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
36 # if __STDC_VERSION__ >= 199901L && !defined(VMS)
47 /* Missing proto on LynxOS */
48 char *gconvert(double, int, int, char *);
51 #ifdef PERL_UTF8_CACHE_ASSERT
52 /* if adding more checks watch out for the following tests:
53 * t/op/index.t t/op/length.t t/op/pat.t t/op/substr.t
54 * lib/utf8.t lib/Unicode/Collate/t/index.t
57 # define ASSERT_UTF8_CACHE(cache) \
58 STMT_START { if (cache) { assert((cache)[0] <= (cache)[1]); \
59 assert((cache)[2] <= (cache)[3]); \
60 assert((cache)[3] <= (cache)[1]);} \
63 # define ASSERT_UTF8_CACHE(cache) NOOP
66 #ifdef PERL_OLD_COPY_ON_WRITE
67 #define SV_COW_NEXT_SV(sv) INT2PTR(SV *,SvUVX(sv))
68 #define SV_COW_NEXT_SV_SET(current,next) SvUV_set(current, PTR2UV(next))
69 /* This is a pessimistic view. Scalar must be purely a read-write PV to copy-
73 /* ============================================================================
75 =head1 Allocation and deallocation of SVs.
77 An SV (or AV, HV, etc.) is allocated in two parts: the head (struct
78 sv, av, hv...) contains type and reference count information, and for
79 many types, a pointer to the body (struct xrv, xpv, xpviv...), which
80 contains fields specific to each type. Some types store all they need
81 in the head, so don't have a body.
83 In all but the most memory-paranoid configurations (ex: PURIFY), heads
84 and bodies are allocated out of arenas, which by default are
85 approximately 4K chunks of memory parcelled up into N heads or bodies.
86 Sv-bodies are allocated by their sv-type, guaranteeing size
87 consistency needed to allocate safely from arrays.
89 For SV-heads, the first slot in each arena is reserved, and holds a
90 link to the next arena, some flags, and a note of the number of slots.
91 Snaked through each arena chain is a linked list of free items; when
92 this becomes empty, an extra arena is allocated and divided up into N
93 items which are threaded into the free list.
95 SV-bodies are similar, but they use arena-sets by default, which
96 separate the link and info from the arena itself, and reclaim the 1st
97 slot in the arena. SV-bodies are further described later.
99 The following global variables are associated with arenas:
101 PL_sv_arenaroot pointer to list of SV arenas
102 PL_sv_root pointer to list of free SV structures
104 PL_body_arenas head of linked-list of body arenas
105 PL_body_roots[] array of pointers to list of free bodies of svtype
106 arrays are indexed by the svtype needed
108 A few special SV heads are not allocated from an arena, but are
109 instead directly created in the interpreter structure, eg PL_sv_undef.
110 The size of arenas can be changed from the default by setting
111 PERL_ARENA_SIZE appropriately at compile time.
113 The SV arena serves the secondary purpose of allowing still-live SVs
114 to be located and destroyed during final cleanup.
116 At the lowest level, the macros new_SV() and del_SV() grab and free
117 an SV head. (If debugging with -DD, del_SV() calls the function S_del_sv()
118 to return the SV to the free list with error checking.) new_SV() calls
119 more_sv() / sv_add_arena() to add an extra arena if the free list is empty.
120 SVs in the free list have their SvTYPE field set to all ones.
122 At the time of very final cleanup, sv_free_arenas() is called from
123 perl_destruct() to physically free all the arenas allocated since the
124 start of the interpreter.
126 The function visit() scans the SV arenas list, and calls a specified
127 function for each SV it finds which is still live - ie which has an SvTYPE
128 other than all 1's, and a non-zero SvREFCNT. visit() is used by the
129 following functions (specified as [function that calls visit()] / [function
130 called by visit() for each SV]):
132 sv_report_used() / do_report_used()
133 dump all remaining SVs (debugging aid)
135 sv_clean_objs() / do_clean_objs(),do_clean_named_objs(),
136 do_clean_named_io_objs(),do_curse()
137 Attempt to free all objects pointed to by RVs,
138 try to do the same for all objects indir-
139 ectly referenced by typeglobs too, and
140 then do a final sweep, cursing any
141 objects that remain. Called once from
142 perl_destruct(), prior to calling sv_clean_all()
145 sv_clean_all() / do_clean_all()
146 SvREFCNT_dec(sv) each remaining SV, possibly
147 triggering an sv_free(). It also sets the
148 SVf_BREAK flag on the SV to indicate that the
149 refcnt has been artificially lowered, and thus
150 stopping sv_free() from giving spurious warnings
151 about SVs which unexpectedly have a refcnt
152 of zero. called repeatedly from perl_destruct()
153 until there are no SVs left.
155 =head2 Arena allocator API Summary
157 Private API to rest of sv.c
161 new_XPVNV(), del_XPVGV(),
166 sv_report_used(), sv_clean_objs(), sv_clean_all(), sv_free_arenas()
170 * ========================================================================= */
173 * "A time to plant, and a time to uproot what was planted..."
177 # define MEM_LOG_NEW_SV(sv, file, line, func) \
178 Perl_mem_log_new_sv(sv, file, line, func)
179 # define MEM_LOG_DEL_SV(sv, file, line, func) \
180 Perl_mem_log_del_sv(sv, file, line, func)
182 # define MEM_LOG_NEW_SV(sv, file, line, func) NOOP
183 # define MEM_LOG_DEL_SV(sv, file, line, func) NOOP
186 #ifdef DEBUG_LEAKING_SCALARS
187 # define FREE_SV_DEBUG_FILE(sv) STMT_START { \
188 if ((sv)->sv_debug_file) PerlMemShared_free((sv)->sv_debug_file); \
190 # define DEBUG_SV_SERIAL(sv) \
191 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) del_SV\n", \
192 PTR2UV(sv), (long)(sv)->sv_debug_serial))
194 # define FREE_SV_DEBUG_FILE(sv)
195 # define DEBUG_SV_SERIAL(sv) NOOP
199 # define SvARENA_CHAIN(sv) ((sv)->sv_u.svu_rv)
200 # define SvARENA_CHAIN_SET(sv,val) (sv)->sv_u.svu_rv = MUTABLE_SV((val))
201 /* Whilst I'd love to do this, it seems that things like to check on
203 # define POSION_SV_HEAD(sv) PoisonNew(sv, 1, struct STRUCT_SV)
205 # define POSION_SV_HEAD(sv) PoisonNew(&SvANY(sv), 1, void *), \
206 PoisonNew(&SvREFCNT(sv), 1, U32)
208 # define SvARENA_CHAIN(sv) SvANY(sv)
209 # define SvARENA_CHAIN_SET(sv,val) SvANY(sv) = (void *)(val)
210 # define POSION_SV_HEAD(sv)
213 /* Mark an SV head as unused, and add to free list.
215 * If SVf_BREAK is set, skip adding it to the free list, as this SV had
216 * its refcount artificially decremented during global destruction, so
217 * there may be dangling pointers to it. The last thing we want in that
218 * case is for it to be reused. */
220 #define plant_SV(p) \
222 const U32 old_flags = SvFLAGS(p); \
223 MEM_LOG_DEL_SV(p, __FILE__, __LINE__, FUNCTION__); \
224 DEBUG_SV_SERIAL(p); \
225 FREE_SV_DEBUG_FILE(p); \
227 SvFLAGS(p) = SVTYPEMASK; \
228 if (!(old_flags & SVf_BREAK)) { \
229 SvARENA_CHAIN_SET(p, PL_sv_root); \
235 #define uproot_SV(p) \
238 PL_sv_root = MUTABLE_SV(SvARENA_CHAIN(p)); \
243 /* make some more SVs by adding another arena */
250 char *chunk; /* must use New here to match call to */
251 Newx(chunk,PERL_ARENA_SIZE,char); /* Safefree() in sv_free_arenas() */
252 sv_add_arena(chunk, PERL_ARENA_SIZE, 0);
257 /* new_SV(): return a new, empty SV head */
259 #ifdef DEBUG_LEAKING_SCALARS
260 /* provide a real function for a debugger to play with */
262 S_new_SV(pTHX_ const char *file, int line, const char *func)
269 sv = S_more_sv(aTHX);
273 sv->sv_debug_optype = PL_op ? PL_op->op_type : 0;
274 sv->sv_debug_line = (U16) (PL_parser && PL_parser->copline != NOLINE
280 sv->sv_debug_inpad = 0;
281 sv->sv_debug_parent = NULL;
282 sv->sv_debug_file = PL_curcop ? savesharedpv(CopFILE(PL_curcop)): NULL;
284 sv->sv_debug_serial = PL_sv_serial++;
286 MEM_LOG_NEW_SV(sv, file, line, func);
287 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) new_SV (from %s:%d [%s])\n",
288 PTR2UV(sv), (long)sv->sv_debug_serial, file, line, func));
292 # define new_SV(p) (p)=S_new_SV(aTHX_ __FILE__, __LINE__, FUNCTION__)
300 (p) = S_more_sv(aTHX); \
304 MEM_LOG_NEW_SV(p, __FILE__, __LINE__, FUNCTION__); \
309 /* del_SV(): return an empty SV head to the free list */
322 S_del_sv(pTHX_ SV *p)
326 PERL_ARGS_ASSERT_DEL_SV;
331 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
332 const SV * const sv = sva + 1;
333 const SV * const svend = &sva[SvREFCNT(sva)];
334 if (p >= sv && p < svend) {
340 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
341 "Attempt to free non-arena SV: 0x%"UVxf
342 pTHX__FORMAT, PTR2UV(p) pTHX__VALUE);
349 #else /* ! DEBUGGING */
351 #define del_SV(p) plant_SV(p)
353 #endif /* DEBUGGING */
357 =head1 SV Manipulation Functions
359 =for apidoc sv_add_arena
361 Given a chunk of memory, link it to the head of the list of arenas,
362 and split it into a list of free SVs.
368 S_sv_add_arena(pTHX_ char *const ptr, const U32 size, const U32 flags)
371 SV *const sva = MUTABLE_SV(ptr);
375 PERL_ARGS_ASSERT_SV_ADD_ARENA;
377 /* The first SV in an arena isn't an SV. */
378 SvANY(sva) = (void *) PL_sv_arenaroot; /* ptr to next arena */
379 SvREFCNT(sva) = size / sizeof(SV); /* number of SV slots */
380 SvFLAGS(sva) = flags; /* FAKE if not to be freed */
382 PL_sv_arenaroot = sva;
383 PL_sv_root = sva + 1;
385 svend = &sva[SvREFCNT(sva) - 1];
388 SvARENA_CHAIN_SET(sv, (sv + 1));
392 /* Must always set typemask because it's always checked in on cleanup
393 when the arenas are walked looking for objects. */
394 SvFLAGS(sv) = SVTYPEMASK;
397 SvARENA_CHAIN_SET(sv, 0);
401 SvFLAGS(sv) = SVTYPEMASK;
404 /* visit(): call the named function for each non-free SV in the arenas
405 * whose flags field matches the flags/mask args. */
408 S_visit(pTHX_ SVFUNC_t f, const U32 flags, const U32 mask)
414 PERL_ARGS_ASSERT_VISIT;
416 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
417 const SV * const svend = &sva[SvREFCNT(sva)];
419 for (sv = sva + 1; sv < svend; ++sv) {
420 if (SvTYPE(sv) != (svtype)SVTYPEMASK
421 && (sv->sv_flags & mask) == flags
434 /* called by sv_report_used() for each live SV */
437 do_report_used(pTHX_ SV *const sv)
439 if (SvTYPE(sv) != (svtype)SVTYPEMASK) {
440 PerlIO_printf(Perl_debug_log, "****\n");
447 =for apidoc sv_report_used
449 Dump the contents of all SVs not yet freed (debugging aid).
455 Perl_sv_report_used(pTHX)
458 visit(do_report_used, 0, 0);
464 /* called by sv_clean_objs() for each live SV */
467 do_clean_objs(pTHX_ SV *const ref)
472 SV * const target = SvRV(ref);
473 if (SvOBJECT(target)) {
474 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning object ref:\n "), sv_dump(ref)));
475 if (SvWEAKREF(ref)) {
476 sv_del_backref(target, ref);
482 SvREFCNT_dec(target);
489 /* clear any slots in a GV which hold objects - except IO;
490 * called by sv_clean_objs() for each live GV */
493 do_clean_named_objs(pTHX_ SV *const sv)
497 assert(SvTYPE(sv) == SVt_PVGV);
498 assert(isGV_with_GP(sv));
502 /* freeing GP entries may indirectly free the current GV;
503 * hold onto it while we mess with the GP slots */
506 if ( ((obj = GvSV(sv) )) && SvOBJECT(obj)) {
507 DEBUG_D((PerlIO_printf(Perl_debug_log,
508 "Cleaning named glob SV object:\n "), sv_dump(obj)));
512 if ( ((obj = MUTABLE_SV(GvAV(sv)) )) && SvOBJECT(obj)) {
513 DEBUG_D((PerlIO_printf(Perl_debug_log,
514 "Cleaning named glob AV object:\n "), sv_dump(obj)));
518 if ( ((obj = MUTABLE_SV(GvHV(sv)) )) && SvOBJECT(obj)) {
519 DEBUG_D((PerlIO_printf(Perl_debug_log,
520 "Cleaning named glob HV object:\n "), sv_dump(obj)));
524 if ( ((obj = MUTABLE_SV(GvCV(sv)) )) && SvOBJECT(obj)) {
525 DEBUG_D((PerlIO_printf(Perl_debug_log,
526 "Cleaning named glob CV object:\n "), sv_dump(obj)));
530 SvREFCNT_dec(sv); /* undo the inc above */
533 /* clear any IO slots in a GV which hold objects (except stderr, defout);
534 * called by sv_clean_objs() for each live GV */
537 do_clean_named_io_objs(pTHX_ SV *const sv)
541 assert(SvTYPE(sv) == SVt_PVGV);
542 assert(isGV_with_GP(sv));
543 if (!GvGP(sv) || sv == (SV*)PL_stderrgv || sv == (SV*)PL_defoutgv)
547 if ( ((obj = MUTABLE_SV(GvIO(sv)) )) && SvOBJECT(obj)) {
548 DEBUG_D((PerlIO_printf(Perl_debug_log,
549 "Cleaning named glob IO object:\n "), sv_dump(obj)));
553 SvREFCNT_dec(sv); /* undo the inc above */
556 /* Void wrapper to pass to visit() */
558 do_curse(pTHX_ SV * const sv) {
559 if ((PL_stderrgv && GvGP(PL_stderrgv) && (SV*)GvIO(PL_stderrgv) == sv)
560 || (PL_defoutgv && GvGP(PL_defoutgv) && (SV*)GvIO(PL_defoutgv) == sv))
566 =for apidoc sv_clean_objs
568 Attempt to destroy all objects not yet freed.
574 Perl_sv_clean_objs(pTHX)
578 PL_in_clean_objs = TRUE;
579 visit(do_clean_objs, SVf_ROK, SVf_ROK);
580 /* Some barnacles may yet remain, clinging to typeglobs.
581 * Run the non-IO destructors first: they may want to output
582 * error messages, close files etc */
583 visit(do_clean_named_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
584 visit(do_clean_named_io_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
585 /* And if there are some very tenacious barnacles clinging to arrays,
586 closures, or what have you.... */
587 visit(do_curse, SVs_OBJECT, SVs_OBJECT);
588 olddef = PL_defoutgv;
589 PL_defoutgv = NULL; /* disable skip of PL_defoutgv */
590 if (olddef && isGV_with_GP(olddef))
591 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olddef));
592 olderr = PL_stderrgv;
593 PL_stderrgv = NULL; /* disable skip of PL_stderrgv */
594 if (olderr && isGV_with_GP(olderr))
595 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olderr));
596 SvREFCNT_dec(olddef);
597 PL_in_clean_objs = FALSE;
600 /* called by sv_clean_all() for each live SV */
603 do_clean_all(pTHX_ SV *const sv)
606 if (sv == (const SV *) PL_fdpid || sv == (const SV *)PL_strtab) {
607 /* don't clean pid table and strtab */
610 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning loops: SV at 0x%"UVxf"\n", PTR2UV(sv)) ));
611 SvFLAGS(sv) |= SVf_BREAK;
616 =for apidoc sv_clean_all
618 Decrement the refcnt of each remaining SV, possibly triggering a
619 cleanup. This function may have to be called multiple times to free
620 SVs which are in complex self-referential hierarchies.
626 Perl_sv_clean_all(pTHX)
630 PL_in_clean_all = TRUE;
631 cleaned = visit(do_clean_all, 0,0);
636 ARENASETS: a meta-arena implementation which separates arena-info
637 into struct arena_set, which contains an array of struct
638 arena_descs, each holding info for a single arena. By separating
639 the meta-info from the arena, we recover the 1st slot, formerly
640 borrowed for list management. The arena_set is about the size of an
641 arena, avoiding the needless malloc overhead of a naive linked-list.
643 The cost is 1 arena-set malloc per ~320 arena-mallocs, + the unused
644 memory in the last arena-set (1/2 on average). In trade, we get
645 back the 1st slot in each arena (ie 1.7% of a CV-arena, less for
646 smaller types). The recovery of the wasted space allows use of
647 small arenas for large, rare body types, by changing array* fields
648 in body_details_by_type[] below.
651 char *arena; /* the raw storage, allocated aligned */
652 size_t size; /* its size ~4k typ */
653 svtype utype; /* bodytype stored in arena */
658 /* Get the maximum number of elements in set[] such that struct arena_set
659 will fit within PERL_ARENA_SIZE, which is probably just under 4K, and
660 therefore likely to be 1 aligned memory page. */
662 #define ARENAS_PER_SET ((PERL_ARENA_SIZE - sizeof(struct arena_set*) \
663 - 2 * sizeof(int)) / sizeof (struct arena_desc))
666 struct arena_set* next;
667 unsigned int set_size; /* ie ARENAS_PER_SET */
668 unsigned int curr; /* index of next available arena-desc */
669 struct arena_desc set[ARENAS_PER_SET];
673 =for apidoc sv_free_arenas
675 Deallocate the memory used by all arenas. Note that all the individual SV
676 heads and bodies within the arenas must already have been freed.
681 Perl_sv_free_arenas(pTHX)
688 /* Free arenas here, but be careful about fake ones. (We assume
689 contiguity of the fake ones with the corresponding real ones.) */
691 for (sva = PL_sv_arenaroot; sva; sva = svanext) {
692 svanext = MUTABLE_SV(SvANY(sva));
693 while (svanext && SvFAKE(svanext))
694 svanext = MUTABLE_SV(SvANY(svanext));
701 struct arena_set *aroot = (struct arena_set*) PL_body_arenas;
704 struct arena_set *current = aroot;
707 assert(aroot->set[i].arena);
708 Safefree(aroot->set[i].arena);
716 i = PERL_ARENA_ROOTS_SIZE;
718 PL_body_roots[i] = 0;
725 Here are mid-level routines that manage the allocation of bodies out
726 of the various arenas. There are 5 kinds of arenas:
728 1. SV-head arenas, which are discussed and handled above
729 2. regular body arenas
730 3. arenas for reduced-size bodies
733 Arena types 2 & 3 are chained by body-type off an array of
734 arena-root pointers, which is indexed by svtype. Some of the
735 larger/less used body types are malloced singly, since a large
736 unused block of them is wasteful. Also, several svtypes dont have
737 bodies; the data fits into the sv-head itself. The arena-root
738 pointer thus has a few unused root-pointers (which may be hijacked
739 later for arena types 4,5)
741 3 differs from 2 as an optimization; some body types have several
742 unused fields in the front of the structure (which are kept in-place
743 for consistency). These bodies can be allocated in smaller chunks,
744 because the leading fields arent accessed. Pointers to such bodies
745 are decremented to point at the unused 'ghost' memory, knowing that
746 the pointers are used with offsets to the real memory.
749 =head1 SV-Body Allocation
751 Allocation of SV-bodies is similar to SV-heads, differing as follows;
752 the allocation mechanism is used for many body types, so is somewhat
753 more complicated, it uses arena-sets, and has no need for still-live
756 At the outermost level, (new|del)_X*V macros return bodies of the
757 appropriate type. These macros call either (new|del)_body_type or
758 (new|del)_body_allocated macro pairs, depending on specifics of the
759 type. Most body types use the former pair, the latter pair is used to
760 allocate body types with "ghost fields".
762 "ghost fields" are fields that are unused in certain types, and
763 consequently don't need to actually exist. They are declared because
764 they're part of a "base type", which allows use of functions as
765 methods. The simplest examples are AVs and HVs, 2 aggregate types
766 which don't use the fields which support SCALAR semantics.
768 For these types, the arenas are carved up into appropriately sized
769 chunks, we thus avoid wasted memory for those unaccessed members.
770 When bodies are allocated, we adjust the pointer back in memory by the
771 size of the part not allocated, so it's as if we allocated the full
772 structure. (But things will all go boom if you write to the part that
773 is "not there", because you'll be overwriting the last members of the
774 preceding structure in memory.)
776 We calculate the correction using the STRUCT_OFFSET macro on the first
777 member present. If the allocated structure is smaller (no initial NV
778 actually allocated) then the net effect is to subtract the size of the NV
779 from the pointer, to return a new pointer as if an initial NV were actually
780 allocated. (We were using structures named *_allocated for this, but
781 this turned out to be a subtle bug, because a structure without an NV
782 could have a lower alignment constraint, but the compiler is allowed to
783 optimised accesses based on the alignment constraint of the actual pointer
784 to the full structure, for example, using a single 64 bit load instruction
785 because it "knows" that two adjacent 32 bit members will be 8-byte aligned.)
787 This is the same trick as was used for NV and IV bodies. Ironically it
788 doesn't need to be used for NV bodies any more, because NV is now at
789 the start of the structure. IV bodies don't need it either, because
790 they are no longer allocated.
792 In turn, the new_body_* allocators call S_new_body(), which invokes
793 new_body_inline macro, which takes a lock, and takes a body off the
794 linked list at PL_body_roots[sv_type], calling Perl_more_bodies() if
795 necessary to refresh an empty list. Then the lock is released, and
796 the body is returned.
798 Perl_more_bodies allocates a new arena, and carves it up into an array of N
799 bodies, which it strings into a linked list. It looks up arena-size
800 and body-size from the body_details table described below, thus
801 supporting the multiple body-types.
803 If PURIFY is defined, or PERL_ARENA_SIZE=0, arenas are not used, and
804 the (new|del)_X*V macros are mapped directly to malloc/free.
806 For each sv-type, struct body_details bodies_by_type[] carries
807 parameters which control these aspects of SV handling:
809 Arena_size determines whether arenas are used for this body type, and if
810 so, how big they are. PURIFY or PERL_ARENA_SIZE=0 set this field to
811 zero, forcing individual mallocs and frees.
813 Body_size determines how big a body is, and therefore how many fit into
814 each arena. Offset carries the body-pointer adjustment needed for
815 "ghost fields", and is used in *_allocated macros.
817 But its main purpose is to parameterize info needed in
818 Perl_sv_upgrade(). The info here dramatically simplifies the function
819 vs the implementation in 5.8.8, making it table-driven. All fields
820 are used for this, except for arena_size.
822 For the sv-types that have no bodies, arenas are not used, so those
823 PL_body_roots[sv_type] are unused, and can be overloaded. In
824 something of a special case, SVt_NULL is borrowed for HE arenas;
825 PL_body_roots[HE_SVSLOT=SVt_NULL] is filled by S_more_he, but the
826 bodies_by_type[SVt_NULL] slot is not used, as the table is not
831 struct body_details {
832 U8 body_size; /* Size to allocate */
833 U8 copy; /* Size of structure to copy (may be shorter) */
835 unsigned int type : 4; /* We have space for a sanity check. */
836 unsigned int cant_upgrade : 1; /* Cannot upgrade this type */
837 unsigned int zero_nv : 1; /* zero the NV when upgrading from this */
838 unsigned int arena : 1; /* Allocated from an arena */
839 size_t arena_size; /* Size of arena to allocate */
847 /* With -DPURFIY we allocate everything directly, and don't use arenas.
848 This seems a rather elegant way to simplify some of the code below. */
849 #define HASARENA FALSE
851 #define HASARENA TRUE
853 #define NOARENA FALSE
855 /* Size the arenas to exactly fit a given number of bodies. A count
856 of 0 fits the max number bodies into a PERL_ARENA_SIZE.block,
857 simplifying the default. If count > 0, the arena is sized to fit
858 only that many bodies, allowing arenas to be used for large, rare
859 bodies (XPVFM, XPVIO) without undue waste. The arena size is
860 limited by PERL_ARENA_SIZE, so we can safely oversize the
863 #define FIT_ARENA0(body_size) \
864 ((size_t)(PERL_ARENA_SIZE / body_size) * body_size)
865 #define FIT_ARENAn(count,body_size) \
866 ( count * body_size <= PERL_ARENA_SIZE) \
867 ? count * body_size \
868 : FIT_ARENA0 (body_size)
869 #define FIT_ARENA(count,body_size) \
871 ? FIT_ARENAn (count, body_size) \
872 : FIT_ARENA0 (body_size)
874 /* Calculate the length to copy. Specifically work out the length less any
875 final padding the compiler needed to add. See the comment in sv_upgrade
876 for why copying the padding proved to be a bug. */
878 #define copy_length(type, last_member) \
879 STRUCT_OFFSET(type, last_member) \
880 + sizeof (((type*)SvANY((const SV *)0))->last_member)
882 static const struct body_details bodies_by_type[] = {
883 /* HEs use this offset for their arena. */
884 { 0, 0, 0, SVt_NULL, FALSE, NONV, NOARENA, 0 },
886 /* The bind placeholder pretends to be an RV for now.
887 Also it's marked as "can't upgrade" to stop anyone using it before it's
889 { 0, 0, 0, SVt_BIND, TRUE, NONV, NOARENA, 0 },
891 /* IVs are in the head, so the allocation size is 0. */
893 sizeof(IV), /* This is used to copy out the IV body. */
894 STRUCT_OFFSET(XPVIV, xiv_iv), SVt_IV, FALSE, NONV,
895 NOARENA /* IVS don't need an arena */, 0
898 { sizeof(NV), sizeof(NV),
899 STRUCT_OFFSET(XPVNV, xnv_u),
900 SVt_NV, FALSE, HADNV, HASARENA, FIT_ARENA(0, sizeof(NV)) },
902 { sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur),
903 copy_length(XPV, xpv_len) - STRUCT_OFFSET(XPV, xpv_cur),
904 + STRUCT_OFFSET(XPV, xpv_cur),
905 SVt_PV, FALSE, NONV, HASARENA,
906 FIT_ARENA(0, sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur)) },
908 { sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur),
909 copy_length(XPVIV, xiv_u) - STRUCT_OFFSET(XPV, xpv_cur),
910 + STRUCT_OFFSET(XPV, xpv_cur),
911 SVt_PVIV, FALSE, NONV, HASARENA,
912 FIT_ARENA(0, sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur)) },
914 { sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur),
915 copy_length(XPVNV, xnv_u) - STRUCT_OFFSET(XPV, xpv_cur),
916 + STRUCT_OFFSET(XPV, xpv_cur),
917 SVt_PVNV, FALSE, HADNV, HASARENA,
918 FIT_ARENA(0, sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur)) },
920 { sizeof(XPVMG), copy_length(XPVMG, xnv_u), 0, SVt_PVMG, FALSE, HADNV,
921 HASARENA, FIT_ARENA(0, sizeof(XPVMG)) },
926 SVt_REGEXP, FALSE, NONV, HASARENA,
927 FIT_ARENA(0, sizeof(regexp))
930 { sizeof(XPVGV), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV,
931 HASARENA, FIT_ARENA(0, sizeof(XPVGV)) },
933 { sizeof(XPVLV), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV,
934 HASARENA, FIT_ARENA(0, sizeof(XPVLV)) },
937 copy_length(XPVAV, xav_alloc),
939 SVt_PVAV, TRUE, NONV, HASARENA,
940 FIT_ARENA(0, sizeof(XPVAV)) },
943 copy_length(XPVHV, xhv_max),
945 SVt_PVHV, TRUE, NONV, HASARENA,
946 FIT_ARENA(0, sizeof(XPVHV)) },
951 SVt_PVCV, TRUE, NONV, HASARENA,
952 FIT_ARENA(0, sizeof(XPVCV)) },
957 SVt_PVFM, TRUE, NONV, NOARENA,
958 FIT_ARENA(20, sizeof(XPVFM)) },
963 SVt_PVIO, TRUE, NONV, HASARENA,
964 FIT_ARENA(24, sizeof(XPVIO)) },
967 #define new_body_allocated(sv_type) \
968 (void *)((char *)S_new_body(aTHX_ sv_type) \
969 - bodies_by_type[sv_type].offset)
971 /* return a thing to the free list */
973 #define del_body(thing, root) \
975 void ** const thing_copy = (void **)thing; \
976 *thing_copy = *root; \
977 *root = (void*)thing_copy; \
982 #define new_XNV() safemalloc(sizeof(XPVNV))
983 #define new_XPVNV() safemalloc(sizeof(XPVNV))
984 #define new_XPVMG() safemalloc(sizeof(XPVMG))
986 #define del_XPVGV(p) safefree(p)
990 #define new_XNV() new_body_allocated(SVt_NV)
991 #define new_XPVNV() new_body_allocated(SVt_PVNV)
992 #define new_XPVMG() new_body_allocated(SVt_PVMG)
994 #define del_XPVGV(p) del_body(p + bodies_by_type[SVt_PVGV].offset, \
995 &PL_body_roots[SVt_PVGV])
999 /* no arena for you! */
1001 #define new_NOARENA(details) \
1002 safemalloc((details)->body_size + (details)->offset)
1003 #define new_NOARENAZ(details) \
1004 safecalloc((details)->body_size + (details)->offset, 1)
1007 Perl_more_bodies (pTHX_ const svtype sv_type, const size_t body_size,
1008 const size_t arena_size)
1011 void ** const root = &PL_body_roots[sv_type];
1012 struct arena_desc *adesc;
1013 struct arena_set *aroot = (struct arena_set *) PL_body_arenas;
1017 const size_t good_arena_size = Perl_malloc_good_size(arena_size);
1018 #if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT_PRIVATE)
1019 static bool done_sanity_check;
1021 /* PERL_GLOBAL_STRUCT_PRIVATE cannot coexist with global
1022 * variables like done_sanity_check. */
1023 if (!done_sanity_check) {
1024 unsigned int i = SVt_LAST;
1026 done_sanity_check = TRUE;
1029 assert (bodies_by_type[i].type == i);
1035 /* may need new arena-set to hold new arena */
1036 if (!aroot || aroot->curr >= aroot->set_size) {
1037 struct arena_set *newroot;
1038 Newxz(newroot, 1, struct arena_set);
1039 newroot->set_size = ARENAS_PER_SET;
1040 newroot->next = aroot;
1042 PL_body_arenas = (void *) newroot;
1043 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot));
1046 /* ok, now have arena-set with at least 1 empty/available arena-desc */
1047 curr = aroot->curr++;
1048 adesc = &(aroot->set[curr]);
1049 assert(!adesc->arena);
1051 Newx(adesc->arena, good_arena_size, char);
1052 adesc->size = good_arena_size;
1053 adesc->utype = sv_type;
1054 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %"UVuf"\n",
1055 curr, (void*)adesc->arena, (UV)good_arena_size));
1057 start = (char *) adesc->arena;
1059 /* Get the address of the byte after the end of the last body we can fit.
1060 Remember, this is integer division: */
1061 end = start + good_arena_size / body_size * body_size;
1063 /* computed count doesn't reflect the 1st slot reservation */
1064 #if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE)
1065 DEBUG_m(PerlIO_printf(Perl_debug_log,
1066 "arena %p end %p arena-size %d (from %d) type %d "
1068 (void*)start, (void*)end, (int)good_arena_size,
1069 (int)arena_size, sv_type, (int)body_size,
1070 (int)good_arena_size / (int)body_size));
1072 DEBUG_m(PerlIO_printf(Perl_debug_log,
1073 "arena %p end %p arena-size %d type %d size %d ct %d\n",
1074 (void*)start, (void*)end,
1075 (int)arena_size, sv_type, (int)body_size,
1076 (int)good_arena_size / (int)body_size));
1078 *root = (void *)start;
1081 /* Where the next body would start: */
1082 char * const next = start + body_size;
1085 /* This is the last body: */
1086 assert(next == end);
1088 *(void **)start = 0;
1092 *(void**) start = (void *)next;
1097 /* grab a new thing from the free list, allocating more if necessary.
1098 The inline version is used for speed in hot routines, and the
1099 function using it serves the rest (unless PURIFY).
1101 #define new_body_inline(xpv, sv_type) \
1103 void ** const r3wt = &PL_body_roots[sv_type]; \
1104 xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \
1105 ? *((void **)(r3wt)) : Perl_more_bodies(aTHX_ sv_type, \
1106 bodies_by_type[sv_type].body_size,\
1107 bodies_by_type[sv_type].arena_size)); \
1108 *(r3wt) = *(void**)(xpv); \
1114 S_new_body(pTHX_ const svtype sv_type)
1118 new_body_inline(xpv, sv_type);
1124 static const struct body_details fake_rv =
1125 { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 };
1128 =for apidoc sv_upgrade
1130 Upgrade an SV to a more complex form. Generally adds a new body type to the
1131 SV, then copies across as much information as possible from the old body.
1132 It croaks if the SV is already in a more complex form than requested. You
1133 generally want to use the C<SvUPGRADE> macro wrapper, which checks the type
1134 before calling C<sv_upgrade>, and hence does not croak. See also
1141 Perl_sv_upgrade(pTHX_ register SV *const sv, svtype new_type)
1146 const svtype old_type = SvTYPE(sv);
1147 const struct body_details *new_type_details;
1148 const struct body_details *old_type_details
1149 = bodies_by_type + old_type;
1150 SV *referant = NULL;
1152 PERL_ARGS_ASSERT_SV_UPGRADE;
1154 if (old_type == new_type)
1157 /* This clause was purposefully added ahead of the early return above to
1158 the shared string hackery for (sort {$a <=> $b} keys %hash), with the
1159 inference by Nick I-S that it would fix other troublesome cases. See
1160 changes 7162, 7163 (f130fd4589cf5fbb24149cd4db4137c8326f49c1 and parent)
1162 Given that shared hash key scalars are no longer PVIV, but PV, there is
1163 no longer need to unshare so as to free up the IVX slot for its proper
1164 purpose. So it's safe to move the early return earlier. */
1166 if (new_type > SVt_PVMG && SvIsCOW(sv)) {
1167 sv_force_normal_flags(sv, 0);
1170 old_body = SvANY(sv);
1172 /* Copying structures onto other structures that have been neatly zeroed
1173 has a subtle gotcha. Consider XPVMG
1175 +------+------+------+------+------+-------+-------+
1176 | NV | CUR | LEN | IV | MAGIC | STASH |
1177 +------+------+------+------+------+-------+-------+
1178 0 4 8 12 16 20 24 28
1180 where NVs are aligned to 8 bytes, so that sizeof that structure is
1181 actually 32 bytes long, with 4 bytes of padding at the end:
1183 +------+------+------+------+------+-------+-------+------+
1184 | NV | CUR | LEN | IV | MAGIC | STASH | ??? |
1185 +------+------+------+------+------+-------+-------+------+
1186 0 4 8 12 16 20 24 28 32
1188 so what happens if you allocate memory for this structure:
1190 +------+------+------+------+------+-------+-------+------+------+...
1191 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME |
1192 +------+------+------+------+------+-------+-------+------+------+...
1193 0 4 8 12 16 20 24 28 32 36
1195 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you
1196 expect, because you copy the area marked ??? onto GP. Now, ??? may have
1197 started out as zero once, but it's quite possible that it isn't. So now,
1198 rather than a nicely zeroed GP, you have it pointing somewhere random.
1201 (In fact, GP ends up pointing at a previous GP structure, because the
1202 principle cause of the padding in XPVMG getting garbage is a copy of
1203 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now
1204 this happens to be moot because XPVGV has been re-ordered, with GP
1205 no longer after STASH)
1207 So we are careful and work out the size of used parts of all the
1215 referant = SvRV(sv);
1216 old_type_details = &fake_rv;
1217 if (new_type == SVt_NV)
1218 new_type = SVt_PVNV;
1220 if (new_type < SVt_PVIV) {
1221 new_type = (new_type == SVt_NV)
1222 ? SVt_PVNV : SVt_PVIV;
1227 if (new_type < SVt_PVNV) {
1228 new_type = SVt_PVNV;
1232 assert(new_type > SVt_PV);
1233 assert(SVt_IV < SVt_PV);
1234 assert(SVt_NV < SVt_PV);
1241 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena,
1242 there's no way that it can be safely upgraded, because perl.c
1243 expects to Safefree(SvANY(PL_mess_sv)) */
1244 assert(sv != PL_mess_sv);
1245 /* This flag bit is used to mean other things in other scalar types.
1246 Given that it only has meaning inside the pad, it shouldn't be set
1247 on anything that can get upgraded. */
1248 assert(!SvPAD_TYPED(sv));
1251 if (old_type_details->cant_upgrade)
1252 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf,
1253 sv_reftype(sv, 0), (UV) old_type, (UV) new_type);
1256 if (old_type > new_type)
1257 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d",
1258 (int)old_type, (int)new_type);
1260 new_type_details = bodies_by_type + new_type;
1262 SvFLAGS(sv) &= ~SVTYPEMASK;
1263 SvFLAGS(sv) |= new_type;
1265 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of
1266 the return statements above will have triggered. */
1267 assert (new_type != SVt_NULL);
1270 assert(old_type == SVt_NULL);
1271 SvANY(sv) = (XPVIV*)((char*)&(sv->sv_u.svu_iv) - STRUCT_OFFSET(XPVIV, xiv_iv));
1275 assert(old_type == SVt_NULL);
1276 SvANY(sv) = new_XNV();
1281 assert(new_type_details->body_size);
1284 assert(new_type_details->arena);
1285 assert(new_type_details->arena_size);
1286 /* This points to the start of the allocated area. */
1287 new_body_inline(new_body, new_type);
1288 Zero(new_body, new_type_details->body_size, char);
1289 new_body = ((char *)new_body) - new_type_details->offset;
1291 /* We always allocated the full length item with PURIFY. To do this
1292 we fake things so that arena is false for all 16 types.. */
1293 new_body = new_NOARENAZ(new_type_details);
1295 SvANY(sv) = new_body;
1296 if (new_type == SVt_PVAV) {
1300 if (old_type_details->body_size) {
1303 /* It will have been zeroed when the new body was allocated.
1304 Lets not write to it, in case it confuses a write-back
1310 #ifndef NODEFAULT_SHAREKEYS
1311 HvSHAREKEYS_on(sv); /* key-sharing on by default */
1313 HvMAX(sv) = 7; /* (start with 8 buckets) */
1316 /* SVt_NULL isn't the only thing upgraded to AV or HV.
1317 The target created by newSVrv also is, and it can have magic.
1318 However, it never has SvPVX set.
1320 if (old_type == SVt_IV) {
1322 } else if (old_type >= SVt_PV) {
1323 assert(SvPVX_const(sv) == 0);
1326 if (old_type >= SVt_PVMG) {
1327 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic);
1328 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash);
1330 sv->sv_u.svu_array = NULL; /* or svu_hash */
1335 /* XXX Is this still needed? Was it ever needed? Surely as there is
1336 no route from NV to PVIV, NOK can never be true */
1337 assert(!SvNOKp(sv));
1349 assert(new_type_details->body_size);
1350 /* We always allocated the full length item with PURIFY. To do this
1351 we fake things so that arena is false for all 16 types.. */
1352 if(new_type_details->arena) {
1353 /* This points to the start of the allocated area. */
1354 new_body_inline(new_body, new_type);
1355 Zero(new_body, new_type_details->body_size, char);
1356 new_body = ((char *)new_body) - new_type_details->offset;
1358 new_body = new_NOARENAZ(new_type_details);
1360 SvANY(sv) = new_body;
1362 if (old_type_details->copy) {
1363 /* There is now the potential for an upgrade from something without
1364 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */
1365 int offset = old_type_details->offset;
1366 int length = old_type_details->copy;
1368 if (new_type_details->offset > old_type_details->offset) {
1369 const int difference
1370 = new_type_details->offset - old_type_details->offset;
1371 offset += difference;
1372 length -= difference;
1374 assert (length >= 0);
1376 Copy((char *)old_body + offset, (char *)new_body + offset, length,
1380 #ifndef NV_ZERO_IS_ALLBITS_ZERO
1381 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a
1382 * correct 0.0 for us. Otherwise, if the old body didn't have an
1383 * NV slot, but the new one does, then we need to initialise the
1384 * freshly created NV slot with whatever the correct bit pattern is
1386 if (old_type_details->zero_nv && !new_type_details->zero_nv
1387 && !isGV_with_GP(sv))
1391 if (new_type == SVt_PVIO) {
1392 IO * const io = MUTABLE_IO(sv);
1393 GV *iogv = gv_fetchpvs("IO::File::", GV_ADD, SVt_PVHV);
1396 /* Clear the stashcache because a new IO could overrule a package
1398 DEBUG_o(Perl_deb(aTHX_ "sv_upgrade clearing PL_stashcache\n"));
1399 hv_clear(PL_stashcache);
1401 SvSTASH_set(io, MUTABLE_HV(SvREFCNT_inc(GvHV(iogv))));
1402 IoPAGE_LEN(sv) = 60;
1404 if (new_type == SVt_REGEXP)
1405 sv->sv_u.svu_rx = (regexp *)new_body;
1406 else if (old_type < SVt_PV) {
1407 /* referant will be NULL unless the old type was SVt_IV emulating
1409 sv->sv_u.svu_rv = referant;
1413 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1414 (unsigned long)new_type);
1417 if (old_type > SVt_IV) {
1421 /* Note that there is an assumption that all bodies of types that
1422 can be upgraded came from arenas. Only the more complex non-
1423 upgradable types are allowed to be directly malloc()ed. */
1424 assert(old_type_details->arena);
1425 del_body((void*)((char*)old_body + old_type_details->offset),
1426 &PL_body_roots[old_type]);
1432 =for apidoc sv_backoff
1434 Remove any string offset. You should normally use the C<SvOOK_off> macro
1441 Perl_sv_backoff(pTHX_ register SV *const sv)
1444 const char * const s = SvPVX_const(sv);
1446 PERL_ARGS_ASSERT_SV_BACKOFF;
1447 PERL_UNUSED_CONTEXT;
1450 assert(SvTYPE(sv) != SVt_PVHV);
1451 assert(SvTYPE(sv) != SVt_PVAV);
1453 SvOOK_offset(sv, delta);
1455 SvLEN_set(sv, SvLEN(sv) + delta);
1456 SvPV_set(sv, SvPVX(sv) - delta);
1457 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
1458 SvFLAGS(sv) &= ~SVf_OOK;
1465 Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1466 upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1467 Use the C<SvGROW> wrapper instead.
1473 Perl_sv_grow(pTHX_ register SV *const sv, register STRLEN newlen)
1477 PERL_ARGS_ASSERT_SV_GROW;
1479 if (PL_madskills && newlen >= 0x100000) {
1480 PerlIO_printf(Perl_debug_log,
1481 "Allocation too large: %"UVxf"\n", (UV)newlen);
1483 #ifdef HAS_64K_LIMIT
1484 if (newlen >= 0x10000) {
1485 PerlIO_printf(Perl_debug_log,
1486 "Allocation too large: %"UVxf"\n", (UV)newlen);
1489 #endif /* HAS_64K_LIMIT */
1492 if (SvTYPE(sv) < SVt_PV) {
1493 sv_upgrade(sv, SVt_PV);
1494 s = SvPVX_mutable(sv);
1496 else if (SvOOK(sv)) { /* pv is offset? */
1498 s = SvPVX_mutable(sv);
1499 if (newlen > SvLEN(sv))
1500 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
1501 #ifdef HAS_64K_LIMIT
1502 if (newlen >= 0x10000)
1507 s = SvPVX_mutable(sv);
1509 if (newlen > SvLEN(sv)) { /* need more room? */
1510 STRLEN minlen = SvCUR(sv);
1511 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
1512 if (newlen < minlen)
1514 #ifndef Perl_safesysmalloc_size
1515 newlen = PERL_STRLEN_ROUNDUP(newlen);
1517 if (SvLEN(sv) && s) {
1518 s = (char*)saferealloc(s, newlen);
1521 s = (char*)safemalloc(newlen);
1522 if (SvPVX_const(sv) && SvCUR(sv)) {
1523 Move(SvPVX_const(sv), s, (newlen < SvCUR(sv)) ? newlen : SvCUR(sv), char);
1527 #ifdef Perl_safesysmalloc_size
1528 /* Do this here, do it once, do it right, and then we will never get
1529 called back into sv_grow() unless there really is some growing
1531 SvLEN_set(sv, Perl_safesysmalloc_size(s));
1533 SvLEN_set(sv, newlen);
1540 =for apidoc sv_setiv
1542 Copies an integer into the given SV, upgrading first if necessary.
1543 Does not handle 'set' magic. See also C<sv_setiv_mg>.
1549 Perl_sv_setiv(pTHX_ register SV *const sv, const IV i)
1553 PERL_ARGS_ASSERT_SV_SETIV;
1555 SV_CHECK_THINKFIRST_COW_DROP(sv);
1556 switch (SvTYPE(sv)) {
1559 sv_upgrade(sv, SVt_IV);
1562 sv_upgrade(sv, SVt_PVIV);
1566 if (!isGV_with_GP(sv))
1573 /* diag_listed_as: Can't coerce %s to %s in %s */
1574 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1578 (void)SvIOK_only(sv); /* validate number */
1584 =for apidoc sv_setiv_mg
1586 Like C<sv_setiv>, but also handles 'set' magic.
1592 Perl_sv_setiv_mg(pTHX_ register SV *const sv, const IV i)
1594 PERL_ARGS_ASSERT_SV_SETIV_MG;
1601 =for apidoc sv_setuv
1603 Copies an unsigned integer into the given SV, upgrading first if necessary.
1604 Does not handle 'set' magic. See also C<sv_setuv_mg>.
1610 Perl_sv_setuv(pTHX_ register SV *const sv, const UV u)
1612 PERL_ARGS_ASSERT_SV_SETUV;
1614 /* With the if statement to ensure that integers are stored as IVs whenever
1616 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
1619 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1621 If you wish to remove the following if statement, so that this routine
1622 (and its callers) always return UVs, please benchmark to see what the
1623 effect is. Modern CPUs may be different. Or may not :-)
1625 if (u <= (UV)IV_MAX) {
1626 sv_setiv(sv, (IV)u);
1635 =for apidoc sv_setuv_mg
1637 Like C<sv_setuv>, but also handles 'set' magic.
1643 Perl_sv_setuv_mg(pTHX_ register SV *const sv, const UV u)
1645 PERL_ARGS_ASSERT_SV_SETUV_MG;
1652 =for apidoc sv_setnv
1654 Copies a double into the given SV, upgrading first if necessary.
1655 Does not handle 'set' magic. See also C<sv_setnv_mg>.
1661 Perl_sv_setnv(pTHX_ register SV *const sv, const NV num)
1665 PERL_ARGS_ASSERT_SV_SETNV;
1667 SV_CHECK_THINKFIRST_COW_DROP(sv);
1668 switch (SvTYPE(sv)) {
1671 sv_upgrade(sv, SVt_NV);
1675 sv_upgrade(sv, SVt_PVNV);
1679 if (!isGV_with_GP(sv))
1686 /* diag_listed_as: Can't coerce %s to %s in %s */
1687 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
1692 (void)SvNOK_only(sv); /* validate number */
1697 =for apidoc sv_setnv_mg
1699 Like C<sv_setnv>, but also handles 'set' magic.
1705 Perl_sv_setnv_mg(pTHX_ register SV *const sv, const NV num)
1707 PERL_ARGS_ASSERT_SV_SETNV_MG;
1713 /* Print an "isn't numeric" warning, using a cleaned-up,
1714 * printable version of the offending string
1718 S_not_a_number(pTHX_ SV *const sv)
1725 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1728 dsv = newSVpvs_flags("", SVs_TEMP);
1729 pv = sv_uni_display(dsv, sv, 10, UNI_DISPLAY_ISPRINT);
1732 const char * const limit = tmpbuf + sizeof(tmpbuf) - 8;
1733 /* each *s can expand to 4 chars + "...\0",
1734 i.e. need room for 8 chars */
1736 const char *s = SvPVX_const(sv);
1737 const char * const end = s + SvCUR(sv);
1738 for ( ; s < end && d < limit; s++ ) {
1740 if (ch & 128 && !isPRINT_LC(ch)) {
1749 else if (ch == '\r') {
1753 else if (ch == '\f') {
1757 else if (ch == '\\') {
1761 else if (ch == '\0') {
1765 else if (isPRINT_LC(ch))
1782 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1783 /* diag_listed_as: Argument "%s" isn't numeric%s */
1784 "Argument \"%s\" isn't numeric in %s", pv,
1787 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1788 /* diag_listed_as: Argument "%s" isn't numeric%s */
1789 "Argument \"%s\" isn't numeric", pv);
1793 =for apidoc looks_like_number
1795 Test if the content of an SV looks like a number (or is a number).
1796 C<Inf> and C<Infinity> are treated as numbers (so will not issue a
1797 non-numeric warning), even if your atof() doesn't grok them. Get-magic is
1804 Perl_looks_like_number(pTHX_ SV *const sv)
1809 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1811 if (SvPOK(sv) || SvPOKp(sv)) {
1812 sbegin = SvPV_nomg_const(sv, len);
1815 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
1816 return grok_number(sbegin, len, NULL);
1820 S_glob_2number(pTHX_ GV * const gv)
1822 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1824 /* We know that all GVs stringify to something that is not-a-number,
1825 so no need to test that. */
1826 if (ckWARN(WARN_NUMERIC))
1828 SV *const buffer = sv_newmortal();
1829 gv_efullname3(buffer, gv, "*");
1830 not_a_number(buffer);
1832 /* We just want something true to return, so that S_sv_2iuv_common
1833 can tail call us and return true. */
1837 /* Actually, ISO C leaves conversion of UV to IV undefined, but
1838 until proven guilty, assume that things are not that bad... */
1843 As 64 bit platforms often have an NV that doesn't preserve all bits of
1844 an IV (an assumption perl has been based on to date) it becomes necessary
1845 to remove the assumption that the NV always carries enough precision to
1846 recreate the IV whenever needed, and that the NV is the canonical form.
1847 Instead, IV/UV and NV need to be given equal rights. So as to not lose
1848 precision as a side effect of conversion (which would lead to insanity
1849 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
1850 1) to distinguish between IV/UV/NV slots that have cached a valid
1851 conversion where precision was lost and IV/UV/NV slots that have a
1852 valid conversion which has lost no precision
1853 2) to ensure that if a numeric conversion to one form is requested that
1854 would lose precision, the precise conversion (or differently
1855 imprecise conversion) is also performed and cached, to prevent
1856 requests for different numeric formats on the same SV causing
1857 lossy conversion chains. (lossless conversion chains are perfectly
1862 SvIOKp is true if the IV slot contains a valid value
1863 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1864 SvNOKp is true if the NV slot contains a valid value
1865 SvNOK is true only if the NV value is accurate
1868 while converting from PV to NV, check to see if converting that NV to an
1869 IV(or UV) would lose accuracy over a direct conversion from PV to
1870 IV(or UV). If it would, cache both conversions, return NV, but mark
1871 SV as IOK NOKp (ie not NOK).
1873 While converting from PV to IV, check to see if converting that IV to an
1874 NV would lose accuracy over a direct conversion from PV to NV. If it
1875 would, cache both conversions, flag similarly.
1877 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
1878 correctly because if IV & NV were set NV *always* overruled.
1879 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
1880 changes - now IV and NV together means that the two are interchangeable:
1881 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
1883 The benefit of this is that operations such as pp_add know that if
1884 SvIOK is true for both left and right operands, then integer addition
1885 can be used instead of floating point (for cases where the result won't
1886 overflow). Before, floating point was always used, which could lead to
1887 loss of precision compared with integer addition.
1889 * making IV and NV equal status should make maths accurate on 64 bit
1891 * may speed up maths somewhat if pp_add and friends start to use
1892 integers when possible instead of fp. (Hopefully the overhead in
1893 looking for SvIOK and checking for overflow will not outweigh the
1894 fp to integer speedup)
1895 * will slow down integer operations (callers of SvIV) on "inaccurate"
1896 values, as the change from SvIOK to SvIOKp will cause a call into
1897 sv_2iv each time rather than a macro access direct to the IV slot
1898 * should speed up number->string conversion on integers as IV is
1899 favoured when IV and NV are equally accurate
1901 ####################################################################
1902 You had better be using SvIOK_notUV if you want an IV for arithmetic:
1903 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
1904 On the other hand, SvUOK is true iff UV.
1905 ####################################################################
1907 Your mileage will vary depending your CPU's relative fp to integer
1911 #ifndef NV_PRESERVES_UV
1912 # define IS_NUMBER_UNDERFLOW_IV 1
1913 # define IS_NUMBER_UNDERFLOW_UV 2
1914 # define IS_NUMBER_IV_AND_UV 2
1915 # define IS_NUMBER_OVERFLOW_IV 4
1916 # define IS_NUMBER_OVERFLOW_UV 5
1918 /* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
1920 /* For sv_2nv these three cases are "SvNOK and don't bother casting" */
1922 S_sv_2iuv_non_preserve(pTHX_ register SV *const sv
1930 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
1932 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));
1933 if (SvNVX(sv) < (NV)IV_MIN) {
1934 (void)SvIOKp_on(sv);
1936 SvIV_set(sv, IV_MIN);
1937 return IS_NUMBER_UNDERFLOW_IV;
1939 if (SvNVX(sv) > (NV)UV_MAX) {
1940 (void)SvIOKp_on(sv);
1943 SvUV_set(sv, UV_MAX);
1944 return IS_NUMBER_OVERFLOW_UV;
1946 (void)SvIOKp_on(sv);
1948 /* Can't use strtol etc to convert this string. (See truth table in
1950 if (SvNVX(sv) <= (UV)IV_MAX) {
1951 SvIV_set(sv, I_V(SvNVX(sv)));
1952 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
1953 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
1955 /* Integer is imprecise. NOK, IOKp */
1957 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
1960 SvUV_set(sv, U_V(SvNVX(sv)));
1961 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
1962 if (SvUVX(sv) == UV_MAX) {
1963 /* As we know that NVs don't preserve UVs, UV_MAX cannot
1964 possibly be preserved by NV. Hence, it must be overflow.
1966 return IS_NUMBER_OVERFLOW_UV;
1968 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
1970 /* Integer is imprecise. NOK, IOKp */
1972 return IS_NUMBER_OVERFLOW_IV;
1974 #endif /* !NV_PRESERVES_UV*/
1977 S_sv_2iuv_common(pTHX_ SV *const sv)
1981 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
1984 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
1985 * without also getting a cached IV/UV from it at the same time
1986 * (ie PV->NV conversion should detect loss of accuracy and cache
1987 * IV or UV at same time to avoid this. */
1988 /* IV-over-UV optimisation - choose to cache IV if possible */
1990 if (SvTYPE(sv) == SVt_NV)
1991 sv_upgrade(sv, SVt_PVNV);
1993 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
1994 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
1995 certainly cast into the IV range at IV_MAX, whereas the correct
1996 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
1998 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
1999 if (Perl_isnan(SvNVX(sv))) {
2005 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2006 SvIV_set(sv, I_V(SvNVX(sv)));
2007 if (SvNVX(sv) == (NV) SvIVX(sv)
2008 #ifndef NV_PRESERVES_UV
2009 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2010 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2011 /* Don't flag it as "accurately an integer" if the number
2012 came from a (by definition imprecise) NV operation, and
2013 we're outside the range of NV integer precision */
2017 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2019 /* scalar has trailing garbage, eg "42a" */
2021 DEBUG_c(PerlIO_printf(Perl_debug_log,
2022 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (precise)\n",
2028 /* IV not precise. No need to convert from PV, as NV
2029 conversion would already have cached IV if it detected
2030 that PV->IV would be better than PV->NV->IV
2031 flags already correct - don't set public IOK. */
2032 DEBUG_c(PerlIO_printf(Perl_debug_log,
2033 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (imprecise)\n",
2038 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2039 but the cast (NV)IV_MIN rounds to a the value less (more
2040 negative) than IV_MIN which happens to be equal to SvNVX ??
2041 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2042 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2043 (NV)UVX == NVX are both true, but the values differ. :-(
2044 Hopefully for 2s complement IV_MIN is something like
2045 0x8000000000000000 which will be exact. NWC */
2048 SvUV_set(sv, U_V(SvNVX(sv)));
2050 (SvNVX(sv) == (NV) SvUVX(sv))
2051 #ifndef NV_PRESERVES_UV
2052 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2053 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2054 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2055 /* Don't flag it as "accurately an integer" if the number
2056 came from a (by definition imprecise) NV operation, and
2057 we're outside the range of NV integer precision */
2063 DEBUG_c(PerlIO_printf(Perl_debug_log,
2064 "0x%"UVxf" 2iv(%"UVuf" => %"IVdf") (as unsigned)\n",
2070 else if (SvPOKp(sv)) {
2072 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2073 /* We want to avoid a possible problem when we cache an IV/ a UV which
2074 may be later translated to an NV, and the resulting NV is not
2075 the same as the direct translation of the initial string
2076 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2077 be careful to ensure that the value with the .456 is around if the
2078 NV value is requested in the future).
2080 This means that if we cache such an IV/a UV, we need to cache the
2081 NV as well. Moreover, we trade speed for space, and do not
2082 cache the NV if we are sure it's not needed.
2085 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2086 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2087 == IS_NUMBER_IN_UV) {
2088 /* It's definitely an integer, only upgrade to PVIV */
2089 if (SvTYPE(sv) < SVt_PVIV)
2090 sv_upgrade(sv, SVt_PVIV);
2092 } else if (SvTYPE(sv) < SVt_PVNV)
2093 sv_upgrade(sv, SVt_PVNV);
2095 /* If NVs preserve UVs then we only use the UV value if we know that
2096 we aren't going to call atof() below. If NVs don't preserve UVs
2097 then the value returned may have more precision than atof() will
2098 return, even though value isn't perfectly accurate. */
2099 if ((numtype & (IS_NUMBER_IN_UV
2100 #ifdef NV_PRESERVES_UV
2103 )) == IS_NUMBER_IN_UV) {
2104 /* This won't turn off the public IOK flag if it was set above */
2105 (void)SvIOKp_on(sv);
2107 if (!(numtype & IS_NUMBER_NEG)) {
2109 if (value <= (UV)IV_MAX) {
2110 SvIV_set(sv, (IV)value);
2112 /* it didn't overflow, and it was positive. */
2113 SvUV_set(sv, value);
2117 /* 2s complement assumption */
2118 if (value <= (UV)IV_MIN) {
2119 SvIV_set(sv, -(IV)value);
2121 /* Too negative for an IV. This is a double upgrade, but
2122 I'm assuming it will be rare. */
2123 if (SvTYPE(sv) < SVt_PVNV)
2124 sv_upgrade(sv, SVt_PVNV);
2128 SvNV_set(sv, -(NV)value);
2129 SvIV_set(sv, IV_MIN);
2133 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2134 will be in the previous block to set the IV slot, and the next
2135 block to set the NV slot. So no else here. */
2137 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2138 != IS_NUMBER_IN_UV) {
2139 /* It wasn't an (integer that doesn't overflow the UV). */
2140 SvNV_set(sv, Atof(SvPVX_const(sv)));
2142 if (! numtype && ckWARN(WARN_NUMERIC))
2145 #if defined(USE_LONG_DOUBLE)
2146 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%" PERL_PRIgldbl ")\n",
2147 PTR2UV(sv), SvNVX(sv)));
2149 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"NVgf")\n",
2150 PTR2UV(sv), SvNVX(sv)));
2153 #ifdef NV_PRESERVES_UV
2154 (void)SvIOKp_on(sv);
2156 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2157 SvIV_set(sv, I_V(SvNVX(sv)));
2158 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2161 NOOP; /* Integer is imprecise. NOK, IOKp */
2163 /* UV will not work better than IV */
2165 if (SvNVX(sv) > (NV)UV_MAX) {
2167 /* Integer is inaccurate. NOK, IOKp, is UV */
2168 SvUV_set(sv, UV_MAX);
2170 SvUV_set(sv, U_V(SvNVX(sv)));
2171 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2172 NV preservse UV so can do correct comparison. */
2173 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2176 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
2181 #else /* NV_PRESERVES_UV */
2182 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2183 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
2184 /* The IV/UV slot will have been set from value returned by
2185 grok_number above. The NV slot has just been set using
2188 assert (SvIOKp(sv));
2190 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2191 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2192 /* Small enough to preserve all bits. */
2193 (void)SvIOKp_on(sv);
2195 SvIV_set(sv, I_V(SvNVX(sv)));
2196 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2198 /* Assumption: first non-preserved integer is < IV_MAX,
2199 this NV is in the preserved range, therefore: */
2200 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2202 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);
2206 0 0 already failed to read UV.
2207 0 1 already failed to read UV.
2208 1 0 you won't get here in this case. IV/UV
2209 slot set, public IOK, Atof() unneeded.
2210 1 1 already read UV.
2211 so there's no point in sv_2iuv_non_preserve() attempting
2212 to use atol, strtol, strtoul etc. */
2214 sv_2iuv_non_preserve (sv, numtype);
2216 sv_2iuv_non_preserve (sv);
2220 #endif /* NV_PRESERVES_UV */
2221 /* It might be more code efficient to go through the entire logic above
2222 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2223 gets complex and potentially buggy, so more programmer efficient
2224 to do it this way, by turning off the public flags: */
2226 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2230 if (isGV_with_GP(sv))
2231 return glob_2number(MUTABLE_GV(sv));
2233 if (!SvPADTMP(sv)) {
2234 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2237 if (SvTYPE(sv) < SVt_IV)
2238 /* Typically the caller expects that sv_any is not NULL now. */
2239 sv_upgrade(sv, SVt_IV);
2240 /* Return 0 from the caller. */
2247 =for apidoc sv_2iv_flags
2249 Return the integer value of an SV, doing any necessary string
2250 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2251 Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2257 Perl_sv_2iv_flags(pTHX_ register SV *const sv, const I32 flags)
2264 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2270 if (flags & SV_SKIP_OVERLOAD)
2272 tmpstr = AMG_CALLunary(sv, numer_amg);
2273 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2274 return SvIV(tmpstr);
2277 return PTR2IV(SvRV(sv));
2280 if (SvVALID(sv) || isREGEXP(sv)) {
2281 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2282 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2283 In practice they are extremely unlikely to actually get anywhere
2284 accessible by user Perl code - the only way that I'm aware of is when
2285 a constant subroutine which is used as the second argument to index.
2287 Regexps have no SvIVX and SvNVX fields.
2289 assert(isREGEXP(sv) || SvPOKp(sv));
2292 const char * const ptr =
2293 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2295 = grok_number(ptr, SvCUR(sv), &value);
2297 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2298 == IS_NUMBER_IN_UV) {
2299 /* It's definitely an integer */
2300 if (numtype & IS_NUMBER_NEG) {
2301 if (value < (UV)IV_MIN)
2304 if (value < (UV)IV_MAX)
2309 if (ckWARN(WARN_NUMERIC))
2312 return I_V(Atof(ptr));
2316 if (SvTHINKFIRST(sv)) {
2317 #ifdef PERL_OLD_COPY_ON_WRITE
2319 sv_force_normal_flags(sv, 0);
2322 if (SvREADONLY(sv) && !SvOK(sv)) {
2323 if (ckWARN(WARN_UNINITIALIZED))
2330 if (S_sv_2iuv_common(aTHX_ sv))
2334 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"IVdf")\n",
2335 PTR2UV(sv),SvIVX(sv)));
2336 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
2340 =for apidoc sv_2uv_flags
2342 Return the unsigned integer value of an SV, doing any necessary string
2343 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2344 Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
2350 Perl_sv_2uv_flags(pTHX_ register SV *const sv, const I32 flags)
2357 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2363 if (flags & SV_SKIP_OVERLOAD)
2365 tmpstr = AMG_CALLunary(sv, numer_amg);
2366 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2367 return SvUV(tmpstr);
2370 return PTR2UV(SvRV(sv));
2373 if (SvVALID(sv) || isREGEXP(sv)) {
2374 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2375 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2376 Regexps have no SvIVX and SvNVX fields. */
2377 assert(isREGEXP(sv) || SvPOKp(sv));
2380 const char * const ptr =
2381 isREGEXP(sv) ? RX_WRAPPED((REGEXP*)sv) : SvPVX_const(sv);
2383 = grok_number(ptr, SvCUR(sv), &value);
2385 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2386 == IS_NUMBER_IN_UV) {
2387 /* It's definitely an integer */
2388 if (!(numtype & IS_NUMBER_NEG))
2392 if (ckWARN(WARN_NUMERIC))
2395 return U_V(Atof(ptr));
2399 if (SvTHINKFIRST(sv)) {
2400 #ifdef PERL_OLD_COPY_ON_WRITE
2402 sv_force_normal_flags(sv, 0);
2405 if (SvREADONLY(sv) && !SvOK(sv)) {
2406 if (ckWARN(WARN_UNINITIALIZED))
2413 if (S_sv_2iuv_common(aTHX_ sv))
2417 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2uv(%"UVuf")\n",
2418 PTR2UV(sv),SvUVX(sv)));
2419 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
2423 =for apidoc sv_2nv_flags
2425 Return the num value of an SV, doing any necessary string or integer
2426 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2427 Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
2433 Perl_sv_2nv_flags(pTHX_ register SV *const sv, const I32 flags)
2438 if (SvGMAGICAL(sv) || SvVALID(sv) || isREGEXP(sv)) {
2439 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2440 the same flag bit as SVf_IVisUV, so must not let them cache NVs.
2441 Regexps have no SvIVX and SvNVX fields. */
2443 if (flags & SV_GMAGIC)
2447 if (SvPOKp(sv) && !SvIOKp(sv)) {
2448 ptr = SvPVX_const(sv);
2450 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
2451 !grok_number(ptr, SvCUR(sv), NULL))
2457 return (NV)SvUVX(sv);
2459 return (NV)SvIVX(sv);
2465 ptr = RX_WRAPPED((REGEXP *)sv);
2468 assert(SvTYPE(sv) >= SVt_PVMG);
2469 /* This falls through to the report_uninit near the end of the
2471 } else if (SvTHINKFIRST(sv)) {
2476 if (flags & SV_SKIP_OVERLOAD)
2478 tmpstr = AMG_CALLunary(sv, numer_amg);
2479 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2480 return SvNV(tmpstr);
2483 return PTR2NV(SvRV(sv));
2485 #ifdef PERL_OLD_COPY_ON_WRITE
2487 sv_force_normal_flags(sv, 0);
2490 if (SvREADONLY(sv) && !SvOK(sv)) {
2491 if (ckWARN(WARN_UNINITIALIZED))
2496 if (SvTYPE(sv) < SVt_NV) {
2497 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2498 sv_upgrade(sv, SVt_NV);
2499 #ifdef USE_LONG_DOUBLE
2501 STORE_NUMERIC_LOCAL_SET_STANDARD();
2502 PerlIO_printf(Perl_debug_log,
2503 "0x%"UVxf" num(%" PERL_PRIgldbl ")\n",
2504 PTR2UV(sv), SvNVX(sv));
2505 RESTORE_NUMERIC_LOCAL();
2509 STORE_NUMERIC_LOCAL_SET_STANDARD();
2510 PerlIO_printf(Perl_debug_log, "0x%"UVxf" num(%"NVgf")\n",
2511 PTR2UV(sv), SvNVX(sv));
2512 RESTORE_NUMERIC_LOCAL();
2516 else if (SvTYPE(sv) < SVt_PVNV)
2517 sv_upgrade(sv, SVt_PVNV);
2522 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
2523 #ifdef NV_PRESERVES_UV
2529 /* Only set the public NV OK flag if this NV preserves the IV */
2530 /* Check it's not 0xFFFFFFFFFFFFFFFF */
2532 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
2533 : (SvIVX(sv) == I_V(SvNVX(sv))))
2539 else if (SvPOKp(sv)) {
2541 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2542 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
2544 #ifdef NV_PRESERVES_UV
2545 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2546 == IS_NUMBER_IN_UV) {
2547 /* It's definitely an integer */
2548 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
2550 SvNV_set(sv, Atof(SvPVX_const(sv)));
2556 SvNV_set(sv, Atof(SvPVX_const(sv)));
2557 /* Only set the public NV OK flag if this NV preserves the value in
2558 the PV at least as well as an IV/UV would.
2559 Not sure how to do this 100% reliably. */
2560 /* if that shift count is out of range then Configure's test is
2561 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2563 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2564 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2565 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
2566 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2567 /* Can't use strtol etc to convert this string, so don't try.
2568 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2571 /* value has been set. It may not be precise. */
2572 if ((numtype & IS_NUMBER_NEG) && (value > (UV)IV_MIN)) {
2573 /* 2s complement assumption for (UV)IV_MIN */
2574 SvNOK_on(sv); /* Integer is too negative. */
2579 if (numtype & IS_NUMBER_NEG) {
2580 SvIV_set(sv, -(IV)value);
2581 } else if (value <= (UV)IV_MAX) {
2582 SvIV_set(sv, (IV)value);
2584 SvUV_set(sv, value);
2588 if (numtype & IS_NUMBER_NOT_INT) {
2589 /* I believe that even if the original PV had decimals,
2590 they are lost beyond the limit of the FP precision.
2591 However, neither is canonical, so both only get p
2592 flags. NWC, 2000/11/25 */
2593 /* Both already have p flags, so do nothing */
2595 const NV nv = SvNVX(sv);
2596 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2597 if (SvIVX(sv) == I_V(nv)) {
2600 /* It had no "." so it must be integer. */
2604 /* between IV_MAX and NV(UV_MAX).
2605 Could be slightly > UV_MAX */
2607 if (numtype & IS_NUMBER_NOT_INT) {
2608 /* UV and NV both imprecise. */
2610 const UV nv_as_uv = U_V(nv);
2612 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2621 /* It might be more code efficient to go through the entire logic above
2622 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2623 gets complex and potentially buggy, so more programmer efficient
2624 to do it this way, by turning off the public flags: */
2626 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2627 #endif /* NV_PRESERVES_UV */
2630 if (isGV_with_GP(sv)) {
2631 glob_2number(MUTABLE_GV(sv));
2635 if (!PL_localizing && !SvPADTMP(sv) && ckWARN(WARN_UNINITIALIZED))
2637 assert (SvTYPE(sv) >= SVt_NV);
2638 /* Typically the caller expects that sv_any is not NULL now. */
2639 /* XXX Ilya implies that this is a bug in callers that assume this
2640 and ideally should be fixed. */
2643 #if defined(USE_LONG_DOUBLE)
2645 STORE_NUMERIC_LOCAL_SET_STANDARD();
2646 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2nv(%" PERL_PRIgldbl ")\n",
2647 PTR2UV(sv), SvNVX(sv));
2648 RESTORE_NUMERIC_LOCAL();
2652 STORE_NUMERIC_LOCAL_SET_STANDARD();
2653 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 1nv(%"NVgf")\n",
2654 PTR2UV(sv), SvNVX(sv));
2655 RESTORE_NUMERIC_LOCAL();
2664 Return an SV with the numeric value of the source SV, doing any necessary
2665 reference or overload conversion. You must use the C<SvNUM(sv)> macro to
2666 access this function.
2672 Perl_sv_2num(pTHX_ register SV *const sv)
2674 PERL_ARGS_ASSERT_SV_2NUM;
2679 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
2680 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
2681 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2682 return sv_2num(tmpsv);
2684 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2687 /* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2688 * UV as a string towards the end of buf, and return pointers to start and
2691 * We assume that buf is at least TYPE_CHARS(UV) long.
2695 S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
2697 char *ptr = buf + TYPE_CHARS(UV);
2698 char * const ebuf = ptr;
2701 PERL_ARGS_ASSERT_UIV_2BUF;
2713 *--ptr = '0' + (char)(uv % 10);
2722 =for apidoc sv_2pv_flags
2724 Returns a pointer to the string value of an SV, and sets *lp to its length.
2725 If flags includes SV_GMAGIC, does an mg_get() first. Coerces sv to a
2726 string if necessary. Normally invoked via the C<SvPV_flags> macro.
2727 C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
2733 Perl_sv_2pv_flags(pTHX_ register SV *const sv, STRLEN *const lp, const I32 flags)
2743 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2748 if (flags & SV_SKIP_OVERLOAD)
2750 tmpstr = AMG_CALLunary(sv, string_amg);
2751 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
2752 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2754 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2758 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2759 if (flags & SV_CONST_RETURN) {
2760 pv = (char *) SvPVX_const(tmpstr);
2762 pv = (flags & SV_MUTABLE_RETURN)
2763 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
2766 *lp = SvCUR(tmpstr);
2768 pv = sv_2pv_flags(tmpstr, lp, flags);
2781 SV *const referent = SvRV(sv);
2785 retval = buffer = savepvn("NULLREF", len);
2786 } else if (SvTYPE(referent) == SVt_REGEXP &&
2787 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
2788 amagic_is_enabled(string_amg))) {
2789 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
2793 /* If the regex is UTF-8 we want the containing scalar to
2794 have an UTF-8 flag too */
2801 *lp = RX_WRAPLEN(re);
2803 return RX_WRAPPED(re);
2805 const char *const typestr = sv_reftype(referent, 0);
2806 const STRLEN typelen = strlen(typestr);
2807 UV addr = PTR2UV(referent);
2808 const char *stashname = NULL;
2809 STRLEN stashnamelen = 0; /* hush, gcc */
2810 const char *buffer_end;
2812 if (SvOBJECT(referent)) {
2813 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
2816 stashname = HEK_KEY(name);
2817 stashnamelen = HEK_LEN(name);
2819 if (HEK_UTF8(name)) {
2825 stashname = "__ANON__";
2828 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
2829 + 2 * sizeof(UV) + 2 /* )\0 */;
2831 len = typelen + 3 /* (0x */
2832 + 2 * sizeof(UV) + 2 /* )\0 */;
2835 Newx(buffer, len, char);
2836 buffer_end = retval = buffer + len;
2838 /* Working backwards */
2842 *--retval = PL_hexdigit[addr & 15];
2843 } while (addr >>= 4);
2849 memcpy(retval, typestr, typelen);
2853 retval -= stashnamelen;
2854 memcpy(retval, stashname, stashnamelen);
2856 /* retval may not necessarily have reached the start of the
2858 assert (retval >= buffer);
2860 len = buffer_end - retval - 1; /* -1 for that \0 */
2872 if (flags & SV_MUTABLE_RETURN)
2873 return SvPVX_mutable(sv);
2874 if (flags & SV_CONST_RETURN)
2875 return (char *)SvPVX_const(sv);
2880 /* I'm assuming that if both IV and NV are equally valid then
2881 converting the IV is going to be more efficient */
2882 const U32 isUIOK = SvIsUV(sv);
2883 char buf[TYPE_CHARS(UV)];
2887 if (SvTYPE(sv) < SVt_PVIV)
2888 sv_upgrade(sv, SVt_PVIV);
2889 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
2891 /* inlined from sv_setpvn */
2892 s = SvGROW_mutable(sv, len + 1);
2893 Move(ptr, s, len, char);
2897 else if (SvNOK(sv)) {
2898 if (SvTYPE(sv) < SVt_PVNV)
2899 sv_upgrade(sv, SVt_PVNV);
2900 if (SvNVX(sv) == 0.0) {
2901 s = SvGROW_mutable(sv, 2);
2906 /* The +20 is pure guesswork. Configure test needed. --jhi */
2907 s = SvGROW_mutable(sv, NV_DIG + 20);
2908 /* some Xenix systems wipe out errno here */
2909 Gconvert(SvNVX(sv), NV_DIG, 0, s);
2918 else if (isGV_with_GP(sv)) {
2919 GV *const gv = MUTABLE_GV(sv);
2920 SV *const buffer = sv_newmortal();
2922 gv_efullname3(buffer, gv, "*");
2924 assert(SvPOK(buffer));
2928 *lp = SvCUR(buffer);
2929 return SvPVX(buffer);
2931 else if (isREGEXP(sv)) {
2932 if (lp) *lp = RX_WRAPLEN((REGEXP *)sv);
2933 return RX_WRAPPED((REGEXP *)sv);
2938 if (flags & SV_UNDEF_RETURNS_NULL)
2940 if (!PL_localizing && !SvPADTMP(sv) && ckWARN(WARN_UNINITIALIZED))
2942 /* Typically the caller expects that sv_any is not NULL now. */
2943 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
2944 sv_upgrade(sv, SVt_PV);
2949 const STRLEN len = s - SvPVX_const(sv);
2955 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
2956 PTR2UV(sv),SvPVX_const(sv)));
2957 if (flags & SV_CONST_RETURN)
2958 return (char *)SvPVX_const(sv);
2959 if (flags & SV_MUTABLE_RETURN)
2960 return SvPVX_mutable(sv);
2965 =for apidoc sv_copypv
2967 Copies a stringified representation of the source SV into the
2968 destination SV. Automatically performs any necessary mg_get and
2969 coercion of numeric values into strings. Guaranteed to preserve
2970 UTF8 flag even from overloaded objects. Similar in nature to
2971 sv_2pv[_flags] but operates directly on an SV instead of just the
2972 string. Mostly uses sv_2pv_flags to do its work, except when that
2973 would lose the UTF-8'ness of the PV.
2975 =for apidoc sv_copypv_nomg
2977 Like sv_copypv, but doesn't invoke get magic first.
2979 =for apidoc sv_copypv_flags
2981 Implementation of sv_copypv and sv_copypv_nomg. Calls get magic iff flags
2988 Perl_sv_copypv(pTHX_ SV *const dsv, register SV *const ssv)
2990 PERL_ARGS_ASSERT_SV_COPYPV;
2992 sv_copypv_flags(dsv, ssv, 0);
2996 Perl_sv_copypv_flags(pTHX_ SV *const dsv, register SV *const ssv, const I32 flags)
3001 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
3003 if ((flags & SV_GMAGIC) && SvGMAGICAL(ssv))
3005 s = SvPV_nomg_const(ssv,len);
3006 sv_setpvn(dsv,s,len);
3014 =for apidoc sv_2pvbyte
3016 Return a pointer to the byte-encoded representation of the SV, and set *lp
3017 to its length. May cause the SV to be downgraded from UTF-8 as a
3020 Usually accessed via the C<SvPVbyte> macro.
3026 Perl_sv_2pvbyte(pTHX_ register SV *sv, STRLEN *const lp)
3028 PERL_ARGS_ASSERT_SV_2PVBYTE;
3030 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3031 || isGV_with_GP(sv) || SvROK(sv)) {
3032 SV *sv2 = sv_newmortal();
3036 else SvGETMAGIC(sv);
3037 sv_utf8_downgrade(sv,0);
3038 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3042 =for apidoc sv_2pvutf8
3044 Return a pointer to the UTF-8-encoded representation of the SV, and set *lp
3045 to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3047 Usually accessed via the C<SvPVutf8> macro.
3053 Perl_sv_2pvutf8(pTHX_ register SV *sv, STRLEN *const lp)
3055 PERL_ARGS_ASSERT_SV_2PVUTF8;
3057 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3058 || isGV_with_GP(sv) || SvROK(sv))
3059 sv = sv_mortalcopy(sv);
3062 sv_utf8_upgrade_nomg(sv);
3063 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3068 =for apidoc sv_2bool
3070 This macro is only used by sv_true() or its macro equivalent, and only if
3071 the latter's argument is neither SvPOK, SvIOK nor SvNOK.
3072 It calls sv_2bool_flags with the SV_GMAGIC flag.
3074 =for apidoc sv_2bool_flags
3076 This function is only used by sv_true() and friends, and only if
3077 the latter's argument is neither SvPOK, SvIOK nor SvNOK. If the flags
3078 contain SV_GMAGIC, then it does an mg_get() first.
3085 Perl_sv_2bool_flags(pTHX_ register SV *const sv, const I32 flags)
3089 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
3091 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
3097 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
3098 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
3099 return cBOOL(SvTRUE(tmpsv));
3101 return SvRV(sv) != 0;
3103 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
3107 =for apidoc sv_utf8_upgrade
3109 Converts the PV of an SV to its UTF-8-encoded form.
3110 Forces the SV to string form if it is not already.
3111 Will C<mg_get> on C<sv> if appropriate.
3112 Always sets the SvUTF8 flag to avoid future validity checks even
3113 if the whole string is the same in UTF-8 as not.
3114 Returns the number of bytes in the converted string
3116 This is not a general purpose byte encoding to Unicode interface:
3117 use the Encode extension for that.
3119 =for apidoc sv_utf8_upgrade_nomg
3121 Like sv_utf8_upgrade, but doesn't do magic on C<sv>.
3123 =for apidoc sv_utf8_upgrade_flags
3125 Converts the PV of an SV to its UTF-8-encoded form.
3126 Forces the SV to string form if it is not already.
3127 Always sets the SvUTF8 flag to avoid future validity checks even
3128 if all the bytes are invariant in UTF-8.
3129 If C<flags> has C<SV_GMAGIC> bit set,
3130 will C<mg_get> on C<sv> if appropriate, else not.
3131 Returns the number of bytes in the converted string
3132 C<sv_utf8_upgrade> and
3133 C<sv_utf8_upgrade_nomg> are implemented in terms of this function.
3135 This is not a general purpose byte encoding to Unicode interface:
3136 use the Encode extension for that.
3140 The grow version is currently not externally documented. It adds a parameter,
3141 extra, which is the number of unused bytes the string of 'sv' is guaranteed to
3142 have free after it upon return. This allows the caller to reserve extra space
3143 that it intends to fill, to avoid extra grows.
3145 Also externally undocumented for the moment is the flag SV_FORCE_UTF8_UPGRADE,
3146 which can be used to tell this function to not first check to see if there are
3147 any characters that are different in UTF-8 (variant characters) which would
3148 force it to allocate a new string to sv, but to assume there are. Typically
3149 this flag is used by a routine that has already parsed the string to find that
3150 there are such characters, and passes this information on so that the work
3151 doesn't have to be repeated.
3153 (One might think that the calling routine could pass in the position of the
3154 first such variant, so it wouldn't have to be found again. But that is not the
3155 case, because typically when the caller is likely to use this flag, it won't be
3156 calling this routine unless it finds something that won't fit into a byte.
3157 Otherwise it tries to not upgrade and just use bytes. But some things that
3158 do fit into a byte are variants in utf8, and the caller may not have been
3159 keeping track of these.)
3161 If the routine itself changes the string, it adds a trailing NUL. Such a NUL
3162 isn't guaranteed due to having other routines do the work in some input cases,
3163 or if the input is already flagged as being in utf8.
3165 The speed of this could perhaps be improved for many cases if someone wanted to
3166 write a fast function that counts the number of variant characters in a string,
3167 especially if it could return the position of the first one.
3172 Perl_sv_utf8_upgrade_flags_grow(pTHX_ register SV *const sv, const I32 flags, STRLEN extra)
3176 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
3178 if (sv == &PL_sv_undef)
3180 if (!SvPOK_nog(sv)) {
3182 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3183 (void) sv_2pv_flags(sv,&len, flags);
3185 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3189 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
3194 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3199 sv_force_normal_flags(sv, 0);
3202 if (PL_encoding && !(flags & SV_UTF8_NO_ENCODING)) {
3203 sv_recode_to_utf8(sv, PL_encoding);
3204 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3208 if (SvCUR(sv) == 0) {
3209 if (extra) SvGROW(sv, extra);
3210 } else { /* Assume Latin-1/EBCDIC */
3211 /* This function could be much more efficient if we
3212 * had a FLAG in SVs to signal if there are any variant
3213 * chars in the PV. Given that there isn't such a flag
3214 * make the loop as fast as possible (although there are certainly ways
3215 * to speed this up, eg. through vectorization) */
3216 U8 * s = (U8 *) SvPVX_const(sv);
3217 U8 * e = (U8 *) SvEND(sv);
3219 STRLEN two_byte_count = 0;
3221 if (flags & SV_FORCE_UTF8_UPGRADE) goto must_be_utf8;
3223 /* See if really will need to convert to utf8. We mustn't rely on our
3224 * incoming SV being well formed and having a trailing '\0', as certain
3225 * code in pp_formline can send us partially built SVs. */
3229 if (NATIVE_IS_INVARIANT(ch)) continue;
3231 t--; /* t already incremented; re-point to first variant */
3236 /* utf8 conversion not needed because all are invariants. Mark as
3237 * UTF-8 even if no variant - saves scanning loop */
3239 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3244 /* Here, the string should be converted to utf8, either because of an
3245 * input flag (two_byte_count = 0), or because a character that
3246 * requires 2 bytes was found (two_byte_count = 1). t points either to
3247 * the beginning of the string (if we didn't examine anything), or to
3248 * the first variant. In either case, everything from s to t - 1 will
3249 * occupy only 1 byte each on output.
3251 * There are two main ways to convert. One is to create a new string
3252 * and go through the input starting from the beginning, appending each
3253 * converted value onto the new string as we go along. It's probably
3254 * best to allocate enough space in the string for the worst possible
3255 * case rather than possibly running out of space and having to
3256 * reallocate and then copy what we've done so far. Since everything
3257 * from s to t - 1 is invariant, the destination can be initialized
3258 * with these using a fast memory copy
3260 * The other way is to figure out exactly how big the string should be
3261 * by parsing the entire input. Then you don't have to make it big
3262 * enough to handle the worst possible case, and more importantly, if
3263 * the string you already have is large enough, you don't have to
3264 * allocate a new string, you can copy the last character in the input
3265 * string to the final position(s) that will be occupied by the
3266 * converted string and go backwards, stopping at t, since everything
3267 * before that is invariant.
3269 * There are advantages and disadvantages to each method.
3271 * In the first method, we can allocate a new string, do the memory
3272 * copy from the s to t - 1, and then proceed through the rest of the
3273 * string byte-by-byte.
3275 * In the second method, we proceed through the rest of the input
3276 * string just calculating how big the converted string will be. Then
3277 * there are two cases:
3278 * 1) if the string has enough extra space to handle the converted
3279 * value. We go backwards through the string, converting until we
3280 * get to the position we are at now, and then stop. If this
3281 * position is far enough along in the string, this method is
3282 * faster than the other method. If the memory copy were the same
3283 * speed as the byte-by-byte loop, that position would be about
3284 * half-way, as at the half-way mark, parsing to the end and back
3285 * is one complete string's parse, the same amount as starting
3286 * over and going all the way through. Actually, it would be
3287 * somewhat less than half-way, as it's faster to just count bytes
3288 * than to also copy, and we don't have the overhead of allocating
3289 * a new string, changing the scalar to use it, and freeing the
3290 * existing one. But if the memory copy is fast, the break-even
3291 * point is somewhere after half way. The counting loop could be
3292 * sped up by vectorization, etc, to move the break-even point
3293 * further towards the beginning.
3294 * 2) if the string doesn't have enough space to handle the converted
3295 * value. A new string will have to be allocated, and one might
3296 * as well, given that, start from the beginning doing the first
3297 * method. We've spent extra time parsing the string and in
3298 * exchange all we've gotten is that we know precisely how big to
3299 * make the new one. Perl is more optimized for time than space,
3300 * so this case is a loser.
3301 * So what I've decided to do is not use the 2nd method unless it is
3302 * guaranteed that a new string won't have to be allocated, assuming
3303 * the worst case. I also decided not to put any more conditions on it
3304 * than this, for now. It seems likely that, since the worst case is
3305 * twice as big as the unknown portion of the string (plus 1), we won't
3306 * be guaranteed enough space, causing us to go to the first method,
3307 * unless the string is short, or the first variant character is near
3308 * the end of it. In either of these cases, it seems best to use the
3309 * 2nd method. The only circumstance I can think of where this would
3310 * be really slower is if the string had once had much more data in it
3311 * than it does now, but there is still a substantial amount in it */
3314 STRLEN invariant_head = t - s;
3315 STRLEN size = invariant_head + (e - t) * 2 + 1 + extra;
3316 if (SvLEN(sv) < size) {
3318 /* Here, have decided to allocate a new string */
3323 Newx(dst, size, U8);
3325 /* If no known invariants at the beginning of the input string,
3326 * set so starts from there. Otherwise, can use memory copy to
3327 * get up to where we are now, and then start from here */
3329 if (invariant_head <= 0) {
3332 Copy(s, dst, invariant_head, char);
3333 d = dst + invariant_head;
3337 const UV uv = NATIVE8_TO_UNI(*t++);
3338 if (UNI_IS_INVARIANT(uv))
3339 *d++ = (U8)UNI_TO_NATIVE(uv);
3341 *d++ = (U8)UTF8_EIGHT_BIT_HI(uv);
3342 *d++ = (U8)UTF8_EIGHT_BIT_LO(uv);
3346 SvPV_free(sv); /* No longer using pre-existing string */
3347 SvPV_set(sv, (char*)dst);
3348 SvCUR_set(sv, d - dst);
3349 SvLEN_set(sv, size);
3352 /* Here, have decided to get the exact size of the string.
3353 * Currently this happens only when we know that there is
3354 * guaranteed enough space to fit the converted string, so
3355 * don't have to worry about growing. If two_byte_count is 0,
3356 * then t points to the first byte of the string which hasn't
3357 * been examined yet. Otherwise two_byte_count is 1, and t
3358 * points to the first byte in the string that will expand to
3359 * two. Depending on this, start examining at t or 1 after t.
3362 U8 *d = t + two_byte_count;
3365 /* Count up the remaining bytes that expand to two */
3368 const U8 chr = *d++;
3369 if (! NATIVE_IS_INVARIANT(chr)) two_byte_count++;
3372 /* The string will expand by just the number of bytes that
3373 * occupy two positions. But we are one afterwards because of
3374 * the increment just above. This is the place to put the
3375 * trailing NUL, and to set the length before we decrement */
3377 d += two_byte_count;
3378 SvCUR_set(sv, d - s);
3382 /* Having decremented d, it points to the position to put the
3383 * very last byte of the expanded string. Go backwards through
3384 * the string, copying and expanding as we go, stopping when we
3385 * get to the part that is invariant the rest of the way down */
3389 const U8 ch = NATIVE8_TO_UNI(*e--);
3390 if (UNI_IS_INVARIANT(ch)) {
3391 *d-- = UNI_TO_NATIVE(ch);
3393 *d-- = (U8)UTF8_EIGHT_BIT_LO(ch);
3394 *d-- = (U8)UTF8_EIGHT_BIT_HI(ch);
3399 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3400 /* Update pos. We do it at the end rather than during
3401 * the upgrade, to avoid slowing down the common case
3402 * (upgrade without pos) */
3403 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3405 I32 pos = mg->mg_len;
3406 if (pos > 0 && (U32)pos > invariant_head) {
3407 U8 *d = (U8*) SvPVX(sv) + invariant_head;
3408 STRLEN n = (U32)pos - invariant_head;
3410 if (UTF8_IS_START(*d))
3415 mg->mg_len = d - (U8*)SvPVX(sv);
3418 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3419 magic_setutf8(sv,mg); /* clear UTF8 cache */
3424 /* Mark as UTF-8 even if no variant - saves scanning loop */
3430 =for apidoc sv_utf8_downgrade
3432 Attempts to convert the PV of an SV from characters to bytes.
3433 If the PV contains a character that cannot fit
3434 in a byte, this conversion will fail;
3435 in this case, either returns false or, if C<fail_ok> is not
3438 This is not a general purpose Unicode to byte encoding interface:
3439 use the Encode extension for that.
3445 Perl_sv_utf8_downgrade(pTHX_ register SV *const sv, const bool fail_ok)
3449 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3451 if (SvPOKp(sv) && SvUTF8(sv)) {
3455 int mg_flags = SV_GMAGIC;
3458 sv_force_normal_flags(sv, 0);
3460 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3462 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3464 I32 pos = mg->mg_len;
3466 sv_pos_b2u(sv, &pos);
3467 mg_flags = 0; /* sv_pos_b2u does get magic */
3471 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3472 magic_setutf8(sv,mg); /* clear UTF8 cache */
3475 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3477 if (!utf8_to_bytes(s, &len)) {
3482 Perl_croak(aTHX_ "Wide character in %s",
3485 Perl_croak(aTHX_ "Wide character");
3496 =for apidoc sv_utf8_encode
3498 Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3499 flag off so that it looks like octets again.
3505 Perl_sv_utf8_encode(pTHX_ register SV *const sv)
3507 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3509 if (SvREADONLY(sv)) {
3510 sv_force_normal_flags(sv, 0);
3512 (void) sv_utf8_upgrade(sv);
3517 =for apidoc sv_utf8_decode
3519 If the PV of the SV is an octet sequence in UTF-8
3520 and contains a multiple-byte character, the C<SvUTF8> flag is turned on
3521 so that it looks like a character. If the PV contains only single-byte
3522 characters, the C<SvUTF8> flag stays off.
3523 Scans PV for validity and returns false if the PV is invalid UTF-8.
3529 Perl_sv_utf8_decode(pTHX_ register SV *const sv)
3531 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3534 const U8 *start, *c;
3537 /* The octets may have got themselves encoded - get them back as
3540 if (!sv_utf8_downgrade(sv, TRUE))
3543 /* it is actually just a matter of turning the utf8 flag on, but
3544 * we want to make sure everything inside is valid utf8 first.
3546 c = start = (const U8 *) SvPVX_const(sv);
3547 if (!is_utf8_string(c, SvCUR(sv)))
3549 e = (const U8 *) SvEND(sv);
3552 if (!UTF8_IS_INVARIANT(ch)) {
3557 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3558 /* adjust pos to the start of a UTF8 char sequence */
3559 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3561 I32 pos = mg->mg_len;
3563 for (c = start + pos; c > start; c--) {
3564 if (UTF8_IS_START(*c))
3567 mg->mg_len = c - start;
3570 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3571 magic_setutf8(sv,mg); /* clear UTF8 cache */
3578 =for apidoc sv_setsv
3580 Copies the contents of the source SV C<ssv> into the destination SV
3581 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3582 function if the source SV needs to be reused. Does not handle 'set' magic.
3583 Loosely speaking, it performs a copy-by-value, obliterating any previous
3584 content of the destination.
3586 You probably want to use one of the assortment of wrappers, such as
3587 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3588 C<SvSetMagicSV_nosteal>.
3590 =for apidoc sv_setsv_flags
3592 Copies the contents of the source SV C<ssv> into the destination SV
3593 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3594 function if the source SV needs to be reused. Does not handle 'set' magic.
3595 Loosely speaking, it performs a copy-by-value, obliterating any previous
3596 content of the destination.
3597 If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
3598 C<ssv> if appropriate, else not. If the C<flags>
3599 parameter has the C<NOSTEAL> bit set then the
3600 buffers of temps will not be stolen. <sv_setsv>
3601 and C<sv_setsv_nomg> are implemented in terms of this function.
3603 You probably want to use one of the assortment of wrappers, such as
3604 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3605 C<SvSetMagicSV_nosteal>.
3607 This is the primary function for copying scalars, and most other
3608 copy-ish functions and macros use this underneath.
3614 S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
3616 I32 mro_changes = 0; /* 1 = method, 2 = isa, 3 = recursive isa */
3617 HV *old_stash = NULL;
3619 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB;
3621 if (dtype != SVt_PVGV && !isGV_with_GP(dstr)) {
3622 const char * const name = GvNAME(sstr);
3623 const STRLEN len = GvNAMELEN(sstr);
3625 if (dtype >= SVt_PV) {
3631 SvUPGRADE(dstr, SVt_PVGV);
3632 (void)SvOK_off(dstr);
3633 /* We have to turn this on here, even though we turn it off
3634 below, as GvSTASH will fail an assertion otherwise. */
3635 isGV_with_GP_on(dstr);
3637 GvSTASH(dstr) = GvSTASH(sstr);
3639 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
3640 gv_name_set(MUTABLE_GV(dstr), name, len,
3641 GV_ADD | (GvNAMEUTF8(sstr) ? SVf_UTF8 : 0 ));
3642 SvFAKE_on(dstr); /* can coerce to non-glob */
3645 if(GvGP(MUTABLE_GV(sstr))) {
3646 /* If source has method cache entry, clear it */
3648 SvREFCNT_dec(GvCV(sstr));
3649 GvCV_set(sstr, NULL);
3652 /* If source has a real method, then a method is
3655 GvCV((const GV *)sstr) && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3661 /* If dest already had a real method, that's a change as well */
3663 !mro_changes && GvGP(MUTABLE_GV(dstr)) && GvCVu((const GV *)dstr)
3664 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3669 /* We don't need to check the name of the destination if it was not a
3670 glob to begin with. */
3671 if(dtype == SVt_PVGV) {
3672 const char * const name = GvNAME((const GV *)dstr);
3675 /* The stash may have been detached from the symbol table, so
3677 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3681 const STRLEN len = GvNAMELEN(dstr);
3682 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
3683 || (len == 1 && name[0] == ':')) {
3686 /* Set aside the old stash, so we can reset isa caches on
3688 if((old_stash = GvHV(dstr)))
3689 /* Make sure we do not lose it early. */
3690 SvREFCNT_inc_simple_void_NN(
3691 sv_2mortal((SV *)old_stash)
3697 gp_free(MUTABLE_GV(dstr));
3698 isGV_with_GP_off(dstr); /* SvOK_off does not like globs. */
3699 (void)SvOK_off(dstr);
3700 isGV_with_GP_on(dstr);
3701 GvINTRO_off(dstr); /* one-shot flag */
3702 GvGP_set(dstr, gp_ref(GvGP(sstr)));
3703 if (SvTAINTED(sstr))
3705 if (GvIMPORTED(dstr) != GVf_IMPORTED
3706 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
3708 GvIMPORTED_on(dstr);
3711 if(mro_changes == 2) {
3712 if (GvAV((const GV *)sstr)) {
3714 SV * const sref = (SV *)GvAV((const GV *)dstr);
3715 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
3716 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
3717 AV * const ary = newAV();
3718 av_push(ary, mg->mg_obj); /* takes the refcount */
3719 mg->mg_obj = (SV *)ary;
3721 av_push((AV *)mg->mg_obj, SvREFCNT_inc_simple_NN(dstr));
3723 else sv_magic(sref, dstr, PERL_MAGIC_isa, NULL, 0);
3725 mro_isa_changed_in(GvSTASH(dstr));
3727 else if(mro_changes == 3) {
3728 HV * const stash = GvHV(dstr);
3729 if(old_stash ? (HV *)HvENAME_get(old_stash) : stash)
3735 else if(mro_changes) mro_method_changed_in(GvSTASH(dstr));
3740 S_glob_assign_ref(pTHX_ SV *const dstr, SV *const sstr)
3742 SV * const sref = SvREFCNT_inc(SvRV(sstr));
3744 const int intro = GvINTRO(dstr);
3747 const U32 stype = SvTYPE(sref);
3749 PERL_ARGS_ASSERT_GLOB_ASSIGN_REF;
3752 GvINTRO_off(dstr); /* one-shot flag */
3753 GvLINE(dstr) = CopLINE(PL_curcop);
3754 GvEGV(dstr) = MUTABLE_GV(dstr);
3759 location = (SV **) &(GvGP(dstr)->gp_cv); /* XXX bypassing GvCV_set */
3760 import_flag = GVf_IMPORTED_CV;
3763 location = (SV **) &GvHV(dstr);
3764 import_flag = GVf_IMPORTED_HV;
3767 location = (SV **) &GvAV(dstr);
3768 import_flag = GVf_IMPORTED_AV;
3771 location = (SV **) &GvIOp(dstr);
3774 location = (SV **) &GvFORM(dstr);
3777 location = &GvSV(dstr);
3778 import_flag = GVf_IMPORTED_SV;
3781 if (stype == SVt_PVCV) {
3782 /*if (GvCVGEN(dstr) && (GvCV(dstr) != (const CV *)sref || GvCVGEN(dstr))) {*/
3783 if (GvCVGEN(dstr)) {
3784 SvREFCNT_dec(GvCV(dstr));
3785 GvCV_set(dstr, NULL);
3786 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
3789 SAVEGENERICSV(*location);
3793 if (stype == SVt_PVCV && (*location != sref || GvCVGEN(dstr))) {
3794 CV* const cv = MUTABLE_CV(*location);
3796 if (!GvCVGEN((const GV *)dstr) &&
3797 (CvROOT(cv) || CvXSUB(cv)) &&
3798 /* redundant check that avoids creating the extra SV
3799 most of the time: */
3800 (CvCONST(cv) || ckWARN(WARN_REDEFINE)))
3802 SV * const new_const_sv =
3803 CvCONST((const CV *)sref)
3804 ? cv_const_sv((const CV *)sref)
3806 report_redefined_cv(
3807 sv_2mortal(Perl_newSVpvf(aTHX_
3810 HvNAME_HEK(GvSTASH((const GV *)dstr))
3812 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr)))
3815 CvCONST((const CV *)sref) ? &new_const_sv : NULL
3819 cv_ckproto_len_flags(cv, (const GV *)dstr,
3820 SvPOK(sref) ? CvPROTO(sref) : NULL,
3821 SvPOK(sref) ? CvPROTOLEN(sref) : 0,
3822 SvPOK(sref) ? SvUTF8(sref) : 0);
3824 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
3825 GvASSUMECV_on(dstr);
3826 if(GvSTASH(dstr)) mro_method_changed_in(GvSTASH(dstr)); /* sub foo { 1 } sub bar { 2 } *bar = \&foo */
3829 if (import_flag && !(GvFLAGS(dstr) & import_flag)
3830 && CopSTASH_ne(PL_curcop, GvSTASH(dstr))) {
3831 GvFLAGS(dstr) |= import_flag;
3833 if (stype == SVt_PVHV) {
3834 const char * const name = GvNAME((GV*)dstr);
3835 const STRLEN len = GvNAMELEN(dstr);
3838 (len > 1 && name[len-2] == ':' && name[len-1] == ':')
3839 || (len == 1 && name[0] == ':')
3841 && (!dref || HvENAME_get(dref))
3844 (HV *)sref, (HV *)dref,
3850 stype == SVt_PVAV && sref != dref
3851 && strEQ(GvNAME((GV*)dstr), "ISA")
3852 /* The stash may have been detached from the symbol table, so
3853 check its name before doing anything. */
3854 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3857 MAGIC * const omg = dref && SvSMAGICAL(dref)
3858 ? mg_find(dref, PERL_MAGIC_isa)
3860 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
3861 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
3862 AV * const ary = newAV();
3863 av_push(ary, mg->mg_obj); /* takes the refcount */
3864 mg->mg_obj = (SV *)ary;
3867 if (SvTYPE(omg->mg_obj) == SVt_PVAV) {
3868 SV **svp = AvARRAY((AV *)omg->mg_obj);
3869 I32 items = AvFILLp((AV *)omg->mg_obj) + 1;
3873 SvREFCNT_inc_simple_NN(*svp++)
3879 SvREFCNT_inc_simple_NN(omg->mg_obj)
3883 av_push((AV *)mg->mg_obj,SvREFCNT_inc_simple_NN(dstr));
3888 sref, omg ? omg->mg_obj : dstr, PERL_MAGIC_isa, NULL, 0
3890 mg = mg_find(sref, PERL_MAGIC_isa);
3892 /* Since the *ISA assignment could have affected more than
3893 one stash, don't call mro_isa_changed_in directly, but let
3894 magic_clearisa do it for us, as it already has the logic for
3895 dealing with globs vs arrays of globs. */
3897 Perl_magic_clearisa(aTHX_ NULL, mg);
3899 else if (stype == SVt_PVIO) {
3900 DEBUG_o(Perl_deb(aTHX_ "glob_assign_ref clearing PL_stashcache\n"));
3901 /* It's a cache. It will rebuild itself quite happily.
3902 It's a lot of effort to work out exactly which key (or keys)
3903 might be invalidated by the creation of the this file handle.
3905 hv_clear(PL_stashcache);
3910 if (SvTAINTED(sstr))
3916 Perl_sv_setsv_flags(pTHX_ SV *dstr, register SV* sstr, const I32 flags)
3923 PERL_ARGS_ASSERT_SV_SETSV_FLAGS;
3928 if (SvIS_FREED(dstr)) {
3929 Perl_croak(aTHX_ "panic: attempt to copy value %" SVf
3930 " to a freed scalar %p", SVfARG(sstr), (void *)dstr);
3932 SV_CHECK_THINKFIRST_COW_DROP(dstr);
3934 sstr = &PL_sv_undef;
3935 if (SvIS_FREED(sstr)) {
3936 Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p to %p",
3937 (void*)sstr, (void*)dstr);
3939 stype = SvTYPE(sstr);
3940 dtype = SvTYPE(dstr);
3942 /* There's a lot of redundancy below but we're going for speed here */
3947 if (dtype != SVt_PVGV && dtype != SVt_PVLV) {
3948 (void)SvOK_off(dstr);
3956 sv_upgrade(dstr, SVt_IV);
3960 sv_upgrade(dstr, SVt_PVIV);
3964 goto end_of_first_switch;
3966 (void)SvIOK_only(dstr);
3967 SvIV_set(dstr, SvIVX(sstr));
3970 /* SvTAINTED can only be true if the SV has taint magic, which in
3971 turn means that the SV type is PVMG (or greater). This is the
3972 case statement for SVt_IV, so this cannot be true (whatever gcov
3974 assert(!SvTAINTED(sstr));
3979 if (dtype < SVt_PV && dtype != SVt_IV)
3980 sv_upgrade(dstr, SVt_IV);
3988 sv_upgrade(dstr, SVt_NV);
3992 sv_upgrade(dstr, SVt_PVNV);
3996 goto end_of_first_switch;
3998 SvNV_set(dstr, SvNVX(sstr));
3999 (void)SvNOK_only(dstr);
4000 /* SvTAINTED can only be true if the SV has taint magic, which in
4001 turn means that the SV type is PVMG (or greater). This is the
4002 case statement for SVt_NV, so this cannot be true (whatever gcov
4004 assert(!SvTAINTED(sstr));
4011 sv_upgrade(dstr, SVt_PV);
4014 if (dtype < SVt_PVIV)
4015 sv_upgrade(dstr, SVt_PVIV);
4018 if (dtype < SVt_PVNV)
4019 sv_upgrade(dstr, SVt_PVNV);
4023 const char * const type = sv_reftype(sstr,0);
4025 /* diag_listed_as: Bizarre copy of %s */
4026 Perl_croak(aTHX_ "Bizarre copy of %s in %s", type, OP_DESC(PL_op));
4028 Perl_croak(aTHX_ "Bizarre copy of %s", type);
4034 if (dtype < SVt_REGEXP)
4036 if (dtype >= SVt_PV) {
4042 sv_upgrade(dstr, SVt_REGEXP);
4046 /* case SVt_BIND: */
4050 if (SvGMAGICAL(sstr) && (flags & SV_GMAGIC)) {
4052 if (SvTYPE(sstr) != stype)
4053 stype = SvTYPE(sstr);
4055 if (isGV_with_GP(sstr) && dtype <= SVt_PVLV) {
4056 glob_assign_glob(dstr, sstr, dtype);
4059 if (stype == SVt_PVLV)
4061 if (isREGEXP(sstr)) goto upgregexp;
4062 SvUPGRADE(dstr, SVt_PVNV);
4065 SvUPGRADE(dstr, (svtype)stype);
4067 end_of_first_switch:
4069 /* dstr may have been upgraded. */
4070 dtype = SvTYPE(dstr);
4071 sflags = SvFLAGS(sstr);
4073 if (dtype == SVt_PVCV) {
4074 /* Assigning to a subroutine sets the prototype. */
4077 const char *const ptr = SvPV_const(sstr, len);
4079 SvGROW(dstr, len + 1);
4080 Copy(ptr, SvPVX(dstr), len + 1, char);
4081 SvCUR_set(dstr, len);
4083 SvFLAGS(dstr) |= sflags & SVf_UTF8;
4084 CvAUTOLOAD_off(dstr);
4089 else if (dtype == SVt_PVAV || dtype == SVt_PVHV || dtype == SVt_PVFM) {
4090 const char * const type = sv_reftype(dstr,0);
4092 /* diag_listed_as: Cannot copy to %s */
4093 Perl_croak(aTHX_ "Cannot copy to %s in %s", type, OP_DESC(PL_op));
4095 Perl_croak(aTHX_ "Cannot copy to %s", type);
4096 } else if (sflags & SVf_ROK) {
4097 if (isGV_with_GP(dstr)
4098 && SvTYPE(SvRV(sstr)) == SVt_PVGV && isGV_with_GP(SvRV(sstr))) {
4101 if (GvIMPORTED(dstr) != GVf_IMPORTED
4102 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
4104 GvIMPORTED_on(dstr);
4109 glob_assign_glob(dstr, sstr, dtype);
4113 if (dtype >= SVt_PV) {
4114 if (isGV_with_GP(dstr)) {
4115 glob_assign_ref(dstr, sstr);
4118 if (SvPVX_const(dstr)) {
4124 (void)SvOK_off(dstr);
4125 SvRV_set(dstr, SvREFCNT_inc(SvRV(sstr)));
4126 SvFLAGS(dstr) |= sflags & SVf_ROK;
4127 assert(!(sflags & SVp_NOK));
4128 assert(!(sflags & SVp_IOK));
4129 assert(!(sflags & SVf_NOK));
4130 assert(!(sflags & SVf_IOK));
4132 else if (isGV_with_GP(dstr)) {
4133 if (!(sflags & SVf_OK)) {
4134 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4135 "Undefined value assigned to typeglob");
4138 GV *gv = gv_fetchsv_nomg(sstr, GV_ADD, SVt_PVGV);
4139 if (dstr != (const SV *)gv) {
4140 const char * const name = GvNAME((const GV *)dstr);
4141 const STRLEN len = GvNAMELEN(dstr);
4142 HV *old_stash = NULL;
4143 bool reset_isa = FALSE;
4144 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
4145 || (len == 1 && name[0] == ':')) {
4146 /* Set aside the old stash, so we can reset isa caches
4147 on its subclasses. */
4148 if((old_stash = GvHV(dstr))) {
4149 /* Make sure we do not lose it early. */
4150 SvREFCNT_inc_simple_void_NN(
4151 sv_2mortal((SV *)old_stash)
4158 gp_free(MUTABLE_GV(dstr));
4159 GvGP_set(dstr, gp_ref(GvGP(gv)));
4162 HV * const stash = GvHV(dstr);
4164 old_stash ? (HV *)HvENAME_get(old_stash) : stash
4174 else if ((dtype == SVt_REGEXP || dtype == SVt_PVLV)
4175 && (stype == SVt_REGEXP || isREGEXP(sstr))) {
4176 reg_temp_copy((REGEXP*)dstr, (REGEXP*)sstr);
4178 else if (sflags & SVp_POK) {
4182 * Check to see if we can just swipe the string. If so, it's a
4183 * possible small lose on short strings, but a big win on long ones.
4184 * It might even be a win on short strings if SvPVX_const(dstr)
4185 * has to be allocated and SvPVX_const(sstr) has to be freed.
4186 * Likewise if we can set up COW rather than doing an actual copy, we
4187 * drop to the else clause, as the swipe code and the COW setup code
4188 * have much in common.
4191 /* Whichever path we take through the next code, we want this true,
4192 and doing it now facilitates the COW check. */
4193 (void)SvPOK_only(dstr);
4196 /* If we're already COW then this clause is not true, and if COW
4197 is allowed then we drop down to the else and make dest COW
4198 with us. If caller hasn't said that we're allowed to COW
4199 shared hash keys then we don't do the COW setup, even if the
4200 source scalar is a shared hash key scalar. */
4201 (((flags & SV_COW_SHARED_HASH_KEYS)
4202 ? !(sflags & SVf_IsCOW)
4203 : 1 /* If making a COW copy is forbidden then the behaviour we
4204 desire is as if the source SV isn't actually already
4205 COW, even if it is. So we act as if the source flags
4206 are not COW, rather than actually testing them. */
4208 #ifndef PERL_OLD_COPY_ON_WRITE
4209 /* The change that added SV_COW_SHARED_HASH_KEYS makes the logic
4210 when PERL_OLD_COPY_ON_WRITE is defined a little wrong.
4211 Conceptually PERL_OLD_COPY_ON_WRITE being defined should
4212 override SV_COW_SHARED_HASH_KEYS, because it means "always COW"
4213 but in turn, it's somewhat dead code, never expected to go
4214 live, but more kept as a placeholder on how to do it better
4215 in a newer implementation. */
4216 /* If we are COW and dstr is a suitable target then we drop down
4217 into the else and make dest a COW of us. */
4218 || (SvFLAGS(dstr) & SVf_BREAK)
4223 (sflags & SVs_TEMP) && /* slated for free anyway? */
4224 !(sflags & SVf_OOK) && /* and not involved in OOK hack? */
4225 (!(flags & SV_NOSTEAL)) &&
4226 /* and we're allowed to steal temps */
4227 SvREFCNT(sstr) == 1 && /* and no other references to it? */
4228 SvLEN(sstr)) /* and really is a string */
4229 #ifdef PERL_OLD_COPY_ON_WRITE
4230 && ((flags & SV_COW_SHARED_HASH_KEYS)
4231 ? (!((sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4232 && (SvFLAGS(dstr) & CAN_COW_MASK) == CAN_COW_FLAGS
4233 && SvTYPE(sstr) >= SVt_PVIV))
4237 /* Failed the swipe test, and it's not a shared hash key either.
4238 Have to copy the string. */
4239 STRLEN len = SvCUR(sstr);
4240 SvGROW(dstr, len + 1); /* inlined from sv_setpvn */
4241 Move(SvPVX_const(sstr),SvPVX(dstr),len,char);
4242 SvCUR_set(dstr, len);
4243 *SvEND(dstr) = '\0';
4245 /* If PERL_OLD_COPY_ON_WRITE is not defined, then isSwipe will always
4247 /* Either it's a shared hash key, or it's suitable for
4248 copy-on-write or we can swipe the string. */
4250 PerlIO_printf(Perl_debug_log, "Copy on write: sstr --> dstr\n");
4254 #ifdef PERL_OLD_COPY_ON_WRITE
4256 if (!(sflags & SVf_IsCOW)) {
4258 /* Make the source SV into a loop of 1.
4259 (about to become 2) */
4260 SV_COW_NEXT_SV_SET(sstr, sstr);
4264 /* Initial code is common. */
4265 if (SvPVX_const(dstr)) { /* we know that dtype >= SVt_PV */
4270 /* making another shared SV. */
4271 STRLEN cur = SvCUR(sstr);
4272 STRLEN len = SvLEN(sstr);
4273 #ifdef PERL_OLD_COPY_ON_WRITE
4275 assert (SvTYPE(dstr) >= SVt_PVIV);
4276 /* SvIsCOW_normal */
4277 /* splice us in between source and next-after-source. */
4278 SV_COW_NEXT_SV_SET(dstr, SV_COW_NEXT_SV(sstr));
4279 SV_COW_NEXT_SV_SET(sstr, dstr);
4280 SvPV_set(dstr, SvPVX_mutable(sstr));
4284 /* SvIsCOW_shared_hash */
4285 DEBUG_C(PerlIO_printf(Perl_debug_log,
4286 "Copy on write: Sharing hash\n"));
4288 assert (SvTYPE(dstr) >= SVt_PV);
4290 HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)))));
4292 SvLEN_set(dstr, len);
4293 SvCUR_set(dstr, cur);
4297 { /* Passes the swipe test. */
4298 SvPV_set(dstr, SvPVX_mutable(sstr));
4299 SvLEN_set(dstr, SvLEN(sstr));
4300 SvCUR_set(dstr, SvCUR(sstr));
4303 (void)SvOK_off(sstr); /* NOTE: nukes most SvFLAGS on sstr */
4304 SvPV_set(sstr, NULL);
4310 if (sflags & SVp_NOK) {
4311 SvNV_set(dstr, SvNVX(sstr));
4313 if (sflags & SVp_IOK) {
4314 SvIV_set(dstr, SvIVX(sstr));
4315 /* Must do this otherwise some other overloaded use of 0x80000000
4316 gets confused. I guess SVpbm_VALID */
4317 if (sflags & SVf_IVisUV)
4320 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_NOK|SVp_NOK|SVf_UTF8);
4322 const MAGIC * const smg = SvVSTRING_mg(sstr);
4324 sv_magic(dstr, NULL, PERL_MAGIC_vstring,
4325 smg->mg_ptr, smg->mg_len);
4326 SvRMAGICAL_on(dstr);
4330 else if (sflags & (SVp_IOK|SVp_NOK)) {
4331 (void)SvOK_off(dstr);
4332 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_IVisUV|SVf_NOK|SVp_NOK);
4333 if (sflags & SVp_IOK) {
4334 /* XXXX Do we want to set IsUV for IV(ROK)? Be extra safe... */
4335 SvIV_set(dstr, SvIVX(sstr));
4337 if (sflags & SVp_NOK) {
4338 SvNV_set(dstr, SvNVX(sstr));
4342 if (isGV_with_GP(sstr)) {
4343 gv_efullname3(dstr, MUTABLE_GV(sstr), "*");
4346 (void)SvOK_off(dstr);
4348 if (SvTAINTED(sstr))
4353 =for apidoc sv_setsv_mg
4355 Like C<sv_setsv>, but also handles 'set' magic.
4361 Perl_sv_setsv_mg(pTHX_ SV *const dstr, register SV *const sstr)
4363 PERL_ARGS_ASSERT_SV_SETSV_MG;
4365 sv_setsv(dstr,sstr);
4369 #ifdef PERL_OLD_COPY_ON_WRITE
4371 Perl_sv_setsv_cow(pTHX_ SV *dstr, SV *sstr)
4373 STRLEN cur = SvCUR(sstr);
4374 STRLEN len = SvLEN(sstr);
4377 PERL_ARGS_ASSERT_SV_SETSV_COW;
4380 PerlIO_printf(Perl_debug_log, "Fast copy on write: %p -> %p\n",
4381 (void*)sstr, (void*)dstr);
4388 if (SvTHINKFIRST(dstr))
4389 sv_force_normal_flags(dstr, SV_COW_DROP_PV);
4390 else if (SvPVX_const(dstr))
4391 Safefree(SvPVX_mutable(dstr));
4395 SvUPGRADE(dstr, SVt_PVIV);
4397 assert (SvPOK(sstr));
4398 assert (SvPOKp(sstr));
4399 assert (!SvIOK(sstr));
4400 assert (!SvIOKp(sstr));
4401 assert (!SvNOK(sstr));
4402 assert (!SvNOKp(sstr));
4404 if (SvIsCOW(sstr)) {
4406 if (SvLEN(sstr) == 0) {
4407 /* source is a COW shared hash key. */
4408 DEBUG_C(PerlIO_printf(Perl_debug_log,
4409 "Fast copy on write: Sharing hash\n"));
4410 new_pv = HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr))));
4413 SV_COW_NEXT_SV_SET(dstr, SV_COW_NEXT_SV(sstr));
4415 assert ((SvFLAGS(sstr) & CAN_COW_MASK) == CAN_COW_FLAGS);
4416 SvUPGRADE(sstr, SVt_PVIV);
4418 DEBUG_C(PerlIO_printf(Perl_debug_log,
4419 "Fast copy on write: Converting sstr to COW\n"));
4420 SV_COW_NEXT_SV_SET(dstr, sstr);
4422 SV_COW_NEXT_SV_SET(sstr, dstr);
4423 new_pv = SvPVX_mutable(sstr);
4426 SvPV_set(dstr, new_pv);
4427 SvFLAGS(dstr) = (SVt_PVIV|SVf_POK|SVp_POK|SVf_IsCOW);
4430 SvLEN_set(dstr, len);
4431 SvCUR_set(dstr, cur);
4440 =for apidoc sv_setpvn
4442 Copies a string into an SV. The C<len> parameter indicates the number of
4443 bytes to be copied. If the C<ptr> argument is NULL the SV will become
4444 undefined. Does not handle 'set' magic. See C<sv_setpvn_mg>.
4450 Perl_sv_setpvn(pTHX_ register SV *const sv, register const char *const ptr, register const STRLEN len)
4455 PERL_ARGS_ASSERT_SV_SETPVN;
4457 SV_CHECK_THINKFIRST_COW_DROP(sv);
4463 /* len is STRLEN which is unsigned, need to copy to signed */
4466 Perl_croak(aTHX_ "panic: sv_setpvn called with negative strlen %"
4469 SvUPGRADE(sv, SVt_PV);
4471 dptr = SvGROW(sv, len + 1);
4472 Move(ptr,dptr,len,char);
4475 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4477 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4481 =for apidoc sv_setpvn_mg
4483 Like C<sv_setpvn>, but also handles 'set' magic.
4489 Perl_sv_setpvn_mg(pTHX_ register SV *const sv, register const char *const ptr, register const STRLEN len)
4491 PERL_ARGS_ASSERT_SV_SETPVN_MG;
4493 sv_setpvn(sv,ptr,len);
4498 =for apidoc sv_setpv
4500 Copies a string into an SV. The string must be null-terminated. Does not
4501 handle 'set' magic. See C<sv_setpv_mg>.
4507 Perl_sv_setpv(pTHX_ register SV *const sv, register const char *const ptr)
4512 PERL_ARGS_ASSERT_SV_SETPV;
4514 SV_CHECK_THINKFIRST_COW_DROP(sv);
4520 SvUPGRADE(sv, SVt_PV);
4522 SvGROW(sv, len + 1);
4523 Move(ptr,SvPVX(sv),len+1,char);
4525 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4527 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4531 =for apidoc sv_setpv_mg
4533 Like C<sv_setpv>, but also handles 'set' magic.
4539 Perl_sv_setpv_mg(pTHX_ register SV *const sv, register const char *const ptr)
4541 PERL_ARGS_ASSERT_SV_SETPV_MG;
4548 Perl_sv_sethek(pTHX_ register SV *const sv, const HEK *const hek)
4552 PERL_ARGS_ASSERT_SV_SETHEK;
4558 if (HEK_LEN(hek) == HEf_SVKEY) {
4559 sv_setsv(sv, *(SV**)HEK_KEY(hek));
4562 const int flags = HEK_FLAGS(hek);
4563 if (flags & HVhek_WASUTF8) {
4564 STRLEN utf8_len = HEK_LEN(hek);
4565 char *as_utf8 = (char *)bytes_to_utf8((U8*)HEK_KEY(hek), &utf8_len);
4566 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
4569 } else if (flags & HVhek_UNSHARED) {
4570 sv_setpvn(sv, HEK_KEY(hek), HEK_LEN(hek));
4573 else SvUTF8_off(sv);
4577 SV_CHECK_THINKFIRST_COW_DROP(sv);
4578 SvUPGRADE(sv, SVt_PV);
4579 Safefree(SvPVX(sv));
4580 SvPV_set(sv,(char *)HEK_KEY(share_hek_hek(hek)));
4581 SvCUR_set(sv, HEK_LEN(hek));
4587 else SvUTF8_off(sv);
4595 =for apidoc sv_usepvn_flags
4597 Tells an SV to use C<ptr> to find its string value. Normally the
4598 string is stored inside the SV but sv_usepvn allows the SV to use an
4599 outside string. The C<ptr> should point to memory that was allocated
4600 by C<malloc>. It must be the start of a mallocked block
4601 of memory, and not a pointer to the middle of it. The
4602 string length, C<len>, must be supplied. By default
4603 this function will realloc (i.e. move) the memory pointed to by C<ptr>,
4604 so that pointer should not be freed or used by the programmer after
4605 giving it to sv_usepvn, and neither should any pointers from "behind"
4606 that pointer (e.g. ptr + 1) be used.
4608 If C<flags> & SV_SMAGIC is true, will call SvSETMAGIC. If C<flags> &
4609 SV_HAS_TRAILING_NUL is true, then C<ptr[len]> must be NUL, and the realloc
4610 will be skipped (i.e. the buffer is actually at least 1 byte longer than
4611 C<len>, and already meets the requirements for storing in C<SvPVX>).
4617 Perl_sv_usepvn_flags(pTHX_ SV *const sv, char *ptr, const STRLEN len, const U32 flags)
4622 PERL_ARGS_ASSERT_SV_USEPVN_FLAGS;
4624 SV_CHECK_THINKFIRST_COW_DROP(sv);
4625 SvUPGRADE(sv, SVt_PV);
4628 if (flags & SV_SMAGIC)
4632 if (SvPVX_const(sv))
4636 if (flags & SV_HAS_TRAILING_NUL)
4637 assert(ptr[len] == '\0');
4640 allocate = (flags & SV_HAS_TRAILING_NUL)
4642 #ifdef Perl_safesysmalloc_size
4645 PERL_STRLEN_ROUNDUP(len + 1);
4647 if (flags & SV_HAS_TRAILING_NUL) {
4648 /* It's long enough - do nothing.
4649 Specifically Perl_newCONSTSUB is relying on this. */
4652 /* Force a move to shake out bugs in callers. */
4653 char *new_ptr = (char*)safemalloc(allocate);
4654 Copy(ptr, new_ptr, len, char);
4655 PoisonFree(ptr,len,char);
4659 ptr = (char*) saferealloc (ptr, allocate);
4662 #ifdef Perl_safesysmalloc_size
4663 SvLEN_set(sv, Perl_safesysmalloc_size(ptr));
4665 SvLEN_set(sv, allocate);