3 * Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000,
4 * 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008 by Larry Wall and others
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
12 * 'It's a big house this, and very peculiar. Always a bit more
13 * to discover, and no knowing what you'll find round a corner.
14 * And Elves, sir!' --Samwise Gamgee
16 * [p.225 of _The Lord of the Rings_, II/i: "Many Meetings"]
19 /* This file contains general pp ("push/pop") functions that execute the
20 * opcodes that make up a perl program. A typical pp function expects to
21 * find its arguments on the stack, and usually pushes its results onto
22 * the stack, hence the 'pp' terminology. Each OP structure contains
23 * a pointer to the relevant pp_foo() function.
32 #include "regcharclass.h"
34 static const STRLEN small_mu_len = sizeof(GREEK_SMALL_LETTER_MU_UTF8) - 1;
35 static const STRLEN capital_iota_len = sizeof(GREEK_CAPITAL_LETTER_IOTA_UTF8) - 1;
37 /* variations on pp_null */
42 if (GIMME_V == G_SCALAR)
54 assert(SvTYPE(TARG) == SVt_PVCV);
69 CV * const protocv = PadnamePROTOCV(
70 PadlistNAMESARRAY(CvPADLIST(find_runcv(NULL)))[ARGTARG]
72 assert(SvTYPE(TARG) == SVt_PVCV);
74 if (CvISXSUB(protocv)) { /* constant */
75 /* XXX Should we clone it here? */
76 /* If this changes to use SAVECLEARSV, we can move the SAVECLEARSV
77 to introcv and remove the SvPADSTALE_off. */
78 SAVEPADSVANDMORTALIZE(ARGTARG);
79 PAD_SVl(ARGTARG) = SvREFCNT_inc_simple_NN(protocv);
82 if (CvROOT(protocv)) {
83 assert(CvCLONE(protocv));
84 assert(!CvCLONED(protocv));
86 cv_clone_into(protocv,(CV *)TARG);
87 SAVECLEARSV(PAD_SVl(ARGTARG));
94 /* In some cases this function inspects PL_op. If this function is called
95 for new op types, more bool parameters may need to be added in place of
98 When noinit is true, the absence of a gv will cause a retval of undef.
99 This is unrelated to the cv-to-gv assignment case.
103 S_rv2gv(pTHX_ SV *sv, const bool vivify_sv, const bool strict,
106 if (!isGV(sv) || SvFAKE(sv)) SvGETMAGIC(sv);
109 sv = amagic_deref_call(sv, to_gv_amg);
113 if (SvTYPE(sv) == SVt_PVIO) {
114 GV * const gv = MUTABLE_GV(sv_newmortal());
115 gv_init(gv, 0, "__ANONIO__", 10, 0);
116 GvIOp(gv) = MUTABLE_IO(sv);
117 SvREFCNT_inc_void_NN(sv);
120 else if (!isGV_with_GP(sv)) {
121 Perl_die(aTHX_ "Not a GLOB reference");
125 if (!isGV_with_GP(sv)) {
127 /* If this is a 'my' scalar and flag is set then vivify
130 if (vivify_sv && sv != &PL_sv_undef) {
134 Perl_croak_no_modify();
135 gv = MUTABLE_GV(newSV(0));
136 stash = CopSTASH(PL_curcop);
137 if (SvTYPE(stash) != SVt_PVHV) stash = NULL;
138 if (cUNOP->op_targ) {
139 SV * const namesv = PAD_SV(cUNOP->op_targ);
140 gv_init_sv(gv, stash, namesv, 0);
143 gv_init_pv(gv, stash, "__ANONIO__", 0);
145 prepare_SV_for_RV(sv);
146 SvRV_set(sv, MUTABLE_SV(gv));
151 if (PL_op->op_flags & OPf_REF || strict) {
152 Perl_die(aTHX_ PL_no_usym, "a symbol");
154 if (ckWARN(WARN_UNINITIALIZED))
160 if (!(sv = MUTABLE_SV(gv_fetchsv_nomg(
161 sv, GV_ADDMG, SVt_PVGV
170 (SvPOKp(sv) && SvCUR(sv)>32 ? "..." : ""),
174 if ((PL_op->op_private & (OPpLVAL_INTRO|OPpDONT_INIT_GV))
175 == OPpDONT_INIT_GV) {
176 /* We are the target of a coderef assignment. Return
177 the scalar unchanged, and let pp_sasssign deal with
181 sv = MUTABLE_SV(gv_fetchsv_nomg(sv, GV_ADD, SVt_PVGV));
183 /* FAKE globs in the symbol table cause weird bugs (#77810) */
187 if (SvFAKE(sv) && !(PL_op->op_private & OPpALLOW_FAKE)) {
188 SV *newsv = sv_newmortal();
189 sv_setsv_flags(newsv, sv, 0);
201 sv, PL_op->op_private & OPpDEREF,
202 PL_op->op_private & HINT_STRICT_REFS,
203 ((PL_op->op_flags & OPf_SPECIAL) && !(PL_op->op_flags & OPf_MOD))
204 || PL_op->op_type == OP_READLINE
206 if (PL_op->op_private & OPpLVAL_INTRO)
207 save_gp(MUTABLE_GV(sv), !(PL_op->op_flags & OPf_SPECIAL));
212 /* Helper function for pp_rv2sv and pp_rv2av */
214 Perl_softref2xv(pTHX_ SV *const sv, const char *const what,
215 const svtype type, SV ***spp)
219 PERL_ARGS_ASSERT_SOFTREF2XV;
221 if (PL_op->op_private & HINT_STRICT_REFS) {
223 Perl_die(aTHX_ PL_no_symref_sv, sv,
224 (SvPOKp(sv) && SvCUR(sv)>32 ? "..." : ""), what);
226 Perl_die(aTHX_ PL_no_usym, what);
230 PL_op->op_flags & OPf_REF
232 Perl_die(aTHX_ PL_no_usym, what);
233 if (ckWARN(WARN_UNINITIALIZED))
235 if (type != SVt_PV && GIMME_V == G_ARRAY) {
239 **spp = &PL_sv_undef;
242 if ((PL_op->op_flags & OPf_SPECIAL) &&
243 !(PL_op->op_flags & OPf_MOD))
245 if (!(gv = gv_fetchsv_nomg(sv, GV_ADDMG, type)))
247 **spp = &PL_sv_undef;
252 gv = gv_fetchsv_nomg(sv, GV_ADD, type);
265 sv = amagic_deref_call(sv, to_sv_amg);
269 if (SvTYPE(sv) >= SVt_PVAV)
270 DIE(aTHX_ "Not a SCALAR reference");
275 if (!isGV_with_GP(gv)) {
276 gv = Perl_softref2xv(aTHX_ sv, "a SCALAR", SVt_PV, &sp);
282 if (PL_op->op_flags & OPf_MOD) {
283 if (PL_op->op_private & OPpLVAL_INTRO) {
284 if (cUNOP->op_first->op_type == OP_NULL)
285 sv = save_scalar(MUTABLE_GV(TOPs));
287 sv = save_scalar(gv);
289 Perl_croak(aTHX_ "%s", PL_no_localize_ref);
291 else if (PL_op->op_private & OPpDEREF)
292 sv = vivify_ref(sv, PL_op->op_private & OPpDEREF);
294 SPAGAIN; /* in case chasing soft refs reallocated the stack */
302 AV * const av = MUTABLE_AV(TOPs);
303 const I32 lvalue = PL_op->op_flags & OPf_MOD || LVRET;
305 SV ** const svp = Perl_av_arylen_p(aTHX_ MUTABLE_AV(av));
307 *svp = newSV_type(SVt_PVMG);
308 sv_magic(*svp, MUTABLE_SV(av), PERL_MAGIC_arylen, NULL, 0);
312 SETs(sv_2mortal(newSViv(AvFILL(MUTABLE_AV(av)))));
321 if (PL_op->op_flags & OPf_MOD || LVRET) {
322 SV * const ret = sv_2mortal(newSV_type(SVt_PVLV));/* Not TARG RT#67838 */
323 sv_magic(ret, NULL, PERL_MAGIC_pos, NULL, 0);
325 LvTARG(ret) = SvREFCNT_inc_simple(sv);
326 SETs(ret); /* no SvSETMAGIC */
329 const MAGIC * const mg = mg_find_mglob(sv);
330 if (mg && mg->mg_len != -1) {
331 STRLEN i = mg->mg_len;
332 if (PL_op->op_private & OPpTRUEBOOL)
333 SETs(i ? &PL_sv_yes : &PL_sv_zero);
336 if (mg->mg_flags & MGf_BYTES && DO_UTF8(sv))
337 i = sv_pos_b2u_flags(sv, i, SV_GMAGIC|SV_CONST_RETURN);
352 const I32 flags = (PL_op->op_flags & OPf_SPECIAL)
354 : ((PL_op->op_private & (OPpLVAL_INTRO|OPpMAY_RETURN_CONSTANT))
355 == OPpMAY_RETURN_CONSTANT)
358 /* We usually try to add a non-existent subroutine in case of AUTOLOAD. */
359 /* (But not in defined().) */
361 CV *cv = sv_2cv(TOPs, &stash_unused, &gv, flags);
363 else if ((flags == (GV_ADD|GV_NOEXPAND)) && gv && SvROK(gv)) {
364 cv = SvTYPE(SvRV(gv)) == SVt_PVCV
365 ? MUTABLE_CV(SvRV(gv))
369 cv = MUTABLE_CV(&PL_sv_undef);
370 SETs(MUTABLE_SV(cv));
380 SV *ret = &PL_sv_undef;
382 if (SvGMAGICAL(TOPs)) SETs(sv_mortalcopy(TOPs));
383 if (SvPOK(TOPs) && SvCUR(TOPs) >= 7) {
384 const char * s = SvPVX_const(TOPs);
385 if (memBEGINs(s, SvCUR(TOPs), "CORE::")) {
386 const int code = keyword(s + 6, SvCUR(TOPs) - 6, 1);
388 DIE(aTHX_ "Can't find an opnumber for \"%" UTF8f "\"",
389 UTF8fARG(SvFLAGS(TOPs) & SVf_UTF8, SvCUR(TOPs)-6, s+6));
391 SV * const sv = core_prototype(NULL, s + 6, code, NULL);
397 cv = sv_2cv(TOPs, &stash, &gv, 0);
399 ret = newSVpvn_flags(
400 CvPROTO(cv), CvPROTOLEN(cv), SVs_TEMP | SvUTF8(cv)
410 CV *cv = MUTABLE_CV(PAD_SV(PL_op->op_targ));
412 cv = MUTABLE_CV(sv_2mortal(MUTABLE_SV(cv_clone(cv))));
414 PUSHs(MUTABLE_SV(cv));
428 if (GIMME_V != G_ARRAY) {
434 *MARK = &PL_sv_undef;
436 *MARK = refto(*MARK);
440 EXTEND_MORTAL(SP - MARK);
442 *MARK = refto(*MARK);
447 S_refto(pTHX_ SV *sv)
451 PERL_ARGS_ASSERT_REFTO;
453 if (SvTYPE(sv) == SVt_PVLV && LvTYPE(sv) == 'y') {
456 if (!(sv = LvTARG(sv)))
459 SvREFCNT_inc_void_NN(sv);
461 else if (SvTYPE(sv) == SVt_PVAV) {
462 if (!AvREAL((const AV *)sv) && AvREIFY((const AV *)sv))
463 av_reify(MUTABLE_AV(sv));
465 SvREFCNT_inc_void_NN(sv);
467 else if (SvPADTMP(sv)) {
470 else if (UNLIKELY(SvSMAGICAL(sv) && mg_find(sv, PERL_MAGIC_nonelem)))
471 sv_unmagic(SvREFCNT_inc_simple_NN(sv), PERL_MAGIC_nonelem);
474 SvREFCNT_inc_void_NN(sv);
477 sv_upgrade(rv, SVt_IV);
486 SV * const sv = TOPs;
494 /* op is in boolean context? */
495 if ( (PL_op->op_private & OPpTRUEBOOL)
496 || ( (PL_op->op_private & OPpMAYBE_TRUEBOOL)
497 && block_gimme() == G_VOID))
499 /* refs are always true - unless it's to an object blessed into a
500 * class with a false name, i.e. "0". So we have to check for
501 * that remote possibility. The following is is basically an
502 * unrolled SvTRUE(sv_reftype(rv)) */
503 SV * const rv = SvRV(sv);
505 HV *stash = SvSTASH(rv);
506 HEK *hek = HvNAME_HEK(stash);
508 I32 len = HEK_LEN(hek);
509 /* bail out and do it the hard way? */
512 || (len == 1 && HEK_KEY(hek)[0] == '0')
525 sv_ref(TARG, SvRV(sv), TRUE);
541 stash = CopSTASH(PL_curcop);
542 if (SvTYPE(stash) != SVt_PVHV)
543 Perl_croak(aTHX_ "Attempt to bless into a freed package");
546 SV * const ssv = POPs;
550 if (!ssv) goto curstash;
553 if (!SvAMAGIC(ssv)) {
555 Perl_croak(aTHX_ "Attempt to bless into a reference");
557 /* SvAMAGIC is on here, but it only means potentially overloaded,
558 so after stringification: */
559 ptr = SvPV_nomg_const(ssv,len);
560 /* We need to check the flag again: */
561 if (!SvAMAGIC(ssv)) goto frog;
563 else ptr = SvPV_nomg_const(ssv,len);
565 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
566 "Explicit blessing to '' (assuming package main)");
567 stash = gv_stashpvn(ptr, len, GV_ADD|SvUTF8(ssv));
570 (void)sv_bless(TOPs, stash);
580 const char * const elem = SvPV_const(sv, len);
581 GV * const gv = MUTABLE_GV(TOPs);
586 /* elem will always be NUL terminated. */
589 if (memEQs(elem, len, "ARRAY"))
591 tmpRef = MUTABLE_SV(GvAV(gv));
592 if (tmpRef && !AvREAL((const AV *)tmpRef)
593 && AvREIFY((const AV *)tmpRef))
594 av_reify(MUTABLE_AV(tmpRef));
598 if (memEQs(elem, len, "CODE"))
599 tmpRef = MUTABLE_SV(GvCVu(gv));
602 if (memEQs(elem, len, "FILEHANDLE")) {
603 tmpRef = MUTABLE_SV(GvIOp(gv));
606 if (memEQs(elem, len, "FORMAT"))
607 tmpRef = MUTABLE_SV(GvFORM(gv));
610 if (memEQs(elem, len, "GLOB"))
611 tmpRef = MUTABLE_SV(gv);
614 if (memEQs(elem, len, "HASH"))
615 tmpRef = MUTABLE_SV(GvHV(gv));
618 if (memEQs(elem, len, "IO"))
619 tmpRef = MUTABLE_SV(GvIOp(gv));
622 if (memEQs(elem, len, "NAME"))
623 sv = newSVhek(GvNAME_HEK(gv));
626 if (memEQs(elem, len, "PACKAGE")) {
627 const HV * const stash = GvSTASH(gv);
628 const HEK * const hek = stash ? HvNAME_HEK(stash) : NULL;
629 sv = hek ? newSVhek(hek) : newSVpvs("__ANON__");
633 if (memEQs(elem, len, "SCALAR"))
648 /* Pattern matching */
656 if (len == 0 || len > I32_MAX || !SvPOK(sv) || SvUTF8(sv) || SvVALID(sv)) {
657 /* Historically, study was skipped in these cases. */
662 /* Make study a no-op. It's no longer useful and its existence
663 complicates matters elsewhere. */
669 /* also used for: pp_transr() */
676 if (PL_op->op_flags & OPf_STACKED)
681 sv = PAD_SV(ARGTARG);
686 if(PL_op->op_type == OP_TRANSR) {
688 const char * const pv = SvPV(sv,len);
689 SV * const newsv = newSVpvn_flags(pv, len, SVs_TEMP|SvUTF8(sv));
694 Size_t i = do_trans(sv);
700 /* Lvalue operators. */
703 S_do_chomp(pTHX_ SV *retval, SV *sv, bool chomping)
709 PERL_ARGS_ASSERT_DO_CHOMP;
711 if (chomping && (RsSNARF(PL_rs) || RsRECORD(PL_rs)))
713 if (SvTYPE(sv) == SVt_PVAV) {
715 AV *const av = MUTABLE_AV(sv);
716 const I32 max = AvFILL(av);
718 for (i = 0; i <= max; i++) {
719 sv = MUTABLE_SV(av_fetch(av, i, FALSE));
720 if (sv && ((sv = *(SV**)sv), sv != &PL_sv_undef))
721 count += do_chomp(retval, sv, chomping);
725 else if (SvTYPE(sv) == SVt_PVHV) {
726 HV* const hv = MUTABLE_HV(sv);
728 (void)hv_iterinit(hv);
729 while ((entry = hv_iternext(hv)))
730 count += do_chomp(retval, hv_iterval(hv,entry), chomping);
733 else if (SvREADONLY(sv)) {
734 Perl_croak_no_modify();
740 char *temp_buffer = NULL;
745 goto nope_free_nothing;
747 while (len && s[-1] == '\n') {
754 STRLEN rslen, rs_charlen;
755 const char *rsptr = SvPV_const(PL_rs, rslen);
757 rs_charlen = SvUTF8(PL_rs)
761 if (SvUTF8(PL_rs) != SvUTF8(sv)) {
762 /* Assumption is that rs is shorter than the scalar. */
764 /* RS is utf8, scalar is 8 bit. */
766 temp_buffer = (char*)bytes_from_utf8((U8*)rsptr,
769 /* Cannot downgrade, therefore cannot possibly match.
770 At this point, temp_buffer is not alloced, and
771 is the buffer inside PL_rs, so dont free it.
773 assert (temp_buffer == rsptr);
779 /* RS is 8 bit, scalar is utf8. */
780 temp_buffer = (char*)bytes_to_utf8((U8*)rsptr, &rslen);
794 if (memNE(s, rsptr, rslen))
799 SvPV_force_nomg_nolen(sv);
806 Safefree(temp_buffer);
808 SvREFCNT_dec(svrecode);
812 if (len && (!SvPOK(sv) || SvIsCOW(sv)))
813 s = SvPV_force_nomg(sv, len);
816 char * const send = s + len;
817 char * const start = s;
819 while (s > start && UTF8_IS_CONTINUATION(*s))
821 if (is_utf8_string((U8*)s, send - s)) {
822 sv_setpvn(retval, s, send - s);
824 SvCUR_set(sv, s - start);
834 sv_setpvn(retval, s, 1);
848 /* also used for: pp_schomp() */
853 const bool chomping = PL_op->op_type == OP_SCHOMP;
855 const size_t count = do_chomp(TARG, TOPs, chomping);
857 sv_setiv(TARG, count);
863 /* also used for: pp_chomp() */
867 dSP; dMARK; dTARGET; dORIGMARK;
868 const bool chomping = PL_op->op_type == OP_CHOMP;
872 count += do_chomp(TARG, *++MARK, chomping);
874 sv_setiv(TARG, count);
885 if (!PL_op->op_private) {
897 if (SvTHINKFIRST(sv))
898 sv_force_normal_flags(sv, SV_COW_DROP_PV|SV_IMMEDIATE_UNREF);
900 switch (SvTYPE(sv)) {
904 av_undef(MUTABLE_AV(sv));
907 hv_undef(MUTABLE_HV(sv));
910 if (cv_const_sv((const CV *)sv))
911 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
912 "Constant subroutine %" SVf " undefined",
913 SVfARG(CvANON((const CV *)sv)
914 ? newSVpvs_flags("(anonymous)", SVs_TEMP)
915 : sv_2mortal(newSVhek(
917 ? CvNAME_HEK((CV *)sv)
918 : GvENAME_HEK(CvGV((const CV *)sv))
923 /* let user-undef'd sub keep its identity */
924 cv_undef_flags(MUTABLE_CV(sv), CV_UNDEF_KEEP_NAME);
927 assert(isGV_with_GP(sv));
933 /* undef *Pkg::meth_name ... */
935 = GvCVu((const GV *)sv) && (stash = GvSTASH((const GV *)sv))
936 && HvENAME_get(stash);
938 if((stash = GvHV((const GV *)sv))) {
939 if(HvENAME_get(stash))
940 SvREFCNT_inc_simple_void_NN(sv_2mortal((SV *)stash));
944 SvREFCNT_inc_simple_void_NN(sv_2mortal(sv));
945 gp_free(MUTABLE_GV(sv));
947 GvGP_set(sv, gp_ref(gp));
948 #ifndef PERL_DONT_CREATE_GVSV
951 GvLINE(sv) = CopLINE(PL_curcop);
952 GvEGV(sv) = MUTABLE_GV(sv);
956 mro_package_moved(NULL, stash, (const GV *)sv, 0);
958 /* undef *Foo::ISA */
959 if( strEQ(GvNAME((const GV *)sv), "ISA")
960 && (stash = GvSTASH((const GV *)sv))
961 && (method_changed || HvENAME(stash)) )
962 mro_isa_changed_in(stash);
963 else if(method_changed)
964 mro_method_changed_in(
965 GvSTASH((const GV *)sv)
971 if (SvTYPE(sv) >= SVt_PV && SvPVX_const(sv) && SvLEN(sv)) {
985 /* common "slow" code for pp_postinc and pp_postdec */
988 S_postincdec_common(pTHX_ SV *sv, SV *targ)
992 PL_op->op_type == OP_POSTINC || PL_op->op_type == OP_I_POSTINC;
995 TARG = sv_newmortal();
1002 /* special case for undef: see thread at 2003-03/msg00536.html in archive */
1003 if (inc && !SvOK(TARG))
1010 /* also used for: pp_i_postinc() */
1017 /* special-case sv being a simple integer */
1018 if (LIKELY(((sv->sv_flags &
1019 (SVf_THINKFIRST|SVs_GMG|SVf_IVisUV|
1020 SVf_IOK|SVf_NOK|SVf_POK|SVp_NOK|SVp_POK|SVf_ROK))
1022 && SvIVX(sv) != IV_MAX)
1025 SvIV_set(sv, iv + 1);
1026 TARGi(iv, 0); /* arg not GMG, so can't be tainted */
1031 return S_postincdec_common(aTHX_ sv, TARG);
1035 /* also used for: pp_i_postdec() */
1042 /* special-case sv being a simple integer */
1043 if (LIKELY(((sv->sv_flags &
1044 (SVf_THINKFIRST|SVs_GMG|SVf_IVisUV|
1045 SVf_IOK|SVf_NOK|SVf_POK|SVp_NOK|SVp_POK|SVf_ROK))
1047 && SvIVX(sv) != IV_MIN)
1050 SvIV_set(sv, iv - 1);
1051 TARGi(iv, 0); /* arg not GMG, so can't be tainted */
1056 return S_postincdec_common(aTHX_ sv, TARG);
1060 /* Ordinary operators. */
1064 dSP; dATARGET; SV *svl, *svr;
1065 #ifdef PERL_PRESERVE_IVUV
1068 tryAMAGICbin_MG(pow_amg, AMGf_assign|AMGf_numeric);
1071 #ifdef PERL_PRESERVE_IVUV
1072 /* For integer to integer power, we do the calculation by hand wherever
1073 we're sure it is safe; otherwise we call pow() and try to convert to
1074 integer afterwards. */
1075 if (SvIV_please_nomg(svr) && SvIV_please_nomg(svl)) {
1083 const IV iv = SvIVX(svr);
1087 goto float_it; /* Can't do negative powers this way. */
1091 baseuok = SvUOK(svl);
1093 baseuv = SvUVX(svl);
1095 const IV iv = SvIVX(svl);
1098 baseuok = TRUE; /* effectively it's a UV now */
1100 baseuv = -iv; /* abs, baseuok == false records sign */
1103 /* now we have integer ** positive integer. */
1106 /* foo & (foo - 1) is zero only for a power of 2. */
1107 if (!(baseuv & (baseuv - 1))) {
1108 /* We are raising power-of-2 to a positive integer.
1109 The logic here will work for any base (even non-integer
1110 bases) but it can be less accurate than
1111 pow (base,power) or exp (power * log (base)) when the
1112 intermediate values start to spill out of the mantissa.
1113 With powers of 2 we know this can't happen.
1114 And powers of 2 are the favourite thing for perl
1115 programmers to notice ** not doing what they mean. */
1117 NV base = baseuok ? baseuv : -(NV)baseuv;
1122 while (power >>= 1) {
1130 SvIV_please_nomg(svr);
1133 unsigned int highbit = 8 * sizeof(UV);
1134 unsigned int diff = 8 * sizeof(UV);
1135 while (diff >>= 1) {
1137 if (baseuv >> highbit) {
1141 /* we now have baseuv < 2 ** highbit */
1142 if (power * highbit <= 8 * sizeof(UV)) {
1143 /* result will definitely fit in UV, so use UV math
1144 on same algorithm as above */
1147 const bool odd_power = cBOOL(power & 1);
1151 while (power >>= 1) {
1158 if (baseuok || !odd_power)
1159 /* answer is positive */
1161 else if (result <= (UV)IV_MAX)
1162 /* answer negative, fits in IV */
1163 SETi( -(IV)result );
1164 else if (result == (UV)IV_MIN)
1165 /* 2's complement assumption: special case IV_MIN */
1168 /* answer negative, doesn't fit */
1169 SETn( -(NV)result );
1177 NV right = SvNV_nomg(svr);
1178 NV left = SvNV_nomg(svl);
1181 #if defined(USE_LONG_DOUBLE) && defined(HAS_AIX_POWL_NEG_BASE_BUG)
1183 We are building perl with long double support and are on an AIX OS
1184 afflicted with a powl() function that wrongly returns NaNQ for any
1185 negative base. This was reported to IBM as PMR #23047-379 on
1186 03/06/2006. The problem exists in at least the following versions
1187 of AIX and the libm fileset, and no doubt others as well:
1189 AIX 4.3.3-ML10 bos.adt.libm 4.3.3.50
1190 AIX 5.1.0-ML04 bos.adt.libm 5.1.0.29
1191 AIX 5.2.0 bos.adt.libm 5.2.0.85
1193 So, until IBM fixes powl(), we provide the following workaround to
1194 handle the problem ourselves. Our logic is as follows: for
1195 negative bases (left), we use fmod(right, 2) to check if the
1196 exponent is an odd or even integer:
1198 - if odd, powl(left, right) == -powl(-left, right)
1199 - if even, powl(left, right) == powl(-left, right)
1201 If the exponent is not an integer, the result is rightly NaNQ, so
1202 we just return that (as NV_NAN).
1206 NV mod2 = Perl_fmod( right, 2.0 );
1207 if (mod2 == 1.0 || mod2 == -1.0) { /* odd integer */
1208 SETn( -Perl_pow( -left, right) );
1209 } else if (mod2 == 0.0) { /* even integer */
1210 SETn( Perl_pow( -left, right) );
1211 } else { /* fractional power */
1215 SETn( Perl_pow( left, right) );
1218 SETn( Perl_pow( left, right) );
1219 #endif /* HAS_AIX_POWL_NEG_BASE_BUG */
1221 #ifdef PERL_PRESERVE_IVUV
1223 SvIV_please_nomg(svr);
1231 dSP; dATARGET; SV *svl, *svr;
1232 tryAMAGICbin_MG(mult_amg, AMGf_assign|AMGf_numeric);
1236 #ifdef PERL_PRESERVE_IVUV
1238 /* special-case some simple common cases */
1239 if (!((svl->sv_flags|svr->sv_flags) & (SVf_IVisUV|SVs_GMG))) {
1241 U32 flags = (svl->sv_flags & svr->sv_flags);
1242 if (flags & SVf_IOK) {
1243 /* both args are simple IVs */
1248 topl = ((UV)il) >> (UVSIZE * 4 - 1);
1249 topr = ((UV)ir) >> (UVSIZE * 4 - 1);
1251 /* if both are in a range that can't under/overflow, do a
1252 * simple integer multiply: if the top halves(*) of both numbers
1253 * are 00...00 or 11...11, then it's safe.
1254 * (*) for 32-bits, the "top half" is the top 17 bits,
1255 * for 64-bits, its 33 bits */
1257 ((topl+1) | (topr+1))
1258 & ( (((UV)1) << (UVSIZE * 4 + 1)) - 2) /* 11..110 */
1261 TARGi(il * ir, 0); /* args not GMG, so can't be tainted */
1267 else if (flags & SVf_NOK) {
1268 /* both args are NVs */
1274 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
1275 !Perl_isnan(nl) && nl == (NV)(il = (IV)nl)
1276 && !Perl_isnan(nr) && nr == (NV)(ir = (IV)nr)
1278 nl == (NV)(il = (IV)nl) && nr == (NV)(ir = (IV)nr)
1281 /* nothing was lost by converting to IVs */
1285 # if defined(__sgi) && defined(USE_LONG_DOUBLE) && LONG_DOUBLEKIND == LONG_DOUBLE_IS_DOUBLEDOUBLE_128_BIT_BE_BE && NVSIZE == 16
1286 if (Perl_isinf(result)) {
1287 Zero((U8*)&result + 8, 8, U8);
1290 TARGn(result, 0); /* args not GMG, so can't be tainted */
1298 if (SvIV_please_nomg(svr)) {
1299 /* Unless the left argument is integer in range we are going to have to
1300 use NV maths. Hence only attempt to coerce the right argument if
1301 we know the left is integer. */
1302 /* Left operand is defined, so is it IV? */
1303 if (SvIV_please_nomg(svl)) {
1304 bool auvok = SvUOK(svl);
1305 bool buvok = SvUOK(svr);
1306 const UV topmask = (~ (UV)0) << (4 * sizeof (UV));
1307 const UV botmask = ~((~ (UV)0) << (4 * sizeof (UV)));
1316 const IV aiv = SvIVX(svl);
1319 auvok = TRUE; /* effectively it's a UV now */
1321 /* abs, auvok == false records sign */
1322 alow = (aiv == IV_MIN) ? (UV)aiv : (UV)(-aiv);
1328 const IV biv = SvIVX(svr);
1331 buvok = TRUE; /* effectively it's a UV now */
1333 /* abs, buvok == false records sign */
1334 blow = (biv == IV_MIN) ? (UV)biv : (UV)(-biv);
1338 /* If this does sign extension on unsigned it's time for plan B */
1339 ahigh = alow >> (4 * sizeof (UV));
1341 bhigh = blow >> (4 * sizeof (UV));
1343 if (ahigh && bhigh) {
1345 /* eg 32 bit is at least 0x10000 * 0x10000 == 0x100000000
1346 which is overflow. Drop to NVs below. */
1347 } else if (!ahigh && !bhigh) {
1348 /* eg 32 bit is at most 0xFFFF * 0xFFFF == 0xFFFE0001
1349 so the unsigned multiply cannot overflow. */
1350 const UV product = alow * blow;
1351 if (auvok == buvok) {
1352 /* -ve * -ve or +ve * +ve gives a +ve result. */
1356 } else if (product <= (UV)IV_MIN) {
1357 /* 2s complement assumption that (UV)-IV_MIN is correct. */
1358 /* -ve result, which could overflow an IV */
1360 /* can't negate IV_MIN, but there are aren't two
1361 * integers such that !ahigh && !bhigh, where the
1362 * product equals 0x800....000 */
1363 assert(product != (UV)IV_MIN);
1364 SETi( -(IV)product );
1366 } /* else drop to NVs below. */
1368 /* One operand is large, 1 small */
1371 /* swap the operands */
1373 bhigh = blow; /* bhigh now the temp var for the swap */
1377 /* now, ((ahigh * blow) << half_UV_len) + (alow * blow)
1378 multiplies can't overflow. shift can, add can, -ve can. */
1379 product_middle = ahigh * blow;
1380 if (!(product_middle & topmask)) {
1381 /* OK, (ahigh * blow) won't lose bits when we shift it. */
1383 product_middle <<= (4 * sizeof (UV));
1384 product_low = alow * blow;
1386 /* as for pp_add, UV + something mustn't get smaller.
1387 IIRC ANSI mandates this wrapping *behaviour* for
1388 unsigned whatever the actual representation*/
1389 product_low += product_middle;
1390 if (product_low >= product_middle) {
1391 /* didn't overflow */
1392 if (auvok == buvok) {
1393 /* -ve * -ve or +ve * +ve gives a +ve result. */
1395 SETu( product_low );
1397 } else if (product_low <= (UV)IV_MIN) {
1398 /* 2s complement assumption again */
1399 /* -ve result, which could overflow an IV */
1401 SETi(product_low == (UV)IV_MIN
1402 ? IV_MIN : -(IV)product_low);
1404 } /* else drop to NVs below. */
1406 } /* product_middle too large */
1407 } /* ahigh && bhigh */
1412 NV right = SvNV_nomg(svr);
1413 NV left = SvNV_nomg(svl);
1414 NV result = left * right;
1417 #if defined(__sgi) && defined(USE_LONG_DOUBLE) && LONG_DOUBLEKIND == LONG_DOUBLE_IS_DOUBLEDOUBLE_128_BIT_BE_BE && NVSIZE == 16
1418 if (Perl_isinf(result)) {
1419 Zero((U8*)&result + 8, 8, U8);
1429 dSP; dATARGET; SV *svl, *svr;
1430 tryAMAGICbin_MG(div_amg, AMGf_assign|AMGf_numeric);
1433 /* Only try to do UV divide first
1434 if ((SLOPPYDIVIDE is true) or
1435 (PERL_PRESERVE_IVUV is true and one or both SV is a UV too large
1437 The assumption is that it is better to use floating point divide
1438 whenever possible, only doing integer divide first if we can't be sure.
1439 If NV_PRESERVES_UV is true then we know at compile time that no UV
1440 can be too large to preserve, so don't need to compile the code to
1441 test the size of UVs. */
1443 #if defined(SLOPPYDIVIDE) || (defined(PERL_PRESERVE_IVUV) && !defined(NV_PRESERVES_UV))
1444 # define PERL_TRY_UV_DIVIDE
1445 /* ensure that 20./5. == 4. */
1448 #ifdef PERL_TRY_UV_DIVIDE
1449 if (SvIV_please_nomg(svr) && SvIV_please_nomg(svl)) {
1450 bool left_non_neg = SvUOK(svl);
1451 bool right_non_neg = SvUOK(svr);
1455 if (right_non_neg) {
1459 const IV biv = SvIVX(svr);
1462 right_non_neg = TRUE; /* effectively it's a UV now */
1465 right = (biv == IV_MIN) ? (UV)biv : (UV)(-biv);
1468 /* historically undef()/0 gives a "Use of uninitialized value"
1469 warning before dieing, hence this test goes here.
1470 If it were immediately before the second SvIV_please, then
1471 DIE() would be invoked before left was even inspected, so
1472 no inspection would give no warning. */
1474 DIE(aTHX_ "Illegal division by zero");
1480 const IV aiv = SvIVX(svl);
1483 left_non_neg = TRUE; /* effectively it's a UV now */
1486 left = (aiv == IV_MIN) ? (UV)aiv : (UV)(-aiv);
1492 /* For sloppy divide we always attempt integer division. */
1494 /* Otherwise we only attempt it if either or both operands
1495 would not be preserved by an NV. If both fit in NVs
1496 we fall through to the NV divide code below. However,
1497 as left >= right to ensure integer result here, we know that
1498 we can skip the test on the right operand - right big
1499 enough not to be preserved can't get here unless left is
1502 && (left > ((UV)1 << NV_PRESERVES_UV_BITS))
1505 /* Integer division can't overflow, but it can be imprecise. */
1506 const UV result = left / right;
1507 if (result * right == left) {
1508 SP--; /* result is valid */
1509 if (left_non_neg == right_non_neg) {
1510 /* signs identical, result is positive. */
1514 /* 2s complement assumption */
1515 if (result <= (UV)IV_MIN)
1516 SETi(result == (UV)IV_MIN ? IV_MIN : -(IV)result);
1518 /* It's exact but too negative for IV. */
1519 SETn( -(NV)result );
1522 } /* tried integer divide but it was not an integer result */
1523 } /* else (PERL_ABS(result) < 1.0) or (both UVs in range for NV) */
1524 } /* one operand wasn't SvIOK */
1525 #endif /* PERL_TRY_UV_DIVIDE */
1527 NV right = SvNV_nomg(svr);
1528 NV left = SvNV_nomg(svl);
1529 (void)POPs;(void)POPs;
1530 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
1531 if (! Perl_isnan(right) && right == 0.0)
1535 DIE(aTHX_ "Illegal division by zero");
1536 PUSHn( left / right );
1544 tryAMAGICbin_MG(modulo_amg, AMGf_assign|AMGf_numeric);
1548 bool left_neg = FALSE;
1549 bool right_neg = FALSE;
1550 bool use_double = FALSE;
1551 bool dright_valid = FALSE;
1554 SV * const svr = TOPs;
1555 SV * const svl = TOPm1s;
1556 if (SvIV_please_nomg(svr)) {
1557 right_neg = !SvUOK(svr);
1561 const IV biv = SvIVX(svr);
1564 right_neg = FALSE; /* effectively it's a UV now */
1566 right = (biv == IV_MIN) ? (UV)biv : (UV)(-biv);
1571 dright = SvNV_nomg(svr);
1572 right_neg = dright < 0;
1575 if (dright < UV_MAX_P1) {
1576 right = U_V(dright);
1577 dright_valid = TRUE; /* In case we need to use double below. */
1583 /* At this point use_double is only true if right is out of range for
1584 a UV. In range NV has been rounded down to nearest UV and
1585 use_double false. */
1586 if (!use_double && SvIV_please_nomg(svl)) {
1587 left_neg = !SvUOK(svl);
1591 const IV aiv = SvIVX(svl);
1594 left_neg = FALSE; /* effectively it's a UV now */
1596 left = (aiv == IV_MIN) ? (UV)aiv : (UV)(-aiv);
1601 dleft = SvNV_nomg(svl);
1602 left_neg = dleft < 0;
1606 /* This should be exactly the 5.6 behaviour - if left and right are
1607 both in range for UV then use U_V() rather than floor. */
1609 if (dleft < UV_MAX_P1) {
1610 /* right was in range, so is dleft, so use UVs not double.
1614 /* left is out of range for UV, right was in range, so promote
1615 right (back) to double. */
1617 /* The +0.5 is used in 5.6 even though it is not strictly
1618 consistent with the implicit +0 floor in the U_V()
1619 inside the #if 1. */
1620 dleft = Perl_floor(dleft + 0.5);
1623 dright = Perl_floor(dright + 0.5);
1634 DIE(aTHX_ "Illegal modulus zero");
1636 dans = Perl_fmod(dleft, dright);
1637 if ((left_neg != right_neg) && dans)
1638 dans = dright - dans;
1641 sv_setnv(TARG, dans);
1647 DIE(aTHX_ "Illegal modulus zero");
1650 if ((left_neg != right_neg) && ans)
1653 /* XXX may warn: unary minus operator applied to unsigned type */
1654 /* could change -foo to be (~foo)+1 instead */
1655 if (ans <= ~((UV)IV_MAX)+1)
1656 sv_setiv(TARG, ~ans+1);
1658 sv_setnv(TARG, -(NV)ans);
1661 sv_setuv(TARG, ans);
1673 bool infnan = FALSE;
1674 const U8 gimme = GIMME_V;
1676 if (gimme == G_ARRAY && PL_op->op_private & OPpREPEAT_DOLIST) {
1677 /* TODO: think of some way of doing list-repeat overloading ??? */
1682 if (UNLIKELY(PL_op->op_private & OPpREPEAT_DOLIST)) {
1683 /* The parser saw this as a list repeat, and there
1684 are probably several items on the stack. But we're
1685 in scalar/void context, and there's no pp_list to save us
1686 now. So drop the rest of the items -- robin@kitsite.com
1689 if (MARK + 1 < SP) {
1695 ASSUME(MARK + 1 == SP);
1697 MARK[1] = &PL_sv_undef;
1701 tryAMAGICbin_MG(repeat_amg, AMGf_assign);
1707 const UV uv = SvUV_nomg(sv);
1709 count = IV_MAX; /* The best we can do? */
1713 count = SvIV_nomg(sv);
1716 else if (SvNOKp(sv)) {
1717 const NV nv = SvNV_nomg(sv);
1718 infnan = Perl_isinfnan(nv);
1719 if (UNLIKELY(infnan)) {
1723 count = -1; /* An arbitrary negative integer */
1729 count = SvIV_nomg(sv);
1732 Perl_ck_warner(aTHX_ packWARN(WARN_NUMERIC),
1733 "Non-finite repeat count does nothing");
1734 } else if (count < 0) {
1736 Perl_ck_warner(aTHX_ packWARN(WARN_NUMERIC),
1737 "Negative repeat count does nothing");
1740 if (gimme == G_ARRAY && PL_op->op_private & OPpREPEAT_DOLIST) {
1742 const SSize_t items = SP - MARK;
1743 const U8 mod = PL_op->op_flags & OPf_MOD;
1748 if ( items > SSize_t_MAX / count /* max would overflow */
1749 /* repeatcpy would overflow */
1750 || items > I32_MAX / (I32)sizeof(SV *)
1752 Perl_croak(aTHX_ "%s","Out of memory during list extend");
1753 max = items * count;
1758 if (mod && SvPADTMP(*SP)) {
1759 *SP = sv_mortalcopy(*SP);
1766 repeatcpy((char*)(MARK + items), (char*)MARK,
1767 items * sizeof(const SV *), count - 1);
1770 else if (count <= 0)
1773 else { /* Note: mark already snarfed by pp_list */
1774 SV * const tmpstr = POPs;
1779 sv_setsv_nomg(TARG, tmpstr);
1780 SvPV_force_nomg(TARG, len);
1781 isutf = DO_UTF8(TARG);
1788 if ( len > (MEM_SIZE_MAX-1) / (UV)count /* max would overflow */
1789 || len > (U32)I32_MAX /* repeatcpy would overflow */
1791 Perl_croak(aTHX_ "%s",
1792 "Out of memory during string extend");
1793 max = (UV)count * len + 1;
1796 repeatcpy(SvPVX(TARG) + len, SvPVX(TARG), len, count - 1);
1797 SvCUR_set(TARG, SvCUR(TARG) * count);
1799 *SvEND(TARG) = '\0';
1802 (void)SvPOK_only_UTF8(TARG);
1804 (void)SvPOK_only(TARG);
1813 dSP; dATARGET; bool useleft; SV *svl, *svr;
1814 tryAMAGICbin_MG(subtr_amg, AMGf_assign|AMGf_numeric);
1818 #ifdef PERL_PRESERVE_IVUV
1820 /* special-case some simple common cases */
1821 if (!((svl->sv_flags|svr->sv_flags) & (SVf_IVisUV|SVs_GMG))) {
1823 U32 flags = (svl->sv_flags & svr->sv_flags);
1824 if (flags & SVf_IOK) {
1825 /* both args are simple IVs */
1830 topl = ((UV)il) >> (UVSIZE * 8 - 2);
1831 topr = ((UV)ir) >> (UVSIZE * 8 - 2);
1833 /* if both are in a range that can't under/overflow, do a
1834 * simple integer subtract: if the top of both numbers
1835 * are 00 or 11, then it's safe */
1836 if (!( ((topl+1) | (topr+1)) & 2)) {
1838 TARGi(il - ir, 0); /* args not GMG, so can't be tainted */
1844 else if (flags & SVf_NOK) {
1845 /* both args are NVs */
1850 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
1851 !Perl_isnan(nl) && nl == (NV)(il = (IV)nl)
1852 && !Perl_isnan(nr) && nr == (NV)(ir = (IV)nr)
1854 nl == (NV)(il = (IV)nl) && nr == (NV)(ir = (IV)nr)
1857 /* nothing was lost by converting to IVs */
1860 TARGn(nl - nr, 0); /* args not GMG, so can't be tainted */
1868 useleft = USE_LEFT(svl);
1869 /* See comments in pp_add (in pp_hot.c) about Overflow, and how
1870 "bad things" happen if you rely on signed integers wrapping. */
1871 if (SvIV_please_nomg(svr)) {
1872 /* Unless the left argument is integer in range we are going to have to
1873 use NV maths. Hence only attempt to coerce the right argument if
1874 we know the left is integer. */
1881 a_valid = auvok = 1;
1882 /* left operand is undef, treat as zero. */
1884 /* Left operand is defined, so is it IV? */
1885 if (SvIV_please_nomg(svl)) {
1886 if ((auvok = SvUOK(svl)))
1889 const IV aiv = SvIVX(svl);
1892 auvok = 1; /* Now acting as a sign flag. */
1893 } else { /* 2s complement assumption for IV_MIN */
1894 auv = (aiv == IV_MIN) ? (UV)aiv : (UV)-aiv;
1901 bool result_good = 0;
1904 bool buvok = SvUOK(svr);
1909 const IV biv = SvIVX(svr);
1914 buv = (biv == IV_MIN) ? (UV)biv : (UV)-biv;
1916 /* ?uvok if value is >= 0. basically, flagged as UV if it's +ve,
1917 else "IV" now, independent of how it came in.
1918 if a, b represents positive, A, B negative, a maps to -A etc
1923 all UV maths. negate result if A negative.
1924 subtract if signs same, add if signs differ. */
1926 if (auvok ^ buvok) {
1935 /* Must get smaller */
1940 if (result <= buv) {
1941 /* result really should be -(auv-buv). as its negation
1942 of true value, need to swap our result flag */
1954 if (result <= (UV)IV_MIN)
1955 SETi(result == (UV)IV_MIN
1956 ? IV_MIN : -(IV)result);
1958 /* result valid, but out of range for IV. */
1959 SETn( -(NV)result );
1963 } /* Overflow, drop through to NVs. */
1967 useleft = USE_LEFT(svl);
1970 NV value = SvNV_nomg(svr);
1974 /* left operand is undef, treat as zero - value */
1978 SETn( SvNV_nomg(svl) - value );
1983 #define IV_BITS (IVSIZE * 8)
1985 static UV S_uv_shift(UV uv, int shift, bool left)
1991 if (shift >= IV_BITS) {
1994 return left ? uv << shift : uv >> shift;
1997 static IV S_iv_shift(IV iv, int shift, bool left)
2003 if (shift >= IV_BITS) {
2004 return iv < 0 && !left ? -1 : 0;
2006 return left ? iv << shift : iv >> shift;
2009 #define UV_LEFT_SHIFT(uv, shift) S_uv_shift(uv, shift, TRUE)
2010 #define UV_RIGHT_SHIFT(uv, shift) S_uv_shift(uv, shift, FALSE)
2011 #define IV_LEFT_SHIFT(iv, shift) S_iv_shift(iv, shift, TRUE)
2012 #define IV_RIGHT_SHIFT(iv, shift) S_iv_shift(iv, shift, FALSE)
2016 dSP; dATARGET; SV *svl, *svr;
2017 tryAMAGICbin_MG(lshift_amg, AMGf_assign|AMGf_numeric);
2021 const IV shift = SvIV_nomg(svr);
2022 if (PL_op->op_private & HINT_INTEGER) {
2023 SETi(IV_LEFT_SHIFT(SvIV_nomg(svl), shift));
2026 SETu(UV_LEFT_SHIFT(SvUV_nomg(svl), shift));
2034 dSP; dATARGET; SV *svl, *svr;
2035 tryAMAGICbin_MG(rshift_amg, AMGf_assign|AMGf_numeric);
2039 const IV shift = SvIV_nomg(svr);
2040 if (PL_op->op_private & HINT_INTEGER) {
2041 SETi(IV_RIGHT_SHIFT(SvIV_nomg(svl), shift));
2044 SETu(UV_RIGHT_SHIFT(SvUV_nomg(svl), shift));
2055 tryAMAGICbin_MG(lt_amg, AMGf_set|AMGf_numeric);
2059 (SvIOK_notUV(left) && SvIOK_notUV(right))
2060 ? (SvIVX(left) < SvIVX(right))
2061 : (do_ncmp(left, right) == -1)
2071 tryAMAGICbin_MG(gt_amg, AMGf_set|AMGf_numeric);
2075 (SvIOK_notUV(left) && SvIOK_notUV(right))
2076 ? (SvIVX(left) > SvIVX(right))
2077 : (do_ncmp(left, right) == 1)
2087 tryAMAGICbin_MG(le_amg, AMGf_set|AMGf_numeric);
2091 (SvIOK_notUV(left) && SvIOK_notUV(right))
2092 ? (SvIVX(left) <= SvIVX(right))
2093 : (do_ncmp(left, right) <= 0)
2103 tryAMAGICbin_MG(ge_amg, AMGf_set|AMGf_numeric);
2107 (SvIOK_notUV(left) && SvIOK_notUV(right))
2108 ? (SvIVX(left) >= SvIVX(right))
2109 : ( (do_ncmp(left, right) & 2) == 0)
2119 tryAMAGICbin_MG(ne_amg, AMGf_set|AMGf_numeric);
2123 (SvIOK_notUV(left) && SvIOK_notUV(right))
2124 ? (SvIVX(left) != SvIVX(right))
2125 : (do_ncmp(left, right) != 0)
2130 /* compare left and right SVs. Returns:
2134 * 2: left or right was a NaN
2137 Perl_do_ncmp(pTHX_ SV* const left, SV * const right)
2139 PERL_ARGS_ASSERT_DO_NCMP;
2140 #ifdef PERL_PRESERVE_IVUV
2141 /* Fortunately it seems NaN isn't IOK */
2142 if (SvIV_please_nomg(right) && SvIV_please_nomg(left)) {
2144 const IV leftiv = SvIVX(left);
2145 if (!SvUOK(right)) {
2146 /* ## IV <=> IV ## */
2147 const IV rightiv = SvIVX(right);
2148 return (leftiv > rightiv) - (leftiv < rightiv);
2150 /* ## IV <=> UV ## */
2152 /* As (b) is a UV, it's >=0, so it must be < */
2155 const UV rightuv = SvUVX(right);
2156 return ((UV)leftiv > rightuv) - ((UV)leftiv < rightuv);
2161 /* ## UV <=> UV ## */
2162 const UV leftuv = SvUVX(left);
2163 const UV rightuv = SvUVX(right);
2164 return (leftuv > rightuv) - (leftuv < rightuv);
2166 /* ## UV <=> IV ## */
2168 const IV rightiv = SvIVX(right);
2170 /* As (a) is a UV, it's >=0, so it cannot be < */
2173 const UV leftuv = SvUVX(left);
2174 return (leftuv > (UV)rightiv) - (leftuv < (UV)rightiv);
2177 NOT_REACHED; /* NOTREACHED */
2181 NV const rnv = SvNV_nomg(right);
2182 NV const lnv = SvNV_nomg(left);
2184 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2185 if (Perl_isnan(lnv) || Perl_isnan(rnv)) {
2188 return (lnv > rnv) - (lnv < rnv);
2207 tryAMAGICbin_MG(ncmp_amg, AMGf_numeric);
2210 value = do_ncmp(left, right);
2222 /* also used for: pp_sge() pp_sgt() pp_slt() */
2228 int amg_type = sle_amg;
2232 switch (PL_op->op_type) {
2251 tryAMAGICbin_MG(amg_type, AMGf_set);
2255 #ifdef USE_LOCALE_COLLATE
2256 (IN_LC_RUNTIME(LC_COLLATE))
2257 ? sv_cmp_locale_flags(left, right, 0)
2260 sv_cmp_flags(left, right, 0);
2261 SETs(boolSV(cmp * multiplier < rhs));
2269 tryAMAGICbin_MG(seq_amg, AMGf_set);
2272 SETs(boolSV(sv_eq_flags(left, right, 0)));
2280 tryAMAGICbin_MG(sne_amg, AMGf_set);
2283 SETs(boolSV(!sv_eq_flags(left, right, 0)));
2291 tryAMAGICbin_MG(scmp_amg, 0);
2295 #ifdef USE_LOCALE_COLLATE
2296 (IN_LC_RUNTIME(LC_COLLATE))
2297 ? sv_cmp_locale_flags(left, right, 0)
2300 sv_cmp_flags(left, right, 0);
2309 tryAMAGICbin_MG(band_amg, AMGf_assign);
2312 if (SvNIOKp(left) || SvNIOKp(right)) {
2313 const bool left_ro_nonnum = !SvNIOKp(left) && SvREADONLY(left);
2314 const bool right_ro_nonnum = !SvNIOKp(right) && SvREADONLY(right);
2315 if (PL_op->op_private & HINT_INTEGER) {
2316 const IV i = SvIV_nomg(left) & SvIV_nomg(right);
2320 const UV u = SvUV_nomg(left) & SvUV_nomg(right);
2323 if (left_ro_nonnum && left != TARG) SvNIOK_off(left);
2324 if (right_ro_nonnum) SvNIOK_off(right);
2327 do_vop(PL_op->op_type, TARG, left, right);
2337 tryAMAGICbin_MG(band_amg, AMGf_assign|AMGf_numarg);
2339 dATARGET; dPOPTOPssrl;
2340 if (PL_op->op_private & HINT_INTEGER) {
2341 const IV i = SvIV_nomg(left) & SvIV_nomg(right);
2345 const UV u = SvUV_nomg(left) & SvUV_nomg(right);
2355 tryAMAGICbin_MG(sband_amg, AMGf_assign);
2357 dATARGET; dPOPTOPssrl;
2358 do_vop(OP_BIT_AND, TARG, left, right);
2363 /* also used for: pp_bit_xor() */
2368 const int op_type = PL_op->op_type;
2370 tryAMAGICbin_MG((op_type == OP_BIT_OR ? bor_amg : bxor_amg), AMGf_assign);
2373 if (SvNIOKp(left) || SvNIOKp(right)) {
2374 const bool left_ro_nonnum = !SvNIOKp(left) && SvREADONLY(left);
2375 const bool right_ro_nonnum = !SvNIOKp(right) && SvREADONLY(right);
2376 if (PL_op->op_private & HINT_INTEGER) {
2377 const IV l = (USE_LEFT(left) ? SvIV_nomg(left) : 0);
2378 const IV r = SvIV_nomg(right);
2379 const IV result = op_type == OP_BIT_OR ? (l | r) : (l ^ r);
2383 const UV l = (USE_LEFT(left) ? SvUV_nomg(left) : 0);
2384 const UV r = SvUV_nomg(right);
2385 const UV result = op_type == OP_BIT_OR ? (l | r) : (l ^ r);
2388 if (left_ro_nonnum && left != TARG) SvNIOK_off(left);
2389 if (right_ro_nonnum) SvNIOK_off(right);
2392 do_vop(op_type, TARG, left, right);
2399 /* also used for: pp_nbit_xor() */
2404 const int op_type = PL_op->op_type;
2406 tryAMAGICbin_MG((op_type == OP_NBIT_OR ? bor_amg : bxor_amg),
2407 AMGf_assign|AMGf_numarg);
2409 dATARGET; dPOPTOPssrl;
2410 if (PL_op->op_private & HINT_INTEGER) {
2411 const IV l = (USE_LEFT(left) ? SvIV_nomg(left) : 0);
2412 const IV r = SvIV_nomg(right);
2413 const IV result = op_type == OP_NBIT_OR ? (l | r) : (l ^ r);
2417 const UV l = (USE_LEFT(left) ? SvUV_nomg(left) : 0);
2418 const UV r = SvUV_nomg(right);
2419 const UV result = op_type == OP_NBIT_OR ? (l | r) : (l ^ r);
2426 /* also used for: pp_sbit_xor() */
2431 const int op_type = PL_op->op_type;
2433 tryAMAGICbin_MG((op_type == OP_SBIT_OR ? sbor_amg : sbxor_amg),
2436 dATARGET; dPOPTOPssrl;
2437 do_vop(op_type == OP_SBIT_OR ? OP_BIT_OR : OP_BIT_XOR, TARG, left,
2443 PERL_STATIC_INLINE bool
2444 S_negate_string(pTHX)
2449 SV * const sv = TOPs;
2450 if (!SvPOKp(sv) || SvNIOK(sv) || (!SvPOK(sv) && SvNIOKp(sv)))
2452 s = SvPV_nomg_const(sv, len);
2453 if (isIDFIRST(*s)) {
2454 sv_setpvs(TARG, "-");
2457 else if (*s == '+' || (*s == '-' && !looks_like_number(sv))) {
2458 sv_setsv_nomg(TARG, sv);
2459 *SvPV_force_nomg(TARG, len) = *s == '-' ? '+' : '-';
2469 tryAMAGICun_MG(neg_amg, AMGf_numeric);
2470 if (S_negate_string(aTHX)) return NORMAL;
2472 SV * const sv = TOPs;
2475 /* It's publicly an integer */
2478 if (SvIVX(sv) == IV_MIN) {
2479 /* 2s complement assumption. */
2480 SETi(SvIVX(sv)); /* special case: -((UV)IV_MAX+1) ==
2484 else if (SvUVX(sv) <= IV_MAX) {
2489 else if (SvIVX(sv) != IV_MIN) {
2493 #ifdef PERL_PRESERVE_IVUV
2500 if (SvNIOKp(sv) && (SvNIOK(sv) || !SvPOK(sv)))
2501 SETn(-SvNV_nomg(sv));
2502 else if (SvPOKp(sv) && SvIV_please_nomg(sv))
2503 goto oops_its_an_int;
2505 SETn(-SvNV_nomg(sv));
2515 tryAMAGICun_MG(not_amg, AMGf_set);
2517 *PL_stack_sp = boolSV(!SvTRUE_nomg_NN(sv));
2522 S_scomplement(pTHX_ SV *targ, SV *sv)
2528 sv_copypv_nomg(TARG, sv);
2529 tmps = (U8*)SvPV_nomg(TARG, len);
2532 if (len && ! utf8_to_bytes(tmps, &len)) {
2533 Perl_croak(aTHX_ FATAL_ABOVE_FF_MSG, PL_op_desc[PL_op->op_type]);
2544 for ( ; anum && (unsigned long)tmps % sizeof(long); anum--, tmps++)
2547 for ( ; anum >= (I32)sizeof(long); anum -= (I32)sizeof(long), tmpl++)
2552 for ( ; anum > 0; anum--, tmps++)
2559 tryAMAGICun_MG(compl_amg, AMGf_numeric);
2563 if (PL_op->op_private & HINT_INTEGER) {
2564 const IV i = ~SvIV_nomg(sv);
2568 const UV u = ~SvUV_nomg(sv);
2573 S_scomplement(aTHX_ TARG, sv);
2583 tryAMAGICun_MG(compl_amg, AMGf_numeric|AMGf_numarg);
2586 if (PL_op->op_private & HINT_INTEGER) {
2587 const IV i = ~SvIV_nomg(sv);
2591 const UV u = ~SvUV_nomg(sv);
2601 tryAMAGICun_MG(scompl_amg, AMGf_numeric);
2604 S_scomplement(aTHX_ TARG, sv);
2610 /* integer versions of some of the above */
2615 tryAMAGICbin_MG(mult_amg, AMGf_assign);
2618 SETi( left * right );
2627 tryAMAGICbin_MG(div_amg, AMGf_assign);
2630 IV value = SvIV_nomg(right);
2632 DIE(aTHX_ "Illegal division by zero");
2633 num = SvIV_nomg(left);
2635 /* avoid FPE_INTOVF on some platforms when num is IV_MIN */
2639 value = num / value;
2647 /* This is the vanilla old i_modulo. */
2649 tryAMAGICbin_MG(modulo_amg, AMGf_assign);
2653 DIE(aTHX_ "Illegal modulus zero");
2654 /* avoid FPE_INTOVF on some platforms when left is IV_MIN */
2658 SETi( left % right );
2663 #if defined(__GLIBC__) && IVSIZE == 8 \
2664 && ( __GLIBC__ < 2 || (__GLIBC__ == 2 && __GLIBC_MINOR__ < 8))
2666 PP(pp_i_modulo_glibc_bugfix)
2668 /* This is the i_modulo with the workaround for the _moddi3 bug
2669 * in (at least) glibc 2.2.5 (the PERL_ABS() the workaround).
2670 * See below for pp_i_modulo. */
2672 tryAMAGICbin_MG(modulo_amg, AMGf_assign);
2676 DIE(aTHX_ "Illegal modulus zero");
2677 /* avoid FPE_INTOVF on some platforms when left is IV_MIN */
2681 SETi( left % PERL_ABS(right) );
2690 tryAMAGICbin_MG(add_amg, AMGf_assign);
2692 dPOPTOPiirl_ul_nomg;
2693 SETi( left + right );
2701 tryAMAGICbin_MG(subtr_amg, AMGf_assign);
2703 dPOPTOPiirl_ul_nomg;
2704 SETi( left - right );
2712 tryAMAGICbin_MG(lt_amg, AMGf_set);
2715 SETs(boolSV(left < right));
2723 tryAMAGICbin_MG(gt_amg, AMGf_set);
2726 SETs(boolSV(left > right));
2734 tryAMAGICbin_MG(le_amg, AMGf_set);
2737 SETs(boolSV(left <= right));
2745 tryAMAGICbin_MG(ge_amg, AMGf_set);
2748 SETs(boolSV(left >= right));
2756 tryAMAGICbin_MG(eq_amg, AMGf_set);
2759 SETs(boolSV(left == right));
2767 tryAMAGICbin_MG(ne_amg, AMGf_set);
2770 SETs(boolSV(left != right));
2778 tryAMAGICbin_MG(ncmp_amg, 0);
2785 else if (left < right)
2797 tryAMAGICun_MG(neg_amg, 0);
2798 if (S_negate_string(aTHX)) return NORMAL;
2800 SV * const sv = TOPs;
2801 IV const i = SvIV_nomg(sv);
2807 /* High falutin' math. */
2812 tryAMAGICbin_MG(atan2_amg, 0);
2815 SETn(Perl_atan2(left, right));
2821 /* also used for: pp_cos() pp_exp() pp_log() pp_sqrt() */
2826 int amg_type = fallback_amg;
2827 const char *neg_report = NULL;
2828 const int op_type = PL_op->op_type;
2831 case OP_SIN: amg_type = sin_amg; break;
2832 case OP_COS: amg_type = cos_amg; break;
2833 case OP_EXP: amg_type = exp_amg; break;
2834 case OP_LOG: amg_type = log_amg; neg_report = "log"; break;
2835 case OP_SQRT: amg_type = sqrt_amg; neg_report = "sqrt"; break;
2838 assert(amg_type != fallback_amg);
2840 tryAMAGICun_MG(amg_type, 0);
2842 SV * const arg = TOPs;
2843 const NV value = SvNV_nomg(arg);
2849 if (neg_report) { /* log or sqrt */
2851 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2852 ! Perl_isnan(value) &&
2854 (op_type == OP_LOG ? (value <= 0.0) : (value < 0.0))) {
2855 SET_NUMERIC_STANDARD();
2856 /* diag_listed_as: Can't take log of %g */
2857 DIE(aTHX_ "Can't take %s of %" NVgf, neg_report, value);
2862 case OP_SIN: result = Perl_sin(value); break;
2863 case OP_COS: result = Perl_cos(value); break;
2864 case OP_EXP: result = Perl_exp(value); break;
2865 case OP_LOG: result = Perl_log(value); break;
2866 case OP_SQRT: result = Perl_sqrt(value); break;
2873 /* Support Configure command-line overrides for rand() functions.
2874 After 5.005, perhaps we should replace this by Configure support
2875 for drand48(), random(), or rand(). For 5.005, though, maintain
2876 compatibility by calling rand() but allow the user to override it.
2877 See INSTALL for details. --Andy Dougherty 15 July 1998
2879 /* Now it's after 5.005, and Configure supports drand48() and random(),
2880 in addition to rand(). So the overrides should not be needed any more.
2881 --Jarkko Hietaniemi 27 September 1998
2886 if (!PL_srand_called) {
2887 (void)seedDrand01((Rand_seed_t)seed());
2888 PL_srand_called = TRUE;
2900 SV * const sv = POPs;
2906 /* 1 of 2 things can be carried through SvNV, SP or TARG, SP was carried */
2907 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2908 if (! Perl_isnan(value) && value == 0.0)
2918 sv_setnv_mg(TARG, value);
2929 if (MAXARG >= 1 && (TOPs || POPs)) {
2936 pv = SvPV(top, len);
2937 flags = grok_number(pv, len, &anum);
2939 if (!(flags & IS_NUMBER_IN_UV)) {
2940 Perl_ck_warner_d(aTHX_ packWARN(WARN_OVERFLOW),
2941 "Integer overflow in srand");
2949 (void)seedDrand01((Rand_seed_t)anum);
2950 PL_srand_called = TRUE;
2954 /* Historically srand always returned true. We can avoid breaking
2956 sv_setpvs(TARG, "0 but true");
2965 tryAMAGICun_MG(int_amg, AMGf_numeric);
2967 SV * const sv = TOPs;
2968 const IV iv = SvIV_nomg(sv);
2969 /* XXX it's arguable that compiler casting to IV might be subtly
2970 different from modf (for numbers inside (IV_MIN,UV_MAX)) in which
2971 else preferring IV has introduced a subtle behaviour change bug. OTOH
2972 relying on floating point to be accurate is a bug. */
2977 else if (SvIOK(sv)) {
2979 SETu(SvUV_nomg(sv));
2984 const NV value = SvNV_nomg(sv);
2985 if (UNLIKELY(Perl_isinfnan(value)))
2987 else if (value >= 0.0) {
2988 if (value < (NV)UV_MAX + 0.5) {
2991 SETn(Perl_floor(value));
2995 if (value > (NV)IV_MIN - 0.5) {
2998 SETn(Perl_ceil(value));
3009 tryAMAGICun_MG(abs_amg, AMGf_numeric);
3011 SV * const sv = TOPs;
3012 /* This will cache the NV value if string isn't actually integer */
3013 const IV iv = SvIV_nomg(sv);
3018 else if (SvIOK(sv)) {
3019 /* IVX is precise */
3021 SETu(SvUV_nomg(sv)); /* force it to be numeric only */
3029 /* 2s complement assumption. Also, not really needed as
3030 IV_MIN and -IV_MIN should both be %100...00 and NV-able */
3036 const NV value = SvNV_nomg(sv);
3047 /* also used for: pp_hex() */
3053 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES;
3057 SV* const sv = TOPs;
3059 tmps = (SvPV_const(sv, len));
3061 /* If Unicode, try to downgrade
3062 * If not possible, croak. */
3063 SV* const tsv = sv_2mortal(newSVsv(sv));
3066 sv_utf8_downgrade(tsv, FALSE);
3067 tmps = SvPV_const(tsv, len);
3069 if (PL_op->op_type == OP_HEX)
3072 while (*tmps && len && isSPACE(*tmps))
3076 if (isALPHA_FOLD_EQ(*tmps, 'x')) {
3078 result_uv = grok_hex (tmps, &len, &flags, &result_nv);
3080 else if (isALPHA_FOLD_EQ(*tmps, 'b'))
3081 result_uv = grok_bin (tmps, &len, &flags, &result_nv);
3083 result_uv = grok_oct (tmps, &len, &flags, &result_nv);
3085 if (flags & PERL_SCAN_GREATER_THAN_UV_MAX) {
3100 SV * const sv = TOPs;
3102 U32 in_bytes = IN_BYTES;
3103 /* Simplest case shortcut:
3104 * set svflags to just the SVf_POK|SVs_GMG|SVf_UTF8 from the SV,
3105 * with the SVf_UTF8 flag inverted if under 'use bytes' (HINT_BYTES
3108 U32 svflags = (SvFLAGS(sv) ^ (in_bytes << 26)) & (SVf_POK|SVs_GMG|SVf_UTF8);
3110 STATIC_ASSERT_STMT(SVf_UTF8 == (HINT_BYTES << 26));
3113 if (LIKELY(svflags == SVf_POK))
3116 if (svflags & SVs_GMG)
3121 if (!IN_BYTES) { /* reread to avoid using an C auto/register */
3122 if ((SvFLAGS(sv) & (SVf_POK|SVf_UTF8)) == SVf_POK)
3124 if ( SvPOK(sv) && (PL_op->op_private & OPpTRUEBOOL)) {
3125 /* no need to convert from bytes to chars */
3129 len = sv_len_utf8_nomg(sv);
3132 /* unrolled SvPV_nomg_const(sv,len) */
3133 if (SvPOK_nog(sv)) {
3136 if (PL_op->op_private & OPpTRUEBOOL) {
3138 SETs(len ? &PL_sv_yes : &PL_sv_zero);
3143 (void)sv_2pv_flags(sv, &len, 0|SV_CONST_RETURN);
3146 TARGi((IV)(len), 1);
3149 if (!SvPADTMP(TARG)) {
3150 /* OPpTARGET_MY: targ is var in '$lex = length()' */
3155 /* TARG is on stack at this point and is overwriten by SETs.
3156 * This branch is the odd one out, so put TARG by default on
3157 * stack earlier to let local SP go out of liveness sooner */
3160 return NORMAL; /* no putback, SP didn't move in this opcode */
3164 /* Returns false if substring is completely outside original string.
3165 No length is indicated by len_iv = 0 and len_is_uv = 0. len_is_uv must
3166 always be true for an explicit 0.
3169 Perl_translate_substr_offsets( STRLEN curlen, IV pos1_iv,
3170 bool pos1_is_uv, IV len_iv,
3171 bool len_is_uv, STRLEN *posp,
3177 PERL_ARGS_ASSERT_TRANSLATE_SUBSTR_OFFSETS;
3179 if (!pos1_is_uv && pos1_iv < 0 && curlen) {
3180 pos1_is_uv = curlen-1 > ~(UV)pos1_iv;
3183 if ((pos1_is_uv || pos1_iv > 0) && (UV)pos1_iv > curlen)
3186 if (len_iv || len_is_uv) {
3187 if (!len_is_uv && len_iv < 0) {
3188 pos2_iv = curlen + len_iv;
3190 pos2_is_uv = curlen-1 > ~(UV)len_iv;
3193 } else { /* len_iv >= 0 */
3194 if (!pos1_is_uv && pos1_iv < 0) {
3195 pos2_iv = pos1_iv + len_iv;
3196 pos2_is_uv = (UV)len_iv > (UV)IV_MAX;
3198 if ((UV)len_iv > curlen-(UV)pos1_iv)
3201 pos2_iv = pos1_iv+len_iv;
3211 if (!pos2_is_uv && pos2_iv < 0) {
3212 if (!pos1_is_uv && pos1_iv < 0)
3216 else if (!pos1_is_uv && pos1_iv < 0)
3219 if ((UV)pos2_iv < (UV)pos1_iv)
3221 if ((UV)pos2_iv > curlen)
3224 /* pos1_iv and pos2_iv both in 0..curlen, so the cast is safe */
3225 *posp = (STRLEN)( (UV)pos1_iv );
3226 *lenp = (STRLEN)( (UV)pos2_iv - (UV)pos1_iv );
3243 I32 lvalue = PL_op->op_flags & OPf_MOD || LVRET;
3244 const bool rvalue = (GIMME_V != G_VOID);
3247 const char *repl = NULL;
3249 int num_args = PL_op->op_private & 7;
3250 bool repl_need_utf8_upgrade = FALSE;
3254 if(!(repl_sv = POPs)) num_args--;
3256 if ((len_sv = POPs)) {
3257 len_iv = SvIV(len_sv);
3258 len_is_uv = len_iv ? SvIOK_UV(len_sv) : 1;
3263 pos1_iv = SvIV(pos_sv);
3264 pos1_is_uv = SvIOK_UV(pos_sv);
3266 if (PL_op->op_private & OPpSUBSTR_REPL_FIRST) {
3270 if (lvalue && !repl_sv) {
3272 ret = sv_2mortal(newSV_type(SVt_PVLV)); /* Not TARG RT#67838 */
3273 sv_magic(ret, NULL, PERL_MAGIC_substr, NULL, 0);
3275 LvTARG(ret) = SvREFCNT_inc_simple(sv);
3277 pos1_is_uv || pos1_iv >= 0
3278 ? (STRLEN)(UV)pos1_iv
3279 : (LvFLAGS(ret) |= LVf_NEG_OFF, (STRLEN)(UV)-pos1_iv);
3281 len_is_uv || len_iv > 0
3282 ? (STRLEN)(UV)len_iv
3283 : (LvFLAGS(ret) |= LVf_NEG_LEN, (STRLEN)(UV)-len_iv);
3285 PUSHs(ret); /* avoid SvSETMAGIC here */
3289 repl = SvPV_const(repl_sv, repl_len);
3292 Perl_ck_warner(aTHX_ packWARN(WARN_SUBSTR),
3293 "Attempt to use reference as lvalue in substr"
3295 tmps = SvPV_force_nomg(sv, curlen);
3296 if (DO_UTF8(repl_sv) && repl_len) {
3298 /* Upgrade the dest, and recalculate tmps in case the buffer
3299 * got reallocated; curlen may also have been changed */
3300 sv_utf8_upgrade_nomg(sv);
3301 tmps = SvPV_nomg(sv, curlen);
3304 else if (DO_UTF8(sv))
3305 repl_need_utf8_upgrade = TRUE;
3307 else tmps = SvPV_const(sv, curlen);
3309 utf8_curlen = sv_or_pv_len_utf8(sv, tmps, curlen);
3310 if (utf8_curlen == curlen)
3313 curlen = utf8_curlen;
3319 STRLEN pos, len, byte_len, byte_pos;
3321 if (!translate_substr_offsets(
3322 curlen, pos1_iv, pos1_is_uv, len_iv, len_is_uv, &pos, &len
3326 byte_pos = utf8_curlen
3327 ? sv_or_pv_pos_u2b(sv, tmps, pos, &byte_len) : pos;
3332 SvTAINTED_off(TARG); /* decontaminate */
3333 SvUTF8_off(TARG); /* decontaminate */
3334 sv_setpvn(TARG, tmps, byte_len);
3335 #ifdef USE_LOCALE_COLLATE
3336 sv_unmagic(TARG, PERL_MAGIC_collxfrm);
3343 SV* repl_sv_copy = NULL;
3345 if (repl_need_utf8_upgrade) {
3346 repl_sv_copy = newSVsv(repl_sv);
3347 sv_utf8_upgrade(repl_sv_copy);
3348 repl = SvPV_const(repl_sv_copy, repl_len);
3352 sv_insert_flags(sv, byte_pos, byte_len, repl, repl_len, 0);
3353 SvREFCNT_dec(repl_sv_copy);
3356 if (PL_op->op_private & OPpSUBSTR_REPL_FIRST)
3366 Perl_croak(aTHX_ "substr outside of string");
3367 Perl_ck_warner(aTHX_ packWARN(WARN_SUBSTR), "substr outside of string");
3374 const IV size = POPi;
3375 SV* offsetsv = POPs;
3376 SV * const src = POPs;
3377 const I32 lvalue = PL_op->op_flags & OPf_MOD || LVRET;
3383 /* extract a STRLEN-ranged integer value from offsetsv into offset,
3384 * or flag that its out of range */
3386 IV iv = SvIV(offsetsv);
3388 /* avoid a large UV being wrapped to a negative value */
3389 if (SvIOK_UV(offsetsv) && SvUVX(offsetsv) > (UV)IV_MAX)
3390 errflags = LVf_OUT_OF_RANGE;
3392 errflags = (LVf_NEG_OFF|LVf_OUT_OF_RANGE);
3393 #if PTRSIZE < IVSIZE
3394 else if (iv > Size_t_MAX)
3395 errflags = LVf_OUT_OF_RANGE;
3398 offset = (STRLEN)iv;
3401 retuv = errflags ? 0 : do_vecget(src, offset, size);
3403 if (lvalue) { /* it's an lvalue! */
3404 ret = sv_2mortal(newSV_type(SVt_PVLV)); /* Not TARG RT#67838 */
3405 sv_magic(ret, NULL, PERL_MAGIC_vec, NULL, 0);
3407 LvTARG(ret) = SvREFCNT_inc_simple(src);
3408 LvTARGOFF(ret) = offset;
3409 LvTARGLEN(ret) = size;
3410 LvFLAGS(ret) = errflags;
3414 SvTAINTED_off(TARG); /* decontaminate */
3418 sv_setuv(ret, retuv);
3426 /* also used for: pp_rindex() */
3439 const char *little_p;
3442 const bool is_index = PL_op->op_type == OP_INDEX;
3443 const bool threeargs = MAXARG >= 3 && (TOPs || ((void)POPs,0));
3449 big_p = SvPV_const(big, biglen);
3450 little_p = SvPV_const(little, llen);
3452 big_utf8 = DO_UTF8(big);
3453 little_utf8 = DO_UTF8(little);
3454 if (big_utf8 ^ little_utf8) {
3455 /* One needs to be upgraded. */
3457 /* Well, maybe instead we might be able to downgrade the small
3459 char * const pv = (char*)bytes_from_utf8((U8 *)little_p, &llen,
3462 /* If the large string is ISO-8859-1, and it's not possible to
3463 convert the small string to ISO-8859-1, then there is no
3464 way that it could be found anywhere by index. */
3469 /* At this point, pv is a malloc()ed string. So donate it to temp
3470 to ensure it will get free()d */
3471 little = temp = newSV(0);
3472 sv_usepvn(temp, pv, llen);
3473 little_p = SvPVX(little);
3475 temp = newSVpvn(little_p, llen);
3477 sv_utf8_upgrade(temp);
3479 little_p = SvPV_const(little, llen);
3482 if (SvGAMAGIC(big)) {
3483 /* Life just becomes a lot easier if I use a temporary here.
3484 Otherwise I need to avoid calls to sv_pos_u2b(), which (dangerously)
3485 will trigger magic and overloading again, as will fbm_instr()
3487 big = newSVpvn_flags(big_p, biglen,
3488 SVs_TEMP | (big_utf8 ? SVf_UTF8 : 0));
3491 if (SvGAMAGIC(little) || (is_index && !SvOK(little))) {
3492 /* index && SvOK() is a hack. fbm_instr() calls SvPV_const, which will
3493 warn on undef, and we've already triggered a warning with the
3494 SvPV_const some lines above. We can't remove that, as we need to
3495 call some SvPV to trigger overloading early and find out if the
3497 This is all getting too messy. The API isn't quite clean enough,
3498 because data access has side effects.
3500 little = newSVpvn_flags(little_p, llen,
3501 SVs_TEMP | (little_utf8 ? SVf_UTF8 : 0));
3502 little_p = SvPVX(little);
3506 offset = is_index ? 0 : biglen;
3508 if (big_utf8 && offset > 0)
3509 offset = sv_pos_u2b_flags(big, offset, 0, SV_CONST_RETURN);
3515 else if (offset > (SSize_t)biglen)
3517 if (!(little_p = is_index
3518 ? fbm_instr((unsigned char*)big_p + offset,
3519 (unsigned char*)big_p + biglen, little, 0)
3520 : rninstr(big_p, big_p + offset,
3521 little_p, little_p + llen)))
3524 retval = little_p - big_p;
3525 if (retval > 1 && big_utf8)
3526 retval = sv_pos_b2u_flags(big, retval, SV_CONST_RETURN);
3531 /* OPpTRUEBOOL indicates an '== -1' has been optimised away */
3532 if (PL_op->op_private & OPpTRUEBOOL) {
3533 PUSHs( ((retval != -1) ^ cBOOL(PL_op->op_private & OPpINDEX_BOOLNEG))
3534 ? &PL_sv_yes : &PL_sv_no);
3535 if (PL_op->op_private & OPpTARGET_MY)
3536 /* $lex = (index() == -1) */
3537 sv_setsv(TARG, TOPs);
3546 dSP; dMARK; dORIGMARK; dTARGET;
3547 SvTAINTED_off(TARG);
3548 do_sprintf(TARG, SP-MARK, MARK+1);
3549 TAINT_IF(SvTAINTED(TARG));
3561 const U8 *s = (U8*)SvPV_const(argsv, len);
3564 ? (len ? utf8n_to_uvchr(s, len, 0, UTF8_ALLOW_ANYUV) : 0)
3578 if (UNLIKELY(SvAMAGIC(top)))
3580 if (UNLIKELY(isinfnansv(top)))
3581 Perl_croak(aTHX_ "Cannot chr %" NVgf, SvNV(top));
3583 if (!IN_BYTES /* under bytes, chr(-1) eq chr(0xff), etc. */
3584 && ((SvIOKp(top) && !SvIsUV(top) && SvIV_nomg(top) < 0)
3586 ((SvNOKp(top) || (SvOK(top) && !SvIsUV(top)))
3587 && SvNV_nomg(top) < 0.0)))
3589 if (ckWARN(WARN_UTF8)) {
3590 if (SvGMAGICAL(top)) {
3591 SV *top2 = sv_newmortal();
3592 sv_setsv_nomg(top2, top);
3595 Perl_warner(aTHX_ packWARN(WARN_UTF8),
3596 "Invalid negative number (%" SVf ") in chr", SVfARG(top));
3598 value = UNICODE_REPLACEMENT;
3600 value = SvUV_nomg(top);
3604 SvUPGRADE(TARG,SVt_PV);
3606 if (value > 255 && !IN_BYTES) {
3607 SvGROW(TARG, (STRLEN)UVCHR_SKIP(value)+1);
3608 tmps = (char*)uvchr_to_utf8_flags((U8*)SvPVX(TARG), value, 0);
3609 SvCUR_set(TARG, tmps - SvPVX_const(TARG));
3611 (void)SvPOK_only(TARG);
3620 *tmps++ = (char)value;
3622 (void)SvPOK_only(TARG);
3634 const char *tmps = SvPV_const(left, len);
3636 if (DO_UTF8(left)) {
3637 /* If Unicode, try to downgrade.
3638 * If not possible, croak.
3639 * Yes, we made this up. */
3640 SV* const tsv = newSVpvn_flags(tmps, len, SVf_UTF8|SVs_TEMP);
3642 sv_utf8_downgrade(tsv, FALSE);
3643 tmps = SvPV_const(tsv, len);
3645 # ifdef USE_ITHREADS
3647 if (!PL_reentrant_buffer->_crypt_struct_buffer) {
3648 /* This should be threadsafe because in ithreads there is only
3649 * one thread per interpreter. If this would not be true,
3650 * we would need a mutex to protect this malloc. */
3651 PL_reentrant_buffer->_crypt_struct_buffer =
3652 (struct crypt_data *)safemalloc(sizeof(struct crypt_data));
3653 #if defined(__GLIBC__) || defined(__EMX__)
3654 if (PL_reentrant_buffer->_crypt_struct_buffer) {
3655 PL_reentrant_buffer->_crypt_struct_buffer->initialized = 0;
3656 #if (defined(__GLIBC__) && __GLIBC__ == 2) && \
3657 (defined(__GLIBC_MINOR__) && __GLIBC_MINOR__ >= 2 && __GLIBC_MINOR__ < 4)
3658 /* work around glibc-2.2.5 bug, has been fixed at some
3659 * time in glibc-2.3.X */
3660 PL_reentrant_buffer->_crypt_struct_buffer->current_saltbits = 0;
3665 # endif /* HAS_CRYPT_R */
3666 # endif /* USE_ITHREADS */
3668 sv_setpv(TARG, fcrypt(tmps, SvPV_nolen_const(right)));
3670 sv_setpv(TARG, PerlProc_crypt(tmps, SvPV_nolen_const(right)));
3677 "The crypt() function is unimplemented due to excessive paranoia.");
3681 /* Generally UTF-8 and UTF-EBCDIC are indistinguishable at this level. So
3682 * most comments below say UTF-8, when in fact they mean UTF-EBCDIC as well */
3685 /* also used for: pp_lcfirst() */
3689 /* Actually is both lcfirst() and ucfirst(). Only the first character
3690 * changes. This means that possibly we can change in-place, ie., just
3691 * take the source and change that one character and store it back, but not
3692 * if read-only etc, or if the length changes */
3696 STRLEN slen; /* slen is the byte length of the whole SV. */
3699 bool inplace; /* ? Convert first char only, in-place */
3700 bool doing_utf8 = FALSE; /* ? using utf8 */
3701 bool convert_source_to_utf8 = FALSE; /* ? need to convert */
3702 const int op_type = PL_op->op_type;
3705 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
3706 STRLEN ulen; /* ulen is the byte length of the original Unicode character
3707 * stored as UTF-8 at s. */
3708 STRLEN tculen; /* tculen is the byte length of the freshly titlecased (or
3709 * lowercased) character stored in tmpbuf. May be either
3710 * UTF-8 or not, but in either case is the number of bytes */
3712 s = (const U8*)SvPV_const(source, slen);
3714 /* We may be able to get away with changing only the first character, in
3715 * place, but not if read-only, etc. Later we may discover more reasons to
3716 * not convert in-place. */
3717 inplace = !SvREADONLY(source) && SvPADTMP(source);
3719 #ifdef USE_LOCALE_CTYPE
3721 if (IN_LC_RUNTIME(LC_CTYPE)) {
3722 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
3727 /* First calculate what the changed first character should be. This affects
3728 * whether we can just swap it out, leaving the rest of the string unchanged,
3729 * or even if have to convert the dest to UTF-8 when the source isn't */
3731 if (! slen) { /* If empty */
3732 need = 1; /* still need a trailing NUL */
3736 else if (DO_UTF8(source)) { /* Is the source utf8? */
3739 if (op_type == OP_UCFIRST) {
3740 #ifdef USE_LOCALE_CTYPE
3741 _toTITLE_utf8_flags(s, s +slen, tmpbuf, &tculen, IN_LC_RUNTIME(LC_CTYPE));
3743 _toTITLE_utf8_flags(s, s +slen, tmpbuf, &tculen, 0);
3747 #ifdef USE_LOCALE_CTYPE
3748 _toLOWER_utf8_flags(s, s + slen, tmpbuf, &tculen, IN_LC_RUNTIME(LC_CTYPE));
3750 _toLOWER_utf8_flags(s, s + slen, tmpbuf, &tculen, 0);
3754 /* we can't do in-place if the length changes. */
3755 if (ulen != tculen) inplace = FALSE;
3756 need = slen + 1 - ulen + tculen;
3758 else { /* Non-zero length, non-UTF-8, Need to consider locale and if
3759 * latin1 is treated as caseless. Note that a locale takes
3761 ulen = 1; /* Original character is 1 byte */
3762 tculen = 1; /* Most characters will require one byte, but this will
3763 * need to be overridden for the tricky ones */
3766 if (op_type == OP_LCFIRST) {
3768 /* lower case the first letter: no trickiness for any character */
3769 #ifdef USE_LOCALE_CTYPE
3770 if (IN_LC_RUNTIME(LC_CTYPE)) {
3771 *tmpbuf = toLOWER_LC(*s);
3776 *tmpbuf = (IN_UNI_8_BIT)
3777 ? toLOWER_LATIN1(*s)
3781 #ifdef USE_LOCALE_CTYPE
3783 else if (IN_LC_RUNTIME(LC_CTYPE)) {
3784 if (IN_UTF8_CTYPE_LOCALE) {
3788 *tmpbuf = (U8) toUPPER_LC(*s); /* This would be a bug if any
3789 locales have upper and title case
3793 else if (! IN_UNI_8_BIT) {
3794 *tmpbuf = toUPPER(*s); /* Returns caseless for non-ascii, or
3795 * on EBCDIC machines whatever the
3796 * native function does */
3799 /* Here, is ucfirst non-UTF-8, not in locale (unless that locale is
3800 * UTF-8, which we treat as not in locale), and cased latin1 */
3802 #ifdef USE_LOCALE_CTYPE
3806 title_ord = _to_upper_title_latin1(*s, tmpbuf, &tculen, 's');
3808 assert(tculen == 2);
3810 /* If the result is an upper Latin1-range character, it can
3811 * still be represented in one byte, which is its ordinal */
3812 if (UTF8_IS_DOWNGRADEABLE_START(*tmpbuf)) {
3813 *tmpbuf = (U8) title_ord;
3817 /* Otherwise it became more than one ASCII character (in
3818 * the case of LATIN_SMALL_LETTER_SHARP_S) or changed to
3819 * beyond Latin1, so the number of bytes changed, so can't
3820 * replace just the first character in place. */
3823 /* If the result won't fit in a byte, the entire result
3824 * will have to be in UTF-8. Assume worst case sizing in
3825 * conversion. (all latin1 characters occupy at most two
3827 if (title_ord > 255) {
3829 convert_source_to_utf8 = TRUE;
3830 need = slen * 2 + 1;
3832 /* The (converted) UTF-8 and UTF-EBCDIC lengths of all
3833 * (both) characters whose title case is above 255 is
3837 else { /* LATIN_SMALL_LETTER_SHARP_S expands by 1 byte */
3838 need = slen + 1 + 1;
3842 } /* End of use Unicode (Latin1) semantics */
3843 } /* End of changing the case of the first character */
3845 /* Here, have the first character's changed case stored in tmpbuf. Ready to
3846 * generate the result */
3849 /* We can convert in place. This means we change just the first
3850 * character without disturbing the rest; no need to grow */
3852 s = d = (U8*)SvPV_force_nomg(source, slen);
3858 /* Here, we can't convert in place; we earlier calculated how much
3859 * space we will need, so grow to accommodate that */
3860 SvUPGRADE(dest, SVt_PV);
3861 d = (U8*)SvGROW(dest, need);
3862 (void)SvPOK_only(dest);
3869 if (! convert_source_to_utf8) {
3871 /* Here both source and dest are in UTF-8, but have to create
3872 * the entire output. We initialize the result to be the
3873 * title/lower cased first character, and then append the rest
3875 sv_setpvn(dest, (char*)tmpbuf, tculen);
3877 sv_catpvn(dest, (char*)(s + ulen), slen - ulen);
3881 const U8 *const send = s + slen;
3883 /* Here the dest needs to be in UTF-8, but the source isn't,
3884 * except we earlier UTF-8'd the first character of the source
3885 * into tmpbuf. First put that into dest, and then append the
3886 * rest of the source, converting it to UTF-8 as we go. */
3888 /* Assert tculen is 2 here because the only two characters that
3889 * get to this part of the code have 2-byte UTF-8 equivalents */
3891 *d++ = *(tmpbuf + 1);
3892 s++; /* We have just processed the 1st char */
3894 for (; s < send; s++) {
3895 d = uvchr_to_utf8(d, *s);
3898 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
3902 else { /* in-place UTF-8. Just overwrite the first character */
3903 Copy(tmpbuf, d, tculen, U8);
3904 SvCUR_set(dest, need - 1);
3908 else { /* Neither source nor dest are in or need to be UTF-8 */
3910 if (inplace) { /* in-place, only need to change the 1st char */
3913 else { /* Not in-place */
3915 /* Copy the case-changed character(s) from tmpbuf */
3916 Copy(tmpbuf, d, tculen, U8);
3917 d += tculen - 1; /* Code below expects d to point to final
3918 * character stored */
3921 else { /* empty source */
3922 /* See bug #39028: Don't taint if empty */
3926 /* In a "use bytes" we don't treat the source as UTF-8, but, still want
3927 * the destination to retain that flag */
3928 if (SvUTF8(source) && ! IN_BYTES)
3931 if (!inplace) { /* Finish the rest of the string, unchanged */
3932 /* This will copy the trailing NUL */
3933 Copy(s + 1, d + 1, slen, U8);
3934 SvCUR_set(dest, need - 1);
3937 #ifdef USE_LOCALE_CTYPE
3938 if (IN_LC_RUNTIME(LC_CTYPE)) {
3943 if (dest != source && SvTAINTED(source))
3949 /* There's so much setup/teardown code common between uc and lc, I wonder if
3950 it would be worth merging the two, and just having a switch outside each
3951 of the three tight loops. There is less and less commonality though */
3964 if ( SvPADTMP(source)
3965 && !SvREADONLY(source) && SvPOK(source)
3968 #ifdef USE_LOCALE_CTYPE
3969 (IN_LC_RUNTIME(LC_CTYPE))
3970 ? ! IN_UTF8_CTYPE_LOCALE
3976 /* We can convert in place. The reason we can't if in UNI_8_BIT is to
3977 * make the loop tight, so we overwrite the source with the dest before
3978 * looking at it, and we need to look at the original source
3979 * afterwards. There would also need to be code added to handle
3980 * switching to not in-place in midstream if we run into characters
3981 * that change the length. Since being in locale overrides UNI_8_BIT,
3982 * that latter becomes irrelevant in the above test; instead for
3983 * locale, the size can't normally change, except if the locale is a
3986 s = d = (U8*)SvPV_force_nomg(source, len);
3993 s = (const U8*)SvPV_nomg_const(source, len);
3996 SvUPGRADE(dest, SVt_PV);
3997 d = (U8*)SvGROW(dest, min);
3998 (void)SvPOK_only(dest);
4003 #ifdef USE_LOCALE_CTYPE
4005 if (IN_LC_RUNTIME(LC_CTYPE)) {
4006 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
4011 /* Overloaded values may have toggled the UTF-8 flag on source, so we need
4012 to check DO_UTF8 again here. */
4014 if (DO_UTF8(source)) {
4015 const U8 *const send = s + len;
4016 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
4018 /* All occurrences of these are to be moved to follow any other marks.
4019 * This is context-dependent. We may not be passed enough context to
4020 * move the iota subscript beyond all of them, but we do the best we can
4021 * with what we're given. The result is always better than if we
4022 * hadn't done this. And, the problem would only arise if we are
4023 * passed a character without all its combining marks, which would be
4024 * the caller's mistake. The information this is based on comes from a
4025 * comment in Unicode SpecialCasing.txt, (and the Standard's text
4026 * itself) and so can't be checked properly to see if it ever gets
4027 * revised. But the likelihood of it changing is remote */
4028 bool in_iota_subscript = FALSE;
4034 if (in_iota_subscript && ! _is_utf8_mark(s)) {
4036 /* A non-mark. Time to output the iota subscript */
4037 Copy(GREEK_CAPITAL_LETTER_IOTA_UTF8, d, capital_iota_len, U8);
4038 d += capital_iota_len;
4039 in_iota_subscript = FALSE;
4042 /* Then handle the current character. Get the changed case value
4043 * and copy it to the output buffer */
4046 #ifdef USE_LOCALE_CTYPE
4047 uv = _toUPPER_utf8_flags(s, send, tmpbuf, &ulen, IN_LC_RUNTIME(LC_CTYPE));
4049 uv = _toUPPER_utf8_flags(s, send, tmpbuf, &ulen, 0);
4051 #define GREEK_CAPITAL_LETTER_IOTA 0x0399
4052 #define COMBINING_GREEK_YPOGEGRAMMENI 0x0345
4053 if (uv == GREEK_CAPITAL_LETTER_IOTA
4054 && utf8_to_uvchr_buf(s, send, 0) == COMBINING_GREEK_YPOGEGRAMMENI)
4056 in_iota_subscript = TRUE;
4059 if (ulen > u && (SvLEN(dest) < (min += ulen - u))) {
4060 /* If the eventually required minimum size outgrows the
4061 * available space, we need to grow. */
4062 const UV o = d - (U8*)SvPVX_const(dest);
4064 /* If someone uppercases one million U+03B0s we SvGROW()
4065 * one million times. Or we could try guessing how much to
4066 * allocate without allocating too much. Such is life.
4067 * See corresponding comment in lc code for another option
4069 d = o + (U8*) SvGROW(dest, min);
4071 Copy(tmpbuf, d, ulen, U8);
4076 if (in_iota_subscript) {
4077 Copy(GREEK_CAPITAL_LETTER_IOTA_UTF8, d, capital_iota_len, U8);
4078 d += capital_iota_len;
4083 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
4085 else { /* Not UTF-8 */
4087 const U8 *const send = s + len;
4089 /* Use locale casing if in locale; regular style if not treating
4090 * latin1 as having case; otherwise the latin1 casing. Do the
4091 * whole thing in a tight loop, for speed, */
4092 #ifdef USE_LOCALE_CTYPE
4093 if (IN_LC_RUNTIME(LC_CTYPE)) {
4094 if (IN_UTF8_CTYPE_LOCALE) {
4097 for (; s < send; d++, s++)
4098 *d = (U8) toUPPER_LC(*s);
4102 if (! IN_UNI_8_BIT) {
4103 for (; s < send; d++, s++) {
4108 #ifdef USE_LOCALE_CTYPE
4111 for (; s < send; d++, s++) {
4112 *d = toUPPER_LATIN1_MOD(*s);
4113 if (LIKELY(*d != LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
4117 /* The mainstream case is the tight loop above. To avoid
4118 * extra tests in that, all three characters that require
4119 * special handling are mapped by the MOD to the one tested
4121 * Use the source to distinguish between the three cases */
4123 #if UNICODE_MAJOR_VERSION > 2 \
4124 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
4125 && UNICODE_DOT_DOT_VERSION >= 8)
4126 if (*s == LATIN_SMALL_LETTER_SHARP_S) {
4128 /* uc() of this requires 2 characters, but they are
4129 * ASCII. If not enough room, grow the string */
4130 if (SvLEN(dest) < ++min) {
4131 const UV o = d - (U8*)SvPVX_const(dest);
4132 d = o + (U8*) SvGROW(dest, min);
4134 *d++ = 'S'; *d = 'S'; /* upper case is 'SS' */
4135 continue; /* Back to the tight loop; still in ASCII */
4139 /* The other two special handling characters have their
4140 * upper cases outside the latin1 range, hence need to be
4141 * in UTF-8, so the whole result needs to be in UTF-8. So,
4142 * here we are somewhere in the middle of processing a
4143 * non-UTF-8 string, and realize that we will have to convert
4144 * the whole thing to UTF-8. What to do? There are
4145 * several possibilities. The simplest to code is to
4146 * convert what we have so far, set a flag, and continue on
4147 * in the loop. The flag would be tested each time through
4148 * the loop, and if set, the next character would be
4149 * converted to UTF-8 and stored. But, I (khw) didn't want
4150 * to slow down the mainstream case at all for this fairly
4151 * rare case, so I didn't want to add a test that didn't
4152 * absolutely have to be there in the loop, besides the
4153 * possibility that it would get too complicated for
4154 * optimizers to deal with. Another possibility is to just
4155 * give up, convert the source to UTF-8, and restart the
4156 * function that way. Another possibility is to convert
4157 * both what has already been processed and what is yet to
4158 * come separately to UTF-8, then jump into the loop that
4159 * handles UTF-8. But the most efficient time-wise of the
4160 * ones I could think of is what follows, and turned out to
4161 * not require much extra code. */
4163 /* Convert what we have so far into UTF-8, telling the
4164 * function that we know it should be converted, and to
4165 * allow extra space for what we haven't processed yet.
4166 * Assume the worst case space requirements for converting
4167 * what we haven't processed so far: that it will require
4168 * two bytes for each remaining source character, plus the
4169 * NUL at the end. This may cause the string pointer to
4170 * move, so re-find it. */
4172 len = d - (U8*)SvPVX_const(dest);
4173 SvCUR_set(dest, len);
4174 len = sv_utf8_upgrade_flags_grow(dest,
4175 SV_GMAGIC|SV_FORCE_UTF8_UPGRADE,
4177 d = (U8*)SvPVX(dest) + len;
4179 /* Now process the remainder of the source, converting to
4180 * upper and UTF-8. If a resulting byte is invariant in
4181 * UTF-8, output it as-is, otherwise convert to UTF-8 and
4182 * append it to the output. */
4183 for (; s < send; s++) {
4184 (void) _to_upper_title_latin1(*s, d, &len, 'S');
4188 /* Here have processed the whole source; no need to continue
4189 * with the outer loop. Each character has been converted
4190 * to upper case and converted to UTF-8 */
4193 } /* End of processing all latin1-style chars */
4194 } /* End of processing all chars */
4195 } /* End of source is not empty */
4197 if (source != dest) {
4198 *d = '\0'; /* Here d points to 1 after last char, add NUL */
4199 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
4201 } /* End of isn't utf8 */
4202 #ifdef USE_LOCALE_CTYPE
4203 if (IN_LC_RUNTIME(LC_CTYPE)) {
4208 if (dest != source && SvTAINTED(source))
4226 if ( SvPADTMP(source)
4227 && !SvREADONLY(source) && SvPOK(source)
4228 && !DO_UTF8(source)) {
4230 /* We can convert in place, as lowercasing anything in the latin1 range
4231 * (or else DO_UTF8 would have been on) doesn't lengthen it */
4233 s = d = (U8*)SvPV_force_nomg(source, len);
4240 s = (const U8*)SvPV_nomg_const(source, len);
4243 SvUPGRADE(dest, SVt_PV);
4244 d = (U8*)SvGROW(dest, min);
4245 (void)SvPOK_only(dest);
4250 #ifdef USE_LOCALE_CTYPE
4252 if (IN_LC_RUNTIME(LC_CTYPE)) {
4253 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
4258 /* Overloaded values may have toggled the UTF-8 flag on source, so we need
4259 to check DO_UTF8 again here. */
4261 if (DO_UTF8(source)) {
4262 const U8 *const send = s + len;
4263 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
4266 const STRLEN u = UTF8SKIP(s);
4269 #ifdef USE_LOCALE_CTYPE
4270 _toLOWER_utf8_flags(s, send, tmpbuf, &ulen, IN_LC_RUNTIME(LC_CTYPE));
4272 _toLOWER_utf8_flags(s, send, tmpbuf, &ulen, 0);
4275 /* Here is where we would do context-sensitive actions. See the
4276 * commit message for 86510fb15 for why there isn't any */
4278 if (ulen > u && (SvLEN(dest) < (min += ulen - u))) {
4280 /* If the eventually required minimum size outgrows the
4281 * available space, we need to grow. */
4282 const UV o = d - (U8*)SvPVX_const(dest);
4284 /* If someone lowercases one million U+0130s we SvGROW() one
4285 * million times. Or we could try guessing how much to
4286 * allocate without allocating too much. Such is life.
4287 * Another option would be to grow an extra byte or two more
4288 * each time we need to grow, which would cut down the million
4289 * to 500K, with little waste */
4290 d = o + (U8*) SvGROW(dest, min);
4293 /* Copy the newly lowercased letter to the output buffer we're
4295 Copy(tmpbuf, d, ulen, U8);
4298 } /* End of looping through the source string */
4301 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
4302 } else { /* Not utf8 */
4304 const U8 *const send = s + len;
4306 /* Use locale casing if in locale; regular style if not treating
4307 * latin1 as having case; otherwise the latin1 casing. Do the
4308 * whole thing in a tight loop, for speed, */
4309 #ifdef USE_LOCALE_CTYPE
4310 if (IN_LC_RUNTIME(LC_CTYPE)) {
4311 for (; s < send; d++, s++)
4312 *d = toLOWER_LC(*s);
4316 if (! IN_UNI_8_BIT) {
4317 for (; s < send; d++, s++) {
4322 for (; s < send; d++, s++) {
4323 *d = toLOWER_LATIN1(*s);
4327 if (source != dest) {
4329 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
4332 #ifdef USE_LOCALE_CTYPE
4333 if (IN_LC_RUNTIME(LC_CTYPE)) {
4338 if (dest != source && SvTAINTED(source))
4347 SV * const sv = TOPs;
4349 const char *s = SvPV_const(sv,len);
4351 SvUTF8_off(TARG); /* decontaminate */
4354 SvUPGRADE(TARG, SVt_PV);
4355 SvGROW(TARG, (len * 2) + 1);
4359 STRLEN ulen = UTF8SKIP(s);
4360 bool to_quote = FALSE;
4362 if (UTF8_IS_INVARIANT(*s)) {
4363 if (_isQUOTEMETA(*s)) {
4367 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, s + len)) {
4369 #ifdef USE_LOCALE_CTYPE
4370 /* In locale, we quote all non-ASCII Latin1 chars.
4371 * Otherwise use the quoting rules */
4373 IN_LC_RUNTIME(LC_CTYPE)
4376 _isQUOTEMETA(EIGHT_BIT_UTF8_TO_NATIVE(*s, *(s + 1))))
4381 else if (is_QUOTEMETA_high(s)) {
4396 else if (IN_UNI_8_BIT) {
4398 if (_isQUOTEMETA(*s))
4404 /* For non UNI_8_BIT (and hence in locale) just quote all \W
4405 * including everything above ASCII */
4407 if (!isWORDCHAR_A(*s))
4413 SvCUR_set(TARG, d - SvPVX_const(TARG));
4414 (void)SvPOK_only_UTF8(TARG);
4417 sv_setpvn(TARG, s, len);
4433 U8 tmpbuf[UTF8_MAXBYTES_CASE + 1];
4434 #if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
4435 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
4436 || UNICODE_DOT_DOT_VERSION > 0)
4437 const bool full_folding = TRUE; /* This variable is here so we can easily
4438 move to more generality later */
4440 const bool full_folding = FALSE;
4442 const U8 flags = ( full_folding ? FOLD_FLAGS_FULL : 0 )
4443 #ifdef USE_LOCALE_CTYPE
4444 | ( IN_LC_RUNTIME(LC_CTYPE) ? FOLD_FLAGS_LOCALE : 0 )
4448 /* This is a facsimile of pp_lc, but with a thousand bugs thanks to me.
4449 * You are welcome(?) -Hugmeir
4457 s = (const U8*)SvPV_nomg_const(source, len);
4459 if (ckWARN(WARN_UNINITIALIZED))
4460 report_uninit(source);
4467 SvUPGRADE(dest, SVt_PV);
4468 d = (U8*)SvGROW(dest, min);
4469 (void)SvPOK_only(dest);
4475 #ifdef USE_LOCALE_CTYPE
4477 if ( IN_LC_RUNTIME(LC_CTYPE) ) { /* Under locale */
4478 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
4483 if (DO_UTF8(source)) { /* UTF-8 flagged string. */
4485 const STRLEN u = UTF8SKIP(s);
4488 _toFOLD_utf8_flags(s, send, tmpbuf, &ulen, flags);
4490 if (ulen > u && (SvLEN(dest) < (min += ulen - u))) {
4491 const UV o = d - (U8*)SvPVX_const(dest);
4492 d = o + (U8*) SvGROW(dest, min);
4495 Copy(tmpbuf, d, ulen, U8);
4500 } /* Unflagged string */
4502 #ifdef USE_LOCALE_CTYPE
4503 if ( IN_LC_RUNTIME(LC_CTYPE) ) { /* Under locale */
4504 if (IN_UTF8_CTYPE_LOCALE) {
4505 goto do_uni_folding;
4507 for (; s < send; d++, s++)
4508 *d = (U8) toFOLD_LC(*s);
4512 if ( !IN_UNI_8_BIT ) { /* Under nothing, or bytes */
4513 for (; s < send; d++, s++)
4517 #ifdef USE_LOCALE_CTYPE
4520 /* For ASCII and the Latin-1 range, there's only two troublesome
4521 * folds, \x{DF} (\N{LATIN SMALL LETTER SHARP S}), which under full
4522 * casefolding becomes 'ss'; and \x{B5} (\N{MICRO SIGN}), which
4523 * under any fold becomes \x{3BC} (\N{GREEK SMALL LETTER MU}) --
4524 * For the rest, the casefold is their lowercase. */
4525 for (; s < send; d++, s++) {
4526 if (*s == MICRO_SIGN) {
4527 /* \N{MICRO SIGN}'s casefold is \N{GREEK SMALL LETTER MU},
4528 * which is outside of the latin-1 range. There's a couple
4529 * of ways to deal with this -- khw discusses them in
4530 * pp_lc/uc, so go there :) What we do here is upgrade what
4531 * we had already casefolded, then enter an inner loop that
4532 * appends the rest of the characters as UTF-8. */
4533 len = d - (U8*)SvPVX_const(dest);
4534 SvCUR_set(dest, len);
4535 len = sv_utf8_upgrade_flags_grow(dest,
4536 SV_GMAGIC|SV_FORCE_UTF8_UPGRADE,
4537 /* The max expansion for latin1
4538 * chars is 1 byte becomes 2 */
4540 d = (U8*)SvPVX(dest) + len;
4542 Copy(GREEK_SMALL_LETTER_MU_UTF8, d, small_mu_len, U8);
4545 for (; s < send; s++) {
4547 UV fc = _to_uni_fold_flags(*s, tmpbuf, &ulen, flags);
4548 if UVCHR_IS_INVARIANT(fc) {
4550 && *s == LATIN_SMALL_LETTER_SHARP_S)
4559 Copy(tmpbuf, d, ulen, U8);
4565 else if (full_folding && *s == LATIN_SMALL_LETTER_SHARP_S) {
4566 /* Under full casefolding, LATIN SMALL LETTER SHARP S
4567 * becomes "ss", which may require growing the SV. */
4568 if (SvLEN(dest) < ++min) {
4569 const UV o = d - (U8*)SvPVX_const(dest);
4570 d = o + (U8*) SvGROW(dest, min);
4575 else { /* If it's not one of those two, the fold is their lower
4577 *d = toLOWER_LATIN1(*s);
4583 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
4585 #ifdef USE_LOCALE_CTYPE
4586 if (IN_LC_RUNTIME(LC_CTYPE)) {
4591 if (SvTAINTED(source))
4601 dSP; dMARK; dORIGMARK;
4602 AV *const av = MUTABLE_AV(POPs);
4603 const I32 lval = (PL_op->op_flags & OPf_MOD || LVRET);
4605 if (SvTYPE(av) == SVt_PVAV) {
4606 const bool localizing = PL_op->op_private & OPpLVAL_INTRO;
4607 bool can_preserve = FALSE;
4613 can_preserve = SvCANEXISTDELETE(av);
4616 if (lval && localizing) {
4619 for (svp = MARK + 1; svp <= SP; svp++) {
4620 const SSize_t elem = SvIV(*svp);
4624 if (max > AvMAX(av))
4628 while (++MARK <= SP) {
4630 SSize_t elem = SvIV(*MARK);
4631 bool preeminent = TRUE;
4633 if (localizing && can_preserve) {
4634 /* If we can determine whether the element exist,
4635 * Try to preserve the existenceness of a tied array
4636 * element by using EXISTS and DELETE if possible.
4637 * Fallback to FETCH and STORE otherwise. */
4638 preeminent = av_exists(av, elem);
4641 svp = av_fetch(av, elem, lval);
4644 DIE(aTHX_ PL_no_aelem, elem);
4647 save_aelem(av, elem, svp);
4649 SAVEADELETE(av, elem);
4652 *MARK = svp ? *svp : &PL_sv_undef;
4655 if (GIMME_V != G_ARRAY) {
4657 *++MARK = SP > ORIGMARK ? *SP : &PL_sv_undef;
4666 AV *const av = MUTABLE_AV(POPs);
4667 I32 lval = (PL_op->op_flags & OPf_MOD);
4668 SSize_t items = SP - MARK;
4670 if (PL_op->op_private & OPpMAYBE_LVSUB) {
4671 const I32 flags = is_lvalue_sub();
4673 if (!(flags & OPpENTERSUB_INARGS))
4674 /* diag_listed_as: Can't modify %s in %s */
4675 Perl_croak(aTHX_ "Can't modify index/value array slice in list assignment");
4682 *(MARK+items*2-1) = *(MARK+items);
4688 while (++MARK <= SP) {
4691 svp = av_fetch(av, SvIV(*MARK), lval);
4693 if (!svp || !*svp || *svp == &PL_sv_undef) {
4694 DIE(aTHX_ PL_no_aelem, SvIV(*MARK));
4696 *MARK = sv_mortalcopy(*MARK);
4698 *++MARK = svp ? *svp : &PL_sv_undef;
4700 if (GIMME_V != G_ARRAY) {
4701 MARK = SP - items*2;
4702 *++MARK = items > 0 ? *SP : &PL_sv_undef;
4712 AV *array = MUTABLE_AV(POPs);
4713 const U8 gimme = GIMME_V;
4714 IV *iterp = Perl_av_iter_p(aTHX_ array);
4715 const IV current = (*iterp)++;
4717 if (current > av_tindex(array)) {
4719 if (gimme == G_SCALAR)
4727 if (gimme == G_ARRAY) {
4728 SV **const element = av_fetch(array, current, 0);
4729 PUSHs(element ? *element : &PL_sv_undef);
4734 /* also used for: pp_avalues()*/
4738 AV *array = MUTABLE_AV(POPs);
4739 const U8 gimme = GIMME_V;
4741 *Perl_av_iter_p(aTHX_ array) = 0;
4743 if (gimme == G_SCALAR) {
4745 PUSHi(av_tindex(array) + 1);
4747 else if (gimme == G_ARRAY) {
4748 if (UNLIKELY(PL_op->op_private & OPpMAYBE_LVSUB)) {
4749 const I32 flags = is_lvalue_sub();
4750 if (flags && !(flags & OPpENTERSUB_INARGS))
4751 /* diag_listed_as: Can't modify %s in %s */
4753 "Can't modify keys on array in list assignment");
4756 IV n = Perl_av_len(aTHX_ array);
4761 if ( PL_op->op_type == OP_AKEYS
4762 || ( PL_op->op_type == OP_AVHVSWITCH
4763 && (PL_op->op_private & 3) + OP_AEACH == OP_AKEYS ))
4765 for (i = 0; i <= n; i++) {
4770 for (i = 0; i <= n; i++) {
4771 SV *const *const elem = Perl_av_fetch(aTHX_ array, i, 0);
4772 PUSHs(elem ? *elem : &PL_sv_undef);
4780 /* Associative arrays. */
4785 HV * hash = MUTABLE_HV(POPs);
4787 const U8 gimme = GIMME_V;
4789 entry = hv_iternext(hash);
4793 SV* const sv = hv_iterkeysv(entry);
4795 if (gimme == G_ARRAY) {
4797 val = hv_iterval(hash, entry);
4801 else if (gimme == G_SCALAR)
4808 S_do_delete_local(pTHX)
4811 const U8 gimme = GIMME_V;
4814 const bool sliced = !!(PL_op->op_private & OPpSLICE);
4815 SV **unsliced_keysv = sliced ? NULL : sp--;
4816 SV * const osv = POPs;
4817 SV **mark = sliced ? PL_stack_base + POPMARK : unsliced_keysv-1;
4819 const bool tied = SvRMAGICAL(osv)
4820 && mg_find((const SV *)osv, PERL_MAGIC_tied);
4821 const bool can_preserve = SvCANEXISTDELETE(osv);
4822 const U32 type = SvTYPE(osv);
4823 SV ** const end = sliced ? SP : unsliced_keysv;
4825 if (type == SVt_PVHV) { /* hash element */
4826 HV * const hv = MUTABLE_HV(osv);
4827 while (++MARK <= end) {
4828 SV * const keysv = *MARK;
4830 bool preeminent = TRUE;
4832 preeminent = hv_exists_ent(hv, keysv, 0);
4834 HE *he = hv_fetch_ent(hv, keysv, 1, 0);
4841 sv = hv_delete_ent(hv, keysv, 0, 0);
4843 SvREFCNT_inc_simple_void(sv); /* De-mortalize */
4846 if (!sv) DIE(aTHX_ PL_no_helem_sv, SVfARG(keysv));
4847 save_helem_flags(hv, keysv, &sv, SAVEf_KEEPOLDELEM);
4849 *MARK = sv_mortalcopy(sv);
4855 SAVEHDELETE(hv, keysv);
4856 *MARK = &PL_sv_undef;
4860 else if (type == SVt_PVAV) { /* array element */
4861 if (PL_op->op_flags & OPf_SPECIAL) {
4862 AV * const av = MUTABLE_AV(osv);
4863 while (++MARK <= end) {
4864 SSize_t idx = SvIV(*MARK);
4866 bool preeminent = TRUE;
4868 preeminent = av_exists(av, idx);
4870 SV **svp = av_fetch(av, idx, 1);
4877 sv = av_delete(av, idx, 0);
4879 SvREFCNT_inc_simple_void(sv); /* De-mortalize */
4882 save_aelem_flags(av, idx, &sv, SAVEf_KEEPOLDELEM);
4884 *MARK = sv_mortalcopy(sv);
4890 SAVEADELETE(av, idx);
4891 *MARK = &PL_sv_undef;
4896 DIE(aTHX_ "panic: avhv_delete no longer supported");
4899 DIE(aTHX_ "Not a HASH reference");
4901 if (gimme == G_VOID)
4903 else if (gimme == G_SCALAR) {
4908 *++MARK = &PL_sv_undef;
4912 else if (gimme != G_VOID)
4913 PUSHs(*unsliced_keysv);
4924 if (PL_op->op_private & OPpLVAL_INTRO)
4925 return do_delete_local();
4928 discard = (gimme == G_VOID) ? G_DISCARD : 0;
4930 if (PL_op->op_private & (OPpSLICE|OPpKVSLICE)) {
4932 HV * const hv = MUTABLE_HV(POPs);
4933 const U32 hvtype = SvTYPE(hv);
4935 if (PL_op->op_private & OPpKVSLICE) {
4936 SSize_t items = SP - MARK;
4940 *(MARK+items*2-1) = *(MARK+items);
4947 if (hvtype == SVt_PVHV) { /* hash element */
4948 while ((MARK += (1+skip)) <= SP) {
4949 SV * const sv = hv_delete_ent(hv, *(MARK-skip), discard, 0);
4950 *MARK = sv ? sv : &PL_sv_undef;
4953 else if (hvtype == SVt_PVAV) { /* array element */
4954 if (PL_op->op_flags & OPf_SPECIAL) {
4955 while ((MARK += (1+skip)) <= SP) {
4956 SV * const sv = av_delete(MUTABLE_AV(hv), SvIV(*(MARK-skip)), discard);
4957 *MARK = sv ? sv : &PL_sv_undef;
4962 DIE(aTHX_ "Not a HASH reference");
4965 else if (gimme == G_SCALAR) {
4970 *++MARK = &PL_sv_undef;
4976 HV * const hv = MUTABLE_HV(POPs);
4978 if (SvTYPE(hv) == SVt_PVHV)
4979 sv = hv_delete_ent(hv, keysv, discard, 0);
4980 else if (SvTYPE(hv) == SVt_PVAV) {
4981 if (PL_op->op_flags & OPf_SPECIAL)
4982 sv = av_delete(MUTABLE_AV(hv), SvIV(keysv), discard);
4984 DIE(aTHX_ "panic: avhv_delete no longer supported");
4987 DIE(aTHX_ "Not a HASH reference");
5002 if (UNLIKELY( PL_op->op_private & OPpEXISTS_SUB )) {
5004 SV * const sv = POPs;
5005 CV * const cv = sv_2cv(sv, &hv, &gv, 0);
5008 if (gv && isGV(gv) && GvCV(gv) && !GvCVGEN(gv))
5013 hv = MUTABLE_HV(POPs);
5014 if (LIKELY( SvTYPE(hv) == SVt_PVHV )) {
5015 if (hv_exists_ent(hv, tmpsv, 0))
5018 else if (SvTYPE(hv) == SVt_PVAV) {
5019 if (PL_op->op_flags & OPf_SPECIAL) { /* array element */
5020 if (av_exists(MUTABLE_AV(hv), SvIV(tmpsv)))
5025 DIE(aTHX_ "Not a HASH reference");
5032 dSP; dMARK; dORIGMARK;
5033 HV * const hv = MUTABLE_HV(POPs);
5034 const I32 lval = (PL_op->op_flags & OPf_MOD || LVRET);
5035 const bool localizing = PL_op->op_private & OPpLVAL_INTRO;
5036 bool can_preserve = FALSE;
5042 if (SvCANEXISTDELETE(hv))
5043 can_preserve = TRUE;
5046 while (++MARK <= SP) {
5047 SV * const keysv = *MARK;
5050 bool preeminent = TRUE;
5052 if (localizing && can_preserve) {
5053 /* If we can determine whether the element exist,
5054 * try to preserve the existenceness of a tied hash
5055 * element by using EXISTS and DELETE if possible.
5056 * Fallback to FETCH and STORE otherwise. */
5057 preeminent = hv_exists_ent(hv, keysv, 0);
5060 he = hv_fetch_ent(hv, keysv, lval, 0);
5061 svp = he ? &HeVAL(he) : NULL;
5064 if (!svp || !*svp || *svp == &PL_sv_undef) {
5065 DIE(aTHX_ PL_no_helem_sv, SVfARG(keysv));
5068 if (HvNAME_get(hv) && isGV_or_RVCV(*svp))
5069 save_gp(MUTABLE_GV(*svp), !(PL_op->op_flags & OPf_SPECIAL));
5070 else if (preeminent)
5071 save_helem_flags(hv, keysv, svp,
5072 (PL_op->op_flags & OPf_SPECIAL) ? 0 : SAVEf_SETMAGIC);
5074 SAVEHDELETE(hv, keysv);
5077 *MARK = svp && *svp ? *svp : &PL_sv_undef;
5079 if (GIMME_V != G_ARRAY) {
5081 *++MARK = SP > ORIGMARK ? *SP : &PL_sv_undef;
5090 HV * const hv = MUTABLE_HV(POPs);
5091 I32 lval = (PL_op->op_flags & OPf_MOD);
5092 SSize_t items = SP - MARK;
5094 if (PL_op->op_private & OPpMAYBE_LVSUB) {
5095 const I32 flags = is_lvalue_sub();
5097 if (!(flags & OPpENTERSUB_INARGS))
5098 /* diag_listed_as: Can't modify %s in %s */
5099 Perl_croak(aTHX_ "Can't modify key/value hash slice in %s assignment",
5100 GIMME_V == G_ARRAY ? "list" : "scalar");
5107 *(MARK+items*2-1) = *(MARK+items);
5113 while (++MARK <= SP) {
5114 SV * const keysv = *MARK;
5118 he = hv_fetch_ent(hv, keysv, lval, 0);
5119 svp = he ? &HeVAL(he) : NULL;
5122 if (!svp || !*svp || *svp == &PL_sv_undef) {
5123 DIE(aTHX_ PL_no_helem_sv, SVfARG(keysv));
5125 *MARK = sv_mortalcopy(*MARK);
5127 *++MARK = svp && *svp ? *svp : &PL_sv_undef;
5129 if (GIMME_V != G_ARRAY) {
5130 MARK = SP - items*2;
5131 *++MARK = items > 0 ? *SP : &PL_sv_undef;
5137 /* List operators. */
5141 I32 markidx = POPMARK;
5142 if (GIMME_V != G_ARRAY) {
5143 /* don't initialize mark here, EXTEND() may move the stack */
5146 EXTEND(SP, 1); /* in case no arguments, as in @empty */
5147 mark = PL_stack_base + markidx;
5149 *MARK = *SP; /* unwanted list, return last item */
5151 *MARK = &PL_sv_undef;
5161 SV ** const lastrelem = PL_stack_sp;
5162 SV ** const lastlelem = PL_stack_base + POPMARK;
5163 SV ** const firstlelem = PL_stack_base + POPMARK + 1;
5164 SV ** const firstrelem = lastlelem + 1;
5165 const U8 mod = PL_op->op_flags & OPf_MOD;
5167 const I32 max = lastrelem - lastlelem;
5170 if (GIMME_V != G_ARRAY) {
5171 if (lastlelem < firstlelem) {
5173 *firstlelem = &PL_sv_undef;
5176 I32 ix = SvIV(*lastlelem);
5179 if (ix < 0 || ix >= max)
5180 *firstlelem = &PL_sv_undef;
5182 *firstlelem = firstrelem[ix];
5189 SP = firstlelem - 1;
5193 for (lelem = firstlelem; lelem <= lastlelem; lelem++) {
5194 I32 ix = SvIV(*lelem);
5197 if (ix < 0 || ix >= max)
5198 *lelem = &PL_sv_undef;
5200 if (!(*lelem = firstrelem[ix]))
5201 *lelem = &PL_sv_undef;
5202 else if (mod && SvPADTMP(*lelem)) {
5203 *lelem = firstrelem[ix] = sv_mortalcopy(*lelem);
5214 const I32 items = SP - MARK;
5215 SV * const av = MUTABLE_SV(av_make(items, MARK+1));
5217 mXPUSHs((PL_op->op_flags & OPf_SPECIAL)
5218 ? newRV_noinc(av) : av);
5224 dSP; dMARK; dORIGMARK;
5225 HV* const hv = newHV();
5226 SV* const retval = sv_2mortal( PL_op->op_flags & OPf_SPECIAL
5227 ? newRV_noinc(MUTABLE_SV(hv))
5232 (MARK++, SvGMAGICAL(*MARK) ? sv_mortalcopy(*MARK) : *MARK);
5239 sv_setsv_nomg(val, *MARK);
5243 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Odd number of elements in anonymous hash");
5246 (void)hv_store_ent(hv,key,val,0);
5255 dSP; dMARK; dORIGMARK;
5256 int num_args = (SP - MARK);
5257 AV *ary = MUTABLE_AV(*++MARK);
5266 const MAGIC * const mg = SvTIED_mg((const SV *)ary, PERL_MAGIC_tied);
5269 return Perl_tied_method(aTHX_ SV_CONST(SPLICE), mark - 1, MUTABLE_SV(ary), mg,
5270 GIMME_V | TIED_METHOD_ARGUMENTS_ON_STACK,
5274 if (SvREADONLY(ary))
5275 Perl_croak_no_modify();
5280 offset = i = SvIV(*MARK);
5282 offset += AvFILLp(ary) + 1;
5284 DIE(aTHX_ PL_no_aelem, i);
5286 length = SvIVx(*MARK++);
5288 length += AvFILLp(ary) - offset + 1;
5294 length = AvMAX(ary) + 1; /* close enough to infinity */
5298 length = AvMAX(ary) + 1;
5300 if (offset > AvFILLp(ary) + 1) {
5302 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "splice() offset past end of array" );
5303 offset = AvFILLp(ary) + 1;
5305 after = AvFILLp(ary) + 1 - (offset + length);
5306 if (after < 0) { /* not that much array */
5307 length += after; /* offset+length now in array */
5313 /* At this point, MARK .. SP-1 is our new LIST */
5316 diff = newlen - length;
5317 if (newlen && !AvREAL(ary) && AvREIFY(ary))
5320 /* make new elements SVs now: avoid problems if they're from the array */
5321 for (dst = MARK, i = newlen; i; i--) {
5322 SV * const h = *dst;
5323 *dst++ = newSVsv(h);
5326 if (diff < 0) { /* shrinking the area */
5327 SV **tmparyval = NULL;
5329 Newx(tmparyval, newlen, SV*); /* so remember insertion */
5330 Copy(MARK, tmparyval, newlen, SV*);
5333 MARK = ORIGMARK + 1;
5334 if (GIMME_V == G_ARRAY) { /* copy return vals to stack */
5335 const bool real = cBOOL(AvREAL(ary));
5336 MEXTEND(MARK, length);
5338 EXTEND_MORTAL(length);
5339 for (i = 0, dst = MARK; i < length; i++) {
5340 if ((*dst = AvARRAY(ary)[i+offset])) {
5342 sv_2mortal(*dst); /* free them eventually */
5345 *dst = &PL_sv_undef;
5351 *MARK = AvARRAY(ary)[offset+length-1];
5354 for (i = length - 1, dst = &AvARRAY(ary)[offset]; i > 0; i--)
5355 SvREFCNT_dec(*dst++); /* free them now */
5358 *MARK = &PL_sv_undef;
5360 AvFILLp(ary) += diff;
5362 /* pull up or down? */
5364 if (offset < after) { /* easier to pull up */
5365 if (offset) { /* esp. if nothing to pull */
5366 src = &AvARRAY(ary)[offset-1];
5367 dst = src - diff; /* diff is negative */
5368 for (i = offset; i > 0; i--) /* can't trust Copy */
5372 AvARRAY(ary) = AvARRAY(ary) - diff; /* diff is negative */
5376 if (after) { /* anything to pull down? */
5377 src = AvARRAY(ary) + offset + length;
5378 dst = src + diff; /* diff is negative */
5379 Move(src, dst, after, SV*);
5381 dst = &AvARRAY(ary)[AvFILLp(ary)+1];
5382 /* avoid later double free */
5389 Copy( tmparyval, AvARRAY(ary) + offset, newlen, SV* );
5390 Safefree(tmparyval);
5393 else { /* no, expanding (or same) */
5394 SV** tmparyval = NULL;
5396 Newx(tmparyval, length, SV*); /* so remember deletion */
5397 Copy(AvARRAY(ary)+offset, tmparyval, length, SV*);
5400 if (diff > 0) { /* expanding */
5401 /* push up or down? */
5402 if (offset < after && diff <= AvARRAY(ary) - AvALLOC(ary)) {
5406 Move(src, dst, offset, SV*);
5408 AvARRAY(ary) = AvARRAY(ary) - diff;/* diff is positive */
5410 AvFILLp(ary) += diff;
5413 if (AvFILLp(ary) + diff >= AvMAX(ary)) /* oh, well */
5414 av_extend(ary, AvFILLp(ary) + diff);
5415 AvFILLp(ary) += diff;
5418 dst = AvARRAY(ary) + AvFILLp(ary);
5420 for (i = after; i; i--) {
5428 Copy( MARK, AvARRAY(ary) + offset, newlen, SV* );
5431 MARK = ORIGMARK + 1;
5432 if (GIMME_V == G_ARRAY) { /* copy return vals to stack */
5434 const bool real = cBOOL(AvREAL(ary));
5436 EXTEND_MORTAL(length);
5437 for (i = 0, dst = MARK; i < length; i++) {
5438 if ((*dst = tmparyval[i])) {
5440 sv_2mortal(*dst); /* free them eventually */
5442 else *dst = &PL_sv_undef;
5448 else if (length--) {
5449 *MARK = tmparyval[length];
5452 while (length-- > 0)
5453 SvREFCNT_dec(tmparyval[length]);
5456 *MARK = &PL_sv_undef;
5459 *MARK = &PL_sv_undef;
5460 Safefree(tmparyval);
5464 mg_set(MUTABLE_SV(ary));
5472 dSP; dMARK; dORIGMARK; dTARGET;
5473 AV * const ary = MUTABLE_AV(*++MARK);
5474 const MAGIC * const mg = SvTIED_mg((const SV *)ary, PERL_MAGIC_tied);
5477 *MARK-- = SvTIED_obj(MUTABLE_SV(ary), mg);
5480 ENTER_with_name("call_PUSH");
5481 call_sv(SV_CONST(PUSH),G_SCALAR|G_DISCARD|G_METHOD_NAMED);
5482 LEAVE_with_name("call_PUSH");
5483 /* SPAGAIN; not needed: SP is assigned to immediately below */
5486 /* PL_delaymagic is restored by JUMPENV_POP on dieing, so we
5487 * only need to save locally, not on the save stack */
5488 U16 old_delaymagic = PL_delaymagic;
5490 if (SvREADONLY(ary) && MARK < SP) Perl_croak_no_modify();
5491 PL_delaymagic = DM_DELAY;
5492 for (++MARK; MARK <= SP; MARK++) {
5494 if (*MARK) SvGETMAGIC(*MARK);
5497 sv_setsv_nomg(sv, *MARK);
5498 av_store(ary, AvFILLp(ary)+1, sv);
5500 if (PL_delaymagic & DM_ARRAY_ISA)
5501 mg_set(MUTABLE_SV(ary));
5502 PL_delaymagic = old_delaymagic;
5505 if (OP_GIMME(PL_op, 0) != G_VOID) {
5506 PUSHi( AvFILL(ary) + 1 );
5511 /* also used for: pp_pop()*/
5515 AV * const av = PL_op->op_flags & OPf_SPECIAL
5516 ? MUTABLE_AV(GvAVn(PL_defgv)) : MUTABLE_AV(POPs);
5517 SV * const sv = PL_op->op_type == OP_SHIFT ? av_shift(av) : av_pop(av);
5521 (void)sv_2mortal(sv);
5528 dSP; dMARK; dORIGMARK; dTARGET;
5529 AV *ary = MUTABLE_AV(*++MARK);
5530 const MAGIC * const mg = SvTIED_mg((const SV *)ary, PERL_MAGIC_tied);
5533 *MARK-- = SvTIED_obj(MUTABLE_SV(ary), mg);
5536 ENTER_with_name("call_UNSHIFT");
5537 call_sv(SV_CONST(UNSHIFT),G_SCALAR|G_DISCARD|G_METHOD_NAMED);
5538 LEAVE_with_name("call_UNSHIFT");
5539 /* SPAGAIN; not needed: SP is assigned to immediately below */
5542 /* PL_delaymagic is restored by JUMPENV_POP on dieing, so we
5543 * only need to save locally, not on the save stack */
5544 U16 old_delaymagic = PL_delaymagic;
5547 av_unshift(ary, SP - MARK);
5548 PL_delaymagic = DM_DELAY;
5550 SV * const sv = newSVsv(*++MARK);
5551 (void)av_store(ary, i++, sv);
5553 if (PL_delaymagic & DM_ARRAY_ISA)
5554 mg_set(MUTABLE_SV(ary));
5555 PL_delaymagic = old_delaymagic;
5558 if (OP_GIMME(PL_op, 0) != G_VOID) {
5559 PUSHi( AvFILL(ary) + 1 );
5568 if (GIMME_V == G_ARRAY) {
5569 if (PL_op->op_private & OPpREVERSE_INPLACE) {
5573 assert( MARK+1 == SP && *SP && SvTYPE(*SP) == SVt_PVAV);
5574 (void)POPMARK; /* remove mark associated with ex-OP_AASSIGN */
5575 av = MUTABLE_AV((*SP));
5576 /* In-place reversing only happens in void context for the array
5577 * assignment. We don't need to push anything on the stack. */
5580 if (SvMAGICAL(av)) {
5582 SV *tmp = sv_newmortal();
5583 /* For SvCANEXISTDELETE */
5586 bool can_preserve = SvCANEXISTDELETE(av);
5588 for (i = 0, j = av_tindex(av); i < j; ++i, --j) {
5592 if (!av_exists(av, i)) {
5593 if (av_exists(av, j)) {
5594 SV *sv = av_delete(av, j, 0);
5595 begin = *av_fetch(av, i, TRUE);
5596 sv_setsv_mg(begin, sv);
5600 else if (!av_exists(av, j)) {
5601 SV *sv = av_delete(av, i, 0);
5602 end = *av_fetch(av, j, TRUE);
5603 sv_setsv_mg(end, sv);
5608 begin = *av_fetch(av, i, TRUE);
5609 end = *av_fetch(av, j, TRUE);
5610 sv_setsv(tmp, begin);
5611 sv_setsv_mg(begin, end);
5612 sv_setsv_mg(end, tmp);
5616 SV **begin = AvARRAY(av);
5619 SV **end = begin + AvFILLp(av);
5621 while (begin < end) {
5622 SV * const tmp = *begin;
5633 SV * const tmp = *MARK;
5637 /* safe as long as stack cannot get extended in the above */
5646 SvUTF8_off(TARG); /* decontaminate */
5647 if (SP - MARK > 1) {
5648 do_join(TARG, &PL_sv_no, MARK, SP);
5651 } else if (SP > MARK) {
5652 sv_setsv(TARG, *SP);
5655 sv_setsv(TARG, DEFSV);
5659 up = SvPV_force(TARG, len);
5662 if (DO_UTF8(TARG)) { /* first reverse each character */
5663 U8* s = (U8*)SvPVX(TARG);
5664 const U8* send = (U8*)(s + len);
5666 if (UTF8_IS_INVARIANT(*s)) {
5671 if (!utf8_to_uvchr_buf(s, send, 0))
5675 down = (char*)(s - 1);
5676 /* reverse this character */
5678 const char tmp = *up;
5686 down = SvPVX(TARG) + len - 1;
5688 const char tmp = *up;
5692 (void)SvPOK_only_UTF8(TARG);
5701 AV *ary = ( (PL_op->op_private & OPpSPLIT_ASSIGN) /* @a = split */
5702 && (PL_op->op_flags & OPf_STACKED)) /* @{expr} = split */
5703 ? (AV *)POPs : NULL;
5704 IV limit = POPi; /* note, negative is forever */
5705 SV * const sv = POPs;
5707 const char *s = SvPV_const(sv, len);
5708 const bool do_utf8 = DO_UTF8(sv);
5709 const bool in_uni_8_bit = IN_UNI_8_BIT;
5710 const char *strend = s + len;
5711 PMOP *pm = cPMOPx(PL_op);
5716 const STRLEN slen = do_utf8
5717 ? utf8_length((U8*)s, (U8*)strend)
5718 : (STRLEN)(strend - s);
5719 SSize_t maxiters = slen + 10;
5720 I32 trailing_empty = 0;
5722 const IV origlimit = limit;
5725 const U8 gimme = GIMME_V;
5727 I32 oldsave = PL_savestack_ix;
5728 U32 make_mortal = SVs_TEMP;
5734 TAINT_IF(get_regex_charset(RX_EXTFLAGS(rx)) == REGEX_LOCALE_CHARSET &&
5735 (RX_EXTFLAGS(rx) & (RXf_WHITE | RXf_SKIPWHITE)));
5737 /* handle @ary = split(...) optimisation */
5738 if (PL_op->op_private & OPpSPLIT_ASSIGN) {
5739 if (!(PL_op->op_flags & OPf_STACKED)) {
5740 if (PL_op->op_private & OPpSPLIT_LEX) {
5741 if (PL_op->op_private & OPpLVAL_INTRO)
5742 SAVECLEARSV(PAD_SVl(pm->op_pmreplrootu.op_pmtargetoff));
5743 ary = (AV *)PAD_SVl(pm->op_pmreplrootu.op_pmtargetoff);
5748 MUTABLE_GV(PAD_SVl(pm->op_pmreplrootu.op_pmtargetoff));
5750 pm->op_pmreplrootu.op_pmtargetgv;
5752 if (PL_op->op_private & OPpLVAL_INTRO)
5757 /* skip anything pushed by OPpLVAL_INTRO above */
5758 oldsave = PL_savestack_ix;
5764 (void)sv_2mortal(SvREFCNT_inc_simple_NN(sv));
5767 if ((mg = SvTIED_mg((const SV *)ary, PERL_MAGIC_tied))) {
5769 XPUSHs(SvTIED_obj(MUTABLE_SV(ary), mg));
5776 for (i = AvFILLp(ary); i >= 0; i--)
5777 AvARRAY(ary)[i] = &PL_sv_undef; /* don't free mere refs */
5779 /* temporarily switch stacks */
5780 SAVESWITCHSTACK(PL_curstack, ary);
5785 base = SP - PL_stack_base;
5787 if (RX_EXTFLAGS(rx) & RXf_SKIPWHITE) {
5789 while (s < strend && isSPACE_utf8_safe(s, strend))
5792 else if (get_regex_charset(RX_EXTFLAGS(rx)) == REGEX_LOCALE_CHARSET) {
5793 while (s < strend && isSPACE_LC(*s))
5796 else if (in_uni_8_bit) {
5797 while (s < strend && isSPACE_L1(*s))
5801 while (s < strend && isSPACE(*s))
5805 if (RX_EXTFLAGS(rx) & RXf_PMf_MULTILINE) {
5809 gimme_scalar = gimme == G_SCALAR && !ary;
5812 limit = maxiters + 2;
5813 if (RX_EXTFLAGS(rx) & RXf_WHITE) {
5816 /* this one uses 'm' and is a negative test */
5818 while (m < strend && ! isSPACE_utf8_safe(m, strend) ) {
5819 const int t = UTF8SKIP(m);
5820 /* isSPACE_utf8_safe returns FALSE for malform utf8 */
5827 else if (get_regex_charset(RX_EXTFLAGS(rx)) == REGEX_LOCALE_CHARSET)
5829 while (m < strend && !isSPACE_LC(*m))
5832 else if (in_uni_8_bit) {
5833 while (m < strend && !isSPACE_L1(*m))
5836 while (m < strend && !isSPACE(*m))
5849 dstr = newSVpvn_flags(s, m-s,
5850 (do_utf8 ? SVf_UTF8 : 0) | make_mortal);
5854 /* skip the whitespace found last */
5856 s = m + UTF8SKIP(m);
5860 /* this one uses 's' and is a positive test */
5862 while (s < strend && isSPACE_utf8_safe(s, strend) )
5865 else if (get_regex_charset(RX_EXTFLAGS(rx)) == REGEX_LOCALE_CHARSET)
5867 while (s < strend && isSPACE_LC(*s))
5870 else if (in_uni_8_bit) {
5871 while (s < strend && isSPACE_L1(*s))
5874 while (s < strend && isSPACE(*s))
5879 else if (RX_EXTFLAGS(rx) & RXf_START_ONLY) {
5881 for (m = s; m < strend && *m != '\n'; m++)
5894 dstr = newSVpvn_flags(s, m-s,
5895 (do_utf8 ? SVf_UTF8 : 0) | make_mortal);
5901 else if (RX_EXTFLAGS(rx) & RXf_NULL && !(s >= strend)) {
5903 Pre-extend the stack, either the number of bytes or
5904 characters in the string or a limited amount, triggered by:
5906 my ($x, $y) = split //, $str;
5910 if (!gimme_scalar) {
5911 const IV items = limit - 1;
5912 /* setting it to -1 will trigger a panic in EXTEND() */
5913 const SSize_t sslen = slen > SSize_t_MAX ? -1 : (SSize_t)slen;
5914 if (items >=0 && items < sslen)
5922 /* keep track of how many bytes we skip over */
5932 dstr = newSVpvn_flags(m, s-m, SVf_UTF8 | make_mortal);
5945 dstr = newSVpvn(s, 1);
5961 else if (do_utf8 == (RX_UTF8(rx) != 0) &&
5962 (RX_EXTFLAGS(rx) & RXf_USE_INTUIT) && !RX_NPARENS(rx)
5963 && (RX_EXTFLAGS(rx) & RXf_CHECK_ALL)
5964 && !(RX_EXTFLAGS(rx) & RXf_IS_ANCHORED)) {
5965 const int tail = (RX_EXTFLAGS(rx) & RXf_INTUIT_TAIL);
5966 SV * const csv = CALLREG_INTUIT_STRING(rx);
5968 len = RX_MINLENRET(rx);
5969 if (len == 1 && !RX_UTF8(rx) && !tail) {
5970 const char c = *SvPV_nolen_const(csv);
5972 for (m = s; m < strend && *m != c; m++)
5983 dstr = newSVpvn_flags(s, m-s,
5984 (do_utf8 ? SVf_UTF8 : 0) | make_mortal);
5987 /* The rx->minlen is in characters but we want to step
5988 * s ahead by bytes. */
5990 s = (char*)utf8_hop((U8*)m, len);
5992 s = m + len; /* Fake \n at the end */
5996 while (s < strend && --limit &&
5997 (m = fbm_instr((unsigned char*)s, (unsigned char*)strend,
5998 csv, multiline ? FBMrf_MULTILINE : 0)) )
6007 dstr = newSVpvn_flags(s, m-s,
6008 (do_utf8 ? SVf_UTF8 : 0) | make_mortal);
6011 /* The rx->minlen is in characters but we want to step
6012 * s ahead by bytes. */
6014 s = (char*)utf8_hop((U8*)m, len);
6016 s = m + len; /* Fake \n at the end */
6021 maxiters += slen * RX_NPARENS(rx);
6022 while (s < strend && --limit)
6026 rex_return = CALLREGEXEC(rx, (char*)s, (char*)strend, (char*)orig, 1,
6029 if (rex_return == 0)
6031 TAINT_IF(RX_MATCH_TAINTED(rx));
6032 /* we never pass the REXEC_COPY_STR flag, so it should
6033 * never get copied */
6034 assert(!RX_MATCH_COPIED(rx));
6035 m = RX_OFFS(rx)[0].start + orig;
6044 dstr = newSVpvn_flags(s, m-s,
6045 (do_utf8 ? SVf_UTF8 : 0) | make_mortal);
6048 if (RX_NPARENS(rx)) {
6050 for (i = 1; i <= (I32)RX_NPARENS(rx); i++) {
6051 s = RX_OFFS(rx)[i].start + orig;
6052 m = RX_OFFS(rx)[i].end + orig;
6054 /* japhy (07/27/01) -- the (m && s) test doesn't catch
6055 parens that didn't match -- they should be set to
6056 undef, not the empty string */
6064 if (m >= orig && s >= orig) {
6065 dstr = newSVpvn_flags(s, m-s,
6066 (do_utf8 ? SVf_UTF8 : 0)
6070 dstr = &PL_sv_undef; /* undef, not "" */
6076 s = RX_OFFS(rx)[0].end + orig;
6080 if (!gimme_scalar) {
6081 iters = (SP - PL_stack_base) - base;
6083 if (iters > maxiters)
6084 DIE(aTHX_ "Split loop");
6086 /* keep field after final delim? */
6087 if (s < strend || (iters && origlimit)) {
6088 if (!gimme_scalar) {
6089 const STRLEN l = strend - s;
6090 dstr = newSVpvn_flags(s, l, (do_utf8 ? SVf_UTF8 : 0) | make_mortal);
6095 else if (!origlimit) {
6097 iters -= trailing_empty;
6099 while (iters > 0 && (!TOPs || !SvANY(TOPs) || SvCUR(TOPs) == 0)) {
6100 if (TOPs && !make_mortal)
6109 LEAVE_SCOPE(oldsave); /* may undo an earlier SWITCHSTACK */
6113 if (SvSMAGICAL(ary)) {
6115 mg_set(MUTABLE_SV(ary));
6118 if (gimme == G_ARRAY) {
6120 Copy(AvARRAY(ary), SP + 1, iters, SV*);
6127 ENTER_with_name("call_PUSH");
6128 call_sv(SV_CONST(PUSH),G_SCALAR|G_DISCARD|G_METHOD_NAMED);
6129 LEAVE_with_name("call_PUSH");
6131 if (gimme == G_ARRAY) {
6133 /* EXTEND should not be needed - we just popped them */
6135 for (i=0; i < iters; i++) {
6136 SV **svp = av_fetch(ary, i, FALSE);
6137 PUSHs((svp) ? *svp : &PL_sv_undef);
6144 if (gimme == G_ARRAY)
6156 SV *const sv = PAD_SVl(PL_op->op_targ);
6158 if (SvPADSTALE(sv)) {
6161 RETURNOP(cLOGOP->op_other);
6163 RETURNOP(cLOGOP->op_next);
6172 if (SvTYPE(retsv) == SVt_PVAV || SvTYPE(retsv) == SVt_PVHV
6173 || SvTYPE(retsv) == SVt_PVCV) {
6174 retsv = refto(retsv);
6181 /* used for: pp_padany(), pp_custom(); plus any system ops
6182 * that aren't implemented on a particular platform */
6184 PP(unimplemented_op)
6186 const Optype op_type = PL_op->op_type;
6187 /* Using OP_NAME() isn't going to be helpful here. Firstly, it doesn't cope
6188 with out of range op numbers - it only "special" cases op_custom.
6189 Secondly, as the three ops we "panic" on are padmy, mapstart and custom,
6190 if we get here for a custom op then that means that the custom op didn't
6191 have an implementation. Given that OP_NAME() looks up the custom op
6192 by its pp_addr, likely it will return NULL, unless someone (unhelpfully)
6193 registers &PL_unimplemented_op as the address of their custom op.
6194 NULL doesn't generate a useful error message. "custom" does. */
6195 const char *const name = op_type >= OP_max
6196 ? "[out of range]" : PL_op_name[PL_op->op_type];
6197 if(OP_IS_SOCKET(op_type))
6198 DIE(aTHX_ PL_no_sock_func, name);
6199 DIE(aTHX_ "panic: unimplemented op %s (#%d) called", name, op_type);
6203 S_maybe_unwind_defav(pTHX)
6205 if (CX_CUR()->cx_type & CXp_HASARGS) {
6206 PERL_CONTEXT *cx = CX_CUR();
6208 assert(CxHASARGS(cx));
6210 cx->cx_type &= ~CXp_HASARGS;
6214 /* For sorting out arguments passed to a &CORE:: subroutine */
6218 int opnum = SvIOK(cSVOP_sv) ? (int)SvUV(cSVOP_sv) : 0;
6219 int defgv = PL_opargs[opnum] & OA_DEFGV ||opnum==OP_GLOB, whicharg = 0;
6220 AV * const at_ = GvAV(PL_defgv);
6221 SV **svp = at_ ? AvARRAY(at_) : NULL;
6222 I32 minargs = 0, maxargs = 0, numargs = at_ ? AvFILLp(at_)+1 : 0;
6223 I32 oa = opnum ? PL_opargs[opnum] >> OASHIFT : 0;
6224 bool seen_question = 0;
6225 const char *err = NULL;
6226 const bool pushmark = PL_op->op_private & OPpCOREARGS_PUSHMARK;
6228 /* Count how many args there are first, to get some idea how far to
6229 extend the stack. */
6231 if ((oa & 7) == OA_LIST) { maxargs = I32_MAX; break; }
6233 if (oa & OA_OPTIONAL) seen_question = 1;
6234 if (!seen_question) minargs++;
6238 if(numargs < minargs) err = "Not enough";
6239 else if(numargs > maxargs) err = "Too many";
6241 /* diag_listed_as: Too many arguments for %s */
6243 "%s arguments for %s", err,
6244 opnum ? PL_op_desc[opnum] : SvPV_nolen_const(cSVOP_sv)
6247 /* Reset the stack pointer. Without this, we end up returning our own
6248 arguments in list context, in addition to the values we are supposed
6249 to return. nextstate usually does this on sub entry, but we need
6250 to run the next op with the caller's hints, so we cannot have a
6252 SP = PL_stack_base + CX_CUR()->blk_oldsp;
6254 if(!maxargs) RETURN;
6256 /* We do this here, rather than with a separate pushmark op, as it has
6257 to come in between two things this function does (stack reset and
6258 arg pushing). This seems the easiest way to do it. */
6261 (void)Perl_pp_pushmark(aTHX);
6264 EXTEND(SP, maxargs == I32_MAX ? numargs : maxargs);
6265 PUTBACK; /* The code below can die in various places. */
6267 oa = PL_opargs[opnum] >> OASHIFT;
6268 for (; oa&&(numargs||!pushmark); (void)(numargs&&(++svp,--numargs))) {
6273 if (!numargs && defgv && whicharg == minargs + 1) {
6276 else PUSHs(numargs ? svp && *svp ? *svp : &PL_sv_undef : NULL);
6280 PUSHs(svp && *svp ? *svp : &PL_sv_undef);
6287 if (CvUNIQUE(find_runcv_where(FIND_RUNCV_level_eq,1,NULL)))
6290 S_maybe_unwind_defav(aTHX);
6293 PUSHs((SV *)GvAVn(gv));
6296 if (!svp || !*svp || !SvROK(*svp)
6297 || SvTYPE(SvRV(*svp)) != SVt_PVAV)
6299 /* diag_listed_as: Type of arg %d to &CORE::%s must be %s*/
6300 "Type of arg %d to &CORE::%s must be array reference",
6301 whicharg, PL_op_desc[opnum]
6306 if (!svp || !*svp || !SvROK(*svp)
6307 || ( SvTYPE(SvRV(*svp)) != SVt_PVHV
6308 && ( opnum == OP_DBMCLOSE || opnum == OP_DBMOPEN
6309 || SvTYPE(SvRV(*svp)) != SVt_PVAV )))
6311 /* diag_listed_as: Type of arg %d to &CORE::%s must be %s*/
6312 "Type of arg %d to &CORE::%s must be hash%s reference",
6313 whicharg, PL_op_desc[opnum],
6314 opnum == OP_DBMCLOSE || opnum == OP_DBMOPEN
6321 if (!numargs) PUSHs(NULL);
6322 else if(svp && *svp && SvROK(*svp) && isGV_with_GP(SvRV(*svp)))
6323 /* no magic here, as the prototype will have added an extra
6324 refgen and we just want what was there before that */
6327 const bool constr = PL_op->op_private & whicharg;
6329 svp && *svp ? *svp : &PL_sv_undef,
6330 constr, cBOOL(CopHINTS_get(PL_curcop) & HINT_STRICT_REFS),
6336 if (!numargs) goto try_defsv;
6338 const bool wantscalar =
6339 PL_op->op_private & OPpCOREARGS_SCALARMOD;
6340 if (!svp || !*svp || !SvROK(*svp)
6341 /* We have to permit globrefs even for the \$ proto, as
6342 *foo is indistinguishable from ${\*foo}, and the proto-
6343 type permits the latter. */
6344 || SvTYPE(SvRV(*svp)) > (
6345 wantscalar ? SVt_PVLV
6346 : opnum == OP_LOCK || opnum == OP_UNDEF
6352 "Type of arg %d to &CORE::%s must be %s",
6353 whicharg, PL_op_name[opnum],
6355 ? "scalar reference"
6356 : opnum == OP_LOCK || opnum == OP_UNDEF
6357 ? "reference to one of [$@%&*]"
6358 : "reference to one of [$@%*]"
6361 if (opnum == OP_UNDEF && SvRV(*svp) == (SV *)PL_defgv) {
6362 /* Undo @_ localisation, so that sub exit does not undo
6363 part of our undeffing. */
6364 S_maybe_unwind_defav(aTHX);
6369 DIE(aTHX_ "panic: unknown OA_*: %x", (unsigned)(oa&7));
6377 /* Implement CORE::keys(),values(),each().
6379 * We won't know until run-time whether the arg is an array or hash,
6382 * pp_keys/pp_values/pp_each
6384 * pp_akeys/pp_avalues/pp_aeach
6386 * as appropriate (or whatever pp function actually implements the OP_FOO
6387 * functionality for each FOO).
6394 (SvTYPE(TOPs) == SVt_PVAV ? OP_AEACH : OP_EACH)
6395 + (PL_op->op_private & OPpAVHVSWITCH_MASK)
6403 if (PL_op->op_private & OPpOFFBYONE) {
6404 cv = find_runcv_where(FIND_RUNCV_level_eq, 1, NULL);
6406 else cv = find_runcv(NULL);
6407 XPUSHs(CvEVAL(cv) ? &PL_sv_undef : sv_2mortal(newRV((SV *)cv)));
6412 S_localise_aelem_lval(pTHX_ AV * const av, SV * const keysv,
6413 const bool can_preserve)
6415 const SSize_t ix = SvIV(keysv);
6416 if (can_preserve ? av_exists(av, ix) : TRUE) {
6417 SV ** const svp = av_fetch(av, ix, 1);
6419 Perl_croak(aTHX_ PL_no_aelem, ix);
6420 save_aelem(av, ix, svp);
6423 SAVEADELETE(av, ix);
6427 S_localise_helem_lval(pTHX_ HV * const hv, SV * const keysv,
6428 const bool can_preserve)
6430 if (can_preserve ? hv_exists_ent(hv, keysv, 0) : TRUE) {
6431 HE * const he = hv_fetch_ent(hv, keysv, 1, 0);
6432 SV ** const svp = he ? &HeVAL(he) : NULL;
6434 Perl_croak(aTHX_ PL_no_helem_sv, SVfARG(keysv));
6435 save_helem_flags(hv, keysv, svp, 0);
6438 SAVEHDELETE(hv, keysv);
6442 S_localise_gv_slot(pTHX_ GV *gv, U8 type)
6444 if (type == OPpLVREF_SV) {
6445 save_pushptrptr(gv, SvREFCNT_inc_simple(GvSV(gv)), SAVEt_GVSV);
6448 else if (type == OPpLVREF_AV)
6449 /* XXX Inefficient, as it creates a new AV, which we are
6450 about to clobber. */
6453 assert(type == OPpLVREF_HV);
6454 /* XXX Likewise inefficient. */
6463 SV * const key = PL_op->op_private & OPpLVREF_ELEM ? POPs : NULL;
6464 SV * const left = PL_op->op_flags & OPf_STACKED ? POPs : NULL;
6466 const char *bad = NULL;
6467 const U8 type = PL_op->op_private & OPpLVREF_TYPE;
6468 if (!SvROK(sv)) DIE(aTHX_ "Assigned value is not a reference");
6471 if (SvTYPE(SvRV(sv)) > SVt_PVLV)
6475 if (SvTYPE(SvRV(sv)) != SVt_PVAV)
6479 if (SvTYPE(SvRV(sv)) != SVt_PVHV)
6483 if (SvTYPE(SvRV(sv)) != SVt_PVCV)
6487 /* diag_listed_as: Assigned value is not %s reference */
6488 DIE(aTHX_ "Assigned value is not a%s reference", bad);
6492 switch (left ? SvTYPE(left) : 0) {
6495 SV * const old = PAD_SV(ARGTARG);
6496 PAD_SETSV(ARGTARG, SvREFCNT_inc_NN(SvRV(sv)));
6498 if ((PL_op->op_private & (OPpLVAL_INTRO|OPpPAD_STATE))
6500 SAVECLEARSV(PAD_SVl(ARGTARG));
6504 if (PL_op->op_private & OPpLVAL_INTRO) {
6505 S_localise_gv_slot(aTHX_ (GV *)left, type);
6507 gv_setref(left, sv);
6512 if (UNLIKELY(PL_op->op_private & OPpLVAL_INTRO)) {
6513 S_localise_aelem_lval(aTHX_ (AV *)left, key,
6514 SvCANEXISTDELETE(left));
6516 av_store((AV *)left, SvIV(key), SvREFCNT_inc_simple_NN(SvRV(sv)));
6519 if (UNLIKELY(PL_op->op_private & OPpLVAL_INTRO)) {
6521 S_localise_helem_lval(aTHX_ (HV *)left, key,
6522 SvCANEXISTDELETE(left));
6524 (void)hv_store_ent((HV *)left, key, SvREFCNT_inc_simple_NN(SvRV(sv)), 0);
6526 if (PL_op->op_flags & OPf_MOD)
6527 SETs(sv_2mortal(newSVsv(sv)));
6528 /* XXX else can weak references go stale before they are read, e.g.,
6537 SV * const ret = sv_2mortal(newSV_type(SVt_PVMG));
6538 SV * const elem = PL_op->op_private & OPpLVREF_ELEM ? POPs : NULL;
6539 SV * const arg = PL_op->op_flags & OPf_STACKED ? POPs : NULL;
6540 MAGIC * const mg = sv_magicext(ret, arg, PERL_MAGIC_lvref,
6541 &PL_vtbl_lvref, (char *)elem,
6542 elem ? HEf_SVKEY : (I32)ARGTARG);
6543 mg->mg_private = PL_op->op_private;
6544 if (PL_op->op_private & OPpLVREF_ITER)
6545 mg->mg_flags |= MGf_PERSIST;
6546 if (UNLIKELY(PL_op->op_private & OPpLVAL_INTRO)) {
6552 const bool can_preserve = SvCANEXISTDELETE(arg);
6553 if (SvTYPE(arg) == SVt_PVAV)
6554 S_localise_aelem_lval(aTHX_ (AV *)arg, elem, can_preserve);
6556 S_localise_helem_lval(aTHX_ (HV *)arg, elem, can_preserve);
6560 S_localise_gv_slot(aTHX_ (GV *)arg,
6561 PL_op->op_private & OPpLVREF_TYPE);
6563 else if (!(PL_op->op_private & OPpPAD_STATE))
6564 SAVECLEARSV(PAD_SVl(ARGTARG));
6573 AV * const av = (AV *)POPs;
6574 const bool localizing = PL_op->op_private & OPpLVAL_INTRO;
6575 bool can_preserve = FALSE;
6577 if (UNLIKELY(localizing)) {
6582 can_preserve = SvCANEXISTDELETE(av);
6584 if (SvTYPE(av) == SVt_PVAV) {
6587 for (svp = MARK + 1; svp <= SP; svp++) {
6588 const SSize_t elem = SvIV(*svp);
6592 if (max > AvMAX(av))
6597 while (++MARK <= SP) {
6598 SV * const elemsv = *MARK;
6599 if (SvTYPE(av) == SVt_PVAV)
6600 S_localise_aelem_lval(aTHX_ av, elemsv, can_preserve);
6602 S_localise_helem_lval(aTHX_ (HV *)av, elemsv, can_preserve);
6603 *MARK = sv_2mortal(newSV_type(SVt_PVMG));
6604 sv_magic(*MARK,(SV *)av,PERL_MAGIC_lvref,(char *)elemsv,HEf_SVKEY);
6611 if (PL_op->op_flags & OPf_STACKED)
6612 Perl_pp_rv2av(aTHX);
6614 Perl_pp_padav(aTHX);
6618 SETs(0); /* special alias marker that aassign recognises */
6628 SETs(sv_2mortal((SV *)newCONSTSUB(SvTYPE(CopSTASH(PL_curcop))==SVt_PVHV
6629 ? CopSTASH(PL_curcop)
6631 NULL, SvREFCNT_inc_simple_NN(sv))));
6636 /* process one subroutine argument - typically when the sub has a signature:
6637 * introduce PL_curpad[op_targ] and assign to it the value
6638 * for $: (OPf_STACKED ? *sp : $_[N])
6639 * for @/%: @_[N..$#_]
6641 * It's equivalent to
6644 * my $foo = (value-on-stack)
6646 * my @foo = @_[N..$#_]
6656 AV *defav = GvAV(PL_defgv); /* @_ */
6657 IV ix = PTR2IV(cUNOP_AUXo->op_aux);
6660 /* do 'my $var, @var or %var' action */
6661 padentry = &(PAD_SVl(o->op_targ));
6662 save_clearsv(padentry);
6665 if ((o->op_private & OPpARGELEM_MASK) == OPpARGELEM_SV) {
6666 if (o->op_flags & OPf_STACKED) {
6673 /* should already have been checked */
6675 #if IVSIZE > PTRSIZE
6676 assert(ix <= SSize_t_MAX);
6679 svp = av_fetch(defav, ix, FALSE);
6680 val = svp ? *svp : &PL_sv_undef;
6685 /* cargo-culted from pp_sassign */
6686 assert(TAINTING_get || !TAINT_get);
6687 if (UNLIKELY(TAINT_get) && !SvTAINTED(val))
6690 SvSetMagicSV(targ, val);
6694 /* must be AV or HV */
6696 assert(!(o->op_flags & OPf_STACKED));
6697 argc = ((IV)AvFILL(defav) + 1) - ix;
6699 /* This is a copy of the relevant parts of pp_aassign().
6701 if ((o->op_private & OPpARGELEM_MASK) == OPpARGELEM_AV) {
6704 if (AvFILL((AV*)targ) > -1) {
6705 /* target should usually be empty. If we get get
6706 * here, someone's been doing some weird closure tricks.
6707 * Make a copy of all args before clearing the array,
6708 * to avoid the equivalent of @a = ($a[0]) prematurely freeing
6709 * elements. See similar code in pp_aassign.
6711 for (i = 0; i < argc; i++) {
6712 SV **svp = av_fetch(defav, ix + i, FALSE);
6713 SV *newsv = newSV(0);
6714 sv_setsv_flags(newsv,
6715 svp ? *svp : &PL_sv_undef,
6716 (SV_DO_COW_SVSETSV|SV_NOSTEAL));
6717 if (!av_store(defav, ix + i, newsv))
6718 SvREFCNT_dec_NN(newsv);
6720 av_clear((AV*)targ);
6726 av_extend((AV*)targ, argc);
6731 SV **svp = av_fetch(defav, ix + i, FALSE);
6732 SV *val = svp ? *svp : &PL_sv_undef;
6734 sv_setsv(tmpsv, val);
6735 av_store((AV*)targ, i++, tmpsv);
6743 assert((o->op_private & OPpARGELEM_MASK) == OPpARGELEM_HV);
6745 if (SvRMAGICAL(targ) || HvUSEDKEYS((HV*)targ)) {
6746 /* see "target should usually be empty" comment above */
6747 for (i = 0; i < argc; i++) {
6748 SV **svp = av_fetch(defav, ix + i, FALSE);
6749 SV *newsv = newSV(0);
6750 sv_setsv_flags(newsv,
6751 svp ? *svp : &PL_sv_undef,
6752 (SV_DO_COW_SVSETSV|SV_NOSTEAL));
6753 if (!av_store(defav, ix + i, newsv))
6754 SvREFCNT_dec_NN(newsv);
6756 hv_clear((HV*)targ);
6761 assert(argc % 2 == 0);
6770 svp = av_fetch(defav, ix + i++, FALSE);
6771 key = svp ? *svp : &PL_sv_undef;
6772 svp = av_fetch(defav, ix + i++, FALSE);
6773 val = svp ? *svp : &PL_sv_undef;
6776 if (UNLIKELY(SvGMAGICAL(key)))
6777 key = sv_mortalcopy(key);
6779 sv_setsv(tmpsv, val);
6780 hv_store_ent((HV*)targ, key, tmpsv, 0);
6788 /* Handle a default value for one subroutine argument (typically as part
6789 * of a subroutine signature).
6790 * It's equivalent to
6791 * @_ > op_targ ? $_[op_targ] : result_of(op_other)
6793 * Intended to be used where op_next is an OP_ARGELEM
6795 * We abuse the op_targ field slightly: it's an index into @_ rather than
6801 OP * const o = PL_op;
6802 AV *defav = GvAV(PL_defgv); /* @_ */
6803 IV ix = (IV)o->op_targ;
6806 #if IVSIZE > PTRSIZE
6807 assert(ix <= SSize_t_MAX);
6810 if (AvFILL(defav) >= ix) {
6812 SV **svp = av_fetch(defav, ix, FALSE);
6813 SV *val = svp ? *svp : &PL_sv_undef;
6817 return cLOGOPo->op_other;
6822 S_find_runcv_name(void)
6837 sv = sv_2mortal(newSV(0));
6838 gv_fullname4(sv, gv, NULL, TRUE);
6842 /* Check a a subs arguments - i.e. that it has the correct number of args
6843 * (and anything else we might think of in future). Typically used with
6849 OP * const o = PL_op;
6850 UNOP_AUX_item *aux = cUNOP_AUXo->op_aux;
6851 IV params = aux[0].iv;
6852 IV opt_params = aux[1].iv;
6853 char slurpy = (char)(aux[2].iv);
6854 AV *defav = GvAV(PL_defgv); /* @_ */
6858 assert(!SvMAGICAL(defav));
6859 argc = (AvFILLp(defav) + 1);
6860 too_few = (argc < (params - opt_params));
6862 if (UNLIKELY(too_few || (!slurpy && argc > params)))
6863 /* diag_listed_as: Too few arguments for subroutine '%s' */
6864 /* diag_listed_as: Too many arguments for subroutine '%s' */
6865 Perl_croak_caller("Too %s arguments for subroutine '%" SVf "'",
6866 too_few ? "few" : "many", S_find_runcv_name());
6868 if (UNLIKELY(slurpy == '%' && argc > params && (argc - params) % 2))
6869 /* diag_listed_as: Odd name/value argument for subroutine '%s' */
6870 Perl_croak_caller("Odd name/value argument for subroutine '%" SVf "'",
6871 S_find_runcv_name());
6877 * ex: set ts=8 sts=4 sw=4 et: