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 *);
43 #ifdef PERL_NEW_COPY_ON_WRITE
44 # ifndef SV_COW_THRESHOLD
45 # define SV_COW_THRESHOLD 0 /* COW iff len > K */
47 # ifndef SV_COWBUF_THRESHOLD
48 # define SV_COWBUF_THRESHOLD 1250 /* COW iff len > K */
50 # ifndef SV_COW_MAX_WASTE_THRESHOLD
51 # define SV_COW_MAX_WASTE_THRESHOLD 80 /* COW iff (len - cur) < K */
53 # ifndef SV_COWBUF_WASTE_THRESHOLD
54 # define SV_COWBUF_WASTE_THRESHOLD 80 /* COW iff (len - cur) < K */
56 # ifndef SV_COW_MAX_WASTE_FACTOR_THRESHOLD
57 # define SV_COW_MAX_WASTE_FACTOR_THRESHOLD 2 /* COW iff len < (cur * K) */
59 # ifndef SV_COWBUF_WASTE_FACTOR_THRESHOLD
60 # define SV_COWBUF_WASTE_FACTOR_THRESHOLD 2 /* COW iff len < (cur * K) */
63 /* Work around compiler warnings about unsigned >= THRESHOLD when thres-
66 # define GE_COW_THRESHOLD(cur) ((cur) >= SV_COW_THRESHOLD)
68 # define GE_COW_THRESHOLD(cur) 1
70 #if SV_COWBUF_THRESHOLD
71 # define GE_COWBUF_THRESHOLD(cur) ((cur) >= SV_COWBUF_THRESHOLD)
73 # define GE_COWBUF_THRESHOLD(cur) 1
75 #if SV_COW_MAX_WASTE_THRESHOLD
76 # define GE_COW_MAX_WASTE_THRESHOLD(cur,len) (((len)-(cur)) < SV_COW_MAX_WASTE_THRESHOLD)
78 # define GE_COW_MAX_WASTE_THRESHOLD(cur,len) 1
80 #if SV_COWBUF_WASTE_THRESHOLD
81 # define GE_COWBUF_WASTE_THRESHOLD(cur,len) (((len)-(cur)) < SV_COWBUF_WASTE_THRESHOLD)
83 # define GE_COWBUF_WASTE_THRESHOLD(cur,len) 1
85 #if SV_COW_MAX_WASTE_FACTOR_THRESHOLD
86 # define GE_COW_MAX_WASTE_FACTOR_THRESHOLD(cur,len) ((len) < SV_COW_MAX_WASTE_FACTOR_THRESHOLD * (cur))
88 # define GE_COW_MAX_WASTE_FACTOR_THRESHOLD(cur,len) 1
90 #if SV_COWBUF_WASTE_FACTOR_THRESHOLD
91 # define GE_COWBUF_WASTE_FACTOR_THRESHOLD(cur,len) ((len) < SV_COWBUF_WASTE_FACTOR_THRESHOLD * (cur))
93 # define GE_COWBUF_WASTE_FACTOR_THRESHOLD(cur,len) 1
96 #define CHECK_COW_THRESHOLD(cur,len) (\
97 GE_COW_THRESHOLD((cur)) && \
98 GE_COW_MAX_WASTE_THRESHOLD((cur),(len)) && \
99 GE_COW_MAX_WASTE_FACTOR_THRESHOLD((cur),(len)) \
101 #define CHECK_COWBUF_THRESHOLD(cur,len) (\
102 GE_COWBUF_THRESHOLD((cur)) && \
103 GE_COWBUF_WASTE_THRESHOLD((cur),(len)) && \
104 GE_COWBUF_WASTE_FACTOR_THRESHOLD((cur),(len)) \
107 #ifdef PERL_UTF8_CACHE_ASSERT
108 /* if adding more checks watch out for the following tests:
109 * t/op/index.t t/op/length.t t/op/pat.t t/op/substr.t
110 * lib/utf8.t lib/Unicode/Collate/t/index.t
113 # define ASSERT_UTF8_CACHE(cache) \
114 STMT_START { if (cache) { assert((cache)[0] <= (cache)[1]); \
115 assert((cache)[2] <= (cache)[3]); \
116 assert((cache)[3] <= (cache)[1]);} \
119 # define ASSERT_UTF8_CACHE(cache) NOOP
122 #ifdef PERL_OLD_COPY_ON_WRITE
123 #define SV_COW_NEXT_SV(sv) INT2PTR(SV *,SvUVX(sv))
124 #define SV_COW_NEXT_SV_SET(current,next) SvUV_set(current, PTR2UV(next))
127 /* ============================================================================
129 =head1 Allocation and deallocation of SVs.
130 An SV (or AV, HV, etc.) is allocated in two parts: the head (struct
131 sv, av, hv...) contains type and reference count information, and for
132 many types, a pointer to the body (struct xrv, xpv, xpviv...), which
133 contains fields specific to each type. Some types store all they need
134 in the head, so don't have a body.
136 In all but the most memory-paranoid configurations (ex: PURIFY), heads
137 and bodies are allocated out of arenas, which by default are
138 approximately 4K chunks of memory parcelled up into N heads or bodies.
139 Sv-bodies are allocated by their sv-type, guaranteeing size
140 consistency needed to allocate safely from arrays.
142 For SV-heads, the first slot in each arena is reserved, and holds a
143 link to the next arena, some flags, and a note of the number of slots.
144 Snaked through each arena chain is a linked list of free items; when
145 this becomes empty, an extra arena is allocated and divided up into N
146 items which are threaded into the free list.
148 SV-bodies are similar, but they use arena-sets by default, which
149 separate the link and info from the arena itself, and reclaim the 1st
150 slot in the arena. SV-bodies are further described later.
152 The following global variables are associated with arenas:
154 PL_sv_arenaroot pointer to list of SV arenas
155 PL_sv_root pointer to list of free SV structures
157 PL_body_arenas head of linked-list of body arenas
158 PL_body_roots[] array of pointers to list of free bodies of svtype
159 arrays are indexed by the svtype needed
161 A few special SV heads are not allocated from an arena, but are
162 instead directly created in the interpreter structure, eg PL_sv_undef.
163 The size of arenas can be changed from the default by setting
164 PERL_ARENA_SIZE appropriately at compile time.
166 The SV arena serves the secondary purpose of allowing still-live SVs
167 to be located and destroyed during final cleanup.
169 At the lowest level, the macros new_SV() and del_SV() grab and free
170 an SV head. (If debugging with -DD, del_SV() calls the function S_del_sv()
171 to return the SV to the free list with error checking.) new_SV() calls
172 more_sv() / sv_add_arena() to add an extra arena if the free list is empty.
173 SVs in the free list have their SvTYPE field set to all ones.
175 At the time of very final cleanup, sv_free_arenas() is called from
176 perl_destruct() to physically free all the arenas allocated since the
177 start of the interpreter.
179 The function visit() scans the SV arenas list, and calls a specified
180 function for each SV it finds which is still live - ie which has an SvTYPE
181 other than all 1's, and a non-zero SvREFCNT. visit() is used by the
182 following functions (specified as [function that calls visit()] / [function
183 called by visit() for each SV]):
185 sv_report_used() / do_report_used()
186 dump all remaining SVs (debugging aid)
188 sv_clean_objs() / do_clean_objs(),do_clean_named_objs(),
189 do_clean_named_io_objs(),do_curse()
190 Attempt to free all objects pointed to by RVs,
191 try to do the same for all objects indir-
192 ectly referenced by typeglobs too, and
193 then do a final sweep, cursing any
194 objects that remain. Called once from
195 perl_destruct(), prior to calling sv_clean_all()
198 sv_clean_all() / do_clean_all()
199 SvREFCNT_dec(sv) each remaining SV, possibly
200 triggering an sv_free(). It also sets the
201 SVf_BREAK flag on the SV to indicate that the
202 refcnt has been artificially lowered, and thus
203 stopping sv_free() from giving spurious warnings
204 about SVs which unexpectedly have a refcnt
205 of zero. called repeatedly from perl_destruct()
206 until there are no SVs left.
208 =head2 Arena allocator API Summary
210 Private API to rest of sv.c
214 new_XPVNV(), del_XPVGV(),
219 sv_report_used(), sv_clean_objs(), sv_clean_all(), sv_free_arenas()
223 * ========================================================================= */
226 * "A time to plant, and a time to uproot what was planted..."
230 # define MEM_LOG_NEW_SV(sv, file, line, func) \
231 Perl_mem_log_new_sv(sv, file, line, func)
232 # define MEM_LOG_DEL_SV(sv, file, line, func) \
233 Perl_mem_log_del_sv(sv, file, line, func)
235 # define MEM_LOG_NEW_SV(sv, file, line, func) NOOP
236 # define MEM_LOG_DEL_SV(sv, file, line, func) NOOP
239 #ifdef DEBUG_LEAKING_SCALARS
240 # define FREE_SV_DEBUG_FILE(sv) STMT_START { \
241 if ((sv)->sv_debug_file) PerlMemShared_free((sv)->sv_debug_file); \
243 # define DEBUG_SV_SERIAL(sv) \
244 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) del_SV\n", \
245 PTR2UV(sv), (long)(sv)->sv_debug_serial))
247 # define FREE_SV_DEBUG_FILE(sv)
248 # define DEBUG_SV_SERIAL(sv) NOOP
252 # define SvARENA_CHAIN(sv) ((sv)->sv_u.svu_rv)
253 # define SvARENA_CHAIN_SET(sv,val) (sv)->sv_u.svu_rv = MUTABLE_SV((val))
254 /* Whilst I'd love to do this, it seems that things like to check on
256 # define POSION_SV_HEAD(sv) PoisonNew(sv, 1, struct STRUCT_SV)
258 # define POSION_SV_HEAD(sv) PoisonNew(&SvANY(sv), 1, void *), \
259 PoisonNew(&SvREFCNT(sv), 1, U32)
261 # define SvARENA_CHAIN(sv) SvANY(sv)
262 # define SvARENA_CHAIN_SET(sv,val) SvANY(sv) = (void *)(val)
263 # define POSION_SV_HEAD(sv)
266 /* Mark an SV head as unused, and add to free list.
268 * If SVf_BREAK is set, skip adding it to the free list, as this SV had
269 * its refcount artificially decremented during global destruction, so
270 * there may be dangling pointers to it. The last thing we want in that
271 * case is for it to be reused. */
273 #define plant_SV(p) \
275 const U32 old_flags = SvFLAGS(p); \
276 MEM_LOG_DEL_SV(p, __FILE__, __LINE__, FUNCTION__); \
277 DEBUG_SV_SERIAL(p); \
278 FREE_SV_DEBUG_FILE(p); \
280 SvFLAGS(p) = SVTYPEMASK; \
281 if (!(old_flags & SVf_BREAK)) { \
282 SvARENA_CHAIN_SET(p, PL_sv_root); \
288 #define uproot_SV(p) \
291 PL_sv_root = MUTABLE_SV(SvARENA_CHAIN(p)); \
296 /* make some more SVs by adding another arena */
302 char *chunk; /* must use New here to match call to */
303 Newx(chunk,PERL_ARENA_SIZE,char); /* Safefree() in sv_free_arenas() */
304 sv_add_arena(chunk, PERL_ARENA_SIZE, 0);
309 /* new_SV(): return a new, empty SV head */
311 #ifdef DEBUG_LEAKING_SCALARS
312 /* provide a real function for a debugger to play with */
314 S_new_SV(pTHX_ const char *file, int line, const char *func)
321 sv = S_more_sv(aTHX);
325 sv->sv_debug_optype = PL_op ? PL_op->op_type : 0;
326 sv->sv_debug_line = (U16) (PL_parser && PL_parser->copline != NOLINE
332 sv->sv_debug_inpad = 0;
333 sv->sv_debug_parent = NULL;
334 sv->sv_debug_file = PL_curcop ? savesharedpv(CopFILE(PL_curcop)): NULL;
336 sv->sv_debug_serial = PL_sv_serial++;
338 MEM_LOG_NEW_SV(sv, file, line, func);
339 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) new_SV (from %s:%d [%s])\n",
340 PTR2UV(sv), (long)sv->sv_debug_serial, file, line, func));
344 # define new_SV(p) (p)=S_new_SV(aTHX_ __FILE__, __LINE__, FUNCTION__)
352 (p) = S_more_sv(aTHX); \
356 MEM_LOG_NEW_SV(p, __FILE__, __LINE__, FUNCTION__); \
361 /* del_SV(): return an empty SV head to the free list */
374 S_del_sv(pTHX_ SV *p)
376 PERL_ARGS_ASSERT_DEL_SV;
381 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
382 const SV * const sv = sva + 1;
383 const SV * const svend = &sva[SvREFCNT(sva)];
384 if (p >= sv && p < svend) {
390 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
391 "Attempt to free non-arena SV: 0x%"UVxf
392 pTHX__FORMAT, PTR2UV(p) pTHX__VALUE);
399 #else /* ! DEBUGGING */
401 #define del_SV(p) plant_SV(p)
403 #endif /* DEBUGGING */
407 =head1 SV Manipulation Functions
409 =for apidoc sv_add_arena
411 Given a chunk of memory, link it to the head of the list of arenas,
412 and split it into a list of free SVs.
418 S_sv_add_arena(pTHX_ char *const ptr, const U32 size, const U32 flags)
420 SV *const sva = MUTABLE_SV(ptr);
424 PERL_ARGS_ASSERT_SV_ADD_ARENA;
426 /* The first SV in an arena isn't an SV. */
427 SvANY(sva) = (void *) PL_sv_arenaroot; /* ptr to next arena */
428 SvREFCNT(sva) = size / sizeof(SV); /* number of SV slots */
429 SvFLAGS(sva) = flags; /* FAKE if not to be freed */
431 PL_sv_arenaroot = sva;
432 PL_sv_root = sva + 1;
434 svend = &sva[SvREFCNT(sva) - 1];
437 SvARENA_CHAIN_SET(sv, (sv + 1));
441 /* Must always set typemask because it's always checked in on cleanup
442 when the arenas are walked looking for objects. */
443 SvFLAGS(sv) = SVTYPEMASK;
446 SvARENA_CHAIN_SET(sv, 0);
450 SvFLAGS(sv) = SVTYPEMASK;
453 /* visit(): call the named function for each non-free SV in the arenas
454 * whose flags field matches the flags/mask args. */
457 S_visit(pTHX_ SVFUNC_t f, const U32 flags, const U32 mask)
462 PERL_ARGS_ASSERT_VISIT;
464 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
465 const SV * const svend = &sva[SvREFCNT(sva)];
467 for (sv = sva + 1; sv < svend; ++sv) {
468 if (SvTYPE(sv) != (svtype)SVTYPEMASK
469 && (sv->sv_flags & mask) == flags
482 /* called by sv_report_used() for each live SV */
485 do_report_used(pTHX_ SV *const sv)
487 if (SvTYPE(sv) != (svtype)SVTYPEMASK) {
488 PerlIO_printf(Perl_debug_log, "****\n");
495 =for apidoc sv_report_used
497 Dump the contents of all SVs not yet freed (debugging aid).
503 Perl_sv_report_used(pTHX)
506 visit(do_report_used, 0, 0);
512 /* called by sv_clean_objs() for each live SV */
515 do_clean_objs(pTHX_ SV *const ref)
519 SV * const target = SvRV(ref);
520 if (SvOBJECT(target)) {
521 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning object ref:\n "), sv_dump(ref)));
522 if (SvWEAKREF(ref)) {
523 sv_del_backref(target, ref);
529 SvREFCNT_dec_NN(target);
536 /* clear any slots in a GV which hold objects - except IO;
537 * called by sv_clean_objs() for each live GV */
540 do_clean_named_objs(pTHX_ SV *const sv)
543 assert(SvTYPE(sv) == SVt_PVGV);
544 assert(isGV_with_GP(sv));
548 /* freeing GP entries may indirectly free the current GV;
549 * hold onto it while we mess with the GP slots */
552 if ( ((obj = GvSV(sv) )) && SvOBJECT(obj)) {
553 DEBUG_D((PerlIO_printf(Perl_debug_log,
554 "Cleaning named glob SV object:\n "), sv_dump(obj)));
556 SvREFCNT_dec_NN(obj);
558 if ( ((obj = MUTABLE_SV(GvAV(sv)) )) && SvOBJECT(obj)) {
559 DEBUG_D((PerlIO_printf(Perl_debug_log,
560 "Cleaning named glob AV object:\n "), sv_dump(obj)));
562 SvREFCNT_dec_NN(obj);
564 if ( ((obj = MUTABLE_SV(GvHV(sv)) )) && SvOBJECT(obj)) {
565 DEBUG_D((PerlIO_printf(Perl_debug_log,
566 "Cleaning named glob HV object:\n "), sv_dump(obj)));
568 SvREFCNT_dec_NN(obj);
570 if ( ((obj = MUTABLE_SV(GvCV(sv)) )) && SvOBJECT(obj)) {
571 DEBUG_D((PerlIO_printf(Perl_debug_log,
572 "Cleaning named glob CV object:\n "), sv_dump(obj)));
574 SvREFCNT_dec_NN(obj);
576 SvREFCNT_dec_NN(sv); /* undo the inc above */
579 /* clear any IO slots in a GV which hold objects (except stderr, defout);
580 * called by sv_clean_objs() for each live GV */
583 do_clean_named_io_objs(pTHX_ SV *const sv)
586 assert(SvTYPE(sv) == SVt_PVGV);
587 assert(isGV_with_GP(sv));
588 if (!GvGP(sv) || sv == (SV*)PL_stderrgv || sv == (SV*)PL_defoutgv)
592 if ( ((obj = MUTABLE_SV(GvIO(sv)) )) && SvOBJECT(obj)) {
593 DEBUG_D((PerlIO_printf(Perl_debug_log,
594 "Cleaning named glob IO object:\n "), sv_dump(obj)));
596 SvREFCNT_dec_NN(obj);
598 SvREFCNT_dec_NN(sv); /* undo the inc above */
601 /* Void wrapper to pass to visit() */
603 do_curse(pTHX_ SV * const sv) {
604 if ((PL_stderrgv && GvGP(PL_stderrgv) && (SV*)GvIO(PL_stderrgv) == sv)
605 || (PL_defoutgv && GvGP(PL_defoutgv) && (SV*)GvIO(PL_defoutgv) == sv))
611 =for apidoc sv_clean_objs
613 Attempt to destroy all objects not yet freed.
619 Perl_sv_clean_objs(pTHX)
622 PL_in_clean_objs = TRUE;
623 visit(do_clean_objs, SVf_ROK, SVf_ROK);
624 /* Some barnacles may yet remain, clinging to typeglobs.
625 * Run the non-IO destructors first: they may want to output
626 * error messages, close files etc */
627 visit(do_clean_named_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
628 visit(do_clean_named_io_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
629 /* And if there are some very tenacious barnacles clinging to arrays,
630 closures, or what have you.... */
631 visit(do_curse, SVs_OBJECT, SVs_OBJECT);
632 olddef = PL_defoutgv;
633 PL_defoutgv = NULL; /* disable skip of PL_defoutgv */
634 if (olddef && isGV_with_GP(olddef))
635 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olddef));
636 olderr = PL_stderrgv;
637 PL_stderrgv = NULL; /* disable skip of PL_stderrgv */
638 if (olderr && isGV_with_GP(olderr))
639 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olderr));
640 SvREFCNT_dec(olddef);
641 PL_in_clean_objs = FALSE;
644 /* called by sv_clean_all() for each live SV */
647 do_clean_all(pTHX_ SV *const sv)
649 if (sv == (const SV *) PL_fdpid || sv == (const SV *)PL_strtab) {
650 /* don't clean pid table and strtab */
653 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning loops: SV at 0x%"UVxf"\n", PTR2UV(sv)) ));
654 SvFLAGS(sv) |= SVf_BREAK;
659 =for apidoc sv_clean_all
661 Decrement the refcnt of each remaining SV, possibly triggering a
662 cleanup. This function may have to be called multiple times to free
663 SVs which are in complex self-referential hierarchies.
669 Perl_sv_clean_all(pTHX)
672 PL_in_clean_all = TRUE;
673 cleaned = visit(do_clean_all, 0,0);
678 ARENASETS: a meta-arena implementation which separates arena-info
679 into struct arena_set, which contains an array of struct
680 arena_descs, each holding info for a single arena. By separating
681 the meta-info from the arena, we recover the 1st slot, formerly
682 borrowed for list management. The arena_set is about the size of an
683 arena, avoiding the needless malloc overhead of a naive linked-list.
685 The cost is 1 arena-set malloc per ~320 arena-mallocs, + the unused
686 memory in the last arena-set (1/2 on average). In trade, we get
687 back the 1st slot in each arena (ie 1.7% of a CV-arena, less for
688 smaller types). The recovery of the wasted space allows use of
689 small arenas for large, rare body types, by changing array* fields
690 in body_details_by_type[] below.
693 char *arena; /* the raw storage, allocated aligned */
694 size_t size; /* its size ~4k typ */
695 svtype utype; /* bodytype stored in arena */
700 /* Get the maximum number of elements in set[] such that struct arena_set
701 will fit within PERL_ARENA_SIZE, which is probably just under 4K, and
702 therefore likely to be 1 aligned memory page. */
704 #define ARENAS_PER_SET ((PERL_ARENA_SIZE - sizeof(struct arena_set*) \
705 - 2 * sizeof(int)) / sizeof (struct arena_desc))
708 struct arena_set* next;
709 unsigned int set_size; /* ie ARENAS_PER_SET */
710 unsigned int curr; /* index of next available arena-desc */
711 struct arena_desc set[ARENAS_PER_SET];
715 =for apidoc sv_free_arenas
717 Deallocate the memory used by all arenas. Note that all the individual SV
718 heads and bodies within the arenas must already have been freed.
724 Perl_sv_free_arenas(pTHX)
730 /* Free arenas here, but be careful about fake ones. (We assume
731 contiguity of the fake ones with the corresponding real ones.) */
733 for (sva = PL_sv_arenaroot; sva; sva = svanext) {
734 svanext = MUTABLE_SV(SvANY(sva));
735 while (svanext && SvFAKE(svanext))
736 svanext = MUTABLE_SV(SvANY(svanext));
743 struct arena_set *aroot = (struct arena_set*) PL_body_arenas;
746 struct arena_set *current = aroot;
749 assert(aroot->set[i].arena);
750 Safefree(aroot->set[i].arena);
758 i = PERL_ARENA_ROOTS_SIZE;
760 PL_body_roots[i] = 0;
767 Here are mid-level routines that manage the allocation of bodies out
768 of the various arenas. There are 5 kinds of arenas:
770 1. SV-head arenas, which are discussed and handled above
771 2. regular body arenas
772 3. arenas for reduced-size bodies
775 Arena types 2 & 3 are chained by body-type off an array of
776 arena-root pointers, which is indexed by svtype. Some of the
777 larger/less used body types are malloced singly, since a large
778 unused block of them is wasteful. Also, several svtypes dont have
779 bodies; the data fits into the sv-head itself. The arena-root
780 pointer thus has a few unused root-pointers (which may be hijacked
781 later for arena types 4,5)
783 3 differs from 2 as an optimization; some body types have several
784 unused fields in the front of the structure (which are kept in-place
785 for consistency). These bodies can be allocated in smaller chunks,
786 because the leading fields arent accessed. Pointers to such bodies
787 are decremented to point at the unused 'ghost' memory, knowing that
788 the pointers are used with offsets to the real memory.
791 =head1 SV-Body Allocation
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 don't need it either, because
834 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) */
879 unsigned int type : 4; /* We have space for a sanity check. */
880 unsigned int cant_upgrade : 1; /* Cannot upgrade this type */
881 unsigned int zero_nv : 1; /* zero the NV when upgrading from this */
882 unsigned int arena : 1; /* Allocated from an arena */
883 size_t 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
937 { sizeof(NV), sizeof(NV),
938 STRUCT_OFFSET(XPVNV, xnv_u),
939 SVt_NV, FALSE, HADNV, HASARENA, FIT_ARENA(0, sizeof(NV)) },
941 { sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur),
942 copy_length(XPV, xpv_len) - STRUCT_OFFSET(XPV, xpv_cur),
943 + STRUCT_OFFSET(XPV, xpv_cur),
944 SVt_PV, FALSE, NONV, HASARENA,
945 FIT_ARENA(0, sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur)) },
947 { sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur),
948 copy_length(XINVLIST, is_offset) - STRUCT_OFFSET(XPV, xpv_cur),
949 + STRUCT_OFFSET(XPV, xpv_cur),
950 SVt_INVLIST, TRUE, NONV, HASARENA,
951 FIT_ARENA(0, sizeof(XINVLIST) - STRUCT_OFFSET(XPV, xpv_cur)) },
953 { sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur),
954 copy_length(XPVIV, xiv_u) - STRUCT_OFFSET(XPV, xpv_cur),
955 + STRUCT_OFFSET(XPV, xpv_cur),
956 SVt_PVIV, FALSE, NONV, HASARENA,
957 FIT_ARENA(0, sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur)) },
959 { sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur),
960 copy_length(XPVNV, xnv_u) - STRUCT_OFFSET(XPV, xpv_cur),
961 + STRUCT_OFFSET(XPV, xpv_cur),
962 SVt_PVNV, FALSE, HADNV, HASARENA,
963 FIT_ARENA(0, sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur)) },
965 { sizeof(XPVMG), copy_length(XPVMG, xnv_u), 0, SVt_PVMG, FALSE, HADNV,
966 HASARENA, FIT_ARENA(0, sizeof(XPVMG)) },
971 SVt_REGEXP, TRUE, NONV, HASARENA,
972 FIT_ARENA(0, sizeof(regexp))
975 { sizeof(XPVGV), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV,
976 HASARENA, FIT_ARENA(0, sizeof(XPVGV)) },
978 { sizeof(XPVLV), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV,
979 HASARENA, FIT_ARENA(0, sizeof(XPVLV)) },
982 copy_length(XPVAV, xav_alloc),
984 SVt_PVAV, TRUE, NONV, HASARENA,
985 FIT_ARENA(0, sizeof(XPVAV)) },
988 copy_length(XPVHV, xhv_max),
990 SVt_PVHV, TRUE, NONV, HASARENA,
991 FIT_ARENA(0, sizeof(XPVHV)) },
996 SVt_PVCV, TRUE, NONV, HASARENA,
997 FIT_ARENA(0, sizeof(XPVCV)) },
1002 SVt_PVFM, TRUE, NONV, NOARENA,
1003 FIT_ARENA(20, sizeof(XPVFM)) },
1008 SVt_PVIO, TRUE, NONV, HASARENA,
1009 FIT_ARENA(24, sizeof(XPVIO)) },
1012 #define new_body_allocated(sv_type) \
1013 (void *)((char *)S_new_body(aTHX_ sv_type) \
1014 - bodies_by_type[sv_type].offset)
1016 /* return a thing to the free list */
1018 #define del_body(thing, root) \
1020 void ** const thing_copy = (void **)thing; \
1021 *thing_copy = *root; \
1022 *root = (void*)thing_copy; \
1027 #define new_XNV() safemalloc(sizeof(XPVNV))
1028 #define new_XPVNV() safemalloc(sizeof(XPVNV))
1029 #define new_XPVMG() safemalloc(sizeof(XPVMG))
1031 #define del_XPVGV(p) safefree(p)
1035 #define new_XNV() new_body_allocated(SVt_NV)
1036 #define new_XPVNV() new_body_allocated(SVt_PVNV)
1037 #define new_XPVMG() new_body_allocated(SVt_PVMG)
1039 #define del_XPVGV(p) del_body(p + bodies_by_type[SVt_PVGV].offset, \
1040 &PL_body_roots[SVt_PVGV])
1044 /* no arena for you! */
1046 #define new_NOARENA(details) \
1047 safemalloc((details)->body_size + (details)->offset)
1048 #define new_NOARENAZ(details) \
1049 safecalloc((details)->body_size + (details)->offset, 1)
1052 Perl_more_bodies (pTHX_ const svtype sv_type, const size_t body_size,
1053 const size_t arena_size)
1055 void ** const root = &PL_body_roots[sv_type];
1056 struct arena_desc *adesc;
1057 struct arena_set *aroot = (struct arena_set *) PL_body_arenas;
1061 const size_t good_arena_size = Perl_malloc_good_size(arena_size);
1062 #if defined(DEBUGGING) && defined(PERL_GLOBAL_STRUCT)
1065 #if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT_PRIVATE)
1066 static bool done_sanity_check;
1068 /* PERL_GLOBAL_STRUCT_PRIVATE cannot coexist with global
1069 * variables like done_sanity_check. */
1070 if (!done_sanity_check) {
1071 unsigned int i = SVt_LAST;
1073 done_sanity_check = TRUE;
1076 assert (bodies_by_type[i].type == i);
1082 /* may need new arena-set to hold new arena */
1083 if (!aroot || aroot->curr >= aroot->set_size) {
1084 struct arena_set *newroot;
1085 Newxz(newroot, 1, struct arena_set);
1086 newroot->set_size = ARENAS_PER_SET;
1087 newroot->next = aroot;
1089 PL_body_arenas = (void *) newroot;
1090 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot));
1093 /* ok, now have arena-set with at least 1 empty/available arena-desc */
1094 curr = aroot->curr++;
1095 adesc = &(aroot->set[curr]);
1096 assert(!adesc->arena);
1098 Newx(adesc->arena, good_arena_size, char);
1099 adesc->size = good_arena_size;
1100 adesc->utype = sv_type;
1101 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %"UVuf"\n",
1102 curr, (void*)adesc->arena, (UV)good_arena_size));
1104 start = (char *) adesc->arena;
1106 /* Get the address of the byte after the end of the last body we can fit.
1107 Remember, this is integer division: */
1108 end = start + good_arena_size / body_size * body_size;
1110 /* computed count doesn't reflect the 1st slot reservation */
1111 #if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE)
1112 DEBUG_m(PerlIO_printf(Perl_debug_log,
1113 "arena %p end %p arena-size %d (from %d) type %d "
1115 (void*)start, (void*)end, (int)good_arena_size,
1116 (int)arena_size, sv_type, (int)body_size,
1117 (int)good_arena_size / (int)body_size));
1119 DEBUG_m(PerlIO_printf(Perl_debug_log,
1120 "arena %p end %p arena-size %d type %d size %d ct %d\n",
1121 (void*)start, (void*)end,
1122 (int)arena_size, sv_type, (int)body_size,
1123 (int)good_arena_size / (int)body_size));
1125 *root = (void *)start;
1128 /* Where the next body would start: */
1129 char * const next = start + body_size;
1132 /* This is the last body: */
1133 assert(next == end);
1135 *(void **)start = 0;
1139 *(void**) start = (void *)next;
1144 /* grab a new thing from the free list, allocating more if necessary.
1145 The inline version is used for speed in hot routines, and the
1146 function using it serves the rest (unless PURIFY).
1148 #define new_body_inline(xpv, sv_type) \
1150 void ** const r3wt = &PL_body_roots[sv_type]; \
1151 xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \
1152 ? *((void **)(r3wt)) : Perl_more_bodies(aTHX_ sv_type, \
1153 bodies_by_type[sv_type].body_size,\
1154 bodies_by_type[sv_type].arena_size)); \
1155 *(r3wt) = *(void**)(xpv); \
1161 S_new_body(pTHX_ const svtype sv_type)
1164 new_body_inline(xpv, sv_type);
1170 static const struct body_details fake_rv =
1171 { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 };
1174 =for apidoc sv_upgrade
1176 Upgrade an SV to a more complex form. Generally adds a new body type to the
1177 SV, then copies across as much information as possible from the old body.
1178 It croaks if the SV is already in a more complex form than requested. You
1179 generally want to use the C<SvUPGRADE> macro wrapper, which checks the type
1180 before calling C<sv_upgrade>, and hence does not croak. See also
1187 Perl_sv_upgrade(pTHX_ SV *const sv, svtype new_type)
1191 const svtype old_type = SvTYPE(sv);
1192 const struct body_details *new_type_details;
1193 const struct body_details *old_type_details
1194 = bodies_by_type + old_type;
1195 SV *referant = NULL;
1197 PERL_ARGS_ASSERT_SV_UPGRADE;
1199 if (old_type == new_type)
1202 /* This clause was purposefully added ahead of the early return above to
1203 the shared string hackery for (sort {$a <=> $b} keys %hash), with the
1204 inference by Nick I-S that it would fix other troublesome cases. See
1205 changes 7162, 7163 (f130fd4589cf5fbb24149cd4db4137c8326f49c1 and parent)
1207 Given that shared hash key scalars are no longer PVIV, but PV, there is
1208 no longer need to unshare so as to free up the IVX slot for its proper
1209 purpose. So it's safe to move the early return earlier. */
1211 if (new_type > SVt_PVMG && SvIsCOW(sv)) {
1212 sv_force_normal_flags(sv, 0);
1215 old_body = SvANY(sv);
1217 /* Copying structures onto other structures that have been neatly zeroed
1218 has a subtle gotcha. Consider XPVMG
1220 +------+------+------+------+------+-------+-------+
1221 | NV | CUR | LEN | IV | MAGIC | STASH |
1222 +------+------+------+------+------+-------+-------+
1223 0 4 8 12 16 20 24 28
1225 where NVs are aligned to 8 bytes, so that sizeof that structure is
1226 actually 32 bytes long, with 4 bytes of padding at the end:
1228 +------+------+------+------+------+-------+-------+------+
1229 | NV | CUR | LEN | IV | MAGIC | STASH | ??? |
1230 +------+------+------+------+------+-------+-------+------+
1231 0 4 8 12 16 20 24 28 32
1233 so what happens if you allocate memory for this structure:
1235 +------+------+------+------+------+-------+-------+------+------+...
1236 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME |
1237 +------+------+------+------+------+-------+-------+------+------+...
1238 0 4 8 12 16 20 24 28 32 36
1240 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you
1241 expect, because you copy the area marked ??? onto GP. Now, ??? may have
1242 started out as zero once, but it's quite possible that it isn't. So now,
1243 rather than a nicely zeroed GP, you have it pointing somewhere random.
1246 (In fact, GP ends up pointing at a previous GP structure, because the
1247 principle cause of the padding in XPVMG getting garbage is a copy of
1248 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now
1249 this happens to be moot because XPVGV has been re-ordered, with GP
1250 no longer after STASH)
1252 So we are careful and work out the size of used parts of all the
1260 referant = SvRV(sv);
1261 old_type_details = &fake_rv;
1262 if (new_type == SVt_NV)
1263 new_type = SVt_PVNV;
1265 if (new_type < SVt_PVIV) {
1266 new_type = (new_type == SVt_NV)
1267 ? SVt_PVNV : SVt_PVIV;
1272 if (new_type < SVt_PVNV) {
1273 new_type = SVt_PVNV;
1277 assert(new_type > SVt_PV);
1278 assert(SVt_IV < SVt_PV);
1279 assert(SVt_NV < SVt_PV);
1286 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena,
1287 there's no way that it can be safely upgraded, because perl.c
1288 expects to Safefree(SvANY(PL_mess_sv)) */
1289 assert(sv != PL_mess_sv);
1290 /* This flag bit is used to mean other things in other scalar types.
1291 Given that it only has meaning inside the pad, it shouldn't be set
1292 on anything that can get upgraded. */
1293 assert(!SvPAD_TYPED(sv));
1296 if (UNLIKELY(old_type_details->cant_upgrade))
1297 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf,
1298 sv_reftype(sv, 0), (UV) old_type, (UV) new_type);
1301 if (UNLIKELY(old_type > new_type))
1302 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d",
1303 (int)old_type, (int)new_type);
1305 new_type_details = bodies_by_type + new_type;
1307 SvFLAGS(sv) &= ~SVTYPEMASK;
1308 SvFLAGS(sv) |= new_type;
1310 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of
1311 the return statements above will have triggered. */
1312 assert (new_type != SVt_NULL);
1315 assert(old_type == SVt_NULL);
1316 SvANY(sv) = (XPVIV*)((char*)&(sv->sv_u.svu_iv) - STRUCT_OFFSET(XPVIV, xiv_iv));
1320 assert(old_type == SVt_NULL);
1321 SvANY(sv) = new_XNV();
1326 assert(new_type_details->body_size);
1329 assert(new_type_details->arena);
1330 assert(new_type_details->arena_size);
1331 /* This points to the start of the allocated area. */
1332 new_body_inline(new_body, new_type);
1333 Zero(new_body, new_type_details->body_size, char);
1334 new_body = ((char *)new_body) - new_type_details->offset;
1336 /* We always allocated the full length item with PURIFY. To do this
1337 we fake things so that arena is false for all 16 types.. */
1338 new_body = new_NOARENAZ(new_type_details);
1340 SvANY(sv) = new_body;
1341 if (new_type == SVt_PVAV) {
1345 if (old_type_details->body_size) {
1348 /* It will have been zeroed when the new body was allocated.
1349 Lets not write to it, in case it confuses a write-back
1355 #ifndef NODEFAULT_SHAREKEYS
1356 HvSHAREKEYS_on(sv); /* key-sharing on by default */
1358 /* start with PERL_HASH_DEFAULT_HvMAX+1 buckets: */
1359 HvMAX(sv) = PERL_HASH_DEFAULT_HvMAX;
1362 /* SVt_NULL isn't the only thing upgraded to AV or HV.
1363 The target created by newSVrv also is, and it can have magic.
1364 However, it never has SvPVX set.
1366 if (old_type == SVt_IV) {
1368 } else if (old_type >= SVt_PV) {
1369 assert(SvPVX_const(sv) == 0);
1372 if (old_type >= SVt_PVMG) {
1373 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic);
1374 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash);
1376 sv->sv_u.svu_array = NULL; /* or svu_hash */
1381 /* XXX Is this still needed? Was it ever needed? Surely as there is
1382 no route from NV to PVIV, NOK can never be true */
1383 assert(!SvNOKp(sv));
1396 assert(new_type_details->body_size);
1397 /* We always allocated the full length item with PURIFY. To do this
1398 we fake things so that arena is false for all 16 types.. */
1399 if(new_type_details->arena) {
1400 /* This points to the start of the allocated area. */
1401 new_body_inline(new_body, new_type);
1402 Zero(new_body, new_type_details->body_size, char);
1403 new_body = ((char *)new_body) - new_type_details->offset;
1405 new_body = new_NOARENAZ(new_type_details);
1407 SvANY(sv) = new_body;
1409 if (old_type_details->copy) {
1410 /* There is now the potential for an upgrade from something without
1411 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */
1412 int offset = old_type_details->offset;
1413 int length = old_type_details->copy;
1415 if (new_type_details->offset > old_type_details->offset) {
1416 const int difference
1417 = new_type_details->offset - old_type_details->offset;
1418 offset += difference;
1419 length -= difference;
1421 assert (length >= 0);
1423 Copy((char *)old_body + offset, (char *)new_body + offset, length,
1427 #ifndef NV_ZERO_IS_ALLBITS_ZERO
1428 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a
1429 * correct 0.0 for us. Otherwise, if the old body didn't have an
1430 * NV slot, but the new one does, then we need to initialise the
1431 * freshly created NV slot with whatever the correct bit pattern is
1433 if (old_type_details->zero_nv && !new_type_details->zero_nv
1434 && !isGV_with_GP(sv))
1438 if (UNLIKELY(new_type == SVt_PVIO)) {
1439 IO * const io = MUTABLE_IO(sv);
1440 GV *iogv = gv_fetchpvs("IO::File::", GV_ADD, SVt_PVHV);
1443 /* Clear the stashcache because a new IO could overrule a package
1445 DEBUG_o(Perl_deb(aTHX_ "sv_upgrade clearing PL_stashcache\n"));
1446 hv_clear(PL_stashcache);
1448 SvSTASH_set(io, MUTABLE_HV(SvREFCNT_inc(GvHV(iogv))));
1449 IoPAGE_LEN(sv) = 60;
1451 if (UNLIKELY(new_type == SVt_REGEXP))
1452 sv->sv_u.svu_rx = (regexp *)new_body;
1453 else if (old_type < SVt_PV) {
1454 /* referant will be NULL unless the old type was SVt_IV emulating
1456 sv->sv_u.svu_rv = referant;
1460 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1461 (unsigned long)new_type);
1464 if (old_type > SVt_IV) {
1468 /* Note that there is an assumption that all bodies of types that
1469 can be upgraded came from arenas. Only the more complex non-
1470 upgradable types are allowed to be directly malloc()ed. */
1471 assert(old_type_details->arena);
1472 del_body((void*)((char*)old_body + old_type_details->offset),
1473 &PL_body_roots[old_type]);
1479 =for apidoc sv_backoff
1481 Remove any string offset. You should normally use the C<SvOOK_off> macro
1488 Perl_sv_backoff(SV *const sv)
1491 const char * const s = SvPVX_const(sv);
1493 PERL_ARGS_ASSERT_SV_BACKOFF;
1496 assert(SvTYPE(sv) != SVt_PVHV);
1497 assert(SvTYPE(sv) != SVt_PVAV);
1499 SvOOK_offset(sv, delta);
1501 SvLEN_set(sv, SvLEN(sv) + delta);
1502 SvPV_set(sv, SvPVX(sv) - delta);
1503 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
1504 SvFLAGS(sv) &= ~SVf_OOK;
1511 Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1512 upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1513 Use the C<SvGROW> wrapper instead.
1518 static void S_sv_uncow(pTHX_ SV * const sv, const U32 flags);
1521 Perl_sv_grow(pTHX_ SV *const sv, STRLEN newlen)
1525 PERL_ARGS_ASSERT_SV_GROW;
1529 if (SvTYPE(sv) < SVt_PV) {
1530 sv_upgrade(sv, SVt_PV);
1531 s = SvPVX_mutable(sv);
1533 else if (SvOOK(sv)) { /* pv is offset? */
1535 s = SvPVX_mutable(sv);
1536 if (newlen > SvLEN(sv))
1537 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
1541 if (SvIsCOW(sv)) S_sv_uncow(aTHX_ sv, 0);
1542 s = SvPVX_mutable(sv);
1545 #ifdef PERL_NEW_COPY_ON_WRITE
1546 /* the new COW scheme uses SvPVX(sv)[SvLEN(sv)-1] (if spare)
1547 * to store the COW count. So in general, allocate one more byte than
1548 * asked for, to make it likely this byte is always spare: and thus
1549 * make more strings COW-able.
1550 * If the new size is a big power of two, don't bother: we assume the
1551 * caller wanted a nice 2^N sized block and will be annoyed at getting
1557 #if defined(PERL_USE_MALLOC_SIZE) && defined(Perl_safesysmalloc_size)
1558 #define PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1561 if (newlen > SvLEN(sv)) { /* need more room? */
1562 STRLEN minlen = SvCUR(sv);
1563 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
1564 if (newlen < minlen)
1566 #ifndef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1568 /* Don't round up on the first allocation, as odds are pretty good that
1569 * the initial request is accurate as to what is really needed */
1571 newlen = PERL_STRLEN_ROUNDUP(newlen);
1574 if (SvLEN(sv) && s) {
1575 s = (char*)saferealloc(s, newlen);
1578 s = (char*)safemalloc(newlen);
1579 if (SvPVX_const(sv) && SvCUR(sv)) {
1580 Move(SvPVX_const(sv), s, (newlen < SvCUR(sv)) ? newlen : SvCUR(sv), char);
1584 #ifdef PERL_UNWARANTED_CHUMMINESS_WITH_MALLOC
1585 /* Do this here, do it once, do it right, and then we will never get
1586 called back into sv_grow() unless there really is some growing
1588 SvLEN_set(sv, Perl_safesysmalloc_size(s));
1590 SvLEN_set(sv, newlen);
1597 =for apidoc sv_setiv
1599 Copies an integer into the given SV, upgrading first if necessary.
1600 Does not handle 'set' magic. See also C<sv_setiv_mg>.
1606 Perl_sv_setiv(pTHX_ SV *const sv, const IV i)
1608 PERL_ARGS_ASSERT_SV_SETIV;
1610 SV_CHECK_THINKFIRST_COW_DROP(sv);
1611 switch (SvTYPE(sv)) {
1614 sv_upgrade(sv, SVt_IV);
1617 sv_upgrade(sv, SVt_PVIV);
1621 if (!isGV_with_GP(sv))
1628 /* diag_listed_as: Can't coerce %s to %s in %s */
1629 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1633 (void)SvIOK_only(sv); /* validate number */
1639 =for apidoc sv_setiv_mg
1641 Like C<sv_setiv>, but also handles 'set' magic.
1647 Perl_sv_setiv_mg(pTHX_ SV *const sv, const IV i)
1649 PERL_ARGS_ASSERT_SV_SETIV_MG;
1656 =for apidoc sv_setuv
1658 Copies an unsigned integer into the given SV, upgrading first if necessary.
1659 Does not handle 'set' magic. See also C<sv_setuv_mg>.
1665 Perl_sv_setuv(pTHX_ SV *const sv, const UV u)
1667 PERL_ARGS_ASSERT_SV_SETUV;
1669 /* With the if statement to ensure that integers are stored as IVs whenever
1671 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
1674 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1676 If you wish to remove the following if statement, so that this routine
1677 (and its callers) always return UVs, please benchmark to see what the
1678 effect is. Modern CPUs may be different. Or may not :-)
1680 if (u <= (UV)IV_MAX) {
1681 sv_setiv(sv, (IV)u);
1690 =for apidoc sv_setuv_mg
1692 Like C<sv_setuv>, but also handles 'set' magic.
1698 Perl_sv_setuv_mg(pTHX_ SV *const sv, const UV u)
1700 PERL_ARGS_ASSERT_SV_SETUV_MG;
1707 =for apidoc sv_setnv
1709 Copies a double into the given SV, upgrading first if necessary.
1710 Does not handle 'set' magic. See also C<sv_setnv_mg>.
1716 Perl_sv_setnv(pTHX_ SV *const sv, const NV num)
1718 PERL_ARGS_ASSERT_SV_SETNV;
1720 SV_CHECK_THINKFIRST_COW_DROP(sv);
1721 switch (SvTYPE(sv)) {
1724 sv_upgrade(sv, SVt_NV);
1728 sv_upgrade(sv, SVt_PVNV);
1732 if (!isGV_with_GP(sv))
1739 /* diag_listed_as: Can't coerce %s to %s in %s */
1740 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
1745 (void)SvNOK_only(sv); /* validate number */
1750 =for apidoc sv_setnv_mg
1752 Like C<sv_setnv>, but also handles 'set' magic.
1758 Perl_sv_setnv_mg(pTHX_ SV *const sv, const NV num)
1760 PERL_ARGS_ASSERT_SV_SETNV_MG;
1766 /* Return a cleaned-up, printable version of sv, for non-numeric, or
1767 * not incrementable warning display.
1768 * Originally part of S_not_a_number().
1769 * The return value may be != tmpbuf.
1773 S_sv_display(pTHX_ SV *const sv, char *tmpbuf, STRLEN tmpbuf_size) {
1776 PERL_ARGS_ASSERT_SV_DISPLAY;
1779 SV *dsv = newSVpvs_flags("", SVs_TEMP);
1780 pv = sv_uni_display(dsv, sv, 10, UNI_DISPLAY_ISPRINT);
1783 const char * const limit = tmpbuf + tmpbuf_size - 8;
1784 /* each *s can expand to 4 chars + "...\0",
1785 i.e. need room for 8 chars */
1787 const char *s = SvPVX_const(sv);
1788 const char * const end = s + SvCUR(sv);
1789 for ( ; s < end && d < limit; s++ ) {
1791 if (! isASCII(ch) && !isPRINT_LC(ch)) {
1795 /* Map to ASCII "equivalent" of Latin1 */
1796 ch = LATIN1_TO_NATIVE(NATIVE_TO_LATIN1(ch) & 127);
1802 else if (ch == '\r') {
1806 else if (ch == '\f') {
1810 else if (ch == '\\') {
1814 else if (ch == '\0') {
1818 else if (isPRINT_LC(ch))
1837 /* Print an "isn't numeric" warning, using a cleaned-up,
1838 * printable version of the offending string
1842 S_not_a_number(pTHX_ SV *const sv)
1847 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1849 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1852 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1853 /* diag_listed_as: Argument "%s" isn't numeric%s */
1854 "Argument \"%s\" isn't numeric in %s", pv,
1857 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1858 /* diag_listed_as: Argument "%s" isn't numeric%s */
1859 "Argument \"%s\" isn't numeric", pv);
1863 S_not_incrementable(pTHX_ SV *const sv) {
1867 PERL_ARGS_ASSERT_NOT_INCREMENTABLE;
1869 pv = sv_display(sv, tmpbuf, sizeof(tmpbuf));
1871 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1872 "Argument \"%s\" treated as 0 in increment (++)", pv);
1876 =for apidoc looks_like_number
1878 Test if the content of an SV looks like a number (or is a number).
1879 C<Inf> and C<Infinity> are treated as numbers (so will not issue a
1880 non-numeric warning), even if your atof() doesn't grok them. Get-magic is
1887 Perl_looks_like_number(pTHX_ SV *const sv)
1892 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1894 if (SvPOK(sv) || SvPOKp(sv)) {
1895 sbegin = SvPV_nomg_const(sv, len);
1898 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
1899 return grok_number(sbegin, len, NULL);
1903 S_glob_2number(pTHX_ GV * const gv)
1905 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1907 /* We know that all GVs stringify to something that is not-a-number,
1908 so no need to test that. */
1909 if (ckWARN(WARN_NUMERIC))
1911 SV *const buffer = sv_newmortal();
1912 gv_efullname3(buffer, gv, "*");
1913 not_a_number(buffer);
1915 /* We just want something true to return, so that S_sv_2iuv_common
1916 can tail call us and return true. */
1920 /* Actually, ISO C leaves conversion of UV to IV undefined, but
1921 until proven guilty, assume that things are not that bad... */
1926 As 64 bit platforms often have an NV that doesn't preserve all bits of
1927 an IV (an assumption perl has been based on to date) it becomes necessary
1928 to remove the assumption that the NV always carries enough precision to
1929 recreate the IV whenever needed, and that the NV is the canonical form.
1930 Instead, IV/UV and NV need to be given equal rights. So as to not lose
1931 precision as a side effect of conversion (which would lead to insanity
1932 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
1933 1) to distinguish between IV/UV/NV slots that have a valid conversion cached
1934 where precision was lost, and IV/UV/NV slots that have a valid conversion
1935 which has lost no precision
1936 2) to ensure that if a numeric conversion to one form is requested that
1937 would lose precision, the precise conversion (or differently
1938 imprecise conversion) is also performed and cached, to prevent
1939 requests for different numeric formats on the same SV causing
1940 lossy conversion chains. (lossless conversion chains are perfectly
1945 SvIOKp is true if the IV slot contains a valid value
1946 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1947 SvNOKp is true if the NV slot contains a valid value
1948 SvNOK is true only if the NV value is accurate
1951 while converting from PV to NV, check to see if converting that NV to an
1952 IV(or UV) would lose accuracy over a direct conversion from PV to
1953 IV(or UV). If it would, cache both conversions, return NV, but mark
1954 SV as IOK NOKp (ie not NOK).
1956 While converting from PV to IV, check to see if converting that IV to an
1957 NV would lose accuracy over a direct conversion from PV to NV. If it
1958 would, cache both conversions, flag similarly.
1960 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
1961 correctly because if IV & NV were set NV *always* overruled.
1962 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
1963 changes - now IV and NV together means that the two are interchangeable:
1964 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
1966 The benefit of this is that operations such as pp_add know that if
1967 SvIOK is true for both left and right operands, then integer addition
1968 can be used instead of floating point (for cases where the result won't
1969 overflow). Before, floating point was always used, which could lead to
1970 loss of precision compared with integer addition.
1972 * making IV and NV equal status should make maths accurate on 64 bit
1974 * may speed up maths somewhat if pp_add and friends start to use
1975 integers when possible instead of fp. (Hopefully the overhead in
1976 looking for SvIOK and checking for overflow will not outweigh the
1977 fp to integer speedup)
1978 * will slow down integer operations (callers of SvIV) on "inaccurate"
1979 values, as the change from SvIOK to SvIOKp will cause a call into
1980 sv_2iv each time rather than a macro access direct to the IV slot
1981 * should speed up number->string conversion on integers as IV is
1982 favoured when IV and NV are equally accurate
1984 ####################################################################
1985 You had better be using SvIOK_notUV if you want an IV for arithmetic:
1986 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
1987 On the other hand, SvUOK is true iff UV.
1988 ####################################################################
1990 Your mileage will vary depending your CPU's relative fp to integer
1994 #ifndef NV_PRESERVES_UV
1995 # define IS_NUMBER_UNDERFLOW_IV 1
1996 # define IS_NUMBER_UNDERFLOW_UV 2
1997 # define IS_NUMBER_IV_AND_UV 2
1998 # define IS_NUMBER_OVERFLOW_IV 4
1999 # define IS_NUMBER_OVERFLOW_UV 5
2001 /* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
2003 /* For sv_2nv these three cases are "SvNOK and don't bother casting" */
2005 S_sv_2iuv_non_preserve(pTHX_ SV *const sv
2011 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
2012 PERL_UNUSED_CONTEXT;
2014 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));
2015 if (SvNVX(sv) < (NV)IV_MIN) {
2016 (void)SvIOKp_on(sv);
2018 SvIV_set(sv, IV_MIN);
2019 return IS_NUMBER_UNDERFLOW_IV;
2021 if (SvNVX(sv) > (NV)UV_MAX) {
2022 (void)SvIOKp_on(sv);
2025 SvUV_set(sv, UV_MAX);
2026 return IS_NUMBER_OVERFLOW_UV;
2028 (void)SvIOKp_on(sv);
2030 /* Can't use strtol etc to convert this string. (See truth table in
2032 if (SvNVX(sv) <= (UV)IV_MAX) {
2033 SvIV_set(sv, I_V(SvNVX(sv)));
2034 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2035 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
2037 /* Integer is imprecise. NOK, IOKp */
2039 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
2042 SvUV_set(sv, U_V(SvNVX(sv)));
2043 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2044 if (SvUVX(sv) == UV_MAX) {
2045 /* As we know that NVs don't preserve UVs, UV_MAX cannot
2046 possibly be preserved by NV. Hence, it must be overflow.
2048 return IS_NUMBER_OVERFLOW_UV;
2050 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
2052 /* Integer is imprecise. NOK, IOKp */
2054 return IS_NUMBER_OVERFLOW_IV;
2056 #endif /* !NV_PRESERVES_UV*/
2059 S_sv_2iuv_common(pTHX_ SV *const sv)
2061 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
2064 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
2065 * without also getting a cached IV/UV from it at the same time
2066 * (ie PV->NV conversion should detect loss of accuracy and cache
2067 * IV or UV at same time to avoid this. */
2068 /* IV-over-UV optimisation - choose to cache IV if possible */
2070 if (SvTYPE(sv) == SVt_NV)
2071 sv_upgrade(sv, SVt_PVNV);
2073 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
2074 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
2075 certainly cast into the IV range at IV_MAX, whereas the correct
2076 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
2078 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2079 if (Perl_isnan(SvNVX(sv))) {
2085 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2086 SvIV_set(sv, I_V(SvNVX(sv)));
2087 if (SvNVX(sv) == (NV) SvIVX(sv)
2088 #ifndef NV_PRESERVES_UV
2089 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2090 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2091 /* Don't flag it as "accurately an integer" if the number
2092 came from a (by definition imprecise) NV operation, and
2093 we're outside the range of NV integer precision */
2097 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2099 /* scalar has trailing garbage, eg "42a" */
2101 DEBUG_c(PerlIO_printf(Perl_debug_log,
2102 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (precise)\n",
2108 /* IV not precise. No need to convert from PV, as NV
2109 conversion would already have cached IV if it detected
2110 that PV->IV would be better than PV->NV->IV
2111 flags already correct - don't set public IOK. */
2112 DEBUG_c(PerlIO_printf(Perl_debug_log,
2113 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (imprecise)\n",
2118 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2119 but the cast (NV)IV_MIN rounds to a the value less (more
2120 negative) than IV_MIN which happens to be equal to SvNVX ??
2121 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2122 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2123 (NV)UVX == NVX are both true, but the values differ. :-(
2124 Hopefully for 2s complement IV_MIN is something like
2125 0x8000000000000000 which will be exact. NWC */
2128 SvUV_set(sv, U_V(SvNVX(sv)));
2130 (SvNVX(sv) == (NV) SvUVX(sv))
2131 #ifndef NV_PRESERVES_UV
2132 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2133 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2134 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2135 /* Don't flag it as "accurately an integer" if the number
2136 came from a (by definition imprecise) NV operation, and
2137 we're outside the range of NV integer precision */
2143 DEBUG_c(PerlIO_printf(Perl_debug_log,
2144 "0x%"UVxf" 2iv(%"UVuf" => %"IVdf") (as unsigned)\n",
2150 else if (SvPOKp(sv)) {
2152 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2153 /* We want to avoid a possible problem when we cache an IV/ a UV which
2154 may be later translated to an NV, and the resulting NV is not
2155 the same as the direct translation of the initial string
2156 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2157 be careful to ensure that the value with the .456 is around if the
2158 NV value is requested in the future).
2160 This means that if we cache such an IV/a UV, we need to cache the
2161 NV as well. Moreover, we trade speed for space, and do not
2162 cache the NV if we are sure it's not needed.
2165 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2166 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2167 == IS_NUMBER_IN_UV) {
2168 /* It's definitely an integer, only upgrade to PVIV */
2169 if (SvTYPE(sv) < SVt_PVIV)
2170 sv_upgrade(sv, SVt_PVIV);
2172 } else if (SvTYPE(sv) < SVt_PVNV)
2173 sv_upgrade(sv, SVt_PVNV);
2175 /* If NVs preserve UVs then we only use the UV value if we know that
2176 we aren't going to call atof() below. If NVs don't preserve UVs
2177 then the value returned may have more precision than atof() will
2178 return, even though value isn't perfectly accurate. */
2179 if ((numtype & (IS_NUMBER_IN_UV
2180 #ifdef NV_PRESERVES_UV
2183 )) == IS_NUMBER_IN_UV) {
2184 /* This won't turn off the public IOK flag if it was set above */
2185 (void)SvIOKp_on(sv);
2187 if (!(numtype & IS_NUMBER_NEG)) {
2189 if (value <= (UV)IV_MAX) {
2190 SvIV_set(sv, (IV)value);
2192 /* it didn't overflow, and it was positive. */
2193 SvUV_set(sv, value);
2197 /* 2s complement assumption */
2198 if (value <= (UV)IV_MIN) {
2199 SvIV_set(sv, -(IV)value);
2201 /* Too negative for an IV. This is a double upgrade, but
2202 I'm assuming it will be rare. */
2203 if (SvTYPE(sv) < SVt_PVNV)
2204 sv_upgrade(sv, SVt_PVNV);
2208 SvNV_set(sv, -(NV)value);
2209 SvIV_set(sv, IV_MIN);
2213 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2214 will be in the previous block to set the IV slot, and the next
2215 block to set the NV slot. So no else here. */
2217 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2218 != IS_NUMBER_IN_UV) {
2219 /* It wasn't an (integer that doesn't overflow the UV). */
2220 SvNV_set(sv, Atof(SvPVX_const(sv)));
2222 if (! numtype && ckWARN(WARN_NUMERIC))
2225 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%" NVgf ")\n",
2226 PTR2UV(sv), SvNVX(sv)));
2228 #ifdef NV_PRESERVES_UV
2229 (void)SvIOKp_on(sv);
2231 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2232 SvIV_set(sv, I_V(SvNVX(sv)));
2233 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2236 NOOP; /* Integer is imprecise. NOK, IOKp */
2238 /* UV will not work better than IV */
2240 if (SvNVX(sv) > (NV)UV_MAX) {
2242 /* Integer is inaccurate. NOK, IOKp, is UV */
2243 SvUV_set(sv, UV_MAX);
2245 SvUV_set(sv, U_V(SvNVX(sv)));
2246 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2247 NV preservse UV so can do correct comparison. */
2248 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2251 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
2256 #else /* NV_PRESERVES_UV */
2257 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2258 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
2259 /* The IV/UV slot will have been set from value returned by
2260 grok_number above. The NV slot has just been set using
2263 assert (SvIOKp(sv));
2265 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2266 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2267 /* Small enough to preserve all bits. */
2268 (void)SvIOKp_on(sv);
2270 SvIV_set(sv, I_V(SvNVX(sv)));
2271 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2273 /* Assumption: first non-preserved integer is < IV_MAX,
2274 this NV is in the preserved range, therefore: */
2275 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2277 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);
2281 0 0 already failed to read UV.
2282 0 1 already failed to read UV.
2283 1 0 you won't get here in this case. IV/UV
2284 slot set, public IOK, Atof() unneeded.
2285 1 1 already read UV.
2286 so there's no point in sv_2iuv_non_preserve() attempting
2287 to use atol, strtol, strtoul etc. */
2289 sv_2iuv_non_preserve (sv, numtype);
2291 sv_2iuv_non_preserve (sv);
2295 #endif /* NV_PRESERVES_UV */
2296 /* It might be more code efficient to go through the entire logic above
2297 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2298 gets complex and potentially buggy, so more programmer efficient
2299 to do it this way, by turning off the public flags: */
2301 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2305 if (isGV_with_GP(sv))
2306 return glob_2number(MUTABLE_GV(sv));
2308 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2310 if (SvTYPE(sv) < SVt_IV)
2311 /* Typically the caller expects that sv_any is not NULL now. */
2312 sv_upgrade(sv, SVt_IV);
2313 /* Return 0 from the caller. */
2320 =for apidoc sv_2iv_flags
2322 Return the integer value of an SV, doing any necessary string
2323 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2324 Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2330 Perl_sv_2iv_flags(pTHX_ SV *const sv, const I32 flags)
2332 PERL_ARGS_ASSERT_SV_2IV_FLAGS;
2334 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2335 && SvTYPE(sv) != SVt_PVFM);
2337 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2343 if (flags & SV_SKIP_OVERLOAD)
2345 tmpstr = AMG_CALLunary(sv, numer_amg);
2346 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2347 return SvIV(tmpstr);
2350 return PTR2IV(SvRV(sv));
2353 if (SvVALID(sv) || isREGEXP(sv)) {
2354 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2355 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2356 In practice they are extremely unlikely to actually get anywhere
2357 accessible by user Perl code - the only way that I'm aware of is when
2358 a constant subroutine which is used as the second argument to index.
2360 Regexps have no SvIVX and SvNVX fields.
2362 assert(isREGEXP(sv) || SvPOKp(sv));
2365 const char * const ptr =
2366 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2368 = grok_number(ptr, SvCUR(sv), &value);
2370 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2371 == IS_NUMBER_IN_UV) {
2372 /* It's definitely an integer */
2373 if (numtype & IS_NUMBER_NEG) {
2374 if (value < (UV)IV_MIN)
2377 if (value < (UV)IV_MAX)
2382 if (ckWARN(WARN_NUMERIC))
2385 return I_V(Atof(ptr));
2389 if (SvTHINKFIRST(sv)) {
2390 #ifdef PERL_OLD_COPY_ON_WRITE
2392 sv_force_normal_flags(sv, 0);
2395 if (SvREADONLY(sv) && !SvOK(sv)) {
2396 if (ckWARN(WARN_UNINITIALIZED))
2403 if (S_sv_2iuv_common(aTHX_ sv))
2407 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"IVdf")\n",
2408 PTR2UV(sv),SvIVX(sv)));
2409 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
2413 =for apidoc sv_2uv_flags
2415 Return the unsigned integer value of an SV, doing any necessary string
2416 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2417 Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
2423 Perl_sv_2uv_flags(pTHX_ SV *const sv, const I32 flags)
2425 PERL_ARGS_ASSERT_SV_2UV_FLAGS;
2427 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2433 if (flags & SV_SKIP_OVERLOAD)
2435 tmpstr = AMG_CALLunary(sv, numer_amg);
2436 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2437 return SvUV(tmpstr);
2440 return PTR2UV(SvRV(sv));
2443 if (SvVALID(sv) || isREGEXP(sv)) {
2444 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2445 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2446 Regexps have no SvIVX and SvNVX fields. */
2447 assert(isREGEXP(sv) || SvPOKp(sv));
2450 const char * const ptr =
2451 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2453 = grok_number(ptr, SvCUR(sv), &value);
2455 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2456 == IS_NUMBER_IN_UV) {
2457 /* It's definitely an integer */
2458 if (!(numtype & IS_NUMBER_NEG))
2462 if (ckWARN(WARN_NUMERIC))
2465 return U_V(Atof(ptr));
2469 if (SvTHINKFIRST(sv)) {
2470 #ifdef PERL_OLD_COPY_ON_WRITE
2472 sv_force_normal_flags(sv, 0);
2475 if (SvREADONLY(sv) && !SvOK(sv)) {
2476 if (ckWARN(WARN_UNINITIALIZED))
2483 if (S_sv_2iuv_common(aTHX_ sv))
2487 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2uv(%"UVuf")\n",
2488 PTR2UV(sv),SvUVX(sv)));
2489 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
2493 =for apidoc sv_2nv_flags
2495 Return the num value of an SV, doing any necessary string or integer
2496 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2497 Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
2503 Perl_sv_2nv_flags(pTHX_ SV *const sv, const I32 flags)
2505 PERL_ARGS_ASSERT_SV_2NV_FLAGS;
2507 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2508 && SvTYPE(sv) != SVt_PVFM);
2509 if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) {
2510 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2511 the same flag bit as SVf_IVisUV, so must not let them cache NVs.
2512 Regexps have no SvIVX and SvNVX fields. */
2514 if (flags & SV_GMAGIC)
2518 if (SvPOKp(sv) && !SvIOKp(sv)) {
2519 ptr = SvPVX_const(sv);
2521 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
2522 !grok_number(ptr, SvCUR(sv), NULL))
2528 return (NV)SvUVX(sv);
2530 return (NV)SvIVX(sv);
2536 ptr = RX_WRAPPED((REGEXP *)sv);
2539 assert(SvTYPE(sv) >= SVt_PVMG);
2540 /* This falls through to the report_uninit near the end of the
2542 } else if (SvTHINKFIRST(sv)) {
2547 if (flags & SV_SKIP_OVERLOAD)
2549 tmpstr = AMG_CALLunary(sv, numer_amg);
2550 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2551 return SvNV(tmpstr);
2554 return PTR2NV(SvRV(sv));
2556 #ifdef PERL_OLD_COPY_ON_WRITE
2558 sv_force_normal_flags(sv, 0);
2561 if (SvREADONLY(sv) && !SvOK(sv)) {
2562 if (ckWARN(WARN_UNINITIALIZED))
2567 if (SvTYPE(sv) < SVt_NV) {
2568 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2569 sv_upgrade(sv, SVt_NV);
2571 STORE_NUMERIC_LOCAL_SET_STANDARD();
2572 PerlIO_printf(Perl_debug_log,
2573 "0x%"UVxf" num(%" NVgf ")\n",
2574 PTR2UV(sv), SvNVX(sv));
2575 RESTORE_NUMERIC_LOCAL();
2578 else if (SvTYPE(sv) < SVt_PVNV)
2579 sv_upgrade(sv, SVt_PVNV);
2584 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
2585 #ifdef NV_PRESERVES_UV
2591 /* Only set the public NV OK flag if this NV preserves the IV */
2592 /* Check it's not 0xFFFFFFFFFFFFFFFF */
2594 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
2595 : (SvIVX(sv) == I_V(SvNVX(sv))))
2601 else if (SvPOKp(sv)) {
2603 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2604 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
2606 #ifdef NV_PRESERVES_UV
2607 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2608 == IS_NUMBER_IN_UV) {
2609 /* It's definitely an integer */
2610 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
2612 SvNV_set(sv, Atof(SvPVX_const(sv)));
2618 SvNV_set(sv, Atof(SvPVX_const(sv)));
2619 /* Only set the public NV OK flag if this NV preserves the value in
2620 the PV at least as well as an IV/UV would.
2621 Not sure how to do this 100% reliably. */
2622 /* if that shift count is out of range then Configure's test is
2623 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2625 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2626 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2627 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
2628 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2629 /* Can't use strtol etc to convert this string, so don't try.
2630 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2633 /* value has been set. It may not be precise. */
2634 if ((numtype & IS_NUMBER_NEG) && (value > (UV)IV_MIN)) {
2635 /* 2s complement assumption for (UV)IV_MIN */
2636 SvNOK_on(sv); /* Integer is too negative. */
2641 if (numtype & IS_NUMBER_NEG) {
2642 SvIV_set(sv, -(IV)value);
2643 } else if (value <= (UV)IV_MAX) {
2644 SvIV_set(sv, (IV)value);
2646 SvUV_set(sv, value);
2650 if (numtype & IS_NUMBER_NOT_INT) {
2651 /* I believe that even if the original PV had decimals,
2652 they are lost beyond the limit of the FP precision.
2653 However, neither is canonical, so both only get p
2654 flags. NWC, 2000/11/25 */
2655 /* Both already have p flags, so do nothing */
2657 const NV nv = SvNVX(sv);
2658 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2659 if (SvIVX(sv) == I_V(nv)) {
2662 /* It had no "." so it must be integer. */
2666 /* between IV_MAX and NV(UV_MAX).
2667 Could be slightly > UV_MAX */
2669 if (numtype & IS_NUMBER_NOT_INT) {
2670 /* UV and NV both imprecise. */
2672 const UV nv_as_uv = U_V(nv);
2674 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2683 /* It might be more code efficient to go through the entire logic above
2684 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2685 gets complex and potentially buggy, so more programmer efficient
2686 to do it this way, by turning off the public flags: */
2688 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2689 #endif /* NV_PRESERVES_UV */
2692 if (isGV_with_GP(sv)) {
2693 glob_2number(MUTABLE_GV(sv));
2697 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2699 assert (SvTYPE(sv) >= SVt_NV);
2700 /* Typically the caller expects that sv_any is not NULL now. */
2701 /* XXX Ilya implies that this is a bug in callers that assume this
2702 and ideally should be fixed. */
2706 STORE_NUMERIC_LOCAL_SET_STANDARD();
2707 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2nv(%" NVgf ")\n",
2708 PTR2UV(sv), SvNVX(sv));
2709 RESTORE_NUMERIC_LOCAL();
2717 Return an SV with the numeric value of the source SV, doing any necessary
2718 reference or overload conversion. You must use the C<SvNUM(sv)> macro to
2719 access this function.
2725 Perl_sv_2num(pTHX_ SV *const sv)
2727 PERL_ARGS_ASSERT_SV_2NUM;
2732 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
2733 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
2734 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2735 return sv_2num(tmpsv);
2737 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2740 /* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2741 * UV as a string towards the end of buf, and return pointers to start and
2744 * We assume that buf is at least TYPE_CHARS(UV) long.
2748 S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
2750 char *ptr = buf + TYPE_CHARS(UV);
2751 char * const ebuf = ptr;
2754 PERL_ARGS_ASSERT_UIV_2BUF;
2766 *--ptr = '0' + (char)(uv % 10);
2774 /* Helper for sv_2pv_flags and sv_vcatpvfn_flags. If the NV is an
2775 * infinity or a not-a-number, writes the appropriate strings to the
2776 * buffer, including a zero byte. On success returns the written length,
2777 * excluding the zero byte, on failure returns zero. */
2779 S_infnan_copy(NV nv, char* buffer, size_t maxlen) {
2784 if (Perl_isinf(nv)) {
2794 else if (Perl_isnan(nv)) {
2798 /* XXX output the payload mantissa bits as "(hhh...)" */
2803 return s - buffer - 1;
2808 =for apidoc sv_2pv_flags
2810 Returns a pointer to the string value of an SV, and sets *lp to its length.
2811 If flags includes SV_GMAGIC, does an mg_get() first. Coerces sv to a
2812 string if necessary. Normally invoked via the C<SvPV_flags> macro.
2813 C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
2819 Perl_sv_2pv_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
2823 PERL_ARGS_ASSERT_SV_2PV_FLAGS;
2825 assert (SvTYPE(sv) != SVt_PVAV && SvTYPE(sv) != SVt_PVHV
2826 && SvTYPE(sv) != SVt_PVFM);
2827 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2832 if (flags & SV_SKIP_OVERLOAD)
2834 tmpstr = AMG_CALLunary(sv, string_amg);
2835 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
2836 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2838 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2842 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2843 if (flags & SV_CONST_RETURN) {
2844 pv = (char *) SvPVX_const(tmpstr);
2846 pv = (flags & SV_MUTABLE_RETURN)
2847 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
2850 *lp = SvCUR(tmpstr);
2852 pv = sv_2pv_flags(tmpstr, lp, flags);
2865 SV *const referent = SvRV(sv);
2869 retval = buffer = savepvn("NULLREF", len);
2870 } else if (SvTYPE(referent) == SVt_REGEXP &&
2871 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
2872 amagic_is_enabled(string_amg))) {
2873 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
2877 /* If the regex is UTF-8 we want the containing scalar to
2878 have an UTF-8 flag too */
2885 *lp = RX_WRAPLEN(re);
2887 return RX_WRAPPED(re);
2889 const char *const typestr = sv_reftype(referent, 0);
2890 const STRLEN typelen = strlen(typestr);
2891 UV addr = PTR2UV(referent);
2892 const char *stashname = NULL;
2893 STRLEN stashnamelen = 0; /* hush, gcc */
2894 const char *buffer_end;
2896 if (SvOBJECT(referent)) {
2897 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
2900 stashname = HEK_KEY(name);
2901 stashnamelen = HEK_LEN(name);
2903 if (HEK_UTF8(name)) {
2909 stashname = "__ANON__";
2912 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
2913 + 2 * sizeof(UV) + 2 /* )\0 */;
2915 len = typelen + 3 /* (0x */
2916 + 2 * sizeof(UV) + 2 /* )\0 */;
2919 Newx(buffer, len, char);
2920 buffer_end = retval = buffer + len;
2922 /* Working backwards */
2926 *--retval = PL_hexdigit[addr & 15];
2927 } while (addr >>= 4);
2933 memcpy(retval, typestr, typelen);
2937 retval -= stashnamelen;
2938 memcpy(retval, stashname, stashnamelen);
2940 /* retval may not necessarily have reached the start of the
2942 assert (retval >= buffer);
2944 len = buffer_end - retval - 1; /* -1 for that \0 */
2956 if (flags & SV_MUTABLE_RETURN)
2957 return SvPVX_mutable(sv);
2958 if (flags & SV_CONST_RETURN)
2959 return (char *)SvPVX_const(sv);
2964 /* I'm assuming that if both IV and NV are equally valid then
2965 converting the IV is going to be more efficient */
2966 const U32 isUIOK = SvIsUV(sv);
2967 char buf[TYPE_CHARS(UV)];
2971 if (SvTYPE(sv) < SVt_PVIV)
2972 sv_upgrade(sv, SVt_PVIV);
2973 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
2975 /* inlined from sv_setpvn */
2976 s = SvGROW_mutable(sv, len + 1);
2977 Move(ptr, s, len, char);
2982 else if (SvNOK(sv)) {
2983 if (SvTYPE(sv) < SVt_PVNV)
2984 sv_upgrade(sv, SVt_PVNV);
2985 if (SvNVX(sv) == 0.0) {
2986 s = SvGROW_mutable(sv, 2);
2991 /* The +20 is pure guesswork. Configure test needed. --jhi */
2992 s = SvGROW_mutable(sv, NV_DIG + 20);
2994 len = S_infnan_copy(SvNVX(sv), s, 5);
2999 /* some Xenix systems wipe out errno here */
3001 #ifndef USE_LOCALE_NUMERIC
3002 PERL_UNUSED_RESULT(Gconvert(SvNVX(sv), NV_DIG, 0, s));
3006 DECLARE_STORE_LC_NUMERIC_SET_TO_NEEDED();
3007 PERL_UNUSED_RESULT(Gconvert(SvNVX(sv), NV_DIG, 0, s));
3009 /* If the radix character is UTF-8, and actually is in the
3010 * output, turn on the UTF-8 flag for the scalar */
3011 if (PL_numeric_local
3012 && PL_numeric_radix_sv && SvUTF8(PL_numeric_radix_sv)
3013 && instr(s, SvPVX_const(PL_numeric_radix_sv)))
3017 RESTORE_LC_NUMERIC();
3020 /* We don't call SvPOK_on(), because it may come to
3021 * pass that the locale changes so that the
3022 * stringification we just did is no longer correct. We
3023 * will have to re-stringify every time it is needed */
3030 else if (isGV_with_GP(sv)) {
3031 GV *const gv = MUTABLE_GV(sv);
3032 SV *const buffer = sv_newmortal();
3034 gv_efullname3(buffer, gv, "*");
3036 assert(SvPOK(buffer));
3040 *lp = SvCUR(buffer);
3041 return SvPVX(buffer);
3043 else if (isREGEXP(sv)) {
3044 if (lp) *lp = RX_WRAPLEN((REGEXP *)sv);
3045 return RX_WRAPPED((REGEXP *)sv);
3050 if (flags & SV_UNDEF_RETURNS_NULL)
3052 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
3054 /* Typically the caller expects that sv_any is not NULL now. */
3055 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
3056 sv_upgrade(sv, SVt_PV);
3061 const STRLEN len = s - SvPVX_const(sv);
3066 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
3067 PTR2UV(sv),SvPVX_const(sv)));
3068 if (flags & SV_CONST_RETURN)
3069 return (char *)SvPVX_const(sv);
3070 if (flags & SV_MUTABLE_RETURN)
3071 return SvPVX_mutable(sv);
3076 =for apidoc sv_copypv
3078 Copies a stringified representation of the source SV into the
3079 destination SV. Automatically performs any necessary mg_get and
3080 coercion of numeric values into strings. Guaranteed to preserve
3081 UTF8 flag even from overloaded objects. Similar in nature to
3082 sv_2pv[_flags] but operates directly on an SV instead of just the
3083 string. Mostly uses sv_2pv_flags to do its work, except when that
3084 would lose the UTF-8'ness of the PV.
3086 =for apidoc sv_copypv_nomg
3088 Like sv_copypv, but doesn't invoke get magic first.
3090 =for apidoc sv_copypv_flags
3092 Implementation of sv_copypv and sv_copypv_nomg. Calls get magic iff flags
3099 Perl_sv_copypv(pTHX_ SV *const dsv, SV *const ssv)
3101 PERL_ARGS_ASSERT_SV_COPYPV;
3103 sv_copypv_flags(dsv, ssv, 0);
3107 Perl_sv_copypv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
3112 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
3114 if ((flags & SV_GMAGIC) && SvGMAGICAL(ssv))
3116 s = SvPV_nomg_const(ssv,len);
3117 sv_setpvn(dsv,s,len);
3125 =for apidoc sv_2pvbyte
3127 Return a pointer to the byte-encoded representation of the SV, and set *lp
3128 to its length. May cause the SV to be downgraded from UTF-8 as a
3131 Usually accessed via the C<SvPVbyte> macro.
3137 Perl_sv_2pvbyte(pTHX_ SV *sv, STRLEN *const lp)
3139 PERL_ARGS_ASSERT_SV_2PVBYTE;
3142 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3143 || isGV_with_GP(sv) || SvROK(sv)) {
3144 SV *sv2 = sv_newmortal();
3145 sv_copypv_nomg(sv2,sv);
3148 sv_utf8_downgrade(sv,0);
3149 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3153 =for apidoc sv_2pvutf8
3155 Return a pointer to the UTF-8-encoded representation of the SV, and set *lp
3156 to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3158 Usually accessed via the C<SvPVutf8> macro.
3164 Perl_sv_2pvutf8(pTHX_ SV *sv, STRLEN *const lp)
3166 PERL_ARGS_ASSERT_SV_2PVUTF8;
3168 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3169 || isGV_with_GP(sv) || SvROK(sv))
3170 sv = sv_mortalcopy(sv);
3173 sv_utf8_upgrade_nomg(sv);
3174 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3179 =for apidoc sv_2bool
3181 This macro is only used by sv_true() or its macro equivalent, and only if
3182 the latter's argument is neither SvPOK, SvIOK nor SvNOK.
3183 It calls sv_2bool_flags with the SV_GMAGIC flag.
3185 =for apidoc sv_2bool_flags
3187 This function is only used by sv_true() and friends, and only if
3188 the latter's argument is neither SvPOK, SvIOK nor SvNOK. If the flags
3189 contain SV_GMAGIC, then it does an mg_get() first.
3196 Perl_sv_2bool_flags(pTHX_ SV *sv, I32 flags)
3198 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
3201 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
3207 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
3208 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv)))) {
3211 if(SvGMAGICAL(sv)) {
3213 goto restart; /* call sv_2bool */
3215 /* expanded SvTRUE_common(sv, (flags = 0, goto restart)) */
3216 else if(!SvOK(sv)) {
3219 else if(SvPOK(sv)) {
3220 svb = SvPVXtrue(sv);
3222 else if((SvFLAGS(sv) & (SVf_IOK|SVf_NOK))) {
3223 svb = (SvIOK(sv) && SvIVX(sv) != 0)
3224 || (SvNOK(sv) && SvNVX(sv) != 0.0);
3228 goto restart; /* call sv_2bool_nomg */
3233 return SvRV(sv) != 0;
3237 RX_WRAPLEN(sv) > 1 || (RX_WRAPLEN(sv) && *RX_WRAPPED(sv) != '0');
3238 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
3242 =for apidoc sv_utf8_upgrade
3244 Converts the PV of an SV to its UTF-8-encoded form.
3245 Forces the SV to string form if it is not already.
3246 Will C<mg_get> on C<sv> if appropriate.
3247 Always sets the SvUTF8 flag to avoid future validity checks even
3248 if the whole string is the same in UTF-8 as not.
3249 Returns the number of bytes in the converted string
3251 This is not a general purpose byte encoding to Unicode interface:
3252 use the Encode extension for that.
3254 =for apidoc sv_utf8_upgrade_nomg
3256 Like sv_utf8_upgrade, but doesn't do magic on C<sv>.
3258 =for apidoc sv_utf8_upgrade_flags
3260 Converts the PV of an SV to its UTF-8-encoded form.
3261 Forces the SV to string form if it is not already.
3262 Always sets the SvUTF8 flag to avoid future validity checks even
3263 if all the bytes are invariant in UTF-8.
3264 If C<flags> has C<SV_GMAGIC> bit set,
3265 will C<mg_get> on C<sv> if appropriate, else not.
3267 If C<flags> has SV_FORCE_UTF8_UPGRADE set, this function assumes that the PV
3268 will expand when converted to UTF-8, and skips the extra work of checking for
3269 that. Typically this flag is used by a routine that has already parsed the
3270 string and found such characters, and passes this information on so that the
3271 work doesn't have to be repeated.
3273 Returns the number of bytes in the converted string.
3275 This is not a general purpose byte encoding to Unicode interface:
3276 use the Encode extension for that.
3278 =for apidoc sv_utf8_upgrade_flags_grow
3280 Like sv_utf8_upgrade_flags, but has an additional parameter C<extra>, which is
3281 the number of unused bytes the string of 'sv' is guaranteed to have free after
3282 it upon return. This allows the caller to reserve extra space that it intends
3283 to fill, to avoid extra grows.
3285 C<sv_utf8_upgrade>, C<sv_utf8_upgrade_nomg>, and C<sv_utf8_upgrade_flags>
3286 are implemented in terms of this function.
3288 Returns the number of bytes in the converted string (not including the spares).
3292 (One might think that the calling routine could pass in the position of the
3293 first variant character when it has set SV_FORCE_UTF8_UPGRADE, so it wouldn't
3294 have to be found again. But that is not the case, because typically when the
3295 caller is likely to use this flag, it won't be calling this routine unless it
3296 finds something that won't fit into a byte. Otherwise it tries to not upgrade
3297 and just use bytes. But some things that do fit into a byte are variants in
3298 utf8, and the caller may not have been keeping track of these.)
3300 If the routine itself changes the string, it adds a trailing C<NUL>. Such a
3301 C<NUL> isn't guaranteed due to having other routines do the work in some input
3302 cases, or if the input is already flagged as being in utf8.
3304 The speed of this could perhaps be improved for many cases if someone wanted to
3305 write a fast function that counts the number of variant characters in a string,
3306 especially if it could return the position of the first one.
3311 Perl_sv_utf8_upgrade_flags_grow(pTHX_ SV *const sv, const I32 flags, STRLEN extra)
3313 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
3315 if (sv == &PL_sv_undef)
3317 if (!SvPOK_nog(sv)) {
3319 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3320 (void) sv_2pv_flags(sv,&len, flags);
3322 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3326 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
3331 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3336 S_sv_uncow(aTHX_ sv, 0);
3339 if (PL_encoding && !(flags & SV_UTF8_NO_ENCODING)) {
3340 sv_recode_to_utf8(sv, PL_encoding);
3341 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3345 if (SvCUR(sv) == 0) {
3346 if (extra) SvGROW(sv, extra);
3347 } else { /* Assume Latin-1/EBCDIC */
3348 /* This function could be much more efficient if we
3349 * had a FLAG in SVs to signal if there are any variant
3350 * chars in the PV. Given that there isn't such a flag
3351 * make the loop as fast as possible (although there are certainly ways
3352 * to speed this up, eg. through vectorization) */
3353 U8 * s = (U8 *) SvPVX_const(sv);
3354 U8 * e = (U8 *) SvEND(sv);
3356 STRLEN two_byte_count = 0;
3358 if (flags & SV_FORCE_UTF8_UPGRADE) goto must_be_utf8;
3360 /* See if really will need to convert to utf8. We mustn't rely on our
3361 * incoming SV being well formed and having a trailing '\0', as certain
3362 * code in pp_formline can send us partially built SVs. */
3366 if (NATIVE_BYTE_IS_INVARIANT(ch)) continue;
3368 t--; /* t already incremented; re-point to first variant */
3373 /* utf8 conversion not needed because all are invariants. Mark as
3374 * UTF-8 even if no variant - saves scanning loop */
3376 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3381 /* Here, the string should be converted to utf8, either because of an
3382 * input flag (two_byte_count = 0), or because a character that
3383 * requires 2 bytes was found (two_byte_count = 1). t points either to
3384 * the beginning of the string (if we didn't examine anything), or to
3385 * the first variant. In either case, everything from s to t - 1 will
3386 * occupy only 1 byte each on output.
3388 * There are two main ways to convert. One is to create a new string
3389 * and go through the input starting from the beginning, appending each
3390 * converted value onto the new string as we go along. It's probably
3391 * best to allocate enough space in the string for the worst possible
3392 * case rather than possibly running out of space and having to
3393 * reallocate and then copy what we've done so far. Since everything
3394 * from s to t - 1 is invariant, the destination can be initialized
3395 * with these using a fast memory copy
3397 * The other way is to figure out exactly how big the string should be
3398 * by parsing the entire input. Then you don't have to make it big
3399 * enough to handle the worst possible case, and more importantly, if
3400 * the string you already have is large enough, you don't have to
3401 * allocate a new string, you can copy the last character in the input
3402 * string to the final position(s) that will be occupied by the
3403 * converted string and go backwards, stopping at t, since everything
3404 * before that is invariant.
3406 * There are advantages and disadvantages to each method.
3408 * In the first method, we can allocate a new string, do the memory
3409 * copy from the s to t - 1, and then proceed through the rest of the
3410 * string byte-by-byte.
3412 * In the second method, we proceed through the rest of the input
3413 * string just calculating how big the converted string will be. Then
3414 * there are two cases:
3415 * 1) if the string has enough extra space to handle the converted
3416 * value. We go backwards through the string, converting until we
3417 * get to the position we are at now, and then stop. If this
3418 * position is far enough along in the string, this method is
3419 * faster than the other method. If the memory copy were the same
3420 * speed as the byte-by-byte loop, that position would be about
3421 * half-way, as at the half-way mark, parsing to the end and back
3422 * is one complete string's parse, the same amount as starting
3423 * over and going all the way through. Actually, it would be
3424 * somewhat less than half-way, as it's faster to just count bytes
3425 * than to also copy, and we don't have the overhead of allocating
3426 * a new string, changing the scalar to use it, and freeing the
3427 * existing one. But if the memory copy is fast, the break-even
3428 * point is somewhere after half way. The counting loop could be
3429 * sped up by vectorization, etc, to move the break-even point
3430 * further towards the beginning.
3431 * 2) if the string doesn't have enough space to handle the converted
3432 * value. A new string will have to be allocated, and one might
3433 * as well, given that, start from the beginning doing the first
3434 * method. We've spent extra time parsing the string and in
3435 * exchange all we've gotten is that we know precisely how big to
3436 * make the new one. Perl is more optimized for time than space,
3437 * so this case is a loser.
3438 * So what I've decided to do is not use the 2nd method unless it is
3439 * guaranteed that a new string won't have to be allocated, assuming
3440 * the worst case. I also decided not to put any more conditions on it
3441 * than this, for now. It seems likely that, since the worst case is
3442 * twice as big as the unknown portion of the string (plus 1), we won't
3443 * be guaranteed enough space, causing us to go to the first method,
3444 * unless the string is short, or the first variant character is near
3445 * the end of it. In either of these cases, it seems best to use the
3446 * 2nd method. The only circumstance I can think of where this would
3447 * be really slower is if the string had once had much more data in it
3448 * than it does now, but there is still a substantial amount in it */
3451 STRLEN invariant_head = t - s;
3452 STRLEN size = invariant_head + (e - t) * 2 + 1 + extra;
3453 if (SvLEN(sv) < size) {
3455 /* Here, have decided to allocate a new string */
3460 Newx(dst, size, U8);
3462 /* If no known invariants at the beginning of the input string,
3463 * set so starts from there. Otherwise, can use memory copy to
3464 * get up to where we are now, and then start from here */
3466 if (invariant_head == 0) {
3469 Copy(s, dst, invariant_head, char);
3470 d = dst + invariant_head;
3474 append_utf8_from_native_byte(*t, &d);
3478 SvPV_free(sv); /* No longer using pre-existing string */
3479 SvPV_set(sv, (char*)dst);
3480 SvCUR_set(sv, d - dst);
3481 SvLEN_set(sv, size);
3484 /* Here, have decided to get the exact size of the string.
3485 * Currently this happens only when we know that there is
3486 * guaranteed enough space to fit the converted string, so
3487 * don't have to worry about growing. If two_byte_count is 0,
3488 * then t points to the first byte of the string which hasn't
3489 * been examined yet. Otherwise two_byte_count is 1, and t
3490 * points to the first byte in the string that will expand to
3491 * two. Depending on this, start examining at t or 1 after t.
3494 U8 *d = t + two_byte_count;
3497 /* Count up the remaining bytes that expand to two */
3500 const U8 chr = *d++;
3501 if (! NATIVE_BYTE_IS_INVARIANT(chr)) two_byte_count++;
3504 /* The string will expand by just the number of bytes that
3505 * occupy two positions. But we are one afterwards because of
3506 * the increment just above. This is the place to put the
3507 * trailing NUL, and to set the length before we decrement */
3509 d += two_byte_count;
3510 SvCUR_set(sv, d - s);
3514 /* Having decremented d, it points to the position to put the
3515 * very last byte of the expanded string. Go backwards through
3516 * the string, copying and expanding as we go, stopping when we
3517 * get to the part that is invariant the rest of the way down */
3521 if (NATIVE_BYTE_IS_INVARIANT(*e)) {
3524 *d-- = UTF8_EIGHT_BIT_LO(*e);
3525 *d-- = UTF8_EIGHT_BIT_HI(*e);
3531 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3532 /* Update pos. We do it at the end rather than during
3533 * the upgrade, to avoid slowing down the common case
3534 * (upgrade without pos).
3535 * pos can be stored as either bytes or characters. Since
3536 * this was previously a byte string we can just turn off
3537 * the bytes flag. */
3538 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3540 mg->mg_flags &= ~MGf_BYTES;
3542 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3543 magic_setutf8(sv,mg); /* clear UTF8 cache */
3548 /* Mark as UTF-8 even if no variant - saves scanning loop */
3554 =for apidoc sv_utf8_downgrade
3556 Attempts to convert the PV of an SV from characters to bytes.
3557 If the PV contains a character that cannot fit
3558 in a byte, this conversion will fail;
3559 in this case, either returns false or, if C<fail_ok> is not
3562 This is not a general purpose Unicode to byte encoding interface:
3563 use the Encode extension for that.
3569 Perl_sv_utf8_downgrade(pTHX_ SV *const sv, const bool fail_ok)
3571 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3573 if (SvPOKp(sv) && SvUTF8(sv)) {
3577 int mg_flags = SV_GMAGIC;
3580 S_sv_uncow(aTHX_ sv, 0);
3582 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3584 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3585 if (mg && mg->mg_len > 0 && mg->mg_flags & MGf_BYTES) {
3586 mg->mg_len = sv_pos_b2u_flags(sv, mg->mg_len,
3587 SV_GMAGIC|SV_CONST_RETURN);
3588 mg_flags = 0; /* sv_pos_b2u does get magic */
3590 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3591 magic_setutf8(sv,mg); /* clear UTF8 cache */
3594 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3596 if (!utf8_to_bytes(s, &len)) {
3601 Perl_croak(aTHX_ "Wide character in %s",
3604 Perl_croak(aTHX_ "Wide character");
3615 =for apidoc sv_utf8_encode
3617 Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3618 flag off so that it looks like octets again.
3624 Perl_sv_utf8_encode(pTHX_ SV *const sv)
3626 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3628 if (SvREADONLY(sv)) {
3629 sv_force_normal_flags(sv, 0);
3631 (void) sv_utf8_upgrade(sv);
3636 =for apidoc sv_utf8_decode
3638 If the PV of the SV is an octet sequence in UTF-8
3639 and contains a multiple-byte character, the C<SvUTF8> flag is turned on
3640 so that it looks like a character. If the PV contains only single-byte
3641 characters, the C<SvUTF8> flag stays off.
3642 Scans PV for validity and returns false if the PV is invalid UTF-8.
3648 Perl_sv_utf8_decode(pTHX_ SV *const sv)
3650 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3653 const U8 *start, *c;
3656 /* The octets may have got themselves encoded - get them back as
3659 if (!sv_utf8_downgrade(sv, TRUE))
3662 /* it is actually just a matter of turning the utf8 flag on, but
3663 * we want to make sure everything inside is valid utf8 first.
3665 c = start = (const U8 *) SvPVX_const(sv);
3666 if (!is_utf8_string(c, SvCUR(sv)))
3668 e = (const U8 *) SvEND(sv);
3671 if (!UTF8_IS_INVARIANT(ch)) {
3676 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3677 /* XXX Is this dead code? XS_utf8_decode calls SvSETMAGIC
3678 after this, clearing pos. Does anything on CPAN
3680 /* adjust pos to the start of a UTF8 char sequence */
3681 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3683 I32 pos = mg->mg_len;
3685 for (c = start + pos; c > start; c--) {
3686 if (UTF8_IS_START(*c))
3689 mg->mg_len = c - start;
3692 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3693 magic_setutf8(sv,mg); /* clear UTF8 cache */
3700 =for apidoc sv_setsv
3702 Copies the contents of the source SV C<ssv> into the destination SV
3703 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3704 function if the source SV needs to be reused. Does not handle 'set' magic on
3705 destination SV. Calls 'get' magic on source SV. Loosely speaking, it
3706 performs a copy-by-value, obliterating any previous content of the
3709 You probably want to use one of the assortment of wrappers, such as
3710 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3711 C<SvSetMagicSV_nosteal>.
3713 =for apidoc sv_setsv_flags
3715 Copies the contents of the source SV C<ssv> into the destination SV
3716 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3717 function if the source SV needs to be reused. Does not handle 'set' magic.
3718 Loosely speaking, it performs a copy-by-value, obliterating any previous
3719 content of the destination.
3720 If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
3721 C<ssv> if appropriate, else not. If the C<flags>
3722 parameter has the C<SV_NOSTEAL> bit set then the
3723 buffers of temps will not be stolen. <sv_setsv>
3724 and C<sv_setsv_nomg> are implemented in terms of this function.
3726 You probably want to use one of the assortment of wrappers, such as
3727 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3728 C<SvSetMagicSV_nosteal>.
3730 This is the primary function for copying scalars, and most other
3731 copy-ish functions and macros use this underneath.
3737 S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
3739 I32 mro_changes = 0; /* 1 = method, 2 = isa, 3 = recursive isa */
3740 HV *old_stash = NULL;
3742 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB;
3744 if (dtype != SVt_PVGV && !isGV_with_GP(dstr)) {
3745 const char * const name = GvNAME(sstr);
3746 const STRLEN len = GvNAMELEN(sstr);
3748 if (dtype >= SVt_PV) {
3754 SvUPGRADE(dstr, SVt_PVGV);
3755 (void)SvOK_off(dstr);
3756 isGV_with_GP_on(dstr);
3758 GvSTASH(dstr) = GvSTASH(sstr);
3760 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
3761 gv_name_set(MUTABLE_GV(dstr), name, len,
3762 GV_ADD | (GvNAMEUTF8(sstr) ? SVf_UTF8 : 0 ));
3763 SvFAKE_on(dstr); /* can coerce to non-glob */
3766 if(GvGP(MUTABLE_GV(sstr))) {
3767 /* If source has method cache entry, clear it */
3769 SvREFCNT_dec(GvCV(sstr));
3770 GvCV_set(sstr, NULL);
3773 /* If source has a real method, then a method is
3776 GvCV((const GV *)sstr) && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3782 /* If dest already had a real method, that's a change as well */
3784 !mro_changes && GvGP(MUTABLE_GV(dstr)) && GvCVu((const GV *)dstr)
3785 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3790 /* We don't need to check the name of the destination if it was not a
3791 glob to begin with. */
3792 if(dtype == SVt_PVGV) {
3793 const char * const name = GvNAME((const GV *)dstr);
3796 /* The stash may have been detached from the symbol table, so
3798 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3802 const STRLEN len = GvNAMELEN(dstr);
3803 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
3804 || (len == 1 && name[0] == ':')) {
3807 /* Set aside the old stash, so we can reset isa caches on
3809 if((old_stash = GvHV(dstr)))
3810 /* Make sure we do not lose it early. */
3811 SvREFCNT_inc_simple_void_NN(
3812 sv_2mortal((SV *)old_stash)
3817 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
3820 gp_free(MUTABLE_GV(dstr));
3821 GvINTRO_off(dstr); /* one-shot flag */
3822 GvGP_set(dstr, gp_ref(GvGP(sstr)));
3823 if (SvTAINTED(sstr))
3825 if (GvIMPORTED(dstr) != GVf_IMPORTED
3826 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
3828 GvIMPORTED_on(dstr);
3831 if(mro_changes == 2) {
3832 if (GvAV((const GV *)sstr)) {
3834 SV * const sref = (SV *)GvAV((const GV *)dstr);
3835 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
3836 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
3837 AV * const ary = newAV();
3838 av_push(ary, mg->mg_obj); /* takes the refcount */
3839 mg->mg_obj = (SV *)ary;
3841 av_push((AV *)mg->mg_obj, SvREFCNT_inc_simple_NN(dstr));
3843 else sv_magic(sref, dstr, PERL_MAGIC_isa, NULL, 0);
3845 mro_isa_changed_in(GvSTASH(dstr));
3847 else if(mro_changes == 3) {
3848 HV * const stash = GvHV(dstr);
3849 if(old_stash ? (HV *)HvENAME_get(old_stash) : stash)
3855 else if(mro_changes) mro_method_changed_in(GvSTASH(dstr));
3856 if (GvIO(dstr) && dtype == SVt_PVGV) {
3857 DEBUG_o(Perl_deb(aTHX_
3858 "glob_assign_glob clearing PL_stashcache\n"));
3859 /* It's a cache. It will rebuild itself quite happily.
3860 It's a lot of effort to work out exactly which key (or keys)
3861 might be invalidated by the creation of the this file handle.
3863 hv_clear(PL_stashcache);
3869 S_glob_assign_ref(pTHX_ SV *const dstr, SV *const sstr)
3871 SV * const sref = SvRV(sstr);
3873 const int intro = GvINTRO(dstr);
3876 const U32 stype = SvTYPE(sref);
3878 PERL_ARGS_ASSERT_GLOB_ASSIGN_REF;
3881 GvINTRO_off(dstr); /* one-shot flag */
3882 GvLINE(dstr) = CopLINE(PL_curcop);
3883 GvEGV(dstr) = MUTABLE_GV(dstr);
3888 location = (SV **) &(GvGP(dstr)->gp_cv); /* XXX bypassing GvCV_set */
3889 import_flag = GVf_IMPORTED_CV;
3892 location = (SV **) &GvHV(dstr);
3893 import_flag = GVf_IMPORTED_HV;
3896 location = (SV **) &GvAV(dstr);
3897 import_flag = GVf_IMPORTED_AV;
3900 location = (SV **) &GvIOp(dstr);
3903 location = (SV **) &GvFORM(dstr);
3906 location = &GvSV(dstr);
3907 import_flag = GVf_IMPORTED_SV;
3910 if (stype == SVt_PVCV) {
3911 /*if (GvCVGEN(dstr) && (GvCV(dstr) != (const CV *)sref || GvCVGEN(dstr))) {*/
3912 if (GvCVGEN(dstr)) {
3913 SvREFCNT_dec(GvCV(dstr));
3914 GvCV_set(dstr, NULL);
3915 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
3918 /* SAVEt_GVSLOT takes more room on the savestack and has more
3919 overhead in leave_scope than SAVEt_GENERIC_SV. But for CVs
3920 leave_scope needs access to the GV so it can reset method
3921 caches. We must use SAVEt_GVSLOT whenever the type is
3922 SVt_PVCV, even if the stash is anonymous, as the stash may
3923 gain a name somehow before leave_scope. */
3924 if (stype == SVt_PVCV) {
3925 /* There is no save_pushptrptrptr. Creating it for this
3926 one call site would be overkill. So inline the ss add
3930 SS_ADD_PTR(location);
3931 SS_ADD_PTR(SvREFCNT_inc(*location));
3932 SS_ADD_UV(SAVEt_GVSLOT);
3935 else SAVEGENERICSV(*location);
3938 if (stype == SVt_PVCV && (*location != sref || GvCVGEN(dstr))) {
3939 CV* const cv = MUTABLE_CV(*location);
3941 if (!GvCVGEN((const GV *)dstr) &&
3942 (CvROOT(cv) || CvXSUB(cv)) &&
3943 /* redundant check that avoids creating the extra SV
3944 most of the time: */
3945 (CvCONST(cv) || ckWARN(WARN_REDEFINE)))
3947 SV * const new_const_sv =
3948 CvCONST((const CV *)sref)
3949 ? cv_const_sv((const CV *)sref)
3951 report_redefined_cv(
3952 sv_2mortal(Perl_newSVpvf(aTHX_
3955 HvNAME_HEK(GvSTASH((const GV *)dstr))
3957 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr)))
3960 CvCONST((const CV *)sref) ? &new_const_sv : NULL
3964 cv_ckproto_len_flags(cv, (const GV *)dstr,
3965 SvPOK(sref) ? CvPROTO(sref) : NULL,
3966 SvPOK(sref) ? CvPROTOLEN(sref) : 0,
3967 SvPOK(sref) ? SvUTF8(sref) : 0);
3969 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
3970 GvASSUMECV_on(dstr);
3971 if(GvSTASH(dstr)) { /* sub foo { 1 } sub bar { 2 } *bar = \&foo */
3972 if (intro && GvREFCNT(dstr) > 1) {
3973 /* temporary remove extra savestack's ref */
3975 gv_method_changed(dstr);
3978 else gv_method_changed(dstr);
3981 *location = SvREFCNT_inc_simple_NN(sref);
3982 if (import_flag && !(GvFLAGS(dstr) & import_flag)
3983 && CopSTASH_ne(PL_curcop, GvSTASH(dstr))) {
3984 GvFLAGS(dstr) |= import_flag;
3986 if (stype == SVt_PVHV) {
3987 const char * const name = GvNAME((GV*)dstr);
3988 const STRLEN len = GvNAMELEN(dstr);
3991 (len > 1 && name[len-2] == ':' && name[len-1] == ':')
3992 || (len == 1 && name[0] == ':')
3994 && (!dref || HvENAME_get(dref))
3997 (HV *)sref, (HV *)dref,
4003 stype == SVt_PVAV && sref != dref
4004 && strEQ(GvNAME((GV*)dstr), "ISA")
4005 /* The stash may have been detached from the symbol table, so
4006 check its name before doing anything. */
4007 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
4010 MAGIC * const omg = dref && SvSMAGICAL(dref)
4011 ? mg_find(dref, PERL_MAGIC_isa)
4013 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
4014 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
4015 AV * const ary = newAV();
4016 av_push(ary, mg->mg_obj); /* takes the refcount */
4017 mg->mg_obj = (SV *)ary;
4020 if (SvTYPE(omg->mg_obj) == SVt_PVAV) {
4021 SV **svp = AvARRAY((AV *)omg->mg_obj);
4022 I32 items = AvFILLp((AV *)omg->mg_obj) + 1;
4026 SvREFCNT_inc_simple_NN(*svp++)
4032 SvREFCNT_inc_simple_NN(omg->mg_obj)
4036 av_push((AV *)mg->mg_obj,SvREFCNT_inc_simple_NN(dstr));
4041 sref, omg ? omg->mg_obj : dstr, PERL_MAGIC_isa, NULL, 0
4043 mg = mg_find(sref, PERL_MAGIC_isa);
4045 /* Since the *ISA assignment could have affected more than
4046 one stash, don't call mro_isa_changed_in directly, but let
4047 magic_clearisa do it for us, as it already has the logic for
4048 dealing with globs vs arrays of globs. */
4050 Perl_magic_clearisa(aTHX_ NULL, mg);
4052 else if (stype == SVt_PVIO) {
4053 DEBUG_o(Perl_deb(aTHX_ "glob_assign_ref clearing PL_stashcache\n"));
4054 /* It's a cache. It will rebuild itself quite happily.
4055 It's a lot of effort to work out exactly which key (or keys)
4056 might be invalidated by the creation of the this file handle.
4058 hv_clear(PL_stashcache);
4062 if (!intro) SvREFCNT_dec(dref);
4063 if (SvTAINTED(sstr))
4071 #ifdef PERL_DEBUG_READONLY_COW
4072 # include <sys/mman.h>
4074 # ifndef PERL_MEMORY_DEBUG_HEADER_SIZE
4075 # define PERL_MEMORY_DEBUG_HEADER_SIZE 0
4079 Perl_sv_buf_to_ro(pTHX_ SV *sv)
4081 struct perl_memory_debug_header * const header =
4082 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4083 const MEM_SIZE len = header->size;
4084 PERL_ARGS_ASSERT_SV_BUF_TO_RO;
4085 # ifdef PERL_TRACK_MEMPOOL
4086 if (!header->readonly) header->readonly = 1;
4088 if (mprotect(header, len, PROT_READ))
4089 Perl_warn(aTHX_ "mprotect RW for COW string %p %lu failed with %d",
4090 header, len, errno);
4094 S_sv_buf_to_rw(pTHX_ SV *sv)
4096 struct perl_memory_debug_header * const header =
4097 (struct perl_memory_debug_header *)(SvPVX(sv)-PERL_MEMORY_DEBUG_HEADER_SIZE);
4098 const MEM_SIZE len = header->size;
4099 PERL_ARGS_ASSERT_SV_BUF_TO_RW;
4100 if (mprotect(header, len, PROT_READ|PROT_WRITE))
4101 Perl_warn(aTHX_ "mprotect for COW string %p %lu failed with %d",
4102 header, len, errno);
4103 # ifdef PERL_TRACK_MEMPOOL
4104 header->readonly = 0;
4109 # define sv_buf_to_ro(sv) NOOP
4110 # define sv_buf_to_rw(sv) NOOP
4114 Perl_sv_setsv_flags(pTHX_ SV *dstr, SV* sstr, const I32 flags)
4120 PERL_ARGS_ASSERT_SV_SETSV_FLAGS;
4125 if (SvIS_FREED(dstr)) {
4126 Perl_croak(aTHX_ "panic: attempt to copy value %" SVf
4127 " to a freed scalar %p", SVfARG(sstr), (void *)dstr);
4129 SV_CHECK_THINKFIRST_COW_DROP(dstr);
4131 sstr = &PL_sv_undef;
4132 if (SvIS_FREED(sstr)) {
4133 Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p to %p",
4134 (void*)sstr, (void*)dstr);
4136 stype = SvTYPE(sstr);
4137 dtype = SvTYPE(dstr);
4139 /* There's a lot of redundancy below but we're going for speed here */
4144 if (dtype != SVt_PVGV && dtype != SVt_PVLV) {
4145 (void)SvOK_off(dstr);
4153 sv_upgrade(dstr, SVt_IV);
4157 sv_upgrade(dstr, SVt_PVIV);
4161 goto end_of_first_switch;
4163 (void)SvIOK_only(dstr);
4164 SvIV_set(dstr, SvIVX(sstr));
4167 /* SvTAINTED can only be true if the SV has taint magic, which in
4168 turn means that the SV type is PVMG (or greater). This is the
4169 case statement for SVt_IV, so this cannot be true (whatever gcov
4171 assert(!SvTAINTED(sstr));
4176 if (dtype < SVt_PV && dtype != SVt_IV)
4177 sv_upgrade(dstr, SVt_IV);
4185 sv_upgrade(dstr, SVt_NV);
4189 sv_upgrade(dstr, SVt_PVNV);
4193 goto end_of_first_switch;
4195 SvNV_set(dstr, SvNVX(sstr));
4196 (void)SvNOK_only(dstr);
4197 /* SvTAINTED can only be true if the SV has taint magic, which in
4198 turn means that the SV type is PVMG (or greater). This is the
4199 case statement for SVt_NV, so this cannot be true (whatever gcov
4201 assert(!SvTAINTED(sstr));
4208 sv_upgrade(dstr, SVt_PV);
4211 if (dtype < SVt_PVIV)
4212 sv_upgrade(dstr, SVt_PVIV);
4215 if (dtype < SVt_PVNV)
4216 sv_upgrade(dstr, SVt_PVNV);
4220 const char * const type = sv_reftype(sstr,0);
4222 /* diag_listed_as: Bizarre copy of %s */
4223 Perl_croak(aTHX_ "Bizarre copy of %s in %s", type, OP_DESC(PL_op));
4225 Perl_croak(aTHX_ "Bizarre copy of %s", type);
4227 NOT_REACHED; /* NOTREACHED */
4231 if (dtype < SVt_REGEXP)
4233 if (dtype >= SVt_PV) {
4239 sv_upgrade(dstr, SVt_REGEXP);
4247 if (SvGMAGICAL(sstr) && (flags & SV_GMAGIC)) {
4249 if (SvTYPE(sstr) != stype)
4250 stype = SvTYPE(sstr);
4252 if (isGV_with_GP(sstr) && dtype <= SVt_PVLV) {
4253 glob_assign_glob(dstr, sstr, dtype);
4256 if (stype == SVt_PVLV)
4258 if (isREGEXP(sstr)) goto upgregexp;
4259 SvUPGRADE(dstr, SVt_PVNV);
4262 SvUPGRADE(dstr, (svtype)stype);
4264 end_of_first_switch:
4266 /* dstr may have been upgraded. */
4267 dtype = SvTYPE(dstr);
4268 sflags = SvFLAGS(sstr);
4270 if (dtype == SVt_PVCV) {
4271 /* Assigning to a subroutine sets the prototype. */
4274 const char *const ptr = SvPV_const(sstr, len);
4276 SvGROW(dstr, len + 1);
4277 Copy(ptr, SvPVX(dstr), len + 1, char);
4278 SvCUR_set(dstr, len);
4280 SvFLAGS(dstr) |= sflags & SVf_UTF8;
4281 CvAUTOLOAD_off(dstr);
4286 else if (dtype == SVt_PVAV || dtype == SVt_PVHV || dtype == SVt_PVFM) {
4287 const char * const type = sv_reftype(dstr,0);
4289 /* diag_listed_as: Cannot copy to %s */
4290 Perl_croak(aTHX_ "Cannot copy to %s in %s", type, OP_DESC(PL_op));
4292 Perl_croak(aTHX_ "Cannot copy to %s", type);
4293 } else if (sflags & SVf_ROK) {
4294 if (isGV_with_GP(dstr)
4295 && SvTYPE(SvRV(sstr)) == SVt_PVGV && isGV_with_GP(SvRV(sstr))) {
4298 if (GvIMPORTED(dstr) != GVf_IMPORTED
4299 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
4301 GvIMPORTED_on(dstr);
4306 glob_assign_glob(dstr, sstr, dtype);
4310 if (dtype >= SVt_PV) {
4311 if (isGV_with_GP(dstr)) {
4312 glob_assign_ref(dstr, sstr);
4315 if (SvPVX_const(dstr)) {
4321 (void)SvOK_off(dstr);
4322 SvRV_set(dstr, SvREFCNT_inc(SvRV(sstr)));
4323 SvFLAGS(dstr) |= sflags & SVf_ROK;
4324 assert(!(sflags & SVp_NOK));
4325 assert(!(sflags & SVp_IOK));
4326 assert(!(sflags & SVf_NOK));
4327 assert(!(sflags & SVf_IOK));
4329 else if (isGV_with_GP(dstr)) {
4330 if (!(sflags & SVf_OK)) {
4331 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4332 "Undefined value assigned to typeglob");
4335 GV *gv = gv_fetchsv_nomg(sstr, GV_ADD, SVt_PVGV);
4336 if (dstr != (const SV *)gv) {
4337 const char * const name = GvNAME((const GV *)dstr);
4338 const STRLEN len = GvNAMELEN(dstr);
4339 HV *old_stash = NULL;
4340 bool reset_isa = FALSE;
4341 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
4342 || (len == 1 && name[0] == ':')) {
4343 /* Set aside the old stash, so we can reset isa caches
4344 on its subclasses. */
4345 if((old_stash = GvHV(dstr))) {
4346 /* Make sure we do not lose it early. */
4347 SvREFCNT_inc_simple_void_NN(
4348 sv_2mortal((SV *)old_stash)
4355 SvREFCNT_inc_simple_void_NN(sv_2mortal(dstr));
4356 gp_free(MUTABLE_GV(dstr));
4358 GvGP_set(dstr, gp_ref(GvGP(gv)));
4361 HV * const stash = GvHV(dstr);
4363 old_stash ? (HV *)HvENAME_get(old_stash) : stash
4373 else if ((dtype == SVt_REGEXP || dtype == SVt_PVLV)
4374 && (stype == SVt_REGEXP || isREGEXP(sstr))) {
4375 reg_temp_copy((REGEXP*)dstr, (REGEXP*)sstr);
4377 else if (sflags & SVp_POK) {
4378 const STRLEN cur = SvCUR(sstr);
4379 const STRLEN len = SvLEN(sstr);
4382 * We have three basic ways to copy the string:
4388 * Which we choose is based on various factors. The following
4389 * things are listed in order of speed, fastest to slowest:
4391 * - Copying a short string
4392 * - Copy-on-write bookkeeping
4394 * - Copying a long string
4396 * We swipe the string (steal the string buffer) if the SV on the
4397 * rhs is about to be freed anyway (TEMP and refcnt==1). This is a
4398 * big win on long strings. It should be a win on short strings if
4399 * SvPVX_const(dstr) has to be allocated. If not, it should not
4400 * slow things down, as SvPVX_const(sstr) would have been freed
4403 * We also steal the buffer from a PADTMP (operator target) if it
4404 * is ‘long enough’. For short strings, a swipe does not help
4405 * here, as it causes more malloc calls the next time the target
4406 * is used. Benchmarks show that even if SvPVX_const(dstr) has to
4407 * be allocated it is still not worth swiping PADTMPs for short
4408 * strings, as the savings here are small.
4410 * If the rhs is already flagged as a copy-on-write string and COW
4411 * is possible here, we use copy-on-write and make both SVs share
4412 * the string buffer.
4414 * If the rhs is not flagged as copy-on-write, then we see whether
4415 * it is worth upgrading it to such. If the lhs already has a buf-
4416 * fer big enough and the string is short, we skip it and fall back
4417 * to method 3, since memcpy is faster for short strings than the
4418 * later bookkeeping overhead that copy-on-write entails.
4420 * If there is no buffer on the left, or the buffer is too small,
4421 * then we use copy-on-write.
4424 /* Whichever path we take through the next code, we want this true,
4425 and doing it now facilitates the COW check. */
4426 (void)SvPOK_only(dstr);
4430 /* slated for free anyway (and not COW)? */
4431 (sflags & (SVs_TEMP|SVf_IsCOW)) == SVs_TEMP
4432 /* or a swipable TARG */
4433 || ((sflags & (SVs_PADTMP|SVf_READONLY|SVf_IsCOW))
4435 /* whose buffer is worth stealing */
4436 && CHECK_COWBUF_THRESHOLD(cur,len)
4439 !(sflags & SVf_OOK) && /* and not involved in OOK hack? */
4440 (!(flags & SV_NOSTEAL)) &&
4441 /* and we're allowed to steal temps */
4442 SvREFCNT(sstr) == 1 && /* and no other references to it? */
4443 len) /* and really is a string */
4444 { /* Passes the swipe test. */
4445 if (SvPVX_const(dstr)) /* we know that dtype >= SVt_PV */
4447 SvPV_set(dstr, SvPVX_mutable(sstr));
4448 SvLEN_set(dstr, SvLEN(sstr));
4449 SvCUR_set(dstr, SvCUR(sstr));
4452 (void)SvOK_off(sstr); /* NOTE: nukes most SvFLAGS on sstr */
4453 SvPV_set(sstr, NULL);
4458 else if (flags & SV_COW_SHARED_HASH_KEYS
4460 #ifdef PERL_OLD_COPY_ON_WRITE
4461 ( sflags & SVf_IsCOW
4462 || ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4463 && (SvFLAGS(dstr) & CAN_COW_MASK) == CAN_COW_FLAGS
4464 && SvTYPE(sstr) >= SVt_PVIV && len
4467 #elif defined(PERL_NEW_COPY_ON_WRITE)
4470 ( (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4471 /* If this is a regular (non-hek) COW, only so
4472 many COW "copies" are possible. */
4473 && CowREFCNT(sstr) != SV_COW_REFCNT_MAX ))
4474 : ( (sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4475 && !(SvFLAGS(dstr) & SVf_BREAK)
4476 && CHECK_COW_THRESHOLD(cur,len) && cur+1 < len
4477 && (CHECK_COWBUF_THRESHOLD(cur,len) || SvLEN(dstr) < cur+1)
4481 && !(SvFLAGS(dstr) & SVf_BREAK)
4484 /* Either it's a shared hash key, or it's suitable for
4487 PerlIO_printf(Perl_debug_log, "Copy on write: sstr --> dstr\n");
4492 if (!(sflags & SVf_IsCOW)) {
4494 # ifdef PERL_OLD_COPY_ON_WRITE
4495 /* Make the source SV into a loop of 1.
4496 (about to become 2) */
4497 SV_COW_NEXT_SV_SET(sstr, sstr);
4499 CowREFCNT(sstr) = 0;
4503 if (SvPVX_const(dstr)) { /* we know that dtype >= SVt_PV */
4509 # ifdef PERL_OLD_COPY_ON_WRITE
4510 assert (SvTYPE(dstr) >= SVt_PVIV);
4511 /* SvIsCOW_normal */
4512 /* splice us in between source and next-after-source. */
4513 SV_COW_NEXT_SV_SET(dstr, SV_COW_NEXT_SV(sstr));
4514 SV_COW_NEXT_SV_SET(sstr, dstr);
4516 if (sflags & SVf_IsCOW) {
4521 SvPV_set(dstr, SvPVX_mutable(sstr));
4526 /* SvIsCOW_shared_hash */
4527 DEBUG_C(PerlIO_printf(Perl_debug_log,
4528 "Copy on write: Sharing hash\n"));
4530 assert (SvTYPE(dstr) >= SVt_PV);
4532 HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)))));
4534 SvLEN_set(dstr, len);
4535 SvCUR_set(dstr, cur);
4538 /* Failed the swipe test, and we cannot do copy-on-write either.
4539 Have to copy the string. */
4540 SvGROW(dstr, cur + 1); /* inlined from sv_setpvn */
4541 Move(SvPVX_const(sstr),SvPVX(dstr),cur,char);
4542 SvCUR_set(dstr, cur);
4543 *SvEND(dstr) = '\0';
4545 if (sflags & SVp_NOK) {
4546 SvNV_set(dstr, SvNVX(sstr));
4548 if (sflags & SVp_IOK) {
4549 SvIV_set(dstr, SvIVX(sstr));
4550 /* Must do this otherwise some other overloaded use of 0x80000000
4551 gets confused. I guess SVpbm_VALID */
4552 if (sflags & SVf_IVisUV)
4555 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_NOK|SVp_NOK|SVf_UTF8);
4557 const MAGIC * const smg = SvVSTRING_mg(sstr);
4559 sv_magic(dstr, NULL, PERL_MAGIC_vstring,
4560 smg->mg_ptr, smg->mg_len);
4561 SvRMAGICAL_on(dstr);
4565 else if (sflags & (SVp_IOK|SVp_NOK)) {
4566 (void)SvOK_off(dstr);
4567 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_IVisUV|SVf_NOK|SVp_NOK);
4568 if (sflags & SVp_IOK) {
4569 /* XXXX Do we want to set IsUV for IV(ROK)? Be extra safe... */
4570 SvIV_set(dstr, SvIVX(sstr));
4572 if (sflags & SVp_NOK) {
4573 SvNV_set(dstr, SvNVX(sstr));
4577 if (isGV_with_GP(sstr)) {
4578 gv_efullname3(dstr, MUTABLE_GV(sstr), "*");
4581 (void)SvOK_off(dstr);
4583 if (SvTAINTED(sstr))
4588 =for apidoc sv_setsv_mg
4590 Like C<sv_setsv>, but also handles 'set' magic.
4596 Perl_sv_setsv_mg(pTHX_ SV *const dstr, SV *const sstr)
4598 PERL_ARGS_ASSERT_SV_SETSV_MG;
4600 sv_setsv(dstr,sstr);
4605 # ifdef PERL_OLD_COPY_ON_WRITE
4606 # define SVt_COW SVt_PVIV
4608 # define SVt_COW SVt_PV
4611 Perl_sv_setsv_cow(pTHX_ SV *dstr, SV *sstr)
4613 STRLEN cur = SvCUR(sstr);
4614 STRLEN len = SvLEN(sstr);
4616 #if defined(PERL_DEBUG_READONLY_COW) && defined(PERL_NEW_COPY_ON_WRITE)
4617 const bool already = cBOOL(SvIsCOW(sstr));
4620 PERL_ARGS_ASSERT_SV_SETSV_COW;
4623 PerlIO_printf(Perl_debug_log, "Fast copy on write: %p -> %p\n",
4624 (void*)sstr, (void*)dstr);
4631 if (SvTHINKFIRST(dstr))
4632 sv_force_normal_flags(dstr, SV_COW_DROP_PV);
4633 else if (SvPVX_const(dstr))
4634 Safefree(SvPVX_mutable(dstr));
4638 SvUPGRADE(dstr, SVt_COW);
4640 assert (SvPOK(sstr));
4641 assert (SvPOKp(sstr));
4642 # ifdef PERL_OLD_COPY_ON_WRITE
4643 assert (!SvIOK(sstr));
4644 assert (!SvIOKp(sstr));
4645 assert (!SvNOK(sstr));
4646 assert (!SvNOKp(sstr));
4649 if (SvIsCOW(sstr)) {
4651 if (SvLEN(sstr) == 0) {
4652 /* source is a COW shared hash key. */
4653 DEBUG_C(PerlIO_printf(Perl_debug_log,
4654 "Fast copy on write: Sharing hash\n"));
4655 new_pv = HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr))));
4658 # ifdef PERL_OLD_COPY_ON_WRITE
4659 SV_COW_NEXT_SV_SET(dstr, SV_COW_NEXT_SV(sstr));
4661 assert(SvCUR(sstr)+1 < SvLEN(sstr));
4662 assert(CowREFCNT(sstr) < SV_COW_REFCNT_MAX);
4665 assert ((SvFLAGS(sstr) & CAN_COW_MASK) == CAN_COW_FLAGS);
4666 SvUPGRADE(sstr, SVt_COW);
4668 DEBUG_C(PerlIO_printf(Perl_debug_log,
4669 "Fast copy on write: Converting sstr to COW\n"));
4670 # ifdef PERL_OLD_COPY_ON_WRITE
4671 SV_COW_NEXT_SV_SET(dstr, sstr);
4673 CowREFCNT(sstr) = 0;
4676 # ifdef PERL_OLD_COPY_ON_WRITE
4677 SV_COW_NEXT_SV_SET(sstr, dstr);
4679 # ifdef PERL_DEBUG_READONLY_COW
4680 if (already) sv_buf_to_rw(sstr);
4684 new_pv = SvPVX_mutable(sstr);
4688 SvPV_set(dstr, new_pv);
4689 SvFLAGS(dstr) = (SVt_COW|SVf_POK|SVp_POK|SVf_IsCOW);
4692 SvLEN_set(dstr, len);
4693 SvCUR_set(dstr, cur);
4702 =for apidoc sv_setpvn
4704 Copies a string (possibly containing embedded C<NUL> characters) into an SV.
4705 The C<len> parameter indicates the number of
4706 bytes to be copied. If the C<ptr> argument is NULL the SV will become
4707 undefined. Does not handle 'set' magic. See C<sv_setpvn_mg>.
4713 Perl_sv_setpvn(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4717 PERL_ARGS_ASSERT_SV_SETPVN;
4719 SV_CHECK_THINKFIRST_COW_DROP(sv);
4725 /* len is STRLEN which is unsigned, need to copy to signed */
4728 Perl_croak(aTHX_ "panic: sv_setpvn called with negative strlen %"
4731 SvUPGRADE(sv, SVt_PV);
4733 dptr = SvGROW(sv, len + 1);
4734 Move(ptr,dptr,len,char);
4737 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4739 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4743 =for apidoc sv_setpvn_mg
4745 Like C<sv_setpvn>, but also handles 'set' magic.
4751 Perl_sv_setpvn_mg(pTHX_ SV *const sv, const char *const ptr, const STRLEN len)
4753 PERL_ARGS_ASSERT_SV_SETPVN_MG;
4755 sv_setpvn(sv,ptr,len);
4760 =for apidoc sv_setpv
4762 Copies a string into an SV. The string must be terminated with a C<NUL>
4764 Does not handle 'set' magic. See C<sv_setpv_mg>.
4770 Perl_sv_setpv(pTHX_ SV *const sv, const char *const ptr)
4774 PERL_ARGS_ASSERT_SV_SETPV;
4776 SV_CHECK_THINKFIRST_COW_DROP(sv);
4782 SvUPGRADE(sv, SVt_PV);
4784 SvGROW(sv, len + 1);
4785 Move(ptr,SvPVX(sv),len+1,char);
4787 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4789 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4793 =for apidoc sv_setpv_mg
4795 Like C<sv_setpv>, but also handles 'set' magic.
4801 Perl_sv_setpv_mg(pTHX_ SV *const sv, const char *const ptr)
4803 PERL_ARGS_ASSERT_SV_SETPV_MG;
4810 Perl_sv_sethek(pTHX_ SV *const sv, const HEK *const hek)
4812 PERL_ARGS_ASSERT_SV_SETHEK;
4818 if (HEK_LEN(hek) == HEf_SVKEY) {
4819 sv_setsv(sv, *(SV**)HEK_KEY(hek));
4822 const int flags = HEK_FLAGS(hek);
4823 if (flags & HVhek_WASUTF8) {
4824 STRLEN utf8_len = HEK_LEN(hek);
4825 char *as_utf8 = (char *)bytes_to_utf8((U8*)HEK_KEY(hek), &utf8_len);
4826 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
4829 } else if (flags & HVhek_UNSHARED) {
4830 sv_setpvn(sv, HEK_KEY(hek), HEK_LEN(hek));
4833 else SvUTF8_off(sv);
4837 SV_CHECK_THINKFIRST_COW_DROP(sv);
4838 SvUPGRADE(sv, SVt_PV);
4840 SvPV_set(sv,(char *)HEK_KEY(share_hek_hek(hek)));
4841 SvCUR_set(sv, HEK_LEN(hek));
4847 else SvUTF8_off(sv);
4855 =for apidoc sv_usepvn_flags
4857 Tells an SV to use C<ptr> to find its string value. Normally the
4858 string is stored inside the SV, but sv_usepvn allows the SV to use an
4859 outside string. The C<ptr> should point to memory that was allocated
4860 by L<Newx|perlclib/Memory Management and String Handling>. It must be
4861 the start of a Newx-ed block of memory, and not a pointer to the
4862 middle of it (beware of L<OOK|perlguts/Offsets> and copy-on-write),
4863 and not be from a non-Newx memory allocator like C<malloc>. The
4864 string length, C<len>, must be supplied. By default this function
4865 will C<Renew> (i.e. realloc, move) the memory pointed to by C<ptr>,
4866 so that pointer should not be freed or used by the programmer after
4867 giving it to sv_usepvn, and neither should any pointers from "behind"
4868 that pointer (e.g. ptr + 1) be used.
4870 If C<flags> & SV_SMAGIC is true, will call SvSETMAGIC. If C<flags> &
4871 SV_HAS_TRAILING_NUL is true, then C<ptr[len]> must be C<NUL>, and the realloc
4872 will be skipped (i.e. the buffer is actually at least 1 byte longer than
4873 C<len>, and already meets the requirements for storing in C<SvPVX>).
4879 Perl_sv_usepvn_flags(pTHX_ SV *const sv, char *ptr, const STRLEN len, const U32 flags)
4883 PERL_ARGS_ASSERT_SV_USEPVN_FLAGS;
4885 SV_CHECK_THINKFIRST_COW_DROP(sv);
4886 SvUPGRADE(sv, SVt_PV);
4889 if (flags & SV_SMAGIC)
4893 if (SvPVX_const(sv))
4897 if (flags & SV_HAS_TRAILING_NUL)
4898 assert(ptr[len] == '\0');
4901 allocate = (flags & SV_HAS_TRAILING_NUL)
4903 #ifdef Perl_safesysmalloc_size
4906 PERL_STRLEN_ROUNDUP(len + 1);
4908 if (flags & SV_HAS_TRAILING_NUL) {
4909 /* It's long enough - do nothing.
4910 Specifically Perl_newCONSTSUB is relying on this. */
4913 /* Force a move to shake out bugs in callers. */
4914 char *new_ptr = (char*)safemalloc(allocate);
4915 Copy(ptr, new_ptr, len, char);
4916 PoisonFree(ptr,len,char);
4920 ptr = (char*) saferealloc (ptr, allocate);
4923 #ifdef Perl_safesysmalloc_size
4924 SvLEN_set(sv, Perl_safesysmalloc_size(ptr));
4926 SvLEN_set(sv, allocate);
4930 if (!(flags & SV_HAS_TRAILING_NUL)) {
4933 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4935 if (flags & SV_SMAGIC)
4939 #ifdef PERL_OLD_COPY_ON_WRITE
4940 /* Need to do this *after* making the SV normal, as we need the buffer
4941 pointer to remain valid until after we've copied it. If we let go too early,
4942 another thread could invalidate it by unsharing last of the same hash key
4943 (which it can do by means other than releasing copy-on-write Svs)
4944 or by changing the other copy-on-write SVs in the loop. */
4946 S_sv_release_COW(pTHX_ SV *sv, const char *pvx, SV *after)
4948 PERL_ARGS_ASSERT_SV_RELEASE_COW;
4950 { /* this SV was SvIsCOW_normal(sv) */
4951 /* we need to find the SV pointing to us. */
4952 SV *current = SV_COW_NEXT_SV(after);
4954 if (current == sv) {
4955 /* The SV we point to points back to us (there were only two of us
4957 Hence other SV is no longer copy on write either. */
4959 sv_buf_to_rw(after);
4961 /* We need to follow the pointers around the loop. */
4963 while ((next = SV_COW_NEXT_SV(current)) != sv) {
4966 /* don't loop forever if the structure is bust, and we have
4967 a pointer into a closed loop. */
4968 assert (current != after);
4969 assert (SvPVX_const(current) == pvx);
4971 /* Make the SV before us point to the SV after us. */
4972 SV_COW_NEXT_SV_SET(current, after);
4978 =for apidoc sv_force_normal_flags
4980 Undo various types of fakery on an SV, where fakery means
4981 "more than" a string: if the PV is a shared string, make
4982 a private copy; if we're a ref, stop refing; if we're a glob, downgrade to
4983 an xpvmg; if we're a copy-on-write scalar, this is the on-write time when
4984 we do the copy, and is also used locally; if this is a
4985 vstring, drop the vstring magic. If C<SV_COW_DROP_PV> is set
4986 then a copy-on-write scalar drops its PV buffer (if any) and becomes
4987 SvPOK_off rather than making a copy. (Used where this
4988 scalar is about to be set to some other value.) In addition,
4989 the C<flags> parameter gets passed to C<sv_unref_flags()>
4990 when unreffing. C<sv_force_normal> calls this function
4991 with flags set to 0.
4993 This function is expected to be used to signal to perl that this SV is
4994 about to be written to, and any extra book-keeping needs to be taken care
4995 of. Hence, it croaks on read-only values.
5001 S_sv_uncow(pTHX_ SV * const sv, const U32 flags)
5003 assert(SvIsCOW(sv));
5006 const char * const pvx = SvPVX_const(sv);
5007 const STRLEN len = SvLEN(sv);
5008 const STRLEN cur = SvCUR(sv);
5009 # ifdef PERL_OLD_COPY_ON_WRITE
5010 /* next COW sv in the loop. If len is 0 then this is a shared-hash
5011 key scalar, so we mustn't attempt to call SV_COW_NEXT_SV(), as
5012 we'll fail an assertion. */
5013 SV * const next = len ? SV_COW_NEXT_SV(sv) : 0;
5017 PerlIO_printf(Perl_debug_log,
5018 "Copy on write: Force normal %ld\n",
5023 # ifdef PERL_NEW_COPY_ON_WRITE
5024 if (len && CowREFCNT(sv) == 0)
5025 /* We own the buffer ourselves. */
5031 /* This SV doesn't own the buffer, so need to Newx() a new one: */
5032 # ifdef PERL_NEW_COPY_ON_WRITE
5033 /* Must do this first, since the macro uses SvPVX. */
5043 if (flags & SV_COW_DROP_PV) {
5044 /* OK, so we don't need to copy our buffer. */
5047 SvGROW(sv, cur + 1);
5048 Move(pvx,SvPVX(sv),cur,char);
5053 # ifdef PERL_OLD_COPY_ON_WRITE
5054 sv_release_COW(sv, pvx, next);
5057 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5064 const char * const pvx = SvPVX_const(sv);
5065 const STRLEN len = SvCUR(sv);
5069 if (flags & SV_COW_DROP_PV) {
5070 /* OK, so we don't need to copy our buffer. */
5073 SvGROW(sv, len + 1);
5074 Move(pvx,SvPVX(sv),len,char);
5077 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
5083 Perl_sv_force_normal_flags(pTHX_ SV *const sv, const U32 flags)
5085 PERL_ARGS_ASSERT_SV_FORCE_NORMAL_FLAGS;
5088 Perl_croak_no_modify();
5089 else if (SvIsCOW(sv))
5090 S_sv_uncow(aTHX_ sv, flags);
5092 sv_unref_flags(sv, flags);
5093 else if (SvFAKE(sv) && isGV_with_GP(sv))
5094 sv_unglob(sv, flags);
5095 else if (SvFAKE(sv) && isREGEXP(sv)) {
5096 /* Need to downgrade the REGEXP to a simple(r) scalar. This is analogous
5097 to sv_unglob. We only need it here, so inline it. */
5098 const bool islv = SvTYPE(sv) == SVt_PVLV;
5099 const svtype new_type =
5100 islv ? SVt_NULL : SvMAGIC(sv) || SvSTASH(sv) ? SVt_PVMG : SVt_PV;
5101 SV *const temp = newSV_type(new_type);
5102 regexp *const temp_p = ReANY((REGEXP *)sv);
5104 if (new_type == SVt_PVMG) {
5105 SvMAGIC_set(temp, SvMAGIC(sv));
5106 SvMAGIC_set(sv, NULL);
5107 SvSTASH_set(temp, SvSTASH(sv));
5108 SvSTASH_set(sv, NULL);
5110 if (!islv) SvCUR_set(temp, SvCUR(sv));
5111 /* Remember that SvPVX is in the head, not the body. But
5112 RX_WRAPPED is in the body. */
5113 assert(ReANY((REGEXP *)sv)->mother_re);
5114 /* Their buffer is already owned by someone else. */
5115 if (flags & SV_COW_DROP_PV) {
5116 /* SvLEN is already 0. For SVt_REGEXP, we have a brand new
5117 zeroed body. For SVt_PVLV, it should have been set to 0
5118 before turning into a regexp. */
5119 assert(!SvLEN(islv ? sv : temp));
5120 sv->sv_u.svu_pv = 0;
5123 sv->sv_u.svu_pv = savepvn(RX_WRAPPED((REGEXP *)sv), SvCUR(sv));
5124 SvLEN_set(islv ? sv : temp, SvCUR(sv)+1);
5128 /* Now swap the rest of the bodies. */
5132 SvFLAGS(sv) &= ~SVTYPEMASK;
5133 SvFLAGS(sv) |= new_type;
5134 SvANY(sv) = SvANY(temp);
5137 SvFLAGS(temp) &= ~(SVTYPEMASK);
5138 SvFLAGS(temp) |= SVt_REGEXP|SVf_FAKE;
5139 SvANY(temp) = temp_p;
5140 temp->sv_u.svu_rx = (regexp *)temp_p;
5142 SvREFCNT_dec_NN(temp);
5144 else if (SvVOK(sv)) sv_unmagic(sv, PERL_MAGIC_vstring);
5150 Efficient removal of characters from the beginning of the string buffer.
5151 SvPOK(sv), or at least SvPOKp(sv), must be true and the C<ptr> must be a
5152 pointer to somewhere inside the string buffer. The C<ptr> becomes the first
5153 character of the adjusted string. Uses the "OOK hack". On return, only
5154 SvPOK(sv) and SvPOKp(sv) among the OK flags will be true.
5156 Beware: after this function returns, C<ptr> and SvPVX_const(sv) may no longer
5157 refer to the same chunk of data.
5159 The unfortunate similarity of this function's name to that of Perl's C<chop>
5160 operator is strictly coincidental. This function works from the left;
5161 C<chop> works from the right.
5167 Perl_sv_chop(pTHX_ SV *const sv, const char *const ptr)
5178 PERL_ARGS_ASSERT_SV_CHOP;
5180 if (!ptr || !SvPOKp(sv))
5182 delta = ptr - SvPVX_const(sv);
5184 /* Nothing to do. */
5187 max_delta = SvLEN(sv) ? SvLEN(sv) : SvCUR(sv);
5188 if (delta > max_delta)
5189 Perl_croak(aTHX_ "panic: sv_chop ptr=%p, start=%p, end=%p",
5190 ptr, SvPVX_const(sv), SvPVX_const(sv) + max_delta);
5191 /* SvPVX(sv) may move in SV_CHECK_THINKFIRST(sv), so don't use ptr any more */
5192 SV_CHECK_THINKFIRST(sv);
5193 SvPOK_only_UTF8(sv);
5196 if (!SvLEN(sv)) { /* make copy of shared string */
5197 const char *pvx = SvPVX_const(sv);
5198 const STRLEN len = SvCUR(sv);
5199 SvGROW(sv, len + 1);
5200 Move(pvx,SvPVX(sv),len,char);
5206 SvOOK_offset(sv, old_delta);
5208 SvLEN_set(sv, SvLEN(sv) - delta);
5209 SvCUR_set(sv, SvCUR(sv) - delta);
5210 SvPV_set(sv, SvPVX(sv) + delta);
5212 p = (U8 *)SvPVX_const(sv);
5215 /* how many bytes were evacuated? we will fill them with sentinel
5216 bytes, except for the part holding the new offset of course. */
5219 evacn += (old_delta < 0x100 ? 1 : 1 + sizeof(STRLEN));
5221 assert(evacn <= delta + old_delta);
5225 /* This sets 'delta' to the accumulated value of all deltas so far */
5229 /* If 'delta' fits in a byte, store it just prior to the new beginning of
5230 * the string; otherwise store a 0 byte there and store 'delta' just prior
5231 * to that, using as many bytes as a STRLEN occupies. Thus it overwrites a
5232 * portion of the chopped part of the string */
5233 if (delta < 0x100) {
5237 p -= sizeof(STRLEN);
5238 Copy((U8*)&delta, p, sizeof(STRLEN), U8);
5242 /* Fill the preceding buffer with sentinals to verify that no-one is
5252 =for apidoc sv_catpvn
5254 Concatenates the string onto the end of the string which is in the SV. The
5255 C<len> indicates number of bytes to copy. If the SV has the UTF-8
5256 status set, then the bytes appended should be valid UTF-8.
5257 Handles 'get' magic, but not 'set' magic. See C<sv_catpvn_mg>.
5259 =for apidoc sv_catpvn_flags
5261 Concatenates the string onto the end of the string which is in the SV. The
5262 C<len> indicates number of bytes to copy. If the SV has the UTF-8
5263 status set, then the bytes appended should be valid UTF-8.
5264 If C<flags> has the C<SV_SMAGIC> bit set, will
5265 C<mg_set> on C<dsv> afterwards if appropriate.
5266 C<sv_catpvn> and C<sv_catpvn_nomg> are implemented
5267 in terms of this function.
5273 Perl_sv_catpvn_flags(pTHX_ SV *const dsv, const char *sstr, const STRLEN slen, const I32 flags)
5276 const char * const dstr = SvPV_force_flags(dsv, dlen, flags);
5278 PERL_ARGS_ASSERT_SV_CATPVN_FLAGS;
5279 assert((flags & (SV_CATBYTES|SV_CATUTF8)) != (SV_CATBYTES|SV_CATUTF8));
5281 if (!(flags & SV_CATBYTES) || !SvUTF8(dsv)) {
5282 if (flags & SV_CATUTF8 && !SvUTF8(dsv)) {
5283 sv_utf8_upgrade_flags_grow(dsv, 0, slen + 1);
5286 else SvGROW(dsv, dlen + slen + 1);
5288 sstr = SvPVX_const(dsv);
5289 Move(sstr, SvPVX(dsv) + dlen, slen, char);
5290 SvCUR_set(dsv, SvCUR(dsv) + slen);
5293 /* We inline bytes_to_utf8, to avoid an extra malloc. */
5294 const char * const send = sstr + slen;
5297 /* Something this code does not account for, which I think is
5298 impossible; it would require the same pv to be treated as
5299 bytes *and* utf8, which would indicate a bug elsewhere. */
5300 assert(sstr != dstr);
5302 SvGROW(dsv, dlen + slen * 2 + 1);
5303 d = (U8 *)SvPVX(dsv) + dlen;
5305 while (sstr < send) {
5306 append_utf8_from_native_byte(*sstr, &d);
5309 SvCUR_set(dsv, d-(const U8 *)SvPVX(dsv));
5312 (void)SvPOK_only_UTF8(dsv); /* validate pointer */
5314 if (flags & SV_SMAGIC)
5319 =for apidoc sv_catsv
5321 Concatenates the string from SV C<ssv> onto the end of the string in SV
5322 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5323 Handles 'get' magic on both SVs, but no 'set' magic. See C<sv_catsv_mg> and
5326 =for apidoc sv_catsv_flags
5328 Concatenates the string from SV C<ssv> onto the end of the string in SV
5329 C<dsv>. If C<ssv> is null, does nothing; otherwise modifies only C<dsv>.
5330 If C<flags> include C<SV_GMAGIC> bit set, will call C<mg_get> on both SVs if
5331 appropriate. If C<flags> include C<SV_SMAGIC>, C<mg_set> will be called on
5332 the modified SV afterward, if appropriate. C<sv_catsv>, C<sv_catsv_nomg>,
5333 and C<sv_catsv_mg> are implemented in terms of this function.
5338 Perl_sv_catsv_flags(pTHX_ SV *const dsv, SV *const ssv, const I32 flags)
5340 PERL_ARGS_ASSERT_SV_CATSV_FLAGS;
5344 const char *spv = SvPV_flags_const(ssv, slen, flags);
5346 if (flags & SV_GMAGIC)
5348 sv_catpvn_flags(dsv, spv, slen,
5349 DO_UTF8(ssv) ? SV_CATUTF8 : SV_CATBYTES);
5350 if (flags & SV_SMAGIC)
5357 =for apidoc sv_catpv
5359 Concatenates the C<NUL>-terminated string onto the end of the string which is
5361 If the SV has the UTF-8 status set, then the bytes appended should be
5362 valid UTF-8. Handles 'get' magic, but not 'set' magic. See C<sv_catpv_mg>.
5367 Perl_sv_catpv(pTHX_ SV *const sv, const char *ptr)
5373 PERL_ARGS_ASSERT_SV_CATPV;
5377 junk = SvPV_force(sv, tlen);
5379 SvGROW(sv, tlen + len + 1);
5381 ptr = SvPVX_const(sv);
5382 Move(ptr,SvPVX(sv)+tlen,len+1,char);
5383 SvCUR_set(sv, SvCUR(sv) + len);
5384 (void)SvPOK_only_UTF8(sv); /* validate pointer */
5389 =for apidoc sv_catpv_flags
5391 Concatenates the C<NUL>-terminated string onto the end of the string which is
5393 If the SV has the UTF-8 status set, then the bytes appended should
5394 be valid UTF-8. If C<flags> has the C<SV_SMAGIC> bit set, will C<mg_set>
5395 on the modified SV if appropriate.
5401 Perl_sv_catpv_flags(pTHX_ SV *dstr, const char *sstr, const I32 flags)
5403 PERL_ARGS_ASSERT_SV_CATPV_FLAGS;
5404 sv_catpvn_flags(dstr, sstr, strlen(sstr), flags);
5408 =for apidoc sv_catpv_mg
5410 Like C<sv_catpv>, but also handles 'set' magic.
5416 Perl_sv_catpv_mg(pTHX_ SV *const sv, const char *const ptr)
5418 PERL_ARGS_ASSERT_SV_CATPV_MG;
5427 Creates a new SV. A non-zero C<len> parameter indicates the number of
5428 bytes of preallocated string space the SV should have. An extra byte for a
5429 trailing C<NUL> is also reserved. (SvPOK is not set for the SV even if string
5430 space is allocated.) The reference count for the new SV is set to 1.
5432 In 5.9.3, newSV() replaces the older NEWSV() API, and drops the first
5433 parameter, I<x>, a debug aid which allowed callers to identify themselves.
5434 This aid has been superseded by a new build option, PERL_MEM_LOG (see
5435 L<perlhacktips/PERL_MEM_LOG>). The older API is still there for use in XS
5436 modules supporting older perls.
5442 Perl_newSV(pTHX_ const STRLEN len)
5448 sv_upgrade(sv, SVt_PV);
5449 SvGROW(sv, len + 1);
5454 =for apidoc sv_magicext
5456 Adds magic to an SV, upgrading it if necessary. Applies the
5457 supplied vtable and returns a pointer to the magic added.
5459 Note that C<sv_magicext> will allow things that C<sv_magic> will not.
5460 In particular, you can add magic to SvREADONLY SVs, and add more than
5461 one instance of the same 'how'.
5463 If C<namlen> is greater than zero then a C<savepvn> I<copy> of C<name> is
5464 stored, if C<namlen> is zero then C<name> is stored as-is and - as another
5465 special case - if C<(name && namlen == HEf_SVKEY)> then C<name> is assumed
5466 to contain an C<SV*> and is stored as-is with its REFCNT incremented.
5468 (This is now used as a subroutine by C<sv_magic>.)
5473 Perl_sv_magicext(pTHX_ SV *const sv, SV *const obj, const int how,
5474 const MGVTBL *const vtable, const char *const name, const I32 namlen)
5478 PERL_ARGS_ASSERT_SV_MAGICEXT;
5480 if (SvTYPE(sv)==SVt_PVAV) { assert (!AvPAD_NAMELIST(sv)); }
5482 SvUPGRADE(sv, SVt_PVMG);
5483 Newxz(mg, 1, MAGIC);
5484 mg->mg_moremagic = SvMAGIC(sv);
5485 SvMAGIC_set(sv, mg);
5487 /* Sometimes a magic contains a reference loop, where the sv and
5488 object refer to each other. To prevent a reference loop that
5489 would prevent such objects being freed, we look for such loops
5490 and if we find one we avoid incrementing the object refcount.
5492 Note we cannot do this to avoid self-tie loops as intervening RV must
5493 have its REFCNT incremented to keep it in existence.
5496 if (!obj || obj == sv ||
5497 how == PERL_MAGIC_arylen ||
5498 how == PERL_MAGIC_symtab ||
5499 (SvTYPE(obj) == SVt_PVGV &&
5500 (GvSV(obj) == sv || GvHV(obj) == (const HV *)sv
5501 || GvAV(obj) == (const AV *)sv || GvCV(obj) == (const CV *)sv
5502 || GvIOp(obj) == (const IO *)sv || GvFORM(obj) == (const CV *)sv)))
5507 mg->mg_obj = SvREFCNT_inc_simple(obj);
5508 mg->mg_flags |= MGf_REFCOUNTED;
5511 /* Normal self-ties simply pass a null object, and instead of
5512 using mg_obj directly, use the SvTIED_obj macro to produce a
5513 new RV as needed. For glob "self-ties", we are tieing the PVIO
5514 with an RV obj pointing to the glob containing the PVIO. In
5515 this case, to avoid a reference loop, we need to weaken the
5519 if (how == PERL_MAGIC_tiedscalar && SvTYPE(sv) == SVt_PVIO &&
5520 obj && SvROK(obj) && GvIO(SvRV(obj)) == (const IO *)sv)
5526 mg->mg_len = namlen;
5529 mg->mg_ptr = savepvn(name, namlen);
5530 else if (namlen == HEf_SVKEY) {
5531 /* Yes, this is casting away const. This is only for the case of
5532 HEf_SVKEY. I think we need to document this aberation of the
5533 constness of the API, rather than making name non-const, as
5534 that change propagating outwards a long way. */
5535 mg->mg_ptr = (char*)SvREFCNT_inc_simple_NN((SV *)name);
5537 mg->mg_ptr = (char *) name;
5539 mg->mg_virtual = (MGVTBL *) vtable;
5546 Perl_sv_magicext_mglob(pTHX_ SV *sv)
5548 PERL_ARGS_ASSERT_SV_MAGICEXT_MGLOB;
5549 if (SvTYPE(sv) == SVt_PVLV && LvTYPE(sv) == 'y') {
5550 /* This sv is only a delegate. //g magic must be attached to
5555 #ifdef PERL_OLD_COPY_ON_WRITE
5557 sv_force_normal_flags(sv, 0);
5559 return sv_magicext(sv, NULL, PERL_MAGIC_regex_global,
5560 &PL_vtbl_mglob, 0, 0);
5564 =for apidoc sv_magic
5566 Adds magic to an SV. First upgrades C<sv> to type C<SVt_PVMG> if
5567 necessary, then adds a new magic item of type C<how> to the head of the
5570 See C<sv_magicext> (which C<sv_magic> now calls) for a description of the
5571 handling of the C<name> and C<namlen> arguments.
5573 You need to use C<sv_magicext> to add magic to SvREADONLY SVs and also
5574 to add more than one instance of the same 'how'.
5580 Perl_sv_magic(pTHX_ SV *const sv, SV *const obj, const int how,
5581 const char *const name, const I32 namlen)
5583 const MGVTBL *vtable;
5586 unsigned int vtable_index;
5588 PERL_ARGS_ASSERT_SV_MAGIC;
5590 if (how < 0 || (unsigned)how >= C_ARRAY_LENGTH(PL_magic_data)
5591 || ((flags = PL_magic_data[how]),
5592 (vtable_index = flags & PERL_MAGIC_VTABLE_MASK)
5593 > magic_vtable_max))
5594 Perl_croak(aTHX_ "Don't know how to handle magic of type \\%o", how);
5596 /* PERL_MAGIC_ext is reserved for use by extensions not perl internals.
5597 Useful for attaching extension internal data to perl vars.
5598 Note that multiple extensions may clash if magical scalars
5599 etc holding private data from one are passed to another. */
5601 vtable = (vtable_index == magic_vtable_max)
5602 ? NULL : PL_magic_vtables + vtable_index;
5604 #ifdef PERL_OLD_COPY_ON_WRITE
5606 sv_force_normal_flags(sv, 0);
5608 if (SvREADONLY(sv)) {
5610 !PERL_MAGIC_TYPE_READONLY_ACCEPTABLE(how)
5613 Perl_croak_no_modify();
5616 if (SvMAGICAL(sv) || (how == PERL_MAGIC_taint && SvTYPE(sv) >= SVt_PVMG)) {
5617 if (SvMAGIC(sv) && (mg = mg_find(sv, how))) {
5618 /* sv_magic() refuses to add a magic of the same 'how' as an
5621 if (how == PERL_MAGIC_taint)
5627 /* Force pos to be stored as characters, not bytes. */
5628 if (SvMAGICAL(sv) && DO_UTF8(sv)
5629 && (mg = mg_find(sv, PERL_MAGIC_regex_global))
5631 && mg->mg_flags & MGf_BYTES) {
5632 mg->mg_len = (SSize_t)sv_pos_b2u_flags(sv, (STRLEN)mg->mg_len,
5634 mg->mg_flags &= ~MGf_BYTES;
5637 /* Rest of work is done else where */
5638 mg = sv_magicext(sv,obj,how,vtable,name,namlen);
5641 case PERL_MAGIC_taint:
5644 case PERL_MAGIC_ext:
5645 case PERL_MAGIC_dbfile:
5652 S_sv_unmagicext_flags(pTHX_ SV *const sv, const int type, MGVTBL *vtbl, const U32 flags)
5659 if (SvTYPE(sv) < SVt_PVMG || !SvMAGIC(sv))
5661 mgp = &(((XPVMG*) SvANY(sv))->xmg_u.xmg_magic);
5662 for (mg = *mgp; mg; mg = *mgp) {
5663 const MGVTBL* const virt = mg->mg_virtual;
5664 if (mg->mg_type == type && (!flags || virt == vtbl)) {
5665 *mgp = mg->mg_moremagic;
5666 if (virt && virt->svt_free)
5667 virt->svt_free(aTHX_ sv, mg);
5668 if (mg->mg_ptr && mg->mg_type != PERL_MAGIC_regex_global) {
5670 Safefree(mg->mg_ptr);
5671 else if (mg->mg_len == HEf_SVKEY)
5672 SvREFCNT_dec(MUTABLE_SV(mg->mg_ptr));
5673 else if (mg->mg_type == PERL_MAGIC_utf8)
5674 Safefree(mg->mg_ptr);
5676 if (mg->mg_flags & MGf_REFCOUNTED)
5677 SvREFCNT_dec(mg->mg_obj);
5681 mgp = &mg->mg_moremagic;
5684 if (SvMAGICAL(sv)) /* if we're under save_magic, wait for restore_magic; */
5685 mg_magical(sv); /* else fix the flags now */
5689 SvFLAGS(sv) |= (SvFLAGS(sv) & (SVp_IOK|SVp_NOK|SVp_POK)) >> PRIVSHIFT;
5695 =for apidoc sv_unmagic
5697 Removes all magic of type C<type> from an SV.
5703 Perl_sv_unmagic(pTHX_ SV *const sv, const int type)
5705 PERL_ARGS_ASSERT_SV_UNMAGIC;
5706 return S_sv_unmagicext_flags(aTHX_ sv, type, NULL, 0);
5710 =for apidoc sv_unmagicext
5712 Removes all magic of type C<type> with the specified C<vtbl> from an SV.
5718 Perl_sv_unmagicext(pTHX_ SV *const sv, const int type, MGVTBL *vtbl)
5720 PERL_ARGS_ASSERT_SV_UNMAGICEXT;
5721 return S_sv_unmagicext_flags(aTHX_ sv, type, vtbl, 1);
5725 =for apidoc sv_rvweaken
5727 Weaken a reference: set the C<SvWEAKREF> flag on this RV; give the
5728 referred-to SV C<PERL_MAGIC_backref> magic if it hasn't already; and
5729 push a back-reference to this RV onto the array of backreferences
5730 associated with that magic. If the RV is magical, set magic will be
5731 called after the RV is cleared.
5737 Perl_sv_rvweaken(pTHX_ SV *const sv)
5741 PERL_ARGS_ASSERT_SV_RVWEAKEN;
5743 if (!SvOK(sv)) /* let undefs pass */
5746 Perl_croak(aTHX_ "Can't weaken a nonreference");
5747 else if (SvWEAKREF(sv)) {
5748 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Reference is already weak");
5751 else if (SvREADONLY(sv)) croak_no_modify();
5753 Perl_sv_add_backref(aTHX_ tsv, sv);
5755 SvREFCNT_dec_NN(tsv);
5759 /* Give tsv backref magic if it hasn't already got it, then push a
5760 * back-reference to sv onto the array associated with the backref magic.
5762 * As an optimisation, if there's only one backref and it's not an AV,
5763 * store it directly in the HvAUX or mg_obj slot, avoiding the need to
5764 * allocate an AV. (Whether the slot holds an AV tells us whether this is
5768 /* A discussion about the backreferences array and its refcount:
5770 * The AV holding the backreferences is pointed to either as the mg_obj of
5771 * PERL_MAGIC_backref, or in the specific case of a HV, from the
5772 * xhv_backreferences field. The array is created with a refcount
5773 * of 2. This means that if during global destruction the array gets
5774 * picked on before its parent to have its refcount decremented by the
5775 * random zapper, it won't actually be freed, meaning it's still there for
5776 * when its parent gets freed.
5778 * When the parent SV is freed, the extra ref is killed by
5779 * Perl_sv_kill_backrefs. The other ref is killed, in the case of magic,
5780 * by mg_free() / MGf_REFCOUNTED, or for a hash, by Perl_hv_kill_backrefs.
5782 * When a single backref SV is stored directly, it is not reference
5787 Perl_sv_add_backref(pTHX_ SV *const tsv, SV *const sv)
5793 PERL_ARGS_ASSERT_SV_ADD_BACKREF;
5795 /* find slot to store array or singleton backref */
5797 if (SvTYPE(tsv) == SVt_PVHV) {
5798 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
5801 mg = mg_find(tsv, PERL_MAGIC_backref);
5803 mg = sv_magicext(tsv, NULL, PERL_MAGIC_backref, &PL_vtbl_backref, NULL, 0);
5804 svp = &(mg->mg_obj);
5807 /* create or retrieve the array */
5809 if ( (!*svp && SvTYPE(sv) == SVt_PVAV)
5810 || (*svp && SvTYPE(*svp) != SVt_PVAV)
5814 mg->mg_flags |= MGf_REFCOUNTED;
5817 SvREFCNT_inc_simple_void_NN(av);
5818 /* av now has a refcnt of 2; see discussion above */
5819 av_extend(av, *svp ? 2 : 1);
5821 /* move single existing backref to the array */
5822 AvARRAY(av)[++AvFILLp(av)] = *svp; /* av_push() */
5827 av = MUTABLE_AV(*svp);
5829 /* optimisation: store single backref directly in HvAUX or mg_obj */
5833 assert(SvTYPE(av) == SVt_PVAV);
5834 if (AvFILLp(av) >= AvMAX(av)) {
5835 av_extend(av, AvFILLp(av)+1);
5838 /* push new backref */
5839 AvARRAY(av)[++AvFILLp(av)] = sv; /* av_push() */
5842 /* delete a back-reference to ourselves from the backref magic associated
5843 * with the SV we point to.
5847 Perl_sv_del_backref(pTHX_ SV *const tsv, SV *const sv)
5851 PERL_ARGS_ASSERT_SV_DEL_BACKREF;
5853 if (SvTYPE(tsv) == SVt_PVHV) {
5855 svp = (SV**)Perl_hv_backreferences_p(aTHX_ MUTABLE_HV(tsv));
5857 else if (SvIS_FREED(tsv) && PL_phase == PERL_PHASE_DESTRUCT) {
5858 /* It's possible for the the last (strong) reference to tsv to have
5859 become freed *before* the last thing holding a weak reference.
5860 If both survive longer than the backreferences array, then when
5861 the referent's reference count drops to 0 and it is freed, it's
5862 not able to chase the backreferences, so they aren't NULLed.
5864 For example, a CV holds a weak reference to its stash. If both the
5865 CV and the stash survive longer than the backreferences array,
5866 and the CV gets picked for the SvBREAK() treatment first,
5867 *and* it turns out that the stash is only being kept alive because
5868 of an our variable in the pad of the CV, then midway during CV
5869 destruction the stash gets freed, but CvSTASH() isn't set to NULL.
5870 It ends up pointing to the freed HV. Hence it's chased in here, and
5871 if this block wasn't here, it would hit the !svp panic just below.
5873 I don't believe that "better" destruction ordering is going to help
5874 here - during global destruction there's always going to be the
5875 chance that something goes out of order. We've tried to make it
5876 foolproof before, and it only resulted in evolutionary pressure on
5877 fools. Which made us look foolish for our hubris. :-(
5883 = SvMAGICAL(tsv) ? mg_find(tsv, PERL_MAGIC_backref) : NULL;
5884 svp = mg ? &(mg->mg_obj) : NULL;
5888 Perl_croak(aTHX_ "panic: del_backref, svp=0");
5890 /* It's possible that sv is being freed recursively part way through the
5891 freeing of tsv. If this happens, the backreferences array of tsv has
5892 already been freed, and so svp will be NULL. If this is the case,
5893 we should not panic. Instead, nothing needs doing, so return. */
5894 if (PL_phase == PERL_PHASE_DESTRUCT && SvREFCNT(tsv) == 0)
5896 Perl_croak(aTHX_ "panic: del_backref, *svp=%p phase=%s refcnt=%" UVuf,
5897 (void*)*svp, PL_phase_names[PL_phase], (UV)SvREFCNT(tsv));
5900 if (SvTYPE(*svp) == SVt_PVAV) {
5904 AV * const av = (AV*)*svp;
5906 assert(!SvIS_FREED(av));
5910 /* for an SV with N weak references to it, if all those
5911 * weak refs are deleted, then sv_del_backref will be called
5912 * N times and O(N^2) compares will be done within the backref
5913 * array. To ameliorate this potential slowness, we:
5914 * 1) make sure this code is as tight as possible;
5915 * 2) when looking for SV, look for it at both the head and tail of the
5916 * array first before searching the rest, since some create/destroy
5917 * patterns will cause the backrefs to be freed in order.
5924 SV **p = &svp[fill];
5925 SV *const topsv = *p;
5932 /* We weren't the last entry.
5933 An unordered list has this property that you
5934 can take the last element off the end to fill
5935 the hole, and it's still an unordered list :-)
5941 break; /* should only be one */
5948 AvFILLp(av) = fill-1;
5950 else if (SvIS_FREED(*svp) && PL_phase == PERL_PHASE_DESTRUCT) {
5951 /* freed AV; skip */
5954 /* optimisation: only a single backref, stored directly */
5956 Perl_croak(aTHX_ "panic: del_backref, *svp=%p, sv=%p",
5957 (void*)*svp, (void*)sv);
5964 Perl_sv_kill_backrefs(pTHX_ SV *const sv, AV *const av)
5970 PERL_ARGS_ASSERT_SV_KILL_BACKREFS;
5975 /* after multiple passes through Perl_sv_clean_all() for a thingy
5976 * that has badly leaked, the backref array may have gotten freed,
5977 * since we only protect it against 1 round of cleanup */
5978 if (SvIS_FREED(av)) {
5979 if (PL_in_clean_all) /* All is fair */
5982 "panic: magic_killbackrefs (freed backref AV/SV)");
5986 is_array = (SvTYPE(av) == SVt_PVAV);
5988 assert(!SvIS_FREED(av));
5991 last = svp + AvFILLp(av);
5994 /* optimisation: only a single backref, stored directly */
6000 while (svp <= last) {
6002 SV *const referrer = *svp;
6003 if (SvWEAKREF(referrer)) {
6004 /* XXX Should we check that it hasn't changed? */
6005 assert(SvROK(referrer));
6006 SvRV_set(referrer, 0);
6008 SvWEAKREF_off(referrer);
6009 SvSETMAGIC(referrer);
6010 } else if (SvTYPE(referrer) == SVt_PVGV ||
6011 SvTYPE(referrer) == SVt_PVLV) {
6012 assert(SvTYPE(sv) == SVt_PVHV); /* stash backref */
6013 /* You lookin' at me? */
6014 assert(GvSTASH(referrer));
6015 assert(GvSTASH(referrer) == (const HV *)sv);
6016 GvSTASH(referrer) = 0;
6017 } else if (SvTYPE(referrer) == SVt_PVCV ||
6018 SvTYPE(referrer) == SVt_PVFM) {
6019 if (SvTYPE(sv) == SVt_PVHV) { /* stash backref */
6020 /* You lookin' at me? */
6021 assert(CvSTASH(referrer));
6022 assert(CvSTASH(referrer) == (const HV *)sv);
6023 SvANY(MUTABLE_CV(referrer))->xcv_stash = 0;
6026 assert(SvTYPE(sv) == SVt_PVGV);
6027 /* You lookin' at me? */
6028 assert(CvGV(referrer));
6029 assert(CvGV(referrer) == (const GV *)sv);
6030 anonymise_cv_maybe(MUTABLE_GV(sv),
6031 MUTABLE_CV(referrer));
6036 "panic: magic_killbackrefs (flags=%"UVxf")",
6037 (UV)SvFLAGS(referrer));
6048 SvREFCNT_dec_NN(av); /* remove extra count added by sv_add_backref() */
6054 =for apidoc sv_insert
6056 Inserts a string at the specified offset/length within the SV. Similar to
6057 the Perl substr() function. Handles get magic.
6059 =for apidoc sv_insert_flags
6061 Same as C<sv_insert>, but the extra C<flags> are passed to the
6062 C<SvPV_force_flags> that applies to C<bigstr>.
6068 Perl_sv_insert_flags(pTHX_ SV *const bigstr, const STRLEN offset, const STRLEN len, const char *const little, const STRLEN littlelen, const U32 flags)
6074 SSize_t i; /* better be sizeof(STRLEN) or bad things happen */
6077 PERL_ARGS_ASSERT_SV_INSERT_FLAGS;
6080 Perl_croak(aTHX_ "Can't modify nonexistent substring");
6081 SvPV_force_flags(bigstr, curlen, flags);
6082 (void)SvPOK_only_UTF8(bigstr);
6083 if (offset + len > curlen) {
6084 SvGROW(bigstr, offset+len+1);
6085 Zero(SvPVX(bigstr)+curlen, offset+len-curlen, char);
6086 SvCUR_set(bigstr, offset+len);
6090 i = littlelen - len;
6091 if (i > 0) { /* string might grow */
6092 big = SvGROW(bigstr, SvCUR(bigstr) + i + 1);
6093 mid = big + offset + len;
6094 midend = bigend = big + SvCUR(bigstr);
6097 while (midend > mid) /* shove everything down */
6098 *--bigend = *--midend;
6099 Move(little,big+offset,littlelen,char);
6100 SvCUR_set(bigstr, SvCUR(bigstr) + i);
6105 Move(little,SvPVX(bigstr)+offset,len,char);
6110 big = SvPVX(bigstr);
6113 bigend = big + SvCUR(bigstr);
6115 if (midend > bigend)
6116 Perl_croak(aTHX_ "panic: sv_insert, midend=%p, bigend=%p",
6119 if (mid - big > bigend - midend) { /* faster to shorten from end */
6121 Move(little, mid, littlelen,char);
6124 i = bigend - midend;
6126 Move(midend, mid, i,char);
6130 SvCUR_set(bigstr, mid - big);
6132 else if ((i = mid - big)) { /* faster from front */
6133 midend -= littlelen;
6135 Move(big, midend - i, i, char);
6136 sv_chop(bigstr,midend-i);
6138 Move(little, mid, littlelen,char);
6140 else if (littlelen) {
6141 midend -= littlelen;
6142 sv_chop(bigstr,midend);
6143 Move(little,midend,littlelen,char);
6146 sv_chop(bigstr,midend);
6152 =for apidoc sv_replace
6154 Make the first argument a copy of the second, then delete the original.
6155 The target SV physically takes over ownership of the body of the source SV
6156 and inherits its flags; however, the target keeps any magic it owns,
6157 and any magic in the source is discarded.
6158 Note that this is a rather specialist SV copying operation; most of the
6159 time you'll want to use C<sv_setsv> or one of its many macro front-ends.
6165 Perl_sv_replace(pTHX_ SV *const sv, SV *const nsv)
6167 const U32 refcnt = SvREFCNT(sv);
6169 PERL_ARGS_ASSERT_SV_REPLACE;
6171 SV_CHECK_THINKFIRST_COW_DROP(sv);
6172 if (SvREFCNT(nsv) != 1) {
6173 Perl_croak(aTHX_ "panic: reference miscount on nsv in sv_replace()"
6174 " (%" UVuf " != 1)", (UV) SvREFCNT(nsv));
6176 if (SvMAGICAL(sv)) {
6180 sv_upgrade(nsv, SVt_PVMG);
6181 SvMAGIC_set(nsv, SvMAGIC(sv));
6182 SvFLAGS(nsv) |= SvMAGICAL(sv);
6184 SvMAGIC_set(sv, NULL);
6188 assert(!SvREFCNT(sv));
6189 #ifdef DEBUG_LEAKING_SCALARS
6190 sv->sv_flags = nsv->sv_flags;
6191 sv->sv_any = nsv->sv_any;
6192 sv->sv_refcnt = nsv->sv_refcnt;
6193 sv->sv_u = nsv->sv_u;
6195 StructCopy(nsv,sv,SV);
6197 if(SvTYPE(sv) == SVt_IV) {
6199 = (XPVIV*)((char*)&(sv->sv_u.svu_iv) - STRUCT_OFFSET(XPVIV, xiv_iv));
6203 #ifdef PERL_OLD_COPY_ON_WRITE
6204 if (SvIsCOW_normal(nsv)) {
6205 /* We need to follow the pointers around the loop to make the
6206 previous SV point to sv, rather than nsv. */
6209 while ((next = SV_COW_NEXT_SV(current)) != nsv) {
6212 assert(SvPVX_const(current) == SvPVX_const(nsv));
6214 /* Make the SV before us point to the SV after us. */
6216 PerlIO_printf(Perl_debug_log, "previous is\n");
6218 PerlIO_printf(Perl_debug_log,
6219 "move it from 0x%"UVxf" to 0x%"UVxf"\n",
6220 (UV) SV_COW_NEXT_SV(current), (UV) sv);
6222 SV_COW_NEXT_SV_SET(current, sv);
6225 SvREFCNT(sv) = refcnt;
6226 SvFLAGS(nsv) |= SVTYPEMASK; /* Mark as freed */
6231 /* We're about to free a GV which has a CV that refers back to us.
6232 * If that CV will outlive us, make it anonymous (i.e. fix up its CvGV
6236 S_anonymise_cv_maybe(pTHX_ GV *gv, CV* cv)
6241 PERL_ARGS_ASSERT_ANONYMISE_CV_MAYBE;
6244 assert(SvREFCNT(gv) == 0);
6245 assert(isGV(gv) && isGV_with_GP(gv));
6247 assert(!CvANON(cv));
6248 assert(CvGV(cv) == gv);
6249 assert(!CvNAMED(cv));
6251 /* will the CV shortly be freed by gp_free() ? */
6252 if (GvCV(gv) == cv && GvGP(gv)->gp_refcnt < 2 && SvREFCNT(cv) < 2) {
6253 SvANY(cv)->xcv_gv_u.xcv_gv = NULL;
6257 /* if not, anonymise: */
6258 gvname = (GvSTASH(gv) && HvNAME(GvSTASH(gv)) && HvENAME(GvSTASH(gv)))
6259 ? newSVhek(HvENAME_HEK(GvSTASH(gv)))
6260 : newSVpvn_flags( "__ANON__", 8, 0 );
6261 sv_catpvs(gvname, "::__ANON__");
6262 anongv = gv_fetchsv(gvname, GV_ADDMULTI, SVt_PVCV);
6263 SvREFCNT_dec_NN(gvname);
6267 SvANY(cv)->xcv_gv_u.xcv_gv = MUTABLE_GV(SvREFCNT_inc(anongv));
6272 =for apidoc sv_clear
6274 Clear an SV: call any destructors, free up any memory used by the body,
6275 and free the body itself. The SV's head is I<not> freed, although
6276 its type is set to all 1's so that it won't inadvertently be assumed
6277 to be live during global destruction etc.
6278 This function should only be called when REFCNT is zero. Most of the time
6279 you'll want to call C<sv_free()> (or its macro wrapper C<SvREFCNT_dec>)
6286 Perl_sv_clear(pTHX_ SV *const orig_sv)
6291 const struct body_details *sv_type_details;
6297 PERL_ARGS_ASSERT_SV_CLEAR;
6299 /* within this loop, sv is the SV currently being freed, and
6300 * iter_sv is the most recent AV or whatever that's being iterated
6301 * over to provide more SVs */
6307 assert(SvREFCNT(sv) == 0);
6308 assert(SvTYPE(sv) != (svtype)SVTYPEMASK);
6310 if (type <= SVt_IV) {
6311 /* See the comment in sv.h about the collusion between this
6312 * early return and the overloading of the NULL slots in the
6316 SvFLAGS(sv) &= SVf_BREAK;
6317 SvFLAGS(sv) |= SVTYPEMASK;
6321 assert(!SvOBJECT(sv) || type >= SVt_PVMG); /* objs are always >= MG */
6323 if (type >= SVt_PVMG) {
6325 if (!curse(sv, 1)) goto get_next_sv;
6326 type = SvTYPE(sv); /* destructor may have changed it */
6328 /* Free back-references before magic, in case the magic calls
6329 * Perl code that has weak references to sv. */
6330 if (type == SVt_PVHV) {
6331 Perl_hv_kill_backrefs(aTHX_ MUTABLE_HV(sv));
6335 else if (type == SVt_PVMG && SvPAD_OUR(sv)) {
6336 SvREFCNT_dec(SvOURSTASH(sv));
6338 else if (type == SVt_PVAV && AvPAD_NAMELIST(sv)) {
6339 assert(!SvMAGICAL(sv));
6340 } else if (SvMAGIC(sv)) {
6341 /* Free back-references before other types of magic. */
6342 sv_unmagic(sv, PERL_MAGIC_backref);
6346 if (type == SVt_PVMG && SvPAD_TYPED(sv))
6347 SvREFCNT_dec(SvSTASH(sv));
6350 /* case SVt_INVLIST: */
6353 IoIFP(sv) != PerlIO_stdin() &&
6354 IoIFP(sv) != PerlIO_stdout() &&
6355 IoIFP(sv) != PerlIO_stderr() &&
6356 !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6358 io_close(MUTABLE_IO(sv), FALSE);
6360 if (IoDIRP(sv) && !(IoFLAGS(sv) & IOf_FAKE_DIRP))
6361 PerlDir_close(IoDIRP(sv));
6362 IoDIRP(sv) = (DIR*)NULL;
6363 Safefree(IoTOP_NAME(sv));
6364 Safefree(IoFMT_NAME(sv));
6365 Safefree(IoBOTTOM_NAME(sv));
6366 if ((const GV *)sv == PL_statgv)
6370 /* FIXME for plugins */
6372 pregfree2((REGEXP*) sv);
6376 cv_undef(MUTABLE_CV(sv));
6377 /* If we're in a stash, we don't own a reference to it.
6378 * However it does have a back reference to us, which needs to
6380 if ((stash = CvSTASH(sv)))
6381 sv_del_backref(MUTABLE_SV(stash), sv);
6384 if (PL_last_swash_hv == (const HV *)sv) {
6385 PL_last_swash_hv = NULL;
6387 if (HvTOTALKEYS((HV*)sv) > 0) {
6389 /* this statement should match the one at the beginning of
6390 * hv_undef_flags() */
6391 if ( PL_phase != PERL_PHASE_DESTRUCT
6392 && (name = HvNAME((HV*)sv)))
6394 if (PL_stashcache) {
6395 DEBUG_o(Perl_deb(aTHX_ "sv_clear clearing PL_stashcache for '%"SVf"'\n",
6397 (void)hv_deletehek(PL_stashcache,
6398 HvNAME_HEK((HV*)sv), G_DISCARD);
6400 hv_name_set((HV*)sv, NULL, 0, 0);
6403 /* save old iter_sv in unused SvSTASH field */
6404 assert(!SvOBJECT(sv));
6405 SvSTASH(sv) = (HV*)iter_sv;
6408 /* save old hash_index in unused SvMAGIC field */
6409 assert(!SvMAGICAL(sv));
6410 assert(!SvMAGIC(sv));
6411 ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index = hash_index;
6414 next_sv = Perl_hfree_next_entry(aTHX_ (HV*)sv, &hash_index);
6415 goto get_next_sv; /* process this new sv */
6417 /* free empty hash */
6418 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6419 assert(!HvARRAY((HV*)sv));
6423 AV* av = MUTABLE_AV(sv);
6424 if (PL_comppad == av) {
6428 if (AvREAL(av) && AvFILLp(av) > -1) {
6429 next_sv = AvARRAY(av)[AvFILLp(av)--];
6430 /* save old iter_sv in top-most slot of AV,
6431 * and pray that it doesn't get wiped in the meantime */
6432 AvARRAY(av)[AvMAX(av)] = iter_sv;
6434 goto get_next_sv; /* process this new sv */
6436 Safefree(AvALLOC(av));
6441 if (LvTYPE(sv) == 'T') { /* for tie: return HE to pool */
6442 SvREFCNT_dec(HeKEY_sv((HE*)LvTARG(sv)));
6443 HeNEXT((HE*)LvTARG(sv)) = PL_hv_fetch_ent_mh;
6444 PL_hv_fetch_ent_mh = (HE*)LvTARG(sv);
6446 else if (LvTYPE(sv) != 't') /* unless tie: unrefcnted fake SV** */
6447 SvREFCNT_dec(LvTARG(sv));
6448 if (isREGEXP(sv)) goto freeregexp;
6450 if (isGV_with_GP(sv)) {
6451 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
6452 && HvENAME_get(stash))
6453 mro_method_changed_in(stash);
6454 gp_free(MUTABLE_GV(sv));
6456 unshare_hek(GvNAME_HEK(sv));
6457 /* If we're in a stash, we don't own a reference to it.
6458 * However it does have a back reference to us, which
6459 * needs to be cleared. */
6460 if (!SvVALID(sv) && (stash = GvSTASH(sv)))
6461 sv_del_backref(MUTABLE_SV(stash), sv);
6463 /* FIXME. There are probably more unreferenced pointers to SVs
6464 * in the interpreter struct that we should check and tidy in
6465 * a similar fashion to this: */
6466 /* See also S_sv_unglob, which does the same thing. */
6467 if ((const GV *)sv == PL_last_in_gv)
6468 PL_last_in_gv = NULL;
6469 else if ((const GV *)sv == PL_statgv)
6471 else if ((const GV *)sv == PL_stderrgv)
6479 /* Don't bother with SvOOK_off(sv); as we're only going to
6483 SvOOK_offset(sv, offset);
6484 SvPV_set(sv, SvPVX_mutable(sv) - offset);
6485 /* Don't even bother with turning off the OOK flag. */
6490 SV * const target = SvRV(sv);
6492 sv_del_backref(target, sv);
6498 else if (SvPVX_const(sv)
6499 && !(SvTYPE(sv) == SVt_PVIO
6500 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6504 PerlIO_printf(Perl_debug_log, "Copy on write: clear\n");
6508 # ifdef PERL_OLD_COPY_ON_WRITE
6509 sv_release_COW(sv, SvPVX_const(sv), SV_COW_NEXT_SV(sv));
6511 if (CowREFCNT(sv)) {
6519 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6523 # ifdef PERL_OLD_COPY_ON_WRITE
6527 Safefree(SvPVX_mutable(sv));
6531 else if (SvPVX_const(sv) && SvLEN(sv)
6532 && !(SvTYPE(sv) == SVt_PVIO
6533 && !(IoFLAGS(sv) & IOf_FAKE_DIRP)))
6534 Safefree(SvPVX_mutable(sv));
6535 else if (SvPVX_const(sv) && SvIsCOW(sv)) {
6536 unshare_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sv)));
6546 SvFLAGS(sv) &= SVf_BREAK;
6547 SvFLAGS(sv) |= SVTYPEMASK;
6549 sv_type_details = bodies_by_type + type;
6550 if (sv_type_details->arena) {
6551 del_body(((char *)SvANY(sv) + sv_type_details->offset),
6552 &PL_body_roots[type]);
6554 else if (sv_type_details->body_size) {
6555 safefree(SvANY(sv));
6559 /* caller is responsible for freeing the head of the original sv */
6560 if (sv != orig_sv && !SvREFCNT(sv))
6563 /* grab and free next sv, if any */
6571 else if (!iter_sv) {
6573 } else if (SvTYPE(iter_sv) == SVt_PVAV) {
6574 AV *const av = (AV*)iter_sv;
6575 if (AvFILLp(av) > -1) {
6576 sv = AvARRAY(av)[AvFILLp(av)--];
6578 else { /* no more elements of current AV to free */
6581 /* restore previous value, squirrelled away */
6582 iter_sv = AvARRAY(av)[AvMAX(av)];
6583 Safefree(AvALLOC(av));
6586 } else if (SvTYPE(iter_sv) == SVt_PVHV) {
6587 sv = Perl_hfree_next_entry(aTHX_ (HV*)iter_sv, &hash_index);
6588 if (!sv && !HvTOTALKEYS((HV *)iter_sv)) {
6589 /* no more elements of current HV to free */
6592 /* Restore previous values of iter_sv and hash_index,
6593 * squirrelled away */
6594 assert(!SvOBJECT(sv));
6595 iter_sv = (SV*)SvSTASH(sv);
6596 assert(!SvMAGICAL(sv));
6597 hash_index = ((XPVMG*) SvANY(sv))->xmg_u.xmg_hash_index;
6599 /* perl -DA does not like rubbish in SvMAGIC. */
6603 /* free any remaining detritus from the hash struct */
6604 Perl_hv_undef_flags(aTHX_ MUTABLE_HV(sv), HV_NAME_SETALL);
6605 assert(!HvARRAY((HV*)sv));
6610 /* unrolled SvREFCNT_dec and sv_free2 follows: */
6614 if (!SvREFCNT(sv)) {
6618 if (--(SvREFCNT(sv)))
6622 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6623 "Attempt to free temp prematurely: SV 0x%"UVxf
6624 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6628 if (SvIMMORTAL(sv)) {
6629 /* make sure SvREFCNT(sv)==0 happens very seldom */
6630 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6639 /* This routine curses the sv itself, not the object referenced by sv. So
6640 sv does not have to be ROK. */
6643 S_curse(pTHX_ SV * const sv, const bool check_refcnt) {
6644 PERL_ARGS_ASSERT_CURSE;
6645 assert(SvOBJECT(sv));
6647 if (PL_defstash && /* Still have a symbol table? */
6653 stash = SvSTASH(sv);
6654 assert(SvTYPE(stash) == SVt_PVHV);
6655 if (HvNAME(stash)) {
6656 CV* destructor = NULL;
6657 assert (SvOOK(stash));
6658 if (!SvOBJECT(stash)) destructor = (CV *)SvSTASH(stash);
6659 if (!destructor || HvMROMETA(stash)->destroy_gen
6660 != PL_sub_generation)
6663 gv_fetchmeth_autoload(stash, "DESTROY", 7, 0);
6664 if (gv) destructor = GvCV(gv);
6665 if (!SvOBJECT(stash))
6668 destructor ? (HV *)destructor : ((HV *)0)+1;
6669 HvAUX(stash)->xhv_mro_meta->destroy_gen =
6673 assert(!destructor || destructor == ((CV *)0)+1
6674 || SvTYPE(destructor) == SVt_PVCV);
6675 if (destructor && destructor != ((CV *)0)+1
6676 /* A constant subroutine can have no side effects, so
6677 don't bother calling it. */
6678 && !CvCONST(destructor)
6679 /* Don't bother calling an empty destructor or one that
6680 returns immediately. */
6681 && (CvISXSUB(destructor)
6682 || (CvSTART(destructor)
6683 && (CvSTART(destructor)->op_next->op_type
6685 && (CvSTART(destructor)->op_next->op_type
6687 || CvSTART(destructor)->op_next->op_next->op_type
6693 SV* const tmpref = newRV(sv);
6694 SvREADONLY_on(tmpref); /* DESTROY() could be naughty */
6696 PUSHSTACKi(PERLSI_DESTROY);
6701 call_sv(MUTABLE_SV(destructor),
6702 G_DISCARD|G_EVAL|G_KEEPERR|G_VOID);
6706 if(SvREFCNT(tmpref) < 2) {
6707 /* tmpref is not kept alive! */
6709 SvRV_set(tmpref, NULL);
6712 SvREFCNT_dec_NN(tmpref);
6715 } while (SvOBJECT(sv) && SvSTASH(sv) != stash);
6718 if (check_refcnt && SvREFCNT(sv)) {
6719 if (PL_in_clean_objs)
6721 "DESTROY created new reference to dead object '%"HEKf"'",
6722 HEKfARG(HvNAME_HEK(stash)));
6723 /* DESTROY gave object new lease on life */
6729 HV * const stash = SvSTASH(sv);
6730 /* Curse before freeing the stash, as freeing the stash could cause
6731 a recursive call into S_curse. */
6732 SvOBJECT_off(sv); /* Curse the object. */
6733 SvSTASH_set(sv,0); /* SvREFCNT_dec may try to read this */
6734 SvREFCNT_dec(stash); /* possibly of changed persuasion */
6740 =for apidoc sv_newref
6742 Increment an SV's reference count. Use the C<SvREFCNT_inc()> wrapper
6749 Perl_sv_newref(pTHX_ SV *const sv)
6751 PERL_UNUSED_CONTEXT;
6760 Decrement an SV's reference count, and if it drops to zero, call
6761 C<sv_clear> to invoke destructors and free up any memory used by
6762 the body; finally, deallocate the SV's head itself.
6763 Normally called via a wrapper macro C<SvREFCNT_dec>.
6769 Perl_sv_free(pTHX_ SV *const sv)
6775 /* Private helper function for SvREFCNT_dec().
6776 * Called with rc set to original SvREFCNT(sv), where rc == 0 or 1 */
6779 Perl_sv_free2(pTHX_ SV *const sv, const U32 rc)
6783 PERL_ARGS_ASSERT_SV_FREE2;
6785 if (LIKELY( rc == 1 )) {
6791 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEBUGGING),
6792 "Attempt to free temp prematurely: SV 0x%"UVxf
6793 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6797 if (SvIMMORTAL(sv)) {
6798 /* make sure SvREFCNT(sv)==0 happens very seldom */
6799 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6803 if (! SvREFCNT(sv)) /* may have have been resurrected */
6808 /* handle exceptional cases */
6812 if (SvFLAGS(sv) & SVf_BREAK)
6813 /* this SV's refcnt has been artificially decremented to
6814 * trigger cleanup */
6816 if (PL_in_clean_all) /* All is fair */
6818 if (SvIMMORTAL(sv)) {
6819 /* make sure SvREFCNT(sv)==0 happens very seldom */
6820 SvREFCNT(sv) = SvREFCNT_IMMORTAL;
6823 if (ckWARN_d(WARN_INTERNAL)) {
6824 #ifdef DEBUG_LEAKING_SCALARS_FORK_DUMP
6825 Perl_dump_sv_child(aTHX_ sv);
6827 #ifdef DEBUG_LEAKING_SCALARS
6830 #ifdef DEBUG_LEAKING_SCALARS_ABORT
6831 if (PL_warnhook == PERL_WARNHOOK_FATAL
6832 || ckDEAD(packWARN(WARN_INTERNAL))) {
6833 /* Don't let Perl_warner cause us to escape our fate: */
6837 /* This may not return: */
6838 Perl_warner(aTHX_ packWARN(WARN_INTERNAL),
6839 "Attempt to free unreferenced scalar: SV 0x%"UVxf
6840 pTHX__FORMAT, PTR2UV(sv) pTHX__VALUE);
6843 #ifdef DEBUG_LEAKING_SCALARS_ABORT
6853 Returns the length of the string in the SV. Handles magic and type
6854 coercion and sets the UTF8 flag appropriately. See also C<SvCUR>, which
6855 gives raw access to the xpv_cur slot.
6861 Perl_sv_len(pTHX_ SV *const sv)
6868 (void)SvPV_const(sv, len);
6873 =for apidoc sv_len_utf8
6875 Returns the number of characters in the string in an SV, counting wide
6876 UTF-8 bytes as a single character. Handles magic and type coercion.
6882 * The length is cached in PERL_MAGIC_utf8, in the mg_len field. Also the
6883 * mg_ptr is used, by sv_pos_u2b() and sv_pos_b2u() - see the comments below.
6884 * (Note that the mg_len is not the length of the mg_ptr field.
6885 * This allows the cache to store the character length of the string without
6886 * needing to malloc() extra storage to attach to the mg_ptr.)
6891 Perl_sv_len_utf8(pTHX_ SV *const sv)
6897 return sv_len_utf8_nomg(sv);
6901 Perl_sv_len_utf8_nomg(pTHX_ SV * const sv)
6904 const U8 *s = (U8*)SvPV_nomg_const(sv, len);
6906 PERL_ARGS_ASSERT_SV_LEN_UTF8_NOMG;
6908 if (PL_utf8cache && SvUTF8(sv)) {
6910 MAGIC *mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_utf8) : NULL;
6912 if (mg && (mg->mg_len != -1 || mg->mg_ptr)) {
6913 if (mg->mg_len != -1)
6916 /* We can use the offset cache for a headstart.
6917 The longer value is stored in the first pair. */
6918 STRLEN *cache = (STRLEN *) mg->mg_ptr;
6920 ulen = cache[0] + Perl_utf8_length(aTHX_ s + cache[1],
6924 if (PL_utf8cache < 0) {
6925 const STRLEN real = Perl_utf8_length(aTHX_ s, s + len);
6926 assert_uft8_cache_coherent("sv_len_utf8", ulen, real, sv);
6930 ulen = Perl_utf8_length(aTHX_ s, s + len);
6931 utf8_mg_len_cache_update(sv, &mg, ulen);
6935 return SvUTF8(sv) ? Perl_utf8_length(aTHX_ s, s + len) : len;
6938 /* Walk forwards to find the byte corresponding to the passed in UTF-8
6941 S_sv_pos_u2b_forwards(const U8 *const start, const U8 *const send,
6942 STRLEN *const uoffset_p, bool *const at_end)
6944 const U8 *s = start;
6945 STRLEN uoffset = *uoffset_p;
6947 PERL_ARGS_ASSERT_SV_POS_U2B_FORWARDS;
6949 while (s < send && uoffset) {
6956 else if (s > send) {
6958 /* This is the existing behaviour. Possibly it should be a croak, as
6959 it's actually a bounds error */
6962 *uoffset_p -= uoffset;
6966 /* Given the length of the string in both bytes and UTF-8 characters, decide
6967 whether to walk forwards or backwards to find the byte corresponding to
6968 the passed in UTF-8 offset. */
6970 S_sv_pos_u2b_midway(const U8 *const start, const U8 *send,
6971 STRLEN uoffset, const STRLEN uend)
6973 STRLEN backw = uend - uoffset;
6975 PERL_ARGS_ASSERT_SV_POS_U2B_MIDWAY;
6977 if (uoffset < 2 * backw) {
6978 /* The assumption is that going forwards is twice the speed of going
6979 forward (that's where the 2 * backw comes from).
6980 (The real figure of course depends on the UTF-8 data.) */
6981 const U8 *s = start;
6983 while (s < send && uoffset--)
6993 while (UTF8_IS_CONTINUATION(*send))
6996 return send - start;
6999 /* For the string representation of the given scalar, find the byte
7000 corresponding to the passed in UTF-8 offset. uoffset0 and boffset0
7001 give another position in the string, *before* the sought offset, which
7002 (which is always true, as 0, 0 is a valid pair of positions), which should
7003 help reduce the amount of linear searching.
7004 If *mgp is non-NULL, it should point to the UTF-8 cache magic, which
7005 will be used to reduce the amount of linear searching. The cache will be
7006 created if necessary, and the found value offered to it for update. */
7008 S_sv_pos_u2b_cached(pTHX_ SV *const sv, MAGIC **const mgp, const U8 *const start,
7009 const U8 *const send, STRLEN uoffset,
7010 STRLEN uoffset0, STRLEN boffset0)
7012 STRLEN boffset = 0; /* Actually always set, but let's keep gcc happy. */
7014 bool at_end = FALSE;
7016 PERL_ARGS_ASSERT_SV_POS_U2B_CACHED;
7018 assert (uoffset >= uoffset0);
7023 if (!SvREADONLY(sv) && !SvGMAGICAL(sv) && SvPOK(sv)
7025 && (*mgp || (SvTYPE(sv) >= SVt_PVMG &&
7026 (*mgp = mg_find(sv, PERL_MAGIC_utf8))))) {
7027 if ((*mgp)->mg_ptr) {
7028 STRLEN *cache = (STRLEN *) (*mgp)->mg_ptr;
7029 if (cache[0] == uoffset) {
7030 /* An exact match. */
7033 if (cache[2] == uoffset) {
7034 /* An exact match. */
7038 if (cache[0] < uoffset) {
7039 /* The cache already knows part of the way. */
7040 if (cache[0] > uoffset0) {
7041 /* The cache knows more than the passed in pair */
7042 uoffset0 = cache[0];
7043 boffset0 = cache[1];
7045 if ((*mgp)->mg_len != -1) {
7046 /* And we know the end too. */
7048 + sv_pos_u2b_midway(start + boffset0, send,
7050 (*mgp)->mg_len - uoffset0);
7052 uoffset -= uoffset0;
7054 + sv_pos_u2b_forwards(start + boffset0,
7055 send, &uoffset, &at_end);
7056 uoffset += uoffset0;
7059 else if (cache[2] < uoffset) {
7060 /* We're between the two cache entries. */
7061 if (cache[2] > uoffset0) {
7062 /* and the cache knows more than the passed in pair */
7063 uoffset0 = cache[2];
7064 boffset0 = cache[3];
7068 + sv_pos_u2b_midway(start + boffset0,
7071 cache[0] - uoffset0);
7074 + sv_pos_u2b_midway(start + boffset0,
7077 cache[2] - uoffset0);
7081 else if ((*mgp)->mg_len != -1) {
7082 /* If we can take advantage of a passed in offset, do so. */
7083 /* In fact, offset0 is either 0, or less than offset, so don't
7084 need to worry about the other possibility. */
7086 + sv_pos_u2b_midway(start + boffset0, send,
7088 (*mgp)->mg_len - uoffset0);
7093 if (!found || PL_utf8cache < 0) {
7094 STRLEN real_boffset;
7095 uoffset -= uoffset0;
7096 real_boffset = boffset0 + sv_pos_u2b_forwards(start + boffset0,
7097 send, &uoffset, &at_end);
7098 uoffset += uoffset0;
7100 if (found && PL_utf8cache < 0)
7101 assert_uft8_cache_coherent("sv_pos_u2b_cache", boffset,
7103 boffset = real_boffset;
7106 if (PL_utf8cache && !SvGMAGICAL(sv) && SvPOK(sv)) {
7108 utf8_mg_len_cache_update(sv, mgp, uoffset);
7110 utf8_mg_pos_cache_update(sv, mgp, boffset, uoffset, send - start);
7117 =for apidoc sv_pos_u2b_flags
7119 Converts the offset from a count of UTF-8 chars from
7120 the start of the string, to a count of the equivalent number of bytes; if
7121 lenp is non-zero, it does the same to lenp, but this time starting from
7122 the offset, rather than from the start
7123 of the string. Handles type coercion.
7124 I<flags> is passed to C<SvPV_flags>, and usually should be
7125 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7131 * sv_pos_u2b_flags() uses, like sv_pos_b2u(), the mg_ptr of the potential
7132 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7133 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7138 Perl_sv_pos_u2b_flags(pTHX_ SV *const sv, STRLEN uoffset, STRLEN *const lenp,
7145 PERL_ARGS_ASSERT_SV_POS_U2B_FLAGS;
7147 start = (U8*)SvPV_flags(sv, len, flags);
7149 const U8 * const send = start + len;
7151 boffset = sv_pos_u2b_cached(sv, &mg, start, send, uoffset, 0, 0);
7154 && *lenp /* don't bother doing work for 0, as its bytes equivalent
7155 is 0, and *lenp is already set to that. */) {
7156 /* Convert the relative offset to absolute. */
7157 const STRLEN uoffset2 = uoffset + *lenp;
7158 const STRLEN boffset2
7159 = sv_pos_u2b_cached(sv, &mg, start, send, uoffset2,
7160 uoffset, boffset) - boffset;
7174 =for apidoc sv_pos_u2b
7176 Converts the value pointed to by offsetp from a count of UTF-8 chars from
7177 the start of the string, to a count of the equivalent number of bytes; if
7178 lenp is non-zero, it does the same to lenp, but this time starting from
7179 the offset, rather than from the start of the string. Handles magic and
7182 Use C<sv_pos_u2b_flags> in preference, which correctly handles strings longer
7189 * sv_pos_u2b() uses, like sv_pos_b2u(), the mg_ptr of the potential
7190 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7191 * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
7195 /* This function is subject to size and sign problems */
7198 Perl_sv_pos_u2b(pTHX_ SV *const sv, I32 *const offsetp, I32 *const lenp)
7200 PERL_ARGS_ASSERT_SV_POS_U2B;
7203 STRLEN ulen = (STRLEN)*lenp;
7204 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, &ulen,
7205 SV_GMAGIC|SV_CONST_RETURN);
7208 *offsetp = (I32)sv_pos_u2b_flags(sv, (STRLEN)*offsetp, NULL,
7209 SV_GMAGIC|SV_CONST_RETURN);
7214 S_utf8_mg_len_cache_update(pTHX_ SV *const sv, MAGIC **const mgp,
7217 PERL_ARGS_ASSERT_UTF8_MG_LEN_CACHE_UPDATE;
7218 if (SvREADONLY(sv) || SvGMAGICAL(sv) || !SvPOK(sv))
7221 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7222 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7223 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, &PL_vtbl_utf8, 0, 0);
7227 (*mgp)->mg_len = ulen;
7230 /* Create and update the UTF8 magic offset cache, with the proffered utf8/
7231 byte length pairing. The (byte) length of the total SV is passed in too,
7232 as blen, because for some (more esoteric) SVs, the call to SvPV_const()
7233 may not have updated SvCUR, so we can't rely on reading it directly.
7235 The proffered utf8/byte length pairing isn't used if the cache already has
7236 two pairs, and swapping either for the proffered pair would increase the
7237 RMS of the intervals between known byte offsets.
7239 The cache itself consists of 4 STRLEN values
7240 0: larger UTF-8 offset
7241 1: corresponding byte offset
7242 2: smaller UTF-8 offset
7243 3: corresponding byte offset
7245 Unused cache pairs have the value 0, 0.
7246 Keeping the cache "backwards" means that the invariant of
7247 cache[0] >= cache[2] is maintained even with empty slots, which means that
7248 the code that uses it doesn't need to worry if only 1 entry has actually
7249 been set to non-zero. It also makes the "position beyond the end of the
7250 cache" logic much simpler, as the first slot is always the one to start
7254 S_utf8_mg_pos_cache_update(pTHX_ SV *const sv, MAGIC **const mgp, const STRLEN byte,
7255 const STRLEN utf8, const STRLEN blen)
7259 PERL_ARGS_ASSERT_UTF8_MG_POS_CACHE_UPDATE;
7264 if (!*mgp && (SvTYPE(sv) < SVt_PVMG ||
7265 !(*mgp = mg_find(sv, PERL_MAGIC_utf8)))) {
7266 *mgp = sv_magicext(sv, 0, PERL_MAGIC_utf8, (MGVTBL*)&PL_vtbl_utf8, 0,
7268 (*mgp)->mg_len = -1;
7272 if (!(cache = (STRLEN *)(*mgp)->mg_ptr)) {
7273 Newxz(cache, PERL_MAGIC_UTF8_CACHESIZE * 2, STRLEN);
7274 (*mgp)->mg_ptr = (char *) cache;
7278 if (PL_utf8cache < 0 && SvPOKp(sv)) {
7279 /* SvPOKp() because it's possible that sv has string overloading, and
7280 therefore is a reference, hence SvPVX() is actually a pointer.
7281 This cures the (very real) symptoms of RT 69422, but I'm not actually
7282 sure whether we should even be caching the results of UTF-8
7283 operations on overloading, given that nothing stops overloading
7284 returning a different value every time it's called. */
7285 const U8 *start = (const U8 *) SvPVX_const(sv);
7286 const STRLEN realutf8 = utf8_length(start, start + byte);
7288 assert_uft8_cache_coherent("utf8_mg_pos_cache_update", utf8, realutf8,
7292 /* Cache is held with the later position first, to simplify the code
7293 that deals with unbounded ends. */
7295 ASSERT_UTF8_CACHE(cache);
7296 if (cache[1] == 0) {
7297 /* Cache is totally empty */
7300 } else if (cache[3] == 0) {
7301 if (byte > cache[1]) {
7302 /* New one is larger, so goes first. */
7303 cache[2] = cache[0];
7304 cache[3] = cache[1];
7312 #define THREEWAY_SQUARE(a,b,c,d) \
7313 ((float)((d) - (c))) * ((float)((d) - (c))) \
7314 + ((float)((c) - (b))) * ((float)((c) - (b))) \
7315 + ((float)((b) - (a))) * ((float)((b) - (a)))
7317 /* Cache has 2 slots in use, and we know three potential pairs.
7318 Keep the two that give the lowest RMS distance. Do the
7319 calculation in bytes simply because we always know the byte
7320 length. squareroot has the same ordering as the positive value,
7321 so don't bother with the actual square root. */
7322 if (byte > cache[1]) {
7323 /* New position is after the existing pair of pairs. */
7324 const float keep_earlier
7325 = THREEWAY_SQUARE(0, cache[3], byte, blen);
7326 const float keep_later
7327 = THREEWAY_SQUARE(0, cache[1], byte, blen);
7329 if (keep_later < keep_earlier) {
7330 cache[2] = cache[0];
7331 cache[3] = cache[1];
7340 else if (byte > cache[3]) {
7341 /* New position is between the existing pair of pairs. */
7342 const float keep_earlier
7343 = THREEWAY_SQUARE(0, cache[3], byte, blen);
7344 const float keep_later
7345 = THREEWAY_SQUARE(0, byte, cache[1], blen);
7347 if (keep_later < keep_earlier) {
7357 /* New position is before the existing pair of pairs. */
7358 const float keep_earlier
7359 = THREEWAY_SQUARE(0, byte, cache[3], blen);
7360 const float keep_later
7361 = THREEWAY_SQUARE(0, byte, cache[1], blen);
7363 if (keep_later < keep_earlier) {
7368 cache[0] = cache[2];
7369 cache[1] = cache[3];
7375 ASSERT_UTF8_CACHE(cache);
7378 /* We already know all of the way, now we may be able to walk back. The same
7379 assumption is made as in S_sv_pos_u2b_midway(), namely that walking
7380 backward is half the speed of walking forward. */
7382 S_sv_pos_b2u_midway(pTHX_ const U8 *const s, const U8 *const target,
7383 const U8 *end, STRLEN endu)
7385 const STRLEN forw = target - s;
7386 STRLEN backw = end - target;
7388 PERL_ARGS_ASSERT_SV_POS_B2U_MIDWAY;
7390 if (forw < 2 * backw) {
7391 return utf8_length(s, target);
7394 while (end > target) {
7396 while (UTF8_IS_CONTINUATION(*end)) {
7405 =for apidoc sv_pos_b2u_flags
7407 Converts the offset from a count of bytes from the start of the string, to
7408 a count of the equivalent number of UTF-8 chars. Handles type coercion.
7409 I<flags> is passed to C<SvPV_flags>, and usually should be
7410 C<SV_GMAGIC|SV_CONST_RETURN> to handle magic.
7416 * sv_pos_b2u_flags() uses, like sv_pos_u2b_flags(), the mg_ptr of the
7417 * potential PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8
7422 Perl_sv_pos_b2u_flags(pTHX_ SV *const sv, STRLEN const offset, U32 flags)
7425 STRLEN len = 0; /* Actually always set, but let's keep gcc happy. */
7431 PERL_ARGS_ASSERT_SV_POS_B2U_FLAGS;
7433 s = (const U8*)SvPV_flags(sv, blen, flags);
7436 Perl_croak(aTHX_ "panic: sv_pos_b2u: bad byte offset, blen=%"UVuf
7437 ", byte=%"UVuf, (UV)blen, (UV)offset);
7443 && SvTYPE(sv) >= SVt_PVMG
7444 && (mg = mg_find(sv, PERL_MAGIC_utf8)))
7447 STRLEN * const cache = (STRLEN *) mg->mg_ptr;
7448 if (cache[1] == offset) {
7449 /* An exact match. */
7452 if (cache[3] == offset) {
7453 /* An exact match. */
7457 if (cache[1] < offset) {
7458 /* We already know part of the way. */
7459 if (mg->mg_len != -1) {
7460 /* Actually, we know the end too. */
7462 + S_sv_pos_b2u_midway(aTHX_ s + cache[1], send,
7463 s + blen, mg->mg_len - cache[0]);
7465 len = cache[0] + utf8_length(s + cache[1], send);
7468 else if (cache[3] < offset) {
7469 /* We're between the two cached pairs, so we do the calculation
7470 offset by the byte/utf-8 positions for the earlier pair,
7471 then add the utf-8 characters from the string start to
7473 len = S_sv_pos_b2u_midway(aTHX_ s + cache[3], send,
7474 s + cache[1], cache[0] - cache[2])
7478 else { /* cache[3] > offset */
7479 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + cache[3],
7483 ASSERT_UTF8_CACHE(cache);
7485 } else if (mg->mg_len != -1) {
7486 len = S_sv_pos_b2u_midway(aTHX_ s, send, s + blen, mg->mg_len);
7490 if (!found || PL_utf8cache < 0) {
7491 const STRLEN real_len = utf8_length(s, send);
7493 if (found && PL_utf8cache < 0)
7494 assert_uft8_cache_coherent("sv_pos_b2u", len, real_len, sv);
7500 utf8_mg_len_cache_update(sv, &mg, len);
7502 utf8_mg_pos_cache_update(sv, &mg, offset, len, blen);
7509 =for apidoc sv_pos_b2u
7511 Converts the value pointed to by offsetp from a count of bytes from the
7512 start of the string, to a count of the equivalent number of UTF-8 chars.
7513 Handles magic and type coercion.
7515 Use C<sv_pos_b2u_flags> in preference, which correctly handles strings
7522 * sv_pos_b2u() uses, like sv_pos_u2b(), the mg_ptr of the potential
7523 * PERL_MAGIC_utf8 of the sv to store the mapping between UTF-8 and
7528 Perl_sv_pos_b2u(pTHX_ SV *const sv, I32 *const offsetp)
7530 PERL_ARGS_ASSERT_SV_POS_B2U;
7535 *offsetp = (I32)sv_pos_b2u_flags(sv, (STRLEN)*offsetp,
7536 SV_GMAGIC|SV_CONST_RETURN);
7540 S_assert_uft8_cache_coherent(pTHX_ const char *const func, STRLEN from_cache,
7541 STRLEN real, SV *const sv)
7543 PERL_ARGS_ASSERT_ASSERT_UFT8_CACHE_COHERENT;
7545 /* As this is debugging only code, save space by keeping this test here,
7546 rather than inlining it in all the callers. */
7547 if (from_cache == real)
7550 /* Need to turn the assertions off otherwise we may recurse infinitely
7551 while printing error messages. */
7552 SAVEI8(PL_utf8cache);
7554 Perl_croak(aTHX_ "panic: %s cache %"UVuf" real %"UVuf" for %"SVf,
7555 func, (UV) from_cache, (UV) real, SVfARG(sv));
7561 Returns a boolean indicating whether the strings in the two SVs are
7562 identical. Is UTF-8 and 'use bytes' aware, handles get magic, and will
7563 coerce its args to strings if necessary.
7565 =for apidoc sv_eq_flags
7567 Returns a boolean indicating whether the strings in the two SVs are
7568 identical. Is UTF-8 and 'use bytes' aware and coerces its args to strings
7569 if necessary. If the flags include SV_GMAGIC, it handles get-magic, too.
7575 Perl_sv_eq_flags(pTHX_ SV *sv1, SV *sv2, const U32 flags)
7582 SV* svrecode = NULL;
7589 /* if pv1 and pv2 are the same, second SvPV_const call may
7590 * invalidate pv1 (if we are handling magic), so we may need to
7592 if (sv1 == sv2 && flags & SV_GMAGIC
7593 && (SvTHINKFIRST(sv1) || SvGMAGICAL(sv1))) {
7594 pv1 = SvPV_const(sv1, cur1);
7595 sv1 = newSVpvn_flags(pv1, cur1, SVs_TEMP | SvUTF8(sv2));
7597 pv1 = SvPV_flags_const(sv1, cur1, flags);
7605 pv2 = SvPV_flags_const(sv2, cur2, flags);
7607 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7608 /* Differing utf8ness.
7609 * Do not UTF8size the comparands as a side-effect. */
7612 svrecode = newSVpvn(pv2, cur2);
7613 sv_recode_to_utf8(svrecode, PL_encoding);
7614 pv2 = SvPV_const(svrecode, cur2);
7617 svrecode = newSVpvn(pv1, cur1);
7618 sv_recode_to_utf8(svrecode, PL_encoding);
7619 pv1 = SvPV_const(svrecode, cur1);
7621 /* Now both are in UTF-8. */
7623 SvREFCNT_dec_NN(svrecode);
7629 /* sv1 is the UTF-8 one */
7630 return bytes_cmp_utf8((const U8*)pv2, cur2,
7631 (const U8*)pv1, cur1) == 0;
7634 /* sv2 is the UTF-8 one */
7635 return bytes_cmp_utf8((const U8*)pv1, cur1,
7636 (const U8*)pv2, cur2) == 0;
7642 eq = (pv1 == pv2) || memEQ(pv1, pv2, cur1);
7644 SvREFCNT_dec(svrecode);
7652 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7653 string in C<sv1> is less than, equal to, or greater than the string in
7654 C<sv2>. Is UTF-8 and 'use bytes' aware, handles get magic, and will
7655 coerce its args to strings if necessary. See also C<sv_cmp_locale>.
7657 =for apidoc sv_cmp_flags
7659 Compares the strings in two SVs. Returns -1, 0, or 1 indicating whether the
7660 string in C<sv1> is less than, equal to, or greater than the string in
7661 C<sv2>. Is UTF-8 and 'use bytes' aware and will coerce its args to strings
7662 if necessary. If the flags include SV_GMAGIC, it handles get magic. See
7663 also C<sv_cmp_locale_flags>.
7669 Perl_sv_cmp(pTHX_ SV *const sv1, SV *const sv2)
7671 return sv_cmp_flags(sv1, sv2, SV_GMAGIC);
7675 Perl_sv_cmp_flags(pTHX_ SV *const sv1, SV *const sv2,
7679 const char *pv1, *pv2;
7681 SV *svrecode = NULL;
7688 pv1 = SvPV_flags_const(sv1, cur1, flags);
7695 pv2 = SvPV_flags_const(sv2, cur2, flags);
7697 if (cur1 && cur2 && SvUTF8(sv1) != SvUTF8(sv2) && !IN_BYTES) {
7698 /* Differing utf8ness.
7699 * Do not UTF8size the comparands as a side-effect. */
7702 svrecode = newSVpvn(pv2, cur2);
7703 sv_recode_to_utf8(svrecode, PL_encoding);
7704 pv2 = SvPV_const(svrecode, cur2);
7707 const int retval = -bytes_cmp_utf8((const U8*)pv2, cur2,
7708 (const U8*)pv1, cur1);
7709 return retval ? retval < 0 ? -1 : +1 : 0;
7714 svrecode = newSVpvn(pv1, cur1);
7715 sv_recode_to_utf8(svrecode, PL_encoding);
7716 pv1 = SvPV_const(svrecode, cur1);
7719 const int retval = bytes_cmp_utf8((const U8*)pv1, cur1,
7720 (const U8*)pv2, cur2);
7721 return retval ? retval < 0 ? -1 : +1 : 0;
7727 cmp = cur2 ? -1 : 0;
7731 const I32 retval = memcmp((const void*)pv1, (const void*)pv2, cur1 < cur2 ? cur1 : cur2);
7734 cmp = retval < 0 ? -1 : 1;
7735 } else if (cur1 == cur2) {
7738 cmp = cur1 < cur2 ? -1 : 1;
7742 SvREFCNT_dec(svrecode);
7748 =for apidoc sv_cmp_locale
7750 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
7751 'use bytes' aware, handles get magic, and will coerce its args to strings
7752 if necessary. See also C<sv_cmp>.
7754 =for apidoc sv_cmp_locale_flags
7756 Compares the strings in two SVs in a locale-aware manner. Is UTF-8 and
7757 'use bytes' aware and will coerce its args to strings if necessary. If the
7758 flags contain SV_GMAGIC, it handles get magic. See also C<sv_cmp_flags>.
7764 Perl_sv_cmp_locale(pTHX_ SV *const sv1, SV *const sv2)
7766 return sv_cmp_locale_flags(sv1, sv2, SV_GMAGIC);
7770 Perl_sv_cmp_locale_flags(pTHX_ SV *const sv1, SV *const sv2,
7773 #ifdef USE_LOCALE_COLLATE
7779 if (PL_collation_standard)
7783 pv1 = sv1 ? sv_collxfrm_flags(sv1, &len1, flags) : (char *) NULL;
7785 pv2 = sv2 ? sv_collxfrm_flags(sv2, &len2, flags) : (char *) NULL;
7787 if (!pv1 || !len1) {
7798 retval = memcmp((void*)pv1, (void*)pv2, len1 < len2 ? len1 : len2);
7801 return retval < 0 ? -1 : 1;
7804 * When the result of collation is equality, that doesn't mean
7805 * that there are no differences -- some locales exclude some
7806 * characters from consideration. So to avoid false equalities,
7807 * we use the raw string as a tiebreaker.
7814 PERL_UNUSED_ARG(flags);
7815 #endif /* USE_LOCALE_COLLATE */
7817 return sv_cmp(sv1, sv2);
7821 #ifdef USE_LOCALE_COLLATE
7824 =for apidoc sv_collxfrm
7826 This calls C<sv_collxfrm_flags> with the SV_GMAGIC flag. See
7827 C<sv_collxfrm_flags>.
7829 =for apidoc sv_collxfrm_flags
7831 Add Collate Transform magic to an SV if it doesn't already have it. If the
7832 flags contain SV_GMAGIC, it handles get-magic.
7834 Any scalar variable may carry PERL_MAGIC_collxfrm magic that contains the
7835 scalar data of the variable, but transformed to such a format that a normal
7836 memory comparison can be used to compare the data according to the locale
7843 Perl_sv_collxfrm_flags(pTHX_ SV *const sv, STRLEN *const nxp, const I32 flags)
7847 PERL_ARGS_ASSERT_SV_COLLXFRM_FLAGS;
7849 mg = SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_collxfrm) : (MAGIC *) NULL;
7850 if (!mg || !mg->mg_ptr || *(U32*)mg->mg_ptr != PL_collation_ix) {
7856 Safefree(mg->mg_ptr);
7857 s = SvPV_flags_const(sv, len, flags);
7858 if ((xf = mem_collxfrm(s, len, &xlen))) {
7860 #ifdef PERL_OLD_COPY_ON_WRITE
7862 sv_force_normal_flags(sv, 0);
7864 mg = sv_magicext(sv, 0, PERL_MAGIC_collxfrm, &PL_vtbl_collxfrm,
7878 if (mg && mg->mg_ptr) {
7880 return mg->mg_ptr + sizeof(PL_collation_ix);
7888 #endif /* USE_LOCALE_COLLATE */
7891 S_sv_gets_append_to_utf8(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
7893 SV * const tsv = newSV(0);
7896 sv_gets(tsv, fp, 0);
7897 sv_utf8_upgrade_nomg(tsv);
7898 SvCUR_set(sv,append);
7901 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
7905 S_sv_gets_read_record(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
7908 const STRLEN recsize = SvUV(SvRV(PL_rs)); /* RsRECORD() guarantees > 0. */
7909 /* Grab the size of the record we're getting */
7910 char *buffer = SvGROW(sv, (STRLEN)(recsize + append + 1)) + append;
7917 /* With a true, record-oriented file on VMS, we need to use read directly
7918 * to ensure that we respect RMS record boundaries. The user is responsible
7919 * for providing a PL_rs value that corresponds to the FAB$W_MRS (maximum
7920 * record size) field. N.B. This is likely to produce invalid results on
7921 * varying-width character data when a record ends mid-character.
7923 fd = PerlIO_fileno(fp);
7925 && PerlLIO_fstat(fd, &st) == 0
7926 && (st.st_fab_rfm == FAB$C_VAR
7927 || st.st_fab_rfm == FAB$C_VFC
7928 || st.st_fab_rfm == FAB$C_FIX)) {
7930 bytesread = PerlLIO_read(fd, buffer, recsize);
7932 else /* in-memory file from PerlIO::Scalar
7933 * or not a record-oriented file
7937 bytesread = PerlIO_read(fp, buffer, recsize);
7939 /* At this point, the logic in sv_get() means that sv will
7940 be treated as utf-8 if the handle is utf8.
7942 if (PerlIO_isutf8(fp) && bytesread > 0) {
7943 char *bend = buffer + bytesread;
7944 char *bufp = buffer;
7945 size_t charcount = 0;
7946 bool charstart = TRUE;
7949 while (charcount < recsize) {
7950 /* count accumulated characters */
7951 while (bufp < bend) {
7953 skip = UTF8SKIP(bufp);
7955 if (bufp + skip > bend) {
7956 /* partial at the end */
7967 if (charcount < recsize) {
7969 STRLEN bufp_offset = bufp - buffer;
7970 SSize_t morebytesread;
7972 /* originally I read enough to fill any incomplete
7973 character and the first byte of the next
7974 character if needed, but if there's many
7975 multi-byte encoded characters we're going to be
7976 making a read call for every character beyond
7977 the original read size.
7979 So instead, read the rest of the character if
7980 any, and enough bytes to match at least the
7981 start bytes for each character we're going to
7985 readsize = recsize - charcount;
7987 readsize = skip - (bend - bufp) + recsize - charcount - 1;
7988 buffer = SvGROW(sv, append + bytesread + readsize + 1) + append;
7989 bend = buffer + bytesread;
7990 morebytesread = PerlIO_read(fp, bend, readsize);
7991 if (morebytesread <= 0) {
7992 /* we're done, if we still have incomplete
7993 characters the check code in sv_gets() will
7996 I'd originally considered doing
7997 PerlIO_ungetc() on all but the lead
7998 character of the incomplete character, but
7999 read() doesn't do that, so I don't.
8004 /* prepare to scan some more */
8005 bytesread += morebytesread;
8006 bend = buffer + bytesread;
8007 bufp = buffer + bufp_offset;
8015 SvCUR_set(sv, bytesread + append);
8016 buffer[bytesread] = '\0';
8017 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8023 Get a line from the filehandle and store it into the SV, optionally
8024 appending to the currently-stored string. If C<append> is not 0, the
8025 line is appended to the SV instead of overwriting it. C<append> should
8026 be set to the byte offset that the appended string should start at
8027 in the SV (typically, C<SvCUR(sv)> is a suitable choice).
8033 Perl_sv_gets(pTHX_ SV *const sv, PerlIO *const fp, I32 append)
8043 PERL_ARGS_ASSERT_SV_GETS;
8045 if (SvTHINKFIRST(sv))
8046 sv_force_normal_flags(sv, append ? 0 : SV_COW_DROP_PV);
8047 /* XXX. If you make this PVIV, then copy on write can copy scalars read
8049 However, perlbench says it's slower, because the existing swipe code
8050 is faster than copy on write.
8051 Swings and roundabouts. */
8052 SvUPGRADE(sv, SVt_PV);
8055 /* line is going to be appended to the existing buffer in the sv */
8056 if (PerlIO_isutf8(fp)) {
8058 sv_utf8_upgrade_nomg(sv);
8059 sv_pos_u2b(sv,&append,0);
8061 } else if (SvUTF8(sv)) {
8062 return S_sv_gets_append_to_utf8(aTHX_ sv, fp, append);
8068 /* not appending - "clear" the string by setting SvCUR to 0,
8069 * the pv is still avaiable. */
8072 if (PerlIO_isutf8(fp))
8075 if (IN_PERL_COMPILETIME) {
8076 /* we always read code in line mode */
8080 else if (RsSNARF(PL_rs)) {
8081 /* If it is a regular disk file use size from stat() as estimate
8082 of amount we are going to read -- may result in mallocing
8083 more memory than we really need if the layers below reduce
8084 the size we read (e.g. CRLF or a gzip layer).
8087 if (!PerlLIO_fstat(PerlIO_fileno(fp), &st) && S_ISREG(st.st_mode)) {
8088 const Off_t offset = PerlIO_tell(fp);
8089 if (offset != (Off_t) -1 && st.st_size + append > offset) {
8090 #ifdef PERL_NEW_COPY_ON_WRITE
8091 /* Add an extra byte for the sake of copy-on-write's
8092 * buffer reference count. */
8093 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 2));
8095 (void) SvGROW(sv, (STRLEN)((st.st_size - offset) + append + 1));
8102 else if (RsRECORD(PL_rs)) {
8103 return S_sv_gets_read_record(aTHX_ sv, fp, append);
8105 else if (RsPARA(PL_rs)) {
8111 /* Get $/ i.e. PL_rs into same encoding as stream wants */
8112 if (PerlIO_isutf8(fp)) {
8113 rsptr = SvPVutf8(PL_rs, rslen);
8116 if (SvUTF8(PL_rs)) {
8117 if (!sv_utf8_downgrade(PL_rs, TRUE)) {
8118 Perl_croak(aTHX_ "Wide character in $/");
8121 /* extract the raw pointer to the record separator */
8122 rsptr = SvPV_const(PL_rs, rslen);
8126 /* rslast is the last character in the record separator
8127 * note we don't use rslast except when rslen is true, so the
8128 * null assign is a placeholder. */
8129 rslast = rslen ? rsptr[rslen - 1] : '\0';
8131 if (rspara) { /* have to do this both before and after */
8132 do { /* to make sure file boundaries work right */
8135 i = PerlIO_getc(fp);
8139 PerlIO_ungetc(fp,i);
8145 /* See if we know enough about I/O mechanism to cheat it ! */
8147 /* This used to be #ifdef test - it is made run-time test for ease
8148 of abstracting out stdio interface. One call should be cheap
8149 enough here - and may even be a macro allowing compile
8153 if (PerlIO_fast_gets(fp)) {
8155 * We can do buffer based IO operations on this filehandle.
8157 * This means we can bypass a lot of subcalls and process
8158 * the buffer directly, it also means we know the upper bound
8159 * on the amount of data we might read of the current buffer
8160 * into our sv. Knowing this allows us to preallocate the pv
8161 * to be able to hold that maximum, which allows us to simplify
8162 * a lot of logic. */
8165 * We're going to steal some values from the stdio struct
8166 * and put EVERYTHING in the innermost loop into registers.
8168 STDCHAR *ptr; /* pointer into fp's read-ahead buffer */
8169 STRLEN bpx; /* length of the data in the target sv
8170 used to fix pointers after a SvGROW */
8171 I32 shortbuffered; /* If the pv buffer is shorter than the amount
8172 of data left in the read-ahead buffer.
8173 If 0 then the pv buffer can hold the full
8174 amount left, otherwise this is the amount it
8177 #if defined(__VMS) && defined(PERLIO_IS_STDIO)
8178 /* An ungetc()d char is handled separately from the regular
8179 * buffer, so we getc() it back out and stuff it in the buffer.
8181 i = PerlIO_getc(fp);
8182 if (i == EOF) return 0;
8183 *(--((*fp)->_ptr)) = (unsigned char) i;
8187 /* Here is some breathtakingly efficient cheating */
8189 /* When you read the following logic resist the urge to think
8190 * of record separators that are 1 byte long. They are an
8191 * uninteresting special (simple) case.
8193 * Instead think of record separators which are at least 2 bytes
8194 * long, and keep in mind that we need to deal with such
8195 * separators when they cross a read-ahead buffer boundary.
8197 * Also consider that we need to gracefully deal with separators
8198 * that may be longer than a single read ahead buffer.
8200 * Lastly do not forget we want to copy the delimiter as well. We
8201 * are copying all data in the file _up_to_and_including_ the separator
8204 * Now that you have all that in mind here is what is happening below:
8206 * 1. When we first enter the loop we do some memory book keeping to see
8207 * how much free space there is in the target SV. (This sub assumes that
8208 * it is operating on the same SV most of the time via $_ and that it is
8209 * going to be able to reuse the same pv buffer each call.) If there is
8210 * "enough" room then we set "shortbuffered" to how much space there is
8211 * and start reading forward.
8213 * 2. When we scan forward we copy from the read-ahead buffer to the target
8214 * SV's pv buffer. While we go we watch for the end of the read-ahead buffer,
8215 * and the end of the of pv, as well as for the "rslast", which is the last
8216 * char of the separator.
8218 * 3. When scanning forward if we see rslast then we jump backwards in *pv*
8219 * (which has a "complete" record up to the point we saw rslast) and check
8220 * it to see if it matches the separator. If it does we are done. If it doesn't
8221 * we continue on with the scan/copy.
8223 * 4. If we run out of read-ahead buffer (cnt goes to 0) then we have to get
8224 * the IO system to read the next buffer. We do this by doing a getc(), which
8225 * returns a single char read (or EOF), and prefills the buffer, and also
8226 * allows us to find out how full the buffer is. We use this information to
8227 * SvGROW() the sv to the size remaining in the buffer, after which we copy
8228 * the returned single char into the target sv, and then go back into scan
8231 * 5. If we run out of write-buffer then we SvGROW() it by the size of the
8232 * remaining space in the read-buffer.
8234 * Note that this code despite its twisty-turny nature is pretty darn slick.
8235 * It manages single byte separators, multi-byte cross boundary separators,
8236 * and cross-read-buffer separators cleanly and efficiently at the cost
8237 * of potentially greatly overallocating the target SV.
8243 /* get the number of bytes remaining in the read-ahead buffer
8244 * on first call on a given fp this will return 0.*/
8245 cnt = PerlIO_get_cnt(fp);
8247 /* make sure we have the room */
8248 if ((I32)(SvLEN(sv) - append) <= cnt + 1) {
8249 /* Not room for all of it
8250 if we are looking for a separator and room for some
8252 if (rslen && cnt > 80 && (I32)SvLEN(sv) > append) {
8253 /* just process what we have room for */
8254 shortbuffered = cnt - SvLEN(sv) + append + 1;
8255 cnt -= shortbuffered;
8258 /* ensure that the target sv has enough room to hold
8259 * the rest of the read-ahead buffer */
8261 /* remember that cnt can be negative */
8262 SvGROW(sv, (STRLEN)(append + (cnt <= 0 ? 2 : (cnt + 1))));
8266 /* we have enough room to hold the full buffer, lets scream */
8270 /* extract the pointer to sv's string buffer, offset by append as necessary */
8271 bp = (STDCHAR*)SvPVX_const(sv) + append; /* move these two too to registers */
8272 /* extract the point to the read-ahead buffer */
8273 ptr = (STDCHAR*)PerlIO_get_ptr(fp);
8275 /* some trace debug output */
8276 DEBUG_P(PerlIO_printf(Perl_debug_log,
8277 "Screamer: entering, ptr=%"UVuf", cnt=%ld\n",PTR2UV(ptr),(long)cnt));
8278 DEBUG_P(PerlIO_printf(Perl_debug_log,
8279 "Screamer: entering: PerlIO * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"
8281 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8282 PTR2UV(PerlIO_has_base(fp) ? PerlIO_get_base(fp) : 0)));
8286 /* if there is stuff left in the read-ahead buffer */
8288 /* if there is a separator */
8290 /* loop until we hit the end of the read-ahead buffer */
8291 while (cnt > 0) { /* this | eat */
8292 /* scan forward copying and searching for rslast as we go */
8294 if ((*bp++ = *ptr++) == rslast) /* really | dust */
8295 goto thats_all_folks; /* screams | sed :-) */
8299 /* no separator, slurp the full buffer */
8300 Copy(ptr, bp, cnt, char); /* this | eat */
8301 bp += cnt; /* screams | dust */
8302 ptr += cnt; /* louder | sed :-) */
8304 assert (!shortbuffered);
8305 goto cannot_be_shortbuffered;
8309 if (shortbuffered) { /* oh well, must extend */
8310 /* we didnt have enough room to fit the line into the target buffer
8311 * so we must extend the target buffer and keep going */
8312 cnt = shortbuffered;
8314 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8316 /* extned the target sv's buffer so it can hold the full read-ahead buffer */
8317 SvGROW(sv, SvLEN(sv) + append + cnt + 2);
8318 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8322 cannot_be_shortbuffered:
8323 /* we need to refill the read-ahead buffer if possible */
8325 DEBUG_P(PerlIO_printf(Perl_debug_log,
8326 "Screamer: going to getc, ptr=%"UVuf", cnt=%"IVdf"\n",
8327 PTR2UV(ptr),(IV)cnt));
8328 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* deregisterize cnt and ptr */
8330 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8331 "Screamer: pre: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf"\n",
8332 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8333 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8336 call PerlIO_getc() to let it prefill the lookahead buffer
8338 This used to call 'filbuf' in stdio form, but as that behaves like
8339 getc when cnt <= 0 we use PerlIO_getc here to avoid introducing
8340 another abstraction.
8342 Note we have to deal with the char in 'i' if we are not at EOF
8344 i = PerlIO_getc(fp); /* get more characters */
8346 DEBUG_Pv(PerlIO_printf(Perl_debug_log,
8347 "Screamer: post: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf"\n",
8348 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8349 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8351 /* find out how much is left in the read-ahead buffer, and rextract its pointer */
8352 cnt = PerlIO_get_cnt(fp);
8353 ptr = (STDCHAR*)PerlIO_get_ptr(fp); /* reregisterize cnt and ptr */
8354 DEBUG_P(PerlIO_printf(Perl_debug_log,
8355 "Screamer: after getc, ptr=%"UVuf", cnt=%"IVdf"\n",
8356 PTR2UV(ptr),(IV)cnt));
8358 if (i == EOF) /* all done for ever? */
8359 goto thats_really_all_folks;
8361 /* make sure we have enough space in the target sv */
8362 bpx = bp - (STDCHAR*)SvPVX_const(sv); /* box up before relocation */
8364 SvGROW(sv, bpx + cnt + 2);
8365 bp = (STDCHAR*)SvPVX_const(sv) + bpx; /* unbox after relocation */
8367 /* copy of the char we got from getc() */
8368 *bp++ = (STDCHAR)i; /* store character from PerlIO_getc */
8370 /* make sure we deal with the i being the last character of a separator */
8371 if (rslen && (STDCHAR)i == rslast) /* all done for now? */
8372 goto thats_all_folks;
8376 /* check if we have actually found the separator - only really applies
8378 if ((rslen > 1 && (STRLEN)(bp - (STDCHAR*)SvPVX_const(sv)) < rslen) ||
8379 memNE((char*)bp - rslen, rsptr, rslen))
8380 goto screamer; /* go back to the fray */
8381 thats_really_all_folks:
8383 cnt += shortbuffered;
8384 DEBUG_P(PerlIO_printf(Perl_debug_log,
8385 "Screamer: quitting, ptr=%"UVuf", cnt=%"IVdf"\n",PTR2UV(ptr),(IV)cnt));
8386 PerlIO_set_ptrcnt(fp, (STDCHAR*)ptr, cnt); /* put these back or we're in trouble */
8387 DEBUG_P(PerlIO_printf(Perl_debug_log,
8388 "Screamer: end: FILE * thinks ptr=%"UVuf", cnt=%"IVdf", base=%"UVuf
8390 PTR2UV(PerlIO_get_ptr(fp)), (IV)PerlIO_get_cnt(fp),
8391 PTR2UV(PerlIO_has_base (fp) ? PerlIO_get_base(fp) : 0)));
8393 SvCUR_set(sv, bp - (STDCHAR*)SvPVX_const(sv)); /* set length */
8394 DEBUG_P(PerlIO_printf(Perl_debug_log,
8395 "Screamer: done, len=%ld, string=|%.*s|\n",
8396 (long)SvCUR(sv),(int)SvCUR(sv),SvPVX_const(sv)));
8400 /*The big, slow, and stupid way. */
8401 #ifdef USE_HEAP_INSTEAD_OF_STACK /* Even slower way. */
8402 STDCHAR *buf = NULL;
8403 Newx(buf, 8192, STDCHAR);
8411 const STDCHAR * const bpe = buf + sizeof(buf);
8413 while ((i = PerlIO_getc(fp)) != EOF && (*bp++ = (STDCHAR)i) != rslast && bp < bpe)
8414 ; /* keep reading */
8418 cnt = PerlIO_read(fp,(char*)buf, sizeof(buf));
8419 /* Accommodate broken VAXC compiler, which applies U8 cast to
8420 * both args of ?: operator, causing EOF to change into 255
8423 i = (U8)buf[cnt - 1];
8429 cnt = 0; /* we do need to re-set the sv even when cnt <= 0 */
8431 sv_catpvn_nomg(sv, (char *) buf, cnt);
8433 sv_setpvn(sv, (char *) buf, cnt); /* "nomg" is implied */
8435 if (i != EOF && /* joy */
8437 SvCUR(sv) < rslen ||
8438 memNE(SvPVX_const(sv) + SvCUR(sv) - rslen, rsptr, rslen)))
8442 * If we're reading from a TTY and we get a short read,
8443 * indicating that the user hit his EOF character, we need
8444 * to notice it now, because if we try to read from the TTY
8445 * again, the EOF condition will disappear.
8447 * The comparison of cnt to sizeof(buf) is an optimization
8448 * that prevents unnecessary calls to feof().
8452 if (!(cnt < (I32)sizeof(buf) && PerlIO_eof(fp)))
8456 #ifdef USE_HEAP_INSTEAD_OF_STACK
8461 if (rspara) { /* have to do this both before and after */
8462 while (i != EOF) { /* to make sure file boundaries work right */
8463 i = PerlIO_getc(fp);
8465 PerlIO_ungetc(fp,i);
8471 return (SvCUR(sv) - append) ? SvPVX(sv) : NULL;
8477 Auto-increment of the value in the SV, doing string to numeric conversion
8478 if necessary. Handles 'get' magic and operator overloading.
8484 Perl_sv_inc(pTHX_ SV *const sv)
8493 =for apidoc sv_inc_nomg
8495 Auto-increment of the value in the SV, doing string to numeric conversion
8496 if necessary. Handles operator overloading. Skips handling 'get' magic.
8502 Perl_sv_inc_nomg(pTHX_ SV *const sv)
8509 if (SvTHINKFIRST(sv)) {
8510 if (SvREADONLY(sv)) {
8511 Perl_croak_no_modify();
8515 if (SvAMAGIC(sv) && AMG_CALLunary(sv, inc_amg))
8517 i = PTR2IV(SvRV(sv));
8521 else sv_force_normal_flags(sv, 0);
8523 flags = SvFLAGS(sv);
8524 if ((flags & (SVp_NOK|SVp_IOK)) == SVp_NOK) {
8525 /* It's (privately or publicly) a float, but not tested as an
8526 integer, so test it to see. */
8528 flags = SvFLAGS(sv);
8530 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8531 /* It's publicly an integer, or privately an integer-not-float */
8532 #ifdef PERL_PRESERVE_IVUV
8536 if (SvUVX(sv) == UV_MAX)
8537 sv_setnv(sv, UV_MAX_P1);
8539 (void)SvIOK_only_UV(sv);
8540 SvUV_set(sv, SvUVX(sv) + 1);
8542 if (SvIVX(sv) == IV_MAX)
8543 sv_setuv(sv, (UV)IV_MAX + 1);
8545 (void)SvIOK_only(sv);
8546 SvIV_set(sv, SvIVX(sv) + 1);
8551 if (flags & SVp_NOK) {
8552 const NV was = SvNVX(sv);
8553 if (NV_OVERFLOWS_INTEGERS_AT &&
8554 was >= NV_OVERFLOWS_INTEGERS_AT) {
8555 /* diag_listed_as: Lost precision when %s %f by 1 */
8556 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8557 "Lost precision when incrementing %" NVff " by 1",
8560 (void)SvNOK_only(sv);
8561 SvNV_set(sv, was + 1.0);
8565 if (!(flags & SVp_POK) || !*SvPVX_const(sv)) {
8566 if ((flags & SVTYPEMASK) < SVt_PVIV)
8567 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV ? SVt_PVIV : SVt_IV));
8568 (void)SvIOK_only(sv);
8573 while (isALPHA(*d)) d++;
8574 while (isDIGIT(*d)) d++;
8575 if (d < SvEND(sv)) {
8576 const int numtype = grok_number_flags(SvPVX_const(sv), SvCUR(sv), NULL, PERL_SCAN_TRAILING);
8577 #ifdef PERL_PRESERVE_IVUV
8578 /* Got to punt this as an integer if needs be, but we don't issue
8579 warnings. Probably ought to make the sv_iv_please() that does
8580 the conversion if possible, and silently. */
8581 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
8582 /* Need to try really hard to see if it's an integer.
8583 9.22337203685478e+18 is an integer.
8584 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
8585 so $a="9.22337203685478e+18"; $a+0; $a++
8586 needs to be the same as $a="9.22337203685478e+18"; $a++
8593 /* sv_2iv *should* have made this an NV */
8594 if (flags & SVp_NOK) {
8595 (void)SvNOK_only(sv);
8596 SvNV_set(sv, SvNVX(sv) + 1.0);
8599 /* I don't think we can get here. Maybe I should assert this
8600 And if we do get here I suspect that sv_setnv will croak. NWC
8602 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_inc punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"NVgf"\n",
8603 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
8605 #endif /* PERL_PRESERVE_IVUV */
8606 if (!numtype && ckWARN(WARN_NUMERIC))
8607 not_incrementable(sv);
8608 sv_setnv(sv,Atof(SvPVX_const(sv)) + 1.0);
8612 while (d >= SvPVX_const(sv)) {
8620 /* MKS: The original code here died if letters weren't consecutive.
8621 * at least it didn't have to worry about non-C locales. The
8622 * new code assumes that ('z'-'a')==('Z'-'A'), letters are
8623 * arranged in order (although not consecutively) and that only
8624 * [A-Za-z] are accepted by isALPHA in the C locale.
8626 if (isALPHA_FOLD_NE(*d, 'z')) {
8627 do { ++*d; } while (!isALPHA(*d));
8630 *(d--) -= 'z' - 'a';
8635 *(d--) -= 'z' - 'a' + 1;
8639 /* oh,oh, the number grew */
8640 SvGROW(sv, SvCUR(sv) + 2);
8641 SvCUR_set(sv, SvCUR(sv) + 1);
8642 for (d = SvPVX(sv) + SvCUR(sv); d > SvPVX_const(sv); d--)
8653 Auto-decrement of the value in the SV, doing string to numeric conversion
8654 if necessary. Handles 'get' magic and operator overloading.
8660 Perl_sv_dec(pTHX_ SV *const sv)
8669 =for apidoc sv_dec_nomg
8671 Auto-decrement of the value in the SV, doing string to numeric conversion
8672 if necessary. Handles operator overloading. Skips handling 'get' magic.
8678 Perl_sv_dec_nomg(pTHX_ SV *const sv)
8684 if (SvTHINKFIRST(sv)) {
8685 if (SvREADONLY(sv)) {
8686 Perl_croak_no_modify();
8690 if (SvAMAGIC(sv) && AMG_CALLunary(sv, dec_amg))
8692 i = PTR2IV(SvRV(sv));
8696 else sv_force_normal_flags(sv, 0);
8698 /* Unlike sv_inc we don't have to worry about string-never-numbers
8699 and keeping them magic. But we mustn't warn on punting */
8700 flags = SvFLAGS(sv);
8701 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
8702 /* It's publicly an integer, or privately an integer-not-float */
8703 #ifdef PERL_PRESERVE_IVUV
8707 if (SvUVX(sv) == 0) {
8708 (void)SvIOK_only(sv);
8712 (void)SvIOK_only_UV(sv);
8713 SvUV_set(sv, SvUVX(sv) - 1);
8716 if (SvIVX(sv) == IV_MIN) {
8717 sv_setnv(sv, (NV)IV_MIN);
8721 (void)SvIOK_only(sv);
8722 SvIV_set(sv, SvIVX(sv) - 1);
8727 if (flags & SVp_NOK) {
8730 const NV was = SvNVX(sv);
8731 if (NV_OVERFLOWS_INTEGERS_AT &&
8732 was <= -NV_OVERFLOWS_INTEGERS_AT) {
8733 /* diag_listed_as: Lost precision when %s %f by 1 */
8734 Perl_ck_warner(aTHX_ packWARN(WARN_IMPRECISION),
8735 "Lost precision when decrementing %" NVff " by 1",
8738 (void)SvNOK_only(sv);
8739 SvNV_set(sv, was - 1.0);
8743 if (!(flags & SVp_POK)) {
8744 if ((flags & SVTYPEMASK) < SVt_PVIV)
8745 sv_upgrade(sv, ((flags & SVTYPEMASK) > SVt_IV) ? SVt_PVIV : SVt_IV);
8747 (void)SvIOK_only(sv);
8750 #ifdef PERL_PRESERVE_IVUV
8752 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), NULL);
8753 if (numtype && !(numtype & IS_NUMBER_INFINITY)) {
8754 /* Need to try really hard to see if it's an integer.
8755 9.22337203685478e+18 is an integer.
8756 but "9.22337203685478e+18" + 0 is UV=9223372036854779904
8757 so $a="9.22337203685478e+18"; $a+0; $a--
8758 needs to be the same as $a="9.22337203685478e+18"; $a--
8765 /* sv_2iv *should* have made this an NV */
8766 if (flags & SVp_NOK) {
8767 (void)SvNOK_only(sv);
8768 SvNV_set(sv, SvNVX(sv) - 1.0);
8771 /* I don't think we can get here. Maybe I should assert this
8772 And if we do get here I suspect that sv_setnv will croak. NWC
8774 DEBUG_c(PerlIO_printf(Perl_debug_log,"sv_dec punt failed to convert '%s' to IOK or NOKp, UV=0x%"UVxf" NV=%"NVgf"\n",
8775 SvPVX_const(sv), SvIVX(sv), SvNVX(sv)));
8778 #endif /* PERL_PRESERVE_IVUV */
8779 sv_setnv(sv,Atof(SvPVX_const(sv)) - 1.0); /* punt */
8782 /* this define is used to eliminate a chunk of duplicated but shared logic
8783 * it has the suffix __SV_C to signal that it isnt API, and isnt meant to be
8784 * used anywhere but here - yves
8786 #define PUSH_EXTEND_MORTAL__SV_C(AnSv) \
8789 PL_tmps_stack[++PL_tmps_ix] = (AnSv); \
8793 =for apidoc sv_mortalcopy
8795 Creates a new SV which is a copy of the original SV (using C<sv_setsv>).
8796 The new SV is marked as mortal. It will be destroyed "soon", either by an
8797 explicit call to FREETMPS, or by an implicit call at places such as
8798 statement boundaries. See also C<sv_newmortal> and C<sv_2mortal>.
8803 /* Make a string that will exist for the duration of the expression
8804 * evaluation. Actually, it may have to last longer than that, but
8805 * hopefully we won't free it until it has been assigned to a
8806 * permanent location. */
8809 Perl_sv_mortalcopy_flags(pTHX_ SV *const oldstr, U32 flags)
8813 if (flags & SV_GMAGIC)
8814 SvGETMAGIC(oldstr); /* before new_SV, in case it dies */
8816 sv_setsv_flags(sv,oldstr,flags & ~SV_GMAGIC);
8817 PUSH_EXTEND_MORTAL__SV_C(sv);
8823 =for apidoc sv_newmortal
8825 Creates a new null SV which is mortal. The reference count of the SV is
8826 set to 1. It will be destroyed "soon", either by an explicit call to
8827 FREETMPS, or by an implicit call at places such as statement boundaries.
8828 See also C<sv_mortalcopy> and C<sv_2mortal>.
8834 Perl_sv_newmortal(pTHX)
8839 SvFLAGS(sv) = SVs_TEMP;
8840 PUSH_EXTEND_MORTAL__SV_C(sv);
8846 =for apidoc newSVpvn_flags
8848 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
8849 characters) into it. The reference count for the
8850 SV is set to 1. Note that if C<len> is zero, Perl will create a zero length
8851 string. You are responsible for ensuring that the source string is at least
8852 C<len> bytes long. If the C<s> argument is NULL the new SV will be undefined.
8853 Currently the only flag bits accepted are C<SVf_UTF8> and C<SVs_TEMP>.
8854 If C<SVs_TEMP> is set, then C<sv_2mortal()> is called on the result before
8855 returning. If C<SVf_UTF8> is set, C<s>
8856 is considered to be in UTF-8 and the
8857 C<SVf_UTF8> flag will be set on the new SV.
8858 C<newSVpvn_utf8()> is a convenience wrapper for this function, defined as
8860 #define newSVpvn_utf8(s, len, u) \
8861 newSVpvn_flags((s), (len), (u) ? SVf_UTF8 : 0)
8867 Perl_newSVpvn_flags(pTHX_ const char *const s, const STRLEN len, const U32 flags)
8871 /* All the flags we don't support must be zero.
8872 And we're new code so I'm going to assert this from the start. */
8873 assert(!(flags & ~(SVf_UTF8|SVs_TEMP)));
8875 sv_setpvn(sv,s,len);
8877 /* This code used to do a sv_2mortal(), however we now unroll the call to
8878 * sv_2mortal() and do what it does ourselves here. Since we have asserted
8879 * that flags can only have the SVf_UTF8 and/or SVs_TEMP flags set above we
8880 * can use it to enable the sv flags directly (bypassing SvTEMP_on), which
8881 * in turn means we dont need to mask out the SVf_UTF8 flag below, which
8882 * means that we eliminate quite a few steps than it looks - Yves
8883 * (explaining patch by gfx) */
8885 SvFLAGS(sv) |= flags;
8887 if(flags & SVs_TEMP){
8888 PUSH_EXTEND_MORTAL__SV_C(sv);
8895 =for apidoc sv_2mortal
8897 Marks an existing SV as mortal. The SV will be destroyed "soon", either
8898 by an explicit call to FREETMPS, or by an implicit call at places such as
8899 statement boundaries. SvTEMP() is turned on which means that the SV's
8900 string buffer can be "stolen" if this SV is copied. See also C<sv_newmortal>
8901 and C<sv_mortalcopy>.
8907 Perl_sv_2mortal(pTHX_ SV *const sv)
8914 PUSH_EXTEND_MORTAL__SV_C(sv);
8922 Creates a new SV and copies a string (which may contain C<NUL> (C<\0>)
8923 characters) into it. The reference count for the
8924 SV is set to 1. If C<len> is zero, Perl will compute the length using
8925 strlen(), (which means if you use this option, that C<s> can't have embedded
8926 C<NUL> characters and has to have a terminating C<NUL> byte).
8928 For efficiency, consider using C<newSVpvn> instead.
8934 Perl_newSVpv(pTHX_ const char *const s, const STRLEN len)
8939 sv_setpvn(sv, s, len || s == NULL ? len : strlen(s));
8944 =for apidoc newSVpvn
8946 Creates a new SV and copies a string into it, which may contain C<NUL> characters
8947 (C<\0>) and other binary data. The reference count for the SV is set to 1.
8948 Note that if C<len> is zero, Perl will create a zero length (Perl) string. You
8949 are responsible for ensuring that the source buffer is at least
8950 C<len> bytes long. If the C<buffer> argument is NULL the new SV will be
8957 Perl_newSVpvn(pTHX_ const char *const buffer, const STRLEN len)
8961 sv_setpvn(sv,buffer,len);
8966 =for apidoc newSVhek
8968 Creates a new SV from the hash key structure. It will generate scalars that
8969 point to the shared string table where possible. Returns a new (undefined)
8970 SV if the hek is NULL.
8976 Perl_newSVhek(pTHX_ const HEK *const hek)
8985 if (HEK_LEN(hek) == HEf_SVKEY) {
8986 return newSVsv(*(SV**)HEK_KEY(hek));
8988 const int flags = HEK_FLAGS(hek);
8989 if (flags & HVhek_WASUTF8) {
8991 Andreas would like keys he put in as utf8 to come back as utf8
8993 STRLEN utf8_len = HEK_LEN(hek);
8994 SV * const sv = newSV_type(SVt_PV);
8995 char *as_utf8 = (char *)bytes_to_utf8 ((U8*)HEK_KEY(hek), &utf8_len);
8996 /* bytes_to_utf8() allocates a new string, which we can repurpose: */
8997 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
9000 } else if (flags & HVhek_UNSHARED) {
9001 /* A hash that isn't using shared hash keys has to have
9002 the flag in every key so that we know not to try to call
9003 share_hek_hek on it. */
9005 SV * const sv = newSVpvn (HEK_KEY(hek), HEK_LEN(hek));
9010 /* This will be overwhelminly the most common case. */
9012 /* Inline most of newSVpvn_share(), because share_hek_hek() is far
9013 more efficient than sharepvn(). */
9017 sv_upgrade(sv, SVt_PV);
9018 SvPV_set(sv, (char *)HEK_KEY(share_hek_hek(hek)));
9019 SvCUR_set(sv, HEK_LEN(hek));
9031 =for apidoc newSVpvn_share
9033 Creates a new SV with its SvPVX_const pointing to a shared string in the string
9034 table. If the string does not already exist in the table, it is
9035 created first. Turns on the SvIsCOW flag (or READONLY
9036 and FAKE in 5.16 and earlier). If the C<hash> parameter
9037 is non-zero, that value is used; otherwise the hash is computed.
9038 The string's hash can later be retrieved from the SV
9039 with the C<SvSHARED_HASH()> macro. The idea here is
9040 that as the string table is used for shared hash keys these strings will have
9041 SvPVX_const == HeKEY and hash lookup will avoid string compare.
9047 Perl_newSVpvn_share(pTHX_ const char *src, I32 len, U32 hash)
9051 bool is_utf8 = FALSE;
9052 const char *const orig_src = src;
9055 STRLEN tmplen = -len;
9057 /* See the note in hv.c:hv_fetch() --jhi */
9058 src = (char*)bytes_from_utf8((const U8*)src, &tmplen, &is_utf8);
9062 PERL_HASH(hash, src, len);
9064 /* The logic for this is inlined in S_mro_get_linear_isa_dfs(), so if it
9065 changes here, update it there too. */
9066 sv_upgrade(sv, SVt_PV);
9067 SvPV_set(sv, sharepvn(src, is_utf8?-len:len, hash));
9074 if (src != orig_src)
9080 =for apidoc newSVpv_share
9082 Like C<newSVpvn_share>, but takes a C<NUL>-terminated string instead of a
9089 Perl_newSVpv_share(pTHX_ const char *src, U32 hash)
9091 return newSVpvn_share(src, strlen(src), hash);
9094 #if defined(PERL_IMPLICIT_CONTEXT)
9096 /* pTHX_ magic can't cope with varargs, so this is a no-context
9097 * version of the main function, (which may itself be aliased to us).
9098 * Don't access this version directly.
9102 Perl_newSVpvf_nocontext(const char *const pat, ...)
9108 PERL_ARGS_ASSERT_NEWSVPVF_NOCONTEXT;
9110 va_start(args, pat);
9111 sv = vnewSVpvf(pat, &args);
9118 =for apidoc newSVpvf
9120 Creates a new SV and initializes it with the string formatted like
9127 Perl_newSVpvf(pTHX_ const char *const pat, ...)
9132 PERL_ARGS_ASSERT_NEWSVPVF;
9134 va_start(args, pat);
9135 sv = vnewSVpvf(pat, &args);
9140 /* backend for newSVpvf() and newSVpvf_nocontext() */
9143 Perl_vnewSVpvf(pTHX_ const char *const pat, va_list *const args)
9147 PERL_ARGS_ASSERT_VNEWSVPVF;
9150 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
9157 Creates a new SV and copies a floating point value into it.
9158 The reference count for the SV is set to 1.
9164 Perl_newSVnv(pTHX_ const NV n)
9176 Creates a new SV and copies an integer into it. The reference count for the
9183 Perl_newSViv(pTHX_ const IV i)
9195 Creates a new SV and copies an unsigned integer into it.
9196 The reference count for the SV is set to 1.
9202 Perl_newSVuv(pTHX_ const UV u)
9212 =for apidoc newSV_type
9214 Creates a new SV, of the type specified. The reference count for the new SV
9221 Perl_newSV_type(pTHX_ const svtype type)
9226 sv_upgrade(sv, type);
9231 =for apidoc newRV_noinc
9233 Creates an RV wrapper for an SV. The reference count for the original
9234 SV is B<not> incremented.
9240 Perl_newRV_noinc(pTHX_ SV *const tmpRef)
9242 SV *sv = newSV_type(SVt_IV);
9244 PERL_ARGS_ASSERT_NEWRV_NOINC;
9247 SvRV_set(sv, tmpRef);
9252 /* newRV_inc is the official function name to use now.
9253 * newRV_inc is in fact #defined to newRV in sv.h
9257 Perl_newRV(pTHX_ SV *const sv)
9259 PERL_ARGS_ASSERT_NEWRV;
9261 return newRV_noinc(SvREFCNT_inc_simple_NN(sv));
9267 Creates a new SV which is an exact duplicate of the original SV.
9274 Perl_newSVsv(pTHX_ SV *const old)
9280 if (SvTYPE(old) == (svtype)SVTYPEMASK) {
9281 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL), "semi-panic: attempt to dup freed string");
9284 /* Do this here, otherwise we leak the new SV if this croaks. */
9287 /* SV_NOSTEAL prevents TEMP buffers being, well, stolen, and saves games
9288 with SvTEMP_off and SvTEMP_on round a call to sv_setsv. */
9289 sv_setsv_flags(sv, old, SV_NOSTEAL);
9294 =for apidoc sv_reset
9296 Underlying implementation for the C<reset> Perl function.
9297 Note that the perl-level function is vaguely deprecated.
9303 Perl_sv_reset(pTHX_ const char *s, HV *const stash)
9305 PERL_ARGS_ASSERT_SV_RESET;
9307 sv_resetpvn(*s ? s : NULL, strlen(s), stash);
9311 Perl_sv_resetpvn(pTHX_ const char *s, STRLEN len, HV * const stash)
9313 char todo[PERL_UCHAR_MAX+1];
9316 if (!stash || SvTYPE(stash) != SVt_PVHV)
9319 if (!s) { /* reset ?? searches */
9320 MAGIC * const mg = mg_find((const SV *)stash, PERL_MAGIC_symtab);
9322 const U32 count = mg->mg_len / sizeof(PMOP**);
9323 PMOP **pmp = (PMOP**) mg->mg_ptr;
9324 PMOP *const *const end = pmp + count;
9328 SvREADONLY_off(PL_regex_pad[(*pmp)->op_pmoffset]);
9330 (*pmp)->op_pmflags &= ~PMf_USED;
9338 /* reset variables */
9340 if (!HvARRAY(stash))
9343 Zero(todo, 256, char);
9347 I32 i = (unsigned char)*s;
9351 max = (unsigned char)*s++;
9352 for ( ; i <= max; i++) {
9355 for (i = 0; i <= (I32) HvMAX(stash); i++) {
9357 for (entry = HvARRAY(stash)[i];
9359 entry = HeNEXT(entry))
9364 if (!todo[(U8)*HeKEY(entry)])
9366 gv = MUTABLE_GV(HeVAL(entry));
9368 if (sv && !SvREADONLY(sv)) {
9369 SV_CHECK_THINKFIRST_COW_DROP(sv);
9370 if (!isGV(sv)) SvOK_off(sv);
9375 if (GvHV(gv) && !HvNAME_get(GvHV(gv))) {
9386 Using various gambits, try to get an IO from an SV: the IO slot if its a
9387 GV; or the recursive result if we're an RV; or the IO slot of the symbol
9388 named after the PV if we're a string.
9390 'Get' magic is ignored on the sv passed in, but will be called on
9391 C<SvRV(sv)> if sv is an RV.
9397 Perl_sv_2io(pTHX_ SV *const sv)
9402 PERL_ARGS_ASSERT_SV_2IO;
9404 switch (SvTYPE(sv)) {
9406 io = MUTABLE_IO(sv);
9410 if (isGV_with_GP(sv)) {
9411 gv = MUTABLE_GV(sv);
9414 Perl_croak(aTHX_ "Bad filehandle: %"HEKf,
9415 HEKfARG(GvNAME_HEK(gv)));
9421 Perl_croak(aTHX_ PL_no_usym, "filehandle");
9423 SvGETMAGIC(SvRV(sv));
9424 return sv_2io(SvRV(sv));
9426 gv = gv_fetchsv_nomg(sv, 0, SVt_PVIO);
9433 if (SvGMAGICAL(sv)) {
9434 newsv = sv_newmortal();
9435 sv_setsv_nomg(newsv, sv);
9437 Perl_croak(aTHX_ "Bad filehandle: %"SVf, SVfARG(newsv));
9447 Using various gambits, try to get a CV from an SV; in addition, try if
9448 possible to set C<*st> and C<*gvp> to the stash and GV associated with it.
9449 The flags in C<lref> are passed to gv_fetchsv.
9455 Perl_sv_2cv(pTHX_ SV *sv, HV **const st, GV **const gvp, const I32 lref)
9460 PERL_ARGS_ASSERT_SV_2CV;
9467 switch (SvTYPE(sv)) {
9471 return MUTABLE_CV(sv);
9481 sv = amagic_deref_call(sv, to_cv_amg);
9484 if (SvTYPE(sv) == SVt_PVCV) {
9485 cv = MUTABLE_CV(sv);
9490 else if(SvGETMAGIC(sv), isGV_with_GP(sv))
9491 gv = MUTABLE_GV(sv);
9493 Perl_croak(aTHX_ "Not a subroutine reference");
9495 else if (isGV_with_GP(sv)) {
9496 gv = MUTABLE_GV(sv);
9499 gv = gv_fetchsv_nomg(sv, lref, SVt_PVCV);
9506 /* Some flags to gv_fetchsv mean don't really create the GV */
9507 if (!isGV_with_GP(gv)) {
9512 if (lref & ~GV_ADDMG && !GvCVu(gv)) {
9513 /* XXX this is probably not what they think they're getting.
9514 * It has the same effect as "sub name;", i.e. just a forward
9525 Returns true if the SV has a true value by Perl's rules.
9526 Use the C<SvTRUE> macro instead, which may call C<sv_true()> or may
9527 instead use an in-line version.
9533 Perl_sv_true(pTHX_ SV *const sv)
9538 const XPV* const tXpv = (XPV*)SvANY(sv);
9540 (tXpv->xpv_cur > 1 ||
9541 (tXpv->xpv_cur && *sv->sv_u.svu_pv != '0')))
9548 return SvIVX(sv) != 0;
9551 return SvNVX(sv) != 0.0;
9553 return sv_2bool(sv);
9559 =for apidoc sv_pvn_force
9561 Get a sensible string out of the SV somehow.
9562 A private implementation of the C<SvPV_force> macro for compilers which
9563 can't cope with complex macro expressions. Always use the macro instead.
9565 =for apidoc sv_pvn_force_flags
9567 Get a sensible string out of the SV somehow.
9568 If C<flags> has C<SV_GMAGIC> bit set, will C<mg_get> on C<sv> if
9569 appropriate, else not. C<sv_pvn_force> and C<sv_pvn_force_nomg> are
9570 implemented in terms of this function.
9571 You normally want to use the various wrapper macros instead: see
9572 C<SvPV_force> and C<SvPV_force_nomg>
9578 Perl_sv_pvn_force_flags(pTHX_ SV *const sv, STRLEN *const lp, const I32 flags)
9580 PERL_ARGS_ASSERT_SV_PVN_FORCE_FLAGS;
9582 if (flags & SV_GMAGIC) SvGETMAGIC(sv);
9583 if (SvTHINKFIRST(sv) && (!SvROK(sv) || SvREADONLY(sv)))
9584 sv_force_normal_flags(sv, 0);
9594 if (SvTYPE(sv) > SVt_PVLV
9595 || isGV_with_GP(sv))
9596 /* diag_listed_as: Can't coerce %s to %s in %s */
9597 Perl_croak(aTHX_ "Can't coerce %s to string in %s", sv_reftype(sv,0),
9599 s = sv_2pv_flags(sv, &len, flags &~ SV_GMAGIC);
9606 if (SvTYPE(sv) < SVt_PV ||
9607 s != SvPVX_const(sv)) { /* Almost, but not quite, sv_setpvn() */
9610 SvUPGRADE(sv, SVt_PV); /* Never FALSE */
9611 SvGROW(sv, len + 1);
9612 Move(s,SvPVX(sv),len,char);
9614 SvPVX(sv)[len] = '\0';
9617 SvPOK_on(sv); /* validate pointer */
9619 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
9620 PTR2UV(sv),SvPVX_const(sv)));
9623 (void)SvPOK_only_UTF8(sv);
9624 return SvPVX_mutable(sv);
9628 =for apidoc sv_pvbyten_force
9630 The backend for the C<SvPVbytex_force> macro. Always use the macro
9637 Perl_sv_pvbyten_force(pTHX_ SV *const sv, STRLEN *const lp)
9639 PERL_ARGS_ASSERT_SV_PVBYTEN_FORCE;
9641 sv_pvn_force(sv,lp);
9642 sv_utf8_downgrade(sv,0);
9648 =for apidoc sv_pvutf8n_force
9650 The backend for the C<SvPVutf8x_force> macro. Always use the macro
9657 Perl_sv_pvutf8n_force(pTHX_ SV *const sv, STRLEN *const lp)
9659 PERL_ARGS_ASSERT_SV_PVUTF8N_FORCE;
9662 sv_utf8_upgrade_nomg(sv);
9668 =for apidoc sv_reftype
9670 Returns a string describing what the SV is a reference to.
9676 Perl_sv_reftype(pTHX_ const SV *const sv, const int ob)
9678 PERL_ARGS_ASSERT_SV_REFTYPE;
9679 if (ob && SvOBJECT(sv)) {
9680 return SvPV_nolen_const(sv_ref(NULL, sv, ob));
9683 /* WARNING - There is code, for instance in mg.c, that assumes that
9684 * the only reason that sv_reftype(sv,0) would return a string starting
9685 * with 'L' or 'S' is that it is a LVALUE or a SCALAR.
9686 * Yes this a dodgy way to do type checking, but it saves practically reimplementing
9687 * this routine inside other subs, and it saves time.
9688 * Do not change this assumption without searching for "dodgy type check" in
9691 switch (SvTYPE(sv)) {
9706 case SVt_PVLV: return (char *) (SvROK(sv) ? "REF"
9707 /* tied lvalues should appear to be
9708 * scalars for backwards compatibility */
9709 : (isALPHA_FOLD_EQ(LvTYPE(sv), 't'))
9710 ? "SCALAR" : "LVALUE");
9711 case SVt_PVAV: return "ARRAY";
9712 case SVt_PVHV: return "HASH";
9713 case SVt_PVCV: return "CODE";
9714 case SVt_PVGV: return (char *) (isGV_with_GP(sv)
9715 ? "GLOB" : "SCALAR");
9716 case SVt_PVFM: return "FORMAT";
9717 case SVt_PVIO: return "IO";
9718 case SVt_INVLIST: return "INVLIST";
9719 case SVt_REGEXP: return "REGEXP";
9720 default: return "UNKNOWN";
9728 Returns a SV describing what the SV passed in is a reference to.
9734 Perl_sv_ref(pTHX_ SV *dst, const SV *const sv, const int ob)
9736 PERL_ARGS_ASSERT_SV_REF;
9739 dst = sv_newmortal();
9741 if (ob && SvOBJECT(sv)) {
9742 HvNAME_get(SvSTASH(sv))
9743 ? sv_sethek(dst, HvNAME_HEK(SvSTASH(sv)))
9744 : sv_setpvn(dst, "__ANON__", 8);
9747 const char * reftype = sv_reftype(sv, 0);
9748 sv_setpv(dst, reftype);
9754 =for apidoc sv_isobject
9756 Returns a boolean indicating whether the SV is an RV pointing to a blessed
9757 object. If the SV is not an RV, or if the object is not blessed, then this
9764 Perl_sv_isobject(pTHX_ SV *sv)
9780 Returns a boolean indicating whether the SV is blessed into the specified
9781 class. This does not check for subtypes; use C<sv_derived_from> to verify
9782 an inheritance relationship.
9788 Perl_sv_isa(pTHX_ SV *sv, const char *const name)
9792 PERL_ARGS_ASSERT_SV_ISA;
9802 hvname = HvNAME_get(SvSTASH(sv));
9806 return strEQ(hvname, name);
9812 Creates a new SV for the existing RV, C<rv>, to point to. If C<rv> is not an
9813 RV then it will be upgraded to one. If C<classname> is non-null then the new
9814 SV will be blessed in the specified package. The new SV is returned and its
9815 reference count is 1. The reference count 1 is owned by C<rv>.
9821 Perl_newSVrv(pTHX_ SV *const rv, const char *const classname)
9825 PERL_ARGS_ASSERT_NEWSVRV;
9829 SV_CHECK_THINKFIRST_COW_DROP(rv);
9831 if (SvTYPE(rv) >= SVt_PVMG) {
9832 const U32 refcnt = SvREFCNT(rv);
9836 SvREFCNT(rv) = refcnt;
9838 sv_upgrade(rv, SVt_IV);
9839 } else if (SvROK(rv)) {
9840 SvREFCNT_dec(SvRV(rv));
9842 prepare_SV_for_RV(rv);
9850 HV* const stash = gv_stashpv(classname, GV_ADD);
9851 (void)sv_bless(rv, stash);
9857 Perl_newSVavdefelem(pTHX_ AV *av, SSize_t ix, bool extendible)
9859 SV * const lv = newSV_type(SVt_PVLV);
9860 PERL_ARGS_ASSERT_NEWSVAVDEFELEM;
9862 sv_magic(lv, NULL, PERL_MAGIC_defelem, NULL, 0);
9863 LvTARG(lv) = SvREFCNT_inc_simple_NN(av);
9864 LvSTARGOFF(lv) = ix;
9865 LvTARGLEN(lv) = extendible ? 1 : (STRLEN)UV_MAX;
9870 =for apidoc sv_setref_pv
9872 Copies a pointer into a new SV, optionally blessing the SV. The C<rv>
9873 argument will be upgraded to an RV. That RV will be modified to point to
9874 the new SV. If the C<pv> argument is NULL then C<PL_sv_undef> will be placed
9875 into the SV. The C<classname> argument indicates the package for the
9876 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
9877 will have a reference count of 1, and the RV will be returned.
9879 Do not use with other Perl types such as HV, AV, SV, CV, because those
9880 objects will become corrupted by the pointer copy process.
9882 Note that C<sv_setref_pvn> copies the string while this copies the pointer.
9888 Perl_sv_setref_pv(pTHX_ SV *const rv, const char *const classname, void *const pv)
9890 PERL_ARGS_ASSERT_SV_SETREF_PV;
9893 sv_setsv(rv, &PL_sv_undef);
9897 sv_setiv(newSVrv(rv,classname), PTR2IV(pv));
9902 =for apidoc sv_setref_iv
9904 Copies an integer into a new SV, optionally blessing the SV. The C<rv>
9905 argument will be upgraded to an RV. That RV will be modified to point to
9906 the new SV. The C<classname> argument indicates the package for the
9907 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
9908 will have a reference count of 1, and the RV will be returned.
9914 Perl_sv_setref_iv(pTHX_ SV *const rv, const char *const classname, const IV iv)
9916 PERL_ARGS_ASSERT_SV_SETREF_IV;
9918 sv_setiv(newSVrv(rv,classname), iv);
9923 =for apidoc sv_setref_uv
9925 Copies an unsigned integer into a new SV, optionally blessing the SV. The C<rv>
9926 argument will be upgraded to an RV. That RV will be modified to point to
9927 the new SV. The C<classname> argument indicates the package for the
9928 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
9929 will have a reference count of 1, and the RV will be returned.
9935 Perl_sv_setref_uv(pTHX_ SV *const rv, const char *const classname, const UV uv)
9937 PERL_ARGS_ASSERT_SV_SETREF_UV;
9939 sv_setuv(newSVrv(rv,classname), uv);
9944 =for apidoc sv_setref_nv
9946 Copies a double into a new SV, optionally blessing the SV. The C<rv>
9947 argument will be upgraded to an RV. That RV will be modified to point to
9948 the new SV. The C<classname> argument indicates the package for the
9949 blessing. Set C<classname> to C<NULL> to avoid the blessing. The new SV
9950 will have a reference count of 1, and the RV will be returned.
9956 Perl_sv_setref_nv(pTHX_ SV *const rv, const char *const classname, const NV nv)
9958 PERL_ARGS_ASSERT_SV_SETREF_NV;
9960 sv_setnv(newSVrv(rv,classname), nv);
9965 =for apidoc sv_setref_pvn
9967 Copies a string into a new SV, optionally blessing the SV. The length of the
9968 string must be specified with C<n>. The C<rv> argument will be upgraded to
9969 an RV. That RV will be modified to point to the new SV. The C<classname>
9970 argument indicates the package for the blessing. Set C<classname> to
9971 C<NULL> to avoid the blessing. The new SV will have a reference count
9972 of 1, and the RV will be returned.
9974 Note that C<sv_setref_pv> copies the pointer while this copies the string.
9980 Perl_sv_setref_pvn(pTHX_ SV *const rv, const char *const classname,
9981 const char *const pv, const STRLEN n)
9983 PERL_ARGS_ASSERT_SV_SETREF_PVN;
9985 sv_setpvn(newSVrv(rv,classname), pv, n);
9990 =for apidoc sv_bless
9992 Blesses an SV into a specified package. The SV must be an RV. The package
9993 must be designated by its stash (see C<gv_stashpv()>). The reference count
9994 of the SV is unaffected.
10000 Perl_sv_bless(pTHX_ SV *const sv, HV *const stash)
10003 HV *oldstash = NULL;
10005 PERL_ARGS_ASSERT_SV_BLESS;
10009 Perl_croak(aTHX_ "Can't bless non-reference value");
10011 if (SvFLAGS(tmpRef) & (SVs_OBJECT|SVf_READONLY)) {
10012 if (SvREADONLY(tmpRef))
10013 Perl_croak_no_modify();
10014 if (SvOBJECT(tmpRef)) {
10015 oldstash = SvSTASH(tmpRef);
10018 SvOBJECT_on(tmpRef);
10019 SvUPGRADE(tmpRef, SVt_PVMG);
10020 SvSTASH_set(tmpRef, MUTABLE_HV(SvREFCNT_inc_simple(stash)));
10021 SvREFCNT_dec(oldstash);
10023 if(SvSMAGICAL(tmpRef))
10024 if(mg_find(tmpRef, PERL_MAGIC_ext) || mg_find(tmpRef, PERL_MAGIC_uvar))
10032 /* Downgrades a PVGV to a PVMG. If it's actually a PVLV, we leave the type
10033 * as it is after unglobbing it.
10036 PERL_STATIC_INLINE void
10037 S_sv_unglob(pTHX_ SV *const sv, U32 flags)
10041 SV * const temp = flags & SV_COW_DROP_PV ? NULL : sv_newmortal();
10043 PERL_ARGS_ASSERT_SV_UNGLOB;
10045 assert(SvTYPE(sv) == SVt_PVGV || SvTYPE(sv) == SVt_PVLV);
10047 if (!(flags & SV_COW_DROP_PV))
10048 gv_efullname3(temp, MUTABLE_GV(sv), "*");
10050 SvREFCNT_inc_simple_void_NN(sv_2mortal(sv));
10052 if(GvCVu((const GV *)sv) && (stash = GvSTASH(MUTABLE_GV(sv)))
10053 && HvNAME_get(stash))
10054 mro_method_changed_in(stash);
10055 gp_free(MUTABLE_GV(sv));
10058 sv_del_backref(MUTABLE_SV(GvSTASH(sv)), sv);
10059 GvSTASH(sv) = NULL;
10062 if (GvNAME_HEK(sv)) {
10063 unshare_hek(GvNAME_HEK(sv));
10065 isGV_with_GP_off(sv);
10067 if(SvTYPE(sv) == SVt_PVGV) {
10068 /* need to keep SvANY(sv) in the right arena */
10069 xpvmg = new_XPVMG();
10070 StructCopy(SvANY(sv), xpvmg, XPVMG);
10071 del_XPVGV(SvANY(sv));
10074 SvFLAGS(sv) &= ~SVTYPEMASK;
10075 SvFLAGS(sv) |= SVt_PVMG;
10078 /* Intentionally not calling any local SET magic, as this isn't so much a
10079 set operation as merely an internal storage change. */
10080 if (flags & SV_COW_DROP_PV) SvOK_off(sv);
10081 else sv_setsv_flags(sv, temp, 0);
10083 if ((const GV *)sv == PL_last_in_gv)
10084 PL_last_in_gv = NULL;
10085 else if ((const GV *)sv == PL_statgv)
10090 =for apidoc sv_unref_flags
10092 Unsets the RV status of the SV, and decrements the reference count of
10093 whatever was being referenced by the RV. This can almost be thought of
10094 as a reversal of C<newSVrv>. The C<cflags> argument can contain
10095 C<SV_IMMEDIATE_UNREF> to force the reference count to be decremented
10096 (otherwise the decrementing is conditional on the reference count being
10097 different from one or the reference being a readonly SV).
10104 Perl_sv_unref_flags(pTHX_ SV *const ref, const U32 flags)
10106 SV* const target = SvRV(ref);
10108 PERL_ARGS_ASSERT_SV_UNREF_FLAGS;
10110 if (SvWEAKREF(ref)) {
10111 sv_del_backref(target, ref);
10112 SvWEAKREF_off(ref);
10113 SvRV_set(ref, NULL);
10116 SvRV_set(ref, NULL);
10118 /* You can't have a || SvREADONLY(target) here, as $a = $$a, where $a was
10119 assigned to as BEGIN {$a = \"Foo"} will fail. */
10120 if (SvREFCNT(target) != 1 || (flags & SV_IMMEDIATE_UNREF))
10121 SvREFCNT_dec_NN(target);
10122 else /* XXX Hack, but hard to make $a=$a->[1] work otherwise */
10123 sv_2mortal(target); /* Schedule for freeing later */
10127 =for apidoc sv_untaint
10129 Untaint an SV. Use C<SvTAINTED_off> instead.
10135 Perl_sv_untaint(pTHX_ SV *const sv)
10137 PERL_ARGS_ASSERT_SV_UNTAINT;
10138 PERL_UNUSED_CONTEXT;
10140 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10141 MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10148 =for apidoc sv_tainted
10150 Test an SV for taintedness. Use C<SvTAINTED> instead.
10156 Perl_sv_tainted(pTHX_ SV *const sv)
10158 PERL_ARGS_ASSERT_SV_TAINTED;
10159 PERL_UNUSED_CONTEXT;
10161 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
10162 const MAGIC * const mg = mg_find(sv, PERL_MAGIC_taint);
10163 if (mg && (mg->mg_len & 1) )
10170 =for apidoc sv_setpviv
10172 Copies an integer into the given SV, also updating its string value.
10173 Does not handle 'set' magic. See C<sv_setpviv_mg>.
10179 Perl_sv_setpviv(pTHX_ SV *const sv, const IV iv)
10181 char buf[TYPE_CHARS(UV)];
10183 char * const ptr = uiv_2buf(buf, iv, 0, 0, &ebuf);
10185 PERL_ARGS_ASSERT_SV_SETPVIV;
10187 sv_setpvn(sv, ptr, ebuf - ptr);
10191 =for apidoc sv_setpviv_mg
10193 Like C<sv_setpviv>, but also handles 'set' magic.
10199 Perl_sv_setpviv_mg(pTHX_ SV *const sv, const IV iv)
10201 PERL_ARGS_ASSERT_SV_SETPVIV_MG;
10203 sv_setpviv(sv, iv);
10207 #if defined(PERL_IMPLICIT_CONTEXT)
10209 /* pTHX_ magic can't cope with varargs, so this is a no-context
10210 * version of the main function, (which may itself be aliased to us).
10211 * Don't access this version directly.
10215 Perl_sv_setpvf_nocontext(SV *const sv, const char *const pat, ...)
10220 PERL_ARGS_ASSERT_SV_SETPVF_NOCONTEXT;
10222 va_start(args, pat);
10223 sv_vsetpvf(sv, pat, &args);
10227 /* pTHX_ magic can't cope with varargs, so this is a no-context
10228 * version of the main function, (which may itself be aliased to us).
10229 * Don't access this version directly.
10233 Perl_sv_setpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10238 PERL_ARGS_ASSERT_SV_SETPVF_MG_NOCONTEXT;
10240 va_start(args, pat);
10241 sv_vsetpvf_mg(sv, pat, &args);
10247 =for apidoc sv_setpvf
10249 Works like C<sv_catpvf> but copies the text into the SV instead of
10250 appending it. Does not handle 'set' magic. See C<sv_setpvf_mg>.
10256 Perl_sv_setpvf(pTHX_ SV *const sv, const char *const pat, ...)
10260 PERL_ARGS_ASSERT_SV_SETPVF;
10262 va_start(args, pat);
10263 sv_vsetpvf(sv, pat, &args);
10268 =for apidoc sv_vsetpvf
10270 Works like C<sv_vcatpvf> but copies the text into the SV instead of
10271 appending it. Does not handle 'set' magic. See C<sv_vsetpvf_mg>.
10273 Usually used via its frontend C<sv_setpvf>.
10279 Perl_sv_vsetpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10281 PERL_ARGS_ASSERT_SV_VSETPVF;
10283 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10287 =for apidoc sv_setpvf_mg
10289 Like C<sv_setpvf>, but also handles 'set' magic.
10295 Perl_sv_setpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10299 PERL_ARGS_ASSERT_SV_SETPVF_MG;
10301 va_start(args, pat);
10302 sv_vsetpvf_mg(sv, pat, &args);
10307 =for apidoc sv_vsetpvf_mg
10309 Like C<sv_vsetpvf>, but also handles 'set' magic.
10311 Usually used via its frontend C<sv_setpvf_mg>.
10317 Perl_sv_vsetpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10319 PERL_ARGS_ASSERT_SV_VSETPVF_MG;
10321 sv_vsetpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10325 #if defined(PERL_IMPLICIT_CONTEXT)
10327 /* pTHX_ magic can't cope with varargs, so this is a no-context
10328 * version of the main function, (which may itself be aliased to us).
10329 * Don't access this version directly.
10333 Perl_sv_catpvf_nocontext(SV *const sv, const char *const pat, ...)
10338 PERL_ARGS_ASSERT_SV_CATPVF_NOCONTEXT;
10340 va_start(args, pat);
10341 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10345 /* pTHX_ magic can't cope with varargs, so this is a no-context
10346 * version of the main function, (which may itself be aliased to us).
10347 * Don't access this version directly.
10351 Perl_sv_catpvf_mg_nocontext(SV *const sv, const char *const pat, ...)
10356 PERL_ARGS_ASSERT_SV_CATPVF_MG_NOCONTEXT;
10358 va_start(args, pat);
10359 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10366 =for apidoc sv_catpvf
10368 Processes its arguments like C<sprintf> and appends the formatted
10369 output to an SV. If the appended data contains "wide" characters
10370 (including, but not limited to, SVs with a UTF-8 PV formatted with %s,
10371 and characters >255 formatted with %c), the original SV might get
10372 upgraded to UTF-8. Handles 'get' magic, but not 'set' magic. See
10373 C<sv_catpvf_mg>. If the original SV was UTF-8, the pattern should be
10374 valid UTF-8; if the original SV was bytes, the pattern should be too.
10379 Perl_sv_catpvf(pTHX_ SV *const sv, const char *const pat, ...)
10383 PERL_ARGS_ASSERT_SV_CATPVF;
10385 va_start(args, pat);
10386 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10391 =for apidoc sv_vcatpvf
10393 Processes its arguments like C<vsprintf> and appends the formatted output
10394 to an SV. Does not handle 'set' magic. See C<sv_vcatpvf_mg>.
10396 Usually used via its frontend C<sv_catpvf>.
10402 Perl_sv_vcatpvf(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10404 PERL_ARGS_ASSERT_SV_VCATPVF;
10406 sv_vcatpvfn_flags(sv, pat, strlen(pat), args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10410 =for apidoc sv_catpvf_mg
10412 Like C<sv_catpvf>, but also handles 'set' magic.
10418 Perl_sv_catpvf_mg(pTHX_ SV *const sv, const char *const pat, ...)
10422 PERL_ARGS_ASSERT_SV_CATPVF_MG;
10424 va_start(args, pat);
10425 sv_vcatpvfn_flags(sv, pat, strlen(pat), &args, NULL, 0, NULL, SV_GMAGIC|SV_SMAGIC);
10431 =for apidoc sv_vcatpvf_mg
10433 Like C<sv_vcatpvf>, but also handles 'set' magic.
10435 Usually used via its frontend C<sv_catpvf_mg>.
10441 Perl_sv_vcatpvf_mg(pTHX_ SV *const sv, const char *const pat, va_list *const args)
10443 PERL_ARGS_ASSERT_SV_VCATPVF_MG;
10445 sv_vcatpvfn(sv, pat, strlen(pat), args, NULL, 0, NULL);
10450 =for apidoc sv_vsetpvfn
10452 Works like C<sv_vcatpvfn> but copies the text into the SV instead of
10455 Usually used via one of its frontends C<sv_vsetpvf> and C<sv_vsetpvf_mg>.
10461 Perl_sv_vsetpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10462 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
10464 PERL_ARGS_ASSERT_SV_VSETPVFN;
10467 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, 0);
10472 * Warn of missing argument to sprintf, and then return a defined value
10473 * to avoid inappropriate "use of uninit" warnings [perl #71000].
10476 S_vcatpvfn_missing_argument(pTHX) {
10477 if (ckWARN(WARN_MISSING)) {
10478 Perl_warner(aTHX_ packWARN(WARN_MISSING), "Missing argument in %s",
10479 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
10486 S_expect_number(pTHX_ char **const pattern)
10490 PERL_ARGS_ASSERT_EXPECT_NUMBER;
10492 switch (**pattern) {
10493 case '1': case '2': case '3':
10494 case '4': case '5': case '6':
10495 case '7': case '8': case '9':
10496 var = *(*pattern)++ - '0';
10497 while (isDIGIT(**pattern)) {
10498 const I32 tmp = var * 10 + (*(*pattern)++ - '0');
10500 Perl_croak(aTHX_ "Integer overflow in format string for %s", (PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn"));
10508 S_F0convert(NV nv, char *const endbuf, STRLEN *const len)
10510 const int neg = nv < 0;
10513 PERL_ARGS_ASSERT_F0CONVERT;
10521 if (uv & 1 && uv == nv)
10522 uv--; /* Round to even */
10524 const unsigned dig = uv % 10;
10526 } while (uv /= 10);
10537 =for apidoc sv_vcatpvfn
10539 =for apidoc sv_vcatpvfn_flags
10541 Processes its arguments like C<vsprintf> and appends the formatted output
10542 to an SV. Uses an array of SVs if the C style variable argument list is
10543 missing (NULL). When running with taint checks enabled, indicates via
10544 C<maybe_tainted> if results are untrustworthy (often due to the use of
10547 If called as C<sv_vcatpvfn> or flags include C<SV_GMAGIC>, calls get magic.
10549 Usually used via one of its frontends C<sv_vcatpvf> and C<sv_vcatpvf_mg>.
10554 #define VECTORIZE_ARGS vecsv = va_arg(*args, SV*);\
10555 vecstr = (U8*)SvPV_const(vecsv,veclen);\
10556 vec_utf8 = DO_UTF8(vecsv);
10558 /* XXX maybe_tainted is never assigned to, so the doc above is lying. */
10561 Perl_sv_vcatpvfn(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10562 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted)
10564 PERL_ARGS_ASSERT_SV_VCATPVFN;
10566 sv_vcatpvfn_flags(sv, pat, patlen, args, svargs, svmax, maybe_tainted, SV_GMAGIC|SV_SMAGIC);
10569 /* vhex will contain the values (0..15) of the hex digits ("nybbles"
10570 * of 4 bits); 1 for the implicit 1, and at most 128 bits of mantissa,
10571 * four bits per xdigit. */
10572 #define VHEX_SIZE (1+128/4)
10574 /* If we do not have a known long double format, (including not using
10575 * long doubles, or long doubles being equal to doubles) then we will
10576 * fall back to the ldexp/frexp route, with which we can retrieve at
10577 * most as many bits as our widest unsigned integer type is. We try
10578 * to get a 64-bit unsigned integer even if we are not having 64-bit
10580 #if defined(HAS_QUAD) && defined(Uquad_t)
10581 # define MANTISSATYPE Uquad_t
10582 # define MANTISSASIZE 8
10584 # define MANTISSATYPE UV /* May lose precision if UVSIZE is not 8. */
10585 # define MANTISSASIZE UVSIZE
10588 /* S_hextract() is a helper for Perl_sv_vcatpvfn_flags, for extracting
10589 * the hexadecimal values (for %a/%A). The nv is the NV where the value
10590 * are being extracted from (either directly from the long double in-memory
10591 * presentation, or from the uquad computed via frexp+ldexp). frexp also
10592 * is used to update the exponent. vhex is the pointer to the beginning
10593 * of the output buffer (of VHEX_SIZE).
10595 * The tricky part is that S_hextract() needs to be called twice:
10596 * the first time with vend as NULL, and the second time with vend as
10597 * the pointer returned by the first call. What happens is that on
10598 * the first round the output size is computed, and the intended
10599 * extraction sanity checked. On the second round the actual output
10600 * (the extraction of the hexadecimal values) takes place.
10601 * Sanity failures cause fatal failures during both rounds. */
10603 S_hextract(pTHX_ const NV nv, int* exponent, U8* vhex, U8* vend)
10607 int ixmin = 0, ixmax = 0;
10609 /* XXX Inf/NaN/denormal handling in the HEXTRACT_IMPLICIT_BIT,
10610 * and elsewhere. */
10612 /* These macros are just to reduce typos, they have multiple
10613 * repetitions below, but usually only one (or sometimes two)
10614 * of them is really being used. */
10615 /* HEXTRACT_OUTPUT() extracts the high nybble first. */
10616 #define HEXTRACT_OUTPUT_HI(ix) (*v++ = nvp[ix] >> 4)
10617 #define HEXTRACT_OUTPUT_LO(ix) (*v++ = nvp[ix] & 0xF)
10618 #define HEXTRACT_OUTPUT(ix) \
10620 HEXTRACT_OUTPUT_HI(ix); \
10621 HEXTRACT_OUTPUT_LO(ix); \
10623 #define HEXTRACT_COUNT(ix, c) \
10628 else if (ix > ixmax) \
10631 #define HEXTRACT_IMPLICIT_BIT() \
10639 /* First see if we are using long doubles. */
10640 #if NVSIZE > DOUBLESIZE && LONG_DOUBLEKIND != LONG_DOUBLE_IS_DOUBLE
10641 const U8* nvp = (const U8*)(&nv);
10642 # define HEXTRACTSIZE NVSIZE
10643 (void)Perl_frexp(PERL_ABS(nv), exponent);
10644 # if LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN
10645 /* Used in e.g. VMS and HP-UX IA-64, e.g. -0.1L:
10646 * 9a 99 99 99 99 99 99 99 99 99 99 99 99 99 fb 3f */
10647 /* The bytes 13..0 are the mantissa/fraction,
10648 * the 15,14 are the sign+exponent. */
10649 HEXTRACT_IMPLICIT_BIT();
10650 for (ix = 13; ix >= 0; ix--) {
10652 HEXTRACT_OUTPUT(ix);
10654 HEXTRACT_COUNT(ix, 2);
10656 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_BIG_ENDIAN
10657 /* Used in e.g. Solaris Sparc and HP-UX PA-RISC, e.g. -0.1L:
10658 * bf fb 99 99 99 99 99 99 99 99 99 99 99 99 99 9a */
10659 /* The bytes 2..15 are the mantissa/fraction,
10660 * the 0,1 are the sign+exponent. */
10661 HEXTRACT_IMPLICIT_BIT();
10662 for (ix = 2; ix <= 15; ix++) {
10664 HEXTRACT_OUTPUT(ix);
10666 HEXTRACT_COUNT(ix, 2);
10668 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN
10669 /* x86 80-bit "extended precision", 64 bits of mantissa / fraction /
10670 * significand, 15 bits of exponent, 1 bit of sign. NVSIZE can
10671 * be either 12 (ILP32, Solaris x86) or 16 (LP64, Linux and OS X),
10672 * meaning that 2 or 6 bytes are empty padding. */
10673 /* The bytes 7..0 are the mantissa/fraction */
10674 /* There explicitly is *no* implicit bit in this case. */
10675 for (ix = 7; ix >= 0; ix--) {
10677 HEXTRACT_OUTPUT(ix);
10679 HEXTRACT_COUNT(ix, 2);
10681 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN
10682 /* (does this format ever happen?) */
10683 /* There explicitly is *no* implicit bit in this case. */
10684 for (ix = 0; ix < 8; ix++) {
10686 HEXTRACT_OUTPUT(ix);
10688 HEXTRACT_COUNT(ix, 2);
10690 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_DOUBLEDOUBLE_128_BIT_LITTLE_ENDIAN
10691 /* Where is this used?
10692 * 9a 99 99 99 99 99 59 bc 9a 99 99 99 99 99 b9 3f */
10693 HEXTRACT_IMPLICIT_BIT();
10695 HEXTRACT_OUTPUT_LO(14);
10697 HEXTRACT_COUNT(14, 1);
10698 for (ix = 13; ix >= 8; ix--) {
10700 HEXTRACT_OUTPUT(ix);
10702 HEXTRACT_COUNT(ix, 2);
10704 /* XXX not extracting from the second double -- see the discussion
10705 * below for the big endian double double. */
10708 HEXTRACT_OUTPUT_LO(6);
10710 HEXTRACT_COUNT(6, 1);
10711 for (ix = 5; ix >= 0; ix--) {
10713 HEXTRACT_OUTPUT(ix);
10715 HEXTRACT_COUNT(ix, 2);
10718 # elif LONG_DOUBLEKIND == LONG_DOUBLE_IS_DOUBLEDOUBLE_128_BIT_BIG_ENDIAN
10719 /* Used in e.g. PPC/Power (AIX) and MIPS.
10721 * The mantissa bits are in two separate stretches, e.g. for -0.1L:
10722 * 3f b9 99 99 99 99 99 9a bc 59 99 99 99 99 99 9a
10724 HEXTRACT_IMPLICIT_BIT();
10726 HEXTRACT_OUTPUT_LO(1);
10728 HEXTRACT_COUNT(1, 1);
10729 for (ix = 2; ix < 8; ix++) {
10731 HEXTRACT_OUTPUT(ix);
10733 HEXTRACT_COUNT(ix, 2);
10735 /* XXX not extracting the second double mantissa bits- this is not
10736 * right nor ideal (we effectively reduce the output format to
10737 * that of a "single double", only 53 bits), but we do not know
10738 * exactly how to do the extraction correctly so that it matches
10739 * the semantics of, say, the IEEE quadruple float. */
10742 HEXTRACT_OUTPUT_LO(9);
10744 HEXTRACT_COUNT(9, 1);
10745 for (ix = 10; ix < 16; ix++) {
10747 HEXTRACT_OUTPUT(ix);
10749 HEXTRACT_COUNT(ix, 2);
10754 "Hexadecimal float: unsupported long double format");
10757 /* If not using long doubles (or if the long double format is
10758 * known but not yet supported), try to retrieve the mantissa bits
10759 * via frexp+ldexp. */
10761 NV norm = Perl_frexp(PERL_ABS(nv), exponent);
10762 /* Theoretically we have all the bytes [0, MANTISSASIZE-1] to
10763 * inspect; but in practice we don't want the leading nybbles that
10764 * are zero. With the common IEEE 754 value for NV_MANT_DIG being
10765 * 53, we want the limit byte to be (int)((53-1)/8) == 6.
10767 * Note that this is _not_ inspecting the in-memory format of the
10768 * nv (as opposed to the long double method), but instead the UV
10769 * retrieved with the frexp+ldexp invocation. */
10770 # if MANTISSASIZE * 8 > NV_MANT_DIG
10771 MANTISSATYPE mantissa = (MANTISSATYPE)Perl_ldexp(norm, NV_MANT_DIG);
10772 int limit_byte = (NV_MANT_DIG - 1) / 8;
10774 /* There will be low-order precision loss. Try to salvage as many
10775 * bits as possible. Will truncate, not round. */
10776 MANTISSATYPE mantissa =
10778 /* The highest possible shift by two that fits in the
10779 * mantissa and is aligned (by four) the same was as
10781 MANTISSASIZE * 8 - (4 - NV_MANT_DIG % 4));
10782 int limit_byte = MANTISSASIZE - 1;
10784 const U8* nvp = (const U8*)(&mantissa);
10785 # define HEXTRACTSIZE MANTISSASIZE
10786 /* We make here the wild assumption that the endianness of doubles
10787 * is similar to the endianness of integers, and that there is no
10788 * middle-endianness. This may come back to haunt us (the rumor
10789 * has it that ARM can be quite haunted).
10791 * We generate 4-bit xdigits (nybble/nibble) instead of 8-bit
10792 * bytes, since we might need to handle printf precision, and also
10793 * insert the radix.
10795 # if BYTEORDER == 0x12345678 || BYTEORDER == 0x1234 || \
10796 LONG_DOUBLEKIND == LONG_DOUBLE_IS_IEEE_754_128_BIT_LITTLE_ENDIAN || \
10797 LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN || \
10798 LONG_DOUBLEKIND == LONG_DOUBLE_IS_DOUBLEDOUBLE_128_BIT_LITTLE_ENDIAN
10799 /* Little endian. */
10800 for (ix = limit_byte; ix >= 0; ix--) {
10802 HEXTRACT_OUTPUT(ix);
10804 HEXTRACT_COUNT(ix, 2);
10808 for (ix = MANTISSASIZE - 1 - limit_byte; ix < MANTISSASIZE; ix++) {
10810 HEXTRACT_OUTPUT(ix);
10812 HEXTRACT_COUNT(ix, 2);
10815 /* If there are not enough bits in MANTISSATYPE, we couldn't get
10816 * all of them, issue a warning.
10818 * Note that NV_PRESERVES_UV_BITS would not help here, it is the
10819 * wrong way around. */
10820 # if NV_MANT_DIG > MANTISSASIZE * 8
10821 Perl_ck_warner(aTHX_ packWARN(WARN_OVERFLOW),
10822 "Hexadecimal float: precision loss");
10825 /* Croak for various reasons: if the output pointer escaped the
10826 * output buffer, if the extraction index escaped the extraction
10827 * buffer, or if the ending output pointer didn't match the
10828 * previously computed value. */
10829 if (v <= vhex || v - vhex >= VHEX_SIZE ||
10830 ixmin < 0 || ixmax >= HEXTRACTSIZE ||
10831 (vend && v != vend))
10832 Perl_croak(aTHX_ "Hexadecimal float: internal error");
10837 Perl_sv_vcatpvfn_flags(pTHX_ SV *const sv, const char *const pat, const STRLEN patlen,
10838 va_list *const args, SV **const svargs, const I32 svmax, bool *const maybe_tainted,
10843 const char *patend;
10846 static const char nullstr[] = "(null)";
10848 bool has_utf8 = DO_UTF8(sv); /* has the result utf8? */
10849 const bool pat_utf8 = has_utf8; /* the pattern is in utf8? */
10851 /* Times 4: a decimal digit takes more than 3 binary digits.
10852 * NV_DIG: mantissa takes than many decimal digits.
10853 * Plus 32: Playing safe. */
10854 char ebuf[IV_DIG * 4 + NV_DIG + 32];
10855 /* large enough for "%#.#f" --chip */
10856 /* what about long double NVs? --jhi */
10857 bool no_redundant_warning = FALSE; /* did we use any explicit format parameter index? */
10858 bool hexfp = FALSE;
10860 DECLARATION_FOR_STORE_LC_NUMERIC_SET_TO_NEEDED;
10862 PERL_ARGS_ASSERT_SV_VCATPVFN_FLAGS;
10863 PERL_UNUSED_ARG(maybe_tainted);
10865 if (flags & SV_GMAGIC)
10868 /* no matter what, this is a string now */
10869 (void)SvPV_force_nomg(sv, origlen);
10871 /* special-case "", "%s", and "%-p" (SVf - see below) */
10873 if (svmax && ckWARN(WARN_REDUNDANT))
10874 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
10875 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
10878 if (patlen == 2 && pat[0] == '%' && pat[1] == 's') {
10879 if (svmax > 1 && ckWARN(WARN_REDUNDANT))
10880 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
10881 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
10884 const char * const s = va_arg(*args, char*);
10885 sv_catpv_nomg(sv, s ? s : nullstr);
10887 else if (svix < svmax) {
10888 /* we want get magic on the source but not the target. sv_catsv can't do that, though */
10889 SvGETMAGIC(*svargs);
10890 sv_catsv_nomg(sv, *svargs);
10893 S_vcatpvfn_missing_argument(aTHX);
10896 if (args && patlen == 3 && pat[0] == '%' &&
10897 pat[1] == '-' && pat[2] == 'p') {
10898 if (svmax > 1 && ckWARN(WARN_REDUNDANT))
10899 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
10900 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
10901 argsv = MUTABLE_SV(va_arg(*args, void*));
10902 sv_catsv_nomg(sv, argsv);
10906 #ifndef USE_LONG_DOUBLE
10907 /* special-case "%.<number>[gf]" */
10908 if ( !args && patlen <= 5 && pat[0] == '%' && pat[1] == '.'
10909 && (pat[patlen-1] == 'g' || pat[patlen-1] == 'f') ) {
10910 unsigned digits = 0;
10914 while (*pp >= '0' && *pp <= '9')
10915 digits = 10 * digits + (*pp++ - '0');
10917 /* XXX: Why do this `svix < svmax` test? Couldn't we just
10918 format the first argument and WARN_REDUNDANT if svmax > 1?
10919 Munged by Nicholas Clark in v5.13.0-209-g95ea86d */
10920 if (pp - pat == (int)patlen - 1 && svix < svmax) {
10921 const NV nv = SvNV(*svargs);
10923 /* Add check for digits != 0 because it seems that some
10924 gconverts are buggy in this case, and we don't yet have
10925 a Configure test for this. */
10926 if (digits && digits < sizeof(ebuf) - NV_DIG - 10) {
10927 /* 0, point, slack */
10928 STORE_LC_NUMERIC_SET_TO_NEEDED();
10929 PERL_UNUSED_RESULT(Gconvert(nv, (int)digits, 0, ebuf));
10930 sv_catpv_nomg(sv, ebuf);
10931 if (*ebuf) /* May return an empty string for digits==0 */
10934 } else if (!digits) {
10937 if ((p = F0convert(nv, ebuf + sizeof ebuf, &l))) {
10938 sv_catpvn_nomg(sv, p, l);
10944 #endif /* !USE_LONG_DOUBLE */
10946 if (!args && svix < svmax && DO_UTF8(*svargs))
10949 patend = (char*)pat + patlen;
10950 for (p = (char*)pat; p < patend; p = q) {
10953 bool vectorize = FALSE;
10954 bool vectorarg = FALSE;
10955 bool vec_utf8 = FALSE;
10961 bool has_precis = FALSE;
10963 const I32 osvix = svix;
10964 bool is_utf8 = FALSE; /* is this item utf8? */
10965 #ifdef HAS_LDBL_SPRINTF_BUG
10966 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
10967 with sfio - Allen <allens@cpan.org> */
10968 bool fix_ldbl_sprintf_bug = FALSE;
10972 U8 utf8buf[UTF8_MAXBYTES+1];
10973 STRLEN esignlen = 0;
10975 const char *eptr = NULL;
10976 const char *fmtstart;
10979 const U8 *vecstr = NULL;
10986 /* we need a long double target in case HAS_LONG_DOUBLE but
10987 not USE_LONG_DOUBLE
10989 #if defined(HAS_LONG_DOUBLE) && LONG_DOUBLESIZE > DOUBLESIZE
10991 # define myNVgf PERL_PRIgldbl
10994 # define myNVgf NVgf
10999 const char *dotstr = ".";
11000 STRLEN dotstrlen = 1;
11001 I32 efix = 0; /* explicit format parameter index */
11002 I32 ewix = 0; /* explicit width index */
11003 I32 epix = 0; /* explicit precision index */
11004 I32 evix = 0; /* explicit vector index */
11005 bool asterisk = FALSE;
11006 bool infnan = FALSE;
11008 /* echo everything up to the next format specification */
11009 for (q = p; q < patend && *q != '%'; ++q) ;
11011 if (has_utf8 && !pat_utf8)
11012 sv_catpvn_nomg_utf8_upgrade(sv, p, q - p, nsv);
11014 sv_catpvn_nomg(sv, p, q - p);
11023 We allow format specification elements in this order:
11024 \d+\$ explicit format parameter index
11026 v|\*(\d+\$)?v vector with optional (optionally specified) arg
11027 0 flag (as above): repeated to allow "v02"
11028 \d+|\*(\d+\$)? width using optional (optionally specified) arg
11029 \.(\d*|\*(\d+\$)?) precision using optional (optionally specified) arg
11031 [%bcdefginopsuxDFOUX] format (mandatory)
11036 As of perl5.9.3, printf format checking is on by default.
11037 Internally, perl uses %p formats to provide an escape to
11038 some extended formatting. This block deals with those
11039 extensions: if it does not match, (char*)q is reset and
11040 the normal format processing code is used.
11042 Currently defined extensions are:
11043 %p include pointer address (standard)
11044 %-p (SVf) include an SV (previously %_)
11045 %-<num>p include an SV with precision <num>
11047 %3p include a HEK with precision of 256
11048 %4p char* preceded by utf8 flag and length
11049 %<num>p (where num is 1 or > 4) reserved for future
11052 Robin Barker 2005-07-14 (but modified since)
11054 %1p (VDf) removed. RMB 2007-10-19
11061 else if (strnEQ(q, UTF8f, sizeof(UTF8f)-1)) { /* UTF8f */
11062 /* The argument has already gone through cBOOL, so the cast
11064 is_utf8 = (bool)va_arg(*args, int);
11065 elen = va_arg(*args, UV);
11066 eptr = va_arg(*args, char *);
11067 q += sizeof(UTF8f)-1;
11070 n = expect_number(&q);
11072 if (sv) { /* SVf */
11077 argsv = MUTABLE_SV(va_arg(*args, void*));
11078 eptr = SvPV_const(argsv, elen);
11079 if (DO_UTF8(argsv))
11083 else if (n==2 || n==3) { /* HEKf */
11084 HEK * const hek = va_arg(*args, HEK *);
11085 eptr = HEK_KEY(hek);
11086 elen = HEK_LEN(hek);
11087 if (HEK_UTF8(hek)) is_utf8 = TRUE;
11088 if (n==3) precis = 256, has_precis = TRUE;
11092 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
11093 "internal %%<num>p might conflict with future printf extensions");
11099 if ( (width = expect_number(&q)) ) {
11103 if (!no_redundant_warning)
11104 /* I've forgotten if it's a better
11105 micro-optimization to always set this or to
11106 only set it if it's unset */
11107 no_redundant_warning = TRUE;
11119 if (plus == '+' && *q == ' ') /* '+' over ' ' */
11148 if ( (ewix = expect_number(&q)) )
11157 if ((vectorarg = asterisk)) {
11170 width = expect_number(&q);
11173 if (vectorize && vectorarg) {
11174 /* vectorizing, but not with the default "." */
11176 vecsv = va_arg(*args, SV*);
11178 vecsv = (evix > 0 && evix <= svmax)
11179 ? svargs[evix-1] : S_vcatpvfn_missing_argument(aTHX);
11181 vecsv = svix < svmax
11182 ? svargs[svix++] : S_vcatpvfn_missing_argument(aTHX);
11184 dotstr = SvPV_const(vecsv, dotstrlen);
11185 /* Keep the DO_UTF8 test *after* the SvPV call, else things go
11186 bad with tied or overloaded values that return UTF8. */
11187 if (DO_UTF8(vecsv))
11189 else if (has_utf8) {
11190 vecsv = sv_mortalcopy(vecsv);
11191 sv_utf8_upgrade(vecsv);
11192 dotstr = SvPV_const(vecsv, dotstrlen);
11199 i = va_arg(*args, int);
11201 i = (ewix ? ewix <= svmax : svix < svmax) ?
11202 SvIVx(svargs[ewix ? ewix-1 : svix++]) : 0;
11204 width = (i < 0) ? -i : i;
11214 if ( ((epix = expect_number(&q))) && (*q++ != '$') )
11216 /* XXX: todo, support specified precision parameter */
11220 i = va_arg(*args, int);
11222 i = (ewix ? ewix <= svmax : svix < svmax)
11223 ? SvIVx(svargs[ewix ? ewix-1 : svix++]) : 0;
11225 has_precis = !(i < 0);
11229 while (isDIGIT(*q))
11230 precis = precis * 10 + (*q++ - '0');
11239 else if (efix ? (efix > 0 && efix <= svmax) : svix < svmax) {
11240 vecsv = svargs[efix ? efix-1 : svix++];
11241 vecstr = (U8*)SvPV_const(vecsv,veclen);
11242 vec_utf8 = DO_UTF8(vecsv);
11244 /* if this is a version object, we need to convert
11245 * back into v-string notation and then let the
11246 * vectorize happen normally
11248 if (sv_isobject(vecsv) && sv_derived_from(vecsv, "version")) {
11249 if ( hv_exists(MUTABLE_HV(SvRV(vecsv)), "alpha", 5 ) ) {
11250 Perl_ck_warner_d(aTHX_ packWARN(WARN_PRINTF),
11251 "vector argument not supported with alpha versions");
11254 vecsv = sv_newmortal();
11255 scan_vstring((char *)vecstr, (char *)vecstr + veclen,
11257 vecstr = (U8*)SvPV_const(vecsv, veclen);
11258 vec_utf8 = DO_UTF8(vecsv);
11272 case 'I': /* Ix, I32x, and I64x */
11273 # ifdef USE_64_BIT_INT
11274 if (q[1] == '6' && q[2] == '4') {
11280 if (q[1] == '3' && q[2] == '2') {
11284 # ifdef USE_64_BIT_INT
11290 #if IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)
11302 #if IVSIZE >= 8 || defined(HAS_LONG_DOUBLE)
11303 if (*q == 'l') { /* lld, llf */
11312 if (*++q == 'h') { /* hhd, hhu */
11341 if (!vectorize && !args) {
11343 const I32 i = efix-1;
11344 argsv = (i >= 0 && i < svmax)
11345 ? svargs[i] : S_vcatpvfn_missing_argument(aTHX);
11347 argsv = (svix >= 0 && svix < svmax)
11348 ? svargs[svix++] : S_vcatpvfn_missing_argument(aTHX);
11352 if (argsv && SvNOK(argsv)) {
11353 /* XXX va_arg(*args) case? */
11354 infnan = Perl_isinfnan(SvNV(argsv));
11357 switch (c = *q++) {
11364 uv = (args) ? va_arg(*args, int) :
11365 infnan ? UNICODE_REPLACEMENT : SvIV(argsv);
11367 (!UVCHR_IS_INVARIANT(uv) && SvUTF8(sv)))
11369 eptr = (char*)utf8buf;
11370 elen = uvchr_to_utf8((U8*)eptr, uv) - utf8buf;
11384 eptr = va_arg(*args, char*);
11386 elen = strlen(eptr);
11388 eptr = (char *)nullstr;
11389 elen = sizeof nullstr - 1;
11393 eptr = SvPV_const(argsv, elen);
11394 if (DO_UTF8(argsv)) {
11395 STRLEN old_precis = precis;
11396 if (has_precis && precis < elen) {
11397 STRLEN ulen = sv_or_pv_len_utf8(argsv, eptr, elen);
11398 STRLEN p = precis > ulen ? ulen : precis;
11399 precis = sv_or_pv_pos_u2b(argsv, eptr, p, 0);
11400 /* sticks at end */
11402 if (width) { /* fudge width (can't fudge elen) */
11403 if (has_precis && precis < elen)
11404 width += precis - old_precis;
11407 elen - sv_or_pv_len_utf8(argsv,eptr,elen);
11414 if (has_precis && precis < elen)
11423 goto floating_point;
11425 if (alt || vectorize)
11427 uv = PTR2UV(args ? va_arg(*args, void*) : argsv);
11442 goto floating_point;
11449 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
11458 esignbuf[esignlen++] = plus;
11462 case 'c': iv = (char)va_arg(*args, int); break;
11463 case 'h': iv = (short)va_arg(*args, int); break;
11464 case 'l': iv = va_arg(*args, long); break;
11465 case 'V': iv = va_arg(*args, IV); break;
11466 case 'z': iv = va_arg(*args, SSize_t); break;
11467 #ifdef HAS_PTRDIFF_T
11468 case 't': iv = va_arg(*args, ptrdiff_t); break;
11470 default: iv = va_arg(*args, int); break;
11472 case 'j': iv = va_arg(*args, intmax_t); break;
11476 iv = va_arg(*args, Quad_t); break;
11483 IV tiv = SvIV(argsv); /* work around GCC bug #13488 */
11485 case 'c': iv = (char)tiv; break;
11486 case 'h': iv = (short)tiv; break;
11487 case 'l': iv = (long)tiv; break;
11489 default: iv = tiv; break;
11492 iv = (Quad_t)tiv; break;
11498 if ( !vectorize ) /* we already set uv above */
11503 esignbuf[esignlen++] = plus;
11507 esignbuf[esignlen++] = '-';
11547 goto floating_point;
11555 uv = utf8n_to_uvchr(vecstr, veclen, &ulen,
11566 case 'c': uv = (unsigned char)va_arg(*args, unsigned); break;
11567 case 'h': uv = (unsigned short)va_arg(*args, unsigned); break;
11568 case 'l': uv = va_arg(*args, unsigned long); break;
11569 case 'V': uv = va_arg(*args, UV); break;
11570 case 'z': uv = va_arg(*args, Size_t); break;
11571 #ifdef HAS_PTRDIFF_T
11572 case 't': uv = va_arg(*args, ptrdiff_t); break; /* will sign extend, but there is no uptrdiff_t, so oh well */
11575 case 'j': uv = va_arg(*args, uintmax_t); break;
11577 default: uv = va_arg(*args, unsigned); break;
11580 uv = va_arg(*args, Uquad_t); break;
11587 UV tuv = SvUV(argsv); /* work around GCC bug #13488 */
11589 case 'c': uv = (unsigned char)tuv; break;
11590 case 'h': uv = (unsigned short)tuv; break;
11591 case 'l': uv = (unsigned long)tuv; break;
11593 default: uv = tuv; break;
11596 uv = (Uquad_t)tuv; break;
11605 char *ptr = ebuf + sizeof ebuf;
11606 bool tempalt = uv ? alt : FALSE; /* Vectors can't change alt */
11612 p = (char *)((c == 'X') ? PL_hexdigit + 16 : PL_hexdigit);
11616 } while (uv >>= 4);
11618 esignbuf[esignlen++] = '0';
11619 esignbuf[esignlen++] = c; /* 'x' or 'X' */
11625 *--ptr = '0' + dig;
11626 } while (uv >>= 3);
11627 if (alt && *ptr != '0')
11633 *--ptr = '0' + dig;
11634 } while (uv >>= 1);
11636 esignbuf[esignlen++] = '0';
11637 esignbuf[esignlen++] = c;
11640 default: /* it had better be ten or less */
11643 *--ptr = '0' + dig;
11644 } while (uv /= base);
11647 elen = (ebuf + sizeof ebuf) - ptr;
11651 zeros = precis - elen;
11652 else if (precis == 0 && elen == 1 && *eptr == '0'
11653 && !(base == 8 && alt)) /* "%#.0o" prints "0" */
11656 /* a precision nullifies the 0 flag. */
11663 /* FLOATING POINT */
11668 c = 'f'; /* maybe %F isn't supported here */
11670 case 'e': case 'E':
11672 case 'g': case 'G':
11673 case 'a': case 'A':
11677 /* This is evil, but floating point is even more evil */
11679 /* for SV-style calling, we can only get NV
11680 for C-style calling, we assume %f is double;
11681 for simplicity we allow any of %Lf, %llf, %qf for long double
11685 #if defined(USE_LONG_DOUBLE)
11689 /* [perl #20339] - we should accept and ignore %lf rather than die */
11693 #if defined(USE_LONG_DOUBLE)
11694 intsize = args ? 0 : 'q';
11698 #if defined(HAS_LONG_DOUBLE)
11711 /* now we need (long double) if intsize == 'q', else (double) */
11713 #if LONG_DOUBLESIZE > DOUBLESIZE
11715 va_arg(*args, long double) :
11716 va_arg(*args, double)
11718 va_arg(*args, double)
11723 /* frexp() (or frexpl) has some unspecified behaviour for
11724 * nan/inf/-inf, so let's avoid calling that on non-finites. */
11725 if (isALPHA_FOLD_NE(c, 'e') && Perl_isfinite(nv)) {
11727 (void)Perl_frexp(nv, &i);
11728 if (i == PERL_INT_MIN)
11729 Perl_die(aTHX_ "panic: frexp: %"myNVgf, nv);
11730 /* Do not set hexfp earlier since we want to printf
11731 * Inf/NaN for Inf/NAN, not their hexfp. */
11732 hexfp = isALPHA_FOLD_EQ(c, 'a');
11733 if (UNLIKELY(hexfp)) {
11734 /* This seriously overshoots in most cases, but
11735 * better the undershooting. Firstly, all bytes
11736 * of the NV are not mantissa, some of them are
11737 * exponent. Secondly, for the reasonably common
11738 * long doubles case, the "80-bit extended", two
11739 * or six bytes of the NV are unused. */
11741 (nv < 0) ? 1 : 0 + /* possible unary minus */
11743 1 + /* the very unlikely carry */
11746 2 * NVSIZE + /* 2 hexdigits for each byte */
11748 BIT_DIGITS(NV_MAX_EXP) + /* exponent */
11750 #if LONG_DOUBLEKIND == LONG_DOUBLE_IS_DOUBLEDOUBLE_128_BIT_LITTLE_ENDIAN || \
11751 LONG_DOUBLEKIND == LONG_DOUBLE_IS_DOUBLEDOUBLE_128_BIT_BIG_ENDIAN
11752 /* However, for the "double double", we need more.
11753 * Since each double has their own exponent, the
11754 * doubles may float (haha) rather far from each
11755 * other, and the number of required bits is much
11756 * larger, up to total of 1028 bits. (NOTE: this
11757 * is not actually implemented properly yet,
11758 * we are using just the first double, see
11759 * S_hextract() for details. But let's prepare
11760 * for the future.) */
11762 /* 2 hexdigits for each byte. */
11763 need += (1028/8 - DOUBLESIZE + 1) * 2;
11765 #ifdef USE_LOCALE_NUMERIC
11766 STORE_LC_NUMERIC_SET_TO_NEEDED();
11767 if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC))
11768 need += SvLEN(PL_numeric_radix_sv);
11769 RESTORE_LC_NUMERIC();
11773 need = BIT_DIGITS(i);
11774 } /* if i < 0, the number of digits is hard to predict. */
11776 need += has_precis ? precis : 6; /* known default */
11781 #ifdef HAS_LDBL_SPRINTF_BUG
11782 /* This is to try to fix a bug with irix/nonstop-ux/powerux and
11783 with sfio - Allen <allens@cpan.org> */
11786 # define MY_DBL_MAX DBL_MAX
11787 # else /* XXX guessing! HUGE_VAL may be defined as infinity, so not using */
11788 # if DOUBLESIZE >= 8
11789 # define MY_DBL_MAX 1.7976931348623157E+308L
11791 # define MY_DBL_MAX 3.40282347E+38L
11795 # ifdef HAS_LDBL_SPRINTF_BUG_LESS1 /* only between -1L & 1L - Allen */
11796 # define MY_DBL_MAX_BUG 1L
11798 # define MY_DBL_MAX_BUG MY_DBL_MAX
11802 # define MY_DBL_MIN DBL_MIN
11803 # else /* XXX guessing! -Allen */
11804 # if DOUBLESIZE >= 8
11805 # define MY_DBL_MIN 2.2250738585072014E-308L
11807 # define MY_DBL_MIN 1.17549435E-38L
11811 if ((intsize == 'q') && (c == 'f') &&
11812 ((nv < MY_DBL_MAX_BUG) && (nv > -MY_DBL_MAX_BUG)) &&
11813 (need < DBL_DIG)) {
11814 /* it's going to be short enough that
11815 * long double precision is not needed */
11817 if ((nv <= 0L) && (nv >= -0L))
11818 fix_ldbl_sprintf_bug = TRUE; /* 0 is 0 - easiest */
11820 /* would use Perl_fp_class as a double-check but not
11821 * functional on IRIX - see perl.h comments */
11823 if ((nv >= MY_DBL_MIN) || (nv <= -MY_DBL_MIN)) {
11824 /* It's within the range that a double can represent */
11825 #if defined(DBL_MAX) && !defined(DBL_MIN)
11826 if ((nv >= ((long double)1/DBL_MAX)) ||
11827 (nv <= (-(long double)1/DBL_MAX)))
11829 fix_ldbl_sprintf_bug = TRUE;
11832 if (fix_ldbl_sprintf_bug == TRUE) {
11842 # undef MY_DBL_MAX_BUG
11845 #endif /* HAS_LDBL_SPRINTF_BUG */
11847 need += 20; /* fudge factor */
11848 if (PL_efloatsize < need) {
11849 Safefree(PL_efloatbuf);
11850 PL_efloatsize = need + 20; /* more fudge */
11851 Newx(PL_efloatbuf, PL_efloatsize, char);
11852 PL_efloatbuf[0] = '\0';
11855 if ( !(width || left || plus || alt) && fill != '0'
11856 && has_precis && intsize != 'q' ) { /* Shortcuts */
11857 /* See earlier comment about buggy Gconvert when digits,
11859 if ( c == 'g' && precis) {
11860 STORE_LC_NUMERIC_SET_TO_NEEDED();
11861 PERL_UNUSED_RESULT(Gconvert((NV)nv, (int)precis, 0, PL_efloatbuf));
11862 /* May return an empty string for digits==0 */
11863 if (*PL_efloatbuf) {
11864 elen = strlen(PL_efloatbuf);
11865 goto float_converted;
11867 } else if ( c == 'f' && !precis) {
11868 if ((eptr = F0convert(nv, ebuf + sizeof ebuf, &elen)))
11873 if (UNLIKELY(hexfp)) {
11874 /* Hexadecimal floating point. */
11875 char* p = PL_efloatbuf;
11876 U8 vhex[VHEX_SIZE];
11877 U8* v = vhex; /* working pointer to vhex */
11878 U8* vend; /* pointer to one beyond last digit of vhex */
11879 U8* vfnz = NULL; /* first non-zero */
11880 const bool lower = (c == 'a');
11881 /* At output the values of vhex (up to vend) will
11882 * be mapped through the xdig to get the actual
11883 * human-readable xdigits. */
11884 const char* xdig = PL_hexdigit;
11885 int zerotail = 0; /* how many extra zeros to append */
11886 int exponent = 0; /* exponent of the floating point input */
11888 /* XXX: denormals, NaN, Inf.
11890 * For example with denormals, (assuming the vanilla
11891 * 64-bit double): the exponent is zero. 1xp-1074 is
11892 * the smallest denormal and the smallest double, it
11893 * should be output as 0x0.0000000000001p-1022 to
11894 * match its internal structure. */
11896 vend = S_hextract(aTHX_ nv, &exponent, vhex, NULL);
11897 S_hextract(aTHX_ nv, &exponent, vhex, vend);
11899 #if NVSIZE > DOUBLESIZE && defined(LONG_DOUBLEKIND)
11900 # if LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN || \
11901 LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_BIG_ENDIAN
11918 xdig += 16; /* Use uppercase hex. */
11921 /* Find the first non-zero xdigit. */
11922 for (v = vhex; v < vend; v++) {
11930 U8* vlnz = NULL; /* The last non-zero. */
11932 /* Find the last non-zero xdigit. */
11933 for (v = vend - 1; v >= vhex; v--) {
11940 #if NVSIZE == DOUBLESIZE
11945 v = vhex + precis + 1;
11947 /* Round away from zero: if the tail
11948 * beyond the precis xdigits is equal to
11949 * or greater than 0x8000... */
11950 bool round = *v > 0x8;
11951 if (!round && *v == 0x8) {
11952 for (v++; v < vend; v++) {
11960 for (v = vhex + precis; v >= vhex; v--) {
11967 /* If the carry goes all the way to
11968 * the front, we need to output
11969 * a single '1'. This goes against
11970 * the "xdigit and then radix"
11971 * but since this is "cannot happen"
11972 * category, that is probably good. */
11977 /* The new effective "last non zero". */
11978 vlnz = vhex + precis;
11981 zerotail = precis - (vlnz - vhex);
11988 /* The radix is always output after the first
11989 * non-zero xdigit, or if alt. */
11990 if (vfnz < vlnz || alt) {
11991 #ifndef USE_LOCALE_NUMERIC
11994 STORE_LC_NUMERIC_SET_TO_NEEDED();
11995 if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC)) {
11997 const char* r = SvPV(PL_numeric_radix_sv, n);
11998 Copy(r, p, n, char);
12004 RESTORE_LC_NUMERIC();
12019 elen = p - PL_efloatbuf;
12020 elen += my_snprintf(p, PL_efloatsize - elen,
12021 "%c%+d", lower ? 'p' : 'P',
12024 if (elen < width) {
12026 /* Pad the back with spaces. */
12027 memset(PL_efloatbuf + elen, ' ', width - elen);
12029 else if (fill == '0') {
12030 /* Insert the zeros between the "0x" and
12031 * the digits, otherwise we end up with
12033 STRLEN nzero = width - elen;
12034 char* zerox = PL_efloatbuf + 2;
12035 Move(zerox, zerox + nzero, elen - 2, char);
12036 memset(zerox, fill, nzero);
12039 /* Move it to the right. */
12040 Move(PL_efloatbuf, PL_efloatbuf + width - elen,
12042 /* Pad the front with spaces. */
12043 memset(PL_efloatbuf, ' ', width - elen);
12049 elen = S_infnan_copy(nv, PL_efloatbuf, 5);
12051 char *ptr = ebuf + sizeof ebuf;
12054 /* FIXME: what to do if HAS_LONG_DOUBLE but not PERL_PRIfldbl? */
12055 #if defined(HAS_LONG_DOUBLE) && defined(PERL_PRIfldbl)
12056 /* Note that this is HAS_LONG_DOUBLE and PERL_PRIfldbl,
12057 * not USE_LONG_DOUBLE and NVff. In other words,
12058 * this needs to work without USE_LONG_DOUBLE. */
12059 if (intsize == 'q') {
12060 /* Copy the one or more characters in a long double
12061 * format before the 'base' ([efgEFG]) character to
12062 * the format string. */
12063 static char const ldblf[] = PERL_PRIfldbl;
12064 char const *p = ldblf + sizeof(ldblf) - 3;
12065 while (p >= ldblf) { *--ptr = *p--; }
12070 do { *--ptr = '0' + (base % 10); } while (base /= 10);
12075 do { *--ptr = '0' + (base % 10); } while (base /= 10);
12087 /* No taint. Otherwise we are in the strange situation
12088 * where printf() taints but print($float) doesn't.
12091 STORE_LC_NUMERIC_SET_TO_NEEDED();
12093 /* hopefully the above makes ptr a very constrained format
12094 * that is safe to use, even though it's not literal */
12095 GCC_DIAG_IGNORE(-Wformat-nonliteral);
12096 #if defined(HAS_LONG_DOUBLE)
12097 elen = ((intsize == 'q')
12098 ? my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, nv)
12099 : my_snprintf(PL_efloatbuf, PL_efloatsize, ptr, (double)nv));
12101 elen = my_sprintf(PL_efloatbuf, ptr, nv);
12107 eptr = PL_efloatbuf;
12109 #ifdef USE_LOCALE_NUMERIC
12110 /* If the decimal point character in the string is UTF-8, make the
12112 if (PL_numeric_radix_sv && SvUTF8(PL_numeric_radix_sv)
12113 && instr(eptr, SvPVX_const(PL_numeric_radix_sv)))
12126 i = SvCUR(sv) - origlen;
12129 case 'c': *(va_arg(*args, char*)) = i; break;
12130 case 'h': *(va_arg(*args, short*)) = i; break;
12131 default: *(va_arg(*args, int*)) = i; break;
12132 case 'l': *(va_arg(*args, long*)) = i; break;
12133 case 'V': *(va_arg(*args, IV*)) = i; break;
12134 case 'z': *(va_arg(*args, SSize_t*)) = i; break;
12135 #ifdef HAS_PTRDIFF_T
12136 case 't': *(va_arg(*args, ptrdiff_t*)) = i; break;
12139 case 'j': *(va_arg(*args, intmax_t*)) = i; break;
12143 *(va_arg(*args, Quad_t*)) = i; break;
12150 sv_setuv_mg(argsv, has_utf8 ? (UV)sv_len_utf8(sv) : (UV)i);
12151 continue; /* not "break" */
12158 && (PL_op->op_type == OP_PRTF || PL_op->op_type == OP_SPRINTF)
12159 && ckWARN(WARN_PRINTF))
12161 SV * const msg = sv_newmortal();
12162 Perl_sv_setpvf(aTHX_ msg, "Invalid conversion in %sprintf: ",
12163 (PL_op->op_type == OP_PRTF) ? "" : "s");
12164 if (fmtstart < patend) {
12165 const char * const fmtend = q < patend ? q : patend;
12167 sv_catpvs(msg, "\"%");
12168 for (f = fmtstart; f < fmtend; f++) {
12170 sv_catpvn_nomg(msg, f, 1);
12172 Perl_sv_catpvf(aTHX_ msg,
12173 "\\%03"UVof, (UV)*f & 0xFF);
12176 sv_catpvs(msg, "\"");
12178 sv_catpvs(msg, "end of string");
12180 Perl_warner(aTHX_ packWARN(WARN_PRINTF), "%"SVf, SVfARG(msg)); /* yes, this is reentrant */
12183 /* output mangled stuff ... */
12189 /* ... right here, because formatting flags should not apply */
12190 SvGROW(sv, SvCUR(sv) + elen + 1);
12192 Copy(eptr, p, elen, char);
12195 SvCUR_set(sv, p - SvPVX_const(sv));
12197 continue; /* not "break" */
12200 if (is_utf8 != has_utf8) {
12203 sv_utf8_upgrade(sv);
12206 const STRLEN old_elen = elen;
12207 SV * const nsv = newSVpvn_flags(eptr, elen, SVs_TEMP);
12208 sv_utf8_upgrade(nsv);
12209 eptr = SvPVX_const(nsv);
12212 if (width) { /* fudge width (can't fudge elen) */
12213 width += elen - old_elen;
12219 have = esignlen + zeros + elen;
12221 croak_memory_wrap();
12223 need = (have > width ? have : width);
12226 if (need >= (((STRLEN)~0) - SvCUR(sv) - dotstrlen - 1))
12227 croak_memory_wrap();
12228 SvGROW(sv, SvCUR(sv) + need + dotstrlen + 1);
12230 if (esignlen && fill == '0') {
12232 for (i = 0; i < (int)esignlen; i++)
12233 *p++ = esignbuf[i];
12235 if (gap && !left) {
12236 memset(p, fill, gap);
12239 if (esignlen && fill != '0') {
12241 for (i = 0; i < (int)esignlen; i++)
12242 *p++ = esignbuf[i];
12246 for (i = zeros; i; i--)
12250 Copy(eptr, p, elen, char);
12254 memset(p, ' ', gap);
12259 Copy(dotstr, p, dotstrlen, char);
12263 vectorize = FALSE; /* done iterating over vecstr */
12270 SvCUR_set(sv, p - SvPVX_const(sv));
12277 /* Now that we've consumed all our printf format arguments (svix)
12278 * do we have things left on the stack that we didn't use?
12280 if (!no_redundant_warning && svmax >= svix + 1 && ckWARN(WARN_REDUNDANT)) {
12281 Perl_warner(aTHX_ packWARN(WARN_REDUNDANT), "Redundant argument in %s",
12282 PL_op ? OP_DESC(PL_op) : "sv_vcatpvfn()");
12287 RESTORE_LC_NUMERIC(); /* Done outside loop, so don't have to save/restore
12291 /* =========================================================================
12293 =head1 Cloning an interpreter
12297 All the macros and functions in this section are for the private use of
12298 the main function, perl_clone().
12300 The foo_dup() functions make an exact copy of an existing foo thingy.
12301 During the course of a cloning, a hash table is used to map old addresses
12302 to new addresses. The table is created and manipulated with the
12303 ptr_table_* functions.
12305 * =========================================================================*/
12308 #if defined(USE_ITHREADS)
12310 /* XXX Remove this so it doesn't have to go thru the macro and return for nothing */
12311 #ifndef GpREFCNT_inc
12312 # define GpREFCNT_inc(gp) ((gp) ? (++(gp)->gp_refcnt, (gp)) : (GP*)NULL)
12316 /* Certain cases in Perl_ss_dup have been merged, by relying on the fact
12317 that currently av_dup, gv_dup and hv_dup are the same as sv_dup.
12318 If this changes, please unmerge ss_dup.
12319 Likewise, sv_dup_inc_multiple() relies on this fact. */
12320 #define sv_dup_inc_NN(s,t) SvREFCNT_inc_NN(sv_dup_inc(s,t))
12321 #define av_dup(s,t) MUTABLE_AV(sv_dup((const SV *)s,t))
12322 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
12323 #define hv_dup(s,t) MUTABLE_HV(sv_dup((const SV *)s,t))
12324 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
12325 #define cv_dup(s,t) MUTABLE_CV(sv_dup((const SV *)s,t))
12326 #define cv_dup_inc(s,t) MUTABLE_CV(sv_dup_inc((const SV *)s,t))
12327 #define io_dup(s,t) MUTABLE_IO(sv_dup((const SV *)s,t))
12328 #define io_dup_inc(s,t) MUTABLE_IO(sv_dup_inc((const SV *)s,t))
12329 #define gv_dup(s,t) MUTABLE_GV(sv_dup((const SV *)s,t))
12330 #define gv_dup_inc(s,t) MUTABLE_GV(sv_dup_inc((const SV *)s,t))
12331 #define SAVEPV(p) ((p) ? savepv(p) : NULL)
12332 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
12334 /* clone a parser */
12337 Perl_parser_dup(pTHX_ const yy_parser *const proto, CLONE_PARAMS *const param)
12341 PERL_ARGS_ASSERT_PARSER_DUP;
12346 /* look for it in the table first */
12347 parser = (yy_parser *)ptr_table_fetch(PL_ptr_table, proto);
12351 /* create anew and remember what it is */
12352 Newxz(parser, 1, yy_parser);
12353 ptr_table_store(PL_ptr_table, proto, parser);
12355 /* XXX these not yet duped */
12356 parser->old_parser = NULL;
12357 parser->stack = NULL;
12359 parser->stack_size = 0;
12360 /* XXX parser->stack->state = 0; */
12362 /* XXX eventually, just Copy() most of the parser struct ? */
12364 parser->lex_brackets = proto->lex_brackets;
12365 parser->lex_casemods = proto->lex_casemods;
12366 parser->lex_brackstack = savepvn(proto->lex_brackstack,
12367 (proto->lex_brackets < 120 ? 120 : proto->lex_brackets));
12368 parser->lex_casestack = savepvn(proto->lex_casestack,
12369 (proto->lex_casemods < 12 ? 12 : proto->lex_casemods));
12370 parser->lex_defer = proto->lex_defer;
12371 parser->lex_dojoin = proto->lex_dojoin;
12372 parser->lex_formbrack = proto->lex_formbrack;
12373 parser->lex_inpat = proto->lex_inpat;
12374 parser->lex_inwhat = proto->lex_inwhat;
12375 parser->lex_op = proto->lex_op;
12376 parser->lex_repl = sv_dup_inc(proto->lex_repl, param);
12377 parser->lex_starts = proto->lex_starts;
12378 parser->lex_stuff = sv_dup_inc(proto->lex_stuff, param);
12379 parser->multi_close = proto->multi_close;
12380 parser->multi_open = proto->multi_open;
12381 parser->multi_start = proto->multi_start;
12382 parser->multi_end = proto->multi_end;
12383 parser->preambled = proto->preambled;
12384 parser->sublex_info = proto->sublex_info; /* XXX not quite right */
12385 parser->linestr = sv_dup_inc(proto->linestr, param);
12386 parser->expect = proto->expect;
12387 parser->copline = proto->copline;
12388 parser->last_lop_op = proto->last_lop_op;
12389 parser->lex_state = proto->lex_state;
12390 parser->rsfp = fp_dup(proto->rsfp, '<', param);
12391 /* rsfp_filters entries have fake IoDIRP() */
12392 parser->rsfp_filters= av_dup_inc(proto->rsfp_filters, param);
12393 parser->in_my = proto->in_my;
12394 parser->in_my_stash = hv_dup(proto->in_my_stash, param);
12395 parser->error_count = proto->error_count;
12398 parser->linestr = sv_dup_inc(proto->linestr, param);
12401 char * const ols = SvPVX(proto->linestr);
12402 char * const ls = SvPVX(parser->linestr);
12404 parser->bufptr = ls + (proto->bufptr >= ols ?
12405 proto->bufptr - ols : 0);
12406 parser->oldbufptr = ls + (proto->oldbufptr >= ols ?
12407 proto->oldbufptr - ols : 0);
12408 parser->oldoldbufptr= ls + (proto->oldoldbufptr >= ols ?
12409 proto->oldoldbufptr - ols : 0);
12410 parser->linestart = ls + (proto->linestart >= ols ?
12411 proto->linestart - ols : 0);
12412 parser->last_uni = ls + (proto->last_uni >= ols ?
12413 proto->last_uni - ols : 0);
12414 parser->last_lop = ls + (proto->last_lop >= ols ?
12415 proto->last_lop - ols : 0);
12417 parser->bufend = ls + SvCUR(parser->linestr);
12420 Copy(proto->tokenbuf, parser->tokenbuf, 256, char);
12423 Copy(proto->nextval, parser->nextval, 5, YYSTYPE);
12424 Copy(proto->nexttype, parser->nexttype, 5, I32);
12425 parser->nexttoke = proto->nexttoke;
12427 /* XXX should clone saved_curcop here, but we aren't passed
12428 * proto_perl; so do it in perl_clone_using instead */
12434 /* duplicate a file handle */
12437 Perl_fp_dup(pTHX_ PerlIO *const fp, const char type, CLONE_PARAMS *const param)
12441 PERL_ARGS_ASSERT_FP_DUP;
12442 PERL_UNUSED_ARG(type);
12445 return (PerlIO*)NULL;
12447 /* look for it in the table first */
12448 ret = (PerlIO*)ptr_table_fetch(PL_ptr_table, fp);
12452 /* create anew and remember what it is */
12453 ret = PerlIO_fdupopen(aTHX_ fp, param, PERLIO_DUP_CLONE);
12454 ptr_table_store(PL_ptr_table, fp, ret);
12458 /* duplicate a directory handle */
12461 Perl_dirp_dup(pTHX_ DIR *const dp, CLONE_PARAMS *const param)
12465 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
12467 const Direntry_t *dirent;
12468 char smallbuf[256];
12474 PERL_UNUSED_CONTEXT;
12475 PERL_ARGS_ASSERT_DIRP_DUP;
12480 /* look for it in the table first */
12481 ret = (DIR*)ptr_table_fetch(PL_ptr_table, dp);
12485 #if defined(HAS_FCHDIR) && defined(HAS_TELLDIR) && defined(HAS_SEEKDIR)
12487 PERL_UNUSED_ARG(param);
12491 /* open the current directory (so we can switch back) */
12492 if (!(pwd = PerlDir_open("."))) return (DIR *)NULL;
12494 /* chdir to our dir handle and open the present working directory */
12495 if (fchdir(my_dirfd(dp)) < 0 || !(ret = PerlDir_open("."))) {
12496 PerlDir_close(pwd);
12497 return (DIR *)NULL;
12499 /* Now we should have two dir handles pointing to the same dir. */
12501 /* Be nice to the calling code and chdir back to where we were. */
12502 /* XXX If this fails, then what? */
12503 PERL_UNUSED_RESULT(fchdir(my_dirfd(pwd)));
12505 /* We have no need of the pwd handle any more. */
12506 PerlDir_close(pwd);
12509 # define d_namlen(d) (d)->d_namlen
12511 # define d_namlen(d) strlen((d)->d_name)
12513 /* Iterate once through dp, to get the file name at the current posi-
12514 tion. Then step back. */
12515 pos = PerlDir_tell(dp);
12516 if ((dirent = PerlDir_read(dp))) {
12517 len = d_namlen(dirent);
12518 if (len <= sizeof smallbuf) name = smallbuf;
12519 else Newx(name, len, char);
12520 Move(dirent->d_name, name, len, char);
12522 PerlDir_seek(dp, pos);
12524 /* Iterate through the new dir handle, till we find a file with the
12526 if (!dirent) /* just before the end */
12528 pos = PerlDir_tell(ret);
12529 if (PerlDir_read(ret)) continue; /* not there yet */
12530 PerlDir_seek(ret, pos); /* step back */
12534 const long pos0 = PerlDir_tell(ret);
12536 pos = PerlDir_tell(ret);
12537 if ((dirent = PerlDir_read(ret))) {
12538 if (len == (STRLEN)d_namlen(dirent)
12539 && memEQ(name, dirent->d_name, len)) {
12541 PerlDir_seek(ret, pos); /* step back */
12544 /* else we are not there yet; keep iterating */
12546 else { /* This is not meant to happen. The best we can do is
12547 reset the iterator to the beginning. */
12548 PerlDir_seek(ret, pos0);
12555 if (name && name != smallbuf)
12560 ret = win32_dirp_dup(dp, param);
12563 /* pop it in the pointer table */
12565 ptr_table_store(PL_ptr_table, dp, ret);
12570 /* duplicate a typeglob */
12573 Perl_gp_dup(pTHX_ GP *const gp, CLONE_PARAMS *const param)
12577 PERL_ARGS_ASSERT_GP_DUP;
12581 /* look for it in the table first */
12582 ret = (GP*)ptr_table_fetch(PL_ptr_table, gp);
12586 /* create anew and remember what it is */
12588 ptr_table_store(PL_ptr_table, gp, ret);
12591 /* ret->gp_refcnt must be 0 before any other dups are called. We're relying
12592 on Newxz() to do this for us. */
12593 ret->gp_sv = sv_dup_inc(gp->gp_sv, param);
12594 ret->gp_io = io_dup_inc(gp->gp_io, param);
12595 ret->gp_form = cv_dup_inc(gp->gp_form, param);
12596 ret->gp_av = av_dup_inc(gp->gp_av, param);
12597 ret->gp_hv = hv_dup_inc(gp->gp_hv, param);
12598 ret->gp_egv = gv_dup(gp->gp_egv, param);/* GvEGV is not refcounted */
12599 ret->gp_cv = cv_dup_inc(gp->gp_cv, param);
12600 ret->gp_cvgen = gp->gp_cvgen;
12601 ret->gp_line = gp->gp_line;
12602 ret->gp_file_hek = hek_dup(gp->gp_file_hek, param);
12606 /* duplicate a chain of magic */
12609 Perl_mg_dup(pTHX_ MAGIC *mg, CLONE_PARAMS *const param)
12611 MAGIC *mgret = NULL;
12612 MAGIC **mgprev_p = &mgret;
12614 PERL_ARGS_ASSERT_MG_DUP;
12616 for (; mg; mg = mg->mg_moremagic) {
12619 if ((param->flags & CLONEf_JOIN_IN)
12620 && mg->mg_type == PERL_MAGIC_backref)
12621 /* when joining, we let the individual SVs add themselves to
12622 * backref as needed. */
12625 Newx(nmg, 1, MAGIC);
12627 mgprev_p = &(nmg->mg_moremagic);
12629 /* There was a comment "XXX copy dynamic vtable?" but as we don't have
12630 dynamic vtables, I'm not sure why Sarathy wrote it. The comment dates
12631 from the original commit adding Perl_mg_dup() - revision 4538.
12632 Similarly there is the annotation "XXX random ptr?" next to the
12633 assignment to nmg->mg_ptr. */
12636 /* FIXME for plugins
12637 if (nmg->mg_type == PERL_MAGIC_qr) {
12638 nmg->mg_obj = MUTABLE_SV(CALLREGDUPE((REGEXP*)nmg->mg_obj, param));
12642 nmg->mg_obj = (nmg->mg_flags & MGf_REFCOUNTED)
12643 ? nmg->mg_type == PERL_MAGIC_backref
12644 /* The backref AV has its reference
12645 * count deliberately bumped by 1 */
12646 ? SvREFCNT_inc(av_dup_inc((const AV *)
12647 nmg->mg_obj, param))
12648 : sv_dup_inc(nmg->mg_obj, param)
12649 : sv_dup(nmg->mg_obj, param);
12651 if (nmg->mg_ptr && nmg->mg_type != PERL_MAGIC_regex_global) {
12652 if (nmg->mg_len > 0) {
12653 nmg->mg_ptr = SAVEPVN(nmg->mg_ptr, nmg->mg_len);
12654 if (nmg->mg_type == PERL_MAGIC_overload_table &&
12655 AMT_AMAGIC((AMT*)nmg->mg_ptr))
12657 AMT * const namtp = (AMT*)nmg->mg_ptr;
12658 sv_dup_inc_multiple((SV**)(namtp->table),
12659 (SV**)(namtp->table), NofAMmeth, param);
12662 else if (nmg->mg_len == HEf_SVKEY)
12663 nmg->mg_ptr = (char*)sv_dup_inc((const SV *)nmg->mg_ptr, param);
12665 if ((nmg->mg_flags & MGf_DUP) && nmg->mg_virtual && nmg->mg_virtual->svt_dup) {
12666 nmg->mg_virtual->svt_dup(aTHX_ nmg, param);
12672 #endif /* USE_ITHREADS */
12674 struct ptr_tbl_arena {
12675 struct ptr_tbl_arena *next;
12676 struct ptr_tbl_ent array[1023/3]; /* as ptr_tbl_ent has 3 pointers. */
12679 /* create a new pointer-mapping table */
12682 Perl_ptr_table_new(pTHX)
12685 PERL_UNUSED_CONTEXT;
12687 Newx(tbl, 1, PTR_TBL_t);
12688 tbl->tbl_max = 511;
12689 tbl->tbl_items = 0;
12690 tbl->tbl_arena = NULL;
12691 tbl->tbl_arena_next = NULL;
12692 tbl->tbl_arena_end = NULL;
12693 Newxz(tbl->tbl_ary, tbl->tbl_max + 1, PTR_TBL_ENT_t*);
12697 #define PTR_TABLE_HASH(ptr) \
12698 ((PTR2UV(ptr) >> 3) ^ (PTR2UV(ptr) >> (3 + 7)) ^ (PTR2UV(ptr) >> (3 + 17)))
12700 /* map an existing pointer using a table */
12702 STATIC PTR_TBL_ENT_t *
12703 S_ptr_table_find(PTR_TBL_t *const tbl, const void *const sv)
12705 PTR_TBL_ENT_t *tblent;
12706 const UV hash = PTR_TABLE_HASH(sv);
12708 PERL_ARGS_ASSERT_PTR_TABLE_FIND;
12710 tblent = tbl->tbl_ary[hash & tbl->tbl_max];
12711 for (; tblent; tblent = tblent->next) {
12712 if (tblent->oldval == sv)
12719 Perl_ptr_table_fetch(pTHX_ PTR_TBL_t *const tbl, const void *const sv)
12721 PTR_TBL_ENT_t const *const tblent = ptr_table_find(tbl, sv);
12723 PERL_ARGS_ASSERT_PTR_TABLE_FETCH;
12724 PERL_UNUSED_CONTEXT;
12726 return tblent ? tblent->newval : NULL;
12729 /* add a new entry to a pointer-mapping table 'tbl'. In hash terms, 'oldsv' is
12730 * the key; 'newsv' is the value. The names "old" and "new" are specific to
12731 * the core's typical use of ptr_tables in thread cloning. */
12734 Perl_ptr_table_store(pTHX_ PTR_TBL_t *const tbl, const void *const oldsv, void *const newsv)
12736 PTR_TBL_ENT_t *tblent = ptr_table_find(tbl, oldsv);
12738 PERL_ARGS_ASSERT_PTR_TABLE_STORE;
12739 PERL_UNUSED_CONTEXT;
12742 tblent->newval = newsv;
12744 const UV entry = PTR_TABLE_HASH(oldsv) & tbl->tbl_max;
12746 if (tbl->tbl_arena_next == tbl->tbl_arena_end) {
12747 struct ptr_tbl_arena *new_arena;
12749 Newx(new_arena, 1, struct ptr_tbl_arena);
12750 new_arena->next = tbl->tbl_arena;
12751 tbl->tbl_arena = new_arena;
12752 tbl->tbl_arena_next = new_arena->array;
12753 tbl->tbl_arena_end = C_ARRAY_END(new_arena->array);
12756 tblent = tbl->tbl_arena_next++;
12758 tblent->oldval = oldsv;
12759 tblent->newval = newsv;
12760 tblent->next = tbl->tbl_ary[entry];
12761 tbl->tbl_ary[entry] = tblent;
12763 if (tblent->next && tbl->tbl_items > tbl->tbl_max)
12764 ptr_table_split(tbl);
12768 /* double the hash bucket size of an existing ptr table */
12771 Perl_ptr_table_split(pTHX_ PTR_TBL_t *const tbl)
12773 PTR_TBL_ENT_t **ary = tbl->tbl_ary;
12774 const UV oldsize = tbl->tbl_max + 1;
12775 UV newsize = oldsize * 2;
12778 PERL_ARGS_ASSERT_PTR_TABLE_SPLIT;
12779 PERL_UNUSED_CONTEXT;
12781 Renew(ary, newsize, PTR_TBL_ENT_t*);
12782 Zero(&ary[oldsize], newsize-oldsize, PTR_TBL_ENT_t*);
12783 tbl->tbl_max = --newsize;
12784 tbl->tbl_ary = ary;
12785 for (i=0; i < oldsize; i++, ary++) {
12786 PTR_TBL_ENT_t **entp = ary;
12787 PTR_TBL_ENT_t *ent = *ary;
12788 PTR_TBL_ENT_t **curentp;
12791 curentp = ary + oldsize;
12793 if ((newsize & PTR_TABLE_HASH(ent->oldval)) != i) {
12795 ent->next = *curentp;
12805 /* remove all the entries from a ptr table */
12806 /* Deprecated - will be removed post 5.14 */
12809 Perl_ptr_table_clear(pTHX_ PTR_TBL_t *const tbl)
12811 PERL_UNUSED_CONTEXT;
12812 if (tbl && tbl->tbl_items) {
12813 struct ptr_tbl_arena *arena = tbl->tbl_arena;
12815 Zero(tbl->tbl_ary, tbl->tbl_max + 1, struct ptr_tbl_ent **);
12818 struct ptr_tbl_arena *next = arena->next;
12824 tbl->tbl_items = 0;
12825 tbl->tbl_arena = NULL;
12826 tbl->tbl_arena_next = NULL;
12827 tbl->tbl_arena_end = NULL;
12831 /* clear and free a ptr table */
12834 Perl_ptr_table_free(pTHX_ PTR_TBL_t *const tbl)
12836 struct ptr_tbl_arena *arena;
12838 PERL_UNUSED_CONTEXT;
12844 arena = tbl->tbl_arena;
12847 struct ptr_tbl_arena *next = arena->next;
12853 Safefree(tbl->tbl_ary);
12857 #if defined(USE_ITHREADS)
12860 Perl_rvpv_dup(pTHX_ SV *const dstr, const SV *const sstr, CLONE_PARAMS *const param)
12862 PERL_ARGS_ASSERT_RVPV_DUP;
12864 assert(!isREGEXP(sstr));
12866 if (SvWEAKREF(sstr)) {
12867 SvRV_set(dstr, sv_dup(SvRV_const(sstr), param));
12868 if (param->flags & CLONEf_JOIN_IN) {
12869 /* if joining, we add any back references individually rather
12870 * than copying the whole backref array */
12871 Perl_sv_add_backref(aTHX_ SvRV(dstr), dstr);
12875 SvRV_set(dstr, sv_dup_inc(SvRV_const(sstr), param));
12877 else if (SvPVX_const(sstr)) {
12878 /* Has something there */
12880 /* Normal PV - clone whole allocated space */
12881 SvPV_set(dstr, SAVEPVN(SvPVX_const(sstr), SvLEN(sstr)-1));
12882 /* sstr may not be that normal, but actually copy on write.
12883 But we are a true, independent SV, so: */
12887 /* Special case - not normally malloced for some reason */
12888 if (isGV_with_GP(sstr)) {
12889 /* Don't need to do anything here. */
12891 else if ((SvIsCOW(sstr))) {
12892 /* A "shared" PV - clone it as "shared" PV */
12894 HEK_KEY(hek_dup(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)),
12898 /* Some other special case - random pointer */
12899 SvPV_set(dstr, (char *) SvPVX_const(sstr));
12904 /* Copy the NULL */
12905 SvPV_set(dstr, NULL);
12909 /* duplicate a list of SVs. source and dest may point to the same memory. */
12911 S_sv_dup_inc_multiple(pTHX_ SV *const *source, SV **dest,
12912 SSize_t items, CLONE_PARAMS *const param)
12914 PERL_ARGS_ASSERT_SV_DUP_INC_MULTIPLE;
12916 while (items-- > 0) {
12917 *dest++ = sv_dup_inc(*source++, param);
12923 /* duplicate an SV of any type (including AV, HV etc) */
12926 S_sv_dup_common(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
12931 PERL_ARGS_ASSERT_SV_DUP_COMMON;
12933 if (SvTYPE(sstr) == (svtype)SVTYPEMASK) {
12934 #ifdef DEBUG_LEAKING_SCALARS_ABORT
12939 /* look for it in the table first */
12940 dstr = MUTABLE_SV(ptr_table_fetch(PL_ptr_table, sstr));
12944 if(param->flags & CLONEf_JOIN_IN) {
12945 /** We are joining here so we don't want do clone
12946 something that is bad **/
12947 if (SvTYPE(sstr) == SVt_PVHV) {
12948 const HEK * const hvname = HvNAME_HEK(sstr);
12950 /** don't clone stashes if they already exist **/
12951 dstr = MUTABLE_SV(gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
12952 HEK_UTF8(hvname) ? SVf_UTF8 : 0));
12953 ptr_table_store(PL_ptr_table, sstr, dstr);
12957 else if (SvTYPE(sstr) == SVt_PVGV && !SvFAKE(sstr)) {
12958 HV *stash = GvSTASH(sstr);
12959 const HEK * hvname;
12960 if (stash && (hvname = HvNAME_HEK(stash))) {
12961 /** don't clone GVs if they already exist **/
12963 stash = gv_stashpvn(HEK_KEY(hvname), HEK_LEN(hvname),
12964 HEK_UTF8(hvname) ? SVf_UTF8 : 0);
12966 stash, GvNAME(sstr),
12972 if (svp && *svp && SvTYPE(*svp) == SVt_PVGV) {
12973 ptr_table_store(PL_ptr_table, sstr, *svp);
12980 /* create anew and remember what it is */
12983 #ifdef DEBUG_LEAKING_SCALARS
12984 dstr->sv_debug_optype = sstr->sv_debug_optype;
12985 dstr->sv_debug_line = sstr->sv_debug_line;
12986 dstr->sv_debug_inpad = sstr->sv_debug_inpad;
12987 dstr->sv_debug_parent = (SV*)sstr;
12988 FREE_SV_DEBUG_FILE(dstr);
12989 dstr->sv_debug_file = savesharedpv(sstr->sv_debug_file);
12992 ptr_table_store(PL_ptr_table, sstr, dstr);
12995 SvFLAGS(dstr) = SvFLAGS(sstr);
12996 SvFLAGS(dstr) &= ~SVf_OOK; /* don't propagate OOK hack */
12997 SvREFCNT(dstr) = 0; /* must be before any other dups! */
13000 if (SvANY(sstr) && PL_watch_pvx && SvPVX_const(sstr) == PL_watch_pvx)
13001 PerlIO_printf(Perl_debug_log, "watch at %p hit, found string \"%s\"\n",
13002 (void*)PL_watch_pvx, SvPVX_const(sstr));
13005 /* don't clone objects whose class has asked us not to */
13006 if (SvOBJECT(sstr) && ! (SvFLAGS(SvSTASH(sstr)) & SVphv_CLONEABLE)) {
13011 switch (SvTYPE(sstr)) {
13013 SvANY(dstr) = NULL;
13016 SvANY(dstr) = (XPVIV*)((char*)&(dstr->sv_u.svu_iv) - STRUCT_OFFSET(XPVIV, xiv_iv));
13018 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
13020 SvIV_set(dstr, SvIVX(sstr));
13024 SvANY(dstr) = new_XNV();
13025 SvNV_set(dstr, SvNVX(sstr));
13029 /* These are all the types that need complex bodies allocating. */
13031 const svtype sv_type = SvTYPE(sstr);
13032 const struct body_details *const sv_type_details
13033 = bodies_by_type + sv_type;
13037 Perl_croak(aTHX_ "Bizarre SvTYPE [%" IVdf "]", (IV)SvTYPE(sstr));
13053 assert(sv_type_details->body_size);
13054 if (sv_type_details->arena) {
13055 new_body_inline(new_body, sv_type);
13057 = (void*)((char*)new_body - sv_type_details->offset);
13059 new_body = new_NOARENA(sv_type_details);
13063 SvANY(dstr) = new_body;
13066 Copy(((char*)SvANY(sstr)) + sv_type_details->offset,
13067 ((char*)SvANY(dstr)) + sv_type_details->offset,
13068 sv_type_details->copy, char);
13070 Copy(((char*)SvANY(sstr)),
13071 ((char*)SvANY(dstr)),
13072 sv_type_details->body_size + sv_type_details->offset, char);
13075 if (sv_type != SVt_PVAV && sv_type != SVt_PVHV
13076 && !isGV_with_GP(dstr)
13078 && !(sv_type == SVt_PVIO && !(IoFLAGS(dstr) & IOf_FAKE_DIRP)))
13079 Perl_rvpv_dup(aTHX_ dstr, sstr, param);
13081 /* The Copy above means that all the source (unduplicated) pointers
13082 are now in the destination. We can check the flags and the
13083 pointers in either, but it's possible that there's less cache
13084 missing by always going for the destination.
13085 FIXME - instrument and check that assumption */
13086 if (sv_type >= SVt_PVMG) {
13087 if ((sv_type == SVt_PVMG) && SvPAD_OUR(dstr)) {
13088 SvOURSTASH_set(dstr, hv_dup_inc(SvOURSTASH(dstr), param));
13089 } else if (sv_type == SVt_PVAV && AvPAD_NAMELIST(dstr)) {
13091 } else if (SvMAGIC(dstr))
13092 SvMAGIC_set(dstr, mg_dup(SvMAGIC(dstr), param));
13093 if (SvOBJECT(dstr) && SvSTASH(dstr))
13094 SvSTASH_set(dstr, hv_dup_inc(SvSTASH(dstr), param));
13095 else SvSTASH_set(dstr, 0); /* don't copy DESTROY cache */
13098 /* The cast silences a GCC warning about unhandled types. */
13099 switch ((int)sv_type) {
13110 /* FIXME for plugins */
13111 dstr->sv_u.svu_rx = ((REGEXP *)dstr)->sv_any;
13112 re_dup_guts((REGEXP*) sstr, (REGEXP*) dstr, param);
13115 /* XXX LvTARGOFF sometimes holds PMOP* when DEBUGGING */
13116 if (LvTYPE(dstr) == 't') /* for tie: unrefcnted fake (SV**) */
13117 LvTARG(dstr) = dstr;
13118 else if (LvTYPE(dstr) == 'T') /* for tie: fake HE */
13119 LvTARG(dstr) = MUTABLE_SV(he_dup((HE*)LvTARG(dstr), 0, param));
13121 LvTARG(dstr) = sv_dup_inc(LvTARG(dstr), param);
13122 if (isREGEXP(sstr)) goto duprex;
13124 /* non-GP case already handled above */
13125 if(isGV_with_GP(sstr)) {
13126 GvNAME_HEK(dstr) = hek_dup(GvNAME_HEK(dstr), param);
13127 /* Don't call sv_add_backref here as it's going to be
13128 created as part of the magic cloning of the symbol
13129 table--unless this is during a join and the stash
13130 is not actually being cloned. */
13131 /* Danger Will Robinson - GvGP(dstr) isn't initialised
13132 at the point of this comment. */
13133 GvSTASH(dstr) = hv_dup(GvSTASH(dstr), param);
13134 if (param->flags & CLONEf_JOIN_IN)
13135 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
13136 GvGP_set(dstr, gp_dup(GvGP(sstr), param));
13137 (void)GpREFCNT_inc(GvGP(dstr));
13141 /* PL_parser->rsfp_filters entries have fake IoDIRP() */
13142 if(IoFLAGS(dstr) & IOf_FAKE_DIRP) {
13143 /* I have no idea why fake dirp (rsfps)
13144 should be treated differently but otherwise
13145 we end up with leaks -- sky*/
13146 IoTOP_GV(dstr) = gv_dup_inc(IoTOP_GV(dstr), param);
13147 IoFMT_GV(dstr) = gv_dup_inc(IoFMT_GV(dstr), param);
13148 IoBOTTOM_GV(dstr) = gv_dup_inc(IoBOTTOM_GV(dstr), param);
13150 IoTOP_GV(dstr) = gv_dup(IoTOP_GV(dstr), param);
13151 IoFMT_GV(dstr) = gv_dup(IoFMT_GV(dstr), param);
13152 IoBOTTOM_GV(dstr) = gv_dup(IoBOTTOM_GV(dstr), param);
13153 if (IoDIRP(dstr)) {
13154 IoDIRP(dstr) = dirp_dup(IoDIRP(dstr), param);
13157 /* IoDIRP(dstr) is already a copy of IoDIRP(sstr) */
13159 IoIFP(dstr) = fp_dup(IoIFP(sstr), IoTYPE(dstr), param);
13161 if (IoOFP(dstr) == IoIFP(sstr))
13162 IoOFP(dstr) = IoIFP(dstr);
13164 IoOFP(dstr) = fp_dup(IoOFP(dstr), IoTYPE(dstr), param);
13165 IoTOP_NAME(dstr) = SAVEPV(IoTOP_NAME(dstr));
13166 IoFMT_NAME(dstr) = SAVEPV(IoFMT_NAME(dstr));
13167 IoBOTTOM_NAME(dstr) = SAVEPV(IoBOTTOM_NAME(dstr));
13170 /* avoid cloning an empty array */
13171 if (AvARRAY((const AV *)sstr) && AvFILLp((const AV *)sstr) >= 0) {
13172 SV **dst_ary, **src_ary;
13173 SSize_t items = AvFILLp((const AV *)sstr) + 1;
13175 src_ary = AvARRAY((const AV *)sstr);
13176 Newxz(dst_ary, AvMAX((const AV *)sstr)+1, SV*);
13177 ptr_table_store(PL_ptr_table, src_ary, dst_ary);
13178 AvARRAY(MUTABLE_AV(dstr)) = dst_ary;
13179 AvALLOC((const AV *)dstr) = dst_ary;
13180 if (AvREAL((const AV *)sstr)) {
13181 dst_ary = sv_dup_inc_multiple(src_ary, dst_ary, items,
13185 while (items-- > 0)
13186 *dst_ary++ = sv_dup(*src_ary++, param);
13188 items = AvMAX((const AV *)sstr) - AvFILLp((const AV *)sstr);
13189 while (items-- > 0) {
13190 *dst_ary++ = &PL_sv_undef;
13194 AvARRAY(MUTABLE_AV(dstr)) = NULL;
13195 AvALLOC((const AV *)dstr) = (SV**)NULL;
13196 AvMAX( (const AV *)dstr) = -1;
13197 AvFILLp((const AV *)dstr) = -1;
13201 if (HvARRAY((const HV *)sstr)) {
13203 const bool sharekeys = !!HvSHAREKEYS(sstr);
13204 XPVHV * const dxhv = (XPVHV*)SvANY(dstr);
13205 XPVHV * const sxhv = (XPVHV*)SvANY(sstr);
13207 Newx(darray, PERL_HV_ARRAY_ALLOC_BYTES(dxhv->xhv_max+1)
13208 + (SvOOK(sstr) ? sizeof(struct xpvhv_aux) : 0),
13210 HvARRAY(dstr) = (HE**)darray;
13211 while (i <= sxhv->xhv_max) {
13212 const HE * const source = HvARRAY(sstr)[i];
13213 HvARRAY(dstr)[i] = source
13214 ? he_dup(source, sharekeys, param) : 0;
13218 const struct xpvhv_aux * const saux = HvAUX(sstr);
13219 struct xpvhv_aux * const daux = HvAUX(dstr);
13220 /* This flag isn't copied. */
13223 if (saux->xhv_name_count) {
13224 HEK ** const sname = saux->xhv_name_u.xhvnameu_names;
13226 = saux->xhv_name_count < 0
13227 ? -saux->xhv_name_count
13228 : saux->xhv_name_count;
13229 HEK **shekp = sname + count;
13231 Newx(daux->xhv_name_u.xhvnameu_names, count, HEK *);
13232 dhekp = daux->xhv_name_u.xhvnameu_names + count;
13233 while (shekp-- > sname) {
13235 *dhekp = hek_dup(*shekp, param);
13239 daux->xhv_name_u.xhvnameu_name
13240 = hek_dup(saux->xhv_name_u.xhvnameu_name,
13243 daux->xhv_name_count = saux->xhv_name_count;
13245 daux->xhv_fill_lazy = saux->xhv_fill_lazy;
13246 daux->xhv_aux_flags = saux->xhv_aux_flags;
13247 #ifdef PERL_HASH_RANDOMIZE_KEYS
13248 daux->xhv_rand = saux->xhv_rand;
13249 daux->xhv_last_rand = saux->xhv_last_rand;
13251 daux->xhv_riter = saux->xhv_riter;
13252 daux->xhv_eiter = saux->xhv_eiter
13253 ? he_dup(saux->xhv_eiter,
13254 cBOOL(HvSHAREKEYS(sstr)), param) : 0;
13255 /* backref array needs refcnt=2; see sv_add_backref */
13256 daux->xhv_backreferences =
13257 (param->flags & CLONEf_JOIN_IN)
13258 /* when joining, we let the individual GVs and
13259 * CVs add themselves to backref as
13260 * needed. This avoids pulling in stuff
13261 * that isn't required, and simplifies the
13262 * case where stashes aren't cloned back
13263 * if they already exist in the parent
13266 : saux->xhv_backreferences
13267 ? (SvTYPE(saux->xhv_backreferences) == SVt_PVAV)
13268 ? MUTABLE_AV(SvREFCNT_inc(
13269 sv_dup_inc((const SV *)
13270 saux->xhv_backreferences, param)))
13271 : MUTABLE_AV(sv_dup((const SV *)
13272 saux->xhv_backreferences, param))
13275 daux->xhv_mro_meta = saux->xhv_mro_meta
13276 ? mro_meta_dup(saux->xhv_mro_meta, param)
13279 /* Record stashes for possible cloning in Perl_clone(). */
13281 av_push(param->stashes, dstr);
13285 HvARRAY(MUTABLE_HV(dstr)) = NULL;
13288 if (!(param->flags & CLONEf_COPY_STACKS)) {
13293 /* NOTE: not refcounted */
13294 SvANY(MUTABLE_CV(dstr))->xcv_stash =
13295 hv_dup(CvSTASH(dstr), param);
13296 if ((param->flags & CLONEf_JOIN_IN) && CvSTASH(dstr))
13297 Perl_sv_add_backref(aTHX_ MUTABLE_SV(CvSTASH(dstr)), dstr);
13298 if (!CvISXSUB(dstr)) {
13300 CvROOT(dstr) = OpREFCNT_inc(CvROOT(dstr));
13302 CvSLABBED_off(dstr);
13303 } else if (CvCONST(dstr)) {
13304 CvXSUBANY(dstr).any_ptr =
13305 sv_dup_inc((const SV *)CvXSUBANY(dstr).any_ptr, param);
13307 assert(!CvSLABBED(dstr));
13308 if (CvDYNFILE(dstr)) CvFILE(dstr) = SAVEPV(CvFILE(dstr));
13310 SvANY((CV *)dstr)->xcv_gv_u.xcv_hek =
13311 share_hek_hek(CvNAME_HEK((CV *)sstr));
13312 /* don't dup if copying back - CvGV isn't refcounted, so the
13313 * duped GV may never be freed. A bit of a hack! DAPM */
13315 SvANY(MUTABLE_CV(dstr))->xcv_gv_u.xcv_gv =
13317 ? gv_dup_inc(CvGV(sstr), param)
13318 : (param->flags & CLONEf_JOIN_IN)
13320 : gv_dup(CvGV(sstr), param);
13322 CvPADLIST(dstr) = padlist_dup(CvPADLIST(sstr), param);
13324 CvWEAKOUTSIDE(sstr)
13325 ? cv_dup( CvOUTSIDE(dstr), param)
13326 : cv_dup_inc(CvOUTSIDE(dstr), param);
13336 Perl_sv_dup_inc(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13338 PERL_ARGS_ASSERT_SV_DUP_INC;
13339 return sstr ? SvREFCNT_inc(sv_dup_common(sstr, param)) : NULL;
13343 Perl_sv_dup(pTHX_ const SV *const sstr, CLONE_PARAMS *const param)
13345 SV *dstr = sstr ? sv_dup_common(sstr, param) : NULL;
13346 PERL_ARGS_ASSERT_SV_DUP;
13348 /* Track every SV that (at least initially) had a reference count of 0.
13349 We need to do this by holding an actual reference to it in this array.
13350 If we attempt to cheat, turn AvREAL_off(), and store only pointers
13351 (akin to the stashes hash, and the perl stack), we come unstuck if
13352 a weak reference (or other SV legitimately SvREFCNT() == 0 for this
13353 thread) is manipulated in a CLONE method, because CLONE runs before the
13354 unreferenced array is walked to find SVs still with SvREFCNT() == 0
13355 (and fix things up by giving each a reference via the temps stack).
13356 Instead, during CLONE, if the 0-referenced SV has SvREFCNT_inc() and
13357 then SvREFCNT_dec(), it will be cleaned up (and added to the free list)
13358 before the walk of unreferenced happens and a reference to that is SV
13359 added to the temps stack. At which point we have the same SV considered
13360 to be in use, and free to be re-used. Not good.
13362 if (dstr && !(param->flags & CLONEf_COPY_STACKS) && !SvREFCNT(dstr)) {
13363 assert(param->unreferenced);
13364 av_push(param->unreferenced, SvREFCNT_inc(dstr));
13370 /* duplicate a context */
13373 Perl_cx_dup(pTHX_ PERL_CONTEXT *cxs, I32 ix, I32 max, CLONE_PARAMS* param)
13375 PERL_CONTEXT *ncxs;
13377 PERL_ARGS_ASSERT_CX_DUP;
13380 return (PERL_CONTEXT*)NULL;
13382 /* look for it in the table first */
13383 ncxs = (PERL_CONTEXT*)ptr_table_fetch(PL_ptr_table, cxs);
13387 /* create anew and remember what it is */
13388 Newx(ncxs, max + 1, PERL_CONTEXT);
13389 ptr_table_store(PL_ptr_table, cxs, ncxs);
13390 Copy(cxs, ncxs, max + 1, PERL_CONTEXT);
13393 PERL_CONTEXT * const ncx = &ncxs[ix];
13394 if (CxTYPE(ncx) == CXt_SUBST) {
13395 Perl_croak(aTHX_ "Cloning substitution context is unimplemented");
13398 ncx->blk_oldcop = (COP*)any_dup(ncx->blk_oldcop, param->proto_perl);
13399 switch (CxTYPE(ncx)) {
13401 ncx->blk_sub.cv = (ncx->blk_sub.olddepth == 0
13402 ? cv_dup_inc(ncx->blk_sub.cv, param)
13403 : cv_dup(ncx->blk_sub.cv,param));
13404 if(CxHASARGS(ncx)){
13405 ncx->blk_sub.argarray = av_dup_inc(ncx->blk_sub.argarray,param);
13406 ncx->blk_sub.savearray = av_dup_inc(ncx->blk_sub.savearray,param);
13408 ncx->blk_sub.argarray = NULL;
13409 ncx->blk_sub.savearray = NULL;
13411 ncx->blk_sub.oldcomppad = (PAD*)ptr_table_fetch(PL_ptr_table,
13412 ncx->blk_sub.oldcomppad);
13415 ncx->blk_eval.old_namesv = sv_dup_inc(ncx->blk_eval.old_namesv,
13417 ncx->blk_eval.cur_text = sv_dup(ncx->blk_eval.cur_text, param);
13418 ncx->blk_eval.cv = cv_dup(ncx->blk_eval.cv, param);
13420 case CXt_LOOP_LAZYSV:
13421 ncx->blk_loop.state_u.lazysv.end
13422 = sv_dup_inc(ncx->blk_loop.state_u.lazysv.end, param);
13423 /* We are taking advantage of av_dup_inc and sv_dup_inc
13424 actually being the same function, and order equivalence of
13426 We can assert the later [but only at run time :-(] */
13427 assert ((void *) &ncx->blk_loop.state_u.ary.ary ==
13428 (void *) &ncx->blk_loop.state_u.lazysv.cur);
13430 ncx->blk_loop.state_u.ary.ary
13431 = av_dup_inc(ncx->blk_loop.state_u.ary.ary, param);
13432 case CXt_LOOP_LAZYIV:
13433 case CXt_LOOP_PLAIN:
13434 if (CxPADLOOP(ncx)) {
13435 ncx->blk_loop.itervar_u.oldcomppad
13436 = (PAD*)ptr_table_fetch(PL_ptr_table,
13437 ncx->blk_loop.itervar_u.oldcomppad);
13439 ncx->blk_loop.itervar_u.gv
13440 = gv_dup((const GV *)ncx->blk_loop.itervar_u.gv,
13445 ncx->blk_format.cv = cv_dup(ncx->blk_format.cv, param);
13446 ncx->blk_format.gv = gv_dup(ncx->blk_format.gv, param);
13447 ncx->blk_format.dfoutgv = gv_dup_inc(ncx->blk_format.dfoutgv,
13462 /* duplicate a stack info structure */
13465 Perl_si_dup(pTHX_ PERL_SI *si, CLONE_PARAMS* param)
13469 PERL_ARGS_ASSERT_SI_DUP;
13472 return (PERL_SI*)NULL;
13474 /* look for it in the table first */
13475 nsi = (PERL_SI*)ptr_table_fetch(PL_ptr_table, si);
13479 /* create anew and remember what it is */
13480 Newxz(nsi, 1, PERL_SI);
13481 ptr_table_store(PL_ptr_table, si, nsi);
13483 nsi->si_stack = av_dup_inc(si->si_stack, param);
13484 nsi->si_cxix = si->si_cxix;
13485 nsi->si_cxmax = si->si_cxmax;
13486 nsi->si_cxstack = cx_dup(si->si_cxstack, si->si_cxix, si->si_cxmax, param);
13487 nsi->si_type = si->si_type;
13488 nsi->si_prev = si_dup(si->si_prev, param);
13489 nsi->si_next = si_dup(si->si_next, param);
13490 nsi->si_markoff = si->si_markoff;
13495 #define POPINT(ss,ix) ((ss)[--(ix)].any_i32)
13496 #define TOPINT(ss,ix) ((ss)[ix].any_i32)
13497 #define POPLONG(ss,ix) ((ss)[--(ix)].any_long)
13498 #define TOPLONG(ss,ix) ((ss)[ix].any_long)
13499 #define POPIV(ss,ix) ((ss)[--(ix)].any_iv)
13500 #define TOPIV(ss,ix) ((ss)[ix].any_iv)
13501 #define POPUV(ss,ix) ((ss)[--(ix)].any_uv)
13502 #define TOPUV(ss,ix) ((ss)[ix].any_uv)
13503 #define POPBOOL(ss,ix) ((ss)[--(ix)].any_bool)
13504 #define TOPBOOL(ss,ix) ((ss)[ix].any_bool)
13505 #define POPPTR(ss,ix) ((ss)[--(ix)].any_ptr)
13506 #define TOPPTR(ss,ix) ((ss)[ix].any_ptr)
13507 #define POPDPTR(ss,ix) ((ss)[--(ix)].any_dptr)
13508 #define TOPDPTR(ss,ix) ((ss)[ix].any_dptr)
13509 #define POPDXPTR(ss,ix) ((ss)[--(ix)].any_dxptr)
13510 #define TOPDXPTR(ss,ix) ((ss)[ix].any_dxptr)
13513 #define pv_dup_inc(p) SAVEPV(p)
13514 #define pv_dup(p) SAVEPV(p)
13515 #define svp_dup_inc(p,pp) any_dup(p,pp)
13517 /* map any object to the new equivent - either something in the
13518 * ptr table, or something in the interpreter structure
13522 Perl_any_dup(pTHX_ void *v, const PerlInterpreter *proto_perl)
13526 PERL_ARGS_ASSERT_ANY_DUP;
13529 return (void*)NULL;
13531 /* look for it in the table first */
13532 ret = ptr_table_fetch(PL_ptr_table, v);
13536 /* see if it is part of the interpreter structure */
13537 if (v >= (void*)proto_perl && v < (void*)(proto_perl+1))
13538 ret = (void*)(((char*)aTHX) + (((char*)v) - (char*)proto_perl));
13546 /* duplicate the save stack */
13549 Perl_ss_dup(pTHX_ PerlInterpreter *proto_perl, CLONE_PARAMS* param)
13552 ANY * const ss = proto_perl->Isavestack;
13553 const I32 max = proto_perl->Isavestack_max;
13554 I32 ix = proto_perl->Isavestack_ix;
13567 void (*dptr) (void*);
13568 void (*dxptr) (pTHX_ void*);
13570 PERL_ARGS_ASSERT_SS_DUP;
13572 Newxz(nss, max, ANY);
13575 const UV uv = POPUV(ss,ix);
13576 const U8 type = (U8)uv & SAVE_MASK;
13578 TOPUV(nss,ix) = uv;
13580 case SAVEt_CLEARSV:
13581 case SAVEt_CLEARPADRANGE:
13583 case SAVEt_HELEM: /* hash element */
13584 sv = (const SV *)POPPTR(ss,ix);
13585 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13587 case SAVEt_ITEM: /* normal string */
13588 case SAVEt_GVSV: /* scalar slot in GV */
13589 case SAVEt_SV: /* scalar reference */
13590 sv = (const SV *)POPPTR(ss,ix);
13591 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13594 case SAVEt_MORTALIZESV:
13595 case SAVEt_READONLY_OFF:
13596 sv = (const SV *)POPPTR(ss,ix);
13597 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13599 case SAVEt_SHARED_PVREF: /* char* in shared space */
13600 c = (char*)POPPTR(ss,ix);
13601 TOPPTR(nss,ix) = savesharedpv(c);
13602 ptr = POPPTR(ss,ix);
13603 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13605 case SAVEt_GENERIC_SVREF: /* generic sv */
13606 case SAVEt_SVREF: /* scalar reference */
13607 sv = (const SV *)POPPTR(ss,ix);
13608 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13609 ptr = POPPTR(ss,ix);
13610 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
13612 case SAVEt_GVSLOT: /* any slot in GV */
13613 sv = (const SV *)POPPTR(ss,ix);
13614 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13615 ptr = POPPTR(ss,ix);
13616 TOPPTR(nss,ix) = svp_dup_inc((SV**)ptr, proto_perl);/* XXXXX */
13617 sv = (const SV *)POPPTR(ss,ix);
13618 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13620 case SAVEt_HV: /* hash reference */
13621 case SAVEt_AV: /* array reference */
13622 sv = (const SV *) POPPTR(ss,ix);
13623 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13625 case SAVEt_COMPPAD:
13627 sv = (const SV *) POPPTR(ss,ix);
13628 TOPPTR(nss,ix) = sv_dup(sv, param);
13630 case SAVEt_INT: /* int reference */
13631 ptr = POPPTR(ss,ix);
13632 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13633 intval = (int)POPINT(ss,ix);
13634 TOPINT(nss,ix) = intval;
13636 case SAVEt_LONG: /* long reference */
13637 ptr = POPPTR(ss,ix);
13638 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13639 longval = (long)POPLONG(ss,ix);
13640 TOPLONG(nss,ix) = longval;
13642 case SAVEt_I32: /* I32 reference */
13643 ptr = POPPTR(ss,ix);
13644 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13646 TOPINT(nss,ix) = i;
13648 case SAVEt_IV: /* IV reference */
13649 case SAVEt_STRLEN: /* STRLEN/size_t ref */
13650 ptr = POPPTR(ss,ix);
13651 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13653 TOPIV(nss,ix) = iv;
13655 case SAVEt_HPTR: /* HV* reference */
13656 case SAVEt_APTR: /* AV* reference */
13657 case SAVEt_SPTR: /* SV* reference */
13658 ptr = POPPTR(ss,ix);
13659 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13660 sv = (const SV *)POPPTR(ss,ix);
13661 TOPPTR(nss,ix) = sv_dup(sv, param);
13663 case SAVEt_VPTR: /* random* reference */
13664 ptr = POPPTR(ss,ix);
13665 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13667 case SAVEt_INT_SMALL:
13668 case SAVEt_I32_SMALL:
13669 case SAVEt_I16: /* I16 reference */
13670 case SAVEt_I8: /* I8 reference */
13672 ptr = POPPTR(ss,ix);
13673 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13675 case SAVEt_GENERIC_PVREF: /* generic char* */
13676 case SAVEt_PPTR: /* char* reference */
13677 ptr = POPPTR(ss,ix);
13678 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13679 c = (char*)POPPTR(ss,ix);
13680 TOPPTR(nss,ix) = pv_dup(c);
13682 case SAVEt_GP: /* scalar reference */
13683 gp = (GP*)POPPTR(ss,ix);
13684 TOPPTR(nss,ix) = gp = gp_dup(gp, param);
13685 (void)GpREFCNT_inc(gp);
13686 gv = (const GV *)POPPTR(ss,ix);
13687 TOPPTR(nss,ix) = gv_dup_inc(gv, param);
13690 ptr = POPPTR(ss,ix);
13691 if (ptr && (((OP*)ptr)->op_private & OPpREFCOUNTED)) {
13692 /* these are assumed to be refcounted properly */
13694 switch (((OP*)ptr)->op_type) {
13696 case OP_LEAVESUBLV:
13700 case OP_LEAVEWRITE:
13701 TOPPTR(nss,ix) = ptr;
13704 (void) OpREFCNT_inc(o);
13708 TOPPTR(nss,ix) = NULL;
13713 TOPPTR(nss,ix) = NULL;
13715 case SAVEt_FREECOPHH:
13716 ptr = POPPTR(ss,ix);
13717 TOPPTR(nss,ix) = cophh_copy((COPHH *)ptr);
13719 case SAVEt_ADELETE:
13720 av = (const AV *)POPPTR(ss,ix);
13721 TOPPTR(nss,ix) = av_dup_inc(av, param);
13723 TOPINT(nss,ix) = i;
13726 hv = (const HV *)POPPTR(ss,ix);
13727 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
13729 TOPINT(nss,ix) = i;
13732 c = (char*)POPPTR(ss,ix);
13733 TOPPTR(nss,ix) = pv_dup_inc(c);
13735 case SAVEt_STACK_POS: /* Position on Perl stack */
13737 TOPINT(nss,ix) = i;
13739 case SAVEt_DESTRUCTOR:
13740 ptr = POPPTR(ss,ix);
13741 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
13742 dptr = POPDPTR(ss,ix);
13743 TOPDPTR(nss,ix) = DPTR2FPTR(void (*)(void*),
13744 any_dup(FPTR2DPTR(void *, dptr),
13747 case SAVEt_DESTRUCTOR_X:
13748 ptr = POPPTR(ss,ix);
13749 TOPPTR(nss,ix) = any_dup(ptr, proto_perl); /* XXX quite arbitrary */
13750 dxptr = POPDXPTR(ss,ix);
13751 TOPDXPTR(nss,ix) = DPTR2FPTR(void (*)(pTHX_ void*),
13752 any_dup(FPTR2DPTR(void *, dxptr),
13755 case SAVEt_REGCONTEXT:
13757 ix -= uv >> SAVE_TIGHT_SHIFT;
13759 case SAVEt_AELEM: /* array element */
13760 sv = (const SV *)POPPTR(ss,ix);
13761 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13763 TOPINT(nss,ix) = i;
13764 av = (const AV *)POPPTR(ss,ix);
13765 TOPPTR(nss,ix) = av_dup_inc(av, param);
13768 ptr = POPPTR(ss,ix);
13769 TOPPTR(nss,ix) = ptr;
13772 ptr = POPPTR(ss,ix);
13773 ptr = cophh_copy((COPHH*)ptr);
13774 TOPPTR(nss,ix) = ptr;
13776 TOPINT(nss,ix) = i;
13777 if (i & HINT_LOCALIZE_HH) {
13778 hv = (const HV *)POPPTR(ss,ix);
13779 TOPPTR(nss,ix) = hv_dup_inc(hv, param);
13782 case SAVEt_PADSV_AND_MORTALIZE:
13783 longval = (long)POPLONG(ss,ix);
13784 TOPLONG(nss,ix) = longval;
13785 ptr = POPPTR(ss,ix);
13786 TOPPTR(nss,ix) = any_dup(ptr, proto_perl);
13787 sv = (const SV *)POPPTR(ss,ix);
13788 TOPPTR(nss,ix) = sv_dup_inc(sv, param);
13790 case SAVEt_SET_SVFLAGS:
13792 TOPINT(nss,ix) = i;
13794 TOPINT(nss,ix) = i;
13795 sv = (const SV *)POPPTR(ss,ix);
13796 TOPPTR(nss,ix) = sv_dup(sv, param);
13798 case SAVEt_COMPILE_WARNINGS:
13799 ptr = POPPTR(ss,ix);
13800 TOPPTR(nss,ix) = DUP_WARNINGS((STRLEN*)ptr);
13803 ptr = POPPTR(ss,ix);
13804 TOPPTR(nss,ix) = parser_dup((const yy_parser*)ptr, param);
13808 "panic: ss_dup inconsistency (%"IVdf")", (IV) type);
13816 /* if sv is a stash, call $class->CLONE_SKIP(), and set the SVphv_CLONEABLE
13817 * flag to the result. This is done for each stash before cloning starts,
13818 * so we know which stashes want their objects cloned */
13821 do_mark_cloneable_stash(pTHX_ SV *const sv)
13823 const HEK * const hvname = HvNAME_HEK((const HV *)sv);
13825 GV* const cloner = gv_fetchmethod_autoload(MUTABLE_HV(sv), "CLONE_SKIP", 0);
13826 SvFLAGS(sv) |= SVphv_CLONEABLE; /* clone objects by default */
13827 if (cloner && GvCV(cloner)) {
13834 mXPUSHs(newSVhek(hvname));
13836 call_sv(MUTABLE_SV(GvCV(cloner)), G_SCALAR);
13843 SvFLAGS(sv) &= ~SVphv_CLONEABLE;
13851 =for apidoc perl_clone
13853 Create and return a new interpreter by cloning the current one.
13855 perl_clone takes these flags as parameters:
13857 CLONEf_COPY_STACKS - is used to, well, copy the stacks also,
13858 without it we only clone the data and zero the stacks,
13859 with it we copy the stacks and the new perl interpreter is
13860 ready to run at the exact same point as the previous one.
13861 The pseudo-fork code uses COPY_STACKS while the
13862 threads->create doesn't.
13864 CLONEf_KEEP_PTR_TABLE -
13865 perl_clone keeps a ptr_table with the pointer of the old
13866 variable as a key and the new variable as a value,
13867 this allows it to check if something has been cloned and not
13868 clone it again but rather just use the value and increase the
13869 refcount. If KEEP_PTR_TABLE is not set then perl_clone will kill
13870 the ptr_table using the function
13871 C<ptr_table_free(PL_ptr_table); PL_ptr_table = NULL;>,
13872 reason to keep it around is if you want to dup some of your own
13873 variable who are outside the graph perl scans, example of this
13874 code is in threads.xs create.
13876 CLONEf_CLONE_HOST -
13877 This is a win32 thing, it is ignored on unix, it tells perls
13878 win32host code (which is c++) to clone itself, this is needed on
13879 win32 if you want to run two threads at the same time,
13880 if you just want to do some stuff in a separate perl interpreter
13881 and then throw it away and return to the original one,
13882 you don't need to do anything.
13887 /* XXX the above needs expanding by someone who actually understands it ! */
13888 EXTERN_C PerlInterpreter *
13889 perl_clone_host(PerlInterpreter* proto_perl, UV flags);
13892 perl_clone(PerlInterpreter *proto_perl, UV flags)
13895 #ifdef PERL_IMPLICIT_SYS
13897 PERL_ARGS_ASSERT_PERL_CLONE;
13899 /* perlhost.h so we need to call into it
13900 to clone the host, CPerlHost should have a c interface, sky */
13902 if (flags & CLONEf_CLONE_HOST) {
13903 return perl_clone_host(proto_perl,flags);
13905 return perl_clone_using(proto_perl, flags,
13907 proto_perl->IMemShared,
13908 proto_perl->IMemParse,
13910 proto_perl->IStdIO,
13914 proto_perl->IProc);
13918 perl_clone_using(PerlInterpreter *proto_perl, UV flags,
13919 struct IPerlMem* ipM, struct IPerlMem* ipMS,
13920 struct IPerlMem* ipMP, struct IPerlEnv* ipE,
13921 struct IPerlStdIO* ipStd, struct IPerlLIO* ipLIO,
13922 struct IPerlDir* ipD, struct IPerlSock* ipS,
13923 struct IPerlProc* ipP)
13925 /* XXX many of the string copies here can be optimized if they're
13926 * constants; they need to be allocated as common memory and just
13927 * their pointers copied. */
13930 CLONE_PARAMS clone_params;
13931 CLONE_PARAMS* const param = &clone_params;
13933 PerlInterpreter * const my_perl = (PerlInterpreter*)(*ipM->pMalloc)(ipM, sizeof(PerlInterpreter));
13935 PERL_ARGS_ASSERT_PERL_CLONE_USING;
13936 #else /* !PERL_IMPLICIT_SYS */
13938 CLONE_PARAMS clone_params;
13939 CLONE_PARAMS* param = &clone_params;
13940 PerlInterpreter * const my_perl = (PerlInterpreter*)PerlMem_malloc(sizeof(PerlInterpreter));
13942 PERL_ARGS_ASSERT_PERL_CLONE;
13943 #endif /* PERL_IMPLICIT_SYS */
13945 /* for each stash, determine whether its objects should be cloned */
13946 S_visit(proto_perl, do_mark_cloneable_stash, SVt_PVHV, SVTYPEMASK);
13947 PERL_SET_THX(my_perl);
13950 PoisonNew(my_perl, 1, PerlInterpreter);
13953 PL_defstash = NULL; /* may be used by perl malloc() */
13956 PL_scopestack_name = 0;
13958 PL_savestack_ix = 0;
13959 PL_savestack_max = -1;
13960 PL_sig_pending = 0;
13962 Zero(&PL_debug_pad, 1, struct perl_debug_pad);
13963 # ifdef DEBUG_LEAKING_SCALARS
13964 PL_sv_serial = (((UV)my_perl >> 2) & 0xfff) * 1000000;
13966 #else /* !DEBUGGING */
13967 Zero(my_perl, 1, PerlInterpreter);
13968 #endif /* DEBUGGING */
13970 #ifdef PERL_IMPLICIT_SYS
13971 /* host pointers */
13973 PL_MemShared = ipMS;
13974 PL_MemParse = ipMP;
13981 #endif /* PERL_IMPLICIT_SYS */
13984 param->flags = flags;
13985 /* Nothing in the core code uses this, but we make it available to
13986 extensions (using mg_dup). */
13987 param->proto_perl = proto_perl;
13988 /* Likely nothing will use this, but it is initialised to be consistent
13989 with Perl_clone_params_new(). */
13990 param->new_perl = my_perl;
13991 param->unreferenced = NULL;
13994 INIT_TRACK_MEMPOOL(my_perl->Imemory_debug_header, my_perl);
13996 PL_body_arenas = NULL;
13997 Zero(&PL_body_roots, 1, PL_body_roots);
14001 PL_sv_arenaroot = NULL;
14003 PL_debug = proto_perl->Idebug;
14005 /* dbargs array probably holds garbage */
14008 PL_compiling = proto_perl->Icompiling;
14010 /* pseudo environmental stuff */
14011 PL_origargc = proto_perl->Iorigargc;
14012 PL_origargv = proto_perl->Iorigargv;
14014 #ifndef NO_TAINT_SUPPORT
14015 /* Set tainting stuff before PerlIO_debug can possibly get called */
14016 PL_tainting = proto_perl->Itainting;
14017 PL_taint_warn = proto_perl->Itaint_warn;
14019 PL_tainting = FALSE;
14020 PL_taint_warn = FALSE;
14023 PL_minus_c = proto_perl->Iminus_c;
14025 PL_localpatches = proto_perl->Ilocalpatches;
14026 PL_splitstr = proto_perl->Isplitstr;
14027 PL_minus_n = proto_perl->Iminus_n;
14028 PL_minus_p = proto_perl->Iminus_p;
14029 PL_minus_l = proto_perl->Iminus_l;
14030 PL_minus_a = proto_perl->Iminus_a;
14031 PL_minus_E = proto_perl->Iminus_E;
14032 PL_minus_F = proto_perl->Iminus_F;
14033 PL_doswitches = proto_perl->Idoswitches;
14034 PL_dowarn = proto_perl->Idowarn;
14035 #ifdef PERL_SAWAMPERSAND
14036 PL_sawampersand = proto_perl->Isawampersand;
14038 PL_unsafe = proto_perl->Iunsafe;
14039 PL_perldb = proto_perl->Iperldb;
14040 PL_perl_destruct_level = proto_perl->Iperl_destruct_level;
14041 PL_exit_flags = proto_perl->Iexit_flags;
14043 /* XXX time(&PL_basetime) when asked for? */
14044 PL_basetime = proto_perl->Ibasetime;
14046 PL_maxsysfd = proto_perl->Imaxsysfd;
14047 PL_statusvalue = proto_perl->Istatusvalue;
14049 PL_statusvalue_vms = proto_perl->Istatusvalue_vms;
14051 PL_statusvalue_posix = proto_perl->Istatusvalue_posix;
14054 /* RE engine related */
14055 PL_regmatch_slab = NULL;
14056 PL_reg_curpm = NULL;
14058 PL_sub_generation = proto_perl->Isub_generation;
14060 /* funky return mechanisms */
14061 PL_forkprocess = proto_perl->Iforkprocess;
14063 /* internal state */
14064 PL_maxo = proto_perl->Imaxo;
14066 PL_main_start = proto_perl->Imain_start;
14067 PL_eval_root = proto_perl->Ieval_root;
14068 PL_eval_start = proto_perl->Ieval_start;
14070 PL_filemode = proto_perl->Ifilemode;
14071 PL_lastfd = proto_perl->Ilastfd;
14072 PL_oldname = proto_perl->Ioldname; /* XXX not quite right */
14075 PL_gensym = proto_perl->Igensym;
14077 PL_laststatval = proto_perl->Ilaststatval;
14078 PL_laststype = proto_perl->Ilaststype;
14081 PL_profiledata = NULL;
14083 PL_generation = proto_perl->Igeneration;
14085 PL_in_clean_objs = proto_perl->Iin_clean_objs;
14086 PL_in_clean_all = proto_perl->Iin_clean_all;
14088 PL_delaymagic_uid = proto_perl->Idelaymagic_uid;
14089 PL_delaymagic_euid = proto_perl->Idelaymagic_euid;
14090 PL_delaymagic_gid = proto_perl->Idelaymagic_gid;
14091 PL_delaymagic_egid = proto_perl->Idelaymagic_egid;
14092 PL_nomemok = proto_perl->Inomemok;
14093 PL_an = proto_perl->Ian;
14094 PL_evalseq = proto_perl->Ievalseq;
14095 PL_origenviron = proto_perl->Iorigenviron; /* XXX not quite right */
14096 PL_origalen = proto_perl->Iorigalen;
14098 PL_sighandlerp = proto_perl->Isighandlerp;
14100 PL_runops = proto_perl->Irunops;
14102 PL_subline = proto_perl->Isubline;
14105 PL_cryptseen = proto_perl->Icryptseen;
14108 #ifdef USE_LOCALE_COLLATE
14109 PL_collation_ix = proto_perl->Icollation_ix;
14110 PL_collation_standard = proto_perl->Icollation_standard;
14111 PL_collxfrm_base = proto_perl->Icollxfrm_base;
14112 PL_collxfrm_mult = proto_perl->Icollxfrm_mult;
14113 #endif /* USE_LOCALE_COLLATE */
14115 #ifdef USE_LOCALE_NUMERIC
14116 PL_numeric_standard = proto_perl->Inumeric_standard;
14117 PL_numeric_local = proto_perl->Inumeric_local;
14118 #endif /* !USE_LOCALE_NUMERIC */
14120 /* Did the locale setup indicate UTF-8? */
14121 PL_utf8locale = proto_perl->Iutf8locale;
14122 PL_in_utf8_CTYPE_locale = proto_perl->Iin_utf8_CTYPE_locale;
14123 /* Unicode features (see perlrun/-C) */
14124 PL_unicode = proto_perl->Iunicode;
14126 /* Pre-5.8 signals control */
14127 PL_signals = proto_perl->Isignals;
14129 /* times() ticks per second */
14130 PL_clocktick = proto_perl->Iclocktick;
14132 /* Recursion stopper for PerlIO_find_layer */
14133 PL_in_load_module = proto_perl->Iin_load_module;
14135 /* sort() routine */
14136 PL_sort_RealCmp = proto_perl->Isort_RealCmp;
14138 /* Not really needed/useful since the reenrant_retint is "volatile",
14139 * but do it for consistency's sake. */
14140 PL_reentrant_retint = proto_perl->Ireentrant_retint;
14142 /* Hooks to shared SVs and locks. */
14143 PL_sharehook = proto_perl->Isharehook;
14144 PL_lockhook = proto_perl->Ilockhook;
14145 PL_unlockhook = proto_perl->Iunlockhook;
14146 PL_threadhook = proto_perl->Ithreadhook;
14147 PL_destroyhook = proto_perl->Idestroyhook;
14148 PL_signalhook = proto_perl->Isignalhook;
14150 PL_globhook = proto_perl->Iglobhook;
14153 PL_last_swash_hv = NULL; /* reinits on demand */
14154 PL_last_swash_klen = 0;
14155 PL_last_swash_key[0]= '\0';
14156 PL_last_swash_tmps = (U8*)NULL;
14157 PL_last_swash_slen = 0;
14159 PL_srand_called = proto_perl->Isrand_called;
14160 Copy(&(proto_perl->Irandom_state), &PL_random_state, 1, PL_RANDOM_STATE_TYPE);
14162 if (flags & CLONEf_COPY_STACKS) {
14163 /* next allocation will be PL_tmps_stack[PL_tmps_ix+1] */
14164 PL_tmps_ix = proto_perl->Itmps_ix;
14165 PL_tmps_max = proto_perl->Itmps_max;
14166 PL_tmps_floor = proto_perl->Itmps_floor;
14168 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
14169 * NOTE: unlike the others! */
14170 PL_scopestack_ix = proto_perl->Iscopestack_ix;
14171 PL_scopestack_max = proto_perl->Iscopestack_max;
14173 /* next SSPUSHFOO() sets PL_savestack[PL_savestack_ix]
14174 * NOTE: unlike the others! */
14175 PL_savestack_ix = proto_perl->Isavestack_ix;
14176 PL_savestack_max = proto_perl->Isavestack_max;
14179 PL_start_env = proto_perl->Istart_env; /* XXXXXX */
14180 PL_top_env = &PL_start_env;
14182 PL_op = proto_perl->Iop;
14185 PL_Xpv = (XPV*)NULL;
14186 my_perl->Ina = proto_perl->Ina;
14188 PL_statbuf = proto_perl->Istatbuf;
14189 PL_statcache = proto_perl->Istatcache;
14191 #ifndef NO_TAINT_SUPPORT
14192 PL_tainted = proto_perl->Itainted;
14194 PL_tainted = FALSE;
14196 PL_curpm = proto_perl->Icurpm; /* XXX No PMOP ref count */
14198 PL_chopset = proto_perl->Ichopset; /* XXX never deallocated */
14200 PL_restartjmpenv = proto_perl->Irestartjmpenv;
14201 PL_restartop = proto_perl->Irestartop;
14202 PL_in_eval = proto_perl->Iin_eval;
14203 PL_delaymagic = proto_perl->Idelaymagic;
14204 PL_phase = proto_perl->Iphase;
14205 PL_localizing = proto_perl->Ilocalizing;
14207 PL_hv_fetch_ent_mh = NULL;
14208 PL_modcount = proto_perl->Imodcount;
14209 PL_lastgotoprobe = NULL;
14210 PL_dumpindent = proto_perl->Idumpindent;
14212 PL_efloatbuf = NULL; /* reinits on demand */
14213 PL_efloatsize = 0; /* reinits on demand */
14217 PL_colorset = 0; /* reinits PL_colors[] */
14218 /*PL_colors[6] = {0,0,0,0,0,0};*/
14220 /* Pluggable optimizer */
14221 PL_peepp = proto_perl->Ipeepp;
14222 PL_rpeepp = proto_perl->Irpeepp;
14223 /* op_free() hook */
14224 PL_opfreehook = proto_perl->Iopfreehook;
14226 #ifdef USE_REENTRANT_API
14227 /* XXX: things like -Dm will segfault here in perlio, but doing
14228 * PERL_SET_CONTEXT(proto_perl);
14229 * breaks too many other things
14231 Perl_reentrant_init(aTHX);
14234 /* create SV map for pointer relocation */
14235 PL_ptr_table = ptr_table_new();
14237 /* initialize these special pointers as early as possible */
14239 ptr_table_store(PL_ptr_table, &proto_perl->Isv_undef, &PL_sv_undef);
14240 ptr_table_store(PL_ptr_table, &proto_perl->Isv_no, &PL_sv_no);
14241 ptr_table_store(PL_ptr_table, &proto_perl->Isv_yes, &PL_sv_yes);
14243 /* create (a non-shared!) shared string table */
14244 PL_strtab = newHV();
14245 HvSHAREKEYS_off(PL_strtab);
14246 hv_ksplit(PL_strtab, HvTOTALKEYS(proto_perl->Istrtab));
14247 ptr_table_store(PL_ptr_table, proto_perl->Istrtab, PL_strtab);
14249 Zero(PL_sv_consts, SV_CONSTS_COUNT, SV*);
14251 /* This PV will be free'd special way so must set it same way op.c does */
14252 PL_compiling.cop_file = savesharedpv(PL_compiling.cop_file);
14253 ptr_table_store(PL_ptr_table, proto_perl->Icompiling.cop_file, PL_compiling.cop_file);
14255 ptr_table_store(PL_ptr_table, &proto_perl->Icompiling, &PL_compiling);
14256 PL_compiling.cop_warnings = DUP_WARNINGS(PL_compiling.cop_warnings);
14257 CopHINTHASH_set(&PL_compiling, cophh_copy(CopHINTHASH_get(&PL_compiling)));
14258 PL_curcop = (COP*)any_dup(proto_perl->Icurcop, proto_perl);
14260 param->stashes = newAV(); /* Setup array of objects to call clone on */
14261 /* This makes no difference to the implementation, as it always pushes
14262 and shifts pointers to other SVs without changing their reference
14263 count, with the array becoming empty before it is freed. However, it
14264 makes it conceptually clear what is going on, and will avoid some
14265 work inside av.c, filling slots between AvFILL() and AvMAX() with
14266 &PL_sv_undef, and SvREFCNT_dec()ing those. */
14267 AvREAL_off(param->stashes);
14269 if (!(flags & CLONEf_COPY_STACKS)) {
14270 param->unreferenced = newAV();
14273 #ifdef PERLIO_LAYERS
14274 /* Clone PerlIO tables as soon as we can handle general xx_dup() */
14275 PerlIO_clone(aTHX_ proto_perl, param);
14278 PL_envgv = gv_dup_inc(proto_perl->Ienvgv, param);
14279 PL_incgv = gv_dup_inc(proto_perl->Iincgv, param);
14280 PL_hintgv = gv_dup_inc(proto_perl->Ihintgv, param);
14281 PL_origfilename = SAVEPV(proto_perl->Iorigfilename);
14282 PL_diehook = sv_dup_inc(proto_perl->Idiehook, param);
14283 PL_warnhook = sv_dup_inc(proto_perl->Iwarnhook, param);
14286 PL_patchlevel = sv_dup_inc(proto_perl->Ipatchlevel, param);
14287 PL_apiversion = sv_dup_inc(proto_perl->Iapiversion, param);
14288 PL_inplace = SAVEPV(proto_perl->Iinplace);
14289 PL_e_script = sv_dup_inc(proto_perl->Ie_script, param);
14291 /* magical thingies */
14293 PL_encoding = sv_dup(proto_perl->Iencoding, param);
14295 sv_setpvs(PERL_DEBUG_PAD(0), ""); /* For regex debugging. */
14296 sv_setpvs(PERL_DEBUG_PAD(1), ""); /* ext/re needs these */
14297 sv_setpvs(PERL_DEBUG_PAD(2), ""); /* even without DEBUGGING. */
14300 /* Clone the regex array */
14301 /* ORANGE FIXME for plugins, probably in the SV dup code.
14302 newSViv(PTR2IV(CALLREGDUPE(
14303 INT2PTR(REGEXP *, SvIVX(regex)), param))))
14305 PL_regex_padav = av_dup_inc(proto_perl->Iregex_padav, param);
14306 PL_regex_pad = AvARRAY(PL_regex_padav);
14308 PL_stashpadmax = proto_perl->Istashpadmax;
14309 PL_stashpadix = proto_perl->Istashpadix ;
14310 Newx(PL_stashpad, PL_stashpadmax, HV *);
14313 for (; o < PL_stashpadmax; ++o)
14314 PL_stashpad[o] = hv_dup(proto_perl->Istashpad[o], param);
14317 /* shortcuts to various I/O objects */
14318 PL_ofsgv = gv_dup_inc(proto_perl->Iofsgv, param);
14319 PL_stdingv = gv_dup(proto_perl->Istdingv, param);
14320 PL_stderrgv = gv_dup(proto_perl->Istderrgv, param);
14321 PL_defgv = gv_dup(proto_perl->Idefgv, param);
14322 PL_argvgv = gv_dup_inc(proto_perl->Iargvgv, param);
14323 PL_argvoutgv = gv_dup(proto_perl->Iargvoutgv, param);
14324 PL_argvout_stack = av_dup_inc(proto_perl->Iargvout_stack, param);
14326 /* shortcuts to regexp stuff */
14327 PL_replgv = gv_dup_inc(proto_perl->Ireplgv, param);
14329 /* shortcuts to misc objects */
14330 PL_errgv = gv_dup(proto_perl->Ierrgv, param);
14332 /* shortcuts to debugging objects */
14333 PL_DBgv = gv_dup_inc(proto_perl->IDBgv, param);
14334 PL_DBline = gv_dup_inc(proto_perl->IDBline, param);
14335 PL_DBsub = gv_dup_inc(proto_perl->IDBsub, param);
14336 PL_DBsingle = sv_dup(proto_perl->IDBsingle, param);
14337 PL_DBtrace = sv_dup(proto_perl->IDBtrace, param);
14338 PL_DBsignal = sv_dup(proto_perl->IDBsignal, param);
14340 /* symbol tables */
14341 PL_defstash = hv_dup_inc(proto_perl->Idefstash, param);
14342 PL_curstash = hv_dup_inc(proto_perl->Icurstash, param);
14343 PL_debstash = hv_dup(proto_perl->Idebstash, param);
14344 PL_globalstash = hv_dup(proto_perl->Iglobalstash, param);
14345 PL_curstname = sv_dup_inc(proto_perl->Icurstname, param);
14347 PL_beginav = av_dup_inc(proto_perl->Ibeginav, param);
14348 PL_beginav_save = av_dup_inc(proto_perl->Ibeginav_save, param);
14349 PL_checkav_save = av_dup_inc(proto_perl->Icheckav_save, param);
14350 PL_unitcheckav = av_dup_inc(proto_perl->Iunitcheckav, param);
14351 PL_unitcheckav_save = av_dup_inc(proto_perl->Iunitcheckav_save, param);
14352 PL_endav = av_dup_inc(proto_perl->Iendav, param);
14353 PL_checkav = av_dup_inc(proto_perl->Icheckav, param);
14354 PL_initav = av_dup_inc(proto_perl->Iinitav, param);
14356 PL_isarev = hv_dup_inc(proto_perl->Iisarev, param);
14358 /* subprocess state */
14359 PL_fdpid = av_dup_inc(proto_perl->Ifdpid, param);
14361 if (proto_perl->Iop_mask)
14362 PL_op_mask = SAVEPVN(proto_perl->Iop_mask, PL_maxo);
14365 /* PL_asserting = proto_perl->Iasserting; */
14367 /* current interpreter roots */
14368 PL_main_cv = cv_dup_inc(proto_perl->Imain_cv, param);
14370 PL_main_root = OpREFCNT_inc(proto_perl->Imain_root);
14373 /* runtime control stuff */
14374 PL_curcopdb = (COP*)any_dup(proto_perl->Icurcopdb, proto_perl);
14376 PL_preambleav = av_dup_inc(proto_perl->Ipreambleav, param);
14378 PL_ors_sv = sv_dup_inc(proto_perl->Iors_sv, param);
14380 /* interpreter atexit processing */
14381 PL_exitlistlen = proto_perl->Iexitlistlen;
14382 if (PL_exitlistlen) {
14383 Newx(PL_exitlist, PL_exitlistlen, PerlExitListEntry);
14384 Copy(proto_perl->Iexitlist, PL_exitlist, PL_exitlistlen, PerlExitListEntry);
14387 PL_exitlist = (PerlExitListEntry*)NULL;
14389 PL_my_cxt_size = proto_perl->Imy_cxt_size;
14390 if (PL_my_cxt_size) {
14391 Newx(PL_my_cxt_list, PL_my_cxt_size, void *);
14392 Copy(proto_perl->Imy_cxt_list, PL_my_cxt_list, PL_my_cxt_size, void *);
14393 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
14394 Newx(PL_my_cxt_keys, PL_my_cxt_size, const char *);
14395 Copy(proto_perl->Imy_cxt_keys, PL_my_cxt_keys, PL_my_cxt_size, char *);
14399 PL_my_cxt_list = (void**)NULL;
14400 #ifdef PERL_GLOBAL_STRUCT_PRIVATE
14401 PL_my_cxt_keys = (const char**)NULL;
14404 PL_modglobal = hv_dup_inc(proto_perl->Imodglobal, param);
14405 PL_custom_op_names = hv_dup_inc(proto_perl->Icustom_op_names,param);
14406 PL_custom_op_descs = hv_dup_inc(proto_perl->Icustom_op_descs,param);
14407 PL_custom_ops = hv_dup_inc(proto_perl->Icustom_ops, param);
14409 PL_compcv = cv_dup(proto_perl->Icompcv, param);
14411 PAD_CLONE_VARS(proto_perl, param);
14413 #ifdef HAVE_INTERP_INTERN
14414 sys_intern_dup(&proto_perl->Isys_intern, &PL_sys_intern);
14417 PL_DBcv = cv_dup(proto_perl->IDBcv, param);
14419 #ifdef PERL_USES_PL_PIDSTATUS
14420 PL_pidstatus = newHV(); /* XXX flag for cloning? */
14422 PL_osname = SAVEPV(proto_perl->Iosname);
14423 PL_parser = parser_dup(proto_perl->Iparser, param);
14425 /* XXX this only works if the saved cop has already been cloned */
14426 if (proto_perl->Iparser) {
14427 PL_parser->saved_curcop = (COP*)any_dup(
14428 proto_perl->Iparser->saved_curcop,
14432 PL_subname = sv_dup_inc(proto_perl->Isubname, param);
14434 #ifdef USE_LOCALE_COLLATE
14435 PL_collation_name = SAVEPV(proto_perl->Icollation_name);
14436 #endif /* USE_LOCALE_COLLATE */
14438 #ifdef USE_LOCALE_NUMERIC
14439 PL_numeric_name = SAVEPV(proto_perl->Inumeric_name);
14440 PL_numeric_radix_sv = sv_dup_inc(proto_perl->Inumeric_radix_sv, param);
14441 #endif /* !USE_LOCALE_NUMERIC */
14443 /* Unicode inversion lists */
14444 PL_Latin1 = sv_dup_inc(proto_perl->ILatin1, param);
14445 PL_UpperLatin1 = sv_dup_inc(proto_perl->IUpperLatin1, param);
14446 PL_AboveLatin1 = sv_dup_inc(proto_perl->IAboveLatin1, param);
14447 PL_InBitmap = sv_dup_inc(proto_perl->IInBitmap, param);
14449 PL_NonL1NonFinalFold = sv_dup_inc(proto_perl->INonL1NonFinalFold, param);
14450 PL_HasMultiCharFold = sv_dup_inc(proto_perl->IHasMultiCharFold, param);
14452 /* utf8 character class swashes */
14453 for (i = 0; i < POSIX_SWASH_COUNT; i++) {
14454 PL_utf8_swash_ptrs[i] = sv_dup_inc(proto_perl->Iutf8_swash_ptrs[i], param);
14456 for (i = 0; i < POSIX_CC_COUNT; i++) {
14457 PL_XPosix_ptrs[i] = sv_dup_inc(proto_perl->IXPosix_ptrs[i], param);
14459 PL_utf8_mark = sv_dup_inc(proto_perl->Iutf8_mark, param);
14460 PL_utf8_X_regular_begin = sv_dup_inc(proto_perl->Iutf8_X_regular_begin, param);
14461 PL_utf8_X_extend = sv_dup_inc(proto_perl->Iutf8_X_extend, param);
14462 PL_utf8_toupper = sv_dup_inc(proto_perl->Iutf8_toupper, param);
14463 PL_utf8_totitle = sv_dup_inc(proto_perl->Iutf8_totitle, param);
14464 PL_utf8_tolower = sv_dup_inc(proto_perl->Iutf8_tolower, param);
14465 PL_utf8_tofold = sv_dup_inc(proto_perl->Iutf8_tofold, param);
14466 PL_utf8_idstart = sv_dup_inc(proto_perl->Iutf8_idstart, param);
14467 PL_utf8_xidstart = sv_dup_inc(proto_perl->Iutf8_xidstart, param);
14468 PL_utf8_perl_idstart = sv_dup_inc(proto_perl->Iutf8_perl_idstart, param);
14469 PL_utf8_perl_idcont = sv_dup_inc(proto_perl->Iutf8_perl_idcont, param);
14470 PL_utf8_idcont = sv_dup_inc(proto_perl->Iutf8_idcont, param);
14471 PL_utf8_xidcont = sv_dup_inc(proto_perl->Iutf8_xidcont, param);
14472 PL_utf8_foldable = sv_dup_inc(proto_perl->Iutf8_foldable, param);
14473 PL_utf8_charname_begin = sv_dup_inc(proto_perl->Iutf8_charname_begin, param);
14474 PL_utf8_charname_continue = sv_dup_inc(proto_perl->Iutf8_charname_continue, param);
14476 if (proto_perl->Ipsig_pend) {
14477 Newxz(PL_psig_pend, SIG_SIZE, int);
14480 PL_psig_pend = (int*)NULL;
14483 if (proto_perl->Ipsig_name) {
14484 Newx(PL_psig_name, 2 * SIG_SIZE, SV*);
14485 sv_dup_inc_multiple(proto_perl->Ipsig_name, PL_psig_name, 2 * SIG_SIZE,
14487 PL_psig_ptr = PL_psig_name + SIG_SIZE;
14490 PL_psig_ptr = (SV**)NULL;
14491 PL_psig_name = (SV**)NULL;
14494 if (flags & CLONEf_COPY_STACKS) {
14495 Newx(PL_tmps_stack, PL_tmps_max, SV*);
14496 sv_dup_inc_multiple(proto_perl->Itmps_stack, PL_tmps_stack,
14497 PL_tmps_ix+1, param);
14499 /* next PUSHMARK() sets *(PL_markstack_ptr+1) */
14500 i = proto_perl->Imarkstack_max - proto_perl->Imarkstack;
14501 Newxz(PL_markstack, i, I32);
14502 PL_markstack_max = PL_markstack + (proto_perl->Imarkstack_max
14503 - proto_perl->Imarkstack);
14504 PL_markstack_ptr = PL_markstack + (proto_perl->Imarkstack_ptr
14505 - proto_perl->Imarkstack);
14506 Copy(proto_perl->Imarkstack, PL_markstack,
14507 PL_markstack_ptr - PL_markstack + 1, I32);
14509 /* next push_scope()/ENTER sets PL_scopestack[PL_scopestack_ix]
14510 * NOTE: unlike the others! */
14511 Newxz(PL_scopestack, PL_scopestack_max, I32);
14512 Copy(proto_perl->Iscopestack, PL_scopestack, PL_scopestack_ix, I32);
14515 Newxz(PL_scopestack_name, PL_scopestack_max, const char *);
14516 Copy(proto_perl->Iscopestack_name, PL_scopestack_name, PL_scopestack_ix, const char *);
14518 /* reset stack AV to correct length before its duped via
14519 * PL_curstackinfo */
14520 AvFILLp(proto_perl->Icurstack) =
14521 proto_perl->Istack_sp - proto_perl->Istack_base;
14523 /* NOTE: si_dup() looks at PL_markstack */
14524 PL_curstackinfo = si_dup(proto_perl->Icurstackinfo, param);
14526 /* PL_curstack = PL_curstackinfo->si_stack; */
14527 PL_curstack = av_dup(proto_perl->Icurstack, param);
14528 PL_mainstack = av_dup(proto_perl->Imainstack, param);
14530 /* next PUSHs() etc. set *(PL_stack_sp+1) */
14531 PL_stack_base = AvARRAY(PL_curstack);
14532 PL_stack_sp = PL_stack_base + (proto_perl->Istack_sp
14533 - proto_perl->Istack_base);
14534 PL_stack_max = PL_stack_base + AvMAX(PL_curstack);
14536 /*Newxz(PL_savestack, PL_savestack_max, ANY);*/
14537 PL_savestack = ss_dup(proto_perl, param);
14541 ENTER; /* perl_destruct() wants to LEAVE; */
14544 PL_statgv = gv_dup(proto_perl->Istatgv, param);
14545 PL_statname = sv_dup_inc(proto_perl->Istatname, param);
14547 PL_rs = sv_dup_inc(proto_perl->Irs, param);
14548 PL_last_in_gv = gv_dup(proto_perl->Ilast_in_gv, param);
14549 PL_defoutgv = gv_dup_inc(proto_perl->Idefoutgv, param);
14550 PL_toptarget = sv_dup_inc(proto_perl->Itoptarget, param);
14551 PL_bodytarget = sv_dup_inc(proto_perl->Ibodytarget, param);
14552 PL_formtarget = sv_dup(proto_perl->Iformtarget, param);
14554 PL_errors = sv_dup_inc(proto_perl->Ierrors, param);
14556 PL_sortcop = (OP*)any_dup(proto_perl->Isortcop, proto_perl);
14557 PL_firstgv = gv_dup_inc(proto_perl->Ifirstgv, param);
14558 PL_secondgv = gv_dup_inc(proto_perl->Isecondgv, param);
14560 PL_stashcache = newHV();
14562 PL_watchaddr = (char **) ptr_table_fetch(PL_ptr_table,
14563 proto_perl->Iwatchaddr);
14564 PL_watchok = PL_watchaddr ? * PL_watchaddr : NULL;
14565 if (PL_debug && PL_watchaddr) {
14566 PerlIO_printf(Perl_debug_log,
14567 "WATCHING: %"UVxf" cloned as %"UVxf" with value %"UVxf"\n",
14568 PTR2UV(proto_perl->Iwatchaddr), PTR2UV(PL_watchaddr),
14569 PTR2UV(PL_watchok));
14572 PL_registered_mros = hv_dup_inc(proto_perl->Iregistered_mros, param);
14573 PL_blockhooks = av_dup_inc(proto_perl->Iblockhooks, param);
14574 PL_utf8_foldclosures = hv_dup_inc(proto_perl->Iutf8_foldclosures, param);
14576 /* Call the ->CLONE method, if it exists, for each of the stashes
14577 identified by sv_dup() above.
14579 while(av_tindex(param->stashes) != -1) {
14580 HV* const stash = MUTABLE_HV(av_shift(param->stashes));
14581 GV* const cloner = gv_fetchmethod_autoload(stash, "CLONE", 0);
14582 if (cloner && GvCV(cloner)) {
14587 mXPUSHs(newSVhek(HvNAME_HEK(stash)));
14589 call_sv(MUTABLE_SV(GvCV(cloner)), G_DISCARD);
14595 if (!(flags & CLONEf_KEEP_PTR_TABLE)) {
14596 ptr_table_free(PL_ptr_table);
14597 PL_ptr_table = NULL;
14600 if (!(flags & CLONEf_COPY_STACKS)) {
14601 unreferenced_to_tmp_stack(param->unreferenced);
14604 SvREFCNT_dec(param->stashes);
14606 /* orphaned? eg threads->new inside BEGIN or use */
14607 if (PL_compcv && ! SvREFCNT(PL_compcv)) {
14608 SvREFCNT_inc_simple_void(PL_compcv);
14609 SAVEFREESV(PL_compcv);
14616 S_unreferenced_to_tmp_stack(pTHX_ AV *const unreferenced)
14618 PERL_ARGS_ASSERT_UNREFERENCED_TO_TMP_STACK;
14620 if (AvFILLp(unreferenced) > -1) {
14621 SV **svp = AvARRAY(unreferenced);
14622 SV **const last = svp + AvFILLp(unreferenced);
14626 if (SvREFCNT(*svp) == 1)
14628 } while (++svp <= last);
14630 EXTEND_MORTAL(count);
14631 svp = AvARRAY(unreferenced);
14634 if (SvREFCNT(*svp) == 1) {
14635 /* Our reference is the only one to this SV. This means that
14636 in this thread, the scalar effectively has a 0 reference.
14637 That doesn't work (cleanup never happens), so donate our
14638 reference to it onto the save stack. */
14639 PL_tmps_stack[++PL_tmps_ix] = *svp;
14641 /* As an optimisation, because we are already walking the
14642 entire array, instead of above doing either
14643 SvREFCNT_inc(*svp) or *svp = &PL_sv_undef, we can instead
14644 release our reference to the scalar, so that at the end of
14645 the array owns zero references to the scalars it happens to
14646 point to. We are effectively converting the array from
14647 AvREAL() on to AvREAL() off. This saves the av_clear()
14648 (triggered by the SvREFCNT_dec(unreferenced) below) from
14649 walking the array a second time. */
14650 SvREFCNT_dec(*svp);
14653 } while (++svp <= last);
14654 AvREAL_off(unreferenced);
14656 SvREFCNT_dec_NN(unreferenced);
14660 Perl_clone_params_del(CLONE_PARAMS *param)
14662 /* This seemingly funky ordering keeps the build with PERL_GLOBAL_STRUCT
14664 PerlInterpreter *const to = param->new_perl;
14666 PerlInterpreter *const was = PERL_GET_THX;
14668 PERL_ARGS_ASSERT_CLONE_PARAMS_DEL;
14674 SvREFCNT_dec(param->stashes);
14675 if (param->unreferenced)
14676 unreferenced_to_tmp_stack(param->unreferenced);
14686 Perl_clone_params_new(PerlInterpreter *const from, PerlInterpreter *const to)
14689 /* Need to play this game, as newAV() can call safesysmalloc(), and that
14690 does a dTHX; to get the context from thread local storage.
14691 FIXME - under PERL_CORE Newx(), Safefree() and friends should expand to
14692 a version that passes in my_perl. */
14693 PerlInterpreter *const was = PERL_GET_THX;
14694 CLONE_PARAMS *param;
14696 PERL_ARGS_ASSERT_CLONE_PARAMS_NEW;
14702 /* Given that we've set the context, we can do this unshared. */
14703 Newx(param, 1, CLONE_PARAMS);
14706 param->proto_perl = from;
14707 param->new_perl = to;
14708 param->stashes = (AV *)Perl_newSV_type(to, SVt_PVAV);
14709 AvREAL_off(param->stashes);
14710 param->unreferenced = (AV *)Perl_newSV_type(to, SVt_PVAV);
14718 #endif /* USE_ITHREADS */
14721 Perl_init_constants(pTHX)
14723 SvREFCNT(&PL_sv_undef) = SvREFCNT_IMMORTAL;
14724 SvFLAGS(&PL_sv_undef) = SVf_READONLY|SVt_NULL;
14725 SvANY(&PL_sv_undef) = NULL;
14727 SvANY(&PL_sv_no) = new_XPVNV();
14728 SvREFCNT(&PL_sv_no) = SvREFCNT_IMMORTAL;
14729 SvFLAGS(&PL_sv_no) = SVt_PVNV|SVf_READONLY
14730 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
14733 SvANY(&PL_sv_yes) = new_XPVNV();
14734 SvREFCNT(&PL_sv_yes) = SvREFCNT_IMMORTAL;
14735 SvFLAGS(&PL_sv_yes) = SVt_PVNV|SVf_READONLY
14736 |SVp_IOK|SVf_IOK|SVp_NOK|SVf_NOK
14739 SvPV_set(&PL_sv_no, (char*)PL_No);
14740 SvCUR_set(&PL_sv_no, 0);
14741 SvLEN_set(&PL_sv_no, 0);
14742 SvIV_set(&PL_sv_no, 0);
14743 SvNV_set(&PL_sv_no, 0);
14745 SvPV_set(&PL_sv_yes, (char*)PL_Yes);
14746 SvCUR_set(&PL_sv_yes, 1);
14747 SvLEN_set(&PL_sv_yes, 0);
14748 SvIV_set(&PL_sv_yes, 1);
14749 SvNV_set(&PL_sv_yes, 1);
14753 =head1 Unicode Support
14755 =for apidoc sv_recode_to_utf8
14757 The encoding is assumed to be an Encode object, on entry the PV
14758 of the sv is assumed to be octets in that encoding, and the sv
14759 will be converted into Unicode (and UTF-8).
14761 If the sv already is UTF-8 (or if it is not POK), or if the encoding
14762 is not a reference, nothing is done to the sv. If the encoding is not
14763 an C<Encode::XS> Encoding object, bad things will happen.
14764 (See F<lib/encoding.pm> and L<Encode>.)
14766 The PV of the sv is returned.
14771 Perl_sv_recode_to_utf8(pTHX_ SV *sv, SV *encoding)
14773 PERL_ARGS_ASSERT_SV_RECODE_TO_UTF8;
14775 if (SvPOK(sv) && !SvUTF8(sv) && !IN_BYTES && SvROK(encoding)) {
14784 if (SvPADTMP(nsv)) {
14785 nsv = sv_newmortal();
14786 SvSetSV_nosteal(nsv, sv);
14795 Passing sv_yes is wrong - it needs to be or'ed set of constants
14796 for Encode::XS, while UTf-8 decode (currently) assumes a true value means
14797 remove converted chars from source.
14799 Both will default the value - let them.
14801 XPUSHs(&PL_sv_yes);
14804 call_method("decode", G_SCALAR);
14808 s = SvPV_const(uni, len);
14809 if (s != SvPVX_const(sv)) {
14810 SvGROW(sv, len + 1);
14811 Move(s, SvPVX(sv), len + 1, char);
14812 SvCUR_set(sv, len);
14817 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
14818 /* clear pos and any utf8 cache */
14819 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
14822 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
14823 magic_setutf8(sv,mg); /* clear UTF8 cache */
14828 return SvPOKp(sv) ? SvPVX(sv) : NULL;
14832 =for apidoc sv_cat_decode
14834 The encoding is assumed to be an Encode object, the PV of the ssv is
14835 assumed to be octets in that encoding and decoding the input starts
14836 from the position which (PV + *offset) pointed to. The dsv will be
14837 concatenated the decoded UTF-8 string from ssv. Decoding will terminate
14838 when the string tstr appears in decoding output or the input ends on
14839 the PV of the ssv. The value which the offset points will be modified
14840 to the last input position on the ssv.
14842 Returns TRUE if the terminator was found, else returns FALSE.
14847 Perl_sv_cat_decode(pTHX_ SV *dsv, SV *encoding,
14848 SV *ssv, int *offset, char *tstr, int tlen)
14852 PERL_ARGS_ASSERT_SV_CAT_DECODE;
14854 if (SvPOK(ssv) && SvPOK(dsv) && SvROK(encoding) && offset) {
14865 offsv = newSViv(*offset);
14867 mPUSHp(tstr, tlen);
14869 call_method("cat_decode", G_SCALAR);
14871 ret = SvTRUE(TOPs);
14872 *offset = SvIV(offsv);
14878 Perl_croak(aTHX_ "Invalid argument to sv_cat_decode");
14883 /* ---------------------------------------------------------------------
14885 * support functions for report_uninit()
14888 /* the maxiumum size of array or hash where we will scan looking
14889 * for the undefined element that triggered the warning */
14891 #define FUV_MAX_SEARCH_SIZE 1000
14893 /* Look for an entry in the hash whose value has the same SV as val;
14894 * If so, return a mortal copy of the key. */
14897 S_find_hash_subscript(pTHX_ const HV *const hv, const SV *const val)
14903 PERL_ARGS_ASSERT_FIND_HASH_SUBSCRIPT;
14905 if (!hv || SvMAGICAL(hv) || !HvARRAY(hv) ||
14906 (HvTOTALKEYS(hv) > FUV_MAX_SEARCH_SIZE))
14909 array = HvARRAY(hv);
14911 for (i=HvMAX(hv); i>=0; i--) {
14913 for (entry = array[i]; entry; entry = HeNEXT(entry)) {
14914 if (HeVAL(entry) != val)
14916 if ( HeVAL(entry) == &PL_sv_undef ||
14917 HeVAL(entry) == &PL_sv_placeholder)
14921 if (HeKLEN(entry) == HEf_SVKEY)
14922 return sv_mortalcopy(HeKEY_sv(entry));
14923 return sv_2mortal(newSVhek(HeKEY_hek(entry)));
14929 /* Look for an entry in the array whose value has the same SV as val;
14930 * If so, return the index, otherwise return -1. */
14933 S_find_array_subscript(pTHX_ const AV *const av, const SV *const val)
14935 PERL_ARGS_ASSERT_FIND_ARRAY_SUBSCRIPT;
14937 if (!av || SvMAGICAL(av) || !AvARRAY(av) ||
14938 (AvFILLp(av) > FUV_MAX_SEARCH_SIZE))
14941 if (val != &PL_sv_undef) {
14942 SV ** const svp = AvARRAY(av);
14945 for (i=AvFILLp(av); i>=0; i--)
14952 /* varname(): return the name of a variable, optionally with a subscript.
14953 * If gv is non-zero, use the name of that global, along with gvtype (one
14954 * of "$", "@", "%"); otherwise use the name of the lexical at pad offset
14955 * targ. Depending on the value of the subscript_type flag, return:
14958 #define FUV_SUBSCRIPT_NONE 1 /* "@foo" */
14959 #define FUV_SUBSCRIPT_ARRAY 2 /* "$foo[aindex]" */
14960 #define FUV_SUBSCRIPT_HASH 3 /* "$foo{keyname}" */
14961 #define FUV_SUBSCRIPT_WITHIN 4 /* "within @foo" */
14964 Perl_varname(pTHX_ const GV *const gv, const char gvtype, PADOFFSET targ,
14965 const SV *const keyname, I32 aindex, int subscript_type)
14968 SV * const name = sv_newmortal();
14969 if (gv && isGV(gv)) {
14971 buffer[0] = gvtype;
14974 /* as gv_fullname4(), but add literal '^' for $^FOO names */
14976 gv_fullname4(name, gv, buffer, 0);
14978 if ((unsigned int)SvPVX(name)[1] <= 26) {
14980 buffer[1] = SvPVX(name)[1] + 'A' - 1;
14982 /* Swap the 1 unprintable control character for the 2 byte pretty
14983 version - ie substr($name, 1, 1) = $buffer; */
14984 sv_insert(name, 1, 1, buffer, 2);
14988 CV * const cv = gv ? ((CV *)gv) : find_runcv(NULL);
14992 assert(!cv || SvTYPE(cv) == SVt_PVCV || SvTYPE(cv) == SVt_PVFM);
14994 if (!cv || !CvPADLIST(cv))
14996 av = *PadlistARRAY(CvPADLIST(cv));
14997 sv = *av_fetch(av, targ, FALSE);
14998 sv_setsv_flags(name, sv, 0);
15001 if (subscript_type == FUV_SUBSCRIPT_HASH) {
15002 SV * const sv = newSV(0);
15003 *SvPVX(name) = '$';
15004 Perl_sv_catpvf(aTHX_ name, "{%s}",
15005 pv_pretty(sv, SvPVX_const(keyname), SvCUR(keyname), 32, NULL, NULL,
15006 PERL_PV_PRETTY_DUMP | PERL_PV_ESCAPE_UNI_DETECT ));
15007 SvREFCNT_dec_NN(sv);
15009 else if (subscript_type == FUV_SUBSCRIPT_ARRAY) {
15010 *SvPVX(name) = '$';
15011 Perl_sv_catpvf(aTHX_ name, "[%"IVdf"]", (IV)aindex);
15013 else if (subscript_type == FUV_SUBSCRIPT_WITHIN) {
15014 /* We know that name has no magic, so can use 0 instead of SV_GMAGIC */
15015 Perl_sv_insert_flags(aTHX_ name, 0, 0, STR_WITH_LEN("within "), 0);
15023 =for apidoc find_uninit_var
15025 Find the name of the undefined variable (if any) that caused the operator
15026 to issue a "Use of uninitialized value" warning.
15027 If match is true, only return a name if its value matches uninit_sv.
15028 So roughly speaking, if a unary operator (such as OP_COS) generates a
15029 warning, then following the direct child of the op may yield an
15030 OP_PADSV or OP_GV that gives the name of the undefined variable. On the
15031 other hand, with OP_ADD there are two branches to follow, so we only print
15032 the variable name if we get an exact match.
15034 The name is returned as a mortal SV.
15036 Assumes that PL_op is the op that originally triggered the error, and that
15037 PL_comppad/PL_curpad points to the currently executing pad.
15043 S_find_uninit_var(pTHX_ const OP *const obase, const SV *const uninit_sv,
15049 const OP *o, *o2, *kid;
15051 if (!obase || (match && (!uninit_sv || uninit_sv == &PL_sv_undef ||
15052 uninit_sv == &PL_sv_placeholder)))
15055 switch (obase->op_type) {
15062 const bool pad = ( obase->op_type == OP_PADAV
15063 || obase->op_type == OP_PADHV
15064 || obase->op_type == OP_PADRANGE
15067 const bool hash = ( obase->op_type == OP_PADHV
15068 || obase->op_type == OP_RV2HV
15069 || (obase->op_type == OP_PADRANGE
15070 && SvTYPE(PAD_SVl(obase->op_targ)) == SVt_PVHV)
15074 int subscript_type = FUV_SUBSCRIPT_WITHIN;
15076 if (pad) { /* @lex, %lex */
15077 sv = PAD_SVl(obase->op_targ);
15081 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
15082 /* @global, %global */
15083 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
15086 sv = hash ? MUTABLE_SV(GvHV(gv)): MUTABLE_SV(GvAV(gv));
15088 else if (obase == PL_op) /* @{expr}, %{expr} */
15089 return find_uninit_var(cUNOPx(obase)->op_first,
15091 else /* @{expr}, %{expr} as a sub-expression */
15095 /* attempt to find a match within the aggregate */
15097 keysv = find_hash_subscript((const HV*)sv, uninit_sv);
15099 subscript_type = FUV_SUBSCRIPT_HASH;
15102 index = find_array_subscript((const AV *)sv, uninit_sv);
15104 subscript_type = FUV_SUBSCRIPT_ARRAY;
15107 if (match && subscript_type == FUV_SUBSCRIPT_WITHIN)
15110 return varname(gv, hash ? '%' : '@', obase->op_targ,
15111 keysv, index, subscript_type);
15115 if (cUNOPx(obase)->op_first->op_type == OP_GV) {
15117 gv = cGVOPx_gv(cUNOPx(obase)->op_first);
15118 if (!gv || !GvSTASH(gv))
15120 if (match && (GvSV(gv) != uninit_sv))
15122 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
15125 return find_uninit_var(cUNOPx(obase)->op_first, uninit_sv, 1);
15128 if (match && PAD_SVl(obase->op_targ) != uninit_sv)
15130 return varname(NULL, '$', obase->op_targ,
15131 NULL, 0, FUV_SUBSCRIPT_NONE);
15134 gv = cGVOPx_gv(obase);
15135 if (!gv || (match && GvSV(gv) != uninit_sv) || !GvSTASH(gv))
15137 return varname(gv, '$', 0, NULL, 0, FUV_SUBSCRIPT_NONE);
15139 case OP_AELEMFAST_LEX:
15142 AV *av = MUTABLE_AV(PAD_SV(obase->op_targ));
15143 if (!av || SvRMAGICAL(av))
15145 svp = av_fetch(av, (I8)obase->op_private, FALSE);
15146 if (!svp || *svp != uninit_sv)
15149 return varname(NULL, '$', obase->op_targ,
15150 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
15153 gv = cGVOPx_gv(obase);
15158 AV *const av = GvAV(gv);
15159 if (!av || SvRMAGICAL(av))
15161 svp = av_fetch(av, (I8)obase->op_private, FALSE);
15162 if (!svp || *svp != uninit_sv)
15165 return varname(gv, '$', 0,
15166 NULL, (I8)obase->op_private, FUV_SUBSCRIPT_ARRAY);
15168 NOT_REACHED; /* NOTREACHED */
15171 o = cUNOPx(obase)->op_first;
15172 if (!o || o->op_type != OP_NULL ||
15173 ! (o->op_targ == OP_AELEM || o->op_targ == OP_HELEM))
15175 return find_uninit_var(cBINOPo->op_last, uninit_sv, match);
15180 bool negate = FALSE;
15182 if (PL_op == obase)
15183 /* $a[uninit_expr] or $h{uninit_expr} */
15184 return find_uninit_var(cBINOPx(obase)->op_last, uninit_sv, match);
15187 o = cBINOPx(obase)->op_first;
15188 kid = cBINOPx(obase)->op_last;
15190 /* get the av or hv, and optionally the gv */
15192 if (o->op_type == OP_PADAV || o->op_type == OP_PADHV) {
15193 sv = PAD_SV(o->op_targ);
15195 else if ((o->op_type == OP_RV2AV || o->op_type == OP_RV2HV)
15196 && cUNOPo->op_first->op_type == OP_GV)
15198 gv = cGVOPx_gv(cUNOPo->op_first);
15202 == OP_RV2HV ? MUTABLE_SV(GvHV(gv)) : MUTABLE_SV(GvAV(gv));
15207 if (kid && kid->op_type == OP_NEGATE) {
15209 kid = cUNOPx(kid)->op_first;
15212 if (kid && kid->op_type == OP_CONST && SvOK(cSVOPx_sv(kid))) {
15213 /* index is constant */
15216 kidsv = sv_2mortal(newSVpvs("-"));
15217 sv_catsv(kidsv, cSVOPx_sv(kid));
15220 kidsv = cSVOPx_sv(kid);
15224 if (obase->op_type == OP_HELEM) {
15225 HE* he = hv_fetch_ent(MUTABLE_HV(sv), kidsv, 0, 0);
15226 if (!he || HeVAL(he) != uninit_sv)
15230 SV * const opsv = cSVOPx_sv(kid);
15231 const IV opsviv = SvIV(opsv);
15232 SV * const * const svp = av_fetch(MUTABLE_AV(sv),
15233 negate ? - opsviv : opsviv,
15235 if (!svp || *svp != uninit_sv)
15239 if (obase->op_type == OP_HELEM)
15240 return varname(gv, '%', o->op_targ,
15241 kidsv, 0, FUV_SUBSCRIPT_HASH);
15243 return varname(gv, '@', o->op_targ, NULL,
15244 negate ? - SvIV(cSVOPx_sv(kid)) : SvIV(cSVOPx_sv(kid)),
15245 FUV_SUBSCRIPT_ARRAY);
15248 /* index is an expression;
15249 * attempt to find a match within the aggregate */
15250 if (obase->op_type == OP_HELEM) {
15251 SV * const keysv = find_hash_subscript((const HV*)sv, uninit_sv);
15253 return varname(gv, '%', o->op_targ,
15254 keysv, 0, FUV_SUBSCRIPT_HASH);
15258 = find_array_subscript((const AV *)sv, uninit_sv);
15260 return varname(gv, '@', o->op_targ,
15261 NULL, index, FUV_SUBSCRIPT_ARRAY);
15266 (o->op_type == OP_PADAV || o->op_type == OP_RV2AV)
15268 o->op_targ, NULL, 0, FUV_SUBSCRIPT_WITHIN);
15270 NOT_REACHED; /* NOTREACHED */
15274 /* only examine RHS */
15275 return find_uninit_var(cBINOPx(obase)->op_first, uninit_sv, match);
15278 o = cUNOPx(obase)->op_first;
15279 if ( o->op_type == OP_PUSHMARK
15280 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)
15284 if (!OP_HAS_SIBLING(o)) {
15285 /* one-arg version of open is highly magical */
15287 if (o->op_type == OP_GV) { /* open FOO; */
15289 if (match && GvSV(gv) != uninit_sv)
15291 return varname(gv, '$', 0,
15292 NULL, 0, FUV_SUBSCRIPT_NONE);
15294 /* other possibilities not handled are:
15295 * open $x; or open my $x; should return '${*$x}'
15296 * open expr; should return '$'.expr ideally
15302 /* ops where $_ may be an implicit arg */
15307 if ( !(obase->op_flags & OPf_STACKED)) {
15308 if (uninit_sv == ((obase->op_private & OPpTARGET_MY)
15309 ? PAD_SVl(obase->op_targ)
15312 sv = sv_newmortal();
15313 sv_setpvs(sv, "$_");
15322 match = 1; /* print etc can return undef on defined args */
15323 /* skip filehandle as it can't produce 'undef' warning */
15324 o = cUNOPx(obase)->op_first;
15325 if ((obase->op_flags & OPf_STACKED)
15327 ( o->op_type == OP_PUSHMARK
15328 || (o->op_type == OP_NULL && o->op_targ == OP_PUSHMARK)))
15329 o = OP_SIBLING(OP_SIBLING(o));
15333 case OP_ENTEREVAL: /* could be eval $undef or $x='$undef'; eval $x */
15334 case OP_CUSTOM: /* XS or custom code could trigger random warnings */
15336 /* the following ops are capable of returning PL_sv_undef even for
15337 * defined arg(s) */
15356 case OP_GETPEERNAME:
15404 case OP_SMARTMATCH:
15413 /* XXX tmp hack: these two may call an XS sub, and currently
15414 XS subs don't have a SUB entry on the context stack, so CV and
15415 pad determination goes wrong, and BAD things happen. So, just
15416 don't try to determine the value under those circumstances.
15417 Need a better fix at dome point. DAPM 11/2007 */
15423 GV * const gv = gv_fetchpvs(".", GV_NOTQUAL, SVt_PV);
15424 if (gv && GvSV(gv) == uninit_sv)
15425 return newSVpvs_flags("$.", SVs_TEMP);
15430 /* def-ness of rval pos() is independent of the def-ness of its arg */
15431 if ( !(obase->op_flags & OPf_MOD))
15436 if (SvROK(PL_rs) && uninit_sv == SvRV(PL_rs))
15437 return newSVpvs_flags("${$/}", SVs_TEMP);
15442 if (!(obase->op_flags & OPf_KIDS))
15444 o = cUNOPx(obase)->op_first;
15450 /* This loop checks all the kid ops, skipping any that cannot pos-
15451 * sibly be responsible for the uninitialized value; i.e., defined
15452 * constants and ops that return nothing. If there is only one op
15453 * left that is not skipped, then we *know* it is responsible for
15454 * the uninitialized value. If there is more than one op left, we
15455 * have to look for an exact match in the while() loop below.
15456 * Note that we skip padrange, because the individual pad ops that
15457 * it replaced are still in the tree, so we work on them instead.
15460 for (kid=o; kid; kid = OP_SIBLING(kid)) {
15461 const OPCODE type = kid->op_type;
15462 if ( (type == OP_CONST && SvOK(cSVOPx_sv(kid)))
15463 || (type == OP_NULL && ! (kid->op_flags & OPf_KIDS))
15464 || (type == OP_PUSHMARK)
15465 || (type == OP_PADRANGE)
15469 if (o2) { /* more than one found */
15476 return find_uninit_var(o2, uninit_sv, match);
15478 /* scan all args */
15480 sv = find_uninit_var(o, uninit_sv, 1);
15492 =for apidoc report_uninit
15494 Print appropriate "Use of uninitialized variable" warning.
15500 Perl_report_uninit(pTHX_ const SV *uninit_sv)
15503 SV* varname = NULL;
15504 if (uninit_sv && PL_curpad) {
15505 varname = find_uninit_var(PL_op, uninit_sv,0);
15507 sv_insert(varname, 0, 0, " ", 1);
15509 /* PL_warn_uninit_sv is constant */
15510 GCC_DIAG_IGNORE(-Wformat-nonliteral);
15511 /* diag_listed_as: Use of uninitialized value%s */
15512 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit_sv,
15513 SVfARG(varname ? varname : &PL_sv_no),
15514 " in ", OP_DESC(PL_op));
15518 /* PL_warn_uninit is constant */
15519 GCC_DIAG_IGNORE(-Wformat-nonliteral);
15520 Perl_warner(aTHX_ packWARN(WARN_UNINITIALIZED), PL_warn_uninit,
15528 * c-indentation-style: bsd
15529 * c-basic-offset: 4
15530 * indent-tabs-mode: nil
15533 * ex: set ts=8 sts=4 sw=4 et: