3 * Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000,
4 * 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009 by Larry Wall
7 * You may distribute under the terms of either the GNU General Public
8 * License or the Artistic License, as specified in the README file.
13 * 'I wonder what the Entish is for "yes" and "no",' he thought.
16 * [p.480 of _The Lord of the Rings_, III/iv: "Treebeard"]
22 * This file contains the code that creates, manipulates and destroys
23 * scalar values (SVs). The other types (AV, HV, GV, etc.) reuse the
24 * structure of an SV, so their creation and destruction is handled
25 * here; higher-level functions are in av.c, hv.c, and so on. Opcode
26 * level functions (eg. substr, split, join) for each of the types are
36 # if __STDC_VERSION__ >= 199901L && !defined(VMS)
47 /* Missing proto on LynxOS */
48 char *gconvert(double, int, int, char *);
51 #ifdef PERL_UTF8_CACHE_ASSERT
52 /* if adding more checks watch out for the following tests:
53 * t/op/index.t t/op/length.t t/op/pat.t t/op/substr.t
54 * lib/utf8.t lib/Unicode/Collate/t/index.t
57 # define ASSERT_UTF8_CACHE(cache) \
58 STMT_START { if (cache) { assert((cache)[0] <= (cache)[1]); \
59 assert((cache)[2] <= (cache)[3]); \
60 assert((cache)[3] <= (cache)[1]);} \
63 # define ASSERT_UTF8_CACHE(cache) NOOP
66 #ifdef PERL_OLD_COPY_ON_WRITE
67 #define SV_COW_NEXT_SV(sv) INT2PTR(SV *,SvUVX(sv))
68 #define SV_COW_NEXT_SV_SET(current,next) SvUV_set(current, PTR2UV(next))
69 /* This is a pessimistic view. Scalar must be purely a read-write PV to copy-
73 /* ============================================================================
75 =head1 Allocation and deallocation of SVs.
77 An SV (or AV, HV, etc.) is allocated in two parts: the head (struct
78 sv, av, hv...) contains type and reference count information, and for
79 many types, a pointer to the body (struct xrv, xpv, xpviv...), which
80 contains fields specific to each type. Some types store all they need
81 in the head, so don't have a body.
83 In all but the most memory-paranoid configurations (ex: PURIFY), heads
84 and bodies are allocated out of arenas, which by default are
85 approximately 4K chunks of memory parcelled up into N heads or bodies.
86 Sv-bodies are allocated by their sv-type, guaranteeing size
87 consistency needed to allocate safely from arrays.
89 For SV-heads, the first slot in each arena is reserved, and holds a
90 link to the next arena, some flags, and a note of the number of slots.
91 Snaked through each arena chain is a linked list of free items; when
92 this becomes empty, an extra arena is allocated and divided up into N
93 items which are threaded into the free list.
95 SV-bodies are similar, but they use arena-sets by default, which
96 separate the link and info from the arena itself, and reclaim the 1st
97 slot in the arena. SV-bodies are further described later.
99 The following global variables are associated with arenas:
101 PL_sv_arenaroot pointer to list of SV arenas
102 PL_sv_root pointer to list of free SV structures
104 PL_body_arenas head of linked-list of body arenas
105 PL_body_roots[] array of pointers to list of free bodies of svtype
106 arrays are indexed by the svtype needed
108 A few special SV heads are not allocated from an arena, but are
109 instead directly created in the interpreter structure, eg PL_sv_undef.
110 The size of arenas can be changed from the default by setting
111 PERL_ARENA_SIZE appropriately at compile time.
113 The SV arena serves the secondary purpose of allowing still-live SVs
114 to be located and destroyed during final cleanup.
116 At the lowest level, the macros new_SV() and del_SV() grab and free
117 an SV head. (If debugging with -DD, del_SV() calls the function S_del_sv()
118 to return the SV to the free list with error checking.) new_SV() calls
119 more_sv() / sv_add_arena() to add an extra arena if the free list is empty.
120 SVs in the free list have their SvTYPE field set to all ones.
122 At the time of very final cleanup, sv_free_arenas() is called from
123 perl_destruct() to physically free all the arenas allocated since the
124 start of the interpreter.
126 The function visit() scans the SV arenas list, and calls a specified
127 function for each SV it finds which is still live - ie which has an SvTYPE
128 other than all 1's, and a non-zero SvREFCNT. visit() is used by the
129 following functions (specified as [function that calls visit()] / [function
130 called by visit() for each SV]):
132 sv_report_used() / do_report_used()
133 dump all remaining SVs (debugging aid)
135 sv_clean_objs() / do_clean_objs(),do_clean_named_objs(),
136 do_clean_named_io_objs()
137 Attempt to free all objects pointed to by RVs,
138 and try to do the same for all objects indirectly
139 referenced by typeglobs too. Called once from
140 perl_destruct(), prior to calling sv_clean_all()
143 sv_clean_all() / do_clean_all()
144 SvREFCNT_dec(sv) each remaining SV, possibly
145 triggering an sv_free(). It also sets the
146 SVf_BREAK flag on the SV to indicate that the
147 refcnt has been artificially lowered, and thus
148 stopping sv_free() from giving spurious warnings
149 about SVs which unexpectedly have a refcnt
150 of zero. called repeatedly from perl_destruct()
151 until there are no SVs left.
153 =head2 Arena allocator API Summary
155 Private API to rest of sv.c
159 new_XPVNV(), del_XPVGV(),
164 sv_report_used(), sv_clean_objs(), sv_clean_all(), sv_free_arenas()
168 * ========================================================================= */
171 * "A time to plant, and a time to uproot what was planted..."
175 # define MEM_LOG_NEW_SV(sv, file, line, func) \
176 Perl_mem_log_new_sv(sv, file, line, func)
177 # define MEM_LOG_DEL_SV(sv, file, line, func) \
178 Perl_mem_log_del_sv(sv, file, line, func)
180 # define MEM_LOG_NEW_SV(sv, file, line, func) NOOP
181 # define MEM_LOG_DEL_SV(sv, file, line, func) NOOP
184 #ifdef DEBUG_LEAKING_SCALARS
185 # define FREE_SV_DEBUG_FILE(sv) Safefree((sv)->sv_debug_file)
186 # define DEBUG_SV_SERIAL(sv) \
187 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) del_SV\n", \
188 PTR2UV(sv), (long)(sv)->sv_debug_serial))
190 # define FREE_SV_DEBUG_FILE(sv)
191 # define DEBUG_SV_SERIAL(sv) NOOP
195 # define SvARENA_CHAIN(sv) ((sv)->sv_u.svu_rv)
196 # define SvARENA_CHAIN_SET(sv,val) (sv)->sv_u.svu_rv = MUTABLE_SV((val))
197 /* Whilst I'd love to do this, it seems that things like to check on
199 # define POSION_SV_HEAD(sv) PoisonNew(sv, 1, struct STRUCT_SV)
201 # define POSION_SV_HEAD(sv) PoisonNew(&SvANY(sv), 1, void *), \
202 PoisonNew(&SvREFCNT(sv), 1, U32)
204 # define SvARENA_CHAIN(sv) SvANY(sv)
205 # define SvARENA_CHAIN_SET(sv,val) SvANY(sv) = (void *)(val)
206 # define POSION_SV_HEAD(sv)
209 /* Mark an SV head as unused, and add to free list.
211 * If SVf_BREAK is set, skip adding it to the free list, as this SV had
212 * its refcount artificially decremented during global destruction, so
213 * there may be dangling pointers to it. The last thing we want in that
214 * case is for it to be reused. */
216 #define plant_SV(p) \
218 const U32 old_flags = SvFLAGS(p); \
219 MEM_LOG_DEL_SV(p, __FILE__, __LINE__, FUNCTION__); \
220 DEBUG_SV_SERIAL(p); \
221 FREE_SV_DEBUG_FILE(p); \
223 SvFLAGS(p) = SVTYPEMASK; \
224 if (!(old_flags & SVf_BREAK)) { \
225 SvARENA_CHAIN_SET(p, PL_sv_root); \
231 #define uproot_SV(p) \
234 PL_sv_root = MUTABLE_SV(SvARENA_CHAIN(p)); \
239 /* make some more SVs by adding another arena */
246 char *chunk; /* must use New here to match call to */
247 Newx(chunk,PERL_ARENA_SIZE,char); /* Safefree() in sv_free_arenas() */
248 sv_add_arena(chunk, PERL_ARENA_SIZE, 0);
253 /* new_SV(): return a new, empty SV head */
255 #ifdef DEBUG_LEAKING_SCALARS
256 /* provide a real function for a debugger to play with */
258 S_new_SV(pTHX_ const char *file, int line, const char *func)
265 sv = S_more_sv(aTHX);
269 sv->sv_debug_optype = PL_op ? PL_op->op_type : 0;
270 sv->sv_debug_line = (U16) (PL_parser && PL_parser->copline != NOLINE
276 sv->sv_debug_inpad = 0;
277 sv->sv_debug_parent = NULL;
278 sv->sv_debug_file = PL_curcop ? savepv(CopFILE(PL_curcop)): NULL;
280 sv->sv_debug_serial = PL_sv_serial++;
282 MEM_LOG_NEW_SV(sv, file, line, func);
283 DEBUG_m(PerlIO_printf(Perl_debug_log, "0x%"UVxf": (%05ld) new_SV (from %s:%d [%s])\n",
284 PTR2UV(sv), (long)sv->sv_debug_serial, file, line, func));
288 # define new_SV(p) (p)=S_new_SV(aTHX_ __FILE__, __LINE__, FUNCTION__)
296 (p) = S_more_sv(aTHX); \
300 MEM_LOG_NEW_SV(p, __FILE__, __LINE__, FUNCTION__); \
305 /* del_SV(): return an empty SV head to the free list */
318 S_del_sv(pTHX_ SV *p)
322 PERL_ARGS_ASSERT_DEL_SV;
327 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
328 const SV * const sv = sva + 1;
329 const SV * const svend = &sva[SvREFCNT(sva)];
330 if (p >= sv && p < svend) {
336 Perl_ck_warner_d(aTHX_ packWARN(WARN_INTERNAL),
337 "Attempt to free non-arena SV: 0x%"UVxf
338 pTHX__FORMAT, PTR2UV(p) pTHX__VALUE);
345 #else /* ! DEBUGGING */
347 #define del_SV(p) plant_SV(p)
349 #endif /* DEBUGGING */
353 =head1 SV Manipulation Functions
355 =for apidoc sv_add_arena
357 Given a chunk of memory, link it to the head of the list of arenas,
358 and split it into a list of free SVs.
364 S_sv_add_arena(pTHX_ char *const ptr, const U32 size, const U32 flags)
367 SV *const sva = MUTABLE_SV(ptr);
371 PERL_ARGS_ASSERT_SV_ADD_ARENA;
373 /* The first SV in an arena isn't an SV. */
374 SvANY(sva) = (void *) PL_sv_arenaroot; /* ptr to next arena */
375 SvREFCNT(sva) = size / sizeof(SV); /* number of SV slots */
376 SvFLAGS(sva) = flags; /* FAKE if not to be freed */
378 PL_sv_arenaroot = sva;
379 PL_sv_root = sva + 1;
381 svend = &sva[SvREFCNT(sva) - 1];
384 SvARENA_CHAIN_SET(sv, (sv + 1));
388 /* Must always set typemask because it's always checked in on cleanup
389 when the arenas are walked looking for objects. */
390 SvFLAGS(sv) = SVTYPEMASK;
393 SvARENA_CHAIN_SET(sv, 0);
397 SvFLAGS(sv) = SVTYPEMASK;
400 /* visit(): call the named function for each non-free SV in the arenas
401 * whose flags field matches the flags/mask args. */
404 S_visit(pTHX_ SVFUNC_t f, const U32 flags, const U32 mask)
410 PERL_ARGS_ASSERT_VISIT;
412 for (sva = PL_sv_arenaroot; sva; sva = MUTABLE_SV(SvANY(sva))) {
413 register const SV * const svend = &sva[SvREFCNT(sva)];
415 for (sv = sva + 1; sv < svend; ++sv) {
416 if (SvTYPE(sv) != (svtype)SVTYPEMASK
417 && (sv->sv_flags & mask) == flags
430 /* called by sv_report_used() for each live SV */
433 do_report_used(pTHX_ SV *const sv)
435 if (SvTYPE(sv) != (svtype)SVTYPEMASK) {
436 PerlIO_printf(Perl_debug_log, "****\n");
443 =for apidoc sv_report_used
445 Dump the contents of all SVs not yet freed (debugging aid).
451 Perl_sv_report_used(pTHX)
454 visit(do_report_used, 0, 0);
460 /* called by sv_clean_objs() for each live SV */
463 do_clean_objs(pTHX_ SV *const ref)
468 SV * const target = SvRV(ref);
469 if (SvOBJECT(target)) {
470 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning object ref:\n "), sv_dump(ref)));
471 if (SvWEAKREF(ref)) {
472 sv_del_backref(target, ref);
478 SvREFCNT_dec(target);
483 /* XXX Might want to check arrays, etc. */
487 /* clear any slots in a GV which hold objects - except IO;
488 * called by sv_clean_objs() for each live GV */
491 do_clean_named_objs(pTHX_ SV *const sv)
495 assert(SvTYPE(sv) == SVt_PVGV);
496 assert(isGV_with_GP(sv));
500 /* freeing GP entries may indirectly free the current GV;
501 * hold onto it while we mess with the GP slots */
504 if ( ((obj = GvSV(sv) )) && SvOBJECT(obj)) {
505 DEBUG_D((PerlIO_printf(Perl_debug_log,
506 "Cleaning named glob SV object:\n "), sv_dump(obj)));
510 if ( ((obj = MUTABLE_SV(GvAV(sv)) )) && SvOBJECT(obj)) {
511 DEBUG_D((PerlIO_printf(Perl_debug_log,
512 "Cleaning named glob AV object:\n "), sv_dump(obj)));
516 if ( ((obj = MUTABLE_SV(GvHV(sv)) )) && SvOBJECT(obj)) {
517 DEBUG_D((PerlIO_printf(Perl_debug_log,
518 "Cleaning named glob HV object:\n "), sv_dump(obj)));
522 if ( ((obj = MUTABLE_SV(GvCV(sv)) )) && SvOBJECT(obj)) {
523 DEBUG_D((PerlIO_printf(Perl_debug_log,
524 "Cleaning named glob CV object:\n "), sv_dump(obj)));
528 SvREFCNT_dec(sv); /* undo the inc above */
531 /* clear any IO slots in a GV which hold objects (except stderr, defout);
532 * called by sv_clean_objs() for each live GV */
535 do_clean_named_io_objs(pTHX_ SV *const sv)
539 assert(SvTYPE(sv) == SVt_PVGV);
540 assert(isGV_with_GP(sv));
541 if (!GvGP(sv) || sv == (SV*)PL_stderrgv || sv == (SV*)PL_defoutgv)
545 if ( ((obj = MUTABLE_SV(GvIO(sv)) )) && SvOBJECT(obj)) {
546 DEBUG_D((PerlIO_printf(Perl_debug_log,
547 "Cleaning named glob IO object:\n "), sv_dump(obj)));
551 SvREFCNT_dec(sv); /* undo the inc above */
554 /* Void wrapper to pass to visit() */
556 do_curse(pTHX_ SV * const sv) {
557 if ((PL_stderrgv && GvGP(PL_stderrgv) && (SV*)GvIO(PL_stderrgv) == sv)
558 || (PL_defoutgv && GvGP(PL_defoutgv) && (SV*)GvIO(PL_defoutgv) == sv))
564 =for apidoc sv_clean_objs
566 Attempt to destroy all objects not yet freed.
572 Perl_sv_clean_objs(pTHX)
576 PL_in_clean_objs = TRUE;
577 visit(do_clean_objs, SVf_ROK, SVf_ROK);
578 /* Some barnacles may yet remain, clinging to typeglobs.
579 * Run the non-IO destructors first: they may want to output
580 * error messages, close files etc */
581 visit(do_clean_named_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
582 visit(do_clean_named_io_objs, SVt_PVGV|SVpgv_GP, SVTYPEMASK|SVp_POK|SVpgv_GP);
583 /* And if there are some very tenacious barnacles clinging to arrays,
584 closures, or what have you.... */
585 visit(do_curse, SVs_OBJECT, SVs_OBJECT);
586 olddef = PL_defoutgv;
587 PL_defoutgv = NULL; /* disable skip of PL_defoutgv */
588 if (olddef && isGV_with_GP(olddef))
589 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olddef));
590 olderr = PL_stderrgv;
591 PL_stderrgv = NULL; /* disable skip of PL_stderrgv */
592 if (olderr && isGV_with_GP(olderr))
593 do_clean_named_io_objs(aTHX_ MUTABLE_SV(olderr));
594 SvREFCNT_dec(olddef);
595 PL_in_clean_objs = FALSE;
598 /* called by sv_clean_all() for each live SV */
601 do_clean_all(pTHX_ SV *const sv)
604 if (sv == (const SV *) PL_fdpid || sv == (const SV *)PL_strtab) {
605 /* don't clean pid table and strtab */
608 DEBUG_D((PerlIO_printf(Perl_debug_log, "Cleaning loops: SV at 0x%"UVxf"\n", PTR2UV(sv)) ));
609 SvFLAGS(sv) |= SVf_BREAK;
614 =for apidoc sv_clean_all
616 Decrement the refcnt of each remaining SV, possibly triggering a
617 cleanup. This function may have to be called multiple times to free
618 SVs which are in complex self-referential hierarchies.
624 Perl_sv_clean_all(pTHX)
628 PL_in_clean_all = TRUE;
629 cleaned = visit(do_clean_all, 0,0);
634 ARENASETS: a meta-arena implementation which separates arena-info
635 into struct arena_set, which contains an array of struct
636 arena_descs, each holding info for a single arena. By separating
637 the meta-info from the arena, we recover the 1st slot, formerly
638 borrowed for list management. The arena_set is about the size of an
639 arena, avoiding the needless malloc overhead of a naive linked-list.
641 The cost is 1 arena-set malloc per ~320 arena-mallocs, + the unused
642 memory in the last arena-set (1/2 on average). In trade, we get
643 back the 1st slot in each arena (ie 1.7% of a CV-arena, less for
644 smaller types). The recovery of the wasted space allows use of
645 small arenas for large, rare body types, by changing array* fields
646 in body_details_by_type[] below.
649 char *arena; /* the raw storage, allocated aligned */
650 size_t size; /* its size ~4k typ */
651 svtype utype; /* bodytype stored in arena */
656 /* Get the maximum number of elements in set[] such that struct arena_set
657 will fit within PERL_ARENA_SIZE, which is probably just under 4K, and
658 therefore likely to be 1 aligned memory page. */
660 #define ARENAS_PER_SET ((PERL_ARENA_SIZE - sizeof(struct arena_set*) \
661 - 2 * sizeof(int)) / sizeof (struct arena_desc))
664 struct arena_set* next;
665 unsigned int set_size; /* ie ARENAS_PER_SET */
666 unsigned int curr; /* index of next available arena-desc */
667 struct arena_desc set[ARENAS_PER_SET];
671 =for apidoc sv_free_arenas
673 Deallocate the memory used by all arenas. Note that all the individual SV
674 heads and bodies within the arenas must already have been freed.
679 Perl_sv_free_arenas(pTHX)
686 /* Free arenas here, but be careful about fake ones. (We assume
687 contiguity of the fake ones with the corresponding real ones.) */
689 for (sva = PL_sv_arenaroot; sva; sva = svanext) {
690 svanext = MUTABLE_SV(SvANY(sva));
691 while (svanext && SvFAKE(svanext))
692 svanext = MUTABLE_SV(SvANY(svanext));
699 struct arena_set *aroot = (struct arena_set*) PL_body_arenas;
702 struct arena_set *current = aroot;
705 assert(aroot->set[i].arena);
706 Safefree(aroot->set[i].arena);
714 i = PERL_ARENA_ROOTS_SIZE;
716 PL_body_roots[i] = 0;
723 Here are mid-level routines that manage the allocation of bodies out
724 of the various arenas. There are 5 kinds of arenas:
726 1. SV-head arenas, which are discussed and handled above
727 2. regular body arenas
728 3. arenas for reduced-size bodies
731 Arena types 2 & 3 are chained by body-type off an array of
732 arena-root pointers, which is indexed by svtype. Some of the
733 larger/less used body types are malloced singly, since a large
734 unused block of them is wasteful. Also, several svtypes dont have
735 bodies; the data fits into the sv-head itself. The arena-root
736 pointer thus has a few unused root-pointers (which may be hijacked
737 later for arena types 4,5)
739 3 differs from 2 as an optimization; some body types have several
740 unused fields in the front of the structure (which are kept in-place
741 for consistency). These bodies can be allocated in smaller chunks,
742 because the leading fields arent accessed. Pointers to such bodies
743 are decremented to point at the unused 'ghost' memory, knowing that
744 the pointers are used with offsets to the real memory.
747 =head1 SV-Body Allocation
749 Allocation of SV-bodies is similar to SV-heads, differing as follows;
750 the allocation mechanism is used for many body types, so is somewhat
751 more complicated, it uses arena-sets, and has no need for still-live
754 At the outermost level, (new|del)_X*V macros return bodies of the
755 appropriate type. These macros call either (new|del)_body_type or
756 (new|del)_body_allocated macro pairs, depending on specifics of the
757 type. Most body types use the former pair, the latter pair is used to
758 allocate body types with "ghost fields".
760 "ghost fields" are fields that are unused in certain types, and
761 consequently don't need to actually exist. They are declared because
762 they're part of a "base type", which allows use of functions as
763 methods. The simplest examples are AVs and HVs, 2 aggregate types
764 which don't use the fields which support SCALAR semantics.
766 For these types, the arenas are carved up into appropriately sized
767 chunks, we thus avoid wasted memory for those unaccessed members.
768 When bodies are allocated, we adjust the pointer back in memory by the
769 size of the part not allocated, so it's as if we allocated the full
770 structure. (But things will all go boom if you write to the part that
771 is "not there", because you'll be overwriting the last members of the
772 preceding structure in memory.)
774 We calculate the correction using the STRUCT_OFFSET macro on the first
775 member present. If the allocated structure is smaller (no initial NV
776 actually allocated) then the net effect is to subtract the size of the NV
777 from the pointer, to return a new pointer as if an initial NV were actually
778 allocated. (We were using structures named *_allocated for this, but
779 this turned out to be a subtle bug, because a structure without an NV
780 could have a lower alignment constraint, but the compiler is allowed to
781 optimised accesses based on the alignment constraint of the actual pointer
782 to the full structure, for example, using a single 64 bit load instruction
783 because it "knows" that two adjacent 32 bit members will be 8-byte aligned.)
785 This is the same trick as was used for NV and IV bodies. Ironically it
786 doesn't need to be used for NV bodies any more, because NV is now at
787 the start of the structure. IV bodies don't need it either, because
788 they are no longer allocated.
790 In turn, the new_body_* allocators call S_new_body(), which invokes
791 new_body_inline macro, which takes a lock, and takes a body off the
792 linked list at PL_body_roots[sv_type], calling Perl_more_bodies() if
793 necessary to refresh an empty list. Then the lock is released, and
794 the body is returned.
796 Perl_more_bodies allocates a new arena, and carves it up into an array of N
797 bodies, which it strings into a linked list. It looks up arena-size
798 and body-size from the body_details table described below, thus
799 supporting the multiple body-types.
801 If PURIFY is defined, or PERL_ARENA_SIZE=0, arenas are not used, and
802 the (new|del)_X*V macros are mapped directly to malloc/free.
804 For each sv-type, struct body_details bodies_by_type[] carries
805 parameters which control these aspects of SV handling:
807 Arena_size determines whether arenas are used for this body type, and if
808 so, how big they are. PURIFY or PERL_ARENA_SIZE=0 set this field to
809 zero, forcing individual mallocs and frees.
811 Body_size determines how big a body is, and therefore how many fit into
812 each arena. Offset carries the body-pointer adjustment needed for
813 "ghost fields", and is used in *_allocated macros.
815 But its main purpose is to parameterize info needed in
816 Perl_sv_upgrade(). The info here dramatically simplifies the function
817 vs the implementation in 5.8.8, making it table-driven. All fields
818 are used for this, except for arena_size.
820 For the sv-types that have no bodies, arenas are not used, so those
821 PL_body_roots[sv_type] are unused, and can be overloaded. In
822 something of a special case, SVt_NULL is borrowed for HE arenas;
823 PL_body_roots[HE_SVSLOT=SVt_NULL] is filled by S_more_he, but the
824 bodies_by_type[SVt_NULL] slot is not used, as the table is not
829 struct body_details {
830 U8 body_size; /* Size to allocate */
831 U8 copy; /* Size of structure to copy (may be shorter) */
833 unsigned int type : 4; /* We have space for a sanity check. */
834 unsigned int cant_upgrade : 1; /* Cannot upgrade this type */
835 unsigned int zero_nv : 1; /* zero the NV when upgrading from this */
836 unsigned int arena : 1; /* Allocated from an arena */
837 size_t arena_size; /* Size of arena to allocate */
845 /* With -DPURFIY we allocate everything directly, and don't use arenas.
846 This seems a rather elegant way to simplify some of the code below. */
847 #define HASARENA FALSE
849 #define HASARENA TRUE
851 #define NOARENA FALSE
853 /* Size the arenas to exactly fit a given number of bodies. A count
854 of 0 fits the max number bodies into a PERL_ARENA_SIZE.block,
855 simplifying the default. If count > 0, the arena is sized to fit
856 only that many bodies, allowing arenas to be used for large, rare
857 bodies (XPVFM, XPVIO) without undue waste. The arena size is
858 limited by PERL_ARENA_SIZE, so we can safely oversize the
861 #define FIT_ARENA0(body_size) \
862 ((size_t)(PERL_ARENA_SIZE / body_size) * body_size)
863 #define FIT_ARENAn(count,body_size) \
864 ( count * body_size <= PERL_ARENA_SIZE) \
865 ? count * body_size \
866 : FIT_ARENA0 (body_size)
867 #define FIT_ARENA(count,body_size) \
869 ? FIT_ARENAn (count, body_size) \
870 : FIT_ARENA0 (body_size)
872 /* Calculate the length to copy. Specifically work out the length less any
873 final padding the compiler needed to add. See the comment in sv_upgrade
874 for why copying the padding proved to be a bug. */
876 #define copy_length(type, last_member) \
877 STRUCT_OFFSET(type, last_member) \
878 + sizeof (((type*)SvANY((const SV *)0))->last_member)
880 static const struct body_details bodies_by_type[] = {
881 /* HEs use this offset for their arena. */
882 { 0, 0, 0, SVt_NULL, FALSE, NONV, NOARENA, 0 },
884 /* The bind placeholder pretends to be an RV for now.
885 Also it's marked as "can't upgrade" to stop anyone using it before it's
887 { 0, 0, 0, SVt_BIND, TRUE, NONV, NOARENA, 0 },
889 /* IVs are in the head, so the allocation size is 0. */
891 sizeof(IV), /* This is used to copy out the IV body. */
892 STRUCT_OFFSET(XPVIV, xiv_iv), SVt_IV, FALSE, NONV,
893 NOARENA /* IVS don't need an arena */, 0
896 { sizeof(NV), sizeof(NV),
897 STRUCT_OFFSET(XPVNV, xnv_u),
898 SVt_NV, FALSE, HADNV, HASARENA, FIT_ARENA(0, sizeof(NV)) },
900 { sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur),
901 copy_length(XPV, xpv_len) - STRUCT_OFFSET(XPV, xpv_cur),
902 + STRUCT_OFFSET(XPV, xpv_cur),
903 SVt_PV, FALSE, NONV, HASARENA,
904 FIT_ARENA(0, sizeof(XPV) - STRUCT_OFFSET(XPV, xpv_cur)) },
906 { sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur),
907 copy_length(XPVIV, xiv_u) - STRUCT_OFFSET(XPV, xpv_cur),
908 + STRUCT_OFFSET(XPV, xpv_cur),
909 SVt_PVIV, FALSE, NONV, HASARENA,
910 FIT_ARENA(0, sizeof(XPVIV) - STRUCT_OFFSET(XPV, xpv_cur)) },
912 { sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur),
913 copy_length(XPVNV, xnv_u) - STRUCT_OFFSET(XPV, xpv_cur),
914 + STRUCT_OFFSET(XPV, xpv_cur),
915 SVt_PVNV, FALSE, HADNV, HASARENA,
916 FIT_ARENA(0, sizeof(XPVNV) - STRUCT_OFFSET(XPV, xpv_cur)) },
918 { sizeof(XPVMG), copy_length(XPVMG, xnv_u), 0, SVt_PVMG, FALSE, HADNV,
919 HASARENA, FIT_ARENA(0, sizeof(XPVMG)) },
924 SVt_REGEXP, FALSE, NONV, HASARENA,
925 FIT_ARENA(0, sizeof(regexp))
928 { sizeof(XPVGV), sizeof(XPVGV), 0, SVt_PVGV, TRUE, HADNV,
929 HASARENA, FIT_ARENA(0, sizeof(XPVGV)) },
931 { sizeof(XPVLV), sizeof(XPVLV), 0, SVt_PVLV, TRUE, HADNV,
932 HASARENA, FIT_ARENA(0, sizeof(XPVLV)) },
935 copy_length(XPVAV, xav_alloc),
937 SVt_PVAV, TRUE, NONV, HASARENA,
938 FIT_ARENA(0, sizeof(XPVAV)) },
941 copy_length(XPVHV, xhv_max),
943 SVt_PVHV, TRUE, NONV, HASARENA,
944 FIT_ARENA(0, sizeof(XPVHV)) },
949 SVt_PVCV, TRUE, NONV, HASARENA,
950 FIT_ARENA(0, sizeof(XPVCV)) },
955 SVt_PVFM, TRUE, NONV, NOARENA,
956 FIT_ARENA(20, sizeof(XPVFM)) },
961 SVt_PVIO, TRUE, NONV, HASARENA,
962 FIT_ARENA(24, sizeof(XPVIO)) },
965 #define new_body_allocated(sv_type) \
966 (void *)((char *)S_new_body(aTHX_ sv_type) \
967 - bodies_by_type[sv_type].offset)
969 /* return a thing to the free list */
971 #define del_body(thing, root) \
973 void ** const thing_copy = (void **)thing; \
974 *thing_copy = *root; \
975 *root = (void*)thing_copy; \
980 #define new_XNV() safemalloc(sizeof(XPVNV))
981 #define new_XPVNV() safemalloc(sizeof(XPVNV))
982 #define new_XPVMG() safemalloc(sizeof(XPVMG))
984 #define del_XPVGV(p) safefree(p)
988 #define new_XNV() new_body_allocated(SVt_NV)
989 #define new_XPVNV() new_body_allocated(SVt_PVNV)
990 #define new_XPVMG() new_body_allocated(SVt_PVMG)
992 #define del_XPVGV(p) del_body(p + bodies_by_type[SVt_PVGV].offset, \
993 &PL_body_roots[SVt_PVGV])
997 /* no arena for you! */
999 #define new_NOARENA(details) \
1000 safemalloc((details)->body_size + (details)->offset)
1001 #define new_NOARENAZ(details) \
1002 safecalloc((details)->body_size + (details)->offset, 1)
1005 Perl_more_bodies (pTHX_ const svtype sv_type, const size_t body_size,
1006 const size_t arena_size)
1009 void ** const root = &PL_body_roots[sv_type];
1010 struct arena_desc *adesc;
1011 struct arena_set *aroot = (struct arena_set *) PL_body_arenas;
1015 const size_t good_arena_size = Perl_malloc_good_size(arena_size);
1016 #if defined(DEBUGGING) && !defined(PERL_GLOBAL_STRUCT_PRIVATE)
1017 static bool done_sanity_check;
1019 /* PERL_GLOBAL_STRUCT_PRIVATE cannot coexist with global
1020 * variables like done_sanity_check. */
1021 if (!done_sanity_check) {
1022 unsigned int i = SVt_LAST;
1024 done_sanity_check = TRUE;
1027 assert (bodies_by_type[i].type == i);
1033 /* may need new arena-set to hold new arena */
1034 if (!aroot || aroot->curr >= aroot->set_size) {
1035 struct arena_set *newroot;
1036 Newxz(newroot, 1, struct arena_set);
1037 newroot->set_size = ARENAS_PER_SET;
1038 newroot->next = aroot;
1040 PL_body_arenas = (void *) newroot;
1041 DEBUG_m(PerlIO_printf(Perl_debug_log, "new arenaset %p\n", (void*)aroot));
1044 /* ok, now have arena-set with at least 1 empty/available arena-desc */
1045 curr = aroot->curr++;
1046 adesc = &(aroot->set[curr]);
1047 assert(!adesc->arena);
1049 Newx(adesc->arena, good_arena_size, char);
1050 adesc->size = good_arena_size;
1051 adesc->utype = sv_type;
1052 DEBUG_m(PerlIO_printf(Perl_debug_log, "arena %d added: %p size %"UVuf"\n",
1053 curr, (void*)adesc->arena, (UV)good_arena_size));
1055 start = (char *) adesc->arena;
1057 /* Get the address of the byte after the end of the last body we can fit.
1058 Remember, this is integer division: */
1059 end = start + good_arena_size / body_size * body_size;
1061 /* computed count doesn't reflect the 1st slot reservation */
1062 #if defined(MYMALLOC) || defined(HAS_MALLOC_GOOD_SIZE)
1063 DEBUG_m(PerlIO_printf(Perl_debug_log,
1064 "arena %p end %p arena-size %d (from %d) type %d "
1066 (void*)start, (void*)end, (int)good_arena_size,
1067 (int)arena_size, sv_type, (int)body_size,
1068 (int)good_arena_size / (int)body_size));
1070 DEBUG_m(PerlIO_printf(Perl_debug_log,
1071 "arena %p end %p arena-size %d type %d size %d ct %d\n",
1072 (void*)start, (void*)end,
1073 (int)arena_size, sv_type, (int)body_size,
1074 (int)good_arena_size / (int)body_size));
1076 *root = (void *)start;
1079 /* Where the next body would start: */
1080 char * const next = start + body_size;
1083 /* This is the last body: */
1084 assert(next == end);
1086 *(void **)start = 0;
1090 *(void**) start = (void *)next;
1095 /* grab a new thing from the free list, allocating more if necessary.
1096 The inline version is used for speed in hot routines, and the
1097 function using it serves the rest (unless PURIFY).
1099 #define new_body_inline(xpv, sv_type) \
1101 void ** const r3wt = &PL_body_roots[sv_type]; \
1102 xpv = (PTR_TBL_ENT_t*) (*((void **)(r3wt)) \
1103 ? *((void **)(r3wt)) : Perl_more_bodies(aTHX_ sv_type, \
1104 bodies_by_type[sv_type].body_size,\
1105 bodies_by_type[sv_type].arena_size)); \
1106 *(r3wt) = *(void**)(xpv); \
1112 S_new_body(pTHX_ const svtype sv_type)
1116 new_body_inline(xpv, sv_type);
1122 static const struct body_details fake_rv =
1123 { 0, 0, 0, SVt_IV, FALSE, NONV, NOARENA, 0 };
1126 =for apidoc sv_upgrade
1128 Upgrade an SV to a more complex form. Generally adds a new body type to the
1129 SV, then copies across as much information as possible from the old body.
1130 It croaks if the SV is already in a more complex form than requested. You
1131 generally want to use the C<SvUPGRADE> macro wrapper, which checks the type
1132 before calling C<sv_upgrade>, and hence does not croak. See also
1139 Perl_sv_upgrade(pTHX_ register SV *const sv, svtype new_type)
1144 const svtype old_type = SvTYPE(sv);
1145 const struct body_details *new_type_details;
1146 const struct body_details *old_type_details
1147 = bodies_by_type + old_type;
1148 SV *referant = NULL;
1150 PERL_ARGS_ASSERT_SV_UPGRADE;
1152 if (old_type == new_type)
1155 /* This clause was purposefully added ahead of the early return above to
1156 the shared string hackery for (sort {$a <=> $b} keys %hash), with the
1157 inference by Nick I-S that it would fix other troublesome cases. See
1158 changes 7162, 7163 (f130fd4589cf5fbb24149cd4db4137c8326f49c1 and parent)
1160 Given that shared hash key scalars are no longer PVIV, but PV, there is
1161 no longer need to unshare so as to free up the IVX slot for its proper
1162 purpose. So it's safe to move the early return earlier. */
1164 if (new_type != SVt_PV && SvIsCOW(sv)) {
1165 sv_force_normal_flags(sv, 0);
1168 old_body = SvANY(sv);
1170 /* Copying structures onto other structures that have been neatly zeroed
1171 has a subtle gotcha. Consider XPVMG
1173 +------+------+------+------+------+-------+-------+
1174 | NV | CUR | LEN | IV | MAGIC | STASH |
1175 +------+------+------+------+------+-------+-------+
1176 0 4 8 12 16 20 24 28
1178 where NVs are aligned to 8 bytes, so that sizeof that structure is
1179 actually 32 bytes long, with 4 bytes of padding at the end:
1181 +------+------+------+------+------+-------+-------+------+
1182 | NV | CUR | LEN | IV | MAGIC | STASH | ??? |
1183 +------+------+------+------+------+-------+-------+------+
1184 0 4 8 12 16 20 24 28 32
1186 so what happens if you allocate memory for this structure:
1188 +------+------+------+------+------+-------+-------+------+------+...
1189 | NV | CUR | LEN | IV | MAGIC | STASH | GP | NAME |
1190 +------+------+------+------+------+-------+-------+------+------+...
1191 0 4 8 12 16 20 24 28 32 36
1193 zero it, then copy sizeof(XPVMG) bytes on top of it? Not quite what you
1194 expect, because you copy the area marked ??? onto GP. Now, ??? may have
1195 started out as zero once, but it's quite possible that it isn't. So now,
1196 rather than a nicely zeroed GP, you have it pointing somewhere random.
1199 (In fact, GP ends up pointing at a previous GP structure, because the
1200 principle cause of the padding in XPVMG getting garbage is a copy of
1201 sizeof(XPVMG) bytes from a XPVGV structure in sv_unglob. Right now
1202 this happens to be moot because XPVGV has been re-ordered, with GP
1203 no longer after STASH)
1205 So we are careful and work out the size of used parts of all the
1213 referant = SvRV(sv);
1214 old_type_details = &fake_rv;
1215 if (new_type == SVt_NV)
1216 new_type = SVt_PVNV;
1218 if (new_type < SVt_PVIV) {
1219 new_type = (new_type == SVt_NV)
1220 ? SVt_PVNV : SVt_PVIV;
1225 if (new_type < SVt_PVNV) {
1226 new_type = SVt_PVNV;
1230 assert(new_type > SVt_PV);
1231 assert(SVt_IV < SVt_PV);
1232 assert(SVt_NV < SVt_PV);
1239 /* Because the XPVMG of PL_mess_sv isn't allocated from the arena,
1240 there's no way that it can be safely upgraded, because perl.c
1241 expects to Safefree(SvANY(PL_mess_sv)) */
1242 assert(sv != PL_mess_sv);
1243 /* This flag bit is used to mean other things in other scalar types.
1244 Given that it only has meaning inside the pad, it shouldn't be set
1245 on anything that can get upgraded. */
1246 assert(!SvPAD_TYPED(sv));
1249 if (old_type_details->cant_upgrade)
1250 Perl_croak(aTHX_ "Can't upgrade %s (%" UVuf ") to %" UVuf,
1251 sv_reftype(sv, 0), (UV) old_type, (UV) new_type);
1254 if (old_type > new_type)
1255 Perl_croak(aTHX_ "sv_upgrade from type %d down to type %d",
1256 (int)old_type, (int)new_type);
1258 new_type_details = bodies_by_type + new_type;
1260 SvFLAGS(sv) &= ~SVTYPEMASK;
1261 SvFLAGS(sv) |= new_type;
1263 /* This can't happen, as SVt_NULL is <= all values of new_type, so one of
1264 the return statements above will have triggered. */
1265 assert (new_type != SVt_NULL);
1268 assert(old_type == SVt_NULL);
1269 SvANY(sv) = (XPVIV*)((char*)&(sv->sv_u.svu_iv) - STRUCT_OFFSET(XPVIV, xiv_iv));
1273 assert(old_type == SVt_NULL);
1274 SvANY(sv) = new_XNV();
1279 assert(new_type_details->body_size);
1282 assert(new_type_details->arena);
1283 assert(new_type_details->arena_size);
1284 /* This points to the start of the allocated area. */
1285 new_body_inline(new_body, new_type);
1286 Zero(new_body, new_type_details->body_size, char);
1287 new_body = ((char *)new_body) - new_type_details->offset;
1289 /* We always allocated the full length item with PURIFY. To do this
1290 we fake things so that arena is false for all 16 types.. */
1291 new_body = new_NOARENAZ(new_type_details);
1293 SvANY(sv) = new_body;
1294 if (new_type == SVt_PVAV) {
1298 if (old_type_details->body_size) {
1301 /* It will have been zeroed when the new body was allocated.
1302 Lets not write to it, in case it confuses a write-back
1308 #ifndef NODEFAULT_SHAREKEYS
1309 HvSHAREKEYS_on(sv); /* key-sharing on by default */
1311 HvMAX(sv) = 7; /* (start with 8 buckets) */
1314 /* SVt_NULL isn't the only thing upgraded to AV or HV.
1315 The target created by newSVrv also is, and it can have magic.
1316 However, it never has SvPVX set.
1318 if (old_type == SVt_IV) {
1320 } else if (old_type >= SVt_PV) {
1321 assert(SvPVX_const(sv) == 0);
1324 if (old_type >= SVt_PVMG) {
1325 SvMAGIC_set(sv, ((XPVMG*)old_body)->xmg_u.xmg_magic);
1326 SvSTASH_set(sv, ((XPVMG*)old_body)->xmg_stash);
1328 sv->sv_u.svu_array = NULL; /* or svu_hash */
1334 /* This ensures that SvTHINKFIRST(sv) is true, and hence that
1335 sv_force_normal_flags(sv) is called. */
1338 /* XXX Is this still needed? Was it ever needed? Surely as there is
1339 no route from NV to PVIV, NOK can never be true */
1340 assert(!SvNOKp(sv));
1351 assert(new_type_details->body_size);
1352 /* We always allocated the full length item with PURIFY. To do this
1353 we fake things so that arena is false for all 16 types.. */
1354 if(new_type_details->arena) {
1355 /* This points to the start of the allocated area. */
1356 new_body_inline(new_body, new_type);
1357 Zero(new_body, new_type_details->body_size, char);
1358 new_body = ((char *)new_body) - new_type_details->offset;
1360 new_body = new_NOARENAZ(new_type_details);
1362 SvANY(sv) = new_body;
1364 if (old_type_details->copy) {
1365 /* There is now the potential for an upgrade from something without
1366 an offset (PVNV or PVMG) to something with one (PVCV, PVFM) */
1367 int offset = old_type_details->offset;
1368 int length = old_type_details->copy;
1370 if (new_type_details->offset > old_type_details->offset) {
1371 const int difference
1372 = new_type_details->offset - old_type_details->offset;
1373 offset += difference;
1374 length -= difference;
1376 assert (length >= 0);
1378 Copy((char *)old_body + offset, (char *)new_body + offset, length,
1382 #ifndef NV_ZERO_IS_ALLBITS_ZERO
1383 /* If NV 0.0 is stores as all bits 0 then Zero() already creates a
1384 * correct 0.0 for us. Otherwise, if the old body didn't have an
1385 * NV slot, but the new one does, then we need to initialise the
1386 * freshly created NV slot with whatever the correct bit pattern is
1388 if (old_type_details->zero_nv && !new_type_details->zero_nv
1389 && !isGV_with_GP(sv))
1393 if (new_type == SVt_PVIO) {
1394 IO * const io = MUTABLE_IO(sv);
1395 GV *iogv = gv_fetchpvs("IO::File::", GV_ADD, SVt_PVHV);
1398 /* Clear the stashcache because a new IO could overrule a package
1400 hv_clear(PL_stashcache);
1402 SvSTASH_set(io, MUTABLE_HV(SvREFCNT_inc(GvHV(iogv))));
1403 IoPAGE_LEN(sv) = 60;
1405 if (old_type < SVt_PV) {
1406 /* referant will be NULL unless the old type was SVt_IV emulating
1408 sv->sv_u.svu_rv = referant;
1412 Perl_croak(aTHX_ "panic: sv_upgrade to unknown type %lu",
1413 (unsigned long)new_type);
1416 if (old_type > SVt_IV) {
1420 /* Note that there is an assumption that all bodies of types that
1421 can be upgraded came from arenas. Only the more complex non-
1422 upgradable types are allowed to be directly malloc()ed. */
1423 assert(old_type_details->arena);
1424 del_body((void*)((char*)old_body + old_type_details->offset),
1425 &PL_body_roots[old_type]);
1431 =for apidoc sv_backoff
1433 Remove any string offset. You should normally use the C<SvOOK_off> macro
1440 Perl_sv_backoff(pTHX_ register SV *const sv)
1443 const char * const s = SvPVX_const(sv);
1445 PERL_ARGS_ASSERT_SV_BACKOFF;
1446 PERL_UNUSED_CONTEXT;
1449 assert(SvTYPE(sv) != SVt_PVHV);
1450 assert(SvTYPE(sv) != SVt_PVAV);
1452 SvOOK_offset(sv, delta);
1454 SvLEN_set(sv, SvLEN(sv) + delta);
1455 SvPV_set(sv, SvPVX(sv) - delta);
1456 Move(s, SvPVX(sv), SvCUR(sv)+1, char);
1457 SvFLAGS(sv) &= ~SVf_OOK;
1464 Expands the character buffer in the SV. If necessary, uses C<sv_unref> and
1465 upgrades the SV to C<SVt_PV>. Returns a pointer to the character buffer.
1466 Use the C<SvGROW> wrapper instead.
1472 Perl_sv_grow(pTHX_ register SV *const sv, register STRLEN newlen)
1476 PERL_ARGS_ASSERT_SV_GROW;
1478 if (PL_madskills && newlen >= 0x100000) {
1479 PerlIO_printf(Perl_debug_log,
1480 "Allocation too large: %"UVxf"\n", (UV)newlen);
1482 #ifdef HAS_64K_LIMIT
1483 if (newlen >= 0x10000) {
1484 PerlIO_printf(Perl_debug_log,
1485 "Allocation too large: %"UVxf"\n", (UV)newlen);
1488 #endif /* HAS_64K_LIMIT */
1491 if (SvTYPE(sv) < SVt_PV) {
1492 sv_upgrade(sv, SVt_PV);
1493 s = SvPVX_mutable(sv);
1495 else if (SvOOK(sv)) { /* pv is offset? */
1497 s = SvPVX_mutable(sv);
1498 if (newlen > SvLEN(sv))
1499 newlen += 10 * (newlen - SvCUR(sv)); /* avoid copy each time */
1500 #ifdef HAS_64K_LIMIT
1501 if (newlen >= 0x10000)
1506 s = SvPVX_mutable(sv);
1508 if (newlen > SvLEN(sv)) { /* need more room? */
1509 STRLEN minlen = SvCUR(sv);
1510 minlen += (minlen >> PERL_STRLEN_EXPAND_SHIFT) + 10;
1511 if (newlen < minlen)
1513 #ifndef Perl_safesysmalloc_size
1514 newlen = PERL_STRLEN_ROUNDUP(newlen);
1516 if (SvLEN(sv) && s) {
1517 s = (char*)saferealloc(s, newlen);
1520 s = (char*)safemalloc(newlen);
1521 if (SvPVX_const(sv) && SvCUR(sv)) {
1522 Move(SvPVX_const(sv), s, (newlen < SvCUR(sv)) ? newlen : SvCUR(sv), char);
1526 #ifdef Perl_safesysmalloc_size
1527 /* Do this here, do it once, do it right, and then we will never get
1528 called back into sv_grow() unless there really is some growing
1530 SvLEN_set(sv, Perl_safesysmalloc_size(s));
1532 SvLEN_set(sv, newlen);
1539 =for apidoc sv_setiv
1541 Copies an integer into the given SV, upgrading first if necessary.
1542 Does not handle 'set' magic. See also C<sv_setiv_mg>.
1548 Perl_sv_setiv(pTHX_ register SV *const sv, const IV i)
1552 PERL_ARGS_ASSERT_SV_SETIV;
1554 SV_CHECK_THINKFIRST_COW_DROP(sv);
1555 switch (SvTYPE(sv)) {
1558 sv_upgrade(sv, SVt_IV);
1561 sv_upgrade(sv, SVt_PVIV);
1565 if (!isGV_with_GP(sv))
1572 /* diag_listed_as: Can't coerce %s to %s in %s */
1573 Perl_croak(aTHX_ "Can't coerce %s to integer in %s", sv_reftype(sv,0),
1577 (void)SvIOK_only(sv); /* validate number */
1583 =for apidoc sv_setiv_mg
1585 Like C<sv_setiv>, but also handles 'set' magic.
1591 Perl_sv_setiv_mg(pTHX_ register SV *const sv, const IV i)
1593 PERL_ARGS_ASSERT_SV_SETIV_MG;
1600 =for apidoc sv_setuv
1602 Copies an unsigned integer into the given SV, upgrading first if necessary.
1603 Does not handle 'set' magic. See also C<sv_setuv_mg>.
1609 Perl_sv_setuv(pTHX_ register SV *const sv, const UV u)
1611 PERL_ARGS_ASSERT_SV_SETUV;
1613 /* With these two if statements:
1614 u=1.49 s=0.52 cu=72.49 cs=10.64 scripts=270 tests=20865
1617 u=1.35 s=0.47 cu=73.45 cs=11.43 scripts=270 tests=20865
1619 If you wish to remove them, please benchmark to see what the effect is
1621 if (u <= (UV)IV_MAX) {
1622 sv_setiv(sv, (IV)u);
1631 =for apidoc sv_setuv_mg
1633 Like C<sv_setuv>, but also handles 'set' magic.
1639 Perl_sv_setuv_mg(pTHX_ register SV *const sv, const UV u)
1641 PERL_ARGS_ASSERT_SV_SETUV_MG;
1648 =for apidoc sv_setnv
1650 Copies a double into the given SV, upgrading first if necessary.
1651 Does not handle 'set' magic. See also C<sv_setnv_mg>.
1657 Perl_sv_setnv(pTHX_ register SV *const sv, const NV num)
1661 PERL_ARGS_ASSERT_SV_SETNV;
1663 SV_CHECK_THINKFIRST_COW_DROP(sv);
1664 switch (SvTYPE(sv)) {
1667 sv_upgrade(sv, SVt_NV);
1671 sv_upgrade(sv, SVt_PVNV);
1675 if (!isGV_with_GP(sv))
1682 /* diag_listed_as: Can't coerce %s to %s in %s */
1683 Perl_croak(aTHX_ "Can't coerce %s to number in %s", sv_reftype(sv,0),
1688 (void)SvNOK_only(sv); /* validate number */
1693 =for apidoc sv_setnv_mg
1695 Like C<sv_setnv>, but also handles 'set' magic.
1701 Perl_sv_setnv_mg(pTHX_ register SV *const sv, const NV num)
1703 PERL_ARGS_ASSERT_SV_SETNV_MG;
1709 /* Print an "isn't numeric" warning, using a cleaned-up,
1710 * printable version of the offending string
1714 S_not_a_number(pTHX_ SV *const sv)
1721 PERL_ARGS_ASSERT_NOT_A_NUMBER;
1724 dsv = newSVpvs_flags("", SVs_TEMP);
1725 pv = sv_uni_display(dsv, sv, 10, UNI_DISPLAY_ISPRINT);
1728 const char * const limit = tmpbuf + sizeof(tmpbuf) - 8;
1729 /* each *s can expand to 4 chars + "...\0",
1730 i.e. need room for 8 chars */
1732 const char *s = SvPVX_const(sv);
1733 const char * const end = s + SvCUR(sv);
1734 for ( ; s < end && d < limit; s++ ) {
1736 if (ch & 128 && !isPRINT_LC(ch)) {
1745 else if (ch == '\r') {
1749 else if (ch == '\f') {
1753 else if (ch == '\\') {
1757 else if (ch == '\0') {
1761 else if (isPRINT_LC(ch))
1778 Perl_warner(aTHX_ packWARN(WARN_NUMERIC),
1779 /* 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)
1802 register const char *sbegin;
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 SV *const buffer = sv_newmortal();
1820 PERL_ARGS_ASSERT_GLOB_2NUMBER;
1822 gv_efullname3(buffer, gv, "*");
1824 /* We know that all GVs stringify to something that is not-a-number,
1825 so no need to test that. */
1826 if (ckWARN(WARN_NUMERIC))
1827 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) && SvLEN(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)
2258 if (SvGMAGICAL(sv) || SvVALID(sv)) {
2259 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2260 the same flag bit as SVf_IVisUV, so must not let them cache IVs.
2261 In practice they are extremely unlikely to actually get anywhere
2262 accessible by user Perl code - the only way that I'm aware of is when
2263 a constant subroutine which is used as the second argument to index.
2265 if (flags & SV_GMAGIC)
2270 return I_V(SvNVX(sv));
2272 if (SvPOKp(sv) && SvLEN(sv)) {
2275 = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2277 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2278 == IS_NUMBER_IN_UV) {
2279 /* It's definitely an integer */
2280 if (numtype & IS_NUMBER_NEG) {
2281 if (value < (UV)IV_MIN)
2284 if (value < (UV)IV_MAX)
2289 if (ckWARN(WARN_NUMERIC))
2292 return I_V(Atof(SvPVX_const(sv)));
2297 assert(SvTYPE(sv) >= SVt_PVMG);
2298 /* This falls through to the report_uninit inside S_sv_2iuv_common. */
2299 } else if (SvTHINKFIRST(sv)) {
2304 if (flags & SV_SKIP_OVERLOAD)
2306 tmpstr = AMG_CALLunary(sv, numer_amg);
2307 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2308 return SvIV(tmpstr);
2311 return PTR2IV(SvRV(sv));
2314 sv_force_normal_flags(sv, 0);
2316 if (SvREADONLY(sv) && !SvOK(sv)) {
2317 if (ckWARN(WARN_UNINITIALIZED))
2323 if (S_sv_2iuv_common(aTHX_ sv))
2326 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2iv(%"IVdf")\n",
2327 PTR2UV(sv),SvIVX(sv)));
2328 return SvIsUV(sv) ? (IV)SvUVX(sv) : SvIVX(sv);
2332 =for apidoc sv_2uv_flags
2334 Return the unsigned integer value of an SV, doing any necessary string
2335 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2336 Normally used via the C<SvUV(sv)> and C<SvUVx(sv)> macros.
2342 Perl_sv_2uv_flags(pTHX_ register SV *const sv, const I32 flags)
2347 if (SvGMAGICAL(sv) || SvVALID(sv)) {
2348 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2349 the same flag bit as SVf_IVisUV, so must not let them cache IVs. */
2350 if (flags & SV_GMAGIC)
2355 return U_V(SvNVX(sv));
2356 if (SvPOKp(sv) && SvLEN(sv)) {
2359 = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2361 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2362 == IS_NUMBER_IN_UV) {
2363 /* It's definitely an integer */
2364 if (!(numtype & IS_NUMBER_NEG))
2368 if (ckWARN(WARN_NUMERIC))
2371 return U_V(Atof(SvPVX_const(sv)));
2376 assert(SvTYPE(sv) >= SVt_PVMG);
2377 /* This falls through to the report_uninit inside S_sv_2iuv_common. */
2378 } else if (SvTHINKFIRST(sv)) {
2383 if (flags & SV_SKIP_OVERLOAD)
2385 tmpstr = AMG_CALLunary(sv, numer_amg);
2386 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2387 return SvUV(tmpstr);
2390 return PTR2UV(SvRV(sv));
2393 sv_force_normal_flags(sv, 0);
2395 if (SvREADONLY(sv) && !SvOK(sv)) {
2396 if (ckWARN(WARN_UNINITIALIZED))
2402 if (S_sv_2iuv_common(aTHX_ sv))
2406 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2uv(%"UVuf")\n",
2407 PTR2UV(sv),SvUVX(sv)));
2408 return SvIsUV(sv) ? SvUVX(sv) : (UV)SvIVX(sv);
2412 =for apidoc sv_2nv_flags
2414 Return the num value of an SV, doing any necessary string or integer
2415 conversion. If flags includes SV_GMAGIC, does an mg_get() first.
2416 Normally used via the C<SvNV(sv)> and C<SvNVx(sv)> macros.
2422 Perl_sv_2nv_flags(pTHX_ register SV *const sv, const I32 flags)
2427 if (SvGMAGICAL(sv) || SvVALID(sv)) {
2428 /* FBMs use the space for SvIVX and SvNVX for other purposes, and use
2429 the same flag bit as SVf_IVisUV, so must not let them cache NVs. */
2430 if (flags & SV_GMAGIC)
2434 if ((SvPOKp(sv) && SvLEN(sv)) && !SvIOKp(sv)) {
2435 if (!SvIOKp(sv) && ckWARN(WARN_NUMERIC) &&
2436 !grok_number(SvPVX_const(sv), SvCUR(sv), NULL))
2438 return Atof(SvPVX_const(sv));
2442 return (NV)SvUVX(sv);
2444 return (NV)SvIVX(sv);
2449 assert(SvTYPE(sv) >= SVt_PVMG);
2450 /* This falls through to the report_uninit near the end of the
2452 } else if (SvTHINKFIRST(sv)) {
2457 if (flags & SV_SKIP_OVERLOAD)
2459 tmpstr = AMG_CALLunary(sv, numer_amg);
2460 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2461 return SvNV(tmpstr);
2464 return PTR2NV(SvRV(sv));
2467 sv_force_normal_flags(sv, 0);
2469 if (SvREADONLY(sv) && !SvOK(sv)) {
2470 if (ckWARN(WARN_UNINITIALIZED))
2475 if (SvTYPE(sv) < SVt_NV) {
2476 /* The logic to use SVt_PVNV if necessary is in sv_upgrade. */
2477 sv_upgrade(sv, SVt_NV);
2478 #ifdef USE_LONG_DOUBLE
2480 STORE_NUMERIC_LOCAL_SET_STANDARD();
2481 PerlIO_printf(Perl_debug_log,
2482 "0x%"UVxf" num(%" PERL_PRIgldbl ")\n",
2483 PTR2UV(sv), SvNVX(sv));
2484 RESTORE_NUMERIC_LOCAL();
2488 STORE_NUMERIC_LOCAL_SET_STANDARD();
2489 PerlIO_printf(Perl_debug_log, "0x%"UVxf" num(%"NVgf")\n",
2490 PTR2UV(sv), SvNVX(sv));
2491 RESTORE_NUMERIC_LOCAL();
2495 else if (SvTYPE(sv) < SVt_PVNV)
2496 sv_upgrade(sv, SVt_PVNV);
2501 SvNV_set(sv, SvIsUV(sv) ? (NV)SvUVX(sv) : (NV)SvIVX(sv));
2502 #ifdef NV_PRESERVES_UV
2508 /* Only set the public NV OK flag if this NV preserves the IV */
2509 /* Check it's not 0xFFFFFFFFFFFFFFFF */
2511 SvIsUV(sv) ? ((SvUVX(sv) != UV_MAX)&&(SvUVX(sv) == U_V(SvNVX(sv))))
2512 : (SvIVX(sv) == I_V(SvNVX(sv))))
2518 else if (SvPOKp(sv) && SvLEN(sv)) {
2520 const int numtype = grok_number(SvPVX_const(sv), SvCUR(sv), &value);
2521 if (!SvIOKp(sv) && !numtype && ckWARN(WARN_NUMERIC))
2523 #ifdef NV_PRESERVES_UV
2524 if ((numtype & (IS_NUMBER_IN_UV | IS_NUMBER_NOT_INT))
2525 == IS_NUMBER_IN_UV) {
2526 /* It's definitely an integer */
2527 SvNV_set(sv, (numtype & IS_NUMBER_NEG) ? -(NV)value : (NV)value);
2529 SvNV_set(sv, Atof(SvPVX_const(sv)));
2535 SvNV_set(sv, Atof(SvPVX_const(sv)));
2536 /* Only set the public NV OK flag if this NV preserves the value in
2537 the PV at least as well as an IV/UV would.
2538 Not sure how to do this 100% reliably. */
2539 /* if that shift count is out of range then Configure's test is
2540 wonky. We shouldn't be in here with NV_PRESERVES_UV_BITS ==
2542 if (((UV)1 << NV_PRESERVES_UV_BITS) >
2543 U_V(SvNVX(sv) > 0 ? SvNVX(sv) : -SvNVX(sv))) {
2544 SvNOK_on(sv); /* Definitely small enough to preserve all bits */
2545 } else if (!(numtype & IS_NUMBER_IN_UV)) {
2546 /* Can't use strtol etc to convert this string, so don't try.
2547 sv_2iv and sv_2uv will use the NV to convert, not the PV. */
2550 /* value has been set. It may not be precise. */
2551 if ((numtype & IS_NUMBER_NEG) && (value > (UV)IV_MIN)) {
2552 /* 2s complement assumption for (UV)IV_MIN */
2553 SvNOK_on(sv); /* Integer is too negative. */
2558 if (numtype & IS_NUMBER_NEG) {
2559 SvIV_set(sv, -(IV)value);
2560 } else if (value <= (UV)IV_MAX) {
2561 SvIV_set(sv, (IV)value);
2563 SvUV_set(sv, value);
2567 if (numtype & IS_NUMBER_NOT_INT) {
2568 /* I believe that even if the original PV had decimals,
2569 they are lost beyond the limit of the FP precision.
2570 However, neither is canonical, so both only get p
2571 flags. NWC, 2000/11/25 */
2572 /* Both already have p flags, so do nothing */
2574 const NV nv = SvNVX(sv);
2575 if (SvNVX(sv) < (NV)IV_MAX + 0.5) {
2576 if (SvIVX(sv) == I_V(nv)) {
2579 /* It had no "." so it must be integer. */
2583 /* between IV_MAX and NV(UV_MAX).
2584 Could be slightly > UV_MAX */
2586 if (numtype & IS_NUMBER_NOT_INT) {
2587 /* UV and NV both imprecise. */
2589 const UV nv_as_uv = U_V(nv);
2591 if (value == nv_as_uv && SvUVX(sv) != UV_MAX) {
2600 /* It might be more code efficient to go through the entire logic above
2601 and conditionally set with SvNOKp_on() rather than SvNOK(), but it
2602 gets complex and potentially buggy, so more programmer efficient
2603 to do it this way, by turning off the public flags: */
2605 SvFLAGS(sv) &= ~(SVf_IOK|SVf_NOK);
2606 #endif /* NV_PRESERVES_UV */
2609 if (isGV_with_GP(sv)) {
2610 glob_2number(MUTABLE_GV(sv));
2614 if (!PL_localizing && !SvPADTMP(sv) && ckWARN(WARN_UNINITIALIZED))
2616 assert (SvTYPE(sv) >= SVt_NV);
2617 /* Typically the caller expects that sv_any is not NULL now. */
2618 /* XXX Ilya implies that this is a bug in callers that assume this
2619 and ideally should be fixed. */
2622 #if defined(USE_LONG_DOUBLE)
2624 STORE_NUMERIC_LOCAL_SET_STANDARD();
2625 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2nv(%" PERL_PRIgldbl ")\n",
2626 PTR2UV(sv), SvNVX(sv));
2627 RESTORE_NUMERIC_LOCAL();
2631 STORE_NUMERIC_LOCAL_SET_STANDARD();
2632 PerlIO_printf(Perl_debug_log, "0x%"UVxf" 1nv(%"NVgf")\n",
2633 PTR2UV(sv), SvNVX(sv));
2634 RESTORE_NUMERIC_LOCAL();
2643 Return an SV with the numeric value of the source SV, doing any necessary
2644 reference or overload conversion. You must use the C<SvNUM(sv)> macro to
2645 access this function.
2651 Perl_sv_2num(pTHX_ register SV *const sv)
2653 PERL_ARGS_ASSERT_SV_2NUM;
2658 SV * const tmpsv = AMG_CALLunary(sv, numer_amg);
2659 TAINT_IF(tmpsv && SvTAINTED(tmpsv));
2660 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
2661 return sv_2num(tmpsv);
2663 return sv_2mortal(newSVuv(PTR2UV(SvRV(sv))));
2666 /* uiv_2buf(): private routine for use by sv_2pv_flags(): print an IV or
2667 * UV as a string towards the end of buf, and return pointers to start and
2670 * We assume that buf is at least TYPE_CHARS(UV) long.
2674 S_uiv_2buf(char *const buf, const IV iv, UV uv, const int is_uv, char **const peob)
2676 char *ptr = buf + TYPE_CHARS(UV);
2677 char * const ebuf = ptr;
2680 PERL_ARGS_ASSERT_UIV_2BUF;
2692 *--ptr = '0' + (char)(uv % 10);
2701 =for apidoc sv_2pv_flags
2703 Returns a pointer to the string value of an SV, and sets *lp to its length.
2704 If flags includes SV_GMAGIC, does an mg_get() first. Coerces sv to a
2705 string if necessary. Normally invoked via the C<SvPV_flags> macro.
2706 C<sv_2pv()> and C<sv_2pv_nomg> usually end up here too.
2712 Perl_sv_2pv_flags(pTHX_ register SV *const sv, STRLEN *const lp, const I32 flags)
2722 if (SvGMAGICAL(sv)) {
2723 if (flags & SV_GMAGIC)
2728 if (flags & SV_MUTABLE_RETURN)
2729 return SvPVX_mutable(sv);
2730 if (flags & SV_CONST_RETURN)
2731 return (char *)SvPVX_const(sv);
2734 if (SvIOKp(sv) || SvNOKp(sv)) {
2735 char tbuf[64]; /* Must fit sprintf/Gconvert of longest IV/NV */
2740 ? my_snprintf(tbuf, sizeof(tbuf), "%"UVuf, (UV)SvUVX(sv))
2741 : my_snprintf(tbuf, sizeof(tbuf), "%"IVdf, (IV)SvIVX(sv));
2742 } else if(SvNVX(sv) == 0.0) {
2747 Gconvert(SvNVX(sv), NV_DIG, 0, tbuf);
2754 SvUPGRADE(sv, SVt_PV);
2757 s = SvGROW_mutable(sv, len + 1);
2760 return (char*)memcpy(s, tbuf, len + 1);
2766 assert(SvTYPE(sv) >= SVt_PVMG);
2767 /* This falls through to the report_uninit near the end of the
2769 } else if (SvTHINKFIRST(sv)) {
2774 if (flags & SV_SKIP_OVERLOAD)
2776 tmpstr = AMG_CALLunary(sv, string_amg);
2777 TAINT_IF(tmpstr && SvTAINTED(tmpstr));
2778 if (tmpstr && (!SvROK(tmpstr) || (SvRV(tmpstr) != SvRV(sv)))) {
2780 /* char *pv = lp ? SvPV(tmpstr, *lp) : SvPV_nolen(tmpstr);
2784 if ((SvFLAGS(tmpstr) & (SVf_POK)) == SVf_POK) {
2785 if (flags & SV_CONST_RETURN) {
2786 pv = (char *) SvPVX_const(tmpstr);
2788 pv = (flags & SV_MUTABLE_RETURN)
2789 ? SvPVX_mutable(tmpstr) : SvPVX(tmpstr);
2792 *lp = SvCUR(tmpstr);
2794 pv = sv_2pv_flags(tmpstr, lp, flags);
2807 SV *const referent = SvRV(sv);
2811 retval = buffer = savepvn("NULLREF", len);
2812 } else if (SvTYPE(referent) == SVt_REGEXP && (
2813 !(PL_curcop->cop_hints & HINT_NO_AMAGIC)
2814 || amagic_is_enabled(string_amg)
2816 REGEXP * const re = (REGEXP *)MUTABLE_PTR(referent);
2821 /* If the regex is UTF-8 we want the containing scalar to
2822 have an UTF-8 flag too */
2828 if ((seen_evals = RX_SEEN_EVALS(re)))
2829 PL_reginterp_cnt += seen_evals;
2832 *lp = RX_WRAPLEN(re);
2834 return RX_WRAPPED(re);
2836 const char *const typestr = sv_reftype(referent, 0);
2837 const STRLEN typelen = strlen(typestr);
2838 UV addr = PTR2UV(referent);
2839 const char *stashname = NULL;
2840 STRLEN stashnamelen = 0; /* hush, gcc */
2841 const char *buffer_end;
2843 if (SvOBJECT(referent)) {
2844 const HEK *const name = HvNAME_HEK(SvSTASH(referent));
2847 stashname = HEK_KEY(name);
2848 stashnamelen = HEK_LEN(name);
2850 if (HEK_UTF8(name)) {
2856 stashname = "__ANON__";
2859 len = stashnamelen + 1 /* = */ + typelen + 3 /* (0x */
2860 + 2 * sizeof(UV) + 2 /* )\0 */;
2862 len = typelen + 3 /* (0x */
2863 + 2 * sizeof(UV) + 2 /* )\0 */;
2866 Newx(buffer, len, char);
2867 buffer_end = retval = buffer + len;
2869 /* Working backwards */
2873 *--retval = PL_hexdigit[addr & 15];
2874 } while (addr >>= 4);
2880 memcpy(retval, typestr, typelen);
2884 retval -= stashnamelen;
2885 memcpy(retval, stashname, stashnamelen);
2887 /* retval may not necessarily have reached the start of the
2889 assert (retval >= buffer);
2891 len = buffer_end - retval - 1; /* -1 for that \0 */
2899 if (SvREADONLY(sv) && !SvOK(sv)) {
2902 if (flags & SV_UNDEF_RETURNS_NULL)
2904 if (ckWARN(WARN_UNINITIALIZED))
2909 if (SvIOK(sv) || ((SvIOKp(sv) && !SvNOKp(sv)))) {
2910 /* I'm assuming that if both IV and NV are equally valid then
2911 converting the IV is going to be more efficient */
2912 const U32 isUIOK = SvIsUV(sv);
2913 char buf[TYPE_CHARS(UV)];
2917 if (SvTYPE(sv) < SVt_PVIV)
2918 sv_upgrade(sv, SVt_PVIV);
2919 ptr = uiv_2buf(buf, SvIVX(sv), SvUVX(sv), isUIOK, &ebuf);
2921 /* inlined from sv_setpvn */
2922 s = SvGROW_mutable(sv, len + 1);
2923 Move(ptr, s, len, char);
2927 else if (SvNOKp(sv)) {
2928 if (SvTYPE(sv) < SVt_PVNV)
2929 sv_upgrade(sv, SVt_PVNV);
2930 if (SvNVX(sv) == 0.0) {
2931 s = SvGROW_mutable(sv, 2);
2936 /* The +20 is pure guesswork. Configure test needed. --jhi */
2937 s = SvGROW_mutable(sv, NV_DIG + 20);
2938 /* some Xenix systems wipe out errno here */
2939 Gconvert(SvNVX(sv), NV_DIG, 0, s);
2949 if (isGV_with_GP(sv)) {
2950 GV *const gv = MUTABLE_GV(sv);
2951 SV *const buffer = sv_newmortal();
2953 gv_efullname3(buffer, gv, "*");
2955 assert(SvPOK(buffer));
2957 *lp = SvCUR(buffer);
2959 if ( SvUTF8(buffer) ) SvUTF8_on(sv);
2960 return SvPVX(buffer);
2965 if (flags & SV_UNDEF_RETURNS_NULL)
2967 if (!PL_localizing && !SvPADTMP(sv) && ckWARN(WARN_UNINITIALIZED))
2969 if (SvTYPE(sv) < SVt_PV)
2970 /* Typically the caller expects that sv_any is not NULL now. */
2971 sv_upgrade(sv, SVt_PV);
2975 const STRLEN len = s - SvPVX_const(sv);
2981 DEBUG_c(PerlIO_printf(Perl_debug_log, "0x%"UVxf" 2pv(%s)\n",
2982 PTR2UV(sv),SvPVX_const(sv)));
2983 if (flags & SV_CONST_RETURN)
2984 return (char *)SvPVX_const(sv);
2985 if (flags & SV_MUTABLE_RETURN)
2986 return SvPVX_mutable(sv);
2991 =for apidoc sv_copypv
2993 Copies a stringified representation of the source SV into the
2994 destination SV. Automatically performs any necessary mg_get and
2995 coercion of numeric values into strings. Guaranteed to preserve
2996 UTF8 flag even from overloaded objects. Similar in nature to
2997 sv_2pv[_flags] but operates directly on an SV instead of just the
2998 string. Mostly uses sv_2pv_flags to do its work, except when that
2999 would lose the UTF-8'ness of the PV.
3005 Perl_sv_copypv(pTHX_ SV *const dsv, register SV *const ssv)
3008 const char * const s = SvPV_const(ssv,len);
3010 PERL_ARGS_ASSERT_SV_COPYPV;
3012 sv_setpvn(dsv,s,len);
3020 =for apidoc sv_2pvbyte
3022 Return a pointer to the byte-encoded representation of the SV, and set *lp
3023 to its length. May cause the SV to be downgraded from UTF-8 as a
3026 Usually accessed via the C<SvPVbyte> macro.
3032 Perl_sv_2pvbyte(pTHX_ register SV *const sv, STRLEN *const lp)
3034 PERL_ARGS_ASSERT_SV_2PVBYTE;
3037 sv_utf8_downgrade(sv,0);
3038 return lp ? SvPV_nomg(sv,*lp) : SvPV_nomg_nolen(sv);
3042 =for apidoc sv_2pvutf8
3044 Return a pointer to the UTF-8-encoded representation of the SV, and set *lp
3045 to its length. May cause the SV to be upgraded to UTF-8 as a side-effect.
3047 Usually accessed via the C<SvPVutf8> macro.
3053 Perl_sv_2pvutf8(pTHX_ register SV *const sv, STRLEN *const lp)
3055 PERL_ARGS_ASSERT_SV_2PVUTF8;
3057 sv_utf8_upgrade(sv);
3058 return lp ? SvPV(sv,*lp) : SvPV_nolen(sv);
3063 =for apidoc sv_2bool
3065 This macro is only used by sv_true() or its macro equivalent, and only if
3066 the latter's argument is neither SvPOK, SvIOK nor SvNOK.
3067 It calls sv_2bool_flags with the SV_GMAGIC flag.
3069 =for apidoc sv_2bool_flags
3071 This function is only used by sv_true() and friends, and only if
3072 the latter's argument is neither SvPOK, SvIOK nor SvNOK. If the flags
3073 contain SV_GMAGIC, then it does an mg_get() first.
3080 Perl_sv_2bool_flags(pTHX_ register SV *const sv, const I32 flags)
3084 PERL_ARGS_ASSERT_SV_2BOOL_FLAGS;
3086 if(flags & SV_GMAGIC) SvGETMAGIC(sv);
3092 SV * const tmpsv = AMG_CALLunary(sv, bool__amg);
3093 if (tmpsv && (!SvROK(tmpsv) || (SvRV(tmpsv) != SvRV(sv))))
3094 return cBOOL(SvTRUE(tmpsv));
3096 return SvRV(sv) != 0;
3099 register XPV* const Xpvtmp = (XPV*)SvANY(sv);
3101 (*sv->sv_u.svu_pv > '0' ||
3102 Xpvtmp->xpv_cur > 1 ||
3103 (Xpvtmp->xpv_cur && *sv->sv_u.svu_pv != '0')))
3110 return SvIVX(sv) != 0;
3113 return SvNVX(sv) != 0.0;
3115 if (isGV_with_GP(sv))
3125 =for apidoc sv_utf8_upgrade
3127 Converts the PV of an SV to its UTF-8-encoded form.
3128 Forces the SV to string form if it is not already.
3129 Will C<mg_get> on C<sv> if appropriate.
3130 Always sets the SvUTF8 flag to avoid future validity checks even
3131 if the whole string is the same in UTF-8 as not.
3132 Returns the number of bytes in the converted string
3134 This is not as a general purpose byte encoding to Unicode interface:
3135 use the Encode extension for that.
3137 =for apidoc sv_utf8_upgrade_nomg
3139 Like sv_utf8_upgrade, but doesn't do magic on C<sv>.
3141 =for apidoc sv_utf8_upgrade_flags
3143 Converts the PV of an SV to its UTF-8-encoded form.
3144 Forces the SV to string form if it is not already.
3145 Always sets the SvUTF8 flag to avoid future validity checks even
3146 if all the bytes are invariant in UTF-8.
3147 If C<flags> has C<SV_GMAGIC> bit set,
3148 will C<mg_get> on C<sv> if appropriate, else not.
3149 Returns the number of bytes in the converted string
3150 C<sv_utf8_upgrade> and
3151 C<sv_utf8_upgrade_nomg> are implemented in terms of this function.
3153 This is not as a general purpose byte encoding to Unicode interface:
3154 use the Encode extension for that.
3158 The grow version is currently not externally documented. It adds a parameter,
3159 extra, which is the number of unused bytes the string of 'sv' is guaranteed to
3160 have free after it upon return. This allows the caller to reserve extra space
3161 that it intends to fill, to avoid extra grows.
3163 Also externally undocumented for the moment is the flag SV_FORCE_UTF8_UPGRADE,
3164 which can be used to tell this function to not first check to see if there are
3165 any characters that are different in UTF-8 (variant characters) which would
3166 force it to allocate a new string to sv, but to assume there are. Typically
3167 this flag is used by a routine that has already parsed the string to find that
3168 there are such characters, and passes this information on so that the work
3169 doesn't have to be repeated.
3171 (One might think that the calling routine could pass in the position of the
3172 first such variant, so it wouldn't have to be found again. But that is not the
3173 case, because typically when the caller is likely to use this flag, it won't be
3174 calling this routine unless it finds something that won't fit into a byte.
3175 Otherwise it tries to not upgrade and just use bytes. But some things that
3176 do fit into a byte are variants in utf8, and the caller may not have been
3177 keeping track of these.)
3179 If the routine itself changes the string, it adds a trailing NUL. Such a NUL
3180 isn't guaranteed due to having other routines do the work in some input cases,
3181 or if the input is already flagged as being in utf8.
3183 The speed of this could perhaps be improved for many cases if someone wanted to
3184 write a fast function that counts the number of variant characters in a string,
3185 especially if it could return the position of the first one.
3190 Perl_sv_utf8_upgrade_flags_grow(pTHX_ register SV *const sv, const I32 flags, STRLEN extra)
3194 PERL_ARGS_ASSERT_SV_UTF8_UPGRADE_FLAGS_GROW;
3196 if (sv == &PL_sv_undef)
3200 if (SvREADONLY(sv) && (SvPOKp(sv) || SvIOKp(sv) || SvNOKp(sv))) {
3201 (void) sv_2pv_flags(sv,&len, flags);
3203 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3207 (void) SvPV_force_flags(sv,len,flags & SV_GMAGIC);
3212 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3217 sv_force_normal_flags(sv, 0);
3220 if (PL_encoding && !(flags & SV_UTF8_NO_ENCODING)) {
3221 sv_recode_to_utf8(sv, PL_encoding);
3222 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3226 if (SvCUR(sv) == 0) {
3227 if (extra) SvGROW(sv, extra);
3228 } else { /* Assume Latin-1/EBCDIC */
3229 /* This function could be much more efficient if we
3230 * had a FLAG in SVs to signal if there are any variant
3231 * chars in the PV. Given that there isn't such a flag
3232 * make the loop as fast as possible (although there are certainly ways
3233 * to speed this up, eg. through vectorization) */
3234 U8 * s = (U8 *) SvPVX_const(sv);
3235 U8 * e = (U8 *) SvEND(sv);
3237 STRLEN two_byte_count = 0;
3239 if (flags & SV_FORCE_UTF8_UPGRADE) goto must_be_utf8;
3241 /* See if really will need to convert to utf8. We mustn't rely on our
3242 * incoming SV being well formed and having a trailing '\0', as certain
3243 * code in pp_formline can send us partially built SVs. */
3247 if (NATIVE_IS_INVARIANT(ch)) continue;
3249 t--; /* t already incremented; re-point to first variant */
3254 /* utf8 conversion not needed because all are invariants. Mark as
3255 * UTF-8 even if no variant - saves scanning loop */
3257 if (extra) SvGROW(sv, SvCUR(sv) + extra);
3262 /* Here, the string should be converted to utf8, either because of an
3263 * input flag (two_byte_count = 0), or because a character that
3264 * requires 2 bytes was found (two_byte_count = 1). t points either to
3265 * the beginning of the string (if we didn't examine anything), or to
3266 * the first variant. In either case, everything from s to t - 1 will
3267 * occupy only 1 byte each on output.
3269 * There are two main ways to convert. One is to create a new string
3270 * and go through the input starting from the beginning, appending each
3271 * converted value onto the new string as we go along. It's probably
3272 * best to allocate enough space in the string for the worst possible
3273 * case rather than possibly running out of space and having to
3274 * reallocate and then copy what we've done so far. Since everything
3275 * from s to t - 1 is invariant, the destination can be initialized
3276 * with these using a fast memory copy
3278 * The other way is to figure out exactly how big the string should be
3279 * by parsing the entire input. Then you don't have to make it big
3280 * enough to handle the worst possible case, and more importantly, if
3281 * the string you already have is large enough, you don't have to
3282 * allocate a new string, you can copy the last character in the input
3283 * string to the final position(s) that will be occupied by the
3284 * converted string and go backwards, stopping at t, since everything
3285 * before that is invariant.
3287 * There are advantages and disadvantages to each method.
3289 * In the first method, we can allocate a new string, do the memory
3290 * copy from the s to t - 1, and then proceed through the rest of the
3291 * string byte-by-byte.
3293 * In the second method, we proceed through the rest of the input
3294 * string just calculating how big the converted string will be. Then
3295 * there are two cases:
3296 * 1) if the string has enough extra space to handle the converted
3297 * value. We go backwards through the string, converting until we
3298 * get to the position we are at now, and then stop. If this
3299 * position is far enough along in the string, this method is
3300 * faster than the other method. If the memory copy were the same
3301 * speed as the byte-by-byte loop, that position would be about
3302 * half-way, as at the half-way mark, parsing to the end and back
3303 * is one complete string's parse, the same amount as starting
3304 * over and going all the way through. Actually, it would be
3305 * somewhat less than half-way, as it's faster to just count bytes
3306 * than to also copy, and we don't have the overhead of allocating
3307 * a new string, changing the scalar to use it, and freeing the
3308 * existing one. But if the memory copy is fast, the break-even
3309 * point is somewhere after half way. The counting loop could be
3310 * sped up by vectorization, etc, to move the break-even point
3311 * further towards the beginning.
3312 * 2) if the string doesn't have enough space to handle the converted
3313 * value. A new string will have to be allocated, and one might
3314 * as well, given that, start from the beginning doing the first
3315 * method. We've spent extra time parsing the string and in
3316 * exchange all we've gotten is that we know precisely how big to
3317 * make the new one. Perl is more optimized for time than space,
3318 * so this case is a loser.
3319 * So what I've decided to do is not use the 2nd method unless it is
3320 * guaranteed that a new string won't have to be allocated, assuming
3321 * the worst case. I also decided not to put any more conditions on it
3322 * than this, for now. It seems likely that, since the worst case is
3323 * twice as big as the unknown portion of the string (plus 1), we won't
3324 * be guaranteed enough space, causing us to go to the first method,
3325 * unless the string is short, or the first variant character is near
3326 * the end of it. In either of these cases, it seems best to use the
3327 * 2nd method. The only circumstance I can think of where this would
3328 * be really slower is if the string had once had much more data in it
3329 * than it does now, but there is still a substantial amount in it */
3332 STRLEN invariant_head = t - s;
3333 STRLEN size = invariant_head + (e - t) * 2 + 1 + extra;
3334 if (SvLEN(sv) < size) {
3336 /* Here, have decided to allocate a new string */
3341 Newx(dst, size, U8);
3343 /* If no known invariants at the beginning of the input string,
3344 * set so starts from there. Otherwise, can use memory copy to
3345 * get up to where we are now, and then start from here */
3347 if (invariant_head <= 0) {
3350 Copy(s, dst, invariant_head, char);
3351 d = dst + invariant_head;
3355 const UV uv = NATIVE8_TO_UNI(*t++);
3356 if (UNI_IS_INVARIANT(uv))
3357 *d++ = (U8)UNI_TO_NATIVE(uv);
3359 *d++ = (U8)UTF8_EIGHT_BIT_HI(uv);
3360 *d++ = (U8)UTF8_EIGHT_BIT_LO(uv);
3364 SvPV_free(sv); /* No longer using pre-existing string */
3365 SvPV_set(sv, (char*)dst);
3366 SvCUR_set(sv, d - dst);
3367 SvLEN_set(sv, size);
3370 /* Here, have decided to get the exact size of the string.
3371 * Currently this happens only when we know that there is
3372 * guaranteed enough space to fit the converted string, so
3373 * don't have to worry about growing. If two_byte_count is 0,
3374 * then t points to the first byte of the string which hasn't
3375 * been examined yet. Otherwise two_byte_count is 1, and t
3376 * points to the first byte in the string that will expand to
3377 * two. Depending on this, start examining at t or 1 after t.
3380 U8 *d = t + two_byte_count;
3383 /* Count up the remaining bytes that expand to two */
3386 const U8 chr = *d++;
3387 if (! NATIVE_IS_INVARIANT(chr)) two_byte_count++;
3390 /* The string will expand by just the number of bytes that
3391 * occupy two positions. But we are one afterwards because of
3392 * the increment just above. This is the place to put the
3393 * trailing NUL, and to set the length before we decrement */
3395 d += two_byte_count;
3396 SvCUR_set(sv, d - s);
3400 /* Having decremented d, it points to the position to put the
3401 * very last byte of the expanded string. Go backwards through
3402 * the string, copying and expanding as we go, stopping when we
3403 * get to the part that is invariant the rest of the way down */
3407 const U8 ch = NATIVE8_TO_UNI(*e--);
3408 if (UNI_IS_INVARIANT(ch)) {
3409 *d-- = UNI_TO_NATIVE(ch);
3411 *d-- = (U8)UTF8_EIGHT_BIT_LO(ch);
3412 *d-- = (U8)UTF8_EIGHT_BIT_HI(ch);
3417 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3418 /* Update pos. We do it at the end rather than during
3419 * the upgrade, to avoid slowing down the common case
3420 * (upgrade without pos) */
3421 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3423 I32 pos = mg->mg_len;
3424 if (pos > 0 && (U32)pos > invariant_head) {
3425 U8 *d = (U8*) SvPVX(sv) + invariant_head;
3426 STRLEN n = (U32)pos - invariant_head;
3428 if (UTF8_IS_START(*d))
3433 mg->mg_len = d - (U8*)SvPVX(sv);
3436 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3437 magic_setutf8(sv,mg); /* clear UTF8 cache */
3442 /* Mark as UTF-8 even if no variant - saves scanning loop */
3448 =for apidoc sv_utf8_downgrade
3450 Attempts to convert the PV of an SV from characters to bytes.
3451 If the PV contains a character that cannot fit
3452 in a byte, this conversion will fail;
3453 in this case, either returns false or, if C<fail_ok> is not
3456 This is not as a general purpose Unicode to byte encoding interface:
3457 use the Encode extension for that.
3463 Perl_sv_utf8_downgrade(pTHX_ register SV *const sv, const bool fail_ok)
3467 PERL_ARGS_ASSERT_SV_UTF8_DOWNGRADE;
3469 if (SvPOKp(sv) && SvUTF8(sv)) {
3473 int mg_flags = SV_GMAGIC;
3476 sv_force_normal_flags(sv, 0);
3478 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3480 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3482 I32 pos = mg->mg_len;
3484 sv_pos_b2u(sv, &pos);
3485 mg_flags = 0; /* sv_pos_b2u does get magic */
3489 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3490 magic_setutf8(sv,mg); /* clear UTF8 cache */
3493 s = (U8 *) SvPV_flags(sv, len, mg_flags);
3495 if (!utf8_to_bytes(s, &len)) {
3500 Perl_croak(aTHX_ "Wide character in %s",
3503 Perl_croak(aTHX_ "Wide character");
3514 =for apidoc sv_utf8_encode
3516 Converts the PV of an SV to UTF-8, but then turns the C<SvUTF8>
3517 flag off so that it looks like octets again.
3523 Perl_sv_utf8_encode(pTHX_ register SV *const sv)
3525 PERL_ARGS_ASSERT_SV_UTF8_ENCODE;
3527 if (SvREADONLY(sv)) {
3528 sv_force_normal_flags(sv, 0);
3530 (void) sv_utf8_upgrade(sv);
3535 =for apidoc sv_utf8_decode
3537 If the PV of the SV is an octet sequence in UTF-8
3538 and contains a multiple-byte character, the C<SvUTF8> flag is turned on
3539 so that it looks like a character. If the PV contains only single-byte
3540 characters, the C<SvUTF8> flag stays off.
3541 Scans PV for validity and returns false if the PV is invalid UTF-8.
3547 Perl_sv_utf8_decode(pTHX_ register SV *const sv)
3549 PERL_ARGS_ASSERT_SV_UTF8_DECODE;
3552 const U8 *start, *c;
3555 /* The octets may have got themselves encoded - get them back as
3558 if (!sv_utf8_downgrade(sv, TRUE))
3561 /* it is actually just a matter of turning the utf8 flag on, but
3562 * we want to make sure everything inside is valid utf8 first.
3564 c = start = (const U8 *) SvPVX_const(sv);
3565 if (!is_utf8_string(c, SvCUR(sv)))
3567 e = (const U8 *) SvEND(sv);
3570 if (!UTF8_IS_INVARIANT(ch)) {
3575 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
3576 /* adjust pos to the start of a UTF8 char sequence */
3577 MAGIC * mg = mg_find(sv, PERL_MAGIC_regex_global);
3579 I32 pos = mg->mg_len;
3581 for (c = start + pos; c > start; c--) {
3582 if (UTF8_IS_START(*c))
3585 mg->mg_len = c - start;
3588 if ((mg = mg_find(sv, PERL_MAGIC_utf8)))
3589 magic_setutf8(sv,mg); /* clear UTF8 cache */
3596 =for apidoc sv_setsv
3598 Copies the contents of the source SV C<ssv> into the destination SV
3599 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3600 function if the source SV needs to be reused. Does not handle 'set' magic.
3601 Loosely speaking, it performs a copy-by-value, obliterating any previous
3602 content of the destination.
3604 You probably want to use one of the assortment of wrappers, such as
3605 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3606 C<SvSetMagicSV_nosteal>.
3608 =for apidoc sv_setsv_flags
3610 Copies the contents of the source SV C<ssv> into the destination SV
3611 C<dsv>. The source SV may be destroyed if it is mortal, so don't use this
3612 function if the source SV needs to be reused. Does not handle 'set' magic.
3613 Loosely speaking, it performs a copy-by-value, obliterating any previous
3614 content of the destination.
3615 If the C<flags> parameter has the C<SV_GMAGIC> bit set, will C<mg_get> on
3616 C<ssv> if appropriate, else not. If the C<flags>
3617 parameter has the C<NOSTEAL> bit set then the
3618 buffers of temps will not be stolen. <sv_setsv>
3619 and C<sv_setsv_nomg> are implemented in terms of this function.
3621 You probably want to use one of the assortment of wrappers, such as
3622 C<SvSetSV>, C<SvSetSV_nosteal>, C<SvSetMagicSV> and
3623 C<SvSetMagicSV_nosteal>.
3625 This is the primary function for copying scalars, and most other
3626 copy-ish functions and macros use this underneath.
3632 S_glob_assign_glob(pTHX_ SV *const dstr, SV *const sstr, const int dtype)
3634 I32 mro_changes = 0; /* 1 = method, 2 = isa, 3 = recursive isa */
3635 HV *old_stash = NULL;
3637 PERL_ARGS_ASSERT_GLOB_ASSIGN_GLOB;
3639 if (dtype != SVt_PVGV && !isGV_with_GP(dstr)) {
3640 const char * const name = GvNAME(sstr);
3641 const STRLEN len = GvNAMELEN(sstr);
3643 if (dtype >= SVt_PV) {
3649 SvUPGRADE(dstr, SVt_PVGV);
3650 (void)SvOK_off(dstr);
3651 /* We have to turn this on here, even though we turn it off
3652 below, as GvSTASH will fail an assertion otherwise. */
3653 isGV_with_GP_on(dstr);
3655 GvSTASH(dstr) = GvSTASH(sstr);
3657 Perl_sv_add_backref(aTHX_ MUTABLE_SV(GvSTASH(dstr)), dstr);
3658 gv_name_set(MUTABLE_GV(dstr), name, len,
3659 GV_ADD | (GvNAMEUTF8(sstr) ? SVf_UTF8 : 0 ));
3660 SvFAKE_on(dstr); /* can coerce to non-glob */
3663 if(GvGP(MUTABLE_GV(sstr))) {
3664 /* If source has method cache entry, clear it */
3666 SvREFCNT_dec(GvCV(sstr));
3667 GvCV_set(sstr, NULL);
3670 /* If source has a real method, then a method is
3673 GvCV((const GV *)sstr) && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3679 /* If dest already had a real method, that's a change as well */
3681 !mro_changes && GvGP(MUTABLE_GV(dstr)) && GvCVu((const GV *)dstr)
3682 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3687 /* We don't need to check the name of the destination if it was not a
3688 glob to begin with. */
3689 if(dtype == SVt_PVGV) {
3690 const char * const name = GvNAME((const GV *)dstr);
3693 /* The stash may have been detached from the symbol table, so
3695 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3696 && GvAV((const GV *)sstr)
3700 const STRLEN len = GvNAMELEN(dstr);
3701 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
3702 || (len == 1 && name[0] == ':')) {
3705 /* Set aside the old stash, so we can reset isa caches on
3707 if((old_stash = GvHV(dstr)))
3708 /* Make sure we do not lose it early. */
3709 SvREFCNT_inc_simple_void_NN(
3710 sv_2mortal((SV *)old_stash)
3716 gp_free(MUTABLE_GV(dstr));
3717 isGV_with_GP_off(dstr); /* SvOK_off does not like globs. */
3718 (void)SvOK_off(dstr);
3719 isGV_with_GP_on(dstr);
3720 GvINTRO_off(dstr); /* one-shot flag */
3721 GvGP_set(dstr, gp_ref(GvGP(sstr)));
3722 if (SvTAINTED(sstr))
3724 if (GvIMPORTED(dstr) != GVf_IMPORTED
3725 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
3727 GvIMPORTED_on(dstr);
3730 if(mro_changes == 2) {
3732 SV * const sref = (SV *)GvAV((const GV *)dstr);
3733 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
3734 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
3735 AV * const ary = newAV();
3736 av_push(ary, mg->mg_obj); /* takes the refcount */
3737 mg->mg_obj = (SV *)ary;
3739 av_push((AV *)mg->mg_obj, SvREFCNT_inc_simple_NN(dstr));
3741 else sv_magic(sref, dstr, PERL_MAGIC_isa, NULL, 0);
3742 mro_isa_changed_in(GvSTASH(dstr));
3744 else if(mro_changes == 3) {
3745 HV * const stash = GvHV(dstr);
3746 if(old_stash ? (HV *)HvENAME_get(old_stash) : stash)
3752 else if(mro_changes) mro_method_changed_in(GvSTASH(dstr));
3757 S_glob_assign_ref(pTHX_ SV *const dstr, SV *const sstr)
3759 SV * const sref = SvREFCNT_inc(SvRV(sstr));
3761 const int intro = GvINTRO(dstr);
3764 const U32 stype = SvTYPE(sref);
3766 PERL_ARGS_ASSERT_GLOB_ASSIGN_REF;
3769 GvINTRO_off(dstr); /* one-shot flag */
3770 GvLINE(dstr) = CopLINE(PL_curcop);
3771 GvEGV(dstr) = MUTABLE_GV(dstr);
3776 location = (SV **) &(GvGP(dstr)->gp_cv); /* XXX bypassing GvCV_set */
3777 import_flag = GVf_IMPORTED_CV;
3780 location = (SV **) &GvHV(dstr);
3781 import_flag = GVf_IMPORTED_HV;
3784 location = (SV **) &GvAV(dstr);
3785 import_flag = GVf_IMPORTED_AV;
3788 location = (SV **) &GvIOp(dstr);
3791 location = (SV **) &GvFORM(dstr);
3794 location = &GvSV(dstr);
3795 import_flag = GVf_IMPORTED_SV;
3798 if (stype == SVt_PVCV) {
3799 /*if (GvCVGEN(dstr) && (GvCV(dstr) != (const CV *)sref || GvCVGEN(dstr))) {*/
3800 if (GvCVGEN(dstr)) {
3801 SvREFCNT_dec(GvCV(dstr));
3802 GvCV_set(dstr, NULL);
3803 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
3806 SAVEGENERICSV(*location);
3810 if (stype == SVt_PVCV && (*location != sref || GvCVGEN(dstr))) {
3811 CV* const cv = MUTABLE_CV(*location);
3813 if (!GvCVGEN((const GV *)dstr) &&
3814 (CvROOT(cv) || CvXSUB(cv)) &&
3815 /* redundant check that avoids creating the extra SV
3816 most of the time: */
3817 (CvCONST(cv) || ckWARN(WARN_REDEFINE)))
3819 SV * const new_const_sv =
3820 CvCONST((const CV *)sref)
3821 ? cv_const_sv((const CV *)sref)
3823 report_redefined_cv(
3824 sv_2mortal(Perl_newSVpvf(aTHX_
3827 HvNAME_HEK(GvSTASH((const GV *)dstr))
3829 HEKfARG(GvENAME_HEK(MUTABLE_GV(dstr)))
3832 CvCONST((const CV *)sref) ? &new_const_sv : NULL
3836 cv_ckproto_len_flags(cv, (const GV *)dstr,
3837 SvPOK(sref) ? CvPROTO(sref) : NULL,
3838 SvPOK(sref) ? CvPROTOLEN(sref) : 0,
3839 SvPOK(sref) ? SvUTF8(sref) : 0);
3841 GvCVGEN(dstr) = 0; /* Switch off cacheness. */
3842 GvASSUMECV_on(dstr);
3843 if(GvSTASH(dstr)) mro_method_changed_in(GvSTASH(dstr)); /* sub foo { 1 } sub bar { 2 } *bar = \&foo */
3846 if (import_flag && !(GvFLAGS(dstr) & import_flag)
3847 && CopSTASH_ne(PL_curcop, GvSTASH(dstr))) {
3848 GvFLAGS(dstr) |= import_flag;
3850 if (stype == SVt_PVHV) {
3851 const char * const name = GvNAME((GV*)dstr);
3852 const STRLEN len = GvNAMELEN(dstr);
3855 (len > 1 && name[len-2] == ':' && name[len-1] == ':')
3856 || (len == 1 && name[0] == ':')
3858 && (!dref || HvENAME_get(dref))
3861 (HV *)sref, (HV *)dref,
3867 stype == SVt_PVAV && sref != dref
3868 && strEQ(GvNAME((GV*)dstr), "ISA")
3869 /* The stash may have been detached from the symbol table, so
3870 check its name before doing anything. */
3871 && GvSTASH(dstr) && HvENAME(GvSTASH(dstr))
3874 MAGIC * const omg = dref && SvSMAGICAL(dref)
3875 ? mg_find(dref, PERL_MAGIC_isa)
3877 if (SvSMAGICAL(sref) && (mg = mg_find(sref, PERL_MAGIC_isa))) {
3878 if (SvTYPE(mg->mg_obj) != SVt_PVAV) {
3879 AV * const ary = newAV();
3880 av_push(ary, mg->mg_obj); /* takes the refcount */
3881 mg->mg_obj = (SV *)ary;
3884 if (SvTYPE(omg->mg_obj) == SVt_PVAV) {
3885 SV **svp = AvARRAY((AV *)omg->mg_obj);
3886 I32 items = AvFILLp((AV *)omg->mg_obj) + 1;
3890 SvREFCNT_inc_simple_NN(*svp++)
3896 SvREFCNT_inc_simple_NN(omg->mg_obj)
3900 av_push((AV *)mg->mg_obj,SvREFCNT_inc_simple_NN(dstr));
3905 sref, omg ? omg->mg_obj : dstr, PERL_MAGIC_isa, NULL, 0
3907 mg = mg_find(sref, PERL_MAGIC_isa);
3909 /* Since the *ISA assignment could have affected more than
3910 one stash, don't call mro_isa_changed_in directly, but let
3911 magic_clearisa do it for us, as it already has the logic for
3912 dealing with globs vs arrays of globs. */
3914 Perl_magic_clearisa(aTHX_ NULL, mg);
3919 if (SvTAINTED(sstr))
3925 Perl_sv_setsv_flags(pTHX_ SV *dstr, register SV* sstr, const I32 flags)
3928 register U32 sflags;
3930 register svtype stype;
3932 PERL_ARGS_ASSERT_SV_SETSV_FLAGS;
3937 if (SvIS_FREED(dstr)) {
3938 Perl_croak(aTHX_ "panic: attempt to copy value %" SVf
3939 " to a freed scalar %p", SVfARG(sstr), (void *)dstr);
3941 SV_CHECK_THINKFIRST_COW_DROP(dstr);
3943 sstr = &PL_sv_undef;
3944 if (SvIS_FREED(sstr)) {
3945 Perl_croak(aTHX_ "panic: attempt to copy freed scalar %p to %p",
3946 (void*)sstr, (void*)dstr);
3948 stype = SvTYPE(sstr);
3949 dtype = SvTYPE(dstr);
3951 (void)SvAMAGIC_off(dstr);
3954 /* need to nuke the magic */
3955 sv_unmagic(dstr, PERL_MAGIC_vstring);
3958 /* There's a lot of redundancy below but we're going for speed here */
3963 if (dtype != SVt_PVGV && dtype != SVt_PVLV) {
3964 (void)SvOK_off(dstr);
3972 sv_upgrade(dstr, SVt_IV);
3976 sv_upgrade(dstr, SVt_PVIV);
3980 goto end_of_first_switch;
3982 (void)SvIOK_only(dstr);
3983 SvIV_set(dstr, SvIVX(sstr));
3986 /* SvTAINTED can only be true if the SV has taint magic, which in
3987 turn means that the SV type is PVMG (or greater). This is the
3988 case statement for SVt_IV, so this cannot be true (whatever gcov
3990 assert(!SvTAINTED(sstr));
3995 if (dtype < SVt_PV && dtype != SVt_IV)
3996 sv_upgrade(dstr, SVt_IV);
4004 sv_upgrade(dstr, SVt_NV);
4008 sv_upgrade(dstr, SVt_PVNV);
4012 goto end_of_first_switch;
4014 SvNV_set(dstr, SvNVX(sstr));
4015 (void)SvNOK_only(dstr);
4016 /* SvTAINTED can only be true if the SV has taint magic, which in
4017 turn means that the SV type is PVMG (or greater). This is the
4018 case statement for SVt_NV, so this cannot be true (whatever gcov
4020 assert(!SvTAINTED(sstr));
4026 #ifdef PERL_OLD_COPY_ON_WRITE
4027 if ((SvFLAGS(sstr) & CAN_COW_MASK) == CAN_COW_FLAGS) {
4028 if (dtype < SVt_PVIV)
4029 sv_upgrade(dstr, SVt_PVIV);
4036 sv_upgrade(dstr, SVt_PV);
4039 if (dtype < SVt_PVIV)
4040 sv_upgrade(dstr, SVt_PVIV);
4043 if (dtype < SVt_PVNV)
4044 sv_upgrade(dstr, SVt_PVNV);
4048 const char * const type = sv_reftype(sstr,0);
4050 /* diag_listed_as: Bizarre copy of %s */
4051 Perl_croak(aTHX_ "Bizarre copy of %s in %s", type, OP_DESC(PL_op));
4053 Perl_croak(aTHX_ "Bizarre copy of %s", type);
4058 if (dtype < SVt_REGEXP)
4059 sv_upgrade(dstr, SVt_REGEXP);
4062 /* case SVt_BIND: */
4066 if (SvGMAGICAL(sstr) && (flags & SV_GMAGIC)) {
4068 if (SvTYPE(sstr) != stype)
4069 stype = SvTYPE(sstr);
4071 if (isGV_with_GP(sstr) && dtype <= SVt_PVLV) {
4072 glob_assign_glob(dstr, sstr, dtype);
4075 if (stype == SVt_PVLV)
4076 SvUPGRADE(dstr, SVt_PVNV);
4078 SvUPGRADE(dstr, (svtype)stype);
4080 end_of_first_switch:
4082 /* dstr may have been upgraded. */
4083 dtype = SvTYPE(dstr);
4084 sflags = SvFLAGS(sstr);
4086 if (dtype == SVt_PVCV || dtype == SVt_PVFM) {
4087 /* Assigning to a subroutine sets the prototype. */
4090 const char *const ptr = SvPV_const(sstr, len);
4092 SvGROW(dstr, len + 1);
4093 Copy(ptr, SvPVX(dstr), len + 1, char);
4094 SvCUR_set(dstr, len);
4096 SvFLAGS(dstr) |= sflags & SVf_UTF8;
4097 CvAUTOLOAD_off(dstr);
4101 } else if (dtype == SVt_PVAV || dtype == SVt_PVHV) {
4102 const char * const type = sv_reftype(dstr,0);
4104 /* diag_listed_as: Cannot copy to %s */
4105 Perl_croak(aTHX_ "Cannot copy to %s in %s", type, OP_DESC(PL_op));
4107 Perl_croak(aTHX_ "Cannot copy to %s", type);
4108 } else if (sflags & SVf_ROK) {
4109 if (isGV_with_GP(dstr)
4110 && SvTYPE(SvRV(sstr)) == SVt_PVGV && isGV_with_GP(SvRV(sstr))) {
4113 if (GvIMPORTED(dstr) != GVf_IMPORTED
4114 && CopSTASH_ne(PL_curcop, GvSTASH(dstr)))
4116 GvIMPORTED_on(dstr);
4121 glob_assign_glob(dstr, sstr, dtype);
4125 if (dtype >= SVt_PV) {
4126 if (isGV_with_GP(dstr)) {
4127 glob_assign_ref(dstr, sstr);
4130 if (SvPVX_const(dstr)) {
4136 (void)SvOK_off(dstr);
4137 SvRV_set(dstr, SvREFCNT_inc(SvRV(sstr)));
4138 SvFLAGS(dstr) |= sflags & SVf_ROK;
4139 assert(!(sflags & SVp_NOK));
4140 assert(!(sflags & SVp_IOK));
4141 assert(!(sflags & SVf_NOK));
4142 assert(!(sflags & SVf_IOK));
4144 else if (isGV_with_GP(dstr)) {
4145 if (!(sflags & SVf_OK)) {
4146 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
4147 "Undefined value assigned to typeglob");
4150 GV *gv = gv_fetchsv_nomg(sstr, GV_ADD, SVt_PVGV);
4151 if (dstr != (const SV *)gv) {
4152 const char * const name = GvNAME((const GV *)dstr);
4153 const STRLEN len = GvNAMELEN(dstr);
4154 HV *old_stash = NULL;
4155 bool reset_isa = FALSE;
4156 if ((len > 1 && name[len-2] == ':' && name[len-1] == ':')
4157 || (len == 1 && name[0] == ':')) {
4158 /* Set aside the old stash, so we can reset isa caches
4159 on its subclasses. */
4160 if((old_stash = GvHV(dstr))) {
4161 /* Make sure we do not lose it early. */
4162 SvREFCNT_inc_simple_void_NN(
4163 sv_2mortal((SV *)old_stash)
4170 gp_free(MUTABLE_GV(dstr));
4171 GvGP_set(dstr, gp_ref(GvGP(gv)));
4174 HV * const stash = GvHV(dstr);
4176 old_stash ? (HV *)HvENAME_get(old_stash) : stash
4186 else if (dtype == SVt_REGEXP && stype == SVt_REGEXP) {
4187 reg_temp_copy((REGEXP*)dstr, (REGEXP*)sstr);
4189 else if (sflags & SVp_POK) {
4193 * Check to see if we can just swipe the string. If so, it's a
4194 * possible small lose on short strings, but a big win on long ones.
4195 * It might even be a win on short strings if SvPVX_const(dstr)
4196 * has to be allocated and SvPVX_const(sstr) has to be freed.
4197 * Likewise if we can set up COW rather than doing an actual copy, we
4198 * drop to the else clause, as the swipe code and the COW setup code
4199 * have much in common.
4202 /* Whichever path we take through the next code, we want this true,
4203 and doing it now facilitates the COW check. */
4204 (void)SvPOK_only(dstr);
4207 /* If we're already COW then this clause is not true, and if COW
4208 is allowed then we drop down to the else and make dest COW
4209 with us. If caller hasn't said that we're allowed to COW
4210 shared hash keys then we don't do the COW setup, even if the
4211 source scalar is a shared hash key scalar. */
4212 (((flags & SV_COW_SHARED_HASH_KEYS)
4213 ? (sflags & (SVf_FAKE|SVf_READONLY)) != (SVf_FAKE|SVf_READONLY)
4214 : 1 /* If making a COW copy is forbidden then the behaviour we
4215 desire is as if the source SV isn't actually already
4216 COW, even if it is. So we act as if the source flags
4217 are not COW, rather than actually testing them. */
4219 #ifndef PERL_OLD_COPY_ON_WRITE
4220 /* The change that added SV_COW_SHARED_HASH_KEYS makes the logic
4221 when PERL_OLD_COPY_ON_WRITE is defined a little wrong.
4222 Conceptually PERL_OLD_COPY_ON_WRITE being defined should
4223 override SV_COW_SHARED_HASH_KEYS, because it means "always COW"
4224 but in turn, it's somewhat dead code, never expected to go
4225 live, but more kept as a placeholder on how to do it better
4226 in a newer implementation. */
4227 /* If we are COW and dstr is a suitable target then we drop down
4228 into the else and make dest a COW of us. */
4229 || (SvFLAGS(dstr) & CAN_COW_MASK) != CAN_COW_FLAGS
4234 (sflags & SVs_TEMP) && /* slated for free anyway? */
4235 !(sflags & SVf_OOK) && /* and not involved in OOK hack? */
4236 (!(flags & SV_NOSTEAL)) &&
4237 /* and we're allowed to steal temps */
4238 SvREFCNT(sstr) == 1 && /* and no other references to it? */
4239 SvLEN(sstr)) /* and really is a string */
4240 #ifdef PERL_OLD_COPY_ON_WRITE
4241 && ((flags & SV_COW_SHARED_HASH_KEYS)
4242 ? (!((sflags & CAN_COW_MASK) == CAN_COW_FLAGS
4243 && (SvFLAGS(dstr) & CAN_COW_MASK) == CAN_COW_FLAGS
4244 && SvTYPE(sstr) >= SVt_PVIV && SvTYPE(sstr) != SVt_PVFM))
4248 /* Failed the swipe test, and it's not a shared hash key either.
4249 Have to copy the string. */
4250 STRLEN len = SvCUR(sstr);
4251 SvGROW(dstr, len + 1); /* inlined from sv_setpvn */
4252 Move(SvPVX_const(sstr),SvPVX(dstr),len,char);
4253 SvCUR_set(dstr, len);
4254 *SvEND(dstr) = '\0';
4256 /* If PERL_OLD_COPY_ON_WRITE is not defined, then isSwipe will always
4258 /* Either it's a shared hash key, or it's suitable for
4259 copy-on-write or we can swipe the string. */
4261 PerlIO_printf(Perl_debug_log, "Copy on write: sstr --> dstr\n");
4265 #ifdef PERL_OLD_COPY_ON_WRITE
4267 if ((sflags & (SVf_FAKE | SVf_READONLY))
4268 != (SVf_FAKE | SVf_READONLY)) {
4269 SvREADONLY_on(sstr);
4271 /* Make the source SV into a loop of 1.
4272 (about to become 2) */
4273 SV_COW_NEXT_SV_SET(sstr, sstr);
4277 /* Initial code is common. */
4278 if (SvPVX_const(dstr)) { /* we know that dtype >= SVt_PV */
4283 /* making another shared SV. */
4284 STRLEN cur = SvCUR(sstr);
4285 STRLEN len = SvLEN(sstr);
4286 #ifdef PERL_OLD_COPY_ON_WRITE
4288 assert (SvTYPE(dstr) >= SVt_PVIV);
4289 /* SvIsCOW_normal */
4290 /* splice us in between source and next-after-source. */
4291 SV_COW_NEXT_SV_SET(dstr, SV_COW_NEXT_SV(sstr));
4292 SV_COW_NEXT_SV_SET(sstr, dstr);
4293 SvPV_set(dstr, SvPVX_mutable(sstr));
4297 /* SvIsCOW_shared_hash */
4298 DEBUG_C(PerlIO_printf(Perl_debug_log,
4299 "Copy on write: Sharing hash\n"));
4301 assert (SvTYPE(dstr) >= SVt_PV);
4303 HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr)))));
4305 SvLEN_set(dstr, len);
4306 SvCUR_set(dstr, cur);
4307 SvREADONLY_on(dstr);
4311 { /* Passes the swipe test. */
4312 SvPV_set(dstr, SvPVX_mutable(sstr));
4313 SvLEN_set(dstr, SvLEN(sstr));
4314 SvCUR_set(dstr, SvCUR(sstr));
4317 (void)SvOK_off(sstr); /* NOTE: nukes most SvFLAGS on sstr */
4318 SvPV_set(sstr, NULL);
4324 if (sflags & SVp_NOK) {
4325 SvNV_set(dstr, SvNVX(sstr));
4327 if (sflags & SVp_IOK) {
4328 SvIV_set(dstr, SvIVX(sstr));
4329 /* Must do this otherwise some other overloaded use of 0x80000000
4330 gets confused. I guess SVpbm_VALID */
4331 if (sflags & SVf_IVisUV)
4334 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_NOK|SVp_NOK|SVf_UTF8);
4336 const MAGIC * const smg = SvVSTRING_mg(sstr);
4338 sv_magic(dstr, NULL, PERL_MAGIC_vstring,
4339 smg->mg_ptr, smg->mg_len);
4340 SvRMAGICAL_on(dstr);
4344 else if (sflags & (SVp_IOK|SVp_NOK)) {
4345 (void)SvOK_off(dstr);
4346 SvFLAGS(dstr) |= sflags & (SVf_IOK|SVp_IOK|SVf_IVisUV|SVf_NOK|SVp_NOK);
4347 if (sflags & SVp_IOK) {
4348 /* XXXX Do we want to set IsUV for IV(ROK)? Be extra safe... */
4349 SvIV_set(dstr, SvIVX(sstr));
4351 if (sflags & SVp_NOK) {
4352 SvNV_set(dstr, SvNVX(sstr));
4356 if (isGV_with_GP(sstr)) {
4357 gv_efullname3(dstr, MUTABLE_GV(sstr), "*");
4360 (void)SvOK_off(dstr);
4362 if (SvTAINTED(sstr))
4367 =for apidoc sv_setsv_mg
4369 Like C<sv_setsv>, but also handles 'set' magic.
4375 Perl_sv_setsv_mg(pTHX_ SV *const dstr, register SV *const sstr)
4377 PERL_ARGS_ASSERT_SV_SETSV_MG;
4379 sv_setsv(dstr,sstr);
4383 #ifdef PERL_OLD_COPY_ON_WRITE
4385 Perl_sv_setsv_cow(pTHX_ SV *dstr, SV *sstr)
4387 STRLEN cur = SvCUR(sstr);
4388 STRLEN len = SvLEN(sstr);
4389 register char *new_pv;
4391 PERL_ARGS_ASSERT_SV_SETSV_COW;
4394 PerlIO_printf(Perl_debug_log, "Fast copy on write: %p -> %p\n",
4395 (void*)sstr, (void*)dstr);
4402 if (SvTHINKFIRST(dstr))
4403 sv_force_normal_flags(dstr, SV_COW_DROP_PV);
4404 else if (SvPVX_const(dstr))
4405 Safefree(SvPVX_const(dstr));
4409 SvUPGRADE(dstr, SVt_PVIV);
4411 assert (SvPOK(sstr));
4412 assert (SvPOKp(sstr));
4413 assert (!SvIOK(sstr));
4414 assert (!SvIOKp(sstr));
4415 assert (!SvNOK(sstr));
4416 assert (!SvNOKp(sstr));
4418 if (SvIsCOW(sstr)) {
4420 if (SvLEN(sstr) == 0) {
4421 /* source is a COW shared hash key. */
4422 DEBUG_C(PerlIO_printf(Perl_debug_log,
4423 "Fast copy on write: Sharing hash\n"));
4424 new_pv = HEK_KEY(share_hek_hek(SvSHARED_HEK_FROM_PV(SvPVX_const(sstr))));
4427 SV_COW_NEXT_SV_SET(dstr, SV_COW_NEXT_SV(sstr));
4429 assert ((SvFLAGS(sstr) & CAN_COW_MASK) == CAN_COW_FLAGS);
4430 SvUPGRADE(sstr, SVt_PVIV);
4431 SvREADONLY_on(sstr);
4433 DEBUG_C(PerlIO_printf(Perl_debug_log,
4434 "Fast copy on write: Converting sstr to COW\n"));
4435 SV_COW_NEXT_SV_SET(dstr, sstr);
4437 SV_COW_NEXT_SV_SET(sstr, dstr);
4438 new_pv = SvPVX_mutable(sstr);
4441 SvPV_set(dstr, new_pv);
4442 SvFLAGS(dstr) = (SVt_PVIV|SVf_POK|SVp_POK|SVf_FAKE|SVf_READONLY);
4445 SvLEN_set(dstr, len);
4446 SvCUR_set(dstr, cur);
4455 =for apidoc sv_setpvn
4457 Copies a string into an SV. The C<len> parameter indicates the number of
4458 bytes to be copied. If the C<ptr> argument is NULL the SV will become
4459 undefined. Does not handle 'set' magic. See C<sv_setpvn_mg>.
4465 Perl_sv_setpvn(pTHX_ register SV *const sv, register const char *const ptr, register const STRLEN len)
4468 register char *dptr;
4470 PERL_ARGS_ASSERT_SV_SETPVN;
4472 SV_CHECK_THINKFIRST_COW_DROP(sv);
4478 /* len is STRLEN which is unsigned, need to copy to signed */
4481 Perl_croak(aTHX_ "panic: sv_setpvn called with negative strlen %"
4484 SvUPGRADE(sv, SVt_PV);
4486 dptr = SvGROW(sv, len + 1);
4487 Move(ptr,dptr,len,char);
4490 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4492 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4496 =for apidoc sv_setpvn_mg
4498 Like C<sv_setpvn>, but also handles 'set' magic.
4504 Perl_sv_setpvn_mg(pTHX_ register SV *const sv, register const char *const ptr, register const STRLEN len)
4506 PERL_ARGS_ASSERT_SV_SETPVN_MG;
4508 sv_setpvn(sv,ptr,len);
4513 =for apidoc sv_setpv
4515 Copies a string into an SV. The string must be null-terminated. Does not
4516 handle 'set' magic. See C<sv_setpv_mg>.
4522 Perl_sv_setpv(pTHX_ register SV *const sv, register const char *const ptr)
4525 register STRLEN len;
4527 PERL_ARGS_ASSERT_SV_SETPV;
4529 SV_CHECK_THINKFIRST_COW_DROP(sv);
4535 SvUPGRADE(sv, SVt_PV);
4537 SvGROW(sv, len + 1);
4538 Move(ptr,SvPVX(sv),len+1,char);
4540 (void)SvPOK_only_UTF8(sv); /* validate pointer */
4542 if (SvTYPE(sv) == SVt_PVCV) CvAUTOLOAD_off(sv);
4546 =for apidoc sv_setpv_mg
4548 Like C<sv_setpv>, but also handles 'set' magic.
4554 Perl_sv_setpv_mg(pTHX_ register SV *const sv, register const char *const ptr)
4556 PERL_ARGS_ASSERT_SV_SETPV_MG;
4563 Perl_sv_sethek(pTHX_ register SV *const sv, const HEK *const hek)
4567 PERL_ARGS_ASSERT_SV_SETHEK;
4573 if (HEK_LEN(hek) == HEf_SVKEY) {
4574 sv_setsv(sv, *(SV**)HEK_KEY(hek));
4577 const int flags = HEK_FLAGS(hek);
4578 if (flags & HVhek_WASUTF8) {
4579 STRLEN utf8_len = HEK_LEN(hek);
4580 char *as_utf8 = (char *)bytes_to_utf8((U8*)HEK_KEY(hek), &utf8_len);
4581 sv_usepvn_flags(sv, as_utf8, utf8_len, SV_HAS_TRAILING_NUL);
4584 } else if (flags & (HVhek_REHASH|HVhek_UNSHARED)) {
4585 sv_setpvn(sv, HEK_KEY(hek), HEK_LEN(hek));
4588 else SvUTF8_off(sv);
4592 SV_CHECK_THINKFIRST_COW_DROP(sv);
4593 SvUPGRADE(sv, SVt_PV);
4594 SvPV_set(sv,(char *)HEK_KEY(share_hek_hek(hek)));
4595 SvCUR_set(sv, HEK_LEN(hek));
4602 else SvUTF8_off(sv);
4610 =for apidoc sv_usepvn_flags
4612 Tells an SV to use C<ptr> to find its string value. Normally the
4613 string is stored inside the SV but sv_usepvn allows the SV to use an
4614 outside string. The C<ptr> should point to memory that was allocated
4615 by C<malloc>. It must be the start of a mallocked block
4616 of memory, and not a pointer to the middle of it. The
4617 string length, C<len>, must be supplied. By default
4618 this function will realloc (i.e. move) the memory pointed to by C<ptr>,
4619 so that pointer should not be freed or used by the programmer after
4620 giving it to sv_usepvn, and neither should any pointers from "behind"
4621 that pointer (e.g. ptr + 1) be used.
4623 If C<flags> & SV_SMAGIC is true, will call SvSETMAGIC. If C<flags> &
4624 SV_HAS_TRAILING_NUL is true, then C<ptr[len]> must be NUL, and the realloc
4625 will be skipped (i.e. the buffer is actually at least 1 byte longer than
4626 C<len>, and already meets the requirements for storing in C<SvPVX>).