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 /* work around glibc-2.2.5 bug */
3657 PL_reentrant_buffer->_crypt_struct_buffer->current_saltbits = 0;
3661 # endif /* HAS_CRYPT_R */
3662 # endif /* USE_ITHREADS */
3664 sv_setpv(TARG, fcrypt(tmps, SvPV_nolen_const(right)));
3666 sv_setpv(TARG, PerlProc_crypt(tmps, SvPV_nolen_const(right)));
3673 "The crypt() function is unimplemented due to excessive paranoia.");
3677 /* Generally UTF-8 and UTF-EBCDIC are indistinguishable at this level. So
3678 * most comments below say UTF-8, when in fact they mean UTF-EBCDIC as well */
3681 /* also used for: pp_lcfirst() */
3685 /* Actually is both lcfirst() and ucfirst(). Only the first character
3686 * changes. This means that possibly we can change in-place, ie., just
3687 * take the source and change that one character and store it back, but not
3688 * if read-only etc, or if the length changes */
3692 STRLEN slen; /* slen is the byte length of the whole SV. */
3695 bool inplace; /* ? Convert first char only, in-place */
3696 bool doing_utf8 = FALSE; /* ? using utf8 */
3697 bool convert_source_to_utf8 = FALSE; /* ? need to convert */
3698 const int op_type = PL_op->op_type;
3701 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
3702 STRLEN ulen; /* ulen is the byte length of the original Unicode character
3703 * stored as UTF-8 at s. */
3704 STRLEN tculen; /* tculen is the byte length of the freshly titlecased (or
3705 * lowercased) character stored in tmpbuf. May be either
3706 * UTF-8 or not, but in either case is the number of bytes */
3708 s = (const U8*)SvPV_const(source, slen);
3710 /* We may be able to get away with changing only the first character, in
3711 * place, but not if read-only, etc. Later we may discover more reasons to
3712 * not convert in-place. */
3713 inplace = !SvREADONLY(source) && SvPADTMP(source);
3715 #ifdef USE_LOCALE_CTYPE
3717 if (IN_LC_RUNTIME(LC_CTYPE)) {
3718 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
3723 /* First calculate what the changed first character should be. This affects
3724 * whether we can just swap it out, leaving the rest of the string unchanged,
3725 * or even if have to convert the dest to UTF-8 when the source isn't */
3727 if (! slen) { /* If empty */
3728 need = 1; /* still need a trailing NUL */
3731 else if (DO_UTF8(source)) { /* Is the source utf8? */
3734 if (op_type == OP_UCFIRST) {
3735 #ifdef USE_LOCALE_CTYPE
3736 _toTITLE_utf8_flags(s, s +slen, tmpbuf, &tculen, IN_LC_RUNTIME(LC_CTYPE));
3738 _toTITLE_utf8_flags(s, s +slen, tmpbuf, &tculen, 0);
3742 #ifdef USE_LOCALE_CTYPE
3743 _toLOWER_utf8_flags(s, s + slen, tmpbuf, &tculen, IN_LC_RUNTIME(LC_CTYPE));
3745 _toLOWER_utf8_flags(s, s + slen, tmpbuf, &tculen, 0);
3749 /* we can't do in-place if the length changes. */
3750 if (ulen != tculen) inplace = FALSE;
3751 need = slen + 1 - ulen + tculen;
3753 else { /* Non-zero length, non-UTF-8, Need to consider locale and if
3754 * latin1 is treated as caseless. Note that a locale takes
3756 ulen = 1; /* Original character is 1 byte */
3757 tculen = 1; /* Most characters will require one byte, but this will
3758 * need to be overridden for the tricky ones */
3761 if (op_type == OP_LCFIRST) {
3763 /* lower case the first letter: no trickiness for any character */
3764 #ifdef USE_LOCALE_CTYPE
3765 if (IN_LC_RUNTIME(LC_CTYPE)) {
3766 *tmpbuf = toLOWER_LC(*s);
3771 *tmpbuf = (IN_UNI_8_BIT)
3772 ? toLOWER_LATIN1(*s)
3776 #ifdef USE_LOCALE_CTYPE
3778 else if (IN_LC_RUNTIME(LC_CTYPE)) {
3779 if (IN_UTF8_CTYPE_LOCALE) {
3783 *tmpbuf = (U8) toUPPER_LC(*s); /* This would be a bug if any
3784 locales have upper and title case
3788 else if (! IN_UNI_8_BIT) {
3789 *tmpbuf = toUPPER(*s); /* Returns caseless for non-ascii, or
3790 * on EBCDIC machines whatever the
3791 * native function does */
3794 /* Here, is ucfirst non-UTF-8, not in locale (unless that locale is
3795 * UTF-8, which we treat as not in locale), and cased latin1 */
3797 #ifdef USE_LOCALE_CTYPE
3801 title_ord = _to_upper_title_latin1(*s, tmpbuf, &tculen, 's');
3803 assert(tculen == 2);
3805 /* If the result is an upper Latin1-range character, it can
3806 * still be represented in one byte, which is its ordinal */
3807 if (UTF8_IS_DOWNGRADEABLE_START(*tmpbuf)) {
3808 *tmpbuf = (U8) title_ord;
3812 /* Otherwise it became more than one ASCII character (in
3813 * the case of LATIN_SMALL_LETTER_SHARP_S) or changed to
3814 * beyond Latin1, so the number of bytes changed, so can't
3815 * replace just the first character in place. */
3818 /* If the result won't fit in a byte, the entire result
3819 * will have to be in UTF-8. Assume worst case sizing in
3820 * conversion. (all latin1 characters occupy at most two
3822 if (title_ord > 255) {
3824 convert_source_to_utf8 = TRUE;
3825 need = slen * 2 + 1;
3827 /* The (converted) UTF-8 and UTF-EBCDIC lengths of all
3828 * (both) characters whose title case is above 255 is
3832 else { /* LATIN_SMALL_LETTER_SHARP_S expands by 1 byte */
3833 need = slen + 1 + 1;
3837 } /* End of use Unicode (Latin1) semantics */
3838 } /* End of changing the case of the first character */
3840 /* Here, have the first character's changed case stored in tmpbuf. Ready to
3841 * generate the result */
3844 /* We can convert in place. This means we change just the first
3845 * character without disturbing the rest; no need to grow */
3847 s = d = (U8*)SvPV_force_nomg(source, slen);
3853 /* Here, we can't convert in place; we earlier calculated how much
3854 * space we will need, so grow to accommodate that */
3855 SvUPGRADE(dest, SVt_PV);
3856 d = (U8*)SvGROW(dest, need);
3857 (void)SvPOK_only(dest);
3864 if (! convert_source_to_utf8) {
3866 /* Here both source and dest are in UTF-8, but have to create
3867 * the entire output. We initialize the result to be the
3868 * title/lower cased first character, and then append the rest
3870 sv_setpvn(dest, (char*)tmpbuf, tculen);
3872 sv_catpvn(dest, (char*)(s + ulen), slen - ulen);
3876 const U8 *const send = s + slen;
3878 /* Here the dest needs to be in UTF-8, but the source isn't,
3879 * except we earlier UTF-8'd the first character of the source
3880 * into tmpbuf. First put that into dest, and then append the
3881 * rest of the source, converting it to UTF-8 as we go. */
3883 /* Assert tculen is 2 here because the only two characters that
3884 * get to this part of the code have 2-byte UTF-8 equivalents */
3886 *d++ = *(tmpbuf + 1);
3887 s++; /* We have just processed the 1st char */
3889 for (; s < send; s++) {
3890 d = uvchr_to_utf8(d, *s);
3893 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
3897 else { /* in-place UTF-8. Just overwrite the first character */
3898 Copy(tmpbuf, d, tculen, U8);
3899 SvCUR_set(dest, need - 1);
3903 else { /* Neither source nor dest are in or need to be UTF-8 */
3905 if (inplace) { /* in-place, only need to change the 1st char */
3908 else { /* Not in-place */
3910 /* Copy the case-changed character(s) from tmpbuf */
3911 Copy(tmpbuf, d, tculen, U8);
3912 d += tculen - 1; /* Code below expects d to point to final
3913 * character stored */
3916 else { /* empty source */
3917 /* See bug #39028: Don't taint if empty */
3921 /* In a "use bytes" we don't treat the source as UTF-8, but, still want
3922 * the destination to retain that flag */
3923 if (SvUTF8(source) && ! IN_BYTES)
3926 if (!inplace) { /* Finish the rest of the string, unchanged */
3927 /* This will copy the trailing NUL */
3928 Copy(s + 1, d + 1, slen, U8);
3929 SvCUR_set(dest, need - 1);
3932 #ifdef USE_LOCALE_CTYPE
3933 if (IN_LC_RUNTIME(LC_CTYPE)) {
3938 if (dest != source && SvTAINTED(source))
3944 /* There's so much setup/teardown code common between uc and lc, I wonder if
3945 it would be worth merging the two, and just having a switch outside each
3946 of the three tight loops. There is less and less commonality though */
3959 if ( SvPADTMP(source)
3960 && !SvREADONLY(source) && SvPOK(source)
3963 #ifdef USE_LOCALE_CTYPE
3964 (IN_LC_RUNTIME(LC_CTYPE))
3965 ? ! IN_UTF8_CTYPE_LOCALE
3971 /* We can convert in place. The reason we can't if in UNI_8_BIT is to
3972 * make the loop tight, so we overwrite the source with the dest before
3973 * looking at it, and we need to look at the original source
3974 * afterwards. There would also need to be code added to handle
3975 * switching to not in-place in midstream if we run into characters
3976 * that change the length. Since being in locale overrides UNI_8_BIT,
3977 * that latter becomes irrelevant in the above test; instead for
3978 * locale, the size can't normally change, except if the locale is a
3981 s = d = (U8*)SvPV_force_nomg(source, len);
3988 s = (const U8*)SvPV_nomg_const(source, len);
3991 SvUPGRADE(dest, SVt_PV);
3992 d = (U8*)SvGROW(dest, min);
3993 (void)SvPOK_only(dest);
3998 #ifdef USE_LOCALE_CTYPE
4000 if (IN_LC_RUNTIME(LC_CTYPE)) {
4001 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
4006 /* Overloaded values may have toggled the UTF-8 flag on source, so we need
4007 to check DO_UTF8 again here. */
4009 if (DO_UTF8(source)) {
4010 const U8 *const send = s + len;
4011 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
4013 /* All occurrences of these are to be moved to follow any other marks.
4014 * This is context-dependent. We may not be passed enough context to
4015 * move the iota subscript beyond all of them, but we do the best we can
4016 * with what we're given. The result is always better than if we
4017 * hadn't done this. And, the problem would only arise if we are
4018 * passed a character without all its combining marks, which would be
4019 * the caller's mistake. The information this is based on comes from a
4020 * comment in Unicode SpecialCasing.txt, (and the Standard's text
4021 * itself) and so can't be checked properly to see if it ever gets
4022 * revised. But the likelihood of it changing is remote */
4023 bool in_iota_subscript = FALSE;
4029 if (in_iota_subscript && ! _is_utf8_mark(s)) {
4031 /* A non-mark. Time to output the iota subscript */
4032 Copy(GREEK_CAPITAL_LETTER_IOTA_UTF8, d, capital_iota_len, U8);
4033 d += capital_iota_len;
4034 in_iota_subscript = FALSE;
4037 /* Then handle the current character. Get the changed case value
4038 * and copy it to the output buffer */
4041 #ifdef USE_LOCALE_CTYPE
4042 uv = _toUPPER_utf8_flags(s, send, tmpbuf, &ulen, IN_LC_RUNTIME(LC_CTYPE));
4044 uv = _toUPPER_utf8_flags(s, send, tmpbuf, &ulen, 0);
4046 #define GREEK_CAPITAL_LETTER_IOTA 0x0399
4047 #define COMBINING_GREEK_YPOGEGRAMMENI 0x0345
4048 if (uv == GREEK_CAPITAL_LETTER_IOTA
4049 && utf8_to_uvchr_buf(s, send, 0) == COMBINING_GREEK_YPOGEGRAMMENI)
4051 in_iota_subscript = TRUE;
4054 if (ulen > u && (SvLEN(dest) < (min += ulen - u))) {
4055 /* If the eventually required minimum size outgrows the
4056 * available space, we need to grow. */
4057 const UV o = d - (U8*)SvPVX_const(dest);
4059 /* If someone uppercases one million U+03B0s we SvGROW()
4060 * one million times. Or we could try guessing how much to
4061 * allocate without allocating too much. Such is life.
4062 * See corresponding comment in lc code for another option
4064 d = o + (U8*) SvGROW(dest, min);
4066 Copy(tmpbuf, d, ulen, U8);
4071 if (in_iota_subscript) {
4072 Copy(GREEK_CAPITAL_LETTER_IOTA_UTF8, d, capital_iota_len, U8);
4073 d += capital_iota_len;
4078 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
4080 else { /* Not UTF-8 */
4082 const U8 *const send = s + len;
4084 /* Use locale casing if in locale; regular style if not treating
4085 * latin1 as having case; otherwise the latin1 casing. Do the
4086 * whole thing in a tight loop, for speed, */
4087 #ifdef USE_LOCALE_CTYPE
4088 if (IN_LC_RUNTIME(LC_CTYPE)) {
4089 if (IN_UTF8_CTYPE_LOCALE) {
4092 for (; s < send; d++, s++)
4093 *d = (U8) toUPPER_LC(*s);
4097 if (! IN_UNI_8_BIT) {
4098 for (; s < send; d++, s++) {
4103 #ifdef USE_LOCALE_CTYPE
4106 for (; s < send; d++, s++) {
4107 *d = toUPPER_LATIN1_MOD(*s);
4108 if (LIKELY(*d != LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
4112 /* The mainstream case is the tight loop above. To avoid
4113 * extra tests in that, all three characters that require
4114 * special handling are mapped by the MOD to the one tested
4116 * Use the source to distinguish between the three cases */
4118 #if UNICODE_MAJOR_VERSION > 2 \
4119 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
4120 && UNICODE_DOT_DOT_VERSION >= 8)
4121 if (*s == LATIN_SMALL_LETTER_SHARP_S) {
4123 /* uc() of this requires 2 characters, but they are
4124 * ASCII. If not enough room, grow the string */
4125 if (SvLEN(dest) < ++min) {
4126 const UV o = d - (U8*)SvPVX_const(dest);
4127 d = o + (U8*) SvGROW(dest, min);
4129 *d++ = 'S'; *d = 'S'; /* upper case is 'SS' */
4130 continue; /* Back to the tight loop; still in ASCII */
4134 /* The other two special handling characters have their
4135 * upper cases outside the latin1 range, hence need to be
4136 * in UTF-8, so the whole result needs to be in UTF-8. So,
4137 * here we are somewhere in the middle of processing a
4138 * non-UTF-8 string, and realize that we will have to convert
4139 * the whole thing to UTF-8. What to do? There are
4140 * several possibilities. The simplest to code is to
4141 * convert what we have so far, set a flag, and continue on
4142 * in the loop. The flag would be tested each time through
4143 * the loop, and if set, the next character would be
4144 * converted to UTF-8 and stored. But, I (khw) didn't want
4145 * to slow down the mainstream case at all for this fairly
4146 * rare case, so I didn't want to add a test that didn't
4147 * absolutely have to be there in the loop, besides the
4148 * possibility that it would get too complicated for
4149 * optimizers to deal with. Another possibility is to just
4150 * give up, convert the source to UTF-8, and restart the
4151 * function that way. Another possibility is to convert
4152 * both what has already been processed and what is yet to
4153 * come separately to UTF-8, then jump into the loop that
4154 * handles UTF-8. But the most efficient time-wise of the
4155 * ones I could think of is what follows, and turned out to
4156 * not require much extra code. */
4158 /* Convert what we have so far into UTF-8, telling the
4159 * function that we know it should be converted, and to
4160 * allow extra space for what we haven't processed yet.
4161 * Assume the worst case space requirements for converting
4162 * what we haven't processed so far: that it will require
4163 * two bytes for each remaining source character, plus the
4164 * NUL at the end. This may cause the string pointer to
4165 * move, so re-find it. */
4167 len = d - (U8*)SvPVX_const(dest);
4168 SvCUR_set(dest, len);
4169 len = sv_utf8_upgrade_flags_grow(dest,
4170 SV_GMAGIC|SV_FORCE_UTF8_UPGRADE,
4172 d = (U8*)SvPVX(dest) + len;
4174 /* Now process the remainder of the source, converting to
4175 * upper and UTF-8. If a resulting byte is invariant in
4176 * UTF-8, output it as-is, otherwise convert to UTF-8 and
4177 * append it to the output. */
4178 for (; s < send; s++) {
4179 (void) _to_upper_title_latin1(*s, d, &len, 'S');
4183 /* Here have processed the whole source; no need to continue
4184 * with the outer loop. Each character has been converted
4185 * to upper case and converted to UTF-8 */
4188 } /* End of processing all latin1-style chars */
4189 } /* End of processing all chars */
4190 } /* End of source is not empty */
4192 if (source != dest) {
4193 *d = '\0'; /* Here d points to 1 after last char, add NUL */
4194 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
4196 } /* End of isn't utf8 */
4197 #ifdef USE_LOCALE_CTYPE
4198 if (IN_LC_RUNTIME(LC_CTYPE)) {
4203 if (dest != source && SvTAINTED(source))
4221 if ( SvPADTMP(source)
4222 && !SvREADONLY(source) && SvPOK(source)
4223 && !DO_UTF8(source)) {
4225 /* We can convert in place, as lowercasing anything in the latin1 range
4226 * (or else DO_UTF8 would have been on) doesn't lengthen it */
4228 s = d = (U8*)SvPV_force_nomg(source, len);
4235 s = (const U8*)SvPV_nomg_const(source, len);
4238 SvUPGRADE(dest, SVt_PV);
4239 d = (U8*)SvGROW(dest, min);
4240 (void)SvPOK_only(dest);
4245 #ifdef USE_LOCALE_CTYPE
4247 if (IN_LC_RUNTIME(LC_CTYPE)) {
4248 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
4253 /* Overloaded values may have toggled the UTF-8 flag on source, so we need
4254 to check DO_UTF8 again here. */
4256 if (DO_UTF8(source)) {
4257 const U8 *const send = s + len;
4258 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
4261 const STRLEN u = UTF8SKIP(s);
4264 #ifdef USE_LOCALE_CTYPE
4265 _toLOWER_utf8_flags(s, send, tmpbuf, &ulen, IN_LC_RUNTIME(LC_CTYPE));
4267 _toLOWER_utf8_flags(s, send, tmpbuf, &ulen, 0);
4270 /* Here is where we would do context-sensitive actions. See the
4271 * commit message for 86510fb15 for why there isn't any */
4273 if (ulen > u && (SvLEN(dest) < (min += ulen - u))) {
4275 /* If the eventually required minimum size outgrows the
4276 * available space, we need to grow. */
4277 const UV o = d - (U8*)SvPVX_const(dest);
4279 /* If someone lowercases one million U+0130s we SvGROW() one
4280 * million times. Or we could try guessing how much to
4281 * allocate without allocating too much. Such is life.
4282 * Another option would be to grow an extra byte or two more
4283 * each time we need to grow, which would cut down the million
4284 * to 500K, with little waste */
4285 d = o + (U8*) SvGROW(dest, min);
4288 /* Copy the newly lowercased letter to the output buffer we're
4290 Copy(tmpbuf, d, ulen, U8);
4293 } /* End of looping through the source string */
4296 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
4297 } else { /* Not utf8 */
4299 const U8 *const send = s + len;
4301 /* Use locale casing if in locale; regular style if not treating
4302 * latin1 as having case; otherwise the latin1 casing. Do the
4303 * whole thing in a tight loop, for speed, */
4304 #ifdef USE_LOCALE_CTYPE
4305 if (IN_LC_RUNTIME(LC_CTYPE)) {
4306 for (; s < send; d++, s++)
4307 *d = toLOWER_LC(*s);
4311 if (! IN_UNI_8_BIT) {
4312 for (; s < send; d++, s++) {
4317 for (; s < send; d++, s++) {
4318 *d = toLOWER_LATIN1(*s);
4322 if (source != dest) {
4324 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
4327 #ifdef USE_LOCALE_CTYPE
4328 if (IN_LC_RUNTIME(LC_CTYPE)) {
4333 if (dest != source && SvTAINTED(source))
4342 SV * const sv = TOPs;
4344 const char *s = SvPV_const(sv,len);
4346 SvUTF8_off(TARG); /* decontaminate */
4349 SvUPGRADE(TARG, SVt_PV);
4350 SvGROW(TARG, (len * 2) + 1);
4354 STRLEN ulen = UTF8SKIP(s);
4355 bool to_quote = FALSE;
4357 if (UTF8_IS_INVARIANT(*s)) {
4358 if (_isQUOTEMETA(*s)) {
4362 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, s + len)) {
4364 #ifdef USE_LOCALE_CTYPE
4365 /* In locale, we quote all non-ASCII Latin1 chars.
4366 * Otherwise use the quoting rules */
4368 IN_LC_RUNTIME(LC_CTYPE)
4371 _isQUOTEMETA(EIGHT_BIT_UTF8_TO_NATIVE(*s, *(s + 1))))
4376 else if (is_QUOTEMETA_high(s)) {
4391 else if (IN_UNI_8_BIT) {
4393 if (_isQUOTEMETA(*s))
4399 /* For non UNI_8_BIT (and hence in locale) just quote all \W
4400 * including everything above ASCII */
4402 if (!isWORDCHAR_A(*s))
4408 SvCUR_set(TARG, d - SvPVX_const(TARG));
4409 (void)SvPOK_only_UTF8(TARG);
4412 sv_setpvn(TARG, s, len);
4428 U8 tmpbuf[UTF8_MAXBYTES_CASE + 1];
4429 #if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
4430 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
4431 || UNICODE_DOT_DOT_VERSION > 0)
4432 const bool full_folding = TRUE; /* This variable is here so we can easily
4433 move to more generality later */
4435 const bool full_folding = FALSE;
4437 const U8 flags = ( full_folding ? FOLD_FLAGS_FULL : 0 )
4438 #ifdef USE_LOCALE_CTYPE
4439 | ( IN_LC_RUNTIME(LC_CTYPE) ? FOLD_FLAGS_LOCALE : 0 )
4443 /* This is a facsimile of pp_lc, but with a thousand bugs thanks to me.
4444 * You are welcome(?) -Hugmeir
4452 s = (const U8*)SvPV_nomg_const(source, len);
4454 if (ckWARN(WARN_UNINITIALIZED))
4455 report_uninit(source);
4462 SvUPGRADE(dest, SVt_PV);
4463 d = (U8*)SvGROW(dest, min);
4464 (void)SvPOK_only(dest);
4470 #ifdef USE_LOCALE_CTYPE
4472 if ( IN_LC_RUNTIME(LC_CTYPE) ) { /* Under locale */
4473 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
4478 if (DO_UTF8(source)) { /* UTF-8 flagged string. */
4480 const STRLEN u = UTF8SKIP(s);
4483 _toFOLD_utf8_flags(s, send, tmpbuf, &ulen, flags);
4485 if (ulen > u && (SvLEN(dest) < (min += ulen - u))) {
4486 const UV o = d - (U8*)SvPVX_const(dest);
4487 d = o + (U8*) SvGROW(dest, min);
4490 Copy(tmpbuf, d, ulen, U8);
4495 } /* Unflagged string */
4497 #ifdef USE_LOCALE_CTYPE
4498 if ( IN_LC_RUNTIME(LC_CTYPE) ) { /* Under locale */
4499 if (IN_UTF8_CTYPE_LOCALE) {
4500 goto do_uni_folding;
4502 for (; s < send; d++, s++)
4503 *d = (U8) toFOLD_LC(*s);
4507 if ( !IN_UNI_8_BIT ) { /* Under nothing, or bytes */
4508 for (; s < send; d++, s++)
4512 #ifdef USE_LOCALE_CTYPE
4515 /* For ASCII and the Latin-1 range, there's only two troublesome
4516 * folds, \x{DF} (\N{LATIN SMALL LETTER SHARP S}), which under full
4517 * casefolding becomes 'ss'; and \x{B5} (\N{MICRO SIGN}), which
4518 * under any fold becomes \x{3BC} (\N{GREEK SMALL LETTER MU}) --
4519 * For the rest, the casefold is their lowercase. */
4520 for (; s < send; d++, s++) {
4521 if (*s == MICRO_SIGN) {
4522 /* \N{MICRO SIGN}'s casefold is \N{GREEK SMALL LETTER MU},
4523 * which is outside of the latin-1 range. There's a couple
4524 * of ways to deal with this -- khw discusses them in
4525 * pp_lc/uc, so go there :) What we do here is upgrade what
4526 * we had already casefolded, then enter an inner loop that
4527 * appends the rest of the characters as UTF-8. */
4528 len = d - (U8*)SvPVX_const(dest);
4529 SvCUR_set(dest, len);
4530 len = sv_utf8_upgrade_flags_grow(dest,
4531 SV_GMAGIC|SV_FORCE_UTF8_UPGRADE,
4532 /* The max expansion for latin1
4533 * chars is 1 byte becomes 2 */
4535 d = (U8*)SvPVX(dest) + len;
4537 Copy(GREEK_SMALL_LETTER_MU_UTF8, d, small_mu_len, U8);
4540 for (; s < send; s++) {
4542 UV fc = _to_uni_fold_flags(*s, tmpbuf, &ulen, flags);
4543 if UVCHR_IS_INVARIANT(fc) {
4545 && *s == LATIN_SMALL_LETTER_SHARP_S)
4554 Copy(tmpbuf, d, ulen, U8);
4560 else if (full_folding && *s == LATIN_SMALL_LETTER_SHARP_S) {
4561 /* Under full casefolding, LATIN SMALL LETTER SHARP S
4562 * becomes "ss", which may require growing the SV. */
4563 if (SvLEN(dest) < ++min) {
4564 const UV o = d - (U8*)SvPVX_const(dest);
4565 d = o + (U8*) SvGROW(dest, min);
4570 else { /* If it's not one of those two, the fold is their lower
4572 *d = toLOWER_LATIN1(*s);
4578 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
4580 #ifdef USE_LOCALE_CTYPE
4581 if (IN_LC_RUNTIME(LC_CTYPE)) {
4586 if (SvTAINTED(source))
4596 dSP; dMARK; dORIGMARK;
4597 AV *const av = MUTABLE_AV(POPs);
4598 const I32 lval = (PL_op->op_flags & OPf_MOD || LVRET);
4600 if (SvTYPE(av) == SVt_PVAV) {
4601 const bool localizing = PL_op->op_private & OPpLVAL_INTRO;
4602 bool can_preserve = FALSE;
4608 can_preserve = SvCANEXISTDELETE(av);
4611 if (lval && localizing) {
4614 for (svp = MARK + 1; svp <= SP; svp++) {
4615 const SSize_t elem = SvIV(*svp);
4619 if (max > AvMAX(av))
4623 while (++MARK <= SP) {
4625 SSize_t elem = SvIV(*MARK);
4626 bool preeminent = TRUE;
4628 if (localizing && can_preserve) {
4629 /* If we can determine whether the element exist,
4630 * Try to preserve the existenceness of a tied array
4631 * element by using EXISTS and DELETE if possible.
4632 * Fallback to FETCH and STORE otherwise. */
4633 preeminent = av_exists(av, elem);
4636 svp = av_fetch(av, elem, lval);
4639 DIE(aTHX_ PL_no_aelem, elem);
4642 save_aelem(av, elem, svp);
4644 SAVEADELETE(av, elem);
4647 *MARK = svp ? *svp : &PL_sv_undef;
4650 if (GIMME_V != G_ARRAY) {
4652 *++MARK = SP > ORIGMARK ? *SP : &PL_sv_undef;
4661 AV *const av = MUTABLE_AV(POPs);
4662 I32 lval = (PL_op->op_flags & OPf_MOD);
4663 SSize_t items = SP - MARK;
4665 if (PL_op->op_private & OPpMAYBE_LVSUB) {
4666 const I32 flags = is_lvalue_sub();
4668 if (!(flags & OPpENTERSUB_INARGS))
4669 /* diag_listed_as: Can't modify %s in %s */
4670 Perl_croak(aTHX_ "Can't modify index/value array slice in list assignment");
4677 *(MARK+items*2-1) = *(MARK+items);
4683 while (++MARK <= SP) {
4686 svp = av_fetch(av, SvIV(*MARK), lval);
4688 if (!svp || !*svp || *svp == &PL_sv_undef) {
4689 DIE(aTHX_ PL_no_aelem, SvIV(*MARK));
4691 *MARK = sv_mortalcopy(*MARK);
4693 *++MARK = svp ? *svp : &PL_sv_undef;
4695 if (GIMME_V != G_ARRAY) {
4696 MARK = SP - items*2;
4697 *++MARK = items > 0 ? *SP : &PL_sv_undef;
4707 AV *array = MUTABLE_AV(POPs);
4708 const U8 gimme = GIMME_V;
4709 IV *iterp = Perl_av_iter_p(aTHX_ array);
4710 const IV current = (*iterp)++;
4712 if (current > av_tindex(array)) {
4714 if (gimme == G_SCALAR)
4722 if (gimme == G_ARRAY) {
4723 SV **const element = av_fetch(array, current, 0);
4724 PUSHs(element ? *element : &PL_sv_undef);
4729 /* also used for: pp_avalues()*/
4733 AV *array = MUTABLE_AV(POPs);
4734 const U8 gimme = GIMME_V;
4736 *Perl_av_iter_p(aTHX_ array) = 0;
4738 if (gimme == G_SCALAR) {
4740 PUSHi(av_tindex(array) + 1);
4742 else if (gimme == G_ARRAY) {
4743 if (UNLIKELY(PL_op->op_private & OPpMAYBE_LVSUB)) {
4744 const I32 flags = is_lvalue_sub();
4745 if (flags && !(flags & OPpENTERSUB_INARGS))
4746 /* diag_listed_as: Can't modify %s in %s */
4748 "Can't modify keys on array in list assignment");
4751 IV n = Perl_av_len(aTHX_ array);
4756 if ( PL_op->op_type == OP_AKEYS
4757 || ( PL_op->op_type == OP_AVHVSWITCH
4758 && (PL_op->op_private & 3) + OP_AEACH == OP_AKEYS ))
4760 for (i = 0; i <= n; i++) {
4765 for (i = 0; i <= n; i++) {
4766 SV *const *const elem = Perl_av_fetch(aTHX_ array, i, 0);
4767 PUSHs(elem ? *elem : &PL_sv_undef);
4775 /* Associative arrays. */
4780 HV * hash = MUTABLE_HV(POPs);
4782 const U8 gimme = GIMME_V;
4784 entry = hv_iternext(hash);
4788 SV* const sv = hv_iterkeysv(entry);
4790 if (gimme == G_ARRAY) {
4792 val = hv_iterval(hash, entry);
4796 else if (gimme == G_SCALAR)
4803 S_do_delete_local(pTHX)
4806 const U8 gimme = GIMME_V;
4809 const bool sliced = !!(PL_op->op_private & OPpSLICE);
4810 SV **unsliced_keysv = sliced ? NULL : sp--;
4811 SV * const osv = POPs;
4812 SV **mark = sliced ? PL_stack_base + POPMARK : unsliced_keysv-1;
4814 const bool tied = SvRMAGICAL(osv)
4815 && mg_find((const SV *)osv, PERL_MAGIC_tied);
4816 const bool can_preserve = SvCANEXISTDELETE(osv);
4817 const U32 type = SvTYPE(osv);
4818 SV ** const end = sliced ? SP : unsliced_keysv;
4820 if (type == SVt_PVHV) { /* hash element */
4821 HV * const hv = MUTABLE_HV(osv);
4822 while (++MARK <= end) {
4823 SV * const keysv = *MARK;
4825 bool preeminent = TRUE;
4827 preeminent = hv_exists_ent(hv, keysv, 0);
4829 HE *he = hv_fetch_ent(hv, keysv, 1, 0);
4836 sv = hv_delete_ent(hv, keysv, 0, 0);
4838 SvREFCNT_inc_simple_void(sv); /* De-mortalize */
4841 if (!sv) DIE(aTHX_ PL_no_helem_sv, SVfARG(keysv));
4842 save_helem_flags(hv, keysv, &sv, SAVEf_KEEPOLDELEM);
4844 *MARK = sv_mortalcopy(sv);
4850 SAVEHDELETE(hv, keysv);
4851 *MARK = &PL_sv_undef;
4855 else if (type == SVt_PVAV) { /* array element */
4856 if (PL_op->op_flags & OPf_SPECIAL) {
4857 AV * const av = MUTABLE_AV(osv);
4858 while (++MARK <= end) {
4859 SSize_t idx = SvIV(*MARK);
4861 bool preeminent = TRUE;
4863 preeminent = av_exists(av, idx);
4865 SV **svp = av_fetch(av, idx, 1);
4872 sv = av_delete(av, idx, 0);
4874 SvREFCNT_inc_simple_void(sv); /* De-mortalize */
4877 save_aelem_flags(av, idx, &sv, SAVEf_KEEPOLDELEM);
4879 *MARK = sv_mortalcopy(sv);
4885 SAVEADELETE(av, idx);
4886 *MARK = &PL_sv_undef;
4891 DIE(aTHX_ "panic: avhv_delete no longer supported");
4894 DIE(aTHX_ "Not a HASH reference");
4896 if (gimme == G_VOID)
4898 else if (gimme == G_SCALAR) {
4903 *++MARK = &PL_sv_undef;
4907 else if (gimme != G_VOID)
4908 PUSHs(*unsliced_keysv);
4919 if (PL_op->op_private & OPpLVAL_INTRO)
4920 return do_delete_local();
4923 discard = (gimme == G_VOID) ? G_DISCARD : 0;
4925 if (PL_op->op_private & (OPpSLICE|OPpKVSLICE)) {
4927 HV * const hv = MUTABLE_HV(POPs);
4928 const U32 hvtype = SvTYPE(hv);
4930 if (PL_op->op_private & OPpKVSLICE) {
4931 SSize_t items = SP - MARK;
4935 *(MARK+items*2-1) = *(MARK+items);
4942 if (hvtype == SVt_PVHV) { /* hash element */
4943 while ((MARK += (1+skip)) <= SP) {
4944 SV * const sv = hv_delete_ent(hv, *(MARK-skip), discard, 0);
4945 *MARK = sv ? sv : &PL_sv_undef;
4948 else if (hvtype == SVt_PVAV) { /* array element */
4949 if (PL_op->op_flags & OPf_SPECIAL) {
4950 while ((MARK += (1+skip)) <= SP) {
4951 SV * const sv = av_delete(MUTABLE_AV(hv), SvIV(*(MARK-skip)), discard);
4952 *MARK = sv ? sv : &PL_sv_undef;
4957 DIE(aTHX_ "Not a HASH reference");
4960 else if (gimme == G_SCALAR) {
4965 *++MARK = &PL_sv_undef;
4971 HV * const hv = MUTABLE_HV(POPs);
4973 if (SvTYPE(hv) == SVt_PVHV)
4974 sv = hv_delete_ent(hv, keysv, discard, 0);
4975 else if (SvTYPE(hv) == SVt_PVAV) {
4976 if (PL_op->op_flags & OPf_SPECIAL)
4977 sv = av_delete(MUTABLE_AV(hv), SvIV(keysv), discard);
4979 DIE(aTHX_ "panic: avhv_delete no longer supported");
4982 DIE(aTHX_ "Not a HASH reference");
4997 if (UNLIKELY( PL_op->op_private & OPpEXISTS_SUB )) {
4999 SV * const sv = POPs;
5000 CV * const cv = sv_2cv(sv, &hv, &gv, 0);
5003 if (gv && isGV(gv) && GvCV(gv) && !GvCVGEN(gv))
5008 hv = MUTABLE_HV(POPs);
5009 if (LIKELY( SvTYPE(hv) == SVt_PVHV )) {
5010 if (hv_exists_ent(hv, tmpsv, 0))
5013 else if (SvTYPE(hv) == SVt_PVAV) {
5014 if (PL_op->op_flags & OPf_SPECIAL) { /* array element */
5015 if (av_exists(MUTABLE_AV(hv), SvIV(tmpsv)))
5020 DIE(aTHX_ "Not a HASH reference");
5027 dSP; dMARK; dORIGMARK;
5028 HV * const hv = MUTABLE_HV(POPs);
5029 const I32 lval = (PL_op->op_flags & OPf_MOD || LVRET);
5030 const bool localizing = PL_op->op_private & OPpLVAL_INTRO;
5031 bool can_preserve = FALSE;
5037 if (SvCANEXISTDELETE(hv))
5038 can_preserve = TRUE;
5041 while (++MARK <= SP) {
5042 SV * const keysv = *MARK;
5045 bool preeminent = TRUE;
5047 if (localizing && can_preserve) {
5048 /* If we can determine whether the element exist,
5049 * try to preserve the existenceness of a tied hash
5050 * element by using EXISTS and DELETE if possible.
5051 * Fallback to FETCH and STORE otherwise. */
5052 preeminent = hv_exists_ent(hv, keysv, 0);
5055 he = hv_fetch_ent(hv, keysv, lval, 0);
5056 svp = he ? &HeVAL(he) : NULL;
5059 if (!svp || !*svp || *svp == &PL_sv_undef) {
5060 DIE(aTHX_ PL_no_helem_sv, SVfARG(keysv));
5063 if (HvNAME_get(hv) && isGV_or_RVCV(*svp))
5064 save_gp(MUTABLE_GV(*svp), !(PL_op->op_flags & OPf_SPECIAL));
5065 else if (preeminent)
5066 save_helem_flags(hv, keysv, svp,
5067 (PL_op->op_flags & OPf_SPECIAL) ? 0 : SAVEf_SETMAGIC);
5069 SAVEHDELETE(hv, keysv);
5072 *MARK = svp && *svp ? *svp : &PL_sv_undef;
5074 if (GIMME_V != G_ARRAY) {
5076 *++MARK = SP > ORIGMARK ? *SP : &PL_sv_undef;
5085 HV * const hv = MUTABLE_HV(POPs);
5086 I32 lval = (PL_op->op_flags & OPf_MOD);
5087 SSize_t items = SP - MARK;
5089 if (PL_op->op_private & OPpMAYBE_LVSUB) {
5090 const I32 flags = is_lvalue_sub();
5092 if (!(flags & OPpENTERSUB_INARGS))
5093 /* diag_listed_as: Can't modify %s in %s */
5094 Perl_croak(aTHX_ "Can't modify key/value hash slice in %s assignment",
5095 GIMME_V == G_ARRAY ? "list" : "scalar");
5102 *(MARK+items*2-1) = *(MARK+items);
5108 while (++MARK <= SP) {
5109 SV * const keysv = *MARK;
5113 he = hv_fetch_ent(hv, keysv, lval, 0);
5114 svp = he ? &HeVAL(he) : NULL;
5117 if (!svp || !*svp || *svp == &PL_sv_undef) {
5118 DIE(aTHX_ PL_no_helem_sv, SVfARG(keysv));
5120 *MARK = sv_mortalcopy(*MARK);
5122 *++MARK = svp && *svp ? *svp : &PL_sv_undef;
5124 if (GIMME_V != G_ARRAY) {
5125 MARK = SP - items*2;
5126 *++MARK = items > 0 ? *SP : &PL_sv_undef;
5132 /* List operators. */
5136 I32 markidx = POPMARK;
5137 if (GIMME_V != G_ARRAY) {
5138 /* don't initialize mark here, EXTEND() may move the stack */
5141 EXTEND(SP, 1); /* in case no arguments, as in @empty */
5142 mark = PL_stack_base + markidx;
5144 *MARK = *SP; /* unwanted list, return last item */
5146 *MARK = &PL_sv_undef;
5156 SV ** const lastrelem = PL_stack_sp;
5157 SV ** const lastlelem = PL_stack_base + POPMARK;
5158 SV ** const firstlelem = PL_stack_base + POPMARK + 1;
5159 SV ** const firstrelem = lastlelem + 1;
5160 const U8 mod = PL_op->op_flags & OPf_MOD;
5162 const I32 max = lastrelem - lastlelem;
5165 if (GIMME_V != G_ARRAY) {
5166 if (lastlelem < firstlelem) {
5168 *firstlelem = &PL_sv_undef;
5171 I32 ix = SvIV(*lastlelem);
5174 if (ix < 0 || ix >= max)
5175 *firstlelem = &PL_sv_undef;
5177 *firstlelem = firstrelem[ix];
5184 SP = firstlelem - 1;
5188 for (lelem = firstlelem; lelem <= lastlelem; lelem++) {
5189 I32 ix = SvIV(*lelem);
5192 if (ix < 0 || ix >= max)
5193 *lelem = &PL_sv_undef;
5195 if (!(*lelem = firstrelem[ix]))
5196 *lelem = &PL_sv_undef;
5197 else if (mod && SvPADTMP(*lelem)) {
5198 *lelem = firstrelem[ix] = sv_mortalcopy(*lelem);
5209 const I32 items = SP - MARK;
5210 SV * const av = MUTABLE_SV(av_make(items, MARK+1));
5212 mXPUSHs((PL_op->op_flags & OPf_SPECIAL)
5213 ? newRV_noinc(av) : av);
5219 dSP; dMARK; dORIGMARK;
5220 HV* const hv = newHV();
5221 SV* const retval = sv_2mortal( PL_op->op_flags & OPf_SPECIAL
5222 ? newRV_noinc(MUTABLE_SV(hv))
5227 (MARK++, SvGMAGICAL(*MARK) ? sv_mortalcopy(*MARK) : *MARK);
5234 sv_setsv_nomg(val, *MARK);
5238 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Odd number of elements in anonymous hash");
5241 (void)hv_store_ent(hv,key,val,0);
5250 dSP; dMARK; dORIGMARK;
5251 int num_args = (SP - MARK);
5252 AV *ary = MUTABLE_AV(*++MARK);
5261 const MAGIC * const mg = SvTIED_mg((const SV *)ary, PERL_MAGIC_tied);
5264 return Perl_tied_method(aTHX_ SV_CONST(SPLICE), mark - 1, MUTABLE_SV(ary), mg,
5265 GIMME_V | TIED_METHOD_ARGUMENTS_ON_STACK,
5269 if (SvREADONLY(ary))
5270 Perl_croak_no_modify();
5275 offset = i = SvIV(*MARK);
5277 offset += AvFILLp(ary) + 1;
5279 DIE(aTHX_ PL_no_aelem, i);
5281 length = SvIVx(*MARK++);
5283 length += AvFILLp(ary) - offset + 1;
5289 length = AvMAX(ary) + 1; /* close enough to infinity */
5293 length = AvMAX(ary) + 1;
5295 if (offset > AvFILLp(ary) + 1) {
5297 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "splice() offset past end of array" );
5298 offset = AvFILLp(ary) + 1;
5300 after = AvFILLp(ary) + 1 - (offset + length);
5301 if (after < 0) { /* not that much array */
5302 length += after; /* offset+length now in array */
5308 /* At this point, MARK .. SP-1 is our new LIST */
5311 diff = newlen - length;
5312 if (newlen && !AvREAL(ary) && AvREIFY(ary))
5315 /* make new elements SVs now: avoid problems if they're from the array */
5316 for (dst = MARK, i = newlen; i; i--) {
5317 SV * const h = *dst;
5318 *dst++ = newSVsv(h);
5321 if (diff < 0) { /* shrinking the area */
5322 SV **tmparyval = NULL;
5324 Newx(tmparyval, newlen, SV*); /* so remember insertion */
5325 Copy(MARK, tmparyval, newlen, SV*);
5328 MARK = ORIGMARK + 1;
5329 if (GIMME_V == G_ARRAY) { /* copy return vals to stack */
5330 const bool real = cBOOL(AvREAL(ary));
5331 MEXTEND(MARK, length);
5333 EXTEND_MORTAL(length);
5334 for (i = 0, dst = MARK; i < length; i++) {
5335 if ((*dst = AvARRAY(ary)[i+offset])) {
5337 sv_2mortal(*dst); /* free them eventually */
5340 *dst = &PL_sv_undef;
5346 *MARK = AvARRAY(ary)[offset+length-1];
5349 for (i = length - 1, dst = &AvARRAY(ary)[offset]; i > 0; i--)
5350 SvREFCNT_dec(*dst++); /* free them now */
5353 *MARK = &PL_sv_undef;
5355 AvFILLp(ary) += diff;
5357 /* pull up or down? */
5359 if (offset < after) { /* easier to pull up */
5360 if (offset) { /* esp. if nothing to pull */
5361 src = &AvARRAY(ary)[offset-1];
5362 dst = src - diff; /* diff is negative */
5363 for (i = offset; i > 0; i--) /* can't trust Copy */
5367 AvARRAY(ary) = AvARRAY(ary) - diff; /* diff is negative */
5371 if (after) { /* anything to pull down? */
5372 src = AvARRAY(ary) + offset + length;
5373 dst = src + diff; /* diff is negative */
5374 Move(src, dst, after, SV*);
5376 dst = &AvARRAY(ary)[AvFILLp(ary)+1];
5377 /* avoid later double free */
5384 Copy( tmparyval, AvARRAY(ary) + offset, newlen, SV* );
5385 Safefree(tmparyval);
5388 else { /* no, expanding (or same) */
5389 SV** tmparyval = NULL;
5391 Newx(tmparyval, length, SV*); /* so remember deletion */
5392 Copy(AvARRAY(ary)+offset, tmparyval, length, SV*);
5395 if (diff > 0) { /* expanding */
5396 /* push up or down? */
5397 if (offset < after && diff <= AvARRAY(ary) - AvALLOC(ary)) {
5401 Move(src, dst, offset, SV*);
5403 AvARRAY(ary) = AvARRAY(ary) - diff;/* diff is positive */
5405 AvFILLp(ary) += diff;
5408 if (AvFILLp(ary) + diff >= AvMAX(ary)) /* oh, well */
5409 av_extend(ary, AvFILLp(ary) + diff);
5410 AvFILLp(ary) += diff;
5413 dst = AvARRAY(ary) + AvFILLp(ary);
5415 for (i = after; i; i--) {
5423 Copy( MARK, AvARRAY(ary) + offset, newlen, SV* );
5426 MARK = ORIGMARK + 1;
5427 if (GIMME_V == G_ARRAY) { /* copy return vals to stack */
5429 const bool real = cBOOL(AvREAL(ary));
5431 EXTEND_MORTAL(length);
5432 for (i = 0, dst = MARK; i < length; i++) {
5433 if ((*dst = tmparyval[i])) {
5435 sv_2mortal(*dst); /* free them eventually */
5437 else *dst = &PL_sv_undef;
5443 else if (length--) {
5444 *MARK = tmparyval[length];
5447 while (length-- > 0)
5448 SvREFCNT_dec(tmparyval[length]);
5451 *MARK = &PL_sv_undef;
5454 *MARK = &PL_sv_undef;
5455 Safefree(tmparyval);
5459 mg_set(MUTABLE_SV(ary));
5467 dSP; dMARK; dORIGMARK; dTARGET;
5468 AV * const ary = MUTABLE_AV(*++MARK);
5469 const MAGIC * const mg = SvTIED_mg((const SV *)ary, PERL_MAGIC_tied);
5472 *MARK-- = SvTIED_obj(MUTABLE_SV(ary), mg);
5475 ENTER_with_name("call_PUSH");
5476 call_sv(SV_CONST(PUSH),G_SCALAR|G_DISCARD|G_METHOD_NAMED);
5477 LEAVE_with_name("call_PUSH");
5478 /* SPAGAIN; not needed: SP is assigned to immediately below */
5481 /* PL_delaymagic is restored by JUMPENV_POP on dieing, so we
5482 * only need to save locally, not on the save stack */
5483 U16 old_delaymagic = PL_delaymagic;
5485 if (SvREADONLY(ary) && MARK < SP) Perl_croak_no_modify();
5486 PL_delaymagic = DM_DELAY;
5487 for (++MARK; MARK <= SP; MARK++) {
5489 if (*MARK) SvGETMAGIC(*MARK);
5492 sv_setsv_nomg(sv, *MARK);
5493 av_store(ary, AvFILLp(ary)+1, sv);
5495 if (PL_delaymagic & DM_ARRAY_ISA)
5496 mg_set(MUTABLE_SV(ary));
5497 PL_delaymagic = old_delaymagic;
5500 if (OP_GIMME(PL_op, 0) != G_VOID) {
5501 PUSHi( AvFILL(ary) + 1 );
5506 /* also used for: pp_pop()*/
5510 AV * const av = PL_op->op_flags & OPf_SPECIAL
5511 ? MUTABLE_AV(GvAVn(PL_defgv)) : MUTABLE_AV(POPs);
5512 SV * const sv = PL_op->op_type == OP_SHIFT ? av_shift(av) : av_pop(av);
5516 (void)sv_2mortal(sv);
5523 dSP; dMARK; dORIGMARK; dTARGET;
5524 AV *ary = MUTABLE_AV(*++MARK);
5525 const MAGIC * const mg = SvTIED_mg((const SV *)ary, PERL_MAGIC_tied);
5528 *MARK-- = SvTIED_obj(MUTABLE_SV(ary), mg);
5531 ENTER_with_name("call_UNSHIFT");
5532 call_sv(SV_CONST(UNSHIFT),G_SCALAR|G_DISCARD|G_METHOD_NAMED);
5533 LEAVE_with_name("call_UNSHIFT");
5534 /* SPAGAIN; not needed: SP is assigned to immediately below */
5537 /* PL_delaymagic is restored by JUMPENV_POP on dieing, so we
5538 * only need to save locally, not on the save stack */
5539 U16 old_delaymagic = PL_delaymagic;
5542 av_unshift(ary, SP - MARK);
5543 PL_delaymagic = DM_DELAY;
5545 SV * const sv = newSVsv(*++MARK);
5546 (void)av_store(ary, i++, sv);
5548 if (PL_delaymagic & DM_ARRAY_ISA)
5549 mg_set(MUTABLE_SV(ary));
5550 PL_delaymagic = old_delaymagic;
5553 if (OP_GIMME(PL_op, 0) != G_VOID) {
5554 PUSHi( AvFILL(ary) + 1 );
5563 if (GIMME_V == G_ARRAY) {
5564 if (PL_op->op_private & OPpREVERSE_INPLACE) {
5568 assert( MARK+1 == SP && *SP && SvTYPE(*SP) == SVt_PVAV);
5569 (void)POPMARK; /* remove mark associated with ex-OP_AASSIGN */
5570 av = MUTABLE_AV((*SP));
5571 /* In-place reversing only happens in void context for the array
5572 * assignment. We don't need to push anything on the stack. */
5575 if (SvMAGICAL(av)) {
5577 SV *tmp = sv_newmortal();
5578 /* For SvCANEXISTDELETE */
5581 bool can_preserve = SvCANEXISTDELETE(av);
5583 for (i = 0, j = av_tindex(av); i < j; ++i, --j) {
5587 if (!av_exists(av, i)) {
5588 if (av_exists(av, j)) {
5589 SV *sv = av_delete(av, j, 0);
5590 begin = *av_fetch(av, i, TRUE);
5591 sv_setsv_mg(begin, sv);
5595 else if (!av_exists(av, j)) {
5596 SV *sv = av_delete(av, i, 0);
5597 end = *av_fetch(av, j, TRUE);
5598 sv_setsv_mg(end, sv);
5603 begin = *av_fetch(av, i, TRUE);
5604 end = *av_fetch(av, j, TRUE);
5605 sv_setsv(tmp, begin);
5606 sv_setsv_mg(begin, end);
5607 sv_setsv_mg(end, tmp);
5611 SV **begin = AvARRAY(av);
5614 SV **end = begin + AvFILLp(av);
5616 while (begin < end) {
5617 SV * const tmp = *begin;
5621 if (tmp && SvWEAKREF(tmp))
5625 /* make sure we catch the middle element */
5626 if (begin == end && *begin && SvWEAKREF(*begin))
5627 sv_rvunweaken(*begin);
5635 SV * const tmp = *MARK;
5639 /* safe as long as stack cannot get extended in the above */
5648 SvUTF8_off(TARG); /* decontaminate */
5649 if (SP - MARK > 1) {
5650 do_join(TARG, &PL_sv_no, MARK, SP);
5653 } else if (SP > MARK) {
5654 sv_setsv(TARG, *SP);
5657 sv_setsv(TARG, DEFSV);
5661 up = SvPV_force(TARG, len);
5664 if (DO_UTF8(TARG)) { /* first reverse each character */
5665 U8* s = (U8*)SvPVX(TARG);
5666 const U8* send = (U8*)(s + len);
5668 if (UTF8_IS_INVARIANT(*s)) {
5673 if (!utf8_to_uvchr_buf(s, send, 0))
5677 down = (char*)(s - 1);
5678 /* reverse this character */
5680 const char tmp = *up;
5688 down = SvPVX(TARG) + len - 1;
5690 const char tmp = *up;
5694 (void)SvPOK_only_UTF8(TARG);
5703 AV *ary = ( (PL_op->op_private & OPpSPLIT_ASSIGN) /* @a = split */
5704 && (PL_op->op_flags & OPf_STACKED)) /* @{expr} = split */
5705 ? (AV *)POPs : NULL;
5706 IV limit = POPi; /* note, negative is forever */
5707 SV * const sv = POPs;
5709 const char *s = SvPV_const(sv, len);
5710 const bool do_utf8 = DO_UTF8(sv);
5711 const bool in_uni_8_bit = IN_UNI_8_BIT;
5712 const char *strend = s + len;
5713 PMOP *pm = cPMOPx(PL_op);
5718 const STRLEN slen = do_utf8
5719 ? utf8_length((U8*)s, (U8*)strend)
5720 : (STRLEN)(strend - s);
5721 SSize_t maxiters = slen + 10;
5722 I32 trailing_empty = 0;
5724 const IV origlimit = limit;
5727 const U8 gimme = GIMME_V;
5729 I32 oldsave = PL_savestack_ix;
5730 U32 make_mortal = SVs_TEMP;
5736 TAINT_IF(get_regex_charset(RX_EXTFLAGS(rx)) == REGEX_LOCALE_CHARSET &&
5737 (RX_EXTFLAGS(rx) & (RXf_WHITE | RXf_SKIPWHITE)));
5739 /* handle @ary = split(...) optimisation */
5740 if (PL_op->op_private & OPpSPLIT_ASSIGN) {
5741 if (!(PL_op->op_flags & OPf_STACKED)) {
5742 if (PL_op->op_private & OPpSPLIT_LEX) {
5743 if (PL_op->op_private & OPpLVAL_INTRO)
5744 SAVECLEARSV(PAD_SVl(pm->op_pmreplrootu.op_pmtargetoff));
5745 ary = (AV *)PAD_SVl(pm->op_pmreplrootu.op_pmtargetoff);
5750 MUTABLE_GV(PAD_SVl(pm->op_pmreplrootu.op_pmtargetoff));
5752 pm->op_pmreplrootu.op_pmtargetgv;
5754 if (PL_op->op_private & OPpLVAL_INTRO)
5759 /* skip anything pushed by OPpLVAL_INTRO above */
5760 oldsave = PL_savestack_ix;
5766 (void)sv_2mortal(SvREFCNT_inc_simple_NN(sv));
5769 if ((mg = SvTIED_mg((const SV *)ary, PERL_MAGIC_tied))) {
5771 XPUSHs(SvTIED_obj(MUTABLE_SV(ary), mg));
5778 for (i = AvFILLp(ary); i >= 0; i--)
5779 AvARRAY(ary)[i] = &PL_sv_undef; /* don't free mere refs */
5781 /* temporarily switch stacks */
5782 SAVESWITCHSTACK(PL_curstack, ary);
5787 base = SP - PL_stack_base;
5789 if (RX_EXTFLAGS(rx) & RXf_SKIPWHITE) {
5791 while (s < strend && isSPACE_utf8_safe(s, strend))
5794 else if (get_regex_charset(RX_EXTFLAGS(rx)) == REGEX_LOCALE_CHARSET) {
5795 while (s < strend && isSPACE_LC(*s))
5798 else if (in_uni_8_bit) {
5799 while (s < strend && isSPACE_L1(*s))
5803 while (s < strend && isSPACE(*s))
5807 if (RX_EXTFLAGS(rx) & RXf_PMf_MULTILINE) {
5811 gimme_scalar = gimme == G_SCALAR && !ary;
5814 limit = maxiters + 2;
5815 if (RX_EXTFLAGS(rx) & RXf_WHITE) {
5818 /* this one uses 'm' and is a negative test */
5820 while (m < strend && ! isSPACE_utf8_safe(m, strend) ) {
5821 const int t = UTF8SKIP(m);
5822 /* isSPACE_utf8_safe returns FALSE for malform utf8 */
5829 else if (get_regex_charset(RX_EXTFLAGS(rx)) == REGEX_LOCALE_CHARSET)
5831 while (m < strend && !isSPACE_LC(*m))
5834 else if (in_uni_8_bit) {
5835 while (m < strend && !isSPACE_L1(*m))
5838 while (m < strend && !isSPACE(*m))
5851 dstr = newSVpvn_flags(s, m-s,
5852 (do_utf8 ? SVf_UTF8 : 0) | make_mortal);
5856 /* skip the whitespace found last */
5858 s = m + UTF8SKIP(m);
5862 /* this one uses 's' and is a positive test */
5864 while (s < strend && isSPACE_utf8_safe(s, strend) )
5867 else if (get_regex_charset(RX_EXTFLAGS(rx)) == REGEX_LOCALE_CHARSET)
5869 while (s < strend && isSPACE_LC(*s))
5872 else if (in_uni_8_bit) {
5873 while (s < strend && isSPACE_L1(*s))
5876 while (s < strend && isSPACE(*s))
5881 else if (RX_EXTFLAGS(rx) & RXf_START_ONLY) {
5883 for (m = s; m < strend && *m != '\n'; m++)
5896 dstr = newSVpvn_flags(s, m-s,
5897 (do_utf8 ? SVf_UTF8 : 0) | make_mortal);
5903 else if (RX_EXTFLAGS(rx) & RXf_NULL && !(s >= strend)) {
5905 Pre-extend the stack, either the number of bytes or
5906 characters in the string or a limited amount, triggered by:
5908 my ($x, $y) = split //, $str;
5912 if (!gimme_scalar) {
5913 const IV items = limit - 1;
5914 /* setting it to -1 will trigger a panic in EXTEND() */
5915 const SSize_t sslen = slen > SSize_t_MAX ? -1 : (SSize_t)slen;
5916 if (items >=0 && items < sslen)
5924 /* keep track of how many bytes we skip over */
5934 dstr = newSVpvn_flags(m, s-m, SVf_UTF8 | make_mortal);
5947 dstr = newSVpvn(s, 1);
5963 else if (do_utf8 == (RX_UTF8(rx) != 0) &&
5964 (RX_EXTFLAGS(rx) & RXf_USE_INTUIT) && !RX_NPARENS(rx)
5965 && (RX_EXTFLAGS(rx) & RXf_CHECK_ALL)
5966 && !(RX_EXTFLAGS(rx) & RXf_IS_ANCHORED)) {
5967 const int tail = (RX_EXTFLAGS(rx) & RXf_INTUIT_TAIL);
5968 SV * const csv = CALLREG_INTUIT_STRING(rx);
5970 len = RX_MINLENRET(rx);
5971 if (len == 1 && !RX_UTF8(rx) && !tail) {
5972 const char c = *SvPV_nolen_const(csv);
5974 for (m = s; m < strend && *m != c; m++)
5985 dstr = newSVpvn_flags(s, m-s,
5986 (do_utf8 ? SVf_UTF8 : 0) | make_mortal);
5989 /* The rx->minlen is in characters but we want to step
5990 * s ahead by bytes. */
5992 s = (char*)utf8_hop((U8*)m, len);
5994 s = m + len; /* Fake \n at the end */
5998 while (s < strend && --limit &&
5999 (m = fbm_instr((unsigned char*)s, (unsigned char*)strend,
6000 csv, multiline ? FBMrf_MULTILINE : 0)) )
6009 dstr = newSVpvn_flags(s, m-s,
6010 (do_utf8 ? SVf_UTF8 : 0) | make_mortal);
6013 /* The rx->minlen is in characters but we want to step
6014 * s ahead by bytes. */
6016 s = (char*)utf8_hop((U8*)m, len);
6018 s = m + len; /* Fake \n at the end */
6023 maxiters += slen * RX_NPARENS(rx);
6024 while (s < strend && --limit)
6028 rex_return = CALLREGEXEC(rx, (char*)s, (char*)strend, (char*)orig, 1,
6031 if (rex_return == 0)
6033 TAINT_IF(RX_MATCH_TAINTED(rx));
6034 /* we never pass the REXEC_COPY_STR flag, so it should
6035 * never get copied */
6036 assert(!RX_MATCH_COPIED(rx));
6037 m = RX_OFFS(rx)[0].start + orig;
6046 dstr = newSVpvn_flags(s, m-s,
6047 (do_utf8 ? SVf_UTF8 : 0) | make_mortal);
6050 if (RX_NPARENS(rx)) {
6052 for (i = 1; i <= (I32)RX_NPARENS(rx); i++) {
6053 s = RX_OFFS(rx)[i].start + orig;
6054 m = RX_OFFS(rx)[i].end + orig;
6056 /* japhy (07/27/01) -- the (m && s) test doesn't catch
6057 parens that didn't match -- they should be set to
6058 undef, not the empty string */
6066 if (m >= orig && s >= orig) {
6067 dstr = newSVpvn_flags(s, m-s,
6068 (do_utf8 ? SVf_UTF8 : 0)
6072 dstr = &PL_sv_undef; /* undef, not "" */
6078 s = RX_OFFS(rx)[0].end + orig;
6082 if (!gimme_scalar) {
6083 iters = (SP - PL_stack_base) - base;
6085 if (iters > maxiters)
6086 DIE(aTHX_ "Split loop");
6088 /* keep field after final delim? */
6089 if (s < strend || (iters && origlimit)) {
6090 if (!gimme_scalar) {
6091 const STRLEN l = strend - s;
6092 dstr = newSVpvn_flags(s, l, (do_utf8 ? SVf_UTF8 : 0) | make_mortal);
6097 else if (!origlimit) {
6099 iters -= trailing_empty;
6101 while (iters > 0 && (!TOPs || !SvANY(TOPs) || SvCUR(TOPs) == 0)) {
6102 if (TOPs && !make_mortal)
6111 LEAVE_SCOPE(oldsave); /* may undo an earlier SWITCHSTACK */
6115 if (SvSMAGICAL(ary)) {
6117 mg_set(MUTABLE_SV(ary));
6120 if (gimme == G_ARRAY) {
6122 Copy(AvARRAY(ary), SP + 1, iters, SV*);
6129 ENTER_with_name("call_PUSH");
6130 call_sv(SV_CONST(PUSH),G_SCALAR|G_DISCARD|G_METHOD_NAMED);
6131 LEAVE_with_name("call_PUSH");
6133 if (gimme == G_ARRAY) {
6135 /* EXTEND should not be needed - we just popped them */
6137 for (i=0; i < iters; i++) {
6138 SV **svp = av_fetch(ary, i, FALSE);
6139 PUSHs((svp) ? *svp : &PL_sv_undef);
6146 if (gimme == G_ARRAY)
6158 SV *const sv = PAD_SVl(PL_op->op_targ);
6160 if (SvPADSTALE(sv)) {
6163 RETURNOP(cLOGOP->op_other);
6165 RETURNOP(cLOGOP->op_next);
6174 if (SvTYPE(retsv) == SVt_PVAV || SvTYPE(retsv) == SVt_PVHV
6175 || SvTYPE(retsv) == SVt_PVCV) {
6176 retsv = refto(retsv);
6183 /* used for: pp_padany(), pp_custom(); plus any system ops
6184 * that aren't implemented on a particular platform */
6186 PP(unimplemented_op)
6188 const Optype op_type = PL_op->op_type;
6189 /* Using OP_NAME() isn't going to be helpful here. Firstly, it doesn't cope
6190 with out of range op numbers - it only "special" cases op_custom.
6191 Secondly, as the three ops we "panic" on are padmy, mapstart and custom,
6192 if we get here for a custom op then that means that the custom op didn't
6193 have an implementation. Given that OP_NAME() looks up the custom op
6194 by its pp_addr, likely it will return NULL, unless someone (unhelpfully)
6195 registers &PL_unimplemented_op as the address of their custom op.
6196 NULL doesn't generate a useful error message. "custom" does. */
6197 const char *const name = op_type >= OP_max
6198 ? "[out of range]" : PL_op_name[PL_op->op_type];
6199 if(OP_IS_SOCKET(op_type))
6200 DIE(aTHX_ PL_no_sock_func, name);
6201 DIE(aTHX_ "panic: unimplemented op %s (#%d) called", name, op_type);
6205 S_maybe_unwind_defav(pTHX)
6207 if (CX_CUR()->cx_type & CXp_HASARGS) {
6208 PERL_CONTEXT *cx = CX_CUR();
6210 assert(CxHASARGS(cx));
6212 cx->cx_type &= ~CXp_HASARGS;
6216 /* For sorting out arguments passed to a &CORE:: subroutine */
6220 int opnum = SvIOK(cSVOP_sv) ? (int)SvUV(cSVOP_sv) : 0;
6221 int defgv = PL_opargs[opnum] & OA_DEFGV ||opnum==OP_GLOB, whicharg = 0;
6222 AV * const at_ = GvAV(PL_defgv);
6223 SV **svp = at_ ? AvARRAY(at_) : NULL;
6224 I32 minargs = 0, maxargs = 0, numargs = at_ ? AvFILLp(at_)+1 : 0;
6225 I32 oa = opnum ? PL_opargs[opnum] >> OASHIFT : 0;
6226 bool seen_question = 0;
6227 const char *err = NULL;
6228 const bool pushmark = PL_op->op_private & OPpCOREARGS_PUSHMARK;
6230 /* Count how many args there are first, to get some idea how far to
6231 extend the stack. */
6233 if ((oa & 7) == OA_LIST) { maxargs = I32_MAX; break; }
6235 if (oa & OA_OPTIONAL) seen_question = 1;
6236 if (!seen_question) minargs++;
6240 if(numargs < minargs) err = "Not enough";
6241 else if(numargs > maxargs) err = "Too many";
6243 /* diag_listed_as: Too many arguments for %s */
6245 "%s arguments for %s", err,
6246 opnum ? PL_op_desc[opnum] : SvPV_nolen_const(cSVOP_sv)
6249 /* Reset the stack pointer. Without this, we end up returning our own
6250 arguments in list context, in addition to the values we are supposed
6251 to return. nextstate usually does this on sub entry, but we need
6252 to run the next op with the caller's hints, so we cannot have a
6254 SP = PL_stack_base + CX_CUR()->blk_oldsp;
6256 if(!maxargs) RETURN;
6258 /* We do this here, rather than with a separate pushmark op, as it has
6259 to come in between two things this function does (stack reset and
6260 arg pushing). This seems the easiest way to do it. */
6263 (void)Perl_pp_pushmark(aTHX);
6266 EXTEND(SP, maxargs == I32_MAX ? numargs : maxargs);
6267 PUTBACK; /* The code below can die in various places. */
6269 oa = PL_opargs[opnum] >> OASHIFT;
6270 for (; oa&&(numargs||!pushmark); (void)(numargs&&(++svp,--numargs))) {
6275 if (!numargs && defgv && whicharg == minargs + 1) {
6278 else PUSHs(numargs ? svp && *svp ? *svp : &PL_sv_undef : NULL);
6282 PUSHs(svp && *svp ? *svp : &PL_sv_undef);
6289 if (CvUNIQUE(find_runcv_where(FIND_RUNCV_level_eq,1,NULL)))
6292 S_maybe_unwind_defav(aTHX);
6295 PUSHs((SV *)GvAVn(gv));
6298 if (!svp || !*svp || !SvROK(*svp)
6299 || SvTYPE(SvRV(*svp)) != SVt_PVAV)
6301 /* diag_listed_as: Type of arg %d to &CORE::%s must be %s*/
6302 "Type of arg %d to &CORE::%s must be array reference",
6303 whicharg, PL_op_desc[opnum]
6308 if (!svp || !*svp || !SvROK(*svp)
6309 || ( SvTYPE(SvRV(*svp)) != SVt_PVHV
6310 && ( opnum == OP_DBMCLOSE || opnum == OP_DBMOPEN
6311 || SvTYPE(SvRV(*svp)) != SVt_PVAV )))
6313 /* diag_listed_as: Type of arg %d to &CORE::%s must be %s*/
6314 "Type of arg %d to &CORE::%s must be hash%s reference",
6315 whicharg, PL_op_desc[opnum],
6316 opnum == OP_DBMCLOSE || opnum == OP_DBMOPEN
6323 if (!numargs) PUSHs(NULL);
6324 else if(svp && *svp && SvROK(*svp) && isGV_with_GP(SvRV(*svp)))
6325 /* no magic here, as the prototype will have added an extra
6326 refgen and we just want what was there before that */
6329 const bool constr = PL_op->op_private & whicharg;
6331 svp && *svp ? *svp : &PL_sv_undef,
6332 constr, cBOOL(CopHINTS_get(PL_curcop) & HINT_STRICT_REFS),
6338 if (!numargs) goto try_defsv;
6340 const bool wantscalar =
6341 PL_op->op_private & OPpCOREARGS_SCALARMOD;
6342 if (!svp || !*svp || !SvROK(*svp)
6343 /* We have to permit globrefs even for the \$ proto, as
6344 *foo is indistinguishable from ${\*foo}, and the proto-
6345 type permits the latter. */
6346 || SvTYPE(SvRV(*svp)) > (
6347 wantscalar ? SVt_PVLV
6348 : opnum == OP_LOCK || opnum == OP_UNDEF
6354 "Type of arg %d to &CORE::%s must be %s",
6355 whicharg, PL_op_name[opnum],
6357 ? "scalar reference"
6358 : opnum == OP_LOCK || opnum == OP_UNDEF
6359 ? "reference to one of [$@%&*]"
6360 : "reference to one of [$@%*]"
6363 if (opnum == OP_UNDEF && SvRV(*svp) == (SV *)PL_defgv) {
6364 /* Undo @_ localisation, so that sub exit does not undo
6365 part of our undeffing. */
6366 S_maybe_unwind_defav(aTHX);
6371 DIE(aTHX_ "panic: unknown OA_*: %x", (unsigned)(oa&7));
6379 /* Implement CORE::keys(),values(),each().
6381 * We won't know until run-time whether the arg is an array or hash,
6384 * pp_keys/pp_values/pp_each
6386 * pp_akeys/pp_avalues/pp_aeach
6388 * as appropriate (or whatever pp function actually implements the OP_FOO
6389 * functionality for each FOO).
6396 (SvTYPE(TOPs) == SVt_PVAV ? OP_AEACH : OP_EACH)
6397 + (PL_op->op_private & OPpAVHVSWITCH_MASK)
6405 if (PL_op->op_private & OPpOFFBYONE) {
6406 cv = find_runcv_where(FIND_RUNCV_level_eq, 1, NULL);
6408 else cv = find_runcv(NULL);
6409 XPUSHs(CvEVAL(cv) ? &PL_sv_undef : sv_2mortal(newRV((SV *)cv)));
6414 S_localise_aelem_lval(pTHX_ AV * const av, SV * const keysv,
6415 const bool can_preserve)
6417 const SSize_t ix = SvIV(keysv);
6418 if (can_preserve ? av_exists(av, ix) : TRUE) {
6419 SV ** const svp = av_fetch(av, ix, 1);
6421 Perl_croak(aTHX_ PL_no_aelem, ix);
6422 save_aelem(av, ix, svp);
6425 SAVEADELETE(av, ix);
6429 S_localise_helem_lval(pTHX_ HV * const hv, SV * const keysv,
6430 const bool can_preserve)
6432 if (can_preserve ? hv_exists_ent(hv, keysv, 0) : TRUE) {
6433 HE * const he = hv_fetch_ent(hv, keysv, 1, 0);
6434 SV ** const svp = he ? &HeVAL(he) : NULL;
6436 Perl_croak(aTHX_ PL_no_helem_sv, SVfARG(keysv));
6437 save_helem_flags(hv, keysv, svp, 0);
6440 SAVEHDELETE(hv, keysv);
6444 S_localise_gv_slot(pTHX_ GV *gv, U8 type)
6446 if (type == OPpLVREF_SV) {
6447 save_pushptrptr(gv, SvREFCNT_inc_simple(GvSV(gv)), SAVEt_GVSV);
6450 else if (type == OPpLVREF_AV)
6451 /* XXX Inefficient, as it creates a new AV, which we are
6452 about to clobber. */
6455 assert(type == OPpLVREF_HV);
6456 /* XXX Likewise inefficient. */
6465 SV * const key = PL_op->op_private & OPpLVREF_ELEM ? POPs : NULL;
6466 SV * const left = PL_op->op_flags & OPf_STACKED ? POPs : NULL;
6468 const char *bad = NULL;
6469 const U8 type = PL_op->op_private & OPpLVREF_TYPE;
6470 if (!SvROK(sv)) DIE(aTHX_ "Assigned value is not a reference");
6473 if (SvTYPE(SvRV(sv)) > SVt_PVLV)
6477 if (SvTYPE(SvRV(sv)) != SVt_PVAV)
6481 if (SvTYPE(SvRV(sv)) != SVt_PVHV)
6485 if (SvTYPE(SvRV(sv)) != SVt_PVCV)
6489 /* diag_listed_as: Assigned value is not %s reference */
6490 DIE(aTHX_ "Assigned value is not a%s reference", bad);
6494 switch (left ? SvTYPE(left) : 0) {
6497 SV * const old = PAD_SV(ARGTARG);
6498 PAD_SETSV(ARGTARG, SvREFCNT_inc_NN(SvRV(sv)));
6500 if ((PL_op->op_private & (OPpLVAL_INTRO|OPpPAD_STATE))
6502 SAVECLEARSV(PAD_SVl(ARGTARG));
6506 if (PL_op->op_private & OPpLVAL_INTRO) {
6507 S_localise_gv_slot(aTHX_ (GV *)left, type);
6509 gv_setref(left, sv);
6514 if (UNLIKELY(PL_op->op_private & OPpLVAL_INTRO)) {
6515 S_localise_aelem_lval(aTHX_ (AV *)left, key,
6516 SvCANEXISTDELETE(left));
6518 av_store((AV *)left, SvIV(key), SvREFCNT_inc_simple_NN(SvRV(sv)));
6521 if (UNLIKELY(PL_op->op_private & OPpLVAL_INTRO)) {
6523 S_localise_helem_lval(aTHX_ (HV *)left, key,
6524 SvCANEXISTDELETE(left));
6526 (void)hv_store_ent((HV *)left, key, SvREFCNT_inc_simple_NN(SvRV(sv)), 0);
6528 if (PL_op->op_flags & OPf_MOD)
6529 SETs(sv_2mortal(newSVsv(sv)));
6530 /* XXX else can weak references go stale before they are read, e.g.,
6539 SV * const ret = sv_2mortal(newSV_type(SVt_PVMG));
6540 SV * const elem = PL_op->op_private & OPpLVREF_ELEM ? POPs : NULL;
6541 SV * const arg = PL_op->op_flags & OPf_STACKED ? POPs : NULL;
6542 MAGIC * const mg = sv_magicext(ret, arg, PERL_MAGIC_lvref,
6543 &PL_vtbl_lvref, (char *)elem,
6544 elem ? HEf_SVKEY : (I32)ARGTARG);
6545 mg->mg_private = PL_op->op_private;
6546 if (PL_op->op_private & OPpLVREF_ITER)
6547 mg->mg_flags |= MGf_PERSIST;
6548 if (UNLIKELY(PL_op->op_private & OPpLVAL_INTRO)) {
6554 const bool can_preserve = SvCANEXISTDELETE(arg);
6555 if (SvTYPE(arg) == SVt_PVAV)
6556 S_localise_aelem_lval(aTHX_ (AV *)arg, elem, can_preserve);
6558 S_localise_helem_lval(aTHX_ (HV *)arg, elem, can_preserve);
6562 S_localise_gv_slot(aTHX_ (GV *)arg,
6563 PL_op->op_private & OPpLVREF_TYPE);
6565 else if (!(PL_op->op_private & OPpPAD_STATE))
6566 SAVECLEARSV(PAD_SVl(ARGTARG));
6575 AV * const av = (AV *)POPs;
6576 const bool localizing = PL_op->op_private & OPpLVAL_INTRO;
6577 bool can_preserve = FALSE;
6579 if (UNLIKELY(localizing)) {
6584 can_preserve = SvCANEXISTDELETE(av);
6586 if (SvTYPE(av) == SVt_PVAV) {
6589 for (svp = MARK + 1; svp <= SP; svp++) {
6590 const SSize_t elem = SvIV(*svp);
6594 if (max > AvMAX(av))
6599 while (++MARK <= SP) {
6600 SV * const elemsv = *MARK;
6601 if (SvTYPE(av) == SVt_PVAV)
6602 S_localise_aelem_lval(aTHX_ av, elemsv, can_preserve);
6604 S_localise_helem_lval(aTHX_ (HV *)av, elemsv, can_preserve);
6605 *MARK = sv_2mortal(newSV_type(SVt_PVMG));
6606 sv_magic(*MARK,(SV *)av,PERL_MAGIC_lvref,(char *)elemsv,HEf_SVKEY);
6613 if (PL_op->op_flags & OPf_STACKED)
6614 Perl_pp_rv2av(aTHX);
6616 Perl_pp_padav(aTHX);
6620 SETs(0); /* special alias marker that aassign recognises */
6630 SETs(sv_2mortal((SV *)newCONSTSUB(SvTYPE(CopSTASH(PL_curcop))==SVt_PVHV
6631 ? CopSTASH(PL_curcop)
6633 NULL, SvREFCNT_inc_simple_NN(sv))));
6638 /* process one subroutine argument - typically when the sub has a signature:
6639 * introduce PL_curpad[op_targ] and assign to it the value
6640 * for $: (OPf_STACKED ? *sp : $_[N])
6641 * for @/%: @_[N..$#_]
6643 * It's equivalent to
6646 * my $foo = (value-on-stack)
6648 * my @foo = @_[N..$#_]
6658 AV *defav = GvAV(PL_defgv); /* @_ */
6659 IV ix = PTR2IV(cUNOP_AUXo->op_aux);
6662 /* do 'my $var, @var or %var' action */
6663 padentry = &(PAD_SVl(o->op_targ));
6664 save_clearsv(padentry);
6667 if ((o->op_private & OPpARGELEM_MASK) == OPpARGELEM_SV) {
6668 if (o->op_flags & OPf_STACKED) {
6675 /* should already have been checked */
6677 #if IVSIZE > PTRSIZE
6678 assert(ix <= SSize_t_MAX);
6681 svp = av_fetch(defav, ix, FALSE);
6682 val = svp ? *svp : &PL_sv_undef;
6687 /* cargo-culted from pp_sassign */
6688 assert(TAINTING_get || !TAINT_get);
6689 if (UNLIKELY(TAINT_get) && !SvTAINTED(val))
6692 SvSetMagicSV(targ, val);
6696 /* must be AV or HV */
6698 assert(!(o->op_flags & OPf_STACKED));
6699 argc = ((IV)AvFILL(defav) + 1) - ix;
6701 /* This is a copy of the relevant parts of pp_aassign().
6703 if ((o->op_private & OPpARGELEM_MASK) == OPpARGELEM_AV) {
6706 if (AvFILL((AV*)targ) > -1) {
6707 /* target should usually be empty. If we get get
6708 * here, someone's been doing some weird closure tricks.
6709 * Make a copy of all args before clearing the array,
6710 * to avoid the equivalent of @a = ($a[0]) prematurely freeing
6711 * elements. See similar code in pp_aassign.
6713 for (i = 0; i < argc; i++) {
6714 SV **svp = av_fetch(defav, ix + i, FALSE);
6715 SV *newsv = newSV(0);
6716 sv_setsv_flags(newsv,
6717 svp ? *svp : &PL_sv_undef,
6718 (SV_DO_COW_SVSETSV|SV_NOSTEAL));
6719 if (!av_store(defav, ix + i, newsv))
6720 SvREFCNT_dec_NN(newsv);
6722 av_clear((AV*)targ);
6728 av_extend((AV*)targ, argc);
6733 SV **svp = av_fetch(defav, ix + i, FALSE);
6734 SV *val = svp ? *svp : &PL_sv_undef;
6736 sv_setsv(tmpsv, val);
6737 av_store((AV*)targ, i++, tmpsv);
6745 assert((o->op_private & OPpARGELEM_MASK) == OPpARGELEM_HV);
6747 if (SvRMAGICAL(targ) || HvUSEDKEYS((HV*)targ)) {
6748 /* see "target should usually be empty" comment above */
6749 for (i = 0; i < argc; i++) {
6750 SV **svp = av_fetch(defav, ix + i, FALSE);
6751 SV *newsv = newSV(0);
6752 sv_setsv_flags(newsv,
6753 svp ? *svp : &PL_sv_undef,
6754 (SV_DO_COW_SVSETSV|SV_NOSTEAL));
6755 if (!av_store(defav, ix + i, newsv))
6756 SvREFCNT_dec_NN(newsv);
6758 hv_clear((HV*)targ);
6763 assert(argc % 2 == 0);
6772 svp = av_fetch(defav, ix + i++, FALSE);
6773 key = svp ? *svp : &PL_sv_undef;
6774 svp = av_fetch(defav, ix + i++, FALSE);
6775 val = svp ? *svp : &PL_sv_undef;
6778 if (UNLIKELY(SvGMAGICAL(key)))
6779 key = sv_mortalcopy(key);
6781 sv_setsv(tmpsv, val);
6782 hv_store_ent((HV*)targ, key, tmpsv, 0);
6790 /* Handle a default value for one subroutine argument (typically as part
6791 * of a subroutine signature).
6792 * It's equivalent to
6793 * @_ > op_targ ? $_[op_targ] : result_of(op_other)
6795 * Intended to be used where op_next is an OP_ARGELEM
6797 * We abuse the op_targ field slightly: it's an index into @_ rather than
6803 OP * const o = PL_op;
6804 AV *defav = GvAV(PL_defgv); /* @_ */
6805 IV ix = (IV)o->op_targ;
6808 #if IVSIZE > PTRSIZE
6809 assert(ix <= SSize_t_MAX);
6812 if (AvFILL(defav) >= ix) {
6814 SV **svp = av_fetch(defav, ix, FALSE);
6815 SV *val = svp ? *svp : &PL_sv_undef;
6819 return cLOGOPo->op_other;
6824 S_find_runcv_name(void)
6839 sv = sv_2mortal(newSV(0));
6840 gv_fullname4(sv, gv, NULL, TRUE);
6844 /* Check a a subs arguments - i.e. that it has the correct number of args
6845 * (and anything else we might think of in future). Typically used with
6851 OP * const o = PL_op;
6852 UNOP_AUX_item *aux = cUNOP_AUXo->op_aux;
6853 IV params = aux[0].iv;
6854 IV opt_params = aux[1].iv;
6855 char slurpy = (char)(aux[2].iv);
6856 AV *defav = GvAV(PL_defgv); /* @_ */
6860 assert(!SvMAGICAL(defav));
6861 argc = (AvFILLp(defav) + 1);
6862 too_few = (argc < (params - opt_params));
6864 if (UNLIKELY(too_few || (!slurpy && argc > params)))
6865 /* diag_listed_as: Too few arguments for subroutine '%s' */
6866 /* diag_listed_as: Too many arguments for subroutine '%s' */
6867 Perl_croak_caller("Too %s arguments for subroutine '%" SVf "'",
6868 too_few ? "few" : "many", S_find_runcv_name());
6870 if (UNLIKELY(slurpy == '%' && argc > params && (argc - params) % 2))
6871 /* diag_listed_as: Odd name/value argument for subroutine '%s' */
6872 Perl_croak_caller("Odd name/value argument for subroutine '%" SVf "'",
6873 S_find_runcv_name());
6879 * ex: set ts=8 sts=4 sw=4 et: