3 * Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000,
4 * 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008 by Larry Wall and others
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
12 * 'It's a big house this, and very peculiar. Always a bit more
13 * to discover, and no knowing what you'll find round a corner.
14 * And Elves, sir!' --Samwise Gamgee
16 * [p.225 of _The Lord of the Rings_, II/i: "Many Meetings"]
19 /* This file contains general pp ("push/pop") functions that execute the
20 * opcodes that make up a perl program. A typical pp function expects to
21 * find its arguments on the stack, and usually pushes its results onto
22 * the stack, hence the 'pp' terminology. Each OP structure contains
23 * a pointer to the relevant pp_foo() function.
32 #include "regcharclass.h"
34 /* XXX I can't imagine anyone who doesn't have this actually _needs_
35 it, since pid_t is an integral type.
38 #ifdef NEED_GETPID_PROTO
39 extern Pid_t getpid (void);
43 * Some BSDs and Cygwin default to POSIX math instead of IEEE.
44 * This switches them over to IEEE.
46 #if defined(LIBM_LIB_VERSION)
47 _LIB_VERSION_TYPE _LIB_VERSION = _IEEE_;
50 static const STRLEN small_mu_len = sizeof(GREEK_SMALL_LETTER_MU_UTF8) - 1;
51 static const STRLEN capital_iota_len = sizeof(GREEK_CAPITAL_LETTER_IOTA_UTF8) - 1;
53 /* variations on pp_null */
58 if (GIMME_V == G_SCALAR)
65 /* This is also called directly by pp_lvavref. */
70 assert(SvTYPE(TARG) == SVt_PVAV);
71 if (UNLIKELY( PL_op->op_private & OPpLVAL_INTRO ))
72 if (LIKELY( !(PL_op->op_private & OPpPAD_STATE) ))
73 SAVECLEARSV(PAD_SVl(PL_op->op_targ));
75 if (PL_op->op_flags & OPf_REF) {
78 } else if (PL_op->op_private & OPpMAYBE_LVSUB) {
79 const I32 flags = is_lvalue_sub();
80 if (flags && !(flags & OPpENTERSUB_INARGS)) {
81 if (GIMME_V == G_SCALAR)
82 /* diag_listed_as: Can't return %s to lvalue scalar context */
83 Perl_croak(aTHX_ "Can't return array to lvalue scalar context");
89 if (gimme == G_ARRAY) {
90 /* XXX see also S_pushav in pp_hot.c */
91 const Size_t maxarg = AvFILL(MUTABLE_AV(TARG)) + 1;
93 if (SvMAGICAL(TARG)) {
95 for (i=0; i < maxarg; i++) {
96 SV * const * const svp = av_fetch(MUTABLE_AV(TARG), i, FALSE);
97 SP[i+1] = (svp) ? *svp : &PL_sv_undef;
102 for (i=0; i < (PADOFFSET)maxarg; i++) {
103 SV * const sv = AvARRAY((const AV *)TARG)[i];
104 SP[i+1] = sv ? sv : &PL_sv_undef;
109 else if (gimme == G_SCALAR) {
110 SV* const sv = sv_newmortal();
111 const SSize_t maxarg = AvFILL(MUTABLE_AV(TARG)) + 1;
112 sv_setiv(sv, maxarg);
123 assert(SvTYPE(TARG) == SVt_PVHV);
125 if (UNLIKELY( PL_op->op_private & OPpLVAL_INTRO ))
126 if (LIKELY( !(PL_op->op_private & OPpPAD_STATE) ))
127 SAVECLEARSV(PAD_SVl(PL_op->op_targ));
128 if (PL_op->op_flags & OPf_REF)
130 else if (PL_op->op_private & OPpMAYBE_LVSUB) {
131 const I32 flags = is_lvalue_sub();
132 if (flags && !(flags & OPpENTERSUB_INARGS)) {
133 if (GIMME_V == G_SCALAR)
134 /* diag_listed_as: Can't return %s to lvalue scalar context */
135 Perl_croak(aTHX_ "Can't return hash to lvalue scalar context");
140 if (gimme == G_ARRAY) {
141 RETURNOP(Perl_do_kv(aTHX));
143 else if ((PL_op->op_private & OPpTRUEBOOL
144 || ( PL_op->op_private & OPpMAYBE_TRUEBOOL
145 && block_gimme() == G_VOID ))
146 && (!SvRMAGICAL(TARG) || !mg_find(TARG, PERL_MAGIC_tied)))
147 SETs(HvUSEDKEYS(TARG) ? &PL_sv_yes : sv_2mortal(newSViv(0)));
148 else if (gimme == G_SCALAR) {
149 SV* const sv = Perl_hv_scalar(aTHX_ MUTABLE_HV(TARG));
158 assert(SvTYPE(TARG) == SVt_PVCV);
166 SvPADSTALE_off(TARG);
173 CV * const protocv = PadnamePROTOCV(
174 PadlistNAMESARRAY(CvPADLIST(find_runcv(NULL)))[ARGTARG]
176 assert(SvTYPE(TARG) == SVt_PVCV);
178 if (CvISXSUB(protocv)) { /* constant */
179 /* XXX Should we clone it here? */
180 /* If this changes to use SAVECLEARSV, we can move the SAVECLEARSV
181 to introcv and remove the SvPADSTALE_off. */
182 SAVEPADSVANDMORTALIZE(ARGTARG);
183 PAD_SVl(ARGTARG) = SvREFCNT_inc_simple_NN(protocv);
186 if (CvROOT(protocv)) {
187 assert(CvCLONE(protocv));
188 assert(!CvCLONED(protocv));
190 cv_clone_into(protocv,(CV *)TARG);
191 SAVECLEARSV(PAD_SVl(ARGTARG));
198 /* In some cases this function inspects PL_op. If this function is called
199 for new op types, more bool parameters may need to be added in place of
202 When noinit is true, the absence of a gv will cause a retval of undef.
203 This is unrelated to the cv-to-gv assignment case.
207 S_rv2gv(pTHX_ SV *sv, const bool vivify_sv, const bool strict,
210 if (!isGV(sv) || SvFAKE(sv)) SvGETMAGIC(sv);
213 sv = amagic_deref_call(sv, to_gv_amg);
217 if (SvTYPE(sv) == SVt_PVIO) {
218 GV * const gv = MUTABLE_GV(sv_newmortal());
219 gv_init(gv, 0, "__ANONIO__", 10, 0);
220 GvIOp(gv) = MUTABLE_IO(sv);
221 SvREFCNT_inc_void_NN(sv);
224 else if (!isGV_with_GP(sv)) {
225 Perl_die(aTHX_ "Not a GLOB reference");
229 if (!isGV_with_GP(sv)) {
231 /* If this is a 'my' scalar and flag is set then vivify
234 if (vivify_sv && sv != &PL_sv_undef) {
237 Perl_croak_no_modify();
238 if (cUNOP->op_targ) {
239 SV * const namesv = PAD_SV(cUNOP->op_targ);
240 HV *stash = CopSTASH(PL_curcop);
241 if (SvTYPE(stash) != SVt_PVHV) stash = NULL;
242 gv = MUTABLE_GV(newSV(0));
243 gv_init_sv(gv, stash, namesv, 0);
246 const char * const name = CopSTASHPV(PL_curcop);
247 gv = newGVgen_flags(name,
248 HvNAMEUTF8(CopSTASH(PL_curcop)) ? SVf_UTF8 : 0 );
249 SvREFCNT_inc_simple_void_NN(gv);
251 prepare_SV_for_RV(sv);
252 SvRV_set(sv, MUTABLE_SV(gv));
257 if (PL_op->op_flags & OPf_REF || strict) {
258 Perl_die(aTHX_ PL_no_usym, "a symbol");
260 if (ckWARN(WARN_UNINITIALIZED))
266 if (!(sv = MUTABLE_SV(gv_fetchsv_nomg(
267 sv, GV_ADDMG, SVt_PVGV
276 (SvPOKp(sv) && SvCUR(sv)>32 ? "..." : ""),
280 if ((PL_op->op_private & (OPpLVAL_INTRO|OPpDONT_INIT_GV))
281 == OPpDONT_INIT_GV) {
282 /* We are the target of a coderef assignment. Return
283 the scalar unchanged, and let pp_sasssign deal with
287 sv = MUTABLE_SV(gv_fetchsv_nomg(sv, GV_ADD, SVt_PVGV));
289 /* FAKE globs in the symbol table cause weird bugs (#77810) */
293 if (SvFAKE(sv) && !(PL_op->op_private & OPpALLOW_FAKE)) {
294 SV *newsv = sv_newmortal();
295 sv_setsv_flags(newsv, sv, 0);
307 sv, PL_op->op_private & OPpDEREF,
308 PL_op->op_private & HINT_STRICT_REFS,
309 ((PL_op->op_flags & OPf_SPECIAL) && !(PL_op->op_flags & OPf_MOD))
310 || PL_op->op_type == OP_READLINE
312 if (PL_op->op_private & OPpLVAL_INTRO)
313 save_gp(MUTABLE_GV(sv), !(PL_op->op_flags & OPf_SPECIAL));
318 /* Helper function for pp_rv2sv and pp_rv2av */
320 Perl_softref2xv(pTHX_ SV *const sv, const char *const what,
321 const svtype type, SV ***spp)
325 PERL_ARGS_ASSERT_SOFTREF2XV;
327 if (PL_op->op_private & HINT_STRICT_REFS) {
329 Perl_die(aTHX_ PL_no_symref_sv, sv,
330 (SvPOKp(sv) && SvCUR(sv)>32 ? "..." : ""), what);
332 Perl_die(aTHX_ PL_no_usym, what);
336 PL_op->op_flags & OPf_REF
338 Perl_die(aTHX_ PL_no_usym, what);
339 if (ckWARN(WARN_UNINITIALIZED))
341 if (type != SVt_PV && GIMME_V == G_ARRAY) {
345 **spp = &PL_sv_undef;
348 if ((PL_op->op_flags & OPf_SPECIAL) &&
349 !(PL_op->op_flags & OPf_MOD))
351 if (!(gv = gv_fetchsv_nomg(sv, GV_ADDMG, type)))
353 **spp = &PL_sv_undef;
358 gv = gv_fetchsv_nomg(sv, GV_ADD, type);
371 sv = amagic_deref_call(sv, to_sv_amg);
375 if (SvTYPE(sv) >= SVt_PVAV)
376 DIE(aTHX_ "Not a SCALAR reference");
381 if (!isGV_with_GP(gv)) {
382 gv = Perl_softref2xv(aTHX_ sv, "a SCALAR", SVt_PV, &sp);
388 if (PL_op->op_flags & OPf_MOD) {
389 if (PL_op->op_private & OPpLVAL_INTRO) {
390 if (cUNOP->op_first->op_type == OP_NULL)
391 sv = save_scalar(MUTABLE_GV(TOPs));
393 sv = save_scalar(gv);
395 Perl_croak(aTHX_ "%s", PL_no_localize_ref);
397 else if (PL_op->op_private & OPpDEREF)
398 sv = vivify_ref(sv, PL_op->op_private & OPpDEREF);
407 AV * const av = MUTABLE_AV(TOPs);
408 const I32 lvalue = PL_op->op_flags & OPf_MOD || LVRET;
410 SV ** const svp = Perl_av_arylen_p(aTHX_ MUTABLE_AV(av));
412 *svp = newSV_type(SVt_PVMG);
413 sv_magic(*svp, MUTABLE_SV(av), PERL_MAGIC_arylen, NULL, 0);
417 SETs(sv_2mortal(newSViv(AvFILL(MUTABLE_AV(av)))));
426 if (PL_op->op_flags & OPf_MOD || LVRET) {
427 SV * const ret = sv_2mortal(newSV_type(SVt_PVLV));/* Not TARG RT#67838 */
428 sv_magic(ret, NULL, PERL_MAGIC_pos, NULL, 0);
430 LvTARG(ret) = SvREFCNT_inc_simple(sv);
431 SETs(ret); /* no SvSETMAGIC */
434 const MAGIC * const mg = mg_find_mglob(sv);
435 if (mg && mg->mg_len != -1) {
437 STRLEN i = mg->mg_len;
438 if (mg->mg_flags & MGf_BYTES && DO_UTF8(sv))
439 i = sv_pos_b2u_flags(sv, i, SV_GMAGIC|SV_CONST_RETURN);
453 const I32 flags = (PL_op->op_flags & OPf_SPECIAL)
455 : ((PL_op->op_private & (OPpLVAL_INTRO|OPpMAY_RETURN_CONSTANT))
456 == OPpMAY_RETURN_CONSTANT)
459 /* We usually try to add a non-existent subroutine in case of AUTOLOAD. */
460 /* (But not in defined().) */
462 CV *cv = sv_2cv(TOPs, &stash_unused, &gv, flags);
464 else if ((flags == (GV_ADD|GV_NOEXPAND)) && gv && SvROK(gv)) {
465 cv = SvTYPE(SvRV(gv)) == SVt_PVCV
466 ? MUTABLE_CV(SvRV(gv))
470 cv = MUTABLE_CV(&PL_sv_undef);
471 SETs(MUTABLE_SV(cv));
481 SV *ret = &PL_sv_undef;
483 if (SvGMAGICAL(TOPs)) SETs(sv_mortalcopy(TOPs));
484 if (SvPOK(TOPs) && SvCUR(TOPs) >= 7) {
485 const char * s = SvPVX_const(TOPs);
486 if (strnEQ(s, "CORE::", 6)) {
487 const int code = keyword(s + 6, SvCUR(TOPs) - 6, 1);
489 DIE(aTHX_ "Can't find an opnumber for \"%"UTF8f"\"",
490 UTF8fARG(SvFLAGS(TOPs) & SVf_UTF8, SvCUR(TOPs)-6, s+6));
492 SV * const sv = core_prototype(NULL, s + 6, code, NULL);
498 cv = sv_2cv(TOPs, &stash, &gv, 0);
500 ret = newSVpvn_flags(
501 CvPROTO(cv), CvPROTOLEN(cv), SVs_TEMP | SvUTF8(cv)
511 CV *cv = MUTABLE_CV(PAD_SV(PL_op->op_targ));
513 cv = MUTABLE_CV(sv_2mortal(MUTABLE_SV(cv_clone(cv))));
515 PUSHs(MUTABLE_SV(cv));
529 if (GIMME_V != G_ARRAY) {
535 *MARK = &PL_sv_undef;
537 *MARK = refto(*MARK);
541 EXTEND_MORTAL(SP - MARK);
543 *MARK = refto(*MARK);
548 S_refto(pTHX_ SV *sv)
552 PERL_ARGS_ASSERT_REFTO;
554 if (SvTYPE(sv) == SVt_PVLV && LvTYPE(sv) == 'y') {
557 if (!(sv = LvTARG(sv)))
560 SvREFCNT_inc_void_NN(sv);
562 else if (SvTYPE(sv) == SVt_PVAV) {
563 if (!AvREAL((const AV *)sv) && AvREIFY((const AV *)sv))
564 av_reify(MUTABLE_AV(sv));
566 SvREFCNT_inc_void_NN(sv);
568 else if (SvPADTMP(sv)) {
573 SvREFCNT_inc_void_NN(sv);
576 sv_upgrade(rv, SVt_IV);
585 SV * const sv = TOPs;
593 /* use the return value that is in a register, its the same as TARG */
594 TARG = sv_ref(TARG,SvRV(sv),TRUE);
609 stash = CopSTASH(PL_curcop);
610 if (SvTYPE(stash) != SVt_PVHV)
611 Perl_croak(aTHX_ "Attempt to bless into a freed package");
614 SV * const ssv = POPs;
618 if (!ssv) goto curstash;
621 if (!SvAMAGIC(ssv)) {
623 Perl_croak(aTHX_ "Attempt to bless into a reference");
625 /* SvAMAGIC is on here, but it only means potentially overloaded,
626 so after stringification: */
627 ptr = SvPV_nomg_const(ssv,len);
628 /* We need to check the flag again: */
629 if (!SvAMAGIC(ssv)) goto frog;
631 else ptr = SvPV_nomg_const(ssv,len);
633 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
634 "Explicit blessing to '' (assuming package main)");
635 stash = gv_stashpvn(ptr, len, GV_ADD|SvUTF8(ssv));
638 (void)sv_bless(TOPs, stash);
648 const char * const elem = SvPV_const(sv, len);
649 GV * const gv = MUTABLE_GV(TOPs);
654 /* elem will always be NUL terminated. */
655 const char * const second_letter = elem + 1;
658 if (len == 5 && strEQ(second_letter, "RRAY"))
660 tmpRef = MUTABLE_SV(GvAV(gv));
661 if (tmpRef && !AvREAL((const AV *)tmpRef)
662 && AvREIFY((const AV *)tmpRef))
663 av_reify(MUTABLE_AV(tmpRef));
667 if (len == 4 && strEQ(second_letter, "ODE"))
668 tmpRef = MUTABLE_SV(GvCVu(gv));
671 if (len == 10 && strEQ(second_letter, "ILEHANDLE")) {
672 /* finally deprecated in 5.8.0 */
673 deprecate("*glob{FILEHANDLE}");
674 tmpRef = MUTABLE_SV(GvIOp(gv));
677 if (len == 6 && strEQ(second_letter, "ORMAT"))
678 tmpRef = MUTABLE_SV(GvFORM(gv));
681 if (len == 4 && strEQ(second_letter, "LOB"))
682 tmpRef = MUTABLE_SV(gv);
685 if (len == 4 && strEQ(second_letter, "ASH"))
686 tmpRef = MUTABLE_SV(GvHV(gv));
689 if (*second_letter == 'O' && !elem[2] && len == 2)
690 tmpRef = MUTABLE_SV(GvIOp(gv));
693 if (len == 4 && strEQ(second_letter, "AME"))
694 sv = newSVhek(GvNAME_HEK(gv));
697 if (len == 7 && strEQ(second_letter, "ACKAGE")) {
698 const HV * const stash = GvSTASH(gv);
699 const HEK * const hek = stash ? HvNAME_HEK(stash) : NULL;
700 sv = hek ? newSVhek(hek) : newSVpvs("__ANON__");
704 if (len == 6 && strEQ(second_letter, "CALAR"))
719 /* Pattern matching */
727 if (len == 0 || len > I32_MAX || !SvPOK(sv) || SvUTF8(sv) || SvVALID(sv)) {
728 /* Historically, study was skipped in these cases. */
733 /* Make study a no-op. It's no longer useful and its existence
734 complicates matters elsewhere. */
740 /* also used for: pp_transr() */
747 if (PL_op->op_flags & OPf_STACKED)
752 sv = PAD_SV(ARGTARG);
757 if(PL_op->op_type == OP_TRANSR) {
759 const char * const pv = SvPV(sv,len);
760 SV * const newsv = newSVpvn_flags(pv, len, SVs_TEMP|SvUTF8(sv));
765 mPUSHi(do_trans(sv));
770 /* Lvalue operators. */
773 S_do_chomp(pTHX_ SV *retval, SV *sv, bool chomping)
779 PERL_ARGS_ASSERT_DO_CHOMP;
781 if (chomping && (RsSNARF(PL_rs) || RsRECORD(PL_rs)))
783 if (SvTYPE(sv) == SVt_PVAV) {
785 AV *const av = MUTABLE_AV(sv);
786 const I32 max = AvFILL(av);
788 for (i = 0; i <= max; i++) {
789 sv = MUTABLE_SV(av_fetch(av, i, FALSE));
790 if (sv && ((sv = *(SV**)sv), sv != &PL_sv_undef))
791 count += do_chomp(retval, sv, chomping);
795 else if (SvTYPE(sv) == SVt_PVHV) {
796 HV* const hv = MUTABLE_HV(sv);
798 (void)hv_iterinit(hv);
799 while ((entry = hv_iternext(hv)))
800 count += do_chomp(retval, hv_iterval(hv,entry), chomping);
803 else if (SvREADONLY(sv)) {
804 Perl_croak_no_modify();
809 /* XXX, here sv is utf8-ized as a side-effect!
810 If encoding.pm is used properly, almost string-generating
811 operations, including literal strings, chr(), input data, etc.
812 should have been utf8-ized already, right?
814 sv_recode_to_utf8(sv, _get_encoding());
820 char *temp_buffer = NULL;
829 while (len && s[-1] == '\n') {
836 STRLEN rslen, rs_charlen;
837 const char *rsptr = SvPV_const(PL_rs, rslen);
839 rs_charlen = SvUTF8(PL_rs)
843 if (SvUTF8(PL_rs) != SvUTF8(sv)) {
844 /* Assumption is that rs is shorter than the scalar. */
846 /* RS is utf8, scalar is 8 bit. */
848 temp_buffer = (char*)bytes_from_utf8((U8*)rsptr,
851 /* Cannot downgrade, therefore cannot possibly match
853 assert (temp_buffer == rsptr);
859 else if (IN_ENCODING) {
860 /* RS is 8 bit, encoding.pm is used.
861 * Do not recode PL_rs as a side-effect. */
862 svrecode = newSVpvn(rsptr, rslen);
863 sv_recode_to_utf8(svrecode, _get_encoding());
864 rsptr = SvPV_const(svrecode, rslen);
865 rs_charlen = sv_len_utf8(svrecode);
868 /* RS is 8 bit, scalar is utf8. */
869 temp_buffer = (char*)bytes_to_utf8((U8*)rsptr, &rslen);
883 if (memNE(s, rsptr, rslen))
888 SvPV_force_nomg_nolen(sv);
896 SvREFCNT_dec(svrecode);
898 Safefree(temp_buffer);
900 if (len && (!SvPOK(sv) || SvIsCOW(sv)))
901 s = SvPV_force_nomg(sv, len);
904 char * const send = s + len;
905 char * const start = s;
907 while (s > start && UTF8_IS_CONTINUATION(*s))
909 if (is_utf8_string((U8*)s, send - s)) {
910 sv_setpvn(retval, s, send - s);
912 SvCUR_set(sv, s - start);
918 sv_setpvs(retval, "");
922 sv_setpvn(retval, s, 1);
929 sv_setpvs(retval, "");
936 /* also used for: pp_schomp() */
941 const bool chomping = PL_op->op_type == OP_SCHOMP;
943 const size_t count = do_chomp(TARG, TOPs, chomping);
945 sv_setiv(TARG, count);
951 /* also used for: pp_chomp() */
955 dSP; dMARK; dTARGET; dORIGMARK;
956 const bool chomping = PL_op->op_type == OP_CHOMP;
960 count += do_chomp(TARG, *++MARK, chomping);
962 sv_setiv(TARG, count);
973 if (!PL_op->op_private) {
985 if (SvTHINKFIRST(sv))
986 sv_force_normal_flags(sv, SV_COW_DROP_PV|SV_IMMEDIATE_UNREF);
988 switch (SvTYPE(sv)) {
992 av_undef(MUTABLE_AV(sv));
995 hv_undef(MUTABLE_HV(sv));
998 if (cv_const_sv((const CV *)sv))
999 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
1000 "Constant subroutine %"SVf" undefined",
1001 SVfARG(CvANON((const CV *)sv)
1002 ? newSVpvs_flags("(anonymous)", SVs_TEMP)
1003 : sv_2mortal(newSVhek(
1005 ? CvNAME_HEK((CV *)sv)
1006 : GvENAME_HEK(CvGV((const CV *)sv))
1011 /* let user-undef'd sub keep its identity */
1012 cv_undef_flags(MUTABLE_CV(sv), CV_UNDEF_KEEP_NAME);
1015 assert(isGV_with_GP(sv));
1016 assert(!SvFAKE(sv));
1021 /* undef *Pkg::meth_name ... */
1023 = GvCVu((const GV *)sv) && (stash = GvSTASH((const GV *)sv))
1024 && HvENAME_get(stash);
1026 if((stash = GvHV((const GV *)sv))) {
1027 if(HvENAME_get(stash))
1028 SvREFCNT_inc_simple_void_NN(sv_2mortal((SV *)stash));
1032 SvREFCNT_inc_simple_void_NN(sv_2mortal(sv));
1033 gp_free(MUTABLE_GV(sv));
1035 GvGP_set(sv, gp_ref(gp));
1036 #ifndef PERL_DONT_CREATE_GVSV
1037 GvSV(sv) = newSV(0);
1039 GvLINE(sv) = CopLINE(PL_curcop);
1040 GvEGV(sv) = MUTABLE_GV(sv);
1044 mro_package_moved(NULL, stash, (const GV *)sv, 0);
1046 /* undef *Foo::ISA */
1047 if( strEQ(GvNAME((const GV *)sv), "ISA")
1048 && (stash = GvSTASH((const GV *)sv))
1049 && (method_changed || HvENAME(stash)) )
1050 mro_isa_changed_in(stash);
1051 else if(method_changed)
1052 mro_method_changed_in(
1053 GvSTASH((const GV *)sv)
1059 if (SvTYPE(sv) >= SVt_PV && SvPVX_const(sv) && SvLEN(sv)) {
1073 /* also used for: pp_i_postdec() pp_i_postinc() pp_postdec() */
1079 PL_op->op_type == OP_POSTINC || PL_op->op_type == OP_I_POSTINC;
1080 if (SvTYPE(TOPs) >= SVt_PVAV || (isGV_with_GP(TOPs) && !SvFAKE(TOPs)))
1081 Perl_croak_no_modify();
1083 TARG = sv_newmortal();
1084 sv_setsv(TARG, TOPs);
1085 if (!SvREADONLY(TOPs) && !SvGMAGICAL(TOPs) && SvIOK_notUV(TOPs) && !SvNOK(TOPs) && !SvPOK(TOPs)
1086 && SvIVX(TOPs) != (inc ? IV_MAX : IV_MIN))
1088 SvIV_set(TOPs, SvIVX(TOPs) + (inc ? 1 : -1));
1089 SvFLAGS(TOPs) &= ~(SVp_NOK|SVp_POK);
1093 else sv_dec_nomg(TOPs);
1095 /* special case for undef: see thread at 2003-03/msg00536.html in archive */
1096 if (inc && !SvOK(TARG))
1102 /* Ordinary operators. */
1106 dSP; dATARGET; SV *svl, *svr;
1107 #ifdef PERL_PRESERVE_IVUV
1110 tryAMAGICbin_MG(pow_amg, AMGf_assign|AMGf_numeric);
1113 #ifdef PERL_PRESERVE_IVUV
1114 /* For integer to integer power, we do the calculation by hand wherever
1115 we're sure it is safe; otherwise we call pow() and try to convert to
1116 integer afterwards. */
1117 if (SvIV_please_nomg(svr) && SvIV_please_nomg(svl)) {
1125 const IV iv = SvIVX(svr);
1129 goto float_it; /* Can't do negative powers this way. */
1133 baseuok = SvUOK(svl);
1135 baseuv = SvUVX(svl);
1137 const IV iv = SvIVX(svl);
1140 baseuok = TRUE; /* effectively it's a UV now */
1142 baseuv = -iv; /* abs, baseuok == false records sign */
1145 /* now we have integer ** positive integer. */
1148 /* foo & (foo - 1) is zero only for a power of 2. */
1149 if (!(baseuv & (baseuv - 1))) {
1150 /* We are raising power-of-2 to a positive integer.
1151 The logic here will work for any base (even non-integer
1152 bases) but it can be less accurate than
1153 pow (base,power) or exp (power * log (base)) when the
1154 intermediate values start to spill out of the mantissa.
1155 With powers of 2 we know this can't happen.
1156 And powers of 2 are the favourite thing for perl
1157 programmers to notice ** not doing what they mean. */
1159 NV base = baseuok ? baseuv : -(NV)baseuv;
1164 while (power >>= 1) {
1172 SvIV_please_nomg(svr);
1175 unsigned int highbit = 8 * sizeof(UV);
1176 unsigned int diff = 8 * sizeof(UV);
1177 while (diff >>= 1) {
1179 if (baseuv >> highbit) {
1183 /* we now have baseuv < 2 ** highbit */
1184 if (power * highbit <= 8 * sizeof(UV)) {
1185 /* result will definitely fit in UV, so use UV math
1186 on same algorithm as above */
1189 const bool odd_power = cBOOL(power & 1);
1193 while (power >>= 1) {
1200 if (baseuok || !odd_power)
1201 /* answer is positive */
1203 else if (result <= (UV)IV_MAX)
1204 /* answer negative, fits in IV */
1205 SETi( -(IV)result );
1206 else if (result == (UV)IV_MIN)
1207 /* 2's complement assumption: special case IV_MIN */
1210 /* answer negative, doesn't fit */
1211 SETn( -(NV)result );
1219 NV right = SvNV_nomg(svr);
1220 NV left = SvNV_nomg(svl);
1223 #if defined(USE_LONG_DOUBLE) && defined(HAS_AIX_POWL_NEG_BASE_BUG)
1225 We are building perl with long double support and are on an AIX OS
1226 afflicted with a powl() function that wrongly returns NaNQ for any
1227 negative base. This was reported to IBM as PMR #23047-379 on
1228 03/06/2006. The problem exists in at least the following versions
1229 of AIX and the libm fileset, and no doubt others as well:
1231 AIX 4.3.3-ML10 bos.adt.libm 4.3.3.50
1232 AIX 5.1.0-ML04 bos.adt.libm 5.1.0.29
1233 AIX 5.2.0 bos.adt.libm 5.2.0.85
1235 So, until IBM fixes powl(), we provide the following workaround to
1236 handle the problem ourselves. Our logic is as follows: for
1237 negative bases (left), we use fmod(right, 2) to check if the
1238 exponent is an odd or even integer:
1240 - if odd, powl(left, right) == -powl(-left, right)
1241 - if even, powl(left, right) == powl(-left, right)
1243 If the exponent is not an integer, the result is rightly NaNQ, so
1244 we just return that (as NV_NAN).
1248 NV mod2 = Perl_fmod( right, 2.0 );
1249 if (mod2 == 1.0 || mod2 == -1.0) { /* odd integer */
1250 SETn( -Perl_pow( -left, right) );
1251 } else if (mod2 == 0.0) { /* even integer */
1252 SETn( Perl_pow( -left, right) );
1253 } else { /* fractional power */
1257 SETn( Perl_pow( left, right) );
1260 SETn( Perl_pow( left, right) );
1261 #endif /* HAS_AIX_POWL_NEG_BASE_BUG */
1263 #ifdef PERL_PRESERVE_IVUV
1265 SvIV_please_nomg(svr);
1273 dSP; dATARGET; SV *svl, *svr;
1274 tryAMAGICbin_MG(mult_amg, AMGf_assign|AMGf_numeric);
1277 #ifdef PERL_PRESERVE_IVUV
1278 if (SvIV_please_nomg(svr)) {
1279 /* Unless the left argument is integer in range we are going to have to
1280 use NV maths. Hence only attempt to coerce the right argument if
1281 we know the left is integer. */
1282 /* Left operand is defined, so is it IV? */
1283 if (SvIV_please_nomg(svl)) {
1284 bool auvok = SvUOK(svl);
1285 bool buvok = SvUOK(svr);
1286 const UV topmask = (~ (UV)0) << (4 * sizeof (UV));
1287 const UV botmask = ~((~ (UV)0) << (4 * sizeof (UV)));
1296 const IV aiv = SvIVX(svl);
1299 auvok = TRUE; /* effectively it's a UV now */
1301 /* abs, auvok == false records sign */
1302 alow = (aiv == IV_MIN) ? (UV)aiv : (UV)(-aiv);
1308 const IV biv = SvIVX(svr);
1311 buvok = TRUE; /* effectively it's a UV now */
1313 /* abs, buvok == false records sign */
1314 blow = (biv == IV_MIN) ? (UV)biv : (UV)(-biv);
1318 /* If this does sign extension on unsigned it's time for plan B */
1319 ahigh = alow >> (4 * sizeof (UV));
1321 bhigh = blow >> (4 * sizeof (UV));
1323 if (ahigh && bhigh) {
1325 /* eg 32 bit is at least 0x10000 * 0x10000 == 0x100000000
1326 which is overflow. Drop to NVs below. */
1327 } else if (!ahigh && !bhigh) {
1328 /* eg 32 bit is at most 0xFFFF * 0xFFFF == 0xFFFE0001
1329 so the unsigned multiply cannot overflow. */
1330 const UV product = alow * blow;
1331 if (auvok == buvok) {
1332 /* -ve * -ve or +ve * +ve gives a +ve result. */
1336 } else if (product <= (UV)IV_MIN) {
1337 /* 2s complement assumption that (UV)-IV_MIN is correct. */
1338 /* -ve result, which could overflow an IV */
1340 /* can't negate IV_MIN, but there are aren't two
1341 * integers such that !ahigh && !bhigh, where the
1342 * product equals 0x800....000 */
1343 assert(product != (UV)IV_MIN);
1344 SETi( -(IV)product );
1346 } /* else drop to NVs below. */
1348 /* One operand is large, 1 small */
1351 /* swap the operands */
1353 bhigh = blow; /* bhigh now the temp var for the swap */
1357 /* now, ((ahigh * blow) << half_UV_len) + (alow * blow)
1358 multiplies can't overflow. shift can, add can, -ve can. */
1359 product_middle = ahigh * blow;
1360 if (!(product_middle & topmask)) {
1361 /* OK, (ahigh * blow) won't lose bits when we shift it. */
1363 product_middle <<= (4 * sizeof (UV));
1364 product_low = alow * blow;
1366 /* as for pp_add, UV + something mustn't get smaller.
1367 IIRC ANSI mandates this wrapping *behaviour* for
1368 unsigned whatever the actual representation*/
1369 product_low += product_middle;
1370 if (product_low >= product_middle) {
1371 /* didn't overflow */
1372 if (auvok == buvok) {
1373 /* -ve * -ve or +ve * +ve gives a +ve result. */
1375 SETu( product_low );
1377 } else if (product_low <= (UV)IV_MIN) {
1378 /* 2s complement assumption again */
1379 /* -ve result, which could overflow an IV */
1381 SETi(product_low == (UV)IV_MIN
1382 ? IV_MIN : -(IV)product_low);
1384 } /* else drop to NVs below. */
1386 } /* product_middle too large */
1387 } /* ahigh && bhigh */
1392 NV right = SvNV_nomg(svr);
1393 NV left = SvNV_nomg(svl);
1395 SETn( left * right );
1402 dSP; dATARGET; SV *svl, *svr;
1403 tryAMAGICbin_MG(div_amg, AMGf_assign|AMGf_numeric);
1406 /* Only try to do UV divide first
1407 if ((SLOPPYDIVIDE is true) or
1408 (PERL_PRESERVE_IVUV is true and one or both SV is a UV too large
1410 The assumption is that it is better to use floating point divide
1411 whenever possible, only doing integer divide first if we can't be sure.
1412 If NV_PRESERVES_UV is true then we know at compile time that no UV
1413 can be too large to preserve, so don't need to compile the code to
1414 test the size of UVs. */
1417 # define PERL_TRY_UV_DIVIDE
1418 /* ensure that 20./5. == 4. */
1420 # ifdef PERL_PRESERVE_IVUV
1421 # ifndef NV_PRESERVES_UV
1422 # define PERL_TRY_UV_DIVIDE
1427 #ifdef PERL_TRY_UV_DIVIDE
1428 if (SvIV_please_nomg(svr) && SvIV_please_nomg(svl)) {
1429 bool left_non_neg = SvUOK(svl);
1430 bool right_non_neg = SvUOK(svr);
1434 if (right_non_neg) {
1438 const IV biv = SvIVX(svr);
1441 right_non_neg = TRUE; /* effectively it's a UV now */
1444 right = (biv == IV_MIN) ? (UV)biv : (UV)(-biv);
1447 /* historically undef()/0 gives a "Use of uninitialized value"
1448 warning before dieing, hence this test goes here.
1449 If it were immediately before the second SvIV_please, then
1450 DIE() would be invoked before left was even inspected, so
1451 no inspection would give no warning. */
1453 DIE(aTHX_ "Illegal division by zero");
1459 const IV aiv = SvIVX(svl);
1462 left_non_neg = TRUE; /* effectively it's a UV now */
1465 left = (aiv == IV_MIN) ? (UV)aiv : (UV)(-aiv);
1471 /* For sloppy divide we always attempt integer division. */
1473 /* Otherwise we only attempt it if either or both operands
1474 would not be preserved by an NV. If both fit in NVs
1475 we fall through to the NV divide code below. However,
1476 as left >= right to ensure integer result here, we know that
1477 we can skip the test on the right operand - right big
1478 enough not to be preserved can't get here unless left is
1481 && (left > ((UV)1 << NV_PRESERVES_UV_BITS))
1484 /* Integer division can't overflow, but it can be imprecise. */
1485 const UV result = left / right;
1486 if (result * right == left) {
1487 SP--; /* result is valid */
1488 if (left_non_neg == right_non_neg) {
1489 /* signs identical, result is positive. */
1493 /* 2s complement assumption */
1494 if (result <= (UV)IV_MIN)
1495 SETi(result == (UV)IV_MIN ? IV_MIN : -(IV)result);
1497 /* It's exact but too negative for IV. */
1498 SETn( -(NV)result );
1501 } /* tried integer divide but it was not an integer result */
1502 } /* else (PERL_ABS(result) < 1.0) or (both UVs in range for NV) */
1503 } /* one operand wasn't SvIOK */
1504 #endif /* PERL_TRY_UV_DIVIDE */
1506 NV right = SvNV_nomg(svr);
1507 NV left = SvNV_nomg(svl);
1508 (void)POPs;(void)POPs;
1509 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
1510 if (! Perl_isnan(right) && right == 0.0)
1514 DIE(aTHX_ "Illegal division by zero");
1515 PUSHn( left / right );
1523 tryAMAGICbin_MG(modulo_amg, AMGf_assign|AMGf_numeric);
1527 bool left_neg = FALSE;
1528 bool right_neg = FALSE;
1529 bool use_double = FALSE;
1530 bool dright_valid = FALSE;
1533 SV * const svr = TOPs;
1534 SV * const svl = TOPm1s;
1535 if (SvIV_please_nomg(svr)) {
1536 right_neg = !SvUOK(svr);
1540 const IV biv = SvIVX(svr);
1543 right_neg = FALSE; /* effectively it's a UV now */
1545 right = (biv == IV_MIN) ? (UV)biv : (UV)(-biv);
1550 dright = SvNV_nomg(svr);
1551 right_neg = dright < 0;
1554 if (dright < UV_MAX_P1) {
1555 right = U_V(dright);
1556 dright_valid = TRUE; /* In case we need to use double below. */
1562 /* At this point use_double is only true if right is out of range for
1563 a UV. In range NV has been rounded down to nearest UV and
1564 use_double false. */
1565 if (!use_double && SvIV_please_nomg(svl)) {
1566 left_neg = !SvUOK(svl);
1570 const IV aiv = SvIVX(svl);
1573 left_neg = FALSE; /* effectively it's a UV now */
1575 left = (aiv == IV_MIN) ? (UV)aiv : (UV)(-aiv);
1580 dleft = SvNV_nomg(svl);
1581 left_neg = dleft < 0;
1585 /* This should be exactly the 5.6 behaviour - if left and right are
1586 both in range for UV then use U_V() rather than floor. */
1588 if (dleft < UV_MAX_P1) {
1589 /* right was in range, so is dleft, so use UVs not double.
1593 /* left is out of range for UV, right was in range, so promote
1594 right (back) to double. */
1596 /* The +0.5 is used in 5.6 even though it is not strictly
1597 consistent with the implicit +0 floor in the U_V()
1598 inside the #if 1. */
1599 dleft = Perl_floor(dleft + 0.5);
1602 dright = Perl_floor(dright + 0.5);
1613 DIE(aTHX_ "Illegal modulus zero");
1615 dans = Perl_fmod(dleft, dright);
1616 if ((left_neg != right_neg) && dans)
1617 dans = dright - dans;
1620 sv_setnv(TARG, dans);
1626 DIE(aTHX_ "Illegal modulus zero");
1629 if ((left_neg != right_neg) && ans)
1632 /* XXX may warn: unary minus operator applied to unsigned type */
1633 /* could change -foo to be (~foo)+1 instead */
1634 if (ans <= ~((UV)IV_MAX)+1)
1635 sv_setiv(TARG, ~ans+1);
1637 sv_setnv(TARG, -(NV)ans);
1640 sv_setuv(TARG, ans);
1652 bool infnan = FALSE;
1654 if (GIMME_V == G_ARRAY && PL_op->op_private & OPpREPEAT_DOLIST) {
1655 /* TODO: think of some way of doing list-repeat overloading ??? */
1660 if (UNLIKELY(PL_op->op_private & OPpREPEAT_DOLIST)) {
1661 /* The parser saw this as a list repeat, and there
1662 are probably several items on the stack. But we're
1663 in scalar/void context, and there's no pp_list to save us
1664 now. So drop the rest of the items -- robin@kitsite.com
1667 if (MARK + 1 < SP) {
1673 ASSUME(MARK + 1 == SP);
1675 MARK[1] = &PL_sv_undef;
1679 tryAMAGICbin_MG(repeat_amg, AMGf_assign);
1685 const UV uv = SvUV_nomg(sv);
1687 count = IV_MAX; /* The best we can do? */
1691 count = SvIV_nomg(sv);
1694 else if (SvNOKp(sv)) {
1695 const NV nv = SvNV_nomg(sv);
1696 infnan = Perl_isinfnan(nv);
1697 if (UNLIKELY(infnan)) {
1701 count = -1; /* An arbitrary negative integer */
1707 count = SvIV_nomg(sv);
1710 Perl_ck_warner(aTHX_ packWARN(WARN_NUMERIC),
1711 "Non-finite repeat count does nothing");
1712 } else if (count < 0) {
1714 Perl_ck_warner(aTHX_ packWARN(WARN_NUMERIC),
1715 "Negative repeat count does nothing");
1718 if (GIMME_V == G_ARRAY && PL_op->op_private & OPpREPEAT_DOLIST) {
1720 static const char* const oom_list_extend = "Out of memory during list extend";
1721 const I32 items = SP - MARK;
1722 const I32 max = items * count;
1723 const U8 mod = PL_op->op_flags & OPf_MOD;
1725 MEM_WRAP_CHECK_1(max, SV*, oom_list_extend);
1726 /* Did the max computation overflow? */
1727 if (items > 0 && max > 0 && (max < items || max < count))
1728 Perl_croak(aTHX_ "%s", oom_list_extend);
1733 if (mod && SvPADTMP(*SP)) {
1734 *SP = sv_mortalcopy(*SP);
1741 repeatcpy((char*)(MARK + items), (char*)MARK,
1742 items * sizeof(const SV *), count - 1);
1745 else if (count <= 0)
1748 else { /* Note: mark already snarfed by pp_list */
1749 SV * const tmpstr = POPs;
1752 static const char* const oom_string_extend =
1753 "Out of memory during string extend";
1756 sv_setsv_nomg(TARG, tmpstr);
1757 SvPV_force_nomg(TARG, len);
1758 isutf = DO_UTF8(TARG);
1763 const STRLEN max = (UV)count * len;
1764 if (len > MEM_SIZE_MAX / count)
1765 Perl_croak(aTHX_ "%s", oom_string_extend);
1766 MEM_WRAP_CHECK_1(max, char, oom_string_extend);
1767 SvGROW(TARG, max + 1);
1768 repeatcpy(SvPVX(TARG) + len, SvPVX(TARG), len, count - 1);
1769 SvCUR_set(TARG, SvCUR(TARG) * count);
1771 *SvEND(TARG) = '\0';
1774 (void)SvPOK_only_UTF8(TARG);
1776 (void)SvPOK_only(TARG);
1785 dSP; dATARGET; bool useleft; SV *svl, *svr;
1786 tryAMAGICbin_MG(subtr_amg, AMGf_assign|AMGf_numeric);
1789 useleft = USE_LEFT(svl);
1790 #ifdef PERL_PRESERVE_IVUV
1791 /* See comments in pp_add (in pp_hot.c) about Overflow, and how
1792 "bad things" happen if you rely on signed integers wrapping. */
1793 if (SvIV_please_nomg(svr)) {
1794 /* Unless the left argument is integer in range we are going to have to
1795 use NV maths. Hence only attempt to coerce the right argument if
1796 we know the left is integer. */
1803 a_valid = auvok = 1;
1804 /* left operand is undef, treat as zero. */
1806 /* Left operand is defined, so is it IV? */
1807 if (SvIV_please_nomg(svl)) {
1808 if ((auvok = SvUOK(svl)))
1811 const IV aiv = SvIVX(svl);
1814 auvok = 1; /* Now acting as a sign flag. */
1815 } else { /* 2s complement assumption for IV_MIN */
1816 auv = (aiv == IV_MIN) ? (UV)aiv : (UV)-aiv;
1823 bool result_good = 0;
1826 bool buvok = SvUOK(svr);
1831 const IV biv = SvIVX(svr);
1836 buv = (biv == IV_MIN) ? (UV)biv : (UV)-biv;
1838 /* ?uvok if value is >= 0. basically, flagged as UV if it's +ve,
1839 else "IV" now, independent of how it came in.
1840 if a, b represents positive, A, B negative, a maps to -A etc
1845 all UV maths. negate result if A negative.
1846 subtract if signs same, add if signs differ. */
1848 if (auvok ^ buvok) {
1857 /* Must get smaller */
1862 if (result <= buv) {
1863 /* result really should be -(auv-buv). as its negation
1864 of true value, need to swap our result flag */
1876 if (result <= (UV)IV_MIN)
1877 SETi(result == (UV)IV_MIN
1878 ? IV_MIN : -(IV)result);
1880 /* result valid, but out of range for IV. */
1881 SETn( -(NV)result );
1885 } /* Overflow, drop through to NVs. */
1890 NV value = SvNV_nomg(svr);
1894 /* left operand is undef, treat as zero - value */
1898 SETn( SvNV_nomg(svl) - value );
1905 dSP; dATARGET; SV *svl, *svr;
1906 tryAMAGICbin_MG(lshift_amg, AMGf_assign|AMGf_numeric);
1910 const IV shift = SvIV_nomg(svr);
1911 if (PL_op->op_private & HINT_INTEGER) {
1912 const IV i = SvIV_nomg(svl);
1916 const UV u = SvUV_nomg(svl);
1925 dSP; dATARGET; SV *svl, *svr;
1926 tryAMAGICbin_MG(rshift_amg, AMGf_assign|AMGf_numeric);
1930 const IV shift = SvIV_nomg(svr);
1931 if (PL_op->op_private & HINT_INTEGER) {
1932 const IV i = SvIV_nomg(svl);
1936 const UV u = SvUV_nomg(svl);
1948 tryAMAGICbin_MG(lt_amg, AMGf_set|AMGf_numeric);
1952 (SvIOK_notUV(left) && SvIOK_notUV(right))
1953 ? (SvIVX(left) < SvIVX(right))
1954 : (do_ncmp(left, right) == -1)
1964 tryAMAGICbin_MG(gt_amg, AMGf_set|AMGf_numeric);
1968 (SvIOK_notUV(left) && SvIOK_notUV(right))
1969 ? (SvIVX(left) > SvIVX(right))
1970 : (do_ncmp(left, right) == 1)
1980 tryAMAGICbin_MG(le_amg, AMGf_set|AMGf_numeric);
1984 (SvIOK_notUV(left) && SvIOK_notUV(right))
1985 ? (SvIVX(left) <= SvIVX(right))
1986 : (do_ncmp(left, right) <= 0)
1996 tryAMAGICbin_MG(ge_amg, AMGf_set|AMGf_numeric);
2000 (SvIOK_notUV(left) && SvIOK_notUV(right))
2001 ? (SvIVX(left) >= SvIVX(right))
2002 : ( (do_ncmp(left, right) & 2) == 0)
2012 tryAMAGICbin_MG(ne_amg, AMGf_set|AMGf_numeric);
2016 (SvIOK_notUV(left) && SvIOK_notUV(right))
2017 ? (SvIVX(left) != SvIVX(right))
2018 : (do_ncmp(left, right) != 0)
2023 /* compare left and right SVs. Returns:
2027 * 2: left or right was a NaN
2030 Perl_do_ncmp(pTHX_ SV* const left, SV * const right)
2032 PERL_ARGS_ASSERT_DO_NCMP;
2033 #ifdef PERL_PRESERVE_IVUV
2034 /* Fortunately it seems NaN isn't IOK */
2035 if (SvIV_please_nomg(right) && SvIV_please_nomg(left)) {
2037 const IV leftiv = SvIVX(left);
2038 if (!SvUOK(right)) {
2039 /* ## IV <=> IV ## */
2040 const IV rightiv = SvIVX(right);
2041 return (leftiv > rightiv) - (leftiv < rightiv);
2043 /* ## IV <=> UV ## */
2045 /* As (b) is a UV, it's >=0, so it must be < */
2048 const UV rightuv = SvUVX(right);
2049 return ((UV)leftiv > rightuv) - ((UV)leftiv < rightuv);
2054 /* ## UV <=> UV ## */
2055 const UV leftuv = SvUVX(left);
2056 const UV rightuv = SvUVX(right);
2057 return (leftuv > rightuv) - (leftuv < rightuv);
2059 /* ## UV <=> IV ## */
2061 const IV rightiv = SvIVX(right);
2063 /* As (a) is a UV, it's >=0, so it cannot be < */
2066 const UV leftuv = SvUVX(left);
2067 return (leftuv > (UV)rightiv) - (leftuv < (UV)rightiv);
2070 NOT_REACHED; /* NOTREACHED */
2074 NV const rnv = SvNV_nomg(right);
2075 NV const lnv = SvNV_nomg(left);
2077 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2078 if (Perl_isnan(lnv) || Perl_isnan(rnv)) {
2081 return (lnv > rnv) - (lnv < rnv);
2100 tryAMAGICbin_MG(ncmp_amg, AMGf_numeric);
2103 value = do_ncmp(left, right);
2115 /* also used for: pp_sge() pp_sgt() pp_slt() */
2121 int amg_type = sle_amg;
2125 switch (PL_op->op_type) {
2144 tryAMAGICbin_MG(amg_type, AMGf_set);
2148 #ifdef USE_LOCALE_COLLATE
2149 (IN_LC_RUNTIME(LC_COLLATE))
2150 ? sv_cmp_locale_flags(left, right, 0)
2153 sv_cmp_flags(left, right, 0);
2154 SETs(boolSV(cmp * multiplier < rhs));
2162 tryAMAGICbin_MG(seq_amg, AMGf_set);
2165 SETs(boolSV(sv_eq_flags(left, right, 0)));
2173 tryAMAGICbin_MG(sne_amg, AMGf_set);
2176 SETs(boolSV(!sv_eq_flags(left, right, 0)));
2184 tryAMAGICbin_MG(scmp_amg, 0);
2188 #ifdef USE_LOCALE_COLLATE
2189 (IN_LC_RUNTIME(LC_COLLATE))
2190 ? sv_cmp_locale_flags(left, right, 0)
2193 sv_cmp_flags(left, right, 0);
2202 tryAMAGICbin_MG(band_amg, AMGf_assign);
2205 if (SvNIOKp(left) || SvNIOKp(right)) {
2206 const bool left_ro_nonnum = !SvNIOKp(left) && SvREADONLY(left);
2207 const bool right_ro_nonnum = !SvNIOKp(right) && SvREADONLY(right);
2208 if (PL_op->op_private & HINT_INTEGER) {
2209 const IV i = SvIV_nomg(left) & SvIV_nomg(right);
2213 const UV u = SvUV_nomg(left) & SvUV_nomg(right);
2216 if (left_ro_nonnum && left != TARG) SvNIOK_off(left);
2217 if (right_ro_nonnum) SvNIOK_off(right);
2220 do_vop(PL_op->op_type, TARG, left, right);
2228 /* also used for: pp_bit_xor() */
2233 const int op_type = PL_op->op_type;
2235 tryAMAGICbin_MG((op_type == OP_BIT_OR ? bor_amg : bxor_amg), AMGf_assign);
2238 if (SvNIOKp(left) || SvNIOKp(right)) {
2239 const bool left_ro_nonnum = !SvNIOKp(left) && SvREADONLY(left);
2240 const bool right_ro_nonnum = !SvNIOKp(right) && SvREADONLY(right);
2241 if (PL_op->op_private & HINT_INTEGER) {
2242 const IV l = (USE_LEFT(left) ? SvIV_nomg(left) : 0);
2243 const IV r = SvIV_nomg(right);
2244 const IV result = op_type == OP_BIT_OR ? (l | r) : (l ^ r);
2248 const UV l = (USE_LEFT(left) ? SvUV_nomg(left) : 0);
2249 const UV r = SvUV_nomg(right);
2250 const UV result = op_type == OP_BIT_OR ? (l | r) : (l ^ r);
2253 if (left_ro_nonnum && left != TARG) SvNIOK_off(left);
2254 if (right_ro_nonnum) SvNIOK_off(right);
2257 do_vop(op_type, TARG, left, right);
2264 PERL_STATIC_INLINE bool
2265 S_negate_string(pTHX)
2270 SV * const sv = TOPs;
2271 if (!SvPOKp(sv) || SvNIOK(sv) || (!SvPOK(sv) && SvNIOKp(sv)))
2273 s = SvPV_nomg_const(sv, len);
2274 if (isIDFIRST(*s)) {
2275 sv_setpvs(TARG, "-");
2278 else if (*s == '+' || (*s == '-' && !looks_like_number(sv))) {
2279 sv_setsv_nomg(TARG, sv);
2280 *SvPV_force_nomg(TARG, len) = *s == '-' ? '+' : '-';
2290 tryAMAGICun_MG(neg_amg, AMGf_numeric);
2291 if (S_negate_string(aTHX)) return NORMAL;
2293 SV * const sv = TOPs;
2296 /* It's publicly an integer */
2299 if (SvIVX(sv) == IV_MIN) {
2300 /* 2s complement assumption. */
2301 SETi(SvIVX(sv)); /* special case: -((UV)IV_MAX+1) ==
2305 else if (SvUVX(sv) <= IV_MAX) {
2310 else if (SvIVX(sv) != IV_MIN) {
2314 #ifdef PERL_PRESERVE_IVUV
2321 if (SvNIOKp(sv) && (SvNIOK(sv) || !SvPOK(sv)))
2322 SETn(-SvNV_nomg(sv));
2323 else if (SvPOKp(sv) && SvIV_please_nomg(sv))
2324 goto oops_its_an_int;
2326 SETn(-SvNV_nomg(sv));
2334 tryAMAGICun_MG(not_amg, AMGf_set);
2335 *PL_stack_sp = boolSV(!SvTRUE_nomg(*PL_stack_sp));
2342 tryAMAGICun_MG(compl_amg, AMGf_numeric);
2346 if (PL_op->op_private & HINT_INTEGER) {
2347 const IV i = ~SvIV_nomg(sv);
2351 const UV u = ~SvUV_nomg(sv);
2360 sv_copypv_nomg(TARG, sv);
2361 tmps = (U8*)SvPV_nomg(TARG, len);
2364 /* Calculate exact length, let's not estimate. */
2369 U8 * const send = tmps + len;
2370 U8 * const origtmps = tmps;
2371 const UV utf8flags = UTF8_ALLOW_ANYUV;
2373 while (tmps < send) {
2374 const UV c = utf8n_to_uvchr(tmps, send-tmps, &l, utf8flags);
2376 targlen += UNISKIP(~c);
2382 /* Now rewind strings and write them. */
2389 Newx(result, targlen + 1, U8);
2391 while (tmps < send) {
2392 const UV c = utf8n_to_uvchr(tmps, send-tmps, &l, utf8flags);
2394 p = uvchr_to_utf8_flags(p, ~c, UNICODE_ALLOW_ANY);
2397 sv_usepvn_flags(TARG, (char*)result, targlen,
2398 SV_HAS_TRAILING_NUL);
2405 Newx(result, nchar + 1, U8);
2407 while (tmps < send) {
2408 const U8 c = (U8)utf8n_to_uvchr(tmps, send-tmps, &l, utf8flags);
2413 sv_usepvn_flags(TARG, (char*)result, nchar, SV_HAS_TRAILING_NUL);
2422 for ( ; anum && (unsigned long)tmps % sizeof(long); anum--, tmps++)
2425 for ( ; anum >= (I32)sizeof(long); anum -= (I32)sizeof(long), tmpl++)
2430 for ( ; anum > 0; anum--, tmps++)
2438 /* integer versions of some of the above */
2443 tryAMAGICbin_MG(mult_amg, AMGf_assign);
2446 SETi( left * right );
2455 tryAMAGICbin_MG(div_amg, AMGf_assign);
2458 IV value = SvIV_nomg(right);
2460 DIE(aTHX_ "Illegal division by zero");
2461 num = SvIV_nomg(left);
2463 /* avoid FPE_INTOVF on some platforms when num is IV_MIN */
2467 value = num / value;
2473 #if defined(__GLIBC__) && IVSIZE == 8 && !defined(PERL_DEBUG_READONLY_OPS) \
2474 && ( __GLIBC__ < 2 || (__GLIBC__ == 2 && __GLIBC_MINOR__ < 8))
2481 /* This is the vanilla old i_modulo. */
2483 tryAMAGICbin_MG(modulo_amg, AMGf_assign);
2487 DIE(aTHX_ "Illegal modulus zero");
2488 /* avoid FPE_INTOVF on some platforms when left is IV_MIN */
2492 SETi( left % right );
2497 #if defined(__GLIBC__) && IVSIZE == 8 && !defined(PERL_DEBUG_READONLY_OPS) \
2498 && ( __GLIBC__ < 2 || (__GLIBC__ == 2 && __GLIBC_MINOR__ < 8))
2503 /* This is the i_modulo with the workaround for the _moddi3 bug
2504 * in (at least) glibc 2.2.5 (the PERL_ABS() the workaround).
2505 * See below for pp_i_modulo. */
2507 tryAMAGICbin_MG(modulo_amg, AMGf_assign);
2511 DIE(aTHX_ "Illegal modulus zero");
2512 /* avoid FPE_INTOVF on some platforms when left is IV_MIN */
2516 SETi( left % PERL_ABS(right) );
2523 dVAR; dSP; dATARGET;
2524 tryAMAGICbin_MG(modulo_amg, AMGf_assign);
2528 DIE(aTHX_ "Illegal modulus zero");
2529 /* The assumption is to use hereafter the old vanilla version... */
2531 PL_ppaddr[OP_I_MODULO] =
2533 /* .. but if we have glibc, we might have a buggy _moddi3
2534 * (at least glibc 2.2.5 is known to have this bug), in other
2535 * words our integer modulus with negative quad as the second
2536 * argument might be broken. Test for this and re-patch the
2537 * opcode dispatch table if that is the case, remembering to
2538 * also apply the workaround so that this first round works
2539 * right, too. See [perl #9402] for more information. */
2543 /* Cannot do this check with inlined IV constants since
2544 * that seems to work correctly even with the buggy glibc. */
2546 /* Yikes, we have the bug.
2547 * Patch in the workaround version. */
2549 PL_ppaddr[OP_I_MODULO] =
2550 &Perl_pp_i_modulo_1;
2551 /* Make certain we work right this time, too. */
2552 right = PERL_ABS(right);
2555 /* avoid FPE_INTOVF on some platforms when left is IV_MIN */
2559 SETi( left % right );
2568 tryAMAGICbin_MG(add_amg, AMGf_assign);
2570 dPOPTOPiirl_ul_nomg;
2571 SETi( left + right );
2579 tryAMAGICbin_MG(subtr_amg, AMGf_assign);
2581 dPOPTOPiirl_ul_nomg;
2582 SETi( left - right );
2590 tryAMAGICbin_MG(lt_amg, AMGf_set);
2593 SETs(boolSV(left < right));
2601 tryAMAGICbin_MG(gt_amg, AMGf_set);
2604 SETs(boolSV(left > right));
2612 tryAMAGICbin_MG(le_amg, AMGf_set);
2615 SETs(boolSV(left <= right));
2623 tryAMAGICbin_MG(ge_amg, AMGf_set);
2626 SETs(boolSV(left >= right));
2634 tryAMAGICbin_MG(eq_amg, AMGf_set);
2637 SETs(boolSV(left == right));
2645 tryAMAGICbin_MG(ne_amg, AMGf_set);
2648 SETs(boolSV(left != right));
2656 tryAMAGICbin_MG(ncmp_amg, 0);
2663 else if (left < right)
2675 tryAMAGICun_MG(neg_amg, 0);
2676 if (S_negate_string(aTHX)) return NORMAL;
2678 SV * const sv = TOPs;
2679 IV const i = SvIV_nomg(sv);
2685 /* High falutin' math. */
2690 tryAMAGICbin_MG(atan2_amg, 0);
2693 SETn(Perl_atan2(left, right));
2699 /* also used for: pp_cos() pp_exp() pp_log() pp_sqrt() */
2704 int amg_type = fallback_amg;
2705 const char *neg_report = NULL;
2706 const int op_type = PL_op->op_type;
2709 case OP_SIN: amg_type = sin_amg; break;
2710 case OP_COS: amg_type = cos_amg; break;
2711 case OP_EXP: amg_type = exp_amg; break;
2712 case OP_LOG: amg_type = log_amg; neg_report = "log"; break;
2713 case OP_SQRT: amg_type = sqrt_amg; neg_report = "sqrt"; break;
2716 assert(amg_type != fallback_amg);
2718 tryAMAGICun_MG(amg_type, 0);
2720 SV * const arg = TOPs;
2721 const NV value = SvNV_nomg(arg);
2723 if (neg_report) { /* log or sqrt */
2725 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2726 ! Perl_isnan(value) &&
2728 (op_type == OP_LOG ? (value <= 0.0) : (value < 0.0))) {
2729 SET_NUMERIC_STANDARD();
2730 /* diag_listed_as: Can't take log of %g */
2731 DIE(aTHX_ "Can't take %s of %"NVgf, neg_report, value);
2736 case OP_SIN: result = Perl_sin(value); break;
2737 case OP_COS: result = Perl_cos(value); break;
2738 case OP_EXP: result = Perl_exp(value); break;
2739 case OP_LOG: result = Perl_log(value); break;
2740 case OP_SQRT: result = Perl_sqrt(value); break;
2747 /* Support Configure command-line overrides for rand() functions.
2748 After 5.005, perhaps we should replace this by Configure support
2749 for drand48(), random(), or rand(). For 5.005, though, maintain
2750 compatibility by calling rand() but allow the user to override it.
2751 See INSTALL for details. --Andy Dougherty 15 July 1998
2753 /* Now it's after 5.005, and Configure supports drand48() and random(),
2754 in addition to rand(). So the overrides should not be needed any more.
2755 --Jarkko Hietaniemi 27 September 1998
2760 if (!PL_srand_called) {
2761 (void)seedDrand01((Rand_seed_t)seed());
2762 PL_srand_called = TRUE;
2774 SV * const sv = POPs;
2780 /* 1 of 2 things can be carried through SvNV, SP or TARG, SP was carried */
2781 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2782 if (! Perl_isnan(value) && value == 0.0)
2792 sv_setnv_mg(TARG, value);
2803 if (MAXARG >= 1 && (TOPs || POPs)) {
2810 pv = SvPV(top, len);
2811 flags = grok_number(pv, len, &anum);
2813 if (!(flags & IS_NUMBER_IN_UV)) {
2814 Perl_ck_warner_d(aTHX_ packWARN(WARN_OVERFLOW),
2815 "Integer overflow in srand");
2823 (void)seedDrand01((Rand_seed_t)anum);
2824 PL_srand_called = TRUE;
2828 /* Historically srand always returned true. We can avoid breaking
2830 sv_setpvs(TARG, "0 but true");
2839 tryAMAGICun_MG(int_amg, AMGf_numeric);
2841 SV * const sv = TOPs;
2842 const IV iv = SvIV_nomg(sv);
2843 /* XXX it's arguable that compiler casting to IV might be subtly
2844 different from modf (for numbers inside (IV_MIN,UV_MAX)) in which
2845 else preferring IV has introduced a subtle behaviour change bug. OTOH
2846 relying on floating point to be accurate is a bug. */
2851 else if (SvIOK(sv)) {
2853 SETu(SvUV_nomg(sv));
2858 const NV value = SvNV_nomg(sv);
2859 if (UNLIKELY(Perl_isinfnan(value)))
2861 else if (value >= 0.0) {
2862 if (value < (NV)UV_MAX + 0.5) {
2865 SETn(Perl_floor(value));
2869 if (value > (NV)IV_MIN - 0.5) {
2872 SETn(Perl_ceil(value));
2883 tryAMAGICun_MG(abs_amg, AMGf_numeric);
2885 SV * const sv = TOPs;
2886 /* This will cache the NV value if string isn't actually integer */
2887 const IV iv = SvIV_nomg(sv);
2892 else if (SvIOK(sv)) {
2893 /* IVX is precise */
2895 SETu(SvUV_nomg(sv)); /* force it to be numeric only */
2903 /* 2s complement assumption. Also, not really needed as
2904 IV_MIN and -IV_MIN should both be %100...00 and NV-able */
2910 const NV value = SvNV_nomg(sv);
2921 /* also used for: pp_hex() */
2927 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES;
2931 SV* const sv = TOPs;
2933 tmps = (SvPV_const(sv, len));
2935 /* If Unicode, try to downgrade
2936 * If not possible, croak. */
2937 SV* const tsv = sv_2mortal(newSVsv(sv));
2940 sv_utf8_downgrade(tsv, FALSE);
2941 tmps = SvPV_const(tsv, len);
2943 if (PL_op->op_type == OP_HEX)
2946 while (*tmps && len && isSPACE(*tmps))
2950 if (isALPHA_FOLD_EQ(*tmps, 'x')) {
2952 result_uv = grok_hex (tmps, &len, &flags, &result_nv);
2954 else if (isALPHA_FOLD_EQ(*tmps, 'b'))
2955 result_uv = grok_bin (tmps, &len, &flags, &result_nv);
2957 result_uv = grok_oct (tmps, &len, &flags, &result_nv);
2959 if (flags & PERL_SCAN_GREATER_THAN_UV_MAX) {
2973 SV * const sv = TOPs;
2975 U32 in_bytes = IN_BYTES;
2976 /* simplest case shortcut */
2977 /* turn off SVf_UTF8 in tmp flags if HINT_BYTES on*/
2978 U32 svflags = (SvFLAGS(sv) ^ (in_bytes << 26)) & (SVf_POK|SVs_GMG|SVf_UTF8);
2979 STATIC_ASSERT_STMT(HINT_BYTES == 0x00000008 && SVf_UTF8 == 0x20000000 && (SVf_UTF8 == HINT_BYTES << 26));
2982 if(LIKELY(svflags == SVf_POK))
2984 if(svflags & SVs_GMG)
2987 if (!IN_BYTES) /* reread to avoid using an C auto/register */
2988 sv_setiv(TARG, (IV)sv_len_utf8_nomg(sv));
2992 /* unrolled SvPV_nomg_const(sv,len) */
2997 (void)sv_2pv_flags(sv, &len, 0|SV_CONST_RETURN);
2999 sv_setiv(TARG, (IV)(len));
3002 if (!SvPADTMP(TARG)) {
3003 sv_setsv_nomg(TARG, &PL_sv_undef);
3004 } else { /* TARG is on stack at this point and is overwriten by SETs.
3005 This branch is the odd one out, so put TARG by default on
3006 stack earlier to let local SP go out of liveness sooner */
3013 return NORMAL; /* no putback, SP didn't move in this opcode */
3016 /* Returns false if substring is completely outside original string.
3017 No length is indicated by len_iv = 0 and len_is_uv = 0. len_is_uv must
3018 always be true for an explicit 0.
3021 Perl_translate_substr_offsets( STRLEN curlen, IV pos1_iv,
3022 bool pos1_is_uv, IV len_iv,
3023 bool len_is_uv, STRLEN *posp,
3029 PERL_ARGS_ASSERT_TRANSLATE_SUBSTR_OFFSETS;
3031 if (!pos1_is_uv && pos1_iv < 0 && curlen) {
3032 pos1_is_uv = curlen-1 > ~(UV)pos1_iv;
3035 if ((pos1_is_uv || pos1_iv > 0) && (UV)pos1_iv > curlen)
3038 if (len_iv || len_is_uv) {
3039 if (!len_is_uv && len_iv < 0) {
3040 pos2_iv = curlen + len_iv;
3042 pos2_is_uv = curlen-1 > ~(UV)len_iv;
3045 } else { /* len_iv >= 0 */
3046 if (!pos1_is_uv && pos1_iv < 0) {
3047 pos2_iv = pos1_iv + len_iv;
3048 pos2_is_uv = (UV)len_iv > (UV)IV_MAX;
3050 if ((UV)len_iv > curlen-(UV)pos1_iv)
3053 pos2_iv = pos1_iv+len_iv;
3063 if (!pos2_is_uv && pos2_iv < 0) {
3064 if (!pos1_is_uv && pos1_iv < 0)
3068 else if (!pos1_is_uv && pos1_iv < 0)
3071 if ((UV)pos2_iv < (UV)pos1_iv)
3073 if ((UV)pos2_iv > curlen)
3076 /* pos1_iv and pos2_iv both in 0..curlen, so the cast is safe */
3077 *posp = (STRLEN)( (UV)pos1_iv );
3078 *lenp = (STRLEN)( (UV)pos2_iv - (UV)pos1_iv );
3095 I32 lvalue = PL_op->op_flags & OPf_MOD || LVRET;
3096 const bool rvalue = (GIMME_V != G_VOID);
3099 const char *repl = NULL;
3101 int num_args = PL_op->op_private & 7;
3102 bool repl_need_utf8_upgrade = FALSE;
3106 if(!(repl_sv = POPs)) num_args--;
3108 if ((len_sv = POPs)) {
3109 len_iv = SvIV(len_sv);
3110 len_is_uv = len_iv ? SvIOK_UV(len_sv) : 1;
3115 pos1_iv = SvIV(pos_sv);
3116 pos1_is_uv = SvIOK_UV(pos_sv);
3118 if (PL_op->op_private & OPpSUBSTR_REPL_FIRST) {
3122 if (lvalue && !repl_sv) {
3124 ret = sv_2mortal(newSV_type(SVt_PVLV)); /* Not TARG RT#67838 */
3125 sv_magic(ret, NULL, PERL_MAGIC_substr, NULL, 0);
3127 LvTARG(ret) = SvREFCNT_inc_simple(sv);
3129 pos1_is_uv || pos1_iv >= 0
3130 ? (STRLEN)(UV)pos1_iv
3131 : (LvFLAGS(ret) |= 1, (STRLEN)(UV)-pos1_iv);
3133 len_is_uv || len_iv > 0
3134 ? (STRLEN)(UV)len_iv
3135 : (LvFLAGS(ret) |= 2, (STRLEN)(UV)-len_iv);
3137 PUSHs(ret); /* avoid SvSETMAGIC here */
3141 repl = SvPV_const(repl_sv, repl_len);
3144 Perl_ck_warner(aTHX_ packWARN(WARN_SUBSTR),
3145 "Attempt to use reference as lvalue in substr"
3147 tmps = SvPV_force_nomg(sv, curlen);
3148 if (DO_UTF8(repl_sv) && repl_len) {
3150 sv_utf8_upgrade_nomg(sv);
3154 else if (DO_UTF8(sv))
3155 repl_need_utf8_upgrade = TRUE;
3157 else tmps = SvPV_const(sv, curlen);
3159 utf8_curlen = sv_or_pv_len_utf8(sv, tmps, curlen);
3160 if (utf8_curlen == curlen)
3163 curlen = utf8_curlen;
3169 STRLEN pos, len, byte_len, byte_pos;
3171 if (!translate_substr_offsets(
3172 curlen, pos1_iv, pos1_is_uv, len_iv, len_is_uv, &pos, &len
3176 byte_pos = utf8_curlen
3177 ? sv_or_pv_pos_u2b(sv, tmps, pos, &byte_len) : pos;
3182 SvTAINTED_off(TARG); /* decontaminate */
3183 SvUTF8_off(TARG); /* decontaminate */
3184 sv_setpvn(TARG, tmps, byte_len);
3185 #ifdef USE_LOCALE_COLLATE
3186 sv_unmagic(TARG, PERL_MAGIC_collxfrm);
3193 SV* repl_sv_copy = NULL;
3195 if (repl_need_utf8_upgrade) {
3196 repl_sv_copy = newSVsv(repl_sv);
3197 sv_utf8_upgrade(repl_sv_copy);
3198 repl = SvPV_const(repl_sv_copy, repl_len);
3202 sv_insert_flags(sv, byte_pos, byte_len, repl, repl_len, 0);
3203 SvREFCNT_dec(repl_sv_copy);
3206 if (PL_op->op_private & OPpSUBSTR_REPL_FIRST)
3216 Perl_croak(aTHX_ "substr outside of string");
3217 Perl_ck_warner(aTHX_ packWARN(WARN_SUBSTR), "substr outside of string");
3224 const IV size = POPi;
3225 const IV offset = POPi;
3226 SV * const src = POPs;
3227 const I32 lvalue = PL_op->op_flags & OPf_MOD || LVRET;
3230 if (lvalue) { /* it's an lvalue! */
3231 ret = sv_2mortal(newSV_type(SVt_PVLV)); /* Not TARG RT#67838 */
3232 sv_magic(ret, NULL, PERL_MAGIC_vec, NULL, 0);
3234 LvTARG(ret) = SvREFCNT_inc_simple(src);
3235 LvTARGOFF(ret) = offset;
3236 LvTARGLEN(ret) = size;
3240 SvTAINTED_off(TARG); /* decontaminate */
3244 sv_setuv(ret, do_vecget(src, offset, size));
3252 /* also used for: pp_rindex() */
3265 const char *little_p;
3268 const bool is_index = PL_op->op_type == OP_INDEX;
3269 const bool threeargs = MAXARG >= 3 && (TOPs || ((void)POPs,0));
3275 big_p = SvPV_const(big, biglen);
3276 little_p = SvPV_const(little, llen);
3278 big_utf8 = DO_UTF8(big);
3279 little_utf8 = DO_UTF8(little);
3280 if (big_utf8 ^ little_utf8) {
3281 /* One needs to be upgraded. */
3282 if (little_utf8 && !IN_ENCODING) {
3283 /* Well, maybe instead we might be able to downgrade the small
3285 char * const pv = (char*)bytes_from_utf8((U8 *)little_p, &llen,
3288 /* If the large string is ISO-8859-1, and it's not possible to
3289 convert the small string to ISO-8859-1, then there is no
3290 way that it could be found anywhere by index. */
3295 /* At this point, pv is a malloc()ed string. So donate it to temp
3296 to ensure it will get free()d */
3297 little = temp = newSV(0);
3298 sv_usepvn(temp, pv, llen);
3299 little_p = SvPVX(little);
3302 ? newSVpvn(big_p, biglen) : newSVpvn(little_p, llen);
3305 sv_recode_to_utf8(temp, _get_encoding());
3307 sv_utf8_upgrade(temp);
3312 big_p = SvPV_const(big, biglen);
3315 little_p = SvPV_const(little, llen);
3319 if (SvGAMAGIC(big)) {
3320 /* Life just becomes a lot easier if I use a temporary here.
3321 Otherwise I need to avoid calls to sv_pos_u2b(), which (dangerously)
3322 will trigger magic and overloading again, as will fbm_instr()
3324 big = newSVpvn_flags(big_p, biglen,
3325 SVs_TEMP | (big_utf8 ? SVf_UTF8 : 0));
3328 if (SvGAMAGIC(little) || (is_index && !SvOK(little))) {
3329 /* index && SvOK() is a hack. fbm_instr() calls SvPV_const, which will
3330 warn on undef, and we've already triggered a warning with the
3331 SvPV_const some lines above. We can't remove that, as we need to
3332 call some SvPV to trigger overloading early and find out if the
3334 This is all getting to messy. The API isn't quite clean enough,
3335 because data access has side effects.
3337 little = newSVpvn_flags(little_p, llen,
3338 SVs_TEMP | (little_utf8 ? SVf_UTF8 : 0));
3339 little_p = SvPVX(little);
3343 offset = is_index ? 0 : biglen;
3345 if (big_utf8 && offset > 0)
3346 offset = sv_pos_u2b_flags(big, offset, 0, SV_CONST_RETURN);
3352 else if (offset > (SSize_t)biglen)
3354 if (!(little_p = is_index
3355 ? fbm_instr((unsigned char*)big_p + offset,
3356 (unsigned char*)big_p + biglen, little, 0)
3357 : rninstr(big_p, big_p + offset,
3358 little_p, little_p + llen)))
3361 retval = little_p - big_p;
3362 if (retval > 1 && big_utf8)
3363 retval = sv_pos_b2u_flags(big, retval, SV_CONST_RETURN);
3373 dSP; dMARK; dORIGMARK; dTARGET;
3374 SvTAINTED_off(TARG);
3375 do_sprintf(TARG, SP-MARK, MARK+1);
3376 TAINT_IF(SvTAINTED(TARG));
3388 const U8 *s = (U8*)SvPV_const(argsv, len);
3390 if (IN_ENCODING && SvPOK(argsv) && !DO_UTF8(argsv)) {
3391 SV * const tmpsv = sv_2mortal(newSVsv(argsv));
3392 s = (U8*)sv_recode_to_utf8(tmpsv, _get_encoding());
3393 len = UTF8SKIP(s); /* Should be well-formed; so this is its length */
3398 ? utf8n_to_uvchr(s, len, 0, UTF8_ALLOW_ANYUV)
3412 if (UNLIKELY(SvAMAGIC(top)))
3414 if (UNLIKELY(isinfnansv(top)))
3415 Perl_croak(aTHX_ "Cannot chr %"NVgf, SvNV(top));
3417 if (!IN_BYTES /* under bytes, chr(-1) eq chr(0xff), etc. */
3418 && ((SvIOKp(top) && !SvIsUV(top) && SvIV_nomg(top) < 0)
3420 ((SvNOKp(top) || (SvOK(top) && !SvIsUV(top)))
3421 && SvNV_nomg(top) < 0.0))) {
3422 if (ckWARN(WARN_UTF8)) {
3423 if (SvGMAGICAL(top)) {
3424 SV *top2 = sv_newmortal();
3425 sv_setsv_nomg(top2, top);
3428 Perl_warner(aTHX_ packWARN(WARN_UTF8),
3429 "Invalid negative number (%"SVf") in chr", SVfARG(top));
3431 value = UNICODE_REPLACEMENT;
3433 value = SvUV_nomg(top);
3437 SvUPGRADE(TARG,SVt_PV);
3439 if (value > 255 && !IN_BYTES) {
3440 SvGROW(TARG, (STRLEN)UNISKIP(value)+1);
3441 tmps = (char*)uvchr_to_utf8_flags((U8*)SvPVX(TARG), value, 0);
3442 SvCUR_set(TARG, tmps - SvPVX_const(TARG));
3444 (void)SvPOK_only(TARG);
3453 *tmps++ = (char)value;
3455 (void)SvPOK_only(TARG);
3457 if (IN_ENCODING && !IN_BYTES) {
3458 sv_recode_to_utf8(TARG, _get_encoding());
3460 if (SvCUR(TARG) == 0
3461 || ! is_utf8_string((U8*)tmps, SvCUR(TARG))
3462 || UTF8_IS_REPLACEMENT((U8*) tmps, (U8*) tmps + SvCUR(TARG)))
3467 *tmps++ = (char)value;
3483 const char *tmps = SvPV_const(left, len);
3485 if (DO_UTF8(left)) {
3486 /* If Unicode, try to downgrade.
3487 * If not possible, croak.
3488 * Yes, we made this up. */
3489 SV* const tsv = newSVpvn_flags(tmps, len, SVf_UTF8|SVs_TEMP);
3491 sv_utf8_downgrade(tsv, FALSE);
3492 tmps = SvPV_const(tsv, len);
3494 # ifdef USE_ITHREADS
3496 if (!PL_reentrant_buffer->_crypt_struct_buffer) {
3497 /* This should be threadsafe because in ithreads there is only
3498 * one thread per interpreter. If this would not be true,
3499 * we would need a mutex to protect this malloc. */
3500 PL_reentrant_buffer->_crypt_struct_buffer =
3501 (struct crypt_data *)safemalloc(sizeof(struct crypt_data));
3502 #if defined(__GLIBC__) || defined(__EMX__)
3503 if (PL_reentrant_buffer->_crypt_struct_buffer) {
3504 PL_reentrant_buffer->_crypt_struct_buffer->initialized = 0;
3505 /* work around glibc-2.2.5 bug */
3506 PL_reentrant_buffer->_crypt_struct_buffer->current_saltbits = 0;
3510 # endif /* HAS_CRYPT_R */
3511 # endif /* USE_ITHREADS */
3513 sv_setpv(TARG, fcrypt(tmps, SvPV_nolen_const(right)));
3515 sv_setpv(TARG, PerlProc_crypt(tmps, SvPV_nolen_const(right)));
3522 "The crypt() function is unimplemented due to excessive paranoia.");
3526 /* Generally UTF-8 and UTF-EBCDIC are indistinguishable at this level. So
3527 * most comments below say UTF-8, when in fact they mean UTF-EBCDIC as well */
3530 /* also used for: pp_lcfirst() */
3534 /* Actually is both lcfirst() and ucfirst(). Only the first character
3535 * changes. This means that possibly we can change in-place, ie., just
3536 * take the source and change that one character and store it back, but not
3537 * if read-only etc, or if the length changes */
3541 STRLEN slen; /* slen is the byte length of the whole SV. */
3544 bool inplace; /* ? Convert first char only, in-place */
3545 bool doing_utf8 = FALSE; /* ? using utf8 */
3546 bool convert_source_to_utf8 = FALSE; /* ? need to convert */
3547 const int op_type = PL_op->op_type;
3550 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
3551 STRLEN ulen; /* ulen is the byte length of the original Unicode character
3552 * stored as UTF-8 at s. */
3553 STRLEN tculen; /* tculen is the byte length of the freshly titlecased (or
3554 * lowercased) character stored in tmpbuf. May be either
3555 * UTF-8 or not, but in either case is the number of bytes */
3557 s = (const U8*)SvPV_const(source, slen);
3559 /* We may be able to get away with changing only the first character, in
3560 * place, but not if read-only, etc. Later we may discover more reasons to
3561 * not convert in-place. */
3562 inplace = !SvREADONLY(source)
3563 && ( SvPADTMP(source)
3564 || ( SvTEMP(source) && !SvSMAGICAL(source)
3565 && SvREFCNT(source) == 1));
3567 /* First calculate what the changed first character should be. This affects
3568 * whether we can just swap it out, leaving the rest of the string unchanged,
3569 * or even if have to convert the dest to UTF-8 when the source isn't */
3571 if (! slen) { /* If empty */
3572 need = 1; /* still need a trailing NUL */
3575 else if (DO_UTF8(source)) { /* Is the source utf8? */
3578 if (op_type == OP_UCFIRST) {
3579 #ifdef USE_LOCALE_CTYPE
3580 _to_utf8_title_flags(s, tmpbuf, &tculen, IN_LC_RUNTIME(LC_CTYPE));
3582 _to_utf8_title_flags(s, tmpbuf, &tculen, 0);
3586 #ifdef USE_LOCALE_CTYPE
3587 _to_utf8_lower_flags(s, tmpbuf, &tculen, IN_LC_RUNTIME(LC_CTYPE));
3589 _to_utf8_lower_flags(s, tmpbuf, &tculen, 0);
3593 /* we can't do in-place if the length changes. */
3594 if (ulen != tculen) inplace = FALSE;
3595 need = slen + 1 - ulen + tculen;
3597 else { /* Non-zero length, non-UTF-8, Need to consider locale and if
3598 * latin1 is treated as caseless. Note that a locale takes
3600 ulen = 1; /* Original character is 1 byte */
3601 tculen = 1; /* Most characters will require one byte, but this will
3602 * need to be overridden for the tricky ones */
3605 if (op_type == OP_LCFIRST) {
3607 /* lower case the first letter: no trickiness for any character */
3608 #ifdef USE_LOCALE_CTYPE
3609 if (IN_LC_RUNTIME(LC_CTYPE)) {
3610 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
3611 *tmpbuf = toLOWER_LC(*s);
3616 *tmpbuf = (IN_UNI_8_BIT)
3617 ? toLOWER_LATIN1(*s)
3621 #ifdef USE_LOCALE_CTYPE
3623 else if (IN_LC_RUNTIME(LC_CTYPE)) {
3624 if (IN_UTF8_CTYPE_LOCALE) {
3628 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
3629 *tmpbuf = (U8) toUPPER_LC(*s); /* This would be a bug if any
3630 locales have upper and title case
3634 else if (! IN_UNI_8_BIT) {
3635 *tmpbuf = toUPPER(*s); /* Returns caseless for non-ascii, or
3636 * on EBCDIC machines whatever the
3637 * native function does */
3640 /* Here, is ucfirst non-UTF-8, not in locale (unless that locale is
3641 * UTF-8, which we treat as not in locale), and cased latin1 */
3643 #ifdef USE_LOCALE_CTYPE
3647 title_ord = _to_upper_title_latin1(*s, tmpbuf, &tculen, 's');
3649 assert(tculen == 2);
3651 /* If the result is an upper Latin1-range character, it can
3652 * still be represented in one byte, which is its ordinal */
3653 if (UTF8_IS_DOWNGRADEABLE_START(*tmpbuf)) {
3654 *tmpbuf = (U8) title_ord;
3658 /* Otherwise it became more than one ASCII character (in
3659 * the case of LATIN_SMALL_LETTER_SHARP_S) or changed to
3660 * beyond Latin1, so the number of bytes changed, so can't
3661 * replace just the first character in place. */
3664 /* If the result won't fit in a byte, the entire result
3665 * will have to be in UTF-8. Assume worst case sizing in
3666 * conversion. (all latin1 characters occupy at most two
3668 if (title_ord > 255) {
3670 convert_source_to_utf8 = TRUE;
3671 need = slen * 2 + 1;
3673 /* The (converted) UTF-8 and UTF-EBCDIC lengths of all
3674 * (both) characters whose title case is above 255 is
3678 else { /* LATIN_SMALL_LETTER_SHARP_S expands by 1 byte */
3679 need = slen + 1 + 1;
3683 } /* End of use Unicode (Latin1) semantics */
3684 } /* End of changing the case of the first character */
3686 /* Here, have the first character's changed case stored in tmpbuf. Ready to
3687 * generate the result */
3690 /* We can convert in place. This means we change just the first
3691 * character without disturbing the rest; no need to grow */
3693 s = d = (U8*)SvPV_force_nomg(source, slen);
3699 /* Here, we can't convert in place; we earlier calculated how much
3700 * space we will need, so grow to accommodate that */
3701 SvUPGRADE(dest, SVt_PV);
3702 d = (U8*)SvGROW(dest, need);
3703 (void)SvPOK_only(dest);
3710 if (! convert_source_to_utf8) {
3712 /* Here both source and dest are in UTF-8, but have to create
3713 * the entire output. We initialize the result to be the
3714 * title/lower cased first character, and then append the rest
3716 sv_setpvn(dest, (char*)tmpbuf, tculen);
3718 sv_catpvn(dest, (char*)(s + ulen), slen - ulen);
3722 const U8 *const send = s + slen;
3724 /* Here the dest needs to be in UTF-8, but the source isn't,
3725 * except we earlier UTF-8'd the first character of the source
3726 * into tmpbuf. First put that into dest, and then append the
3727 * rest of the source, converting it to UTF-8 as we go. */
3729 /* Assert tculen is 2 here because the only two characters that
3730 * get to this part of the code have 2-byte UTF-8 equivalents */
3732 *d++ = *(tmpbuf + 1);
3733 s++; /* We have just processed the 1st char */
3735 for (; s < send; s++) {
3736 d = uvchr_to_utf8(d, *s);
3739 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
3743 else { /* in-place UTF-8. Just overwrite the first character */
3744 Copy(tmpbuf, d, tculen, U8);
3745 SvCUR_set(dest, need - 1);
3749 else { /* Neither source nor dest are in or need to be UTF-8 */
3751 if (inplace) { /* in-place, only need to change the 1st char */
3754 else { /* Not in-place */
3756 /* Copy the case-changed character(s) from tmpbuf */
3757 Copy(tmpbuf, d, tculen, U8);
3758 d += tculen - 1; /* Code below expects d to point to final
3759 * character stored */
3762 else { /* empty source */
3763 /* See bug #39028: Don't taint if empty */
3767 /* In a "use bytes" we don't treat the source as UTF-8, but, still want
3768 * the destination to retain that flag */
3769 if (SvUTF8(source) && ! IN_BYTES)
3772 if (!inplace) { /* Finish the rest of the string, unchanged */
3773 /* This will copy the trailing NUL */
3774 Copy(s + 1, d + 1, slen, U8);
3775 SvCUR_set(dest, need - 1);
3778 #ifdef USE_LOCALE_CTYPE
3779 if (IN_LC_RUNTIME(LC_CTYPE)) {
3784 if (dest != source && SvTAINTED(source))
3790 /* There's so much setup/teardown code common between uc and lc, I wonder if
3791 it would be worth merging the two, and just having a switch outside each
3792 of the three tight loops. There is less and less commonality though */
3805 if ((SvPADTMP(source)
3807 (SvTEMP(source) && !SvSMAGICAL(source) && SvREFCNT(source) == 1))
3808 && !SvREADONLY(source) && SvPOK(source)
3811 #ifdef USE_LOCALE_CTYPE
3812 (IN_LC_RUNTIME(LC_CTYPE))
3813 ? ! IN_UTF8_CTYPE_LOCALE
3819 /* We can convert in place. The reason we can't if in UNI_8_BIT is to
3820 * make the loop tight, so we overwrite the source with the dest before
3821 * looking at it, and we need to look at the original source
3822 * afterwards. There would also need to be code added to handle
3823 * switching to not in-place in midstream if we run into characters
3824 * that change the length. Since being in locale overrides UNI_8_BIT,
3825 * that latter becomes irrelevant in the above test; instead for
3826 * locale, the size can't normally change, except if the locale is a
3829 s = d = (U8*)SvPV_force_nomg(source, len);
3836 s = (const U8*)SvPV_nomg_const(source, len);
3839 SvUPGRADE(dest, SVt_PV);
3840 d = (U8*)SvGROW(dest, min);
3841 (void)SvPOK_only(dest);
3846 /* Overloaded values may have toggled the UTF-8 flag on source, so we need
3847 to check DO_UTF8 again here. */
3849 if (DO_UTF8(source)) {
3850 const U8 *const send = s + len;
3851 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
3853 /* All occurrences of these are to be moved to follow any other marks.
3854 * This is context-dependent. We may not be passed enough context to
3855 * move the iota subscript beyond all of them, but we do the best we can
3856 * with what we're given. The result is always better than if we
3857 * hadn't done this. And, the problem would only arise if we are
3858 * passed a character without all its combining marks, which would be
3859 * the caller's mistake. The information this is based on comes from a
3860 * comment in Unicode SpecialCasing.txt, (and the Standard's text
3861 * itself) and so can't be checked properly to see if it ever gets
3862 * revised. But the likelihood of it changing is remote */
3863 bool in_iota_subscript = FALSE;
3869 if (in_iota_subscript && ! _is_utf8_mark(s)) {
3871 /* A non-mark. Time to output the iota subscript */
3872 Copy(GREEK_CAPITAL_LETTER_IOTA_UTF8, d, capital_iota_len, U8);
3873 d += capital_iota_len;
3874 in_iota_subscript = FALSE;
3877 /* Then handle the current character. Get the changed case value
3878 * and copy it to the output buffer */
3881 #ifdef USE_LOCALE_CTYPE
3882 uv = _to_utf8_upper_flags(s, tmpbuf, &ulen, IN_LC_RUNTIME(LC_CTYPE));
3884 uv = _to_utf8_upper_flags(s, tmpbuf, &ulen, 0);
3886 #define GREEK_CAPITAL_LETTER_IOTA 0x0399
3887 #define COMBINING_GREEK_YPOGEGRAMMENI 0x0345
3888 if (uv == GREEK_CAPITAL_LETTER_IOTA
3889 && utf8_to_uvchr_buf(s, send, 0) == COMBINING_GREEK_YPOGEGRAMMENI)
3891 in_iota_subscript = TRUE;
3894 if (ulen > u && (SvLEN(dest) < (min += ulen - u))) {
3895 /* If the eventually required minimum size outgrows the
3896 * available space, we need to grow. */
3897 const UV o = d - (U8*)SvPVX_const(dest);
3899 /* If someone uppercases one million U+03B0s we SvGROW()
3900 * one million times. Or we could try guessing how much to
3901 * allocate without allocating too much. Such is life.
3902 * See corresponding comment in lc code for another option
3905 d = (U8*)SvPVX(dest) + o;
3907 Copy(tmpbuf, d, ulen, U8);
3912 if (in_iota_subscript) {
3913 Copy(GREEK_CAPITAL_LETTER_IOTA_UTF8, d, capital_iota_len, U8);
3914 d += capital_iota_len;
3919 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
3921 else { /* Not UTF-8 */
3923 const U8 *const send = s + len;
3925 /* Use locale casing if in locale; regular style if not treating
3926 * latin1 as having case; otherwise the latin1 casing. Do the
3927 * whole thing in a tight loop, for speed, */
3928 #ifdef USE_LOCALE_CTYPE
3929 if (IN_LC_RUNTIME(LC_CTYPE)) {
3930 if (IN_UTF8_CTYPE_LOCALE) {
3933 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
3934 for (; s < send; d++, s++)
3935 *d = (U8) toUPPER_LC(*s);
3939 if (! IN_UNI_8_BIT) {
3940 for (; s < send; d++, s++) {
3945 #ifdef USE_LOCALE_CTYPE
3948 for (; s < send; d++, s++) {
3949 *d = toUPPER_LATIN1_MOD(*s);
3950 if (LIKELY(*d != LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
3954 /* The mainstream case is the tight loop above. To avoid
3955 * extra tests in that, all three characters that require
3956 * special handling are mapped by the MOD to the one tested
3958 * Use the source to distinguish between the three cases */
3960 if (*s == LATIN_SMALL_LETTER_SHARP_S) {
3962 /* uc() of this requires 2 characters, but they are
3963 * ASCII. If not enough room, grow the string */
3964 if (SvLEN(dest) < ++min) {
3965 const UV o = d - (U8*)SvPVX_const(dest);
3967 d = (U8*)SvPVX(dest) + o;
3969 *d++ = 'S'; *d = 'S'; /* upper case is 'SS' */
3970 continue; /* Back to the tight loop; still in ASCII */
3973 /* The other two special handling characters have their
3974 * upper cases outside the latin1 range, hence need to be
3975 * in UTF-8, so the whole result needs to be in UTF-8. So,
3976 * here we are somewhere in the middle of processing a
3977 * non-UTF-8 string, and realize that we will have to convert
3978 * the whole thing to UTF-8. What to do? There are
3979 * several possibilities. The simplest to code is to
3980 * convert what we have so far, set a flag, and continue on
3981 * in the loop. The flag would be tested each time through
3982 * the loop, and if set, the next character would be
3983 * converted to UTF-8 and stored. But, I (khw) didn't want
3984 * to slow down the mainstream case at all for this fairly
3985 * rare case, so I didn't want to add a test that didn't
3986 * absolutely have to be there in the loop, besides the
3987 * possibility that it would get too complicated for
3988 * optimizers to deal with. Another possibility is to just
3989 * give up, convert the source to UTF-8, and restart the
3990 * function that way. Another possibility is to convert
3991 * both what has already been processed and what is yet to
3992 * come separately to UTF-8, then jump into the loop that
3993 * handles UTF-8. But the most efficient time-wise of the
3994 * ones I could think of is what follows, and turned out to
3995 * not require much extra code. */
3997 /* Convert what we have so far into UTF-8, telling the
3998 * function that we know it should be converted, and to
3999 * allow extra space for what we haven't processed yet.
4000 * Assume the worst case space requirements for converting
4001 * what we haven't processed so far: that it will require
4002 * two bytes for each remaining source character, plus the
4003 * NUL at the end. This may cause the string pointer to
4004 * move, so re-find it. */
4006 len = d - (U8*)SvPVX_const(dest);
4007 SvCUR_set(dest, len);
4008 len = sv_utf8_upgrade_flags_grow(dest,
4009 SV_GMAGIC|SV_FORCE_UTF8_UPGRADE,
4011 d = (U8*)SvPVX(dest) + len;
4013 /* Now process the remainder of the source, converting to
4014 * upper and UTF-8. If a resulting byte is invariant in
4015 * UTF-8, output it as-is, otherwise convert to UTF-8 and
4016 * append it to the output. */
4017 for (; s < send; s++) {
4018 (void) _to_upper_title_latin1(*s, d, &len, 'S');
4022 /* Here have processed the whole source; no need to continue
4023 * with the outer loop. Each character has been converted
4024 * to upper case and converted to UTF-8 */
4027 } /* End of processing all latin1-style chars */
4028 } /* End of processing all chars */
4029 } /* End of source is not empty */
4031 if (source != dest) {
4032 *d = '\0'; /* Here d points to 1 after last char, add NUL */
4033 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
4035 } /* End of isn't utf8 */
4036 #ifdef USE_LOCALE_CTYPE
4037 if (IN_LC_RUNTIME(LC_CTYPE)) {
4042 if (dest != source && SvTAINTED(source))
4060 if ( ( SvPADTMP(source)
4061 || ( SvTEMP(source) && !SvSMAGICAL(source)
4062 && SvREFCNT(source) == 1 )
4064 && !SvREADONLY(source) && SvPOK(source)
4065 && !DO_UTF8(source)) {
4067 /* We can convert in place, as lowercasing anything in the latin1 range
4068 * (or else DO_UTF8 would have been on) doesn't lengthen it */
4070 s = d = (U8*)SvPV_force_nomg(source, len);
4077 s = (const U8*)SvPV_nomg_const(source, len);
4080 SvUPGRADE(dest, SVt_PV);
4081 d = (U8*)SvGROW(dest, min);
4082 (void)SvPOK_only(dest);
4087 /* Overloaded values may have toggled the UTF-8 flag on source, so we need
4088 to check DO_UTF8 again here. */
4090 if (DO_UTF8(source)) {
4091 const U8 *const send = s + len;
4092 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
4095 const STRLEN u = UTF8SKIP(s);
4098 #ifdef USE_LOCALE_CTYPE
4099 _to_utf8_lower_flags(s, tmpbuf, &ulen, IN_LC_RUNTIME(LC_CTYPE));
4101 _to_utf8_lower_flags(s, tmpbuf, &ulen, 0);
4104 /* Here is where we would do context-sensitive actions. See the
4105 * commit message for 86510fb15 for why there isn't any */
4107 if (ulen > u && (SvLEN(dest) < (min += ulen - u))) {
4109 /* If the eventually required minimum size outgrows the
4110 * available space, we need to grow. */
4111 const UV o = d - (U8*)SvPVX_const(dest);
4113 /* If someone lowercases one million U+0130s we SvGROW() one
4114 * million times. Or we could try guessing how much to
4115 * allocate without allocating too much. Such is life.
4116 * Another option would be to grow an extra byte or two more
4117 * each time we need to grow, which would cut down the million
4118 * to 500K, with little waste */
4120 d = (U8*)SvPVX(dest) + o;
4123 /* Copy the newly lowercased letter to the output buffer we're
4125 Copy(tmpbuf, d, ulen, U8);
4128 } /* End of looping through the source string */
4131 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
4132 } else { /* Not utf8 */
4134 const U8 *const send = s + len;
4136 /* Use locale casing if in locale; regular style if not treating
4137 * latin1 as having case; otherwise the latin1 casing. Do the
4138 * whole thing in a tight loop, for speed, */
4139 #ifdef USE_LOCALE_CTYPE
4140 if (IN_LC_RUNTIME(LC_CTYPE)) {
4141 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
4142 for (; s < send; d++, s++)
4143 *d = toLOWER_LC(*s);
4147 if (! IN_UNI_8_BIT) {
4148 for (; s < send; d++, s++) {
4153 for (; s < send; d++, s++) {
4154 *d = toLOWER_LATIN1(*s);
4158 if (source != dest) {
4160 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
4163 #ifdef USE_LOCALE_CTYPE
4164 if (IN_LC_RUNTIME(LC_CTYPE)) {
4169 if (dest != source && SvTAINTED(source))
4178 SV * const sv = TOPs;
4180 const char *s = SvPV_const(sv,len);
4182 SvUTF8_off(TARG); /* decontaminate */
4185 SvUPGRADE(TARG, SVt_PV);
4186 SvGROW(TARG, (len * 2) + 1);
4190 STRLEN ulen = UTF8SKIP(s);
4191 bool to_quote = FALSE;
4193 if (UTF8_IS_INVARIANT(*s)) {
4194 if (_isQUOTEMETA(*s)) {
4198 else if (UTF8_IS_DOWNGRADEABLE_START(*s)) {
4200 #ifdef USE_LOCALE_CTYPE
4201 /* In locale, we quote all non-ASCII Latin1 chars.
4202 * Otherwise use the quoting rules */
4204 IN_LC_RUNTIME(LC_CTYPE)
4207 _isQUOTEMETA(TWO_BYTE_UTF8_TO_NATIVE(*s, *(s + 1))))
4212 else if (is_QUOTEMETA_high(s)) {
4227 else if (IN_UNI_8_BIT) {
4229 if (_isQUOTEMETA(*s))
4235 /* For non UNI_8_BIT (and hence in locale) just quote all \W
4236 * including everything above ASCII */
4238 if (!isWORDCHAR_A(*s))
4244 SvCUR_set(TARG, d - SvPVX_const(TARG));
4245 (void)SvPOK_only_UTF8(TARG);
4248 sv_setpvn(TARG, s, len);
4264 U8 tmpbuf[UTF8_MAXBYTES_CASE + 1];
4265 const bool full_folding = TRUE; /* This variable is here so we can easily
4266 move to more generality later */
4267 const U8 flags = ( full_folding ? FOLD_FLAGS_FULL : 0 )
4268 #ifdef USE_LOCALE_CTYPE
4269 | ( IN_LC_RUNTIME(LC_CTYPE) ? FOLD_FLAGS_LOCALE : 0 )
4273 /* This is a facsimile of pp_lc, but with a thousand bugs thanks to me.
4274 * You are welcome(?) -Hugmeir
4282 s = (const U8*)SvPV_nomg_const(source, len);
4284 if (ckWARN(WARN_UNINITIALIZED))
4285 report_uninit(source);
4292 SvUPGRADE(dest, SVt_PV);
4293 d = (U8*)SvGROW(dest, min);
4294 (void)SvPOK_only(dest);
4299 if (DO_UTF8(source)) { /* UTF-8 flagged string. */
4301 const STRLEN u = UTF8SKIP(s);
4304 _to_utf8_fold_flags(s, tmpbuf, &ulen, flags);
4306 if (ulen > u && (SvLEN(dest) < (min += ulen - u))) {
4307 const UV o = d - (U8*)SvPVX_const(dest);
4309 d = (U8*)SvPVX(dest) + o;
4312 Copy(tmpbuf, d, ulen, U8);
4317 } /* Unflagged string */
4319 #ifdef USE_LOCALE_CTYPE
4320 if ( IN_LC_RUNTIME(LC_CTYPE) ) { /* Under locale */
4321 if (IN_UTF8_CTYPE_LOCALE) {
4322 goto do_uni_folding;
4324 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
4325 for (; s < send; d++, s++)
4326 *d = (U8) toFOLD_LC(*s);
4330 if ( !IN_UNI_8_BIT ) { /* Under nothing, or bytes */
4331 for (; s < send; d++, s++)
4335 #ifdef USE_LOCALE_CTYPE
4338 /* For ASCII and the Latin-1 range, there's only two troublesome
4339 * folds, \x{DF} (\N{LATIN SMALL LETTER SHARP S}), which under full
4340 * casefolding becomes 'ss'; and \x{B5} (\N{MICRO SIGN}), which
4341 * under any fold becomes \x{3BC} (\N{GREEK SMALL LETTER MU}) --
4342 * For the rest, the casefold is their lowercase. */
4343 for (; s < send; d++, s++) {
4344 if (*s == MICRO_SIGN) {
4345 /* \N{MICRO SIGN}'s casefold is \N{GREEK SMALL LETTER MU},
4346 * which is outside of the latin-1 range. There's a couple
4347 * of ways to deal with this -- khw discusses them in
4348 * pp_lc/uc, so go there :) What we do here is upgrade what
4349 * we had already casefolded, then enter an inner loop that
4350 * appends the rest of the characters as UTF-8. */
4351 len = d - (U8*)SvPVX_const(dest);
4352 SvCUR_set(dest, len);
4353 len = sv_utf8_upgrade_flags_grow(dest,
4354 SV_GMAGIC|SV_FORCE_UTF8_UPGRADE,
4355 /* The max expansion for latin1
4356 * chars is 1 byte becomes 2 */
4358 d = (U8*)SvPVX(dest) + len;
4360 Copy(GREEK_SMALL_LETTER_MU_UTF8, d, small_mu_len, U8);
4363 for (; s < send; s++) {
4365 UV fc = _to_uni_fold_flags(*s, tmpbuf, &ulen, flags);
4366 if UVCHR_IS_INVARIANT(fc) {
4368 && *s == LATIN_SMALL_LETTER_SHARP_S)
4377 Copy(tmpbuf, d, ulen, U8);
4383 else if (full_folding && *s == LATIN_SMALL_LETTER_SHARP_S) {
4384 /* Under full casefolding, LATIN SMALL LETTER SHARP S
4385 * becomes "ss", which may require growing the SV. */
4386 if (SvLEN(dest) < ++min) {
4387 const UV o = d - (U8*)SvPVX_const(dest);
4389 d = (U8*)SvPVX(dest) + o;
4394 else { /* If it's not one of those two, the fold is their lower
4396 *d = toLOWER_LATIN1(*s);
4402 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
4404 #ifdef USE_LOCALE_CTYPE
4405 if (IN_LC_RUNTIME(LC_CTYPE)) {
4410 if (SvTAINTED(source))
4420 dSP; dMARK; dORIGMARK;
4421 AV *const av = MUTABLE_AV(POPs);
4422 const I32 lval = (PL_op->op_flags & OPf_MOD || LVRET);
4424 if (SvTYPE(av) == SVt_PVAV) {
4425 const bool localizing = PL_op->op_private & OPpLVAL_INTRO;
4426 bool can_preserve = FALSE;
4432 can_preserve = SvCANEXISTDELETE(av);
4435 if (lval && localizing) {
4438 for (svp = MARK + 1; svp <= SP; svp++) {
4439 const SSize_t elem = SvIV(*svp);
4443 if (max > AvMAX(av))
4447 while (++MARK <= SP) {
4449 SSize_t elem = SvIV(*MARK);
4450 bool preeminent = TRUE;
4452 if (localizing && can_preserve) {
4453 /* If we can determine whether the element exist,
4454 * Try to preserve the existenceness of a tied array
4455 * element by using EXISTS and DELETE if possible.
4456 * Fallback to FETCH and STORE otherwise. */
4457 preeminent = av_exists(av, elem);
4460 svp = av_fetch(av, elem, lval);
4463 DIE(aTHX_ PL_no_aelem, elem);
4466 save_aelem(av, elem, svp);
4468 SAVEADELETE(av, elem);
4471 *MARK = svp ? *svp : &PL_sv_undef;
4474 if (GIMME_V != G_ARRAY) {
4476 *++MARK = SP > ORIGMARK ? *SP : &PL_sv_undef;
4485 AV *const av = MUTABLE_AV(POPs);
4486 I32 lval = (PL_op->op_flags & OPf_MOD);
4487 SSize_t items = SP - MARK;
4489 if (PL_op->op_private & OPpMAYBE_LVSUB) {
4490 const I32 flags = is_lvalue_sub();
4492 if (!(flags & OPpENTERSUB_INARGS))
4493 /* diag_listed_as: Can't modify %s in %s */
4494 Perl_croak(aTHX_ "Can't modify index/value array slice in list assignment");
4501 *(MARK+items*2-1) = *(MARK+items);
4507 while (++MARK <= SP) {
4510 svp = av_fetch(av, SvIV(*MARK), lval);
4512 if (!svp || !*svp || *svp == &PL_sv_undef) {
4513 DIE(aTHX_ PL_no_aelem, SvIV(*MARK));
4515 *MARK = sv_mortalcopy(*MARK);
4517 *++MARK = svp ? *svp : &PL_sv_undef;
4519 if (GIMME_V != G_ARRAY) {
4520 MARK = SP - items*2;
4521 *++MARK = items > 0 ? *SP : &PL_sv_undef;
4528 /* Smart dereferencing for keys, values and each */
4530 /* also used for: pp_reach() pp_rvalues() */
4542 (SvTYPE(sv) != SVt_PVHV && SvTYPE(sv) != SVt_PVAV)
4547 "Type of argument to %s must be unblessed hashref or arrayref",
4548 PL_op_desc[PL_op->op_type] );
4551 if (PL_op->op_flags & OPf_SPECIAL && SvTYPE(sv) == SVt_PVAV)
4553 "Can't modify %s in %s",
4554 PL_op_desc[PL_op->op_type], PL_op_desc[PL_op->op_next->op_type]
4557 /* Delegate to correct function for op type */
4559 if (PL_op->op_type == OP_RKEYS || PL_op->op_type == OP_RVALUES) {
4560 return (SvTYPE(sv) == SVt_PVHV) ? Perl_do_kv(aTHX) : Perl_pp_akeys(aTHX);
4563 return (SvTYPE(sv) == SVt_PVHV)
4564 ? Perl_pp_each(aTHX)
4565 : Perl_pp_aeach(aTHX);
4572 AV *array = MUTABLE_AV(POPs);
4573 const I32 gimme = GIMME_V;
4574 IV *iterp = Perl_av_iter_p(aTHX_ array);
4575 const IV current = (*iterp)++;
4577 if (current > av_tindex(array)) {
4579 if (gimme == G_SCALAR)
4587 if (gimme == G_ARRAY) {
4588 SV **const element = av_fetch(array, current, 0);
4589 PUSHs(element ? *element : &PL_sv_undef);
4594 /* also used for: pp_avalues()*/
4598 AV *array = MUTABLE_AV(POPs);
4599 const I32 gimme = GIMME_V;
4601 *Perl_av_iter_p(aTHX_ array) = 0;
4603 if (gimme == G_SCALAR) {
4605 PUSHi(av_tindex(array) + 1);
4607 else if (gimme == G_ARRAY) {
4608 IV n = Perl_av_len(aTHX_ array);
4613 if (PL_op->op_type == OP_AKEYS || PL_op->op_type == OP_RKEYS) {
4614 for (i = 0; i <= n; i++) {
4619 for (i = 0; i <= n; i++) {
4620 SV *const *const elem = Perl_av_fetch(aTHX_ array, i, 0);
4621 PUSHs(elem ? *elem : &PL_sv_undef);
4628 /* Associative arrays. */
4633 HV * hash = MUTABLE_HV(POPs);
4635 const I32 gimme = GIMME_V;
4637 entry = hv_iternext(hash);
4641 SV* const sv = hv_iterkeysv(entry);
4643 if (gimme == G_ARRAY) {
4645 val = hv_iterval(hash, entry);
4649 else if (gimme == G_SCALAR)
4656 S_do_delete_local(pTHX)
4659 const I32 gimme = GIMME_V;
4662 const bool sliced = !!(PL_op->op_private & OPpSLICE);
4663 SV **unsliced_keysv = sliced ? NULL : sp--;
4664 SV * const osv = POPs;
4665 SV **mark = sliced ? PL_stack_base + POPMARK : unsliced_keysv-1;