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 /* int2str_table: lookup table containing string representations of all
2850 * two digit numbers. For example, int2str_table.arr[0] is "00" and
2851 * int2str_table.arr[12*2] is "12".
2853 * We are going to read two bytes at a time, so we have to ensure that
2854 * the array is aligned to a 2 byte boundary. That's why it was made a
2855 * union with a dummy U16 member. */
2856 static const union {
2859 } int2str_table = {{
2860 '0', '0', '0', '1', '0', '2', '0', '3', '0', '4', '0', '5', '0', '6',
2861 '0', '7', '0', '8', '0', '9', '1', '0', '1', '1', '1', '2', '1', '3',
2862 '1', '4', '1', '5', '1', '6', '1', '7', '1', '8', '1', '9', '2', '0',
2863 '2', '1', '2', '2', '2', '3', '2', '4', '2', '5', '2', '6', '2', '7',
2864 '2', '8', '2', '9', '3', '0', '3', '1', '3', '2', '3', '3', '3', '4',
2865 '3', '5', '3', '6', '3', '7', '3', '8', '3', '9', '4', '0', '4', '1',
2866 '4', '2', '4', '3', '4', '4', '4', '5', '4', '6', '4', '7', '4', '8',
2867 '4', '9', '5', '0', '5', '1', '5', '2', '5', '3', '5', '4', '5', '5',
2868 '5', '6', '5', '7', '5', '8', '5', '9', '6', '0', '6', '1', '6', '2',
2869 '6', '3', '6', '4', '6', '5', '6', '6', '6', '7', '6', '8', '6', '9',
2870 '7', '0', '7', '1', '7', '2', '7', '3', '7', '4', '7', '5', '7', '6',
2871 '7', '7', '7', '8', '7', '9', '8', '0', '8', '1', '8', '2', '8', '3',
2872 '8', '4', '8', '5', '8', '6', '8', '7', '8', '8', '8', '9', '9', '0',
2873 '9', '1', '9', '2', '9', '3', '9', '4', '9', '5', '9', '6', '9', '7',
2877 /* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2878 * UV as a string towards the end of buf, and return pointers to start and
2881 * We assume that buf is at least TYPE_CHARS(UV) long.
2884 PERL_STATIC_INLINE char *
2885 S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
2887 char *ptr = buf + TYPE_CHARS(UV);
2888 char * const ebuf = ptr;
2890 U16 *word_ptr, *word_table;
2892 PERL_ARGS_ASSERT_UIV_2BUF;
2894 /* ptr has to be properly aligned, because we will cast it to U16* */
2895 assert(PTR2nat(ptr) % 2 == 0);
2896 /* we are going to read/write two bytes at a time */
2897 word_ptr = (U16*)ptr;
2898 word_table = (U16*)int2str_table.arr;
2900 if (UNLIKELY(is_uv))
2911 *--word_ptr = word_table[uv % 100];
2914 ptr = (char*)word_ptr;
2917 *--ptr = (char)uv + '0';
2919 *--word_ptr = word_table[uv];
2920 ptr = (char*)word_ptr;
2930 /* Helper for sv_2pv_flags and sv_vcatpvfn_flags. If the NV is an
2931 * infinity or a not-a-number, writes the appropriate strings to the
2932 * buffer, including a zero byte. On success returns the written length,
2933 * excluding the zero byte, on failure (not an infinity, not a nan)
2934 * returns zero, assert-fails on maxlen being too short.
2936 * XXX for "Inf", "-Inf", and "NaN", we could have three read-only
2937 * shared string constants we point to, instead of generating a new
2938 * string for each instance. */
2940 S_infnan_2pv(NV nv, char* buffer, size_t maxlen, char plus) {
2942 assert(maxlen >= 4);
2943 if (Perl_isinf(nv)) {
2945 if (maxlen < 5) /* "-Inf\0" */
2955 else if (Perl_isnan(nv)) {
2959 /* XXX optionally output the payload mantissa bits as
2960 * "(unsigned)" (to match the nan("...") C99 function,
2961 * or maybe as "(0xhhh...)" would make more sense...
2962 * provide a format string so that the user can decide?
2963 * NOTE: would affect the maxlen and assert() logic.*/
2968 assert((s == buffer + 3) || (s == buffer + 4));
2974 =for apidoc sv_2pv_flags
2976 Returns a pointer to the string value of an SV, and sets C<*lp> to its length.
2977 If flags has the C<SV_GMAGIC> bit set, does an C<mg_get()> first. Coerces C<sv> to a
2978 string if necessary. Normally invoked via the C<SvPV_flags> macro.
2979 C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
2985 Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
2989 PERL_ARGS_ASSERT_SV_2PV_FLAGS;
2991 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2992 && SvTYPE(sv) != SVt_PVFM);
2993 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2998 if (flags & SV_SKIP_OVERLOAD)
3000 tmpstr = AMG_CALLunary(sv, string_amg);
3001 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
3002 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
3004 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
3008 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
3009 if (flags & SV_CONST_RETURN) {
3010 pv = (char *) SvPVX_const(tmpstr);
3012 pv = (flags & SV_MUTABLE_RETURN)
3013 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
3016 *lp = SvCUR(tmpstr);
3018 pv = sv_2pv_flags(tmpstr, lp, flags);
3031 SV *const referent = SvRV(sv);
3035 retval = buffer = savepvn("NULLREF", len);
3036 } else if (SvTYPE(referent) == SVt_REGEXP &&
3037 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
3038 amagic_is_enabled(string_amg))) {
3039 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
3043 /* If the regex is UTF-8 we want the containing scalar to
3044 have an UTF-8 flag too */
3051 *lp = RX_WRAPLEN(re);
3053 return RX_WRAPPED(re);
3055 const char *const typestr = sv_reftype(referent, 0);
3056 const STRLEN typelen = strlen(typestr);
3057 UV addr = PTR2UV(referent);
3058 const char *stashname = NULL;
3059 STRLEN stashnamelen = 0; /* hush, gcc */
3060 const char *buffer_end;
3062 if (SvOBJECT(referent)) {
3063 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
3066 stashname = HEK_KEY(name);
3067 stashnamelen = HEK_LEN(name);
3069 if (HEK_UTF8(name)) {
3075 stashname = "__ANON__";
3078 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
3079 + 2 * sizeof(UV) + 2 /* )\0 */;
3081 len = typelen + 3 /* (0x */
3082 + 2 * sizeof(UV) + 2 /* )\0 */;
3085 Newx(buffer, len, char);
3086 buffer_end = retval = buffer + len;
3088 /* Working backwards */
3092 *--retval = PL_hexdigit[addr & 15];
3093 } while (addr >>= 4);
3099 memcpy(retval, typestr, typelen);
3103 retval -= stashnamelen;
3104 memcpy(retval, stashname, stashnamelen);
3106 /* retval may not necessarily have reached the start of the
3108 assert (retval >= buffer);
3110 len = buffer_end - retval - 1; /* -1 for that \0 */
3122 if (flags & SV_MUTABLE_RETURN)
3123 return SvPVX_mutable(sv);
3124 if (flags & SV_CONST_RETURN)
3125 return (char *)SvPVX_const(sv);
3130 /* I'm assuming that if both IV and NV are equally valid then
3131 converting the IV is going to be more efficient */
3132 const U32 isUIOK = SvIsUV(sv);
3133 /* The purpose of this union is to ensure that arr is aligned on
3134 a 2 byte boundary, because that is what uiv_2buf() requires */
3136 char arr[TYPE_CHARS(UV)];
3142 if (SvTYPE(sv) < SVt_PVIV)
3143 sv_upgrade(sv, SVt_PVIV);
3144 ptr = uiv_2buf(buf.arr, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
3146 /* inlined from sv_setpvn */
3147 s = SvGROW_mutable(sv, len + 1);
3148 Move(ptr, s, len, char);
3153 else if (SvNOK(sv)) {
3154 if (SvTYPE(sv) < SVt_PVNV)
3155 sv_upgrade(sv, SVt_PVNV);
3156 if (SvNVX(sv) == 0.0
3157 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
3158 && !Perl_isnan(SvNVX(sv))
3161 s = SvGROW_mutable(sv, 2);
3166 STRLEN size = 5; /* "-Inf\0" */
3168 s = SvGROW_mutable(sv, size);
3169 len = S_infnan_2pv(SvNVX(sv), s, size, 0);
3175 /* some Xenix systems wipe out errno here */
3184 5 + /* exponent digits */
3188 s = SvGROW_mutable(sv, size);
3189 #ifndef USE_LOCALE_NUMERIC
3190 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3196 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
3197 STORE_LC_NUMERIC_SET_TO_NEEDED();
3199 local_radix = _NOT_IN_NUMERIC_STANDARD;
3200 if (local_radix && SvCUR(PL_numeric_radix_sv) > 1) {
3201 size += SvCUR(PL_numeric_radix_sv) - 1;
3202 s = SvGROW_mutable(sv, size);
3205 SNPRINTF_G(SvNVX(sv), s, SvLEN(sv), NV_DIG);
3207 /* If the radix character is UTF-8, and actually is in the
3208 * output, turn on the UTF-8 flag for the scalar */
3210 && SvUTF8(PL_numeric_radix_sv)
3211 && instr(s, SvPVX_const(PL_numeric_radix_sv)))
3216 RESTORE_LC_NUMERIC();
3219 /* We don't call SvPOK_on(), because it may come to
3220 * pass that the locale changes so that the
3221 * stringification we just did is no longer correct. We
3222 * will have to re-stringify every time it is needed */
3229 else if (isGV_with_GP(sv)) {
3230 GV *const gv = MUTABLE_GV(sv);
3231 SV *const buffer = sv_newmortal();
3233 gv_efullname3(buffer, gv, "*");
3235 assert(SvPOK(buffer));
3241 *lp = SvCUR(buffer);
3242 return SvPVX(buffer);
3247 if (flags & SV_UNDEF_RETURNS_NULL)
3249 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
3251 /* Typically the caller expects that sv_any is not NULL now. */
3252 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
3253 sv_upgrade(sv, SVt_PV);
3258 const STRLEN len = s - SvPVX_const(sv);
3263 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2pv(%s)\n",
3264 PTR2UV(sv),SvPVX_const(sv)));
3265 if (flags & SV_CONST_RETURN)
3266 return (char *)SvPVX_const(sv);
3267 if (flags & SV_MUTABLE_RETURN)
3268 return SvPVX_mutable(sv);
3273 =for apidoc sv_copypv
3275 Copies a stringified representation of the source SV into the
3276 destination SV. Automatically performs any necessary C<mg_get> and
3277 coercion of numeric values into strings. Guaranteed to preserve
3278 C<UTF8> flag even from overloaded objects. Similar in nature to
3279 C<sv_2pv[_flags]> but operates directly on an SV instead of just the
3280 string. Mostly uses C<sv_2pv_flags> to do its work, except when that
3281 would lose the UTF-8'ness of the PV.
3283 =for apidoc sv_copypv_nomg
3285 Like C<sv_copypv>, but doesn't invoke get magic first.
3287 =for apidoc sv_copypv_flags
3289 Implementation of C<sv_copypv> and C<sv_copypv_nomg>. Calls get magic iff flags
3290 has the C<SV_GMAGIC> bit set.
3296 Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
3301 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
3303 s = SvPV_flags_const(ssv,len,(flags & SV_GMAGIC));
3304 sv_setpvn(dsv,s,len);
3312 =for apidoc sv_2pvbyte
3314 Return a pointer to the byte-encoded representation of the SV, and set C<*lp>
3315 to its length. May cause the SV to be downgraded from UTF-8 as a
3318 Usually accessed via the C<SvPVbyte> macro.
3324 Perl_sv_2pvbyte(pTHX_ SV *sv, STRLEN *const lp)
3326 PERL_ARGS_ASSERT_SV_2PVBYTE;
3329 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3330 || isGV_with_GP(sv) || SvROK(sv)) {
3331 SV *sv2 = sv_newmortal();
3332 sv_copypv_nomg(sv2,sv);
3335 sv_utf8_downgrade(sv,0);
3336 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3340 =for apidoc sv_2pvutf8
3342 Return a pointer to the UTF-8-encoded representation of the SV, and set C<*lp>
3343 to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3345 Usually accessed via the C<SvPVutf8> macro.
3351 Perl_sv_2pvutf8(pTHX_ SV *sv, STRLEN *const lp)
3353 PERL_ARGS_ASSERT_SV_2PVUTF8;
3355 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3356 || isGV_with_GP(sv) || SvROK(sv))
3357 sv = sv_mortalcopy(sv);
3360 sv_utf8_upgrade_nomg(sv);
3361 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3366 =for apidoc sv_2bool
3368 This macro is only used by C<sv_true()> or its macro equivalent, and only if
3369 the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>.
3370 It calls C<sv_2bool_flags> with the C<SV_GMAGIC> flag.
3372 =for apidoc sv_2bool_flags
3374 This function is only used by C<sv_true()> and friends, and only if
3375 the latter's argument is neither C<SvPOK>, C<SvIOK> nor C<SvNOK>. If the flags
3376 contain C<SV_GMAGIC>, then it does an C<mg_get()> first.
3383 Perl_sv_2bool_flags(pTHX_ SV *sv, I32 flags)
3385 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
3388 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
3394 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
3395 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv)))) {
3398 if(SvGMAGICAL(sv)) {
3400 goto restart; /* call sv_2bool */
3402 /* expanded SvTRUE_common(sv, (flags = 0, goto restart)) */
3403 else if(!SvOK(sv)) {
3406 else if(SvPOK(sv)) {
3407 svb = SvPVXtrue(sv);
3409 else if((SvFLAGS(sv) & (SVf_IOK|SVf_NOK))) {
3410 svb = (SvIOK(sv) && SvIVX(sv) != 0)
3411 || (SvNOK(sv) && SvNVX(sv) != 0.0);
3415 goto restart; /* call sv_2bool_nomg */
3425 RX_WRAPLEN(sv) > 1 || (RX_WRAPLEN(sv) && *RX_WRAPPED(sv) != '0');
3427 if (SvNOK(sv) && !SvPOK(sv))
3428 return SvNVX(sv) != 0.0;
3430 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
3434 =for apidoc sv_utf8_upgrade
3436 Converts the PV of an SV to its UTF-8-encoded form.
3437 Forces the SV to string form if it is not already.
3438 Will C<mg_get> on C<sv> if appropriate.
3439 Always sets the C<SvUTF8> flag to avoid future validity checks even
3440 if the whole string is the same in UTF-8 as not.
3441 Returns the number of bytes in the converted string
3443 This is not a general purpose byte encoding to Unicode interface:
3444 use the Encode extension for that.
3446 =for apidoc sv_utf8_upgrade_nomg
3448 Like C<sv_utf8_upgrade>, but doesn't do magic on C<sv>.
3450 =for apidoc sv_utf8_upgrade_flags
3452 Converts the PV of an SV to its UTF-8-encoded form.
3453 Forces the SV to string form if it is not already.
3454 Always sets the SvUTF8 flag to avoid future validity checks even
3455 if all the bytes are invariant in UTF-8.
3456 If C<flags> has C<SV_GMAGIC> bit set,
3457 will C<mg_get> on C<sv> if appropriate, else not.
3459 The C<SV_FORCE_UTF8_UPGRADE> flag is now ignored.
3461 Returns the number of bytes in the converted string.
3463 This is not a general purpose byte encoding to Unicode interface:
3464 use the Encode extension for that.
3466 =for apidoc sv_utf8_upgrade_flags_grow
3468 Like C<sv_utf8_upgrade_flags>, but has an additional parameter C<extra>, which is
3469 the number of unused bytes the string of C<sv> is guaranteed to have free after
3470 it upon return. This allows the caller to reserve extra space that it intends
3471 to fill, to avoid extra grows.
3473 C<sv_utf8_upgrade>, C<sv_utf8_upgrade_nomg>, and C<sv_utf8_upgrade_flags>
3474 are implemented in terms of this function.
3476 Returns the number of bytes in the converted string (not including the spares).
3480 If the routine itself changes the string, it adds a trailing C<NUL>. Such a
3481 C<NUL> isn't guaranteed due to having other routines do the work in some input
3482 cases, or if the input is already flagged as being in utf8.
3487 Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV *const sv, const I32 flags, STRLEN extra)
3489 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
3491 if (sv == &PL_sv_undef)
3493 if (!SvPOK_nog(sv)) {
3495 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3496 (void) sv_2pv_flags(sv,&len, flags);
3498 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3502 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
3506 /* SVt_REGEXP's shouldn't be upgraded to UTF8 - they're already
3507 * compiled and individual nodes will remain non-utf8 even if the
3508 * stringified version of the pattern gets upgraded. Whether the
3509 * PVX of a REGEXP should be grown or we should just croak, I don't
3511 if (SvUTF8(sv) || isREGEXP(sv)) {
3512 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3517 S_sv_uncow(aTHX_ sv, 0);
3520 if (SvCUR(sv) == 0) {
3521 if (extra) SvGROW(sv, extra + 1); /* Make sure is room for a trailing
3523 } else { /* Assume Latin-1/EBCDIC */
3524 /* This function could be much more efficient if we
3525 * had a FLAG in SVs to signal if there are any variant
3526 * chars in the PV. Given that there isn't such a flag
3527 * make the loop as fast as possible. */
3528 U8 * s = (U8 *) SvPVX_const(sv);
3531 if (is_utf8_invariant_string_loc(s, SvCUR(sv), (const U8 **) &t)) {
3533 /* utf8 conversion not needed because all are invariants. Mark
3534 * as UTF-8 even if no variant - saves scanning loop */
3536 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3540 /* Here, there is at least one variant (t points to the first one), so
3541 * the string should be converted to utf8. Everything from 's' to
3542 * 't - 1' will occupy only 1 byte each on output.
3544 * Note that the incoming SV may not have a trailing '\0', as certain
3545 * code in pp_formline can send us partially built SVs.
3547 * There are two main ways to convert. One is to create a new string
3548 * and go through the input starting from the beginning, appending each
3549 * converted value onto the new string as we go along. Going this
3550 * route, it's probably best to initially allocate enough space in the
3551 * string rather than possibly running out of space and having to
3552 * reallocate and then copy what we've done so far. Since everything
3553 * from 's' to 't - 1' is invariant, the destination can be initialized
3554 * with these using a fast memory copy. To be sure to allocate enough
3555 * space, one could use the worst case scenario, where every remaining
3556 * byte expands to two under UTF-8, or one could parse it and count
3557 * exactly how many do expand.
3559 * The other way is to unconditionally parse the remainder of the
3560 * string to figure out exactly how big the expanded string will be,
3561 * growing if needed. Then start at the end of the string and place
3562 * the character there at the end of the unfilled space in the expanded
3563 * one, working backwards until reaching 't'.
3565 * The problem with assuming the worst case scenario is that for very
3566 * long strings, we could allocate much more memory than actually
3567 * needed, which can create performance problems. If we have to parse
3568 * anyway, the second method is the winner as it may avoid an extra
3569 * copy. The code used to use the first method under some
3570 * circumstances, but now that there is faster variant counting on
3571 * ASCII platforms, the second method is used exclusively, eliminating
3572 * some code that no longer has to be maintained. */
3575 /* Count the total number of variants there are. We can start
3576 * just beyond the first one, which is known to be at 't' */
3577 const Size_t invariant_length = t - s;
3578 U8 * e = (U8 *) SvEND(sv);
3580 /* The length of the left overs, plus 1. */
3581 const Size_t remaining_length_p1 = e - t;
3583 /* We expand by 1 for the variant at 't' and one for each remaining
3584 * variant (we start looking at 't+1') */
3585 Size_t expansion = 1 + variant_under_utf8_count(t + 1, e);
3587 /* +1 = trailing NUL */
3588 Size_t need = SvCUR(sv) + expansion + extra + 1;
3591 /* Grow if needed */
3592 if (SvLEN(sv) < need) {
3593 t = invariant_length + (U8*) SvGROW(sv, need);
3594 e = t + remaining_length_p1;
3596 SvCUR_set(sv, invariant_length + remaining_length_p1 + expansion);
3598 /* Set the NUL at the end */
3599 d = (U8 *) SvEND(sv);
3602 /* Having decremented d, it points to the position to put the
3603 * very last byte of the expanded string. Go backwards through
3604 * the string, copying and expanding as we go, stopping when we
3605 * get to the part that is invariant the rest of the way down */
3609 if (NATIVE_BYTE_IS_INVARIANT(*e)) {
3612 *d-- = UTF8_EIGHT_BIT_LO(*e);
3613 *d-- = UTF8_EIGHT_BIT_HI(*e);
3618 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3619 /* Update pos. We do it at the end rather than during
3620 * the upgrade, to avoid slowing down the common case
3621 * (upgrade without pos).
3622 * pos can be stored as either bytes or characters. Since
3623 * this was previously a byte string we can just turn off
3624 * the bytes flag. */
3625 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3627 mg->mg_flags &= ~MGf_BYTES;
3629 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3630 magic_setutf8(sv,mg); /* clear UTF8 cache */
3640 =for apidoc sv_utf8_downgrade
3642 Attempts to convert the PV of an SV from characters to bytes.
3643 If the PV contains a character that cannot fit
3644 in a byte, this conversion will fail;
3645 in this case, either returns false or, if C<fail_ok> is not
3648 This is not a general purpose Unicode to byte encoding interface:
3649 use the C<Encode> extension for that.
3655 Perl_sv_utf8_downgrade(pTHX_ SV *const sv, const bool fail_ok)
3657 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3659 if (SvPOKp(sv) && SvUTF8(sv)) {
3663 int mg_flags = SV_GMAGIC;
3666 S_sv_uncow(aTHX_ sv, 0);
3668 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3670 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3671 if (mg && mg->mg_len > 0 && mg->mg_flags & MGf_BYTES) {
3672 mg->mg_len = sv_pos_b2u_flags(sv, mg->mg_len,
3673 SV_GMAGIC|SV_CONST_RETURN);
3674 mg_flags = 0; /* sv_pos_b2u does get magic */
3676 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3677 magic_setutf8(sv,mg); /* clear UTF8 cache */
3680 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3682 if (!utf8_to_bytes(s, &len)) {
3687 Perl_croak(aTHX_ "Wide character in %s",
3690 Perl_croak(aTHX_ "Wide character");
3701 =for apidoc sv_utf8_encode
3703 Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3704 flag off so that it looks like octets again.
3710 Perl_sv_utf8_encode(pTHX_ SV *const sv)
3712 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3714 if (SvREADONLY(sv)) {
3715 sv_force_normal_flags(sv, 0);
3717 (void) sv_utf8_upgrade(sv);
3722 =for apidoc sv_utf8_decode
3724 If the PV of the SV is an octet sequence in Perl's extended UTF-8
3725 and contains a multiple-byte character, the C<SvUTF8> flag is turned on
3726 so that it looks like a character. If the PV contains only single-byte
3727 characters, the C<SvUTF8> flag stays off.
3728 Scans PV for validity and returns FALSE if the PV is invalid UTF-8.
3734 Perl_sv_utf8_decode(pTHX_ SV *const sv)
3736 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3739 const U8 *start, *c, *first_variant;
3741 /* The octets may have got themselves encoded - get them back as
3744 if (!sv_utf8_downgrade(sv, TRUE))
3747 /* it is actually just a matter of turning the utf8 flag on, but
3748 * we want to make sure everything inside is valid utf8 first.
3750 c = start = (const U8 *) SvPVX_const(sv);
3751 if (! is_utf8_invariant_string_loc(c, SvCUR(sv), &first_variant)) {
3752 if (!is_utf8_string(first_variant, SvCUR(sv) - (first_variant -c)))
3756 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3757 /* XXX Is this dead code? XS_utf8_decode calls SvSETMAGIC
3758 after this, clearing pos. Does anything on CPAN
3760 /* adjust pos to the start of a UTF8 char sequence */
3761 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3763 I32 pos = mg->mg_len;
3765 for (c = start + pos; c > start; c--) {
3766 if (UTF8_IS_START(*c))
3769 mg->mg_len = c - start;
3772 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3773 magic_setutf8(sv,mg); /* clear UTF8 cache */
3780 =for apidoc sv_setsv
3782 Copies the contents of the source SV C<ssv> into the destination SV
3783 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3784 function if the source SV needs to be reused. Does not handle 'set' magic on
3785 destination SV. Calls 'get' magic on source SV. Loosely speaking, it
3786 performs a copy-by-value, obliterating any previous content of the
3789 You probably want to use one of the assortment of wrappers, such as
3790 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3791 C<SvSetMagicSV_nosteal>.
3793 =for apidoc sv_setsv_flags
3795 Copies the contents of the source SV C<ssv> into the destination SV
3796 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3797 function if the source SV needs to be reused. Does not handle 'set' magic.
3798 Loosely speaking, it performs a copy-by-value, obliterating any previous
3799 content of the destination.
3800 If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
3801 C<ssv> if appropriate, else not. If the C<flags>
3802 parameter has the C<SV_NOSTEAL> bit set then the
3803 buffers of temps will not be stolen. C<sv_setsv>
3804 and C<sv_setsv_nomg> are implemented in terms of this function.
3806 You probably want to use one of the assortment of wrappers, such as
3807 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3808 C<SvSetMagicSV_nosteal>.
3810 This is the primary function for copying scalars, and most other
3811 copy-ish functions and macros use this underneath.
3817 S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
3819 I32 mro_changes = 0; /* 1 = method, 2 = isa, 3 = recursive isa */
3820 HV *old_stash = NULL;
3822 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB;
3824 if (dtype != SVt_PVGV && !isGV_with_GP(dstr)) {
3825 const char * const name = GvNAME(sstr);
3826 const STRLEN len = GvNAMELEN(sstr);
3828 if (dtype >= SVt_PV) {
3834 SvUPGRADE(dstr, SVt_PVGV);
3835 (void)SvOK_off(dstr);
3836 isGV_with_GP_on(dstr);
3838 GvSTASH(dstr) = GvSTASH(sstr);
3840 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
3841 gv_name_set(MUTABLE_GV(dstr), name, len,
3842 GV_ADD | (GvNAMEUTF8(sstr) ? SVf_UTF8 : 0 ));
3843 SvFAKE_on(dstr); /* can coerce to non-glob */
3846 if(GvGP(MUTABLE_GV(sstr))) {
3847 /* If source has method cache entry, clear it */
3849 SvREFCNT_dec(GvCV(sstr));
3850 GvCV_set(sstr, NULL);
3853 /* If source has a real method, then a method is
3856 GvCV((const GV *)sstr) && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3862 /* If dest already had a real method, that's a change as well */
3864 !mro_changes && GvGP(MUTABLE_GV(dstr)) && GvCVu((const GV *)dstr)
3865 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3870 /* We don't need to check the name of the destination if it was not a
3871 glob to begin with. */
3872 if(dtype == SVt_PVGV) {
3873 const char * const name = GvNAME((const GV *)dstr);
3874 const STRLEN len = GvNAMELEN(dstr);
3875 if(memEQs(name, len, "ISA")
3876 /* The stash may have been detached from the symbol table, so
3878 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3882 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
3883 || (len == 1 && name[0] == ':')) {
3886 /* Set aside the old stash, so we can reset isa caches on
3888 if((old_stash = GvHV(dstr)))
3889 /* Make sure we do not lose it early. */
3890 SvREFCNT_inc_simple_void_NN(
3891 sv_2mortal((SV *)old_stash)
3896 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
3899 /* freeing dstr's GP might free sstr (e.g. *x = $x),
3900 * so temporarily protect it */
3902 SAVEFREESV(SvREFCNT_inc_simple_NN(sstr));
3903 gp_free(MUTABLE_GV(dstr));
3904 GvINTRO_off(dstr); /* one-shot flag */
3905 GvGP_set(dstr, gp_ref(GvGP(sstr)));
3908 if (SvTAINTED(sstr))
3910 if (GvIMPORTED(dstr) != GVf_IMPORTED
3911 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
3913 GvIMPORTED_on(dstr);
3916 if(mro_changes == 2) {
3917 if (GvAV((const GV *)sstr)) {
3919 SV * const sref = (SV *)GvAV((const GV *)dstr);
3920 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
3921 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
3922 AV * const ary = newAV();
3923 av_push(ary, mg->mg_obj); /* takes the refcount */
3924 mg->mg_obj = (SV *)ary;
3926 av_push((AV *)mg->mg_obj, SvREFCNT_inc_simple_NN(dstr));
3928 else sv_magic(sref, dstr, PERL_MAGIC_isa, NULL, 0);
3930 mro_isa_changed_in(GvSTASH(dstr));
3932 else if(mro_changes == 3) {
3933 HV * const stash = GvHV(dstr);
3934 if(old_stash ? (HV *)HvENAME_get(old_stash) : stash)
3940 else if(mro_changes) mro_method_changed_in(GvSTASH(dstr));
3941 if (GvIO(dstr) && dtype == SVt_PVGV) {
3942 DEBUG_o(Perl_deb(aTHX_
3943 "glob_assign_glob clearing PL_stashcache\n"));
3944 /* It's a cache. It will rebuild itself quite happily.
3945 It's a lot of effort to work out exactly which key (or keys)
3946 might be invalidated by the creation of the this file handle.
3948 hv_clear(PL_stashcache);
3954 Perl_gv_setref(pTHX_ SV *const dstr, SV *const sstr)
3956 SV * const sref = SvRV(sstr);
3958 const int intro = GvINTRO(dstr);
3961 const U32 stype = SvTYPE(sref);
3963 PERL_ARGS_ASSERT_GV_SETREF;
3966 GvINTRO_off(dstr); /* one-shot flag */
3967 GvLINE(dstr) = CopLINE(PL_curcop);
3968 GvEGV(dstr) = MUTABLE_GV(dstr);
3973 location = (SV **) &(GvGP(dstr)->gp_cv); /* XXX bypassing GvCV_set */
3974 import_flag = GVf_IMPORTED_CV;
3977 location = (SV **) &GvHV(dstr);
3978 import_flag = GVf_IMPORTED_HV;
3981 location = (SV **) &GvAV(dstr);
3982 import_flag = GVf_IMPORTED_AV;
3985 location = (SV **) &GvIOp(dstr);
3988 location = (SV **) &GvFORM(dstr);
3991 location = &GvSV(dstr);
3992 import_flag = GVf_IMPORTED_SV;
3995 if (stype == SVt_PVCV) {
3996 /*if (GvCVGEN(dstr) && (GvCV(dstr) != (const CV *)sref || GvCVGEN(dstr))) {*/
3997 if (GvCVGEN(dstr)) {
3998 SvREFCNT_dec(GvCV(dstr));
3999 GvCV_set(dstr, NULL);
4000 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4003 /* SAVEt_GVSLOT takes more room on the savestack and has more
4004 overhead in leave_scope than SAVEt_GENERIC_SV. But for CVs
4005 leave_scope needs access to the GV so it can reset method
4006 caches. We must use SAVEt_GVSLOT whenever the type is
4007 SVt_PVCV, even if the stash is anonymous, as the stash may
4008 gain a name somehow before leave_scope. */
4009 if (stype == SVt_PVCV) {
4010 /* There is no save_pushptrptrptr. Creating it for this
4011 one call site would be overkill. So inline the ss add
4015 SS_ADD_PTR(location);
4016 SS_ADD_PTR(SvREFCNT_inc(*location));
4017 SS_ADD_UV(SAVEt_GVSLOT);
4020 else SAVEGENERICSV(*location);
4023 if (stype == SVt_PVCV && (*location != sref || GvCVGEN(dstr))) {
4024 CV* const cv = MUTABLE_CV(*location);
4026 if (!GvCVGEN((const GV *)dstr) &&
4027 (CvROOT(cv) || CvXSUB(cv)) &&
4028 /* redundant check that avoids creating the extra SV
4029 most of the time: */
4030 (CvCONST(cv) || ckWARN(WARN_REDEFINE)))
4032 SV * const new_const_sv =
4033 CvCONST((const CV *)sref)
4034 ? cv_const_sv((const CV *)sref)
4036 HV * const stash = GvSTASH((const GV *)dstr);
4037 report_redefined_cv(
4040 ? Perl_newSVpvf(aTHX_
4041 "%" HEKf "::%" HEKf,
4042 HEKfARG(HvNAME_HEK(stash)),
4043 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr))))
4044 : Perl_newSVpvf(aTHX_
4046 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr))))
4049 CvCONST((const CV *)sref) ? &new_const_sv : NULL
4053 cv_ckproto_len_flags(cv, (const GV *)dstr,
4054 SvPOK(sref) ? CvPROTO(sref) : NULL,
4055 SvPOK(sref) ? CvPROTOLEN(sref) : 0,
4056 SvPOK(sref) ? SvUTF8(sref) : 0);
4058 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
4059 GvASSUMECV_on(dstr);
4060 if(GvSTASH(dstr)) { /* sub foo { 1 } sub bar { 2 } *bar = \&foo */
4061 if (intro && GvREFCNT(dstr) > 1) {
4062 /* temporary remove extra savestack's ref */
4064 gv_method_changed(dstr);
4067 else gv_method_changed(dstr);
4070 *location = SvREFCNT_inc_simple_NN(sref);
4071 if (import_flag && !(GvFLAGS(dstr) & import_flag)
4072 && CopSTASH_ne(PL_curcop, GvSTASH(dstr))) {
4073 GvFLAGS(dstr) |= import_flag;
4076 if (stype == SVt_PVHV) {
4077 const char * const name = GvNAME((GV*)dstr);
4078 const STRLEN len = GvNAMELEN(dstr);
4081 (len > 1 && name[len-2] == ':' && name[len-1] == ':')
4082 || (len == 1 && name[0] == ':')
4084 && (!dref || HvENAME_get(dref))
4087 (HV *)sref, (HV *)dref,
4093 stype == SVt_PVAV && sref != dref
4094 && memEQs(GvNAME((GV*)dstr), GvNAMELEN((GV*)dstr), "ISA")
4095 /* The stash may have been detached from the symbol table, so
4096 check its name before doing anything. */
4097 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
4100 MAGIC * const omg = dref && SvSMAGICAL(dref)
4101 ? mg_find(dref, PERL_MAGIC_isa)
4103 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
4104 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
4105 AV * const ary = newAV();
4106 av_push(ary, mg->mg_obj); /* takes the refcount */
4107 mg->mg_obj = (SV *)ary;
4110 if (SvTYPE(omg->mg_obj) == SVt_PVAV) {
4111 SV **svp = AvARRAY((AV *)omg->mg_obj);
4112 I32 items = AvFILLp((AV *)omg->mg_obj) + 1;
4116 SvREFCNT_inc_simple_NN(*svp++)
4122 SvREFCNT_inc_simple_NN(omg->mg_obj)
4126 av_push((AV *)mg->mg_obj,SvREFCNT_inc_simple_NN(dstr));
4132 sref, omg ? omg->mg_obj : dstr, PERL_MAGIC_isa, NULL, 0
4134 for (i = 0; i <= AvFILL(sref); ++i) {
4135 SV **elem = av_fetch ((AV*)sref, i, 0);
4138 *elem, sref, PERL_MAGIC_isaelem, NULL, i
4142 mg = mg_find(sref, PERL_MAGIC_isa);
4144 /* Since the *ISA assignment could have affected more than
4145 one stash, don't call mro_isa_changed_in directly, but let
4146 magic_clearisa do it for us, as it already has the logic for
4147 dealing with globs vs arrays of globs. */
4149 Perl_magic_clearisa(aTHX_ NULL, mg);
4151 else if (stype == SVt_PVIO) {
4152 DEBUG_o(Perl_deb(aTHX_ "gv_setref clearing PL_stashcache\n"));
4153 /* It's a cache. It will rebuild itself quite happily.
4154 It's a lot of effort to work out exactly which key (or keys)
4155 might be invalidated by the creation of the this file handle.
4157 hv_clear(PL_stashcache);
4161 if (!intro) SvREFCNT_dec(dref);
4162 if (SvTAINTED(sstr))
4170 #ifdef PERL_DEBUG_READONLY_COW
4171 # include <sys/mman.h>
4173 # ifndef PERL_MEMORY_DEBUG_HEADER_SIZE
4174 # define PERL_MEMORY_DEBUG_HEADER_SIZE 0
4178 Perl_sv_buf_to_ro(pTHX_ SV *sv)
4180 struct perl_memory_debug_header * const header =
4181 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4182 const MEM_SIZE len = header->size;
4183 PERL_ARGS_ASSERT_SV_BUF_TO_RO;
4184 # ifdef PERL_TRACK_MEMPOOL
4185 if (!header->readonly) header->readonly = 1;
4187 if (mprotect(header, len, PROT_READ))
4188 Perl_warn(aTHX_ "mprotect RW for COW string %p %lu failed with %d",
4189 header, len, errno);
4193 S_sv_buf_to_rw(pTHX_ SV *sv)
4195 struct perl_memory_debug_header * const header =
4196 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4197 const MEM_SIZE len = header->size;
4198 PERL_ARGS_ASSERT_SV_BUF_TO_RW;
4199 if (mprotect(header, len, PROT_READ|PROT_WRITE))
4200 Perl_warn(aTHX_ "mprotect for COW string %p %lu failed with %d",
4201 header, len, errno);
4202 # ifdef PERL_TRACK_MEMPOOL
4203 header->readonly = 0;
4208 # define sv_buf_to_ro(sv) NOOP
4209 # define sv_buf_to_rw(sv) NOOP
4213 Perl_sv_setsv_flags(pTHX_ SV *dstr, SV* sstr, const I32 flags)
4218 unsigned int both_type;
4220 PERL_ARGS_ASSERT_SV_SETSV_FLAGS;
4222 if (UNLIKELY( sstr == dstr ))
4225 if (UNLIKELY( !sstr ))
4226 sstr = &PL_sv_undef;
4228 stype = SvTYPE(sstr);
4229 dtype = SvTYPE(dstr);
4230 both_type = (stype | dtype);
4232 /* with these values, we can check that both SVs are NULL/IV (and not
4233 * freed) just by testing the or'ed types */
4234 STATIC_ASSERT_STMT(SVt_NULL == 0);
4235 STATIC_ASSERT_STMT(SVt_IV == 1);
4236 if (both_type <= 1) {
4237 /* both src and dst are UNDEF/IV/RV, so we can do a lot of
4243 /* minimal subset of SV_CHECK_THINKFIRST_COW_DROP(dstr) */
4244 if (SvREADONLY(dstr))
4245 Perl_croak_no_modify();
4247 if (SvWEAKREF(dstr))
4248 sv_unref_flags(dstr, 0);
4250 old_rv = SvRV(dstr);
4253 assert(!SvGMAGICAL(sstr));
4254 assert(!SvGMAGICAL(dstr));
4256 sflags = SvFLAGS(sstr);
4257 if (sflags & (SVf_IOK|SVf_ROK)) {
4258 SET_SVANY_FOR_BODYLESS_IV(dstr);
4259 new_dflags = SVt_IV;
4261 if (sflags & SVf_ROK) {
4262 dstr->sv_u.svu_rv = SvREFCNT_inc(SvRV(sstr));
4263 new_dflags |= SVf_ROK;
4266 /* both src and dst are <= SVt_IV, so sv_any points to the
4267 * head; so access the head directly
4269 assert( &(sstr->sv_u.svu_iv)
4270 == &(((XPVIV*) SvANY(sstr))->xiv_iv));
4271 assert( &(dstr->sv_u.svu_iv)
4272 == &(((XPVIV*) SvANY(dstr))->xiv_iv));
4273 dstr->sv_u.svu_iv = sstr->sv_u.svu_iv;
4274 new_dflags |= (SVf_IOK|SVp_IOK|(sflags & SVf_IVisUV));
4278 new_dflags = dtype; /* turn off everything except the type */
4280 SvFLAGS(dstr) = new_dflags;
4281 SvREFCNT_dec(old_rv);
4286 if (UNLIKELY(both_type == SVTYPEMASK)) {
4287 if (SvIS_FREED(dstr)) {
4288 Perl_croak(aTHX_ "panic: attempt to copy value %" SVf
4289 " to a freed scalar %p", SVfARG(sstr), (void *)dstr);
4291 if (SvIS_FREED(sstr)) {
4292 Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p to %p",
4293 (void*)sstr, (void*)dstr);
4299 SV_CHECK_THINKFIRST_COW_DROP(dstr);
4300 dtype = SvTYPE(dstr); /* THINKFIRST may have changed type */
4302 /* There's a lot of redundancy below but we're going for speed here */
4307 if (LIKELY( dtype != SVt_PVGV && dtype != SVt_PVLV )) {
4308 (void)SvOK_off(dstr);
4316 /* For performance, we inline promoting to type SVt_IV. */
4317 /* We're starting from SVt_NULL, so provided that define is
4318 * actual 0, we don't have to unset any SV type flags
4319 * to promote to SVt_IV. */
4320 STATIC_ASSERT_STMT(SVt_NULL == 0);
4321 SET_SVANY_FOR_BODYLESS_IV(dstr);
4322 SvFLAGS(dstr) |= SVt_IV;
4326 sv_upgrade(dstr, SVt_PVIV);
4330 goto end_of_first_switch;
4332 (void)SvIOK_only(dstr);
4333 SvIV_set(dstr, SvIVX(sstr));
4336 /* SvTAINTED can only be true if the SV has taint magic, which in
4337 turn means that the SV type is PVMG (or greater). This is the
4338 case statement for SVt_IV, so this cannot be true (whatever gcov
4340 assert(!SvTAINTED(sstr));
4345 if (dtype < SVt_PV && dtype != SVt_IV)
4346 sv_upgrade(dstr, SVt_IV);
4350 if (LIKELY( SvNOK(sstr) )) {
4354 sv_upgrade(dstr, SVt_NV);
4358 sv_upgrade(dstr, SVt_PVNV);
4362 goto end_of_first_switch;
4364 SvNV_set(dstr, SvNVX(sstr));
4365 (void)SvNOK_only(dstr);
4366 /* SvTAINTED can only be true if the SV has taint magic, which in
4367 turn means that the SV type is PVMG (or greater). This is the
4368 case statement for SVt_NV, so this cannot be true (whatever gcov
4370 assert(!SvTAINTED(sstr));
4377 sv_upgrade(dstr, SVt_PV);
4380 if (dtype < SVt_PVIV)
4381 sv_upgrade(dstr, SVt_PVIV);
4384 if (dtype < SVt_PVNV)
4385 sv_upgrade(dstr, SVt_PVNV);
4389 invlist_clone(sstr, dstr);
4393 const char * const type = sv_reftype(sstr,0);
4395 /* diag_listed_as: Bizarre copy of %s */
4396 Perl_croak(aTHX_ "Bizarre copy of %s in %s", type, OP_DESC(PL_op));
4398 Perl_croak(aTHX_ "Bizarre copy of %s", type);
4400 NOT_REACHED; /* NOTREACHED */
4404 if (dtype < SVt_REGEXP)
4405 sv_upgrade(dstr, SVt_REGEXP);
4411 if (SvGMAGICAL(sstr) && (flags & SV_GMAGIC)) {
4413 if (SvTYPE(sstr) != stype)
4414 stype = SvTYPE(sstr);
4416 if (isGV_with_GP(sstr) && dtype <= SVt_PVLV) {
4417 glob_assign_glob(dstr, sstr, dtype);
4420 if (stype == SVt_PVLV)
4422 if (isREGEXP(sstr)) goto upgregexp;
4423 SvUPGRADE(dstr, SVt_PVNV);
4426 SvUPGRADE(dstr, (svtype)stype);
4428 end_of_first_switch:
4430 /* dstr may have been upgraded. */
4431 dtype = SvTYPE(dstr);
4432 sflags = SvFLAGS(sstr);
4434 if (UNLIKELY( dtype == SVt_PVCV )) {
4435 /* Assigning to a subroutine sets the prototype. */
4438 const char *const ptr = SvPV_const(sstr, len);
4440 SvGROW(dstr, len + 1);
4441 Copy(ptr, SvPVX(dstr), len + 1, char);
4442 SvCUR_set(dstr, len);
4444 SvFLAGS(dstr) |= sflags & SVf_UTF8;
4445 CvAUTOLOAD_off(dstr);
4450 else if (UNLIKELY(dtype == SVt_PVAV || dtype == SVt_PVHV
4451 || dtype == SVt_PVFM))
4453 const char * const type = sv_reftype(dstr,0);
4455 /* diag_listed_as: Cannot copy to %s */
4456 Perl_croak(aTHX_ "Cannot copy to %s in %s", type, OP_DESC(PL_op));
4458 Perl_croak(aTHX_ "Cannot copy to %s", type);
4459 } else if (sflags & SVf_ROK) {
4460 if (isGV_with_GP(dstr)
4461 && SvTYPE(SvRV(sstr)) == SVt_PVGV && isGV_with_GP(SvRV(sstr))) {
4464 if (GvIMPORTED(dstr) != GVf_IMPORTED
4465 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
4467 GvIMPORTED_on(dstr);
4472 glob_assign_glob(dstr, sstr, dtype);
4476 if (dtype >= SVt_PV) {
4477 if (isGV_with_GP(dstr)) {
4478 gv_setref(dstr, sstr);
4481 if (SvPVX_const(dstr)) {
4487 (void)SvOK_off(dstr);
4488 SvRV_set(dstr, SvREFCNT_inc(SvRV(sstr)));
4489 SvFLAGS(dstr) |= sflags & SVf_ROK;
4490 assert(!(sflags & SVp_NOK));
4491 assert(!(sflags & SVp_IOK));
4492 assert(!(sflags & SVf_NOK));
4493 assert(!(sflags & SVf_IOK));
4495 else if (isGV_with_GP(dstr)) {
4496 if (!(sflags & SVf_OK)) {
4497 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4498 "Undefined value assigned to typeglob");
4501 GV *gv = gv_fetchsv_nomg(sstr, GV_ADD, SVt_PVGV);
4502 if (dstr != (const SV *)gv) {
4503 const char * const name = GvNAME((const GV *)dstr);
4504 const STRLEN len = GvNAMELEN(dstr);
4505 HV *old_stash = NULL;
4506 bool reset_isa = FALSE;
4507 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
4508 || (len == 1 && name[0] == ':')) {
4509 /* Set aside the old stash, so we can reset isa caches
4510 on its subclasses. */
4511 if((old_stash = GvHV(dstr))) {
4512 /* Make sure we do not lose it early. */
4513 SvREFCNT_inc_simple_void_NN(
4514 sv_2mortal((SV *)old_stash)
4521 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
4522 gp_free(MUTABLE_GV(dstr));
4524 GvGP_set(dstr, gp_ref(GvGP(gv)));
4527 HV * const stash = GvHV(dstr);
4529 old_stash ? (HV *)HvENAME_get(old_stash) : stash
4539 else if ((dtype == SVt_REGEXP || dtype == SVt_PVLV)
4540 && (stype == SVt_REGEXP || isREGEXP(sstr))) {
4541 reg_temp_copy((REGEXP*)dstr, (REGEXP*)sstr);
4543 else if (sflags & SVp_POK) {
4544 const STRLEN cur = SvCUR(sstr);
4545 const STRLEN len = SvLEN(sstr);
4548 * We have three basic ways to copy the string:
4554 * Which we choose is based on various factors. The following
4555 * things are listed in order of speed, fastest to slowest:
4557 * - Copying a short string
4558 * - Copy-on-write bookkeeping
4560 * - Copying a long string
4562 * We swipe the string (steal the string buffer) if the SV on the
4563 * rhs is about to be freed anyway (TEMP and refcnt==1). This is a
4564 * big win on long strings. It should be a win on short strings if
4565 * SvPVX_const(dstr) has to be allocated. If not, it should not
4566 * slow things down, as SvPVX_const(sstr) would have been freed
4569 * We also steal the buffer from a PADTMP (operator target) if it
4570 * is ‘long enough’. For short strings, a swipe does not help
4571 * here, as it causes more malloc calls the next time the target
4572 * is used. Benchmarks show that even if SvPVX_const(dstr) has to
4573 * be allocated it is still not worth swiping PADTMPs for short
4574 * strings, as the savings here are small.
4576 * If swiping is not an option, then we see whether it is
4577 * worth using copy-on-write. If the lhs already has a buf-
4578 * fer big enough and the string is short, we skip it and fall back
4579 * to method 3, since memcpy is faster for short strings than the
4580 * later bookkeeping overhead that copy-on-write entails.
4582 * If the rhs is not a copy-on-write string yet, then we also
4583 * consider whether the buffer is too large relative to the string
4584 * it holds. Some operations such as readline allocate a large
4585 * buffer in the expectation of reusing it. But turning such into
4586 * a COW buffer is counter-productive because it increases memory
4587 * usage by making readline allocate a new large buffer the sec-
4588 * ond time round. So, if the buffer is too large, again, we use
4591 * Finally, if there is no buffer on the left, or the buffer is too
4592 * small, then we use copy-on-write and make both SVs share the
4597 /* Whichever path we take through the next code, we want this true,
4598 and doing it now facilitates the COW check. */
4599 (void)SvPOK_only(dstr);
4603 /* slated for free anyway (and not COW)? */
4604 (sflags & (SVs_TEMP|SVf_IsCOW)) == SVs_TEMP
4605 /* or a swipable TARG */
4607 (SVs_PADTMP|SVf_READONLY|SVf_PROTECT|SVf_IsCOW))
4609 /* whose buffer is worth stealing */
4610 && CHECK_COWBUF_THRESHOLD(cur,len)
4613 !(sflags & SVf_OOK) && /* and not involved in OOK hack? */
4614 (!(flags & SV_NOSTEAL)) &&
4615 /* and we're allowed to steal temps */
4616 SvREFCNT(sstr) == 1 && /* and no other references to it? */
4617 len) /* and really is a string */
4618 { /* Passes the swipe test. */
4619 if (SvPVX_const(dstr)) /* we know that dtype >= SVt_PV */
4621 SvPV_set(dstr, SvPVX_mutable(sstr));
4622 SvLEN_set(dstr, SvLEN(sstr));
4623 SvCUR_set(dstr, SvCUR(sstr));
4626 (void)SvOK_off(sstr); /* NOTE: nukes most SvFLAGS on sstr */
4627 SvPV_set(sstr, NULL);
4632 else if (flags & SV_COW_SHARED_HASH_KEYS
4634 #ifdef PERL_COPY_ON_WRITE
4637 ( (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4638 /* If this is a regular (non-hek) COW, only so
4639 many COW "copies" are possible. */
4640 && CowREFCNT(sstr) != SV_COW_REFCNT_MAX ))
4641 : ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4642 && !(SvFLAGS(dstr) & SVf_BREAK)
4643 && CHECK_COW_THRESHOLD(cur,len) && cur+1 < len
4644 && (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4648 && !(SvFLAGS(dstr) & SVf_BREAK)
4651 /* Either it's a shared hash key, or it's suitable for
4655 PerlIO_printf(Perl_debug_log, "Copy on write: sstr --> dstr\n");
4661 if (!(sflags & SVf_IsCOW)) {
4663 CowREFCNT(sstr) = 0;
4666 if (SvPVX_const(dstr)) { /* we know that dtype >= SVt_PV */
4672 if (sflags & SVf_IsCOW) {
4676 SvPV_set(dstr, SvPVX_mutable(sstr));
4681 /* SvIsCOW_shared_hash */
4682 DEBUG_C(PerlIO_printf(Perl_debug_log,
4683 "Copy on write: Sharing hash\n"));
4685 assert (SvTYPE(dstr) >= SVt_PV);
4687 HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)))));
4689 SvLEN_set(dstr, len);
4690 SvCUR_set(dstr, cur);
4693 /* Failed the swipe test, and we cannot do copy-on-write either.
4694 Have to copy the string. */
4695 SvGROW(dstr, cur + 1); /* inlined from sv_setpvn */
4696 Move(SvPVX_const(sstr),SvPVX(dstr),cur,char);
4697 SvCUR_set(dstr, cur);
4698 *SvEND(dstr) = '\0';
4700 if (sflags & SVp_NOK) {
4701 SvNV_set(dstr, SvNVX(sstr));
4703 if (sflags & SVp_IOK) {
4704 SvIV_set(dstr, SvIVX(sstr));
4705 if (sflags & SVf_IVisUV)
4708 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_NOK|SVp_NOK|SVf_UTF8);
4710 const MAGIC * const smg = SvVSTRING_mg(sstr);
4712 sv_magic(dstr, NULL, PERL_MAGIC_vstring,
4713 smg->mg_ptr, smg->mg_len);
4714 SvRMAGICAL_on(dstr);
4718 else if (sflags & (SVp_IOK|SVp_NOK)) {
4719 (void)SvOK_off(dstr);
4720 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_IVisUV|SVf_NOK|SVp_NOK);
4721 if (sflags & SVp_IOK) {
4722 /* XXXX Do we want to set IsUV for IV(ROK)? Be extra safe... */
4723 SvIV_set(dstr, SvIVX(sstr));
4725 if (sflags & SVp_NOK) {
4726 SvNV_set(dstr, SvNVX(sstr));
4730 if (isGV_with_GP(sstr)) {
4731 gv_efullname3(dstr, MUTABLE_GV(sstr), "*");
4734 (void)SvOK_off(dstr);
4736 if (SvTAINTED(sstr))
4742 =for apidoc sv_set_undef
4744 Equivalent to C<sv_setsv(sv, &PL_sv_undef)>, but more efficient.
4745 Doesn't handle set magic.
4747 The perl equivalent is C<$sv = undef;>. Note that it doesn't free any string
4748 buffer, unlike C<undef $sv>.
4750 Introduced in perl 5.25.12.
4756 Perl_sv_set_undef(pTHX_ SV *sv)
4758 U32 type = SvTYPE(sv);
4760 PERL_ARGS_ASSERT_SV_SET_UNDEF;
4762 /* shortcut, NULL, IV, RV */
4764 if (type <= SVt_IV) {
4765 assert(!SvGMAGICAL(sv));
4766 if (SvREADONLY(sv)) {
4767 /* does undeffing PL_sv_undef count as modifying a read-only
4768 * variable? Some XS code does this */
4769 if (sv == &PL_sv_undef)
4771 Perl_croak_no_modify();
4776 sv_unref_flags(sv, 0);
4779 SvFLAGS(sv) = type; /* quickly turn off all flags */
4780 SvREFCNT_dec_NN(rv);
4784 SvFLAGS(sv) = type; /* quickly turn off all flags */
4789 Perl_croak(aTHX_ "panic: attempt to undefine a freed scalar %p",
4792 SV_CHECK_THINKFIRST_COW_DROP(sv);
4794 if (isGV_with_GP(sv))
4795 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4796 "Undefined value assigned to typeglob");
4804 =for apidoc sv_setsv_mg
4806 Like C<sv_setsv>, but also handles 'set' magic.
4812 Perl_sv_setsv_mg(pTHX_ SV *const dstr, SV *const sstr)
4814 PERL_ARGS_ASSERT_SV_SETSV_MG;
4816 sv_setsv(dstr,sstr);
4821 # define SVt_COW SVt_PV
4823 Perl_sv_setsv_cow(pTHX_ SV *dstr, SV *sstr)
4825 STRLEN cur = SvCUR(sstr);
4826 STRLEN len = SvLEN(sstr);
4828 #if defined(PERL_DEBUG_READONLY_COW) && defined(PERL_COPY_ON_WRITE)
4829 const bool already = cBOOL(SvIsCOW(sstr));
4832 PERL_ARGS_ASSERT_SV_SETSV_COW;
4835 PerlIO_printf(Perl_debug_log, "Fast copy on write: %p -> %p\n",
4836 (void*)sstr, (void*)dstr);
4843 if (SvTHINKFIRST(dstr))
4844 sv_force_normal_flags(dstr, SV_COW_DROP_PV);
4845 else if (SvPVX_const(dstr))
4846 Safefree(SvPVX_mutable(dstr));
4850 SvUPGRADE(dstr, SVt_COW);
4852 assert (SvPOK(sstr));
4853 assert (SvPOKp(sstr));
4855 if (SvIsCOW(sstr)) {
4857 if (SvLEN(sstr) == 0) {
4858 /* source is a COW shared hash key. */
4859 DEBUG_C(PerlIO_printf(Perl_debug_log,
4860 "Fast copy on write: Sharing hash\n"));
4861 new_pv = HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr))));
4864 assert(SvCUR(sstr)+1 < SvLEN(sstr));
4865 assert(CowREFCNT(sstr) < SV_COW_REFCNT_MAX);
4867 assert ((SvFLAGS(sstr) & CAN_COW_MASK) == CAN_COW_FLAGS);
4868 SvUPGRADE(sstr, SVt_COW);
4870 DEBUG_C(PerlIO_printf(Perl_debug_log,
4871 "Fast copy on write: Converting sstr to COW\n"));
4872 CowREFCNT(sstr) = 0;
4874 # ifdef PERL_DEBUG_READONLY_COW
4875 if (already) sv_buf_to_rw(sstr);
4878 new_pv = SvPVX_mutable(sstr);
4882 SvPV_set(dstr, new_pv);
4883 SvFLAGS(dstr) = (SVt_COW|SVf_POK|SVp_POK|SVf_IsCOW);
4886 SvLEN_set(dstr, len);
4887 SvCUR_set(dstr, cur);
4897 =for apidoc sv_setpv_bufsize
4899 Sets the SV to be a string of cur bytes length, with at least
4900 len bytes available. Ensures that there is a null byte at SvEND.
4901 Returns a char * pointer to the SvPV buffer.
4907 Perl_sv_setpv_bufsize(pTHX_ SV *const sv, const STRLEN cur, const STRLEN len)
4911 PERL_ARGS_ASSERT_SV_SETPV_BUFSIZE;
4913 SV_CHECK_THINKFIRST_COW_DROP(sv);
4914 SvUPGRADE(sv, SVt_PV);
4915 pv = SvGROW(sv, len + 1);
4918 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4921 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4926 =for apidoc sv_setpvn
4928 Copies a string (possibly containing embedded C<NUL> characters) into an SV.
4929 The C<len> parameter indicates the number of
4930 bytes to be copied. If the C<ptr> argument is NULL the SV will become
4931 undefined. Does not handle 'set' magic. See C<L</sv_setpvn_mg>>.
4937 Perl_sv_setpvn(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4941 PERL_ARGS_ASSERT_SV_SETPVN;
4943 SV_CHECK_THINKFIRST_COW_DROP(sv);
4944 if (isGV_with_GP(sv))
4945 Perl_croak_no_modify();
4951 /* len is STRLEN which is unsigned, need to copy to signed */
4954 Perl_croak(aTHX_ "panic: sv_setpvn called with negative strlen %"
4957 SvUPGRADE(sv, SVt_PV);
4959 dptr = SvGROW(sv, len + 1);
4960 Move(ptr,dptr,len,char);
4963 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4965 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4969 =for apidoc sv_setpvn_mg
4971 Like C<sv_setpvn>, but also handles 'set' magic.
4977 Perl_sv_setpvn_mg(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4979 PERL_ARGS_ASSERT_SV_SETPVN_MG;
4981 sv_setpvn(sv,ptr,len);
4986 =for apidoc sv_setpv
4988 Copies a string into an SV. The string must be terminated with a C<NUL>
4989 character, and not contain embeded C<NUL>'s.
4990 Does not handle 'set' magic. See C<L</sv_setpv_mg>>.
4996 Perl_sv_setpv(pTHX_ SV *const sv, const char *const ptr)
5000 PERL_ARGS_ASSERT_SV_SETPV;
5002 SV_CHECK_THINKFIRST_COW_DROP(sv);
5008 SvUPGRADE(sv, SVt_PV);
5010 SvGROW(sv, len + 1);
5011 Move(ptr,SvPVX(sv),len+1,char);
5013 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5015 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
5019 =for apidoc sv_setpv_mg
5021 Like C<sv_setpv>, but also handles 'set' magic.
5027 Perl_sv_setpv_mg(pTHX_ SV *const sv, const char *const ptr)
5029 PERL_ARGS_ASSERT_SV_SETPV_MG;
5036 Perl_sv_sethek(pTHX_ SV *const sv, const HEK *const hek)
5038 PERL_ARGS_ASSERT_SV_SETHEK;
5044 if (HEK_LEN(hek) == HEf_SVKEY) {
5045 sv_setsv(sv, *(SV**)HEK_KEY(hek));
5048 const int flags = HEK_FLAGS(hek);
5049 if (flags & HVhek_WASUTF8) {
5050 STRLEN utf8_len = HEK_LEN(hek);
5051 char *as_utf8 = (char *)bytes_to_utf8((U8*)HEK_KEY(hek), &utf8_len);
5052 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
5055 } else if (flags & HVhek_UNSHARED) {
5056 sv_setpvn(sv, HEK_KEY(hek), HEK_LEN(hek));
5059 else SvUTF8_off(sv);
5063 SV_CHECK_THINKFIRST_COW_DROP(sv);
5064 SvUPGRADE(sv, SVt_PV);
5066 SvPV_set(sv,(char *)HEK_KEY(share_hek_hek(hek)));
5067 SvCUR_set(sv, HEK_LEN(hek));
5073 else SvUTF8_off(sv);
5081 =for apidoc sv_usepvn_flags
5083 Tells an SV to use C<ptr> to find its string value. Normally the
5084 string is stored inside the SV, but sv_usepvn allows the SV to use an
5085 outside string. C<ptr> should point to memory that was allocated
5086 by L<C<Newx>|perlclib/Memory Management and String Handling>. It must be
5087 the start of a C<Newx>-ed block of memory, and not a pointer to the
5088 middle of it (beware of L<C<OOK>|perlguts/Offsets> and copy-on-write),
5089 and not be from a non-C<Newx> memory allocator like C<malloc>. The
5090 string length, C<len>, must be supplied. By default this function
5091 will C<Renew> (i.e. realloc, move) the memory pointed to by C<ptr>,
5092 so that pointer should not be freed or used by the programmer after
5093 giving it to C<sv_usepvn>, and neither should any pointers from "behind"
5094 that pointer (e.g. ptr + 1) be used.
5096 If S<C<flags & SV_SMAGIC>> is true, will call C<SvSETMAGIC>. If
5097 S<C<flags & SV_HAS_TRAILING_NUL>> is true, then C<ptr[len]> must be C<NUL>,
5099 will be skipped (i.e. the buffer is actually at least 1 byte longer than
5100 C<len>, and already meets the requirements for storing in C<SvPVX>).
5106 Perl_sv_usepvn_flags(pTHX_ SV *const sv, char *ptr, const STRLEN len, const U32 flags)
5110 PERL_ARGS_ASSERT_SV_USEPVN_FLAGS;
5112 SV_CHECK_THINKFIRST_COW_DROP(sv);
5113 SvUPGRADE(sv, SVt_PV);
5116 if (flags & SV_SMAGIC)
5120 if (SvPVX_const(sv))
5124 if (flags & SV_HAS_TRAILING_NUL)
5125 assert(ptr[len] == '\0');
5128 allocate = (flags & SV_HAS_TRAILING_NUL)
5130 #ifdef Perl_safesysmalloc_size
5133 PERL_STRLEN_ROUNDUP(len + 1);
5135 if (flags & SV_HAS_TRAILING_NUL) {
5136 /* It's long enough - do nothing.
5137 Specifically Perl_newCONSTSUB is relying on this. */
5140 /* Force a move to shake out bugs in callers. */
5141 char *new_ptr = (char*)safemalloc(allocate);
5142 Copy(ptr, new_ptr, len, char);
5143 PoisonFree(ptr,len,char);
5147 ptr = (char*) saferealloc (ptr, allocate);
5150 #ifdef Perl_safesysmalloc_size
5151 SvLEN_set(sv, Perl_safesysmalloc_size(ptr));
5153 SvLEN_set(sv, allocate);
5157 if (!(flags & SV_HAS_TRAILING_NUL)) {
5160 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5162 if (flags & SV_SMAGIC)
5168 S_sv_uncow(pTHX_ SV * const sv, const U32 flags)
5170 assert(SvIsCOW(sv));
5173 const char * const pvx = SvPVX_const(sv);
5174 const STRLEN len = SvLEN(sv);
5175 const STRLEN cur = SvCUR(sv);
5179 PerlIO_printf(Perl_debug_log,
5180 "Copy on write: Force normal %ld\n",
5186 # ifdef PERL_COPY_ON_WRITE
5188 /* Must do this first, since the CowREFCNT uses SvPVX and
5189 we need to write to CowREFCNT, or de-RO the whole buffer if we are
5190 the only owner left of the buffer. */
5191 sv_buf_to_rw(sv); /* NOOP if RO-ing not supported */
5193 U8 cowrefcnt = CowREFCNT(sv);
5194 if(cowrefcnt != 0) {
5196 CowREFCNT(sv) = cowrefcnt;
5201 /* Else we are the only owner of the buffer. */
5206 /* This SV doesn't own the buffer, so need to Newx() a new one: */
5211 if (flags & SV_COW_DROP_PV) {
5212 /* OK, so we don't need to copy our buffer. */
5215 SvGROW(sv, cur + 1);
5216 Move(pvx,SvPVX(sv),cur,char);
5221 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5229 const char * const pvx = SvPVX_const(sv);
5230 const STRLEN len = SvCUR(sv);
5234 if (flags & SV_COW_DROP_PV) {
5235 /* OK, so we don't need to copy our buffer. */
5238 SvGROW(sv, len + 1);
5239 Move(pvx,SvPVX(sv),len,char);
5242 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5249 =for apidoc sv_force_normal_flags
5251 Undo various types of fakery on an SV, where fakery means
5252 "more than" a string: if the PV is a shared string, make
5253 a private copy; if we're a ref, stop refing; if we're a glob, downgrade to
5254 an C<xpvmg>; if we're a copy-on-write scalar, this is the on-write time when
5255 we do the copy, and is also used locally; if this is a
5256 vstring, drop the vstring magic. If C<SV_COW_DROP_PV> is set
5257 then a copy-on-write scalar drops its PV buffer (if any) and becomes
5258 C<SvPOK_off> rather than making a copy. (Used where this
5259 scalar is about to be set to some other value.) In addition,
5260 the C<flags> parameter gets passed to C<sv_unref_flags()>
5261 when unreffing. C<sv_force_normal> calls this function
5262 with flags set to 0.
5264 This function is expected to be used to signal to perl that this SV is
5265 about to be written to, and any extra book-keeping needs to be taken care
5266 of. Hence, it croaks on read-only values.
5272 Perl_sv_force_normal_flags(pTHX_ SV *const sv, const U32 flags)
5274 PERL_ARGS_ASSERT_SV_FORCE_NORMAL_FLAGS;
5277 Perl_croak_no_modify();
5278 else if (SvIsCOW(sv) && LIKELY(SvTYPE(sv) != SVt_PVHV))
5279 S_sv_uncow(aTHX_ sv, flags);
5281 sv_unref_flags(sv, flags);
5282 else if (SvFAKE(sv) && isGV_with_GP(sv))
5283 sv_unglob(sv, flags);
5284 else if (SvFAKE(sv) && isREGEXP(sv)) {
5285 /* Need to downgrade the REGEXP to a simple(r) scalar. This is analogous
5286 to sv_unglob. We only need it here, so inline it. */
5287 const bool islv = SvTYPE(sv) == SVt_PVLV;
5288 const svtype new_type =
5289 islv ? SVt_NULL : SvMAGIC(sv) || SvSTASH(sv) ? SVt_PVMG : SVt_PV;
5290 SV *const temp = newSV_type(new_type);
5291 regexp *old_rx_body;
5293 if (new_type == SVt_PVMG) {
5294 SvMAGIC_set(temp, SvMAGIC(sv));
5295 SvMAGIC_set(sv, NULL);
5296 SvSTASH_set(temp, SvSTASH(sv));
5297 SvSTASH_set(sv, NULL);
5300 SvCUR_set(temp, SvCUR(sv));
5301 /* Remember that SvPVX is in the head, not the body. */
5302 assert(ReANY((REGEXP *)sv)->mother_re);
5305 /* LV-as-regex has sv->sv_any pointing to an XPVLV body,
5306 * whose xpvlenu_rx field points to the regex body */
5307 XPV *xpv = (XPV*)(SvANY(sv));
5308 old_rx_body = xpv->xpv_len_u.xpvlenu_rx;
5309 xpv->xpv_len_u.xpvlenu_rx = NULL;
5312 old_rx_body = ReANY((REGEXP *)sv);
5314 /* Their buffer is already owned by someone else. */
5315 if (flags & SV_COW_DROP_PV) {
5316 /* SvLEN is already 0. For SVt_REGEXP, we have a brand new
5317 zeroed body. For SVt_PVLV, we zeroed it above (len field
5318 a union with xpvlenu_rx) */
5319 assert(!SvLEN(islv ? sv : temp));
5320 sv->sv_u.svu_pv = 0;
5323 sv->sv_u.svu_pv = savepvn(RX_WRAPPED((REGEXP *)sv), SvCUR(sv));
5324 SvLEN_set(islv ? sv : temp, SvCUR(sv)+1);
5328 /* Now swap the rest of the bodies. */
5332 SvFLAGS(sv) &= ~SVTYPEMASK;
5333 SvFLAGS(sv) |= new_type;
5334 SvANY(sv) = SvANY(temp);
5337 SvFLAGS(temp) &= ~(SVTYPEMASK);
5338 SvFLAGS(temp) |= SVt_REGEXP|SVf_FAKE;
5339 SvANY(temp) = old_rx_body;
5341 SvREFCNT_dec_NN(temp);
5343 else if (SvVOK(sv)) sv_unmagic(sv, PERL_MAGIC_vstring);
5349 Efficient removal of characters from the beginning of the string buffer.
5350 C<SvPOK(sv)>, or at least C<SvPOKp(sv)>, must be true and C<ptr> must be a
5351 pointer to somewhere inside the string buffer. C<ptr> becomes the first
5352 character of the adjusted string. Uses the C<OOK> hack. On return, only
5353 C<SvPOK(sv)> and C<SvPOKp(sv)> among the C<OK> flags will be true.
5355 Beware: after this function returns, C<ptr> and SvPVX_const(sv) may no longer
5356 refer to the same chunk of data.
5358 The unfortunate similarity of this function's name to that of Perl's C<chop>
5359 operator is strictly coincidental. This function works from the left;
5360 C<chop> works from the right.
5366 Perl_sv_chop(pTHX_ SV *const sv, const char *const ptr)
5377 PERL_ARGS_ASSERT_SV_CHOP;
5379 if (!ptr || !SvPOKp(sv))
5381 delta = ptr - SvPVX_const(sv);
5383 /* Nothing to do. */
5386 max_delta = SvLEN(sv) ? SvLEN(sv) : SvCUR(sv);
5387 if (delta > max_delta)
5388 Perl_croak(aTHX_ "panic: sv_chop ptr=%p, start=%p, end=%p",
5389 ptr, SvPVX_const(sv), SvPVX_const(sv) + max_delta);
5390 /* SvPVX(sv) may move in SV_CHECK_THINKFIRST(sv), so don't use ptr any more */
5391 SV_CHECK_THINKFIRST(sv);
5392 SvPOK_only_UTF8(sv);
5395 if (!SvLEN(sv)) { /* make copy of shared string */
5396 const char *pvx = SvPVX_const(sv);
5397 const STRLEN len = SvCUR(sv);
5398 SvGROW(sv, len + 1);
5399 Move(pvx,SvPVX(sv),len,char);
5405 SvOOK_offset(sv, old_delta);
5407 SvLEN_set(sv, SvLEN(sv) - delta);
5408 SvCUR_set(sv, SvCUR(sv) - delta);
5409 SvPV_set(sv, SvPVX(sv) + delta);
5411 p = (U8 *)SvPVX_const(sv);
5414 /* how many bytes were evacuated? we will fill them with sentinel
5415 bytes, except for the part holding the new offset of course. */
5418 evacn += (old_delta < 0x100 ? 1 : 1 + sizeof(STRLEN));
5420 assert(evacn <= delta + old_delta);
5424 /* This sets 'delta' to the accumulated value of all deltas so far */
5428 /* If 'delta' fits in a byte, store it just prior to the new beginning of
5429 * the string; otherwise store a 0 byte there and store 'delta' just prior
5430 * to that, using as many bytes as a STRLEN occupies. Thus it overwrites a
5431 * portion of the chopped part of the string */
5432 if (delta < 0x100) {
5436 p -= sizeof(STRLEN);
5437 Copy((U8*)&delta, p, sizeof(STRLEN), U8);
5441 /* Fill the preceding buffer with sentinals to verify that no-one is
5451 =for apidoc sv_catpvn
5453 Concatenates the string onto the end of the string which is in the SV.
5454 C<len> indicates number of bytes to copy. If the SV has the UTF-8
5455 status set, then the bytes appended should be valid UTF-8.
5456 Handles 'get' magic, but not 'set' magic. See C<L</sv_catpvn_mg>>.
5458 =for apidoc sv_catpvn_flags
5460 Concatenates the string onto the end of the string which is in the SV. The
5461 C<len> indicates number of bytes to copy.
5463 By default, the string appended is assumed to be valid UTF-8 if the SV has
5464 the UTF-8 status set, and a string of bytes otherwise. One can force the
5465 appended string to be interpreted as UTF-8 by supplying the C<SV_CATUTF8>
5466 flag, and as bytes by supplying the C<SV_CATBYTES> flag; the SV or the
5467 string appended will be upgraded to UTF-8 if necessary.
5469 If C<flags> has the C<SV_SMAGIC> bit set, will
5470 C<mg_set> on C<dsv> afterwards if appropriate.
5471 C<sv_catpvn> and C<sv_catpvn_nomg> are implemented
5472 in terms of this function.
5478 Perl_sv_catpvn_flags(pTHX_ SV *const dsv, const char *sstr, const STRLEN slen, const I32 flags)
5481 const char * const dstr = SvPV_force_flags(dsv, dlen, flags);
5483 PERL_ARGS_ASSERT_SV_CATPVN_FLAGS;
5484 assert((flags & (SV_CATBYTES|SV_CATUTF8)) != (SV_CATBYTES|SV_CATUTF8));
5486 if (!(flags & SV_CATBYTES) || !SvUTF8(dsv)) {
5487 if (flags & SV_CATUTF8 && !SvUTF8(dsv)) {
5488 sv_utf8_upgrade_flags_grow(dsv, 0, slen + 1);
5491 else SvGROW(dsv, dlen + slen + 3);
5493 sstr = SvPVX_const(dsv);
5494 Move(sstr, SvPVX(dsv) + dlen, slen, char);
5495 SvCUR_set(dsv, SvCUR(dsv) + slen);
5498 /* We inline bytes_to_utf8, to avoid an extra malloc. */
5499 const char * const send = sstr + slen;
5502 /* Something this code does not account for, which I think is
5503 impossible; it would require the same pv to be treated as
5504 bytes *and* utf8, which would indicate a bug elsewhere. */
5505 assert(sstr != dstr);
5507 SvGROW(dsv, dlen + slen * 2 + 3);
5508 d = (U8 *)SvPVX(dsv) + dlen;
5510 while (sstr < send) {
5511 append_utf8_from_native_byte(*sstr, &d);
5514 SvCUR_set(dsv, d-(const U8 *)SvPVX(dsv));
5517 (void)SvPOK_only_UTF8(dsv); /* validate pointer */
5519 if (flags & SV_SMAGIC)
5524 =for apidoc sv_catsv
5526 Concatenates the string from SV C<ssv> onto the end of the string in SV
5527 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5528 Handles 'get' magic on both SVs, but no 'set' magic. See C<L</sv_catsv_mg>>
5529 and C<L</sv_catsv_nomg>>.
5531 =for apidoc sv_catsv_flags
5533 Concatenates the string from SV C<ssv> onto the end of the string in SV
5534 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5535 If C<flags> has the C<SV_GMAGIC> bit set, will call C<mg_get> on both SVs if
5536 appropriate. If C<flags> has the C<SV_SMAGIC> bit set, C<mg_set> will be called on
5537 the modified SV afterward, if appropriate. C<sv_catsv>, C<sv_catsv_nomg>,
5538 and C<sv_catsv_mg> are implemented in terms of this function.
5543 Perl_sv_catsv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
5545 PERL_ARGS_ASSERT_SV_CATSV_FLAGS;
5549 const char *spv = SvPV_flags_const(ssv, slen, flags);
5550 if (flags & SV_GMAGIC)
5552 sv_catpvn_flags(dsv, spv, slen,
5553 DO_UTF8(ssv) ? SV_CATUTF8 : SV_CATBYTES);
5554 if (flags & SV_SMAGIC)
5560 =for apidoc sv_catpv
5562 Concatenates the C<NUL>-terminated string onto the end of the string which is
5564 If the SV has the UTF-8 status set, then the bytes appended should be
5565 valid UTF-8. Handles 'get' magic, but not 'set' magic. See
5571 Perl_sv_catpv(pTHX_ SV *const sv, const char *ptr)
5577 PERL_ARGS_ASSERT_SV_CATPV;
5581 junk = SvPV_force(sv, tlen);
5583 SvGROW(sv, tlen + len + 1);
5585 ptr = SvPVX_const(sv);
5586 Move(ptr,SvPVX(sv)+tlen,len+1,char);
5587 SvCUR_set(sv, SvCUR(sv) + len);
5588 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5593 =for apidoc sv_catpv_flags
5595 Concatenates the C<NUL>-terminated string onto the end of the string which is
5597 If the SV has the UTF-8 status set, then the bytes appended should
5598 be valid UTF-8. If C<flags> has the C<SV_SMAGIC> bit set, will C<mg_set>
5599 on the modified SV if appropriate.
5605 Perl_sv_catpv_flags(pTHX_ SV *dstr, const char *sstr, const I32 flags)
5607 PERL_ARGS_ASSERT_SV_CATPV_FLAGS;
5608 sv_catpvn_flags(dstr, sstr, strlen(sstr), flags);
5612 =for apidoc sv_catpv_mg
5614 Like C<sv_catpv>, but also handles 'set' magic.
5620 Perl_sv_catpv_mg(pTHX_ SV *const sv, const char *const ptr)
5622 PERL_ARGS_ASSERT_SV_CATPV_MG;
5631 Creates a new SV. A non-zero C<len> parameter indicates the number of
5632 bytes of preallocated string space the SV should have. An extra byte for a
5633 trailing C<NUL> is also reserved. (C<SvPOK> is not set for the SV even if string
5634 space is allocated.) The reference count for the new SV is set to 1.
5636 In 5.9.3, C<newSV()> replaces the older C<NEWSV()> API, and drops the first
5637 parameter, I<x>, a debug aid which allowed callers to identify themselves.
5638 This aid has been superseded by a new build option, C<PERL_MEM_LOG> (see
5639 L<perlhacktips/PERL_MEM_LOG>). The older API is still there for use in XS
5640 modules supporting older perls.
5646 Perl_newSV(pTHX_ const STRLEN len)
5652 sv_grow(sv, len + 1);
5657 =for apidoc sv_magicext
5659 Adds magic to an SV, upgrading it if necessary. Applies the
5660 supplied C<vtable> and returns a pointer to the magic added.
5662 Note that C<sv_magicext> will allow things that C<sv_magic> will not.
5663 In particular, you can add magic to C<SvREADONLY> SVs, and add more than
5664 one instance of the same C<how>.
5666 If C<namlen> is greater than zero then a C<savepvn> I<copy> of C<name> is
5667 stored, if C<namlen> is zero then C<name> is stored as-is and - as another
5668 special case - if C<(name && namlen == HEf_SVKEY)> then C<name> is assumed
5669 to contain an SV* and is stored as-is with its C<REFCNT> incremented.
5671 (This is now used as a subroutine by C<sv_magic>.)
5676 Perl_sv_magicext(pTHX_ SV *const sv, SV *const obj, const int how,
5677 const MGVTBL *const vtable, const char *const name, const I32 namlen)
5681 PERL_ARGS_ASSERT_SV_MAGICEXT;
5683 SvUPGRADE(sv, SVt_PVMG);
5684 Newxz(mg, 1, MAGIC);
5685 mg->mg_moremagic = SvMAGIC(sv);
5686 SvMAGIC_set(sv, mg);
5688 /* Sometimes a magic contains a reference loop, where the sv and
5689 object refer to each other. To prevent a reference loop that
5690 would prevent such objects being freed, we look for such loops
5691 and if we find one we avoid incrementing the object refcount.
5693 Note we cannot do this to avoid self-tie loops as intervening RV must
5694 have its REFCNT incremented to keep it in existence.
5697 if (!obj || obj == sv ||
5698 how == PERL_MAGIC_arylen ||
5699 how == PERL_MAGIC_regdata ||
5700 how == PERL_MAGIC_regdatum ||
5701 how == PERL_MAGIC_symtab ||
5702 (SvTYPE(obj) == SVt_PVGV &&
5703 (GvSV(obj) == sv || GvHV(obj) == (const HV *)sv
5704 || GvAV(obj) == (const AV *)sv || GvCV(obj) == (const CV *)sv
5705 || GvIOp(obj) == (const IO *)sv || GvFORM(obj) == (const CV *)sv)))
5710 mg->mg_obj = SvREFCNT_inc_simple(obj);
5711 mg->mg_flags |= MGf_REFCOUNTED;
5714 /* Normal self-ties simply pass a null object, and instead of
5715 using mg_obj directly, use the SvTIED_obj macro to produce a
5716 new RV as needed. For glob "self-ties", we are tieing the PVIO
5717 with an RV obj pointing to the glob containing the PVIO. In
5718 this case, to avoid a reference loop, we need to weaken the
5722 if (how == PERL_MAGIC_tiedscalar && SvTYPE(sv) == SVt_PVIO &&
5723 obj && SvROK(obj) && GvIO(SvRV(obj)) == (const IO *)sv)
5729 mg->mg_len = namlen;
5732 mg->mg_ptr = savepvn(name, namlen);
5733 else if (namlen == HEf_SVKEY) {
5734 /* Yes, this is casting away const. This is only for the case of
5735 HEf_SVKEY. I think we need to document this aberation of the
5736 constness of the API, rather than making name non-const, as
5737 that change propagating outwards a long way. */
5738 mg->mg_ptr = (char*)SvREFCNT_inc_simple_NN((SV *)name);
5740 mg->mg_ptr = (char *) name;
5742 mg->mg_virtual = (MGVTBL *) vtable;
5749 Perl_sv_magicext_mglob(pTHX_ SV *sv)
5751 PERL_ARGS_ASSERT_SV_MAGICEXT_MGLOB;
5752 if (SvTYPE(sv) == SVt_PVLV && LvTYPE(sv) == 'y') {
5753 /* This sv is only a delegate. //g magic must be attached to
5758 return sv_magicext(sv, NULL, PERL_MAGIC_regex_global,
5759 &PL_vtbl_mglob, 0, 0);
5763 =for apidoc sv_magic
5765 Adds magic to an SV. First upgrades C<sv> to type C<SVt_PVMG> if
5766 necessary, then adds a new magic item of type C<how> to the head of the
5769 See C<L</sv_magicext>> (which C<sv_magic> now calls) for a description of the
5770 handling of the C<name> and C<namlen> arguments.
5772 You need to use C<sv_magicext> to add magic to C<SvREADONLY> SVs and also
5773 to add more than one instance of the same C<how>.
5779 Perl_sv_magic(pTHX_ SV *const sv, SV *const obj, const int how,
5780 const char *const name, const I32 namlen)
5782 const MGVTBL *vtable;
5785 unsigned int vtable_index;
5787 PERL_ARGS_ASSERT_SV_MAGIC;
5789 if (how < 0 || (unsigned)how >= C_ARRAY_LENGTH(PL_magic_data)
5790 || ((flags = PL_magic_data[how]),
5791 (vtable_index = flags & PERL_MAGIC_VTABLE_MASK)
5792 > magic_vtable_max))
5793 Perl_croak(aTHX_ "Don't know how to handle magic of type \\%o", how);
5795 /* PERL_MAGIC_ext is reserved for use by extensions not perl internals.
5796 Useful for attaching extension internal data to perl vars.
5797 Note that multiple extensions may clash if magical scalars
5798 etc holding private data from one are passed to another. */
5800 vtable = (vtable_index == magic_vtable_max)
5801 ? NULL : PL_magic_vtables + vtable_index;
5803 if (SvREADONLY(sv)) {
5805 !PERL_MAGIC_TYPE_READONLY_ACCEPTABLE(how)
5808 Perl_croak_no_modify();
5811 if (SvMAGICAL(sv) || (how == PERL_MAGIC_taint && SvTYPE(sv) >= SVt_PVMG)) {
5812 if (SvMAGIC(sv) && (mg = mg_find(sv, how))) {
5813 /* sv_magic() refuses to add a magic of the same 'how' as an
5816 if (how == PERL_MAGIC_taint)
5822 /* Force pos to be stored as characters, not bytes. */
5823 if (SvMAGICAL(sv) && DO_UTF8(sv)
5824 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
5826 && mg->mg_flags & MGf_BYTES) {
5827 mg->mg_len = (SSize_t)sv_pos_b2u_flags(sv, (STRLEN)mg->mg_len,
5829 mg->mg_flags &= ~MGf_BYTES;
5832 /* Rest of work is done else where */
5833 mg = sv_magicext(sv,obj,how,vtable,name,namlen);
5836 case PERL_MAGIC_taint:
5839 case PERL_MAGIC_ext:
5840 case PERL_MAGIC_dbfile:
5847 S_sv_unmagicext_flags(pTHX_ SV *const sv, const int type, MGVTBL *vtbl, const U32 flags)
5854 if (SvTYPE(sv) < SVt_PVMG || !SvMAGIC(sv))
5856 mgp = &(((XPVMG*) SvANY(sv))->xmg_u.xmg_magic);
5857 for (mg = *mgp; mg; mg = *mgp) {
5858 const MGVTBL* const virt = mg->mg_virtual;
5859 if (mg->mg_type == type && (!flags || virt == vtbl)) {
5860 *mgp = mg->mg_moremagic;
5861 if (virt && virt->svt_free)
5862 virt->svt_free(aTHX_ sv, mg);
5863 if (mg->mg_ptr && mg->mg_type != PERL_MAGIC_regex_global) {
5865 Safefree(mg->mg_ptr);
5866 else if (mg->mg_len == HEf_SVKEY)
5867 SvREFCNT_dec(MUTABLE_SV(mg->mg_ptr));
5868 else if (mg->mg_type == PERL_MAGIC_utf8)
5869 Safefree(mg->mg_ptr);
5871 if (mg->mg_flags & MGf_REFCOUNTED)
5872 SvREFCNT_dec(mg->mg_obj);
5876 mgp = &mg->mg_moremagic;
5879 if (SvMAGICAL(sv)) /* if we're under save_magic, wait for restore_magic; */
5880 mg_magical(sv); /* else fix the flags now */
5889 =for apidoc sv_unmagic
5891 Removes all magic of type C<type> from an SV.
5897 Perl_sv_unmagic(pTHX_ SV *const sv, const int type)
5899 PERL_ARGS_ASSERT_SV_UNMAGIC;
5900 return S_sv_unmagicext_flags(aTHX_ sv, type, NULL, 0);
5904 =for apidoc sv_unmagicext
5906 Removes all magic of type C<type> with the specified C<vtbl> from an SV.
5912 Perl_sv_unmagicext(pTHX_ SV *const sv, const int type, MGVTBL *vtbl)
5914 PERL_ARGS_ASSERT_SV_UNMAGICEXT;
5915 return S_sv_unmagicext_flags(aTHX_ sv, type, vtbl, 1);
5919 =for apidoc sv_rvweaken
5921 Weaken a reference: set the C<SvWEAKREF> flag on this RV; give the
5922 referred-to SV C<PERL_MAGIC_backref> magic if it hasn't already; and
5923 push a back-reference to this RV onto the array of backreferences
5924 associated with that magic. If the RV is magical, set magic will be
5925 called after the RV is cleared. Silently ignores C<undef> and warns
5926 on already-weak references.
5932 Perl_sv_rvweaken(pTHX_ SV *const sv)
5936 PERL_ARGS_ASSERT_SV_RVWEAKEN;
5938 if (!SvOK(sv)) /* let undefs pass */
5941 Perl_croak(aTHX_ "Can't weaken a nonreference");
5942 else if (SvWEAKREF(sv)) {
5943 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is already weak");
5946 else if (SvREADONLY(sv)) croak_no_modify();
5948 Perl_sv_add_backref(aTHX_ tsv, sv);
5950 SvREFCNT_dec_NN(tsv);
5955 =for apidoc sv_rvunweaken
5957 Unweaken a reference: Clear the C<SvWEAKREF> flag on this RV; remove
5958 the backreference to this RV from the array of backreferences
5959 associated with the target SV, increment the refcount of the target.
5960 Silently ignores C<undef> and warns on non-weak references.
5966 Perl_sv_rvunweaken(pTHX_ SV *const sv)
5970 PERL_ARGS_ASSERT_SV_RVUNWEAKEN;
5972 if (!SvOK(sv)) /* let undefs pass */
5975 Perl_croak(aTHX_ "Can't unweaken a nonreference");
5976 else if (!SvWEAKREF(sv)) {
5977 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is not weak");
5980 else if (SvREADONLY(sv)) croak_no_modify();
5985 SvREFCNT_inc_NN(tsv);
5986 Perl_sv_del_backref(aTHX_ tsv, sv);
5991 =for apidoc sv_get_backrefs
5993 If C<sv> is the target of a weak reference then it returns the back
5994 references structure associated with the sv; otherwise return C<NULL>.
5996 When returning a non-null result the type of the return is relevant. If it
5997 is an AV then the elements of the AV are the weak reference RVs which
5998 point at this item. If it is any other type then the item itself is the
6001 See also C<Perl_sv_add_backref()>, C<Perl_sv_del_backref()>,
6002 C<Perl_sv_kill_backrefs()>
6008 Perl_sv_get_backrefs(SV *const sv)
6012 PERL_ARGS_ASSERT_SV_GET_BACKREFS;
6014 /* find slot to store array or singleton backref */
6016 if (SvTYPE(sv) == SVt_PVHV) {
6018 struct xpvhv_aux * const iter = HvAUX((HV *)sv);
6019 backrefs = (SV *)iter->xhv_backreferences;
6021 } else if (SvMAGICAL(sv)) {
6022 MAGIC *mg = mg_find(sv, PERL_MAGIC_backref);
6024 backrefs = mg->mg_obj;
6029 /* Give tsv backref magic if it hasn't already got it, then push a
6030 * back-reference to sv onto the array associated with the backref magic.
6032 * As an optimisation, if there's only one backref and it's not an AV,
6033 * store it directly in the HvAUX or mg_obj slot, avoiding the need to
6034 * allocate an AV. (Whether the slot holds an AV tells us whether this is
6038 /* A discussion about the backreferences array and its refcount:
6040 * The AV holding the backreferences is pointed to either as the mg_obj of
6041 * PERL_MAGIC_backref, or in the specific case of a HV, from the
6042 * xhv_backreferences field. The array is created with a refcount
6043 * of 2. This means that if during global destruction the array gets
6044 * picked on before its parent to have its refcount decremented by the
6045 * random zapper, it won't actually be freed, meaning it's still there for
6046 * when its parent gets freed.
6048 * When the parent SV is freed, the extra ref is killed by
6049 * Perl_sv_kill_backrefs. The other ref is killed, in the case of magic,
6050 * by mg_free() / MGf_REFCOUNTED, or for a hash, by Perl_hv_kill_backrefs.
6052 * When a single backref SV is stored directly, it is not reference
6057 Perl_sv_add_backref(pTHX_ SV *const tsv, SV *const sv)
6063 PERL_ARGS_ASSERT_SV_ADD_BACKREF;
6065 /* find slot to store array or singleton backref */
6067 if (SvTYPE(tsv) == SVt_PVHV) {
6068 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
6071 mg = mg_find(tsv, PERL_MAGIC_backref);
6073 mg = sv_magicext(tsv, NULL, PERL_MAGIC_backref, &PL_vtbl_backref, NULL, 0);
6074 svp = &(mg->mg_obj);
6077 /* create or retrieve the array */
6079 if ( (!*svp && SvTYPE(sv) == SVt_PVAV)
6080 || (*svp && SvTYPE(*svp) != SVt_PVAV)
6084 mg->mg_flags |= MGf_REFCOUNTED;
6087 SvREFCNT_inc_simple_void_NN(av);
6088 /* av now has a refcnt of 2; see discussion above */
6089 av_extend(av, *svp ? 2 : 1);
6091 /* move single existing backref to the array */
6092 AvARRAY(av)[++AvFILLp(av)] = *svp; /* av_push() */
6097 av = MUTABLE_AV(*svp);
6099 /* optimisation: store single backref directly in HvAUX or mg_obj */
6103 assert(SvTYPE(av) == SVt_PVAV);
6104 if (AvFILLp(av) >= AvMAX(av)) {
6105 av_extend(av, AvFILLp(av)+1);
6108 /* push new backref */
6109 AvARRAY(av)[++AvFILLp(av)] = sv; /* av_push() */
6112 /* delete a back-reference to ourselves from the backref magic associated
6113 * with the SV we point to.
6117 Perl_sv_del_backref(pTHX_ SV *const tsv, SV *const sv)
6121 PERL_ARGS_ASSERT_SV_DEL_BACKREF;
6123 if (SvTYPE(tsv) == SVt_PVHV) {
6125 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
6127 else if (SvIS_FREED(tsv) && PL_phase == PERL_PHASE_DESTRUCT) {
6128 /* It's possible for the the last (strong) reference to tsv to have
6129 become freed *before* the last thing holding a weak reference.
6130 If both survive longer than the backreferences array, then when
6131 the referent's reference count drops to 0 and it is freed, it's
6132 not able to chase the backreferences, so they aren't NULLed.
6134 For example, a CV holds a weak reference to its stash. If both the
6135 CV and the stash survive longer than the backreferences array,
6136 and the CV gets picked for the SvBREAK() treatment first,
6137 *and* it turns out that the stash is only being kept alive because
6138 of an our variable in the pad of the CV, then midway during CV
6139 destruction the stash gets freed, but CvSTASH() isn't set to NULL.
6140 It ends up pointing to the freed HV. Hence it's chased in here, and
6141 if this block wasn't here, it would hit the !svp panic just below.
6143 I don't believe that "better" destruction ordering is going to help
6144 here - during global destruction there's always going to be the
6145 chance that something goes out of order. We've tried to make it
6146 foolproof before, and it only resulted in evolutionary pressure on
6147 fools. Which made us look foolish for our hubris. :-(
6153 = SvMAGICAL(tsv) ? mg_find(tsv, PERL_MAGIC_backref) : NULL;
6154 svp = mg ? &(mg->mg_obj) : NULL;
6158 Perl_croak(aTHX_ "panic: del_backref, svp=0");
6160 /* It's possible that sv is being freed recursively part way through the
6161 freeing of tsv. If this happens, the backreferences array of tsv has
6162 already been freed, and so svp will be NULL. If this is the case,
6163 we should not panic. Instead, nothing needs doing, so return. */
6164 if (PL_phase == PERL_PHASE_DESTRUCT && SvREFCNT(tsv) == 0)
6166 Perl_croak(aTHX_ "panic: del_backref, *svp=%p phase=%s refcnt=%" UVuf,
6167 (void*)*svp, PL_phase_names[PL_phase], (UV)SvREFCNT(tsv));
6170 if (SvTYPE(*svp) == SVt_PVAV) {
6174 AV * const av = (AV*)*svp;
6176 assert(!SvIS_FREED(av));
6180 /* for an SV with N weak references to it, if all those
6181 * weak refs are deleted, then sv_del_backref will be called
6182 * N times and O(N^2) compares will be done within the backref
6183 * array. To ameliorate this potential slowness, we:
6184 * 1) make sure this code is as tight as possible;
6185 * 2) when looking for SV, look for it at both the head and tail of the
6186 * array first before searching the rest, since some create/destroy
6187 * patterns will cause the backrefs to be freed in order.
6194 SV **p = &svp[fill];
6195 SV *const topsv = *p;
6202 /* We weren't the last entry.
6203 An unordered list has this property that you
6204 can take the last element off the end to fill
6205 the hole, and it's still an unordered list :-)
6211 break; /* should only be one */
6218 AvFILLp(av) = fill-1;
6220 else if (SvIS_FREED(*svp) && PL_phase == PERL_PHASE_DESTRUCT) {
6221 /* freed AV; skip */
6224 /* optimisation: only a single backref, stored directly */
6226 Perl_croak(aTHX_ "panic: del_backref, *svp=%p, sv=%p",
6227 (void*)*svp, (void*)sv);
6234 Perl_sv_kill_backrefs(pTHX_ SV *const sv, AV *const av)
6240 PERL_ARGS_ASSERT_SV_KILL_BACKREFS;
6245 /* after multiple passes through Perl_sv_clean_all() for a thingy
6246 * that has badly leaked, the backref array may have gotten freed,
6247 * since we only protect it against 1 round of cleanup */
6248 if (SvIS_FREED(av)) {
6249 if (PL_in_clean_all) /* All is fair */
6252 "panic: magic_killbackrefs (freed backref AV/SV)");
6256 is_array = (SvTYPE(av) == SVt_PVAV);
6258 assert(!SvIS_FREED(av));
6261 last = svp + AvFILLp(av);
6264 /* optimisation: only a single backref, stored directly */
6270 while (svp <= last) {
6272 SV *const referrer = *svp;
6273 if (SvWEAKREF(referrer)) {
6274 /* XXX Should we check that it hasn't changed? */
6275 assert(SvROK(referrer));
6276 SvRV_set(referrer, 0);
6278 SvWEAKREF_off(referrer);
6279 SvSETMAGIC(referrer);
6280 } else if (SvTYPE(referrer) == SVt_PVGV ||
6281 SvTYPE(referrer) == SVt_PVLV) {
6282 assert(SvTYPE(sv) == SVt_PVHV); /* stash backref */
6283 /* You lookin' at me? */
6284 assert(GvSTASH(referrer));
6285 assert(GvSTASH(referrer) == (const HV *)sv);
6286 GvSTASH(referrer) = 0;
6287 } else if (SvTYPE(referrer) == SVt_PVCV ||
6288 SvTYPE(referrer) == SVt_PVFM) {
6289 if (SvTYPE(sv) == SVt_PVHV) { /* stash backref */
6290 /* You lookin' at me? */
6291 assert(CvSTASH(referrer));
6292 assert(CvSTASH(referrer) == (const HV *)sv);
6293 SvANY(MUTABLE_CV(referrer))->xcv_stash = 0;
6296 assert(SvTYPE(sv) == SVt_PVGV);
6297 /* You lookin' at me? */
6298 assert(CvGV(referrer));
6299 assert(CvGV(referrer) == (const GV *)sv);
6300 anonymise_cv_maybe(MUTABLE_GV(sv),
6301 MUTABLE_CV(referrer));
6306 "panic: magic_killbackrefs (flags=%" UVxf ")",
6307 (UV)SvFLAGS(referrer));
6318 SvREFCNT_dec_NN(av); /* remove extra count added by sv_add_backref() */
6324 =for apidoc sv_insert
6326 Inserts and/or replaces a string at the specified offset/length within the SV.
6327 Similar to the Perl C<substr()> function, with C<littlelen> bytes starting at
6328 C<little> replacing C<len> bytes of the string in C<bigstr> starting at
6329 C<offset>. Handles get magic.
6331 =for apidoc sv_insert_flags
6333 Same as C<sv_insert>, but the extra C<flags> are passed to the
6334 C<SvPV_force_flags> that applies to C<bigstr>.
6340 Perl_sv_insert_flags(pTHX_ SV *const bigstr, const STRLEN offset, const STRLEN len, const char *little, const STRLEN littlelen, const U32 flags)
6346 SSize_t i; /* better be sizeof(STRLEN) or bad things happen */
6349 PERL_ARGS_ASSERT_SV_INSERT_FLAGS;
6351 SvPV_force_flags(bigstr, curlen, flags);
6352 (void)SvPOK_only_UTF8(bigstr);
6354 if (little >= SvPVX(bigstr) &&
6355 little < SvPVX(bigstr) + (SvLEN(bigstr) ? SvLEN(bigstr) : SvCUR(bigstr))) {
6356 /* little is a pointer to within bigstr, since we can reallocate bigstr,
6357 or little...little+littlelen might overlap offset...offset+len we make a copy
6359 little = savepvn(little, littlelen);
6363 if (offset + len > curlen) {
6364 SvGROW(bigstr, offset+len+1);
6365 Zero(SvPVX(bigstr)+curlen, offset+len-curlen, char);
6366 SvCUR_set(bigstr, offset+len);
6370 i = littlelen - len;
6371 if (i > 0) { /* string might grow */
6372 big = SvGROW(bigstr, SvCUR(bigstr) + i + 1);
6373 mid = big + offset + len;
6374 midend = bigend = big + SvCUR(bigstr);
6377 while (midend > mid) /* shove everything down */
6378 *--bigend = *--midend;
6379 Move(little,big+offset,littlelen,char);
6380 SvCUR_set(bigstr, SvCUR(bigstr) + i);
6385 Move(little,SvPVX(bigstr)+offset,len,char);
6390 big = SvPVX(bigstr);
6393 bigend = big + SvCUR(bigstr);
6395 if (midend > bigend)
6396 Perl_croak(aTHX_ "panic: sv_insert, midend=%p, bigend=%p",
6399 if (mid - big > bigend - midend) { /* faster to shorten from end */
6401 Move(little, mid, littlelen,char);
6404 i = bigend - midend;
6406 Move(midend, mid, i,char);
6410 SvCUR_set(bigstr, mid - big);
6412 else if ((i = mid - big)) { /* faster from front */
6413 midend -= littlelen;
6415 Move(big, midend - i, i, char);
6416 sv_chop(bigstr,midend-i);
6418 Move(little, mid, littlelen,char);
6420 else if (littlelen) {
6421 midend -= littlelen;
6422 sv_chop(bigstr,midend);
6423 Move(little,midend,littlelen,char);
6426 sv_chop(bigstr,midend);
6432 =for apidoc sv_replace
6434 Make the first argument a copy of the second, then delete the original.
6435 The target SV physically takes over ownership of the body of the source SV
6436 and inherits its flags; however, the target keeps any magic it owns,
6437 and any magic in the source is discarded.
6438 Note that this is a rather specialist SV copying operation; most of the
6439 time you'll want to use C<sv_setsv> or one of its many macro front-ends.
6445 Perl_sv_replace(pTHX_ SV *const sv, SV *const nsv)
6447 const U32 refcnt = SvREFCNT(sv);
6449 PERL_ARGS_ASSERT_SV_REPLACE;
6451 SV_CHECK_THINKFIRST_COW_DROP(sv);
6452 if (SvREFCNT(nsv) != 1) {
6453 Perl_croak(aTHX_ "panic: reference miscount on nsv in sv_replace()"
6454 " (%" UVuf " != 1)", (UV) SvREFCNT(nsv));
6456 if (SvMAGICAL(sv)) {
6460 sv_upgrade(nsv, SVt_PVMG);
6461 SvMAGIC_set(nsv, SvMAGIC(sv));
6462 SvFLAGS(nsv) |= SvMAGICAL(sv);
6464 SvMAGIC_set(sv, NULL);
6468 assert(!SvREFCNT(sv));
6469 #ifdef DEBUG_LEAKING_SCALARS
6470 sv->sv_flags = nsv->sv_flags;
6471 sv->sv_any = nsv->sv_any;
6472 sv->sv_refcnt = nsv->sv_refcnt;
6473 sv->sv_u = nsv->sv_u;
6475 StructCopy(nsv,sv,SV);
6477 if(SvTYPE(sv) == SVt_IV) {
6478 SET_SVANY_FOR_BODYLESS_IV(sv);
6482 SvREFCNT(sv) = refcnt;
6483 SvFLAGS(nsv) |= SVTYPEMASK; /* Mark as freed */
6488 /* We're about to free a GV which has a CV that refers back to us.
6489 * If that CV will outlive us, make it anonymous (i.e. fix up its CvGV
6493 S_anonymise_cv_maybe(pTHX_ GV *gv, CV* cv)
6498 PERL_ARGS_ASSERT_ANONYMISE_CV_MAYBE;
6501 assert(SvREFCNT(gv) == 0);
6502 assert(isGV(gv) && isGV_with_GP(gv));
6504 assert(!CvANON(cv));
6505 assert(CvGV(cv) == gv);
6506 assert(!CvNAMED(cv));
6508 /* will the CV shortly be freed by gp_free() ? */
6509 if (GvCV(gv) == cv && GvGP(gv)->gp_refcnt < 2 && SvREFCNT(cv) < 2) {
6510 SvANY(cv)->xcv_gv_u.xcv_gv = NULL;
6514 /* if not, anonymise: */
6515 gvname = (GvSTASH(gv) && HvNAME(GvSTASH(gv)) && HvENAME(GvSTASH(gv)))
6516 ? newSVhek(HvENAME_HEK(GvSTASH(gv)))
6517 : newSVpvn_flags( "__ANON__", 8, 0 );
6518 sv_catpvs(gvname, "::__ANON__");
6519 anongv = gv_fetchsv(gvname, GV_ADDMULTI, SVt_PVCV);
6520 SvREFCNT_dec_NN(gvname);
6524 SvANY(cv)->xcv_gv_u.xcv_gv = MUTABLE_GV(SvREFCNT_inc(anongv));
6529 =for apidoc sv_clear
6531 Clear an SV: call any destructors, free up any memory used by the body,
6532 and free the body itself. The SV's head is I<not> freed, although
6533 its type is set to all 1's so that it won't inadvertently be assumed
6534 to be live during global destruction etc.
6535 This function should only be called when C<REFCNT> is zero. Most of the time
6536 you'll want to call C<sv_free()> (or its macro wrapper C<SvREFCNT_dec>)
6543 Perl_sv_clear(pTHX_ SV *const orig_sv)
6548 const struct body_details *sv_type_details;
6552 STRLEN hash_index = 0; /* initialise to make Coverity et al happy.
6553 Not strictly necessary */
6555 PERL_ARGS_ASSERT_SV_CLEAR;
6557 /* within this loop, sv is the SV currently being freed, and
6558 * iter_sv is the most recent AV or whatever that's being iterated
6559 * over to provide more SVs */
6565 assert(SvREFCNT(sv) == 0);
6566 assert(SvTYPE(sv) != (svtype)SVTYPEMASK);
6568 if (type <= SVt_IV) {
6569 /* See the comment in sv.h about the collusion between this
6570 * early return and the overloading of the NULL slots in the
6574 SvFLAGS(sv) &= SVf_BREAK;
6575 SvFLAGS(sv) |= SVTYPEMASK;
6579 /* objs are always >= MG, but pad names use the SVs_OBJECT flag
6580 for another purpose */
6581 assert(!SvOBJECT(sv) || type >= SVt_PVMG);
6583 if (type >= SVt_PVMG) {
6585 if (!curse(sv, 1)) goto get_next_sv;
6586 type = SvTYPE(sv); /* destructor may have changed it */
6588 /* Free back-references before magic, in case the magic calls
6589 * Perl code that has weak references to sv. */
6590 if (type == SVt_PVHV) {
6591 Perl_hv_kill_backrefs(aTHX_ MUTABLE_HV(sv));
6595 else if (SvMAGIC(sv)) {
6596 /* Free back-references before other types of magic. */
6597 sv_unmagic(sv, PERL_MAGIC_backref);
6603 /* case SVt_INVLIST: */
6606 IoIFP(sv) != PerlIO_stdin() &&
6607 IoIFP(sv) != PerlIO_stdout() &&
6608 IoIFP(sv) != PerlIO_stderr() &&
6609 !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6611 io_close(MUTABLE_IO(sv), NULL, FALSE,
6612 (IoTYPE(sv) == IoTYPE_WRONLY ||
6613 IoTYPE(sv) == IoTYPE_RDWR ||
6614 IoTYPE(sv) == IoTYPE_APPEND));
6616 if (IoDIRP(sv) && !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6617 PerlDir_close(IoDIRP(sv));
6618 IoDIRP(sv) = (DIR*)NULL;
6619 Safefree(IoTOP_NAME(sv));
6620 Safefree(IoFMT_NAME(sv));
6621 Safefree(IoBOTTOM_NAME(sv));
6622 if ((const GV *)sv == PL_statgv)
6626 /* FIXME for plugins */
6627 pregfree2((REGEXP*) sv);
6631 cv_undef(MUTABLE_CV(sv));
6632 /* If we're in a stash, we don't own a reference to it.
6633 * However it does have a back reference to us, which needs to
6635 if ((stash = CvSTASH(sv)))
6636 sv_del_backref(MUTABLE_SV(stash), sv);
6639 if (PL_last_swash_hv == (const HV *)sv) {
6640 PL_last_swash_hv = NULL;
6642 if (HvTOTALKEYS((HV*)sv) > 0) {
6644 /* this statement should match the one at the beginning of
6645 * hv_undef_flags() */
6646 if ( PL_phase != PERL_PHASE_DESTRUCT
6647 && (hek = HvNAME_HEK((HV*)sv)))
6649 if (PL_stashcache) {
6650 DEBUG_o(Perl_deb(aTHX_
6651 "sv_clear clearing PL_stashcache for '%" HEKf
6654 (void)hv_deletehek(PL_stashcache,
6657 hv_name_set((HV*)sv, NULL, 0, 0);
6660 /* save old iter_sv in unused SvSTASH field */
6661 assert(!SvOBJECT(sv));
6662 SvSTASH(sv) = (HV*)iter_sv;
6665 /* save old hash_index in unused SvMAGIC field */
6666 assert(!SvMAGICAL(sv));
6667 assert(!SvMAGIC(sv));
6668 ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index = hash_index;
6671 next_sv = Perl_hfree_next_entry(aTHX_ (HV*)sv, &hash_index);
6672 goto get_next_sv; /* process this new sv */
6674 /* free empty hash */
6675 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6676 assert(!HvARRAY((HV*)sv));
6680 AV* av = MUTABLE_AV(sv);
6681 if (PL_comppad == av) {
6685 if (AvREAL(av) && AvFILLp(av) > -1) {
6686 next_sv = AvARRAY(av)[AvFILLp(av)--];
6687 /* save old iter_sv in top-most slot of AV,
6688 * and pray that it doesn't get wiped in the meantime */
6689 AvARRAY(av)[AvMAX(av)] = iter_sv;
6691 goto get_next_sv; /* process this new sv */
6693 Safefree(AvALLOC(av));
6698 if (LvTYPE(sv) == 'T') { /* for tie: return HE to pool */
6699 SvREFCNT_dec(HeKEY_sv((HE*)LvTARG(sv)));
6700 HeNEXT((HE*)LvTARG(sv)) = PL_hv_fetch_ent_mh;
6701 PL_hv_fetch_ent_mh = (HE*)LvTARG(sv);
6703 else if (LvTYPE(sv) != 't') /* unless tie: unrefcnted fake SV** */
6704 SvREFCNT_dec(LvTARG(sv));
6706 /* SvLEN points to a regex body. Free the body, then
6707 * set SvLEN to whatever value was in the now-freed
6708 * regex body. The PVX buffer is shared by multiple re's
6709 * and only freed once, by the re whose len in non-null */
6710 STRLEN len = ReANY(sv)->xpv_len;
6711 pregfree2((REGEXP*) sv);
6712 SvLEN_set((sv), len);
6717 if (isGV_with_GP(sv)) {
6718 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
6719 && HvENAME_get(stash))
6720 mro_method_changed_in(stash);
6721 gp_free(MUTABLE_GV(sv));
6723 unshare_hek(GvNAME_HEK(sv));
6724 /* If we're in a stash, we don't own a reference to it.
6725 * However it does have a back reference to us, which
6726 * needs to be cleared. */
6727 if ((stash = GvSTASH(sv)))
6728 sv_del_backref(MUTABLE_SV(stash), sv);
6730 /* FIXME. There are probably more unreferenced pointers to SVs
6731 * in the interpreter struct that we should check and tidy in
6732 * a similar fashion to this: */
6733 /* See also S_sv_unglob, which does the same thing. */
6734 if ((const GV *)sv == PL_last_in_gv)
6735 PL_last_in_gv = NULL;
6736 else if ((const GV *)sv == PL_statgv)
6738 else if ((const GV *)sv == PL_stderrgv)
6747 /* Don't bother with SvOOK_off(sv); as we're only going to
6751 SvOOK_offset(sv, offset);
6752 SvPV_set(sv, SvPVX_mutable(sv) - offset);
6753 /* Don't even bother with turning off the OOK flag. */
6758 SV * const target = SvRV(sv);
6760 sv_del_backref(target, sv);
6766 else if (SvPVX_const(sv)
6767 && !(SvTYPE(sv) == SVt_PVIO
6768 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6773 PerlIO_printf(Perl_debug_log, "Copy on write: clear\n");
6778 if (CowREFCNT(sv)) {
6785 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6790 Safefree(SvPVX_mutable(sv));
6794 else if (SvPVX_const(sv) && SvLEN(sv)
6795 && !(SvTYPE(sv) == SVt_PVIO
6796 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6797 Safefree(SvPVX_mutable(sv));
6798 else if (SvPVX_const(sv) && SvIsCOW(sv)) {
6799 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6809 SvFLAGS(sv) &= SVf_BREAK;
6810 SvFLAGS(sv) |= SVTYPEMASK;
6812 sv_type_details = bodies_by_type + type;
6813 if (sv_type_details->arena) {
6814 del_body(((char *)SvANY(sv) + sv_type_details->offset),
6815 &PL_body_roots[type]);
6817 else if (sv_type_details->body_size) {
6818 safefree(SvANY(sv));
6822 /* caller is responsible for freeing the head of the original sv */
6823 if (sv != orig_sv && !SvREFCNT(sv))
6826 /* grab and free next sv, if any */
6834 else if (!iter_sv) {
6836 } else if (SvTYPE(iter_sv) == SVt_PVAV) {
6837 AV *const av = (AV*)iter_sv;
6838 if (AvFILLp(av) > -1) {
6839 sv = AvARRAY(av)[AvFILLp(av)--];
6841 else { /* no more elements of current AV to free */
6844 /* restore previous value, squirrelled away */
6845 iter_sv = AvARRAY(av)[AvMAX(av)];
6846 Safefree(AvALLOC(av));
6849 } else if (SvTYPE(iter_sv) == SVt_PVHV) {
6850 sv = Perl_hfree_next_entry(aTHX_ (HV*)iter_sv, &hash_index);
6851 if (!sv && !HvTOTALKEYS((HV *)iter_sv)) {
6852 /* no more elements of current HV to free */
6855 /* Restore previous values of iter_sv and hash_index,
6856 * squirrelled away */
6857 assert(!SvOBJECT(sv));
6858 iter_sv = (SV*)SvSTASH(sv);
6859 assert(!SvMAGICAL(sv));
6860 hash_index = ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index;
6862 /* perl -DA does not like rubbish in SvMAGIC. */
6866 /* free any remaining detritus from the hash struct */
6867 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6868 assert(!HvARRAY((HV*)sv));
6873 /* unrolled SvREFCNT_dec and sv_free2 follows: */
6877 if (!SvREFCNT(sv)) {
6881 if (--(SvREFCNT(sv)))
6885 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6886 "Attempt to free temp prematurely: SV 0x%" UVxf
6887 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6891 if (SvIMMORTAL(sv)) {
6892 /* make sure SvREFCNT(sv)==0 happens very seldom */
6893 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6902 /* This routine curses the sv itself, not the object referenced by sv. So
6903 sv does not have to be ROK. */
6906 S_curse(pTHX_ SV * const sv, const bool check_refcnt) {
6907 PERL_ARGS_ASSERT_CURSE;
6908 assert(SvOBJECT(sv));
6910 if (PL_defstash && /* Still have a symbol table? */
6916 stash = SvSTASH(sv);
6917 assert(SvTYPE(stash) == SVt_PVHV);
6918 if (HvNAME(stash)) {
6919 CV* destructor = NULL;
6920 struct mro_meta *meta;
6922 assert (SvOOK(stash));
6924 DEBUG_o( Perl_deb(aTHX_ "Looking for DESTROY method for %s\n",
6927 /* don't make this an initialization above the assert, since it needs
6929 meta = HvMROMETA(stash);
6930 if (meta->destroy_gen && meta->destroy_gen == PL_sub_generation) {
6931 destructor = meta->destroy;
6932 DEBUG_o( Perl_deb(aTHX_ "Using cached DESTROY method %p for %s\n",
6933 (void *)destructor, HvNAME(stash)) );
6936 bool autoload = FALSE;
6938 gv_fetchmeth_pvn(stash, S_destroy, S_destroy_len, -1, 0);
6940 destructor = GvCV(gv);
6942 gv = gv_autoload_pvn(stash, S_destroy, S_destroy_len,
6943 GV_AUTOLOAD_ISMETHOD);
6945 destructor = GvCV(gv);
6949 /* we don't cache AUTOLOAD for DESTROY, since this code
6950 would then need to set $__PACKAGE__::AUTOLOAD, or the
6951 equivalent for XS AUTOLOADs */
6953 meta->destroy_gen = PL_sub_generation;
6954 meta->destroy = destructor;
6956 DEBUG_o( Perl_deb(aTHX_ "Set cached DESTROY method %p for %s\n",
6957 (void *)destructor, HvNAME(stash)) );
6960 DEBUG_o( Perl_deb(aTHX_ "Not caching AUTOLOAD for DESTROY method for %s\n",
6964 assert(!destructor || SvTYPE(destructor) == SVt_PVCV);
6966 /* A constant subroutine can have no side effects, so
6967 don't bother calling it. */
6968 && !CvCONST(destructor)
6969 /* Don't bother calling an empty destructor or one that
6970 returns immediately. */
6971 && (CvISXSUB(destructor)
6972 || (CvSTART(destructor)
6973 && (CvSTART(destructor)->op_next->op_type
6975 && (CvSTART(destructor)->op_next->op_type
6977 || CvSTART(destructor)->op_next->op_next->op_type
6983 SV* const tmpref = newRV(sv);
6984 SvREADONLY_on(tmpref); /* DESTROY() could be naughty */
6986 PUSHSTACKi(PERLSI_DESTROY);
6991 call_sv(MUTABLE_SV(destructor),
6992 G_DISCARD|G_EVAL|G_KEEPERR|G_VOID);
6996 if(SvREFCNT(tmpref) < 2) {
6997 /* tmpref is not kept alive! */
6999 SvRV_set(tmpref, NULL);
7002 SvREFCNT_dec_NN(tmpref);
7005 } while (SvOBJECT(sv) && SvSTASH(sv) != stash);
7008 if (check_refcnt && SvREFCNT(sv)) {
7009 if (PL_in_clean_objs)
7011 "DESTROY created new reference to dead object '%" HEKf "'",
7012 HEKfARG(HvNAME_HEK(stash)));
7013 /* DESTROY gave object new lease on life */
7019 HV * const stash = SvSTASH(sv);
7020 /* Curse before freeing the stash, as freeing the stash could cause
7021 a recursive call into S_curse. */
7022 SvOBJECT_off(sv); /* Curse the object. */
7023 SvSTASH_set(sv,0); /* SvREFCNT_dec may try to read this */
7024 SvREFCNT_dec(stash); /* possibly of changed persuasion */
7030 =for apidoc sv_newref
7032 Increment an SV's reference count. Use the C<SvREFCNT_inc()> wrapper
7039 Perl_sv_newref(pTHX_ SV *const sv)
7041 PERL_UNUSED_CONTEXT;
7050 Decrement an SV's reference count, and if it drops to zero, call
7051 C<sv_clear> to invoke destructors and free up any memory used by
7052 the body; finally, deallocating the SV's head itself.
7053 Normally called via a wrapper macro C<SvREFCNT_dec>.
7059 Perl_sv_free(pTHX_ SV *const sv)
7065 /* Private helper function for SvREFCNT_dec().
7066 * Called with rc set to original SvREFCNT(sv), where rc == 0 or 1 */
7069 Perl_sv_free2(pTHX_ SV *const sv, const U32 rc)
7073 PERL_ARGS_ASSERT_SV_FREE2;
7075 if (LIKELY( rc == 1 )) {
7081 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
7082 "Attempt to free temp prematurely: SV 0x%" UVxf
7083 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
7087 if (SvIMMORTAL(sv)) {
7088 /* make sure SvREFCNT(sv)==0 happens very seldom */
7089 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
7093 if (! SvREFCNT(sv)) /* may have have been resurrected */
7098 /* handle exceptional cases */
7102 if (SvFLAGS(sv) & SVf_BREAK)
7103 /* this SV's refcnt has been artificially decremented to
7104 * trigger cleanup */
7106 if (PL_in_clean_all) /* All is fair */
7108 if (SvIMMORTAL(sv)) {
7109 /* make sure SvREFCNT(sv)==0 happens very seldom */
7110 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
7113 if (ckWARN_d(WARN_INTERNAL)) {
7114 #ifdef DEBUG_LEAKING_SCALARS_FORK_DUMP
7115 Perl_dump_sv_child(aTHX_ sv);
7117 #ifdef DEBUG_LEAKING_SCALARS
7120 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7121 if (PL_warnhook == PERL_WARNHOOK_FATAL
7122 || ckDEAD(packWARN(WARN_INTERNAL))) {
7123 /* Don't let Perl_warner cause us to escape our fate: */
7127 /* This may not return: */
7128 Perl_warner(aTHX_ packWARN(WARN_INTERNAL),
7129 "Attempt to free unreferenced scalar: SV 0x%" UVxf
7130 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
7133 #ifdef DEBUG_LEAKING_SCALARS_ABORT
7143 Returns the length of the string in the SV. Handles magic and type
7144 coercion and sets the UTF8 flag appropriately. See also C<L</SvCUR>>, which
7145 gives raw access to the C<xpv_cur> slot.
7151 Perl_sv_len(pTHX_ SV *const sv)
7158 (void)SvPV_const(sv, len);
7163 =for apidoc sv_len_utf8
7165 Returns the number of characters in the string in an SV, counting wide
7166 UTF-8 bytes as a single character. Handles magic and type coercion.
7172 * The length is cached in PERL_MAGIC_utf8, in the mg_len field. Also the
7173 * mg_ptr is used, by sv_pos_u2b() and sv_pos_b2u() - see the comments below.
7174 * (Note that the mg_len is not the length of the mg_ptr field.
7175 * This allows the cache to store the character length of the string without
7176 * needing to malloc() extra storage to attach to the mg_ptr.)
7181 Perl_sv_len_utf8(pTHX_ SV *const sv)
7187 return sv_len_utf8_nomg(sv);
7191 Perl_sv_len_utf8_nomg(pTHX_ SV * const sv)
7194 const U8 *s = (U8*)SvPV_nomg_const(sv, len);
7196 PERL_ARGS_ASSERT_SV_LEN_UTF8_NOMG;
7198 if (PL_utf8cache && SvUTF8(sv)) {
7200 MAGIC *mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_utf8) : NULL;
7202 if (mg && (mg->mg_len != -1 || mg->mg_ptr)) {
7203 if (mg->mg_len != -1)
7206 /* We can use the offset cache for a headstart.
7207 The longer value is stored in the first pair. */
7208 STRLEN *cache = (STRLEN *) mg->mg_ptr;
7210 ulen = cache[0] + Perl_utf8_length(aTHX_ s + cache[1],
7214 if (PL_utf8cache < 0) {
7215 const STRLEN real = Perl_utf8_length(aTHX_ s, s + len);
7216 assert_uft8_cache_coherent("sv_len_utf8", ulen, real, sv);
7220 ulen = Perl_utf8_length(aTHX_ s, s + len);
7221 utf8_mg_len_cache_update(sv, &mg, ulen);
7225 return SvUTF8(sv) ? Perl_utf8_length(aTHX_ s, s + len) : len;
7228 /* Walk forwards to find the byte corresponding to the passed in UTF-8
7231 S_sv_pos_u2b_forwards(const U8 *const start, const U8 *const send,
7232 STRLEN *const uoffset_p, bool *const at_end)
7234 const U8 *s = start;
7235 STRLEN uoffset = *uoffset_p;
7237 PERL_ARGS_ASSERT_SV_POS_U2B_FORWARDS;
7239 while (s < send && uoffset) {
7246 else if (s > send) {
7248 /* This is the existing behaviour. Possibly it should be a croak, as
7249 it's actually a bounds error */
7252 *uoffset_p -= uoffset;
7256 /* Given the length of the string in both bytes and UTF-8 characters, decide
7257 whether to walk forwards or backwards to find the byte corresponding to
7258 the passed in UTF-8 offset. */
7260 S_sv_pos_u2b_midway(const U8 *const start, const U8 *send,
7261 STRLEN uoffset, const STRLEN uend)
7263 STRLEN backw = uend - uoffset;
7265 PERL_ARGS_ASSERT_SV_POS_U2B_MIDWAY;
7267 if (uoffset < 2 * backw) {
7268 /* The assumption is that going forwards is twice the speed of going
7269 forward (that's where the 2 * backw comes from).
7270 (The real figure of course depends on the UTF-8 data.) */
7271 const U8 *s = start;
7273 while (s < send && uoffset--)
7283 while (UTF8_IS_CONTINUATION(*send))
7286 return send - start;
7289 /* For the string representation of the given scalar, find the byte
7290 corresponding to the passed in UTF-8 offset. uoffset0 and boffset0
7291 give another position in the string, *before* the sought offset, which
7292 (which is always true, as 0, 0 is a valid pair of positions), which should
7293 help reduce the amount of linear searching.
7294 If *mgp is non-NULL, it should point to the UTF-8 cache magic, which
7295 will be used to reduce the amount of linear searching. The cache will be
7296 created if necessary, and the found value offered to it for update. */
7298 S_sv_pos_u2b_cached(pTHX_ SV *const sv, MAGIC **const mgp, const U8 *const start,
7299 const U8 *const send, STRLEN uoffset,
7300 STRLEN uoffset0, STRLEN boffset0)
7302 STRLEN boffset = 0; /* Actually always set, but let's keep gcc happy. */
7304 bool at_end = FALSE;
7306 PERL_ARGS_ASSERT_SV_POS_U2B_CACHED;
7308 assert (uoffset >= uoffset0);
7313 if (!SvREADONLY(sv) && !SvGMAGICAL(sv) && SvPOK(sv)
7315 && (*mgp || (SvTYPE(sv) >= SVt_PVMG &&
7316 (*mgp = mg_find(sv, PERL_MAGIC_utf8))))) {
7317 if ((*mgp)->mg_ptr) {
7318 STRLEN *cache = (STRLEN *) (*mgp)->mg_ptr;
7319 if (cache[0] == uoffset) {
7320 /* An exact match. */
7323 if (cache[2] == uoffset) {
7324 /* An exact match. */
7328 if (cache[0] < uoffset) {
7329 /* The cache already knows part of the way. */
7330 if (cache[0] > uoffset0) {
7331 /* The cache knows more than the passed in pair */
7332 uoffset0 = cache[0];
7333 boffset0 = cache[1];
7335 if ((*mgp)->mg_len != -1) {
7336 /* And we know the end too. */
7338 + sv_pos_u2b_midway(start + boffset0, send,
7340 (*mgp)->mg_len - uoffset0);
7342 uoffset -= uoffset0;
7344 + sv_pos_u2b_forwards(start + boffset0,
7345 send, &uoffset, &at_end);
7346 uoffset += uoffset0;
7349 else if (cache[2] < uoffset) {
7350 /* We're between the two cache entries. */
7351 if (cache[2] > uoffset0) {
7352 /* and the cache knows more than the passed in pair */
7353 uoffset0 = cache[2];
7354 boffset0 = cache[3];
7358 + sv_pos_u2b_midway(start + boffset0,
7361 cache[0] - uoffset0);
7364 + sv_pos_u2b_midway(start + boffset0,
7367 cache[2] - uoffset0);
7371 else if ((*mgp)->mg_len != -1) {
7372 /* If we can take advantage of a passed in offset, do so. */
7373 /* In fact, offset0 is either 0, or less than offset, so don't
7374 need to worry about the other possibility. */
7376 + sv_pos_u2b_midway(start + boffset0, send,
7378 (*mgp)->mg_len - uoffset0);
7383 if (!found || PL_utf8cache < 0) {
7384 STRLEN real_boffset;
7385 uoffset -= uoffset0;
7386 real_boffset = boffset0 + sv_pos_u2b_forwards(start + boffset0,
7387 send, &uoffset, &at_end);
7388 uoffset += uoffset0;
7390 if (found && PL_utf8cache < 0)
7391 assert_uft8_cache_coherent("sv_pos_u2b_cache", boffset,
7393 boffset = real_boffset;
7396 if (PL_utf8cache && !SvGMAGICAL(sv) && SvPOK(sv)) {
7398 utf8_mg_len_cache_update(sv, mgp, uoffset);
7400 utf8_mg_pos_cache_update(sv, mgp, boffset, uoffset, send - start);
7407 =for apidoc sv_pos_u2b_flags
7409 Converts the offset from a count of UTF-8 chars from
7410 the start of the string, to a count of the equivalent number of bytes; if
7411 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7412 C<offset>, rather than from the start
7413 of the string. Handles type coercion.
7414 C<flags> is passed to C<SvPV_flags>, and usually should be
7415 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7421 * sv_pos_u2b_flags() uses, like sv_pos_b2u(), the mg_ptr of the potential
7422 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7423 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7428 Perl_sv_pos_u2b_flags(pTHX_ SV *const sv, STRLEN uoffset, STRLEN *const lenp,
7435 PERL_ARGS_ASSERT_SV_POS_U2B_FLAGS;
7437 start = (U8*)SvPV_flags(sv, len, flags);
7439 const U8 * const send = start + len;
7441 boffset = sv_pos_u2b_cached(sv, &mg, start, send, uoffset, 0, 0);
7444 && *lenp /* don't bother doing work for 0, as its bytes equivalent
7445 is 0, and *lenp is already set to that. */) {
7446 /* Convert the relative offset to absolute. */
7447 const STRLEN uoffset2 = uoffset + *lenp;
7448 const STRLEN boffset2
7449 = sv_pos_u2b_cached(sv, &mg, start, send, uoffset2,
7450 uoffset, boffset) - boffset;
7464 =for apidoc sv_pos_u2b
7466 Converts the value pointed to by C<offsetp> from a count of UTF-8 chars from
7467 the start of the string, to a count of the equivalent number of bytes; if
7468 C<lenp> is non-zero, it does the same to C<lenp>, but this time starting from
7469 the offset, rather than from the start of the string. Handles magic and
7472 Use C<sv_pos_u2b_flags> in preference, which correctly handles strings longer
7479 * sv_pos_u2b() uses, like sv_pos_b2u(), the mg_ptr of the potential
7480 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7481 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7485 /* This function is subject to size and sign problems */
7488 Perl_sv_pos_u2b(pTHX_ SV *const sv, I32 *const offsetp, I32 *const lenp)
7490 PERL_ARGS_ASSERT_SV_POS_U2B;
7493 STRLEN ulen = (STRLEN)*lenp;
7494 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, &ulen,
7495 SV_GMAGIC|SV_CONST_RETURN);
7498 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, NULL,
7499 SV_GMAGIC|SV_CONST_RETURN);
7504 S_utf8_mg_len_cache_update(pTHX_ SV *const sv, MAGIC **const mgp,
7507 PERL_ARGS_ASSERT_UTF8_MG_LEN_CACHE_UPDATE;
7508 if (SvREADONLY(sv) || SvGMAGICAL(sv) || !SvPOK(sv))
7511 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7512 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7513 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, &PL_vtbl_utf8, 0, 0);
7517 (*mgp)->mg_len = ulen;
7520 /* Create and update the UTF8 magic offset cache, with the proffered utf8/
7521 byte length pairing. The (byte) length of the total SV is passed in too,
7522 as blen, because for some (more esoteric) SVs, the call to SvPV_const()
7523 may not have updated SvCUR, so we can't rely on reading it directly.
7525 The proffered utf8/byte length pairing isn't used if the cache already has
7526 two pairs, and swapping either for the proffered pair would increase the
7527 RMS of the intervals between known byte offsets.
7529 The cache itself consists of 4 STRLEN values
7530 0: larger UTF-8 offset
7531 1: corresponding byte offset
7532 2: smaller UTF-8 offset
7533 3: corresponding byte offset
7535 Unused cache pairs have the value 0, 0.
7536 Keeping the cache "backwards" means that the invariant of
7537 cache[0] >= cache[2] is maintained even with empty slots, which means that
7538 the code that uses it doesn't need to worry if only 1 entry has actually
7539 been set to non-zero. It also makes the "position beyond the end of the
7540 cache" logic much simpler, as the first slot is always the one to start
7544 S_utf8_mg_pos_cache_update(pTHX_ SV *const sv, MAGIC **const mgp, const STRLEN byte,
7545 const STRLEN utf8, const STRLEN blen)
7549 PERL_ARGS_ASSERT_UTF8_MG_POS_CACHE_UPDATE;
7554 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7555 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7556 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, (MGVTBL*)&PL_vtbl_utf8, 0,
7558 (*mgp)->mg_len = -1;
7562 if (!(cache = (STRLEN *)(*mgp)->mg_ptr)) {
7563 Newxz(cache, PERL_MAGIC_UTF8_CACHESIZE * 2, STRLEN);
7564 (*mgp)->mg_ptr = (char *) cache;
7568 if (PL_utf8cache < 0 && SvPOKp(sv)) {
7569 /* SvPOKp() because, if sv is a reference, then SvPVX() is actually
7570 a pointer. Note that we no longer cache utf8 offsets on refer-
7571 ences, but this check is still a good idea, for robustness. */
7572 const U8 *start = (const U8 *) SvPVX_const(sv);
7573 const STRLEN realutf8 = utf8_length(start, start + byte);
7575 assert_uft8_cache_coherent("utf8_mg_pos_cache_update", utf8, realutf8,
7579 /* Cache is held with the later position first, to simplify the code
7580 that deals with unbounded ends. */
7582 ASSERT_UTF8_CACHE(cache);
7583 if (cache[1] == 0) {
7584 /* Cache is totally empty */
7587 } else if (cache[3] == 0) {
7588 if (byte > cache[1]) {
7589 /* New one is larger, so goes first. */
7590 cache[2] = cache[0];
7591 cache[3] = cache[1];
7599 /* float casts necessary? XXX */
7600 #define THREEWAY_SQUARE(a,b,c,d) \
7601 ((float)((d) - (c))) * ((float)((d) - (c))) \
7602 + ((float)((c) - (b))) * ((float)((c) - (b))) \
7603 + ((float)((b) - (a))) * ((float)((b) - (a)))
7605 /* Cache has 2 slots in use, and we know three potential pairs.
7606 Keep the two that give the lowest RMS distance. Do the
7607 calculation in bytes simply because we always know the byte
7608 length. squareroot has the same ordering as the positive value,
7609 so don't bother with the actual square root. */
7610 if (byte > cache[1]) {
7611 /* New position is after the existing pair of pairs. */
7612 const float keep_earlier
7613 = THREEWAY_SQUARE(0, cache[3], byte, blen);
7614 const float keep_later
7615 = THREEWAY_SQUARE(0, cache[1], byte, blen);
7617 if (keep_later < keep_earlier) {
7618 cache[2] = cache[0];
7619 cache[3] = cache[1];
7625 const float keep_later = THREEWAY_SQUARE(0, byte, cache[1], blen);
7626 float b, c, keep_earlier;
7627 if (byte > cache[3]) {
7628 /* New position is between the existing pair of pairs. */
7629 b = (float)cache[3];
7632 /* New position is before the existing pair of pairs. */
7634 c = (float)cache[3];
7636 keep_earlier = THREEWAY_SQUARE(0, b, c, blen);
7637 if (byte > cache[3]) {
7638 if (keep_later < keep_earlier) {
7648 if (! (keep_later < keep_earlier)) {
7649 cache[0] = cache[2];
7650 cache[1] = cache[3];
7657 ASSERT_UTF8_CACHE(cache);
7660 /* We already know all of the way, now we may be able to walk back. The same
7661 assumption is made as in S_sv_pos_u2b_midway(), namely that walking
7662 backward is half the speed of walking forward. */
7664 S_sv_pos_b2u_midway(pTHX_ const U8 *const s, const U8 *const target,
7665 const U8 *end, STRLEN endu)
7667 const STRLEN forw = target - s;
7668 STRLEN backw = end - target;
7670 PERL_ARGS_ASSERT_SV_POS_B2U_MIDWAY;
7672 if (forw < 2 * backw) {
7673 return utf8_length(s, target);
7676 while (end > target) {
7678 while (UTF8_IS_CONTINUATION(*end)) {
7687 =for apidoc sv_pos_b2u_flags
7689 Converts C<offset> from a count of bytes from the start of the string, to
7690 a count of the equivalent number of UTF-8 chars. Handles type coercion.
7691 C<flags> is passed to C<SvPV_flags>, and usually should be
7692 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7698 * sv_pos_b2u_flags() uses, like sv_pos_u2b_flags(), the mg_ptr of the
7699 * potential PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8
7704 Perl_sv_pos_b2u_flags(pTHX_ SV *const sv, STRLEN const offset, U32 flags)
7707 STRLEN len = 0; /* Actually always set, but let's keep gcc happy. */
7713 PERL_ARGS_ASSERT_SV_POS_B2U_FLAGS;
7715 s = (const U8*)SvPV_flags(sv, blen, flags);
7718 Perl_croak(aTHX_ "panic: sv_pos_b2u: bad byte offset, blen=%" UVuf
7719 ", byte=%" UVuf, (UV)blen, (UV)offset);
7725 && SvTYPE(sv) >= SVt_PVMG
7726 && (mg = mg_find(sv, PERL_MAGIC_utf8)))
7729 STRLEN * const cache = (STRLEN *) mg->mg_ptr;
7730 if (cache[1] == offset) {
7731 /* An exact match. */
7734 if (cache[3] == offset) {
7735 /* An exact match. */
7739 if (cache[1] < offset) {
7740 /* We already know part of the way. */
7741 if (mg->mg_len != -1) {
7742 /* Actually, we know the end too. */
7744 + S_sv_pos_b2u_midway(aTHX_ s + cache[1], send,
7745 s + blen, mg->mg_len - cache[0]);
7747 len = cache[0] + utf8_length(s + cache[1], send);
7750 else if (cache[3] < offset) {
7751 /* We're between the two cached pairs, so we do the calculation
7752 offset by the byte/utf-8 positions for the earlier pair,
7753 then add the utf-8 characters from the string start to
7755 len = S_sv_pos_b2u_midway(aTHX_ s + cache[3], send,
7756 s + cache[1], cache[0] - cache[2])
7760 else { /* cache[3] > offset */
7761 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + cache[3],
7765 ASSERT_UTF8_CACHE(cache);
7767 } else if (mg->mg_len != -1) {
7768 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + blen, mg->mg_len);
7772 if (!found || PL_utf8cache < 0) {
7773 const STRLEN real_len = utf8_length(s, send);
7775 if (found && PL_utf8cache < 0)
7776 assert_uft8_cache_coherent("sv_pos_b2u", len, real_len, sv);
7782 utf8_mg_len_cache_update(sv, &mg, len);
7784 utf8_mg_pos_cache_update(sv, &mg, offset, len, blen);
7791 =for apidoc sv_pos_b2u
7793 Converts the value pointed to by C<offsetp> from a count of bytes from the
7794 start of the string, to a count of the equivalent number of UTF-8 chars.
7795 Handles magic and type coercion.
7797 Use C<sv_pos_b2u_flags> in preference, which correctly handles strings
7804 * sv_pos_b2u() uses, like sv_pos_u2b(), the mg_ptr of the potential
7805 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7810 Perl_sv_pos_b2u(pTHX_ SV *const sv, I32 *const offsetp)
7812 PERL_ARGS_ASSERT_SV_POS_B2U;
7817 *offsetp = (I32)sv_pos_b2u_flags(sv, (STRLEN)*offsetp,
7818 SV_GMAGIC|SV_CONST_RETURN);
7822 S_assert_uft8_cache_coherent(pTHX_ const char *const func, STRLEN from_cache,
7823 STRLEN real, SV *const sv)
7825 PERL_ARGS_ASSERT_ASSERT_UFT8_CACHE_COHERENT;
7827 /* As this is debugging only code, save space by keeping this test here,
7828 rather than inlining it in all the callers. */
7829 if (from_cache == real)
7832 /* Need to turn the assertions off otherwise we may recurse infinitely
7833 while printing error messages. */
7834 SAVEI8(PL_utf8cache);
7836 Perl_croak(aTHX_ "panic: %s cache %" UVuf " real %" UVuf " for %" SVf,
7837 func, (UV) from_cache, (UV) real, SVfARG(sv));
7843 Returns a boolean indicating whether the strings in the two SVs are
7844 identical. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7845 coerce its args to strings if necessary.
7847 =for apidoc sv_eq_flags
7849 Returns a boolean indicating whether the strings in the two SVs are
7850 identical. Is UTF-8 and S<C<'use bytes'>> aware and coerces its args to strings
7851 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get-magic, too.
7857 Perl_sv_eq_flags(pTHX_ SV *sv1, SV *sv2, const U32 flags)
7869 /* if pv1 and pv2 are the same, second SvPV_const call may
7870 * invalidate pv1 (if we are handling magic), so we may need to
7872 if (sv1 == sv2 && flags & SV_GMAGIC
7873 && (SvTHINKFIRST(sv1) || SvGMAGICAL(sv1))) {
7874 pv1 = SvPV_const(sv1, cur1);
7875 sv1 = newSVpvn_flags(pv1, cur1, SVs_TEMP | SvUTF8(sv2));
7877 pv1 = SvPV_flags_const(sv1, cur1, flags);
7885 pv2 = SvPV_flags_const(sv2, cur2, flags);
7887 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7888 /* Differing utf8ness. */
7890 /* sv1 is the UTF-8 one */
7891 return bytes_cmp_utf8((const U8*)pv2, cur2,
7892 (const U8*)pv1, cur1) == 0;
7895 /* sv2 is the UTF-8 one */
7896 return bytes_cmp_utf8((const U8*)pv1, cur1,
7897 (const U8*)pv2, cur2) == 0;
7902 return (pv1 == pv2) || memEQ(pv1, pv2, cur1);
7910 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7911 string in C<sv1> is less than, equal to, or greater than the string in
7912 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware, handles get magic, and will
7913 coerce its args to strings if necessary. See also C<L</sv_cmp_locale>>.
7915 =for apidoc sv_cmp_flags
7917 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7918 string in C<sv1> is less than, equal to, or greater than the string in
7919 C<sv2>. Is UTF-8 and S<C<'use bytes'>> aware and will coerce its args to strings
7920 if necessary. If the flags has the C<SV_GMAGIC> bit set, it handles get magic. See
7921 also C<L</sv_cmp_locale_flags>>.
7927 Perl_sv_cmp(pTHX_ SV *const sv1, SV *const sv2)
7929 return sv_cmp_flags(sv1, sv2, SV_GMAGIC);
7933 Perl_sv_cmp_flags(pTHX_ SV *const sv1, SV *const sv2,
7937 const char *pv1, *pv2;
7939 SV *svrecode = NULL;
7946 pv1 = SvPV_flags_const(sv1, cur1, flags);
7953 pv2 = SvPV_flags_const(sv2, cur2, flags);
7955 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7956 /* Differing utf8ness. */
7958 const int retval = -bytes_cmp_utf8((const U8*)pv2, cur2,
7959 (const U8*)pv1, cur1);
7960 return retval ? retval < 0 ? -1 : +1 : 0;
7963 const int retval = bytes_cmp_utf8((const U8*)pv1, cur1,
7964 (const U8*)pv2, cur2);
7965 return retval ? retval < 0 ? -1 : +1 : 0;
7969 /* Here, if both are non-NULL, then they have the same UTF8ness. */
7972 cmp = cur2 ? -1 : 0;
7976 STRLEN shortest_len = cur1 < cur2 ? cur1 : cur2;
7979 if (! DO_UTF8(sv1)) {
7981 const I32 retval = memcmp((const void*)pv1,
7985 cmp = retval < 0 ? -1 : 1;
7986 } else if (cur1 == cur2) {
7989 cmp = cur1 < cur2 ? -1 : 1;
7993 else { /* Both are to be treated as UTF-EBCDIC */
7995 /* EBCDIC UTF-8 is complicated by the fact that it is based on I8
7996 * which remaps code points 0-255. We therefore generally have to
7997 * unmap back to the original values to get an accurate comparison.
7998 * But we don't have to do that for UTF-8 invariants, as by
7999 * definition, they aren't remapped, nor do we have to do it for
8000 * above-latin1 code points, as they also aren't remapped. (This
8001 * code also works on ASCII platforms, but the memcmp() above is
8004 const char *e = pv1 + shortest_len;
8006 /* Find the first bytes that differ between the two strings */
8007 while (pv1 < e && *pv1 == *pv2) {
8013 if (pv1 == e) { /* Are the same all the way to the end */
8017 cmp = cur1 < cur2 ? -1 : 1;
8020 else /* Here *pv1 and *pv2 are not equal, but all bytes earlier
8021 * in the strings were. The current bytes may or may not be
8022 * at the beginning of a character. But neither or both are
8023 * (or else earlier bytes would have been different). And
8024 * if we are in the middle of a character, the two
8025 * characters are comprised of the same number of bytes
8026 * (because in this case the start bytes are the same, and
8027 * the start bytes encode the character's length). */
8028 if (UTF8_IS_INVARIANT(*pv1))
8030 /* If both are invariants; can just compare directly */
8031 if (UTF8_IS_INVARIANT(*pv2)) {
8032 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8034 else /* Since *pv1 is invariant, it is the whole character,
8035 which means it is at the beginning of a character.
8036 That means pv2 is also at the beginning of a
8037 character (see earlier comment). Since it isn't
8038 invariant, it must be a start byte. If it starts a
8039 character whose code point is above 255, that
8040 character is greater than any single-byte char, which
8042 if (UTF8_IS_ABOVE_LATIN1_START(*pv2))
8047 /* Here, pv2 points to a character composed of 2 bytes
8048 * whose code point is < 256. Get its code point and
8049 * compare with *pv1 */
8050 cmp = ((U8) *pv1 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
8055 else /* The code point starting at pv1 isn't a single byte */
8056 if (UTF8_IS_INVARIANT(*pv2))
8058 /* But here, the code point starting at *pv2 is a single byte,
8059 * and so *pv1 must begin a character, hence is a start byte.
8060 * If that character is above 255, it is larger than any
8061 * single-byte char, which *pv2 is */
8062 if (UTF8_IS_ABOVE_LATIN1_START(*pv1)) {
8066 /* Here, pv1 points to a character composed of 2 bytes
8067 * whose code point is < 256. Get its code point and
8068 * compare with the single byte character *pv2 */
8069 cmp = (EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1)) < (U8) *pv2)
8074 else /* Here, we've ruled out either *pv1 and *pv2 being
8075 invariant. That means both are part of variants, but not
8076 necessarily at the start of a character */
8077 if ( UTF8_IS_ABOVE_LATIN1_START(*pv1)
8078 || UTF8_IS_ABOVE_LATIN1_START(*pv2))
8080 /* Here, at least one is the start of a character, which means
8081 * the other is also a start byte. And the code point of at
8082 * least one of the characters is above 255. It is a
8083 * characteristic of UTF-EBCDIC that all start bytes for
8084 * above-latin1 code points are well behaved as far as code
8085 * point comparisons go, and all are larger than all other
8086 * start bytes, so the comparison with those is also well
8088 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8091 /* Here both *pv1 and *pv2 are part of variant characters.
8092 * They could be both continuations, or both start characters.
8093 * (One or both could even be an illegal start character (for
8094 * an overlong) which for the purposes of sorting we treat as
8096 if (UTF8_IS_CONTINUATION(*pv1)) {
8098 /* If they are continuations for code points above 255,
8099 * then comparing the current byte is sufficient, as there
8100 * is no remapping of these and so the comparison is
8101 * well-behaved. We determine if they are such
8102 * continuations by looking at the preceding byte. It
8103 * could be a start byte, from which we can tell if it is
8104 * for an above 255 code point. Or it could be a
8105 * continuation, which means the character occupies at
8106 * least 3 bytes, so must be above 255. */
8107 if ( UTF8_IS_CONTINUATION(*(pv2 - 1))
8108 || UTF8_IS_ABOVE_LATIN1_START(*(pv2 -1)))
8110 cmp = ((U8) *pv1 < (U8) *pv2) ? -1 : 1;
8114 /* Here, the continuations are for code points below 256;
8115 * back up one to get to the start byte */
8120 /* We need to get the actual native code point of each of these
8121 * variants in order to compare them */
8122 cmp = ( EIGHT_BIT_UTF8_TO_NATIVE(*pv1, *(pv1 + 1))
8123 < EIGHT_BIT_UTF8_TO_NATIVE(*pv2, *(pv2 + 1)))
8132 SvREFCNT_dec(svrecode);
8138 =for apidoc sv_cmp_locale
8140 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
8141 S<C<'use bytes'>> aware, handles get magic, and will coerce its args to strings
8142 if necessary. See also C<L</sv_cmp>>.
8144 =for apidoc sv_cmp_locale_flags
8146 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
8147 S<C<'use bytes'>> aware and will coerce its args to strings if necessary. If
8148 the flags contain C<SV_GMAGIC>, it handles get magic. See also
8149 C<L</sv_cmp_flags>>.
8155 Perl_sv_cmp_locale(pTHX_ SV *const sv1, SV *const sv2)
8157 return sv_cmp_locale_flags(sv1, sv2, SV_GMAGIC);
8161 Perl_sv_cmp_locale_flags(pTHX_ SV *const sv1, SV *const sv2,
8164 #ifdef USE_LOCALE_COLLATE
8170 if (PL_collation_standard)
8175 /* Revert to using raw compare if both operands exist, but either one
8176 * doesn't transform properly for collation */
8178 pv1 = sv_collxfrm_flags(sv1, &len1, flags);
8182 pv2 = sv_collxfrm_flags(sv2, &len2, flags);
8188 pv1 = sv1 ? sv_collxfrm_flags(sv1, &len1, flags) : (char *) NULL;
8189 pv2 = sv2 ? sv_collxfrm_flags(sv2, &len2, flags) : (char *) NULL;
8192 if (!pv1 || !len1) {
8203 retval = memcmp((void*)pv1, (void*)pv2, len1 < len2 ? len1 : len2);
8206 return retval < 0 ? -1 : 1;
8209 * When the result of collation is equality, that doesn't mean
8210 * that there are no differences -- some locales exclude some
8211 * characters from consideration. So to avoid false equalities,
8212 * we use the raw string as a tiebreaker.
8219 PERL_UNUSED_ARG(flags);
8220 #endif /* USE_LOCALE_COLLATE */
8222 return sv_cmp(sv1, sv2);
8226 #ifdef USE_LOCALE_COLLATE
8229 =for apidoc sv_collxfrm
8231 This calls C<sv_collxfrm_flags> with the SV_GMAGIC flag. See
8232 C<L</sv_collxfrm_flags>>.
8234 =for apidoc sv_collxfrm_flags
8236 Add Collate Transform magic to an SV if it doesn't already have it. If the
8237 flags contain C<SV_GMAGIC>, it handles get-magic.
8239 Any scalar variable may carry C<PERL_MAGIC_collxfrm> magic that contains the
8240 scalar data of the variable, but transformed to such a format that a normal
8241 memory comparison can be used to compare the data according to the locale
8248 Perl_sv_collxfrm_flags(pTHX_ SV *const sv, STRLEN *const nxp, const I32 flags)
8252 PERL_ARGS_ASSERT_SV_COLLXFRM_FLAGS;
8254 mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_collxfrm) : (MAGIC *) NULL;
8256 /* If we don't have collation magic on 'sv', or the locale has changed
8257 * since the last time we calculated it, get it and save it now */
8258 if (!mg || !mg->mg_ptr || *(U32*)mg->mg_ptr != PL_collation_ix) {
8263 /* Free the old space */
8265 Safefree(mg->mg_ptr);
8267 s = SvPV_flags_const(sv, len, flags);
8268 if ((xf = _mem_collxfrm(s, len, &xlen, cBOOL(SvUTF8(sv))))) {
8270 mg = sv_magicext(sv, 0, PERL_MAGIC_collxfrm, &PL_vtbl_collxfrm,
8285 if (mg && mg->mg_ptr) {
8287 return mg->mg_ptr + sizeof(PL_collation_ix);
8295 #endif /* USE_LOCALE_COLLATE */
8298 S_sv_gets_append_to_utf8(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8300 SV * const tsv = newSV(0);
8303 sv_gets(tsv, fp, 0);
8304 sv_utf8_upgrade_nomg(tsv);
8305 SvCUR_set(sv,append);
8308 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8312 S_sv_gets_read_record(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8315 const STRLEN recsize = SvUV(SvRV(PL_rs)); /* RsRECORD() guarantees > 0. */
8316 /* Grab the size of the record we're getting */
8317 char *buffer = SvGROW(sv, (STRLEN)(recsize + append + 1)) + append;
8324 /* With a true, record-oriented file on VMS, we need to use read directly
8325 * to ensure that we respect RMS record boundaries. The user is responsible
8326 * for providing a PL_rs value that corresponds to the FAB$W_MRS (maximum
8327 * record size) field. N.B. This is likely to produce invalid results on
8328 * varying-width character data when a record ends mid-character.
8330 fd = PerlIO_fileno(fp);
8332 && PerlLIO_fstat(fd, &st) == 0
8333 && (st.st_fab_rfm == FAB$C_VAR
8334 || st.st_fab_rfm == FAB$C_VFC
8335 || st.st_fab_rfm == FAB$C_FIX)) {
8337 bytesread = PerlLIO_read(fd, buffer, recsize);
8339 else /* in-memory file from PerlIO::Scalar
8340 * or not a record-oriented file
8344 bytesread = PerlIO_read(fp, buffer, recsize);
8346 /* At this point, the logic in sv_get() means that sv will
8347 be treated as utf-8 if the handle is utf8.
8349 if (PerlIO_isutf8(fp) && bytesread > 0) {
8350 char *bend = buffer + bytesread;
8351 char *bufp = buffer;
8352 size_t charcount = 0;
8353 bool charstart = TRUE;
8356 while (charcount < recsize) {
8357 /* count accumulated characters */
8358 while (bufp < bend) {
8360 skip = UTF8SKIP(bufp);
8362 if (bufp + skip > bend) {
8363 /* partial at the end */
8374 if (charcount < recsize) {
8376 STRLEN bufp_offset = bufp - buffer;
8377 SSize_t morebytesread;
8379 /* originally I read enough to fill any incomplete
8380 character and the first byte of the next
8381 character if needed, but if there's many
8382 multi-byte encoded characters we're going to be
8383 making a read call for every character beyond
8384 the original read size.
8386 So instead, read the rest of the character if
8387 any, and enough bytes to match at least the
8388 start bytes for each character we're going to
8392 readsize = recsize - charcount;
8394 readsize = skip - (bend - bufp) + recsize - charcount - 1;
8395 buffer = SvGROW(sv, append + bytesread + readsize + 1) + append;
8396 bend = buffer + bytesread;
8397 morebytesread = PerlIO_read(fp, bend, readsize);
8398 if (morebytesread <= 0) {
8399 /* we're done, if we still have incomplete
8400 characters the check code in sv_gets() will
8403 I'd originally considered doing
8404 PerlIO_ungetc() on all but the lead
8405 character of the incomplete character, but
8406 read() doesn't do that, so I don't.
8411 /* prepare to scan some more */
8412 bytesread += morebytesread;
8413 bend = buffer + bytesread;
8414 bufp = buffer + bufp_offset;
8422 SvCUR_set(sv, bytesread + append);
8423 buffer[bytesread] = '\0';
8424 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8430 Get a line from the filehandle and store it into the SV, optionally
8431 appending to the currently-stored string. If C<append> is not 0, the
8432 line is appended to the SV instead of overwriting it. C<append> should
8433 be set to the byte offset that the appended string should start at
8434 in the SV (typically, C<SvCUR(sv)> is a suitable choice).
8440 Perl_sv_gets(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8450 PERL_ARGS_ASSERT_SV_GETS;
8452 if (SvTHINKFIRST(sv))
8453 sv_force_normal_flags(sv, append ? 0 : SV_COW_DROP_PV);
8454 /* XXX. If you make this PVIV, then copy on write can copy scalars read
8456 However, perlbench says it's slower, because the existing swipe code
8457 is faster than copy on write.
8458 Swings and roundabouts. */
8459 SvUPGRADE(sv, SVt_PV);
8462 /* line is going to be appended to the existing buffer in the sv */
8463 if (PerlIO_isutf8(fp)) {
8465 sv_utf8_upgrade_nomg(sv);
8466 sv_pos_u2b(sv,&append,0);
8468 } else if (SvUTF8(sv)) {
8469 return S_sv_gets_append_to_utf8(aTHX_ sv, fp, append);
8475 /* not appending - "clear" the string by setting SvCUR to 0,
8476 * the pv is still avaiable. */
8479 if (PerlIO_isutf8(fp))
8482 if (IN_PERL_COMPILETIME) {
8483 /* we always read code in line mode */
8487 else if (RsSNARF(PL_rs)) {
8488 /* If it is a regular disk file use size from stat() as estimate
8489 of amount we are going to read -- may result in mallocing
8490 more memory than we really need if the layers below reduce
8491 the size we read (e.g. CRLF or a gzip layer).
8494 int fd = PerlIO_fileno(fp);
8495 if (fd >= 0 && (PerlLIO_fstat(fd, &st) == 0) && S_ISREG(st.st_mode)) {
8496 const Off_t offset = PerlIO_tell(fp);
8497 if (offset != (Off_t) -1 && st.st_size + append > offset) {
8498 #ifdef PERL_COPY_ON_WRITE
8499 /* Add an extra byte for the sake of copy-on-write's
8500 * buffer reference count. */
8501 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 2));
8503 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 1));
8510 else if (RsRECORD(PL_rs)) {
8511 return S_sv_gets_read_record(aTHX_ sv, fp, append);
8513 else if (RsPARA(PL_rs)) {
8519 /* Get $/ i.e. PL_rs into same encoding as stream wants */
8520 if (PerlIO_isutf8(fp)) {
8521 rsptr = SvPVutf8(PL_rs, rslen);
8524 if (SvUTF8(PL_rs)) {
8525 if (!sv_utf8_downgrade(PL_rs, TRUE)) {
8526 Perl_croak(aTHX_ "Wide character in $/");
8529 /* extract the raw pointer to the record separator */
8530 rsptr = SvPV_const(PL_rs, rslen);
8534 /* rslast is the last character in the record separator
8535 * note we don't use rslast except when rslen is true, so the
8536 * null assign is a placeholder. */
8537 rslast = rslen ? rsptr[rslen - 1] : '\0';
8539 if (rspara) { /* have to do this both before and after */
8540 /* to make sure file boundaries work right */
8544 i = PerlIO_getc(fp);
8548 PerlIO_ungetc(fp,i);
8554 /* See if we know enough about I/O mechanism to cheat it ! */
8556 /* This used to be #ifdef test - it is made run-time test for ease
8557 of abstracting out stdio interface. One call should be cheap
8558 enough here - and may even be a macro allowing compile
8562 if (PerlIO_fast_gets(fp)) {
8564 * We can do buffer based IO operations on this filehandle.
8566 * This means we can bypass a lot of subcalls and process
8567 * the buffer directly, it also means we know the upper bound
8568 * on the amount of data we might read of the current buffer
8569 * into our sv. Knowing this allows us to preallocate the pv
8570 * to be able to hold that maximum, which allows us to simplify
8571 * a lot of logic. */
8574 * We're going to steal some values from the stdio struct
8575 * and put EVERYTHING in the innermost loop into registers.
8577 STDCHAR *ptr; /* pointer into fp's read-ahead buffer */
8578 STRLEN bpx; /* length of the data in the target sv
8579 used to fix pointers after a SvGROW */
8580 I32 shortbuffered; /* If the pv buffer is shorter than the amount
8581 of data left in the read-ahead buffer.
8582 If 0 then the pv buffer can hold the full
8583 amount left, otherwise this is the amount it
8586 /* Here is some breathtakingly efficient cheating */
8588 /* When you read the following logic resist the urge to think
8589 * of record separators that are 1 byte long. They are an
8590 * uninteresting special (simple) case.
8592 * Instead think of record separators which are at least 2 bytes
8593 * long, and keep in mind that we need to deal with such
8594 * separators when they cross a read-ahead buffer boundary.
8596 * Also consider that we need to gracefully deal with separators
8597 * that may be longer than a single read ahead buffer.
8599 * Lastly do not forget we want to copy the delimiter as well. We
8600 * are copying all data in the file _up_to_and_including_ the separator
8603 * Now that you have all that in mind here is what is happening below:
8605 * 1. When we first enter the loop we do some memory book keeping to see
8606 * how much free space there is in the target SV. (This sub assumes that
8607 * it is operating on the same SV most of the time via $_ and that it is
8608 * going to be able to reuse the same pv buffer each call.) If there is
8609 * "enough" room then we set "shortbuffered" to how much space there is
8610 * and start reading forward.
8612 * 2. When we scan forward we copy from the read-ahead buffer to the target
8613 * SV's pv buffer. While we go we watch for the end of the read-ahead buffer,
8614 * and the end of the of pv, as well as for the "rslast", which is the last
8615 * char of the separator.
8617 * 3. When scanning forward if we see rslast then we jump backwards in *pv*
8618 * (which has a "complete" record up to the point we saw rslast) and check
8619 * it to see if it matches the separator. If it does we are done. If it doesn't
8620 * we continue on with the scan/copy.
8622 * 4. If we run out of read-ahead buffer (cnt goes to 0) then we have to get
8623 * the IO system to read the next buffer. We do this by doing a getc(), which
8624 * returns a single char read (or EOF), and prefills the buffer, and also
8625 * allows us to find out how full the buffer is. We use this information to
8626 * SvGROW() the sv to the size remaining in the buffer, after which we copy
8627 * the returned single char into the target sv, and then go back into scan
8630 * 5. If we run out of write-buffer then we SvGROW() it by the size of the
8631 * remaining space in the read-buffer.
8633 * Note that this code despite its twisty-turny nature is pretty darn slick.
8634 * It manages single byte separators, multi-byte cross boundary separators,
8635 * and cross-read-buffer separators cleanly and efficiently at the cost
8636 * of potentially greatly overallocating the target SV.
8642 /* get the number of bytes remaining in the read-ahead buffer
8643 * on first call on a given fp this will return 0.*/
8644 cnt = PerlIO_get_cnt(fp);
8646 /* make sure we have the room */
8647 if ((I32)(SvLEN(sv) - append) <= cnt + 1) {
8648 /* Not room for all of it
8649 if we are looking for a separator and room for some
8651 if (rslen && cnt > 80 && (I32)SvLEN(sv) > append) {
8652 /* just process what we have room for */
8653 shortbuffered = cnt - SvLEN(sv) + append + 1;
8654 cnt -= shortbuffered;
8657 /* ensure that the target sv has enough room to hold
8658 * the rest of the read-ahead buffer */
8660 /* remember that cnt can be negative */
8661 SvGROW(sv, (STRLEN)(append + (cnt <= 0 ? 2 : (cnt + 1))));
8665 /* we have enough room to hold the full buffer, lets scream */
8669 /* extract the pointer to sv's string buffer, offset by append as necessary */
8670 bp = (STDCHAR*)SvPVX_const(sv) + append; /* move these two too to registers */
8671 /* extract the point to the read-ahead buffer */
8672 ptr = (STDCHAR*)PerlIO_get_ptr(fp);
8674 /* some trace debug output */
8675 DEBUG_P(PerlIO_printf(Perl_debug_log,
8676 "Screamer: entering, ptr=%" UVuf ", cnt=%ld\n",PTR2UV(ptr),(long)cnt));
8677 DEBUG_P(PerlIO_printf(Perl_debug_log,
8678 "Screamer: entering: PerlIO * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%"
8680 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8681 PTR2UV(PerlIO_has_base(fp) ? PerlIO_get_base(fp) : 0)));
8685 /* if there is stuff left in the read-ahead buffer */
8687 /* if there is a separator */
8689 /* find next rslast */
8692 /* shortcut common case of blank line */
8694 if ((*bp++ = *ptr++) == rslast)
8695 goto thats_all_folks;
8697 p = (STDCHAR *)memchr(ptr, rslast, cnt);
8699 SSize_t got = p - ptr + 1;
8700 Copy(ptr, bp, got, STDCHAR);
8704 goto thats_all_folks;
8706 Copy(ptr, bp, cnt, STDCHAR);
8712 /* no separator, slurp the full buffer */
8713 Copy(ptr, bp, cnt, char); /* this | eat */
8714 bp += cnt; /* screams | dust */
8715 ptr += cnt; /* louder | sed :-) */
8717 assert (!shortbuffered);
8718 goto cannot_be_shortbuffered;
8722 if (shortbuffered) { /* oh well, must extend */
8723 /* we didnt have enough room to fit the line into the target buffer
8724 * so we must extend the target buffer and keep going */
8725 cnt = shortbuffered;
8727 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8729 /* extned the target sv's buffer so it can hold the full read-ahead buffer */
8730 SvGROW(sv, SvLEN(sv) + append + cnt + 2);
8731 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8735 cannot_be_shortbuffered:
8736 /* we need to refill the read-ahead buffer if possible */
8738 DEBUG_P(PerlIO_printf(Perl_debug_log,
8739 "Screamer: going to getc, ptr=%" UVuf ", cnt=%" IVdf "\n",
8740 PTR2UV(ptr),(IV)cnt));
8741 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* deregisterize cnt and ptr */
8743 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8744 "Screamer: pre: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf "\n",
8745 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8746 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8749 call PerlIO_getc() to let it prefill the lookahead buffer
8751 This used to call 'filbuf' in stdio form, but as that behaves like
8752 getc when cnt <= 0 we use PerlIO_getc here to avoid introducing
8753 another abstraction.
8755 Note we have to deal with the char in 'i' if we are not at EOF
8757 i = PerlIO_getc(fp); /* get more characters */
8759 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8760 "Screamer: post: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf "\n",
8761 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8762 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8764 /* find out how much is left in the read-ahead buffer, and rextract its pointer */
8765 cnt = PerlIO_get_cnt(fp);
8766 ptr = (STDCHAR*)PerlIO_get_ptr(fp); /* reregisterize cnt and ptr */
8767 DEBUG_P(PerlIO_printf(Perl_debug_log,
8768 "Screamer: after getc, ptr=%" UVuf ", cnt=%" IVdf "\n",
8769 PTR2UV(ptr),(IV)cnt));
8771 if (i == EOF) /* all done for ever? */
8772 goto thats_really_all_folks;
8774 /* make sure we have enough space in the target sv */
8775 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8777 SvGROW(sv, bpx + cnt + 2);
8778 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8780 /* copy of the char we got from getc() */
8781 *bp++ = (STDCHAR)i; /* store character from PerlIO_getc */
8783 /* make sure we deal with the i being the last character of a separator */
8784 if (rslen && (STDCHAR)i == rslast) /* all done for now? */
8785 goto thats_all_folks;
8789 /* check if we have actually found the separator - only really applies
8791 if ((rslen > 1 && (STRLEN)(bp - (STDCHAR*)SvPVX_const(sv)) < rslen) ||
8792 memNE((char*)bp - rslen, rsptr, rslen))
8793 goto screamer; /* go back to the fray */
8794 thats_really_all_folks:
8796 cnt += shortbuffered;
8797 DEBUG_P(PerlIO_printf(Perl_debug_log,
8798 "Screamer: quitting, ptr=%" UVuf ", cnt=%" IVdf "\n",PTR2UV(ptr),(IV)cnt));
8799 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* put these back or we're in trouble */
8800 DEBUG_P(PerlIO_printf(Perl_debug_log,
8801 "Screamer: end: FILE * thinks ptr=%" UVuf ", cnt=%" IVdf ", base=%" UVuf
8803 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8804 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8806 SvCUR_set(sv, bp - (STDCHAR*)SvPVX_const(sv)); /* set length */
8807 DEBUG_P(PerlIO_printf(Perl_debug_log,
8808 "Screamer: done, len=%ld, string=|%.*s|\n",
8809 (long)SvCUR(sv),(int)SvCUR(sv),SvPVX_const(sv)));
8813 /*The big, slow, and stupid way. */
8814 #ifdef USE_HEAP_INSTEAD_OF_STACK /* Even slower way. */
8815 STDCHAR *buf = NULL;
8816 Newx(buf, 8192, STDCHAR);
8824 const STDCHAR * const bpe = buf + sizeof(buf);
8826 while ((i = PerlIO_getc(fp)) != EOF && (*bp++ = (STDCHAR)i) != rslast && bp < bpe)
8827 ; /* keep reading */
8831 cnt = PerlIO_read(fp,(char*)buf, sizeof(buf));
8832 /* Accommodate broken VAXC compiler, which applies U8 cast to
8833 * both args of ?: operator, causing EOF to change into 255
8836 i = (U8)buf[cnt - 1];
8842 cnt = 0; /* we do need to re-set the sv even when cnt <= 0 */
8844 sv_catpvn_nomg(sv, (char *) buf, cnt);
8846 sv_setpvn(sv, (char *) buf, cnt); /* "nomg" is implied */
8848 if (i != EOF && /* joy */
8850 SvCUR(sv) < rslen ||
8851 memNE(SvPVX_const(sv) + SvCUR(sv) - rslen, rsptr, rslen)))
8855 * If we're reading from a TTY and we get a short read,
8856 * indicating that the user hit his EOF character, we need
8857 * to notice it now, because if we try to read from the TTY
8858 * again, the EOF condition will disappear.
8860 * The comparison of cnt to sizeof(buf) is an optimization
8861 * that prevents unnecessary calls to feof().
8865 if (!(cnt < (I32)sizeof(buf) && PerlIO_eof(fp)))
8869 #ifdef USE_HEAP_INSTEAD_OF_STACK
8874 if (rspara) { /* have to do this both before and after */
8875 while (i != EOF) { /* to make sure file boundaries work right */
8876 i = PerlIO_getc(fp);
8878 PerlIO_ungetc(fp,i);
8884 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8890 Auto-increment of the value in the SV, doing string to numeric conversion
8891 if necessary. Handles 'get' magic and operator overloading.
8897 Perl_sv_inc(pTHX_ SV *const sv)
8906 =for apidoc sv_inc_nomg
8908 Auto-increment of the value in the SV, doing string to numeric conversion
8909 if necessary. Handles operator overloading. Skips handling 'get' magic.
8915 Perl_sv_inc_nomg(pTHX_ SV *const sv)
8922 if (SvTHINKFIRST(sv)) {
8923 if (SvREADONLY(sv)) {
8924 Perl_croak_no_modify();
8928 if (SvAMAGIC(sv) && AMG_CALLunary(sv, inc_amg))
8930 i = PTR2IV(SvRV(sv));
8934 else sv_force_normal_flags(sv, 0);
8936 flags = SvFLAGS(sv);
8937 if ((flags & (SVp_NOK|SVp_IOK)) == SVp_NOK) {
8938 /* It's (privately or publicly) a float, but not tested as an
8939 integer, so test it to see. */
8941 flags = SvFLAGS(sv);
8943 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8944 /* It's publicly an integer, or privately an integer-not-float */
8945 #ifdef PERL_PRESERVE_IVUV
8949 if (SvUVX(sv) == UV_MAX)
8950 sv_setnv(sv, UV_MAX_P1);
8952 (void)SvIOK_only_UV(sv);
8953 SvUV_set(sv, SvUVX(sv) + 1);
8955 if (SvIVX(sv) == IV_MAX)
8956 sv_setuv(sv, (UV)IV_MAX + 1);
8958 (void)SvIOK_only(sv);
8959 SvIV_set(sv, SvIVX(sv) + 1);
8964 if (flags & SVp_NOK) {
8965 const NV was = SvNVX(sv);
8966 if (LIKELY(!Perl_isinfnan(was)) &&
8967 NV_OVERFLOWS_INTEGERS_AT != 0.0 &&
8968 was >= NV_OVERFLOWS_INTEGERS_AT) {
8969 /* diag_listed_as: Lost precision when %s %f by 1 */
8970 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8971 "Lost precision when incrementing %" NVff " by 1",
8974 (void)SvNOK_only(sv);
8975 SvNV_set(sv, was + 1.0);
8979 /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */
8980 if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv)))
8981 Perl_croak_no_modify();
8983 if (!(flags & SVp_POK) || !*SvPVX_const(sv)) {
8984 if ((flags & SVTYPEMASK) < SVt_PVIV)
8985 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV ? SVt_PVIV : SVt_IV));
8986 (void)SvIOK_only(sv);
8991 while (isALPHA(*d)) d++;
8992 while (isDIGIT(*d)) d++;
8993 if (d < SvEND(sv)) {
8994 const int numtype = grok_number_flags(SvPVX_const(sv), SvCUR(sv), NULL, PERL_SCAN_TRAILING);
8995 #ifdef PERL_PRESERVE_IVUV
8996 /* Got to punt this as an integer if needs be, but we don't issue
8997 warnings. Probably ought to make the sv_iv_please() that does
8998 the conversion if possible, and silently. */
8999 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
9000 /* Need to try really hard to see if it's an integer.
9001 9.22337203685478e+18 is an integer.
9002 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
9003 so $a="9.22337203685478e+18"; $a+0; $a++
9004 needs to be the same as $a="9.22337203685478e+18"; $a++
9011 /* sv_2iv *should* have made this an NV */
9012 if (flags & SVp_NOK) {
9013 (void)SvNOK_only(sv);
9014 SvNV_set(sv, SvNVX(sv) + 1.0);
9017 /* I don't think we can get here. Maybe I should assert this
9018 And if we do get here I suspect that sv_setnv will croak. NWC
9020 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_inc punt failed to convert '%s' to IOK or NOKp, UV=0x%" UVxf " NV=%" NVgf "\n",
9021 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
9023 #endif /* PERL_PRESERVE_IVUV */
9024 if (!numtype && ckWARN(WARN_NUMERIC))
9025 not_incrementable(sv);
9026 sv_setnv(sv,Atof(SvPVX_const(sv)) + 1.0);
9030 while (d >= SvPVX_const(sv)) {
9038 /* MKS: The original code here died if letters weren't consecutive.
9039 * at least it didn't have to worry about non-C locales. The
9040 * new code assumes that ('z'-'a')==('Z'-'A'), letters are
9041 * arranged in order (although not consecutively) and that only
9042 * [A-Za-z] are accepted by isALPHA in the C locale.
9044 if (isALPHA_FOLD_NE(*d, 'z')) {
9045 do { ++*d; } while (!isALPHA(*d));
9048 *(d--) -= 'z' - 'a';
9053 *(d--) -= 'z' - 'a' + 1;
9057 /* oh,oh, the number grew */
9058 SvGROW(sv, SvCUR(sv) + 2);
9059 SvCUR_set(sv, SvCUR(sv) + 1);
9060 for (d = SvPVX(sv) + SvCUR(sv); d > SvPVX_const(sv); d--)
9071 Auto-decrement of the value in the SV, doing string to numeric conversion
9072 if necessary. Handles 'get' magic and operator overloading.
9078 Perl_sv_dec(pTHX_ SV *const sv)
9087 =for apidoc sv_dec_nomg
9089 Auto-decrement of the value in the SV, doing string to numeric conversion
9090 if necessary. Handles operator overloading. Skips handling 'get' magic.
9096 Perl_sv_dec_nomg(pTHX_ SV *const sv)
9102 if (SvTHINKFIRST(sv)) {
9103 if (SvREADONLY(sv)) {
9104 Perl_croak_no_modify();
9108 if (SvAMAGIC(sv) && AMG_CALLunary(sv, dec_amg))
9110 i = PTR2IV(SvRV(sv));
9114 else sv_force_normal_flags(sv, 0);
9116 /* Unlike sv_inc we don't have to worry about string-never-numbers
9117 and keeping them magic. But we mustn't warn on punting */
9118 flags = SvFLAGS(sv);
9119 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
9120 /* It's publicly an integer, or privately an integer-not-float */
9121 #ifdef PERL_PRESERVE_IVUV
9125 if (SvUVX(sv) == 0) {
9126 (void)SvIOK_only(sv);
9130 (void)SvIOK_only_UV(sv);
9131 SvUV_set(sv, SvUVX(sv) - 1);
9134 if (SvIVX(sv) == IV_MIN) {
9135 sv_setnv(sv, (NV)IV_MIN);
9139 (void)SvIOK_only(sv);
9140 SvIV_set(sv, SvIVX(sv) - 1);
9145 if (flags & SVp_NOK) {
9148 const NV was = SvNVX(sv);
9149 if (LIKELY(!Perl_isinfnan(was)) &&
9150 NV_OVERFLOWS_INTEGERS_AT != 0.0 &&
9151 was <= -NV_OVERFLOWS_INTEGERS_AT) {
9152 /* diag_listed_as: Lost precision when %s %f by 1 */
9153 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
9154 "Lost precision when decrementing %" NVff " by 1",
9157 (void)SvNOK_only(sv);
9158 SvNV_set(sv, was - 1.0);
9163 /* treat AV/HV/CV/FM/IO and non-fake GVs as immutable */
9164 if (SvTYPE(sv) >= SVt_PVAV || (isGV_with_GP(sv) && !SvFAKE(sv)))
9165 Perl_croak_no_modify();
9167 if (!(flags & SVp_POK)) {
9168 if ((flags & SVTYPEMASK) < SVt_PVIV)
9169 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV) ? SVt_PVIV : SVt_IV);
9171 (void)SvIOK_only(sv);
9174 #ifdef PERL_PRESERVE_IVUV
9176 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), NULL);
9177 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
9178 /* Need to try really hard to see if it's an integer.
9179 9.22337203685478e+18 is an integer.
9180 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
9181 so $a="9.22337203685478e+18"; $a+0; $a--
9182 needs to be the same as $a="9.22337203685478e+18"; $a--
9189 /* sv_2iv *should* have made this an NV */
9190 if (flags & SVp_NOK) {
9191 (void)SvNOK_only(sv);
9192 SvNV_set(sv, SvNVX(sv) - 1.0);
9195 /* I don't think we can get here. Maybe I should assert this
9196 And if we do get here I suspect that sv_setnv will croak. NWC
9198 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_dec punt failed to convert '%s' to IOK or NOKp, UV=0x%" UVxf " NV=%" NVgf "\n",
9199 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
9202 #endif /* PERL_PRESERVE_IVUV */
9203 sv_setnv(sv,Atof(SvPVX_const(sv)) - 1.0); /* punt */
9206 /* this define is used to eliminate a chunk of duplicated but shared logic
9207 * it has the suffix __SV_C to signal that it isnt API, and isnt meant to be
9208 * used anywhere but here - yves
9210 #define PUSH_EXTEND_MORTAL__SV_C(AnSv) \
9212 SSize_t ix = ++PL_tmps_ix; \
9213 if (UNLIKELY(ix >= PL_tmps_max)) \
9214 ix = tmps_grow_p(ix); \
9215 PL_tmps_stack[ix] = (AnSv); \
9219 =for apidoc sv_mortalcopy
9221 Creates a new SV which is a copy of the original SV (using C<sv_setsv>).
9222 The new SV is marked as mortal. It will be destroyed "soon", either by an
9223 explicit call to C<FREETMPS>, or by an implicit call at places such as
9224 statement boundaries. See also C<L</sv_newmortal>> and C<L</sv_2mortal>>.
9229 /* Make a string that will exist for the duration of the expression
9230 * evaluation. Actually, it may have to last longer than that, but
9231 * hopefully we won't free it until it has been assigned to a
9232 * permanent location. */
9235 Perl_sv_mortalcopy_flags(pTHX_ SV *const oldstr, U32 flags)
9239 if (flags & SV_GMAGIC)
9240 SvGETMAGIC(oldstr); /* before new_SV, in case it dies */
9242 sv_setsv_flags(sv,oldstr,flags & ~SV_GMAGIC);
9243 PUSH_EXTEND_MORTAL__SV_C(sv);
9249 =for apidoc sv_newmortal
9251 Creates a new null SV which is mortal. The reference count of the SV is
9252 set to 1. It will be destroyed "soon", either by an explicit call to
9253 C<FREETMPS>, or by an implicit call at places such as statement boundaries.
9254 See also C<L</sv_mortalcopy>> and C<L</sv_2mortal>>.
9260 Perl_sv_newmortal(pTHX)
9265 SvFLAGS(sv) = SVs_TEMP;
9266 PUSH_EXTEND_MORTAL__SV_C(sv);
9272 =for apidoc newSVpvn_flags
9274 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9275 characters) into it. The reference count for the
9276 SV is set to 1. Note that if C<len> is zero, Perl will create a zero length
9277 string. You are responsible for ensuring that the source string is at least
9278 C<len> bytes long. If the C<s> argument is NULL the new SV will be undefined.
9279 Currently the only flag bits accepted are C<SVf_UTF8> and C<SVs_TEMP>.
9280 If C<SVs_TEMP> is set, then C<sv_2mortal()> is called on the result before
9281 returning. If C<SVf_UTF8> is set, C<s>
9282 is considered to be in UTF-8 and the
9283 C<SVf_UTF8> flag will be set on the new SV.
9284 C<newSVpvn_utf8()> is a convenience wrapper for this function, defined as
9286 #define newSVpvn_utf8(s, len, u) \
9287 newSVpvn_flags((s), (len), (u) ? SVf_UTF8 : 0)
9293 Perl_newSVpvn_flags(pTHX_ const char *const s, const STRLEN len, const U32 flags)
9297 /* All the flags we don't support must be zero.
9298 And we're new code so I'm going to assert this from the start. */
9299 assert(!(flags & ~(SVf_UTF8|SVs_TEMP)));
9301 sv_setpvn(sv,s,len);
9303 /* This code used to do a sv_2mortal(), however we now unroll the call to
9304 * sv_2mortal() and do what it does ourselves here. Since we have asserted
9305 * that flags can only have the SVf_UTF8 and/or SVs_TEMP flags set above we
9306 * can use it to enable the sv flags directly (bypassing SvTEMP_on), which
9307 * in turn means we dont need to mask out the SVf_UTF8 flag below, which
9308 * means that we eliminate quite a few steps than it looks - Yves
9309 * (explaining patch by gfx) */
9311 SvFLAGS(sv) |= flags;
9313 if(flags & SVs_TEMP){
9314 PUSH_EXTEND_MORTAL__SV_C(sv);
9321 =for apidoc sv_2mortal
9323 Marks an existing SV as mortal. The SV will be destroyed "soon", either
9324 by an explicit call to C<FREETMPS>, or by an implicit call at places such as
9325 statement boundaries. C<SvTEMP()> is turned on which means that the SV's
9326 string buffer can be "stolen" if this SV is copied. See also
9327 C<L</sv_newmortal>> and C<L</sv_mortalcopy>>.
9333 Perl_sv_2mortal(pTHX_ SV *const sv)
9340 PUSH_EXTEND_MORTAL__SV_C(sv);
9348 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
9349 characters) into it. The reference count for the
9350 SV is set to 1. If C<len> is zero, Perl will compute the length using
9351 C<strlen()>, (which means if you use this option, that C<s> can't have embedded
9352 C<NUL> characters and has to have a terminating C<NUL> byte).
9354 This function can cause reliability issues if you are likely to pass in
9355 empty strings that are not null terminated, because it will run
9356 strlen on the string and potentially run past valid memory.
9358 Using L</newSVpvn> is a safer alternative for non C<NUL> terminated strings.
9359 For string literals use L</newSVpvs> instead. This function will work fine for
9360 C<NUL> terminated strings, but if you want to avoid the if statement on whether
9361 to call C<strlen> use C<newSVpvn> instead (calling C<strlen> yourself).
9367 Perl_newSVpv(pTHX_ const char *const s, const STRLEN len)
9372 sv_setpvn(sv, s, len || s == NULL ? len : strlen(s));
9377 =for apidoc newSVpvn
9379 Creates a new SV and copies a string into it, which may contain C<NUL> characters
9380 (C<\0>) and other binary data. The reference count for the SV is set to 1.
9381 Note that if C<len> is zero, Perl will create a zero length (Perl) string. You
9382 are responsible for ensuring that the source buffer is at least
9383 C<len> bytes long. If the C<buffer> argument is NULL the new SV will be
9390 Perl_newSVpvn(pTHX_ const char *const buffer, const STRLEN len)
9394 sv_setpvn(sv,buffer,len);
9399 =for apidoc newSVhek
9401 Creates a new SV from the hash key structure. It will generate scalars that
9402 point to the shared string table where possible. Returns a new (undefined)
9403 SV if C<hek> is NULL.
9409 Perl_newSVhek(pTHX_ const HEK *const hek)
9418 if (HEK_LEN(hek) == HEf_SVKEY) {
9419 return newSVsv(*(SV**)HEK_KEY(hek));
9421 const int flags = HEK_FLAGS(hek);
9422 if (flags & HVhek_WASUTF8) {
9424 Andreas would like keys he put in as utf8 to come back as utf8
9426 STRLEN utf8_len = HEK_LEN(hek);
9427 SV * const sv = newSV_type(SVt_PV);
9428 char *as_utf8 = (char *)bytes_to_utf8 ((U8*)HEK_KEY(hek), &utf8_len);
9429 /* bytes_to_utf8() allocates a new string, which we can repurpose: */
9430 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
9433 } else if (flags & HVhek_UNSHARED) {
9434 /* A hash that isn't using shared hash keys has to have
9435 the flag in every key so that we know not to try to call
9436 share_hek_hek on it. */
9438 SV * const sv = newSVpvn (HEK_KEY(hek), HEK_LEN(hek));
9443 /* This will be overwhelminly the most common case. */
9445 /* Inline most of newSVpvn_share(), because share_hek_hek() is far
9446 more efficient than sharepvn(). */
9450 sv_upgrade(sv, SVt_PV);
9451 SvPV_set(sv, (char *)HEK_KEY(share_hek_hek(hek)));
9452 SvCUR_set(sv, HEK_LEN(hek));
9464 =for apidoc newSVpvn_share
9466 Creates a new SV with its C<SvPVX_const> pointing to a shared string in the string
9467 table. If the string does not already exist in the table, it is
9468 created first. Turns on the C<SvIsCOW> flag (or C<READONLY>
9469 and C<FAKE> in 5.16 and earlier). If the C<hash> parameter
9470 is non-zero, that value is used; otherwise the hash is computed.
9471 The string's hash can later be retrieved from the SV
9472 with the C<SvSHARED_HASH()> macro. The idea here is
9473 that as the string table is used for shared hash keys these strings will have
9474 C<SvPVX_const == HeKEY> and hash lookup will avoid string compare.
9480 Perl_newSVpvn_share(pTHX_ const char *src, I32 len, U32 hash)
9484 bool is_utf8 = FALSE;
9485 const char *const orig_src = src;
9488 STRLEN tmplen = -len;
9490 /* See the note in hv.c:hv_fetch() --jhi */
9491 src = (char*)bytes_from_utf8((const U8*)src, &tmplen, &is_utf8);
9495 PERL_HASH(hash, src, len);
9497 /* The logic for this is inlined in S_mro_get_linear_isa_dfs(), so if it
9498 changes here, update it there too. */
9499 sv_upgrade(sv, SVt_PV);
9500 SvPV_set(sv, sharepvn(src, is_utf8?-len:len, hash));
9507 if (src != orig_src)
9513 =for apidoc newSVpv_share
9515 Like C<newSVpvn_share>, but takes a C<NUL>-terminated string instead of a
9522 Perl_newSVpv_share(pTHX_ const char *src, U32 hash)
9524 return newSVpvn_share(src, strlen(src), hash);
9527 #if defined(PERL_IMPLICIT_CONTEXT)
9529 /* pTHX_ magic can't cope with varargs, so this is a no-context
9530 * version of the main function, (which may itself be aliased to us).
9531 * Don't access this version directly.
9535 Perl_newSVpvf_nocontext(const char *const pat, ...)
9541 PERL_ARGS_ASSERT_NEWSVPVF_NOCONTEXT;
9543 va_start(args, pat);
9544 sv = vnewSVpvf(pat, &args);
9551 =for apidoc newSVpvf
9553 Creates a new SV and initializes it with the string formatted like
9560 Perl_newSVpvf(pTHX_ const char *const pat, ...)
9565 PERL_ARGS_ASSERT_NEWSVPVF;
9567 va_start(args, pat);
9568 sv = vnewSVpvf(pat, &args);
9573 /* backend for newSVpvf() and newSVpvf_nocontext() */
9576 Perl_vnewSVpvf(pTHX_ const char *const pat, va_list *const args)
9580 PERL_ARGS_ASSERT_VNEWSVPVF;
9583 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
9590 Creates a new SV and copies a floating point value into it.
9591 The reference count for the SV is set to 1.
9597 Perl_newSVnv(pTHX_ const NV n)
9609 Creates a new SV and copies an integer into it. The reference count for the
9616 Perl_newSViv(pTHX_ const IV i)
9622 /* Inlining ONLY the small relevant subset of sv_setiv here
9623 * for performance. Makes a significant difference. */
9625 /* We're starting from SVt_FIRST, so provided that's
9626 * actual 0, we don't have to unset any SV type flags
9627 * to promote to SVt_IV. */
9628 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9630 SET_SVANY_FOR_BODYLESS_IV(sv);
9631 SvFLAGS(sv) |= SVt_IV;
9643 Creates a new SV and copies an unsigned integer into it.
9644 The reference count for the SV is set to 1.
9650 Perl_newSVuv(pTHX_ const UV u)
9654 /* Inlining ONLY the small relevant subset of sv_setuv here
9655 * for performance. Makes a significant difference. */
9657 /* Using ivs is more efficient than using uvs - see sv_setuv */
9658 if (u <= (UV)IV_MAX) {
9659 return newSViv((IV)u);
9664 /* We're starting from SVt_FIRST, so provided that's
9665 * actual 0, we don't have to unset any SV type flags
9666 * to promote to SVt_IV. */
9667 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9669 SET_SVANY_FOR_BODYLESS_IV(sv);
9670 SvFLAGS(sv) |= SVt_IV;
9672 (void)SvIsUV_on(sv);
9681 =for apidoc newSV_type
9683 Creates a new SV, of the type specified. The reference count for the new SV
9690 Perl_newSV_type(pTHX_ const svtype type)
9695 ASSUME(SvTYPE(sv) == SVt_FIRST);
9696 if(type != SVt_FIRST)
9697 sv_upgrade(sv, type);
9702 =for apidoc newRV_noinc
9704 Creates an RV wrapper for an SV. The reference count for the original
9705 SV is B<not> incremented.
9711 Perl_newRV_noinc(pTHX_ SV *const tmpRef)
9715 PERL_ARGS_ASSERT_NEWRV_NOINC;
9719 /* We're starting from SVt_FIRST, so provided that's
9720 * actual 0, we don't have to unset any SV type flags
9721 * to promote to SVt_IV. */
9722 STATIC_ASSERT_STMT(SVt_FIRST == 0);
9724 SET_SVANY_FOR_BODYLESS_IV(sv);
9725 SvFLAGS(sv) |= SVt_IV;
9730 SvRV_set(sv, tmpRef);
9735 /* newRV_inc is the official function name to use now.
9736 * newRV_inc is in fact #defined to newRV in sv.h
9740 Perl_newRV(pTHX_ SV *const sv)
9742 PERL_ARGS_ASSERT_NEWRV;
9744 return newRV_noinc(SvREFCNT_inc_simple_NN(sv));
9750 Creates a new SV which is an exact duplicate of the original SV.
9757 Perl_newSVsv(pTHX_ SV *const old)
9763 if (SvTYPE(old) == (svtype)SVTYPEMASK) {
9764 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), "semi-panic: attempt to dup freed string");
9767 /* Do this here, otherwise we leak the new SV if this croaks. */
9770 /* SV_NOSTEAL prevents TEMP buffers being, well, stolen, and saves games
9771 with SvTEMP_off and SvTEMP_on round a call to sv_setsv. */
9772 sv_setsv_flags(sv, old, SV_NOSTEAL);
9777 =for apidoc sv_reset
9779 Underlying implementation for the C<reset> Perl function.
9780 Note that the perl-level function is vaguely deprecated.
9786 Perl_sv_reset(pTHX_ const char *s, HV *const stash)
9788 PERL_ARGS_ASSERT_SV_RESET;
9790 sv_resetpvn(*s ? s : NULL, strlen(s), stash);
9794 Perl_sv_resetpvn(pTHX_ const char *s, STRLEN len, HV * const stash)
9796 char todo[PERL_UCHAR_MAX+1];
9799 if (!stash || SvTYPE(stash) != SVt_PVHV)
9802 if (!s) { /* reset ?? searches */
9803 MAGIC * const mg = mg_find((const SV *)stash, PERL_MAGIC_symtab);
9805 const U32 count = mg->mg_len / sizeof(PMOP**);
9806 PMOP **pmp = (PMOP**) mg->mg_ptr;
9807 PMOP *const *const end = pmp + count;
9811 SvREADONLY_off(PL_regex_pad[(*pmp)->op_pmoffset]);
9813 (*pmp)->op_pmflags &= ~PMf_USED;
9821 /* reset variables */
9823 if (!HvARRAY(stash))
9826 Zero(todo, 256, char);
9830 I32 i = (unsigned char)*s;
9834 max = (unsigned char)*s++;
9835 for ( ; i <= max; i++) {
9838 for (i = 0; i <= (I32) HvMAX(stash); i++) {
9840 for (entry = HvARRAY(stash)[i];
9842 entry = HeNEXT(entry))
9847 if (!todo[(U8)*HeKEY(entry)])
9849 gv = MUTABLE_GV(HeVAL(entry));
9853 if (sv && !SvREADONLY(sv)) {
9854 SV_CHECK_THINKFIRST_COW_DROP(sv);
9855 if (!isGV(sv)) SvOK_off(sv);
9860 if (GvHV(gv) && !HvNAME_get(GvHV(gv))) {
9871 Using various gambits, try to get an IO from an SV: the IO slot if its a
9872 GV; or the recursive result if we're an RV; or the IO slot of the symbol
9873 named after the PV if we're a string.
9875 'Get' magic is ignored on the C<sv> passed in, but will be called on
9876 C<SvRV(sv)> if C<sv> is an RV.
9882 Perl_sv_2io(pTHX_ SV *const sv)
9887 PERL_ARGS_ASSERT_SV_2IO;
9889 switch (SvTYPE(sv)) {
9891 io = MUTABLE_IO(sv);
9895 if (isGV_with_GP(sv)) {
9896 gv = MUTABLE_GV(sv);
9899 Perl_croak(aTHX_ "Bad filehandle: %" HEKf,
9900 HEKfARG(GvNAME_HEK(gv)));
9906 Perl_croak(aTHX_ PL_no_usym, "filehandle");
9908 SvGETMAGIC(SvRV(sv));
9909 return sv_2io(SvRV(sv));
9911 gv = gv_fetchsv_nomg(sv, 0, SVt_PVIO);
9918 if (SvGMAGICAL(sv)) {
9919 newsv = sv_newmortal();
9920 sv_setsv_nomg(newsv, sv);
9922 Perl_croak(aTHX_ "Bad filehandle: %" SVf, SVfARG(newsv));
9932 Using various gambits, try to get a CV from an SV; in addition, try if
9933 possible to set C<*st> and C<*gvp> to the stash and GV associated with it.
9934 The flags in C<lref> are passed to C<gv_fetchsv>.
9940 Perl_sv_2cv(pTHX_ SV *sv, HV **const st, GV **const gvp, const I32 lref)
9945 PERL_ARGS_ASSERT_SV_2CV;
9952 switch (SvTYPE(sv)) {
9956 return MUTABLE_CV(sv);
9966 sv = amagic_deref_call(sv, to_cv_amg);
9969 if (SvTYPE(sv) == SVt_PVCV) {
9970 cv = MUTABLE_CV(sv);
9975 else if(SvGETMAGIC(sv), isGV_with_GP(sv))
9976 gv = MUTABLE_GV(sv);
9978 Perl_croak(aTHX_ "Not a subroutine reference");
9980 else if (isGV_with_GP(sv)) {
9981 gv = MUTABLE_GV(sv);
9984 gv = gv_fetchsv_nomg(sv, lref, SVt_PVCV);
9991 /* Some flags to gv_fetchsv mean don't really create the GV */
9992 if (!isGV_with_GP(gv)) {
9997 if (lref & ~GV_ADDMG && !GvCVu(gv)) {
9998 /* XXX this is probably not what they think they're getting.
9999 * It has the same effect as "sub name;", i.e. just a forward
10008 =for apidoc sv_true
10010 Returns true if the SV has a true value by Perl's rules.
10011 Use the C<SvTRUE> macro instead, which may call C<sv_true()> or may
10012 instead use an in-line version.
10018 Perl_sv_true(pTHX_ SV *const sv)
10023 const XPV* const tXpv = (XPV*)SvANY(sv);
10025 (tXpv->xpv_cur > 1 ||
10026 (tXpv->xpv_cur && *sv->sv_u.svu_pv != '0')))
10033 return SvIVX(sv) != 0;
10036 return SvNVX(sv) != 0.0;
10038 return sv_2bool(sv);
10044 =for apidoc sv_pvn_force
10046 Get a sensible string out of the SV somehow.
10047 A private implementation of the C<SvPV_force> macro for compilers which
10048 can't cope with complex macro expressions. Always use the macro instead.
10050 =for apidoc sv_pvn_force_flags
10052 Get a sensible string out of the SV somehow.
10053 If C<flags> has the C<SV_GMAGIC> bit set, will C<mg_get> on C<sv> if
10054 appropriate, else not. C<sv_pvn_force> and C<sv_pvn_force_nomg> are
10055 implemented in terms of this function.
10056 You normally want to use the various wrapper macros instead: see
10057 C<L</SvPV_force>> and C<L</SvPV_force_nomg>>.
10063 Perl_sv_pvn_force_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
10065 PERL_ARGS_ASSERT_SV_PVN_FORCE_FLAGS;
10067 if (flags & SV_GMAGIC) SvGETMAGIC(sv);
10068 if (SvTHINKFIRST(sv) && (!SvROK(sv) || SvREADONLY(sv)))
10069 sv_force_normal_flags(sv, 0);
10079 if (SvTYPE(sv) > SVt_PVLV
10080 || isGV_with_GP(sv))
10081 /* diag_listed_as: Can't coerce %s to %s in %s */
10082 Perl_croak(aTHX_ "Can't coerce %s to string in %s", sv_reftype(sv,0),
10084 s = sv_2pv_flags(sv, &len, flags &~ SV_GMAGIC);
10091 if (SvTYPE(sv) < SVt_PV ||
10092 s != SvPVX_const(sv)) { /* Almost, but not quite, sv_setpvn() */
10095 SvUPGRADE(sv, SVt_PV); /* Never FALSE */
10096 SvGROW(sv, len + 1);
10097 Move(s,SvPVX(sv),len,char);
10098 SvCUR_set(sv, len);
10099 SvPVX(sv)[len] = '\0';
10102 SvPOK_on(sv); /* validate pointer */
10104 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%" UVxf " 2pv(%s)\n",
10105 PTR2UV(sv),SvPVX_const(sv)));
10108 (void)SvPOK_only_UTF8(sv);
10109 return SvPVX_mutable(sv);
10113 =for apidoc sv_pvbyten_force
10115 The backend for the C<SvPVbytex_force> macro. Always use the macro
10122 Perl_sv_pvbyten_force(pTHX_ SV *const sv, STRLEN *const lp)
10124 PERL_ARGS_ASSERT_SV_PVBYTEN_FORCE;
10126 sv_pvn_force(sv,lp);
10127 sv_utf8_downgrade(sv,0);
10133 =for apidoc sv_pvutf8n_force
10135 The backend for the C<SvPVutf8x_force> macro. Always use the macro
10142 Perl_sv_pvutf8n_force(pTHX_ SV *const sv, STRLEN *const lp)
10144 PERL_ARGS_ASSERT_SV_PVUTF8N_FORCE;
10146 sv_pvn_force(sv,0);
10147 sv_utf8_upgrade_nomg(sv);
10153 =for apidoc sv_reftype
10155 Returns a string describing what the SV is a reference to.
10157 If ob is true and the SV is blessed, the string is the class name,
10158 otherwise it is the type of the SV, "SCALAR", "ARRAY" etc.
10164 Perl_sv_reftype(pTHX_ const SV *const sv, const int ob)
10166 PERL_ARGS_ASSERT_SV_REFTYPE;
10167 if (ob && SvOBJECT(sv)) {
10168 return SvPV_nolen_const(sv_ref(NULL, sv, ob));
10171 /* WARNING - There is code, for instance in mg.c, that assumes that
10172 * the only reason that sv_reftype(sv,0) would return a string starting
10173 * with 'L' or 'S' is that it is a LVALUE or a SCALAR.
10174 * Yes this a dodgy way to do type checking, but it saves practically reimplementing
10175 * this routine inside other subs, and it saves time.
10176 * Do not change this assumption without searching for "dodgy type check" in
10179 switch (SvTYPE(sv)) {
10194 case SVt_PVLV: return (char *) (SvROK(sv) ? "REF"
10195 /* tied lvalues should appear to be
10196 * scalars for backwards compatibility */
10197 : (isALPHA_FOLD_EQ(LvTYPE(sv), 't'))
10198 ? "SCALAR" : "LVALUE");
10199 case SVt_PVAV: return "ARRAY";
10200 case SVt_PVHV: return "HASH";
10201 case SVt_PVCV: return "CODE";
10202 case SVt_PVGV: return (char *) (isGV_with_GP(sv)
10203 ? "GLOB" : "SCALAR");
10204 case SVt_PVFM: return "FORMAT";
10205 case SVt_PVIO: return "IO";
10206 case SVt_INVLIST: return "INVLIST";
10207 case SVt_REGEXP: return "REGEXP";
10208 default: return "UNKNOWN";
10216 Returns a SV describing what the SV passed in is a reference to.
10218 dst can be a SV to be set to the description or NULL, in which case a
10219 mortal SV is returned.
10221 If ob is true and the SV is blessed, the description is the class
10222 name, otherwise it is the type of the SV, "SCALAR", "ARRAY" etc.
10228 Perl_sv_ref(pTHX_ SV *dst, const SV *const sv, const int ob)
10230 PERL_ARGS_ASSERT_SV_REF;
10233 dst = sv_newmortal();
10235 if (ob && SvOBJECT(sv)) {
10236 HvNAME_get(SvSTASH(sv))
10237 ? sv_sethek(dst, HvNAME_HEK(SvSTASH(sv)))
10238 : sv_setpvs(dst, "__ANON__");
10241 const char * reftype = sv_reftype(sv, 0);
10242 sv_setpv(dst, reftype);
10248 =for apidoc sv_isobject
10250 Returns a boolean indicating whether the SV is an RV pointing to a blessed
10251 object. If the SV is not an RV, or if the object is not blessed, then this
10258 Perl_sv_isobject(pTHX_ SV *sv)
10274 Returns a boolean indicating whether the SV is blessed into the specified
10275 class. This does not check for subtypes; use C<sv_derived_from> to verify
10276 an inheritance relationship.
10282 Perl_sv_isa(pTHX_ SV *sv, const char *const name)
10284 const char *hvname;
10286 PERL_ARGS_ASSERT_SV_ISA;
10296 hvname = HvNAME_get(SvSTASH(sv));
10300 return strEQ(hvname, name);
10304 =for apidoc newSVrv
10306 Creates a new SV for the existing RV, C<rv>, to point to. If C<rv> is not an
10307 RV then it will be upgraded to one. If C<classname> is non-null then the new
10308 SV will be blessed in the specified package. The new SV is returned and its
10309 reference count is 1. The reference count 1 is owned by C<rv>.
10315 Perl_newSVrv(pTHX_ SV *const rv, const char *const classname)
10319 PERL_ARGS_ASSERT_NEWSVRV;
10323 SV_CHECK_THINKFIRST_COW_DROP(rv);
10325 if (UNLIKELY( SvTYPE(rv) >= SVt_PVMG )) {
10326 const U32 refcnt = SvREFCNT(rv);
10330 SvREFCNT(rv) = refcnt;
10332 sv_upgrade(rv, SVt_IV);
10333 } else if (SvROK(rv)) {
10334 SvREFCNT_dec(SvRV(rv));
10336 prepare_SV_for_RV(rv);
10344 HV* const stash = gv_stashpv(classname, GV_ADD);
10345 (void)sv_bless(rv, stash);
10351 Perl_newSVavdefelem(pTHX_ AV *av, SSize_t ix, bool extendible)
10353 SV * const lv = newSV_type(SVt_PVLV);
10354 PERL_ARGS_ASSERT_NEWSVAVDEFELEM;
10356 sv_magic(lv, NULL, PERL_MAGIC_defelem, NULL, 0);
10357 LvTARG(lv) = SvREFCNT_inc_simple_NN(av);
10358 LvSTARGOFF(lv) = ix;
10359 LvTARGLEN(lv) = extendible ? 1 : (STRLEN)UV_MAX;
10364 =for apidoc sv_setref_pv
10366 Copies a pointer 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. If the C<pv> argument is C<NULL>, then C<PL_sv_undef> will be placed
10369 into the SV. The C<classname> argument indicates the package for the
10370 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10371 will have a reference count of 1, and the RV will be returned.
10373 Do not use with other Perl types such as HV, AV, SV, CV, because those
10374 objects will become corrupted by the pointer copy process.
10376 Note that C<sv_setref_pvn> copies the string while this copies the pointer.
10382 Perl_sv_setref_pv(pTHX_ SV *const rv, const char *const classname, void *const pv)
10384 PERL_ARGS_ASSERT_SV_SETREF_PV;
10391 sv_setiv(newSVrv(rv,classname), PTR2IV(pv));
10396 =for apidoc sv_setref_iv
10398 Copies an integer into a new SV, optionally blessing the SV. The C<rv>
10399 argument will be upgraded to an RV. That RV will be modified to point to
10400 the new SV. The C<classname> argument indicates the package for the
10401 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10402 will have a reference count of 1, and the RV will be returned.
10408 Perl_sv_setref_iv(pTHX_ SV *const rv, const char *const classname, const IV iv)
10410 PERL_ARGS_ASSERT_SV_SETREF_IV;
10412 sv_setiv(newSVrv(rv,classname), iv);
10417 =for apidoc sv_setref_uv
10419 Copies an unsigned integer into a new SV, optionally blessing the SV. The C<rv>
10420 argument will be upgraded to an RV. That RV will be modified to point to
10421 the new SV. The C<classname> argument indicates the package for the
10422 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10423 will have a reference count of 1, and the RV will be returned.
10429 Perl_sv_setref_uv(pTHX_ SV *const rv, const char *const classname, const UV uv)
10431 PERL_ARGS_ASSERT_SV_SETREF_UV;
10433 sv_setuv(newSVrv(rv,classname), uv);
10438 =for apidoc sv_setref_nv
10440 Copies a double into a new SV, optionally blessing the SV. The C<rv>
10441 argument will be upgraded to an RV. That RV will be modified to point to
10442 the new SV. The C<classname> argument indicates the package for the
10443 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
10444 will have a reference count of 1, and the RV will be returned.
10450 Perl_sv_setref_nv(pTHX_ SV *const rv, const char *const classname, const NV nv)
10452 PERL_ARGS_ASSERT_SV_SETREF_NV;
10454 sv_setnv(newSVrv(rv,classname), nv);
10459 =for apidoc sv_setref_pvn
10461 Copies a string into a new SV, optionally blessing the SV. The length of the
10462 string must be specified with C<n>. The C<rv> argument will be upgraded to
10463 an RV. That RV will be modified to point to the new SV. The C<classname>
10464 argument indicates the package for the blessing. Set C<classname> to
10465 C<NULL> to avoid the blessing. The new SV will have a reference count
10466 of 1, and the RV will be returned.
10468 Note that C<sv_setref_pv> copies the pointer while this copies the string.
10474 Perl_sv_setref_pvn(pTHX_ SV *const rv, const char *const classname,
10475 const char *const pv, const STRLEN n)
10477 PERL_ARGS_ASSERT_SV_SETREF_PVN;
10479 sv_setpvn(newSVrv(rv,classname), pv, n);
10484 =for apidoc sv_bless
10486 Blesses an SV into a specified package. The SV must be an RV. The package
10487 must be designated by its stash (see C<L</gv_stashpv>>). The reference count
10488 of the SV is unaffected.
10494 Perl_sv_bless(pTHX_ SV *const sv, HV *const stash)
10497 HV *oldstash = NULL;
10499 PERL_ARGS_ASSERT_SV_BLESS;
10503 Perl_croak(aTHX_ "Can't bless non-reference value");
10505 if (SvFLAGS(tmpRef) & (SVs_OBJECT|SVf_READONLY|SVf_PROTECT)) {
10506 if (SvREADONLY(tmpRef))
10507 Perl_croak_no_modify();
10508 if (SvOBJECT(tmpRef)) {
10509 oldstash = SvSTASH(tmpRef);
10512 SvOBJECT_on(tmpRef);
10513 SvUPGRADE(tmpRef, SVt_PVMG);
10514 SvSTASH_set(tmpRef, MUTABLE_HV(SvREFCNT_inc_simple(stash)));
10515 SvREFCNT_dec(oldstash);
10517 if(SvSMAGICAL(tmpRef))
10518 if(mg_find(tmpRef, PERL_MAGIC_ext) || mg_find(tmpRef, PERL_MAGIC_uvar))
10526 /* Downgrades a PVGV to a PVMG. If it's actually a PVLV, we leave the type
10527 * as it is after unglobbing it.
10530 PERL_STATIC_INLINE void
10531 S_sv_unglob(pTHX_ SV *const sv, U32 flags)
10535 SV * const temp = flags & SV_COW_DROP_PV ? NULL : sv_newmortal();
10537 PERL_ARGS_ASSERT_SV_UNGLOB;
10539 assert(SvTYPE(sv) == SVt_PVGV || SvTYPE(sv) == SVt_PVLV);
10541 if (!(flags & SV_COW_DROP_PV))
10542 gv_efullname3(temp, MUTABLE_GV(sv), "*");
10544 SvREFCNT_inc_simple_void_NN(sv_2mortal(sv));
10546 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
10547 && HvNAME_get(stash))
10548 mro_method_changed_in(stash);
10549 gp_free(MUTABLE_GV(sv));
10552 sv_del_backref(MUTABLE_SV(GvSTASH(sv)), sv);
10553 GvSTASH(sv) = NULL;
10556 if (GvNAME_HEK(sv)) {
10557 unshare_hek(GvNAME_HEK(sv));
10559 isGV_with_GP_off(sv);
10561 if(SvTYPE(sv) == SVt_PVGV) {
10562 /* need to keep SvANY(sv) in the right arena */
10563 xpvmg = new_XPVMG();
10564 StructCopy(SvANY(sv), xpvmg, XPVMG);
10565 del_XPVGV(SvANY(sv));
10568 SvFLAGS(sv) &= ~SVTYPEMASK;
10569 SvFLAGS(sv) |= SVt_PVMG;
10572 /* Intentionally not calling any local SET magic, as this isn't so much a
10573 set operation as merely an internal storage change. */
10574 if (flags & SV_COW_DROP_PV) SvOK_off(sv);
10575 else sv_setsv_flags(sv, temp, 0);
10577 if ((const GV *)sv == PL_last_in_gv)
10578 PL_last_in_gv = NULL;
10579 else if ((const GV *)sv == PL_statgv)
10584 =for apidoc sv_unref_flags
10586 Unsets the RV status of the SV, and decrements the reference count of
10587 whatever was being referenced by the RV. This can almost be thought of
10588 as a reversal of C<newSVrv>. The C<cflags> argument can contain
10589 C<SV_IMMEDIATE_UNREF> to force the reference count to be decremented
10590 (otherwise the decrementing is conditional on the reference count being
10591 different from one or the reference being a readonly SV).
10592 See C<L</SvROK_off>>.
10598 Perl_sv_unref_flags(pTHX_ SV *const ref, const U32 flags)
10600 SV* const target = SvRV(ref);
10602 PERL_ARGS_ASSERT_SV_UNREF_FLAGS;
10604 if (SvWEAKREF(ref)) {
10605 sv_del_backref(target, ref);
10606 SvWEAKREF_off(ref);
10607 SvRV_set(ref, NULL);
10610 SvRV_set(ref, NULL);
10612 /* You can't have a || SvREADONLY(target) here, as $a = $$a, where $a was
10613 assigned to as BEGIN {$a = \"Foo"} will fail. */
10614 if (SvREFCNT(target) != 1 || (flags & SV_IMMEDIATE_UNREF))
10615 SvREFCNT_dec_NN(target);
10616 else /* XXX Hack, but hard to make $a=$a->[1] work otherwise */
10617 sv_2mortal(target); /* Schedule for freeing later */
10621 =for apidoc sv_untaint
10623 Untaint an SV. Use C<SvTAINTED_off> instead.
10629 Perl_sv_untaint(pTHX_ SV *const sv)
10631 PERL_ARGS_ASSERT_SV_UNTAINT;
10632 PERL_UNUSED_CONTEXT;
10634 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10635 MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10642 =for apidoc sv_tainted
10644 Test an SV for taintedness. Use C<SvTAINTED> instead.
10650 Perl_sv_tainted(pTHX_ SV *const sv)
10652 PERL_ARGS_ASSERT_SV_TAINTED;
10653 PERL_UNUSED_CONTEXT;
10655 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10656 const MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10657 if (mg && (mg->mg_len & 1) )
10663 #ifndef NO_MATHOMS /* Can't move these to mathoms.c because call uiv_2buf(),
10664 private to this file */
10667 =for apidoc sv_setpviv
10669 Copies an integer into the given SV, also updating its string value.
10670 Does not handle 'set' magic. See C<L</sv_setpviv_mg>>.
10676 Perl_sv_setpviv(pTHX_ SV *const sv, const IV iv)
10678 /* The purpose of this union is to ensure that arr is aligned on
10679 a 2 byte boundary, because that is what uiv_2buf() requires */
10681 char arr[TYPE_CHARS(UV)];
10685 char * const ptr = uiv_2buf(buf.arr, iv, 0, 0, &ebuf);
10687 PERL_ARGS_ASSERT_SV_SETPVIV;
10689 sv_setpvn(sv, ptr, ebuf - ptr);
10693 =for apidoc sv_setpviv_mg
10695 Like C<sv_setpviv>, but also handles 'set' magic.
10701 Perl_sv_setpviv_mg(pTHX_ SV *const sv, const IV iv)
10703 PERL_ARGS_ASSERT_SV_SETPVIV_MG;
10705 sv_setpviv(sv, iv);
10709 #endif /* NO_MATHOMS */
10711 #if defined(PERL_IMPLICIT_CONTEXT)
10713 /* pTHX_ magic can't cope with varargs, so this is a no-context
10714 * version of the main function, (which may itself be aliased to us).
10715 * Don't access this version directly.
10719 Perl_sv_setpvf_nocontext(SV *const sv, const char *const pat, ...)
10724 PERL_ARGS_ASSERT_SV_SETPVF_NOCONTEXT;
10726 va_start(args, pat);
10727 sv_vsetpvf(sv, pat, &args);
10731 /* pTHX_ magic can't cope with varargs, so this is a no-context
10732 * version of the main function, (which may itself be aliased to us).
10733 * Don't access this version directly.
10737 Perl_sv_setpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10742 PERL_ARGS_ASSERT_SV_SETPVF_MG_NOCONTEXT;
10744 va_start(args, pat);
10745 sv_vsetpvf_mg(sv, pat, &args);
10751 =for apidoc sv_setpvf
10753 Works like C<sv_catpvf> but copies the text into the SV instead of
10754 appending it. Does not handle 'set' magic. See C<L</sv_setpvf_mg>>.
10760 Perl_sv_setpvf(pTHX_ SV *const sv, const char *const pat, ...)
10764 PERL_ARGS_ASSERT_SV_SETPVF;
10766 va_start(args, pat);
10767 sv_vsetpvf(sv, pat, &args);
10772 =for apidoc sv_vsetpvf
10774 Works like C<sv_vcatpvf> but copies the text into the SV instead of
10775 appending it. Does not handle 'set' magic. See C<L</sv_vsetpvf_mg>>.
10777 Usually used via its frontend C<sv_setpvf>.
10783 Perl_sv_vsetpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10785 PERL_ARGS_ASSERT_SV_VSETPVF;
10787 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10791 =for apidoc sv_setpvf_mg
10793 Like C<sv_setpvf>, but also handles 'set' magic.
10799 Perl_sv_setpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10803 PERL_ARGS_ASSERT_SV_SETPVF_MG;
10805 va_start(args, pat);
10806 sv_vsetpvf_mg(sv, pat, &args);
10811 =for apidoc sv_vsetpvf_mg
10813 Like C<sv_vsetpvf>, but also handles 'set' magic.
10815 Usually used via its frontend C<sv_setpvf_mg>.
10821 Perl_sv_vsetpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10823 PERL_ARGS_ASSERT_SV_VSETPVF_MG;
10825 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10829 #if defined(PERL_IMPLICIT_CONTEXT)
10831 /* pTHX_ magic can't cope with varargs, so this is a no-context
10832 * version of the main function, (which may itself be aliased to us).
10833 * Don't access this version directly.
10837 Perl_sv_catpvf_nocontext(SV *const sv, const char *const pat, ...)
10842 PERL_ARGS_ASSERT_SV_CATPVF_NOCONTEXT;
10844 va_start(args, pat);
10845 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10849 /* pTHX_ magic can't cope with varargs, so this is a no-context
10850 * version of the main function, (which may itself be aliased to us).
10851 * Don't access this version directly.
10855 Perl_sv_catpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10860 PERL_ARGS_ASSERT_SV_CATPVF_MG_NOCONTEXT;
10862 va_start(args, pat);
10863 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10870 =for apidoc sv_catpvf
10872 Processes its arguments like C<sv_catpvfn>, and appends the formatted
10873 output to an SV. As with C<sv_catpvfn> called with a non-null C-style
10874 variable argument list, argument reordering is not supported.
10875 If the appended data contains "wide" characters
10876 (including, but not limited to, SVs with a UTF-8 PV formatted with C<%s>,
10877 and characters >255 formatted with C<%c>), the original SV might get
10878 upgraded to UTF-8. Handles 'get' magic, but not 'set' magic. See
10879 C<L</sv_catpvf_mg>>. If the original SV was UTF-8, the pattern should be
10880 valid UTF-8; if the original SV was bytes, the pattern should be too.
10885 Perl_sv_catpvf(pTHX_ SV *const sv, const char *const pat, ...)
10889 PERL_ARGS_ASSERT_SV_CATPVF;
10891 va_start(args, pat);
10892 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10897 =for apidoc sv_vcatpvf
10899 Processes its arguments like C<sv_catpvfn> called with a non-null C-style
10900 variable argument list, and appends the formatted output
10901 to an SV. Does not handle 'set' magic. See C<L</sv_vcatpvf_mg>>.
10903 Usually used via its frontend C<sv_catpvf>.
10909 Perl_sv_vcatpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10911 PERL_ARGS_ASSERT_SV_VCATPVF;
10913 sv_vcatpvfn_flags(sv, pat, strlen(pat), args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10917 =for apidoc sv_catpvf_mg
10919 Like C<sv_catpvf>, but also handles 'set' magic.
10925 Perl_sv_catpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10929 PERL_ARGS_ASSERT_SV_CATPVF_MG;
10931 va_start(args, pat);
10932 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10938 =for apidoc sv_vcatpvf_mg
10940 Like C<sv_vcatpvf>, but also handles 'set' magic.
10942 Usually used via its frontend C<sv_catpvf_mg>.
10948 Perl_sv_vcatpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10950 PERL_ARGS_ASSERT_SV_VCATPVF_MG;
10952 sv_vcatpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10957 =for apidoc sv_vsetpvfn
10959 Works like C<sv_vcatpvfn> but copies the text into the SV instead of
10962 Usually used via one of its frontends C<sv_vsetpvf> and C<sv_vsetpvf_mg>.
10968 Perl_sv_vsetpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10969 va_list *const args, SV **const svargs, const Size_t sv_count, bool *const maybe_tainted)
10971 PERL_ARGS_ASSERT_SV_VSETPVFN;
10974 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, sv_count, maybe_tainted, 0);
10978 /* simplified inline Perl_sv_catpvn_nomg() when you know the SV's SvPOK */
10980 PERL_STATIC_INLINE void
10981 S_sv_catpvn_simple(pTHX_ SV *const sv, const char* const buf, const STRLEN len)
10983 STRLEN const need = len + SvCUR(sv) + 1;
10986 /* can't wrap as both len and SvCUR() are allocated in
10987 * memory and together can't consume all the address space
10989 assert(need > len);
10994 Copy(buf, end, len, char);
10997 SvCUR_set(sv, need - 1);
11002 * Warn of missing argument to sprintf. The value used in place of such
11003 * arguments should be &PL_sv_no; an undefined value would yield
11004 * inappropriate "use of uninit" warnings [perl #71000].
11007 S_warn_vcatpvfn_missing_argument(pTHX) {
11008 if (ckWARN(WARN_MISSING)) {
11009 Perl_warner(aTHX_ packWARN(WARN_MISSING), "Missing argument in %s",
11010 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
11019 Perl_croak(aTHX_ "Integer overflow in format string for %s",
11020 (PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn"));
11024 /* Given an int i from the next arg (if args is true) or an sv from an arg
11025 * (if args is false), try to extract a STRLEN-ranged value from the arg,
11026 * with overflow checking.
11027 * Sets *neg to true if the value was negative (untouched otherwise.
11028 * Returns the absolute value.
11029 * As an extra margin of safety, it croaks if the returned value would
11030 * exceed the maximum value of a STRLEN / 4.
11034 S_sprintf_arg_num_val(pTHX_ va_list *const args, int i, SV *sv, bool *neg)
11048 if (UNLIKELY(SvIsUV(sv))) {
11049 UV uv = SvUV_nomg(sv);
11051 S_croak_overflow();
11055 iv = SvIV_nomg(sv);
11059 S_croak_overflow();
11065 if (iv > (IV)(((STRLEN)~0) / 4))
11066 S_croak_overflow();
11072 /* Returns true if c is in the range '1'..'9'
11073 * Written with the cast so it only needs one conditional test
11075 #define IS_1_TO_9(c) ((U8)(c - '1') <= 8)
11077 /* Read in and return a number. Updates *pattern to point to the char
11078 * following the number. Expects the first char to 1..9.
11079 * Croaks if the number exceeds 1/4 of the maximum value of STRLEN.
11080 * This is a belt-and-braces safety measure to complement any
11081 * overflow/wrap checks done in the main body of sv_vcatpvfn_flags.
11082 * It means that e.g. on a 32-bit system the width/precision can't be more
11083 * than 1G, which seems reasonable.
11087 S_expect_number(pTHX_ const char **const pattern)
11091 PERL_ARGS_ASSERT_EXPECT_NUMBER;
11093 assert(IS_1_TO_9(**pattern));
11095 var = *(*pattern)++ - '0';
11096 while (isDIGIT(**pattern)) {
11097 /* if var * 10 + 9 would exceed 1/4 max strlen, croak */
11098 if (var > ((((STRLEN)~0) / 4 - 9) / 10))
11099 S_croak_overflow();
11100 var = var * 10 + (*(*pattern)++ - '0');
11105 /* Implement a fast "%.0f": given a pointer to the end of a buffer (caller
11106 * ensures it's big enough), back fill it with the rounded integer part of
11107 * nv. Returns ptr to start of string, and sets *len to its length.
11108 * Returns NULL if not convertible.
11112 S_F0convert(NV nv, char *const endbuf, STRLEN *const len)
11114 const int neg = nv < 0;
11117 PERL_ARGS_ASSERT_F0CONVERT;
11119 assert(!Perl_isinfnan(nv));
11122 if (nv != 0.0 && nv < UV_MAX) {
11128 if (uv & 1 && uv == nv)
11129 uv--; /* Round to even */
11132 const unsigned dig = uv % 10;
11134 } while (uv /= 10);
11144 /* XXX maybe_tainted is never assigned to, so the doc above is lying. */
11147 Perl_sv_vcatpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11148 va_list *const args, SV **const svargs, const Size_t sv_count, bool *const maybe_tainted)
11150 PERL_ARGS_ASSERT_SV_VCATPVFN;
11152 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, sv_count, maybe_tainted, SV_GMAGIC|SV_SMAGIC);
11156 /* For the vcatpvfn code, we need a long double target in case
11157 * HAS_LONG_DOUBLE, even without USE_LONG_DOUBLE, so that we can printf
11158 * with long double formats, even without NV being long double. But we
11159 * call the target 'fv' instead of 'nv', since most of the time it is not
11160 * (most compilers these days recognize "long double", even if only as a
11161 * synonym for "double").
11163 #if defined(HAS_LONG_DOUBLE) && LONG_DOUBLESIZE > DOUBLESIZE && \
11164 defined(PERL_PRIgldbl) && !defined(USE_QUADMATH)
11165 # define VCATPVFN_FV_GF PERL_PRIgldbl
11166 # if defined(__VMS) && defined(__ia64) && defined(__IEEE_FLOAT)
11167 /* Work around breakage in OTS$CVT_FLOAT_T_X */
11168 # define VCATPVFN_NV_TO_FV(nv,fv) \
11171 fv = Perl_isnan(_dv) ? LDBL_QNAN : _dv; \
11174 # define VCATPVFN_NV_TO_FV(nv,fv) (fv)=(nv)
11176 typedef long double vcatpvfn_long_double_t;
11178 # define VCATPVFN_FV_GF NVgf
11179 # define VCATPVFN_NV_TO_FV(nv,fv) (fv)=(nv)
11180 typedef NV vcatpvfn_long_double_t;
11183 #ifdef LONGDOUBLE_DOUBLEDOUBLE
11184 /* The first double can be as large as 2**1023, or '1' x '0' x 1023.
11185 * The second double can be as small as 2**-1074, or '0' x 1073 . '1'.
11186 * The sum of them can be '1' . '0' x 2096 . '1', with implied radix point
11187 * after the first 1023 zero bits.
11189 * XXX The 2098 is quite large (262.25 bytes) and therefore some sort
11190 * of dynamically growing buffer might be better, start at just 16 bytes
11191 * (for example) and grow only when necessary. Or maybe just by looking
11192 * at the exponents of the two doubles? */
11193 # define DOUBLEDOUBLE_MAXBITS 2098
11196 /* vhex will contain the values (0..15) of the hex digits ("nybbles"
11197 * of 4 bits); 1 for the implicit 1, and the mantissa bits, four bits
11198 * per xdigit. For the double-double case, this can be rather many.
11199 * The non-double-double-long-double overshoots since all bits of NV
11200 * are not mantissa bits, there are also exponent bits. */
11201 #ifdef LONGDOUBLE_DOUBLEDOUBLE
11202 # define VHEX_SIZE (3+DOUBLEDOUBLE_MAXBITS/4)
11204 # define VHEX_SIZE (1+(NVSIZE * 8)/4)
11207 /* If we do not have a known long double format, (including not using
11208 * long doubles, or long doubles being equal to doubles) then we will
11209 * fall back to the ldexp/frexp route, with which we can retrieve at
11210 * most as many bits as our widest unsigned integer type is. We try
11211 * to get a 64-bit unsigned integer even if we are not using a 64-bit UV.
11213 * (If you want to test the case of UVSIZE == 4, NVSIZE == 8,
11214 * set the MANTISSATYPE to int and the MANTISSASIZE to 4.)
11216 #if defined(HAS_QUAD) && defined(Uquad_t)
11217 # define MANTISSATYPE Uquad_t
11218 # define MANTISSASIZE 8
11220 # define MANTISSATYPE UV
11221 # define MANTISSASIZE UVSIZE
11224 #if defined(DOUBLE_LITTLE_ENDIAN) || defined(LONGDOUBLE_LITTLE_ENDIAN)
11225 # define HEXTRACT_LITTLE_ENDIAN
11226 #elif defined(DOUBLE_BIG_ENDIAN) || defined(LONGDOUBLE_BIG_ENDIAN)
11227 # define HEXTRACT_BIG_ENDIAN
11229 # define HEXTRACT_MIX_ENDIAN
11232 /* S_hextract() is a helper for S_format_hexfp, for extracting
11233 * the hexadecimal values (for %a/%A). The nv is the NV where the value
11234 * are being extracted from (either directly from the long double in-memory
11235 * presentation, or from the uquad computed via frexp+ldexp). frexp also
11236 * is used to update the exponent. The subnormal is set to true
11237 * for IEEE 754 subnormals/denormals (including the x86 80-bit format).
11238 * The vhex is the pointer to the beginning of the output buffer of VHEX_SIZE.
11240 * The tricky part is that S_hextract() needs to be called twice:
11241 * the first time with vend as NULL, and the second time with vend as
11242 * the pointer returned by the first call. What happens is that on
11243 * the first round the output size is computed, and the intended
11244 * extraction sanity checked. On the second round the actual output
11245 * (the extraction of the hexadecimal values) takes place.
11246 * Sanity failures cause fatal failures during both rounds. */
11248 S_hextract(pTHX_ const NV nv, int* exponent, bool *subnormal,
11249 U8* vhex, U8* vend)
11253 int ixmin = 0, ixmax = 0;
11255 /* XXX Inf/NaN are not handled here, since it is
11256 * assumed they are to be output as "Inf" and "NaN". */
11258 /* These macros are just to reduce typos, they have multiple
11259 * repetitions below, but usually only one (or sometimes two)
11260 * of them is really being used. */
11261 /* HEXTRACT_OUTPUT() extracts the high nybble first. */
11262 #define HEXTRACT_OUTPUT_HI(ix) (*v++ = nvp[ix] >> 4)
11263 #define HEXTRACT_OUTPUT_LO(ix) (*v++ = nvp[ix] & 0xF)
11264 #define HEXTRACT_OUTPUT(ix) \
11266 HEXTRACT_OUTPUT_HI(ix); HEXTRACT_OUTPUT_LO(ix); \
11268 #define HEXTRACT_COUNT(ix, c) \
11270 v += c; if (ix < ixmin) ixmin = ix; else if (ix > ixmax) ixmax = ix; \
11272 #define HEXTRACT_BYTE(ix) \
11274 if (vend) HEXTRACT_OUTPUT(ix); else HEXTRACT_COUNT(ix, 2); \
11276 #define HEXTRACT_LO_NYBBLE(ix) \
11278 if (vend) HEXTRACT_OUTPUT_LO(ix); else HEXTRACT_COUNT(ix, 1); \
11280 /* HEXTRACT_TOP_NYBBLE is just convenience disguise,
11281 * to make it look less odd when the top bits of a NV
11282 * are extracted using HEXTRACT_LO_NYBBLE: the highest
11283 * order bits can be in the "low nybble" of a byte. */
11284 #define HEXTRACT_TOP_NYBBLE(ix) HEXTRACT_LO_NYBBLE(ix)
11285 #define HEXTRACT_BYTES_LE(a, b) \
11286 for (ix = a; ix >= b; ix--) { HEXTRACT_BYTE(ix); }
11287 #define HEXTRACT_BYTES_BE(a, b) \
11288 for (ix = a; ix <= b; ix++) { HEXTRACT_BYTE(ix); }
11289 #define HEXTRACT_GET_SUBNORMAL(nv) *subnormal = Perl_fp_class_denorm(nv)
11290 #define HEXTRACT_IMPLICIT_BIT(nv) \
11292 if (!*subnormal) { \
11293 if (vend) *v++ = ((nv) == 0.0) ? 0 : 1; else v++; \
11297 /* Most formats do. Those which don't should undef this.
11299 * But also note that IEEE 754 subnormals do not have it, or,
11300 * expressed alternatively, their implicit bit is zero. */
11301 #define HEXTRACT_HAS_IMPLICIT_BIT
11303 /* Many formats do. Those which don't should undef this. */
11304 #define HEXTRACT_HAS_TOP_NYBBLE
11306 /* HEXTRACTSIZE is the maximum number of xdigits. */
11307 #if defined(USE_LONG_DOUBLE) && defined(LONGDOUBLE_DOUBLEDOUBLE)
11308 # define HEXTRACTSIZE (2+DOUBLEDOUBLE_MAXBITS/4)
11310 # define HEXTRACTSIZE 2 * NVSIZE
11313 const U8* vmaxend = vhex + HEXTRACTSIZE;
11315 assert(HEXTRACTSIZE <= VHEX_SIZE);
11317 PERL_UNUSED_VAR(ix); /* might happen */
11318 (void)Perl_frexp(PERL_ABS(nv), exponent);
11319 *subnormal = FALSE;
11320 if (vend && (vend <= vhex || vend > vmaxend)) {
11321 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11322 Perl_croak(aTHX_ "Hexadecimal float: internal error (entry)");
11325 /* First check if using long doubles. */
11326 #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE)
11327 # if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN
11328 /* Used in e.g. VMS and HP-UX IA-64, e.g. -0.1L:
11329 * 9a 99 99 99 99 99 99 99 99 99 99 99 99 99 fb bf */
11330 /* The bytes 13..0 are the mantissa/fraction,
11331 * the 15,14 are the sign+exponent. */
11332 const U8* nvp = (const U8*)(&nv);
11333 HEXTRACT_GET_SUBNORMAL(nv);
11334 HEXTRACT_IMPLICIT_BIT(nv);
11335 # undef HEXTRACT_HAS_TOP_NYBBLE
11336 HEXTRACT_BYTES_LE(13, 0);
11337 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_BIG_ENDIAN
11338 /* Used in e.g. Solaris Sparc and HP-UX PA-RISC, e.g. -0.1L:
11339 * bf fb 99 99 99 99 99 99 99 99 99 99 99 99 99 9a */
11340 /* The bytes 2..15 are the mantissa/fraction,
11341 * the 0,1 are the sign+exponent. */
11342 const U8* nvp = (const U8*)(&nv);
11343 HEXTRACT_GET_SUBNORMAL(nv);
11344 HEXTRACT_IMPLICIT_BIT(nv);
11345 # undef HEXTRACT_HAS_TOP_NYBBLE
11346 HEXTRACT_BYTES_BE(2, 15);
11347 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN
11348 /* x86 80-bit "extended precision", 64 bits of mantissa / fraction /
11349 * significand, 15 bits of exponent, 1 bit of sign. No implicit bit.
11350 * NVSIZE can be either 12 (ILP32, Solaris x86) or 16 (LP64, Linux
11351 * and OS X), meaning that 2 or 6 bytes are empty padding. */
11352 /* The bytes 0..1 are the sign+exponent,
11353 * the bytes 2..9 are the mantissa/fraction. */
11354 const U8* nvp = (const U8*)(&nv);
11355 # undef HEXTRACT_HAS_IMPLICIT_BIT
11356 # undef HEXTRACT_HAS_TOP_NYBBLE
11357 HEXTRACT_GET_SUBNORMAL(nv);
11358 HEXTRACT_BYTES_LE(7, 0);
11359 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN
11360 /* Does this format ever happen? (Wikipedia says the Motorola
11361 * 6888x math coprocessors used format _like_ this but padded
11362 * to 96 bits with 16 unused bits between the exponent and the
11364 const U8* nvp = (const U8*)(&nv);
11365 # undef HEXTRACT_HAS_IMPLICIT_BIT
11366 # undef HEXTRACT_HAS_TOP_NYBBLE
11367 HEXTRACT_GET_SUBNORMAL(nv);
11368 HEXTRACT_BYTES_BE(0, 7);
11370 # define HEXTRACT_FALLBACK
11371 /* Double-double format: two doubles next to each other.
11372 * The first double is the high-order one, exactly like
11373 * it would be for a "lone" double. The second double
11374 * is shifted down using the exponent so that that there
11375 * are no common bits. The tricky part is that the value
11376 * of the double-double is the SUM of the two doubles and
11377 * the second one can be also NEGATIVE.
11379 * Because of this tricky construction the bytewise extraction we
11380 * use for the other long double formats doesn't work, we must
11381 * extract the values bit by bit.
11383 * The little-endian double-double is used .. somewhere?
11385 * The big endian double-double is used in e.g. PPC/Power (AIX)
11388 * The mantissa bits are in two separate stretches, e.g. for -0.1L:
11389 * 9a 99 99 99 99 99 59 bc 9a 99 99 99 99 99 b9 3f (LE)
11390 * 3f b9 99 99 99 99 99 9a bc 59 99 99 99 99 99 9a (BE)
11393 #else /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) */
11394 /* Using normal doubles, not long doubles.
11396 * We generate 4-bit xdigits (nybble/nibble) instead of 8-bit
11397 * bytes, since we might need to handle printf precision, and
11398 * also need to insert the radix. */
11400 # ifdef HEXTRACT_LITTLE_ENDIAN
11401 /* 0 1 2 3 4 5 6 7 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11402 const U8* nvp = (const U8*)(&nv);
11403 HEXTRACT_GET_SUBNORMAL(nv);
11404 HEXTRACT_IMPLICIT_BIT(nv);
11405 HEXTRACT_TOP_NYBBLE(6);
11406 HEXTRACT_BYTES_LE(5, 0);
11407 # elif defined(HEXTRACT_BIG_ENDIAN)
11408 /* 7 6 5 4 3 2 1 0 (MSB = 7, LSB = 0, 6+7 = exponent+sign) */
11409 const U8* nvp = (const U8*)(&nv);
11410 HEXTRACT_GET_SUBNORMAL(nv);
11411 HEXTRACT_IMPLICIT_BIT(nv);
11412 HEXTRACT_TOP_NYBBLE(1);
11413 HEXTRACT_BYTES_BE(2, 7);
11414 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_LE_BE
11415 /* 4 5 6 7 0 1 2 3 (MSB = 7, LSB = 0, 6:7 = nybble:exponent:sign) */
11416 const U8* nvp = (const U8*)(&nv);
11417 HEXTRACT_GET_SUBNORMAL(nv);
11418 HEXTRACT_IMPLICIT_BIT(nv);
11419 HEXTRACT_TOP_NYBBLE(2); /* 6 */
11420 HEXTRACT_BYTE(1); /* 5 */
11421 HEXTRACT_BYTE(0); /* 4 */
11422 HEXTRACT_BYTE(7); /* 3 */
11423 HEXTRACT_BYTE(6); /* 2 */
11424 HEXTRACT_BYTE(5); /* 1 */
11425 HEXTRACT_BYTE(4); /* 0 */
11426 # elif DOUBLEKIND == DOUBLE_IS_IEEE_754_64_BIT_MIXED_ENDIAN_BE_LE
11427 /* 3 2 1 0 7 6 5 4 (MSB = 7, LSB = 0, 7:6 = sign:exponent:nybble) */
11428 const U8* nvp = (const U8*)(&nv);
11429 HEXTRACT_GET_SUBNORMAL(nv);
11430 HEXTRACT_IMPLICIT_BIT(nv);
11431 HEXTRACT_TOP_NYBBLE(5); /* 6 */
11432 HEXTRACT_BYTE(6); /* 5 */
11433 HEXTRACT_BYTE(7); /* 4 */
11434 HEXTRACT_BYTE(0); /* 3 */
11435 HEXTRACT_BYTE(1); /* 2 */
11436 HEXTRACT_BYTE(2); /* 1 */
11437 HEXTRACT_BYTE(3); /* 0 */
11439 # define HEXTRACT_FALLBACK
11442 # define HEXTRACT_FALLBACK
11444 #endif /* #if defined(USE_LONG_DOUBLE) && (NVSIZE > DOUBLESIZE) #else */
11446 #ifdef HEXTRACT_FALLBACK
11447 HEXTRACT_GET_SUBNORMAL(nv);
11448 # undef HEXTRACT_HAS_TOP_NYBBLE /* Meaningless, but consistent. */
11449 /* The fallback is used for the double-double format, and
11450 * for unknown long double formats, and for unknown double
11451 * formats, or in general unknown NV formats. */
11452 if (nv == (NV)0.0) {
11460 NV d = nv < 0 ? -nv : nv;
11462 U8 ha = 0x0; /* hexvalue accumulator */
11463 U8 hd = 0x8; /* hexvalue digit */
11465 /* Shift d and e (and update exponent) so that e <= d < 2*e,
11466 * this is essentially manual frexp(). Multiplying by 0.5 and
11467 * doubling should be lossless in binary floating point. */
11477 while (d >= e + e) {
11481 /* Now e <= d < 2*e */
11483 /* First extract the leading hexdigit (the implicit bit). */
11499 /* Then extract the remaining hexdigits. */
11500 while (d > (NV)0.0) {
11506 /* Output or count in groups of four bits,
11507 * that is, when the hexdigit is down to one. */
11512 /* Reset the hexvalue. */
11521 /* Flush possible pending hexvalue. */
11531 /* Croak for various reasons: if the output pointer escaped the
11532 * output buffer, if the extraction index escaped the extraction
11533 * buffer, or if the ending output pointer didn't match the
11534 * previously computed value. */
11535 if (v <= vhex || v - vhex >= VHEX_SIZE ||
11536 /* For double-double the ixmin and ixmax stay at zero,
11537 * which is convenient since the HEXTRACTSIZE is tricky
11538 * for double-double. */
11539 ixmin < 0 || ixmax >= NVSIZE ||
11540 (vend && v != vend)) {
11541 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11542 Perl_croak(aTHX_ "Hexadecimal float: internal error (overflow)");
11548 /* S_format_hexfp(): helper function for Perl_sv_vcatpvfn_flags().
11550 * Processes the %a/%A hexadecimal floating-point format, since the
11551 * built-in snprintf()s which are used for most of the f/p formats, don't
11552 * universally handle %a/%A.
11553 * Populates buf of length bufsize, and returns the length of the created
11555 * The rest of the args have the same meaning as the local vars of the
11556 * same name within Perl_sv_vcatpvfn_flags().
11558 * It assumes the caller has already done STORE_LC_NUMERIC_SET_TO_NEEDED();
11560 * It requires the caller to make buf large enough.
11564 S_format_hexfp(pTHX_ char * const buf, const STRLEN bufsize, const char c,
11565 const NV nv, const vcatpvfn_long_double_t fv,
11566 bool has_precis, STRLEN precis, STRLEN width,
11567 bool alt, char plus, bool left, bool fill)
11569 /* Hexadecimal floating point. */
11571 U8 vhex[VHEX_SIZE];
11572 U8* v = vhex; /* working pointer to vhex */
11573 U8* vend; /* pointer to one beyond last digit of vhex */
11574 U8* vfnz = NULL; /* first non-zero */
11575 U8* vlnz = NULL; /* last non-zero */
11576 U8* v0 = NULL; /* first output */
11577 const bool lower = (c == 'a');
11578 /* At output the values of vhex (up to vend) will
11579 * be mapped through the xdig to get the actual
11580 * human-readable xdigits. */
11581 const char* xdig = PL_hexdigit;
11582 STRLEN zerotail = 0; /* how many extra zeros to append */
11583 int exponent = 0; /* exponent of the floating point input */
11584 bool hexradix = FALSE; /* should we output the radix */
11585 bool subnormal = FALSE; /* IEEE 754 subnormal/denormal */
11586 bool negative = FALSE;
11589 /* XXX: NaN, Inf -- though they are printed as "NaN" and "Inf".
11591 * For example with denormals, (assuming the vanilla
11592 * 64-bit double): the exponent is zero. 1xp-1074 is
11593 * the smallest denormal and the smallest double, it
11594 * could be output also as 0x0.0000000000001p-1022 to
11595 * match its internal structure. */
11597 vend = S_hextract(aTHX_ nv, &exponent, &subnormal, vhex, NULL);
11598 S_hextract(aTHX_ nv, &exponent, &subnormal, vhex, vend);
11600 #if NVSIZE > DOUBLESIZE
11601 # ifdef HEXTRACT_HAS_IMPLICIT_BIT
11602 /* In this case there is an implicit bit,
11603 * and therefore the exponent is shifted by one. */
11605 # elif defined(NV_X86_80_BIT)
11607 /* The subnormals of the x86-80 have a base exponent of -16382,
11608 * (while the physical exponent bits are zero) but the frexp()
11609 * returned the scientific-style floating exponent. We want
11610 * to map the last one as:
11611 * -16831..-16384 -> -16382 (the last normal is 0x1p-16382)
11612 * -16835..-16388 -> -16384
11613 * since we want to keep the first hexdigit
11614 * as one of the [8421]. */
11615 exponent = -4 * ( (exponent + 1) / -4) - 2;
11619 /* TBD: other non-implicit-bit platforms than the x86-80. */
11623 negative = fv < 0 || Perl_signbit(nv);
11634 xdig += 16; /* Use uppercase hex. */
11637 /* Find the first non-zero xdigit. */
11638 for (v = vhex; v < vend; v++) {
11646 /* Find the last non-zero xdigit. */
11647 for (v = vend - 1; v >= vhex; v--) {
11654 #if NVSIZE == DOUBLESIZE
11660 #ifndef NV_X86_80_BIT
11662 /* IEEE 754 subnormals (but not the x86 80-bit):
11663 * we want "normalize" the subnormal,
11664 * so we need to right shift the hex nybbles
11665 * so that the output of the subnormal starts
11666 * from the first true bit. (Another, equally
11667 * valid, policy would be to dump the subnormal
11668 * nybbles as-is, to display the "physical" layout.) */
11671 /* Find the ceil(log2(v[0])) of
11672 * the top non-zero nybble. */
11673 for (i = vfnz[0], n = 0; i > 1; i >>= 1, n++) { }
11677 for (vshr = vlnz; vshr >= vfnz; vshr--) {
11678 vshr[1] |= (vshr[0] & (0xF >> (4 - n))) << (4 - n);
11692 U8* ve = (subnormal ? vlnz + 1 : vend);
11693 SSize_t vn = ve - v0;
11695 if (precis < (Size_t)(vn - 1)) {
11696 bool overflow = FALSE;
11697 if (v0[precis + 1] < 0x8) {
11698 /* Round down, nothing to do. */
11699 } else if (v0[precis + 1] > 0x8) {
11702 overflow = v0[precis] > 0xF;
11704 } else { /* v0[precis] == 0x8 */
11705 /* Half-point: round towards the one
11706 * with the even least-significant digit:
11714 * 78 -> 8 f8 -> 10 */
11715 if ((v0[precis] & 0x1)) {
11718 overflow = v0[precis] > 0xF;
11723 for (v = v0 + precis - 1; v >= v0; v--) {
11725 overflow = *v > 0xF;
11731 if (v == v0 - 1 && overflow) {
11732 /* If the overflow goes all the
11733 * way to the front, we need to
11734 * insert 0x1 in front, and adjust
11736 Move(v0, v0 + 1, vn - 1, char);
11742 /* The new effective "last non zero". */
11743 vlnz = v0 + precis;
11747 subnormal ? precis - vn + 1 :
11748 precis - (vlnz - vhex);
11755 /* If there are non-zero xdigits, the radix
11756 * is output after the first one. */
11767 /* The radix is always output if precis, or if alt. */
11768 if (precis > 0 || alt) {
11773 #ifndef USE_LOCALE_NUMERIC
11776 if (IN_LC(LC_NUMERIC)) {
11778 const char* r = SvPV(PL_numeric_radix_sv, n);
11779 Copy(r, p, n, char);
11793 if (zerotail > 0) {
11794 while (zerotail--) {
11801 /* sanity checks */
11802 if (elen >= bufsize || width >= bufsize)
11803 /* diag_listed_as: Hexadecimal float: internal error (%s) */
11804 Perl_croak(aTHX_ "Hexadecimal float: internal error (overflow)");
11806 elen += my_snprintf(p, bufsize - elen,
11807 "%c%+d", lower ? 'p' : 'P',
11810 if (elen < width) {
11811 STRLEN gap = (STRLEN)(width - elen);
11813 /* Pad the back with spaces. */
11814 memset(buf + elen, ' ', gap);
11817 /* Insert the zeros after the "0x" and the
11818 * the potential sign, but before the digits,
11819 * otherwise we end up with "0000xH.HHH...",
11820 * when we want "0x000H.HHH..." */
11821 STRLEN nzero = gap;
11822 char* zerox = buf + 2;
11823 STRLEN nmove = elen - 2;
11824 if (negative || plus) {
11828 Move(zerox, zerox + nzero, nmove, char);
11829 memset(zerox, fill ? '0' : ' ', nzero);
11832 /* Move it to the right. */
11833 Move(buf, buf + gap,
11835 /* Pad the front with spaces. */
11836 memset(buf, ' ', gap);
11845 =for apidoc sv_vcatpvfn
11847 =for apidoc sv_vcatpvfn_flags
11849 Processes its arguments like C<vsprintf> and appends the formatted output
11850 to an SV. Uses an array of SVs if the C-style variable argument list is
11851 missing (C<NULL>). Argument reordering (using format specifiers like C<%2$d>
11852 or C<%*2$d>) is supported only when using an array of SVs; using a C-style
11853 C<va_list> argument list with a format string that uses argument reordering
11854 will yield an exception.
11856 When running with taint checks enabled, indicates via
11857 C<maybe_tainted> if results are untrustworthy (often due to the use of
11860 If called as C<sv_vcatpvfn> or flags has the C<SV_GMAGIC> bit set, calls get magic.
11862 It assumes that pat has the same utf8-ness as sv. It's the caller's
11863 responsibility to ensure that this is so.
11865 Usually used via one of its frontends C<sv_vcatpvf> and C<sv_vcatpvf_mg>.
11872 Perl_sv_vcatpvfn_flags(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
11873 va_list *const args, SV **const svargs, const Size_t sv_count, bool *const maybe_tainted,
11876 const char *fmtstart; /* character following the current '%' */
11877 const char *q; /* current position within format */
11878 const char *patend;
11881 static const char nullstr[] = "(null)";
11882 bool has_utf8 = DO_UTF8(sv); /* has the result utf8? */
11883 const bool pat_utf8 = has_utf8; /* the pattern is in utf8? */
11884 /* Times 4: a decimal digit takes more than 3 binary digits.
11885 * NV_DIG: mantissa takes that many decimal digits.
11886 * Plus 32: Playing safe. */
11887 char ebuf[IV_DIG * 4 + NV_DIG + 32];
11888 bool no_redundant_warning = FALSE; /* did we use any explicit format parameter index? */
11889 #ifdef USE_LOCALE_NUMERIC
11890 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
11891 bool lc_numeric_set = FALSE; /* called STORE_LC_NUMERIC_SET_TO_NEEDED? */
11894 PERL_ARGS_ASSERT_SV_VCATPVFN_FLAGS;
11895 PERL_UNUSED_ARG(maybe_tainted);
11897 if (flags & SV_GMAGIC)
11900 /* no matter what, this is a string now */
11901 (void)SvPV_force_nomg(sv, origlen);
11903 /* the code that scans for flags etc following a % relies on
11904 * a '\0' being present to avoid falling off the end. Ideally that
11905 * should be fixed */
11906 assert(pat[patlen] == '\0');
11909 /* Special-case "", "%s", "%-p" (SVf - see below) and "%.0f".
11910 * In each case, if there isn't the correct number of args, instead
11911 * fall through to the main code to handle the issuing of any
11915 if (patlen == 0 && (args || sv_count == 0))
11918 if (patlen <= 4 && pat[0] == '%' && (args || sv_count == 1)) {
11921 if (patlen == 2 && pat[1] == 's') {
11923 const char * const s = va_arg(*args, char*);
11924 sv_catpv_nomg(sv, s ? s : nullstr);
11927 /* we want get magic on the source but not the target.
11928 * sv_catsv can't do that, though */
11929 SvGETMAGIC(*svargs);
11930 sv_catsv_nomg(sv, *svargs);
11937 if (patlen == 3 && pat[1] == '-' && pat[2] == 'p') {
11938 SV *asv = MUTABLE_SV(va_arg(*args, void*));
11939 sv_catsv_nomg(sv, asv);
11943 #if !defined(USE_LONG_DOUBLE) && !defined(USE_QUADMATH)
11944 /* special-case "%.0f" */
11945 else if ( patlen == 4
11946 && pat[1] == '.' && pat[2] == '0' && pat[3] == 'f')
11948 const NV nv = SvNV(*svargs);
11949 if (LIKELY(!Perl_isinfnan(nv))) {
11953 if ((p = F0convert(nv, ebuf + sizeof ebuf, &l))) {
11954 sv_catpvn_nomg(sv, p, l);
11959 #endif /* !USE_LONG_DOUBLE */
11963 patend = (char*)pat + patlen;
11964 for (fmtstart = pat; fmtstart < patend; fmtstart = q) {
11965 char intsize = 0; /* size qualifier in "%hi..." etc */
11966 bool alt = FALSE; /* has "%#..." */
11967 bool left = FALSE; /* has "%-..." */
11968 bool fill = FALSE; /* has "%0..." */
11969 char plus = 0; /* has "%+..." */
11970 STRLEN width = 0; /* value of "%NNN..." */
11971 bool has_precis = FALSE; /* has "%.NNN..." */
11972 STRLEN precis = 0; /* value of "%.NNN..." */
11973 int base = 0; /* base to print in, e.g. 8 for %o */
11974 UV uv = 0; /* the value to print of int-ish args */
11976 bool vectorize = FALSE; /* has "%v..." */
11977 bool vec_utf8 = FALSE; /* SvUTF8(vec arg) */
11978 const U8 *vecstr = NULL; /* SvPVX(vec arg) */
11979 STRLEN veclen = 0; /* SvCUR(vec arg) */
11980 const char *dotstr = NULL; /* separator string for %v */
11981 STRLEN dotstrlen; /* length of separator string for %v */
11983 Size_t efix = 0; /* explicit format parameter index */
11984 const Size_t osvix = svix; /* original index in case of bad fmt */
11987 bool is_utf8 = FALSE; /* is this item utf8? */
11988 bool arg_missing = FALSE; /* give "Missing argument" warning */
11989 char esignbuf[4]; /* holds sign prefix, e.g. "-0x" */
11990 STRLEN esignlen = 0; /* length of e.g. "-0x" */
11991 STRLEN zeros = 0; /* how many '0' to prepend */
11993 const char *eptr = NULL; /* the address of the element string */
11994 STRLEN elen = 0; /* the length of the element string */
11996 char c; /* the actual format ('d', s' etc) */
11999 /* echo everything up to the next format specification */
12000 for (q = fmtstart; q < patend && *q != '%'; ++q)
12003 if (q > fmtstart) {
12004 if (has_utf8 && !pat_utf8) {
12005 /* upgrade and copy the bytes of fmtstart..q-1 to utf8 on
12009 STRLEN need = SvCUR(sv) + (q - fmtstart) + 1;
12011 for (p = fmtstart; p < q; p++)
12012 if (!NATIVE_BYTE_IS_INVARIANT(*p))
12017 for (p = fmtstart; p < q; p++)
12018 append_utf8_from_native_byte((U8)*p, (U8**)&dst);
12020 SvCUR_set(sv, need - 1);
12023 S_sv_catpvn_simple(aTHX_ sv, fmtstart, q - fmtstart);
12028 fmtstart = q; /* fmtstart is char following the '%' */
12031 We allow format specification elements in this order:
12032 \d+\$ explicit format parameter index
12034 v|\*(\d+\$)?v vector with optional (optionally specified) arg
12035 0 flag (as above): repeated to allow "v02"
12036 \d+|\*(\d+\$)? width using optional (optionally specified) arg
12037 \.(\d*|\*(\d+\$)?) precision using optional (optionally specified) arg
12039 [%bcdefginopsuxDFOUX] format (mandatory)
12042 if (IS_1_TO_9(*q)) {
12043 width = expect_number(&q);
12046 Perl_croak_nocontext(
12047 "Cannot yet reorder sv_catpvfn() arguments from va_list");
12049 efix = (Size_t)width;
12051 no_redundant_warning = TRUE;
12063 if (plus == '+' && *q == ' ') /* '+' over ' ' */
12090 /* at this point we can expect one of:
12092 * 123 an explicit width
12093 * * width taken from next arg
12094 * *12$ width taken from 12th arg
12097 * But any width specification may be preceded by a v, in one of its
12102 * So an asterisk may be either a width specifier or a vector
12103 * separator arg specifier, and we don't know which initially
12108 STRLEN ix; /* explicit width/vector separator index */
12110 if (IS_1_TO_9(*q)) {
12111 ix = expect_number(&q);
12114 Perl_croak_nocontext(
12115 "Cannot yet reorder sv_catpvfn() arguments from va_list");
12116 no_redundant_warning = TRUE;
12125 /* The asterisk was for *v, *NNN$v: vectorizing, but not
12126 * with the default "." */
12131 vecsv = va_arg(*args, SV*);
12133 ix = ix ? ix - 1 : svix++;
12134 vecsv = ix < sv_count ? svargs[ix]
12135 : (arg_missing = TRUE, &PL_sv_no);
12137 dotstr = SvPV_const(vecsv, dotstrlen);
12138 /* Keep the DO_UTF8 test *after* the SvPV call, else things go
12139 bad with tied or overloaded values that return UTF8. */
12140 if (DO_UTF8(vecsv))
12142 else if (has_utf8) {
12143 vecsv = sv_mortalcopy(vecsv);
12144 sv_utf8_upgrade(vecsv);
12145 dotstr = SvPV_const(vecsv, dotstrlen);
12152 /* the asterisk specified a width */
12157 i = va_arg(*args, int);
12159 ix = ix ? ix - 1 : svix++;
12160 sv = (ix < sv_count) ? svargs[ix]
12161 : (arg_missing = TRUE, (SV*)NULL);
12163 width = S_sprintf_arg_num_val(aTHX_ args, i, sv, &left);
12166 else if (*q == 'v') {
12177 /* explicit width? */
12183 width = expect_number(&q);
12193 STRLEN ix; /* explicit precision index */
12195 if (IS_1_TO_9(*q)) {
12196 ix = expect_number(&q);
12199 Perl_croak_nocontext(
12200 "Cannot yet reorder sv_catpvfn() arguments from va_list");
12201 no_redundant_warning = TRUE;
12214 i = va_arg(*args, int);
12216 ix = ix ? ix - 1 : svix++;
12217 sv = (ix < sv_count) ? svargs[ix]
12218 : (arg_missing = TRUE, (SV*)NULL);
12220 precis = S_sprintf_arg_num_val(aTHX_ args, i, sv, &neg);
12225 /* although it doesn't seem documented, this code has long
12227 * no digits following the '.' is treated like '.0'
12228 * the number may be preceded by any number of zeroes,
12229 * e.g. "%.0001f", which is the same as "%.1f"
12230 * so I've kept that behaviour. DAPM May 2017
12234 precis = IS_1_TO_9(*q) ? expect_number(&q) : 0;
12243 case 'I': /* Ix, I32x, and I64x */
12244 # ifdef USE_64_BIT_INT
12245 if (q[1] == '6' && q[2] == '4') {
12251 if (q[1] == '3' && q[2] == '2') {
12255 # ifdef USE_64_BIT_INT
12261 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
12262 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
12265 # ifdef USE_QUADMATH
12278 #if (IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)) || \
12279 (IVSIZE == 4 && !defined(HAS_LONG_DOUBLE))
12280 if (*q == 'l') { /* lld, llf */
12289 if (*++q == 'h') { /* hhd, hhu */
12306 c = *q++; /* c now holds the conversion type */
12308 /* '%' doesn't have an arg, so skip arg processing */
12317 if (vectorize && !strchr("BbDdiOouUXx", c))
12320 /* get next arg (individual branches do their own va_arg()
12321 * handling for the args case) */
12324 efix = efix ? efix - 1 : svix++;
12325 argsv = efix < sv_count ? svargs[efix]
12326 : (arg_missing = TRUE, &PL_sv_no);
12336 eptr = va_arg(*args, char*);
12339 elen = my_strnlen(eptr, precis);
12341 elen = strlen(eptr);
12343 eptr = (char *)nullstr;
12344 elen = sizeof nullstr - 1;
12348 eptr = SvPV_const(argsv, elen);
12349 if (DO_UTF8(argsv)) {
12350 STRLEN old_precis = precis;
12351 if (has_precis && precis < elen) {
12352 STRLEN ulen = sv_or_pv_len_utf8(argsv, eptr, elen);
12353 STRLEN p = precis > ulen ? ulen : precis;
12354 precis = sv_or_pv_pos_u2b(argsv, eptr, p, 0);
12355 /* sticks at end */
12357 if (width) { /* fudge width (can't fudge elen) */
12358 if (has_precis && precis < elen)
12359 width += precis - old_precis;
12362 elen - sv_or_pv_len_utf8(argsv,eptr,elen);
12369 if (has_precis && precis < elen)
12381 * "%...p" is normally treated like "%...x", except that the
12382 * number to print is the SV's address (or a pointer address
12383 * for C-ish sprintf).
12385 * However, the C-ish sprintf variant allows a few special
12386 * extensions. These are currently:
12388 * %-p (SVf) Like %s, but gets the string from an SV*
12389 * arg rather than a char* arg.
12390 * (This was previously %_).
12392 * %-<num>p Ditto but like %.<num>s (i.e. num is max width)
12394 * %2p (HEKf) Like %s, but using the key string in a HEK
12396 * %3p (HEKf256) Ditto but like %.256s
12398 * %d%lu%4p (UTF8f) A utf8 string. Consumes 3 args:
12399 * (cBOOL(utf8), len, string_buf).
12400 * It's handled by the "case 'd'" branch
12401 * rather than here.
12403 * %<num>p where num is 1 or > 4: reserved for future
12404 * extensions. Warns, but then is treated as a
12405 * general %p (print hex address) format.
12413 /* not %*p or %*1$p - any width was explicit */
12417 if (left) { /* %-p (SVf), %-NNNp */
12422 argsv = MUTABLE_SV(va_arg(*args, void*));
12423 eptr = SvPV_const(argsv, elen);
12424 if (DO_UTF8(argsv))
12429 else if (width == 2 || width == 3) { /* HEKf, HEKf256 */
12430 HEK * const hek = va_arg(*args, HEK *);
12431 eptr = HEK_KEY(hek);
12432 elen = HEK_LEN(hek);
12443 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
12444 "internal %%<num>p might conflict with future printf extensions");
12448 /* treat as normal %...p */
12450 uv = PTR2UV(args ? va_arg(*args, void*) : argsv);
12455 /* Ignore any size specifiers, since they're not documented as
12456 * being allowed for %c (ideally we should warn on e.g. '%hc').
12457 * Setting a default intsize, along with a positive
12458 * (which signals unsigned) base, causes, for C-ish use, the
12459 * va_arg to be interpreted as as unsigned int, when it's
12460 * actually signed, which will convert -ve values to high +ve
12461 * values. Note that unlike the libc %c, values > 255 will
12462 * convert to high unicode points rather than being truncated
12463 * to 8 bits. For perlish use, it will do SvUV(argsv), which
12464 * will again convert -ve args to high -ve values.
12467 base = 1; /* special value that indicates we're doing a 'c' */
12468 goto get_int_arg_val;
12477 goto get_int_arg_val;
12480 /* probably just a plain %d, but it might be the start of the
12481 * special UTF8f format, which usually looks something like
12482 * "%d%lu%4p" (the lu may vary by platform)
12484 assert((UTF8f)[0] == 'd');
12485 assert((UTF8f)[1] == '%');
12487 if ( args /* UTF8f only valid for C-ish sprintf */
12488 && q == fmtstart + 1 /* plain %d, not %....d */
12489 && patend >= fmtstart + sizeof(UTF8f) - 1 /* long enough */
12491 && strnEQ(q + 1, UTF8f + 2, sizeof(UTF8f) - 3))
12493 /* The argument has already gone through cBOOL, so the cast
12495 is_utf8 = (bool)va_arg(*args, int);
12496 elen = va_arg(*args, UV);
12497 /* if utf8 length is larger than 0x7ffff..., then it might
12498 * have been a signed value that wrapped */
12499 if (elen > ((~(STRLEN)0) >> 1)) {
12500 assert(0); /* in DEBUGGING build we want to crash */
12501 elen = 0; /* otherwise we want to treat this as an empty string */
12503 eptr = va_arg(*args, char *);
12504 q += sizeof(UTF8f) - 2;
12511 goto get_int_arg_val;
12522 goto get_int_arg_val;
12527 goto get_int_arg_val;
12538 goto get_int_arg_val;
12553 esignbuf[esignlen++] = plus;
12556 /* initialise the vector string to iterate over */
12558 vecsv = args ? va_arg(*args, SV*) : argsv;
12560 /* if this is a version object, we need to convert
12561 * back into v-string notation and then let the
12562 * vectorize happen normally
12564 if (sv_isobject(vecsv) && sv_derived_from(vecsv, "version")) {
12565 if ( hv_existss(MUTABLE_HV(SvRV(vecsv)), "alpha") ) {
12566 Perl_ck_warner_d(aTHX_ packWARN(WARN_PRINTF),
12567 "vector argument not supported with alpha versions");
12571 vecstr = (U8*)SvPV_const(vecsv,veclen);
12572 vecsv = sv_newmortal();
12573 scan_vstring((char *)vecstr, (char *)vecstr + veclen,
12577 vecstr = (U8*)SvPV_const(vecsv, veclen);
12578 vec_utf8 = DO_UTF8(vecsv);
12580 /* This is the re-entry point for when we're iterating
12581 * over the individual characters of a vector arg */
12584 goto done_valid_conversion;
12586 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
12596 /* test arg for inf/nan. This can trigger an unwanted
12597 * 'str' overload, so manually force 'num' overload first
12601 if (UNLIKELY(SvAMAGIC(argsv)))
12602 argsv = sv_2num(argsv);
12603 if (UNLIKELY(isinfnansv(argsv)))
12604 goto handle_infnan_argsv;
12608 /* signed int type */
12613 case 'c': iv = (char)va_arg(*args, int); break;
12614 case 'h': iv = (short)va_arg(*args, int); break;
12615 case 'l': iv = va_arg(*args, long); break;
12616 case 'V': iv = va_arg(*args, IV); break;
12617 case 'z': iv = va_arg(*args, SSize_t); break;
12618 #ifdef HAS_PTRDIFF_T
12619 case 't': iv = va_arg(*args, ptrdiff_t); break;
12621 default: iv = va_arg(*args, int); break;
12622 case 'j': iv = va_arg(*args, PERL_INTMAX_T); break;
12625 iv = va_arg(*args, Quad_t); break;
12632 /* assign to tiv then cast to iv to work around
12633 * 2003 GCC cast bug (gnu.org bugzilla #13488) */
12634 IV tiv = SvIV_nomg(argsv);
12636 case 'c': iv = (char)tiv; break;
12637 case 'h': iv = (short)tiv; break;
12638 case 'l': iv = (long)tiv; break;
12640 default: iv = tiv; break;
12643 iv = (Quad_t)tiv; break;
12650 /* now convert iv to uv */
12654 esignbuf[esignlen++] = plus;
12658 esignbuf[esignlen++] = '-';
12662 /* unsigned int type */
12665 case 'c': uv = (unsigned char)va_arg(*args, unsigned);
12667 case 'h': uv = (unsigned short)va_arg(*args, unsigned);
12669 case 'l': uv = va_arg(*args, unsigned long); break;
12670 case 'V': uv = va_arg(*args, UV); break;
12671 case 'z': uv = va_arg(*args, Size_t); break;
12672 #ifdef HAS_PTRDIFF_T
12673 /* will sign extend, but there is no
12674 * uptrdiff_t, so oh well */
12675 case 't': uv = va_arg(*args, ptrdiff_t); break;
12677 case 'j': uv = va_arg(*args, PERL_UINTMAX_T); break;
12678 default: uv = va_arg(*args, unsigned); break;
12681 uv = va_arg(*args, Uquad_t); break;
12688 /* assign to tiv then cast to iv to work around
12689 * 2003 GCC cast bug (gnu.org bugzilla #13488) */
12690 UV tuv = SvUV_nomg(argsv);
12692 case 'c': uv = (unsigned char)tuv; break;
12693 case 'h': uv = (unsigned short)tuv; break;
12694 case 'l': uv = (unsigned long)tuv; break;
12696 default: uv = tuv; break;
12699 uv = (Uquad_t)tuv; break;
12710 char *ptr = ebuf + sizeof ebuf;
12717 const char * const p =
12718 (c == 'X') ? PL_hexdigit + 16 : PL_hexdigit;
12723 } while (uv >>= 4);
12724 if (alt && *ptr != '0') {
12725 esignbuf[esignlen++] = '0';
12726 esignbuf[esignlen++] = c; /* 'x' or 'X' */
12733 *--ptr = '0' + dig;
12734 } while (uv >>= 3);
12735 if (alt && *ptr != '0')
12741 *--ptr = '0' + dig;
12742 } while (uv >>= 1);
12743 if (alt && *ptr != '0') {
12744 esignbuf[esignlen++] = '0';
12745 esignbuf[esignlen++] = c; /* 'b' or 'B' */
12750 /* special-case: base 1 indicates a 'c' format:
12751 * we use the common code for extracting a uv,
12752 * but handle that value differently here than
12753 * all the other int types */
12755 (!UVCHR_IS_INVARIANT(uv) && SvUTF8(sv)))
12758 assert(sizeof(ebuf) >= UTF8_MAXBYTES + 1);
12760 elen = uvchr_to_utf8((U8*)eptr, uv) - (U8*)ebuf;
12765 ebuf[0] = (char)uv;
12770 default: /* it had better be ten or less */
12773 *--ptr = '0' + dig;
12774 } while (uv /= base);
12777 elen = (ebuf + sizeof ebuf) - ptr;
12781 zeros = precis - elen;
12782 else if (precis == 0 && elen == 1 && *eptr == '0'
12783 && !(base == 8 && alt)) /* "%#.0o" prints "0" */
12786 /* a precision nullifies the 0 flag. */
12792 /* FLOATING POINT */
12795 c = 'f'; /* maybe %F isn't supported here */
12797 case 'e': case 'E':
12799 case 'g': case 'G':
12800 case 'a': case 'A':
12803 STRLEN float_need; /* what PL_efloatsize needs to become */
12804 bool hexfp; /* hexadecimal floating point? */
12806 vcatpvfn_long_double_t fv;
12809 /* This is evil, but floating point is even more evil */
12811 /* for SV-style calling, we can only get NV
12812 for C-style calling, we assume %f is double;
12813 for simplicity we allow any of %Lf, %llf, %qf for long double
12817 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12821 /* [perl #20339] - we should accept and ignore %lf rather than die */
12825 #if defined(USE_LONG_DOUBLE) || defined(USE_QUADMATH)
12826 intsize = args ? 0 : 'q';
12830 #if defined(HAS_LONG_DOUBLE)
12843 /* Now we need (long double) if intsize == 'q', else (double). */
12845 /* Note: do not pull NVs off the va_list with va_arg()
12846 * (pull doubles instead) because if you have a build
12847 * with long doubles, you would always be pulling long
12848 * doubles, which would badly break anyone using only
12849 * doubles (i.e. the majority of builds). In other
12850 * words, you cannot mix doubles and long doubles.
12851 * The only case where you can pull off long doubles
12852 * is when the format specifier explicitly asks so with
12854 #ifdef USE_QUADMATH
12855 fv = intsize == 'q' ?
12856 va_arg(*args, NV) : va_arg(*args, double);
12858 #elif LONG_DOUBLESIZE > DOUBLESIZE
12859 if (intsize == 'q') {
12860 fv = va_arg(*args, long double);
12863 nv = va_arg(*args, double);
12864 VCATPVFN_NV_TO_FV(nv, fv);
12867 nv = va_arg(*args, double);
12874 /* we jump here if an int-ish format encountered an
12875 * infinite/Nan argsv. After setting nv/fv, it falls
12876 * into the isinfnan block which follows */
12877 handle_infnan_argsv:
12878 nv = SvNV_nomg(argsv);
12879 VCATPVFN_NV_TO_FV(nv, fv);
12882 if (Perl_isinfnan(nv)) {
12884 Perl_croak(aTHX_ "Cannot printf %" NVgf " with '%c'",
12885 SvNV_nomg(argsv), (int)c);
12887 elen = S_infnan_2pv(nv, ebuf, sizeof(ebuf), plus);
12896 /* special-case "%.0f" */
12900 && !(width || left || plus || alt)
12903 && ((eptr = F0convert(nv, ebuf + sizeof ebuf, &elen)))
12907 /* Determine the buffer size needed for the various
12908 * floating-point formats.
12910 * The basic possibilities are:
12913 * %f 1111111.123456789
12914 * %e 1.111111123e+06
12915 * %a 0x1.0f4471f9bp+20
12917 * %g 1.11111112e+15
12919 * where P is the value of the precision in the format, or 6
12920 * if not specified. Note the two possible output formats of
12921 * %g; in both cases the number of significant digits is <=
12924 * For most of the format types the maximum buffer size needed
12925 * is precision, plus: any leading 1 or 0x1, the radix
12926 * point, and an exponent. The difficult one is %f: for a
12927 * large positive exponent it can have many leading digits,
12928 * which needs to be calculated specially. Also %a is slightly
12929 * different in that in the absence of a specified precision,
12930 * it uses as many digits as necessary to distinguish
12931 * different values.
12933 * First, here are the constant bits. For ease of calculation
12934 * we over-estimate the needed buffer size, for example by
12935 * assuming all formats have an exponent and a leading 0x1.
12937 * Also for production use, add a little extra overhead for
12938 * safety's sake. Under debugging don't, as it means we're
12939 * more likely to quickly spot issues during development.
12942 float_need = 1 /* possible unary minus */
12943 + 4 /* "0x1" plus very unlikely carry */
12944 + 1 /* default radix point '.' */
12945 + 2 /* "e-", "p+" etc */
12946 + 6 /* exponent: up to 16383 (quad fp) */
12948 + 20 /* safety net */
12953 /* determine the radix point len, e.g. length(".") in "1.2" */
12954 #ifdef USE_LOCALE_NUMERIC
12955 /* note that we may either explicitly use PL_numeric_radix_sv
12956 * below, or implicitly, via an snprintf() variant.
12957 * Note also things like ps_AF.utf8 which has
12958 * "\N{ARABIC DECIMAL SEPARATOR} as a radix point */
12959 if (!lc_numeric_set) {
12960 /* only set once and reuse in-locale value on subsequent
12962 * XXX what happens if we die in an eval?
12964 STORE_LC_NUMERIC_SET_TO_NEEDED();
12965 lc_numeric_set = TRUE;
12968 if (IN_LC(LC_NUMERIC)) {
12969 /* this can't wrap unless PL_numeric_radix_sv is a string
12970 * consuming virtually all the 32-bit or 64-bit address
12973 float_need += (SvCUR(PL_numeric_radix_sv) - 1);
12975 /* floating-point formats only get utf8 if the radix point
12976 * is utf8. All other characters in the string are < 128
12977 * and so can be safely appended to both a non-utf8 and utf8
12979 * Note that this will convert the output to utf8 even if
12980 * the radix point didn't get output.
12982 if (SvUTF8(PL_numeric_radix_sv) && !has_utf8) {
12983 sv_utf8_upgrade(sv);
12991 if (isALPHA_FOLD_EQ(c, 'f')) {
12992 /* Determine how many digits before the radix point
12993 * might be emitted. frexp() (or frexpl) has some
12994 * unspecified behaviour for nan/inf/-inf, so lucky we've
12995 * already handled them above */
12997 int i = PERL_INT_MIN;
12998 (void)Perl_frexp((NV)fv, &i);
12999 if (i == PERL_INT_MIN)
13000 Perl_die(aTHX_ "panic: frexp: %" VCATPVFN_FV_GF, fv);
13003 digits = BIT_DIGITS(i);
13004 /* this can't overflow. 'digits' will only be a few
13005 * thousand even for the largest floating-point types.
13006 * And up until now float_need is just some small
13007 * constants plus radix len, which can't be in
13008 * overflow territory unless the radix SV is consuming
13009 * over 1/2 the address space */
13010 assert(float_need < ((STRLEN)~0) - digits);
13011 float_need += digits;
13014 else if (UNLIKELY(isALPHA_FOLD_EQ(c, 'a'))) {
13017 /* %a in the absence of precision may print as many
13018 * digits as needed to represent the entire mantissa
13020 * This estimate seriously overshoots in most cases,
13021 * but better the undershooting. Firstly, all bytes
13022 * of the NV are not mantissa, some of them are
13023 * exponent. Secondly, for the reasonably common
13024 * long doubles case, the "80-bit extended", two
13025 * or six bytes of the NV are unused. Also, we'll
13026 * still pick up an extra +6 from the default
13027 * precision calculation below. */
13029 #ifdef LONGDOUBLE_DOUBLEDOUBLE
13030 /* For the "double double", we need more.
13031 * Since each double has their own exponent, the
13032 * doubles may float (haha) rather far from each
13033 * other, and the number of required bits is much
13034 * larger, up to total of DOUBLEDOUBLE_MAXBITS bits.
13035 * See the definition of DOUBLEDOUBLE_MAXBITS.
13037 * Need 2 hexdigits for each byte. */
13038 (DOUBLEDOUBLE_MAXBITS/8 + 1) * 2;
13040 NVSIZE * 2; /* 2 hexdigits for each byte */
13042 /* see "this can't overflow" comment above */
13043 assert(float_need < ((STRLEN)~0) - digits);
13044 float_need += digits;
13047 /* special-case "%.<number>g" if it will fit in ebuf */
13049 && precis /* See earlier comment about buggy Gconvert
13050 when digits, aka precis, is 0 */
13052 /* check, in manner not involving wrapping, that it will
13054 && float_need < sizeof(ebuf)
13055 && sizeof(ebuf) - float_need > precis
13056 && !(width || left || plus || alt)
13060 SNPRINTF_G(fv, ebuf, sizeof(ebuf), precis);
13061 elen = strlen(ebuf);
13068 STRLEN pr = has_precis ? precis : 6; /* known default */
13069 /* this probably can't wrap, since precis is limited
13070 * to 1/4 address space size, but better safe than sorry
13072 if (float_need >= ((STRLEN)~0) - pr)
13073 croak_memory_wrap();
13077 if (float_need < width)
13078 float_need = width;
13080 if (PL_efloatsize <= float_need) {
13081 /* PL_efloatbuf should be at least 1 greater than
13082 * float_need to allow a trailing \0 to be returned by
13083 * snprintf(). If we need to grow, overgrow for the
13084 * benefit of future generations */
13085 const STRLEN extra = 0x20;
13086 if (float_need >= ((STRLEN)~0) - extra)
13087 croak_memory_wrap();
13088 float_need += extra;
13089 Safefree(PL_efloatbuf);
13090 PL_efloatsize = float_need;
13091 Newx(PL_efloatbuf, PL_efloatsize, char);
13092 PL_efloatbuf[0] = '\0';
13095 if (UNLIKELY(hexfp)) {
13096 elen = S_format_hexfp(aTHX_ PL_efloatbuf, PL_efloatsize, c,
13097 nv, fv, has_precis, precis, width,
13098 alt, plus, left, fill);
13101 char *ptr = ebuf + sizeof ebuf;
13104 #if defined(USE_QUADMATH)
13105 if (intsize == 'q') {
13109 /* FIXME: what to do if HAS_LONG_DOUBLE but not PERL_PRIfldbl? */
13110 #elif defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
13111 /* Note that this is HAS_LONG_DOUBLE and PERL_PRIfldbl,
13112 * not USE_LONG_DOUBLE and NVff. In other words,
13113 * this needs to work without USE_LONG_DOUBLE. */
13114 if (intsize == 'q') {
13115 /* Copy the one or more characters in a long double
13116 * format before the 'base' ([efgEFG]) character to
13117 * the format string. */
13118 static char const ldblf[] = PERL_PRIfldbl;
13119 char const *p = ldblf + sizeof(ldblf) - 3;
13120 while (p >= ldblf) { *--ptr = *p--; }
13125 do { *--ptr = '0' + (base % 10); } while (base /= 10);
13130 do { *--ptr = '0' + (base % 10); } while (base /= 10);
13142 /* No taint. Otherwise we are in the strange situation
13143 * where printf() taints but print($float) doesn't.
13146 /* hopefully the above makes ptr a very constrained format
13147 * that is safe to use, even though it's not literal */
13148 GCC_DIAG_IGNORE_STMT(-Wformat-nonliteral);
13149 #ifdef USE_QUADMATH
13151 const char* qfmt = quadmath_format_single(ptr);
13153 Perl_croak_nocontext("panic: quadmath invalid format \"%s\"", ptr);
13154 elen = quadmath_snprintf(PL_efloatbuf, PL_efloatsize,
13156 if ((IV)elen == -1) {
13159 Perl_croak_nocontext("panic: quadmath_snprintf failed, format \"%s\"", qfmt);
13164 #elif defined(HAS_LONG_DOUBLE)
13165 elen = ((intsize == 'q')
13166 ? my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv)
13167 : my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, (double)fv));
13169 elen = my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, fv);
13171 GCC_DIAG_RESTORE_STMT;
13174 eptr = PL_efloatbuf;
13178 /* Since floating-point formats do their own formatting and
13179 * padding, we skip the main block of code at the end of this
13180 * loop which handles appending eptr to sv, and do our own
13181 * stripped-down version */
13186 assert(elen >= width);
13188 S_sv_catpvn_simple(aTHX_ sv, eptr, elen);
13190 goto done_valid_conversion;
13198 /* XXX ideally we should warn if any flags etc have been
13199 * set, e.g. "%-4.5n" */
13200 /* XXX if sv was originally non-utf8 with a char in the
13201 * range 0x80-0xff, then if it got upgraded, we should
13202 * calculate char len rather than byte len here */
13203 len = SvCUR(sv) - origlen;
13205 int i = (len > PERL_INT_MAX) ? PERL_INT_MAX : (int)len;
13208 case 'c': *(va_arg(*args, char*)) = i; break;
13209 case 'h': *(va_arg(*args, short*)) = i; break;
13210 default: *(va_arg(*args, int*)) = i; break;
13211 case 'l': *(va_arg(*args, long*)) = i; break;
13212 case 'V': *(va_arg(*args, IV*)) = i; break;
13213 case 'z': *(va_arg(*args, SSize_t*)) = i; break;
13214 #ifdef HAS_PTRDIFF_T
13215 case 't': *(va_arg(*args, ptrdiff_t*)) = i; break;
13217 case 'j': *(va_arg(*args, PERL_INTMAX_T*)) = i; break;
13220 *(va_arg(*args, Quad_t*)) = i; break;
13228 Perl_croak_nocontext(
13229 "Missing argument for %%n in %s",
13230 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
13231 sv_setuv_mg(argsv, has_utf8 ? (UV)sv_len_utf8(sv) : (UV)len);
13233 goto done_valid_conversion;
13241 && (PL_op->op_type == OP_PRTF || PL_op->op_type == OP_SPRINTF)
13242 && ckWARN(WARN_PRINTF))
13244 SV * const msg = sv_newmortal();
13245 Perl_sv_setpvf(aTHX_ msg, "Invalid conversion in %sprintf: ",
13246 (PL_op->op_type == OP_PRTF) ? "" : "s");
13247 if (fmtstart < patend) {
13248 const char * const fmtend = q < patend ? q : patend;
13250 sv_catpvs(msg, "\"%");
13251 for (f = fmtstart; f < fmtend; f++) {
13253 sv_catpvn_nomg(msg, f, 1);
13255 Perl_sv_catpvf(aTHX_ msg,
13256 "\\%03" UVof, (UV)*f & 0xFF);
13259 sv_catpvs(msg, "\"");
13261 sv_catpvs(msg, "end of string");
13263 Perl_warner(aTHX_ packWARN(WARN_PRINTF), "%" SVf, SVfARG(msg)); /* yes, this is reentrant */
13266 /* mangled format: output the '%', then continue from the
13267 * character following that */
13268 sv_catpvn_nomg(sv, fmtstart-1, 1);
13271 /* Any "redundant arg" warning from now onwards will probably
13272 * just be misleading, so don't bother. */
13273 no_redundant_warning = TRUE;
13274 continue; /* not "break" */
13277 if (is_utf8 != has_utf8) {
13280 sv_utf8_upgrade(sv);
13283 const STRLEN old_elen = elen;
13284 SV * const nsv = newSVpvn_flags(eptr, elen, SVs_TEMP);
13285 sv_utf8_upgrade(nsv);
13286 eptr = SvPVX_const(nsv);
13289 if (width) { /* fudge width (can't fudge elen) */
13290 width += elen - old_elen;
13297 /* append esignbuf, filler, zeros, eptr and dotstr to sv */
13300 STRLEN need, have, gap;
13304 /* signed value that's wrapped? */
13305 assert(elen <= ((~(STRLEN)0) >> 1));
13307 /* if zeros is non-zero, then it represents filler between
13308 * elen and precis. So adding elen and zeros together will
13309 * always be <= precis, and the addition can never wrap */
13310 assert(!zeros || (precis > elen && precis - elen == zeros));
13311 have = elen + zeros;
13313 if (have >= (((STRLEN)~0) - esignlen))
13314 croak_memory_wrap();
13317 need = (have > width ? have : width);
13320 if (need >= (((STRLEN)~0) - (SvCUR(sv) + 1)))
13321 croak_memory_wrap();
13322 need += (SvCUR(sv) + 1);
13329 for (i = 0; i < esignlen; i++)
13330 *s++ = esignbuf[i];
13331 for (i = zeros; i; i--)
13333 Copy(eptr, s, elen, char);
13335 for (i = gap; i; i--)
13340 for (i = 0; i < esignlen; i++)
13341 *s++ = esignbuf[i];
13346 for (i = gap; i; i--)
13348 for (i = 0; i < esignlen; i++)
13349 *s++ = esignbuf[i];
13352 for (i = zeros; i; i--)
13354 Copy(eptr, s, elen, char);
13359 SvCUR_set(sv, s - SvPVX_const(sv));
13367 if (vectorize && veclen) {
13368 /* we append the vector separator separately since %v isn't
13369 * very common: don't slow down the general case by adding
13370 * dotstrlen to need etc */
13371 sv_catpvn_nomg(sv, dotstr, dotstrlen);
13373 goto vector; /* do next iteration */
13376 done_valid_conversion:
13379 S_warn_vcatpvfn_missing_argument(aTHX);
13382 /* Now that we've consumed all our printf format arguments (svix)
13383 * do we have things left on the stack that we didn't use?
13385 if (!no_redundant_warning && sv_count >= svix + 1 && ckWARN(WARN_REDUNDANT)) {
13386 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
13387 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
13392 #ifdef USE_LOCALE_NUMERIC
13394 if (lc_numeric_set) {
13395 RESTORE_LC_NUMERIC(); /* Done outside loop, so don't have to
13396 save/restore each iteration. */
13403 /* =========================================================================
13405 =head1 Cloning an interpreter
13409 All the macros and functions in this section are for the private use of
13410 the main function, perl_clone().
13412 The foo_dup() functions make an exact copy of an existing foo thingy.
13413 During the course of a cloning, a hash table is used to map old addresses
13414 to new addresses. The table is created and manipulated with the
13415 ptr_table_* functions.
13417 * =========================================================================*/
13420 #if defined(USE_ITHREADS)
13422 /* XXX Remove this so it doesn't have to go thru the macro and return for nothing */
13423 #ifndef GpREFCNT_inc
13424 # define GpREFCNT_inc(gp) ((gp) ? (++(gp)->gp_refcnt, (gp)) : (GP*)NULL)
13428 /* Certain cases in Perl_ss_dup have been merged, by relying on the fact
13429 that currently av_dup, gv_dup and hv_dup are the same as sv_dup.
13430 If this changes, please unmerge ss_dup.
13431 Likewise, sv_dup_inc_multiple() relies on this fact. */
13432 #define sv_dup_inc_NN(s,t) SvREFCNT_inc_NN(sv_dup_inc(s,t))
13433 #define av_dup(s,t) MUTABLE_AV(sv_dup((const SV *)s,t))
13434 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
13435 #define hv_dup(s,t) MUTABLE_HV(sv_dup((const SV *)s,t))
13436 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
13437 #define cv_dup(s,t) MUTABLE_CV(sv_dup((const SV *)s,t))
13438 #define cv_dup_inc(s,t) MUTABLE_CV(sv_dup_inc((const SV *)s,t))
13439 #define io_dup(s,t) MUTABLE_IO(sv_dup((const SV *)s,t))
13440 #define io_dup_inc(s,t) MUTABLE_IO(sv_dup_inc((const SV *)s,t))
13441 #define gv_dup(s,t) MUTABLE_GV(sv_dup((const SV *)s,t))
13442 #define gv_dup_inc(s,t) MUTABLE_GV(sv_dup_inc((const SV *)s,t))
13443 #define SAVEPV(p) ((p) ? savepv(p) : NULL)
13444 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
13446 /* clone a parser */
13449 Perl_parser_dup(pTHX_ const yy_parser *const proto, CLONE_PARAMS *const param)
13453 PERL_ARGS_ASSERT_PARSER_DUP;
13458 /* look for it in the table first */
13459 parser = (yy_parser *)ptr_table_fetch(PL_ptr_table, proto);
13463 /* create anew and remember what it is */
13464 Newxz(parser, 1, yy_parser);
13465 ptr_table_store(PL_ptr_table, proto, parser);
13467 /* XXX eventually, just Copy() most of the parser struct ? */
13469 parser->lex_brackets = proto->lex_brackets;
13470 parser->lex_casemods = proto->lex_casemods;
13471 parser->lex_brackstack = savepvn(proto->lex_brackstack,
13472 (proto->lex_brackets < 120 ? 120 : proto->lex_brackets));
13473 parser->lex_casestack = savepvn(proto->lex_casestack,
13474 (proto->lex_casemods < 12 ? 12 : proto->lex_casemods));
13475 parser->lex_defer = proto->lex_defer;
13476 parser->lex_dojoin = proto->lex_dojoin;
13477 parser->lex_formbrack = proto->lex_formbrack;
13478 parser->lex_inpat = proto->lex_inpat;
13479 parser->lex_inwhat = proto->lex_inwhat;
13480 parser->lex_op = proto->lex_op;
13481 parser->lex_repl = sv_dup_inc(proto->lex_repl, param);
13482 parser->lex_starts = proto->lex_starts;
13483 parser->lex_stuff = sv_dup_inc(proto->lex_stuff, param);
13484 parser->multi_close = proto->multi_close;
13485 parser->multi_open = proto->multi_open;
13486 parser->multi_start = proto->multi_start;
13487 parser->multi_end = proto->multi_end;
13488 parser->preambled = proto->preambled;
13489 parser->lex_super_state = proto->lex_super_state;
13490 parser->lex_sub_inwhat = proto->lex_sub_inwhat;
13491 parser->lex_sub_op = proto->lex_sub_op;
13492 parser->lex_sub_repl= sv_dup_inc(proto->lex_sub_repl, param);
13493 parser->linestr = sv_dup_inc(proto->linestr, param);
13494 parser->expect = proto->expect;
13495 parser->copline = proto->copline;
13496 parser->last_lop_op = proto->last_lop_op;
13497 parser->lex_state = proto->lex_state;
13498 parser->rsfp = fp_dup(proto->rsfp, '<', param);
13499 /* rsfp_filters entries have fake IoDIRP() */
13500 parser->rsfp_filters= av_dup_inc(proto->rsfp_filters, param);
13501 parser->in_my = proto->in_my;
13502 parser->in_my_stash = hv_dup(proto->in_my_stash, param);
13503 parser->error_count = proto->error_count;
13504 parser->sig_elems = proto->sig_elems;
13505 parser->sig_optelems= proto->sig_optelems;
13506 parser->sig_slurpy = proto->sig_slurpy;
13507 parser->recheck_utf8_validity = proto->recheck_utf8_validity;
13510 char * const ols = SvPVX(proto->linestr);
13511 char * const ls = SvPVX(parser->linestr);
13513 parser->bufptr = ls + (proto->bufptr >= ols ?
13514 proto->bufptr - ols : 0);
13515 parser->oldbufptr = ls + (proto->oldbufptr >= ols ?
13516 proto->oldbufptr - ols : 0);
13517 parser->oldoldbufptr= ls + (proto->oldoldbufptr >= ols ?
13518 proto->oldoldbufptr - ols : 0);
13519 parser->linestart = ls + (proto->linestart >= ols ?
13520 proto->linestart - ols : 0);
13521 parser->last_uni = ls + (proto->last_uni >= ols ?
13522 proto->last_uni - ols : 0);
13523 parser->last_lop = ls + (proto->last_lop >= ols ?
13524 proto->last_lop - ols : 0);
13526 parser->bufend = ls + SvCUR(parser->linestr);
13529 Copy(proto->tokenbuf, parser->tokenbuf, 256, char);
13532 Copy(proto->nextval, parser->nextval, 5, YYSTYPE);
13533 Copy(proto->nexttype, parser->nexttype, 5, I32);
13534 parser->nexttoke = proto->nexttoke;
13536 /* XXX should clone saved_curcop here, but we aren't passed
13537 * proto_perl; so do it in perl_clone_using instead */
13543 /* duplicate a file handle */
13546 Perl_fp_dup(pTHX_ PerlIO *const fp, const char type, CLONE_PARAMS *const param)
13550 PERL_ARGS_ASSERT_FP_DUP;
13551 PERL_UNUSED_ARG(type);
13554 return (PerlIO*)NULL;
13556 /* look for it in the table first */
13557 ret = (PerlIO*)ptr_table_fetch(PL_ptr_table, fp);
13561 /* create anew and remember what it is */
13562 #ifdef __amigaos4__
13563 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE|PERLIO_DUP_FD);
13565 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE);
13567 ptr_table_store(PL_ptr_table, fp, ret);
13571 /* duplicate a directory handle */
13574 Perl_dirp_dup(pTHX_ DIR *const dp, CLONE_PARAMS *const param)
13578 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
13580 const Direntry_t *dirent;
13581 char smallbuf[256]; /* XXX MAXPATHLEN, surely? */
13587 PERL_UNUSED_CONTEXT;
13588 PERL_ARGS_ASSERT_DIRP_DUP;
13593 /* look for it in the table first */
13594 ret = (DIR*)ptr_table_fetch(PL_ptr_table, dp);
13598 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
13600 PERL_UNUSED_ARG(param);
13604 /* open the current directory (so we can switch back) */
13605 if (!(pwd = PerlDir_open("."))) return (DIR *)NULL;
13607 /* chdir to our dir handle and open the present working directory */
13608 if (fchdir(my_dirfd(dp)) < 0 || !(ret = PerlDir_open("."))) {
13609 PerlDir_close(pwd);
13610 return (DIR *)NULL;
13612 /* Now we should have two dir handles pointing to the same dir. */
13614 /* Be nice to the calling code and chdir back to where we were. */
13615 /* XXX If this fails, then what? */
13616 PERL_UNUSED_RESULT(fchdir(my_dirfd(pwd)));
13618 /* We have no need of the pwd handle any more. */
13619 PerlDir_close(pwd);
13622 # define d_namlen(d) (d)->d_namlen
13624 # define d_namlen(d) strlen((d)->d_name)
13626 /* Iterate once through dp, to get the file name at the current posi-
13627 tion. Then step back. */
13628 pos = PerlDir_tell(dp);
13629 if ((dirent = PerlDir_read(dp))) {
13630 len = d_namlen(dirent);
13631 if (len > sizeof(dirent->d_name) && sizeof(dirent->d_name) > PTRSIZE) {
13632 /* If the len is somehow magically longer than the
13633 * maximum length of the directory entry, even though
13634 * we could fit it in a buffer, we could not copy it
13635 * from the dirent. Bail out. */
13636 PerlDir_close(ret);
13639 if (len <= sizeof smallbuf) name = smallbuf;
13640 else Newx(name, len, char);
13641 Move(dirent->d_name, name, len, char);
13643 PerlDir_seek(dp, pos);
13645 /* Iterate through the new dir handle, till we find a file with the
13647 if (!dirent) /* just before the end */
13649 pos = PerlDir_tell(ret);
13650 if (PerlDir_read(ret)) continue; /* not there yet */
13651 PerlDir_seek(ret, pos); /* step back */
13655 const long pos0 = PerlDir_tell(ret);
13657 pos = PerlDir_tell(ret);
13658 if ((dirent = PerlDir_read(ret))) {
13659 if (len == (STRLEN)d_namlen(dirent)
13660 && memEQ(name, dirent->d_name, len)) {
13662 PerlDir_seek(ret, pos); /* step back */
13665 /* else we are not there yet; keep iterating */
13667 else { /* This is not meant to happen. The best we can do is
13668 reset the iterator to the beginning. */
13669 PerlDir_seek(ret, pos0);
13676 if (name && name != smallbuf)
13681 ret = win32_dirp_dup(dp, param);
13684 /* pop it in the pointer table */
13686 ptr_table_store(PL_ptr_table, dp, ret);
13691 /* duplicate a typeglob */
13694 Perl_gp_dup(pTHX_ GP *const gp, CLONE_PARAMS *const param)
13698 PERL_ARGS_ASSERT_GP_DUP;
13702 /* look for it in the table first */
13703 ret = (GP*)ptr_table_fetch(PL_ptr_table, gp);
13707 /* create anew and remember what it is */
13709 ptr_table_store(PL_ptr_table, gp, ret);
13712 /* ret->gp_refcnt must be 0 before any other dups are called. We're relying
13713 on Newxz() to do this for us. */
13714 ret->gp_sv = sv_dup_inc(gp->gp_sv, param);
13715 ret->gp_io = io_dup_inc(gp->gp_io, param);
13716 ret->gp_form = cv_dup_inc(gp->gp_form, param);
13717 ret->gp_av = av_dup_inc(gp->gp_av, param);
13718 ret->gp_hv = hv_dup_inc(gp->gp_hv, param);
13719 ret->gp_egv = gv_dup(gp->gp_egv, param);/* GvEGV is not refcounted */
13720 ret->gp_cv = cv_dup_inc(gp->gp_cv, param);
13721 ret->gp_cvgen = gp->gp_cvgen;
13722 ret->gp_line = gp->gp_line;
13723 ret->gp_file_hek = hek_dup(gp->gp_file_hek, param);
13727 /* duplicate a chain of magic */
13730 Perl_mg_dup(pTHX_ MAGIC *mg, CLONE_PARAMS *const param)
13732 MAGIC *mgret = NULL;
13733 MAGIC **mgprev_p = &mgret;
13735 PERL_ARGS_ASSERT_MG_DUP;
13737 for (; mg; mg = mg->mg_moremagic) {
13740 if ((param->flags & CLONEf_JOIN_IN)
13741 && mg->mg_type == PERL_MAGIC_backref)
13742 /* when joining, we let the individual SVs add themselves to
13743 * backref as needed. */
13746 Newx(nmg, 1, MAGIC);
13748 mgprev_p = &(nmg->mg_moremagic);
13750 /* There was a comment "XXX copy dynamic vtable?" but as we don't have
13751 dynamic vtables, I'm not sure why Sarathy wrote it. The comment dates
13752 from the original commit adding Perl_mg_dup() - revision 4538.
13753 Similarly there is the annotation "XXX random ptr?" next to the
13754 assignment to nmg->mg_ptr. */
13757 /* FIXME for plugins
13758 if (nmg->mg_type == PERL_MAGIC_qr) {
13759 nmg->mg_obj = MUTABLE_SV(CALLREGDUPE((REGEXP*)nmg->mg_obj, param));
13763 nmg->mg_obj = (nmg->mg_flags & MGf_REFCOUNTED)
13764 ? nmg->mg_type == PERL_MAGIC_backref
13765 /* The backref AV has its reference
13766 * count deliberately bumped by 1 */
13767 ? SvREFCNT_inc(av_dup_inc((const AV *)
13768 nmg->mg_obj, param))
13769 : sv_dup_inc(nmg->mg_obj, param)
13770 : (nmg->mg_type == PERL_MAGIC_regdatum ||
13771 nmg->mg_type == PERL_MAGIC_regdata)
13773 : sv_dup(nmg->mg_obj, param);
13775 if (nmg->mg_ptr && nmg->mg_type != PERL_MAGIC_regex_global) {
13776 if (nmg->mg_len > 0) {
13777 nmg->mg_ptr = SAVEPVN(nmg->mg_ptr, nmg->mg_len);
13778 if (nmg->mg_type == PERL_MAGIC_overload_table &&
13779 AMT_AMAGIC((AMT*)nmg->mg_ptr))
13781 AMT * const namtp = (AMT*)nmg->mg_ptr;
13782 sv_dup_inc_multiple((SV**)(namtp->table),
13783 (SV**)(namtp->table), NofAMmeth, param);
13786 else if (nmg->mg_len == HEf_SVKEY)
13787 nmg->mg_ptr = (char*)sv_dup_inc((const SV *)nmg->mg_ptr, param);
13789 if ((nmg->mg_flags & MGf_DUP) && nmg->mg_virtual && nmg->mg_virtual->svt_dup) {
13790 nmg->mg_virtual->svt_dup(aTHX_ nmg, param);
13796 #endif /* USE_ITHREADS */
13798 struct ptr_tbl_arena {
13799 struct ptr_tbl_arena *next;
13800 struct ptr_tbl_ent array[1023/3]; /* as ptr_tbl_ent has 3 pointers. */
13803 /* create a new pointer-mapping table */
13806 Perl_ptr_table_new(pTHX)
13809 PERL_UNUSED_CONTEXT;
13811 Newx(tbl, 1, PTR_TBL_t);
13812 tbl->tbl_max = 511;
13813 tbl->tbl_items = 0;
13814 tbl->tbl_arena = NULL;
13815 tbl->tbl_arena_next = NULL;
13816 tbl->tbl_arena_end = NULL;
13817 Newxz(tbl->tbl_ary, tbl->tbl_max + 1, PTR_TBL_ENT_t*);
13821 #define PTR_TABLE_HASH(ptr) \
13822 ((PTR2UV(ptr) >> 3) ^ (PTR2UV(ptr) >> (3 + 7)) ^ (PTR2UV(ptr) >> (3 + 17)))
13824 /* map an existing pointer using a table */
13826 STATIC PTR_TBL_ENT_t *
13827 S_ptr_table_find(PTR_TBL_t *const tbl, const void *const sv)
13829 PTR_TBL_ENT_t *tblent;
13830 const UV hash = PTR_TABLE_HASH(sv);
13832 PERL_ARGS_ASSERT_PTR_TABLE_FIND;
13834 tblent = tbl->tbl_ary[hash & tbl->tbl_max];
13835 for (; tblent; tblent = tblent->next) {
13836 if (tblent->oldval == sv)
13843 Perl_ptr_table_fetch(pTHX_ PTR_TBL_t *const tbl, const void *const sv)
13845 PTR_TBL_ENT_t const *const tblent = ptr_table_find(tbl, sv);
13847 PERL_ARGS_ASSERT_PTR_TABLE_FETCH;
13848 PERL_UNUSED_CONTEXT;
13850 return tblent ? tblent->newval : NULL;
13853 /* add a new entry to a pointer-mapping table 'tbl'. In hash terms, 'oldsv' is
13854 * the key; 'newsv' is the value. The names "old" and "new" are specific to
13855 * the core's typical use of ptr_tables in thread cloning. */
13858 Perl_ptr_table_store(pTHX_ PTR_TBL_t *const tbl, const void *const oldsv, void *const newsv)
13860 PTR_TBL_ENT_t *tblent = ptr_table_find(tbl, oldsv);
13862 PERL_ARGS_ASSERT_PTR_TABLE_STORE;
13863 PERL_UNUSED_CONTEXT;
13866 tblent->newval = newsv;
13868 const UV entry = PTR_TABLE_HASH(oldsv) & tbl->tbl_max;
13870 if (tbl->tbl_arena_next == tbl->tbl_arena_end) {
13871 struct ptr_tbl_arena *new_arena;
13873 Newx(new_arena, 1, struct ptr_tbl_arena);
13874 new_arena->next = tbl->tbl_arena;
13875 tbl->tbl_arena = new_arena;
13876 tbl->tbl_arena_next = new_arena->array;
13877 tbl->tbl_arena_end = C_ARRAY_END(new_arena->array);
13880 tblent = tbl->tbl_arena_next++;
13882 tblent->oldval = oldsv;
13883 tblent->newval = newsv;
13884 tblent->next = tbl->tbl_ary[entry];
13885 tbl->tbl_ary[entry] = tblent;
13887 if (tblent->next && tbl->tbl_items > tbl->tbl_max)
13888 ptr_table_split(tbl);
13892 /* double the hash bucket size of an existing ptr table */
13895 Perl_ptr_table_split(pTHX_ PTR_TBL_t *const tbl)
13897 PTR_TBL_ENT_t **ary = tbl->tbl_ary;
13898 const UV oldsize = tbl->tbl_max + 1;
13899 UV newsize = oldsize * 2;
13902 PERL_ARGS_ASSERT_PTR_TABLE_SPLIT;
13903 PERL_UNUSED_CONTEXT;
13905 Renew(ary, newsize, PTR_TBL_ENT_t*);
13906 Zero(&ary[oldsize], newsize-oldsize, PTR_TBL_ENT_t*);
13907 tbl->tbl_max = --newsize;
13908 tbl->tbl_ary = ary;
13909 for (i=0; i < oldsize; i++, ary++) {
13910 PTR_TBL_ENT_t **entp = ary;
13911 PTR_TBL_ENT_t *ent = *ary;
13912 PTR_TBL_ENT_t **curentp;
13915 curentp = ary + oldsize;
13917 if ((newsize & PTR_TABLE_HASH(ent->oldval)) != i) {
13919 ent->next = *curentp;
13929 /* remove all the entries from a ptr table */
13930 /* Deprecated - will be removed post 5.14 */
13933 Perl_ptr_table_clear(pTHX_ PTR_TBL_t *const tbl)
13935 PERL_UNUSED_CONTEXT;
13936 if (tbl && tbl->tbl_items) {
13937 struct ptr_tbl_arena *arena = tbl->tbl_arena;
13939 Zero(tbl->tbl_ary, tbl->tbl_max + 1, struct ptr_tbl_ent *);
13942 struct ptr_tbl_arena *next = arena->next;
13948 tbl->tbl_items = 0;
13949 tbl->tbl_arena = NULL;
13950 tbl->tbl_arena_next = NULL;
13951 tbl->tbl_arena_end = NULL;
13955 /* clear and free a ptr table */
13958 Perl_ptr_table_free(pTHX_ PTR_TBL_t *const tbl)
13960 struct ptr_tbl_arena *arena;
13962 PERL_UNUSED_CONTEXT;
13968 arena = tbl->tbl_arena;
13971 struct ptr_tbl_arena *next = arena->next;
13977 Safefree(tbl->tbl_ary);
13981 #if defined(USE_ITHREADS)
13984 Perl_rvpv_dup(pTHX_ SV *const dstr, const SV *const sstr, CLONE_PARAMS *const param)
13986 PERL_ARGS_ASSERT_RVPV_DUP;
13988 assert(!isREGEXP(sstr));
13990 if (SvWEAKREF(sstr)) {
13991 SvRV_set(dstr, sv_dup(SvRV_const(sstr), param));
13992 if (param->flags & CLONEf_JOIN_IN) {
13993 /* if joining, we add any back references individually rather
13994 * than copying the whole backref array */
13995 Perl_sv_add_backref(aTHX_ SvRV(dstr), dstr);
13999 SvRV_set(dstr, sv_dup_inc(SvRV_const(sstr), param));
14001 else if (SvPVX_const(sstr)) {
14002 /* Has something there */
14004 /* Normal PV - clone whole allocated space */
14005 SvPV_set(dstr, SAVEPVN(SvPVX_const(sstr), SvLEN(sstr)-1));
14006 /* sstr may not be that normal, but actually copy on write.
14007 But we are a true, independent SV, so: */
14011 /* Special case - not normally malloced for some reason */
14012 if (isGV_with_GP(sstr)) {
14013 /* Don't need to do anything here. */
14015 else if ((SvIsCOW(sstr))) {
14016 /* A "shared" PV - clone it as "shared" PV */
14018 HEK_KEY(hek_dup(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)),
14022 /* Some other special case - random pointer */
14023 SvPV_set(dstr, (char *) SvPVX_const(sstr));
14028 /* Copy the NULL */
14029 SvPV_set(dstr, NULL);
14033 /* duplicate a list of SVs. source and dest may point to the same memory. */
14035 S_sv_dup_inc_multiple(pTHX_ SV *const *source, SV **dest,
14036 SSize_t items, CLONE_PARAMS *const param)
14038 PERL_ARGS_ASSERT_SV_DUP_INC_MULTIPLE;
14040 while (items-- > 0) {
14041 *dest++ = sv_dup_inc(*source++, param);
14047 /* duplicate an SV of any type (including AV, HV etc) */
14050 S_sv_dup_common(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14055 PERL_ARGS_ASSERT_SV_DUP_COMMON;
14057 if (SvTYPE(sstr) == (svtype)SVTYPEMASK) {
14058 #ifdef DEBUG_LEAKING_SCALARS_ABORT
14063 /* look for it in the table first */
14064 dstr = MUTABLE_SV(ptr_table_fetch(PL_ptr_table, sstr));
14068 if(param->flags & CLONEf_JOIN_IN) {
14069 /** We are joining here so we don't want do clone
14070 something that is bad **/
14071 if (SvTYPE(sstr) == SVt_PVHV) {
14072 const HEK * const hvname = HvNAME_HEK(sstr);
14074 /** don't clone stashes if they already exist **/
14075 dstr = MUTABLE_SV(gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
14076 HEK_UTF8(hvname) ? SVf_UTF8 : 0));
14077 ptr_table_store(PL_ptr_table, sstr, dstr);
14081 else if (SvTYPE(sstr) == SVt_PVGV && !SvFAKE(sstr)) {
14082 HV *stash = GvSTASH(sstr);
14083 const HEK * hvname;
14084 if (stash && (hvname = HvNAME_HEK(stash))) {
14085 /** don't clone GVs if they already exist **/
14087 stash = gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
14088 HEK_UTF8(hvname) ? SVf_UTF8 : 0);
14090 stash, GvNAME(sstr),
14096 if (svp && *svp && SvTYPE(*svp) == SVt_PVGV) {
14097 ptr_table_store(PL_ptr_table, sstr, *svp);
14104 /* create anew and remember what it is */
14107 #ifdef DEBUG_LEAKING_SCALARS
14108 dstr->sv_debug_optype = sstr->sv_debug_optype;
14109 dstr->sv_debug_line = sstr->sv_debug_line;
14110 dstr->sv_debug_inpad = sstr->sv_debug_inpad;
14111 dstr->sv_debug_parent = (SV*)sstr;
14112 FREE_SV_DEBUG_FILE(dstr);
14113 dstr->sv_debug_file = savesharedpv(sstr->sv_debug_file);
14116 ptr_table_store(PL_ptr_table, sstr, dstr);
14119 SvFLAGS(dstr) = SvFLAGS(sstr);
14120 SvFLAGS(dstr) &= ~SVf_OOK; /* don't propagate OOK hack */
14121 SvREFCNT(dstr) = 0; /* must be before any other dups! */
14124 if (SvANY(sstr) && PL_watch_pvx && SvPVX_const(sstr) == PL_watch_pvx)
14125 PerlIO_printf(Perl_debug_log, "watch at %p hit, found string \"%s\"\n",
14126 (void*)PL_watch_pvx, SvPVX_const(sstr));
14129 /* don't clone objects whose class has asked us not to */
14131 && ! (SvFLAGS(SvSTASH(sstr)) & SVphv_CLONEABLE))
14137 switch (SvTYPE(sstr)) {
14139 SvANY(dstr) = NULL;
14142 SET_SVANY_FOR_BODYLESS_IV(dstr);
14144 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
14146 SvIV_set(dstr, SvIVX(sstr));
14150 #if NVSIZE <= IVSIZE
14151 SET_SVANY_FOR_BODYLESS_NV(dstr);
14153 SvANY(dstr) = new_XNV();
14155 SvNV_set(dstr, SvNVX(sstr));
14159 /* These are all the types that need complex bodies allocating. */
14161 const svtype sv_type = SvTYPE(sstr);
14162 const struct body_details *const sv_type_details
14163 = bodies_by_type + sv_type;
14167 Perl_croak(aTHX_ "Bizarre SvTYPE [%" IVdf "]", (IV)SvTYPE(sstr));
14168 NOT_REACHED; /* NOTREACHED */
14184 assert(sv_type_details->body_size);
14185 if (sv_type_details->arena) {
14186 new_body_inline(new_body, sv_type);
14188 = (void*)((char*)new_body - sv_type_details->offset);
14190 new_body = new_NOARENA(sv_type_details);
14194 SvANY(dstr) = new_body;
14197 Copy(((char*)SvANY(sstr)) + sv_type_details->offset,
14198 ((char*)SvANY(dstr)) + sv_type_details->offset,
14199 sv_type_details->copy, char);
14201 Copy(((char*)SvANY(sstr)),
14202 ((char*)SvANY(dstr)),
14203 sv_type_details->body_size + sv_type_details->offset, char);
14206 if (sv_type != SVt_PVAV && sv_type != SVt_PVHV
14207 && !isGV_with_GP(dstr)
14209 && !(sv_type == SVt_PVIO && !(IoFLAGS(dstr) & IOf_FAKE_DIRP)))
14210 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
14212 /* The Copy above means that all the source (unduplicated) pointers
14213 are now in the destination. We can check the flags and the
14214 pointers in either, but it's possible that there's less cache
14215 missing by always going for the destination.
14216 FIXME - instrument and check that assumption */
14217 if (sv_type >= SVt_PVMG) {
14219 SvMAGIC_set(dstr, mg_dup(SvMAGIC(dstr), param));
14220 if (SvOBJECT(dstr) && SvSTASH(dstr))
14221 SvSTASH_set(dstr, hv_dup_inc(SvSTASH(dstr), param));
14222 else SvSTASH_set(dstr, 0); /* don't copy DESTROY cache */
14225 /* The cast silences a GCC warning about unhandled types. */
14226 switch ((int)sv_type) {
14237 /* FIXME for plugins */
14238 re_dup_guts((REGEXP*) sstr, (REGEXP*) dstr, param);
14241 /* XXX LvTARGOFF sometimes holds PMOP* when DEBUGGING */
14242 if (LvTYPE(dstr) == 't') /* for tie: unrefcnted fake (SV**) */
14243 LvTARG(dstr) = dstr;
14244 else if (LvTYPE(dstr) == 'T') /* for tie: fake HE */
14245 LvTARG(dstr) = MUTABLE_SV(he_dup((HE*)LvTARG(dstr), 0, param));
14247 LvTARG(dstr) = sv_dup_inc(LvTARG(dstr), param);
14248 if (isREGEXP(sstr)) goto duprex;
14251 /* non-GP case already handled above */
14252 if(isGV_with_GP(sstr)) {
14253 GvNAME_HEK(dstr) = hek_dup(GvNAME_HEK(dstr), param);
14254 /* Don't call sv_add_backref here as it's going to be
14255 created as part of the magic cloning of the symbol
14256 table--unless this is during a join and the stash
14257 is not actually being cloned. */
14258 /* Danger Will Robinson - GvGP(dstr) isn't initialised
14259 at the point of this comment. */
14260 GvSTASH(dstr) = hv_dup(GvSTASH(dstr), param);
14261 if (param->flags & CLONEf_JOIN_IN)
14262 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
14263 GvGP_set(dstr, gp_dup(GvGP(sstr), param));
14264 (void)GpREFCNT_inc(GvGP(dstr));
14268 /* PL_parser->rsfp_filters entries have fake IoDIRP() */
14269 if(IoFLAGS(dstr) & IOf_FAKE_DIRP) {
14270 /* I have no idea why fake dirp (rsfps)
14271 should be treated differently but otherwise
14272 we end up with leaks -- sky*/
14273 IoTOP_GV(dstr) = gv_dup_inc(IoTOP_GV(dstr), param);
14274 IoFMT_GV(dstr) = gv_dup_inc(IoFMT_GV(dstr), param);
14275 IoBOTTOM_GV(dstr) = gv_dup_inc(IoBOTTOM_GV(dstr), param);
14277 IoTOP_GV(dstr) = gv_dup(IoTOP_GV(dstr), param);
14278 IoFMT_GV(dstr) = gv_dup(IoFMT_GV(dstr), param);
14279 IoBOTTOM_GV(dstr) = gv_dup(IoBOTTOM_GV(dstr), param);
14280 if (IoDIRP(dstr)) {
14281 IoDIRP(dstr) = dirp_dup(IoDIRP(dstr), param);
14284 /* IoDIRP(dstr) is already a copy of IoDIRP(sstr) */
14286 IoIFP(dstr) = fp_dup(IoIFP(sstr), IoTYPE(dstr), param);
14288 if (IoOFP(dstr) == IoIFP(sstr))
14289 IoOFP(dstr) = IoIFP(dstr);
14291 IoOFP(dstr) = fp_dup(IoOFP(dstr), IoTYPE(dstr), param);
14292 IoTOP_NAME(dstr) = SAVEPV(IoTOP_NAME(dstr));
14293 IoFMT_NAME(dstr) = SAVEPV(IoFMT_NAME(dstr));
14294 IoBOTTOM_NAME(dstr) = SAVEPV(IoBOTTOM_NAME(dstr));
14297 /* avoid cloning an empty array */
14298 if (AvARRAY((const AV *)sstr) && AvFILLp((const AV *)sstr) >= 0) {
14299 SV **dst_ary, **src_ary;
14300 SSize_t items = AvFILLp((const AV *)sstr) + 1;
14302 src_ary = AvARRAY((const AV *)sstr);
14303 Newx(dst_ary, AvMAX((const AV *)sstr)+1, SV*);
14304 ptr_table_store(PL_ptr_table, src_ary, dst_ary);
14305 AvARRAY(MUTABLE_AV(dstr)) = dst_ary;
14306 AvALLOC((const AV *)dstr) = dst_ary;
14307 if (AvREAL((const AV *)sstr)) {
14308 dst_ary = sv_dup_inc_multiple(src_ary, dst_ary, items,
14312 while (items-- > 0)
14313 *dst_ary++ = sv_dup(*src_ary++, param);
14315 items = AvMAX((const AV *)sstr) - AvFILLp((const AV *)sstr);
14316 while (items-- > 0) {
14321 AvARRAY(MUTABLE_AV(dstr)) = NULL;
14322 AvALLOC((const AV *)dstr) = (SV**)NULL;
14323 AvMAX( (const AV *)dstr) = -1;
14324 AvFILLp((const AV *)dstr) = -1;
14328 if (HvARRAY((const HV *)sstr)) {
14330 const bool sharekeys = !!HvSHAREKEYS(sstr);
14331 XPVHV * const dxhv = (XPVHV*)SvANY(dstr);
14332 XPVHV * const sxhv = (XPVHV*)SvANY(sstr);
14334 Newx(darray, PERL_HV_ARRAY_ALLOC_BYTES(dxhv->xhv_max+1)
14335 + (SvOOK(sstr) ? sizeof(struct xpvhv_aux) : 0),
14337 HvARRAY(dstr) = (HE**)darray;
14338 while (i <= sxhv->xhv_max) {
14339 const HE * const source = HvARRAY(sstr)[i];
14340 HvARRAY(dstr)[i] = source
14341 ? he_dup(source, sharekeys, param) : 0;
14345 const struct xpvhv_aux * const saux = HvAUX(sstr);
14346 struct xpvhv_aux * const daux = HvAUX(dstr);
14347 /* This flag isn't copied. */
14350 if (saux->xhv_name_count) {
14351 HEK ** const sname = saux->xhv_name_u.xhvnameu_names;
14353 = saux->xhv_name_count < 0
14354 ? -saux->xhv_name_count
14355 : saux->xhv_name_count;
14356 HEK **shekp = sname + count;
14358 Newx(daux->xhv_name_u.xhvnameu_names, count, HEK *);
14359 dhekp = daux->xhv_name_u.xhvnameu_names + count;
14360 while (shekp-- > sname) {
14362 *dhekp = hek_dup(*shekp, param);
14366 daux->xhv_name_u.xhvnameu_name
14367 = hek_dup(saux->xhv_name_u.xhvnameu_name,
14370 daux->xhv_name_count = saux->xhv_name_count;
14372 daux->xhv_aux_flags = saux->xhv_aux_flags;
14373 #ifdef PERL_HASH_RANDOMIZE_KEYS
14374 daux->xhv_rand = saux->xhv_rand;
14375 daux->xhv_last_rand = saux->xhv_last_rand;
14377 daux->xhv_riter = saux->xhv_riter;
14378 daux->xhv_eiter = saux->xhv_eiter
14379 ? he_dup(saux->xhv_eiter,
14380 cBOOL(HvSHAREKEYS(sstr)), param) : 0;
14381 /* backref array needs refcnt=2; see sv_add_backref */
14382 daux->xhv_backreferences =
14383 (param->flags & CLONEf_JOIN_IN)
14384 /* when joining, we let the individual GVs and
14385 * CVs add themselves to backref as
14386 * needed. This avoids pulling in stuff
14387 * that isn't required, and simplifies the
14388 * case where stashes aren't cloned back
14389 * if they already exist in the parent
14392 : saux->xhv_backreferences
14393 ? (SvTYPE(saux->xhv_backreferences) == SVt_PVAV)
14394 ? MUTABLE_AV(SvREFCNT_inc(
14395 sv_dup_inc((const SV *)
14396 saux->xhv_backreferences, param)))
14397 : MUTABLE_AV(sv_dup((const SV *)
14398 saux->xhv_backreferences, param))
14401 daux->xhv_mro_meta = saux->xhv_mro_meta
14402 ? mro_meta_dup(saux->xhv_mro_meta, param)
14405 /* Record stashes for possible cloning in Perl_clone(). */
14407 av_push(param->stashes, dstr);
14411 HvARRAY(MUTABLE_HV(dstr)) = NULL;
14414 if (!(param->flags & CLONEf_COPY_STACKS)) {
14419 /* NOTE: not refcounted */
14420 SvANY(MUTABLE_CV(dstr))->xcv_stash =
14421 hv_dup(CvSTASH(dstr), param);
14422 if ((param->flags & CLONEf_JOIN_IN) && CvSTASH(dstr))
14423 Perl_sv_add_backref(aTHX_ MUTABLE_SV(CvSTASH(dstr)), dstr);
14424 if (!CvISXSUB(dstr)) {
14426 CvROOT(dstr) = OpREFCNT_inc(CvROOT(dstr));
14428 CvSLABBED_off(dstr);
14429 } else if (CvCONST(dstr)) {
14430 CvXSUBANY(dstr).any_ptr =
14431 sv_dup_inc((const SV *)CvXSUBANY(dstr).any_ptr, param);
14433 assert(!CvSLABBED(dstr));
14434 if (CvDYNFILE(dstr)) CvFILE(dstr) = SAVEPV(CvFILE(dstr));
14436 SvANY((CV *)dstr)->xcv_gv_u.xcv_hek =
14437 hek_dup(CvNAME_HEK((CV *)sstr), param);
14438 /* don't dup if copying back - CvGV isn't refcounted, so the
14439 * duped GV may never be freed. A bit of a hack! DAPM */
14441 SvANY(MUTABLE_CV(dstr))->xcv_gv_u.xcv_gv =
14443 ? gv_dup_inc(CvGV(sstr), param)
14444 : (param->flags & CLONEf_JOIN_IN)
14446 : gv_dup(CvGV(sstr), param);
14448 if (!CvISXSUB(sstr)) {
14449 PADLIST * padlist = CvPADLIST(sstr);
14451 padlist = padlist_dup(padlist, param);
14452 CvPADLIST_set(dstr, padlist);
14454 /* unthreaded perl can't sv_dup so we dont support unthreaded's CvHSCXT */
14455 PoisonPADLIST(dstr);
14458 CvWEAKOUTSIDE(sstr)
14459 ? cv_dup( CvOUTSIDE(dstr), param)
14460 : cv_dup_inc(CvOUTSIDE(dstr), param);
14470 Perl_sv_dup_inc(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14472 PERL_ARGS_ASSERT_SV_DUP_INC;
14473 return sstr ? SvREFCNT_inc(sv_dup_common(sstr, param)) : NULL;
14477 Perl_sv_dup(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
14479 SV *dstr = sstr ? sv_dup_common(sstr, param) : NULL;
14480 PERL_ARGS_ASSERT_SV_DUP;
14482 /* Track every SV that (at least initially) had a reference count of 0.
14483 We need to do this by holding an actual reference to it in this array.
14484 If we attempt to cheat, turn AvREAL_off(), and store only pointers
14485 (akin to the stashes hash, and the perl stack), we come unstuck if
14486 a weak reference (or other SV legitimately SvREFCNT() == 0 for this
14487 thread) is manipulated in a CLONE method, because CLONE runs before the
14488 unreferenced array is walked to find SVs still with SvREFCNT() == 0
14489 (and fix things up by giving each a reference via the temps stack).
14490 Instead, during CLONE, if the 0-referenced SV has SvREFCNT_inc() and
14491 then SvREFCNT_dec(), it will be cleaned up (and added to the free list)
14492 before the walk of unreferenced happens and a reference to that is SV
14493 added to the temps stack. At which point we have the same SV considered
14494 to be in use, and free to be re-used. Not good.
14496 if (dstr && !(param->flags & CLONEf_COPY_STACKS) && !SvREFCNT(dstr)) {
14497 assert(param->unreferenced);
14498 av_push(param->unreferenced, SvREFCNT_inc(dstr));
14504 /* duplicate a context */
14507 Perl_cx_dup(pTHX_ PERL_CONTEXT *cxs, I32 ix, I32 max, CLONE_PARAMS* param)
14509 PERL_CONTEXT *ncxs;
14511 PERL_ARGS_ASSERT_CX_DUP;
14514 return (PERL_CONTEXT*)NULL;
14516 /* look for it in the table first */
14517 ncxs = (PERL_CONTEXT*)ptr_table_fetch(PL_ptr_table, cxs);
14521 /* create anew and remember what it is */
14522 Newx(ncxs, max + 1, PERL_CONTEXT);
14523 ptr_table_store(PL_ptr_table, cxs, ncxs);
14524 Copy(cxs, ncxs, max + 1, PERL_CONTEXT);
14527 PERL_CONTEXT * const ncx = &ncxs[ix];
14528 if (CxTYPE(ncx) == CXt_SUBST) {
14529 Perl_croak(aTHX_ "Cloning substitution context is unimplemented");
14532 ncx->blk_oldcop = (COP*)any_dup(ncx->blk_oldcop, param->proto_perl);
14533 switch (CxTYPE(ncx)) {
14535 ncx->blk_sub.cv = cv_dup_inc(ncx->blk_sub.cv, param);
14536 if(CxHASARGS(ncx)){
14537 ncx->blk_sub.savearray = av_dup_inc(ncx->blk_sub.savearray,param);
14539 ncx->blk_sub.savearray = NULL;
14541 ncx->blk_sub.prevcomppad = (PAD*)ptr_table_fetch(PL_ptr_table,
14542 ncx->blk_sub.prevcomppad);
14545 ncx->blk_eval.old_namesv = sv_dup_inc(ncx->blk_eval.old_namesv,
14547 /* XXX should this sv_dup_inc? Or only if CxEVAL_TXT_REFCNTED ???? */
14548 ncx->blk_eval.cur_text = sv_dup(ncx->blk_eval.cur_text, param);
14549 ncx->blk_eval.cv = cv_dup(ncx->blk_eval.cv, param);
14550 /* XXX what do do with cur_top_env ???? */
14552 case CXt_LOOP_LAZYSV:
14553 ncx->blk_loop.state_u.lazysv.end
14554 = sv_dup_inc(ncx->blk_loop.state_u.lazysv.end, param);
14555 /* Fallthrough: duplicate lazysv.cur by using the ary.ary
14556 duplication code instead.
14557 We are taking advantage of (1) av_dup_inc and sv_dup_inc
14558 actually being the same function, and (2) order
14559 equivalence of the two unions.
14560 We can assert the later [but only at run time :-(] */
14561 assert ((void *) &ncx->blk_loop.state_u.ary.ary ==
14562 (void *) &ncx->blk_loop.state_u.lazysv.cur);
14565 ncx->blk_loop.state_u.ary.ary
14566 = av_dup_inc(ncx->blk_loop.state_u.ary.ary, param);
14568 case CXt_LOOP_LIST:
14569 case CXt_LOOP_LAZYIV:
14570 /* code common to all 'for' CXt_LOOP_* types */
14571 ncx->blk_loop.itersave =
14572 sv_dup_inc(ncx->blk_loop.itersave, param);
14573 if (CxPADLOOP(ncx)) {
14574 PADOFFSET off = ncx->blk_loop.itervar_u.svp
14575 - &CX_CURPAD_SV(ncx->blk_loop, 0);
14576 ncx->blk_loop.oldcomppad =
14577 (PAD*)ptr_table_fetch(PL_ptr_table,
14578 ncx->blk_loop.oldcomppad);
14579 ncx->blk_loop.itervar_u.svp =
14580 &CX_CURPAD_SV(ncx->blk_loop, off);
14583 /* this copies the GV if CXp_FOR_GV, or the SV for an
14584 * alias (for \$x (...)) - relies on gv_dup being the
14585 * same as sv_dup */
14586 ncx->blk_loop.itervar_u.gv
14587 = gv_dup((const GV *)ncx->blk_loop.itervar_u.gv,
14591 case CXt_LOOP_PLAIN:
14594 ncx->blk_format.prevcomppad =
14595 (PAD*)ptr_table_fetch(PL_ptr_table,
14596 ncx->blk_format.prevcomppad);
14597 ncx->blk_format.cv = cv_dup_inc(ncx->blk_format.cv, param);
14598 ncx->blk_format.gv = gv_dup(ncx->blk_format.gv, param);
14599 ncx->blk_format.dfoutgv = gv_dup_inc(ncx->blk_format.dfoutgv,
14603 ncx->blk_givwhen.defsv_save =
14604 sv_dup_inc(ncx->blk_givwhen.defsv_save, param);
14617 /* duplicate a stack info structure */
14620 Perl_si_dup(pTHX_ PERL_SI *si, CLONE_PARAMS* param)
14624 PERL_ARGS_ASSERT_SI_DUP;
14627 return (PERL_SI*)NULL;
14629 /* look for it in the table first */
14630 nsi = (PERL_SI*)ptr_table_fetch(PL_ptr_table, si);
14634 /* create anew and remember what it is */
14635 Newx(nsi, 1, PERL_SI);
14636 ptr_table_store(PL_ptr_table, si, nsi);
14638 nsi->si_stack = av_dup_inc(si->si_stack, param);
14639 nsi->si_cxix = si->si_cxix;
14640 nsi->si_cxmax = si->si_cxmax;
14641 nsi->si_cxstack = cx_dup(si->si_cxstack, si->si_cxix, si->si_cxmax, param);
14642 nsi->si_type = si->si_type;
14643 nsi->si_prev = si_dup(si->si_prev, param);
14644 nsi->si_next = si_dup(si->si_next, param);
14645 nsi->si_markoff = si->si_markoff;
14646 #if defined DEBUGGING && !defined DEBUGGING_RE_ONLY
14647 nsi->si_stack_hwm = 0;
14653 #define POPINT(ss,ix) ((ss)[--(ix)].any_i32)
14654 #define TOPINT(ss,ix) ((ss)[ix].any_i32)
14655 #define POPLONG(ss,ix) ((ss)[--(ix)].any_long)
14656 #define TOPLONG(ss,ix) ((ss)[ix].any_long)
14657 #define POPIV(ss,ix) ((ss)[--(ix)].any_iv)
14658 #define TOPIV(ss,ix) ((ss)[ix].any_iv)
14659 #define POPUV(ss,ix) ((ss)[--(ix)].any_uv)
14660 #define TOPUV(ss,ix) ((ss)[ix].any_uv)
14661 #define POPBOOL(ss,ix) ((ss)[--(ix)].any_bool)
14662 #define TOPBOOL(ss,ix) ((ss)[ix].any_bool)
14663 #define POPPTR(ss,ix) ((ss)[--(ix)].any_ptr)
14664 #define TOPPTR(ss,ix) ((ss)[ix].any_ptr)
14665 #define POPDPTR(ss,ix) ((ss)[--(ix)].any_dptr)
14666 #define TOPDPTR(ss,ix) ((ss)[ix].any_dptr)
14667 #define POPDXPTR(ss,ix) ((ss)[--(ix)].any_dxptr)
14668 #define TOPDXPTR(ss,ix) ((ss)[ix].any_dxptr)
14671 #define pv_dup_inc(p) SAVEPV(p)
14672 #define pv_dup(p) SAVEPV(p)
14673 #define svp_dup_inc(p,pp) any_dup(p,pp)
14675 /* map any object to the new equivent - either something in the
14676 * ptr table, or something in the interpreter structure
14680 Perl_any_dup(pTHX_ void *v, const PerlInterpreter *proto_perl)
14684 PERL_ARGS_ASSERT_ANY_DUP;
14687 return (void*)NULL;
14689 /* look for it in the table first */
14690 ret = ptr_table_fetch(PL_ptr_table, v);
14694 /* see if it is part of the interpreter structure */
14695 if (v >= (void*)proto_perl && v < (void*)(proto_perl+1))
14696 ret = (void*)(((char*)aTHX) + (((char*)v) - (char*)proto_perl));
14704 /* duplicate the save stack */
14707 Perl_ss_dup(pTHX_ PerlInterpreter *proto_perl, CLONE_PARAMS* param)
14710 ANY * const ss = proto_perl->Isavestack;
14711 const I32 max = proto_perl->Isavestack_max + SS_MAXPUSH;
14712 I32 ix = proto_perl->Isavestack_ix;
14725 void (*dptr) (void*);
14726 void (*dxptr) (pTHX_ void*);
14728 PERL_ARGS_ASSERT_SS_DUP;
14730 Newx(nss, max, ANY);
14733 const UV uv = POPUV(ss,ix);
14734 const U8 type = (U8)uv & SAVE_MASK;
14736 TOPUV(nss,ix) = uv;
14738 case SAVEt_CLEARSV:
14739 case SAVEt_CLEARPADRANGE:
14741 case SAVEt_HELEM: /* hash element */
14742 case SAVEt_SV: /* scalar reference */
14743 sv = (const SV *)POPPTR(ss,ix);
14744 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14746 case SAVEt_ITEM: /* normal string */
14747 case SAVEt_GVSV: /* scalar slot in GV */
14748 sv = (const SV *)POPPTR(ss,ix);
14749 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14750 if (type == SAVEt_SV)
14754 case SAVEt_MORTALIZESV:
14755 case SAVEt_READONLY_OFF:
14756 sv = (const SV *)POPPTR(ss,ix);
14757 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14759 case SAVEt_FREEPADNAME:
14760 ptr = POPPTR(ss,ix);
14761 TOPPTR(nss,ix) = padname_dup((PADNAME *)ptr, param);
14762 PadnameREFCNT((PADNAME *)TOPPTR(nss,ix))++;
14764 case SAVEt_SHARED_PVREF: /* char* in shared space */
14765 c = (char*)POPPTR(ss,ix);
14766 TOPPTR(nss,ix) = savesharedpv(c);
14767 ptr = POPPTR(ss,ix);
14768 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14770 case SAVEt_GENERIC_SVREF: /* generic sv */
14771 case SAVEt_SVREF: /* scalar reference */
14772 sv = (const SV *)POPPTR(ss,ix);
14773 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14774 if (type == SAVEt_SVREF)
14775 SvREFCNT_inc_simple_void((SV *)TOPPTR(nss,ix));
14776 ptr = POPPTR(ss,ix);
14777 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14779 case SAVEt_GVSLOT: /* any slot in GV */
14780 sv = (const SV *)POPPTR(ss,ix);
14781 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14782 ptr = POPPTR(ss,ix);
14783 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
14784 sv = (const SV *)POPPTR(ss,ix);
14785 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14787 case SAVEt_HV: /* hash reference */
14788 case SAVEt_AV: /* array reference */
14789 sv = (const SV *) POPPTR(ss,ix);
14790 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14792 case SAVEt_COMPPAD:
14794 sv = (const SV *) POPPTR(ss,ix);
14795 TOPPTR(nss,ix) = sv_dup(sv, param);
14797 case SAVEt_INT: /* int reference */
14798 ptr = POPPTR(ss,ix);
14799 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14800 intval = (int)POPINT(ss,ix);
14801 TOPINT(nss,ix) = intval;
14803 case SAVEt_LONG: /* long reference */
14804 ptr = POPPTR(ss,ix);
14805 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14806 longval = (long)POPLONG(ss,ix);
14807 TOPLONG(nss,ix) = longval;
14809 case SAVEt_I32: /* I32 reference */
14810 ptr = POPPTR(ss,ix);
14811 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14813 TOPINT(nss,ix) = i;
14815 case SAVEt_IV: /* IV reference */
14816 case SAVEt_STRLEN: /* STRLEN/size_t ref */
14817 ptr = POPPTR(ss,ix);
14818 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14820 TOPIV(nss,ix) = iv;
14822 case SAVEt_TMPSFLOOR:
14824 TOPIV(nss,ix) = iv;
14826 case SAVEt_HPTR: /* HV* reference */
14827 case SAVEt_APTR: /* AV* reference */
14828 case SAVEt_SPTR: /* SV* reference */
14829 ptr = POPPTR(ss,ix);
14830 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14831 sv = (const SV *)POPPTR(ss,ix);
14832 TOPPTR(nss,ix) = sv_dup(sv, param);
14834 case SAVEt_VPTR: /* random* reference */
14835 ptr = POPPTR(ss,ix);
14836 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14838 case SAVEt_INT_SMALL:
14839 case SAVEt_I32_SMALL:
14840 case SAVEt_I16: /* I16 reference */
14841 case SAVEt_I8: /* I8 reference */
14843 ptr = POPPTR(ss,ix);
14844 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14846 case SAVEt_GENERIC_PVREF: /* generic char* */
14847 case SAVEt_PPTR: /* char* reference */
14848 ptr = POPPTR(ss,ix);
14849 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14850 c = (char*)POPPTR(ss,ix);
14851 TOPPTR(nss,ix) = pv_dup(c);
14853 case SAVEt_GP: /* scalar reference */
14854 gp = (GP*)POPPTR(ss,ix);
14855 TOPPTR(nss,ix) = gp = gp_dup(gp, param);
14856 (void)GpREFCNT_inc(gp);
14857 gv = (const GV *)POPPTR(ss,ix);
14858 TOPPTR(nss,ix) = gv_dup_inc(gv, param);
14861 ptr = POPPTR(ss,ix);
14862 if (ptr && (((OP*)ptr)->op_private & OPpREFCOUNTED)) {
14863 /* these are assumed to be refcounted properly */
14865 switch (((OP*)ptr)->op_type) {
14867 case OP_LEAVESUBLV:
14871 case OP_LEAVEWRITE:
14872 TOPPTR(nss,ix) = ptr;
14875 (void) OpREFCNT_inc(o);
14879 TOPPTR(nss,ix) = NULL;
14884 TOPPTR(nss,ix) = NULL;
14886 case SAVEt_FREECOPHH:
14887 ptr = POPPTR(ss,ix);
14888 TOPPTR(nss,ix) = cophh_copy((COPHH *)ptr);
14890 case SAVEt_ADELETE:
14891 av = (const AV *)POPPTR(ss,ix);
14892 TOPPTR(nss,ix) = av_dup_inc(av, param);
14894 TOPINT(nss,ix) = i;
14897 hv = (const HV *)POPPTR(ss,ix);
14898 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14900 TOPINT(nss,ix) = i;
14903 c = (char*)POPPTR(ss,ix);
14904 TOPPTR(nss,ix) = pv_dup_inc(c);
14906 case SAVEt_STACK_POS: /* Position on Perl stack */
14908 TOPINT(nss,ix) = i;
14910 case SAVEt_DESTRUCTOR:
14911 ptr = POPPTR(ss,ix);
14912 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14913 dptr = POPDPTR(ss,ix);
14914 TOPDPTR(nss,ix) = DPTR2FPTR(void (*)(void*),
14915 any_dup(FPTR2DPTR(void *, dptr),
14918 case SAVEt_DESTRUCTOR_X:
14919 ptr = POPPTR(ss,ix);
14920 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
14921 dxptr = POPDXPTR(ss,ix);
14922 TOPDXPTR(nss,ix) = DPTR2FPTR(void (*)(pTHX_ void*),
14923 any_dup(FPTR2DPTR(void *, dxptr),
14926 case SAVEt_REGCONTEXT:
14928 ix -= uv >> SAVE_TIGHT_SHIFT;
14930 case SAVEt_AELEM: /* array element */
14931 sv = (const SV *)POPPTR(ss,ix);
14932 TOPPTR(nss,ix) = SvREFCNT_inc(sv_dup_inc(sv, param));
14934 TOPIV(nss,ix) = iv;
14935 av = (const AV *)POPPTR(ss,ix);
14936 TOPPTR(nss,ix) = av_dup_inc(av, param);
14939 ptr = POPPTR(ss,ix);
14940 TOPPTR(nss,ix) = ptr;
14943 ptr = POPPTR(ss,ix);
14944 ptr = cophh_copy((COPHH*)ptr);
14945 TOPPTR(nss,ix) = ptr;
14947 TOPINT(nss,ix) = i;
14948 if (i & HINT_LOCALIZE_HH) {
14949 hv = (const HV *)POPPTR(ss,ix);
14950 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
14953 case SAVEt_PADSV_AND_MORTALIZE:
14954 longval = (long)POPLONG(ss,ix);
14955 TOPLONG(nss,ix) = longval;
14956 ptr = POPPTR(ss,ix);
14957 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
14958 sv = (const SV *)POPPTR(ss,ix);
14959 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
14961 case SAVEt_SET_SVFLAGS:
14963 TOPINT(nss,ix) = i;
14965 TOPINT(nss,ix) = i;
14966 sv = (const SV *)POPPTR(ss,ix);
14967 TOPPTR(nss,ix) = sv_dup(sv, param);
14969 case SAVEt_COMPILE_WARNINGS:
14970 ptr = POPPTR(ss,ix);
14971 TOPPTR(nss,ix) = DUP_WARNINGS((STRLEN*)ptr);
14974 ptr = POPPTR(ss,ix);
14975 TOPPTR(nss,ix) = parser_dup((const yy_parser*)ptr, param);
14979 "panic: ss_dup inconsistency (%" IVdf ")", (IV) type);
14987 /* if sv is a stash, call $class->CLONE_SKIP(), and set the SVphv_CLONEABLE
14988 * flag to the result. This is done for each stash before cloning starts,
14989 * so we know which stashes want their objects cloned */
14992 do_mark_cloneable_stash(pTHX_ SV *const sv)
14994 const HEK * const hvname = HvNAME_HEK((const HV *)sv);
14996 GV* const cloner = gv_fetchmethod_autoload(MUTABLE_HV(sv), "CLONE_SKIP", 0);
14997 SvFLAGS(sv) |= SVphv_CLONEABLE; /* clone objects by default */
14998 if (cloner && GvCV(cloner)) {
15005 mXPUSHs(newSVhek(hvname));
15007 call_sv(MUTABLE_SV(GvCV(cloner)), G_SCALAR);
15014 SvFLAGS(sv) &= ~SVphv_CLONEABLE;
15022 =for apidoc perl_clone
15024 Create and return a new interpreter by cloning the current one.
15026 C<perl_clone> takes these flags as parameters:
15028 C<CLONEf_COPY_STACKS> - is used to, well, copy the stacks also,
15029 without it we only clone the data and zero the stacks,
15030 with it we copy the stacks and the new perl interpreter is
15031 ready to run at the exact same point as the previous one.
15032 The pseudo-fork code uses C<COPY_STACKS> while the
15033 threads->create doesn't.
15035 C<CLONEf_KEEP_PTR_TABLE> -
15036 C<perl_clone> keeps a ptr_table with the pointer of the old
15037 variable as a key and the new variable as a value,
15038 this allows it to check if something has been cloned and not
15039 clone it again but rather just use the value and increase the
15040 refcount. If C<KEEP_PTR_TABLE> is not set then C<perl_clone> will kill
15041 the ptr_table using the function
15042 C<ptr_table_free(PL_ptr_table); PL_ptr_table = NULL;>,
15043 reason to keep it around is if you want to dup some of your own
15044 variable who are outside the graph perl scans, an example of this
15045 code is in F<threads.xs> create.
15047 C<CLONEf_CLONE_HOST> -
15048 This is a win32 thing, it is ignored on unix, it tells perls
15049 win32host code (which is c++) to clone itself, this is needed on
15050 win32 if you want to run two threads at the same time,
15051 if you just want to do some stuff in a separate perl interpreter
15052 and then throw it away and return to the original one,
15053 you don't need to do anything.
15058 /* XXX the above needs expanding by someone who actually understands it ! */
15059 EXTERN_C PerlInterpreter *
15060 perl_clone_host(PerlInterpreter* proto_perl, UV flags);
15063 perl_clone(PerlInterpreter *proto_perl, UV flags)
15066 #ifdef PERL_IMPLICIT_SYS
15068 PERL_ARGS_ASSERT_PERL_CLONE;
15070 /* perlhost.h so we need to call into it
15071 to clone the host, CPerlHost should have a c interface, sky */
15073 #ifndef __amigaos4__
15074 if (flags & CLONEf_CLONE_HOST) {
15075 return perl_clone_host(proto_perl,flags);
15078 return perl_clone_using(proto_perl, flags,
15080 proto_perl->IMemShared,
15081 proto_perl->IMemParse,
15083 proto_perl->IStdIO,
15087 proto_perl->IProc);
15091 perl_clone_using(PerlInterpreter *proto_perl, UV flags,
15092 struct IPerlMem* ipM, struct IPerlMem* ipMS,
15093 struct IPerlMem* ipMP, struct IPerlEnv* ipE,
15094 struct IPerlStdIO* ipStd, struct IPerlLIO* ipLIO,
15095 struct IPerlDir* ipD, struct IPerlSock* ipS,
15096 struct IPerlProc* ipP)
15098 /* XXX many of the string copies here can be optimized if they're
15099 * constants; they need to be allocated as common memory and just
15100 * their pointers copied. */
15103 CLONE_PARAMS clone_params;
15104 CLONE_PARAMS* const param = &clone_params;
15106 PerlInterpreter * const my_perl = (PerlInterpreter*)(*ipM->pMalloc)(ipM, sizeof(PerlInterpreter));
15108 PERL_ARGS_ASSERT_PERL_CLONE_USING;
15109 #else /* !PERL_IMPLICIT_SYS */
15111 CLONE_PARAMS clone_params;
15112 CLONE_PARAMS* param = &clone_params;
15113 PerlInterpreter * const my_perl = (PerlInterpreter*)PerlMem_malloc(sizeof(PerlInterpreter));
15115 PERL_ARGS_ASSERT_PERL_CLONE;
15116 #endif /* PERL_IMPLICIT_SYS */
15118 /* for each stash, determine whether its objects should be cloned */
15119 S_visit(proto_perl, do_mark_cloneable_stash, SVt_PVHV, SVTYPEMASK);
15120 PERL_SET_THX(my_perl);
15123 PoisonNew(my_perl, 1, PerlInterpreter);
15126 PL_defstash = NULL; /* may be used by perl malloc() */
15129 PL_scopestack_name = 0;
15131 PL_savestack_ix = 0;
15132 PL_savestack_max = -1;
15133 PL_sig_pending = 0;
15135 Zero(&PL_debug_pad, 1, struct perl_debug_pad);
15136 Zero(&PL_padname_undef, 1, PADNAME);
15137 Zero(&PL_padname_const, 1, PADNAME);
15138 # ifdef DEBUG_LEAKING_SCALARS
15139 PL_sv_serial = (((UV)my_perl >> 2) & 0xfff) * 1000000;
15141 # ifdef PERL_TRACE_OPS
15142 Zero(PL_op_exec_cnt, OP_max+2, UV);
15144 #else /* !DEBUGGING */
15145 Zero(my_perl, 1, PerlInterpreter);
15146 #endif /* DEBUGGING */
15148 #ifdef PERL_IMPLICIT_SYS
15149 /* host pointers */
15151 PL_MemShared = ipMS;
15152 PL_MemParse = ipMP;
15159 #endif /* PERL_IMPLICIT_SYS */
15162 param->flags = flags;
15163 /* Nothing in the core code uses this, but we make it available to
15164 extensions (using mg_dup). */
15165 param->proto_perl = proto_perl;
15166 /* Likely nothing will use this, but it is initialised to be consistent
15167 with Perl_clone_params_new(). */
15168 param->new_perl = my_perl;
15169 param->unreferenced = NULL;
15172 INIT_TRACK_MEMPOOL(my_perl->Imemory_debug_header, my_perl);
15174 PL_body_arenas = NULL;
15175 Zero(&PL_body_roots, 1, PL_body_roots);
15179 PL_sv_arenaroot = NULL;
15181 PL_debug = proto_perl->Idebug;
15183 /* dbargs array probably holds garbage */
15186 PL_compiling = proto_perl->Icompiling;
15188 /* pseudo environmental stuff */
15189 PL_origargc = proto_perl->Iorigargc;
15190 PL_origargv = proto_perl->Iorigargv;
15192 #ifndef NO_TAINT_SUPPORT
15193 /* Set tainting stuff before PerlIO_debug can possibly get called */
15194 PL_tainting = proto_perl->Itainting;
15195 PL_taint_warn = proto_perl->Itaint_warn;
15197 PL_tainting = FALSE;
15198 PL_taint_warn = FALSE;
15201 PL_minus_c = proto_perl->Iminus_c;
15203 PL_localpatches = proto_perl->Ilocalpatches;
15204 PL_splitstr = proto_perl->Isplitstr;
15205 PL_minus_n = proto_perl->Iminus_n;
15206 PL_minus_p = proto_perl->Iminus_p;
15207 PL_minus_l = proto_perl->Iminus_l;
15208 PL_minus_a = proto_perl->Iminus_a;
15209 PL_minus_E = proto_perl->Iminus_E;
15210 PL_minus_F = proto_perl->Iminus_F;
15211 PL_doswitches = proto_perl->Idoswitches;
15212 PL_dowarn = proto_perl->Idowarn;
15213 #ifdef PERL_SAWAMPERSAND
15214 PL_sawampersand = proto_perl->Isawampersand;
15216 PL_unsafe = proto_perl->Iunsafe;
15217 PL_perldb = proto_perl->Iperldb;
15218 PL_perl_destruct_level = proto_perl->Iperl_destruct_level;
15219 PL_exit_flags = proto_perl->Iexit_flags;
15221 /* XXX time(&PL_basetime) when asked for? */
15222 PL_basetime = proto_perl->Ibasetime;
15224 PL_maxsysfd = proto_perl->Imaxsysfd;
15225 PL_statusvalue = proto_perl->Istatusvalue;
15227 PL_statusvalue_vms = proto_perl->Istatusvalue_vms;
15229 PL_statusvalue_posix = proto_perl->Istatusvalue_posix;
15232 /* RE engine related */
15233 PL_regmatch_slab = NULL;
15234 PL_reg_curpm = NULL;
15236 PL_sub_generation = proto_perl->Isub_generation;
15238 /* funky return mechanisms */
15239 PL_forkprocess = proto_perl->Iforkprocess;
15241 /* internal state */
15242 PL_main_start = proto_perl->Imain_start;
15243 PL_eval_root = proto_perl->Ieval_root;
15244 PL_eval_start = proto_perl->Ieval_start;
15246 PL_filemode = proto_perl->Ifilemode;
15247 PL_lastfd = proto_perl->Ilastfd;
15248 PL_oldname = proto_perl->Ioldname; /* XXX not quite right */
15249 PL_gensym = proto_perl->Igensym;
15251 PL_laststatval = proto_perl->Ilaststatval;
15252 PL_laststype = proto_perl->Ilaststype;
15255 PL_profiledata = NULL;
15257 PL_generation = proto_perl->Igeneration;
15259 PL_in_clean_objs = proto_perl->Iin_clean_objs;
15260 PL_in_clean_all = proto_perl->Iin_clean_all;
15262 PL_delaymagic_uid = proto_perl->Idelaymagic_uid;
15263 PL_delaymagic_euid = proto_perl->Idelaymagic_euid;
15264 PL_delaymagic_gid = proto_perl->Idelaymagic_gid;
15265 PL_delaymagic_egid = proto_perl->Idelaymagic_egid;
15266 PL_nomemok = proto_perl->Inomemok;
15267 PL_an = proto_perl->Ian;
15268 PL_evalseq = proto_perl->Ievalseq;
15269 PL_origenviron = proto_perl->Iorigenviron; /* XXX not quite right */
15270 PL_origalen = proto_perl->Iorigalen;
15272 PL_sighandlerp = proto_perl->Isighandlerp;
15274 PL_runops = proto_perl->Irunops;
15276 PL_subline = proto_perl->Isubline;
15278 PL_cv_has_eval = proto_perl->Icv_has_eval;
15281 PL_cryptseen = proto_perl->Icryptseen;
15284 #ifdef USE_LOCALE_COLLATE
15285 PL_collation_ix = proto_perl->Icollation_ix;
15286 PL_collation_standard = proto_perl->Icollation_standard;
15287 PL_collxfrm_base = proto_perl->Icollxfrm_base;
15288 PL_collxfrm_mult = proto_perl->Icollxfrm_mult;
15289 PL_strxfrm_max_cp = proto_perl->Istrxfrm_max_cp;
15290 #endif /* USE_LOCALE_COLLATE */
15292 #ifdef USE_LOCALE_NUMERIC
15293 PL_numeric_standard = proto_perl->Inumeric_standard;
15294 PL_numeric_underlying = proto_perl->Inumeric_underlying;
15295 PL_numeric_underlying_is_standard = proto_perl->Inumeric_underlying_is_standard;
15296 #endif /* !USE_LOCALE_NUMERIC */
15298 /* Did the locale setup indicate UTF-8? */
15299 PL_utf8locale = proto_perl->Iutf8locale;
15300 PL_in_utf8_CTYPE_locale = proto_perl->Iin_utf8_CTYPE_locale;
15301 PL_in_utf8_COLLATE_locale = proto_perl->Iin_utf8_COLLATE_locale;
15302 my_strlcpy(PL_locale_utf8ness, proto_perl->Ilocale_utf8ness, sizeof(PL_locale_utf8ness));
15303 #if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
15304 PL_lc_numeric_mutex_depth = 0;
15306 /* Unicode features (see perlrun/-C) */
15307 PL_unicode = proto_perl->Iunicode;
15309 /* Pre-5.8 signals control */
15310 PL_signals = proto_perl->Isignals;
15312 /* times() ticks per second */
15313 PL_clocktick = proto_perl->Iclocktick;
15315 /* Recursion stopper for PerlIO_find_layer */
15316 PL_in_load_module = proto_perl->Iin_load_module;
15318 /* sort() routine */
15319 PL_sort_RealCmp = proto_perl->Isort_RealCmp;
15321 /* Not really needed/useful since the reenrant_retint is "volatile",
15322 * but do it for consistency's sake. */
15323 PL_reentrant_retint = proto_perl->Ireentrant_retint;
15325 /* Hooks to shared SVs and locks. */
15326 PL_sharehook = proto_perl->Isharehook;
15327 PL_lockhook = proto_perl->Ilockhook;
15328 PL_unlockhook = proto_perl->Iunlockhook;
15329 PL_threadhook = proto_perl->Ithreadhook;
15330 PL_destroyhook = proto_perl->Idestroyhook;
15331 PL_signalhook = proto_perl->Isignalhook;
15333 PL_globhook = proto_perl->Iglobhook;
15336 PL_last_swash_hv = NULL; /* reinits on demand */
15337 PL_last_swash_klen = 0;
15338 PL_last_swash_key[0]= '\0';
15339 PL_last_swash_tmps = (U8*)NULL;
15340 PL_last_swash_slen = 0;
15342 PL_srand_called = proto_perl->Isrand_called;
15343 Copy(&(proto_perl->Irandom_state), &PL_random_state, 1, PL_RANDOM_STATE_TYPE);
15345 if (flags & CLONEf_COPY_STACKS) {
15346 /* next allocation will be PL_tmps_stack[PL_tmps_ix+1] */
15347 PL_tmps_ix = proto_perl->Itmps_ix;
15348 PL_tmps_max = proto_perl->Itmps_max;
15349 PL_tmps_floor = proto_perl->Itmps_floor;
15351 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
15352 * NOTE: unlike the others! */
15353 PL_scopestack_ix = proto_perl->Iscopestack_ix;
15354 PL_scopestack_max = proto_perl->Iscopestack_max;
15356 /* next SSPUSHFOO() sets PL_savestack[PL_savestack_ix]
15357 * NOTE: unlike the others! */
15358 PL_savestack_ix = proto_perl->Isavestack_ix;
15359 PL_savestack_max = proto_perl->Isavestack_max;
15362 PL_start_env = proto_perl->Istart_env; /* XXXXXX */
15363 PL_top_env = &PL_start_env;
15365 PL_op = proto_perl->Iop;
15368 PL_Xpv = (XPV*)NULL;
15369 my_perl->Ina = proto_perl->Ina;
15371 PL_statcache = proto_perl->Istatcache;
15373 #ifndef NO_TAINT_SUPPORT
15374 PL_tainted = proto_perl->Itainted;
15376 PL_tainted = FALSE;
15378 PL_curpm = proto_perl->Icurpm; /* XXX No PMOP ref count */
15380 PL_chopset = proto_perl->Ichopset; /* XXX never deallocated */
15382 PL_restartjmpenv = proto_perl->Irestartjmpenv;
15383 PL_restartop = proto_perl->Irestartop;
15384 PL_in_eval = proto_perl->Iin_eval;
15385 PL_delaymagic = proto_perl->Idelaymagic;
15386 PL_phase = proto_perl->Iphase;
15387 PL_localizing = proto_perl->Ilocalizing;
15389 PL_hv_fetch_ent_mh = NULL;
15390 PL_modcount = proto_perl->Imodcount;
15391 PL_lastgotoprobe = NULL;
15392 PL_dumpindent = proto_perl->Idumpindent;
15394 PL_efloatbuf = NULL; /* reinits on demand */
15395 PL_efloatsize = 0; /* reinits on demand */
15399 PL_colorset = 0; /* reinits PL_colors[] */
15400 /*PL_colors[6] = {0,0,0,0,0,0};*/
15402 /* Pluggable optimizer */
15403 PL_peepp = proto_perl->Ipeepp;
15404 PL_rpeepp = proto_perl->Irpeepp;
15405 /* op_free() hook */
15406 PL_opfreehook = proto_perl->Iopfreehook;
15408 #ifdef USE_REENTRANT_API
15409 /* XXX: things like -Dm will segfault here in perlio, but doing
15410 * PERL_SET_CONTEXT(proto_perl);
15411 * breaks too many other things
15413 Perl_reentrant_init(aTHX);
15416 /* create SV map for pointer relocation */
15417 PL_ptr_table = ptr_table_new();
15419 /* initialize these special pointers as early as possible */
15421 ptr_table_store(PL_ptr_table, &proto_perl->Isv_undef, &PL_sv_undef);
15422 ptr_table_store(PL_ptr_table, &proto_perl->Isv_no, &PL_sv_no);
15423 ptr_table_store(PL_ptr_table, &proto_perl->Isv_zero, &PL_sv_zero);
15424 ptr_table_store(PL_ptr_table, &proto_perl->Isv_yes, &PL_sv_yes);
15425 ptr_table_store(PL_ptr_table, &proto_perl->Ipadname_const,
15426 &PL_padname_const);
15428 /* create (a non-shared!) shared string table */
15429 PL_strtab = newHV();
15430 HvSHAREKEYS_off(PL_strtab);
15431 hv_ksplit(PL_strtab, HvTOTALKEYS(proto_perl->Istrtab));
15432 ptr_table_store(PL_ptr_table, proto_perl->Istrtab, PL_strtab);
15434 Zero(PL_sv_consts, SV_CONSTS_COUNT, SV*);
15436 /* This PV will be free'd special way so must set it same way op.c does */
15437 PL_compiling.cop_file = savesharedpv(PL_compiling.cop_file);
15438 ptr_table_store(PL_ptr_table, proto_perl->Icompiling.cop_file, PL_compiling.cop_file);
15440 ptr_table_store(PL_ptr_table, &proto_perl->Icompiling, &PL_compiling);
15441 PL_compiling.cop_warnings = DUP_WARNINGS(PL_compiling.cop_warnings);
15442 CopHINTHASH_set(&PL_compiling, cophh_copy(CopHINTHASH_get(&PL_compiling)));
15443 PL_curcop = (COP*)any_dup(proto_perl->Icurcop, proto_perl);
15445 param->stashes = newAV(); /* Setup array of objects to call clone on */
15446 /* This makes no difference to the implementation, as it always pushes
15447 and shifts pointers to other SVs without changing their reference
15448 count, with the array becoming empty before it is freed. However, it
15449 makes it conceptually clear what is going on, and will avoid some
15450 work inside av.c, filling slots between AvFILL() and AvMAX() with
15451 &PL_sv_undef, and SvREFCNT_dec()ing those. */
15452 AvREAL_off(param->stashes);
15454 if (!(flags & CLONEf_COPY_STACKS)) {
15455 param->unreferenced = newAV();
15458 #ifdef PERLIO_LAYERS
15459 /* Clone PerlIO tables as soon as we can handle general xx_dup() */
15460 PerlIO_clone(aTHX_ proto_perl, param);
15463 PL_envgv = gv_dup_inc(proto_perl->Ienvgv, param);
15464 PL_incgv = gv_dup_inc(proto_perl->Iincgv, param);
15465 PL_hintgv = gv_dup_inc(proto_perl->Ihintgv, param);
15466 PL_origfilename = SAVEPV(proto_perl->Iorigfilename);
15467 PL_xsubfilename = proto_perl->Ixsubfilename;
15468 PL_diehook = sv_dup_inc(proto_perl->Idiehook, param);
15469 PL_warnhook = sv_dup_inc(proto_perl->Iwarnhook, param);
15472 PL_patchlevel = sv_dup_inc(proto_perl->Ipatchlevel, param);
15473 PL_inplace = SAVEPV(proto_perl->Iinplace);
15474 PL_e_script = sv_dup_inc(proto_perl->Ie_script, param);
15476 /* magical thingies */
15478 SvPVCLEAR(PERL_DEBUG_PAD(0)); /* For regex debugging. */
15479 SvPVCLEAR(PERL_DEBUG_PAD(1)); /* ext/re needs these */
15480 SvPVCLEAR(PERL_DEBUG_PAD(2)); /* even without DEBUGGING. */
15483 /* Clone the regex array */
15484 /* ORANGE FIXME for plugins, probably in the SV dup code.
15485 newSViv(PTR2IV(CALLREGDUPE(
15486 INT2PTR(REGEXP *, SvIVX(regex)), param))))
15488 PL_regex_padav = av_dup_inc(proto_perl->Iregex_padav, param);
15489 PL_regex_pad = AvARRAY(PL_regex_padav);
15491 PL_stashpadmax = proto_perl->Istashpadmax;
15492 PL_stashpadix = proto_perl->Istashpadix ;
15493 Newx(PL_stashpad, PL_stashpadmax, HV *);
15496 for (; o < PL_stashpadmax; ++o)
15497 PL_stashpad[o] = hv_dup(proto_perl->Istashpad[o], param);
15500 /* shortcuts to various I/O objects */
15501 PL_ofsgv = gv_dup_inc(proto_perl->Iofsgv, param);
15502 PL_stdingv = gv_dup(proto_perl->Istdingv, param);
15503 PL_stderrgv = gv_dup(proto_perl->Istderrgv, param);
15504 PL_defgv = gv_dup(proto_perl->Idefgv, param);
15505 PL_argvgv = gv_dup_inc(proto_perl->Iargvgv, param);
15506 PL_argvoutgv = gv_dup(proto_perl->Iargvoutgv, param);
15507 PL_argvout_stack = av_dup_inc(proto_perl->Iargvout_stack, param);
15509 /* shortcuts to regexp stuff */
15510 PL_replgv = gv_dup_inc(proto_perl->Ireplgv, param);
15512 /* shortcuts to misc objects */
15513 PL_errgv = gv_dup(proto_perl->Ierrgv, param);
15515 /* shortcuts to debugging objects */
15516 PL_DBgv = gv_dup_inc(proto_perl->IDBgv, param);
15517 PL_DBline = gv_dup_inc(proto_perl->IDBline, param);
15518 PL_DBsub = gv_dup_inc(proto_perl->IDBsub, param);
15519 PL_DBsingle = sv_dup(proto_perl->IDBsingle, param);
15520 PL_DBtrace = sv_dup(proto_perl->IDBtrace, param);
15521 PL_DBsignal = sv_dup(proto_perl->IDBsignal, param);
15522 Copy(proto_perl->IDBcontrol, PL_DBcontrol, DBVARMG_COUNT, IV);
15524 /* symbol tables */
15525 PL_defstash = hv_dup_inc(proto_perl->Idefstash, param);
15526 PL_curstash = hv_dup_inc(proto_perl->Icurstash, param);
15527 PL_debstash = hv_dup(proto_perl->Idebstash, param);
15528 PL_globalstash = hv_dup(proto_perl->Iglobalstash, param);
15529 PL_curstname = sv_dup_inc(proto_perl->Icurstname, param);
15531 PL_beginav = av_dup_inc(proto_perl->Ibeginav, param);
15532 PL_beginav_save = av_dup_inc(proto_perl->Ibeginav_save, param);
15533 PL_checkav_save = av_dup_inc(proto_perl->Icheckav_save, param);
15534 PL_unitcheckav = av_dup_inc(proto_perl->Iunitcheckav, param);
15535 PL_unitcheckav_save = av_dup_inc(proto_perl->Iunitcheckav_save, param);
15536 PL_endav = av_dup_inc(proto_perl->Iendav, param);
15537 PL_checkav = av_dup_inc(proto_perl->Icheckav, param);
15538 PL_initav = av_dup_inc(proto_perl->Iinitav, param);
15539 PL_savebegin = proto_perl->Isavebegin;
15541 PL_isarev = hv_dup_inc(proto_perl->Iisarev, param);
15543 /* subprocess state */
15544 PL_fdpid = av_dup_inc(proto_perl->Ifdpid, param);
15546 if (proto_perl->Iop_mask)
15547 PL_op_mask = SAVEPVN(proto_perl->Iop_mask, PL_maxo);
15550 /* PL_asserting = proto_perl->Iasserting; */
15552 /* current interpreter roots */
15553 PL_main_cv = cv_dup_inc(proto_perl->Imain_cv, param);
15555 PL_main_root = OpREFCNT_inc(proto_perl->Imain_root);
15558 /* runtime control stuff */
15559 PL_curcopdb = (COP*)any_dup(proto_perl->Icurcopdb, proto_perl);
15561 PL_preambleav = av_dup_inc(proto_perl->Ipreambleav, param);
15563 PL_ors_sv = sv_dup_inc(proto_perl->Iors_sv, param);
15565 /* interpreter atexit processing */
15566 PL_exitlistlen = proto_perl->Iexitlistlen;
15567 if (PL_exitlistlen) {
15568 Newx(PL_exitlist, PL_exitlistlen, PerlExitListEntry);
15569 Copy(proto_perl->Iexitlist, PL_exitlist, PL_exitlistlen, PerlExitListEntry);
15572 PL_exitlist = (PerlExitListEntry*)NULL;
15574 PL_my_cxt_size = proto_perl->Imy_cxt_size;
15575 if (PL_my_cxt_size) {
15576 Newx(PL_my_cxt_list, PL_my_cxt_size, void *);
15577 Copy(proto_perl->Imy_cxt_list, PL_my_cxt_list, PL_my_cxt_size, void *);
15578 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
15579 Newx(PL_my_cxt_keys, PL_my_cxt_size, const char *);
15580 Copy(proto_perl->Imy_cxt_keys, PL_my_cxt_keys, PL_my_cxt_size, char *);
15584 PL_my_cxt_list = (void**)NULL;
15585 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
15586 PL_my_cxt_keys = (const char**)NULL;
15589 PL_modglobal = hv_dup_inc(proto_perl->Imodglobal, param);
15590 PL_custom_op_names = hv_dup_inc(proto_perl->Icustom_op_names,param);
15591 PL_custom_op_descs = hv_dup_inc(proto_perl->Icustom_op_descs,param);
15592 PL_custom_ops = hv_dup_inc(proto_perl->Icustom_ops, param);
15594 PL_compcv = cv_dup(proto_perl->Icompcv, param);
15596 PAD_CLONE_VARS(proto_perl, param);
15598 #ifdef HAVE_INTERP_INTERN
15599 sys_intern_dup(&proto_perl->Isys_intern, &PL_sys_intern);
15602 PL_DBcv = cv_dup(proto_perl->IDBcv, param);
15604 #ifdef PERL_USES_PL_PIDSTATUS
15605 PL_pidstatus = newHV(); /* XXX flag for cloning? */
15607 PL_osname = SAVEPV(proto_perl->Iosname);
15608 PL_parser = parser_dup(proto_perl->Iparser, param);
15610 /* XXX this only works if the saved cop has already been cloned */
15611 if (proto_perl->Iparser) {
15612 PL_parser->saved_curcop = (COP*)any_dup(
15613 proto_perl->Iparser->saved_curcop,
15617 PL_subname = sv_dup_inc(proto_perl->Isubname, param);
15619 #if defined(USE_POSIX_2008_LOCALE) \
15620 && defined(USE_THREAD_SAFE_LOCALE) \
15621 && ! defined(HAS_QUERYLOCALE)
15622 for (i = 0; i < (int) C_ARRAY_LENGTH(PL_curlocales); i++) {
15623 PL_curlocales[i] = savepv("."); /* An illegal value */
15626 #ifdef USE_LOCALE_CTYPE
15627 /* Should we warn if uses locale? */
15628 PL_warn_locale = sv_dup_inc(proto_perl->Iwarn_locale, param);
15631 #ifdef USE_LOCALE_COLLATE
15632 PL_collation_name = SAVEPV(proto_perl->Icollation_name);
15633 #endif /* USE_LOCALE_COLLATE */
15635 #ifdef USE_LOCALE_NUMERIC
15636 PL_numeric_name = SAVEPV(proto_perl->Inumeric_name);
15637 PL_numeric_radix_sv = sv_dup_inc(proto_perl->Inumeric_radix_sv, param);
15639 # if defined(HAS_POSIX_2008_LOCALE)
15640 PL_underlying_numeric_obj = NULL;
15642 #endif /* !USE_LOCALE_NUMERIC */
15644 PL_langinfo_buf = NULL;
15645 PL_langinfo_bufsize = 0;
15647 PL_setlocale_buf = NULL;
15648 PL_setlocale_bufsize = 0;
15650 /* utf8 character class swashes */
15651 PL_seen_deprecated_macro = hv_dup_inc(proto_perl->Iseen_deprecated_macro, param);
15653 if (proto_perl->Ipsig_pend) {
15654 Newxz(PL_psig_pend, SIG_SIZE, int);
15657 PL_psig_pend = (int*)NULL;
15660 if (proto_perl->Ipsig_name) {
15661 Newx(PL_psig_name, 2 * SIG_SIZE, SV*);
15662 sv_dup_inc_multiple(proto_perl->Ipsig_name, PL_psig_name, 2 * SIG_SIZE,
15664 PL_psig_ptr = PL_psig_name + SIG_SIZE;
15667 PL_psig_ptr = (SV**)NULL;
15668 PL_psig_name = (SV**)NULL;
15671 if (flags & CLONEf_COPY_STACKS) {
15672 Newx(PL_tmps_stack, PL_tmps_max, SV*);
15673 sv_dup_inc_multiple(proto_perl->Itmps_stack, PL_tmps_stack,
15674 PL_tmps_ix+1, param);
15676 /* next PUSHMARK() sets *(PL_markstack_ptr+1) */
15677 i = proto_perl->Imarkstack_max - proto_perl->Imarkstack;
15678 Newx(PL_markstack, i, I32);
15679 PL_markstack_max = PL_markstack + (proto_perl->Imarkstack_max
15680 - proto_perl->Imarkstack);
15681 PL_markstack_ptr = PL_markstack + (proto_perl->Imarkstack_ptr
15682 - proto_perl->Imarkstack);
15683 Copy(proto_perl->Imarkstack, PL_markstack,
15684 PL_markstack_ptr - PL_markstack + 1, I32);
15686 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
15687 * NOTE: unlike the others! */
15688 Newx(PL_scopestack, PL_scopestack_max, I32);
15689 Copy(proto_perl->Iscopestack, PL_scopestack, PL_scopestack_ix, I32);
15692 Newx(PL_scopestack_name, PL_scopestack_max, const char *);
15693 Copy(proto_perl->Iscopestack_name, PL_scopestack_name, PL_scopestack_ix, const char *);
15695 /* reset stack AV to correct length before its duped via
15696 * PL_curstackinfo */
15697 AvFILLp(proto_perl->Icurstack) =
15698 proto_perl->Istack_sp - proto_perl->Istack_base;
15700 /* NOTE: si_dup() looks at PL_markstack */
15701 PL_curstackinfo = si_dup(proto_perl->Icurstackinfo, param);
15703 /* PL_curstack = PL_curstackinfo->si_stack; */
15704 PL_curstack = av_dup(proto_perl->Icurstack, param);
15705 PL_mainstack = av_dup(proto_perl->Imainstack, param);
15707 /* next PUSHs() etc. set *(PL_stack_sp+1) */
15708 PL_stack_base = AvARRAY(PL_curstack);
15709 PL_stack_sp = PL_stack_base + (proto_perl->Istack_sp
15710 - proto_perl->Istack_base);
15711 PL_stack_max = PL_stack_base + AvMAX(PL_curstack);
15713 /*Newxz(PL_savestack, PL_savestack_max, ANY);*/
15714 PL_savestack = ss_dup(proto_perl, param);
15718 ENTER; /* perl_destruct() wants to LEAVE; */
15721 PL_statgv = gv_dup(proto_perl->Istatgv, param);
15722 PL_statname = sv_dup_inc(proto_perl->Istatname, param);
15724 PL_rs = sv_dup_inc(proto_perl->Irs, param);
15725 PL_last_in_gv = gv_dup(proto_perl->Ilast_in_gv, param);
15726 PL_defoutgv = gv_dup_inc(proto_perl->Idefoutgv, param);
15727 PL_toptarget = sv_dup_inc(proto_perl->Itoptarget, param);
15728 PL_bodytarget = sv_dup_inc(proto_perl->Ibodytarget, param);
15729 PL_formtarget = sv_dup(proto_perl->Iformtarget, param);
15731 PL_errors = sv_dup_inc(proto_perl->Ierrors, param);
15733 PL_sortcop = (OP*)any_dup(proto_perl->Isortcop, proto_perl);
15734 PL_firstgv = gv_dup_inc(proto_perl->Ifirstgv, param);
15735 PL_secondgv = gv_dup_inc(proto_perl->Isecondgv, param);
15737 PL_stashcache = newHV();
15739 PL_watchaddr = (char **) ptr_table_fetch(PL_ptr_table,
15740 proto_perl->Iwatchaddr);
15741 PL_watchok = PL_watchaddr ? * PL_watchaddr : NULL;
15742 if (PL_debug && PL_watchaddr) {
15743 PerlIO_printf(Perl_debug_log,
15744 "WATCHING: %" UVxf " cloned as %" UVxf " with value %" UVxf "\n",
15745 PTR2UV(proto_perl->Iwatchaddr), PTR2UV(PL_watchaddr),
15746 PTR2UV(PL_watchok));
15749 PL_registered_mros = hv_dup_inc(proto_perl->Iregistered_mros, param);
15750 PL_blockhooks = av_dup_inc(proto_perl->Iblockhooks, param);
15752 /* Call the ->CLONE method, if it exists, for each of the stashes
15753 identified by sv_dup() above.
15755 while(av_tindex(param->stashes) != -1) {
15756 HV* const stash = MUTABLE_HV(av_shift(param->stashes));
15757 GV* const cloner = gv_fetchmethod_autoload(stash, "CLONE", 0);
15758 if (cloner && GvCV(cloner)) {
15763 mXPUSHs(newSVhek(HvNAME_HEK(stash)));
15765 call_sv(MUTABLE_SV(GvCV(cloner)), G_DISCARD);
15771 if (!(flags & CLONEf_KEEP_PTR_TABLE)) {
15772 ptr_table_free(PL_ptr_table);
15773 PL_ptr_table = NULL;
15776 if (!(flags & CLONEf_COPY_STACKS)) {
15777 unreferenced_to_tmp_stack(param->unreferenced);
15780 SvREFCNT_dec(param->stashes);
15782 /* orphaned? eg threads->new inside BEGIN or use */
15783 if (PL_compcv && ! SvREFCNT(PL_compcv)) {
15784 SvREFCNT_inc_simple_void(PL_compcv);
15785 SAVEFREESV(PL_compcv);
15792 S_unreferenced_to_tmp_stack(pTHX_ AV *const unreferenced)
15794 PERL_ARGS_ASSERT_UNREFERENCED_TO_TMP_STACK;
15796 if (AvFILLp(unreferenced) > -1) {
15797 SV **svp = AvARRAY(unreferenced);
15798 SV **const last = svp + AvFILLp(unreferenced);
15802 if (SvREFCNT(*svp) == 1)
15804 } while (++svp <= last);
15806 EXTEND_MORTAL(count);
15807 svp = AvARRAY(unreferenced);
15810 if (SvREFCNT(*svp) == 1) {
15811 /* Our reference is the only one to this SV. This means that
15812 in this thread, the scalar effectively has a 0 reference.
15813 That doesn't work (cleanup never happens), so donate our
15814 reference to it onto the save stack. */
15815 PL_tmps_stack[++PL_tmps_ix] = *svp;
15817 /* As an optimisation, because we are already walking the
15818 entire array, instead of above doing either
15819 SvREFCNT_inc(*svp) or *svp = &PL_sv_undef, we can instead
15820 release our reference to the scalar, so that at the end of
15821 the array owns zero references to the scalars it happens to
15822 point to. We are effectively converting the array from
15823 AvREAL() on to AvREAL() off. This saves the av_clear()
15824 (triggered by the SvREFCNT_dec(unreferenced) below) from
15825 walking the array a second time. */
15826 SvREFCNT_dec(*svp);
15829 } while (++svp <= last);
15830 AvREAL_off(unreferenced);
15832 SvREFCNT_dec_NN(unreferenced);
15836 Perl_clone_params_del(CLONE_PARAMS *param)
15838 /* This seemingly funky ordering keeps the build with PERL_GLOBAL_STRUCT
15840 PerlInterpreter *const to = param->new_perl;
15842 PerlInterpreter *const was = PERL_GET_THX;
15844 PERL_ARGS_ASSERT_CLONE_PARAMS_DEL;
15850 SvREFCNT_dec(param->stashes);
15851 if (param->unreferenced)
15852 unreferenced_to_tmp_stack(param->unreferenced);
15862 Perl_clone_params_new(PerlInterpreter *const from, PerlInterpreter *const to)
15865 /* Need to play this game, as newAV() can call safesysmalloc(), and that
15866 does a dTHX; to get the context from thread local storage.
15867 FIXME - under PERL_CORE Newx(), Safefree() and friends should expand to
15868 a version that passes in my_perl. */
15869 PerlInterpreter *const was = PERL_GET_THX;
15870 CLONE_PARAMS *param;
15872 PERL_ARGS_ASSERT_CLONE_PARAMS_NEW;
15878 /* Given that we've set the context, we can do this unshared. */
15879 Newx(param, 1, CLONE_PARAMS);
15882 param->proto_perl = from;
15883 param->new_perl = to;
15884 param->stashes = (AV *)Perl_newSV_type(to, SVt_PVAV);
15885 AvREAL_off(param->stashes);
15886 param->unreferenced = (AV *)Perl_newSV_type(to, SVt_PVAV);
15894 #endif /* USE_ITHREADS */
15897 Perl_init_constants(pTHX)
15901 SvREFCNT(&PL_sv_undef) = SvREFCNT_IMMORTAL;
15902 SvFLAGS(&PL_sv_undef) = SVf_READONLY|SVf_PROTECT|SVt_NULL;
15903 SvANY(&PL_sv_undef) = NULL;
15905 SvANY(&PL_sv_no) = new_XPVNV();
15906 SvREFCNT(&PL_sv_no) = SvREFCNT_IMMORTAL;
15907 SvFLAGS(&PL_sv_no) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15908 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15911 SvANY(&PL_sv_yes) = new_XPVNV();
15912 SvREFCNT(&PL_sv_yes) = SvREFCNT_IMMORTAL;
15913 SvFLAGS(&PL_sv_yes) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15914 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15917 SvANY(&PL_sv_zero) = new_XPVNV();
15918 SvREFCNT(&PL_sv_zero) = SvREFCNT_IMMORTAL;
15919 SvFLAGS(&PL_sv_zero) = SVt_PVNV|SVf_READONLY|SVf_PROTECT
15920 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
15924 SvPV_set(&PL_sv_no, (char*)PL_No);
15925 SvCUR_set(&PL_sv_no, 0);
15926 SvLEN_set(&PL_sv_no, 0);
15927 SvIV_set(&PL_sv_no, 0);
15928 SvNV_set(&PL_sv_no, 0);
15930 SvPV_set(&PL_sv_yes, (char*)PL_Yes);
15931 SvCUR_set(&PL_sv_yes, 1);
15932 SvLEN_set(&PL_sv_yes, 0);
15933 SvIV_set(&PL_sv_yes, 1);
15934 SvNV_set(&PL_sv_yes, 1);
15936 SvPV_set(&PL_sv_zero, (char*)PL_Zero);
15937 SvCUR_set(&PL_sv_zero, 1);
15938 SvLEN_set(&PL_sv_zero, 0);
15939 SvIV_set(&PL_sv_zero, 0);
15940 SvNV_set(&PL_sv_zero, 0);
15942 PadnamePV(&PL_padname_const) = (char *)PL_No;
15944 assert(SvIMMORTAL_INTERP(&PL_sv_yes));
15945 assert(SvIMMORTAL_INTERP(&PL_sv_undef));
15946 assert(SvIMMORTAL_INTERP(&PL_sv_no));
15947 assert(SvIMMORTAL_INTERP(&PL_sv_zero));
15949 assert(SvIMMORTAL(&PL_sv_yes));
15950 assert(SvIMMORTAL(&PL_sv_undef));
15951 assert(SvIMMORTAL(&PL_sv_no));
15952 assert(SvIMMORTAL(&PL_sv_zero));
15954 assert( SvIMMORTAL_TRUE(&PL_sv_yes));
15955 assert(!SvIMMORTAL_TRUE(&PL_sv_undef));
15956 assert(!SvIMMORTAL_TRUE(&PL_sv_no));
15957 assert(!SvIMMORTAL_TRUE(&PL_sv_zero));
15959 assert( SvTRUE_nomg_NN(&PL_sv_yes));
15960 assert(!SvTRUE_nomg_NN(&PL_sv_undef));
15961 assert(!SvTRUE_nomg_NN(&PL_sv_no));
15962 assert(!SvTRUE_nomg_NN(&PL_sv_zero));
15966 =head1 Unicode Support
15968 =for apidoc sv_recode_to_utf8
15970 C<encoding> is assumed to be an C<Encode> object, on entry the PV
15971 of C<sv> is assumed to be octets in that encoding, and C<sv>
15972 will be converted into Unicode (and UTF-8).
15974 If C<sv> already is UTF-8 (or if it is not C<POK>), or if C<encoding>
15975 is not a reference, nothing is done to C<sv>. If C<encoding> is not
15976 an C<Encode::XS> Encoding object, bad things will happen.
15977 (See F<cpan/Encode/encoding.pm> and L<Encode>.)
15979 The PV of C<sv> is returned.
15984 Perl_sv_recode_to_utf8(pTHX_ SV *sv, SV *encoding)
15986 PERL_ARGS_ASSERT_SV_RECODE_TO_UTF8;
15988 if (SvPOK(sv) && !SvUTF8(sv) && !IN_BYTES && SvROK(encoding)) {
15997 if (SvPADTMP(nsv)) {
15998 nsv = sv_newmortal();
15999 SvSetSV_nosteal(nsv, sv);
16008 Passing sv_yes is wrong - it needs to be or'ed set of constants
16009 for Encode::XS, while UTf-8 decode (currently) assumes a true value means
16010 remove converted chars from source.
16012 Both will default the value - let them.
16014 XPUSHs(&PL_sv_yes);
16017 call_method("decode", G_SCALAR);
16021 s = SvPV_const(uni, len);
16022 if (s != SvPVX_const(sv)) {
16023 SvGROW(sv, len + 1);
16024 Move(s, SvPVX(sv), len + 1, char);
16025 SvCUR_set(sv, len);
16030 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
16031 /* clear pos and any utf8 cache */
16032 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
16035 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
16036 magic_setutf8(sv,mg); /* clear UTF8 cache */
16041 return SvPOKp(sv) ? SvPVX(sv) : NULL;
16045 =for apidoc sv_cat_decode
16047 C<encoding> is assumed to be an C<Encode> object, the PV of C<ssv> is
16048 assumed to be octets in that encoding and decoding the input starts
16049 from the position which S<C<(PV + *offset)>> pointed to. C<dsv> will be
16050 concatenated with the decoded UTF-8 string from C<ssv>. Decoding will terminate
16051 when the string C<tstr> appears in decoding output or the input ends on
16052 the PV of C<ssv>. The value which C<offset> points will be modified
16053 to the last input position on C<ssv>.
16055 Returns TRUE if the terminator was found, else returns FALSE.
16060 Perl_sv_cat_decode(pTHX_ SV *dsv, SV *encoding,
16061 SV *ssv, int *offset, char *tstr, int tlen)
16065 PERL_ARGS_ASSERT_SV_CAT_DECODE;
16067 if (SvPOK(ssv) && SvPOK(dsv) && SvROK(encoding)) {
16078 offsv = newSViv(*offset);
16080 mPUSHp(tstr, tlen);
16082 call_method("cat_decode", G_SCALAR);
16084 ret = SvTRUE(TOPs);
16085 *offset = SvIV(offsv);
16091 Perl_croak(aTHX_ "Invalid argument to sv_cat_decode");
16096 /* ---------------------------------------------------------------------
16098 * support functions for report_uninit()
16101 /* the maxiumum size of array or hash where we will scan looking
16102 * for the undefined element that triggered the warning */
16104 #define FUV_MAX_SEARCH_SIZE 1000
16106 /* Look for an entry in the hash whose value has the same SV as val;
16107 * If so, return a mortal copy of the key. */
16110 S_find_hash_subscript(pTHX_ const HV *const hv, const SV *const val)
16116 PERL_ARGS_ASSERT_FIND_HASH_SUBSCRIPT;
16118 if (!hv || SvMAGICAL(hv) || !HvARRAY(hv) ||
16119 (HvTOTALKEYS(hv) > FUV_MAX_SEARCH_SIZE))
16122 array = HvARRAY(hv);
16124 for (i=HvMAX(hv); i>=0; i--) {
16126 for (entry = array[i]; entry; entry = HeNEXT(entry)) {
16127 if (HeVAL(entry) != val)
16129 if ( HeVAL(entry) == &PL_sv_undef ||
16130 HeVAL(entry) == &PL_sv_placeholder)
16134 if (HeKLEN(entry) == HEf_SVKEY)
16135 return sv_mortalcopy(HeKEY_sv(entry));
16136 return sv_2mortal(newSVhek(HeKEY_hek(entry)));
16142 /* Look for an entry in the array whose value has the same SV as val;
16143 * If so, return the index, otherwise return -1. */
16146 S_find_array_subscript(pTHX_ const AV *const av, const SV *const val)
16148 PERL_ARGS_ASSERT_FIND_ARRAY_SUBSCRIPT;
16150 if (!av || SvMAGICAL(av) || !AvARRAY(av) ||
16151 (AvFILLp(av) > FUV_MAX_SEARCH_SIZE))
16154 if (val != &PL_sv_undef) {
16155 SV ** const svp = AvARRAY(av);
16158 for (i=AvFILLp(av); i>=0; i--)
16165 /* varname(): return the name of a variable, optionally with a subscript.
16166 * If gv is non-zero, use the name of that global, along with gvtype (one
16167 * of "$", "@", "%"); otherwise use the name of the lexical at pad offset
16168 * targ. Depending on the value of the subscript_type flag, return:
16171 #define FUV_SUBSCRIPT_NONE 1 /* "@foo" */
16172 #define FUV_SUBSCRIPT_ARRAY 2 /* "$foo[aindex]" */
16173 #define FUV_SUBSCRIPT_HASH 3 /* "$foo{keyname}" */
16174 #define FUV_SUBSCRIPT_WITHIN 4 /* "within @foo" */
16177 Perl_varname(pTHX_ const GV *const gv, const char gvtype, PADOFFSET targ,
16178 const SV *const keyname, SSize_t aindex, int subscript_type)
16181 SV * const name = sv_newmortal();
16182 if (gv && isGV(gv)) {
16184 buffer[0] = gvtype;
16187 /* as gv_fullname4(), but add literal '^' for $^FOO names */
16189 gv_fullname4(name, gv, buffer, 0);
16191 if ((unsigned int)SvPVX(name)[1] <= 26) {
16193 buffer[1] = SvPVX(name)[1] + 'A' - 1;
16195 /* Swap the 1 unprintable control character for the 2 byte pretty
16196 version - ie substr($name, 1, 1) = $buffer; */
16197 sv_insert(name, 1, 1, buffer, 2);
16201 CV * const cv = gv ? ((CV *)gv) : find_runcv(NULL);
16204 assert(!cv || SvTYPE(cv) == SVt_PVCV || SvTYPE(cv) == SVt_PVFM);
16206 if (!cv || !CvPADLIST(cv))
16208 sv = padnamelist_fetch(PadlistNAMES(CvPADLIST(cv)), targ);
16209 sv_setpvn(name, PadnamePV(sv), PadnameLEN(sv));
16213 if (subscript_type == FUV_SUBSCRIPT_HASH) {
16214 SV * const sv = newSV(0);
16216 const char * const pv = SvPV_nomg_const((SV*)keyname, len);
16218 *SvPVX(name) = '$';
16219 Perl_sv_catpvf(aTHX_ name, "{%s}",
16220 pv_pretty(sv, pv, len, 32, NULL, NULL,
16221 PERL_PV_PRETTY_DUMP | PERL_PV_ESCAPE_UNI_DETECT ));
16222 SvREFCNT_dec_NN(sv);
16224 else if (subscript_type == FUV_SUBSCRIPT_ARRAY) {
16225 *SvPVX(name) = '$';
16226 Perl_sv_catpvf(aTHX_ name, "[%" IVdf "]", (IV)aindex);
16228 else if (subscript_type == FUV_SUBSCRIPT_WITHIN) {
16229 /* We know that name has no magic, so can use 0 instead of SV_GMAGIC */
16230 Perl_sv_insert_flags(aTHX_ name, 0, 0, STR_WITH_LEN("within "), 0);
16238 =for apidoc find_uninit_var
16240 Find the name of the undefined variable (if any) that caused the operator
16241 to issue a "Use of uninitialized value" warning.
16242 If match is true, only return a name if its value matches C<uninit_sv>.
16243 So roughly speaking, if a unary operator (such as C<OP_COS>) generates a
16244 warning, then following the direct child of the op may yield an
16245 C<OP_PADSV> or C<OP_GV> that gives the name of the undefined variable. On the
16246 other hand, with C<OP_ADD> there are two branches to follow, so we only print
16247 the variable name if we get an exact match.
16248 C<desc_p> points to a string pointer holding the description of the op.
16249 This may be updated if needed.
16251 The name is returned as a mortal SV.
16253 Assumes that C<PL_op> is the OP that originally triggered the error, and that
16254 C<PL_comppad>/C<PL_curpad> points to the currently executing pad.
16260 S_find_uninit_var(pTHX_ const OP *const obase, const SV *const uninit_sv,
16261 bool match, const char **desc_p)
16266 const OP *o, *o2, *kid;
16268 PERL_ARGS_ASSERT_FIND_UNINIT_VAR;
16270 if (!obase || (match && (!uninit_sv || uninit_sv == &PL_sv_undef ||
16271 uninit_sv == &PL_sv_placeholder)))
16274 switch (obase->op_type) {
16277 /* undef should care if its args are undef - any warnings
16278 * will be from tied/magic vars */
16286 const bool pad = ( obase->op_type == OP_PADAV
16287 || obase->op_type == OP_PADHV
16288 || obase->op_type == OP_PADRANGE
16291 const bool hash = ( obase->op_type == OP_PADHV
16292 || obase->op_type == OP_RV2HV
16293 || (obase->op_type == OP_PADRANGE
16294 && SvTYPE(PAD_SVl(obase->op_targ)) == SVt_PVHV)
16298 int subscript_type = FUV_SUBSCRIPT_WITHIN;
16300 if (pad) { /* @lex, %lex */
16301 sv = PAD_SVl(obase->op_targ);
16305 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
16306 /* @global, %global */
16307 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
16310 sv = hash ? MUTABLE_SV(GvHV(gv)): MUTABLE_SV(GvAV(gv));
16312 else if (obase == PL_op) /* @{expr}, %{expr} */
16313 return find_uninit_var(cUNOPx(obase)->op_first,
16314 uninit_sv, match, desc_p);
16315 else /* @{expr}, %{expr} as a sub-expression */
16319 /* attempt to find a match within the aggregate */
16321 keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16323 subscript_type = FUV_SUBSCRIPT_HASH;
16326 index = find_array_subscript((const AV *)sv, uninit_sv);
16328 subscript_type = FUV_SUBSCRIPT_ARRAY;
16331 if (match && subscript_type == FUV_SUBSCRIPT_WITHIN)
16334 return varname(gv, (char)(hash ? '%' : '@'), obase->op_targ,
16335 keysv, index, subscript_type);
16339 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
16341 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
16342 if (!gv || !GvSTASH(gv))
16344 if (match && (GvSV(gv) != uninit_sv))
16346 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
16349 return find_uninit_var(cUNOPx(obase)->op_first, uninit_sv, 1, desc_p);
16352 if (match && PAD_SVl(obase->op_targ) != uninit_sv)
16354 return varname(NULL, '$', obase->op_targ,
16355 NULL, 0, FUV_SUBSCRIPT_NONE);
16358 gv = cGVOPx_gv(obase);
16359 if (!gv || (match && GvSV(gv) != uninit_sv) || !GvSTASH(gv))
16361 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
16363 case OP_AELEMFAST_LEX:
16366 AV *av = MUTABLE_AV(PAD_SV(obase->op_targ));
16367 if (!av || SvRMAGICAL(av))
16369 svp = av_fetch(av, (I8)obase->op_private, FALSE);
16370 if (!svp || *svp != uninit_sv)
16373 return varname(NULL, '$', obase->op_targ,
16374 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
16377 gv = cGVOPx_gv(obase);
16382 AV *const av = GvAV(gv);
16383 if (!av || SvRMAGICAL(av))
16385 svp = av_fetch(av, (I8)obase->op_private, FALSE);
16386 if (!svp || *svp != uninit_sv)
16389 return varname(gv, '$', 0,
16390 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
16392 NOT_REACHED; /* NOTREACHED */
16395 o = cUNOPx(obase)->op_first;
16396 if (!o || o->op_type != OP_NULL ||
16397 ! (o->op_targ == OP_AELEM || o->op_targ == OP_HELEM))
16399 return find_uninit_var(cBINOPo->op_last, uninit_sv, match, desc_p);
16404 bool negate = FALSE;
16406 if (PL_op == obase)
16407 /* $a[uninit_expr] or $h{uninit_expr} */
16408 return find_uninit_var(cBINOPx(obase)->op_last,
16409 uninit_sv, match, desc_p);
16412 o = cBINOPx(obase)->op_first;
16413 kid = cBINOPx(obase)->op_last;
16415 /* get the av or hv, and optionally the gv */
16417 if (o->op_type == OP_PADAV || o->op_type == OP_PADHV) {
16418 sv = PAD_SV(o->op_targ);
16420 else if ((o->op_type == OP_RV2AV || o->op_type == OP_RV2HV)
16421 && cUNOPo->op_first->op_type == OP_GV)
16423 gv = cGVOPx_gv(cUNOPo->op_first);
16427 == OP_RV2HV ? MUTABLE_SV(GvHV(gv)) : MUTABLE_SV(GvAV(gv));
16432 if (kid && kid->op_type == OP_NEGATE) {
16434 kid = cUNOPx(kid)->op_first;
16437 if (kid && kid->op_type == OP_CONST && SvOK(cSVOPx_sv(kid))) {
16438 /* index is constant */
16441 kidsv = newSVpvs_flags("-", SVs_TEMP);
16442 sv_catsv(kidsv, cSVOPx_sv(kid));
16445 kidsv = cSVOPx_sv(kid);
16449 if (obase->op_type == OP_HELEM) {
16450 HE* he = hv_fetch_ent(MUTABLE_HV(sv), kidsv, 0, 0);
16451 if (!he || HeVAL(he) != uninit_sv)
16455 SV * const opsv = cSVOPx_sv(kid);
16456 const IV opsviv = SvIV(opsv);
16457 SV * const * const svp = av_fetch(MUTABLE_AV(sv),
16458 negate ? - opsviv : opsviv,
16460 if (!svp || *svp != uninit_sv)
16464 if (obase->op_type == OP_HELEM)
16465 return varname(gv, '%', o->op_targ,
16466 kidsv, 0, FUV_SUBSCRIPT_HASH);
16468 return varname(gv, '@', o->op_targ, NULL,
16469 negate ? - SvIV(cSVOPx_sv(kid)) : SvIV(cSVOPx_sv(kid)),
16470 FUV_SUBSCRIPT_ARRAY);
16473 /* index is an expression;
16474 * attempt to find a match within the aggregate */
16475 if (obase->op_type == OP_HELEM) {
16476 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16478 return varname(gv, '%', o->op_targ,
16479 keysv, 0, FUV_SUBSCRIPT_HASH);
16482 const SSize_t index
16483 = find_array_subscript((const AV *)sv, uninit_sv);
16485 return varname(gv, '@', o->op_targ,
16486 NULL, index, FUV_SUBSCRIPT_ARRAY);
16491 (char)((o->op_type == OP_PADAV || o->op_type == OP_RV2AV)
16493 o->op_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
16495 NOT_REACHED; /* NOTREACHED */
16498 case OP_MULTIDEREF: {
16499 /* If we were executing OP_MULTIDEREF when the undef warning
16500 * triggered, then it must be one of the index values within
16501 * that triggered it. If not, then the only possibility is that
16502 * the value retrieved by the last aggregate index might be the
16503 * culprit. For the former, we set PL_multideref_pc each time before
16504 * using an index, so work though the item list until we reach
16505 * that point. For the latter, just work through the entire item
16506 * list; the last aggregate retrieved will be the candidate.
16507 * There is a third rare possibility: something triggered
16508 * magic while fetching an array/hash element. Just display
16509 * nothing in this case.
16512 /* the named aggregate, if any */
16513 PADOFFSET agg_targ = 0;
16515 /* the last-seen index */
16517 PADOFFSET index_targ;
16519 IV index_const_iv = 0; /* init for spurious compiler warn */
16520 SV *index_const_sv;
16521 int depth = 0; /* how many array/hash lookups we've done */
16523 UNOP_AUX_item *items = cUNOP_AUXx(obase)->op_aux;
16524 UNOP_AUX_item *last = NULL;
16525 UV actions = items->uv;
16528 if (PL_op == obase) {
16529 last = PL_multideref_pc;
16530 assert(last >= items && last <= items + items[-1].uv);
16537 switch (actions & MDEREF_ACTION_MASK) {
16539 case MDEREF_reload:
16540 actions = (++items)->uv;
16543 case MDEREF_HV_padhv_helem: /* $lex{...} */
16546 case MDEREF_AV_padav_aelem: /* $lex[...] */
16547 agg_targ = (++items)->pad_offset;
16551 case MDEREF_HV_gvhv_helem: /* $pkg{...} */
16554 case MDEREF_AV_gvav_aelem: /* $pkg[...] */
16556 agg_gv = (GV*)UNOP_AUX_item_sv(++items);
16557 assert(isGV_with_GP(agg_gv));
16560 case MDEREF_HV_gvsv_vivify_rv2hv_helem: /* $pkg->{...} */
16561 case MDEREF_HV_padsv_vivify_rv2hv_helem: /* $lex->{...} */
16564 case MDEREF_HV_pop_rv2hv_helem: /* expr->{...} */
16565 case MDEREF_HV_vivify_rv2hv_helem: /* vivify, ->{...} */
16571 case MDEREF_AV_gvsv_vivify_rv2av_aelem: /* $pkg->[...] */
16572 case MDEREF_AV_padsv_vivify_rv2av_aelem: /* $lex->[...] */
16575 case MDEREF_AV_pop_rv2av_aelem: /* expr->[...] */
16576 case MDEREF_AV_vivify_rv2av_aelem: /* vivify, ->[...] */
16583 index_const_sv = NULL;
16585 index_type = (actions & MDEREF_INDEX_MASK);
16586 switch (index_type) {
16587 case MDEREF_INDEX_none:
16589 case MDEREF_INDEX_const:
16591 index_const_sv = UNOP_AUX_item_sv(++items)
16593 index_const_iv = (++items)->iv;
16595 case MDEREF_INDEX_padsv:
16596 index_targ = (++items)->pad_offset;
16598 case MDEREF_INDEX_gvsv:
16599 index_gv = (GV*)UNOP_AUX_item_sv(++items);
16600 assert(isGV_with_GP(index_gv));
16604 if (index_type != MDEREF_INDEX_none)
16607 if ( index_type == MDEREF_INDEX_none
16608 || (actions & MDEREF_FLAG_last)
16609 || (last && items >= last)
16613 actions >>= MDEREF_SHIFT;
16616 if (PL_op == obase) {
16617 /* most likely index was undef */
16619 *desc_p = ( (actions & MDEREF_FLAG_last)
16620 && (obase->op_private
16621 & (OPpMULTIDEREF_EXISTS|OPpMULTIDEREF_DELETE)))
16623 (obase->op_private & OPpMULTIDEREF_EXISTS)
16626 : is_hv ? "hash element" : "array element";
16627 assert(index_type != MDEREF_INDEX_none);
16629 if (GvSV(index_gv) == uninit_sv)
16630 return varname(index_gv, '$', 0, NULL, 0,
16631 FUV_SUBSCRIPT_NONE);
16636 if (PL_curpad[index_targ] == uninit_sv)
16637 return varname(NULL, '$', index_targ,
16638 NULL, 0, FUV_SUBSCRIPT_NONE);
16642 /* If we got to this point it was undef on a const subscript,
16643 * so magic probably involved, e.g. $ISA[0]. Give up. */
16647 /* the SV returned by pp_multideref() was undef, if anything was */
16653 sv = PAD_SV(agg_targ);
16655 sv = is_hv ? MUTABLE_SV(GvHV(agg_gv)) : MUTABLE_SV(GvAV(agg_gv));
16659 if (index_type == MDEREF_INDEX_const) {
16664 HE* he = hv_fetch_ent(MUTABLE_HV(sv), index_const_sv, 0, 0);
16665 if (!he || HeVAL(he) != uninit_sv)
16669 SV * const * const svp =
16670 av_fetch(MUTABLE_AV(sv), index_const_iv, FALSE);
16671 if (!svp || *svp != uninit_sv)
16676 ? varname(agg_gv, '%', agg_targ,
16677 index_const_sv, 0, FUV_SUBSCRIPT_HASH)
16678 : varname(agg_gv, '@', agg_targ,
16679 NULL, index_const_iv, FUV_SUBSCRIPT_ARRAY);
16682 /* index is an var */
16684 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
16686 return varname(agg_gv, '%', agg_targ,
16687 keysv, 0, FUV_SUBSCRIPT_HASH);
16690 const SSize_t index
16691 = find_array_subscript((const AV *)sv, uninit_sv);
16693 return varname(agg_gv, '@', agg_targ,
16694 NULL, index, FUV_SUBSCRIPT_ARRAY);
16698 return varname(agg_gv,
16700 agg_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
16702 NOT_REACHED; /* NOTREACHED */
16706 /* only examine RHS */
16707 return find_uninit_var(cBINOPx(obase)->op_first, uninit_sv,
16711 o = cUNOPx(obase)->op_first;
16712 if ( o->op_type == OP_PUSHMARK
16713 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)
16717 if (!OpHAS_SIBLING(o)) {
16718 /* one-arg version of open is highly magical */
16720 if (o->op_type == OP_GV) { /* open FOO; */
16722 if (match && GvSV(gv) != uninit_sv)
16724 return varname(gv, '$', 0,
16725 NULL, 0, FUV_SUBSCRIPT_NONE);
16727 /* other possibilities not handled are:
16728 * open $x; or open my $x; should return '${*$x}'
16729 * open expr; should return '$'.expr ideally
16736 /* ops where $_ may be an implicit arg */
16741 if ( !(obase->op_flags & OPf_STACKED)) {
16742 if (uninit_sv == DEFSV)
16743 return newSVpvs_flags("$_", SVs_TEMP);
16744 else if (obase->op_targ
16745 && uninit_sv == PAD_SVl(obase->op_targ))
16746 return varname(NULL, '$', obase->op_targ, NULL, 0,
16747 FUV_SUBSCRIPT_NONE);
16754 match = 1; /* print etc can return undef on defined args */
16755 /* skip filehandle as it can't produce 'undef' warning */
16756 o = cUNOPx(obase)->op_first;
16757 if ((obase->op_flags & OPf_STACKED)
16759 ( o->op_type == OP_PUSHMARK
16760 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)))
16761 o = OpSIBLING(OpSIBLING(o));
16765 case OP_ENTEREVAL: /* could be eval $undef or $x='$undef'; eval $x */
16766 case OP_CUSTOM: /* XS or custom code could trigger random warnings */
16768 /* the following ops are capable of returning PL_sv_undef even for
16769 * defined arg(s) */
16788 case OP_GETPEERNAME:
16835 case OP_SMARTMATCH:
16844 /* XXX tmp hack: these two may call an XS sub, and currently
16845 XS subs don't have a SUB entry on the context stack, so CV and
16846 pad determination goes wrong, and BAD things happen. So, just
16847 don't try to determine the value under those circumstances.
16848 Need a better fix at dome point. DAPM 11/2007 */
16854 GV * const gv = gv_fetchpvs(".", GV_NOTQUAL, SVt_PV);
16855 if (gv && GvSV(gv) == uninit_sv)
16856 return newSVpvs_flags("$.", SVs_TEMP);
16861 /* def-ness of rval pos() is independent of the def-ness of its arg */
16862 if ( !(obase->op_flags & OPf_MOD))
16868 if (SvROK(PL_rs) && uninit_sv == SvRV(PL_rs))
16869 return newSVpvs_flags("${$/}", SVs_TEMP);
16874 if (!(obase->op_flags & OPf_KIDS))
16876 o = cUNOPx(obase)->op_first;
16882 /* This loop checks all the kid ops, skipping any that cannot pos-
16883 * sibly be responsible for the uninitialized value; i.e., defined
16884 * constants and ops that return nothing. If there is only one op
16885 * left that is not skipped, then we *know* it is responsible for
16886 * the uninitialized value. If there is more than one op left, we
16887 * have to look for an exact match in the while() loop below.
16888 * Note that we skip padrange, because the individual pad ops that
16889 * it replaced are still in the tree, so we work on them instead.
16892 for (kid=o; kid; kid = OpSIBLING(kid)) {
16893 const OPCODE type = kid->op_type;
16894 if ( (type == OP_CONST && SvOK(cSVOPx_sv(kid)))
16895 || (type == OP_NULL && ! (kid->op_flags & OPf_KIDS))
16896 || (type == OP_PUSHMARK)
16897 || (type == OP_PADRANGE)
16901 if (o2) { /* more than one found */
16908 return find_uninit_var(o2, uninit_sv, match, desc_p);
16910 /* scan all args */
16912 sv = find_uninit_var(o, uninit_sv, 1, desc_p);
16924 =for apidoc report_uninit
16926 Print appropriate "Use of uninitialized variable" warning.
16932 Perl_report_uninit(pTHX_ const SV *uninit_sv)
16934 const char *desc = NULL;
16935 SV* varname = NULL;
16938 desc = PL_op->op_type == OP_STRINGIFY && PL_op->op_folded
16940 : PL_op->op_type == OP_MULTICONCAT
16941 && (PL_op->op_private & OPpMULTICONCAT_FAKE)
16944 if (uninit_sv && PL_curpad) {
16945 varname = find_uninit_var(PL_op, uninit_sv, 0, &desc);
16947 sv_insert(varname, 0, 0, " ", 1);
16950 else if (PL_curstackinfo->si_type == PERLSI_SORT && cxstack_ix == 0)
16951 /* we've reached the end of a sort block or sub,
16952 * and the uninit value is probably what that code returned */
16955 /* PL_warn_uninit_sv is constant */
16956 GCC_DIAG_IGNORE_STMT(-Wformat-nonliteral);
16958 /* diag_listed_as: Use of uninitialized value%s */
16959 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit_sv,
16960 SVfARG(varname ? varname : &PL_sv_no),
16963 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit,
16965 GCC_DIAG_RESTORE_STMT;
16969 * ex: set ts=8 sts=4 sw=4 et: