * of 2. This means that if during global destruction the array gets
5912 * picked on before its parent to have its refcount decremented by the
5913 * random zapper, it won't actually be freed, meaning it's still there for
5914 * when its parent gets freed.
5916 * When the parent SV is freed, the extra ref is killed by
5917 * Perl_sv_kill_backrefs. The other ref is killed, in the case of magic,
5918 * by mg_free() / MGf_REFCOUNTED, or for a hash, by Perl_hv_kill_backrefs.
5920 * When a single backref SV is stored directly, it is not reference
5925 Perl_sv_add_backref(pTHX_ SV *const tsv, SV *const sv)
5931 PERL_ARGS_ASSERT_SV_ADD_BACKREF;
5933 /* find slot to store array or singleton backref */
5935 if (SvTYPE(tsv) == SVt_PVHV) {
5936 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
5939 mg = mg_find(tsv, PERL_MAGIC_backref);
5941 mg = sv_magicext(tsv, NULL, PERL_MAGIC_backref, &PL_vtbl_backref, NULL, 0);
5942 svp = &(mg->mg_obj);
5945 /* create or retrieve the array */
5947 if ( (!*svp && SvTYPE(sv) == SVt_PVAV)
5948 || (*svp && SvTYPE(*svp) != SVt_PVAV)
5952 mg->mg_flags |= MGf_REFCOUNTED;
5955 SvREFCNT_inc_simple_void_NN(av);
5956 /* av now has a refcnt of 2; see discussion above */
5957 av_extend(av, *svp ? 2 : 1);
5959 /* move single existing backref to the array */
5960 AvARRAY(av)[++AvFILLp(av)] = *svp; /* av_push() */
5965 av = MUTABLE_AV(*svp);
5967 /* optimisation: store single backref directly in HvAUX or mg_obj */
5971 assert(SvTYPE(av) == SVt_PVAV);
5972 if (AvFILLp(av) >= AvMAX(av)) {
5973 av_extend(av, AvFILLp(av)+1);
5976 /* push new backref */
5977 AvARRAY(av)[++AvFILLp(av)] = sv; /* av_push() */
5980 /* delete a back-reference to ourselves from the backref magic associated
5981 * with the SV we point to.
5985 Perl_sv_del_backref(pTHX_ SV *const tsv, SV *const sv)
5989 PERL_ARGS_ASSERT_SV_DEL_BACKREF;
5991 if (SvTYPE(tsv) == SVt_PVHV) {
5993 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
5995 else if (SvIS_FREED(tsv) && PL_phase == PERL_PHASE_DESTRUCT) {
5996 /* It's possible for the the last (strong) reference to tsv to have
5997 become freed *before* the last thing holding a weak reference.
5998 If both survive longer than the backreferences array, then when
5999 the referent's reference count drops to 0 and it is freed, it's
6000 not able to chase the backreferences, so they aren't NULLed.
6002 For example, a CV holds a weak reference to its stash. If both the
6003 CV and the stash survive longer than the backreferences array,
6004 and the CV gets picked for the SvBREAK() treatment first,
6005 *and* it turns out that the stash is only being kept alive because
6006 of an our variable in the pad of the CV, then midway during CV
6007 destruction the stash gets freed, but CvSTASH() isn't set to NULL.
6008 It ends up pointing to the freed HV. Hence it's chased in here, and
6009 if this block wasn't here, it would hit the !svp panic just below.
6011 I don't believe that "better" destruction ordering is going to help
6012 here - during global destruction there's always going to be the
6013 chance that something goes out of order. We've tried to make it
6014 foolproof before, and it only resulted in evolutionary pressure on
6015 fools. Which made us look foolish for our hubris. :-(
6021 = SvMAGICAL(tsv) ? mg_find(tsv, PERL_MAGIC_backref) : NULL;
6022 svp = mg ? &(mg->mg_obj) : NULL;
6026 Perl_croak(aTHX_ "panic: del_backref, svp=0");
6028 /* It's possible that sv is being freed recursively part way through the
6029 freeing of tsv. If this happens, the backreferences array of tsv has
6030 already been freed, and so svp will be NULL. If this is the case,
6031 we should not panic. Instead, nothing needs doing, so return. */
6032 if (PL_phase == PERL_PHASE_DESTRUCT && SvREFCNT(tsv) == 0)
6034 Perl_croak(aTHX_ "panic: del_backref, *svp=%p phase=%s refcnt=%" UVuf,
6035 (void*)*svp, PL_phase_names[PL_phase], (UV)SvREFCNT(tsv));
6038 if (SvTYPE(*svp) == SVt_PVAV) {
6042 AV * const av = (AV*)*svp;
6044 assert(!SvIS_FREED(av));
6048 /* for an SV with N weak references to it, if all those
6049 * weak refs are deleted, then sv_del_backref will be called
6050 * N times and O(N^2) compares will be done within the backref
6051 * array. To ameliorate this potential slowness, we:
6052 * 1) make sure this code is as tight as possible;
6053 * 2) when looking for SV, look for it at both the head and tail of the
6054 * array first before searching the rest, since some create/destroy
6055 * patterns will cause the backrefs to be freed in order.
6062 SV **p = &svp[fill];
6063 SV *const topsv = *p;
6070 /* We weren't the last entry.
6071 An unordered list has this property that you
6072 can take the last element off the end to fill
6073 the hole, and it's still an unordered list :-)
6079 break; /* should only be one */
6086 AvFILLp(av) = fill-1;
6088 else if (SvIS_FREED(*svp) && PL_phase == PERL_PHASE_DESTRUCT) {
6089 /* freed AV; skip */
6092 /* optimisation: only a single backref, stored directly */
6094 Perl_croak(aTHX_ "panic: del_backref, *svp=%p, sv=%p",
6095 (void*)*svp, (void*)sv);
6102 Perl_sv_kill_backrefs(pTHX_ SV *const sv, AV *const av)
6108 PERL_ARGS_ASSERT_SV_KILL_BACKREFS;
6113 /* after multiple passes through Perl_sv_clean_all() for a thingy
6114 * that has badly leaked, the backref array may have gotten freed,
6115 * since we only protect it against 1 round of cleanup */
6116 if (SvIS_FREED(av)) {
6117 if (PL_in_clean_all) /* All is fair */
6120 "panic: magic_killbackrefs (freed backref AV/SV)");
6124 is_array = (SvTYPE(av) == SVt_PVAV);
6126 assert(!SvIS_FREED(av));
6129 last = svp + AvFILLp(av);
6132 /* optimisation: only a single backref, stored directly */
6138 while (svp <= last) {
6140 SV *const referrer = *svp;
6141 if (SvWEAKREF(referrer)) {
6142 /* XXX Should we check that it hasn't changed? */
6143 assert(SvROK(referrer));
6144 SvRV_set(referrer, 0);
6146 SvWEAKREF_off(referrer);
6147 SvSETMAGIC(referrer);
6148 } else if (SvTYPE(referrer) == SVt_PVGV ||
6149 SvTYPE(referrer) == SVt_PVLV) {
6150 assert(SvTYPE(sv) == SVt_PVHV); /* stash backref */
6151 /* You lookin' at me? */
6152 assert(GvSTASH(referrer));
6153 assert(GvSTASH(referrer) == (const HV *)sv);
6154 GvSTASH(referrer) = 0;
6155 } else if (SvTYPE(referrer) == SVt_PVCV ||
6156 SvTYPE(referrer) == SVt_PVFM) {
6157 if (SvTYPE(sv) == SVt_PVHV) { /* stash backref */
6158 /* You lookin' at me? */
6159 assert(CvSTASH(referrer));
6160 assert(CvSTASH(referrer) == (const HV *)sv);
6161 SvANY(MUTABLE_CV(referrer))->xcv_stash = 0;
6164 assert(SvTYPE(sv) == SVt_PVGV);
6165 /* You lookin' at me? */
6166 assert(CvGV(referrer));
6167 assert(CvGV(referrer) == (const GV *)sv);
6168 anonymise_cv_maybe(MUTABLE_GV(sv),
6169 MUTABLE_CV(referrer));
6174 "panic: magic_killbackrefs (flags=%"UVxf")",
6175 (UV)SvFLAGS(referrer));
6186 SvREFCNT_dec_NN(av); /* remove extra count added by sv_add_backref() */
6192 =for apidoc sv_insert
6194 Inserts a string at the specified offset/length within the SV. Similar to
6195 the Perl substr() function. Handles get magic.
6197 =for apidoc sv_insert_flags
6199 Same as C<sv_insert>, but the extra C<flags> are passed to the
6200 C<SvPV_force_flags> that applies to C<bigstr>.
6206 Perl_sv_insert_flags(pTHX_ SV *const bigstr, const STRLEN offset, const STRLEN len, const char *const little, const STRLEN littlelen, const U32 flags)
6212 SSize_t i; /* better be sizeof(STRLEN) or bad things happen */
6215 PERL_ARGS_ASSERT_SV_INSERT_FLAGS;
6218 Perl_croak(aTHX_ "Can't modify nonexistent substring");
6219 SvPV_force_flags(bigstr, curlen, flags);
6220 (void)SvPOK_only_UTF8(bigstr);
6221 if (offset + len > curlen) {
6222 SvGROW(bigstr, offset+len+1);
6223 Zero(SvPVX(bigstr)+curlen, offset+len-curlen, char);
6224 SvCUR_set(bigstr, offset+len);
6228 i = littlelen - len;
6229 if (i > 0) { /* string might grow */
6230 big = SvGROW(bigstr, SvCUR(bigstr) + i + 1);
6231 mid = big + offset + len;
6232 midend = bigend = big + SvCUR(bigstr);
6235 while (midend > mid) /* shove everything down */
6236 *--bigend = *--midend;
6237 Move(little,big+offset,littlelen,char);
6238 SvCUR_set(bigstr, SvCUR(bigstr) + i);
6243 Move(little,SvPVX(bigstr)+offset,len,char);
6248 big = SvPVX(bigstr);
6251 bigend = big + SvCUR(bigstr);
6253 if (midend > bigend)
6254 Perl_croak(aTHX_ "panic: sv_insert, midend=%p, bigend=%p",
6257 if (mid - big > bigend - midend) { /* faster to shorten from end */
6259 Move(little, mid, littlelen,char);
6262 i = bigend - midend;
6264 Move(midend, mid, i,char);
6268 SvCUR_set(bigstr, mid - big);
6270 else if ((i = mid - big)) { /* faster from front */
6271 midend -= littlelen;
6273 Move(big, midend - i, i, char);
6274 sv_chop(bigstr,midend-i);
6276 Move(little, mid, littlelen,char);
6278 else if (littlelen) {
6279 midend -= littlelen;
6280 sv_chop(bigstr,midend);
6281 Move(little,midend,littlelen,char);
6284 sv_chop(bigstr,midend);
6290 =for apidoc sv_replace
6292 Make the first argument a copy of the second, then delete the original.
6293 The target SV physically takes over ownership of the body of the source SV
6294 and inherits its flags; however, the target keeps any magic it owns,
6295 and any magic in the source is discarded.
6296 Note that this is a rather specialist SV copying operation; most of the
6297 time you'll want to use C<sv_setsv> or one of its many macro front-ends.
6303 Perl_sv_replace(pTHX_ SV *const sv, SV *const nsv)
6305 const U32 refcnt = SvREFCNT(sv);
6307 PERL_ARGS_ASSERT_SV_REPLACE;
6309 SV_CHECK_THINKFIRST_COW_DROP(sv);
6310 if (SvREFCNT(nsv) != 1) {
6311 Perl_croak(aTHX_ "panic: reference miscount on nsv in sv_replace()"
6312 " (%" UVuf " != 1)", (UV) SvREFCNT(nsv));
6314 if (SvMAGICAL(sv)) {
6318 sv_upgrade(nsv, SVt_PVMG);
6319 SvMAGIC_set(nsv, SvMAGIC(sv));
6320 SvFLAGS(nsv) |= SvMAGICAL(sv);
6322 SvMAGIC_set(sv, NULL);
6326 assert(!SvREFCNT(sv));
6327 #ifdef DEBUG_LEAKING_SCALARS
6328 sv->sv_flags = nsv->sv_flags;
6329 sv->sv_any = nsv->sv_any;
6330 sv->sv_refcnt = nsv->sv_refcnt;
6331 sv->sv_u = nsv->sv_u;
6333 StructCopy(nsv,sv,SV);
6335 if(SvTYPE(sv) == SVt_IV) {
6337 = (XPVIV*)((char*)&(sv->sv_u.svu_iv) - STRUCT_OFFSET(XPVIV, xiv_iv));
6341 #ifdef PERL_OLD_COPY_ON_WRITE
6342 if (SvIsCOW_normal(nsv)) {
6343 /* We need to follow the pointers around the loop to make the
6344 previous SV point to sv, rather than nsv. */
6347 while ((next = SV_COW_NEXT_SV(current)) != nsv) {
6350 assert(SvPVX_const(current) == SvPVX_const(nsv));
6352 /* Make the SV before us point to the SV after us. */
6354 PerlIO_printf(Perl_debug_log, "previous is\n");
6356 PerlIO_printf(Perl_debug_log,
6357 "move it from 0x%"UVxf" to 0x%"UVxf"\n",
6358 (UV) SV_COW_NEXT_SV(current), (UV) sv);
6360 SV_COW_NEXT_SV_SET(current, sv);
6363 SvREFCNT(sv) = refcnt;
6364 SvFLAGS(nsv) |= SVTYPEMASK; /* Mark as freed */
6369 /* We're about to free a GV which has a CV that refers back to us.
6370 * If that CV will outlive us, make it anonymous (i.e. fix up its CvGV
6374 S_anonymise_cv_maybe(pTHX_ GV *gv, CV* cv)
6379 PERL_ARGS_ASSERT_ANONYMISE_CV_MAYBE;
6382 assert(SvREFCNT(gv) == 0);
6383 assert(isGV(gv) && isGV_with_GP(gv));
6385 assert(!CvANON(cv));
6386 assert(CvGV(cv) == gv);
6387 assert(!CvNAMED(cv));
6389 /* will the CV shortly be freed by gp_free() ? */
6390 if (GvCV(gv) == cv && GvGP(gv)->gp_refcnt < 2 && SvREFCNT(cv) < 2) {
6391 SvANY(cv)->xcv_gv_u.xcv_gv = NULL;
6395 /* if not, anonymise: */
6396 gvname = (GvSTASH(gv) && HvNAME(GvSTASH(gv)) && HvENAME(GvSTASH(gv)))
6397 ? newSVhek(HvENAME_HEK(GvSTASH(gv)))
6398 : newSVpvn_flags( "__ANON__", 8, 0 );
6399 sv_catpvs(gvname, "::__ANON__");
6400 anongv = gv_fetchsv(gvname, GV_ADDMULTI, SVt_PVCV);
6401 SvREFCNT_dec_NN(gvname);
6405 SvANY(cv)->xcv_gv_u.xcv_gv = MUTABLE_GV(SvREFCNT_inc(anongv));
6410 =for apidoc sv_clear
6412 Clear an SV: call any destructors, free up any memory used by the body,
6413 and free the body itself. The SV's head is I<not> freed, although
6414 its type is set to all 1's so that it won't inadvertently be assumed
6415 to be live during global destruction etc.
6416 This function should only be called when REFCNT is zero. Most of the time
6417 you'll want to call C<sv_free()> (or its macro wrapper C<SvREFCNT_dec>)
6424 Perl_sv_clear(pTHX_ SV *const orig_sv)
6429 const struct body_details *sv_type_details;
6435 PERL_ARGS_ASSERT_SV_CLEAR;
6437 /* within this loop, sv is the SV currently being freed, and
6438 * iter_sv is the most recent AV or whatever that's being iterated
6439 * over to provide more SVs */
6445 assert(SvREFCNT(sv) == 0);
6446 assert(SvTYPE(sv) != (svtype)SVTYPEMASK);
6448 if (type <= SVt_IV) {
6449 /* See the comment in sv.h about the collusion between this
6450 * early return and the overloading of the NULL slots in the
6454 SvFLAGS(sv) &= SVf_BREAK;
6455 SvFLAGS(sv) |= SVTYPEMASK;
6459 assert(!SvOBJECT(sv) || type >= SVt_PVMG); /* objs are always >= MG */
6461 if (type >= SVt_PVMG) {
6463 if (!curse(sv, 1)) goto get_next_sv;
6464 type = SvTYPE(sv); /* destructor may have changed it */
6466 /* Free back-references before magic, in case the magic calls
6467 * Perl code that has weak references to sv. */
6468 if (type == SVt_PVHV) {
6469 Perl_hv_kill_backrefs(aTHX_ MUTABLE_HV(sv));
6473 else if (type == SVt_PVMG && SvPAD_OUR(sv)) {
6474 SvREFCNT_dec(SvOURSTASH(sv));
6476 else if (type == SVt_PVAV && AvPAD_NAMELIST(sv)) {
6477 assert(!SvMAGICAL(sv));
6478 } else if (SvMAGIC(sv)) {
6479 /* Free back-references before other types of magic. */
6480 sv_unmagic(sv, PERL_MAGIC_backref);
6484 if (type == SVt_PVMG && SvPAD_TYPED(sv))
6485 SvREFCNT_dec(SvSTASH(sv));
6488 /* case SVt_INVLIST: */
6491 IoIFP(sv) != PerlIO_stdin() &&
6492 IoIFP(sv) != PerlIO_stdout() &&
6493 IoIFP(sv) != PerlIO_stderr() &&
6494 !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6496 io_close(MUTABLE_IO(sv), NULL, FALSE,
6497 (IoTYPE(sv) == IoTYPE_WRONLY ||
6498 IoTYPE(sv) == IoTYPE_RDWR ||
6499 IoTYPE(sv) == IoTYPE_APPEND));
6501 if (IoDIRP(sv) && !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6502 PerlDir_close(IoDIRP(sv));
6503 IoDIRP(sv) = (DIR*)NULL;
6504 Safefree(IoTOP_NAME(sv));
6505 Safefree(IoFMT_NAME(sv));
6506 Safefree(IoBOTTOM_NAME(sv));
6507 if ((const GV *)sv == PL_statgv)
6511 /* FIXME for plugins */
6513 pregfree2((REGEXP*) sv);
6517 cv_undef(MUTABLE_CV(sv));
6518 /* If we're in a stash, we don't own a reference to it.
6519 * However it does have a back reference to us, which needs to
6521 if ((stash = CvSTASH(sv)))
6522 sv_del_backref(MUTABLE_SV(stash), sv);
6525 if (PL_last_swash_hv == (const HV *)sv) {
6526 PL_last_swash_hv = NULL;
6528 if (HvTOTALKEYS((HV*)sv) > 0) {
6530 /* this statement should match the one at the beginning of
6531 * hv_undef_flags() */
6532 if ( PL_phase != PERL_PHASE_DESTRUCT
6533 && (name = HvNAME((HV*)sv)))
6535 if (PL_stashcache) {
6536 DEBUG_o(Perl_deb(aTHX_ "sv_clear clearing PL_stashcache for '%"SVf"'\n",
6538 (void)hv_deletehek(PL_stashcache,
6539 HvNAME_HEK((HV*)sv), G_DISCARD);
6541 hv_name_set((HV*)sv, NULL, 0, 0);
6544 /* save old iter_sv in unused SvSTASH field */
6545 assert(!SvOBJECT(sv));
6546 SvSTASH(sv) = (HV*)iter_sv;
6549 /* save old hash_index in unused SvMAGIC field */
6550 assert(!SvMAGICAL(sv));
6551 assert(!SvMAGIC(sv));
6552 ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index = hash_index;
6555 next_sv = Perl_hfree_next_entry(aTHX_ (HV*)sv, &hash_index);
6556 goto get_next_sv; /* process this new sv */
6558 /* free empty hash */
6559 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6560 assert(!HvARRAY((HV*)sv));
6564 AV* av = MUTABLE_AV(sv);
6565 if (PL_comppad == av) {
6569 if (AvREAL(av) && AvFILLp(av) > -1) {
6570 next_sv = AvARRAY(av)[AvFILLp(av)--];
6571 /* save old iter_sv in top-most slot of AV,
6572 * and pray that it doesn't get wiped in the meantime */
6573 AvARRAY(av)[AvMAX(av)] = iter_sv;
6575 goto get_next_sv; /* process this new sv */
6577 Safefree(AvALLOC(av));
6582 if (LvTYPE(sv) == 'T') { /* for tie: return HE to pool */
6583 SvREFCNT_dec(HeKEY_sv((HE*)LvTARG(sv)));
6584 HeNEXT((HE*)LvTARG(sv)) = PL_hv_fetch_ent_mh;
6585 PL_hv_fetch_ent_mh = (HE*)LvTARG(sv);
6587 else if (LvTYPE(sv) != 't') /* unless tie: unrefcnted fake SV** */
6588 SvREFCNT_dec(LvTARG(sv));
6589 if (isREGEXP(sv)) goto freeregexp;
6591 if (isGV_with_GP(sv)) {
6592 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
6593 && HvENAME_get(stash))
6594 mro_method_changed_in(stash);
6595 gp_free(MUTABLE_GV(sv));
6597 unshare_hek(GvNAME_HEK(sv));
6598 /* If we're in a stash, we don't own a reference to it.
6599 * However it does have a back reference to us, which
6600 * needs to be cleared. */
6601 if (!SvVALID(sv) && (stash = GvSTASH(sv)))
6602 sv_del_backref(MUTABLE_SV(stash), sv);
6604 /* FIXME. There are probably more unreferenced pointers to SVs
6605 * in the interpreter struct that we should check and tidy in
6606 * a similar fashion to this: */
6607 /* See also S_sv_unglob, which does the same thing. */
6608 if ((const GV *)sv == PL_last_in_gv)
6609 PL_last_in_gv = NULL;
6610 else if ((const GV *)sv == PL_statgv)
6612 else if ((const GV *)sv == PL_stderrgv)
6620 /* Don't bother with SvOOK_off(sv); as we're only going to
6624 SvOOK_offset(sv, offset);
6625 SvPV_set(sv, SvPVX_mutable(sv) - offset);
6626 /* Don't even bother with turning off the OOK flag. */
6631 SV * const target = SvRV(sv);
6633 sv_del_backref(target, sv);
6639 else if (SvPVX_const(sv)
6640 && !(SvTYPE(sv) == SVt_PVIO
6641 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6645 PerlIO_printf(Perl_debug_log, "Copy on write: clear\n");
6649 # ifdef PERL_OLD_COPY_ON_WRITE
6650 sv_release_COW(sv, SvPVX_const(sv), SV_COW_NEXT_SV(sv));
6652 if (CowREFCNT(sv)) {
6660 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6664 # ifdef PERL_OLD_COPY_ON_WRITE
6668 Safefree(SvPVX_mutable(sv));
6672 else if (SvPVX_const(sv) && SvLEN(sv)
6673 && !(SvTYPE(sv) == SVt_PVIO
6674 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6675 Safefree(SvPVX_mutable(sv));
6676 else if (SvPVX_const(sv) && SvIsCOW(sv)) {
6677 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6687 SvFLAGS(sv) &= SVf_BREAK;
6688 SvFLAGS(sv) |= SVTYPEMASK;
6690 sv_type_details = bodies_by_type + type;
6691 if (sv_type_details->arena) {
6692 del_body(((char *)SvANY(sv) + sv_type_details->offset),
6693 &PL_body_roots[type]);
6695 else if (sv_type_details->body_size) {
6696 safefree(SvANY(sv));
6700 /* caller is responsible for freeing the head of the original sv */
6701 if (sv != orig_sv && !SvREFCNT(sv))
6704 /* grab and free next sv, if any */
6712 else if (!iter_sv) {
6714 } else if (SvTYPE(iter_sv) == SVt_PVAV) {
6715 AV *const av = (AV*)iter_sv;
6716 if (AvFILLp(av) > -1) {
6717 sv = AvARRAY(av)[AvFILLp(av)--];
6719 else { /* no more elements of current AV to free */
6722 /* restore previous value, squirrelled away */
6723 iter_sv = AvARRAY(av)[AvMAX(av)];
6724 Safefree(AvALLOC(av));
6727 } else if (SvTYPE(iter_sv) == SVt_PVHV) {
6728 sv = Perl_hfree_next_entry(aTHX_ (HV*)iter_sv, &hash_index);
6729 if (!sv && !HvTOTALKEYS((HV *)iter_sv)) {
6730 /* no more elements of current HV to free */
6733 /* Restore previous values of iter_sv and hash_index,
6734 * squirrelled away */
6735 assert(!SvOBJECT(sv));
6736 iter_sv = (SV*)SvSTASH(sv);
6737 assert(!SvMAGICAL(sv));
6738 hash_index = ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index;
6740 /* perl -DA does not like rubbish in SvMAGIC. */
6744 /* free any remaining detritus from the hash struct */
6745 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6746 assert(!HvARRAY((HV*)sv));
6751 /* unrolled SvREFCNT_dec and sv_free2 follows: */
6755 if (!SvREFCNT(sv)) {
6759 if (--(SvREFCNT(sv)))
6763 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6764 "Attempt to free temp prematurely: SV 0x%"UVxf
6765 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6769 if (SvIMMORTAL(sv)) {
6770 /* make sure SvREFCNT(sv)==0 happens very seldom */
6771 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6780 /* This routine curses the sv itself, not the object referenced by sv. So
6781 sv does not have to be ROK. */
6784 S_curse(pTHX_ SV * const sv, const bool check_refcnt) {
6785 PERL_ARGS_ASSERT_CURSE;
6786 assert(SvOBJECT(sv));
6788 if (PL_defstash && /* Still have a symbol table? */
6794 stash = SvSTASH(sv);
6795 assert(SvTYPE(stash) == SVt_PVHV);
6796 if (HvNAME(stash)) {
6797 CV* destructor = NULL;
6798 assert (SvOOK(stash));
6799 if (!SvOBJECT(stash)) destructor = (CV *)SvSTASH(stash);
6800 if (!destructor || HvMROMETA(stash)->destroy_gen
6801 != PL_sub_generation)
6804 gv_fetchmeth_autoload(stash, "DESTROY", 7, 0);
6805 if (gv) destructor = GvCV(gv);
6806 if (!SvOBJECT(stash))
6809 destructor ? (HV *)destructor : ((HV *)0)+1;
6810 HvAUX(stash)->xhv_mro_meta->destroy_gen =
6814 assert(!destructor || destructor == ((CV *)0)+1
6815 || SvTYPE(destructor) == SVt_PVCV);
6816 if (destructor && destructor != ((CV *)0)+1
6817 /* A constant subroutine can have no side effects, so
6818 don't bother calling it. */
6819 && !CvCONST(destructor)
6820 /* Don't bother calling an empty destructor or one that
6821 returns immediately. */
6822 && (CvISXSUB(destructor)
6823 || (CvSTART(destructor)
6824 && (CvSTART(destructor)->op_next->op_type
6826 && (CvSTART(destructor)->op_next->op_type
6828 || CvSTART(destructor)->op_next->op_next->op_type
6834 SV* const tmpref = newRV(sv);
6835 SvREADONLY_on(tmpref); /* DESTROY() could be naughty */
6837 PUSHSTACKi(PERLSI_DESTROY);
6842 call_sv(MUTABLE_SV(destructor),
6843 G_DISCARD|G_EVAL|G_KEEPERR|G_VOID);
6847 if(SvREFCNT(tmpref) < 2) {
6848 /* tmpref is not kept alive! */
6850 SvRV_set(tmpref, NULL);
6853 SvREFCNT_dec_NN(tmpref);
6856 } while (SvOBJECT(sv) && SvSTASH(sv) != stash);
6859 if (check_refcnt && SvREFCNT(sv)) {
6860 if (PL_in_clean_objs)
6862 "DESTROY created new reference to dead object '%"HEKf"'",
6863 HEKfARG(HvNAME_HEK(stash)));
6864 /* DESTROY gave object new lease on life */
6870 HV * const stash = SvSTASH(sv);
6871 /* Curse before freeing the stash, as freeing the stash could cause
6872 a recursive call into S_curse. */
6873 SvOBJECT_off(sv); /* Curse the object. */
6874 SvSTASH_set(sv,0); /* SvREFCNT_dec may try to read this */
6875 SvREFCNT_dec(stash); /* possibly of changed persuasion */
6881 =for apidoc sv_newref
6883 Increment an SV's reference count. Use the C<SvREFCNT_inc()> wrapper
6890 Perl_sv_newref(pTHX_ SV *const sv)
6892 PERL_UNUSED_CONTEXT;
6901 Decrement an SV's reference count, and if it drops to zero, call
6902 C<sv_clear> to invoke destructors and free up any memory used by
6903 the body; finally, deallocate the SV's head itself.
6904 Normally called via a wrapper macro C<SvREFCNT_dec>.
6910 Perl_sv_free(pTHX_ SV *const sv)
6916 /* Private helper function for SvREFCNT_dec().
6917 * Called with rc set to original SvREFCNT(sv), where rc == 0 or 1 */
6920 Perl_sv_free2(pTHX_ SV *const sv, const U32 rc)
6924 PERL_ARGS_ASSERT_SV_FREE2;
6926 if (LIKELY( rc == 1 )) {
6932 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6933 "Attempt to free temp prematurely: SV 0x%"UVxf
6934 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6938 if (SvIMMORTAL(sv)) {
6939 /* make sure SvREFCNT(sv)==0 happens very seldom */
6940 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6944 if (! SvREFCNT(sv)) /* may have have been resurrected */
6949 /* handle exceptional cases */
6953 if (SvFLAGS(sv) & SVf_BREAK)
6954 /* this SV's refcnt has been artificially decremented to
6955 * trigger cleanup */
6957 if (PL_in_clean_all) /* All is fair */
6959 if (SvIMMORTAL(sv)) {
6960 /* make sure SvREFCNT(sv)==0 happens very seldom */
6961 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6964 if (ckWARN_d(WARN_INTERNAL)) {
6965 #ifdef DEBUG_LEAKING_SCALARS_FORK_DUMP
6966 Perl_dump_sv_child(aTHX_ sv);
6968 #ifdef DEBUG_LEAKING_SCALARS
6971 #ifdef DEBUG_LEAKING_SCALARS_ABORT
6972 if (PL_warnhook == PERL_WARNHOOK_FATAL
6973 || ckDEAD(packWARN(WARN_INTERNAL))) {
6974 /* Don't let Perl_warner cause us to escape our fate: */
6978 /* This may not return: */
6979 Perl_warner(aTHX_ packWARN(WARN_INTERNAL),
6980 "Attempt to free unreferenced scalar: SV 0x%"UVxf
6981 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6984 #ifdef DEBUG_LEAKING_SCALARS_ABORT
6994 Returns the length of the string in the SV. Handles magic and type
6995 coercion and sets the UTF8 flag appropriately. See also C<SvCUR>, which
6996 gives raw access to the xpv_cur slot.
7002 Perl_sv_len(pTHX_ SV *const sv)
7009 (void)SvPV_const(sv, len);
7014 =for apidoc sv_len_utf8
7016 Returns the number of characters in the string in an SV, counting wide
7017 UTF-8 bytes as a single character. Handles magic and type coercion.
7023 * The length is cached in PERL_MAGIC_utf8, in the mg_len field. Also the
7024 * mg_ptr is used, by sv_pos_u2b() and sv_pos_b2u() - see the comments below.
7025 * (Note that the mg_len is not the length of the mg_ptr field.
7026 * This allows the cache to store the character length of the string without
7027 * needing to malloc() extra storage to attach to the mg_ptr.)
7032 Perl_sv_len_utf8(pTHX_ SV *const sv)
7038 return sv_len_utf8_nomg(sv);
7042 Perl_sv_len_utf8_nomg(pTHX_ SV * const sv)
7045 const U8 *s = (U8*)SvPV_nomg_const(sv, len);
7047 PERL_ARGS_ASSERT_SV_LEN_UTF8_NOMG;
7049 if (PL_utf8cache && SvUTF8(sv)) {
7051 MAGIC *mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_utf8) : NULL;
7053 if (mg && (mg->mg_len != -1 || mg->mg_ptr)) {
7054 if (mg->mg_len != -1)
7057 /* We can use the offset cache for a headstart.
7058 The longer value is stored in the first pair. */
7059 STRLEN *cache = (STRLEN *) mg->mg_ptr;
7061 ulen = cache[0] + Perl_utf8_length(aTHX_ s + cache[1],
7065 if (PL_utf8cache < 0) {
7066 const STRLEN real = Perl_utf8_length(aTHX_ s, s + len);
7067 assert_uft8_cache_coherent("sv_len_utf8", ulen, real, sv);
7071 ulen = Perl_utf8_length(aTHX_ s, s + len);
7072 utf8_mg_len_cache_update(sv, &mg, ulen);
7076 return SvUTF8(sv) ? Perl_utf8_length(aTHX_ s, s + len) : len;
7079 /* Walk forwards to find the byte corresponding to the passed in UTF-8
7082 S_sv_pos_u2b_forwards(const U8 *const start, const U8 *const send,
7083 STRLEN *const uoffset_p, bool *const at_end)
7085 const U8 *s = start;
7086 STRLEN uoffset = *uoffset_p;
7088 PERL_ARGS_ASSERT_SV_POS_U2B_FORWARDS;
7090 while (s < send && uoffset) {
7097 else if (s > send) {
7099 /* This is the existing behaviour. Possibly it should be a croak, as
7100 it's actually a bounds error */
7103 *uoffset_p -= uoffset;
7107 /* Given the length of the string in both bytes and UTF-8 characters, decide
7108 whether to walk forwards or backwards to find the byte corresponding to
7109 the passed in UTF-8 offset. */
7111 S_sv_pos_u2b_midway(const U8 *const start, const U8 *send,
7112 STRLEN uoffset, const STRLEN uend)
7114 STRLEN backw = uend - uoffset;
7116 PERL_ARGS_ASSERT_SV_POS_U2B_MIDWAY;
7118 if (uoffset < 2 * backw) {
7119 /* The assumption is that going forwards is twice the speed of going
7120 forward (that's where the 2 * backw comes from).
7121 (The real figure of course depends on the UTF-8 data.) */
7122 const U8 *s = start;
7124 while (s < send && uoffset--)
7134 while (UTF8_IS_CONTINUATION(*send))
7137 return send - start;
7140 /* For the string representation of the given scalar, find the byte
7141 corresponding to the passed in UTF-8 offset. uoffset0 and boffset0
7142 give another position in the string, *before* the sought offset, which
7143 (which is always true, as 0, 0 is a valid pair of positions), which should
7144 help reduce the amount of linear searching.
7145 If *mgp is non-NULL, it should point to the UTF-8 cache magic, which
7146 will be used to reduce the amount of linear searching. The cache will be
7147 created if necessary, and the found value offered to it for update. */
7149 S_sv_pos_u2b_cached(pTHX_ SV *const sv, MAGIC **const mgp, const U8 *const start,
7150 const U8 *const send, STRLEN uoffset,
7151 STRLEN uoffset0, STRLEN boffset0)
7153 STRLEN boffset = 0; /* Actually always set, but let's keep gcc happy. */
7155 bool at_end = FALSE;
7157 PERL_ARGS_ASSERT_SV_POS_U2B_CACHED;
7159 assert (uoffset >= uoffset0);
7164 if (!SvREADONLY(sv) && !SvGMAGICAL(sv) && SvPOK(sv)
7166 && (*mgp || (SvTYPE(sv) >= SVt_PVMG &&
7167 (*mgp = mg_find(sv, PERL_MAGIC_utf8))))) {
7168 if ((*mgp)->mg_ptr) {
7169 STRLEN *cache = (STRLEN *) (*mgp)->mg_ptr;
7170 if (cache[0] == uoffset) {
7171 /* An exact match. */
7174 if (cache[2] == uoffset) {
7175 /* An exact match. */
7179 if (cache[0] < uoffset) {
7180 /* The cache already knows part of the way. */
7181 if (cache[0] > uoffset0) {
7182 /* The cache knows more than the passed in pair */
7183 uoffset0 = cache[0];
7184 boffset0 = cache[1];
7186 if ((*mgp)->mg_len != -1) {
7187 /* And we know the end too. */
7189 + sv_pos_u2b_midway(start + boffset0, send,
7191 (*mgp)->mg_len - uoffset0);
7193 uoffset -= uoffset0;
7195 + sv_pos_u2b_forwards(start + boffset0,
7196 send, &uoffset, &at_end);
7197 uoffset += uoffset0;
7200 else if (cache[2] < uoffset) {
7201 /* We're between the two cache entries. */
7202 if (cache[2] > uoffset0) {
7203 /* and the cache knows more than the passed in pair */
7204 uoffset0 = cache[2];
7205 boffset0 = cache[3];
7209 + sv_pos_u2b_midway(start + boffset0,
7212 cache[0] - uoffset0);
7215 + sv_pos_u2b_midway(start + boffset0,
7218 cache[2] - uoffset0);
7222 else if ((*mgp)->mg_len != -1) {
7223 /* If we can take advantage of a passed in offset, do so. */
7224 /* In fact, offset0 is either 0, or less than offset, so don't
7225 need to worry about the other possibility. */
7227 + sv_pos_u2b_midway(start + boffset0, send,
7229 (*mgp)->mg_len - uoffset0);
7234 if (!found || PL_utf8cache < 0) {
7235 STRLEN real_boffset;
7236 uoffset -= uoffset0;
7237 real_boffset = boffset0 + sv_pos_u2b_forwards(start + boffset0,
7238 send, &uoffset, &at_end);
7239 uoffset += uoffset0;
7241 if (found && PL_utf8cache < 0)
7242 assert_uft8_cache_coherent("sv_pos_u2b_cache", boffset,
7244 boffset = real_boffset;
7247 if (PL_utf8cache && !SvGMAGICAL(sv) && SvPOK(sv)) {
7249 utf8_mg_len_cache_update(sv, mgp, uoffset);
7251 utf8_mg_pos_cache_update(sv, mgp, boffset, uoffset, send - start);
7258 =for apidoc sv_pos_u2b_flags
7260 Converts the offset from a count of UTF-8 chars from
7261 the start of the string, to a count of the equivalent number of bytes; if
7262 lenp is non-zero, it does the same to lenp, but this time starting from
7263 the offset, rather than from the start
7264 of the string. Handles type coercion.
7265 I<flags> is passed to C<SvPV_flags>, and usually should be
7266 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7272 * sv_pos_u2b_flags() 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().
7279 Perl_sv_pos_u2b_flags(pTHX_ SV *const sv, STRLEN uoffset, STRLEN *const lenp,
7286 PERL_ARGS_ASSERT_SV_POS_U2B_FLAGS;
7288 start = (U8*)SvPV_flags(sv, len, flags);
7290 const U8 * const send = start + len;
7292 boffset = sv_pos_u2b_cached(sv, &mg, start, send, uoffset, 0, 0);
7295 && *lenp /* don't bother doing work for 0, as its bytes equivalent
7296 is 0, and *lenp is already set to that. */) {
7297 /* Convert the relative offset to absolute. */
7298 const STRLEN uoffset2 = uoffset + *lenp;
7299 const STRLEN boffset2
7300 = sv_pos_u2b_cached(sv, &mg, start, send, uoffset2,
7301 uoffset, boffset) - boffset;
7315 =for apidoc sv_pos_u2b
7317 Converts the value pointed to by offsetp from a count of UTF-8 chars from
7318 the start of the string, to a count of the equivalent number of bytes; if
7319 lenp is non-zero, it does the same to lenp, but this time starting from
7320 the offset, rather than from the start of the string. Handles magic and
7323 Use C<sv_pos_u2b_flags> in preference, which correctly handles strings longer
7330 * sv_pos_u2b() uses, like sv_pos_b2u(), the mg_ptr of the potential
7331 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7332 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7336 /* This function is subject to size and sign problems */
7339 Perl_sv_pos_u2b(pTHX_ SV *const sv, I32 *const offsetp, I32 *const lenp)
7341 PERL_ARGS_ASSERT_SV_POS_U2B;
7344 STRLEN ulen = (STRLEN)*lenp;
7345 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, &ulen,
7346 SV_GMAGIC|SV_CONST_RETURN);
7349 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, NULL,
7350 SV_GMAGIC|SV_CONST_RETURN);
7355 S_utf8_mg_len_cache_update(pTHX_ SV *const sv, MAGIC **const mgp,
7358 PERL_ARGS_ASSERT_UTF8_MG_LEN_CACHE_UPDATE;
7359 if (SvREADONLY(sv) || SvGMAGICAL(sv) || !SvPOK(sv))
7362 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7363 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7364 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, &PL_vtbl_utf8, 0, 0);
7368 (*mgp)->mg_len = ulen;
7371 /* Create and update the UTF8 magic offset cache, with the proffered utf8/
7372 byte length pairing. The (byte) length of the total SV is passed in too,
7373 as blen, because for some (more esoteric) SVs, the call to SvPV_const()
7374 may not have updated SvCUR, so we can't rely on reading it directly.
7376 The proffered utf8/byte length pairing isn't used if the cache already has
7377 two pairs, and swapping either for the proffered pair would increase the
7378 RMS of the intervals between known byte offsets.
7380 The cache itself consists of 4 STRLEN values
7381 0: larger UTF-8 offset
7382 1: corresponding byte offset
7383 2: smaller UTF-8 offset
7384 3: corresponding byte offset
7386 Unused cache pairs have the value 0, 0.
7387 Keeping the cache "backwards" means that the invariant of
7388 cache[0] >= cache[2] is maintained even with empty slots, which means that
7389 the code that uses it doesn't need to worry if only 1 entry has actually
7390 been set to non-zero. It also makes the "position beyond the end of the
7391 cache" logic much simpler, as the first slot is always the one to start
7395 S_utf8_mg_pos_cache_update(pTHX_ SV *const sv, MAGIC **const mgp, const STRLEN byte,
7396 const STRLEN utf8, const STRLEN blen)
7400 PERL_ARGS_ASSERT_UTF8_MG_POS_CACHE_UPDATE;
7405 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7406 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7407 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, (MGVTBL*)&PL_vtbl_utf8, 0,
7409 (*mgp)->mg_len = -1;
7413 if (!(cache = (STRLEN *)(*mgp)->mg_ptr)) {
7414 Newxz(cache, PERL_MAGIC_UTF8_CACHESIZE * 2, STRLEN);
7415 (*mgp)->mg_ptr = (char *) cache;
7419 if (PL_utf8cache < 0 && SvPOKp(sv)) {
7420 /* SvPOKp() because, if sv is a reference, then SvPVX() is actually
7421 a pointer. Note that we no longer cache utf8 offsets on refer-
7422 ences, but this check is still a good idea, for robustness. */
7423 const U8 *start = (const U8 *) SvPVX_const(sv);
7424 const STRLEN realutf8 = utf8_length(start, start + byte);
7426 assert_uft8_cache_coherent("utf8_mg_pos_cache_update", utf8, realutf8,
7430 /* Cache is held with the later position first, to simplify the code
7431 that deals with unbounded ends. */
7433 ASSERT_UTF8_CACHE(cache);
7434 if (cache[1] == 0) {
7435 /* Cache is totally empty */
7438 } else if (cache[3] == 0) {
7439 if (byte > cache[1]) {
7440 /* New one is larger, so goes first. */
7441 cache[2] = cache[0];
7442 cache[3] = cache[1];
7450 /* float casts necessary? XXX */
7451 #define THREEWAY_SQUARE(a,b,c,d) \
7452 ((float)((d) - (c))) * ((float)((d) - (c))) \
7453 + ((float)((c) - (b))) * ((float)((c) - (b))) \
7454 + ((float)((b) - (a))) * ((float)((b) - (a)))
7456 /* Cache has 2 slots in use, and we know three potential pairs.
7457 Keep the two that give the lowest RMS distance. Do the
7458 calculation in bytes simply because we always know the byte
7459 length. squareroot has the same ordering as the positive value,
7460 so don't bother with the actual square root. */
7461 if (byte > cache[1]) {
7462 /* New position is after the existing pair of pairs. */
7463 const float keep_earlier
7464 = THREEWAY_SQUARE(0, cache[3], byte, blen);
7465 const float keep_later
7466 = THREEWAY_SQUARE(0, cache[1], byte, blen);
7468 if (keep_later < keep_earlier) {
7469 cache[2] = cache[0];
7470 cache[3] = cache[1];
7476 const float keep_later = THREEWAY_SQUARE(0, byte, cache[1], blen);
7477 float b, c, keep_earlier;
7478 if (byte > cache[3]) {
7479 /* New position is between the existing pair of pairs. */
7480 b = (float)cache[3];
7483 /* New position is before the existing pair of pairs. */
7485 c = (float)cache[3];
7487 keep_earlier = THREEWAY_SQUARE(0, b, c, blen);
7488 if (byte > cache[3]) {
7489 if (keep_later < keep_earlier) {
7499 if (! (keep_later < keep_earlier)) {
7500 cache[0] = cache[2];
7501 cache[1] = cache[3];
7508 ASSERT_UTF8_CACHE(cache);
7511 /* We already know all of the way, now we may be able to walk back. The same
7512 assumption is made as in S_sv_pos_u2b_midway(), namely that walking
7513 backward is half the speed of walking forward. */
7515 S_sv_pos_b2u_midway(pTHX_ const U8 *const s, const U8 *const target,
7516 const U8 *end, STRLEN endu)
7518 const STRLEN forw = target - s;
7519 STRLEN backw = end - target;
7521 PERL_ARGS_ASSERT_SV_POS_B2U_MIDWAY;
7523 if (forw < 2 * backw) {
7524 return utf8_length(s, target);
7527 while (end > target) {
7529 while (UTF8_IS_CONTINUATION(*end)) {
7538 =for apidoc sv_pos_b2u_flags
7540 Converts the offset from a count of bytes from the start of the string, to
7541 a count of the equivalent number of UTF-8 chars. Handles type coercion.
7542 I<flags> is passed to C<SvPV_flags>, and usually should be
7543 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7549 * sv_pos_b2u_flags() uses, like sv_pos_u2b_flags(), the mg_ptr of the
7550 * potential PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8
7555 Perl_sv_pos_b2u_flags(pTHX_ SV *const sv, STRLEN const offset, U32 flags)
7558 STRLEN len = 0; /* Actually always set, but let's keep gcc happy. */
7564 PERL_ARGS_ASSERT_SV_POS_B2U_FLAGS;
7566 s = (const U8*)SvPV_flags(sv, blen, flags);
7569 Perl_croak(aTHX_ "panic: sv_pos_b2u: bad byte offset, blen=%"UVuf
7570 ", byte=%"UVuf, (UV)blen, (UV)offset);
7576 && SvTYPE(sv) >= SVt_PVMG
7577 && (mg = mg_find(sv, PERL_MAGIC_utf8)))
7580 STRLEN * const cache = (STRLEN *) mg->mg_ptr;
7581 if (cache[1] == offset) {
7582 /* An exact match. */
7585 if (cache[3] == offset) {
7586 /* An exact match. */
7590 if (cache[1] < offset) {
7591 /* We already know part of the way. */
7592 if (mg->mg_len != -1) {
7593 /* Actually, we know the end too. */
7595 + S_sv_pos_b2u_midway(aTHX_ s + cache[1], send,
7596 s + blen, mg->mg_len - cache[0]);
7598 len = cache[0] + utf8_length(s + cache[1], send);
7601 else if (cache[3] < offset) {
7602 /* We're between the two cached pairs, so we do the calculation
7603 offset by the byte/utf-8 positions for the earlier pair,
7604 then add the utf-8 characters from the string start to
7606 len = S_sv_pos_b2u_midway(aTHX_ s + cache[3], send,
7607 s + cache[1], cache[0] - cache[2])
7611 else { /* cache[3] > offset */
7612 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + cache[3],
7616 ASSERT_UTF8_CACHE(cache);
7618 } else if (mg->mg_len != -1) {
7619 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + blen, mg->mg_len);
7623 if (!found || PL_utf8cache < 0) {
7624 const STRLEN real_len = utf8_length(s, send);
7626 if (found && PL_utf8cache < 0)
7627 assert_uft8_cache_coherent("sv_pos_b2u", len, real_len, sv);
7633 utf8_mg_len_cache_update(sv, &mg, len);
7635 utf8_mg_pos_cache_update(sv, &mg, offset, len, blen);
7642 =for apidoc sv_pos_b2u
7644 Converts the value pointed to by offsetp from a count of bytes from the
7645 start of the string, to a count of the equivalent number of UTF-8 chars.
7646 Handles magic and type coercion.
7648 Use C<sv_pos_b2u_flags> in preference, which correctly handles strings
7655 * sv_pos_b2u() uses, like sv_pos_u2b(), the mg_ptr of the potential
7656 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7661 Perl_sv_pos_b2u(pTHX_ SV *const sv, I32 *const offsetp)
7663 PERL_ARGS_ASSERT_SV_POS_B2U;
7668 *offsetp = (I32)sv_pos_b2u_flags(sv, (STRLEN)*offsetp,
7669 SV_GMAGIC|SV_CONST_RETURN);
7673 S_assert_uft8_cache_coherent(pTHX_ const char *const func, STRLEN from_cache,
7674 STRLEN real, SV *const sv)
7676 PERL_ARGS_ASSERT_ASSERT_UFT8_CACHE_COHERENT;
7678 /* As this is debugging only code, save space by keeping this test here,
7679 rather than inlining it in all the callers. */
7680 if (from_cache == real)
7683 /* Need to turn the assertions off otherwise we may recurse infinitely
7684 while printing error messages. */
7685 SAVEI8(PL_utf8cache);
7687 Perl_croak(aTHX_ "panic: %s cache %"UVuf" real %"UVuf" for %"SVf,
7688 func, (UV) from_cache, (UV) real, SVfARG(sv));
7694 Returns a boolean indicating whether the strings in the two SVs are
7695 identical. Is UTF-8 and 'use bytes' aware, handles get magic, and will
7696 coerce its args to strings if necessary.
7698 =for apidoc sv_eq_flags
7700 Returns a boolean indicating whether the strings in the two SVs are
7701 identical. Is UTF-8 and 'use bytes' aware and coerces its args to strings
7702 if necessary. If the flags include SV_GMAGIC, it handles get-magic, too.
7708 Perl_sv_eq_flags(pTHX_ SV *sv1, SV *sv2, const U32 flags)
7715 SV* svrecode = NULL;
7722 /* if pv1 and pv2 are the same, second SvPV_const call may
7723 * invalidate pv1 (if we are handling magic), so we may need to
7725 if (sv1 == sv2 && flags & SV_GMAGIC
7726 && (SvTHINKFIRST(sv1) || SvGMAGICAL(sv1))) {
7727 pv1 = SvPV_const(sv1, cur1);
7728 sv1 = newSVpvn_flags(pv1, cur1, SVs_TEMP | SvUTF8(sv2));
7730 pv1 = SvPV_flags_const(sv1, cur1, flags);
7738 pv2 = SvPV_flags_const(sv2, cur2, flags);
7740 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7741 /* Differing utf8ness.
7742 * Do not UTF8size the comparands as a side-effect. */
7745 svrecode = newSVpvn(pv2, cur2);
7746 sv_recode_to_utf8(svrecode, PL_encoding);
7747 pv2 = SvPV_const(svrecode, cur2);
7750 svrecode = newSVpvn(pv1, cur1);
7751 sv_recode_to_utf8(svrecode, PL_encoding);
7752 pv1 = SvPV_const(svrecode, cur1);
7754 /* Now both are in UTF-8. */
7756 SvREFCNT_dec_NN(svrecode);
7762 /* sv1 is the UTF-8 one */
7763 return bytes_cmp_utf8((const U8*)pv2, cur2,
7764 (const U8*)pv1, cur1) == 0;
7767 /* sv2 is the UTF-8 one */
7768 return bytes_cmp_utf8((const U8*)pv1, cur1,
7769 (const U8*)pv2, cur2) == 0;
7775 eq = (pv1 == pv2) || memEQ(pv1, pv2, cur1);
7777 SvREFCNT_dec(svrecode);
7785 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7786 string in C<sv1> is less than, equal to, or greater than the string in
7787 C<sv2>. Is UTF-8 and 'use bytes' aware, handles get magic, and will
7788 coerce its args to strings if necessary. See also C<sv_cmp_locale>.
7790 =for apidoc sv_cmp_flags
7792 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7793 string in C<sv1> is less than, equal to, or greater than the string in
7794 C<sv2>. Is UTF-8 and 'use bytes' aware and will coerce its args to strings
7795 if necessary. If the flags include SV_GMAGIC, it handles get magic. See
7796 also C<sv_cmp_locale_flags>.
7802 Perl_sv_cmp(pTHX_ SV *const sv1, SV *const sv2)
7804 return sv_cmp_flags(sv1, sv2, SV_GMAGIC);
7808 Perl_sv_cmp_flags(pTHX_ SV *const sv1, SV *const sv2,
7812 const char *pv1, *pv2;
7814 SV *svrecode = NULL;
7821 pv1 = SvPV_flags_const(sv1, cur1, flags);
7828 pv2 = SvPV_flags_const(sv2, cur2, flags);
7830 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7831 /* Differing utf8ness.
7832 * Do not UTF8size the comparands as a side-effect. */
7835 svrecode = newSVpvn(pv2, cur2);
7836 sv_recode_to_utf8(svrecode, PL_encoding);
7837 pv2 = SvPV_const(svrecode, cur2);
7840 const int retval = -bytes_cmp_utf8((const U8*)pv2, cur2,
7841 (const U8*)pv1, cur1);
7842 return retval ? retval < 0 ? -1 : +1 : 0;
7847 svrecode = newSVpvn(pv1, cur1);
7848 sv_recode_to_utf8(svrecode, PL_encoding);
7849 pv1 = SvPV_const(svrecode, cur1);
7852 const int retval = bytes_cmp_utf8((const U8*)pv1, cur1,
7853 (const U8*)pv2, cur2);
7854 return retval ? retval < 0 ? -1 : +1 : 0;
7860 cmp = cur2 ? -1 : 0;
7864 const I32 retval = memcmp((const void*)pv1, (const void*)pv2, cur1 < cur2 ? cur1 : cur2);
7867 cmp = retval < 0 ? -1 : 1;
7868 } else if (cur1 == cur2) {
7871 cmp = cur1 < cur2 ? -1 : 1;
7875 SvREFCNT_dec(svrecode);
7881 =for apidoc sv_cmp_locale
7883 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
7884 'use bytes' aware, handles get magic, and will coerce its args to strings
7885 if necessary. See also C<sv_cmp>.
7887 =for apidoc sv_cmp_locale_flags
7889 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
7890 'use bytes' aware and will coerce its args to strings if necessary. If the
7891 flags contain SV_GMAGIC, it handles get magic. See also C<sv_cmp_flags>.
7897 Perl_sv_cmp_locale(pTHX_ SV *const sv1, SV *const sv2)
7899 return sv_cmp_locale_flags(sv1, sv2, SV_GMAGIC);
7903 Perl_sv_cmp_locale_flags(pTHX_ SV *const sv1, SV *const sv2,
7906 #ifdef USE_LOCALE_COLLATE
7912 if (PL_collation_standard)
7916 pv1 = sv1 ? sv_collxfrm_flags(sv1, &len1, flags) : (char *) NULL;
7918 pv2 = sv2 ? sv_collxfrm_flags(sv2, &len2, flags) : (char *) NULL;
7920 if (!pv1 || !len1) {
7931 retval = memcmp((void*)pv1, (void*)pv2, len1 < len2 ? len1 : len2);
7934 return retval < 0 ? -1 : 1;
7937 * When the result of collation is equality, that doesn't mean
7938 * that there are no differences -- some locales exclude some
7939 * characters from consideration. So to avoid false equalities,
7940 * we use the raw string as a tiebreaker.
7947 PERL_UNUSED_ARG(flags);
7948 #endif /* USE_LOCALE_COLLATE */
7950 return sv_cmp(sv1, sv2);
7954 #ifdef USE_LOCALE_COLLATE
7957 =for apidoc sv_collxfrm
7959 This calls C<sv_collxfrm_flags> with the SV_GMAGIC flag. See
7960 C<sv_collxfrm_flags>.
7962 =for apidoc sv_collxfrm_flags
7964 Add Collate Transform magic to an SV if it doesn't already have it. If the
7965 flags contain SV_GMAGIC, it handles get-magic.
7967 Any scalar variable may carry PERL_MAGIC_collxfrm magic that contains the
7968 scalar data of the variable, but transformed to such a format that a normal
7969 memory comparison can be used to compare the data according to the locale
7976 Perl_sv_collxfrm_flags(pTHX_ SV *const sv, STRLEN *const nxp, const I32 flags)
7980 PERL_ARGS_ASSERT_SV_COLLXFRM_FLAGS;
7982 mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_collxfrm) : (MAGIC *) NULL;
7983 if (!mg || !mg->mg_ptr || *(U32*)mg->mg_ptr != PL_collation_ix) {
7989 Safefree(mg->mg_ptr);
7990 s = SvPV_flags_const(sv, len, flags);
7991 if ((xf = mem_collxfrm(s, len, &xlen))) {
7993 #ifdef PERL_OLD_COPY_ON_WRITE
7995 sv_force_normal_flags(sv, 0);
7997 mg = sv_magicext(sv, 0, PERL_MAGIC_collxfrm, &PL_vtbl_collxfrm,
8011 if (mg && mg->mg_ptr) {
8013 return mg->mg_ptr + sizeof(PL_collation_ix);
8021 #endif /* USE_LOCALE_COLLATE */
8024 S_sv_gets_append_to_utf8(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8026 SV * const tsv = newSV(0);
8029 sv_gets(tsv, fp, 0);
8030 sv_utf8_upgrade_nomg(tsv);
8031 SvCUR_set(sv,append);
8034 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8038 S_sv_gets_read_record(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8041 const STRLEN recsize = SvUV(SvRV(PL_rs)); /* RsRECORD() guarantees > 0. */
8042 /* Grab the size of the record we're getting */
8043 char *buffer = SvGROW(sv, (STRLEN)(recsize + append + 1)) + append;
8050 /* With a true, record-oriented file on VMS, we need to use read directly
8051 * to ensure that we respect RMS record boundaries. The user is responsible
8052 * for providing a PL_rs value that corresponds to the FAB$W_MRS (maximum
8053 * record size) field. N.B. This is likely to produce invalid results on
8054 * varying-width character data when a record ends mid-character.
8056 fd = PerlIO_fileno(fp);
8058 && PerlLIO_fstat(fd, &st) == 0
8059 && (st.st_fab_rfm == FAB$C_VAR
8060 || st.st_fab_rfm == FAB$C_VFC
8061 || st.st_fab_rfm == FAB$C_FIX)) {
8063 bytesread = PerlLIO_read(fd, buffer, recsize);
8065 else /* in-memory file from PerlIO::Scalar
8066 * or not a record-oriented file
8070 bytesread = PerlIO_read(fp, buffer, recsize);
8072 /* At this point, the logic in sv_get() means that sv will
8073 be treated as utf-8 if the handle is utf8.
8075 if (PerlIO_isutf8(fp) && bytesread > 0) {
8076 char *bend = buffer + bytesread;
8077 char *bufp = buffer;
8078 size_t charcount = 0;
8079 bool charstart = TRUE;
8082 while (charcount < recsize) {
8083 /* count accumulated characters */
8084 while (bufp < bend) {
8086 skip = UTF8SKIP(bufp);
8088 if (bufp + skip > bend) {
8089 /* partial at the end */
8100 if (charcount < recsize) {
8102 STRLEN bufp_offset = bufp - buffer;
8103 SSize_t morebytesread;
8105 /* originally I read enough to fill any incomplete
8106 character and the first byte of the next
8107 character if needed, but if there's many
8108 multi-byte encoded characters we're going to be
8109 making a read call for every character beyond
8110 the original read size.
8112 So instead, read the rest of the character if
8113 any, and enough bytes to match at least the
8114 start bytes for each character we're going to
8118 readsize = recsize - charcount;
8120 readsize = skip - (bend - bufp) + recsize - charcount - 1;
8121 buffer = SvGROW(sv, append + bytesread + readsize + 1) + append;
8122 bend = buffer + bytesread;
8123 morebytesread = PerlIO_read(fp, bend, readsize);
8124 if (morebytesread <= 0) {
8125 /* we're done, if we still have incomplete
8126 characters the check code in sv_gets() will
8129 I'd originally considered doing
8130 PerlIO_ungetc() on all but the lead
8131 character of the incomplete character, but
8132 read() doesn't do that, so I don't.
8137 /* prepare to scan some more */
8138 bytesread += morebytesread;
8139 bend = buffer + bytesread;
8140 bufp = buffer + bufp_offset;
8148 SvCUR_set(sv, bytesread + append);
8149 buffer[bytesread] = '\0';
8150 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8156 Get a line from the filehandle and store it into the SV, optionally
8157 appending to the currently-stored string. If C<append> is not 0, the
8158 line is appended to the SV instead of overwriting it. C<append> should
8159 be set to the byte offset that the appended string should start at
8160 in the SV (typically, C<SvCUR(sv)> is a suitable choice).
8166 Perl_sv_gets(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8176 PERL_ARGS_ASSERT_SV_GETS;
8178 if (SvTHINKFIRST(sv))
8179 sv_force_normal_flags(sv, append ? 0 : SV_COW_DROP_PV);
8180 /* XXX. If you make this PVIV, then copy on write can copy scalars read
8182 However, perlbench says it's slower, because the existing swipe code
8183 is faster than copy on write.
8184 Swings and roundabouts. */
8185 SvUPGRADE(sv, SVt_PV);
8188 /* line is going to be appended to the existing buffer in the sv */
8189 if (PerlIO_isutf8(fp)) {
8191 sv_utf8_upgrade_nomg(sv);
8192 sv_pos_u2b(sv,&append,0);
8194 } else if (SvUTF8(sv)) {
8195 return S_sv_gets_append_to_utf8(aTHX_ sv, fp, append);
8201 /* not appending - "clear" the string by setting SvCUR to 0,
8202 * the pv is still avaiable. */
8205 if (PerlIO_isutf8(fp))
8208 if (IN_PERL_COMPILETIME) {
8209 /* we always read code in line mode */
8213 else if (RsSNARF(PL_rs)) {
8214 /* If it is a regular disk file use size from stat() as estimate
8215 of amount we are going to read -- may result in mallocing
8216 more memory than we really need if the layers below reduce
8217 the size we read (e.g. CRLF or a gzip layer).
8220 if (!PerlLIO_fstat(PerlIO_fileno(fp), &st) && S_ISREG(st.st_mode)) {
8221 const Off_t offset = PerlIO_tell(fp);
8222 if (offset != (Off_t) -1 && st.st_size + append > offset) {
8223 #ifdef PERL_NEW_COPY_ON_WRITE
8224 /* Add an extra byte for the sake of copy-on-write's
8225 * buffer reference count. */
8226 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 2));
8228 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 1));
8235 else if (RsRECORD(PL_rs)) {
8236 return S_sv_gets_read_record(aTHX_ sv, fp, append);
8238 else if (RsPARA(PL_rs)) {
8244 /* Get $/ i.e. PL_rs into same encoding as stream wants */
8245 if (PerlIO_isutf8(fp)) {
8246 rsptr = SvPVutf8(PL_rs, rslen);
8249 if (SvUTF8(PL_rs)) {
8250 if (!sv_utf8_downgrade(PL_rs, TRUE)) {
8251 Perl_croak(aTHX_ "Wide character in $/");
8254 /* extract the raw pointer to the record separator */
8255 rsptr = SvPV_const(PL_rs, rslen);
8259 /* rslast is the last character in the record separator
8260 * note we don't use rslast except when rslen is true, so the
8261 * null assign is a placeholder. */
8262 rslast = rslen ? rsptr[rslen - 1] : '\0';
8264 if (rspara) { /* have to do this both before and after */
8265 do { /* to make sure file boundaries work right */
8268 i = PerlIO_getc(fp);
8272 PerlIO_ungetc(fp,i);
8278 /* See if we know enough about I/O mechanism to cheat it ! */
8280 /* This used to be #ifdef test - it is made run-time test for ease
8281 of abstracting out stdio interface. One call should be cheap
8282 enough here - and may even be a macro allowing compile
8286 if (PerlIO_fast_gets(fp)) {
8288 * We can do buffer based IO operations on this filehandle.
8290 * This means we can bypass a lot of subcalls and process
8291 * the buffer directly, it also means we know the upper bound
8292 * on the amount of data we might read of the current buffer
8293 * into our sv. Knowing this allows us to preallocate the pv
8294 * to be able to hold that maximum, which allows us to simplify
8295 * a lot of logic. */
8298 * We're going to steal some values from the stdio struct
8299 * and put EVERYTHING in the innermost loop into registers.
8301 STDCHAR *ptr; /* pointer into fp's read-ahead buffer */
8302 STRLEN bpx; /* length of the data in the target sv
8303 used to fix pointers after a SvGROW */
8304 I32 shortbuffered; /* If the pv buffer is shorter than the amount
8305 of data left in the read-ahead buffer.
8306 If 0 then the pv buffer can hold the full
8307 amount left, otherwise this is the amount it
8310 #if defined(__VMS) && defined(PERLIO_IS_STDIO)
8311 /* An ungetc()d char is handled separately from the regular
8312 * buffer, so we getc() it back out and stuff it in the buffer.
8314 i = PerlIO_getc(fp);
8315 if (i == EOF) return 0;
8316 *(--((*fp)->_ptr)) = (unsigned char) i;
8320 /* Here is some breathtakingly efficient cheating */
8322 /* When you read the following logic resist the urge to think
8323 * of record separators that are 1 byte long. They are an
8324 * uninteresting special (simple) case.
8326 * Instead think of record separators which are at least 2 bytes
8327 * long, and keep in mind that we need to deal with such
8328 * separators when they cross a read-ahead buffer boundary.
8330 * Also consider that we need to gracefully deal with separators
8331 * that may be longer than a single read ahead buffer.
8333 * Lastly do not forget we want to copy the delimiter as well. We
8334 * are copying all data in the file _up_to_and_including_ the separator
8337 * Now that you have all that in mind here is what is happening below:
8339 * 1. When we first enter the loop we do some memory book keeping to see
8340 * how much free space there is in the target SV. (This sub assumes that
8341 * it is operating on the same SV most of the time via $_ and that it is
8342 * going to be able to reuse the same pv buffer each call.) If there is
8343 * "enough" room then we set "shortbuffered" to how much space there is
8344 * and start reading forward.
8346 * 2. When we scan forward we copy from the read-ahead buffer to the target
8347 * SV's pv buffer. While we go we watch for the end of the read-ahead buffer,
8348 * and the end of the of pv, as well as for the "rslast", which is the last
8349 * char of the separator.
8351 * 3. When scanning forward if we see rslast then we jump backwards in *pv*
8352 * (which has a "complete" record up to the point we saw rslast) and check
8353 * it to see if it matches the separator. If it does we are done. If it doesn't
8354 * we continue on with the scan/copy.
8356 * 4. If we run out of read-ahead buffer (cnt goes to 0) then we have to get
8357 * the IO system to read the next buffer. We do this by doing a getc(), which
8358 * returns a single char read (or EOF), and prefills the buffer, and also
8359 * allows us to find out how full the buffer is. We use this information to
8360 * SvGROW() the sv to the size remaining in the buffer, after which we copy
8361 * the returned single char into the target sv, and then go back into scan
8364 * 5. If we run out of write-buffer then we SvGROW() it by the size of the
8365 * remaining space in the read-buffer.
8367 * Note that this code despite its twisty-turny nature is pretty darn slick.
8368 * It manages single byte separators, multi-byte cross boundary separators,
8369 * and cross-read-buffer separators cleanly and efficiently at the cost
8370 * of potentially greatly overallocating the target SV.
8376 /* get the number of bytes remaining in the read-ahead buffer
8377 * on first call on a given fp this will return 0.*/
8378 cnt = PerlIO_get_cnt(fp);
8380 /* make sure we have the room */
8381 if ((I32)(SvLEN(sv) - append) <= cnt + 1) {
8382 /* Not room for all of it
8383 if we are looking for a separator and room for some
8385 if (rslen && cnt > 80 && (I32)SvLEN(sv) > append) {
8386 /* just process what we have room for */
8387 shortbuffered = cnt - SvLEN(sv) + append + 1;
8388 cnt -= shortbuffered;
8391 /* ensure that the target sv has enough room to hold
8392 * the rest of the read-ahead buffer */
8394 /* remember that cnt can be negative */
8395 SvGROW(sv, (STRLEN)(append + (cnt <= 0 ? 2 : (cnt + 1))));
8399 /* we have enough room to hold the full buffer, lets scream */
8403 /* extract the pointer to sv's string buffer, offset by append as necessary */
8404 bp = (STDCHAR*)SvPVX_const(sv) + append; /* move these two too to registers */
8405 /* extract the point to the read-ahead buffer */
8406 ptr = (STDCHAR*)PerlIO_get_ptr(fp);
8408 /* some trace debug output */
8409 DEBUG_P(PerlIO_printf(Perl_debug_log,
8410 "Screamer: entering, ptr=%"UVuf", cnt=%ld\n",PTR2UV(ptr),(long)cnt));
8411 DEBUG_P(PerlIO_printf(Perl_debug_log,
8412 "Screamer: entering: PerlIO * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"
8414 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8415 PTR2UV(PerlIO_has_base(fp) ? PerlIO_get_base(fp) : 0)));
8419 /* if there is stuff left in the read-ahead buffer */
8421 /* if there is a separator */
8423 /* loop until we hit the end of the read-ahead buffer */
8424 while (cnt > 0) { /* this | eat */
8425 /* scan forward copying and searching for rslast as we go */
8427 if ((*bp++ = *ptr++) == rslast) /* really | dust */
8428 goto thats_all_folks; /* screams | sed :-) */
8432 /* no separator, slurp the full buffer */
8433 Copy(ptr, bp, cnt, char); /* this | eat */
8434 bp += cnt; /* screams | dust */
8435 ptr += cnt; /* louder | sed :-) */
8437 assert (!shortbuffered);
8438 goto cannot_be_shortbuffered;
8442 if (shortbuffered) { /* oh well, must extend */
8443 /* we didnt have enough room to fit the line into the target buffer
8444 * so we must extend the target buffer and keep going */
8445 cnt = shortbuffered;
8447 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8449 /* extned the target sv's buffer so it can hold the full read-ahead buffer */
8450 SvGROW(sv, SvLEN(sv) + append + cnt + 2);
8451 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8455 cannot_be_shortbuffered:
8456 /* we need to refill the read-ahead buffer if possible */
8458 DEBUG_P(PerlIO_printf(Perl_debug_log,
8459 "Screamer: going to getc, ptr=%"UVuf", cnt=%"IVdf"\n",
8460 PTR2UV(ptr),(IV)cnt));
8461 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* deregisterize cnt and ptr */
8463 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8464 "Screamer: pre: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf"\n",
8465 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8466 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8469 call PerlIO_getc() to let it prefill the lookahead buffer
8471 This used to call 'filbuf' in stdio form, but as that behaves like
8472 getc when cnt <= 0 we use PerlIO_getc here to avoid introducing
8473 another abstraction.
8475 Note we have to deal with the char in 'i' if we are not at EOF
8477 i = PerlIO_getc(fp); /* get more characters */
8479 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8480 "Screamer: post: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf"\n",
8481 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8482 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8484 /* find out how much is left in the read-ahead buffer, and rextract its pointer */
8485 cnt = PerlIO_get_cnt(fp);
8486 ptr = (STDCHAR*)PerlIO_get_ptr(fp); /* reregisterize cnt and ptr */
8487 DEBUG_P(PerlIO_printf(Perl_debug_log,
8488 "Screamer: after getc, ptr=%"UVuf", cnt=%"IVdf"\n",
8489 PTR2UV(ptr),(IV)cnt));
8491 if (i == EOF) /* all done for ever? */
8492 goto thats_really_all_folks;
8494 /* make sure we have enough space in the target sv */
8495 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8497 SvGROW(sv, bpx + cnt + 2);
8498 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8500 /* copy of the char we got from getc() */
8501 *bp++ = (STDCHAR)i; /* store character from PerlIO_getc */
8503 /* make sure we deal with the i being the last character of a separator */
8504 if (rslen && (STDCHAR)i == rslast) /* all done for now? */
8505 goto thats_all_folks;
8509 /* check if we have actually found the separator - only really applies
8511 if ((rslen > 1 && (STRLEN)(bp - (STDCHAR*)SvPVX_const(sv)) < rslen) ||
8512 memNE((char*)bp - rslen, rsptr, rslen))
8513 goto screamer; /* go back to the fray */
8514 thats_really_all_folks:
8516 cnt += shortbuffered;
8517 DEBUG_P(PerlIO_printf(Perl_debug_log,
8518 "Screamer: quitting, ptr=%"UVuf", cnt=%"IVdf"\n",PTR2UV(ptr),(IV)cnt));
8519 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* put these back or we're in trouble */
8520 DEBUG_P(PerlIO_printf(Perl_debug_log,
8521 "Screamer: end: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf
8523 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8524 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8526 SvCUR_set(sv, bp - (STDCHAR*)SvPVX_const(sv)); /* set length */
8527 DEBUG_P(PerlIO_printf(Perl_debug_log,
8528 "Screamer: done, len=%ld, string=|%.*s|\n",
8529 (long)SvCUR(sv),(int)SvCUR(sv),SvPVX_const(sv)));
8533 /*The big, slow, and stupid way. */
8534 #ifdef USE_HEAP_INSTEAD_OF_STACK /* Even slower way. */
8535 STDCHAR *buf = NULL;
8536 Newx(buf, 8192, STDCHAR);
8544 const STDCHAR * const bpe = buf + sizeof(buf);
8546 while ((i = PerlIO_getc(fp)) != EOF && (*bp++ = (STDCHAR)i) != rslast && bp < bpe)
8547 ; /* keep reading */
8551 cnt = PerlIO_read(fp,(char*)buf, sizeof(buf));
8552 /* Accommodate broken VAXC compiler, which applies U8 cast to
8553 * both args of ?: operator, causing EOF to change into 255
8556 i = (U8)buf[cnt - 1];
8562 cnt = 0; /* we do need to re-set the sv even when cnt <= 0 */
8564 sv_catpvn_nomg(sv, (char *) buf, cnt);
8566 sv_setpvn(sv, (char *) buf, cnt); /* "nomg" is implied */
8568 if (i != EOF && /* joy */
8570 SvCUR(sv) < rslen ||
8571 memNE(SvPVX_const(sv) + SvCUR(sv) - rslen, rsptr, rslen)))
8575 * If we're reading from a TTY and we get a short read,
8576 * indicating that the user hit his EOF character, we need
8577 * to notice it now, because if we try to read from the TTY
8578 * again, the EOF condition will disappear.
8580 * The comparison of cnt to sizeof(buf) is an optimization
8581 * that prevents unnecessary calls to feof().
8585 if (!(cnt < (I32)sizeof(buf) && PerlIO_eof(fp)))
8589 #ifdef USE_HEAP_INSTEAD_OF_STACK
8594 if (rspara) { /* have to do this both before and after */
8595 while (i != EOF) { /* to make sure file boundaries work right */
8596 i = PerlIO_getc(fp);
8598 PerlIO_ungetc(fp,i);
8604 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8610 Auto-increment of the value in the SV, doing string to numeric conversion
8611 if necessary. Handles 'get' magic and operator overloading.
8617 Perl_sv_inc(pTHX_ SV *const sv)
8626 =for apidoc sv_inc_nomg
8628 Auto-increment of the value in the SV, doing string to numeric conversion
8629 if necessary. Handles operator overloading. Skips handling 'get' magic.
8635 Perl_sv_inc_nomg(pTHX_ SV *const sv)
8642 if (SvTHINKFIRST(sv)) {
8643 if (SvREADONLY(sv)) {
8644 Perl_croak_no_modify();
8648 if (SvAMAGIC(sv) && AMG_CALLunary(sv, inc_amg))
8650 i = PTR2IV(SvRV(sv));
8654 else sv_force_normal_flags(sv, 0);
8656 flags = SvFLAGS(sv);
8657 if ((flags & (SVp_NOK|SVp_IOK)) == SVp_NOK) {
8658 /* It's (privately or publicly) a float, but not tested as an
8659 integer, so test it to see. */
8661 flags = SvFLAGS(sv);
8663 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8664 /* It's publicly an integer, or privately an integer-not-float */
8665 #ifdef PERL_PRESERVE_IVUV
8669 if (SvUVX(sv) == UV_MAX)
8670 sv_setnv(sv, UV_MAX_P1);
8672 (void)SvIOK_only_UV(sv);
8673 SvUV_set(sv, SvUVX(sv) + 1);
8675 if (SvIVX(sv) == IV_MAX)
8676 sv_setuv(sv, (UV)IV_MAX + 1);
8678 (void)SvIOK_only(sv);
8679 SvIV_set(sv, SvIVX(sv) + 1);
8684 if (flags & SVp_NOK) {
8685 const NV was = SvNVX(sv);
8686 if (LIKELY(!Perl_isinfnan(was)) &&
8687 NV_OVERFLOWS_INTEGERS_AT &&
8688 was >= NV_OVERFLOWS_INTEGERS_AT) {
8689 /* diag_listed_as: Lost precision when %s %f by 1 */
8690 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8691 "Lost precision when incrementing %" NVff " by 1",
8694 (void)SvNOK_only(sv);
8695 SvNV_set(sv, was + 1.0);
8699 if (!(flags & SVp_POK) || !*SvPVX_const(sv)) {
8700 if ((flags & SVTYPEMASK) < SVt_PVIV)
8701 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV ? SVt_PVIV : SVt_IV));
8702 (void)SvIOK_only(sv);
8707 while (isALPHA(*d)) d++;
8708 while (isDIGIT(*d)) d++;
8709 if (d < SvEND(sv)) {
8710 const int numtype = grok_number_flags(SvPVX_const(sv), SvCUR(sv), NULL, PERL_SCAN_TRAILING);
8711 #ifdef PERL_PRESERVE_IVUV
8712 /* Got to punt this as an integer if needs be, but we don't issue
8713 warnings. Probably ought to make the sv_iv_please() that does
8714 the conversion if possible, and silently. */
8715 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
8716 /* Need to try really hard to see if it's an integer.
8717 9.22337203685478e+18 is an integer.
8718 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
8719 so $a="9.22337203685478e+18"; $a+0; $a++
8720 needs to be the same as $a="9.22337203685478e+18"; $a++
8727 /* sv_2iv *should* have made this an NV */
8728 if (flags & SVp_NOK) {
8729 (void)SvNOK_only(sv);
8730 SvNV_set(sv, SvNVX(sv) + 1.0);
8733 /* I don't think we can get here. Maybe I should assert this
8734 And if we do get here I suspect that sv_setnv will croak. NWC
8736 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_inc punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"NVgf"\n",
8737 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
8739 #endif /* PERL_PRESERVE_IVUV */
8740 if (!numtype && ckWARN(WARN_NUMERIC))
8741 not_incrementable(sv);
8742 sv_setnv(sv,Atof(SvPVX_const(sv)) + 1.0);
8746 while (d >= SvPVX_const(sv)) {
8754 /* MKS: The original code here died if letters weren't consecutive.
8755 * at least it didn't have to worry about non-C locales. The
8756 * new code assumes that ('z'-'a')==('Z'-'A'), letters are
8757 * arranged in order (although not consecutively) and that only
8758 * [A-Za-z] are accepted by isALPHA in the C locale.
8760 if (isALPHA_FOLD_NE(*d, 'z')) {
8761 do { ++*d; } while (!isALPHA(*d));
8764 *(d--) -= 'z' - 'a';
8769 *(d--) -= 'z' - 'a' + 1;
8773 /* oh,oh, the number grew */
8774 SvGROW(sv, SvCUR(sv) + 2);
8775 SvCUR_set(sv, SvCUR(sv) + 1);
8776 for (d = SvPVX(sv) + SvCUR(sv); d > SvPVX_const(sv); d--)
8787 Auto-decrement of the value in the SV, doing string to numeric conversion
8788 if necessary. Handles 'get' magic and operator overloading.
8794 Perl_sv_dec(pTHX_ SV *const sv)
8803 =for apidoc sv_dec_nomg
8805 Auto-decrement of the value in the SV, doing string to numeric conversion
8806 if necessary. Handles operator overloading. Skips handling 'get' magic.
8812 Perl_sv_dec_nomg(pTHX_ SV *const sv)
8818 if (SvTHINKFIRST(sv)) {
8819 if (SvREADONLY(sv)) {
8820 Perl_croak_no_modify();
8824 if (SvAMAGIC(sv) && AMG_CALLunary(sv, dec_amg))
8826 i = PTR2IV(SvRV(sv));
8830 else sv_force_normal_flags(sv, 0);
8832 /* Unlike sv_inc we don't have to worry about string-never-numbers
8833 and keeping them magic. But we mustn't warn on punting */
8834 flags = SvFLAGS(sv);
8835 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8836 /* It's publicly an integer, or privately an integer-not-float */
8837 #ifdef PERL_PRESERVE_IVUV
8841 if (SvUVX(sv) == 0) {
8842 (void)SvIOK_only(sv);
8846 (void)SvIOK_only_UV(sv);
8847 SvUV_set(sv, SvUVX(sv) - 1);
8850 if (SvIVX(sv) == IV_MIN) {
8851 sv_setnv(sv, (NV)IV_MIN);
8855 (void)SvIOK_only(sv);
8856 SvIV_set(sv, SvIVX(sv) - 1);
8861 if (flags & SVp_NOK) {
8864 const NV was = SvNVX(sv);
8865 if (LIKELY(!Perl_isinfnan(was)) &&
8866 NV_OVERFLOWS_INTEGERS_AT &&
8867 was <= -NV_OVERFLOWS_INTEGERS_AT) {
8868 /* diag_listed_as: Lost precision when %s %f by 1 */
8869 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8870 "Lost precision when decrementing %" NVff " by 1",
8873 (void)SvNOK_only(sv);
8874 SvNV_set(sv, was - 1.0);
8878 if (!(flags & SVp_POK)) {
8879 if ((flags & SVTYPEMASK) < SVt_PVIV)
8880 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV) ? SVt_PVIV : SVt_IV);
8882 (void)SvIOK_only(sv);
8885 #ifdef PERL_PRESERVE_IVUV
8887 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), NULL);
8888 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
8889 /* Need to try really hard to see if it's an integer.
8890 9.22337203685478e+18 is an integer.
8891 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
8892 so $a="9.22337203685478e+18"; $a+0; $a--
8893 needs to be the same as $a="9.22337203685478e+18"; $a--
8900 /* sv_2iv *should* have made this an NV */
8901 if (flags & SVp_NOK) {
8902 (void)SvNOK_only(sv);
8903 SvNV_set(sv, SvNVX(sv) - 1.0);
8906 /* I don't think we can get here. Maybe I should assert this
8907 And if we do get here I suspect that sv_setnv will croak. NWC
8909 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_dec punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"NVgf"\n",
8910 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
8913 #endif /* PERL_PRESERVE_IVUV */
8914 sv_setnv(sv,Atof(SvPVX_const(sv)) - 1.0); /* punt */
8917 /* this define is used to eliminate a chunk of duplicated but shared logic
8918 * it has the suffix __SV_C to signal that it isnt API, and isnt meant to be
8919 * used anywhere but here - yves
8921 #define PUSH_EXTEND_MORTAL__SV_C(AnSv) \
8923 SSize_t ix = ++PL_tmps_ix; \
8924 if (UNLIKELY(ix >= PL_tmps_max)) \
8925 ix = tmps_grow_p(ix); \
8926 PL_tmps_stack[ix] = (AnSv); \
8930 =for apidoc sv_mortalcopy
8932 Creates a new SV which is a copy of the original SV (using C<sv_setsv>).
8933 The new SV is marked as mortal. It will be destroyed "soon", either by an
8934 explicit call to FREETMPS, or by an implicit call at places such as
8935 statement boundaries. See also C<sv_newmortal> and C<sv_2mortal>.
8940 /* Make a string that will exist for the duration of the expression
8941 * evaluation. Actually, it may have to last longer than that, but
8942 * hopefully we won't free it until it has been assigned to a
8943 * permanent location. */
8946 Perl_sv_mortalcopy_flags(pTHX_ SV *const oldstr, U32 flags)
8950 if (flags & SV_GMAGIC)
8951 SvGETMAGIC(oldstr); /* before new_SV, in case it dies */
8953 sv_setsv_flags(sv,oldstr,flags & ~SV_GMAGIC);
8954 PUSH_EXTEND_MORTAL__SV_C(sv);
8960 =for apidoc sv_newmortal
8962 Creates a new null SV which is mortal. The reference count of the SV is
8963 set to 1. It will be destroyed "soon", either by an explicit call to
8964 FREETMPS, or by an implicit call at places such as statement boundaries.
8965 See also C<sv_mortalcopy> and C<sv_2mortal>.
8971 Perl_sv_newmortal(pTHX)
8976 SvFLAGS(sv) = SVs_TEMP;
8977 PUSH_EXTEND_MORTAL__SV_C(sv);
8983 =for apidoc newSVpvn_flags
8985 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
8986 characters) into it. The reference count for the
8987 SV is set to 1. Note that if C<len> is zero, Perl will create a zero length
8988 string. You are responsible for ensuring that the source string is at least
8989 C<len> bytes long. If the C<s> argument is NULL the new SV will be undefined.
8990 Currently the only flag bits accepted are C<SVf_UTF8> and C<SVs_TEMP>.
8991 If C<SVs_TEMP> is set, then C<sv_2mortal()> is called on the result before
8992 returning. If C<SVf_UTF8> is set, C<s>
8993 is considered to be in UTF-8 and the
8994 C<SVf_UTF8> flag will be set on the new SV.
8995 C<newSVpvn_utf8()> is a convenience wrapper for this function, defined as
8997 #define newSVpvn_utf8(s, len, u) \
8998 newSVpvn_flags((s), (len), (u) ? SVf_UTF8 : 0)
9004 Perl_newSVpvn_flags(pTHX_ const char *const s, const STRLEN len, const U32 flags)
9008 /* All the flags we don't support must be zero.
9009 And we're new code so I'm going to assert this from the start. */
9010 assert(!(flags & ~(SVf_UTF8|SVs_TEMP)));
9012 sv_setpvn(sv,s,len);
9014 /* This code used to do a sv_2mortal(), however we now unroll the call to
9015 * sv_2mortal() and do what it does ourselves here. Since we have asserted
9016 * that flags can only have the SVf_UTF8 and/or SVs_TEMP flags set above we
9017 * can use it to enable the sv flags directly (bypassing SvTEMP_on), which
9018 * in turn means we dont need to mask out the SVf_UTF8 flag below, which
9019 * means that we eliminate quite a few steps than it looks - Yves
9020 * (explaining patch by gfx) */
9022 SvFLAGS(sv) |= flags;
9024 if(flags & SVs_TEMP){
9025 PUSH_EXTEND_MORTAL__SV_C(sv);
9032 =for apidoc sv_2mortal
9034 Marks an existing SV as mortal. The SV will be destroyed "soon", either
9035 by an explicit call to FREETMPS, or by an implicit call at places such as
9036 statement boundaries. SvTEMP() is turned on which means that the SV's
9037 string buffer can be "stolen" if this SV is copied. See also C<sv_newmortal>
9038 and C<sv_mortalcopy>.
9044 Perl_sv_2mortal(pTHX_ SV *const sv)
9051 PUSH_EXTEND_MORTAL__SV_C(sv);
9059 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9060 characters) into it. The reference count for the
9061 SV is set to 1. If C<len> is zero, Perl will compute the length using
9062 strlen(), (which means if you use this option, that C<s> can't have embedded
9063 C<NUL> characters and has to have a terminating C<NUL> byte).
9065 For efficiency, consider using C<newSVpvn> instead.
9071 Perl_newSVpv(pTHX_ const char *const s, const STRLEN len)
9076 sv_setpvn(sv, s, len || s == NULL ? len : strlen(s));
9081 =for apidoc newSVpvn
9083 Creates a new SV and copies a string into it, which may contain C<NUL> characters
9084 (C<\0>) and other binary data. The reference count for the SV is set to 1.
9085 Note that if C<len> is zero, Perl will create a zero length (Perl) string. You
9086 are responsible for ensuring that the source buffer is at least
9087 C<len> bytes long. If the C<buffer> argument is NULL the new SV will be
9094 Perl_newSVpvn(pTHX_ const char *const buffer, const STRLEN len)
9098 sv_setpvn(sv,buffer,len);
9103 =for apidoc newSVhek
9105 Creates a new SV from the hash key structure. It will generate scalars that
9106 point to the shared string table where possible. Returns a new (undefined)
9107 SV if the hek is NULL.
9113 Perl_newSVhek(pTHX_ const HEK *const hek)
9122 if (HEK_LEN(hek) == HEf_SVKEY) {
9123 return newSVsv(*(SV**)HEK_KEY(hek));
9125 const int flags = HEK_FLAGS(hek);
9126 if (flags & HVhek_WASUTF8) {
9128 Andreas would like keys he put in as utf8 to come back as utf8
9130 STRLEN utf8_len = HEK_LEN(hek);
9131 SV * const sv = newSV_type(SVt_PV);
9132 char *as_utf8 = (char *)bytes_to_utf8 ((U8*)HEK_KEY(hek), &utf8_len);
9133 /* bytes_to_utf8() allocates a new string, which we can repurpose: */
9134 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
9137 } else if (flags & HVhek_UNSHARED) {
9138 /* A hash that isn't using shared hash keys has to have
9139 the flag in every key so that we know not to try to call
9140 share_hek_hek on it. */
9142 SV * const sv = newSVpvn (HEK_KEY(hek), HEK_LEN(hek));
9147 /* This will be overwhelminly the most common case. */
9149 /* Inline most of newSVpvn_share(), because share_hek_hek() is far
9150 more efficient than sharepvn(). */
9154 sv_upgrade(sv, SVt_PV);
9155 SvPV_set(sv, (char *)HEK_KEY(share_hek_hek(hek)));
9156 SvCUR_set(sv, HEK_LEN(hek));
9168 =for apidoc newSVpvn_share
9170 Creates a new SV with its SvPVX_const pointing to a shared string in the string
9171 table. If the string does not already exist in the table, it is
9172 created first. Turns on the SvIsCOW flag (or READONLY
9173 and FAKE in 5.16 and earlier). If the C<hash> parameter
9174 is non-zero, that value is used; otherwise the hash is computed.
9175 The string's hash can later be retrieved from the SV
9176 with the C<SvSHARED_HASH()> macro. The idea here is
9177 that as the string table is used for shared hash keys these strings will have
9178 SvPVX_const == HeKEY and hash lookup will avoid string compare.
9184 Perl_newSVpvn_share(pTHX_ const char *src, I32 len, U32 hash)
9188 bool is_utf8 = FALSE;
9189 const char *const orig_src = src;
9192 STRLEN tmplen = -len;
9194 /* See the note in hv.c:hv_fetch() --jhi */
9195 src = (char*)bytes_from_utf8((const U8*)src, &tmplen, &is_utf8);
9199 PERL_HASH(hash, src, len);
9201 /* The logic for this is inlined in S_mro_get_linear_isa_dfs(), so if it
9202 changes here, update it there too. */
9203 sv_upgrade(sv, SVt_PV);
9204 SvPV_set(sv, sharepvn(src, is_utf8?-len:len, hash));
9211 if (src != orig_src)
9217 =for apidoc newSVpv_share
9219 Like C<newSVpvn_share>, but takes a C<NUL>-terminated string instead of a
9226 Perl_newSVpv_share(pTHX_ const char *src, U32 hash)
9228 return newSVpvn_share(src, strlen(src), hash);
9231 #if defined(PERL_IMPLICIT_CONTEXT)
9233 /* pTHX_ magic can't cope with varargs, so this is a no-context
9234 * version of the main function, (which may itself be aliased to us).
9235 * Don't access this version directly.
9239 Perl_newSVpvf_nocontext(const char *const pat, ...)
9245 PERL_ARGS_ASSERT_NEWSVPVF_NOCONTEXT;
9247 va_start(args, pat);
9248 sv = vnewSVpvf(pat, &args);
9255 =for apidoc newSVpvf
9257 Creates a new SV and initializes it with the string formatted like
9264 Perl_newSVpvf(pTHX_ const char *const pat, ...)
9269 PERL_ARGS_ASSERT_NEWSVPVF;
9271 va_start(args, pat);
9272 sv = vnewSVpvf(pat, &args);
9277 /* backend for newSVpvf() and newSVpvf_nocontext() */
9280 Perl_vnewSVpvf(pTHX_ const char *const pat, va_list *const args)
9284 PERL_ARGS_ASSERT_VNEWSVPVF;
9287 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
9294 Creates a new SV and copies a floating point value into it.
9295 The reference count for the SV is set to 1.
9301 Perl_newSVnv(pTHX_ const NV n)
9313 Creates a new SV and copies an integer into it. The reference count for the
9320 Perl_newSViv(pTHX_ const IV i)
9332 Creates a new SV and copies an unsigned integer into it.
9333 The reference count for the SV is set to 1.
9339 Perl_newSVuv(pTHX_ const UV u)
9349 =for apidoc newSV_type
9351 Creates a new SV, of the type specified. The reference count for the new SV
9358 Perl_newSV_type(pTHX_ const svtype type)
9363 ASSUME(SvTYPE(sv) == SVt_FIRST);
9364 if(type != SVt_FIRST)
9365 sv_upgrade(sv, type);
9370 =for apidoc newRV_noinc
9372 Creates an RV wrapper for an SV. The reference count for the original
9373 SV is B<not> incremented.
9379 Perl_newRV_noinc(pTHX_ SV *const tmpRef)
9381 SV *sv = newSV_type(SVt_IV);
9383 PERL_ARGS_ASSERT_NEWRV_NOINC;
9386 SvRV_set(sv, tmpRef);
9391 /* newRV_inc is the official function name to use now.
9392 * newRV_inc is in fact #defined to newRV in sv.h
9396 Perl_newRV(pTHX_ SV *const sv)
9398 PERL_ARGS_ASSERT_NEWRV;
9400 return newRV_noinc(SvREFCNT_inc_simple_NN(sv));
9406 Creates a new SV which is an exact duplicate of the original SV.
9413 Perl_newSVsv(pTHX_ SV *const old)
9419 if (SvTYPE(old) == (svtype)SVTYPEMASK) {
9420 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), "semi-panic: attempt to dup freed string");
9423 /* Do this here, otherwise we leak the new SV if this croaks. */
9426 /* SV_NOSTEAL prevents TEMP buffers being, well, stolen, and saves games
9427 with SvTEMP_off and SvTEMP_on round a call to sv_setsv. */
9428 sv_setsv_flags(sv, old, SV_NOSTEAL);
9433 =for apidoc sv_reset
9435 Underlying implementation for the C<reset> Perl function.
9436 Note that the perl-level function is vaguely deprecated.
9442 Perl_sv_reset(pTHX_ const char *s, HV *const stash)
9444 PERL_ARGS_ASSERT_SV_RESET;
9446 sv_resetpvn(*s ? s : NULL, strlen(s), stash);
9450 Perl_sv_resetpvn(pTHX_ const char *s, STRLEN len, HV * const stash)
9452 char todo[PERL_UCHAR_MAX+1];
9455 if (!stash || SvTYPE(stash) != SVt_PVHV)
9458 if (!s) { /* reset ?? searches */
9459 MAGIC * const mg = mg_find((const SV *)stash, PERL_MAGIC_symtab);
9461 const U32 count = mg->mg_len / sizeof(PMOP**);
9462 PMOP **pmp = (PMOP**) mg->mg_ptr;
9463 PMOP *const *const end = pmp + count;
9467 SvREADONLY_off(PL_regex_pad[(*pmp)->op_pmoffset]);
9469 (*pmp)->op_pmflags &= ~PMf_USED;
9477 /* reset variables */
9479 if (!HvARRAY(stash))
9482 Zero(todo, 256, char);
9486 I32 i = (unsigned char)*s;
9490 max = (unsigned char)*s++;
9491 for ( ; i <= max; i++) {
9494 for (i = 0; i <= (I32) HvMAX(stash); i++) {
9496 for (entry = HvARRAY(stash)[i];
9498 entry = HeNEXT(entry))
9503 if (!todo[(U8)*HeKEY(entry)])
9505 gv = MUTABLE_GV(HeVAL(entry));
9507 if (sv && !SvREADONLY(sv)) {
9508 SV_CHECK_THINKFIRST_COW_DROP(sv);
9509 if (!isGV(sv)) SvOK_off(sv);
9514 if (GvHV(gv) && !HvNAME_get(GvHV(gv))) {
9525 Using various gambits, try to get an IO from an SV: the IO slot if its a
9526 GV; or the recursive result if we're an RV; or the IO slot of the symbol
9527 named after the PV if we're a string.
9529 'Get' magic is ignored on the sv passed in, but will be called on
9530 C<SvRV(sv)> if sv is an RV.
9536 Perl_sv_2io(pTHX_ SV *const sv)
9541 PERL_ARGS_ASSERT_SV_2IO;
9543 switch (SvTYPE(sv)) {
9545 io = MUTABLE_IO(sv);
9549 if (isGV_with_GP(sv)) {
9550 gv = MUTABLE_GV(sv);
9553 Perl_croak(aTHX_ "Bad filehandle: %"HEKf,
9554 HEKfARG(GvNAME_HEK(gv)));
9560 Perl_croak(aTHX_ PL_no_usym, "filehandle");
9562 SvGETMAGIC(SvRV(sv));
9563 return sv_2io(SvRV(sv));
9565 gv = gv_fetchsv_nomg(sv, 0, SVt_PVIO);
9572 if (SvGMAGICAL(sv)) {
9573 newsv = sv_newmortal();
9574 sv_setsv_nomg(newsv, sv);
9576 Perl_croak(aTHX_ "Bad filehandle: %"SVf, SVfARG(newsv));
9586 Using various gambits, try to get a CV from an SV; in addition, try if
9587 possible to set C<*st> and C<*gvp> to the stash and GV associated with it.
9588 The flags in C<lref> are passed to gv_fetchsv.
9594 Perl_sv_2cv(pTHX_ SV *sv, HV **const st, GV **const gvp, const I32 lref)
9599 PERL_ARGS_ASSERT_SV_2CV;
9606 switch (SvTYPE(sv)) {
9610 return MUTABLE_CV(sv);
9620 sv = amagic_deref_call(sv, to_cv_amg);
9623 if (SvTYPE(sv) == SVt_PVCV) {
9624 cv = MUTABLE_CV(sv);
9629 else if(SvGETMAGIC(sv), isGV_with_GP(sv))
9630 gv = MUTABLE_GV(sv);
9632 Perl_croak(aTHX_ "Not a subroutine reference");
9634 else if (isGV_with_GP(sv)) {
9635 gv = MUTABLE_GV(sv);
9638 gv = gv_fetchsv_nomg(sv, lref, SVt_PVCV);
9645 /* Some flags to gv_fetchsv mean don't really create the GV */
9646 if (!isGV_with_GP(gv)) {
9651 if (lref & ~GV_ADDMG && !GvCVu(gv)) {
9652 /* XXX this is probably not what they think they're getting.
9653 * It has the same effect as "sub name;", i.e. just a forward
9664 Returns true if the SV has a true value by Perl's rules.
9665 Use the C<SvTRUE> macro instead, which may call C<sv_true()> or may
9666 instead use an in-line version.
9672 Perl_sv_true(pTHX_ SV *const sv)
9677 const XPV* const tXpv = (XPV*)SvANY(sv);
9679 (tXpv->xpv_cur > 1 ||
9680 (tXpv->xpv_cur && *sv->sv_u.svu_pv != '0')))
9687 return SvIVX(sv) != 0;
9690 return SvNVX(sv) != 0.0;
9692 return sv_2bool(sv);
9698 =for apidoc sv_pvn_force
9700 Get a sensible string out of the SV somehow.
9701 A private implementation of the C<SvPV_force> macro for compilers which
9702 can't cope with complex macro expressions. Always use the macro instead.
9704 =for apidoc sv_pvn_force_flags
9706 Get a sensible string out of the SV somehow.
9707 If C<flags> has C<SV_GMAGIC> bit set, will C<mg_get> on C<sv> if
9708 appropriate, else not. C<sv_pvn_force> and C<sv_pvn_force_nomg> are
9709 implemented in terms of this function.
9710 You normally want to use the various wrapper macros instead: see
9711 C<SvPV_force> and C<SvPV_force_nomg>
9717 Perl_sv_pvn_force_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
9719 PERL_ARGS_ASSERT_SV_PVN_FORCE_FLAGS;
9721 if (flags & SV_GMAGIC) SvGETMAGIC(sv);
9722 if (SvTHINKFIRST(sv) && (!SvROK(sv) || SvREADONLY(sv)))
9723 sv_force_normal_flags(sv, 0);
9733 if (SvTYPE(sv) > SVt_PVLV
9734 || isGV_with_GP(sv))
9735 /* diag_listed_as: Can't coerce %s to %s in %s */
9736 Perl_croak(aTHX_ "Can't coerce %s to string in %s", sv_reftype(sv,0),
9738 s = sv_2pv_flags(sv, &len, flags &~ SV_GMAGIC);
9745 if (SvTYPE(sv) < SVt_PV ||
9746 s != SvPVX_const(sv)) { /* Almost, but not quite, sv_setpvn() */
9749 SvUPGRADE(sv, SVt_PV); /* Never FALSE */
9750 SvGROW(sv, len + 1);
9751 Move(s,SvPVX(sv),len,char);
9753 SvPVX(sv)[len] = '\0';
9756 SvPOK_on(sv); /* validate pointer */
9758 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
9759 PTR2UV(sv),SvPVX_const(sv)));
9762 (void)SvPOK_only_UTF8(sv);
9763 return SvPVX_mutable(sv);
9767 =for apidoc sv_pvbyten_force
9769 The backend for the C<SvPVbytex_force> macro. Always use the macro
9776 Perl_sv_pvbyten_force(pTHX_ SV *const sv, STRLEN *const lp)
9778 PERL_ARGS_ASSERT_SV_PVBYTEN_FORCE;
9780 sv_pvn_force(sv,lp);
9781 sv_utf8_downgrade(sv,0);
9787 =for apidoc sv_pvutf8n_force
9789 The backend for the C<SvPVutf8x_force> macro. Always use the macro
9796 Perl_sv_pvutf8n_force(pTHX_ SV *const sv, STRLEN *const lp)
9798 PERL_ARGS_ASSERT_SV_PVUTF8N_FORCE;
9801 sv_utf8_upgrade_nomg(sv);
9807 =for apidoc sv_reftype
9809 Returns a string describing what the SV is a reference to.
9815 Perl_sv_reftype(pTHX_ const SV *const sv, const int ob)
9817 PERL_ARGS_ASSERT_SV_REFTYPE;
9818 if (ob && SvOBJECT(sv)) {
9819 return SvPV_nolen_const(sv_ref(NULL, sv, ob));
9822 /* WARNING - There is code, for instance in mg.c, that assumes that
9823 * the only reason that sv_reftype(sv,0) would return a string starting
9824 * with 'L' or 'S' is that it is a LVALUE or a SCALAR.
9825 * Yes this a dodgy way to do type checking, but it saves practically reimplementing
9826 * this routine inside other subs, and it saves time.
9827 * Do not change this assumption without searching for "dodgy type check" in
9830 switch (SvTYPE(sv)) {
9845 case SVt_PVLV: return (char *) (SvROK(sv) ? "REF"
9846 /* tied lvalues should appear to be
9847 * scalars for backwards compatibility */
9848 : (isALPHA_FOLD_EQ(LvTYPE(sv), 't'))
9849 ? "SCALAR" : "LVALUE");
9850 case SVt_PVAV: return "ARRAY";
9851 case SVt_PVHV: return "HASH";
9852 case SVt_PVCV: return "CODE";
9853 case SVt_PVGV: return (char *) (isGV_with_GP(sv)
9854 ? "GLOB" : "SCALAR");
9855 case SVt_PVFM: return "FORMAT";
9856 case SVt_PVIO: return "IO";
9857 case SVt_INVLIST: return "INVLIST";
9858 case SVt_REGEXP: return "REGEXP";
9859 default: return "UNKNOWN";
9867 Returns a SV describing what the SV passed in is a reference to.
9873 Perl_sv_ref(pTHX_ SV *dst, const SV *const sv, const int ob)
9875 PERL_ARGS_ASSERT_SV_REF;
9878 dst = sv_newmortal();
9880 if (ob && SvOBJECT(sv)) {
9881 HvNAME_get(SvSTASH(sv))
9882 ? sv_sethek(dst, HvNAME_HEK(SvSTASH(sv)))
9883 : sv_setpvn(dst, "__ANON__", 8);
9886 const char * reftype = sv_reftype(sv, 0);
9887 sv_setpv(dst, reftype);
9893 =for apidoc sv_isobject
9895 Returns a boolean indicating whether the SV is an RV pointing to a blessed
9896 object. If the SV is not an RV, or if the object is not blessed, then this
9903 Perl_sv_isobject(pTHX_ SV *sv)
9919 Returns a boolean indicating whether the SV is blessed into the specified
9920 class. This does not check for subtypes; use C<sv_derived_from> to verify
9921 an inheritance relationship.
9927 Perl_sv_isa(pTHX_ SV *sv, const char *const name)
9931 PERL_ARGS_ASSERT_SV_ISA;
9941 hvname = HvNAME_get(SvSTASH(sv));
9945 return strEQ(hvname, name);
9951 Creates a new SV for the existing RV, C<rv>, to point to. If C<rv> is not an
9952 RV then it will be upgraded to one. If C<classname> is non-null then the new
9953 SV will be blessed in the specified package. The new SV is returned and its
9954 reference count is 1. The reference count 1 is owned by C<rv>.
9960 Perl_newSVrv(pTHX_ SV *const rv, const char *const classname)
9964 PERL_ARGS_ASSERT_NEWSVRV;
9968 SV_CHECK_THINKFIRST_COW_DROP(rv);
9970 if (SvTYPE(rv) >= SVt_PVMG) {
9971 const U32 refcnt = SvREFCNT(rv);
9975 SvREFCNT(rv) = refcnt;
9977 sv_upgrade(rv, SVt_IV);
9978 } else if (SvROK(rv)) {
9979 SvREFCNT_dec(SvRV(rv));
9981 prepare_SV_for_RV(rv);
9989 HV* const stash = gv_stashpv(classname, GV_ADD);
9990 (void)sv_bless(rv, stash);
9996 Perl_newSVavdefelem(pTHX_ AV *av, SSize_t ix, bool extendible)
9998 SV * const lv = newSV_type(SVt_PVLV);
9999 PERL_ARGS_ASSERT_NEWSVAVDEFELEM;
10001 sv_magic(lv, NULL, PERL_MAGIC_defelem, NULL, 0);
10002 LvTARG(lv) = SvREFCNT_inc_simple_NN(av);
10003 LvSTARGOFF(lv) = ix;
10004 LvTARGLEN(lv) = extendible ? 1 : (STRLEN)UV_MAX;
10009 =for apidoc sv_setref_pv
10011 Copies a pointer into a new SV, optionally blessing the SV. The C<rv>
10012 argument will be upgraded to an RV. That RV will be modified to point to
10013 the new SV. If the C<pv> argument is NULL then C<PL_sv_undef> will be placed
10014 into the SV. The C<classname> argument indicates the package for the
10015 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10016 will have a reference count of 1, and the RV will be returned.
10018 Do not use with other Perl types such as HV, AV, SV, CV, because those
10019 objects will become corrupted by the pointer copy process.
10021 Note that C<sv_setref_pvn> copies the string while this copies the pointer.
10027 Perl_sv_setref_pv(pTHX_ SV *const rv, const char *const classname, void *const pv)
10029 PERL_ARGS_ASSERT_SV_SETREF_PV;
10032 sv_setsv(rv, &PL_sv_undef);
10036 sv_setiv(newSVrv(rv,classname), PTR2IV(pv));
10041 =for apidoc sv_setref_iv
10043 Copies an integer into a new SV, optionally blessing the SV. The C<rv>
10044 argument will be upgraded to an RV. That RV will be modified to point to
10045 the new SV. The C<classname> argument indicates the package for the
10046 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10047 will have a reference count of 1, and the RV will be returned.
10053 Perl_sv_setref_iv(pTHX_ SV *const rv, const char *const classname, const IV iv)
10055 PERL_ARGS_ASSERT_SV_SETREF_IV;
10057 sv_setiv(newSVrv(rv,classname), iv);
10062 =for apidoc sv_setref_uv
10064 Copies an unsigned integer into a new SV, optionally blessing the SV. The C<rv>
10065 argument will be upgraded to an RV. That RV will be modified to point to
10066 the new SV. The C<classname> argument indicates the package for the
10067 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10068 will have a reference count of 1, and the RV will be returned.
10074 Perl_sv_setref_uv(pTHX_ SV *const rv, const char *const classname, const UV uv)
10076 PERL_ARGS_ASSERT_SV_SETREF_UV;
10078 sv_setuv(newSVrv(rv,classname), uv);
10083 =for apidoc sv_setref_nv
10085 Copies a double into a new SV, optionally blessing the SV. The C<rv>
10086 argument will be upgraded to an RV. That RV will be modified to point to
10087 the new SV. The C<classname> argument indicates the package for the
10088 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10089 will have a reference count of 1, and the RV will be returned.
10095 Perl_sv_setref_nv(pTHX_ SV *const rv, const char *const classname, const NV nv)
10097 PERL_ARGS_ASSERT_SV_SETREF_NV;
10099 sv_setnv(newSVrv(rv,classname), nv);
10104 =for apidoc sv_setref_pvn
10106 Copies a string into a new SV, optionally blessing the SV. The length of the
10107 string must be specified with C<n>. The C<rv> argument will be upgraded to
10108 an RV. That RV will be modified to point to the new SV. The C<classname>
10109 argument indicates the package for the blessing. Set C<classname> to
10110 C<NULL> to avoid the blessing. The new SV will have a reference count
10111 of 1, and the RV will be returned.
10113 Note that C<sv_setref_pv> copies the pointer while this copies the string.
10119 Perl_sv_setref_pvn(pTHX_ SV *const rv, const char *const classname,
10120 const char *const pv, const STRLEN n)
10122 PERL_ARGS_ASSERT_SV_SETREF_PVN;
10124 sv_setpvn(newSVrv(rv,classname), pv, n);
10129 =for apidoc sv_bless
10131 Blesses an SV into a specified package. The SV must be an RV. The package
10132 must be designated by its stash (see C<gv_stashpv()>). The reference count
10133 of the SV is unaffected.
10139 Perl_sv_bless(pTHX_ SV *const sv, HV *const stash)
10142 HV *oldstash = NULL;
10144 PERL_ARGS_ASSERT_SV_BLESS;
10148 Perl_croak(aTHX_ "Can't bless non-reference value");
10150 if (SvFLAGS(tmpRef) & (SVs_OBJECT|SVf_READONLY|SVf_PROTECT)) {
10151 if (SvREADONLY(tmpRef))
10152 Perl_croak_no_modify();
10153 if (SvOBJECT(tmpRef)) {
10154 oldstash = SvSTASH(tmpRef);
10157 SvOBJECT_on(tmpRef);
10158 SvUPGRADE(tmpRef, SVt_PVMG);
10159 SvSTASH_set(tmpRef, MUTABLE_HV(SvREFCNT_inc_simple(stash)));
10160 SvREFCNT_dec(oldstash);
10162 if(SvSMAGICAL(tmpRef))
10163 if(mg_find(tmpRef, PERL_MAGIC_ext) || mg_find(tmpRef, PERL_MAGIC_uvar))
10171 /* Downgrades a PVGV to a PVMG. If it's actually a PVLV, we leave the type
10172 * as it is after unglobbing it.
10175 PERL_STATIC_INLINE void
10176 S_sv_unglob(pTHX_ SV *const sv, U32 flags)
10180 SV * const temp = flags & SV_COW_DROP_PV ? NULL : sv_newmortal();
10182 PERL_ARGS_ASSERT_SV_UNGLOB;
10184 assert(SvTYPE(sv) == SVt_PVGV || SvTYPE(sv) == SVt_PVLV);
10186 if (!(flags & SV_COW_DROP_PV))
10187 gv_efullname3(temp, MUTABLE_GV(sv), "*");
10189 SvREFCNT_inc_simple_void_NN(sv_2mortal(sv));
10191 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
10192 && HvNAME_get(stash))
10193 mro_method_changed_in(stash);
10194 gp_free(MUTABLE_GV(sv));
10197 sv_del_backref(MUTABLE_SV(GvSTASH(sv)), sv);
10198 GvSTASH(sv) = NULL;
10201 if (GvNAME_HEK(sv)) {
10202 unshare_hek(GvNAME_HEK(sv));
10204 isGV_with_GP_off(sv);
10206 if(SvTYPE(sv) == SVt_PVGV) {
10207 /* need to keep SvANY(sv) in the right arena */
10208 xpvmg = new_XPVMG();
10209 StructCopy(SvANY(sv), xpvmg, XPVMG);
10210 del_XPVGV(SvANY(sv));
10213 SvFLAGS(sv) &= ~SVTYPEMASK;
10214 SvFLAGS(sv) |= SVt_PVMG;
10217 /* Intentionally not calling any local SET magic, as this isn't so much a
10218 set operation as merely an internal storage change. */
10219 if (flags & SV_COW_DROP_PV) SvOK_off(sv);
10220 else sv_setsv_flags(sv, temp, 0);
10222 if ((const GV *)sv == PL_last_in_gv)
10223 PL_last_in_gv = NULL;
10224 else if ((const GV *)sv == PL_statgv)
10229 =for apidoc sv_unref_flags
10231 Unsets the RV status of the SV, and decrements the reference count of
10232 whatever was being referenced by the RV. This can almost be thought of
10233 as a reversal of C<newSVrv>. The C<cflags> argument can contain
10234 C<SV_IMMEDIATE_UNREF> to force the reference count to be decremented
10235 (otherwise the decrementing is conditional on the reference count being
10236 different from one or the reference being a readonly SV).
10243 Perl_sv_unref_flags(pTHX_ SV *const ref, const U32 flags)
10245 SV* const target = SvRV(ref);
10247 PERL_ARGS_ASSERT_SV_UNREF_FLAGS;
10249 if (SvWEAKREF(ref)) {
10250 sv_del_backref(target, ref);
10251 SvWEAKREF_off(ref);
10252 SvRV_set(ref, NULL);
10255 SvRV_set(ref, NULL);
10257 /* You can't have a || SvREADONLY(target) here, as $a = $$a, where $a was
10258 assigned to as BEGIN {$a = \"Foo"} will fail. */
10259 if (SvREFCNT(target) != 1 || (flags & SV_IMMEDIATE_UNREF))
10260 SvREFCNT_dec_NN(target);
10261 else /* XXX Hack, but hard to make $a=$a->[1] work otherwise */
10262 sv_2mortal(target); /* Schedule for freeing later */
10266 =for apidoc sv_untaint
10268 Untaint an SV. Use C<SvTAINTED_off> instead.
10274 Perl_sv_untaint(pTHX_ SV *const sv)
10276 PERL_ARGS_ASSERT_SV_UNTAINT;
10277 PERL_UNUSED_CONTEXT;
10279 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10280 MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10287 =for apidoc sv_tainted
10289 Test an SV for taintedness. Use C<SvTAINTED> instead.
10295 Perl_sv_tainted(pTHX_ SV *const sv)
10297 PERL_ARGS_ASSERT_SV_TAINTED;
10298 PERL_UNUSED_CONTEXT;
10300 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10301 const MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10302 if (mg && (mg->mg_len & 1) )
10309 =for apidoc sv_setpviv
10311 Copies an integer into the given SV, also updating its string value.
10312 Does not handle 'set' magic. See C<sv_setpviv_mg>.
10318 Perl_sv_setpviv(pTHX_ SV *const sv, const IV iv)
10320 char buf[TYPE_CHARS(UV)];
10322 char * const ptr = uiv_2buf(buf, iv, 0, 0, &ebuf);
10324 PERL_ARGS_ASSERT_SV_SETPVIV;
10326 sv_setpvn(sv, ptr, ebuf - ptr);
10330 =for apidoc sv_setpviv_mg
10332 Like C<sv_setpviv>, but also handles 'set' magic.
10338 Perl_sv_setpviv_mg(pTHX_ SV *const sv, const IV iv)
10340 PERL_ARGS_ASSERT_SV_SETPVIV_MG;
10342 sv_setpviv(sv, iv);
10346 #if defined(PERL_IMPLICIT_CONTEXT)
10348 /* pTHX_ magic can't cope with varargs, so this is a no-context
10349 * version of the main function, (which may itself be aliased to us).
10350 * Don't access this version directly.
10354 Perl_sv_setpvf_nocontext(SV *const sv, const char *const pat, ...)
10359 PERL_ARGS_ASSERT_SV_SETPVF_NOCONTEXT;
10361 va_start(args, pat);
10362 sv_vsetpvf(sv, pat, &args);
10366 /* pTHX_ magic can't cope with varargs, so this is a no-context
10367 * version of the main function, (which may itself be aliased to us).
10368 * Don't access this version directly.
10372 Perl_sv_setpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10377 PERL_ARGS_ASSERT_SV_SETPVF_MG_NOCONTEXT;
10379 va_start(args, pat);
10380 sv_vsetpvf_mg(sv, pat, &args);
10386 =for apidoc sv_setpvf
10388 Works like C<sv_catpvf> but copies the text into the SV instead of
10389 appending it. Does not handle 'set' magic. See C<sv_setpvf_mg>.
10395 Perl_sv_setpvf(pTHX_ SV *const sv, const char *const pat, ...)
10399 PERL_ARGS_ASSERT_SV_SETPVF;
10401 va_start(args, pat);
10402 sv_vsetpvf(sv, pat, &args);
10407 =for apidoc sv_vsetpvf
10409 Works like C<sv_vcatpvf> but copies the text into the SV instead of
10410 appending it. Does not handle 'set' magic. See C<sv_vsetpvf_mg>.
10412 Usually used via its frontend C<sv_setpvf>.
10418 Perl_sv_vsetpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10420 PERL_ARGS_ASSERT_SV_VSETPVF;
10422 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10426 =for apidoc sv_setpvf_mg
10428 Like C<sv_setpvf>, but also handles 'set' magic.
10434 Perl_sv_setpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10438 PERL_ARGS_ASSERT_SV_SETPVF_MG;
10440 va_start(args, pat);
10441 sv_vsetpvf_mg(sv, pat, &args);
10446 =for apidoc sv_vsetpvf_mg
10448 Like C<sv_vsetpvf>, but also handles 'set' magic.
10450 Usually used via its frontend C<sv_setpvf_mg>.
10456 Perl_sv_vsetpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10458 PERL_ARGS_ASSERT_SV_VSETPVF_MG;
10460 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10464 #if defined(PERL_IMPLICIT_CONTEXT)
10466 /* pTHX_ magic can't cope with varargs, so this is a no-context
10467 * version of the main function, (which may itself be aliased to us).
10468 * Don't access this version directly.
10472 Perl_sv_catpvf_nocontext(SV *const sv, const char *const pat, ...)
10477 PERL_ARGS_ASSERT_SV_CATPVF_NOCONTEXT;
10479 va_start(args, pat);
10480 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10484 /* pTHX_ magic can't cope with varargs, so this is a no-context
10485 * version of the main function, (which may itself be aliased to us).
10486 * Don't access this version directly.
10490 Perl_sv_catpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10495 PERL_ARGS_ASSERT_SV_CATPVF_MG_NOCONTEXT;
10497 va_start(args, pat);
10498 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10505 =for apidoc sv_catpvf
10507 Processes its arguments like C<sprintf> and appends the formatted
10508 output to an SV. If the appended data contains "wide" characters
10509 (including, but not limited to, SVs with a UTF-8 PV formatted with %s,
10510 and characters >255 formatted with %c), the original SV might get
10511 upgraded to UTF-8. Handles 'get' magic, but not 'set' magic. See
10512 C<sv_catpvf_mg>. If the original SV was UTF-8, the pattern should be
10513 valid UTF-8; if the original SV was bytes, the pattern should be too.
10518 Perl_sv_catpvf(pTHX_ SV *const sv, const char *const pat, ...)
10522 PERL_ARGS_ASSERT_SV_CATPVF;
10524 va_start(args, pat);
10525 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10530 =for apidoc sv_vcatpvf
10532 Processes its arguments like C<vsprintf> and appends the formatted output
10533 to an SV. Does not handle 'set' magic. See C<sv_vcatpvf_mg>.
10535 Usually used via its frontend C<sv_catpvf>.
10541 Perl_sv_vcatpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10543 PERL_ARGS_ASSERT_SV_VCATPVF;
10545 sv_vcatpvfn_flags(sv, pat, strlen(pat), args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10549 =for apidoc sv_catpvf_mg
10551 Like C<sv_catpvf>, but also handles 'set' magic.
10557 Perl_sv_catpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10561 PERL_ARGS_ASSERT_SV_CATPVF_MG;
10563 va_start(args, pat);
10564 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10570 =for apidoc sv_vcatpvf_mg
10572 Like C<sv_vcatpvf>, but also handles 'set' magic.
10574 Usually used via its frontend C<sv_catpvf_mg>.
10580 Perl_sv_vcatpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10582 PERL_ARGS_ASSERT_SV_VCATPVF_MG;
10584 sv_vcatpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10589 =for apidoc sv_vsetpvfn
10591 Works like C<sv_vcatpvfn> but copies the text into the SV instead of
10594 Usually used via one of its frontends C<sv_vsetpvf> and C<sv_vsetpvf_mg>.
10600 Perl_sv_vsetpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10601 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
10603 PERL_ARGS_ASSERT_SV_VSETPVFN;
10606 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, 0);
10611 * Warn of missing argument to sprintf, and then return a defined value
10612 * to avoid inappropriate "use of uninit" warnings [perl #71000].
10615 S_vcatpvfn_missing_argument(pTHX) {
10616 if (ckWARN(WARN_MISSING)) {
10617 Perl_warner(aTHX_ packWARN(WARN_MISSING), "Missing argument in %s",
10618 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
10625 S_expect_number(pTHX_ char **const pattern)
10629 PERL_ARGS_ASSERT_EXPECT_NUMBER;
10631 switch (**pattern) {
10632 case '1': case '2': case '3':
10633 case '4': case '5': case '6':
10634 case '7': case '8': case '9':
10635 var = *(*pattern)++ - '0';
10636 while (isDIGIT(**pattern)) {
10637 const I32 tmp = var * 10 + (*(*pattern)++ - '0');
10639 Perl_croak(aTHX_ "Integer overflow in format string for %s", (PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn"));
10647 S_F0convert(NV nv, char *const endbuf, STRLEN *const len)
10649 const int neg = nv < 0;
10652 PERL_ARGS_ASSERT_F0CONVERT;
10654 if (UNLIKELY(Perl_isinfnan(nv))) {
10655 STRLEN n = S_infnan_2pv(nv, endbuf - *len, *len);
10665 if (uv & 1 && uv == nv)
10666 uv--; /* Round to even */
10668 const unsigned dig = uv % 10;
10670 } while (uv /= 10);
10681 =for apidoc sv_vcatpvfn
10683 =for apidoc sv_vcatpvfn_flags
10685 Processes its arguments like C<vsprintf> and appends the formatted output
10686 to an SV. Uses an array of SVs if the C style variable argument list is
10687 missing (NULL). When running with taint checks enabled, indicates via
10688 C<maybe_tainted> if results are untrustworthy (often due to the use of
10691 If called as C<sv_vcatpvfn> or flags include C<SV_GMAGIC>, calls get magic.
10693 Usually used via one of its frontends C<sv_vcatpvf> and C<sv_vcatpvf_mg>.
10698 #define VECTORIZE_ARGS vecsv = va_arg(*args, SV*);\
10699 vecstr = (U8*)SvPV_const(vecsv,veclen);\
10700 vec_utf8 = DO_UTF8(vecsv);
10702 /* XXX maybe_tainted is never assigned to, so the doc above is lying. */
10705 Perl_sv_vcatpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10706 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
10708 PERL_ARGS_ASSERT_SV_VCATPVFN;
10710 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, SV_GMAGIC|SV_SMAGIC);
10713 #if DOUBLEKIND == DOUBLE_IS_IEEE_754_32_BIT_LITTLE_ENDIAN || \
10714 DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_LITTLE_ENDIAN || \
10715 DOUBLEKIND == DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN
10716 # define DOUBLE_LITTLE_ENDIAN
10719 #if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN || \
10720 LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN || \
10721 LONG_DOUBLEKIND == LONG_DOUBLE_IS_DOUBLEDOUBLE_128_BIT_LITTLE_ENDIAN
10722 # define LONGDOUBLE_LITTLE_ENDIAN
10725 #if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_BIG_ENDIAN || \
10726 LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN || \
10727 LONG_DOUBLEKIND == LONG_DOUBLE_IS_DOUBLEDOUBLE_128_BIT_BIG_ENDIAN
10728 # define LONGDOUBLE_BIG_ENDIAN
10731 #if LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN || \
10732 LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN
10733 # define LONGDOUBLE_X86_80_BIT
10736 #if LONG_DOUBLEKIND == LONG_DOUBLE_IS_DOUBLEDOUBLE_128_BIT_LITTLE_ENDIAN || \
10737 LONG_DOUBLEKIND == LONG_DOUBLE_IS_DOUBLEDOUBLE_128_BIT_BIG_ENDIAN
10738 # define LONGDOUBLE_DOUBLEDOUBLE
10739 /* The first double can be as large as 2**1023, or '1' x '0' x 1023.
10740 * The second double can be as small as 2**-1074, or '0' x 1073 . '1'.
10741 * The sum of them can be '1' . '0' x 2096 . '1', with implied radix point
10742 * after the first 1023 zero bits.
10744 * XXX The 2098 is quite large (262.25 bytes) and therefore some sort
10745 * of dynamically growing buffer might be better, start at just 16 bytes
10746 * (for example) and grow only when necessary. Or maybe just by looking
10747 * at the exponents of the two doubles? */
10748 # define DOUBLEDOUBLE_MAXBITS 2098
10751 /* vhex will contain the values (0..15) of the hex digits ("nybbles"
10752 * of 4 bits); 1 for the implicit 1, and the mantissa bits, four bits
10753 * per xdigit. For the double-double case, this can be rather many.
10754 * The non-double-double-long-double overshoots since all bits of NV
10755 * are not mantissa bits, there are also exponent bits. */
10756 #ifdef LONGDOUBLE_DOUBLEDOUBLE
10757 # define VHEX_SIZE (1+DOUBLEDOUBLE_MAXBITS/4)
10759 # define VHEX_SIZE (1+(NVSIZE * 8)/4)
10762 /* If we do not have a known long double format, (including not using
10763 * long doubles, or long doubles being equal to doubles) then we will
10764 * fall back to the ldexp/frexp route, with which we can retrieve at
10765 * most as many bits as our widest unsigned integer type is. We try
10766 * to get a 64-bit unsigned integer even if we are not using a 64-bit UV.
10768 * (If you want to test the case of UVSIZE == 4, NVSIZE == 8,
10769 * set the MANTISSATYPE to int and the MANTISSASIZE to 4.)
10771 #if defined(HAS_QUAD) && defined(Uquad_t)
10772 # define MANTISSATYPE Uquad_t
10773 # define MANTISSASIZE 8
10775 # define MANTISSATYPE UV
10776 # define MANTISSASIZE UVSIZE
10779 #if defined(DOUBLE_LITTLE_ENDIAN) || defined(LONGDOUBLE_LITTLE_ENDIAN)
10780 # define HEXTRACT_LITTLE_ENDIAN
10781 #elif defined(DOUBLE_BIG_ENDIAN) || defined(LONGDOUBLE_BIG_ENDIAN)
10782 # define HEXTRACT_BIG_ENDIAN
10784 # define HEXTRACT_MIX_ENDIAN
10787 /* S_hextract() is a helper for Perl_sv_vcatpvfn_flags, for extracting
10788 * the hexadecimal values (for %a/%A). The nv is the NV where the value
10789 * are being extracted from (either directly from the long double in-memory
10790 * presentation, or from the uquad computed via frexp+ldexp). frexp also
10791 * is used to update the exponent. vhex is the pointer to the beginning
10792 * of the output buffer (of VHEX_SIZE).
10794 * The tricky part is that S_hextract() needs to be called twice:
10795 * the first time with vend as NULL, and the second time with vend as
10796 * the pointer returned by the first call. What happens is that on
10797 * the first round the output size is computed, and the intended
10798 * extraction sanity checked. On the second round the actual output
10799 * (the extraction of the hexadecimal values) takes place.
10800 * Sanity failures cause fatal failures during both rounds. */
10802 S_hextract(pTHX_ const NV nv, int* exponent, U8* vhex, U8* vend)
10806 int ixmin = 0, ixmax = 0;
10808 /* XXX Inf/NaN/denormal handling in the HEXTRACT_IMPLICIT_BIT,
10809 * and elsewhere. */
10811 /* These macros are just to reduce typos, they have multiple
10812 * repetitions below, but usually only one (or sometimes two)
10813 * of them is really being used. */
10814 /* HEXTRACT_OUTPUT() extracts the high nybble first. */
10815 #define HEXTRACT_OUTPUT_HI(ix) (*v++ = nvp[ix] >> 4)
10816 #define HEXTRACT_OUTPUT_LO(ix) (*v++ = nvp[ix] & 0xF)
10817 #define HEXTRACT_OUTPUT(ix) \
10819 HEXTRACT_OUTPUT_HI(ix); HEXTRACT_OUTPUT_LO(ix); \
10821 #define HEXTRACT_COUNT(ix, c) \
10823 v += c; if (ix < ixmin) ixmin = ix; else if (ix > ixmax) ixmax = ix; \
10825 #define HEXTRACT_BYTE(ix) \
10827 if (vend) HEXTRACT_OUTPUT(ix); else HEXTRACT_COUNT(ix, 2); \
10829 #define HEXTRACT_LO_NYBBLE(ix) \
10831 if (vend) HEXTRACT_OUTPUT_LO(ix); else HEXTRACT_COUNT(ix, 1); \
10833 /* HEXTRACT_TOP_NYBBLE is just convenience disguise,
10834 * to make it look less odd when the top bits of a NV
10835 * are extracted using HEXTRACT_LO_NYBBLE: the highest
10836 * order bits can be in the "low nybble" of a byte. */
10837 #define HEXTRACT_TOP_NYBBLE(ix) HEXTRACT_LO_NYBBLE(ix)
10838 #define HEXTRACT_BYTES_LE(a, b) \
10839 for (ix = a; ix >= b; ix--) { HEXTRACT_BYTE(ix); }
10840 #define HEXTRACT_BYTES_BE(a, b) \
10841 for (ix = a; ix <= b; ix++) { HEXTRACT_BYTE(ix); }
10842 #define HEXTRACT_IMPLICIT_BIT(nv) \
10844 if (vend) *v++ = ((nv) == 0.0) ? 0 : 1; else v++; \
10847 /* Most formats do. Those which don't should undef this. */
10848 #define HEXTRACT_HAS_IMPLICIT_BIT
10849 /* Many formats do. Those which don't should undef this. */
10850 #define HEXTRACT_HAS_TOP_NYBBLE
10852 /* HEXTRACTSIZE is the maximum number of xdigits. */
10853 #if defined(USE_LONG_DOUBLE) && defined(LONGDOUBLE_DOUBLEDOUBLE)
10854 # define HEXTRACTSIZE (DOUBLEDOUBLE_MAXBITS/4)
10856 # define HEXTRACTSIZE 2 * NVSIZE
10859 const U8* vmaxend = vhex + HEXTRACTSIZE;
10860 PERL_UNUSED_VAR(ix); /* might happen */
10861 (void)Perl_frexp(PERL_ABS(nv), exponent);
10862 if (vend && (vend <= vhex || vend > vmaxend))
10863 Perl_croak(aTHX_ "Hexadecimal float: internal error");
10865 /* First check if using long doubles. */
10866 #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE)
10867 # if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN
10868 /* Used in e.g. VMS and HP-UX IA-64, e.g. -0.1L:
10869 * 9a 99 99 99 99 99 99 99 99 99 99 99 99 99 fb 3f */
10870 /* The bytes 13..0 are the mantissa/fraction,
10871 * the 15,14 are the sign+exponent. */
10872 const U8* nvp = (const U8*)(&nv);
10873 HEXTRACT_IMPLICIT_BIT(nv);
10874 # undef HEXTRACT_HAS_TOP_NYBBLE
10875 HEXTRACT_BYTES_LE(13, 0);
10876 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_BIG_ENDIAN
10877 /* Used in e.g. Solaris Sparc and HP-UX PA-RISC, e.g. -0.1L:
10878 * bf fb 99 99 99 99 99 99 99 99 99 99 99 99 99 9a */
10879 /* The bytes 2..15 are the mantissa/fraction,
10880 * the 0,1 are the sign+exponent. */
10881 const U8* nvp = (const U8*)(&nv);
10882 HEXTRACT_IMPLICIT_BIT(nv);
10883 # undef HEXTRACT_HAS_TOP_NYBBLE
10884 HEXTRACT_BYTES_BE(2, 15);
10885 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN
10886 /* x86 80-bit "extended precision", 64 bits of mantissa / fraction /
10887 * significand, 15 bits of exponent, 1 bit of sign. NVSIZE can
10888 * be either 12 (ILP32, Solaris x86) or 16 (LP64, Linux and OS X),
10889 * meaning that 2 or 6 bytes are empty padding. */
10890 /* The bytes 7..0 are the mantissa/fraction */
10891 const U8* nvp = (const U8*)(&nv);
10892 # undef HEXTRACT_HAS_IMPLICIT_BIT
10893 # undef HEXTRACT_HAS_TOP_NYBBLE
10894 HEXTRACT_BYTES_LE(7, 0);
10895 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN
10896 /* Does this format ever happen? (Wikipedia says the Motorola
10897 * 6888x math coprocessors used format _like_ this but padded
10898 * to 96 bits with 16 unused bits between the exponent and the
10900 const U8* nvp = (const U8*)(&nv);
10901 # undef HEXTRACT_HAS_IMPLICIT_BIT
10902 # undef HEXTRACT_HAS_TOP_NYBBLE
10903 HEXTRACT_BYTES_BE(0, 7);
10905 # define HEXTRACT_FALLBACK
10906 /* Double-double format: two doubles next to each other.
10907 * The first double is the high-order one, exactly like
10908 * it would be for a "lone" double. The second double
10909 * is shifted down using the exponent so that that there
10910 * are no common bits. The tricky part is that the value
10911 * of the double-double is the SUM of the two doubles and
10912 * the second one can be also NEGATIVE.
10914 * Because of this tricky construction the bytewise extraction we
10915 * use for the other long double formats doesn't work, we must
10916 * extract the values bit by bit.
10918 * The little-endian double-double is used .. somewhere?
10920 * The big endian double-double is used in e.g. PPC/Power (AIX)
10923 * The mantissa bits are in two separate stretches, e.g. for -0.1L:
10924 * 9a 99 99 99 99 99 59 bc 9a 99 99 99 99 99 b9 3f (LE)
10925 * 3f b9 99 99 99 99 99 9a bc 59 99 99 99 99 99 9a (BE)
10928 #else /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) */
10929 /* Using normal doubles, not long doubles.
10931 * We generate 4-bit xdigits (nybble/nibble) instead of 8-bit
10932 * bytes, since we might need to handle printf precision, and
10933 * also need to insert the radix. */
10935 # ifdef HEXTRACT_LITTLE_ENDIAN
10936 /* 0 1 2 3 4 5 6 7 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
10937 const U8* nvp = (const U8*)(&nv);
10938 HEXTRACT_IMPLICIT_BIT(nv);
10939 HEXTRACT_TOP_NYBBLE(6);
10940 HEXTRACT_BYTES_LE(5, 0);
10941 # elif defined(HEXTRACT_BIG_ENDIAN)
10942 /* 7 6 5 4 3 2 1 0 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
10943 const U8* nvp = (const U8*)(&nv);
10944 HEXTRACT_IMPLICIT_BIT(nv);
10945 HEXTRACT_TOP_NYBBLE(1);
10946 HEXTRACT_BYTES_BE(2, 7);
10947 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_LE_BE
10948 /* 4 5 6 7 0 1 2 3 (MSB = 7, LSB = 0, 6:7 = nybble:exponent:sign) */
10949 const U8* nvp = (const U8*)(&nv);
10950 HEXTRACT_IMPLICIT_BIT(nv);
10951 HEXTRACT_TOP_NYBBLE(2); /* 6 */
10952 HEXTRACT_BYTE(1); /* 5 */
10953 HEXTRACT_BYTE(0); /* 4 */
10954 HEXTRACT_BYTE(7); /* 3 */
10955 HEXTRACT_BYTE(6); /* 2 */
10956 HEXTRACT_BYTE(5); /* 1 */
10957 HEXTRACT_BYTE(4); /* 0 */
10958 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_BE_LE
10959 /* 3 2 1 0 7 6 5 4 (MSB = 7, LSB = 0, 7:6 = sign:exponent:nybble) */
10960 const U8* nvp = (const U8*)(&nv);
10961 HEXTRACT_IMPLICIT_BIT(nv);
10962 HEXTRACT_TOP_NYBBLE(5); /* 6 */
10963 HEXTRACT_BYTE(6); /* 5 */
10964 HEXTRACT_BYTE(7); /* 4 */
10965 HEXTRACT_BYTE(0); /* 3 */
10966 HEXTRACT_BYTE(1); /* 2 */
10967 HEXTRACT_BYTE(2); /* 1 */
10968 HEXTRACT_BYTE(3); /* 0 */
10970 # define HEXTRACT_FALLBACK
10973 # define HEXTRACT_FALLBACK
10975 #endif /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) #else */
10976 # ifdef HEXTRACT_FALLBACK
10977 # undef HEXTRACT_HAS_TOP_NYBBLE /* Meaningless, but consistent. */
10978 /* The fallback is used for the double-double format, and
10979 * for unknown long double formats, and for unknown double
10980 * formats, or in general unknown NV formats. */
10981 if (nv == (NV)0.0) {
10989 NV d = nv < 0 ? -nv : nv;
10991 U8 ha = 0x0; /* hexvalue accumulator */
10992 U8 hd = 0x8; /* hexvalue digit */
10994 /* Shift d and e (and update exponent) so that e <= d < 2*e,
10995 * this is essentially manual frexp(). Multiplying by 0.5 and
10996 * doubling should be lossless in binary floating point. */
11006 while (d >= e + e) {
11010 /* Now e <= d < 2*e */
11012 /* First extract the leading hexdigit (the implicit bit). */
11028 /* Then extract the remaining hexdigits. */
11029 while (d > (NV)0.0) {
11035 /* Output or count in groups of four bits,
11036 * that is, when the hexdigit is down to one. */
11041 /* Reset the hexvalue. */
11050 /* Flush possible pending hexvalue. */
11060 /* Croak for various reasons: if the output pointer escaped the
11061 * output buffer, if the extraction index escaped the extraction
11062 * buffer, or if the ending output pointer didn't match the
11063 * previously computed value. */
11064 if (v <= vhex || v - vhex >= VHEX_SIZE ||
11065 /* For double-double the ixmin and ixmax stay at zero,
11066 * which is convenient since the HEXTRACTSIZE is tricky
11067 * for double-double. */
11068 ixmin < 0 || ixmax >= NVSIZE ||
11069 (vend && v != vend))
11070 Perl_croak(aTHX_ "Hexadecimal float: internal error");
11075 Perl_sv_vcatpvfn_flags(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11076 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted,
11081 const char *patend;
11084 static const char nullstr[] = "(null)";
11086 bool has_utf8 = DO_UTF8(sv); /* has the result utf8? */
11087 const bool pat_utf8 = has_utf8; /* the pattern is in utf8? */
11089 /* Times 4: a decimal digit takes more than 3 binary digits.
11090 * NV_DIG: mantissa takes than many decimal digits.
11091 * Plus 32: Playing safe. */
11092 char ebuf[IV_DIG * 4 + NV_DIG + 32];
11093 bool no_redundant_warning = FALSE; /* did we use any explicit format parameter index? */
11094 bool hexfp = FALSE; /* hexadecimal floating point? */
11096 DECLARATION_FOR_STORE_LC_NUMERIC_SET_TO_NEEDED;
11098 PERL_ARGS_ASSERT_SV_VCATPVFN_FLAGS;
11099 PERL_UNUSED_ARG(maybe_tainted);
11101 if (flags & SV_GMAGIC)
11104 /* no matter what, this is a string now */
11105 (void)SvPV_force_nomg(sv, origlen);
11107 /* special-case "", "%s", and "%-p" (SVf - see below) */
11109 if (svmax && ckWARN(WARN_REDUNDANT))
11110 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11111 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11114 if (patlen == 2 && pat[0] == '%' && pat[1] == 's') {
11115 if (svmax > 1 && ckWARN(WARN_REDUNDANT))
11116 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11117 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11120 const char * const s = va_arg(*args, char*);
11121 sv_catpv_nomg(sv, s ? s : nullstr);
11123 else if (svix < svmax) {
11124 /* we want get magic on the source but not the target. sv_catsv can't do that, though */
11125 SvGETMAGIC(*svargs);
11126 sv_catsv_nomg(sv, *svargs);
11129 S_vcatpvfn_missing_argument(aTHX);
11132 if (args && patlen == 3 && pat[0] == '%' &&
11133 pat[1] == '-' && pat[2] == 'p') {
11134 if (svmax > 1 && ckWARN(WARN_REDUNDANT))
11135 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
11136 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11137 argsv = MUTABLE_SV(va_arg(*args, void*));
11138 sv_catsv_nomg(sv, argsv);
11142 #if !defined(USE_LONG_DOUBLE) && !defined(USE_QUADMATH)
11143 /* special-case "%.<number>[gf]" */
11144 if ( !args && patlen <= 5 && pat[0] == '%' && pat[1] == '.'
11145 && (pat[patlen-1] == 'g' || pat[patlen-1] == 'f') ) {
11146 unsigned digits = 0;
11150 while (*pp >= '0' && *pp <= '9')
11151 digits = 10 * digits + (*pp++ - '0');
11153 /* XXX: Why do this `svix < svmax` test? Couldn't we just
11154 format the first argument and WARN_REDUNDANT if svmax > 1?
11155 Munged by Nicholas Clark in v5.13.0-209-g95ea86d */
11156 if (pp - pat == (int)patlen - 1 && svix < svmax) {
11157 const NV nv = SvNV(*svargs);
11158 if (LIKELY(!Perl_isinfnan(nv))) {
11160 /* Add check for digits != 0 because it seems that some
11161 gconverts are buggy in this case, and we don't yet have
11162 a Configure test for this. */
11163 if (digits && digits < sizeof(ebuf) - NV_DIG - 10) {
11164 /* 0, point, slack */
11165 STORE_LC_NUMERIC_SET_TO_NEEDED();
11166 SNPRINTF_G(nv, ebuf, size, digits);
11167 sv_catpv_nomg(sv, ebuf);
11168 if (*ebuf) /* May return an empty string for digits==0 */
11171 } else if (!digits) {
11174 if ((p = F0convert(nv, ebuf + sizeof ebuf, &l))) {
11175 sv_catpvn_nomg(sv, p, l);
11182 #endif /* !USE_LONG_DOUBLE */
11184 if (!args && svix < svmax && DO_UTF8(*svargs))
11187 patend = (char*)pat + patlen;
11188 for (p = (char*)pat; p < patend; p = q) {
11191 bool vectorize = FALSE;
11192 bool vectorarg = FALSE;
11193 bool vec_utf8 = FALSE;
11199 bool has_precis = FALSE;
11201 const I32 osvix = svix;
11202 bool is_utf8 = FALSE; /* is this item utf8? */
11203 #ifdef HAS_LDBL_SPRINTF_BUG
11204 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
11205 with sfio - Allen <allens@cpan.org> */
11206 bool fix_ldbl_sprintf_bug = FALSE;
11210 U8 utf8buf[UTF8_MAXBYTES+1];
11211 STRLEN esignlen = 0;
11213 const char *eptr = NULL;
11214 const char *fmtstart;
11217 const U8 *vecstr = NULL;
11224 /* We need a long double target in case HAS_LONG_DOUBLE,
11225 * even without USE_LONG_DOUBLE, so that we can printf with
11226 * long double formats, even without NV being long double.
11227 * But we call the target 'fv' instead of 'nv', since most of
11228 * the time it is not (most compilers these days recognize
11229 * "long double", even if only as a synonym for "double").
11231 #if defined(HAS_LONG_DOUBLE) && LONG_DOUBLESIZE > DOUBLESIZE && \
11232 defined(PERL_PRIgldbl) && !defined(USE_QUADMATH)
11234 # define FV_ISFINITE(x) Perl_isfinitel(x)
11235 # define FV_GF PERL_PRIgldbl
11236 # if defined(__VMS) && defined(__ia64) && defined(__IEEE_FLOAT)
11237 /* Work around breakage in OTS$CVT_FLOAT_T_X */
11238 # define NV_TO_FV(nv,fv) STMT_START { \
11240 fv = Perl_isnan(_dv) ? LDBL_QNAN : _dv; \
11243 # define NV_TO_FV(nv,fv) (fv)=(nv)
11247 # define FV_ISFINITE(x) Perl_isfinite((NV)(x))
11248 # define FV_GF NVgf
11249 # define NV_TO_FV(nv,fv) (fv)=(nv)
11254 const char *dotstr = ".";
11255 STRLEN dotstrlen = 1;
11256 I32 efix = 0; /* explicit format parameter index */
11257 I32 ewix = 0; /* explicit width index */
11258 I32 epix = 0; /* explicit precision index */
11259 I32 evix = 0; /* explicit vector index */
11260 bool asterisk = FALSE;
11261 bool infnan = FALSE;
11263 /* echo everything up to the next format specification */
11264 for (q = p; q < patend && *q != '%'; ++q) ;
11266 if (has_utf8 && !pat_utf8)
11267 sv_catpvn_nomg_utf8_upgrade(sv, p, q - p, nsv);
11269 sv_catpvn_nomg(sv, p, q - p);
11278 We allow format specification elements in this order:
11279 \d+\$ explicit format parameter index
11281 v|\*(\d+\$)?v vector with optional (optionally specified) arg
11282 0 flag (as above): repeated to allow "v02"
11283 \d+|\*(\d+\$)? width using optional (optionally specified) arg
11284 \.(\d*|\*(\d+\$)?) precision using optional (optionally specified) arg
11286 [%bcdefginopsuxDFOUX] format (mandatory)
11291 As of perl5.9.3, printf format checking is on by default.
11292 Internally, perl uses %p formats to provide an escape to
11293 some extended formatting. This block deals with those
11294 extensions: if it does not match, (char*)q is reset and
11295 the normal format processing code is used.
11297 Currently defined extensions are:
11298 %p include pointer address (standard)
11299 %-p (SVf) include an SV (previously %_)
11300 %-<num>p include an SV with precision <num>
11302 %3p include a HEK with precision of 256
11303 %4p char* preceded by utf8 flag and length
11304 %<num>p (where num is 1 or > 4) reserved for future
11307 Robin Barker 2005-07-14 (but modified since)
11309 %1p (VDf) removed. RMB 2007-10-19
11316 else if (strnEQ(q, UTF8f, sizeof(UTF8f)-1)) { /* UTF8f */
11317 /* The argument has already gone through cBOOL, so the cast
11319 is_utf8 = (bool)va_arg(*args, int);
11320 elen = va_arg(*args, UV);
11321 if ((IV)elen < 0) {
11322 /* check if utf8 length is larger than 0 when cast to IV */
11323 assert( (IV)elen >= 0 ); /* in DEBUGGING build we want to crash */
11324 elen= 0; /* otherwise we want to treat this as an empty string */
11326 eptr = va_arg(*args, char *);
11327 q += sizeof(UTF8f)-1;
11330 n = expect_number(&q);
11332 if (sv) { /* SVf */
11337 argsv = MUTABLE_SV(va_arg(*args, void*));
11338 eptr = SvPV_const(argsv, elen);
11339 if (DO_UTF8(argsv))
11343 else if (n==2 || n==3) { /* HEKf */
11344 HEK * const hek = va_arg(*args, HEK *);
11345 eptr = HEK_KEY(hek);
11346 elen = HEK_LEN(hek);
11347 if (HEK_UTF8(hek)) is_utf8 = TRUE;
11348 if (n==3) precis = 256, has_precis = TRUE;
11352 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
11353 "internal %%<num>p might conflict with future printf extensions");
11359 if ( (width = expect_number(&q)) ) {
11363 if (!no_redundant_warning)
11364 /* I've forgotten if it's a better
11365 micro-optimization to always set this or to
11366 only set it if it's unset */
11367 no_redundant_warning = TRUE;
11379 if (plus == '+' && *q == ' ') /* '+' over ' ' */
11408 if ( (ewix = expect_number(&q)) )
11417 if ((vectorarg = asterisk)) {
11430 width = expect_number(&q);
11433 if (vectorize && vectorarg) {
11434 /* vectorizing, but not with the default "." */
11436 vecsv = va_arg(*args, SV*);
11438 vecsv = (evix > 0 && evix <= svmax)
11439 ? svargs[evix-1] : S_vcatpvfn_missing_argument(aTHX);
11441 vecsv = svix < svmax
11442 ? svargs[svix++] : S_vcatpvfn_missing_argument(aTHX);
11444 dotstr = SvPV_const(vecsv, dotstrlen);
11445 /* Keep the DO_UTF8 test *after* the SvPV call, else things go
11446 bad with tied or overloaded values that return UTF8. */
11447 if (DO_UTF8(vecsv))
11449 else if (has_utf8) {
11450 vecsv = sv_mortalcopy(vecsv);
11451 sv_utf8_upgrade(vecsv);
11452 dotstr = SvPV_const(vecsv, dotstrlen);
11459 i = va_arg(*args, int);
11461 i = (ewix ? ewix <= svmax : svix < svmax) ?
11462 SvIVx(svargs[ewix ? ewix-1 : svix++]) : 0;
11464 width = (i < 0) ? -i : i;
11474 if ( ((epix = expect_number(&q))) && (*q++ != '$') )
11476 /* XXX: todo, support specified precision parameter */
11480 i = va_arg(*args, int);
11482 i = (ewix ? ewix <= svmax : svix < svmax)
11483 ? SvIVx(svargs[ewix ? ewix-1 : svix++]) : 0;
11485 has_precis = !(i < 0);
11489 while (isDIGIT(*q))
11490 precis = precis * 10 + (*q++ - '0');
11499 else if (efix ? (efix > 0 && efix <= svmax) : svix < svmax) {
11500 vecsv = svargs[efix ? efix-1 : svix++];
11501 vecstr = (U8*)SvPV_const(vecsv,veclen);
11502 vec_utf8 = DO_UTF8(vecsv);
11504 /* if this is a version object, we need to convert
11505 * back into v-string notation and then let the
11506 * vectorize happen normally
11508 if (sv_isobject(vecsv) && sv_derived_from(vecsv, "version")) {
11509 if ( hv_exists(MUTABLE_HV(SvRV(vecsv)), "alpha", 5 ) ) {
11510 Perl_ck_warner_d(aTHX_ packWARN(WARN_PRINTF),
11511 "vector argument not supported with alpha versions");
11514 vecsv = sv_newmortal();
11515 scan_vstring((char *)vecstr, (char *)vecstr + veclen,
11517 vecstr = (U8*)SvPV_const(vecsv, veclen);
11518 vec_utf8 = DO_UTF8(vecsv);
11532 case 'I': /* Ix, I32x, and I64x */
11533 # ifdef USE_64_BIT_INT
11534 if (q[1] == '6' && q[2] == '4') {
11540 if (q[1] == '3' && q[2] == '2') {
11544 # ifdef USE_64_BIT_INT
11550 #if IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)
11553 #ifdef USE_QUADMATH
11566 #if IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)
11567 if (*q == 'l') { /* lld, llf */
11576 if (*++q == 'h') { /* hhd, hhu */
11605 if (!vectorize && !args) {
11607 const I32 i = efix-1;
11608 argsv = (i >= 0 && i < svmax)
11609 ? svargs[i] : S_vcatpvfn_missing_argument(aTHX);
11611 argsv = (svix >= 0 && svix < svmax)
11612 ? svargs[svix++] : S_vcatpvfn_missing_argument(aTHX);
11616 if (argsv && strchr("BbcDdiOopuUXx",*q)) {
11617 /* XXX va_arg(*args) case? need peek, use va_copy? */
11619 infnan = UNLIKELY(isinfnansv(argsv));
11622 switch (c = *q++) {
11630 Perl_croak(aTHX_ "Cannot printf %"NVgf" with '%c'",
11631 /* no va_arg() case */
11632 SvNV_nomg(argsv), (int)c);
11633 uv = (args) ? va_arg(*args, int) : SvIV_nomg(argsv);
11635 (!UVCHR_IS_INVARIANT(uv) && SvUTF8(sv)))
11637 eptr = (char*)utf8buf;
11638 elen = uvchr_to_utf8((U8*)eptr, uv) - utf8buf;
11652 eptr = va_arg(*args, char*);
11654 elen = strlen(eptr);
11656 eptr = (char *)nullstr;
11657 elen = sizeof nullstr - 1;
11661 eptr = SvPV_const(argsv, elen);
11662 if (DO_UTF8(argsv)) {
11663 STRLEN old_precis = precis;
11664 if (has_precis && precis < elen) {
11665 STRLEN ulen = sv_or_pv_len_utf8(argsv, eptr, elen);
11666 STRLEN p = precis > ulen ? ulen : precis;
11667 precis = sv_or_pv_pos_u2b(argsv, eptr, p, 0);
11668 /* sticks at end */
11670 if (width) { /* fudge width (can't fudge elen) */
11671 if (has_precis && precis < elen)
11672 width += precis - old_precis;
11675 elen - sv_or_pv_len_utf8(argsv,eptr,elen);
11682 if (has_precis && precis < elen)
11690 goto floating_point;
11692 if (alt || vectorize)
11694 uv = PTR2UV(args ? va_arg(*args, void*) : argsv);
11708 goto floating_point;
11715 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
11724 esignbuf[esignlen++] = plus;
11728 case 'c': iv = (char)va_arg(*args, int); break;
11729 case 'h': iv = (short)va_arg(*args, int); break;
11730 case 'l': iv = va_arg(*args, long); break;
11731 case 'V': iv = va_arg(*args, IV); break;
11732 case 'z': iv = va_arg(*args, SSize_t); break;
11733 #ifdef HAS_PTRDIFF_T
11734 case 't': iv = va_arg(*args, ptrdiff_t); break;
11736 default: iv = va_arg(*args, int); break;
11738 case 'j': iv = va_arg(*args, intmax_t); break;
11742 iv = va_arg(*args, Quad_t); break;
11749 IV tiv = SvIV_nomg(argsv); /* work around GCC bug #13488 */
11751 case 'c': iv = (char)tiv; break;
11752 case 'h': iv = (short)tiv; break;
11753 case 'l': iv = (long)tiv; break;
11755 default: iv = tiv; break;
11758 iv = (Quad_t)tiv; break;
11764 if ( !vectorize ) /* we already set uv above */
11769 esignbuf[esignlen++] = plus;
11773 esignbuf[esignlen++] = '-';
11812 goto floating_point;
11820 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
11831 case 'c': uv = (unsigned char)va_arg(*args, unsigned); break;
11832 case 'h': uv = (unsigned short)va_arg(*args, unsigned); break;
11833 case 'l': uv = va_arg(*args, unsigned long); break;
11834 case 'V': uv = va_arg(*args, UV); break;
11835 case 'z': uv = va_arg(*args, Size_t); break;
11836 #ifdef HAS_PTRDIFF_T
11837 case 't': uv = va_arg(*args, ptrdiff_t); break; /* will sign extend, but there is no uptrdiff_t, so oh well */
11840 case 'j': uv = va_arg(*args, uintmax_t); break;
11842 default: uv = va_arg(*args, unsigned); break;
11845 uv = va_arg(*args, Uquad_t); break;
11852 UV tuv = SvUV_nomg(argsv); /* work around GCC bug #13488 */
11854 case 'c': uv = (unsigned char)tuv; break;
11855 case 'h': uv = (unsigned short)tuv; break;
11856 case 'l': uv = (unsigned long)tuv; break;
11858 default: uv = tuv; break;
11861 uv = (Uquad_t)tuv; break;
11870 char *ptr = ebuf + sizeof ebuf;
11871 bool tempalt = uv ? alt : FALSE; /* Vectors can't change alt */
11877 p = (char *)((c == 'X') ? PL_hexdigit + 16 : PL_hexdigit);
11881 } while (uv >>= 4);
11883 esignbuf[esignlen++] = '0';
11884 esignbuf[esignlen++] = c; /* 'x' or 'X' */
11890 *--ptr = '0' + dig;
11891 } while (uv >>= 3);
11892 if (alt && *ptr != '0')
11898 *--ptr = '0' + dig;
11899 } while (uv >>= 1);
11901 esignbuf[esignlen++] = '0';
11902 esignbuf[esignlen++] = c;
11905 default: /* it had better be ten or less */
11908 *--ptr = '0' + dig;
11909 } while (uv /= base);
11912 elen = (ebuf + sizeof ebuf) - ptr;
11916 zeros = precis - elen;
11917 else if (precis == 0 && elen == 1 && *eptr == '0'
11918 && !(base == 8 && alt)) /* "%#.0o" prints "0" */
11921 /* a precision nullifies the 0 flag. */
11928 /* FLOATING POINT */
11933 c = 'f'; /* maybe %F isn't supported here */
11935 case 'e': case 'E':
11937 case 'g': case 'G':
11938 case 'a': case 'A':
11942 /* This is evil, but floating point is even more evil */
11944 /* for SV-style calling, we can only get NV
11945 for C-style calling, we assume %f is double;
11946 for simplicity we allow any of %Lf, %llf, %qf for long double
11950 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
11954 /* [perl #20339] - we should accept and ignore %lf rather than die */
11958 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
11959 intsize = args ? 0 : 'q';
11963 #if defined(HAS_LONG_DOUBLE)
11976 /* Now we need (long double) if intsize == 'q', else (double). */
11978 /* Note: do not pull NVs off the va_list with va_arg()
11979 * (pull doubles instead) because if you have a build
11980 * with long doubles, you would always be pulling long
11981 * doubles, which would badly break anyone using only
11982 * doubles (i.e. the majority of builds). In other
11983 * words, you cannot mix doubles and long doubles.
11984 * The only case where you can pull off long doubles
11985 * is when the format specifier explicitly asks so with
11987 #ifdef USE_QUADMATH
11988 fv = intsize == 'q' ?
11989 va_arg(*args, NV) : va_arg(*args, double);
11990 #elif LONG_DOUBLESIZE > DOUBLESIZE
11991 if (intsize == 'q')
11992 fv = va_arg(*args, long double);
11994 NV_TO_FV(va_arg(*args, double), fv);
11996 fv = va_arg(*args, double);
12001 if (!infnan) SvGETMAGIC(argsv);
12002 NV_TO_FV(SvNV_nomg(argsv), fv);
12006 /* frexp() (or frexpl) has some unspecified behaviour for
12007 * nan/inf/-inf, so let's avoid calling that on non-finites. */
12008 if (isALPHA_FOLD_NE(c, 'e') && FV_ISFINITE(fv)) {
12010 (void)Perl_frexp((NV)fv, &i);
12011 if (i == PERL_INT_MIN)
12012 Perl_die(aTHX_ "panic: frexp: %"FV_GF, fv);
12013 /* Do not set hexfp earlier since we want to printf
12014 * Inf/NaN for Inf/NaN, not their hexfp. */
12015 hexfp = isALPHA_FOLD_EQ(c, 'a');
12016 if (UNLIKELY(hexfp)) {
12017 /* This seriously overshoots in most cases, but
12018 * better the undershooting. Firstly, all bytes
12019 * of the NV are not mantissa, some of them are
12020 * exponent. Secondly, for the reasonably common
12021 * long doubles case, the "80-bit extended", two
12022 * or six bytes of the NV are unused. */
12024 (fv < 0) ? 1 : 0 + /* possible unary minus */
12026 1 + /* the very unlikely carry */
12029 2 * NVSIZE + /* 2 hexdigits for each byte */
12031 6 + /* exponent: sign, plus up to 16383 (quad fp) */
12033 #ifdef LONGDOUBLE_DOUBLEDOUBLE
12034 /* However, for the "double double", we need more.
12035 * Since each double has their own exponent, the
12036 * doubles may float (haha) rather far from each
12037 * other, and the number of required bits is much
12038 * larger, up to total of DOUBLEDOUBLE_MAXBITS bits.
12039 * See the definition of DOUBLEDOUBLE_MAXBITS.
12041 * Need 2 hexdigits for each byte. */
12042 need += (DOUBLEDOUBLE_MAXBITS/8 + 1) * 2;
12043 /* the size for the exponent already added */
12045 #ifdef USE_LOCALE_NUMERIC
12046 STORE_LC_NUMERIC_SET_TO_NEEDED();
12047 if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC))
12048 need += SvLEN(PL_numeric_radix_sv);
12049 RESTORE_LC_NUMERIC();
12053 need = BIT_DIGITS(i);
12054 } /* if i < 0, the number of digits is hard to predict. */
12056 need += has_precis ? precis : 6; /* known default */
12061 #ifdef HAS_LDBL_SPRINTF_BUG
12062 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
12063 with sfio - Allen <allens@cpan.org> */
12066 # define MY_DBL_MAX DBL_MAX
12067 # else /* XXX guessing! HUGE_VAL may be defined as infinity, so not using */
12068 # if DOUBLESIZE >= 8
12069 # define MY_DBL_MAX 1.7976931348623157E+308L
12071 # define MY_DBL_MAX 3.40282347E+38L
12075 # ifdef HAS_LDBL_SPRINTF_BUG_LESS1 /* only between -1L & 1L - Allen */
12076 # define MY_DBL_MAX_BUG 1L
12078 # define MY_DBL_MAX_BUG MY_DBL_MAX
12082 # define MY_DBL_MIN DBL_MIN
12083 # else /* XXX guessing! -Allen */
12084 # if DOUBLESIZE >= 8
12085 # define MY_DBL_MIN 2.2250738585072014E-308L
12087 # define MY_DBL_MIN 1.17549435E-38L
12091 if ((intsize == 'q') && (c == 'f') &&
12092 ((fv < MY_DBL_MAX_BUG) && (fv > -MY_DBL_MAX_BUG)) &&
12093 (need < DBL_DIG)) {
12094 /* it's going to be short enough that
12095 * long double precision is not needed */
12097 if ((fv <= 0L) && (fv >= -0L))
12098 fix_ldbl_sprintf_bug = TRUE; /* 0 is 0 - easiest */
12100 /* would use Perl_fp_class as a double-check but not
12101 * functional on IRIX - see perl.h comments */
12103 if ((fv >= MY_DBL_MIN) || (fv <= -MY_DBL_MIN)) {
12104 /* It's within the range that a double can represent */
12105 #if defined(DBL_MAX) && !defined(DBL_MIN)
12106 if ((fv >= ((long double)1/DBL_MAX)) ||
12107 (fv <= (-(long double)1/DBL_MAX)))
12109 fix_ldbl_sprintf_bug = TRUE;
12112 if (fix_ldbl_sprintf_bug == TRUE) {
12122 # undef MY_DBL_MAX_BUG
12125 #endif /* HAS_LDBL_SPRINTF_BUG */
12127 need += 20; /* fudge factor */
12128 if (PL_efloatsize < need) {
12129 Safefree(PL_efloatbuf);
12130 PL_efloatsize = need + 20; /* more fudge */
12131 Newx(PL_efloatbuf, PL_efloatsize, char);
12132 PL_efloatbuf[0] = '\0';
12135 if ( !(width || left || plus || alt) && fill != '0'
12136 && has_precis && intsize != 'q' /* Shortcuts */
12137 && LIKELY(!Perl_isinfnan((NV)fv)) ) {
12138 /* See earlier comment about buggy Gconvert when digits,
12140 if ( c == 'g' && precis ) {
12141 STORE_LC_NUMERIC_SET_TO_NEEDED();
12142 SNPRINTF_G(fv, PL_efloatbuf, PL_efloatsize, precis);
12143 /* May return an empty string for digits==0 */
12144 if (*PL_efloatbuf) {
12145 elen = strlen(PL_efloatbuf);
12146 goto float_converted;
12148 } else if ( c == 'f' && !precis ) {
12149 if ((eptr = F0convert(fv, ebuf + sizeof ebuf, &elen)))
12154 if (UNLIKELY(hexfp)) {
12155 /* Hexadecimal floating point. */
12156 char* p = PL_efloatbuf;
12157 U8 vhex[VHEX_SIZE];
12158 U8* v = vhex; /* working pointer to vhex */
12159 U8* vend; /* pointer to one beyond last digit of vhex */
12160 U8* vfnz = NULL; /* first non-zero */
12161 const bool lower = (c == 'a');
12162 /* At output the values of vhex (up to vend) will
12163 * be mapped through the xdig to get the actual
12164 * human-readable xdigits. */
12165 const char* xdig = PL_hexdigit;
12166 int zerotail = 0; /* how many extra zeros to append */
12167 int exponent = 0; /* exponent of the floating point input */
12169 /* XXX: denormals, NaN, Inf.
12171 * For example with denormals, (assuming the vanilla
12172 * 64-bit double): the exponent is zero. 1xp-1074 is
12173 * the smallest denormal and the smallest double, it
12174 * should be output as 0x0.0000000000001p-1022 to
12175 * match its internal structure. */
12177 /* Note: fv can be (and often is) long double.
12178 * Here it is explicitly cast to NV. */
12179 vend = S_hextract(aTHX_ (NV)fv, &exponent, vhex, NULL);
12180 S_hextract(aTHX_ (NV)fv, &exponent, vhex, vend);
12182 #if NVSIZE > DOUBLESIZE
12183 # ifdef HEXTRACT_HAS_IMPLICIT_BIT
12184 /* In this case there is an implicit bit,
12185 * and therefore the exponent is shifted shift by one. */
12188 /* In this case there is no implicit bit,
12189 * and the exponent is shifted by the first xdigit. */
12204 xdig += 16; /* Use uppercase hex. */
12207 /* Find the first non-zero xdigit. */
12208 for (v = vhex; v < vend; v++) {
12216 U8* vlnz = NULL; /* The last non-zero. */
12218 /* Find the last non-zero xdigit. */
12219 for (v = vend - 1; v >= vhex; v--) {
12226 #if NVSIZE == DOUBLESIZE
12232 v = vhex + precis + 1;
12234 /* Round away from zero: if the tail
12235 * beyond the precis xdigits is equal to
12236 * or greater than 0x8000... */
12237 bool round = *v > 0x8;
12238 if (!round && *v == 0x8) {
12239 for (v++; v < vend; v++) {
12247 for (v = vhex + precis; v >= vhex; v--) {
12254 /* If the carry goes all the way to
12255 * the front, we need to output
12256 * a single '1'. This goes against
12257 * the "xdigit and then radix"
12258 * but since this is "cannot happen"
12259 * category, that is probably good. */
12264 /* The new effective "last non zero". */
12265 vlnz = vhex + precis;
12268 zerotail = precis - (vlnz - vhex);
12275 /* The radix is always output after the first
12276 * non-zero xdigit, or if alt. */
12277 if (vfnz < vlnz || alt) {
12278 #ifndef USE_LOCALE_NUMERIC
12281 STORE_LC_NUMERIC_SET_TO_NEEDED();
12282 if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC)) {
12284 const char* r = SvPV(PL_numeric_radix_sv, n);
12285 Copy(r, p, n, char);
12291 RESTORE_LC_NUMERIC();
12306 elen = p - PL_efloatbuf;
12307 elen += my_snprintf(p, PL_efloatsize - elen,
12308 "%c%+d", lower ? 'p' : 'P',
12311 if (elen < width) {
12313 /* Pad the back with spaces. */
12314 memset(PL_efloatbuf + elen, ' ', width - elen);
12316 else if (fill == '0') {
12317 /* Insert the zeros between the "0x" and
12318 * the digits, otherwise we end up with
12320 STRLEN nzero = width - elen;
12321 char* zerox = PL_efloatbuf + 2;
12322 Move(zerox, zerox + nzero, elen - 2, char);
12323 memset(zerox, fill, nzero);
12326 /* Move it to the right. */
12327 Move(PL_efloatbuf, PL_efloatbuf + width - elen,
12329 /* Pad the front with spaces. */
12330 memset(PL_efloatbuf, ' ', width - elen);
12336 elen = S_infnan_2pv(fv, PL_efloatbuf, PL_efloatsize);
12339 char *ptr = ebuf + sizeof ebuf;
12342 #if defined(USE_QUADMATH)
12343 if (intsize == 'q') {
12347 /* FIXME: what to do if HAS_LONG_DOUBLE but not PERL_PRIfldbl? */
12348 #elif defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
12349 /* Note that this is HAS_LONG_DOUBLE and PERL_PRIfldbl,
12350 * not USE_LONG_DOUBLE and NVff. In other words,
12351 * this needs to work without USE_LONG_DOUBLE. */
12352 if (intsize == 'q') {
12353 /* Copy the one or more characters in a long double
12354 * format before the 'base' ([efgEFG]) character to
12355 * the format string. */
12356 static char const ldblf[] = PERL_PRIfldbl;
12357 char const *p = ldblf + sizeof(ldblf) - 3;
12358 while (p >= ldblf) { *--ptr = *p--; }
12363 do { *--ptr = '0' + (base % 10); } while (base /= 10);
12368 do { *--ptr = '0' + (base % 10); } while (base /= 10);
12380 /* No taint. Otherwise we are in the strange situation
12381 * where printf() taints but print($float) doesn't.
12384 STORE_LC_NUMERIC_SET_TO_NEEDED();
12386 /* hopefully the above makes ptr a very constrained format
12387 * that is safe to use, even though it's not literal */
12388 GCC_DIAG_IGNORE(-Wformat-nonliteral);
12389 #ifdef USE_QUADMATH
12391 const char* qfmt = quadmath_format_single(ptr);
12393 Perl_croak_nocontext("panic: quadmath invalid format \"%s\"", ptr);
12394 elen = quadmath_snprintf(PL_efloatbuf, PL_efloatsize,
12396 if ((IV)elen == -1)
12397 Perl_croak_nocontext("panic: quadmath_snprintf failed, format \"%s|'", qfmt);
12401 #elif defined(HAS_LONG_DOUBLE)
12402 elen = ((intsize == 'q')
12403 ? my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv)
12404 : my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, (double)fv));
12406 elen = my_sprintf(PL_efloatbuf, ptr, fv);
12412 eptr = PL_efloatbuf;
12413 assert((IV)elen > 0); /* here zero elen is bad */
12415 #ifdef USE_LOCALE_NUMERIC
12416 /* If the decimal point character in the string is UTF-8, make the
12418 if (PL_numeric_radix_sv && SvUTF8(PL_numeric_radix_sv)
12419 && instr(eptr, SvPVX_const(PL_numeric_radix_sv)))
12432 i = SvCUR(sv) - origlen;
12435 case 'c': *(va_arg(*args, char*)) = i; break;
12436 case 'h': *(va_arg(*args, short*)) = i; break;
12437 default: *(va_arg(*args, int*)) = i; break;
12438 case 'l': *(va_arg(*args, long*)) = i; break;
12439 case 'V': *(va_arg(*args, IV*)) = i; break;
12440 case 'z': *(va_arg(*args, SSize_t*)) = i; break;
12441 #ifdef HAS_PTRDIFF_T
12442 case 't': *(va_arg(*args, ptrdiff_t*)) = i; break;
12445 case 'j': *(va_arg(*args, intmax_t*)) = i; break;
12449 *(va_arg(*args, Quad_t*)) = i; break;
12456 sv_setuv_mg(argsv, has_utf8 ? (UV)sv_len_utf8(sv) : (UV)i);
12457 continue; /* not "break" */
12464 && (PL_op->op_type == OP_PRTF || PL_op->op_type == OP_SPRINTF)
12465 && ckWARN(WARN_PRINTF))
12467 SV * const msg = sv_newmortal();
12468 Perl_sv_setpvf(aTHX_ msg, "Invalid conversion in %sprintf: ",
12469 (PL_op->op_type == OP_PRTF) ? "" : "s");
12470 if (fmtstart < patend) {
12471 const char * const fmtend = q < patend ? q : patend;
12473 sv_catpvs(msg, "\"%");
12474 for (f = fmtstart; f < fmtend; f++) {
12476 sv_catpvn_nomg(msg, f, 1);
12478 Perl_sv_catpvf(aTHX_ msg,
12479 "\\%03"UVof, (UV)*f & 0xFF);
12482 sv_catpvs(msg, "\"");
12484 sv_catpvs(msg, "end of string");
12486 Perl_warner(aTHX_ packWARN(WARN_PRINTF), "%"SVf, SVfARG(msg)); /* yes, this is reentrant */
12489 /* output mangled stuff ... */
12495 /* ... right here, because formatting flags should not apply */
12496 SvGROW(sv, SvCUR(sv) + elen + 1);
12498 Copy(eptr, p, elen, char);
12501 SvCUR_set(sv, p - SvPVX_const(sv));
12503 continue; /* not "break" */
12506 if (is_utf8 != has_utf8) {
12509 sv_utf8_upgrade(sv);
12512 const STRLEN old_elen = elen;
12513 SV * const nsv = newSVpvn_flags(eptr, elen, SVs_TEMP);
12514 sv_utf8_upgrade(nsv);
12515 eptr = SvPVX_const(nsv);
12518 if (width) { /* fudge width (can't fudge elen) */
12519 width += elen - old_elen;
12525 assert((IV)elen >= 0); /* here zero elen is fine */
12526 have = esignlen + zeros + elen;
12528 croak_memory_wrap();
12530 need = (have > width ? have : width);
12533 if (need >= (((STRLEN)~0) - SvCUR(sv) - dotstrlen - 1))
12534 croak_memory_wrap();
12535 SvGROW(sv, SvCUR(sv) + need + dotstrlen + 1);
12537 if (esignlen && fill == '0') {
12539 for (i = 0; i < (int)esignlen; i++)
12540 *p++ = esignbuf[i];
12542 if (gap && !left) {
12543 memset(p, fill, gap);
12546 if (esignlen && fill != '0') {
12548 for (i = 0; i < (int)esignlen; i++)
12549 *p++ = esignbuf[i];
12553 for (i = zeros; i; i--)
12557 Copy(eptr, p, elen, char);
12561 memset(p, ' ', gap);
12566 Copy(dotstr, p, dotstrlen, char);
12570 vectorize = FALSE; /* done iterating over vecstr */
12577 SvCUR_set(sv, p - SvPVX_const(sv));
12584 /* Now that we've consumed all our printf format arguments (svix)
12585 * do we have things left on the stack that we didn't use?
12587 if (!no_redundant_warning && svmax >= svix + 1 && ckWARN(WARN_REDUNDANT)) {
12588 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
12589 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
12594 RESTORE_LC_NUMERIC(); /* Done outside loop, so don't have to save/restore
12598 /* =========================================================================
12600 =head1 Cloning an interpreter
12604 All the macros and functions in this section are for the private use of
12605 the main function, perl_clone().
12607 The foo_dup() functions make an exact copy of an existing foo thingy.
12608 During the course of a cloning, a hash table is used to map old addresses
12609 to new addresses. The table is created and manipulated with the
12610 ptr_table_* functions.
12612 * =========================================================================*/
12615 #if defined(USE_ITHREADS)
12617 /* XXX Remove this so it doesn't have to go thru the macro and return for nothing */
12618 #ifndef GpREFCNT_inc
12619 # define GpREFCNT_inc(gp) ((gp) ? (++(gp)->gp_refcnt, (gp)) : (GP*)NULL)
12623 /* Certain cases in Perl_ss_dup have been merged, by relying on the fact
12624 that currently av_dup, gv_dup and hv_dup are the same as sv_dup.
12625 If this changes, please unmerge ss_dup.
12626 Likewise, sv_dup_inc_multiple() relies on this fact. */
12627 #define sv_dup_inc_NN(s,t) SvREFCNT_inc_NN(sv_dup_inc(s,t))
12628 #define av_dup(s,t) MUTABLE_AV(sv_dup((const SV *)s,t))
12629 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
12630 #define hv_dup(s,t) MUTABLE_HV(sv_dup((const SV *)s,t))
12631 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
12632 #define cv_dup(s,t) MUTABLE_CV(sv_dup((const SV *)s,t))
12633 #define cv_dup_inc(s,t) MUTABLE_CV(sv_dup_inc((const SV *)s,t))
12634 #define io_dup(s,t) MUTABLE_IO(sv_dup((const SV *)s,t))
12635 #define io_dup_inc(s,t) MUTABLE_IO(sv_dup_inc((const SV *)s,t))
12636 #define gv_dup(s,t) MUTABLE_GV(sv_dup((const SV *)s,t))
12637 #define gv_dup_inc(s,t) MUTABLE_GV(sv_dup_inc((const SV *)s,t))
12638 #define SAVEPV(p) ((p) ? savepv(p) : NULL)
12639 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
12641 /* clone a parser */
12644 Perl_parser_dup(pTHX_ const yy_parser *const proto, CLONE_PARAMS *const param)
12648 PERL_ARGS_ASSERT_PARSER_DUP;
12653 /* look for it in the table first */
12654 parser = (yy_parser *)ptr_table_fetch(PL_ptr_table, proto);
12658 /* create anew and remember what it is */
12659 Newxz(parser, 1, yy_parser);
12660 ptr_table_store(PL_ptr_table, proto, parser);
12662 /* XXX these not yet duped */
12663 parser->old_parser = NULL;
12664 parser->stack = NULL;
12666 parser->stack_size = 0;
12667 /* XXX parser->stack->state = 0; */
12669 /* XXX eventually, just Copy() most of the parser struct ? */
12671 parser->lex_brackets = proto->lex_brackets;
12672 parser->lex_casemods = proto->lex_casemods;
12673 parser->lex_brackstack = savepvn(proto->lex_brackstack,
12674 (proto->lex_brackets < 120 ? 120 : proto->lex_brackets));
12675 parser->lex_casestack = savepvn(proto->lex_casestack,
12676 (proto->lex_casemods < 12 ? 12 : proto->lex_casemods));
12677 parser->lex_defer = proto->lex_defer;
12678 parser->lex_dojoin = proto->lex_dojoin;
12679 parser->lex_formbrack = proto->lex_formbrack;
12680 parser->lex_inpat = proto->lex_inpat;
12681 parser->lex_inwhat = proto->lex_inwhat;
12682 parser->lex_op = proto->lex_op;
12683 parser->lex_repl = sv_dup_inc(proto->lex_repl, param);
12684 parser->lex_starts = proto->lex_starts;
12685 parser->lex_stuff = sv_dup_inc(proto->lex_stuff, param);
12686 parser->multi_close = proto->multi_close;
12687 parser->multi_open = proto->multi_open;
12688 parser->multi_start = proto->multi_start;
12689 parser->multi_end = proto->multi_end;
12690 parser->preambled = proto->preambled;
12691 parser->sublex_info = proto->sublex_info; /* XXX not quite right */
12692 parser->linestr = sv_dup_inc(proto->linestr, param);
12693 parser->expect = proto->expect;
12694 parser->copline = proto->copline;
12695 parser->last_lop_op = proto->last_lop_op;
12696 parser->lex_state = proto->lex_state;
12697 parser->rsfp = fp_dup(proto->rsfp, '<', param);
12698 /* rsfp_filters entries have fake IoDIRP() */
12699 parser->rsfp_filters= av_dup_inc(proto->rsfp_filters, param);
12700 parser->in_my = proto->in_my;
12701 parser->in_my_stash = hv_dup(proto->in_my_stash, param);
12702 parser->error_count = proto->error_count;
12705 parser->linestr = sv_dup_inc(proto->linestr, param);
12708 char * const ols = SvPVX(proto->linestr);
12709 char * const ls = SvPVX(parser->linestr);
12711 parser->bufptr = ls + (proto->bufptr >= ols ?
12712 proto->bufptr - ols : 0);
12713 parser->oldbufptr = ls + (proto->oldbufptr >= ols ?
12714 proto->oldbufptr - ols : 0);
12715 parser->oldoldbufptr= ls + (proto->oldoldbufptr >= ols ?
12716 proto->oldoldbufptr - ols : 0);
12717 parser->linestart = ls + (proto->linestart >= ols ?
12718 proto->linestart - ols : 0);
12719 parser->last_uni = ls + (proto->last_uni >= ols ?
12720 proto->last_uni - ols : 0);
12721 parser->last_lop = ls + (proto->last_lop >= ols ?
12722 proto->last_lop - ols : 0);
12724 parser->bufend = ls + SvCUR(parser->linestr);
12727 Copy(proto->tokenbuf, parser->tokenbuf, 256, char);
12730 Copy(proto->nextval, parser->nextval, 5, YYSTYPE);
12731 Copy(proto->nexttype, parser->nexttype, 5, I32);
12732 parser->nexttoke = proto->nexttoke;
12734 /* XXX should clone saved_curcop here, but we aren't passed
12735 * proto_perl; so do it in perl_clone_using instead */
12741 /* duplicate a file handle */
12744 Perl_fp_dup(pTHX_ PerlIO *const fp, const char type, CLONE_PARAMS *const param)
12748 PERL_ARGS_ASSERT_FP_DUP;
12749 PERL_UNUSED_ARG(type);
12752 return (PerlIO*)NULL;
12754 /* look for it in the table first */
12755 ret = (PerlIO*)ptr_table_fetch(PL_ptr_table, fp);
12759 /* create anew and remember what it is */
12760 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE);
12761 ptr_table_store(PL_ptr_table, fp, ret);
12765 /* duplicate a directory handle */
12768 Perl_dirp_dup(pTHX_ DIR *const dp, CLONE_PARAMS *const param)
12772 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
12774 const Direntry_t *dirent;
12775 char smallbuf[256];
12781 PERL_UNUSED_CONTEXT;
12782 PERL_ARGS_ASSERT_DIRP_DUP;
12787 /* look for it in the table first */
12788 ret = (DIR*)ptr_table_fetch(PL_ptr_table, dp);
12792 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
12794 PERL_UNUSED_ARG(param);
12798 /* open the current directory (so we can switch back) */
12799 if (!(pwd = PerlDir_open("."))) return (DIR *)NULL;
12801 /* chdir to our dir handle and open the present working directory */
12802 if (fchdir(my_dirfd(dp)) < 0 || !(ret = PerlDir_open("."))) {
12803 PerlDir_close(pwd);
12804 return (DIR *)NULL;
12806 /* Now we should have two dir handles pointing to the same dir. */
12808 /* Be nice to the calling code and chdir back to where we were. */
12809 /* XXX If this fails, then what? */
12810 PERL_UNUSED_RESULT(fchdir(my_dirfd(pwd)));
12812 /* We have no need of the pwd handle any more. */
12813 PerlDir_close(pwd);
12816 # define d_namlen(d) (d)->d_namlen
12818 # define d_namlen(d) strlen((d)->d_name)
12820 /* Iterate once through dp, to get the file name at the current posi-
12821 tion. Then step back. */
12822 pos = PerlDir_tell(dp);
12823 if ((dirent = PerlDir_read(dp))) {
12824 len = d_namlen(dirent);
12825 if (len <= sizeof smallbuf) name = smallbuf;
12826 else Newx(name, len, char);
12827 Move(dirent->d_name, name, len, char);
12829 PerlDir_seek(dp, pos);
12831 /* Iterate through the new dir handle, till we find a file with the
12833 if (!dirent) /* just before the end */
12835 pos = PerlDir_tell(ret);
12836 if (PerlDir_read(ret)) continue; /* not there yet */
12837 PerlDir_seek(ret, pos); /* step back */
12841 const long pos0 = PerlDir_tell(ret);
12843 pos = PerlDir_tell(ret);
12844 if ((dirent = PerlDir_read(ret))) {
12845 if (len == (STRLEN)d_namlen(dirent)
12846 && memEQ(name, dirent->d_name, len)) {
12848 PerlDir_seek(ret, pos); /* step back */
12851 /* else we are not there yet; keep iterating */
12853 else { /* This is not meant to happen. The best we can do is
12854 reset the iterator to the beginning. */
12855 PerlDir_seek(ret, pos0);
12862 if (name && name != smallbuf)
12867 ret = win32_dirp_dup(dp, param);
12870 /* pop it in the pointer table */
12872 ptr_table_store(PL_ptr_table, dp, ret);
12877 /* duplicate a typeglob */
12880 Perl_gp_dup(pTHX_ GP *const gp, CLONE_PARAMS *const param)
12884 PERL_ARGS_ASSERT_GP_DUP;
12888 /* look for it in the table first */
12889 ret = (GP*)ptr_table_fetch(PL_ptr_table, gp);
12893 /* create anew and remember what it is */
12895 ptr_table_store(PL_ptr_table, gp, ret);
12898 /* ret->gp_refcnt must be 0 before any other dups are called. We're relying
12899 on Newxz() to do this for us. */
12900 ret->gp_sv = sv_dup_inc(gp->gp_sv, param);
12901 ret->gp_io = io_dup_inc(gp->gp_io, param);
12902 ret->gp_form = cv_dup_inc(gp->gp_form, param);
12903 ret->gp_av = av_dup_inc(gp->gp_av, param);
12904 ret->gp_hv = hv_dup_inc(gp->gp_hv, param);
12905 ret->gp_egv = gv_dup(gp->gp_egv, param);/* GvEGV is not refcounted */
12906 ret->gp_cv = cv_dup_inc(gp->gp_cv, param);
12907 ret->gp_cvgen = gp->gp_cvgen;
12908 ret->gp_line = gp->gp_line;
12909 ret->gp_file_hek = hek_dup(gp->gp_file_hek, param);
12913 /* duplicate a chain of magic */
12916 Perl_mg_dup(pTHX_ MAGIC *mg, CLONE_PARAMS *const param)
12918 MAGIC *mgret = NULL;
12919 MAGIC **mgprev_p = &mgret;
12921 PERL_ARGS_ASSERT_MG_DUP;
12923 for (; mg; mg = mg->mg_moremagic) {
12926 if ((param->flags & CLONEf_JOIN_IN)
12927 && mg->mg_type == PERL_MAGIC_backref)
12928 /* when joining, we let the individual SVs add themselves to
12929 * backref as needed. */
12932 Newx(nmg, 1, MAGIC);
12934 mgprev_p = &(nmg->mg_moremagic);
12936 /* There was a comment "XXX copy dynamic vtable?" but as we don't have
12937 dynamic vtables, I'm not sure why Sarathy wrote it. The comment dates
12938 from the original commit adding Perl_mg_dup() - revision 4538.
12939 Similarly there is the annotation "XXX random ptr?" next to the
12940 assignment to nmg->mg_ptr. */
12943 /* FIXME for plugins
12944 if (nmg->mg_type == PERL_MAGIC_qr) {
12945 nmg->mg_obj = MUTABLE_SV(CALLREGDUPE((REGEXP*)nmg->mg_obj, param));
12949 nmg->mg_obj = (nmg->mg_flags & MGf_REFCOUNTED)
12950 ? nmg->mg_type == PERL_MAGIC_backref
12951 /* The backref AV has its reference
12952 * count deliberately bumped by 1 */
12953 ? SvREFCNT_inc(av_dup_inc((const AV *)
12954 nmg->mg_obj, param))
12955 : sv_dup_inc(nmg->mg_obj, param)
12956 : sv_dup(nmg->mg_obj, param);
12958 if (nmg->mg_ptr && nmg->mg_type != PERL_MAGIC_regex_global) {
12959 if (nmg->mg_len > 0) {
12960 nmg->mg_ptr = SAVEPVN(nmg->mg_ptr, nmg->mg_len);
12961 if (nmg->mg_type == PERL_MAGIC_overload_table &&
12962 AMT_AMAGIC((AMT*)nmg->mg_ptr))
12964 AMT * const namtp = (AMT*)nmg->mg_ptr;
12965 sv_dup_inc_multiple((SV**)(namtp->table),
12966 (SV**)(namtp->table), NofAMmeth, param);
12969 else if (nmg->mg_len == HEf_SVKEY)
12970 nmg->mg_ptr = (char*)sv_dup_inc((const SV *)nmg->mg_ptr, param);
12972 if ((nmg->mg_flags & MGf_DUP) && nmg->mg_virtual && nmg->mg_virtual->svt_dup) {
12973 nmg->mg_virtual->svt_dup(aTHX_ nmg, param);
12979 #endif /* USE_ITHREADS */
12981 struct ptr_tbl_arena {
12982 struct ptr_tbl_arena *next;
12983 struct ptr_tbl_ent array[1023/3]; /* as ptr_tbl_ent has 3 pointers. */
12986 /* create a new pointer-mapping table */
12989 Perl_ptr_table_new(pTHX)
12992 PERL_UNUSED_CONTEXT;
12994 Newx(tbl, 1, PTR_TBL_t);
12995 tbl->tbl_max = 511;
12996 tbl->tbl_items = 0;
12997 tbl->tbl_arena = NULL;
12998 tbl->tbl_arena_next = NULL;
12999 tbl->tbl_arena_end = NULL;
13000 Newxz(tbl->tbl_ary, tbl->tbl_max + 1, PTR_TBL_ENT_t*);
13004 #define PTR_TABLE_HASH(ptr) \
13005 ((PTR2UV(ptr) >> 3) ^ (PTR2UV(ptr) >> (3 + 7)) ^ (PTR2UV(ptr) >> (3 + 17)))
13007 /* map an existing pointer using a table */
13009 STATIC PTR_TBL_ENT_t *
13010 S_ptr_table_find(PTR_TBL_t *const tbl, const void *const sv)
13012 PTR_TBL_ENT_t *tblent;
13013 const UV hash = PTR_TABLE_HASH(sv);
13015 PERL_ARGS_ASSERT_PTR_TABLE_FIND;
13017 tblent = tbl->tbl_ary[hash & tbl->tbl_max];
13018 for (; tblent; tblent = tblent->next) {
13019 if (tblent->oldval == sv)
13026 Perl_ptr_table_fetch(pTHX_ PTR_TBL_t *const tbl, const void *const sv)
13028 PTR_TBL_ENT_t const *const tblent = ptr_table_find(tbl, sv);
13030 PERL_ARGS_ASSERT_PTR_TABLE_FETCH;
13031 PERL_UNUSED_CONTEXT;
13033 return tblent ? tblent->newval : NULL;
13036 /* add a new entry to a pointer-mapping table 'tbl'. In hash terms, 'oldsv' is
13037 * the key; 'newsv' is the value. The names "old" and "new" are specific to
13038 * the core's typical use of ptr_tables in thread cloning. */
13041 Perl_ptr_table_store(pTHX_ PTR_TBL_t *const tbl, const void *const oldsv, void *const newsv)
13043 PTR_TBL_ENT_t *tblent = ptr_table_find(tbl, oldsv);
13045 PERL_ARGS_ASSERT_PTR_TABLE_STORE;
13046 PERL_UNUSED_CONTEXT;
13049 tblent->newval = newsv;
13051 const UV entry = PTR_TABLE_HASH(oldsv) & tbl->tbl_max;
13053 if (tbl->tbl_arena_next == tbl->tbl_arena_end) {
13054 struct ptr_tbl_arena *new_arena;
13056 Newx(new_arena, 1, struct ptr_tbl_arena);
13057 new_arena->next = tbl->tbl_arena;
13058 tbl->tbl_arena = new_arena;
13059 tbl->tbl_arena_next = new_arena->array;
13060 tbl->tbl_arena_end = C_ARRAY_END(new_arena->array);
13063 tblent = tbl->tbl_arena_next++;
13065 tblent->oldval = oldsv;
13066 tblent->newval = newsv;
13067 tblent->next = tbl->tbl_ary[entry];
13068 tbl->tbl_ary[entry] = tblent;
13070 if (tblent->next && tbl->tbl_items > tbl->tbl_max)
13071 ptr_table_split(tbl);
13075 /* double the hash bucket size of an existing ptr table */
13078 Perl_ptr_table_split(pTHX_ PTR_TBL_t *const tbl)
13080 PTR_TBL_ENT_t **ary = tbl->tbl_ary;
13081 const UV oldsize = tbl->tbl_max + 1;
13082 UV newsize = oldsize * 2;
13085 PERL_ARGS_ASSERT_PTR_TABLE_SPLIT;
13086 PERL_UNUSED_CONTEXT;
13088 Renew(ary, newsize, PTR_TBL_ENT_t*);
13089 Zero(&ary[oldsize], newsize-oldsize, PTR_TBL_ENT_t*);
13090 tbl->tbl_max = --newsize;
13091 tbl->tbl_ary = ary;
13092 for (i=0; i < oldsize; i++, ary++) {
13093 PTR_TBL_ENT_t **entp = ary;
13094 PTR_TBL_ENT_t *ent = *ary;
13095 PTR_TBL_ENT_t **curentp;
13098 curentp = ary + oldsize;
13100 if ((newsize & PTR_TABLE_HASH(ent->oldval)) != i) {
13102 ent->next = *curentp;
13112 /* remove all the entries from a ptr table */
13113 /* Deprecated - will be removed post 5.14 */
13116 Perl_ptr_table_clear(pTHX_ PTR_TBL_t *const tbl)
13118 PERL_UNUSED_CONTEXT;
13119 if (tbl && tbl->tbl_items) {
13120 struct ptr_tbl_arena *arena = tbl->tbl_arena;
13122 Zero(tbl->tbl_ary, tbl->tbl_max + 1, struct ptr_tbl_ent **);
13125 struct ptr_tbl_arena *next = arena->next;
13131 tbl->tbl_items = 0;
13132 tbl->tbl_arena = NULL;
13133 tbl->tbl_arena_next = NULL;
13134 tbl->tbl_arena_end = NULL;
13138 /* clear and free a ptr table */
13141 Perl_ptr_table_free(pTHX_ PTR_TBL_t *const tbl)
13143 struct ptr_tbl_arena *arena;
13145 PERL_UNUSED_CONTEXT;
13151 arena = tbl->tbl_arena;
13154 struct ptr_tbl_arena *next = arena->next;
13160 Safefree(tbl->tbl_ary);
13164 #if defined(USE_ITHREADS)
13167 Perl_rvpv_dup(pTHX_ SV *const dstr, const SV *const sstr, CLONE_PARAMS *const param)
13169 PERL_ARGS_ASSERT_RVPV_DUP;
13171 assert(!isREGEXP(sstr));
13173 if (SvWEAKREF(sstr)) {
13174 SvRV_set(dstr, sv_dup(SvRV_const(sstr), param));
13175 if (param->flags & CLONEf_JOIN_IN) {
13176 /* if joining, we add any back references individually rather
13177 * than copying the whole backref array */
13178 Perl_sv_add_backref(aTHX_ SvRV(dstr), dstr);
13182 SvRV_set(dstr, sv_dup_inc(SvRV_const(sstr), param));
13184 else if (SvPVX_const(sstr)) {
13185 /* Has something there */
13187 /* Normal PV - clone whole allocated space */
13188 SvPV_set(dstr, SAVEPVN(SvPVX_const(sstr), SvLEN(sstr)-1));
13189 /* sstr may not be that normal, but actually copy on write.
13190 But we are a true, independent SV, so: */
13194 /* Special case - not normally malloced for some reason */
13195 if (isGV_with_GP(sstr)) {
13196 /* Don't need to do anything here. */
13198 else if ((SvIsCOW(sstr))) {
13199 /* A "shared" PV - clone it as "shared" PV */
13201 HEK_KEY(hek_dup(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)),
13205 /* Some other special case - random pointer */
13206 SvPV_set(dstr, (char *) SvPVX_const(sstr));
13211 /* Copy the NULL */
13212 SvPV_set(dstr, NULL);
13216 /* duplicate a list of SVs. source and dest may point to the same memory. */
13218 S_sv_dup_inc_multiple(pTHX_ SV *const *source, SV **dest,
13219 SSize_t items, CLONE_PARAMS *const param)
13221 PERL_ARGS_ASSERT_SV_DUP_INC_MULTIPLE;
13223 while (items-- > 0) {
13224 *dest++ = sv_dup_inc(*source++, param);
13230 /* duplicate an SV of any type (including AV, HV etc) */
13233 S_sv_dup_common(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13238 PERL_ARGS_ASSERT_SV_DUP_COMMON;
13240 if (SvTYPE(sstr) == (svtype)SVTYPEMASK) {
13241 #ifdef DEBUG_LEAKING_SCALARS_ABORT
13246 /* look for it in the table first */
13247 dstr = MUTABLE_SV(ptr_table_fetch(PL_ptr_table, sstr));
13251 if(param->flags & CLONEf_JOIN_IN) {
13252 /** We are joining here so we don't want do clone
13253 something that is bad **/
13254 if (SvTYPE(sstr) == SVt_PVHV) {
13255 const HEK * const hvname = HvNAME_HEK(sstr);
13257 /** don't clone stashes if they already exist **/
13258 dstr = MUTABLE_SV(gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
13259 HEK_UTF8(hvname) ? SVf_UTF8 : 0));
13260 ptr_table_store(PL_ptr_table, sstr, dstr);
13264 else if (SvTYPE(sstr) == SVt_PVGV && !SvFAKE(sstr)) {
13265 HV *stash = GvSTASH(sstr);
13266 const HEK * hvname;
13267 if (stash && (hvname = HvNAME_HEK(stash))) {
13268 /** don't clone GVs if they already exist **/
13270 stash = gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
13271 HEK_UTF8(hvname) ? SVf_UTF8 : 0);
13273 stash, GvNAME(sstr),
13279 if (svp && *svp && SvTYPE(*svp) == SVt_PVGV) {
13280 ptr_table_store(PL_ptr_table, sstr, *svp);
13287 /* create anew and remember what it is */
13290 #ifdef DEBUG_LEAKING_SCALARS
13291 dstr->sv_debug_optype = sstr->sv_debug_optype;
13292 dstr->sv_debug_line = sstr->sv_debug_line;
13293 dstr->sv_debug_inpad = sstr->sv_debug_inpad;
13294 dstr->sv_debug_parent = (SV*)sstr;
13295 FREE_SV_DEBUG_FILE(dstr);
13296 dstr->sv_debug_file = savesharedpv(sstr->sv_debug_file);
13299 ptr_table_store(PL_ptr_table, sstr, dstr);
13302 SvFLAGS(dstr) = SvFLAGS(sstr);
13303 SvFLAGS(dstr) &= ~SVf_OOK; /* don't propagate OOK hack */
13304 SvREFCNT(dstr) = 0; /* must be before any other dups! */
13307 if (SvANY(sstr) && PL_watch_pvx && SvPVX_const(sstr) == PL_watch_pvx)
13308 PerlIO_printf(Perl_debug_log, "watch at %p hit, found string \"%s\"\n",
13309 (void*)PL_watch_pvx, SvPVX_const(sstr));
13312 /* don't clone objects whose class has asked us not to */
13313 if (SvOBJECT(sstr) && ! (SvFLAGS(SvSTASH(sstr)) & SVphv_CLONEABLE)) {
13318 switch (SvTYPE(sstr)) {
13320 SvANY(dstr) = NULL;
13323 SvANY(dstr) = (XPVIV*)((char*)&(dstr->sv_u.svu_iv) - STRUCT_OFFSET(XPVIV, xiv_iv));
13325 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
13327 SvIV_set(dstr, SvIVX(sstr));
13331 #if NVSIZE <= IVSIZE
13332 SvANY(dstr) = (XPVNV*)((char*)&(dstr->sv_u.svu_nv) - STRUCT_OFFSET(XPVNV, xnv_u.xnv_nv));
13334 SvANY(dstr) = new_XNV();
13336 SvNV_set(dstr, SvNVX(sstr));
13340 /* These are all the types that need complex bodies allocating. */
13342 const svtype sv_type = SvTYPE(sstr);
13343 const struct body_details *const sv_type_details
13344 = bodies_by_type + sv_type;
13348 Perl_croak(aTHX_ "Bizarre SvTYPE [%" IVdf "]", (IV)SvTYPE(sstr));
13364 assert(sv_type_details->body_size);
13365 if (sv_type_details->arena) {
13366 new_body_inline(new_body, sv_type);
13368 = (void*)((char*)new_body - sv_type_details->offset);
13370 new_body = new_NOARENA(sv_type_details);
13374 SvANY(dstr) = new_body;
13377 Copy(((char*)SvANY(sstr)) + sv_type_details->offset,
13378 ((char*)SvANY(dstr)) + sv_type_details->offset,
13379 sv_type_details->copy, char);
13381 Copy(((char*)SvANY(sstr)),
13382 ((char*)SvANY(dstr)),
13383 sv_type_details->body_size + sv_type_details->offset, char);
13386 if (sv_type != SVt_PVAV && sv_type != SVt_PVHV
13387 && !isGV_with_GP(dstr)
13389 && !(sv_type == SVt_PVIO && !(IoFLAGS(dstr) & IOf_FAKE_DIRP)))
13390 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
13392 /* The Copy above means that all the source (unduplicated) pointers
13393 are now in the destination. We can check the flags and the
13394 pointers in either, but it's possible that there's less cache
13395 missing by always going for the destination.
13396 FIXME - instrument and check that assumption */
13397 if (sv_type >= SVt_PVMG) {
13398 if ((sv_type == SVt_PVMG) && SvPAD_OUR(dstr)) {
13399 SvOURSTASH_set(dstr, hv_dup_inc(SvOURSTASH(dstr), param));
13400 } else if (sv_type == SVt_PVAV && AvPAD_NAMELIST(dstr)) {
13402 } else if (SvMAGIC(dstr))
13403 SvMAGIC_set(dstr, mg_dup(SvMAGIC(dstr), param));
13404 if (SvOBJECT(dstr) && SvSTASH(dstr))
13405 SvSTASH_set(dstr, hv_dup_inc(SvSTASH(dstr), param));
13406 else SvSTASH_set(dstr, 0); /* don't copy DESTROY cache */
13409 /* The cast silences a GCC warning about unhandled types. */
13410 switch ((int)sv_type) {
13421 /* FIXME for plugins */
13422 dstr->sv_u.svu_rx = ((REGEXP *)dstr)->sv_any;
13423 re_dup_guts((REGEXP*) sstr, (REGEXP*) dstr, param);
13426 /* XXX LvTARGOFF sometimes holds PMOP* when DEBUGGING */
13427 if (LvTYPE(dstr) == 't') /* for tie: unrefcnted fake (SV**) */
13428 LvTARG(dstr) = dstr;
13429 else if (LvTYPE(dstr) == 'T') /* for tie: fake HE */
13430 LvTARG(dstr) = MUTABLE_SV(he_dup((HE*)LvTARG(dstr), 0, param));
13432 LvTARG(dstr) = sv_dup_inc(LvTARG(dstr), param);
13433 if (isREGEXP(sstr)) goto duprex;
13435 /* non-GP case already handled above */
13436 if(isGV_with_GP(sstr)) {
13437 GvNAME_HEK(dstr) = hek_dup(GvNAME_HEK(dstr), param);
13438 /* Don't call sv_add_backref here as it's going to be
13439 created as part of the magic cloning of the symbol
13440 table--unless this is during a join and the stash
13441 is not actually being cloned. */
13442 /* Danger Will Robinson - GvGP(dstr) isn't initialised
13443 at the point of this comment. */
13444 GvSTASH(dstr) = hv_dup(GvSTASH(dstr), param);
13445 if (param->flags & CLONEf_JOIN_IN)
13446 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
13447 GvGP_set(dstr, gp_dup(GvGP(sstr), param));
13448 (void)GpREFCNT_inc(GvGP(dstr));
13452 /* PL_parser->rsfp_filters entries have fake IoDIRP() */
13453 if(IoFLAGS(dstr) & IOf_FAKE_DIRP) {
13454 /* I have no idea why fake dirp (rsfps)
13455 should be treated differently but otherwise
13456 we end up with leaks -- sky*/
13457 IoTOP_GV(dstr) = gv_dup_inc(IoTOP_GV(dstr), param);
13458 IoFMT_GV(dstr) = gv_dup_inc(IoFMT_GV(dstr), param);
13459 IoBOTTOM_GV(dstr) = gv_dup_inc(IoBOTTOM_GV(dstr), param);
13461 IoTOP_GV(dstr) = gv_dup(IoTOP_GV(dstr), param);
13462 IoFMT_GV(dstr) = gv_dup(IoFMT_GV(dstr), param);
13463 IoBOTTOM_GV(dstr) = gv_dup(IoBOTTOM_GV(dstr), param);
13464 if (IoDIRP(dstr)) {
13465 IoDIRP(dstr) = dirp_dup(IoDIRP(dstr), param);
13468 /* IoDIRP(dstr) is already a copy of IoDIRP(sstr) */
13470 IoIFP(dstr) = fp_dup(IoIFP(sstr), IoTYPE(dstr), param);
13472 if (IoOFP(dstr) == IoIFP(sstr))
13473 IoOFP(dstr) = IoIFP(dstr);
13475 IoOFP(dstr) = fp_dup(IoOFP(dstr), IoTYPE(dstr), param);
13476 IoTOP_NAME(dstr) = SAVEPV(IoTOP_NAME(dstr));
13477 IoFMT_NAME(dstr) = SAVEPV(IoFMT_NAME(dstr));
13478 IoBOTTOM_NAME(dstr) = SAVEPV(IoBOTTOM_NAME(dstr));
13481 /* avoid cloning an empty array */
13482 if (AvARRAY((const AV *)sstr) && AvFILLp((const AV *)sstr) >= 0) {
13483 SV **dst_ary, **src_ary;
13484 SSize_t items = AvFILLp((const AV *)sstr) + 1;
13486 src_ary = AvARRAY((const AV *)sstr);
13487 Newxz(dst_ary, AvMAX((const AV *)sstr)+1, SV*);
13488 ptr_table_store(PL_ptr_table, src_ary, dst_ary);
13489 AvARRAY(MUTABLE_AV(dstr)) = dst_ary;
13490 AvALLOC((const AV *)dstr) = dst_ary;
13491 if (AvREAL((const AV *)sstr)) {
13492 dst_ary = sv_dup_inc_multiple(src_ary, dst_ary, items,
13496 while (items-- > 0)
13497 *dst_ary++ = sv_dup(*src_ary++, param);
13499 items = AvMAX((const AV *)sstr) - AvFILLp((const AV *)sstr);
13500 while (items-- > 0) {
13501 *dst_ary++ = &PL_sv_undef;
13505 AvARRAY(MUTABLE_AV(dstr)) = NULL;
13506 AvALLOC((const AV *)dstr) = (SV**)NULL;
13507 AvMAX( (const AV *)dstr) = -1;
13508 AvFILLp((const AV *)dstr) = -1;
13512 if (HvARRAY((const HV *)sstr)) {
13514 const bool sharekeys = !!HvSHAREKEYS(sstr);
13515 XPVHV * const dxhv = (XPVHV*)SvANY(dstr);
13516 XPVHV * const sxhv = (XPVHV*)SvANY(sstr);
13518 Newx(darray, PERL_HV_ARRAY_ALLOC_BYTES(dxhv->xhv_max+1)
13519 + (SvOOK(sstr) ? sizeof(struct xpvhv_aux) : 0),
13521 HvARRAY(dstr) = (HE**)darray;
13522 while (i <= sxhv->xhv_max) {
13523 const HE * const source = HvARRAY(sstr)[i];
13524 HvARRAY(dstr)[i] = source
13525 ? he_dup(source, sharekeys, param) : 0;
13529 const struct xpvhv_aux * const saux = HvAUX(sstr);
13530 struct xpvhv_aux * const daux = HvAUX(dstr);
13531 /* This flag isn't copied. */
13534 if (saux->xhv_name_count) {
13535 HEK ** const sname = saux->xhv_name_u.xhvnameu_names;
13537 = saux->xhv_name_count < 0
13538 ? -saux->xhv_name_count
13539 : saux->xhv_name_count;
13540 HEK **shekp = sname + count;
13542 Newx(daux->xhv_name_u.xhvnameu_names, count, HEK *);
13543 dhekp = daux->xhv_name_u.xhvnameu_names + count;
13544 while (shekp-- > sname) {
13546 *dhekp = hek_dup(*shekp, param);
13550 daux->xhv_name_u.xhvnameu_name
13551 = hek_dup(saux->xhv_name_u.xhvnameu_name,
13554 daux->xhv_name_count = saux->xhv_name_count;
13556 daux->xhv_fill_lazy = saux->xhv_fill_lazy;
13557 daux->xhv_aux_flags = saux->xhv_aux_flags;
13558 #ifdef PERL_HASH_RANDOMIZE_KEYS
13559 daux->xhv_rand = saux->xhv_rand;
13560 daux->xhv_last_rand = saux->xhv_last_rand;
13562 daux->xhv_riter = saux->xhv_riter;
13563 daux->xhv_eiter = saux->xhv_eiter
13564 ? he_dup(saux->xhv_eiter,
13565 cBOOL(HvSHAREKEYS(sstr)), param) : 0;
13566 /* backref array needs refcnt=2; see sv_add_backref */
13567 daux->xhv_backreferences =
13568 (param->flags & CLONEf_JOIN_IN)
13569 /* when joining, we let the individual GVs and
13570 * CVs add themselves to backref as
13571 * needed. This avoids pulling in stuff
13572 * that isn't required, and simplifies the
13573 * case where stashes aren't cloned back
13574 * if they already exist in the parent
13577 : saux->xhv_backreferences
13578 ? (SvTYPE(saux->xhv_backreferences) == SVt_PVAV)
13579 ? MUTABLE_AV(SvREFCNT_inc(
13580 sv_dup_inc((const SV *)
13581 saux->xhv_backreferences, param)))
13582 : MUTABLE_AV(sv_dup((const SV *)
13583 saux->xhv_backreferences, param))
13586 daux->xhv_mro_meta = saux->xhv_mro_meta
13587 ? mro_meta_dup(saux->xhv_mro_meta, param)
13590 /* Record stashes for possible cloning in Perl_clone(). */
13592 av_push(param->stashes, dstr);
13596 HvARRAY(MUTABLE_HV(dstr)) = NULL;
13599 if (!(param->flags & CLONEf_COPY_STACKS)) {
13604 /* NOTE: not refcounted */
13605 SvANY(MUTABLE_CV(dstr))->xcv_stash =
13606 hv_dup(CvSTASH(dstr), param);
13607 if ((param->flags & CLONEf_JOIN_IN) && CvSTASH(dstr))
13608 Perl_sv_add_backref(aTHX_ MUTABLE_SV(CvSTASH(dstr)), dstr);
13609 if (!CvISXSUB(dstr)) {
13611 CvROOT(dstr) = OpREFCNT_inc(CvROOT(dstr));
13613 CvSLABBED_off(dstr);
13614 } else if (CvCONST(dstr)) {
13615 CvXSUBANY(dstr).any_ptr =
13616 sv_dup_inc((const SV *)CvXSUBANY(dstr).any_ptr, param);
13618 assert(!CvSLABBED(dstr));
13619 if (CvDYNFILE(dstr)) CvFILE(dstr) = SAVEPV(CvFILE(dstr));
13621 SvANY((CV *)dstr)->xcv_gv_u.xcv_hek =
13622 hek_dup(CvNAME_HEK((CV *)sstr), param);
13623 /* don't dup if copying back - CvGV isn't refcounted, so the
13624 * duped GV may never be freed. A bit of a hack! DAPM */
13626 SvANY(MUTABLE_CV(dstr))->xcv_gv_u.xcv_gv =
13628 ? gv_dup_inc(CvGV(sstr), param)
13629 : (param->flags & CLONEf_JOIN_IN)
13631 : gv_dup(CvGV(sstr), param);
13633 if (!CvISXSUB(sstr)) {
13634 if(CvPADLIST(sstr))
13635 CvPADLIST_set(dstr, padlist_dup(CvPADLIST(sstr), param));
13637 CvPADLIST_set(dstr, NULL);
13638 } else { /* future union here */
13639 CvRESERVED(dstr) = NULL;
13642 CvWEAKOUTSIDE(sstr)
13643 ? cv_dup( CvOUTSIDE(dstr), param)
13644 : cv_dup_inc(CvOUTSIDE(dstr), param);
13654 Perl_sv_dup_inc(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13656 PERL_ARGS_ASSERT_SV_DUP_INC;
13657 return sstr ? SvREFCNT_inc(sv_dup_common(sstr, param)) : NULL;
13661 Perl_sv_dup(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13663 SV *dstr = sstr ? sv_dup_common(sstr, param) : NULL;
13664 PERL_ARGS_ASSERT_SV_DUP;
13666 /* Track every SV that (at least initially) had a reference count of 0.
13667 We need to do this by holding an actual reference to it in this array.
13668 If we attempt to cheat, turn AvREAL_off(), and store only pointers
13669 (akin to the stashes hash, and the perl stack), we come unstuck if
13670 a weak reference (or other SV legitimately SvREFCNT() == 0 for this
13671 thread) is manipulated in a CLONE method, because CLONE runs before the
13672 unreferenced array is walked to find SVs still with SvREFCNT() == 0
13673 (and fix things up by giving each a reference via the temps stack).
13674 Instead, during CLONE, if the 0-referenced SV has SvREFCNT_inc() and
13675 then SvREFCNT_dec(), it will be cleaned up (and added to the free list)
13676 before the walk of unreferenced happens and a reference to that is SV
13677 added to the temps stack. At which point we have the same SV considered
13678 to be in use, and free to be re-used. Not good.
13680 if (dstr && !(param->flags & CLONEf_COPY_STACKS) && !SvREFCNT(dstr)) {
13681 assert(param->unreferenced);
13682 av_push(param->unreferenced, SvREFCNT_inc(dstr));
13688 /* duplicate a context */
13691 Perl_cx_dup(pTHX_ PERL_CONTEXT *cxs, I32 ix, I32 max, CLONE_PARAMS* param)
13693 PERL_CONTEXT *ncxs;
13695 PERL_ARGS_ASSERT_CX_DUP;
13698 return (PERL_CONTEXT*)NULL;
13700 /* look for it in the table first */
13701 ncxs = (PERL_CONTEXT*)ptr_table_fetch(PL_ptr_table, cxs);
13705 /* create anew and remember what it is */
13706 Newx(ncxs, max + 1, PERL_CONTEXT);
13707 ptr_table_store(PL_ptr_table, cxs, ncxs);
13708 Copy(cxs, ncxs, max + 1, PERL_CONTEXT);
13711 PERL_CONTEXT * const ncx = &ncxs[ix];
13712 if (CxTYPE(ncx) == CXt_SUBST) {
13713 Perl_croak(aTHX_ "Cloning substitution context is unimplemented");
13716 ncx->blk_oldcop = (COP*)any_dup(ncx->blk_oldcop, param->proto_perl);
13717 switch (CxTYPE(ncx)) {
13719 ncx->blk_sub.cv = (ncx->blk_sub.olddepth == 0
13720 ? cv_dup_inc(ncx->blk_sub.cv, param)
13721 : cv_dup(ncx->blk_sub.cv,param));
13722 if(CxHASARGS(ncx)){
13723 ncx->blk_sub.argarray = av_dup_inc(ncx->blk_sub.argarray,param);
13724 ncx->blk_sub.savearray = av_dup_inc(ncx->blk_sub.savearray,param);
13726 ncx->blk_sub.argarray = NULL;
13727 ncx->blk_sub.savearray = NULL;
13729 ncx->blk_sub.oldcomppad = (PAD*)ptr_table_fetch(PL_ptr_table,
13730 ncx->blk_sub.oldcomppad);
13733 ncx->blk_eval.old_namesv = sv_dup_inc(ncx->blk_eval.old_namesv,
13735 ncx->blk_eval.cur_text = sv_dup(ncx->blk_eval.cur_text, param);
13736 ncx->blk_eval.cv = cv_dup(ncx->blk_eval.cv, param);
13738 case CXt_LOOP_LAZYSV:
13739 ncx->blk_loop.state_u.lazysv.end
13740 = sv_dup_inc(ncx->blk_loop.state_u.lazysv.end, param);
13741 /* We are taking advantage of av_dup_inc and sv_dup_inc
13742 actually being the same function, and order equivalence of
13744 We can assert the later [but only at run time :-(] */
13745 assert ((void *) &ncx->blk_loop.state_u.ary.ary ==
13746 (void *) &ncx->blk_loop.state_u.lazysv.cur);
13748 ncx->blk_loop.state_u.ary.ary
13749 = av_dup_inc(ncx->blk_loop.state_u.ary.ary, param);
13750 case CXt_LOOP_LAZYIV:
13751 case CXt_LOOP_PLAIN:
13752 if (CxPADLOOP(ncx)) {
13753 ncx->blk_loop.itervar_u.oldcomppad
13754 = (PAD*)ptr_table_fetch(PL_ptr_table,
13755 ncx->blk_loop.itervar_u.oldcomppad);
13757 ncx->blk_loop.itervar_u.gv
13758 = gv_dup((const GV *)ncx->blk_loop.itervar_u.gv,
13763 ncx->blk_format.cv = cv_dup(ncx->blk_format.cv, param);
13764 ncx->blk_format.gv = gv_dup(ncx->blk_format.gv, param);
13765 ncx->blk_format.dfoutgv = gv_dup_inc(ncx->blk_format.dfoutgv,
13780 /* duplicate a stack info structure */
13783 Perl_si_dup(pTHX_ PERL_SI *si, CLONE_PARAMS* param)
13787 PERL_ARGS_ASSERT_SI_DUP;
13790 return (PERL_SI*)NULL;
13792 /* look for it in the table first */
13793 nsi = (PERL_SI*)ptr_table_fetch(PL_ptr_table, si);
13797 /* create anew and remember what it is */
13798 Newxz(nsi, 1, PERL_SI);
13799 ptr_table_store(PL_ptr_table, si, nsi);
13801 nsi->si_stack = av_dup_inc(si->si_stack, param);
13802 nsi->si_cxix = si->si_cxix;
13803 nsi->si_cxmax = si->si_cxmax;
13804 nsi->si_cxstack = cx_dup(si->si_cxstack, si->si_cxix, si->si_cxmax, param);
13805 nsi->si_type = si->si_type;
13806 nsi->si_prev = si_dup(si->si_prev, param);
13807 nsi->si_next = si_dup(si->si_next, param);
13808 nsi->si_markoff = si->si_markoff;
13813 #define POPINT(ss,ix) ((ss)[--(ix)].any_i32)
13814 #define TOPINT(ss,ix) ((ss)[ix].any_i32)
13815 #define POPLONG(ss,ix) ((ss)[--(ix)].any_long)
13816 #define TOPLONG(ss,ix) ((ss)[ix].any_long)
13817 #define POPIV(ss,ix) ((ss)[--(ix)].any_iv)
13818 #define TOPIV(ss,ix) ((ss)[ix].any_iv)
13819 #define POPUV(ss,ix) ((ss)[--(ix)].any_uv)
13820 #define TOPUV(ss,ix) ((ss)[ix].any_uv)
13821 #define POPBOOL(ss,ix) ((ss)[--(ix)].any_bool)
13822 #define TOPBOOL(ss,ix) ((ss)[ix].any_bool)
13823 #define POPPTR(ss,ix) ((ss)[--(ix)].any_ptr)
13824 #define TOPPTR(ss,ix) ((ss)[ix].any_ptr)
13825 #define POPDPTR(ss,ix) ((ss)[--(ix)].any_dptr)
13826 #define TOPDPTR(ss,ix) ((ss)[ix].any_dptr)
13827 #define POPDXPTR(ss,ix) ((ss)[--(ix)].any_dxptr)
13828 #define TOPDXPTR(ss,ix) ((ss)[ix].any_dxptr)
13831 #define pv_dup_inc(p) SAVEPV(p)
13832 #define pv_dup(p) SAVEPV(p)
13833 #define svp_dup_inc(p,pp) any_dup(p,pp)
13835 /* map any object to the new equivent - either something in the
13836 * ptr table, or something in the interpreter structure
13840 Perl_any_dup(pTHX_ void *v, const PerlInterpreter *proto_perl)
13844 PERL_ARGS_ASSERT_ANY_DUP;
13847 return (void*)NULL;
13849 /* look for it in the table first */
13850 ret = ptr_table_fetch(PL_ptr_table, v);
13854 /* see if it is part of the interpreter structure */
13855 if (v >= (void*)proto_perl && v < (void*)(proto_perl+1))
13856 ret = (void*)(((char*)aTHX) + (((char*)v) - (char*)proto_perl));
13864 /* duplicate the save stack */
13867 Perl_ss_dup(pTHX_ PerlInterpreter *proto_perl, CLONE_PARAMS* param)
13870 ANY * const ss = proto_perl->Isavestack;
13871 const I32 max = proto_perl->Isavestack_max;
13872 I32 ix = proto_perl->Isavestack_ix;
13885 void (*dptr) (void*);
13886 void (*dxptr) (pTHX_ void*);
13888 PERL_ARGS_ASSERT_SS_DUP;
13890 Newxz(nss, max, ANY);
13893 const UV uv = POPUV(ss,ix);
13894 const U8 type = (U8)uv & SAVE_MASK;
13896 TOPUV(nss,ix) = uv;
13898 case SAVEt_CLEARSV:
13899 case SAVEt_CLEARPADRANGE:
13901 case SAVEt_HELEM: /* hash element */
13902 sv = (const SV *)POPPTR(ss,ix);
13903 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13905 case SAVEt_ITEM: /* normal string */
13906 case SAVEt_GVSV: /* scalar slot in GV */
13907 case SAVEt_SV: /* scalar reference */
13908 sv = (const SV *)POPPTR(ss,ix);
13909 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13912 case SAVEt_MORTALIZESV:
13913 case SAVEt_READONLY_OFF:
13914 sv = (const SV *)POPPTR(ss,ix);
13915 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13917 case SAVEt_SHARED_PVREF: /* char* in shared space */
13918 c = (char*)POPPTR(ss,ix);
13919 TOPPTR(nss,ix) = savesharedpv(c);
13920 ptr = POPPTR(ss,ix);
13921 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13923 case SAVEt_GENERIC_SVREF: /* generic sv */
13924 case SAVEt_SVREF: /* scalar reference */
13925 sv = (const SV *)POPPTR(ss,ix);
13926 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13927 ptr = POPPTR(ss,ix);
13928 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
13930 case SAVEt_GVSLOT: /* any slot in GV */
13931 sv = (const SV *)POPPTR(ss,ix);
13932 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13933 ptr = POPPTR(ss,ix);
13934 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
13935 sv = (const SV *)POPPTR(ss,ix);
13936 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13938 case SAVEt_HV: /* hash reference */
13939 case SAVEt_AV: /* array reference */
13940 sv = (const SV *) POPPTR(ss,ix);
13941 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13943 case SAVEt_COMPPAD:
13945 sv = (const SV *) POPPTR(ss,ix);
13946 TOPPTR(nss,ix) = sv_dup(sv, param);
13948 case SAVEt_INT: /* int reference */
13949 ptr = POPPTR(ss,ix);
13950 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13951 intval = (int)POPINT(ss,ix);
13952 TOPINT(nss,ix) = intval;
13954 case SAVEt_LONG: /* long reference */
13955 ptr = POPPTR(ss,ix);
13956 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13957 longval = (long)POPLONG(ss,ix);
13958 TOPLONG(nss,ix) = longval;
13960 case SAVEt_I32: /* I32 reference */
13961 ptr = POPPTR(ss,ix);
13962 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13964 TOPINT(nss,ix) = i;
13966 case SAVEt_IV: /* IV reference */
13967 case SAVEt_STRLEN: /* STRLEN/size_t ref */
13968 ptr = POPPTR(ss,ix);
13969 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13971 TOPIV(nss,ix) = iv;
13973 case SAVEt_HPTR: /* HV* reference */
13974 case SAVEt_APTR: /* AV* reference */
13975 case SAVEt_SPTR: /* SV* reference */
13976 ptr = POPPTR(ss,ix);
13977 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13978 sv = (const SV *)POPPTR(ss,ix);
13979 TOPPTR(nss,ix) = sv_dup(sv, param);
13981 case SAVEt_VPTR: /* random* reference */
13982 ptr = POPPTR(ss,ix);
13983 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13985 case SAVEt_INT_SMALL:
13986 case SAVEt_I32_SMALL:
13987 case SAVEt_I16: /* I16 reference */
13988 case SAVEt_I8: /* I8 reference */
13990 ptr = POPPTR(ss,ix);
13991 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13993 case SAVEt_GENERIC_PVREF: /* generic char* */
13994 case SAVEt_PPTR: /* char* reference */
13995 ptr = POPPTR(ss,ix);
13996 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13997 c = (char*)POPPTR(ss,ix);
13998 TOPPTR(nss,ix) = pv_dup(c);
14000 case SAVEt_GP: /* scalar reference */
14001 gp = (GP*)POPPTR(ss,ix);
14002 TOPPTR(nss,ix) = gp = gp_dup(gp, param);
14003 (void)GpREFCNT_inc(gp);
14004 gv = (const GV *)POPPTR(ss,ix);
14005 TOPPTR(nss,ix) = gv_dup_inc(gv, param);
14008 ptr = POPPTR(ss,ix);
14009 if (ptr && (((OP*)ptr)->op_private & OPpREFCOUNTED)) {
14010 /* these are assumed to be refcounted properly */
14012 switch (((OP*)ptr)->op_type) {
14014 case OP_LEAVESUBLV:
14018 case OP_LEAVEWRITE:
14019 TOPPTR(nss,ix) = ptr;
14022 (void) OpREFCNT_inc(o);
14026 TOPPTR(nss,ix) = NULL;
14031 TOPPTR(nss,ix) = NULL;
14033 case SAVEt_FREECOPHH:
14034 ptr = POPPTR(ss,ix);
14035 TOPPTR(nss,ix) = cophh_copy((COPHH *)ptr);
14037 case SAVEt_ADELETE:
14038 av = (const AV *)POPPTR(ss,ix);
14039 TOPPTR(nss,ix) = av_dup_inc(av, param);
14041 TOPINT(nss,ix) = i;
14044 hv = (const HV *)POPPTR(ss,ix);
14045 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14047 TOPINT(nss,ix) = i;
14050 c = (char*)POPPTR(ss,ix);
14051 TOPPTR(nss,ix) = pv_dup_inc(c);
14053 case SAVEt_STACK_POS: /* Position on Perl stack */
14055 TOPINT(nss,ix) = i;
14057 case SAVEt_DESTRUCTOR:
14058 ptr = POPPTR(ss,ix);
14059 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14060 dptr = POPDPTR(ss,ix);
14061 TOPDPTR(nss,ix) = DPTR2FPTR(void (*)(void*),
14062 any_dup(FPTR2DPTR(void *, dptr),
14065 case SAVEt_DESTRUCTOR_X:
14066 ptr = POPPTR(ss,ix);
14067 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14068 dxptr = POPDXPTR(ss,ix);
14069 TOPDXPTR(nss,ix) = DPTR2FPTR(void (*)(pTHX_ void*),
14070 any_dup(FPTR2DPTR(void *, dxptr),
14073 case SAVEt_REGCONTEXT:
14075 ix -= uv >> SAVE_TIGHT_SHIFT;
14077 case SAVEt_AELEM: /* array element */
14078 sv = (const SV *)POPPTR(ss,ix);
14079 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14081 TOPINT(nss,ix) = i;
14082 av = (const AV *)POPPTR(ss,ix);
14083 TOPPTR(nss,ix) = av_dup_inc(av, param);
14086 ptr = POPPTR(ss,ix);
14087 TOPPTR(nss,ix) = ptr;
14090 ptr = POPPTR(ss,ix);
14091 ptr = cophh_copy((COPHH*)ptr);
14092 TOPPTR(nss,ix) = ptr;
14094 TOPINT(nss,ix) = i;
14095 if (i & HINT_LOCALIZE_HH) {
14096 hv = (const HV *)POPPTR(ss,ix);
14097 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14100 case SAVEt_PADSV_AND_MORTALIZE:
14101 longval = (long)POPLONG(ss,ix);
14102 TOPLONG(nss,ix) = longval;
14103 ptr = POPPTR(ss,ix);
14104 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14105 sv = (const SV *)POPPTR(ss,ix);
14106 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14108 case SAVEt_SET_SVFLAGS:
14110 TOPINT(nss,ix) = i;
14112 TOPINT(nss,ix) = i;
14113 sv = (const SV *)POPPTR(ss,ix);
14114 TOPPTR(nss,ix) = sv_dup(sv, param);
14116 case SAVEt_COMPILE_WARNINGS:
14117 ptr = POPPTR(ss,ix);
14118 TOPPTR(nss,ix) = DUP_WARNINGS((STRLEN*)ptr);
14121 ptr = POPPTR(ss,ix);
14122 TOPPTR(nss,ix) = parser_dup((const yy_parser*)ptr, param);
14124 case SAVEt_GP_ALIASED_SV:
14125 ptr = POPPTR(ss,ix);
14126 TOPPTR(nss,ix) = gp_dup((GP *)ptr, param);
14127 ((GP *)ptr)->gp_refcnt++;
14131 "panic: ss_dup inconsistency (%"IVdf")", (IV) type);
14139 /* if sv is a stash, call $class->CLONE_SKIP(), and set the SVphv_CLONEABLE
14140 * flag to the result. This is done for each stash before cloning starts,
14141 * so we know which stashes want their objects cloned */
14144 do_mark_cloneable_stash(pTHX_ SV *const sv)
14146 const HEK * const hvname = HvNAME_HEK((const HV *)sv);
14148 GV* const cloner = gv_fetchmethod_autoload(MUTABLE_HV(sv), "CLONE_SKIP", 0);
14149 SvFLAGS(sv) |= SVphv_CLONEABLE; /* clone objects by default */
14150 if (cloner && GvCV(cloner)) {
14157 mXPUSHs(newSVhek(hvname));
14159 call_sv(MUTABLE_SV(GvCV(cloner)), G_SCALAR);
14166 SvFLAGS(sv) &= ~SVphv_CLONEABLE;
14174 =for apidoc perl_clone
14176 Create and return a new interpreter by cloning the current one.
14178 perl_clone takes these flags as parameters:
14180 CLONEf_COPY_STACKS - is used to, well, copy the stacks also,
14181 without it we only clone the data and zero the stacks,
14182 with it we copy the stacks and the new perl interpreter is
14183 ready to run at the exact same point as the previous one.
14184 The pseudo-fork code uses COPY_STACKS while the
14185 threads->create doesn't.
14187 CLONEf_KEEP_PTR_TABLE -
14188 perl_clone keeps a ptr_table with the pointer of the old
14189 variable as a key and the new variable as a value,
14190 this allows it to check if something has been cloned and not
14191 clone it again but rather just use the value and increase the
14192 refcount. If KEEP_PTR_TABLE is not set then perl_clone will kill
14193 the ptr_table using the function
14194 C<ptr_table_free(PL_ptr_table); PL_ptr_table = NULL;>,
14195 reason to keep it around is if you want to dup some of your own
14196 variable who are outside the graph perl scans, example of this
14197 code is in threads.xs create.
14199 CLONEf_CLONE_HOST -
14200 This is a win32 thing, it is ignored on unix, it tells perls
14201 win32host code (which is c++) to clone itself, this is needed on
14202 win32 if you want to run two threads at the same time,
14203 if you just want to do some stuff in a separate perl interpreter
14204 and then throw it away and return to the original one,
14205 you don't need to do anything.
14210 /* XXX the above needs expanding by someone who actually understands it ! */
14211 EXTERN_C PerlInterpreter *
14212 perl_clone_host(PerlInterpreter* proto_perl, UV flags);
14215 perl_clone(PerlInterpreter *proto_perl, UV flags)
14218 #ifdef PERL_IMPLICIT_SYS
14220 PERL_ARGS_ASSERT_PERL_CLONE;
14222 /* perlhost.h so we need to call into it
14223 to clone the host, CPerlHost should have a c interface, sky */
14225 if (flags & CLONEf_CLONE_HOST) {
14226 return perl_clone_host(proto_perl,flags);
14228 return perl_clone_using(proto_perl, flags,
14230 proto_perl->IMemShared,
14231 proto_perl->IMemParse,
14233 proto_perl->IStdIO,
14237 proto_perl->IProc);
14241 perl_clone_using(PerlInterpreter *proto_perl, UV flags,
14242 struct IPerlMem* ipM, struct IPerlMem* ipMS,
14243 struct IPerlMem* ipMP, struct IPerlEnv* ipE,
14244 struct IPerlStdIO* ipStd, struct IPerlLIO* ipLIO,
14245 struct IPerlDir* ipD, struct IPerlSock* ipS,
14246 struct IPerlProc* ipP)
14248 /* XXX many of the string copies here can be optimized if they're
14249 * constants; they need to be allocated as common memory and just
14250 * their pointers copied. */
14253 CLONE_PARAMS clone_params;
14254 CLONE_PARAMS* const param = &clone_params;
14256 PerlInterpreter * const my_perl = (PerlInterpreter*)(*ipM->pMalloc)(ipM, sizeof(PerlInterpreter));
14258 PERL_ARGS_ASSERT_PERL_CLONE_USING;
14259 #else /* !PERL_IMPLICIT_SYS */
14261 CLONE_PARAMS clone_params;
14262 CLONE_PARAMS* param = &clone_params;
14263 PerlInterpreter * const my_perl = (PerlInterpreter*)PerlMem_malloc(sizeof(PerlInterpreter));
14265 PERL_ARGS_ASSERT_PERL_CLONE;
14266 #endif /* PERL_IMPLICIT_SYS */
14268 /* for each stash, determine whether its objects should be cloned */
14269 S_visit(proto_perl, do_mark_cloneable_stash, SVt_PVHV, SVTYPEMASK);
14270 PERL_SET_THX(my_perl);
14273 PoisonNew(my_perl, 1, PerlInterpreter);
14276 PL_defstash = NULL; /* may be used by perl malloc() */
14279 PL_scopestack_name = 0;
14281 PL_savestack_ix = 0;
14282 PL_savestack_max = -1;
14283 PL_sig_pending = 0;
14285 Zero(&PL_debug_pad, 1, struct perl_debug_pad);
14286 # ifdef DEBUG_LEAKING_SCALARS
14287 PL_sv_serial = (((UV)my_perl >> 2) & 0xfff) * 1000000;
14289 #else /* !DEBUGGING */
14290 Zero(my_perl, 1, PerlInterpreter);
14291 #endif /* DEBUGGING */
14293 #ifdef PERL_IMPLICIT_SYS
14294 /* host pointers */
14296 PL_MemShared = ipMS;
14297 PL_MemParse = ipMP;
14304 #endif /* PERL_IMPLICIT_SYS */
14307 param->flags = flags;
14308 /* Nothing in the core code uses this, but we make it available to
14309 extensions (using mg_dup). */
14310 param->proto_perl = proto_perl;
14311 /* Likely nothing will use this, but it is initialised to be consistent
14312 with Perl_clone_params_new(). */
14313 param->new_perl = my_perl;
14314 param->unreferenced = NULL;
14317 INIT_TRACK_MEMPOOL(my_perl->Imemory_debug_header, my_perl);
14319 PL_body_arenas = NULL;
14320 Zero(&PL_body_roots, 1, PL_body_roots);
14324 PL_sv_arenaroot = NULL;
14326 PL_debug = proto_perl->Idebug;
14328 /* dbargs array probably holds garbage */
14331 PL_compiling = proto_perl->Icompiling;
14333 /* pseudo environmental stuff */
14334 PL_origargc = proto_perl->Iorigargc;
14335 PL_origargv = proto_perl->Iorigargv;
14337 #ifndef NO_TAINT_SUPPORT
14338 /* Set tainting stuff before PerlIO_debug can possibly get called */
14339 PL_tainting = proto_perl->Itainting;
14340 PL_taint_warn = proto_perl->Itaint_warn;
14342 PL_tainting = FALSE;
14343 PL_taint_warn = FALSE;
14346 PL_minus_c = proto_perl->Iminus_c;
14348 PL_localpatches = proto_perl->Ilocalpatches;
14349 PL_splitstr = proto_perl->Isplitstr;
14350 PL_minus_n = proto_perl->Iminus_n;
14351 PL_minus_p = proto_perl->Iminus_p;
14352 PL_minus_l = proto_perl->Iminus_l;
14353 PL_minus_a = proto_perl->Iminus_a;
14354 PL_minus_E = proto_perl->Iminus_E;
14355 PL_minus_F = proto_perl->Iminus_F;
14356 PL_doswitches = proto_perl->Idoswitches;
14357 PL_dowarn = proto_perl->Idowarn;
14358 PL_sawalias = proto_perl->Isawalias;
14359 #ifdef PERL_SAWAMPERSAND
14360 PL_sawampersand = proto_perl->Isawampersand;
14362 PL_unsafe = proto_perl->Iunsafe;
14363 PL_perldb = proto_perl->Iperldb;
14364 PL_perl_destruct_level = proto_perl->Iperl_destruct_level;
14365 PL_exit_flags = proto_perl->Iexit_flags;
14367 /* XXX time(&PL_basetime) when asked for? */
14368 PL_basetime = proto_perl->Ibasetime;
14370 PL_maxsysfd = proto_perl->Imaxsysfd;
14371 PL_statusvalue = proto_perl->Istatusvalue;
14373 PL_statusvalue_vms = proto_perl->Istatusvalue_vms;
14375 PL_statusvalue_posix = proto_perl->Istatusvalue_posix;
14378 /* RE engine related */
14379 PL_regmatch_slab = NULL;
14380 PL_reg_curpm = NULL;
14382 PL_sub_generation = proto_perl->Isub_generation;
14384 /* funky return mechanisms */
14385 PL_forkprocess = proto_perl->Iforkprocess;
14387 /* internal state */
14388 PL_maxo = proto_perl->Imaxo;
14390 PL_main_start = proto_perl->Imain_start;
14391 PL_eval_root = proto_perl->Ieval_root;
14392 PL_eval_start = proto_perl->Ieval_start;
14394 PL_filemode = proto_perl->Ifilemode;
14395 PL_lastfd = proto_perl->Ilastfd;
14396 PL_oldname = proto_perl->Ioldname; /* XXX not quite right */
14399 PL_gensym = proto_perl->Igensym;
14401 PL_laststatval = proto_perl->Ilaststatval;
14402 PL_laststype = proto_perl->Ilaststype;
14405 PL_profiledata = NULL;
14407 PL_generation = proto_perl->Igeneration;
14409 PL_in_clean_objs = proto_perl->Iin_clean_objs;
14410 PL_in_clean_all = proto_perl->Iin_clean_all;
14412 PL_delaymagic_uid = proto_perl->Idelaymagic_uid;
14413 PL_delaymagic_euid = proto_perl->Idelaymagic_euid;
14414 PL_delaymagic_gid = proto_perl->Idelaymagic_gid;
14415 PL_delaymagic_egid = proto_perl->Idelaymagic_egid;
14416 PL_nomemok = proto_perl->Inomemok;
14417 PL_an = proto_perl->Ian;
14418 PL_evalseq = proto_perl->Ievalseq;
14419 PL_origenviron = proto_perl->Iorigenviron; /* XXX not quite right */
14420 PL_origalen = proto_perl->Iorigalen;
14422 PL_sighandlerp = proto_perl->Isighandlerp;
14424 PL_runops = proto_perl->Irunops;
14426 PL_subline = proto_perl->Isubline;
14429 PL_cryptseen = proto_perl->Icryptseen;
14432 #ifdef USE_LOCALE_COLLATE
14433 PL_collation_ix = proto_perl->Icollation_ix;
14434 PL_collation_standard = proto_perl->Icollation_standard;
14435 PL_collxfrm_base = proto_perl->Icollxfrm_base;
14436 PL_collxfrm_mult = proto_perl->Icollxfrm_mult;
14437 #endif /* USE_LOCALE_COLLATE */
14439 #ifdef USE_LOCALE_NUMERIC
14440 PL_numeric_standard = proto_perl->Inumeric_standard;
14441 PL_numeric_local = proto_perl->Inumeric_local;
14442 #endif /* !USE_LOCALE_NUMERIC */
14444 /* Did the locale setup indicate UTF-8? */
14445 PL_utf8locale = proto_perl->Iutf8locale;
14446 PL_in_utf8_CTYPE_locale = proto_perl->Iin_utf8_CTYPE_locale;
14447 /* Unicode features (see perlrun/-C) */
14448 PL_unicode = proto_perl->Iunicode;
14450 /* Pre-5.8 signals control */
14451 PL_signals = proto_perl->Isignals;
14453 /* times() ticks per second */
14454 PL_clocktick = proto_perl->Iclocktick;
14456 /* Recursion stopper for PerlIO_find_layer */
14457 PL_in_load_module = proto_perl->Iin_load_module;
14459 /* sort() routine */
14460 PL_sort_RealCmp = proto_perl->Isort_RealCmp;
14462 /* Not really needed/useful since the reenrant_retint is "volatile",
14463 * but do it for consistency's sake. */
14464 PL_reentrant_retint = proto_perl->Ireentrant_retint;
14466 /* Hooks to shared SVs and locks. */
14467 PL_sharehook = proto_perl->Isharehook;
14468 PL_lockhook = proto_perl->Ilockhook;
14469 PL_unlockhook = proto_perl->Iunlockhook;
14470 PL_threadhook = proto_perl->Ithreadhook;
14471 PL_destroyhook = proto_perl->Idestroyhook;
14472 PL_signalhook = proto_perl->Isignalhook;
14474 PL_globhook = proto_perl->Iglobhook;
14477 PL_last_swash_hv = NULL; /* reinits on demand */
14478 PL_last_swash_klen = 0;
14479 PL_last_swash_key[0]= '\0';
14480 PL_last_swash_tmps = (U8*)NULL;
14481 PL_last_swash_slen = 0;
14483 PL_srand_called = proto_perl->Isrand_called;
14484 Copy(&(proto_perl->Irandom_state), &PL_random_state, 1, PL_RANDOM_STATE_TYPE);
14486 if (flags & CLONEf_COPY_STACKS) {
14487 /* next allocation will be PL_tmps_stack[PL_tmps_ix+1] */
14488 PL_tmps_ix = proto_perl->Itmps_ix;
14489 PL_tmps_max = proto_perl->Itmps_max;
14490 PL_tmps_floor = proto_perl->Itmps_floor;
14492 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
14493 * NOTE: unlike the others! */
14494 PL_scopestack_ix = proto_perl->Iscopestack_ix;
14495 PL_scopestack_max = proto_perl->Iscopestack_max;
14497 /* next SSPUSHFOO() sets PL_savestack[PL_savestack_ix]
14498 * NOTE: unlike the others! */
14499 PL_savestack_ix = proto_perl->Isavestack_ix;
14500 PL_savestack_max = proto_perl->Isavestack_max;
14503 PL_start_env = proto_perl->Istart_env; /* XXXXXX */
14504 PL_top_env = &PL_start_env;
14506 PL_op = proto_perl->Iop;
14509 PL_Xpv = (XPV*)NULL;
14510 my_perl->Ina = proto_perl->Ina;
14512 PL_statbuf = proto_perl->Istatbuf;
14513 PL_statcache = proto_perl->Istatcache;
14515 #ifndef NO_TAINT_SUPPORT
14516 PL_tainted = proto_perl->Itainted;
14518 PL_tainted = FALSE;
14520 PL_curpm = proto_perl->Icurpm; /* XXX No PMOP ref count */
14522 PL_chopset = proto_perl->Ichopset; /* XXX never deallocated */
14524 PL_restartjmpenv = proto_perl->Irestartjmpenv;
14525 PL_restartop = proto_perl->Irestartop;
14526 PL_in_eval = proto_perl->Iin_eval;
14527 PL_delaymagic = proto_perl->Idelaymagic;
14528 PL_phase = proto_perl->Iphase;
14529 PL_localizing = proto_perl->Ilocalizing;
14531 PL_hv_fetch_ent_mh = NULL;
14532 PL_modcount = proto_perl->Imodcount;
14533 PL_lastgotoprobe = NULL;
14534 PL_dumpindent = proto_perl->Idumpindent;
14536 PL_efloatbuf = NULL; /* reinits on demand */
14537 PL_efloatsize = 0; /* reinits on demand */
14541 PL_colorset = 0; /* reinits PL_colors[] */
14542 /*PL_colors[6] = {0,0,0,0,0,0};*/
14544 /* Pluggable optimizer */
14545 PL_peepp = proto_perl->Ipeepp;
14546 PL_rpeepp = proto_perl->Irpeepp;
14547 /* op_free() hook */
14548 PL_opfreehook = proto_perl->Iopfreehook;
14550 #ifdef USE_REENTRANT_API
14551 /* XXX: things like -Dm will segfault here in perlio, but doing
14552 * PERL_SET_CONTEXT(proto_perl);
14553 * breaks too many other things
14555 Perl_reentrant_init(aTHX);
14558 /* create SV map for pointer relocation */
14559 PL_ptr_table = ptr_table_new();
14561 /* initialize these special pointers as early as possible */
14563 ptr_table_store(PL_ptr_table, &proto_perl->Isv_undef, &PL_sv_undef);
14564 ptr_table_store(PL_ptr_table, &proto_perl->Isv_no, &PL_sv_no);
14565 ptr_table_store(PL_ptr_table, &proto_perl->Isv_yes, &PL_sv_yes);
14567 /* create (a non-shared!) shared string table */
14568 PL_strtab = newHV();
14569 HvSHAREKEYS_off(PL_strtab);
14570 hv_ksplit(PL_strtab, HvTOTALKEYS(proto_perl->Istrtab));
14571 ptr_table_store(PL_ptr_table, proto_perl->Istrtab, PL_strtab);
14573 Zero(PL_sv_consts, SV_CONSTS_COUNT, SV*);
14575 /* This PV will be free'd special way so must set it same way op.c does */
14576 PL_compiling.cop_file = savesharedpv(PL_compiling.cop_file);
14577 ptr_table_store(PL_ptr_table, proto_perl->Icompiling.cop_file, PL_compiling.cop_file);
14579 ptr_table_store(PL_ptr_table, &proto_perl->Icompiling, &PL_compiling);
14580 PL_compiling.cop_warnings = DUP_WARNINGS(PL_compiling.cop_warnings);
14581 CopHINTHASH_set(&PL_compiling, cophh_copy(CopHINTHASH_get(&PL_compiling)));
14582 PL_curcop = (COP*)any_dup(proto_perl->Icurcop, proto_perl);
14584 param->stashes = newAV(); /* Setup array of objects to call clone on */
14585 /* This makes no difference to the implementation, as it always pushes
14586 and shifts pointers to other SVs without changing their reference
14587 count, with the array becoming empty before it is freed. However, it
14588 makes it conceptually clear what is going on, and will avoid some
14589 work inside av.c, filling slots between AvFILL() and AvMAX() with
14590 &PL_sv_undef, and SvREFCNT_dec()ing those. */
14591 AvREAL_off(param->stashes);
14593 if (!(flags & CLONEf_COPY_STACKS)) {
14594 param->unreferenced = newAV();
14597 #ifdef PERLIO_LAYERS
14598 /* Clone PerlIO tables as soon as we can handle general xx_dup() */
14599 PerlIO_clone(aTHX_ proto_perl, param);
14602 PL_envgv = gv_dup_inc(proto_perl->Ienvgv, param);
14603 PL_incgv = gv_dup_inc(proto_perl->Iincgv, param);
14604 PL_hintgv = gv_dup_inc(proto_perl->Ihintgv, param);
14605 PL_origfilename = SAVEPV(proto_perl->Iorigfilename);
14606 PL_diehook = sv_dup_inc(proto_perl->Idiehook, param);
14607 PL_warnhook = sv_dup_inc(proto_perl->Iwarnhook, param);
14610 PL_patchlevel = sv_dup_inc(proto_perl->Ipatchlevel, param);
14611 PL_inplace = SAVEPV(proto_perl->Iinplace);
14612 PL_e_script = sv_dup_inc(proto_perl->Ie_script, param);
14614 /* magical thingies */
14616 PL_encoding = sv_dup(proto_perl->Iencoding, param);
14618 sv_setpvs(PERL_DEBUG_PAD(0), ""); /* For regex debugging. */
14619 sv_setpvs(PERL_DEBUG_PAD(1), ""); /* ext/re needs these */
14620 sv_setpvs(PERL_DEBUG_PAD(2), ""); /* even without DEBUGGING. */
14623 /* Clone the regex array */
14624 /* ORANGE FIXME for plugins, probably in the SV dup code.
14625 newSViv(PTR2IV(CALLREGDUPE(
14626 INT2PTR(REGEXP *, SvIVX(regex)), param))))
14628 PL_regex_padav = av_dup_inc(proto_perl->Iregex_padav, param);
14629 PL_regex_pad = AvARRAY(PL_regex_padav);
14631 PL_stashpadmax = proto_perl->Istashpadmax;
14632 PL_stashpadix = proto_perl->Istashpadix ;
14633 Newx(PL_stashpad, PL_stashpadmax, HV *);
14636 for (; o < PL_stashpadmax; ++o)
14637 PL_stashpad[o] = hv_dup(proto_perl->Istashpad[o], param);
14640 /* shortcuts to various I/O objects */
14641 PL_ofsgv = gv_dup_inc(proto_perl->Iofsgv, param);
14642 PL_stdingv = gv_dup(proto_perl->Istdingv, param);
14643 PL_stderrgv = gv_dup(proto_perl->Istderrgv, param);
14644 PL_defgv = gv_dup(proto_perl->Idefgv, param);
14645 PL_argvgv = gv_dup_inc(proto_perl->Iargvgv, param);
14646 PL_argvoutgv = gv_dup(proto_perl->Iargvoutgv, param);
14647 PL_argvout_stack = av_dup_inc(proto_perl->Iargvout_stack, param);
14649 /* shortcuts to regexp stuff */
14650 PL_replgv = gv_dup_inc(proto_perl->Ireplgv, param);
14652 /* shortcuts to misc objects */
14653 PL_errgv = gv_dup(proto_perl->Ierrgv, param);
14655 /* shortcuts to debugging objects */
14656 PL_DBgv = gv_dup_inc(proto_perl->IDBgv, param);
14657 PL_DBline = gv_dup_inc(proto_perl->IDBline, param);
14658 PL_DBsub = gv_dup_inc(proto_perl->IDBsub, param);
14659 PL_DBsingle = sv_dup(proto_perl->IDBsingle, param);
14660 PL_DBtrace = sv_dup(proto_perl->IDBtrace, param);
14661 PL_DBsignal = sv_dup(proto_perl->IDBsignal, param);
14662 Copy(proto_perl->IDBcontrol, PL_DBcontrol, DBVARMG_COUNT, IV);
14664 /* symbol tables */
14665 PL_defstash = hv_dup_inc(proto_perl->Idefstash, param);
14666 PL_curstash = hv_dup_inc(proto_perl->Icurstash, param);
14667 PL_debstash = hv_dup(proto_perl->Idebstash, param);
14668 PL_globalstash = hv_dup(proto_perl->Iglobalstash, param);
14669 PL_curstname = sv_dup_inc(proto_perl->Icurstname, param);
14671 PL_beginav = av_dup_inc(proto_perl->Ibeginav, param);
14672 PL_beginav_save = av_dup_inc(proto_perl->Ibeginav_save, param);
14673 PL_checkav_save = av_dup_inc(proto_perl->Icheckav_save, param);
14674 PL_unitcheckav = av_dup_inc(proto_perl->Iunitcheckav, param);
14675 PL_unitcheckav_save = av_dup_inc(proto_perl->Iunitcheckav_save, param);
14676 PL_endav = av_dup_inc(proto_perl->Iendav, param);
14677 PL_checkav = av_dup_inc(proto_perl->Icheckav, param);
14678 PL_initav = av_dup_inc(proto_perl->Iinitav, param);
14680 PL_isarev = hv_dup_inc(proto_perl->Iisarev, param);
14682 /* subprocess state */
14683 PL_fdpid = av_dup_inc(proto_perl->Ifdpid, param);
14685 if (proto_perl->Iop_mask)
14686 PL_op_mask = SAVEPVN(proto_perl->Iop_mask, PL_maxo);
14689 /* PL_asserting = proto_perl->Iasserting; */
14691 /* current interpreter roots */
14692 PL_main_cv = cv_dup_inc(proto_perl->Imain_cv, param);
14694 PL_main_root = OpREFCNT_inc(proto_perl->Imain_root);
14697 /* runtime control stuff */
14698 PL_curcopdb = (COP*)any_dup(proto_perl->Icurcopdb, proto_perl);
14700 PL_preambleav = av_dup_inc(proto_perl->Ipreambleav, param);
14702 PL_ors_sv = sv_dup_inc(proto_perl->Iors_sv, param);
14704 /* interpreter atexit processing */
14705 PL_exitlistlen = proto_perl->Iexitlistlen;
14706 if (PL_exitlistlen) {
14707 Newx(PL_exitlist, PL_exitlistlen, PerlExitListEntry);
14708 Copy(proto_perl->Iexitlist, PL_exitlist, PL_exitlistlen, PerlExitListEntry);
14711 PL_exitlist = (PerlExitListEntry*)NULL;
14713 PL_my_cxt_size = proto_perl->Imy_cxt_size;
14714 if (PL_my_cxt_size) {
14715 Newx(PL_my_cxt_list, PL_my_cxt_size, void *);
14716 Copy(proto_perl->Imy_cxt_list, PL_my_cxt_list, PL_my_cxt_size, void *);
14717 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
14718 Newx(PL_my_cxt_keys, PL_my_cxt_size, const char *);
14719 Copy(proto_perl->Imy_cxt_keys, PL_my_cxt_keys, PL_my_cxt_size, char *);
14723 PL_my_cxt_list = (void**)NULL;
14724 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
14725 PL_my_cxt_keys = (const char**)NULL;
14728 PL_modglobal = hv_dup_inc(proto_perl->Imodglobal, param);
14729 PL_custom_op_names = hv_dup_inc(proto_perl->Icustom_op_names,param);
14730 PL_custom_op_descs = hv_dup_inc(proto_perl->Icustom_op_descs,param);
14731 PL_custom_ops = hv_dup_inc(proto_perl->Icustom_ops, param);
14733 PL_compcv = cv_dup(proto_perl->Icompcv, param);
14735 PAD_CLONE_VARS(proto_perl, param);
14737 #ifdef HAVE_INTERP_INTERN
14738 sys_intern_dup(&proto_perl->Isys_intern, &PL_sys_intern);
14741 PL_DBcv = cv_dup(proto_perl->IDBcv, param);
14743 #ifdef PERL_USES_PL_PIDSTATUS
14744 PL_pidstatus = newHV(); /* XXX flag for cloning? */
14746 PL_osname = SAVEPV(proto_perl->Iosname);
14747 PL_parser = parser_dup(proto_perl->Iparser, param);
14749 /* XXX this only works if the saved cop has already been cloned */
14750 if (proto_perl->Iparser) {
14751 PL_parser->saved_curcop = (COP*)any_dup(
14752 proto_perl->Iparser->saved_curcop,
14756 PL_subname = sv_dup_inc(proto_perl->Isubname, param);
14758 #ifdef USE_LOCALE_COLLATE
14759 PL_collation_name = SAVEPV(proto_perl->Icollation_name);
14760 #endif /* USE_LOCALE_COLLATE */
14762 #ifdef USE_LOCALE_NUMERIC
14763 PL_numeric_name = SAVEPV(proto_perl->Inumeric_name);
14764 PL_numeric_radix_sv = sv_dup_inc(proto_perl->Inumeric_radix_sv, param);
14765 #endif /* !USE_LOCALE_NUMERIC */
14767 /* Unicode inversion lists */
14768 PL_Latin1 = sv_dup_inc(proto_perl->ILatin1, param);
14769 PL_UpperLatin1 = sv_dup_inc(proto_perl->IUpperLatin1, param);
14770 PL_AboveLatin1 = sv_dup_inc(proto_perl->IAboveLatin1, param);
14771 PL_InBitmap = sv_dup_inc(proto_perl->IInBitmap, param);
14773 PL_NonL1NonFinalFold = sv_dup_inc(proto_perl->INonL1NonFinalFold, param);
14774 PL_HasMultiCharFold = sv_dup_inc(proto_perl->IHasMultiCharFold, param);
14776 /* utf8 character class swashes */
14777 for (i = 0; i < POSIX_SWASH_COUNT; i++) {
14778 PL_utf8_swash_ptrs[i] = sv_dup_inc(proto_perl->Iutf8_swash_ptrs[i], param);
14780 for (i = 0; i < POSIX_CC_COUNT; i++) {
14781 PL_XPosix_ptrs[i] = sv_dup_inc(proto_perl->IXPosix_ptrs[i], param);
14783 PL_utf8_mark = sv_dup_inc(proto_perl->Iutf8_mark, param);
14784 PL_utf8_X_regular_begin = sv_dup_inc(proto_perl->Iutf8_X_regular_begin, param);
14785 PL_utf8_X_extend = sv_dup_inc(proto_perl->Iutf8_X_extend, param);
14786 PL_utf8_toupper = sv_dup_inc(proto_perl->Iutf8_toupper, param);
14787 PL_utf8_totitle = sv_dup_inc(proto_perl->Iutf8_totitle, param);
14788 PL_utf8_tolower = sv_dup_inc(proto_perl->Iutf8_tolower, param);
14789 PL_utf8_tofold = sv_dup_inc(proto_perl->Iutf8_tofold, param);
14790 PL_utf8_idstart = sv_dup_inc(proto_perl->Iutf8_idstart, param);
14791 PL_utf8_xidstart = sv_dup_inc(proto_perl->Iutf8_xidstart, param);
14792 PL_utf8_perl_idstart = sv_dup_inc(proto_perl->Iutf8_perl_idstart, param);
14793 PL_utf8_perl_idcont = sv_dup_inc(proto_perl->Iutf8_perl_idcont, param);
14794 PL_utf8_idcont = sv_dup_inc(proto_perl->Iutf8_idcont, param);
14795 PL_utf8_xidcont = sv_dup_inc(proto_perl->Iutf8_xidcont, param);
14796 PL_utf8_foldable = sv_dup_inc(proto_perl->Iutf8_foldable, param);
14797 PL_utf8_charname_begin = sv_dup_inc(proto_perl->Iutf8_charname_begin, param);
14798 PL_utf8_charname_continue = sv_dup_inc(proto_perl->Iutf8_charname_continue, param);
14800 if (proto_perl->Ipsig_pend) {
14801 Newxz(PL_psig_pend, SIG_SIZE, int);
14804 PL_psig_pend = (int*)NULL;
14807 if (proto_perl->Ipsig_name) {
14808 Newx(PL_psig_name, 2 * SIG_SIZE, SV*);
14809 sv_dup_inc_multiple(proto_perl->Ipsig_name, PL_psig_name, 2 * SIG_SIZE,
14811 PL_psig_ptr = PL_psig_name + SIG_SIZE;
14814 PL_psig_ptr = (SV**)NULL;
14815 PL_psig_name = (SV**)NULL;
14818 if (flags & CLONEf_COPY_STACKS) {
14819 Newx(PL_tmps_stack, PL_tmps_max, SV*);
14820 sv_dup_inc_multiple(proto_perl->Itmps_stack, PL_tmps_stack,
14821 PL_tmps_ix+1, param);
14823 /* next PUSHMARK() sets *(PL_markstack_ptr+1) */
14824 i = proto_perl->Imarkstack_max - proto_perl->Imarkstack;
14825 Newxz(PL_markstack, i, I32);
14826 PL_markstack_max = PL_markstack + (proto_perl->Imarkstack_max
14827 - proto_perl->Imarkstack);
14828 PL_markstack_ptr = PL_markstack + (proto_perl->Imarkstack_ptr
14829 - proto_perl->Imarkstack);
14830 Copy(proto_perl->Imarkstack, PL_markstack,
14831 PL_markstack_ptr - PL_markstack + 1, I32);
14833 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
14834 * NOTE: unlike the others! */
14835 Newxz(PL_scopestack, PL_scopestack_max, I32);
14836 Copy(proto_perl->Iscopestack, PL_scopestack, PL_scopestack_ix, I32);
14839 Newxz(PL_scopestack_name, PL_scopestack_max, const char *);
14840 Copy(proto_perl->Iscopestack_name, PL_scopestack_name, PL_scopestack_ix, const char *);
14842 /* reset stack AV to correct length before its duped via
14843 * PL_curstackinfo */
14844 AvFILLp(proto_perl->Icurstack) =
14845 proto_perl->Istack_sp - proto_perl->Istack_base;
14847 /* NOTE: si_dup() looks at PL_markstack */
14848 PL_curstackinfo = si_dup(proto_perl->Icurstackinfo, param);
14850 /* PL_curstack = PL_curstackinfo->si_stack; */
14851 PL_curstack = av_dup(proto_perl->Icurstack, param);
14852 PL_mainstack = av_dup(proto_perl->Imainstack, param);
14854 /* next PUSHs() etc. set *(PL_stack_sp+1) */
14855 PL_stack_base = AvARRAY(PL_curstack);
14856 PL_stack_sp = PL_stack_base + (proto_perl->Istack_sp
14857 - proto_perl->Istack_base);
14858 PL_stack_max = PL_stack_base + AvMAX(PL_curstack);
14860 /*Newxz(PL_savestack, PL_savestack_max, ANY);*/
14861 PL_savestack = ss_dup(proto_perl, param);
14865 ENTER; /* perl_destruct() wants to LEAVE; */
14868 PL_statgv = gv_dup(proto_perl->Istatgv, param);
14869 PL_statname = sv_dup_inc(proto_perl->Istatname, param);
14871 PL_rs = sv_dup_inc(proto_perl->Irs, param);
14872 PL_last_in_gv = gv_dup(proto_perl->Ilast_in_gv, param);
14873 PL_defoutgv = gv_dup_inc(proto_perl->Idefoutgv, param);
14874 PL_toptarget = sv_dup_inc(proto_perl->Itoptarget, param);
14875 PL_bodytarget = sv_dup_inc(proto_perl->Ibodytarget, param);
14876 PL_formtarget = sv_dup(proto_perl->Iformtarget, param);
14878 PL_errors = sv_dup_inc(proto_perl->Ierrors, param);
14880 PL_sortcop = (OP*)any_dup(proto_perl->Isortcop, proto_perl);
14881 PL_firstgv = gv_dup_inc(proto_perl->Ifirstgv, param);
14882 PL_secondgv = gv_dup_inc(proto_perl->Isecondgv, param);
14884 PL_stashcache = newHV();
14886 PL_watchaddr = (char **) ptr_table_fetch(PL_ptr_table,
14887 proto_perl->Iwatchaddr);
14888 PL_watchok = PL_watchaddr ? * PL_watchaddr : NULL;
14889 if (PL_debug && PL_watchaddr) {
14890 PerlIO_printf(Perl_debug_log,
14891 "WATCHING: %"UVxf" cloned as %"UVxf" with value %"UVxf"\n",
14892 PTR2UV(proto_perl->Iwatchaddr), PTR2UV(PL_watchaddr),
14893 PTR2UV(PL_watchok));
14896 PL_registered_mros = hv_dup_inc(proto_perl->Iregistered_mros, param);
14897 PL_blockhooks = av_dup_inc(proto_perl->Iblockhooks, param);
14898 PL_utf8_foldclosures = hv_dup_inc(proto_perl->Iutf8_foldclosures, param);
14900 /* Call the ->CLONE method, if it exists, for each of the stashes
14901 identified by sv_dup() above.
14903 while(av_tindex(param->stashes) != -1) {
14904 HV* const stash = MUTABLE_HV(av_shift(param->stashes));
14905 GV* const cloner = gv_fetchmethod_autoload(stash, "CLONE", 0);
14906 if (cloner && GvCV(cloner)) {
14911 mXPUSHs(newSVhek(HvNAME_HEK(stash)));
14913 call_sv(MUTABLE_SV(GvCV(cloner)), G_DISCARD);
14919 if (!(flags & CLONEf_KEEP_PTR_TABLE)) {
14920 ptr_table_free(PL_ptr_table);
14921 PL_ptr_table = NULL;
14924 if (!(flags & CLONEf_COPY_STACKS)) {
14925 unreferenced_to_tmp_stack(param->unreferenced);
14928 SvREFCNT_dec(param->stashes);
14930 /* orphaned? eg threads->new inside BEGIN or use */
14931 if (PL_compcv && ! SvREFCNT(PL_compcv)) {
14932 SvREFCNT_inc_simple_void(PL_compcv);
14933 SAVEFREESV(PL_compcv);
14940 S_unreferenced_to_tmp_stack(pTHX_ AV *const unreferenced)
14942 PERL_ARGS_ASSERT_UNREFERENCED_TO_TMP_STACK;
14944 if (AvFILLp(unreferenced) > -1) {
14945 SV **svp = AvARRAY(unreferenced);
14946 SV **const last = svp + AvFILLp(unreferenced);
14950 if (SvREFCNT(*svp) == 1)
14952 } while (++svp <= last);
14954 EXTEND_MORTAL(count);
14955 svp = AvARRAY(unreferenced);
14958 if (SvREFCNT(*svp) == 1) {
14959 /* Our reference is the only one to this SV. This means that
14960 in this thread, the scalar effectively has a 0 reference.
14961 That doesn't work (cleanup never happens), so donate our
14962 reference to it onto the save stack. */
14963 PL_tmps_stack[++PL_tmps_ix] = *svp;
14965 /* As an optimisation, because we are already walking the
14966 entire array, instead of above doing either
14967 SvREFCNT_inc(*svp) or *svp = &PL_sv_undef, we can instead
14968 release our reference to the scalar, so that at the end of
14969 the array owns zero references to the scalars it happens to
14970 point to. We are effectively converting the array from
14971 AvREAL() on to AvREAL() off. This saves the av_clear()
14972 (triggered by the SvREFCNT_dec(unreferenced) below) from
14973 walking the array a second time. */
14974 SvREFCNT_dec(*svp);
14977 } while (++svp <= last);
14978 AvREAL_off(unreferenced);
14980 SvREFCNT_dec_NN(unreferenced);
14984 Perl_clone_params_del(CLONE_PARAMS *param)
14986 /* This seemingly funky ordering keeps the build with PERL_GLOBAL_STRUCT
14988 PerlInterpreter *const to = param->new_perl;
14990 PerlInterpreter *const was = PERL_GET_THX;
14992 PERL_ARGS_ASSERT_CLONE_PARAMS_DEL;
14998 SvREFCNT_dec(param->stashes);
14999 if (param->unreferenced)
15000 unreferenced_to_tmp_stack(param->unreferenced);
15010 Perl_clone_params_new(PerlInterpreter *const from, PerlInterpreter *const to)
15013 /* Need to play this game, as newAV() can call safesysmalloc(), and that
15014 does a dTHX; to get the context from thread local storage.
15015 FIXME - under PERL_CORE Newx(), Safefree() and friends should expand to
15016 a version that passes in my_perl. */
15017 PerlInterpreter *const was = PERL_GET_THX;
15018 CLONE_PARAMS *param;
15020 PERL_ARGS_ASSERT_CLONE_PARAMS_NEW;
15026 /* Given that we've set the context, we can do this unshared. */
15027 Newx(param, 1, CLONE_PARAMS);
15030 param->proto_perl = from;
15031 param->new_perl = to;
15032 param->stashes = (AV *)Perl_newSV_type(to, SVt_PVAV);
15033 AvREAL_off(param->stashes);
15034 param->unreferenced = (AV *)Perl_newSV_type(to, SVt_PVAV);
15042 #endif /* USE_ITHREADS */
15045 Perl_init_constants(pTHX)
15047 SvREFCNT(&PL_sv_undef) = SvREFCNT_IMMORTAL;
15048 SvFLAGS(&PL_sv_undef) = SVf_READONLY|SVf_PROTECT|SVt_NULL;
15049 SvANY(&PL_sv_undef) = NULL;
15051 SvANY(&PL_sv_no) = new_XPVNV();
15052 SvREFCNT(&PL_sv_no) = SvREFCNT_IMMORTAL;
15053 SvFLAGS(&PL_sv_no) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15054 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15057 SvANY(&PL_sv_yes) = new_XPVNV();
15058 SvREFCNT(&PL_sv_yes) = SvREFCNT_IMMORTAL;
15059 SvFLAGS(&PL_sv_yes) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15060 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15063 SvPV_set(&PL_sv_no, (char*)PL_No);
15064 SvCUR_set(&PL_sv_no, 0);
15065 SvLEN_set(&PL_sv_no, 0);
15066 SvIV_set(&PL_sv_no, 0);
15067 SvNV_set(&PL_sv_no, 0);
15069 SvPV_set(&PL_sv_yes, (char*)PL_Yes);
15070 SvCUR_set(&PL_sv_yes, 1);
15071 SvLEN_set(&PL_sv_yes, 0);
15072 SvIV_set(&PL_sv_yes, 1);
15073 SvNV_set(&PL_sv_yes, 1);
15077 =head1 Unicode Support
15079 =for apidoc sv_recode_to_utf8
15081 The encoding is assumed to be an Encode object, on entry the PV
15082 of the sv is assumed to be octets in that encoding, and the sv
15083 will be converted into Unicode (and UTF-8).
15085 If the sv already is UTF-8 (or if it is not POK), or if the encoding
15086 is not a reference, nothing is done to the sv. If the encoding is not
15087 an C<Encode::XS> Encoding object, bad things will happen.
15088 (See F<lib/encoding.pm> and L<Encode>.)
15090 The PV of the sv is returned.
15095 Perl_sv_recode_to_utf8(pTHX_ SV *sv, SV *encoding)
15097 PERL_ARGS_ASSERT_SV_RECODE_TO_UTF8;
15099 if (SvPOK(sv) && !SvUTF8(sv) && !IN_BYTES && SvROK(encoding)) {
15108 if (SvPADTMP(nsv)) {
15109 nsv = sv_newmortal();
15110 SvSetSV_nosteal(nsv, sv);
15118 Passing sv_yes is wrong - it needs to be or'ed set of constants
15119 for Encode::XS, while UTf-8 decode (currently) assumes a true value means
15120 remove converted chars from source.
15122 Both will default the value - let them.
15124 XPUSHs(&PL_sv_yes);
15127 call_method("decode", G_SCALAR);
15131 s = SvPV_const(uni, len);
15132 if (s != SvPVX_const(sv)) {
15133 SvGROW(sv, len + 1);
15134 Move(s, SvPVX(sv), len + 1, char);
15135 SvCUR_set(sv, len);
15140 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
15141 /* clear pos and any utf8 cache */
15142 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
15145 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
15146 magic_setutf8(sv,mg); /* clear UTF8 cache */
15151 return SvPOKp(sv) ? SvPVX(sv) : NULL;
15155 =for apidoc sv_cat_decode
15157 The encoding is assumed to be an Encode object, the PV of the ssv is
15158 assumed to be octets in that encoding and decoding the input starts
15159 from the position which (PV + *offset) pointed to. The dsv will be
15160 concatenated the decoded UTF-8 string from ssv. Decoding will terminate
15161 when the string tstr appears in decoding output or the input ends on
15162 the PV of the ssv. The value which the offset points will be modified
15163 to the last input position on the ssv.
15165 Returns TRUE if the terminator was found, else returns FALSE.
15170 Perl_sv_cat_decode(pTHX_ SV *dsv, SV *encoding,
15171 SV *ssv, int *offset, char *tstr, int tlen)
15175 PERL_ARGS_ASSERT_SV_CAT_DECODE;
15177 if (SvPOK(ssv) && SvPOK(dsv) && SvROK(encoding) && offset) {
15187 offsv = newSViv(*offset);
15189 mPUSHp(tstr, tlen);
15191 call_method("cat_decode", G_SCALAR);
15193 ret = SvTRUE(TOPs);
15194 *offset = SvIV(offsv);
15200 Perl_croak(aTHX_ "Invalid argument to sv_cat_decode");
15205 /* ---------------------------------------------------------------------
15207 * support functions for report_uninit()
15210 /* the maxiumum size of array or hash where we will scan looking
15211 * for the undefined element that triggered the warning */
15213 #define FUV_MAX_SEARCH_SIZE 1000
15215 /* Look for an entry in the hash whose value has the same SV as val;
15216 * If so, return a mortal copy of the key. */
15219 S_find_hash_subscript(pTHX_ const HV *const hv, const SV *const val)
15225 PERL_ARGS_ASSERT_FIND_HASH_SUBSCRIPT;
15227 if (!hv || SvMAGICAL(hv) || !HvARRAY(hv) ||
15228 (HvTOTALKEYS(hv) > FUV_MAX_SEARCH_SIZE))
15231 array = HvARRAY(hv);
15233 for (i=HvMAX(hv); i>=0; i--) {
15235 for (entry = array[i]; entry; entry = HeNEXT(entry)) {
15236 if (HeVAL(entry) != val)
15238 if ( HeVAL(entry) == &PL_sv_undef ||
15239 HeVAL(entry) == &PL_sv_placeholder)
15243 if (HeKLEN(entry) == HEf_SVKEY)
15244 return sv_mortalcopy(HeKEY_sv(entry));
15245 return sv_2mortal(newSVhek(HeKEY_hek(entry)));
15251 /* Look for an entry in the array whose value has the same SV as val;
15252 * If so, return the index, otherwise return -1. */
15255 S_find_array_subscript(pTHX_ const AV *const av, const SV *const val)
15257 PERL_ARGS_ASSERT_FIND_ARRAY_SUBSCRIPT;
15259 if (!av || SvMAGICAL(av) || !AvARRAY(av) ||
15260 (AvFILLp(av) > FUV_MAX_SEARCH_SIZE))
15263 if (val != &PL_sv_undef) {
15264 SV ** const svp = AvARRAY(av);
15267 for (i=AvFILLp(av); i>=0; i--)
15274 /* varname(): return the name of a variable, optionally with a subscript.
15275 * If gv is non-zero, use the name of that global, along with gvtype (one
15276 * of "$", "@", "%"); otherwise use the name of the lexical at pad offset
15277 * targ. Depending on the value of the subscript_type flag, return:
15280 #define FUV_SUBSCRIPT_NONE 1 /* "@foo" */
15281 #define FUV_SUBSCRIPT_ARRAY 2 /* "$foo[aindex]" */
15282 #define FUV_SUBSCRIPT_HASH 3 /* "$foo{keyname}" */
15283 #define FUV_SUBSCRIPT_WITHIN 4 /* "within @foo" */
15286 Perl_varname(pTHX_ const GV *const gv, const char gvtype, PADOFFSET targ,
15287 const SV *const keyname, I32 aindex, int subscript_type)
15290 SV * const name = sv_newmortal();
15291 if (gv && isGV(gv)) {
15293 buffer[0] = gvtype;
15296 /* as gv_fullname4(), but add literal '^' for $^FOO names */
15298 gv_fullname4(name, gv, buffer, 0);
15300 if ((unsigned int)SvPVX(name)[1] <= 26) {
15302 buffer[1] = SvPVX(name)[1] + 'A' - 1;
15304 /* Swap the 1 unprintable control character for the 2 byte pretty
15305 version - ie substr($name, 1, 1) = $buffer; */
15306 sv_insert(name, 1, 1, buffer, 2);
15310 CV * const cv = gv ? ((CV *)gv) : find_runcv(NULL);
15314 assert(!cv || SvTYPE(cv) == SVt_PVCV || SvTYPE(cv) == SVt_PVFM);
15316 if (!cv || !CvPADLIST(cv))
15318 av = *PadlistARRAY(CvPADLIST(cv));
15319 sv = *av_fetch(av, targ, FALSE);
15320 sv_setsv_flags(name, sv, 0);
15323 if (subscript_type == FUV_SUBSCRIPT_HASH) {
15324 SV * const sv = newSV(0);
15325 *SvPVX(name) = '$';
15326 Perl_sv_catpvf(aTHX_ name, "{%s}",
15327 pv_pretty(sv, SvPVX_const(keyname), SvCUR(keyname), 32, NULL, NULL,
15328 PERL_PV_PRETTY_DUMP | PERL_PV_ESCAPE_UNI_DETECT ));
15329 SvREFCNT_dec_NN(sv);
15331 else if (subscript_type == FUV_SUBSCRIPT_ARRAY) {
15332 *SvPVX(name) = '$';
15333 Perl_sv_catpvf(aTHX_ name, "[%"IVdf"]", (IV)aindex);
15335 else if (subscript_type == FUV_SUBSCRIPT_WITHIN) {
15336 /* We know that name has no magic, so can use 0 instead of SV_GMAGIC */
15337 Perl_sv_insert_flags(aTHX_ name, 0, 0, STR_WITH_LEN("within "), 0);
15345 =for apidoc find_uninit_var
15347 Find the name of the undefined variable (if any) that caused the operator
15348 to issue a "Use of uninitialized value" warning.
15349 If match is true, only return a name if its value matches uninit_sv.
15350 So roughly speaking, if a unary operator (such as OP_COS) generates a
15351 warning, then following the direct child of the op may yield an
15352 OP_PADSV or OP_GV that gives the name of the undefined variable. On the
15353 other hand, with OP_ADD there are two branches to follow, so we only print
15354 the variable name if we get an exact match.
15356 The name is returned as a mortal SV.
15358 Assumes that PL_op is the op that originally triggered the error, and that
15359 PL_comppad/PL_curpad points to the currently executing pad.
15365 S_find_uninit_var(pTHX_ const OP *const obase, const SV *const uninit_sv,
15371 const OP *o, *o2, *kid;
15373 if (!obase || (match && (!uninit_sv || uninit_sv == &PL_sv_undef ||
15374 uninit_sv == &PL_sv_placeholder)))
15377 switch (obase->op_type) {
15384 const bool pad = ( obase->op_type == OP_PADAV
15385 || obase->op_type == OP_PADHV
15386 || obase->op_type == OP_PADRANGE
15389 const bool hash = ( obase->op_type == OP_PADHV
15390 || obase->op_type == OP_RV2HV
15391 || (obase->op_type == OP_PADRANGE
15392 && SvTYPE(PAD_SVl(obase->op_targ)) == SVt_PVHV)
15396 int subscript_type = FUV_SUBSCRIPT_WITHIN;
15398 if (pad) { /* @lex, %lex */
15399 sv = PAD_SVl(obase->op_targ);
15403 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
15404 /* @global, %global */
15405 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
15408 sv = hash ? MUTABLE_SV(GvHV(gv)): MUTABLE_SV(GvAV(gv));
15410 else if (obase == PL_op) /* @{expr}, %{expr} */
15411 return find_uninit_var(cUNOPx(obase)->op_first,
15413 else /* @{expr}, %{expr} as a sub-expression */
15417 /* attempt to find a match within the aggregate */
15419 keysv = find_hash_subscript((const HV*)sv, uninit_sv);
15421 subscript_type = FUV_SUBSCRIPT_HASH;
15424 index = find_array_subscript((const AV *)sv, uninit_sv);
15426 subscript_type = FUV_SUBSCRIPT_ARRAY;
15429 if (match && subscript_type == FUV_SUBSCRIPT_WITHIN)
15432 return varname(gv, (char)(hash ? '%' : '@'), obase->op_targ,
15433 keysv, index, subscript_type);
15437 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
15439 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
15440 if (!gv || !GvSTASH(gv))
15442 if (match && (GvSV(gv) != uninit_sv))
15444 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
15447 return find_uninit_var(cUNOPx(obase)->op_first, uninit_sv, 1);
15450 if (match && PAD_SVl(obase->op_targ) != uninit_sv)
15452 return varname(NULL, '$', obase->op_targ,
15453 NULL, 0, FUV_SUBSCRIPT_NONE);
15456 gv = cGVOPx_gv(obase);
15457 if (!gv || (match && GvSV(gv) != uninit_sv) || !GvSTASH(gv))
15459 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
15461 case OP_AELEMFAST_LEX:
15464 AV *av = MUTABLE_AV(PAD_SV(obase->op_targ));
15465 if (!av || SvRMAGICAL(av))
15467 svp = av_fetch(av, (I8)obase->op_private, FALSE);
15468 if (!svp || *svp != uninit_sv)
15471 return varname(NULL, '$', obase->op_targ,
15472 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
15475 gv = cGVOPx_gv(obase);
15480 AV *const av = GvAV(gv);
15481 if (!av || SvRMAGICAL(av))
15483 svp = av_fetch(av, (I8)obase->op_private, FALSE);
15484 if (!svp || *svp != uninit_sv)
15487 return varname(gv, '$', 0,
15488 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
15490 NOT_REACHED; /* NOTREACHED */
15493 o = cUNOPx(obase)->op_first;
15494 if (!o || o->op_type != OP_NULL ||
15495 ! (o->op_targ == OP_AELEM || o->op_targ == OP_HELEM))
15497 return find_uninit_var(cBINOPo->op_last, uninit_sv, match);
15502 bool negate = FALSE;
15504 if (PL_op == obase)
15505 /* $a[uninit_expr] or $h{uninit_expr} */
15506 return find_uninit_var(cBINOPx(obase)->op_last, uninit_sv, match);
15509 o = cBINOPx(obase)->op_first;
15510 kid = cBINOPx(obase)->op_last;
15512 /* get the av or hv, and optionally the gv */
15514 if (o->op_type == OP_PADAV || o->op_type == OP_PADHV) {
15515 sv = PAD_SV(o->op_targ);
15517 else if ((o->op_type == OP_RV2AV || o->op_type == OP_RV2HV)
15518 && cUNOPo->op_first->op_type == OP_GV)
15520 gv = cGVOPx_gv(cUNOPo->op_first);
15524 == OP_RV2HV ? MUTABLE_SV(GvHV(gv)) : MUTABLE_SV(GvAV(gv));
15529 if (kid && kid->op_type == OP_NEGATE) {
15531 kid = cUNOPx(kid)->op_first;
15534 if (kid && kid->op_type == OP_CONST && SvOK(cSVOPx_sv(kid))) {
15535 /* index is constant */
15538 kidsv = sv_2mortal(newSVpvs("-"));
15539 sv_catsv(kidsv, cSVOPx_sv(kid));
15542 kidsv = cSVOPx_sv(kid);
15546 if (obase->op_type == OP_HELEM) {
15547 HE* he = hv_fetch_ent(MUTABLE_HV(sv), kidsv, 0, 0);
15548 if (!he || HeVAL(he) != uninit_sv)
15552 SV * const opsv = cSVOPx_sv(kid);
15553 const IV opsviv = SvIV(opsv);
15554 SV * const * const svp = av_fetch(MUTABLE_AV(sv),
15555 negate ? - opsviv : opsviv,
15557 if (!svp || *svp != uninit_sv)
15561 if (obase->op_type == OP_HELEM)
15562 return varname(gv, '%', o->op_targ,
15563 kidsv, 0, FUV_SUBSCRIPT_HASH);
15565 return varname(gv, '@', o->op_targ, NULL,
15566 negate ? - SvIV(cSVOPx_sv(kid)) : SvIV(cSVOPx_sv(kid)),
15567 FUV_SUBSCRIPT_ARRAY);
15570 /* index is an expression;
15571 * attempt to find a match within the aggregate */
15572 if (obase->op_type == OP_HELEM) {
15573 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
15575 return varname(gv, '%', o->op_targ,
15576 keysv, 0, FUV_SUBSCRIPT_HASH);
15580 = find_array_subscript((const AV *)sv, uninit_sv);
15582 return varname(gv, '@', o->op_targ,
15583 NULL, index, FUV_SUBSCRIPT_ARRAY);
15588 (char)((o->op_type == OP_PADAV || o->op_type == OP_RV2AV)
15590 o->op_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
15592 NOT_REACHED; /* NOTREACHED */
15596 /* only examine RHS */
15597 return find_uninit_var(cBINOPx(obase)->op_first, uninit_sv, match);
15600 o = cUNOPx(obase)->op_first;
15601 if ( o->op_type == OP_PUSHMARK
15602 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)
15606 if (!OP_HAS_SIBLING(o)) {
15607 /* one-arg version of open is highly magical */
15609 if (o->op_type == OP_GV) { /* open FOO; */
15611 if (match && GvSV(gv) != uninit_sv)
15613 return varname(gv, '$', 0,
15614 NULL, 0, FUV_SUBSCRIPT_NONE);
15616 /* other possibilities not handled are:
15617 * open $x; or open my $x; should return '${*$x}'
15618 * open expr; should return '$'.expr ideally
15624 /* ops where $_ may be an implicit arg */
15629 if ( !(obase->op_flags & OPf_STACKED)) {
15630 if (uninit_sv == ((obase->op_private & OPpTARGET_MY)
15631 ? PAD_SVl(obase->op_targ)
15634 sv = sv_newmortal();
15635 sv_setpvs(sv, "$_");
15644 match = 1; /* print etc can return undef on defined args */
15645 /* skip filehandle as it can't produce 'undef' warning */
15646 o = cUNOPx(obase)->op_first;
15647 if ((obase->op_flags & OPf_STACKED)
15649 ( o->op_type == OP_PUSHMARK
15650 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)))
15651 o = OP_SIBLING(OP_SIBLING(o));
15655 case OP_ENTEREVAL: /* could be eval $undef or $x='$undef'; eval $x */
15656 case OP_CUSTOM: /* XS or custom code could trigger random warnings */
15658 /* the following ops are capable of returning PL_sv_undef even for
15659 * defined arg(s) */
15678 case OP_GETPEERNAME:
15726 case OP_SMARTMATCH:
15735 /* XXX tmp hack: these two may call an XS sub, and currently
15736 XS subs don't have a SUB entry on the context stack, so CV and
15737 pad determination goes wrong, and BAD things happen. So, just
15738 don't try to determine the value under those circumstances.
15739 Need a better fix at dome point. DAPM 11/2007 */
15745 GV * const gv = gv_fetchpvs(".", GV_NOTQUAL, SVt_PV);
15746 if (gv && GvSV(gv) == uninit_sv)
15747 return newSVpvs_flags("$.", SVs_TEMP);
15752 /* def-ness of rval pos() is independent of the def-ness of its arg */
15753 if ( !(obase->op_flags & OPf_MOD))
15758 if (SvROK(PL_rs) && uninit_sv == SvRV(PL_rs))
15759 return newSVpvs_flags("${$/}", SVs_TEMP);
15764 if (!(obase->op_flags & OPf_KIDS))
15766 o = cUNOPx(obase)->op_first;
15772 /* This loop checks all the kid ops, skipping any that cannot pos-
15773 * sibly be responsible for the uninitialized value; i.e., defined
15774 * constants and ops that return nothing. If there is only one op
15775 * left that is not skipped, then we *know* it is responsible for
15776 * the uninitialized value. If there is more than one op left, we
15777 * have to look for an exact match in the while() loop below.
15778 * Note that we skip padrange, because the individual pad ops that
15779 * it replaced are still in the tree, so we work on them instead.
15782 for (kid=o; kid; kid = OP_SIBLING(kid)) {
15783 const OPCODE type = kid->op_type;
15784 if ( (type == OP_CONST && SvOK(cSVOPx_sv(kid)))
15785 || (type == OP_NULL && ! (kid->op_flags & OPf_KIDS))
15786 || (type == OP_PUSHMARK)
15787 || (type == OP_PADRANGE)
15791 if (o2) { /* more than one found */
15798 return find_uninit_var(o2, uninit_sv, match);
15800 /* scan all args */
15802 sv = find_uninit_var(o, uninit_sv, 1);
15814 =for apidoc report_uninit
15816 Print appropriate "Use of uninitialized variable" warning.
15822 Perl_report_uninit(pTHX_ const SV *uninit_sv)
15825 SV* varname = NULL;
15827 if (uninit_sv && PL_curpad) {
15828 varname = find_uninit_var(PL_op, uninit_sv,0);
15830 sv_insert(varname, 0, 0, " ", 1);
15832 desc = PL_op->op_type == OP_STRINGIFY && PL_op->op_folded
15835 /* PL_warn_uninit_sv is constant */
15836 GCC_DIAG_IGNORE(-Wformat-nonliteral);
15837 /* diag_listed_as: Use of uninitialized value%s */
15838 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit_sv,
15839 SVfARG(varname ? varname : &PL_sv_no),
15844 /* PL_warn_uninit is constant */
15845 GCC_DIAG_IGNORE(-Wformat-nonliteral);
15846 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit,
15854 * c-indentation-style: bsd
15855 * c-basic-offset: 4
15856 * indent-tabs-mode: nil
15859 * ex: set ts=8 sts=4 sw=4 et: