3 * Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
4 * 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, by Larry Wall and others
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
9 * "I wonder what the Entish is for 'yes' and 'no'," he thought.
12 * This file contains the code that creates, manipulates and destroys
13 * scalar values (SVs). The other types (AV, HV, GV, etc.) reuse the
14 * structure of an SV, so their creation and destruction is handled
15 * here; higher-level functions are in av.c, hv.c, and so on. Opcode
16 * level functions (eg. substr, split, join) for each of the types are
28 /* Missing proto on LynxOS */
29 char *gconvert(double, int, int, char *);
32 #ifdef PERL_UTF8_CACHE_ASSERT
33 /* if adding more checks watch out for the following tests:
34 * t/op/index.t t/op/length.t t/op/pat.t t/op/substr.t
35 * lib/utf8.t lib/Unicode/Collate/t/index.t
38 # define ASSERT_UTF8_CACHE(cache) \
39 STMT_START { if (cache) { assert((cache)[0] <= (cache)[1]); \
40 assert((cache)[2] <= (cache)[3]); \
41 assert((cache)[3] <= (cache)[1]);} \
44 # define ASSERT_UTF8_CACHE(cache) NOOP
47 #ifdef PERL_OLD_COPY_ON_WRITE
48 #define SV_COW_NEXT_SV(sv) INT2PTR(SV *,SvUVX(sv))
49 #define SV_COW_NEXT_SV_SET(current,next) SvUV_set(current, PTR2UV(next))
50 /* This is a pessimistic view. Scalar must be purely a read-write PV to copy-
54 /* ============================================================================
56 =head1 Allocation and deallocation of SVs.
58 An SV (or AV, HV, etc.) is allocated in two parts: the head (struct
59 sv, av, hv...) contains type and reference count information, and for
60 many types, a pointer to the body (struct xrv, xpv, xpviv...), which
61 contains fields specific to each type. Some types store all they need
62 in the head, so don't have a body.
64 In all but the most memory-paranoid configuations (ex: PURIFY), heads
65 and bodies are allocated out of arenas, which by default are
66 approximately 4K chunks of memory parcelled up into N heads or bodies.
67 Sv-bodies are allocated by their sv-type, guaranteeing size
68 consistency needed to allocate safely from arrays.
70 For SV-heads, the first slot in each arena is reserved, and holds a
71 link to the next arena, some flags, and a note of the number of slots.
72 Snaked through each arena chain is a linked list of free items; when
73 this becomes empty, an extra arena is allocated and divided up into N
74 items which are threaded into the free list.
76 SV-bodies are similar, but they use arena-sets by default, which
77 separate the link and info from the arena itself, and reclaim the 1st
78 slot in the arena. SV-bodies are further described later.
80 The following global variables are associated with arenas:
82 PL_sv_arenaroot pointer to list of SV arenas
83 PL_sv_root pointer to list of free SV structures
85 PL_body_arenas head of linked-list of body arenas
86 PL_body_roots[] array of pointers to list of free bodies of svtype
87 arrays are indexed by the svtype needed
89 A few special SV heads are not allocated from an arena, but are
90 instead directly created in the interpreter structure, eg PL_sv_undef.
91 The size of arenas can be changed from the default by setting
92 PERL_ARENA_SIZE appropriately at compile time.
94 The SV arena serves the secondary purpose of allowing still-live SVs
95 to be located and destroyed during final cleanup.
97 At the lowest level, the macros new_SV() and del_SV() grab and free
98 an SV head. (If debugging with -DD, del_SV() calls the function S_del_sv()
99 to return the SV to the free list with error checking.) new_SV() calls
100 more_sv() / sv_add_arena() to add an extra arena if the free list is empty.
101 SVs in the free list have their SvTYPE field set to all ones.
103 At the time of very final cleanup, sv_free_arenas() is called from
104 perl_destruct() to physically free all the arenas allocated since the
105 start of the interpreter.
107 The function visit() scans the SV arenas list, and calls a specified
108 function for each SV it finds which is still live - ie which has an SvTYPE
109 other than all 1's, and a non-zero SvREFCNT. visit() is used by the
110 following functions (specified as [function that calls visit()] / [function
111 called by visit() for each SV]):
113 sv_report_used() / do_report_used()
114 dump all remaining SVs (debugging aid)
116 sv_clean_objs() / do_clean_objs(),do_clean_named_objs()
117 Attempt to free all objects pointed to by RVs,
118 and, unless DISABLE_DESTRUCTOR_KLUDGE is defined,
119 try to do the same for all objects indirectly
120 referenced by typeglobs too. Called once from
121 perl_destruct(), prior to calling sv_clean_all()
124 sv_clean_all() / do_clean_all()
125 SvREFCNT_dec(sv) each remaining SV, possibly
126 triggering an sv_free(). It also sets the
127 SVf_BREAK flag on the SV to indicate that the
128 refcnt has been artificially lowered, and thus
129 stopping sv_free() from giving spurious warnings
130 about SVs which unexpectedly have a refcnt
131 of zero. called repeatedly from perl_destruct()
132 until there are no SVs left.
134 =head2 Arena allocator API Summary
136 Private API to rest of sv.c
140 new_XIV(), del_XIV(),
141 new_XNV(), del_XNV(),
146 sv_report_used(), sv_clean_objs(), sv_clean_all(), sv_free_arenas()
150 ============================================================================ */
153 * "A time to plant, and a time to uproot what was planted..."
157 Perl_offer_nice_chunk(pTHX_ void *const chunk, const U32 chunk_size)
163 PERL_ARGS_ASSERT_OFFER_NICE_CHUNK;
165 new_chunk = (void *)(chunk);
166 new_chunk_size = (chunk_size);
167 if (new_chunk_size > PL_nice_chunk_size) {
168 Safefree(PL_nice_chunk);
169 PL_nice_chunk = (char *) new_chunk;
170 PL_nice_chunk_size = new_chunk_size;
176 #ifdef DEBUG_LEAKING_SCALARS
177 # define FREE_SV_DEBUG_FILE(sv) Safefree((sv)->sv_debug_file)
179 # define FREE_SV_DEBUG_FILE(sv)
183 # define SvARENA_CHAIN(sv) ((sv)->sv_u.svu_rv)
184 /* Whilst I'd love to do this, it seems that things like to check on
186 # define POSION_SV_HEAD(sv) PoisonNew(sv, 1, struct STRUCT_SV)
188 # define POSION_SV_HEAD(sv) PoisonNew(&SvANY(sv), 1, void *), \
189 PoisonNew(&SvREFCNT(sv), 1, U32)
191 # define SvARENA_CHAIN(sv) SvANY(sv)
192 # define POSION_SV_HEAD(sv)
195 #define plant_SV(p) \
197 FREE_SV_DEBUG_FILE(p); \
199 SvARENA_CHAIN(p) = (void *)PL_sv_root; \
200 SvFLAGS(p) = SVTYPEMASK; \
205 #define uproot_SV(p) \
208 PL_sv_root = (SV*)SvARENA_CHAIN(p); \
213 /* make some more SVs by adding another arena */
222 sv_add_arena(PL_nice_chunk, PL_nice_chunk_size, 0);
223 PL_nice_chunk = NULL;
224 PL_nice_chunk_size = 0;
227 char *chunk; /* must use New here to match call to */
228 Newx(chunk,PERL_ARENA_SIZE,char); /* Safefree() in sv_free_arenas() */
229 sv_add_arena(chunk, PERL_ARENA_SIZE, 0);
235 /* new_SV(): return a new, empty SV head */
237 #ifdef DEBUG_LEAKING_SCALARS
238 /* provide a real function for a debugger to play with */
247 sv = S_more_sv(aTHX);
251 sv->sv_debug_optype = PL_op ? PL_op->op_type : 0;
252 sv->sv_debug_line = (U16) (PL_parser
253 ? PL_parser->copline == NOLINE
259 sv->sv_debug_inpad = 0;
260 sv->sv_debug_cloned = 0;
261 sv->sv_debug_file = PL_curcop ? savepv(CopFILE(PL_curcop)): NULL;
265 # define new_SV(p) (p)=S_new_SV(aTHX)
273 (p) = S_more_sv(aTHX); \
281 /* del_SV(): return an empty SV head to the free list */
294 S_del_sv(pTHX_ SV *p)
298 PERL_ARGS_ASSERT_DEL_SV;
303 for (sva = PL_sv_arenaroot; sva; sva = (SV *) SvANY(sva)) {
304 const SV * const sv = sva + 1;
305 const SV * const svend = &sva[SvREFCNT(sva)];
306 if (p >= sv && p < svend) {
312 if (ckWARN_d(WARN_INTERNAL))
313 Perl_warner(aTHX_ packWARN(WARN_INTERNAL),
314 "Attempt to free non-arena SV: 0x%"UVxf
315 pTHX__FORMAT, PTR2UV(p) pTHX__VALUE);
322 #else /* ! DEBUGGING */
324 #define del_SV(p) plant_SV(p)
326 #endif /* DEBUGGING */
330 =head1 SV Manipulation Functions
332 =for apidoc sv_add_arena
334 Given a chunk of memory, link it to the head of the list of arenas,
335 and split it into a list of free SVs.
341 Perl_sv_add_arena(pTHX_ char *const ptr, const U32 size, const U32 flags)
344 SV* const sva = (SV*)ptr;
348 PERL_ARGS_ASSERT_SV_ADD_ARENA;
350 /* The first SV in an arena isn't an SV. */
351 SvANY(sva) = (void *) PL_sv_arenaroot; /* ptr to next arena */
352 SvREFCNT(sva) = size / sizeof(SV); /* number of SV slots */
353 SvFLAGS(sva) = flags; /* FAKE if not to be freed */
355 PL_sv_arenaroot = sva;
356 PL_sv_root = sva + 1;
358 svend = &sva[SvREFCNT(sva) - 1];
361 SvARENA_CHAIN(sv) = (void *)(SV*)(sv + 1);
365 /* Must always set typemask because it's always checked in on cleanup
366 when the arenas are walked looking for objects. */
367 SvFLAGS(sv) = SVTYPEMASK;
370 SvARENA_CHAIN(sv) = 0;
374 SvFLAGS(sv) = SVTYPEMASK;
377 /* visit(): call the named function for each non-free SV in the arenas
378 * whose flags field matches the flags/mask args. */
381 S_visit(pTHX_ SVFUNC_t f, const U32 flags, const U32 mask)
387 PERL_ARGS_ASSERT_VISIT;
389 for (sva = PL_sv_arenaroot; sva; sva = (SV*)SvANY(sva)) {
390 register const SV * const svend = &sva[SvREFCNT(sva)];
392 for (sv = sva + 1; sv < svend; ++sv) {
393 if (SvTYPE(sv) != SVTYPEMASK
394 && (sv->sv_flags & mask) == flags
407 /* called by sv_report_used() for each live SV */
410 do_report_used(pTHX_ SV *const sv)
412 if (SvTYPE(sv) != SVTYPEMASK) {
413 PerlIO_printf(Perl_debug_log, "****\n");
420 =for apidoc sv_report_used
422 Dump the contents of all SVs not yet freed. (Debugging aid).
428 Perl_sv_report_used(pTHX)
431 visit(do_report_used, 0, 0);
437 /* called by sv_clean_objs() for each live SV */
440 do_clean_objs(pTHX_ SV *const ref)
445 SV * const target = SvRV(ref);
446 if (SvOBJECT(target)) {
447 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning object ref:\n "), sv_dump(ref)));
448 if (SvWEAKREF(ref)) {
449 sv_del_backref(target, ref);
455 SvREFCNT_dec(target);
460 /* XXX Might want to check arrays, etc. */
463 /* called by sv_clean_objs() for each live SV */
465 #ifndef DISABLE_DESTRUCTOR_KLUDGE
467 do_clean_named_objs(pTHX_ SV *const sv)
470 assert(SvTYPE(sv) == SVt_PVGV);
471 assert(isGV_with_GP(sv));
474 #ifdef PERL_DONT_CREATE_GVSV
477 SvOBJECT(GvSV(sv))) ||
478 (GvAV(sv) && SvOBJECT(GvAV(sv))) ||
479 (GvHV(sv) && SvOBJECT(GvHV(sv))) ||
480 /* In certain rare cases GvIOp(sv) can be NULL, which would make SvOBJECT(GvIO(sv)) dereference NULL. */
481 (GvIO(sv) ? (SvFLAGS(GvIOp(sv)) & SVs_OBJECT) : 0) ||
482 (GvCV(sv) && SvOBJECT(GvCV(sv))) )
484 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning named glob object:\n "), sv_dump(sv)));
485 SvFLAGS(sv) |= SVf_BREAK;
493 =for apidoc sv_clean_objs
495 Attempt to destroy all objects not yet freed
501 Perl_sv_clean_objs(pTHX)
504 PL_in_clean_objs = TRUE;
505 visit(do_clean_objs, SVf_ROK, SVf_ROK);
506 #ifndef DISABLE_DESTRUCTOR_KLUDGE
507 /* some barnacles may yet remain, clinging to typeglobs */
508 visit(do_clean_named_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
510 PL_in_clean_objs = FALSE;
513 /* called by sv_clean_all() for each live SV */
516 do_clean_all(pTHX_ SV *const sv)
519 if (sv == (SV*) PL_fdpid || sv == (SV *)PL_strtab) {
520 /* don't clean pid table and strtab */
523 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning loops: SV at 0x%"UVxf"\n", PTR2UV(sv)) ));
524 SvFLAGS(sv) |= SVf_BREAK;
529 =for apidoc sv_clean_all
531 Decrement the refcnt of each remaining SV, possibly triggering a
532 cleanup. This function may have to be called multiple times to free
533 SVs which are in complex self-referential hierarchies.
539 Perl_sv_clean_all(pTHX)
543 PL_in_clean_all = TRUE;
544 cleaned = visit(do_clean_all, 0,0);
545 PL_in_clean_all = FALSE;
550 ARENASETS: a meta-arena implementation which separates arena-info
551 into struct arena_set, which contains an array of struct
552 arena_descs, each holding info for a single arena. By separating
553 the meta-info from the arena, we recover the 1st slot, formerly
554 borrowed for list management. The arena_set is about the size of an
555 arena, avoiding the needless malloc overhead of a naive linked-list.
557 The cost is 1 arena-set malloc per ~320 arena-mallocs, + the unused
558 memory in the last arena-set (1/2 on average). In trade, we get
559 back the 1st slot in each arena (ie 1.7% of a CV-arena, less for
560 smaller types). The recovery of the wasted space allows use of
561 small arenas for large, rare body types, by changing array* fields
562 in body_details_by_type[] below.
565 char *arena; /* the raw storage, allocated aligned */
566 size_t size; /* its size ~4k typ */
567 U32 misc; /* type, and in future other things. */
572 /* Get the maximum number of elements in set[] such that struct arena_set
573 will fit within PERL_ARENA_SIZE, which is probably just under 4K, and
574 therefore likely to be 1 aligned memory page. */
576 #define ARENAS_PER_SET ((PERL_ARENA_SIZE - sizeof(struct arena_set*) \
577 - 2 * sizeof(int)) / sizeof (struct arena_desc))
580 struct arena_set* next;
581 unsigned int set_size; /* ie ARENAS_PER_SET */
582 unsigned int curr; /* index of next available arena-desc */
583 struct arena_desc set[ARENAS_PER_SET];
587 =for apidoc sv_free_arenas
589 Deallocate the memory used by all arenas. Note that all the individual SV
590 heads and bodies within the arenas must already have been freed.
595 Perl_sv_free_arenas(pTHX)
602 /* Free arenas here, but be careful about fake ones. (We assume
603 contiguity of the fake ones with the corresponding real ones.) */
605 for (sva = PL_sv_arenaroot; sva; sva = svanext) {
606 svanext = (SV*) SvANY(sva);
607 while (svanext && SvFAKE(svanext))
608 svanext = (SV*) SvANY(svanext);
615 struct arena_set *aroot = (struct arena_set*) PL_body_arenas;
618 struct arena_set *current = aroot;
621 assert(aroot->set[i].arena);
622 Safefree(aroot->set[i].arena);
630 i = PERL_ARENA_ROOTS_SIZE;
632 PL_body_roots[i] = 0;
634 Safefree(PL_nice_chunk);
635 PL_nice_chunk = NULL;
636 PL_nice_chunk_size = 0;
642 Here are mid-level routines that manage the allocation of bodies out
643 of the various arenas. There are 5 kinds of arenas:
645 1. SV-head arenas, which are discussed and handled above
646 2. regular body arenas
647 3. arenas for reduced-size bodies
649 5. pte arenas (thread related)
651 Arena types 2 & 3 are chained by body-type off an array of
652 arena-root pointers, which is indexed by svtype. Some of the
653 larger/less used body types are malloced singly, since a large
654 unused block of them is wasteful. Also, several svtypes dont have
655 bodies; the data fits into the sv-head itself. The arena-root
656 pointer thus has a few unused root-pointers (which may be hijacked
657 later for arena types 4,5)
659 3 differs from 2 as an optimization; some body types have several
660 unused fields in the front of the structure (which are kept in-place
661 for consistency). These bodies can be allocated in smaller chunks,
662 because the leading fields arent accessed. Pointers to such bodies
663 are decremented to point at the unused 'ghost' memory, knowing that
664 the pointers are used with offsets to the real memory.
666 HE, HEK arenas are managed separately, with separate code, but may
667 be merge-able later..
669 PTE arenas are not sv-bodies, but they share these mid-level
670 mechanics, so are considered here. The new mid-level mechanics rely
671 on the sv_type of the body being allocated, so we just reserve one
672 of the unused body-slots for PTEs, then use it in those (2) PTE
673 contexts below (line ~10k)
676 /* get_arena(size): this creates custom-sized arenas
677 TBD: export properly for hv.c: S_more_he().
680 Perl_get_arena(pTHX_ const size_t arena_size, const U32 misc)
683 struct arena_desc* adesc;
684 struct arena_set *aroot = (struct arena_set*) PL_body_arenas;
687 /* shouldnt need this
688 if (!arena_size) arena_size = PERL_ARENA_SIZE;
691 /* may need new arena-set to hold new arena */
692 if (!aroot || aroot->curr >= aroot->set_size) {
693 struct arena_set *newroot;
694 Newxz(newroot, 1, struct arena_set);
695 newroot->set_size = ARENAS_PER_SET;
696 newroot->next = aroot;
698 PL_body_arenas = (void *) newroot;
699 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot));
702 /* ok, now have arena-set with at least 1 empty/available arena-desc */
703 curr = aroot->curr++;
704 adesc = &(aroot->set[curr]);
705 assert(!adesc->arena);
707 Newx(adesc->arena, arena_size, char);
708 adesc->size = arena_size;
710 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %"UVuf"\n",
711 curr, (void*)adesc->arena, (UV)arena_size));
717 /* return a thing to the free list */
719 #define del_body(thing, root) \
721 void ** const thing_copy = (void **)thing;\
722 *thing_copy = *root; \
723 *root = (void*)thing_copy; \
728 =head1 SV-Body Allocation
730 Allocation of SV-bodies is similar to SV-heads, differing as follows;
731 the allocation mechanism is used for many body types, so is somewhat
732 more complicated, it uses arena-sets, and has no need for still-live
735 At the outermost level, (new|del)_X*V macros return bodies of the
736 appropriate type. These macros call either (new|del)_body_type or
737 (new|del)_body_allocated macro pairs, depending on specifics of the
738 type. Most body types use the former pair, the latter pair is used to
739 allocate body types with "ghost fields".
741 "ghost fields" are fields that are unused in certain types, and
742 consequently dont need to actually exist. They are declared because
743 they're part of a "base type", which allows use of functions as
744 methods. The simplest examples are AVs and HVs, 2 aggregate types
745 which don't use the fields which support SCALAR semantics.
747 For these types, the arenas are carved up into *_allocated size
748 chunks, we thus avoid wasted memory for those unaccessed members.
749 When bodies are allocated, we adjust the pointer back in memory by the
750 size of the bit not allocated, so it's as if we allocated the full
751 structure. (But things will all go boom if you write to the part that
752 is "not there", because you'll be overwriting the last members of the
753 preceding structure in memory.)
755 We calculate the correction using the STRUCT_OFFSET macro. For
756 example, if xpv_allocated is the same structure as XPV then the two
757 OFFSETs sum to zero, and the pointer is unchanged. If the allocated
758 structure is smaller (no initial NV actually allocated) then the net
759 effect is to subtract the size of the NV from the pointer, to return a
760 new pointer as if an initial NV were actually allocated.
762 This is the same trick as was used for NV and IV bodies. Ironically it
763 doesn't need to be used for NV bodies any more, because NV is now at
764 the start of the structure. IV bodies don't need it either, because
765 they are no longer allocated.
767 In turn, the new_body_* allocators call S_new_body(), which invokes
768 new_body_inline macro, which takes a lock, and takes a body off the
769 linked list at PL_body_roots[sv_type], calling S_more_bodies() if
770 necessary to refresh an empty list. Then the lock is released, and
771 the body is returned.
773 S_more_bodies calls get_arena(), and carves it up into an array of N
774 bodies, which it strings into a linked list. It looks up arena-size
775 and body-size from the body_details table described below, thus
776 supporting the multiple body-types.
778 If PURIFY is defined, or PERL_ARENA_SIZE=0, arenas are not used, and
779 the (new|del)_X*V macros are mapped directly to malloc/free.
785 For each sv-type, struct body_details bodies_by_type[] carries
786 parameters which control these aspects of SV handling:
788 Arena_size determines whether arenas are used for this body type, and if
789 so, how big they are. PURIFY or PERL_ARENA_SIZE=0 set this field to
790 zero, forcing individual mallocs and frees.
792 Body_size determines how big a body is, and therefore how many fit into
793 each arena. Offset carries the body-pointer adjustment needed for
794 *_allocated body types, and is used in *_allocated macros.
796 But its main purpose is to parameterize info needed in
797 Perl_sv_upgrade(). The info here dramatically simplifies the function
798 vs the implementation in 5.8.7, making it table-driven. All fields
799 are used for this, except for arena_size.
801 For the sv-types that have no bodies, arenas are not used, so those
802 PL_body_roots[sv_type] are unused, and can be overloaded. In
803 something of a special case, SVt_NULL is borrowed for HE arenas;
804 PL_body_roots[HE_SVSLOT=SVt_NULL] is filled by S_more_he, but the
805 bodies_by_type[SVt_NULL] slot is not used, as the table is not
808 PTEs also use arenas, but are never seen in Perl_sv_upgrade. Nonetheless,
809 they get their own slot in bodies_by_type[PTE_SVSLOT =SVt_IV], so they can
810 just use the same allocation semantics. At first, PTEs were also
811 overloaded to a non-body sv-type, but this yielded hard-to-find malloc
812 bugs, so was simplified by claiming a new slot. This choice has no
813 consequence at this time.
817 struct body_details {
818 U8 body_size; /* Size to allocate */
819 U8 copy; /* Size of structure to copy (may be shorter) */
821 unsigned int type : 4; /* We have space for a sanity check. */
822 unsigned int cant_upgrade : 1; /* Cannot upgrade this type */
823 unsigned int zero_nv : 1; /* zero the NV when upgrading from this */
824 unsigned int arena : 1; /* Allocated from an arena */
825 size_t arena_size; /* Size of arena to allocate */
833 /* With -DPURFIY we allocate everything directly, and don't use arenas.
834 This seems a rather elegant way to simplify some of the code below. */
835 #define HASARENA FALSE
837 #define HASARENA TRUE
839 #define NOARENA FALSE
841 /* Size the arenas to exactly fit a given number of bodies. A count
842 of 0 fits the max number bodies into a PERL_ARENA_SIZE.block,
843 simplifying the default. If count > 0, the arena is sized to fit
844 only that many bodies, allowing arenas to be used for large, rare
845 bodies (XPVFM, XPVIO) without undue waste. The arena size is
846 limited by PERL_ARENA_SIZE, so we can safely oversize the
849 #define FIT_ARENA0(body_size) \
850 ((size_t)(PERL_ARENA_SIZE / body_size) * body_size)
851 #define FIT_ARENAn(count,body_size) \
852 ( count * body_size <= PERL_ARENA_SIZE) \
853 ? count * body_size \
854 : FIT_ARENA0 (body_size)
855 #define FIT_ARENA(count,body_size) \
857 ? FIT_ARENAn (count, body_size) \
858 : FIT_ARENA0 (body_size)
860 /* A macro to work out the offset needed to subtract from a pointer to (say)
867 to make its members accessible via a pointer to (say)
877 #define relative_STRUCT_OFFSET(longer, shorter, member) \
878 (STRUCT_OFFSET(shorter, member) - STRUCT_OFFSET(longer, member))
880 /* Calculate the length to copy. Specifically work out the length less any
881 final padding the compiler needed to add. See the comment in sv_upgrade
882 for why copying the padding proved to be a bug. */
884 #define copy_length(type, last_member) \
885 STRUCT_OFFSET(type, last_member) \
886 + sizeof (((type*)SvANY((SV*)0))->last_member)
888 static const struct body_details bodies_by_type[] = {
889 { sizeof(HE), 0, 0, SVt_NULL,
890 FALSE, NONV, NOARENA, FIT_ARENA(0, sizeof(HE)) },
892 /* The bind placeholder pretends to be an RV for now.
893 Also it's marked as "can't upgrade" to stop anyone using it before it's
895 { 0, 0, 0, SVt_BIND, TRUE, NONV, NOARENA, 0 },
897 /* IVs are in the head, so the allocation size is 0.
898 However, the slot is overloaded for PTEs. */
899 { sizeof(struct ptr_tbl_ent), /* This is used for PTEs. */
900 sizeof(IV), /* This is used to copy out the IV body. */
901 STRUCT_OFFSET(XPVIV, xiv_iv), SVt_IV, FALSE, NONV,
902 NOARENA /* IVS don't need an arena */,
903 /* But PTEs need to know the size of their arena */
904 FIT_ARENA(0, sizeof(struct ptr_tbl_ent))
907 /* 8 bytes on most ILP32 with IEEE doubles */
908 { sizeof(NV), sizeof(NV), 0, SVt_NV, FALSE, HADNV, HASARENA,
909 FIT_ARENA(0, sizeof(NV)) },
911 /* 8 bytes on most ILP32 with IEEE doubles */
912 { sizeof(xpv_allocated),
913 copy_length(XPV, xpv_len)
914 - relative_STRUCT_OFFSET(xpv_allocated, XPV, xpv_cur),
915 + relative_STRUCT_OFFSET(xpv_allocated, XPV, xpv_cur),
916 SVt_PV, FALSE, NONV, HASARENA, FIT_ARENA(0, sizeof(xpv_allocated)) },
919 { sizeof(xpviv_allocated),
920 copy_length(XPVIV, xiv_u)
921 - relative_STRUCT_OFFSET(xpviv_allocated, XPVIV, xpv_cur),
922 + relative_STRUCT_OFFSET(xpviv_allocated, XPVIV, xpv_cur),
923 SVt_PVIV, FALSE, NONV, HASARENA, FIT_ARENA(0, sizeof(xpviv_allocated)) },
926 { sizeof(XPVNV), copy_length(XPVNV, xiv_u), 0, SVt_PVNV, FALSE, HADNV,
927 HASARENA, FIT_ARENA(0, sizeof(XPVNV)) },
930 { sizeof(XPVMG), copy_length(XPVMG, xmg_stash), 0, SVt_PVMG, FALSE, HADNV,
931 HASARENA, FIT_ARENA(0, sizeof(XPVMG)) },
934 { sizeof(struct regexp_allocated), sizeof(struct regexp_allocated),
935 + relative_STRUCT_OFFSET(struct regexp_allocated, regexp, xpv_cur),
936 SVt_REGEXP, FALSE, NONV, HASARENA,
937 FIT_ARENA(0, sizeof(struct regexp_allocated))
941 { sizeof(XPVGV), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV,
942 HASARENA, FIT_ARENA(0, sizeof(XPVGV)) },
945 { sizeof(XPVLV), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV,
946 HASARENA, FIT_ARENA(0, sizeof(XPVLV)) },
948 { sizeof(xpvav_allocated),
949 copy_length(XPVAV, xmg_stash)
950 - relative_STRUCT_OFFSET(xpvav_allocated, XPVAV, xav_fill),
951 + relative_STRUCT_OFFSET(xpvav_allocated, XPVAV, xav_fill),
952 SVt_PVAV, TRUE, NONV, HASARENA, FIT_ARENA(0, sizeof(xpvav_allocated)) },
954 { sizeof(xpvhv_allocated),
955 copy_length(XPVHV, xmg_stash)
956 - relative_STRUCT_OFFSET(xpvhv_allocated, XPVHV, xhv_fill),
957 + relative_STRUCT_OFFSET(xpvhv_allocated, XPVHV, xhv_fill),
958 SVt_PVHV, TRUE, NONV, HASARENA, FIT_ARENA(0, sizeof(xpvhv_allocated)) },
961 { sizeof(xpvcv_allocated), sizeof(xpvcv_allocated),
962 + relative_STRUCT_OFFSET(xpvcv_allocated, XPVCV, xpv_cur),
963 SVt_PVCV, TRUE, NONV, HASARENA, FIT_ARENA(0, sizeof(xpvcv_allocated)) },
965 { sizeof(xpvfm_allocated), sizeof(xpvfm_allocated),
966 + relative_STRUCT_OFFSET(xpvfm_allocated, XPVFM, xpv_cur),
967 SVt_PVFM, TRUE, NONV, NOARENA, FIT_ARENA(20, sizeof(xpvfm_allocated)) },
969 /* XPVIO is 84 bytes, fits 48x */
970 { sizeof(xpvio_allocated), sizeof(xpvio_allocated),
971 + relative_STRUCT_OFFSET(xpvio_allocated, XPVIO, xpv_cur),
972 SVt_PVIO, TRUE, NONV, HASARENA, FIT_ARENA(24, sizeof(xpvio_allocated)) },
975 #define new_body_type(sv_type) \
976 (void *)((char *)S_new_body(aTHX_ sv_type))
978 #define del_body_type(p, sv_type) \
979 del_body(p, &PL_body_roots[sv_type])
982 #define new_body_allocated(sv_type) \
983 (void *)((char *)S_new_body(aTHX_ sv_type) \
984 - bodies_by_type[sv_type].offset)
986 #define del_body_allocated(p, sv_type) \
987 del_body(p + bodies_by_type[sv_type].offset, &PL_body_roots[sv_type])
990 #define my_safemalloc(s) (void*)safemalloc(s)
991 #define my_safecalloc(s) (void*)safecalloc(s, 1)
992 #define my_safefree(p) safefree((char*)p)
996 #define new_XNV() my_safemalloc(sizeof(XPVNV))
997 #define del_XNV(p) my_safefree(p)
999 #define new_XPVNV() my_safemalloc(sizeof(XPVNV))
1000 #define del_XPVNV(p) my_safefree(p)
1002 #define new_XPVAV() my_safemalloc(sizeof(XPVAV))
1003 #define del_XPVAV(p) my_safefree(p)
1005 #define new_XPVHV() my_safemalloc(sizeof(XPVHV))
1006 #define del_XPVHV(p) my_safefree(p)
1008 #define new_XPVMG() my_safemalloc(sizeof(XPVMG))
1009 #define del_XPVMG(p) my_safefree(p)
1011 #define new_XPVGV() my_safemalloc(sizeof(XPVGV))
1012 #define del_XPVGV(p) my_safefree(p)
1016 #define new_XNV() new_body_type(SVt_NV)
1017 #define del_XNV(p) del_body_type(p, SVt_NV)
1019 #define new_XPVNV() new_body_type(SVt_PVNV)
1020 #define del_XPVNV(p) del_body_type(p, SVt_PVNV)
1022 #define new_XPVAV() new_body_allocated(SVt_PVAV)
1023 #define del_XPVAV(p) del_body_allocated(p, SVt_PVAV)
1025 #define new_XPVHV() new_body_allocated(SVt_PVHV)
1026 #define del_XPVHV(p) del_body_allocated(p, SVt_PVHV)
1028 #define new_XPVMG() new_body_type(SVt_PVMG)
1029 #define del_XPVMG(p) del_body_type(p, SVt_PVMG)
1031 #define new_XPVGV() new_body_type(SVt_PVGV)
1032 #define del_XPVGV(p) del_body_type(p, SVt_PVGV)
1036 /* no arena for you! */
1038 #define new_NOARENA(details) \
1039 my_safemalloc((details)->body_size + (details)->offset)
1040 #define new_NOARENAZ(details) \
1041 my_safecalloc((details)->body_size + (details)->offset)
1044 S_more_bodies (pTHX_ const svtype sv_type)
1047 void ** const root = &PL_body_roots[sv_type];
1048 const struct body_details * const bdp = &bodies_by_type[sv_type];
1049 const size_t body_size = bdp->body_size;
1052 const size_t arena_size = Perl_malloc_good_size(bdp->arena_size);
1053 #if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT_PRIVATE)
1054 static bool done_sanity_check;
1056 /* PERL_GLOBAL_STRUCT_PRIVATE cannot coexist with global
1057 * variables like done_sanity_check. */
1058 if (!done_sanity_check) {
1059 unsigned int i = SVt_LAST;
1061 done_sanity_check = TRUE;
1064 assert (bodies_by_type[i].type == i);
1068 assert(bdp->arena_size);
1070 start = (char*) Perl_get_arena(aTHX_ arena_size, sv_type);
1072 end = start + arena_size - 2 * body_size;
1074 /* computed count doesnt reflect the 1st slot reservation */
1075 #if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE)
1076 DEBUG_m(PerlIO_printf(Perl_debug_log,
1077 "arena %p end %p arena-size %d (from %d) type %d "
1079 (void*)start, (void*)end, (int)arena_size,
1080 (int)bdp->arena_size, sv_type, (int)body_size,
1081 (int)arena_size / (int)body_size));
1083 DEBUG_m(PerlIO_printf(Perl_debug_log,
1084 "arena %p end %p arena-size %d type %d size %d ct %d\n",
1085 (void*)start, (void*)end,
1086 (int)bdp->arena_size, sv_type, (int)body_size,
1087 (int)bdp->arena_size / (int)body_size));
1089 *root = (void *)start;
1091 while (start <= end) {
1092 char * const next = start + body_size;
1093 *(void**) start = (void *)next;
1096 *(void **)start = 0;
1101 /* grab a new thing from the free list, allocating more if necessary.
1102 The inline version is used for speed in hot routines, and the
1103 function using it serves the rest (unless PURIFY).
1105 #define new_body_inline(xpv, sv_type) \
1107 void ** const r3wt = &PL_body_roots[sv_type]; \
1108 xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \
1109 ? *((void **)(r3wt)) : more_bodies(sv_type)); \
1110 *(r3wt) = *(void**)(xpv); \
1116 S_new_body(pTHX_ const svtype sv_type)
1120 new_body_inline(xpv, sv_type);
1126 static const struct body_details fake_rv =
1127 { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 };
1130 =for apidoc sv_upgrade
1132 Upgrade an SV to a more complex form. Generally adds a new body type to the
1133 SV, then copies across as much information as possible from the old body.
1134 You generally want to use the C<SvUPGRADE> macro wrapper. See also C<svtype>.
1140 Perl_sv_upgrade(pTHX_ register SV *const sv, svtype new_type)
1145 const svtype old_type = SvTYPE(sv);
1146 const struct body_details *new_type_details;
1147 const struct body_details *old_type_details
1148 = bodies_by_type + old_type;
1149 SV *referant = NULL;
1151 PERL_ARGS_ASSERT_SV_UPGRADE;
1153 if (new_type != SVt_PV && SvIsCOW(sv)) {
1154 sv_force_normal_flags(sv, 0);
1157 if (old_type == new_type)
1160 old_body = SvANY(sv);
1162 /* Copying structures onto other structures that have been neatly zeroed
1163 has a subtle gotcha. Consider XPVMG
1165 +------+------+------+------+------+-------+-------+
1166 | NV | CUR | LEN | IV | MAGIC | STASH |
1167 +------+------+------+------+------+-------+-------+
1168 0 4 8 12 16 20 24 28
1170 where NVs are aligned to 8 bytes, so that sizeof that structure is
1171 actually 32 bytes long, with 4 bytes of padding at the end:
1173 +------+------+------+------+------+-------+-------+------+
1174 | NV | CUR | LEN | IV | MAGIC | STASH | ??? |
1175 +------+------+------+------+------+-------+-------+------+
1176 0 4 8 12 16 20 24 28 32
1178 so what happens if you allocate memory for this structure:
1180 +------+------+------+------+------+-------+-------+------+------+...
1181 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME |
1182 +------+------+------+------+------+-------+-------+------+------+...
1183 0 4 8 12 16 20 24 28 32 36
1185 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you
1186 expect, because you copy the area marked ??? onto GP. Now, ??? may have
1187 started out as zero once, but it's quite possible that it isn't. So now,
1188 rather than a nicely zeroed GP, you have it pointing somewhere random.
1191 (In fact, GP ends up pointing at a previous GP structure, because the
1192 principle cause of the padding in XPVMG getting garbage is a copy of
1193 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now
1194 this happens to be moot because XPVGV has been re-ordered, with GP
1195 no longer after STASH)
1197 So we are careful and work out the size of used parts of all the
1205 referant = SvRV(sv);
1206 old_type_details = &fake_rv;
1207 if (new_type == SVt_NV)
1208 new_type = SVt_PVNV;
1210 if (new_type < SVt_PVIV) {
1211 new_type = (new_type == SVt_NV)
1212 ? SVt_PVNV : SVt_PVIV;
1217 if (new_type < SVt_PVNV) {
1218 new_type = SVt_PVNV;
1222 assert(new_type > SVt_PV);
1223 assert(SVt_IV < SVt_PV);
1224 assert(SVt_NV < SVt_PV);
1231 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena,
1232 there's no way that it can be safely upgraded, because perl.c
1233 expects to Safefree(SvANY(PL_mess_sv)) */
1234 assert(sv != PL_mess_sv);
1235 /* This flag bit is used to mean other things in other scalar types.
1236 Given that it only has meaning inside the pad, it shouldn't be set
1237 on anything that can get upgraded. */
1238 assert(!SvPAD_TYPED(sv));
1241 if (old_type_details->cant_upgrade)
1242 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf,
1243 sv_reftype(sv, 0), (UV) old_type, (UV) new_type);
1246 if (old_type > new_type)
1247 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d",
1248 (int)old_type, (int)new_type);
1250 new_type_details = bodies_by_type + new_type;
1252 SvFLAGS(sv) &= ~SVTYPEMASK;
1253 SvFLAGS(sv) |= new_type;
1255 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of
1256 the return statements above will have triggered. */
1257 assert (new_type != SVt_NULL);
1260 assert(old_type == SVt_NULL);
1261 SvANY(sv) = (XPVIV*)((char*)&(sv->sv_u.svu_iv) - STRUCT_OFFSET(XPVIV, xiv_iv));
1265 assert(old_type == SVt_NULL);
1266 SvANY(sv) = new_XNV();
1271 assert(new_type_details->body_size);
1274 assert(new_type_details->arena);
1275 assert(new_type_details->arena_size);
1276 /* This points to the start of the allocated area. */
1277 new_body_inline(new_body, new_type);
1278 Zero(new_body, new_type_details->body_size, char);
1279 new_body = ((char *)new_body) - new_type_details->offset;
1281 /* We always allocated the full length item with PURIFY. To do this
1282 we fake things so that arena is false for all 16 types.. */
1283 new_body = new_NOARENAZ(new_type_details);
1285 SvANY(sv) = new_body;
1286 if (new_type == SVt_PVAV) {
1290 if (old_type_details->body_size) {
1293 /* It will have been zeroed when the new body was allocated.
1294 Lets not write to it, in case it confuses a write-back
1300 #ifndef NODEFAULT_SHAREKEYS
1301 HvSHAREKEYS_on(sv); /* key-sharing on by default */
1303 HvMAX(sv) = 7; /* (start with 8 buckets) */
1304 if (old_type_details->body_size) {
1307 /* It will have been zeroed when the new body was allocated.
1308 Lets not write to it, in case it confuses a write-back
1313 /* SVt_NULL isn't the only thing upgraded to AV or HV.
1314 The target created by newSVrv also is, and it can have magic.
1315 However, it never has SvPVX set.
1317 if (old_type == SVt_IV) {
1319 } else if (old_type >= SVt_PV) {
1320 assert(SvPVX_const(sv) == 0);
1323 if (old_type >= SVt_PVMG) {
1324 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic);
1325 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash);
1327 sv->sv_u.svu_array = NULL; /* or svu_hash */
1333 /* XXX Is this still needed? Was it ever needed? Surely as there is
1334 no route from NV to PVIV, NOK can never be true */
1335 assert(!SvNOKp(sv));
1347 assert(new_type_details->body_size);
1348 /* We always allocated the full length item with PURIFY. To do this
1349 we fake things so that arena is false for all 16 types.. */
1350 if(new_type_details->arena) {
1351 /* This points to the start of the allocated area. */
1352 new_body_inline(new_body, new_type);
1353 Zero(new_body, new_type_details->body_size, char);
1354 new_body = ((char *)new_body) - new_type_details->offset;
1356 new_body = new_NOARENAZ(new_type_details);
1358 SvANY(sv) = new_body;
1360 if (old_type_details->copy) {
1361 /* There is now the potential for an upgrade from something without
1362 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */
1363 int offset = old_type_details->offset;
1364 int length = old_type_details->copy;
1366 if (new_type_details->offset > old_type_details->offset) {
1367 const int difference
1368 = new_type_details->offset - old_type_details->offset;
1369 offset += difference;
1370 length -= difference;
1372 assert (length >= 0);
1374 Copy((char *)old_body + offset, (char *)new_body + offset, length,
1378 #ifndef NV_ZERO_IS_ALLBITS_ZERO
1379 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a
1380 * correct 0.0 for us. Otherwise, if the old body didn't have an
1381 * NV slot, but the new one does, then we need to initialise the
1382 * freshly created NV slot with whatever the correct bit pattern is
1384 if (old_type_details->zero_nv && !new_type_details->zero_nv
1385 && !isGV_with_GP(sv))
1389 if (new_type == SVt_PVIO)
1390 IoPAGE_LEN(sv) = 60;
1391 if (old_type < SVt_PV) {
1392 /* referant will be NULL unless the old type was SVt_IV emulating
1394 sv->sv_u.svu_rv = referant;
1398 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1399 (unsigned long)new_type);
1402 if (old_type_details->arena) {
1403 /* If there was an old body, then we need to free it.
1404 Note that there is an assumption that all bodies of types that
1405 can be upgraded came from arenas. Only the more complex non-
1406 upgradable types are allowed to be directly malloc()ed. */
1408 my_safefree(old_body);
1410 del_body((void*)((char*)old_body + old_type_details->offset),
1411 &PL_body_roots[old_type]);
1417 =for apidoc sv_backoff
1419 Remove any string offset. You should normally use the C<SvOOK_off> macro
1426 Perl_sv_backoff(pTHX_ register SV *const sv)
1429 const char * const s = SvPVX_const(sv);
1431 PERL_ARGS_ASSERT_SV_BACKOFF;
1432 PERL_UNUSED_CONTEXT;
1435 assert(SvTYPE(sv) != SVt_PVHV);
1436 assert(SvTYPE(sv) != SVt_PVAV);
1438 SvOOK_offset(sv, delta);
1440 SvLEN_set(sv, SvLEN(sv) + delta);
1441 SvPV_set(sv, SvPVX(sv) - delta);
1442 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
1443 SvFLAGS(sv) &= ~SVf_OOK;
1450 Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1451 upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1452 Use the C<SvGROW> wrapper instead.
1458 Perl_sv_grow(pTHX_ register SV *const sv, register STRLEN newlen)
1462 PERL_ARGS_ASSERT_SV_GROW;
1464 if (PL_madskills && newlen >= 0x100000) {
1465 PerlIO_printf(Perl_debug_log,
1466 "Allocation too large: %"UVxf"\n", (UV)newlen);
1468 #ifdef HAS_64K_LIMIT
1469 if (newlen >= 0x10000) {
1470 PerlIO_printf(Perl_debug_log,
1471 "Allocation too large: %"UVxf"\n", (UV)newlen);
1474 #endif /* HAS_64K_LIMIT */
1477 if (SvTYPE(sv) < SVt_PV) {
1478 sv_upgrade(sv, SVt_PV);
1479 s = SvPVX_mutable(sv);
1481 else if (SvOOK(sv)) { /* pv is offset? */
1483 s = SvPVX_mutable(sv);
1484 if (newlen > SvLEN(sv))
1485 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
1486 #ifdef HAS_64K_LIMIT
1487 if (newlen >= 0x10000)
1492 s = SvPVX_mutable(sv);
1494 if (newlen > SvLEN(sv)) { /* need more room? */
1496 newlen = PERL_STRLEN_ROUNDUP(newlen);
1498 if (SvLEN(sv) && s) {
1499 s = (char*)saferealloc(s, newlen);
1502 s = (char*)safemalloc(newlen);
1503 if (SvPVX_const(sv) && SvCUR(sv)) {
1504 Move(SvPVX_const(sv), s, (newlen < SvCUR(sv)) ? newlen : SvCUR(sv), char);
1508 #ifdef Perl_safesysmalloc_size
1509 /* Do this here, do it once, do it right, and then we will never get
1510 called back into sv_grow() unless there really is some growing
1512 SvLEN_set(sv, Perl_safesysmalloc_size(s));
1514 SvLEN_set(sv, newlen);
1521 =for apidoc sv_setiv
1523 Copies an integer into the given SV, upgrading first if necessary.
1524 Does not handle 'set' magic. See also C<sv_setiv_mg>.
1530 Perl_sv_setiv(pTHX_ register SV *const sv, const IV i)
1534 PERL_ARGS_ASSERT_SV_SETIV;
1536 SV_CHECK_THINKFIRST_COW_DROP(sv);
1537 switch (SvTYPE(sv)) {
1540 sv_upgrade(sv, SVt_IV);
1543 sv_upgrade(sv, SVt_PVIV);
1552 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1556 (void)SvIOK_only(sv); /* validate number */
1562 =for apidoc sv_setiv_mg
1564 Like C<sv_setiv>, but also handles 'set' magic.
1570 Perl_sv_setiv_mg(pTHX_ register SV *const sv, const IV i)
1572 PERL_ARGS_ASSERT_SV_SETIV_MG;
1579 =for apidoc sv_setuv
1581 Copies an unsigned integer into the given SV, upgrading first if necessary.
1582 Does not handle 'set' magic. See also C<sv_setuv_mg>.
1588 Perl_sv_setuv(pTHX_ register SV *const sv, const UV u)
1590 PERL_ARGS_ASSERT_SV_SETUV;
1592 /* With these two if statements:
1593 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
1596 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1598 If you wish to remove them, please benchmark to see what the effect is
1600 if (u <= (UV)IV_MAX) {
1601 sv_setiv(sv, (IV)u);
1610 =for apidoc sv_setuv_mg
1612 Like C<sv_setuv>, but also handles 'set' magic.
1618 Perl_sv_setuv_mg(pTHX_ register SV *const sv, const UV u)
1620 PERL_ARGS_ASSERT_SV_SETUV_MG;
1627 =for apidoc sv_setnv
1629 Copies a double into the given SV, upgrading first if necessary.
1630 Does not handle 'set' magic. See also C<sv_setnv_mg>.
1636 Perl_sv_setnv(pTHX_ register SV *const sv, const NV num)
1640 PERL_ARGS_ASSERT_SV_SETNV;
1642 SV_CHECK_THINKFIRST_COW_DROP(sv);
1643 switch (SvTYPE(sv)) {
1646 sv_upgrade(sv, SVt_NV);
1650 sv_upgrade(sv, SVt_PVNV);
1659 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
1664 (void)SvNOK_only(sv); /* validate number */
1669 =for apidoc sv_setnv_mg
1671 Like C<sv_setnv>, but also handles 'set' magic.
1677 Perl_sv_setnv_mg(pTHX_ register SV *const sv, const NV num)
1679 PERL_ARGS_ASSERT_SV_SETNV_MG;
1685 /* Print an "isn't numeric" warning, using a cleaned-up,
1686 * printable version of the offending string
1690 S_not_a_number(pTHX_ SV *const sv)
1697 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1700 dsv = newSVpvs_flags("", SVs_TEMP);
1701 pv = sv_uni_display(dsv, sv, 10, 0);
1704 const char * const limit = tmpbuf + sizeof(tmpbuf) - 8;
1705 /* each *s can expand to 4 chars + "...\0",
1706 i.e. need room for 8 chars */
1708 const char *s = SvPVX_const(sv);
1709 const char * const end = s + SvCUR(sv);
1710 for ( ; s < end && d < limit; s++ ) {
1712 if (ch & 128 && !isPRINT_LC(ch)) {
1721 else if (ch == '\r') {
1725 else if (ch == '\f') {
1729 else if (ch == '\\') {
1733 else if (ch == '\0') {
1737 else if (isPRINT_LC(ch))
1754 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1755 "Argument \"%s\" isn't numeric in %s", pv,
1758 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1759 "Argument \"%s\" isn't numeric", pv);
1763 =for apidoc looks_like_number
1765 Test if the content of an SV looks like a number (or is a number).
1766 C<Inf> and C<Infinity> are treated as numbers (so will not issue a
1767 non-numeric warning), even if your atof() doesn't grok them.
1773 Perl_looks_like_number(pTHX_ SV *const sv)
1775 register const char *sbegin;
1778 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1781 sbegin = SvPVX_const(sv);
1784 else if (SvPOKp(sv))
1785 sbegin = SvPV_const(sv, len);
1787 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
1788 return grok_number(sbegin, len, NULL);
1792 S_glob_2number(pTHX_ GV * const gv)
1794 const U32 wasfake = SvFLAGS(gv) & SVf_FAKE;
1795 SV *const buffer = sv_newmortal();
1797 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1799 /* FAKE globs can get coerced, so need to turn this off temporarily if it
1802 gv_efullname3(buffer, gv, "*");
1803 SvFLAGS(gv) |= wasfake;
1805 /* We know that all GVs stringify to something that is not-a-number,
1806 so no need to test that. */
1807 if (ckWARN(WARN_NUMERIC))
1808 not_a_number(buffer);
1809 /* We just want something true to return, so that S_sv_2iuv_common
1810 can tail call us and return true. */
1815 S_glob_2pv(pTHX_ GV * const gv, STRLEN * const len)
1817 const U32 wasfake = SvFLAGS(gv) & SVf_FAKE;
1818 SV *const buffer = sv_newmortal();
1820 PERL_ARGS_ASSERT_GLOB_2PV;
1822 /* FAKE globs can get coerced, so need to turn this off temporarily if it
1825 gv_efullname3(buffer, gv, "*");
1826 SvFLAGS(gv) |= wasfake;
1828 assert(SvPOK(buffer));
1830 *len = SvCUR(buffer);
1832 return SvPVX(buffer);
1835 /* Actually, ISO C leaves conversion of UV to IV undefined, but
1836 until proven guilty, assume that things are not that bad... */
1841 As 64 bit platforms often have an NV that doesn't preserve all bits of
1842 an IV (an assumption perl has been based on to date) it becomes necessary
1843 to remove the assumption that the NV always carries enough precision to
1844 recreate the IV whenever needed, and that the NV is the canonical form.
1845 Instead, IV/UV and NV need to be given equal rights. So as to not lose
1846 precision as a side effect of conversion (which would lead to insanity
1847 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
1848 1) to distinguish between IV/UV/NV slots that have cached a valid
1849 conversion where precision was lost and IV/UV/NV slots that have a
1850 valid conversion which has lost no precision
1851 2) to ensure that if a numeric conversion to one form is requested that
1852 would lose precision, the precise conversion (or differently
1853 imprecise conversion) is also performed and cached, to prevent
1854 requests for different numeric formats on the same SV causing
1855 lossy conversion chains. (lossless conversion chains are perfectly
1860 SvIOKp is true if the IV slot contains a valid value
1861 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1862 SvNOKp is true if the NV slot contains a valid value
1863 SvNOK is true only if the NV value is accurate
1866 while converting from PV to NV, check to see if converting that NV to an
1867 IV(or UV) would lose accuracy over a direct conversion from PV to
1868 IV(or UV). If it would, cache both conversions, return NV, but mark
1869 SV as IOK NOKp (ie not NOK).
1871 While converting from PV to IV, check to see if converting that IV to an
1872 NV would lose accuracy over a direct conversion from PV to NV. If it
1873 would, cache both conversions, flag similarly.
1875 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
1876 correctly because if IV & NV were set NV *always* overruled.
1877 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
1878 changes - now IV and NV together means that the two are interchangeable:
1879 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
1881 The benefit of this is that operations such as pp_add know that if
1882 SvIOK is true for both left and right operands, then integer addition
1883 can be used instead of floating point (for cases where the result won't
1884 overflow). Before, floating point was always used, which could lead to
1885 loss of precision compared with integer addition.
1887 * making IV and NV equal status should make maths accurate on 64 bit
1889 * may speed up maths somewhat if pp_add and friends start to use
1890 integers when possible instead of fp. (Hopefully the overhead in
1891 looking for SvIOK and checking for overflow will not outweigh the
1892 fp to integer speedup)
1893 * will slow down integer operations (callers of SvIV) on "inaccurate"
1894 values, as the change from SvIOK to SvIOKp will cause a call into
1895 sv_2iv each time rather than a macro access direct to the IV slot
1896 * should speed up number->string conversion on integers as IV is
1897 favoured when IV and NV are equally accurate
1899 ####################################################################
1900 You had better be using SvIOK_notUV if you want an IV for arithmetic:
1901 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
1902 On the other hand, SvUOK is true iff UV.
1903 ####################################################################
1905 Your mileage will vary depending your CPU's relative fp to integer
1909 #ifndef NV_PRESERVES_UV
1910 # define IS_NUMBER_UNDERFLOW_IV 1
1911 # define IS_NUMBER_UNDERFLOW_UV 2
1912 # define IS_NUMBER_IV_AND_UV 2
1913 # define IS_NUMBER_OVERFLOW_IV 4
1914 # define IS_NUMBER_OVERFLOW_UV 5
1916 /* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
1918 /* For sv_2nv these three cases are "SvNOK and don't bother casting" */
1920 S_sv_2iuv_non_preserve(pTHX_ register SV *const sv
1928 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
1930 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));
1931 if (SvNVX(sv) < (NV)IV_MIN) {
1932 (void)SvIOKp_on(sv);
1934 SvIV_set(sv, IV_MIN);
1935 return IS_NUMBER_UNDERFLOW_IV;
1937 if (SvNVX(sv) > (NV)UV_MAX) {
1938 (void)SvIOKp_on(sv);
1941 SvUV_set(sv, UV_MAX);
1942 return IS_NUMBER_OVERFLOW_UV;
1944 (void)SvIOKp_on(sv);
1946 /* Can't use strtol etc to convert this string. (See truth table in
1948 if (SvNVX(sv) <= (UV)IV_MAX) {
1949 SvIV_set(sv, I_V(SvNVX(sv)));
1950 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
1951 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
1953 /* Integer is imprecise. NOK, IOKp */
1955 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
1958 SvUV_set(sv, U_V(SvNVX(sv)));
1959 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
1960 if (SvUVX(sv) == UV_MAX) {
1961 /* As we know that NVs don't preserve UVs, UV_MAX cannot
1962 possibly be preserved by NV. Hence, it must be overflow.
1964 return IS_NUMBER_OVERFLOW_UV;
1966 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
1968 /* Integer is imprecise. NOK, IOKp */
1970 return IS_NUMBER_OVERFLOW_IV;
1972 #endif /* !NV_PRESERVES_UV*/
1975 S_sv_2iuv_common(pTHX_ SV *const sv)
1979 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
1982 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
1983 * without also getting a cached IV/UV from it at the same time
1984 * (ie PV->NV conversion should detect loss of accuracy and cache
1985 * IV or UV at same time to avoid this. */
1986 /* IV-over-UV optimisation - choose to cache IV if possible */
1988 if (SvTYPE(sv) == SVt_NV)
1989 sv_upgrade(sv, SVt_PVNV);
1991 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
1992 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
1993 certainly cast into the IV range at IV_MAX, whereas the correct
1994 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
1996 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
1997 if (Perl_isnan(SvNVX(sv))) {
2003 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2004 SvIV_set(sv, I_V(SvNVX(sv)));
2005 if (SvNVX(sv) == (NV) SvIVX(sv)
2006 #ifndef NV_PRESERVES_UV
2007 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2008 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2009 /* Don't flag it as "accurately an integer" if the number
2010 came from a (by definition imprecise) NV operation, and
2011 we're outside the range of NV integer precision */
2015 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2017 /* scalar has trailing garbage, eg "42a" */
2019 DEBUG_c(PerlIO_printf(Perl_debug_log,
2020 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (precise)\n",
2026 /* IV not precise. No need to convert from PV, as NV
2027 conversion would already have cached IV if it detected
2028 that PV->IV would be better than PV->NV->IV
2029 flags already correct - don't set public IOK. */
2030 DEBUG_c(PerlIO_printf(Perl_debug_log,
2031 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (imprecise)\n",
2036 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2037 but the cast (NV)IV_MIN rounds to a the value less (more
2038 negative) than IV_MIN which happens to be equal to SvNVX ??
2039 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2040 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2041 (NV)UVX == NVX are both true, but the values differ. :-(
2042 Hopefully for 2s complement IV_MIN is something like
2043 0x8000000000000000 which will be exact. NWC */
2046 SvUV_set(sv, U_V(SvNVX(sv)));
2048 (SvNVX(sv) == (NV) SvUVX(sv))
2049 #ifndef NV_PRESERVES_UV
2050 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2051 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2052 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2053 /* Don't flag it as "accurately an integer" if the number
2054 came from a (by definition imprecise) NV operation, and
2055 we're outside the range of NV integer precision */
2061 DEBUG_c(PerlIO_printf(Perl_debug_log,
2062 "0x%"UVxf" 2iv(%"UVuf" => %"IVdf") (as unsigned)\n",
2068 else if (SvPOKp(sv) && SvLEN(sv)) {
2070 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2071 /* We want to avoid a possible problem when we cache an IV/ a UV which
2072 may be later translated to an NV, and the resulting NV is not
2073 the same as the direct translation of the initial string
2074 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2075 be careful to ensure that the value with the .456 is around if the
2076 NV value is requested in the future).
2078 This means that if we cache such an IV/a UV, we need to cache the
2079 NV as well. Moreover, we trade speed for space, and do not
2080 cache the NV if we are sure it's not needed.
2083 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2084 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2085 == IS_NUMBER_IN_UV) {
2086 /* It's definitely an integer, only upgrade to PVIV */
2087 if (SvTYPE(sv) < SVt_PVIV)
2088 sv_upgrade(sv, SVt_PVIV);
2090 } else if (SvTYPE(sv) < SVt_PVNV)
2091 sv_upgrade(sv, SVt_PVNV);
2093 /* If NVs preserve UVs then we only use the UV value if we know that
2094 we aren't going to call atof() below. If NVs don't preserve UVs
2095 then the value returned may have more precision than atof() will
2096 return, even though value isn't perfectly accurate. */
2097 if ((numtype & (IS_NUMBER_IN_UV
2098 #ifdef NV_PRESERVES_UV
2101 )) == IS_NUMBER_IN_UV) {
2102 /* This won't turn off the public IOK flag if it was set above */
2103 (void)SvIOKp_on(sv);
2105 if (!(numtype & IS_NUMBER_NEG)) {
2107 if (value <= (UV)IV_MAX) {
2108 SvIV_set(sv, (IV)value);
2110 /* it didn't overflow, and it was positive. */
2111 SvUV_set(sv, value);
2115 /* 2s complement assumption */
2116 if (value <= (UV)IV_MIN) {
2117 SvIV_set(sv, -(IV)value);
2119 /* Too negative for an IV. This is a double upgrade, but
2120 I'm assuming it will be rare. */
2121 if (SvTYPE(sv) < SVt_PVNV)
2122 sv_upgrade(sv, SVt_PVNV);
2126 SvNV_set(sv, -(NV)value);
2127 SvIV_set(sv, IV_MIN);
2131 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2132 will be in the previous block to set the IV slot, and the next
2133 block to set the NV slot. So no else here. */
2135 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2136 != IS_NUMBER_IN_UV) {
2137 /* It wasn't an (integer that doesn't overflow the UV). */
2138 SvNV_set(sv, Atof(SvPVX_const(sv)));
2140 if (! numtype && ckWARN(WARN_NUMERIC))
2143 #if defined(USE_LONG_DOUBLE)
2144 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%" PERL_PRIgldbl ")\n",
2145 PTR2UV(sv), SvNVX(sv)));
2147 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"NVgf")\n",
2148 PTR2UV(sv), SvNVX(sv)));
2151 #ifdef NV_PRESERVES_UV
2152 (void)SvIOKp_on(sv);
2154 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2155 SvIV_set(sv, I_V(SvNVX(sv)));
2156 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2159 NOOP; /* Integer is imprecise. NOK, IOKp */
2161 /* UV will not work better than IV */
2163 if (SvNVX(sv) > (NV)UV_MAX) {
2165 /* Integer is inaccurate. NOK, IOKp, is UV */
2166 SvUV_set(sv, UV_MAX);
2168 SvUV_set(sv, U_V(SvNVX(sv)));
2169 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2170 NV preservse UV so can do correct comparison. */
2171 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2174 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
2179 #else /* NV_PRESERVES_UV */
2180 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2181 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
2182 /* The IV/UV slot will have been set from value returned by
2183 grok_number above. The NV slot has just been set using
2186 assert (SvIOKp(sv));
2188 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2189 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2190 /* Small enough to preserve all bits. */
2191 (void)SvIOKp_on(sv);
2193 SvIV_set(sv, I_V(SvNVX(sv)));
2194 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2196 /* Assumption: first non-preserved integer is < IV_MAX,
2197 this NV is in the preserved range, therefore: */
2198 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2200 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);
2204 0 0 already failed to read UV.
2205 0 1 already failed to read UV.
2206 1 0 you won't get here in this case. IV/UV
2207 slot set, public IOK, Atof() unneeded.
2208 1 1 already read UV.
2209 so there's no point in sv_2iuv_non_preserve() attempting
2210 to use atol, strtol, strtoul etc. */
2212 sv_2iuv_non_preserve (sv, numtype);
2214 sv_2iuv_non_preserve (sv);
2218 #endif /* NV_PRESERVES_UV */
2219 /* It might be more code efficient to go through the entire logic above
2220 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2221 gets complex and potentially buggy, so more programmer efficient
2222 to do it this way, by turning off the public flags: */
2224 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2228 if (isGV_with_GP(sv))
2229 return glob_2number((GV *)sv);
2231 if (!(SvFLAGS(sv) & SVs_PADTMP)) {
2232 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2235 if (SvTYPE(sv) < SVt_IV)
2236 /* Typically the caller expects that sv_any is not NULL now. */
2237 sv_upgrade(sv, SVt_IV);
2238 /* Return 0 from the caller. */
2245 =for apidoc sv_2iv_flags
2247 Return the integer value of an SV, doing any necessary string
2248 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2249 Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2255 Perl_sv_2iv_flags(pTHX_ register SV *const sv, const I32 flags)
2260 if (SvGMAGICAL(sv) || (SvTYPE(sv) == SVt_PVGV && SvVALID(sv))) {
2261 /* FBMs use the same flag bit as SVf_IVisUV, so must let them
2262 cache IVs just in case. In practice it seems that they never
2263 actually anywhere accessible by user Perl code, let alone get used
2264 in anything other than a string context. */
2265 if (flags & SV_GMAGIC)
2270 return I_V(SvNVX(sv));
2272 if (SvPOKp(sv) && SvLEN(sv)) {
2275 = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2277 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2278 == IS_NUMBER_IN_UV) {
2279 /* It's definitely an integer */
2280 if (numtype & IS_NUMBER_NEG) {
2281 if (value < (UV)IV_MIN)
2284 if (value < (UV)IV_MAX)
2289 if (ckWARN(WARN_NUMERIC))
2292 return I_V(Atof(SvPVX_const(sv)));
2297 assert(SvTYPE(sv) >= SVt_PVMG);
2298 /* This falls through to the report_uninit inside S_sv_2iuv_common. */
2299 } else if (SvTHINKFIRST(sv)) {
2303 SV * const tmpstr=AMG_CALLun(sv,numer);
2304 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2305 return SvIV(tmpstr);
2308 return PTR2IV(SvRV(sv));
2311 sv_force_normal_flags(sv, 0);
2313 if (SvREADONLY(sv) && !SvOK(sv)) {
2314 if (ckWARN(WARN_UNINITIALIZED))
2320 if (S_sv_2iuv_common(aTHX_ sv))
2323 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"IVdf")\n",
2324 PTR2UV(sv),SvIVX(sv)));
2325 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
2329 =for apidoc sv_2uv_flags
2331 Return the unsigned integer value of an SV, doing any necessary string
2332 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2333 Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
2339 Perl_sv_2uv_flags(pTHX_ register SV *const sv, const I32 flags)
2344 if (SvGMAGICAL(sv) || (SvTYPE(sv) == SVt_PVGV && SvVALID(sv))) {
2345 /* FBMs use the same flag bit as SVf_IVisUV, so must let them
2346 cache IVs just in case. */
2347 if (flags & SV_GMAGIC)
2352 return U_V(SvNVX(sv));
2353 if (SvPOKp(sv) && SvLEN(sv)) {
2356 = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2358 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2359 == IS_NUMBER_IN_UV) {
2360 /* It's definitely an integer */
2361 if (!(numtype & IS_NUMBER_NEG))
2365 if (ckWARN(WARN_NUMERIC))
2368 return U_V(Atof(SvPVX_const(sv)));
2373 assert(SvTYPE(sv) >= SVt_PVMG);
2374 /* This falls through to the report_uninit inside S_sv_2iuv_common. */
2375 } else if (SvTHINKFIRST(sv)) {
2379 SV *const tmpstr = AMG_CALLun(sv,numer);
2380 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2381 return SvUV(tmpstr);
2384 return PTR2UV(SvRV(sv));
2387 sv_force_normal_flags(sv, 0);
2389 if (SvREADONLY(sv) && !SvOK(sv)) {
2390 if (ckWARN(WARN_UNINITIALIZED))
2396 if (S_sv_2iuv_common(aTHX_ sv))
2400 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2uv(%"UVuf")\n",
2401 PTR2UV(sv),SvUVX(sv)));
2402 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
2408 Return the num value of an SV, doing any necessary string or integer
2409 conversion, magic etc. Normally used via the C<SvNV(sv)> and C<SvNVx(sv)>
2416 Perl_sv_2nv(pTHX_ register SV *const sv)
2421 if (SvGMAGICAL(sv) || (SvTYPE(sv) == SVt_PVGV && SvVALID(sv))) {
2422 /* FBMs use the same flag bit as SVf_IVisUV, so must let them
2423 cache IVs just in case. */
2427 if ((SvPOKp(sv) && SvLEN(sv)) && !SvIOKp(sv)) {
2428 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
2429 !grok_number(SvPVX_const(sv), SvCUR(sv), NULL))
2431 return Atof(SvPVX_const(sv));
2435 return (NV)SvUVX(sv);
2437 return (NV)SvIVX(sv);
2442 assert(SvTYPE(sv) >= SVt_PVMG);
2443 /* This falls through to the report_uninit near the end of the
2445 } else if (SvTHINKFIRST(sv)) {
2449 SV *const tmpstr = AMG_CALLun(sv,numer);
2450 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2451 return SvNV(tmpstr);
2454 return PTR2NV(SvRV(sv));
2457 sv_force_normal_flags(sv, 0);
2459 if (SvREADONLY(sv) && !SvOK(sv)) {
2460 if (ckWARN(WARN_UNINITIALIZED))
2465 if (SvTYPE(sv) < SVt_NV) {
2466 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2467 sv_upgrade(sv, SVt_NV);
2468 #ifdef USE_LONG_DOUBLE
2470 STORE_NUMERIC_LOCAL_SET_STANDARD();
2471 PerlIO_printf(Perl_debug_log,
2472 "0x%"UVxf" num(%" PERL_PRIgldbl ")\n",
2473 PTR2UV(sv), SvNVX(sv));
2474 RESTORE_NUMERIC_LOCAL();
2478 STORE_NUMERIC_LOCAL_SET_STANDARD();
2479 PerlIO_printf(Perl_debug_log, "0x%"UVxf" num(%"NVgf")\n",
2480 PTR2UV(sv), SvNVX(sv));
2481 RESTORE_NUMERIC_LOCAL();
2485 else if (SvTYPE(sv) < SVt_PVNV)
2486 sv_upgrade(sv, SVt_PVNV);
2491 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
2492 #ifdef NV_PRESERVES_UV
2498 /* Only set the public NV OK flag if this NV preserves the IV */
2499 /* Check it's not 0xFFFFFFFFFFFFFFFF */
2501 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
2502 : (SvIVX(sv) == I_V(SvNVX(sv))))
2508 else if (SvPOKp(sv) && SvLEN(sv)) {
2510 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2511 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
2513 #ifdef NV_PRESERVES_UV
2514 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2515 == IS_NUMBER_IN_UV) {
2516 /* It's definitely an integer */
2517 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
2519 SvNV_set(sv, Atof(SvPVX_const(sv)));
2525 SvNV_set(sv, Atof(SvPVX_const(sv)));
2526 /* Only set the public NV OK flag if this NV preserves the value in
2527 the PV at least as well as an IV/UV would.
2528 Not sure how to do this 100% reliably. */
2529 /* if that shift count is out of range then Configure's test is
2530 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2532 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2533 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2534 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
2535 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2536 /* Can't use strtol etc to convert this string, so don't try.
2537 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2540 /* value has been set. It may not be precise. */
2541 if ((numtype & IS_NUMBER_NEG) && (value > (UV)IV_MIN)) {
2542 /* 2s complement assumption for (UV)IV_MIN */
2543 SvNOK_on(sv); /* Integer is too negative. */
2548 if (numtype & IS_NUMBER_NEG) {
2549 SvIV_set(sv, -(IV)value);
2550 } else if (value <= (UV)IV_MAX) {
2551 SvIV_set(sv, (IV)value);
2553 SvUV_set(sv, value);
2557 if (numtype & IS_NUMBER_NOT_INT) {
2558 /* I believe that even if the original PV had decimals,
2559 they are lost beyond the limit of the FP precision.
2560 However, neither is canonical, so both only get p
2561 flags. NWC, 2000/11/25 */
2562 /* Both already have p flags, so do nothing */
2564 const NV nv = SvNVX(sv);
2565 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2566 if (SvIVX(sv) == I_V(nv)) {
2569 /* It had no "." so it must be integer. */
2573 /* between IV_MAX and NV(UV_MAX).
2574 Could be slightly > UV_MAX */
2576 if (numtype & IS_NUMBER_NOT_INT) {
2577 /* UV and NV both imprecise. */
2579 const UV nv_as_uv = U_V(nv);
2581 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2590 /* It might be more code efficient to go through the entire logic above
2591 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2592 gets complex and potentially buggy, so more programmer efficient
2593 to do it this way, by turning off the public flags: */
2595 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2596 #endif /* NV_PRESERVES_UV */
2599 if (isGV_with_GP(sv)) {
2600 glob_2number((GV *)sv);
2604 if (!PL_localizing && !(SvFLAGS(sv) & SVs_PADTMP) && ckWARN(WARN_UNINITIALIZED))
2606 assert (SvTYPE(sv) >= SVt_NV);
2607 /* Typically the caller expects that sv_any is not NULL now. */
2608 /* XXX Ilya implies that this is a bug in callers that assume this
2609 and ideally should be fixed. */
2612 #if defined(USE_LONG_DOUBLE)
2614 STORE_NUMERIC_LOCAL_SET_STANDARD();
2615 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2nv(%" PERL_PRIgldbl ")\n",
2616 PTR2UV(sv), SvNVX(sv));
2617 RESTORE_NUMERIC_LOCAL();
2621 STORE_NUMERIC_LOCAL_SET_STANDARD();
2622 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 1nv(%"NVgf")\n",
2623 PTR2UV(sv), SvNVX(sv));
2624 RESTORE_NUMERIC_LOCAL();
2633 Return an SV with the numeric value of the source SV, doing any necessary
2634 reference or overload conversion. You must use the C<SvNUM(sv)> macro to
2635 access this function.
2641 Perl_sv_2num(pTHX_ register SV *const sv)
2643 PERL_ARGS_ASSERT_SV_2NUM;
2648 SV * const tmpsv = AMG_CALLun(sv,numer);
2649 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2650 return sv_2num(tmpsv);
2652 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2655 /* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2656 * UV as a string towards the end of buf, and return pointers to start and
2659 * We assume that buf is at least TYPE_CHARS(UV) long.
2663 S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
2665 char *ptr = buf + TYPE_CHARS(UV);
2666 char * const ebuf = ptr;
2669 PERL_ARGS_ASSERT_UIV_2BUF;
2681 *--ptr = '0' + (char)(uv % 10);
2690 =for apidoc sv_2pv_flags
2692 Returns a pointer to the string value of an SV, and sets *lp to its length.
2693 If flags includes SV_GMAGIC, does an mg_get() first. Coerces sv to a string
2695 Normally invoked via the C<SvPV_flags> macro. C<sv_2pv()> and C<sv_2pv_nomg>
2696 usually end up here too.
2702 Perl_sv_2pv_flags(pTHX_ register SV *const sv, STRLEN *const lp, const I32 flags)
2712 if (SvGMAGICAL(sv)) {
2713 if (flags & SV_GMAGIC)
2718 if (flags & SV_MUTABLE_RETURN)
2719 return SvPVX_mutable(sv);
2720 if (flags & SV_CONST_RETURN)
2721 return (char *)SvPVX_const(sv);
2724 if (SvIOKp(sv) || SvNOKp(sv)) {
2725 char tbuf[64]; /* Must fit sprintf/Gconvert of longest IV/NV */
2730 ? my_snprintf(tbuf, sizeof(tbuf), "%"UVuf, (UV)SvUVX(sv))
2731 : my_snprintf(tbuf, sizeof(tbuf), "%"IVdf, (IV)SvIVX(sv));
2733 Gconvert(SvNVX(sv), NV_DIG, 0, tbuf);
2740 #ifdef FIXNEGATIVEZERO
2741 if (len == 2 && tbuf[0] == '-' && tbuf[1] == '0') {
2747 SvUPGRADE(sv, SVt_PV);
2750 s = SvGROW_mutable(sv, len + 1);
2753 return (char*)memcpy(s, tbuf, len + 1);
2759 assert(SvTYPE(sv) >= SVt_PVMG);
2760 /* This falls through to the report_uninit near the end of the
2762 } else if (SvTHINKFIRST(sv)) {
2766 SV *const tmpstr = AMG_CALLun(sv,string);
2767 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2769 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2773 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2774 if (flags & SV_CONST_RETURN) {
2775 pv = (char *) SvPVX_const(tmpstr);
2777 pv = (flags & SV_MUTABLE_RETURN)
2778 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
2781 *lp = SvCUR(tmpstr);
2783 pv = sv_2pv_flags(tmpstr, lp, flags);
2796 const SV *const referent = (SV*)SvRV(sv);
2800 retval = buffer = savepvn("NULLREF", len);
2801 } else if (SvTYPE(referent) == SVt_REGEXP) {
2802 const REGEXP * const re = (REGEXP *)referent;
2807 /* If the regex is UTF-8 we want the containing scalar to
2808 have an UTF-8 flag too */
2814 if ((seen_evals = RX_SEEN_EVALS(re)))
2815 PL_reginterp_cnt += seen_evals;
2818 *lp = RX_WRAPLEN(re);
2820 return RX_WRAPPED(re);
2822 const char *const typestr = sv_reftype(referent, 0);
2823 const STRLEN typelen = strlen(typestr);
2824 UV addr = PTR2UV(referent);
2825 const char *stashname = NULL;
2826 STRLEN stashnamelen = 0; /* hush, gcc */
2827 const char *buffer_end;
2829 if (SvOBJECT(referent)) {
2830 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
2833 stashname = HEK_KEY(name);
2834 stashnamelen = HEK_LEN(name);
2836 if (HEK_UTF8(name)) {
2842 stashname = "__ANON__";
2845 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
2846 + 2 * sizeof(UV) + 2 /* )\0 */;
2848 len = typelen + 3 /* (0x */
2849 + 2 * sizeof(UV) + 2 /* )\0 */;
2852 Newx(buffer, len, char);
2853 buffer_end = retval = buffer + len;
2855 /* Working backwards */
2859 *--retval = PL_hexdigit[addr & 15];
2860 } while (addr >>= 4);
2866 memcpy(retval, typestr, typelen);
2870 retval -= stashnamelen;
2871 memcpy(retval, stashname, stashnamelen);
2873 /* retval may not neccesarily have reached the start of the
2875 assert (retval >= buffer);
2877 len = buffer_end - retval - 1; /* -1 for that \0 */
2885 if (SvREADONLY(sv) && !SvOK(sv)) {
2888 if (flags & SV_UNDEF_RETURNS_NULL)
2890 if (ckWARN(WARN_UNINITIALIZED))
2895 if (SvIOK(sv) || ((SvIOKp(sv) && !SvNOKp(sv)))) {
2896 /* I'm assuming that if both IV and NV are equally valid then
2897 converting the IV is going to be more efficient */
2898 const U32 isUIOK = SvIsUV(sv);
2899 char buf[TYPE_CHARS(UV)];
2903 if (SvTYPE(sv) < SVt_PVIV)
2904 sv_upgrade(sv, SVt_PVIV);
2905 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
2907 /* inlined from sv_setpvn */
2908 s = SvGROW_mutable(sv, len + 1);
2909 Move(ptr, s, len, char);
2913 else if (SvNOKp(sv)) {
2914 const int olderrno = errno;
2915 if (SvTYPE(sv) < SVt_PVNV)
2916 sv_upgrade(sv, SVt_PVNV);
2917 /* The +20 is pure guesswork. Configure test needed. --jhi */
2918 s = SvGROW_mutable(sv, NV_DIG + 20);
2919 /* some Xenix systems wipe out errno here */
2921 if (SvNVX(sv) == 0.0)
2922 my_strlcpy(s, "0", SvLEN(sv));
2926 Gconvert(SvNVX(sv), NV_DIG, 0, s);
2929 #ifdef FIXNEGATIVEZERO
2930 if (*s == '-' && s[1] == '0' && !s[2]) {
2942 if (isGV_with_GP(sv))
2943 return glob_2pv((GV *)sv, lp);
2947 if (flags & SV_UNDEF_RETURNS_NULL)
2949 if (!PL_localizing && !(SvFLAGS(sv) & SVs_PADTMP) && ckWARN(WARN_UNINITIALIZED))
2951 if (SvTYPE(sv) < SVt_PV)
2952 /* Typically the caller expects that sv_any is not NULL now. */
2953 sv_upgrade(sv, SVt_PV);
2957 const STRLEN len = s - SvPVX_const(sv);
2963 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
2964 PTR2UV(sv),SvPVX_const(sv)));
2965 if (flags & SV_CONST_RETURN)
2966 return (char *)SvPVX_const(sv);
2967 if (flags & SV_MUTABLE_RETURN)
2968 return SvPVX_mutable(sv);
2973 =for apidoc sv_copypv
2975 Copies a stringified representation of the source SV into the
2976 destination SV. Automatically performs any necessary mg_get and
2977 coercion of numeric values into strings. Guaranteed to preserve
2978 UTF8 flag even from overloaded objects. Similar in nature to
2979 sv_2pv[_flags] but operates directly on an SV instead of just the
2980 string. Mostly uses sv_2pv_flags to do its work, except when that
2981 would lose the UTF-8'ness of the PV.
2987 Perl_sv_copypv(pTHX_ SV *const dsv, register SV *const ssv)
2990 const char * const s = SvPV_const(ssv,len);
2992 PERL_ARGS_ASSERT_SV_COPYPV;
2994 sv_setpvn(dsv,s,len);
3002 =for apidoc sv_2pvbyte
3004 Return a pointer to the byte-encoded representation of the SV, and set *lp
3005 to its length. May cause the SV to be downgraded from UTF-8 as a
3008 Usually accessed via the C<SvPVbyte> macro.
3014 Perl_sv_2pvbyte(pTHX_ register SV *const sv, STRLEN *const lp)
3016 PERL_ARGS_ASSERT_SV_2PVBYTE;
3018 sv_utf8_downgrade(sv,0);
3019 return lp ? SvPV(sv,*lp) : SvPV_nolen(sv);
3023 =for apidoc sv_2pvutf8
3025 Return a pointer to the UTF-8-encoded representation of the SV, and set *lp
3026 to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3028 Usually accessed via the C<SvPVutf8> macro.
3034 Perl_sv_2pvutf8(pTHX_ register SV *const sv, STRLEN *const lp)
3036 PERL_ARGS_ASSERT_SV_2PVUTF8;
3038 sv_utf8_upgrade(sv);
3039 return lp ? SvPV(sv,*lp) : SvPV_nolen(sv);
3044 =for apidoc sv_2bool
3046 This function is only called on magical items, and is only used by
3047 sv_true() or its macro equivalent.
3053 Perl_sv_2bool(pTHX_ register SV *const sv)
3057 PERL_ARGS_ASSERT_SV_2BOOL;
3065 SV * const tmpsv = AMG_CALLun(sv,bool_);
3066 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
3067 return (bool)SvTRUE(tmpsv);
3069 return SvRV(sv) != 0;
3072 register XPV* const Xpvtmp = (XPV*)SvANY(sv);
3074 (*sv->sv_u.svu_pv > '0' ||
3075 Xpvtmp->xpv_cur > 1 ||
3076 (Xpvtmp->xpv_cur && *sv->sv_u.svu_pv != '0')))
3083 return SvIVX(sv) != 0;
3086 return SvNVX(sv) != 0.0;
3088 if (isGV_with_GP(sv))
3098 =for apidoc sv_utf8_upgrade
3100 Converts the PV of an SV to its UTF-8-encoded form.
3101 Forces the SV to string form if it is not already.
3102 Always sets the SvUTF8 flag to avoid future validity checks even
3103 if all the bytes have hibit clear.
3105 This is not as a general purpose byte encoding to Unicode interface:
3106 use the Encode extension for that.
3108 =for apidoc sv_utf8_upgrade_flags
3110 Converts the PV of an SV to its UTF-8-encoded form.
3111 Forces the SV to string form if it is not already.
3112 Always sets the SvUTF8 flag to avoid future validity checks even
3113 if all the bytes have hibit clear. If C<flags> has C<SV_GMAGIC> bit set,
3114 will C<mg_get> on C<sv> if appropriate, else not. C<sv_utf8_upgrade> and
3115 C<sv_utf8_upgrade_nomg> are implemented in terms of this function.
3117 This is not as a general purpose byte encoding to Unicode interface:
3118 use the Encode extension for that.
3124 Perl_sv_utf8_upgrade_flags(pTHX_ register SV *const sv, const I32 flags)
3128 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS;
3130 if (sv == &PL_sv_undef)
3134 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3135 (void) sv_2pv_flags(sv,&len, flags);
3139 (void) SvPV_force(sv,len);
3148 sv_force_normal_flags(sv, 0);
3151 if (PL_encoding && !(flags & SV_UTF8_NO_ENCODING))
3152 sv_recode_to_utf8(sv, PL_encoding);
3153 else { /* Assume Latin-1/EBCDIC */
3154 /* This function could be much more efficient if we
3155 * had a FLAG in SVs to signal if there are any hibit
3156 * chars in the PV. Given that there isn't such a flag
3157 * make the loop as fast as possible. */
3158 const U8 * const s = (U8 *) SvPVX_const(sv);
3159 const U8 * const e = (U8 *) SvEND(sv);
3164 /* Check for hi bit */
3165 if (!NATIVE_IS_INVARIANT(ch)) {
3166 STRLEN len = SvCUR(sv) + 1; /* Plus the \0 */
3167 U8 * const recoded = bytes_to_utf8((U8*)s, &len);
3169 SvPV_free(sv); /* No longer using what was there before. */
3170 SvPV_set(sv, (char*)recoded);
3171 SvCUR_set(sv, len - 1);
3172 SvLEN_set(sv, len); /* No longer know the real size. */
3176 /* Mark as UTF-8 even if no hibit - saves scanning loop */
3183 =for apidoc sv_utf8_downgrade
3185 Attempts to convert the PV of an SV from characters to bytes.
3186 If the PV contains a character beyond byte, this conversion will fail;
3187 in this case, either returns false or, if C<fail_ok> is not
3190 This is not as a general purpose Unicode to byte encoding interface:
3191 use the Encode extension for that.
3197 Perl_sv_utf8_downgrade(pTHX_ register SV *const sv, const bool fail_ok)
3201 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3203 if (SvPOKp(sv) && SvUTF8(sv)) {
3209 sv_force_normal_flags(sv, 0);
3211 s = (U8 *) SvPV(sv, len);
3212 if (!utf8_to_bytes(s, &len)) {
3217 Perl_croak(aTHX_ "Wide character in %s",
3220 Perl_croak(aTHX_ "Wide character");
3231 =for apidoc sv_utf8_encode
3233 Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3234 flag off so that it looks like octets again.
3240 Perl_sv_utf8_encode(pTHX_ register SV *const sv)
3242 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3245 sv_force_normal_flags(sv, 0);
3247 if (SvREADONLY(sv)) {
3248 Perl_croak(aTHX_ PL_no_modify);
3250 (void) sv_utf8_upgrade(sv);
3255 =for apidoc sv_utf8_decode
3257 If the PV of the SV is an octet sequence in UTF-8
3258 and contains a multiple-byte character, the C<SvUTF8> flag is turned on
3259 so that it looks like a character. If the PV contains only single-byte
3260 characters, the C<SvUTF8> flag stays being off.
3261 Scans PV for validity and returns false if the PV is invalid UTF-8.
3267 Perl_sv_utf8_decode(pTHX_ register SV *const sv)
3269 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3275 /* The octets may have got themselves encoded - get them back as
3278 if (!sv_utf8_downgrade(sv, TRUE))
3281 /* it is actually just a matter of turning the utf8 flag on, but
3282 * we want to make sure everything inside is valid utf8 first.
3284 c = (const U8 *) SvPVX_const(sv);
3285 if (!is_utf8_string(c, SvCUR(sv)+1))
3287 e = (const U8 *) SvEND(sv);
3290 if (!UTF8_IS_INVARIANT(ch)) {
3300 =for apidoc sv_setsv
3302 Copies the contents of the source SV C<ssv> into the destination SV
3303 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3304 function if the source SV needs to be reused. Does not handle 'set' magic.
3305 Loosely speaking, it performs a copy-by-value, obliterating any previous
3306 content of the destination.
3308 You probably want to use one of the assortment of wrappers, such as
3309 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3310 C<SvSetMagicSV_nosteal>.
3312 =for apidoc sv_setsv_flags
3314 Copies the contents of the source SV C<ssv> into the destination SV
3315 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3316 function if the source SV needs to be reused. Does not handle 'set' magic.
3317 Loosely speaking, it performs a copy-by-value, obliterating any previous
3318 content of the destination.
3319 If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
3320 C<ssv> if appropriate, else not. If the C<flags> parameter has the
3321 C<NOSTEAL> bit set then the buffers of temps will not be stolen. <sv_setsv>
3322 and C<sv_setsv_nomg> are implemented in terms of this function.
3324 You probably want to use one of the assortment of wrappers, such as
3325 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3326 C<SvSetMagicSV_nosteal>.
3328 This is the primary function for copying scalars, and most other
3329 copy-ish functions and macros use this underneath.
3335 S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
3337 I32 mro_changes = 0; /* 1 = method, 2 = isa */
3339 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB;
3341 if (dtype != SVt_PVGV) {
3342 const char * const name = GvNAME(sstr);
3343 const STRLEN len = GvNAMELEN(sstr);
3345 if (dtype >= SVt_PV) {
3351 SvUPGRADE(dstr, SVt_PVGV);
3352 (void)SvOK_off(dstr);
3353 /* FIXME - why are we doing this, then turning it off and on again
3355 isGV_with_GP_on(dstr);
3357 GvSTASH(dstr) = GvSTASH(sstr);
3359 Perl_sv_add_backref(aTHX_ (SV*)GvSTASH(dstr), dstr);
3360 gv_name_set((GV *)dstr, name, len, GV_ADD);
3361 SvFAKE_on(dstr); /* can coerce to non-glob */
3364 #ifdef GV_UNIQUE_CHECK
3365 if (GvUNIQUE((GV*)dstr)) {
3366 Perl_croak(aTHX_ PL_no_modify);
3370 if(GvGP((GV*)sstr)) {
3371 /* If source has method cache entry, clear it */
3373 SvREFCNT_dec(GvCV(sstr));
3377 /* If source has a real method, then a method is
3379 else if(GvCV((GV*)sstr)) {
3384 /* If dest already had a real method, that's a change as well */
3385 if(!mro_changes && GvGP((GV*)dstr) && GvCVu((GV*)dstr)) {
3389 if(strEQ(GvNAME((GV*)dstr),"ISA"))
3393 isGV_with_GP_off(dstr);
3394 (void)SvOK_off(dstr);
3395 isGV_with_GP_on(dstr);
3396 GvINTRO_off(dstr); /* one-shot flag */
3397 GvGP(dstr) = gp_ref(GvGP(sstr));
3398 if (SvTAINTED(sstr))
3400 if (GvIMPORTED(dstr) != GVf_IMPORTED
3401 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
3403 GvIMPORTED_on(dstr);
3406 if(mro_changes == 2) mro_isa_changed_in(GvSTASH(dstr));
3407 else if(mro_changes) mro_method_changed_in(GvSTASH(dstr));
3412 S_glob_assign_ref(pTHX_ SV *const dstr, SV *const sstr)
3414 SV * const sref = SvREFCNT_inc(SvRV(sstr));
3416 const int intro = GvINTRO(dstr);
3419 const U32 stype = SvTYPE(sref);
3421 PERL_ARGS_ASSERT_GLOB_ASSIGN_REF;
3423 #ifdef GV_UNIQUE_CHECK
3424 if (GvUNIQUE((GV*)dstr)) {
3425 Perl_croak(aTHX_ PL_no_modify);
3430 GvINTRO_off(dstr); /* one-shot flag */
3431 GvLINE(dstr) = CopLINE(PL_curcop);
3432 GvEGV(dstr) = (GV*)dstr;
3437 location = (SV **) &GvCV(dstr);
3438 import_flag = GVf_IMPORTED_CV;
3441 location = (SV **) &GvHV(dstr);
3442 import_flag = GVf_IMPORTED_HV;
3445 location = (SV **) &GvAV(dstr);
3446 import_flag = GVf_IMPORTED_AV;
3449 location = (SV **) &GvIOp(dstr);
3452 location = (SV **) &GvFORM(dstr);
3454 location = &GvSV(dstr);
3455 import_flag = GVf_IMPORTED_SV;
3458 if (stype == SVt_PVCV) {
3459 /*if (GvCVGEN(dstr) && (GvCV(dstr) != (CV*)sref || GvCVGEN(dstr))) {*/
3460 if (GvCVGEN(dstr)) {
3461 SvREFCNT_dec(GvCV(dstr));
3463 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
3466 SAVEGENERICSV(*location);
3470 if (stype == SVt_PVCV && (*location != sref || GvCVGEN(dstr))) {
3471 CV* const cv = (CV*)*location;
3473 if (!GvCVGEN((GV*)dstr) &&
3474 (CvROOT(cv) || CvXSUB(cv)))
3476 /* Redefining a sub - warning is mandatory if
3477 it was a const and its value changed. */
3478 if (CvCONST(cv) && CvCONST((CV*)sref)
3479 && cv_const_sv(cv) == cv_const_sv((CV*)sref)) {
3481 /* They are 2 constant subroutines generated from
3482 the same constant. This probably means that
3483 they are really the "same" proxy subroutine
3484 instantiated in 2 places. Most likely this is
3485 when a constant is exported twice. Don't warn.
3488 else if (ckWARN(WARN_REDEFINE)
3490 && (!CvCONST((CV*)sref)
3491 || sv_cmp(cv_const_sv(cv),
3492 cv_const_sv((CV*)sref))))) {
3493 Perl_warner(aTHX_ packWARN(WARN_REDEFINE),
3496 ? "Constant subroutine %s::%s redefined"
3497 : "Subroutine %s::%s redefined"),
3498 HvNAME_get(GvSTASH((GV*)dstr)),
3499 GvENAME((GV*)dstr));
3503 cv_ckproto_len(cv, (GV*)dstr,
3504 SvPOK(sref) ? SvPVX_const(sref) : NULL,
3505 SvPOK(sref) ? SvCUR(sref) : 0);
3507 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
3508 GvASSUMECV_on(dstr);
3509 if(GvSTASH(dstr)) mro_method_changed_in(GvSTASH(dstr)); /* sub foo { 1 } sub bar { 2 } *bar = \&foo */
3512 if (import_flag && !(GvFLAGS(dstr) & import_flag)
3513 && CopSTASH_ne(PL_curcop, GvSTASH(dstr))) {
3514 GvFLAGS(dstr) |= import_flag;
3519 if (SvTAINTED(sstr))
3525 Perl_sv_setsv_flags(pTHX_ SV *dstr, register SV* sstr, const I32 flags)
3528 register U32 sflags;
3530 register svtype stype;
3532 PERL_ARGS_ASSERT_SV_SETSV_FLAGS;
3537 if (SvIS_FREED(dstr)) {
3538 Perl_croak(aTHX_ "panic: attempt to copy value %" SVf
3539 " to a freed scalar %p", SVfARG(sstr), (void *)dstr);
3541 SV_CHECK_THINKFIRST_COW_DROP(dstr);
3543 sstr = &PL_sv_undef;
3544 if (SvIS_FREED(sstr)) {
3545 Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p to %p",
3546 (void*)sstr, (void*)dstr);
3548 stype = SvTYPE(sstr);
3549 dtype = SvTYPE(dstr);
3551 (void)SvAMAGIC_off(dstr);
3554 /* need to nuke the magic */
3556 SvRMAGICAL_off(dstr);
3559 /* There's a lot of redundancy below but we're going for speed here */
3564 if (dtype != SVt_PVGV) {
3565 (void)SvOK_off(dstr);
3573 sv_upgrade(dstr, SVt_IV);
3577 sv_upgrade(dstr, SVt_PVIV);
3580 goto end_of_first_switch;
3582 (void)SvIOK_only(dstr);
3583 SvIV_set(dstr, SvIVX(sstr));
3586 /* SvTAINTED can only be true if the SV has taint magic, which in
3587 turn means that the SV type is PVMG (or greater). This is the
3588 case statement for SVt_IV, so this cannot be true (whatever gcov
3590 assert(!SvTAINTED(sstr));
3595 if (dtype < SVt_PV && dtype != SVt_IV)
3596 sv_upgrade(dstr, SVt_IV);
3604 sv_upgrade(dstr, SVt_NV);
3608 sv_upgrade(dstr, SVt_PVNV);
3611 goto end_of_first_switch;
3613 SvNV_set(dstr, SvNVX(sstr));
3614 (void)SvNOK_only(dstr);
3615 /* SvTAINTED can only be true if the SV has taint magic, which in
3616 turn means that the SV type is PVMG (or greater). This is the
3617 case statement for SVt_NV, so this cannot be true (whatever gcov
3619 assert(!SvTAINTED(sstr));
3625 #ifdef PERL_OLD_COPY_ON_WRITE
3626 if ((SvFLAGS(sstr) & CAN_COW_MASK) == CAN_COW_FLAGS) {
3627 if (dtype < SVt_PVIV)
3628 sv_upgrade(dstr, SVt_PVIV);
3636 sv_upgrade(dstr, SVt_PV);
3639 if (dtype < SVt_PVIV)
3640 sv_upgrade(dstr, SVt_PVIV);
3643 if (dtype < SVt_PVNV)
3644 sv_upgrade(dstr, SVt_PVNV);
3648 const char * const type = sv_reftype(sstr,0);
3650 Perl_croak(aTHX_ "Bizarre copy of %s in %s", type, OP_NAME(PL_op));
3652 Perl_croak(aTHX_ "Bizarre copy of %s", type);
3656 /* case SVt_BIND: */
3659 if (isGV_with_GP(sstr) && dtype <= SVt_PVGV) {
3660 glob_assign_glob(dstr, sstr, dtype);
3663 /* SvVALID means that this PVGV is playing at being an FBM. */
3667 if (SvGMAGICAL(sstr) && (flags & SV_GMAGIC)) {
3669 if (SvTYPE(sstr) != stype) {
3670 stype = SvTYPE(sstr);
3671 if (isGV_with_GP(sstr) && stype == SVt_PVGV && dtype <= SVt_PVGV) {
3672 glob_assign_glob(dstr, sstr, dtype);
3677 if (stype == SVt_PVLV)
3678 SvUPGRADE(dstr, SVt_PVNV);
3680 SvUPGRADE(dstr, (svtype)stype);
3682 end_of_first_switch:
3684 /* dstr may have been upgraded. */
3685 dtype = SvTYPE(dstr);
3686 sflags = SvFLAGS(sstr);
3688 if (dtype == SVt_PVCV || dtype == SVt_PVFM) {
3689 /* Assigning to a subroutine sets the prototype. */
3692 const char *const ptr = SvPV_const(sstr, len);
3694 SvGROW(dstr, len + 1);
3695 Copy(ptr, SvPVX(dstr), len + 1, char);
3696 SvCUR_set(dstr, len);
3698 SvFLAGS(dstr) |= sflags & SVf_UTF8;
3702 } else if (dtype == SVt_PVAV || dtype == SVt_PVHV) {
3703 const char * const type = sv_reftype(dstr,0);
3705 Perl_croak(aTHX_ "Cannot copy to %s in %s", type, OP_NAME(PL_op));
3707 Perl_croak(aTHX_ "Cannot copy to %s", type);
3708 } else if (sflags & SVf_ROK) {
3709 if (isGV_with_GP(dstr) && dtype == SVt_PVGV
3710 && SvTYPE(SvRV(sstr)) == SVt_PVGV) {
3713 if (GvIMPORTED(dstr) != GVf_IMPORTED
3714 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
3716 GvIMPORTED_on(dstr);
3721 glob_assign_glob(dstr, sstr, dtype);
3725 if (dtype >= SVt_PV) {
3726 if (dtype == SVt_PVGV && isGV_with_GP(dstr)) {
3727 glob_assign_ref(dstr, sstr);
3730 if (SvPVX_const(dstr)) {
3736 (void)SvOK_off(dstr);
3737 SvRV_set(dstr, SvREFCNT_inc(SvRV(sstr)));
3738 SvFLAGS(dstr) |= sflags & SVf_ROK;
3739 assert(!(sflags & SVp_NOK));
3740 assert(!(sflags & SVp_IOK));
3741 assert(!(sflags & SVf_NOK));
3742 assert(!(sflags & SVf_IOK));
3744 else if (dtype == SVt_PVGV && isGV_with_GP(dstr)) {
3745 if (!(sflags & SVf_OK)) {
3746 if (ckWARN(WARN_MISC))
3747 Perl_warner(aTHX_ packWARN(WARN_MISC),
3748 "Undefined value assigned to typeglob");
3751 GV *gv = gv_fetchsv(sstr, GV_ADD, SVt_PVGV);
3752 if (dstr != (SV*)gv) {
3755 GvGP(dstr) = gp_ref(GvGP(gv));
3759 else if (sflags & SVp_POK) {
3763 * Check to see if we can just swipe the string. If so, it's a
3764 * possible small lose on short strings, but a big win on long ones.
3765 * It might even be a win on short strings if SvPVX_const(dstr)
3766 * has to be allocated and SvPVX_const(sstr) has to be freed.
3767 * Likewise if we can set up COW rather than doing an actual copy, we
3768 * drop to the else clause, as the swipe code and the COW setup code
3769 * have much in common.
3772 /* Whichever path we take through the next code, we want this true,
3773 and doing it now facilitates the COW check. */
3774 (void)SvPOK_only(dstr);
3777 /* If we're already COW then this clause is not true, and if COW
3778 is allowed then we drop down to the else and make dest COW
3779 with us. If caller hasn't said that we're allowed to COW
3780 shared hash keys then we don't do the COW setup, even if the
3781 source scalar is a shared hash key scalar. */
3782 (((flags & SV_COW_SHARED_HASH_KEYS)
3783 ? (sflags & (SVf_FAKE|SVf_READONLY)) != (SVf_FAKE|SVf_READONLY)
3784 : 1 /* If making a COW copy is forbidden then the behaviour we
3785 desire is as if the source SV isn't actually already
3786 COW, even if it is. So we act as if the source flags
3787 are not COW, rather than actually testing them. */
3789 #ifndef PERL_OLD_COPY_ON_WRITE
3790 /* The change that added SV_COW_SHARED_HASH_KEYS makes the logic
3791 when PERL_OLD_COPY_ON_WRITE is defined a little wrong.
3792 Conceptually PERL_OLD_COPY_ON_WRITE being defined should
3793 override SV_COW_SHARED_HASH_KEYS, because it means "always COW"
3794 but in turn, it's somewhat dead code, never expected to go
3795 live, but more kept as a placeholder on how to do it better
3796 in a newer implementation. */
3797 /* If we are COW and dstr is a suitable target then we drop down
3798 into the else and make dest a COW of us. */
3799 || (SvFLAGS(dstr) & CAN_COW_MASK) != CAN_COW_FLAGS
3804 (sflags & SVs_TEMP) && /* slated for free anyway? */
3805 !(sflags & SVf_OOK) && /* and not involved in OOK hack? */
3806 (!(flags & SV_NOSTEAL)) &&
3807 /* and we're allowed to steal temps */
3808 SvREFCNT(sstr) == 1 && /* and no other references to it? */
3809 SvLEN(sstr) && /* and really is a string */
3810 /* and won't be needed again, potentially */
3811 !(PL_op && PL_op->op_type == OP_AASSIGN))
3812 #ifdef PERL_OLD_COPY_ON_WRITE
3813 && ((flags & SV_COW_SHARED_HASH_KEYS)
3814 ? (!((sflags & CAN_COW_MASK) == CAN_COW_FLAGS
3815 && (SvFLAGS(dstr) & CAN_COW_MASK) == CAN_COW_FLAGS
3816 && SvTYPE(sstr) >= SVt_PVIV))
3820 /* Failed the swipe test, and it's not a shared hash key either.
3821 Have to copy the string. */
3822 STRLEN len = SvCUR(sstr);
3823 SvGROW(dstr, len + 1); /* inlined from sv_setpvn */
3824 Move(SvPVX_const(sstr),SvPVX(dstr),len,char);
3825 SvCUR_set(dstr, len);
3826 *SvEND(dstr) = '\0';
3828 /* If PERL_OLD_COPY_ON_WRITE is not defined, then isSwipe will always
3830 /* Either it's a shared hash key, or it's suitable for
3831 copy-on-write or we can swipe the string. */
3833 PerlIO_printf(Perl_debug_log, "Copy on write: sstr --> dstr\n");
3837 #ifdef PERL_OLD_COPY_ON_WRITE
3839 /* I believe I should acquire a global SV mutex if
3840 it's a COW sv (not a shared hash key) to stop
3841 it going un copy-on-write.
3842 If the source SV has gone un copy on write between up there
3843 and down here, then (assert() that) it is of the correct
3844 form to make it copy on write again */
3845 if ((sflags & (SVf_FAKE | SVf_READONLY))
3846 != (SVf_FAKE | SVf_READONLY)) {
3847 SvREADONLY_on(sstr);
3849 /* Make the source SV into a loop of 1.
3850 (about to become 2) */
3851 SV_COW_NEXT_SV_SET(sstr, sstr);
3855 /* Initial code is common. */
3856 if (SvPVX_const(dstr)) { /* we know that dtype >= SVt_PV */
3861 /* making another shared SV. */
3862 STRLEN cur = SvCUR(sstr);
3863 STRLEN len = SvLEN(sstr);
3864 #ifdef PERL_OLD_COPY_ON_WRITE
3866 assert (SvTYPE(dstr) >= SVt_PVIV);
3867 /* SvIsCOW_normal */
3868 /* splice us in between source and next-after-source. */
3869 SV_COW_NEXT_SV_SET(dstr, SV_COW_NEXT_SV(sstr));
3870 SV_COW_NEXT_SV_SET(sstr, dstr);
3871 SvPV_set(dstr, SvPVX_mutable(sstr));
3875 /* SvIsCOW_shared_hash */
3876 DEBUG_C(PerlIO_printf(Perl_debug_log,
3877 "Copy on write: Sharing hash\n"));
3879 assert (SvTYPE(dstr) >= SVt_PV);
3881 HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)))));
3883 SvLEN_set(dstr, len);
3884 SvCUR_set(dstr, cur);
3885 SvREADONLY_on(dstr);
3887 /* Relesase a global SV mutex. */
3890 { /* Passes the swipe test. */
3891 SvPV_set(dstr, SvPVX_mutable(sstr));
3892 SvLEN_set(dstr, SvLEN(sstr));
3893 SvCUR_set(dstr, SvCUR(sstr));
3896 (void)SvOK_off(sstr); /* NOTE: nukes most SvFLAGS on sstr */
3897 SvPV_set(sstr, NULL);
3903 if (sflags & SVp_NOK) {
3904 SvNV_set(dstr, SvNVX(sstr));
3906 if (sflags & SVp_IOK) {
3907 SvIV_set(dstr, SvIVX(sstr));
3908 /* Must do this otherwise some other overloaded use of 0x80000000
3909 gets confused. I guess SVpbm_VALID */
3910 if (sflags & SVf_IVisUV)
3913 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_NOK|SVp_NOK|SVf_UTF8);
3915 const MAGIC * const smg = SvVSTRING_mg(sstr);
3917 sv_magic(dstr, NULL, PERL_MAGIC_vstring,
3918 smg->mg_ptr, smg->mg_len);
3919 SvRMAGICAL_on(dstr);
3923 else if (sflags & (SVp_IOK|SVp_NOK)) {
3924 (void)SvOK_off(dstr);
3925 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_IVisUV|SVf_NOK|SVp_NOK);
3926 if (sflags & SVp_IOK) {
3927 /* XXXX Do we want to set IsUV for IV(ROK)? Be extra safe... */
3928 SvIV_set(dstr, SvIVX(sstr));
3930 if (sflags & SVp_NOK) {
3931 SvNV_set(dstr, SvNVX(sstr));
3935 if (isGV_with_GP(sstr)) {
3936 /* This stringification rule for globs is spread in 3 places.
3937 This feels bad. FIXME. */
3938 const U32 wasfake = sflags & SVf_FAKE;
3940 /* FAKE globs can get coerced, so need to turn this off
3941 temporarily if it is on. */
3943 gv_efullname3(dstr, (GV *)sstr, "*");
3944 SvFLAGS(sstr) |= wasfake;
3947 (void)SvOK_off(dstr);
3949 if (SvTAINTED(sstr))
3954 =for apidoc sv_setsv_mg
3956 Like C<sv_setsv>, but also handles 'set' magic.
3962 Perl_sv_setsv_mg(pTHX_ SV *const dstr, register SV *const sstr)
3964 PERL_ARGS_ASSERT_SV_SETSV_MG;
3966 sv_setsv(dstr,sstr);
3970 #ifdef PERL_OLD_COPY_ON_WRITE
3972 Perl_sv_setsv_cow(pTHX_ SV *dstr, SV *sstr)
3974 STRLEN cur = SvCUR(sstr);
3975 STRLEN len = SvLEN(sstr);
3976 register char *new_pv;
3978 PERL_ARGS_ASSERT_SV_SETSV_COW;
3981 PerlIO_printf(Perl_debug_log, "Fast copy on write: %p -> %p\n",
3982 (void*)sstr, (void*)dstr);
3989 if (SvTHINKFIRST(dstr))
3990 sv_force_normal_flags(dstr, SV_COW_DROP_PV);
3991 else if (SvPVX_const(dstr))
3992 Safefree(SvPVX_const(dstr));
3996 SvUPGRADE(dstr, SVt_PVIV);
3998 assert (SvPOK(sstr));
3999 assert (SvPOKp(sstr));
4000 assert (!SvIOK(sstr));
4001 assert (!SvIOKp(sstr));
4002 assert (!SvNOK(sstr));
4003 assert (!SvNOKp(sstr));
4005 if (SvIsCOW(sstr)) {
4007 if (SvLEN(sstr) == 0) {
4008 /* source is a COW shared hash key. */
4009 DEBUG_C(PerlIO_printf(Perl_debug_log,
4010 "Fast copy on write: Sharing hash\n"));
4011 new_pv = HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr))));
4014 SV_COW_NEXT_SV_SET(dstr, SV_COW_NEXT_SV(sstr));
4016 assert ((SvFLAGS(sstr) & CAN_COW_MASK) == CAN_COW_FLAGS);
4017 SvUPGRADE(sstr, SVt_PVIV);
4018 SvREADONLY_on(sstr);
4020 DEBUG_C(PerlIO_printf(Perl_debug_log,
4021 "Fast copy on write: Converting sstr to COW\n"));
4022 SV_COW_NEXT_SV_SET(dstr, sstr);
4024 SV_COW_NEXT_SV_SET(sstr, dstr);
4025 new_pv = SvPVX_mutable(sstr);
4028 SvPV_set(dstr, new_pv);
4029 SvFLAGS(dstr) = (SVt_PVIV|SVf_POK|SVp_POK|SVf_FAKE|SVf_READONLY);
4032 SvLEN_set(dstr, len);
4033 SvCUR_set(dstr, cur);
4042 =for apidoc sv_setpvn
4044 Copies a string into an SV. The C<len> parameter indicates the number of
4045 bytes to be copied. If the C<ptr> argument is NULL the SV will become
4046 undefined. Does not handle 'set' magic. See C<sv_setpvn_mg>.
4052 Perl_sv_setpvn(pTHX_ register SV *const sv, register const char *const ptr, register const STRLEN len)
4055 register char *dptr;
4057 PERL_ARGS_ASSERT_SV_SETPVN;
4059 SV_CHECK_THINKFIRST_COW_DROP(sv);
4065 /* len is STRLEN which is unsigned, need to copy to signed */
4068 Perl_croak(aTHX_ "panic: sv_setpvn called with negative strlen");
4070 SvUPGRADE(sv, SVt_PV);
4072 dptr = SvGROW(sv, len + 1);
4073 Move(ptr,dptr,len,char);
4076 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4081 =for apidoc sv_setpvn_mg
4083 Like C<sv_setpvn>, but also handles 'set' magic.
4089 Perl_sv_setpvn_mg(pTHX_ register SV *const sv, register const char *const ptr, register const STRLEN len)
4091 PERL_ARGS_ASSERT_SV_SETPVN_MG;
4093 sv_setpvn(sv,ptr,len);
4098 =for apidoc sv_setpv
4100 Copies a string into an SV. The string must be null-terminated. Does not
4101 handle 'set' magic. See C<sv_setpv_mg>.
4107 Perl_sv_setpv(pTHX_ register SV *const sv, register const char *const ptr)
4110 register STRLEN len;
4112 PERL_ARGS_ASSERT_SV_SETPV;
4114 SV_CHECK_THINKFIRST_COW_DROP(sv);
4120 SvUPGRADE(sv, SVt_PV);
4122 SvGROW(sv, len + 1);
4123 Move(ptr,SvPVX(sv),len+1,char);
4125 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4130 =for apidoc sv_setpv_mg
4132 Like C<sv_setpv>, but also handles 'set' magic.
4138 Perl_sv_setpv_mg(pTHX_ register SV *const sv, register const char *const ptr)
4140 PERL_ARGS_ASSERT_SV_SETPV_MG;
4147 =for apidoc sv_usepvn_flags
4149 Tells an SV to use C<ptr> to find its string value. Normally the
4150 string is stored inside the SV but sv_usepvn allows the SV to use an
4151 outside string. The C<ptr> should point to memory that was allocated
4152 by C<malloc>. The string length, C<len>, must be supplied. By default
4153 this function will realloc (i.e. move) the memory pointed to by C<ptr>,
4154 so that pointer should not be freed or used by the programmer after
4155 giving it to sv_usepvn, and neither should any pointers from "behind"
4156 that pointer (e.g. ptr + 1) be used.
4158 If C<flags> & SV_SMAGIC is true, will call SvSETMAGIC. If C<flags> &
4159 SV_HAS_TRAILING_NUL is true, then C<ptr[len]> must be NUL, and the realloc
4160 will be skipped. (i.e. the buffer is actually at least 1 byte longer than
4161 C<len>, and already meets the requirements for storing in C<SvPVX>)
4167 Perl_sv_usepvn_flags(pTHX_ SV *const sv, char *ptr, const STRLEN len, const U32 flags)
4172 PERL_ARGS_ASSERT_SV_USEPVN_FLAGS;
4174 SV_CHECK_THINKFIRST_COW_DROP(sv);
4175 SvUPGRADE(sv, SVt_PV);
4178 if (flags & SV_SMAGIC)
4182 if (SvPVX_const(sv))
4186 if (flags & SV_HAS_TRAILING_NUL)
4187 assert(ptr[len] == '\0');
4190 allocate = (flags & SV_HAS_TRAILING_NUL)
4192 #ifdef Perl_safesysmalloc_size
4195 PERL_STRLEN_ROUNDUP(len + 1);
4197 if (flags & SV_HAS_TRAILING_NUL) {
4198 /* It's long enough - do nothing.
4199 Specfically Perl_newCONSTSUB is relying on this. */
4202 /* Force a move to shake out bugs in callers. */
4203 char *new_ptr = (char*)safemalloc(allocate);
4204 Copy(ptr, new_ptr, len, char);
4205 PoisonFree(ptr,len,char);
4209 ptr = (char*) saferealloc (ptr, allocate);
4212 #ifdef Perl_safesysmalloc_size
4213 SvLEN_set(sv, Perl_safesysmalloc_size(ptr));
4215 SvLEN_set(sv, allocate);
4219 if (!(flags & SV_HAS_TRAILING_NUL)) {
4222 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4224 if (flags & SV_SMAGIC)
4228 #ifdef PERL_OLD_COPY_ON_WRITE
4229 /* Need to do this *after* making the SV normal, as we need the buffer
4230 pointer to remain valid until after we've copied it. If we let go too early,
4231 another thread could invalidate it by unsharing last of the same hash key
4232 (which it can do by means other than releasing copy-on-write Svs)
4233 or by changing the other copy-on-write SVs in the loop. */
4235 S_sv_release_COW(pTHX_ register SV *sv, const char *pvx, SV *after)
4237 PERL_ARGS_ASSERT_SV_RELEASE_COW;
4239 { /* this SV was SvIsCOW_normal(sv) */
4240 /* we need to find the SV pointing to us. */
4241 SV *current = SV_COW_NEXT_SV(after);
4243 if (current == sv) {
4244 /* The SV we point to points back to us (there were only two of us
4246 Hence other SV is no longer copy on write either. */
4248 SvREADONLY_off(after);
4250 /* We need to follow the pointers around the loop. */
4252 while ((next = SV_COW_NEXT_SV(current)) != sv) {
4255 /* don't loop forever if the structure is bust, and we have
4256 a pointer into a closed loop. */
4257 assert (current != after);
4258 assert (SvPVX_const(current) == pvx);
4260 /* Make the SV before us point to the SV after us. */
4261 SV_COW_NEXT_SV_SET(current, after);
4267 =for apidoc sv_force_normal_flags
4269 Undo various types of fakery on an SV: if the PV is a shared string, make
4270 a private copy; if we're a ref, stop refing; if we're a glob, downgrade to
4271 an xpvmg; if we're a copy-on-write scalar, this is the on-write time when
4272 we do the copy, and is also used locally. If C<SV_COW_DROP_PV> is set
4273 then a copy-on-write scalar drops its PV buffer (if any) and becomes
4274 SvPOK_off rather than making a copy. (Used where this scalar is about to be
4275 set to some other value.) In addition, the C<flags> parameter gets passed to
4276 C<sv_unref_flags()> when unrefing. C<sv_force_normal> calls this function
4277 with flags set to 0.
4283 Perl_sv_force_normal_flags(pTHX_ register SV *const sv, const U32 flags)
4287 PERL_ARGS_ASSERT_SV_FORCE_NORMAL_FLAGS;
4289 #ifdef PERL_OLD_COPY_ON_WRITE
4290 if (SvREADONLY(sv)) {
4291 /* At this point I believe I should acquire a global SV mutex. */
4293 const char * const pvx = SvPVX_const(sv);
4294 const STRLEN len = SvLEN(sv);
4295 const STRLEN cur = SvCUR(sv);
4296 /* next COW sv in the loop. If len is 0 then this is a shared-hash
4297 key scalar, so we mustn't attempt to call SV_COW_NEXT_SV(), as
4298 we'll fail an assertion. */
4299 SV * const next = len ? SV_COW_NEXT_SV(sv) : 0;
4302 PerlIO_printf(Perl_debug_log,
4303 "Copy on write: Force normal %ld\n",
4309 /* This SV doesn't own the buffer, so need to Newx() a new one: */
4312 if (flags & SV_COW_DROP_PV) {
4313 /* OK, so we don't need to copy our buffer. */
4316 SvGROW(sv, cur + 1);
4317 Move(pvx,SvPVX(sv),cur,char);
4322 sv_release_COW(sv, pvx, next);
4324 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
4330 else if (IN_PERL_RUNTIME)
4331 Perl_croak(aTHX_ PL_no_modify);
4332 /* At this point I believe that I can drop the global SV mutex. */
4335 if (SvREADONLY(sv)) {
4337 const char * const pvx = SvPVX_const(sv);
4338 const STRLEN len = SvCUR(sv);
4343 SvGROW(sv, len + 1);
4344 Move(pvx,SvPVX(sv),len,char);
4346 unshare_hek(SvSHARED_HEK_FROM_PV(pvx));
4348 else if (IN_PERL_RUNTIME)
4349 Perl_croak(aTHX_ PL_no_modify);
4353 sv_unref_flags(sv, flags);
4354 else if (SvFAKE(sv) && SvTYPE(sv) == SVt_PVGV)
4361 Efficient removal of characters from the beginning of the string buffer.
4362 SvPOK(sv) must be true and the C<ptr> must be a pointer to somewhere inside
4363 the string buffer. The C<ptr> becomes the first character of the adjusted
4364 string. Uses the "OOK hack".
4365 Beware: after this function returns, C<ptr> and SvPVX_const(sv) may no longer
4366 refer to the same chunk of data.
4372 Perl_sv_chop(pTHX_ register SV *const sv, register const char *const ptr)
4378 const U8 *real_start;
4381 PERL_ARGS_ASSERT_SV_CHOP;
4383 if (!ptr || !SvPOKp(sv))
4385 delta = ptr - SvPVX_const(sv);
4387 /* Nothing to do. */
4390 assert(ptr > SvPVX_const(sv));
4391 SV_CHECK_THINKFIRST(sv);
4394 if (!SvLEN(sv)) { /* make copy of shared string */
4395 const char *pvx = SvPVX_const(sv);
4396 const STRLEN len = SvCUR(sv);
4397 SvGROW(sv, len + 1);
4398 Move(pvx,SvPVX(sv),len,char);
4401 SvFLAGS(sv) |= SVf_OOK;
4404 SvOOK_offset(sv, old_delta);
4406 SvLEN_set(sv, SvLEN(sv) - delta);
4407 SvCUR_set(sv, SvCUR(sv) - delta);
4408 SvPV_set(sv, SvPVX(sv) + delta);
4410 p = (U8 *)SvPVX_const(sv);
4415 real_start = p - delta;
4419 if (delta < 0x100) {
4423 p -= sizeof(STRLEN);
4424 Copy((U8*)&delta, p, sizeof(STRLEN), U8);
4428 /* Fill the preceding buffer with sentinals to verify that no-one is
4430 while (p > real_start) {
4438 =for apidoc sv_catpvn
4440 Concatenates the string onto the end of the string which is in the SV. The
4441 C<len> indicates number of bytes to copy. If the SV has the UTF-8
4442 status set, then the bytes appended should be valid UTF-8.
4443 Handles 'get' magic, but not 'set' magic. See C<sv_catpvn_mg>.
4445 =for apidoc sv_catpvn_flags
4447 Concatenates the string onto the end of the string which is in the SV. The
4448 C<len> indicates number of bytes to copy. If the SV has the UTF-8
4449 status set, then the bytes appended should be valid UTF-8.
4450 If C<flags> has C<SV_GMAGIC> bit set, will C<mg_get> on C<dsv> if
4451 appropriate, else not. C<sv_catpvn> and C<sv_catpvn_nomg> are implemented
4452 in terms of this function.
4458 Perl_sv_catpvn_flags(pTHX_ register SV *const dsv, register const char *sstr, register const STRLEN slen, const I32 flags)
4462 const char * const dstr = SvPV_force_flags(dsv, dlen, flags);
4464 PERL_ARGS_ASSERT_SV_CATPVN_FLAGS;
4466 SvGROW(dsv, dlen + slen + 1);
4468 sstr = SvPVX_const(dsv);
4469 Move(sstr, SvPVX(dsv) + dlen, slen, char);
4470 SvCUR_set(dsv, SvCUR(dsv) + slen);
4472 (void)SvPOK_only_UTF8(dsv); /* validate pointer */
4474 if (flags & SV_SMAGIC)
4479 =for apidoc sv_catsv
4481 Concatenates the string from SV C<ssv> onto the end of the string in
4482 SV C<dsv>. Modifies C<dsv> but not C<ssv>. Handles 'get' magic, but
4483 not 'set' magic. See C<sv_catsv_mg>.
4485 =for apidoc sv_catsv_flags
4487 Concatenates the string from SV C<ssv> onto the end of the string in
4488 SV C<dsv>. Modifies C<dsv> but not C<ssv>. If C<flags> has C<SV_GMAGIC>
4489 bit set, will C<mg_get> on the SVs if appropriate, else not. C<sv_catsv>
4490 and C<sv_catsv_nomg> are implemented in terms of this function.
4495 Perl_sv_catsv_flags(pTHX_ SV *const dsv, register SV *const ssv, const I32 flags)
4499 PERL_ARGS_ASSERT_SV_CATSV_FLAGS;
4503 const char *spv = SvPV_const(ssv, slen);
4505 /* sutf8 and dutf8 were type bool, but under USE_ITHREADS,
4506 gcc version 2.95.2 20000220 (Debian GNU/Linux) for
4507 Linux xxx 2.2.17 on sparc64 with gcc -O2, we erroneously
4508 get dutf8 = 0x20000000, (i.e. SVf_UTF8) even though
4509 dsv->sv_flags doesn't have that bit set.
4510 Andy Dougherty 12 Oct 2001
4512 const I32 sutf8 = DO_UTF8(ssv);
4515 if (SvGMAGICAL(dsv) && (flags & SV_GMAGIC))
4517 dutf8 = DO_UTF8(dsv);
4519 if (dutf8 != sutf8) {
4521 /* Not modifying source SV, so taking a temporary copy. */
4522 SV* const csv = newSVpvn_flags(spv, slen, SVs_TEMP);
4524 sv_utf8_upgrade(csv);
4525 spv = SvPV_const(csv, slen);
4528 sv_utf8_upgrade_nomg(dsv);
4530 sv_catpvn_nomg(dsv, spv, slen);
4533 if (flags & SV_SMAGIC)
4538 =for apidoc sv_catpv
4540 Concatenates the string onto the end of the string which is in the SV.
4541 If the SV has the UTF-8 status set, then the bytes appended should be
4542 valid UTF-8. Handles 'get' magic, but not 'set' magic. See C<sv_catpv_mg>.
4547 Perl_sv_catpv(pTHX_ register SV *const sv, register const char *ptr)
4550 register STRLEN len;
4554 PERL_ARGS_ASSERT_SV_CATPV;
4558 junk = SvPV_force(sv, tlen);
4560 SvGROW(sv, tlen + len + 1);
4562 ptr = SvPVX_const(sv);
4563 Move(ptr,SvPVX(sv)+tlen,len+1,char);
4564 SvCUR_set(sv, SvCUR(sv) + len);
4565 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4570 =for apidoc sv_catpv_mg
4572 Like C<sv_catpv>, but also handles 'set' magic.
4578 Perl_sv_catpv_mg(pTHX_ register SV *const sv, register const char *const ptr)
4580 PERL_ARGS_ASSERT_SV_CATPV_MG;
4589 Creates a new SV. A non-zero C<len> parameter indicates the number of
4590 bytes of preallocated string space the SV should have. An extra byte for a
4591 trailing NUL is also reserved. (SvPOK is not set for the SV even if string
4592 space is allocated.) The reference count for the new SV is set to 1.
4594 In 5.9.3, newSV() replaces the older NEWSV() API, and drops the first
4595 parameter, I<x>, a debug aid which allowed callers to identify themselves.
4596 This aid has been superseded by a new build option, PERL_MEM_LOG (see
4597 L<perlhack/PERL_MEM_LOG>). The older API is still there for use in XS
4598 modules supporting older perls.
4604 Perl_newSV(pTHX_ const STRLEN len)
4611 sv_upgrade(sv, SVt_PV);
4612 SvGROW(sv, len + 1);
4617 =for apidoc sv_magicext
4619 Adds magic to an SV, upgrading it if necessary. Applies the
4620 supplied vtable and returns a pointer to the magic added.
4622 Note that C<sv_magicext> will allow things that C<sv_magic> will not.
4623 In particular, you can add magic to SvREADONLY SVs, and add more than
4624 one instance of the same 'how'.
4626 If C<namlen> is greater than zero then a C<savepvn> I<copy> of C<name> is
4627 stored, if C<namlen> is zero then C<name> is stored as-is and - as another
4628 special case - if C<(name && namlen == HEf_SVKEY)> then C<name> is assumed
4629 to contain an C<SV*> and is stored as-is with its REFCNT incremented.
4631 (This is now used as a subroutine by C<sv_magic>.)
4636 Perl_sv_magicext(pTHX_ SV *const sv, SV *const obj, const int how,
4637 const MGVTBL *const vtable, const char *const name, const I32 namlen)
4642 PERL_ARGS_ASSERT_SV_MAGICEXT;
4644 SvUPGRADE(sv, SVt_PVMG);
4645 Newxz(mg, 1, MAGIC);
4646 mg->mg_moremagic = SvMAGIC(sv);
4647 SvMAGIC_set(sv, mg);
4649 /* Sometimes a magic contains a reference loop, where the sv and
4650 object refer to each other. To prevent a reference loop that
4651 would prevent such objects being freed, we look for such loops
4652 and if we find one we avoid incrementing the object refcount.
4654 Note we cannot do this to avoid self-tie loops as intervening RV must
4655 have its REFCNT incremented to keep it in existence.
4658 if (!obj || obj == sv ||
4659 how == PERL_MAGIC_arylen ||
4660 how == PERL_MAGIC_symtab ||
4661 (SvTYPE(obj) == SVt_PVGV &&
4662 (GvSV(obj) == sv || GvHV(obj) == (HV*)sv || GvAV(obj) == (AV*)sv ||
4663 GvCV(obj) == (CV*)sv || GvIOp(obj) == (IO*)sv ||
4664 GvFORM(obj) == (CV*)sv)))
4669 mg->mg_obj = SvREFCNT_inc_simple(obj);
4670 mg->mg_flags |= MGf_REFCOUNTED;
4673 /* Normal self-ties simply pass a null object, and instead of
4674 using mg_obj directly, use the SvTIED_obj macro to produce a
4675 new RV as needed. For glob "self-ties", we are tieing the PVIO
4676 with an RV obj pointing to the glob containing the PVIO. In
4677 this case, to avoid a reference loop, we need to weaken the
4681 if (how == PERL_MAGIC_tiedscalar && SvTYPE(sv) == SVt_PVIO &&
4682 obj && SvROK(obj) && GvIO(SvRV(obj)) == (IO*)sv)
4688 mg->mg_len = namlen;
4691 mg->mg_ptr = savepvn(name, namlen);
4692 else if (namlen == HEf_SVKEY)
4693 mg->mg_ptr = (char*)SvREFCNT_inc_simple_NN((SV*)name);
4695 mg->mg_ptr = (char *) name;
4697 mg->mg_virtual = (MGVTBL *) vtable;
4701 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK|SVf_POK);
4706 =for apidoc sv_magic
4708 Adds magic to an SV. First upgrades C<sv> to type C<SVt_PVMG> if necessary,
4709 then adds a new magic item of type C<how> to the head of the magic list.
4711 See C<sv_magicext> (which C<sv_magic> now calls) for a description of the
4712 handling of the C<name> and C<namlen> arguments.
4714 You need to use C<sv_magicext> to add magic to SvREADONLY SVs and also
4715 to add more than one instance of the same 'how'.
4721 Perl_sv_magic(pTHX_ register SV *const sv, SV *const obj, const int how,
4722 const char *const name, const I32 namlen)
4725 const MGVTBL *vtable;
4728 PERL_ARGS_ASSERT_SV_MAGIC;
4730 #ifdef PERL_OLD_COPY_ON_WRITE
4732 sv_force_normal_flags(sv, 0);
4734 if (SvREADONLY(sv)) {
4736 /* its okay to attach magic to shared strings; the subsequent
4737 * upgrade to PVMG will unshare the string */
4738 !(SvFAKE(sv) && SvTYPE(sv) < SVt_PVMG)
4741 && how != PERL_MAGIC_regex_global
4742 && how != PERL_MAGIC_bm
4743 && how != PERL_MAGIC_fm
4744 && how != PERL_MAGIC_sv
4745 && how != PERL_MAGIC_backref
4748 Perl_croak(aTHX_ PL_no_modify);
4751 if (SvMAGICAL(sv) || (how == PERL_MAGIC_taint && SvTYPE(sv) >= SVt_PVMG)) {
4752 if (SvMAGIC(sv) && (mg = mg_find(sv, how))) {
4753 /* sv_magic() refuses to add a magic of the same 'how' as an
4756 if (how == PERL_MAGIC_taint) {
4758 /* Any scalar which already had taint magic on which someone