3 * Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000,
4 * 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008 by Larry Wall and others
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
12 * 'It's a big house this, and very peculiar. Always a bit more
13 * to discover, and no knowing what you'll find round a corner.
14 * And Elves, sir!' --Samwise Gamgee
16 * [p.225 of _The Lord of the Rings_, II/i: "Many Meetings"]
19 /* This file contains general pp ("push/pop") functions that execute the
20 * opcodes that make up a perl program. A typical pp function expects to
21 * find its arguments on the stack, and usually pushes its results onto
22 * the stack, hence the 'pp' terminology. Each OP structure contains
23 * a pointer to the relevant pp_foo() function.
32 #include "regcharclass.h"
34 static const STRLEN small_mu_len = sizeof(GREEK_SMALL_LETTER_MU_UTF8) - 1;
35 static const STRLEN capital_iota_len = sizeof(GREEK_CAPITAL_LETTER_IOTA_UTF8) - 1;
37 /* variations on pp_null */
42 if (GIMME_V == G_SCALAR)
54 assert(SvTYPE(TARG) == SVt_PVCV);
69 CV * const protocv = PadnamePROTOCV(
70 PadlistNAMESARRAY(CvPADLIST(find_runcv(NULL)))[ARGTARG]
72 assert(SvTYPE(TARG) == SVt_PVCV);
74 if (CvISXSUB(protocv)) { /* constant */
75 /* XXX Should we clone it here? */
76 /* If this changes to use SAVECLEARSV, we can move the SAVECLEARSV
77 to introcv and remove the SvPADSTALE_off. */
78 SAVEPADSVANDMORTALIZE(ARGTARG);
79 PAD_SVl(ARGTARG) = SvREFCNT_inc_simple_NN(protocv);
82 if (CvROOT(protocv)) {
83 assert(CvCLONE(protocv));
84 assert(!CvCLONED(protocv));
86 cv_clone_into(protocv,(CV *)TARG);
87 SAVECLEARSV(PAD_SVl(ARGTARG));
94 /* In some cases this function inspects PL_op. If this function is called
95 for new op types, more bool parameters may need to be added in place of
98 When noinit is true, the absence of a gv will cause a retval of undef.
99 This is unrelated to the cv-to-gv assignment case.
103 S_rv2gv(pTHX_ SV *sv, const bool vivify_sv, const bool strict,
106 if (!isGV(sv) || SvFAKE(sv)) SvGETMAGIC(sv);
109 sv = amagic_deref_call(sv, to_gv_amg);
113 if (SvTYPE(sv) == SVt_PVIO) {
114 GV * const gv = MUTABLE_GV(sv_newmortal());
115 gv_init(gv, 0, "__ANONIO__", 10, 0);
116 GvIOp(gv) = MUTABLE_IO(sv);
117 SvREFCNT_inc_void_NN(sv);
120 else if (!isGV_with_GP(sv)) {
121 Perl_die(aTHX_ "Not a GLOB reference");
125 if (!isGV_with_GP(sv)) {
127 /* If this is a 'my' scalar and flag is set then vivify
130 if (vivify_sv && sv != &PL_sv_undef) {
134 Perl_croak_no_modify();
135 gv = MUTABLE_GV(newSV(0));
136 stash = CopSTASH(PL_curcop);
137 if (SvTYPE(stash) != SVt_PVHV) stash = NULL;
138 if (cUNOP->op_targ) {
139 SV * const namesv = PAD_SV(cUNOP->op_targ);
140 gv_init_sv(gv, stash, namesv, 0);
143 gv_init_pv(gv, stash, "__ANONIO__", 0);
145 prepare_SV_for_RV(sv);
146 SvRV_set(sv, MUTABLE_SV(gv));
151 if (PL_op->op_flags & OPf_REF || strict) {
152 Perl_die(aTHX_ PL_no_usym, "a symbol");
154 if (ckWARN(WARN_UNINITIALIZED))
160 if (!(sv = MUTABLE_SV(gv_fetchsv_nomg(
161 sv, GV_ADDMG, SVt_PVGV
170 (SvPOKp(sv) && SvCUR(sv)>32 ? "..." : ""),
174 if ((PL_op->op_private & (OPpLVAL_INTRO|OPpDONT_INIT_GV))
175 == OPpDONT_INIT_GV) {
176 /* We are the target of a coderef assignment. Return
177 the scalar unchanged, and let pp_sasssign deal with
181 sv = MUTABLE_SV(gv_fetchsv_nomg(sv, GV_ADD, SVt_PVGV));
183 /* FAKE globs in the symbol table cause weird bugs (#77810) */
187 if (SvFAKE(sv) && !(PL_op->op_private & OPpALLOW_FAKE)) {
188 SV *newsv = sv_newmortal();
189 sv_setsv_flags(newsv, sv, 0);
201 sv, PL_op->op_private & OPpDEREF,
202 PL_op->op_private & HINT_STRICT_REFS,
203 ((PL_op->op_flags & OPf_SPECIAL) && !(PL_op->op_flags & OPf_MOD))
204 || PL_op->op_type == OP_READLINE
206 if (PL_op->op_private & OPpLVAL_INTRO)
207 save_gp(MUTABLE_GV(sv), !(PL_op->op_flags & OPf_SPECIAL));
212 /* Helper function for pp_rv2sv and pp_rv2av */
214 Perl_softref2xv(pTHX_ SV *const sv, const char *const what,
215 const svtype type, SV ***spp)
219 PERL_ARGS_ASSERT_SOFTREF2XV;
221 if (PL_op->op_private & HINT_STRICT_REFS) {
223 Perl_die(aTHX_ PL_no_symref_sv, sv,
224 (SvPOKp(sv) && SvCUR(sv)>32 ? "..." : ""), what);
226 Perl_die(aTHX_ PL_no_usym, what);
230 PL_op->op_flags & OPf_REF
232 Perl_die(aTHX_ PL_no_usym, what);
233 if (ckWARN(WARN_UNINITIALIZED))
235 if (type != SVt_PV && GIMME_V == G_ARRAY) {
239 **spp = &PL_sv_undef;
242 if ((PL_op->op_flags & OPf_SPECIAL) &&
243 !(PL_op->op_flags & OPf_MOD))
245 if (!(gv = gv_fetchsv_nomg(sv, GV_ADDMG, type)))
247 **spp = &PL_sv_undef;
252 gv = gv_fetchsv_nomg(sv, GV_ADD, type);
265 sv = amagic_deref_call(sv, to_sv_amg);
269 if (SvTYPE(sv) >= SVt_PVAV)
270 DIE(aTHX_ "Not a SCALAR reference");
275 if (!isGV_with_GP(gv)) {
276 gv = Perl_softref2xv(aTHX_ sv, "a SCALAR", SVt_PV, &sp);
282 if (PL_op->op_flags & OPf_MOD) {
283 if (PL_op->op_private & OPpLVAL_INTRO) {
284 if (cUNOP->op_first->op_type == OP_NULL)
285 sv = save_scalar(MUTABLE_GV(TOPs));
287 sv = save_scalar(gv);
289 Perl_croak(aTHX_ "%s", PL_no_localize_ref);
291 else if (PL_op->op_private & OPpDEREF)
292 sv = vivify_ref(sv, PL_op->op_private & OPpDEREF);
294 SPAGAIN; /* in case chasing soft refs reallocated the stack */
302 AV * const av = MUTABLE_AV(TOPs);
303 const I32 lvalue = PL_op->op_flags & OPf_MOD || LVRET;
305 SV ** const svp = Perl_av_arylen_p(aTHX_ MUTABLE_AV(av));
307 *svp = newSV_type(SVt_PVMG);
308 sv_magic(*svp, MUTABLE_SV(av), PERL_MAGIC_arylen, NULL, 0);
312 SETs(sv_2mortal(newSViv(AvFILL(MUTABLE_AV(av)))));
321 if (PL_op->op_flags & OPf_MOD || LVRET) {
322 SV * const ret = sv_2mortal(newSV_type(SVt_PVLV));/* Not TARG RT#67838 */
323 sv_magic(ret, NULL, PERL_MAGIC_pos, NULL, 0);
325 LvTARG(ret) = SvREFCNT_inc_simple(sv);
326 SETs(ret); /* no SvSETMAGIC */
329 const MAGIC * const mg = mg_find_mglob(sv);
330 if (mg && mg->mg_len != -1) {
331 STRLEN i = mg->mg_len;
332 if (PL_op->op_private & OPpTRUEBOOL)
333 SETs(i ? &PL_sv_yes : &PL_sv_zero);
336 if (mg->mg_flags & MGf_BYTES && DO_UTF8(sv))
337 i = sv_pos_b2u_flags(sv, i, SV_GMAGIC|SV_CONST_RETURN);
352 const I32 flags = (PL_op->op_flags & OPf_SPECIAL)
354 : ((PL_op->op_private & (OPpLVAL_INTRO|OPpMAY_RETURN_CONSTANT))
355 == OPpMAY_RETURN_CONSTANT)
358 /* We usually try to add a non-existent subroutine in case of AUTOLOAD. */
359 /* (But not in defined().) */
361 CV *cv = sv_2cv(TOPs, &stash_unused, &gv, flags);
363 else if ((flags == (GV_ADD|GV_NOEXPAND)) && gv && SvROK(gv)) {
364 cv = SvTYPE(SvRV(gv)) == SVt_PVCV
365 ? MUTABLE_CV(SvRV(gv))
369 cv = MUTABLE_CV(&PL_sv_undef);
370 SETs(MUTABLE_SV(cv));
380 SV *ret = &PL_sv_undef;
382 if (SvGMAGICAL(TOPs)) SETs(sv_mortalcopy(TOPs));
383 if (SvPOK(TOPs) && SvCUR(TOPs) >= 7) {
384 const char * s = SvPVX_const(TOPs);
385 if (memBEGINs(s, SvCUR(TOPs), "CORE::")) {
386 const int code = keyword(s + 6, SvCUR(TOPs) - 6, 1);
388 DIE(aTHX_ "Can't find an opnumber for \"%" UTF8f "\"",
389 UTF8fARG(SvFLAGS(TOPs) & SVf_UTF8, SvCUR(TOPs)-6, s+6));
391 SV * const sv = core_prototype(NULL, s + 6, code, NULL);
397 cv = sv_2cv(TOPs, &stash, &gv, 0);
399 ret = newSVpvn_flags(
400 CvPROTO(cv), CvPROTOLEN(cv), SVs_TEMP | SvUTF8(cv)
410 CV *cv = MUTABLE_CV(PAD_SV(PL_op->op_targ));
412 cv = MUTABLE_CV(sv_2mortal(MUTABLE_SV(cv_clone(cv))));
414 PUSHs(MUTABLE_SV(cv));
428 if (GIMME_V != G_ARRAY) {
434 *MARK = &PL_sv_undef;
436 *MARK = refto(*MARK);
440 EXTEND_MORTAL(SP - MARK);
442 *MARK = refto(*MARK);
447 S_refto(pTHX_ SV *sv)
451 PERL_ARGS_ASSERT_REFTO;
453 if (SvTYPE(sv) == SVt_PVLV && LvTYPE(sv) == 'y') {
456 if (!(sv = LvTARG(sv)))
459 SvREFCNT_inc_void_NN(sv);
461 else if (SvTYPE(sv) == SVt_PVAV) {
462 if (!AvREAL((const AV *)sv) && AvREIFY((const AV *)sv))
463 av_reify(MUTABLE_AV(sv));
465 SvREFCNT_inc_void_NN(sv);
467 else if (SvPADTMP(sv)) {
472 SvREFCNT_inc_void_NN(sv);
475 sv_upgrade(rv, SVt_IV);
484 SV * const sv = TOPs;
492 /* op is in boolean context? */
493 if ( (PL_op->op_private & OPpTRUEBOOL)
494 || ( (PL_op->op_private & OPpMAYBE_TRUEBOOL)
495 && block_gimme() == G_VOID))
497 /* refs are always true - unless it's to an object blessed into a
498 * class with a false name, i.e. "0". So we have to check for
499 * that remote possibility. The following is is basically an
500 * unrolled SvTRUE(sv_reftype(rv)) */
501 SV * const rv = SvRV(sv);
503 HV *stash = SvSTASH(rv);
504 HEK *hek = HvNAME_HEK(stash);
506 I32 len = HEK_LEN(hek);
507 /* bail out and do it the hard way? */
510 || (len == 1 && HEK_KEY(hek)[0] == '0')
523 sv_ref(TARG, SvRV(sv), TRUE);
539 stash = CopSTASH(PL_curcop);
540 if (SvTYPE(stash) != SVt_PVHV)
541 Perl_croak(aTHX_ "Attempt to bless into a freed package");
544 SV * const ssv = POPs;
548 if (!ssv) goto curstash;
551 if (!SvAMAGIC(ssv)) {
553 Perl_croak(aTHX_ "Attempt to bless into a reference");
555 /* SvAMAGIC is on here, but it only means potentially overloaded,
556 so after stringification: */
557 ptr = SvPV_nomg_const(ssv,len);
558 /* We need to check the flag again: */
559 if (!SvAMAGIC(ssv)) goto frog;
561 else ptr = SvPV_nomg_const(ssv,len);
563 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
564 "Explicit blessing to '' (assuming package main)");
565 stash = gv_stashpvn(ptr, len, GV_ADD|SvUTF8(ssv));
568 (void)sv_bless(TOPs, stash);
578 const char * const elem = SvPV_const(sv, len);
579 GV * const gv = MUTABLE_GV(TOPs);
584 /* elem will always be NUL terminated. */
587 if (memEQs(elem, len, "ARRAY"))
589 tmpRef = MUTABLE_SV(GvAV(gv));
590 if (tmpRef && !AvREAL((const AV *)tmpRef)
591 && AvREIFY((const AV *)tmpRef))
592 av_reify(MUTABLE_AV(tmpRef));
596 if (memEQs(elem, len, "CODE"))
597 tmpRef = MUTABLE_SV(GvCVu(gv));
600 if (memEQs(elem, len, "FILEHANDLE")) {
601 tmpRef = MUTABLE_SV(GvIOp(gv));
604 if (memEQs(elem, len, "FORMAT"))
605 tmpRef = MUTABLE_SV(GvFORM(gv));
608 if (memEQs(elem, len, "GLOB"))
609 tmpRef = MUTABLE_SV(gv);
612 if (memEQs(elem, len, "HASH"))
613 tmpRef = MUTABLE_SV(GvHV(gv));
616 if (memEQs(elem, len, "IO"))
617 tmpRef = MUTABLE_SV(GvIOp(gv));
620 if (memEQs(elem, len, "NAME"))
621 sv = newSVhek(GvNAME_HEK(gv));
624 if (memEQs(elem, len, "PACKAGE")) {
625 const HV * const stash = GvSTASH(gv);
626 const HEK * const hek = stash ? HvNAME_HEK(stash) : NULL;
627 sv = hek ? newSVhek(hek) : newSVpvs("__ANON__");
631 if (memEQs(elem, len, "SCALAR"))
646 /* Pattern matching */
654 if (len == 0 || len > I32_MAX || !SvPOK(sv) || SvUTF8(sv) || SvVALID(sv)) {
655 /* Historically, study was skipped in these cases. */
660 /* Make study a no-op. It's no longer useful and its existence
661 complicates matters elsewhere. */
667 /* also used for: pp_transr() */
674 if (PL_op->op_flags & OPf_STACKED)
679 sv = PAD_SV(ARGTARG);
684 if(PL_op->op_type == OP_TRANSR) {
686 const char * const pv = SvPV(sv,len);
687 SV * const newsv = newSVpvn_flags(pv, len, SVs_TEMP|SvUTF8(sv));
692 Size_t i = do_trans(sv);
698 /* Lvalue operators. */
701 S_do_chomp(pTHX_ SV *retval, SV *sv, bool chomping)
707 PERL_ARGS_ASSERT_DO_CHOMP;
709 if (chomping && (RsSNARF(PL_rs) || RsRECORD(PL_rs)))
711 if (SvTYPE(sv) == SVt_PVAV) {
713 AV *const av = MUTABLE_AV(sv);
714 const I32 max = AvFILL(av);
716 for (i = 0; i <= max; i++) {
717 sv = MUTABLE_SV(av_fetch(av, i, FALSE));
718 if (sv && ((sv = *(SV**)sv), sv != &PL_sv_undef))
719 count += do_chomp(retval, sv, chomping);
723 else if (SvTYPE(sv) == SVt_PVHV) {
724 HV* const hv = MUTABLE_HV(sv);
726 (void)hv_iterinit(hv);
727 while ((entry = hv_iternext(hv)))
728 count += do_chomp(retval, hv_iterval(hv,entry), chomping);
731 else if (SvREADONLY(sv)) {
732 Perl_croak_no_modify();
738 char *temp_buffer = NULL;
743 goto nope_free_nothing;
745 while (len && s[-1] == '\n') {
752 STRLEN rslen, rs_charlen;
753 const char *rsptr = SvPV_const(PL_rs, rslen);
755 rs_charlen = SvUTF8(PL_rs)
759 if (SvUTF8(PL_rs) != SvUTF8(sv)) {
760 /* Assumption is that rs is shorter than the scalar. */
762 /* RS is utf8, scalar is 8 bit. */
764 temp_buffer = (char*)bytes_from_utf8((U8*)rsptr,
767 /* Cannot downgrade, therefore cannot possibly match.
768 At this point, temp_buffer is not alloced, and
769 is the buffer inside PL_rs, so dont free it.
771 assert (temp_buffer == rsptr);
777 /* RS is 8 bit, scalar is utf8. */
778 temp_buffer = (char*)bytes_to_utf8((U8*)rsptr, &rslen);
792 if (memNE(s, rsptr, rslen))
797 SvPV_force_nomg_nolen(sv);
804 Safefree(temp_buffer);
806 SvREFCNT_dec(svrecode);
810 if (len && (!SvPOK(sv) || SvIsCOW(sv)))
811 s = SvPV_force_nomg(sv, len);
814 char * const send = s + len;
815 char * const start = s;
817 while (s > start && UTF8_IS_CONTINUATION(*s))
819 if (is_utf8_string((U8*)s, send - s)) {
820 sv_setpvn(retval, s, send - s);
822 SvCUR_set(sv, s - start);
832 sv_setpvn(retval, s, 1);
846 /* also used for: pp_schomp() */
851 const bool chomping = PL_op->op_type == OP_SCHOMP;
853 const size_t count = do_chomp(TARG, TOPs, chomping);
855 sv_setiv(TARG, count);
861 /* also used for: pp_chomp() */
865 dSP; dMARK; dTARGET; dORIGMARK;
866 const bool chomping = PL_op->op_type == OP_CHOMP;
870 count += do_chomp(TARG, *++MARK, chomping);
872 sv_setiv(TARG, count);
883 if (!PL_op->op_private) {
895 if (SvTHINKFIRST(sv))
896 sv_force_normal_flags(sv, SV_COW_DROP_PV|SV_IMMEDIATE_UNREF);
898 switch (SvTYPE(sv)) {
902 av_undef(MUTABLE_AV(sv));
905 hv_undef(MUTABLE_HV(sv));
908 if (cv_const_sv((const CV *)sv))
909 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
910 "Constant subroutine %" SVf " undefined",
911 SVfARG(CvANON((const CV *)sv)
912 ? newSVpvs_flags("(anonymous)", SVs_TEMP)
913 : sv_2mortal(newSVhek(
915 ? CvNAME_HEK((CV *)sv)
916 : GvENAME_HEK(CvGV((const CV *)sv))
921 /* let user-undef'd sub keep its identity */
922 cv_undef_flags(MUTABLE_CV(sv), CV_UNDEF_KEEP_NAME);
925 assert(isGV_with_GP(sv));
931 /* undef *Pkg::meth_name ... */
933 = GvCVu((const GV *)sv) && (stash = GvSTASH((const GV *)sv))
934 && HvENAME_get(stash);
936 if((stash = GvHV((const GV *)sv))) {
937 if(HvENAME_get(stash))
938 SvREFCNT_inc_simple_void_NN(sv_2mortal((SV *)stash));
942 SvREFCNT_inc_simple_void_NN(sv_2mortal(sv));
943 gp_free(MUTABLE_GV(sv));
945 GvGP_set(sv, gp_ref(gp));
946 #ifndef PERL_DONT_CREATE_GVSV
949 GvLINE(sv) = CopLINE(PL_curcop);
950 GvEGV(sv) = MUTABLE_GV(sv);
954 mro_package_moved(NULL, stash, (const GV *)sv, 0);
956 /* undef *Foo::ISA */
957 if( strEQ(GvNAME((const GV *)sv), "ISA")
958 && (stash = GvSTASH((const GV *)sv))
959 && (method_changed || HvENAME(stash)) )
960 mro_isa_changed_in(stash);
961 else if(method_changed)
962 mro_method_changed_in(
963 GvSTASH((const GV *)sv)
969 if (SvTYPE(sv) >= SVt_PV && SvPVX_const(sv) && SvLEN(sv)) {
983 /* common "slow" code for pp_postinc and pp_postdec */
986 S_postincdec_common(pTHX_ SV *sv, SV *targ)
990 PL_op->op_type == OP_POSTINC || PL_op->op_type == OP_I_POSTINC;
993 TARG = sv_newmortal();
1000 /* special case for undef: see thread at 2003-03/msg00536.html in archive */
1001 if (inc && !SvOK(TARG))
1008 /* also used for: pp_i_postinc() */
1015 /* special-case sv being a simple integer */
1016 if (LIKELY(((sv->sv_flags &
1017 (SVf_THINKFIRST|SVs_GMG|SVf_IVisUV|
1018 SVf_IOK|SVf_NOK|SVf_POK|SVp_NOK|SVp_POK|SVf_ROK))
1020 && SvIVX(sv) != IV_MAX)
1023 SvIV_set(sv, iv + 1);
1024 TARGi(iv, 0); /* arg not GMG, so can't be tainted */
1029 return S_postincdec_common(aTHX_ sv, TARG);
1033 /* also used for: pp_i_postdec() */
1040 /* special-case sv being a simple integer */
1041 if (LIKELY(((sv->sv_flags &
1042 (SVf_THINKFIRST|SVs_GMG|SVf_IVisUV|
1043 SVf_IOK|SVf_NOK|SVf_POK|SVp_NOK|SVp_POK|SVf_ROK))
1045 && SvIVX(sv) != IV_MIN)
1048 SvIV_set(sv, iv - 1);
1049 TARGi(iv, 0); /* arg not GMG, so can't be tainted */
1054 return S_postincdec_common(aTHX_ sv, TARG);
1058 /* Ordinary operators. */
1062 dSP; dATARGET; SV *svl, *svr;
1063 #ifdef PERL_PRESERVE_IVUV
1066 tryAMAGICbin_MG(pow_amg, AMGf_assign|AMGf_numeric);
1069 #ifdef PERL_PRESERVE_IVUV
1070 /* For integer to integer power, we do the calculation by hand wherever
1071 we're sure it is safe; otherwise we call pow() and try to convert to
1072 integer afterwards. */
1073 if (SvIV_please_nomg(svr) && SvIV_please_nomg(svl)) {
1081 const IV iv = SvIVX(svr);
1085 goto float_it; /* Can't do negative powers this way. */
1089 baseuok = SvUOK(svl);
1091 baseuv = SvUVX(svl);
1093 const IV iv = SvIVX(svl);
1096 baseuok = TRUE; /* effectively it's a UV now */
1098 baseuv = -iv; /* abs, baseuok == false records sign */
1101 /* now we have integer ** positive integer. */
1104 /* foo & (foo - 1) is zero only for a power of 2. */
1105 if (!(baseuv & (baseuv - 1))) {
1106 /* We are raising power-of-2 to a positive integer.
1107 The logic here will work for any base (even non-integer
1108 bases) but it can be less accurate than
1109 pow (base,power) or exp (power * log (base)) when the
1110 intermediate values start to spill out of the mantissa.
1111 With powers of 2 we know this can't happen.
1112 And powers of 2 are the favourite thing for perl
1113 programmers to notice ** not doing what they mean. */
1115 NV base = baseuok ? baseuv : -(NV)baseuv;
1120 while (power >>= 1) {
1128 SvIV_please_nomg(svr);
1131 unsigned int highbit = 8 * sizeof(UV);
1132 unsigned int diff = 8 * sizeof(UV);
1133 while (diff >>= 1) {
1135 if (baseuv >> highbit) {
1139 /* we now have baseuv < 2 ** highbit */
1140 if (power * highbit <= 8 * sizeof(UV)) {
1141 /* result will definitely fit in UV, so use UV math
1142 on same algorithm as above */
1145 const bool odd_power = cBOOL(power & 1);
1149 while (power >>= 1) {
1156 if (baseuok || !odd_power)
1157 /* answer is positive */
1159 else if (result <= (UV)IV_MAX)
1160 /* answer negative, fits in IV */
1161 SETi( -(IV)result );
1162 else if (result == (UV)IV_MIN)
1163 /* 2's complement assumption: special case IV_MIN */
1166 /* answer negative, doesn't fit */
1167 SETn( -(NV)result );
1175 NV right = SvNV_nomg(svr);
1176 NV left = SvNV_nomg(svl);
1179 #if defined(USE_LONG_DOUBLE) && defined(HAS_AIX_POWL_NEG_BASE_BUG)
1181 We are building perl with long double support and are on an AIX OS
1182 afflicted with a powl() function that wrongly returns NaNQ for any
1183 negative base. This was reported to IBM as PMR #23047-379 on
1184 03/06/2006. The problem exists in at least the following versions
1185 of AIX and the libm fileset, and no doubt others as well:
1187 AIX 4.3.3-ML10 bos.adt.libm 4.3.3.50
1188 AIX 5.1.0-ML04 bos.adt.libm 5.1.0.29
1189 AIX 5.2.0 bos.adt.libm 5.2.0.85
1191 So, until IBM fixes powl(), we provide the following workaround to
1192 handle the problem ourselves. Our logic is as follows: for
1193 negative bases (left), we use fmod(right, 2) to check if the
1194 exponent is an odd or even integer:
1196 - if odd, powl(left, right) == -powl(-left, right)
1197 - if even, powl(left, right) == powl(-left, right)
1199 If the exponent is not an integer, the result is rightly NaNQ, so
1200 we just return that (as NV_NAN).
1204 NV mod2 = Perl_fmod( right, 2.0 );
1205 if (mod2 == 1.0 || mod2 == -1.0) { /* odd integer */
1206 SETn( -Perl_pow( -left, right) );
1207 } else if (mod2 == 0.0) { /* even integer */
1208 SETn( Perl_pow( -left, right) );
1209 } else { /* fractional power */
1213 SETn( Perl_pow( left, right) );
1216 SETn( Perl_pow( left, right) );
1217 #endif /* HAS_AIX_POWL_NEG_BASE_BUG */
1219 #ifdef PERL_PRESERVE_IVUV
1221 SvIV_please_nomg(svr);
1229 dSP; dATARGET; SV *svl, *svr;
1230 tryAMAGICbin_MG(mult_amg, AMGf_assign|AMGf_numeric);
1234 #ifdef PERL_PRESERVE_IVUV
1236 /* special-case some simple common cases */
1237 if (!((svl->sv_flags|svr->sv_flags) & (SVf_IVisUV|SVs_GMG))) {
1239 U32 flags = (svl->sv_flags & svr->sv_flags);
1240 if (flags & SVf_IOK) {
1241 /* both args are simple IVs */
1246 topl = ((UV)il) >> (UVSIZE * 4 - 1);
1247 topr = ((UV)ir) >> (UVSIZE * 4 - 1);
1249 /* if both are in a range that can't under/overflow, do a
1250 * simple integer multiply: if the top halves(*) of both numbers
1251 * are 00...00 or 11...11, then it's safe.
1252 * (*) for 32-bits, the "top half" is the top 17 bits,
1253 * for 64-bits, its 33 bits */
1255 ((topl+1) | (topr+1))
1256 & ( (((UV)1) << (UVSIZE * 4 + 1)) - 2) /* 11..110 */
1259 TARGi(il * ir, 0); /* args not GMG, so can't be tainted */
1265 else if (flags & SVf_NOK) {
1266 /* both args are NVs */
1272 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
1273 !Perl_isnan(nl) && nl == (NV)(il = (IV)nl)
1274 && !Perl_isnan(nr) && nr == (NV)(ir = (IV)nr)
1276 nl == (NV)(il = (IV)nl) && nr == (NV)(ir = (IV)nr)
1279 /* nothing was lost by converting to IVs */
1283 # if defined(__sgi) && defined(USE_LONG_DOUBLE) && LONG_DOUBLEKIND == LONG_DOUBLE_IS_DOUBLEDOUBLE_128_BIT_BE_BE && NVSIZE == 16
1284 if (Perl_isinf(result)) {
1285 Zero((U8*)&result + 8, 8, U8);
1288 TARGn(result, 0); /* args not GMG, so can't be tainted */
1296 if (SvIV_please_nomg(svr)) {
1297 /* Unless the left argument is integer in range we are going to have to
1298 use NV maths. Hence only attempt to coerce the right argument if
1299 we know the left is integer. */
1300 /* Left operand is defined, so is it IV? */
1301 if (SvIV_please_nomg(svl)) {
1302 bool auvok = SvUOK(svl);
1303 bool buvok = SvUOK(svr);
1304 const UV topmask = (~ (UV)0) << (4 * sizeof (UV));
1305 const UV botmask = ~((~ (UV)0) << (4 * sizeof (UV)));
1314 const IV aiv = SvIVX(svl);
1317 auvok = TRUE; /* effectively it's a UV now */
1319 /* abs, auvok == false records sign */
1320 alow = (aiv == IV_MIN) ? (UV)aiv : (UV)(-aiv);
1326 const IV biv = SvIVX(svr);
1329 buvok = TRUE; /* effectively it's a UV now */
1331 /* abs, buvok == false records sign */
1332 blow = (biv == IV_MIN) ? (UV)biv : (UV)(-biv);
1336 /* If this does sign extension on unsigned it's time for plan B */
1337 ahigh = alow >> (4 * sizeof (UV));
1339 bhigh = blow >> (4 * sizeof (UV));
1341 if (ahigh && bhigh) {
1343 /* eg 32 bit is at least 0x10000 * 0x10000 == 0x100000000
1344 which is overflow. Drop to NVs below. */
1345 } else if (!ahigh && !bhigh) {
1346 /* eg 32 bit is at most 0xFFFF * 0xFFFF == 0xFFFE0001
1347 so the unsigned multiply cannot overflow. */
1348 const UV product = alow * blow;
1349 if (auvok == buvok) {
1350 /* -ve * -ve or +ve * +ve gives a +ve result. */
1354 } else if (product <= (UV)IV_MIN) {
1355 /* 2s complement assumption that (UV)-IV_MIN is correct. */
1356 /* -ve result, which could overflow an IV */
1358 /* can't negate IV_MIN, but there are aren't two
1359 * integers such that !ahigh && !bhigh, where the
1360 * product equals 0x800....000 */
1361 assert(product != (UV)IV_MIN);
1362 SETi( -(IV)product );
1364 } /* else drop to NVs below. */
1366 /* One operand is large, 1 small */
1369 /* swap the operands */
1371 bhigh = blow; /* bhigh now the temp var for the swap */
1375 /* now, ((ahigh * blow) << half_UV_len) + (alow * blow)
1376 multiplies can't overflow. shift can, add can, -ve can. */
1377 product_middle = ahigh * blow;
1378 if (!(product_middle & topmask)) {
1379 /* OK, (ahigh * blow) won't lose bits when we shift it. */
1381 product_middle <<= (4 * sizeof (UV));
1382 product_low = alow * blow;
1384 /* as for pp_add, UV + something mustn't get smaller.
1385 IIRC ANSI mandates this wrapping *behaviour* for
1386 unsigned whatever the actual representation*/
1387 product_low += product_middle;
1388 if (product_low >= product_middle) {
1389 /* didn't overflow */
1390 if (auvok == buvok) {
1391 /* -ve * -ve or +ve * +ve gives a +ve result. */
1393 SETu( product_low );
1395 } else if (product_low <= (UV)IV_MIN) {
1396 /* 2s complement assumption again */
1397 /* -ve result, which could overflow an IV */
1399 SETi(product_low == (UV)IV_MIN
1400 ? IV_MIN : -(IV)product_low);
1402 } /* else drop to NVs below. */
1404 } /* product_middle too large */
1405 } /* ahigh && bhigh */
1410 NV right = SvNV_nomg(svr);
1411 NV left = SvNV_nomg(svl);
1412 NV result = left * right;
1415 #if defined(__sgi) && defined(USE_LONG_DOUBLE) && LONG_DOUBLEKIND == LONG_DOUBLE_IS_DOUBLEDOUBLE_128_BIT_BE_BE && NVSIZE == 16
1416 if (Perl_isinf(result)) {
1417 Zero((U8*)&result + 8, 8, U8);
1427 dSP; dATARGET; SV *svl, *svr;
1428 tryAMAGICbin_MG(div_amg, AMGf_assign|AMGf_numeric);
1431 /* Only try to do UV divide first
1432 if ((SLOPPYDIVIDE is true) or
1433 (PERL_PRESERVE_IVUV is true and one or both SV is a UV too large
1435 The assumption is that it is better to use floating point divide
1436 whenever possible, only doing integer divide first if we can't be sure.
1437 If NV_PRESERVES_UV is true then we know at compile time that no UV
1438 can be too large to preserve, so don't need to compile the code to
1439 test the size of UVs. */
1441 #if defined(SLOPPYDIVIDE) || (defined(PERL_PRESERVE_IVUV) && !defined(NV_PRESERVES_UV))
1442 # define PERL_TRY_UV_DIVIDE
1443 /* ensure that 20./5. == 4. */
1446 #ifdef PERL_TRY_UV_DIVIDE
1447 if (SvIV_please_nomg(svr) && SvIV_please_nomg(svl)) {
1448 bool left_non_neg = SvUOK(svl);
1449 bool right_non_neg = SvUOK(svr);
1453 if (right_non_neg) {
1457 const IV biv = SvIVX(svr);
1460 right_non_neg = TRUE; /* effectively it's a UV now */
1463 right = (biv == IV_MIN) ? (UV)biv : (UV)(-biv);
1466 /* historically undef()/0 gives a "Use of uninitialized value"
1467 warning before dieing, hence this test goes here.
1468 If it were immediately before the second SvIV_please, then
1469 DIE() would be invoked before left was even inspected, so
1470 no inspection would give no warning. */
1472 DIE(aTHX_ "Illegal division by zero");
1478 const IV aiv = SvIVX(svl);
1481 left_non_neg = TRUE; /* effectively it's a UV now */
1484 left = (aiv == IV_MIN) ? (UV)aiv : (UV)(-aiv);
1490 /* For sloppy divide we always attempt integer division. */
1492 /* Otherwise we only attempt it if either or both operands
1493 would not be preserved by an NV. If both fit in NVs
1494 we fall through to the NV divide code below. However,
1495 as left >= right to ensure integer result here, we know that
1496 we can skip the test on the right operand - right big
1497 enough not to be preserved can't get here unless left is
1500 && (left > ((UV)1 << NV_PRESERVES_UV_BITS))
1503 /* Integer division can't overflow, but it can be imprecise. */
1504 const UV result = left / right;
1505 if (result * right == left) {
1506 SP--; /* result is valid */
1507 if (left_non_neg == right_non_neg) {
1508 /* signs identical, result is positive. */
1512 /* 2s complement assumption */
1513 if (result <= (UV)IV_MIN)
1514 SETi(result == (UV)IV_MIN ? IV_MIN : -(IV)result);
1516 /* It's exact but too negative for IV. */
1517 SETn( -(NV)result );
1520 } /* tried integer divide but it was not an integer result */
1521 } /* else (PERL_ABS(result) < 1.0) or (both UVs in range for NV) */
1522 } /* one operand wasn't SvIOK */
1523 #endif /* PERL_TRY_UV_DIVIDE */
1525 NV right = SvNV_nomg(svr);
1526 NV left = SvNV_nomg(svl);
1527 (void)POPs;(void)POPs;
1528 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
1529 if (! Perl_isnan(right) && right == 0.0)
1533 DIE(aTHX_ "Illegal division by zero");
1534 PUSHn( left / right );
1542 tryAMAGICbin_MG(modulo_amg, AMGf_assign|AMGf_numeric);
1546 bool left_neg = FALSE;
1547 bool right_neg = FALSE;
1548 bool use_double = FALSE;
1549 bool dright_valid = FALSE;
1552 SV * const svr = TOPs;
1553 SV * const svl = TOPm1s;
1554 if (SvIV_please_nomg(svr)) {
1555 right_neg = !SvUOK(svr);
1559 const IV biv = SvIVX(svr);
1562 right_neg = FALSE; /* effectively it's a UV now */
1564 right = (biv == IV_MIN) ? (UV)biv : (UV)(-biv);
1569 dright = SvNV_nomg(svr);
1570 right_neg = dright < 0;
1573 if (dright < UV_MAX_P1) {
1574 right = U_V(dright);
1575 dright_valid = TRUE; /* In case we need to use double below. */
1581 /* At this point use_double is only true if right is out of range for
1582 a UV. In range NV has been rounded down to nearest UV and
1583 use_double false. */
1584 if (!use_double && SvIV_please_nomg(svl)) {
1585 left_neg = !SvUOK(svl);
1589 const IV aiv = SvIVX(svl);
1592 left_neg = FALSE; /* effectively it's a UV now */
1594 left = (aiv == IV_MIN) ? (UV)aiv : (UV)(-aiv);
1599 dleft = SvNV_nomg(svl);
1600 left_neg = dleft < 0;
1604 /* This should be exactly the 5.6 behaviour - if left and right are
1605 both in range for UV then use U_V() rather than floor. */
1607 if (dleft < UV_MAX_P1) {
1608 /* right was in range, so is dleft, so use UVs not double.
1612 /* left is out of range for UV, right was in range, so promote
1613 right (back) to double. */
1615 /* The +0.5 is used in 5.6 even though it is not strictly
1616 consistent with the implicit +0 floor in the U_V()
1617 inside the #if 1. */
1618 dleft = Perl_floor(dleft + 0.5);
1621 dright = Perl_floor(dright + 0.5);
1632 DIE(aTHX_ "Illegal modulus zero");
1634 dans = Perl_fmod(dleft, dright);
1635 if ((left_neg != right_neg) && dans)
1636 dans = dright - dans;
1639 sv_setnv(TARG, dans);
1645 DIE(aTHX_ "Illegal modulus zero");
1648 if ((left_neg != right_neg) && ans)
1651 /* XXX may warn: unary minus operator applied to unsigned type */
1652 /* could change -foo to be (~foo)+1 instead */
1653 if (ans <= ~((UV)IV_MAX)+1)
1654 sv_setiv(TARG, ~ans+1);
1656 sv_setnv(TARG, -(NV)ans);
1659 sv_setuv(TARG, ans);
1671 bool infnan = FALSE;
1673 if (GIMME_V == G_ARRAY && PL_op->op_private & OPpREPEAT_DOLIST) {
1674 /* TODO: think of some way of doing list-repeat overloading ??? */
1679 if (UNLIKELY(PL_op->op_private & OPpREPEAT_DOLIST)) {
1680 /* The parser saw this as a list repeat, and there
1681 are probably several items on the stack. But we're
1682 in scalar/void context, and there's no pp_list to save us
1683 now. So drop the rest of the items -- robin@kitsite.com
1686 if (MARK + 1 < SP) {
1692 ASSUME(MARK + 1 == SP);
1694 MARK[1] = &PL_sv_undef;
1698 tryAMAGICbin_MG(repeat_amg, AMGf_assign);
1704 const UV uv = SvUV_nomg(sv);
1706 count = IV_MAX; /* The best we can do? */
1710 count = SvIV_nomg(sv);
1713 else if (SvNOKp(sv)) {
1714 const NV nv = SvNV_nomg(sv);
1715 infnan = Perl_isinfnan(nv);
1716 if (UNLIKELY(infnan)) {
1720 count = -1; /* An arbitrary negative integer */
1726 count = SvIV_nomg(sv);
1729 Perl_ck_warner(aTHX_ packWARN(WARN_NUMERIC),
1730 "Non-finite repeat count does nothing");
1731 } else if (count < 0) {
1733 Perl_ck_warner(aTHX_ packWARN(WARN_NUMERIC),
1734 "Negative repeat count does nothing");
1737 if (GIMME_V == G_ARRAY && PL_op->op_private & OPpREPEAT_DOLIST) {
1739 const SSize_t items = SP - MARK;
1740 const U8 mod = PL_op->op_flags & OPf_MOD;
1745 if ( items > SSize_t_MAX / count /* max would overflow */
1746 /* repeatcpy would overflow */
1747 || items > I32_MAX / (I32)sizeof(SV *)
1749 Perl_croak(aTHX_ "%s","Out of memory during list extend");
1750 max = items * count;
1755 if (mod && SvPADTMP(*SP)) {
1756 *SP = sv_mortalcopy(*SP);
1763 repeatcpy((char*)(MARK + items), (char*)MARK,
1764 items * sizeof(const SV *), count - 1);
1767 else if (count <= 0)
1770 else { /* Note: mark already snarfed by pp_list */
1771 SV * const tmpstr = POPs;
1776 sv_setsv_nomg(TARG, tmpstr);
1777 SvPV_force_nomg(TARG, len);
1778 isutf = DO_UTF8(TARG);
1785 if ( len > (MEM_SIZE_MAX-1) / (UV)count /* max would overflow */
1786 || len > (U32)I32_MAX /* repeatcpy would overflow */
1788 Perl_croak(aTHX_ "%s",
1789 "Out of memory during string extend");
1790 max = (UV)count * len + 1;
1793 repeatcpy(SvPVX(TARG) + len, SvPVX(TARG), len, count - 1);
1794 SvCUR_set(TARG, SvCUR(TARG) * count);
1796 *SvEND(TARG) = '\0';
1799 (void)SvPOK_only_UTF8(TARG);
1801 (void)SvPOK_only(TARG);
1810 dSP; dATARGET; bool useleft; SV *svl, *svr;
1811 tryAMAGICbin_MG(subtr_amg, AMGf_assign|AMGf_numeric);
1815 #ifdef PERL_PRESERVE_IVUV
1817 /* special-case some simple common cases */
1818 if (!((svl->sv_flags|svr->sv_flags) & (SVf_IVisUV|SVs_GMG))) {
1820 U32 flags = (svl->sv_flags & svr->sv_flags);
1821 if (flags & SVf_IOK) {
1822 /* both args are simple IVs */
1827 topl = ((UV)il) >> (UVSIZE * 8 - 2);
1828 topr = ((UV)ir) >> (UVSIZE * 8 - 2);
1830 /* if both are in a range that can't under/overflow, do a
1831 * simple integer subtract: if the top of both numbers
1832 * are 00 or 11, then it's safe */
1833 if (!( ((topl+1) | (topr+1)) & 2)) {
1835 TARGi(il - ir, 0); /* args not GMG, so can't be tainted */
1841 else if (flags & SVf_NOK) {
1842 /* both args are NVs */
1847 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
1848 !Perl_isnan(nl) && nl == (NV)(il = (IV)nl)
1849 && !Perl_isnan(nr) && nr == (NV)(ir = (IV)nr)
1851 nl == (NV)(il = (IV)nl) && nr == (NV)(ir = (IV)nr)
1854 /* nothing was lost by converting to IVs */
1857 TARGn(nl - nr, 0); /* args not GMG, so can't be tainted */
1865 useleft = USE_LEFT(svl);
1866 /* See comments in pp_add (in pp_hot.c) about Overflow, and how
1867 "bad things" happen if you rely on signed integers wrapping. */
1868 if (SvIV_please_nomg(svr)) {
1869 /* Unless the left argument is integer in range we are going to have to
1870 use NV maths. Hence only attempt to coerce the right argument if
1871 we know the left is integer. */
1878 a_valid = auvok = 1;
1879 /* left operand is undef, treat as zero. */
1881 /* Left operand is defined, so is it IV? */
1882 if (SvIV_please_nomg(svl)) {
1883 if ((auvok = SvUOK(svl)))
1886 const IV aiv = SvIVX(svl);
1889 auvok = 1; /* Now acting as a sign flag. */
1890 } else { /* 2s complement assumption for IV_MIN */
1891 auv = (aiv == IV_MIN) ? (UV)aiv : (UV)-aiv;
1898 bool result_good = 0;
1901 bool buvok = SvUOK(svr);
1906 const IV biv = SvIVX(svr);
1911 buv = (biv == IV_MIN) ? (UV)biv : (UV)-biv;
1913 /* ?uvok if value is >= 0. basically, flagged as UV if it's +ve,
1914 else "IV" now, independent of how it came in.
1915 if a, b represents positive, A, B negative, a maps to -A etc
1920 all UV maths. negate result if A negative.
1921 subtract if signs same, add if signs differ. */
1923 if (auvok ^ buvok) {
1932 /* Must get smaller */
1937 if (result <= buv) {
1938 /* result really should be -(auv-buv). as its negation
1939 of true value, need to swap our result flag */
1951 if (result <= (UV)IV_MIN)
1952 SETi(result == (UV)IV_MIN
1953 ? IV_MIN : -(IV)result);
1955 /* result valid, but out of range for IV. */
1956 SETn( -(NV)result );
1960 } /* Overflow, drop through to NVs. */
1964 useleft = USE_LEFT(svl);
1967 NV value = SvNV_nomg(svr);
1971 /* left operand is undef, treat as zero - value */
1975 SETn( SvNV_nomg(svl) - value );
1980 #define IV_BITS (IVSIZE * 8)
1982 static UV S_uv_shift(UV uv, int shift, bool left)
1988 if (shift >= IV_BITS) {
1991 return left ? uv << shift : uv >> shift;
1994 static IV S_iv_shift(IV iv, int shift, bool left)
2000 if (shift >= IV_BITS) {
2001 return iv < 0 && !left ? -1 : 0;
2003 return left ? iv << shift : iv >> shift;
2006 #define UV_LEFT_SHIFT(uv, shift) S_uv_shift(uv, shift, TRUE)
2007 #define UV_RIGHT_SHIFT(uv, shift) S_uv_shift(uv, shift, FALSE)
2008 #define IV_LEFT_SHIFT(iv, shift) S_iv_shift(iv, shift, TRUE)
2009 #define IV_RIGHT_SHIFT(iv, shift) S_iv_shift(iv, shift, FALSE)
2013 dSP; dATARGET; SV *svl, *svr;
2014 tryAMAGICbin_MG(lshift_amg, AMGf_assign|AMGf_numeric);
2018 const IV shift = SvIV_nomg(svr);
2019 if (PL_op->op_private & HINT_INTEGER) {
2020 SETi(IV_LEFT_SHIFT(SvIV_nomg(svl), shift));
2023 SETu(UV_LEFT_SHIFT(SvUV_nomg(svl), shift));
2031 dSP; dATARGET; SV *svl, *svr;
2032 tryAMAGICbin_MG(rshift_amg, AMGf_assign|AMGf_numeric);
2036 const IV shift = SvIV_nomg(svr);
2037 if (PL_op->op_private & HINT_INTEGER) {
2038 SETi(IV_RIGHT_SHIFT(SvIV_nomg(svl), shift));
2041 SETu(UV_RIGHT_SHIFT(SvUV_nomg(svl), shift));
2052 tryAMAGICbin_MG(lt_amg, AMGf_set|AMGf_numeric);
2056 (SvIOK_notUV(left) && SvIOK_notUV(right))
2057 ? (SvIVX(left) < SvIVX(right))
2058 : (do_ncmp(left, right) == -1)
2068 tryAMAGICbin_MG(gt_amg, AMGf_set|AMGf_numeric);
2072 (SvIOK_notUV(left) && SvIOK_notUV(right))
2073 ? (SvIVX(left) > SvIVX(right))
2074 : (do_ncmp(left, right) == 1)
2084 tryAMAGICbin_MG(le_amg, AMGf_set|AMGf_numeric);
2088 (SvIOK_notUV(left) && SvIOK_notUV(right))
2089 ? (SvIVX(left) <= SvIVX(right))
2090 : (do_ncmp(left, right) <= 0)
2100 tryAMAGICbin_MG(ge_amg, AMGf_set|AMGf_numeric);
2104 (SvIOK_notUV(left) && SvIOK_notUV(right))
2105 ? (SvIVX(left) >= SvIVX(right))
2106 : ( (do_ncmp(left, right) & 2) == 0)
2116 tryAMAGICbin_MG(ne_amg, AMGf_set|AMGf_numeric);
2120 (SvIOK_notUV(left) && SvIOK_notUV(right))
2121 ? (SvIVX(left) != SvIVX(right))
2122 : (do_ncmp(left, right) != 0)
2127 /* compare left and right SVs. Returns:
2131 * 2: left or right was a NaN
2134 Perl_do_ncmp(pTHX_ SV* const left, SV * const right)
2136 PERL_ARGS_ASSERT_DO_NCMP;
2137 #ifdef PERL_PRESERVE_IVUV
2138 /* Fortunately it seems NaN isn't IOK */
2139 if (SvIV_please_nomg(right) && SvIV_please_nomg(left)) {
2141 const IV leftiv = SvIVX(left);
2142 if (!SvUOK(right)) {
2143 /* ## IV <=> IV ## */
2144 const IV rightiv = SvIVX(right);
2145 return (leftiv > rightiv) - (leftiv < rightiv);
2147 /* ## IV <=> UV ## */
2149 /* As (b) is a UV, it's >=0, so it must be < */
2152 const UV rightuv = SvUVX(right);
2153 return ((UV)leftiv > rightuv) - ((UV)leftiv < rightuv);
2158 /* ## UV <=> UV ## */
2159 const UV leftuv = SvUVX(left);
2160 const UV rightuv = SvUVX(right);
2161 return (leftuv > rightuv) - (leftuv < rightuv);
2163 /* ## UV <=> IV ## */
2165 const IV rightiv = SvIVX(right);
2167 /* As (a) is a UV, it's >=0, so it cannot be < */
2170 const UV leftuv = SvUVX(left);
2171 return (leftuv > (UV)rightiv) - (leftuv < (UV)rightiv);
2174 NOT_REACHED; /* NOTREACHED */
2178 NV const rnv = SvNV_nomg(right);
2179 NV const lnv = SvNV_nomg(left);
2181 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2182 if (Perl_isnan(lnv) || Perl_isnan(rnv)) {
2185 return (lnv > rnv) - (lnv < rnv);
2204 tryAMAGICbin_MG(ncmp_amg, AMGf_numeric);
2207 value = do_ncmp(left, right);
2219 /* also used for: pp_sge() pp_sgt() pp_slt() */
2225 int amg_type = sle_amg;
2229 switch (PL_op->op_type) {
2248 tryAMAGICbin_MG(amg_type, AMGf_set);
2252 #ifdef USE_LOCALE_COLLATE
2253 (IN_LC_RUNTIME(LC_COLLATE))
2254 ? sv_cmp_locale_flags(left, right, 0)
2257 sv_cmp_flags(left, right, 0);
2258 SETs(boolSV(cmp * multiplier < rhs));
2266 tryAMAGICbin_MG(seq_amg, AMGf_set);
2269 SETs(boolSV(sv_eq_flags(left, right, 0)));
2277 tryAMAGICbin_MG(sne_amg, AMGf_set);
2280 SETs(boolSV(!sv_eq_flags(left, right, 0)));
2288 tryAMAGICbin_MG(scmp_amg, 0);
2292 #ifdef USE_LOCALE_COLLATE
2293 (IN_LC_RUNTIME(LC_COLLATE))
2294 ? sv_cmp_locale_flags(left, right, 0)
2297 sv_cmp_flags(left, right, 0);
2306 tryAMAGICbin_MG(band_amg, AMGf_assign);
2309 if (SvNIOKp(left) || SvNIOKp(right)) {
2310 const bool left_ro_nonnum = !SvNIOKp(left) && SvREADONLY(left);
2311 const bool right_ro_nonnum = !SvNIOKp(right) && SvREADONLY(right);
2312 if (PL_op->op_private & HINT_INTEGER) {
2313 const IV i = SvIV_nomg(left) & SvIV_nomg(right);
2317 const UV u = SvUV_nomg(left) & SvUV_nomg(right);
2320 if (left_ro_nonnum && left != TARG) SvNIOK_off(left);
2321 if (right_ro_nonnum) SvNIOK_off(right);
2324 do_vop(PL_op->op_type, TARG, left, right);
2334 tryAMAGICbin_MG(band_amg, AMGf_assign|AMGf_numarg);
2336 dATARGET; dPOPTOPssrl;
2337 if (PL_op->op_private & HINT_INTEGER) {
2338 const IV i = SvIV_nomg(left) & SvIV_nomg(right);
2342 const UV u = SvUV_nomg(left) & SvUV_nomg(right);
2352 tryAMAGICbin_MG(sband_amg, AMGf_assign);
2354 dATARGET; dPOPTOPssrl;
2355 do_vop(OP_BIT_AND, TARG, left, right);
2360 /* also used for: pp_bit_xor() */
2365 const int op_type = PL_op->op_type;
2367 tryAMAGICbin_MG((op_type == OP_BIT_OR ? bor_amg : bxor_amg), AMGf_assign);
2370 if (SvNIOKp(left) || SvNIOKp(right)) {
2371 const bool left_ro_nonnum = !SvNIOKp(left) && SvREADONLY(left);
2372 const bool right_ro_nonnum = !SvNIOKp(right) && SvREADONLY(right);
2373 if (PL_op->op_private & HINT_INTEGER) {
2374 const IV l = (USE_LEFT(left) ? SvIV_nomg(left) : 0);
2375 const IV r = SvIV_nomg(right);
2376 const IV result = op_type == OP_BIT_OR ? (l | r) : (l ^ r);
2380 const UV l = (USE_LEFT(left) ? SvUV_nomg(left) : 0);
2381 const UV r = SvUV_nomg(right);
2382 const UV result = op_type == OP_BIT_OR ? (l | r) : (l ^ r);
2385 if (left_ro_nonnum && left != TARG) SvNIOK_off(left);
2386 if (right_ro_nonnum) SvNIOK_off(right);
2389 do_vop(op_type, TARG, left, right);
2396 /* also used for: pp_nbit_xor() */
2401 const int op_type = PL_op->op_type;
2403 tryAMAGICbin_MG((op_type == OP_NBIT_OR ? bor_amg : bxor_amg),
2404 AMGf_assign|AMGf_numarg);
2406 dATARGET; dPOPTOPssrl;
2407 if (PL_op->op_private & HINT_INTEGER) {
2408 const IV l = (USE_LEFT(left) ? SvIV_nomg(left) : 0);
2409 const IV r = SvIV_nomg(right);
2410 const IV result = op_type == OP_NBIT_OR ? (l | r) : (l ^ r);
2414 const UV l = (USE_LEFT(left) ? SvUV_nomg(left) : 0);
2415 const UV r = SvUV_nomg(right);
2416 const UV result = op_type == OP_NBIT_OR ? (l | r) : (l ^ r);
2423 /* also used for: pp_sbit_xor() */
2428 const int op_type = PL_op->op_type;
2430 tryAMAGICbin_MG((op_type == OP_SBIT_OR ? sbor_amg : sbxor_amg),
2433 dATARGET; dPOPTOPssrl;
2434 do_vop(op_type == OP_SBIT_OR ? OP_BIT_OR : OP_BIT_XOR, TARG, left,
2440 PERL_STATIC_INLINE bool
2441 S_negate_string(pTHX)
2446 SV * const sv = TOPs;
2447 if (!SvPOKp(sv) || SvNIOK(sv) || (!SvPOK(sv) && SvNIOKp(sv)))
2449 s = SvPV_nomg_const(sv, len);
2450 if (isIDFIRST(*s)) {
2451 sv_setpvs(TARG, "-");
2454 else if (*s == '+' || (*s == '-' && !looks_like_number(sv))) {
2455 sv_setsv_nomg(TARG, sv);
2456 *SvPV_force_nomg(TARG, len) = *s == '-' ? '+' : '-';
2466 tryAMAGICun_MG(neg_amg, AMGf_numeric);
2467 if (S_negate_string(aTHX)) return NORMAL;
2469 SV * const sv = TOPs;
2472 /* It's publicly an integer */
2475 if (SvIVX(sv) == IV_MIN) {
2476 /* 2s complement assumption. */
2477 SETi(SvIVX(sv)); /* special case: -((UV)IV_MAX+1) ==
2481 else if (SvUVX(sv) <= IV_MAX) {
2486 else if (SvIVX(sv) != IV_MIN) {
2490 #ifdef PERL_PRESERVE_IVUV
2497 if (SvNIOKp(sv) && (SvNIOK(sv) || !SvPOK(sv)))
2498 SETn(-SvNV_nomg(sv));
2499 else if (SvPOKp(sv) && SvIV_please_nomg(sv))
2500 goto oops_its_an_int;
2502 SETn(-SvNV_nomg(sv));
2512 tryAMAGICun_MG(not_amg, AMGf_set);
2514 *PL_stack_sp = boolSV(!SvTRUE_nomg_NN(sv));
2519 S_scomplement(pTHX_ SV *targ, SV *sv)
2525 sv_copypv_nomg(TARG, sv);
2526 tmps = (U8*)SvPV_nomg(TARG, len);
2529 if (len && ! utf8_to_bytes(tmps, &len)) {
2530 Perl_croak(aTHX_ fatal_above_ff_msg, PL_op_desc[PL_op->op_type]);
2541 for ( ; anum && (unsigned long)tmps % sizeof(long); anum--, tmps++)
2544 for ( ; anum >= (I32)sizeof(long); anum -= (I32)sizeof(long), tmpl++)
2549 for ( ; anum > 0; anum--, tmps++)
2556 tryAMAGICun_MG(compl_amg, AMGf_numeric);
2560 if (PL_op->op_private & HINT_INTEGER) {
2561 const IV i = ~SvIV_nomg(sv);
2565 const UV u = ~SvUV_nomg(sv);
2570 S_scomplement(aTHX_ TARG, sv);
2580 tryAMAGICun_MG(compl_amg, AMGf_numeric|AMGf_numarg);
2583 if (PL_op->op_private & HINT_INTEGER) {
2584 const IV i = ~SvIV_nomg(sv);
2588 const UV u = ~SvUV_nomg(sv);
2598 tryAMAGICun_MG(scompl_amg, AMGf_numeric);
2601 S_scomplement(aTHX_ TARG, sv);
2607 /* integer versions of some of the above */
2612 tryAMAGICbin_MG(mult_amg, AMGf_assign);
2615 SETi( left * right );
2624 tryAMAGICbin_MG(div_amg, AMGf_assign);
2627 IV value = SvIV_nomg(right);
2629 DIE(aTHX_ "Illegal division by zero");
2630 num = SvIV_nomg(left);
2632 /* avoid FPE_INTOVF on some platforms when num is IV_MIN */
2636 value = num / value;
2644 /* This is the vanilla old i_modulo. */
2646 tryAMAGICbin_MG(modulo_amg, AMGf_assign);
2650 DIE(aTHX_ "Illegal modulus zero");
2651 /* avoid FPE_INTOVF on some platforms when left is IV_MIN */
2655 SETi( left % right );
2660 #if defined(__GLIBC__) && IVSIZE == 8 \
2661 && ( __GLIBC__ < 2 || (__GLIBC__ == 2 && __GLIBC_MINOR__ < 8))
2663 PP(pp_i_modulo_glibc_bugfix)
2665 /* This is the i_modulo with the workaround for the _moddi3 bug
2666 * in (at least) glibc 2.2.5 (the PERL_ABS() the workaround).
2667 * See below for pp_i_modulo. */
2669 tryAMAGICbin_MG(modulo_amg, AMGf_assign);
2673 DIE(aTHX_ "Illegal modulus zero");
2674 /* avoid FPE_INTOVF on some platforms when left is IV_MIN */
2678 SETi( left % PERL_ABS(right) );
2687 tryAMAGICbin_MG(add_amg, AMGf_assign);
2689 dPOPTOPiirl_ul_nomg;
2690 SETi( left + right );
2698 tryAMAGICbin_MG(subtr_amg, AMGf_assign);
2700 dPOPTOPiirl_ul_nomg;
2701 SETi( left - right );
2709 tryAMAGICbin_MG(lt_amg, AMGf_set);
2712 SETs(boolSV(left < right));
2720 tryAMAGICbin_MG(gt_amg, AMGf_set);
2723 SETs(boolSV(left > right));
2731 tryAMAGICbin_MG(le_amg, AMGf_set);
2734 SETs(boolSV(left <= right));
2742 tryAMAGICbin_MG(ge_amg, AMGf_set);
2745 SETs(boolSV(left >= right));
2753 tryAMAGICbin_MG(eq_amg, AMGf_set);
2756 SETs(boolSV(left == right));
2764 tryAMAGICbin_MG(ne_amg, AMGf_set);
2767 SETs(boolSV(left != right));
2775 tryAMAGICbin_MG(ncmp_amg, 0);
2782 else if (left < right)
2794 tryAMAGICun_MG(neg_amg, 0);
2795 if (S_negate_string(aTHX)) return NORMAL;
2797 SV * const sv = TOPs;
2798 IV const i = SvIV_nomg(sv);
2804 /* High falutin' math. */
2809 tryAMAGICbin_MG(atan2_amg, 0);
2812 SETn(Perl_atan2(left, right));
2818 /* also used for: pp_cos() pp_exp() pp_log() pp_sqrt() */
2823 int amg_type = fallback_amg;
2824 const char *neg_report = NULL;
2825 const int op_type = PL_op->op_type;
2828 case OP_SIN: amg_type = sin_amg; break;
2829 case OP_COS: amg_type = cos_amg; break;
2830 case OP_EXP: amg_type = exp_amg; break;
2831 case OP_LOG: amg_type = log_amg; neg_report = "log"; break;
2832 case OP_SQRT: amg_type = sqrt_amg; neg_report = "sqrt"; break;
2835 assert(amg_type != fallback_amg);
2837 tryAMAGICun_MG(amg_type, 0);
2839 SV * const arg = TOPs;
2840 const NV value = SvNV_nomg(arg);
2846 if (neg_report) { /* log or sqrt */
2848 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2849 ! Perl_isnan(value) &&
2851 (op_type == OP_LOG ? (value <= 0.0) : (value < 0.0))) {
2852 SET_NUMERIC_STANDARD();
2853 /* diag_listed_as: Can't take log of %g */
2854 DIE(aTHX_ "Can't take %s of %" NVgf, neg_report, value);
2859 case OP_SIN: result = Perl_sin(value); break;
2860 case OP_COS: result = Perl_cos(value); break;
2861 case OP_EXP: result = Perl_exp(value); break;
2862 case OP_LOG: result = Perl_log(value); break;
2863 case OP_SQRT: result = Perl_sqrt(value); break;
2870 /* Support Configure command-line overrides for rand() functions.
2871 After 5.005, perhaps we should replace this by Configure support
2872 for drand48(), random(), or rand(). For 5.005, though, maintain
2873 compatibility by calling rand() but allow the user to override it.
2874 See INSTALL for details. --Andy Dougherty 15 July 1998
2876 /* Now it's after 5.005, and Configure supports drand48() and random(),
2877 in addition to rand(). So the overrides should not be needed any more.
2878 --Jarkko Hietaniemi 27 September 1998
2883 if (!PL_srand_called) {
2884 (void)seedDrand01((Rand_seed_t)seed());
2885 PL_srand_called = TRUE;
2897 SV * const sv = POPs;
2903 /* 1 of 2 things can be carried through SvNV, SP or TARG, SP was carried */
2904 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2905 if (! Perl_isnan(value) && value == 0.0)
2915 sv_setnv_mg(TARG, value);
2926 if (MAXARG >= 1 && (TOPs || POPs)) {
2933 pv = SvPV(top, len);
2934 flags = grok_number(pv, len, &anum);
2936 if (!(flags & IS_NUMBER_IN_UV)) {
2937 Perl_ck_warner_d(aTHX_ packWARN(WARN_OVERFLOW),
2938 "Integer overflow in srand");
2946 (void)seedDrand01((Rand_seed_t)anum);
2947 PL_srand_called = TRUE;
2951 /* Historically srand always returned true. We can avoid breaking
2953 sv_setpvs(TARG, "0 but true");
2962 tryAMAGICun_MG(int_amg, AMGf_numeric);
2964 SV * const sv = TOPs;
2965 const IV iv = SvIV_nomg(sv);
2966 /* XXX it's arguable that compiler casting to IV might be subtly
2967 different from modf (for numbers inside (IV_MIN,UV_MAX)) in which
2968 else preferring IV has introduced a subtle behaviour change bug. OTOH
2969 relying on floating point to be accurate is a bug. */
2974 else if (SvIOK(sv)) {
2976 SETu(SvUV_nomg(sv));
2981 const NV value = SvNV_nomg(sv);
2982 if (UNLIKELY(Perl_isinfnan(value)))
2984 else if (value >= 0.0) {
2985 if (value < (NV)UV_MAX + 0.5) {
2988 SETn(Perl_floor(value));
2992 if (value > (NV)IV_MIN - 0.5) {
2995 SETn(Perl_ceil(value));
3006 tryAMAGICun_MG(abs_amg, AMGf_numeric);
3008 SV * const sv = TOPs;
3009 /* This will cache the NV value if string isn't actually integer */
3010 const IV iv = SvIV_nomg(sv);
3015 else if (SvIOK(sv)) {
3016 /* IVX is precise */
3018 SETu(SvUV_nomg(sv)); /* force it to be numeric only */
3026 /* 2s complement assumption. Also, not really needed as
3027 IV_MIN and -IV_MIN should both be %100...00 and NV-able */
3033 const NV value = SvNV_nomg(sv);
3044 /* also used for: pp_hex() */
3050 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES;
3054 SV* const sv = TOPs;
3056 tmps = (SvPV_const(sv, len));
3058 /* If Unicode, try to downgrade
3059 * If not possible, croak. */
3060 SV* const tsv = sv_2mortal(newSVsv(sv));
3063 sv_utf8_downgrade(tsv, FALSE);
3064 tmps = SvPV_const(tsv, len);
3066 if (PL_op->op_type == OP_HEX)
3069 while (*tmps && len && isSPACE(*tmps))
3073 if (isALPHA_FOLD_EQ(*tmps, 'x')) {
3075 result_uv = grok_hex (tmps, &len, &flags, &result_nv);
3077 else if (isALPHA_FOLD_EQ(*tmps, 'b'))
3078 result_uv = grok_bin (tmps, &len, &flags, &result_nv);
3080 result_uv = grok_oct (tmps, &len, &flags, &result_nv);
3082 if (flags & PERL_SCAN_GREATER_THAN_UV_MAX) {
3097 SV * const sv = TOPs;
3099 U32 in_bytes = IN_BYTES;
3100 /* Simplest case shortcut:
3101 * set svflags to just the SVf_POK|SVs_GMG|SVf_UTF8 from the SV,
3102 * with the SVf_UTF8 flag inverted if under 'use bytes' (HINT_BYTES
3105 U32 svflags = (SvFLAGS(sv) ^ (in_bytes << 26)) & (SVf_POK|SVs_GMG|SVf_UTF8);
3107 STATIC_ASSERT_STMT(SVf_UTF8 == (HINT_BYTES << 26));
3110 if (LIKELY(svflags == SVf_POK))
3113 if (svflags & SVs_GMG)
3118 if (!IN_BYTES) { /* reread to avoid using an C auto/register */
3119 if ((SvFLAGS(sv) & (SVf_POK|SVf_UTF8)) == SVf_POK)
3121 if ( SvPOK(sv) && (PL_op->op_private & OPpTRUEBOOL)) {
3122 /* no need to convert from bytes to chars */
3126 len = sv_len_utf8_nomg(sv);
3129 /* unrolled SvPV_nomg_const(sv,len) */
3130 if (SvPOK_nog(sv)) {
3133 if (PL_op->op_private & OPpTRUEBOOL) {
3135 SETs(len ? &PL_sv_yes : &PL_sv_zero);
3140 (void)sv_2pv_flags(sv, &len, 0|SV_CONST_RETURN);
3143 TARGi((IV)(len), 1);
3146 if (!SvPADTMP(TARG)) {
3147 /* OPpTARGET_MY: targ is var in '$lex = length()' */
3152 /* TARG is on stack at this point and is overwriten by SETs.
3153 * This branch is the odd one out, so put TARG by default on
3154 * stack earlier to let local SP go out of liveness sooner */
3157 return NORMAL; /* no putback, SP didn't move in this opcode */
3161 /* Returns false if substring is completely outside original string.
3162 No length is indicated by len_iv = 0 and len_is_uv = 0. len_is_uv must
3163 always be true for an explicit 0.
3166 Perl_translate_substr_offsets( STRLEN curlen, IV pos1_iv,
3167 bool pos1_is_uv, IV len_iv,
3168 bool len_is_uv, STRLEN *posp,
3174 PERL_ARGS_ASSERT_TRANSLATE_SUBSTR_OFFSETS;
3176 if (!pos1_is_uv && pos1_iv < 0 && curlen) {
3177 pos1_is_uv = curlen-1 > ~(UV)pos1_iv;
3180 if ((pos1_is_uv || pos1_iv > 0) && (UV)pos1_iv > curlen)
3183 if (len_iv || len_is_uv) {
3184 if (!len_is_uv && len_iv < 0) {
3185 pos2_iv = curlen + len_iv;
3187 pos2_is_uv = curlen-1 > ~(UV)len_iv;
3190 } else { /* len_iv >= 0 */
3191 if (!pos1_is_uv && pos1_iv < 0) {
3192 pos2_iv = pos1_iv + len_iv;
3193 pos2_is_uv = (UV)len_iv > (UV)IV_MAX;
3195 if ((UV)len_iv > curlen-(UV)pos1_iv)
3198 pos2_iv = pos1_iv+len_iv;
3208 if (!pos2_is_uv && pos2_iv < 0) {
3209 if (!pos1_is_uv && pos1_iv < 0)
3213 else if (!pos1_is_uv && pos1_iv < 0)
3216 if ((UV)pos2_iv < (UV)pos1_iv)
3218 if ((UV)pos2_iv > curlen)
3221 /* pos1_iv and pos2_iv both in 0..curlen, so the cast is safe */
3222 *posp = (STRLEN)( (UV)pos1_iv );
3223 *lenp = (STRLEN)( (UV)pos2_iv - (UV)pos1_iv );
3240 I32 lvalue = PL_op->op_flags & OPf_MOD || LVRET;
3241 const bool rvalue = (GIMME_V != G_VOID);
3244 const char *repl = NULL;
3246 int num_args = PL_op->op_private & 7;
3247 bool repl_need_utf8_upgrade = FALSE;
3251 if(!(repl_sv = POPs)) num_args--;
3253 if ((len_sv = POPs)) {
3254 len_iv = SvIV(len_sv);
3255 len_is_uv = len_iv ? SvIOK_UV(len_sv) : 1;
3260 pos1_iv = SvIV(pos_sv);
3261 pos1_is_uv = SvIOK_UV(pos_sv);
3263 if (PL_op->op_private & OPpSUBSTR_REPL_FIRST) {
3267 if (lvalue && !repl_sv) {
3269 ret = sv_2mortal(newSV_type(SVt_PVLV)); /* Not TARG RT#67838 */
3270 sv_magic(ret, NULL, PERL_MAGIC_substr, NULL, 0);
3272 LvTARG(ret) = SvREFCNT_inc_simple(sv);
3274 pos1_is_uv || pos1_iv >= 0
3275 ? (STRLEN)(UV)pos1_iv
3276 : (LvFLAGS(ret) |= LVf_NEG_OFF, (STRLEN)(UV)-pos1_iv);
3278 len_is_uv || len_iv > 0
3279 ? (STRLEN)(UV)len_iv
3280 : (LvFLAGS(ret) |= LVf_NEG_LEN, (STRLEN)(UV)-len_iv);
3282 PUSHs(ret); /* avoid SvSETMAGIC here */
3286 repl = SvPV_const(repl_sv, repl_len);
3289 Perl_ck_warner(aTHX_ packWARN(WARN_SUBSTR),
3290 "Attempt to use reference as lvalue in substr"
3292 tmps = SvPV_force_nomg(sv, curlen);
3293 if (DO_UTF8(repl_sv) && repl_len) {
3295 /* Upgrade the dest, and recalculate tmps in case the buffer
3296 * got reallocated; curlen may also have been changed */
3297 sv_utf8_upgrade_nomg(sv);
3298 tmps = SvPV_nomg(sv, curlen);
3301 else if (DO_UTF8(sv))
3302 repl_need_utf8_upgrade = TRUE;
3304 else tmps = SvPV_const(sv, curlen);
3306 utf8_curlen = sv_or_pv_len_utf8(sv, tmps, curlen);
3307 if (utf8_curlen == curlen)
3310 curlen = utf8_curlen;
3316 STRLEN pos, len, byte_len, byte_pos;
3318 if (!translate_substr_offsets(
3319 curlen, pos1_iv, pos1_is_uv, len_iv, len_is_uv, &pos, &len
3323 byte_pos = utf8_curlen
3324 ? sv_or_pv_pos_u2b(sv, tmps, pos, &byte_len) : pos;
3329 SvTAINTED_off(TARG); /* decontaminate */
3330 SvUTF8_off(TARG); /* decontaminate */
3331 sv_setpvn(TARG, tmps, byte_len);
3332 #ifdef USE_LOCALE_COLLATE
3333 sv_unmagic(TARG, PERL_MAGIC_collxfrm);
3340 SV* repl_sv_copy = NULL;
3342 if (repl_need_utf8_upgrade) {
3343 repl_sv_copy = newSVsv(repl_sv);
3344 sv_utf8_upgrade(repl_sv_copy);
3345 repl = SvPV_const(repl_sv_copy, repl_len);
3349 sv_insert_flags(sv, byte_pos, byte_len, repl, repl_len, 0);
3350 SvREFCNT_dec(repl_sv_copy);
3353 if (PL_op->op_private & OPpSUBSTR_REPL_FIRST)
3363 Perl_croak(aTHX_ "substr outside of string");
3364 Perl_ck_warner(aTHX_ packWARN(WARN_SUBSTR), "substr outside of string");
3371 const IV size = POPi;
3372 SV* offsetsv = POPs;
3373 SV * const src = POPs;
3374 const I32 lvalue = PL_op->op_flags & OPf_MOD || LVRET;
3380 /* extract a STRLEN-ranged integer value from offsetsv into offset,
3381 * or flag that its out of range */
3383 IV iv = SvIV(offsetsv);
3385 /* avoid a large UV being wrapped to a negative value */
3386 if (SvIOK_UV(offsetsv) && SvUVX(offsetsv) > (UV)IV_MAX)
3387 errflags = LVf_OUT_OF_RANGE;
3389 errflags = (LVf_NEG_OFF|LVf_OUT_OF_RANGE);
3390 #if PTRSIZE < IVSIZE
3391 else if (iv > Size_t_MAX)
3392 errflags = LVf_OUT_OF_RANGE;
3395 offset = (STRLEN)iv;
3398 retuv = errflags ? 0 : do_vecget(src, offset, size);
3400 if (lvalue) { /* it's an lvalue! */
3401 ret = sv_2mortal(newSV_type(SVt_PVLV)); /* Not TARG RT#67838 */
3402 sv_magic(ret, NULL, PERL_MAGIC_vec, NULL, 0);
3404 LvTARG(ret) = SvREFCNT_inc_simple(src);
3405 LvTARGOFF(ret) = offset;
3406 LvTARGLEN(ret) = size;
3407 LvFLAGS(ret) = errflags;
3411 SvTAINTED_off(TARG); /* decontaminate */
3415 sv_setuv(ret, retuv);
3423 /* also used for: pp_rindex() */
3436 const char *little_p;
3439 const bool is_index = PL_op->op_type == OP_INDEX;
3440 const bool threeargs = MAXARG >= 3 && (TOPs || ((void)POPs,0));
3446 big_p = SvPV_const(big, biglen);
3447 little_p = SvPV_const(little, llen);
3449 big_utf8 = DO_UTF8(big);
3450 little_utf8 = DO_UTF8(little);
3451 if (big_utf8 ^ little_utf8) {
3452 /* One needs to be upgraded. */
3454 /* Well, maybe instead we might be able to downgrade the small
3456 char * const pv = (char*)bytes_from_utf8((U8 *)little_p, &llen,
3459 /* If the large string is ISO-8859-1, and it's not possible to
3460 convert the small string to ISO-8859-1, then there is no
3461 way that it could be found anywhere by index. */
3466 /* At this point, pv is a malloc()ed string. So donate it to temp
3467 to ensure it will get free()d */
3468 little = temp = newSV(0);
3469 sv_usepvn(temp, pv, llen);
3470 little_p = SvPVX(little);
3472 temp = newSVpvn(little_p, llen);
3474 sv_utf8_upgrade(temp);
3476 little_p = SvPV_const(little, llen);
3479 if (SvGAMAGIC(big)) {
3480 /* Life just becomes a lot easier if I use a temporary here.
3481 Otherwise I need to avoid calls to sv_pos_u2b(), which (dangerously)
3482 will trigger magic and overloading again, as will fbm_instr()
3484 big = newSVpvn_flags(big_p, biglen,
3485 SVs_TEMP | (big_utf8 ? SVf_UTF8 : 0));
3488 if (SvGAMAGIC(little) || (is_index && !SvOK(little))) {
3489 /* index && SvOK() is a hack. fbm_instr() calls SvPV_const, which will
3490 warn on undef, and we've already triggered a warning with the
3491 SvPV_const some lines above. We can't remove that, as we need to
3492 call some SvPV to trigger overloading early and find out if the
3494 This is all getting too messy. The API isn't quite clean enough,
3495 because data access has side effects.
3497 little = newSVpvn_flags(little_p, llen,
3498 SVs_TEMP | (little_utf8 ? SVf_UTF8 : 0));
3499 little_p = SvPVX(little);
3503 offset = is_index ? 0 : biglen;
3505 if (big_utf8 && offset > 0)
3506 offset = sv_pos_u2b_flags(big, offset, 0, SV_CONST_RETURN);
3512 else if (offset > (SSize_t)biglen)
3514 if (!(little_p = is_index
3515 ? fbm_instr((unsigned char*)big_p + offset,
3516 (unsigned char*)big_p + biglen, little, 0)
3517 : rninstr(big_p, big_p + offset,
3518 little_p, little_p + llen)))
3521 retval = little_p - big_p;
3522 if (retval > 1 && big_utf8)
3523 retval = sv_pos_b2u_flags(big, retval, SV_CONST_RETURN);
3528 /* OPpTRUEBOOL indicates an '== -1' has been optimised away */
3529 if (PL_op->op_private & OPpTRUEBOOL) {
3530 PUSHs( ((retval != -1) ^ cBOOL(PL_op->op_private & OPpINDEX_BOOLNEG))
3531 ? &PL_sv_yes : &PL_sv_no);
3532 if (PL_op->op_private & OPpTARGET_MY)
3533 /* $lex = (index() == -1) */
3534 sv_setsv(TARG, TOPs);
3543 dSP; dMARK; dORIGMARK; dTARGET;
3544 SvTAINTED_off(TARG);
3545 do_sprintf(TARG, SP-MARK, MARK+1);
3546 TAINT_IF(SvTAINTED(TARG));
3558 const U8 *s = (U8*)SvPV_const(argsv, len);
3561 ? (len ? utf8n_to_uvchr(s, len, 0, UTF8_ALLOW_ANYUV) : 0)
3575 if (UNLIKELY(SvAMAGIC(top)))
3577 if (UNLIKELY(isinfnansv(top)))
3578 Perl_croak(aTHX_ "Cannot chr %" NVgf, SvNV(top));
3580 if (!IN_BYTES /* under bytes, chr(-1) eq chr(0xff), etc. */
3581 && ((SvIOKp(top) && !SvIsUV(top) && SvIV_nomg(top) < 0)
3583 ((SvNOKp(top) || (SvOK(top) && !SvIsUV(top)))
3584 && SvNV_nomg(top) < 0.0)))
3586 if (ckWARN(WARN_UTF8)) {
3587 if (SvGMAGICAL(top)) {
3588 SV *top2 = sv_newmortal();
3589 sv_setsv_nomg(top2, top);
3592 Perl_warner(aTHX_ packWARN(WARN_UTF8),
3593 "Invalid negative number (%" SVf ") in chr", SVfARG(top));
3595 value = UNICODE_REPLACEMENT;
3597 value = SvUV_nomg(top);
3601 SvUPGRADE(TARG,SVt_PV);
3603 if (value > 255 && !IN_BYTES) {
3604 SvGROW(TARG, (STRLEN)UVCHR_SKIP(value)+1);
3605 tmps = (char*)uvchr_to_utf8_flags((U8*)SvPVX(TARG), value, 0);
3606 SvCUR_set(TARG, tmps - SvPVX_const(TARG));
3608 (void)SvPOK_only(TARG);
3617 *tmps++ = (char)value;
3619 (void)SvPOK_only(TARG);
3631 const char *tmps = SvPV_const(left, len);
3633 if (DO_UTF8(left)) {
3634 /* If Unicode, try to downgrade.
3635 * If not possible, croak.
3636 * Yes, we made this up. */
3637 SV* const tsv = newSVpvn_flags(tmps, len, SVf_UTF8|SVs_TEMP);
3639 sv_utf8_downgrade(tsv, FALSE);
3640 tmps = SvPV_const(tsv, len);
3642 # ifdef USE_ITHREADS
3644 if (!PL_reentrant_buffer->_crypt_struct_buffer) {
3645 /* This should be threadsafe because in ithreads there is only
3646 * one thread per interpreter. If this would not be true,
3647 * we would need a mutex to protect this malloc. */
3648 PL_reentrant_buffer->_crypt_struct_buffer =
3649 (struct crypt_data *)safemalloc(sizeof(struct crypt_data));
3650 #if defined(__GLIBC__) || defined(__EMX__)
3651 if (PL_reentrant_buffer->_crypt_struct_buffer) {
3652 PL_reentrant_buffer->_crypt_struct_buffer->initialized = 0;
3653 /* work around glibc-2.2.5 bug */
3654 PL_reentrant_buffer->_crypt_struct_buffer->current_saltbits = 0;
3658 # endif /* HAS_CRYPT_R */
3659 # endif /* USE_ITHREADS */
3661 sv_setpv(TARG, fcrypt(tmps, SvPV_nolen_const(right)));
3663 sv_setpv(TARG, PerlProc_crypt(tmps, SvPV_nolen_const(right)));
3670 "The crypt() function is unimplemented due to excessive paranoia.");
3674 /* Generally UTF-8 and UTF-EBCDIC are indistinguishable at this level. So
3675 * most comments below say UTF-8, when in fact they mean UTF-EBCDIC as well */
3678 /* also used for: pp_lcfirst() */
3682 /* Actually is both lcfirst() and ucfirst(). Only the first character
3683 * changes. This means that possibly we can change in-place, ie., just
3684 * take the source and change that one character and store it back, but not
3685 * if read-only etc, or if the length changes */
3689 STRLEN slen; /* slen is the byte length of the whole SV. */
3692 bool inplace; /* ? Convert first char only, in-place */
3693 bool doing_utf8 = FALSE; /* ? using utf8 */
3694 bool convert_source_to_utf8 = FALSE; /* ? need to convert */
3695 const int op_type = PL_op->op_type;
3698 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
3699 STRLEN ulen; /* ulen is the byte length of the original Unicode character
3700 * stored as UTF-8 at s. */
3701 STRLEN tculen; /* tculen is the byte length of the freshly titlecased (or
3702 * lowercased) character stored in tmpbuf. May be either
3703 * UTF-8 or not, but in either case is the number of bytes */
3705 s = (const U8*)SvPV_const(source, slen);
3707 /* We may be able to get away with changing only the first character, in
3708 * place, but not if read-only, etc. Later we may discover more reasons to
3709 * not convert in-place. */
3710 inplace = !SvREADONLY(source) && SvPADTMP(source);
3712 /* First calculate what the changed first character should be. This affects
3713 * whether we can just swap it out, leaving the rest of the string unchanged,
3714 * or even if have to convert the dest to UTF-8 when the source isn't */
3716 if (! slen) { /* If empty */
3717 need = 1; /* still need a trailing NUL */
3720 else if (DO_UTF8(source)) { /* Is the source utf8? */
3723 if (op_type == OP_UCFIRST) {
3724 #ifdef USE_LOCALE_CTYPE
3725 _toTITLE_utf8_flags(s, s +slen, tmpbuf, &tculen, IN_LC_RUNTIME(LC_CTYPE));
3727 _toTITLE_utf8_flags(s, s +slen, tmpbuf, &tculen, 0);
3731 #ifdef USE_LOCALE_CTYPE
3732 _toLOWER_utf8_flags(s, s + slen, tmpbuf, &tculen, IN_LC_RUNTIME(LC_CTYPE));
3734 _toLOWER_utf8_flags(s, s + slen, tmpbuf, &tculen, 0);
3738 /* we can't do in-place if the length changes. */
3739 if (ulen != tculen) inplace = FALSE;
3740 need = slen + 1 - ulen + tculen;
3742 else { /* Non-zero length, non-UTF-8, Need to consider locale and if
3743 * latin1 is treated as caseless. Note that a locale takes
3745 ulen = 1; /* Original character is 1 byte */
3746 tculen = 1; /* Most characters will require one byte, but this will
3747 * need to be overridden for the tricky ones */
3750 if (op_type == OP_LCFIRST) {
3752 /* lower case the first letter: no trickiness for any character */
3753 #ifdef USE_LOCALE_CTYPE
3754 if (IN_LC_RUNTIME(LC_CTYPE)) {
3755 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
3756 *tmpbuf = toLOWER_LC(*s);
3761 *tmpbuf = (IN_UNI_8_BIT)
3762 ? toLOWER_LATIN1(*s)
3766 #ifdef USE_LOCALE_CTYPE
3768 else if (IN_LC_RUNTIME(LC_CTYPE)) {
3769 if (IN_UTF8_CTYPE_LOCALE) {
3773 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
3774 *tmpbuf = (U8) toUPPER_LC(*s); /* This would be a bug if any
3775 locales have upper and title case
3779 else if (! IN_UNI_8_BIT) {
3780 *tmpbuf = toUPPER(*s); /* Returns caseless for non-ascii, or
3781 * on EBCDIC machines whatever the
3782 * native function does */
3785 /* Here, is ucfirst non-UTF-8, not in locale (unless that locale is
3786 * UTF-8, which we treat as not in locale), and cased latin1 */
3788 #ifdef USE_LOCALE_CTYPE
3792 title_ord = _to_upper_title_latin1(*s, tmpbuf, &tculen, 's');
3794 assert(tculen == 2);
3796 /* If the result is an upper Latin1-range character, it can
3797 * still be represented in one byte, which is its ordinal */
3798 if (UTF8_IS_DOWNGRADEABLE_START(*tmpbuf)) {
3799 *tmpbuf = (U8) title_ord;
3803 /* Otherwise it became more than one ASCII character (in
3804 * the case of LATIN_SMALL_LETTER_SHARP_S) or changed to
3805 * beyond Latin1, so the number of bytes changed, so can't
3806 * replace just the first character in place. */
3809 /* If the result won't fit in a byte, the entire result
3810 * will have to be in UTF-8. Assume worst case sizing in
3811 * conversion. (all latin1 characters occupy at most two
3813 if (title_ord > 255) {
3815 convert_source_to_utf8 = TRUE;
3816 need = slen * 2 + 1;
3818 /* The (converted) UTF-8 and UTF-EBCDIC lengths of all
3819 * (both) characters whose title case is above 255 is
3823 else { /* LATIN_SMALL_LETTER_SHARP_S expands by 1 byte */
3824 need = slen + 1 + 1;
3828 } /* End of use Unicode (Latin1) semantics */
3829 } /* End of changing the case of the first character */
3831 /* Here, have the first character's changed case stored in tmpbuf. Ready to
3832 * generate the result */
3835 /* We can convert in place. This means we change just the first
3836 * character without disturbing the rest; no need to grow */
3838 s = d = (U8*)SvPV_force_nomg(source, slen);
3844 /* Here, we can't convert in place; we earlier calculated how much
3845 * space we will need, so grow to accommodate that */
3846 SvUPGRADE(dest, SVt_PV);
3847 d = (U8*)SvGROW(dest, need);
3848 (void)SvPOK_only(dest);
3855 if (! convert_source_to_utf8) {
3857 /* Here both source and dest are in UTF-8, but have to create
3858 * the entire output. We initialize the result to be the
3859 * title/lower cased first character, and then append the rest
3861 sv_setpvn(dest, (char*)tmpbuf, tculen);
3863 sv_catpvn(dest, (char*)(s + ulen), slen - ulen);
3867 const U8 *const send = s + slen;
3869 /* Here the dest needs to be in UTF-8, but the source isn't,
3870 * except we earlier UTF-8'd the first character of the source
3871 * into tmpbuf. First put that into dest, and then append the
3872 * rest of the source, converting it to UTF-8 as we go. */
3874 /* Assert tculen is 2 here because the only two characters that
3875 * get to this part of the code have 2-byte UTF-8 equivalents */
3877 *d++ = *(tmpbuf + 1);
3878 s++; /* We have just processed the 1st char */
3880 for (; s < send; s++) {
3881 d = uvchr_to_utf8(d, *s);
3884 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
3888 else { /* in-place UTF-8. Just overwrite the first character */
3889 Copy(tmpbuf, d, tculen, U8);
3890 SvCUR_set(dest, need - 1);
3894 else { /* Neither source nor dest are in or need to be UTF-8 */
3896 if (inplace) { /* in-place, only need to change the 1st char */
3899 else { /* Not in-place */
3901 /* Copy the case-changed character(s) from tmpbuf */
3902 Copy(tmpbuf, d, tculen, U8);
3903 d += tculen - 1; /* Code below expects d to point to final
3904 * character stored */
3907 else { /* empty source */
3908 /* See bug #39028: Don't taint if empty */
3912 /* In a "use bytes" we don't treat the source as UTF-8, but, still want
3913 * the destination to retain that flag */
3914 if (SvUTF8(source) && ! IN_BYTES)
3917 if (!inplace) { /* Finish the rest of the string, unchanged */
3918 /* This will copy the trailing NUL */
3919 Copy(s + 1, d + 1, slen, U8);
3920 SvCUR_set(dest, need - 1);
3923 #ifdef USE_LOCALE_CTYPE
3924 if (IN_LC_RUNTIME(LC_CTYPE)) {
3929 if (dest != source && SvTAINTED(source))
3935 /* There's so much setup/teardown code common between uc and lc, I wonder if
3936 it would be worth merging the two, and just having a switch outside each
3937 of the three tight loops. There is less and less commonality though */
3950 if ( SvPADTMP(source)
3951 && !SvREADONLY(source) && SvPOK(source)
3954 #ifdef USE_LOCALE_CTYPE
3955 (IN_LC_RUNTIME(LC_CTYPE))
3956 ? ! IN_UTF8_CTYPE_LOCALE
3962 /* We can convert in place. The reason we can't if in UNI_8_BIT is to
3963 * make the loop tight, so we overwrite the source with the dest before
3964 * looking at it, and we need to look at the original source
3965 * afterwards. There would also need to be code added to handle
3966 * switching to not in-place in midstream if we run into characters
3967 * that change the length. Since being in locale overrides UNI_8_BIT,
3968 * that latter becomes irrelevant in the above test; instead for
3969 * locale, the size can't normally change, except if the locale is a
3972 s = d = (U8*)SvPV_force_nomg(source, len);
3979 s = (const U8*)SvPV_nomg_const(source, len);
3982 SvUPGRADE(dest, SVt_PV);
3983 d = (U8*)SvGROW(dest, min);
3984 (void)SvPOK_only(dest);
3989 /* Overloaded values may have toggled the UTF-8 flag on source, so we need
3990 to check DO_UTF8 again here. */
3992 if (DO_UTF8(source)) {
3993 const U8 *const send = s + len;
3994 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
3996 /* All occurrences of these are to be moved to follow any other marks.
3997 * This is context-dependent. We may not be passed enough context to
3998 * move the iota subscript beyond all of them, but we do the best we can
3999 * with what we're given. The result is always better than if we
4000 * hadn't done this. And, the problem would only arise if we are
4001 * passed a character without all its combining marks, which would be
4002 * the caller's mistake. The information this is based on comes from a
4003 * comment in Unicode SpecialCasing.txt, (and the Standard's text
4004 * itself) and so can't be checked properly to see if it ever gets
4005 * revised. But the likelihood of it changing is remote */
4006 bool in_iota_subscript = FALSE;
4012 if (in_iota_subscript && ! _is_utf8_mark(s)) {
4014 /* A non-mark. Time to output the iota subscript */
4015 Copy(GREEK_CAPITAL_LETTER_IOTA_UTF8, d, capital_iota_len, U8);
4016 d += capital_iota_len;
4017 in_iota_subscript = FALSE;
4020 /* Then handle the current character. Get the changed case value
4021 * and copy it to the output buffer */
4024 #ifdef USE_LOCALE_CTYPE
4025 uv = _toUPPER_utf8_flags(s, send, tmpbuf, &ulen, IN_LC_RUNTIME(LC_CTYPE));
4027 uv = _toUPPER_utf8_flags(s, send, tmpbuf, &ulen, 0);
4029 #define GREEK_CAPITAL_LETTER_IOTA 0x0399
4030 #define COMBINING_GREEK_YPOGEGRAMMENI 0x0345
4031 if (uv == GREEK_CAPITAL_LETTER_IOTA
4032 && utf8_to_uvchr_buf(s, send, 0) == COMBINING_GREEK_YPOGEGRAMMENI)
4034 in_iota_subscript = TRUE;
4037 if (ulen > u && (SvLEN(dest) < (min += ulen - u))) {
4038 /* If the eventually required minimum size outgrows the
4039 * available space, we need to grow. */
4040 const UV o = d - (U8*)SvPVX_const(dest);
4042 /* If someone uppercases one million U+03B0s we SvGROW()
4043 * one million times. Or we could try guessing how much to
4044 * allocate without allocating too much. Such is life.
4045 * See corresponding comment in lc code for another option
4047 d = o + (U8*) SvGROW(dest, min);
4049 Copy(tmpbuf, d, ulen, U8);
4054 if (in_iota_subscript) {
4055 Copy(GREEK_CAPITAL_LETTER_IOTA_UTF8, d, capital_iota_len, U8);
4056 d += capital_iota_len;
4061 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
4063 else { /* Not UTF-8 */
4065 const U8 *const send = s + len;
4067 /* Use locale casing if in locale; regular style if not treating
4068 * latin1 as having case; otherwise the latin1 casing. Do the
4069 * whole thing in a tight loop, for speed, */
4070 #ifdef USE_LOCALE_CTYPE
4071 if (IN_LC_RUNTIME(LC_CTYPE)) {
4072 if (IN_UTF8_CTYPE_LOCALE) {
4075 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
4076 for (; s < send; d++, s++)
4077 *d = (U8) toUPPER_LC(*s);
4081 if (! IN_UNI_8_BIT) {
4082 for (; s < send; d++, s++) {
4087 #ifdef USE_LOCALE_CTYPE
4090 for (; s < send; d++, s++) {
4091 *d = toUPPER_LATIN1_MOD(*s);
4092 if (LIKELY(*d != LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
4096 /* The mainstream case is the tight loop above. To avoid
4097 * extra tests in that, all three characters that require
4098 * special handling are mapped by the MOD to the one tested
4100 * Use the source to distinguish between the three cases */
4102 #if UNICODE_MAJOR_VERSION > 2 \
4103 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
4104 && UNICODE_DOT_DOT_VERSION >= 8)
4105 if (*s == LATIN_SMALL_LETTER_SHARP_S) {
4107 /* uc() of this requires 2 characters, but they are
4108 * ASCII. If not enough room, grow the string */
4109 if (SvLEN(dest) < ++min) {
4110 const UV o = d - (U8*)SvPVX_const(dest);
4111 d = o + (U8*) SvGROW(dest, min);
4113 *d++ = 'S'; *d = 'S'; /* upper case is 'SS' */
4114 continue; /* Back to the tight loop; still in ASCII */
4118 /* The other two special handling characters have their
4119 * upper cases outside the latin1 range, hence need to be
4120 * in UTF-8, so the whole result needs to be in UTF-8. So,
4121 * here we are somewhere in the middle of processing a
4122 * non-UTF-8 string, and realize that we will have to convert
4123 * the whole thing to UTF-8. What to do? There are
4124 * several possibilities. The simplest to code is to
4125 * convert what we have so far, set a flag, and continue on
4126 * in the loop. The flag would be tested each time through
4127 * the loop, and if set, the next character would be
4128 * converted to UTF-8 and stored. But, I (khw) didn't want
4129 * to slow down the mainstream case at all for this fairly
4130 * rare case, so I didn't want to add a test that didn't
4131 * absolutely have to be there in the loop, besides the
4132 * possibility that it would get too complicated for
4133 * optimizers to deal with. Another possibility is to just
4134 * give up, convert the source to UTF-8, and restart the
4135 * function that way. Another possibility is to convert
4136 * both what has already been processed and what is yet to
4137 * come separately to UTF-8, then jump into the loop that
4138 * handles UTF-8. But the most efficient time-wise of the
4139 * ones I could think of is what follows, and turned out to
4140 * not require much extra code. */
4142 /* Convert what we have so far into UTF-8, telling the
4143 * function that we know it should be converted, and to
4144 * allow extra space for what we haven't processed yet.
4145 * Assume the worst case space requirements for converting
4146 * what we haven't processed so far: that it will require
4147 * two bytes for each remaining source character, plus the
4148 * NUL at the end. This may cause the string pointer to
4149 * move, so re-find it. */
4151 len = d - (U8*)SvPVX_const(dest);
4152 SvCUR_set(dest, len);
4153 len = sv_utf8_upgrade_flags_grow(dest,
4154 SV_GMAGIC|SV_FORCE_UTF8_UPGRADE,
4156 d = (U8*)SvPVX(dest) + len;
4158 /* Now process the remainder of the source, converting to
4159 * upper and UTF-8. If a resulting byte is invariant in
4160 * UTF-8, output it as-is, otherwise convert to UTF-8 and
4161 * append it to the output. */
4162 for (; s < send; s++) {
4163 (void) _to_upper_title_latin1(*s, d, &len, 'S');
4167 /* Here have processed the whole source; no need to continue
4168 * with the outer loop. Each character has been converted
4169 * to upper case and converted to UTF-8 */
4172 } /* End of processing all latin1-style chars */
4173 } /* End of processing all chars */
4174 } /* End of source is not empty */
4176 if (source != dest) {
4177 *d = '\0'; /* Here d points to 1 after last char, add NUL */
4178 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
4180 } /* End of isn't utf8 */
4181 #ifdef USE_LOCALE_CTYPE
4182 if (IN_LC_RUNTIME(LC_CTYPE)) {
4187 if (dest != source && SvTAINTED(source))
4205 if ( SvPADTMP(source)
4206 && !SvREADONLY(source) && SvPOK(source)
4207 && !DO_UTF8(source)) {
4209 /* We can convert in place, as lowercasing anything in the latin1 range
4210 * (or else DO_UTF8 would have been on) doesn't lengthen it */
4212 s = d = (U8*)SvPV_force_nomg(source, len);
4219 s = (const U8*)SvPV_nomg_const(source, len);
4222 SvUPGRADE(dest, SVt_PV);
4223 d = (U8*)SvGROW(dest, min);
4224 (void)SvPOK_only(dest);
4229 /* Overloaded values may have toggled the UTF-8 flag on source, so we need
4230 to check DO_UTF8 again here. */
4232 if (DO_UTF8(source)) {
4233 const U8 *const send = s + len;
4234 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
4237 const STRLEN u = UTF8SKIP(s);
4240 #ifdef USE_LOCALE_CTYPE
4241 _toLOWER_utf8_flags(s, send, tmpbuf, &ulen, IN_LC_RUNTIME(LC_CTYPE));
4243 _toLOWER_utf8_flags(s, send, tmpbuf, &ulen, 0);
4246 /* Here is where we would do context-sensitive actions. See the
4247 * commit message for 86510fb15 for why there isn't any */
4249 if (ulen > u && (SvLEN(dest) < (min += ulen - u))) {
4251 /* If the eventually required minimum size outgrows the
4252 * available space, we need to grow. */
4253 const UV o = d - (U8*)SvPVX_const(dest);
4255 /* If someone lowercases one million U+0130s we SvGROW() one
4256 * million times. Or we could try guessing how much to
4257 * allocate without allocating too much. Such is life.
4258 * Another option would be to grow an extra byte or two more
4259 * each time we need to grow, which would cut down the million
4260 * to 500K, with little waste */
4261 d = o + (U8*) SvGROW(dest, min);
4264 /* Copy the newly lowercased letter to the output buffer we're
4266 Copy(tmpbuf, d, ulen, U8);
4269 } /* End of looping through the source string */
4272 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
4273 } else { /* Not utf8 */
4275 const U8 *const send = s + len;
4277 /* Use locale casing if in locale; regular style if not treating
4278 * latin1 as having case; otherwise the latin1 casing. Do the
4279 * whole thing in a tight loop, for speed, */
4280 #ifdef USE_LOCALE_CTYPE
4281 if (IN_LC_RUNTIME(LC_CTYPE)) {
4282 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
4283 for (; s < send; d++, s++)
4284 *d = toLOWER_LC(*s);
4288 if (! IN_UNI_8_BIT) {
4289 for (; s < send; d++, s++) {
4294 for (; s < send; d++, s++) {
4295 *d = toLOWER_LATIN1(*s);
4299 if (source != dest) {
4301 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
4304 #ifdef USE_LOCALE_CTYPE
4305 if (IN_LC_RUNTIME(LC_CTYPE)) {
4310 if (dest != source && SvTAINTED(source))
4319 SV * const sv = TOPs;
4321 const char *s = SvPV_const(sv,len);
4323 SvUTF8_off(TARG); /* decontaminate */
4326 SvUPGRADE(TARG, SVt_PV);
4327 SvGROW(TARG, (len * 2) + 1);
4331 STRLEN ulen = UTF8SKIP(s);
4332 bool to_quote = FALSE;
4334 if (UTF8_IS_INVARIANT(*s)) {
4335 if (_isQUOTEMETA(*s)) {
4339 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, s + len)) {
4341 #ifdef USE_LOCALE_CTYPE
4342 /* In locale, we quote all non-ASCII Latin1 chars.
4343 * Otherwise use the quoting rules */
4345 IN_LC_RUNTIME(LC_CTYPE)
4348 _isQUOTEMETA(EIGHT_BIT_UTF8_TO_NATIVE(*s, *(s + 1))))
4353 else if (is_QUOTEMETA_high(s)) {
4368 else if (IN_UNI_8_BIT) {
4370 if (_isQUOTEMETA(*s))
4376 /* For non UNI_8_BIT (and hence in locale) just quote all \W
4377 * including everything above ASCII */
4379 if (!isWORDCHAR_A(*s))
4385 SvCUR_set(TARG, d - SvPVX_const(TARG));
4386 (void)SvPOK_only_UTF8(TARG);
4389 sv_setpvn(TARG, s, len);
4405 U8 tmpbuf[UTF8_MAXBYTES_CASE + 1];
4406 #if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
4407 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
4408 || UNICODE_DOT_DOT_VERSION > 0)
4409 const bool full_folding = TRUE; /* This variable is here so we can easily
4410 move to more generality later */
4412 const bool full_folding = FALSE;
4414 const U8 flags = ( full_folding ? FOLD_FLAGS_FULL : 0 )
4415 #ifdef USE_LOCALE_CTYPE
4416 | ( IN_LC_RUNTIME(LC_CTYPE) ? FOLD_FLAGS_LOCALE : 0 )
4420 /* This is a facsimile of pp_lc, but with a thousand bugs thanks to me.
4421 * You are welcome(?) -Hugmeir
4429 s = (const U8*)SvPV_nomg_const(source, len);
4431 if (ckWARN(WARN_UNINITIALIZED))
4432 report_uninit(source);
4439 SvUPGRADE(dest, SVt_PV);
4440 d = (U8*)SvGROW(dest, min);
4441 (void)SvPOK_only(dest);
4446 if (DO_UTF8(source)) { /* UTF-8 flagged string. */
4448 const STRLEN u = UTF8SKIP(s);
4451 _toFOLD_utf8_flags(s, send, tmpbuf, &ulen, flags);
4453 if (ulen > u && (SvLEN(dest) < (min += ulen - u))) {
4454 const UV o = d - (U8*)SvPVX_const(dest);
4455 d = o + (U8*) SvGROW(dest, min);
4458 Copy(tmpbuf, d, ulen, U8);
4463 } /* Unflagged string */
4465 #ifdef USE_LOCALE_CTYPE
4466 if ( IN_LC_RUNTIME(LC_CTYPE) ) { /* Under locale */
4467 if (IN_UTF8_CTYPE_LOCALE) {
4468 goto do_uni_folding;
4470 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
4471 for (; s < send; d++, s++)
4472 *d = (U8) toFOLD_LC(*s);
4476 if ( !IN_UNI_8_BIT ) { /* Under nothing, or bytes */
4477 for (; s < send; d++, s++)
4481 #ifdef USE_LOCALE_CTYPE
4484 /* For ASCII and the Latin-1 range, there's only two troublesome
4485 * folds, \x{DF} (\N{LATIN SMALL LETTER SHARP S}), which under full
4486 * casefolding becomes 'ss'; and \x{B5} (\N{MICRO SIGN}), which
4487 * under any fold becomes \x{3BC} (\N{GREEK SMALL LETTER MU}) --
4488 * For the rest, the casefold is their lowercase. */
4489 for (; s < send; d++, s++) {
4490 if (*s == MICRO_SIGN) {
4491 /* \N{MICRO SIGN}'s casefold is \N{GREEK SMALL LETTER MU},
4492 * which is outside of the latin-1 range. There's a couple
4493 * of ways to deal with this -- khw discusses them in
4494 * pp_lc/uc, so go there :) What we do here is upgrade what
4495 * we had already casefolded, then enter an inner loop that
4496 * appends the rest of the characters as UTF-8. */
4497 len = d - (U8*)SvPVX_const(dest);
4498 SvCUR_set(dest, len);
4499 len = sv_utf8_upgrade_flags_grow(dest,
4500 SV_GMAGIC|SV_FORCE_UTF8_UPGRADE,
4501 /* The max expansion for latin1
4502 * chars is 1 byte becomes 2 */
4504 d = (U8*)SvPVX(dest) + len;
4506 Copy(GREEK_SMALL_LETTER_MU_UTF8, d, small_mu_len, U8);
4509 for (; s < send; s++) {
4511 UV fc = _to_uni_fold_flags(*s, tmpbuf, &ulen, flags);
4512 if UVCHR_IS_INVARIANT(fc) {
4514 && *s == LATIN_SMALL_LETTER_SHARP_S)
4523 Copy(tmpbuf, d, ulen, U8);
4529 else if (full_folding && *s == LATIN_SMALL_LETTER_SHARP_S) {
4530 /* Under full casefolding, LATIN SMALL LETTER SHARP S
4531 * becomes "ss", which may require growing the SV. */
4532 if (SvLEN(dest) < ++min) {
4533 const UV o = d - (U8*)SvPVX_const(dest);
4534 d = o + (U8*) SvGROW(dest, min);
4539 else { /* If it's not one of those two, the fold is their lower
4541 *d = toLOWER_LATIN1(*s);
4547 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
4549 #ifdef USE_LOCALE_CTYPE
4550 if (IN_LC_RUNTIME(LC_CTYPE)) {
4555 if (SvTAINTED(source))
4565 dSP; dMARK; dORIGMARK;
4566 AV *const av = MUTABLE_AV(POPs);
4567 const I32 lval = (PL_op->op_flags & OPf_MOD || LVRET);
4569 if (SvTYPE(av) == SVt_PVAV) {
4570 const bool localizing = PL_op->op_private & OPpLVAL_INTRO;
4571 bool can_preserve = FALSE;
4577 can_preserve = SvCANEXISTDELETE(av);
4580 if (lval && localizing) {
4583 for (svp = MARK + 1; svp <= SP; svp++) {
4584 const SSize_t elem = SvIV(*svp);
4588 if (max > AvMAX(av))
4592 while (++MARK <= SP) {
4594 SSize_t elem = SvIV(*MARK);
4595 bool preeminent = TRUE;
4597 if (localizing && can_preserve) {
4598 /* If we can determine whether the element exist,
4599 * Try to preserve the existenceness of a tied array
4600 * element by using EXISTS and DELETE if possible.
4601 * Fallback to FETCH and STORE otherwise. */
4602 preeminent = av_exists(av, elem);
4605 svp = av_fetch(av, elem, lval);
4608 DIE(aTHX_ PL_no_aelem, elem);
4611 save_aelem(av, elem, svp);
4613 SAVEADELETE(av, elem);
4616 *MARK = svp ? *svp : &PL_sv_undef;
4619 if (GIMME_V != G_ARRAY) {
4621 *++MARK = SP > ORIGMARK ? *SP : &PL_sv_undef;
4630 AV *const av = MUTABLE_AV(POPs);
4631 I32 lval = (PL_op->op_flags & OPf_MOD);
4632 SSize_t items = SP - MARK;
4634 if (PL_op->op_private & OPpMAYBE_LVSUB) {
4635 const I32 flags = is_lvalue_sub();
4637 if (!(flags & OPpENTERSUB_INARGS))
4638 /* diag_listed_as: Can't modify %s in %s */
4639 Perl_croak(aTHX_ "Can't modify index/value array slice in list assignment");
4646 *(MARK+items*2-1) = *(MARK+items);
4652 while (++MARK <= SP) {
4655 svp = av_fetch(av, SvIV(*MARK), lval);
4657 if (!svp || !*svp || *svp == &PL_sv_undef) {
4658 DIE(aTHX_ PL_no_aelem, SvIV(*MARK));
4660 *MARK = sv_mortalcopy(*MARK);
4662 *++MARK = svp ? *svp : &PL_sv_undef;
4664 if (GIMME_V != G_ARRAY) {
4665 MARK = SP - items*2;
4666 *++MARK = items > 0 ? *SP : &PL_sv_undef;
4676 AV *array = MUTABLE_AV(POPs);
4677 const U8 gimme = GIMME_V;
4678 IV *iterp = Perl_av_iter_p(aTHX_ array);
4679 const IV current = (*iterp)++;
4681 if (current > av_tindex(array)) {
4683 if (gimme == G_SCALAR)
4691 if (gimme == G_ARRAY) {
4692 SV **const element = av_fetch(array, current, 0);
4693 PUSHs(element ? *element : &PL_sv_undef);
4698 /* also used for: pp_avalues()*/
4702 AV *array = MUTABLE_AV(POPs);
4703 const U8 gimme = GIMME_V;
4705 *Perl_av_iter_p(aTHX_ array) = 0;
4707 if (gimme == G_SCALAR) {
4709 PUSHi(av_tindex(array) + 1);
4711 else if (gimme == G_ARRAY) {
4712 if (UNLIKELY(PL_op->op_private & OPpMAYBE_LVSUB)) {
4713 const I32 flags = is_lvalue_sub();
4714 if (flags && !(flags & OPpENTERSUB_INARGS))
4715 /* diag_listed_as: Can't modify %s in %s */
4717 "Can't modify keys on array in list assignment");
4720 IV n = Perl_av_len(aTHX_ array);
4725 if ( PL_op->op_type == OP_AKEYS
4726 || ( PL_op->op_type == OP_AVHVSWITCH
4727 && (PL_op->op_private & 3) + OP_AEACH == OP_AKEYS ))
4729 for (i = 0; i <= n; i++) {
4734 for (i = 0; i <= n; i++) {
4735 SV *const *const elem = Perl_av_fetch(aTHX_ array, i, 0);
4736 PUSHs(elem ? *elem : &PL_sv_undef);
4744 /* Associative arrays. */
4749 HV * hash = MUTABLE_HV(POPs);
4751 const U8 gimme = GIMME_V;
4753 entry = hv_iternext(hash);
4757 SV* const sv = hv_iterkeysv(entry);
4759 if (gimme == G_ARRAY) {
4761 val = hv_iterval(hash, entry);
4765 else if (gimme == G_SCALAR)
4772 S_do_delete_local(pTHX)
4775 const U8 gimme = GIMME_V;
4778 const bool sliced = !!(PL_op->op_private & OPpSLICE);
4779 SV **unsliced_keysv = sliced ? NULL : sp--;
4780 SV * const osv = POPs;
4781 SV **mark = sliced ? PL_stack_base + POPMARK : unsliced_keysv-1;
4783 const bool tied = SvRMAGICAL(osv)
4784 && mg_find((const SV *)osv, PERL_MAGIC_tied);
4785 const bool can_preserve = SvCANEXISTDELETE(osv);
4786 const U32 type = SvTYPE(osv);
4787 SV ** const end = sliced ? SP : unsliced_keysv;
4789 if (type == SVt_PVHV) { /* hash element */
4790 HV * const hv = MUTABLE_HV(osv);
4791 while (++MARK <= end) {
4792 SV * const keysv = *MARK;
4794 bool preeminent = TRUE;
4796 preeminent = hv_exists_ent(hv, keysv, 0);
4798 HE *he = hv_fetch_ent(hv, keysv, 1, 0);
4805 sv = hv_delete_ent(hv, keysv, 0, 0);
4807 SvREFCNT_inc_simple_void(sv); /* De-mortalize */
4810 if (!sv) DIE(aTHX_ PL_no_helem_sv, SVfARG(keysv));
4811 save_helem_flags(hv, keysv, &sv, SAVEf_KEEPOLDELEM);
4813 *MARK = sv_mortalcopy(sv);
4819 SAVEHDELETE(hv, keysv);
4820 *MARK = &PL_sv_undef;
4824 else if (type == SVt_PVAV) { /* array element */
4825 if (PL_op->op_flags & OPf_SPECIAL) {
4826 AV * const av = MUTABLE_AV(osv);
4827 while (++MARK <= end) {
4828 SSize_t idx = SvIV(*MARK);
4830 bool preeminent = TRUE;
4832 preeminent = av_exists(av, idx);
4834 SV **svp = av_fetch(av, idx, 1);
4841 sv = av_delete(av, idx, 0);
4843 SvREFCNT_inc_simple_void(sv); /* De-mortalize */
4846 save_aelem_flags(av, idx, &sv, SAVEf_KEEPOLDELEM);
4848 *MARK = sv_mortalcopy(sv);
4854 SAVEADELETE(av, idx);
4855 *MARK = &PL_sv_undef;
4860 DIE(aTHX_ "panic: avhv_delete no longer supported");
4863 DIE(aTHX_ "Not a HASH reference");
4865 if (gimme == G_VOID)
4867 else if (gimme == G_SCALAR) {
4872 *++MARK = &PL_sv_undef;
4876 else if (gimme != G_VOID)
4877 PUSHs(*unsliced_keysv);
4888 if (PL_op->op_private & OPpLVAL_INTRO)
4889 return do_delete_local();
4892 discard = (gimme == G_VOID) ? G_DISCARD : 0;
4894 if (PL_op->op_private & (OPpSLICE|OPpKVSLICE)) {
4896 HV * const hv = MUTABLE_HV(POPs);
4897 const U32 hvtype = SvTYPE(hv);
4899 if (PL_op->op_private & OPpKVSLICE) {
4900 SSize_t items = SP - MARK;
4904 *(MARK+items*2-1) = *(MARK+items);
4911 if (hvtype == SVt_PVHV) { /* hash element */
4912 while ((MARK += (1+skip)) <= SP) {
4913 SV * const sv = hv_delete_ent(hv, *(MARK-skip), discard, 0);
4914 *MARK = sv ? sv : &PL_sv_undef;
4917 else if (hvtype == SVt_PVAV) { /* array element */
4918 if (PL_op->op_flags & OPf_SPECIAL) {
4919 while ((MARK += (1+skip)) <= SP) {
4920 SV * const sv = av_delete(MUTABLE_AV(hv), SvIV(*(MARK-skip)), discard);
4921 *MARK = sv ? sv : &PL_sv_undef;
4926 DIE(aTHX_ "Not a HASH reference");
4929 else if (gimme == G_SCALAR) {
4934 *++MARK = &PL_sv_undef;
4940 HV * const hv = MUTABLE_HV(POPs);
4942 if (SvTYPE(hv) == SVt_PVHV)
4943 sv = hv_delete_ent(hv, keysv, discard, 0);
4944 else if (SvTYPE(hv) == SVt_PVAV) {
4945 if (PL_op->op_flags & OPf_SPECIAL)
4946 sv = av_delete(MUTABLE_AV(hv), SvIV(keysv), discard);
4948 DIE(aTHX_ "panic: avhv_delete no longer supported");
4951 DIE(aTHX_ "Not a HASH reference");
4966 if (UNLIKELY( PL_op->op_private & OPpEXISTS_SUB )) {
4968 SV * const sv = POPs;
4969 CV * const cv = sv_2cv(sv, &hv, &gv, 0);
4972 if (gv && isGV(gv) && GvCV(gv) && !GvCVGEN(gv))
4977 hv = MUTABLE_HV(POPs);
4978 if (LIKELY( SvTYPE(hv) == SVt_PVHV )) {
4979 if (hv_exists_ent(hv, tmpsv, 0))
4982 else if (SvTYPE(hv) == SVt_PVAV) {
4983 if (PL_op->op_flags & OPf_SPECIAL) { /* array element */
4984 if (av_exists(MUTABLE_AV(hv), SvIV(tmpsv)))
4989 DIE(aTHX_ "Not a HASH reference");
4996 dSP; dMARK; dORIGMARK;
4997 HV * const hv = MUTABLE_HV(POPs);
4998 const I32 lval = (PL_op->op_flags & OPf_MOD || LVRET);
4999 const bool localizing = PL_op->op_private & OPpLVAL_INTRO;
5000 bool can_preserve = FALSE;
5006 if (SvCANEXISTDELETE(hv))
5007 can_preserve = TRUE;
5010 while (++MARK <= SP) {
5011 SV * const keysv = *MARK;
5014 bool preeminent = TRUE;
5016 if (localizing && can_preserve) {
5017 /* If we can determine whether the element exist,
5018 * try to preserve the existenceness of a tied hash
5019 * element by using EXISTS and DELETE if possible.
5020 * Fallback to FETCH and STORE otherwise. */
5021 preeminent = hv_exists_ent(hv, keysv, 0);
5024 he = hv_fetch_ent(hv, keysv, lval, 0);
5025 svp = he ? &HeVAL(he) : NULL;
5028 if (!svp || !*svp || *svp == &PL_sv_undef) {
5029 DIE(aTHX_ PL_no_helem_sv, SVfARG(keysv));
5032 if (HvNAME_get(hv) && isGV_or_RVCV(*svp))
5033 save_gp(MUTABLE_GV(*svp), !(PL_op->op_flags & OPf_SPECIAL));
5034 else if (preeminent)
5035 save_helem_flags(hv, keysv, svp,
5036 (PL_op->op_flags & OPf_SPECIAL) ? 0 : SAVEf_SETMAGIC);
5038 SAVEHDELETE(hv, keysv);
5041 *MARK = svp && *svp ? *svp : &PL_sv_undef;
5043 if (GIMME_V != G_ARRAY) {
5045 *++MARK = SP > ORIGMARK ? *SP : &PL_sv_undef;
5054 HV * const hv = MUTABLE_HV(POPs);
5055 I32 lval = (PL_op->op_flags & OPf_MOD);
5056 SSize_t items = SP - MARK;
5058 if (PL_op->op_private & OPpMAYBE_LVSUB) {
5059 const I32 flags = is_lvalue_sub();
5061 if (!(flags & OPpENTERSUB_INARGS))
5062 /* diag_listed_as: Can't modify %s in %s */
5063 Perl_croak(aTHX_ "Can't modify key/value hash slice in %s assignment",
5064 GIMME_V == G_ARRAY ? "list" : "scalar");
5071 *(MARK+items*2-1) = *(MARK+items);
5077 while (++MARK <= SP) {
5078 SV * const keysv = *MARK;
5082 he = hv_fetch_ent(hv, keysv, lval, 0);
5083 svp = he ? &HeVAL(he) : NULL;
5086 if (!svp || !*svp || *svp == &PL_sv_undef) {
5087 DIE(aTHX_ PL_no_helem_sv, SVfARG(keysv));
5089 *MARK = sv_mortalcopy(*MARK);
5091 *++MARK = svp && *svp ? *svp : &PL_sv_undef;
5093 if (GIMME_V != G_ARRAY) {
5094 MARK = SP - items*2;
5095 *++MARK = items > 0 ? *SP : &PL_sv_undef;
5101 /* List operators. */
5105 I32 markidx = POPMARK;
5106 if (GIMME_V != G_ARRAY) {
5107 /* don't initialize mark here, EXTEND() may move the stack */
5110 EXTEND(SP, 1); /* in case no arguments, as in @empty */
5111 mark = PL_stack_base + markidx;
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 */
5618 if (SP - MARK > 1) {
5619 do_join(TARG, &PL_sv_no, MARK, SP);
5622 } else if (SP > MARK) {
5623 sv_setsv(TARG, *SP);
5626 sv_setsv(TARG, DEFSV);
5630 up = SvPV_force(TARG, len);
5633 if (DO_UTF8(TARG)) { /* first reverse each character */
5634 U8* s = (U8*)SvPVX(TARG);
5635 const U8* send = (U8*)(s + len);
5637 if (UTF8_IS_INVARIANT(*s)) {
5642 if (!utf8_to_uvchr_buf(s, send, 0))
5646 down = (char*)(s - 1);
5647 /* reverse this character */
5649 const char tmp = *up;
5657 down = SvPVX(TARG) + len - 1;
5659 const char tmp = *up;
5663 (void)SvPOK_only_UTF8(TARG);
5672 AV *ary = ( (PL_op->op_private & OPpSPLIT_ASSIGN) /* @a = split */
5673 && (PL_op->op_flags & OPf_STACKED)) /* @{expr} = split */
5674 ? (AV *)POPs : NULL;
5675 IV limit = POPi; /* note, negative is forever */
5676 SV * const sv = POPs;
5678 const char *s = SvPV_const(sv, len);
5679 const bool do_utf8 = DO_UTF8(sv);
5680 const bool in_uni_8_bit = IN_UNI_8_BIT;
5681 const char *strend = s + len;
5682 PMOP *pm = cPMOPx(PL_op);
5687 const STRLEN slen = do_utf8
5688 ? utf8_length((U8*)s, (U8*)strend)
5689 : (STRLEN)(strend - s);
5690 SSize_t maxiters = slen + 10;
5691 I32 trailing_empty = 0;
5693 const IV origlimit = limit;
5696 const U8 gimme = GIMME_V;
5698 I32 oldsave = PL_savestack_ix;
5699 U32 make_mortal = SVs_TEMP;
5705 TAINT_IF(get_regex_charset(RX_EXTFLAGS(rx)) == REGEX_LOCALE_CHARSET &&
5706 (RX_EXTFLAGS(rx) & (RXf_WHITE | RXf_SKIPWHITE)));
5708 /* handle @ary = split(...) optimisation */
5709 if (PL_op->op_private & OPpSPLIT_ASSIGN) {
5710 if (!(PL_op->op_flags & OPf_STACKED)) {
5711 if (PL_op->op_private & OPpSPLIT_LEX) {
5712 if (PL_op->op_private & OPpLVAL_INTRO)
5713 SAVECLEARSV(PAD_SVl(pm->op_pmreplrootu.op_pmtargetoff));
5714 ary = (AV *)PAD_SVl(pm->op_pmreplrootu.op_pmtargetoff);
5719 MUTABLE_GV(PAD_SVl(pm->op_pmreplrootu.op_pmtargetoff));
5721 pm->op_pmreplrootu.op_pmtargetgv;
5723 if (PL_op->op_private & OPpLVAL_INTRO)
5728 /* skip anything pushed by OPpLVAL_INTRO above */
5729 oldsave = PL_savestack_ix;
5735 (void)sv_2mortal(SvREFCNT_inc_simple_NN(sv));
5738 if ((mg = SvTIED_mg((const SV *)ary, PERL_MAGIC_tied))) {
5740 XPUSHs(SvTIED_obj(MUTABLE_SV(ary), mg));
5747 for (i = AvFILLp(ary); i >= 0; i--)
5748 AvARRAY(ary)[i] = &PL_sv_undef; /* don't free mere refs */
5750 /* temporarily switch stacks */
5751 SAVESWITCHSTACK(PL_curstack, ary);
5756 base = SP - PL_stack_base;
5758 if (RX_EXTFLAGS(rx) & RXf_SKIPWHITE) {
5760 while (s < strend && isSPACE_utf8_safe(s, strend))
5763 else if (get_regex_charset(RX_EXTFLAGS(rx)) == REGEX_LOCALE_CHARSET) {
5764 while (s < strend && isSPACE_LC(*s))
5767 else if (in_uni_8_bit) {
5768 while (s < strend && isSPACE_L1(*s))
5772 while (s < strend && isSPACE(*s))
5776 if (RX_EXTFLAGS(rx) & RXf_PMf_MULTILINE) {
5780 gimme_scalar = gimme == G_SCALAR && !ary;
5783 limit = maxiters + 2;
5784 if (RX_EXTFLAGS(rx) & RXf_WHITE) {
5787 /* this one uses 'm' and is a negative test */
5789 while (m < strend && ! isSPACE_utf8_safe(m, strend) ) {
5790 const int t = UTF8SKIP(m);
5791 /* isSPACE_utf8_safe returns FALSE for malform utf8 */
5798 else if (get_regex_charset(RX_EXTFLAGS(rx)) == REGEX_LOCALE_CHARSET)
5800 while (m < strend && !isSPACE_LC(*m))
5803 else if (in_uni_8_bit) {
5804 while (m < strend && !isSPACE_L1(*m))
5807 while (m < strend && !isSPACE(*m))
5820 dstr = newSVpvn_flags(s, m-s,
5821 (do_utf8 ? SVf_UTF8 : 0) | make_mortal);
5825 /* skip the whitespace found last */
5827 s = m + UTF8SKIP(m);
5831 /* this one uses 's' and is a positive test */
5833 while (s < strend && isSPACE_utf8_safe(s, strend) )
5836 else if (get_regex_charset(RX_EXTFLAGS(rx)) == REGEX_LOCALE_CHARSET)
5838 while (s < strend && isSPACE_LC(*s))
5841 else if (in_uni_8_bit) {
5842 while (s < strend && isSPACE_L1(*s))
5845 while (s < strend && isSPACE(*s))
5850 else if (RX_EXTFLAGS(rx) & RXf_START_ONLY) {
5852 for (m = s; m < strend && *m != '\n'; m++)
5865 dstr = newSVpvn_flags(s, m-s,
5866 (do_utf8 ? SVf_UTF8 : 0) | make_mortal);
5872 else if (RX_EXTFLAGS(rx) & RXf_NULL && !(s >= strend)) {
5874 Pre-extend the stack, either the number of bytes or
5875 characters in the string or a limited amount, triggered by:
5877 my ($x, $y) = split //, $str;
5881 if (!gimme_scalar) {
5882 const IV items = limit - 1;
5883 /* setting it to -1 will trigger a panic in EXTEND() */
5884 const SSize_t sslen = slen > SSize_t_MAX ? -1 : (SSize_t)slen;
5885 if (items >=0 && items < sslen)
5893 /* keep track of how many bytes we skip over */
5903 dstr = newSVpvn_flags(m, s-m, SVf_UTF8 | make_mortal);
5916 dstr = newSVpvn(s, 1);
5932 else if (do_utf8 == (RX_UTF8(rx) != 0) &&
5933 (RX_EXTFLAGS(rx) & RXf_USE_INTUIT) && !RX_NPARENS(rx)
5934 && (RX_EXTFLAGS(rx) & RXf_CHECK_ALL)
5935 && !(RX_EXTFLAGS(rx) & RXf_IS_ANCHORED)) {
5936 const int tail = (RX_EXTFLAGS(rx) & RXf_INTUIT_TAIL);
5937 SV * const csv = CALLREG_INTUIT_STRING(rx);
5939 len = RX_MINLENRET(rx);
5940 if (len == 1 && !RX_UTF8(rx) && !tail) {
5941 const char c = *SvPV_nolen_const(csv);
5943 for (m = s; m < strend && *m != c; m++)
5954 dstr = newSVpvn_flags(s, m-s,
5955 (do_utf8 ? SVf_UTF8 : 0) | make_mortal);
5958 /* The rx->minlen is in characters but we want to step
5959 * s ahead by bytes. */
5961 s = (char*)utf8_hop((U8*)m, len);
5963 s = m + len; /* Fake \n at the end */
5967 while (s < strend && --limit &&
5968 (m = fbm_instr((unsigned char*)s, (unsigned char*)strend,
5969 csv, multiline ? FBMrf_MULTILINE : 0)) )
5978 dstr = newSVpvn_flags(s, m-s,
5979 (do_utf8 ? SVf_UTF8 : 0) | make_mortal);
5982 /* The rx->minlen is in characters but we want to step
5983 * s ahead by bytes. */
5985 s = (char*)utf8_hop((U8*)m, len);
5987 s = m + len; /* Fake \n at the end */
5992 maxiters += slen * RX_NPARENS(rx);
5993 while (s < strend && --limit)
5997 rex_return = CALLREGEXEC(rx, (char*)s, (char*)strend, (char*)orig, 1,
6000 if (rex_return == 0)
6002 TAINT_IF(RX_MATCH_TAINTED(rx));
6003 /* we never pass the REXEC_COPY_STR flag, so it should
6004 * never get copied */
6005 assert(!RX_MATCH_COPIED(rx));
6006 m = RX_OFFS(rx)[0].start + orig;
6015 dstr = newSVpvn_flags(s, m-s,
6016 (do_utf8 ? SVf_UTF8 : 0) | make_mortal);
6019 if (RX_NPARENS(rx)) {
6021 for (i = 1; i <= (I32)RX_NPARENS(rx); i++) {
6022 s = RX_OFFS(rx)[i].start + orig;
6023 m = RX_OFFS(rx)[i].end + orig;
6025 /* japhy (07/27/01) -- the (m && s) test doesn't catch
6026 parens that didn't match -- they should be set to
6027 undef, not the empty string */
6035 if (m >= orig && s >= orig) {
6036 dstr = newSVpvn_flags(s, m-s,
6037 (do_utf8 ? SVf_UTF8 : 0)
6041 dstr = &PL_sv_undef; /* undef, not "" */
6047 s = RX_OFFS(rx)[0].end + orig;
6051 if (!gimme_scalar) {
6052 iters = (SP - PL_stack_base) - base;
6054 if (iters > maxiters)
6055 DIE(aTHX_ "Split loop");
6057 /* keep field after final delim? */
6058 if (s < strend || (iters && origlimit)) {
6059 if (!gimme_scalar) {
6060 const STRLEN l = strend - s;
6061 dstr = newSVpvn_flags(s, l, (do_utf8 ? SVf_UTF8 : 0) | make_mortal);
6066 else if (!origlimit) {
6068 iters -= trailing_empty;
6070 while (iters > 0 && (!TOPs || !SvANY(TOPs) || SvCUR(TOPs) == 0)) {
6071 if (TOPs && !make_mortal)
6080 LEAVE_SCOPE(oldsave); /* may undo an earlier SWITCHSTACK */
6084 if (SvSMAGICAL(ary)) {
6086 mg_set(MUTABLE_SV(ary));
6089 if (gimme == G_ARRAY) {
6091 Copy(AvARRAY(ary), SP + 1, iters, SV*);
6098 ENTER_with_name("call_PUSH");
6099 call_sv(SV_CONST(PUSH),G_SCALAR|G_DISCARD|G_METHOD_NAMED);
6100 LEAVE_with_name("call_PUSH");
6102 if (gimme == G_ARRAY) {
6104 /* EXTEND should not be needed - we just popped them */
6106 for (i=0; i < iters; i++) {
6107 SV **svp = av_fetch(ary, i, FALSE);
6108 PUSHs((svp) ? *svp : &PL_sv_undef);
6115 if (gimme == G_ARRAY)
6127 SV *const sv = PAD_SVl(PL_op->op_targ);
6129 if (SvPADSTALE(sv)) {
6132 RETURNOP(cLOGOP->op_other);
6134 RETURNOP(cLOGOP->op_next);
6143 if (SvTYPE(retsv) == SVt_PVAV || SvTYPE(retsv) == SVt_PVHV
6144 || SvTYPE(retsv) == SVt_PVCV) {
6145 retsv = refto(retsv);
6152 /* used for: pp_padany(), pp_custom(); plus any system ops
6153 * that aren't implemented on a particular platform */
6155 PP(unimplemented_op)
6157 const Optype op_type = PL_op->op_type;
6158 /* Using OP_NAME() isn't going to be helpful here. Firstly, it doesn't cope
6159 with out of range op numbers - it only "special" cases op_custom.
6160 Secondly, as the three ops we "panic" on are padmy, mapstart and custom,
6161 if we get here for a custom op then that means that the custom op didn't
6162 have an implementation. Given that OP_NAME() looks up the custom op
6163 by its pp_addr, likely it will return NULL, unless someone (unhelpfully)
6164 registers &PL_unimplemented_op as the address of their custom op.
6165 NULL doesn't generate a useful error message. "custom" does. */
6166 const char *const name = op_type >= OP_max
6167 ? "[out of range]" : PL_op_name[PL_op->op_type];
6168 if(OP_IS_SOCKET(op_type))
6169 DIE(aTHX_ PL_no_sock_func, name);
6170 DIE(aTHX_ "panic: unimplemented op %s (#%d) called", name, op_type);
6174 S_maybe_unwind_defav(pTHX)
6176 if (CX_CUR()->cx_type & CXp_HASARGS) {
6177 PERL_CONTEXT *cx = CX_CUR();
6179 assert(CxHASARGS(cx));
6181 cx->cx_type &= ~CXp_HASARGS;
6185 /* For sorting out arguments passed to a &CORE:: subroutine */
6189 int opnum = SvIOK(cSVOP_sv) ? (int)SvUV(cSVOP_sv) : 0;
6190 int defgv = PL_opargs[opnum] & OA_DEFGV ||opnum==OP_GLOB, whicharg = 0;
6191 AV * const at_ = GvAV(PL_defgv);
6192 SV **svp = at_ ? AvARRAY(at_) : NULL;
6193 I32 minargs = 0, maxargs = 0, numargs = at_ ? AvFILLp(at_)+1 : 0;
6194 I32 oa = opnum ? PL_opargs[opnum] >> OASHIFT : 0;
6195 bool seen_question = 0;
6196 const char *err = NULL;
6197 const bool pushmark = PL_op->op_private & OPpCOREARGS_PUSHMARK;
6199 /* Count how many args there are first, to get some idea how far to
6200 extend the stack. */
6202 if ((oa & 7) == OA_LIST) { maxargs = I32_MAX; break; }
6204 if (oa & OA_OPTIONAL) seen_question = 1;
6205 if (!seen_question) minargs++;
6209 if(numargs < minargs) err = "Not enough";
6210 else if(numargs > maxargs) err = "Too many";
6212 /* diag_listed_as: Too many arguments for %s */
6214 "%s arguments for %s", err,
6215 opnum ? PL_op_desc[opnum] : SvPV_nolen_const(cSVOP_sv)
6218 /* Reset the stack pointer. Without this, we end up returning our own
6219 arguments in list context, in addition to the values we are supposed
6220 to return. nextstate usually does this on sub entry, but we need
6221 to run the next op with the caller's hints, so we cannot have a
6223 SP = PL_stack_base + CX_CUR()->blk_oldsp;
6225 if(!maxargs) RETURN;
6227 /* We do this here, rather than with a separate pushmark op, as it has
6228 to come in between two things this function does (stack reset and
6229 arg pushing). This seems the easiest way to do it. */
6232 (void)Perl_pp_pushmark(aTHX);
6235 EXTEND(SP, maxargs == I32_MAX ? numargs : maxargs);
6236 PUTBACK; /* The code below can die in various places. */
6238 oa = PL_opargs[opnum] >> OASHIFT;
6239 for (; oa&&(numargs||!pushmark); (void)(numargs&&(++svp,--numargs))) {
6244 if (!numargs && defgv && whicharg == minargs + 1) {
6247 else PUSHs(numargs ? svp && *svp ? *svp : &PL_sv_undef : NULL);
6251 PUSHs(svp && *svp ? *svp : &PL_sv_undef);
6258 if (CvUNIQUE(find_runcv_where(FIND_RUNCV_level_eq,1,NULL)))
6261 S_maybe_unwind_defav(aTHX);
6264 PUSHs((SV *)GvAVn(gv));
6267 if (!svp || !*svp || !SvROK(*svp)
6268 || SvTYPE(SvRV(*svp)) != SVt_PVAV)
6270 /* diag_listed_as: Type of arg %d to &CORE::%s must be %s*/
6271 "Type of arg %d to &CORE::%s must be array reference",
6272 whicharg, PL_op_desc[opnum]
6277 if (!svp || !*svp || !SvROK(*svp)
6278 || ( SvTYPE(SvRV(*svp)) != SVt_PVHV
6279 && ( opnum == OP_DBMCLOSE || opnum == OP_DBMOPEN
6280 || SvTYPE(SvRV(*svp)) != SVt_PVAV )))
6282 /* diag_listed_as: Type of arg %d to &CORE::%s must be %s*/
6283 "Type of arg %d to &CORE::%s must be hash%s reference",
6284 whicharg, PL_op_desc[opnum],
6285 opnum == OP_DBMCLOSE || opnum == OP_DBMOPEN
6292 if (!numargs) PUSHs(NULL);
6293 else if(svp && *svp && SvROK(*svp) && isGV_with_GP(SvRV(*svp)))
6294 /* no magic here, as the prototype will have added an extra
6295 refgen and we just want what was there before that */
6298 const bool constr = PL_op->op_private & whicharg;
6300 svp && *svp ? *svp : &PL_sv_undef,
6301 constr, cBOOL(CopHINTS_get(PL_curcop) & HINT_STRICT_REFS),
6307 if (!numargs) goto try_defsv;
6309 const bool wantscalar =
6310 PL_op->op_private & OPpCOREARGS_SCALARMOD;
6311 if (!svp || !*svp || !SvROK(*svp)
6312 /* We have to permit globrefs even for the \$ proto, as
6313 *foo is indistinguishable from ${\*foo}, and the proto-
6314 type permits the latter. */
6315 || SvTYPE(SvRV(*svp)) > (
6316 wantscalar ? SVt_PVLV
6317 : opnum == OP_LOCK || opnum == OP_UNDEF
6323 "Type of arg %d to &CORE::%s must be %s",
6324 whicharg, PL_op_name[opnum],
6326 ? "scalar reference"
6327 : opnum == OP_LOCK || opnum == OP_UNDEF
6328 ? "reference to one of [$@%&*]"
6329 : "reference to one of [$@%*]"
6332 if (opnum == OP_UNDEF && SvRV(*svp) == (SV *)PL_defgv) {
6333 /* Undo @_ localisation, so that sub exit does not undo
6334 part of our undeffing. */
6335 S_maybe_unwind_defav(aTHX);
6340 DIE(aTHX_ "panic: unknown OA_*: %x", (unsigned)(oa&7));
6348 /* Implement CORE::keys(),values(),each().
6350 * We won't know until run-time whether the arg is an array or hash,
6353 * pp_keys/pp_values/pp_each
6355 * pp_akeys/pp_avalues/pp_aeach
6357 * as appropriate (or whatever pp function actually implements the OP_FOO
6358 * functionality for each FOO).
6365 (SvTYPE(TOPs) == SVt_PVAV ? OP_AEACH : OP_EACH)
6366 + (PL_op->op_private & OPpAVHVSWITCH_MASK)
6374 if (PL_op->op_private & OPpOFFBYONE) {
6375 cv = find_runcv_where(FIND_RUNCV_level_eq, 1, NULL);
6377 else cv = find_runcv(NULL);
6378 XPUSHs(CvEVAL(cv) ? &PL_sv_undef : sv_2mortal(newRV((SV *)cv)));
6383 S_localise_aelem_lval(pTHX_ AV * const av, SV * const keysv,
6384 const bool can_preserve)
6386 const SSize_t ix = SvIV(keysv);
6387 if (can_preserve ? av_exists(av, ix) : TRUE) {
6388 SV ** const svp = av_fetch(av, ix, 1);
6390 Perl_croak(aTHX_ PL_no_aelem, ix);
6391 save_aelem(av, ix, svp);
6394 SAVEADELETE(av, ix);
6398 S_localise_helem_lval(pTHX_ HV * const hv, SV * const keysv,
6399 const bool can_preserve)
6401 if (can_preserve ? hv_exists_ent(hv, keysv, 0) : TRUE) {
6402 HE * const he = hv_fetch_ent(hv, keysv, 1, 0);
6403 SV ** const svp = he ? &HeVAL(he) : NULL;
6405 Perl_croak(aTHX_ PL_no_helem_sv, SVfARG(keysv));
6406 save_helem_flags(hv, keysv, svp, 0);
6409 SAVEHDELETE(hv, keysv);
6413 S_localise_gv_slot(pTHX_ GV *gv, U8 type)
6415 if (type == OPpLVREF_SV) {
6416 save_pushptrptr(gv, SvREFCNT_inc_simple(GvSV(gv)), SAVEt_GVSV);
6419 else if (type == OPpLVREF_AV)
6420 /* XXX Inefficient, as it creates a new AV, which we are
6421 about to clobber. */
6424 assert(type == OPpLVREF_HV);
6425 /* XXX Likewise inefficient. */
6434 SV * const key = PL_op->op_private & OPpLVREF_ELEM ? POPs : NULL;
6435 SV * const left = PL_op->op_flags & OPf_STACKED ? POPs : NULL;
6437 const char *bad = NULL;
6438 const U8 type = PL_op->op_private & OPpLVREF_TYPE;
6439 if (!SvROK(sv)) DIE(aTHX_ "Assigned value is not a reference");
6442 if (SvTYPE(SvRV(sv)) > SVt_PVLV)
6446 if (SvTYPE(SvRV(sv)) != SVt_PVAV)
6450 if (SvTYPE(SvRV(sv)) != SVt_PVHV)
6454 if (SvTYPE(SvRV(sv)) != SVt_PVCV)
6458 /* diag_listed_as: Assigned value is not %s reference */
6459 DIE(aTHX_ "Assigned value is not a%s reference", bad);
6463 switch (left ? SvTYPE(left) : 0) {
6466 SV * const old = PAD_SV(ARGTARG);
6467 PAD_SETSV(ARGTARG, SvREFCNT_inc_NN(SvRV(sv)));
6469 if ((PL_op->op_private & (OPpLVAL_INTRO|OPpPAD_STATE))
6471 SAVECLEARSV(PAD_SVl(ARGTARG));
6475 if (PL_op->op_private & OPpLVAL_INTRO) {
6476 S_localise_gv_slot(aTHX_ (GV *)left, type);
6478 gv_setref(left, sv);
6483 if (UNLIKELY(PL_op->op_private & OPpLVAL_INTRO)) {
6484 S_localise_aelem_lval(aTHX_ (AV *)left, key,
6485 SvCANEXISTDELETE(left));
6487 av_store((AV *)left, SvIV(key), SvREFCNT_inc_simple_NN(SvRV(sv)));
6490 if (UNLIKELY(PL_op->op_private & OPpLVAL_INTRO)) {
6492 S_localise_helem_lval(aTHX_ (HV *)left, key,
6493 SvCANEXISTDELETE(left));
6495 (void)hv_store_ent((HV *)left, key, SvREFCNT_inc_simple_NN(SvRV(sv)), 0);
6497 if (PL_op->op_flags & OPf_MOD)
6498 SETs(sv_2mortal(newSVsv(sv)));
6499 /* XXX else can weak references go stale before they are read, e.g.,
6508 SV * const ret = sv_2mortal(newSV_type(SVt_PVMG));
6509 SV * const elem = PL_op->op_private & OPpLVREF_ELEM ? POPs : NULL;
6510 SV * const arg = PL_op->op_flags & OPf_STACKED ? POPs : NULL;
6511 MAGIC * const mg = sv_magicext(ret, arg, PERL_MAGIC_lvref,
6512 &PL_vtbl_lvref, (char *)elem,
6513 elem ? HEf_SVKEY : (I32)ARGTARG);
6514 mg->mg_private = PL_op->op_private;
6515 if (PL_op->op_private & OPpLVREF_ITER)
6516 mg->mg_flags |= MGf_PERSIST;
6517 if (UNLIKELY(PL_op->op_private & OPpLVAL_INTRO)) {
6523 const bool can_preserve = SvCANEXISTDELETE(arg);
6524 if (SvTYPE(arg) == SVt_PVAV)
6525 S_localise_aelem_lval(aTHX_ (AV *)arg, elem, can_preserve);
6527 S_localise_helem_lval(aTHX_ (HV *)arg, elem, can_preserve);
6531 S_localise_gv_slot(aTHX_ (GV *)arg,
6532 PL_op->op_private & OPpLVREF_TYPE);
6534 else if (!(PL_op->op_private & OPpPAD_STATE))
6535 SAVECLEARSV(PAD_SVl(ARGTARG));
6544 AV * const av = (AV *)POPs;
6545 const bool localizing = PL_op->op_private & OPpLVAL_INTRO;
6546 bool can_preserve = FALSE;
6548 if (UNLIKELY(localizing)) {
6553 can_preserve = SvCANEXISTDELETE(av);
6555 if (SvTYPE(av) == SVt_PVAV) {
6558 for (svp = MARK + 1; svp <= SP; svp++) {
6559 const SSize_t elem = SvIV(*svp);
6563 if (max > AvMAX(av))
6568 while (++MARK <= SP) {
6569 SV * const elemsv = *MARK;
6570 if (SvTYPE(av) == SVt_PVAV)
6571 S_localise_aelem_lval(aTHX_ av, elemsv, can_preserve);
6573 S_localise_helem_lval(aTHX_ (HV *)av, elemsv, can_preserve);
6574 *MARK = sv_2mortal(newSV_type(SVt_PVMG));
6575 sv_magic(*MARK,(SV *)av,PERL_MAGIC_lvref,(char *)elemsv,HEf_SVKEY);
6582 if (PL_op->op_flags & OPf_STACKED)
6583 Perl_pp_rv2av(aTHX);
6585 Perl_pp_padav(aTHX);
6589 SETs(0); /* special alias marker that aassign recognises */
6599 SETs(sv_2mortal((SV *)newCONSTSUB(SvTYPE(CopSTASH(PL_curcop))==SVt_PVHV
6600 ? CopSTASH(PL_curcop)
6602 NULL, SvREFCNT_inc_simple_NN(sv))));
6607 /* process one subroutine argument - typically when the sub has a signature:
6608 * introduce PL_curpad[op_targ] and assign to it the value
6609 * for $: (OPf_STACKED ? *sp : $_[N])
6610 * for @/%: @_[N..$#_]
6612 * It's equivalent to
6615 * my $foo = (value-on-stack)
6617 * my @foo = @_[N..$#_]
6627 AV *defav = GvAV(PL_defgv); /* @_ */
6628 IV ix = PTR2IV(cUNOP_AUXo->op_aux);
6631 /* do 'my $var, @var or %var' action */
6632 padentry = &(PAD_SVl(o->op_targ));
6633 save_clearsv(padentry);
6636 if ((o->op_private & OPpARGELEM_MASK) == OPpARGELEM_SV) {
6637 if (o->op_flags & OPf_STACKED) {
6644 /* should already have been checked */
6646 #if IVSIZE > PTRSIZE
6647 assert(ix <= SSize_t_MAX);
6650 svp = av_fetch(defav, ix, FALSE);
6651 val = svp ? *svp : &PL_sv_undef;
6656 /* cargo-culted from pp_sassign */
6657 assert(TAINTING_get || !TAINT_get);
6658 if (UNLIKELY(TAINT_get) && !SvTAINTED(val))
6661 SvSetMagicSV(targ, val);
6665 /* must be AV or HV */
6667 assert(!(o->op_flags & OPf_STACKED));
6668 argc = ((IV)AvFILL(defav) + 1) - ix;
6670 /* This is a copy of the relevant parts of pp_aassign().
6672 if ((o->op_private & OPpARGELEM_MASK) == OPpARGELEM_AV) {
6675 if (AvFILL((AV*)targ) > -1) {
6676 /* target should usually be empty. If we get get
6677 * here, someone's been doing some weird closure tricks.
6678 * Make a copy of all args before clearing the array,
6679 * to avoid the equivalent of @a = ($a[0]) prematurely freeing
6680 * elements. See similar code in pp_aassign.
6682 for (i = 0; i < argc; i++) {
6683 SV **svp = av_fetch(defav, ix + i, FALSE);
6684 SV *newsv = newSV(0);
6685 sv_setsv_flags(newsv,
6686 svp ? *svp : &PL_sv_undef,
6687 (SV_DO_COW_SVSETSV|SV_NOSTEAL));
6688 if (!av_store(defav, ix + i, newsv))
6689 SvREFCNT_dec_NN(newsv);
6691 av_clear((AV*)targ);
6697 av_extend((AV*)targ, argc);
6702 SV **svp = av_fetch(defav, ix + i, FALSE);
6703 SV *val = svp ? *svp : &PL_sv_undef;
6705 sv_setsv(tmpsv, val);
6706 av_store((AV*)targ, i++, tmpsv);
6714 assert((o->op_private & OPpARGELEM_MASK) == OPpARGELEM_HV);
6716 if (SvRMAGICAL(targ) || HvUSEDKEYS((HV*)targ)) {
6717 /* see "target should usually be empty" comment above */
6718 for (i = 0; i < argc; i++) {
6719 SV **svp = av_fetch(defav, ix + i, FALSE);
6720 SV *newsv = newSV(0);
6721 sv_setsv_flags(newsv,
6722 svp ? *svp : &PL_sv_undef,
6723 (SV_DO_COW_SVSETSV|SV_NOSTEAL));
6724 if (!av_store(defav, ix + i, newsv))
6725 SvREFCNT_dec_NN(newsv);
6727 hv_clear((HV*)targ);
6732 assert(argc % 2 == 0);
6741 svp = av_fetch(defav, ix + i++, FALSE);
6742 key = svp ? *svp : &PL_sv_undef;
6743 svp = av_fetch(defav, ix + i++, FALSE);
6744 val = svp ? *svp : &PL_sv_undef;
6747 if (UNLIKELY(SvGMAGICAL(key)))
6748 key = sv_mortalcopy(key);
6750 sv_setsv(tmpsv, val);
6751 hv_store_ent((HV*)targ, key, tmpsv, 0);
6759 /* Handle a default value for one subroutine argument (typically as part
6760 * of a subroutine signature).
6761 * It's equivalent to
6762 * @_ > op_targ ? $_[op_targ] : result_of(op_other)
6764 * Intended to be used where op_next is an OP_ARGELEM
6766 * We abuse the op_targ field slightly: it's an index into @_ rather than
6772 OP * const o = PL_op;
6773 AV *defav = GvAV(PL_defgv); /* @_ */
6774 IV ix = (IV)o->op_targ;
6777 #if IVSIZE > PTRSIZE
6778 assert(ix <= SSize_t_MAX);
6781 if (AvFILL(defav) >= ix) {
6783 SV **svp = av_fetch(defav, ix, FALSE);
6784 SV *val = svp ? *svp : &PL_sv_undef;
6788 return cLOGOPo->op_other;
6793 S_find_runcv_name(void)
6808 sv = sv_2mortal(newSV(0));
6809 gv_fullname4(sv, gv, NULL, TRUE);
6813 /* Check a a subs arguments - i.e. that it has the correct number of args
6814 * (and anything else we might think of in future). Typically used with
6820 OP * const o = PL_op;
6821 UNOP_AUX_item *aux = cUNOP_AUXo->op_aux;
6822 IV params = aux[0].iv;
6823 IV opt_params = aux[1].iv;
6824 char slurpy = (char)(aux[2].iv);
6825 AV *defav = GvAV(PL_defgv); /* @_ */
6829 assert(!SvMAGICAL(defav));
6830 argc = (AvFILLp(defav) + 1);
6831 too_few = (argc < (params - opt_params));
6833 if (UNLIKELY(too_few || (!slurpy && argc > params)))
6834 /* diag_listed_as: Too few arguments for subroutine '%s' */
6835 /* diag_listed_as: Too many arguments for subroutine '%s' */
6836 Perl_croak_caller("Too %s arguments for subroutine '%" SVf "'",
6837 too_few ? "few" : "many", S_find_runcv_name());
6839 if (UNLIKELY(slurpy == '%' && argc > params && (argc - params) % 2))
6840 /* diag_listed_as: Odd name/value argument for subroutine '%s' */
6841 Perl_croak_caller("Odd name/value argument for subroutine '%" SVf "'",
6842 S_find_runcv_name());
6848 * ex: set ts=8 sts=4 sw=4 et: