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 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_PVMG && 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 */
1333 /* XXX Is this still needed? Was it ever needed? Surely as there is
1334 no route from NV to PVIV, NOK can never be true */
1335 assert(!SvNOKp(sv));
1347 assert(new_type_details->body_size);
1348 /* We always allocated the full length item with PURIFY. To do this
1349 we fake things so that arena is false for all 16 types.. */
1350 if(new_type_details->arena) {
1351 /* This points to the start of the allocated area. */
1352 new_body_inline(new_body, new_type);
1353 Zero(new_body, new_type_details->body_size, char);
1354 new_body = ((char *)new_body) - new_type_details->offset;
1356 new_body = new_NOARENAZ(new_type_details);
1358 SvANY(sv) = new_body;
1360 if (old_type_details->copy) {
1361 /* There is now the potential for an upgrade from something without
1362 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */
1363 int offset = old_type_details->offset;
1364 int length = old_type_details->copy;
1366 if (new_type_details->offset > old_type_details->offset) {
1367 const int difference
1368 = new_type_details->offset - old_type_details->offset;
1369 offset += difference;
1370 length -= difference;
1372 assert (length >= 0);
1374 Copy((char *)old_body + offset, (char *)new_body + offset, length,
1378 #ifndef NV_ZERO_IS_ALLBITS_ZERO
1379 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a
1380 * correct 0.0 for us. Otherwise, if the old body didn't have an
1381 * NV slot, but the new one does, then we need to initialise the
1382 * freshly created NV slot with whatever the correct bit pattern is
1384 if (old_type_details->zero_nv && !new_type_details->zero_nv
1385 && !isGV_with_GP(sv))
1389 if (new_type == SVt_PVIO) {
1390 IO * const io = MUTABLE_IO(sv);
1391 GV *iogv = gv_fetchpvs("IO::File::", GV_ADD, SVt_PVHV);
1394 /* Clear the stashcache because a new IO could overrule a package
1396 DEBUG_o(Perl_deb(aTHX_ "sv_upgrade clearing PL_stashcache\n"));
1397 hv_clear(PL_stashcache);
1399 SvSTASH_set(io, MUTABLE_HV(SvREFCNT_inc(GvHV(iogv))));
1400 IoPAGE_LEN(sv) = 60;
1402 if (old_type < SVt_PV) {
1403 /* referant will be NULL unless the old type was SVt_IV emulating
1405 sv->sv_u.svu_rv = referant;
1409 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1410 (unsigned long)new_type);
1413 if (old_type > SVt_IV) {
1417 /* Note that there is an assumption that all bodies of types that
1418 can be upgraded came from arenas. Only the more complex non-
1419 upgradable types are allowed to be directly malloc()ed. */
1420 assert(old_type_details->arena);
1421 del_body((void*)((char*)old_body + old_type_details->offset),
1422 &PL_body_roots[old_type]);
1428 =for apidoc sv_backoff
1430 Remove any string offset. You should normally use the C<SvOOK_off> macro
1437 Perl_sv_backoff(pTHX_ register SV *const sv)
1440 const char * const s = SvPVX_const(sv);
1442 PERL_ARGS_ASSERT_SV_BACKOFF;
1443 PERL_UNUSED_CONTEXT;
1446 assert(SvTYPE(sv) != SVt_PVHV);
1447 assert(SvTYPE(sv) != SVt_PVAV);
1449 SvOOK_offset(sv, delta);
1451 SvLEN_set(sv, SvLEN(sv) + delta);
1452 SvPV_set(sv, SvPVX(sv) - delta);
1453 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
1454 SvFLAGS(sv) &= ~SVf_OOK;
1461 Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1462 upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1463 Use the C<SvGROW> wrapper instead.
1469 Perl_sv_grow(pTHX_ register SV *const sv, register STRLEN newlen)
1473 PERL_ARGS_ASSERT_SV_GROW;
1475 if (PL_madskills && newlen >= 0x100000) {
1476 PerlIO_printf(Perl_debug_log,
1477 "Allocation too large: %"UVxf"\n", (UV)newlen);
1479 #ifdef HAS_64K_LIMIT
1480 if (newlen >= 0x10000) {
1481 PerlIO_printf(Perl_debug_log,
1482 "Allocation too large: %"UVxf"\n", (UV)newlen);
1485 #endif /* HAS_64K_LIMIT */
1488 if (SvTYPE(sv) < SVt_PV) {
1489 sv_upgrade(sv, SVt_PV);
1490 s = SvPVX_mutable(sv);
1492 else if (SvOOK(sv)) { /* pv is offset? */
1494 s = SvPVX_mutable(sv);
1495 if (newlen > SvLEN(sv))
1496 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
1497 #ifdef HAS_64K_LIMIT
1498 if (newlen >= 0x10000)
1503 s = SvPVX_mutable(sv);
1505 if (newlen > SvLEN(sv)) { /* need more room? */
1506 STRLEN minlen = SvCUR(sv);
1507 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
1508 if (newlen < minlen)
1510 #ifndef Perl_safesysmalloc_size
1511 newlen = PERL_STRLEN_ROUNDUP(newlen);
1513 if (SvLEN(sv) && s) {
1514 s = (char*)saferealloc(s, newlen);
1517 s = (char*)safemalloc(newlen);
1518 if (SvPVX_const(sv) && SvCUR(sv)) {
1519 Move(SvPVX_const(sv), s, (newlen < SvCUR(sv)) ? newlen : SvCUR(sv), char);
1523 #ifdef Perl_safesysmalloc_size
1524 /* Do this here, do it once, do it right, and then we will never get
1525 called back into sv_grow() unless there really is some growing
1527 SvLEN_set(sv, Perl_safesysmalloc_size(s));
1529 SvLEN_set(sv, newlen);
1536 =for apidoc sv_setiv
1538 Copies an integer into the given SV, upgrading first if necessary.
1539 Does not handle 'set' magic. See also C<sv_setiv_mg>.
1545 Perl_sv_setiv(pTHX_ register SV *const sv, const IV i)
1549 PERL_ARGS_ASSERT_SV_SETIV;
1551 SV_CHECK_THINKFIRST_COW_DROP(sv);
1552 switch (SvTYPE(sv)) {
1555 sv_upgrade(sv, SVt_IV);
1558 sv_upgrade(sv, SVt_PVIV);
1562 if (!isGV_with_GP(sv))
1569 /* diag_listed_as: Can't coerce %s to %s in %s */
1570 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1574 (void)SvIOK_only(sv); /* validate number */
1580 =for apidoc sv_setiv_mg
1582 Like C<sv_setiv>, but also handles 'set' magic.
1588 Perl_sv_setiv_mg(pTHX_ register SV *const sv, const IV i)
1590 PERL_ARGS_ASSERT_SV_SETIV_MG;
1597 =for apidoc sv_setuv
1599 Copies an unsigned integer into the given SV, upgrading first if necessary.
1600 Does not handle 'set' magic. See also C<sv_setuv_mg>.
1606 Perl_sv_setuv(pTHX_ register SV *const sv, const UV u)
1608 PERL_ARGS_ASSERT_SV_SETUV;
1610 /* With the if statement to ensure that integers are stored as IVs whenever
1612 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
1615 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1617 If you wish to remove the following if statement, so that this routine
1618 (and its callers) always return UVs, please benchmark to see what the
1619 effect is. Modern CPUs may be different. Or may not :-)
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 /* diag_listed_as: Argument "%s" isn't numeric%s */
1780 "Argument \"%s\" isn't numeric in %s", pv,
1783 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1784 /* diag_listed_as: Argument "%s" isn't numeric%s */
1785 "Argument \"%s\" isn't numeric", pv);
1789 =for apidoc looks_like_number
1791 Test if the content of an SV looks like a number (or is a number).
1792 C<Inf> and C<Infinity> are treated as numbers (so will not issue a
1793 non-numeric warning), even if your atof() doesn't grok them. Get-magic is
1800 Perl_looks_like_number(pTHX_ SV *const sv)
1805 PERL_ARGS_ASSERT_LOOKS_LIKE_NUMBER;
1807 if (SvPOK(sv) || SvPOKp(sv)) {
1808 sbegin = SvPV_nomg_const(sv, len);
1811 return SvFLAGS(sv) & (SVf_NOK|SVp_NOK|SVf_IOK|SVp_IOK);
1812 return grok_number(sbegin, len, NULL);
1816 S_glob_2number(pTHX_ GV * const gv)
1818 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1820 /* We know that all GVs stringify to something that is not-a-number,
1821 so no need to test that. */
1822 if (ckWARN(WARN_NUMERIC))
1824 SV *const buffer = sv_newmortal();
1825 gv_efullname3(buffer, gv, "*");
1826 not_a_number(buffer);
1828 /* We just want something true to return, so that S_sv_2iuv_common
1829 can tail call us and return true. */
1833 /* Actually, ISO C leaves conversion of UV to IV undefined, but
1834 until proven guilty, assume that things are not that bad... */
1839 As 64 bit platforms often have an NV that doesn't preserve all bits of
1840 an IV (an assumption perl has been based on to date) it becomes necessary
1841 to remove the assumption that the NV always carries enough precision to
1842 recreate the IV whenever needed, and that the NV is the canonical form.
1843 Instead, IV/UV and NV need to be given equal rights. So as to not lose
1844 precision as a side effect of conversion (which would lead to insanity
1845 and the dragon(s) in t/op/numconvert.t getting very angry) the intent is
1846 1) to distinguish between IV/UV/NV slots that have cached a valid
1847 conversion where precision was lost and IV/UV/NV slots that have a
1848 valid conversion which has lost no precision
1849 2) to ensure that if a numeric conversion to one form is requested that
1850 would lose precision, the precise conversion (or differently
1851 imprecise conversion) is also performed and cached, to prevent
1852 requests for different numeric formats on the same SV causing
1853 lossy conversion chains. (lossless conversion chains are perfectly
1858 SvIOKp is true if the IV slot contains a valid value
1859 SvIOK is true only if the IV value is accurate (UV if SvIOK_UV true)
1860 SvNOKp is true if the NV slot contains a valid value
1861 SvNOK is true only if the NV value is accurate
1864 while converting from PV to NV, check to see if converting that NV to an
1865 IV(or UV) would lose accuracy over a direct conversion from PV to
1866 IV(or UV). If it would, cache both conversions, return NV, but mark
1867 SV as IOK NOKp (ie not NOK).
1869 While converting from PV to IV, check to see if converting that IV to an
1870 NV would lose accuracy over a direct conversion from PV to NV. If it
1871 would, cache both conversions, flag similarly.
1873 Before, the SV value "3.2" could become NV=3.2 IV=3 NOK, IOK quite
1874 correctly because if IV & NV were set NV *always* overruled.
1875 Now, "3.2" will become NV=3.2 IV=3 NOK, IOKp, because the flag's meaning
1876 changes - now IV and NV together means that the two are interchangeable:
1877 SvIVX == (IV) SvNVX && SvNVX == (NV) SvIVX;
1879 The benefit of this is that operations such as pp_add know that if
1880 SvIOK is true for both left and right operands, then integer addition
1881 can be used instead of floating point (for cases where the result won't
1882 overflow). Before, floating point was always used, which could lead to
1883 loss of precision compared with integer addition.
1885 * making IV and NV equal status should make maths accurate on 64 bit
1887 * may speed up maths somewhat if pp_add and friends start to use
1888 integers when possible instead of fp. (Hopefully the overhead in
1889 looking for SvIOK and checking for overflow will not outweigh the
1890 fp to integer speedup)
1891 * will slow down integer operations (callers of SvIV) on "inaccurate"
1892 values, as the change from SvIOK to SvIOKp will cause a call into
1893 sv_2iv each time rather than a macro access direct to the IV slot
1894 * should speed up number->string conversion on integers as IV is
1895 favoured when IV and NV are equally accurate
1897 ####################################################################
1898 You had better be using SvIOK_notUV if you want an IV for arithmetic:
1899 SvIOK is true if (IV or UV), so you might be getting (IV)SvUV.
1900 On the other hand, SvUOK is true iff UV.
1901 ####################################################################
1903 Your mileage will vary depending your CPU's relative fp to integer
1907 #ifndef NV_PRESERVES_UV
1908 # define IS_NUMBER_UNDERFLOW_IV 1
1909 # define IS_NUMBER_UNDERFLOW_UV 2
1910 # define IS_NUMBER_IV_AND_UV 2
1911 # define IS_NUMBER_OVERFLOW_IV 4
1912 # define IS_NUMBER_OVERFLOW_UV 5
1914 /* sv_2iuv_non_preserve(): private routine for use by sv_2iv() and sv_2uv() */
1916 /* For sv_2nv these three cases are "SvNOK and don't bother casting" */
1918 S_sv_2iuv_non_preserve(pTHX_ register SV *const sv
1926 PERL_ARGS_ASSERT_SV_2IUV_NON_PRESERVE;
1928 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));
1929 if (SvNVX(sv) < (NV)IV_MIN) {
1930 (void)SvIOKp_on(sv);
1932 SvIV_set(sv, IV_MIN);
1933 return IS_NUMBER_UNDERFLOW_IV;
1935 if (SvNVX(sv) > (NV)UV_MAX) {
1936 (void)SvIOKp_on(sv);
1939 SvUV_set(sv, UV_MAX);
1940 return IS_NUMBER_OVERFLOW_UV;
1942 (void)SvIOKp_on(sv);
1944 /* Can't use strtol etc to convert this string. (See truth table in
1946 if (SvNVX(sv) <= (UV)IV_MAX) {
1947 SvIV_set(sv, I_V(SvNVX(sv)));
1948 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
1949 SvIOK_on(sv); /* Integer is precise. NOK, IOK */
1951 /* Integer is imprecise. NOK, IOKp */
1953 return SvNVX(sv) < 0 ? IS_NUMBER_UNDERFLOW_UV : IS_NUMBER_IV_AND_UV;
1956 SvUV_set(sv, U_V(SvNVX(sv)));
1957 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
1958 if (SvUVX(sv) == UV_MAX) {
1959 /* As we know that NVs don't preserve UVs, UV_MAX cannot
1960 possibly be preserved by NV. Hence, it must be overflow.
1962 return IS_NUMBER_OVERFLOW_UV;
1964 SvIOK_on(sv); /* Integer is precise. NOK, UOK */
1966 /* Integer is imprecise. NOK, IOKp */
1968 return IS_NUMBER_OVERFLOW_IV;
1970 #endif /* !NV_PRESERVES_UV*/
1973 S_sv_2iuv_common(pTHX_ SV *const sv)
1977 PERL_ARGS_ASSERT_SV_2IUV_COMMON;
1980 /* erm. not sure. *should* never get NOKp (without NOK) from sv_2nv
1981 * without also getting a cached IV/UV from it at the same time
1982 * (ie PV->NV conversion should detect loss of accuracy and cache
1983 * IV or UV at same time to avoid this. */
1984 /* IV-over-UV optimisation - choose to cache IV if possible */
1986 if (SvTYPE(sv) == SVt_NV)
1987 sv_upgrade(sv, SVt_PVNV);
1989 (void)SvIOKp_on(sv); /* Must do this first, to clear any SvOOK */
1990 /* < not <= as for NV doesn't preserve UV, ((NV)IV_MAX+1) will almost
1991 certainly cast into the IV range at IV_MAX, whereas the correct
1992 answer is the UV IV_MAX +1. Hence < ensures that dodgy boundary
1994 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
1995 if (Perl_isnan(SvNVX(sv))) {
2001 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2002 SvIV_set(sv, I_V(SvNVX(sv)));
2003 if (SvNVX(sv) == (NV) SvIVX(sv)
2004 #ifndef NV_PRESERVES_UV
2005 && (((UV)1 << NV_PRESERVES_UV_BITS) >
2006 (UV)(SvIVX(sv) > 0 ? SvIVX(sv) : -SvIVX(sv)))
2007 /* Don't flag it as "accurately an integer" if the number
2008 came from a (by definition imprecise) NV operation, and
2009 we're outside the range of NV integer precision */
2013 SvIOK_on(sv); /* Can this go wrong with rounding? NWC */
2015 /* scalar has trailing garbage, eg "42a" */
2017 DEBUG_c(PerlIO_printf(Perl_debug_log,
2018 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (precise)\n",
2024 /* IV not precise. No need to convert from PV, as NV
2025 conversion would already have cached IV if it detected
2026 that PV->IV would be better than PV->NV->IV
2027 flags already correct - don't set public IOK. */
2028 DEBUG_c(PerlIO_printf(Perl_debug_log,
2029 "0x%"UVxf" iv(%"NVgf" => %"IVdf") (imprecise)\n",
2034 /* Can the above go wrong if SvIVX == IV_MIN and SvNVX < IV_MIN,
2035 but the cast (NV)IV_MIN rounds to a the value less (more
2036 negative) than IV_MIN which happens to be equal to SvNVX ??
2037 Analogous to 0xFFFFFFFFFFFFFFFF rounding up to NV (2**64) and
2038 NV rounding back to 0xFFFFFFFFFFFFFFFF, so UVX == UV(NVX) and
2039 (NV)UVX == NVX are both true, but the values differ. :-(
2040 Hopefully for 2s complement IV_MIN is something like
2041 0x8000000000000000 which will be exact. NWC */
2044 SvUV_set(sv, U_V(SvNVX(sv)));
2046 (SvNVX(sv) == (NV) SvUVX(sv))
2047 #ifndef NV_PRESERVES_UV
2048 /* Make sure it's not 0xFFFFFFFFFFFFFFFF */
2049 /*&& (SvUVX(sv) != UV_MAX) irrelevant with code below */
2050 && (((UV)1 << NV_PRESERVES_UV_BITS) > SvUVX(sv))
2051 /* Don't flag it as "accurately an integer" if the number
2052 came from a (by definition imprecise) NV operation, and
2053 we're outside the range of NV integer precision */
2059 DEBUG_c(PerlIO_printf(Perl_debug_log,
2060 "0x%"UVxf" 2iv(%"UVuf" => %"IVdf") (as unsigned)\n",
2066 else if (SvPOKp(sv)) {
2068 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2069 /* We want to avoid a possible problem when we cache an IV/ a UV which
2070 may be later translated to an NV, and the resulting NV is not
2071 the same as the direct translation of the initial string
2072 (eg 123.456 can shortcut to the IV 123 with atol(), but we must
2073 be careful to ensure that the value with the .456 is around if the
2074 NV value is requested in the future).
2076 This means that if we cache such an IV/a UV, we need to cache the
2077 NV as well. Moreover, we trade speed for space, and do not
2078 cache the NV if we are sure it's not needed.
2081 /* SVt_PVNV is one higher than SVt_PVIV, hence this order */
2082 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2083 == IS_NUMBER_IN_UV) {
2084 /* It's definitely an integer, only upgrade to PVIV */
2085 if (SvTYPE(sv) < SVt_PVIV)
2086 sv_upgrade(sv, SVt_PVIV);
2088 } else if (SvTYPE(sv) < SVt_PVNV)
2089 sv_upgrade(sv, SVt_PVNV);
2091 /* If NVs preserve UVs then we only use the UV value if we know that
2092 we aren't going to call atof() below. If NVs don't preserve UVs
2093 then the value returned may have more precision than atof() will
2094 return, even though value isn't perfectly accurate. */
2095 if ((numtype & (IS_NUMBER_IN_UV
2096 #ifdef NV_PRESERVES_UV
2099 )) == IS_NUMBER_IN_UV) {
2100 /* This won't turn off the public IOK flag if it was set above */
2101 (void)SvIOKp_on(sv);
2103 if (!(numtype & IS_NUMBER_NEG)) {
2105 if (value <= (UV)IV_MAX) {
2106 SvIV_set(sv, (IV)value);
2108 /* it didn't overflow, and it was positive. */
2109 SvUV_set(sv, value);
2113 /* 2s complement assumption */
2114 if (value <= (UV)IV_MIN) {
2115 SvIV_set(sv, -(IV)value);
2117 /* Too negative for an IV. This is a double upgrade, but
2118 I'm assuming it will be rare. */
2119 if (SvTYPE(sv) < SVt_PVNV)
2120 sv_upgrade(sv, SVt_PVNV);
2124 SvNV_set(sv, -(NV)value);
2125 SvIV_set(sv, IV_MIN);
2129 /* For !NV_PRESERVES_UV and IS_NUMBER_IN_UV and IS_NUMBER_NOT_INT we
2130 will be in the previous block to set the IV slot, and the next
2131 block to set the NV slot. So no else here. */
2133 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2134 != IS_NUMBER_IN_UV) {
2135 /* It wasn't an (integer that doesn't overflow the UV). */
2136 SvNV_set(sv, Atof(SvPVX_const(sv)));
2138 if (! numtype && ckWARN(WARN_NUMERIC))
2141 #if defined(USE_LONG_DOUBLE)
2142 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%" PERL_PRIgldbl ")\n",
2143 PTR2UV(sv), SvNVX(sv)));
2145 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"NVgf")\n",
2146 PTR2UV(sv), SvNVX(sv)));
2149 #ifdef NV_PRESERVES_UV
2150 (void)SvIOKp_on(sv);
2152 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2153 SvIV_set(sv, I_V(SvNVX(sv)));
2154 if ((NV)(SvIVX(sv)) == SvNVX(sv)) {
2157 NOOP; /* Integer is imprecise. NOK, IOKp */
2159 /* UV will not work better than IV */
2161 if (SvNVX(sv) > (NV)UV_MAX) {
2163 /* Integer is inaccurate. NOK, IOKp, is UV */
2164 SvUV_set(sv, UV_MAX);
2166 SvUV_set(sv, U_V(SvNVX(sv)));
2167 /* 0xFFFFFFFFFFFFFFFF not an issue in here, NVs
2168 NV preservse UV so can do correct comparison. */
2169 if ((NV)(SvUVX(sv)) == SvNVX(sv)) {
2172 NOOP; /* Integer is imprecise. NOK, IOKp, is UV */
2177 #else /* NV_PRESERVES_UV */
2178 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2179 == (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT)) {
2180 /* The IV/UV slot will have been set from value returned by
2181 grok_number above. The NV slot has just been set using
2184 assert (SvIOKp(sv));
2186 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2187 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2188 /* Small enough to preserve all bits. */
2189 (void)SvIOKp_on(sv);
2191 SvIV_set(sv, I_V(SvNVX(sv)));
2192 if ((NV)(SvIVX(sv)) == SvNVX(sv))
2194 /* Assumption: first non-preserved integer is < IV_MAX,
2195 this NV is in the preserved range, therefore: */
2196 if (!(U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))
2198 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);
2202 0 0 already failed to read UV.
2203 0 1 already failed to read UV.
2204 1 0 you won't get here in this case. IV/UV
2205 slot set, public IOK, Atof() unneeded.
2206 1 1 already read UV.
2207 so there's no point in sv_2iuv_non_preserve() attempting
2208 to use atol, strtol, strtoul etc. */
2210 sv_2iuv_non_preserve (sv, numtype);
2212 sv_2iuv_non_preserve (sv);
2216 #endif /* NV_PRESERVES_UV */
2217 /* It might be more code efficient to go through the entire logic above
2218 and conditionally set with SvIOKp_on() rather than SvIOK(), but it
2219 gets complex and potentially buggy, so more programmer efficient
2220 to do it this way, by turning off the public flags: */
2222 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2226 if (isGV_with_GP(sv))
2227 return glob_2number(MUTABLE_GV(sv));
2229 if (!SvPADTMP(sv)) {
2230 if (!PL_localizing && ckWARN(WARN_UNINITIALIZED))
2233 if (SvTYPE(sv) < SVt_IV)
2234 /* Typically the caller expects that sv_any is not NULL now. */
2235 sv_upgrade(sv, SVt_IV);
2236 /* Return 0 from the caller. */
2243 =for apidoc sv_2iv_flags
2245 Return the integer value of an SV, doing any necessary string
2246 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2247 Normally used via the C<SvIV(sv)> and C<SvIVx(sv)> macros.
2253 Perl_sv_2iv_flags(pTHX_ register SV *const sv, const I32 flags)
2260 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2266 if (flags & SV_SKIP_OVERLOAD)
2268 tmpstr = AMG_CALLunary(sv, numer_amg);
2269 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2270 return SvIV(tmpstr);
2273 return PTR2IV(SvRV(sv));
2276 if (SvVALID(sv) || SvTYPE(sv) == SVt_REGEXP) {
2277 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2278 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2279 In practice they are extremely unlikely to actually get anywhere
2280 accessible by user Perl code - the only way that I'm aware of is when
2281 a constant subroutine which is used as the second argument to index.
2283 Regexps have no SvIVX and SvNVX fields.
2289 = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2291 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2292 == IS_NUMBER_IN_UV) {
2293 /* It's definitely an integer */
2294 if (numtype & IS_NUMBER_NEG) {
2295 if (value < (UV)IV_MIN)
2298 if (value < (UV)IV_MAX)
2303 if (ckWARN(WARN_NUMERIC))
2306 return I_V(Atof(SvPVX_const(sv)));
2310 if (SvTHINKFIRST(sv)) {
2311 #ifdef PERL_OLD_COPY_ON_WRITE
2313 sv_force_normal_flags(sv, 0);
2316 if (SvREADONLY(sv) && !SvOK(sv)) {
2317 if (ckWARN(WARN_UNINITIALIZED))
2324 if (S_sv_2iuv_common(aTHX_ sv))
2328 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"IVdf")\n",
2329 PTR2UV(sv),SvIVX(sv)));
2330 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
2334 =for apidoc sv_2uv_flags
2336 Return the unsigned integer value of an SV, doing any necessary string
2337 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2338 Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
2344 Perl_sv_2uv_flags(pTHX_ register SV *const sv, const I32 flags)
2351 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2357 if (flags & SV_SKIP_OVERLOAD)
2359 tmpstr = AMG_CALLunary(sv, numer_amg);
2360 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2361 return SvUV(tmpstr);
2364 return PTR2UV(SvRV(sv));
2367 if (SvVALID(sv) || SvTYPE(sv) == SVt_REGEXP) {
2368 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2369 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2370 Regexps have no SvIVX and SvNVX fields. */
2375 = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2377 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2378 == IS_NUMBER_IN_UV) {
2379 /* It's definitely an integer */
2380 if (!(numtype & IS_NUMBER_NEG))
2384 if (ckWARN(WARN_NUMERIC))
2387 return U_V(Atof(SvPVX_const(sv)));
2391 if (SvTHINKFIRST(sv)) {
2392 #ifdef PERL_OLD_COPY_ON_WRITE
2394 sv_force_normal_flags(sv, 0);
2397 if (SvREADONLY(sv) && !SvOK(sv)) {
2398 if (ckWARN(WARN_UNINITIALIZED))
2405 if (S_sv_2iuv_common(aTHX_ sv))
2409 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2uv(%"UVuf")\n",
2410 PTR2UV(sv),SvUVX(sv)));
2411 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
2415 =for apidoc sv_2nv_flags
2417 Return the num value of an SV, doing any necessary string or integer
2418 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2419 Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
2425 Perl_sv_2nv_flags(pTHX_ register SV *const sv, const I32 flags)
2430 if (SvGMAGICAL(sv) || SvVALID(sv) || SvTYPE(sv) == SVt_REGEXP) {
2431 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2432 the same flag bit as SVf_IVisUV, so must not let them cache NVs.
2433 Regexps have no SvIVX and SvNVX fields. */
2434 if (flags & SV_GMAGIC)
2438 if (SvPOKp(sv) && !SvIOKp(sv)) {
2439 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
2440 !grok_number(SvPVX_const(sv), SvCUR(sv), NULL))
2442 return Atof(SvPVX_const(sv));
2446 return (NV)SvUVX(sv);
2448 return (NV)SvIVX(sv);
2453 assert(SvTYPE(sv) >= SVt_PVMG);
2454 /* This falls through to the report_uninit near the end of the
2456 } else if (SvTHINKFIRST(sv)) {
2461 if (flags & SV_SKIP_OVERLOAD)
2463 tmpstr = AMG_CALLunary(sv, numer_amg);
2464 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2465 return SvNV(tmpstr);
2468 return PTR2NV(SvRV(sv));
2470 #ifdef PERL_OLD_COPY_ON_WRITE
2472 sv_force_normal_flags(sv, 0);
2475 if (SvREADONLY(sv) && !SvOK(sv)) {
2476 if (ckWARN(WARN_UNINITIALIZED))
2481 if (SvTYPE(sv) < SVt_NV) {
2482 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2483 sv_upgrade(sv, SVt_NV);
2484 #ifdef USE_LONG_DOUBLE
2486 STORE_NUMERIC_LOCAL_SET_STANDARD();
2487 PerlIO_printf(Perl_debug_log,
2488 "0x%"UVxf" num(%" PERL_PRIgldbl ")\n",
2489 PTR2UV(sv), SvNVX(sv));
2490 RESTORE_NUMERIC_LOCAL();
2494 STORE_NUMERIC_LOCAL_SET_STANDARD();
2495 PerlIO_printf(Perl_debug_log, "0x%"UVxf" num(%"NVgf")\n",
2496 PTR2UV(sv), SvNVX(sv));
2497 RESTORE_NUMERIC_LOCAL();
2501 else if (SvTYPE(sv) < SVt_PVNV)
2502 sv_upgrade(sv, SVt_PVNV);
2507 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
2508 #ifdef NV_PRESERVES_UV
2514 /* Only set the public NV OK flag if this NV preserves the IV */
2515 /* Check it's not 0xFFFFFFFFFFFFFFFF */
2517 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
2518 : (SvIVX(sv) == I_V(SvNVX(sv))))
2524 else if (SvPOKp(sv)) {
2526 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2527 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
2529 #ifdef NV_PRESERVES_UV
2530 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2531 == IS_NUMBER_IN_UV) {
2532 /* It's definitely an integer */
2533 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
2535 SvNV_set(sv, Atof(SvPVX_const(sv)));
2541 SvNV_set(sv, Atof(SvPVX_const(sv)));
2542 /* Only set the public NV OK flag if this NV preserves the value in
2543 the PV at least as well as an IV/UV would.
2544 Not sure how to do this 100% reliably. */
2545 /* if that shift count is out of range then Configure's test is
2546 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2548 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2549 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2550 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
2551 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2552 /* Can't use strtol etc to convert this string, so don't try.
2553 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2556 /* value has been set. It may not be precise. */
2557 if ((numtype & IS_NUMBER_NEG) && (value > (UV)IV_MIN)) {
2558 /* 2s complement assumption for (UV)IV_MIN */
2559 SvNOK_on(sv); /* Integer is too negative. */
2564 if (numtype & IS_NUMBER_NEG) {
2565 SvIV_set(sv, -(IV)value);
2566 } else if (value <= (UV)IV_MAX) {
2567 SvIV_set(sv, (IV)value);
2569 SvUV_set(sv, value);
2573 if (numtype & IS_NUMBER_NOT_INT) {
2574 /* I believe that even if the original PV had decimals,
2575 they are lost beyond the limit of the FP precision.
2576 However, neither is canonical, so both only get p
2577 flags. NWC, 2000/11/25 */
2578 /* Both already have p flags, so do nothing */
2580 const NV nv = SvNVX(sv);
2581 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2582 if (SvIVX(sv) == I_V(nv)) {
2585 /* It had no "." so it must be integer. */
2589 /* between IV_MAX and NV(UV_MAX).
2590 Could be slightly > UV_MAX */
2592 if (numtype & IS_NUMBER_NOT_INT) {
2593 /* UV and NV both imprecise. */
2595 const UV nv_as_uv = U_V(nv);
2597 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2606 /* It might be more code efficient to go through the entire logic above
2607 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2608 gets complex and potentially buggy, so more programmer efficient
2609 to do it this way, by turning off the public flags: */
2611 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2612 #endif /* NV_PRESERVES_UV */
2615 if (isGV_with_GP(sv)) {
2616 glob_2number(MUTABLE_GV(sv));
2620 if (!PL_localizing && !SvPADTMP(sv) && ckWARN(WARN_UNINITIALIZED))
2622 assert (SvTYPE(sv) >= SVt_NV);
2623 /* Typically the caller expects that sv_any is not NULL now. */
2624 /* XXX Ilya implies that this is a bug in callers that assume this
2625 and ideally should be fixed. */
2628 #if defined(USE_LONG_DOUBLE)
2630 STORE_NUMERIC_LOCAL_SET_STANDARD();
2631 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2nv(%" PERL_PRIgldbl ")\n",
2632 PTR2UV(sv), SvNVX(sv));
2633 RESTORE_NUMERIC_LOCAL();
2637 STORE_NUMERIC_LOCAL_SET_STANDARD();
2638 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 1nv(%"NVgf")\n",
2639 PTR2UV(sv), SvNVX(sv));
2640 RESTORE_NUMERIC_LOCAL();
2649 Return an SV with the numeric value of the source SV, doing any necessary
2650 reference or overload conversion. You must use the C<SvNUM(sv)> macro to
2651 access this function.
2657 Perl_sv_2num(pTHX_ register SV *const sv)
2659 PERL_ARGS_ASSERT_SV_2NUM;
2664 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
2665 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
2666 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2667 return sv_2num(tmpsv);
2669 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2672 /* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2673 * UV as a string towards the end of buf, and return pointers to start and
2676 * We assume that buf is at least TYPE_CHARS(UV) long.
2680 S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
2682 char *ptr = buf + TYPE_CHARS(UV);
2683 char * const ebuf = ptr;
2686 PERL_ARGS_ASSERT_UIV_2BUF;
2698 *--ptr = '0' + (char)(uv % 10);
2707 =for apidoc sv_2pv_flags
2709 Returns a pointer to the string value of an SV, and sets *lp to its length.
2710 If flags includes SV_GMAGIC, does an mg_get() first. Coerces sv to a
2711 string if necessary. Normally invoked via the C<SvPV_flags> macro.
2712 C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
2718 Perl_sv_2pv_flags(pTHX_ register SV *const sv, STRLEN *const lp, const I32 flags)
2728 if (SvGMAGICAL(sv) && (flags & SV_GMAGIC))
2733 if (flags & SV_SKIP_OVERLOAD)
2735 tmpstr = AMG_CALLunary(sv, string_amg);
2736 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
2737 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2739 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2743 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2744 if (flags & SV_CONST_RETURN) {
2745 pv = (char *) SvPVX_const(tmpstr);
2747 pv = (flags & SV_MUTABLE_RETURN)
2748 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
2751 *lp = SvCUR(tmpstr);
2753 pv = sv_2pv_flags(tmpstr, lp, flags);
2766 SV *const referent = SvRV(sv);
2770 retval = buffer = savepvn("NULLREF", len);
2771 } else if (SvTYPE(referent) == SVt_REGEXP &&
2772 (!(PL_curcop->cop_hints & HINT_NO_AMAGIC) ||
2773 amagic_is_enabled(string_amg))) {
2774 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
2778 /* If the regex is UTF-8 we want the containing scalar to
2779 have an UTF-8 flag too */
2786 *lp = RX_WRAPLEN(re);
2788 return RX_WRAPPED(re);
2790 const char *const typestr = sv_reftype(referent, 0);
2791 const STRLEN typelen = strlen(typestr);
2792 UV addr = PTR2UV(referent);
2793 const char *stashname = NULL;
2794 STRLEN stashnamelen = 0; /* hush, gcc */
2795 const char *buffer_end;
2797 if (SvOBJECT(referent)) {
2798 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
2801 stashname = HEK_KEY(name);
2802 stashnamelen = HEK_LEN(name);
2804 if (HEK_UTF8(name)) {
2810 stashname = "__ANON__";
2813 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
2814 + 2 * sizeof(UV) + 2 /* )\0 */;
2816 len = typelen + 3 /* (0x */
2817 + 2 * sizeof(UV) + 2 /* )\0 */;
2820 Newx(buffer, len, char);
2821 buffer_end = retval = buffer + len;
2823 /* Working backwards */
2827 *--retval = PL_hexdigit[addr & 15];
2828 } while (addr >>= 4);
2834 memcpy(retval, typestr, typelen);
2838 retval -= stashnamelen;
2839 memcpy(retval, stashname, stashnamelen);
2841 /* retval may not necessarily have reached the start of the
2843 assert (retval >= buffer);
2845 len = buffer_end - retval - 1; /* -1 for that \0 */
2857 if (flags & SV_MUTABLE_RETURN)
2858 return SvPVX_mutable(sv);
2859 if (flags & SV_CONST_RETURN)
2860 return (char *)SvPVX_const(sv);
2865 /* I'm assuming that if both IV and NV are equally valid then
2866 converting the IV is going to be more efficient */
2867 const U32 isUIOK = SvIsUV(sv);
2868 char buf[TYPE_CHARS(UV)];
2872 if (SvTYPE(sv) < SVt_PVIV)
2873 sv_upgrade(sv, SVt_PVIV);
2874 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
2876 /* inlined from sv_setpvn */
2877 s = SvGROW_mutable(sv, len + 1);
2878 Move(ptr, s, len, char);
2882 else if (SvNOK(sv)) {
2883 if (SvTYPE(sv) < SVt_PVNV)
2884 sv_upgrade(sv, SVt_PVNV);
2885 if (SvNVX(sv) == 0.0) {
2886 s = SvGROW_mutable(sv, 2);
2891 /* The +20 is pure guesswork. Configure test needed. --jhi */
2892 s = SvGROW_mutable(sv, NV_DIG + 20);
2893 /* some Xenix systems wipe out errno here */
2894 Gconvert(SvNVX(sv), NV_DIG, 0, s);
2903 else if (isGV_with_GP(sv)) {
2904 GV *const gv = MUTABLE_GV(sv);
2905 SV *const buffer = sv_newmortal();
2907 gv_efullname3(buffer, gv, "*");
2909 assert(SvPOK(buffer));
2913 *lp = SvCUR(buffer);
2914 return SvPVX(buffer);
2919 if (flags & SV_UNDEF_RETURNS_NULL)
2921 if (!PL_localizing && !SvPADTMP(sv) && ckWARN(WARN_UNINITIALIZED))
2923 /* Typically the caller expects that sv_any is not NULL now. */
2924 if (!SvREADONLY(sv) && SvTYPE(sv) < SVt_PV)
2925 sv_upgrade(sv, SVt_PV);
2930 const STRLEN len = s - SvPVX_const(sv);
2936 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
2937 PTR2UV(sv),SvPVX_const(sv)));
2938 if (flags & SV_CONST_RETURN)
2939 return (char *)SvPVX_const(sv);
2940 if (flags & SV_MUTABLE_RETURN)
2941 return SvPVX_mutable(sv);
2946 =for apidoc sv_copypv
2948 Copies a stringified representation of the source SV into the
2949 destination SV. Automatically performs any necessary mg_get and
2950 coercion of numeric values into strings. Guaranteed to preserve
2951 UTF8 flag even from overloaded objects. Similar in nature to
2952 sv_2pv[_flags] but operates directly on an SV instead of just the
2953 string. Mostly uses sv_2pv_flags to do its work, except when that
2954 would lose the UTF-8'ness of the PV.
2956 =for apidoc sv_copypv_nomg
2958 Like sv_copypv, but doesn't invoke get magic first.
2960 =for apidoc sv_copypv_flags
2962 Implementation of sv_copypv and sv_copypv_nomg. Calls get magic iff flags
2969 Perl_sv_copypv(pTHX_ SV *const dsv, register SV *const ssv)
2971 PERL_ARGS_ASSERT_SV_COPYPV;
2973 sv_copypv_flags(dsv, ssv, 0);
2977 Perl_sv_copypv_flags(pTHX_ SV *const dsv, register SV *const ssv, const I32 flags)
2982 PERL_ARGS_ASSERT_SV_COPYPV_FLAGS;
2984 if ((flags & SV_GMAGIC) && SvGMAGICAL(ssv))
2986 s = SvPV_nomg_const(ssv,len);
2987 sv_setpvn(dsv,s,len);
2995 =for apidoc sv_2pvbyte
2997 Return a pointer to the byte-encoded representation of the SV, and set *lp
2998 to its length. May cause the SV to be downgraded from UTF-8 as a
3001 Usually accessed via the C<SvPVbyte> macro.
3007 Perl_sv_2pvbyte(pTHX_ register SV *sv, STRLEN *const lp)
3009 PERL_ARGS_ASSERT_SV_2PVBYTE;
3011 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3012 || isGV_with_GP(sv) || SvROK(sv)) {
3013 SV *sv2 = sv_newmortal();
3017 else SvGETMAGIC(sv);
3018 sv_utf8_downgrade(sv,0);
3019 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3023 =for apidoc sv_2pvutf8
3025 Return a pointer to the UTF-8-encoded representation of the SV, and set *lp
3026 to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3028 Usually accessed via the C<SvPVutf8> macro.
3034 Perl_sv_2pvutf8(pTHX_ register SV *sv, STRLEN *const lp)
3036 PERL_ARGS_ASSERT_SV_2PVUTF8;
3038 if (((SvREADONLY(sv) || SvFAKE(sv)) && !SvIsCOW(sv))
3039 || isGV_with_GP(sv) || SvROK(sv))
3040 sv = sv_mortalcopy(sv);
3043 sv_utf8_upgrade_nomg(sv);
3044 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3049 =for apidoc sv_2bool
3051 This macro is only used by sv_true() or its macro equivalent, and only if
3052 the latter's argument is neither SvPOK, SvIOK nor SvNOK.
3053 It calls sv_2bool_flags with the SV_GMAGIC flag.
3055 =for apidoc sv_2bool_flags
3057 This function is only used by sv_true() and friends, and only if
3058 the latter's argument is neither SvPOK, SvIOK nor SvNOK. If the flags
3059 contain SV_GMAGIC, then it does an mg_get() first.
3066 Perl_sv_2bool_flags(pTHX_ register SV *const sv, const I32 flags)
3070 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
3072 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
3078 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
3079 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
3080 return cBOOL(SvTRUE(tmpsv));
3082 return SvRV(sv) != 0;
3084 return SvTRUE_common(sv, isGV_with_GP(sv) ? 1 : 0);
3088 =for apidoc sv_utf8_upgrade
3090 Converts the PV of an SV to its UTF-8-encoded form.
3091 Forces the SV to string form if it is not already.
3092 Will C<mg_get> on C<sv> if appropriate.
3093 Always sets the SvUTF8 flag to avoid future validity checks even
3094 if the whole string is the same in UTF-8 as not.
3095 Returns the number of bytes in the converted string
3097 This is not a general purpose byte encoding to Unicode interface:
3098 use the Encode extension for that.
3100 =for apidoc sv_utf8_upgrade_nomg
3102 Like sv_utf8_upgrade, but doesn't do magic on C<sv>.
3104 =for apidoc sv_utf8_upgrade_flags
3106 Converts the PV of an SV to its UTF-8-encoded form.
3107 Forces the SV to string form if it is not already.
3108 Always sets the SvUTF8 flag to avoid future validity checks even
3109 if all the bytes are invariant in UTF-8.
3110 If C<flags> has C<SV_GMAGIC> bit set,
3111 will C<mg_get> on C<sv> if appropriate, else not.
3112 Returns the number of bytes in the converted string
3113 C<sv_utf8_upgrade> and
3114 C<sv_utf8_upgrade_nomg> are implemented in terms of this function.
3116 This is not a general purpose byte encoding to Unicode interface:
3117 use the Encode extension for that.
3121 The grow version is currently not externally documented. It adds a parameter,
3122 extra, which is the number of unused bytes the string of 'sv' is guaranteed to
3123 have free after it upon return. This allows the caller to reserve extra space
3124 that it intends to fill, to avoid extra grows.
3126 Also externally undocumented for the moment is the flag SV_FORCE_UTF8_UPGRADE,
3127 which can be used to tell this function to not first check to see if there are
3128 any characters that are different in UTF-8 (variant characters) which would
3129 force it to allocate a new string to sv, but to assume there are. Typically
3130 this flag is used by a routine that has already parsed the string to find that
3131 there are such characters, and passes this information on so that the work
3132 doesn't have to be repeated.
3134 (One might think that the calling routine could pass in the position of the
3135 first such variant, so it wouldn't have to be found again. But that is not the
3136 case, because typically when the caller is likely to use this flag, it won't be
3137 calling this routine unless it finds something that won't fit into a byte.
3138 Otherwise it tries to not upgrade and just use bytes. But some things that
3139 do fit into a byte are variants in utf8, and the caller may not have been
3140 keeping track of these.)
3142 If the routine itself changes the string, it adds a trailing NUL. Such a NUL
3143 isn't guaranteed due to having other routines do the work in some input cases,
3144 or if the input is already flagged as being in utf8.
3146 The speed of this could perhaps be improved for many cases if someone wanted to
3147 write a fast function that counts the number of variant characters in a string,
3148 especially if it could return the position of the first one.
3153 Perl_sv_utf8_upgrade_flags_grow(pTHX_ register SV *const sv, const I32 flags, STRLEN extra)
3157 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
3159 if (sv == &PL_sv_undef)
3161 if (!SvPOK_nog(sv)) {
3163 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3164 (void) sv_2pv_flags(sv,&len, flags);
3166 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3170 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
3175 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3180 sv_force_normal_flags(sv, 0);
3183 if (PL_encoding && !(flags & SV_UTF8_NO_ENCODING)) {
3184 sv_recode_to_utf8(sv, PL_encoding);
3185 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3189 if (SvCUR(sv) == 0) {
3190 if (extra) SvGROW(sv, extra);
3191 } else { /* Assume Latin-1/EBCDIC */
3192 /* This function could be much more efficient if we
3193 * had a FLAG in SVs to signal if there are any variant
3194 * chars in the PV. Given that there isn't such a flag
3195 * make the loop as fast as possible (although there are certainly ways
3196 * to speed this up, eg. through vectorization) */
3197 U8 * s = (U8 *) SvPVX_const(sv);
3198 U8 * e = (U8 *) SvEND(sv);
3200 STRLEN two_byte_count = 0;
3202 if (flags & SV_FORCE_UTF8_UPGRADE) goto must_be_utf8;
3204 /* See if really will need to convert to utf8. We mustn't rely on our
3205 * incoming SV being well formed and having a trailing '\0', as certain
3206 * code in pp_formline can send us partially built SVs. */
3210 if (NATIVE_IS_INVARIANT(ch)) continue;
3212 t--; /* t already incremented; re-point to first variant */
3217 /* utf8 conversion not needed because all are invariants. Mark as
3218 * UTF-8 even if no variant - saves scanning loop */
3220 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3225 /* Here, the string should be converted to utf8, either because of an
3226 * input flag (two_byte_count = 0), or because a character that
3227 * requires 2 bytes was found (two_byte_count = 1). t points either to
3228 * the beginning of the string (if we didn't examine anything), or to
3229 * the first variant. In either case, everything from s to t - 1 will
3230 * occupy only 1 byte each on output.
3232 * There are two main ways to convert. One is to create a new string
3233 * and go through the input starting from the beginning, appending each
3234 * converted value onto the new string as we go along. It's probably
3235 * best to allocate enough space in the string for the worst possible
3236 * case rather than possibly running out of space and having to
3237 * reallocate and then copy what we've done so far. Since everything
3238 * from s to t - 1 is invariant, the destination can be initialized
3239 * with these using a fast memory copy
3241 * The other way is to figure out exactly how big the string should be
3242 * by parsing the entire input. Then you don't have to make it big
3243 * enough to handle the worst possible case, and more importantly, if
3244 * the string you already have is large enough, you don't have to
3245 * allocate a new string, you can copy the last character in the input
3246 * string to the final position(s) that will be occupied by the
3247 * converted string and go backwards, stopping at t, since everything
3248 * before that is invariant.
3250 * There are advantages and disadvantages to each method.
3252 * In the first method, we can allocate a new string, do the memory
3253 * copy from the s to t - 1, and then proceed through the rest of the
3254 * string byte-by-byte.
3256 * In the second method, we proceed through the rest of the input
3257 * string just calculating how big the converted string will be. Then
3258 * there are two cases:
3259 * 1) if the string has enough extra space to handle the converted
3260 * value. We go backwards through the string, converting until we
3261 * get to the position we are at now, and then stop. If this
3262 * position is far enough along in the string, this method is
3263 * faster than the other method. If the memory copy were the same
3264 * speed as the byte-by-byte loop, that position would be about
3265 * half-way, as at the half-way mark, parsing to the end and back
3266 * is one complete string's parse, the same amount as starting
3267 * over and going all the way through. Actually, it would be
3268 * somewhat less than half-way, as it's faster to just count bytes
3269 * than to also copy, and we don't have the overhead of allocating
3270 * a new string, changing the scalar to use it, and freeing the
3271 * existing one. But if the memory copy is fast, the break-even
3272 * point is somewhere after half way. The counting loop could be
3273 * sped up by vectorization, etc, to move the break-even point
3274 * further towards the beginning.
3275 * 2) if the string doesn't have enough space to handle the converted
3276 * value. A new string will have to be allocated, and one might
3277 * as well, given that, start from the beginning doing the first
3278 * method. We've spent extra time parsing the string and in
3279 * exchange all we've gotten is that we know precisely how big to
3280 * make the new one. Perl is more optimized for time than space,
3281 * so this case is a loser.
3282 * So what I've decided to do is not use the 2nd method unless it is
3283 * guaranteed that a new string won't have to be allocated, assuming
3284 * the worst case. I also decided not to put any more conditions on it
3285 * than this, for now. It seems likely that, since the worst case is
3286 * twice as big as the unknown portion of the string (plus 1), we won't
3287 * be guaranteed enough space, causing us to go to the first method,
3288 * unless the string is short, or the first variant character is near
3289 * the end of it. In either of these cases, it seems best to use the
3290 * 2nd method. The only circumstance I can think of where this would
3291 * be really slower is if the string had once had much more data in it
3292 * than it does now, but there is still a substantial amount in it */
3295 STRLEN invariant_head = t - s;
3296 STRLEN size = invariant_head + (e - t) * 2 + 1 + extra;
3297 if (SvLEN(sv) < size) {
3299 /* Here, have decided to allocate a new string */
3304 Newx(dst, size, U8);
3306 /* If no known invariants at the beginning of the input string,
3307 * set so starts from there. Otherwise, can use memory copy to
3308 * get up to where we are now, and then start from here */
3310 if (invariant_head <= 0) {
3313 Copy(s, dst, invariant_head, char);
3314 d = dst + invariant_head;
3318 const UV uv = NATIVE8_TO_UNI(*t++);
3319 if (UNI_IS_INVARIANT(uv))
3320 *d++ = (U8)UNI_TO_NATIVE(uv);
3322 *d++ = (U8)UTF8_EIGHT_BIT_HI(uv);
3323 *d++ = (U8)UTF8_EIGHT_BIT_LO(uv);
3327 SvPV_free(sv); /* No longer using pre-existing string */
3328 SvPV_set(sv, (char*)dst);
3329 SvCUR_set(sv, d - dst);
3330 SvLEN_set(sv, size);
3333 /* Here, have decided to get the exact size of the string.
3334 * Currently this happens only when we know that there is
3335 * guaranteed enough space to fit the converted string, so
3336 * don't have to worry about growing. If two_byte_count is 0,
3337 * then t points to the first byte of the string which hasn't
3338 * been examined yet. Otherwise two_byte_count is 1, and t
3339 * points to the first byte in the string that will expand to
3340 * two. Depending on this, start examining at t or 1 after t.
3343 U8 *d = t + two_byte_count;
3346 /* Count up the remaining bytes that expand to two */
3349 const U8 chr = *d++;
3350 if (! NATIVE_IS_INVARIANT(chr)) two_byte_count++;
3353 /* The string will expand by just the number of bytes that
3354 * occupy two positions. But we are one afterwards because of
3355 * the increment just above. This is the place to put the
3356 * trailing NUL, and to set the length before we decrement */
3358 d += two_byte_count;
3359 SvCUR_set(sv, d - s);
3363 /* Having decremented d, it points to the position to put the
3364 * very last byte of the expanded string. Go backwards through
3365 * the string, copying and expanding as we go, stopping when we
3366 * get to the part that is invariant the rest of the way down */
3370 const U8 ch = NATIVE8_TO_UNI(*e--);
3371 if (UNI_IS_INVARIANT(ch)) {
3372 *d-- = UNI_TO_NATIVE(ch);
3374 *d-- = (U8)UTF8_EIGHT_BIT_LO(ch);
3375 *d-- = (U8)UTF8_EIGHT_BIT_HI(ch);
3380 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3381 /* Update pos. We do it at the end rather than during
3382 * the upgrade, to avoid slowing down the common case
3383 * (upgrade without pos) */
3384 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3386 I32 pos = mg->mg_len;
3387 if (pos > 0 && (U32)pos > invariant_head) {
3388 U8 *d = (U8*) SvPVX(sv) + invariant_head;
3389 STRLEN n = (U32)pos - invariant_head;
3391 if (UTF8_IS_START(*d))
3396 mg->mg_len = d - (U8*)SvPVX(sv);
3399 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3400 magic_setutf8(sv,mg); /* clear UTF8 cache */
3405 /* Mark as UTF-8 even if no variant - saves scanning loop */
3411 =for apidoc sv_utf8_downgrade
3413 Attempts to convert the PV of an SV from characters to bytes.
3414 If the PV contains a character that cannot fit
3415 in a byte, this conversion will fail;
3416 in this case, either returns false or, if C<fail_ok> is not
3419 This is not a general purpose Unicode to byte encoding interface:
3420 use the Encode extension for that.
3426 Perl_sv_utf8_downgrade(pTHX_ register SV *const sv, const bool fail_ok)
3430 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3432 if (SvPOKp(sv) && SvUTF8(sv)) {
3436 int mg_flags = SV_GMAGIC;
3439 sv_force_normal_flags(sv, 0);
3441 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3443 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3445 I32 pos = mg->mg_len;
3447 sv_pos_b2u(sv, &pos);
3448 mg_flags = 0; /* sv_pos_b2u does get magic */
3452 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3453 magic_setutf8(sv,mg); /* clear UTF8 cache */
3456 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3458 if (!utf8_to_bytes(s, &len)) {
3463 Perl_croak(aTHX_ "Wide character in %s",
3466 Perl_croak(aTHX_ "Wide character");
3477 =for apidoc sv_utf8_encode
3479 Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3480 flag off so that it looks like octets again.
3486 Perl_sv_utf8_encode(pTHX_ register SV *const sv)
3488 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3490 if (SvREADONLY(sv)) {
3491 sv_force_normal_flags(sv, 0);
3493 (void) sv_utf8_upgrade(sv);
3498 =for apidoc sv_utf8_decode
3500 If the PV of the SV is an octet sequence in UTF-8
3501 and contains a multiple-byte character, the C<SvUTF8> flag is turned on
3502 so that it looks like a character. If the PV contains only single-byte
3503 characters, the C<SvUTF8> flag stays off.
3504 Scans PV for validity and returns false if the PV is invalid UTF-8.
3510 Perl_sv_utf8_decode(pTHX_ register SV *const sv)
3512 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3515 const U8 *start, *c;
3518 /* The octets may have got themselves encoded - get them back as
3521 if (!sv_utf8_downgrade(sv, TRUE))
3524 /* it is actually just a matter of turning the utf8 flag on, but
3525 * we want to make sure everything inside is valid utf8 first.
3527 c = start = (const U8 *) SvPVX_const(sv);
3528 if (!is_utf8_string(c, SvCUR(sv)))
3530 e = (const U8 *) SvEND(sv);
3533 if (!UTF8_IS_INVARIANT(ch)) {
3538 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3539 /* adjust pos to the start of a UTF8 char sequence */
3540 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3542 I32 pos = mg->mg_len;
3544 for (c = start + pos; c > start; c--) {
3545 if (UTF8_IS_START(*c))
3548 mg->mg_len = c - start;
3551 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3552 magic_setutf8(sv,mg); /* clear UTF8 cache */
3559 =for apidoc sv_setsv
3561 Copies the contents of the source SV C<ssv> into the destination SV
3562 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3563 function if the source SV needs to be reused. Does not handle 'set' magic.
3564 Loosely speaking, it performs a copy-by-value, obliterating any previous
3565 content of the destination.
3567 You probably want to use one of the assortment of wrappers, such as
3568 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3569 C<SvSetMagicSV_nosteal>.
3571 =for apidoc sv_setsv_flags
3573 Copies the contents of the source SV C<ssv> into the destination SV
3574 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3575 function if the source SV needs to be reused. Does not handle 'set' magic.
3576 Loosely speaking, it performs a copy-by-value, obliterating any previous
3577 content of the destination.
3578 If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
3579 C<ssv> if appropriate, else not. If the C<flags>
3580 parameter has the C<NOSTEAL> bit set then the
3581 buffers of temps will not be stolen. <sv_setsv>
3582 and C<sv_setsv_nomg> are implemented in terms of this function.
3584 You probably want to use one of the assortment of wrappers, such as
3585 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3586 C<SvSetMagicSV_nosteal>.
3588 This is the primary function for copying scalars, and most other
3589 copy-ish functions and macros use this underneath.
3595 S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
3597 I32 mro_changes = 0; /* 1 = method, 2 = isa, 3 = recursive isa */
3598 HV *old_stash = NULL;
3600 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB;
3602 if (dtype != SVt_PVGV && !isGV_with_GP(dstr)) {
3603 const char * const name = GvNAME(sstr);
3604 const STRLEN len = GvNAMELEN(sstr);
3606 if (dtype >= SVt_PV) {
3612 SvUPGRADE(dstr, SVt_PVGV);
3613 (void)SvOK_off(dstr);
3614 /* We have to turn this on here, even though we turn it off
3615 below, as GvSTASH will fail an assertion otherwise. */
3616 isGV_with_GP_on(dstr);
3618 GvSTASH(dstr) = GvSTASH(sstr);
3620 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
3621 gv_name_set(MUTABLE_GV(dstr), name, len,
3622 GV_ADD | (GvNAMEUTF8(sstr) ? SVf_UTF8 : 0 ));
3623 SvFAKE_on(dstr); /* can coerce to non-glob */
3626 if(GvGP(MUTABLE_GV(sstr))) {
3627 /* If source has method cache entry, clear it */
3629 SvREFCNT_dec(GvCV(sstr));
3630 GvCV_set(sstr, NULL);
3633 /* If source has a real method, then a method is
3636 GvCV((const GV *)sstr) && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3642 /* If dest already had a real method, that's a change as well */
3644 !mro_changes && GvGP(MUTABLE_GV(dstr)) && GvCVu((const GV *)dstr)
3645 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3650 /* We don't need to check the name of the destination if it was not a
3651 glob to begin with. */
3652 if(dtype == SVt_PVGV) {
3653 const char * const name = GvNAME((const GV *)dstr);
3656 /* The stash may have been detached from the symbol table, so
3658 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3662 const STRLEN len = GvNAMELEN(dstr);
3663 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
3664 || (len == 1 && name[0] == ':')) {
3667 /* Set aside the old stash, so we can reset isa caches on
3669 if((old_stash = GvHV(dstr)))
3670 /* Make sure we do not lose it early. */
3671 SvREFCNT_inc_simple_void_NN(
3672 sv_2mortal((SV *)old_stash)
3678 gp_free(MUTABLE_GV(dstr));
3679 isGV_with_GP_off(dstr); /* SvOK_off does not like globs. */
3680 (void)SvOK_off(dstr);
3681 isGV_with_GP_on(dstr);
3682 GvINTRO_off(dstr); /* one-shot flag */
3683 GvGP_set(dstr, gp_ref(GvGP(sstr)));
3684 if (SvTAINTED(sstr))
3686 if (GvIMPORTED(dstr) != GVf_IMPORTED
3687 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
3689 GvIMPORTED_on(dstr);
3692 if(mro_changes == 2) {
3693 if (GvAV((const GV *)sstr)) {
3695 SV * const sref = (SV *)GvAV((const GV *)dstr);
3696 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
3697 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
3698 AV * const ary = newAV();
3699 av_push(ary, mg->mg_obj); /* takes the refcount */
3700 mg->mg_obj = (SV *)ary;
3702 av_push((AV *)mg->mg_obj, SvREFCNT_inc_simple_NN(dstr));
3704 else sv_magic(sref, dstr, PERL_MAGIC_isa, NULL, 0);
3706 mro_isa_changed_in(GvSTASH(dstr));
3708 else if(mro_changes == 3) {
3709 HV * const stash = GvHV(dstr);
3710 if(old_stash ? (HV *)HvENAME_get(old_stash) : stash)
3716 else if(mro_changes) mro_method_changed_in(GvSTASH(dstr));
3721 S_glob_assign_ref(pTHX_ SV *const dstr, SV *const sstr)
3723 SV * const sref = SvREFCNT_inc(SvRV(sstr));
3725 const int intro = GvINTRO(dstr);
3728 const U32 stype = SvTYPE(sref);
3730 PERL_ARGS_ASSERT_GLOB_ASSIGN_REF;
3733 GvINTRO_off(dstr); /* one-shot flag */
3734 GvLINE(dstr) = CopLINE(PL_curcop);
3735 GvEGV(dstr) = MUTABLE_GV(dstr);
3740 location = (SV **) &(GvGP(dstr)->gp_cv); /* XXX bypassing GvCV_set */
3741 import_flag = GVf_IMPORTED_CV;
3744 location = (SV **) &GvHV(dstr);
3745 import_flag = GVf_IMPORTED_HV;
3748 location = (SV **) &GvAV(dstr);
3749 import_flag = GVf_IMPORTED_AV;
3752 location = (SV **) &GvIOp(dstr);
3755 location = (SV **) &GvFORM(dstr);
3758 location = &GvSV(dstr);
3759 import_flag = GVf_IMPORTED_SV;
3762 if (stype == SVt_PVCV) {
3763 /*if (GvCVGEN(dstr) && (GvCV(dstr) != (const CV *)sref || GvCVGEN(dstr))) {*/
3764 if (GvCVGEN(dstr)) {
3765 SvREFCNT_dec(GvCV(dstr));
3766 GvCV_set(dstr, NULL);
3767 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
3770 SAVEGENERICSV(*location);
3774 if (stype == SVt_PVCV && (*location != sref || GvCVGEN(dstr))) {
3775 CV* const cv = MUTABLE_CV(*location);
3777 if (!GvCVGEN((const GV *)dstr) &&
3778 (CvROOT(cv) || CvXSUB(cv)) &&
3779 /* redundant check that avoids creating the extra SV
3780 most of the time: */
3781 (CvCONST(cv) || ckWARN(WARN_REDEFINE)))
3783 SV * const new_const_sv =
3784 CvCONST((const CV *)sref)
3785 ? cv_const_sv((const CV *)sref)
3787 report_redefined_cv(
3788 sv_2mortal(Perl_newSVpvf(aTHX_
3791 HvNAME_HEK(GvSTASH((const GV *)dstr))
3793 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr)))
3796 CvCONST((const CV *)sref) ? &new_const_sv : NULL
3800 cv_ckproto_len_flags(cv, (const GV *)dstr,
3801 SvPOK(sref) ? CvPROTO(sref) : NULL,
3802 SvPOK(sref) ? CvPROTOLEN(sref) : 0,
3803 SvPOK(sref) ? SvUTF8(sref) : 0);
3805 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
3806 GvASSUMECV_on(dstr);
3807 if(GvSTASH(dstr)) mro_method_changed_in(GvSTASH(dstr)); /* sub foo { 1 } sub bar { 2 } *bar = \&foo */
3810 if (import_flag && !(GvFLAGS(dstr) & import_flag)
3811 && CopSTASH_ne(PL_curcop, GvSTASH(dstr))) {
3812 GvFLAGS(dstr) |= import_flag;
3814 if (stype == SVt_PVHV) {
3815 const char * const name = GvNAME((GV*)dstr);
3816 const STRLEN len = GvNAMELEN(dstr);
3819 (len > 1 && name[len-2] == ':' && name[len-1] == ':')
3820 || (len == 1 && name[0] == ':')
3822 && (!dref || HvENAME_get(dref))
3825 (HV *)sref, (HV *)dref,
3831 stype == SVt_PVAV && sref != dref
3832 && strEQ(GvNAME((GV*)dstr), "ISA")
3833 /* The stash may have been detached from the symbol table, so
3834 check its name before doing anything. */
3835 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3838 MAGIC * const omg = dref && SvSMAGICAL(dref)
3839 ? mg_find(dref, PERL_MAGIC_isa)
3841 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
3842 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
3843 AV * const ary = newAV();
3844 av_push(ary, mg->mg_obj); /* takes the refcount */
3845 mg->mg_obj = (SV *)ary;
3848 if (SvTYPE(omg->mg_obj) == SVt_PVAV) {
3849 SV **svp = AvARRAY((AV *)omg->mg_obj);
3850 I32 items = AvFILLp((AV *)omg->mg_obj) + 1;
3854 SvREFCNT_inc_simple_NN(*svp++)
3860 SvREFCNT_inc_simple_NN(omg->mg_obj)
3864 av_push((AV *)mg->mg_obj,SvREFCNT_inc_simple_NN(dstr));
3869 sref, omg ? omg->mg_obj : dstr, PERL_MAGIC_isa, NULL, 0
3871 mg = mg_find(sref, PERL_MAGIC_isa);
3873 /* Since the *ISA assignment could have affected more than
3874 one stash, don't call mro_isa_changed_in directly, but let
3875 magic_clearisa do it for us, as it already has the logic for
3876 dealing with globs vs arrays of globs. */
3878 Perl_magic_clearisa(aTHX_ NULL, mg);
3880 else if (stype == SVt_PVIO) {
3881 DEBUG_o(Perl_deb(aTHX_ "glob_assign_ref clearing PL_stashcache\n"));
3882 /* It's a cache. It will rebuild itself quite happily.
3883 It's a lot of effort to work out exactly which key (or keys)
3884 might be invalidated by the creation of the this file handle.
3886 hv_clear(PL_stashcache);
3891 if (SvTAINTED(sstr))
3897 Perl_sv_setsv_flags(pTHX_ SV *dstr, register SV* sstr, const I32 flags)
3904 PERL_ARGS_ASSERT_SV_SETSV_FLAGS;
3909 if (SvIS_FREED(dstr)) {
3910 Perl_croak(aTHX_ "panic: attempt to copy value %" SVf
3911 " to a freed scalar %p", SVfARG(sstr), (void *)dstr);
3913 SV_CHECK_THINKFIRST_COW_DROP(dstr);
3915 sstr = &PL_sv_undef;
3916 if (SvIS_FREED(sstr)) {
3917 Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p to %p",
3918 (void*)sstr, (void*)dstr);
3920 stype = SvTYPE(sstr);
3921 dtype = SvTYPE(dstr);
3923 /* There's a lot of redundancy below but we're going for speed here */
3928 if (dtype != SVt_PVGV && dtype != SVt_PVLV) {
3929 (void)SvOK_off(dstr);
3937 sv_upgrade(dstr, SVt_IV);
3941 sv_upgrade(dstr, SVt_PVIV);
3945 goto end_of_first_switch;
3947 (void)SvIOK_only(dstr);
3948 SvIV_set(dstr, SvIVX(sstr));
3951 /* SvTAINTED can only be true if the SV has taint magic, which in
3952 turn means that the SV type is PVMG (or greater). This is the
3953 case statement for SVt_IV, so this cannot be true (whatever gcov
3955 assert(!SvTAINTED(sstr));
3960 if (dtype < SVt_PV && dtype != SVt_IV)
3961 sv_upgrade(dstr, SVt_IV);
3969 sv_upgrade(dstr, SVt_NV);
3973 sv_upgrade(dstr, SVt_PVNV);
3977 goto end_of_first_switch;
3979 SvNV_set(dstr, SvNVX(sstr));
3980 (void)SvNOK_only(dstr);
3981 /* SvTAINTED can only be true if the SV has taint magic, which in
3982 turn means that the SV type is PVMG (or greater). This is the
3983 case statement for SVt_NV, so this cannot be true (whatever gcov
3985 assert(!SvTAINTED(sstr));
3992 sv_upgrade(dstr, SVt_PV);
3995 if (dtype < SVt_PVIV)
3996 sv_upgrade(dstr, SVt_PVIV);
3999 if (dtype < SVt_PVNV)
4000 sv_upgrade(dstr, SVt_PVNV);
4004 const char * const type = sv_reftype(sstr,0);
4006 /* diag_listed_as: Bizarre copy of %s */
4007 Perl_croak(aTHX_ "Bizarre copy of %s in %s", type, OP_DESC(PL_op));
4009 Perl_croak(aTHX_ "Bizarre copy of %s", type);
4014 if (dtype < SVt_REGEXP)
4015 sv_upgrade(dstr, SVt_REGEXP);
4018 /* case SVt_BIND: */
4022 if (SvGMAGICAL(sstr) && (flags & SV_GMAGIC)) {
4024 if (SvTYPE(sstr) != stype)
4025 stype = SvTYPE(sstr);
4027 if (isGV_with_GP(sstr) && dtype <= SVt_PVLV) {
4028 glob_assign_glob(dstr, sstr, dtype);
4031 if (stype == SVt_PVLV)
4032 SvUPGRADE(dstr, SVt_PVNV);
4034 SvUPGRADE(dstr, (svtype)stype);
4036 end_of_first_switch:
4038 /* dstr may have been upgraded. */
4039 dtype = SvTYPE(dstr);
4040 sflags = SvFLAGS(sstr);
4042 if (dtype == SVt_PVCV) {
4043 /* Assigning to a subroutine sets the prototype. */
4046 const char *const ptr = SvPV_const(sstr, len);
4048 SvGROW(dstr, len + 1);
4049 Copy(ptr, SvPVX(dstr), len + 1, char);
4050 SvCUR_set(dstr, len);
4052 SvFLAGS(dstr) |= sflags & SVf_UTF8;
4053 CvAUTOLOAD_off(dstr);
4058 else if (dtype == SVt_PVAV || dtype == SVt_PVHV || dtype == SVt_PVFM) {
4059 const char * const type = sv_reftype(dstr,0);
4061 /* diag_listed_as: Cannot copy to %s */
4062 Perl_croak(aTHX_ "Cannot copy to %s in %s", type, OP_DESC(PL_op));
4064 Perl_croak(aTHX_ "Cannot copy to %s", type);
4065 } else if (sflags & SVf_ROK) {
4066 if (isGV_with_GP(dstr)
4067 && SvTYPE(SvRV(sstr)) == SVt_PVGV && isGV_with_GP(SvRV(sstr))) {
4070 if (GvIMPORTED(dstr) != GVf_IMPORTED
4071 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
4073 GvIMPORTED_on(dstr);
4078 glob_assign_glob(dstr, sstr, dtype);
4082 if (dtype >= SVt_PV) {
4083 if (isGV_with_GP(dstr)) {
4084 glob_assign_ref(dstr, sstr);
4087 if (SvPVX_const(dstr)) {
4093 (void)SvOK_off(dstr);
4094 SvRV_set(dstr, SvREFCNT_inc(SvRV(sstr)));
4095 SvFLAGS(dstr) |= sflags & SVf_ROK;
4096 assert(!(sflags & SVp_NOK));
4097 assert(!(sflags & SVp_IOK));
4098 assert(!(sflags & SVf_NOK));
4099 assert(!(sflags & SVf_IOK));
4101 else if (isGV_with_GP(dstr)) {
4102 if (!(sflags & SVf_OK)) {
4103 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4104 "Undefined value assigned to typeglob");
4107 GV *gv = gv_fetchsv_nomg(sstr, GV_ADD, SVt_PVGV);
4108 if (dstr != (const SV *)gv) {
4109 const char * const name = GvNAME((const GV *)dstr);
4110 const STRLEN len = GvNAMELEN(dstr);
4111 HV *old_stash = NULL;
4112 bool reset_isa = FALSE;
4113 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
4114 || (len == 1 && name[0] == ':')) {
4115 /* Set aside the old stash, so we can reset isa caches
4116 on its subclasses. */
4117 if((old_stash = GvHV(dstr))) {
4118 /* Make sure we do not lose it early. */
4119 SvREFCNT_inc_simple_void_NN(
4120 sv_2mortal((SV *)old_stash)
4127 gp_free(MUTABLE_GV(dstr));
4128 GvGP_set(dstr, gp_ref(GvGP(gv)));
4131 HV * const stash = GvHV(dstr);
4133 old_stash ? (HV *)HvENAME_get(old_stash) : stash
4143 else if (dtype == SVt_REGEXP && stype == SVt_REGEXP) {
4144 reg_temp_copy((REGEXP*)dstr, (REGEXP*)sstr);
4146 else if (sflags & SVp_POK) {
4150 * Check to see if we can just swipe the string. If so, it's a
4151 * possible small lose on short strings, but a big win on long ones.
4152 * It might even be a win on short strings if SvPVX_const(dstr)
4153 * has to be allocated and SvPVX_const(sstr) has to be freed.
4154 * Likewise if we can set up COW rather than doing an actual copy, we
4155 * drop to the else clause, as the swipe code and the COW setup code
4156 * have much in common.
4159 /* Whichever path we take through the next code, we want this true,
4160 and doing it now facilitates the COW check. */
4161 (void)SvPOK_only(dstr);
4164 /* If we're already COW then this clause is not true, and if COW
4165 is allowed then we drop down to the else and make dest COW
4166 with us. If caller hasn't said that we're allowed to COW
4167 shared hash keys then we don't do the COW setup, even if the
4168 source scalar is a shared hash key scalar. */
4169 (((flags & SV_COW_SHARED_HASH_KEYS)
4170 ? (sflags & (SVf_FAKE|SVf_READONLY)) != (SVf_FAKE|SVf_READONLY)
4171 : 1 /* If making a COW copy is forbidden then the behaviour we
4172 desire is as if the source SV isn't actually already
4173 COW, even if it is. So we act as if the source flags
4174 are not COW, rather than actually testing them. */
4176 #ifndef PERL_OLD_COPY_ON_WRITE
4177 /* The change that added SV_COW_SHARED_HASH_KEYS makes the logic
4178 when PERL_OLD_COPY_ON_WRITE is defined a little wrong.
4179 Conceptually PERL_OLD_COPY_ON_WRITE being defined should
4180 override SV_COW_SHARED_HASH_KEYS, because it means "always COW"
4181 but in turn, it's somewhat dead code, never expected to go
4182 live, but more kept as a placeholder on how to do it better
4183 in a newer implementation. */
4184 /* If we are COW and dstr is a suitable target then we drop down
4185 into the else and make dest a COW of us. */
4186 || (SvFLAGS(dstr) & SVf_BREAK)
4191 (sflags & SVs_TEMP) && /* slated for free anyway? */
4192 !(sflags & SVf_OOK) && /* and not involved in OOK hack? */
4193 (!(flags & SV_NOSTEAL)) &&
4194 /* and we're allowed to steal temps */
4195 SvREFCNT(sstr) == 1 && /* and no other references to it? */
4196 SvLEN(sstr)) /* and really is a string */
4197 #ifdef PERL_OLD_COPY_ON_WRITE
4198 && ((flags & SV_COW_SHARED_HASH_KEYS)
4199 ? (!((sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4200 && (SvFLAGS(dstr) & CAN_COW_MASK) == CAN_COW_FLAGS
4201 && SvTYPE(sstr) >= SVt_PVIV))
4205 /* Failed the swipe test, and it's not a shared hash key either.
4206 Have to copy the string. */
4207 STRLEN len = SvCUR(sstr);
4208 SvGROW(dstr, len + 1); /* inlined from sv_setpvn */
4209 Move(SvPVX_const(sstr),SvPVX(dstr),len,char);
4210 SvCUR_set(dstr, len);
4211 *SvEND(dstr) = '\0';
4213 /* If PERL_OLD_COPY_ON_WRITE is not defined, then isSwipe will always
4215 /* Either it's a shared hash key, or it's suitable for
4216 copy-on-write or we can swipe the string. */
4218 PerlIO_printf(Perl_debug_log, "Copy on write: sstr --> dstr\n");
4222 #ifdef PERL_OLD_COPY_ON_WRITE
4224 if ((sflags & (SVf_FAKE | SVf_READONLY))
4225 != (SVf_FAKE | SVf_READONLY)) {
4226 SvREADONLY_on(sstr);
4228 /* Make the source SV into a loop of 1.
4229 (about to become 2) */
4230 SV_COW_NEXT_SV_SET(sstr, sstr);
4234 /* Initial code is common. */
4235 if (SvPVX_const(dstr)) { /* we know that dtype >= SVt_PV */
4240 /* making another shared SV. */
4241 STRLEN cur = SvCUR(sstr);
4242 STRLEN len = SvLEN(sstr);
4243 #ifdef PERL_OLD_COPY_ON_WRITE
4245 assert (SvTYPE(dstr) >= SVt_PVIV);
4246 /* SvIsCOW_normal */
4247 /* splice us in between source and next-after-source. */
4248 SV_COW_NEXT_SV_SET(dstr, SV_COW_NEXT_SV(sstr));
4249 SV_COW_NEXT_SV_SET(sstr, dstr);
4250 SvPV_set(dstr, SvPVX_mutable(sstr));
4254 /* SvIsCOW_shared_hash */
4255 DEBUG_C(PerlIO_printf(Perl_debug_log,
4256 "Copy on write: Sharing hash\n"));
4258 assert (SvTYPE(dstr) >= SVt_PV);
4260 HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)))));
4262 SvLEN_set(dstr, len);
4263 SvCUR_set(dstr, cur);
4264 SvREADONLY_on(dstr);
4268 { /* Passes the swipe test. */
4269 SvPV_set(dstr, SvPVX_mutable(sstr));
4270 SvLEN_set(dstr, SvLEN(sstr));
4271 SvCUR_set(dstr, SvCUR(sstr));
4274 (void)SvOK_off(sstr); /* NOTE: nukes most SvFLAGS on sstr */
4275 SvPV_set(sstr, NULL);
4281 if (sflags & SVp_NOK) {
4282 SvNV_set(dstr, SvNVX(sstr));
4284 if (sflags & SVp_IOK) {
4285 SvIV_set(dstr, SvIVX(sstr));
4286 /* Must do this otherwise some other overloaded use of 0x80000000
4287 gets confused. I guess SVpbm_VALID */
4288 if (sflags & SVf_IVisUV)
4291 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_NOK|SVp_NOK|SVf_UTF8);
4293 const MAGIC * const smg = SvVSTRING_mg(sstr);
4295 sv_magic(dstr, NULL, PERL_MAGIC_vstring,
4296 smg->mg_ptr, smg->mg_len);
4297 SvRMAGICAL_on(dstr);
4301 else if (sflags & (SVp_IOK|SVp_NOK)) {
4302 (void)SvOK_off(dstr);
4303 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_IVisUV|SVf_NOK|SVp_NOK);
4304 if (sflags & SVp_IOK) {
4305 /* XXXX Do we want to set IsUV for IV(ROK)? Be extra safe... */
4306 SvIV_set(dstr, SvIVX(sstr));
4308 if (sflags & SVp_NOK) {
4309 SvNV_set(dstr, SvNVX(sstr));
4313 if (isGV_with_GP(sstr)) {
4314 gv_efullname3(dstr, MUTABLE_GV(sstr), "*");
4317 (void)SvOK_off(dstr);
4319 if (SvTAINTED(sstr))
4324 =for apidoc sv_setsv_mg
4326 Like C<sv_setsv>, but also handles 'set' magic.
4332 Perl_sv_setsv_mg(pTHX_ SV *const dstr, register SV *const sstr)
4334 PERL_ARGS_ASSERT_SV_SETSV_MG;
4336 sv_setsv(dstr,sstr);
4340 #ifdef PERL_OLD_COPY_ON_WRITE
4342 Perl_sv_setsv_cow(pTHX_ SV *dstr, SV *sstr)
4344 STRLEN cur = SvCUR(sstr);
4345 STRLEN len = SvLEN(sstr);
4348 PERL_ARGS_ASSERT_SV_SETSV_COW;
4351 PerlIO_printf(Perl_debug_log, "Fast copy on write: %p -> %p\n",
4352 (void*)sstr, (void*)dstr);
4359 if (SvTHINKFIRST(dstr))
4360 sv_force_normal_flags(dstr, SV_COW_DROP_PV);
4361 else if (SvPVX_const(dstr))
4362 Safefree(SvPVX_mutable(dstr));
4366 SvUPGRADE(dstr, SVt_PVIV);
4368 assert (SvPOK(sstr));
4369 assert (SvPOKp(sstr));
4370 assert (!SvIOK(sstr));
4371 assert (!SvIOKp(sstr));
4372 assert (!SvNOK(sstr));
4373 assert (!SvNOKp(sstr));
4375 if (SvIsCOW(sstr)) {
4377 if (SvLEN(sstr) == 0) {
4378 /* source is a COW shared hash key. */
4379 DEBUG_C(PerlIO_printf(Perl_debug_log,
4380 "Fast copy on write: Sharing hash\n"));
4381 new_pv = HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr))));
4384 SV_COW_NEXT_SV_SET(dstr, SV_COW_NEXT_SV(sstr));
4386 assert ((SvFLAGS(sstr) & CAN_COW_MASK) == CAN_COW_FLAGS);
4387 SvUPGRADE(sstr, SVt_PVIV);
4388 SvREADONLY_on(sstr);
4390 DEBUG_C(PerlIO_printf(Perl_debug_log,
4391 "Fast copy on write: Converting sstr to COW\n"));
4392 SV_COW_NEXT_SV_SET(dstr, sstr);
4394 SV_COW_NEXT_SV_SET(sstr, dstr);
4395 new_pv = SvPVX_mutable(sstr);
4398 SvPV_set(dstr, new_pv);
4399 SvFLAGS(dstr) = (SVt_PVIV|SVf_POK|SVp_POK|SVf_FAKE|SVf_READONLY);
4402 SvLEN_set(dstr, len);
4403 SvCUR_set(dstr, cur);
4412 =for apidoc sv_setpvn
4414 Copies a string into an SV. The C<len> parameter indicates the number of
4415 bytes to be copied. If the C<ptr> argument is NULL the SV will become
4416 undefined. Does not handle 'set' magic. See C<sv_setpvn_mg>.
4422 Perl_sv_setpvn(pTHX_ register SV *const sv, register const char *const ptr, register const STRLEN len)
4427 PERL_ARGS_ASSERT_SV_SETPVN;
4429 SV_CHECK_THINKFIRST_COW_DROP(sv);
4435 /* len is STRLEN which is unsigned, need to copy to signed */
4438 Perl_croak(aTHX_ "panic: sv_setpvn called with negative strlen %"
4441 SvUPGRADE(sv, SVt_PV);
4443 dptr = SvGROW(sv, len + 1);
4444 Move(ptr,dptr,len,char);
4447 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4449 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4453 =for apidoc sv_setpvn_mg
4455 Like C<sv_setpvn>, but also handles 'set' magic.
4461 Perl_sv_setpvn_mg(pTHX_ register SV *const sv, register const char *const ptr, register const STRLEN len)
4463 PERL_ARGS_ASSERT_SV_SETPVN_MG;
4465 sv_setpvn(sv,ptr,len);
4470 =for apidoc sv_setpv
4472 Copies a string into an SV. The string must be null-terminated. Does not
4473 handle 'set' magic. See C<sv_setpv_mg>.
4479 Perl_sv_setpv(pTHX_ register SV *const sv, register const char *const ptr)
4484 PERL_ARGS_ASSERT_SV_SETPV;
4486 SV_CHECK_THINKFIRST_COW_DROP(sv);
4492 SvUPGRADE(sv, SVt_PV);
4494 SvGROW(sv, len + 1);
4495 Move(ptr,SvPVX(sv),len+1,char);
4497 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4499 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4503 =for apidoc sv_setpv_mg
4505 Like C<sv_setpv>, but also handles 'set' magic.
4511 Perl_sv_setpv_mg(pTHX_ register SV *const sv, register const char *const ptr)
4513 PERL_ARGS_ASSERT_SV_SETPV_MG;
4520 Perl_sv_sethek(pTHX_ register SV *const sv, const HEK *const hek)
4524 PERL_ARGS_ASSERT_SV_SETHEK;
4530 if (HEK_LEN(hek) == HEf_SVKEY) {
4531 sv_setsv(sv, *(SV**)HEK_KEY(hek));
4534 const int flags = HEK_FLAGS(hek);
4535 if (flags & HVhek_WASUTF8) {
4536 STRLEN utf8_len = HEK_LEN(hek);
4537 char *as_utf8 = (char *)bytes_to_utf8((U8*)HEK_KEY(hek), &utf8_len);
4538 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
4541 } else if (flags & (HVhek_REHASH|HVhek_UNSHARED)) {
4542 sv_setpvn(sv, HEK_KEY(hek), HEK_LEN(hek));
4545 else SvUTF8_off(sv);
4549 SV_CHECK_THINKFIRST_COW_DROP(sv);
4550 SvUPGRADE(sv, SVt_PV);
4551 Safefree(SvPVX(sv));
4552 SvPV_set(sv,(char *)HEK_KEY(share_hek_hek(hek)));
4553 SvCUR_set(sv, HEK_LEN(hek));
4560 else SvUTF8_off(sv);
4568 =for apidoc sv_usepvn_flags
4570 Tells an SV to use C<ptr> to find its string value. Normally the
4571 string is stored inside the SV but sv_usepvn allows the SV to use an
4572 outside string. The C<ptr> should point to memory that was allocated
4573 by C<malloc>. It must be the start of a mallocked block
4574 of memory, and not a pointer to the middle of it. The
4575 string length, C<len>, must be supplied. By default
4576 this function will realloc (i.e. move) the memory pointed to by C<ptr>,
4577 so that pointer should not be freed or used by the programmer after
4578 giving it to sv_usepvn, and neither should any pointers from "behind"
4579 that pointer (e.g. ptr + 1) be used.
4581 If C<flags> & SV_SMAGIC is true, will call SvSETMAGIC. If C<flags> &
4582 SV_HAS_TRAILING_NUL is true, then C<ptr[len]> must be NUL, and the realloc
4583 will be skipped (i.e. the buffer is actually at least 1 byte longer than
4584 C<len>, and already meets the requirements for storing in C<SvPVX>).
4590 Perl_sv_usepvn_flags(pTHX_ SV *const sv, char *ptr, const STRLEN len, const U32 flags)
4595 PERL_ARGS_ASSERT_SV_USEPVN_FLAGS;
4597 SV_CHECK_THINKFIRST_COW_DROP(sv);
4598 SvUPGRADE(sv, SVt_PV);
4601 if (flags & SV_SMAGIC)
4605 if (SvPVX_const(sv))
4609 if (flags & SV_HAS_TRAILING_NUL)
4610 assert(ptr[len] == '\0');
4613 allocate = (flags & SV_HAS_TRAILING_NUL)
4615 #ifdef Perl_safesysmalloc_size
4618 PERL_STRLEN_ROUNDUP(len + 1);
4620 if (flags & SV_HAS_TRAILING_NUL) {
4621 /* It's long enough - do nothing.
4622 Specifically Perl_newCONSTSUB is relying on this. */
4625 /* Force a move to shake out bugs in callers. */
4626 char *new_ptr = (char*)safemalloc(allocate);
4627 Copy(ptr, new_ptr, len, char);
4628 PoisonFree(ptr,len,char);
4632 ptr = (char*) saferealloc (ptr, allocate);
4635 #ifdef Perl_safesysmalloc_size
4636 SvLEN_set(sv, Perl_safesysmalloc_size(ptr));
4638 SvLEN_set(sv, allocate);
4642 if (!(flags & SV_HAS_TRAILING_NUL)) {
4645 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4647 if (flags & SV_SMAGIC)
4651 #ifdef PERL_OLD_COPY_ON_WRITE
4652 /* Need to do this *after* making the SV normal, as we need the buffer
4653 pointer to remain valid until after we've copied it. If we let go too early,
4654 another thread could invalidate it by unsharing last of the same hash key
4655 (which it can do by means other than releasing copy-on-write Svs)
4656 or by changing the other copy-on-write SVs in the loop. */
4658 S_sv_release_COW(pTHX_ register SV *sv, const char *pvx, SV *after)
4660 PERL_ARGS_ASSERT_SV_RELEASE_COW;
4662 { /* this SV was SvIsCOW_normal(sv) */
4663 /* we need to find the SV pointing to us. */
4664 SV *current = SV_COW_NEXT_SV(after);
4666 if (current == sv) {
4667 /* The SV we point to points back to us (there were only two of us