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.
31 #include "invlist_inline.h"
33 #include "regcharclass.h"
35 /* variations on pp_null */
40 if (GIMME_V == G_SCALAR)
52 assert(SvTYPE(TARG) == SVt_PVCV);
67 CV * const protocv = PadnamePROTOCV(
68 PadlistNAMESARRAY(CvPADLIST(find_runcv(NULL)))[ARGTARG]
70 assert(SvTYPE(TARG) == SVt_PVCV);
72 if (CvISXSUB(protocv)) { /* constant */
73 /* XXX Should we clone it here? */
74 /* If this changes to use SAVECLEARSV, we can move the SAVECLEARSV
75 to introcv and remove the SvPADSTALE_off. */
76 SAVEPADSVANDMORTALIZE(ARGTARG);
77 PAD_SVl(ARGTARG) = SvREFCNT_inc_simple_NN(protocv);
80 if (CvROOT(protocv)) {
81 assert(CvCLONE(protocv));
82 assert(!CvCLONED(protocv));
84 cv_clone_into(protocv,(CV *)TARG);
85 SAVECLEARSV(PAD_SVl(ARGTARG));
92 /* In some cases this function inspects PL_op. If this function is called
93 for new op types, more bool parameters may need to be added in place of
96 When noinit is true, the absence of a gv will cause a retval of undef.
97 This is unrelated to the cv-to-gv assignment case.
101 S_rv2gv(pTHX_ SV *sv, const bool vivify_sv, const bool strict,
104 if (!isGV(sv) || SvFAKE(sv)) SvGETMAGIC(sv);
107 sv = amagic_deref_call(sv, to_gv_amg);
111 if (SvTYPE(sv) == SVt_PVIO) {
112 GV * const gv = MUTABLE_GV(sv_newmortal());
113 gv_init(gv, 0, "__ANONIO__", 10, 0);
114 GvIOp(gv) = MUTABLE_IO(sv);
115 SvREFCNT_inc_void_NN(sv);
118 else if (!isGV_with_GP(sv)) {
119 Perl_die(aTHX_ "Not a GLOB reference");
123 if (!isGV_with_GP(sv)) {
125 /* If this is a 'my' scalar and flag is set then vivify
128 if (vivify_sv && sv != &PL_sv_undef) {
132 Perl_croak_no_modify();
133 gv = MUTABLE_GV(newSV(0));
134 stash = CopSTASH(PL_curcop);
135 if (SvTYPE(stash) != SVt_PVHV) stash = NULL;
136 if (cUNOP->op_targ) {
137 SV * const namesv = PAD_SV(cUNOP->op_targ);
138 gv_init_sv(gv, stash, namesv, 0);
141 gv_init_pv(gv, stash, "__ANONIO__", 0);
143 prepare_SV_for_RV(sv);
144 SvRV_set(sv, MUTABLE_SV(gv));
149 if (PL_op->op_flags & OPf_REF || strict) {
150 Perl_die(aTHX_ PL_no_usym, "a symbol");
152 if (ckWARN(WARN_UNINITIALIZED))
158 if (!(sv = MUTABLE_SV(gv_fetchsv_nomg(
159 sv, GV_ADDMG, SVt_PVGV
168 (SvPOKp(sv) && SvCUR(sv)>32 ? "..." : ""),
172 if ((PL_op->op_private & (OPpLVAL_INTRO|OPpDONT_INIT_GV))
173 == OPpDONT_INIT_GV) {
174 /* We are the target of a coderef assignment. Return
175 the scalar unchanged, and let pp_sasssign deal with
179 sv = MUTABLE_SV(gv_fetchsv_nomg(sv, GV_ADD, SVt_PVGV));
181 /* FAKE globs in the symbol table cause weird bugs (#77810) */
185 if (SvFAKE(sv) && !(PL_op->op_private & OPpALLOW_FAKE)) {
186 SV *newsv = sv_newmortal();
187 sv_setsv_flags(newsv, sv, 0);
199 sv, PL_op->op_private & OPpDEREF,
200 PL_op->op_private & HINT_STRICT_REFS,
201 ((PL_op->op_flags & OPf_SPECIAL) && !(PL_op->op_flags & OPf_MOD))
202 || PL_op->op_type == OP_READLINE
204 if (PL_op->op_private & OPpLVAL_INTRO)
205 save_gp(MUTABLE_GV(sv), !(PL_op->op_flags & OPf_SPECIAL));
210 /* Helper function for pp_rv2sv and pp_rv2av */
212 Perl_softref2xv(pTHX_ SV *const sv, const char *const what,
213 const svtype type, SV ***spp)
217 PERL_ARGS_ASSERT_SOFTREF2XV;
219 if (PL_op->op_private & HINT_STRICT_REFS) {
221 Perl_die(aTHX_ PL_no_symref_sv, sv,
222 (SvPOKp(sv) && SvCUR(sv)>32 ? "..." : ""), what);
224 Perl_die(aTHX_ PL_no_usym, what);
228 PL_op->op_flags & OPf_REF
230 Perl_die(aTHX_ PL_no_usym, what);
231 if (ckWARN(WARN_UNINITIALIZED))
233 if (type != SVt_PV && GIMME_V == G_ARRAY) {
237 **spp = &PL_sv_undef;
240 if ((PL_op->op_flags & OPf_SPECIAL) &&
241 !(PL_op->op_flags & OPf_MOD))
243 if (!(gv = gv_fetchsv_nomg(sv, GV_ADDMG, type)))
245 **spp = &PL_sv_undef;
250 gv = gv_fetchsv_nomg(sv, GV_ADD, type);
263 sv = amagic_deref_call(sv, to_sv_amg);
267 if (SvTYPE(sv) >= SVt_PVAV)
268 DIE(aTHX_ "Not a SCALAR reference");
273 if (!isGV_with_GP(gv)) {
274 gv = Perl_softref2xv(aTHX_ sv, "a SCALAR", SVt_PV, &sp);
280 if (PL_op->op_flags & OPf_MOD) {
281 if (PL_op->op_private & OPpLVAL_INTRO) {
282 if (cUNOP->op_first->op_type == OP_NULL)
283 sv = save_scalar(MUTABLE_GV(TOPs));
285 sv = save_scalar(gv);
287 Perl_croak(aTHX_ "%s", PL_no_localize_ref);
289 else if (PL_op->op_private & OPpDEREF)
290 sv = vivify_ref(sv, PL_op->op_private & OPpDEREF);
292 SPAGAIN; /* in case chasing soft refs reallocated the stack */
300 AV * const av = MUTABLE_AV(TOPs);
301 const I32 lvalue = PL_op->op_flags & OPf_MOD || LVRET;
303 SV ** const svp = Perl_av_arylen_p(aTHX_ MUTABLE_AV(av));
305 *svp = newSV_type(SVt_PVMG);
306 sv_magic(*svp, MUTABLE_SV(av), PERL_MAGIC_arylen, NULL, 0);
310 SETs(sv_2mortal(newSViv(AvFILL(MUTABLE_AV(av)))));
319 if (PL_op->op_flags & OPf_MOD || LVRET) {
320 SV * const ret = sv_2mortal(newSV_type(SVt_PVLV));/* Not TARG RT#67838 */
321 sv_magic(ret, NULL, PERL_MAGIC_pos, NULL, 0);
323 LvTARG(ret) = SvREFCNT_inc_simple(sv);
324 SETs(ret); /* no SvSETMAGIC */
327 const MAGIC * const mg = mg_find_mglob(sv);
328 if (mg && mg->mg_len != -1) {
329 STRLEN i = mg->mg_len;
330 if (PL_op->op_private & OPpTRUEBOOL)
331 SETs(i ? &PL_sv_yes : &PL_sv_zero);
334 if (mg->mg_flags & MGf_BYTES && DO_UTF8(sv))
335 i = sv_pos_b2u_flags(sv, i, SV_GMAGIC|SV_CONST_RETURN);
350 const I32 flags = (PL_op->op_flags & OPf_SPECIAL)
352 : ((PL_op->op_private & (OPpLVAL_INTRO|OPpMAY_RETURN_CONSTANT))
353 == OPpMAY_RETURN_CONSTANT)
356 /* We usually try to add a non-existent subroutine in case of AUTOLOAD. */
357 /* (But not in defined().) */
359 CV *cv = sv_2cv(TOPs, &stash_unused, &gv, flags);
361 else if ((flags == (GV_ADD|GV_NOEXPAND)) && gv && SvROK(gv)) {
362 cv = SvTYPE(SvRV(gv)) == SVt_PVCV
363 ? MUTABLE_CV(SvRV(gv))
367 cv = MUTABLE_CV(&PL_sv_undef);
368 SETs(MUTABLE_SV(cv));
378 SV *ret = &PL_sv_undef;
380 if (SvGMAGICAL(TOPs)) SETs(sv_mortalcopy(TOPs));
381 if (SvPOK(TOPs) && SvCUR(TOPs) >= 7) {
382 const char * s = SvPVX_const(TOPs);
383 if (memBEGINs(s, SvCUR(TOPs), "CORE::")) {
384 const int code = keyword(s + 6, SvCUR(TOPs) - 6, 1);
386 DIE(aTHX_ "Can't find an opnumber for \"%" UTF8f "\"",
387 UTF8fARG(SvFLAGS(TOPs) & SVf_UTF8, SvCUR(TOPs)-6, s+6));
389 SV * const sv = core_prototype(NULL, s + 6, code, NULL);
395 cv = sv_2cv(TOPs, &stash, &gv, 0);
397 ret = newSVpvn_flags(
398 CvPROTO(cv), CvPROTOLEN(cv), SVs_TEMP | SvUTF8(cv)
408 CV *cv = MUTABLE_CV(PAD_SV(PL_op->op_targ));
410 cv = MUTABLE_CV(sv_2mortal(MUTABLE_SV(cv_clone(cv))));
412 PUSHs(MUTABLE_SV(cv));
426 if (GIMME_V != G_ARRAY) {
432 *MARK = &PL_sv_undef;
434 *MARK = refto(*MARK);
438 EXTEND_MORTAL(SP - MARK);
440 *MARK = refto(*MARK);
445 S_refto(pTHX_ SV *sv)
449 PERL_ARGS_ASSERT_REFTO;
451 if (SvTYPE(sv) == SVt_PVLV && LvTYPE(sv) == 'y') {
454 if (!(sv = LvTARG(sv)))
457 SvREFCNT_inc_void_NN(sv);
459 else if (SvTYPE(sv) == SVt_PVAV) {
460 if (!AvREAL((const AV *)sv) && AvREIFY((const AV *)sv))
461 av_reify(MUTABLE_AV(sv));
463 SvREFCNT_inc_void_NN(sv);
465 else if (SvPADTMP(sv)) {
468 else if (UNLIKELY(SvSMAGICAL(sv) && mg_find(sv, PERL_MAGIC_nonelem)))
469 sv_unmagic(SvREFCNT_inc_simple_NN(sv), PERL_MAGIC_nonelem);
472 SvREFCNT_inc_void_NN(sv);
475 sv_upgrade(rv, SVt_IV);
484 SV * const sv = TOPs;
492 /* op is in boolean context? */
493 if ( (PL_op->op_private & OPpTRUEBOOL)
494 || ( (PL_op->op_private & OPpMAYBE_TRUEBOOL)
495 && block_gimme() == G_VOID))
497 /* refs are always true - unless it's to an object blessed into a
498 * class with a false name, i.e. "0". So we have to check for
499 * that remote possibility. The following is is basically an
500 * unrolled SvTRUE(sv_reftype(rv)) */
501 SV * const rv = SvRV(sv);
503 HV *stash = SvSTASH(rv);
504 HEK *hek = HvNAME_HEK(stash);
506 I32 len = HEK_LEN(hek);
507 /* bail out and do it the hard way? */
510 || (len == 1 && HEK_KEY(hek)[0] == '0')
523 sv_ref(TARG, SvRV(sv), TRUE);
539 stash = CopSTASH(PL_curcop);
540 if (SvTYPE(stash) != SVt_PVHV)
541 Perl_croak(aTHX_ "Attempt to bless into a freed package");
544 SV * const ssv = POPs;
548 if (!ssv) goto curstash;
551 if (!SvAMAGIC(ssv)) {
553 Perl_croak(aTHX_ "Attempt to bless into a reference");
555 /* SvAMAGIC is on here, but it only means potentially overloaded,
556 so after stringification: */
557 ptr = SvPV_nomg_const(ssv,len);
558 /* We need to check the flag again: */
559 if (!SvAMAGIC(ssv)) goto frog;
561 else ptr = SvPV_nomg_const(ssv,len);
563 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
564 "Explicit blessing to '' (assuming package main)");
565 stash = gv_stashpvn(ptr, len, GV_ADD|SvUTF8(ssv));
568 (void)sv_bless(TOPs, stash);
578 const char * const elem = SvPV_const(sv, len);
579 GV * const gv = MUTABLE_GV(TOPs);
584 /* elem will always be NUL terminated. */
587 if (memEQs(elem, len, "ARRAY"))
589 tmpRef = MUTABLE_SV(GvAV(gv));
590 if (tmpRef && !AvREAL((const AV *)tmpRef)
591 && AvREIFY((const AV *)tmpRef))
592 av_reify(MUTABLE_AV(tmpRef));
596 if (memEQs(elem, len, "CODE"))
597 tmpRef = MUTABLE_SV(GvCVu(gv));
600 if (memEQs(elem, len, "FILEHANDLE")) {
601 tmpRef = MUTABLE_SV(GvIOp(gv));
604 if (memEQs(elem, len, "FORMAT"))
605 tmpRef = MUTABLE_SV(GvFORM(gv));
608 if (memEQs(elem, len, "GLOB"))
609 tmpRef = MUTABLE_SV(gv);
612 if (memEQs(elem, len, "HASH"))
613 tmpRef = MUTABLE_SV(GvHV(gv));
616 if (memEQs(elem, len, "IO"))
617 tmpRef = MUTABLE_SV(GvIOp(gv));
620 if (memEQs(elem, len, "NAME"))
621 sv = newSVhek(GvNAME_HEK(gv));
624 if (memEQs(elem, len, "PACKAGE")) {
625 const HV * const stash = GvSTASH(gv);
626 const HEK * const hek = stash ? HvNAME_HEK(stash) : NULL;
627 sv = hek ? newSVhek(hek) : newSVpvs("__ANON__");
631 if (memEQs(elem, len, "SCALAR"))
646 /* Pattern matching */
654 if (len == 0 || len > I32_MAX || !SvPOK(sv) || SvUTF8(sv) || SvVALID(sv)) {
655 /* Historically, study was skipped in these cases. */
660 /* Make study a no-op. It's no longer useful and its existence
661 complicates matters elsewhere. */
667 /* also used for: pp_transr() */
674 if (PL_op->op_flags & OPf_STACKED)
679 sv = PAD_SV(ARGTARG);
684 if(PL_op->op_type == OP_TRANSR) {
686 const char * const pv = SvPV(sv,len);
687 SV * const newsv = newSVpvn_flags(pv, len, SVs_TEMP|SvUTF8(sv));
692 Size_t i = do_trans(sv);
698 /* Lvalue operators. */
701 S_do_chomp(pTHX_ SV *retval, SV *sv, bool chomping)
707 PERL_ARGS_ASSERT_DO_CHOMP;
709 if (chomping && (RsSNARF(PL_rs) || RsRECORD(PL_rs)))
711 if (SvTYPE(sv) == SVt_PVAV) {
713 AV *const av = MUTABLE_AV(sv);
714 const I32 max = AvFILL(av);
716 for (i = 0; i <= max; i++) {
717 sv = MUTABLE_SV(av_fetch(av, i, FALSE));
718 if (sv && ((sv = *(SV**)sv), sv != &PL_sv_undef))
719 count += do_chomp(retval, sv, chomping);
723 else if (SvTYPE(sv) == SVt_PVHV) {
724 HV* const hv = MUTABLE_HV(sv);
726 (void)hv_iterinit(hv);
727 while ((entry = hv_iternext(hv)))
728 count += do_chomp(retval, hv_iterval(hv,entry), chomping);
731 else if (SvREADONLY(sv)) {
732 Perl_croak_no_modify();
738 char *temp_buffer = NULL;
743 goto nope_free_nothing;
745 while (len && s[-1] == '\n') {
752 STRLEN rslen, rs_charlen;
753 const char *rsptr = SvPV_const(PL_rs, rslen);
755 rs_charlen = SvUTF8(PL_rs)
759 if (SvUTF8(PL_rs) != SvUTF8(sv)) {
760 /* Assumption is that rs is shorter than the scalar. */
762 /* RS is utf8, scalar is 8 bit. */
764 temp_buffer = (char*)bytes_from_utf8((U8*)rsptr,
767 /* Cannot downgrade, therefore cannot possibly match.
768 At this point, temp_buffer is not alloced, and
769 is the buffer inside PL_rs, so dont free it.
771 assert (temp_buffer == rsptr);
777 /* RS is 8 bit, scalar is utf8. */
778 temp_buffer = (char*)bytes_to_utf8((U8*)rsptr, &rslen);
792 if (memNE(s, rsptr, rslen))
797 SvPV_force_nomg_nolen(sv);
804 Safefree(temp_buffer);
806 SvREFCNT_dec(svrecode);
810 if (len && (!SvPOK(sv) || SvIsCOW(sv)))
811 s = SvPV_force_nomg(sv, len);
814 char * const send = s + len;
815 char * const start = s;
817 while (s > start && UTF8_IS_CONTINUATION(*s))
819 if (is_utf8_string((U8*)s, send - s)) {
820 sv_setpvn(retval, s, send - s);
822 SvCUR_set(sv, s - start);
832 sv_setpvn(retval, s, 1);
846 /* also used for: pp_schomp() */
851 const bool chomping = PL_op->op_type == OP_SCHOMP;
853 const size_t count = do_chomp(TARG, TOPs, chomping);
855 sv_setiv(TARG, count);
861 /* also used for: pp_chomp() */
865 dSP; dMARK; dTARGET; dORIGMARK;
866 const bool chomping = PL_op->op_type == OP_CHOMP;
870 count += do_chomp(TARG, *++MARK, chomping);
872 sv_setiv(TARG, count);
883 if (!PL_op->op_private) {
895 if (SvTHINKFIRST(sv))
896 sv_force_normal_flags(sv, SV_COW_DROP_PV|SV_IMMEDIATE_UNREF);
898 switch (SvTYPE(sv)) {
902 av_undef(MUTABLE_AV(sv));
905 hv_undef(MUTABLE_HV(sv));
908 if (cv_const_sv((const CV *)sv))
909 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
910 "Constant subroutine %" SVf " undefined",
911 SVfARG(CvANON((const CV *)sv)
912 ? newSVpvs_flags("(anonymous)", SVs_TEMP)
913 : sv_2mortal(newSVhek(
915 ? CvNAME_HEK((CV *)sv)
916 : GvENAME_HEK(CvGV((const CV *)sv))
921 /* let user-undef'd sub keep its identity */
922 cv_undef_flags(MUTABLE_CV(sv), CV_UNDEF_KEEP_NAME);
925 assert(isGV_with_GP(sv));
931 /* undef *Pkg::meth_name ... */
933 = GvCVu((const GV *)sv) && (stash = GvSTASH((const GV *)sv))
934 && HvENAME_get(stash);
936 if((stash = GvHV((const GV *)sv))) {
937 if(HvENAME_get(stash))
938 SvREFCNT_inc_simple_void_NN(sv_2mortal((SV *)stash));
942 SvREFCNT_inc_simple_void_NN(sv_2mortal(sv));
943 gp_free(MUTABLE_GV(sv));
945 GvGP_set(sv, gp_ref(gp));
946 #ifndef PERL_DONT_CREATE_GVSV
949 GvLINE(sv) = CopLINE(PL_curcop);
950 GvEGV(sv) = MUTABLE_GV(sv);
954 mro_package_moved(NULL, stash, (const GV *)sv, 0);
956 /* undef *Foo::ISA */
957 if( strEQ(GvNAME((const GV *)sv), "ISA")
958 && (stash = GvSTASH((const GV *)sv))
959 && (method_changed || HvENAME(stash)) )
960 mro_isa_changed_in(stash);
961 else if(method_changed)
962 mro_method_changed_in(
963 GvSTASH((const GV *)sv)
969 if (SvTYPE(sv) >= SVt_PV && SvPVX_const(sv) && SvLEN(sv)) {
983 /* common "slow" code for pp_postinc and pp_postdec */
986 S_postincdec_common(pTHX_ SV *sv, SV *targ)
990 PL_op->op_type == OP_POSTINC || PL_op->op_type == OP_I_POSTINC;
993 TARG = sv_newmortal();
1000 /* special case for undef: see thread at 2003-03/msg00536.html in archive */
1001 if (inc && !SvOK(TARG))
1008 /* also used for: pp_i_postinc() */
1015 /* special-case sv being a simple integer */
1016 if (LIKELY(((sv->sv_flags &
1017 (SVf_THINKFIRST|SVs_GMG|SVf_IVisUV|
1018 SVf_IOK|SVf_NOK|SVf_POK|SVp_NOK|SVp_POK|SVf_ROK))
1020 && SvIVX(sv) != IV_MAX)
1023 SvIV_set(sv, iv + 1);
1024 TARGi(iv, 0); /* arg not GMG, so can't be tainted */
1029 return S_postincdec_common(aTHX_ sv, TARG);
1033 /* also used for: pp_i_postdec() */
1040 /* special-case sv being a simple integer */
1041 if (LIKELY(((sv->sv_flags &
1042 (SVf_THINKFIRST|SVs_GMG|SVf_IVisUV|
1043 SVf_IOK|SVf_NOK|SVf_POK|SVp_NOK|SVp_POK|SVf_ROK))
1045 && SvIVX(sv) != IV_MIN)
1048 SvIV_set(sv, iv - 1);
1049 TARGi(iv, 0); /* arg not GMG, so can't be tainted */
1054 return S_postincdec_common(aTHX_ sv, TARG);
1058 /* Ordinary operators. */
1062 dSP; dATARGET; SV *svl, *svr;
1063 #ifdef PERL_PRESERVE_IVUV
1066 tryAMAGICbin_MG(pow_amg, AMGf_assign|AMGf_numeric);
1069 #ifdef PERL_PRESERVE_IVUV
1070 /* For integer to integer power, we do the calculation by hand wherever
1071 we're sure it is safe; otherwise we call pow() and try to convert to
1072 integer afterwards. */
1073 if (SvIV_please_nomg(svr) && SvIV_please_nomg(svl)) {
1081 const IV iv = SvIVX(svr);
1085 goto float_it; /* Can't do negative powers this way. */
1089 baseuok = SvUOK(svl);
1091 baseuv = SvUVX(svl);
1093 const IV iv = SvIVX(svl);
1096 baseuok = TRUE; /* effectively it's a UV now */
1098 baseuv = -iv; /* abs, baseuok == false records sign */
1101 /* now we have integer ** positive integer. */
1104 /* foo & (foo - 1) is zero only for a power of 2. */
1105 if (!(baseuv & (baseuv - 1))) {
1106 /* We are raising power-of-2 to a positive integer.
1107 The logic here will work for any base (even non-integer
1108 bases) but it can be less accurate than
1109 pow (base,power) or exp (power * log (base)) when the
1110 intermediate values start to spill out of the mantissa.
1111 With powers of 2 we know this can't happen.
1112 And powers of 2 are the favourite thing for perl
1113 programmers to notice ** not doing what they mean. */
1115 NV base = baseuok ? baseuv : -(NV)baseuv;
1120 while (power >>= 1) {
1128 SvIV_please_nomg(svr);
1131 unsigned int highbit = 8 * sizeof(UV);
1132 unsigned int diff = 8 * sizeof(UV);
1133 while (diff >>= 1) {
1135 if (baseuv >> highbit) {
1139 /* we now have baseuv < 2 ** highbit */
1140 if (power * highbit <= 8 * sizeof(UV)) {
1141 /* result will definitely fit in UV, so use UV math
1142 on same algorithm as above */
1145 const bool odd_power = cBOOL(power & 1);
1149 while (power >>= 1) {
1156 if (baseuok || !odd_power)
1157 /* answer is positive */
1159 else if (result <= (UV)IV_MAX)
1160 /* answer negative, fits in IV */
1161 SETi( -(IV)result );
1162 else if (result == (UV)IV_MIN)
1163 /* 2's complement assumption: special case IV_MIN */
1166 /* answer negative, doesn't fit */
1167 SETn( -(NV)result );
1175 NV right = SvNV_nomg(svr);
1176 NV left = SvNV_nomg(svl);
1179 #if defined(USE_LONG_DOUBLE) && defined(HAS_AIX_POWL_NEG_BASE_BUG)
1181 We are building perl with long double support and are on an AIX OS
1182 afflicted with a powl() function that wrongly returns NaNQ for any
1183 negative base. This was reported to IBM as PMR #23047-379 on
1184 03/06/2006. The problem exists in at least the following versions
1185 of AIX and the libm fileset, and no doubt others as well:
1187 AIX 4.3.3-ML10 bos.adt.libm 4.3.3.50
1188 AIX 5.1.0-ML04 bos.adt.libm 5.1.0.29
1189 AIX 5.2.0 bos.adt.libm 5.2.0.85
1191 So, until IBM fixes powl(), we provide the following workaround to
1192 handle the problem ourselves. Our logic is as follows: for
1193 negative bases (left), we use fmod(right, 2) to check if the
1194 exponent is an odd or even integer:
1196 - if odd, powl(left, right) == -powl(-left, right)
1197 - if even, powl(left, right) == powl(-left, right)
1199 If the exponent is not an integer, the result is rightly NaNQ, so
1200 we just return that (as NV_NAN).
1204 NV mod2 = Perl_fmod( right, 2.0 );
1205 if (mod2 == 1.0 || mod2 == -1.0) { /* odd integer */
1206 SETn( -Perl_pow( -left, right) );
1207 } else if (mod2 == 0.0) { /* even integer */
1208 SETn( Perl_pow( -left, right) );
1209 } else { /* fractional power */
1213 SETn( Perl_pow( left, right) );
1216 SETn( Perl_pow( left, right) );
1217 #endif /* HAS_AIX_POWL_NEG_BASE_BUG */
1219 #ifdef PERL_PRESERVE_IVUV
1221 SvIV_please_nomg(svr);
1229 dSP; dATARGET; SV *svl, *svr;
1230 tryAMAGICbin_MG(mult_amg, AMGf_assign|AMGf_numeric);
1234 #ifdef PERL_PRESERVE_IVUV
1236 /* special-case some simple common cases */
1237 if (!((svl->sv_flags|svr->sv_flags) & (SVf_IVisUV|SVs_GMG))) {
1239 U32 flags = (svl->sv_flags & svr->sv_flags);
1240 if (flags & SVf_IOK) {
1241 /* both args are simple IVs */
1246 topl = ((UV)il) >> (UVSIZE * 4 - 1);
1247 topr = ((UV)ir) >> (UVSIZE * 4 - 1);
1249 /* if both are in a range that can't under/overflow, do a
1250 * simple integer multiply: if the top halves(*) of both numbers
1251 * are 00...00 or 11...11, then it's safe.
1252 * (*) for 32-bits, the "top half" is the top 17 bits,
1253 * for 64-bits, its 33 bits */
1255 ((topl+1) | (topr+1))
1256 & ( (((UV)1) << (UVSIZE * 4 + 1)) - 2) /* 11..110 */
1259 TARGi(il * ir, 0); /* args not GMG, so can't be tainted */
1265 else if (flags & SVf_NOK) {
1266 /* both args are NVs */
1271 if (lossless_NV_to_IV(nl, &il) && lossless_NV_to_IV(nr, &ir)) {
1272 /* nothing was lost by converting to IVs */
1277 # if defined(__sgi) && defined(USE_LONG_DOUBLE) && LONG_DOUBLEKIND == LONG_DOUBLE_IS_DOUBLEDOUBLE_128_BIT_BE_BE && NVSIZE == 16
1278 if (Perl_isinf(result)) {
1279 Zero((U8*)&result + 8, 8, U8);
1282 TARGn(result, 0); /* args not GMG, so can't be tainted */
1290 if (SvIV_please_nomg(svr)) {
1291 /* Unless the left argument is integer in range we are going to have to
1292 use NV maths. Hence only attempt to coerce the right argument if
1293 we know the left is integer. */
1294 /* Left operand is defined, so is it IV? */
1295 if (SvIV_please_nomg(svl)) {
1296 bool auvok = SvUOK(svl);
1297 bool buvok = SvUOK(svr);
1298 const UV topmask = (~ (UV)0) << (4 * sizeof (UV));
1299 const UV botmask = ~((~ (UV)0) << (4 * sizeof (UV)));
1308 const IV aiv = SvIVX(svl);
1311 auvok = TRUE; /* effectively it's a UV now */
1313 /* abs, auvok == false records sign; Using 0- here and
1314 * later to silence bogus warning from MS VC */
1315 alow = (UV) (0 - (UV) aiv);
1321 const IV biv = SvIVX(svr);
1324 buvok = TRUE; /* effectively it's a UV now */
1326 /* abs, buvok == false records sign */
1327 blow = (UV) (0 - (UV) biv);
1331 /* If this does sign extension on unsigned it's time for plan B */
1332 ahigh = alow >> (4 * sizeof (UV));
1334 bhigh = blow >> (4 * sizeof (UV));
1336 if (ahigh && bhigh) {
1338 /* eg 32 bit is at least 0x10000 * 0x10000 == 0x100000000
1339 which is overflow. Drop to NVs below. */
1340 } else if (!ahigh && !bhigh) {
1341 /* eg 32 bit is at most 0xFFFF * 0xFFFF == 0xFFFE0001
1342 so the unsigned multiply cannot overflow. */
1343 const UV product = alow * blow;
1344 if (auvok == buvok) {
1345 /* -ve * -ve or +ve * +ve gives a +ve result. */
1349 } else if (product <= (UV)IV_MIN) {
1350 /* 2s complement assumption that (UV)-IV_MIN is correct. */
1351 /* -ve result, which could overflow an IV */
1353 /* can't negate IV_MIN, but there are aren't two
1354 * integers such that !ahigh && !bhigh, where the
1355 * product equals 0x800....000 */
1356 assert(product != (UV)IV_MIN);
1357 SETi( -(IV)product );
1359 } /* else drop to NVs below. */
1361 /* One operand is large, 1 small */
1364 /* swap the operands */
1366 bhigh = blow; /* bhigh now the temp var for the swap */
1370 /* now, ((ahigh * blow) << half_UV_len) + (alow * blow)
1371 multiplies can't overflow. shift can, add can, -ve can. */
1372 product_middle = ahigh * blow;
1373 if (!(product_middle & topmask)) {
1374 /* OK, (ahigh * blow) won't lose bits when we shift it. */
1376 product_middle <<= (4 * sizeof (UV));
1377 product_low = alow * blow;
1379 /* as for pp_add, UV + something mustn't get smaller.
1380 IIRC ANSI mandates this wrapping *behaviour* for
1381 unsigned whatever the actual representation*/
1382 product_low += product_middle;
1383 if (product_low >= product_middle) {
1384 /* didn't overflow */
1385 if (auvok == buvok) {
1386 /* -ve * -ve or +ve * +ve gives a +ve result. */
1388 SETu( product_low );
1390 } else if (product_low <= (UV)IV_MIN) {
1391 /* 2s complement assumption again */
1392 /* -ve result, which could overflow an IV */
1394 SETi(product_low == (UV)IV_MIN
1395 ? IV_MIN : -(IV)product_low);
1397 } /* else drop to NVs below. */
1399 } /* product_middle too large */
1400 } /* ahigh && bhigh */
1405 NV right = SvNV_nomg(svr);
1406 NV left = SvNV_nomg(svl);
1407 NV result = left * right;
1410 #if defined(__sgi) && defined(USE_LONG_DOUBLE) && LONG_DOUBLEKIND == LONG_DOUBLE_IS_DOUBLEDOUBLE_128_BIT_BE_BE && NVSIZE == 16
1411 if (Perl_isinf(result)) {
1412 Zero((U8*)&result + 8, 8, U8);
1422 dSP; dATARGET; SV *svl, *svr;
1423 tryAMAGICbin_MG(div_amg, AMGf_assign|AMGf_numeric);
1426 /* Only try to do UV divide first
1427 if ((SLOPPYDIVIDE is true) or
1428 (PERL_PRESERVE_IVUV is true and one or both SV is a UV too large
1430 The assumption is that it is better to use floating point divide
1431 whenever possible, only doing integer divide first if we can't be sure.
1432 If NV_PRESERVES_UV is true then we know at compile time that no UV
1433 can be too large to preserve, so don't need to compile the code to
1434 test the size of UVs. */
1436 #if defined(SLOPPYDIVIDE) || (defined(PERL_PRESERVE_IVUV) && !defined(NV_PRESERVES_UV))
1437 # define PERL_TRY_UV_DIVIDE
1438 /* ensure that 20./5. == 4. */
1441 #ifdef PERL_TRY_UV_DIVIDE
1442 if (SvIV_please_nomg(svr) && SvIV_please_nomg(svl)) {
1443 bool left_non_neg = SvUOK(svl);
1444 bool right_non_neg = SvUOK(svr);
1448 if (right_non_neg) {
1452 const IV biv = SvIVX(svr);
1455 right_non_neg = TRUE; /* effectively it's a UV now */
1461 /* historically undef()/0 gives a "Use of uninitialized value"
1462 warning before dieing, hence this test goes here.
1463 If it were immediately before the second SvIV_please, then
1464 DIE() would be invoked before left was even inspected, so
1465 no inspection would give no warning. */
1467 DIE(aTHX_ "Illegal division by zero");
1473 const IV aiv = SvIVX(svl);
1476 left_non_neg = TRUE; /* effectively it's a UV now */
1485 /* For sloppy divide we always attempt integer division. */
1487 /* Otherwise we only attempt it if either or both operands
1488 would not be preserved by an NV. If both fit in NVs
1489 we fall through to the NV divide code below. However,
1490 as left >= right to ensure integer result here, we know that
1491 we can skip the test on the right operand - right big
1492 enough not to be preserved can't get here unless left is
1495 && (left > ((UV)1 << NV_PRESERVES_UV_BITS))
1498 /* Integer division can't overflow, but it can be imprecise. */
1500 /* Modern compilers optimize division followed by
1501 * modulo into a single div instruction */
1502 const UV result = left / right;
1503 if (left % right == 0) {
1504 SP--; /* result is valid */
1505 if (left_non_neg == right_non_neg) {
1506 /* signs identical, result is positive. */
1510 /* 2s complement assumption */
1511 if (result <= (UV)IV_MIN)
1512 SETi(result == (UV)IV_MIN ? IV_MIN : -(IV)result);
1514 /* It's exact but too negative for IV. */
1515 SETn( -(NV)result );
1518 } /* tried integer divide but it was not an integer result */
1519 } /* else (PERL_ABS(result) < 1.0) or (both UVs in range for NV) */
1520 } /* one operand wasn't SvIOK */
1521 #endif /* PERL_TRY_UV_DIVIDE */
1523 NV right = SvNV_nomg(svr);
1524 NV left = SvNV_nomg(svl);
1525 (void)POPs;(void)POPs;
1526 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
1527 if (! Perl_isnan(right) && right == 0.0)
1531 DIE(aTHX_ "Illegal division by zero");
1532 PUSHn( left / right );
1540 tryAMAGICbin_MG(modulo_amg, AMGf_assign|AMGf_numeric);
1544 bool left_neg = FALSE;
1545 bool right_neg = FALSE;
1546 bool use_double = FALSE;
1547 bool dright_valid = FALSE;
1550 SV * const svr = TOPs;
1551 SV * const svl = TOPm1s;
1552 if (SvIV_please_nomg(svr)) {
1553 right_neg = !SvUOK(svr);
1557 const IV biv = SvIVX(svr);
1560 right_neg = FALSE; /* effectively it's a UV now */
1562 right = (UV) (0 - (UV) biv);
1567 dright = SvNV_nomg(svr);
1568 right_neg = dright < 0;
1571 if (dright < UV_MAX_P1) {
1572 right = U_V(dright);
1573 dright_valid = TRUE; /* In case we need to use double below. */
1579 /* At this point use_double is only true if right is out of range for
1580 a UV. In range NV has been rounded down to nearest UV and
1581 use_double false. */
1582 if (!use_double && SvIV_please_nomg(svl)) {
1583 left_neg = !SvUOK(svl);
1587 const IV aiv = SvIVX(svl);
1590 left_neg = FALSE; /* effectively it's a UV now */
1592 left = (UV) (0 - (UV) aiv);
1597 dleft = SvNV_nomg(svl);
1598 left_neg = dleft < 0;
1602 /* This should be exactly the 5.6 behaviour - if left and right are
1603 both in range for UV then use U_V() rather than floor. */
1605 if (dleft < UV_MAX_P1) {
1606 /* right was in range, so is dleft, so use UVs not double.
1610 /* left is out of range for UV, right was in range, so promote
1611 right (back) to double. */
1613 /* The +0.5 is used in 5.6 even though it is not strictly
1614 consistent with the implicit +0 floor in the U_V()
1615 inside the #if 1. */
1616 dleft = Perl_floor(dleft + 0.5);
1619 dright = Perl_floor(dright + 0.5);
1630 DIE(aTHX_ "Illegal modulus zero");
1632 dans = Perl_fmod(dleft, dright);
1633 if ((left_neg != right_neg) && dans)
1634 dans = dright - dans;
1637 sv_setnv(TARG, dans);
1643 DIE(aTHX_ "Illegal modulus zero");
1646 if ((left_neg != right_neg) && ans)
1649 /* XXX may warn: unary minus operator applied to unsigned type */
1650 /* could change -foo to be (~foo)+1 instead */
1651 if (ans <= ~((UV)IV_MAX)+1)
1652 sv_setiv(TARG, ~ans+1);
1654 sv_setnv(TARG, -(NV)ans);
1657 sv_setuv(TARG, ans);
1669 bool infnan = FALSE;
1670 const U8 gimme = GIMME_V;
1672 if (gimme == G_ARRAY && PL_op->op_private & OPpREPEAT_DOLIST) {
1673 /* TODO: think of some way of doing list-repeat overloading ??? */
1678 if (UNLIKELY(PL_op->op_private & OPpREPEAT_DOLIST)) {
1679 /* The parser saw this as a list repeat, and there
1680 are probably several items on the stack. But we're
1681 in scalar/void context, and there's no pp_list to save us
1682 now. So drop the rest of the items -- robin@kitsite.com
1685 if (MARK + 1 < SP) {
1691 ASSUME(MARK + 1 == SP);
1694 MARK[1] = &PL_sv_undef;
1698 tryAMAGICbin_MG(repeat_amg, AMGf_assign);
1704 const UV uv = SvUV_nomg(sv);
1706 count = IV_MAX; /* The best we can do? */
1710 count = SvIV_nomg(sv);
1713 else if (SvNOKp(sv)) {
1714 const NV nv = SvNV_nomg(sv);
1715 infnan = Perl_isinfnan(nv);
1716 if (UNLIKELY(infnan)) {
1720 count = -1; /* An arbitrary negative integer */
1726 count = SvIV_nomg(sv);
1729 Perl_ck_warner(aTHX_ packWARN(WARN_NUMERIC),
1730 "Non-finite repeat count does nothing");
1731 } else if (count < 0) {
1733 Perl_ck_warner(aTHX_ packWARN(WARN_NUMERIC),
1734 "Negative repeat count does nothing");
1737 if (gimme == G_ARRAY && PL_op->op_private & OPpREPEAT_DOLIST) {
1739 const SSize_t items = SP - MARK;
1740 const U8 mod = PL_op->op_flags & OPf_MOD;
1745 if ( items > SSize_t_MAX / count /* max would overflow */
1746 /* repeatcpy would overflow */
1747 || items > I32_MAX / (I32)sizeof(SV *)
1749 Perl_croak(aTHX_ "%s","Out of memory during list extend");
1750 max = items * count;
1755 if (mod && SvPADTMP(*SP)) {
1756 *SP = sv_mortalcopy(*SP);
1763 repeatcpy((char*)(MARK + items), (char*)MARK,
1764 items * sizeof(const SV *), count - 1);
1767 else if (count <= 0)
1770 else { /* Note: mark already snarfed by pp_list */
1771 SV * const tmpstr = POPs;
1776 sv_setsv_nomg(TARG, tmpstr);
1777 SvPV_force_nomg(TARG, len);
1778 isutf = DO_UTF8(TARG);
1785 if ( len > (MEM_SIZE_MAX-1) / (UV)count /* max would overflow */
1786 || len > (U32)I32_MAX /* repeatcpy would overflow */
1788 Perl_croak(aTHX_ "%s",
1789 "Out of memory during string extend");
1790 max = (UV)count * len + 1;
1793 repeatcpy(SvPVX(TARG) + len, SvPVX(TARG), len, count - 1);
1794 SvCUR_set(TARG, SvCUR(TARG) * count);
1796 *SvEND(TARG) = '\0';
1799 (void)SvPOK_only_UTF8(TARG);
1801 (void)SvPOK_only(TARG);
1810 dSP; dATARGET; bool useleft; SV *svl, *svr;
1811 tryAMAGICbin_MG(subtr_amg, AMGf_assign|AMGf_numeric);
1815 #ifdef PERL_PRESERVE_IVUV
1817 /* special-case some simple common cases */
1818 if (!((svl->sv_flags|svr->sv_flags) & (SVf_IVisUV|SVs_GMG))) {
1820 U32 flags = (svl->sv_flags & svr->sv_flags);
1821 if (flags & SVf_IOK) {
1822 /* both args are simple IVs */
1827 topl = ((UV)il) >> (UVSIZE * 8 - 2);
1828 topr = ((UV)ir) >> (UVSIZE * 8 - 2);
1830 /* if both are in a range that can't under/overflow, do a
1831 * simple integer subtract: if the top of both numbers
1832 * are 00 or 11, then it's safe */
1833 if (!( ((topl+1) | (topr+1)) & 2)) {
1835 TARGi(il - ir, 0); /* args not GMG, so can't be tainted */
1841 else if (flags & SVf_NOK) {
1842 /* both args are NVs */
1846 if (lossless_NV_to_IV(nl, &il) && lossless_NV_to_IV(nr, &ir)) {
1847 /* nothing was lost by converting to IVs */
1851 TARGn(nl - nr, 0); /* args not GMG, so can't be tainted */
1859 useleft = USE_LEFT(svl);
1860 /* See comments in pp_add (in pp_hot.c) about Overflow, and how
1861 "bad things" happen if you rely on signed integers wrapping. */
1862 if (SvIV_please_nomg(svr)) {
1863 /* Unless the left argument is integer in range we are going to have to
1864 use NV maths. Hence only attempt to coerce the right argument if
1865 we know the left is integer. */
1872 a_valid = auvok = 1;
1873 /* left operand is undef, treat as zero. */
1875 /* Left operand is defined, so is it IV? */
1876 if (SvIV_please_nomg(svl)) {
1877 if ((auvok = SvUOK(svl)))
1880 const IV aiv = SvIVX(svl);
1883 auvok = 1; /* Now acting as a sign flag. */
1885 auv = (UV) (0 - (UV) aiv);
1892 bool result_good = 0;
1895 bool buvok = SvUOK(svr);
1900 const IV biv = SvIVX(svr);
1905 buv = (UV) (0 - (UV) biv);
1907 /* ?uvok if value is >= 0. basically, flagged as UV if it's +ve,
1908 else "IV" now, independent of how it came in.
1909 if a, b represents positive, A, B negative, a maps to -A etc
1914 all UV maths. negate result if A negative.
1915 subtract if signs same, add if signs differ. */
1917 if (auvok ^ buvok) {
1926 /* Must get smaller */
1931 if (result <= buv) {
1932 /* result really should be -(auv-buv). as its negation
1933 of true value, need to swap our result flag */
1945 if (result <= (UV)IV_MIN)
1946 SETi(result == (UV)IV_MIN
1947 ? IV_MIN : -(IV)result);
1949 /* result valid, but out of range for IV. */
1950 SETn( -(NV)result );
1954 } /* Overflow, drop through to NVs. */
1958 useleft = USE_LEFT(svl);
1961 NV value = SvNV_nomg(svr);
1965 /* left operand is undef, treat as zero - value */
1969 SETn( SvNV_nomg(svl) - value );
1974 #define IV_BITS (IVSIZE * 8)
1976 static UV S_uv_shift(UV uv, int shift, bool left)
1982 if (UNLIKELY(shift >= IV_BITS)) {
1985 return left ? uv << shift : uv >> shift;
1988 static IV S_iv_shift(IV iv, int shift, bool left)
1995 if (UNLIKELY(shift >= IV_BITS)) {
1996 return iv < 0 && !left ? -1 : 0;
1999 /* For left shifts, perl 5 has chosen to treat the value as unsigned for
2000 * the * purposes of shifting, then cast back to signed. This is very
2001 * different from Raku:
2003 * $ raku -e 'say -2 +< 5'
2006 * $ ./perl -le 'print -2 << 5'
2007 * 18446744073709551552
2010 if (iv == IV_MIN) { /* Casting this to a UV is undefined behavior */
2013 return (IV) (((UV) iv) << shift);
2016 /* Here is right shift */
2020 #define UV_LEFT_SHIFT(uv, shift) S_uv_shift(uv, shift, TRUE)
2021 #define UV_RIGHT_SHIFT(uv, shift) S_uv_shift(uv, shift, FALSE)
2022 #define IV_LEFT_SHIFT(iv, shift) S_iv_shift(iv, shift, TRUE)
2023 #define IV_RIGHT_SHIFT(iv, shift) S_iv_shift(iv, shift, FALSE)
2027 dSP; dATARGET; SV *svl, *svr;
2028 tryAMAGICbin_MG(lshift_amg, AMGf_assign|AMGf_numeric);
2032 const IV shift = SvIV_nomg(svr);
2033 if (PL_op->op_private & HINT_INTEGER) {
2034 SETi(IV_LEFT_SHIFT(SvIV_nomg(svl), shift));
2037 SETu(UV_LEFT_SHIFT(SvUV_nomg(svl), shift));
2045 dSP; dATARGET; SV *svl, *svr;
2046 tryAMAGICbin_MG(rshift_amg, AMGf_assign|AMGf_numeric);
2050 const IV shift = SvIV_nomg(svr);
2051 if (PL_op->op_private & HINT_INTEGER) {
2052 SETi(IV_RIGHT_SHIFT(SvIV_nomg(svl), shift));
2055 SETu(UV_RIGHT_SHIFT(SvUV_nomg(svl), shift));
2065 U32 flags_and, flags_or;
2067 tryAMAGICbin_MG(lt_amg, AMGf_numeric);
2070 flags_and = SvFLAGS(left) & SvFLAGS(right);
2071 flags_or = SvFLAGS(left) | SvFLAGS(right);
2074 ( (flags_and & SVf_IOK) && ((flags_or & SVf_IVisUV) ==0 ) )
2075 ? (SvIVX(left) < SvIVX(right))
2076 : (flags_and & SVf_NOK)
2077 ? (SvNVX(left) < SvNVX(right))
2078 : (do_ncmp(left, right) == -1)
2087 U32 flags_and, flags_or;
2089 tryAMAGICbin_MG(gt_amg, AMGf_numeric);
2092 flags_and = SvFLAGS(left) & SvFLAGS(right);
2093 flags_or = SvFLAGS(left) | SvFLAGS(right);
2096 ( (flags_and & SVf_IOK) && ((flags_or & SVf_IVisUV) ==0 ) )
2097 ? (SvIVX(left) > SvIVX(right))
2098 : (flags_and & SVf_NOK)
2099 ? (SvNVX(left) > SvNVX(right))
2100 : (do_ncmp(left, right) == 1)
2109 U32 flags_and, flags_or;
2111 tryAMAGICbin_MG(le_amg, AMGf_numeric);
2114 flags_and = SvFLAGS(left) & SvFLAGS(right);
2115 flags_or = SvFLAGS(left) | SvFLAGS(right);
2118 ( (flags_and & SVf_IOK) && ((flags_or & SVf_IVisUV) ==0 ) )
2119 ? (SvIVX(left) <= SvIVX(right))
2120 : (flags_and & SVf_NOK)
2121 ? (SvNVX(left) <= SvNVX(right))
2122 : (do_ncmp(left, right) <= 0)
2131 U32 flags_and, flags_or;
2133 tryAMAGICbin_MG(ge_amg, AMGf_numeric);
2136 flags_and = SvFLAGS(left) & SvFLAGS(right);
2137 flags_or = SvFLAGS(left) | SvFLAGS(right);
2140 ( (flags_and & SVf_IOK) && ((flags_or & SVf_IVisUV) ==0 ) )
2141 ? (SvIVX(left) >= SvIVX(right))
2142 : (flags_and & SVf_NOK)
2143 ? (SvNVX(left) >= SvNVX(right))
2144 : ( (do_ncmp(left, right) & 2) == 0)
2153 U32 flags_and, flags_or;
2155 tryAMAGICbin_MG(ne_amg, AMGf_numeric);
2158 flags_and = SvFLAGS(left) & SvFLAGS(right);
2159 flags_or = SvFLAGS(left) | SvFLAGS(right);
2162 ( (flags_and & SVf_IOK) && ((flags_or & SVf_IVisUV) ==0 ) )
2163 ? (SvIVX(left) != SvIVX(right))
2164 : (flags_and & SVf_NOK)
2165 ? (SvNVX(left) != SvNVX(right))
2166 : (do_ncmp(left, right) != 0)
2171 /* compare left and right SVs. Returns:
2175 * 2: left or right was a NaN
2178 Perl_do_ncmp(pTHX_ SV* const left, SV * const right)
2180 PERL_ARGS_ASSERT_DO_NCMP;
2181 #ifdef PERL_PRESERVE_IVUV
2182 /* Fortunately it seems NaN isn't IOK */
2183 if (SvIV_please_nomg(right) && SvIV_please_nomg(left)) {
2185 const IV leftiv = SvIVX(left);
2186 if (!SvUOK(right)) {
2187 /* ## IV <=> IV ## */
2188 const IV rightiv = SvIVX(right);
2189 return (leftiv > rightiv) - (leftiv < rightiv);
2191 /* ## IV <=> UV ## */
2193 /* As (b) is a UV, it's >=0, so it must be < */
2196 const UV rightuv = SvUVX(right);
2197 return ((UV)leftiv > rightuv) - ((UV)leftiv < rightuv);
2202 /* ## UV <=> UV ## */
2203 const UV leftuv = SvUVX(left);
2204 const UV rightuv = SvUVX(right);
2205 return (leftuv > rightuv) - (leftuv < rightuv);
2207 /* ## UV <=> IV ## */
2209 const IV rightiv = SvIVX(right);
2211 /* As (a) is a UV, it's >=0, so it cannot be < */
2214 const UV leftuv = SvUVX(left);
2215 return (leftuv > (UV)rightiv) - (leftuv < (UV)rightiv);
2218 NOT_REACHED; /* NOTREACHED */
2222 NV const rnv = SvNV_nomg(right);
2223 NV const lnv = SvNV_nomg(left);
2225 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2226 if (Perl_isnan(lnv) || Perl_isnan(rnv)) {
2229 return (lnv > rnv) - (lnv < rnv);
2248 tryAMAGICbin_MG(ncmp_amg, AMGf_numeric);
2251 value = do_ncmp(left, right);
2263 /* also used for: pp_sge() pp_sgt() pp_slt() */
2269 int amg_type = sle_amg;
2273 switch (PL_op->op_type) {
2292 tryAMAGICbin_MG(amg_type, 0);
2296 #ifdef USE_LOCALE_COLLATE
2297 (IN_LC_RUNTIME(LC_COLLATE))
2298 ? sv_cmp_locale_flags(left, right, 0)
2301 sv_cmp_flags(left, right, 0);
2302 SETs(boolSV(cmp * multiplier < rhs));
2310 tryAMAGICbin_MG(seq_amg, 0);
2313 SETs(boolSV(sv_eq_flags(left, right, 0)));
2321 tryAMAGICbin_MG(sne_amg, 0);
2324 SETs(boolSV(!sv_eq_flags(left, right, 0)));
2332 tryAMAGICbin_MG(scmp_amg, 0);
2336 #ifdef USE_LOCALE_COLLATE
2337 (IN_LC_RUNTIME(LC_COLLATE))
2338 ? sv_cmp_locale_flags(left, right, 0)
2341 sv_cmp_flags(left, right, 0);
2350 tryAMAGICbin_MG(band_amg, AMGf_assign);
2353 if (SvNIOKp(left) || SvNIOKp(right)) {
2354 const bool left_ro_nonnum = !SvNIOKp(left) && SvREADONLY(left);
2355 const bool right_ro_nonnum = !SvNIOKp(right) && SvREADONLY(right);
2356 if (PL_op->op_private & HINT_INTEGER) {
2357 const IV i = SvIV_nomg(left) & SvIV_nomg(right);
2361 const UV u = SvUV_nomg(left) & SvUV_nomg(right);
2364 if (left_ro_nonnum && left != TARG) SvNIOK_off(left);
2365 if (right_ro_nonnum) SvNIOK_off(right);
2368 do_vop(PL_op->op_type, TARG, left, right);
2378 tryAMAGICbin_MG(band_amg, AMGf_assign|AMGf_numarg);
2380 dATARGET; dPOPTOPssrl;
2381 if (PL_op->op_private & HINT_INTEGER) {
2382 const IV i = SvIV_nomg(left) & SvIV_nomg(right);
2386 const UV u = SvUV_nomg(left) & SvUV_nomg(right);
2396 tryAMAGICbin_MG(sband_amg, AMGf_assign);
2398 dATARGET; dPOPTOPssrl;
2399 do_vop(OP_BIT_AND, TARG, left, right);
2404 /* also used for: pp_bit_xor() */
2409 const int op_type = PL_op->op_type;
2411 tryAMAGICbin_MG((op_type == OP_BIT_OR ? bor_amg : bxor_amg), AMGf_assign);
2414 if (SvNIOKp(left) || SvNIOKp(right)) {
2415 const bool left_ro_nonnum = !SvNIOKp(left) && SvREADONLY(left);
2416 const bool right_ro_nonnum = !SvNIOKp(right) && SvREADONLY(right);
2417 if (PL_op->op_private & HINT_INTEGER) {
2418 const IV l = (USE_LEFT(left) ? SvIV_nomg(left) : 0);
2419 const IV r = SvIV_nomg(right);
2420 const IV result = op_type == OP_BIT_OR ? (l | r) : (l ^ r);
2424 const UV l = (USE_LEFT(left) ? SvUV_nomg(left) : 0);
2425 const UV r = SvUV_nomg(right);
2426 const UV result = op_type == OP_BIT_OR ? (l | r) : (l ^ r);
2429 if (left_ro_nonnum && left != TARG) SvNIOK_off(left);
2430 if (right_ro_nonnum) SvNIOK_off(right);
2433 do_vop(op_type, TARG, left, right);
2440 /* also used for: pp_nbit_xor() */
2445 const int op_type = PL_op->op_type;
2447 tryAMAGICbin_MG((op_type == OP_NBIT_OR ? bor_amg : bxor_amg),
2448 AMGf_assign|AMGf_numarg);
2450 dATARGET; dPOPTOPssrl;
2451 if (PL_op->op_private & HINT_INTEGER) {
2452 const IV l = (USE_LEFT(left) ? SvIV_nomg(left) : 0);
2453 const IV r = SvIV_nomg(right);
2454 const IV result = op_type == OP_NBIT_OR ? (l | r) : (l ^ r);
2458 const UV l = (USE_LEFT(left) ? SvUV_nomg(left) : 0);
2459 const UV r = SvUV_nomg(right);
2460 const UV result = op_type == OP_NBIT_OR ? (l | r) : (l ^ r);
2467 /* also used for: pp_sbit_xor() */
2472 const int op_type = PL_op->op_type;
2474 tryAMAGICbin_MG((op_type == OP_SBIT_OR ? sbor_amg : sbxor_amg),
2477 dATARGET; dPOPTOPssrl;
2478 do_vop(op_type == OP_SBIT_OR ? OP_BIT_OR : OP_BIT_XOR, TARG, left,
2484 PERL_STATIC_INLINE bool
2485 S_negate_string(pTHX)
2490 SV * const sv = TOPs;
2491 if (!SvPOKp(sv) || SvNIOK(sv) || (!SvPOK(sv) && SvNIOKp(sv)))
2493 s = SvPV_nomg_const(sv, len);
2494 if (isIDFIRST(*s)) {
2495 sv_setpvs(TARG, "-");
2498 else if (*s == '+' || (*s == '-' && !looks_like_number(sv))) {
2499 sv_setsv_nomg(TARG, sv);
2500 *SvPV_force_nomg(TARG, len) = *s == '-' ? '+' : '-';
2510 tryAMAGICun_MG(neg_amg, AMGf_numeric);
2511 if (S_negate_string(aTHX)) return NORMAL;
2513 SV * const sv = TOPs;
2516 /* It's publicly an integer */
2519 if (SvIVX(sv) == IV_MIN) {
2520 /* 2s complement assumption. */
2521 SETi(SvIVX(sv)); /* special case: -((UV)IV_MAX+1) ==
2525 else if (SvUVX(sv) <= IV_MAX) {
2530 else if (SvIVX(sv) != IV_MIN) {
2534 #ifdef PERL_PRESERVE_IVUV
2541 if (SvNIOKp(sv) && (SvNIOK(sv) || !SvPOK(sv)))
2542 SETn(-SvNV_nomg(sv));
2543 else if (SvPOKp(sv) && SvIV_please_nomg(sv))
2544 goto oops_its_an_int;
2546 SETn(-SvNV_nomg(sv));
2556 tryAMAGICun_MG(not_amg, 0);
2558 *PL_stack_sp = boolSV(!SvTRUE_nomg_NN(sv));
2563 S_scomplement(pTHX_ SV *targ, SV *sv)
2569 sv_copypv_nomg(TARG, sv);
2570 tmps = (U8*)SvPV_nomg(TARG, len);
2573 if (len && ! utf8_to_bytes(tmps, &len)) {
2574 Perl_croak(aTHX_ FATAL_ABOVE_FF_MSG, PL_op_desc[PL_op->op_type]);
2576 SvCUR_set(TARG, len);
2584 for ( ; anum && PTR2nat(tmps) % sizeof(long); anum--, tmps++)
2587 for ( ; anum >= (I32)sizeof(long); anum -= (I32)sizeof(long), tmpl++)
2592 for ( ; anum > 0; anum--, tmps++)
2599 tryAMAGICun_MG(compl_amg, AMGf_numeric);
2603 if (PL_op->op_private & HINT_INTEGER) {
2604 const IV i = ~SvIV_nomg(sv);
2608 const UV u = ~SvUV_nomg(sv);
2613 S_scomplement(aTHX_ TARG, sv);
2623 tryAMAGICun_MG(compl_amg, AMGf_numeric|AMGf_numarg);
2626 if (PL_op->op_private & HINT_INTEGER) {
2627 const IV i = ~SvIV_nomg(sv);
2631 const UV u = ~SvUV_nomg(sv);
2641 tryAMAGICun_MG(scompl_amg, AMGf_numeric);
2644 S_scomplement(aTHX_ TARG, sv);
2650 /* integer versions of some of the above */
2655 tryAMAGICbin_MG(mult_amg, AMGf_assign);
2658 SETi( left * right );
2667 tryAMAGICbin_MG(div_amg, AMGf_assign);
2670 IV value = SvIV_nomg(right);
2672 DIE(aTHX_ "Illegal division by zero");
2673 num = SvIV_nomg(left);
2675 /* avoid FPE_INTOVF on some platforms when num is IV_MIN */
2679 value = num / value;
2688 tryAMAGICbin_MG(modulo_amg, AMGf_assign);
2692 DIE(aTHX_ "Illegal modulus zero");
2693 /* avoid FPE_INTOVF on some platforms when left is IV_MIN */
2697 SETi( left % right );
2705 tryAMAGICbin_MG(add_amg, AMGf_assign);
2707 dPOPTOPiirl_ul_nomg;
2708 SETi( left + right );
2716 tryAMAGICbin_MG(subtr_amg, AMGf_assign);
2718 dPOPTOPiirl_ul_nomg;
2719 SETi( left - right );
2727 tryAMAGICbin_MG(lt_amg, 0);
2730 SETs(boolSV(left < right));
2738 tryAMAGICbin_MG(gt_amg, 0);
2741 SETs(boolSV(left > right));
2749 tryAMAGICbin_MG(le_amg, 0);
2752 SETs(boolSV(left <= right));
2760 tryAMAGICbin_MG(ge_amg, 0);
2763 SETs(boolSV(left >= right));
2771 tryAMAGICbin_MG(eq_amg, 0);
2774 SETs(boolSV(left == right));
2782 tryAMAGICbin_MG(ne_amg, 0);
2785 SETs(boolSV(left != right));
2793 tryAMAGICbin_MG(ncmp_amg, 0);
2800 else if (left < right)
2812 tryAMAGICun_MG(neg_amg, 0);
2813 if (S_negate_string(aTHX)) return NORMAL;
2815 SV * const sv = TOPs;
2816 IV const i = SvIV_nomg(sv);
2822 /* High falutin' math. */
2827 tryAMAGICbin_MG(atan2_amg, 0);
2830 SETn(Perl_atan2(left, right));
2836 /* also used for: pp_cos() pp_exp() pp_log() pp_sqrt() */
2841 int amg_type = fallback_amg;
2842 const char *neg_report = NULL;
2843 const int op_type = PL_op->op_type;
2846 case OP_SIN: amg_type = sin_amg; break;
2847 case OP_COS: amg_type = cos_amg; break;
2848 case OP_EXP: amg_type = exp_amg; break;
2849 case OP_LOG: amg_type = log_amg; neg_report = "log"; break;
2850 case OP_SQRT: amg_type = sqrt_amg; neg_report = "sqrt"; break;
2853 assert(amg_type != fallback_amg);
2855 tryAMAGICun_MG(amg_type, 0);
2857 SV * const arg = TOPs;
2858 const NV value = SvNV_nomg(arg);
2864 if (neg_report) { /* log or sqrt */
2866 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2867 ! Perl_isnan(value) &&
2869 (op_type == OP_LOG ? (value <= 0.0) : (value < 0.0))) {
2870 SET_NUMERIC_STANDARD();
2871 /* diag_listed_as: Can't take log of %g */
2872 DIE(aTHX_ "Can't take %s of %" NVgf, neg_report, value);
2877 case OP_SIN: result = Perl_sin(value); break;
2878 case OP_COS: result = Perl_cos(value); break;
2879 case OP_EXP: result = Perl_exp(value); break;
2880 case OP_LOG: result = Perl_log(value); break;
2881 case OP_SQRT: result = Perl_sqrt(value); break;
2888 /* Support Configure command-line overrides for rand() functions.
2889 After 5.005, perhaps we should replace this by Configure support
2890 for drand48(), random(), or rand(). For 5.005, though, maintain
2891 compatibility by calling rand() but allow the user to override it.
2892 See INSTALL for details. --Andy Dougherty 15 July 1998
2894 /* Now it's after 5.005, and Configure supports drand48() and random(),
2895 in addition to rand(). So the overrides should not be needed any more.
2896 --Jarkko Hietaniemi 27 September 1998
2901 if (!PL_srand_called) {
2902 (void)seedDrand01((Rand_seed_t)seed());
2903 PL_srand_called = TRUE;
2915 SV * const sv = POPs;
2921 /* 1 of 2 things can be carried through SvNV, SP or TARG, SP was carried */
2922 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2923 if (! Perl_isnan(value) && value == 0.0)
2933 sv_setnv_mg(TARG, value);
2944 if (MAXARG >= 1 && (TOPs || POPs)) {
2951 pv = SvPV(top, len);
2952 flags = grok_number(pv, len, &anum);
2954 if (!(flags & IS_NUMBER_IN_UV)) {
2955 Perl_ck_warner_d(aTHX_ packWARN(WARN_OVERFLOW),
2956 "Integer overflow in srand");
2964 (void)seedDrand01((Rand_seed_t)anum);
2965 PL_srand_called = TRUE;
2969 /* Historically srand always returned true. We can avoid breaking
2971 sv_setpvs(TARG, "0 but true");
2980 tryAMAGICun_MG(int_amg, AMGf_numeric);
2982 SV * const sv = TOPs;
2983 const IV iv = SvIV_nomg(sv);
2984 /* XXX it's arguable that compiler casting to IV might be subtly
2985 different from modf (for numbers inside (IV_MIN,UV_MAX)) in which
2986 else preferring IV has introduced a subtle behaviour change bug. OTOH
2987 relying on floating point to be accurate is a bug. */
2992 else if (SvIOK(sv)) {
2994 SETu(SvUV_nomg(sv));
2999 const NV value = SvNV_nomg(sv);
3000 if (UNLIKELY(Perl_isinfnan(value)))
3002 else if (value >= 0.0) {
3003 if (value < (NV)UV_MAX + 0.5) {
3006 SETn(Perl_floor(value));
3010 if (value > (NV)IV_MIN - 0.5) {
3013 SETn(Perl_ceil(value));
3024 tryAMAGICun_MG(abs_amg, AMGf_numeric);
3026 SV * const sv = TOPs;
3027 /* This will cache the NV value if string isn't actually integer */
3028 const IV iv = SvIV_nomg(sv);
3033 else if (SvIOK(sv)) {
3034 /* IVX is precise */
3036 SETu(SvUV_nomg(sv)); /* force it to be numeric only */
3044 /* 2s complement assumption. Also, not really needed as
3045 IV_MIN and -IV_MIN should both be %100...00 and NV-able */
3051 const NV value = SvNV_nomg(sv);
3062 /* also used for: pp_hex() */
3068 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES;
3072 SV* const sv = TOPs;
3074 tmps = (SvPV_const(sv, len));
3076 /* If Unicode, try to downgrade
3077 * If not possible, croak. */
3078 SV* const tsv = sv_2mortal(newSVsv(sv));
3081 sv_utf8_downgrade(tsv, FALSE);
3082 tmps = SvPV_const(tsv, len);
3084 if (PL_op->op_type == OP_HEX)
3087 while (*tmps && len && isSPACE(*tmps))
3091 if (isALPHA_FOLD_EQ(*tmps, 'x')) {
3093 flags |= PERL_SCAN_DISALLOW_PREFIX;
3095 result_uv = grok_hex (tmps, &len, &flags, &result_nv);
3097 else if (isALPHA_FOLD_EQ(*tmps, 'b')) {
3099 flags |= PERL_SCAN_DISALLOW_PREFIX;
3100 result_uv = grok_bin (tmps, &len, &flags, &result_nv);
3103 if (isALPHA_FOLD_EQ(*tmps, 'o')) {
3106 result_uv = grok_oct (tmps, &len, &flags, &result_nv);
3109 if (flags & PERL_SCAN_GREATER_THAN_UV_MAX) {
3124 SV * const sv = TOPs;
3126 U32 in_bytes = IN_BYTES;
3127 /* Simplest case shortcut:
3128 * set svflags to just the SVf_POK|SVs_GMG|SVf_UTF8 from the SV,
3129 * with the SVf_UTF8 flag inverted if under 'use bytes' (HINT_BYTES
3132 U32 svflags = (SvFLAGS(sv) ^ (in_bytes << 26)) & (SVf_POK|SVs_GMG|SVf_UTF8);
3134 STATIC_ASSERT_STMT(SVf_UTF8 == (HINT_BYTES << 26));
3137 if (LIKELY(svflags == SVf_POK))
3140 if (svflags & SVs_GMG)
3145 if (!IN_BYTES) { /* reread to avoid using an C auto/register */
3146 if ((SvFLAGS(sv) & (SVf_POK|SVf_UTF8)) == SVf_POK)
3148 if ( SvPOK(sv) && (PL_op->op_private & OPpTRUEBOOL)) {
3149 /* no need to convert from bytes to chars */
3153 len = sv_len_utf8_nomg(sv);
3156 /* unrolled SvPV_nomg_const(sv,len) */
3157 if (SvPOK_nog(sv)) {
3160 if (PL_op->op_private & OPpTRUEBOOL) {
3162 SETs(len ? &PL_sv_yes : &PL_sv_zero);
3167 (void)sv_2pv_flags(sv, &len, 0|SV_CONST_RETURN);
3170 TARGi((IV)(len), 1);
3173 if (!SvPADTMP(TARG)) {
3174 /* OPpTARGET_MY: targ is var in '$lex = length()' */
3179 /* TARG is on stack at this point and is overwriten by SETs.
3180 * This branch is the odd one out, so put TARG by default on
3181 * stack earlier to let local SP go out of liveness sooner */
3184 return NORMAL; /* no putback, SP didn't move in this opcode */
3188 /* Returns false if substring is completely outside original string.
3189 No length is indicated by len_iv = 0 and len_is_uv = 0. len_is_uv must
3190 always be true for an explicit 0.
3193 Perl_translate_substr_offsets( STRLEN curlen, IV pos1_iv,
3194 bool pos1_is_uv, IV len_iv,
3195 bool len_is_uv, STRLEN *posp,
3201 PERL_ARGS_ASSERT_TRANSLATE_SUBSTR_OFFSETS;
3203 if (!pos1_is_uv && pos1_iv < 0 && curlen) {
3204 pos1_is_uv = curlen-1 > ~(UV)pos1_iv;
3207 if ((pos1_is_uv || pos1_iv > 0) && (UV)pos1_iv > curlen)
3210 if (len_iv || len_is_uv) {
3211 if (!len_is_uv && len_iv < 0) {
3212 pos2_iv = curlen + len_iv;
3214 pos2_is_uv = curlen-1 > ~(UV)len_iv;
3217 } else { /* len_iv >= 0 */
3218 if (!pos1_is_uv && pos1_iv < 0) {
3219 pos2_iv = pos1_iv + len_iv;
3220 pos2_is_uv = (UV)len_iv > (UV)IV_MAX;
3222 if ((UV)len_iv > curlen-(UV)pos1_iv)
3225 pos2_iv = pos1_iv+len_iv;
3235 if (!pos2_is_uv && pos2_iv < 0) {
3236 if (!pos1_is_uv && pos1_iv < 0)
3240 else if (!pos1_is_uv && pos1_iv < 0)
3243 if ((UV)pos2_iv < (UV)pos1_iv)
3245 if ((UV)pos2_iv > curlen)
3248 /* pos1_iv and pos2_iv both in 0..curlen, so the cast is safe */
3249 *posp = (STRLEN)( (UV)pos1_iv );
3250 *lenp = (STRLEN)( (UV)pos2_iv - (UV)pos1_iv );
3267 I32 lvalue = PL_op->op_flags & OPf_MOD || LVRET;
3268 const bool rvalue = (GIMME_V != G_VOID);
3271 const char *repl = NULL;
3273 int num_args = PL_op->op_private & 7;
3274 bool repl_need_utf8_upgrade = FALSE;
3278 if(!(repl_sv = POPs)) num_args--;
3280 if ((len_sv = POPs)) {
3281 len_iv = SvIV(len_sv);
3282 len_is_uv = len_iv ? SvIOK_UV(len_sv) : 1;
3287 pos1_iv = SvIV(pos_sv);
3288 pos1_is_uv = SvIOK_UV(pos_sv);
3290 if (PL_op->op_private & OPpSUBSTR_REPL_FIRST) {
3294 if (lvalue && !repl_sv) {
3296 ret = sv_2mortal(newSV_type(SVt_PVLV)); /* Not TARG RT#67838 */
3297 sv_magic(ret, NULL, PERL_MAGIC_substr, NULL, 0);
3299 LvTARG(ret) = SvREFCNT_inc_simple(sv);
3301 pos1_is_uv || pos1_iv >= 0
3302 ? (STRLEN)(UV)pos1_iv
3303 : (LvFLAGS(ret) |= LVf_NEG_OFF, (STRLEN)(UV)-pos1_iv);
3305 len_is_uv || len_iv > 0
3306 ? (STRLEN)(UV)len_iv
3307 : (LvFLAGS(ret) |= LVf_NEG_LEN, (STRLEN)(UV)-len_iv);
3309 PUSHs(ret); /* avoid SvSETMAGIC here */
3313 repl = SvPV_const(repl_sv, repl_len);
3316 Perl_ck_warner(aTHX_ packWARN(WARN_SUBSTR),
3317 "Attempt to use reference as lvalue in substr"
3319 tmps = SvPV_force_nomg(sv, curlen);
3320 if (DO_UTF8(repl_sv) && repl_len) {
3322 /* Upgrade the dest, and recalculate tmps in case the buffer
3323 * got reallocated; curlen may also have been changed */
3324 sv_utf8_upgrade_nomg(sv);
3325 tmps = SvPV_nomg(sv, curlen);
3328 else if (DO_UTF8(sv))
3329 repl_need_utf8_upgrade = TRUE;
3331 else tmps = SvPV_const(sv, curlen);
3333 utf8_curlen = sv_or_pv_len_utf8(sv, tmps, curlen);
3334 if (utf8_curlen == curlen)
3337 curlen = utf8_curlen;
3343 STRLEN pos, len, byte_len, byte_pos;
3345 if (!translate_substr_offsets(
3346 curlen, pos1_iv, pos1_is_uv, len_iv, len_is_uv, &pos, &len
3350 byte_pos = utf8_curlen
3351 ? sv_or_pv_pos_u2b(sv, tmps, pos, &byte_len) : pos;
3356 SvTAINTED_off(TARG); /* decontaminate */
3357 SvUTF8_off(TARG); /* decontaminate */
3358 sv_setpvn(TARG, tmps, byte_len);
3359 #ifdef USE_LOCALE_COLLATE
3360 sv_unmagic(TARG, PERL_MAGIC_collxfrm);
3367 SV* repl_sv_copy = NULL;
3369 if (repl_need_utf8_upgrade) {
3370 repl_sv_copy = newSVsv(repl_sv);
3371 sv_utf8_upgrade(repl_sv_copy);
3372 repl = SvPV_const(repl_sv_copy, repl_len);
3376 sv_insert_flags(sv, byte_pos, byte_len, repl, repl_len, 0);
3377 SvREFCNT_dec(repl_sv_copy);
3380 if (PL_op->op_private & OPpSUBSTR_REPL_FIRST)
3390 Perl_croak(aTHX_ "substr outside of string");
3391 Perl_ck_warner(aTHX_ packWARN(WARN_SUBSTR), "substr outside of string");
3398 const IV size = POPi;
3399 SV* offsetsv = POPs;
3400 SV * const src = POPs;
3401 const I32 lvalue = PL_op->op_flags & OPf_MOD || LVRET;
3407 /* extract a STRLEN-ranged integer value from offsetsv into offset,
3408 * or flag that its out of range */
3410 IV iv = SvIV(offsetsv);
3412 /* avoid a large UV being wrapped to a negative value */
3413 if (SvIOK_UV(offsetsv) && SvUVX(offsetsv) > (UV)IV_MAX)
3414 errflags = LVf_OUT_OF_RANGE;
3416 errflags = (LVf_NEG_OFF|LVf_OUT_OF_RANGE);
3417 #if PTRSIZE < IVSIZE
3418 else if (iv > Size_t_MAX)
3419 errflags = LVf_OUT_OF_RANGE;
3422 offset = (STRLEN)iv;
3425 retuv = errflags ? 0 : do_vecget(src, offset, size);
3427 if (lvalue) { /* it's an lvalue! */
3428 ret = sv_2mortal(newSV_type(SVt_PVLV)); /* Not TARG RT#67838 */
3429 sv_magic(ret, NULL, PERL_MAGIC_vec, NULL, 0);
3431 LvTARG(ret) = SvREFCNT_inc_simple(src);
3432 LvTARGOFF(ret) = offset;
3433 LvTARGLEN(ret) = size;
3434 LvFLAGS(ret) = errflags;
3438 SvTAINTED_off(TARG); /* decontaminate */
3442 sv_setuv(ret, retuv);
3450 /* also used for: pp_rindex() */
3463 const char *little_p;
3466 const bool is_index = PL_op->op_type == OP_INDEX;
3467 const bool threeargs = MAXARG >= 3 && (TOPs || ((void)POPs,0));
3473 big_p = SvPV_const(big, biglen);
3474 little_p = SvPV_const(little, llen);
3476 big_utf8 = DO_UTF8(big);
3477 little_utf8 = DO_UTF8(little);
3478 if (big_utf8 ^ little_utf8) {
3479 /* One needs to be upgraded. */
3481 /* Well, maybe instead we might be able to downgrade the small
3483 char * const pv = (char*)bytes_from_utf8((U8 *)little_p, &llen,
3486 /* If the large string is ISO-8859-1, and it's not possible to
3487 convert the small string to ISO-8859-1, then there is no
3488 way that it could be found anywhere by index. */
3493 /* At this point, pv is a malloc()ed string. So donate it to temp
3494 to ensure it will get free()d */
3495 little = temp = newSV(0);
3496 sv_usepvn(temp, pv, llen);
3497 little_p = SvPVX(little);
3499 temp = newSVpvn(little_p, llen);
3501 sv_utf8_upgrade(temp);
3503 little_p = SvPV_const(little, llen);
3506 if (SvGAMAGIC(big)) {
3507 /* Life just becomes a lot easier if I use a temporary here.
3508 Otherwise I need to avoid calls to sv_pos_u2b(), which (dangerously)
3509 will trigger magic and overloading again, as will fbm_instr()
3511 big = newSVpvn_flags(big_p, biglen,
3512 SVs_TEMP | (big_utf8 ? SVf_UTF8 : 0));
3515 if (SvGAMAGIC(little) || (is_index && !SvOK(little))) {
3516 /* index && SvOK() is a hack. fbm_instr() calls SvPV_const, which will
3517 warn on undef, and we've already triggered a warning with the
3518 SvPV_const some lines above. We can't remove that, as we need to
3519 call some SvPV to trigger overloading early and find out if the
3521 This is all getting too messy. The API isn't quite clean enough,
3522 because data access has side effects.
3524 little = newSVpvn_flags(little_p, llen,
3525 SVs_TEMP | (little_utf8 ? SVf_UTF8 : 0));
3526 little_p = SvPVX(little);
3530 offset = is_index ? 0 : biglen;
3532 if (big_utf8 && offset > 0)
3533 offset = sv_pos_u2b_flags(big, offset, 0, SV_CONST_RETURN);
3539 else if (offset > (SSize_t)biglen)
3541 if (!(little_p = is_index
3542 ? fbm_instr((unsigned char*)big_p + offset,
3543 (unsigned char*)big_p + biglen, little, 0)
3544 : rninstr(big_p, big_p + offset,
3545 little_p, little_p + llen)))
3548 retval = little_p - big_p;
3549 if (retval > 1 && big_utf8)
3550 retval = sv_pos_b2u_flags(big, retval, SV_CONST_RETURN);
3555 /* OPpTRUEBOOL indicates an '== -1' has been optimised away */
3556 if (PL_op->op_private & OPpTRUEBOOL) {
3557 SV *result = ((retval != -1) ^ cBOOL(PL_op->op_private & OPpINDEX_BOOLNEG))
3558 ? &PL_sv_yes : &PL_sv_no;
3559 if (PL_op->op_private & OPpTARGET_MY) {
3560 /* $lex = (index() == -1) */
3561 sv_setsv_mg(TARG, result);
3575 dSP; dMARK; dORIGMARK; dTARGET;
3576 SvTAINTED_off(TARG);
3577 do_sprintf(TARG, SP-MARK, MARK+1);
3578 TAINT_IF(SvTAINTED(TARG));
3590 const U8 *s = (U8*)SvPV_const(argsv, len);
3593 ? (len ? utf8n_to_uvchr(s, len, 0, UTF8_ALLOW_ANYUV) : 0)
3607 if (UNLIKELY(SvAMAGIC(top)))
3609 if (UNLIKELY(isinfnansv(top)))
3610 Perl_croak(aTHX_ "Cannot chr %" NVgf, SvNV(top));
3612 if (!IN_BYTES /* under bytes, chr(-1) eq chr(0xff), etc. */
3613 && ((SvIOKp(top) && !SvIsUV(top) && SvIV_nomg(top) < 0)
3615 ((SvNOKp(top) || (SvOK(top) && !SvIsUV(top)))
3616 && SvNV_nomg(top) < 0.0)))
3618 if (ckWARN(WARN_UTF8)) {
3619 if (SvGMAGICAL(top)) {
3620 SV *top2 = sv_newmortal();
3621 sv_setsv_nomg(top2, top);
3624 Perl_warner(aTHX_ packWARN(WARN_UTF8),
3625 "Invalid negative number (%" SVf ") in chr", SVfARG(top));
3627 value = UNICODE_REPLACEMENT;
3629 value = SvUV_nomg(top);
3633 SvUPGRADE(TARG,SVt_PV);
3635 if (value > 255 && !IN_BYTES) {
3636 SvGROW(TARG, (STRLEN)UVCHR_SKIP(value)+1);
3637 tmps = (char*)uvchr_to_utf8_flags((U8*)SvPVX(TARG), value, 0);
3638 SvCUR_set(TARG, tmps - SvPVX_const(TARG));
3640 (void)SvPOK_only(TARG);
3649 *tmps++ = (char)value;
3651 (void)SvPOK_only(TARG);
3663 const char *tmps = SvPV_const(left, len);
3665 if (DO_UTF8(left)) {
3666 /* If Unicode, try to downgrade.
3667 * If not possible, croak.
3668 * Yes, we made this up. */
3669 SV* const tsv = newSVpvn_flags(tmps, len, SVf_UTF8|SVs_TEMP);
3671 sv_utf8_downgrade(tsv, FALSE);
3672 tmps = SvPV_const(tsv, len);
3674 # ifdef USE_ITHREADS
3676 if (!PL_reentrant_buffer->_crypt_struct_buffer) {
3677 /* This should be threadsafe because in ithreads there is only
3678 * one thread per interpreter. If this would not be true,
3679 * we would need a mutex to protect this malloc. */
3680 PL_reentrant_buffer->_crypt_struct_buffer =
3681 (struct crypt_data *)safemalloc(sizeof(struct crypt_data));
3682 # if defined(__GLIBC__) || defined(__EMX__)
3683 if (PL_reentrant_buffer->_crypt_struct_buffer) {
3684 PL_reentrant_buffer->_crypt_struct_buffer->initialized = 0;
3688 # endif /* HAS_CRYPT_R */
3689 # endif /* USE_ITHREADS */
3691 sv_setpv(TARG, PerlProc_crypt(tmps, SvPV_nolen_const(right)));
3698 "The crypt() function is unimplemented due to excessive paranoia.");
3702 /* Generally UTF-8 and UTF-EBCDIC are indistinguishable at this level. So
3703 * most comments below say UTF-8, when in fact they mean UTF-EBCDIC as well */
3706 /* also used for: pp_lcfirst() */
3710 /* Actually is both lcfirst() and ucfirst(). Only the first character
3711 * changes. This means that possibly we can change in-place, ie., just
3712 * take the source and change that one character and store it back, but not
3713 * if read-only etc, or if the length changes */
3717 STRLEN slen; /* slen is the byte length of the whole SV. */
3720 bool inplace; /* ? Convert first char only, in-place */
3721 bool doing_utf8 = FALSE; /* ? using utf8 */
3722 bool convert_source_to_utf8 = FALSE; /* ? need to convert */
3723 const int op_type = PL_op->op_type;
3726 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
3727 STRLEN ulen; /* ulen is the byte length of the original Unicode character
3728 * stored as UTF-8 at s. */
3729 STRLEN tculen; /* tculen is the byte length of the freshly titlecased (or
3730 * lowercased) character stored in tmpbuf. May be either
3731 * UTF-8 or not, but in either case is the number of bytes */
3732 bool remove_dot_above = FALSE;
3734 s = (const U8*)SvPV_const(source, slen);
3736 /* We may be able to get away with changing only the first character, in
3737 * place, but not if read-only, etc. Later we may discover more reasons to
3738 * not convert in-place. */
3739 inplace = !SvREADONLY(source) && SvPADTMP(source);
3741 #ifdef USE_LOCALE_CTYPE
3743 if (IN_LC_RUNTIME(LC_CTYPE)) {
3744 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
3749 /* First calculate what the changed first character should be. This affects
3750 * whether we can just swap it out, leaving the rest of the string unchanged,
3751 * or even if have to convert the dest to UTF-8 when the source isn't */
3753 if (! slen) { /* If empty */
3754 need = 1; /* still need a trailing NUL */
3758 else if (DO_UTF8(source)) { /* Is the source utf8? */
3762 if (op_type == OP_UCFIRST) {
3763 #ifdef USE_LOCALE_CTYPE
3764 _toTITLE_utf8_flags(s, s +slen, tmpbuf, &tculen, IN_LC_RUNTIME(LC_CTYPE));
3766 _toTITLE_utf8_flags(s, s +slen, tmpbuf, &tculen, 0);
3771 #ifdef USE_LOCALE_CTYPE
3773 _toLOWER_utf8_flags(s, s + slen, tmpbuf, &tculen, IN_LC_RUNTIME(LC_CTYPE));
3775 /* In turkic locales, lower casing an 'I' normally yields U+0131,
3776 * LATIN SMALL LETTER DOTLESS I, but not if the grapheme also
3777 * contains a COMBINING DOT ABOVE. Instead it is treated like
3778 * LATIN CAPITAL LETTER I WITH DOT ABOVE lowercased to 'i'. The
3779 * call to lowercase above has handled this. But SpecialCasing.txt
3780 * says we are supposed to remove the COMBINING DOT ABOVE. We can
3781 * tell if we have this situation if I ==> i in a turkic locale. */
3782 if ( UNLIKELY(PL_in_utf8_turkic_locale)
3783 && IN_LC_RUNTIME(LC_CTYPE)
3784 && (UNLIKELY(*s == 'I' && tmpbuf[0] == 'i')))
3786 /* Here, we know there was a COMBINING DOT ABOVE. We won't be
3787 * able to handle this in-place. */
3790 /* It seems likely that the DOT will immediately follow the
3791 * 'I'. If so, we can remove it simply by indicating to the
3792 * code below to start copying the source just beyond the DOT.
3793 * We know its length is 2 */
3794 if (LIKELY(memBEGINs(s + 1, s + slen, COMBINING_DOT_ABOVE_UTF8))) {
3797 else { /* But if it doesn't follow immediately, set a flag for
3799 remove_dot_above = TRUE;
3803 PERL_UNUSED_VAR(remove_dot_above);
3805 _toLOWER_utf8_flags(s, s + slen, tmpbuf, &tculen, 0);
3810 /* we can't do in-place if the length changes. */
3811 if (ulen != tculen) inplace = FALSE;
3812 need = slen + 1 - ulen + tculen;
3814 else { /* Non-zero length, non-UTF-8, Need to consider locale and if
3815 * latin1 is treated as caseless. Note that a locale takes
3817 ulen = 1; /* Original character is 1 byte */
3818 tculen = 1; /* Most characters will require one byte, but this will
3819 * need to be overridden for the tricky ones */
3823 #ifdef USE_LOCALE_CTYPE
3825 if (IN_LC_RUNTIME(LC_CTYPE)) {
3826 if ( UNLIKELY(PL_in_utf8_turkic_locale)
3827 && ( (op_type == OP_LCFIRST && UNLIKELY(*s == 'I'))
3828 || (op_type == OP_UCFIRST && UNLIKELY(*s == 'i'))))
3830 if (*s == 'I') { /* lcfirst('I') */
3831 tmpbuf[0] = UTF8_TWO_BYTE_HI(LATIN_SMALL_LETTER_DOTLESS_I);
3832 tmpbuf[1] = UTF8_TWO_BYTE_LO(LATIN_SMALL_LETTER_DOTLESS_I);
3834 else { /* ucfirst('i') */
3835 tmpbuf[0] = UTF8_TWO_BYTE_HI(LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE);
3836 tmpbuf[1] = UTF8_TWO_BYTE_LO(LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE);
3841 convert_source_to_utf8 = TRUE;
3842 need += variant_under_utf8_count(s, s + slen);
3844 else if (op_type == OP_LCFIRST) {
3846 /* For lc, there are no gotchas for UTF-8 locales (other than
3847 * the turkish ones already handled above) */
3848 *tmpbuf = toLOWER_LC(*s);
3850 else { /* ucfirst */
3852 /* But for uc, some characters require special handling */
3853 if (IN_UTF8_CTYPE_LOCALE) {
3857 /* This would be a bug if any locales have upper and title case
3859 *tmpbuf = (U8) toUPPER_LC(*s);
3864 /* Here, not in locale. If not using Unicode rules, is a simple
3865 * lower/upper, depending */
3866 if (! IN_UNI_8_BIT) {
3867 *tmpbuf = (op_type == OP_LCFIRST)
3871 else if (op_type == OP_LCFIRST) {
3872 /* lower case the first letter: no trickiness for any character */
3873 *tmpbuf = toLOWER_LATIN1(*s);
3876 /* Here, is ucfirst non-UTF-8, not in locale (unless that locale is
3877 * non-turkic UTF-8, which we treat as not in locale), and cased
3880 #ifdef USE_LOCALE_CTYPE
3884 title_ord = _to_upper_title_latin1(*s, tmpbuf, &tculen, 's');
3886 assert(tculen == 2);
3888 /* If the result is an upper Latin1-range character, it can
3889 * still be represented in one byte, which is its ordinal */
3890 if (UTF8_IS_DOWNGRADEABLE_START(*tmpbuf)) {
3891 *tmpbuf = (U8) title_ord;
3895 /* Otherwise it became more than one ASCII character (in
3896 * the case of LATIN_SMALL_LETTER_SHARP_S) or changed to
3897 * beyond Latin1, so the number of bytes changed, so can't
3898 * replace just the first character in place. */
3901 /* If the result won't fit in a byte, the entire result
3902 * will have to be in UTF-8. Allocate enough space for the
3903 * expanded first byte, and if UTF-8, the rest of the input
3904 * string, some or all of which may also expand to two
3905 * bytes, plus the terminating NUL. */
3906 if (title_ord > 255) {
3908 convert_source_to_utf8 = TRUE;
3910 + variant_under_utf8_count(s, s + slen)
3913 /* The (converted) UTF-8 and UTF-EBCDIC lengths of all
3914 * characters whose title case is above 255 is
3918 else { /* LATIN_SMALL_LETTER_SHARP_S expands by 1 byte */
3919 need = slen + 1 + 1;
3923 } /* End of use Unicode (Latin1) semantics */
3924 } /* End of changing the case of the first character */
3926 /* Here, have the first character's changed case stored in tmpbuf. Ready to
3927 * generate the result */
3930 /* We can convert in place. This means we change just the first
3931 * character without disturbing the rest; no need to grow */
3933 s = d = (U8*)SvPV_force_nomg(source, slen);
3939 /* Here, we can't convert in place; we earlier calculated how much
3940 * space we will need, so grow to accommodate that */
3941 SvUPGRADE(dest, SVt_PV);
3942 d = (U8*)SvGROW(dest, need);
3943 (void)SvPOK_only(dest);
3950 if (! convert_source_to_utf8) {
3952 /* Here both source and dest are in UTF-8, but have to create
3953 * the entire output. We initialize the result to be the
3954 * title/lower cased first character, and then append the rest
3956 sv_setpvn(dest, (char*)tmpbuf, tculen);
3959 /* But this boolean being set means we are in a turkic
3960 * locale, and there is a DOT character that needs to be
3961 * removed, and it isn't immediately after the current
3962 * character. Keep concatenating characters to the output
3963 * one at a time, until we find the DOT, which we simply
3965 if (UNLIKELY(remove_dot_above)) {
3967 Size_t this_len = UTF8SKIP(s + ulen);
3969 sv_catpvn(dest, (char*)(s + ulen), this_len);
3972 if (memBEGINs(s + ulen, s + slen, COMBINING_DOT_ABOVE_UTF8)) {
3976 } while (s + ulen < s + slen);
3979 /* The rest of the string can be concatenated unchanged,
3981 sv_catpvn(dest, (char*)(s + ulen), slen - ulen);
3985 const U8 *const send = s + slen;
3987 /* Here the dest needs to be in UTF-8, but the source isn't,
3988 * except we earlier UTF-8'd the first character of the source
3989 * into tmpbuf. First put that into dest, and then append the
3990 * rest of the source, converting it to UTF-8 as we go. */
3992 /* Assert tculen is 2 here because the only characters that
3993 * get to this part of the code have 2-byte UTF-8 equivalents */
3994 assert(tculen == 2);
3996 *d++ = *(tmpbuf + 1);
3997 s++; /* We have just processed the 1st char */
4000 append_utf8_from_native_byte(*s, &d);
4005 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
4009 else { /* in-place UTF-8. Just overwrite the first character */
4010 Copy(tmpbuf, d, tculen, U8);
4011 SvCUR_set(dest, need - 1);
4015 else { /* Neither source nor dest are, nor need to be UTF-8 */
4017 if (inplace) { /* in-place, only need to change the 1st char */
4020 else { /* Not in-place */
4022 /* Copy the case-changed character(s) from tmpbuf */
4023 Copy(tmpbuf, d, tculen, U8);
4024 d += tculen - 1; /* Code below expects d to point to final
4025 * character stored */
4028 else { /* empty source */
4029 /* See bug #39028: Don't taint if empty */
4033 /* In a "use bytes" we don't treat the source as UTF-8, but, still want
4034 * the destination to retain that flag */
4035 if (DO_UTF8(source))
4038 if (!inplace) { /* Finish the rest of the string, unchanged */
4039 /* This will copy the trailing NUL */
4040 Copy(s + 1, d + 1, slen, U8);
4041 SvCUR_set(dest, need - 1);
4044 #ifdef USE_LOCALE_CTYPE
4045 if (IN_LC_RUNTIME(LC_CTYPE)) {
4050 if (dest != source && SvTAINTED(source))
4068 if ( SvPADTMP(source)
4069 && !SvREADONLY(source) && SvPOK(source)
4072 #ifdef USE_LOCALE_CTYPE
4073 (IN_LC_RUNTIME(LC_CTYPE))
4074 ? ! IN_UTF8_CTYPE_LOCALE
4080 /* We can convert in place. The reason we can't if in UNI_8_BIT is to
4081 * make the loop tight, so we overwrite the source with the dest before
4082 * looking at it, and we need to look at the original source
4083 * afterwards. There would also need to be code added to handle
4084 * switching to not in-place in midstream if we run into characters
4085 * that change the length. Since being in locale overrides UNI_8_BIT,
4086 * that latter becomes irrelevant in the above test; instead for
4087 * locale, the size can't normally change, except if the locale is a
4090 s = d = (U8*)SvPV_force_nomg(source, len);
4097 s = (const U8*)SvPV_nomg_const(source, len);
4100 SvUPGRADE(dest, SVt_PV);
4101 d = (U8*)SvGROW(dest, min);
4102 (void)SvPOK_only(dest);
4107 #ifdef USE_LOCALE_CTYPE
4109 if (IN_LC_RUNTIME(LC_CTYPE)) {
4110 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
4115 /* Overloaded values may have toggled the UTF-8 flag on source, so we need
4116 to check DO_UTF8 again here. */
4118 if (DO_UTF8(source)) {
4119 const U8 *const send = s + len;
4120 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
4122 #define GREEK_CAPITAL_LETTER_IOTA 0x0399
4123 #define COMBINING_GREEK_YPOGEGRAMMENI 0x0345
4124 /* All occurrences of these are to be moved to follow any other marks.
4125 * This is context-dependent. We may not be passed enough context to
4126 * move the iota subscript beyond all of them, but we do the best we can
4127 * with what we're given. The result is always better than if we
4128 * hadn't done this. And, the problem would only arise if we are
4129 * passed a character without all its combining marks, which would be
4130 * the caller's mistake. The information this is based on comes from a
4131 * comment in Unicode SpecialCasing.txt, (and the Standard's text
4132 * itself) and so can't be checked properly to see if it ever gets
4133 * revised. But the likelihood of it changing is remote */
4134 bool in_iota_subscript = FALSE;
4140 if (UNLIKELY(in_iota_subscript)) {
4141 UV cp = utf8_to_uvchr_buf(s, send, NULL);
4143 if (! _invlist_contains_cp(PL_utf8_mark, cp)) {
4145 /* A non-mark. Time to output the iota subscript */
4146 *d++ = UTF8_TWO_BYTE_HI(GREEK_CAPITAL_LETTER_IOTA);
4147 *d++ = UTF8_TWO_BYTE_LO(GREEK_CAPITAL_LETTER_IOTA);
4148 in_iota_subscript = FALSE;
4152 /* Then handle the current character. Get the changed case value
4153 * and copy it to the output buffer */
4156 #ifdef USE_LOCALE_CTYPE
4157 uv = _toUPPER_utf8_flags(s, send, tmpbuf, &ulen, IN_LC_RUNTIME(LC_CTYPE));
4159 uv = _toUPPER_utf8_flags(s, send, tmpbuf, &ulen, 0);
4161 if (uv == GREEK_CAPITAL_LETTER_IOTA
4162 && utf8_to_uvchr_buf(s, send, 0) == COMBINING_GREEK_YPOGEGRAMMENI)
4164 in_iota_subscript = TRUE;
4167 if (ulen > u && (SvLEN(dest) < (min += ulen - u))) {
4168 /* If the eventually required minimum size outgrows the
4169 * available space, we need to grow. */
4170 const UV o = d - (U8*)SvPVX_const(dest);
4172 /* If someone uppercases one million U+03B0s we SvGROW()
4173 * one million times. Or we could try guessing how much to
4174 * allocate without allocating too much. But we can't
4175 * really guess without examining the rest of the string.
4176 * Such is life. See corresponding comment in lc code for
4178 d = o + (U8*) SvGROW(dest, min);
4180 Copy(tmpbuf, d, ulen, U8);
4185 if (in_iota_subscript) {
4186 *d++ = UTF8_TWO_BYTE_HI(GREEK_CAPITAL_LETTER_IOTA);
4187 *d++ = UTF8_TWO_BYTE_LO(GREEK_CAPITAL_LETTER_IOTA);
4192 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
4194 else { /* Not UTF-8 */
4196 const U8 *const send = s + len;
4198 /* Use locale casing if in locale; regular style if not treating
4199 * latin1 as having case; otherwise the latin1 casing. Do the
4200 * whole thing in a tight loop, for speed, */
4201 #ifdef USE_LOCALE_CTYPE
4202 if (IN_LC_RUNTIME(LC_CTYPE)) {
4203 if (IN_UTF8_CTYPE_LOCALE) {
4206 for (; s < send; d++, s++)
4207 *d = (U8) toUPPER_LC(*s);
4211 if (! IN_UNI_8_BIT) {
4212 for (; s < send; d++, s++) {
4217 #ifdef USE_LOCALE_CTYPE
4220 for (; s < send; d++, s++) {
4223 *d = toUPPER_LATIN1_MOD(*s);
4224 if ( LIKELY(*d != LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)
4226 #ifdef USE_LOCALE_CTYPE
4228 && (LIKELY( ! PL_in_utf8_turkic_locale
4229 || ! IN_LC_RUNTIME(LC_CTYPE))
4237 /* The mainstream case is the tight loop above. To avoid
4238 * extra tests in that, all three characters that always
4239 * require special handling are mapped by the MOD to the
4240 * one tested just above. Use the source to distinguish
4241 * between those cases */
4243 #if UNICODE_MAJOR_VERSION > 2 \
4244 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
4245 && UNICODE_DOT_DOT_VERSION >= 8)
4246 if (*s == LATIN_SMALL_LETTER_SHARP_S) {
4248 /* uc() of this requires 2 characters, but they are
4249 * ASCII. If not enough room, grow the string */
4250 if (SvLEN(dest) < ++min) {
4251 const UV o = d - (U8*)SvPVX_const(dest);
4252 d = o + (U8*) SvGROW(dest, min);
4254 *d++ = 'S'; *d = 'S'; /* upper case is 'SS' */
4255 continue; /* Back to the tight loop; still in ASCII */
4259 /* The other special handling characters have their
4260 * upper cases outside the latin1 range, hence need to be
4261 * in UTF-8, so the whole result needs to be in UTF-8.
4263 * So, here we are somewhere in the middle of processing a
4264 * non-UTF-8 string, and realize that we will have to
4265 * convert the whole thing to UTF-8. What to do? There
4266 * are several possibilities. The simplest to code is to
4267 * convert what we have so far, set a flag, and continue on
4268 * in the loop. The flag would be tested each time through
4269 * the loop, and if set, the next character would be
4270 * converted to UTF-8 and stored. But, I (khw) didn't want
4271 * to slow down the mainstream case at all for this fairly
4272 * rare case, so I didn't want to add a test that didn't
4273 * absolutely have to be there in the loop, besides the
4274 * possibility that it would get too complicated for
4275 * optimizers to deal with. Another possibility is to just
4276 * give up, convert the source to UTF-8, and restart the
4277 * function that way. Another possibility is to convert
4278 * both what has already been processed and what is yet to
4279 * come separately to UTF-8, then jump into the loop that
4280 * handles UTF-8. But the most efficient time-wise of the
4281 * ones I could think of is what follows, and turned out to
4282 * not require much extra code.
4284 * First, calculate the extra space needed for the
4285 * remainder of the source needing to be in UTF-8. Except
4286 * for the 'i' in Turkic locales, in UTF-8 strings, the
4287 * uppercase of a character below 256 occupies the same
4288 * number of bytes as the original. Therefore, the space
4289 * needed is the that number plus the number of characters
4290 * that become two bytes when converted to UTF-8, plus, in
4291 * turkish locales, the number of 'i's. */
4293 extra = send - s + variant_under_utf8_count(s, send);
4295 #ifdef USE_LOCALE_CTYPE
4297 if (UNLIKELY(*s == 'i')) { /* We wouldn't get an 'i' here
4298 unless are in a Turkic
4300 const U8 * s_peek = s;
4305 s_peek = (U8 *) memchr(s_peek + 1, 'i',
4306 send - (s_peek + 1));
4307 } while (s_peek != NULL);
4311 /* Convert what we have so far into UTF-8, telling the
4312 * function that we know it should be converted, and to
4313 * allow extra space for what we haven't processed yet.
4315 * This may cause the string pointer to move, so need to
4316 * save and re-find it. */
4318 len = d - (U8*)SvPVX_const(dest);
4319 SvCUR_set(dest, len);
4320 len = sv_utf8_upgrade_flags_grow(dest,
4321 SV_GMAGIC|SV_FORCE_UTF8_UPGRADE,
4323 + 1 /* trailing NUL */ );
4324 d = (U8*)SvPVX(dest) + len;
4326 /* Now process the remainder of the source, simultaneously
4327 * converting to upper and UTF-8.
4329 * To avoid extra tests in the loop body, and since the
4330 * loop is so simple, split out the rare Turkic case into
4333 #ifdef USE_LOCALE_CTYPE
4334 if ( UNLIKELY(PL_in_utf8_turkic_locale)
4335 && UNLIKELY(IN_LC_RUNTIME(LC_CTYPE)))
4337 for (; s < send; s++) {
4339 *d++ = UTF8_TWO_BYTE_HI(LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE);
4340 *d++ = UTF8_TWO_BYTE_LO(LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE);
4343 (void) _to_upper_title_latin1(*s, d, &len, 'S');
4350 for (; s < send; s++) {
4351 (void) _to_upper_title_latin1(*s, d, &len, 'S');
4355 /* Here have processed the whole source; no need to
4356 * continue with the outer loop. Each character has been
4357 * converted to upper case and converted to UTF-8. */
4359 } /* End of processing all latin1-style chars */
4360 } /* End of processing all chars */
4361 } /* End of source is not empty */
4363 if (source != dest) {
4364 *d = '\0'; /* Here d points to 1 after last char, add NUL */
4365 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
4367 } /* End of isn't utf8 */
4368 #ifdef USE_LOCALE_CTYPE
4369 if (IN_LC_RUNTIME(LC_CTYPE)) {
4374 if (dest != source && SvTAINTED(source))
4389 bool has_turkic_I = FALSE;
4393 if ( SvPADTMP(source)
4394 && !SvREADONLY(source) && SvPOK(source)
4397 #ifdef USE_LOCALE_CTYPE
4399 && ( LIKELY(! IN_LC_RUNTIME(LC_CTYPE))
4400 || LIKELY(! PL_in_utf8_turkic_locale))
4406 /* We can convert in place, as, outside of Turkic UTF-8 locales,
4407 * lowercasing anything in the latin1 range (or else DO_UTF8 would have
4408 * been on) doesn't lengthen it. */
4410 s = d = (U8*)SvPV_force_nomg(source, len);
4417 s = (const U8*)SvPV_nomg_const(source, len);
4420 SvUPGRADE(dest, SVt_PV);
4421 d = (U8*)SvGROW(dest, min);
4422 (void)SvPOK_only(dest);
4427 #ifdef USE_LOCALE_CTYPE
4429 if (IN_LC_RUNTIME(LC_CTYPE)) {
4432 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
4434 /* Lowercasing in a Turkic locale can cause non-UTF-8 to need to become
4435 * UTF-8 for the single case of the character 'I' */
4436 if ( UNLIKELY(PL_in_utf8_turkic_locale)
4437 && ! DO_UTF8(source)
4438 && (next_I = (U8 *) memchr(s, 'I', len)))
4441 const U8 *const send = s + len;
4446 next_I = (U8 *) memchr(next_I + 1, 'I',
4447 send - (next_I + 1));
4448 } while (next_I != NULL);
4450 /* Except for the 'I', in UTF-8 strings, the lower case of a
4451 * character below 256 occupies the same number of bytes as the
4452 * original. Therefore, the space needed is the original length
4453 * plus I_count plus the number of characters that become two bytes
4454 * when converted to UTF-8 */
4455 sv_utf8_upgrade_flags_grow(dest, 0, len
4457 + variant_under_utf8_count(s, send)
4458 + 1 /* Trailing NUL */ );
4459 d = (U8*)SvPVX(dest);
4460 has_turkic_I = TRUE;
4466 /* Overloaded values may have toggled the UTF-8 flag on source, so we need
4467 to check DO_UTF8 again here. */
4469 if (DO_UTF8(source)) {
4470 const U8 *const send = s + len;
4471 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
4472 bool remove_dot_above = FALSE;
4475 const STRLEN u = UTF8SKIP(s);
4478 #ifdef USE_LOCALE_CTYPE
4480 _toLOWER_utf8_flags(s, send, tmpbuf, &ulen, IN_LC_RUNTIME(LC_CTYPE));
4482 /* If we are in a Turkic locale, we have to do more work. As noted
4483 * in the comments for lcfirst, there is a special case if a 'I'
4484 * is in a grapheme with COMBINING DOT ABOVE UTF8. It turns into a
4485 * 'i', and the DOT must be removed. We check for that situation,
4486 * and set a flag if the DOT is there. Then each time through the
4487 * loop, we have to see if we need to remove the next DOT above,
4488 * and if so, do it. We know that there is a DOT because
4489 * _toLOWER_utf8_flags() wouldn't have returned 'i' unless there
4490 * was one in a proper position. */
4491 if ( UNLIKELY(PL_in_utf8_turkic_locale)
4492 && IN_LC_RUNTIME(LC_CTYPE))
4494 if ( UNLIKELY(remove_dot_above)
4495 && memBEGINs(tmpbuf, sizeof(tmpbuf), COMBINING_DOT_ABOVE_UTF8))
4498 remove_dot_above = FALSE;
4501 else if (UNLIKELY(*s == 'I' && tmpbuf[0] == 'i')) {
4502 remove_dot_above = TRUE;
4506 PERL_UNUSED_VAR(remove_dot_above);
4508 _toLOWER_utf8_flags(s, send, tmpbuf, &ulen, 0);
4511 /* Here is where we would do context-sensitive actions for the
4512 * Greek final sigma. See the commit message for 86510fb15 for why
4513 * there isn't any */
4515 if (ulen > u && (SvLEN(dest) < (min += ulen - u))) {
4517 /* If the eventually required minimum size outgrows the
4518 * available space, we need to grow. */
4519 const UV o = d - (U8*)SvPVX_const(dest);
4521 /* If someone lowercases one million U+0130s we SvGROW() one
4522 * million times. Or we could try guessing how much to
4523 * allocate without allocating too much. Such is life.
4524 * Another option would be to grow an extra byte or two more
4525 * each time we need to grow, which would cut down the million
4526 * to 500K, with little waste */
4527 d = o + (U8*) SvGROW(dest, min);
4530 /* Copy the newly lowercased letter to the output buffer we're
4532 Copy(tmpbuf, d, ulen, U8);
4535 } /* End of looping through the source string */
4538 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
4539 } else { /* 'source' not utf8 */
4541 const U8 *const send = s + len;
4543 /* Use locale casing if in locale; regular style if not treating
4544 * latin1 as having case; otherwise the latin1 casing. Do the
4545 * whole thing in a tight loop, for speed, */
4546 #ifdef USE_LOCALE_CTYPE
4547 if (IN_LC_RUNTIME(LC_CTYPE)) {
4548 if (LIKELY( ! has_turkic_I)) {
4549 for (; s < send; d++, s++)
4550 *d = toLOWER_LC(*s);
4552 else { /* This is the only case where lc() converts 'dest'
4553 into UTF-8 from a non-UTF-8 'source' */
4554 for (; s < send; s++) {
4556 *d++ = UTF8_TWO_BYTE_HI(LATIN_SMALL_LETTER_DOTLESS_I);
4557 *d++ = UTF8_TWO_BYTE_LO(LATIN_SMALL_LETTER_DOTLESS_I);
4560 append_utf8_from_native_byte(toLOWER_LATIN1(*s), &d);
4567 if (! IN_UNI_8_BIT) {
4568 for (; s < send; d++, s++) {
4573 for (; s < send; d++, s++) {
4574 *d = toLOWER_LATIN1(*s);
4578 if (source != dest) {
4580 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
4583 #ifdef USE_LOCALE_CTYPE
4584 if (IN_LC_RUNTIME(LC_CTYPE)) {
4589 if (dest != source && SvTAINTED(source))
4598 SV * const sv = TOPs;
4600 const char *s = SvPV_const(sv,len);
4602 SvUTF8_off(TARG); /* decontaminate */
4605 SvUPGRADE(TARG, SVt_PV);
4606 SvGROW(TARG, (len * 2) + 1);
4610 STRLEN ulen = UTF8SKIP(s);
4611 bool to_quote = FALSE;
4613 if (UTF8_IS_INVARIANT(*s)) {
4614 if (_isQUOTEMETA(*s)) {
4618 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, s + len)) {
4620 #ifdef USE_LOCALE_CTYPE
4621 /* In locale, we quote all non-ASCII Latin1 chars.
4622 * Otherwise use the quoting rules */
4624 IN_LC_RUNTIME(LC_CTYPE)
4627 _isQUOTEMETA(EIGHT_BIT_UTF8_TO_NATIVE(*s, *(s + 1))))
4632 else if (is_QUOTEMETA_high(s)) {
4647 else if (IN_UNI_8_BIT) {
4649 if (_isQUOTEMETA(*s))
4655 /* For non UNI_8_BIT (and hence in locale) just quote all \W
4656 * including everything above ASCII */
4658 if (!isWORDCHAR_A(*s))
4664 SvCUR_set(TARG, d - SvPVX_const(TARG));
4665 (void)SvPOK_only_UTF8(TARG);
4668 sv_setpvn(TARG, s, len);
4684 U8 tmpbuf[UTF8_MAXBYTES_CASE + 1];
4685 #if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
4686 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
4687 || UNICODE_DOT_DOT_VERSION > 0)
4688 const bool full_folding = TRUE; /* This variable is here so we can easily
4689 move to more generality later */
4691 const bool full_folding = FALSE;
4693 const U8 flags = ( full_folding ? FOLD_FLAGS_FULL : 0 )
4694 #ifdef USE_LOCALE_CTYPE
4695 | ( IN_LC_RUNTIME(LC_CTYPE) ? FOLD_FLAGS_LOCALE : 0 )
4699 /* This is a facsimile of pp_lc, but with a thousand bugs thanks to me.
4700 * You are welcome(?) -Hugmeir
4708 s = (const U8*)SvPV_nomg_const(source, len);
4710 if (ckWARN(WARN_UNINITIALIZED))
4711 report_uninit(source);
4718 SvUPGRADE(dest, SVt_PV);
4719 d = (U8*)SvGROW(dest, min);
4720 (void)SvPOK_only(dest);
4726 #ifdef USE_LOCALE_CTYPE
4728 if ( IN_LC_RUNTIME(LC_CTYPE) ) { /* Under locale */
4729 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
4734 if (DO_UTF8(source)) { /* UTF-8 flagged string. */
4736 const STRLEN u = UTF8SKIP(s);
4739 _toFOLD_utf8_flags(s, send, tmpbuf, &ulen, flags);
4741 if (ulen > u && (SvLEN(dest) < (min += ulen - u))) {
4742 const UV o = d - (U8*)SvPVX_const(dest);
4743 d = o + (U8*) SvGROW(dest, min);
4746 Copy(tmpbuf, d, ulen, U8);
4751 } /* Unflagged string */
4753 #ifdef USE_LOCALE_CTYPE
4754 if ( IN_LC_RUNTIME(LC_CTYPE) ) { /* Under locale */
4755 if (IN_UTF8_CTYPE_LOCALE) {
4756 goto do_uni_folding;
4758 for (; s < send; d++, s++)
4759 *d = (U8) toFOLD_LC(*s);
4763 if ( !IN_UNI_8_BIT ) { /* Under nothing, or bytes */
4764 for (; s < send; d++, s++)
4768 #ifdef USE_LOCALE_CTYPE
4771 /* For ASCII and the Latin-1 range, there's potentially three
4772 * troublesome folds:
4773 * \x{DF} (\N{LATIN SMALL LETTER SHARP S}), which under full
4774 * casefolding becomes 'ss';
4775 * \x{B5} (\N{MICRO SIGN}), which under any fold becomes
4776 * \x{3BC} (\N{GREEK SMALL LETTER MU})
4777 * I only in Turkic locales, this folds to \x{131}
4778 * \N{LATIN SMALL LETTER DOTLESS I}
4779 * For the rest, the casefold is their lowercase. */
4780 for (; s < send; d++, s++) {
4781 if ( UNLIKELY(*s == MICRO_SIGN)
4782 #ifdef USE_LOCALE_CTYPE
4783 || ( UNLIKELY(PL_in_utf8_turkic_locale)
4784 && UNLIKELY(IN_LC_RUNTIME(LC_CTYPE))
4785 && UNLIKELY(*s == 'I'))
4788 Size_t extra = send - s
4789 + variant_under_utf8_count(s, send);
4791 /* \N{MICRO SIGN}'s casefold is \N{GREEK SMALL LETTER MU},
4792 * and 'I' in Turkic locales is \N{LATIN SMALL LETTER
4793 * DOTLESS I} both of which are outside of the latin-1
4794 * range. There's a couple of ways to deal with this -- khw
4795 * discusses them in pp_lc/uc, so go there :) What we do
4796 * here is upgrade what we had already casefolded, then
4797 * enter an inner loop that appends the rest of the
4798 * characters as UTF-8.
4800 * First we calculate the needed size of the upgraded dest
4801 * beyond what's been processed already (the upgrade
4802 * function figures that out). Except for the 'I' in
4803 * Turkic locales, in UTF-8 strings, the fold case of a
4804 * character below 256 occupies the same number of bytes as
4805 * the original (even the Sharp S). Therefore, the space
4806 * needed is the number of bytes remaining plus the number
4807 * of characters that become two bytes when converted to
4808 * UTF-8 plus, in turkish locales, the number of 'I's */
4810 if (UNLIKELY(*s == 'I')) {
4811 const U8 * s_peek = s;
4816 s_peek = (U8 *) memchr(s_peek + 1, 'I',
4817 send - (s_peek + 1));
4818 } while (s_peek != NULL);
4821 /* Growing may move things, so have to save and recalculate
4823 len = d - (U8*)SvPVX_const(dest);
4824 SvCUR_set(dest, len);
4825 len = sv_utf8_upgrade_flags_grow(dest,
4826 SV_GMAGIC|SV_FORCE_UTF8_UPGRADE,
4828 + 1 /* Trailing NUL */ );
4829 d = (U8*)SvPVX(dest) + len;
4832 *d++ = UTF8_TWO_BYTE_HI(LATIN_SMALL_LETTER_DOTLESS_I);
4833 *d++ = UTF8_TWO_BYTE_LO(LATIN_SMALL_LETTER_DOTLESS_I);
4836 *d++ = UTF8_TWO_BYTE_HI(GREEK_SMALL_LETTER_MU);
4837 *d++ = UTF8_TWO_BYTE_LO(GREEK_SMALL_LETTER_MU);
4841 for (; s < send; s++) {
4843 _to_uni_fold_flags(*s, d, &ulen, flags);
4848 else if ( UNLIKELY(*s == LATIN_SMALL_LETTER_SHARP_S)
4851 /* Under full casefolding, LATIN SMALL LETTER SHARP S
4852 * becomes "ss", which may require growing the SV. */
4853 if (SvLEN(dest) < ++min) {
4854 const UV o = d - (U8*)SvPVX_const(dest);
4855 d = o + (U8*) SvGROW(dest, min);
4860 else { /* Else, the fold is the lower case */
4861 *d = toLOWER_LATIN1(*s);
4867 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
4869 #ifdef USE_LOCALE_CTYPE
4870 if (IN_LC_RUNTIME(LC_CTYPE)) {
4875 if (SvTAINTED(source))
4885 dSP; dMARK; dORIGMARK;
4886 AV *const av = MUTABLE_AV(POPs);
4887 const I32 lval = (PL_op->op_flags & OPf_MOD || LVRET);
4889 if (SvTYPE(av) == SVt_PVAV) {
4890 const bool localizing = PL_op->op_private & OPpLVAL_INTRO;
4891 bool can_preserve = FALSE;
4897 can_preserve = SvCANEXISTDELETE(av);
4900 if (lval && localizing) {
4903 for (svp = MARK + 1; svp <= SP; svp++) {
4904 const SSize_t elem = SvIV(*svp);
4908 if (max > AvMAX(av))
4912 while (++MARK <= SP) {
4914 SSize_t elem = SvIV(*MARK);
4915 bool preeminent = TRUE;
4917 if (localizing && can_preserve) {
4918 /* If we can determine whether the element exist,
4919 * Try to preserve the existenceness of a tied array
4920 * element by using EXISTS and DELETE if possible.
4921 * Fallback to FETCH and STORE otherwise. */
4922 preeminent = av_exists(av, elem);
4925 svp = av_fetch(av, elem, lval);
4928 DIE(aTHX_ PL_no_aelem, elem);
4931 save_aelem(av, elem, svp);
4933 SAVEADELETE(av, elem);
4936 *MARK = svp ? *svp : &PL_sv_undef;
4939 if (GIMME_V != G_ARRAY) {
4941 *++MARK = SP > ORIGMARK ? *SP : &PL_sv_undef;
4950 AV *const av = MUTABLE_AV(POPs);
4951 I32 lval = (PL_op->op_flags & OPf_MOD);
4952 SSize_t items = SP - MARK;
4954 if (PL_op->op_private & OPpMAYBE_LVSUB) {
4955 const I32 flags = is_lvalue_sub();
4957 if (!(flags & OPpENTERSUB_INARGS))
4958 /* diag_listed_as: Can't modify %s in %s */
4959 Perl_croak(aTHX_ "Can't modify index/value array slice in list assignment");
4966 *(MARK+items*2-1) = *(MARK+items);
4972 while (++MARK <= SP) {
4975 svp = av_fetch(av, SvIV(*MARK), lval);
4977 if (!svp || !*svp || *svp == &PL_sv_undef) {
4978 DIE(aTHX_ PL_no_aelem, SvIV(*MARK));
4980 *MARK = sv_mortalcopy(*MARK);
4982 *++MARK = svp ? *svp : &PL_sv_undef;
4984 if (GIMME_V != G_ARRAY) {
4985 MARK = SP - items*2;
4986 *++MARK = items > 0 ? *SP : &PL_sv_undef;
4996 AV *array = MUTABLE_AV(POPs);
4997 const U8 gimme = GIMME_V;
4998 IV *iterp = Perl_av_iter_p(aTHX_ array);
4999 const IV current = (*iterp)++;
5001 if (current > av_top_index(array)) {
5003 if (gimme == G_SCALAR)
5011 if (gimme == G_ARRAY) {
5012 SV **const element = av_fetch(array, current, 0);
5013 PUSHs(element ? *element : &PL_sv_undef);
5018 /* also used for: pp_avalues()*/
5022 AV *array = MUTABLE_AV(POPs);
5023 const U8 gimme = GIMME_V;
5025 *Perl_av_iter_p(aTHX_ array) = 0;
5027 if (gimme == G_SCALAR) {
5029 PUSHi(av_count(array));
5031 else if (gimme == G_ARRAY) {
5032 if (UNLIKELY(PL_op->op_private & OPpMAYBE_LVSUB)) {
5033 const I32 flags = is_lvalue_sub();
5034 if (flags && !(flags & OPpENTERSUB_INARGS))
5035 /* diag_listed_as: Can't modify %s in %s */
5037 "Can't modify keys on array in list assignment");
5040 IV n = av_top_index(array);
5045 if ( PL_op->op_type == OP_AKEYS
5046 || ( PL_op->op_type == OP_AVHVSWITCH
5047 && (PL_op->op_private & 3) + OP_AEACH == OP_AKEYS ))
5049 for (i = 0; i <= n; i++) {
5054 for (i = 0; i <= n; i++) {
5055 SV *const *const elem = Perl_av_fetch(aTHX_ array, i, 0);
5056 PUSHs(elem ? *elem : &PL_sv_undef);
5064 /* Associative arrays. */
5069 HV * hash = MUTABLE_HV(POPs);
5071 const U8 gimme = GIMME_V;
5073 entry = hv_iternext(hash);
5077 SV* const sv = hv_iterkeysv(entry);
5079 if (gimme == G_ARRAY) {
5081 val = hv_iterval(hash, entry);
5085 else if (gimme == G_SCALAR)
5092 S_do_delete_local(pTHX)
5095 const U8 gimme = GIMME_V;
5098 const bool sliced = !!(PL_op->op_private & OPpSLICE);
5099 SV **unsliced_keysv = sliced ? NULL : sp--;
5100 SV * const osv = POPs;
5101 SV **mark = sliced ? PL_stack_base + POPMARK : unsliced_keysv-1;
5103 const bool tied = SvRMAGICAL(osv)
5104 && mg_find((const SV *)osv, PERL_MAGIC_tied);
5105 const bool can_preserve = SvCANEXISTDELETE(osv);
5106 const U32 type = SvTYPE(osv);
5107 SV ** const end = sliced ? SP : unsliced_keysv;
5109 if (type == SVt_PVHV) { /* hash element */
5110 HV * const hv = MUTABLE_HV(osv);
5111 while (++MARK <= end) {
5112 SV * const keysv = *MARK;
5114 bool preeminent = TRUE;
5116 preeminent = hv_exists_ent(hv, keysv, 0);
5118 HE *he = hv_fetch_ent(hv, keysv, 1, 0);
5125 sv = hv_delete_ent(hv, keysv, 0, 0);
5127 SvREFCNT_inc_simple_void(sv); /* De-mortalize */
5130 if (!sv) DIE(aTHX_ PL_no_helem_sv, SVfARG(keysv));
5131 save_helem_flags(hv, keysv, &sv, SAVEf_KEEPOLDELEM);
5133 *MARK = sv_mortalcopy(sv);
5139 SAVEHDELETE(hv, keysv);
5140 *MARK = &PL_sv_undef;
5144 else if (type == SVt_PVAV) { /* array element */
5145 if (PL_op->op_flags & OPf_SPECIAL) {
5146 AV * const av = MUTABLE_AV(osv);
5147 while (++MARK <= end) {
5148 SSize_t idx = SvIV(*MARK);
5150 bool preeminent = TRUE;
5152 preeminent = av_exists(av, idx);
5154 SV **svp = av_fetch(av, idx, 1);
5161 sv = av_delete(av, idx, 0);
5163 SvREFCNT_inc_simple_void(sv); /* De-mortalize */
5166 save_aelem_flags(av, idx, &sv, SAVEf_KEEPOLDELEM);
5168 *MARK = sv_mortalcopy(sv);
5174 SAVEADELETE(av, idx);
5175 *MARK = &PL_sv_undef;
5180 DIE(aTHX_ "panic: avhv_delete no longer supported");
5183 DIE(aTHX_ "Not a HASH reference");
5185 if (gimme == G_VOID)
5187 else if (gimme == G_SCALAR) {
5192 *++MARK = &PL_sv_undef;
5196 else if (gimme != G_VOID)
5197 PUSHs(*unsliced_keysv);
5208 if (PL_op->op_private & OPpLVAL_INTRO)
5209 return do_delete_local();
5212 discard = (gimme == G_VOID) ? G_DISCARD : 0;
5214 if (PL_op->op_private & (OPpSLICE|OPpKVSLICE)) {
5216 HV * const hv = MUTABLE_HV(POPs);
5217 const U32 hvtype = SvTYPE(hv);
5219 if (PL_op->op_private & OPpKVSLICE) {
5220 SSize_t items = SP - MARK;
5224 *(MARK+items*2-1) = *(MARK+items);
5231 if (hvtype == SVt_PVHV) { /* hash element */
5232 while ((MARK += (1+skip)) <= SP) {
5233 SV * const sv = hv_delete_ent(hv, *(MARK-skip), discard, 0);
5234 *MARK = sv ? sv : &PL_sv_undef;
5237 else if (hvtype == SVt_PVAV) { /* array element */
5238 if (PL_op->op_flags & OPf_SPECIAL) {
5239 while ((MARK += (1+skip)) <= SP) {
5240 SV * const sv = av_delete(MUTABLE_AV(hv), SvIV(*(MARK-skip)), discard);
5241 *MARK = sv ? sv : &PL_sv_undef;
5246 DIE(aTHX_ "Not a HASH reference");
5249 else if (gimme == G_SCALAR) {
5254 *++MARK = &PL_sv_undef;
5260 HV * const hv = MUTABLE_HV(POPs);
5262 if (SvTYPE(hv) == SVt_PVHV)
5263 sv = hv_delete_ent(hv, keysv, discard, 0);
5264 else if (SvTYPE(hv) == SVt_PVAV) {
5265 if (PL_op->op_flags & OPf_SPECIAL)
5266 sv = av_delete(MUTABLE_AV(hv), SvIV(keysv), discard);
5268 DIE(aTHX_ "panic: avhv_delete no longer supported");
5271 DIE(aTHX_ "Not a HASH reference");
5286 if (UNLIKELY( PL_op->op_private & OPpEXISTS_SUB )) {
5288 SV * const sv = POPs;
5289 CV * const cv = sv_2cv(sv, &hv, &gv, 0);
5292 if (gv && isGV(gv) && GvCV(gv) && !GvCVGEN(gv))
5297 hv = MUTABLE_HV(POPs);
5298 if (LIKELY( SvTYPE(hv) == SVt_PVHV )) {
5299 if (hv_exists_ent(hv, tmpsv, 0))
5302 else if (SvTYPE(hv) == SVt_PVAV) {
5303 if (PL_op->op_flags & OPf_SPECIAL) { /* array element */
5304 if (av_exists(MUTABLE_AV(hv), SvIV(tmpsv)))
5309 DIE(aTHX_ "Not a HASH reference");
5316 dSP; dMARK; dORIGMARK;
5317 HV * const hv = MUTABLE_HV(POPs);
5318 const I32 lval = (PL_op->op_flags & OPf_MOD || LVRET);
5319 const bool localizing = PL_op->op_private & OPpLVAL_INTRO;
5320 bool can_preserve = FALSE;
5326 if (SvCANEXISTDELETE(hv))
5327 can_preserve = TRUE;
5330 while (++MARK <= SP) {
5331 SV * const keysv = *MARK;
5334 bool preeminent = TRUE;
5336 if (localizing && can_preserve) {
5337 /* If we can determine whether the element exist,
5338 * try to preserve the existenceness of a tied hash
5339 * element by using EXISTS and DELETE if possible.
5340 * Fallback to FETCH and STORE otherwise. */
5341 preeminent = hv_exists_ent(hv, keysv, 0);
5344 he = hv_fetch_ent(hv, keysv, lval, 0);
5345 svp = he ? &HeVAL(he) : NULL;
5348 if (!svp || !*svp || *svp == &PL_sv_undef) {
5349 DIE(aTHX_ PL_no_helem_sv, SVfARG(keysv));
5352 if (HvNAME_get(hv) && isGV_or_RVCV(*svp))
5353 save_gp(MUTABLE_GV(*svp), !(PL_op->op_flags & OPf_SPECIAL));
5354 else if (preeminent)
5355 save_helem_flags(hv, keysv, svp,
5356 (PL_op->op_flags & OPf_SPECIAL) ? 0 : SAVEf_SETMAGIC);
5358 SAVEHDELETE(hv, keysv);
5361 *MARK = svp && *svp ? *svp : &PL_sv_undef;
5363 if (GIMME_V != G_ARRAY) {
5365 *++MARK = SP > ORIGMARK ? *SP : &PL_sv_undef;
5374 HV * const hv = MUTABLE_HV(POPs);
5375 I32 lval = (PL_op->op_flags & OPf_MOD);
5376 SSize_t items = SP - MARK;
5378 if (PL_op->op_private & OPpMAYBE_LVSUB) {
5379 const I32 flags = is_lvalue_sub();
5381 if (!(flags & OPpENTERSUB_INARGS))
5382 /* diag_listed_as: Can't modify %s in %s */
5383 Perl_croak(aTHX_ "Can't modify key/value hash slice in %s assignment",
5384 GIMME_V == G_ARRAY ? "list" : "scalar");
5391 *(MARK+items*2-1) = *(MARK+items);
5397 while (++MARK <= SP) {
5398 SV * const keysv = *MARK;
5402 he = hv_fetch_ent(hv, keysv, lval, 0);
5403 svp = he ? &HeVAL(he) : NULL;
5406 if (!svp || !*svp || *svp == &PL_sv_undef) {
5407 DIE(aTHX_ PL_no_helem_sv, SVfARG(keysv));
5409 *MARK = sv_mortalcopy(*MARK);
5411 *++MARK = svp && *svp ? *svp : &PL_sv_undef;
5413 if (GIMME_V != G_ARRAY) {
5414 MARK = SP - items*2;
5415 *++MARK = items > 0 ? *SP : &PL_sv_undef;
5421 /* List operators. */
5425 I32 markidx = POPMARK;
5426 if (GIMME_V != G_ARRAY) {
5427 /* don't initialize mark here, EXTEND() may move the stack */
5430 EXTEND(SP, 1); /* in case no arguments, as in @empty */
5431 mark = PL_stack_base + markidx;
5433 *MARK = *SP; /* unwanted list, return last item */
5435 *MARK = &PL_sv_undef;
5445 SV ** const lastrelem = PL_stack_sp;
5446 SV ** const lastlelem = PL_stack_base + POPMARK;
5447 SV ** const firstlelem = PL_stack_base + POPMARK + 1;
5448 SV ** const firstrelem = lastlelem + 1;
5449 const U8 mod = PL_op->op_flags & OPf_MOD;
5451 const I32 max = lastrelem - lastlelem;
5454 if (GIMME_V != G_ARRAY) {
5455 if (lastlelem < firstlelem) {
5457 *firstlelem = &PL_sv_undef;
5460 I32 ix = SvIV(*lastlelem);
5463 if (ix < 0 || ix >= max)
5464 *firstlelem = &PL_sv_undef;
5466 *firstlelem = firstrelem[ix];
5473 SP = firstlelem - 1;
5477 for (lelem = firstlelem; lelem <= lastlelem; lelem++) {
5478 I32 ix = SvIV(*lelem);
5481 if (ix < 0 || ix >= max)
5482 *lelem = &PL_sv_undef;
5484 if (!(*lelem = firstrelem[ix]))
5485 *lelem = &PL_sv_undef;
5486 else if (mod && SvPADTMP(*lelem)) {
5487 *lelem = firstrelem[ix] = sv_mortalcopy(*lelem);
5498 const I32 items = SP - MARK;
5499 SV * const av = MUTABLE_SV(av_make(items, MARK+1));
5501 mXPUSHs((PL_op->op_flags & OPf_SPECIAL)
5502 ? newRV_noinc(av) : av);
5508 dSP; dMARK; dORIGMARK;
5509 HV* const hv = newHV();
5510 SV* const retval = sv_2mortal( PL_op->op_flags & OPf_SPECIAL
5511 ? newRV_noinc(MUTABLE_SV(hv))
5516 (MARK++, SvGMAGICAL(*MARK) ? sv_mortalcopy(*MARK) : *MARK);
5523 sv_setsv_nomg(val, *MARK);