3 * Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000,
4 * 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009 by Larry Wall
7 * You may distribute under the terms of either the GNU General Public
8 * License or the Artistic License, as specified in the README file.
13 * 'I wonder what the Entish is for "yes" and "no",' he thought.
16 * [p.480 of _The Lord of the Rings_, III/iv: "Treebeard"]
22 * This file contains the code that creates, manipulates and destroys
23 * scalar values (SVs). The other types (AV, HV, GV, etc.) reuse the
24 * structure of an SV, so their creation and destruction is handled
25 * here; higher-level functions are in av.c, hv.c, and so on. Opcode
26 * level functions (eg. substr, split, join) for each of the types are
39 /* Missing proto on LynxOS */
40 char *gconvert(double, int, int, char *);
44 # define SNPRINTF_G(nv, buffer, size, ndig) \
45 quadmath_snprintf(buffer, size, "%.*Qg", (int)ndig, (NV)(nv))
47 # define SNPRINTF_G(nv, buffer, size, ndig) \
48 PERL_UNUSED_RESULT(Gconvert((NV)(nv), (int)ndig, 0, buffer))
51 #ifndef SV_COW_THRESHOLD
52 # define SV_COW_THRESHOLD 0 /* COW iff len > K */
54 #ifndef SV_COWBUF_THRESHOLD
55 # define SV_COWBUF_THRESHOLD 1250 /* COW iff len > K */
57 #ifndef SV_COW_MAX_WASTE_THRESHOLD
58 # define SV_COW_MAX_WASTE_THRESHOLD 80 /* COW iff (len - cur) < K */
60 #ifndef SV_COWBUF_WASTE_THRESHOLD
61 # define SV_COWBUF_WASTE_THRESHOLD 80 /* COW iff (len - cur) < K */
63 #ifndef SV_COW_MAX_WASTE_FACTOR_THRESHOLD
64 # define SV_COW_MAX_WASTE_FACTOR_THRESHOLD 2 /* COW iff len < (cur * K) */
66 #ifndef SV_COWBUF_WASTE_FACTOR_THRESHOLD
67 # define SV_COWBUF_WASTE_FACTOR_THRESHOLD 2 /* COW iff len < (cur * K) */
69 /* Work around compiler warnings about unsigned >= THRESHOLD when thres-
72 # define GE_COW_THRESHOLD(cur) ((cur) >= SV_COW_THRESHOLD)
74 # define GE_COW_THRESHOLD(cur) 1
76 #if SV_COWBUF_THRESHOLD
77 # define GE_COWBUF_THRESHOLD(cur) ((cur) >= SV_COWBUF_THRESHOLD)
79 # define GE_COWBUF_THRESHOLD(cur) 1
81 #if SV_COW_MAX_WASTE_THRESHOLD
82 # define GE_COW_MAX_WASTE_THRESHOLD(cur,len) (((len)-(cur)) < SV_COW_MAX_WASTE_THRESHOLD)
84 # define GE_COW_MAX_WASTE_THRESHOLD(cur,len) 1
86 #if SV_COWBUF_WASTE_THRESHOLD
87 # define GE_COWBUF_WASTE_THRESHOLD(cur,len) (((len)-(cur)) < SV_COWBUF_WASTE_THRESHOLD)
89 # define GE_COWBUF_WASTE_THRESHOLD(cur,len) 1
91 #if SV_COW_MAX_WASTE_FACTOR_THRESHOLD
92 # define GE_COW_MAX_WASTE_FACTOR_THRESHOLD(cur,len) ((len) < SV_COW_MAX_WASTE_FACTOR_THRESHOLD * (cur))
94 # define GE_COW_MAX_WASTE_FACTOR_THRESHOLD(cur,len) 1
96 #if SV_COWBUF_WASTE_FACTOR_THRESHOLD
97 # define GE_COWBUF_WASTE_FACTOR_THRESHOLD(cur,len) ((len) < SV_COWBUF_WASTE_FACTOR_THRESHOLD * (cur))
99 # define GE_COWBUF_WASTE_FACTOR_THRESHOLD(cur,len) 1
102 #define CHECK_COW_THRESHOLD(cur,len) (\
103 GE_COW_THRESHOLD((cur)) && \
104 GE_COW_MAX_WASTE_THRESHOLD((cur),(len)) && \
105 GE_COW_MAX_WASTE_FACTOR_THRESHOLD((cur),(len)) \
107 #define CHECK_COWBUF_THRESHOLD(cur,len) (\
108 GE_COWBUF_THRESHOLD((cur)) && \
109 GE_COWBUF_WASTE_THRESHOLD((cur),(len)) && \
110 GE_COWBUF_WASTE_FACTOR_THRESHOLD((cur),(len)) \
113 #ifdef PERL_UTF8_CACHE_ASSERT
114 /* if adding more checks watch out for the following tests:
115 * t/op/index.t t/op/length.t t/op/pat.t t/op/substr.t
116 * lib/utf8.t lib/Unicode/Collate/t/index.t
119 # define ASSERT_UTF8_CACHE(cache) \
120 STMT_START { if (cache) { assert((cache)[0] <= (cache)[1]); \
121 assert((cache)[2] <= (cache)[3]); \
122 assert((cache)[3] <= (cache)[1]);} \
125 # define ASSERT_UTF8_CACHE(cache) NOOP
128 static const char S_destroy[] = "DESTROY";
129 #define S_destroy_len (sizeof(S_destroy)-1)
131 /* ============================================================================
133 =head1 Allocation and deallocation of SVs.
135 An SV (or AV, HV, etc.) is allocated in two parts: the head (struct
136 sv, av, hv...) contains type and reference count information, and for
137 many types, a pointer to the body (struct xrv, xpv, xpviv...), which
138 contains fields specific to each type. Some types store all they need
139 in the head, so don't have a body.
141 In all but the most memory-paranoid configurations (ex: PURIFY), heads
142 and bodies are allocated out of arenas, which by default are
143 approximately 4K chunks of memory parcelled up into N heads or bodies.
144 Sv-bodies are allocated by their sv-type, guaranteeing size
145 consistency needed to allocate safely from arrays.
147 For SV-heads, the first slot in each arena is reserved, and holds a
148 link to the next arena, some flags, and a note of the number of slots.
149 Snaked through each arena chain is a linked list of free items; when
150 this becomes empty, an extra arena is allocated and divided up into N
151 items which are threaded into the free list.
153 SV-bodies are similar, but they use arena-sets by default, which
154 separate the link and info from the arena itself, and reclaim the 1st
155 slot in the arena. SV-bodies are further described later.
157 The following global variables are associated with arenas:
159 PL_sv_arenaroot pointer to list of SV arenas
160 PL_sv_root pointer to list of free SV structures
162 PL_body_arenas head of linked-list of body arenas
163 PL_body_roots[] array of pointers to list of free bodies of svtype
164 arrays are indexed by the svtype needed
166 A few special SV heads are not allocated from an arena, but are
167 instead directly created in the interpreter structure, eg PL_sv_undef.
168 The size of arenas can be changed from the default by setting
169 PERL_ARENA_SIZE appropriately at compile time.
171 The SV arena serves the secondary purpose of allowing still-live SVs
172 to be located and destroyed during final cleanup.
174 At the lowest level, the macros new_SV() and del_SV() grab and free
175 an SV head. (If debugging with -DD, del_SV() calls the function S_del_sv()
176 to return the SV to the free list with error checking.) new_SV() calls
177 more_sv() / sv_add_arena() to add an extra arena if the free list is empty.
178 SVs in the free list have their SvTYPE field set to all ones.
180 At the time of very final cleanup, sv_free_arenas() is called from
181 perl_destruct() to physically free all the arenas allocated since the
182 start of the interpreter.
184 The function visit() scans the SV arenas list, and calls a specified
185 function for each SV it finds which is still live - ie which has an SvTYPE
186 other than all 1's, and a non-zero SvREFCNT. visit() is used by the
187 following functions (specified as [function that calls visit()] / [function
188 called by visit() for each SV]):
190 sv_report_used() / do_report_used()
191 dump all remaining SVs (debugging aid)
193 sv_clean_objs() / do_clean_objs(),do_clean_named_objs(),
194 do_clean_named_io_objs(),do_curse()
195 Attempt to free all objects pointed to by RVs,
196 try to do the same for all objects indir-
197 ectly referenced by typeglobs too, and
198 then do a final sweep, cursing any
199 objects that remain. Called once from
200 perl_destruct(), prior to calling sv_clean_all()
203 sv_clean_all() / do_clean_all()
204 SvREFCNT_dec(sv) each remaining SV, possibly
205 triggering an sv_free(). It also sets the
206 SVf_BREAK flag on the SV to indicate that the
207 refcnt has been artificially lowered, and thus
208 stopping sv_free() from giving spurious warnings
209 about SVs which unexpectedly have a refcnt
210 of zero. called repeatedly from perl_destruct()
211 until there are no SVs left.
213 =head2 Arena allocator API Summary
215 Private API to rest of sv.c
219 new_XPVNV(), del_XPVGV(),
224 sv_report_used(), sv_clean_objs(), sv_clean_all(), sv_free_arenas()
228 * ========================================================================= */
231 * "A time to plant, and a time to uproot what was planted..."
235 # define MEM_LOG_NEW_SV(sv, file, line, func) \
236 Perl_mem_log_new_sv(sv, file, line, func)
237 # define MEM_LOG_DEL_SV(sv, file, line, func) \
238 Perl_mem_log_del_sv(sv, file, line, func)
240 # define MEM_LOG_NEW_SV(sv, file, line, func) NOOP
241 # define MEM_LOG_DEL_SV(sv, file, line, func) NOOP
244 #ifdef DEBUG_LEAKING_SCALARS
245 # define FREE_SV_DEBUG_FILE(sv) STMT_START { \
246 if ((sv)->sv_debug_file) PerlMemShared_free((sv)->sv_debug_file); \
248 # define DEBUG_SV_SERIAL(sv) \
249 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%" UVxf ": (%05ld) del_SV\n", \
250 PTR2UV(sv), (long)(sv)->sv_debug_serial))
252 # define FREE_SV_DEBUG_FILE(sv)
253 # define DEBUG_SV_SERIAL(sv) NOOP
257 # define SvARENA_CHAIN(sv) ((sv)->sv_u.svu_rv)
258 # define SvARENA_CHAIN_SET(sv,val) (sv)->sv_u.svu_rv = MUTABLE_SV((val))
259 /* Whilst I'd love to do this, it seems that things like to check on
261 # define POISON_SV_HEAD(sv) PoisonNew(sv, 1, struct STRUCT_SV)
263 # define POISON_SV_HEAD(sv) PoisonNew(&SvANY(sv), 1, void *), \
264 PoisonNew(&SvREFCNT(sv), 1, U32)
266 # define SvARENA_CHAIN(sv) SvANY(sv)
267 # define SvARENA_CHAIN_SET(sv,val) SvANY(sv) = (void *)(val)
268 # define POISON_SV_HEAD(sv)
271 /* Mark an SV head as unused, and add to free list.
273 * If SVf_BREAK is set, skip adding it to the free list, as this SV had
274 * its refcount artificially decremented during global destruction, so
275 * there may be dangling pointers to it. The last thing we want in that
276 * case is for it to be reused. */
278 #define plant_SV(p) \
280 const U32 old_flags = SvFLAGS(p); \
281 MEM_LOG_DEL_SV(p, __FILE__, __LINE__, FUNCTION__); \
282 DEBUG_SV_SERIAL(p); \
283 FREE_SV_DEBUG_FILE(p); \
285 SvFLAGS(p) = SVTYPEMASK; \
286 if (!(old_flags & SVf_BREAK)) { \
287 SvARENA_CHAIN_SET(p, PL_sv_root); \
293 #define uproot_SV(p) \
296 PL_sv_root = MUTABLE_SV(SvARENA_CHAIN(p)); \
301 /* make some more SVs by adding another arena */
307 char *chunk; /* must use New here to match call to */
308 Newx(chunk,PERL_ARENA_SIZE,char); /* Safefree() in sv_free_arenas() */
309 sv_add_arena(chunk, PERL_ARENA_SIZE, 0);
314 /* new_SV(): return a new, empty SV head */
316 #ifdef DEBUG_LEAKING_SCALARS
317 /* provide a real function for a debugger to play with */
319 S_new_SV(pTHX_ const char *file, int line, const char *func)
326 sv = S_more_sv(aTHX);
330 sv->sv_debug_optype = PL_op ? PL_op->op_type : 0;
331 sv->sv_debug_line = (U16) (PL_parser && PL_parser->copline != NOLINE
337 sv->sv_debug_inpad = 0;
338 sv->sv_debug_parent = NULL;
339 sv->sv_debug_file = PL_curcop ? savesharedpv(CopFILE(PL_curcop)): NULL;
341 sv->sv_debug_serial = PL_sv_serial++;
343 MEM_LOG_NEW_SV(sv, file, line, func);
344 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%" UVxf ": (%05ld) new_SV (from %s:%d [%s])\n",
345 PTR2UV(sv), (long)sv->sv_debug_serial, file, line, func));
349 # define new_SV(p) (p)=S_new_SV(aTHX_ __FILE__, __LINE__, FUNCTION__)
357 (p) = S_more_sv(aTHX); \
361 MEM_LOG_NEW_SV(p, __FILE__, __LINE__, FUNCTION__); \
366 /* del_SV(): return an empty SV head to the free list */
379 S_del_sv(pTHX_ SV *p)
381 PERL_ARGS_ASSERT_DEL_SV;
386 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
387 const SV * const sv = sva + 1;
388 const SV * const svend = &sva[SvREFCNT(sva)];
389 if (p >= sv && p < svend) {
395 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
396 "Attempt to free non-arena SV: 0x%" UVxf
397 pTHX__FORMAT, PTR2UV(p) pTHX__VALUE);
404 #else /* ! DEBUGGING */
406 #define del_SV(p) plant_SV(p)
408 #endif /* DEBUGGING */
412 =head1 SV Manipulation Functions
414 =for apidoc sv_add_arena
416 Given a chunk of memory, link it to the head of the list of arenas,
417 and split it into a list of free SVs.
423 S_sv_add_arena(pTHX_ char *const ptr, const U32 size, const U32 flags)
425 SV *const sva = MUTABLE_SV(ptr);
429 PERL_ARGS_ASSERT_SV_ADD_ARENA;
431 /* The first SV in an arena isn't an SV. */
432 SvANY(sva) = (void *) PL_sv_arenaroot; /* ptr to next arena */
433 SvREFCNT(sva) = size / sizeof(SV); /* number of SV slots */
434 SvFLAGS(sva) = flags; /* FAKE if not to be freed */
436 PL_sv_arenaroot = sva;
437 PL_sv_root = sva + 1;
439 svend = &sva[SvREFCNT(sva) - 1];
442 SvARENA_CHAIN_SET(sv, (sv + 1));
446 /* Must always set typemask because it's always checked in on cleanup
447 when the arenas are walked looking for objects. */
448 SvFLAGS(sv) = SVTYPEMASK;
451 SvARENA_CHAIN_SET(sv, 0);
455 SvFLAGS(sv) = SVTYPEMASK;
458 /* visit(): call the named function for each non-free SV in the arenas
459 * whose flags field matches the flags/mask args. */
462 S_visit(pTHX_ SVFUNC_t f, const U32 flags, const U32 mask)
467 PERL_ARGS_ASSERT_VISIT;
469 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
470 const SV * const svend = &sva[SvREFCNT(sva)];
472 for (sv = sva + 1; sv < svend; ++sv) {
473 if (SvTYPE(sv) != (svtype)SVTYPEMASK
474 && (sv->sv_flags & mask) == flags
487 /* called by sv_report_used() for each live SV */
490 do_report_used(pTHX_ SV *const sv)
492 if (SvTYPE(sv) != (svtype)SVTYPEMASK) {
493 PerlIO_printf(Perl_debug_log, "****\n");
500 =for apidoc sv_report_used
502 Dump the contents of all SVs not yet freed (debugging aid).
508 Perl_sv_report_used(pTHX)
511 visit(do_report_used, 0, 0);
517 /* called by sv_clean_objs() for each live SV */
520 do_clean_objs(pTHX_ SV *const ref)
524 SV * const target = SvRV(ref);
525 if (SvOBJECT(target)) {
526 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning object ref:\n "), sv_dump(ref)));
527 if (SvWEAKREF(ref)) {
528 sv_del_backref(target, ref);
534 SvREFCNT_dec_NN(target);
541 /* clear any slots in a GV which hold objects - except IO;
542 * called by sv_clean_objs() for each live GV */
545 do_clean_named_objs(pTHX_ SV *const sv)
548 assert(SvTYPE(sv) == SVt_PVGV);
549 assert(isGV_with_GP(sv));
553 /* freeing GP entries may indirectly free the current GV;
554 * hold onto it while we mess with the GP slots */
557 if ( ((obj = GvSV(sv) )) && SvOBJECT(obj)) {
558 DEBUG_D((PerlIO_printf(Perl_debug_log,
559 "Cleaning named glob SV object:\n "), sv_dump(obj)));
561 SvREFCNT_dec_NN(obj);
563 if ( ((obj = MUTABLE_SV(GvAV(sv)) )) && SvOBJECT(obj)) {
564 DEBUG_D((PerlIO_printf(Perl_debug_log,
565 "Cleaning named glob AV object:\n "), sv_dump(obj)));
567 SvREFCNT_dec_NN(obj);
569 if ( ((obj = MUTABLE_SV(GvHV(sv)) )) && SvOBJECT(obj)) {
570 DEBUG_D((PerlIO_printf(Perl_debug_log,
571 "Cleaning named glob HV object:\n "), sv_dump(obj)));
573 SvREFCNT_dec_NN(obj);
575 if ( ((obj = MUTABLE_SV(GvCV(sv)) )) && SvOBJECT(obj)) {
576 DEBUG_D((PerlIO_printf(Perl_debug_log,
577 "Cleaning named glob CV object:\n "), sv_dump(obj)));
579 SvREFCNT_dec_NN(obj);
581 SvREFCNT_dec_NN(sv); /* undo the inc above */
584 /* clear any IO slots in a GV which hold objects (except stderr, defout);
585 * called by sv_clean_objs() for each live GV */
588 do_clean_named_io_objs(pTHX_ SV *const sv)
591 assert(SvTYPE(sv) == SVt_PVGV);
592 assert(isGV_with_GP(sv));
593 if (!GvGP(sv) || sv == (SV*)PL_stderrgv || sv == (SV*)PL_defoutgv)
597 if ( ((obj = MUTABLE_SV(GvIO(sv)) )) && SvOBJECT(obj)) {
598 DEBUG_D((PerlIO_printf(Perl_debug_log,
599 "Cleaning named glob IO object:\n "), sv_dump(obj)));
601 SvREFCNT_dec_NN(obj);
603 SvREFCNT_dec_NN(sv); /* undo the inc above */
606 /* Void wrapper to pass to visit() */
608 do_curse(pTHX_ SV * const sv) {
609 if ((PL_stderrgv && GvGP(PL_stderrgv) && (SV*)GvIO(PL_stderrgv) == sv)
610 || (PL_defoutgv && GvGP(PL_defoutgv) && (SV*)GvIO(PL_defoutgv) == sv))
616 =for apidoc sv_clean_objs
618 Attempt to destroy all objects not yet freed.
624 Perl_sv_clean_objs(pTHX)
627 PL_in_clean_objs = TRUE;
628 visit(do_clean_objs, SVf_ROK, SVf_ROK);
629 /* Some barnacles may yet remain, clinging to typeglobs.
630 * Run the non-IO destructors first: they may want to output
631 * error messages, close files etc */
632 visit(do_clean_named_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
633 visit(do_clean_named_io_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
634 /* And if there are some very tenacious barnacles clinging to arrays,
635 closures, or what have you.... */
636 visit(do_curse, SVs_OBJECT, SVs_OBJECT);
637 olddef = PL_defoutgv;
638 PL_defoutgv = NULL; /* disable skip of PL_defoutgv */
639 if (olddef && isGV_with_GP(olddef))
640 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olddef));
641 olderr = PL_stderrgv;
642 PL_stderrgv = NULL; /* disable skip of PL_stderrgv */
643 if (olderr && isGV_with_GP(olderr))
644 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olderr));
645 SvREFCNT_dec(olddef);
646 PL_in_clean_objs = FALSE;
649 /* called by sv_clean_all() for each live SV */
652 do_clean_all(pTHX_ SV *const sv)
654 if (sv == (const SV *) PL_fdpid || sv == (const SV *)PL_strtab) {
655 /* don't clean pid table and strtab */
658 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning loops: SV at 0x%" UVxf "\n", PTR2UV(sv)) ));
659 SvFLAGS(sv) |= SVf_BREAK;
664 =for apidoc sv_clean_all
666 Decrement the refcnt of each remaining SV, possibly triggering a
667 cleanup. This function may have to be called multiple times to free
668 SVs which are in complex self-referential hierarchies.
674 Perl_sv_clean_all(pTHX)
677 PL_in_clean_all = TRUE;
678 cleaned = visit(do_clean_all, 0,0);
683 ARENASETS: a meta-arena implementation which separates arena-info
684 into struct arena_set, which contains an array of struct
685 arena_descs, each holding info for a single arena. By separating
686 the meta-info from the arena, we recover the 1st slot, formerly
687 borrowed for list management. The arena_set is about the size of an
688 arena, avoiding the needless malloc overhead of a naive linked-list.
690 The cost is 1 arena-set malloc per ~320 arena-mallocs, + the unused
691 memory in the last arena-set (1/2 on average). In trade, we get
692 back the 1st slot in each arena (ie 1.7% of a CV-arena, less for
693 smaller types). The recovery of the wasted space allows use of
694 small arenas for large, rare body types, by changing array* fields
695 in body_details_by_type[] below.
698 char *arena; /* the raw storage, allocated aligned */
699 size_t size; /* its size ~4k typ */
700 svtype utype; /* bodytype stored in arena */
705 /* Get the maximum number of elements in set[] such that struct arena_set
706 will fit within PERL_ARENA_SIZE, which is probably just under 4K, and
707 therefore likely to be 1 aligned memory page. */
709 #define ARENAS_PER_SET ((PERL_ARENA_SIZE - sizeof(struct arena_set*) \
710 - 2 * sizeof(int)) / sizeof (struct arena_desc))
713 struct arena_set* next;
714 unsigned int set_size; /* ie ARENAS_PER_SET */
715 unsigned int curr; /* index of next available arena-desc */
716 struct arena_desc set[ARENAS_PER_SET];
720 =for apidoc sv_free_arenas
722 Deallocate the memory used by all arenas. Note that all the individual SV
723 heads and bodies within the arenas must already have been freed.
729 Perl_sv_free_arenas(pTHX)
735 /* Free arenas here, but be careful about fake ones. (We assume
736 contiguity of the fake ones with the corresponding real ones.) */
738 for (sva = PL_sv_arenaroot; sva; sva = svanext) {
739 svanext = MUTABLE_SV(SvANY(sva));
740 while (svanext && SvFAKE(svanext))
741 svanext = MUTABLE_SV(SvANY(svanext));
748 struct arena_set *aroot = (struct arena_set*) PL_body_arenas;
751 struct arena_set *current = aroot;
754 assert(aroot->set[i].arena);
755 Safefree(aroot->set[i].arena);
763 i = PERL_ARENA_ROOTS_SIZE;
765 PL_body_roots[i] = 0;
772 Here are mid-level routines that manage the allocation of bodies out
773 of the various arenas. There are 4 kinds of arenas:
775 1. SV-head arenas, which are discussed and handled above
776 2. regular body arenas
777 3. arenas for reduced-size bodies
780 Arena types 2 & 3 are chained by body-type off an array of
781 arena-root pointers, which is indexed by svtype. Some of the
782 larger/less used body types are malloced singly, since a large
783 unused block of them is wasteful. Also, several svtypes dont have
784 bodies; the data fits into the sv-head itself. The arena-root
785 pointer thus has a few unused root-pointers (which may be hijacked
786 later for arena type 4)
788 3 differs from 2 as an optimization; some body types have several
789 unused fields in the front of the structure (which are kept in-place
790 for consistency). These bodies can be allocated in smaller chunks,
791 because the leading fields arent accessed. Pointers to such bodies
792 are decremented to point at the unused 'ghost' memory, knowing that
793 the pointers are used with offsets to the real memory.
795 Allocation of SV-bodies is similar to SV-heads, differing as follows;
796 the allocation mechanism is used for many body types, so is somewhat
797 more complicated, it uses arena-sets, and has no need for still-live
800 At the outermost level, (new|del)_X*V macros return bodies of the
801 appropriate type. These macros call either (new|del)_body_type or
802 (new|del)_body_allocated macro pairs, depending on specifics of the
803 type. Most body types use the former pair, the latter pair is used to
804 allocate body types with "ghost fields".
806 "ghost fields" are fields that are unused in certain types, and
807 consequently don't need to actually exist. They are declared because
808 they're part of a "base type", which allows use of functions as
809 methods. The simplest examples are AVs and HVs, 2 aggregate types
810 which don't use the fields which support SCALAR semantics.
812 For these types, the arenas are carved up into appropriately sized
813 chunks, we thus avoid wasted memory for those unaccessed members.
814 When bodies are allocated, we adjust the pointer back in memory by the
815 size of the part not allocated, so it's as if we allocated the full
816 structure. (But things will all go boom if you write to the part that
817 is "not there", because you'll be overwriting the last members of the
818 preceding structure in memory.)
820 We calculate the correction using the STRUCT_OFFSET macro on the first
821 member present. If the allocated structure is smaller (no initial NV
822 actually allocated) then the net effect is to subtract the size of the NV
823 from the pointer, to return a new pointer as if an initial NV were actually
824 allocated. (We were using structures named *_allocated for this, but
825 this turned out to be a subtle bug, because a structure without an NV
826 could have a lower alignment constraint, but the compiler is allowed to
827 optimised accesses based on the alignment constraint of the actual pointer
828 to the full structure, for example, using a single 64 bit load instruction
829 because it "knows" that two adjacent 32 bit members will be 8-byte aligned.)
831 This is the same trick as was used for NV and IV bodies. Ironically it
832 doesn't need to be used for NV bodies any more, because NV is now at
833 the start of the structure. IV bodies, and also in some builds NV bodies,
834 don't need it either, because they are no longer allocated.
836 In turn, the new_body_* allocators call S_new_body(), which invokes
837 new_body_inline macro, which takes a lock, and takes a body off the
838 linked list at PL_body_roots[sv_type], calling Perl_more_bodies() if
839 necessary to refresh an empty list. Then the lock is released, and
840 the body is returned.
842 Perl_more_bodies allocates a new arena, and carves it up into an array of N
843 bodies, which it strings into a linked list. It looks up arena-size
844 and body-size from the body_details table described below, thus
845 supporting the multiple body-types.
847 If PURIFY is defined, or PERL_ARENA_SIZE=0, arenas are not used, and
848 the (new|del)_X*V macros are mapped directly to malloc/free.
850 For each sv-type, struct body_details bodies_by_type[] carries
851 parameters which control these aspects of SV handling:
853 Arena_size determines whether arenas are used for this body type, and if
854 so, how big they are. PURIFY or PERL_ARENA_SIZE=0 set this field to
855 zero, forcing individual mallocs and frees.
857 Body_size determines how big a body is, and therefore how many fit into
858 each arena. Offset carries the body-pointer adjustment needed for
859 "ghost fields", and is used in *_allocated macros.
861 But its main purpose is to parameterize info needed in
862 Perl_sv_upgrade(). The info here dramatically simplifies the function
863 vs the implementation in 5.8.8, making it table-driven. All fields
864 are used for this, except for arena_size.
866 For the sv-types that have no bodies, arenas are not used, so those
867 PL_body_roots[sv_type] are unused, and can be overloaded. In
868 something of a special case, SVt_NULL is borrowed for HE arenas;
869 PL_body_roots[HE_SVSLOT=SVt_NULL] is filled by S_more_he, but the
870 bodies_by_type[SVt_NULL] slot is not used, as the table is not
875 struct body_details {
876 U8 body_size; /* Size to allocate */
877 U8 copy; /* Size of structure to copy (may be shorter) */
878 U8 offset; /* Size of unalloced ghost fields to first alloced field*/
879 PERL_BITFIELD8 type : 4; /* We have space for a sanity check. */
880 PERL_BITFIELD8 cant_upgrade : 1;/* Cannot upgrade this type */
881 PERL_BITFIELD8 zero_nv : 1; /* zero the NV when upgrading from this */
882 PERL_BITFIELD8 arena : 1; /* Allocated from an arena */
883 U32 arena_size; /* Size of arena to allocate */
891 /* With -DPURFIY we allocate everything directly, and don't use arenas.
892 This seems a rather elegant way to simplify some of the code below. */
893 #define HASARENA FALSE
895 #define HASARENA TRUE
897 #define NOARENA FALSE
899 /* Size the arenas to exactly fit a given number of bodies. A count
900 of 0 fits the max number bodies into a PERL_ARENA_SIZE.block,
901 simplifying the default. If count > 0, the arena is sized to fit
902 only that many bodies, allowing arenas to be used for large, rare
903 bodies (XPVFM, XPVIO) without undue waste. The arena size is
904 limited by PERL_ARENA_SIZE, so we can safely oversize the
907 #define FIT_ARENA0(body_size) \
908 ((size_t)(PERL_ARENA_SIZE / body_size) * body_size)
909 #define FIT_ARENAn(count,body_size) \
910 ( count * body_size <= PERL_ARENA_SIZE) \
911 ? count * body_size \
912 : FIT_ARENA0 (body_size)
913 #define FIT_ARENA(count,body_size) \
915 ? FIT_ARENAn (count, body_size) \
916 : FIT_ARENA0 (body_size))
918 /* Calculate the length to copy. Specifically work out the length less any
919 final padding the compiler needed to add. See the comment in sv_upgrade
920 for why copying the padding proved to be a bug. */
922 #define copy_length(type, last_member) \
923 STRUCT_OFFSET(type, last_member) \
924 + sizeof (((type*)SvANY((const SV *)0))->last_member)
926 static const struct body_details bodies_by_type[] = {
927 /* HEs use this offset for their arena. */
928 { 0, 0, 0, SVt_NULL, FALSE, NONV, NOARENA, 0 },
930 /* IVs are in the head, so the allocation size is 0. */
932 sizeof(IV), /* This is used to copy out the IV body. */
933 STRUCT_OFFSET(XPVIV, xiv_iv), SVt_IV, FALSE, NONV,
934 NOARENA /* IVS don't need an arena */, 0
939 STRUCT_OFFSET(XPVNV, xnv_u),
940 SVt_NV, FALSE, HADNV, NOARENA, 0 },
942 { sizeof(NV), sizeof(NV),
943 STRUCT_OFFSET(XPVNV, xnv_u),
944 SVt_NV, FALSE, HADNV, HASARENA, FIT_ARENA(0, sizeof(NV)) },
947 { sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur),
948 copy_length(XPV, xpv_len) - STRUCT_OFFSET(XPV, xpv_cur),
949 + STRUCT_OFFSET(XPV, xpv_cur),
950 SVt_PV, FALSE, NONV, HASARENA,
951 FIT_ARENA(0, sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur)) },
953 { sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur),
954 copy_length(XINVLIST, is_offset) - STRUCT_OFFSET(XPV, xpv_cur),
955 + STRUCT_OFFSET(XPV, xpv_cur),
956 SVt_INVLIST, TRUE, NONV, HASARENA,
957 FIT_ARENA(0, sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur)) },
959 { sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur),
960 copy_length(XPVIV, xiv_u) - STRUCT_OFFSET(XPV, xpv_cur),
961 + STRUCT_OFFSET(XPV, xpv_cur),
962 SVt_PVIV, FALSE, NONV, HASARENA,
963 FIT_ARENA(0, sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur)) },
965 { sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur),
966 copy_length(XPVNV, xnv_u) - STRUCT_OFFSET(XPV, xpv_cur),
967 + STRUCT_OFFSET(XPV, xpv_cur),
968 SVt_PVNV, FALSE, HADNV, HASARENA,
969 FIT_ARENA(0, sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur)) },
971 { sizeof(XPVMG), copy_length(XPVMG, xnv_u), 0, SVt_PVMG, FALSE, HADNV,
972 HASARENA, FIT_ARENA(0, sizeof(XPVMG)) },
977 SVt_REGEXP, TRUE, NONV, HASARENA,
978 FIT_ARENA(0, sizeof(regexp))
981 { sizeof(XPVGV), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV,
982 HASARENA, FIT_ARENA(0, sizeof(XPVGV)) },
984 { sizeof(XPVLV), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV,
985 HASARENA, FIT_ARENA(0, sizeof(XPVLV)) },
988 copy_length(XPVAV, xav_alloc),
990 SVt_PVAV, TRUE, NONV, HASARENA,
991 FIT_ARENA(0, sizeof(XPVAV)) },
994 copy_length(XPVHV, xhv_max),
996 SVt_PVHV, TRUE, NONV, HASARENA,
997 FIT_ARENA(0, sizeof(XPVHV)) },
1002 SVt_PVCV, TRUE, NONV, HASARENA,
1003 FIT_ARENA(0, sizeof(XPVCV)) },
1008 SVt_PVFM, TRUE, NONV, NOARENA,
1009 FIT_ARENA(20, sizeof(XPVFM)) },
1014 SVt_PVIO, TRUE, NONV, HASARENA,
1015 FIT_ARENA(24, sizeof(XPVIO)) },
1018 #define new_body_allocated(sv_type) \
1019 (void *)((char *)S_new_body(aTHX_ sv_type) \
1020 - bodies_by_type[sv_type].offset)
1022 /* return a thing to the free list */
1024 #define del_body(thing, root) \
1026 void ** const thing_copy = (void **)thing; \
1027 *thing_copy = *root; \
1028 *root = (void*)thing_copy; \
1032 #if !(NVSIZE <= IVSIZE)
1033 # define new_XNV() safemalloc(sizeof(XPVNV))
1035 #define new_XPVNV() safemalloc(sizeof(XPVNV))
1036 #define new_XPVMG() safemalloc(sizeof(XPVMG))
1038 #define del_XPVGV(p) safefree(p)
1042 #if !(NVSIZE <= IVSIZE)
1043 # define new_XNV() new_body_allocated(SVt_NV)
1045 #define new_XPVNV() new_body_allocated(SVt_PVNV)
1046 #define new_XPVMG() new_body_allocated(SVt_PVMG)
1048 #define del_XPVGV(p) del_body(p + bodies_by_type[SVt_PVGV].offset, \
1049 &PL_body_roots[SVt_PVGV])
1053 /* no arena for you! */
1055 #define new_NOARENA(details) \
1056 safemalloc((details)->body_size + (details)->offset)
1057 #define new_NOARENAZ(details) \
1058 safecalloc((details)->body_size + (details)->offset, 1)
1061 Perl_more_bodies (pTHX_ const svtype sv_type, const size_t body_size,
1062 const size_t arena_size)
1064 void ** const root = &PL_body_roots[sv_type];
1065 struct arena_desc *adesc;
1066 struct arena_set *aroot = (struct arena_set *) PL_body_arenas;
1070 const size_t good_arena_size = Perl_malloc_good_size(arena_size);
1071 #if defined(DEBUGGING) && defined(PERL_GLOBAL_STRUCT)
1074 #if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT_PRIVATE)
1075 static bool done_sanity_check;
1077 /* PERL_GLOBAL_STRUCT_PRIVATE cannot coexist with global
1078 * variables like done_sanity_check. */
1079 if (!done_sanity_check) {
1080 unsigned int i = SVt_LAST;
1082 done_sanity_check = TRUE;
1085 assert (bodies_by_type[i].type == i);
1091 /* may need new arena-set to hold new arena */
1092 if (!aroot || aroot->curr >= aroot->set_size) {
1093 struct arena_set *newroot;
1094 Newxz(newroot, 1, struct arena_set);
1095 newroot->set_size = ARENAS_PER_SET;
1096 newroot->next = aroot;
1098 PL_body_arenas = (void *) newroot;
1099 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot));
1102 /* ok, now have arena-set with at least 1 empty/available arena-desc */
1103 curr = aroot->curr++;
1104 adesc = &(aroot->set[curr]);
1105 assert(!adesc->arena);
1107 Newx(adesc->arena, good_arena_size, char);
1108 adesc->size = good_arena_size;
1109 adesc->utype = sv_type;
1110 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %" UVuf "\n",
1111 curr, (void*)adesc->arena, (UV)good_arena_size));
1113 start = (char *) adesc->arena;
1115 /* Get the address of the byte after the end of the last body we can fit.
1116 Remember, this is integer division: */
1117 end = start + good_arena_size / body_size * body_size;
1119 /* computed count doesn't reflect the 1st slot reservation */
1120 #if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE)
1121 DEBUG_m(PerlIO_printf(Perl_debug_log,
1122 "arena %p end %p arena-size %d (from %d) type %d "
1124 (void*)start, (void*)end, (int)good_arena_size,
1125 (int)arena_size, sv_type, (int)body_size,
1126 (int)good_arena_size / (int)body_size));
1128 DEBUG_m(PerlIO_printf(Perl_debug_log,
1129 "arena %p end %p arena-size %d type %d size %d ct %d\n",
1130 (void*)start, (void*)end,
1131 (int)arena_size, sv_type, (int)body_size,
1132 (int)good_arena_size / (int)body_size));
1134 *root = (void *)start;
1137 /* Where the next body would start: */
1138 char * const next = start + body_size;
1141 /* This is the last body: */
1142 assert(next == end);
1144 *(void **)start = 0;
1148 *(void**) start = (void *)next;
1153 /* grab a new thing from the free list, allocating more if necessary.
1154 The inline version is used for speed in hot routines, and the
1155 function using it serves the rest (unless PURIFY).
1157 #define new_body_inline(xpv, sv_type) \
1159 void ** const r3wt = &PL_body_roots[sv_type]; \
1160 xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \
1161 ? *((void **)(r3wt)) : Perl_more_bodies(aTHX_ sv_type, \
1162 bodies_by_type[sv_type].body_size,\
1163 bodies_by_type[sv_type].arena_size)); \
1164 *(r3wt) = *(void**)(xpv); \
1170 S_new_body(pTHX_ const svtype sv_type)
1173 new_body_inline(xpv, sv_type);
1179 static const struct body_details fake_rv =
1180 { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 };
1183 =for apidoc sv_upgrade
1185 Upgrade an SV to a more complex form. Generally adds a new body type to the
1186 SV, then copies across as much information as possible from the old body.
1187 It croaks if the SV is already in a more complex form than requested. You
1188 generally want to use the C<SvUPGRADE> macro wrapper, which checks the type
1189 before calling C<sv_upgrade>, and hence does not croak. See also
1196 Perl_sv_upgrade(pTHX_ SV *const sv, svtype new_type)
1200 const svtype old_type = SvTYPE(sv);
1201 const struct body_details *new_type_details;
1202 const struct body_details *old_type_details
1203 = bodies_by_type + old_type;
1204 SV *referent = NULL;
1206 PERL_ARGS_ASSERT_SV_UPGRADE;
1208 if (old_type == new_type)
1211 /* This clause was purposefully added ahead of the early return above to
1212 the shared string hackery for (sort {$a <=> $b} keys %hash), with the
1213 inference by Nick I-S that it would fix other troublesome cases. See
1214 changes 7162, 7163 (f130fd4589cf5fbb24149cd4db4137c8326f49c1 and parent)
1216 Given that shared hash key scalars are no longer PVIV, but PV, there is
1217 no longer need to unshare so as to free up the IVX slot for its proper
1218 purpose. So it's safe to move the early return earlier. */
1220 if (new_type > SVt_PVMG && SvIsCOW(sv)) {
1221 sv_force_normal_flags(sv, 0);
1224 old_body = SvANY(sv);
1226 /* Copying structures onto other structures that have been neatly zeroed
1227 has a subtle gotcha. Consider XPVMG
1229 +------+------+------+------+------+-------+-------+
1230 | NV | CUR | LEN | IV | MAGIC | STASH |
1231 +------+------+------+------+------+-------+-------+
1232 0 4 8 12 16 20 24 28
1234 where NVs are aligned to 8 bytes, so that sizeof that structure is
1235 actually 32 bytes long, with 4 bytes of padding at the end:
1237 +------+------+------+------+------+-------+-------+------+
1238 | NV | CUR | LEN | IV | MAGIC | STASH | ??? |
1239 +------+------+------+------+------+-------+-------+------+
1240 0 4 8 12 16 20 24 28 32
1242 so what happens if you allocate memory for this structure:
1244 +------+------+------+------+------+-------+-------+------+------+...
1245 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME |
1246 +------+------+------+------+------+-------+-------+------+------+...
1247 0 4 8 12 16 20 24 28 32 36
1249 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you
1250 expect, because you copy the area marked ??? onto GP. Now, ??? may have
1251 started out as zero once, but it's quite possible that it isn't. So now,
1252 rather than a nicely zeroed GP, you have it pointing somewhere random.
1255 (In fact, GP ends up pointing at a previous GP structure, because the
1256 principle cause of the padding in XPVMG getting garbage is a copy of
1257 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now
1258 this happens to be moot because XPVGV has been re-ordered, with GP
1259 no longer after STASH)
1261 So we are careful and work out the size of used parts of all the
1269 referent = SvRV(sv);
1270 old_type_details = &fake_rv;
1271 if (new_type == SVt_NV)
1272 new_type = SVt_PVNV;
1274 if (new_type < SVt_PVIV) {
1275 new_type = (new_type == SVt_NV)
1276 ? SVt_PVNV : SVt_PVIV;
1281 if (new_type < SVt_PVNV) {
1282 new_type = SVt_PVNV;
1286 assert(new_type > SVt_PV);
1287 STATIC_ASSERT_STMT(SVt_IV < SVt_PV);
1288 STATIC_ASSERT_STMT(SVt_NV < SVt_PV);
1295 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena,
1296 there's no way that it can be safely upgraded, because perl.c
1297 expects to Safefree(SvANY(PL_mess_sv)) */
1298 assert(sv != PL_mess_sv);
1301 if (UNLIKELY(old_type_details->cant_upgrade))
1302 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf,
1303 sv_reftype(sv, 0), (UV) old_type, (UV) new_type);
1306 if (UNLIKELY(old_type > new_type))
1307 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d",
1308 (int)old_type, (int)new_type);
1310 new_type_details = bodies_by_type + new_type;
1312 SvFLAGS(sv) &= ~SVTYPEMASK;
1313 SvFLAGS(sv) |= new_type;
1315 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of
1316 the return statements above will have triggered. */
1317 assert (new_type != SVt_NULL);
1320 assert(old_type == SVt_NULL);
1321 SET_SVANY_FOR_BODYLESS_IV(sv);
1325 assert(old_type == SVt_NULL);
1326 #if NVSIZE <= IVSIZE
1327 SET_SVANY_FOR_BODYLESS_NV(sv);
1329 SvANY(sv) = new_XNV();
1335 assert(new_type_details->body_size);
1338 assert(new_type_details->arena);
1339 assert(new_type_details->arena_size);
1340 /* This points to the start of the allocated area. */
1341 new_body_inline(new_body, new_type);
1342 Zero(new_body, new_type_details->body_size, char);
1343 new_body = ((char *)new_body) - new_type_details->offset;
1345 /* We always allocated the full length item with PURIFY. To do this
1346 we fake things so that arena is false for all 16 types.. */
1347 new_body = new_NOARENAZ(new_type_details);
1349 SvANY(sv) = new_body;
1350 if (new_type == SVt_PVAV) {
1354 if (old_type_details->body_size) {
1357 /* It will have been zeroed when the new body was allocated.
1358 Lets not write to it, in case it confuses a write-back
1364 #ifndef NODEFAULT_SHAREKEYS
1365 HvSHAREKEYS_on(sv); /* key-sharing on by default */
1367 /* start with PERL_HASH_DEFAULT_HvMAX+1 buckets: */
1368 HvMAX(sv) = PERL_HASH_DEFAULT_HvMAX;
1371 /* SVt_NULL isn't the only thing upgraded to AV or HV.
1372 The target created by newSVrv also is, and it can have magic.
1373 However, it never has SvPVX set.
1375 if (old_type == SVt_IV) {
1377 } else if (old_type >= SVt_PV) {
1378 assert(SvPVX_const(sv) == 0);
1381 if (old_type >= SVt_PVMG) {
1382 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic);
1383 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash);
1385 sv->sv_u.svu_array = NULL; /* or svu_hash */
1390 /* XXX Is this still needed? Was it ever needed? Surely as there is
1391 no route from NV to PVIV, NOK can never be true */
1392 assert(!SvNOKp(sv));
1406 assert(new_type_details->body_size);
1407 /* We always allocated the full length item with PURIFY. To do this
1408 we fake things so that arena is false for all 16 types.. */
1409 if(new_type_details->arena) {
1410 /* This points to the start of the allocated area. */
1411 new_body_inline(new_body, new_type);
1412 Zero(new_body, new_type_details->body_size, char);
1413 new_body = ((char *)new_body) - new_type_details->offset;
1415 new_body = new_NOARENAZ(new_type_details);
1417 SvANY(sv) = new_body;
1419 if (old_type_details->copy) {
1420 /* There is now the potential for an upgrade from something without
1421 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */
1422 int offset = old_type_details->offset;
1423 int length = old_type_details->copy;
1425 if (new_type_details->offset > old_type_details->offset) {
1426 const int difference
1427 = new_type_details->offset - old_type_details->offset;
1428 offset += difference;
1429 length -= difference;
1431 assert (length >= 0);
1433 Copy((char *)old_body + offset, (char *)new_body + offset, length,
1437 #ifndef NV_ZERO_IS_ALLBITS_ZERO
1438 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a
1439 * correct 0.0 for us. Otherwise, if the old body didn't have an
1440 * NV slot, but the new one does, then we need to initialise the
1441 * freshly created NV slot with whatever the correct bit pattern is
1443 if (old_type_details->zero_nv && !new_type_details->zero_nv
1444 && !isGV_with_GP(sv))
1448 if (UNLIKELY(new_type == SVt_PVIO)) {
1449 IO * const io = MUTABLE_IO(sv);
1450 GV *iogv = gv_fetchpvs("IO::File::", GV_ADD, SVt_PVHV);
1453 /* Clear the stashcache because a new IO could overrule a package
1455 DEBUG_o(Perl_deb(aTHX_ "sv_upgrade clearing PL_stashcache\n"));
1456 hv_clear(PL_stashcache);
1458 SvSTASH_set(io, MUTABLE_HV(SvREFCNT_inc(GvHV(iogv))));
1459 IoPAGE_LEN(sv) = 60;
1461 if (old_type < SVt_PV) {
1462 /* referent will be NULL unless the old type was SVt_IV emulating
1464 sv->sv_u.svu_rv = referent;
1468 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1469 (unsigned long)new_type);
1472 /* if this is zero, this is a body-less SVt_NULL, SVt_IV/SVt_RV,
1473 and sometimes SVt_NV */
1474 if (old_type_details->body_size) {
1478 /* Note that there is an assumption that all bodies of types that
1479 can be upgraded came from arenas. Only the more complex non-
1480 upgradable types are allowed to be directly malloc()ed. */
1481 assert(old_type_details->arena);
1482 del_body((void*)((char*)old_body + old_type_details->offset),
1483 &PL_body_roots[old_type]);
1489 =for apidoc sv_backoff
1491 Remove any string offset. You should normally use the C<SvOOK_off> macro
1497 /* prior to 5.000 stable, this function returned the new OOK-less SvFLAGS
1498 prior to 5.23.4 this function always returned 0
1502 Perl_sv_backoff(SV *const sv)
1505 const char * const s = SvPVX_const(sv);
1507 PERL_ARGS_ASSERT_SV_BACKOFF;
1510 assert(SvTYPE(sv) != SVt_PVHV);
1511 assert(SvTYPE(sv) != SVt_PVAV);
1513 SvOOK_offset(sv, delta);
1515 SvLEN_set(sv, SvLEN(sv) + delta);
1516 SvPV_set(sv, SvPVX(sv) - delta);
1517 SvFLAGS(sv) &= ~SVf_OOK;
1518 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
1523 /* forward declaration */
1524 static void S_sv_uncow(pTHX_ SV * const sv, const U32 flags);
1530 Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1531 upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1532 Use the C<SvGROW> wrapper instead.
1539 Perl_sv_grow(pTHX_ SV *const sv, STRLEN newlen)
1543 PERL_ARGS_ASSERT_SV_GROW;
1547 if (SvTYPE(sv) < SVt_PV) {
1548 sv_upgrade(sv, SVt_PV);
1549 s = SvPVX_mutable(sv);
1551 else if (SvOOK(sv)) { /* pv is offset? */
1553 s = SvPVX_mutable(sv);
1554 if (newlen > SvLEN(sv))
1555 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
1559 if (SvIsCOW(sv)) S_sv_uncow(aTHX_ sv, 0);
1560 s = SvPVX_mutable(sv);
1563 #ifdef PERL_COPY_ON_WRITE
1564 /* the new COW scheme uses SvPVX(sv)[SvLEN(sv)-1] (if spare)
1565 * to store the COW count. So in general, allocate one more byte than
1566 * asked for, to make it likely this byte is always spare: and thus
1567 * make more strings COW-able.
1569 * Only increment if the allocation isn't MEM_SIZE_MAX,
1570 * otherwise it will wrap to 0.
1572 if ( newlen != MEM_SIZE_MAX )
1576 #if defined(PERL_USE_MALLOC_SIZE) && defined(Perl_safesysmalloc_size)
1577 #define PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1580 if (newlen > SvLEN(sv)) { /* need more room? */
1581 STRLEN minlen = SvCUR(sv);
1582 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
1583 if (newlen < minlen)
1585 #ifndef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1587 /* Don't round up on the first allocation, as odds are pretty good that
1588 * the initial request is accurate as to what is really needed */
1590 STRLEN rounded = PERL_STRLEN_ROUNDUP(newlen);
1591 if (rounded > 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, 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<L</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))
1650 /* diag_listed_as: Can't coerce %s to %s in %s */
1651 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1653 NOT_REACHED; /* NOTREACHED */
1657 (void)SvIOK_only(sv); /* validate number */
1663 =for apidoc sv_setiv_mg
1665 Like C<sv_setiv>, but also handles 'set' magic.
1671 Perl_sv_setiv_mg(pTHX_ SV *const sv, const IV i)
1673 PERL_ARGS_ASSERT_SV_SETIV_MG;
1680 =for apidoc sv_setuv
1682 Copies an unsigned integer into the given SV, upgrading first if necessary.
1683 Does not handle 'set' magic. See also C<L</sv_setuv_mg>>.
1689 Perl_sv_setuv(pTHX_ SV *const sv, const UV u)
1691 PERL_ARGS_ASSERT_SV_SETUV;
1693 /* With the if statement to ensure that integers are stored as IVs whenever
1695 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
1698 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1700 If you wish to remove the following if statement, so that this routine
1701 (and its callers) always return UVs, please benchmark to see what the
1702 effect is. Modern CPUs may be different. Or may not :-)
1704 if (u <= (UV)IV_MAX) {
1705 sv_setiv(sv, (IV)u);
1714 =for apidoc sv_setuv_mg
1716 Like C<sv_setuv>, but also handles 'set' magic.
1722 Perl_sv_setuv_mg(pTHX_ SV *const sv, const UV u)
1724 PERL_ARGS_ASSERT_SV_SETUV_MG;
1731 =for apidoc sv_setnv
1733 Copies a double into the given SV, upgrading first if necessary.
1734 Does not handle 'set' magic. See also C<L</sv_setnv_mg>>.
1740 Perl_sv_setnv(pTHX_ SV *const sv, const NV num)
1742 PERL_ARGS_ASSERT_SV_SETNV;
1744 SV_CHECK_THINKFIRST_COW_DROP(sv);
1745 switch (SvTYPE(sv)) {
1748 sv_upgrade(sv, SVt_NV);
1752 sv_upgrade(sv, SVt_PVNV);
1756 if (!isGV_with_GP(sv))
1764 /* diag_listed_as: Can't coerce %s to %s in %s */
1765 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
1767 NOT_REACHED; /* NOTREACHED */
1772 (void)SvNOK_only(sv); /* validate number */
1777 =for apidoc sv_setnv_mg
1779 Like C<sv_setnv>, but also handles 'set' magic.
1785 Perl_sv_setnv_mg(pTHX_ SV *const sv, const NV num)
1787 PERL_ARGS_ASSERT_SV_SETNV_MG;
1793 /* Return a cleaned-up, printable version of sv, for non-numeric, or
1794 * not incrementable warning display.
1795 * Originally part of S_not_a_number().
1796 * The return value may be != tmpbuf.
1800 S_sv_display(pTHX_ SV *const sv, char *tmpbuf, STRLEN tmpbuf_size) {
1803 PERL_ARGS_ASSERT_SV_DISPLAY;
1806 SV *dsv = newSVpvs_flags("", SVs_TEMP);
1807 pv = sv_uni_display(dsv, sv, 32, UNI_DISPLAY_ISPRINT);
1810 const char * const limit = tmpbuf + tmpbuf_size - 8;
1811 /* each *s can expand to 4 chars + "...\0",
1812 i.e. need room for 8 chars */
1814 const char *s = SvPVX_const(sv);
1815 const char * const end = s + SvCUR(sv);
1816 for ( ; s < end && d < limit; s++ ) {
1818 if (! isASCII(ch) && !isPRINT_LC(ch)) {
1822 /* Map to ASCII "equivalent" of Latin1 */
1823 ch = LATIN1_TO_NATIVE(NATIVE_TO_LATIN1(ch) & 127);
1829 else if (ch == '\r') {
1833 else if (ch == '\f') {
1837 else if (ch == '\\') {
1841 else if (ch == '\0') {
1845 else if (isPRINT_LC(ch))
1864 /* Print an "isn't numeric" warning, using a cleaned-up,
1865 * printable version of the offending string
1869 S_not_a_number(pTHX_ SV *const sv)
1874 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1876 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1879 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1880 /* diag_listed_as: Argument "%s" isn't numeric%s */
1881 "Argument \"%s\" isn't numeric in %s", pv,
1884 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1885 /* diag_listed_as: Argument "%s" isn't numeric%s */
1886 "Argument \"%s\" isn't numeric", pv);
1890 S_not_incrementable(pTHX_ SV *const sv) {
1894 PERL_ARGS_ASSERT_NOT_INCREMENTABLE;
1896 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1898 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1899 "Argument \"%s\" treated as 0 in increment (++)", pv);
1903 =for apidoc looks_like_number
1905 Test if the content of an SV looks like a number (or is a number).
1906 C<Inf> and C<Infinity> are treated as numbers (so will not issue a
1907 non-numeric warning), even if your C<atof()> doesn't grok them. Get-magic is
1914 Perl_looks_like_number(pTHX_ SV *const sv)
1920 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1922 if (SvPOK(sv) || SvPOKp(sv)) {
1923 sbegin = SvPV_nomg_const(sv, len);
1926 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
1927 numtype = grok_number(sbegin, len, NULL);
1928 return ((numtype & IS_NUMBER_TRAILING)) ? 0 : numtype;
1932 S_glob_2number(pTHX_ GV * const gv)
1934 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1936 /* We know that all GVs stringify to something that is not-a-number,
1937 so no need to test that. */
1938 if (ckWARN(WARN_NUMERIC))
1940 SV *const buffer = sv_newmortal();
1941 gv_efullname3(buffer, gv, "*");
1942 not_a_number(buffer);
1944 /* We just want something true to return, so that S_sv_2iuv_common
1945 can tail call us and return true. */
1949 /* Actually, ISO C leaves conversion of UV to IV undefined, but
1950 until proven guilty, assume that things are not that bad... */
1955 As 64 bit platforms often have an NV that doesn't preserve all bits of
1956 an IV (an assumption perl has been based on to date) it becomes necessary
1957 to remove the assumption that the NV always carries enough precision to
1958 recreate the IV whenever needed, and that the NV is the canonical form.
1959 Instead, IV/UV and NV need to be given equal rights. So as to not lose
1960 precision as a side effect of conversion (which would lead to insanity
1961 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
1962 1) to distinguish between IV/UV/NV slots that have a valid conversion cached
1963 where precision was lost, and IV/UV/NV slots that have a valid conversion
1964 which has lost no precision
1965 2) to ensure that if a numeric conversion to one form is requested that
1966 would lose precision, the precise conversion (or differently
1967 imprecise conversion) is also performed and cached, to prevent
1968 requests for different numeric formats on the same SV causing
1969 lossy conversion chains. (lossless conversion chains are perfectly
1974 SvIOKp is true if the IV slot contains a valid value
1975 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1976 SvNOKp is true if the NV slot contains a valid value
1977 SvNOK is true only if the NV value is accurate
1980 while converting from PV to NV, check to see if converting that NV to an
1981 IV(or UV) would lose accuracy over a direct conversion from PV to
1982 IV(or UV). If it would, cache both conversions, return NV, but mark
1983 SV as IOK NOKp (ie not NOK).
1985 While converting from PV to IV, check to see if converting that IV to an
1986 NV would lose accuracy over a direct conversion from PV to NV. If it
1987 would, cache both conversions, flag similarly.
1989 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
1990 correctly because if IV & NV were set NV *always* overruled.
1991 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
1992 changes - now IV and NV together means that the two are interchangeable:
1993 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
1995 The benefit of this is that operations such as pp_add know that if
1996 SvIOK is true for both left and right operands, then integer addition
1997 can be used instead of floating point (for cases where the result won't
1998 overflow). Before, floating point was always used, which could lead to
1999 loss of precision compared with integer addition.
2001 * making IV and NV equal status should make maths accurate on 64 bit
2003 * may speed up maths somewhat if pp_add and friends start to use
2004 integers when possible instead of fp. (Hopefully the overhead in
2005 looking for SvIOK and checking for overflow will not outweigh the
2006 fp to integer speedup)
2007 * will slow down integer operations (callers of SvIV) on "inaccurate"
2008 values, as the change from SvIOK to SvIOKp will cause a call into
2009 sv_2iv each time rather than a macro access direct to the IV slot
2010 * should speed up number->string conversion on integers as IV is
2011 favoured when IV and NV are equally accurate
2013 ####################################################################
2014 You had better be using SvIOK_notUV if you want an IV for arithmetic:
2015 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
2016 On the other hand, SvUOK is true iff UV.
2017 ####################################################################
2019 Your mileage will vary depending your CPU's relative fp to integer
2023 #ifndef NV_PRESERVES_UV
2024 # define IS_NUMBER_UNDERFLOW_IV 1
2025 # define IS_NUMBER_UNDERFLOW_UV 2
2026 # define IS_NUMBER_IV_AND_UV 2
2027 # define IS_NUMBER_OVERFLOW_IV 4
2028 # define IS_NUMBER_OVERFLOW_UV 5
2030 /* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
2032 /* For sv_2nv these three cases are "SvNOK and don't bother casting" */
2034 S_sv_2iuv_non_preserve(pTHX_ SV *const sv
2040 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
2041 PERL_UNUSED_CONTEXT;
2043 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));
2044 if (SvNVX(sv) < (NV)IV_MIN) {
2045 (void)SvIOKp_on(sv);
2047 SvIV_set(sv, IV_MIN);
2048 return IS_NUMBER_UNDERFLOW_IV;
2050 if (SvNVX(sv) > (NV)UV_MAX) {
2051 (void)SvIOKp_on(sv);
2054 SvUV_set(sv, UV_MAX);
2055 return IS_NUMBER_OVERFLOW_UV;
2057 (void)SvIOKp_on(sv);
2059 /* Can't use strtol etc to convert this string. (See truth table in
2061 if (SvNVX(sv) <= (UV)IV_MAX) {
2062 SvIV_set(sv, I_V(SvNVX(sv)));
2063 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2064 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
2066 /* Integer is imprecise. NOK, IOKp */
2068 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
2071 SvUV_set(sv, U_V(SvNVX(sv)));
2072 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2073 if (SvUVX(sv) == UV_MAX) {
2074 /* As we know that NVs don't preserve UVs, UV_MAX cannot
2075 possibly be preserved by NV. Hence, it must be overflow.
2077 return IS_NUMBER_OVERFLOW_UV;
2079 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
2081 /* Integer is imprecise. NOK, IOKp */
2083 return IS_NUMBER_OVERFLOW_IV;
2085 #endif /* !NV_PRESERVES_UV*/
2087 /* If numtype is infnan, set the NV of the sv accordingly.
2088 * If numtype is anything else, try setting the NV using Atof(PV). */
2090 # pragma warning(push)
2091 # pragma warning(disable:4756;disable:4056)
2094 S_sv_setnv(pTHX_ SV* sv, int numtype)
2096 bool pok = cBOOL(SvPOK(sv));
2099 if ((numtype & IS_NUMBER_INFINITY)) {
2100 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -NV_INF : NV_INF);
2105 if ((numtype & IS_NUMBER_NAN)) {
2106 SvNV_set(sv, NV_NAN);
2111 SvNV_set(sv, Atof(SvPVX_const(sv)));
2112 /* Purposefully no true nok here, since we don't want to blow
2113 * away the possible IOK/UV of an existing sv. */
2116 SvNOK_only(sv); /* No IV or UV please, this is pure infnan. */
2118 SvPOK_on(sv); /* PV is okay, though. */
2122 # pragma warning(pop)
2126 S_sv_2iuv_common(pTHX_ SV *const sv)
2128 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
2131 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
2132 * without also getting a cached IV/UV from it at the same time
2133 * (ie PV->NV conversion should detect loss of accuracy and cache
2134 * IV or UV at same time to avoid this. */
2135 /* IV-over-UV optimisation - choose to cache IV if possible */
2137 if (SvTYPE(sv) == SVt_NV)
2138 sv_upgrade(sv, SVt_PVNV);
2140 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
2141 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
2142 certainly cast into the IV range at IV_MAX, whereas the correct
2143 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
2145 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2146 if (Perl_isnan(SvNVX(sv))) {
2152 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2153 SvIV_set(sv, I_V(SvNVX(sv)));
2154 if (SvNVX(sv) == (NV) SvIVX(sv)
2155 #ifndef NV_PRESERVES_UV
2156 && SvIVX(sv) != IV_MIN /* avoid negating IV_MIN below */
2157 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2158 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2159 /* Don't flag it as "accurately an integer" if the number
2160 came from a (by definition imprecise) NV operation, and
2161 we're outside the range of NV integer precision */
2165 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2167 /* scalar has trailing garbage, eg "42a" */
2169 DEBUG_c(PerlIO_printf(Perl_debug_log,
2170 "0x%" UVxf " iv(%" NVgf " => %" IVdf ") (precise)\n",
2176 /* IV not precise. No need to convert from PV, as NV
2177 conversion would already have cached IV if it detected
2178 that PV->IV would be better than PV->NV->IV
2179 flags already correct - don't set public IOK. */
2180 DEBUG_c(PerlIO_printf(Perl_debug_log,
2181 "0x%" UVxf " iv(%" NVgf " => %" IVdf ") (imprecise)\n",
2186 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2187 but the cast (NV)IV_MIN rounds to a the value less (more
2188 negative) than IV_MIN which happens to be equal to SvNVX ??
2189 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2190 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2191 (NV)UVX == NVX are both true, but the values differ. :-(
2192 Hopefully for 2s complement IV_MIN is something like
2193 0x8000000000000000 which will be exact. NWC */
2196 SvUV_set(sv, U_V(SvNVX(sv)));
2198 (SvNVX(sv) == (NV) SvUVX(sv))
2199 #ifndef NV_PRESERVES_UV
2200 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2201 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2202 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2203 /* Don't flag it as "accurately an integer" if the number
2204 came from a (by definition imprecise) NV operation, and
2205 we're outside the range of NV integer precision */
2211 DEBUG_c(PerlIO_printf(Perl_debug_log,
2212 "0x%" UVxf " 2iv(%" UVuf " => %" IVdf ") (as unsigned)\n",
2218 else if (SvPOKp(sv)) {
2221 const char *s = SvPVX_const(sv);
2222 const STRLEN cur = SvCUR(sv);
2224 /* short-cut for a single digit string like "1" */
2229 if (SvTYPE(sv) < SVt_PVIV)
2230 sv_upgrade(sv, SVt_PVIV);
2232 SvIV_set(sv, (IV)(c - '0'));
2237 numtype = grok_number(s, cur, &value);
2238 /* We want to avoid a possible problem when we cache an IV/ a UV which
2239 may be later translated to an NV, and the resulting NV is not
2240 the same as the direct translation of the initial string
2241 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2242 be careful to ensure that the value with the .456 is around if the
2243 NV value is requested in the future).
2245 This means that if we cache such an IV/a UV, we need to cache the
2246 NV as well. Moreover, we trade speed for space, and do not
2247 cache the NV if we are sure it's not needed.
2250 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2251 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2252 == IS_NUMBER_IN_UV) {
2253 /* It's definitely an integer, only upgrade to PVIV */
2254 if (SvTYPE(sv) < SVt_PVIV)
2255 sv_upgrade(sv, SVt_PVIV);
2257 } else if (SvTYPE(sv) < SVt_PVNV)
2258 sv_upgrade(sv, SVt_PVNV);
2260 if ((numtype & (IS_NUMBER_INFINITY | IS_NUMBER_NAN))) {
2261 if (ckWARN(WARN_NUMERIC) && ((numtype & IS_NUMBER_TRAILING)))
2263 S_sv_setnv(aTHX_ sv, numtype);
2267 /* If NVs preserve UVs then we only use the UV value if we know that
2268 we aren't going to call atof() below. If NVs don't preserve UVs
2269 then the value returned may have more precision than atof() will
2270 return, even though value isn't perfectly accurate. */
2271 if ((numtype & (IS_NUMBER_IN_UV
2272 #ifdef NV_PRESERVES_UV
2275 )) == IS_NUMBER_IN_UV) {
2276 /* This won't turn off the public IOK flag if it was set above */
2277 (void)SvIOKp_on(sv);
2279 if (!(numtype & IS_NUMBER_NEG)) {
2281 if (value <= (UV)IV_MAX) {
2282 SvIV_set(sv, (IV)value);
2284 /* it didn't overflow, and it was positive. */
2285 SvUV_set(sv, value);
2289 /* 2s complement assumption */
2290 if (value <= (UV)IV_MIN) {
2291 SvIV_set(sv, value == (UV)IV_MIN
2292 ? IV_MIN : -(IV)value);
2294 /* Too negative for an IV. This is a double upgrade, but
2295 I'm assuming it will be rare. */
2296 if (SvTYPE(sv) < SVt_PVNV)
2297 sv_upgrade(sv, SVt_PVNV);
2301 SvNV_set(sv, -(NV)value);
2302 SvIV_set(sv, IV_MIN);
2306 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2307 will be in the previous block to set the IV slot, and the next
2308 block to set the NV slot. So no else here. */
2310 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2311 != IS_NUMBER_IN_UV) {
2312 /* It wasn't an (integer that doesn't overflow the UV). */
2313 S_sv_setnv(aTHX_ sv, numtype);
2315 if (! numtype && ckWARN(WARN_NUMERIC))
2318 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2iv(%" NVgf ")\n",
2319 PTR2UV(sv), SvNVX(sv)));
2321 #ifdef NV_PRESERVES_UV
2322 (void)SvIOKp_on(sv);
2324 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2325 if (Perl_isnan(SvNVX(sv))) {
2331 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2332 SvIV_set(sv, I_V(SvNVX(sv)));
2333 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2336 NOOP; /* Integer is imprecise. NOK, IOKp */
2338 /* UV will not work better than IV */
2340 if (SvNVX(sv) > (NV)UV_MAX) {
2342 /* Integer is inaccurate. NOK, IOKp, is UV */
2343 SvUV_set(sv, UV_MAX);
2345 SvUV_set(sv, U_V(SvNVX(sv)));
2346 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2347 NV preservse UV so can do correct comparison. */
2348 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2351 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
2356 #else /* NV_PRESERVES_UV */
2357 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2358 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
2359 /* The IV/UV slot will have been set from value returned by
2360 grok_number above. The NV slot has just been set using
2363 assert (SvIOKp(sv));
2365 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2366 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2367 /* Small enough to preserve all bits. */
2368 (void)SvIOKp_on(sv);
2370 SvIV_set(sv, I_V(SvNVX(sv)));
2371 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2373 /* Assumption: first non-preserved integer is < IV_MAX,
2374 this NV is in the preserved range, therefore: */
2375 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2377 Perl_croak(aTHX_ "sv_2iv assumed (U_V(fabs((double)SvNVX(sv))) < (UV)IV_MAX) but SvNVX(sv)=%" NVgf " U_V is 0x%" UVxf ", IV_MAX is 0x%" UVxf "\n", SvNVX(sv), U_V(SvNVX(sv)), (UV)IV_MAX);
2381 0 0 already failed to read UV.
2382 0 1 already failed to read UV.
2383 1 0 you won't get here in this case. IV/UV
2384 slot set, public IOK, Atof() unneeded.
2385 1 1 already read UV.
2386 so there's no point in sv_2iuv_non_preserve() attempting
2387 to use atol, strtol, strtoul etc. */
2389 sv_2iuv_non_preserve (sv, numtype);
2391 sv_2iuv_non_preserve (sv);
2395 #endif /* NV_PRESERVES_UV */
2396 /* It might be more code efficient to go through the entire logic above
2397 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2398 gets complex and potentially buggy, so more programmer efficient
2399 to do it this way, by turning off the public flags: */
2401 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2405 if (isGV_with_GP(sv))
2406 return glob_2number(MUTABLE_GV(sv));
2408 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2410 if (SvTYPE(sv) < SVt_IV)
2411 /* Typically the caller expects that sv_any is not NULL now. */
2412 sv_upgrade(sv, SVt_IV);
2413 /* Return 0 from the caller. */
2420 =for apidoc sv_2iv_flags
2422 Return the integer value of an SV, doing any necessary string
2423 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2424 Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2430 Perl_sv_2iv_flags(pTHX_ SV *const sv, const I32 flags)
2432 PERL_ARGS_ASSERT_SV_2IV_FLAGS;
2434 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2435 && SvTYPE(sv) != SVt_PVFM);
2437 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2443 if (flags & SV_SKIP_OVERLOAD)
2445 tmpstr = AMG_CALLunary(sv, numer_amg);
2446 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2447 return SvIV(tmpstr);
2450 return PTR2IV(SvRV(sv));
2453 if (SvVALID(sv) || isREGEXP(sv)) {
2454 /* FBMs use the space for SvIVX and SvNVX for other purposes, so
2455 must not let them cache IVs.
2456 In practice they are extremely unlikely to actually get anywhere
2457 accessible by user Perl code - the only way that I'm aware of is when
2458 a constant subroutine which is used as the second argument to index.
2460 Regexps have no SvIVX and SvNVX fields.
2465 const char * const ptr =
2466 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2468 = grok_number(ptr, SvCUR(sv), &value);
2470 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2471 == IS_NUMBER_IN_UV) {
2472 /* It's definitely an integer */
2473 if (numtype & IS_NUMBER_NEG) {
2474 if (value < (UV)IV_MIN)
2477 if (value < (UV)IV_MAX)
2482 /* Quite wrong but no good choices. */
2483 if ((numtype & IS_NUMBER_INFINITY)) {
2484 return (numtype & IS_NUMBER_NEG) ? IV_MIN : IV_MAX;
2485 } else if ((numtype & IS_NUMBER_NAN)) {
2486 return 0; /* So wrong. */
2490 if (ckWARN(WARN_NUMERIC))
2493 return I_V(Atof(ptr));
2497 if (SvTHINKFIRST(sv)) {
2498 if (SvREADONLY(sv) && !SvOK(sv)) {
2499 if (ckWARN(WARN_UNINITIALIZED))
2506 if (S_sv_2iuv_common(aTHX_ sv))
2510 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2iv(%" IVdf ")\n",
2511 PTR2UV(sv),SvIVX(sv)));
2512 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
2516 =for apidoc sv_2uv_flags
2518 Return the unsigned integer value of an SV, doing any necessary string
2519 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2520 Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
2526 Perl_sv_2uv_flags(pTHX_ SV *const sv, const I32 flags)
2528 PERL_ARGS_ASSERT_SV_2UV_FLAGS;
2530 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2536 if (flags & SV_SKIP_OVERLOAD)
2538 tmpstr = AMG_CALLunary(sv, numer_amg);
2539 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2540 return SvUV(tmpstr);
2543 return PTR2UV(SvRV(sv));
2546 if (SvVALID(sv) || isREGEXP(sv)) {
2547 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2548 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2549 Regexps have no SvIVX and SvNVX fields. */
2553 const char * const ptr =
2554 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2556 = grok_number(ptr, SvCUR(sv), &value);
2558 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2559 == IS_NUMBER_IN_UV) {
2560 /* It's definitely an integer */
2561 if (!(numtype & IS_NUMBER_NEG))
2565 /* Quite wrong but no good choices. */
2566 if ((numtype & IS_NUMBER_INFINITY)) {
2567 return UV_MAX; /* So wrong. */
2568 } else if ((numtype & IS_NUMBER_NAN)) {
2569 return 0; /* So wrong. */
2573 if (ckWARN(WARN_NUMERIC))
2576 return U_V(Atof(ptr));
2580 if (SvTHINKFIRST(sv)) {
2581 if (SvREADONLY(sv) && !SvOK(sv)) {
2582 if (ckWARN(WARN_UNINITIALIZED))
2589 if (S_sv_2iuv_common(aTHX_ sv))
2593 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2uv(%" UVuf ")\n",
2594 PTR2UV(sv),SvUVX(sv)));
2595 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
2599 =for apidoc sv_2nv_flags
2601 Return the num value of an SV, doing any necessary string or integer
2602 conversion. If C<flags> has the C<SV_GMAGIC> bit set, does an C<mg_get()> first.
2603 Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
2609 Perl_sv_2nv_flags(pTHX_ SV *const sv, const I32 flags)
2611 PERL_ARGS_ASSERT_SV_2NV_FLAGS;
2613 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2614 && SvTYPE(sv) != SVt_PVFM);
2615 if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) {
2616 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2617 the same flag bit as SVf_IVisUV, so must not let them cache NVs.
2618 Regexps have no SvIVX and SvNVX fields. */
2620 if (flags & SV_GMAGIC)
2624 if (SvPOKp(sv) && !SvIOKp(sv)) {
2625 ptr = SvPVX_const(sv);
2626 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
2627 !grok_number(ptr, SvCUR(sv), NULL))
2633 return (NV)SvUVX(sv);
2635 return (NV)SvIVX(sv);
2640 assert(SvTYPE(sv) >= SVt_PVMG);
2641 /* This falls through to the report_uninit near the end of the
2643 } else if (SvTHINKFIRST(sv)) {
2648 if (flags & SV_SKIP_OVERLOAD)
2650 tmpstr = AMG_CALLunary(sv, numer_amg);
2651 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2652 return SvNV(tmpstr);
2655 return PTR2NV(SvRV(sv));
2657 if (SvREADONLY(sv) && !SvOK(sv)) {
2658 if (ckWARN(WARN_UNINITIALIZED))
2663 if (SvTYPE(sv) < SVt_NV) {
2664 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2665 sv_upgrade(sv, SVt_NV);
2666 CLANG_DIAG_IGNORE_STMT(-Wthread-safety);
2668 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2669 STORE_LC_NUMERIC_SET_STANDARD();
2670 PerlIO_printf(Perl_debug_log,
2671 "0x%" UVxf " num(%" NVgf ")\n",
2672 PTR2UV(sv), SvNVX(sv));
2673 RESTORE_LC_NUMERIC();
2675 CLANG_DIAG_RESTORE_STMT;
2678 else if (SvTYPE(sv) < SVt_PVNV)
2679 sv_upgrade(sv, SVt_PVNV);
2684 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
2685 #ifdef NV_PRESERVES_UV
2691 /* Only set the public NV OK flag if this NV preserves the IV */
2692 /* Check it's not 0xFFFFFFFFFFFFFFFF */
2694 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
2695 : (SvIVX(sv) == I_V(SvNVX(sv))))
2701 else if (SvPOKp(sv)) {
2703 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2704 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
2706 #ifdef NV_PRESERVES_UV
2707 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2708 == IS_NUMBER_IN_UV) {
2709 /* It's definitely an integer */
2710 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
2712 S_sv_setnv(aTHX_ sv, numtype);
2719 SvNV_set(sv, Atof(SvPVX_const(sv)));
2720 /* Only set the public NV OK flag if this NV preserves the value in
2721 the PV at least as well as an IV/UV would.
2722 Not sure how to do this 100% reliably. */
2723 /* if that shift count is out of range then Configure's test is
2724 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2726 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2727 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2728 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
2729 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2730 /* Can't use strtol etc to convert this string, so don't try.
2731 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2734 /* value has been set. It may not be precise. */
2735 if ((numtype & IS_NUMBER_NEG) && (value >= (UV)IV_MIN)) {
2736 /* 2s complement assumption for (UV)IV_MIN */
2737 SvNOK_on(sv); /* Integer is too negative. */
2742 if (numtype & IS_NUMBER_NEG) {
2743 /* -IV_MIN is undefined, but we should never reach
2744 * this point with both IS_NUMBER_NEG and value ==
2746 assert(value != (UV)IV_MIN);
2747 SvIV_set(sv, -(IV)value);
2748 } else if (value <= (UV)IV_MAX) {
2749 SvIV_set(sv, (IV)value);
2751 SvUV_set(sv, value);
2755 if (numtype & IS_NUMBER_NOT_INT) {
2756 /* I believe that even if the original PV had decimals,
2757 they are lost beyond the limit of the FP precision.
2758 However, neither is canonical, so both only get p
2759 flags. NWC, 2000/11/25 */
2760 /* Both already have p flags, so do nothing */
2762 const NV nv = SvNVX(sv);
2763 /* XXX should this spot have NAN_COMPARE_BROKEN, too? */
2764 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2765 if (SvIVX(sv) == I_V(nv)) {
2768 /* It had no "." so it must be integer. */
2772 /* between IV_MAX and NV(UV_MAX).
2773 Could be slightly > UV_MAX */
2775 if (numtype & IS_NUMBER_NOT_INT) {
2776 /* UV and NV both imprecise. */
2778 const UV nv_as_uv = U_V(nv);
2780 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2789 /* It might be more code efficient to go through the entire logic above
2790 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2791 gets complex and potentially buggy, so more programmer efficient
2792 to do it this way, by turning off the public flags: */
2794 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2795 #endif /* NV_PRESERVES_UV */
2798 if (isGV_with_GP(sv)) {
2799 glob_2number(MUTABLE_GV(sv));
2803 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2805 assert (SvTYPE(sv) >= SVt_NV);
2806 /* Typically the caller expects that sv_any is not NULL now. */
2807 /* XXX Ilya implies that this is a bug in callers that assume this
2808 and ideally should be fixed. */
2811 CLANG_DIAG_IGNORE_STMT(-Wthread-safety);
2813 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2814 STORE_LC_NUMERIC_SET_STANDARD();
2815 PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2nv(%" NVgf ")\n",
2816 PTR2UV(sv), SvNVX(sv));
2817 RESTORE_LC_NUMERIC();
2819 CLANG_DIAG_RESTORE_STMT;
2826 Return an SV with the numeric value of the source SV, doing any necessary
2827 reference or overload conversion. The caller is expected to have handled
2834 Perl_sv_2num(pTHX_ SV *const sv)
2836 PERL_ARGS_ASSERT_SV_2NUM;
2841 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
2842 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
2843 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2844 return sv_2num(tmpsv);
2846 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2849 /* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2850 * UV as a string towards the end of buf, and return pointers to start and
2853 * We assume that buf is at least TYPE_CHARS(UV) long.
2857 S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
2859 char *ptr = buf + TYPE_CHARS(UV);
2860 char * const ebuf = ptr;
2863 PERL_ARGS_ASSERT_UIV_2BUF;
2871 uv = (iv == IV_MIN) ? (UV)iv : (UV)(-iv);
2875 *--ptr = '0' + (char)(uv % 10);
2883 /* Helper for sv_2pv_flags and sv_vcatpvfn_flags. If the NV is an
2884 * infinity or a not-a-number, writes the appropriate strings to the
2885 * buffer, including a zero byte. On success returns the written length,
2886 * excluding the zero byte, on failure (not an infinity, not a nan)
2887 * returns zero, assert-fails on maxlen being too short.
2889 * XXX for "Inf", "-Inf", and "NaN", we could have three read-only
2890 * shared string constants we point to, instead of generating a new
2891 * string for each instance. */
2893 S_infnan_2pv(NV nv, char* buffer, size_t maxlen, char plus) {
2895 assert(maxlen >= 4);
2896 if (Perl_isinf(nv)) {
2898 if (maxlen < 5) /* "-Inf\0" */
2908 else if (Perl_isnan(nv)) {
2912 /* XXX optionally output the payload mantissa bits as
2913 * "(unsigned)" (to match the nan("...") C99 function,
2914 * or maybe as "(0xhhh...)" would make more sense...
2915 * provide a format string so that the user can decide?
2916 * NOTE: would affect the maxlen and assert() logic.*/
2921 assert((s == buffer + 3) || (s == buffer + 4));
2927 =for apidoc sv_2pv_flags
2929 Returns a pointer to the string value of an SV, and sets C<*lp> to its length.
2930 If flags has the C<SV_GMAGIC> bit set, does an C<mg_get()> first. Coerces C<sv> to a
2931 string if necessary. Normally invoked via the C<SvPV_flags> macro.
2932 C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
2938 Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
2942 PERL_ARGS_ASSERT_SV_2PV_FLAGS;
2944 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2945 && SvTYPE(sv) != SVt_PVFM);
2946 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2951 if (flags & SV_SKIP_OVERLOAD)
2953 tmpstr = AMG_CALLunary(sv, string_amg);
2954 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
2955 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2957 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2961 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2962 if (flags & SV_CONST_RETURN) {
2963 pv = (char *) SvPVX_const(tmpstr);
2965 pv = (flags & SV_MUTABLE_RETURN)
2966 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
2969 *lp = SvCUR(tmpstr);
2971 pv = sv_2pv_flags(tmpstr, lp, flags);
2984 SV *const referent = SvRV(sv);
2988 retval = buffer = savepvn("NULLREF", len);
2989 } else if (SvTYPE(referent) == SVt_REGEXP &&
2990 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
2991 amagic_is_enabled(string_amg))) {
2992 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
2996 /* If the regex is UTF-8 we want the containing scalar to
2997 have an UTF-8 flag too */
3004 *lp = RX_WRAPLEN(re);
3006 return RX_WRAPPED(re);
3008 const char *const typestr = sv_reftype(referent, 0);
3009 const STRLEN typelen = strlen(typestr);
3010 UV addr = PTR2UV(referent);
3011 const char *stashname = NULL;
3012 STRLEN stashnamelen = 0; /* hush, gcc */
3013 const char *buffer_end;
3015 if (SvOBJECT(referent)) {
3016 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
3019 stashname = HEK_KEY(name);
3020 stashnamelen = HEK_LEN(name);
3022 if (HEK_UTF8(name)) {
3028 stashname = "__ANON__";
3031 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
3032 + 2 * sizeof(UV) + 2 /* )\0 */;
3034 len = typelen + 3 /* (0x */
3035 + 2 * sizeof(UV) + 2 /* )\0 */;
3038 Newx(buffer, len, char);
3039 buffer_end = retval = buffer + len;
3041 /* Working backwards */
3045 *--retval = PL_hexdigit[addr & 15];
3046 } while (addr >>= 4);
3052 memcpy(retval, typestr, typelen);
3056 retval -= stashnamelen;
3057 memcpy(retval, stashname, stashnamelen);
3059 /* retval may not necessarily have reached the start of the
3061 assert (retval >= buffer);
3063 len = buffer_end - retval - 1; /* -1 for that \0 */
3075 if (flags & SV_MUTABLE_RETURN)
3076 return SvPVX_mutable(sv);
3077 if (flags & SV_CONST_RETURN)
3078 return (char *)SvPVX_const(sv);
3083 /* I'm assuming that if both IV and NV are equally valid then
3084 converting the IV is going to be more efficient */
3085 const U32 isUIOK = SvIsUV(sv);
3086 char buf[TYPE_CHARS(UV)];
3090 if (SvTYPE(sv) < SVt_PVIV)
3091 sv_upgrade(sv, SVt_PVIV);
3092 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
3094 /* inlined from sv_setpvn */
3095 s = SvGROW_mutable(sv, len + 1);
3096 Move(ptr, s, len, char);
3101 else if (SvNOK(sv)) {
3102 if (SvTYPE(sv) < SVt_PVNV)
3103 sv_upgrade(sv, SVt_PVNV);
3104 if (SvNVX(sv) == 0.0
3105 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
3106 && !Perl_isnan(SvNVX(sv))
3109 s = SvGROW_mutable(sv, 2);
3114 STRLEN size = 5; /* "-Inf\0" */
3116 s = SvGROW_mutable(sv, size);
3117 len = S_infnan_2pv(SvNVX(sv), s, size, 0);
3123 /* some Xenix systems wipe out errno here */
3132 5 + /* exponent digits */
3136 s = SvGROW_mutable(sv, size);
3137 #ifndef USE_LOCALE_NUMERIC
3138 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3144 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
3145 STORE_LC_NUMERIC_SET_TO_NEEDED();
3147 local_radix = _NOT_IN_NUMERIC_STANDARD;
3148 if (local_radix && SvCUR(PL_numeric_radix_sv) > 1) {
3149 size += SvCUR(PL_numeric_radix_sv) - 1;
3150 s = SvGROW_mutable(sv, size);
3153 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3155 /* If the radix character is UTF-8, and actually is in the
3156 * output, turn on the UTF-8 flag for the scalar */
3158 && SvUTF8(PL_numeric_radix_sv)
3159 && instr(s, SvPVX_const(PL_numeric_radix_sv)))
3164 RESTORE_LC_NUMERIC();
3167 /* We don't call SvPOK_on(), because it may come to
3168 * pass that the locale changes so that the
3169 * stringification we just did is no longer correct. We
3170 * will have to re-stringify every time it is needed */
3177 else if (isGV_with_GP(sv)) {
3178 GV *const gv = MUTABLE_GV(sv);
3179 SV *const buffer = sv_newmortal();
3181 gv_efullname3(buffer, gv, "*");
3183 assert(SvPOK(buffer));
3189 *lp = SvCUR(buffer);
3190 return SvPVX(buffer);
3195 if (flags & SV_UNDEF_RETURNS_NULL)
3197 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
3199 /* Typically the caller expects that sv_any is not NULL now. */
3200 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
3201 sv_upgrade(sv, SVt_PV);
3206 const STRLEN len = s - SvPVX_const(sv);
3211 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2pv(%s)\n",
3212 PTR2UV(sv),SvPVX_const(sv)));
3213 if (flags & SV_CONST_RETURN)
3214 return (char *)SvPVX_const(sv);
3215 if (flags & SV_MUTABLE_RETURN)
3216 return SvPVX_mutable(sv);
3221 =for apidoc sv_copypv
3223 Copies a stringified representation of the source SV into the
3224 destination SV. Automatically performs any necessary C<mg_get> and
3225 coercion of numeric values into strings. Guaranteed to preserve
3226 C<UTF8> flag even from overloaded objects. Similar in nature to
3227 C<sv_2pv[_flags]> but operates directly on an SV instead of just the
3228 string. Mostly uses C<sv_2pv_flags> to do its work, except when that
3229 would lose the UTF-8'ness of the PV.
3231 =for apidoc sv_copypv_nomg
3233 Like C<sv_copypv>, but doesn't invoke get magic first.
3235 =for apidoc sv_copypv_flags
3237 Implementation of C<sv_copypv> and C<sv_copypv_nomg>. Calls get magic iff flags
3238 has the C<SV_GMAGIC> bit set.
3244 Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
3249 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
3251 s = SvPV_flags_const(ssv,len,(flags & SV_GMAGIC));
3252 sv_setpvn(dsv,s,len);
3260 =for apidoc sv_2pvbyte
3262 Return a pointer to the byte-encoded representation of the SV, and set C<*lp>
3263 to its length. May cause the SV to be downgraded from UTF-8 as a
3266 Usually accessed via the C<SvPVbyte> macro.
3272 Perl_sv_2pvbyte(pTHX_ SV *sv, STRLEN *const lp)
3274 PERL_ARGS_ASSERT_SV_2PVBYTE;
3277 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3278 || isGV_with_GP(sv) || SvROK(sv)) {
3279 SV *sv2 = sv_newmortal();
3280 sv_copypv_nomg(sv2,sv);
3283 sv_utf8_downgrade(sv,0);
3284 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3288 =for apidoc sv_2pvutf8
3290 Return a pointer to the UTF-8-encoded representation of the SV, and set C<*lp>
3291 to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3293 Usually accessed via the C<SvPVutf8> macro.
3299 Perl_sv_2pvutf8(pTHX_ SV *sv, STRLEN *const lp)
3301 PERL_ARGS_ASSERT_SV_2PVUTF8;
3303 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3304 || isGV_with_GP(sv) || SvROK(sv))
3305 sv = sv_mortalcopy(sv);
3308 sv_utf8_upgrade_nomg(sv);
3309 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3314 =for apidoc sv_2bool
3316 This macro is only used by C<sv_true()> or its macro equivalent, and only if
3317 the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>.
3318 It calls C<sv_2bool_flags> with the C<SV_GMAGIC> flag.
3320 =for apidoc sv_2bool_flags
3322 This function is only used by C<sv_true()> and friends, and only if
3323 the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>. If the flags
3324 contain C<SV_GMAGIC>, then it does an C<mg_get()> first.
3331 Perl_sv_2bool_flags(pTHX_ SV *sv, I32 flags)
3333 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
3336 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
3342 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
3343 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv)))) {
3346 if(SvGMAGICAL(sv)) {
3348 goto restart; /* call sv_2bool */
3350 /* expanded SvTRUE_common(sv, (flags = 0, goto restart)) */
3351 else if(!SvOK(sv)) {
3354 else if(SvPOK(sv)) {
3355 svb = SvPVXtrue(sv);
3357 else if((SvFLAGS(sv) & (SVf_IOK|SVf_NOK))) {
3358 svb = (SvIOK(sv) && SvIVX(sv) != 0)
3359 || (SvNOK(sv) && SvNVX(sv) != 0.0);
3363 goto restart; /* call sv_2bool_nomg */
3373 RX_WRAPLEN(sv) > 1 || (RX_WRAPLEN(sv) && *RX_WRAPPED(sv) != '0');
3375 if (SvNOK(sv) && !SvPOK(sv))
3376 return SvNVX(sv) != 0.0;
3378 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
3382 =for apidoc sv_utf8_upgrade
3384 Converts the PV of an SV to its UTF-8-encoded form.
3385 Forces the SV to string form if it is not already.
3386 Will C<mg_get> on C<sv> if appropriate.
3387 Always sets the C<SvUTF8> flag to avoid future validity checks even
3388 if the whole string is the same in UTF-8 as not.
3389 Returns the number of bytes in the converted string
3391 This is not a general purpose byte encoding to Unicode interface:
3392 use the Encode extension for that.
3394 =for apidoc sv_utf8_upgrade_nomg
3396 Like C<sv_utf8_upgrade>, but doesn't do magic on C<sv>.
3398 =for apidoc sv_utf8_upgrade_flags
3400 Converts the PV of an SV to its UTF-8-encoded form.
3401 Forces the SV to string form if it is not already.
3402 Always sets the SvUTF8 flag to avoid future validity checks even
3403 if all the bytes are invariant in UTF-8.
3404 If C<flags> has C<SV_GMAGIC> bit set,
3405 will C<mg_get> on C<sv> if appropriate, else not.
3407 The C<SV_FORCE_UTF8_UPGRADE> flag is now ignored.
3409 Returns the number of bytes in the converted string.
3411 This is not a general purpose byte encoding to Unicode interface:
3412 use the Encode extension for that.
3414 =for apidoc sv_utf8_upgrade_flags_grow
3416 Like C<sv_utf8_upgrade_flags>, but has an additional parameter C<extra>, which is
3417 the number of unused bytes the string of C<sv> is guaranteed to have free after
3418 it upon return. This allows the caller to reserve extra space that it intends
3419 to fill, to avoid extra grows.
3421 C<sv_utf8_upgrade>, C<sv_utf8_upgrade_nomg>, and C<sv_utf8_upgrade_flags>
3422 are implemented in terms of this function.
3424 Returns the number of bytes in the converted string (not including the spares).
3428 If the routine itself changes the string, it adds a trailing C<NUL>. Such a
3429 C<NUL> isn't guaranteed due to having other routines do the work in some input
3430 cases, or if the input is already flagged as being in utf8.
3435 Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV *const sv, const I32 flags, STRLEN extra)
3437 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
3439 if (sv == &PL_sv_undef)
3441 if (!SvPOK_nog(sv)) {
3443 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3444 (void) sv_2pv_flags(sv,&len, flags);
3446 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3450 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
3454 /* SVt_REGEXP's shouldn't be upgraded to UTF8 - they're already
3455 * compiled and individual nodes will remain non-utf8 even if the
3456 * stringified version of the pattern gets upgraded. Whether the
3457 * PVX of a REGEXP should be grown or we should just croak, I don't
3459 if (SvUTF8(sv) || isREGEXP(sv)) {
3460 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3465 S_sv_uncow(aTHX_ sv, 0);
3468 if (SvCUR(sv) == 0) {
3469 if (extra) SvGROW(sv, extra);
3470 } else { /* Assume Latin-1/EBCDIC */
3471 /* This function could be much more efficient if we
3472 * had a FLAG in SVs to signal if there are any variant
3473 * chars in the PV. Given that there isn't such a flag
3474 * make the loop as fast as possible. */
3475 U8 * s = (U8 *) SvPVX_const(sv);
3478 if (is_utf8_invariant_string_loc(s, SvCUR(sv), (const U8 **) &t)) {
3480 /* utf8 conversion not needed because all are invariants. Mark
3481 * as UTF-8 even if no variant - saves scanning loop */
3483 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3487 /* Here, there is at least one variant (t points to the first one), so
3488 * the string should be converted to utf8. Everything from 's' to
3489 * 't - 1' will occupy only 1 byte each on output.
3491 * Note that the incoming SV may not have a trailing '\0', as certain
3492 * code in pp_formline can send us partially built SVs.
3494 * There are two main ways to convert. One is to create a new string
3495 * and go through the input starting from the beginning, appending each
3496 * converted value onto the new string as we go along. Going this
3497 * route, it's probably best to initially allocate enough space in the
3498 * string rather than possibly running out of space and having to
3499 * reallocate and then copy what we've done so far. Since everything
3500 * from 's' to 't - 1' is invariant, the destination can be initialized
3501 * with these using a fast memory copy. To be sure to allocate enough
3502 * space, one could use the worst case scenario, where every remaining
3503 * byte expands to two under UTF-8, or one could parse it and count
3504 * exactly how many do expand.
3506 * The other way is to unconditionally parse the remainder of the
3507 * string to figure out exactly how big the expanded string will be,
3508 * growing if needed. Then start at the end of the string and place
3509 * the character there at the end of the unfilled space in the expanded
3510 * one, working backwards until reaching 't'.
3512 * The problem with assuming the worst case scenario is that for very
3513 * long strings, we could allocate much more memory than actually
3514 * needed, which can create performance problems. If we have to parse
3515 * anyway, the second method is the winner as it may avoid an extra
3516 * copy. The code used to use the first method under some
3517 * circumstances, but now that there is faster variant counting on
3518 * ASCII platforms, the second method is used exclusively, eliminating
3519 * some code that no longer has to be maintained. */
3522 /* Count the total number of variants there are. We can start
3523 * just beyond the first one, which is known to be at 't' */
3524 const Size_t invariant_length = t - s;
3525 U8 * e = (U8 *) SvEND(sv);
3527 /* The length of the left overs, plus 1. */
3528 const Size_t remaining_length_p1 = e - t;
3530 /* We expand by 1 for the variant at 't' and one for each remaining
3531 * variant (we start looking at 't+1') */
3532 Size_t expansion = 1 + variant_under_utf8_count(t + 1, e);
3534 /* +1 = trailing NUL */
3535 Size_t need = SvCUR(sv) + expansion + extra + 1;
3538 /* Grow if needed */
3539 if (SvLEN(sv) < need) {
3540 t = invariant_length + (U8*) SvGROW(sv, need);
3541 e = t + remaining_length_p1;
3543 SvCUR_set(sv, invariant_length + remaining_length_p1 + expansion);
3545 /* Set the NUL at the end */
3546 d = (U8 *) SvEND(sv);
3549 /* Having decremented d, it points to the position to put the
3550 * very last byte of the expanded string. Go backwards through
3551 * the string, copying and expanding as we go, stopping when we
3552 * get to the part that is invariant the rest of the way down */
3556 if (NATIVE_BYTE_IS_INVARIANT(*e)) {
3559 *d-- = UTF8_EIGHT_BIT_LO(*e);
3560 *d-- = UTF8_EIGHT_BIT_HI(*e);
3565 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3566 /* Update pos. We do it at the end rather than during
3567 * the upgrade, to avoid slowing down the common case
3568 * (upgrade without pos).
3569 * pos can be stored as either bytes or characters. Since
3570 * this was previously a byte string we can just turn off
3571 * the bytes flag. */
3572 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3574 mg->mg_flags &= ~MGf_BYTES;
3576 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3577 magic_setutf8(sv,mg); /* clear UTF8 cache */
3587 =for apidoc sv_utf8_downgrade
3589 Attempts to convert the PV of an SV from characters to bytes.
3590 If the PV contains a character that cannot fit
3591 in a byte, this conversion will fail;
3592 in this case, either returns false or, if C<fail_ok> is not
3595 This is not a general purpose Unicode to byte encoding interface:
3596 use the C<Encode> extension for that.
3602 Perl_sv_utf8_downgrade(pTHX_ SV *const sv, const bool fail_ok)
3604 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3606 if (SvPOKp(sv) && SvUTF8(sv)) {
3610 int mg_flags = SV_GMAGIC;
3613 S_sv_uncow(aTHX_ sv, 0);
3615 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3617 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3618 if (mg && mg->mg_len > 0 && mg->mg_flags & MGf_BYTES) {
3619 mg->mg_len = sv_pos_b2u_flags(sv, mg->mg_len,
3620 SV_GMAGIC|SV_CONST_RETURN);
3621 mg_flags = 0; /* sv_pos_b2u does get magic */
3623 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3624 magic_setutf8(sv,mg); /* clear UTF8 cache */
3627 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3629 if (!utf8_to_bytes(s, &len)) {
3634 Perl_croak(aTHX_ "Wide character in %s",
3637 Perl_croak(aTHX_ "Wide character");
3648 =for apidoc sv_utf8_encode
3650 Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3651 flag off so that it looks like octets again.
3657 Perl_sv_utf8_encode(pTHX_ SV *const sv)
3659 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3661 if (SvREADONLY(sv)) {
3662 sv_force_normal_flags(sv, 0);
3664 (void) sv_utf8_upgrade(sv);
3669 =for apidoc sv_utf8_decode
3671 If the PV of the SV is an octet sequence in Perl's extended UTF-8
3672 and contains a multiple-byte character, the C<SvUTF8> flag is turned on
3673 so that it looks like a character. If the PV contains only single-byte
3674 characters, the C<SvUTF8> flag stays off.
3675 Scans PV for validity and returns FALSE if the PV is invalid UTF-8.
3681 Perl_sv_utf8_decode(pTHX_ SV *const sv)
3683 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3686 const U8 *start, *c, *first_variant;
3688 /* The octets may have got themselves encoded - get them back as
3691 if (!sv_utf8_downgrade(sv, TRUE))
3694 /* it is actually just a matter of turning the utf8 flag on, but
3695 * we want to make sure everything inside is valid utf8 first.
3697 c = start = (const U8 *) SvPVX_const(sv);
3698 if (! is_utf8_invariant_string_loc(c, SvCUR(sv), &first_variant)) {
3699 if (!is_utf8_string(first_variant, SvCUR(sv) - (first_variant -c)))
3703 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3704 /* XXX Is this dead code? XS_utf8_decode calls SvSETMAGIC
3705 after this, clearing pos. Does anything on CPAN
3707 /* adjust pos to the start of a UTF8 char sequence */
3708 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3710 I32 pos = mg->mg_len;
3712 for (c = start + pos; c > start; c--) {
3713 if (UTF8_IS_START(*c))
3716 mg->mg_len = c - start;
3719 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3720 magic_setutf8(sv,mg); /* clear UTF8 cache */
3727 =for apidoc sv_setsv
3729 Copies the contents of the source SV C<ssv> into the destination SV
3730 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3731 function if the source SV needs to be reused. Does not handle 'set' magic on
3732 destination SV. Calls 'get' magic on source SV. Loosely speaking, it
3733 performs a copy-by-value, obliterating any previous content of the
3736 You probably want to use one of the assortment of wrappers, such as
3737 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3738 C<SvSetMagicSV_nosteal>.
3740 =for apidoc sv_setsv_flags
3742 Copies the contents of the source SV C<ssv> into the destination SV
3743 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3744 function if the source SV needs to be reused. Does not handle 'set' magic.
3745 Loosely speaking, it performs a copy-by-value, obliterating any previous
3746 content of the destination.
3747 If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
3748 C<ssv> if appropriate, else not. If the C<flags>
3749 parameter has the C<SV_NOSTEAL> bit set then the
3750 buffers of temps will not be stolen. C<sv_setsv>
3751 and C<sv_setsv_nomg> are implemented in terms of this function.
3753 You probably want to use one of the assortment of wrappers, such as
3754 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3755 C<SvSetMagicSV_nosteal>.
3757 This is the primary function for copying scalars, and most other
3758 copy-ish functions and macros use this underneath.
3764 S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
3766 I32 mro_changes = 0; /* 1 = method, 2 = isa, 3 = recursive isa */
3767 HV *old_stash = NULL;
3769 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB;
3771 if (dtype != SVt_PVGV && !isGV_with_GP(dstr)) {
3772 const char * const name = GvNAME(sstr);
3773 const STRLEN len = GvNAMELEN(sstr);
3775 if (dtype >= SVt_PV) {
3781 SvUPGRADE(dstr, SVt_PVGV);
3782 (void)SvOK_off(dstr);
3783 isGV_with_GP_on(dstr);
3785 GvSTASH(dstr) = GvSTASH(sstr);
3787 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
3788 gv_name_set(MUTABLE_GV(dstr), name, len,
3789 GV_ADD | (GvNAMEUTF8(sstr) ? SVf_UTF8 : 0 ));
3790 SvFAKE_on(dstr); /* can coerce to non-glob */
3793 if(GvGP(MUTABLE_GV(sstr))) {
3794 /* If source has method cache entry, clear it */
3796 SvREFCNT_dec(GvCV(sstr));
3797 GvCV_set(sstr, NULL);
3800 /* If source has a real method, then a method is
3803 GvCV((const GV *)sstr) && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3809 /* If dest already had a real method, that's a change as well */
3811 !mro_changes && GvGP(MUTABLE_GV(dstr)) && GvCVu((const GV *)dstr)
3812 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3817 /* We don't need to check the name of the destination if it was not a
3818 glob to begin with. */
3819 if(dtype == SVt_PVGV) {
3820 const char * const name = GvNAME((const GV *)dstr);
3821 const STRLEN len = GvNAMELEN(dstr);
3822 if(memEQs(name, len, "ISA")
3823 /* The stash may have been detached from the symbol table, so
3825 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3829 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
3830 || (len == 1 && name[0] == ':')) {
3833 /* Set aside the old stash, so we can reset isa caches on
3835 if((old_stash = GvHV(dstr)))
3836 /* Make sure we do not lose it early. */
3837 SvREFCNT_inc_simple_void_NN(
3838 sv_2mortal((SV *)old_stash)
3843 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
3846 /* freeing dstr's GP might free sstr (e.g. *x = $x),
3847 * so temporarily protect it */
3849 SAVEFREESV(SvREFCNT_inc_simple_NN(sstr));
3850 gp_free(MUTABLE_GV(dstr));
3851 GvINTRO_off(dstr); /* one-shot flag */
3852 GvGP_set(dstr, gp_ref(GvGP(sstr)));
3855 if (SvTAINTED(sstr))
3857 if (GvIMPORTED(dstr) != GVf_IMPORTED
3858 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
3860 GvIMPORTED_on(dstr);
3863 if(mro_changes == 2) {
3864 if (GvAV((const GV *)sstr)) {
3866 SV * const sref = (SV *)GvAV((const GV *)dstr);
3867 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
3868 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
3869 AV * const ary = newAV();
3870 av_push(ary, mg->mg_obj); /* takes the refcount */
3871 mg->mg_obj = (SV *)ary;
3873 av_push((AV *)mg->mg_obj, SvREFCNT_inc_simple_NN(dstr));
3875 else sv_magic(sref, dstr, PERL_MAGIC_isa, NULL, 0);
3877 mro_isa_changed_in(GvSTASH(dstr));
3879 else if(mro_changes == 3) {
3880 HV * const stash = GvHV(dstr);
3881 if(old_stash ? (HV *)HvENAME_get(old_stash) : stash)
3887 else if(mro_changes) mro_method_changed_in(GvSTASH(dstr));
3888 if (GvIO(dstr) && dtype == SVt_PVGV) {
3889 DEBUG_o(Perl_deb(aTHX_
3890 "glob_assign_glob clearing PL_stashcache\n"));
3891 /* It's a cache. It will rebuild itself quite happily.
3892 It's a lot of effort to work out exactly which key (or keys)
3893 might be invalidated by the creation of the this file handle.
3895 hv_clear(PL_stashcache);
3901 Perl_gv_setref(pTHX_ SV *const dstr, SV *const sstr)
3903 SV * const sref = SvRV(sstr);
3905 const int intro = GvINTRO(dstr);
3908 const U32 stype = SvTYPE(sref);
3910 PERL_ARGS_ASSERT_GV_SETREF;
3913 GvINTRO_off(dstr); /* one-shot flag */
3914 GvLINE(dstr) = CopLINE(PL_curcop);
3915 GvEGV(dstr) = MUTABLE_GV(dstr);
3920 location = (SV **) &(GvGP(dstr)->gp_cv); /* XXX bypassing GvCV_set */
3921 import_flag = GVf_IMPORTED_CV;
3924 location = (SV **) &GvHV(dstr);
3925 import_flag = GVf_IMPORTED_HV;
3928 location = (SV **) &GvAV(dstr);
3929 import_flag = GVf_IMPORTED_AV;
3932 location = (SV **) &GvIOp(dstr);
3935 location = (SV **) &GvFORM(dstr);
3938 location = &GvSV(dstr);
3939 import_flag = GVf_IMPORTED_SV;
3942 if (stype == SVt_PVCV) {
3943 /*if (GvCVGEN(dstr) && (GvCV(dstr) != (const CV *)sref || GvCVGEN(dstr))) {*/
3944 if (GvCVGEN(dstr)) {
3945 SvREFCNT_dec(GvCV(dstr));
3946 GvCV_set(dstr, NULL);
3947 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
3950 /* SAVEt_GVSLOT takes more room on the savestack and has more
3951 overhead in leave_scope than SAVEt_GENERIC_SV. But for CVs
3952 leave_scope needs access to the GV so it can reset method
3953 caches. We must use SAVEt_GVSLOT whenever the type is
3954 SVt_PVCV, even if the stash is anonymous, as the stash may
3955 gain a name somehow before leave_scope. */
3956 if (stype == SVt_PVCV) {
3957 /* There is no save_pushptrptrptr. Creating it for this
3958 one call site would be overkill. So inline the ss add
3962 SS_ADD_PTR(location);
3963 SS_ADD_PTR(SvREFCNT_inc(*location));
3964 SS_ADD_UV(SAVEt_GVSLOT);
3967 else SAVEGENERICSV(*location);
3970 if (stype == SVt_PVCV && (*location != sref || GvCVGEN(dstr))) {
3971 CV* const cv = MUTABLE_CV(*location);
3973 if (!GvCVGEN((const GV *)dstr) &&
3974 (CvROOT(cv) || CvXSUB(cv)) &&
3975 /* redundant check that avoids creating the extra SV
3976 most of the time: */
3977 (CvCONST(cv) || ckWARN(WARN_REDEFINE)))
3979 SV * const new_const_sv =
3980 CvCONST((const CV *)sref)
3981 ? cv_const_sv((const CV *)sref)
3983 HV * const stash = GvSTASH((const GV *)dstr);
3984 report_redefined_cv(
3987 ? Perl_newSVpvf(aTHX_
3988 "%" HEKf "::%" HEKf,
3989 HEKfARG(HvNAME_HEK(stash)),
3990 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr))))
3991 : Perl_newSVpvf(aTHX_
3993 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr))))
3996 CvCONST((const CV *)sref) ? &new_const_sv : NULL
4000 cv_ckproto_len_flags(cv, (const GV *)dstr,
4001 SvPOK(sref) ? CvPROTO(sref) : NULL,
4002 SvPOK(sref) ? CvPROTOLEN(sref) : 0,
4003 SvPOK(sref) ? SvUTF8(sref) : 0);
4005 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4006 GvASSUMECV_on(dstr);
4007 if(GvSTASH(dstr)) { /* sub foo { 1 } sub bar { 2 } *bar = \&foo */
4008 if (intro && GvREFCNT(dstr) > 1) {
4009 /* temporary remove extra savestack's ref */
4011 gv_method_changed(dstr);
4014 else gv_method_changed(dstr);
4017 *location = SvREFCNT_inc_simple_NN(sref);
4018 if (import_flag && !(GvFLAGS(dstr) & import_flag)
4019 && CopSTASH_ne(PL_curcop, GvSTASH(dstr))) {
4020 GvFLAGS(dstr) |= import_flag;
4023 if (stype == SVt_PVHV) {
4024 const char * const name = GvNAME((GV*)dstr);
4025 const STRLEN len = GvNAMELEN(dstr);
4028 (len > 1 && name[len-2] == ':' && name[len-1] == ':')
4029 || (len == 1 && name[0] == ':')
4031 && (!dref || HvENAME_get(dref))
4034 (HV *)sref, (HV *)dref,
4040 stype == SVt_PVAV && sref != dref
4041 && memEQs(GvNAME((GV*)dstr), GvNAMELEN((GV*)dstr), "ISA")
4042 /* The stash may have been detached from the symbol table, so
4043 check its name before doing anything. */
4044 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
4047 MAGIC * const omg = dref && SvSMAGICAL(dref)
4048 ? mg_find(dref, PERL_MAGIC_isa)
4050 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
4051 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
4052 AV * const ary = newAV();
4053 av_push(ary, mg->mg_obj); /* takes the refcount */
4054 mg->mg_obj = (SV *)ary;
4057 if (SvTYPE(omg->mg_obj) == SVt_PVAV) {
4058 SV **svp = AvARRAY((AV *)omg->mg_obj);
4059 I32 items = AvFILLp((AV *)omg->mg_obj) + 1;
4063 SvREFCNT_inc_simple_NN(*svp++)
4069 SvREFCNT_inc_simple_NN(omg->mg_obj)
4073 av_push((AV *)mg->mg_obj,SvREFCNT_inc_simple_NN(dstr));
4079 sref, omg ? omg->mg_obj : dstr, PERL_MAGIC_isa, NULL, 0
4081 for (i = 0; i <= AvFILL(sref); ++i) {
4082 SV **elem = av_fetch ((AV*)sref, i, 0);
4085 *elem, sref, PERL_MAGIC_isaelem, NULL, i
4089 mg = mg_find(sref, PERL_MAGIC_isa);
4091 /* Since the *ISA assignment could have affected more than
4092 one stash, don't call mro_isa_changed_in directly, but let
4093 magic_clearisa do it for us, as it already has the logic for
4094 dealing with globs vs arrays of globs. */
4096 Perl_magic_clearisa(aTHX_ NULL, mg);
4098 else if (stype == SVt_PVIO) {
4099 DEBUG_o(Perl_deb(aTHX_ "gv_setref clearing PL_stashcache\n"));
4100 /* It's a cache. It will rebuild itself quite happily.
4101 It's a lot of effort to work out exactly which key (or keys)
4102 might be invalidated by the creation of the this file handle.
4104 hv_clear(PL_stashcache);
4108 if (!intro) SvREFCNT_dec(dref);
4109 if (SvTAINTED(sstr))
4117 #ifdef PERL_DEBUG_READONLY_COW
4118 # include <sys/mman.h>
4120 # ifndef PERL_MEMORY_DEBUG_HEADER_SIZE
4121 # define PERL_MEMORY_DEBUG_HEADER_SIZE 0
4125 Perl_sv_buf_to_ro(pTHX_ SV *sv)
4127 struct perl_memory_debug_header * const header =
4128 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4129 const MEM_SIZE len = header->size;
4130 PERL_ARGS_ASSERT_SV_BUF_TO_RO;
4131 # ifdef PERL_TRACK_MEMPOOL
4132 if (!header->readonly) header->readonly = 1;
4134 if (mprotect(header, len, PROT_READ))
4135 Perl_warn(aTHX_ "mprotect RW for COW string %p %lu failed with %d",
4136 header, len, errno);
4140 S_sv_buf_to_rw(pTHX_ SV *sv)
4142 struct perl_memory_debug_header * const header =
4143 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4144 const MEM_SIZE len = header->size;
4145 PERL_ARGS_ASSERT_SV_BUF_TO_RW;
4146 if (mprotect(header, len, PROT_READ|PROT_WRITE))
4147 Perl_warn(aTHX_ "mprotect for COW string %p %lu failed with %d",
4148 header, len, errno);
4149 # ifdef PERL_TRACK_MEMPOOL
4150 header->readonly = 0;
4155 # define sv_buf_to_ro(sv) NOOP
4156 # define sv_buf_to_rw(sv) NOOP
4160 Perl_sv_setsv_flags(pTHX_ SV *dstr, SV* sstr, const I32 flags)
4165 unsigned int both_type;
4167 PERL_ARGS_ASSERT_SV_SETSV_FLAGS;
4169 if (UNLIKELY( sstr == dstr ))
4172 if (UNLIKELY( !sstr ))
4173 sstr = &PL_sv_undef;
4175 stype = SvTYPE(sstr);
4176 dtype = SvTYPE(dstr);
4177 both_type = (stype | dtype);
4179 /* with these values, we can check that both SVs are NULL/IV (and not
4180 * freed) just by testing the or'ed types */
4181 STATIC_ASSERT_STMT(SVt_NULL == 0);
4182 STATIC_ASSERT_STMT(SVt_IV == 1);
4183 if (both_type <= 1) {
4184 /* both src and dst are UNDEF/IV/RV, so we can do a lot of
4190 /* minimal subset of SV_CHECK_THINKFIRST_COW_DROP(dstr) */
4191 if (SvREADONLY(dstr))
4192 Perl_croak_no_modify();
4194 if (SvWEAKREF(dstr))
4195 sv_unref_flags(dstr, 0);
4197 old_rv = SvRV(dstr);
4200 assert(!SvGMAGICAL(sstr));
4201 assert(!SvGMAGICAL(dstr));
4203 sflags = SvFLAGS(sstr);
4204 if (sflags & (SVf_IOK|SVf_ROK)) {
4205 SET_SVANY_FOR_BODYLESS_IV(dstr);
4206 new_dflags = SVt_IV;
4208 if (sflags & SVf_ROK) {
4209 dstr->sv_u.svu_rv = SvREFCNT_inc(SvRV(sstr));
4210 new_dflags |= SVf_ROK;
4213 /* both src and dst are <= SVt_IV, so sv_any points to the
4214 * head; so access the head directly
4216 assert( &(sstr->sv_u.svu_iv)
4217 == &(((XPVIV*) SvANY(sstr))->xiv_iv));
4218 assert( &(dstr->sv_u.svu_iv)
4219 == &(((XPVIV*) SvANY(dstr))->xiv_iv));
4220 dstr->sv_u.svu_iv = sstr->sv_u.svu_iv;
4221 new_dflags |= (SVf_IOK|SVp_IOK|(sflags & SVf_IVisUV));
4225 new_dflags = dtype; /* turn off everything except the type */
4227 SvFLAGS(dstr) = new_dflags;
4228 SvREFCNT_dec(old_rv);
4233 if (UNLIKELY(both_type == SVTYPEMASK)) {
4234 if (SvIS_FREED(dstr)) {
4235 Perl_croak(aTHX_ "panic: attempt to copy value %" SVf
4236 " to a freed scalar %p", SVfARG(sstr), (void *)dstr);
4238 if (SvIS_FREED(sstr)) {
4239 Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p to %p",
4240 (void*)sstr, (void*)dstr);
4246 SV_CHECK_THINKFIRST_COW_DROP(dstr);
4247 dtype = SvTYPE(dstr); /* THINKFIRST may have changed type */
4249 /* There's a lot of redundancy below but we're going for speed here */
4254 if (LIKELY( dtype != SVt_PVGV && dtype != SVt_PVLV )) {
4255 (void)SvOK_off(dstr);
4263 /* For performance, we inline promoting to type SVt_IV. */
4264 /* We're starting from SVt_NULL, so provided that define is
4265 * actual 0, we don't have to unset any SV type flags
4266 * to promote to SVt_IV. */
4267 STATIC_ASSERT_STMT(SVt_NULL == 0);
4268 SET_SVANY_FOR_BODYLESS_IV(dstr);
4269 SvFLAGS(dstr) |= SVt_IV;
4273 sv_upgrade(dstr, SVt_PVIV);
4277 goto end_of_first_switch;
4279 (void)SvIOK_only(dstr);
4280 SvIV_set(dstr, SvIVX(sstr));
4283 /* SvTAINTED can only be true if the SV has taint magic, which in
4284 turn means that the SV type is PVMG (or greater). This is the
4285 case statement for SVt_IV, so this cannot be true (whatever gcov
4287 assert(!SvTAINTED(sstr));
4292 if (dtype < SVt_PV && dtype != SVt_IV)
4293 sv_upgrade(dstr, SVt_IV);
4297 if (LIKELY( SvNOK(sstr) )) {
4301 sv_upgrade(dstr, SVt_NV);
4305 sv_upgrade(dstr, SVt_PVNV);
4309 goto end_of_first_switch;
4311 SvNV_set(dstr, SvNVX(sstr));
4312 (void)SvNOK_only(dstr);
4313 /* SvTAINTED can only be true if the SV has taint magic, which in
4314 turn means that the SV type is PVMG (or greater). This is the
4315 case statement for SVt_NV, so this cannot be true (whatever gcov
4317 assert(!SvTAINTED(sstr));
4324 sv_upgrade(dstr, SVt_PV);
4327 if (dtype < SVt_PVIV)
4328 sv_upgrade(dstr, SVt_PVIV);
4331 if (dtype < SVt_PVNV)
4332 sv_upgrade(dstr, SVt_PVNV);
4336 invlist_clone(sstr, dstr);
4340 const char * const type = sv_reftype(sstr,0);
4342 /* diag_listed_as: Bizarre copy of %s */
4343 Perl_croak(aTHX_ "Bizarre copy of %s in %s", type, OP_DESC(PL_op));
4345 Perl_croak(aTHX_ "Bizarre copy of %s", type);
4347 NOT_REACHED; /* NOTREACHED */
4351 if (dtype < SVt_REGEXP)
4352 sv_upgrade(dstr, SVt_REGEXP);
4358 if (SvGMAGICAL(sstr) && (flags & SV_GMAGIC)) {
4360 if (SvTYPE(sstr) != stype)
4361 stype = SvTYPE(sstr);
4363 if (isGV_with_GP(sstr) && dtype <= SVt_PVLV) {
4364 glob_assign_glob(dstr, sstr, dtype);
4367 if (stype == SVt_PVLV)
4369 if (isREGEXP(sstr)) goto upgregexp;
4370 SvUPGRADE(dstr, SVt_PVNV);
4373 SvUPGRADE(dstr, (svtype)stype);
4375 end_of_first_switch:
4377 /* dstr may have been upgraded. */
4378 dtype = SvTYPE(dstr);
4379 sflags = SvFLAGS(sstr);
4381 if (UNLIKELY( dtype == SVt_PVCV )) {
4382 /* Assigning to a subroutine sets the prototype. */
4385 const char *const ptr = SvPV_const(sstr, len);
4387 SvGROW(dstr, len + 1);
4388 Copy(ptr, SvPVX(dstr), len + 1, char);
4389 SvCUR_set(dstr, len);
4391 SvFLAGS(dstr) |= sflags & SVf_UTF8;
4392 CvAUTOLOAD_off(dstr);
4397 else if (UNLIKELY(dtype == SVt_PVAV || dtype == SVt_PVHV
4398 || dtype == SVt_PVFM))
4400 const char * const type = sv_reftype(dstr,0);
4402 /* diag_listed_as: Cannot copy to %s */
4403 Perl_croak(aTHX_ "Cannot copy to %s in %s", type, OP_DESC(PL_op));
4405 Perl_croak(aTHX_ "Cannot copy to %s", type);
4406 } else if (sflags & SVf_ROK) {
4407 if (isGV_with_GP(dstr)
4408 && SvTYPE(SvRV(sstr)) == SVt_PVGV && isGV_with_GP(SvRV(sstr))) {
4411 if (GvIMPORTED(dstr) != GVf_IMPORTED
4412 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
4414 GvIMPORTED_on(dstr);
4419 glob_assign_glob(dstr, sstr, dtype);
4423 if (dtype >= SVt_PV) {
4424 if (isGV_with_GP(dstr)) {
4425 gv_setref(dstr, sstr);
4428 if (SvPVX_const(dstr)) {
4434 (void)SvOK_off(dstr);
4435 SvRV_set(dstr, SvREFCNT_inc(SvRV(sstr)));
4436 SvFLAGS(dstr) |= sflags & SVf_ROK;
4437 assert(!(sflags & SVp_NOK));
4438 assert(!(sflags & SVp_IOK));
4439 assert(!(sflags & SVf_NOK));
4440 assert(!(sflags & SVf_IOK));
4442 else if (isGV_with_GP(dstr)) {
4443 if (!(sflags & SVf_OK)) {
4444 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4445 "Undefined value assigned to typeglob");
4448 GV *gv = gv_fetchsv_nomg(sstr, GV_ADD, SVt_PVGV);
4449 if (dstr != (const SV *)gv) {
4450 const char * const name = GvNAME((const GV *)dstr);
4451 const STRLEN len = GvNAMELEN(dstr);
4452 HV *old_stash = NULL;
4453 bool reset_isa = FALSE;
4454 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
4455 || (len == 1 && name[0] == ':')) {
4456 /* Set aside the old stash, so we can reset isa caches
4457 on its subclasses. */
4458 if((old_stash = GvHV(dstr))) {
4459 /* Make sure we do not lose it early. */
4460 SvREFCNT_inc_simple_void_NN(
4461 sv_2mortal((SV *)old_stash)
4468 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
4469 gp_free(MUTABLE_GV(dstr));
4471 GvGP_set(dstr, gp_ref(GvGP(gv)));
4474 HV * const stash = GvHV(dstr);
4476 old_stash ? (HV *)HvENAME_get(old_stash) : stash
4486 else if ((dtype == SVt_REGEXP || dtype == SVt_PVLV)
4487 && (stype == SVt_REGEXP || isREGEXP(sstr))) {
4488 reg_temp_copy((REGEXP*)dstr, (REGEXP*)sstr);
4490 else if (sflags & SVp_POK) {
4491 const STRLEN cur = SvCUR(sstr);
4492 const STRLEN len = SvLEN(sstr);
4495 * We have three basic ways to copy the string:
4501 * Which we choose is based on various factors. The following
4502 * things are listed in order of speed, fastest to slowest:
4504 * - Copying a short string
4505 * - Copy-on-write bookkeeping
4507 * - Copying a long string
4509 * We swipe the string (steal the string buffer) if the SV on the
4510 * rhs is about to be freed anyway (TEMP and refcnt==1). This is a
4511 * big win on long strings. It should be a win on short strings if
4512 * SvPVX_const(dstr) has to be allocated. If not, it should not
4513 * slow things down, as SvPVX_const(sstr) would have been freed
4516 * We also steal the buffer from a PADTMP (operator target) if it
4517 * is ‘long enough’. For short strings, a swipe does not help
4518 * here, as it causes more malloc calls the next time the target
4519 * is used. Benchmarks show that even if SvPVX_const(dstr) has to
4520 * be allocated it is still not worth swiping PADTMPs for short
4521 * strings, as the savings here are small.
4523 * If swiping is not an option, then we see whether it is
4524 * worth using copy-on-write. If the lhs already has a buf-
4525 * fer big enough and the string is short, we skip it and fall back
4526 * to method 3, since memcpy is faster for short strings than the
4527 * later bookkeeping overhead that copy-on-write entails.
4529 * If the rhs is not a copy-on-write string yet, then we also
4530 * consider whether the buffer is too large relative to the string
4531 * it holds. Some operations such as readline allocate a large
4532 * buffer in the expectation of reusing it. But turning such into
4533 * a COW buffer is counter-productive because it increases memory
4534 * usage by making readline allocate a new large buffer the sec-
4535 * ond time round. So, if the buffer is too large, again, we use
4538 * Finally, if there is no buffer on the left, or the buffer is too
4539 * small, then we use copy-on-write and make both SVs share the
4544 /* Whichever path we take through the next code, we want this true,
4545 and doing it now facilitates the COW check. */
4546 (void)SvPOK_only(dstr);
4550 /* slated for free anyway (and not COW)? */
4551 (sflags & (SVs_TEMP|SVf_IsCOW)) == SVs_TEMP
4552 /* or a swipable TARG */
4554 (SVs_PADTMP|SVf_READONLY|SVf_PROTECT|SVf_IsCOW))
4556 /* whose buffer is worth stealing */
4557 && CHECK_COWBUF_THRESHOLD(cur,len)
4560 !(sflags & SVf_OOK) && /* and not involved in OOK hack? */
4561 (!(flags & SV_NOSTEAL)) &&
4562 /* and we're allowed to steal temps */
4563 SvREFCNT(sstr) == 1 && /* and no other references to it? */
4564 len) /* and really is a string */
4565 { /* Passes the swipe test. */
4566 if (SvPVX_const(dstr)) /* we know that dtype >= SVt_PV */
4568 SvPV_set(dstr, SvPVX_mutable(sstr));
4569 SvLEN_set(dstr, SvLEN(sstr));
4570 SvCUR_set(dstr, SvCUR(sstr));
4573 (void)SvOK_off(sstr); /* NOTE: nukes most SvFLAGS on sstr */
4574 SvPV_set(sstr, NULL);
4579 else if (flags & SV_COW_SHARED_HASH_KEYS
4581 #ifdef PERL_COPY_ON_WRITE
4584 ( (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4585 /* If this is a regular (non-hek) COW, only so
4586 many COW "copies" are possible. */
4587 && CowREFCNT(sstr) != SV_COW_REFCNT_MAX ))
4588 : ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4589 && !(SvFLAGS(dstr) & SVf_BREAK)
4590 && CHECK_COW_THRESHOLD(cur,len) && cur+1 < len
4591 && (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4595 && !(SvFLAGS(dstr) & SVf_BREAK)
4598 /* Either it's a shared hash key, or it's suitable for
4602 PerlIO_printf(Perl_debug_log, "Copy on write: sstr --> dstr\n");
4608 if (!(sflags & SVf_IsCOW)) {
4610 CowREFCNT(sstr) = 0;
4613 if (SvPVX_const(dstr)) { /* we know that dtype >= SVt_PV */
4619 if (sflags & SVf_IsCOW) {
4623 SvPV_set(dstr, SvPVX_mutable(sstr));
4628 /* SvIsCOW_shared_hash */
4629 DEBUG_C(PerlIO_printf(Perl_debug_log,
4630 "Copy on write: Sharing hash\n"));
4632 assert (SvTYPE(dstr) >= SVt_PV);
4634 HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)))));
4636 SvLEN_set(dstr, len);
4637 SvCUR_set(dstr, cur);
4640 /* Failed the swipe test, and we cannot do copy-on-write either.
4641 Have to copy the string. */
4642 SvGROW(dstr, cur + 1); /* inlined from sv_setpvn */
4643 Move(SvPVX_const(sstr),SvPVX(dstr),cur,char);
4644 SvCUR_set(dstr, cur);
4645 *SvEND(dstr) = '\0';
4647 if (sflags & SVp_NOK) {
4648 SvNV_set(dstr, SvNVX(sstr));
4650 if (sflags & SVp_IOK) {
4651 SvIV_set(dstr, SvIVX(sstr));
4652 if (sflags & SVf_IVisUV)
4655 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_NOK|SVp_NOK|SVf_UTF8);
4657 const MAGIC * const smg = SvVSTRING_mg(sstr);
4659 sv_magic(dstr, NULL, PERL_MAGIC_vstring,
4660 smg->mg_ptr, smg->mg_len);
4661 SvRMAGICAL_on(dstr);
4665 else if (sflags & (SVp_IOK|SVp_NOK)) {
4666 (void)SvOK_off(dstr);
4667 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_IVisUV|SVf_NOK|SVp_NOK);
4668 if (sflags & SVp_IOK) {
4669 /* XXXX Do we want to set IsUV for IV(ROK)? Be extra safe... */
4670 SvIV_set(dstr, SvIVX(sstr));
4672 if (sflags & SVp_NOK) {
4673 SvNV_set(dstr, SvNVX(sstr));
4677 if (isGV_with_GP(sstr)) {
4678 gv_efullname3(dstr, MUTABLE_GV(sstr), "*");
4681 (void)SvOK_off(dstr);
4683 if (SvTAINTED(sstr))
4689 =for apidoc sv_set_undef
4691 Equivalent to C<sv_setsv(sv, &PL_sv_undef)>, but more efficient.
4692 Doesn't handle set magic.
4694 The perl equivalent is C<$sv = undef;>. Note that it doesn't free any string
4695 buffer, unlike C<undef $sv>.
4697 Introduced in perl 5.25.12.
4703 Perl_sv_set_undef(pTHX_ SV *sv)
4705 U32 type = SvTYPE(sv);
4707 PERL_ARGS_ASSERT_SV_SET_UNDEF;
4709 /* shortcut, NULL, IV, RV */
4711 if (type <= SVt_IV) {
4712 assert(!SvGMAGICAL(sv));
4713 if (SvREADONLY(sv)) {
4714 /* does undeffing PL_sv_undef count as modifying a read-only
4715 * variable? Some XS code does this */
4716 if (sv == &PL_sv_undef)
4718 Perl_croak_no_modify();
4723 sv_unref_flags(sv, 0);
4726 SvFLAGS(sv) = type; /* quickly turn off all flags */
4727 SvREFCNT_dec_NN(rv);
4731 SvFLAGS(sv) = type; /* quickly turn off all flags */
4736 Perl_croak(aTHX_ "panic: attempt to undefine a freed scalar %p",
4739 SV_CHECK_THINKFIRST_COW_DROP(sv);
4741 if (isGV_with_GP(sv))
4742 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4743 "Undefined value assigned to typeglob");
4751 =for apidoc sv_setsv_mg
4753 Like C<sv_setsv>, but also handles 'set' magic.
4759 Perl_sv_setsv_mg(pTHX_ SV *const dstr, SV *const sstr)
4761 PERL_ARGS_ASSERT_SV_SETSV_MG;
4763 sv_setsv(dstr,sstr);
4768 # define SVt_COW SVt_PV
4770 Perl_sv_setsv_cow(pTHX_ SV *dstr, SV *sstr)
4772 STRLEN cur = SvCUR(sstr);
4773 STRLEN len = SvLEN(sstr);
4775 #if defined(PERL_DEBUG_READONLY_COW) && defined(PERL_COPY_ON_WRITE)
4776 const bool already = cBOOL(SvIsCOW(sstr));
4779 PERL_ARGS_ASSERT_SV_SETSV_COW;
4782 PerlIO_printf(Perl_debug_log, "Fast copy on write: %p -> %p\n",
4783 (void*)sstr, (void*)dstr);
4790 if (SvTHINKFIRST(dstr))
4791 sv_force_normal_flags(dstr, SV_COW_DROP_PV);
4792 else if (SvPVX_const(dstr))
4793 Safefree(SvPVX_mutable(dstr));
4797 SvUPGRADE(dstr, SVt_COW);
4799 assert (SvPOK(sstr));
4800 assert (SvPOKp(sstr));
4802 if (SvIsCOW(sstr)) {
4804 if (SvLEN(sstr) == 0) {
4805 /* source is a COW shared hash key. */
4806 DEBUG_C(PerlIO_printf(Perl_debug_log,
4807 "Fast copy on write: Sharing hash\n"));
4808 new_pv = HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr))));
4811 assert(SvCUR(sstr)+1 < SvLEN(sstr));
4812 assert(CowREFCNT(sstr) < SV_COW_REFCNT_MAX);
4814 assert ((SvFLAGS(sstr) & CAN_COW_MASK) == CAN_COW_FLAGS);
4815 SvUPGRADE(sstr, SVt_COW);
4817 DEBUG_C(PerlIO_printf(Perl_debug_log,
4818 "Fast copy on write: Converting sstr to COW\n"));
4819 CowREFCNT(sstr) = 0;
4821 # ifdef PERL_DEBUG_READONLY_COW
4822 if (already) sv_buf_to_rw(sstr);
4825 new_pv = SvPVX_mutable(sstr);
4829 SvPV_set(dstr, new_pv);
4830 SvFLAGS(dstr) = (SVt_COW|SVf_POK|SVp_POK|SVf_IsCOW);
4833 SvLEN_set(dstr, len);
4834 SvCUR_set(dstr, cur);
4844 =for apidoc sv_setpv_bufsize
4846 Sets the SV to be a string of cur bytes length, with at least
4847 len bytes available. Ensures that there is a null byte at SvEND.
4848 Returns a char * pointer to the SvPV buffer.
4854 Perl_sv_setpv_bufsize(pTHX_ SV *const sv, const STRLEN cur, const STRLEN len)
4858 PERL_ARGS_ASSERT_SV_SETPV_BUFSIZE;
4860 SV_CHECK_THINKFIRST_COW_DROP(sv);
4861 SvUPGRADE(sv, SVt_PV);
4862 pv = SvGROW(sv, len + 1);
4865 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4868 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4873 =for apidoc sv_setpvn
4875 Copies a string (possibly containing embedded C<NUL> characters) into an SV.
4876 The C<len> parameter indicates the number of
4877 bytes to be copied. If the C<ptr> argument is NULL the SV will become
4878 undefined. Does not handle 'set' magic. See C<L</sv_setpvn_mg>>.
4884 Perl_sv_setpvn(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4888 PERL_ARGS_ASSERT_SV_SETPVN;
4890 SV_CHECK_THINKFIRST_COW_DROP(sv);
4891 if (isGV_with_GP(sv))
4892 Perl_croak_no_modify();
4898 /* len is STRLEN which is unsigned, need to copy to signed */
4901 Perl_croak(aTHX_ "panic: sv_setpvn called with negative strlen %"
4904 SvUPGRADE(sv, SVt_PV);
4906 dptr = SvGROW(sv, len + 1);
4907 Move(ptr,dptr,len,char);
4910 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4912 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4916 =for apidoc sv_setpvn_mg
4918 Like C<sv_setpvn>, but also handles 'set' magic.
4924 Perl_sv_setpvn_mg(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4926 PERL_ARGS_ASSERT_SV_SETPVN_MG;
4928 sv_setpvn(sv,ptr,len);
4933 =for apidoc sv_setpv
4935 Copies a string into an SV. The string must be terminated with a C<NUL>
4936 character, and not contain embeded C<NUL>'s.
4937 Does not handle 'set' magic. See C<L</sv_setpv_mg>>.
4943 Perl_sv_setpv(pTHX_ SV *const sv, const char *const ptr)
4947 PERL_ARGS_ASSERT_SV_SETPV;
4949 SV_CHECK_THINKFIRST_COW_DROP(sv);
4955 SvUPGRADE(sv, SVt_PV);
4957 SvGROW(sv, len + 1);
4958 Move(ptr,SvPVX(sv),len+1,char);
4960 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4962 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4966 =for apidoc sv_setpv_mg
4968 Like C<sv_setpv>, but also handles 'set' magic.
4974 Perl_sv_setpv_mg(pTHX_ SV *const sv, const char *const ptr)
4976 PERL_ARGS_ASSERT_SV_SETPV_MG;
4983 Perl_sv_sethek(pTHX_ SV *const sv, const HEK *const hek)
4985 PERL_ARGS_ASSERT_SV_SETHEK;
4991 if (HEK_LEN(hek) == HEf_SVKEY) {
4992 sv_setsv(sv, *(SV**)HEK_KEY(hek));
4995 const int flags = HEK_FLAGS(hek);
4996 if (flags & HVhek_WASUTF8) {
4997 STRLEN utf8_len = HEK_LEN(hek);
4998 char *as_utf8 = (char *)bytes_to_utf8((U8*)HEK_KEY(hek), &utf8_len);
4999 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
5002 } else if (flags & HVhek_UNSHARED) {
5003 sv_setpvn(sv, HEK_KEY(hek), HEK_LEN(hek));
5006 else SvUTF8_off(sv);
5010 SV_CHECK_THINKFIRST_COW_DROP(sv);
5011 SvUPGRADE(sv, SVt_PV);
5013 SvPV_set(sv,(char *)HEK_KEY(share_hek_hek(hek)));
5014 SvCUR_set(sv, HEK_LEN(hek));
5020 else SvUTF8_off(sv);
5028 =for apidoc sv_usepvn_flags
5030 Tells an SV to use C<ptr> to find its string value. Normally the
5031 string is stored inside the SV, but sv_usepvn allows the SV to use an
5032 outside string. C<ptr> should point to memory that was allocated
5033 by L<C<Newx>|perlclib/Memory Management and String Handling>. It must be
5034 the start of a C<Newx>-ed block of memory, and not a pointer to the
5035 middle of it (beware of L<C<OOK>|perlguts/Offsets> and copy-on-write),
5036 and not be from a non-C<Newx> memory allocator like C<malloc>. The
5037 string length, C<len>, must be supplied. By default this function
5038 will C<Renew> (i.e. realloc, move) the memory pointed to by C<ptr>,
5039 so that pointer should not be freed or used by the programmer after
5040 giving it to C<sv_usepvn>, and neither should any pointers from "behind"
5041 that pointer (e.g. ptr + 1) be used.
5043 If S<C<flags & SV_SMAGIC>> is true, will call C<SvSETMAGIC>. If
5044 S<C<flags & SV_HAS_TRAILING_NUL>> is true, then C<ptr[len]> must be C<NUL>,
5046 will be skipped (i.e. the buffer is actually at least 1 byte longer than
5047 C<len>, and already meets the requirements for storing in C<SvPVX>).
5053 Perl_sv_usepvn_flags(pTHX_ SV *const sv, char *ptr, const STRLEN len, const U32 flags)
5057 PERL_ARGS_ASSERT_SV_USEPVN_FLAGS;
5059 SV_CHECK_THINKFIRST_COW_DROP(sv);
5060 SvUPGRADE(sv, SVt_PV);
5063 if (flags & SV_SMAGIC)
5067 if (SvPVX_const(sv))
5071 if (flags & SV_HAS_TRAILING_NUL)
5072 assert(ptr[len] == '\0');
5075 allocate = (flags & SV_HAS_TRAILING_NUL)
5077 #ifdef Perl_safesysmalloc_size
5080 PERL_STRLEN_ROUNDUP(len + 1);
5082 if (flags & SV_HAS_TRAILING_NUL) {
5083 /* It's long enough - do nothing.
5084 Specifically Perl_newCONSTSUB is relying on this. */
5087 /* Force a move to shake out bugs in callers. */
5088 char *new_ptr = (char*)safemalloc(allocate);
5089 Copy(ptr, new_ptr, len, char);
5090 PoisonFree(ptr,len,char);
5094 ptr = (char*) saferealloc (ptr, allocate);
5097 #ifdef Perl_safesysmalloc_size
5098 SvLEN_set(sv, Perl_safesysmalloc_size(ptr));
5100 SvLEN_set(sv, allocate);
5104 if (!(flags & SV_HAS_TRAILING_NUL)) {
5107 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5109 if (flags & SV_SMAGIC)
5115 S_sv_uncow(pTHX_ SV * const sv, const U32 flags)
5117 assert(SvIsCOW(sv));
5120 const char * const pvx = SvPVX_const(sv);
5121 const STRLEN len = SvLEN(sv);
5122 const STRLEN cur = SvCUR(sv);
5126 PerlIO_printf(Perl_debug_log,
5127 "Copy on write: Force normal %ld\n",
5133 # ifdef PERL_COPY_ON_WRITE
5135 /* Must do this first, since the CowREFCNT uses SvPVX and
5136 we need to write to CowREFCNT, or de-RO the whole buffer if we are
5137 the only owner left of the buffer. */
5138 sv_buf_to_rw(sv); /* NOOP if RO-ing not supported */
5140 U8 cowrefcnt = CowREFCNT(sv);
5141 if(cowrefcnt != 0) {
5143 CowREFCNT(sv) = cowrefcnt;
5148 /* Else we are the only owner of the buffer. */
5153 /* This SV doesn't own the buffer, so need to Newx() a new one: */
5158 if (flags & SV_COW_DROP_PV) {
5159 /* OK, so we don't need to copy our buffer. */
5162 SvGROW(sv, cur + 1);
5163 Move(pvx,SvPVX(sv),cur,char);
5169 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5177 const char * const pvx = SvPVX_const(sv);
5178 const STRLEN len = SvCUR(sv);
5182 if (flags & SV_COW_DROP_PV) {
5183 /* OK, so we don't need to copy our buffer. */
5186 SvGROW(sv, len + 1);
5187 Move(pvx,SvPVX(sv),len,char);
5190 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5197 =for apidoc sv_force_normal_flags
5199 Undo various types of fakery on an SV, where fakery means
5200 "more than" a string: if the PV is a shared string, make
5201 a private copy; if we're a ref, stop refing; if we're a glob, downgrade to
5202 an C<xpvmg>; if we're a copy-on-write scalar, this is the on-write time when
5203 we do the copy, and is also used locally; if this is a
5204 vstring, drop the vstring magic. If C<SV_COW_DROP_PV> is set
5205 then a copy-on-write scalar drops its PV buffer (if any) and becomes
5206 C<SvPOK_off> rather than making a copy. (Used where this
5207 scalar is about to be set to some other value.) In addition,
5208 the C<flags> parameter gets passed to C<sv_unref_flags()>
5209 when unreffing. C<sv_force_normal> calls this function
5210 with flags set to 0.
5212 This function is expected to be used to signal to perl that this SV is
5213 about to be written to, and any extra book-keeping needs to be taken care
5214 of. Hence, it croaks on read-only values.
5220 Perl_sv_force_normal_flags(pTHX_ SV *const sv, const U32 flags)
5222 PERL_ARGS_ASSERT_SV_FORCE_NORMAL_FLAGS;
5225 Perl_croak_no_modify();
5226 else if (SvIsCOW(sv) && LIKELY(SvTYPE(sv) != SVt_PVHV))
5227 S_sv_uncow(aTHX_ sv, flags);
5229 sv_unref_flags(sv, flags);
5230 else if (SvFAKE(sv) && isGV_with_GP(sv))
5231 sv_unglob(sv, flags);
5232 else if (SvFAKE(sv) && isREGEXP(sv)) {
5233 /* Need to downgrade the REGEXP to a simple(r) scalar. This is analogous
5234 to sv_unglob. We only need it here, so inline it. */
5235 const bool islv = SvTYPE(sv) == SVt_PVLV;
5236 const svtype new_type =
5237 islv ? SVt_NULL : SvMAGIC(sv) || SvSTASH(sv) ? SVt_PVMG : SVt_PV;
5238 SV *const temp = newSV_type(new_type);
5239 regexp *old_rx_body;
5241 if (new_type == SVt_PVMG) {
5242 SvMAGIC_set(temp, SvMAGIC(sv));
5243 SvMAGIC_set(sv, NULL);
5244 SvSTASH_set(temp, SvSTASH(sv));
5245 SvSTASH_set(sv, NULL);
5248 SvCUR_set(temp, SvCUR(sv));
5249 /* Remember that SvPVX is in the head, not the body. */
5250 assert(ReANY((REGEXP *)sv)->mother_re);
5253 /* LV-as-regex has sv->sv_any pointing to an XPVLV body,
5254 * whose xpvlenu_rx field points to the regex body */
5255 XPV *xpv = (XPV*)(SvANY(sv));
5256 old_rx_body = xpv->xpv_len_u.xpvlenu_rx;
5257 xpv->xpv_len_u.xpvlenu_rx = NULL;
5260 old_rx_body = ReANY((REGEXP *)sv);
5262 /* Their buffer is already owned by someone else. */
5263 if (flags & SV_COW_DROP_PV) {
5264 /* SvLEN is already 0. For SVt_REGEXP, we have a brand new
5265 zeroed body. For SVt_PVLV, we zeroed it above (len field
5266 a union with xpvlenu_rx) */
5267 assert(!SvLEN(islv ? sv : temp));
5268 sv->sv_u.svu_pv = 0;
5271 sv->sv_u.svu_pv = savepvn(RX_WRAPPED((REGEXP *)sv), SvCUR(sv));
5272 SvLEN_set(islv ? sv : temp, SvCUR(sv)+1);
5276 /* Now swap the rest of the bodies. */
5280 SvFLAGS(sv) &= ~SVTYPEMASK;
5281 SvFLAGS(sv) |= new_type;
5282 SvANY(sv) = SvANY(temp);
5285 SvFLAGS(temp) &= ~(SVTYPEMASK);
5286 SvFLAGS(temp) |= SVt_REGEXP|SVf_FAKE;
5287 SvANY(temp) = old_rx_body;
5289 SvREFCNT_dec_NN(temp);
5291 else if (SvVOK(sv)) sv_unmagic(sv, PERL_MAGIC_vstring);
5297 Efficient removal of characters from the beginning of the string buffer.
5298 C<SvPOK(sv)>, or at least C<SvPOKp(sv)>, must be true and C<ptr> must be a
5299 pointer to somewhere inside the string buffer. C<ptr> becomes the first
5300 character of the adjusted string. Uses the C<OOK> hack. On return, only
5301 C<SvPOK(sv)> and C<SvPOKp(sv)> among the C<OK> flags will be true.
5303 Beware: after this function returns, C<ptr> and SvPVX_const(sv) may no longer
5304 refer to the same chunk of data.
5306 The unfortunate similarity of this function's name to that of Perl's C<chop>
5307 operator is strictly coincidental. This function works from the left;
5308 C<chop> works from the right.
5314 Perl_sv_chop(pTHX_ SV *const sv, const char *const ptr)
5325 PERL_ARGS_ASSERT_SV_CHOP;
5327 if (!ptr || !SvPOKp(sv))
5329 delta = ptr - SvPVX_const(sv);
5331 /* Nothing to do. */
5334 max_delta = SvLEN(sv) ? SvLEN(sv) : SvCUR(sv);
5335 if (delta > max_delta)
5336 Perl_croak(aTHX_ "panic: sv_chop ptr=%p, start=%p, end=%p",
5337 ptr, SvPVX_const(sv), SvPVX_const(sv) + max_delta);
5338 /* SvPVX(sv) may move in SV_CHECK_THINKFIRST(sv), so don't use ptr any more */
5339 SV_CHECK_THINKFIRST(sv);
5340 SvPOK_only_UTF8(sv);
5343 if (!SvLEN(sv)) { /* make copy of shared string */
5344 const char *pvx = SvPVX_const(sv);
5345 const STRLEN len = SvCUR(sv);
5346 SvGROW(sv, len + 1);
5347 Move(pvx,SvPVX(sv),len,char);
5353 SvOOK_offset(sv, old_delta);
5355 SvLEN_set(sv, SvLEN(sv) - delta);
5356 SvCUR_set(sv, SvCUR(sv) - delta);
5357 SvPV_set(sv, SvPVX(sv) + delta);
5359 p = (U8 *)SvPVX_const(sv);
5362 /* how many bytes were evacuated? we will fill them with sentinel
5363 bytes, except for the part holding the new offset of course. */
5366 evacn += (old_delta < 0x100 ? 1 : 1 + sizeof(STRLEN));
5368 assert(evacn <= delta + old_delta);
5372 /* This sets 'delta' to the accumulated value of all deltas so far */
5376 /* If 'delta' fits in a byte, store it just prior to the new beginning of
5377 * the string; otherwise store a 0 byte there and store 'delta' just prior
5378 * to that, using as many bytes as a STRLEN occupies. Thus it overwrites a
5379 * portion of the chopped part of the string */
5380 if (delta < 0x100) {
5384 p -= sizeof(STRLEN);
5385 Copy((U8*)&delta, p, sizeof(STRLEN), U8);
5389 /* Fill the preceding buffer with sentinals to verify that no-one is
5399 =for apidoc sv_catpvn
5401 Concatenates the string onto the end of the string which is in the SV.
5402 C<len> indicates number of bytes to copy. If the SV has the UTF-8
5403 status set, then the bytes appended should be valid UTF-8.
5404 Handles 'get' magic, but not 'set' magic. See C<L</sv_catpvn_mg>>.
5406 =for apidoc sv_catpvn_flags
5408 Concatenates the string onto the end of the string which is in the SV. The
5409 C<len> indicates number of bytes to copy.
5411 By default, the string appended is assumed to be valid UTF-8 if the SV has
5412 the UTF-8 status set, and a string of bytes otherwise. One can force the
5413 appended string to be interpreted as UTF-8 by supplying the C<SV_CATUTF8>
5414 flag, and as bytes by supplying the C<SV_CATBYTES> flag; the SV or the
5415 string appended will be upgraded to UTF-8 if necessary.
5417 If C<flags> has the C<SV_SMAGIC> bit set, will
5418 C<mg_set> on C<dsv> afterwards if appropriate.
5419 C<sv_catpvn> and C<sv_catpvn_nomg> are implemented
5420 in terms of this function.
5426 Perl_sv_catpvn_flags(pTHX_ SV *const dsv, const char *sstr, const STRLEN slen, const I32 flags)
5429 const char * const dstr = SvPV_force_flags(dsv, dlen, flags);
5431 PERL_ARGS_ASSERT_SV_CATPVN_FLAGS;
5432 assert((flags & (SV_CATBYTES|SV_CATUTF8)) != (SV_CATBYTES|SV_CATUTF8));
5434 if (!(flags & SV_CATBYTES) || !SvUTF8(dsv)) {
5435 if (flags & SV_CATUTF8 && !SvUTF8(dsv)) {
5436 sv_utf8_upgrade_flags_grow(dsv, 0, slen + 1);
5439 else SvGROW(dsv, dlen + slen + 3);
5441 sstr = SvPVX_const(dsv);
5442 Move(sstr, SvPVX(dsv) + dlen, slen, char);
5443 SvCUR_set(dsv, SvCUR(dsv) + slen);
5446 /* We inline bytes_to_utf8, to avoid an extra malloc. */
5447 const char * const send = sstr + slen;
5450 /* Something this code does not account for, which I think is
5451 impossible; it would require the same pv to be treated as
5452 bytes *and* utf8, which would indicate a bug elsewhere. */
5453 assert(sstr != dstr);
5455 SvGROW(dsv, dlen + slen * 2 + 3);
5456 d = (U8 *)SvPVX(dsv) + dlen;
5458 while (sstr < send) {
5459 append_utf8_from_native_byte(*sstr, &d);
5462 SvCUR_set(dsv, d-(const U8 *)SvPVX(dsv));
5465 (void)SvPOK_only_UTF8(dsv); /* validate pointer */
5467 if (flags & SV_SMAGIC)
5472 =for apidoc sv_catsv
5474 Concatenates the string from SV C<ssv> onto the end of the string in SV
5475 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5476 Handles 'get' magic on both SVs, but no 'set' magic. See C<L</sv_catsv_mg>>
5477 and C<L</sv_catsv_nomg>>.
5479 =for apidoc sv_catsv_flags
5481 Concatenates the string from SV C<ssv> onto the end of the string in SV
5482 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5483 If C<flags> has the C<SV_GMAGIC> bit set, will call C<mg_get> on both SVs if
5484 appropriate. If C<flags> has the C<SV_SMAGIC> bit set, C<mg_set> will be called on
5485 the modified SV afterward, if appropriate. C<sv_catsv>, C<sv_catsv_nomg>,
5486 and C<sv_catsv_mg> are implemented in terms of this function.
5491 Perl_sv_catsv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
5493 PERL_ARGS_ASSERT_SV_CATSV_FLAGS;
5497 const char *spv = SvPV_flags_const(ssv, slen, flags);
5498 if (flags & SV_GMAGIC)
5500 sv_catpvn_flags(dsv, spv, slen,
5501 DO_UTF8(ssv) ? SV_CATUTF8 : SV_CATBYTES);
5502 if (flags & SV_SMAGIC)
5508 =for apidoc sv_catpv
5510 Concatenates the C<NUL>-terminated string onto the end of the string which is
5512 If the SV has the UTF-8 status set, then the bytes appended should be
5513 valid UTF-8. Handles 'get' magic, but not 'set' magic. See
5519 Perl_sv_catpv(pTHX_ SV *const sv, const char *ptr)
5525 PERL_ARGS_ASSERT_SV_CATPV;
5529 junk = SvPV_force(sv, tlen);
5531 SvGROW(sv, tlen + len + 1);
5533 ptr = SvPVX_const(sv);
5534 Move(ptr,SvPVX(sv)+tlen,len+1,char);
5535 SvCUR_set(sv, SvCUR(sv) + len);
5536 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5541 =for apidoc sv_catpv_flags
5543 Concatenates the C<NUL>-terminated string onto the end of the string which is
5545 If the SV has the UTF-8 status set, then the bytes appended should
5546 be valid UTF-8. If C<flags> has the C<SV_SMAGIC> bit set, will C<mg_set>
5547 on the modified SV if appropriate.
5553 Perl_sv_catpv_flags(pTHX_ SV *dstr, const char *sstr, const I32 flags)
5555 PERL_ARGS_ASSERT_SV_CATPV_FLAGS;
5556 sv_catpvn_flags(dstr, sstr, strlen(sstr), flags);
5560 =for apidoc sv_catpv_mg
5562 Like C<sv_catpv>, but also handles 'set' magic.
5568 Perl_sv_catpv_mg(pTHX_ SV *const sv, const char *const ptr)
5570 PERL_ARGS_ASSERT_SV_CATPV_MG;
5579 Creates a new SV. A non-zero C<len> parameter indicates the number of
5580 bytes of preallocated string space the SV should have. An extra byte for a
5581 trailing C<NUL> is also reserved. (C<SvPOK> is not set for the SV even if string
5582 space is allocated.) The reference count for the new SV is set to 1.
5584 In 5.9.3, C<newSV()> replaces the older C<NEWSV()> API, and drops the first
5585 parameter, I<x>, a debug aid which allowed callers to identify themselves.
5586 This aid has been superseded by a new build option, C<PERL_MEM_LOG> (see
5587 L<perlhacktips/PERL_MEM_LOG>). The older API is still there for use in XS
5588 modules supporting older perls.
5594 Perl_newSV(pTHX_ const STRLEN len)
5600 sv_grow(sv, len + 1);
5605 =for apidoc sv_magicext
5607 Adds magic to an SV, upgrading it if necessary. Applies the
5608 supplied C<vtable> and returns a pointer to the magic added.
5610 Note that C<sv_magicext> will allow things that C<sv_magic> will not.
5611 In particular, you can add magic to C<SvREADONLY> SVs, and add more than
5612 one instance of the same C<how>.
5614 If C<namlen> is greater than zero then a C<savepvn> I<copy> of C<name> is
5615 stored, if C<namlen> is zero then C<name> is stored as-is and - as another
5616 special case - if C<(name && namlen == HEf_SVKEY)> then C<name> is assumed
5617 to contain an SV* and is stored as-is with its C<REFCNT> incremented.
5619 (This is now used as a subroutine by C<sv_magic>.)
5624 Perl_sv_magicext(pTHX_ SV *const sv, SV *const obj, const int how,
5625 const MGVTBL *const vtable, const char *const name, const I32 namlen)
5629 PERL_ARGS_ASSERT_SV_MAGICEXT;
5631 SvUPGRADE(sv, SVt_PVMG);
5632 Newxz(mg, 1, MAGIC);
5633 mg->mg_moremagic = SvMAGIC(sv);
5634 SvMAGIC_set(sv, mg);
5636 /* Sometimes a magic contains a reference loop, where the sv and
5637 object refer to each other. To prevent a reference loop that
5638 would prevent such objects being freed, we look for such loops
5639 and if we find one we avoid incrementing the object refcount.
5641 Note we cannot do this to avoid self-tie loops as intervening RV must
5642 have its REFCNT incremented to keep it in existence.
5645 if (!obj || obj == sv ||
5646 how == PERL_MAGIC_arylen ||
5647 how == PERL_MAGIC_regdata ||
5648 how == PERL_MAGIC_regdatum ||
5649 how == PERL_MAGIC_symtab ||
5650 (SvTYPE(obj) == SVt_PVGV &&
5651 (GvSV(obj) == sv || GvHV(obj) == (const HV *)sv
5652 || GvAV(obj) == (const AV *)sv || GvCV(obj) == (const CV *)sv
5653 || GvIOp(obj) == (const IO *)sv || GvFORM(obj) == (const CV *)sv)))
5658 mg->mg_obj = SvREFCNT_inc_simple(obj);
5659 mg->mg_flags |= MGf_REFCOUNTED;
5662 /* Normal self-ties simply pass a null object, and instead of
5663 using mg_obj directly, use the SvTIED_obj macro to produce a
5664 new RV as needed. For glob "self-ties", we are tieing the PVIO
5665 with an RV obj pointing to the glob containing the PVIO. In
5666 this case, to avoid a reference loop, we need to weaken the
5670 if (how == PERL_MAGIC_tiedscalar && SvTYPE(sv) == SVt_PVIO &&
5671 obj && SvROK(obj) && GvIO(SvRV(obj)) == (const IO *)sv)
5677 mg->mg_len = namlen;
5680 mg->mg_ptr = savepvn(name, namlen);
5681 else if (namlen == HEf_SVKEY) {
5682 /* Yes, this is casting away const. This is only for the case of
5683 HEf_SVKEY. I think we need to document this aberation of the
5684 constness of the API, rather than making name non-const, as
5685 that change propagating outwards a long way. */
5686 mg->mg_ptr = (char*)SvREFCNT_inc_simple_NN((SV *)name);
5688 mg->mg_ptr = (char *) name;
5690 mg->mg_virtual = (MGVTBL *) vtable;
5697 Perl_sv_magicext_mglob(pTHX_ SV *sv)
5699 PERL_ARGS_ASSERT_SV_MAGICEXT_MGLOB;
5700 if (SvTYPE(sv) == SVt_PVLV && LvTYPE(sv) == 'y') {
5701 /* This sv is only a delegate. //g magic must be attached to
5706 return sv_magicext(sv, NULL, PERL_MAGIC_regex_global,
5707 &PL_vtbl_mglob, 0, 0);
5711 =for apidoc sv_magic
5713 Adds magic to an SV. First upgrades C<sv> to type C<SVt_PVMG> if
5714 necessary, then adds a new magic item of type C<how> to the head of the
5717 See C<L</sv_magicext>> (which C<sv_magic> now calls) for a description of the
5718 handling of the C<name> and C<namlen> arguments.
5720 You need to use C<sv_magicext> to add magic to C<SvREADONLY> SVs and also
5721 to add more than one instance of the same C<how>.
5727 Perl_sv_magic(pTHX_ SV *const sv, SV *const obj, const int how,
5728 const char *const name, const I32 namlen)
5730 const MGVTBL *vtable;
5733 unsigned int vtable_index;
5735 PERL_ARGS_ASSERT_SV_MAGIC;
5737 if (how < 0 || (unsigned)how >= C_ARRAY_LENGTH(PL_magic_data)
5738 || ((flags = PL_magic_data[how]),
5739 (vtable_index = flags & PERL_MAGIC_VTABLE_MASK)
5740 > magic_vtable_max))
5741 Perl_croak(aTHX_ "Don't know how to handle magic of type \\%o", how);
5743 /* PERL_MAGIC_ext is reserved for use by extensions not perl internals.
5744 Useful for attaching extension internal data to perl vars.
5745 Note that multiple extensions may clash if magical scalars
5746 etc holding private data from one are passed to another. */
5748 vtable = (vtable_index == magic_vtable_max)
5749 ? NULL : PL_magic_vtables + vtable_index;
5751 if (SvREADONLY(sv)) {
5753 !PERL_MAGIC_TYPE_READONLY_ACCEPTABLE(how)
5756 Perl_croak_no_modify();
5759 if (SvMAGICAL(sv) || (how == PERL_MAGIC_taint && SvTYPE(sv) >= SVt_PVMG)) {
5760 if (SvMAGIC(sv) && (mg = mg_find(sv, how))) {
5761 /* sv_magic() refuses to add a magic of the same 'how' as an
5764 if (how == PERL_MAGIC_taint)
5770 /* Force pos to be stored as characters, not bytes. */
5771 if (SvMAGICAL(sv) && DO_UTF8(sv)
5772 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
5774 && mg->mg_flags & MGf_BYTES) {
5775 mg->mg_len = (SSize_t)sv_pos_b2u_flags(sv, (STRLEN)mg->mg_len,
5777 mg->mg_flags &= ~MGf_BYTES;
5780 /* Rest of work is done else where */
5781 mg = sv_magicext(sv,obj,how,vtable,name,namlen);
5784 case PERL_MAGIC_taint:
5787 case PERL_MAGIC_ext:
5788 case PERL_MAGIC_dbfile:
5795 S_sv_unmagicext_flags(pTHX_ SV *const sv, const int type, MGVTBL *vtbl, const U32 flags)
5802 if (SvTYPE(sv) < SVt_PVMG || !SvMAGIC(sv))
5804 mgp = &(((XPVMG*) SvANY(sv))->xmg_u.xmg_magic);
5805 for (mg = *mgp; mg; mg = *mgp) {
5806 const MGVTBL* const virt = mg->mg_virtual;
5807 if (mg->mg_type == type && (!flags || virt == vtbl)) {
5808 *mgp = mg->mg_moremagic;
5809 if (virt && virt->svt_free)
5810 virt->svt_free(aTHX_ sv, mg);
5811 if (mg->mg_ptr && mg->mg_type != PERL_MAGIC_regex_global) {
5813 Safefree(mg->mg_ptr);
5814 else if (mg->mg_len == HEf_SVKEY)
5815 SvREFCNT_dec(MUTABLE_SV(mg->mg_ptr));
5816 else if (mg->mg_type == PERL_MAGIC_utf8)
5817 Safefree(mg->mg_ptr);
5819 if (mg->mg_flags & MGf_REFCOUNTED)
5820 SvREFCNT_dec(mg->mg_obj);
5824 mgp = &mg->mg_moremagic;
5827 if (SvMAGICAL(sv)) /* if we're under save_magic, wait for restore_magic; */
5828 mg_magical(sv); /* else fix the flags now */
5837 =for apidoc sv_unmagic
5839 Removes all magic of type C<type> from an SV.
5845 Perl_sv_unmagic(pTHX_ SV *const sv, const int type)
5847 PERL_ARGS_ASSERT_SV_UNMAGIC;
5848 return S_sv_unmagicext_flags(aTHX_ sv, type, NULL, 0);
5852 =for apidoc sv_unmagicext
5854 Removes all magic of type C<type> with the specified C<vtbl> from an SV.
5860 Perl_sv_unmagicext(pTHX_ SV *const sv, const int type, MGVTBL *vtbl)
5862 PERL_ARGS_ASSERT_SV_UNMAGICEXT;
5863 return S_sv_unmagicext_flags(aTHX_ sv, type, vtbl, 1);
5867 =for apidoc sv_rvweaken
5869 Weaken a reference: set the C<SvWEAKREF> flag on this RV; give the
5870 referred-to SV C<PERL_MAGIC_backref> magic if it hasn't already; and
5871 push a back-reference to this RV onto the array of backreferences
5872 associated with that magic. If the RV is magical, set magic will be
5873 called after the RV is cleared. Silently ignores C<undef> and warns
5874 on already-weak references.
5880 Perl_sv_rvweaken(pTHX_ SV *const sv)
5884 PERL_ARGS_ASSERT_SV_RVWEAKEN;
5886 if (!SvOK(sv)) /* let undefs pass */
5889 Perl_croak(aTHX_ "Can't weaken a nonreference");
5890 else if (SvWEAKREF(sv)) {
5891 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is already weak");
5894 else if (SvREADONLY(sv)) croak_no_modify();
5896 Perl_sv_add_backref(aTHX_ tsv, sv);
5898 SvREFCNT_dec_NN(tsv);
5903 =for apidoc sv_rvunweaken
5905 Unweaken a reference: Clear the C<SvWEAKREF> flag on this RV; remove
5906 the backreference to this RV from the array of backreferences
5907 associated with the target SV, increment the refcount of the target.
5908 Silently ignores C<undef> and warns on non-weak references.
5914 Perl_sv_rvunweaken(pTHX_ SV *const sv)
5918 PERL_ARGS_ASSERT_SV_RVUNWEAKEN;
5920 if (!SvOK(sv)) /* let undefs pass */
5923 Perl_croak(aTHX_ "Can't unweaken a nonreference");
5924 else if (!SvWEAKREF(sv)) {
5925 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is not weak");
5928 else if (SvREADONLY(sv)) croak_no_modify();
5933 SvREFCNT_inc_NN(tsv);
5934 Perl_sv_del_backref(aTHX_ tsv, sv);
5939 =for apidoc sv_get_backrefs
5941 If C<sv> is the target of a weak reference then it returns the back
5942 references structure associated with the sv; otherwise return C<NULL>.
5944 When returning a non-null result the type of the return is relevant. If it
5945 is an AV then the elements of the AV are the weak reference RVs which
5946 point at this item. If it is any other type then the item itself is the
5949 See also C<Perl_sv_add_backref()>, C<Perl_sv_del_backref()>,
5950 C<Perl_sv_kill_backrefs()>
5956 Perl_sv_get_backrefs(SV *const sv)
5960 PERL_ARGS_ASSERT_SV_GET_BACKREFS;
5962 /* find slot to store array or singleton backref */
5964 if (SvTYPE(sv) == SVt_PVHV) {
5966 struct xpvhv_aux * const iter = HvAUX((HV *)sv);
5967 backrefs = (SV *)iter->xhv_backreferences;
5969 } else if (SvMAGICAL(sv)) {
5970 MAGIC *mg = mg_find(sv, PERL_MAGIC_backref);
5972 backrefs = mg->mg_obj;
5977 /* Give tsv backref magic if it hasn't already got it, then push a
5978 * back-reference to sv onto the array associated with the backref magic.
5980 * As an optimisation, if there's only one backref and it's not an AV,
5981 * store it directly in the HvAUX or mg_obj slot, avoiding the need to
5982 * allocate an AV. (Whether the slot holds an AV tells us whether this is
5986 /* A discussion about the backreferences array and its refcount:
5988 * The AV holding the backreferences is pointed to either as the mg_obj of
5989 * PERL_MAGIC_backref, or in the specific case of a HV, from the
5990 * xhv_backreferences field. The array is created with a refcount
5991 * of 2. This means that if during global destruction the array gets
5992 * picked on before its parent to have its refcount decremented by the
5993 * random zapper, it won't actually be freed, meaning it's still there for
5994 * when its parent gets freed.
5996 * When the parent SV is freed, the extra ref is killed by
5997 * Perl_sv_kill_backrefs. The other ref is killed, in the case of magic,
5998 * by mg_free() / MGf_REFCOUNTED, or for a hash, by Perl_hv_kill_backrefs.
6000 * When a single backref SV is stored directly, it is not reference
6005 Perl_sv_add_backref(pTHX_ SV *const tsv, SV *const sv)
6011 PERL_ARGS_ASSERT_SV_ADD_BACKREF;
6013 /* find slot to store array or singleton backref */
6015 if (SvTYPE(tsv) == SVt_PVHV) {
6016 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
6019 mg = mg_find(tsv, PERL_MAGIC_backref);
6021 mg = sv_magicext(tsv, NULL, PERL_MAGIC_backref, &PL_vtbl_backref, NULL, 0);
6022 svp = &(mg->mg_obj);
6025 /* create or retrieve the array */
6027 if ( (!*svp && SvTYPE(sv) == SVt_PVAV)
6028 || (*svp && SvTYPE(*svp) != SVt_PVAV)
6032 mg->mg_flags |= MGf_REFCOUNTED;
6035 SvREFCNT_inc_simple_void_NN(av);
6036 /* av now has a refcnt of 2; see discussion above */
6037 av_extend(av, *svp ? 2 : 1);
6039 /* move single existing backref to the array */
6040 AvARRAY(av)[++AvFILLp(av)] = *svp; /* av_push() */
6045 av = MUTABLE_AV(*svp);
6047 /* optimisation: store single backref directly in HvAUX or mg_obj */
6051 assert(SvTYPE(av) == SVt_PVAV);
6052 if (AvFILLp(av) >= AvMAX(av)) {
6053 av_extend(av, AvFILLp(av)+1);
6056 /* push new backref */
6057 AvARRAY(av)[++AvFILLp(av)] = sv; /* av_push() */
6060 /* delete a back-reference to ourselves from the backref magic associated
6061 * with the SV we point to.
6065 Perl_sv_del_backref(pTHX_ SV *const tsv, SV *const sv)
6069 PERL_ARGS_ASSERT_SV_DEL_BACKREF;
6071 if (SvTYPE(tsv) == SVt_PVHV) {
6073 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
6075 else if (SvIS_FREED(tsv) && PL_phase == PERL_PHASE_DESTRUCT) {
6076 /* It's possible for the the last (strong) reference to tsv to have
6077 become freed *before* the last thing holding a weak reference.
6078 If both survive longer than the backreferences array, then when
6079 the referent's reference count drops to 0 and it is freed, it's
6080 not able to chase the backreferences, so they aren't NULLed.
6082 For example, a CV holds a weak reference to its stash. If both the
6083 CV and the stash survive longer than the backreferences array,
6084 and the CV gets picked for the SvBREAK() treatment first,
6085 *and* it turns out that the stash is only being kept alive because
6086 of an our variable in the pad of the CV, then midway during CV
6087 destruction the stash gets freed, but CvSTASH() isn't set to NULL.
6088 It ends up pointing to the freed HV. Hence it's chased in here, and
6089 if this block wasn't here, it would hit the !svp panic just below.
6091 I don't believe that "better" destruction ordering is going to help
6092 here - during global destruction there's always going to be the
6093 chance that something goes out of order. We've tried to make it
6094 foolproof before, and it only resulted in evolutionary pressure on
6095 fools. Which made us look foolish for our hubris. :-(
6101 = SvMAGICAL(tsv) ? mg_find(tsv, PERL_MAGIC_backref) : NULL;
6102 svp = mg ? &(mg->mg_obj) : NULL;
6106 Perl_croak(aTHX_ "panic: del_backref, svp=0");
6108 /* It's possible that sv is being freed recursively part way through the
6109 freeing of tsv. If this happens, the backreferences array of tsv has
6110 already been freed, and so svp will be NULL. If this is the case,
6111 we should not panic. Instead, nothing needs doing, so return. */
6112 if (PL_phase == PERL_PHASE_DESTRUCT && SvREFCNT(tsv) == 0)
6114 Perl_croak(aTHX_ "panic: del_backref, *svp=%p phase=%s refcnt=%" UVuf,
6115 (void*)*svp, PL_phase_names[PL_phase], (UV)SvREFCNT(tsv));
6118 if (SvTYPE(*svp) == SVt_PVAV) {
6122 AV * const av = (AV*)*svp;
6124 assert(!SvIS_FREED(av));
6128 /* for an SV with N weak references to it, if all those
6129 * weak refs are deleted, then sv_del_backref will be called
6130 * N times and O(N^2) compares will be done within the backref
6131 * array. To ameliorate this potential slowness, we:
6132 * 1) make sure this code is as tight as possible;
6133 * 2) when looking for SV, look for it at both the head and tail of the
6134 * array first before searching the rest, since some create/destroy
6135 * patterns will cause the backrefs to be freed in order.
6142 SV **p = &svp[fill];
6143 SV *const topsv = *p;
6150 /* We weren't the last entry.
6151 An unordered list has this property that you
6152 can take the last element off the end to fill
6153 the hole, and it's still an unordered list :-)
6159 break; /* should only be one */
6166 AvFILLp(av) = fill-1;
6168 else if (SvIS_FREED(*svp) && PL_phase == PERL_PHASE_DESTRUCT) {
6169 /* freed AV; skip */
6172 /* optimisation: only a single backref, stored directly */
6174 Perl_croak(aTHX_ "panic: del_backref, *svp=%p, sv=%p",
6175 (void*)*svp, (void*)sv);
6182 Perl_sv_kill_backrefs(pTHX_ SV *const sv, AV *const av)
6188 PERL_ARGS_ASSERT_SV_KILL_BACKREFS;
6193 /* after multiple passes through Perl_sv_clean_all() for a thingy
6194 * that has badly leaked, the backref array may have gotten freed,
6195 * since we only protect it against 1 round of cleanup */
6196 if (SvIS_FREED(av)) {
6197 if (PL_in_clean_all) /* All is fair */
6200 "panic: magic_killbackrefs (freed backref AV/SV)");
6204 is_array = (SvTYPE(av) == SVt_PVAV);
6206 assert(!SvIS_FREED(av));
6209 last = svp + AvFILLp(av);
6212 /* optimisation: only a single backref, stored directly */
6218 while (svp <= last) {
6220 SV *const referrer = *svp;
6221 if (SvWEAKREF(referrer)) {
6222 /* XXX Should we check that it hasn't changed? */
6223 assert(SvROK(referrer));
6224 SvRV_set(referrer, 0);
6226 SvWEAKREF_off(referrer);
6227 SvSETMAGIC(referrer);
6228 } else if (SvTYPE(referrer) == SVt_PVGV ||
6229 SvTYPE(referrer) == SVt_PVLV) {
6230 assert(SvTYPE(sv) == SVt_PVHV); /* stash backref */
6231 /* You lookin' at me? */
6232 assert(GvSTASH(referrer));
6233 assert(GvSTASH(referrer) == (const HV *)sv);
6234 GvSTASH(referrer) = 0;
6235 } else if (SvTYPE(referrer) == SVt_PVCV ||
6236 SvTYPE(referrer) == SVt_PVFM) {
6237 if (SvTYPE(sv) == SVt_PVHV) { /* stash backref */
6238 /* You lookin' at me? */
6239 assert(CvSTASH(referrer));
6240 assert(CvSTASH(referrer) == (const HV *)sv);
6241 SvANY(MUTABLE_CV(referrer))->xcv_stash = 0;
6244 assert(SvTYPE(sv) == SVt_PVGV);
6245 /* You lookin' at me? */
6246 assert(CvGV(referrer));
6247 assert(CvGV(referrer) == (const GV *)sv);
6248 anonymise_cv_maybe(MUTABLE_GV(sv),
6249 MUTABLE_CV(referrer));
6254 "panic: magic_killbackrefs (flags=%" UVxf ")",
6255 (UV)SvFLAGS(referrer));
6266 SvREFCNT_dec_NN(av); /* remove extra count added by sv_add_backref() */
6272 =for apidoc sv_insert
6274 Inserts and/or replaces a string at the specified offset/length within the SV.
6275 Similar to the Perl C<substr()> function, with C<littlelen> bytes starting at
6276 C<little> replacing C<len> bytes of the string in C<bigstr> starting at
6277 C<offset>. Handles get magic.
6279 =for apidoc sv_insert_flags
6281 Same as C<sv_insert>, but the extra C<flags> are passed to the
6282 C<SvPV_force_flags> that applies to C<bigstr>.
6288 Perl_sv_insert_flags(pTHX_ SV *const bigstr, const STRLEN offset, const STRLEN len, const char *little, const STRLEN littlelen, const U32 flags)
6294 SSize_t i; /* better be sizeof(STRLEN) or bad things happen */
6297 PERL_ARGS_ASSERT_SV_INSERT_FLAGS;
6299 SvPV_force_flags(bigstr, curlen, flags);
6300 (void)SvPOK_only_UTF8(bigstr);
6302 if (little >= SvPVX(bigstr) &&
6303 little < SvPVX(bigstr) + (SvLEN(bigstr) ? SvLEN(bigstr) : SvCUR(bigstr))) {
6304 /* little is a pointer to within bigstr, since we can reallocate bigstr,
6305 or little...little+littlelen might overlap offset...offset+len we make a copy
6307 little = savepvn(little, littlelen);
6311 if (offset + len > curlen) {
6312 SvGROW(bigstr, offset+len+1);
6313 Zero(SvPVX(bigstr)+curlen, offset+len-curlen, char);
6314 SvCUR_set(bigstr, offset+len);
6318 i = littlelen - len;
6319 if (i > 0) { /* string might grow */
6320 big = SvGROW(bigstr, SvCUR(bigstr) + i + 1);
6321 mid = big + offset + len;
6322 midend = bigend = big + SvCUR(bigstr);
6325 while (midend > mid) /* shove everything down */
6326 *--bigend = *--midend;
6327 Move(little,big+offset,littlelen,char);
6328 SvCUR_set(bigstr, SvCUR(bigstr) + i);
6333 Move(little,SvPVX(bigstr)+offset,len,char);
6338 big = SvPVX(bigstr);
6341 bigend = big + SvCUR(bigstr);
6343 if (midend > bigend)
6344 Perl_croak(aTHX_ "panic: sv_insert, midend=%p, bigend=%p",
6347 if (mid - big > bigend - midend) { /* faster to shorten from end */
6349 Move(little, mid, littlelen,char);
6352 i = bigend - midend;
6354 Move(midend, mid, i,char);
6358 SvCUR_set(bigstr, mid - big);
6360 else if ((i = mid - big)) { /* faster from front */
6361 midend -= littlelen;
6363 Move(big, midend - i, i, char);
6364 sv_chop(bigstr,midend-i);
6366 Move(little, mid, littlelen,char);
6368 else if (littlelen) {
6369 midend -= littlelen;
6370 sv_chop(bigstr,midend);
6371 Move(little,midend,littlelen,char);
6374 sv_chop(bigstr,midend);
6380 =for apidoc sv_replace
6382 Make the first argument a copy of the second, then delete the original.
6383 The target SV physically takes over ownership of the body of the source SV
6384 and inherits its flags; however, the target keeps any magic it owns,
6385 and any magic in the source is discarded.
6386 Note that this is a rather specialist SV copying operation; most of the
6387 time you'll want to use C<sv_setsv> or one of its many macro front-ends.
6393 Perl_sv_replace(pTHX_ SV *const sv, SV *const nsv)
6395 const U32 refcnt = SvREFCNT(sv);
6397 PERL_ARGS_ASSERT_SV_REPLACE;
6399 SV_CHECK_THINKFIRST_COW_DROP(sv);
6400 if (SvREFCNT(nsv) != 1) {
6401 Perl_croak(aTHX_ "panic: reference miscount on nsv in sv_replace()"
6402 " (%" UVuf " != 1)", (UV) SvREFCNT(nsv));
6404 if (SvMAGICAL(sv)) {
6408 sv_upgrade(nsv, SVt_PVMG);
6409 SvMAGIC_set(nsv, SvMAGIC(sv));
6410 SvFLAGS(nsv) |= SvMAGICAL(sv);
6412 SvMAGIC_set(sv, NULL);
6416 assert(!SvREFCNT(sv));
6417 #ifdef DEBUG_LEAKING_SCALARS
6418 sv->sv_flags = nsv->sv_flags;
6419 sv->sv_any = nsv->sv_any;
6420 sv->sv_refcnt = nsv->sv_refcnt;
6421 sv->sv_u = nsv->sv_u;
6423 StructCopy(nsv,sv,SV);
6425 if(SvTYPE(sv) == SVt_IV) {
6426 SET_SVANY_FOR_BODYLESS_IV(sv);
6430 SvREFCNT(sv) = refcnt;
6431 SvFLAGS(nsv) |= SVTYPEMASK; /* Mark as freed */
6436 /* We're about to free a GV which has a CV that refers back to us.
6437 * If that CV will outlive us, make it anonymous (i.e. fix up its CvGV
6441 S_anonymise_cv_maybe(pTHX_ GV *gv, CV* cv)
6446 PERL_ARGS_ASSERT_ANONYMISE_CV_MAYBE;
6449 assert(SvREFCNT(gv) == 0);
6450 assert(isGV(gv) && isGV_with_GP(gv));
6452 assert(!CvANON(cv));
6453 assert(CvGV(cv) == gv);
6454 assert(!CvNAMED(cv));
6456 /* will the CV shortly be freed by gp_free() ? */
6457 if (GvCV(gv) == cv && GvGP(gv)->gp_refcnt < 2 && SvREFCNT(cv) < 2) {
6458 SvANY(cv)->xcv_gv_u.xcv_gv = NULL;
6462 /* if not, anonymise: */
6463 gvname = (GvSTASH(gv) && HvNAME(GvSTASH(gv)) && HvENAME(GvSTASH(gv)))
6464 ? newSVhek(HvENAME_HEK(GvSTASH(gv)))
6465 : newSVpvn_flags( "__ANON__", 8, 0 );
6466 sv_catpvs(gvname, "::__ANON__");
6467 anongv = gv_fetchsv(gvname, GV_ADDMULTI, SVt_PVCV);
6468 SvREFCNT_dec_NN(gvname);
6472 SvANY(cv)->xcv_gv_u.xcv_gv = MUTABLE_GV(SvREFCNT_inc(anongv));
6477 =for apidoc sv_clear
6479 Clear an SV: call any destructors, free up any memory used by the body,
6480 and free the body itself. The SV's head is I<not> freed, although
6481 its type is set to all 1's so that it won't inadvertently be assumed
6482 to be live during global destruction etc.
6483 This function should only be called when C<REFCNT> is zero. Most of the time
6484 you'll want to call C<sv_free()> (or its macro wrapper C<SvREFCNT_dec>)
6491 Perl_sv_clear(pTHX_ SV *const orig_sv)
6496 const struct body_details *sv_type_details;
6500 STRLEN hash_index = 0; /* initialise to make Coverity et al happy.
6501 Not strictly necessary */
6503 PERL_ARGS_ASSERT_SV_CLEAR;
6505 /* within this loop, sv is the SV currently being freed, and
6506 * iter_sv is the most recent AV or whatever that's being iterated
6507 * over to provide more SVs */
6513 assert(SvREFCNT(sv) == 0);
6514 assert(SvTYPE(sv) != (svtype)SVTYPEMASK);
6516 if (type <= SVt_IV) {
6517 /* See the comment in sv.h about the collusion between this
6518 * early return and the overloading of the NULL slots in the
6522 SvFLAGS(sv) &= SVf_BREAK;
6523 SvFLAGS(sv) |= SVTYPEMASK;
6527 /* objs are always >= MG, but pad names use the SVs_OBJECT flag
6528 for another purpose */
6529 assert(!SvOBJECT(sv) || type >= SVt_PVMG);
6531 if (type >= SVt_PVMG) {
6533 if (!curse(sv, 1)) goto get_next_sv;
6534 type = SvTYPE(sv); /* destructor may have changed it */
6536 /* Free back-references before magic, in case the magic calls
6537 * Perl code that has weak references to sv. */
6538 if (type == SVt_PVHV) {
6539 Perl_hv_kill_backrefs(aTHX_ MUTABLE_HV(sv));
6543 else if (SvMAGIC(sv)) {
6544 /* Free back-references before other types of magic. */
6545 sv_unmagic(sv, PERL_MAGIC_backref);
6551 /* case SVt_INVLIST: */
6554 IoIFP(sv) != PerlIO_stdin() &&
6555 IoIFP(sv) != PerlIO_stdout() &&
6556 IoIFP(sv) != PerlIO_stderr() &&
6557 !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6559 io_close(MUTABLE_IO(sv), NULL, FALSE,
6560 (IoTYPE(sv) == IoTYPE_WRONLY ||
6561 IoTYPE(sv) == IoTYPE_RDWR ||
6562 IoTYPE(sv) == IoTYPE_APPEND));
6564 if (IoDIRP(sv) && !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6565 PerlDir_close(IoDIRP(sv));
6566 IoDIRP(sv) = (DIR*)NULL;
6567 Safefree(IoTOP_NAME(sv));
6568 Safefree(IoFMT_NAME(sv));
6569 Safefree(IoBOTTOM_NAME(sv));
6570 if ((const GV *)sv == PL_statgv)
6574 /* FIXME for plugins */
6575 pregfree2((REGEXP*) sv);
6579 cv_undef(MUTABLE_CV(sv));
6580 /* If we're in a stash, we don't own a reference to it.
6581 * However it does have a back reference to us, which needs to
6583 if ((stash = CvSTASH(sv)))
6584 sv_del_backref(MUTABLE_SV(stash), sv);
6587 if (PL_last_swash_hv == (const HV *)sv) {
6588 PL_last_swash_hv = NULL;
6590 if (HvTOTALKEYS((HV*)sv) > 0) {
6592 /* this statement should match the one at the beginning of
6593 * hv_undef_flags() */
6594 if ( PL_phase != PERL_PHASE_DESTRUCT
6595 && (hek = HvNAME_HEK((HV*)sv)))
6597 if (PL_stashcache) {
6598 DEBUG_o(Perl_deb(aTHX_
6599 "sv_clear clearing PL_stashcache for '%" HEKf
6602 (void)hv_deletehek(PL_stashcache,
6605 hv_name_set((HV*)sv, NULL, 0, 0);
6608 /* save old iter_sv in unused SvSTASH field */
6609 assert(!SvOBJECT(sv));
6610 SvSTASH(sv) = (HV*)iter_sv;
6613 /* save old hash_index in unused SvMAGIC field */
6614 assert(!SvMAGICAL(sv));
6615 assert(!SvMAGIC(sv));
6616 ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index = hash_index;
6619 next_sv = Perl_hfree_next_entry(aTHX_ (HV*)sv, &hash_index);
6620 goto get_next_sv; /* process this new sv */
6622 /* free empty hash */
6623 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6624 assert(!HvARRAY((HV*)sv));
6628 AV* av = MUTABLE_AV(sv);
6629 if (PL_comppad == av) {
6633 if (AvREAL(av) && AvFILLp(av) > -1) {
6634 next_sv = AvARRAY(av)[AvFILLp(av)--];
6635 /* save old iter_sv in top-most slot of AV,
6636 * and pray that it doesn't get wiped in the meantime */
6637 AvARRAY(av)[AvMAX(av)] = iter_sv;
6639 goto get_next_sv; /* process this new sv */
6641 Safefree(AvALLOC(av));
6646 if (LvTYPE(sv) == 'T') { /* for tie: return HE to pool */
6647 SvREFCNT_dec(HeKEY_sv((HE*)LvTARG(sv)));
6648 HeNEXT((HE*)LvTARG(sv)) = PL_hv_fetch_ent_mh;
6649 PL_hv_fetch_ent_mh = (HE*)LvTARG(sv);
6651 else if (LvTYPE(sv) != 't') /* unless tie: unrefcnted fake SV** */
6652 SvREFCNT_dec(LvTARG(sv));
6654 /* SvLEN points to a regex body. Free the body, then
6655 * set SvLEN to whatever value was in the now-freed
6656 * regex body. The PVX buffer is shared by multiple re's
6657 * and only freed once, by the re whose len in non-null */
6658 STRLEN len = ReANY(sv)->xpv_len;
6659 pregfree2((REGEXP*) sv);
6660 SvLEN_set((sv), len);
6665 if (isGV_with_GP(sv)) {
6666 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
6667 && HvENAME_get(stash))
6668 mro_method_changed_in(stash);
6669 gp_free(MUTABLE_GV(sv));
6671 unshare_hek(GvNAME_HEK(sv));
6672 /* If we're in a stash, we don't own a reference to it.
6673 * However it does have a back reference to us, which
6674 * needs to be cleared. */
6675 if ((stash = GvSTASH(sv)))
6676 sv_del_backref(MUTABLE_SV(stash), sv);
6678 /* FIXME. There are probably more unreferenced pointers to SVs
6679 * in the interpreter struct that we should check and tidy in
6680 * a similar fashion to this: */
6681 /* See also S_sv_unglob, which does the same thing. */
6682 if ((const GV *)sv == PL_last_in_gv)
6683 PL_last_in_gv = NULL;
6684 else if ((const GV *)sv == PL_statgv)
6686 else if ((const GV *)sv == PL_stderrgv)
6695 /* Don't bother with SvOOK_off(sv); as we're only going to
6699 SvOOK_offset(sv, offset);
6700 SvPV_set(sv, SvPVX_mutable(sv) - offset);
6701 /* Don't even bother with turning off the OOK flag. */
6706 SV * const target = SvRV(sv);
6708 sv_del_backref(target, sv);
6714 else if (SvPVX_const(sv)
6715 && !(SvTYPE(sv) == SVt_PVIO
6716 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6721 PerlIO_printf(Perl_debug_log, "Copy on write: clear\n");
6726 if (CowREFCNT(sv)) {
6733 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6738 Safefree(SvPVX_mutable(sv));
6742 else if (SvPVX_const(sv) && SvLEN(sv)
6743 && !(SvTYPE(sv) == SVt_PVIO
6744 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6745 Safefree(SvPVX_mutable(sv));
6746 else if (SvPVX_const(sv) && SvIsCOW(sv)) {
6747 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6757 SvFLAGS(sv) &= SVf_BREAK;
6758 SvFLAGS(sv) |= SVTYPEMASK;
6760 sv_type_details = bodies_by_type + type;
6761 if (sv_type_details->arena) {
6762 del_body(((char *)SvANY(sv) + sv_type_details->offset),
6763 &PL_body_roots[type]);
6765 else if (sv_type_details->body_size) {
6766 safefree(SvANY(sv));
6770 /* caller is responsible for freeing the head of the original sv */
6771 if (sv != orig_sv && !SvREFCNT(sv))
6774 /* grab and free next sv, if any */
6782 else if (!iter_sv) {
6784 } else if (SvTYPE(iter_sv) == SVt_PVAV) {
6785 AV *const av = (AV*)iter_sv;
6786 if (AvFILLp(av) > -1) {
6787 sv = AvARRAY(av)[AvFILLp(av)--];
6789 else { /* no more elements of current AV to free */
6792 /* restore previous value, squirrelled away */
6793 iter_sv = AvARRAY(av)[AvMAX(av)];
6794 Safefree(AvALLOC(av));
6797 } else if (SvTYPE(iter_sv) == SVt_PVHV) {
6798 sv = Perl_hfree_next_entry(aTHX_ (HV*)iter_sv, &hash_index);
6799 if (!sv && !HvTOTALKEYS((HV *)iter_sv)) {
6800 /* no more elements of current HV to free */
6803 /* Restore previous values of iter_sv and hash_index,
6804 * squirrelled away */
6805 assert(!SvOBJECT(sv));
6806 iter_sv = (SV*)SvSTASH(sv);
6807 assert(!SvMAGICAL(sv));
6808 hash_index = ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index;
6810 /* perl -DA does not like rubbish in SvMAGIC. */
6814 /* free any remaining detritus from the hash struct */
6815 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6816 assert(!HvARRAY((HV*)sv));
6821 /* unrolled SvREFCNT_dec and sv_free2 follows: */
6825 if (!SvREFCNT(sv)) {
6829 if (--(SvREFCNT(sv)))
6833 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6834 "Attempt to free temp prematurely: SV 0x%" UVxf
6835 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6839 if (SvIMMORTAL(sv)) {
6840 /* make sure SvREFCNT(sv)==0 happens very seldom */
6841 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6850 /* This routine curses the sv itself, not the object referenced by sv. So
6851 sv does not have to be ROK. */
6854 S_curse(pTHX_ SV * const sv, const bool check_refcnt) {
6855 PERL_ARGS_ASSERT_CURSE;
6856 assert(SvOBJECT(sv));
6858 if (PL_defstash && /* Still have a symbol table? */
6864 stash = SvSTASH(sv);
6865 assert(SvTYPE(stash) == SVt_PVHV);
6866 if (HvNAME(stash)) {
6867 CV* destructor = NULL;
6868 struct mro_meta *meta;
6870 assert (SvOOK(stash));
6872 DEBUG_o( Perl_deb(aTHX_ "Looking for DESTROY method for %s\n",
6875 /* don't make this an initialization above the assert, since it needs
6877 meta = HvMROMETA(stash);
6878 if (meta->destroy_gen && meta->destroy_gen == PL_sub_generation) {
6879 destructor = meta->destroy;
6880 DEBUG_o( Perl_deb(aTHX_ "Using cached DESTROY method %p for %s\n",
6881 (void *)destructor, HvNAME(stash)) );
6884 bool autoload = FALSE;
6886 gv_fetchmeth_pvn(stash, S_destroy, S_destroy_len, -1, 0);
6888 destructor = GvCV(gv);
6890 gv = gv_autoload_pvn(stash, S_destroy, S_destroy_len,
6891 GV_AUTOLOAD_ISMETHOD);
6893 destructor = GvCV(gv);
6897 /* we don't cache AUTOLOAD for DESTROY, since this code
6898 would then need to set $__PACKAGE__::AUTOLOAD, or the
6899 equivalent for XS AUTOLOADs */
6901 meta->destroy_gen = PL_sub_generation;
6902 meta->destroy = destructor;
6904 DEBUG_o( Perl_deb(aTHX_ "Set cached DESTROY method %p for %s\n",
6905 (void *)destructor, HvNAME(stash)) );
6908 DEBUG_o( Perl_deb(aTHX_ "Not caching AUTOLOAD for DESTROY method for %s\n",
6912 assert(!destructor || SvTYPE(destructor) == SVt_PVCV);
6914 /* A constant subroutine can have no side effects, so
6915 don't bother calling it. */
6916 && !CvCONST(destructor)
6917 /* Don't bother calling an empty destructor or one that
6918 returns immediately. */
6919 && (CvISXSUB(destructor)
6920 || (CvSTART(destructor)
6921 && (CvSTART(destructor)->op_next->op_type
6923 && (CvSTART(destructor)->op_next->op_type
6925 || CvSTART(destructor)->op_next->op_next->op_type
6931 SV* const tmpref = newRV(sv);
6932 SvREADONLY_on(tmpref); /* DESTROY() could be naughty */
6934 PUSHSTACKi(PERLSI_DESTROY);
6939 call_sv(MUTABLE_SV(destructor),
6940 G_DISCARD|G_EVAL|G_KEEPERR|G_VOID);
6944 if(SvREFCNT(tmpref) < 2) {
6945 /* tmpref is not kept alive! */
6947 SvRV_set(tmpref, NULL);
6950 SvREFCNT_dec_NN(tmpref);
6953 } while (SvOBJECT(sv) && SvSTASH(sv) != stash);
6956 if (check_refcnt && SvREFCNT(sv)) {
6957 if (PL_in_clean_objs)
6959 "DESTROY created new reference to dead object '%" HEKf "'",
6960 HEKfARG(HvNAME_HEK(stash)));
6961 /* DESTROY gave object new lease on life */
6967 HV * const stash = SvSTASH(sv);
6968 /* Curse before freeing the stash, as freeing the stash could cause
6969 a recursive call into S_curse. */
6970 SvOBJECT_off(sv); /* Curse the object. */
6971 SvSTASH_set(sv,0); /* SvREFCNT_dec may try to read this */
6972 SvREFCNT_dec(stash); /* possibly of changed persuasion */
6978 =for apidoc sv_newref
6980 Increment an SV's reference count. Use the C<SvREFCNT_inc()> wrapper
6987 Perl_sv_newref(pTHX_ SV *const sv)
6989 PERL_UNUSED_CONTEXT;
6998 Decrement an SV's reference count, and if it drops to zero, call
6999 C<sv_clear> to invoke destructors and free up any memory used by
7000 the body; finally, deallocating the SV's head itself.
7001 Normally called via a wrapper macro C<SvREFCNT_dec>.
7007 Perl_sv_free(pTHX_ SV *const sv)
7013 /* Private helper function for SvREFCNT_dec().
7014 * Called with rc set to original SvREFCNT(sv), where rc == 0 or 1 */
7017 Perl_sv_free2(pTHX_ SV *const sv, const U32 rc)
7021 PERL_ARGS_ASSERT_SV_FREE2;
7023 if (LIKELY( rc == 1 )) {
7029 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
7030 "Attempt to free temp prematurely: SV 0x%" UVxf
7031 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
7035 if (SvIMMORTAL(sv)) {
7036 /* make sure SvREFCNT(sv)==0 happens very seldom */
7037 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
7041 if (! SvREFCNT(sv)) /* may have have been resurrected */
7046 /* handle exceptional cases */
7050 if (SvFLAGS(sv) & SVf_BREAK)
7051 /* this SV's refcnt has been artificially decremented to
7052 * trigger cleanup */
7054 if (PL_in_clean_all) /* All is fair */
7056 if (SvIMMORTAL(sv)) {
7057 /* make sure SvREFCNT(sv)==0 happens very seldom */
7058 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
7061 if (ckWARN_d(WARN_INTERNAL)) {
7062 #ifdef DEBUG_LEAKING_SCALARS_FORK_DUMP
7063 Perl_dump_sv_child(aTHX_ sv);
7065 #ifdef DEBUG_LEAKING_SCALARS
7068 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7069 if (PL_warnhook == PERL_WARNHOOK_FATAL
7070 || ckDEAD(packWARN(WARN_INTERNAL))) {
7071 /* Don't let Perl_warner cause us to escape our fate: */
7075 /* This may not return: */
7076 Perl_warner(aTHX_ packWARN(WARN_INTERNAL),
7077 "Attempt to free unreferenced scalar: SV 0x%" UVxf
7078 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
7081 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7091 Returns the length of the string in the SV. Handles magic and type
7092 coercion and sets the UTF8 flag appropriately. See also C<L</SvCUR>>, which
7093 gives raw access to the C<xpv_cur> slot.
7099 Perl_sv_len(pTHX_ SV *const sv)
7106 (void)SvPV_const(sv, len);
7111 =for apidoc sv_len_utf8
7113 Returns the number of characters in the string in an SV, counting wide
7114 UTF-8 bytes as a single character. Handles magic and type coercion.
7120 * The length is cached in PERL_MAGIC_utf8, in the mg_len field. Also the
7121 * mg_ptr is used, by sv_pos_u2b() and sv_pos_b2u() - see the comments below.
7122 * (Note that the mg_len is not the length of the mg_ptr field.
7123 * This allows the cache to store the character length of the string without
7124 * needing to malloc() extra storage to attach to the mg_ptr.)
7129 Perl_sv_len_utf8(pTHX_ SV *const sv)
7135 return sv_len_utf8_nomg(sv);
7139 Perl_sv_len_utf8_nomg(pTHX_ SV * const sv)
7142 const U8 *s = (U8*)SvPV_nomg_const(sv, len);
7144 PERL_ARGS_ASSERT_SV_LEN_UTF8_NOMG;
7146 if (PL_utf8cache && SvUTF8(sv)) {
7148 MAGIC *mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_utf8) : NULL;
7150 if (mg && (mg->mg_len != -1 || mg->mg_ptr)) {
7151 if (mg->mg_len != -1)
7154 /* We can use the offset cache for a headstart.
7155 The longer value is stored in the first pair. */
7156 STRLEN *cache = (STRLEN *) mg->mg_ptr;
7158 ulen = cache[0] + Perl_utf8_length(aTHX_ s + cache[1],
7162 if (PL_utf8cache < 0) {
7163 const STRLEN real = Perl_utf8_length(aTHX_ s, s + len);
7164 assert_uft8_cache_coherent("sv_len_utf8", ulen, real, sv);
7168 ulen = Perl_utf8_length(aTHX_ s, s + len);
7169 utf8_mg_len_cache_update(sv, &mg, ulen);
7173 return SvUTF8(sv) ? Perl_utf8_length(aTHX_ s, s + len) : len;
7176 /* Walk forwards to find the byte corresponding to the passed in UTF-8
7179 S_sv_pos_u2b_forwards(const U8 *const start, const U8 *const send,
7180 STRLEN *const uoffset_p, bool *const at_end)
7182 const U8 *s = start;
7183 STRLEN uoffset = *uoffset_p;
7185 PERL_ARGS_ASSERT_SV_POS_U2B_FORWARDS;
7187 while (s < send && uoffset) {
7194 else if (s > send) {
7196 /* This is the existing behaviour. Possibly it should be a croak, as
7197 it's actually a bounds error */
7200 *uoffset_p -= uoffset;
7204 /* Given the length of the string in both bytes and UTF-8 characters, decide
7205 whether to walk forwards or backwards to find the byte corresponding to
7206 the passed in UTF-8 offset. */
7208 S_sv_pos_u2b_midway(const U8 *const start, const U8 *send,
7209 STRLEN uoffset, const STRLEN uend)
7211 STRLEN backw = uend - uoffset;
7213 PERL_ARGS_ASSERT_SV_POS_U2B_MIDWAY;
7215 if (uoffset < 2 * backw) {
7216 /* The assumption is that going forwards is twice the speed of going
7217 forward (that's where the 2 * backw comes from).
7218 (The real figure of course depends on the UTF-8 data.) */
7219 const U8 *s = start;
7221 while (s < send && uoffset--)
7231 while (UTF8_IS_CONTINUATION(*send))
7234 return send - start;
7237 /* For the string representation of the given scalar, find the byte
7238 corresponding to the passed in UTF-8 offset. uoffset0 and boffset0
7239 give another position in the string, *before* the sought offset, which
7240 (which is always true, as 0, 0 is a valid pair of positions), which should
7241 help reduce the amount of linear searching.
7242 If *mgp is non-NULL, it should point to the UTF-8 cache magic, which
7243 will be used to reduce the amount of linear searching. The cache will be
7244 created if necessary, and the found value offered to it for update. */
7246 S_sv_pos_u2b_cached(pTHX_ SV *const sv, MAGIC **const mgp, const U8 *const start,
7247 const U8 *const send, STRLEN uoffset,
7248 STRLEN uoffset0, STRLEN boffset0)
7250 STRLEN boffset = 0; /* Actually always set, but let's keep gcc happy. */
7252 bool at_end = FALSE;
7254 PERL_ARGS_ASSERT_SV_POS_U2B_CACHED;
7256 assert (uoffset >= uoffset0);
7261 if (!SvREADONLY(sv) && !SvGMAGICAL(sv) && SvPOK(sv)
7263 && (*mgp || (SvTYPE(sv) >= SVt_PVMG &&
7264 (*mgp = mg_find(sv, PERL_MAGIC_utf8))))) {
7265 if ((*mgp)->mg_ptr) {
7266 STRLEN *cache = (STRLEN *) (*mgp)->mg_ptr;
7267 if (cache[0] == uoffset) {
7268 /* An exact match. */
7271 if (cache[2] == uoffset) {
7272 /* An exact match. */
7276 if (cache[0] < uoffset) {
7277 /* The cache already knows part of the way. */
7278 if (cache[0] > uoffset0) {
7279 /* The cache knows more than the passed in pair */
7280 uoffset0 = cache[0];
7281 boffset0 = cache[1];
7283 if ((*mgp)->mg_len != -1) {
7284 /* And we know the end too. */
7286 + sv_pos_u2b_midway(start + boffset0, send,
7288 (*mgp)->mg_len - uoffset0);
7290 uoffset -= uoffset0;
7292 + sv_pos_u2b_forwards(start + boffset0,
7293 send, &uoffset, &at_end);
7294 uoffset += uoffset0;
7297 else if (cache[2] < uoffset) {
7298 /* We're between the two cache entries. */
7299 if (cache[2] > uoffset0) {
7300 /* and the cache knows more than the passed in pair */
7301 uoffset0 = cache[2];
7302 boffset0 = cache[3];
7306 + sv_pos_u2b_midway(start + boffset0,
7309 cache[0] - uoffset0);
7312 + sv_pos_u2b_midway(start + boffset0,
7315 cache[2] - uoffset0);
7319 else if ((*mgp)->mg_len != -1) {
7320 /* If we can take advantage of a passed in offset, do so. */
7321 /* In fact, offset0 is either 0, or less than offset, so don't
7322 need to worry about the other possibility. */
7324 + sv_pos_u2b_midway(start + boffset0, send,
7326 (*mgp)->mg_len - uoffset0);
7331 if (!found || PL_utf8cache < 0) {
7332 STRLEN real_boffset;
7333 uoffset -= uoffset0;
7334 real_boffset = boffset0 + sv_pos_u2b_forwards(start + boffset0,
7335 send, &uoffset, &at_end);
7336 uoffset += uoffset0;
7338 if (found && PL_utf8cache < 0)
7339 assert_uft8_cache_coherent("sv_pos_u2b_cache", boffset,
7341 boffset = real_boffset;
7344 if (PL_utf8cache && !SvGMAGICAL(sv) && SvPOK(sv)) {
7346 utf8_mg_len_cache_update(sv, mgp, uoffset);
7348 utf8_mg_pos_cache_update(sv, mgp, boffset, uoffset, send - start);
7355 =for apidoc sv_pos_u2b_flags
7357 Converts the offset from a count of UTF-8 chars from
7358 the start of the string, to a count of the equivalent number of bytes; if
7359 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7360 C<offset>, rather than from the start
7361 of the string. Handles type coercion.
7362 C<flags> is passed to C<SvPV_flags>, and usually should be
7363 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7369 * sv_pos_u2b_flags() uses, like sv_pos_b2u(), the mg_ptr of the potential
7370 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7371 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7376 Perl_sv_pos_u2b_flags(pTHX_ SV *const sv, STRLEN uoffset, STRLEN *const lenp,
7383 PERL_ARGS_ASSERT_SV_POS_U2B_FLAGS;
7385 start = (U8*)SvPV_flags(sv, len, flags);
7387 const U8 * const send = start + len;
7389 boffset = sv_pos_u2b_cached(sv, &mg, start, send, uoffset, 0, 0);
7392 && *lenp /* don't bother doing work for 0, as its bytes equivalent
7393 is 0, and *lenp is already set to that. */) {
7394 /* Convert the relative offset to absolute. */
7395 const STRLEN uoffset2 = uoffset + *lenp;
7396 const STRLEN boffset2
7397 = sv_pos_u2b_cached(sv, &mg, start, send, uoffset2,
7398 uoffset, boffset) - boffset;
7412 =for apidoc sv_pos_u2b
7414 Converts the value pointed to by C<offsetp> from a count of UTF-8 chars from
7415 the start of the string, to a count of the equivalent number of bytes; if
7416 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7417 the offset, rather than from the start of the string. Handles magic and
7420 Use C<sv_pos_u2b_flags> in preference, which correctly handles strings longer
7427 * sv_pos_u2b() uses, like sv_pos_b2u(), the mg_ptr of the potential
7428 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7429 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7433 /* This function is subject to size and sign problems */
7436 Perl_sv_pos_u2b(pTHX_ SV *const sv, I32 *const offsetp, I32 *const lenp)
7438 PERL_ARGS_ASSERT_SV_POS_U2B;
7441 STRLEN ulen = (STRLEN)*lenp;
7442 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, &ulen,
7443 SV_GMAGIC|SV_CONST_RETURN);
7446 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, NULL,
7447 SV_GMAGIC|SV_CONST_RETURN);
7452 S_utf8_mg_len_cache_update(pTHX_ SV *const sv, MAGIC **const mgp,
7455 PERL_ARGS_ASSERT_UTF8_MG_LEN_CACHE_UPDATE;
7456 if (SvREADONLY(sv) || SvGMAGICAL(sv) || !SvPOK(sv))
7459 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7460 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7461 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, &PL_vtbl_utf8, 0, 0);
7465 (*mgp)->mg_len = ulen;
7468 /* Create and update the UTF8 magic offset cache, with the proffered utf8/
7469 byte length pairing. The (byte) length of the total SV is passed in too,
7470 as blen, because for some (more esoteric) SVs, the call to SvPV_const()
7471 may not have updated SvCUR, so we can't rely on reading it directly.
7473 The proffered utf8/byte length pairing isn't used if the cache already has
7474 two pairs, and swapping either for the proffered pair would increase the
7475 RMS of the intervals between known byte offsets.
7477 The cache itself consists of 4 STRLEN values
7478 0: larger UTF-8 offset
7479 1: corresponding byte offset
7480 2: smaller UTF-8 offset
7481 3: corresponding byte offset
7483 Unused cache pairs have the value 0, 0.
7484 Keeping the cache "backwards" means that the invariant of
7485 cache[0] >= cache[2] is maintained even with empty slots, which means that
7486 the code that uses it doesn't need to worry if only 1 entry has actually
7487 been set to non-zero. It also makes the "position beyond the end of the
7488 cache" logic much simpler, as the first slot is always the one to start
7492 S_utf8_mg_pos_cache_update(pTHX_ SV *const sv, MAGIC **const mgp, const STRLEN byte,
7493 const STRLEN utf8, const STRLEN blen)
7497 PERL_ARGS_ASSERT_UTF8_MG_POS_CACHE_UPDATE;
7502 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7503 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7504 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, (MGVTBL*)&PL_vtbl_utf8, 0,
7506 (*mgp)->mg_len = -1;
7510 if (!(cache = (STRLEN *)(*mgp)->mg_ptr)) {
7511 Newxz(cache, PERL_MAGIC_UTF8_CACHESIZE * 2, STRLEN);
7512 (*mgp)->mg_ptr = (char *) cache;
7516 if (PL_utf8cache < 0 && SvPOKp(sv)) {
7517 /* SvPOKp() because, if sv is a reference, then SvPVX() is actually
7518 a pointer. Note that we no longer cache utf8 offsets on refer-
7519 ences, but this check is still a good idea, for robustness. */
7520 const U8 *start = (const U8 *) SvPVX_const(sv);
7521 const STRLEN realutf8 = utf8_length(start, start + byte);
7523 assert_uft8_cache_coherent("utf8_mg_pos_cache_update", utf8, realutf8,
7527 /* Cache is held with the later position first, to simplify the code
7528 that deals with unbounded ends. */
7530 ASSERT_UTF8_CACHE(cache);
7531 if (cache[1] == 0) {
7532 /* Cache is totally empty */
7535 } else if (cache[3] == 0) {
7536 if (byte > cache[1]) {
7537 /* New one is larger, so goes first. */
7538 cache[2] = cache[0];
7539 cache[3] = cache[1];
7547 /* float casts necessary? XXX */
7548 #define THREEWAY_SQUARE(a,b,c,d) \
7549 ((float)((d) - (c))) * ((float)((d) - (c))) \
7550 + ((float)((c) - (b))) * ((float)((c) - (b))) \
7551 + ((float)((b) - (a))) * ((float)((b) - (a)))
7553 /* Cache has 2 slots in use, and we know three potential pairs.
7554 Keep the two that give the lowest RMS distance. Do the
7555 calculation in bytes simply because we always know the byte
7556 length. squareroot has the same ordering as the positive value,
7557 so don't bother with the actual square root. */
7558 if (byte > cache[1]) {
7559 /* New position is after the existing pair of pairs. */
7560 const float keep_earlier
7561 = THREEWAY_SQUARE(0, cache[3], byte, blen);
7562 const float keep_later
7563 = THREEWAY_SQUARE(0, cache[1], byte, blen);
7565 if (keep_later < keep_earlier) {
7566 cache[2] = cache[0];
7567 cache[3] = cache[1];
7573 const float keep_later = THREEWAY_SQUARE(0, byte, cache[1], blen);
7574 float b, c, keep_earlier;
7575 if (byte > cache[3]) {
7576 /* New position is between the existing pair of pairs. */
7577 b = (float)cache[3];
7580 /* New position is before the existing pair of pairs. */
7582 c = (float)cache[3];
7584 keep_earlier = THREEWAY_SQUARE(0, b, c, blen);
7585 if (byte > cache[3]) {
7586 if (keep_later < keep_earlier) {
7596 if (! (keep_later < keep_earlier)) {
7597 cache[0] = cache[2];
7598 cache[1] = cache[3];
7605 ASSERT_UTF8_CACHE(cache);
7608 /* We already know all of the way, now we may be able to walk back. The same
7609 assumption is made as in S_sv_pos_u2b_midway(), namely that walking
7610 backward is half the speed of walking forward. */
7612 S_sv_pos_b2u_midway(pTHX_ const U8 *const s, const U8 *const target,
7613 const U8 *end, STRLEN endu)
7615 const STRLEN forw = target - s;
7616 STRLEN backw = end - target;
7618 PERL_ARGS_ASSERT_SV_POS_B2U_MIDWAY;
7620 if (forw < 2 * backw) {
7621 return utf8_length(s, target);
7624 while (end > target) {
7626 while (UTF8_IS_CONTINUATION(*end)) {
7635 =for apidoc sv_pos_b2u_flags
7637 Converts C<offset> from a count of bytes from the start of the string, to
7638 a count of the equivalent number of UTF-8 chars. Handles type coercion.
7639 C<flags> is passed to C<SvPV_flags>, and usually should be
7640 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7646 * sv_pos_b2u_flags() uses, like sv_pos_u2b_flags(), the mg_ptr of the
7647 * potential PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8
7652 Perl_sv_pos_b2u_flags(pTHX_ SV *const sv, STRLEN const offset, U32 flags)
7655 STRLEN len = 0; /* Actually always set, but let's keep gcc happy. */
7661 PERL_ARGS_ASSERT_SV_POS_B2U_FLAGS;
7663 s = (const U8*)SvPV_flags(sv, blen, flags);
7666 Perl_croak(aTHX_ "panic: sv_pos_b2u: bad byte offset, blen=%" UVuf
7667 ", byte=%" UVuf, (UV)blen, (UV)offset);
7673 && SvTYPE(sv) >= SVt_PVMG
7674 && (mg = mg_find(sv, PERL_MAGIC_utf8)))
7677 STRLEN * const cache = (STRLEN *) mg->mg_ptr;
7678 if (cache[1] == offset) {
7679 /* An exact match. */
7682 if (cache[3] == offset) {
7683 /* An exact match. */
7687 if (cache[1] < offset) {
7688 /* We already know part of the way. */
7689 if (mg->mg_len != -1) {
7690 /* Actually, we know the end too. */
7692 + S_sv_pos_b2u_midway(aTHX_ s + cache[1], send,
7693 s + blen, mg->mg_len - cache[0]);
7695 len = cache[0] + utf8_length(s + cache[1], send);
7698 else if (cache[3] < offset) {
7699 /* We're between the two cached pairs, so we do the calculation
7700 offset by the byte/utf-8 positions for the earlier pair,
7701 then add the utf-8 characters from the string start to
7703 len = S_sv_pos_b2u_midway(aTHX_ s + cache[3], send,
7704 s + cache[1], cache[0] - cache[2])
7708 else { /* cache[3] > offset */
7709 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + cache[3],
7713 ASSERT_UTF8_CACHE(cache);
7715 } else if (mg->mg_len != -1) {
7716 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + blen, mg->mg_len);
7720 if (!found || PL_utf8cache < 0) {
7721 const STRLEN real_len = utf8_length(s, send);
7723 if (found && PL_utf8cache < 0)
7724 assert_uft8_cache_coherent("sv_pos_b2u", len, real_len, sv);
7730 utf8_mg_len_cache_update(sv, &mg, len);
7732 utf8_mg_pos_cache_update(sv, &mg, offset, len, blen);
7739 =for apidoc sv_pos_b2u
7741 Converts the value pointed to by C<offsetp> from a count of bytes from the
7742 start of the string, to a count of the equivalent number of UTF-8 chars.
7743 Handles magic and type coercion.
7745 Use C<sv_pos_b2u_flags> in preference, which correctly handles strings
7752 * sv_pos_b2u() uses, like sv_pos_u2b(), the mg_ptr of the potential
7753 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7758 Perl_sv_pos_b2u(pTHX_ SV *const sv, I32 *const offsetp)
7760 PERL_ARGS_ASSERT_SV_POS_B2U;
7765 *offsetp = (I32)sv_pos_b2u_flags(sv, (STRLEN)*offsetp,
7766 SV_GMAGIC|SV_CONST_RETURN);
7770 S_assert_uft8_cache_coherent(pTHX_ const char *const func, STRLEN from_cache,
7771 STRLEN real, SV *const sv)
7773 PERL_ARGS_ASSERT_ASSERT_UFT8_CACHE_COHERENT;
7775 /* As this is debugging only code, save space by keeping this test here,
7776 rather than inlining it in all the callers. */
7777 if (from_cache == real)
7780 /* Need to turn the assertions off otherwise we may recurse infinitely
7781 while printing error messages. */
7782 SAVEI8(PL_utf8cache);
7784 Perl_croak(aTHX_ "panic: %s cache %" UVuf " real %" UVuf " for %" SVf,
7785 func, (UV) from_cache, (UV) real, SVfARG(sv));
7791 Returns a boolean indicating whether the strings in the two SVs are
7792 identical. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7793 coerce its args to strings if necessary.
7795 =for apidoc sv_eq_flags
7797 Returns a boolean indicating whether the strings in the two SVs are
7798 identical. Is UTF-8 and S<C<'use bytes'>> aware and coerces its args to strings
7799 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get-magic, too.
7805 Perl_sv_eq_flags(pTHX_ SV *sv1, SV *sv2, const U32 flags)
7817 /* if pv1 and pv2 are the same, second SvPV_const call may
7818 * invalidate pv1 (if we are handling magic), so we may need to
7820 if (sv1 == sv2 && flags & SV_GMAGIC
7821 && (SvTHINKFIRST(sv1) || SvGMAGICAL(sv1))) {
7822 pv1 = SvPV_const(sv1, cur1);
7823 sv1 = newSVpvn_flags(pv1, cur1, SVs_TEMP | SvUTF8(sv2));
7825 pv1 = SvPV_flags_const(sv1, cur1, flags);
7833 pv2 = SvPV_flags_const(sv2, cur2, flags);
7835 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7836 /* Differing utf8ness. */
7838 /* sv1 is the UTF-8 one */
7839 return bytes_cmp_utf8((const U8*)pv2, cur2,
7840 (const U8*)pv1, cur1) == 0;
7843 /* sv2 is the UTF-8 one */
7844 return bytes_cmp_utf8((const U8*)pv1, cur1,
7845 (const U8*)pv2, cur2) == 0;
7850 return (pv1 == pv2) || memEQ(pv1, pv2, cur1);
7858 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7859 string in C<sv1> is less than, equal to, or greater than the string in
7860 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7861 coerce its args to strings if necessary. See also C<L</sv_cmp_locale>>.
7863 =for apidoc sv_cmp_flags
7865 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7866 string in C<sv1> is less than, equal to, or greater than the string in
7867 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware and will coerce its args to strings
7868 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get magic. See
7869 also C<L</sv_cmp_locale_flags>>.
7875 Perl_sv_cmp(pTHX_ SV *const sv1, SV *const sv2)
7877 return sv_cmp_flags(sv1, sv2, SV_GMAGIC);
7881 Perl_sv_cmp_flags(pTHX_ SV *const sv1, SV *const sv2,
7885 const char *pv1, *pv2;
7887 SV *svrecode = NULL;
7894 pv1 = SvPV_flags_const(sv1, cur1, flags);
7901 pv2 = SvPV_flags_const(sv2, cur2, flags);
7903 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7904 /* Differing utf8ness. */
7906 const int retval = -bytes_cmp_utf8((const U8*)pv2, cur2,
7907 (const U8*)pv1, cur1);
7908 return retval ? retval < 0 ? -1 : +1 : 0;
7911 const int retval = bytes_cmp_utf8((const U8*)pv1, cur1,
7912 (const U8*)pv2, cur2);
7913 return retval ? retval < 0 ? -1 : +1 : 0;
7917 /* Here, if both are non-NULL, then they have the same UTF8ness. */
7920 cmp = cur2 ? -1 : 0;
7924 STRLEN shortest_len = cur1 < cur2 ? cur1 : cur2;
7927 if (! DO_UTF8(sv1)) {
7929 const I32 retval = memcmp((const void*)pv1,
7933 cmp = retval < 0 ? -1 : 1;
7934 } else if (cur1 == cur2) {
7937 cmp = cur1 < cur2 ? -1 : 1;
7941 else { /* Both are to be treated as UTF-EBCDIC */
7943 /* EBCDIC UTF-8 is complicated by the fact that it is based on I8
7944 * which remaps code points 0-255. We therefore generally have to
7945 * unmap back to the original values to get an accurate comparison.
7946 * But we don't have to do that for UTF-8 invariants, as by
7947 * definition, they aren't remapped, nor do we have to do it for
7948 * above-latin1 code points, as they also aren't remapped. (This
7949 * code also works on ASCII platforms, but the memcmp() above is
7952 const char *e = pv1 + shortest_len;
7954 /* Find the first bytes that differ between the two strings */
7955 while (pv1 < e && *pv1 == *pv2) {
7961 if (pv1 == e) { /* Are the same all the way to the end */
7965 cmp = cur1 < cur2 ? -1 : 1;
7968 else /* Here *pv1 and *pv2 are not equal, but all bytes earlier
7969 * in the strings were. The current bytes may or may not be
7970 * at the beginning of a character. But neither or both are
7971 * (or else earlier bytes would have been different). And
7972 * if we are in the middle of a character, the two
7973 * characters are comprised of the same number of bytes
7974 * (because in this case the start bytes are the same, and
7975 * the start bytes encode the character's length). */
7976 if (UTF8_IS_INVARIANT(*pv1))
7978 /* If both are invariants; can just compare directly */
7979 if (UTF8_IS_INVARIANT(*pv2)) {
7980 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
7982 else /* Since *pv1 is invariant, it is the whole character,
7983 which means it is at the beginning of a character.
7984 That means pv2 is also at the beginning of a
7985 character (see earlier comment). Since it isn't
7986 invariant, it must be a start byte. If it starts a
7987 character whose code point is above 255, that
7988 character is greater than any single-byte char, which
7990 if (UTF8_IS_ABOVE_LATIN1_START(*pv2))
7995 /* Here, pv2 points to a character composed of 2 bytes
7996 * whose code point is < 256. Get its code point and
7997 * compare with *pv1 */
7998 cmp = ((U8) *pv1 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
8003 else /* The code point starting at pv1 isn't a single byte */
8004 if (UTF8_IS_INVARIANT(*pv2))
8006 /* But here, the code point starting at *pv2 is a single byte,
8007 * and so *pv1 must begin a character, hence is a start byte.
8008 * If that character is above 255, it is larger than any
8009 * single-byte char, which *pv2 is */
8010 if (UTF8_IS_ABOVE_LATIN1_START(*pv1)) {
8014 /* Here, pv1 points to a character composed of 2 bytes
8015 * whose code point is < 256. Get its code point and
8016 * compare with the single byte character *pv2 */
8017 cmp = (EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1)) < (U8) *pv2)
8022 else /* Here, we've ruled out either *pv1 and *pv2 being
8023 invariant. That means both are part of variants, but not
8024 necessarily at the start of a character */
8025 if ( UTF8_IS_ABOVE_LATIN1_START(*pv1)
8026 || UTF8_IS_ABOVE_LATIN1_START(*pv2))
8028 /* Here, at least one is the start of a character, which means
8029 * the other is also a start byte. And the code point of at
8030 * least one of the characters is above 255. It is a
8031 * characteristic of UTF-EBCDIC that all start bytes for
8032 * above-latin1 code points are well behaved as far as code
8033 * point comparisons go, and all are larger than all other
8034 * start bytes, so the comparison with those is also well
8036 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8039 /* Here both *pv1 and *pv2 are part of variant characters.
8040 * They could be both continuations, or both start characters.
8041 * (One or both could even be an illegal start character (for
8042 * an overlong) which for the purposes of sorting we treat as
8044 if (UTF8_IS_CONTINUATION(*pv1)) {
8046 /* If they are continuations for code points above 255,
8047 * then comparing the current byte is sufficient, as there
8048 * is no remapping of these and so the comparison is
8049 * well-behaved. We determine if they are such
8050 * continuations by looking at the preceding byte. It
8051 * could be a start byte, from which we can tell if it is
8052 * for an above 255 code point. Or it could be a
8053 * continuation, which means the character occupies at
8054 * least 3 bytes, so must be above 255. */
8055 if ( UTF8_IS_CONTINUATION(*(pv2 - 1))
8056 || UTF8_IS_ABOVE_LATIN1_START(*(pv2 -1)))
8058 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8062 /* Here, the continuations are for code points below 256;
8063 * back up one to get to the start byte */
8068 /* We need to get the actual native code point of each of these
8069 * variants in order to compare them */
8070 cmp = ( EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1))
8071 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
8080 SvREFCNT_dec(svrecode);
8086 =for apidoc sv_cmp_locale
8088 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
8089 S<C<'use bytes'>> aware, handles get magic, and will coerce its args to strings
8090 if necessary. See also C<L</sv_cmp>>.
8092 =for apidoc sv_cmp_locale_flags
8094 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
8095 S<C<'use bytes'>> aware and will coerce its args to strings if necessary. If
8096 the flags contain C<SV_GMAGIC>, it handles get magic. See also
8097 C<L</sv_cmp_flags>>.
8103 Perl_sv_cmp_locale(pTHX_ SV *const sv1, SV *const sv2)
8105 return sv_cmp_locale_flags(sv1, sv2, SV_GMAGIC);
8109 Perl_sv_cmp_locale_flags(pTHX_ SV *const sv1, SV *const sv2,
8112 #ifdef USE_LOCALE_COLLATE
8118 if (PL_collation_standard)
8123 /* Revert to using raw compare if both operands exist, but either one
8124 * doesn't transform properly for collation */
8126 pv1 = sv_collxfrm_flags(sv1, &len1, flags);
8130 pv2 = sv_collxfrm_flags(sv2, &len2, flags);
8136 pv1 = sv1 ? sv_collxfrm_flags(sv1, &len1, flags) : (char *) NULL;
8137 pv2 = sv2 ? sv_collxfrm_flags(sv2, &len2, flags) : (char *) NULL;
8140 if (!pv1 || !len1) {
8151 retval = memcmp((void*)pv1, (void*)pv2, len1 < len2 ? len1 : len2);
8154 return retval < 0 ? -1 : 1;
8157 * When the result of collation is equality, that doesn't mean
8158 * that there are no differences -- some locales exclude some
8159 * characters from consideration. So to avoid false equalities,
8160 * we use the raw string as a tiebreaker.
8167 PERL_UNUSED_ARG(flags);
8168 #endif /* USE_LOCALE_COLLATE */
8170 return sv_cmp(sv1, sv2);
8174 #ifdef USE_LOCALE_COLLATE
8177 =for apidoc sv_collxfrm
8179 This calls C<sv_collxfrm_flags> with the SV_GMAGIC flag. See
8180 C<L</sv_collxfrm_flags>>.
8182 =for apidoc sv_collxfrm_flags
8184 Add Collate Transform magic to an SV if it doesn't already have it. If the
8185 flags contain C<SV_GMAGIC>, it handles get-magic.
8187 Any scalar variable may carry C<PERL_MAGIC_collxfrm> magic that contains the
8188 scalar data of the variable, but transformed to such a format that a normal
8189 memory comparison can be used to compare the data according to the locale
8196 Perl_sv_collxfrm_flags(pTHX_ SV *const sv, STRLEN *const nxp, const I32 flags)
8200 PERL_ARGS_ASSERT_SV_COLLXFRM_FLAGS;
8202 mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_collxfrm) : (MAGIC *) NULL;
8204 /* If we don't have collation magic on 'sv', or the locale has changed
8205 * since the last time we calculated it, get it and save it now */
8206 if (!mg || !mg->mg_ptr || *(U32*)mg->mg_ptr != PL_collation_ix) {
8211 /* Free the old space */
8213 Safefree(mg->mg_ptr);
8215 s = SvPV_flags_const(sv, len, flags);
8216 if ((xf = _mem_collxfrm(s, len, &xlen, cBOOL(SvUTF8(sv))))) {
8218 mg = sv_magicext(sv, 0, PERL_MAGIC_collxfrm, &PL_vtbl_collxfrm,
8233 if (mg && mg->mg_ptr) {
8235 return mg->mg_ptr + sizeof(PL_collation_ix);
8243 #endif /* USE_LOCALE_COLLATE */
8246 S_sv_gets_append_to_utf8(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8248 SV * const tsv = newSV(0);
8251 sv_gets(tsv, fp, 0);
8252 sv_utf8_upgrade_nomg(tsv);
8253 SvCUR_set(sv,append);
8256 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8260 S_sv_gets_read_record(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8263 const STRLEN recsize = SvUV(SvRV(PL_rs)); /* RsRECORD() guarantees > 0. */
8264 /* Grab the size of the record we're getting */
8265 char *buffer = SvGROW(sv, (STRLEN)(recsize + append + 1)) + append;
8272 /* With a true, record-oriented file on VMS, we need to use read directly
8273 * to ensure that we respect RMS record boundaries. The user is responsible
8274 * for providing a PL_rs value that corresponds to the FAB$W_MRS (maximum
8275 * record size) field. N.B. This is likely to produce invalid results on
8276 * varying-width character data when a record ends mid-character.
8278 fd = PerlIO_fileno(fp);
8280 && PerlLIO_fstat(fd, &st) == 0
8281 && (st.st_fab_rfm == FAB$C_VAR
8282 || st.st_fab_rfm == FAB$C_VFC
8283 || st.st_fab_rfm == FAB$C_FIX)) {
8285 bytesread = PerlLIO_read(fd, buffer, recsize);
8287 else /* in-memory file from PerlIO::Scalar
8288 * or not a record-oriented file
8292 bytesread = PerlIO_read(fp, buffer, recsize);
8294 /* At this point, the logic in sv_get() means that sv will
8295 be treated as utf-8 if the handle is utf8.
8297 if (PerlIO_isutf8(fp) && bytesread > 0) {
8298 char *bend = buffer + bytesread;
8299 char *bufp = buffer;
8300 size_t charcount = 0;
8301 bool charstart = TRUE;
8304 while (charcount < recsize) {
8305 /* count accumulated characters */
8306 while (bufp < bend) {
8308 skip = UTF8SKIP(bufp);
8310 if (bufp + skip > bend) {
8311 /* partial at the end */
8322 if (charcount < recsize) {
8324 STRLEN bufp_offset = bufp - buffer;
8325 SSize_t morebytesread;
8327 /* originally I read enough to fill any incomplete
8328 character and the first byte of the next
8329 character if needed, but if there's many
8330 multi-byte encoded characters we're going to be
8331 making a read call for every character beyond
8332 the original read size.
8334 So instead, read the rest of the character if
8335 any, and enough bytes to match at least the
8336 start bytes for each character we're going to
8340 readsize = recsize - charcount;
8342 readsize = skip - (bend - bufp) + recsize - charcount - 1;
8343 buffer = SvGROW(sv, append + bytesread + readsize + 1) + append;
8344 bend = buffer + bytesread;
8345 morebytesread = PerlIO_read(fp, bend, readsize);
8346 if (morebytesread <= 0) {
8347 /* we're done, if we still have incomplete
8348 characters the check code in sv_gets() will
8351 I'd originally considered doing
8352 PerlIO_ungetc() on all but the lead
8353 character of the incomplete character, but
8354 read() doesn't do that, so I don't.
8359 /* prepare to scan some more */
8360 bytesread += morebytesread;
8361 bend = buffer + bytesread;
8362 bufp = buffer + bufp_offset;
8370 SvCUR_set(sv, bytesread + append);
8371 buffer[bytesread] = '\0';
8372 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8378 Get a line from the filehandle and store it into the SV, optionally
8379 appending to the currently-stored string. If C<append> is not 0, the
8380 line is appended to the SV instead of overwriting it. C<append> should
8381 be set to the byte offset that the appended string should start at
8382 in the SV (typically, C<SvCUR(sv)> is a suitable choice).
8388 Perl_sv_gets(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8398 PERL_ARGS_ASSERT_SV_GETS;
8400 if (SvTHINKFIRST(sv))
8401 sv_force_normal_flags(sv, append ? 0 : SV_COW_DROP_PV);
8402 /* XXX. If you make this PVIV, then copy on write can copy scalars read
8404 However, perlbench says it's slower, because the existing swipe code
8405 is faster than copy on write.
8406 Swings and roundabouts. */
8407 SvUPGRADE(sv, SVt_PV);
8410 /* line is going to be appended to the existing buffer in the sv */
8411 if (PerlIO_isutf8(fp)) {
8413 sv_utf8_upgrade_nomg(sv);
8414 sv_pos_u2b(sv,&append,0);
8416 } else if (SvUTF8(sv)) {
8417 return S_sv_gets_append_to_utf8(aTHX_ sv, fp, append);
8423 /* not appending - "clear" the string by setting SvCUR to 0,
8424 * the pv is still avaiable. */
8427 if (PerlIO_isutf8(fp))
8430 if (IN_PERL_COMPILETIME) {
8431 /* we always read code in line mode */
8435 else if (RsSNARF(PL_rs)) {
8436 /* If it is a regular disk file use size from stat() as estimate
8437 of amount we are going to read -- may result in mallocing
8438 more memory than we really need if the layers below reduce
8439 the size we read (e.g. CRLF or a gzip layer).
8442 int fd = PerlIO_fileno(fp);
8443 if (fd >= 0 && (PerlLIO_fstat(fd, &st) == 0) && S_ISREG(st.st_mode)) {
8444 const Off_t offset = PerlIO_tell(fp);
8445 if (offset != (Off_t) -1 && st.st_size + append > offset) {
8446 #ifdef PERL_COPY_ON_WRITE
8447 /* Add an extra byte for the sake of copy-on-write's
8448 * buffer reference count. */
8449 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 2));
8451 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 1));
8458 else if (RsRECORD(PL_rs)) {
8459 return S_sv_gets_read_record(aTHX_ sv, fp, append);
8461 else if (RsPARA(PL_rs)) {
8467 /* Get $/ i.e. PL_rs into same encoding as stream wants */
8468 if (PerlIO_isutf8(fp)) {
8469 rsptr = SvPVutf8(PL_rs, rslen);
8472 if (SvUTF8(PL_rs)) {
8473 if (!sv_utf8_downgrade(PL_rs, TRUE)) {
8474 Perl_croak(aTHX_ "Wide character in $/");
8477 /* extract the raw pointer to the record separator */
8478 rsptr = SvPV_const(PL_rs, rslen);
8482 /* rslast is the last character in the record separator
8483 * note we don't use rslast except when rslen is true, so the
8484 * null assign is a placeholder. */
8485 rslast = rslen ? rsptr[rslen - 1] : '\0';
8487 if (rspara) { /* have to do this both before and after */
8488 do { /* to make sure file boundaries work right */
8491 i = PerlIO_getc(fp);
8495 PerlIO_ungetc(fp,i);
8501 /* See if we know enough about I/O mechanism to cheat it ! */
8503 /* This used to be #ifdef test - it is made run-time test for ease
8504 of abstracting out stdio interface. One call should be cheap
8505 enough here - and may even be a macro allowing compile
8509 if (PerlIO_fast_gets(fp)) {
8511 * We can do buffer based IO operations on this filehandle.
8513 * This means we can bypass a lot of subcalls and process
8514 * the buffer directly, it also means we know the upper bound
8515 * on the amount of data we might read of the current buffer
8516 * into our sv. Knowing this allows us to preallocate the pv
8517 * to be able to hold that maximum, which allows us to simplify
8518 * a lot of logic. */
8521 * We're going to steal some values from the stdio struct
8522 * and put EVERYTHING in the innermost loop into registers.
8524 STDCHAR *ptr; /* pointer into fp's read-ahead buffer */
8525 STRLEN bpx; /* length of the data in the target sv
8526 used to fix pointers after a SvGROW */
8527 I32 shortbuffered; /* If the pv buffer is shorter than the amount
8528 of data left in the read-ahead buffer.
8529 If 0 then the pv buffer can hold the full
8530 amount left, otherwise this is the amount it
8533 /* Here is some breathtakingly efficient cheating */
8535 /* When you read the following logic resist the urge to think
8536 * of record separators that are 1 byte long. They are an
8537 * uninteresting special (simple) case.
8539 * Instead think of record separators which are at least 2 bytes
8540 * long, and keep in mind that we need to deal with such
8541 * separators when they cross a read-ahead buffer boundary.
8543 * Also consider that we need to gracefully deal with separators
8544 * that may be longer than a single read ahead buffer.
8546 * Lastly do not forget we want to copy the delimiter as well. We
8547 * are copying all data in the file _up_to_and_including_ the separator
8550 * Now that you have all that in mind here is what is happening below:
8552 * 1. When we first enter the loop we do some memory book keeping to see
8553 * how much free space there is in the target SV. (This sub assumes that
8554 * it is operating on the same SV most of the time via $_ and that it is
8555 * going to be able to reuse the same pv buffer each call.) If there is
8556 * "enough" room then we set "shortbuffered" to how much space there is
8557 * and start reading forward.
8559 * 2. When we scan forward we copy from the read-ahead buffer to the target
8560 * SV's pv buffer. While we go we watch for the end of the read-ahead buffer,
8561 * and the end of the of pv, as well as for the "rslast", which is the last
8562 * char of the separator.
8564 * 3. When scanning forward if we see rslast then we jump backwards in *pv*
8565 * (which has a "complete" record up to the point we saw rslast) and check
8566 * it to see if it matches the separator. If it does we are done. If it doesn't
8567 * we continue on with the scan/copy.
8569 * 4. If we run out of read-ahead buffer (cnt goes to 0) then we have to get
8570 * the IO system to read the next buffer. We do this by doing a getc(), which
8571 * returns a single char read (or EOF), and prefills the buffer, and also
8572 * allows us to find out how full the buffer is. We use this information to
8573 * SvGROW() the sv to the size remaining in the buffer, after which we copy
8574 * the returned single char into the target sv, and then go back into scan
8577 * 5. If we run out of write-buffer then we SvGROW() it by the size of the
8578 * remaining space in the read-buffer.
8580 * Note that this code despite its twisty-turny nature is pretty darn slick.
8581 * It manages single byte separators, multi-byte cross boundary separators,
8582 * and cross-read-buffer separators cleanly and efficiently at the cost
8583 * of potentially greatly overallocating the target SV.
8589 /* get the number of bytes remaining in the read-ahead buffer
8590 * on first call on a given fp this will return 0.*/
8591 cnt = PerlIO_get_cnt(fp);
8593 /* make sure we have the room */
8594 if ((I32)(SvLEN(sv) - append) <= cnt + 1) {
8595 /* Not room for all of it
8596 if we are looking for a separator and room for some
8598 if (rslen && cnt > 80 && (I32)SvLEN(sv) > append) {
8599 /* just process what we have room for */
8600 shortbuffered = cnt - SvLEN(sv) + append + 1;
8601 cnt -= shortbuffered;
8604 /* ensure that the target sv has enough room to hold
8605 * the rest of the read-ahead buffer */
8607 /* remember that cnt can be negative */
8608 SvGROW(sv, (STRLEN)(append + (cnt <= 0 ? 2 : (cnt + 1))));
8612 /* we have enough room to hold the full buffer, lets scream */
8616 /* extract the pointer to sv's string buffer, offset by append as necessary */
8617 bp = (STDCHAR*)SvPVX_const(sv) + append; /* move these two too to registers */
8618 /* extract the point to the read-ahead buffer */
8619 ptr = (STDCHAR*)PerlIO_get_ptr(fp);
8621 /* some trace debug output */
8622 DEBUG_P(PerlIO_printf(Perl_debug_log,
8623 "Screamer: entering, ptr=%" UVuf ", cnt=%ld\n",PTR2UV(ptr),(long)cnt));
8624 DEBUG_P(PerlIO_printf(Perl_debug_log,
8625 "Screamer: entering: PerlIO * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%"
8627 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8628 PTR2UV(PerlIO_has_base(fp) ? PerlIO_get_base(fp) : 0)));
8632 /* if there is stuff left in the read-ahead buffer */
8634 /* if there is a separator */
8636 /* find next rslast */
8639 /* shortcut common case of blank line */
8641 if ((*bp++ = *ptr++) == rslast)
8642 goto thats_all_folks;
8644 p = (STDCHAR *)memchr(ptr, rslast, cnt);
8646 SSize_t got = p - ptr + 1;
8647 Copy(ptr, bp, got, STDCHAR);
8651 goto thats_all_folks;
8653 Copy(ptr, bp, cnt, STDCHAR);
8659 /* no separator, slurp the full buffer */
8660 Copy(ptr, bp, cnt, char); /* this | eat */
8661 bp += cnt; /* screams | dust */
8662 ptr += cnt; /* louder | sed :-) */
8664 assert (!shortbuffered);
8665 goto cannot_be_shortbuffered;
8669 if (shortbuffered) { /* oh well, must extend */
8670 /* we didnt have enough room to fit the line into the target buffer
8671 * so we must extend the target buffer and keep going */
8672 cnt = shortbuffered;
8674 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8676 /* extned the target sv's buffer so it can hold the full read-ahead buffer */
8677 SvGROW(sv, SvLEN(sv) + append + cnt + 2);
8678 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8682 cannot_be_shortbuffered:
8683 /* we need to refill the read-ahead buffer if possible */
8685 DEBUG_P(PerlIO_printf(Perl_debug_log,
8686 "Screamer: going to getc, ptr=%" UVuf ", cnt=%" IVdf "\n",
8687 PTR2UV(ptr),(IV)cnt));
8688 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* deregisterize cnt and ptr */
8690 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8691 "Screamer: pre: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf "\n",
8692 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8693 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8696 call PerlIO_getc() to let it prefill the lookahead buffer
8698 This used to call 'filbuf' in stdio form, but as that behaves like
8699 getc when cnt <= 0 we use PerlIO_getc here to avoid introducing
8700 another abstraction.
8702 Note we have to deal with the char in 'i' if we are not at EOF
8704 i = PerlIO_getc(fp); /* get more characters */
8706 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8707 "Screamer: post: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf "\n",
8708 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8709 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8711 /* find out how much is left in the read-ahead buffer, and rextract its pointer */
8712 cnt = PerlIO_get_cnt(fp);
8713 ptr = (STDCHAR*)PerlIO_get_ptr(fp); /* reregisterize cnt and ptr */
8714 DEBUG_P(PerlIO_printf(Perl_debug_log,
8715 "Screamer: after getc, ptr=%" UVuf ", cnt=%" IVdf "\n",
8716 PTR2UV(ptr),(IV)cnt));
8718 if (i == EOF) /* all done for ever? */
8719 goto thats_really_all_folks;
8721 /* make sure we have enough space in the target sv */
8722 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8724 SvGROW(sv, bpx + cnt + 2);
8725 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8727 /* copy of the char we got from getc() */
8728 *bp++ = (STDCHAR)i; /* store character from PerlIO_getc */
8730 /* make sure we deal with the i being the last character of a separator */
8731 if (rslen && (STDCHAR)i == rslast) /* all done for now? */
8732 goto thats_all_folks;
8736 /* check if we have actually found the separator - only really applies
8738 if ((rslen > 1 && (STRLEN)(bp - (STDCHAR*)SvPVX_const(sv)) < rslen) ||
8739 memNE((char*)bp - rslen, rsptr, rslen))
8740 goto screamer; /* go back to the fray */
8741 thats_really_all_folks:
8743 cnt += shortbuffered;
8744 DEBUG_P(PerlIO_printf(Perl_debug_log,
8745 "Screamer: quitting, ptr=%" UVuf ", cnt=%" IVdf "\n",PTR2UV(ptr),(IV)cnt));
8746 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* put these back or we're in trouble */
8747 DEBUG_P(PerlIO_printf(Perl_debug_log,
8748 "Screamer: end: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf
8750 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8751 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8753 SvCUR_set(sv, bp - (STDCHAR*)SvPVX_const(sv)); /* set length */
8754 DEBUG_P(PerlIO_printf(Perl_debug_log,
8755 "Screamer: done, len=%ld, string=|%.*s|\n",
8756 (long)SvCUR(sv),(int)SvCUR(sv),SvPVX_const(sv)));
8760 /*The big, slow, and stupid way. */
8761 #ifdef USE_HEAP_INSTEAD_OF_STACK /* Even slower way. */
8762 STDCHAR *buf = NULL;
8763 Newx(buf, 8192, STDCHAR);
8771 const STDCHAR * const bpe = buf + sizeof(buf);
8773 while ((i = PerlIO_getc(fp)) != EOF && (*bp++ = (STDCHAR)i) != rslast && bp < bpe)
8774 ; /* keep reading */
8778 cnt = PerlIO_read(fp,(char*)buf, sizeof(buf));
8779 /* Accommodate broken VAXC compiler, which applies U8 cast to
8780 * both args of ?: operator, causing EOF to change into 255
8783 i = (U8)buf[cnt - 1];
8789 cnt = 0; /* we do need to re-set the sv even when cnt <= 0 */
8791 sv_catpvn_nomg(sv, (char *) buf, cnt);
8793 sv_setpvn(sv, (char *) buf, cnt); /* "nomg" is implied */
8795 if (i != EOF && /* joy */
8797 SvCUR(sv) < rslen ||
8798 memNE(SvPVX_const(sv) + SvCUR(sv) - rslen, rsptr, rslen)))
8802 * If we're reading from a TTY and we get a short read,
8803 * indicating that the user hit his EOF character, we need
8804 * to notice it now, because if we try to read from the TTY
8805 * again, the EOF condition will disappear.
8807 * The comparison of cnt to sizeof(buf) is an optimization
8808 * that prevents unnecessary calls to feof().
8812 if (!(cnt < (I32)sizeof(buf) && PerlIO_eof(fp)))
8816 #ifdef USE_HEAP_INSTEAD_OF_STACK
8821 if (rspara) { /* have to do this both before and after */
8822 while (i != EOF) { /* to make sure file boundaries work right */
8823 i = PerlIO_getc(fp);
8825 PerlIO_ungetc(fp,i);
8831 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8837 Auto-increment of the value in the SV, doing string to numeric conversion
8838 if necessary. Handles 'get' magic and operator overloading.
8844 Perl_sv_inc(pTHX_ SV *const sv)
8853 =for apidoc sv_inc_nomg
8855 Auto-increment of the value in the SV, doing string to numeric conversion
8856 if necessary. Handles operator overloading. Skips handling 'get' magic.
8862 Perl_sv_inc_nomg(pTHX_ SV *const sv)
8869 if (SvTHINKFIRST(sv)) {
8870 if (SvREADONLY(sv)) {
8871 Perl_croak_no_modify();
8875 if (SvAMAGIC(sv) && AMG_CALLunary(sv, inc_amg))
8877 i = PTR2IV(SvRV(sv));
8881 else sv_force_normal_flags(sv, 0);
8883 flags = SvFLAGS(sv);
8884 if ((flags & (SVp_NOK|SVp_IOK)) == SVp_NOK) {
8885 /* It's (privately or publicly) a float, but not tested as an
8886 integer, so test it to see. */
8888 flags = SvFLAGS(sv);
8890 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8891 /* It's publicly an integer, or privately an integer-not-float */
8892 #ifdef PERL_PRESERVE_IVUV
8896 if (SvUVX(sv) == UV_MAX)
8897 sv_setnv(sv, UV_MAX_P1);
8899 (void)SvIOK_only_UV(sv);
8900 SvUV_set(sv, SvUVX(sv) + 1);
8902 if (SvIVX(sv) == IV_MAX)
8903 sv_setuv(sv, (UV)IV_MAX + 1);
8905 (void)SvIOK_only(sv);
8906 SvIV_set(sv, SvIVX(sv) + 1);
8911 if (flags & SVp_NOK) {
8912 const NV was = SvNVX(sv);
8913 if (LIKELY(!Perl_isinfnan(was)) &&
8914 NV_OVERFLOWS_INTEGERS_AT != 0.0 &&
8915 was >= NV_OVERFLOWS_INTEGERS_AT) {
8916 /* diag_listed_as: Lost precision when %s %f by 1 */
8917 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8918 "Lost precision when incrementing %" NVff " by 1",
8921 (void)SvNOK_only(sv);
8922 SvNV_set(sv, was + 1.0);
8926 /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */
8927 if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv)))
8928 Perl_croak_no_modify();
8930 if (!(flags & SVp_POK) || !*SvPVX_const(sv)) {
8931 if ((flags & SVTYPEMASK) < SVt_PVIV)
8932 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV ? SVt_PVIV : SVt_IV));
8933 (void)SvIOK_only(sv);
8938 while (isALPHA(*d)) d++;
8939 while (isDIGIT(*d)) d++;
8940 if (d < SvEND(sv)) {
8941 const int numtype = grok_number_flags(SvPVX_const(sv), SvCUR(sv), NULL, PERL_SCAN_TRAILING);
8942 #ifdef PERL_PRESERVE_IVUV
8943 /* Got to punt this as an integer if needs be, but we don't issue
8944 warnings. Probably ought to make the sv_iv_please() that does
8945 the conversion if possible, and silently. */
8946 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
8947 /* Need to try really hard to see if it's an integer.
8948 9.22337203685478e+18 is an integer.
8949 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
8950 so $a="9.22337203685478e+18"; $a+0; $a++
8951 needs to be the same as $a="9.22337203685478e+18"; $a++
8958 /* sv_2iv *should* have made this an NV */
8959 if (flags & SVp_NOK) {
8960 (void)SvNOK_only(sv);
8961 SvNV_set(sv, SvNVX(sv) + 1.0);
8964 /* I don't think we can get here. Maybe I should assert this
8965 And if we do get here I suspect that sv_setnv will croak. NWC
8967 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_inc punt failed to convert '%s' to IOK or NOKp, UV=0x%" UVxf " NV=%" NVgf "\n",
8968 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
8970 #endif /* PERL_PRESERVE_IVUV */
8971 if (!numtype && ckWARN(WARN_NUMERIC))
8972 not_incrementable(sv);
8973 sv_setnv(sv,Atof(SvPVX_const(sv)) + 1.0);
8977 while (d >= SvPVX_const(sv)) {
8985 /* MKS: The original code here died if letters weren't consecutive.
8986 * at least it didn't have to worry about non-C locales. The
8987 * new code assumes that ('z'-'a')==('Z'-'A'), letters are
8988 * arranged in order (although not consecutively) and that only
8989 * [A-Za-z] are accepted by isALPHA in the C locale.
8991 if (isALPHA_FOLD_NE(*d, 'z')) {
8992 do { ++*d; } while (!isALPHA(*d));
8995 *(d--) -= 'z' - 'a';
9000 *(d--) -= 'z' - 'a' + 1;
9004 /* oh,oh, the number grew */
9005 SvGROW(sv, SvCUR(sv) + 2);
9006 SvCUR_set(sv, SvCUR(sv) + 1);
9007 for (d = SvPVX(sv) + SvCUR(sv); d > SvPVX_const(sv); d--)
9018 Auto-decrement of the value in the SV, doing string to numeric conversion
9019 if necessary. Handles 'get' magic and operator overloading.
9025 Perl_sv_dec(pTHX_ SV *const sv)
9034 =for apidoc sv_dec_nomg
9036 Auto-decrement of the value in the SV, doing string to numeric conversion
9037 if necessary. Handles operator overloading. Skips handling 'get' magic.
9043 Perl_sv_dec_nomg(pTHX_ SV *const sv)
9049 if (SvTHINKFIRST(sv)) {
9050 if (SvREADONLY(sv)) {
9051 Perl_croak_no_modify();
9055 if (SvAMAGIC(sv) && AMG_CALLunary(sv, dec_amg))
9057 i = PTR2IV(SvRV(sv));
9061 else sv_force_normal_flags(sv, 0);
9063 /* Unlike sv_inc we don't have to worry about string-never-numbers
9064 and keeping them magic. But we mustn't warn on punting */
9065 flags = SvFLAGS(sv);
9066 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
9067 /* It's publicly an integer, or privately an integer-not-float */
9068 #ifdef PERL_PRESERVE_IVUV
9072 if (SvUVX(sv) == 0) {
9073 (void)SvIOK_only(sv);
9077 (void)SvIOK_only_UV(sv);
9078 SvUV_set(sv, SvUVX(sv) - 1);
9081 if (SvIVX(sv) == IV_MIN) {
9082 sv_setnv(sv, (NV)IV_MIN);
9086 (void)SvIOK_only(sv);
9087 SvIV_set(sv, SvIVX(sv) - 1);
9092 if (flags & SVp_NOK) {
9095 const NV was = SvNVX(sv);
9096 if (LIKELY(!Perl_isinfnan(was)) &&
9097 NV_OVERFLOWS_INTEGERS_AT != 0.0 &&
9098 was <= -NV_OVERFLOWS_INTEGERS_AT) {
9099 /* diag_listed_as: Lost precision when %s %f by 1 */
9100 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
9101 "Lost precision when decrementing %" NVff " by 1",
9104 (void)SvNOK_only(sv);
9105 SvNV_set(sv, was - 1.0);
9110 /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */
9111 if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv)))
9112 Perl_croak_no_modify();
9114 if (!(flags & SVp_POK)) {
9115 if ((flags & SVTYPEMASK) < SVt_PVIV)
9116 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV) ? SVt_PVIV : SVt_IV);
9118 (void)SvIOK_only(sv);
9121 #ifdef PERL_PRESERVE_IVUV
9123 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), NULL);
9124 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
9125 /* Need to try really hard to see if it's an integer.
9126 9.22337203685478e+18 is an integer.
9127 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
9128 so $a="9.22337203685478e+18"; $a+0; $a--
9129 needs to be the same as $a="9.22337203685478e+18"; $a--
9136 /* sv_2iv *should* have made this an NV */
9137 if (flags & SVp_NOK) {
9138 (void)SvNOK_only(sv);
9139 SvNV_set(sv, SvNVX(sv) - 1.0);
9142 /* I don't think we can get here. Maybe I should assert this
9143 And if we do get here I suspect that sv_setnv will croak. NWC
9145 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_dec punt failed to convert '%s' to IOK or NOKp, UV=0x%" UVxf " NV=%" NVgf "\n",
9146 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
9149 #endif /* PERL_PRESERVE_IVUV */
9150 sv_setnv(sv,Atof(SvPVX_const(sv)) - 1.0); /* punt */
9153 /* this define is used to eliminate a chunk of duplicated but shared logic
9154 * it has the suffix __SV_C to signal that it isnt API, and isnt meant to be
9155 * used anywhere but here - yves
9157 #define PUSH_EXTEND_MORTAL__SV_C(AnSv) \
9159 SSize_t ix = ++PL_tmps_ix; \
9160 if (UNLIKELY(ix >= PL_tmps_max)) \
9161 ix = tmps_grow_p(ix); \
9162 PL_tmps_stack[ix] = (AnSv); \
9166 =for apidoc sv_mortalcopy
9168 Creates a new SV which is a copy of the original SV (using C<sv_setsv>).
9169 The new SV is marked as mortal. It will be destroyed "soon", either by an
9170 explicit call to C<FREETMPS>, or by an implicit call at places such as
9171 statement boundaries. See also C<L</sv_newmortal>> and C<L</sv_2mortal>>.
9176 /* Make a string that will exist for the duration of the expression
9177 * evaluation. Actually, it may have to last longer than that, but
9178 * hopefully we won't free it until it has been assigned to a
9179 * permanent location. */
9182 Perl_sv_mortalcopy_flags(pTHX_ SV *const oldstr, U32 flags)
9186 if (flags & SV_GMAGIC)
9187 SvGETMAGIC(oldstr); /* before new_SV, in case it dies */
9189 sv_setsv_flags(sv,oldstr,flags & ~SV_GMAGIC);
9190 PUSH_EXTEND_MORTAL__SV_C(sv);
9196 =for apidoc sv_newmortal
9198 Creates a new null SV which is mortal. The reference count of the SV is
9199 set to 1. It will be destroyed "soon", either by an explicit call to
9200 C<FREETMPS>, or by an implicit call at places such as statement boundaries.
9201 See also C<L</sv_mortalcopy>> and C<L</sv_2mortal>>.
9207 Perl_sv_newmortal(pTHX)
9212 SvFLAGS(sv) = SVs_TEMP;
9213 PUSH_EXTEND_MORTAL__SV_C(sv);
9219 =for apidoc newSVpvn_flags
9221 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9222 characters) into it. The reference count for the
9223 SV is set to 1. Note that if C<len> is zero, Perl will create a zero length
9224 string. You are responsible for ensuring that the source string is at least
9225 C<len> bytes long. If the C<s> argument is NULL the new SV will be undefined.
9226 Currently the only flag bits accepted are C<SVf_UTF8> and C<SVs_TEMP>.
9227 If C<SVs_TEMP> is set, then C<sv_2mortal()> is called on the result before
9228 returning. If C<SVf_UTF8> is set, C<s>
9229 is considered to be in UTF-8 and the
9230 C<SVf_UTF8> flag will be set on the new SV.
9231 C<newSVpvn_utf8()> is a convenience wrapper for this function, defined as
9233 #define newSVpvn_utf8(s, len, u) \
9234 newSVpvn_flags((s), (len), (u) ? SVf_UTF8 : 0)
9240 Perl_newSVpvn_flags(pTHX_ const char *const s, const STRLEN len, const U32 flags)
9244 /* All the flags we don't support must be zero.
9245 And we're new code so I'm going to assert this from the start. */
9246 assert(!(flags & ~(SVf_UTF8|SVs_TEMP)));
9248 sv_setpvn(sv,s,len);
9250 /* This code used to do a sv_2mortal(), however we now unroll the call to
9251 * sv_2mortal() and do what it does ourselves here. Since we have asserted
9252 * that flags can only have the SVf_UTF8 and/or SVs_TEMP flags set above we
9253 * can use it to enable the sv flags directly (bypassing SvTEMP_on), which
9254 * in turn means we dont need to mask out the SVf_UTF8 flag below, which
9255 * means that we eliminate quite a few steps than it looks - Yves
9256 * (explaining patch by gfx) */
9258 SvFLAGS(sv) |= flags;
9260 if(flags & SVs_TEMP){
9261 PUSH_EXTEND_MORTAL__SV_C(sv);
9268 =for apidoc sv_2mortal
9270 Marks an existing SV as mortal. The SV will be destroyed "soon", either
9271 by an explicit call to C<FREETMPS>, or by an implicit call at places such as
9272 statement boundaries. C<SvTEMP()> is turned on which means that the SV's
9273 string buffer can be "stolen" if this SV is copied. See also
9274 C<L</sv_newmortal>> and C<L</sv_mortalcopy>>.
9280 Perl_sv_2mortal(pTHX_ SV *const sv)
9287 PUSH_EXTEND_MORTAL__SV_C(sv);
9295 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9296 characters) into it. The reference count for the
9297 SV is set to 1. If C<len> is zero, Perl will compute the length using
9298 C<strlen()>, (which means if you use this option, that C<s> can't have embedded
9299 C<NUL> characters and has to have a terminating C<NUL> byte).
9301 This function can cause reliability issues if you are likely to pass in
9302 empty strings that are not null terminated, because it will run
9303 strlen on the string and potentially run past valid memory.
9305 Using L</newSVpvn> is a safer alternative for non C<NUL> terminated strings.
9306 For string literals use L</newSVpvs> instead. This function will work fine for
9307 C<NUL> terminated strings, but if you want to avoid the if statement on whether
9308 to call C<strlen> use C<newSVpvn> instead (calling C<strlen> yourself).
9314 Perl_newSVpv(pTHX_ const char *const s, const STRLEN len)
9319 sv_setpvn(sv, s, len || s == NULL ? len : strlen(s));
9324 =for apidoc newSVpvn
9326 Creates a new SV and copies a string into it, which may contain C<NUL> characters
9327 (C<\0>) and other binary data. The reference count for the SV is set to 1.
9328 Note that if C<len> is zero, Perl will create a zero length (Perl) string. You
9329 are responsible for ensuring that the source buffer is at least
9330 C<len> bytes long. If the C<s> argument is NULL the new SV will be
9337 Perl_newSVpvn(pTHX_ const char *const buffer, const STRLEN len)
9341 sv_setpvn(sv,buffer,len);
9346 =for apidoc newSVhek
9348 Creates a new SV from the hash key structure. It will generate scalars that
9349 point to the shared string table where possible. Returns a new (undefined)
9350 SV if C<hek> is NULL.
9356 Perl_newSVhek(pTHX_ const HEK *const hek)
9365 if (HEK_LEN(hek) == HEf_SVKEY) {
9366 return newSVsv(*(SV**)HEK_KEY(hek));
9368 const int flags = HEK_FLAGS(hek);
9369 if (flags & HVhek_WASUTF8) {
9371 Andreas would like keys he put in as utf8 to come back as utf8
9373 STRLEN utf8_len = HEK_LEN(hek);
9374 SV * const sv = newSV_type(SVt_PV);
9375 char *as_utf8 = (char *)bytes_to_utf8 ((U8*)HEK_KEY(hek), &utf8_len);
9376 /* bytes_to_utf8() allocates a new string, which we can repurpose: */
9377 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
9380 } else if (flags & HVhek_UNSHARED) {
9381 /* A hash that isn't using shared hash keys has to have
9382 the flag in every key so that we know not to try to call
9383 share_hek_hek on it. */
9385 SV * const sv = newSVpvn (HEK_KEY(hek), HEK_LEN(hek));
9390 /* This will be overwhelminly the most common case. */
9392 /* Inline most of newSVpvn_share(), because share_hek_hek() is far
9393 more efficient than sharepvn(). */
9397 sv_upgrade(sv, SVt_PV);
9398 SvPV_set(sv, (char *)HEK_KEY(share_hek_hek(hek)));
9399 SvCUR_set(sv, HEK_LEN(hek));
9411 =for apidoc newSVpvn_share
9413 Creates a new SV with its C<SvPVX_const> pointing to a shared string in the string
9414 table. If the string does not already exist in the table, it is
9415 created first. Turns on the C<SvIsCOW> flag (or C<READONLY>
9416 and C<FAKE> in 5.16 and earlier). If the C<hash> parameter
9417 is non-zero, that value is used; otherwise the hash is computed.
9418 The string's hash can later be retrieved from the SV
9419 with the C<SvSHARED_HASH()> macro. The idea here is
9420 that as the string table is used for shared hash keys these strings will have
9421 C<SvPVX_const == HeKEY> and hash lookup will avoid string compare.
9427 Perl_newSVpvn_share(pTHX_ const char *src, I32 len, U32 hash)
9431 bool is_utf8 = FALSE;
9432 const char *const orig_src = src;
9435 STRLEN tmplen = -len;
9437 /* See the note in hv.c:hv_fetch() --jhi */
9438 src = (char*)bytes_from_utf8((const U8*)src, &tmplen, &is_utf8);
9442 PERL_HASH(hash, src, len);
9444 /* The logic for this is inlined in S_mro_get_linear_isa_dfs(), so if it
9445 changes here, update it there too. */
9446 sv_upgrade(sv, SVt_PV);
9447 SvPV_set(sv, sharepvn(src, is_utf8?-len:len, hash));
9454 if (src != orig_src)
9460 =for apidoc newSVpv_share
9462 Like C<newSVpvn_share>, but takes a C<NUL>-terminated string instead of a
9469 Perl_newSVpv_share(pTHX_ const char *src, U32 hash)
9471 return newSVpvn_share(src, strlen(src), hash);
9474 #if defined(PERL_IMPLICIT_CONTEXT)
9476 /* pTHX_ magic can't cope with varargs, so this is a no-context
9477 * version of the main function, (which may itself be aliased to us).
9478 * Don't access this version directly.
9482 Perl_newSVpvf_nocontext(const char *const pat, ...)
9488 PERL_ARGS_ASSERT_NEWSVPVF_NOCONTEXT;
9490 va_start(args, pat);
9491 sv = vnewSVpvf(pat, &args);
9498 =for apidoc newSVpvf
9500 Creates a new SV and initializes it with the string formatted like
9507 Perl_newSVpvf(pTHX_ const char *const pat, ...)
9512 PERL_ARGS_ASSERT_NEWSVPVF;
9514 va_start(args, pat);
9515 sv = vnewSVpvf(pat, &args);
9520 /* backend for newSVpvf() and newSVpvf_nocontext() */
9523 Perl_vnewSVpvf(pTHX_ const char *const pat, va_list *const args)
9527 PERL_ARGS_ASSERT_VNEWSVPVF;
9530 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
9537 Creates a new SV and copies a floating point value into it.
9538 The reference count for the SV is set to 1.
9544 Perl_newSVnv(pTHX_ const NV n)
9556 Creates a new SV and copies an integer into it. The reference count for the
9563 Perl_newSViv(pTHX_ const IV i)
9569 /* Inlining ONLY the small relevant subset of sv_setiv here
9570 * for performance. Makes a significant difference. */
9572 /* We're starting from SVt_FIRST, so provided that's
9573 * actual 0, we don't have to unset any SV type flags
9574 * to promote to SVt_IV. */
9575 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9577 SET_SVANY_FOR_BODYLESS_IV(sv);
9578 SvFLAGS(sv) |= SVt_IV;
9590 Creates a new SV and copies an unsigned integer into it.
9591 The reference count for the SV is set to 1.
9597 Perl_newSVuv(pTHX_ const UV u)
9601 /* Inlining ONLY the small relevant subset of sv_setuv here
9602 * for performance. Makes a significant difference. */
9604 /* Using ivs is more efficient than using uvs - see sv_setuv */
9605 if (u <= (UV)IV_MAX) {
9606 return newSViv((IV)u);
9611 /* We're starting from SVt_FIRST, so provided that's
9612 * actual 0, we don't have to unset any SV type flags
9613 * to promote to SVt_IV. */
9614 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9616 SET_SVANY_FOR_BODYLESS_IV(sv);
9617 SvFLAGS(sv) |= SVt_IV;
9619 (void)SvIsUV_on(sv);
9628 =for apidoc newSV_type
9630 Creates a new SV, of the type specified. The reference count for the new SV
9637 Perl_newSV_type(pTHX_ const svtype type)
9642 ASSUME(SvTYPE(sv) == SVt_FIRST);
9643 if(type != SVt_FIRST)
9644 sv_upgrade(sv, type);
9649 =for apidoc newRV_noinc
9651 Creates an RV wrapper for an SV. The reference count for the original
9652 SV is B<not> incremented.
9658 Perl_newRV_noinc(pTHX_ SV *const tmpRef)
9662 PERL_ARGS_ASSERT_NEWRV_NOINC;
9666 /* We're starting from SVt_FIRST, so provided that's
9667 * actual 0, we don't have to unset any SV type flags
9668 * to promote to SVt_IV. */
9669 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9671 SET_SVANY_FOR_BODYLESS_IV(sv);
9672 SvFLAGS(sv) |= SVt_IV;
9677 SvRV_set(sv, tmpRef);
9682 /* newRV_inc is the official function name to use now.
9683 * newRV_inc is in fact #defined to newRV in sv.h
9687 Perl_newRV(pTHX_ SV *const sv)
9689 PERL_ARGS_ASSERT_NEWRV;
9691 return newRV_noinc(SvREFCNT_inc_simple_NN(sv));
9697 Creates a new SV which is an exact duplicate of the original SV.
9704 Perl_newSVsv(pTHX_ SV *const old)
9710 if (SvTYPE(old) == (svtype)SVTYPEMASK) {
9711 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), "semi-panic: attempt to dup freed string");
9714 /* Do this here, otherwise we leak the new SV if this croaks. */
9717 /* SV_NOSTEAL prevents TEMP buffers being, well, stolen, and saves games
9718 with SvTEMP_off and SvTEMP_on round a call to sv_setsv. */
9719 sv_setsv_flags(sv, old, SV_NOSTEAL);
9724 =for apidoc sv_reset
9726 Underlying implementation for the C<reset> Perl function.
9727 Note that the perl-level function is vaguely deprecated.
9733 Perl_sv_reset(pTHX_ const char *s, HV *const stash)
9735 PERL_ARGS_ASSERT_SV_RESET;
9737 sv_resetpvn(*s ? s : NULL, strlen(s), stash);
9741 Perl_sv_resetpvn(pTHX_ const char *s, STRLEN len, HV * const stash)
9743 char todo[PERL_UCHAR_MAX+1];
9746 if (!stash || SvTYPE(stash) != SVt_PVHV)
9749 if (!s) { /* reset ?? searches */
9750 MAGIC * const mg = mg_find((const SV *)stash, PERL_MAGIC_symtab);
9752 const U32 count = mg->mg_len / sizeof(PMOP**);
9753 PMOP **pmp = (PMOP**) mg->mg_ptr;
9754 PMOP *const *const end = pmp + count;
9758 SvREADONLY_off(PL_regex_pad[(*pmp)->op_pmoffset]);
9760 (*pmp)->op_pmflags &= ~PMf_USED;
9768 /* reset variables */
9770 if (!HvARRAY(stash))
9773 Zero(todo, 256, char);
9777 I32 i = (unsigned char)*s;
9781 max = (unsigned char)*s++;
9782 for ( ; i <= max; i++) {
9785 for (i = 0; i <= (I32) HvMAX(stash); i++) {
9787 for (entry = HvARRAY(stash)[i];
9789 entry = HeNEXT(entry))
9794 if (!todo[(U8)*HeKEY(entry)])
9796 gv = MUTABLE_GV(HeVAL(entry));
9800 if (sv && !SvREADONLY(sv)) {
9801 SV_CHECK_THINKFIRST_COW_DROP(sv);
9802 if (!isGV(sv)) SvOK_off(sv);
9807 if (GvHV(gv) && !HvNAME_get(GvHV(gv))) {
9818 Using various gambits, try to get an IO from an SV: the IO slot if its a
9819 GV; or the recursive result if we're an RV; or the IO slot of the symbol
9820 named after the PV if we're a string.
9822 'Get' magic is ignored on the C<sv> passed in, but will be called on
9823 C<SvRV(sv)> if C<sv> is an RV.
9829 Perl_sv_2io(pTHX_ SV *const sv)
9834 PERL_ARGS_ASSERT_SV_2IO;
9836 switch (SvTYPE(sv)) {
9838 io = MUTABLE_IO(sv);
9842 if (isGV_with_GP(sv)) {
9843 gv = MUTABLE_GV(sv);
9846 Perl_croak(aTHX_ "Bad filehandle: %" HEKf,
9847 HEKfARG(GvNAME_HEK(gv)));
9853 Perl_croak(aTHX_ PL_no_usym, "filehandle");
9855 SvGETMAGIC(SvRV(sv));
9856 return sv_2io(SvRV(sv));
9858 gv = gv_fetchsv_nomg(sv, 0, SVt_PVIO);
9865 if (SvGMAGICAL(sv)) {
9866 newsv = sv_newmortal();
9867 sv_setsv_nomg(newsv, sv);
9869 Perl_croak(aTHX_ "Bad filehandle: %" SVf, SVfARG(newsv));
9879 Using various gambits, try to get a CV from an SV; in addition, try if
9880 possible to set C<*st> and C<*gvp> to the stash and GV associated with it.
9881 The flags in C<lref> are passed to C<gv_fetchsv>.
9887 Perl_sv_2cv(pTHX_ SV *sv, HV **const st, GV **const gvp, const I32 lref)
9892 PERL_ARGS_ASSERT_SV_2CV;
9899 switch (SvTYPE(sv)) {
9903 return MUTABLE_CV(sv);
9913 sv = amagic_deref_call(sv, to_cv_amg);
9916 if (SvTYPE(sv) == SVt_PVCV) {
9917 cv = MUTABLE_CV(sv);
9922 else if(SvGETMAGIC(sv), isGV_with_GP(sv))
9923 gv = MUTABLE_GV(sv);
9925 Perl_croak(aTHX_ "Not a subroutine reference");
9927 else if (isGV_with_GP(sv)) {
9928 gv = MUTABLE_GV(sv);
9931 gv = gv_fetchsv_nomg(sv, lref, SVt_PVCV);
9938 /* Some flags to gv_fetchsv mean don't really create the GV */
9939 if (!isGV_with_GP(gv)) {
9944 if (lref & ~GV_ADDMG && !GvCVu(gv)) {
9945 /* XXX this is probably not what they think they're getting.
9946 * It has the same effect as "sub name;", i.e. just a forward
9957 Returns true if the SV has a true value by Perl's rules.
9958 Use the C<SvTRUE> macro instead, which may call C<sv_true()> or may
9959 instead use an in-line version.
9965 Perl_sv_true(pTHX_ SV *const sv)
9970 const XPV* const tXpv = (XPV*)SvANY(sv);
9972 (tXpv->xpv_cur > 1 ||
9973 (tXpv->xpv_cur && *sv->sv_u.svu_pv != '0')))
9980 return SvIVX(sv) != 0;
9983 return SvNVX(sv) != 0.0;
9985 return sv_2bool(sv);
9991 =for apidoc sv_pvn_force
9993 Get a sensible string out of the SV somehow.
9994 A private implementation of the C<SvPV_force> macro for compilers which
9995 can't cope with complex macro expressions. Always use the macro instead.
9997 =for apidoc sv_pvn_force_flags
9999 Get a sensible string out of the SV somehow.
10000 If C<flags> has the C<SV_GMAGIC> bit set, will C<mg_get> on C<sv> if
10001 appropriate, else not. C<sv_pvn_force> and C<sv_pvn_force_nomg> are
10002 implemented in terms of this function.
10003 You normally want to use the various wrapper macros instead: see
10004 C<L</SvPV_force>> and C<L</SvPV_force_nomg>>.
10010 Perl_sv_pvn_force_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
10012 PERL_ARGS_ASSERT_SV_PVN_FORCE_FLAGS;
10014 if (flags & SV_GMAGIC) SvGETMAGIC(sv);
10015 if (SvTHINKFIRST(sv) && (!SvROK(sv) || SvREADONLY(sv)))
10016 sv_force_normal_flags(sv, 0);
10026 if (SvTYPE(sv) > SVt_PVLV
10027 || isGV_with_GP(sv))
10028 /* diag_listed_as: Can't coerce %s to %s in %s */
10029 Perl_croak(aTHX_ "Can't coerce %s to string in %s", sv_reftype(sv,0),
10031 s = sv_2pv_flags(sv, &len, flags &~ SV_GMAGIC);
10038 if (SvTYPE(sv) < SVt_PV ||
10039 s != SvPVX_const(sv)) { /* Almost, but not quite, sv_setpvn() */
10042 SvUPGRADE(sv, SVt_PV); /* Never FALSE */
10043 SvGROW(sv, len + 1);
10044 Move(s,SvPVX(sv),len,char);
10045 SvCUR_set(sv, len);
10046 SvPVX(sv)[len] = '\0';
10049 SvPOK_on(sv); /* validate pointer */
10051 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2pv(%s)\n",
10052 PTR2UV(sv),SvPVX_const(sv)));
10055 (void)SvPOK_only_UTF8(sv);
10056 return SvPVX_mutable(sv);
10060 =for apidoc sv_pvbyten_force
10062 The backend for the C<SvPVbytex_force> macro. Always use the macro
10069 Perl_sv_pvbyten_force(pTHX_ SV *const sv, STRLEN *const lp)
10071 PERL_ARGS_ASSERT_SV_PVBYTEN_FORCE;
10073 sv_pvn_force(sv,lp);
10074 sv_utf8_downgrade(sv,0);
10080 =for apidoc sv_pvutf8n_force
10082 The backend for the C<SvPVutf8x_force> macro. Always use the macro
10089 Perl_sv_pvutf8n_force(pTHX_ SV *const sv, STRLEN *const lp)
10091 PERL_ARGS_ASSERT_SV_PVUTF8N_FORCE;
10093 sv_pvn_force(sv,0);
10094 sv_utf8_upgrade_nomg(sv);
10100 =for apidoc sv_reftype
10102 Returns a string describing what the SV is a reference to.
10104 If ob is true and the SV is blessed, the string is the class name,
10105 otherwise it is the type of the SV, "SCALAR", "ARRAY" etc.
10111 Perl_sv_reftype(pTHX_ const SV *const sv, const int ob)
10113 PERL_ARGS_ASSERT_SV_REFTYPE;
10114 if (ob && SvOBJECT(sv)) {
10115 return SvPV_nolen_const(sv_ref(NULL, sv, ob));
10118 /* WARNING - There is code, for instance in mg.c, that assumes that
10119 * the only reason that sv_reftype(sv,0) would return a string starting
10120 * with 'L' or 'S' is that it is a LVALUE or a SCALAR.
10121 * Yes this a dodgy way to do type checking, but it saves practically reimplementing
10122 * this routine inside other subs, and it saves time.
10123 * Do not change this assumption without searching for "dodgy type check" in
10126 switch (SvTYPE(sv)) {
10141 case SVt_PVLV: return (char *) (SvROK(sv) ? "REF"
10142 /* tied lvalues should appear to be
10143 * scalars for backwards compatibility */
10144 : (isALPHA_FOLD_EQ(LvTYPE(sv), 't'))
10145 ? "SCALAR" : "LVALUE");
10146 case SVt_PVAV: return "ARRAY";
10147 case SVt_PVHV: return "HASH";
10148 case SVt_PVCV: return "CODE";
10149 case SVt_PVGV: return (char *) (isGV_with_GP(sv)
10150 ? "GLOB" : "SCALAR");
10151 case SVt_PVFM: return "FORMAT";
10152 case SVt_PVIO: return "IO";
10153 case SVt_INVLIST: return "INVLIST";
10154 case SVt_REGEXP: return "REGEXP";
10155 default: return "UNKNOWN";
10163 Returns a SV describing what the SV passed in is a reference to.
10165 dst can be a SV to be set to the description or NULL, in which case a
10166 mortal SV is returned.
10168 If ob is true and the SV is blessed, the description is the class
10169 name, otherwise it is the type of the SV, "SCALAR", "ARRAY" etc.
10175 Perl_sv_ref(pTHX_ SV *dst, const SV *const sv, const int ob)
10177 PERL_ARGS_ASSERT_SV_REF;
10180 dst = sv_newmortal();
10182 if (ob && SvOBJECT(sv)) {
10183 HvNAME_get(SvSTASH(sv))
10184 ? sv_sethek(dst, HvNAME_HEK(SvSTASH(sv)))
10185 : sv_setpvs(dst, "__ANON__");
10188 const char * reftype = sv_reftype(sv, 0);
10189 sv_setpv(dst, reftype);
10195 =for apidoc sv_isobject
10197 Returns a boolean indicating whether the SV is an RV pointing to a blessed
10198 object. If the SV is not an RV, or if the object is not blessed, then this
10205 Perl_sv_isobject(pTHX_ SV *sv)
10221 Returns a boolean indicating whether the SV is blessed into the specified
10222 class. This does not check for subtypes; use C<sv_derived_from> to verify
10223 an inheritance relationship.
10229 Perl_sv_isa(pTHX_ SV *sv, const char *const name)
10231 const char *hvname;
10233 PERL_ARGS_ASSERT_SV_ISA;
10243 hvname = HvNAME_get(SvSTASH(sv));
10247 return strEQ(hvname, name);
10251 =for apidoc newSVrv
10253 Creates a new SV for the existing RV, C<rv>, to point to. If C<rv> is not an
10254 RV then it will be upgraded to one. If C<classname> is non-null then the new
10255 SV will be blessed in the specified package. The new SV is returned and its
10256 reference count is 1. The reference count 1 is owned by C<rv>.
10262 Perl_newSVrv(pTHX_ SV *const rv, const char *const classname)
10266 PERL_ARGS_ASSERT_NEWSVRV;
10270 SV_CHECK_THINKFIRST_COW_DROP(rv);
10272 if (UNLIKELY( SvTYPE(rv) >= SVt_PVMG )) {
10273 const U32 refcnt = SvREFCNT(rv);
10277 SvREFCNT(rv) = refcnt;
10279 sv_upgrade(rv, SVt_IV);
10280 } else if (SvROK(rv)) {
10281 SvREFCNT_dec(SvRV(rv));
10283 prepare_SV_for_RV(rv);
10291 HV* const stash = gv_stashpv(classname, GV_ADD);
10292 (void)sv_bless(rv, stash);
10298 Perl_newSVavdefelem(pTHX_ AV *av, SSize_t ix, bool extendible)
10300 SV * const lv = newSV_type(SVt_PVLV);
10301 PERL_ARGS_ASSERT_NEWSVAVDEFELEM;
10303 sv_magic(lv, NULL, PERL_MAGIC_defelem, NULL, 0);
10304 LvTARG(lv) = SvREFCNT_inc_simple_NN(av);
10305 LvSTARGOFF(lv) = ix;
10306 LvTARGLEN(lv) = extendible ? 1 : (STRLEN)UV_MAX;
10311 =for apidoc sv_setref_pv
10313 Copies a pointer into a new SV, optionally blessing the SV. The C<rv>
10314 argument will be upgraded to an RV. That RV will be modified to point to
10315 the new SV. If the C<pv> argument is C<NULL>, then C<PL_sv_undef> will be placed
10316 into the SV. The C<classname> argument indicates the package for the
10317 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10318 will have a reference count of 1, and the RV will be returned.
10320 Do not use with other Perl types such as HV, AV, SV, CV, because those
10321 objects will become corrupted by the pointer copy process.
10323 Note that C<sv_setref_pvn> copies the string while this copies the pointer.
10329 Perl_sv_setref_pv(pTHX_ SV *const rv, const char *const classname, void *const pv)
10331 PERL_ARGS_ASSERT_SV_SETREF_PV;
10338 sv_setiv(newSVrv(rv,classname), PTR2IV(pv));
10343 =for apidoc sv_setref_iv
10345 Copies an integer into a new SV, optionally blessing the SV. The C<rv>
10346 argument will be upgraded to an RV. That RV will be modified to point to
10347 the new SV. The C<classname> argument indicates the package for the
10348 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10349 will have a reference count of 1, and the RV will be returned.
10355 Perl_sv_setref_iv(pTHX_ SV *const rv, const char *const classname, const IV iv)
10357 PERL_ARGS_ASSERT_SV_SETREF_IV;
10359 sv_setiv(newSVrv(rv,classname), iv);
10364 =for apidoc sv_setref_uv
10366 Copies an unsigned integer into a new SV, optionally blessing the SV. The C<rv>
10367 argument will be upgraded to an RV. That RV will be modified to point to
10368 the new SV. The C<classname> argument indicates the package for the
10369 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10370 will have a reference count of 1, and the RV will be returned.
10376 Perl_sv_setref_uv(pTHX_ SV *const rv, const char *const classname, const UV uv)
10378 PERL_ARGS_ASSERT_SV_SETREF_UV;
10380 sv_setuv(newSVrv(rv,classname), uv);
10385 =for apidoc sv_setref_nv
10387 Copies a double into a new SV, optionally blessing the SV. The C<rv>
10388 argument will be upgraded to an RV. That RV will be modified to point to
10389 the new SV. The C<classname> argument indicates the package for the
10390 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10391 will have a reference count of 1, and the RV will be returned.
10397 Perl_sv_setref_nv(pTHX_ SV *const rv, const char *const classname, const NV nv)
10399 PERL_ARGS_ASSERT_SV_SETREF_NV;
10401 sv_setnv(newSVrv(rv,classname), nv);
10406 =for apidoc sv_setref_pvn
10408 Copies a string into a new SV, optionally blessing the SV. The length of the
10409 string must be specified with C<n>. The C<rv> argument will be upgraded to
10410 an RV. That RV will be modified to point to the new SV. The C<classname>
10411 argument indicates the package for the blessing. Set C<classname> to
10412 C<NULL> to avoid the blessing. The new SV will have a reference count
10413 of 1, and the RV will be returned.
10415 Note that C<sv_setref_pv> copies the pointer while this copies the string.
10421 Perl_sv_setref_pvn(pTHX_ SV *const rv, const char *const classname,
10422 const char *const pv, const STRLEN n)
10424 PERL_ARGS_ASSERT_SV_SETREF_PVN;
10426 sv_setpvn(newSVrv(rv,classname), pv, n);
10431 =for apidoc sv_bless
10433 Blesses an SV into a specified package. The SV must be an RV. The package
10434 must be designated by its stash (see C<L</gv_stashpv>>). The reference count
10435 of the SV is unaffected.
10441 Perl_sv_bless(pTHX_ SV *const sv, HV *const stash)
10444 HV *oldstash = NULL;
10446 PERL_ARGS_ASSERT_SV_BLESS;
10450 Perl_croak(aTHX_ "Can't bless non-reference value");
10452 if (SvFLAGS(tmpRef) & (SVs_OBJECT|SVf_READONLY|SVf_PROTECT)) {
10453 if (SvREADONLY(tmpRef))
10454 Perl_croak_no_modify();
10455 if (SvOBJECT(tmpRef)) {
10456 oldstash = SvSTASH(tmpRef);
10459 SvOBJECT_on(tmpRef);
10460 SvUPGRADE(tmpRef, SVt_PVMG);
10461 SvSTASH_set(tmpRef, MUTABLE_HV(SvREFCNT_inc_simple(stash)));
10462 SvREFCNT_dec(oldstash);
10464 if(SvSMAGICAL(tmpRef))
10465 if(mg_find(tmpRef, PERL_MAGIC_ext) || mg_find(tmpRef, PERL_MAGIC_uvar))
10473 /* Downgrades a PVGV to a PVMG. If it's actually a PVLV, we leave the type
10474 * as it is after unglobbing it.
10477 PERL_STATIC_INLINE void
10478 S_sv_unglob(pTHX_ SV *const sv, U32 flags)
10482 SV * const temp = flags & SV_COW_DROP_PV ? NULL : sv_newmortal();
10484 PERL_ARGS_ASSERT_SV_UNGLOB;
10486 assert(SvTYPE(sv) == SVt_PVGV || SvTYPE(sv) == SVt_PVLV);
10488 if (!(flags & SV_COW_DROP_PV))
10489 gv_efullname3(temp, MUTABLE_GV(sv), "*");
10491 SvREFCNT_inc_simple_void_NN(sv_2mortal(sv));
10493 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
10494 && HvNAME_get(stash))
10495 mro_method_changed_in(stash);
10496 gp_free(MUTABLE_GV(sv));
10499 sv_del_backref(MUTABLE_SV(GvSTASH(sv)), sv);
10500 GvSTASH(sv) = NULL;
10503 if (GvNAME_HEK(sv)) {
10504 unshare_hek(GvNAME_HEK(sv));
10506 isGV_with_GP_off(sv);
10508 if(SvTYPE(sv) == SVt_PVGV) {
10509 /* need to keep SvANY(sv) in the right arena */
10510 xpvmg = new_XPVMG();
10511 StructCopy(SvANY(sv), xpvmg, XPVMG);
10512 del_XPVGV(SvANY(sv));
10515 SvFLAGS(sv) &= ~SVTYPEMASK;
10516 SvFLAGS(sv) |= SVt_PVMG;
10519 /* Intentionally not calling any local SET magic, as this isn't so much a
10520 set operation as merely an internal storage change. */
10521 if (flags & SV_COW_DROP_PV) SvOK_off(sv);
10522 else sv_setsv_flags(sv, temp, 0);
10524 if ((const GV *)sv == PL_last_in_gv)
10525 PL_last_in_gv = NULL;
10526 else if ((const GV *)sv == PL_statgv)
10531 =for apidoc sv_unref_flags
10533 Unsets the RV status of the SV, and decrements the reference count of
10534 whatever was being referenced by the RV. This can almost be thought of
10535 as a reversal of C<newSVrv>. The C<cflags> argument can contain
10536 C<SV_IMMEDIATE_UNREF> to force the reference count to be decremented
10537 (otherwise the decrementing is conditional on the reference count being
10538 different from one or the reference being a readonly SV).
10539 See C<L</SvROK_off>>.
10545 Perl_sv_unref_flags(pTHX_ SV *const ref, const U32 flags)
10547 SV* const target = SvRV(ref);
10549 PERL_ARGS_ASSERT_SV_UNREF_FLAGS;
10551 if (SvWEAKREF(ref)) {
10552 sv_del_backref(target, ref);
10553 SvWEAKREF_off(ref);
10554 SvRV_set(ref, NULL);
10557 SvRV_set(ref, NULL);
10559 /* You can't have a || SvREADONLY(target) here, as $a = $$a, where $a was
10560 assigned to as BEGIN {$a = \"Foo"} will fail. */
10561 if (SvREFCNT(target) != 1 || (flags & SV_IMMEDIATE_UNREF))
10562 SvREFCNT_dec_NN(target);
10563 else /* XXX Hack, but hard to make $a=$a->[1] work otherwise */
10564 sv_2mortal(target); /* Schedule for freeing later */
10568 =for apidoc sv_untaint
10570 Untaint an SV. Use C<SvTAINTED_off> instead.
10576 Perl_sv_untaint(pTHX_ SV *const sv)
10578 PERL_ARGS_ASSERT_SV_UNTAINT;
10579 PERL_UNUSED_CONTEXT;
10581 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10582 MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10589 =for apidoc sv_tainted
10591 Test an SV for taintedness. Use C<SvTAINTED> instead.
10597 Perl_sv_tainted(pTHX_ SV *const sv)
10599 PERL_ARGS_ASSERT_SV_TAINTED;
10600 PERL_UNUSED_CONTEXT;
10602 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10603 const MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10604 if (mg && (mg->mg_len & 1) )
10610 #ifndef NO_MATHOMS /* Can't move these to mathoms.c because call uiv_2buf(),
10611 private to this file */
10614 =for apidoc sv_setpviv
10616 Copies an integer into the given SV, also updating its string value.
10617 Does not handle 'set' magic. See C<L</sv_setpviv_mg>>.
10623 Perl_sv_setpviv(pTHX_ SV *const sv, const IV iv)
10625 char buf[TYPE_CHARS(UV)];
10627 char * const ptr = uiv_2buf(buf, iv, 0, 0, &ebuf);
10629 PERL_ARGS_ASSERT_SV_SETPVIV;
10631 sv_setpvn(sv, ptr, ebuf - ptr);
10635 =for apidoc sv_setpviv_mg
10637 Like C<sv_setpviv>, but also handles 'set' magic.
10643 Perl_sv_setpviv_mg(pTHX_ SV *const sv, const IV iv)
10645 PERL_ARGS_ASSERT_SV_SETPVIV_MG;
10647 sv_setpviv(sv, iv);
10651 #endif /* NO_MATHOMS */
10653 #if defined(PERL_IMPLICIT_CONTEXT)
10655 /* pTHX_ magic can't cope with varargs, so this is a no-context
10656 * version of the main function, (which may itself be aliased to us).
10657 * Don't access this version directly.
10661 Perl_sv_setpvf_nocontext(SV *const sv, const char *const pat, ...)
10666 PERL_ARGS_ASSERT_SV_SETPVF_NOCONTEXT;
10668 va_start(args, pat);
10669 sv_vsetpvf(sv, pat, &args);
10673 /* pTHX_ magic can't cope with varargs, so this is a no-context
10674 * version of the main function, (which may itself be aliased to us).
10675 * Don't access this version directly.
10679 Perl_sv_setpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10684 PERL_ARGS_ASSERT_SV_SETPVF_MG_NOCONTEXT;
10686 va_start(args, pat);
10687 sv_vsetpvf_mg(sv, pat, &args);
10693 =for apidoc sv_setpvf
10695 Works like C<sv_catpvf> but copies the text into the SV instead of
10696 appending it. Does not handle 'set' magic. See C<L</sv_setpvf_mg>>.
10702 Perl_sv_setpvf(pTHX_ SV *const sv, const char *const pat, ...)
10706 PERL_ARGS_ASSERT_SV_SETPVF;
10708 va_start(args, pat);
10709 sv_vsetpvf(sv, pat, &args);
10714 =for apidoc sv_vsetpvf
10716 Works like C<sv_vcatpvf> but copies the text into the SV instead of
10717 appending it. Does not handle 'set' magic. See C<L</sv_vsetpvf_mg>>.
10719 Usually used via its frontend C<sv_setpvf>.
10725 Perl_sv_vsetpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10727 PERL_ARGS_ASSERT_SV_VSETPVF;
10729 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10733 =for apidoc sv_setpvf_mg
10735 Like C<sv_setpvf>, but also handles 'set' magic.
10741 Perl_sv_setpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10745 PERL_ARGS_ASSERT_SV_SETPVF_MG;
10747 va_start(args, pat);
10748 sv_vsetpvf_mg(sv, pat, &args);
10753 =for apidoc sv_vsetpvf_mg
10755 Like C<sv_vsetpvf>, but also handles 'set' magic.
10757 Usually used via its frontend C<sv_setpvf_mg>.
10763 Perl_sv_vsetpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10765 PERL_ARGS_ASSERT_SV_VSETPVF_MG;
10767 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10771 #if defined(PERL_IMPLICIT_CONTEXT)
10773 /* pTHX_ magic can't cope with varargs, so this is a no-context
10774 * version of the main function, (which may itself be aliased to us).
10775 * Don't access this version directly.
10779 Perl_sv_catpvf_nocontext(SV *const sv, const char *const pat, ...)
10784 PERL_ARGS_ASSERT_SV_CATPVF_NOCONTEXT;
10786 va_start(args, pat);
10787 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10791 /* pTHX_ magic can't cope with varargs, so this is a no-context
10792 * version of the main function, (which may itself be aliased to us).
10793 * Don't access this version directly.
10797 Perl_sv_catpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10802 PERL_ARGS_ASSERT_SV_CATPVF_MG_NOCONTEXT;
10804 va_start(args, pat);
10805 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10812 =for apidoc sv_catpvf
10814 Processes its arguments like C<sv_catpvfn>, and appends the formatted
10815 output to an SV. As with C<sv_catpvfn> called with a non-null C-style
10816 variable argument list, argument reordering is not supported.
10817 If the appended data contains "wide" characters
10818 (including, but not limited to, SVs with a UTF-8 PV formatted with C<%s>,
10819 and characters >255 formatted with C<%c>), the original SV might get
10820 upgraded to UTF-8. Handles 'get' magic, but not 'set' magic. See
10821 C<L</sv_catpvf_mg>>. If the original SV was UTF-8, the pattern should be
10822 valid UTF-8; if the original SV was bytes, the pattern should be too.
10827 Perl_sv_catpvf(pTHX_ SV *const sv, const char *const pat, ...)
10831 PERL_ARGS_ASSERT_SV_CATPVF;
10833 va_start(args, pat);
10834 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10839 =for apidoc sv_vcatpvf
10841 Processes its arguments like C<sv_catpvfn> called with a non-null C-style
10842 variable argument list, and appends the formatted output
10843 to an SV. Does not handle 'set' magic. See C<L</sv_vcatpvf_mg>>.
10845 Usually used via its frontend C<sv_catpvf>.
10851 Perl_sv_vcatpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10853 PERL_ARGS_ASSERT_SV_VCATPVF;
10855 sv_vcatpvfn_flags(sv, pat, strlen(pat), args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10859 =for apidoc sv_catpvf_mg
10861 Like C<sv_catpvf>, but also handles 'set' magic.
10867 Perl_sv_catpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10871 PERL_ARGS_ASSERT_SV_CATPVF_MG;
10873 va_start(args, pat);
10874 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10880 =for apidoc sv_vcatpvf_mg
10882 Like C<sv_vcatpvf>, but also handles 'set' magic.
10884 Usually used via its frontend C<sv_catpvf_mg>.
10890 Perl_sv_vcatpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10892 PERL_ARGS_ASSERT_SV_VCATPVF_MG;
10894 sv_vcatpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10899 =for apidoc sv_vsetpvfn
10901 Works like C<sv_vcatpvfn> but copies the text into the SV instead of
10904 Usually used via one of its frontends C<sv_vsetpvf> and C<sv_vsetpvf_mg>.
10910 Perl_sv_vsetpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10911 va_list *const args, SV **const svargs, const Size_t sv_count, bool *const maybe_tainted)
10913 PERL_ARGS_ASSERT_SV_VSETPVFN;
10916 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, sv_count, maybe_tainted, 0);
10920 /* simplified inline Perl_sv_catpvn_nomg() when you know the SV's SvPOK */
10922 PERL_STATIC_INLINE void
10923 S_sv_catpvn_simple(pTHX_ SV *const sv, const char* const buf, const STRLEN len)
10925 STRLEN const need = len + SvCUR(sv) + 1;
10928 /* can't wrap as both len and SvCUR() are allocated in
10929 * memory and together can't consume all the address space
10931 assert(need > len);
10936 Copy(buf, end, len, char);
10939 SvCUR_set(sv, need - 1);
10944 * Warn of missing argument to sprintf. The value used in place of such
10945 * arguments should be &PL_sv_no; an undefined value would yield
10946 * inappropriate "use of uninit" warnings [perl #71000].
10949 S_warn_vcatpvfn_missing_argument(pTHX) {
10950 if (ckWARN(WARN_MISSING)) {
10951 Perl_warner(aTHX_ packWARN(WARN_MISSING), "Missing argument in %s",
10952 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
10961 Perl_croak(aTHX_ "Integer overflow in format string for %s",
10962 (PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn"));
10966 /* Given an int i from the next arg (if args is true) or an sv from an arg
10967 * (if args is false), try to extract a STRLEN-ranged value from the arg,
10968 * with overflow checking.
10969 * Sets *neg to true if the value was negative (untouched otherwise.
10970 * Returns the absolute value.
10971 * As an extra margin of safety, it croaks if the returned value would
10972 * exceed the maximum value of a STRLEN / 4.
10976 S_sprintf_arg_num_val(pTHX_ va_list *const args, int i, SV *sv, bool *neg)
10990 if (UNLIKELY(SvIsUV(sv))) {
10991 UV uv = SvUV_nomg(sv);
10993 S_croak_overflow();
10997 iv = SvIV_nomg(sv);
11001 S_croak_overflow();
11007 if (iv > (IV)(((STRLEN)~0) / 4))
11008 S_croak_overflow();
11014 /* Returns true if c is in the range '1'..'9'
11015 * Written with the cast so it only needs one conditional test
11017 #define IS_1_TO_9(c) ((U8)(c - '1') <= 8)
11019 /* Read in and return a number. Updates *pattern to point to the char
11020 * following the number. Expects the first char to 1..9.
11021 * Croaks if the number exceeds 1/4 of the maximum value of STRLEN.
11022 * This is a belt-and-braces safety measure to complement any
11023 * overflow/wrap checks done in the main body of sv_vcatpvfn_flags.
11024 * It means that e.g. on a 32-bit system the width/precision can't be more
11025 * than 1G, which seems reasonable.
11029 S_expect_number(pTHX_ const char **const pattern)
11033 PERL_ARGS_ASSERT_EXPECT_NUMBER;
11035 assert(IS_1_TO_9(**pattern));
11037 var = *(*pattern)++ - '0';
11038 while (isDIGIT(**pattern)) {
11039 /* if var * 10 + 9 would exceed 1/4 max strlen, croak */
11040 if (var > ((((STRLEN)~0) / 4 - 9) / 10))
11041 S_croak_overflow();
11042 var = var * 10 + (*(*pattern)++ - '0');
11047 /* Implement a fast "%.0f": given a pointer to the end of a buffer (caller
11048 * ensures it's big enough), back fill it with the rounded integer part of
11049 * nv. Returns ptr to start of string, and sets *len to its length.
11050 * Returns NULL if not convertible.
11054 S_F0convert(NV nv, char *const endbuf, STRLEN *const len)
11056 const int neg = nv < 0;
11059 PERL_ARGS_ASSERT_F0CONVERT;
11061 assert(!Perl_isinfnan(nv));
11064 if (nv != 0.0 && nv < UV_MAX) {
11070 if (uv & 1 && uv == nv)
11071 uv--; /* Round to even */
11074 const unsigned dig = uv % 10;
11076 } while (uv /= 10);
11086 /* XXX maybe_tainted is never assigned to, so the doc above is lying. */
11089 Perl_sv_vcatpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11090 va_list *const args, SV **const svargs, const Size_t sv_count, bool *const maybe_tainted)
11092 PERL_ARGS_ASSERT_SV_VCATPVFN;
11094 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, sv_count, maybe_tainted, SV_GMAGIC|SV_SMAGIC);
11098 /* For the vcatpvfn code, we need a long double target in case
11099 * HAS_LONG_DOUBLE, even without USE_LONG_DOUBLE, so that we can printf
11100 * with long double formats, even without NV being long double. But we
11101 * call the target 'fv' instead of 'nv', since most of the time it is not
11102 * (most compilers these days recognize "long double", even if only as a
11103 * synonym for "double").
11105 #if defined(HAS_LONG_DOUBLE) && LONG_DOUBLESIZE > DOUBLESIZE && \
11106 defined(PERL_PRIgldbl) && !defined(USE_QUADMATH)
11107 # define VCATPVFN_FV_GF PERL_PRIgldbl
11108 # if defined(__VMS) && defined(__ia64) && defined(__IEEE_FLOAT)
11109 /* Work around breakage in OTS$CVT_FLOAT_T_X */
11110 # define VCATPVFN_NV_TO_FV(nv,fv) \
11113 fv = Perl_isnan(_dv) ? LDBL_QNAN : _dv; \
11116 # define VCATPVFN_NV_TO_FV(nv,fv) (fv)=(nv)
11118 typedef long double vcatpvfn_long_double_t;
11120 # define VCATPVFN_FV_GF NVgf
11121 # define VCATPVFN_NV_TO_FV(nv,fv) (fv)=(nv)
11122 typedef NV vcatpvfn_long_double_t;
11125 #ifdef LONGDOUBLE_DOUBLEDOUBLE
11126 /* The first double can be as large as 2**1023, or '1' x '0' x 1023.
11127 * The second double can be as small as 2**-1074, or '0' x 1073 . '1'.
11128 * The sum of them can be '1' . '0' x 2096 . '1', with implied radix point
11129 * after the first 1023 zero bits.
11131 * XXX The 2098 is quite large (262.25 bytes) and therefore some sort
11132 * of dynamically growing buffer might be better, start at just 16 bytes
11133 * (for example) and grow only when necessary. Or maybe just by looking
11134 * at the exponents of the two doubles? */
11135 # define DOUBLEDOUBLE_MAXBITS 2098
11138 /* vhex will contain the values (0..15) of the hex digits ("nybbles"
11139 * of 4 bits); 1 for the implicit 1, and the mantissa bits, four bits
11140 * per xdigit. For the double-double case, this can be rather many.
11141 * The non-double-double-long-double overshoots since all bits of NV
11142 * are not mantissa bits, there are also exponent bits. */
11143 #ifdef LONGDOUBLE_DOUBLEDOUBLE
11144 # define VHEX_SIZE (3+DOUBLEDOUBLE_MAXBITS/4)
11146 # define VHEX_SIZE (1+(NVSIZE * 8)/4)
11149 /* If we do not have a known long double format, (including not using
11150 * long doubles, or long doubles being equal to doubles) then we will
11151 * fall back to the ldexp/frexp route, with which we can retrieve at
11152 * most as many bits as our widest unsigned integer type is. We try
11153 * to get a 64-bit unsigned integer even if we are not using a 64-bit UV.
11155 * (If you want to test the case of UVSIZE == 4, NVSIZE == 8,
11156 * set the MANTISSATYPE to int and the MANTISSASIZE to 4.)
11158 #if defined(HAS_QUAD) && defined(Uquad_t)
11159 # define MANTISSATYPE Uquad_t
11160 # define MANTISSASIZE 8
11162 # define MANTISSATYPE UV
11163 # define MANTISSASIZE UVSIZE
11166 #if defined(DOUBLE_LITTLE_ENDIAN) || defined(LONGDOUBLE_LITTLE_ENDIAN)
11167 # define HEXTRACT_LITTLE_ENDIAN
11168 #elif defined(DOUBLE_BIG_ENDIAN) || defined(LONGDOUBLE_BIG_ENDIAN)
11169 # define HEXTRACT_BIG_ENDIAN
11171 # define HEXTRACT_MIX_ENDIAN
11174 /* S_hextract() is a helper for S_format_hexfp, for extracting
11175 * the hexadecimal values (for %a/%A). The nv is the NV where the value
11176 * are being extracted from (either directly from the long double in-memory
11177 * presentation, or from the uquad computed via frexp+ldexp). frexp also
11178 * is used to update the exponent. The subnormal is set to true
11179 * for IEEE 754 subnormals/denormals (including the x86 80-bit format).
11180 * The vhex is the pointer to the beginning of the output buffer of VHEX_SIZE.
11182 * The tricky part is that S_hextract() needs to be called twice:
11183 * the first time with vend as NULL, and the second time with vend as
11184 * the pointer returned by the first call. What happens is that on
11185 * the first round the output size is computed, and the intended
11186 * extraction sanity checked. On the second round the actual output
11187 * (the extraction of the hexadecimal values) takes place.
11188 * Sanity failures cause fatal failures during both rounds. */
11190 S_hextract(pTHX_ const NV nv, int* exponent, bool *subnormal,
11191 U8* vhex, U8* vend)
11195 int ixmin = 0, ixmax = 0;
11197 /* XXX Inf/NaN are not handled here, since it is
11198 * assumed they are to be output as "Inf" and "NaN". */
11200 /* These macros are just to reduce typos, they have multiple
11201 * repetitions below, but usually only one (or sometimes two)
11202 * of them is really being used. */
11203 /* HEXTRACT_OUTPUT() extracts the high nybble first. */
11204 #define HEXTRACT_OUTPUT_HI(ix) (*v++ = nvp[ix] >> 4)
11205 #define HEXTRACT_OUTPUT_LO(ix) (*v++ = nvp[ix] & 0xF)
11206 #define HEXTRACT_OUTPUT(ix) \
11208 HEXTRACT_OUTPUT_HI(ix); HEXTRACT_OUTPUT_LO(ix); \
11210 #define HEXTRACT_COUNT(ix, c) \
11212 v += c; if (ix < ixmin) ixmin = ix; else if (ix > ixmax) ixmax = ix; \
11214 #define HEXTRACT_BYTE(ix) \
11216 if (vend) HEXTRACT_OUTPUT(ix); else HEXTRACT_COUNT(ix, 2); \
11218 #define HEXTRACT_LO_NYBBLE(ix) \
11220 if (vend) HEXTRACT_OUTPUT_LO(ix); else HEXTRACT_COUNT(ix, 1); \
11222 /* HEXTRACT_TOP_NYBBLE is just convenience disguise,
11223 * to make it look less odd when the top bits of a NV
11224 * are extracted using HEXTRACT_LO_NYBBLE: the highest
11225 * order bits can be in the "low nybble" of a byte. */
11226 #define HEXTRACT_TOP_NYBBLE(ix) HEXTRACT_LO_NYBBLE(ix)
11227 #define HEXTRACT_BYTES_LE(a, b) \
11228 for (ix = a; ix >= b; ix--) { HEXTRACT_BYTE(ix); }
11229 #define HEXTRACT_BYTES_BE(a, b) \
11230 for (ix = a; ix <= b; ix++) { HEXTRACT_BYTE(ix); }
11231 #define HEXTRACT_GET_SUBNORMAL(nv) *subnormal = Perl_fp_class_denorm(nv)
11232 #define HEXTRACT_IMPLICIT_BIT(nv) \
11234 if (!*subnormal) { \
11235 if (vend) *v++ = ((nv) == 0.0) ? 0 : 1; else v++; \
11239 /* Most formats do. Those which don't should undef this.
11241 * But also note that IEEE 754 subnormals do not have it, or,
11242 * expressed alternatively, their implicit bit is zero. */
11243 #define HEXTRACT_HAS_IMPLICIT_BIT
11245 /* Many formats do. Those which don't should undef this. */
11246 #define HEXTRACT_HAS_TOP_NYBBLE
11248 /* HEXTRACTSIZE is the maximum number of xdigits. */
11249 #if defined(USE_LONG_DOUBLE) && defined(LONGDOUBLE_DOUBLEDOUBLE)
11250 # define HEXTRACTSIZE (2+DOUBLEDOUBLE_MAXBITS/4)
11252 # define HEXTRACTSIZE 2 * NVSIZE
11255 const U8* vmaxend = vhex + HEXTRACTSIZE;
11257 assert(HEXTRACTSIZE <= VHEX_SIZE);
11259 PERL_UNUSED_VAR(ix); /* might happen */
11260 (void)Perl_frexp(PERL_ABS(nv), exponent);
11261 *subnormal = FALSE;
11262 if (vend && (vend <= vhex || vend > vmaxend)) {
11263 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11264 Perl_croak(aTHX_ "Hexadecimal float: internal error (entry)");
11267 /* First check if using long doubles. */
11268 #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE)
11269 # if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN
11270 /* Used in e.g. VMS and HP-UX IA-64, e.g. -0.1L:
11271 * 9a 99 99 99 99 99 99 99 99 99 99 99 99 99 fb bf */
11272 /* The bytes 13..0 are the mantissa/fraction,
11273 * the 15,14 are the sign+exponent. */
11274 const U8* nvp = (const U8*)(&nv);
11275 HEXTRACT_GET_SUBNORMAL(nv);
11276 HEXTRACT_IMPLICIT_BIT(nv);
11277 # undef HEXTRACT_HAS_TOP_NYBBLE
11278 HEXTRACT_BYTES_LE(13, 0);
11279 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_BIG_ENDIAN
11280 /* Used in e.g. Solaris Sparc and HP-UX PA-RISC, e.g. -0.1L:
11281 * bf fb 99 99 99 99 99 99 99 99 99 99 99 99 99 9a */
11282 /* The bytes 2..15 are the mantissa/fraction,
11283 * the 0,1 are the sign+exponent. */
11284 const U8* nvp = (const U8*)(&nv);
11285 HEXTRACT_GET_SUBNORMAL(nv);
11286 HEXTRACT_IMPLICIT_BIT(nv);
11287 # undef HEXTRACT_HAS_TOP_NYBBLE
11288 HEXTRACT_BYTES_BE(2, 15);
11289 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN
11290 /* x86 80-bit "extended precision", 64 bits of mantissa / fraction /
11291 * significand, 15 bits of exponent, 1 bit of sign. No implicit bit.
11292 * NVSIZE can be either 12 (ILP32, Solaris x86) or 16 (LP64, Linux
11293 * and OS X), meaning that 2 or 6 bytes are empty padding. */
11294 /* The bytes 0..1 are the sign+exponent,
11295 * the bytes 2..9 are the mantissa/fraction. */
11296 const U8* nvp = (const U8*)(&nv);
11297 # undef HEXTRACT_HAS_IMPLICIT_BIT
11298 # undef HEXTRACT_HAS_TOP_NYBBLE
11299 HEXTRACT_GET_SUBNORMAL(nv);
11300 HEXTRACT_BYTES_LE(7, 0);
11301 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN
11302 /* Does this format ever happen? (Wikipedia says the Motorola
11303 * 6888x math coprocessors used format _like_ this but padded
11304 * to 96 bits with 16 unused bits between the exponent and the
11306 const U8* nvp = (const U8*)(&nv);
11307 # undef HEXTRACT_HAS_IMPLICIT_BIT
11308 # undef HEXTRACT_HAS_TOP_NYBBLE
11309 HEXTRACT_GET_SUBNORMAL(nv);
11310 HEXTRACT_BYTES_BE(0, 7);
11312 # define HEXTRACT_FALLBACK
11313 /* Double-double format: two doubles next to each other.
11314 * The first double is the high-order one, exactly like
11315 * it would be for a "lone" double. The second double
11316 * is shifted down using the exponent so that that there
11317 * are no common bits. The tricky part is that the value
11318 * of the double-double is the SUM of the two doubles and
11319 * the second one can be also NEGATIVE.
11321 * Because of this tricky construction the bytewise extraction we
11322 * use for the other long double formats doesn't work, we must
11323 * extract the values bit by bit.
11325 * The little-endian double-double is used .. somewhere?
11327 * The big endian double-double is used in e.g. PPC/Power (AIX)
11330 * The mantissa bits are in two separate stretches, e.g. for -0.1L:
11331 * 9a 99 99 99 99 99 59 bc 9a 99 99 99 99 99 b9 3f (LE)
11332 * 3f b9 99 99 99 99 99 9a bc 59 99 99 99 99 99 9a (BE)
11335 #else /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) */
11336 /* Using normal doubles, not long doubles.
11338 * We generate 4-bit xdigits (nybble/nibble) instead of 8-bit
11339 * bytes, since we might need to handle printf precision, and
11340 * also need to insert the radix. */
11342 # ifdef HEXTRACT_LITTLE_ENDIAN
11343 /* 0 1 2 3 4 5 6 7 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11344 const U8* nvp = (const U8*)(&nv);
11345 HEXTRACT_GET_SUBNORMAL(nv);
11346 HEXTRACT_IMPLICIT_BIT(nv);
11347 HEXTRACT_TOP_NYBBLE(6);
11348 HEXTRACT_BYTES_LE(5, 0);
11349 # elif defined(HEXTRACT_BIG_ENDIAN)
11350 /* 7 6 5 4 3 2 1 0 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11351 const U8* nvp = (const U8*)(&nv);
11352 HEXTRACT_GET_SUBNORMAL(nv);
11353 HEXTRACT_IMPLICIT_BIT(nv);
11354 HEXTRACT_TOP_NYBBLE(1);
11355 HEXTRACT_BYTES_BE(2, 7);
11356 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_LE_BE
11357 /* 4 5 6 7 0 1 2 3 (MSB = 7, LSB = 0, 6:7 = nybble:exponent:sign) */
11358 const U8* nvp = (const U8*)(&nv);
11359 HEXTRACT_GET_SUBNORMAL(nv);
11360 HEXTRACT_IMPLICIT_BIT(nv);
11361 HEXTRACT_TOP_NYBBLE(2); /* 6 */
11362 HEXTRACT_BYTE(1); /* 5 */
11363 HEXTRACT_BYTE(0); /* 4 */
11364 HEXTRACT_BYTE(7); /* 3 */
11365 HEXTRACT_BYTE(6); /* 2 */
11366 HEXTRACT_BYTE(5); /* 1 */
11367 HEXTRACT_BYTE(4); /* 0 */
11368 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_BE_LE
11369 /* 3 2 1 0 7 6 5 4 (MSB = 7, LSB = 0, 7:6 = sign:exponent:nybble) */
11370 const U8* nvp = (const U8*)(&nv);
11371 HEXTRACT_GET_SUBNORMAL(nv);
11372 HEXTRACT_IMPLICIT_BIT(nv);
11373 HEXTRACT_TOP_NYBBLE(5); /* 6 */
11374 HEXTRACT_BYTE(6); /* 5 */
11375 HEXTRACT_BYTE(7); /* 4 */
11376 HEXTRACT_BYTE(0); /* 3 */
11377 HEXTRACT_BYTE(1); /* 2 */
11378 HEXTRACT_BYTE(2); /* 1 */
11379 HEXTRACT_BYTE(3); /* 0 */
11381 # define HEXTRACT_FALLBACK
11384 # define HEXTRACT_FALLBACK
11386 #endif /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) #else */
11388 #ifdef HEXTRACT_FALLBACK
11389 HEXTRACT_GET_SUBNORMAL(nv);
11390 # undef HEXTRACT_HAS_TOP_NYBBLE /* Meaningless, but consistent. */
11391 /* The fallback is used for the double-double format, and
11392 * for unknown long double formats, and for unknown double
11393 * formats, or in general unknown NV formats. */
11394 if (nv == (NV)0.0) {
11402 NV d = nv < 0 ? -nv : nv;
11404 U8 ha = 0x0; /* hexvalue accumulator */
11405 U8 hd = 0x8; /* hexvalue digit */
11407 /* Shift d and e (and update exponent) so that e <= d < 2*e,
11408 * this is essentially manual frexp(). Multiplying by 0.5 and
11409 * doubling should be lossless in binary floating point. */
11419 while (d >= e + e) {
11423 /* Now e <= d < 2*e */
11425 /* First extract the leading hexdigit (the implicit bit). */
11441 /* Then extract the remaining hexdigits. */
11442 while (d > (NV)0.0) {
11448 /* Output or count in groups of four bits,
11449 * that is, when the hexdigit is down to one. */
11454 /* Reset the hexvalue. */
11463 /* Flush possible pending hexvalue. */
11473 /* Croak for various reasons: if the output pointer escaped the
11474 * output buffer, if the extraction index escaped the extraction
11475 * buffer, or if the ending output pointer didn't match the
11476 * previously computed value. */
11477 if (v <= vhex || v - vhex >= VHEX_SIZE ||
11478 /* For double-double the ixmin and ixmax stay at zero,
11479 * which is convenient since the HEXTRACTSIZE is tricky
11480 * for double-double. */
11481 ixmin < 0 || ixmax >= NVSIZE ||
11482 (vend && v != vend)) {
11483 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11484 Perl_croak(aTHX_ "Hexadecimal float: internal error (overflow)");
11490 /* S_format_hexfp(): helper function for Perl_sv_vcatpvfn_flags().
11492 * Processes the %a/%A hexadecimal floating-point format, since the
11493 * built-in snprintf()s which are used for most of the f/p formats, don't
11494 * universally handle %a/%A.
11495 * Populates buf of length bufsize, and returns the length of the created
11497 * The rest of the args have the same meaning as the local vars of the
11498 * same name within Perl_sv_vcatpvfn_flags().
11500 * It assumes the caller has already done STORE_LC_NUMERIC_SET_TO_NEEDED();
11502 * It requires the caller to make buf large enough.
11506 S_format_hexfp(pTHX_ char * const buf, const STRLEN bufsize, const char c,
11507 const NV nv, const vcatpvfn_long_double_t fv,
11508 bool has_precis, STRLEN precis, STRLEN width,
11509 bool alt, char plus, bool left, bool fill)
11511 /* Hexadecimal floating point. */
11513 U8 vhex[VHEX_SIZE];
11514 U8* v = vhex; /* working pointer to vhex */
11515 U8* vend; /* pointer to one beyond last digit of vhex */
11516 U8* vfnz = NULL; /* first non-zero */
11517 U8* vlnz = NULL; /* last non-zero */
11518 U8* v0 = NULL; /* first output */
11519 const bool lower = (c == 'a');
11520 /* At output the values of vhex (up to vend) will
11521 * be mapped through the xdig to get the actual
11522 * human-readable xdigits. */
11523 const char* xdig = PL_hexdigit;
11524 STRLEN zerotail = 0; /* how many extra zeros to append */
11525 int exponent = 0; /* exponent of the floating point input */
11526 bool hexradix = FALSE; /* should we output the radix */
11527 bool subnormal = FALSE; /* IEEE 754 subnormal/denormal */
11528 bool negative = FALSE;
11531 /* XXX: NaN, Inf -- though they are printed as "NaN" and "Inf".
11533 * For example with denormals, (assuming the vanilla
11534 * 64-bit double): the exponent is zero. 1xp-1074 is
11535 * the smallest denormal and the smallest double, it
11536 * could be output also as 0x0.0000000000001p-1022 to
11537 * match its internal structure. */
11539 vend = S_hextract(aTHX_ nv, &exponent, &subnormal, vhex, NULL);
11540 S_hextract(aTHX_ nv, &exponent, &subnormal, vhex, vend);
11542 #if NVSIZE > DOUBLESIZE
11543 # ifdef HEXTRACT_HAS_IMPLICIT_BIT
11544 /* In this case there is an implicit bit,
11545 * and therefore the exponent is shifted by one. */
11547 # elif defined(NV_X86_80_BIT)
11549 /* The subnormals of the x86-80 have a base exponent of -16382,
11550 * (while the physical exponent bits are zero) but the frexp()
11551 * returned the scientific-style floating exponent. We want
11552 * to map the last one as:
11553 * -16831..-16384 -> -16382 (the last normal is 0x1p-16382)
11554 * -16835..-16388 -> -16384
11555 * since we want to keep the first hexdigit
11556 * as one of the [8421]. */
11557 exponent = -4 * ( (exponent + 1) / -4) - 2;
11561 /* TBD: other non-implicit-bit platforms than the x86-80. */
11565 negative = fv < 0 || Perl_signbit(nv);
11576 xdig += 16; /* Use uppercase hex. */
11579 /* Find the first non-zero xdigit. */
11580 for (v = vhex; v < vend; v++) {
11588 /* Find the last non-zero xdigit. */
11589 for (v = vend - 1; v >= vhex; v--) {
11596 #if NVSIZE == DOUBLESIZE
11602 #ifndef NV_X86_80_BIT
11604 /* IEEE 754 subnormals (but not the x86 80-bit):
11605 * we want "normalize" the subnormal,
11606 * so we need to right shift the hex nybbles
11607 * so that the output of the subnormal starts
11608 * from the first true bit. (Another, equally
11609 * valid, policy would be to dump the subnormal
11610 * nybbles as-is, to display the "physical" layout.) */
11613 /* Find the ceil(log2(v[0])) of
11614 * the top non-zero nybble. */
11615 for (i = vfnz[0], n = 0; i > 1; i >>= 1, n++) { }
11619 for (vshr = vlnz; vshr >= vfnz; vshr--) {
11620 vshr[1] |= (vshr[0] & (0xF >> (4 - n))) << (4 - n);
11634 U8* ve = (subnormal ? vlnz + 1 : vend);
11635 SSize_t vn = ve - v0;
11637 if (precis < (Size_t)(vn - 1)) {
11638 bool overflow = FALSE;
11639 if (v0[precis + 1] < 0x8) {
11640 /* Round down, nothing to do. */
11641 } else if (v0[precis + 1] > 0x8) {
11644 overflow = v0[precis] > 0xF;
11646 } else { /* v0[precis] == 0x8 */
11647 /* Half-point: round towards the one
11648 * with the even least-significant digit:
11656 * 78 -> 8 f8 -> 10 */
11657 if ((v0[precis] & 0x1)) {
11660 overflow = v0[precis] > 0xF;
11665 for (v = v0 + precis - 1; v >= v0; v--) {
11667 overflow = *v > 0xF;
11673 if (v == v0 - 1 && overflow) {
11674 /* If the overflow goes all the
11675 * way to the front, we need to
11676 * insert 0x1 in front, and adjust
11678 Move(v0, v0 + 1, vn - 1, char);
11684 /* The new effective "last non zero". */
11685 vlnz = v0 + precis;
11689 subnormal ? precis - vn + 1 :
11690 precis - (vlnz - vhex);
11697 /* If there are non-zero xdigits, the radix
11698 * is output after the first one. */
11709 /* The radix is always output if precis, or if alt. */
11710 if (precis > 0 || alt) {
11715 #ifndef USE_LOCALE_NUMERIC
11718 if (IN_LC(LC_NUMERIC)) {
11720 const char* r = SvPV(PL_numeric_radix_sv, n);
11721 Copy(r, p, n, char);
11735 if (zerotail > 0) {
11736 while (zerotail--) {
11743 /* sanity checks */
11744 if (elen >= bufsize || width >= bufsize)
11745 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11746 Perl_croak(aTHX_ "Hexadecimal float: internal error (overflow)");
11748 elen += my_snprintf(p, bufsize - elen,
11749 "%c%+d", lower ? 'p' : 'P',
11752 if (elen < width) {
11753 STRLEN gap = (STRLEN)(width - elen);
11755 /* Pad the back with spaces. */
11756 memset(buf + elen, ' ', gap);
11759 /* Insert the zeros after the "0x" and the
11760 * the potential sign, but before the digits,
11761 * otherwise we end up with "0000xH.HHH...",
11762 * when we want "0x000H.HHH..." */
11763 STRLEN nzero = gap;
11764 char* zerox = buf + 2;
11765 STRLEN nmove = elen - 2;
11766 if (negative || plus) {
11770 Move(zerox, zerox + nzero, nmove, char);
11771 memset(zerox, fill ? '0' : ' ', nzero);
11774 /* Move it to the right. */
11775 Move(buf, buf + gap,
11777 /* Pad the front with spaces. */
11778 memset(buf, ' ', gap);
11787 =for apidoc sv_vcatpvfn
11789 =for apidoc sv_vcatpvfn_flags
11791 Processes its arguments like C<vsprintf> and appends the formatted output
11792 to an SV. Uses an array of SVs if the C-style variable argument list is
11793 missing (C<NULL>). Argument reordering (using format specifiers like C<%2$d>
11794 or C<%*2$d>) is supported only when using an array of SVs; using a C-style
11795 C<va_list> argument list with a format string that uses argument reordering
11796 will yield an exception.
11798 When running with taint checks enabled, indicates via
11799 C<maybe_tainted> if results are untrustworthy (often due to the use of
11802 If called as C<sv_vcatpvfn> or flags has the C<SV_GMAGIC> bit set, calls get magic.
11804 It assumes that pat has the same utf8-ness as sv. It's the caller's
11805 responsibility to ensure that this is so.
11807 Usually used via one of its frontends C<sv_vcatpvf> and C<sv_vcatpvf_mg>.
11814 Perl_sv_vcatpvfn_flags(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11815 va_list *const args, SV **const svargs, const Size_t sv_count, bool *const maybe_tainted,
11818 const char *fmtstart; /* character following the current '%' */
11819 const char *q; /* current position within format */
11820 const char *patend;
11823 static const char nullstr[] = "(null)";
11824 bool has_utf8 = DO_UTF8(sv); /* has the result utf8? */
11825 const bool pat_utf8 = has_utf8; /* the pattern is in utf8? */
11826 /* Times 4: a decimal digit takes more than 3 binary digits.
11827 * NV_DIG: mantissa takes that many decimal digits.
11828 * Plus 32: Playing safe. */
11829 char ebuf[IV_DIG * 4 + NV_DIG + 32];
11830 bool no_redundant_warning = FALSE; /* did we use any explicit format parameter index? */
11831 #ifdef USE_LOCALE_NUMERIC
11832 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
11833 bool lc_numeric_set = FALSE; /* called STORE_LC_NUMERIC_SET_TO_NEEDED? */
11836 PERL_ARGS_ASSERT_SV_VCATPVFN_FLAGS;
11837 PERL_UNUSED_ARG(maybe_tainted);
11839 if (flags & SV_GMAGIC)
11842 /* no matter what, this is a string now */
11843 (void)SvPV_force_nomg(sv, origlen);
11845 /* the code that scans for flags etc following a % relies on
11846 * a '\0' being present to avoid falling off the end. Ideally that
11847 * should be fixed */
11848 assert(pat[patlen] == '\0');
11851 /* Special-case "", "%s", "%-p" (SVf - see below) and "%.0f".
11852 * In each case, if there isn't the correct number of args, instead
11853 * fall through to the main code to handle the issuing of any
11857 if (patlen == 0 && (args || sv_count == 0))
11860 if (patlen <= 4 && pat[0] == '%' && (args || sv_count == 1)) {
11863 if (patlen == 2 && pat[1] == 's') {
11865 const char * const s = va_arg(*args, char*);
11866 sv_catpv_nomg(sv, s ? s : nullstr);
11869 /* we want get magic on the source but not the target.
11870 * sv_catsv can't do that, though */
11871 SvGETMAGIC(*svargs);
11872 sv_catsv_nomg(sv, *svargs);
11879 if (patlen == 3 && pat[1] == '-' && pat[2] == 'p') {
11880 SV *asv = MUTABLE_SV(va_arg(*args, void*));
11881 sv_catsv_nomg(sv, asv);
11885 #if !defined(USE_LONG_DOUBLE) && !defined(USE_QUADMATH)
11886 /* special-case "%.0f" */
11887 else if ( patlen == 4
11888 && pat[1] == '.' && pat[2] == '0' && pat[3] == 'f')
11890 const NV nv = SvNV(*svargs);
11891 if (LIKELY(!Perl_isinfnan(nv))) {
11895 if ((p = F0convert(nv, ebuf + sizeof ebuf, &l))) {
11896 sv_catpvn_nomg(sv, p, l);
11901 #endif /* !USE_LONG_DOUBLE */
11905 patend = (char*)pat + patlen;
11906 for (fmtstart = pat; fmtstart < patend; fmtstart = q) {
11907 char intsize = 0; /* size qualifier in "%hi..." etc */
11908 bool alt = FALSE; /* has "%#..." */
11909 bool left = FALSE; /* has "%-..." */
11910 bool fill = FALSE; /* has "%0..." */
11911 char plus = 0; /* has "%+..." */
11912 STRLEN width = 0; /* value of "%NNN..." */
11913 bool has_precis = FALSE; /* has "%.NNN..." */
11914 STRLEN precis = 0; /* value of "%.NNN..." */
11915 int base = 0; /* base to print in, e.g. 8 for %o */
11916 UV uv = 0; /* the value to print of int-ish args */
11918 bool vectorize = FALSE; /* has "%v..." */
11919 bool vec_utf8 = FALSE; /* SvUTF8(vec arg) */
11920 const U8 *vecstr = NULL; /* SvPVX(vec arg) */
11921 STRLEN veclen = 0; /* SvCUR(vec arg) */
11922 const char *dotstr = NULL; /* separator string for %v */
11923 STRLEN dotstrlen; /* length of separator string for %v */
11925 Size_t efix = 0; /* explicit format parameter index */
11926 const Size_t osvix = svix; /* original index in case of bad fmt */
11929 bool is_utf8 = FALSE; /* is this item utf8? */
11930 bool arg_missing = FALSE; /* give "Missing argument" warning */
11931 char esignbuf[4]; /* holds sign prefix, e.g. "-0x" */
11932 STRLEN esignlen = 0; /* length of e.g. "-0x" */
11933 STRLEN zeros = 0; /* how many '0' to prepend */
11935 const char *eptr = NULL; /* the address of the element string */
11936 STRLEN elen = 0; /* the length of the element string */
11938 char c; /* the actual format ('d', s' etc) */
11941 /* echo everything up to the next format specification */
11942 for (q = fmtstart; q < patend && *q != '%'; ++q)
11945 if (q > fmtstart) {
11946 if (has_utf8 && !pat_utf8) {
11947 /* upgrade and copy the bytes of fmtstart..q-1 to utf8 on
11951 STRLEN need = SvCUR(sv) + (q - fmtstart) + 1;
11953 for (p = fmtstart; p < q; p++)
11954 if (!NATIVE_BYTE_IS_INVARIANT(*p))
11959 for (p = fmtstart; p < q; p++)
11960 append_utf8_from_native_byte((U8)*p, (U8**)&dst);
11962 SvCUR_set(sv, need - 1);
11965 S_sv_catpvn_simple(aTHX_ sv, fmtstart, q - fmtstart);
11970 fmtstart = q; /* fmtstart is char following the '%' */
11973 We allow format specification elements in this order:
11974 \d+\$ explicit format parameter index
11976 v|\*(\d+\$)?v vector with optional (optionally specified) arg
11977 0 flag (as above): repeated to allow "v02"
11978 \d+|\*(\d+\$)? width using optional (optionally specified) arg
11979 \.(\d*|\*(\d+\$)?) precision using optional (optionally specified) arg
11981 [%bcdefginopsuxDFOUX] format (mandatory)
11984 if (IS_1_TO_9(*q)) {
11985 width = expect_number(&q);
11988 Perl_croak_nocontext(
11989 "Cannot yet reorder sv_catpvfn() arguments from va_list");
11991 efix = (Size_t)width;
11993 no_redundant_warning = TRUE;
12005 if (plus == '+' && *q == ' ') /* '+' over ' ' */
12032 /* at this point we can expect one of:
12034 * 123 an explicit width
12035 * * width taken from next arg
12036 * *12$ width taken from 12th arg
12039 * But any width specification may be preceded by a v, in one of its
12044 * So an asterisk may be either a width specifier or a vector
12045 * separator arg specifier, and we don't know which initially
12050 STRLEN ix; /* explicit width/vector separator index */
12052 if (IS_1_TO_9(*q)) {
12053 ix = expect_number(&q);
12056 Perl_croak_nocontext(
12057 "Cannot yet reorder sv_catpvfn() arguments from va_list");
12058 no_redundant_warning = TRUE;
12067 /* The asterisk was for *v, *NNN$v: vectorizing, but not
12068 * with the default "." */
12073 vecsv = va_arg(*args, SV*);
12075 ix = ix ? ix - 1 : svix++;
12076 vecsv = ix < sv_count ? svargs[ix]
12077 : (arg_missing = TRUE, &PL_sv_no);
12079 dotstr = SvPV_const(vecsv, dotstrlen);
12080 /* Keep the DO_UTF8 test *after* the SvPV call, else things go
12081 bad with tied or overloaded values that return UTF8. */
12082 if (DO_UTF8(vecsv))
12084 else if (has_utf8) {
12085 vecsv = sv_mortalcopy(vecsv);
12086 sv_utf8_upgrade(vecsv);
12087 dotstr = SvPV_const(vecsv, dotstrlen);
12094 /* the asterisk specified a width */
12099 i = va_arg(*args, int);
12101 ix = ix ? ix - 1 : svix++;
12102 sv = (ix < sv_count) ? svargs[ix]
12103 : (arg_missing = TRUE, (SV*)NULL);
12105 width = S_sprintf_arg_num_val(aTHX_ args, i, sv, &left);
12108 else if (*q == 'v') {
12119 /* explicit width? */
12125 width = expect_number(&q);
12135 STRLEN ix; /* explicit precision index */
12137 if (IS_1_TO_9(*q)) {
12138 ix = expect_number(&q);
12141 Perl_croak_nocontext(
12142 "Cannot yet reorder sv_catpvfn() arguments from va_list");
12143 no_redundant_warning = TRUE;
12156 i = va_arg(*args, int);
12158 ix = ix ? ix - 1 : svix++;
12159 sv = (ix < sv_count) ? svargs[ix]
12160 : (arg_missing = TRUE, (SV*)NULL);
12162 precis = S_sprintf_arg_num_val(aTHX_ args, i, sv, &neg);
12167 /* although it doesn't seem documented, this code has long
12169 * no digits following the '.' is treated like '.0'
12170 * the number may be preceded by any number of zeroes,
12171 * e.g. "%.0001f", which is the same as "%.1f"
12172 * so I've kept that behaviour. DAPM May 2017
12176 precis = IS_1_TO_9(*q) ? expect_number(&q) : 0;
12185 case 'I': /* Ix, I32x, and I64x */
12186 # ifdef USE_64_BIT_INT
12187 if (q[1] == '6' && q[2] == '4') {
12193 if (q[1] == '3' && q[2] == '2') {
12197 # ifdef USE_64_BIT_INT
12203 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
12204 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
12207 # ifdef USE_QUADMATH
12220 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
12221 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
12222 if (*q == 'l') { /* lld, llf */
12231 if (*++q == 'h') { /* hhd, hhu */
12248 c = *q++; /* c now holds the conversion type */
12250 /* '%' doesn't have an arg, so skip arg processing */
12259 if (vectorize && !strchr("BbDdiOouUXx", c))
12262 /* get next arg (individual branches do their own va_arg()
12263 * handling for the args case) */
12266 efix = efix ? efix - 1 : svix++;
12267 argsv = efix < sv_count ? svargs[efix]
12268 : (arg_missing = TRUE, &PL_sv_no);
12278 eptr = va_arg(*args, char*);
12281 elen = my_strnlen(eptr, precis);
12283 elen = strlen(eptr);
12285 eptr = (char *)nullstr;
12286 elen = sizeof nullstr - 1;
12290 eptr = SvPV_const(argsv, elen);
12291 if (DO_UTF8(argsv)) {
12292 STRLEN old_precis = precis;
12293 if (has_precis && precis < elen) {
12294 STRLEN ulen = sv_or_pv_len_utf8(argsv, eptr, elen);
12295 STRLEN p = precis > ulen ? ulen : precis;
12296 precis = sv_or_pv_pos_u2b(argsv, eptr, p, 0);
12297 /* sticks at end */
12299 if (width) { /* fudge width (can't fudge elen) */
12300 if (has_precis && precis < elen)
12301 width += precis - old_precis;
12304 elen - sv_or_pv_len_utf8(argsv,eptr,elen);
12311 if (has_precis && precis < elen)
12323 * "%...p" is normally treated like "%...x", except that the
12324 * number to print is the SV's address (or a pointer address
12325 * for C-ish sprintf).
12327 * However, the C-ish sprintf variant allows a few special
12328 * extensions. These are currently:
12330 * %-p (SVf) Like %s, but gets the string from an SV*
12331 * arg rather than a char* arg.
12332 * (This was previously %_).
12334 * %-<num>p Ditto but like %.<num>s (i.e. num is max width)
12336 * %2p (HEKf) Like %s, but using the key string in a HEK
12338 * %3p (HEKf256) Ditto but like %.256s
12340 * %d%lu%4p (UTF8f) A utf8 string. Consumes 3 args:
12341 * (cBOOL(utf8), len, string_buf).
12342 * It's handled by the "case 'd'" branch
12343 * rather than here.
12345 * %<num>p where num is 1 or > 4: reserved for future
12346 * extensions. Warns, but then is treated as a
12347 * general %p (print hex address) format.
12355 /* not %*p or %*1$p - any width was explicit */
12359 if (left) { /* %-p (SVf), %-NNNp */
12364 argsv = MUTABLE_SV(va_arg(*args, void*));
12365 eptr = SvPV_const(argsv, elen);
12366 if (DO_UTF8(argsv))
12371 else if (width == 2 || width == 3) { /* HEKf, HEKf256 */
12372 HEK * const hek = va_arg(*args, HEK *);
12373 eptr = HEK_KEY(hek);
12374 elen = HEK_LEN(hek);
12385 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
12386 "internal %%<num>p might conflict with future printf extensions");
12390 /* treat as normal %...p */
12392 uv = PTR2UV(args ? va_arg(*args, void*) : argsv);
12397 /* Ignore any size specifiers, since they're not documented as
12398 * being allowed for %c (ideally we should warn on e.g. '%hc').
12399 * Setting a default intsize, along with a positive
12400 * (which signals unsigned) base, causes, for C-ish use, the
12401 * va_arg to be interpreted as as unsigned int, when it's
12402 * actually signed, which will convert -ve values to high +ve
12403 * values. Note that unlike the libc %c, values > 255 will
12404 * convert to high unicode points rather than being truncated
12405 * to 8 bits. For perlish use, it will do SvUV(argsv), which
12406 * will again convert -ve args to high -ve values.
12409 base = 1; /* special value that indicates we're doing a 'c' */
12410 goto get_int_arg_val;
12419 goto get_int_arg_val;
12422 /* probably just a plain %d, but it might be the start of the
12423 * special UTF8f format, which usually looks something like
12424 * "%d%lu%4p" (the lu may vary by platform)
12426 assert((UTF8f)[0] == 'd');
12427 assert((UTF8f)[1] == '%');
12429 if ( args /* UTF8f only valid for C-ish sprintf */
12430 && q == fmtstart + 1 /* plain %d, not %....d */
12431 && patend >= fmtstart + sizeof(UTF8f) - 1 /* long enough */
12433 && strnEQ(q + 1, UTF8f + 2, sizeof(UTF8f) - 3))
12435 /* The argument has already gone through cBOOL, so the cast
12437 is_utf8 = (bool)va_arg(*args, int);
12438 elen = va_arg(*args, UV);
12439 /* if utf8 length is larger than 0x7ffff..., then it might
12440 * have been a signed value that wrapped */
12441 if (elen > ((~(STRLEN)0) >> 1)) {
12442 assert(0); /* in DEBUGGING build we want to crash */
12443 elen = 0; /* otherwise we want to treat this as an empty string */
12445 eptr = va_arg(*args, char *);
12446 q += sizeof(UTF8f) - 2;
12453 goto get_int_arg_val;
12464 goto get_int_arg_val;
12469 goto get_int_arg_val;
12480 goto get_int_arg_val;
12495 esignbuf[esignlen++] = plus;
12498 /* initialise the vector string to iterate over */
12500 vecsv = args ? va_arg(*args, SV*) : argsv;
12502 /* if this is a version object, we need to convert
12503 * back into v-string notation and then let the
12504 * vectorize happen normally
12506 if (sv_isobject(vecsv) && sv_derived_from(vecsv, "version")) {
12507 if ( hv_existss(MUTABLE_HV(SvRV(vecsv)), "alpha") ) {
12508 Perl_ck_warner_d(aTHX_ packWARN(WARN_PRINTF),
12509 "vector argument not supported with alpha versions");
12513 vecstr = (U8*)SvPV_const(vecsv,veclen);
12514 vecsv = sv_newmortal();
12515 scan_vstring((char *)vecstr, (char *)vecstr + veclen,
12519 vecstr = (U8*)SvPV_const(vecsv, veclen);
12520 vec_utf8 = DO_UTF8(vecsv);
12522 /* This is the re-entry point for when we're iterating
12523 * over the individual characters of a vector arg */
12526 goto done_valid_conversion;
12528 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
12538 /* test arg for inf/nan. This can trigger an unwanted
12539 * 'str' overload, so manually force 'num' overload first
12543 if (UNLIKELY(SvAMAGIC(argsv)))
12544 argsv = sv_2num(argsv);
12545 if (UNLIKELY(isinfnansv(argsv)))
12546 goto handle_infnan_argsv;
12550 /* signed int type */
12555 case 'c': iv = (char)va_arg(*args, int); break;
12556 case 'h': iv = (short)va_arg(*args, int); break;
12557 case 'l': iv = va_arg(*args, long); break;
12558 case 'V': iv = va_arg(*args, IV); break;
12559 case 'z': iv = va_arg(*args, SSize_t); break;
12560 #ifdef HAS_PTRDIFF_T
12561 case 't': iv = va_arg(*args, ptrdiff_t); break;
12563 default: iv = va_arg(*args, int); break;
12564 case 'j': iv = va_arg(*args, PERL_INTMAX_T); break;
12567 iv = va_arg(*args, Quad_t); break;
12574 /* assign to tiv then cast to iv to work around
12575 * 2003 GCC cast bug (gnu.org bugzilla #13488) */
12576 IV tiv = SvIV_nomg(argsv);
12578 case 'c': iv = (char)tiv; break;
12579 case 'h': iv = (short)tiv; break;
12580 case 'l': iv = (long)tiv; break;
12582 default: iv = tiv; break;
12585 iv = (Quad_t)tiv; break;
12592 /* now convert iv to uv */
12596 esignbuf[esignlen++] = plus;
12599 uv = (iv == IV_MIN) ? (UV)iv : (UV)(-iv);
12600 esignbuf[esignlen++] = '-';
12604 /* unsigned int type */
12607 case 'c': uv = (unsigned char)va_arg(*args, unsigned);
12609 case 'h': uv = (unsigned short)va_arg(*args, unsigned);
12611 case 'l': uv = va_arg(*args, unsigned long); break;
12612 case 'V': uv = va_arg(*args, UV); break;
12613 case 'z': uv = va_arg(*args, Size_t); break;
12614 #ifdef HAS_PTRDIFF_T
12615 /* will sign extend, but there is no
12616 * uptrdiff_t, so oh well */
12617 case 't': uv = va_arg(*args, ptrdiff_t); break;
12619 case 'j': uv = va_arg(*args, PERL_UINTMAX_T); break;
12620 default: uv = va_arg(*args, unsigned); break;
12623 uv = va_arg(*args, Uquad_t); break;
12630 /* assign to tiv then cast to iv to work around
12631 * 2003 GCC cast bug (gnu.org bugzilla #13488) */
12632 UV tuv = SvUV_nomg(argsv);
12634 case 'c': uv = (unsigned char)tuv; break;
12635 case 'h': uv = (unsigned short)tuv; break;
12636 case 'l': uv = (unsigned long)tuv; break;
12638 default: uv = tuv; break;
12641 uv = (Uquad_t)tuv; break;
12652 char *ptr = ebuf + sizeof ebuf;
12659 const char * const p =
12660 (c == 'X') ? PL_hexdigit + 16 : PL_hexdigit;
12665 } while (uv >>= 4);
12666 if (alt && *ptr != '0') {
12667 esignbuf[esignlen++] = '0';
12668 esignbuf[esignlen++] = c; /* 'x' or 'X' */
12675 *--ptr = '0' + dig;
12676 } while (uv >>= 3);
12677 if (alt && *ptr != '0')
12683 *--ptr = '0' + dig;
12684 } while (uv >>= 1);
12685 if (alt && *ptr != '0') {
12686 esignbuf[esignlen++] = '0';
12687 esignbuf[esignlen++] = c; /* 'b' or 'B' */
12692 /* special-case: base 1 indicates a 'c' format:
12693 * we use the common code for extracting a uv,
12694 * but handle that value differently here than
12695 * all the other int types */
12697 (!UVCHR_IS_INVARIANT(uv) && SvUTF8(sv)))
12700 assert(sizeof(ebuf) >= UTF8_MAXBYTES + 1);
12702 elen = uvchr_to_utf8((U8*)eptr, uv) - (U8*)ebuf;
12707 ebuf[0] = (char)uv;
12712 default: /* it had better be ten or less */
12715 *--ptr = '0' + dig;
12716 } while (uv /= base);
12719 elen = (ebuf + sizeof ebuf) - ptr;
12723 zeros = precis - elen;
12724 else if (precis == 0 && elen == 1 && *eptr == '0'
12725 && !(base == 8 && alt)) /* "%#.0o" prints "0" */
12728 /* a precision nullifies the 0 flag. */
12734 /* FLOATING POINT */
12737 c = 'f'; /* maybe %F isn't supported here */
12739 case 'e': case 'E':
12741 case 'g': case 'G':
12742 case 'a': case 'A':
12745 STRLEN float_need; /* what PL_efloatsize needs to become */
12746 bool hexfp; /* hexadecimal floating point? */
12748 vcatpvfn_long_double_t fv;
12751 /* This is evil, but floating point is even more evil */
12753 /* for SV-style calling, we can only get NV
12754 for C-style calling, we assume %f is double;
12755 for simplicity we allow any of %Lf, %llf, %qf for long double
12759 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12763 /* [perl #20339] - we should accept and ignore %lf rather than die */
12767 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12768 intsize = args ? 0 : 'q';
12772 #if defined(HAS_LONG_DOUBLE)
12785 /* Now we need (long double) if intsize == 'q', else (double). */
12787 /* Note: do not pull NVs off the va_list with va_arg()
12788 * (pull doubles instead) because if you have a build
12789 * with long doubles, you would always be pulling long
12790 * doubles, which would badly break anyone using only
12791 * doubles (i.e. the majority of builds). In other
12792 * words, you cannot mix doubles and long doubles.
12793 * The only case where you can pull off long doubles
12794 * is when the format specifier explicitly asks so with
12796 #ifdef USE_QUADMATH
12797 fv = intsize == 'q' ?
12798 va_arg(*args, NV) : va_arg(*args, double);
12800 #elif LONG_DOUBLESIZE > DOUBLESIZE
12801 if (intsize == 'q') {
12802 fv = va_arg(*args, long double);
12805 nv = va_arg(*args, double);
12806 VCATPVFN_NV_TO_FV(nv, fv);
12809 nv = va_arg(*args, double);
12816 /* we jump here if an int-ish format encountered an
12817 * infinite/Nan argsv. After setting nv/fv, it falls
12818 * into the isinfnan block which follows */
12819 handle_infnan_argsv:
12820 nv = SvNV_nomg(argsv);
12821 VCATPVFN_NV_TO_FV(nv, fv);
12824 if (Perl_isinfnan(nv)) {
12826 Perl_croak(aTHX_ "Cannot printf %" NVgf " with '%c'",
12827 SvNV_nomg(argsv), (int)c);
12829 elen = S_infnan_2pv(nv, ebuf, sizeof(ebuf), plus);
12838 /* special-case "%.0f" */
12842 && !(width || left || plus || alt)
12845 && ((eptr = F0convert(nv, ebuf + sizeof ebuf, &elen)))
12849 /* Determine the buffer size needed for the various
12850 * floating-point formats.
12852 * The basic possibilities are:
12855 * %f 1111111.123456789
12856 * %e 1.111111123e+06
12857 * %a 0x1.0f4471f9bp+20
12859 * %g 1.11111112e+15
12861 * where P is the value of the precision in the format, or 6
12862 * if not specified. Note the two possible output formats of
12863 * %g; in both cases the number of significant digits is <=
12866 * For most of the format types the maximum buffer size needed
12867 * is precision, plus: any leading 1 or 0x1, the radix
12868 * point, and an exponent. The difficult one is %f: for a
12869 * large positive exponent it can have many leading digits,
12870 * which needs to be calculated specially. Also %a is slightly
12871 * different in that in the absence of a specified precision,
12872 * it uses as many digits as necessary to distinguish
12873 * different values.
12875 * First, here are the constant bits. For ease of calculation
12876 * we over-estimate the needed buffer size, for example by
12877 * assuming all formats have an exponent and a leading 0x1.
12879 * Also for production use, add a little extra overhead for
12880 * safety's sake. Under debugging don't, as it means we're
12881 * more likely to quickly spot issues during development.
12884 float_need = 1 /* possible unary minus */
12885 + 4 /* "0x1" plus very unlikely carry */
12886 + 1 /* default radix point '.' */
12887 + 2 /* "e-", "p+" etc */
12888 + 6 /* exponent: up to 16383 (quad fp) */
12890 + 20 /* safety net */
12895 /* determine the radix point len, e.g. length(".") in "1.2" */
12896 #ifdef USE_LOCALE_NUMERIC
12897 /* note that we may either explicitly use PL_numeric_radix_sv
12898 * below, or implicitly, via an snprintf() variant.
12899 * Note also things like ps_AF.utf8 which has
12900 * "\N{ARABIC DECIMAL SEPARATOR} as a radix point */
12901 if (!lc_numeric_set) {
12902 /* only set once and reuse in-locale value on subsequent
12904 * XXX what happens if we die in an eval?
12906 STORE_LC_NUMERIC_SET_TO_NEEDED();
12907 lc_numeric_set = TRUE;
12910 if (IN_LC(LC_NUMERIC)) {
12911 /* this can't wrap unless PL_numeric_radix_sv is a string
12912 * consuming virtually all the 32-bit or 64-bit address
12915 float_need += (SvCUR(PL_numeric_radix_sv) - 1);
12917 /* floating-point formats only get utf8 if the radix point
12918 * is utf8. All other characters in the string are < 128
12919 * and so can be safely appended to both a non-utf8 and utf8
12921 * Note that this will convert the output to utf8 even if
12922 * the radix point didn't get output.
12924 if (SvUTF8(PL_numeric_radix_sv) && !has_utf8) {
12925 sv_utf8_upgrade(sv);
12933 if (isALPHA_FOLD_EQ(c, 'f')) {
12934 /* Determine how many digits before the radix point
12935 * might be emitted. frexp() (or frexpl) has some
12936 * unspecified behaviour for nan/inf/-inf, so lucky we've
12937 * already handled them above */
12939 int i = PERL_INT_MIN;
12940 (void)Perl_frexp((NV)fv, &i);
12941 if (i == PERL_INT_MIN)
12942 Perl_die(aTHX_ "panic: frexp: %" VCATPVFN_FV_GF, fv);
12945 digits = BIT_DIGITS(i);
12946 /* this can't overflow. 'digits' will only be a few
12947 * thousand even for the largest floating-point types.
12948 * And up until now float_need is just some small
12949 * constants plus radix len, which can't be in
12950 * overflow territory unless the radix SV is consuming
12951 * over 1/2 the address space */
12952 assert(float_need < ((STRLEN)~0) - digits);
12953 float_need += digits;
12956 else if (UNLIKELY(isALPHA_FOLD_EQ(c, 'a'))) {
12959 /* %a in the absence of precision may print as many
12960 * digits as needed to represent the entire mantissa
12962 * This estimate seriously overshoots in most cases,
12963 * but better the undershooting. Firstly, all bytes
12964 * of the NV are not mantissa, some of them are
12965 * exponent. Secondly, for the reasonably common
12966 * long doubles case, the "80-bit extended", two
12967 * or six bytes of the NV are unused. Also, we'll
12968 * still pick up an extra +6 from the default
12969 * precision calculation below. */
12971 #ifdef LONGDOUBLE_DOUBLEDOUBLE
12972 /* For the "double double", we need more.
12973 * Since each double has their own exponent, the
12974 * doubles may float (haha) rather far from each
12975 * other, and the number of required bits is much
12976 * larger, up to total of DOUBLEDOUBLE_MAXBITS bits.
12977 * See the definition of DOUBLEDOUBLE_MAXBITS.
12979 * Need 2 hexdigits for each byte. */
12980 (DOUBLEDOUBLE_MAXBITS/8 + 1) * 2;
12982 NVSIZE * 2; /* 2 hexdigits for each byte */
12984 /* see "this can't overflow" comment above */
12985 assert(float_need < ((STRLEN)~0) - digits);
12986 float_need += digits;
12989 /* special-case "%.<number>g" if it will fit in ebuf */
12991 && precis /* See earlier comment about buggy Gconvert
12992 when digits, aka precis, is 0 */
12994 /* check, in manner not involving wrapping, that it will
12996 && float_need < sizeof(ebuf)
12997 && sizeof(ebuf) - float_need > precis
12998 && !(width || left || plus || alt)
13002 SNPRINTF_G(fv, ebuf, sizeof(ebuf), precis);
13003 elen = strlen(ebuf);
13010 STRLEN pr = has_precis ? precis : 6; /* known default */
13011 /* this probably can't wrap, since precis is limited
13012 * to 1/4 address space size, but better safe than sorry
13014 if (float_need >= ((STRLEN)~0) - pr)
13015 croak_memory_wrap();
13019 if (float_need < width)
13020 float_need = width;
13022 if (PL_efloatsize <= float_need) {
13023 /* PL_efloatbuf should be at least 1 greater than
13024 * float_need to allow a trailing \0 to be returned by
13025 * snprintf(). If we need to grow, overgrow for the
13026 * benefit of future generations */
13027 const STRLEN extra = 0x20;
13028 if (float_need >= ((STRLEN)~0) - extra)
13029 croak_memory_wrap();
13030 float_need += extra;
13031 Safefree(PL_efloatbuf);
13032 PL_efloatsize = float_need;
13033 Newx(PL_efloatbuf, PL_efloatsize, char);
13034 PL_efloatbuf[0] = '\0';
13037 if (UNLIKELY(hexfp)) {
13038 elen = S_format_hexfp(aTHX_ PL_efloatbuf, PL_efloatsize, c,
13039 nv, fv, has_precis, precis, width,
13040 alt, plus, left, fill);
13043 char *ptr = ebuf + sizeof ebuf;
13046 #if defined(USE_QUADMATH)
13047 if (intsize == 'q') {
13051 /* FIXME: what to do if HAS_LONG_DOUBLE but not PERL_PRIfldbl? */
13052 #elif defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
13053 /* Note that this is HAS_LONG_DOUBLE and PERL_PRIfldbl,
13054 * not USE_LONG_DOUBLE and NVff. In other words,
13055 * this needs to work without USE_LONG_DOUBLE. */
13056 if (intsize == 'q') {
13057 /* Copy the one or more characters in a long double
13058 * format before the 'base' ([efgEFG]) character to
13059 * the format string. */
13060 static char const ldblf[] = PERL_PRIfldbl;
13061 char const *p = ldblf + sizeof(ldblf) - 3;
13062 while (p >= ldblf) { *--ptr = *p--; }
13067 do { *--ptr = '0' + (base % 10); } while (base /= 10);
13072 do { *--ptr = '0' + (base % 10); } while (base /= 10);
13084 /* No taint. Otherwise we are in the strange situation
13085 * where printf() taints but print($float) doesn't.
13088 /* hopefully the above makes ptr a very constrained format
13089 * that is safe to use, even though it's not literal */
13090 GCC_DIAG_IGNORE_STMT(-Wformat-nonliteral);
13091 #ifdef USE_QUADMATH
13093 const char* qfmt = quadmath_format_single(ptr);
13095 Perl_croak_nocontext("panic: quadmath invalid format \"%s\"", ptr);
13096 elen = quadmath_snprintf(PL_efloatbuf, PL_efloatsize,
13098 if ((IV)elen == -1) {
13101 Perl_croak_nocontext("panic: quadmath_snprintf failed, format \"%s\"", qfmt);
13106 #elif defined(HAS_LONG_DOUBLE)
13107 elen = ((intsize == 'q')
13108 ? my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv)
13109 : my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, (double)fv));
13111 elen = my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv);
13113 GCC_DIAG_RESTORE_STMT;
13116 eptr = PL_efloatbuf;
13120 /* Since floating-point formats do their own formatting and
13121 * padding, we skip the main block of code at the end of this
13122 * loop which handles appending eptr to sv, and do our own
13123 * stripped-down version */
13128 assert(elen >= width);
13130 S_sv_catpvn_simple(aTHX_ sv, eptr, elen);
13132 goto done_valid_conversion;
13140 /* XXX ideally we should warn if any flags etc have been
13141 * set, e.g. "%-4.5n" */
13142 /* XXX if sv was originally non-utf8 with a char in the
13143 * range 0x80-0xff, then if it got upgraded, we should
13144 * calculate char len rather than byte len here */
13145 len = SvCUR(sv) - origlen;
13147 int i = (len > PERL_INT_MAX) ? PERL_INT_MAX : (int)len;
13150 case 'c': *(va_arg(*args, char*)) = i; break;
13151 case 'h': *(va_arg(*args, short*)) = i; break;
13152 default: *(va_arg(*args, int*)) = i; break;
13153 case 'l': *(va_arg(*args, long*)) = i; break;
13154 case 'V': *(va_arg(*args, IV*)) = i; break;
13155 case 'z': *(va_arg(*args, SSize_t*)) = i; break;
13156 #ifdef HAS_PTRDIFF_T
13157 case 't': *(va_arg(*args, ptrdiff_t*)) = i; break;
13159 case 'j': *(va_arg(*args, PERL_INTMAX_T*)) = i; break;
13162 *(va_arg(*args, Quad_t*)) = i; break;
13170 Perl_croak_nocontext(
13171 "Missing argument for %%n in %s",
13172 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
13173 sv_setuv_mg(argsv, has_utf8 ? (UV)sv_len_utf8(sv) : (UV)len);
13175 goto done_valid_conversion;
13183 && (PL_op->op_type == OP_PRTF || PL_op->op_type == OP_SPRINTF)
13184 && ckWARN(WARN_PRINTF))
13186 SV * const msg = sv_newmortal();
13187 Perl_sv_setpvf(aTHX_ msg, "Invalid conversion in %sprintf: ",
13188 (PL_op->op_type == OP_PRTF) ? "" : "s");
13189 if (fmtstart < patend) {
13190 const char * const fmtend = q < patend ? q : patend;
13192 sv_catpvs(msg, "\"%");
13193 for (f = fmtstart; f < fmtend; f++) {
13195 sv_catpvn_nomg(msg, f, 1);
13197 Perl_sv_catpvf(aTHX_ msg,
13198 "\\%03" UVof, (UV)*f & 0xFF);
13201 sv_catpvs(msg, "\"");
13203 sv_catpvs(msg, "end of string");
13205 Perl_warner(aTHX_ packWARN(WARN_PRINTF), "%" SVf, SVfARG(msg)); /* yes, this is reentrant */
13208 /* mangled format: output the '%', then continue from the
13209 * character following that */
13210 sv_catpvn_nomg(sv, fmtstart-1, 1);
13213 /* Any "redundant arg" warning from now onwards will probably
13214 * just be misleading, so don't bother. */
13215 no_redundant_warning = TRUE;
13216 continue; /* not "break" */
13219 if (is_utf8 != has_utf8) {
13222 sv_utf8_upgrade(sv);
13225 const STRLEN old_elen = elen;
13226 SV * const nsv = newSVpvn_flags(eptr, elen, SVs_TEMP);
13227 sv_utf8_upgrade(nsv);
13228 eptr = SvPVX_const(nsv);
13231 if (width) { /* fudge width (can't fudge elen) */
13232 width += elen - old_elen;
13239 /* append esignbuf, filler, zeros, eptr and dotstr to sv */
13242 STRLEN need, have, gap;
13246 /* signed value that's wrapped? */
13247 assert(elen <= ((~(STRLEN)0) >> 1));
13249 /* if zeros is non-zero, then it represents filler between
13250 * elen and precis. So adding elen and zeros together will
13251 * always be <= precis, and the addition can never wrap */
13252 assert(!zeros || (precis > elen && precis - elen == zeros));
13253 have = elen + zeros;
13255 if (have >= (((STRLEN)~0) - esignlen))
13256 croak_memory_wrap();
13259 need = (have > width ? have : width);
13262 if (need >= (((STRLEN)~0) - (SvCUR(sv) + 1)))
13263 croak_memory_wrap();
13264 need += (SvCUR(sv) + 1);
13271 for (i = 0; i < esignlen; i++)
13272 *s++ = esignbuf[i];
13273 for (i = zeros; i; i--)
13275 Copy(eptr, s, elen, char);
13277 for (i = gap; i; i--)
13282 for (i = 0; i < esignlen; i++)
13283 *s++ = esignbuf[i];
13288 for (i = gap; i; i--)
13290 for (i = 0; i < esignlen; i++)
13291 *s++ = esignbuf[i];
13294 for (i = zeros; i; i--)
13296 Copy(eptr, s, elen, char);
13301 SvCUR_set(sv, s - SvPVX_const(sv));
13309 if (vectorize && veclen) {
13310 /* we append the vector separator separately since %v isn't
13311 * very common: don't slow down the general case by adding
13312 * dotstrlen to need etc */
13313 sv_catpvn_nomg(sv, dotstr, dotstrlen);
13315 goto vector; /* do next iteration */
13318 done_valid_conversion:
13321 S_warn_vcatpvfn_missing_argument(aTHX);
13324 /* Now that we've consumed all our printf format arguments (svix)
13325 * do we have things left on the stack that we didn't use?
13327 if (!no_redundant_warning && sv_count >= svix + 1 && ckWARN(WARN_REDUNDANT)) {
13328 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
13329 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
13334 #ifdef USE_LOCALE_NUMERIC
13336 if (lc_numeric_set) {
13337 RESTORE_LC_NUMERIC(); /* Done outside loop, so don't have to
13338 save/restore each iteration. */
13345 /* =========================================================================
13347 =head1 Cloning an interpreter
13351 All the macros and functions in this section are for the private use of
13352 the main function, perl_clone().
13354 The foo_dup() functions make an exact copy of an existing foo thingy.
13355 During the course of a cloning, a hash table is used to map old addresses
13356 to new addresses. The table is created and manipulated with the
13357 ptr_table_* functions.
13359 * =========================================================================*/
13362 #if defined(USE_ITHREADS)
13364 /* XXX Remove this so it doesn't have to go thru the macro and return for nothing */
13365 #ifndef GpREFCNT_inc
13366 # define GpREFCNT_inc(gp) ((gp) ? (++(gp)->gp_refcnt, (gp)) : (GP*)NULL)
13370 /* Certain cases in Perl_ss_dup have been merged, by relying on the fact
13371 that currently av_dup, gv_dup and hv_dup are the same as sv_dup.
13372 If this changes, please unmerge ss_dup.
13373 Likewise, sv_dup_inc_multiple() relies on this fact. */
13374 #define sv_dup_inc_NN(s,t) SvREFCNT_inc_NN(sv_dup_inc(s,t))
13375 #define av_dup(s,t) MUTABLE_AV(sv_dup((const SV *)s,t))
13376 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
13377 #define hv_dup(s,t) MUTABLE_HV(sv_dup((const SV *)s,t))
13378 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
13379 #define cv_dup(s,t) MUTABLE_CV(sv_dup((const SV *)s,t))
13380 #define cv_dup_inc(s,t) MUTABLE_CV(sv_dup_inc((const SV *)s,t))
13381 #define io_dup(s,t) MUTABLE_IO(sv_dup((const SV *)s,t))
13382 #define io_dup_inc(s,t) MUTABLE_IO(sv_dup_inc((const SV *)s,t))
13383 #define gv_dup(s,t) MUTABLE_GV(sv_dup((const SV *)s,t))
13384 #define gv_dup_inc(s,t) MUTABLE_GV(sv_dup_inc((const SV *)s,t))
13385 #define SAVEPV(p) ((p) ? savepv(p) : NULL)
13386 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
13388 /* clone a parser */
13391 Perl_parser_dup(pTHX_ const yy_parser *const proto, CLONE_PARAMS *const param)
13395 PERL_ARGS_ASSERT_PARSER_DUP;
13400 /* look for it in the table first */
13401 parser = (yy_parser *)ptr_table_fetch(PL_ptr_table, proto);
13405 /* create anew and remember what it is */
13406 Newxz(parser, 1, yy_parser);
13407 ptr_table_store(PL_ptr_table, proto, parser);
13409 /* XXX eventually, just Copy() most of the parser struct ? */
13411 parser->lex_brackets = proto->lex_brackets;
13412 parser->lex_casemods = proto->lex_casemods;
13413 parser->lex_brackstack = savepvn(proto->lex_brackstack,
13414 (proto->lex_brackets < 120 ? 120 : proto->lex_brackets));
13415 parser->lex_casestack = savepvn(proto->lex_casestack,
13416 (proto->lex_casemods < 12 ? 12 : proto->lex_casemods));
13417 parser->lex_defer = proto->lex_defer;
13418 parser->lex_dojoin = proto->lex_dojoin;
13419 parser->lex_formbrack = proto->lex_formbrack;
13420 parser->lex_inpat = proto->lex_inpat;
13421 parser->lex_inwhat = proto->lex_inwhat;
13422 parser->lex_op = proto->lex_op;
13423 parser->lex_repl = sv_dup_inc(proto->lex_repl, param);
13424 parser->lex_starts = proto->lex_starts;
13425 parser->lex_stuff = sv_dup_inc(proto->lex_stuff, param);
13426 parser->multi_close = proto->multi_close;
13427 parser->multi_open = proto->multi_open;
13428 parser->multi_start = proto->multi_start;
13429 parser->multi_end = proto->multi_end;
13430 parser->preambled = proto->preambled;
13431 parser->lex_super_state = proto->lex_super_state;
13432 parser->lex_sub_inwhat = proto->lex_sub_inwhat;
13433 parser->lex_sub_op = proto->lex_sub_op;
13434 parser->lex_sub_repl= sv_dup_inc(proto->lex_sub_repl, param);
13435 parser->linestr = sv_dup_inc(proto->linestr, param);
13436 parser->expect = proto->expect;
13437 parser->copline = proto->copline;
13438 parser->last_lop_op = proto->last_lop_op;
13439 parser->lex_state = proto->lex_state;
13440 parser->rsfp = fp_dup(proto->rsfp, '<', param);
13441 /* rsfp_filters entries have fake IoDIRP() */
13442 parser->rsfp_filters= av_dup_inc(proto->rsfp_filters, param);
13443 parser->in_my = proto->in_my;
13444 parser->in_my_stash = hv_dup(proto->in_my_stash, param);
13445 parser->error_count = proto->error_count;
13446 parser->sig_elems = proto->sig_elems;
13447 parser->sig_optelems= proto->sig_optelems;
13448 parser->sig_slurpy = proto->sig_slurpy;
13449 parser->recheck_utf8_validity = proto->recheck_utf8_validity;
13452 char * const ols = SvPVX(proto->linestr);
13453 char * const ls = SvPVX(parser->linestr);
13455 parser->bufptr = ls + (proto->bufptr >= ols ?
13456 proto->bufptr - ols : 0);
13457 parser->oldbufptr = ls + (proto->oldbufptr >= ols ?
13458 proto->oldbufptr - ols : 0);
13459 parser->oldoldbufptr= ls + (proto->oldoldbufptr >= ols ?
13460 proto->oldoldbufptr - ols : 0);
13461 parser->linestart = ls + (proto->linestart >= ols ?
13462 proto->linestart - ols : 0);
13463 parser->last_uni = ls + (proto->last_uni >= ols ?
13464 proto->last_uni - ols : 0);
13465 parser->last_lop = ls + (proto->last_lop >= ols ?
13466 proto->last_lop - ols : 0);
13468 parser->bufend = ls + SvCUR(parser->linestr);
13471 Copy(proto->tokenbuf, parser->tokenbuf, 256, char);
13474 Copy(proto->nextval, parser->nextval, 5, YYSTYPE);
13475 Copy(proto->nexttype, parser->nexttype, 5, I32);
13476 parser->nexttoke = proto->nexttoke;
13478 /* XXX should clone saved_curcop here, but we aren't passed
13479 * proto_perl; so do it in perl_clone_using instead */
13485 /* duplicate a file handle */
13488 Perl_fp_dup(pTHX_ PerlIO *const fp, const char type, CLONE_PARAMS *const param)
13492 PERL_ARGS_ASSERT_FP_DUP;
13493 PERL_UNUSED_ARG(type);
13496 return (PerlIO*)NULL;
13498 /* look for it in the table first */
13499 ret = (PerlIO*)ptr_table_fetch(PL_ptr_table, fp);
13503 /* create anew and remember what it is */
13504 #ifdef __amigaos4__
13505 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE|PERLIO_DUP_FD);
13507 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE);
13509 ptr_table_store(PL_ptr_table, fp, ret);
13513 /* duplicate a directory handle */
13516 Perl_dirp_dup(pTHX_ DIR *const dp, CLONE_PARAMS *const param)
13520 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
13522 const Direntry_t *dirent;
13523 char smallbuf[256]; /* XXX MAXPATHLEN, surely? */
13529 PERL_UNUSED_CONTEXT;
13530 PERL_ARGS_ASSERT_DIRP_DUP;
13535 /* look for it in the table first */
13536 ret = (DIR*)ptr_table_fetch(PL_ptr_table, dp);
13540 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
13542 PERL_UNUSED_ARG(param);
13546 /* open the current directory (so we can switch back) */
13547 if (!(pwd = PerlDir_open("."))) return (DIR *)NULL;
13549 /* chdir to our dir handle and open the present working directory */
13550 if (fchdir(my_dirfd(dp)) < 0 || !(ret = PerlDir_open("."))) {
13551 PerlDir_close(pwd);
13552 return (DIR *)NULL;
13554 /* Now we should have two dir handles pointing to the same dir. */
13556 /* Be nice to the calling code and chdir back to where we were. */
13557 /* XXX If this fails, then what? */
13558 PERL_UNUSED_RESULT(fchdir(my_dirfd(pwd)));
13560 /* We have no need of the pwd handle any more. */
13561 PerlDir_close(pwd);
13564 # define d_namlen(d) (d)->d_namlen
13566 # define d_namlen(d) strlen((d)->d_name)
13568 /* Iterate once through dp, to get the file name at the current posi-
13569 tion. Then step back. */
13570 pos = PerlDir_tell(dp);
13571 if ((dirent = PerlDir_read(dp))) {
13572 len = d_namlen(dirent);
13573 if (len > sizeof(dirent->d_name) && sizeof(dirent->d_name) > PTRSIZE) {
13574 /* If the len is somehow magically longer than the
13575 * maximum length of the directory entry, even though
13576 * we could fit it in a buffer, we could not copy it
13577 * from the dirent. Bail out. */
13578 PerlDir_close(ret);
13581 if (len <= sizeof smallbuf) name = smallbuf;
13582 else Newx(name, len, char);
13583 Move(dirent->d_name, name, len, char);
13585 PerlDir_seek(dp, pos);
13587 /* Iterate through the new dir handle, till we find a file with the
13589 if (!dirent) /* just before the end */
13591 pos = PerlDir_tell(ret);
13592 if (PerlDir_read(ret)) continue; /* not there yet */
13593 PerlDir_seek(ret, pos); /* step back */
13597 const long pos0 = PerlDir_tell(ret);
13599 pos = PerlDir_tell(ret);
13600 if ((dirent = PerlDir_read(ret))) {
13601 if (len == (STRLEN)d_namlen(dirent)
13602 && memEQ(name, dirent->d_name, len)) {
13604 PerlDir_seek(ret, pos); /* step back */
13607 /* else we are not there yet; keep iterating */
13609 else { /* This is not meant to happen. The best we can do is
13610 reset the iterator to the beginning. */
13611 PerlDir_seek(ret, pos0);
13618 if (name && name != smallbuf)
13623 ret = win32_dirp_dup(dp, param);
13626 /* pop it in the pointer table */
13628 ptr_table_store(PL_ptr_table, dp, ret);
13633 /* duplicate a typeglob */
13636 Perl_gp_dup(pTHX_ GP *const gp, CLONE_PARAMS *const param)
13640 PERL_ARGS_ASSERT_GP_DUP;
13644 /* look for it in the table first */
13645 ret = (GP*)ptr_table_fetch(PL_ptr_table, gp);
13649 /* create anew and remember what it is */
13651 ptr_table_store(PL_ptr_table, gp, ret);
13654 /* ret->gp_refcnt must be 0 before any other dups are called. We're relying
13655 on Newxz() to do this for us. */
13656 ret->gp_sv = sv_dup_inc(gp->gp_sv, param);
13657 ret->gp_io = io_dup_inc(gp->gp_io, param);
13658 ret->gp_form = cv_dup_inc(gp->gp_form, param);
13659 ret->gp_av = av_dup_inc(gp->gp_av, param);
13660 ret->gp_hv = hv_dup_inc(gp->gp_hv, param);
13661 ret->gp_egv = gv_dup(gp->gp_egv, param);/* GvEGV is not refcounted */
13662 ret->gp_cv = cv_dup_inc(gp->gp_cv, param);
13663 ret->gp_cvgen = gp->gp_cvgen;
13664 ret->gp_line = gp->gp_line;
13665 ret->gp_file_hek = hek_dup(gp->gp_file_hek, param);
13669 /* duplicate a chain of magic */
13672 Perl_mg_dup(pTHX_ MAGIC *mg, CLONE_PARAMS *const param)
13674 MAGIC *mgret = NULL;
13675 MAGIC **mgprev_p = &mgret;
13677 PERL_ARGS_ASSERT_MG_DUP;
13679 for (; mg; mg = mg->mg_moremagic) {
13682 if ((param->flags & CLONEf_JOIN_IN)
13683 && mg->mg_type == PERL_MAGIC_backref)
13684 /* when joining, we let the individual SVs add themselves to
13685 * backref as needed. */
13688 Newx(nmg, 1, MAGIC);
13690 mgprev_p = &(nmg->mg_moremagic);
13692 /* There was a comment "XXX copy dynamic vtable?" but as we don't have
13693 dynamic vtables, I'm not sure why Sarathy wrote it. The comment dates
13694 from the original commit adding Perl_mg_dup() - revision 4538.
13695 Similarly there is the annotation "XXX random ptr?" next to the
13696 assignment to nmg->mg_ptr. */
13699 /* FIXME for plugins
13700 if (nmg->mg_type == PERL_MAGIC_qr) {
13701 nmg->mg_obj = MUTABLE_SV(CALLREGDUPE((REGEXP*)nmg->mg_obj, param));
13705 nmg->mg_obj = (nmg->mg_flags & MGf_REFCOUNTED)
13706 ? nmg->mg_type == PERL_MAGIC_backref
13707 /* The backref AV has its reference
13708 * count deliberately bumped by 1 */
13709 ? SvREFCNT_inc(av_dup_inc((const AV *)
13710 nmg->mg_obj, param))
13711 : sv_dup_inc(nmg->mg_obj, param)
13712 : (nmg->mg_type == PERL_MAGIC_regdatum ||
13713 nmg->mg_type == PERL_MAGIC_regdata)
13715 : sv_dup(nmg->mg_obj, param);
13717 if (nmg->mg_ptr && nmg->mg_type != PERL_MAGIC_regex_global) {
13718 if (nmg->mg_len > 0) {
13719 nmg->mg_ptr = SAVEPVN(nmg->mg_ptr, nmg->mg_len);
13720 if (nmg->mg_type == PERL_MAGIC_overload_table &&
13721 AMT_AMAGIC((AMT*)nmg->mg_ptr))
13723 AMT * const namtp = (AMT*)nmg->mg_ptr;
13724 sv_dup_inc_multiple((SV**)(namtp->table),
13725 (SV**)(namtp->table), NofAMmeth, param);
13728 else if (nmg->mg_len == HEf_SVKEY)
13729 nmg->mg_ptr = (char*)sv_dup_inc((const SV *)nmg->mg_ptr, param);
13731 if ((nmg->mg_flags & MGf_DUP) && nmg->mg_virtual && nmg->mg_virtual->svt_dup) {
13732 nmg->mg_virtual->svt_dup(aTHX_ nmg, param);
13738 #endif /* USE_ITHREADS */
13740 struct ptr_tbl_arena {
13741 struct ptr_tbl_arena *next;
13742 struct ptr_tbl_ent array[1023/3]; /* as ptr_tbl_ent has 3 pointers. */
13745 /* create a new pointer-mapping table */
13748 Perl_ptr_table_new(pTHX)
13751 PERL_UNUSED_CONTEXT;
13753 Newx(tbl, 1, PTR_TBL_t);
13754 tbl->tbl_max = 511;
13755 tbl->tbl_items = 0;
13756 tbl->tbl_arena = NULL;
13757 tbl->tbl_arena_next = NULL;
13758 tbl->tbl_arena_end = NULL;
13759 Newxz(tbl->tbl_ary, tbl->tbl_max + 1, PTR_TBL_ENT_t*);
13763 #define PTR_TABLE_HASH(ptr) \
13764 ((PTR2UV(ptr) >> 3) ^ (PTR2UV(ptr) >> (3 + 7)) ^ (PTR2UV(ptr) >> (3 + 17)))
13766 /* map an existing pointer using a table */
13768 STATIC PTR_TBL_ENT_t *
13769 S_ptr_table_find(PTR_TBL_t *const tbl, const void *const sv)
13771 PTR_TBL_ENT_t *tblent;
13772 const UV hash = PTR_TABLE_HASH(sv);
13774 PERL_ARGS_ASSERT_PTR_TABLE_FIND;
13776 tblent = tbl->tbl_ary[hash & tbl->tbl_max];
13777 for (; tblent; tblent = tblent->next) {
13778 if (tblent->oldval == sv)
13785 Perl_ptr_table_fetch(pTHX_ PTR_TBL_t *const tbl, const void *const sv)
13787 PTR_TBL_ENT_t const *const tblent = ptr_table_find(tbl, sv);
13789 PERL_ARGS_ASSERT_PTR_TABLE_FETCH;
13790 PERL_UNUSED_CONTEXT;
13792 return tblent ? tblent->newval : NULL;
13795 /* add a new entry to a pointer-mapping table 'tbl'. In hash terms, 'oldsv' is
13796 * the key; 'newsv' is the value. The names "old" and "new" are specific to
13797 * the core's typical use of ptr_tables in thread cloning. */
13800 Perl_ptr_table_store(pTHX_ PTR_TBL_t *const tbl, const void *const oldsv, void *const newsv)
13802 PTR_TBL_ENT_t *tblent = ptr_table_find(tbl, oldsv);
13804 PERL_ARGS_ASSERT_PTR_TABLE_STORE;
13805 PERL_UNUSED_CONTEXT;
13808 tblent->newval = newsv;
13810 const UV entry = PTR_TABLE_HASH(oldsv) & tbl->tbl_max;
13812 if (tbl->tbl_arena_next == tbl->tbl_arena_end) {
13813 struct ptr_tbl_arena *new_arena;
13815 Newx(new_arena, 1, struct ptr_tbl_arena);
13816 new_arena->next = tbl->tbl_arena;
13817 tbl->tbl_arena = new_arena;
13818 tbl->tbl_arena_next = new_arena->array;
13819 tbl->tbl_arena_end = C_ARRAY_END(new_arena->array);
13822 tblent = tbl->tbl_arena_next++;
13824 tblent->oldval = oldsv;
13825 tblent->newval = newsv;
13826 tblent->next = tbl->tbl_ary[entry];
13827 tbl->tbl_ary[entry] = tblent;
13829 if (tblent->next && tbl->tbl_items > tbl->tbl_max)
13830 ptr_table_split(tbl);
13834 /* double the hash bucket size of an existing ptr table */
13837 Perl_ptr_table_split(pTHX_ PTR_TBL_t *const tbl)
13839 PTR_TBL_ENT_t **ary = tbl->tbl_ary;
13840 const UV oldsize = tbl->tbl_max + 1;
13841 UV newsize = oldsize * 2;
13844 PERL_ARGS_ASSERT_PTR_TABLE_SPLIT;
13845 PERL_UNUSED_CONTEXT;
13847 Renew(ary, newsize, PTR_TBL_ENT_t*);
13848 Zero(&ary[oldsize], newsize-oldsize, PTR_TBL_ENT_t*);
13849 tbl->tbl_max = --newsize;
13850 tbl->tbl_ary = ary;
13851 for (i=0; i < oldsize; i++, ary++) {
13852 PTR_TBL_ENT_t **entp = ary;
13853 PTR_TBL_ENT_t *ent = *ary;
13854 PTR_TBL_ENT_t **curentp;
13857 curentp = ary + oldsize;
13859 if ((newsize & PTR_TABLE_HASH(ent->oldval)) != i) {
13861 ent->next = *curentp;
13871 /* remove all the entries from a ptr table */
13872 /* Deprecated - will be removed post 5.14 */
13875 Perl_ptr_table_clear(pTHX_ PTR_TBL_t *const tbl)
13877 PERL_UNUSED_CONTEXT;
13878 if (tbl && tbl->tbl_items) {
13879 struct ptr_tbl_arena *arena = tbl->tbl_arena;
13881 Zero(tbl->tbl_ary, tbl->tbl_max + 1, struct ptr_tbl_ent *);
13884 struct ptr_tbl_arena *next = arena->next;
13890 tbl->tbl_items = 0;
13891 tbl->tbl_arena = NULL;
13892 tbl->tbl_arena_next = NULL;
13893 tbl->tbl_arena_end = NULL;
13897 /* clear and free a ptr table */
13900 Perl_ptr_table_free(pTHX_ PTR_TBL_t *const tbl)
13902 struct ptr_tbl_arena *arena;
13904 PERL_UNUSED_CONTEXT;
13910 arena = tbl->tbl_arena;
13913 struct ptr_tbl_arena *next = arena->next;
13919 Safefree(tbl->tbl_ary);
13923 #if defined(USE_ITHREADS)
13926 Perl_rvpv_dup(pTHX_ SV *const dstr, const SV *const sstr, CLONE_PARAMS *const param)
13928 PERL_ARGS_ASSERT_RVPV_DUP;
13930 assert(!isREGEXP(sstr));
13932 if (SvWEAKREF(sstr)) {
13933 SvRV_set(dstr, sv_dup(SvRV_const(sstr), param));
13934 if (param->flags & CLONEf_JOIN_IN) {
13935 /* if joining, we add any back references individually rather
13936 * than copying the whole backref array */
13937 Perl_sv_add_backref(aTHX_ SvRV(dstr), dstr);
13941 SvRV_set(dstr, sv_dup_inc(SvRV_const(sstr), param));
13943 else if (SvPVX_const(sstr)) {
13944 /* Has something there */
13946 /* Normal PV - clone whole allocated space */
13947 SvPV_set(dstr, SAVEPVN(SvPVX_const(sstr), SvLEN(sstr)-1));
13948 /* sstr may not be that normal, but actually copy on write.
13949 But we are a true, independent SV, so: */
13953 /* Special case - not normally malloced for some reason */
13954 if (isGV_with_GP(sstr)) {
13955 /* Don't need to do anything here. */
13957 else if ((SvIsCOW(sstr))) {
13958 /* A "shared" PV - clone it as "shared" PV */
13960 HEK_KEY(hek_dup(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)),
13964 /* Some other special case - random pointer */
13965 SvPV_set(dstr, (char *) SvPVX_const(sstr));
13970 /* Copy the NULL */
13971 SvPV_set(dstr, NULL);
13975 /* duplicate a list of SVs. source and dest may point to the same memory. */
13977 S_sv_dup_inc_multiple(pTHX_ SV *const *source, SV **dest,
13978 SSize_t items, CLONE_PARAMS *const param)
13980 PERL_ARGS_ASSERT_SV_DUP_INC_MULTIPLE;
13982 while (items-- > 0) {
13983 *dest++ = sv_dup_inc(*source++, param);
13989 /* duplicate an SV of any type (including AV, HV etc) */
13992 S_sv_dup_common(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13997 PERL_ARGS_ASSERT_SV_DUP_COMMON;
13999 if (SvTYPE(sstr) == (svtype)SVTYPEMASK) {
14000 #ifdef DEBUG_LEAKING_SCALARS_ABORT
14005 /* look for it in the table first */
14006 dstr = MUTABLE_SV(ptr_table_fetch(PL_ptr_table, sstr));
14010 if(param->flags & CLONEf_JOIN_IN) {
14011 /** We are joining here so we don't want do clone
14012 something that is bad **/
14013 if (SvTYPE(sstr) == SVt_PVHV) {
14014 const HEK * const hvname = HvNAME_HEK(sstr);
14016 /** don't clone stashes if they already exist **/
14017 dstr = MUTABLE_SV(gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
14018 HEK_UTF8(hvname) ? SVf_UTF8 : 0));
14019 ptr_table_store(PL_ptr_table, sstr, dstr);
14023 else if (SvTYPE(sstr) == SVt_PVGV && !SvFAKE(sstr)) {
14024 HV *stash = GvSTASH(sstr);
14025 const HEK * hvname;
14026 if (stash && (hvname = HvNAME_HEK(stash))) {
14027 /** don't clone GVs if they already exist **/
14029 stash = gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
14030 HEK_UTF8(hvname) ? SVf_UTF8 : 0);
14032 stash, GvNAME(sstr),
14038 if (svp && *svp && SvTYPE(*svp) == SVt_PVGV) {
14039 ptr_table_store(PL_ptr_table, sstr, *svp);
14046 /* create anew and remember what it is */
14049 #ifdef DEBUG_LEAKING_SCALARS
14050 dstr->sv_debug_optype = sstr->sv_debug_optype;
14051 dstr->sv_debug_line = sstr->sv_debug_line;
14052 dstr->sv_debug_inpad = sstr->sv_debug_inpad;
14053 dstr->sv_debug_parent = (SV*)sstr;
14054 FREE_SV_DEBUG_FILE(dstr);
14055 dstr->sv_debug_file = savesharedpv(sstr->sv_debug_file);
14058 ptr_table_store(PL_ptr_table, sstr, dstr);
14061 SvFLAGS(dstr) = SvFLAGS(sstr);
14062 SvFLAGS(dstr) &= ~SVf_OOK; /* don't propagate OOK hack */
14063 SvREFCNT(dstr) = 0; /* must be before any other dups! */
14066 if (SvANY(sstr) && PL_watch_pvx && SvPVX_const(sstr) == PL_watch_pvx)
14067 PerlIO_printf(Perl_debug_log, "watch at %p hit, found string \"%s\"\n",
14068 (void*)PL_watch_pvx, SvPVX_const(sstr));
14071 /* don't clone objects whose class has asked us not to */
14073 && ! (SvFLAGS(SvSTASH(sstr)) & SVphv_CLONEABLE))
14079 switch (SvTYPE(sstr)) {
14081 SvANY(dstr) = NULL;
14084 SET_SVANY_FOR_BODYLESS_IV(dstr);
14086 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
14088 SvIV_set(dstr, SvIVX(sstr));
14092 #if NVSIZE <= IVSIZE
14093 SET_SVANY_FOR_BODYLESS_NV(dstr);
14095 SvANY(dstr) = new_XNV();
14097 SvNV_set(dstr, SvNVX(sstr));
14101 /* These are all the types that need complex bodies allocating. */
14103 const svtype sv_type = SvTYPE(sstr);
14104 const struct body_details *const sv_type_details
14105 = bodies_by_type + sv_type;
14109 Perl_croak(aTHX_ "Bizarre SvTYPE [%" IVdf "]", (IV)SvTYPE(sstr));
14110 NOT_REACHED; /* NOTREACHED */
14126 assert(sv_type_details->body_size);
14127 if (sv_type_details->arena) {
14128 new_body_inline(new_body, sv_type);
14130 = (void*)((char*)new_body - sv_type_details->offset);
14132 new_body = new_NOARENA(sv_type_details);
14136 SvANY(dstr) = new_body;
14139 Copy(((char*)SvANY(sstr)) + sv_type_details->offset,
14140 ((char*)SvANY(dstr)) + sv_type_details->offset,
14141 sv_type_details->copy, char);
14143 Copy(((char*)SvANY(sstr)),
14144 ((char*)SvANY(dstr)),
14145 sv_type_details->body_size + sv_type_details->offset, char);
14148 if (sv_type != SVt_PVAV && sv_type != SVt_PVHV
14149 && !isGV_with_GP(dstr)
14151 && !(sv_type == SVt_PVIO && !(IoFLAGS(dstr) & IOf_FAKE_DIRP)))
14152 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
14154 /* The Copy above means that all the source (unduplicated) pointers
14155 are now in the destination. We can check the flags and the
14156 pointers in either, but it's possible that there's less cache
14157 missing by always going for the destination.
14158 FIXME - instrument and check that assumption */
14159 if (sv_type >= SVt_PVMG) {
14161 SvMAGIC_set(dstr, mg_dup(SvMAGIC(dstr), param));
14162 if (SvOBJECT(dstr) && SvSTASH(dstr))
14163 SvSTASH_set(dstr, hv_dup_inc(SvSTASH(dstr), param));
14164 else SvSTASH_set(dstr, 0); /* don't copy DESTROY cache */
14167 /* The cast silences a GCC warning about unhandled types. */
14168 switch ((int)sv_type) {
14179 /* FIXME for plugins */
14180 re_dup_guts((REGEXP*) sstr, (REGEXP*) dstr, param);
14183 /* XXX LvTARGOFF sometimes holds PMOP* when DEBUGGING */
14184 if (LvTYPE(dstr) == 't') /* for tie: unrefcnted fake (SV**) */
14185 LvTARG(dstr) = dstr;
14186 else if (LvTYPE(dstr) == 'T') /* for tie: fake HE */
14187 LvTARG(dstr) = MUTABLE_SV(he_dup((HE*)LvTARG(dstr), 0, param));
14189 LvTARG(dstr) = sv_dup_inc(LvTARG(dstr), param);
14190 if (isREGEXP(sstr)) goto duprex;
14193 /* non-GP case already handled above */
14194 if(isGV_with_GP(sstr)) {
14195 GvNAME_HEK(dstr) = hek_dup(GvNAME_HEK(dstr), param);
14196 /* Don't call sv_add_backref here as it's going to be
14197 created as part of the magic cloning of the symbol
14198 table--unless this is during a join and the stash
14199 is not actually being cloned. */
14200 /* Danger Will Robinson - GvGP(dstr) isn't initialised
14201 at the point of this comment. */
14202 GvSTASH(dstr) = hv_dup(GvSTASH(dstr), param);
14203 if (param->flags & CLONEf_JOIN_IN)
14204 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
14205 GvGP_set(dstr, gp_dup(GvGP(sstr), param));
14206 (void)GpREFCNT_inc(GvGP(dstr));
14210 /* PL_parser->rsfp_filters entries have fake IoDIRP() */
14211 if(IoFLAGS(dstr) & IOf_FAKE_DIRP) {
14212 /* I have no idea why fake dirp (rsfps)
14213 should be treated differently but otherwise
14214 we end up with leaks -- sky*/
14215 IoTOP_GV(dstr) = gv_dup_inc(IoTOP_GV(dstr), param);
14216 IoFMT_GV(dstr) = gv_dup_inc(IoFMT_GV(dstr), param);
14217 IoBOTTOM_GV(dstr) = gv_dup_inc(IoBOTTOM_GV(dstr), param);
14219 IoTOP_GV(dstr) = gv_dup(IoTOP_GV(dstr), param);
14220 IoFMT_GV(dstr) = gv_dup(IoFMT_GV(dstr), param);
14221 IoBOTTOM_GV(dstr) = gv_dup(IoBOTTOM_GV(dstr), param);
14222 if (IoDIRP(dstr)) {
14223 IoDIRP(dstr) = dirp_dup(IoDIRP(dstr), param);
14226 /* IoDIRP(dstr) is already a copy of IoDIRP(sstr) */
14228 IoIFP(dstr) = fp_dup(IoIFP(sstr), IoTYPE(dstr), param);
14230 if (IoOFP(dstr) == IoIFP(sstr))
14231 IoOFP(dstr) = IoIFP(dstr);
14233 IoOFP(dstr) = fp_dup(IoOFP(dstr), IoTYPE(dstr), param);
14234 IoTOP_NAME(dstr) = SAVEPV(IoTOP_NAME(dstr));
14235 IoFMT_NAME(dstr) = SAVEPV(IoFMT_NAME(dstr));
14236 IoBOTTOM_NAME(dstr) = SAVEPV(IoBOTTOM_NAME(dstr));
14239 /* avoid cloning an empty array */
14240 if (AvARRAY((const AV *)sstr) && AvFILLp((const AV *)sstr) >= 0) {
14241 SV **dst_ary, **src_ary;
14242 SSize_t items = AvFILLp((const AV *)sstr) + 1;
14244 src_ary = AvARRAY((const AV *)sstr);
14245 Newx(dst_ary, AvMAX((const AV *)sstr)+1, SV*);
14246 ptr_table_store(PL_ptr_table, src_ary, dst_ary);
14247 AvARRAY(MUTABLE_AV(dstr)) = dst_ary;
14248 AvALLOC((const AV *)dstr) = dst_ary;
14249 if (AvREAL((const AV *)sstr)) {
14250 dst_ary = sv_dup_inc_multiple(src_ary, dst_ary, items,
14254 while (items-- > 0)
14255 *dst_ary++ = sv_dup(*src_ary++, param);
14257 items = AvMAX((const AV *)sstr) - AvFILLp((const AV *)sstr);
14258 while (items-- > 0) {
14263 AvARRAY(MUTABLE_AV(dstr)) = NULL;
14264 AvALLOC((const AV *)dstr) = (SV**)NULL;
14265 AvMAX( (const AV *)dstr) = -1;
14266 AvFILLp((const AV *)dstr) = -1;
14270 if (HvARRAY((const HV *)sstr)) {
14272 const bool sharekeys = !!HvSHAREKEYS(sstr);
14273 XPVHV * const dxhv = (XPVHV*)SvANY(dstr);
14274 XPVHV * const sxhv = (XPVHV*)SvANY(sstr);
14276 Newx(darray, PERL_HV_ARRAY_ALLOC_BYTES(dxhv->xhv_max+1)
14277 + (SvOOK(sstr) ? sizeof(struct xpvhv_aux) : 0),
14279 HvARRAY(dstr) = (HE**)darray;
14280 while (i <= sxhv->xhv_max) {
14281 const HE * const source = HvARRAY(sstr)[i];
14282 HvARRAY(dstr)[i] = source
14283 ? he_dup(source, sharekeys, param) : 0;
14287 const struct xpvhv_aux * const saux = HvAUX(sstr);
14288 struct xpvhv_aux * const daux = HvAUX(dstr);
14289 /* This flag isn't copied. */
14292 if (saux->xhv_name_count) {
14293 HEK ** const sname = saux->xhv_name_u.xhvnameu_names;
14295 = saux->xhv_name_count < 0
14296 ? -saux->xhv_name_count
14297 : saux->xhv_name_count;
14298 HEK **shekp = sname + count;
14300 Newx(daux->xhv_name_u.xhvnameu_names, count, HEK *);
14301 dhekp = daux->xhv_name_u.xhvnameu_names + count;
14302 while (shekp-- > sname) {
14304 *dhekp = hek_dup(*shekp, param);
14308 daux->xhv_name_u.xhvnameu_name
14309 = hek_dup(saux->xhv_name_u.xhvnameu_name,
14312 daux->xhv_name_count = saux->xhv_name_count;
14314 daux->xhv_aux_flags = saux->xhv_aux_flags;
14315 #ifdef PERL_HASH_RANDOMIZE_KEYS
14316 daux->xhv_rand = saux->xhv_rand;
14317 daux->xhv_last_rand = saux->xhv_last_rand;
14319 daux->xhv_riter = saux->xhv_riter;
14320 daux->xhv_eiter = saux->xhv_eiter
14321 ? he_dup(saux->xhv_eiter,
14322 cBOOL(HvSHAREKEYS(sstr)), param) : 0;
14323 /* backref array needs refcnt=2; see sv_add_backref */
14324 daux->xhv_backreferences =
14325 (param->flags & CLONEf_JOIN_IN)
14326 /* when joining, we let the individual GVs and
14327 * CVs add themselves to backref as
14328 * needed. This avoids pulling in stuff
14329 * that isn't required, and simplifies the
14330 * case where stashes aren't cloned back
14331 * if they already exist in the parent
14334 : saux->xhv_backreferences
14335 ? (SvTYPE(saux->xhv_backreferences) == SVt_PVAV)
14336 ? MUTABLE_AV(SvREFCNT_inc(
14337 sv_dup_inc((const SV *)
14338 saux->xhv_backreferences, param)))
14339 : MUTABLE_AV(sv_dup((const SV *)
14340 saux->xhv_backreferences, param))
14343 daux->xhv_mro_meta = saux->xhv_mro_meta
14344 ? mro_meta_dup(saux->xhv_mro_meta, param)
14347 /* Record stashes for possible cloning in Perl_clone(). */
14349 av_push(param->stashes, dstr);
14353 HvARRAY(MUTABLE_HV(dstr)) = NULL;
14356 if (!(param->flags & CLONEf_COPY_STACKS)) {
14361 /* NOTE: not refcounted */
14362 SvANY(MUTABLE_CV(dstr))->xcv_stash =
14363 hv_dup(CvSTASH(dstr), param);
14364 if ((param->flags & CLONEf_JOIN_IN) && CvSTASH(dstr))
14365 Perl_sv_add_backref(aTHX_ MUTABLE_SV(CvSTASH(dstr)), dstr);
14366 if (!CvISXSUB(dstr)) {
14368 CvROOT(dstr) = OpREFCNT_inc(CvROOT(dstr));
14370 CvSLABBED_off(dstr);
14371 } else if (CvCONST(dstr)) {
14372 CvXSUBANY(dstr).any_ptr =
14373 sv_dup_inc((const SV *)CvXSUBANY(dstr).any_ptr, param);
14375 assert(!CvSLABBED(dstr));
14376 if (CvDYNFILE(dstr)) CvFILE(dstr) = SAVEPV(CvFILE(dstr));
14378 SvANY((CV *)dstr)->xcv_gv_u.xcv_hek =
14379 hek_dup(CvNAME_HEK((CV *)sstr), param);
14380 /* don't dup if copying back - CvGV isn't refcounted, so the
14381 * duped GV may never be freed. A bit of a hack! DAPM */
14383 SvANY(MUTABLE_CV(dstr))->xcv_gv_u.xcv_gv =
14385 ? gv_dup_inc(CvGV(sstr), param)
14386 : (param->flags & CLONEf_JOIN_IN)
14388 : gv_dup(CvGV(sstr), param);
14390 if (!CvISXSUB(sstr)) {
14391 PADLIST * padlist = CvPADLIST(sstr);
14393 padlist = padlist_dup(padlist, param);
14394 CvPADLIST_set(dstr, padlist);
14396 /* unthreaded perl can't sv_dup so we dont support unthreaded's CvHSCXT */
14397 PoisonPADLIST(dstr);
14400 CvWEAKOUTSIDE(sstr)
14401 ? cv_dup( CvOUTSIDE(dstr), param)
14402 : cv_dup_inc(CvOUTSIDE(dstr), param);
14412 Perl_sv_dup_inc(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14414 PERL_ARGS_ASSERT_SV_DUP_INC;
14415 return sstr ? SvREFCNT_inc(sv_dup_common(sstr, param)) : NULL;
14419 Perl_sv_dup(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14421 SV *dstr = sstr ? sv_dup_common(sstr, param) : NULL;
14422 PERL_ARGS_ASSERT_SV_DUP;
14424 /* Track every SV that (at least initially) had a reference count of 0.
14425 We need to do this by holding an actual reference to it in this array.
14426 If we attempt to cheat, turn AvREAL_off(), and store only pointers
14427 (akin to the stashes hash, and the perl stack), we come unstuck if
14428 a weak reference (or other SV legitimately SvREFCNT() == 0 for this
14429 thread) is manipulated in a CLONE method, because CLONE runs before the
14430 unreferenced array is walked to find SVs still with SvREFCNT() == 0
14431 (and fix things up by giving each a reference via the temps stack).
14432 Instead, during CLONE, if the 0-referenced SV has SvREFCNT_inc() and
14433 then SvREFCNT_dec(), it will be cleaned up (and added to the free list)
14434 before the walk of unreferenced happens and a reference to that is SV
14435 added to the temps stack. At which point we have the same SV considered
14436 to be in use, and free to be re-used. Not good.
14438 if (dstr && !(param->flags & CLONEf_COPY_STACKS) && !SvREFCNT(dstr)) {
14439 assert(param->unreferenced);
14440 av_push(param->unreferenced, SvREFCNT_inc(dstr));
14446 /* duplicate a context */
14449 Perl_cx_dup(pTHX_ PERL_CONTEXT *cxs, I32 ix, I32 max, CLONE_PARAMS* param)
14451 PERL_CONTEXT *ncxs;
14453 PERL_ARGS_ASSERT_CX_DUP;
14456 return (PERL_CONTEXT*)NULL;
14458 /* look for it in the table first */
14459 ncxs = (PERL_CONTEXT*)ptr_table_fetch(PL_ptr_table, cxs);
14463 /* create anew and remember what it is */
14464 Newx(ncxs, max + 1, PERL_CONTEXT);
14465 ptr_table_store(PL_ptr_table, cxs, ncxs);
14466 Copy(cxs, ncxs, max + 1, PERL_CONTEXT);
14469 PERL_CONTEXT * const ncx = &ncxs[ix];
14470 if (CxTYPE(ncx) == CXt_SUBST) {
14471 Perl_croak(aTHX_ "Cloning substitution context is unimplemented");
14474 ncx->blk_oldcop = (COP*)any_dup(ncx->blk_oldcop, param->proto_perl);
14475 switch (CxTYPE(ncx)) {
14477 ncx->blk_sub.cv = cv_dup_inc(ncx->blk_sub.cv, param);
14478 if(CxHASARGS(ncx)){
14479 ncx->blk_sub.savearray = av_dup_inc(ncx->blk_sub.savearray,param);
14481 ncx->blk_sub.savearray = NULL;
14483 ncx->blk_sub.prevcomppad = (PAD*)ptr_table_fetch(PL_ptr_table,
14484 ncx->blk_sub.prevcomppad);
14487 ncx->blk_eval.old_namesv = sv_dup_inc(ncx->blk_eval.old_namesv,
14489 /* XXX should this sv_dup_inc? Or only if CxEVAL_TXT_REFCNTED ???? */
14490 ncx->blk_eval.cur_text = sv_dup(ncx->blk_eval.cur_text, param);
14491 ncx->blk_eval.cv = cv_dup(ncx->blk_eval.cv, param);
14492 /* XXX what do do with cur_top_env ???? */
14494 case CXt_LOOP_LAZYSV:
14495 ncx->blk_loop.state_u.lazysv.end
14496 = sv_dup_inc(ncx->blk_loop.state_u.lazysv.end, param);
14497 /* Fallthrough: duplicate lazysv.cur by using the ary.ary
14498 duplication code instead.
14499 We are taking advantage of (1) av_dup_inc and sv_dup_inc
14500 actually being the same function, and (2) order
14501 equivalence of the two unions.
14502 We can assert the later [but only at run time :-(] */
14503 assert ((void *) &ncx->blk_loop.state_u.ary.ary ==
14504 (void *) &ncx->blk_loop.state_u.lazysv.cur);
14507 ncx->blk_loop.state_u.ary.ary
14508 = av_dup_inc(ncx->blk_loop.state_u.ary.ary, param);
14510 case CXt_LOOP_LIST:
14511 case CXt_LOOP_LAZYIV:
14512 /* code common to all 'for' CXt_LOOP_* types */
14513 ncx->blk_loop.itersave =
14514 sv_dup_inc(ncx->blk_loop.itersave, param);
14515 if (CxPADLOOP(ncx)) {
14516 PADOFFSET off = ncx->blk_loop.itervar_u.svp
14517 - &CX_CURPAD_SV(ncx->blk_loop, 0);
14518 ncx->blk_loop.oldcomppad =
14519 (PAD*)ptr_table_fetch(PL_ptr_table,
14520 ncx->blk_loop.oldcomppad);
14521 ncx->blk_loop.itervar_u.svp =
14522 &CX_CURPAD_SV(ncx->blk_loop, off);
14525 /* this copies the GV if CXp_FOR_GV, or the SV for an
14526 * alias (for \$x (...)) - relies on gv_dup being the
14527 * same as sv_dup */
14528 ncx->blk_loop.itervar_u.gv
14529 = gv_dup((const GV *)ncx->blk_loop.itervar_u.gv,
14533 case CXt_LOOP_PLAIN:
14536 ncx->blk_format.prevcomppad =
14537 (PAD*)ptr_table_fetch(PL_ptr_table,
14538 ncx->blk_format.prevcomppad);
14539 ncx->blk_format.cv = cv_dup_inc(ncx->blk_format.cv, param);
14540 ncx->blk_format.gv = gv_dup(ncx->blk_format.gv, param);
14541 ncx->blk_format.dfoutgv = gv_dup_inc(ncx->blk_format.dfoutgv,
14545 ncx->blk_givwhen.defsv_save =
14546 sv_dup_inc(ncx->blk_givwhen.defsv_save, param);
14559 /* duplicate a stack info structure */
14562 Perl_si_dup(pTHX_ PERL_SI *si, CLONE_PARAMS* param)
14566 PERL_ARGS_ASSERT_SI_DUP;
14569 return (PERL_SI*)NULL;
14571 /* look for it in the table first */
14572 nsi = (PERL_SI*)ptr_table_fetch(PL_ptr_table, si);
14576 /* create anew and remember what it is */
14577 Newx(nsi, 1, PERL_SI);
14578 ptr_table_store(PL_ptr_table, si, nsi);
14580 nsi->si_stack = av_dup_inc(si->si_stack, param);
14581 nsi->si_cxix = si->si_cxix;
14582 nsi->si_cxmax = si->si_cxmax;
14583 nsi->si_cxstack = cx_dup(si->si_cxstack, si->si_cxix, si->si_cxmax, param);
14584 nsi->si_type = si->si_type;
14585 nsi->si_prev = si_dup(si->si_prev, param);
14586 nsi->si_next = si_dup(si->si_next, param);
14587 nsi->si_markoff = si->si_markoff;
14588 #if defined DEBUGGING && !defined DEBUGGING_RE_ONLY
14589 nsi->si_stack_hwm = 0;
14595 #define POPINT(ss,ix) ((ss)[--(ix)].any_i32)
14596 #define TOPINT(ss,ix) ((ss)[ix].any_i32)
14597 #define POPLONG(ss,ix) ((ss)[--(ix)].any_long)
14598 #define TOPLONG(ss,ix) ((ss)[ix].any_long)
14599 #define POPIV(ss,ix) ((ss)[--(ix)].any_iv)
14600 #define TOPIV(ss,ix) ((ss)[ix].any_iv)
14601 #define POPUV(ss,ix) ((ss)[--(ix)].any_uv)
14602 #define TOPUV(ss,ix) ((ss)[ix].any_uv)
14603 #define POPBOOL(ss,ix) ((ss)[--(ix)].any_bool)
14604 #define TOPBOOL(ss,ix) ((ss)[ix].any_bool)
14605 #define POPPTR(ss,ix) ((ss)[--(ix)].any_ptr)
14606 #define TOPPTR(ss,ix) ((ss)[ix].any_ptr)
14607 #define POPDPTR(ss,ix) ((ss)[--(ix)].any_dptr)
14608 #define TOPDPTR(ss,ix) ((ss)[ix].any_dptr)
14609 #define POPDXPTR(ss,ix) ((ss)[--(ix)].any_dxptr)
14610 #define TOPDXPTR(ss,ix) ((ss)[ix].any_dxptr)
14613 #define pv_dup_inc(p) SAVEPV(p)
14614 #define pv_dup(p) SAVEPV(p)
14615 #define svp_dup_inc(p,pp) any_dup(p,pp)
14617 /* map any object to the new equivent - either something in the
14618 * ptr table, or something in the interpreter structure
14622 Perl_any_dup(pTHX_ void *v, const PerlInterpreter *proto_perl)
14626 PERL_ARGS_ASSERT_ANY_DUP;
14629 return (void*)NULL;
14631 /* look for it in the table first */
14632 ret = ptr_table_fetch(PL_ptr_table, v);
14636 /* see if it is part of the interpreter structure */
14637 if (v >= (void*)proto_perl && v < (void*)(proto_perl+1))
14638 ret = (void*)(((char*)aTHX) + (((char*)v) - (char*)proto_perl));
14646 /* duplicate the save stack */
14649 Perl_ss_dup(pTHX_ PerlInterpreter *proto_perl, CLONE_PARAMS* param)
14652 ANY * const ss = proto_perl->Isavestack;
14653 const I32 max = proto_perl->Isavestack_max + SS_MAXPUSH;
14654 I32 ix = proto_perl->Isavestack_ix;
14667 void (*dptr) (void*);
14668 void (*dxptr) (pTHX_ void*);
14670 PERL_ARGS_ASSERT_SS_DUP;
14672 Newx(nss, max, ANY);
14675 const UV uv = POPUV(ss,ix);
14676 const U8 type = (U8)uv & SAVE_MASK;
14678 TOPUV(nss,ix) = uv;
14680 case SAVEt_CLEARSV:
14681 case SAVEt_CLEARPADRANGE:
14683 case SAVEt_HELEM: /* hash element */
14684 case SAVEt_SV: /* scalar reference */
14685 sv = (const SV *)POPPTR(ss,ix);
14686 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14688 case SAVEt_ITEM: /* normal string */
14689 case SAVEt_GVSV: /* scalar slot in GV */
14690 sv = (const SV *)POPPTR(ss,ix);
14691 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14692 if (type == SAVEt_SV)
14696 case SAVEt_MORTALIZESV:
14697 case SAVEt_READONLY_OFF:
14698 sv = (const SV *)POPPTR(ss,ix);
14699 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14701 case SAVEt_FREEPADNAME:
14702 ptr = POPPTR(ss,ix);
14703 TOPPTR(nss,ix) = padname_dup((PADNAME *)ptr, param);
14704 PadnameREFCNT((PADNAME *)TOPPTR(nss,ix))++;
14706 case SAVEt_SHARED_PVREF: /* char* in shared space */
14707 c = (char*)POPPTR(ss,ix);
14708 TOPPTR(nss,ix) = savesharedpv(c);
14709 ptr = POPPTR(ss,ix);
14710 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14712 case SAVEt_GENERIC_SVREF: /* generic sv */
14713 case SAVEt_SVREF: /* scalar reference */
14714 sv = (const SV *)POPPTR(ss,ix);
14715 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14716 if (type == SAVEt_SVREF)
14717 SvREFCNT_inc_simple_void((SV *)TOPPTR(nss,ix));
14718 ptr = POPPTR(ss,ix);
14719 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14721 case SAVEt_GVSLOT: /* any slot in GV */
14722 sv = (const SV *)POPPTR(ss,ix);
14723 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14724 ptr = POPPTR(ss,ix);
14725 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14726 sv = (const SV *)POPPTR(ss,ix);
14727 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14729 case SAVEt_HV: /* hash reference */
14730 case SAVEt_AV: /* array reference */
14731 sv = (const SV *) POPPTR(ss,ix);
14732 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14734 case SAVEt_COMPPAD:
14736 sv = (const SV *) POPPTR(ss,ix);
14737 TOPPTR(nss,ix) = sv_dup(sv, param);
14739 case SAVEt_INT: /* int reference */
14740 ptr = POPPTR(ss,ix);
14741 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14742 intval = (int)POPINT(ss,ix);
14743 TOPINT(nss,ix) = intval;
14745 case SAVEt_LONG: /* long reference */
14746 ptr = POPPTR(ss,ix);
14747 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14748 longval = (long)POPLONG(ss,ix);
14749 TOPLONG(nss,ix) = longval;
14751 case SAVEt_I32: /* I32 reference */
14752 ptr = POPPTR(ss,ix);
14753 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14755 TOPINT(nss,ix) = i;
14757 case SAVEt_IV: /* IV reference */
14758 case SAVEt_STRLEN: /* STRLEN/size_t ref */
14759 ptr = POPPTR(ss,ix);
14760 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14762 TOPIV(nss,ix) = iv;
14764 case SAVEt_TMPSFLOOR:
14766 TOPIV(nss,ix) = iv;
14768 case SAVEt_HPTR: /* HV* reference */
14769 case SAVEt_APTR: /* AV* reference */
14770 case SAVEt_SPTR: /* SV* reference */
14771 ptr = POPPTR(ss,ix);
14772 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14773 sv = (const SV *)POPPTR(ss,ix);
14774 TOPPTR(nss,ix) = sv_dup(sv, param);
14776 case SAVEt_VPTR: /* random* reference */
14777 ptr = POPPTR(ss,ix);
14778 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14780 case SAVEt_INT_SMALL:
14781 case SAVEt_I32_SMALL:
14782 case SAVEt_I16: /* I16 reference */
14783 case SAVEt_I8: /* I8 reference */
14785 ptr = POPPTR(ss,ix);
14786 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14788 case SAVEt_GENERIC_PVREF: /* generic char* */
14789 case SAVEt_PPTR: /* char* reference */
14790 ptr = POPPTR(ss,ix);
14791 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14792 c = (char*)POPPTR(ss,ix);
14793 TOPPTR(nss,ix) = pv_dup(c);
14795 case SAVEt_GP: /* scalar reference */
14796 gp = (GP*)POPPTR(ss,ix);
14797 TOPPTR(nss,ix) = gp = gp_dup(gp, param);
14798 (void)GpREFCNT_inc(gp);
14799 gv = (const GV *)POPPTR(ss,ix);
14800 TOPPTR(nss,ix) = gv_dup_inc(gv, param);
14803 ptr = POPPTR(ss,ix);
14804 if (ptr && (((OP*)ptr)->op_private & OPpREFCOUNTED)) {
14805 /* these are assumed to be refcounted properly */
14807 switch (((OP*)ptr)->op_type) {
14809 case OP_LEAVESUBLV:
14813 case OP_LEAVEWRITE:
14814 TOPPTR(nss,ix) = ptr;
14817 (void) OpREFCNT_inc(o);
14821 TOPPTR(nss,ix) = NULL;
14826 TOPPTR(nss,ix) = NULL;
14828 case SAVEt_FREECOPHH:
14829 ptr = POPPTR(ss,ix);
14830 TOPPTR(nss,ix) = cophh_copy((COPHH *)ptr);
14832 case SAVEt_ADELETE:
14833 av = (const AV *)POPPTR(ss,ix);
14834 TOPPTR(nss,ix) = av_dup_inc(av, param);
14836 TOPINT(nss,ix) = i;
14839 hv = (const HV *)POPPTR(ss,ix);
14840 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14842 TOPINT(nss,ix) = i;
14845 c = (char*)POPPTR(ss,ix);
14846 TOPPTR(nss,ix) = pv_dup_inc(c);
14848 case SAVEt_STACK_POS: /* Position on Perl stack */
14850 TOPINT(nss,ix) = i;
14852 case SAVEt_DESTRUCTOR:
14853 ptr = POPPTR(ss,ix);
14854 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14855 dptr = POPDPTR(ss,ix);
14856 TOPDPTR(nss,ix) = DPTR2FPTR(void (*)(void*),
14857 any_dup(FPTR2DPTR(void *, dptr),
14860 case SAVEt_DESTRUCTOR_X:
14861 ptr = POPPTR(ss,ix);
14862 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14863 dxptr = POPDXPTR(ss,ix);
14864 TOPDXPTR(nss,ix) = DPTR2FPTR(void (*)(pTHX_ void*),
14865 any_dup(FPTR2DPTR(void *, dxptr),
14868 case SAVEt_REGCONTEXT:
14870 ix -= uv >> SAVE_TIGHT_SHIFT;
14872 case SAVEt_AELEM: /* array element */
14873 sv = (const SV *)POPPTR(ss,ix);
14874 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14876 TOPIV(nss,ix) = iv;
14877 av = (const AV *)POPPTR(ss,ix);
14878 TOPPTR(nss,ix) = av_dup_inc(av, param);
14881 ptr = POPPTR(ss,ix);
14882 TOPPTR(nss,ix) = ptr;
14885 ptr = POPPTR(ss,ix);
14886 ptr = cophh_copy((COPHH*)ptr);
14887 TOPPTR(nss,ix) = ptr;
14889 TOPINT(nss,ix) = i;
14890 if (i & HINT_LOCALIZE_HH) {
14891 hv = (const HV *)POPPTR(ss,ix);
14892 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14895 case SAVEt_PADSV_AND_MORTALIZE:
14896 longval = (long)POPLONG(ss,ix);
14897 TOPLONG(nss,ix) = longval;
14898 ptr = POPPTR(ss,ix);
14899 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14900 sv = (const SV *)POPPTR(ss,ix);
14901 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14903 case SAVEt_SET_SVFLAGS:
14905 TOPINT(nss,ix) = i;
14907 TOPINT(nss,ix) = i;
14908 sv = (const SV *)POPPTR(ss,ix);
14909 TOPPTR(nss,ix) = sv_dup(sv, param);
14911 case SAVEt_COMPILE_WARNINGS:
14912 ptr = POPPTR(ss,ix);
14913 TOPPTR(nss,ix) = DUP_WARNINGS((STRLEN*)ptr);
14916 ptr = POPPTR(ss,ix);
14917 TOPPTR(nss,ix) = parser_dup((const yy_parser*)ptr, param);
14921 "panic: ss_dup inconsistency (%" IVdf ")", (IV) type);
14929 /* if sv is a stash, call $class->CLONE_SKIP(), and set the SVphv_CLONEABLE
14930 * flag to the result. This is done for each stash before cloning starts,
14931 * so we know which stashes want their objects cloned */
14934 do_mark_cloneable_stash(pTHX_ SV *const sv)
14936 const HEK * const hvname = HvNAME_HEK((const HV *)sv);
14938 GV* const cloner = gv_fetchmethod_autoload(MUTABLE_HV(sv), "CLONE_SKIP", 0);
14939 SvFLAGS(sv) |= SVphv_CLONEABLE; /* clone objects by default */
14940 if (cloner && GvCV(cloner)) {
14947 mXPUSHs(newSVhek(hvname));
14949 call_sv(MUTABLE_SV(GvCV(cloner)), G_SCALAR);
14956 SvFLAGS(sv) &= ~SVphv_CLONEABLE;
14964 =for apidoc perl_clone
14966 Create and return a new interpreter by cloning the current one.
14968 C<perl_clone> takes these flags as parameters:
14970 C<CLONEf_COPY_STACKS> - is used to, well, copy the stacks also,
14971 without it we only clone the data and zero the stacks,
14972 with it we copy the stacks and the new perl interpreter is
14973 ready to run at the exact same point as the previous one.
14974 The pseudo-fork code uses C<COPY_STACKS> while the
14975 threads->create doesn't.
14977 C<CLONEf_KEEP_PTR_TABLE> -
14978 C<perl_clone> keeps a ptr_table with the pointer of the old
14979 variable as a key and the new variable as a value,
14980 this allows it to check if something has been cloned and not
14981 clone it again but rather just use the value and increase the
14982 refcount. If C<KEEP_PTR_TABLE> is not set then C<perl_clone> will kill
14983 the ptr_table using the function
14984 C<ptr_table_free(PL_ptr_table); PL_ptr_table = NULL;>,
14985 reason to keep it around is if you want to dup some of your own
14986 variable who are outside the graph perl scans, an example of this
14987 code is in F<threads.xs> create.
14989 C<CLONEf_CLONE_HOST> -
14990 This is a win32 thing, it is ignored on unix, it tells perls
14991 win32host code (which is c++) to clone itself, this is needed on
14992 win32 if you want to run two threads at the same time,
14993 if you just want to do some stuff in a separate perl interpreter
14994 and then throw it away and return to the original one,
14995 you don't need to do anything.
15000 /* XXX the above needs expanding by someone who actually understands it ! */
15001 EXTERN_C PerlInterpreter *
15002 perl_clone_host(PerlInterpreter* proto_perl, UV flags);
15005 perl_clone(PerlInterpreter *proto_perl, UV flags)
15008 #ifdef PERL_IMPLICIT_SYS
15010 PERL_ARGS_ASSERT_PERL_CLONE;
15012 /* perlhost.h so we need to call into it
15013 to clone the host, CPerlHost should have a c interface, sky */
15015 #ifndef __amigaos4__
15016 if (flags & CLONEf_CLONE_HOST) {
15017 return perl_clone_host(proto_perl,flags);
15020 return perl_clone_using(proto_perl, flags,
15022 proto_perl->IMemShared,
15023 proto_perl->IMemParse,
15025 proto_perl->IStdIO,
15029 proto_perl->IProc);
15033 perl_clone_using(PerlInterpreter *proto_perl, UV flags,
15034 struct IPerlMem* ipM, struct IPerlMem* ipMS,
15035 struct IPerlMem* ipMP, struct IPerlEnv* ipE,
15036 struct IPerlStdIO* ipStd, struct IPerlLIO* ipLIO,
15037 struct IPerlDir* ipD, struct IPerlSock* ipS,
15038 struct IPerlProc* ipP)
15040 /* XXX many of the string copies here can be optimized if they're
15041 * constants; they need to be allocated as common memory and just
15042 * their pointers copied. */
15045 CLONE_PARAMS clone_params;
15046 CLONE_PARAMS* const param = &clone_params;
15048 PerlInterpreter * const my_perl = (PerlInterpreter*)(*ipM->pMalloc)(ipM, sizeof(PerlInterpreter));
15050 PERL_ARGS_ASSERT_PERL_CLONE_USING;
15051 #else /* !PERL_IMPLICIT_SYS */
15053 CLONE_PARAMS clone_params;
15054 CLONE_PARAMS* param = &clone_params;
15055 PerlInterpreter * const my_perl = (PerlInterpreter*)PerlMem_malloc(sizeof(PerlInterpreter));
15057 PERL_ARGS_ASSERT_PERL_CLONE;
15058 #endif /* PERL_IMPLICIT_SYS */
15060 /* for each stash, determine whether its objects should be cloned */
15061 S_visit(proto_perl, do_mark_cloneable_stash, SVt_PVHV, SVTYPEMASK);
15062 PERL_SET_THX(my_perl);
15065 PoisonNew(my_perl, 1, PerlInterpreter);
15068 PL_defstash = NULL; /* may be used by perl malloc() */
15071 PL_scopestack_name = 0;
15073 PL_savestack_ix = 0;
15074 PL_savestack_max = -1;
15075 PL_sig_pending = 0;
15077 Zero(&PL_debug_pad, 1, struct perl_debug_pad);
15078 Zero(&PL_padname_undef, 1, PADNAME);
15079 Zero(&PL_padname_const, 1, PADNAME);
15080 # ifdef DEBUG_LEAKING_SCALARS
15081 PL_sv_serial = (((UV)my_perl >> 2) & 0xfff) * 1000000;
15083 # ifdef PERL_TRACE_OPS
15084 Zero(PL_op_exec_cnt, OP_max+2, UV);
15086 #else /* !DEBUGGING */
15087 Zero(my_perl, 1, PerlInterpreter);
15088 #endif /* DEBUGGING */
15090 #ifdef PERL_IMPLICIT_SYS
15091 /* host pointers */
15093 PL_MemShared = ipMS;
15094 PL_MemParse = ipMP;
15101 #endif /* PERL_IMPLICIT_SYS */
15104 param->flags = flags;
15105 /* Nothing in the core code uses this, but we make it available to
15106 extensions (using mg_dup). */
15107 param->proto_perl = proto_perl;
15108 /* Likely nothing will use this, but it is initialised to be consistent
15109 with Perl_clone_params_new(). */
15110 param->new_perl = my_perl;
15111 param->unreferenced = NULL;
15114 INIT_TRACK_MEMPOOL(my_perl->Imemory_debug_header, my_perl);
15116 PL_body_arenas = NULL;
15117 Zero(&PL_body_roots, 1, PL_body_roots);
15121 PL_sv_arenaroot = NULL;
15123 PL_debug = proto_perl->Idebug;
15125 /* dbargs array probably holds garbage */
15128 PL_compiling = proto_perl->Icompiling;
15130 /* pseudo environmental stuff */
15131 PL_origargc = proto_perl->Iorigargc;
15132 PL_origargv = proto_perl->Iorigargv;
15134 #ifndef NO_TAINT_SUPPORT
15135 /* Set tainting stuff before PerlIO_debug can possibly get called */
15136 PL_tainting = proto_perl->Itainting;
15137 PL_taint_warn = proto_perl->Itaint_warn;
15139 PL_tainting = FALSE;
15140 PL_taint_warn = FALSE;
15143 PL_minus_c = proto_perl->Iminus_c;
15145 PL_localpatches = proto_perl->Ilocalpatches;
15146 PL_splitstr = proto_perl->Isplitstr;
15147 PL_minus_n = proto_perl->Iminus_n;
15148 PL_minus_p = proto_perl->Iminus_p;
15149 PL_minus_l = proto_perl->Iminus_l;
15150 PL_minus_a = proto_perl->Iminus_a;
15151 PL_minus_E = proto_perl->Iminus_E;
15152 PL_minus_F = proto_perl->Iminus_F;
15153 PL_doswitches = proto_perl->Idoswitches;
15154 PL_dowarn = proto_perl->Idowarn;
15155 #ifdef PERL_SAWAMPERSAND
15156 PL_sawampersand = proto_perl->Isawampersand;
15158 PL_unsafe = proto_perl->Iunsafe;
15159 PL_perldb = proto_perl->Iperldb;
15160 PL_perl_destruct_level = proto_perl->Iperl_destruct_level;
15161 PL_exit_flags = proto_perl->Iexit_flags;
15163 /* XXX time(&PL_basetime) when asked for? */
15164 PL_basetime = proto_perl->Ibasetime;
15166 PL_maxsysfd = proto_perl->Imaxsysfd;
15167 PL_statusvalue = proto_perl->Istatusvalue;
15169 PL_statusvalue_vms = proto_perl->Istatusvalue_vms;
15171 PL_statusvalue_posix = proto_perl->Istatusvalue_posix;
15174 /* RE engine related */
15175 PL_regmatch_slab = NULL;
15176 PL_reg_curpm = NULL;
15178 PL_sub_generation = proto_perl->Isub_generation;
15180 /* funky return mechanisms */
15181 PL_forkprocess = proto_perl->Iforkprocess;
15183 /* internal state */
15184 PL_main_start = proto_perl->Imain_start;
15185 PL_eval_root = proto_perl->Ieval_root;
15186 PL_eval_start = proto_perl->Ieval_start;
15188 PL_filemode = proto_perl->Ifilemode;
15189 PL_lastfd = proto_perl->Ilastfd;
15190 PL_oldname = proto_perl->Ioldname; /* XXX not quite right */
15191 PL_gensym = proto_perl->Igensym;
15193 PL_laststatval = proto_perl->Ilaststatval;
15194 PL_laststype = proto_perl->Ilaststype;
15197 PL_profiledata = NULL;
15199 PL_generation = proto_perl->Igeneration;
15201 PL_in_clean_objs = proto_perl->Iin_clean_objs;
15202 PL_in_clean_all = proto_perl->Iin_clean_all;
15204 PL_delaymagic_uid = proto_perl->Idelaymagic_uid;
15205 PL_delaymagic_euid = proto_perl->Idelaymagic_euid;
15206 PL_delaymagic_gid = proto_perl->Idelaymagic_gid;
15207 PL_delaymagic_egid = proto_perl->Idelaymagic_egid;
15208 PL_nomemok = proto_perl->Inomemok;
15209 PL_an = proto_perl->Ian;
15210 PL_evalseq = proto_perl->Ievalseq;
15211 PL_origenviron = proto_perl->Iorigenviron; /* XXX not quite right */
15212 PL_origalen = proto_perl->Iorigalen;
15214 PL_sighandlerp = proto_perl->Isighandlerp;
15216 PL_runops = proto_perl->Irunops;
15218 PL_subline = proto_perl->Isubline;
15220 PL_cv_has_eval = proto_perl->Icv_has_eval;
15223 PL_cryptseen = proto_perl->Icryptseen;
15226 #ifdef USE_LOCALE_COLLATE
15227 PL_collation_ix = proto_perl->Icollation_ix;
15228 PL_collation_standard = proto_perl->Icollation_standard;
15229 PL_collxfrm_base = proto_perl->Icollxfrm_base;
15230 PL_collxfrm_mult = proto_perl->Icollxfrm_mult;
15231 PL_strxfrm_max_cp = proto_perl->Istrxfrm_max_cp;
15232 #endif /* USE_LOCALE_COLLATE */
15234 #ifdef USE_LOCALE_NUMERIC
15235 PL_numeric_standard = proto_perl->Inumeric_standard;
15236 PL_numeric_underlying = proto_perl->Inumeric_underlying;
15237 PL_numeric_underlying_is_standard = proto_perl->Inumeric_underlying_is_standard;
15238 #endif /* !USE_LOCALE_NUMERIC */
15240 /* Did the locale setup indicate UTF-8? */
15241 PL_utf8locale = proto_perl->Iutf8locale;
15242 PL_in_utf8_CTYPE_locale = proto_perl->Iin_utf8_CTYPE_locale;
15243 PL_in_utf8_COLLATE_locale = proto_perl->Iin_utf8_COLLATE_locale;
15244 my_strlcpy(PL_locale_utf8ness, proto_perl->Ilocale_utf8ness, sizeof(PL_locale_utf8ness));
15245 #if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
15246 PL_lc_numeric_mutex_depth = 0;
15248 /* Unicode features (see perlrun/-C) */
15249 PL_unicode = proto_perl->Iunicode;
15251 /* Pre-5.8 signals control */
15252 PL_signals = proto_perl->Isignals;
15254 /* times() ticks per second */
15255 PL_clocktick = proto_perl->Iclocktick;
15257 /* Recursion stopper for PerlIO_find_layer */
15258 PL_in_load_module = proto_perl->Iin_load_module;
15260 /* sort() routine */
15261 PL_sort_RealCmp = proto_perl->Isort_RealCmp;
15263 /* Not really needed/useful since the reenrant_retint is "volatile",
15264 * but do it for consistency's sake. */
15265 PL_reentrant_retint = proto_perl->Ireentrant_retint;
15267 /* Hooks to shared SVs and locks. */
15268 PL_sharehook = proto_perl->Isharehook;
15269 PL_lockhook = proto_perl->Ilockhook;
15270 PL_unlockhook = proto_perl->Iunlockhook;
15271 PL_threadhook = proto_perl->Ithreadhook;
15272 PL_destroyhook = proto_perl->Idestroyhook;
15273 PL_signalhook = proto_perl->Isignalhook;
15275 PL_globhook = proto_perl->Iglobhook;
15278 PL_last_swash_hv = NULL; /* reinits on demand */
15279 PL_last_swash_klen = 0;
15280 PL_last_swash_key[0]= '\0';
15281 PL_last_swash_tmps = (U8*)NULL;
15282 PL_last_swash_slen = 0;
15284 PL_srand_called = proto_perl->Isrand_called;
15285 Copy(&(proto_perl->Irandom_state), &PL_random_state, 1, PL_RANDOM_STATE_TYPE);
15287 if (flags & CLONEf_COPY_STACKS) {
15288 /* next allocation will be PL_tmps_stack[PL_tmps_ix+1] */
15289 PL_tmps_ix = proto_perl->Itmps_ix;
15290 PL_tmps_max = proto_perl->Itmps_max;
15291 PL_tmps_floor = proto_perl->Itmps_floor;
15293 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
15294 * NOTE: unlike the others! */
15295 PL_scopestack_ix = proto_perl->Iscopestack_ix;
15296 PL_scopestack_max = proto_perl->Iscopestack_max;
15298 /* next SSPUSHFOO() sets PL_savestack[PL_savestack_ix]
15299 * NOTE: unlike the others! */
15300 PL_savestack_ix = proto_perl->Isavestack_ix;
15301 PL_savestack_max = proto_perl->Isavestack_max;
15304 PL_start_env = proto_perl->Istart_env; /* XXXXXX */
15305 PL_top_env = &PL_start_env;
15307 PL_op = proto_perl->Iop;
15310 PL_Xpv = (XPV*)NULL;
15311 my_perl->Ina = proto_perl->Ina;
15313 PL_statcache = proto_perl->Istatcache;
15315 #ifndef NO_TAINT_SUPPORT
15316 PL_tainted = proto_perl->Itainted;
15318 PL_tainted = FALSE;
15320 PL_curpm = proto_perl->Icurpm; /* XXX No PMOP ref count */
15322 PL_chopset = proto_perl->Ichopset; /* XXX never deallocated */
15324 PL_restartjmpenv = proto_perl->Irestartjmpenv;
15325 PL_restartop = proto_perl->Irestartop;
15326 PL_in_eval = proto_perl->Iin_eval;
15327 PL_delaymagic = proto_perl->Idelaymagic;
15328 PL_phase = proto_perl->Iphase;
15329 PL_localizing = proto_perl->Ilocalizing;
15331 PL_hv_fetch_ent_mh = NULL;
15332 PL_modcount = proto_perl->Imodcount;
15333 PL_lastgotoprobe = NULL;
15334 PL_dumpindent = proto_perl->Idumpindent;
15336 PL_efloatbuf = NULL; /* reinits on demand */
15337 PL_efloatsize = 0; /* reinits on demand */
15341 PL_colorset = 0; /* reinits PL_colors[] */
15342 /*PL_colors[6] = {0,0,0,0,0,0};*/
15344 /* Pluggable optimizer */
15345 PL_peepp = proto_perl->Ipeepp;
15346 PL_rpeepp = proto_perl->Irpeepp;
15347 /* op_free() hook */
15348 PL_opfreehook = proto_perl->Iopfreehook;
15350 #ifdef USE_REENTRANT_API
15351 /* XXX: things like -Dm will segfault here in perlio, but doing
15352 * PERL_SET_CONTEXT(proto_perl);
15353 * breaks too many other things
15355 Perl_reentrant_init(aTHX);
15358 /* create SV map for pointer relocation */
15359 PL_ptr_table = ptr_table_new();
15361 /* initialize these special pointers as early as possible */
15363 ptr_table_store(PL_ptr_table, &proto_perl->Isv_undef, &PL_sv_undef);
15364 ptr_table_store(PL_ptr_table, &proto_perl->Isv_no, &PL_sv_no);
15365 ptr_table_store(PL_ptr_table, &proto_perl->Isv_zero, &PL_sv_zero);
15366 ptr_table_store(PL_ptr_table, &proto_perl->Isv_yes, &PL_sv_yes);
15367 ptr_table_store(PL_ptr_table, &proto_perl->Ipadname_const,
15368 &PL_padname_const);
15370 /* create (a non-shared!) shared string table */
15371 PL_strtab = newHV();
15372 HvSHAREKEYS_off(PL_strtab);
15373 hv_ksplit(PL_strtab, HvTOTALKEYS(proto_perl->Istrtab));
15374 ptr_table_store(PL_ptr_table, proto_perl->Istrtab, PL_strtab);
15376 Zero(PL_sv_consts, SV_CONSTS_COUNT, SV*);
15378 /* This PV will be free'd special way so must set it same way op.c does */
15379 PL_compiling.cop_file = savesharedpv(PL_compiling.cop_file);
15380 ptr_table_store(PL_ptr_table, proto_perl->Icompiling.cop_file, PL_compiling.cop_file);
15382 ptr_table_store(PL_ptr_table, &proto_perl->Icompiling, &PL_compiling);
15383 PL_compiling.cop_warnings = DUP_WARNINGS(PL_compiling.cop_warnings);
15384 CopHINTHASH_set(&PL_compiling, cophh_copy(CopHINTHASH_get(&PL_compiling)));
15385 PL_curcop = (COP*)any_dup(proto_perl->Icurcop, proto_perl);
15387 param->stashes = newAV(); /* Setup array of objects to call clone on */
15388 /* This makes no difference to the implementation, as it always pushes
15389 and shifts pointers to other SVs without changing their reference
15390 count, with the array becoming empty before it is freed. However, it
15391 makes it conceptually clear what is going on, and will avoid some
15392 work inside av.c, filling slots between AvFILL() and AvMAX() with
15393 &PL_sv_undef, and SvREFCNT_dec()ing those. */
15394 AvREAL_off(param->stashes);
15396 if (!(flags & CLONEf_COPY_STACKS)) {
15397 param->unreferenced = newAV();
15400 #ifdef PERLIO_LAYERS
15401 /* Clone PerlIO tables as soon as we can handle general xx_dup() */
15402 PerlIO_clone(aTHX_ proto_perl, param);
15405 PL_envgv = gv_dup_inc(proto_perl->Ienvgv, param);
15406 PL_incgv = gv_dup_inc(proto_perl->Iincgv, param);
15407 PL_hintgv = gv_dup_inc(proto_perl->Ihintgv, param);
15408 PL_origfilename = SAVEPV(proto_perl->Iorigfilename);
15409 PL_xsubfilename = proto_perl->Ixsubfilename;
15410 PL_diehook = sv_dup_inc(proto_perl->Idiehook, param);
15411 PL_warnhook = sv_dup_inc(proto_perl->Iwarnhook, param);
15414 PL_patchlevel = sv_dup_inc(proto_perl->Ipatchlevel, param);
15415 PL_inplace = SAVEPV(proto_perl->Iinplace);
15416 PL_e_script = sv_dup_inc(proto_perl->Ie_script, param);
15418 /* magical thingies */
15420 SvPVCLEAR(PERL_DEBUG_PAD(0)); /* For regex debugging. */
15421 SvPVCLEAR(PERL_DEBUG_PAD(1)); /* ext/re needs these */
15422 SvPVCLEAR(PERL_DEBUG_PAD(2)); /* even without DEBUGGING. */
15425 /* Clone the regex array */
15426 /* ORANGE FIXME for plugins, probably in the SV dup code.
15427 newSViv(PTR2IV(CALLREGDUPE(
15428 INT2PTR(REGEXP *, SvIVX(regex)), param))))
15430 PL_regex_padav = av_dup_inc(proto_perl->Iregex_padav, param);
15431 PL_regex_pad = AvARRAY(PL_regex_padav);
15433 PL_stashpadmax = proto_perl->Istashpadmax;
15434 PL_stashpadix = proto_perl->Istashpadix ;
15435 Newx(PL_stashpad, PL_stashpadmax, HV *);
15438 for (; o < PL_stashpadmax; ++o)
15439 PL_stashpad[o] = hv_dup(proto_perl->Istashpad[o], param);
15442 /* shortcuts to various I/O objects */
15443 PL_ofsgv = gv_dup_inc(proto_perl->Iofsgv, param);
15444 PL_stdingv = gv_dup(proto_perl->Istdingv, param);
15445 PL_stderrgv = gv_dup(proto_perl->Istderrgv, param);
15446 PL_defgv = gv_dup(proto_perl->Idefgv, param);
15447 PL_argvgv = gv_dup_inc(proto_perl->Iargvgv, param);
15448 PL_argvoutgv = gv_dup(proto_perl->Iargvoutgv, param);
15449 PL_argvout_stack = av_dup_inc(proto_perl->Iargvout_stack, param);
15451 /* shortcuts to regexp stuff */
15452 PL_replgv = gv_dup_inc(proto_perl->Ireplgv, param);
15454 /* shortcuts to misc objects */
15455 PL_errgv = gv_dup(proto_perl->Ierrgv, param);
15457 /* shortcuts to debugging objects */
15458 PL_DBgv = gv_dup_inc(proto_perl->IDBgv, param);
15459 PL_DBline = gv_dup_inc(proto_perl->IDBline, param);
15460 PL_DBsub = gv_dup_inc(proto_perl->IDBsub, param);
15461 PL_DBsingle = sv_dup(proto_perl->IDBsingle, param);
15462 PL_DBtrace = sv_dup(proto_perl->IDBtrace, param);
15463 PL_DBsignal = sv_dup(proto_perl->IDBsignal, param);
15464 Copy(proto_perl->IDBcontrol, PL_DBcontrol, DBVARMG_COUNT, IV);
15466 /* symbol tables */
15467 PL_defstash = hv_dup_inc(proto_perl->Idefstash, param);
15468 PL_curstash = hv_dup_inc(proto_perl->Icurstash, param);
15469 PL_debstash = hv_dup(proto_perl->Idebstash, param);
15470 PL_globalstash = hv_dup(proto_perl->Iglobalstash, param);
15471 PL_curstname = sv_dup_inc(proto_perl->Icurstname, param);
15473 PL_beginav = av_dup_inc(proto_perl->Ibeginav, param);
15474 PL_beginav_save = av_dup_inc(proto_perl->Ibeginav_save, param);
15475 PL_checkav_save = av_dup_inc(proto_perl->Icheckav_save, param);
15476 PL_unitcheckav = av_dup_inc(proto_perl->Iunitcheckav, param);
15477 PL_unitcheckav_save = av_dup_inc(proto_perl->Iunitcheckav_save, param);
15478 PL_endav = av_dup_inc(proto_perl->Iendav, param);
15479 PL_checkav = av_dup_inc(proto_perl->Icheckav, param);
15480 PL_initav = av_dup_inc(proto_perl->Iinitav, param);
15481 PL_savebegin = proto_perl->Isavebegin;
15483 PL_isarev = hv_dup_inc(proto_perl->Iisarev, param);
15485 /* subprocess state */
15486 PL_fdpid = av_dup_inc(proto_perl->Ifdpid, param);
15488 if (proto_perl->Iop_mask)
15489 PL_op_mask = SAVEPVN(proto_perl->Iop_mask, PL_maxo);
15492 /* PL_asserting = proto_perl->Iasserting; */
15494 /* current interpreter roots */
15495 PL_main_cv = cv_dup_inc(proto_perl->Imain_cv, param);
15497 PL_main_root = OpREFCNT_inc(proto_perl->Imain_root);
15500 /* runtime control stuff */
15501 PL_curcopdb = (COP*)any_dup(proto_perl->Icurcopdb, proto_perl);
15503 PL_preambleav = av_dup_inc(proto_perl->Ipreambleav, param);
15505 PL_ors_sv = sv_dup_inc(proto_perl->Iors_sv, param);
15507 /* interpreter atexit processing */
15508 PL_exitlistlen = proto_perl->Iexitlistlen;
15509 if (PL_exitlistlen) {
15510 Newx(PL_exitlist, PL_exitlistlen, PerlExitListEntry);
15511 Copy(proto_perl->Iexitlist, PL_exitlist, PL_exitlistlen, PerlExitListEntry);
15514 PL_exitlist = (PerlExitListEntry*)NULL;
15516 PL_my_cxt_size = proto_perl->Imy_cxt_size;
15517 if (PL_my_cxt_size) {
15518 Newx(PL_my_cxt_list, PL_my_cxt_size, void *);
15519 Copy(proto_perl->Imy_cxt_list, PL_my_cxt_list, PL_my_cxt_size, void *);
15520 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
15521 Newx(PL_my_cxt_keys, PL_my_cxt_size, const char *);
15522 Copy(proto_perl->Imy_cxt_keys, PL_my_cxt_keys, PL_my_cxt_size, char *);
15526 PL_my_cxt_list = (void**)NULL;
15527 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
15528 PL_my_cxt_keys = (const char**)NULL;
15531 PL_modglobal = hv_dup_inc(proto_perl->Imodglobal, param);
15532 PL_custom_op_names = hv_dup_inc(proto_perl->Icustom_op_names,param);
15533 PL_custom_op_descs = hv_dup_inc(proto_perl->Icustom_op_descs,param);
15534 PL_custom_ops = hv_dup_inc(proto_perl->Icustom_ops, param);
15536 PL_compcv = cv_dup(proto_perl->Icompcv, param);
15538 PAD_CLONE_VARS(proto_perl, param);
15540 #ifdef HAVE_INTERP_INTERN
15541 sys_intern_dup(&proto_perl->Isys_intern, &PL_sys_intern);
15544 PL_DBcv = cv_dup(proto_perl->IDBcv, param);
15546 #ifdef PERL_USES_PL_PIDSTATUS
15547 PL_pidstatus = newHV(); /* XXX flag for cloning? */
15549 PL_osname = SAVEPV(proto_perl->Iosname);
15550 PL_parser = parser_dup(proto_perl->Iparser, param);
15552 /* XXX this only works if the saved cop has already been cloned */
15553 if (proto_perl->Iparser) {
15554 PL_parser->saved_curcop = (COP*)any_dup(
15555 proto_perl->Iparser->saved_curcop,
15559 PL_subname = sv_dup_inc(proto_perl->Isubname, param);
15561 #if defined(USE_POSIX_2008_LOCALE) \
15562 && defined(USE_THREAD_SAFE_LOCALE) \
15563 && ! defined(HAS_QUERYLOCALE)
15564 for (i = 0; i < (int) C_ARRAY_LENGTH(PL_curlocales); i++) {
15565 PL_curlocales[i] = savepv("."); /* An illegal value */
15568 #ifdef USE_LOCALE_CTYPE
15569 /* Should we warn if uses locale? */
15570 PL_warn_locale = sv_dup_inc(proto_perl->Iwarn_locale, param);
15573 #ifdef USE_LOCALE_COLLATE
15574 PL_collation_name = SAVEPV(proto_perl->Icollation_name);
15575 #endif /* USE_LOCALE_COLLATE */
15577 #ifdef USE_LOCALE_NUMERIC
15578 PL_numeric_name = SAVEPV(proto_perl->Inumeric_name);
15579 PL_numeric_radix_sv = sv_dup_inc(proto_perl->Inumeric_radix_sv, param);
15581 # if defined(HAS_POSIX_2008_LOCALE)
15582 PL_underlying_numeric_obj = NULL;
15584 #endif /* !USE_LOCALE_NUMERIC */
15586 PL_langinfo_buf = NULL;
15587 PL_langinfo_bufsize = 0;
15589 PL_setlocale_buf = NULL;
15590 PL_setlocale_bufsize = 0;
15592 /* utf8 character class swashes */
15593 PL_seen_deprecated_macro = hv_dup_inc(proto_perl->Iseen_deprecated_macro, param);
15595 if (proto_perl->Ipsig_pend) {
15596 Newxz(PL_psig_pend, SIG_SIZE, int);
15599 PL_psig_pend = (int*)NULL;
15602 if (proto_perl->Ipsig_name) {
15603 Newx(PL_psig_name, 2 * SIG_SIZE, SV*);
15604 sv_dup_inc_multiple(proto_perl->Ipsig_name, PL_psig_name, 2 * SIG_SIZE,
15606 PL_psig_ptr = PL_psig_name + SIG_SIZE;
15609 PL_psig_ptr = (SV**)NULL;
15610 PL_psig_name = (SV**)NULL;
15613 if (flags & CLONEf_COPY_STACKS) {
15614 Newx(PL_tmps_stack, PL_tmps_max, SV*);
15615 sv_dup_inc_multiple(proto_perl->Itmps_stack, PL_tmps_stack,
15616 PL_tmps_ix+1, param);
15618 /* next PUSHMARK() sets *(PL_markstack_ptr+1) */
15619 i = proto_perl->Imarkstack_max - proto_perl->Imarkstack;
15620 Newx(PL_markstack, i, I32);
15621 PL_markstack_max = PL_markstack + (proto_perl->Imarkstack_max
15622 - proto_perl->Imarkstack);
15623 PL_markstack_ptr = PL_markstack + (proto_perl->Imarkstack_ptr
15624 - proto_perl->Imarkstack);
15625 Copy(proto_perl->Imarkstack, PL_markstack,
15626 PL_markstack_ptr - PL_markstack + 1, I32);
15628 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
15629 * NOTE: unlike the others! */
15630 Newx(PL_scopestack, PL_scopestack_max, I32);
15631 Copy(proto_perl->Iscopestack, PL_scopestack, PL_scopestack_ix, I32);
15634 Newx(PL_scopestack_name, PL_scopestack_max, const char *);
15635 Copy(proto_perl->Iscopestack_name, PL_scopestack_name, PL_scopestack_ix, const char *);
15637 /* reset stack AV to correct length before its duped via
15638 * PL_curstackinfo */
15639 AvFILLp(proto_perl->Icurstack) =
15640 proto_perl->Istack_sp - proto_perl->Istack_base;
15642 /* NOTE: si_dup() looks at PL_markstack */
15643 PL_curstackinfo = si_dup(proto_perl->Icurstackinfo, param);
15645 /* PL_curstack = PL_curstackinfo->si_stack; */
15646 PL_curstack = av_dup(proto_perl->Icurstack, param);
15647 PL_mainstack = av_dup(proto_perl->Imainstack, param);
15649 /* next PUSHs() etc. set *(PL_stack_sp+1) */
15650 PL_stack_base = AvARRAY(PL_curstack);
15651 PL_stack_sp = PL_stack_base + (proto_perl->Istack_sp
15652 - proto_perl->Istack_base);
15653 PL_stack_max = PL_stack_base + AvMAX(PL_curstack);
15655 /*Newxz(PL_savestack, PL_savestack_max, ANY);*/
15656 PL_savestack = ss_dup(proto_perl, param);
15660 ENTER; /* perl_destruct() wants to LEAVE; */
15663 PL_statgv = gv_dup(proto_perl->Istatgv, param);
15664 PL_statname = sv_dup_inc(proto_perl->Istatname, param);
15666 PL_rs = sv_dup_inc(proto_perl->Irs, param);
15667 PL_last_in_gv = gv_dup(proto_perl->Ilast_in_gv, param);
15668 PL_defoutgv = gv_dup_inc(proto_perl->Idefoutgv, param);
15669 PL_toptarget = sv_dup_inc(proto_perl->Itoptarget, param);
15670 PL_bodytarget = sv_dup_inc(proto_perl->Ibodytarget, param);
15671 PL_formtarget = sv_dup(proto_perl->Iformtarget, param);
15673 PL_errors = sv_dup_inc(proto_perl->Ierrors, param);
15675 PL_sortcop = (OP*)any_dup(proto_perl->Isortcop, proto_perl);
15676 PL_firstgv = gv_dup_inc(proto_perl->Ifirstgv, param);
15677 PL_secondgv = gv_dup_inc(proto_perl->Isecondgv, param);
15679 PL_stashcache = newHV();
15681 PL_watchaddr = (char **) ptr_table_fetch(PL_ptr_table,
15682 proto_perl->Iwatchaddr);
15683 PL_watchok = PL_watchaddr ? * PL_watchaddr : NULL;
15684 if (PL_debug && PL_watchaddr) {
15685 PerlIO_printf(Perl_debug_log,
15686 "WATCHING: %" UVxf " cloned as %" UVxf " with value %" UVxf "\n",
15687 PTR2UV(proto_perl->Iwatchaddr), PTR2UV(PL_watchaddr),
15688 PTR2UV(PL_watchok));
15691 PL_registered_mros = hv_dup_inc(proto_perl->Iregistered_mros, param);
15692 PL_blockhooks = av_dup_inc(proto_perl->Iblockhooks, param);
15694 /* Call the ->CLONE method, if it exists, for each of the stashes
15695 identified by sv_dup() above.
15697 while(av_tindex(param->stashes) != -1) {
15698 HV* const stash = MUTABLE_HV(av_shift(param->stashes));
15699 GV* const cloner = gv_fetchmethod_autoload(stash, "CLONE", 0);
15700 if (cloner && GvCV(cloner)) {
15705 mXPUSHs(newSVhek(HvNAME_HEK(stash)));
15707 call_sv(MUTABLE_SV(GvCV(cloner)), G_DISCARD);
15713 if (!(flags & CLONEf_KEEP_PTR_TABLE)) {
15714 ptr_table_free(PL_ptr_table);
15715 PL_ptr_table = NULL;
15718 if (!(flags & CLONEf_COPY_STACKS)) {
15719 unreferenced_to_tmp_stack(param->unreferenced);
15722 SvREFCNT_dec(param->stashes);
15724 /* orphaned? eg threads->new inside BEGIN or use */
15725 if (PL_compcv && ! SvREFCNT(PL_compcv)) {
15726 SvREFCNT_inc_simple_void(PL_compcv);
15727 SAVEFREESV(PL_compcv);
15734 S_unreferenced_to_tmp_stack(pTHX_ AV *const unreferenced)
15736 PERL_ARGS_ASSERT_UNREFERENCED_TO_TMP_STACK;
15738 if (AvFILLp(unreferenced) > -1) {
15739 SV **svp = AvARRAY(unreferenced);
15740 SV **const last = svp + AvFILLp(unreferenced);
15744 if (SvREFCNT(*svp) == 1)
15746 } while (++svp <= last);
15748 EXTEND_MORTAL(count);
15749 svp = AvARRAY(unreferenced);
15752 if (SvREFCNT(*svp) == 1) {
15753 /* Our reference is the only one to this SV. This means that
15754 in this thread, the scalar effectively has a 0 reference.
15755 That doesn't work (cleanup never happens), so donate our
15756 reference to it onto the save stack. */
15757 PL_tmps_stack[++PL_tmps_ix] = *svp;
15759 /* As an optimisation, because we are already walking the
15760 entire array, instead of above doing either
15761 SvREFCNT_inc(*svp) or *svp = &PL_sv_undef, we can instead
15762 release our reference to the scalar, so that at the end of
15763 the array owns zero references to the scalars it happens to
15764 point to. We are effectively converting the array from
15765 AvREAL() on to AvREAL() off. This saves the av_clear()
15766 (triggered by the SvREFCNT_dec(unreferenced) below) from
15767 walking the array a second time. */
15768 SvREFCNT_dec(*svp);
15771 } while (++svp <= last);
15772 AvREAL_off(unreferenced);
15774 SvREFCNT_dec_NN(unreferenced);
15778 Perl_clone_params_del(CLONE_PARAMS *param)
15780 /* This seemingly funky ordering keeps the build with PERL_GLOBAL_STRUCT
15782 PerlInterpreter *const to = param->new_perl;
15784 PerlInterpreter *const was = PERL_GET_THX;
15786 PERL_ARGS_ASSERT_CLONE_PARAMS_DEL;
15792 SvREFCNT_dec(param->stashes);
15793 if (param->unreferenced)
15794 unreferenced_to_tmp_stack(param->unreferenced);
15804 Perl_clone_params_new(PerlInterpreter *const from, PerlInterpreter *const to)
15807 /* Need to play this game, as newAV() can call safesysmalloc(), and that
15808 does a dTHX; to get the context from thread local storage.
15809 FIXME - under PERL_CORE Newx(), Safefree() and friends should expand to
15810 a version that passes in my_perl. */
15811 PerlInterpreter *const was = PERL_GET_THX;
15812 CLONE_PARAMS *param;
15814 PERL_ARGS_ASSERT_CLONE_PARAMS_NEW;
15820 /* Given that we've set the context, we can do this unshared. */
15821 Newx(param, 1, CLONE_PARAMS);
15824 param->proto_perl = from;
15825 param->new_perl = to;
15826 param->stashes = (AV *)Perl_newSV_type(to, SVt_PVAV);
15827 AvREAL_off(param->stashes);
15828 param->unreferenced = (AV *)Perl_newSV_type(to, SVt_PVAV);
15836 #endif /* USE_ITHREADS */
15839 Perl_init_constants(pTHX)
15841 SvREFCNT(&PL_sv_undef) = SvREFCNT_IMMORTAL;
15842 SvFLAGS(&PL_sv_undef) = SVf_READONLY|SVf_PROTECT|SVt_NULL;
15843 SvANY(&PL_sv_undef) = NULL;
15845 SvANY(&PL_sv_no) = new_XPVNV();
15846 SvREFCNT(&PL_sv_no) = SvREFCNT_IMMORTAL;
15847 SvFLAGS(&PL_sv_no) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15848 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15851 SvANY(&PL_sv_yes) = new_XPVNV();
15852 SvREFCNT(&PL_sv_yes) = SvREFCNT_IMMORTAL;
15853 SvFLAGS(&PL_sv_yes) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15854 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15857 SvANY(&PL_sv_zero) = new_XPVNV();
15858 SvREFCNT(&PL_sv_zero) = SvREFCNT_IMMORTAL;
15859 SvFLAGS(&PL_sv_zero) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15860 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15864 SvPV_set(&PL_sv_no, (char*)PL_No);
15865 SvCUR_set(&PL_sv_no, 0);
15866 SvLEN_set(&PL_sv_no, 0);
15867 SvIV_set(&PL_sv_no, 0);
15868 SvNV_set(&PL_sv_no, 0);
15870 SvPV_set(&PL_sv_yes, (char*)PL_Yes);
15871 SvCUR_set(&PL_sv_yes, 1);
15872 SvLEN_set(&PL_sv_yes, 0);
15873 SvIV_set(&PL_sv_yes, 1);
15874 SvNV_set(&PL_sv_yes, 1);
15876 SvPV_set(&PL_sv_zero, (char*)PL_Zero);
15877 SvCUR_set(&PL_sv_zero, 1);
15878 SvLEN_set(&PL_sv_zero, 0);
15879 SvIV_set(&PL_sv_zero, 0);
15880 SvNV_set(&PL_sv_zero, 0);
15882 PadnamePV(&PL_padname_const) = (char *)PL_No;
15884 assert(SvIMMORTAL_INTERP(&PL_sv_yes));
15885 assert(SvIMMORTAL_INTERP(&PL_sv_undef));
15886 assert(SvIMMORTAL_INTERP(&PL_sv_no));
15887 assert(SvIMMORTAL_INTERP(&PL_sv_zero));
15889 assert(SvIMMORTAL(&PL_sv_yes));
15890 assert(SvIMMORTAL(&PL_sv_undef));
15891 assert(SvIMMORTAL(&PL_sv_no));
15892 assert(SvIMMORTAL(&PL_sv_zero));
15894 assert( SvIMMORTAL_TRUE(&PL_sv_yes));
15895 assert(!SvIMMORTAL_TRUE(&PL_sv_undef));
15896 assert(!SvIMMORTAL_TRUE(&PL_sv_no));
15897 assert(!SvIMMORTAL_TRUE(&PL_sv_zero));
15899 assert( SvTRUE_nomg_NN(&PL_sv_yes));
15900 assert(!SvTRUE_nomg_NN(&PL_sv_undef));
15901 assert(!SvTRUE_nomg_NN(&PL_sv_no));
15902 assert(!SvTRUE_nomg_NN(&PL_sv_zero));
15906 =head1 Unicode Support
15908 =for apidoc sv_recode_to_utf8
15910 C<encoding> is assumed to be an C<Encode> object, on entry the PV
15911 of C<sv> is assumed to be octets in that encoding, and C<sv>
15912 will be converted into Unicode (and UTF-8).
15914 If C<sv> already is UTF-8 (or if it is not C<POK>), or if C<encoding>
15915 is not a reference, nothing is done to C<sv>. If C<encoding> is not
15916 an C<Encode::XS> Encoding object, bad things will happen.
15917 (See F<cpan/Encode/encoding.pm> and L<Encode>.)
15919 The PV of C<sv> is returned.
15924 Perl_sv_recode_to_utf8(pTHX_ SV *sv, SV *encoding)
15926 PERL_ARGS_ASSERT_SV_RECODE_TO_UTF8;
15928 if (SvPOK(sv) && !SvUTF8(sv) && !IN_BYTES && SvROK(encoding)) {
15937 if (SvPADTMP(nsv)) {
15938 nsv = sv_newmortal();
15939 SvSetSV_nosteal(nsv, sv);
15948 Passing sv_yes is wrong - it needs to be or'ed set of constants
15949 for Encode::XS, while UTf-8 decode (currently) assumes a true value means
15950 remove converted chars from source.
15952 Both will default the value - let them.
15954 XPUSHs(&PL_sv_yes);
15957 call_method("decode", G_SCALAR);
15961 s = SvPV_const(uni, len);
15962 if (s != SvPVX_const(sv)) {
15963 SvGROW(sv, len + 1);
15964 Move(s, SvPVX(sv), len + 1, char);
15965 SvCUR_set(sv, len);
15970 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
15971 /* clear pos and any utf8 cache */
15972 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
15975 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
15976 magic_setutf8(sv,mg); /* clear UTF8 cache */
15981 return SvPOKp(sv) ? SvPVX(sv) : NULL;
15985 =for apidoc sv_cat_decode
15987 C<encoding> is assumed to be an C<Encode> object, the PV of C<ssv> is
15988 assumed to be octets in that encoding and decoding the input starts
15989 from the position which S<C<(PV + *offset)>> pointed to. C<dsv> will be
15990 concatenated with the decoded UTF-8 string from C<ssv>. Decoding will terminate
15991 when the string C<tstr> appears in decoding output or the input ends on
15992 the PV of C<ssv>. The value which C<offset> points will be modified
15993 to the last input position on C<ssv>.
15995 Returns TRUE if the terminator was found, else returns FALSE.
16000 Perl_sv_cat_decode(pTHX_ SV *dsv, SV *encoding,
16001 SV *ssv, int *offset, char *tstr, int tlen)
16005 PERL_ARGS_ASSERT_SV_CAT_DECODE;
16007 if (SvPOK(ssv) && SvPOK(dsv) && SvROK(encoding)) {
16018 offsv = newSViv(*offset);
16020 mPUSHp(tstr, tlen);
16022 call_method("cat_decode", G_SCALAR);
16024 ret = SvTRUE(TOPs);
16025 *offset = SvIV(offsv);
16031 Perl_croak(aTHX_ "Invalid argument to sv_cat_decode");
16036 /* ---------------------------------------------------------------------
16038 * support functions for report_uninit()
16041 /* the maxiumum size of array or hash where we will scan looking
16042 * for the undefined element that triggered the warning */
16044 #define FUV_MAX_SEARCH_SIZE 1000
16046 /* Look for an entry in the hash whose value has the same SV as val;
16047 * If so, return a mortal copy of the key. */
16050 S_find_hash_subscript(pTHX_ const HV *const hv, const SV *const val)
16056 PERL_ARGS_ASSERT_FIND_HASH_SUBSCRIPT;
16058 if (!hv || SvMAGICAL(hv) || !HvARRAY(hv) ||
16059 (HvTOTALKEYS(hv) > FUV_MAX_SEARCH_SIZE))
16062 array = HvARRAY(hv);
16064 for (i=HvMAX(hv); i>=0; i--) {
16066 for (entry = array[i]; entry; entry = HeNEXT(entry)) {
16067 if (HeVAL(entry) != val)
16069 if ( HeVAL(entry) == &PL_sv_undef ||
16070 HeVAL(entry) == &PL_sv_placeholder)
16074 if (HeKLEN(entry) == HEf_SVKEY)
16075 return sv_mortalcopy(HeKEY_sv(entry));
16076 return sv_2mortal(newSVhek(HeKEY_hek(entry)));
16082 /* Look for an entry in the array whose value has the same SV as val;
16083 * If so, return the index, otherwise return -1. */
16086 S_find_array_subscript(pTHX_ const AV *const av, const SV *const val)
16088 PERL_ARGS_ASSERT_FIND_ARRAY_SUBSCRIPT;
16090 if (!av || SvMAGICAL(av) || !AvARRAY(av) ||
16091 (AvFILLp(av) > FUV_MAX_SEARCH_SIZE))
16094 if (val != &PL_sv_undef) {
16095 SV ** const svp = AvARRAY(av);
16098 for (i=AvFILLp(av); i>=0; i--)
16105 /* varname(): return the name of a variable, optionally with a subscript.
16106 * If gv is non-zero, use the name of that global, along with gvtype (one
16107 * of "$", "@", "%"); otherwise use the name of the lexical at pad offset
16108 * targ. Depending on the value of the subscript_type flag, return:
16111 #define FUV_SUBSCRIPT_NONE 1 /* "@foo" */
16112 #define FUV_SUBSCRIPT_ARRAY 2 /* "$foo[aindex]" */
16113 #define FUV_SUBSCRIPT_HASH 3 /* "$foo{keyname}" */
16114 #define FUV_SUBSCRIPT_WITHIN 4 /* "within @foo" */
16117 Perl_varname(pTHX_ const GV *const gv, const char gvtype, PADOFFSET targ,
16118 const SV *const keyname, SSize_t aindex, int subscript_type)
16121 SV * const name = sv_newmortal();
16122 if (gv && isGV(gv)) {
16124 buffer[0] = gvtype;
16127 /* as gv_fullname4(), but add literal '^' for $^FOO names */
16129 gv_fullname4(name, gv, buffer, 0);
16131 if ((unsigned int)SvPVX(name)[1] <= 26) {
16133 buffer[1] = SvPVX(name)[1] + 'A' - 1;
16135 /* Swap the 1 unprintable control character for the 2 byte pretty
16136 version - ie substr($name, 1, 1) = $buffer; */
16137 sv_insert(name, 1, 1, buffer, 2);
16141 CV * const cv = gv ? ((CV *)gv) : find_runcv(NULL);
16144 assert(!cv || SvTYPE(cv) == SVt_PVCV || SvTYPE(cv) == SVt_PVFM);
16146 if (!cv || !CvPADLIST(cv))
16148 sv = padnamelist_fetch(PadlistNAMES(CvPADLIST(cv)), targ);
16149 sv_setpvn(name, PadnamePV(sv), PadnameLEN(sv));
16153 if (subscript_type == FUV_SUBSCRIPT_HASH) {
16154 SV * const sv = newSV(0);
16156 const char * const pv = SvPV_nomg_const((SV*)keyname, len);
16158 *SvPVX(name) = '$';
16159 Perl_sv_catpvf(aTHX_ name, "{%s}",
16160 pv_pretty(sv, pv, len, 32, NULL, NULL,
16161 PERL_PV_PRETTY_DUMP | PERL_PV_ESCAPE_UNI_DETECT ));
16162 SvREFCNT_dec_NN(sv);
16164 else if (subscript_type == FUV_SUBSCRIPT_ARRAY) {
16165 *SvPVX(name) = '$';
16166 Perl_sv_catpvf(aTHX_ name, "[%" IVdf "]", (IV)aindex);
16168 else if (subscript_type == FUV_SUBSCRIPT_WITHIN) {
16169 /* We know that name has no magic, so can use 0 instead of SV_GMAGIC */
16170 Perl_sv_insert_flags(aTHX_ name, 0, 0, STR_WITH_LEN("within "), 0);
16178 =for apidoc find_uninit_var
16180 Find the name of the undefined variable (if any) that caused the operator
16181 to issue a "Use of uninitialized value" warning.
16182 If match is true, only return a name if its value matches C<uninit_sv>.
16183 So roughly speaking, if a unary operator (such as C<OP_COS>) generates a
16184 warning, then following the direct child of the op may yield an
16185 C<OP_PADSV> or C<OP_GV> that gives the name of the undefined variable. On the
16186 other hand, with C<OP_ADD> there are two branches to follow, so we only print
16187 the variable name if we get an exact match.
16188 C<desc_p> points to a string pointer holding the description of the op.
16189 This may be updated if needed.
16191 The name is returned as a mortal SV.
16193 Assumes that C<PL_op> is the OP that originally triggered the error, and that
16194 C<PL_comppad>/C<PL_curpad> points to the currently executing pad.
16200 S_find_uninit_var(pTHX_ const OP *const obase, const SV *const uninit_sv,
16201 bool match, const char **desc_p)
16206 const OP *o, *o2, *kid;
16208 PERL_ARGS_ASSERT_FIND_UNINIT_VAR;
16210 if (!obase || (match && (!uninit_sv || uninit_sv == &PL_sv_undef ||
16211 uninit_sv == &PL_sv_placeholder)))
16214 switch (obase->op_type) {
16217 /* undef should care if its args are undef - any warnings
16218 * will be from tied/magic vars */
16226 const bool pad = ( obase->op_type == OP_PADAV
16227 || obase->op_type == OP_PADHV
16228 || obase->op_type == OP_PADRANGE
16231 const bool hash = ( obase->op_type == OP_PADHV
16232 || obase->op_type == OP_RV2HV
16233 || (obase->op_type == OP_PADRANGE
16234 && SvTYPE(PAD_SVl(obase->op_targ)) == SVt_PVHV)
16238 int subscript_type = FUV_SUBSCRIPT_WITHIN;
16240 if (pad) { /* @lex, %lex */
16241 sv = PAD_SVl(obase->op_targ);
16245 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
16246 /* @global, %global */
16247 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
16250 sv = hash ? MUTABLE_SV(GvHV(gv)): MUTABLE_SV(GvAV(gv));
16252 else if (obase == PL_op) /* @{expr}, %{expr} */
16253 return find_uninit_var(cUNOPx(obase)->op_first,
16254 uninit_sv, match, desc_p);
16255 else /* @{expr}, %{expr} as a sub-expression */
16259 /* attempt to find a match within the aggregate */
16261 keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16263 subscript_type = FUV_SUBSCRIPT_HASH;
16266 index = find_array_subscript((const AV *)sv, uninit_sv);
16268 subscript_type = FUV_SUBSCRIPT_ARRAY;
16271 if (match && subscript_type == FUV_SUBSCRIPT_WITHIN)
16274 return varname(gv, (char)(hash ? '%' : '@'), obase->op_targ,
16275 keysv, index, subscript_type);
16279 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
16281 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
16282 if (!gv || !GvSTASH(gv))
16284 if (match && (GvSV(gv) != uninit_sv))
16286 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
16289 return find_uninit_var(cUNOPx(obase)->op_first, uninit_sv, 1, desc_p);
16292 if (match && PAD_SVl(obase->op_targ) != uninit_sv)
16294 return varname(NULL, '$', obase->op_targ,
16295 NULL, 0, FUV_SUBSCRIPT_NONE);
16298 gv = cGVOPx_gv(obase);
16299 if (!gv || (match && GvSV(gv) != uninit_sv) || !GvSTASH(gv))
16301 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
16303 case OP_AELEMFAST_LEX:
16306 AV *av = MUTABLE_AV(PAD_SV(obase->op_targ));
16307 if (!av || SvRMAGICAL(av))
16309 svp = av_fetch(av, (I8)obase->op_private, FALSE);
16310 if (!svp || *svp != uninit_sv)
16313 return varname(NULL, '$', obase->op_targ,
16314 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
16317 gv = cGVOPx_gv(obase);
16322 AV *const av = GvAV(gv);
16323 if (!av || SvRMAGICAL(av))
16325 svp = av_fetch(av, (I8)obase->op_private, FALSE);
16326 if (!svp || *svp != uninit_sv)
16329 return varname(gv, '$', 0,
16330 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
16332 NOT_REACHED; /* NOTREACHED */
16335 o = cUNOPx(obase)->op_first;
16336 if (!o || o->op_type != OP_NULL ||
16337 ! (o->op_targ == OP_AELEM || o->op_targ == OP_HELEM))
16339 return find_uninit_var(cBINOPo->op_last, uninit_sv, match, desc_p);
16344 bool negate = FALSE;
16346 if (PL_op == obase)
16347 /* $a[uninit_expr] or $h{uninit_expr} */
16348 return find_uninit_var(cBINOPx(obase)->op_last,
16349 uninit_sv, match, desc_p);
16352 o = cBINOPx(obase)->op_first;
16353 kid = cBINOPx(obase)->op_last;
16355 /* get the av or hv, and optionally the gv */
16357 if (o->op_type == OP_PADAV || o->op_type == OP_PADHV) {
16358 sv = PAD_SV(o->op_targ);
16360 else if ((o->op_type == OP_RV2AV || o->op_type == OP_RV2HV)
16361 && cUNOPo->op_first->op_type == OP_GV)
16363 gv = cGVOPx_gv(cUNOPo->op_first);
16367 == OP_RV2HV ? MUTABLE_SV(GvHV(gv)) : MUTABLE_SV(GvAV(gv));
16372 if (kid && kid->op_type == OP_NEGATE) {
16374 kid = cUNOPx(kid)->op_first;
16377 if (kid && kid->op_type == OP_CONST && SvOK(cSVOPx_sv(kid))) {
16378 /* index is constant */
16381 kidsv = newSVpvs_flags("-", SVs_TEMP);
16382 sv_catsv(kidsv, cSVOPx_sv(kid));
16385 kidsv = cSVOPx_sv(kid);
16389 if (obase->op_type == OP_HELEM) {
16390 HE* he = hv_fetch_ent(MUTABLE_HV(sv), kidsv, 0, 0);
16391 if (!he || HeVAL(he) != uninit_sv)
16395 SV * const opsv = cSVOPx_sv(kid);
16396 const IV opsviv = SvIV(opsv);
16397 SV * const * const svp = av_fetch(MUTABLE_AV(sv),
16398 negate ? - opsviv : opsviv,
16400 if (!svp || *svp != uninit_sv)
16404 if (obase->op_type == OP_HELEM)
16405 return varname(gv, '%', o->op_targ,
16406 kidsv, 0, FUV_SUBSCRIPT_HASH);
16408 return varname(gv, '@', o->op_targ, NULL,
16409 negate ? - SvIV(cSVOPx_sv(kid)) : SvIV(cSVOPx_sv(kid)),
16410 FUV_SUBSCRIPT_ARRAY);
16413 /* index is an expression;
16414 * attempt to find a match within the aggregate */
16415 if (obase->op_type == OP_HELEM) {
16416 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16418 return varname(gv, '%', o->op_targ,
16419 keysv, 0, FUV_SUBSCRIPT_HASH);
16422 const SSize_t index
16423 = find_array_subscript((const AV *)sv, uninit_sv);
16425 return varname(gv, '@', o->op_targ,
16426 NULL, index, FUV_SUBSCRIPT_ARRAY);
16431 (char)((o->op_type == OP_PADAV || o->op_type == OP_RV2AV)
16433 o->op_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
16435 NOT_REACHED; /* NOTREACHED */
16438 case OP_MULTIDEREF: {
16439 /* If we were executing OP_MULTIDEREF when the undef warning
16440 * triggered, then it must be one of the index values within
16441 * that triggered it. If not, then the only possibility is that
16442 * the value retrieved by the last aggregate index might be the
16443 * culprit. For the former, we set PL_multideref_pc each time before
16444 * using an index, so work though the item list until we reach
16445 * that point. For the latter, just work through the entire item
16446 * list; the last aggregate retrieved will be the candidate.
16447 * There is a third rare possibility: something triggered
16448 * magic while fetching an array/hash element. Just display
16449 * nothing in this case.
16452 /* the named aggregate, if any */
16453 PADOFFSET agg_targ = 0;
16455 /* the last-seen index */
16457 PADOFFSET index_targ;
16459 IV index_const_iv = 0; /* init for spurious compiler warn */
16460 SV *index_const_sv;
16461 int depth = 0; /* how many array/hash lookups we've done */
16463 UNOP_AUX_item *items = cUNOP_AUXx(obase)->op_aux;
16464 UNOP_AUX_item *last = NULL;
16465 UV actions = items->uv;
16468 if (PL_op == obase) {
16469 last = PL_multideref_pc;
16470 assert(last >= items && last <= items + items[-1].uv);
16477 switch (actions & MDEREF_ACTION_MASK) {
16479 case MDEREF_reload:
16480 actions = (++items)->uv;
16483 case MDEREF_HV_padhv_helem: /* $lex{...} */
16486 case MDEREF_AV_padav_aelem: /* $lex[...] */
16487 agg_targ = (++items)->pad_offset;
16491 case MDEREF_HV_gvhv_helem: /* $pkg{...} */
16494 case MDEREF_AV_gvav_aelem: /* $pkg[...] */
16496 agg_gv = (GV*)UNOP_AUX_item_sv(++items);
16497 assert(isGV_with_GP(agg_gv));
16500 case MDEREF_HV_gvsv_vivify_rv2hv_helem: /* $pkg->{...} */
16501 case MDEREF_HV_padsv_vivify_rv2hv_helem: /* $lex->{...} */
16504 case MDEREF_HV_pop_rv2hv_helem: /* expr->{...} */
16505 case MDEREF_HV_vivify_rv2hv_helem: /* vivify, ->{...} */
16511 case MDEREF_AV_gvsv_vivify_rv2av_aelem: /* $pkg->[...] */
16512 case MDEREF_AV_padsv_vivify_rv2av_aelem: /* $lex->[...] */
16515 case MDEREF_AV_pop_rv2av_aelem: /* expr->[...] */
16516 case MDEREF_AV_vivify_rv2av_aelem: /* vivify, ->[...] */
16523 index_const_sv = NULL;
16525 index_type = (actions & MDEREF_INDEX_MASK);
16526 switch (index_type) {
16527 case MDEREF_INDEX_none:
16529 case MDEREF_INDEX_const:
16531 index_const_sv = UNOP_AUX_item_sv(++items)
16533 index_const_iv = (++items)->iv;
16535 case MDEREF_INDEX_padsv:
16536 index_targ = (++items)->pad_offset;
16538 case MDEREF_INDEX_gvsv:
16539 index_gv = (GV*)UNOP_AUX_item_sv(++items);
16540 assert(isGV_with_GP(index_gv));
16544 if (index_type != MDEREF_INDEX_none)
16547 if ( index_type == MDEREF_INDEX_none
16548 || (actions & MDEREF_FLAG_last)
16549 || (last && items >= last)
16553 actions >>= MDEREF_SHIFT;
16556 if (PL_op == obase) {
16557 /* most likely index was undef */
16559 *desc_p = ( (actions & MDEREF_FLAG_last)
16560 && (obase->op_private
16561 & (OPpMULTIDEREF_EXISTS|OPpMULTIDEREF_DELETE)))
16563 (obase->op_private & OPpMULTIDEREF_EXISTS)
16566 : is_hv ? "hash element" : "array element";
16567 assert(index_type != MDEREF_INDEX_none);
16569 if (GvSV(index_gv) == uninit_sv)
16570 return varname(index_gv, '$', 0, NULL, 0,
16571 FUV_SUBSCRIPT_NONE);
16576 if (PL_curpad[index_targ] == uninit_sv)
16577 return varname(NULL, '$', index_targ,
16578 NULL, 0, FUV_SUBSCRIPT_NONE);
16582 /* If we got to this point it was undef on a const subscript,
16583 * so magic probably involved, e.g. $ISA[0]. Give up. */
16587 /* the SV returned by pp_multideref() was undef, if anything was */
16593 sv = PAD_SV(agg_targ);
16595 sv = is_hv ? MUTABLE_SV(GvHV(agg_gv)) : MUTABLE_SV(GvAV(agg_gv));
16599 if (index_type == MDEREF_INDEX_const) {
16604 HE* he = hv_fetch_ent(MUTABLE_HV(sv), index_const_sv, 0, 0);
16605 if (!he || HeVAL(he) != uninit_sv)
16609 SV * const * const svp =
16610 av_fetch(MUTABLE_AV(sv), index_const_iv, FALSE);
16611 if (!svp || *svp != uninit_sv)
16616 ? varname(agg_gv, '%', agg_targ,
16617 index_const_sv, 0, FUV_SUBSCRIPT_HASH)
16618 : varname(agg_gv, '@', agg_targ,
16619 NULL, index_const_iv, FUV_SUBSCRIPT_ARRAY);
16622 /* index is an var */
16624 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16626 return varname(agg_gv, '%', agg_targ,
16627 keysv, 0, FUV_SUBSCRIPT_HASH);
16630 const SSize_t index
16631 = find_array_subscript((const AV *)sv, uninit_sv);
16633 return varname(agg_gv, '@', agg_targ,
16634 NULL, index, FUV_SUBSCRIPT_ARRAY);
16638 return varname(agg_gv,
16640 agg_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
16642 NOT_REACHED; /* NOTREACHED */
16646 /* only examine RHS */
16647 return find_uninit_var(cBINOPx(obase)->op_first, uninit_sv,
16651 o = cUNOPx(obase)->op_first;
16652 if ( o->op_type == OP_PUSHMARK
16653 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)
16657 if (!OpHAS_SIBLING(o)) {
16658 /* one-arg version of open is highly magical */
16660 if (o->op_type == OP_GV) { /* open FOO; */
16662 if (match && GvSV(gv) != uninit_sv)
16664 return varname(gv, '$', 0,
16665 NULL, 0, FUV_SUBSCRIPT_NONE);
16667 /* other possibilities not handled are:
16668 * open $x; or open my $x; should return '${*$x}'
16669 * open expr; should return '$'.expr ideally
16676 /* ops where $_ may be an implicit arg */
16681 if ( !(obase->op_flags & OPf_STACKED)) {
16682 if (uninit_sv == DEFSV)
16683 return newSVpvs_flags("$_", SVs_TEMP);
16684 else if (obase->op_targ
16685 && uninit_sv == PAD_SVl(obase->op_targ))
16686 return varname(NULL, '$', obase->op_targ, NULL, 0,
16687 FUV_SUBSCRIPT_NONE);
16694 match = 1; /* print etc can return undef on defined args */
16695 /* skip filehandle as it can't produce 'undef' warning */
16696 o = cUNOPx(obase)->op_first;
16697 if ((obase->op_flags & OPf_STACKED)
16699 ( o->op_type == OP_PUSHMARK
16700 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)))
16701 o = OpSIBLING(OpSIBLING(o));
16705 case OP_ENTEREVAL: /* could be eval $undef or $x='$undef'; eval $x */
16706 case OP_CUSTOM: /* XS or custom code could trigger random warnings */
16708 /* the following ops are capable of returning PL_sv_undef even for
16709 * defined arg(s) */
16728 case OP_GETPEERNAME:
16775 case OP_SMARTMATCH:
16784 /* XXX tmp hack: these two may call an XS sub, and currently
16785 XS subs don't have a SUB entry on the context stack, so CV and
16786 pad determination goes wrong, and BAD things happen. So, just
16787 don't try to determine the value under those circumstances.
16788 Need a better fix at dome point. DAPM 11/2007 */
16794 GV * const gv = gv_fetchpvs(".", GV_NOTQUAL, SVt_PV);
16795 if (gv && GvSV(gv) == uninit_sv)
16796 return newSVpvs_flags("$.", SVs_TEMP);
16801 /* def-ness of rval pos() is independent of the def-ness of its arg */
16802 if ( !(obase->op_flags & OPf_MOD))
16808 if (SvROK(PL_rs) && uninit_sv == SvRV(PL_rs))
16809 return newSVpvs_flags("${$/}", SVs_TEMP);
16814 if (!(obase->op_flags & OPf_KIDS))
16816 o = cUNOPx(obase)->op_first;
16822 /* This loop checks all the kid ops, skipping any that cannot pos-
16823 * sibly be responsible for the uninitialized value; i.e., defined
16824 * constants and ops that return nothing. If there is only one op
16825 * left that is not skipped, then we *know* it is responsible for
16826 * the uninitialized value. If there is more than one op left, we
16827 * have to look for an exact match in the while() loop below.
16828 * Note that we skip padrange, because the individual pad ops that
16829 * it replaced are still in the tree, so we work on them instead.
16832 for (kid=o; kid; kid = OpSIBLING(kid)) {
16833 const OPCODE type = kid->op_type;
16834 if ( (type == OP_CONST && SvOK(cSVOPx_sv(kid)))
16835 || (type == OP_NULL && ! (kid->op_flags & OPf_KIDS))
16836 || (type == OP_PUSHMARK)
16837 || (type == OP_PADRANGE)
16841 if (o2) { /* more than one found */
16848 return find_uninit_var(o2, uninit_sv, match, desc_p);
16850 /* scan all args */
16852 sv = find_uninit_var(o, uninit_sv, 1, desc_p);
16864 =for apidoc report_uninit
16866 Print appropriate "Use of uninitialized variable" warning.
16872 Perl_report_uninit(pTHX_ const SV *uninit_sv)
16874 const char *desc = NULL;
16875 SV* varname = NULL;
16878 desc = PL_op->op_type == OP_STRINGIFY && PL_op->op_folded
16880 : PL_op->op_type == OP_MULTICONCAT
16881 && (PL_op->op_private & OPpMULTICONCAT_FAKE)
16884 if (uninit_sv && PL_curpad) {
16885 varname = find_uninit_var(PL_op, uninit_sv, 0, &desc);
16887 sv_insert(varname, 0, 0, " ", 1);
16890 else if (PL_curstackinfo->si_type == PERLSI_SORT && cxstack_ix == 0)
16891 /* we've reached the end of a sort block or sub,
16892 * and the uninit value is probably what that code returned */
16895 /* PL_warn_uninit_sv is constant */
16896 GCC_DIAG_IGNORE_STMT(-Wformat-nonliteral);
16898 /* diag_listed_as: Use of uninitialized value%s */
16899 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit_sv,
16900 SVfARG(varname ? varname : &PL_sv_no),
16903 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit,
16905 GCC_DIAG_RESTORE_STMT;
16909 * ex: set ts=8 sts=4 sw=4 et: