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) {
133 Perl_croak_no_modify();
134 if (cUNOP->op_targ) {
135 SV * const namesv = PAD_SV(cUNOP->op_targ);
136 HV *stash = CopSTASH(PL_curcop);
137 if (SvTYPE(stash) != SVt_PVHV) stash = NULL;
138 gv = MUTABLE_GV(newSV(0));
139 gv_init_sv(gv, stash, namesv, 0);
142 const char * const name = CopSTASHPV(PL_curcop);
143 gv = newGVgen_flags(name,
144 HvNAMEUTF8(CopSTASH(PL_curcop)) ? SVf_UTF8 : 0 );
145 SvREFCNT_inc_simple_void_NN(gv);
147 prepare_SV_for_RV(sv);
148 SvRV_set(sv, MUTABLE_SV(gv));
153 if (PL_op->op_flags & OPf_REF || strict) {
154 Perl_die(aTHX_ PL_no_usym, "a symbol");
156 if (ckWARN(WARN_UNINITIALIZED))
162 if (!(sv = MUTABLE_SV(gv_fetchsv_nomg(
163 sv, GV_ADDMG, SVt_PVGV
172 (SvPOKp(sv) && SvCUR(sv)>32 ? "..." : ""),
176 if ((PL_op->op_private & (OPpLVAL_INTRO|OPpDONT_INIT_GV))
177 == OPpDONT_INIT_GV) {
178 /* We are the target of a coderef assignment. Return
179 the scalar unchanged, and let pp_sasssign deal with
183 sv = MUTABLE_SV(gv_fetchsv_nomg(sv, GV_ADD, SVt_PVGV));
185 /* FAKE globs in the symbol table cause weird bugs (#77810) */
189 if (SvFAKE(sv) && !(PL_op->op_private & OPpALLOW_FAKE)) {
190 SV *newsv = sv_newmortal();
191 sv_setsv_flags(newsv, sv, 0);
203 sv, PL_op->op_private & OPpDEREF,
204 PL_op->op_private & HINT_STRICT_REFS,
205 ((PL_op->op_flags & OPf_SPECIAL) && !(PL_op->op_flags & OPf_MOD))
206 || PL_op->op_type == OP_READLINE
208 if (PL_op->op_private & OPpLVAL_INTRO)
209 save_gp(MUTABLE_GV(sv), !(PL_op->op_flags & OPf_SPECIAL));
214 /* Helper function for pp_rv2sv and pp_rv2av */
216 Perl_softref2xv(pTHX_ SV *const sv, const char *const what,
217 const svtype type, SV ***spp)
221 PERL_ARGS_ASSERT_SOFTREF2XV;
223 if (PL_op->op_private & HINT_STRICT_REFS) {
225 Perl_die(aTHX_ PL_no_symref_sv, sv,
226 (SvPOKp(sv) && SvCUR(sv)>32 ? "..." : ""), what);
228 Perl_die(aTHX_ PL_no_usym, what);
232 PL_op->op_flags & OPf_REF
234 Perl_die(aTHX_ PL_no_usym, what);
235 if (ckWARN(WARN_UNINITIALIZED))
237 if (type != SVt_PV && GIMME_V == G_ARRAY) {
241 **spp = &PL_sv_undef;
244 if ((PL_op->op_flags & OPf_SPECIAL) &&
245 !(PL_op->op_flags & OPf_MOD))
247 if (!(gv = gv_fetchsv_nomg(sv, GV_ADDMG, type)))
249 **spp = &PL_sv_undef;
254 gv = gv_fetchsv_nomg(sv, GV_ADD, type);
267 sv = amagic_deref_call(sv, to_sv_amg);
271 if (SvTYPE(sv) >= SVt_PVAV)
272 DIE(aTHX_ "Not a SCALAR reference");
277 if (!isGV_with_GP(gv)) {
278 gv = Perl_softref2xv(aTHX_ sv, "a SCALAR", SVt_PV, &sp);
284 if (PL_op->op_flags & OPf_MOD) {
285 if (PL_op->op_private & OPpLVAL_INTRO) {
286 if (cUNOP->op_first->op_type == OP_NULL)
287 sv = save_scalar(MUTABLE_GV(TOPs));
289 sv = save_scalar(gv);
291 Perl_croak(aTHX_ "%s", PL_no_localize_ref);
293 else if (PL_op->op_private & OPpDEREF)
294 sv = vivify_ref(sv, PL_op->op_private & OPpDEREF);
296 SPAGAIN; /* in case chasing soft refs reallocated the stack */
304 AV * const av = MUTABLE_AV(TOPs);
305 const I32 lvalue = PL_op->op_flags & OPf_MOD || LVRET;
307 SV ** const svp = Perl_av_arylen_p(aTHX_ MUTABLE_AV(av));
309 *svp = newSV_type(SVt_PVMG);
310 sv_magic(*svp, MUTABLE_SV(av), PERL_MAGIC_arylen, NULL, 0);
314 SETs(sv_2mortal(newSViv(AvFILL(MUTABLE_AV(av)))));
323 if (PL_op->op_flags & OPf_MOD || LVRET) {
324 SV * const ret = sv_2mortal(newSV_type(SVt_PVLV));/* Not TARG RT#67838 */
325 sv_magic(ret, NULL, PERL_MAGIC_pos, NULL, 0);
327 LvTARG(ret) = SvREFCNT_inc_simple(sv);
328 SETs(ret); /* no SvSETMAGIC */
331 const MAGIC * const mg = mg_find_mglob(sv);
332 if (mg && mg->mg_len != -1) {
333 STRLEN i = mg->mg_len;
334 if (PL_op->op_private & OPpTRUEBOOL)
335 SETs(i ? &PL_sv_yes : &PL_sv_zero);
338 if (mg->mg_flags & MGf_BYTES && DO_UTF8(sv))
339 i = sv_pos_b2u_flags(sv, i, SV_GMAGIC|SV_CONST_RETURN);
354 const I32 flags = (PL_op->op_flags & OPf_SPECIAL)
356 : ((PL_op->op_private & (OPpLVAL_INTRO|OPpMAY_RETURN_CONSTANT))
357 == OPpMAY_RETURN_CONSTANT)
360 /* We usually try to add a non-existent subroutine in case of AUTOLOAD. */
361 /* (But not in defined().) */
363 CV *cv = sv_2cv(TOPs, &stash_unused, &gv, flags);
365 else if ((flags == (GV_ADD|GV_NOEXPAND)) && gv && SvROK(gv)) {
366 cv = SvTYPE(SvRV(gv)) == SVt_PVCV
367 ? MUTABLE_CV(SvRV(gv))
371 cv = MUTABLE_CV(&PL_sv_undef);
372 SETs(MUTABLE_SV(cv));
382 SV *ret = &PL_sv_undef;
384 if (SvGMAGICAL(TOPs)) SETs(sv_mortalcopy(TOPs));
385 if (SvPOK(TOPs) && SvCUR(TOPs) >= 7) {
386 const char * s = SvPVX_const(TOPs);
387 if (strnEQ(s, "CORE::", 6)) {
388 const int code = keyword(s + 6, SvCUR(TOPs) - 6, 1);
390 DIE(aTHX_ "Can't find an opnumber for \"%" UTF8f "\"",
391 UTF8fARG(SvFLAGS(TOPs) & SVf_UTF8, SvCUR(TOPs)-6, s+6));
393 SV * const sv = core_prototype(NULL, s + 6, code, NULL);
399 cv = sv_2cv(TOPs, &stash, &gv, 0);
401 ret = newSVpvn_flags(
402 CvPROTO(cv), CvPROTOLEN(cv), SVs_TEMP | SvUTF8(cv)
412 CV *cv = MUTABLE_CV(PAD_SV(PL_op->op_targ));
414 cv = MUTABLE_CV(sv_2mortal(MUTABLE_SV(cv_clone(cv))));
416 PUSHs(MUTABLE_SV(cv));
430 if (GIMME_V != G_ARRAY) {
436 *MARK = &PL_sv_undef;
438 *MARK = refto(*MARK);
442 EXTEND_MORTAL(SP - MARK);
444 *MARK = refto(*MARK);
449 S_refto(pTHX_ SV *sv)
453 PERL_ARGS_ASSERT_REFTO;
455 if (SvTYPE(sv) == SVt_PVLV && LvTYPE(sv) == 'y') {
458 if (!(sv = LvTARG(sv)))
461 SvREFCNT_inc_void_NN(sv);
463 else if (SvTYPE(sv) == SVt_PVAV) {
464 if (!AvREAL((const AV *)sv) && AvREIFY((const AV *)sv))
465 av_reify(MUTABLE_AV(sv));
467 SvREFCNT_inc_void_NN(sv);
469 else if (SvPADTMP(sv)) {
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 I32 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;
1675 if (GIMME_V == G_ARRAY && PL_op->op_private & OPpREPEAT_DOLIST) {
1676 /* TODO: think of some way of doing list-repeat overloading ??? */
1681 if (UNLIKELY(PL_op->op_private & OPpREPEAT_DOLIST)) {
1682 /* The parser saw this as a list repeat, and there
1683 are probably several items on the stack. But we're
1684 in scalar/void context, and there's no pp_list to save us
1685 now. So drop the rest of the items -- robin@kitsite.com
1688 if (MARK + 1 < SP) {
1694 ASSUME(MARK + 1 == SP);
1696 MARK[1] = &PL_sv_undef;
1700 tryAMAGICbin_MG(repeat_amg, AMGf_assign);
1706 const UV uv = SvUV_nomg(sv);
1708 count = IV_MAX; /* The best we can do? */
1712 count = SvIV_nomg(sv);
1715 else if (SvNOKp(sv)) {
1716 const NV nv = SvNV_nomg(sv);
1717 infnan = Perl_isinfnan(nv);
1718 if (UNLIKELY(infnan)) {
1722 count = -1; /* An arbitrary negative integer */
1728 count = SvIV_nomg(sv);
1731 Perl_ck_warner(aTHX_ packWARN(WARN_NUMERIC),
1732 "Non-finite repeat count does nothing");
1733 } else if (count < 0) {
1735 Perl_ck_warner(aTHX_ packWARN(WARN_NUMERIC),
1736 "Negative repeat count does nothing");
1739 if (GIMME_V == G_ARRAY && PL_op->op_private & OPpREPEAT_DOLIST) {
1741 const SSize_t items = SP - MARK;
1742 const U8 mod = PL_op->op_flags & OPf_MOD;
1747 if ( items > SSize_t_MAX / count /* max would overflow */
1748 /* repeatcpy would overflow */
1749 || items > I32_MAX / (I32)sizeof(SV *)
1751 Perl_croak(aTHX_ "%s","Out of memory during list extend");
1752 max = items * count;
1757 if (mod && SvPADTMP(*SP)) {
1758 *SP = sv_mortalcopy(*SP);
1765 repeatcpy((char*)(MARK + items), (char*)MARK,
1766 items * sizeof(const SV *), count - 1);
1769 else if (count <= 0)
1772 else { /* Note: mark already snarfed by pp_list */
1773 SV * const tmpstr = POPs;
1778 sv_setsv_nomg(TARG, tmpstr);
1779 SvPV_force_nomg(TARG, len);
1780 isutf = DO_UTF8(TARG);
1787 if ( len > (MEM_SIZE_MAX-1) / (UV)count /* max would overflow */
1788 || len > (U32)I32_MAX /* repeatcpy would overflow */
1790 Perl_croak(aTHX_ "%s",
1791 "Out of memory during string extend");
1792 max = (UV)count * len + 1;
1795 repeatcpy(SvPVX(TARG) + len, SvPVX(TARG), len, count - 1);
1796 SvCUR_set(TARG, SvCUR(TARG) * count);
1798 *SvEND(TARG) = '\0';
1801 (void)SvPOK_only_UTF8(TARG);
1803 (void)SvPOK_only(TARG);
1812 dSP; dATARGET; bool useleft; SV *svl, *svr;
1813 tryAMAGICbin_MG(subtr_amg, AMGf_assign|AMGf_numeric);
1817 #ifdef PERL_PRESERVE_IVUV
1819 /* special-case some simple common cases */
1820 if (!((svl->sv_flags|svr->sv_flags) & (SVf_IVisUV|SVs_GMG))) {
1822 U32 flags = (svl->sv_flags & svr->sv_flags);
1823 if (flags & SVf_IOK) {
1824 /* both args are simple IVs */
1829 topl = ((UV)il) >> (UVSIZE * 8 - 2);
1830 topr = ((UV)ir) >> (UVSIZE * 8 - 2);
1832 /* if both are in a range that can't under/overflow, do a
1833 * simple integer subtract: if the top of both numbers
1834 * are 00 or 11, then it's safe */
1835 if (!( ((topl+1) | (topr+1)) & 2)) {
1837 TARGi(il - ir, 0); /* args not GMG, so can't be tainted */
1843 else if (flags & SVf_NOK) {
1844 /* both args are NVs */
1849 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
1850 !Perl_isnan(nl) && nl == (NV)(il = (IV)nl)
1851 && !Perl_isnan(nr) && nr == (NV)(ir = (IV)nr)
1853 nl == (NV)(il = (IV)nl) && nr == (NV)(ir = (IV)nr)
1856 /* nothing was lost by converting to IVs */
1859 TARGn(nl - nr, 0); /* args not GMG, so can't be tainted */
1867 useleft = USE_LEFT(svl);
1868 /* See comments in pp_add (in pp_hot.c) about Overflow, and how
1869 "bad things" happen if you rely on signed integers wrapping. */
1870 if (SvIV_please_nomg(svr)) {
1871 /* Unless the left argument is integer in range we are going to have to
1872 use NV maths. Hence only attempt to coerce the right argument if
1873 we know the left is integer. */
1880 a_valid = auvok = 1;
1881 /* left operand is undef, treat as zero. */
1883 /* Left operand is defined, so is it IV? */
1884 if (SvIV_please_nomg(svl)) {
1885 if ((auvok = SvUOK(svl)))
1888 const IV aiv = SvIVX(svl);
1891 auvok = 1; /* Now acting as a sign flag. */
1892 } else { /* 2s complement assumption for IV_MIN */
1893 auv = (aiv == IV_MIN) ? (UV)aiv : (UV)-aiv;
1900 bool result_good = 0;
1903 bool buvok = SvUOK(svr);
1908 const IV biv = SvIVX(svr);
1913 buv = (biv == IV_MIN) ? (UV)biv : (UV)-biv;
1915 /* ?uvok if value is >= 0. basically, flagged as UV if it's +ve,
1916 else "IV" now, independent of how it came in.
1917 if a, b represents positive, A, B negative, a maps to -A etc
1922 all UV maths. negate result if A negative.
1923 subtract if signs same, add if signs differ. */
1925 if (auvok ^ buvok) {
1934 /* Must get smaller */
1939 if (result <= buv) {
1940 /* result really should be -(auv-buv). as its negation
1941 of true value, need to swap our result flag */
1953 if (result <= (UV)IV_MIN)
1954 SETi(result == (UV)IV_MIN
1955 ? IV_MIN : -(IV)result);
1957 /* result valid, but out of range for IV. */
1958 SETn( -(NV)result );
1962 } /* Overflow, drop through to NVs. */
1966 useleft = USE_LEFT(svl);
1969 NV value = SvNV_nomg(svr);
1973 /* left operand is undef, treat as zero - value */
1977 SETn( SvNV_nomg(svl) - value );
1982 #define IV_BITS (IVSIZE * 8)
1984 static UV S_uv_shift(UV uv, int shift, bool left)
1990 if (shift >= IV_BITS) {
1993 return left ? uv << shift : uv >> shift;
1996 static IV S_iv_shift(IV iv, int shift, bool left)
2002 if (shift >= IV_BITS) {
2003 return iv < 0 && !left ? -1 : 0;
2005 return left ? iv << shift : iv >> shift;
2008 #define UV_LEFT_SHIFT(uv, shift) S_uv_shift(uv, shift, TRUE)
2009 #define UV_RIGHT_SHIFT(uv, shift) S_uv_shift(uv, shift, FALSE)
2010 #define IV_LEFT_SHIFT(iv, shift) S_iv_shift(iv, shift, TRUE)
2011 #define IV_RIGHT_SHIFT(iv, shift) S_iv_shift(iv, shift, FALSE)
2015 dSP; dATARGET; SV *svl, *svr;
2016 tryAMAGICbin_MG(lshift_amg, AMGf_assign|AMGf_numeric);
2020 const IV shift = SvIV_nomg(svr);
2021 if (PL_op->op_private & HINT_INTEGER) {
2022 SETi(IV_LEFT_SHIFT(SvIV_nomg(svl), shift));
2025 SETu(UV_LEFT_SHIFT(SvUV_nomg(svl), shift));
2033 dSP; dATARGET; SV *svl, *svr;
2034 tryAMAGICbin_MG(rshift_amg, AMGf_assign|AMGf_numeric);
2038 const IV shift = SvIV_nomg(svr);
2039 if (PL_op->op_private & HINT_INTEGER) {
2040 SETi(IV_RIGHT_SHIFT(SvIV_nomg(svl), shift));
2043 SETu(UV_RIGHT_SHIFT(SvUV_nomg(svl), shift));
2054 tryAMAGICbin_MG(lt_amg, AMGf_set|AMGf_numeric);
2058 (SvIOK_notUV(left) && SvIOK_notUV(right))
2059 ? (SvIVX(left) < SvIVX(right))
2060 : (do_ncmp(left, right) == -1)
2070 tryAMAGICbin_MG(gt_amg, AMGf_set|AMGf_numeric);
2074 (SvIOK_notUV(left) && SvIOK_notUV(right))
2075 ? (SvIVX(left) > SvIVX(right))
2076 : (do_ncmp(left, right) == 1)
2086 tryAMAGICbin_MG(le_amg, AMGf_set|AMGf_numeric);
2090 (SvIOK_notUV(left) && SvIOK_notUV(right))
2091 ? (SvIVX(left) <= SvIVX(right))
2092 : (do_ncmp(left, right) <= 0)
2102 tryAMAGICbin_MG(ge_amg, AMGf_set|AMGf_numeric);
2106 (SvIOK_notUV(left) && SvIOK_notUV(right))
2107 ? (SvIVX(left) >= SvIVX(right))
2108 : ( (do_ncmp(left, right) & 2) == 0)
2118 tryAMAGICbin_MG(ne_amg, AMGf_set|AMGf_numeric);
2122 (SvIOK_notUV(left) && SvIOK_notUV(right))
2123 ? (SvIVX(left) != SvIVX(right))
2124 : (do_ncmp(left, right) != 0)
2129 /* compare left and right SVs. Returns:
2133 * 2: left or right was a NaN
2136 Perl_do_ncmp(pTHX_ SV* const left, SV * const right)
2138 PERL_ARGS_ASSERT_DO_NCMP;
2139 #ifdef PERL_PRESERVE_IVUV
2140 /* Fortunately it seems NaN isn't IOK */
2141 if (SvIV_please_nomg(right) && SvIV_please_nomg(left)) {
2143 const IV leftiv = SvIVX(left);
2144 if (!SvUOK(right)) {
2145 /* ## IV <=> IV ## */
2146 const IV rightiv = SvIVX(right);
2147 return (leftiv > rightiv) - (leftiv < rightiv);
2149 /* ## IV <=> UV ## */
2151 /* As (b) is a UV, it's >=0, so it must be < */
2154 const UV rightuv = SvUVX(right);
2155 return ((UV)leftiv > rightuv) - ((UV)leftiv < rightuv);
2160 /* ## UV <=> UV ## */
2161 const UV leftuv = SvUVX(left);
2162 const UV rightuv = SvUVX(right);
2163 return (leftuv > rightuv) - (leftuv < rightuv);
2165 /* ## UV <=> IV ## */
2167 const IV rightiv = SvIVX(right);
2169 /* As (a) is a UV, it's >=0, so it cannot be < */
2172 const UV leftuv = SvUVX(left);
2173 return (leftuv > (UV)rightiv) - (leftuv < (UV)rightiv);
2176 NOT_REACHED; /* NOTREACHED */
2180 NV const rnv = SvNV_nomg(right);
2181 NV const lnv = SvNV_nomg(left);
2183 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2184 if (Perl_isnan(lnv) || Perl_isnan(rnv)) {
2187 return (lnv > rnv) - (lnv < rnv);
2206 tryAMAGICbin_MG(ncmp_amg, AMGf_numeric);
2209 value = do_ncmp(left, right);
2221 /* also used for: pp_sge() pp_sgt() pp_slt() */
2227 int amg_type = sle_amg;
2231 switch (PL_op->op_type) {
2250 tryAMAGICbin_MG(amg_type, AMGf_set);
2254 #ifdef USE_LOCALE_COLLATE
2255 (IN_LC_RUNTIME(LC_COLLATE))
2256 ? sv_cmp_locale_flags(left, right, 0)
2259 sv_cmp_flags(left, right, 0);
2260 SETs(boolSV(cmp * multiplier < rhs));
2268 tryAMAGICbin_MG(seq_amg, AMGf_set);
2271 SETs(boolSV(sv_eq_flags(left, right, 0)));
2279 tryAMAGICbin_MG(sne_amg, AMGf_set);
2282 SETs(boolSV(!sv_eq_flags(left, right, 0)));
2290 tryAMAGICbin_MG(scmp_amg, 0);
2294 #ifdef USE_LOCALE_COLLATE
2295 (IN_LC_RUNTIME(LC_COLLATE))
2296 ? sv_cmp_locale_flags(left, right, 0)
2299 sv_cmp_flags(left, right, 0);
2308 tryAMAGICbin_MG(band_amg, AMGf_assign);
2311 if (SvNIOKp(left) || SvNIOKp(right)) {
2312 const bool left_ro_nonnum = !SvNIOKp(left) && SvREADONLY(left);
2313 const bool right_ro_nonnum = !SvNIOKp(right) && SvREADONLY(right);
2314 if (PL_op->op_private & HINT_INTEGER) {
2315 const IV i = SvIV_nomg(left) & SvIV_nomg(right);
2319 const UV u = SvUV_nomg(left) & SvUV_nomg(right);
2322 if (left_ro_nonnum && left != TARG) SvNIOK_off(left);
2323 if (right_ro_nonnum) SvNIOK_off(right);
2326 do_vop(PL_op->op_type, TARG, left, right);
2336 tryAMAGICbin_MG(band_amg, AMGf_assign|AMGf_numarg);
2338 dATARGET; dPOPTOPssrl;
2339 if (PL_op->op_private & HINT_INTEGER) {
2340 const IV i = SvIV_nomg(left) & SvIV_nomg(right);
2344 const UV u = SvUV_nomg(left) & SvUV_nomg(right);
2354 tryAMAGICbin_MG(sband_amg, AMGf_assign);
2356 dATARGET; dPOPTOPssrl;
2357 do_vop(OP_BIT_AND, TARG, left, right);
2362 /* also used for: pp_bit_xor() */
2367 const int op_type = PL_op->op_type;
2369 tryAMAGICbin_MG((op_type == OP_BIT_OR ? bor_amg : bxor_amg), AMGf_assign);
2372 if (SvNIOKp(left) || SvNIOKp(right)) {
2373 const bool left_ro_nonnum = !SvNIOKp(left) && SvREADONLY(left);
2374 const bool right_ro_nonnum = !SvNIOKp(right) && SvREADONLY(right);
2375 if (PL_op->op_private & HINT_INTEGER) {
2376 const IV l = (USE_LEFT(left) ? SvIV_nomg(left) : 0);
2377 const IV r = SvIV_nomg(right);
2378 const IV result = op_type == OP_BIT_OR ? (l | r) : (l ^ r);
2382 const UV l = (USE_LEFT(left) ? SvUV_nomg(left) : 0);
2383 const UV r = SvUV_nomg(right);
2384 const UV result = op_type == OP_BIT_OR ? (l | r) : (l ^ r);
2387 if (left_ro_nonnum && left != TARG) SvNIOK_off(left);
2388 if (right_ro_nonnum) SvNIOK_off(right);
2391 do_vop(op_type, TARG, left, right);
2398 /* also used for: pp_nbit_xor() */
2403 const int op_type = PL_op->op_type;
2405 tryAMAGICbin_MG((op_type == OP_NBIT_OR ? bor_amg : bxor_amg),
2406 AMGf_assign|AMGf_numarg);
2408 dATARGET; dPOPTOPssrl;
2409 if (PL_op->op_private & HINT_INTEGER) {
2410 const IV l = (USE_LEFT(left) ? SvIV_nomg(left) : 0);
2411 const IV r = SvIV_nomg(right);
2412 const IV result = op_type == OP_NBIT_OR ? (l | r) : (l ^ r);
2416 const UV l = (USE_LEFT(left) ? SvUV_nomg(left) : 0);
2417 const UV r = SvUV_nomg(right);
2418 const UV result = op_type == OP_NBIT_OR ? (l | r) : (l ^ r);
2425 /* also used for: pp_sbit_xor() */
2430 const int op_type = PL_op->op_type;
2432 tryAMAGICbin_MG((op_type == OP_SBIT_OR ? sbor_amg : sbxor_amg),
2435 dATARGET; dPOPTOPssrl;
2436 do_vop(op_type == OP_SBIT_OR ? OP_BIT_OR : OP_BIT_XOR, TARG, left,
2442 PERL_STATIC_INLINE bool
2443 S_negate_string(pTHX)
2448 SV * const sv = TOPs;
2449 if (!SvPOKp(sv) || SvNIOK(sv) || (!SvPOK(sv) && SvNIOKp(sv)))
2451 s = SvPV_nomg_const(sv, len);
2452 if (isIDFIRST(*s)) {
2453 sv_setpvs(TARG, "-");
2456 else if (*s == '+' || (*s == '-' && !looks_like_number(sv))) {
2457 sv_setsv_nomg(TARG, sv);
2458 *SvPV_force_nomg(TARG, len) = *s == '-' ? '+' : '-';
2468 tryAMAGICun_MG(neg_amg, AMGf_numeric);
2469 if (S_negate_string(aTHX)) return NORMAL;
2471 SV * const sv = TOPs;
2474 /* It's publicly an integer */
2477 if (SvIVX(sv) == IV_MIN) {
2478 /* 2s complement assumption. */
2479 SETi(SvIVX(sv)); /* special case: -((UV)IV_MAX+1) ==
2483 else if (SvUVX(sv) <= IV_MAX) {
2488 else if (SvIVX(sv) != IV_MIN) {
2492 #ifdef PERL_PRESERVE_IVUV
2499 if (SvNIOKp(sv) && (SvNIOK(sv) || !SvPOK(sv)))
2500 SETn(-SvNV_nomg(sv));
2501 else if (SvPOKp(sv) && SvIV_please_nomg(sv))
2502 goto oops_its_an_int;
2504 SETn(-SvNV_nomg(sv));
2514 tryAMAGICun_MG(not_amg, AMGf_set);
2516 *PL_stack_sp = boolSV(!SvTRUE_nomg_NN(sv));
2521 S_scomplement(pTHX_ SV *targ, SV *sv)
2527 sv_copypv_nomg(TARG, sv);
2528 tmps = (U8*)SvPV_nomg(TARG, len);
2531 if (len && ! utf8_to_bytes(tmps, &len)) {
2532 Perl_croak(aTHX_ fatal_above_ff_msg, PL_op_desc[PL_op->op_type]);
2543 for ( ; anum && (unsigned long)tmps % sizeof(long); anum--, tmps++)
2546 for ( ; anum >= (I32)sizeof(long); anum -= (I32)sizeof(long), tmpl++)
2551 for ( ; anum > 0; anum--, tmps++)
2558 tryAMAGICun_MG(compl_amg, AMGf_numeric);
2562 if (PL_op->op_private & HINT_INTEGER) {
2563 const IV i = ~SvIV_nomg(sv);
2567 const UV u = ~SvUV_nomg(sv);
2572 S_scomplement(aTHX_ TARG, sv);
2582 tryAMAGICun_MG(compl_amg, AMGf_numeric|AMGf_numarg);
2585 if (PL_op->op_private & HINT_INTEGER) {
2586 const IV i = ~SvIV_nomg(sv);
2590 const UV u = ~SvUV_nomg(sv);
2600 tryAMAGICun_MG(scompl_amg, AMGf_numeric);
2603 S_scomplement(aTHX_ TARG, sv);
2609 /* integer versions of some of the above */
2614 tryAMAGICbin_MG(mult_amg, AMGf_assign);
2617 SETi( left * right );
2626 tryAMAGICbin_MG(div_amg, AMGf_assign);
2629 IV value = SvIV_nomg(right);
2631 DIE(aTHX_ "Illegal division by zero");
2632 num = SvIV_nomg(left);
2634 /* avoid FPE_INTOVF on some platforms when num is IV_MIN */
2638 value = num / value;
2646 /* This is the vanilla old i_modulo. */
2648 tryAMAGICbin_MG(modulo_amg, AMGf_assign);
2652 DIE(aTHX_ "Illegal modulus zero");
2653 /* avoid FPE_INTOVF on some platforms when left is IV_MIN */
2657 SETi( left % right );
2662 #if defined(__GLIBC__) && IVSIZE == 8 \
2663 && ( __GLIBC__ < 2 || (__GLIBC__ == 2 && __GLIBC_MINOR__ < 8))
2665 PP(pp_i_modulo_glibc_bugfix)
2667 /* This is the i_modulo with the workaround for the _moddi3 bug
2668 * in (at least) glibc 2.2.5 (the PERL_ABS() the workaround).
2669 * See below for pp_i_modulo. */
2671 tryAMAGICbin_MG(modulo_amg, AMGf_assign);
2675 DIE(aTHX_ "Illegal modulus zero");
2676 /* avoid FPE_INTOVF on some platforms when left is IV_MIN */
2680 SETi( left % PERL_ABS(right) );
2689 tryAMAGICbin_MG(add_amg, AMGf_assign);
2691 dPOPTOPiirl_ul_nomg;
2692 SETi( left + right );
2700 tryAMAGICbin_MG(subtr_amg, AMGf_assign);
2702 dPOPTOPiirl_ul_nomg;
2703 SETi( left - right );
2711 tryAMAGICbin_MG(lt_amg, AMGf_set);
2714 SETs(boolSV(left < right));
2722 tryAMAGICbin_MG(gt_amg, AMGf_set);
2725 SETs(boolSV(left > right));
2733 tryAMAGICbin_MG(le_amg, AMGf_set);
2736 SETs(boolSV(left <= right));
2744 tryAMAGICbin_MG(ge_amg, AMGf_set);
2747 SETs(boolSV(left >= right));
2755 tryAMAGICbin_MG(eq_amg, AMGf_set);
2758 SETs(boolSV(left == right));
2766 tryAMAGICbin_MG(ne_amg, AMGf_set);
2769 SETs(boolSV(left != right));
2777 tryAMAGICbin_MG(ncmp_amg, 0);
2784 else if (left < right)
2796 tryAMAGICun_MG(neg_amg, 0);
2797 if (S_negate_string(aTHX)) return NORMAL;
2799 SV * const sv = TOPs;
2800 IV const i = SvIV_nomg(sv);
2806 /* High falutin' math. */
2811 tryAMAGICbin_MG(atan2_amg, 0);
2814 SETn(Perl_atan2(left, right));
2820 /* also used for: pp_cos() pp_exp() pp_log() pp_sqrt() */
2825 int amg_type = fallback_amg;
2826 const char *neg_report = NULL;
2827 const int op_type = PL_op->op_type;
2830 case OP_SIN: amg_type = sin_amg; break;
2831 case OP_COS: amg_type = cos_amg; break;
2832 case OP_EXP: amg_type = exp_amg; break;
2833 case OP_LOG: amg_type = log_amg; neg_report = "log"; break;
2834 case OP_SQRT: amg_type = sqrt_amg; neg_report = "sqrt"; break;
2837 assert(amg_type != fallback_amg);
2839 tryAMAGICun_MG(amg_type, 0);
2841 SV * const arg = TOPs;
2842 const NV value = SvNV_nomg(arg);
2848 if (neg_report) { /* log or sqrt */
2850 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2851 ! Perl_isnan(value) &&
2853 (op_type == OP_LOG ? (value <= 0.0) : (value < 0.0))) {
2854 SET_NUMERIC_STANDARD();
2855 /* diag_listed_as: Can't take log of %g */
2856 DIE(aTHX_ "Can't take %s of %" NVgf, neg_report, value);
2861 case OP_SIN: result = Perl_sin(value); break;
2862 case OP_COS: result = Perl_cos(value); break;
2863 case OP_EXP: result = Perl_exp(value); break;
2864 case OP_LOG: result = Perl_log(value); break;
2865 case OP_SQRT: result = Perl_sqrt(value); break;
2872 /* Support Configure command-line overrides for rand() functions.
2873 After 5.005, perhaps we should replace this by Configure support
2874 for drand48(), random(), or rand(). For 5.005, though, maintain
2875 compatibility by calling rand() but allow the user to override it.
2876 See INSTALL for details. --Andy Dougherty 15 July 1998
2878 /* Now it's after 5.005, and Configure supports drand48() and random(),
2879 in addition to rand(). So the overrides should not be needed any more.
2880 --Jarkko Hietaniemi 27 September 1998
2885 if (!PL_srand_called) {
2886 (void)seedDrand01((Rand_seed_t)seed());
2887 PL_srand_called = TRUE;
2899 SV * const sv = POPs;
2905 /* 1 of 2 things can be carried through SvNV, SP or TARG, SP was carried */
2906 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2907 if (! Perl_isnan(value) && value == 0.0)
2917 sv_setnv_mg(TARG, value);
2928 if (MAXARG >= 1 && (TOPs || POPs)) {
2935 pv = SvPV(top, len);
2936 flags = grok_number(pv, len, &anum);
2938 if (!(flags & IS_NUMBER_IN_UV)) {
2939 Perl_ck_warner_d(aTHX_ packWARN(WARN_OVERFLOW),
2940 "Integer overflow in srand");
2948 (void)seedDrand01((Rand_seed_t)anum);
2949 PL_srand_called = TRUE;
2953 /* Historically srand always returned true. We can avoid breaking
2955 sv_setpvs(TARG, "0 but true");
2964 tryAMAGICun_MG(int_amg, AMGf_numeric);
2966 SV * const sv = TOPs;
2967 const IV iv = SvIV_nomg(sv);
2968 /* XXX it's arguable that compiler casting to IV might be subtly
2969 different from modf (for numbers inside (IV_MIN,UV_MAX)) in which
2970 else preferring IV has introduced a subtle behaviour change bug. OTOH
2971 relying on floating point to be accurate is a bug. */
2976 else if (SvIOK(sv)) {
2978 SETu(SvUV_nomg(sv));
2983 const NV value = SvNV_nomg(sv);
2984 if (UNLIKELY(Perl_isinfnan(value)))
2986 else if (value >= 0.0) {
2987 if (value < (NV)UV_MAX + 0.5) {
2990 SETn(Perl_floor(value));
2994 if (value > (NV)IV_MIN - 0.5) {
2997 SETn(Perl_ceil(value));
3008 tryAMAGICun_MG(abs_amg, AMGf_numeric);
3010 SV * const sv = TOPs;
3011 /* This will cache the NV value if string isn't actually integer */
3012 const IV iv = SvIV_nomg(sv);
3017 else if (SvIOK(sv)) {
3018 /* IVX is precise */
3020 SETu(SvUV_nomg(sv)); /* force it to be numeric only */
3028 /* 2s complement assumption. Also, not really needed as
3029 IV_MIN and -IV_MIN should both be %100...00 and NV-able */
3035 const NV value = SvNV_nomg(sv);
3046 /* also used for: pp_hex() */
3052 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES;
3056 SV* const sv = TOPs;
3058 tmps = (SvPV_const(sv, len));
3060 /* If Unicode, try to downgrade
3061 * If not possible, croak. */
3062 SV* const tsv = sv_2mortal(newSVsv(sv));
3065 sv_utf8_downgrade(tsv, FALSE);
3066 tmps = SvPV_const(tsv, len);
3068 if (PL_op->op_type == OP_HEX)
3071 while (*tmps && len && isSPACE(*tmps))
3075 if (isALPHA_FOLD_EQ(*tmps, 'x')) {
3077 result_uv = grok_hex (tmps, &len, &flags, &result_nv);
3079 else if (isALPHA_FOLD_EQ(*tmps, 'b'))
3080 result_uv = grok_bin (tmps, &len, &flags, &result_nv);
3082 result_uv = grok_oct (tmps, &len, &flags, &result_nv);
3084 if (flags & PERL_SCAN_GREATER_THAN_UV_MAX) {
3099 SV * const sv = TOPs;
3101 U32 in_bytes = IN_BYTES;
3102 /* Simplest case shortcut:
3103 * set svflags to just the SVf_POK|SVs_GMG|SVf_UTF8 from the SV,
3104 * with the SVf_UTF8 flag inverted if under 'use bytes' (HINT_BYTES
3107 U32 svflags = (SvFLAGS(sv) ^ (in_bytes << 26)) & (SVf_POK|SVs_GMG|SVf_UTF8);
3109 STATIC_ASSERT_STMT(SVf_UTF8 == (HINT_BYTES << 26));
3112 if (LIKELY(svflags == SVf_POK))
3115 if (svflags & SVs_GMG)
3120 if (!IN_BYTES) { /* reread to avoid using an C auto/register */
3121 if ((SvFLAGS(sv) & (SVf_POK|SVf_UTF8)) == SVf_POK)
3123 if ( SvPOK(sv) && (PL_op->op_private & OPpTRUEBOOL)) {
3124 /* no need to convert from bytes to chars */
3128 len = sv_len_utf8_nomg(sv);
3131 /* unrolled SvPV_nomg_const(sv,len) */
3132 if (SvPOK_nog(sv)) {
3135 if (PL_op->op_private & OPpTRUEBOOL) {
3137 SETs(len ? &PL_sv_yes : &PL_sv_zero);
3142 (void)sv_2pv_flags(sv, &len, 0|SV_CONST_RETURN);
3145 TARGi((IV)(len), 1);
3148 if (!SvPADTMP(TARG)) {
3149 /* OPpTARGET_MY: targ is var in '$lex = length()' */
3154 /* TARG is on stack at this point and is overwriten by SETs.
3155 * This branch is the odd one out, so put TARG by default on
3156 * stack earlier to let local SP go out of liveness sooner */
3159 return NORMAL; /* no putback, SP didn't move in this opcode */
3163 /* Returns false if substring is completely outside original string.
3164 No length is indicated by len_iv = 0 and len_is_uv = 0. len_is_uv must
3165 always be true for an explicit 0.
3168 Perl_translate_substr_offsets( STRLEN curlen, IV pos1_iv,
3169 bool pos1_is_uv, IV len_iv,
3170 bool len_is_uv, STRLEN *posp,
3176 PERL_ARGS_ASSERT_TRANSLATE_SUBSTR_OFFSETS;
3178 if (!pos1_is_uv && pos1_iv < 0 && curlen) {
3179 pos1_is_uv = curlen-1 > ~(UV)pos1_iv;
3182 if ((pos1_is_uv || pos1_iv > 0) && (UV)pos1_iv > curlen)
3185 if (len_iv || len_is_uv) {
3186 if (!len_is_uv && len_iv < 0) {
3187 pos2_iv = curlen + len_iv;
3189 pos2_is_uv = curlen-1 > ~(UV)len_iv;
3192 } else { /* len_iv >= 0 */
3193 if (!pos1_is_uv && pos1_iv < 0) {
3194 pos2_iv = pos1_iv + len_iv;
3195 pos2_is_uv = (UV)len_iv > (UV)IV_MAX;
3197 if ((UV)len_iv > curlen-(UV)pos1_iv)
3200 pos2_iv = pos1_iv+len_iv;
3210 if (!pos2_is_uv && pos2_iv < 0) {
3211 if (!pos1_is_uv && pos1_iv < 0)
3215 else if (!pos1_is_uv && pos1_iv < 0)
3218 if ((UV)pos2_iv < (UV)pos1_iv)
3220 if ((UV)pos2_iv > curlen)
3223 /* pos1_iv and pos2_iv both in 0..curlen, so the cast is safe */
3224 *posp = (STRLEN)( (UV)pos1_iv );
3225 *lenp = (STRLEN)( (UV)pos2_iv - (UV)pos1_iv );
3242 I32 lvalue = PL_op->op_flags & OPf_MOD || LVRET;
3243 const bool rvalue = (GIMME_V != G_VOID);
3246 const char *repl = NULL;
3248 int num_args = PL_op->op_private & 7;
3249 bool repl_need_utf8_upgrade = FALSE;
3253 if(!(repl_sv = POPs)) num_args--;
3255 if ((len_sv = POPs)) {
3256 len_iv = SvIV(len_sv);
3257 len_is_uv = len_iv ? SvIOK_UV(len_sv) : 1;
3262 pos1_iv = SvIV(pos_sv);
3263 pos1_is_uv = SvIOK_UV(pos_sv);
3265 if (PL_op->op_private & OPpSUBSTR_REPL_FIRST) {
3269 if (lvalue && !repl_sv) {
3271 ret = sv_2mortal(newSV_type(SVt_PVLV)); /* Not TARG RT#67838 */
3272 sv_magic(ret, NULL, PERL_MAGIC_substr, NULL, 0);
3274 LvTARG(ret) = SvREFCNT_inc_simple(sv);
3276 pos1_is_uv || pos1_iv >= 0
3277 ? (STRLEN)(UV)pos1_iv
3278 : (LvFLAGS(ret) |= LVf_NEG_OFF, (STRLEN)(UV)-pos1_iv);
3280 len_is_uv || len_iv > 0
3281 ? (STRLEN)(UV)len_iv
3282 : (LvFLAGS(ret) |= LVf_NEG_LEN, (STRLEN)(UV)-len_iv);
3284 PUSHs(ret); /* avoid SvSETMAGIC here */
3288 repl = SvPV_const(repl_sv, repl_len);
3291 Perl_ck_warner(aTHX_ packWARN(WARN_SUBSTR),
3292 "Attempt to use reference as lvalue in substr"
3294 tmps = SvPV_force_nomg(sv, curlen);
3295 if (DO_UTF8(repl_sv) && repl_len) {
3297 /* Upgrade the dest, and recalculate tmps in case the buffer
3298 * got reallocated; curlen may also have been changed */
3299 sv_utf8_upgrade_nomg(sv);
3300 tmps = SvPV_nomg(sv, curlen);
3303 else if (DO_UTF8(sv))
3304 repl_need_utf8_upgrade = TRUE;
3306 else tmps = SvPV_const(sv, curlen);
3308 utf8_curlen = sv_or_pv_len_utf8(sv, tmps, curlen);
3309 if (utf8_curlen == curlen)
3312 curlen = utf8_curlen;
3318 STRLEN pos, len, byte_len, byte_pos;
3320 if (!translate_substr_offsets(
3321 curlen, pos1_iv, pos1_is_uv, len_iv, len_is_uv, &pos, &len
3325 byte_pos = utf8_curlen
3326 ? sv_or_pv_pos_u2b(sv, tmps, pos, &byte_len) : pos;
3331 SvTAINTED_off(TARG); /* decontaminate */
3332 SvUTF8_off(TARG); /* decontaminate */
3333 sv_setpvn(TARG, tmps, byte_len);
3334 #ifdef USE_LOCALE_COLLATE
3335 sv_unmagic(TARG, PERL_MAGIC_collxfrm);
3342 SV* repl_sv_copy = NULL;
3344 if (repl_need_utf8_upgrade) {
3345 repl_sv_copy = newSVsv(repl_sv);
3346 sv_utf8_upgrade(repl_sv_copy);
3347 repl = SvPV_const(repl_sv_copy, repl_len);
3351 sv_insert_flags(sv, byte_pos, byte_len, repl, repl_len, 0);
3352 SvREFCNT_dec(repl_sv_copy);
3355 if (PL_op->op_private & OPpSUBSTR_REPL_FIRST)
3365 Perl_croak(aTHX_ "substr outside of string");
3366 Perl_ck_warner(aTHX_ packWARN(WARN_SUBSTR), "substr outside of string");
3373 const IV size = POPi;
3374 SV* offsetsv = POPs;
3375 SV * const src = POPs;
3376 const I32 lvalue = PL_op->op_flags & OPf_MOD || LVRET;
3382 /* extract a STRLEN-ranged integer value from offsetsv into offset,
3383 * or flag that its out of range */
3385 IV iv = SvIV(offsetsv);
3387 /* avoid a large UV being wrapped to a negative value */
3388 if (SvIOK_UV(offsetsv) && SvUVX(offsetsv) > (UV)IV_MAX)
3389 errflags = LVf_OUT_OF_RANGE;
3391 errflags = (LVf_NEG_OFF|LVf_OUT_OF_RANGE);
3392 #if PTRSIZE < IVSIZE
3393 else if (iv > Size_t_MAX)
3394 errflags = LVf_OUT_OF_RANGE;
3397 offset = (STRLEN)iv;
3400 retuv = errflags ? 0 : do_vecget(src, offset, size);
3402 if (lvalue) { /* it's an lvalue! */
3403 ret = sv_2mortal(newSV_type(SVt_PVLV)); /* Not TARG RT#67838 */
3404 sv_magic(ret, NULL, PERL_MAGIC_vec, NULL, 0);
3406 LvTARG(ret) = SvREFCNT_inc_simple(src);
3407 LvTARGOFF(ret) = offset;
3408 LvTARGLEN(ret) = size;
3409 LvFLAGS(ret) = errflags;
3413 SvTAINTED_off(TARG); /* decontaminate */
3417 sv_setuv(ret, retuv);
3425 /* also used for: pp_rindex() */
3438 const char *little_p;
3441 const bool is_index = PL_op->op_type == OP_INDEX;
3442 const bool threeargs = MAXARG >= 3 && (TOPs || ((void)POPs,0));
3448 big_p = SvPV_const(big, biglen);
3449 little_p = SvPV_const(little, llen);
3451 big_utf8 = DO_UTF8(big);
3452 little_utf8 = DO_UTF8(little);
3453 if (big_utf8 ^ little_utf8) {
3454 /* One needs to be upgraded. */
3456 /* Well, maybe instead we might be able to downgrade the small
3458 char * const pv = (char*)bytes_from_utf8((U8 *)little_p, &llen,
3461 /* If the large string is ISO-8859-1, and it's not possible to
3462 convert the small string to ISO-8859-1, then there is no
3463 way that it could be found anywhere by index. */
3468 /* At this point, pv is a malloc()ed string. So donate it to temp
3469 to ensure it will get free()d */
3470 little = temp = newSV(0);
3471 sv_usepvn(temp, pv, llen);
3472 little_p = SvPVX(little);
3474 temp = newSVpvn(little_p, llen);
3476 sv_utf8_upgrade(temp);
3478 little_p = SvPV_const(little, llen);
3481 if (SvGAMAGIC(big)) {
3482 /* Life just becomes a lot easier if I use a temporary here.
3483 Otherwise I need to avoid calls to sv_pos_u2b(), which (dangerously)
3484 will trigger magic and overloading again, as will fbm_instr()
3486 big = newSVpvn_flags(big_p, biglen,
3487 SVs_TEMP | (big_utf8 ? SVf_UTF8 : 0));
3490 if (SvGAMAGIC(little) || (is_index && !SvOK(little))) {
3491 /* index && SvOK() is a hack. fbm_instr() calls SvPV_const, which will
3492 warn on undef, and we've already triggered a warning with the
3493 SvPV_const some lines above. We can't remove that, as we need to
3494 call some SvPV to trigger overloading early and find out if the
3496 This is all getting too messy. The API isn't quite clean enough,
3497 because data access has side effects.
3499 little = newSVpvn_flags(little_p, llen,
3500 SVs_TEMP | (little_utf8 ? SVf_UTF8 : 0));
3501 little_p = SvPVX(little);
3505 offset = is_index ? 0 : biglen;
3507 if (big_utf8 && offset > 0)
3508 offset = sv_pos_u2b_flags(big, offset, 0, SV_CONST_RETURN);
3514 else if (offset > (SSize_t)biglen)
3516 if (!(little_p = is_index
3517 ? fbm_instr((unsigned char*)big_p + offset,
3518 (unsigned char*)big_p + biglen, little, 0)
3519 : rninstr(big_p, big_p + offset,
3520 little_p, little_p + llen)))
3523 retval = little_p - big_p;
3524 if (retval > 1 && big_utf8)
3525 retval = sv_pos_b2u_flags(big, retval, SV_CONST_RETURN);
3530 /* OPpTRUEBOOL indicates an '== -1' has been optimised away */
3531 if (PL_op->op_private & OPpTRUEBOOL) {
3532 PUSHs( ((retval != -1) ^ cBOOL(PL_op->op_private & OPpINDEX_BOOLNEG))
3533 ? &PL_sv_yes : &PL_sv_no);
3534 if (PL_op->op_private & OPpTARGET_MY)
3535 /* $lex = (index() == -1) */
3536 sv_setsv(TARG, TOPs);
3545 dSP; dMARK; dORIGMARK; dTARGET;
3546 SvTAINTED_off(TARG);
3547 do_sprintf(TARG, SP-MARK, MARK+1);
3548 TAINT_IF(SvTAINTED(TARG));
3560 const U8 *s = (U8*)SvPV_const(argsv, len);
3563 ? (len ? utf8n_to_uvchr(s, len, 0, UTF8_ALLOW_ANYUV) : 0)
3577 if (UNLIKELY(SvAMAGIC(top)))
3579 if (UNLIKELY(isinfnansv(top)))
3580 Perl_croak(aTHX_ "Cannot chr %" NVgf, SvNV(top));
3582 if (!IN_BYTES /* under bytes, chr(-1) eq chr(0xff), etc. */
3583 && ((SvIOKp(top) && !SvIsUV(top) && SvIV_nomg(top) < 0)
3585 ((SvNOKp(top) || (SvOK(top) && !SvIsUV(top)))
3586 && SvNV_nomg(top) < 0.0)))
3588 if (ckWARN(WARN_UTF8)) {
3589 if (SvGMAGICAL(top)) {
3590 SV *top2 = sv_newmortal();
3591 sv_setsv_nomg(top2, top);
3594 Perl_warner(aTHX_ packWARN(WARN_UTF8),
3595 "Invalid negative number (%" SVf ") in chr", SVfARG(top));
3597 value = UNICODE_REPLACEMENT;
3599 value = SvUV_nomg(top);
3603 SvUPGRADE(TARG,SVt_PV);
3605 if (value > 255 && !IN_BYTES) {
3606 SvGROW(TARG, (STRLEN)UVCHR_SKIP(value)+1);
3607 tmps = (char*)uvchr_to_utf8_flags((U8*)SvPVX(TARG), value, 0);
3608 SvCUR_set(TARG, tmps - SvPVX_const(TARG));
3610 (void)SvPOK_only(TARG);
3619 *tmps++ = (char)value;
3621 (void)SvPOK_only(TARG);
3633 const char *tmps = SvPV_const(left, len);
3635 if (DO_UTF8(left)) {
3636 /* If Unicode, try to downgrade.
3637 * If not possible, croak.
3638 * Yes, we made this up. */
3639 SV* const tsv = newSVpvn_flags(tmps, len, SVf_UTF8|SVs_TEMP);
3641 sv_utf8_downgrade(tsv, FALSE);
3642 tmps = SvPV_const(tsv, len);
3644 # ifdef USE_ITHREADS
3646 if (!PL_reentrant_buffer->_crypt_struct_buffer) {
3647 /* This should be threadsafe because in ithreads there is only
3648 * one thread per interpreter. If this would not be true,
3649 * we would need a mutex to protect this malloc. */
3650 PL_reentrant_buffer->_crypt_struct_buffer =
3651 (struct crypt_data *)safemalloc(sizeof(struct crypt_data));
3652 #if defined(__GLIBC__) || defined(__EMX__)
3653 if (PL_reentrant_buffer->_crypt_struct_buffer) {
3654 PL_reentrant_buffer->_crypt_struct_buffer->initialized = 0;
3655 /* work around glibc-2.2.5 bug */
3656 PL_reentrant_buffer->_crypt_struct_buffer->current_saltbits = 0;
3660 # endif /* HAS_CRYPT_R */
3661 # endif /* USE_ITHREADS */
3663 sv_setpv(TARG, fcrypt(tmps, SvPV_nolen_const(right)));
3665 sv_setpv(TARG, PerlProc_crypt(tmps, SvPV_nolen_const(right)));
3672 "The crypt() function is unimplemented due to excessive paranoia.");
3676 /* Generally UTF-8 and UTF-EBCDIC are indistinguishable at this level. So
3677 * most comments below say UTF-8, when in fact they mean UTF-EBCDIC as well */
3680 /* also used for: pp_lcfirst() */
3684 /* Actually is both lcfirst() and ucfirst(). Only the first character
3685 * changes. This means that possibly we can change in-place, ie., just
3686 * take the source and change that one character and store it back, but not
3687 * if read-only etc, or if the length changes */
3691 STRLEN slen; /* slen is the byte length of the whole SV. */
3694 bool inplace; /* ? Convert first char only, in-place */
3695 bool doing_utf8 = FALSE; /* ? using utf8 */
3696 bool convert_source_to_utf8 = FALSE; /* ? need to convert */
3697 const int op_type = PL_op->op_type;
3700 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
3701 STRLEN ulen; /* ulen is the byte length of the original Unicode character
3702 * stored as UTF-8 at s. */
3703 STRLEN tculen; /* tculen is the byte length of the freshly titlecased (or
3704 * lowercased) character stored in tmpbuf. May be either
3705 * UTF-8 or not, but in either case is the number of bytes */
3707 s = (const U8*)SvPV_const(source, slen);
3709 /* We may be able to get away with changing only the first character, in
3710 * place, but not if read-only, etc. Later we may discover more reasons to
3711 * not convert in-place. */
3712 inplace = !SvREADONLY(source) && SvPADTMP(source);
3714 /* First calculate what the changed first character should be. This affects
3715 * whether we can just swap it out, leaving the rest of the string unchanged,
3716 * or even if have to convert the dest to UTF-8 when the source isn't */
3718 if (! slen) { /* If empty */
3719 need = 1; /* still need a trailing NUL */
3722 else if (DO_UTF8(source)) { /* Is the source utf8? */
3725 if (op_type == OP_UCFIRST) {
3726 #ifdef USE_LOCALE_CTYPE
3727 _toTITLE_utf8_flags(s, s +slen, tmpbuf, &tculen, IN_LC_RUNTIME(LC_CTYPE));
3729 _toTITLE_utf8_flags(s, s +slen, tmpbuf, &tculen, 0);
3733 #ifdef USE_LOCALE_CTYPE
3734 _toLOWER_utf8_flags(s, s + slen, tmpbuf, &tculen, IN_LC_RUNTIME(LC_CTYPE));
3736 _toLOWER_utf8_flags(s, s + slen, tmpbuf, &tculen, 0);
3740 /* we can't do in-place if the length changes. */
3741 if (ulen != tculen) inplace = FALSE;
3742 need = slen + 1 - ulen + tculen;
3744 else { /* Non-zero length, non-UTF-8, Need to consider locale and if
3745 * latin1 is treated as caseless. Note that a locale takes
3747 ulen = 1; /* Original character is 1 byte */
3748 tculen = 1; /* Most characters will require one byte, but this will
3749 * need to be overridden for the tricky ones */
3752 if (op_type == OP_LCFIRST) {
3754 /* lower case the first letter: no trickiness for any character */
3755 #ifdef USE_LOCALE_CTYPE
3756 if (IN_LC_RUNTIME(LC_CTYPE)) {
3757 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
3758 *tmpbuf = toLOWER_LC(*s);
3763 *tmpbuf = (IN_UNI_8_BIT)
3764 ? toLOWER_LATIN1(*s)
3768 #ifdef USE_LOCALE_CTYPE
3770 else if (IN_LC_RUNTIME(LC_CTYPE)) {
3771 if (IN_UTF8_CTYPE_LOCALE) {
3775 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
3776 *tmpbuf = (U8) toUPPER_LC(*s); /* This would be a bug if any
3777 locales have upper and title case
3781 else if (! IN_UNI_8_BIT) {
3782 *tmpbuf = toUPPER(*s); /* Returns caseless for non-ascii, or
3783 * on EBCDIC machines whatever the
3784 * native function does */
3787 /* Here, is ucfirst non-UTF-8, not in locale (unless that locale is
3788 * UTF-8, which we treat as not in locale), and cased latin1 */
3790 #ifdef USE_LOCALE_CTYPE
3794 title_ord = _to_upper_title_latin1(*s, tmpbuf, &tculen, 's');
3796 assert(tculen == 2);
3798 /* If the result is an upper Latin1-range character, it can
3799 * still be represented in one byte, which is its ordinal */
3800 if (UTF8_IS_DOWNGRADEABLE_START(*tmpbuf)) {
3801 *tmpbuf = (U8) title_ord;
3805 /* Otherwise it became more than one ASCII character (in
3806 * the case of LATIN_SMALL_LETTER_SHARP_S) or changed to
3807 * beyond Latin1, so the number of bytes changed, so can't
3808 * replace just the first character in place. */
3811 /* If the result won't fit in a byte, the entire result
3812 * will have to be in UTF-8. Assume worst case sizing in
3813 * conversion. (all latin1 characters occupy at most two
3815 if (title_ord > 255) {
3817 convert_source_to_utf8 = TRUE;
3818 need = slen * 2 + 1;
3820 /* The (converted) UTF-8 and UTF-EBCDIC lengths of all
3821 * (both) characters whose title case is above 255 is
3825 else { /* LATIN_SMALL_LETTER_SHARP_S expands by 1 byte */
3826 need = slen + 1 + 1;
3830 } /* End of use Unicode (Latin1) semantics */
3831 } /* End of changing the case of the first character */
3833 /* Here, have the first character's changed case stored in tmpbuf. Ready to
3834 * generate the result */
3837 /* We can convert in place. This means we change just the first
3838 * character without disturbing the rest; no need to grow */
3840 s = d = (U8*)SvPV_force_nomg(source, slen);
3846 /* Here, we can't convert in place; we earlier calculated how much
3847 * space we will need, so grow to accommodate that */
3848 SvUPGRADE(dest, SVt_PV);
3849 d = (U8*)SvGROW(dest, need);
3850 (void)SvPOK_only(dest);
3857 if (! convert_source_to_utf8) {
3859 /* Here both source and dest are in UTF-8, but have to create
3860 * the entire output. We initialize the result to be the
3861 * title/lower cased first character, and then append the rest
3863 sv_setpvn(dest, (char*)tmpbuf, tculen);
3865 sv_catpvn(dest, (char*)(s + ulen), slen - ulen);
3869 const U8 *const send = s + slen;
3871 /* Here the dest needs to be in UTF-8, but the source isn't,
3872 * except we earlier UTF-8'd the first character of the source
3873 * into tmpbuf. First put that into dest, and then append the
3874 * rest of the source, converting it to UTF-8 as we go. */
3876 /* Assert tculen is 2 here because the only two characters that
3877 * get to this part of the code have 2-byte UTF-8 equivalents */
3879 *d++ = *(tmpbuf + 1);
3880 s++; /* We have just processed the 1st char */
3882 for (; s < send; s++) {
3883 d = uvchr_to_utf8(d, *s);
3886 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
3890 else { /* in-place UTF-8. Just overwrite the first character */
3891 Copy(tmpbuf, d, tculen, U8);
3892 SvCUR_set(dest, need - 1);
3896 else { /* Neither source nor dest are in or need to be UTF-8 */
3898 if (inplace) { /* in-place, only need to change the 1st char */
3901 else { /* Not in-place */
3903 /* Copy the case-changed character(s) from tmpbuf */
3904 Copy(tmpbuf, d, tculen, U8);
3905 d += tculen - 1; /* Code below expects d to point to final
3906 * character stored */
3909 else { /* empty source */
3910 /* See bug #39028: Don't taint if empty */
3914 /* In a "use bytes" we don't treat the source as UTF-8, but, still want
3915 * the destination to retain that flag */
3916 if (SvUTF8(source) && ! IN_BYTES)
3919 if (!inplace) { /* Finish the rest of the string, unchanged */
3920 /* This will copy the trailing NUL */
3921 Copy(s + 1, d + 1, slen, U8);
3922 SvCUR_set(dest, need - 1);
3925 #ifdef USE_LOCALE_CTYPE
3926 if (IN_LC_RUNTIME(LC_CTYPE)) {
3931 if (dest != source && SvTAINTED(source))
3937 /* There's so much setup/teardown code common between uc and lc, I wonder if
3938 it would be worth merging the two, and just having a switch outside each
3939 of the three tight loops. There is less and less commonality though */
3952 if ( SvPADTMP(source)
3953 && !SvREADONLY(source) && SvPOK(source)
3956 #ifdef USE_LOCALE_CTYPE
3957 (IN_LC_RUNTIME(LC_CTYPE))
3958 ? ! IN_UTF8_CTYPE_LOCALE
3964 /* We can convert in place. The reason we can't if in UNI_8_BIT is to
3965 * make the loop tight, so we overwrite the source with the dest before
3966 * looking at it, and we need to look at the original source
3967 * afterwards. There would also need to be code added to handle
3968 * switching to not in-place in midstream if we run into characters
3969 * that change the length. Since being in locale overrides UNI_8_BIT,
3970 * that latter becomes irrelevant in the above test; instead for
3971 * locale, the size can't normally change, except if the locale is a
3974 s = d = (U8*)SvPV_force_nomg(source, len);
3981 s = (const U8*)SvPV_nomg_const(source, len);
3984 SvUPGRADE(dest, SVt_PV);
3985 d = (U8*)SvGROW(dest, min);
3986 (void)SvPOK_only(dest);
3991 /* Overloaded values may have toggled the UTF-8 flag on source, so we need
3992 to check DO_UTF8 again here. */
3994 if (DO_UTF8(source)) {
3995 const U8 *const send = s + len;
3996 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
3998 /* All occurrences of these are to be moved to follow any other marks.
3999 * This is context-dependent. We may not be passed enough context to
4000 * move the iota subscript beyond all of them, but we do the best we can
4001 * with what we're given. The result is always better than if we
4002 * hadn't done this. And, the problem would only arise if we are
4003 * passed a character without all its combining marks, which would be
4004 * the caller's mistake. The information this is based on comes from a
4005 * comment in Unicode SpecialCasing.txt, (and the Standard's text
4006 * itself) and so can't be checked properly to see if it ever gets
4007 * revised. But the likelihood of it changing is remote */
4008 bool in_iota_subscript = FALSE;
4014 if (in_iota_subscript && ! _is_utf8_mark(s)) {
4016 /* A non-mark. Time to output the iota subscript */
4017 Copy(GREEK_CAPITAL_LETTER_IOTA_UTF8, d, capital_iota_len, U8);
4018 d += capital_iota_len;
4019 in_iota_subscript = FALSE;
4022 /* Then handle the current character. Get the changed case value
4023 * and copy it to the output buffer */
4026 #ifdef USE_LOCALE_CTYPE
4027 uv = _toUPPER_utf8_flags(s, send, tmpbuf, &ulen, IN_LC_RUNTIME(LC_CTYPE));
4029 uv = _toUPPER_utf8_flags(s, send, tmpbuf, &ulen, 0);
4031 #define GREEK_CAPITAL_LETTER_IOTA 0x0399
4032 #define COMBINING_GREEK_YPOGEGRAMMENI 0x0345
4033 if (uv == GREEK_CAPITAL_LETTER_IOTA
4034 && utf8_to_uvchr_buf(s, send, 0) == COMBINING_GREEK_YPOGEGRAMMENI)
4036 in_iota_subscript = TRUE;
4039 if (ulen > u && (SvLEN(dest) < (min += ulen - u))) {
4040 /* If the eventually required minimum size outgrows the
4041 * available space, we need to grow. */
4042 const UV o = d - (U8*)SvPVX_const(dest);
4044 /* If someone uppercases one million U+03B0s we SvGROW()
4045 * one million times. Or we could try guessing how much to
4046 * allocate without allocating too much. Such is life.
4047 * See corresponding comment in lc code for another option
4049 d = o + (U8*) SvGROW(dest, min);
4051 Copy(tmpbuf, d, ulen, U8);
4056 if (in_iota_subscript) {
4057 Copy(GREEK_CAPITAL_LETTER_IOTA_UTF8, d, capital_iota_len, U8);
4058 d += capital_iota_len;
4063 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
4065 else { /* Not UTF-8 */
4067 const U8 *const send = s + len;
4069 /* Use locale casing if in locale; regular style if not treating
4070 * latin1 as having case; otherwise the latin1 casing. Do the
4071 * whole thing in a tight loop, for speed, */
4072 #ifdef USE_LOCALE_CTYPE
4073 if (IN_LC_RUNTIME(LC_CTYPE)) {
4074 if (IN_UTF8_CTYPE_LOCALE) {
4077 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
4078 for (; s < send; d++, s++)
4079 *d = (U8) toUPPER_LC(*s);
4083 if (! IN_UNI_8_BIT) {
4084 for (; s < send; d++, s++) {
4089 #ifdef USE_LOCALE_CTYPE
4092 for (; s < send; d++, s++) {
4093 *d = toUPPER_LATIN1_MOD(*s);
4094 if (LIKELY(*d != LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
4098 /* The mainstream case is the tight loop above. To avoid
4099 * extra tests in that, all three characters that require
4100 * special handling are mapped by the MOD to the one tested
4102 * Use the source to distinguish between the three cases */
4104 #if UNICODE_MAJOR_VERSION > 2 \
4105 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
4106 && UNICODE_DOT_DOT_VERSION >= 8)
4107 if (*s == LATIN_SMALL_LETTER_SHARP_S) {
4109 /* uc() of this requires 2 characters, but they are
4110 * ASCII. If not enough room, grow the string */
4111 if (SvLEN(dest) < ++min) {
4112 const UV o = d - (U8*)SvPVX_const(dest);
4113 d = o + (U8*) SvGROW(dest, min);
4115 *d++ = 'S'; *d = 'S'; /* upper case is 'SS' */
4116 continue; /* Back to the tight loop; still in ASCII */
4120 /* The other two special handling characters have their
4121 * upper cases outside the latin1 range, hence need to be
4122 * in UTF-8, so the whole result needs to be in UTF-8. So,
4123 * here we are somewhere in the middle of processing a
4124 * non-UTF-8 string, and realize that we will have to convert
4125 * the whole thing to UTF-8. What to do? There are
4126 * several possibilities. The simplest to code is to
4127 * convert what we have so far, set a flag, and continue on
4128 * in the loop. The flag would be tested each time through
4129 * the loop, and if set, the next character would be
4130 * converted to UTF-8 and stored. But, I (khw) didn't want
4131 * to slow down the mainstream case at all for this fairly
4132 * rare case, so I didn't want to add a test that didn't
4133 * absolutely have to be there in the loop, besides the
4134 * possibility that it would get too complicated for
4135 * optimizers to deal with. Another possibility is to just
4136 * give up, convert the source to UTF-8, and restart the
4137 * function that way. Another possibility is to convert
4138 * both what has already been processed and what is yet to
4139 * come separately to UTF-8, then jump into the loop that
4140 * handles UTF-8. But the most efficient time-wise of the
4141 * ones I could think of is what follows, and turned out to
4142 * not require much extra code. */
4144 /* Convert what we have so far into UTF-8, telling the
4145 * function that we know it should be converted, and to
4146 * allow extra space for what we haven't processed yet.
4147 * Assume the worst case space requirements for converting
4148 * what we haven't processed so far: that it will require
4149 * two bytes for each remaining source character, plus the
4150 * NUL at the end. This may cause the string pointer to
4151 * move, so re-find it. */
4153 len = d - (U8*)SvPVX_const(dest);
4154 SvCUR_set(dest, len);
4155 len = sv_utf8_upgrade_flags_grow(dest,
4156 SV_GMAGIC|SV_FORCE_UTF8_UPGRADE,
4158 d = (U8*)SvPVX(dest) + len;
4160 /* Now process the remainder of the source, converting to
4161 * upper and UTF-8. If a resulting byte is invariant in
4162 * UTF-8, output it as-is, otherwise convert to UTF-8 and
4163 * append it to the output. */
4164 for (; s < send; s++) {
4165 (void) _to_upper_title_latin1(*s, d, &len, 'S');
4169 /* Here have processed the whole source; no need to continue
4170 * with the outer loop. Each character has been converted
4171 * to upper case and converted to UTF-8 */
4174 } /* End of processing all latin1-style chars */
4175 } /* End of processing all chars */
4176 } /* End of source is not empty */
4178 if (source != dest) {
4179 *d = '\0'; /* Here d points to 1 after last char, add NUL */
4180 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
4182 } /* End of isn't utf8 */
4183 #ifdef USE_LOCALE_CTYPE
4184 if (IN_LC_RUNTIME(LC_CTYPE)) {
4189 if (dest != source && SvTAINTED(source))
4207 if ( SvPADTMP(source)
4208 && !SvREADONLY(source) && SvPOK(source)
4209 && !DO_UTF8(source)) {
4211 /* We can convert in place, as lowercasing anything in the latin1 range
4212 * (or else DO_UTF8 would have been on) doesn't lengthen it */
4214 s = d = (U8*)SvPV_force_nomg(source, len);
4221 s = (const U8*)SvPV_nomg_const(source, len);
4224 SvUPGRADE(dest, SVt_PV);
4225 d = (U8*)SvGROW(dest, min);
4226 (void)SvPOK_only(dest);
4231 /* Overloaded values may have toggled the UTF-8 flag on source, so we need
4232 to check DO_UTF8 again here. */
4234 if (DO_UTF8(source)) {
4235 const U8 *const send = s + len;
4236 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
4239 const STRLEN u = UTF8SKIP(s);
4242 #ifdef USE_LOCALE_CTYPE
4243 _toLOWER_utf8_flags(s, send, tmpbuf, &ulen, IN_LC_RUNTIME(LC_CTYPE));
4245 _toLOWER_utf8_flags(s, send, tmpbuf, &ulen, 0);
4248 /* Here is where we would do context-sensitive actions. See the
4249 * commit message for 86510fb15 for why there isn't any */
4251 if (ulen > u && (SvLEN(dest) < (min += ulen - u))) {
4253 /* If the eventually required minimum size outgrows the
4254 * available space, we need to grow. */
4255 const UV o = d - (U8*)SvPVX_const(dest);
4257 /* If someone lowercases one million U+0130s we SvGROW() one
4258 * million times. Or we could try guessing how much to
4259 * allocate without allocating too much. Such is life.
4260 * Another option would be to grow an extra byte or two more
4261 * each time we need to grow, which would cut down the million
4262 * to 500K, with little waste */
4263 d = o + (U8*) SvGROW(dest, min);
4266 /* Copy the newly lowercased letter to the output buffer we're
4268 Copy(tmpbuf, d, ulen, U8);
4271 } /* End of looping through the source string */
4274 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
4275 } else { /* Not utf8 */
4277 const U8 *const send = s + len;
4279 /* Use locale casing if in locale; regular style if not treating
4280 * latin1 as having case; otherwise the latin1 casing. Do the
4281 * whole thing in a tight loop, for speed, */
4282 #ifdef USE_LOCALE_CTYPE
4283 if (IN_LC_RUNTIME(LC_CTYPE)) {
4284 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
4285 for (; s < send; d++, s++)
4286 *d = toLOWER_LC(*s);
4290 if (! IN_UNI_8_BIT) {
4291 for (; s < send; d++, s++) {
4296 for (; s < send; d++, s++) {
4297 *d = toLOWER_LATIN1(*s);
4301 if (source != dest) {
4303 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
4306 #ifdef USE_LOCALE_CTYPE
4307 if (IN_LC_RUNTIME(LC_CTYPE)) {
4312 if (dest != source && SvTAINTED(source))
4321 SV * const sv = TOPs;
4323 const char *s = SvPV_const(sv,len);
4325 SvUTF8_off(TARG); /* decontaminate */
4328 SvUPGRADE(TARG, SVt_PV);
4329 SvGROW(TARG, (len * 2) + 1);
4333 STRLEN ulen = UTF8SKIP(s);
4334 bool to_quote = FALSE;
4336 if (UTF8_IS_INVARIANT(*s)) {
4337 if (_isQUOTEMETA(*s)) {
4341 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, s + len)) {
4343 #ifdef USE_LOCALE_CTYPE
4344 /* In locale, we quote all non-ASCII Latin1 chars.
4345 * Otherwise use the quoting rules */
4347 IN_LC_RUNTIME(LC_CTYPE)
4350 _isQUOTEMETA(EIGHT_BIT_UTF8_TO_NATIVE(*s, *(s + 1))))
4355 else if (is_QUOTEMETA_high(s)) {
4370 else if (IN_UNI_8_BIT) {
4372 if (_isQUOTEMETA(*s))
4378 /* For non UNI_8_BIT (and hence in locale) just quote all \W
4379 * including everything above ASCII */
4381 if (!isWORDCHAR_A(*s))
4387 SvCUR_set(TARG, d - SvPVX_const(TARG));
4388 (void)SvPOK_only_UTF8(TARG);
4391 sv_setpvn(TARG, s, len);
4407 U8 tmpbuf[UTF8_MAXBYTES_CASE + 1];
4408 #if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
4409 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
4410 || UNICODE_DOT_DOT_VERSION > 0)
4411 const bool full_folding = TRUE; /* This variable is here so we can easily
4412 move to more generality later */
4414 const bool full_folding = FALSE;
4416 const U8 flags = ( full_folding ? FOLD_FLAGS_FULL : 0 )
4417 #ifdef USE_LOCALE_CTYPE
4418 | ( IN_LC_RUNTIME(LC_CTYPE) ? FOLD_FLAGS_LOCALE : 0 )
4422 /* This is a facsimile of pp_lc, but with a thousand bugs thanks to me.
4423 * You are welcome(?) -Hugmeir
4431 s = (const U8*)SvPV_nomg_const(source, len);
4433 if (ckWARN(WARN_UNINITIALIZED))
4434 report_uninit(source);
4441 SvUPGRADE(dest, SVt_PV);
4442 d = (U8*)SvGROW(dest, min);
4443 (void)SvPOK_only(dest);
4448 if (DO_UTF8(source)) { /* UTF-8 flagged string. */
4450 const STRLEN u = UTF8SKIP(s);
4453 _toFOLD_utf8_flags(s, send, tmpbuf, &ulen, flags);
4455 if (ulen > u && (SvLEN(dest) < (min += ulen - u))) {
4456 const UV o = d - (U8*)SvPVX_const(dest);
4457 d = o + (U8*) SvGROW(dest, min);
4460 Copy(tmpbuf, d, ulen, U8);
4465 } /* Unflagged string */
4467 #ifdef USE_LOCALE_CTYPE
4468 if ( IN_LC_RUNTIME(LC_CTYPE) ) { /* Under locale */
4469 if (IN_UTF8_CTYPE_LOCALE) {
4470 goto do_uni_folding;
4472 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
4473 for (; s < send; d++, s++)
4474 *d = (U8) toFOLD_LC(*s);
4478 if ( !IN_UNI_8_BIT ) { /* Under nothing, or bytes */
4479 for (; s < send; d++, s++)
4483 #ifdef USE_LOCALE_CTYPE
4486 /* For ASCII and the Latin-1 range, there's only two troublesome
4487 * folds, \x{DF} (\N{LATIN SMALL LETTER SHARP S}), which under full
4488 * casefolding becomes 'ss'; and \x{B5} (\N{MICRO SIGN}), which
4489 * under any fold becomes \x{3BC} (\N{GREEK SMALL LETTER MU}) --
4490 * For the rest, the casefold is their lowercase. */
4491 for (; s < send; d++, s++) {
4492 if (*s == MICRO_SIGN) {
4493 /* \N{MICRO SIGN}'s casefold is \N{GREEK SMALL LETTER MU},
4494 * which is outside of the latin-1 range. There's a couple
4495 * of ways to deal with this -- khw discusses them in
4496 * pp_lc/uc, so go there :) What we do here is upgrade what
4497 * we had already casefolded, then enter an inner loop that
4498 * appends the rest of the characters as UTF-8. */
4499 len = d - (U8*)SvPVX_const(dest);
4500 SvCUR_set(dest, len);
4501 len = sv_utf8_upgrade_flags_grow(dest,
4502 SV_GMAGIC|SV_FORCE_UTF8_UPGRADE,
4503 /* The max expansion for latin1
4504 * chars is 1 byte becomes 2 */
4506 d = (U8*)SvPVX(dest) + len;
4508 Copy(GREEK_SMALL_LETTER_MU_UTF8, d, small_mu_len, U8);
4511 for (; s < send; s++) {
4513 UV fc = _to_uni_fold_flags(*s, tmpbuf, &ulen, flags);
4514 if UVCHR_IS_INVARIANT(fc) {
4516 && *s == LATIN_SMALL_LETTER_SHARP_S)
4525 Copy(tmpbuf, d, ulen, U8);
4531 else if (full_folding && *s == LATIN_SMALL_LETTER_SHARP_S) {
4532 /* Under full casefolding, LATIN SMALL LETTER SHARP S
4533 * becomes "ss", which may require growing the SV. */
4534 if (SvLEN(dest) < ++min) {
4535 const UV o = d - (U8*)SvPVX_const(dest);
4536 d = o + (U8*) SvGROW(dest, min);
4541 else { /* If it's not one of those two, the fold is their lower
4543 *d = toLOWER_LATIN1(*s);
4549 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
4551 #ifdef USE_LOCALE_CTYPE
4552 if (IN_LC_RUNTIME(LC_CTYPE)) {
4557 if (SvTAINTED(source))
4567 dSP; dMARK; dORIGMARK;
4568 AV *const av = MUTABLE_AV(POPs);
4569 const I32 lval = (PL_op->op_flags & OPf_MOD || LVRET);
4571 if (SvTYPE(av) == SVt_PVAV) {
4572 const bool localizing = PL_op->op_private & OPpLVAL_INTRO;
4573 bool can_preserve = FALSE;
4579 can_preserve = SvCANEXISTDELETE(av);
4582 if (lval && localizing) {
4585 for (svp = MARK + 1; svp <= SP; svp++) {
4586 const SSize_t elem = SvIV(*svp);
4590 if (max > AvMAX(av))
4594 while (++MARK <= SP) {
4596 SSize_t elem = SvIV(*MARK);
4597 bool preeminent = TRUE;
4599 if (localizing && can_preserve) {
4600 /* If we can determine whether the element exist,
4601 * Try to preserve the existenceness of a tied array
4602 * element by using EXISTS and DELETE if possible.
4603 * Fallback to FETCH and STORE otherwise. */
4604 preeminent = av_exists(av, elem);
4607 svp = av_fetch(av, elem, lval);
4610 DIE(aTHX_ PL_no_aelem, elem);
4613 save_aelem(av, elem, svp);
4615 SAVEADELETE(av, elem);
4618 *MARK = svp ? *svp : &PL_sv_undef;
4621 if (GIMME_V != G_ARRAY) {
4623 *++MARK = SP > ORIGMARK ? *SP : &PL_sv_undef;
4632 AV *const av = MUTABLE_AV(POPs);
4633 I32 lval = (PL_op->op_flags & OPf_MOD);
4634 SSize_t items = SP - MARK;
4636 if (PL_op->op_private & OPpMAYBE_LVSUB) {
4637 const I32 flags = is_lvalue_sub();
4639 if (!(flags & OPpENTERSUB_INARGS))
4640 /* diag_listed_as: Can't modify %s in %s */
4641 Perl_croak(aTHX_ "Can't modify index/value array slice in list assignment");
4648 *(MARK+items*2-1) = *(MARK+items);
4654 while (++MARK <= SP) {
4657 svp = av_fetch(av, SvIV(*MARK), lval);
4659 if (!svp || !*svp || *svp == &PL_sv_undef) {
4660 DIE(aTHX_ PL_no_aelem, SvIV(*MARK));
4662 *MARK = sv_mortalcopy(*MARK);
4664 *++MARK = svp ? *svp : &PL_sv_undef;
4666 if (GIMME_V != G_ARRAY) {
4667 MARK = SP - items*2;
4668 *++MARK = items > 0 ? *SP : &PL_sv_undef;
4678 AV *array = MUTABLE_AV(POPs);
4679 const U8 gimme = GIMME_V;
4680 IV *iterp = Perl_av_iter_p(aTHX_ array);
4681 const IV current = (*iterp)++;
4683 if (current > av_tindex(array)) {
4685 if (gimme == G_SCALAR)
4693 if (gimme == G_ARRAY) {
4694 SV **const element = av_fetch(array, current, 0);
4695 PUSHs(element ? *element : &PL_sv_undef);
4700 /* also used for: pp_avalues()*/
4704 AV *array = MUTABLE_AV(POPs);
4705 const U8 gimme = GIMME_V;
4707 *Perl_av_iter_p(aTHX_ array) = 0;
4709 if (gimme == G_SCALAR) {
4711 PUSHi(av_tindex(array) + 1);
4713 else if (gimme == G_ARRAY) {
4714 if (UNLIKELY(PL_op->op_private & OPpMAYBE_LVSUB)) {
4715 const I32 flags = is_lvalue_sub();
4716 if (flags && !(flags & OPpENTERSUB_INARGS))
4717 /* diag_listed_as: Can't modify %s in %s */
4719 "Can't modify keys on array in list assignment");
4722 IV n = Perl_av_len(aTHX_ array);
4727 if ( PL_op->op_type == OP_AKEYS
4728 || ( PL_op->op_type == OP_AVHVSWITCH
4729 && (PL_op->op_private & 3) + OP_AEACH == OP_AKEYS ))
4731 for (i = 0; i <= n; i++) {
4736 for (i = 0; i <= n; i++) {
4737 SV *const *const elem = Perl_av_fetch(aTHX_ array, i, 0);
4738 PUSHs(elem ? *elem : &PL_sv_undef);
4746 /* Associative arrays. */
4751 HV * hash = MUTABLE_HV(POPs);
4753 const U8 gimme = GIMME_V;
4755 entry = hv_iternext(hash);
4759 SV* const sv = hv_iterkeysv(entry);
4761 if (gimme == G_ARRAY) {
4763 val = hv_iterval(hash, entry);
4767 else if (gimme == G_SCALAR)
4774 S_do_delete_local(pTHX)
4777 const U8 gimme = GIMME_V;
4780 const bool sliced = !!(PL_op->op_private & OPpSLICE);
4781 SV **unsliced_keysv = sliced ? NULL : sp--;
4782 SV * const osv = POPs;
4783 SV **mark = sliced ? PL_stack_base + POPMARK : unsliced_keysv-1;
4785 const bool tied = SvRMAGICAL(osv)
4786 && mg_find((const SV *)osv, PERL_MAGIC_tied);
4787 const bool can_preserve = SvCANEXISTDELETE(osv);
4788 const U32 type = SvTYPE(osv);
4789 SV ** const end = sliced ? SP : unsliced_keysv;
4791 if (type == SVt_PVHV) { /* hash element */
4792 HV * const hv = MUTABLE_HV(osv);
4793 while (++MARK <= end) {
4794 SV * const keysv = *MARK;
4796 bool preeminent = TRUE;
4798 preeminent = hv_exists_ent(hv, keysv, 0);
4800 HE *he = hv_fetch_ent(hv, keysv, 1, 0);
4807 sv = hv_delete_ent(hv, keysv, 0, 0);
4809 SvREFCNT_inc_simple_void(sv); /* De-mortalize */
4812 if (!sv) DIE(aTHX_ PL_no_helem_sv, SVfARG(keysv));
4813 save_helem_flags(hv, keysv, &sv, SAVEf_KEEPOLDELEM);
4815 *MARK = sv_mortalcopy(sv);
4821 SAVEHDELETE(hv, keysv);
4822 *MARK = &PL_sv_undef;
4826 else if (type == SVt_PVAV) { /* array element */
4827 if (PL_op->op_flags & OPf_SPECIAL) {
4828 AV * const av = MUTABLE_AV(osv);
4829 while (++MARK <= end) {
4830 SSize_t idx = SvIV(*MARK);
4832 bool preeminent = TRUE;
4834 preeminent = av_exists(av, idx);
4836 SV **svp = av_fetch(av, idx, 1);
4843 sv = av_delete(av, idx, 0);
4845 SvREFCNT_inc_simple_void(sv); /* De-mortalize */
4848 save_aelem_flags(av, idx, &sv, SAVEf_KEEPOLDELEM);
4850 *MARK = sv_mortalcopy(sv);
4856 SAVEADELETE(av, idx);
4857 *MARK = &PL_sv_undef;
4862 DIE(aTHX_ "panic: avhv_delete no longer supported");
4865 DIE(aTHX_ "Not a HASH reference");
4867 if (gimme == G_VOID)
4869 else if (gimme == G_SCALAR) {
4874 *++MARK = &PL_sv_undef;
4878 else if (gimme != G_VOID)
4879 PUSHs(*unsliced_keysv);
4890 if (PL_op->op_private & OPpLVAL_INTRO)
4891 return do_delete_local();
4894 discard = (gimme == G_VOID) ? G_DISCARD : 0;
4896 if (PL_op->op_private & (OPpSLICE|OPpKVSLICE)) {
4898 HV * const hv = MUTABLE_HV(POPs);
4899 const U32 hvtype = SvTYPE(hv);
4901 if (PL_op->op_private & OPpKVSLICE) {
4902 SSize_t items = SP - MARK;
4906 *(MARK+items*2-1) = *(MARK+items);
4913 if (hvtype == SVt_PVHV) { /* hash element */
4914 while ((MARK += (1+skip)) <= SP) {
4915 SV * const sv = hv_delete_ent(hv, *(MARK-skip), discard, 0);
4916 *MARK = sv ? sv : &PL_sv_undef;
4919 else if (hvtype == SVt_PVAV) { /* array element */
4920 if (PL_op->op_flags & OPf_SPECIAL) {
4921 while ((MARK += (1+skip)) <= SP) {
4922 SV * const sv = av_delete(MUTABLE_AV(hv), SvIV(*(MARK-skip)), discard);
4923 *MARK = sv ? sv : &PL_sv_undef;
4928 DIE(aTHX_ "Not a HASH reference");
4931 else if (gimme == G_SCALAR) {
4936 *++MARK = &PL_sv_undef;
4942 HV * const hv = MUTABLE_HV(POPs);
4944 if (SvTYPE(hv) == SVt_PVHV)
4945 sv = hv_delete_ent(hv, keysv, discard, 0);
4946 else if (SvTYPE(hv) == SVt_PVAV) {
4947 if (PL_op->op_flags & OPf_SPECIAL)
4948 sv = av_delete(MUTABLE_AV(hv), SvIV(keysv), discard);
4950 DIE(aTHX_ "panic: avhv_delete no longer supported");
4953 DIE(aTHX_ "Not a HASH reference");
4968 if (UNLIKELY( PL_op->op_private & OPpEXISTS_SUB )) {
4970 SV * const sv = POPs;
4971 CV * const cv = sv_2cv(sv, &hv, &gv, 0);
4974 if (gv && isGV(gv) && GvCV(gv) && !GvCVGEN(gv))
4979 hv = MUTABLE_HV(POPs);
4980 if (LIKELY( SvTYPE(hv) == SVt_PVHV )) {
4981 if (hv_exists_ent(hv, tmpsv, 0))
4984 else if (SvTYPE(hv) == SVt_PVAV) {
4985 if (PL_op->op_flags & OPf_SPECIAL) { /* array element */
4986 if (av_exists(MUTABLE_AV(hv), SvIV(tmpsv)))
4991 DIE(aTHX_ "Not a HASH reference");
4998 dSP; dMARK; dORIGMARK;
4999 HV * const hv = MUTABLE_HV(POPs);
5000 const I32 lval = (PL_op->op_flags & OPf_MOD || LVRET);
5001 const bool localizing = PL_op->op_private & OPpLVAL_INTRO;
5002 bool can_preserve = FALSE;
5008 if (SvCANEXISTDELETE(hv))
5009 can_preserve = TRUE;
5012 while (++MARK <= SP) {
5013 SV * const keysv = *MARK;
5016 bool preeminent = TRUE;
5018 if (localizing && can_preserve) {
5019 /* If we can determine whether the element exist,
5020 * try to preserve the existenceness of a tied hash
5021 * element by using EXISTS and DELETE if possible.
5022 * Fallback to FETCH and STORE otherwise. */
5023 preeminent = hv_exists_ent(hv, keysv, 0);
5026 he = hv_fetch_ent(hv, keysv, lval, 0);
5027 svp = he ? &HeVAL(he) : NULL;
5030 if (!svp || !*svp || *svp == &PL_sv_undef) {
5031 DIE(aTHX_ PL_no_helem_sv, SVfARG(keysv));
5034 if (HvNAME_get(hv) && isGV_or_RVCV(*svp))
5035 save_gp(MUTABLE_GV(*svp), !(PL_op->op_flags & OPf_SPECIAL));
5036 else if (preeminent)
5037 save_helem_flags(hv, keysv, svp,
5038 (PL_op->op_flags & OPf_SPECIAL) ? 0 : SAVEf_SETMAGIC);
5040 SAVEHDELETE(hv, keysv);
5043 *MARK = svp && *svp ? *svp : &PL_sv_undef;
5045 if (GIMME_V != G_ARRAY) {
5047 *++MARK = SP > ORIGMARK ? *SP : &PL_sv_undef;
5056 HV * const hv = MUTABLE_HV(POPs);
5057 I32 lval = (PL_op->op_flags & OPf_MOD);
5058 SSize_t items = SP - MARK;
5060 if (PL_op->op_private & OPpMAYBE_LVSUB) {
5061 const I32 flags = is_lvalue_sub();
5063 if (!(flags & OPpENTERSUB_INARGS))
5064 /* diag_listed_as: Can't modify %s in %s */
5065 Perl_croak(aTHX_ "Can't modify key/value hash slice in %s assignment",
5066 GIMME_V == G_ARRAY ? "list" : "scalar");
5073 *(MARK+items*2-1) = *(MARK+items);
5079 while (++MARK <= SP) {
5080 SV * const keysv = *MARK;
5084 he = hv_fetch_ent(hv, keysv, lval, 0);
5085 svp = he ? &HeVAL(he) : NULL;
5088 if (!svp || !*svp || *svp == &PL_sv_undef) {
5089 DIE(aTHX_ PL_no_helem_sv, SVfARG(keysv));
5091 *MARK = sv_mortalcopy(*MARK);
5093 *++MARK = svp && *svp ? *svp : &PL_sv_undef;
5095 if (GIMME_V != G_ARRAY) {
5096 MARK = SP - items*2;
5097 *++MARK = items > 0 ? *SP : &PL_sv_undef;
5103 /* List operators. */
5107 I32 markidx = POPMARK;
5108 if (GIMME_V != G_ARRAY) {
5109 SV **mark = PL_stack_base + markidx;
5111 EXTEND(SP, 1); /* in case no arguments, as in @empty */
5113 *MARK = *SP; /* unwanted list, return last item */
5115 *MARK = &PL_sv_undef;
5125 SV ** const lastrelem = PL_stack_sp;
5126 SV ** const lastlelem = PL_stack_base + POPMARK;
5127 SV ** const firstlelem = PL_stack_base + POPMARK + 1;
5128 SV ** const firstrelem = lastlelem + 1;
5129 const U8 mod = PL_op->op_flags & OPf_MOD;
5131 const I32 max = lastrelem - lastlelem;
5134 if (GIMME_V != G_ARRAY) {
5135 if (lastlelem < firstlelem) {
5137 *firstlelem = &PL_sv_undef;
5140 I32 ix = SvIV(*lastlelem);
5143 if (ix < 0 || ix >= max)
5144 *firstlelem = &PL_sv_undef;
5146 *firstlelem = firstrelem[ix];
5153 SP = firstlelem - 1;
5157 for (lelem = firstlelem; lelem <= lastlelem; lelem++) {
5158 I32 ix = SvIV(*lelem);
5161 if (ix < 0 || ix >= max)
5162 *lelem = &PL_sv_undef;
5164 if (!(*lelem = firstrelem[ix]))
5165 *lelem = &PL_sv_undef;
5166 else if (mod && SvPADTMP(*lelem)) {
5167 *lelem = firstrelem[ix] = sv_mortalcopy(*lelem);
5178 const I32 items = SP - MARK;
5179 SV * const av = MUTABLE_SV(av_make(items, MARK+1));
5181 mXPUSHs((PL_op->op_flags & OPf_SPECIAL)
5182 ? newRV_noinc(av) : av);
5188 dSP; dMARK; dORIGMARK;
5189 HV* const hv = newHV();
5190 SV* const retval = sv_2mortal( PL_op->op_flags & OPf_SPECIAL
5191 ? newRV_noinc(MUTABLE_SV(hv))
5196 (MARK++, SvGMAGICAL(*MARK) ? sv_mortalcopy(*MARK) : *MARK);
5203 sv_setsv_nomg(val, *MARK);
5207 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Odd number of elements in anonymous hash");
5210 (void)hv_store_ent(hv,key,val,0);
5219 dSP; dMARK; dORIGMARK;
5220 int num_args = (SP - MARK);
5221 AV *ary = MUTABLE_AV(*++MARK);
5230 const MAGIC * const mg = SvTIED_mg((const SV *)ary, PERL_MAGIC_tied);
5233 return Perl_tied_method(aTHX_ SV_CONST(SPLICE), mark - 1, MUTABLE_SV(ary), mg,
5234 GIMME_V | TIED_METHOD_ARGUMENTS_ON_STACK,
5238 if (SvREADONLY(ary))
5239 Perl_croak_no_modify();
5244 offset = i = SvIV(*MARK);
5246 offset += AvFILLp(ary) + 1;
5248 DIE(aTHX_ PL_no_aelem, i);
5250 length = SvIVx(*MARK++);
5252 length += AvFILLp(ary) - offset + 1;
5258 length = AvMAX(ary) + 1; /* close enough to infinity */
5262 length = AvMAX(ary) + 1;
5264 if (offset > AvFILLp(ary) + 1) {
5266 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "splice() offset past end of array" );
5267 offset = AvFILLp(ary) + 1;
5269 after = AvFILLp(ary) + 1 - (offset + length);
5270 if (after < 0) { /* not that much array */
5271 length += after; /* offset+length now in array */
5277 /* At this point, MARK .. SP-1 is our new LIST */
5280 diff = newlen - length;
5281 if (newlen && !AvREAL(ary) && AvREIFY(ary))
5284 /* make new elements SVs now: avoid problems if they're from the array */
5285 for (dst = MARK, i = newlen; i; i--) {
5286 SV * const h = *dst;
5287 *dst++ = newSVsv(h);
5290 if (diff < 0) { /* shrinking the area */
5291 SV **tmparyval = NULL;
5293 Newx(tmparyval, newlen, SV*); /* so remember insertion */
5294 Copy(MARK, tmparyval, newlen, SV*);
5297 MARK = ORIGMARK + 1;
5298 if (GIMME_V == G_ARRAY) { /* copy return vals to stack */
5299 const bool real = cBOOL(AvREAL(ary));
5300 MEXTEND(MARK, length);
5302 EXTEND_MORTAL(length);
5303 for (i = 0, dst = MARK; i < length; i++) {
5304 if ((*dst = AvARRAY(ary)[i+offset])) {
5306 sv_2mortal(*dst); /* free them eventually */
5309 *dst = &PL_sv_undef;
5315 *MARK = AvARRAY(ary)[offset+length-1];
5318 for (i = length - 1, dst = &AvARRAY(ary)[offset]; i > 0; i--)
5319 SvREFCNT_dec(*dst++); /* free them now */
5322 *MARK = &PL_sv_undef;
5324 AvFILLp(ary) += diff;
5326 /* pull up or down? */
5328 if (offset < after) { /* easier to pull up */
5329 if (offset) { /* esp. if nothing to pull */
5330 src = &AvARRAY(ary)[offset-1];
5331 dst = src - diff; /* diff is negative */
5332 for (i = offset; i > 0; i--) /* can't trust Copy */
5336 AvARRAY(ary) = AvARRAY(ary) - diff; /* diff is negative */
5340 if (after) { /* anything to pull down? */
5341 src = AvARRAY(ary) + offset + length;
5342 dst = src + diff; /* diff is negative */
5343 Move(src, dst, after, SV*);
5345 dst = &AvARRAY(ary)[AvFILLp(ary)+1];
5346 /* avoid later double free */
5353 Copy( tmparyval, AvARRAY(ary) + offset, newlen, SV* );
5354 Safefree(tmparyval);
5357 else { /* no, expanding (or same) */
5358 SV** tmparyval = NULL;
5360 Newx(tmparyval, length, SV*); /* so remember deletion */
5361 Copy(AvARRAY(ary)+offset, tmparyval, length, SV*);
5364 if (diff > 0) { /* expanding */
5365 /* push up or down? */
5366 if (offset < after && diff <= AvARRAY(ary) - AvALLOC(ary)) {
5370 Move(src, dst, offset, SV*);
5372 AvARRAY(ary) = AvARRAY(ary) - diff;/* diff is positive */
5374 AvFILLp(ary) += diff;
5377 if (AvFILLp(ary) + diff >= AvMAX(ary)) /* oh, well */
5378 av_extend(ary, AvFILLp(ary) + diff);
5379 AvFILLp(ary) += diff;
5382 dst = AvARRAY(ary) + AvFILLp(ary);
5384 for (i = after; i; i--) {
5392 Copy( MARK, AvARRAY(ary) + offset, newlen, SV* );
5395 MARK = ORIGMARK + 1;
5396 if (GIMME_V == G_ARRAY) { /* copy return vals to stack */
5398 const bool real = cBOOL(AvREAL(ary));
5400 EXTEND_MORTAL(length);
5401 for (i = 0, dst = MARK; i < length; i++) {
5402 if ((*dst = tmparyval[i])) {
5404 sv_2mortal(*dst); /* free them eventually */
5406 else *dst = &PL_sv_undef;
5412 else if (length--) {
5413 *MARK = tmparyval[length];
5416 while (length-- > 0)
5417 SvREFCNT_dec(tmparyval[length]);
5420 *MARK = &PL_sv_undef;
5423 *MARK = &PL_sv_undef;
5424 Safefree(tmparyval);
5428 mg_set(MUTABLE_SV(ary));
5436 dSP; dMARK; dORIGMARK; dTARGET;
5437 AV * const ary = MUTABLE_AV(*++MARK);
5438 const MAGIC * const mg = SvTIED_mg((const SV *)ary, PERL_MAGIC_tied);
5441 *MARK-- = SvTIED_obj(MUTABLE_SV(ary), mg);
5444 ENTER_with_name("call_PUSH");
5445 call_sv(SV_CONST(PUSH),G_SCALAR|G_DISCARD|G_METHOD_NAMED);
5446 LEAVE_with_name("call_PUSH");
5447 /* SPAGAIN; not needed: SP is assigned to immediately below */
5450 /* PL_delaymagic is restored by JUMPENV_POP on dieing, so we
5451 * only need to save locally, not on the save stack */
5452 U16 old_delaymagic = PL_delaymagic;
5454 if (SvREADONLY(ary) && MARK < SP) Perl_croak_no_modify();
5455 PL_delaymagic = DM_DELAY;
5456 for (++MARK; MARK <= SP; MARK++) {
5458 if (*MARK) SvGETMAGIC(*MARK);
5461 sv_setsv_nomg(sv, *MARK);
5462 av_store(ary, AvFILLp(ary)+1, sv);
5464 if (PL_delaymagic & DM_ARRAY_ISA)
5465 mg_set(MUTABLE_SV(ary));
5466 PL_delaymagic = old_delaymagic;
5469 if (OP_GIMME(PL_op, 0) != G_VOID) {
5470 PUSHi( AvFILL(ary) + 1 );
5475 /* also used for: pp_pop()*/
5479 AV * const av = PL_op->op_flags & OPf_SPECIAL
5480 ? MUTABLE_AV(GvAVn(PL_defgv)) : MUTABLE_AV(POPs);
5481 SV * const sv = PL_op->op_type == OP_SHIFT ? av_shift(av) : av_pop(av);
5485 (void)sv_2mortal(sv);
5492 dSP; dMARK; dORIGMARK; dTARGET;
5493 AV *ary = MUTABLE_AV(*++MARK);
5494 const MAGIC * const mg = SvTIED_mg((const SV *)ary, PERL_MAGIC_tied);
5497 *MARK-- = SvTIED_obj(MUTABLE_SV(ary), mg);
5500 ENTER_with_name("call_UNSHIFT");
5501 call_sv(SV_CONST(UNSHIFT),G_SCALAR|G_DISCARD|G_METHOD_NAMED);
5502 LEAVE_with_name("call_UNSHIFT");
5503 /* SPAGAIN; not needed: SP is assigned to immediately below */
5506 /* PL_delaymagic is restored by JUMPENV_POP on dieing, so we
5507 * only need to save locally, not on the save stack */
5508 U16 old_delaymagic = PL_delaymagic;
5511 av_unshift(ary, SP - MARK);
5512 PL_delaymagic = DM_DELAY;
5514 SV * const sv = newSVsv(*++MARK);
5515 (void)av_store(ary, i++, sv);
5517 if (PL_delaymagic & DM_ARRAY_ISA)
5518 mg_set(MUTABLE_SV(ary));
5519 PL_delaymagic = old_delaymagic;
5522 if (OP_GIMME(PL_op, 0) != G_VOID) {
5523 PUSHi( AvFILL(ary) + 1 );
5532 if (GIMME_V == G_ARRAY) {
5533 if (PL_op->op_private & OPpREVERSE_INPLACE) {
5537 assert( MARK+1 == SP && *SP && SvTYPE(*SP) == SVt_PVAV);
5538 (void)POPMARK; /* remove mark associated with ex-OP_AASSIGN */
5539 av = MUTABLE_AV((*SP));
5540 /* In-place reversing only happens in void context for the array
5541 * assignment. We don't need to push anything on the stack. */
5544 if (SvMAGICAL(av)) {
5546 SV *tmp = sv_newmortal();
5547 /* For SvCANEXISTDELETE */
5550 bool can_preserve = SvCANEXISTDELETE(av);
5552 for (i = 0, j = av_tindex(av); i < j; ++i, --j) {
5556 if (!av_exists(av, i)) {
5557 if (av_exists(av, j)) {
5558 SV *sv = av_delete(av, j, 0);
5559 begin = *av_fetch(av, i, TRUE);
5560 sv_setsv_mg(begin, sv);
5564 else if (!av_exists(av, j)) {
5565 SV *sv = av_delete(av, i, 0);
5566 end = *av_fetch(av, j, TRUE);
5567 sv_setsv_mg(end, sv);
5572 begin = *av_fetch(av, i, TRUE);
5573 end = *av_fetch(av, j, TRUE);
5574 sv_setsv(tmp, begin);
5575 sv_setsv_mg(begin, end);
5576 sv_setsv_mg(end, tmp);
5580 SV **begin = AvARRAY(av);
5583 SV **end = begin + AvFILLp(av);
5585 while (begin < end) {
5586 SV * const tmp = *begin;
5590 if (tmp && SvWEAKREF(tmp))
5594 /* make sure we catch the middle element */
5595 if (begin == end && *begin && SvWEAKREF(*begin))
5596 sv_rvunweaken(*begin);
5604 SV * const tmp = *MARK;
5608 /* safe as long as stack cannot get extended in the above */
5617 SvUTF8_off(TARG); /* decontaminate */
5619 do_join(TARG, &PL_sv_no, MARK, SP);
5621 sv_setsv(TARG, *SP);
5623 sv_setsv(TARG, DEFSV);
5627 up = SvPV_force(TARG, len);
5630 if (DO_UTF8(TARG)) { /* first reverse each character */
5631 U8* s = (U8*)SvPVX(TARG);
5632 const U8* send = (U8*)(s + len);
5634 if (UTF8_IS_INVARIANT(*s)) {
5639 if (!utf8_to_uvchr_buf(s, send, 0))
5643 down = (char*)(s - 1);
5644 /* reverse this character */
5646 const char tmp = *up;
5654 down = SvPVX(TARG) + len - 1;
5656 const char tmp = *up;
5660 (void)SvPOK_only_UTF8(TARG);
5671 AV *ary = ( (PL_op->op_private & OPpSPLIT_ASSIGN) /* @a = split */
5672 && (PL_op->op_flags & OPf_STACKED)) /* @{expr} = split */
5673 ? (AV *)POPs : NULL;
5674 IV limit = POPi; /* note, negative is forever */
5675 SV * const sv = POPs;
5677 const char *s = SvPV_const(sv, len);
5678 const bool do_utf8 = DO_UTF8(sv);
5679 const bool in_uni_8_bit = IN_UNI_8_BIT;
5680 const char *strend = s + len;
5681 PMOP *pm = cPMOPx(PL_op);
5686 const STRLEN slen = do_utf8
5687 ? utf8_length((U8*)s, (U8*)strend)
5688 : (STRLEN)(strend - s);
5689 SSize_t maxiters = slen + 10;
5690 I32 trailing_empty = 0;
5692 const IV origlimit = limit;
5695 const U8 gimme = GIMME_V;
5697 I32 oldsave = PL_savestack_ix;
5698 U32 make_mortal = SVs_TEMP;
5704 TAINT_IF(get_regex_charset(RX_EXTFLAGS(rx)) == REGEX_LOCALE_CHARSET &&
5705 (RX_EXTFLAGS(rx) & (RXf_WHITE | RXf_SKIPWHITE)));
5707 /* handle @ary = split(...) optimisation */
5708 if (PL_op->op_private & OPpSPLIT_ASSIGN) {
5709 if (!(PL_op->op_flags & OPf_STACKED)) {
5710 if (PL_op->op_private & OPpSPLIT_LEX) {
5711 if (PL_op->op_private & OPpLVAL_INTRO)
5712 SAVECLEARSV(PAD_SVl(pm->op_pmreplrootu.op_pmtargetoff));
5713 ary = (AV *)PAD_SVl(pm->op_pmreplrootu.op_pmtargetoff);
5718 MUTABLE_GV(PAD_SVl(pm->op_pmreplrootu.op_pmtargetoff));
5720 pm->op_pmreplrootu.op_pmtargetgv;
5722 if (PL_op->op_private & OPpLVAL_INTRO)
5727 /* skip anything pushed by OPpLVAL_INTRO above */
5728 oldsave = PL_savestack_ix;
5734 (void)sv_2mortal(SvREFCNT_inc_simple_NN(sv));
5737 if ((mg = SvTIED_mg((const SV *)ary, PERL_MAGIC_tied))) {
5739 XPUSHs(SvTIED_obj(MUTABLE_SV(ary), mg));
5746 for (i = AvFILLp(ary); i >= 0; i--)
5747 AvARRAY(ary)[i] = &PL_sv_undef; /* don't free mere refs */
5749 /* temporarily switch stacks */
5750 SAVESWITCHSTACK(PL_curstack, ary);
5755 base = SP - PL_stack_base;
5757 if (RX_EXTFLAGS(rx) & RXf_SKIPWHITE) {
5759 while (s < strend && isSPACE_utf8_safe(s, strend))
5762 else if (get_regex_charset(RX_EXTFLAGS(rx)) == REGEX_LOCALE_CHARSET) {
5763 while (s < strend && isSPACE_LC(*s))
5766 else if (in_uni_8_bit) {
5767 while (s < strend && isSPACE_L1(*s))
5771 while (s < strend && isSPACE(*s))
5775 if (RX_EXTFLAGS(rx) & RXf_PMf_MULTILINE) {
5779 gimme_scalar = gimme == G_SCALAR && !ary;
5782 limit = maxiters + 2;
5783 if (RX_EXTFLAGS(rx) & RXf_WHITE) {
5786 /* this one uses 'm' and is a negative test */
5788 while (m < strend && ! isSPACE_utf8_safe(m, strend) ) {
5789 const int t = UTF8SKIP(m);
5790 /* isSPACE_utf8_safe returns FALSE for malform utf8 */
5797 else if (get_regex_charset(RX_EXTFLAGS(rx)) == REGEX_LOCALE_CHARSET)
5799 while (m < strend && !isSPACE_LC(*m))
5802 else if (in_uni_8_bit) {
5803 while (m < strend && !isSPACE_L1(*m))
5806 while (m < strend && !isSPACE(*m))
5819 dstr = newSVpvn_flags(s, m-s,
5820 (do_utf8 ? SVf_UTF8 : 0) | make_mortal);
5824 /* skip the whitespace found last */
5826 s = m + UTF8SKIP(m);
5830 /* this one uses 's' and is a positive test */
5832 while (s < strend && isSPACE_utf8_safe(s, strend) )
5835 else if (get_regex_charset(RX_EXTFLAGS(rx)) == REGEX_LOCALE_CHARSET)
5837 while (s < strend && isSPACE_LC(*s))
5840 else if (in_uni_8_bit) {
5841 while (s < strend && isSPACE_L1(*s))
5844 while (s < strend && isSPACE(*s))
5849 else if (RX_EXTFLAGS(rx) & RXf_START_ONLY) {
5851 for (m = s; m < strend && *m != '\n'; m++)
5864 dstr = newSVpvn_flags(s, m-s,
5865 (do_utf8 ? SVf_UTF8 : 0) | make_mortal);
5871 else if (RX_EXTFLAGS(rx) & RXf_NULL && !(s >= strend)) {
5873 Pre-extend the stack, either the number of bytes or
5874 characters in the string or a limited amount, triggered by:
5876 my ($x, $y) = split //, $str;
5880 if (!gimme_scalar) {
5881 const IV items = limit - 1;
5882 /* setting it to -1 will trigger a panic in EXTEND() */
5883 const SSize_t sslen = slen > SSize_t_MAX ? -1 : (SSize_t)slen;
5884 if (items >=0 && items < sslen)
5892 /* keep track of how many bytes we skip over */
5902 dstr = newSVpvn_flags(m, s-m, SVf_UTF8 | make_mortal);
5915 dstr = newSVpvn(s, 1);
5931 else if (do_utf8 == (RX_UTF8(rx) != 0) &&
5932 (RX_EXTFLAGS(rx) & RXf_USE_INTUIT) && !RX_NPARENS(rx)
5933 && (RX_EXTFLAGS(rx) & RXf_CHECK_ALL)
5934 && !(RX_EXTFLAGS(rx) & RXf_IS_ANCHORED)) {
5935 const int tail = (RX_EXTFLAGS(rx) & RXf_INTUIT_TAIL);
5936 SV * const csv = CALLREG_INTUIT_STRING(rx);
5938 len = RX_MINLENRET(rx);
5939 if (len == 1 && !RX_UTF8(rx) && !tail) {
5940 const char c = *SvPV_nolen_const(csv);
5942 for (m = s; m < strend && *m != c; m++)
5953 dstr = newSVpvn_flags(s, m-s,
5954 (do_utf8 ? SVf_UTF8 : 0) | make_mortal);
5957 /* The rx->minlen is in characters but we want to step
5958 * s ahead by bytes. */
5960 s = (char*)utf8_hop((U8*)m, len);
5962 s = m + len; /* Fake \n at the end */
5966 while (s < strend && --limit &&
5967 (m = fbm_instr((unsigned char*)s, (unsigned char*)strend,
5968 csv, multiline ? FBMrf_MULTILINE : 0)) )
5977 dstr = newSVpvn_flags(s, m-s,
5978 (do_utf8 ? SVf_UTF8 : 0) | make_mortal);
5981 /* The rx->minlen is in characters but we want to step
5982 * s ahead by bytes. */
5984 s = (char*)utf8_hop((U8*)m, len);
5986 s = m + len; /* Fake \n at the end */
5991 maxiters += slen * RX_NPARENS(rx);
5992 while (s < strend && --limit)
5996 rex_return = CALLREGEXEC(rx, (char*)s, (char*)strend, (char*)orig, 1,
5999 if (rex_return == 0)
6001 TAINT_IF(RX_MATCH_TAINTED(rx));
6002 /* we never pass the REXEC_COPY_STR flag, so it should
6003 * never get copied */
6004 assert(!RX_MATCH_COPIED(rx));
6005 m = RX_OFFS(rx)[0].start + orig;
6014 dstr = newSVpvn_flags(s, m-s,
6015 (do_utf8 ? SVf_UTF8 : 0) | make_mortal);
6018 if (RX_NPARENS(rx)) {
6020 for (i = 1; i <= (I32)RX_NPARENS(rx); i++) {
6021 s = RX_OFFS(rx)[i].start + orig;
6022 m = RX_OFFS(rx)[i].end + orig;
6024 /* japhy (07/27/01) -- the (m && s) test doesn't catch
6025 parens that didn't match -- they should be set to
6026 undef, not the empty string */
6034 if (m >= orig && s >= orig) {
6035 dstr = newSVpvn_flags(s, m-s,
6036 (do_utf8 ? SVf_UTF8 : 0)
6040 dstr = &PL_sv_undef; /* undef, not "" */
6046 s = RX_OFFS(rx)[0].end + orig;
6050 if (!gimme_scalar) {
6051 iters = (SP - PL_stack_base) - base;
6053 if (iters > maxiters)
6054 DIE(aTHX_ "Split loop");
6056 /* keep field after final delim? */
6057 if (s < strend || (iters && origlimit)) {
6058 if (!gimme_scalar) {
6059 const STRLEN l = strend - s;
6060 dstr = newSVpvn_flags(s, l, (do_utf8 ? SVf_UTF8 : 0) | make_mortal);
6065 else if (!origlimit) {
6067 iters -= trailing_empty;
6069 while (iters > 0 && (!TOPs || !SvANY(TOPs) || SvCUR(TOPs) == 0)) {
6070 if (TOPs && !make_mortal)
6079 LEAVE_SCOPE(oldsave); /* may undo an earlier SWITCHSTACK */
6083 if (SvSMAGICAL(ary)) {
6085 mg_set(MUTABLE_SV(ary));
6088 if (gimme == G_ARRAY) {
6090 Copy(AvARRAY(ary), SP + 1, iters, SV*);
6097 ENTER_with_name("call_PUSH");
6098 call_sv(SV_CONST(PUSH),G_SCALAR|G_DISCARD|G_METHOD_NAMED);
6099 LEAVE_with_name("call_PUSH");
6101 if (gimme == G_ARRAY) {
6103 /* EXTEND should not be needed - we just popped them */
6105 for (i=0; i < iters; i++) {
6106 SV **svp = av_fetch(ary, i, FALSE);
6107 PUSHs((svp) ? *svp : &PL_sv_undef);
6114 if (gimme == G_ARRAY)
6126 SV *const sv = PAD_SVl(PL_op->op_targ);
6128 if (SvPADSTALE(sv)) {
6131 RETURNOP(cLOGOP->op_other);
6133 RETURNOP(cLOGOP->op_next);
6142 if (SvTYPE(retsv) == SVt_PVAV || SvTYPE(retsv) == SVt_PVHV
6143 || SvTYPE(retsv) == SVt_PVCV) {
6144 retsv = refto(retsv);
6151 /* used for: pp_padany(), pp_custom(); plus any system ops
6152 * that aren't implemented on a particular platform */
6154 PP(unimplemented_op)
6156 const Optype op_type = PL_op->op_type;
6157 /* Using OP_NAME() isn't going to be helpful here. Firstly, it doesn't cope
6158 with out of range op numbers - it only "special" cases op_custom.
6159 Secondly, as the three ops we "panic" on are padmy, mapstart and custom,
6160 if we get here for a custom op then that means that the custom op didn't
6161 have an implementation. Given that OP_NAME() looks up the custom op
6162 by its pp_addr, likely it will return NULL, unless someone (unhelpfully)
6163 registers &PL_unimplemented_op as the address of their custom op.
6164 NULL doesn't generate a useful error message. "custom" does. */
6165 const char *const name = op_type >= OP_max
6166 ? "[out of range]" : PL_op_name[PL_op->op_type];
6167 if(OP_IS_SOCKET(op_type))
6168 DIE(aTHX_ PL_no_sock_func, name);
6169 DIE(aTHX_ "panic: unimplemented op %s (#%d) called", name, op_type);
6173 S_maybe_unwind_defav(pTHX)
6175 if (CX_CUR()->cx_type & CXp_HASARGS) {
6176 PERL_CONTEXT *cx = CX_CUR();
6178 assert(CxHASARGS(cx));
6180 cx->cx_type &= ~CXp_HASARGS;
6184 /* For sorting out arguments passed to a &CORE:: subroutine */
6188 int opnum = SvIOK(cSVOP_sv) ? (int)SvUV(cSVOP_sv) : 0;
6189 int defgv = PL_opargs[opnum] & OA_DEFGV ||opnum==OP_GLOB, whicharg = 0;
6190 AV * const at_ = GvAV(PL_defgv);
6191 SV **svp = at_ ? AvARRAY(at_) : NULL;
6192 I32 minargs = 0, maxargs = 0, numargs = at_ ? AvFILLp(at_)+1 : 0;
6193 I32 oa = opnum ? PL_opargs[opnum] >> OASHIFT : 0;
6194 bool seen_question = 0;
6195 const char *err = NULL;
6196 const bool pushmark = PL_op->op_private & OPpCOREARGS_PUSHMARK;
6198 /* Count how many args there are first, to get some idea how far to
6199 extend the stack. */
6201 if ((oa & 7) == OA_LIST) { maxargs = I32_MAX; break; }
6203 if (oa & OA_OPTIONAL) seen_question = 1;
6204 if (!seen_question) minargs++;
6208 if(numargs < minargs) err = "Not enough";
6209 else if(numargs > maxargs) err = "Too many";
6211 /* diag_listed_as: Too many arguments for %s */
6213 "%s arguments for %s", err,
6214 opnum ? PL_op_desc[opnum] : SvPV_nolen_const(cSVOP_sv)
6217 /* Reset the stack pointer. Without this, we end up returning our own
6218 arguments in list context, in addition to the values we are supposed
6219 to return. nextstate usually does this on sub entry, but we need
6220 to run the next op with the caller's hints, so we cannot have a
6222 SP = PL_stack_base + CX_CUR()->blk_oldsp;
6224 if(!maxargs) RETURN;
6226 /* We do this here, rather than with a separate pushmark op, as it has
6227 to come in between two things this function does (stack reset and
6228 arg pushing). This seems the easiest way to do it. */
6231 (void)Perl_pp_pushmark(aTHX);
6234 EXTEND(SP, maxargs == I32_MAX ? numargs : maxargs);
6235 PUTBACK; /* The code below can die in various places. */
6237 oa = PL_opargs[opnum] >> OASHIFT;
6238 for (; oa&&(numargs||!pushmark); (void)(numargs&&(++svp,--numargs))) {
6243 if (!numargs && defgv && whicharg == minargs + 1) {
6246 else PUSHs(numargs ? svp && *svp ? *svp : &PL_sv_undef : NULL);
6250 PUSHs(svp && *svp ? *svp : &PL_sv_undef);
6257 if (CvUNIQUE(find_runcv_where(FIND_RUNCV_level_eq,1,NULL)))
6260 S_maybe_unwind_defav(aTHX);
6263 PUSHs((SV *)GvAVn(gv));
6266 if (!svp || !*svp || !SvROK(*svp)
6267 || SvTYPE(SvRV(*svp)) != SVt_PVAV)
6269 /* diag_listed_as: Type of arg %d to &CORE::%s must be %s*/
6270 "Type of arg %d to &CORE::%s must be array reference",
6271 whicharg, PL_op_desc[opnum]
6276 if (!svp || !*svp || !SvROK(*svp)
6277 || ( SvTYPE(SvRV(*svp)) != SVt_PVHV
6278 && ( opnum == OP_DBMCLOSE || opnum == OP_DBMOPEN
6279 || SvTYPE(SvRV(*svp)) != SVt_PVAV )))
6281 /* diag_listed_as: Type of arg %d to &CORE::%s must be %s*/
6282 "Type of arg %d to &CORE::%s must be hash%s reference",
6283 whicharg, PL_op_desc[opnum],
6284 opnum == OP_DBMCLOSE || opnum == OP_DBMOPEN
6291 if (!numargs) PUSHs(NULL);
6292 else if(svp && *svp && SvROK(*svp) && isGV_with_GP(SvRV(*svp)))
6293 /* no magic here, as the prototype will have added an extra
6294 refgen and we just want what was there before that */
6297 const bool constr = PL_op->op_private & whicharg;
6299 svp && *svp ? *svp : &PL_sv_undef,
6300 constr, cBOOL(CopHINTS_get(PL_curcop) & HINT_STRICT_REFS),
6306 if (!numargs) goto try_defsv;
6308 const bool wantscalar =
6309 PL_op->op_private & OPpCOREARGS_SCALARMOD;
6310 if (!svp || !*svp || !SvROK(*svp)
6311 /* We have to permit globrefs even for the \$ proto, as
6312 *foo is indistinguishable from ${\*foo}, and the proto-
6313 type permits the latter. */
6314 || SvTYPE(SvRV(*svp)) > (
6315 wantscalar ? SVt_PVLV
6316 : opnum == OP_LOCK || opnum == OP_UNDEF
6322 "Type of arg %d to &CORE::%s must be %s",
6323 whicharg, PL_op_name[opnum],
6325 ? "scalar reference"
6326 : opnum == OP_LOCK || opnum == OP_UNDEF
6327 ? "reference to one of [$@%&*]"
6328 : "reference to one of [$@%*]"
6331 if (opnum == OP_UNDEF && SvRV(*svp) == (SV *)PL_defgv) {
6332 /* Undo @_ localisation, so that sub exit does not undo
6333 part of our undeffing. */
6334 S_maybe_unwind_defav(aTHX);
6339 DIE(aTHX_ "panic: unknown OA_*: %x", (unsigned)(oa&7));
6347 /* Implement CORE::keys(),values(),each().
6349 * We won't know until run-time whether the arg is an array or hash,
6352 * pp_keys/pp_values/pp_each
6354 * pp_akeys/pp_avalues/pp_aeach
6356 * as appropriate (or whatever pp function actually implements the OP_FOO
6357 * functionality for each FOO).
6364 (SvTYPE(TOPs) == SVt_PVAV ? OP_AEACH : OP_EACH)
6365 + (PL_op->op_private & OPpAVHVSWITCH_MASK)
6373 if (PL_op->op_private & OPpOFFBYONE) {
6374 cv = find_runcv_where(FIND_RUNCV_level_eq, 1, NULL);
6376 else cv = find_runcv(NULL);
6377 XPUSHs(CvEVAL(cv) ? &PL_sv_undef : sv_2mortal(newRV((SV *)cv)));
6382 S_localise_aelem_lval(pTHX_ AV * const av, SV * const keysv,
6383 const bool can_preserve)
6385 const SSize_t ix = SvIV(keysv);
6386 if (can_preserve ? av_exists(av, ix) : TRUE) {
6387 SV ** const svp = av_fetch(av, ix, 1);
6389 Perl_croak(aTHX_ PL_no_aelem, ix);
6390 save_aelem(av, ix, svp);
6393 SAVEADELETE(av, ix);
6397 S_localise_helem_lval(pTHX_ HV * const hv, SV * const keysv,
6398 const bool can_preserve)
6400 if (can_preserve ? hv_exists_ent(hv, keysv, 0) : TRUE) {
6401 HE * const he = hv_fetch_ent(hv, keysv, 1, 0);
6402 SV ** const svp = he ? &HeVAL(he) : NULL;
6404 Perl_croak(aTHX_ PL_no_helem_sv, SVfARG(keysv));
6405 save_helem_flags(hv, keysv, svp, 0);
6408 SAVEHDELETE(hv, keysv);
6412 S_localise_gv_slot(pTHX_ GV *gv, U8 type)
6414 if (type == OPpLVREF_SV) {
6415 save_pushptrptr(gv, SvREFCNT_inc_simple(GvSV(gv)), SAVEt_GVSV);
6418 else if (type == OPpLVREF_AV)
6419 /* XXX Inefficient, as it creates a new AV, which we are
6420 about to clobber. */
6423 assert(type == OPpLVREF_HV);
6424 /* XXX Likewise inefficient. */
6433 SV * const key = PL_op->op_private & OPpLVREF_ELEM ? POPs : NULL;
6434 SV * const left = PL_op->op_flags & OPf_STACKED ? POPs : NULL;
6436 const char *bad = NULL;
6437 const U8 type = PL_op->op_private & OPpLVREF_TYPE;
6438 if (!SvROK(sv)) DIE(aTHX_ "Assigned value is not a reference");
6441 if (SvTYPE(SvRV(sv)) > SVt_PVLV)
6445 if (SvTYPE(SvRV(sv)) != SVt_PVAV)
6449 if (SvTYPE(SvRV(sv)) != SVt_PVHV)
6453 if (SvTYPE(SvRV(sv)) != SVt_PVCV)
6457 /* diag_listed_as: Assigned value is not %s reference */
6458 DIE(aTHX_ "Assigned value is not a%s reference", bad);
6462 switch (left ? SvTYPE(left) : 0) {
6465 SV * const old = PAD_SV(ARGTARG);
6466 PAD_SETSV(ARGTARG, SvREFCNT_inc_NN(SvRV(sv)));
6468 if ((PL_op->op_private & (OPpLVAL_INTRO|OPpPAD_STATE))
6470 SAVECLEARSV(PAD_SVl(ARGTARG));
6474 if (PL_op->op_private & OPpLVAL_INTRO) {
6475 S_localise_gv_slot(aTHX_ (GV *)left, type);
6477 gv_setref(left, sv);
6482 if (UNLIKELY(PL_op->op_private & OPpLVAL_INTRO)) {
6483 S_localise_aelem_lval(aTHX_ (AV *)left, key,
6484 SvCANEXISTDELETE(left));
6486 av_store((AV *)left, SvIV(key), SvREFCNT_inc_simple_NN(SvRV(sv)));
6489 if (UNLIKELY(PL_op->op_private & OPpLVAL_INTRO)) {
6491 S_localise_helem_lval(aTHX_ (HV *)left, key,
6492 SvCANEXISTDELETE(left));
6494 (void)hv_store_ent((HV *)left, key, SvREFCNT_inc_simple_NN(SvRV(sv)), 0);
6496 if (PL_op->op_flags & OPf_MOD)
6497 SETs(sv_2mortal(newSVsv(sv)));
6498 /* XXX else can weak references go stale before they are read, e.g.,
6507 SV * const ret = sv_2mortal(newSV_type(SVt_PVMG));
6508 SV * const elem = PL_op->op_private & OPpLVREF_ELEM ? POPs : NULL;
6509 SV * const arg = PL_op->op_flags & OPf_STACKED ? POPs : NULL;
6510 MAGIC * const mg = sv_magicext(ret, arg, PERL_MAGIC_lvref,
6511 &PL_vtbl_lvref, (char *)elem,
6512 elem ? HEf_SVKEY : (I32)ARGTARG);
6513 mg->mg_private = PL_op->op_private;
6514 if (PL_op->op_private & OPpLVREF_ITER)
6515 mg->mg_flags |= MGf_PERSIST;
6516 if (UNLIKELY(PL_op->op_private & OPpLVAL_INTRO)) {
6522 const bool can_preserve = SvCANEXISTDELETE(arg);
6523 if (SvTYPE(arg) == SVt_PVAV)
6524 S_localise_aelem_lval(aTHX_ (AV *)arg, elem, can_preserve);
6526 S_localise_helem_lval(aTHX_ (HV *)arg, elem, can_preserve);
6530 S_localise_gv_slot(aTHX_ (GV *)arg,
6531 PL_op->op_private & OPpLVREF_TYPE);
6533 else if (!(PL_op->op_private & OPpPAD_STATE))
6534 SAVECLEARSV(PAD_SVl(ARGTARG));
6543 AV * const av = (AV *)POPs;
6544 const bool localizing = PL_op->op_private & OPpLVAL_INTRO;
6545 bool can_preserve = FALSE;
6547 if (UNLIKELY(localizing)) {
6552 can_preserve = SvCANEXISTDELETE(av);
6554 if (SvTYPE(av) == SVt_PVAV) {
6557 for (svp = MARK + 1; svp <= SP; svp++) {
6558 const SSize_t elem = SvIV(*svp);
6562 if (max > AvMAX(av))
6567 while (++MARK <= SP) {
6568 SV * const elemsv = *MARK;
6569 if (SvTYPE(av) == SVt_PVAV)
6570 S_localise_aelem_lval(aTHX_ av, elemsv, can_preserve);
6572 S_localise_helem_lval(aTHX_ (HV *)av, elemsv, can_preserve);
6573 *MARK = sv_2mortal(newSV_type(SVt_PVMG));
6574 sv_magic(*MARK,(SV *)av,PERL_MAGIC_lvref,(char *)elemsv,HEf_SVKEY);
6581 if (PL_op->op_flags & OPf_STACKED)
6582 Perl_pp_rv2av(aTHX);
6584 Perl_pp_padav(aTHX);
6588 SETs(0); /* special alias marker that aassign recognises */
6598 SETs(sv_2mortal((SV *)newCONSTSUB(SvTYPE(CopSTASH(PL_curcop))==SVt_PVHV
6599 ? CopSTASH(PL_curcop)
6601 NULL, SvREFCNT_inc_simple_NN(sv))));
6606 /* process one subroutine argument - typically when the sub has a signature:
6607 * introduce PL_curpad[op_targ] and assign to it the value
6608 * for $: (OPf_STACKED ? *sp : $_[N])
6609 * for @/%: @_[N..$#_]
6611 * It's equivalent to
6614 * my $foo = (value-on-stack)
6616 * my @foo = @_[N..$#_]
6626 AV *defav = GvAV(PL_defgv); /* @_ */
6627 IV ix = PTR2IV(cUNOP_AUXo->op_aux);
6630 /* do 'my $var, @var or %var' action */
6631 padentry = &(PAD_SVl(o->op_targ));
6632 save_clearsv(padentry);
6635 if ((o->op_private & OPpARGELEM_MASK) == OPpARGELEM_SV) {
6636 if (o->op_flags & OPf_STACKED) {
6643 /* should already have been checked */
6645 #if IVSIZE > PTRSIZE
6646 assert(ix <= SSize_t_MAX);
6649 svp = av_fetch(defav, ix, FALSE);
6650 val = svp ? *svp : &PL_sv_undef;
6655 /* cargo-culted from pp_sassign */
6656 assert(TAINTING_get || !TAINT_get);
6657 if (UNLIKELY(TAINT_get) && !SvTAINTED(val))
6660 SvSetMagicSV(targ, val);
6664 /* must be AV or HV */
6666 assert(!(o->op_flags & OPf_STACKED));
6667 argc = ((IV)AvFILL(defav) + 1) - ix;
6669 /* This is a copy of the relevant parts of pp_aassign().
6671 if ((o->op_private & OPpARGELEM_MASK) == OPpARGELEM_AV) {
6674 if (AvFILL((AV*)targ) > -1) {
6675 /* target should usually be empty. If we get get
6676 * here, someone's been doing some weird closure tricks.
6677 * Make a copy of all args before clearing the array,
6678 * to avoid the equivalent of @a = ($a[0]) prematurely freeing
6679 * elements. See similar code in pp_aassign.
6681 for (i = 0; i < argc; i++) {
6682 SV **svp = av_fetch(defav, ix + i, FALSE);
6683 SV *newsv = newSV(0);
6684 sv_setsv_flags(newsv,
6685 svp ? *svp : &PL_sv_undef,
6686 (SV_DO_COW_SVSETSV|SV_NOSTEAL));
6687 if (!av_store(defav, ix + i, newsv))
6688 SvREFCNT_dec_NN(newsv);
6690 av_clear((AV*)targ);
6696 av_extend((AV*)targ, argc);
6701 SV **svp = av_fetch(defav, ix + i, FALSE);
6702 SV *val = svp ? *svp : &PL_sv_undef;
6704 sv_setsv(tmpsv, val);
6705 av_store((AV*)targ, i++, tmpsv);
6713 assert((o->op_private & OPpARGELEM_MASK) == OPpARGELEM_HV);
6715 if (SvRMAGICAL(targ) || HvUSEDKEYS((HV*)targ)) {
6716 /* see "target should usually be empty" comment above */
6717 for (i = 0; i < argc; i++) {
6718 SV **svp = av_fetch(defav, ix + i, FALSE);
6719 SV *newsv = newSV(0);
6720 sv_setsv_flags(newsv,
6721 svp ? *svp : &PL_sv_undef,
6722 (SV_DO_COW_SVSETSV|SV_NOSTEAL));
6723 if (!av_store(defav, ix + i, newsv))
6724 SvREFCNT_dec_NN(newsv);
6726 hv_clear((HV*)targ);
6731 assert(argc % 2 == 0);
6740 svp = av_fetch(defav, ix + i++, FALSE);
6741 key = svp ? *svp : &PL_sv_undef;
6742 svp = av_fetch(defav, ix + i++, FALSE);
6743 val = svp ? *svp : &PL_sv_undef;
6746 if (UNLIKELY(SvGMAGICAL(key)))
6747 key = sv_mortalcopy(key);
6749 sv_setsv(tmpsv, val);
6750 hv_store_ent((HV*)targ, key, tmpsv, 0);
6758 /* Handle a default value for one subroutine argument (typically as part
6759 * of a subroutine signature).
6760 * It's equivalent to
6761 * @_ > op_targ ? $_[op_targ] : result_of(op_other)
6763 * Intended to be used where op_next is an OP_ARGELEM
6765 * We abuse the op_targ field slightly: it's an index into @_ rather than
6771 OP * const o = PL_op;
6772 AV *defav = GvAV(PL_defgv); /* @_ */
6773 IV ix = (IV)o->op_targ;
6776 #if IVSIZE > PTRSIZE
6777 assert(ix <= SSize_t_MAX);
6780 if (AvFILL(defav) >= ix) {
6782 SV **svp = av_fetch(defav, ix, FALSE);
6783 SV *val = svp ? *svp : &PL_sv_undef;
6787 return cLOGOPo->op_other;
6792 S_find_runcv_name(void)
6807 sv = sv_2mortal(newSV(0));
6808 gv_fullname4(sv, gv, NULL, TRUE);
6812 /* Check a a subs arguments - i.e. that it has the correct number of args
6813 * (and anything else we might think of in future). Typically used with
6819 OP * const o = PL_op;
6820 UNOP_AUX_item *aux = cUNOP_AUXo->op_aux;
6821 IV params = aux[0].iv;
6822 IV opt_params = aux[1].iv;
6823 char slurpy = (char)(aux[2].iv);
6824 AV *defav = GvAV(PL_defgv); /* @_ */
6828 assert(!SvMAGICAL(defav));
6829 argc = (AvFILLp(defav) + 1);
6830 too_few = (argc < (params - opt_params));
6832 if (UNLIKELY(too_few || (!slurpy && argc > params)))
6833 /* diag_listed_as: Too few arguments for subroutine '%s' */
6834 /* diag_listed_as: Too many arguments for subroutine '%s' */
6835 Perl_croak_caller("Too %s arguments for subroutine '%" SVf "'",
6836 too_few ? "few" : "many", S_find_runcv_name());
6838 if (UNLIKELY(slurpy == '%' && argc > params && (argc - params) % 2))
6839 /* diag_listed_as: Odd name/value argument for subroutine '%s' */
6840 Perl_croak_caller("Odd name/value argument for subroutine '%" SVf "'",
6841 S_find_runcv_name());
6847 * ex: set ts=8 sts=4 sw=4 et: