3 * Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000,
4 * 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009 by Larry Wall
7 * You may distribute under the terms of either the GNU General Public
8 * License or the Artistic License, as specified in the README file.
13 * 'I wonder what the Entish is for "yes" and "no",' he thought.
16 * [p.480 of _The Lord of the Rings_, III/iv: "Treebeard"]
22 * This file contains the code that creates, manipulates and destroys
23 * scalar values (SVs). The other types (AV, HV, GV, etc.) reuse the
24 * structure of an SV, so their creation and destruction is handled
25 * here; higher-level functions are in av.c, hv.c, and so on. Opcode
26 * level functions (eg. substr, split, join) for each of the types are
36 # if __STDC_VERSION__ >= 199901L && !defined(VMS)
47 /* Missing proto on LynxOS */
48 char *gconvert(double, int, int, char *);
51 #ifdef PERL_UTF8_CACHE_ASSERT
52 /* if adding more checks watch out for the following tests:
53 * t/op/index.t t/op/length.t t/op/pat.t t/op/substr.t
54 * lib/utf8.t lib/Unicode/Collate/t/index.t
57 # define ASSERT_UTF8_CACHE(cache) \
58 STMT_START { if (cache) { assert((cache)[0] <= (cache)[1]); \
59 assert((cache)[2] <= (cache)[3]); \
60 assert((cache)[3] <= (cache)[1]);} \
63 # define ASSERT_UTF8_CACHE(cache) NOOP
66 #ifdef PERL_OLD_COPY_ON_WRITE
67 #define SV_COW_NEXT_SV(sv) INT2PTR(SV *,SvUVX(sv))
68 #define SV_COW_NEXT_SV_SET(current,next) SvUV_set(current, PTR2UV(next))
69 /* This is a pessimistic view. Scalar must be purely a read-write PV to copy-
73 /* ============================================================================
75 =head1 Allocation and deallocation of SVs.
77 An SV (or AV, HV, etc.) is allocated in two parts: the head (struct
78 sv, av, hv...) contains type and reference count information, and for
79 many types, a pointer to the body (struct xrv, xpv, xpviv...), which
80 contains fields specific to each type. Some types store all they need
81 in the head, so don't have a body.
83 In all but the most memory-paranoid configurations (ex: PURIFY), heads
84 and bodies are allocated out of arenas, which by default are
85 approximately 4K chunks of memory parcelled up into N heads or bodies.
86 Sv-bodies are allocated by their sv-type, guaranteeing size
87 consistency needed to allocate safely from arrays.
89 For SV-heads, the first slot in each arena is reserved, and holds a
90 link to the next arena, some flags, and a note of the number of slots.
91 Snaked through each arena chain is a linked list of free items; when
92 this becomes empty, an extra arena is allocated and divided up into N
93 items which are threaded into the free list.
95 SV-bodies are similar, but they use arena-sets by default, which
96 separate the link and info from the arena itself, and reclaim the 1st
97 slot in the arena. SV-bodies are further described later.
99 The following global variables are associated with arenas:
101 PL_sv_arenaroot pointer to list of SV arenas
102 PL_sv_root pointer to list of free SV structures
104 PL_body_arenas head of linked-list of body arenas
105 PL_body_roots[] array of pointers to list of free bodies of svtype
106 arrays are indexed by the svtype needed
108 A few special SV heads are not allocated from an arena, but are
109 instead directly created in the interpreter structure, eg PL_sv_undef.
110 The size of arenas can be changed from the default by setting
111 PERL_ARENA_SIZE appropriately at compile time.
113 The SV arena serves the secondary purpose of allowing still-live SVs
114 to be located and destroyed during final cleanup.
116 At the lowest level, the macros new_SV() and del_SV() grab and free
117 an SV head. (If debugging with -DD, del_SV() calls the function S_del_sv()
118 to return the SV to the free list with error checking.) new_SV() calls
119 more_sv() / sv_add_arena() to add an extra arena if the free list is empty.
120 SVs in the free list have their SvTYPE field set to all ones.
122 At the time of very final cleanup, sv_free_arenas() is called from
123 perl_destruct() to physically free all the arenas allocated since the
124 start of the interpreter.
126 The function visit() scans the SV arenas list, and calls a specified
127 function for each SV it finds which is still live - ie which has an SvTYPE
128 other than all 1's, and a non-zero SvREFCNT. visit() is used by the
129 following functions (specified as [function that calls visit()] / [function
130 called by visit() for each SV]):
132 sv_report_used() / do_report_used()
133 dump all remaining SVs (debugging aid)
135 sv_clean_objs() / do_clean_objs(),do_clean_named_objs(),
136 do_clean_named_io_objs()
137 Attempt to free all objects pointed to by RVs,
138 and try to do the same for all objects indirectly
139 referenced by typeglobs too. Called once from
140 perl_destruct(), prior to calling sv_clean_all()
143 sv_clean_all() / do_clean_all()
144 SvREFCNT_dec(sv) each remaining SV, possibly
145 triggering an sv_free(). It also sets the
146 SVf_BREAK flag on the SV to indicate that the
147 refcnt has been artificially lowered, and thus
148 stopping sv_free() from giving spurious warnings
149 about SVs which unexpectedly have a refcnt
150 of zero. called repeatedly from perl_destruct()
151 until there are no SVs left.
153 =head2 Arena allocator API Summary
155 Private API to rest of sv.c
159 new_XPVNV(), del_XPVGV(),
164 sv_report_used(), sv_clean_objs(), sv_clean_all(), sv_free_arenas()
168 * ========================================================================= */
171 * "A time to plant, and a time to uproot what was planted..."
175 # define MEM_LOG_NEW_SV(sv, file, line, func) \
176 Perl_mem_log_new_sv(sv, file, line, func)
177 # define MEM_LOG_DEL_SV(sv, file, line, func) \
178 Perl_mem_log_del_sv(sv, file, line, func)
180 # define MEM_LOG_NEW_SV(sv, file, line, func) NOOP
181 # define MEM_LOG_DEL_SV(sv, file, line, func) NOOP
184 #ifdef DEBUG_LEAKING_SCALARS
185 # define FREE_SV_DEBUG_FILE(sv) Safefree((sv)->sv_debug_file)
186 # define DEBUG_SV_SERIAL(sv) \
187 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) del_SV\n", \
188 PTR2UV(sv), (long)(sv)->sv_debug_serial))
190 # define FREE_SV_DEBUG_FILE(sv)
191 # define DEBUG_SV_SERIAL(sv) NOOP
195 # define SvARENA_CHAIN(sv) ((sv)->sv_u.svu_rv)
196 # define SvARENA_CHAIN_SET(sv,val) (sv)->sv_u.svu_rv = MUTABLE_SV((val))
197 /* Whilst I'd love to do this, it seems that things like to check on
199 # define POSION_SV_HEAD(sv) PoisonNew(sv, 1, struct STRUCT_SV)
201 # define POSION_SV_HEAD(sv) PoisonNew(&SvANY(sv), 1, void *), \
202 PoisonNew(&SvREFCNT(sv), 1, U32)
204 # define SvARENA_CHAIN(sv) SvANY(sv)
205 # define SvARENA_CHAIN_SET(sv,val) SvANY(sv) = (void *)(val)
206 # define POSION_SV_HEAD(sv)
209 /* Mark an SV head as unused, and add to free list.
211 * If SVf_BREAK is set, skip adding it to the free list, as this SV had
212 * its refcount artificially decremented during global destruction, so
213 * there may be dangling pointers to it. The last thing we want in that
214 * case is for it to be reused. */
216 #define plant_SV(p) \
218 const U32 old_flags = SvFLAGS(p); \
219 MEM_LOG_DEL_SV(p, __FILE__, __LINE__, FUNCTION__); \
220 DEBUG_SV_SERIAL(p); \
221 FREE_SV_DEBUG_FILE(p); \
223 SvFLAGS(p) = SVTYPEMASK; \
224 if (!(old_flags & SVf_BREAK)) { \
225 SvARENA_CHAIN_SET(p, PL_sv_root); \
231 #define uproot_SV(p) \
234 PL_sv_root = MUTABLE_SV(SvARENA_CHAIN(p)); \
239 /* make some more SVs by adding another arena */
246 char *chunk; /* must use New here to match call to */
247 Newx(chunk,PERL_ARENA_SIZE,char); /* Safefree() in sv_free_arenas() */
248 sv_add_arena(chunk, PERL_ARENA_SIZE, 0);
253 /* new_SV(): return a new, empty SV head */
255 #ifdef DEBUG_LEAKING_SCALARS
256 /* provide a real function for a debugger to play with */
258 S_new_SV(pTHX_ const char *file, int line, const char *func)
265 sv = S_more_sv(aTHX);
269 sv->sv_debug_optype = PL_op ? PL_op->op_type : 0;
270 sv->sv_debug_line = (U16) (PL_parser && PL_parser->copline != NOLINE
276 sv->sv_debug_inpad = 0;
277 sv->sv_debug_parent = NULL;
278 sv->sv_debug_file = PL_curcop ? savepv(CopFILE(PL_curcop)): NULL;
280 sv->sv_debug_serial = PL_sv_serial++;
282 MEM_LOG_NEW_SV(sv, file, line, func);
283 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) new_SV (from %s:%d [%s])\n",
284 PTR2UV(sv), (long)sv->sv_debug_serial, file, line, func));
288 # define new_SV(p) (p)=S_new_SV(aTHX_ __FILE__, __LINE__, FUNCTION__)
296 (p) = S_more_sv(aTHX); \
300 MEM_LOG_NEW_SV(p, __FILE__, __LINE__, FUNCTION__); \
305 /* del_SV(): return an empty SV head to the free list */
318 S_del_sv(pTHX_ SV *p)
322 PERL_ARGS_ASSERT_DEL_SV;
327 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
328 const SV * const sv = sva + 1;
329 const SV * const svend = &sva[SvREFCNT(sva)];
330 if (p >= sv && p < svend) {
336 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
337 "Attempt to free non-arena SV: 0x%"UVxf
338 pTHX__FORMAT, PTR2UV(p) pTHX__VALUE);
345 #else /* ! DEBUGGING */
347 #define del_SV(p) plant_SV(p)
349 #endif /* DEBUGGING */
353 =head1 SV Manipulation Functions
355 =for apidoc sv_add_arena
357 Given a chunk of memory, link it to the head of the list of arenas,
358 and split it into a list of free SVs.
364 S_sv_add_arena(pTHX_ char *const ptr, const U32 size, const U32 flags)
367 SV *const sva = MUTABLE_SV(ptr);
371 PERL_ARGS_ASSERT_SV_ADD_ARENA;
373 /* The first SV in an arena isn't an SV. */
374 SvANY(sva) = (void *) PL_sv_arenaroot; /* ptr to next arena */
375 SvREFCNT(sva) = size / sizeof(SV); /* number of SV slots */
376 SvFLAGS(sva) = flags; /* FAKE if not to be freed */
378 PL_sv_arenaroot = sva;
379 PL_sv_root = sva + 1;
381 svend = &sva[SvREFCNT(sva) - 1];
384 SvARENA_CHAIN_SET(sv, (sv + 1));
388 /* Must always set typemask because it's always checked in on cleanup
389 when the arenas are walked looking for objects. */
390 SvFLAGS(sv) = SVTYPEMASK;
393 SvARENA_CHAIN_SET(sv, 0);
397 SvFLAGS(sv) = SVTYPEMASK;
400 /* visit(): call the named function for each non-free SV in the arenas
401 * whose flags field matches the flags/mask args. */
404 S_visit(pTHX_ SVFUNC_t f, const U32 flags, const U32 mask)
410 PERL_ARGS_ASSERT_VISIT;
412 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
413 register const SV * const svend = &sva[SvREFCNT(sva)];
415 for (sv = sva + 1; sv < svend; ++sv) {
416 if (SvTYPE(sv) != (svtype)SVTYPEMASK
417 && (sv->sv_flags & mask) == flags
430 /* called by sv_report_used() for each live SV */
433 do_report_used(pTHX_ SV *const sv)
435 if (SvTYPE(sv) != (svtype)SVTYPEMASK) {
436 PerlIO_printf(Perl_debug_log, "****\n");
443 =for apidoc sv_report_used
445 Dump the contents of all SVs not yet freed. (Debugging aid).
451 Perl_sv_report_used(pTHX)
454 visit(do_report_used, 0, 0);
460 /* called by sv_clean_objs() for each live SV */
463 do_clean_objs(pTHX_ SV *const ref)
468 SV * const target = SvRV(ref);
469 if (SvOBJECT(target)) {
470 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning object ref:\n "), sv_dump(ref)));
471 if (SvWEAKREF(ref)) {
472 sv_del_backref(target, ref);
478 SvREFCNT_dec(target);
483 /* XXX Might want to check arrays, etc. */
487 /* clear any slots in a GV which hold objects - except IO;
488 * called by sv_clean_objs() for each live GV */
491 do_clean_named_objs(pTHX_ SV *const sv)
495 assert(SvTYPE(sv) == SVt_PVGV);
496 assert(isGV_with_GP(sv));
500 /* freeing GP entries may indirectly free the current GV;
501 * hold onto it while we mess with the GP slots */
504 if ( ((obj = GvSV(sv) )) && SvOBJECT(obj)) {
505 DEBUG_D((PerlIO_printf(Perl_debug_log,
506 "Cleaning named glob SV object:\n "), sv_dump(obj)));
510 if ( ((obj = MUTABLE_SV(GvAV(sv)) )) && SvOBJECT(obj)) {
511 DEBUG_D((PerlIO_printf(Perl_debug_log,
512 "Cleaning named glob AV object:\n "), sv_dump(obj)));
516 if ( ((obj = MUTABLE_SV(GvHV(sv)) )) && SvOBJECT(obj)) {
517 DEBUG_D((PerlIO_printf(Perl_debug_log,
518 "Cleaning named glob HV object:\n "), sv_dump(obj)));
522 if ( ((obj = MUTABLE_SV(GvCV(sv)) )) && SvOBJECT(obj)) {
523 DEBUG_D((PerlIO_printf(Perl_debug_log,
524 "Cleaning named glob CV object:\n "), sv_dump(obj)));
528 SvREFCNT_dec(sv); /* undo the inc above */
531 /* clear any IO slots in a GV which hold objects (except stderr, defout);
532 * called by sv_clean_objs() for each live GV */
535 do_clean_named_io_objs(pTHX_ SV *const sv)
539 assert(SvTYPE(sv) == SVt_PVGV);
540 assert(isGV_with_GP(sv));
541 if (!GvGP(sv) || sv == (SV*)PL_stderrgv || sv == (SV*)PL_defoutgv)
545 if ( ((obj = MUTABLE_SV(GvIO(sv)) )) && SvOBJECT(obj)) {
546 DEBUG_D((PerlIO_printf(Perl_debug_log,
547 "Cleaning named glob IO object:\n "), sv_dump(obj)));
551 SvREFCNT_dec(sv); /* undo the inc above */
554 /* Void wrapper to pass to visit() */
556 do_curse(pTHX_ SV * const sv) {
557 if ((PL_stderrgv && GvGP(PL_stderrgv) && (SV*)GvIO(PL_stderrgv) == sv)
558 || (PL_defoutgv && GvGP(PL_defoutgv) && (SV*)GvIO(PL_defoutgv) == sv))
564 =for apidoc sv_clean_objs
566 Attempt to destroy all objects not yet freed
572 Perl_sv_clean_objs(pTHX)
576 PL_in_clean_objs = TRUE;
577 visit(do_clean_objs, SVf_ROK, SVf_ROK);
578 /* Some barnacles may yet remain, clinging to typeglobs.
579 * Run the non-IO destructors first: they may want to output
580 * error messages, close files etc */
581 visit(do_clean_named_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
582 visit(do_clean_named_io_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
583 /* And if there are some very tenacious barnacles clinging to arrays,
584 closures, or what have you.... */
585 visit(do_curse, SVs_OBJECT, SVs_OBJECT);
586 olddef = PL_defoutgv;
587 PL_defoutgv = NULL; /* disable skip of PL_defoutgv */
588 if (olddef && isGV_with_GP(olddef))
589 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olddef));
590 olderr = PL_stderrgv;
591 PL_stderrgv = NULL; /* disable skip of PL_stderrgv */
592 if (olderr && isGV_with_GP(olderr))
593 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olderr));
594 SvREFCNT_dec(olddef);
595 PL_in_clean_objs = FALSE;
598 /* called by sv_clean_all() for each live SV */
601 do_clean_all(pTHX_ SV *const sv)
604 if (sv == (const SV *) PL_fdpid || sv == (const SV *)PL_strtab) {
605 /* don't clean pid table and strtab */
608 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning loops: SV at 0x%"UVxf"\n", PTR2UV(sv)) ));
609 SvFLAGS(sv) |= SVf_BREAK;
614 =for apidoc sv_clean_all
616 Decrement the refcnt of each remaining SV, possibly triggering a
617 cleanup. This function may have to be called multiple times to free
618 SVs which are in complex self-referential hierarchies.
624 Perl_sv_clean_all(pTHX)
628 PL_in_clean_all = TRUE;
629 cleaned = visit(do_clean_all, 0,0);
634 ARENASETS: a meta-arena implementation which separates arena-info
635 into struct arena_set, which contains an array of struct
636 arena_descs, each holding info for a single arena. By separating
637 the meta-info from the arena, we recover the 1st slot, formerly
638 borrowed for list management. The arena_set is about the size of an
639 arena, avoiding the needless malloc overhead of a naive linked-list.
641 The cost is 1 arena-set malloc per ~320 arena-mallocs, + the unused
642 memory in the last arena-set (1/2 on average). In trade, we get
643 back the 1st slot in each arena (ie 1.7% of a CV-arena, less for
644 smaller types). The recovery of the wasted space allows use of
645 small arenas for large, rare body types, by changing array* fields
646 in body_details_by_type[] below.
649 char *arena; /* the raw storage, allocated aligned */
650 size_t size; /* its size ~4k typ */
651 svtype utype; /* bodytype stored in arena */
656 /* Get the maximum number of elements in set[] such that struct arena_set
657 will fit within PERL_ARENA_SIZE, which is probably just under 4K, and
658 therefore likely to be 1 aligned memory page. */
660 #define ARENAS_PER_SET ((PERL_ARENA_SIZE - sizeof(struct arena_set*) \
661 - 2 * sizeof(int)) / sizeof (struct arena_desc))
664 struct arena_set* next;
665 unsigned int set_size; /* ie ARENAS_PER_SET */
666 unsigned int curr; /* index of next available arena-desc */
667 struct arena_desc set[ARENAS_PER_SET];
671 =for apidoc sv_free_arenas
673 Deallocate the memory used by all arenas. Note that all the individual SV
674 heads and bodies within the arenas must already have been freed.
679 Perl_sv_free_arenas(pTHX)
686 /* Free arenas here, but be careful about fake ones. (We assume
687 contiguity of the fake ones with the corresponding real ones.) */
689 for (sva = PL_sv_arenaroot; sva; sva = svanext) {
690 svanext = MUTABLE_SV(SvANY(sva));
691 while (svanext && SvFAKE(svanext))
692 svanext = MUTABLE_SV(SvANY(svanext));
699 struct arena_set *aroot = (struct arena_set*) PL_body_arenas;
702 struct arena_set *current = aroot;
705 assert(aroot->set[i].arena);
706 Safefree(aroot->set[i].arena);
714 i = PERL_ARENA_ROOTS_SIZE;
716 PL_body_roots[i] = 0;
723 Here are mid-level routines that manage the allocation of bodies out
724 of the various arenas. There are 5 kinds of arenas:
726 1. SV-head arenas, which are discussed and handled above
727 2. regular body arenas
728 3. arenas for reduced-size bodies
731 Arena types 2 & 3 are chained by body-type off an array of
732 arena-root pointers, which is indexed by svtype. Some of the
733 larger/less used body types are malloced singly, since a large
734 unused block of them is wasteful. Also, several svtypes dont have
735 bodies; the data fits into the sv-head itself. The arena-root
736 pointer thus has a few unused root-pointers (which may be hijacked
737 later for arena types 4,5)
739 3 differs from 2 as an optimization; some body types have several
740 unused fields in the front of the structure (which are kept in-place
741 for consistency). These bodies can be allocated in smaller chunks,
742 because the leading fields arent accessed. Pointers to such bodies
743 are decremented to point at the unused 'ghost' memory, knowing that
744 the pointers are used with offsets to the real memory.
747 =head1 SV-Body Allocation
749 Allocation of SV-bodies is similar to SV-heads, differing as follows;
750 the allocation mechanism is used for many body types, so is somewhat
751 more complicated, it uses arena-sets, and has no need for still-live
754 At the outermost level, (new|del)_X*V macros return bodies of the
755 appropriate type. These macros call either (new|del)_body_type or
756 (new|del)_body_allocated macro pairs, depending on specifics of the
757 type. Most body types use the former pair, the latter pair is used to
758 allocate body types with "ghost fields".
760 "ghost fields" are fields that are unused in certain types, and
761 consequently don't need to actually exist. They are declared because
762 they're part of a "base type", which allows use of functions as
763 methods. The simplest examples are AVs and HVs, 2 aggregate types
764 which don't use the fields which support SCALAR semantics.
766 For these types, the arenas are carved up into appropriately sized
767 chunks, we thus avoid wasted memory for those unaccessed members.
768 When bodies are allocated, we adjust the pointer back in memory by the
769 size of the part not allocated, so it's as if we allocated the full
770 structure. (But things will all go boom if you write to the part that
771 is "not there", because you'll be overwriting the last members of the
772 preceding structure in memory.)
774 We calculate the correction using the STRUCT_OFFSET macro on the first
775 member present. If the allocated structure is smaller (no initial NV
776 actually allocated) then the net effect is to subtract the size of the NV
777 from the pointer, to return a new pointer as if an initial NV were actually
778 allocated. (We were using structures named *_allocated for this, but
779 this turned out to be a subtle bug, because a structure without an NV
780 could have a lower alignment constraint, but the compiler is allowed to
781 optimised accesses based on the alignment constraint of the actual pointer
782 to the full structure, for example, using a single 64 bit load instruction
783 because it "knows" that two adjacent 32 bit members will be 8-byte aligned.)
785 This is the same trick as was used for NV and IV bodies. Ironically it
786 doesn't need to be used for NV bodies any more, because NV is now at
787 the start of the structure. IV bodies don't need it either, because
788 they are no longer allocated.
790 In turn, the new_body_* allocators call S_new_body(), which invokes
791 new_body_inline macro, which takes a lock, and takes a body off the
792 linked list at PL_body_roots[sv_type], calling Perl_more_bodies() if
793 necessary to refresh an empty list. Then the lock is released, and
794 the body is returned.
796 Perl_more_bodies allocates a new arena, and carves it up into an array of N
797 bodies, which it strings into a linked list. It looks up arena-size
798 and body-size from the body_details table described below, thus
799 supporting the multiple body-types.
801 If PURIFY is defined, or PERL_ARENA_SIZE=0, arenas are not used, and
802 the (new|del)_X*V macros are mapped directly to malloc/free.
804 For each sv-type, struct body_details bodies_by_type[] carries
805 parameters which control these aspects of SV handling:
807 Arena_size determines whether arenas are used for this body type, and if
808 so, how big they are. PURIFY or PERL_ARENA_SIZE=0 set this field to
809 zero, forcing individual mallocs and frees.
811 Body_size determines how big a body is, and therefore how many fit into
812 each arena. Offset carries the body-pointer adjustment needed for
813 "ghost fields", and is used in *_allocated macros.
815 But its main purpose is to parameterize info needed in
816 Perl_sv_upgrade(). The info here dramatically simplifies the function
817 vs the implementation in 5.8.8, making it table-driven. All fields
818 are used for this, except for arena_size.
820 For the sv-types that have no bodies, arenas are not used, so those
821 PL_body_roots[sv_type] are unused, and can be overloaded. In
822 something of a special case, SVt_NULL is borrowed for HE arenas;
823 PL_body_roots[HE_SVSLOT=SVt_NULL] is filled by S_more_he, but the
824 bodies_by_type[SVt_NULL] slot is not used, as the table is not
829 struct body_details {
830 U8 body_size; /* Size to allocate */
831 U8 copy; /* Size of structure to copy (may be shorter) */
833 unsigned int type : 4; /* We have space for a sanity check. */
834 unsigned int cant_upgrade : 1; /* Cannot upgrade this type */
835 unsigned int zero_nv : 1; /* zero the NV when upgrading from this */
836 unsigned int arena : 1; /* Allocated from an arena */
837 size_t arena_size; /* Size of arena to allocate */
845 /* With -DPURFIY we allocate everything directly, and don't use arenas.
846 This seems a rather elegant way to simplify some of the code below. */
847 #define HASARENA FALSE
849 #define HASARENA TRUE
851 #define NOARENA FALSE
853 /* Size the arenas to exactly fit a given number of bodies. A count
854 of 0 fits the max number bodies into a PERL_ARENA_SIZE.block,
855 simplifying the default. If count > 0, the arena is sized to fit
856 only that many bodies, allowing arenas to be used for large, rare
857 bodies (XPVFM, XPVIO) without undue waste. The arena size is
858 limited by PERL_ARENA_SIZE, so we can safely oversize the
861 #define FIT_ARENA0(body_size) \
862 ((size_t)(PERL_ARENA_SIZE / body_size) * body_size)
863 #define FIT_ARENAn(count,body_size) \
864 ( count * body_size <= PERL_ARENA_SIZE) \
865 ? count * body_size \
866 : FIT_ARENA0 (body_size)
867 #define FIT_ARENA(count,body_size) \
869 ? FIT_ARENAn (count, body_size) \
870 : FIT_ARENA0 (body_size)
872 /* Calculate the length to copy. Specifically work out the length less any
873 final padding the compiler needed to add. See the comment in sv_upgrade
874 for why copying the padding proved to be a bug. */
876 #define copy_length(type, last_member) \
877 STRUCT_OFFSET(type, last_member) \
878 + sizeof (((type*)SvANY((const SV *)0))->last_member)
880 static const struct body_details bodies_by_type[] = {
881 /* HEs use this offset for their arena. */
882 { 0, 0, 0, SVt_NULL, FALSE, NONV, NOARENA, 0 },
884 /* The bind placeholder pretends to be an RV for now.
885 Also it's marked as "can't upgrade" to stop anyone using it before it's
887 { 0, 0, 0, SVt_BIND, TRUE, NONV, NOARENA, 0 },
889 /* IVs are in the head, so the allocation size is 0. */
891 sizeof(IV), /* This is used to copy out the IV body. */
892 STRUCT_OFFSET(XPVIV, xiv_iv), SVt_IV, FALSE, NONV,
893 NOARENA /* IVS don't need an arena */, 0
896 { sizeof(NV), sizeof(NV),
897 STRUCT_OFFSET(XPVNV, xnv_u),
898 SVt_NV, FALSE, HADNV, HASARENA, FIT_ARENA(0, sizeof(NV)) },
900 { sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur),
901 copy_length(XPV, xpv_len) - STRUCT_OFFSET(XPV, xpv_cur),
902 + STRUCT_OFFSET(XPV, xpv_cur),
903 SVt_PV, FALSE, NONV, HASARENA,
904 FIT_ARENA(0, sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur)) },
906 { sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur),
907 copy_length(XPVIV, xiv_u) - STRUCT_OFFSET(XPV, xpv_cur),
908 + STRUCT_OFFSET(XPV, xpv_cur),
909 SVt_PVIV, FALSE, NONV, HASARENA,
910 FIT_ARENA(0, sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur)) },
912 { sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur),
913 copy_length(XPVNV, xnv_u) - STRUCT_OFFSET(XPV, xpv_cur),
914 + STRUCT_OFFSET(XPV, xpv_cur),
915 SVt_PVNV, FALSE, HADNV, HASARENA,
916 FIT_ARENA(0, sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur)) },
918 { sizeof(XPVMG), copy_length(XPVMG, xnv_u), 0, SVt_PVMG, FALSE, HADNV,
919 HASARENA, FIT_ARENA(0, sizeof(XPVMG)) },
924 SVt_REGEXP, FALSE, NONV, HASARENA,
925 FIT_ARENA(0, sizeof(regexp))
928 { sizeof(XPVGV), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV,
929 HASARENA, FIT_ARENA(0, sizeof(XPVGV)) },
931 { sizeof(XPVLV), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV,
932 HASARENA, FIT_ARENA(0, sizeof(XPVLV)) },
935 copy_length(XPVAV, xav_alloc),
937 SVt_PVAV, TRUE, NONV, HASARENA,
938 FIT_ARENA(0, sizeof(XPVAV)) },
941 copy_length(XPVHV, xhv_max),
943 SVt_PVHV, TRUE, NONV, HASARENA,
944 FIT_ARENA(0, sizeof(XPVHV)) },
949 SVt_PVCV, TRUE, NONV, HASARENA,
950 FIT_ARENA(0, sizeof(XPVCV)) },
955 SVt_PVFM, TRUE, NONV, NOARENA,
956 FIT_ARENA(20, sizeof(XPVFM)) },
961 SVt_PVIO, TRUE, NONV, HASARENA,
962 FIT_ARENA(24, sizeof(XPVIO)) },
965 #define new_body_allocated(sv_type) \
966 (void *)((char *)S_new_body(aTHX_ sv_type) \
967 - bodies_by_type[sv_type].offset)
969 /* return a thing to the free list */
971 #define del_body(thing, root) \
973 void ** const thing_copy = (void **)thing; \
974 *thing_copy = *root; \
975 *root = (void*)thing_copy; \
980 #define new_XNV() safemalloc(sizeof(XPVNV))
981 #define new_XPVNV() safemalloc(sizeof(XPVNV))
982 #define new_XPVMG() safemalloc(sizeof(XPVMG))
984 #define del_XPVGV(p) safefree(p)
988 #define new_XNV() new_body_allocated(SVt_NV)
989 #define new_XPVNV() new_body_allocated(SVt_PVNV)
990 #define new_XPVMG() new_body_allocated(SVt_PVMG)
992 #define del_XPVGV(p) del_body(p + bodies_by_type[SVt_PVGV].offset, \
993 &PL_body_roots[SVt_PVGV])
997 /* no arena for you! */
999 #define new_NOARENA(details) \
1000 safemalloc((details)->body_size + (details)->offset)
1001 #define new_NOARENAZ(details) \
1002 safecalloc((details)->body_size + (details)->offset, 1)
1005 Perl_more_bodies (pTHX_ const svtype sv_type, const size_t body_size,
1006 const size_t arena_size)
1009 void ** const root = &PL_body_roots[sv_type];
1010 struct arena_desc *adesc;
1011 struct arena_set *aroot = (struct arena_set *) PL_body_arenas;
1015 const size_t good_arena_size = Perl_malloc_good_size(arena_size);
1016 #if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT_PRIVATE)
1017 static bool done_sanity_check;
1019 /* PERL_GLOBAL_STRUCT_PRIVATE cannot coexist with global
1020 * variables like done_sanity_check. */
1021 if (!done_sanity_check) {
1022 unsigned int i = SVt_LAST;
1024 done_sanity_check = TRUE;
1027 assert (bodies_by_type[i].type == i);
1033 /* may need new arena-set to hold new arena */
1034 if (!aroot || aroot->curr >= aroot->set_size) {
1035 struct arena_set *newroot;
1036 Newxz(newroot, 1, struct arena_set);
1037 newroot->set_size = ARENAS_PER_SET;
1038 newroot->next = aroot;
1040 PL_body_arenas = (void *) newroot;
1041 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot));
1044 /* ok, now have arena-set with at least 1 empty/available arena-desc */
1045 curr = aroot->curr++;
1046 adesc = &(aroot->set[curr]);
1047 assert(!adesc->arena);
1049 Newx(adesc->arena, good_arena_size, char);
1050 adesc->size = good_arena_size;
1051 adesc->utype = sv_type;
1052 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %"UVuf"\n",
1053 curr, (void*)adesc->arena, (UV)good_arena_size));
1055 start = (char *) adesc->arena;
1057 /* Get the address of the byte after the end of the last body we can fit.
1058 Remember, this is integer division: */
1059 end = start + good_arena_size / body_size * body_size;
1061 /* computed count doesn't reflect the 1st slot reservation */
1062 #if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE)
1063 DEBUG_m(PerlIO_printf(Perl_debug_log,
1064 "arena %p end %p arena-size %d (from %d) type %d "
1066 (void*)start, (void*)end, (int)good_arena_size,
1067 (int)arena_size, sv_type, (int)body_size,
1068 (int)good_arena_size / (int)body_size));
1070 DEBUG_m(PerlIO_printf(Perl_debug_log,
1071 "arena %p end %p arena-size %d type %d size %d ct %d\n",
1072 (void*)start, (void*)end,
1073 (int)arena_size, sv_type, (int)body_size,
1074 (int)good_arena_size / (int)body_size));
1076 *root = (void *)start;
1079 /* Where the next body would start: */
1080 char * const next = start + body_size;
1083 /* This is the last body: */
1084 assert(next == end);
1086 *(void **)start = 0;
1090 *(void**) start = (void *)next;
1095 /* grab a new thing from the free list, allocating more if necessary.
1096 The inline version is used for speed in hot routines, and the
1097 function using it serves the rest (unless PURIFY).
1099 #define new_body_inline(xpv, sv_type) \
1101 void ** const r3wt = &PL_body_roots[sv_type]; \
1102 xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \
1103 ? *((void **)(r3wt)) : Perl_more_bodies(aTHX_ sv_type, \
1104 bodies_by_type[sv_type].body_size,\
1105 bodies_by_type[sv_type].arena_size)); \
1106 *(r3wt) = *(void**)(xpv); \
1112 S_new_body(pTHX_ const svtype sv_type)
1116 new_body_inline(xpv, sv_type);
1122 static const struct body_details fake_rv =
1123 { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 };
1126 =for apidoc sv_upgrade
1128 Upgrade an SV to a more complex form. Generally adds a new body type to the
1129 SV, then copies across as much information as possible from the old body.
1130 It croaks if the SV is already in a more complex form than requested. You
1131 generally want to use the C<SvUPGRADE> macro wrapper, which checks the type
1132 before calling C<sv_upgrade>, and hence does not croak. See also
1139 Perl_sv_upgrade(pTHX_ register SV *const sv, svtype new_type)
1144 const svtype old_type = SvTYPE(sv);
1145 const struct body_details *new_type_details;
1146 const struct body_details *old_type_details
1147 = bodies_by_type + old_type;
1148 SV *referant = NULL;
1150 PERL_ARGS_ASSERT_SV_UPGRADE;
1152 if (old_type == new_type)
1155 /* This clause was purposefully added ahead of the early return above to
1156 the shared string hackery for (sort {$a <=> $b} keys %hash), with the
1157 inference by Nick I-S that it would fix other troublesome cases. See
1158 changes 7162, 7163 (f130fd4589cf5fbb24149cd4db4137c8326f49c1 and parent)
1160 Given that shared hash key scalars are no longer PVIV, but PV, there is
1161 no longer need to unshare so as to free up the IVX slot for its proper
1162 purpose. So it's safe to move the early return earlier. */
1164 if (new_type != SVt_PV && SvIsCOW(sv)) {
1165 sv_force_normal_flags(sv, 0);
1168 old_body = SvANY(sv);
1170 /* Copying structures onto other structures that have been neatly zeroed
1171 has a subtle gotcha. Consider XPVMG
1173 +------+------+------+------+------+-------+-------+
1174 | NV | CUR | LEN | IV | MAGIC | STASH |
1175 +------+------+------+------+------+-------+-------+
1176 0 4 8 12 16 20 24 28
1178 where NVs are aligned to 8 bytes, so that sizeof that structure is
1179 actually 32 bytes long, with 4 bytes of padding at the end:
1181 +------+------+------+------+------+-------+-------+------+
1182 | NV | CUR | LEN | IV | MAGIC | STASH | ??? |
1183 +------+------+------+------+------+-------+-------+------+
1184 0 4 8 12 16 20 24 28 32
1186 so what happens if you allocate memory for this structure:
1188 +------+------+------+------+------+-------+-------+------+------+...
1189 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME |
1190 +------+------+------+------+------+-------+-------+------+------+...
1191 0 4 8 12 16 20 24 28 32 36
1193 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you
1194 expect, because you copy the area marked ??? onto GP. Now, ??? may have
1195 started out as zero once, but it's quite possible that it isn't. So now,
1196 rather than a nicely zeroed GP, you have it pointing somewhere random.
1199 (In fact, GP ends up pointing at a previous GP structure, because the
1200 principle cause of the padding in XPVMG getting garbage is a copy of
1201 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now
1202 this happens to be moot because XPVGV has been re-ordered, with GP
1203 no longer after STASH)
1205 So we are careful and work out the size of used parts of all the
1213 referant = SvRV(sv);
1214 old_type_details = &fake_rv;
1215 if (new_type == SVt_NV)
1216 new_type = SVt_PVNV;
1218 if (new_type < SVt_PVIV) {
1219 new_type = (new_type == SVt_NV)
1220 ? SVt_PVNV : SVt_PVIV;
1225 if (new_type < SVt_PVNV) {
1226 new_type = SVt_PVNV;
1230 assert(new_type > SVt_PV);
1231 assert(SVt_IV < SVt_PV);
1232 assert(SVt_NV < SVt_PV);
1239 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena,
1240 there's no way that it can be safely upgraded, because perl.c
1241 expects to Safefree(SvANY(PL_mess_sv)) */
1242 assert(sv != PL_mess_sv);
1243 /* This flag bit is used to mean other things in other scalar types.
1244 Given that it only has meaning inside the pad, it shouldn't be set
1245 on anything that can get upgraded. */
1246 assert(!SvPAD_TYPED(sv));
1249 if (old_type_details->cant_upgrade)
1250 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf,
1251 sv_reftype(sv, 0), (UV) old_type, (UV) new_type);
1254 if (old_type > new_type)
1255 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d",
1256 (int)old_type, (int)new_type);
1258 new_type_details = bodies_by_type + new_type;
1260 SvFLAGS(sv) &= ~SVTYPEMASK;
1261 SvFLAGS(sv) |= new_type;
1263 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of
1264 the return statements above will have triggered. */
1265 assert (new_type != SVt_NULL);
1268 assert(old_type == SVt_NULL);
1269 SvANY(sv) = (XPVIV*)((char*)&(sv->sv_u.svu_iv) - STRUCT_OFFSET(XPVIV, xiv_iv));
1273 assert(old_type == SVt_NULL);
1274 SvANY(sv) = new_XNV();
1279 assert(new_type_details->body_size);
1282 assert(new_type_details->arena);
1283 assert(new_type_details->arena_size);
1284 /* This points to the start of the allocated area. */
1285 new_body_inline(new_body, new_type);
1286 Zero(new_body, new_type_details->body_size, char);
1287 new_body = ((char *)new_body) - new_type_details->offset;
1289 /* We always allocated the full length item with PURIFY. To do this
1290 we fake things so that arena is false for all 16 types.. */
1291 new_body = new_NOARENAZ(new_type_details);
1293 SvANY(sv) = new_body;
1294 if (new_type == SVt_PVAV) {
1298 if (old_type_details->body_size) {
1301 /* It will have been zeroed when the new body was allocated.
1302 Lets not write to it, in case it confuses a write-back
1308 #ifndef NODEFAULT_SHAREKEYS
1309 HvSHAREKEYS_on(sv); /* key-sharing on by default */
1311 HvMAX(sv) = 7; /* (start with 8 buckets) */
1314 /* SVt_NULL isn't the only thing upgraded to AV or HV.
1315 The target created by newSVrv also is, and it can have magic.
1316 However, it never has SvPVX set.
1318 if (old_type == SVt_IV) {
1320 } else if (old_type >= SVt_PV) {
1321 assert(SvPVX_const(sv) == 0);
1324 if (old_type >= SVt_PVMG) {
1325 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic);
1326 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash);
1328 sv->sv_u.svu_array = NULL; /* or svu_hash */
1334 /* This ensures that SvTHINKFIRST(sv) is true, and hence that
1335 sv_force_normal_flags(sv) is called. */
1338 /* XXX Is this still needed? Was it ever needed? Surely as there is
1339 no route from NV to PVIV, NOK can never be true */
1340 assert(!SvNOKp(sv));
1351 assert(new_type_details->body_size);
1352 /* We always allocated the full length item with PURIFY. To do this
1353 we fake things so that arena is false for all 16 types.. */
1354 if(new_type_details->arena) {
1355 /* This points to the start of the allocated area. */
1356 new_body_inline(new_body, new_type);
1357 Zero(new_body, new_type_details->body_size, char);
1358 new_body = ((char *)new_body) - new_type_details->offset;
1360 new_body = new_NOARENAZ(new_type_details);
1362 SvANY(sv) = new_body;
1364 if (old_type_details->copy) {
1365 /* There is now the potential for an upgrade from something without
1366 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */
1367 int offset = old_type_details->offset;
1368 int length = old_type_details->copy;
1370 if (new_type_details->offset > old_type_details->offset) {
1371 const int difference
1372 = new_type_details->offset - old_type_details->offset;
1373 offset += difference;
1374 length -= difference;
1376 assert (length >= 0);
1378 Copy((char *)old_body + offset, (char *)new_body + offset, length,
1382 #ifndef NV_ZERO_IS_ALLBITS_ZERO
1383 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a
1384 * correct 0.0 for us. Otherwise, if the old body didn't have an
1385 * NV slot, but the new one does, then we need to initialise the
1386 * freshly created NV slot with whatever the correct bit pattern is
1388 if (old_type_details->zero_nv && !new_type_details->zero_nv
1389 && !isGV_with_GP(sv))
1393 if (new_type == SVt_PVIO) {
1394 IO * const io = MUTABLE_IO(sv);
1395 GV *iogv = gv_fetchpvs("IO::File::", GV_ADD, SVt_PVHV);
1398 /* Clear the stashcache because a new IO could overrule a package
1400 hv_clear(PL_stashcache);
1402 SvSTASH_set(io, MUTABLE_HV(SvREFCNT_inc(GvHV(iogv))));
1403 IoPAGE_LEN(sv) = 60;
1405 if (old_type < SVt_PV) {
1406 /* referant will be NULL unless the old type was SVt_IV emulating
1408 sv->sv_u.svu_rv = referant;
1412 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1413 (unsigned long)new_type);
1416 if (old_type > SVt_IV) {
1420 /* Note that there is an assumption that all bodies of types that
1421 can be upgraded came from arenas. Only the more complex non-
1422 upgradable types are allowed to be directly malloc()ed. */
1423 assert(old_type_details->arena);
1424 del_body((void*)((char*)old_body + old_type_details->offset),
1425 &PL_body_roots[old_type]);
1431 =for apidoc sv_backoff
1433 Remove any string offset. You should normally use the C<SvOOK_off> macro
1440 Perl_sv_backoff(pTHX_ register SV *const sv)
1443 const char * const s = SvPVX_const(sv);
1445 PERL_ARGS_ASSERT_SV_BACKOFF;
1446 PERL_UNUSED_CONTEXT;
1449 assert(SvTYPE(sv) != SVt_PVHV);
1450 assert(SvTYPE(sv) != SVt_PVAV);
1452 SvOOK_offset(sv, delta);
1454 SvLEN_set(sv, SvLEN(sv) + delta);
1455 SvPV_set(sv, SvPVX(sv) - delta);
1456 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
1457 SvFLAGS(sv) &= ~SVf_OOK;
1464 Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1465 upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1466 Use the C<SvGROW> wrapper instead.
1472 Perl_sv_grow(pTHX_ register SV *const sv, register STRLEN newlen)
1476 PERL_ARGS_ASSERT_SV_GROW;
1478 if (PL_madskills && newlen >= 0x100000) {
1479 PerlIO_printf(Perl_debug_log,
1480 "Allocation too large: %"UVxf"\n", (UV)newlen);
1482 #ifdef HAS_64K_LIMIT
1483 if (newlen >= 0x10000) {
1484 PerlIO_printf(Perl_debug_log,
1485 "Allocation too large: %"UVxf"\n", (UV)newlen);
1488 #endif /* HAS_64K_LIMIT */
1491 if (SvTYPE(sv) < SVt_PV) {
1492 sv_upgrade(sv, SVt_PV);
1493 s = SvPVX_mutable(sv);
1495 else if (SvOOK(sv)) { /* pv is offset? */
1497 s = SvPVX_mutable(sv);
1498 if (newlen > SvLEN(sv))
1499 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
1500 #ifdef HAS_64K_LIMIT
1501 if (newlen >= 0x10000)
1506 s = SvPVX_mutable(sv);
1508 if (newlen > SvLEN(sv)) { /* need more room? */
1509 STRLEN minlen = SvCUR(sv);
1510 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
1511 if (newlen < minlen)
1513 #ifndef Perl_safesysmalloc_size
1514 newlen = PERL_STRLEN_ROUNDUP(newlen);
1516 if (SvLEN(sv) && s) {
1517 s = (char*)saferealloc(s, newlen);
1520 s = (char*)safemalloc(newlen);
1521 if (SvPVX_const(sv) && SvCUR(sv)) {
1522 Move(SvPVX_const(sv), s, (newlen < SvCUR(sv)) ? newlen : SvCUR(sv), char);
1526 #ifdef Perl_safesysmalloc_size
1527 /* Do this here, do it once, do it right, and then we will never get
1528 called back into sv_grow() unless there really is some growing
1530 SvLEN_set(sv, Perl_safesysmalloc_size(s));
1532 SvLEN_set(sv, newlen);
1539 =for apidoc sv_setiv
1541 Copies an integer into the given SV, upgrading first if necessary.
1542 Does not handle 'set' magic. See also C<sv_setiv_mg>.
1548 Perl_sv_setiv(pTHX_ register SV *const sv, const IV i)
1552 PERL_ARGS_ASSERT_SV_SETIV;
1554 SV_CHECK_THINKFIRST_COW_DROP(sv);
1555 switch (SvTYPE(sv)) {
1558 sv_upgrade(sv, SVt_IV);
1561 sv_upgrade(sv, SVt_PVIV);
1565 if (!isGV_with_GP(sv))
1572 /* diag_listed_as: Can't coerce %s to %s in %s */
1573 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1577 (void)SvIOK_only(sv); /* validate number */
1583 =for apidoc sv_setiv_mg
1585 Like C<sv_setiv>, but also handles 'set' magic.
1591 Perl_sv_setiv_mg(pTHX_ register SV *const sv, const IV i)
1593 PERL_ARGS_ASSERT_SV_SETIV_MG;
1600 =for apidoc sv_setuv
1602 Copies an unsigned integer into the given SV, upgrading first if necessary.
1603 Does not handle 'set' magic. See also C<sv_setuv_mg>.
1609 Perl_sv_setuv(pTHX_ register SV *const sv, const UV u)
1611 PERL_ARGS_ASSERT_SV_SETUV;
1613 /* With these two if statements:
1614 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
1617 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1619 If you wish to remove them, please benchmark to see what the effect is
1621 if (u <= (UV)IV_MAX) {
1622 sv_setiv(sv, (IV)u);
1631 =for apidoc sv_setuv_mg
1633 Like C<sv_setuv>, but also handles 'set' magic.
1639 Perl_sv_setuv_mg(pTHX_ register SV *const sv, const UV u)
1641 PERL_ARGS_ASSERT_SV_SETUV_MG;
1648 =for apidoc sv_setnv
1650 Copies a double into the given SV, upgrading first if necessary.
1651 Does not handle 'set' magic. See also C<sv_setnv_mg>.
1657 Perl_sv_setnv(pTHX_ register SV *const sv, const NV num)
1661 PERL_ARGS_ASSERT_SV_SETNV;
1663 SV_CHECK_THINKFIRST_COW_DROP(sv);
1664 switch (SvTYPE(sv)) {
1667 sv_upgrade(sv, SVt_NV);
1671 sv_upgrade(sv, SVt_PVNV);
1675 if (!isGV_with_GP(sv))
1682 /* diag_listed_as: Can't coerce %s to %s in %s */
1683 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
1688 (void)SvNOK_only(sv); /* validate number */
1693 =for apidoc sv_setnv_mg
1695 Like C<sv_setnv>, but also handles 'set' magic.
1701 Perl_sv_setnv_mg(pTHX_ register SV *const sv, const NV num)
1703 PERL_ARGS_ASSERT_SV_SETNV_MG;
1709 /* Print an "isn't numeric" warning, using a cleaned-up,
1710 * printable version of the offending string
1714 S_not_a_number(pTHX_ SV *const sv)
1721 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1724 dsv = newSVpvs_flags("", SVs_TEMP);
1725 pv = sv_uni_display(dsv, sv, 10, UNI_DISPLAY_ISPRINT);
1728 const char * const limit = tmpbuf + sizeof(tmpbuf) - 8;
1729 /* each *s can expand to 4 chars + "...\0",
1730 i.e. need room for 8 chars */
1732 const char *s = SvPVX_const(sv);
1733 const char * const end = s + SvCUR(sv);
1734 for ( ; s < end && d < limit; s++ ) {
1736 if (ch & 128 && !isPRINT_LC(ch)) {
1745 else if (ch == '\r') {
1749 else if (ch == '\f') {
1753 else if (ch == '\\') {
1757 else if (ch == '\0') {
1761 else if (isPRINT_LC(ch))
1778 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1779 "Argument \"%s\" isn't numeric in %s", pv,
1782 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1783 "Argument \"%s\" isn't numeric", pv);
1787 =for apidoc looks_like_number
1789 Test if the content of an SV looks like a number (or is a number).
1790 C<Inf> and C<Infinity> are treated as numbers (so will not issue a
1791 non-numeric warning), even if your atof() doesn't grok them. Get-magic is
1798 Perl_looks_like_number(pTHX_ SV *const sv)
1800 register const char *sbegin;
1803 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1805 if (SvPOK(sv) || SvPOKp(sv)) {
1806 sbegin = SvPV_nomg_const(sv, len);
1809 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
1810 return grok_number(sbegin, len, NULL);
1814 S_glob_2number(pTHX_ GV * const gv)
1816 SV *const buffer = sv_newmortal();
1818 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1820 gv_efullname3(buffer, gv, "*");
1822 /* We know that all GVs stringify to something that is not-a-number,
1823 so no need to test that. */
1824 if (ckWARN(WARN_NUMERIC))
1825 not_a_number(buffer);
1826 /* We just want something true to return, so that S_sv_2iuv_common
1827 can tail call us and return true. */
1831 /* Actually, ISO C leaves conversion of UV to IV undefined, but
1832 until proven guilty, assume that things are not that bad... */
1837 As 64 bit platforms often have an NV that doesn't preserve all bits of
1838 an IV (an assumption perl has been based on to date) it becomes necessary
1839 to remove the assumption that the NV always carries enough precision to
1840 recreate the IV whenever needed, and that the NV is the canonical form.
1841 Instead, IV/UV and NV need to be given equal rights. So as to not lose
1842 precision as a side effect of conversion (which would lead to insanity
1843 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
1844 1) to distinguish between IV/UV/NV slots that have cached a valid
1845 conversion where precision was lost and IV/UV/NV slots that have a
1846 valid conversion which has lost no precision
1847 2) to ensure that if a numeric conversion to one form is requested that
1848 would lose precision, the precise conversion (or differently
1849 imprecise conversion) is also performed and cached, to prevent
1850 requests for different numeric formats on the same SV causing
1851 lossy conversion chains. (lossless conversion chains are perfectly
1856 SvIOKp is true if the IV slot contains a valid value
1857 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1858 SvNOKp is true if the NV slot contains a valid value
1859 SvNOK is true only if the NV value is accurate
1862 while converting from PV to NV, check to see if converting that NV to an
1863 IV(or UV) would lose accuracy over a direct conversion from PV to
1864 IV(or UV). If it would, cache both conversions, return NV, but mark
1865 SV as IOK NOKp (ie not NOK).
1867 While converting from PV to IV, check to see if converting that IV to an
1868 NV would lose accuracy over a direct conversion from PV to NV. If it
1869 would, cache both conversions, flag similarly.
1871 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
1872 correctly because if IV & NV were set NV *always* overruled.
1873 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
1874 changes - now IV and NV together means that the two are interchangeable:
1875 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
1877 The benefit of this is that operations such as pp_add know that if
1878 SvIOK is true for both left and right operands, then integer addition
1879 can be used instead of floating point (for cases where the result won't
1880 overflow). Before, floating point was always used, which could lead to
1881 loss of precision compared with integer addition.
1883 * making IV and NV equal status should make maths accurate on 64 bit
1885 * may speed up maths somewhat if pp_add and friends start to use
1886 integers when possible instead of fp. (Hopefully the overhead in
1887 looking for SvIOK and checking for overflow will not outweigh the
1888 fp to integer speedup)
1889 * will slow down integer operations (callers of SvIV) on "inaccurate"
1890 values, as the change from SvIOK to SvIOKp will cause a call into
1891 sv_2iv each time rather than a macro access direct to the IV slot
1892 * should speed up number->string conversion on integers as IV is
1893 favoured when IV and NV are equally accurate
1895 ####################################################################
1896 You had better be using SvIOK_notUV if you want an IV for arithmetic:
1897 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
1898 On the other hand, SvUOK is true iff UV.
1899 ####################################################################
1901 Your mileage will vary depending your CPU's relative fp to integer
1905 #ifndef NV_PRESERVES_UV
1906 # define IS_NUMBER_UNDERFLOW_IV 1
1907 # define IS_NUMBER_UNDERFLOW_UV 2
1908 # define IS_NUMBER_IV_AND_UV 2
1909 # define IS_NUMBER_OVERFLOW_IV 4
1910 # define IS_NUMBER_OVERFLOW_UV 5
1912 /* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
1914 /* For sv_2nv these three cases are "SvNOK and don't bother casting" */
1916 S_sv_2iuv_non_preserve(pTHX_ register SV *const sv
1924 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
1926 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));
1927 if (SvNVX(sv) < (NV)IV_MIN) {
1928 (void)SvIOKp_on(sv);
1930 SvIV_set(sv, IV_MIN);
1931 return IS_NUMBER_UNDERFLOW_IV;
1933 if (SvNVX(sv) > (NV)UV_MAX) {
1934 (void)SvIOKp_on(sv);
1937 SvUV_set(sv, UV_MAX);
1938 return IS_NUMBER_OVERFLOW_UV;
1940 (void)SvIOKp_on(sv);
1942 /* Can't use strtol etc to convert this string. (See truth table in
1944 if (SvNVX(sv) <= (UV)IV_MAX) {
1945 SvIV_set(sv, I_V(SvNVX(sv)));
1946 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
1947 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
1949 /* Integer is imprecise. NOK, IOKp */
1951 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
1954 SvUV_set(sv, U_V(SvNVX(sv)));
1955 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
1956 if (SvUVX(sv) == UV_MAX) {
1957 /* As we know that NVs don't preserve UVs, UV_MAX cannot
1958 possibly be preserved by NV. Hence, it must be overflow.
1960 return IS_NUMBER_OVERFLOW_UV;
1962 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
1964 /* Integer is imprecise. NOK, IOKp */
1966 return IS_NUMBER_OVERFLOW_IV;
1968 #endif /* !NV_PRESERVES_UV*/
1971 S_sv_2iuv_common(pTHX_ SV *const sv)
1975 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
1978 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
1979 * without also getting a cached IV/UV from it at the same time
1980 * (ie PV->NV conversion should detect loss of accuracy and cache
1981 * IV or UV at same time to avoid this. */
1982 /* IV-over-UV optimisation - choose to cache IV if possible */
1984 if (SvTYPE(sv) == SVt_NV)
1985 sv_upgrade(sv, SVt_PVNV);
1987 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
1988 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
1989 certainly cast into the IV range at IV_MAX, whereas the correct
1990 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
1992 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
1993 if (Perl_isnan(SvNVX(sv))) {
1999 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2000 SvIV_set(sv, I_V(SvNVX(sv)));
2001 if (SvNVX(sv) == (NV) SvIVX(sv)
2002 #ifndef NV_PRESERVES_UV
2003 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2004 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2005 /* Don't flag it as "accurately an integer" if the number
2006 came from a (by definition imprecise) NV operation, and
2007 we're outside the range of NV integer precision */
2011 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2013 /* scalar has trailing garbage, eg "42a" */
2015 DEBUG_c(PerlIO_printf(Perl_debug_log,
2016 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (precise)\n",
2022 /* IV not precise. No need to convert from PV, as NV
2023 conversion would already have cached IV if it detected
2024 that PV->IV would be better than PV->NV->IV
2025 flags already correct - don't set public IOK. */
2026 DEBUG_c(PerlIO_printf(Perl_debug_log,
2027 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (imprecise)\n",
2032 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2033 but the cast (NV)IV_MIN rounds to a the value less (more
2034 negative) than IV_MIN which happens to be equal to SvNVX ??
2035 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2036 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2037 (NV)UVX == NVX are both true, but the values differ. :-(
2038 Hopefully for 2s complement IV_MIN is something like
2039 0x8000000000000000 which will be exact. NWC */
2042 SvUV_set(sv, U_V(SvNVX(sv)));
2044 (SvNVX(sv) == (NV) SvUVX(sv))
2045 #ifndef NV_PRESERVES_UV
2046 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2047 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2048 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2049 /* Don't flag it as "accurately an integer" if the number
2050 came from a (by definition imprecise) NV operation, and
2051 we're outside the range of NV integer precision */
2057 DEBUG_c(PerlIO_printf(Perl_debug_log,
2058 "0x%"UVxf" 2iv(%"UVuf" => %"IVdf") (as unsigned)\n",
2064 else if (SvPOKp(sv) && SvLEN(sv)) {
2066 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2067 /* We want to avoid a possible problem when we cache an IV/ a UV which
2068 may be later translated to an NV, and the resulting NV is not
2069 the same as the direct translation of the initial string
2070 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2071 be careful to ensure that the value with the .456 is around if the
2072 NV value is requested in the future).
2074 This means that if we cache such an IV/a UV, we need to cache the
2075 NV as well. Moreover, we trade speed for space, and do not
2076 cache the NV if we are sure it's not needed.
2079 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2080 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2081 == IS_NUMBER_IN_UV) {
2082 /* It's definitely an integer, only upgrade to PVIV */
2083 if (SvTYPE(sv) < SVt_PVIV)
2084 sv_upgrade(sv, SVt_PVIV);
2086 } else if (SvTYPE(sv) < SVt_PVNV)
2087 sv_upgrade(sv, SVt_PVNV);
2089 /* If NVs preserve UVs then we only use the UV value if we know that
2090 we aren't going to call atof() below. If NVs don't preserve UVs
2091 then the value returned may have more precision than atof() will
2092 return, even though value isn't perfectly accurate. */
2093 if ((numtype & (IS_NUMBER_IN_UV
2094 #ifdef NV_PRESERVES_UV
2097 )) == IS_NUMBER_IN_UV) {
2098 /* This won't turn off the public IOK flag if it was set above */
2099 (void)SvIOKp_on(sv);
2101 if (!(numtype & IS_NUMBER_NEG)) {
2103 if (value <= (UV)IV_MAX) {
2104 SvIV_set(sv, (IV)value);
2106 /* it didn't overflow, and it was positive. */
2107 SvUV_set(sv, value);
2111 /* 2s complement assumption */
2112 if (value <= (UV)IV_MIN) {
2113 SvIV_set(sv, -(IV)value);
2115 /* Too negative for an IV. This is a double upgrade, but
2116 I'm assuming it will be rare. */
2117 if (SvTYPE(sv) < SVt_PVNV)
2118 sv_upgrade(sv, SVt_PVNV);
2122 SvNV_set(sv, -(NV)value);
2123 SvIV_set(sv, IV_MIN);
2127 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2128 will be in the previous block to set the IV slot, and the next
2129 block to set the NV slot. So no else here. */
2131 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2132 != IS_NUMBER_IN_UV) {
2133 /* It wasn't an (integer that doesn't overflow the UV). */
2134 SvNV_set(sv, Atof(SvPVX_const(sv)));
2136 if (! numtype && ckWARN(WARN_NUMERIC))
2139 #if defined(USE_LONG_DOUBLE)
2140 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%" PERL_PRIgldbl ")\n",
2141 PTR2UV(sv), SvNVX(sv)));
2143 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"NVgf")\n",
2144 PTR2UV(sv), SvNVX(sv)));
2147 #ifdef NV_PRESERVES_UV
2148 (void)SvIOKp_on(sv);
2150 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2151 SvIV_set(sv, I_V(SvNVX(sv)));
2152 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2155 NOOP; /* Integer is imprecise. NOK, IOKp */
2157 /* UV will not work better than IV */
2159 if (SvNVX(sv) > (NV)UV_MAX) {
2161 /* Integer is inaccurate. NOK, IOKp, is UV */
2162 SvUV_set(sv, UV_MAX);
2164 SvUV_set(sv, U_V(SvNVX(sv)));
2165 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2166 NV preservse UV so can do correct comparison. */
2167 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2170 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
2175 #else /* NV_PRESERVES_UV */
2176 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2177 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
2178 /* The IV/UV slot will have been set from value returned by
2179 grok_number above. The NV slot has just been set using
2182 assert (SvIOKp(sv));
2184 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2185 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2186 /* Small enough to preserve all bits. */
2187 (void)SvIOKp_on(sv);
2189 SvIV_set(sv, I_V(SvNVX(sv)));
2190 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2192 /* Assumption: first non-preserved integer is < IV_MAX,
2193 this NV is in the preserved range, therefore: */
2194 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2196 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);
2200 0 0 already failed to read UV.
2201 0 1 already failed to read UV.
2202 1 0 you won't get here in this case. IV/UV
2203 slot set, public IOK, Atof() unneeded.
2204 1 1 already read UV.
2205 so there's no point in sv_2iuv_non_preserve() attempting
2206 to use atol, strtol, strtoul etc. */
2208 sv_2iuv_non_preserve (sv, numtype);
2210 sv_2iuv_non_preserve (sv);
2214 #endif /* NV_PRESERVES_UV */
2215 /* It might be more code efficient to go through the entire logic above
2216 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2217 gets complex and potentially buggy, so more programmer efficient
2218 to do it this way, by turning off the public flags: */
2220 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2224 if (isGV_with_GP(sv))
2225 return glob_2number(MUTABLE_GV(sv));
2227 if (!SvPADTMP(sv)) {
2228 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2231 if (SvTYPE(sv) < SVt_IV)
2232 /* Typically the caller expects that sv_any is not NULL now. */
2233 sv_upgrade(sv, SVt_IV);
2234 /* Return 0 from the caller. */
2241 =for apidoc sv_2iv_flags
2243 Return the integer value of an SV, doing any necessary string
2244 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2245 Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2251 Perl_sv_2iv_flags(pTHX_ register SV *const sv, const I32 flags)
2256 if (SvGMAGICAL(sv) || SvVALID(sv)) {
2257 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2258 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2259 In practice they are extremely unlikely to actually get anywhere
2260 accessible by user Perl code - the only way that I'm aware of is when
2261 a constant subroutine which is used as the second argument to index.
2263 if (flags & SV_GMAGIC)
2268 return I_V(SvNVX(sv));
2270 if (SvPOKp(sv) && SvLEN(sv)) {
2273 = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2275 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2276 == IS_NUMBER_IN_UV) {
2277 /* It's definitely an integer */
2278 if (numtype & IS_NUMBER_NEG) {
2279 if (value < (UV)IV_MIN)
2282 if (value < (UV)IV_MAX)
2287 if (ckWARN(WARN_NUMERIC))
2290 return I_V(Atof(SvPVX_const(sv)));
2295 assert(SvTYPE(sv) >= SVt_PVMG);
2296 /* This falls through to the report_uninit inside S_sv_2iuv_common. */
2297 } else if (SvTHINKFIRST(sv)) {
2302 if (flags & SV_SKIP_OVERLOAD)
2304 tmpstr = AMG_CALLunary(sv, numer_amg);
2305 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2306 return SvIV(tmpstr);
2309 return PTR2IV(SvRV(sv));
2312 sv_force_normal_flags(sv, 0);
2314 if (SvREADONLY(sv) && !SvOK(sv)) {
2315 if (ckWARN(WARN_UNINITIALIZED))
2321 if (S_sv_2iuv_common(aTHX_ sv))
2324 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"IVdf")\n",
2325 PTR2UV(sv),SvIVX(sv)));
2326 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
2330 =for apidoc sv_2uv_flags
2332 Return the unsigned integer value of an SV, doing any necessary string
2333 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2334 Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
2340 Perl_sv_2uv_flags(pTHX_ register SV *const sv, const I32 flags)
2345 if (SvGMAGICAL(sv) || SvVALID(sv)) {
2346 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2347 the same flag bit as SVf_IVisUV, so must not let them cache IVs. */
2348 if (flags & SV_GMAGIC)
2353 return U_V(SvNVX(sv));
2354 if (SvPOKp(sv) && SvLEN(sv)) {
2357 = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2359 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2360 == IS_NUMBER_IN_UV) {
2361 /* It's definitely an integer */
2362 if (!(numtype & IS_NUMBER_NEG))
2366 if (ckWARN(WARN_NUMERIC))
2369 return U_V(Atof(SvPVX_const(sv)));
2374 assert(SvTYPE(sv) >= SVt_PVMG);
2375 /* This falls through to the report_uninit inside S_sv_2iuv_common. */
2376 } else if (SvTHINKFIRST(sv)) {
2381 if (flags & SV_SKIP_OVERLOAD)
2383 tmpstr = AMG_CALLunary(sv, numer_amg);
2384 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2385 return SvUV(tmpstr);
2388 return PTR2UV(SvRV(sv));
2391 sv_force_normal_flags(sv, 0);
2393 if (SvREADONLY(sv) && !SvOK(sv)) {
2394 if (ckWARN(WARN_UNINITIALIZED))
2400 if (S_sv_2iuv_common(aTHX_ sv))
2404 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2uv(%"UVuf")\n",
2405 PTR2UV(sv),SvUVX(sv)));
2406 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
2410 =for apidoc sv_2nv_flags
2412 Return the num value of an SV, doing any necessary string or integer
2413 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2414 Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
2420 Perl_sv_2nv_flags(pTHX_ register SV *const sv, const I32 flags)
2425 if (SvGMAGICAL(sv) || SvVALID(sv)) {
2426 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2427 the same flag bit as SVf_IVisUV, so must not let them cache NVs. */
2428 if (flags & SV_GMAGIC)
2432 if ((SvPOKp(sv) && SvLEN(sv)) && !SvIOKp(sv)) {
2433 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
2434 !grok_number(SvPVX_const(sv), SvCUR(sv), NULL))
2436 return Atof(SvPVX_const(sv));
2440 return (NV)SvUVX(sv);
2442 return (NV)SvIVX(sv);
2447 assert(SvTYPE(sv) >= SVt_PVMG);
2448 /* This falls through to the report_uninit near the end of the
2450 } else if (SvTHINKFIRST(sv)) {
2455 if (flags & SV_SKIP_OVERLOAD)
2457 tmpstr = AMG_CALLunary(sv, numer_amg);
2458 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2459 return SvNV(tmpstr);
2462 return PTR2NV(SvRV(sv));
2465 sv_force_normal_flags(sv, 0);
2467 if (SvREADONLY(sv) && !SvOK(sv)) {
2468 if (ckWARN(WARN_UNINITIALIZED))
2473 if (SvTYPE(sv) < SVt_NV) {
2474 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2475 sv_upgrade(sv, SVt_NV);
2476 #ifdef USE_LONG_DOUBLE
2478 STORE_NUMERIC_LOCAL_SET_STANDARD();
2479 PerlIO_printf(Perl_debug_log,
2480 "0x%"UVxf" num(%" PERL_PRIgldbl ")\n",
2481 PTR2UV(sv), SvNVX(sv));
2482 RESTORE_NUMERIC_LOCAL();
2486 STORE_NUMERIC_LOCAL_SET_STANDARD();
2487 PerlIO_printf(Perl_debug_log, "0x%"UVxf" num(%"NVgf")\n",
2488 PTR2UV(sv), SvNVX(sv));
2489 RESTORE_NUMERIC_LOCAL();
2493 else if (SvTYPE(sv) < SVt_PVNV)
2494 sv_upgrade(sv, SVt_PVNV);
2499 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
2500 #ifdef NV_PRESERVES_UV
2506 /* Only set the public NV OK flag if this NV preserves the IV */
2507 /* Check it's not 0xFFFFFFFFFFFFFFFF */
2509 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
2510 : (SvIVX(sv) == I_V(SvNVX(sv))))
2516 else if (SvPOKp(sv) && SvLEN(sv)) {
2518 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2519 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
2521 #ifdef NV_PRESERVES_UV
2522 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2523 == IS_NUMBER_IN_UV) {
2524 /* It's definitely an integer */
2525 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
2527 SvNV_set(sv, Atof(SvPVX_const(sv)));
2533 SvNV_set(sv, Atof(SvPVX_const(sv)));
2534 /* Only set the public NV OK flag if this NV preserves the value in
2535 the PV at least as well as an IV/UV would.
2536 Not sure how to do this 100% reliably. */
2537 /* if that shift count is out of range then Configure's test is
2538 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2540 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2541 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2542 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
2543 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2544 /* Can't use strtol etc to convert this string, so don't try.
2545 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2548 /* value has been set. It may not be precise. */
2549 if ((numtype & IS_NUMBER_NEG) && (value > (UV)IV_MIN)) {
2550 /* 2s complement assumption for (UV)IV_MIN */
2551 SvNOK_on(sv); /* Integer is too negative. */
2556 if (numtype & IS_NUMBER_NEG) {
2557 SvIV_set(sv, -(IV)value);
2558 } else if (value <= (UV)IV_MAX) {
2559 SvIV_set(sv, (IV)value);
2561 SvUV_set(sv, value);
2565 if (numtype & IS_NUMBER_NOT_INT) {
2566 /* I believe that even if the original PV had decimals,
2567 they are lost beyond the limit of the FP precision.
2568 However, neither is canonical, so both only get p
2569 flags. NWC, 2000/11/25 */
2570 /* Both already have p flags, so do nothing */
2572 const NV nv = SvNVX(sv);
2573 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2574 if (SvIVX(sv) == I_V(nv)) {
2577 /* It had no "." so it must be integer. */
2581 /* between IV_MAX and NV(UV_MAX).
2582 Could be slightly > UV_MAX */
2584 if (numtype & IS_NUMBER_NOT_INT) {
2585 /* UV and NV both imprecise. */
2587 const UV nv_as_uv = U_V(nv);
2589 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2598 /* It might be more code efficient to go through the entire logic above
2599 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2600 gets complex and potentially buggy, so more programmer efficient
2601 to do it this way, by turning off the public flags: */
2603 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2604 #endif /* NV_PRESERVES_UV */
2607 if (isGV_with_GP(sv)) {
2608 glob_2number(MUTABLE_GV(sv));
2612 if (!PL_localizing && !SvPADTMP(sv) && ckWARN(WARN_UNINITIALIZED))
2614 assert (SvTYPE(sv) >= SVt_NV);
2615 /* Typically the caller expects that sv_any is not NULL now. */
2616 /* XXX Ilya implies that this is a bug in callers that assume this
2617 and ideally should be fixed. */
2620 #if defined(USE_LONG_DOUBLE)
2622 STORE_NUMERIC_LOCAL_SET_STANDARD();
2623 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2nv(%" PERL_PRIgldbl ")\n",
2624 PTR2UV(sv), SvNVX(sv));
2625 RESTORE_NUMERIC_LOCAL();
2629 STORE_NUMERIC_LOCAL_SET_STANDARD();
2630 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 1nv(%"NVgf")\n",
2631 PTR2UV(sv), SvNVX(sv));
2632 RESTORE_NUMERIC_LOCAL();
2641 Return an SV with the numeric value of the source SV, doing any necessary
2642 reference or overload conversion. You must use the C<SvNUM(sv)> macro to
2643 access this function.
2649 Perl_sv_2num(pTHX_ register SV *const sv)
2651 PERL_ARGS_ASSERT_SV_2NUM;
2656 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
2657 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
2658 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2659 return sv_2num(tmpsv);
2661 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2664 /* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2665 * UV as a string towards the end of buf, and return pointers to start and
2668 * We assume that buf is at least TYPE_CHARS(UV) long.
2672 S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
2674 char *ptr = buf + TYPE_CHARS(UV);
2675 char * const ebuf = ptr;
2678 PERL_ARGS_ASSERT_UIV_2BUF;
2690 *--ptr = '0' + (char)(uv % 10);
2699 =for apidoc sv_2pv_flags
2701 Returns a pointer to the string value of an SV, and sets *lp to its length.
2702 If flags includes SV_GMAGIC, does an mg_get() first. Coerces sv to a string
2704 Normally invoked via the C<SvPV_flags> macro. C<sv_2pv()> and C<sv_2pv_nomg>
2705 usually end up here too.
2711 Perl_sv_2pv_flags(pTHX_ register SV *const sv, STRLEN *const lp, const I32 flags)
2721 if (SvGMAGICAL(sv)) {
2722 if (flags & SV_GMAGIC)
2727 if (flags & SV_MUTABLE_RETURN)
2728 return SvPVX_mutable(sv);
2729 if (flags & SV_CONST_RETURN)
2730 return (char *)SvPVX_const(sv);
2733 if (SvIOKp(sv) || SvNOKp(sv)) {
2734 char tbuf[64]; /* Must fit sprintf/Gconvert of longest IV/NV */
2739 ? my_snprintf(tbuf, sizeof(tbuf), "%"UVuf, (UV)SvUVX(sv))
2740 : my_snprintf(tbuf, sizeof(tbuf), "%"IVdf, (IV)SvIVX(sv));
2741 } else if(SvNVX(sv) == 0.0) {
2746 Gconvert(SvNVX(sv), NV_DIG, 0, tbuf);
2753 SvUPGRADE(sv, SVt_PV);
2756 s = SvGROW_mutable(sv, len + 1);
2759 return (char*)memcpy(s, tbuf, len + 1);
2765 assert(SvTYPE(sv) >= SVt_PVMG);
2766 /* This falls through to the report_uninit near the end of the
2768 } else if (SvTHINKFIRST(sv)) {
2773 if (flags & SV_SKIP_OVERLOAD)
2775 tmpstr = AMG_CALLunary(sv, string_amg);
2776 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
2777 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2779 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2783 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2784 if (flags & SV_CONST_RETURN) {
2785 pv = (char *) SvPVX_const(tmpstr);
2787 pv = (flags & SV_MUTABLE_RETURN)
2788 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
2791 *lp = SvCUR(tmpstr);
2793 pv = sv_2pv_flags(tmpstr, lp, flags);
2806 SV *const referent = SvRV(sv);
2810 retval = buffer = savepvn("NULLREF", len);
2811 } else if (SvTYPE(referent) == SVt_REGEXP) {
2812 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
2817 /* If the regex is UTF-8 we want the containing scalar to
2818 have an UTF-8 flag too */
2824 if ((seen_evals = RX_SEEN_EVALS(re)))
2825 PL_reginterp_cnt += seen_evals;
2828 *lp = RX_WRAPLEN(re);
2830 return RX_WRAPPED(re);
2832 const char *const typestr = sv_reftype(referent, 0);
2833 const STRLEN typelen = strlen(typestr);
2834 UV addr = PTR2UV(referent);
2835 const char *stashname = NULL;
2836 STRLEN stashnamelen = 0; /* hush, gcc */
2837 const char *buffer_end;
2839 if (SvOBJECT(referent)) {
2840 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
2843 stashname = HEK_KEY(name);
2844 stashnamelen = HEK_LEN(name);
2846 if (HEK_UTF8(name)) {
2852 stashname = "__ANON__";
2855 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
2856 + 2 * sizeof(UV) + 2 /* )\0 */;
2858 len = typelen + 3 /* (0x */
2859 + 2 * sizeof(UV) + 2 /* )\0 */;
2862 Newx(buffer, len, char);
2863 buffer_end = retval = buffer + len;
2865 /* Working backwards */
2869 *--retval = PL_hexdigit[addr & 15];
2870 } while (addr >>= 4);
2876 memcpy(retval, typestr, typelen);
2880 retval -= stashnamelen;
2881 memcpy(retval, stashname, stashnamelen);
2883 /* retval may not necessarily have reached the start of the
2885 assert (retval >= buffer);
2887 len = buffer_end - retval - 1; /* -1 for that \0 */
2895 if (SvREADONLY(sv) && !SvOK(sv)) {
2898 if (flags & SV_UNDEF_RETURNS_NULL)
2900 if (ckWARN(WARN_UNINITIALIZED))
2905 if (SvIOK(sv) || ((SvIOKp(sv) && !SvNOKp(sv)))) {
2906 /* I'm assuming that if both IV and NV are equally valid then
2907 converting the IV is going to be more efficient */
2908 const U32 isUIOK = SvIsUV(sv);
2909 char buf[TYPE_CHARS(UV)];
2913 if (SvTYPE(sv) < SVt_PVIV)
2914 sv_upgrade(sv, SVt_PVIV);
2915 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
2917 /* inlined from sv_setpvn */
2918 s = SvGROW_mutable(sv, len + 1);
2919 Move(ptr, s, len, char);
2923 else if (SvNOKp(sv)) {
2924 if (SvTYPE(sv) < SVt_PVNV)
2925 sv_upgrade(sv, SVt_PVNV);
2926 if (SvNVX(sv) == 0.0) {
2927 s = SvGROW_mutable(sv, 2);
2932 /* The +20 is pure guesswork. Configure test needed. --jhi */
2933 s = SvGROW_mutable(sv, NV_DIG + 20);
2934 /* some Xenix systems wipe out errno here */
2935 Gconvert(SvNVX(sv), NV_DIG, 0, s);
2945 if (isGV_with_GP(sv)) {
2946 GV *const gv = MUTABLE_GV(sv);
2947 SV *const buffer = sv_newmortal();
2949 gv_efullname3(buffer, gv, "*");
2951 assert(SvPOK(buffer));
2953 *lp = SvCUR(buffer);
2955 if ( SvUTF8(buffer) ) SvUTF8_on(sv);
2956 return SvPVX(buffer);
2961 if (flags & SV_UNDEF_RETURNS_NULL)
2963 if (!PL_localizing && !SvPADTMP(sv) && ckWARN(WARN_UNINITIALIZED))
2965 if (SvTYPE(sv) < SVt_PV)
2966 /* Typically the caller expects that sv_any is not NULL now. */
2967 sv_upgrade(sv, SVt_PV);
2971 const STRLEN len = s - SvPVX_const(sv);
2977 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
2978 PTR2UV(sv),SvPVX_const(sv)));
2979 if (flags & SV_CONST_RETURN)
2980 return (char *)SvPVX_const(sv);
2981 if (flags & SV_MUTABLE_RETURN)
2982 return SvPVX_mutable(sv);
2987 =for apidoc sv_copypv
2989 Copies a stringified representation of the source SV into the
2990 destination SV. Automatically performs any necessary mg_get and
2991 coercion of numeric values into strings. Guaranteed to preserve
2992 UTF8 flag even from overloaded objects. Similar in nature to
2993 sv_2pv[_flags] but operates directly on an SV instead of just the
2994 string. Mostly uses sv_2pv_flags to do its work, except when that
2995 would lose the UTF-8'ness of the PV.
3001 Perl_sv_copypv(pTHX_ SV *const dsv, register SV *const ssv)
3004 const char * const s = SvPV_const(ssv,len);
3006 PERL_ARGS_ASSERT_SV_COPYPV;
3008 sv_setpvn(dsv,s,len);
3016 =for apidoc sv_2pvbyte
3018 Return a pointer to the byte-encoded representation of the SV, and set *lp
3019 to its length. May cause the SV to be downgraded from UTF-8 as a
3022 Usually accessed via the C<SvPVbyte> macro.
3028 Perl_sv_2pvbyte(pTHX_ register SV *const sv, STRLEN *const lp)
3030 PERL_ARGS_ASSERT_SV_2PVBYTE;
3033 sv_utf8_downgrade(sv,0);
3034 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3038 =for apidoc sv_2pvutf8
3040 Return a pointer to the UTF-8-encoded representation of the SV, and set *lp
3041 to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3043 Usually accessed via the C<SvPVutf8> macro.
3049 Perl_sv_2pvutf8(pTHX_ register SV *const sv, STRLEN *const lp)
3051 PERL_ARGS_ASSERT_SV_2PVUTF8;
3053 sv_utf8_upgrade(sv);
3054 return lp ? SvPV(sv,*lp) : SvPV_nolen(sv);
3059 =for apidoc sv_2bool
3061 This macro is only used by sv_true() or its macro equivalent, and only if
3062 the latter's argument is neither SvPOK, SvIOK nor SvNOK.
3063 It calls sv_2bool_flags with the SV_GMAGIC flag.
3065 =for apidoc sv_2bool_flags
3067 This function is only used by sv_true() and friends, and only if
3068 the latter's argument is neither SvPOK, SvIOK nor SvNOK. If the flags
3069 contain SV_GMAGIC, then it does an mg_get() first.
3076 Perl_sv_2bool_flags(pTHX_ register SV *const sv, const I32 flags)
3080 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
3082 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
3088 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
3089 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
3090 return cBOOL(SvTRUE(tmpsv));
3092 return SvRV(sv) != 0;
3095 register XPV* const Xpvtmp = (XPV*)SvANY(sv);
3097 (*sv->sv_u.svu_pv > '0' ||
3098 Xpvtmp->xpv_cur > 1 ||
3099 (Xpvtmp->xpv_cur && *sv->sv_u.svu_pv != '0')))
3106 return SvIVX(sv) != 0;
3109 return SvNVX(sv) != 0.0;
3111 if (isGV_with_GP(sv))
3121 =for apidoc sv_utf8_upgrade
3123 Converts the PV of an SV to its UTF-8-encoded form.
3124 Forces the SV to string form if it is not already.
3125 Will C<mg_get> on C<sv> if appropriate.
3126 Always sets the SvUTF8 flag to avoid future validity checks even
3127 if the whole string is the same in UTF-8 as not.
3128 Returns the number of bytes in the converted string
3130 This is not as a general purpose byte encoding to Unicode interface:
3131 use the Encode extension for that.
3133 =for apidoc sv_utf8_upgrade_nomg
3135 Like sv_utf8_upgrade, but doesn't do magic on C<sv>
3137 =for apidoc sv_utf8_upgrade_flags
3139 Converts the PV of an SV to its UTF-8-encoded form.
3140 Forces the SV to string form if it is not already.
3141 Always sets the SvUTF8 flag to avoid future validity checks even
3142 if all the bytes are invariant in UTF-8. If C<flags> has C<SV_GMAGIC> bit set,
3143 will C<mg_get> on C<sv> if appropriate, else not.
3144 Returns the number of bytes in the converted string
3145 C<sv_utf8_upgrade> and
3146 C<sv_utf8_upgrade_nomg> are implemented in terms of this function.
3148 This is not as a general purpose byte encoding to Unicode interface:
3149 use the Encode extension for that.
3153 The grow version is currently not externally documented. It adds a parameter,
3154 extra, which is the number of unused bytes the string of 'sv' is guaranteed to
3155 have free after it upon return. This allows the caller to reserve extra space
3156 that it intends to fill, to avoid extra grows.
3158 Also externally undocumented for the moment is the flag SV_FORCE_UTF8_UPGRADE,
3159 which can be used to tell this function to not first check to see if there are
3160 any characters that are different in UTF-8 (variant characters) which would
3161 force it to allocate a new string to sv, but to assume there are. Typically
3162 this flag is used by a routine that has already parsed the string to find that
3163 there are such characters, and passes this information on so that the work
3164 doesn't have to be repeated.
3166 (One might think that the calling routine could pass in the position of the
3167 first such variant, so it wouldn't have to be found again. But that is not the
3168 case, because typically when the caller is likely to use this flag, it won't be
3169 calling this routine unless it finds something that won't fit into a byte.
3170 Otherwise it tries to not upgrade and just use bytes. But some things that
3171 do fit into a byte are variants in utf8, and the caller may not have been
3172 keeping track of these.)
3174 If the routine itself changes the string, it adds a trailing NUL. Such a NUL
3175 isn't guaranteed due to having other routines do the work in some input cases,
3176 or if the input is already flagged as being in utf8.
3178 The speed of this could perhaps be improved for many cases if someone wanted to
3179 write a fast function that counts the number of variant characters in a string,
3180 especially if it could return the position of the first one.
3185 Perl_sv_utf8_upgrade_flags_grow(pTHX_ register SV *const sv, const I32 flags, STRLEN extra)
3189 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
3191 if (sv == &PL_sv_undef)
3195 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3196 (void) sv_2pv_flags(sv,&len, flags);
3198 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3202 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
3207 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3212 sv_force_normal_flags(sv, 0);
3215 if (PL_encoding && !(flags & SV_UTF8_NO_ENCODING)) {
3216 sv_recode_to_utf8(sv, PL_encoding);
3217 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3221 if (SvCUR(sv) == 0) {
3222 if (extra) SvGROW(sv, extra);
3223 } else { /* Assume Latin-1/EBCDIC */
3224 /* This function could be much more efficient if we
3225 * had a FLAG in SVs to signal if there are any variant
3226 * chars in the PV. Given that there isn't such a flag
3227 * make the loop as fast as possible (although there are certainly ways
3228 * to speed this up, eg. through vectorization) */
3229 U8 * s = (U8 *) SvPVX_const(sv);
3230 U8 * e = (U8 *) SvEND(sv);
3232 STRLEN two_byte_count = 0;
3234 if (flags & SV_FORCE_UTF8_UPGRADE) goto must_be_utf8;
3236 /* See if really will need to convert to utf8. We mustn't rely on our
3237 * incoming SV being well formed and having a trailing '\0', as certain
3238 * code in pp_formline can send us partially built SVs. */
3242 if (NATIVE_IS_INVARIANT(ch)) continue;
3244 t--; /* t already incremented; re-point to first variant */
3249 /* utf8 conversion not needed because all are invariants. Mark as
3250 * UTF-8 even if no variant - saves scanning loop */
3252 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3257 /* Here, the string should be converted to utf8, either because of an
3258 * input flag (two_byte_count = 0), or because a character that
3259 * requires 2 bytes was found (two_byte_count = 1). t points either to
3260 * the beginning of the string (if we didn't examine anything), or to
3261 * the first variant. In either case, everything from s to t - 1 will
3262 * occupy only 1 byte each on output.
3264 * There are two main ways to convert. One is to create a new string
3265 * and go through the input starting from the beginning, appending each
3266 * converted value onto the new string as we go along. It's probably
3267 * best to allocate enough space in the string for the worst possible
3268 * case rather than possibly running out of space and having to
3269 * reallocate and then copy what we've done so far. Since everything
3270 * from s to t - 1 is invariant, the destination can be initialized
3271 * with these using a fast memory copy
3273 * The other way is to figure out exactly how big the string should be
3274 * by parsing the entire input. Then you don't have to make it big
3275 * enough to handle the worst possible case, and more importantly, if
3276 * the string you already have is large enough, you don't have to
3277 * allocate a new string, you can copy the last character in the input
3278 * string to the final position(s) that will be occupied by the
3279 * converted string and go backwards, stopping at t, since everything
3280 * before that is invariant.
3282 * There are advantages and disadvantages to each method.
3284 * In the first method, we can allocate a new string, do the memory
3285 * copy from the s to t - 1, and then proceed through the rest of the
3286 * string byte-by-byte.
3288 * In the second method, we proceed through the rest of the input
3289 * string just calculating how big the converted string will be. Then
3290 * there are two cases:
3291 * 1) if the string has enough extra space to handle the converted
3292 * value. We go backwards through the string, converting until we
3293 * get to the position we are at now, and then stop. If this
3294 * position is far enough along in the string, this method is
3295 * faster than the other method. If the memory copy were the same
3296 * speed as the byte-by-byte loop, that position would be about
3297 * half-way, as at the half-way mark, parsing to the end and back
3298 * is one complete string's parse, the same amount as starting
3299 * over and going all the way through. Actually, it would be
3300 * somewhat less than half-way, as it's faster to just count bytes
3301 * than to also copy, and we don't have the overhead of allocating
3302 * a new string, changing the scalar to use it, and freeing the
3303 * existing one. But if the memory copy is fast, the break-even
3304 * point is somewhere after half way. The counting loop could be
3305 * sped up by vectorization, etc, to move the break-even point
3306 * further towards the beginning.
3307 * 2) if the string doesn't have enough space to handle the converted
3308 * value. A new string will have to be allocated, and one might
3309 * as well, given that, start from the beginning doing the first
3310 * method. We've spent extra time parsing the string and in
3311 * exchange all we've gotten is that we know precisely how big to
3312 * make the new one. Perl is more optimized for time than space,
3313 * so this case is a loser.
3314 * So what I've decided to do is not use the 2nd method unless it is
3315 * guaranteed that a new string won't have to be allocated, assuming
3316 * the worst case. I also decided not to put any more conditions on it
3317 * than this, for now. It seems likely that, since the worst case is
3318 * twice as big as the unknown portion of the string (plus 1), we won't
3319 * be guaranteed enough space, causing us to go to the first method,
3320 * unless the string is short, or the first variant character is near
3321 * the end of it. In either of these cases, it seems best to use the
3322 * 2nd method. The only circumstance I can think of where this would
3323 * be really slower is if the string had once had much more data in it
3324 * than it does now, but there is still a substantial amount in it */
3327 STRLEN invariant_head = t - s;
3328 STRLEN size = invariant_head + (e - t) * 2 + 1 + extra;
3329 if (SvLEN(sv) < size) {
3331 /* Here, have decided to allocate a new string */
3336 Newx(dst, size, U8);
3338 /* If no known invariants at the beginning of the input string,
3339 * set so starts from there. Otherwise, can use memory copy to
3340 * get up to where we are now, and then start from here */
3342 if (invariant_head <= 0) {
3345 Copy(s, dst, invariant_head, char);
3346 d = dst + invariant_head;
3350 const UV uv = NATIVE8_TO_UNI(*t++);
3351 if (UNI_IS_INVARIANT(uv))
3352 *d++ = (U8)UNI_TO_NATIVE(uv);
3354 *d++ = (U8)UTF8_EIGHT_BIT_HI(uv);
3355 *d++ = (U8)UTF8_EIGHT_BIT_LO(uv);
3359 SvPV_free(sv); /* No longer using pre-existing string */
3360 SvPV_set(sv, (char*)dst);
3361 SvCUR_set(sv, d - dst);
3362 SvLEN_set(sv, size);
3365 /* Here, have decided to get the exact size of the string.
3366 * Currently this happens only when we know that there is
3367 * guaranteed enough space to fit the converted string, so
3368 * don't have to worry about growing. If two_byte_count is 0,
3369 * then t points to the first byte of the string which hasn't
3370 * been examined yet. Otherwise two_byte_count is 1, and t
3371 * points to the first byte in the string that will expand to
3372 * two. Depending on this, start examining at t or 1 after t.
3375 U8 *d = t + two_byte_count;
3378 /* Count up the remaining bytes that expand to two */
3381 const U8 chr = *d++;
3382 if (! NATIVE_IS_INVARIANT(chr)) two_byte_count++;
3385 /* The string will expand by just the number of bytes that
3386 * occupy two positions. But we are one afterwards because of
3387 * the increment just above. This is the place to put the
3388 * trailing NUL, and to set the length before we decrement */
3390 d += two_byte_count;
3391 SvCUR_set(sv, d - s);
3395 /* Having decremented d, it points to the position to put the
3396 * very last byte of the expanded string. Go backwards through
3397 * the string, copying and expanding as we go, stopping when we
3398 * get to the part that is invariant the rest of the way down */
3402 const U8 ch = NATIVE8_TO_UNI(*e--);
3403 if (UNI_IS_INVARIANT(ch)) {
3404 *d-- = UNI_TO_NATIVE(ch);
3406 *d-- = (U8)UTF8_EIGHT_BIT_LO(ch);
3407 *d-- = (U8)UTF8_EIGHT_BIT_HI(ch);
3412 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3413 /* Update pos. We do it at the end rather than during
3414 * the upgrade, to avoid slowing down the common case
3415 * (upgrade without pos) */
3416 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3418 I32 pos = mg->mg_len;
3419 if (pos > 0 && (U32)pos > invariant_head) {
3420 U8 *d = (U8*) SvPVX(sv) + invariant_head;
3421 STRLEN n = (U32)pos - invariant_head;
3423 if (UTF8_IS_START(*d))
3428 mg->mg_len = d - (U8*)SvPVX(sv);
3431 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3432 magic_setutf8(sv,mg); /* clear UTF8 cache */
3437 /* Mark as UTF-8 even if no variant - saves scanning loop */
3443 =for apidoc sv_utf8_downgrade
3445 Attempts to convert the PV of an SV from characters to bytes.
3446 If the PV contains a character that cannot fit
3447 in a byte, this conversion will fail;
3448 in this case, either returns false or, if C<fail_ok> is not
3451 This is not as a general purpose Unicode to byte encoding interface:
3452 use the Encode extension for that.
3458 Perl_sv_utf8_downgrade(pTHX_ register SV *const sv, const bool fail_ok)
3462 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3464 if (SvPOKp(sv) && SvUTF8(sv)) {
3468 int mg_flags = SV_GMAGIC;
3471 sv_force_normal_flags(sv, 0);
3473 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3475 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3477 I32 pos = mg->mg_len;
3479 sv_pos_b2u(sv, &pos);
3480 mg_flags = 0; /* sv_pos_b2u does get magic */
3484 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3485 magic_setutf8(sv,mg); /* clear UTF8 cache */
3488 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3490 if (!utf8_to_bytes(s, &len)) {
3495 Perl_croak(aTHX_ "Wide character in %s",
3498 Perl_croak(aTHX_ "Wide character");
3509 =for apidoc sv_utf8_encode
3511 Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3512 flag off so that it looks like octets again.
3518 Perl_sv_utf8_encode(pTHX_ register SV *const sv)
3520 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3523 sv_force_normal_flags(sv, 0);
3525 if (SvREADONLY(sv)) {
3526 Perl_croak_no_modify(aTHX);
3528 (void) sv_utf8_upgrade(sv);
3533 =for apidoc sv_utf8_decode
3535 If the PV of the SV is an octet sequence in UTF-8
3536 and contains a multiple-byte character, the C<SvUTF8> flag is turned on
3537 so that it looks like a character. If the PV contains only single-byte
3538 characters, the C<SvUTF8> flag stays off.
3539 Scans PV for validity and returns false if the PV is invalid UTF-8.
3545 Perl_sv_utf8_decode(pTHX_ register SV *const sv)
3547 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3550 const U8 *start, *c;
3553 /* The octets may have got themselves encoded - get them back as
3556 if (!sv_utf8_downgrade(sv, TRUE))
3559 /* it is actually just a matter of turning the utf8 flag on, but
3560 * we want to make sure everything inside is valid utf8 first.
3562 c = start = (const U8 *) SvPVX_const(sv);
3563 if (!is_utf8_string(c, SvCUR(sv)+1))
3565 e = (const U8 *) SvEND(sv);
3568 if (!UTF8_IS_INVARIANT(ch)) {
3573 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3574 /* adjust pos to the start of a UTF8 char sequence */
3575 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3577 I32 pos = mg->mg_len;
3579 for (c = start + pos; c > start; c--) {
3580 if (UTF8_IS_START(*c))
3583 mg->mg_len = c - start;
3586 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3587 magic_setutf8(sv,mg); /* clear UTF8 cache */
3594 =for apidoc sv_setsv
3596 Copies the contents of the source SV C<ssv> into the destination SV
3597 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3598 function if the source SV needs to be reused. Does not handle 'set' magic.
3599 Loosely speaking, it performs a copy-by-value, obliterating any previous
3600 content of the destination.
3602 You probably want to use one of the assortment of wrappers, such as
3603 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3604 C<SvSetMagicSV_nosteal>.
3606 =for apidoc sv_setsv_flags
3608 Copies the contents of the source SV C<ssv> into the destination SV
3609 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3610 function if the source SV needs to be reused. Does not handle 'set' magic.
3611 Loosely speaking, it performs a copy-by-value, obliterating any previous
3612 content of the destination.
3613 If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
3614 C<ssv> if appropriate, else not. If the C<flags> parameter has the
3615 C<NOSTEAL> bit set then the buffers of temps will not be stolen. <sv_setsv>
3616 and C<sv_setsv_nomg> are implemented in terms of this function.
3618 You probably want to use one of the assortment of wrappers, such as
3619 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3620 C<SvSetMagicSV_nosteal>.
3622 This is the primary function for copying scalars, and most other
3623 copy-ish functions and macros use this underneath.
3629 S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
3631 I32 mro_changes = 0; /* 1 = method, 2 = isa, 3 = recursive isa */
3632 HV *old_stash = NULL;
3634 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB;
3636 if (dtype != SVt_PVGV && !isGV_with_GP(dstr)) {
3637 const char * const name = GvNAME(sstr);
3638 const STRLEN len = GvNAMELEN(sstr);
3640 if (dtype >= SVt_PV) {
3646 SvUPGRADE(dstr, SVt_PVGV);
3647 (void)SvOK_off(dstr);
3648 /* FIXME - why are we doing this, then turning it off and on again
3650 isGV_with_GP_on(dstr);
3652 GvSTASH(dstr) = GvSTASH(sstr);
3654 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
3655 gv_name_set(MUTABLE_GV(dstr), name, len,
3656 GV_ADD | (GvNAMEUTF8(sstr) ? SVf_UTF8 : 0 ));
3657 SvFAKE_on(dstr); /* can coerce to non-glob */
3660 if(GvGP(MUTABLE_GV(sstr))) {
3661 /* If source has method cache entry, clear it */
3663 SvREFCNT_dec(GvCV(sstr));
3664 GvCV_set(sstr, NULL);
3667 /* If source has a real method, then a method is
3670 GvCV((const GV *)sstr) && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3676 /* If dest already had a real method, that's a change as well */
3678 !mro_changes && GvGP(MUTABLE_GV(dstr)) && GvCVu((const GV *)dstr)
3679 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3684 /* We don't need to check the name of the destination if it was not a
3685 glob to begin with. */
3686 if(dtype == SVt_PVGV) {
3687 const char * const name = GvNAME((const GV *)dstr);
3690 /* The stash may have been detached from the symbol table, so
3692 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3693 && GvAV((const GV *)sstr)
3697 const STRLEN len = GvNAMELEN(dstr);
3698 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
3699 || (len == 1 && name[0] == ':')) {
3702 /* Set aside the old stash, so we can reset isa caches on
3704 if((old_stash = GvHV(dstr)))
3705 /* Make sure we do not lose it early. */
3706 SvREFCNT_inc_simple_void_NN(
3707 sv_2mortal((SV *)old_stash)
3713 gp_free(MUTABLE_GV(dstr));
3714 isGV_with_GP_off(dstr);
3715 (void)SvOK_off(dstr);
3716 isGV_with_GP_on(dstr);
3717 GvINTRO_off(dstr); /* one-shot flag */
3718 GvGP_set(dstr, gp_ref(GvGP(sstr)));
3719 if (SvTAINTED(sstr))
3721 if (GvIMPORTED(dstr) != GVf_IMPORTED
3722 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
3724 GvIMPORTED_on(dstr);
3727 if(mro_changes == 2) {
3729 SV * const sref = (SV *)GvAV((const GV *)dstr);
3730 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
3731 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
3732 AV * const ary = newAV();
3733 av_push(ary, mg->mg_obj); /* takes the refcount */
3734 mg->mg_obj = (SV *)ary;
3736 av_push((AV *)mg->mg_obj, SvREFCNT_inc_simple_NN(dstr));
3738 else sv_magic(sref, dstr, PERL_MAGIC_isa, NULL, 0);
3739 mro_isa_changed_in(GvSTASH(dstr));
3741 else if(mro_changes == 3) {
3742 HV * const stash = GvHV(dstr);
3743 if(old_stash ? (HV *)HvENAME_get(old_stash) : stash)
3749 else if(mro_changes) mro_method_changed_in(GvSTASH(dstr));
3754 S_glob_assign_ref(pTHX_ SV *const dstr, SV *const sstr)
3756 SV * const sref = SvREFCNT_inc(SvRV(sstr));
3758 const int intro = GvINTRO(dstr);
3761 const U32 stype = SvTYPE(sref);
3763 PERL_ARGS_ASSERT_GLOB_ASSIGN_REF;
3766 GvINTRO_off(dstr); /* one-shot flag */
3767 GvLINE(dstr) = CopLINE(PL_curcop);
3768 GvEGV(dstr) = MUTABLE_GV(dstr);
3773 location = (SV **) &(GvGP(dstr)->gp_cv); /* XXX bypassing GvCV_set */
3774 import_flag = GVf_IMPORTED_CV;
3777 location = (SV **) &GvHV(dstr);
3778 import_flag = GVf_IMPORTED_HV;
3781 location = (SV **) &GvAV(dstr);
3782 import_flag = GVf_IMPORTED_AV;
3785 location = (SV **) &GvIOp(dstr);
3788 location = (SV **) &GvFORM(dstr);
3791 location = &GvSV(dstr);
3792 import_flag = GVf_IMPORTED_SV;
3795 if (stype == SVt_PVCV) {
3796 /*if (GvCVGEN(dstr) && (GvCV(dstr) != (const CV *)sref || GvCVGEN(dstr))) {*/
3797 if (GvCVGEN(dstr)) {
3798 SvREFCNT_dec(GvCV(dstr));
3799 GvCV_set(dstr, NULL);
3800 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
3803 SAVEGENERICSV(*location);
3807 if (stype == SVt_PVCV && (*location != sref || GvCVGEN(dstr))) {
3808 CV* const cv = MUTABLE_CV(*location);
3810 if (!GvCVGEN((const GV *)dstr) &&
3811 (CvROOT(cv) || CvXSUB(cv)))
3813 /* Redefining a sub - warning is mandatory if
3814 it was a const and its value changed. */
3815 if (CvCONST(cv) && CvCONST((const CV *)sref)
3817 == cv_const_sv((const CV *)sref)) {
3819 /* They are 2 constant subroutines generated from
3820 the same constant. This probably means that
3821 they are really the "same" proxy subroutine
3822 instantiated in 2 places. Most likely this is
3823 when a constant is exported twice. Don't warn.
3826 else if (ckWARN(WARN_REDEFINE)
3828 && (!CvCONST((const CV *)sref)
3829 || sv_cmp(cv_const_sv(cv),
3830 cv_const_sv((const CV *)
3832 Perl_warner(aTHX_ packWARN(WARN_REDEFINE),
3835 ? "Constant subroutine %"HEKf
3836 "::%"HEKf" redefined"
3837 : "Subroutine %"HEKf"::%"HEKf
3840 HvNAME_HEK(GvSTASH((const GV *)dstr))
3842 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr))));
3846 cv_ckproto_len_flags(cv, (const GV *)dstr,
3847 SvPOK(sref) ? CvPROTO(sref) : NULL,
3848 SvPOK(sref) ? CvPROTOLEN(sref) : 0,
3849 SvPOK(sref) ? SvUTF8(sref) : 0);
3851 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
3852 GvASSUMECV_on(dstr);
3853 if(GvSTASH(dstr)) mro_method_changed_in(GvSTASH(dstr)); /* sub foo { 1 } sub bar { 2 } *bar = \&foo */
3856 if (import_flag && !(GvFLAGS(dstr) & import_flag)
3857 && CopSTASH_ne(PL_curcop, GvSTASH(dstr))) {
3858 GvFLAGS(dstr) |= import_flag;
3860 if (stype == SVt_PVHV) {
3861 const char * const name = GvNAME((GV*)dstr);
3862 const STRLEN len = GvNAMELEN(dstr);
3865 (len > 1 && name[len-2] == ':' && name[len-1] == ':')
3866 || (len == 1 && name[0] == ':')
3868 && (!dref || HvENAME_get(dref))
3871 (HV *)sref, (HV *)dref,
3877 stype == SVt_PVAV && sref != dref
3878 && strEQ(GvNAME((GV*)dstr), "ISA")
3879 /* The stash may have been detached from the symbol table, so
3880 check its name before doing anything. */
3881 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3884 MAGIC * const omg = dref && SvSMAGICAL(dref)
3885 ? mg_find(dref, PERL_MAGIC_isa)
3887 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
3888 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
3889 AV * const ary = newAV();
3890 av_push(ary, mg->mg_obj); /* takes the refcount */
3891 mg->mg_obj = (SV *)ary;
3894 if (SvTYPE(omg->mg_obj) == SVt_PVAV) {
3895 SV **svp = AvARRAY((AV *)omg->mg_obj);
3896 I32 items = AvFILLp((AV *)omg->mg_obj) + 1;
3900 SvREFCNT_inc_simple_NN(*svp++)
3906 SvREFCNT_inc_simple_NN(omg->mg_obj)
3910 av_push((AV *)mg->mg_obj,SvREFCNT_inc_simple_NN(dstr));
3915 sref, omg ? omg->mg_obj : dstr, PERL_MAGIC_isa, NULL, 0
3917 mg = mg_find(sref, PERL_MAGIC_isa);
3919 /* Since the *ISA assignment could have affected more than
3920 one stash, don't call mro_isa_changed_in directly, but let
3921 magic_clearisa do it for us, as it already has the logic for
3922 dealing with globs vs arrays of globs. */
3924 Perl_magic_clearisa(aTHX_ NULL, mg);
3929 if (SvTAINTED(sstr))
3935 Perl_sv_setsv_flags(pTHX_ SV *dstr, register SV* sstr, const I32 flags)
3938 register U32 sflags;
3940 register svtype stype;
3942 PERL_ARGS_ASSERT_SV_SETSV_FLAGS;
3947 if (SvIS_FREED(dstr)) {
3948 Perl_croak(aTHX_ "panic: attempt to copy value %" SVf
3949 " to a freed scalar %p", SVfARG(sstr), (void *)dstr);
3951 SV_CHECK_THINKFIRST_COW_DROP(dstr);
3953 sstr = &PL_sv_undef;
3954 if (SvIS_FREED(sstr)) {
3955 Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p to %p",
3956 (void*)sstr, (void*)dstr);
3958 stype = SvTYPE(sstr);
3959 dtype = SvTYPE(dstr);
3961 (void)SvAMAGIC_off(dstr);
3964 /* need to nuke the magic */
3968 /* There's a lot of redundancy below but we're going for speed here */
3973 if (dtype != SVt_PVGV && dtype != SVt_PVLV) {
3974 (void)SvOK_off(dstr);
3982 sv_upgrade(dstr, SVt_IV);
3986 sv_upgrade(dstr, SVt_PVIV);
3990 goto end_of_first_switch;
3992 (void)SvIOK_only(dstr);
3993 SvIV_set(dstr, SvIVX(sstr));
3996 /* SvTAINTED can only be true if the SV has taint magic, which in
3997 turn means that the SV type is PVMG (or greater). This is the
3998 case statement for SVt_IV, so this cannot be true (whatever gcov
4000 assert(!SvTAINTED(sstr));
4005 if (dtype < SVt_PV && dtype != SVt_IV)
4006 sv_upgrade(dstr, SVt_IV);
4014 sv_upgrade(dstr, SVt_NV);
4018 sv_upgrade(dstr, SVt_PVNV);
4022 goto end_of_first_switch;
4024 SvNV_set(dstr, SvNVX(sstr));
4025 (void)SvNOK_only(dstr);
4026 /* SvTAINTED can only be true if the SV has taint magic, which in
4027 turn means that the SV type is PVMG (or greater). This is the
4028 case statement for SVt_NV, so this cannot be true (whatever gcov
4030 assert(!SvTAINTED(sstr));
4036 #ifdef PERL_OLD_COPY_ON_WRITE
4037 if ((SvFLAGS(sstr) & CAN_COW_MASK) == CAN_COW_FLAGS) {
4038 if (dtype < SVt_PVIV)
4039 sv_upgrade(dstr, SVt_PVIV);
4046 sv_upgrade(dstr, SVt_PV);
4049 if (dtype < SVt_PVIV)
4050 sv_upgrade(dstr, SVt_PVIV);
4053 if (dtype < SVt_PVNV)
4054 sv_upgrade(dstr, SVt_PVNV);
4058 const char * const type = sv_reftype(sstr,0);
4060 Perl_croak(aTHX_ "Bizarre copy of %s in %s", type, OP_DESC(PL_op));
4062 Perl_croak(aTHX_ "Bizarre copy of %s", type);
4067 if (dtype < SVt_REGEXP)
4068 sv_upgrade(dstr, SVt_REGEXP);
4071 /* case SVt_BIND: */
4075 if (SvGMAGICAL(sstr) && (flags & SV_GMAGIC)) {
4077 if (SvTYPE(sstr) != stype)
4078 stype = SvTYPE(sstr);
4080 if (isGV_with_GP(sstr) && dtype <= SVt_PVLV) {
4081 glob_assign_glob(dstr, sstr, dtype);
4084 if (stype == SVt_PVLV)
4085 SvUPGRADE(dstr, SVt_PVNV);
4087 SvUPGRADE(dstr, (svtype)stype);
4089 end_of_first_switch:
4091 /* dstr may have been upgraded. */
4092 dtype = SvTYPE(dstr);
4093 sflags = SvFLAGS(sstr);
4095 if (dtype == SVt_PVCV || dtype == SVt_PVFM) {
4096 /* Assigning to a subroutine sets the prototype. */
4099 const char *const ptr = SvPV_const(sstr, len);
4101 SvGROW(dstr, len + 1);
4102 Copy(ptr, SvPVX(dstr), len + 1, char);
4103 SvCUR_set(dstr, len);
4105 SvFLAGS(dstr) |= sflags & SVf_UTF8;
4106 CvAUTOLOAD_off(dstr);
4110 } else if (dtype == SVt_PVAV || dtype == SVt_PVHV) {
4111 const char * const type = sv_reftype(dstr,0);
4113 Perl_croak(aTHX_ "Cannot copy to %s in %s", type, OP_DESC(PL_op));
4115 Perl_croak(aTHX_ "Cannot copy to %s", type);
4116 } else if (sflags & SVf_ROK) {
4117 if (isGV_with_GP(dstr)
4118 && SvTYPE(SvRV(sstr)) == SVt_PVGV && isGV_with_GP(SvRV(sstr))) {
4121 if (GvIMPORTED(dstr) != GVf_IMPORTED
4122 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
4124 GvIMPORTED_on(dstr);
4129 glob_assign_glob(dstr, sstr, dtype);
4133 if (dtype >= SVt_PV) {
4134 if (isGV_with_GP(dstr)) {
4135 glob_assign_ref(dstr, sstr);
4138 if (SvPVX_const(dstr)) {
4144 (void)SvOK_off(dstr);
4145 SvRV_set(dstr, SvREFCNT_inc(SvRV(sstr)));
4146 SvFLAGS(dstr) |= sflags & SVf_ROK;
4147 assert(!(sflags & SVp_NOK));
4148 assert(!(sflags & SVp_IOK));
4149 assert(!(sflags & SVf_NOK));
4150 assert(!(sflags & SVf_IOK));
4152 else if (isGV_with_GP(dstr)) {
4153 if (!(sflags & SVf_OK)) {
4154 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4155 "Undefined value assigned to typeglob");
4158 GV *gv = gv_fetchsv_nomg(sstr, GV_ADD, SVt_PVGV);
4159 if (dstr != (const SV *)gv) {
4160 const char * const name = GvNAME((const GV *)dstr);
4161 const STRLEN len = GvNAMELEN(dstr);
4162 HV *old_stash = NULL;
4163 bool reset_isa = FALSE;
4164 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
4165 || (len == 1 && name[0] == ':')) {
4166 /* Set aside the old stash, so we can reset isa caches
4167 on its subclasses. */
4168 if((old_stash = GvHV(dstr))) {
4169 /* Make sure we do not lose it early. */
4170 SvREFCNT_inc_simple_void_NN(
4171 sv_2mortal((SV *)old_stash)
4178 gp_free(MUTABLE_GV(dstr));
4179 GvGP_set(dstr, gp_ref(GvGP(gv)));
4182 HV * const stash = GvHV(dstr);
4184 old_stash ? (HV *)HvENAME_get(old_stash) : stash
4194 else if (dtype == SVt_REGEXP && stype == SVt_REGEXP) {
4195 reg_temp_copy((REGEXP*)dstr, (REGEXP*)sstr);
4197 else if (sflags & SVp_POK) {
4201 * Check to see if we can just swipe the string. If so, it's a
4202 * possible small lose on short strings, but a big win on long ones.
4203 * It might even be a win on short strings if SvPVX_const(dstr)
4204 * has to be allocated and SvPVX_const(sstr) has to be freed.
4205 * Likewise if we can set up COW rather than doing an actual copy, we
4206 * drop to the else clause, as the swipe code and the COW setup code
4207 * have much in common.
4210 /* Whichever path we take through the next code, we want this true,
4211 and doing it now facilitates the COW check. */
4212 (void)SvPOK_only(dstr);
4215 /* If we're already COW then this clause is not true, and if COW
4216 is allowed then we drop down to the else and make dest COW
4217 with us. If caller hasn't said that we're allowed to COW
4218 shared hash keys then we don't do the COW setup, even if the
4219 source scalar is a shared hash key scalar. */
4220 (((flags & SV_COW_SHARED_HASH_KEYS)
4221 ? (sflags & (SVf_FAKE|SVf_READONLY)) != (SVf_FAKE|SVf_READONLY)
4222 : 1 /* If making a COW copy is forbidden then the behaviour we
4223 desire is as if the source SV isn't actually already
4224 COW, even if it is. So we act as if the source flags
4225 are not COW, rather than actually testing them. */
4227 #ifndef PERL_OLD_COPY_ON_WRITE
4228 /* The change that added SV_COW_SHARED_HASH_KEYS makes the logic
4229 when PERL_OLD_COPY_ON_WRITE is defined a little wrong.
4230 Conceptually PERL_OLD_COPY_ON_WRITE being defined should
4231 override SV_COW_SHARED_HASH_KEYS, because it means "always COW"
4232 but in turn, it's somewhat dead code, never expected to go
4233 live, but more kept as a placeholder on how to do it better
4234 in a newer implementation. */
4235 /* If we are COW and dstr is a suitable target then we drop down
4236 into the else and make dest a COW of us. */
4237 || (SvFLAGS(dstr) & CAN_COW_MASK) != CAN_COW_FLAGS
4242 (sflags & SVs_TEMP) && /* slated for free anyway? */
4243 !(sflags & SVf_OOK) && /* and not involved in OOK hack? */
4244 (!(flags & SV_NOSTEAL)) &&
4245 /* and we're allowed to steal temps */
4246 SvREFCNT(sstr) == 1 && /* and no other references to it? */
4247 SvLEN(sstr)) /* and really is a string */
4248 #ifdef PERL_OLD_COPY_ON_WRITE
4249 && ((flags & SV_COW_SHARED_HASH_KEYS)
4250 ? (!((sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4251 && (SvFLAGS(dstr) & CAN_COW_MASK) == CAN_COW_FLAGS
4252 && SvTYPE(sstr) >= SVt_PVIV && SvTYPE(sstr) != SVt_PVFM))
4256 /* Failed the swipe test, and it's not a shared hash key either.
4257 Have to copy the string. */
4258 STRLEN len = SvCUR(sstr);
4259 SvGROW(dstr, len + 1); /* inlined from sv_setpvn */
4260 Move(SvPVX_const(sstr),SvPVX(dstr),len,char);
4261 SvCUR_set(dstr, len);
4262 *SvEND(dstr) = '\0';
4264 /* If PERL_OLD_COPY_ON_WRITE is not defined, then isSwipe will always
4266 /* Either it's a shared hash key, or it's suitable for
4267 copy-on-write or we can swipe the string. */
4269 PerlIO_printf(Perl_debug_log, "Copy on write: sstr --> dstr\n");
4273 #ifdef PERL_OLD_COPY_ON_WRITE
4275 if ((sflags & (SVf_FAKE | SVf_READONLY))
4276 != (SVf_FAKE | SVf_READONLY)) {
4277 SvREADONLY_on(sstr);
4279 /* Make the source SV into a loop of 1.
4280 (about to become 2) */
4281 SV_COW_NEXT_SV_SET(sstr, sstr);
4285 /* Initial code is common. */
4286 if (SvPVX_const(dstr)) { /* we know that dtype >= SVt_PV */
4291 /* making another shared SV. */
4292 STRLEN cur = SvCUR(sstr);
4293 STRLEN len = SvLEN(sstr);
4294 #ifdef PERL_OLD_COPY_ON_WRITE
4296 assert (SvTYPE(dstr) >= SVt_PVIV);
4297 /* SvIsCOW_normal */
4298 /* splice us in between source and next-after-source. */
4299 SV_COW_NEXT_SV_SET(dstr, SV_COW_NEXT_SV(sstr));
4300 SV_COW_NEXT_SV_SET(sstr, dstr);
4301 SvPV_set(dstr, SvPVX_mutable(sstr));
4305 /* SvIsCOW_shared_hash */
4306 DEBUG_C(PerlIO_printf(Perl_debug_log,
4307 "Copy on write: Sharing hash\n"));
4309 assert (SvTYPE(dstr) >= SVt_PV);
4311 HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)))));
4313 SvLEN_set(dstr, len);
4314 SvCUR_set(dstr, cur);
4315 SvREADONLY_on(dstr);
4319 { /* Passes the swipe test. */
4320 SvPV_set(dstr, SvPVX_mutable(sstr));
4321 SvLEN_set(dstr, SvLEN(sstr));
4322 SvCUR_set(dstr, SvCUR(sstr));
4325 (void)SvOK_off(sstr); /* NOTE: nukes most SvFLAGS on sstr */
4326 SvPV_set(sstr, NULL);
4332 if (sflags & SVp_NOK) {
4333 SvNV_set(dstr, SvNVX(sstr));
4335 if (sflags & SVp_IOK) {
4336 SvIV_set(dstr, SvIVX(sstr));
4337 /* Must do this otherwise some other overloaded use of 0x80000000
4338 gets confused. I guess SVpbm_VALID */
4339 if (sflags & SVf_IVisUV)
4342 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_NOK|SVp_NOK|SVf_UTF8);
4344 const MAGIC * const smg = SvVSTRING_mg(sstr);
4346 sv_magic(dstr, NULL, PERL_MAGIC_vstring,
4347 smg->mg_ptr, smg->mg_len);
4348 SvRMAGICAL_on(dstr);
4352 else if (sflags & (SVp_IOK|SVp_NOK)) {
4353 (void)SvOK_off(dstr);
4354 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_IVisUV|SVf_NOK|SVp_NOK);
4355 if (sflags & SVp_IOK) {
4356 /* XXXX Do we want to set IsUV for IV(ROK)? Be extra safe... */
4357 SvIV_set(dstr, SvIVX(sstr));
4359 if (sflags & SVp_NOK) {
4360 SvNV_set(dstr, SvNVX(sstr));
4364 if (isGV_with_GP(sstr)) {
4365 gv_efullname3(dstr, MUTABLE_GV(sstr), "*");
4368 (void)SvOK_off(dstr);
4370 if (SvTAINTED(sstr))
4375 =for apidoc sv_setsv_mg
4377 Like C<sv_setsv>, but also handles 'set' magic.
4383 Perl_sv_setsv_mg(pTHX_ SV *const dstr, register SV *const sstr)
4385 PERL_ARGS_ASSERT_SV_SETSV_MG;
4387 sv_setsv(dstr,sstr);
4391 #ifdef PERL_OLD_COPY_ON_WRITE
4393 Perl_sv_setsv_cow(pTHX_ SV *dstr, SV *sstr)
4395 STRLEN cur = SvCUR(sstr);
4396 STRLEN len = SvLEN(sstr);
4397 register char *new_pv;
4399 PERL_ARGS_ASSERT_SV_SETSV_COW;
4402 PerlIO_printf(Perl_debug_log, "Fast copy on write: %p -> %p\n",
4403 (void*)sstr, (void*)dstr);
4410 if (SvTHINKFIRST(dstr))
4411 sv_force_normal_flags(dstr, SV_COW_DROP_PV);
4412 else if (SvPVX_const(dstr))
4413 Safefree(SvPVX_const(dstr));
4417 SvUPGRADE(dstr, SVt_PVIV);
4419 assert (SvPOK(sstr));
4420 assert (SvPOKp(sstr));
4421 assert (!SvIOK(sstr));
4422 assert (!SvIOKp(sstr));
4423 assert (!SvNOK(sstr));
4424 assert (!SvNOKp(sstr));
4426 if (SvIsCOW(sstr)) {
4428 if (SvLEN(sstr) == 0) {
4429 /* source is a COW shared hash key. */
4430 DEBUG_C(PerlIO_printf(Perl_debug_log,
4431 "Fast copy on write: Sharing hash\n"));
4432 new_pv = HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr))));
4435 SV_COW_NEXT_SV_SET(dstr, SV_COW_NEXT_SV(sstr));
4437 assert ((SvFLAGS(sstr) & CAN_COW_MASK) == CAN_COW_FLAGS);
4438 SvUPGRADE(sstr, SVt_PVIV);
4439 SvREADONLY_on(sstr);
4441 DEBUG_C(PerlIO_printf(Perl_debug_log,
4442 "Fast copy on write: Converting sstr to COW\n"));
4443 SV_COW_NEXT_SV_SET(dstr, sstr);
4445 SV_COW_NEXT_SV_SET(sstr, dstr);
4446 new_pv = SvPVX_mutable(sstr);
4449 SvPV_set(dstr, new_pv);
4450 SvFLAGS(dstr) = (SVt_PVIV|SVf_POK|SVp_POK|SVf_FAKE|SVf_READONLY);
4453 SvLEN_set(dstr, len);
4454 SvCUR_set(dstr, cur);
4463 =for apidoc sv_setpvn
4465 Copies a string into an SV. The C<len> parameter indicates the number of
4466 bytes to be copied. If the C<ptr> argument is NULL the SV will become
4467 undefined. Does not handle 'set' magic. See C<sv_setpvn_mg>.
4473 Perl_sv_setpvn(pTHX_ register SV *const sv, register const char *const ptr, register const STRLEN len)
4476 register char *dptr;
4478 PERL_ARGS_ASSERT_SV_SETPVN;
4480 SV_CHECK_THINKFIRST_COW_DROP(sv);
4486 /* len is STRLEN which is unsigned, need to copy to signed */
4489 Perl_croak(aTHX_ "panic: sv_setpvn called with negative strlen");
4491 SvUPGRADE(sv, SVt_PV);
4493 dptr = SvGROW(sv, len + 1);
4494 Move(ptr,dptr,len,char);
4497 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4499 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4503 =for apidoc sv_setpvn_mg
4505 Like C<sv_setpvn>, but also handles 'set' magic.
4511 Perl_sv_setpvn_mg(pTHX_ register SV *const sv, register const char *const ptr, register const STRLEN len)
4513 PERL_ARGS_ASSERT_SV_SETPVN_MG;
4515 sv_setpvn(sv,ptr,len);
4520 =for apidoc sv_setpv
4522 Copies a string into an SV. The string must be null-terminated. Does not
4523 handle 'set' magic. See C<sv_setpv_mg>.
4529 Perl_sv_setpv(pTHX_ register SV *const sv, register const char *const ptr)
4532 register STRLEN len;
4534 PERL_ARGS_ASSERT_SV_SETPV;
4536 SV_CHECK_THINKFIRST_COW_DROP(sv);
4542 SvUPGRADE(sv, SVt_PV);
4544 SvGROW(sv, len + 1);
4545 Move(ptr,SvPVX(sv),len+1,char);
4547 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4549 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4553 =for apidoc sv_setpv_mg
4555 Like C<sv_setpv>, but also handles 'set' magic.
4561 Perl_sv_setpv_mg(pTHX_ register SV *const sv, register const char *const ptr)
4563 PERL_ARGS_ASSERT_SV_SETPV_MG;
4570 Perl_sv_sethek(pTHX_ register SV *const sv, const HEK *const hek)
4574 PERL_ARGS_ASSERT_SV_SETHEK;
4580 if (HEK_LEN(hek) == HEf_SVKEY) {
4581 sv_setsv(sv, *(SV**)HEK_KEY(hek));
4584 const int flags = HEK_FLAGS(hek);
4585 if (flags & HVhek_WASUTF8) {
4586 STRLEN utf8_len = HEK_LEN(hek);
4587 char *as_utf8 = (char *)bytes_to_utf8((U8*)HEK_KEY(hek), &utf8_len);
4588 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
4591 } else if (flags & (HVhek_REHASH|HVhek_UNSHARED)) {
4592 sv_setpvn(sv, HEK_KEY(hek), HEK_LEN(hek));
4595 else SvUTF8_off(sv);
4599 SvUPGRADE(sv, SVt_PV);
4600 sv_usepvn_flags(sv, (char *)HEK_KEY(share_hek_hek(hek)), HEK_LEN(hek), SV_HAS_TRAILING_NUL);
4607 else SvUTF8_off(sv);
4615 =for apidoc sv_usepvn_flags
4617 Tells an SV to use C<ptr> to find its string value. Normally the
4618 string is stored inside the SV but sv_usepvn allows the SV to use an
4619 outside string. The C<ptr> should point to memory that was allocated