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.
33 /* XXX I can't imagine anyone who doesn't have this actually _needs_
34 it, since pid_t is an integral type.
37 #ifdef NEED_GETPID_PROTO
38 extern Pid_t getpid (void);
42 * Some BSDs and Cygwin default to POSIX math instead of IEEE.
43 * This switches them over to IEEE.
45 #if defined(LIBM_LIB_VERSION)
46 _LIB_VERSION_TYPE _LIB_VERSION = _IEEE_;
49 /* variations on pp_null */
55 if (GIMME_V == G_SCALAR)
66 assert(SvTYPE(TARG) == SVt_PVAV);
67 if (PL_op->op_private & OPpLVAL_INTRO)
68 if (!(PL_op->op_private & OPpPAD_STATE))
69 SAVECLEARSV(PAD_SVl(PL_op->op_targ));
71 if (PL_op->op_flags & OPf_REF) {
75 if (GIMME == G_SCALAR)
76 Perl_croak(aTHX_ "Can't return array to lvalue scalar context");
81 if (gimme == G_ARRAY) {
82 const I32 maxarg = AvFILL(MUTABLE_AV(TARG)) + 1;
84 if (SvMAGICAL(TARG)) {
86 for (i=0; i < (U32)maxarg; i++) {
87 SV * const * const svp = av_fetch(MUTABLE_AV(TARG), i, FALSE);
88 SP[i+1] = (svp) ? *svp : &PL_sv_undef;
92 Copy(AvARRAY((const AV *)TARG), SP+1, maxarg, SV*);
96 else if (gimme == G_SCALAR) {
97 SV* const sv = sv_newmortal();
98 const I32 maxarg = AvFILL(MUTABLE_AV(TARG)) + 1;
110 assert(SvTYPE(TARG) == SVt_PVHV);
112 if (PL_op->op_private & OPpLVAL_INTRO)
113 if (!(PL_op->op_private & OPpPAD_STATE))
114 SAVECLEARSV(PAD_SVl(PL_op->op_targ));
115 if (PL_op->op_flags & OPf_REF)
118 if (GIMME == G_SCALAR)
119 Perl_croak(aTHX_ "Can't return hash to lvalue scalar context");
123 if (gimme == G_ARRAY) {
124 RETURNOP(Perl_do_kv(aTHX));
126 else if (gimme == G_SCALAR) {
127 SV* const sv = Perl_hv_scalar(aTHX_ MUTABLE_HV(TARG));
135 static const char S_no_symref_sv[] =
136 "Can't use string (\"%" SVf32 "\"%s) as %s ref while \"strict refs\" in use";
142 if (!isGV(sv) || SvFAKE(sv)) SvGETMAGIC(sv);
146 sv = amagic_deref_call(sv, to_gv_amg);
150 if (SvTYPE(sv) == SVt_PVIO) {
151 GV * const gv = MUTABLE_GV(sv_newmortal());
152 gv_init(gv, 0, "", 0, 0);
153 GvIOp(gv) = MUTABLE_IO(sv);
154 SvREFCNT_inc_void_NN(sv);
157 else if (!isGV_with_GP(sv))
158 DIE(aTHX_ "Not a GLOB reference");
161 if (!isGV_with_GP(sv)) {
162 if (!SvOK(sv) && sv != &PL_sv_undef) {
163 /* If this is a 'my' scalar and flag is set then vivify
167 Perl_croak_no_modify(aTHX);
168 if (PL_op->op_private & OPpDEREF) {
170 if (cUNOP->op_targ) {
172 SV * const namesv = PAD_SV(cUNOP->op_targ);
173 const char * const name = SvPV(namesv, len);
174 gv = MUTABLE_GV(newSV(0));
175 gv_init(gv, CopSTASH(PL_curcop), name, len, 0);
178 const char * const name = CopSTASHPV(PL_curcop);
181 prepare_SV_for_RV(sv);
182 SvRV_set(sv, MUTABLE_SV(gv));
187 if (PL_op->op_flags & OPf_REF ||
188 PL_op->op_private & HINT_STRICT_REFS)
189 DIE(aTHX_ PL_no_usym, "a symbol");
190 if (ckWARN(WARN_UNINITIALIZED))
194 if ((PL_op->op_flags & OPf_SPECIAL) &&
195 !(PL_op->op_flags & OPf_MOD))
197 SV * const temp = MUTABLE_SV(gv_fetchsv(sv, 0, SVt_PVGV));
199 && (!is_gv_magical_sv(sv,0)
200 || !(sv = MUTABLE_SV(gv_fetchsv(sv, GV_ADD,
207 if (PL_op->op_private & HINT_STRICT_REFS)
208 DIE(aTHX_ S_no_symref_sv, sv, (SvPOK(sv) && SvCUR(sv)>32 ? "..." : ""), "a symbol");
209 if ((PL_op->op_private & (OPpLVAL_INTRO|OPpDONT_INIT_GV))
210 == OPpDONT_INIT_GV) {
211 /* We are the target of a coderef assignment. Return
212 the scalar unchanged, and let pp_sasssign deal with
216 sv = MUTABLE_SV(gv_fetchsv(sv, GV_ADD, SVt_PVGV));
218 /* FAKE globs in the symbol table cause weird bugs (#77810) */
219 if (sv) SvFAKE_off(sv);
222 if (sv && SvFAKE(sv)) {
223 SV *newsv = sv_newmortal();
224 sv_setsv_flags(newsv, sv, 0);
228 if (PL_op->op_private & OPpLVAL_INTRO)
229 save_gp(MUTABLE_GV(sv), !(PL_op->op_flags & OPf_SPECIAL));
234 /* Helper function for pp_rv2sv and pp_rv2av */
236 Perl_softref2xv(pTHX_ SV *const sv, const char *const what,
237 const svtype type, SV ***spp)
242 PERL_ARGS_ASSERT_SOFTREF2XV;
244 if (PL_op->op_private & HINT_STRICT_REFS) {
246 Perl_die(aTHX_ S_no_symref_sv, sv, (SvPOK(sv) && SvCUR(sv)>32 ? "..." : ""), what);
248 Perl_die(aTHX_ PL_no_usym, what);
252 PL_op->op_flags & OPf_REF &&
253 PL_op->op_next->op_type != OP_BOOLKEYS
255 Perl_die(aTHX_ PL_no_usym, what);
256 if (ckWARN(WARN_UNINITIALIZED))
258 if (type != SVt_PV && GIMME_V == G_ARRAY) {
262 **spp = &PL_sv_undef;
265 if ((PL_op->op_flags & OPf_SPECIAL) &&
266 !(PL_op->op_flags & OPf_MOD))
268 gv = gv_fetchsv(sv, 0, type);
270 && (!is_gv_magical_sv(sv,0)
271 || !(gv = gv_fetchsv(sv, GV_ADD, type))))
273 **spp = &PL_sv_undef;
278 gv = gv_fetchsv(sv, GV_ADD, type);
288 if (!(PL_op->op_private & OPpDEREFed))
292 sv = amagic_deref_call(sv, to_sv_amg);
297 switch (SvTYPE(sv)) {
303 DIE(aTHX_ "Not a SCALAR reference");
310 if (!isGV_with_GP(gv)) {
311 gv = Perl_softref2xv(aTHX_ sv, "a SCALAR", SVt_PV, &sp);
317 if (PL_op->op_flags & OPf_MOD) {
318 if (PL_op->op_private & OPpLVAL_INTRO) {
319 if (cUNOP->op_first->op_type == OP_NULL)
320 sv = save_scalar(MUTABLE_GV(TOPs));
322 sv = save_scalar(gv);
324 Perl_croak(aTHX_ "%s", PL_no_localize_ref);
326 else if (PL_op->op_private & OPpDEREF)
327 vivify_ref(sv, PL_op->op_private & OPpDEREF);
336 AV * const av = MUTABLE_AV(TOPs);
337 const I32 lvalue = PL_op->op_flags & OPf_MOD || LVRET;
339 SV ** const sv = Perl_av_arylen_p(aTHX_ MUTABLE_AV(av));
341 *sv = newSV_type(SVt_PVMG);
342 sv_magic(*sv, MUTABLE_SV(av), PERL_MAGIC_arylen, NULL, 0);
346 SETs(sv_2mortal(newSViv(
347 AvFILL(MUTABLE_AV(av)) + CopARYBASE_get(PL_curcop)
357 if (PL_op->op_flags & OPf_MOD || LVRET) {
358 SV * const ret = sv_2mortal(newSV_type(SVt_PVLV)); /* Not TARG RT#67838 */
359 sv_magic(ret, NULL, PERL_MAGIC_pos, NULL, 0);
361 LvTARG(ret) = SvREFCNT_inc_simple(sv);
362 PUSHs(ret); /* no SvSETMAGIC */
366 if (SvTYPE(sv) >= SVt_PVMG && SvMAGIC(sv)) {
367 const MAGIC * const mg = mg_find(sv, PERL_MAGIC_regex_global);
368 if (mg && mg->mg_len >= 0) {
373 PUSHi(i + CopARYBASE_get(PL_curcop));
386 const I32 flags = (PL_op->op_flags & OPf_SPECIAL)
388 : ((PL_op->op_private & (OPpLVAL_INTRO|OPpMAY_RETURN_CONSTANT)) == OPpMAY_RETURN_CONSTANT)
391 /* We usually try to add a non-existent subroutine in case of AUTOLOAD. */
392 /* (But not in defined().) */
394 CV *cv = sv_2cv(TOPs, &stash_unused, &gv, flags);
397 cv = MUTABLE_CV(sv_2mortal(MUTABLE_SV(cv_clone(cv))));
398 if ((PL_op->op_private & OPpLVAL_INTRO)) {
399 if (gv && GvCV(gv) == cv && (gv = gv_autoload4(GvSTASH(gv), GvNAME(gv), GvNAMELEN(gv), FALSE)))
402 DIE(aTHX_ "Can't modify non-lvalue subroutine call");
405 else if ((flags == (GV_ADD|GV_NOEXPAND)) && gv && SvROK(gv)) {
409 cv = MUTABLE_CV(&PL_sv_undef);
410 SETs(MUTABLE_SV(cv));
420 SV *ret = &PL_sv_undef;
422 if (SvPOK(TOPs) && SvCUR(TOPs) >= 7) {
423 const char * s = SvPVX_const(TOPs);
424 if (strnEQ(s, "CORE::", 6)) {
425 const int code = keyword(s + 6, SvCUR(TOPs) - 6, 1);
426 if (code < 0) { /* Overridable. */
427 #define MAX_ARGS_OP ((sizeof(I32) - 1) * 2)
428 int i = 0, n = 0, seen_question = 0, defgv = 0;
430 char str[ MAX_ARGS_OP * 2 + 2 ]; /* One ';', one '\0' */
432 if (code == -KEY_chop || code == -KEY_chomp
433 || code == -KEY_exec || code == -KEY_system)
435 if (code == -KEY_mkdir) {
436 ret = newSVpvs_flags("_;$", SVs_TEMP);
439 if (code == -KEY_keys || code == -KEY_values || code == -KEY_each) {
440 ret = newSVpvs_flags("+", SVs_TEMP);
443 if (code == -KEY_push || code == -KEY_unshift) {
444 ret = newSVpvs_flags("+@", SVs_TEMP);
447 if (code == -KEY_pop || code == -KEY_shift) {
448 ret = newSVpvs_flags(";+", SVs_TEMP);
451 if (code == -KEY_splice) {
452 ret = newSVpvs_flags("+;$$@", SVs_TEMP);
455 if (code == -KEY_tied || code == -KEY_untie) {
456 ret = newSVpvs_flags("\\[$@%*]", SVs_TEMP);
459 if (code == -KEY_tie) {
460 ret = newSVpvs_flags("\\[$@%*]$@", SVs_TEMP);
463 if (code == -KEY_readpipe) {
464 s = "CORE::backtick";
466 while (i < MAXO) { /* The slow way. */
467 if (strEQ(s + 6, PL_op_name[i])
468 || strEQ(s + 6, PL_op_desc[i]))
474 goto nonesuch; /* Should not happen... */
476 defgv = PL_opargs[i] & OA_DEFGV;
477 oa = PL_opargs[i] >> OASHIFT;
479 if (oa & OA_OPTIONAL && !seen_question && !defgv) {
483 if ((oa & (OA_OPTIONAL - 1)) >= OA_AVREF
484 && (oa & (OA_OPTIONAL - 1)) <= OA_SCALARREF
485 /* But globs are already references (kinda) */
486 && (oa & (OA_OPTIONAL - 1)) != OA_FILEREF
490 str[n++] = ("?$@@%&*$")[oa & (OA_OPTIONAL - 1)];
493 if (defgv && str[n - 1] == '$')
496 ret = newSVpvn_flags(str, n - 1, SVs_TEMP);
498 else if (code) /* Non-Overridable */
500 else { /* None such */
502 DIE(aTHX_ "Can't find an opnumber for \"%s\"", s+6);
506 cv = sv_2cv(TOPs, &stash, &gv, 0);
508 ret = newSVpvn_flags(SvPVX_const(cv), SvCUR(cv), SVs_TEMP);
517 CV *cv = MUTABLE_CV(PAD_SV(PL_op->op_targ));
519 cv = MUTABLE_CV(sv_2mortal(MUTABLE_SV(cv_clone(cv))));
521 PUSHs(MUTABLE_SV(cv));
535 if (GIMME != G_ARRAY) {
539 *MARK = &PL_sv_undef;
540 *MARK = refto(*MARK);
544 EXTEND_MORTAL(SP - MARK);
546 *MARK = refto(*MARK);
551 S_refto(pTHX_ SV *sv)
556 PERL_ARGS_ASSERT_REFTO;
558 if (SvTYPE(sv) == SVt_PVLV && LvTYPE(sv) == 'y') {
561 if (!(sv = LvTARG(sv)))
564 SvREFCNT_inc_void_NN(sv);
566 else if (SvTYPE(sv) == SVt_PVAV) {
567 if (!AvREAL((const AV *)sv) && AvREIFY((const AV *)sv))
568 av_reify(MUTABLE_AV(sv));
570 SvREFCNT_inc_void_NN(sv);
572 else if (SvPADTMP(sv) && !IS_PADGV(sv))
576 SvREFCNT_inc_void_NN(sv);
579 sv_upgrade(rv, SVt_IV);
589 SV * const sv = POPs;
594 if (!sv || !SvROK(sv))
597 pv = sv_reftype(SvRV(sv),TRUE);
598 PUSHp(pv, strlen(pv));
608 stash = CopSTASH(PL_curcop);
610 SV * const ssv = POPs;
614 if (ssv && !SvGMAGICAL(ssv) && !SvAMAGIC(ssv) && SvROK(ssv))
615 Perl_croak(aTHX_ "Attempt to bless into a reference");
616 ptr = SvPV_const(ssv,len);
618 Perl_ck_warner(aTHX_ packWARN(WARN_MISC),
619 "Explicit blessing to '' (assuming package main)");
620 stash = gv_stashpvn(ptr, len, GV_ADD);
623 (void)sv_bless(TOPs, stash);
632 const char * const elem = SvPV_nolen_const(sv);
633 GV * const gv = MUTABLE_GV(POPs);
638 /* elem will always be NUL terminated. */
639 const char * const second_letter = elem + 1;
642 if (strEQ(second_letter, "RRAY"))
643 tmpRef = MUTABLE_SV(GvAV(gv));
646 if (strEQ(second_letter, "ODE"))
647 tmpRef = MUTABLE_SV(GvCVu(gv));
650 if (strEQ(second_letter, "ILEHANDLE")) {
651 /* finally deprecated in 5.8.0 */
652 deprecate("*glob{FILEHANDLE}");
653 tmpRef = MUTABLE_SV(GvIOp(gv));
656 if (strEQ(second_letter, "ORMAT"))
657 tmpRef = MUTABLE_SV(GvFORM(gv));
660 if (strEQ(second_letter, "LOB"))
661 tmpRef = MUTABLE_SV(gv);
664 if (strEQ(second_letter, "ASH"))
665 tmpRef = MUTABLE_SV(GvHV(gv));
668 if (*second_letter == 'O' && !elem[2])
669 tmpRef = MUTABLE_SV(GvIOp(gv));
672 if (strEQ(second_letter, "AME"))
673 sv = newSVhek(GvNAME_HEK(gv));
676 if (strEQ(second_letter, "ACKAGE")) {
677 const HV * const stash = GvSTASH(gv);
678 const HEK * const hek = stash ? HvNAME_HEK(stash) : NULL;
679 sv = hek ? newSVhek(hek) : newSVpvs("__ANON__");
683 if (strEQ(second_letter, "CALAR"))
698 /* Pattern matching */
703 register unsigned char *s;
706 register I32 *sfirst;
710 if (sv == PL_lastscream) {
714 s = (unsigned char*)(SvPV(sv, len));
716 if (pos <= 0 || !SvPOK(sv) || SvUTF8(sv)) {
717 /* No point in studying a zero length string, and not safe to study
718 anything that doesn't appear to be a simple scalar (and hence might
719 change between now and when the regexp engine runs without our set
720 magic ever running) such as a reference to an object with overloaded
726 SvSCREAM_off(PL_lastscream);
727 SvREFCNT_dec(PL_lastscream);
729 PL_lastscream = SvREFCNT_inc_simple(sv);
731 s = (unsigned char*)(SvPV(sv, len));
735 if (pos > PL_maxscream) {
736 if (PL_maxscream < 0) {
737 PL_maxscream = pos + 80;
738 Newx(PL_screamfirst, 256, I32);
739 Newx(PL_screamnext, PL_maxscream, I32);
742 PL_maxscream = pos + pos / 4;
743 Renew(PL_screamnext, PL_maxscream, I32);
747 sfirst = PL_screamfirst;
748 snext = PL_screamnext;
750 if (!sfirst || !snext)
751 DIE(aTHX_ "do_study: out of memory");
753 for (ch = 256; ch; --ch)
758 register const I32 ch = s[pos];
760 snext[pos] = sfirst[ch] - pos;
767 /* piggyback on m//g magic */
768 sv_magic(sv, NULL, PERL_MAGIC_regex_global, NULL, 0);
777 if (PL_op->op_flags & OPf_STACKED)
779 else if (PL_op->op_private & OPpTARGET_MY)
785 TARG = sv_newmortal();
786 if(PL_op->op_type == OP_TRANSR) {
787 SV * const newsv = newSVsv(sv);
791 else PUSHi(do_trans(sv));
795 /* Lvalue operators. */
798 S_do_chomp(pTHX_ SV *retval, SV *sv, bool chomping)
804 PERL_ARGS_ASSERT_DO_CHOMP;
806 if (chomping && (RsSNARF(PL_rs) || RsRECORD(PL_rs)))
808 if (SvTYPE(sv) == SVt_PVAV) {
810 AV *const av = MUTABLE_AV(sv);
811 const I32 max = AvFILL(av);
813 for (i = 0; i <= max; i++) {
814 sv = MUTABLE_SV(av_fetch(av, i, FALSE));
815 if (sv && ((sv = *(SV**)sv), sv != &PL_sv_undef))
816 do_chomp(retval, sv, chomping);
820 else if (SvTYPE(sv) == SVt_PVHV) {
821 HV* const hv = MUTABLE_HV(sv);
823 (void)hv_iterinit(hv);
824 while ((entry = hv_iternext(hv)))
825 do_chomp(retval, hv_iterval(hv,entry), chomping);
828 else if (SvREADONLY(sv)) {
830 /* SV is copy-on-write */
831 sv_force_normal_flags(sv, 0);
834 Perl_croak_no_modify(aTHX);
839 /* XXX, here sv is utf8-ized as a side-effect!
840 If encoding.pm is used properly, almost string-generating
841 operations, including literal strings, chr(), input data, etc.
842 should have been utf8-ized already, right?
844 sv_recode_to_utf8(sv, PL_encoding);
850 char *temp_buffer = NULL;
859 while (len && s[-1] == '\n') {
866 STRLEN rslen, rs_charlen;
867 const char *rsptr = SvPV_const(PL_rs, rslen);
869 rs_charlen = SvUTF8(PL_rs)
873 if (SvUTF8(PL_rs) != SvUTF8(sv)) {
874 /* Assumption is that rs is shorter than the scalar. */
876 /* RS is utf8, scalar is 8 bit. */
878 temp_buffer = (char*)bytes_from_utf8((U8*)rsptr,
881 /* Cannot downgrade, therefore cannot possibly match
883 assert (temp_buffer == rsptr);
889 else if (PL_encoding) {
890 /* RS is 8 bit, encoding.pm is used.
891 * Do not recode PL_rs as a side-effect. */
892 svrecode = newSVpvn(rsptr, rslen);
893 sv_recode_to_utf8(svrecode, PL_encoding);
894 rsptr = SvPV_const(svrecode, rslen);
895 rs_charlen = sv_len_utf8(svrecode);
898 /* RS is 8 bit, scalar is utf8. */
899 temp_buffer = (char*)bytes_to_utf8((U8*)rsptr, &rslen);
913 if (memNE(s, rsptr, rslen))
915 SvIVX(retval) += rs_charlen;
918 s = SvPV_force_nolen(sv);
926 SvREFCNT_dec(svrecode);
928 Safefree(temp_buffer);
930 if (len && !SvPOK(sv))
931 s = SvPV_force_nomg(sv, len);
934 char * const send = s + len;
935 char * const start = s;
937 while (s > start && UTF8_IS_CONTINUATION(*s))
939 if (is_utf8_string((U8*)s, send - s)) {
940 sv_setpvn(retval, s, send - s);
942 SvCUR_set(sv, s - start);
948 sv_setpvs(retval, "");
952 sv_setpvn(retval, s, 1);
959 sv_setpvs(retval, "");
967 const bool chomping = PL_op->op_type == OP_SCHOMP;
971 do_chomp(TARG, TOPs, chomping);
978 dVAR; dSP; dMARK; dTARGET; dORIGMARK;
979 const bool chomping = PL_op->op_type == OP_CHOMP;
984 do_chomp(TARG, *++MARK, chomping);
995 if (!PL_op->op_private) {
1004 SV_CHECK_THINKFIRST_COW_DROP(sv);
1006 switch (SvTYPE(sv)) {
1010 av_undef(MUTABLE_AV(sv));
1013 hv_undef(MUTABLE_HV(sv));
1016 if (cv_const_sv((const CV *)sv))
1017 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Constant subroutine %s undefined",
1018 CvANON((const CV *)sv) ? "(anonymous)"
1019 : GvENAME(CvGV((const CV *)sv)));
1023 /* let user-undef'd sub keep its identity */
1024 GV* const gv = CvGV((const CV *)sv);
1025 cv_undef(MUTABLE_CV(sv));
1026 CvGV_set(MUTABLE_CV(sv), gv);
1031 SvSetMagicSV(sv, &PL_sv_undef);
1034 else if (isGV_with_GP(sv)) {
1038 /* undef *Pkg::meth_name ... */
1040 = GvCVu((const GV *)sv) && (stash = GvSTASH((const GV *)sv))
1041 && HvENAME_get(stash);
1043 if((stash = GvHV((const GV *)sv))) {
1044 if(HvENAME_get(stash))
1045 SvREFCNT_inc_simple_void_NN(sv_2mortal((SV *)stash));
1049 gp_free(MUTABLE_GV(sv));
1051 GvGP_set(sv, gp_ref(gp));
1052 GvSV(sv) = newSV(0);
1053 GvLINE(sv) = CopLINE(PL_curcop);
1054 GvEGV(sv) = MUTABLE_GV(sv);
1058 mro_package_moved(NULL, stash, (const GV *)sv, 0);
1060 /* undef *Foo::ISA */
1061 if( strEQ(GvNAME((const GV *)sv), "ISA")
1062 && (stash = GvSTASH((const GV *)sv))
1063 && (method_changed || HvENAME(stash)) )
1064 mro_isa_changed_in(stash);
1065 else if(method_changed)
1066 mro_method_changed_in(
1067 GvSTASH((const GV *)sv)
1074 if (SvTYPE(sv) >= SVt_PV && SvPVX_const(sv) && SvLEN(sv)) {
1089 if (SvTYPE(TOPs) >= SVt_PVAV || isGV_with_GP(TOPs))
1090 Perl_croak_no_modify(aTHX);
1091 if (!SvREADONLY(TOPs) && SvIOK_notUV(TOPs) && !SvNOK(TOPs) && !SvPOK(TOPs)
1092 && SvIVX(TOPs) != IV_MIN)
1094 SvIV_set(TOPs, SvIVX(TOPs) - 1);
1095 SvFLAGS(TOPs) &= ~(SVp_NOK|SVp_POK);
1106 if (SvTYPE(TOPs) >= SVt_PVAV || isGV_with_GP(TOPs))
1107 Perl_croak_no_modify(aTHX);
1109 TARG = sv_newmortal();
1110 sv_setsv(TARG, TOPs);
1111 if (!SvREADONLY(TOPs) && SvIOK_notUV(TOPs) && !SvNOK(TOPs) && !SvPOK(TOPs)
1112 && SvIVX(TOPs) != IV_MAX)
1114 SvIV_set(TOPs, SvIVX(TOPs) + 1);
1115 SvFLAGS(TOPs) &= ~(SVp_NOK|SVp_POK);
1120 /* special case for undef: see thread at 2003-03/msg00536.html in archive */
1130 if (SvTYPE(TOPs) >= SVt_PVAV || isGV_with_GP(TOPs))
1131 Perl_croak_no_modify(aTHX);
1133 TARG = sv_newmortal();
1134 sv_setsv(TARG, TOPs);
1135 if (!SvREADONLY(TOPs) && SvIOK_notUV(TOPs) && !SvNOK(TOPs) && !SvPOK(TOPs)
1136 && SvIVX(TOPs) != IV_MIN)
1138 SvIV_set(TOPs, SvIVX(TOPs) - 1);
1139 SvFLAGS(TOPs) &= ~(SVp_NOK|SVp_POK);
1148 /* Ordinary operators. */
1152 dVAR; dSP; dATARGET; SV *svl, *svr;
1153 #ifdef PERL_PRESERVE_IVUV
1156 tryAMAGICbin_MG(pow_amg, AMGf_assign|AMGf_numeric);
1159 #ifdef PERL_PRESERVE_IVUV
1160 /* For integer to integer power, we do the calculation by hand wherever
1161 we're sure it is safe; otherwise we call pow() and try to convert to
1162 integer afterwards. */
1164 SvIV_please_nomg(svr);
1166 SvIV_please_nomg(svl);
1175 const IV iv = SvIVX(svr);
1179 goto float_it; /* Can't do negative powers this way. */
1183 baseuok = SvUOK(svl);
1185 baseuv = SvUVX(svl);
1187 const IV iv = SvIVX(svl);
1190 baseuok = TRUE; /* effectively it's a UV now */
1192 baseuv = -iv; /* abs, baseuok == false records sign */
1195 /* now we have integer ** positive integer. */
1198 /* foo & (foo - 1) is zero only for a power of 2. */
1199 if (!(baseuv & (baseuv - 1))) {
1200 /* We are raising power-of-2 to a positive integer.
1201 The logic here will work for any base (even non-integer
1202 bases) but it can be less accurate than
1203 pow (base,power) or exp (power * log (base)) when the
1204 intermediate values start to spill out of the mantissa.
1205 With powers of 2 we know this can't happen.
1206 And powers of 2 are the favourite thing for perl
1207 programmers to notice ** not doing what they mean. */
1209 NV base = baseuok ? baseuv : -(NV)baseuv;
1214 while (power >>= 1) {
1222 SvIV_please_nomg(svr);
1225 register unsigned int highbit = 8 * sizeof(UV);
1226 register unsigned int diff = 8 * sizeof(UV);
1227 while (diff >>= 1) {
1229 if (baseuv >> highbit) {
1233 /* we now have baseuv < 2 ** highbit */
1234 if (power * highbit <= 8 * sizeof(UV)) {
1235 /* result will definitely fit in UV, so use UV math
1236 on same algorithm as above */
1237 register UV result = 1;
1238 register UV base = baseuv;
1239 const bool odd_power = cBOOL(power & 1);
1243 while (power >>= 1) {
1250 if (baseuok || !odd_power)
1251 /* answer is positive */
1253 else if (result <= (UV)IV_MAX)
1254 /* answer negative, fits in IV */
1255 SETi( -(IV)result );
1256 else if (result == (UV)IV_MIN)
1257 /* 2's complement assumption: special case IV_MIN */
1260 /* answer negative, doesn't fit */
1261 SETn( -(NV)result );
1271 NV right = SvNV_nomg(svr);
1272 NV left = SvNV_nomg(svl);
1275 #if defined(USE_LONG_DOUBLE) && defined(HAS_AIX_POWL_NEG_BASE_BUG)
1277 We are building perl with long double support and are on an AIX OS
1278 afflicted with a powl() function that wrongly returns NaNQ for any
1279 negative base. This was reported to IBM as PMR #23047-379 on
1280 03/06/2006. The problem exists in at least the following versions
1281 of AIX and the libm fileset, and no doubt others as well:
1283 AIX 4.3.3-ML10 bos.adt.libm 4.3.3.50
1284 AIX 5.1.0-ML04 bos.adt.libm 5.1.0.29
1285 AIX 5.2.0 bos.adt.libm 5.2.0.85
1287 So, until IBM fixes powl(), we provide the following workaround to
1288 handle the problem ourselves. Our logic is as follows: for
1289 negative bases (left), we use fmod(right, 2) to check if the
1290 exponent is an odd or even integer:
1292 - if odd, powl(left, right) == -powl(-left, right)
1293 - if even, powl(left, right) == powl(-left, right)
1295 If the exponent is not an integer, the result is rightly NaNQ, so
1296 we just return that (as NV_NAN).
1300 NV mod2 = Perl_fmod( right, 2.0 );
1301 if (mod2 == 1.0 || mod2 == -1.0) { /* odd integer */
1302 SETn( -Perl_pow( -left, right) );
1303 } else if (mod2 == 0.0) { /* even integer */
1304 SETn( Perl_pow( -left, right) );
1305 } else { /* fractional power */
1309 SETn( Perl_pow( left, right) );
1312 SETn( Perl_pow( left, right) );
1313 #endif /* HAS_AIX_POWL_NEG_BASE_BUG */
1315 #ifdef PERL_PRESERVE_IVUV
1317 SvIV_please_nomg(svr);
1325 dVAR; dSP; dATARGET; SV *svl, *svr;
1326 tryAMAGICbin_MG(mult_amg, AMGf_assign|AMGf_numeric);
1329 #ifdef PERL_PRESERVE_IVUV
1330 SvIV_please_nomg(svr);
1332 /* Unless the left argument is integer in range we are going to have to
1333 use NV maths. Hence only attempt to coerce the right argument if
1334 we know the left is integer. */
1335 /* Left operand is defined, so is it IV? */
1336 SvIV_please_nomg(svl);
1338 bool auvok = SvUOK(svl);
1339 bool buvok = SvUOK(svr);
1340 const UV topmask = (~ (UV)0) << (4 * sizeof (UV));
1341 const UV botmask = ~((~ (UV)0) << (4 * sizeof (UV)));
1350 const IV aiv = SvIVX(svl);
1353 auvok = TRUE; /* effectively it's a UV now */
1355 alow = -aiv; /* abs, auvok == false records sign */
1361 const IV biv = SvIVX(svr);
1364 buvok = TRUE; /* effectively it's a UV now */
1366 blow = -biv; /* abs, buvok == false records sign */
1370 /* If this does sign extension on unsigned it's time for plan B */
1371 ahigh = alow >> (4 * sizeof (UV));
1373 bhigh = blow >> (4 * sizeof (UV));
1375 if (ahigh && bhigh) {
1377 /* eg 32 bit is at least 0x10000 * 0x10000 == 0x100000000
1378 which is overflow. Drop to NVs below. */
1379 } else if (!ahigh && !bhigh) {
1380 /* eg 32 bit is at most 0xFFFF * 0xFFFF == 0xFFFE0001
1381 so the unsigned multiply cannot overflow. */
1382 const UV product = alow * blow;
1383 if (auvok == buvok) {
1384 /* -ve * -ve or +ve * +ve gives a +ve result. */
1388 } else if (product <= (UV)IV_MIN) {
1389 /* 2s complement assumption that (UV)-IV_MIN is correct. */
1390 /* -ve result, which could overflow an IV */
1392 SETi( -(IV)product );
1394 } /* else drop to NVs below. */
1396 /* One operand is large, 1 small */
1399 /* swap the operands */
1401 bhigh = blow; /* bhigh now the temp var for the swap */
1405 /* now, ((ahigh * blow) << half_UV_len) + (alow * blow)
1406 multiplies can't overflow. shift can, add can, -ve can. */
1407 product_middle = ahigh * blow;
1408 if (!(product_middle & topmask)) {
1409 /* OK, (ahigh * blow) won't lose bits when we shift it. */
1411 product_middle <<= (4 * sizeof (UV));
1412 product_low = alow * blow;
1414 /* as for pp_add, UV + something mustn't get smaller.
1415 IIRC ANSI mandates this wrapping *behaviour* for
1416 unsigned whatever the actual representation*/
1417 product_low += product_middle;
1418 if (product_low >= product_middle) {
1419 /* didn't overflow */
1420 if (auvok == buvok) {
1421 /* -ve * -ve or +ve * +ve gives a +ve result. */
1423 SETu( product_low );
1425 } else if (product_low <= (UV)IV_MIN) {
1426 /* 2s complement assumption again */
1427 /* -ve result, which could overflow an IV */
1429 SETi( -(IV)product_low );
1431 } /* else drop to NVs below. */
1433 } /* product_middle too large */
1434 } /* ahigh && bhigh */
1439 NV right = SvNV_nomg(svr);
1440 NV left = SvNV_nomg(svl);
1442 SETn( left * right );
1449 dVAR; dSP; dATARGET; SV *svl, *svr;
1450 tryAMAGICbin_MG(div_amg, AMGf_assign|AMGf_numeric);
1453 /* Only try to do UV divide first
1454 if ((SLOPPYDIVIDE is true) or
1455 (PERL_PRESERVE_IVUV is true and one or both SV is a UV too large
1457 The assumption is that it is better to use floating point divide
1458 whenever possible, only doing integer divide first if we can't be sure.
1459 If NV_PRESERVES_UV is true then we know at compile time that no UV
1460 can be too large to preserve, so don't need to compile the code to
1461 test the size of UVs. */
1464 # define PERL_TRY_UV_DIVIDE
1465 /* ensure that 20./5. == 4. */
1467 # ifdef PERL_PRESERVE_IVUV
1468 # ifndef NV_PRESERVES_UV
1469 # define PERL_TRY_UV_DIVIDE
1474 #ifdef PERL_TRY_UV_DIVIDE
1475 SvIV_please_nomg(svr);
1477 SvIV_please_nomg(svl);
1479 bool left_non_neg = SvUOK(svl);
1480 bool right_non_neg = SvUOK(svr);
1484 if (right_non_neg) {
1488 const IV biv = SvIVX(svr);
1491 right_non_neg = TRUE; /* effectively it's a UV now */
1497 /* historically undef()/0 gives a "Use of uninitialized value"
1498 warning before dieing, hence this test goes here.
1499 If it were immediately before the second SvIV_please, then
1500 DIE() would be invoked before left was even inspected, so
1501 no inspection would give no warning. */
1503 DIE(aTHX_ "Illegal division by zero");
1509 const IV aiv = SvIVX(svl);
1512 left_non_neg = TRUE; /* effectively it's a UV now */
1521 /* For sloppy divide we always attempt integer division. */
1523 /* Otherwise we only attempt it if either or both operands
1524 would not be preserved by an NV. If both fit in NVs
1525 we fall through to the NV divide code below. However,
1526 as left >= right to ensure integer result here, we know that
1527 we can skip the test on the right operand - right big
1528 enough not to be preserved can't get here unless left is
1531 && (left > ((UV)1 << NV_PRESERVES_UV_BITS))
1534 /* Integer division can't overflow, but it can be imprecise. */
1535 const UV result = left / right;
1536 if (result * right == left) {
1537 SP--; /* result is valid */
1538 if (left_non_neg == right_non_neg) {
1539 /* signs identical, result is positive. */
1543 /* 2s complement assumption */
1544 if (result <= (UV)IV_MIN)
1545 SETi( -(IV)result );
1547 /* It's exact but too negative for IV. */
1548 SETn( -(NV)result );
1551 } /* tried integer divide but it was not an integer result */
1552 } /* else (PERL_ABS(result) < 1.0) or (both UVs in range for NV) */
1553 } /* left wasn't SvIOK */
1554 } /* right wasn't SvIOK */
1555 #endif /* PERL_TRY_UV_DIVIDE */
1557 NV right = SvNV_nomg(svr);
1558 NV left = SvNV_nomg(svl);
1559 (void)POPs;(void)POPs;
1560 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
1561 if (! Perl_isnan(right) && right == 0.0)
1565 DIE(aTHX_ "Illegal division by zero");
1566 PUSHn( left / right );
1573 dVAR; dSP; dATARGET;
1574 tryAMAGICbin_MG(modulo_amg, AMGf_assign|AMGf_numeric);
1578 bool left_neg = FALSE;
1579 bool right_neg = FALSE;
1580 bool use_double = FALSE;
1581 bool dright_valid = FALSE;
1584 SV * const svr = TOPs;
1585 SV * const svl = TOPm1s;
1586 SvIV_please_nomg(svr);
1588 right_neg = !SvUOK(svr);
1592 const IV biv = SvIVX(svr);
1595 right_neg = FALSE; /* effectively it's a UV now */
1602 dright = SvNV_nomg(svr);
1603 right_neg = dright < 0;
1606 if (dright < UV_MAX_P1) {
1607 right = U_V(dright);
1608 dright_valid = TRUE; /* In case we need to use double below. */
1614 /* At this point use_double is only true if right is out of range for
1615 a UV. In range NV has been rounded down to nearest UV and
1616 use_double false. */
1617 SvIV_please_nomg(svl);
1618 if (!use_double && SvIOK(svl)) {
1620 left_neg = !SvUOK(svl);
1624 const IV aiv = SvIVX(svl);
1627 left_neg = FALSE; /* effectively it's a UV now */
1635 dleft = SvNV_nomg(svl);
1636 left_neg = dleft < 0;
1640 /* This should be exactly the 5.6 behaviour - if left and right are
1641 both in range for UV then use U_V() rather than floor. */
1643 if (dleft < UV_MAX_P1) {
1644 /* right was in range, so is dleft, so use UVs not double.
1648 /* left is out of range for UV, right was in range, so promote
1649 right (back) to double. */
1651 /* The +0.5 is used in 5.6 even though it is not strictly
1652 consistent with the implicit +0 floor in the U_V()
1653 inside the #if 1. */
1654 dleft = Perl_floor(dleft + 0.5);
1657 dright = Perl_floor(dright + 0.5);
1668 DIE(aTHX_ "Illegal modulus zero");
1670 dans = Perl_fmod(dleft, dright);
1671 if ((left_neg != right_neg) && dans)
1672 dans = dright - dans;
1675 sv_setnv(TARG, dans);
1681 DIE(aTHX_ "Illegal modulus zero");
1684 if ((left_neg != right_neg) && ans)
1687 /* XXX may warn: unary minus operator applied to unsigned type */
1688 /* could change -foo to be (~foo)+1 instead */
1689 if (ans <= ~((UV)IV_MAX)+1)
1690 sv_setiv(TARG, ~ans+1);
1692 sv_setnv(TARG, -(NV)ans);
1695 sv_setuv(TARG, ans);
1704 dVAR; dSP; dATARGET;
1708 if (GIMME == G_ARRAY && PL_op->op_private & OPpREPEAT_DOLIST) {
1709 /* TODO: think of some way of doing list-repeat overloading ??? */
1714 tryAMAGICbin_MG(repeat_amg, AMGf_assign);
1720 const UV uv = SvUV_nomg(sv);
1722 count = IV_MAX; /* The best we can do? */
1726 const IV iv = SvIV_nomg(sv);
1733 else if (SvNOKp(sv)) {
1734 const NV nv = SvNV_nomg(sv);
1741 count = SvIV_nomg(sv);
1743 if (GIMME == G_ARRAY && PL_op->op_private & OPpREPEAT_DOLIST) {
1745 static const char oom_list_extend[] = "Out of memory during list extend";
1746 const I32 items = SP - MARK;
1747 const I32 max = items * count;
1749 MEM_WRAP_CHECK_1(max, SV*, oom_list_extend);
1750 /* Did the max computation overflow? */
1751 if (items > 0 && max > 0 && (max < items || max < count))
1752 Perl_croak(aTHX_ oom_list_extend);
1757 /* This code was intended to fix 20010809.028:
1760 for (($x =~ /./g) x 2) {
1761 print chop; # "abcdabcd" expected as output.
1764 * but that change (#11635) broke this code:
1766 $x = [("foo")x2]; # only one "foo" ended up in the anonlist.
1768 * I can't think of a better fix that doesn't introduce
1769 * an efficiency hit by copying the SVs. The stack isn't
1770 * refcounted, and mortalisation obviously doesn't
1771 * Do The Right Thing when the stack has more than
1772 * one pointer to the same mortal value.
1776 *SP = sv_2mortal(newSVsv(*SP));
1786 repeatcpy((char*)(MARK + items), (char*)MARK,
1787 items * sizeof(const SV *), count - 1);
1790 else if (count <= 0)
1793 else { /* Note: mark already snarfed by pp_list */
1794 SV * const tmpstr = POPs;
1797 static const char oom_string_extend[] =
1798 "Out of memory during string extend";
1801 sv_setsv_nomg(TARG, tmpstr);
1802 SvPV_force_nomg(TARG, len);
1803 isutf = DO_UTF8(TARG);
1808 const STRLEN max = (UV)count * len;
1809 if (len > MEM_SIZE_MAX / count)
1810 Perl_croak(aTHX_ oom_string_extend);
1811 MEM_WRAP_CHECK_1(max, char, oom_string_extend);
1812 SvGROW(TARG, max + 1);
1813 repeatcpy(SvPVX(TARG) + len, SvPVX(TARG), len, count - 1);
1814 SvCUR_set(TARG, SvCUR(TARG) * count);
1816 *SvEND(TARG) = '\0';
1819 (void)SvPOK_only_UTF8(TARG);
1821 (void)SvPOK_only(TARG);
1823 if (PL_op->op_private & OPpREPEAT_DOLIST) {
1824 /* The parser saw this as a list repeat, and there
1825 are probably several items on the stack. But we're
1826 in scalar context, and there's no pp_list to save us
1827 now. So drop the rest of the items -- robin@kitsite.com
1839 dVAR; dSP; dATARGET; bool useleft; SV *svl, *svr;
1840 tryAMAGICbin_MG(subtr_amg, AMGf_assign|AMGf_numeric);
1843 useleft = USE_LEFT(svl);
1844 #ifdef PERL_PRESERVE_IVUV
1845 /* See comments in pp_add (in pp_hot.c) about Overflow, and how
1846 "bad things" happen if you rely on signed integers wrapping. */
1847 SvIV_please_nomg(svr);
1849 /* Unless the left argument is integer in range we are going to have to
1850 use NV maths. Hence only attempt to coerce the right argument if
1851 we know the left is integer. */
1852 register UV auv = 0;
1858 a_valid = auvok = 1;
1859 /* left operand is undef, treat as zero. */
1861 /* Left operand is defined, so is it IV? */
1862 SvIV_please_nomg(svl);
1864 if ((auvok = SvUOK(svl)))
1867 register const IV aiv = SvIVX(svl);
1870 auvok = 1; /* Now acting as a sign flag. */
1871 } else { /* 2s complement assumption for IV_MIN */
1879 bool result_good = 0;
1882 bool buvok = SvUOK(svr);
1887 register const IV biv = SvIVX(svr);
1894 /* ?uvok if value is >= 0. basically, flagged as UV if it's +ve,
1895 else "IV" now, independent of how it came in.
1896 if a, b represents positive, A, B negative, a maps to -A etc
1901 all UV maths. negate result if A negative.
1902 subtract if signs same, add if signs differ. */
1904 if (auvok ^ buvok) {
1913 /* Must get smaller */
1918 if (result <= buv) {
1919 /* result really should be -(auv-buv). as its negation
1920 of true value, need to swap our result flag */
1932 if (result <= (UV)IV_MIN)
1933 SETi( -(IV)result );
1935 /* result valid, but out of range for IV. */
1936 SETn( -(NV)result );
1940 } /* Overflow, drop through to NVs. */
1945 NV value = SvNV_nomg(svr);
1949 /* left operand is undef, treat as zero - value */
1953 SETn( SvNV_nomg(svl) - value );
1960 dVAR; dSP; dATARGET; SV *svl, *svr;
1961 tryAMAGICbin_MG(lshift_amg, AMGf_assign|AMGf_numeric);
1965 const IV shift = SvIV_nomg(svr);
1966 if (PL_op->op_private & HINT_INTEGER) {
1967 const IV i = SvIV_nomg(svl);
1971 const UV u = SvUV_nomg(svl);
1980 dVAR; dSP; dATARGET; SV *svl, *svr;
1981 tryAMAGICbin_MG(rshift_amg, AMGf_assign|AMGf_numeric);
1985 const IV shift = SvIV_nomg(svr);
1986 if (PL_op->op_private & HINT_INTEGER) {
1987 const IV i = SvIV_nomg(svl);
1991 const UV u = SvUV_nomg(svl);
2001 tryAMAGICbin_MG(lt_amg, AMGf_set|AMGf_numeric);
2002 #ifdef PERL_PRESERVE_IVUV
2003 SvIV_please_nomg(TOPs);
2005 SvIV_please_nomg(TOPm1s);
2006 if (SvIOK(TOPm1s)) {
2007 bool auvok = SvUOK(TOPm1s);
2008 bool buvok = SvUOK(TOPs);
2010 if (!auvok && !buvok) { /* ## IV < IV ## */
2011 const IV aiv = SvIVX(TOPm1s);
2012 const IV biv = SvIVX(TOPs);
2015 SETs(boolSV(aiv < biv));
2018 if (auvok && buvok) { /* ## UV < UV ## */
2019 const UV auv = SvUVX(TOPm1s);
2020 const UV buv = SvUVX(TOPs);
2023 SETs(boolSV(auv < buv));
2026 if (auvok) { /* ## UV < IV ## */
2028 const IV biv = SvIVX(TOPs);
2031 /* As (a) is a UV, it's >=0, so it cannot be < */
2036 SETs(boolSV(auv < (UV)biv));
2039 { /* ## IV < UV ## */
2040 const IV aiv = SvIVX(TOPm1s);
2044 /* As (b) is a UV, it's >=0, so it must be < */
2051 SETs(boolSV((UV)aiv < buv));
2057 #ifndef NV_PRESERVES_UV
2058 #ifdef PERL_PRESERVE_IVUV
2061 if (SvROK(TOPs) && !SvAMAGIC(TOPs) && SvROK(TOPm1s) && !SvAMAGIC(TOPm1s)) {
2063 SETs(boolSV(SvRV(TOPs) < SvRV(TOPp1s)));
2068 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2070 if (Perl_isnan(left) || Perl_isnan(right))
2072 SETs(boolSV(left < right));
2075 SETs(boolSV(SvNV_nomg(TOPs) < value));
2084 tryAMAGICbin_MG(gt_amg, AMGf_set|AMGf_numeric);
2085 #ifdef PERL_PRESERVE_IVUV
2086 SvIV_please_nomg(TOPs);
2088 SvIV_please_nomg(TOPm1s);
2089 if (SvIOK(TOPm1s)) {
2090 bool auvok = SvUOK(TOPm1s);
2091 bool buvok = SvUOK(TOPs);
2093 if (!auvok && !buvok) { /* ## IV > IV ## */
2094 const IV aiv = SvIVX(TOPm1s);
2095 const IV biv = SvIVX(TOPs);
2098 SETs(boolSV(aiv > biv));
2101 if (auvok && buvok) { /* ## UV > UV ## */
2102 const UV auv = SvUVX(TOPm1s);
2103 const UV buv = SvUVX(TOPs);
2106 SETs(boolSV(auv > buv));
2109 if (auvok) { /* ## UV > IV ## */
2111 const IV biv = SvIVX(TOPs);
2115 /* As (a) is a UV, it's >=0, so it must be > */
2120 SETs(boolSV(auv > (UV)biv));
2123 { /* ## IV > UV ## */
2124 const IV aiv = SvIVX(TOPm1s);
2128 /* As (b) is a UV, it's >=0, so it cannot be > */
2135 SETs(boolSV((UV)aiv > buv));
2141 #ifndef NV_PRESERVES_UV
2142 #ifdef PERL_PRESERVE_IVUV
2145 if (SvROK(TOPs) && !SvAMAGIC(TOPs) && SvROK(TOPm1s) && !SvAMAGIC(TOPm1s)) {
2147 SETs(boolSV(SvRV(TOPs) > SvRV(TOPp1s)));
2152 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2154 if (Perl_isnan(left) || Perl_isnan(right))
2156 SETs(boolSV(left > right));
2159 SETs(boolSV(SvNV_nomg(TOPs) > value));
2168 tryAMAGICbin_MG(le_amg, AMGf_set|AMGf_numeric);
2169 #ifdef PERL_PRESERVE_IVUV
2170 SvIV_please_nomg(TOPs);
2172 SvIV_please_nomg(TOPm1s);
2173 if (SvIOK(TOPm1s)) {
2174 bool auvok = SvUOK(TOPm1s);
2175 bool buvok = SvUOK(TOPs);
2177 if (!auvok && !buvok) { /* ## IV <= IV ## */
2178 const IV aiv = SvIVX(TOPm1s);
2179 const IV biv = SvIVX(TOPs);
2182 SETs(boolSV(aiv <= biv));
2185 if (auvok && buvok) { /* ## UV <= UV ## */
2186 UV auv = SvUVX(TOPm1s);
2187 UV buv = SvUVX(TOPs);
2190 SETs(boolSV(auv <= buv));
2193 if (auvok) { /* ## UV <= IV ## */
2195 const IV biv = SvIVX(TOPs);
2199 /* As (a) is a UV, it's >=0, so a cannot be <= */
2204 SETs(boolSV(auv <= (UV)biv));
2207 { /* ## IV <= UV ## */
2208 const IV aiv = SvIVX(TOPm1s);
2212 /* As (b) is a UV, it's >=0, so a must be <= */
2219 SETs(boolSV((UV)aiv <= buv));
2225 #ifndef NV_PRESERVES_UV
2226 #ifdef PERL_PRESERVE_IVUV
2229 if (SvROK(TOPs) && !SvAMAGIC(TOPs) && SvROK(TOPm1s) && !SvAMAGIC(TOPm1s)) {
2231 SETs(boolSV(SvRV(TOPs) <= SvRV(TOPp1s)));
2236 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2238 if (Perl_isnan(left) || Perl_isnan(right))
2240 SETs(boolSV(left <= right));
2243 SETs(boolSV(SvNV_nomg(TOPs) <= value));
2252 tryAMAGICbin_MG(ge_amg,AMGf_set|AMGf_numeric);
2253 #ifdef PERL_PRESERVE_IVUV
2254 SvIV_please_nomg(TOPs);
2256 SvIV_please_nomg(TOPm1s);
2257 if (SvIOK(TOPm1s)) {
2258 bool auvok = SvUOK(TOPm1s);
2259 bool buvok = SvUOK(TOPs);
2261 if (!auvok && !buvok) { /* ## IV >= IV ## */
2262 const IV aiv = SvIVX(TOPm1s);
2263 const IV biv = SvIVX(TOPs);
2266 SETs(boolSV(aiv >= biv));
2269 if (auvok && buvok) { /* ## UV >= UV ## */
2270 const UV auv = SvUVX(TOPm1s);
2271 const UV buv = SvUVX(TOPs);
2274 SETs(boolSV(auv >= buv));
2277 if (auvok) { /* ## UV >= IV ## */
2279 const IV biv = SvIVX(TOPs);
2283 /* As (a) is a UV, it's >=0, so it must be >= */
2288 SETs(boolSV(auv >= (UV)biv));
2291 { /* ## IV >= UV ## */
2292 const IV aiv = SvIVX(TOPm1s);
2296 /* As (b) is a UV, it's >=0, so a cannot be >= */
2303 SETs(boolSV((UV)aiv >= buv));
2309 #ifndef NV_PRESERVES_UV
2310 #ifdef PERL_PRESERVE_IVUV
2313 if (SvROK(TOPs) && !SvAMAGIC(TOPs) && SvROK(TOPm1s) && !SvAMAGIC(TOPm1s)) {
2315 SETs(boolSV(SvRV(TOPs) >= SvRV(TOPp1s)));
2320 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2322 if (Perl_isnan(left) || Perl_isnan(right))
2324 SETs(boolSV(left >= right));
2327 SETs(boolSV(SvNV_nomg(TOPs) >= value));
2336 tryAMAGICbin_MG(ne_amg,AMGf_set|AMGf_numeric);
2337 #ifndef NV_PRESERVES_UV
2338 if (SvROK(TOPs) && !SvAMAGIC(TOPs) && SvROK(TOPm1s) && !SvAMAGIC(TOPm1s)) {
2340 SETs(boolSV(SvRV(TOPs) != SvRV(TOPp1s)));
2344 #ifdef PERL_PRESERVE_IVUV
2345 SvIV_please_nomg(TOPs);
2347 SvIV_please_nomg(TOPm1s);
2348 if (SvIOK(TOPm1s)) {
2349 const bool auvok = SvUOK(TOPm1s);
2350 const bool buvok = SvUOK(TOPs);
2352 if (auvok == buvok) { /* ## IV == IV or UV == UV ## */
2353 /* Casting IV to UV before comparison isn't going to matter
2354 on 2s complement. On 1s complement or sign&magnitude
2355 (if we have any of them) it could make negative zero
2356 differ from normal zero. As I understand it. (Need to
2357 check - is negative zero implementation defined behaviour
2359 const UV buv = SvUVX(POPs);
2360 const UV auv = SvUVX(TOPs);
2362 SETs(boolSV(auv != buv));
2365 { /* ## Mixed IV,UV ## */
2369 /* != is commutative so swap if needed (save code) */
2371 /* swap. top of stack (b) is the iv */
2375 /* As (a) is a UV, it's >0, so it cannot be == */
2384 /* As (b) is a UV, it's >0, so it cannot be == */
2388 uv = SvUVX(*(SP+1)); /* Do I want TOPp1s() ? */
2390 SETs(boolSV((UV)iv != uv));
2397 #if defined(NAN_COMPARE_BROKEN) && defined(Perl_isnan)
2399 if (Perl_isnan(left) || Perl_isnan(right))
2401 SETs(boolSV(left != right));
2404 SETs(boolSV(SvNV_nomg(TOPs) != value));
2413 tryAMAGICbin_MG(ncmp_amg, AMGf_numeric);
2414 #ifndef NV_PRESERVES_UV
2415 if (SvROK(TOPs) && !SvAMAGIC(TOPs) && SvROK(TOPm1s) && !SvAMAGIC(TOPm1s)) {
2416 const UV right = PTR2UV(SvRV(POPs));
2417 const UV left = PTR2UV(SvRV(TOPs));
2418 SETi((left > right) - (left < right));
2422 #ifdef PERL_PRESERVE_IVUV
2423 /* Fortunately it seems NaN isn't IOK */
2424 SvIV_please_nomg(TOPs);
2426 SvIV_please_nomg(TOPm1s);
2427 if (SvIOK(TOPm1s)) {
2428 const bool leftuvok = SvUOK(TOPm1s);
2429 const bool rightuvok = SvUOK(TOPs);
2431 if (!leftuvok && !rightuvok) { /* ## IV <=> IV ## */
2432 const IV leftiv = SvIVX(TOPm1s);
2433 const IV rightiv = SvIVX(TOPs);
2435 if (leftiv > rightiv)
2437 else if (leftiv < rightiv)
2441 } else if (leftuvok && rightuvok) { /* ## UV <=> UV ## */
2442 const UV leftuv = SvUVX(TOPm1s);
2443 const UV rightuv = SvUVX(TOPs);
2445 if (leftuv > rightuv)
2447 else if (leftuv < rightuv)
2451 } else if (leftuvok) { /* ## UV <=> IV ## */
2452 const IV rightiv = SvIVX(TOPs);
2454 /* As (a) is a UV, it's >=0, so it cannot be < */
2457 const UV leftuv = SvUVX(TOPm1s);
2458 if (leftuv > (UV)rightiv) {
2460 } else if (leftuv < (UV)rightiv) {
2466 } else { /* ## IV <=> UV ## */
2467 const IV leftiv = SvIVX(TOPm1s);
2469 /* As (b) is a UV, it's >=0, so it must be < */
2472 const UV rightuv = SvUVX(TOPs);
2473 if ((UV)leftiv > rightuv) {
2475 } else if ((UV)leftiv < rightuv) {
2493 if (Perl_isnan(left) || Perl_isnan(right)) {
2497 value = (left > right) - (left < right);
2501 else if (left < right)
2503 else if (left > right)
2519 int amg_type = sle_amg;
2523 switch (PL_op->op_type) {
2542 tryAMAGICbin_MG(amg_type, AMGf_set);
2545 const int cmp = (IN_LOCALE_RUNTIME
2546 ? sv_cmp_locale_flags(left, right, 0)
2547 : sv_cmp_flags(left, right, 0));
2548 SETs(boolSV(cmp * multiplier < rhs));
2556 tryAMAGICbin_MG(seq_amg, AMGf_set);
2559 SETs(boolSV(sv_eq_flags(left, right, 0)));
2567 tryAMAGICbin_MG(sne_amg, AMGf_set);
2570 SETs(boolSV(!sv_eq_flags(left, right, 0)));
2578 tryAMAGICbin_MG(scmp_amg, 0);
2581 const int cmp = (IN_LOCALE_RUNTIME
2582 ? sv_cmp_locale_flags(left, right, 0)
2583 : sv_cmp_flags(left, right, 0));
2591 dVAR; dSP; dATARGET;
2592 tryAMAGICbin_MG(band_amg, AMGf_assign);
2595 if (SvNIOKp(left) || SvNIOKp(right)) {
2596 const bool left_ro_nonnum = !SvNIOKp(left) && SvREADONLY(left);
2597 const bool right_ro_nonnum = !SvNIOKp(right) && SvREADONLY(right);
2598 if (PL_op->op_private & HINT_INTEGER) {
2599 const IV i = SvIV_nomg(left) & SvIV_nomg(right);
2603 const UV u = SvUV_nomg(left) & SvUV_nomg(right);
2606 if (left_ro_nonnum) SvNIOK_off(left);
2607 if (right_ro_nonnum) SvNIOK_off(right);
2610 do_vop(PL_op->op_type, TARG, left, right);
2619 dVAR; dSP; dATARGET;
2620 const int op_type = PL_op->op_type;
2622 tryAMAGICbin_MG((op_type == OP_BIT_OR ? bor_amg : bxor_amg), AMGf_assign);
2625 if (SvNIOKp(left) || SvNIOKp(right)) {
2626 const bool left_ro_nonnum = !SvNIOKp(left) && SvREADONLY(left);
2627 const bool right_ro_nonnum = !SvNIOKp(right) && SvREADONLY(right);
2628 if (PL_op->op_private & HINT_INTEGER) {
2629 const IV l = (USE_LEFT(left) ? SvIV_nomg(left) : 0);
2630 const IV r = SvIV_nomg(right);
2631 const IV result = op_type == OP_BIT_OR ? (l | r) : (l ^ r);
2635 const UV l = (USE_LEFT(left) ? SvUV_nomg(left) : 0);
2636 const UV r = SvUV_nomg(right);
2637 const UV result = op_type == OP_BIT_OR ? (l | r) : (l ^ r);
2640 if (left_ro_nonnum) SvNIOK_off(left);
2641 if (right_ro_nonnum) SvNIOK_off(right);
2644 do_vop(op_type, TARG, left, right);
2654 tryAMAGICun_MG(neg_amg, AMGf_numeric);
2656 SV * const sv = TOPs;
2657 const int flags = SvFLAGS(sv);
2659 if( !SvNIOK( sv ) && looks_like_number( sv ) ){
2663 if ((flags & SVf_IOK) || ((flags & (SVp_IOK | SVp_NOK)) == SVp_IOK)) {
2664 /* It's publicly an integer, or privately an integer-not-float */
2667 if (SvIVX(sv) == IV_MIN) {
2668 /* 2s complement assumption. */
2669 SETi(SvIVX(sv)); /* special case: -((UV)IV_MAX+1) == IV_MIN */
2672 else if (SvUVX(sv) <= IV_MAX) {
2677 else if (SvIVX(sv) != IV_MIN) {
2681 #ifdef PERL_PRESERVE_IVUV
2689 SETn(-SvNV_nomg(sv));
2690 else if (SvPOKp(sv)) {
2692 const char * const s = SvPV_nomg_const(sv, len);
2693 if (isIDFIRST(*s)) {
2694 sv_setpvs(TARG, "-");
2697 else if (*s == '+' || *s == '-') {
2698 sv_setsv_nomg(TARG, sv);
2699 *SvPV_force_nomg(TARG, len) = *s == '-' ? '+' : '-';
2701 else if (DO_UTF8(sv)) {
2702 SvIV_please_nomg(sv);
2704 goto oops_its_an_int;
2706 sv_setnv(TARG, -SvNV_nomg(sv));
2708 sv_setpvs(TARG, "-");
2713 SvIV_please_nomg(sv);
2715 goto oops_its_an_int;
2716 sv_setnv(TARG, -SvNV_nomg(sv));
2721 SETn(-SvNV_nomg(sv));
2729 tryAMAGICun_MG(not_amg, AMGf_set);
2730 *PL_stack_sp = boolSV(!SvTRUE_nomg(*PL_stack_sp));
2737 tryAMAGICun_MG(compl_amg, AMGf_numeric);
2741 if (PL_op->op_private & HINT_INTEGER) {
2742 const IV i = ~SvIV_nomg(sv);
2746 const UV u = ~SvUV_nomg(sv);
2755 (void)SvPV_nomg_const(sv,len); /* force check for uninit var */
2756 sv_setsv_nomg(TARG, sv);
2757 tmps = (U8*)SvPV_force_nomg(TARG, len);
2760 /* Calculate exact length, let's not estimate. */
2765 U8 * const send = tmps + len;
2766 U8 * const origtmps = tmps;
2767 const UV utf8flags = UTF8_ALLOW_ANYUV;
2769 while (tmps < send) {
2770 const UV c = utf8n_to_uvchr(tmps, send-tmps, &l, utf8flags);
2772 targlen += UNISKIP(~c);
2778 /* Now rewind strings and write them. */
2785 Newx(result, targlen + 1, U8);
2787 while (tmps < send) {
2788 const UV c = utf8n_to_uvchr(tmps, send-tmps, &l, utf8flags);
2790 p = uvchr_to_utf8_flags(p, ~c, UNICODE_ALLOW_ANY);
2793 sv_usepvn_flags(TARG, (char*)result, targlen,
2794 SV_HAS_TRAILING_NUL);
2801 Newx(result, nchar + 1, U8);
2803 while (tmps < send) {
2804 const U8 c = (U8)utf8n_to_uvchr(tmps, send-tmps, &l, utf8flags);
2809 sv_usepvn_flags(TARG, (char*)result, nchar, SV_HAS_TRAILING_NUL);
2817 register long *tmpl;
2818 for ( ; anum && (unsigned long)tmps % sizeof(long); anum--, tmps++)
2821 for ( ; anum >= (I32)sizeof(long); anum -= (I32)sizeof(long), tmpl++)
2826 for ( ; anum > 0; anum--, tmps++)
2834 /* integer versions of some of the above */
2838 dVAR; dSP; dATARGET;
2839 tryAMAGICbin_MG(mult_amg, AMGf_assign);
2842 SETi( left * right );
2849 dVAR; dSP; dATARGET;
2850 tryAMAGICbin_MG(div_amg, AMGf_assign);
2853 IV num = SvIV_nomg(left);
2854 IV value = left==right ? SvIV(right) : SvIV_nomg(right);
2856 DIE(aTHX_ "Illegal division by zero");
2858 /* avoid FPE_INTOVF on some platforms when num is IV_MIN */
2862 value = num / value;
2868 #if defined(__GLIBC__) && IVSIZE == 8
2875 /* This is the vanilla old i_modulo. */
2876 dVAR; dSP; dATARGET;
2877 tryAMAGICbin_MG(modulo_amg, AMGf_assign);
2881 DIE(aTHX_ "Illegal modulus zero");
2882 /* avoid FPE_INTOVF on some platforms when left is IV_MIN */
2886 SETi( left % right );
2891 #if defined(__GLIBC__) && IVSIZE == 8
2896 /* This is the i_modulo with the workaround for the _moddi3 bug
2897 * in (at least) glibc 2.2.5 (the PERL_ABS() the workaround).
2898 * See below for pp_i_modulo. */
2899 dVAR; dSP; dATARGET;
2900 tryAMAGICbin_MG(modulo_amg, AMGf_assign);
2904 DIE(aTHX_ "Illegal modulus zero");
2905 /* avoid FPE_INTOVF on some platforms when left is IV_MIN */
2909 SETi( left % PERL_ABS(right) );
2916 dVAR; dSP; dATARGET;
2917 tryAMAGICbin_MG(modulo_amg, AMGf_assign);
2921 DIE(aTHX_ "Illegal modulus zero");
2922 /* The assumption is to use hereafter the old vanilla version... */
2924 PL_ppaddr[OP_I_MODULO] =
2926 /* .. but if we have glibc, we might have a buggy _moddi3
2927 * (at least glicb 2.2.5 is known to have this bug), in other
2928 * words our integer modulus with negative quad as the second
2929 * argument might be broken. Test for this and re-patch the
2930 * opcode dispatch table if that is the case, remembering to
2931 * also apply the workaround so that this first round works
2932 * right, too. See [perl #9402] for more information. */
2936 /* Cannot do this check with inlined IV constants since
2937 * that seems to work correctly even with the buggy glibc. */
2939 /* Yikes, we have the bug.
2940 * Patch in the workaround version. */
2942 PL_ppaddr[OP_I_MODULO] =
2943 &Perl_pp_i_modulo_1;
2944 /* Make certain we work right this time, too. */
2945 right = PERL_ABS(right);
2948 /* avoid FPE_INTOVF on some platforms when left is IV_MIN */
2952 SETi( left % right );
2960 dVAR; dSP; dATARGET;
2961 tryAMAGICbin_MG(add_amg, AMGf_assign);
2963 dPOPTOPiirl_ul_nomg;
2964 SETi( left + right );
2971 dVAR; dSP; dATARGET;
2972 tryAMAGICbin_MG(subtr_amg, AMGf_assign);
2974 dPOPTOPiirl_ul_nomg;
2975 SETi( left - right );
2983 tryAMAGICbin_MG(lt_amg, AMGf_set);
2986 SETs(boolSV(left < right));
2994 tryAMAGICbin_MG(gt_amg, AMGf_set);
2997 SETs(boolSV(left > right));
3005 tryAMAGICbin_MG(le_amg, AMGf_set);
3008 SETs(boolSV(left <= right));
3016 tryAMAGICbin_MG(ge_amg, AMGf_set);
3019 SETs(boolSV(left >= right));
3027 tryAMAGICbin_MG(eq_amg, AMGf_set);
3030 SETs(boolSV(left == right));
3038 tryAMAGICbin_MG(ne_amg, AMGf_set);
3041 SETs(boolSV(left != right));
3049 tryAMAGICbin_MG(ncmp_amg, 0);
3056 else if (left < right)
3068 tryAMAGICun_MG(neg_amg, 0);
3070 SV * const sv = TOPs;
3071 IV const i = SvIV_nomg(sv);
3077 /* High falutin' math. */
3082 tryAMAGICbin_MG(atan2_amg, 0);
3085 SETn(Perl_atan2(left, right));
3093 int amg_type = sin_amg;
3094 const char *neg_report = NULL;
3095 NV (*func)(NV) = Perl_sin;
3096 const int op_type = PL_op->op_type;
3113 amg_type = sqrt_amg;
3115 neg_report = "sqrt";
3120 tryAMAGICun_MG(amg_type, 0);
3122 SV * const arg = POPs;
3123 const NV value = SvNV_nomg(arg);
3125 if (op_type == OP_LOG ? (value <= 0.0) : (value < 0.0)) {
3126 SET_NUMERIC_STANDARD();
3127 DIE(aTHX_ "Can't take %s of %"NVgf, neg_report, value);
3130 XPUSHn(func(value));
3135 /* Support Configure command-line overrides for rand() functions.
3136 After 5.005, perhaps we should replace this by Configure support
3137 for drand48(), random(), or rand(). For 5.005, though, maintain
3138 compatibility by calling rand() but allow the user to override it.
3139 See INSTALL for details. --Andy Dougherty 15 July 1998
3141 /* Now it's after 5.005, and Configure supports drand48() and random(),
3142 in addition to rand(). So the overrides should not be needed any more.
3143 --Jarkko Hietaniemi 27 September 1998
3146 #ifndef HAS_DRAND48_PROTO
3147 extern double drand48 (void);
3160 if (!PL_srand_called) {
3161 (void)seedDrand01((Rand_seed_t)seed());
3162 PL_srand_called = TRUE;
3172 const UV anum = (MAXARG < 1) ? seed() : POPu;
3173 (void)seedDrand01((Rand_seed_t)anum);
3174 PL_srand_called = TRUE;
3178 /* Historically srand always returned true. We can avoid breaking
3180 sv_setpvs(TARG, "0 but true");
3189 tryAMAGICun_MG(int_amg, AMGf_numeric);
3191 SV * const sv = TOPs;
3192 const IV iv = SvIV_nomg(sv);
3193 /* XXX it's arguable that compiler casting to IV might be subtly
3194 different from modf (for numbers inside (IV_MIN,UV_MAX)) in which
3195 else preferring IV has introduced a subtle behaviour change bug. OTOH
3196 relying on floating point to be accurate is a bug. */
3201 else if (SvIOK(sv)) {
3203 SETu(SvUV_nomg(sv));
3208 const NV value = SvNV_nomg(sv);
3210 if (value < (NV)UV_MAX + 0.5) {
3213 SETn(Perl_floor(value));
3217 if (value > (NV)IV_MIN - 0.5) {
3220 SETn(Perl_ceil(value));
3231 tryAMAGICun_MG(abs_amg, AMGf_numeric);
3233 SV * const sv = TOPs;
3234 /* This will cache the NV value if string isn't actually integer */
3235 const IV iv = SvIV_nomg(sv);
3240 else if (SvIOK(sv)) {
3241 /* IVX is precise */
3243 SETu(SvUV_nomg(sv)); /* force it to be numeric only */
3251 /* 2s complement assumption. Also, not really needed as
3252 IV_MIN and -IV_MIN should both be %100...00 and NV-able */
3258 const NV value = SvNV_nomg(sv);
3272 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES;
3276 SV* const sv = POPs;
3278 tmps = (SvPV_const(sv, len));
3280 /* If Unicode, try to downgrade
3281 * If not possible, croak. */
3282 SV* const tsv = sv_2mortal(newSVsv(sv));
3285 sv_utf8_downgrade(tsv, FALSE);
3286 tmps = SvPV_const(tsv, len);
3288 if (PL_op->op_type == OP_HEX)
3291 while (*tmps && len && isSPACE(*tmps))
3295 if (*tmps == 'x' || *tmps == 'X') {
3297 result_uv = grok_hex (tmps, &len, &flags, &result_nv);
3299 else if (*tmps == 'b' || *tmps == 'B')
3300 result_uv = grok_bin (tmps, &len, &flags, &result_nv);
3302 result_uv = grok_oct (tmps, &len, &flags, &result_nv);
3304 if (flags & PERL_SCAN_GREATER_THAN_UV_MAX) {
3318 SV * const sv = TOPs;
3320 if (SvGAMAGIC(sv)) {
3321 /* For an overloaded or magic scalar, we can't know in advance if
3322 it's going to be UTF-8 or not. Also, we can't call sv_len_utf8 as
3323 it likes to cache the length. Maybe that should be a documented
3328 = sv_2pv_flags(sv, &len,
3329 SV_UNDEF_RETURNS_NULL|SV_CONST_RETURN|SV_GMAGIC);
3332 if (!SvPADTMP(TARG)) {
3333 sv_setsv(TARG, &PL_sv_undef);
3338 else if (DO_UTF8(sv)) {
3339 SETi(utf8_length((U8*)p, (U8*)p + len));
3343 } else if (SvOK(sv)) {
3344 /* Neither magic nor overloaded. */
3346 SETi(sv_len_utf8(sv));
3350 if (!SvPADTMP(TARG)) {
3351 sv_setsv_nomg(TARG, &PL_sv_undef);
3373 const I32 lvalue = PL_op->op_flags & OPf_MOD || LVRET;
3375 const IV arybase = CopARYBASE_get(PL_curcop);
3377 const char *repl = NULL;
3379 const int num_args = PL_op->op_private & 7;
3380 bool repl_need_utf8_upgrade = FALSE;
3381 bool repl_is_utf8 = FALSE;
3386 repl = SvPV_const(repl_sv, repl_len);
3387 repl_is_utf8 = DO_UTF8(repl_sv) && SvCUR(repl_sv);
3390 len_iv = SvIV(len_sv);
3391 len_is_uv = SvIOK_UV(len_sv);
3394 pos1_iv = SvIV(pos_sv);
3395 pos1_is_uv = SvIOK_UV(pos_sv);
3401 sv_utf8_upgrade(sv);
3403 else if (DO_UTF8(sv))
3404 repl_need_utf8_upgrade = TRUE;
3406 tmps = SvPV_const(sv, curlen);
3408 utf8_curlen = sv_len_utf8(sv);
3409 if (utf8_curlen == curlen)
3412 curlen = utf8_curlen;
3417 if ( (pos1_is_uv && arybase < 0) || (pos1_iv >= arybase) ) { /* pos >= $[ */
3418 UV pos1_uv = pos1_iv-arybase;
3419 /* Overflow can occur when $[ < 0 */
3420 if (arybase < 0 && pos1_uv < (UV)pos1_iv)
3425 else if (pos1_is_uv ? (UV)pos1_iv > 0 : pos1_iv > 0) {
3426 goto bound_fail; /* $[=3; substr($_,2,...) */
3428 else { /* pos < $[ */
3429 if (pos1_iv == 0) { /* $[=1; substr($_,0,...) */
3434 pos1_is_uv = curlen-1 > ~(UV)pos1_iv;
3439 if (pos1_is_uv || pos1_iv > 0) {
3440 if ((UV)pos1_iv > curlen)
3445 if (!len_is_uv && len_iv < 0) {
3446 pos2_iv = curlen + len_iv;
3448 pos2_is_uv = curlen-1 > ~(UV)len_iv;
3451 } else { /* len_iv >= 0 */
3452 if (!pos1_is_uv && pos1_iv < 0) {
3453 pos2_iv = pos1_iv + len_iv;
3454 pos2_is_uv = (UV)len_iv > (UV)IV_MAX;
3456 if ((UV)len_iv > curlen-(UV)pos1_iv)
3459 pos2_iv = pos1_iv+len_iv;
3469 if (!pos2_is_uv && pos2_iv < 0) {
3470 if (!pos1_is_uv && pos1_iv < 0)
3474 else if (!pos1_is_uv && pos1_iv < 0)
3477 if ((UV)pos2_iv < (UV)pos1_iv)
3479 if ((UV)pos2_iv > curlen)
3483 /* pos1_iv and pos2_iv both in 0..curlen, so the cast is safe */
3484 const STRLEN pos = (STRLEN)( (UV)pos1_iv );
3485 const STRLEN len = (STRLEN)( (UV)pos2_iv - (UV)pos1_iv );
3486 STRLEN byte_len = len;
3487 STRLEN byte_pos = utf8_curlen
3488 ? sv_pos_u2b_flags(sv, pos, &byte_len, SV_CONST_RETURN) : pos;
3490 if (lvalue && !repl) {
3493 if (!SvGMAGICAL(sv)) {
3495 SvPV_force_nolen(sv);
3496 Perl_ck_warner(aTHX_ packWARN(WARN_SUBSTR),
3497 "Attempt to use reference as lvalue in substr");
3499 if (isGV_with_GP(sv))
3500 SvPV_force_nolen(sv);
3501 else if (SvOK(sv)) /* is it defined ? */
3502 (void)SvPOK_only_UTF8(sv);
3504 sv_setpvs(sv, ""); /* avoid lexical reincarnation */
3507 ret = sv_2mortal(newSV_type(SVt_PVLV)); /* Not TARG RT#67838 */
3508 sv_magic(ret, NULL, PERL_MAGIC_substr, NULL, 0);
3510 LvTARG(ret) = SvREFCNT_inc_simple(sv);
3511 LvTARGOFF(ret) = pos;
3512 LvTARGLEN(ret) = len;
3515 PUSHs(ret); /* avoid SvSETMAGIC here */
3519 SvTAINTED_off(TARG); /* decontaminate */
3520 SvUTF8_off(TARG); /* decontaminate */
3523 sv_setpvn(TARG, tmps, byte_len);
3524 #ifdef USE_LOCALE_COLLATE
3525 sv_unmagic(TARG, PERL_MAGIC_collxfrm);
3531 SV* repl_sv_copy = NULL;
3533 if (repl_need_utf8_upgrade) {
3534 repl_sv_copy = newSVsv(repl_sv);
3535 sv_utf8_upgrade(repl_sv_copy);
3536 repl = SvPV_const(repl_sv_copy, repl_len);
3537 repl_is_utf8 = DO_UTF8(repl_sv_copy) && SvCUR(sv);
3541 sv_insert_flags(sv, byte_pos, byte_len, repl, repl_len, 0);
3544 SvREFCNT_dec(repl_sv_copy);
3554 Perl_croak(aTHX_ "substr outside of string");
3555 Perl_ck_warner(aTHX_ packWARN(WARN_SUBSTR), "substr outside of string");
3562 register const IV size = POPi;
3563 register const IV offset = POPi;
3564 register SV * const src = POPs;
3565 const I32 lvalue = PL_op->op_flags & OPf_MOD || LVRET;
3568 if (lvalue) { /* it's an lvalue! */
3569 ret = sv_2mortal(newSV_type(SVt_PVLV)); /* Not TARG RT#67838 */
3570 sv_magic(ret, NULL, PERL_MAGIC_vec, NULL, 0);
3572 LvTARG(ret) = SvREFCNT_inc_simple(src);
3573 LvTARGOFF(ret) = offset;
3574 LvTARGLEN(ret) = size;
3578 SvTAINTED_off(TARG); /* decontaminate */
3582 sv_setuv(ret, do_vecget(src, offset, size));
3598 const char *little_p;
3599 const I32 arybase = CopARYBASE_get(PL_curcop);
3602 const bool is_index = PL_op->op_type == OP_INDEX;
3605 /* arybase is in characters, like offset, so combine prior to the
3606 UTF-8 to bytes calculation. */
3607 offset = POPi - arybase;
3611 big_p = SvPV_const(big, biglen);
3612 little_p = SvPV_const(little, llen);
3614 big_utf8 = DO_UTF8(big);
3615 little_utf8 = DO_UTF8(little);
3616 if (big_utf8 ^ little_utf8) {
3617 /* One needs to be upgraded. */
3618 if (little_utf8 && !PL_encoding) {
3619 /* Well, maybe instead we might be able to downgrade the small
3621 char * const pv = (char*)bytes_from_utf8((U8 *)little_p, &llen,
3624 /* If the large string is ISO-8859-1, and it's not possible to
3625 convert the small string to ISO-8859-1, then there is no
3626 way that it could be found anywhere by index. */
3631 /* At this point, pv is a malloc()ed string. So donate it to temp
3632 to ensure it will get free()d */
3633 little = temp = newSV(0);
3634 sv_usepvn(temp, pv, llen);
3635 little_p = SvPVX(little);
3638 ? newSVpvn(big_p, biglen) : newSVpvn(little_p, llen);
3641 sv_recode_to_utf8(temp, PL_encoding);
3643 sv_utf8_upgrade(temp);
3648 big_p = SvPV_const(big, biglen);
3651 little_p = SvPV_const(little, llen);
3655 if (SvGAMAGIC(big)) {
3656 /* Life just becomes a lot easier if I use a temporary here.
3657 Otherwise I need to avoid calls to sv_pos_u2b(), which (dangerously)
3658 will trigger magic and overloading again, as will fbm_instr()
3660 big = newSVpvn_flags(big_p, biglen,
3661 SVs_TEMP | (big_utf8 ? SVf_UTF8 : 0));
3664 if (SvGAMAGIC(little) || (is_index && !SvOK(little))) {
3665 /* index && SvOK() is a hack. fbm_instr() calls SvPV_const, which will
3666 warn on undef, and we've already triggered a warning with the
3667 SvPV_const some lines above. We can't remove that, as we need to
3668 call some SvPV to trigger overloading early and find out if the
3670 This is all getting to messy. The API isn't quite clean enough,
3671 because data access has side effects.
3673 little = newSVpvn_flags(little_p, llen,
3674 SVs_TEMP | (little_utf8 ? SVf_UTF8 : 0));
3675 little_p = SvPVX(little);
3679 offset = is_index ? 0 : biglen;
3681 if (big_utf8 && offset > 0)
3682 sv_pos_u2b(big, &offset, 0);
3688 else if (offset > (I32)biglen)
3690 if (!(little_p = is_index
3691 ? fbm_instr((unsigned char*)big_p + offset,
3692 (unsigned char*)big_p + biglen, little, 0)
3693 : rninstr(big_p, big_p + offset,
3694 little_p, little_p + llen)))
3697 retval = little_p - big_p;
3698 if (retval > 0 && big_utf8)
3699 sv_pos_b2u(big, &retval);
3703 PUSHi(retval + arybase);
3709 dVAR; dSP; dMARK; dORIGMARK; dTARGET;
3710 SvTAINTED_off(TARG);
3711 do_sprintf(TARG, SP-MARK, MARK+1);
3712 TAINT_IF(SvTAINTED(TARG));
3724 const U8 *s = (U8*)SvPV_const(argsv, len);
3726 if (PL_encoding && SvPOK(argsv) && !DO_UTF8(argsv)) {
3727 SV * const tmpsv = sv_2mortal(newSVsv(argsv));
3728 s = (U8*)sv_recode_to_utf8(tmpsv, PL_encoding);
3732 XPUSHu(DO_UTF8(argsv) ?
3733 utf8n_to_uvchr(s, UTF8_MAXBYTES, 0, UTF8_ALLOW_ANYUV) :
3745 if (((SvIOK_notUV(TOPs) && SvIV(TOPs) < 0)
3747 (SvNOK(TOPs) && SvNV(TOPs) < 0.0))) {
3749 value = POPu; /* chr(-1) eq chr(0xff), etc. */
3751 (void) POPs; /* Ignore the argument value. */
3752 value = UNICODE_REPLACEMENT;
3758 SvUPGRADE(TARG,SVt_PV);
3760 if (value > 255 && !IN_BYTES) {
3761 SvGROW(TARG, (STRLEN)UNISKIP(value)+1);
3762 tmps = (char*)uvchr_to_utf8_flags((U8*)SvPVX(TARG), value, 0);
3763 SvCUR_set(TARG, tmps - SvPVX_const(TARG));
3765 (void)SvPOK_only(TARG);
3774 *tmps++ = (char)value;
3776 (void)SvPOK_only(TARG);
3778 if (PL_encoding && !IN_BYTES) {
3779 sv_recode_to_utf8(TARG, PL_encoding);
3781 if (SvCUR(TARG) == 0 || !is_utf8_string((U8*)tmps, SvCUR(TARG)) ||
3782 UNICODE_IS_REPLACEMENT(utf8_to_uvchr((U8*)tmps, NULL))) {
3786 *tmps++ = (char)value;
3802 const char *tmps = SvPV_const(left, len);
3804 if (DO_UTF8(left)) {
3805 /* If Unicode, try to downgrade.
3806 * If not possible, croak.
3807 * Yes, we made this up. */
3808 SV* const tsv = sv_2mortal(newSVsv(left));
3811 sv_utf8_downgrade(tsv, FALSE);
3812 tmps = SvPV_const(tsv, len);
3814 # ifdef USE_ITHREADS
3816 if (!PL_reentrant_buffer->_crypt_struct_buffer) {
3817 /* This should be threadsafe because in ithreads there is only
3818 * one thread per interpreter. If this would not be true,
3819 * we would need a mutex to protect this malloc. */
3820 PL_reentrant_buffer->_crypt_struct_buffer =
3821 (struct crypt_data *)safemalloc(sizeof(struct crypt_data));
3822 #if defined(__GLIBC__) || defined(__EMX__)
3823 if (PL_reentrant_buffer->_crypt_struct_buffer) {
3824 PL_reentrant_buffer->_crypt_struct_buffer->initialized = 0;
3825 /* work around glibc-2.2.5 bug */
3826 PL_reentrant_buffer->_crypt_struct_buffer->current_saltbits = 0;
3830 # endif /* HAS_CRYPT_R */
3831 # endif /* USE_ITHREADS */
3833 sv_setpv(TARG, fcrypt(tmps, SvPV_nolen_const(right)));
3835 sv_setpv(TARG, PerlProc_crypt(tmps, SvPV_nolen_const(right)));
3841 "The crypt() function is unimplemented due to excessive paranoia.");
3845 /* Generally UTF-8 and UTF-EBCDIC are indistinguishable at this level. So
3846 * most comments below say UTF-8, when in fact they mean UTF-EBCDIC as well */
3848 /* Below are several macros that generate code */
3849 /* Generates code to store a unicode codepoint c that is known to occupy
3850 * exactly two UTF-8 and UTF-EBCDIC bytes; it is stored into p and p+1. */
3851 #define STORE_UNI_TO_UTF8_TWO_BYTE(p, c) \
3853 *(p) = UTF8_TWO_BYTE_HI(c); \
3854 *((p)+1) = UTF8_TWO_BYTE_LO(c); \
3857 /* Like STORE_UNI_TO_UTF8_TWO_BYTE, but advances p to point to the next
3858 * available byte after the two bytes */
3859 #define CAT_UNI_TO_UTF8_TWO_BYTE(p, c) \
3861 *(p)++ = UTF8_TWO_BYTE_HI(c); \
3862 *((p)++) = UTF8_TWO_BYTE_LO(c); \
3865 /* Generates code to store the upper case of latin1 character l which is known
3866 * to have its upper case be non-latin1 into the two bytes p and p+1. There
3867 * are only two characters that fit this description, and this macro knows
3868 * about them, and that the upper case values fit into two UTF-8 or UTF-EBCDIC
3870 #define STORE_NON_LATIN1_UC(p, l) \
3872 if ((l) == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS) { \
3873 STORE_UNI_TO_UTF8_TWO_BYTE((p), LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS); \
3874 } else { /* Must be the following letter */ \
3875 STORE_UNI_TO_UTF8_TWO_BYTE((p), GREEK_CAPITAL_LETTER_MU); \
3879 /* Like STORE_NON_LATIN1_UC, but advances p to point to the next available byte
3880 * after the character stored */
3881 #define CAT_NON_LATIN1_UC(p, l) \
3883 if ((l) == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS) { \
3884 CAT_UNI_TO_UTF8_TWO_BYTE((p), LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS); \
3886 CAT_UNI_TO_UTF8_TWO_BYTE((p), GREEK_CAPITAL_LETTER_MU); \
3890 /* Generates code to add the two UTF-8 bytes (probably u) that are the upper
3891 * case of l into p and p+1. u must be the result of toUPPER_LATIN1_MOD(l),
3892 * and must require two bytes to store it. Advances p to point to the next
3893 * available position */
3894 #define CAT_TWO_BYTE_UNI_UPPER_MOD(p, l, u) \
3896 if ((u) != LATIN_SMALL_LETTER_Y_WITH_DIAERESIS) { \
3897 CAT_UNI_TO_UTF8_TWO_BYTE((p), (u)); /* not special, just save it */ \
3898 } else if (l == LATIN_SMALL_LETTER_SHARP_S) { \
3899 *(p)++ = 'S'; *(p)++ = 'S'; /* upper case is 'SS' */ \
3900 } else {/* else is one of the other two special cases */ \
3901 CAT_NON_LATIN1_UC((p), (l)); \
3907 /* Actually is both lcfirst() and ucfirst(). Only the first character
3908 * changes. This means that possibly we can change in-place, ie., just
3909 * take the source and change that one character and store it back, but not
3910 * if read-only etc, or if the length changes */
3915 STRLEN slen; /* slen is the byte length of the whole SV. */
3918 bool inplace; /* ? Convert first char only, in-place */
3919 bool doing_utf8 = FALSE; /* ? using utf8 */
3920 bool convert_source_to_utf8 = FALSE; /* ? need to convert */
3921 const int op_type = PL_op->op_type;
3924 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
3925 STRLEN ulen; /* ulen is the byte length of the original Unicode character
3926 * stored as UTF-8 at s. */
3927 STRLEN tculen; /* tculen is the byte length of the freshly titlecased (or
3928 * lowercased) character stored in tmpbuf. May be either
3929 * UTF-8 or not, but in either case is the number of bytes */
3933 s = (const U8*)SvPV_nomg_const(source, slen);
3935 if (ckWARN(WARN_UNINITIALIZED))
3936 report_uninit(source);
3941 /* We may be able to get away with changing only the first character, in
3942 * place, but not if read-only, etc. Later we may discover more reasons to
3943 * not convert in-place. */
3944 inplace = SvPADTMP(source) && !SvREADONLY(source) && SvTEMP(source);
3946 /* First calculate what the changed first character should be. This affects
3947 * whether we can just swap it out, leaving the rest of the string unchanged,
3948 * or even if have to convert the dest to UTF-8 when the source isn't */
3950 if (! slen) { /* If empty */
3951 need = 1; /* still need a trailing NUL */
3953 else if (DO_UTF8(source)) { /* Is the source utf8? */
3956 /* TODO: This is #ifdefd out because it has hard-coded the standard mappings,
3957 * and doesn't allow for the user to specify their own. When code is added to
3958 * detect if there is a user-defined mapping in force here, and if so to use
3959 * that, then the code below can be compiled. The detection would be a good
3960 * thing anyway, as currently the user-defined mappings only work on utf8
3961 * strings, and thus depend on the chosen internal storage method, which is a
3963 #ifdef GO_AHEAD_AND_BREAK_USER_DEFINED_CASE_MAPPINGS
3964 if (UTF8_IS_INVARIANT(*s)) {
3966 /* An invariant source character is either ASCII or, in EBCDIC, an
3967 * ASCII equivalent or a caseless C1 control. In both these cases,
3968 * the lower and upper cases of any character are also invariants
3969 * (and title case is the same as upper case). So it is safe to
3970 * use the simple case change macros which avoid the overhead of
3971 * the general functions. Note that if perl were to be extended to
3972 * do locale handling in UTF-8 strings, this wouldn't be true in,
3973 * for example, Lithuanian or Turkic. */
3974 *tmpbuf = (op_type == OP_LCFIRST) ? toLOWER(*s) : toUPPER(*s);
3978 else if (UTF8_IS_DOWNGRADEABLE_START(*s)) {
3981 /* Similarly, if the source character isn't invariant but is in the
3982 * latin1 range (or EBCDIC equivalent thereof), we have the case
3983 * changes compiled into perl, and can avoid the overhead of the
3984 * general functions. In this range, the characters are stored as
3985 * two UTF-8 bytes, and it so happens that any changed-case version
3986 * is also two bytes (in both ASCIIish and EBCDIC machines). */
3990 /* Convert the two source bytes to a single Unicode code point
3991 * value, change case and save for below */
3992 chr = TWO_BYTE_UTF8_TO_UNI(*s, *(s+1));
3993 if (op_type == OP_LCFIRST) { /* lower casing is easy */
3994 U8 lower = toLOWER_LATIN1(chr);
3995 STORE_UNI_TO_UTF8_TWO_BYTE(tmpbuf, lower);
3997 else { /* ucfirst */
3998 U8 upper = toUPPER_LATIN1_MOD(chr);
4000 /* Most of the latin1 range characters are well-behaved. Their
4001 * title and upper cases are the same, and are also in the
4002 * latin1 range. The macro above returns their upper (hence
4003 * title) case, and all that need be done is to save the result
4004 * for below. However, several characters are problematic, and
4005 * have to be handled specially. The MOD in the macro name
4006 * above means that these tricky characters all get mapped to
4007 * the single character LATIN_SMALL_LETTER_Y_WITH_DIAERESIS.
4008 * This mapping saves some tests for the majority of the
4011 if (upper != LATIN_SMALL_LETTER_Y_WITH_DIAERESIS) {
4013 /* Not tricky. Just save it. */
4014 STORE_UNI_TO_UTF8_TWO_BYTE(tmpbuf, upper);
4016 else if (chr == LATIN_SMALL_LETTER_SHARP_S) {
4018 /* This one is tricky because it is two characters long,
4019 * though the UTF-8 is still two bytes, so the stored
4020 * length doesn't change */
4021 *tmpbuf = 'S'; /* The UTF-8 is 'Ss' */
4022 *(tmpbuf + 1) = 's';
4026 /* The other two have their title and upper cases the same,
4027 * but are tricky because the changed-case characters
4028 * aren't in the latin1 range. They, however, do fit into
4029 * two UTF-8 bytes */
4030 STORE_NON_LATIN1_UC(tmpbuf, chr);
4035 #endif /* end of dont want to break user-defined casing */
4037 /* Here, can't short-cut the general case */
4039 utf8_to_uvchr(s, &ulen);
4040 if (op_type == OP_UCFIRST) toTITLE_utf8(s, tmpbuf, &tculen);
4041 else toLOWER_utf8(s, tmpbuf, &tculen);
4043 /* we can't do in-place if the length changes. */
4044 if (ulen != tculen) inplace = FALSE;
4045 need = slen + 1 - ulen + tculen;
4046 #ifdef GO_AHEAD_AND_BREAK_USER_DEFINED_CASE_MAPPINGS
4050 else { /* Non-zero length, non-UTF-8, Need to consider locale and if
4051 * latin1 is treated as caseless. Note that a locale takes
4053 tculen = 1; /* Most characters will require one byte, but this will
4054 * need to be overridden for the tricky ones */
4057 if (op_type == OP_LCFIRST) {
4059 /* lower case the first letter: no trickiness for any character */
4060 *tmpbuf = (IN_LOCALE_RUNTIME) ? toLOWER_LC(*s) :
4061 ((IN_UNI_8_BIT) ? toLOWER_LATIN1(*s) : toLOWER(*s));
4064 else if (IN_LOCALE_RUNTIME) {
4065 *tmpbuf = toUPPER_LC(*s); /* This would be a bug if any locales
4066 * have upper and title case different
4069 else if (! IN_UNI_8_BIT) {
4070 *tmpbuf = toUPPER(*s); /* Returns caseless for non-ascii, or
4071 * on EBCDIC machines whatever the
4072 * native function does */
4074 else { /* is ucfirst non-UTF-8, not in locale, and cased latin1 */
4075 *tmpbuf = toUPPER_LATIN1_MOD(*s);
4077 /* tmpbuf now has the correct title case for all latin1 characters
4078 * except for the several ones that have tricky handling. All
4079 * of these are mapped by the MOD to the letter below. */
4080 if (*tmpbuf == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS) {
4082 /* The length is going to change, with all three of these, so
4083 * can't replace just the first character */
4086 /* We use the original to distinguish between these tricky
4088 if (*s == LATIN_SMALL_LETTER_SHARP_S) {
4089 /* Two character title case 'Ss', but can remain non-UTF-8 */
4092 *(tmpbuf + 1) = 's'; /* Assert: length(tmpbuf) >= 2 */
4097 /* The other two tricky ones have their title case outside
4098 * latin1. It is the same as their upper case. */
4100 STORE_NON_LATIN1_UC(tmpbuf, *s);
4102 /* The UTF-8 and UTF-EBCDIC lengths of both these characters
4103 * and their upper cases is 2. */
4106 /* The entire result will have to be in UTF-8. Assume worst
4107 * case sizing in conversion. (all latin1 characters occupy
4108 * at most two bytes in utf8) */
4109 convert_source_to_utf8 = TRUE;
4110 need = slen * 2 + 1;
4112 } /* End of is one of the three special chars */
4113 } /* End of use Unicode (Latin1) semantics */
4114 } /* End of changing the case of the first character */
4116 /* Here, have the first character's changed case stored in tmpbuf. Ready to
4117 * generate the result */
4120 /* We can convert in place. This means we change just the first
4121 * character without disturbing the rest; no need to grow */
4123 s = d = (U8*)SvPV_force_nomg(source, slen);
4129 /* Here, we can't convert in place; we earlier calculated how much
4130 * space we will need, so grow to accommodate that */
4131 SvUPGRADE(dest, SVt_PV);
4132 d = (U8*)SvGROW(dest, need);
4133 (void)SvPOK_only(dest);
4140 if (! convert_source_to_utf8) {
4142 /* Here both source and dest are in UTF-8, but have to create
4143 * the entire output. We initialize the result to be the
4144 * title/lower cased first character, and then append the rest
4146 sv_setpvn(dest, (char*)tmpbuf, tculen);
4148 sv_catpvn(dest, (char*)(s + ulen), slen - ulen);
4152 const U8 *const send = s + slen;
4154 /* Here the dest needs to be in UTF-8, but the source isn't,
4155 * except we earlier UTF-8'd the first character of the source
4156 * into tmpbuf. First put that into dest, and then append the
4157 * rest of the source, converting it to UTF-8 as we go. */
4159 /* Assert tculen is 2 here because the only two characters that
4160 * get to this part of the code have 2-byte UTF-8 equivalents */
4162 *d++ = *(tmpbuf + 1);
4163 s++; /* We have just processed the 1st char */
4165 for (; s < send; s++) {
4166 d = uvchr_to_utf8(d, *s);
4169 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
4173 else { /* in-place UTF-8. Just overwrite the first character */
4174 Copy(tmpbuf, d, tculen, U8);
4175 SvCUR_set(dest, need - 1);
4178 else { /* Neither source nor dest are in or need to be UTF-8 */
4180 if (IN_LOCALE_RUNTIME) {
4184 if (inplace) { /* in-place, only need to change the 1st char */
4187 else { /* Not in-place */
4189 /* Copy the case-changed character(s) from tmpbuf */
4190 Copy(tmpbuf, d, tculen, U8);
4191 d += tculen - 1; /* Code below expects d to point to final
4192 * character stored */
4195 else { /* empty source */
4196 /* See bug #39028: Don't taint if empty */
4200 /* In a "use bytes" we don't treat the source as UTF-8, but, still want
4201 * the destination to retain that flag */
4205 if (!inplace) { /* Finish the rest of the string, unchanged */
4206 /* This will copy the trailing NUL */
4207 Copy(s + 1, d + 1, slen, U8);
4208 SvCUR_set(dest, need - 1);
4211 if (dest != source && SvTAINTED(source))
4217 /* There's so much setup/teardown code common between uc and lc, I wonder if
4218 it would be worth merging the two, and just having a switch outside each
4219 of the three tight loops. There is less and less commonality though */
4233 if (SvPADTMP(source) && !SvREADONLY(source) && !SvAMAGIC(source)
4234 && SvTEMP(source) && !DO_UTF8(source)
4235 && (IN_LOCALE_RUNTIME || ! IN_UNI_8_BIT)) {
4237 /* We can convert in place. The reason we can't if in UNI_8_BIT is to
4238 * make the loop tight, so we overwrite the source with the dest before
4239 * looking at it, and we need to look at the original source
4240 * afterwards. There would also need to be code added to handle
4241 * switching to not in-place in midstream if we run into characters
4242 * that change the length.
4245 s = d = (U8*)SvPV_force_nomg(source, len);
4252 /* The old implementation would copy source into TARG at this point.
4253 This had the side effect that if source was undef, TARG was now
4254 an undefined SV with PADTMP set, and they don't warn inside
4255 sv_2pv_flags(). However, we're now getting the PV direct from
4256 source, which doesn't have PADTMP set, so it would warn. Hence the
4260 s = (const U8*)SvPV_nomg_const(source, len);
4262 if (ckWARN(WARN_UNINITIALIZED))
4263 report_uninit(source);
4269 SvUPGRADE(dest, SVt_PV);
4270 d = (U8*)SvGROW(dest, min);
4271 (void)SvPOK_only(dest);
4276 /* Overloaded values may have toggled the UTF-8 flag on source, so we need
4277 to check DO_UTF8 again here. */
4279 if (DO_UTF8(source)) {
4280 const U8 *const send = s + len;
4281 U8 tmpbuf[UTF8_MAXBYTES+1];
4283 /* All occurrences of these are to be moved to follow any other marks.
4284 * This is context-dependent. We may not be passed enough context to
4285 * move the iota subscript beyond all of them, but we do the best we can
4286 * with what we're given. The result is always better than if we
4287 * hadn't done this. And, the problem would only arise if we are
4288 * passed a character without all its combining marks, which would be
4289 * the caller's mistake. The information this is based on comes from a
4290 * comment in Unicode SpecialCasing.txt, (and the Standard's text
4291 * itself) and so can't be checked properly to see if it ever gets
4292 * revised. But the likelihood of it changing is remote */
4293 bool in_iota_subscript = FALSE;
4296 if (in_iota_subscript && ! is_utf8_mark(s)) {
4297 /* A non-mark. Time to output the iota subscript */
4298 #define GREEK_CAPITAL_LETTER_IOTA 0x0399
4299 #define COMBINING_GREEK_YPOGEGRAMMENI 0x0345
4301 CAT_UNI_TO_UTF8_TWO_BYTE(d, GREEK_CAPITAL_LETTER_IOTA);
4302 in_iota_subscript = FALSE;
4306 /* See comments at the first instance in this file of this ifdef */
4307 #ifdef GO_AHEAD_AND_BREAK_USER_DEFINED_CASE_MAPPINGS
4309 /* If the UTF-8 character is invariant, then it is in the range
4310 * known by the standard macro; result is only one byte long */
4311 if (UTF8_IS_INVARIANT(*s)) {
4315 else if (UTF8_IS_DOWNGRADEABLE_START(*s)) {
4317 /* Likewise, if it fits in a byte, its case change is in our
4319 U8 orig = TWO_BYTE_UTF8_TO_UNI(*s, *s++);
4320 U8 upper = toUPPER_LATIN1_MOD(orig);
4321 CAT_TWO_BYTE_UNI_UPPER_MOD(d, orig, upper);
4329 /* Otherwise, need the general UTF-8 case. Get the changed
4330 * case value and copy it to the output buffer */
4332 const STRLEN u = UTF8SKIP(s);
4335 const UV uv = toUPPER_utf8(s, tmpbuf, &ulen);
4336 if (uv == GREEK_CAPITAL_LETTER_IOTA
4337 && utf8_to_uvchr(s, 0) == COMBINING_GREEK_YPOGEGRAMMENI)
4339 in_iota_subscript = TRUE;
4342 if (ulen > u && (SvLEN(dest) < (min += ulen - u))) {
4343 /* If the eventually required minimum size outgrows
4344 * the available space, we need to grow. */
4345 const UV o = d - (U8*)SvPVX_const(dest);
4347 /* If someone uppercases one million U+03B0s we
4348 * SvGROW() one million times. Or we could try
4349 * guessing how much to allocate without allocating too
4350 * much. Such is life. See corresponding comment in
4351 * lc code for another option */
4353 d = (U8*)SvPVX(dest) + o;
4355 Copy(tmpbuf, d, ulen, U8);
4361 if (in_iota_subscript) {
4362 CAT_UNI_TO_UTF8_TWO_BYTE(d, GREEK_CAPITAL_LETTER_IOTA);
4366 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
4368 else { /* Not UTF-8 */
4370 const U8 *const send = s + len;
4372 /* Use locale casing if in locale; regular style if not treating
4373 * latin1 as having case; otherwise the latin1 casing. Do the
4374 * whole thing in a tight loop, for speed, */
4375 if (IN_LOCALE_RUNTIME) {
4378 for (; s < send; d++, s++)
4379 *d = toUPPER_LC(*s);
4381 else if (! IN_UNI_8_BIT) {
4382 for (; s < send; d++, s++) {
4387 for (; s < send; d++, s++) {
4388 *d = toUPPER_LATIN1_MOD(*s);
4389 if (*d != LATIN_SMALL_LETTER_Y_WITH_DIAERESIS) continue;
4391 /* The mainstream case is the tight loop above. To avoid
4392 * extra tests in that, all three characters that require
4393 * special handling are mapped by the MOD to the one tested
4395 * Use the source to distinguish between the three cases */
4397 if (*s == LATIN_SMALL_LETTER_SHARP_S) {
4399 /* uc() of this requires 2 characters, but they are
4400 * ASCII. If not enough room, grow the string */
4401 if (SvLEN(dest) < ++min) {
4402 const UV o = d - (U8*)SvPVX_const(dest);
4404 d = (U8*)SvPVX(dest) + o;
4406 *d++ = 'S'; *d = 'S'; /* upper case is 'SS' */
4407 continue; /* Back to the tight loop; still in ASCII */
4410 /* The other two special handling characters have their
4411 * upper cases outside the latin1 range, hence need to be
4412 * in UTF-8, so the whole result needs to be in UTF-8. So,
4413 * here we are somewhere in the middle of processing a
4414 * non-UTF-8 string, and realize that we will have to convert
4415 * the whole thing to UTF-8. What to do? There are
4416 * several possibilities. The simplest to code is to
4417 * convert what we have so far, set a flag, and continue on
4418 * in the loop. The flag would be tested each time through
4419 * the loop, and if set, the next character would be
4420 * converted to UTF-8 and stored. But, I (khw) didn't want
4421 * to slow down the mainstream case at all for this fairly
4422 * rare case, so I didn't want to add a test that didn't
4423 * absolutely have to be there in the loop, besides the
4424 * possibility that it would get too complicated for
4425 * optimizers to deal with. Another possibility is to just
4426 * give up, convert the source to UTF-8, and restart the
4427 * function that way. Another possibility is to convert
4428 * both what has already been processed and what is yet to
4429 * come separately to UTF-8, then jump into the loop that
4430 * handles UTF-8. But the most efficient time-wise of the
4431 * ones I could think of is what follows, and turned out to
4432 * not require much extra code. */
4434 /* Convert what we have so far into UTF-8, telling the
4435 * function that we know it should be converted, and to
4436 * allow extra space for what we haven't processed yet.
4437 * Assume the worst case space requirements for converting
4438 * what we haven't processed so far: that it will require
4439 * two bytes for each remaining source character, plus the
4440 * NUL at the end. This may cause the string pointer to
4441 * move, so re-find it. */
4443 len = d - (U8*)SvPVX_const(dest);
4444 SvCUR_set(dest, len);
4445 len = sv_utf8_upgrade_flags_grow(dest,
4446 SV_GMAGIC|SV_FORCE_UTF8_UPGRADE,
4448 d = (U8*)SvPVX(dest) + len;
4450 /* And append the current character's upper case in UTF-8 */
4451 CAT_NON_LATIN1_UC(d, *s);
4453 /* Now process the remainder of the source, converting to
4454 * upper and UTF-8. If a resulting byte is invariant in
4455 * UTF-8, output it as-is, otherwise convert to UTF-8 and
4456 * append it to the output. */
4459 for (; s < send; s++) {
4460 U8 upper = toUPPER_LATIN1_MOD(*s);
4461 if UTF8_IS_INVARIANT(upper) {
4465 CAT_TWO_BYTE_UNI_UPPER_MOD(d, *s, upper);
4469 /* Here have processed the whole source; no need to continue
4470 * with the outer loop. Each character has been converted
4471 * to upper case and converted to UTF-8 */
4474 } /* End of processing all latin1-style chars */
4475 } /* End of processing all chars */
4476 } /* End of source is not empty */
4478 if (source != dest) {
4479 *d = '\0'; /* Here d points to 1 after last char, add NUL */
4480 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
4482 } /* End of isn't utf8 */
4483 if (dest != source && SvTAINTED(source))
4502 if (SvPADTMP(source) && !SvREADONLY(source) && !SvAMAGIC(source)
4503 && SvTEMP(source) && !DO_UTF8(source)) {
4505 /* We can convert in place, as lowercasing anything in the latin1 range
4506 * (or else DO_UTF8 would have been on) doesn't lengthen it */
4508 s = d = (U8*)SvPV_force_nomg(source, len);
4515 /* The old implementation would copy source into TARG at this point.
4516 This had the side effect that if source was undef, TARG was now
4517 an undefined SV with PADTMP set, and they don't warn inside
4518 sv_2pv_flags(). However, we're now getting the PV direct from
4519 source, which doesn't have PADTMP set, so it would warn. Hence the
4523 s = (const U8*)SvPV_nomg_const(source, len);
4525 if (ckWARN(WARN_UNINITIALIZED))
4526 report_uninit(source);
4532 SvUPGRADE(dest, SVt_PV);
4533 d = (U8*)SvGROW(dest, min);
4534 (void)SvPOK_only(dest);
4539 /* Overloaded values may have toggled the UTF-8 flag on source, so we need
4540 to check DO_UTF8 again here. */
4542 if (DO_UTF8(source)) {
4543 const U8 *const send = s + len;
4544 U8 tmpbuf[UTF8_MAXBYTES_CASE+1];
4547 /* See comments at the first instance in this file of this ifdef */
4548 #ifdef GO_AHEAD_AND_BREAK_USER_DEFINED_CASE_MAPPINGS
4549 if (UTF8_IS_INVARIANT(*s)) {
4551 /* Invariant characters use the standard mappings compiled in.
4556 else if (UTF8_IS_DOWNGRADEABLE_START(*s)) {
4558 /* As do the ones in the Latin1 range */
4559 U8 lower = toLOWER_LATIN1(TWO_BYTE_UTF8_TO_UNI(*s, *s++));
4560 CAT_UNI_TO_UTF8_TWO_BYTE(d, lower);
4565 /* Here, is utf8 not in Latin-1 range, have to go out and get
4566 * the mappings from the tables. */
4568 const STRLEN u = UTF8SKIP(s);
4571 #ifndef CONTEXT_DEPENDENT_CASING
4572 toLOWER_utf8(s, tmpbuf, &ulen);
4574 /* This is ifdefd out because it needs more work and thought. It isn't clear
4575 * that we should do it.
4576 * A minor objection is that this is based on a hard-coded rule from the
4577 * Unicode standard, and may change, but this is not very likely at all.
4578 * mktables should check and warn if it does.
4579 * More importantly, if the sigma occurs at the end of the string, we don't
4580 * have enough context to know whether it is part of a larger string or going
4581 * to be or not. It may be that we are passed a subset of the context, via
4582 * a \U...\E, for example, and we could conceivably know the larger context if
4583 * code were changed to pass that in. But, if the string passed in is an
4584 * intermediate result, and the user concatenates two strings together
4585 * after we have made a final sigma, that would be wrong. If the final sigma
4586 * occurs in the middle of the string we are working on, then we know that it
4587 * should be a final sigma, but otherwise we can't be sure. */
4589 const UV uv = toLOWER_utf8(s, tmpbuf, &ulen);
4591 /* If the lower case is a small sigma, it may be that we need
4592 * to change it to a final sigma. This happens at the end of
4593 * a word that contains more than just this character, and only
4594 * when we started with a capital sigma. */
4595 if (uv == UNICODE_GREEK_SMALL_LETTER_SIGMA &&
4596 s > send - len && /* Makes sure not the first letter */
4597 utf8_to_uvchr(s, 0) == UNICODE_GREEK_CAPITAL_LETTER_SIGMA
4600 /* We use the algorithm in:
4601 * http://www.unicode.org/versions/Unicode5.0.0/ch03.pdf (C
4602 * is a CAPITAL SIGMA): If C is preceded by a sequence
4603 * consisting of a cased letter and a case-ignorable
4604 * sequence, and C is not followed by a sequence consisting
4605 * of a case ignorable sequence and then a cased letter,
4606 * then when lowercasing C, C becomes a final sigma */
4608 /* To determine if this is the end of a word, need to peek
4609 * ahead. Look at the next character */
4610 const U8 *peek = s + u;
4612 /* Skip any case ignorable characters */
4613 while (peek < send && is_utf8_case_ignorable(peek)) {
4614 peek += UTF8SKIP(peek);
4617 /* If we reached the end of the string without finding any
4618 * non-case ignorable characters, or if the next such one
4619 * is not-cased, then we have met the conditions for it
4620 * being a final sigma with regards to peek ahead, and so
4621 * must do peek behind for the remaining conditions. (We
4622 * know there is stuff behind to look at since we tested
4623 * above that this isn't the first letter) */
4624 if (peek >= send || ! is_utf8_cased(peek)) {
4625 peek = utf8_hop(s, -1);
4627 /* Here are at the beginning of the first character
4628 * before the original upper case sigma. Keep backing
4629 * up, skipping any case ignorable characters */
4630 while (is_utf8_case_ignorable(peek)) {
4631 peek = utf8_hop(peek, -1);
4634 /* Here peek points to the first byte of the closest
4635 * non-case-ignorable character before the capital
4636 * sigma. If it is cased, then by the Unicode
4637 * algorithm, we should use a small final sigma instead
4638 * of what we have */
4639 if (is_utf8_cased(peek)) {
4640 STORE_UNI_TO_UTF8_TWO_BYTE(tmpbuf,
4641 UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA);
4645 else { /* Not a context sensitive mapping */
4646 #endif /* End of commented out context sensitive */
4647 if (ulen > u && (SvLEN(dest) < (min += ulen - u))) {
4649 /* If the eventually required minimum size outgrows
4650 * the available space, we need to grow. */
4651 const UV o = d - (U8*)SvPVX_const(dest);
4653 /* If someone lowercases one million U+0130s we
4654 * SvGROW() one million times. Or we could try
4655 * guessing how much to allocate without allocating too
4656 * much. Such is life. Another option would be to
4657 * grow an extra byte or two more each time we need to
4658 * grow, which would cut down the million to 500K, with
4661 d = (U8*)SvPVX(dest) + o;
4663 #ifdef CONTEXT_DEPENDENT_CASING
4666 /* Copy the newly lowercased letter to the output buffer we're
4668 Copy(tmpbuf, d, ulen, U8);
4671 #ifdef GO_AHEAD_AND_BREAK_USER_DEFINED_CASE_MAPPINGS
4674 } /* End of looping through the source string */
4677 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
4678 } else { /* Not utf8 */
4680 const U8 *const send = s + len;
4682 /* Use locale casing if in locale; regular style if not treating
4683 * latin1 as having case; otherwise the latin1 casing. Do the
4684 * whole thing in a tight loop, for speed, */
4685 if (IN_LOCALE_RUNTIME) {
4688 for (; s < send; d++, s++)
4689 *d = toLOWER_LC(*s);
4691 else if (! IN_UNI_8_BIT) {
4692 for (; s < send; d++, s++) {
4697 for (; s < send; d++, s++) {
4698 *d = toLOWER_LATIN1(*s);
4702 if (source != dest) {
4704 SvCUR_set(dest, d - (U8*)SvPVX_const(dest));
4707 if (dest != source && SvTAINTED(source))
4716 SV * const sv = TOPs;
4718 register const char *s = SvPV_const(sv,len);
4720 SvUTF8_off(TARG); /* decontaminate */
4723 SvUPGRADE(TARG, SVt_PV);
4724 SvGROW(TARG, (len * 2) + 1);
4728 if (UTF8_IS_CONTINUED(*s)) {
4729 STRLEN ulen = UTF8SKIP(s);
4753 SvCUR_set(TARG, d - SvPVX_const(TARG));
4754 (void)SvPOK_only_UTF8(TARG);
4757 sv_setpvn(TARG, s, len);
4766 dVAR; dSP; dMARK; dORIGMARK;
4767 register AV *const av = MUTABLE_AV(POPs);
4768 register const I32 lval = (PL_op->op_flags & OPf_MOD || LVRET);
4770 if (SvTYPE(av) == SVt_PVAV) {
4771 const I32 arybase = CopARYBASE_get(PL_curcop);
4772 const bool localizing = PL_op->op_private & OPpLVAL_INTRO;
4773 bool can_preserve = FALSE;
4779 can_preserve = SvCANEXISTDELETE(av);
4782 if (lval && localizing) {
4785 for (svp = MARK + 1; svp <= SP; svp++) {
4786 const I32 elem = SvIV(*svp);
4790 if (max > AvMAX(av))
4794 while (++MARK <= SP) {
4796 I32 elem = SvIV(*MARK);
4797 bool preeminent = TRUE;
4801 if (localizing && can_preserve) {
4802 /* If we can determine whether the element exist,
4803 * Try to preserve the existenceness of a tied array
4804 * element by using EXISTS and DELETE if possible.
4805 * Fallback to FETCH and STORE otherwise. */
4806 preeminent = av_exists(av, elem);
4809 svp = av_fetch(av, elem, lval);
4811 if (!svp || *svp == &PL_sv_undef)
4812 DIE(aTHX_ PL_no_aelem, elem);
4815 save_aelem(av, elem, svp);
4817 SAVEADELETE(av, elem);
4820 *MARK = svp ? *svp : &PL_sv_undef;
4823 if (GIMME != G_ARRAY) {
4825 *++MARK = SP > ORIGMARK ? *SP : &PL_sv_undef;
4831 /* Smart dereferencing for keys, values and each */
4844 /* N.B.: AMG macros return sv if no overloading is found */
4845 SV *maybe_hv = AMG_CALLunary(sv, to_hv_amg);
4846 SV *maybe_av = AMG_CALLunary(sv, to_av_amg);
4847 if ( maybe_hv != sv && maybe_av != sv ) {
4848 Perl_ck_warner(aTHX_ packWARN(WARN_AMBIGUOUS), "%s",
4849 Perl_form(aTHX_ "Ambiguous overloaded argument to %s resolved as %%{}",
4850 PL_op_desc[PL_op->op_type]
4855 else if ( maybe_av != sv ) {
4856 if ( SvTYPE(SvRV(sv)) == SVt_PVHV ) {
4857 /* @{} overload, but underlying reftype is HV */
4858 Perl_ck_warner(aTHX_ packWARN(WARN_AMBIGUOUS), "%s",
4859 Perl_form(aTHX_ "Ambiguous overloaded argument to %s resolved as @{}",
4860 PL_op_desc[PL_op->op_type]
4866 else if ( maybe_hv != sv ) {
4867 if ( SvTYPE(SvRV(sv)) == SVt_PVAV ) {
4868 /* %{} overload, but underlying reftype is AV */
4869 Perl_ck_warner(aTHX_ packWARN(WARN_AMBIGUOUS), "%s",
4870 Perl_form(aTHX_ "Ambiguous overloaded argument to %s resolved as %%{}",
4871 PL_op_desc[PL_op->op_type]
4881 if ( SvTYPE(sv) != SVt_PVHV && SvTYPE(sv) != SVt_PVAV ) {
4882 DIE(aTHX_ "Type of argument to %s must be hashref or arrayref",
4883 PL_op_desc[PL_op->op_type] );
4886 /* Delegate to correct function for op type */
4888 if (PL_op->op_type == OP_RKEYS || PL_op->op_type == OP_RVALUES) {
4889 return (SvTYPE(sv) == SVt_PVHV) ? Perl_do_kv(aTHX) : Perl_pp_akeys(aTHX);
4892 return (SvTYPE(sv) == SVt_PVHV) ? Perl_pp_each(aTHX) : Perl_pp_aeach(aTHX);
4900 AV *array = MUTABLE_AV(POPs);
4901 const I32 gimme = GIMME_V;
4902 IV *iterp = Perl_av_iter_p(aTHX_ array);
4903 const IV current = (*iterp)++;
4905 if (current > av_len(array)) {
4907 if (gimme == G_SCALAR)
4914 mPUSHi(CopARYBASE_get(PL_curcop) + current);
4915 if (gimme == G_ARRAY) {
4916 SV **const element = av_fetch(array, current, 0);
4917 PUSHs(element ? *element : &PL_sv_undef);
4926 AV *array = MUTABLE_AV(POPs);
4927 const I32 gimme = GIMME_V;
4929 *Perl_av_iter_p(aTHX_ array) = 0;
4931 if (gimme == G_SCALAR) {
4933 PUSHi(av_len(array) + 1);
4935 else if (gimme == G_ARRAY) {
4936 IV n = Perl_av_len(aTHX_ array);
4937 IV i = CopARYBASE_get(PL_curcop);
4941 if (PL_op->op_type == OP_AKEYS || PL_op->op_type == OP_RKEYS) {
4943 for (; i <= n; i++) {
4948 for (i = 0; i <= n; i++) {
4949 SV *const *const elem = Perl_av_fetch(aTHX_ array, i, 0);
4950 PUSHs(elem ? *elem : &PL_sv_undef);
4957 /* Associative arrays. */
4963 HV * hash = MUTABLE_HV(POPs);
4965 const I32 gimme = GIMME_V;
4968 /* might clobber stack_sp */
4969 entry = hv_iternext(hash);
4974 SV* const sv = hv_iterkeysv(entry);
4975 PUSHs(sv); /* won't clobber stack_sp */
4976 if (gimme == G_ARRAY) {
4979 /* might clobber stack_sp */
4980 val = hv_iterval(hash, entry);
4985 else if (gimme == G_SCALAR)
4992 S_do_delete_local(pTHX)
4996 const I32 gimme = GIMME_V;
5000 if (PL_op->op_private & OPpSLICE) {
5002 SV * const osv = POPs;
5003 const bool tied = SvRMAGICAL(osv)
5004 && mg_find((const SV *)osv, PERL_MAGIC_tied);
5005 const bool can_preserve = SvCANEXISTDELETE(osv)
5006 || mg_find((const SV *)osv, PERL_MAGIC_env);
5007 const U32 type = SvTYPE(osv);
5008 if (type == SVt_PVHV) { /* hash element */
5009 HV * const hv = MUTABLE_HV(osv);
5010 while (++MARK <= SP) {
5011 SV * const keysv = *MARK;
5013 bool preeminent = TRUE;
5015 preeminent = hv_exists_ent(hv, keysv, 0);
5017 HE *he = hv_fetch_ent(hv, keysv, 1, 0);
5024 sv = hv_delete_ent(hv, keysv, 0, 0);
5025 SvREFCNT_inc_simple_void(sv); /* De-mortalize */
5028 save_helem_flags(hv, keysv, &sv, SAVEf_KEEPOLDELEM);
5030 *MARK = sv_mortalcopy(sv);
5036 SAVEHDELETE(hv, keysv);
5037 *MARK = &PL_sv_undef;
5041 else if (type == SVt_PVAV) { /* array element */
5042 if (PL_op->op_flags & OPf_SPECIAL) {
5043 AV * const av = MUTABLE_AV(osv);
5044 while (++MARK <= SP) {
5045 I32 idx = SvIV(*MARK);
5047 bool preeminent = TRUE;
5049 preeminent = av_exists(av, idx);
5051 SV **svp = av_fetch(av, idx, 1);
5058 sv = av_delete(av, idx, 0);
5059 SvREFCNT_inc_simple_void(sv); /* De-mortalize */
5062 save_aelem_flags(av, idx, &sv, SAVEf_KEEPOLDELEM);
5064 *MARK = sv_mortalcopy(sv);
5070 SAVEADELETE(av, idx);
5071 *MARK = &PL_sv_undef;
5077 DIE(aTHX_ "Not a HASH reference");
5078 if (gimme == G_VOID)
5080 else if (gimme == G_SCALAR) {
5085 *++MARK = &PL_sv_undef;
5090 SV * const keysv = POPs;
5091 SV * const osv = POPs;
5092 const bool tied = SvRMAGICAL(osv)
5093 && mg_find((const SV *)osv, PERL_MAGIC_tied);
5094 const bool can_preserve = SvCANEXISTDELETE(osv)
5095 || mg_find((const SV *)osv, PERL_MAGIC_env);
5096 const U32 type = SvTYPE(osv);
5098 if (type == SVt_PVHV) {
5099 HV * const hv = MUTABLE_HV(osv);
5100 bool preeminent = TRUE;
5102 preeminent = hv_exists_ent(hv, keysv, 0);
5104 HE *he = hv_fetch_ent(hv, keysv, 1, 0);
5111 sv = hv_delete_ent(hv, keysv, 0, 0);
5112 SvREFCNT_inc_simple_void(sv); /* De-mortalize */
5115 save_helem_flags(hv, keysv, &sv, SAVEf_KEEPOLDELEM);
5117 SV *nsv = sv_mortalcopy(sv);
5123 SAVEHDELETE(hv, keysv);
5125 else if (type == SVt_PVAV) {
5126 if (PL_op->op_flags & OPf_SPECIAL) {
5127 AV * const av = MUTABLE_AV(osv);
5128 I32 idx = SvIV(keysv);
5129 bool preeminent = TRUE;
5131 preeminent = av_exists(av, idx);
5133 SV **svp = av_fetch(av, idx, 1);
5140 sv = av_delete(av, idx, 0);
5141 SvREFCNT_inc_simple_void(sv); /* De-mortalize */
5144 save_aelem_flags(av, idx, &sv, SAVEf_KEEPOLDELEM);
5146 SV *nsv = sv_mortalcopy(sv);
5152 SAVEADELETE(av, idx);
5155 DIE(aTHX_ "panic: avhv_delete no longer supported");
5158 DIE(aTHX_ "Not a HASH reference");
5161 if (gimme != G_VOID)
5175 if (PL_op->op_private & OPpLVAL_INTRO)
5176 return do_delete_local();
5179 discard = (gimme == G_VOID) ? G_DISCARD : 0;
5181 if (PL_op->op_private & OPpSLICE) {
5183 HV * const hv = MUTABLE_HV(POPs);
5184 const U32 hvtype = SvTYPE(hv);
5185 if (hvtype == SVt_PVHV) { /* hash element */
5186 while (++MARK <= SP) {
5187 SV * const sv = hv_delete_ent(hv, *MARK, discard, 0);
5188 *MARK = sv ? sv : &PL_sv_undef;
5191 else if (hvtype == SVt_PVAV) { /* array element */
5192 if (PL_op->op_flags & OPf_SPECIAL) {
5193 while (++MARK <= SP) {
5194 SV * const sv = av_delete(MUTABLE_AV(hv), SvIV(*MARK), discard);
5195 *MARK = sv ? sv : &PL_sv_undef;
5200 DIE(aTHX_ "Not a HASH reference");
5203 else if (gimme == G_SCALAR) {
5208 *++MARK = &PL_sv_undef;
5214 HV * const hv = MUTABLE_HV(POPs);
5216 if (SvTYPE(hv) == SVt_PVHV)
5217 sv = hv_delete_ent(hv, keysv, discard, 0);
5218 else if (SvTYPE(hv) == SVt_PVAV) {
5219 if (PL_op->op_flags & OPf_SPECIAL)
5220 sv = av_delete(MUTABLE_AV(hv), SvIV(keysv), discard);
5222 DIE(aTHX_ "panic: avhv_delete no longer supported");
5225 DIE(aTHX_ "Not a HASH reference");
5241 if (PL_op->op_private & OPpEXISTS_SUB) {
5243 SV * const sv = POPs;
5244 CV * const cv = sv_2cv(sv, &hv, &gv, 0);
5247 if (gv && isGV(gv) && GvCV(gv) && !GvCVGEN(gv))
5252 hv = MUTABLE_HV(POPs);
5253 if (SvTYPE(hv) == SVt_PVHV) {
5254 if (hv_exists_ent(hv, tmpsv, 0))
5257 else if (SvTYPE(hv) == SVt_PVAV) {
5258 if (PL_op->op_flags & OPf_SPECIAL) { /* array element */
5259 if (av_exists(MUTABLE_AV(hv), SvIV(tmpsv)))
5264 DIE(aTHX_ "Not a HASH reference");
5271 dVAR; dSP; dMARK; dORIGMARK;
5272 register HV * const hv = MUTABLE_HV(POPs);
5273 register const I32 lval = (PL_op->op_flags & OPf_MOD || LVRET);
5274 const bool localizing = PL_op->op_private & OPpLVAL_INTRO;
5275 bool can_preserve = FALSE;
5281 if (SvCANEXISTDELETE(hv) || mg_find((const SV *)hv, PERL_MAGIC_env))
5282 can_preserve = TRUE;
5285 while (++MARK <= SP) {
5286 SV * const keysv = *MARK;
5289 bool preeminent = TRUE;
5291 if (localizing && can_preserve) {
5292 /* If we can determine whether the element exist,
5293 * try to preserve the existenceness of a tied hash
5294 * element by using EXISTS and DELETE if possible.
5295 * Fallback to FETCH and STORE otherwise. */
5296 preeminent = hv_exists_ent(hv, keysv, 0);
5299 he = hv_fetch_ent(hv, keysv, lval, 0);
5300 svp = he ? &HeVAL(he) : NULL;
5303 if (!svp || *svp == &PL_sv_undef) {
5304 DIE(aTHX_ PL_no_helem_sv, SVfARG(keysv));
5307 if (HvNAME_get(hv) && isGV(*svp))
5308 save_gp(MUTABLE_GV(*svp), !(PL_op->op_flags & OPf_SPECIAL));
5309 else if (preeminent)
5310 save_helem_flags(hv, keysv, svp,
5311 (PL_op->op_flags & OPf_SPECIAL) ? 0 : SAVEf_SETMAGIC);
5313 SAVEHDELETE(hv, keysv);
5316 *MARK = svp ? *svp : &PL_sv_undef;
5318 if (GIMME != G_ARRAY) {
5320 *++MARK = SP > ORIGMARK ? *SP : &PL_sv_undef;
5326 /* List operators. */
5331 if (GIMME != G_ARRAY) {
5333 *MARK = *SP; /* unwanted list, return last item */
5335 *MARK = &PL_sv_undef;
5345 SV ** const lastrelem = PL_stack_sp;
5346 SV ** const lastlelem = PL_stack_base + POPMARK;
5347 SV ** const firstlelem = PL_stack_base + POPMARK + 1;
5348 register SV ** const firstrelem = lastlelem + 1;
5349 const I32 arybase = CopARYBASE_get(PL_curcop);
5350 I32 is_something_there = FALSE;
5352 register const I32 max = lastrelem - lastlelem;
5353 register SV **lelem;
5355 if (GIMME != G_ARRAY) {
5356 I32 ix = SvIV(*lastlelem);
5361 if (ix < 0 || ix >= max)
5362 *firstlelem = &PL_sv_undef;
5364 *firstlelem = firstrelem[ix];
5370 SP = firstlelem - 1;
5374 for (lelem = firstlelem; lelem <= lastlelem; lelem++) {
5375 I32 ix = SvIV(*lelem);
5380 if (ix < 0 || ix >= max)
5381 *lelem = &PL_sv_undef;
5383 is_something_there = TRUE;
5384 if (!(*lelem = firstrelem[ix]))
5385 *lelem = &PL_sv_undef;
5388 if (is_something_there)
5391 SP = firstlelem - 1;
5397 dVAR; dSP; dMARK; dORIGMARK;
5398 const I32 items = SP - MARK;
5399 SV * const av = MUTABLE_SV(av_make(items, MARK+1));
5400 SP = ORIGMARK; /* av_make() might realloc stack_sp */
5401 mXPUSHs((PL_op->op_flags & OPf_SPECIAL)
5402 ? newRV_noinc(av) : av);
5408 dVAR; dSP; dMARK; dORIGMARK;
5409 HV* const hv = newHV();
5412 SV * const key = *++MARK;
5413 SV * const val = newSV(0);
5415 sv_setsv(val, *++MARK);
5417 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "Odd number of elements in anonymous hash");
5418 (void)hv_store_ent(hv,key,val,0);
5421 mXPUSHs((PL_op->op_flags & OPf_SPECIAL)
5422 ? newRV_noinc(MUTABLE_SV(hv)) : MUTABLE_SV(hv));
5428 dVAR; dSP; dMARK; dORIGMARK;
5429 register AV *ary = MUTABLE_AV(*++MARK);
5433 register I32 offset;
5434 register I32 length;
5438 const MAGIC * const mg = SvTIED_mg((const SV *)ary, PERL_MAGIC_tied);
5441 return Perl_tied_method(aTHX_ "SPLICE", mark - 1, MUTABLE_SV(ary), mg,
5442 GIMME_V | TIED_METHOD_ARGUMENTS_ON_STACK,
5449 offset = i = SvIV(*MARK);
5451 offset += AvFILLp(ary) + 1;
5453 offset -= CopARYBASE_get(PL_curcop);
5455 DIE(aTHX_ PL_no_aelem, i);
5457 length = SvIVx(*MARK++);
5459 length += AvFILLp(ary) - offset + 1;
5465 length = AvMAX(ary) + 1; /* close enough to infinity */
5469 length = AvMAX(ary) + 1;
5471 if (offset > AvFILLp(ary) + 1) {
5472 Perl_ck_warner(aTHX_ packWARN(WARN_MISC), "splice() offset past end of array" );
5473 offset = AvFILLp(ary) + 1;
5475 after = AvFILLp(ary) + 1 - (offset + length);
5476 if (after < 0) { /* not that much array */
5477 length += after; /* offset+length now in array */
5483 /* At this point, MARK .. SP-1 is our new LIST */
5486 diff = newlen - length;
5487 if (newlen && !AvREAL(ary) && AvREIFY(ary))
5490 /* make new elements SVs now: avoid problems if they're from the array */
5491 for (dst = MARK, i = newlen; i; i--) {
5492 SV * const h = *dst;
5493 *dst++ = newSVsv(h);
5496 if (diff < 0) { /* shrinking the area */
5497 SV **tmparyval = NULL;
5499 Newx(tmparyval, newlen, SV*); /* so remember insertion */
5500 Copy(MARK, tmparyval, newlen, SV*);
5503 MARK = ORIGMARK + 1;
5504 if (GIMME == G_ARRAY) { /* copy return vals to stack */
5505 MEXTEND(MARK, length);
5506 Copy(AvARRAY(ary)+offset, MARK, length, SV*);
5508 EXTEND_MORTAL(length);
5509 for (i = length, dst = MARK; i; i--) {
5510 sv_2mortal(*dst); /* free them eventually */
5517 *MARK = AvARRAY(ary)[offset+length-1];
5520 for (i = length - 1, dst = &AvARRAY(ary)[offset]; i > 0; i--)
5521 SvREFCNT_dec(*dst++); /* free them now */
5524 AvFILLp(ary) += diff;
5526 /* pull up or down? */
5528 if (offset < after) { /* easier to pull up */
5529 if (offset) { /* esp. if nothing to pull */
5530 src = &AvARRAY(ary)[offset-1];
5531 dst = src - diff; /* diff is negative */
5532 for (i = offset; i > 0; i--) /* can't trust Copy */
5536 AvARRAY(ary) = AvARRAY(ary) - diff; /* diff is negative */
5540 if (after) { /* anything to pull down? */
5541 src = AvARRAY(ary) + offset + length;
5542 dst = src + diff; /* diff is negative */
5543 Move(src, dst, after, SV*);
5545 dst = &AvARRAY(ary)[AvFILLp(ary)+1];
5546 /* avoid later double free */
5550 dst[--i] = &PL_sv_undef;
5553 Copy( tmparyval, AvARRAY(ary) + offset, newlen, SV* );
5554 Safefree(tmparyval);
5557 else { /* no, expanding (or same) */
5558 SV** tmparyval = NULL;
5560 Newx(tmparyval, length, SV*); /* so remember deletion */
5561 Copy(AvARRAY(ary)+offset, tmparyval, length, SV*);
5564 if (diff > 0) { /* expanding */
5565 /* push up or down? */
5566 if (offset < after && diff <= AvARRAY(ary) - AvALLOC(ary)) {
5570 Move(src, dst, offset, SV*);
5572 AvARRAY(ary) = AvARRAY(ary) - diff;/* diff is positive */
5574 AvFILLp(ary) += diff;
5577 if (AvFILLp(ary) + diff >= AvMAX(ary)) /* oh, well */
5578 av_extend(ary, AvFILLp(ary) + diff);
5579 AvFILLp(ary) += diff;
5582 dst = AvARRAY(ary) + AvFILLp(ary);
5584 for (i = after; i; i--) {
5592 Copy( MARK, AvARRAY(ary) + offset, newlen, SV* );
5595 MARK = ORIGMARK + 1;
5596 if (GIMME == G_ARRAY) { /* copy return vals to stack */
5598 Copy(tmparyval, MARK, length, SV*);
5600 EXTEND_MORTAL(length);
5601 for (i = length, dst = MARK; i; i--) {
5602 sv_2mortal(*dst); /* free them eventually */
5609 else if (length--) {
5610 *MARK = tmparyval[length];
5613 while (length-- > 0)
5614 SvREFCNT_dec(tmparyval[length]);
5618 *MARK = &PL_sv_undef;
5619 Safefree(tmparyval);
5623 mg_set(MUTABLE_SV(ary));
5631 dVAR; dSP; dMARK; dORIGMARK; dTARGET;
5632 register AV * const ary = MUTABLE_AV(*++MARK);
5633 const MAGIC * const mg = SvTIED_mg((const SV *)ary, PERL_MAGIC_tied);
5636 *MARK-- = SvTIED_obj(MUTABLE_SV(ary), mg);
5639 ENTER_with_name("call_PUSH");
5640 call_method("PUSH",G_SCALAR|G_DISCARD);
5641 LEAVE_with_name("call_PUSH");
5645 PL_delaymagic = DM_DELAY;
5646 for (++MARK; MARK <= SP; MARK++) {
5647 SV * const sv = newSV(0);
5649 sv_setsv(sv, *MARK);
5650 av_store(ary, AvFILLp(ary)+1, sv);
5652 if (PL_delaymagic & DM_ARRAY_ISA)
5653 mg_set(MUTABLE_SV(ary));
5658 if (OP_GIMME(PL_op, 0) != G_VOID) {
5659 PUSHi( AvFILL(ary) + 1 );
5668 AV * const av = PL_op->op_flags & OPf_SPECIAL
5669 ? MUTABLE_AV(GvAV(PL_defgv)) : MUTABLE_AV(POPs);
5670 SV * const sv = PL_op->op_type == OP_SHIFT ? av_shift(av) : av_pop(av);
5674 (void)sv_2mortal(sv);
5681 dVAR; dSP; dMARK; dORIGMARK; dTARGET;
5682 register AV *ary = MUTABLE_AV(*++MARK);
5683 const MAGIC * const mg = SvTIED_mg((const SV *)ary, PERL_MAGIC_tied);
5686 *MARK-- = SvTIED_obj(MUTABLE_SV(ary), mg);
5689 ENTER_with_name("call_UNSHIFT");
5690 call_method("UNSHIFT",G_SCALAR|G_DISCARD);
5691 LEAVE_with_name("call_UNSHIFT");
5696 av_unshift(ary, SP - MARK);
5698 SV * const sv = newSVsv(*++MARK);
5699 (void)av_store(ary, i++, sv);
5703 if (OP_GIMME(PL_op, 0) != G_VOID) {
5704 PUSHi( AvFILL(ary) + 1 );
5713 if (GIMME == G_ARRAY) {
5714 if (PL_op->op_private & OPpREVERSE_INPLACE) {
5718 assert( MARK+1 == SP && *SP && SvTYPE(*SP) == SVt_PVAV);
5719 (void)POPMARK; /* remove mark associated with ex-OP_AASSIGN */
5720 av = MUTABLE_AV((*SP));
5721 /* In-place reversing only happens in void context for the array
5722 * assignment. We don't need to push anything on the stack. */
5725 if (SvMAGICAL(av)) {
5727 register SV *tmp = sv_newmortal();
5728 /* For SvCANEXISTDELETE */
5731 bool can_preserve = SvCANEXISTDELETE(av);
5733 for (i = 0, j = av_len(av); i < j; ++i, --j) {
5734 register SV *begin, *end;
5737 if (!av_exists(av, i)) {
5738 if (av_exists(av, j)) {
5739 register SV *sv = av_delete(av, j, 0);
5740 begin = *av_fetch(av, i, TRUE);
5741 sv_setsv_mg(begin, sv);
5745 else if (!av_exists(av, j)) {
5746 register SV *sv = av_delete(av, i, 0);
5747 end = *av_fetch(av, j, TRUE);
5748 sv_setsv_mg(end, sv);
5753 begin = *av_fetch(av, i, TRUE);
5754 end = *av_fetch(av, j, TRUE);
5755 sv_setsv(tmp, begin);
5756 sv_setsv_mg(begin, end);
5757 sv_setsv_mg(end, tmp);
5761 SV **begin = AvARRAY(av);
5764 SV **end = begin + AvFILLp(av);
5766 while (begin < end) {
5767 register SV * const tmp = *begin;
5778 register SV * const tmp = *MARK;
5782 /* safe as long as stack cannot get extended in the above */
5788 register char *down;
5793 SvUTF8_off(TARG); /* decontaminate */
5795 do_join(TARG, &PL_sv_no, MARK, SP);
5797 sv_setsv(TARG, SP > MARK ? *SP : find_rundefsv());
5798 if (! SvOK(TARG) && ckWARN(WARN_UNINITIALIZED))
5799 report_uninit(TARG);
5802 up = SvPV_force(TARG, len);
5804 if (DO_UTF8(TARG)) { /* first reverse each character */
5805 U8* s = (U8*)SvPVX(TARG);
5806 const U8* send = (U8*)(s + len);
5808 if (UTF8_IS_INVARIANT(*s)) {
5813 if (!utf8_to_uvchr(s, 0))
5817 down = (char*)(s - 1);
5818 /* reverse this character */
5822 *down-- = (char)tmp;
5828 down = SvPVX(TARG) + len - 1;
5832 *down-- = (char)tmp;
5834 (void)SvPOK_only_UTF8(TARG);
5846 register IV limit = POPi; /* note, negative is forever */
5847 SV * const sv = POPs;
5849 register const char *s = SvPV_const(sv, len);
5850 const bool do_utf8 = DO_UTF8(sv);
5851 const char *strend = s + len;
5853 register REGEXP *rx;
5855 register const char *m;
5857 const STRLEN slen = do_utf8 ? utf8_length((U8*)s, (U8*)strend) : (STRLEN)(strend - s);
5858 I32 maxiters = slen + 10;
5859 I32 trailing_empty = 0;
5861 const I32 origlimit = limit;
5864 const I32 gimme = GIMME_V;
5866 const I32 oldsave = PL_savestack_ix;
5867 U32 make_mortal = SVs_TEMP;
5872 Copy(&LvTARGOFF(POPs), &pm, 1, PMOP*);
5877 DIE(aTHX_ "panic: pp_split");
5880 TAINT_IF(get_regex_charset(RX_EXTFLAGS(rx)) == REGEX_LOCALE_CHARSET &&
5881 (RX_EXTFLAGS(rx) & (RXf_WHITE | RXf_SKIPWHITE)));
5883 RX_MATCH_UTF8_set(rx, do_utf8);
5886 if (pm->op_pmreplrootu.op_pmtargetoff) {
5887 ary = GvAVn(MUTABLE_GV(PAD_SVl(pm->op_pmreplrootu.op_pmtargetoff)));
5890 if (pm->op_pmreplrootu.op_pmtargetgv) {
5891 ary = GvAVn(pm->op_pmreplrootu.op_pmtargetgv);
5896 if (ary && (gimme != G_ARRAY || (pm->op_pmflags & PMf_ONCE))) {
5902 if ((mg = SvTIED_mg((const SV *)ary, PERL_MAGIC_tied))) {
5904 XPUSHs(SvTIED_obj(MUTABLE_SV(ary), mg));
5911 for (i = AvFILLp(ary); i >= 0; i--)
5912 AvARRAY(ary)[i] = &PL_sv_undef; /* don't free mere refs */
5914 /* temporarily switch stacks */
5915 SAVESWITCHSTACK(PL_curstack, ary);
5919 base = SP - PL_stack_base;
5921 if (RX_EXTFLAGS(rx) & RXf_SKIPWHITE) {
5923 while (*s == ' ' || is_utf8_space((U8*)s))
5926 else if (get_regex_charset(RX_EXTFLAGS(rx)) == REGEX_LOCALE_CHARSET) {
5927 while (isSPACE_LC(*s))
5935 if (RX_EXTFLAGS(rx) & RXf_PMf_MULTILINE) {
5939 gimme_scalar = gimme == G_SCALAR && !ary;
5942 limit = maxiters + 2;
5943 if (RX_EXTFLAGS(rx) & RXf_WHITE) {
5946 /* this one uses 'm' and is a negative test */
5948 while (m < strend && !( *m == ' ' || is_utf8_space((U8*)m) )) {
5949 const int t = UTF8SKIP(m);
5950 /* is_utf8_space returns FALSE for malform utf8 */
5957 else if (get_regex_charset(RX_EXTFLAGS(rx)) == REGEX_LOCALE_CHARSET) {
5958 while (m < strend && !isSPACE_LC(*m))
5961 while (m < strend && !isSPACE(*m))
5974 dstr = newSVpvn_flags(s, m-s,
5975 (do_utf8 ? SVf_UTF8 : 0) | make_mortal);
5979 /* skip the whitespace found last */
5981 s = m + UTF8SKIP(m);
5985 /* this one uses 's' and is a positive test */
5987 while (s < strend && ( *s == ' ' || is_utf8_space((U8*)s) ))
5990 else if (get_regex_charset(RX_EXTFLAGS(rx)) == REGEX_LOCALE_CHARSET) {
5991 while (s < strend && isSPACE_LC(*s))
5994 while (s < strend && isSPACE(*s))
5999 else if (RX_EXTFLAGS(rx) & RXf_START_ONLY) {
6001 for (m = s; m < strend && *m != '\n'; m++)
6014 dstr = newSVpvn_flags(s, m-s,
6015 (do_utf8 ? SVf_UTF8 : 0) | make_mortal);
6021 else if (RX_EXTFLAGS(rx) & RXf_NULL && !(s >= strend)) {
6023 Pre-extend the stack, either the number of bytes or
6024 characters in the string or a limited amount, triggered by:
6026 my ($x, $y) = split //, $str;
6030 if (!gimme_scalar) {
6031 const U32 items = limit - 1;
6040 /* keep track of how many bytes we skip over */
6050 dstr = newSVpvn_flags(m, s-m, SVf_UTF8 | make_mortal);
6063 dstr = newSVpvn(s, 1);
6079 else if (do_utf8 == (RX_UTF8(rx) != 0) &&
6080 (RX_EXTFLAGS(rx) & RXf_USE_INTUIT) && !RX_NPARENS(rx)
6081 && (RX_EXTFLAGS(rx) & RXf_CHECK_ALL)
6082 && !(RX_EXTFLAGS(rx) & RXf_ANCH)) {
6083 const int tail = (RX_EXTFLAGS(rx) & RXf_INTUIT_TAIL);
6084 SV * const csv = CALLREG_INTUIT_STRING(rx);
6086 len = RX_MINLENRET(rx);
6087 if (len == 1 && !RX_UTF8(rx) && !tail) {
6088 const char c = *SvPV_nolen_const(csv);
6090 for (m = s; m < strend && *m != c; m++)
6101 dstr = newSVpvn_flags(s, m-s,
6102 (do_utf8 ? SVf_UTF8 : 0) | make_mortal);
6105 /* The rx->minlen is in characters but we want to step
6106 * s ahead by bytes. */
6108 s = (char*)utf8_hop((U8*)m, len);
6110 s = m + len; /* Fake \n at the end */
6114 while (s < strend && --limit &&
6115 (m = fbm_instr((unsigned char*)s, (unsigned char*)strend,
6116 csv, multiline ? FBMrf_MULTILINE : 0)) )
6125 dstr = newSVpvn_flags(s, m-s,
6126 (do_utf8 ? SVf_UTF8 : 0) | make_mortal);
6129 /* The rx->minlen is in characters but we want to step
6130 * s ahead by bytes. */
6132 s = (char*)utf8_hop((U8*)m, len);
6134 s = m + len; /* Fake \n at the end */
6139 maxiters += slen * RX_NPARENS(rx);
6140 while (s < strend && --limit)
6144 rex_return = CALLREGEXEC(rx, (char*)s, (char*)strend, (char*)orig, 1 ,
6147 if (rex_return == 0)
6149 TAINT_IF(RX_MATCH_TAINTED(rx));
6150 if (RX_MATCH_COPIED(rx) && RX_SUBBEG(rx) != orig) {
6153 orig = RX_SUBBEG(rx);
6155 strend = s + (strend - m);
6157 m = RX_OFFS(rx)[0].start + orig;
6166 dstr = newSVpvn_flags(s, m-s,
6167 (do_utf8 ? SVf_UTF8 : 0) | make_mortal);
6170 if (RX_NPARENS(rx)) {
6172 for (i = 1; i <= (I32)RX_NPARENS(rx); i++) {
6173 s = RX_OFFS(rx)[i].start + orig;
6174 m = RX_OFFS(rx)[i].end + orig;
6176 /* japhy (07/27/01) -- the (m && s) test doesn't catch
6177 parens that didn't match -- they should be set to
6178 undef, not the empty string */
6186 if (m >= orig && s >= orig) {
6187 dstr = newSVpvn_flags(s, m-s,
6188 (do_utf8 ? SVf_UTF8 : 0)
6192 dstr = &PL_sv_undef; /* undef, not "" */
6198 s = RX_OFFS(rx)[0].end + orig;
6202 if (!gimme_scalar) {
6203 iters = (SP - PL_stack_base) - base;
6205 if (iters > maxiters)
6206 DIE(aTHX_ "Split loop");
6208 /* keep field after final delim? */
6209 if (s < strend || (iters && origlimit)) {
6210 if (!gimme_scalar) {
6211 const STRLEN l = strend - s;
6212 dstr = newSVpvn_flags(s, l, (do_utf8 ? SVf_UTF8 : 0) | make_mortal);
6217 else if (!origlimit) {
6219 iters -= trailing_empty;
6221 while (iters > 0 && (!TOPs || !SvANY(TOPs) || SvCUR(TOPs) == 0)) {
6222 if (TOPs && !make_mortal)
6224 *SP-- = &PL_sv_undef;
6231 LEAVE_SCOPE(oldsave); /* may undo an earlier SWITCHSTACK */
6235 if (SvSMAGICAL(ary)) {
6237 mg_set(MUTABLE_SV(ary));
6240 if (gimme == G_ARRAY) {
6242 Copy(AvARRAY(ary), SP + 1, iters, SV*);
6249 ENTER_with_name("call_PUSH");
6250 call_method("PUSH",G_SCALAR|G_DISCARD);
6251 LEAVE_with_name("call_PUSH");
6253 if (gimme == G_ARRAY) {
6255 /* EXTEND should not be needed - we just popped them */
6257 for (i=0; i < iters; i++) {
6258 SV **svp = av_fetch(ary, i, FALSE);
6259 PUSHs((svp) ? *svp : &PL_sv_undef);
6266 if (gimme == G_ARRAY)
6278 SV *const sv = PAD_SVl(PL_op->op_targ);
6280 if (SvPADSTALE(sv)) {
6283 RETURNOP(cLOGOP->op_other);
6285 RETURNOP(cLOGOP->op_next);
6294 assert(SvTYPE(retsv) != SVt_PVCV);
6296 if (SvTYPE(retsv) == SVt_PVAV || SvTYPE(retsv) == SVt_PVHV) {
6297 retsv = refto(retsv);
6304 PP(unimplemented_op)
6307 const Optype op_type = PL_op->op_type;
6308 /* Using OP_NAME() isn't going to be helpful here. Firstly, it doesn't cope
6309 with out of range op numbers - it only "special" cases op_custom.
6310 Secondly, as the three ops we "panic" on are padmy, mapstart and custom,
6311 if we get here for a custom op then that means that the custom op didn't
6312 have an implementation. Given that OP_NAME() looks up the custom op
6313 by its pp_addr, likely it will return NULL, unless someone (unhelpfully)
6314 registers &PL_unimplemented_op as the address of their custom op.
6315 NULL doesn't generate a useful error message. "custom" does. */
6316 const char *const name = op_type >= OP_max
6317 ? "[out of range]" : PL_op_name[PL_op->op_type];
6318 if(OP_IS_SOCKET(op_type))
6319 DIE(aTHX_ PL_no_sock_func, name);
6320 DIE(aTHX_ "panic: unimplemented op %s (#%d) called", name, op_type);
6327 HV * const hv = (HV*)POPs;
6329 if (SvTYPE(hv) != SVt_PVHV) { XPUSHs(&PL_sv_no); RETURN; }
6331 if (SvRMAGICAL(hv)) {
6332 MAGIC * const mg = mg_find((SV*)hv, PERL_MAGIC_tied);
6334 XPUSHs(magic_scalarpack(hv, mg));
6339 XPUSHs(boolSV(HvKEYS(hv) != 0));
6345 * c-indentation-style: bsd
6347 * indent-tabs-mode: t
6350 * ex: set ts=8 sts=4 sw=4 noet: