X-Git-Url: https://perl5.git.perl.org/perl5.git/blobdiff_plain/13a24baded0bcfd57a5060f46aead50d004b1374..ae0beba11d18ba5a4fa40f476873ab82ee11127b:/regexec.c diff --git a/regexec.c b/regexec.c index f93e17e..60ec4ff 100644 --- a/regexec.c +++ b/regexec.c @@ -13,7 +13,6 @@ * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT. * This causes the main functions to be compiled under new names and with * debugging support added, which makes "use re 'debug'" work. - */ /* NOTE: this is derived from Henry Spencer's regexp code, and should not @@ -78,8 +77,8 @@ #define RF_tainted 1 /* tainted information used? */ #define RF_warned 2 /* warned about big count? */ -#define RF_evaled 4 /* Did an EVAL with setting? */ -#define RF_utf8 8 /* String contains multibyte chars? */ + +#define RF_utf8 8 /* Pattern contains multibyte chars? */ #define UTF ((PL_reg_flags & RF_utf8) != 0) @@ -110,7 +109,7 @@ ? (U8*)pos - off \ : NULL) -#define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)pos, off, (U8*)lim) : (U8*)(pos + off)) +#define HOP3(pos,off,lim) (PL_reg_match_utf8 ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off)) #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim)) #define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \ @@ -151,6 +150,7 @@ } \ } STMT_END + static void restore_pos(pTHX_ void *arg); STATIC CHECKPOINT @@ -161,11 +161,12 @@ S_regcppush(pTHX_ I32 parenfloor) #define REGCP_PAREN_ELEMS 4 const int paren_elems_to_push = (PL_regsize - parenfloor) * REGCP_PAREN_ELEMS; int p; + GET_RE_DEBUG_FLAGS_DECL; if (paren_elems_to_push < 0) Perl_croak(aTHX_ "panic: paren_elems_to_push < 0"); -#define REGCP_OTHER_ELEMS 6 +#define REGCP_OTHER_ELEMS 8 SSGROW(paren_elems_to_push + REGCP_OTHER_ELEMS); for (p = PL_regsize; p > parenfloor; p--) { /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */ @@ -173,8 +174,16 @@ S_regcppush(pTHX_ I32 parenfloor) SSPUSHINT(PL_regstartp[p]); SSPUSHPTR(PL_reg_start_tmp[p]); SSPUSHINT(p); + DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, + " saving \\%"UVuf" %"IVdf"(%"IVdf")..%"IVdf"\n", + (UV)p, (IV)PL_regstartp[p], + (IV)(PL_reg_start_tmp[p] - PL_bostr), + (IV)PL_regendp[p] + )); } /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */ + SSPUSHPTR(PL_regstartp); + SSPUSHPTR(PL_regendp); SSPUSHINT(PL_regsize); SSPUSHINT(*PL_reglastparen); SSPUSHINT(*PL_reglastcloseparen); @@ -189,14 +198,20 @@ S_regcppush(pTHX_ I32 parenfloor) } /* These are needed since we do not localize EVAL nodes: */ -# define REGCP_SET(cp) DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, \ - " Setting an EVAL scope, savestack=%"IVdf"\n", \ - (IV)PL_savestack_ix)); cp = PL_savestack_ix - -# define REGCP_UNWIND(cp) DEBUG_EXECUTE_r(cp != PL_savestack_ix ? \ - PerlIO_printf(Perl_debug_log, \ - " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \ - (IV)(cp), (IV)PL_savestack_ix) : 0); regcpblow(cp) +#define REGCP_SET(cp) \ + DEBUG_STATE_r( \ + PerlIO_printf(Perl_debug_log, \ + " Setting an EVAL scope, savestack=%"IVdf"\n", \ + (IV)PL_savestack_ix)); \ + cp = PL_savestack_ix + +#define REGCP_UNWIND(cp) \ + DEBUG_STATE_r( \ + if (cp != PL_savestack_ix) \ + PerlIO_printf(Perl_debug_log, \ + " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \ + (IV)(cp), (IV)PL_savestack_ix)); \ + regcpblow(cp) STATIC char * S_regcppop(pTHX_ const regexp *rex) @@ -215,7 +230,10 @@ S_regcppop(pTHX_ const regexp *rex) *PL_reglastcloseparen = SSPOPINT; *PL_reglastparen = SSPOPINT; PL_regsize = SSPOPINT; + PL_regendp=(I32 *) SSPOPPTR; + PL_regstartp=(I32 *) SSPOPPTR; + /* Now restore the parentheses context. */ for (i -= (REGCP_OTHER_ELEMS - REGCP_FRAME_ELEMS); i > 0; i -= REGCP_PAREN_ELEMS) { @@ -264,23 +282,6 @@ S_regcppop(pTHX_ const regexp *rex) #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */ -#define TRYPAREN(paren, n, input, where) { \ - if (paren) { \ - if (n) { \ - PL_regstartp[paren] = HOPc(input, -1) - PL_bostr; \ - PL_regendp[paren] = input - PL_bostr; \ - } \ - else \ - PL_regendp[paren] = -1; \ - } \ - REGMATCH(next, where); \ - if (result) \ - sayYES; \ - if (paren && n) \ - PL_regendp[paren] = -1; \ -} - - /* * pregexec and friends */ @@ -361,14 +362,13 @@ Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos, register SV *check; char *strbeg; char *t; - const int do_utf8 = sv ? SvUTF8(sv) : 0; /* if no sv we have to assume bytes */ + const bool do_utf8 = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */ I32 ml_anch; register char *other_last = NULL; /* other substr checked before this */ char *check_at = NULL; /* check substr found at this pos */ const I32 multiline = prog->reganch & PMf_MULTILINE; #ifdef DEBUGGING const char * const i_strpos = strpos; - SV * const dsv = PERL_DEBUG_PAD_ZERO(0); #endif GET_RE_DEBUG_FLAGS_DECL; @@ -376,34 +376,13 @@ Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos, RX_MATCH_UTF8_set(prog,do_utf8); if (prog->reganch & ROPT_UTF8) { - DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, - "UTF-8 regex...\n")); PL_reg_flags |= RF_utf8; } - - DEBUG_EXECUTE_r({ - const char *s = PL_reg_match_utf8 ? - sv_uni_display(dsv, sv, 60, UNI_DISPLAY_REGEX) : - strpos; - const int len = PL_reg_match_utf8 ? - (int)strlen(s) : strend - strpos; - if (!PL_colorset) - reginitcolors(); - if (PL_reg_match_utf8) - DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, - "UTF-8 target...\n")); - PerlIO_printf(Perl_debug_log, - "%sGuessing start of match, REx%s \"%s%.60s%s%s\" against \"%s%.*s%s%s\"...\n", - PL_colors[4], PL_colors[5], PL_colors[0], - prog->precomp, - PL_colors[1], - (strlen(prog->precomp) > 60 ? "..." : ""), - PL_colors[0], - (int)(len > 60 ? 60 : len), - s, PL_colors[1], - (len > 60 ? "..." : "") + DEBUG_EXECUTE_r( + debug_start_match(prog, do_utf8, strpos, strend, + sv ? "Guessing start of match in sv for" + : "Guessing start of match in string for"); ); - }); /* CHR_DIST() would be more correct here but it makes things slow. */ if (prog->minlen > strend - strpos) { @@ -411,6 +390,7 @@ Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos, "String too short... [re_intuit_start]\n")); goto fail; } + strbeg = (sv && SvPOK(sv)) ? strend - SvCUR(sv) : strpos; PL_regeol = strend; if (do_utf8) { @@ -422,9 +402,9 @@ Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos, to_byte_substr(prog); check = prog->check_substr; } - if (check == &PL_sv_undef) { + if (check == &PL_sv_undef) { DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, - "Non-utf string cannot match utf check string\n")); + "Non-utf8 string cannot match utf8 check string\n")); goto fail; } if (prog->reganch & ROPT_ANCH) { /* Match at beg-of-str or after \n */ @@ -447,6 +427,7 @@ Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos, I32 slen; s = HOP3c(strpos, prog->check_offset_min, strend); + if (SvTAIL(check)) { slen = SvCUR(check); /* >= 1 */ @@ -476,8 +457,8 @@ Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos, /* Match is anchored, but substr is not anchored wrt beg-of-str. */ s = strpos; start_shift = prog->check_offset_min; /* okay to underestimate on CC */ - end_shift = prog->minlen - start_shift - - CHR_SVLEN(check) + (SvTAIL(check) != 0); + end_shift = prog->check_end_shift; + if (!ml_anch) { const I32 end = prog->check_offset_max + CHR_SVLEN(check) - (SvTAIL(check) != 0); @@ -490,20 +471,37 @@ Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos, else { /* Can match at random position */ ml_anch = 0; s = strpos; - start_shift = prog->check_offset_min; /* okay to underestimate on CC */ - /* Should be nonnegative! */ - end_shift = prog->minlen - start_shift - - CHR_SVLEN(check) + (SvTAIL(check) != 0); + start_shift = prog->check_offset_min; /* okay to underestimate on CC */ + end_shift = prog->check_end_shift; + + /* end shift should be non negative here */ } #ifdef DEBUGGING /* 7/99: reports of failure (with the older version) */ if (end_shift < 0) - Perl_croak(aTHX_ "panic: end_shift"); + Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ", + (IV)end_shift, prog->precomp); #endif restart: /* Find a possible match in the region s..strend by looking for the "check" substring in the region corrected by start/end_shift. */ + + { + I32 srch_start_shift = start_shift; + I32 srch_end_shift = end_shift; + if (srch_start_shift < 0 && strbeg - s > srch_start_shift) { + srch_end_shift -= ((strbeg - s) - srch_start_shift); + srch_start_shift = strbeg - s; + } + DEBUG_OPTIMISE_MORE_r({ + PerlIO_printf(Perl_debug_log, "Check offset min: %"IVdf" Start shift: %"IVdf" End shift %"IVdf" Real End Shift: %"IVdf"\n", + (IV)prog->check_offset_min, + (IV)srch_start_shift, + (IV)srch_end_shift, + (IV)prog->check_end_shift); + }); + if (flags & REXEC_SCREAM) { I32 p = -1; /* Internal iterator of scream. */ I32 * const pp = data ? data->scream_pos : &p; @@ -513,7 +511,7 @@ Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos, && (BmPREVIOUS(check) == SvCUR(check) - 1) && SvTAIL(check) )) s = screaminstr(sv, check, - start_shift + (s - strbeg), end_shift, pp, 0); + srch_start_shift + (s - strbeg), srch_end_shift, pp, 0); else goto fail_finish; /* we may be pointing at the wrong string */ @@ -522,35 +520,58 @@ Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos, if (data) *data->scream_olds = s; } - else if (prog->reganch & ROPT_CANY_SEEN) - s = fbm_instr((U8*)(s + start_shift), - (U8*)(strend - end_shift), - check, multiline ? FBMrf_MULTILINE : 0); - else - s = fbm_instr(HOP3(s, start_shift, strend), - HOP3(strend, -end_shift, strbeg), + else { + U8* start_point; + U8* end_point; + if (prog->reganch & ROPT_CANY_SEEN) { + start_point= (U8*)(s + srch_start_shift); + end_point= (U8*)(strend - srch_end_shift); + } else { + start_point= HOP3(s, srch_start_shift, srch_start_shift < 0 ? strbeg : strend); + end_point= HOP3(strend, -srch_end_shift, strbeg); + } + DEBUG_OPTIMISE_MORE_r({ + PerlIO_printf(Perl_debug_log, "fbm_instr len=%d str=<%.*s>\n", + (int)(end_point - start_point), + (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point), + start_point); + }); + + s = fbm_instr( start_point, end_point, check, multiline ? FBMrf_MULTILINE : 0); - + } + } /* Update the count-of-usability, remove useless subpatterns, unshift s. */ - DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s %s substr \"%s%.*s%s\"%s%s", + DEBUG_EXECUTE_r({ + RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0), + SvPVX_const(check), RE_SV_DUMPLEN(check), 30); + PerlIO_printf(Perl_debug_log, "%s %s substr %s%s%s", (s ? "Found" : "Did not find"), - (check == (do_utf8 ? prog->anchored_utf8 : prog->anchored_substr) ? "anchored" : "floating"), - PL_colors[0], - (int)(SvCUR(check) - (SvTAIL(check)!=0)), - SvPVX_const(check), - PL_colors[1], (SvTAIL(check) ? "$" : ""), - (s ? " at offset " : "...\n") ) ); + (check == (do_utf8 ? prog->anchored_utf8 : prog->anchored_substr) + ? "anchored" : "floating"), + quoted, + RE_SV_TAIL(check), + (s ? " at offset " : "...\n") ); + }); if (!s) goto fail_finish; - - check_at = s; - /* Finish the diagnostic message */ DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(s - i_strpos)) ); + /* XXX dmq: first branch is for positive lookbehind... + Our check string is offset from the beginning of the pattern. + So we need to do any stclass tests offset forward from that + point. I think. :-( + */ + + + + check_at=s; + + /* Got a candidate. Check MBOL anchoring, and the *other* substr. Start with the other substr. XXXX no SCREAM optimization yet - and a very coarse implementation @@ -559,7 +580,9 @@ Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos, Probably it is right to do no SCREAM here... */ - if (do_utf8 ? (prog->float_utf8 && prog->anchored_utf8) : (prog->float_substr && prog->anchored_substr)) { + if (do_utf8 ? (prog->float_utf8 && prog->anchored_utf8) + : (prog->float_substr && prog->anchored_substr)) + { /* Take into account the "other" substring. */ /* XXXX May be hopelessly wrong for UTF... */ if (!other_last) @@ -586,12 +609,17 @@ Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos, last2 = last1 = HOP3c(strend, -prog->minlen, strbeg); if (last < last1) last1 = last; - /* XXXX It is not documented what units *_offsets are in. Assume bytes. */ + /* XXXX It is not documented what units *_offsets are in. + We assume bytes, but this is clearly wrong. + Meaning this code needs to be carefully reviewed for errors. + dmq. + */ + /* On end-of-str: see comment below. */ must = do_utf8 ? prog->anchored_utf8 : prog->anchored_substr; if (must == &PL_sv_undef) { s = (char*)NULL; - DEBUG_EXECUTE_r(must = prog->anchored_utf8); /* for debug */ + DEBUG_r(must = prog->anchored_utf8); /* for debug */ } else s = fbm_instr( @@ -601,14 +629,15 @@ Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos, must, multiline ? FBMrf_MULTILINE : 0 ); - DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, - "%s anchored substr \"%s%.*s%s\"%s", + DEBUG_EXECUTE_r({ + RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0), + SvPVX_const(must), RE_SV_DUMPLEN(must), 30); + PerlIO_printf(Perl_debug_log, "%s anchored substr %s%s", (s ? "Found" : "Contradicts"), - PL_colors[0], - (int)(SvCUR(must) - - (SvTAIL(must)!=0)), - SvPVX_const(must), - PL_colors[1], (SvTAIL(must) ? "$" : ""))); + quoted, RE_SV_TAIL(must)); + }); + + if (!s) { if (last1 >= last2) { DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, @@ -654,19 +683,20 @@ Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos, and end-of-str is not later than strend we are OK. */ if (must == &PL_sv_undef) { s = (char*)NULL; - DEBUG_EXECUTE_r(must = prog->float_utf8); /* for debug message */ + DEBUG_r(must = prog->float_utf8); /* for debug message */ } else s = fbm_instr((unsigned char*)s, (unsigned char*)last + SvCUR(must) - (SvTAIL(must)!=0), must, multiline ? FBMrf_MULTILINE : 0); - DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s floating substr \"%s%.*s%s\"%s", + DEBUG_EXECUTE_r({ + RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0), + SvPVX_const(must), RE_SV_DUMPLEN(must), 30); + PerlIO_printf(Perl_debug_log, "%s floating substr %s%s", (s ? "Found" : "Contradicts"), - PL_colors[0], - (int)(SvCUR(must) - (SvTAIL(must)!=0)), - SvPVX_const(must), - PL_colors[1], (SvTAIL(must) ? "$" : ""))); + quoted, RE_SV_TAIL(must)); + }); if (!s) { if (last1 == last) { DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, @@ -692,11 +722,26 @@ Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos, } } - t = s - prog->check_offset_max; + + t= (char*)HOP3( s, -prog->check_offset_max, (prog->check_offset_max<0) ? strend : strpos); + + DEBUG_OPTIMISE_MORE_r( + PerlIO_printf(Perl_debug_log, + "Check offset min:%"IVdf" max:%"IVdf" S:%"IVdf" t:%"IVdf" D:%"IVdf" end:%"IVdf"\n", + (IV)prog->check_offset_min, + (IV)prog->check_offset_max, + (IV)(s-strpos), + (IV)(t-strpos), + (IV)(t-s), + (IV)(strend-strpos) + ) + ); + if (s - strpos > prog->check_offset_max /* signed-corrected t > strpos */ && (!do_utf8 - || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*)strpos)) - && t > strpos))) { + || ((t = (char*)reghopmaybe3((U8*)s, -prog->check_offset_max, (U8*) ((prog->check_offset_max<0) ? strend : strpos))) + && t > strpos))) + { /* Fixed substring is found far enough so that the match cannot start at strpos. */ try_at_offset: @@ -805,7 +850,9 @@ Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos, /* Last resort... */ /* XXXX BmUSEFUL already changed, maybe multiple change is meaningful... */ - if (prog->regstclass && OP(prog->regstclass)!=TRIE) { + /* trie stclasses are too expensive to use here, we are better off to + leave it to regmatch itself */ + if (prog->regstclass && PL_regkind[OP(prog->regstclass)]!=TRIE) { /* minlen == 0 is possible if regstclass is \b or \B, and the fixed substr is ''$. Since minlen is already taken into account, s+1 is before strend; @@ -818,14 +865,17 @@ Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos, const int cl_l = (PL_regkind[OP(prog->regstclass)] == EXACT ? CHR_DIST(str+STR_LEN(prog->regstclass), str) : 1); - const char * endpos = (prog->anchored_substr || prog->anchored_utf8 || ml_anch) - ? HOP3c(s, (prog->minlen ? cl_l : 0), strend) - : (prog->float_substr || prog->float_utf8 - ? HOP3c(HOP3c(check_at, -start_shift, strbeg), - cl_l, strend) - : strend); - /*if (OP(prog->regstclass) == TRIE) - endpos++;*/ + char * endpos; + if (prog->anchored_substr || prog->anchored_utf8 || ml_anch) + endpos= HOP3c(s, (prog->minlen ? cl_l : 0), strend); + else if (prog->float_substr || prog->float_utf8) + endpos= HOP3c(HOP3c(check_at, -start_shift, strbeg), cl_l, strend); + else + endpos= strend; + + DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "start_shift: %"IVdf" check_at: %d s: %d endpos: %d\n", + (IV)start_shift, check_at - strbeg, s - strbeg, endpos - strbeg)); + t = s; s = find_byclass(prog, prog->regstclass, s, endpos, NULL); if (!s) { @@ -918,6 +968,149 @@ Perl_re_intuit_start(pTHX_ regexp *prog, SV *sv, char *strpos, return NULL; } + + +#define REXEC_TRIE_READ_CHAR(trie_type, trie, uc, uscan, len, uvc, charid, \ +foldlen, foldbuf, uniflags) STMT_START { \ + switch (trie_type) { \ + case trie_utf8_fold: \ + if ( foldlen>0 ) { \ + uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); \ + foldlen -= len; \ + uscan += len; \ + len=0; \ + } else { \ + uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \ + uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \ + foldlen -= UNISKIP( uvc ); \ + uscan = foldbuf + UNISKIP( uvc ); \ + } \ + break; \ + case trie_utf8: \ + uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \ + break; \ + case trie_plain: \ + uvc = (UV)*uc; \ + len = 1; \ + } \ + \ + if (uvc < 256) { \ + charid = trie->charmap[ uvc ]; \ + } \ + else { \ + charid = 0; \ + if (trie->widecharmap) { \ + SV** const svpp = hv_fetch(trie->widecharmap, \ + (char*)&uvc, sizeof(UV), 0); \ + if (svpp) \ + charid = (U16)SvIV(*svpp); \ + } \ + } \ +} STMT_END + +#define REXEC_FBC_EXACTISH_CHECK(CoNd) \ + if ( (CoNd) \ + && (ln == len || \ + ibcmp_utf8(s, NULL, 0, do_utf8, \ + m, NULL, ln, (bool)UTF)) \ + && (!reginfo || regtry(reginfo, s)) ) \ + goto got_it; \ + else { \ + U8 foldbuf[UTF8_MAXBYTES_CASE+1]; \ + uvchr_to_utf8(tmpbuf, c); \ + f = to_utf8_fold(tmpbuf, foldbuf, &foldlen); \ + if ( f != c \ + && (f == c1 || f == c2) \ + && (ln == foldlen || \ + !ibcmp_utf8((char *) foldbuf, \ + NULL, foldlen, do_utf8, \ + m, \ + NULL, ln, (bool)UTF)) \ + && (!reginfo || regtry(reginfo, s)) ) \ + goto got_it; \ + } \ + s += len + +#define REXEC_FBC_EXACTISH_SCAN(CoNd) \ +STMT_START { \ + while (s <= e) { \ + if ( (CoNd) \ + && (ln == 1 || !(OP(c) == EXACTF \ + ? ibcmp(s, m, ln) \ + : ibcmp_locale(s, m, ln))) \ + && (!reginfo || regtry(reginfo, s)) ) \ + goto got_it; \ + s++; \ + } \ +} STMT_END + +#define REXEC_FBC_UTF8_SCAN(CoDe) \ +STMT_START { \ + while (s + (uskip = UTF8SKIP(s)) <= strend) { \ + CoDe \ + s += uskip; \ + } \ +} STMT_END + +#define REXEC_FBC_SCAN(CoDe) \ +STMT_START { \ + while (s < strend) { \ + CoDe \ + s++; \ + } \ +} STMT_END + +#define REXEC_FBC_UTF8_CLASS_SCAN(CoNd) \ +REXEC_FBC_UTF8_SCAN( \ + if (CoNd) { \ + if (tmp && (!reginfo || regtry(reginfo, s))) \ + goto got_it; \ + else \ + tmp = doevery; \ + } \ + else \ + tmp = 1; \ +) + +#define REXEC_FBC_CLASS_SCAN(CoNd) \ +REXEC_FBC_SCAN( \ + if (CoNd) { \ + if (tmp && (!reginfo || regtry(reginfo, s))) \ + goto got_it; \ + else \ + tmp = doevery; \ + } \ + else \ + tmp = 1; \ +) + +#define REXEC_FBC_TRYIT \ +if ((!reginfo || regtry(reginfo, s))) \ + goto got_it + +#define REXEC_FBC_CSCAN_PRELOAD(UtFpReLoAd,CoNdUtF8,CoNd) \ + if (do_utf8) { \ + UtFpReLoAd; \ + REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \ + } \ + else { \ + REXEC_FBC_CLASS_SCAN(CoNd); \ + } \ + break + +#define REXEC_FBC_CSCAN_TAINT(CoNdUtF8,CoNd) \ + PL_reg_flags |= RF_tainted; \ + if (do_utf8) { \ + REXEC_FBC_UTF8_CLASS_SCAN(CoNdUtF8); \ + } \ + else { \ + REXEC_FBC_CLASS_SCAN(CoNd); \ + } \ + break + +#define DUMP_EXEC_POS(li,s,doutf8) \ + dump_exec_pos(li,s,(PL_regeol),(PL_bostr),(PL_reg_starttry),doutf8) + /* We know what class REx starts with. Try to find this position... */ /* if reginfo is NULL, its a dryrun */ /* annoyingly all the vars in this routine have different names from their counterparts @@ -943,20 +1136,10 @@ S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s, switch (OP(c)) { case ANYOF: if (do_utf8) { - while (s + (uskip = UTF8SKIP(s)) <= strend) { - if ((ANYOF_FLAGS(c) & ANYOF_UNICODE) || + REXEC_FBC_UTF8_CLASS_SCAN((ANYOF_FLAGS(c) & ANYOF_UNICODE) || !UTF8_IS_INVARIANT((U8)s[0]) ? reginclass(prog, c, (U8*)s, 0, do_utf8) : - REGINCLASS(prog, c, (U8*)s)) { - if (tmp && (!reginfo || regtry(reginfo, s))) - goto got_it; - else - tmp = doevery; - } - else - tmp = 1; - s += uskip; - } + REGINCLASS(prog, c, (U8*)s)); } else { while (s < strend) { @@ -979,13 +1162,12 @@ S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s, } break; case CANY: - while (s < strend) { + REXEC_FBC_SCAN( if (tmp && (!reginfo || regtry(reginfo, s))) goto got_it; else tmp = doevery; - s++; - } + ); break; case EXACTF: m = STRING(c); @@ -1050,27 +1232,7 @@ S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s, while (s <= e) { c = utf8n_to_uvchr((U8*)s, UTF8_MAXBYTES, &len, uniflags); - if ( c == c1 - && (ln == len || - ibcmp_utf8(s, NULL, 0, do_utf8, - m, NULL, ln, (bool)UTF)) - && (!reginfo || regtry(reginfo, s)) ) - goto got_it; - else { - U8 foldbuf[UTF8_MAXBYTES_CASE+1]; - uvchr_to_utf8(tmpbuf, c); - f = to_utf8_fold(tmpbuf, foldbuf, &foldlen); - if ( f != c - && (f == c1 || f == c2) - && (ln == foldlen || - !ibcmp_utf8((char *) foldbuf, - NULL, foldlen, do_utf8, - m, - NULL, ln, (bool)UTF)) - && (!reginfo || regtry(reginfo, s)) ) - goto got_it; - } - s += len; + REXEC_FBC_EXACTISH_CHECK(c == c1); } } else { @@ -1089,51 +1251,15 @@ S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s, c == (UV)UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA) c = (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA; - if ( (c == c1 || c == c2) - && (ln == len || - ibcmp_utf8(s, NULL, 0, do_utf8, - m, NULL, ln, (bool)UTF)) - && (!reginfo || regtry(reginfo, s)) ) - goto got_it; - else { - U8 foldbuf[UTF8_MAXBYTES_CASE+1]; - uvchr_to_utf8(tmpbuf, c); - f = to_utf8_fold(tmpbuf, foldbuf, &foldlen); - if ( f != c - && (f == c1 || f == c2) - && (ln == foldlen || - !ibcmp_utf8((char *) foldbuf, - NULL, foldlen, do_utf8, - m, - NULL, ln, (bool)UTF)) - && (!reginfo || regtry(reginfo, s)) ) - goto got_it; - } - s += len; + REXEC_FBC_EXACTISH_CHECK(c == c1 || c == c2); } } } else { if (c1 == c2) - while (s <= e) { - if ( *(U8*)s == c1 - && (ln == 1 || !(OP(c) == EXACTF - ? ibcmp(s, m, ln) - : ibcmp_locale(s, m, ln))) - && (!reginfo || regtry(reginfo, s)) ) - goto got_it; - s++; - } + REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1); else - while (s <= e) { - if ( (*(U8*)s == c1 || *(U8*)s == c2) - && (ln == 1 || !(OP(c) == EXACTF - ? ibcmp(s, m, ln) - : ibcmp_locale(s, m, ln))) - && (!reginfo || regtry(reginfo, s)) ) - goto got_it; - s++; - } + REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2); } break; case BOUNDL: @@ -1150,30 +1276,26 @@ S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s, tmp = ((OP(c) == BOUND ? isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0); LOAD_UTF8_CHARCLASS_ALNUM(); - while (s + (uskip = UTF8SKIP(s)) <= strend) { + REXEC_FBC_UTF8_SCAN( if (tmp == !(OP(c) == BOUND ? (bool)swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8) : isALNUM_LC_utf8((U8*)s))) { tmp = !tmp; - if ((!reginfo || regtry(reginfo, s))) - goto got_it; - } - s += uskip; + REXEC_FBC_TRYIT; } + ); } else { tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; tmp = ((OP(c) == BOUND ? isALNUM(tmp) : isALNUM_LC(tmp)) != 0); - while (s < strend) { + REXEC_FBC_SCAN( if (tmp == !(OP(c) == BOUND ? isALNUM(*s) : isALNUM_LC(*s))) { tmp = !tmp; - if ((!reginfo || regtry(reginfo, s))) - goto got_it; - } - s++; + REXEC_FBC_TRYIT; } + ); } if ((!prog->minlen && tmp) && (!reginfo || regtry(reginfo, s))) goto got_it; @@ -1192,382 +1314,96 @@ S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s, tmp = ((OP(c) == NBOUND ? isALNUM_uni(tmp) : isALNUM_LC_uvchr(UNI_TO_NATIVE(tmp))) != 0); LOAD_UTF8_CHARCLASS_ALNUM(); - while (s + (uskip = UTF8SKIP(s)) <= strend) { + REXEC_FBC_UTF8_SCAN( if (tmp == !(OP(c) == NBOUND ? (bool)swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8) : isALNUM_LC_utf8((U8*)s))) tmp = !tmp; - else if ((!reginfo || regtry(reginfo, s))) - goto got_it; - s += uskip; - } + else REXEC_FBC_TRYIT; + ); } else { tmp = (s != PL_bostr) ? UCHARAT(s - 1) : '\n'; tmp = ((OP(c) == NBOUND ? isALNUM(tmp) : isALNUM_LC(tmp)) != 0); - while (s < strend) { + REXEC_FBC_SCAN( if (tmp == !(OP(c) == NBOUND ? isALNUM(*s) : isALNUM_LC(*s))) tmp = !tmp; - else if ((!reginfo || regtry(reginfo, s))) - goto got_it; - s++; - } + else REXEC_FBC_TRYIT; + ); } if ((!prog->minlen && !tmp) && (!reginfo || regtry(reginfo, s))) goto got_it; break; case ALNUM: - if (do_utf8) { - LOAD_UTF8_CHARCLASS_ALNUM(); - while (s + (uskip = UTF8SKIP(s)) <= strend) { - if (swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8)) { - if (tmp && (!reginfo || regtry(reginfo, s))) - goto got_it; - else - tmp = doevery; - } - else - tmp = 1; - s += uskip; - } - } - else { - while (s < strend) { - if (isALNUM(*s)) { - if (tmp && (!reginfo || regtry(reginfo, s))) - goto got_it; - else - tmp = doevery; - } - else - tmp = 1; - s++; - } - } - break; + REXEC_FBC_CSCAN_PRELOAD( + LOAD_UTF8_CHARCLASS_ALNUM(), + swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8), + isALNUM(*s) + ); case ALNUML: - PL_reg_flags |= RF_tainted; - if (do_utf8) { - while (s + (uskip = UTF8SKIP(s)) <= strend) { - if (isALNUM_LC_utf8((U8*)s)) { - if (tmp && (!reginfo || regtry(reginfo, s))) - goto got_it; - else - tmp = doevery; - } - else - tmp = 1; - s += uskip; - } - } - else { - while (s < strend) { - if (isALNUM_LC(*s)) { - if (tmp && (!reginfo || regtry(reginfo, s))) - goto got_it; - else - tmp = doevery; - } - else - tmp = 1; - s++; - } - } - break; + REXEC_FBC_CSCAN_TAINT( + isALNUM_LC_utf8((U8*)s), + isALNUM_LC(*s) + ); case NALNUM: - if (do_utf8) { - LOAD_UTF8_CHARCLASS_ALNUM(); - while (s + (uskip = UTF8SKIP(s)) <= strend) { - if (!swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8)) { - if (tmp && (!reginfo || regtry(reginfo, s))) - goto got_it; - else - tmp = doevery; - } - else - tmp = 1; - s += uskip; - } - } - else { - while (s < strend) { - if (!isALNUM(*s)) { - if (tmp && (!reginfo || regtry(reginfo, s))) - goto got_it; - else - tmp = doevery; - } - else - tmp = 1; - s++; - } - } - break; + REXEC_FBC_CSCAN_PRELOAD( + LOAD_UTF8_CHARCLASS_ALNUM(), + !swash_fetch(PL_utf8_alnum, (U8*)s, do_utf8), + !isALNUM(*s) + ); case NALNUML: - PL_reg_flags |= RF_tainted; - if (do_utf8) { - while (s + (uskip = UTF8SKIP(s)) <= strend) { - if (!isALNUM_LC_utf8((U8*)s)) { - if (tmp && (!reginfo || regtry(reginfo, s))) - goto got_it; - else - tmp = doevery; - } - else - tmp = 1; - s += uskip; - } - } - else { - while (s < strend) { - if (!isALNUM_LC(*s)) { - if (tmp && (!reginfo || regtry(reginfo, s))) - goto got_it; - else - tmp = doevery; - } - else - tmp = 1; - s++; - } - } - break; + REXEC_FBC_CSCAN_TAINT( + !isALNUM_LC_utf8((U8*)s), + !isALNUM_LC(*s) + ); case SPACE: - if (do_utf8) { - LOAD_UTF8_CHARCLASS_SPACE(); - while (s + (uskip = UTF8SKIP(s)) <= strend) { - if (*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, do_utf8)) { - if (tmp && (!reginfo || regtry(reginfo, s))) - goto got_it; - else - tmp = doevery; - } - else - tmp = 1; - s += uskip; - } - } - else { - while (s < strend) { - if (isSPACE(*s)) { - if (tmp && (!reginfo || regtry(reginfo, s))) - goto got_it; - else - tmp = doevery; - } - else - tmp = 1; - s++; - } - } - break; + REXEC_FBC_CSCAN_PRELOAD( + LOAD_UTF8_CHARCLASS_SPACE(), + *s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, do_utf8), + isSPACE(*s) + ); case SPACEL: - PL_reg_flags |= RF_tainted; - if (do_utf8) { - while (s + (uskip = UTF8SKIP(s)) <= strend) { - if (*s == ' ' || isSPACE_LC_utf8((U8*)s)) { - if (tmp && (!reginfo || regtry(reginfo, s))) - goto got_it; - else - tmp = doevery; - } - else - tmp = 1; - s += uskip; - } - } - else { - while (s < strend) { - if (isSPACE_LC(*s)) { - if (tmp && (!reginfo || regtry(reginfo, s))) - goto got_it; - else - tmp = doevery; - } - else - tmp = 1; - s++; - } - } - break; + REXEC_FBC_CSCAN_TAINT( + *s == ' ' || isSPACE_LC_utf8((U8*)s), + isSPACE_LC(*s) + ); case NSPACE: - if (do_utf8) { - LOAD_UTF8_CHARCLASS_SPACE(); - while (s + (uskip = UTF8SKIP(s)) <= strend) { - if (!(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, do_utf8))) { - if (tmp && (!reginfo || regtry(reginfo, s))) - goto got_it; - else - tmp = doevery; - } - else - tmp = 1; - s += uskip; - } - } - else { - while (s < strend) { - if (!isSPACE(*s)) { - if (tmp && (!reginfo || regtry(reginfo, s))) - goto got_it; - else - tmp = doevery; - } - else - tmp = 1; - s++; - } - } - break; + REXEC_FBC_CSCAN_PRELOAD( + LOAD_UTF8_CHARCLASS_SPACE(), + !(*s == ' ' || swash_fetch(PL_utf8_space,(U8*)s, do_utf8)), + !isSPACE(*s) + ); case NSPACEL: - PL_reg_flags |= RF_tainted; - if (do_utf8) { - while (s + (uskip = UTF8SKIP(s)) <= strend) { - if (!(*s == ' ' || isSPACE_LC_utf8((U8*)s))) { - if (tmp && (!reginfo || regtry(reginfo, s))) - goto got_it; - else - tmp = doevery; - } - else - tmp = 1; - s += uskip; - } - } - else { - while (s < strend) { - if (!isSPACE_LC(*s)) { - if (tmp && (!reginfo || regtry(reginfo, s))) - goto got_it; - else - tmp = doevery; - } - else - tmp = 1; - s++; - } - } - break; + REXEC_FBC_CSCAN_TAINT( + !(*s == ' ' || isSPACE_LC_utf8((U8*)s)), + !isSPACE_LC(*s) + ); case DIGIT: - if (do_utf8) { - LOAD_UTF8_CHARCLASS_DIGIT(); - while (s + (uskip = UTF8SKIP(s)) <= strend) { - if (swash_fetch(PL_utf8_digit,(U8*)s, do_utf8)) { - if (tmp && (!reginfo || regtry(reginfo, s))) - goto got_it; - else - tmp = doevery; - } - else - tmp = 1; - s += uskip; - } - } - else { - while (s < strend) { - if (isDIGIT(*s)) { - if (tmp && (!reginfo || regtry(reginfo, s))) - goto got_it; - else - tmp = doevery; - } - else - tmp = 1; - s++; - } - } - break; + REXEC_FBC_CSCAN_PRELOAD( + LOAD_UTF8_CHARCLASS_DIGIT(), + swash_fetch(PL_utf8_digit,(U8*)s, do_utf8), + isDIGIT(*s) + ); case DIGITL: - PL_reg_flags |= RF_tainted; - if (do_utf8) { - while (s + (uskip = UTF8SKIP(s)) <= strend) { - if (isDIGIT_LC_utf8((U8*)s)) { - if (tmp && (!reginfo || regtry(reginfo, s))) - goto got_it; - else - tmp = doevery; - } - else - tmp = 1; - s += uskip; - } - } - else { - while (s < strend) { - if (isDIGIT_LC(*s)) { - if (tmp && (!reginfo || regtry(reginfo, s))) - goto got_it; - else - tmp = doevery; - } - else - tmp = 1; - s++; - } - } - break; + REXEC_FBC_CSCAN_TAINT( + isDIGIT_LC_utf8((U8*)s), + isDIGIT_LC(*s) + ); case NDIGIT: - if (do_utf8) { - LOAD_UTF8_CHARCLASS_DIGIT(); - while (s + (uskip = UTF8SKIP(s)) <= strend) { - if (!swash_fetch(PL_utf8_digit,(U8*)s, do_utf8)) { - if (tmp && (!reginfo || regtry(reginfo, s))) - goto got_it; - else - tmp = doevery; - } - else - tmp = 1; - s += uskip; - } - } - else { - while (s < strend) { - if (!isDIGIT(*s)) { - if (tmp && (!reginfo || regtry(reginfo, s))) - goto got_it; - else - tmp = doevery; - } - else - tmp = 1; - s++; - } - } - break; + REXEC_FBC_CSCAN_PRELOAD( + LOAD_UTF8_CHARCLASS_DIGIT(), + !swash_fetch(PL_utf8_digit,(U8*)s, do_utf8), + !isDIGIT(*s) + ); case NDIGITL: - PL_reg_flags |= RF_tainted; - if (do_utf8) { - while (s + (uskip = UTF8SKIP(s)) <= strend) { - if (!isDIGIT_LC_utf8((U8*)s)) { - if (tmp && (!reginfo || regtry(reginfo, s))) - goto got_it; - else - tmp = doevery; - } - else - tmp = 1; - s += uskip; - } - } - else { - while (s < strend) { - if (!isDIGIT_LC(*s)) { - if (tmp && (!reginfo || regtry(reginfo, s))) - goto got_it; - else - tmp = doevery; - } - else - tmp = 1; - s++; - } - } - break; - case TRIE: - /*Perl_croak(aTHX_ "panic: unknown regstclass TRIE");*/ + REXEC_FBC_CSCAN_TAINT( + !isDIGIT_LC_utf8((U8*)s), + !isDIGIT_LC(*s) + ); + case AHOCORASICKC: + case AHOCORASICK: { const enum { trie_plain, trie_utf8, trie_utf8_fold } trie_type = do_utf8 ? @@ -1579,14 +1415,19 @@ S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s, reg_trie_data *trie=aho->trie; const char *last_start = strend - trie->minlen; +#ifdef DEBUGGING const char *real_start = s; +#endif STRLEN maxlen = trie->maxlen; SV *sv_points; U8 **points; /* map of where we were in the input string - when reading a given string. For ASCII this + when reading a given char. For ASCII this is unnecessary overhead as the relationship is always 1:1, but for unicode, especially case folded unicode this is not true. */ + U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ]; + U8 *bitmap=NULL; + GET_RE_DEBUG_FLAGS_DECL; @@ -1601,13 +1442,31 @@ S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s, SvPOK_on(sv_points); sv_2mortal(sv_points); points=(U8**)SvPV_nolen(sv_points ); - - if (trie->bitmap && trie_type != trie_utf8_fold) { - while (s <= last_start && !TRIE_BITMAP_TEST(trie,*s) ) { - s++; - } + if ( trie_type != trie_utf8_fold + && (trie->bitmap || OP(c)==AHOCORASICKC) ) + { + if (trie->bitmap) + bitmap=(U8*)trie->bitmap; + else + bitmap=(U8*)ANYOF_BITMAP(c); } - + /* this is the Aho-Corasick algorithm modified a touch + to include special handling for long "unknown char" + sequences. The basic idea being that we use AC as long + as we are dealing with a possible matching char, when + we encounter an unknown char (and we have not encountered + an accepting state) we scan forward until we find a legal + starting char. + AC matching is basically that of trie matching, except + that when we encounter a failing transition, we fall back + to the current states "fail state", and try the current char + again, a process we repeat until we reach the root state, + state 1, or a legal transition. If we fail on the root state + then we can either terminate if we have reached an accepting + state previously, or restart the entire process from the beginning + if we have not. + + */ while (s <= last_start) { const U32 uniflags = UTF8_ALLOW_DEFAULT; U8 *uc = (U8*)s; @@ -1619,71 +1478,68 @@ S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s, STRLEN foldlen = 0; U8 *uscan = (U8*)NULL; U8 *leftmost = NULL; - +#ifdef DEBUGGING + U32 accepted_word= 0; +#endif U32 pointpos = 0; while ( state && uc <= (U8*)strend ) { int failed=0; - if (aho->states[ state ].wordnum) { - U8 *lpos= points[ (pointpos - trie->wordlen[aho->states[ state ].wordnum-1] ) % maxlen ]; - if (!leftmost || lpos < leftmost) - leftmost= lpos; - if (base==0) break; - } - points[pointpos++ % maxlen]= uc; - switch (trie_type) { - case trie_utf8_fold: - if ( foldlen>0 ) { - uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); - foldlen -= len; - uscan += len; - len=0; - } else { - U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ]; - uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); - uvc = to_uni_fold( uvc, foldbuf, &foldlen ); - foldlen -= UNISKIP( uvc ); - uscan = foldbuf + UNISKIP( uvc ); + U32 word = aho->states[ state ].wordnum; + + if( state==1 ) { + if ( bitmap ) { + DEBUG_TRIE_EXECUTE_r( + if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) { + dump_exec_pos( (char *)uc, c, strend, real_start, + (char *)uc, do_utf8 ); + PerlIO_printf( Perl_debug_log, + " Scanning for legal start char...\n"); + } + ); + while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) { + uc++; + } + s= (char *)uc; } - break; - case trie_utf8: - uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, - &len, uniflags ); - break; - case trie_plain: - uvc = (UV)*uc; - len = 1; - } - - if (uvc < 256) { - charid = trie->charmap[ uvc ]; + if (uc >(U8*)last_start) break; } - else { - charid = 0; - if (trie->widecharmap) { - SV** const svpp = hv_fetch(trie->widecharmap, - (char*)&uvc, sizeof(UV), 0); - if (svpp) - charid = (U16)SvIV(*svpp); + + if ( word ) { + U8 *lpos= points[ (pointpos - trie->wordlen[word-1] ) % maxlen ]; + if (!leftmost || lpos < leftmost) { + DEBUG_r(accepted_word=word); + leftmost= lpos; } + if (base==0) break; + } - DEBUG_TRIE_EXECUTE_r( + points[pointpos++ % maxlen]= uc; + REXEC_TRIE_READ_CHAR(trie_type, trie, uc, uscan, len, + uvc, charid, foldlen, foldbuf, uniflags); + DEBUG_TRIE_EXECUTE_r({ + dump_exec_pos( (char *)uc, c, strend, real_start, + s, do_utf8 ); PerlIO_printf(Perl_debug_log, - "Pos: %d Charid:%3x CV:%4"UVxf" ", - (int)((const char*)uc - real_start), charid, uvc) - ); - uc += len; + " Charid:%3u CP:%4"UVxf" ", + charid, uvc); + }); do { - U32 word = aho->states[ state ].wordnum; +#ifdef DEBUGGING + word = aho->states[ state ].wordnum; +#endif base = aho->states[ state ].trans.base; - DEBUG_TRIE_EXECUTE_r( + DEBUG_TRIE_EXECUTE_r({ + if (failed) + dump_exec_pos( (char *)uc, c, strend, real_start, + s, do_utf8 ); PerlIO_printf( Perl_debug_log, - "%sState: %4"UVxf", Base: 0x%-4"UVxf" uvc=%"UVxf" word=%"UVxf"\n", - failed ? "Fail transition to " : "", - state, base, uvc, word) - ); + "%sState: %4"UVxf", word=%"UVxf, + failed ? " Fail transition to " : "", + (UV)state, (UV)word); + }); if ( base ) { U32 tmp; if (charid && @@ -1695,53 +1551,60 @@ S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s, && (tmp=trie->trans[base + charid - 1 - trie->uniquecharcount ].next)) { + DEBUG_TRIE_EXECUTE_r( + PerlIO_printf( Perl_debug_log," - legal\n")); state = tmp; break; } else { - failed++; - if ( state == 1 ) - break; - else - state = aho->fail[state]; + DEBUG_TRIE_EXECUTE_r( + PerlIO_printf( Perl_debug_log," - fail\n")); + failed = 1; + state = aho->fail[state]; } } else { /* we must be accepting here */ - failed++; + DEBUG_TRIE_EXECUTE_r( + PerlIO_printf( Perl_debug_log," - accepting\n")); + failed = 1; break; } } while(state); + uc += len; if (failed) { if (leftmost) break; - else if (!charid && trie->bitmap && trie_type != trie_utf8_fold) { - while ( uc <= (U8*)last_start && !TRIE_BITMAP_TEST(trie,*uc) ) { - uc++; - } - } + if (!state) state = 1; } } if ( aho->states[ state ].wordnum ) { U8 *lpos = points[ (pointpos - trie->wordlen[aho->states[ state ].wordnum-1]) % maxlen ]; - if (!leftmost || lpos < leftmost) + if (!leftmost || lpos < leftmost) { + DEBUG_r(accepted_word=aho->states[ state ].wordnum); leftmost = lpos; + } } - DEBUG_TRIE_EXECUTE_r( - PerlIO_printf( Perl_debug_log, - "%sState: %4"UVxf", Base: 0x%-4"UVxf" uvc=%"UVxf"\n", - "All done: ", - state, base, uvc) - ); if (leftmost) { s = (char*)leftmost; + DEBUG_TRIE_EXECUTE_r({ + PerlIO_printf( + Perl_debug_log,"Matches word #%"UVxf" at position %d. Trying full pattern...\n", + (UV)accepted_word, s - real_start + ); + }); if (!reginfo || regtry(reginfo, s)) { FREETMPS; LEAVE; goto got_it; } s = HOPc(s,1); + DEBUG_TRIE_EXECUTE_r({ + PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n"); + }); } else { + DEBUG_TRIE_EXECUTE_r( + PerlIO_printf( Perl_debug_log,"No match.\n")); break; } } @@ -1780,12 +1643,9 @@ Perl_regexec_flags(pTHX_ register regexp *prog, char *stringarg, register char * I32 scream_pos = -1; /* Internal iterator of scream. */ char *scream_olds = NULL; SV* const oreplsv = GvSV(PL_replgv); - const bool do_utf8 = DO_UTF8(sv); + const bool do_utf8 = (bool)DO_UTF8(sv); I32 multiline; -#ifdef DEBUGGING - SV* dsv0; - SV* dsv1; -#endif + regmatch_info reginfo; /* create some info to pass to regtry etc */ GET_RE_DEBUG_FLAGS_DECL; @@ -1801,20 +1661,21 @@ Perl_regexec_flags(pTHX_ register regexp *prog, char *stringarg, register char * multiline = prog->reganch & PMf_MULTILINE; reginfo.prog = prog; -#ifdef DEBUGGING - dsv0 = PERL_DEBUG_PAD_ZERO(0); - dsv1 = PERL_DEBUG_PAD_ZERO(1); -#endif - RX_MATCH_UTF8_set(prog, do_utf8); + DEBUG_EXECUTE_r( + debug_start_match(prog, do_utf8, startpos, strend, + "Matching"); + ); minlen = prog->minlen; - if (strend - startpos < minlen) { + + if (strend - startpos < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) { DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "String too short [regexec_flags]...\n")); goto phooey; } + /* Check validity of program. */ if (UCHARAT(prog->program) != REG_MAGIC) { Perl_croak(aTHX_ "corrupted regexp program"); @@ -1873,29 +1734,7 @@ Perl_regexec_flags(pTHX_ register regexp *prog, char *stringarg, register char * } } - DEBUG_EXECUTE_r({ - const char * const s0 = UTF - ? pv_uni_display(dsv0, (U8*)prog->precomp, prog->prelen, 60, - UNI_DISPLAY_REGEX) - : prog->precomp; - const int len0 = UTF ? (int)SvCUR(dsv0) : prog->prelen; - const char * const s1 = do_utf8 ? sv_uni_display(dsv1, sv, 60, - UNI_DISPLAY_REGEX) : startpos; - const int len1 = do_utf8 ? (int)SvCUR(dsv1) : strend - startpos; - if (!PL_colorset) - reginitcolors(); - PerlIO_printf(Perl_debug_log, - "%sMatching REx%s \"%s%*.*s%s%s\" against \"%s%.*s%s%s\"\n", - PL_colors[4], PL_colors[5], PL_colors[0], - len0, len0, s0, - PL_colors[1], - len0 > 60 ? "..." : "", - PL_colors[0], - (int)(len1 > 60 ? 60 : len1), - s1, PL_colors[1], - (len1 > 60 ? "..." : "") - ); - }); + /* Simplest case: anchored match need be tried only once. */ /* [unless only anchor is BOL and multiline is set] */ @@ -1940,7 +1779,11 @@ Perl_regexec_flags(pTHX_ register regexp *prog, char *stringarg, register char * } } goto phooey; - } else if (prog->reganch & ROPT_ANCH_GPOS) { + } else if (ROPT_GPOS_CHECK == (prog->reganch & ROPT_GPOS_CHECK)) + { + /* the warning about reginfo.ganch being used without intialization + is bogus -- we set it above, when prog->reganch & ROPT_GPOS_SEEN + and we only enter this block when the same bit is set. */ if (regtry(®info, reginfo.ganch)) goto got_it; goto phooey; @@ -1959,7 +1802,7 @@ Perl_regexec_flags(pTHX_ register regexp *prog, char *stringarg, register char * ch = SvPVX_const(do_utf8 ? prog->anchored_utf8 : prog->anchored_substr)[0]; if (do_utf8) { - while (s < strend) { + REXEC_FBC_SCAN( if (*s == ch) { DEBUG_EXECUTE_r( did_match = 1 ); if (regtry(®info, s)) goto got_it; @@ -1967,11 +1810,10 @@ Perl_regexec_flags(pTHX_ register regexp *prog, char *stringarg, register char * while (s < strend && *s == ch) s += UTF8SKIP(s); } - s += UTF8SKIP(s); - } + ); } else { - while (s < strend) { + REXEC_FBC_SCAN( if (*s == ch) { DEBUG_EXECUTE_r( did_match = 1 ); if (regtry(®info, s)) goto got_it; @@ -1979,8 +1821,7 @@ Perl_regexec_flags(pTHX_ register regexp *prog, char *stringarg, register char * while (s < strend && *s == ch) s++; } - s++; - } + ); } DEBUG_EXECUTE_r(if (!did_match) PerlIO_printf(Perl_debug_log, @@ -2011,14 +1852,19 @@ Perl_regexec_flags(pTHX_ register regexp *prog, char *stringarg, register char * back_max = prog->float_max_offset; back_min = prog->float_min_offset; } + + if (must == &PL_sv_undef) /* could not downgrade utf8 check substring, so must fail */ goto phooey; - last = HOP3c(strend, /* Cannot start after this */ - -(I32)(CHR_SVLEN(must) - - (SvTAIL(must) != 0) + back_min), strbeg); - + if (back_min<0) { + last = strend; + } else { + last = HOP3c(strend, /* Cannot start after this */ + -(I32)(CHR_SVLEN(must) + - (SvTAIL(must) != 0) + back_min), strbeg); + } if (s > PL_bostr) last1 = HOPc(s, -1); else @@ -2031,9 +1877,9 @@ Perl_regexec_flags(pTHX_ register regexp *prog, char *stringarg, register char * strend = HOPc(strend, -dontbother); while ( (s <= last) && ((flags & REXEC_SCREAM) - ? (s = screaminstr(sv, must, HOP3c(s, back_min, strend) - strbeg, + ? (s = screaminstr(sv, must, HOP3c(s, back_min, (back_min<0 ? strbeg : strend)) - strbeg, end_shift, &scream_pos, 0)) - : (s = fbm_instr((unsigned char*)HOP3(s, back_min, strend), + : (s = fbm_instr((unsigned char*)HOP3(s, back_min, (back_min<0 ? strbeg : strend)), (unsigned char*)strend, must, multiline ? FBMrf_MULTILINE : 0))) ) { /* we may be pointing at the wrong string */ @@ -2065,45 +1911,34 @@ Perl_regexec_flags(pTHX_ register regexp *prog, char *stringarg, register char * } } } - DEBUG_EXECUTE_r(if (!did_match) - PerlIO_printf(Perl_debug_log, - "Did not find %s substr \"%s%.*s%s\"%s...\n", + DEBUG_EXECUTE_r(if (!did_match) { + RE_PV_QUOTED_DECL(quoted, do_utf8, PERL_DEBUG_PAD_ZERO(0), + SvPVX_const(must), RE_SV_DUMPLEN(must), 30); + PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n", ((must == prog->anchored_substr || must == prog->anchored_utf8) ? "anchored" : "floating"), - PL_colors[0], - (int)(SvCUR(must) - (SvTAIL(must)!=0)), - SvPVX_const(must), - PL_colors[1], (SvTAIL(must) ? "$" : "")) - ); + quoted, RE_SV_TAIL(must)); + }); goto phooey; } - else if ((c = prog->regstclass)) { + else if ( (c = prog->regstclass) ) { if (minlen) { const OPCODE op = OP(prog->regstclass); /* don't bother with what can't match */ - if (PL_regkind[op] != EXACT && op != CANY && op != TRIE) + if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE) strend = HOPc(strend, -(minlen - 1)); } DEBUG_EXECUTE_r({ SV * const prop = sv_newmortal(); - const char *s0; - const char *s1; - int len0; - int len1; - regprop(prog, prop, c); - s0 = UTF ? - pv_uni_display(dsv0, (U8*)SvPVX_const(prop), SvCUR(prop), 60, - UNI_DISPLAY_REGEX) : - SvPVX_const(prop); - len0 = UTF ? SvCUR(dsv0) : SvCUR(prop); - s1 = UTF ? - sv_uni_display(dsv1, sv, 60, UNI_DISPLAY_REGEX) : s; - len1 = UTF ? (int)SvCUR(dsv1) : strend - s; - PerlIO_printf(Perl_debug_log, - "Matching stclass \"%*.*s\" against \"%*.*s\" (%d chars)\n", - len0, len0, s0, - len1, len1, s1, (int)(strend - s)); + { + RE_PV_QUOTED_DECL(quoted,UTF,PERL_DEBUG_PAD_ZERO(1), + s,strend-s,60); + PerlIO_printf(Perl_debug_log, + "Matching stclass %.*s against %s (%d chars)\n", + (int)SvCUR(prop), SvPVX_const(prop), + quoted, (int)(strend - s)); + } }); if (find_byclass(prog, c, s, strend, ®info)) goto got_it; @@ -2150,9 +1985,10 @@ Perl_regexec_flags(pTHX_ register regexp *prog, char *stringarg, register char * } } if (last == NULL) { - DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, - "%sCan't trim the tail, match fails (should not happen)%s\n", - PL_colors[4], PL_colors[5])); + DEBUG_EXECUTE_r( + PerlIO_printf(Perl_debug_log, + "%sCan't trim the tail, match fails (should not happen)%s\n", + PL_colors[4], PL_colors[5])); goto phooey; /* Should not happen! */ } dontbother = strend - last + prog->float_min_offset; @@ -2192,6 +2028,8 @@ got_it: the same. */ restore_pos(aTHX_ prog); } + if (prog->paren_names) + (void)hv_iterinit(prog->paren_names); /* make sure $`, $&, $', and $digit will work later */ if ( !(flags & REXEC_NOT_FIRST) ) { @@ -2234,6 +2072,7 @@ phooey: return 0; } + /* - regtry - try match at specific point */ @@ -2247,9 +2086,6 @@ S_regtry(pTHX_ const regmatch_info *reginfo, char *startpos) regexp *prog = reginfo->prog; GET_RE_DEBUG_FLAGS_DECL; -#ifdef DEBUGGING - PL_regindent = 0; /* XXXX Not good when matches are reenterable... */ -#endif if ((prog->reganch & ROPT_EVAL_SEEN) && !PL_reg_eval_set) { MAGIC *mg; @@ -2258,7 +2094,7 @@ S_regtry(pTHX_ const regmatch_info *reginfo, char *startpos) PerlIO_printf(Perl_debug_log, " setting stack tmpbase at %"IVdf"\n", (IV)(PL_stack_sp - PL_stack_base)); )); - SAVEI32(cxstack[cxstack_ix].blk_oldsp); + SAVESTACK_CXPOS(); cxstack[cxstack_ix].blk_oldsp = PL_stack_sp - PL_stack_base; /* Otherwise OP_NEXTSTATE will free whatever on stack now. */ SAVETMPS; @@ -2320,16 +2156,16 @@ S_regtry(pTHX_ const regmatch_info *reginfo, char *startpos) prog->subbeg = PL_bostr; prog->sublen = PL_regeol - PL_bostr; /* strend may have been modified */ } + DEBUG_EXECUTE_r(PL_reg_starttry = startpos); prog->startp[0] = startpos - PL_bostr; PL_reginput = startpos; - PL_regstartp = prog->startp; - PL_regendp = prog->endp; PL_reglastparen = &prog->lastparen; PL_reglastcloseparen = &prog->lastcloseparen; prog->lastparen = 0; prog->lastcloseparen = 0; PL_regsize = 0; - DEBUG_EXECUTE_r(PL_reg_starttry = startpos); + PL_regstartp = prog->startp; + PL_regendp = prog->endp; if (PL_reg_start_tmpl <= prog->nparens) { PL_reg_start_tmpl = prog->nparens*3/2 + 3; if(PL_reg_start_tmp) @@ -2372,107 +2208,30 @@ S_regtry(pTHX_ const regmatch_info *reginfo, char *startpos) return 0; } -#define RE_UNWIND_BRANCH 1 -#define RE_UNWIND_BRANCHJ 2 - -union re_unwind_t; - -typedef struct { /* XX: makes sense to enlarge it... */ - I32 type; - I32 prev; - CHECKPOINT lastcp; -} re_unwind_generic_t; - -typedef struct { - I32 type; - I32 prev; - CHECKPOINT lastcp; - I32 lastparen; - regnode *next; - char *locinput; - I32 nextchr; - int minmod; -#ifdef DEBUGGING - int regindent; -#endif -} re_unwind_branch_t; - -typedef union re_unwind_t { - I32 type; - re_unwind_generic_t generic; - re_unwind_branch_t branch; -} re_unwind_t; #define sayYES goto yes #define sayNO goto no -#define sayNO_ANYOF goto no_anyof -#define sayYES_FINAL goto yes_final -#define sayNO_FINAL goto no_final -#define sayNO_SILENT goto do_no -#define saySAME(x) if (x) goto yes; else goto no - -#define POSCACHE_SUCCESS 0 /* caching success rather than failure */ -#define POSCACHE_SEEN 1 /* we know what we're caching */ -#define POSCACHE_START 2 /* the real cache: this bit maps to pos 0 */ - -#define CACHEsayYES STMT_START { \ - if (st->u.whilem.cache_offset | st->u.whilem.cache_bit) { \ - if (!(PL_reg_poscache[0] & (1<u.whilem.cache_offset] |= (1<u.whilem.cache_bit); \ - } \ - else if (PL_reg_poscache[0] & (1<u.whilem.cache_offset] |= (1<u.whilem.cache_bit); \ - } \ - else { \ - /* cache records failure, but this is success */ \ - DEBUG_r( \ - PerlIO_printf(Perl_debug_log, \ - "%*s (remove success from failure cache)\n", \ - REPORT_CODE_OFF+PL_regindent*2, "") \ - ); \ - PL_reg_poscache[st->u.whilem.cache_offset] &= ~(1<u.whilem.cache_bit); \ - } \ - } \ - sayYES; \ -} STMT_END +#define sayNO_SILENT goto no_silent -#define CACHEsayNO STMT_START { \ - if (st->u.whilem.cache_offset | st->u.whilem.cache_bit) { \ - if (!(PL_reg_poscache[0] & (1<u.whilem.cache_offset] |= (1<u.whilem.cache_bit); \ - } \ - else if (!(PL_reg_poscache[0] & (1<u.whilem.cache_offset] |= (1<u.whilem.cache_bit); \ - } \ - else { \ - /* cache records success, but this is failure */ \ - DEBUG_r( \ - PerlIO_printf(Perl_debug_log, \ - "%*s (remove failure from success cache)\n", \ - REPORT_CODE_OFF+PL_regindent*2, "") \ - ); \ - PL_reg_poscache[st->u.whilem.cache_offset] &= ~(1<u.whilem.cache_bit); \ - } \ - } \ - sayNO; \ -} STMT_END +/* we dont use STMT_START/END here because it leads to + "unreachable code" warnings, which are bogus, but distracting. */ +#define CACHEsayNO \ + if (ST.cache_mask) \ + PL_reg_poscache[ST.cache_offset] |= ST.cache_mask; \ + sayNO /* this is used to determine how far from the left messages like - 'failed...' are printed. Currently 29 makes these messages line - up with the opcode they refer to. Earlier perls used 25 which - left these messages outdented making reviewing a debug output - quite difficult. + 'failed...' are printed. It should be set such that messages + are inline with the regop output that created them. */ -#define REPORT_CODE_OFF 29 +#define REPORT_CODE_OFF 32 /* Make sure there is a test for this +1 options in re_tests */ #define TRIE_INITAL_ACCEPT_BUFFLEN 4; -/* this value indiciates that the c1/c2 "next char" test should be skipped */ -#define CHRTEST_VOID -1000 +#define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */ +#define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */ #define SLAB_FIRST(s) (&(s)->states[0]) #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1]) @@ -2496,183 +2255,225 @@ S_push_slab(pTHX) return SLAB_FIRST(s); } -/* simulate a recursive call to regmatch */ - -#define REGMATCH(ns, where) \ - st->scan = scan; \ - scan = (ns); \ - st->resume_state = resume_##where; \ - goto start_recurse; \ - resume_point_##where: - - -/* push a new regex state. Set newst to point to it */ - -#define PUSH_STATE(newst, resume) \ - depth++; \ - DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "PUSH STATE(%d)\n", depth)); \ - st->scan = scan; \ - st->next = next; \ - st->n = n; \ - st->locinput = locinput; \ - st->resume_state = resume; \ - newst = st+1; \ - if (newst > SLAB_LAST(PL_regmatch_slab)) \ - newst = S_push_slab(aTHX); \ - PL_regmatch_state = newst; \ - newst->cc = 0; \ - newst->minmod = 0; \ - newst->sw = 0; \ - newst->logical = 0; \ - newst->unwind = 0; \ - locinput = PL_reginput; \ - nextchr = UCHARAT(locinput); - -#define POP_STATE \ - DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "POP STATE(%d)\n", depth)); \ - depth--; \ - st--; \ - if (st < SLAB_FIRST(PL_regmatch_slab)) { \ - PL_regmatch_slab = PL_regmatch_slab->prev; \ - st = SLAB_LAST(PL_regmatch_slab); \ - } \ - PL_regmatch_state = st; \ - scan = st->scan; \ - next = st->next; \ - n = st->n; \ - locinput = st->locinput; \ - nextchr = UCHARAT(locinput); + +/* push a new state then goto it */ + +#define PUSH_STATE_GOTO(state, node) \ + scan = node; \ + st->resume_state = state; \ + goto push_state; + +/* push a new state with success backtracking, then goto it */ + +#define PUSH_YES_STATE_GOTO(state, node) \ + scan = node; \ + st->resume_state = state; \ + goto push_yes_state; + + /* - - regmatch - main matching routine - * - * Conceptually the strategy is simple: check to see whether the current - * node matches, call self recursively to see whether the rest matches, - * and then act accordingly. In practice we make some effort to avoid - * recursion, in particular by going through "ordinary" nodes (that don't - * need to know whether the rest of the match failed) by a loop instead of - * by recursion. - */ -/* [lwall] I've hoisted the register declarations to the outer block in order to - * maybe save a little bit of pushing and popping on the stack. It also takes - * advantage of machines that use a register save mask on subroutine entry. - * - * This function used to be heavily recursive, but since this had the - * effect of blowing the CPU stack on complex regexes, it has been - * restructured to be iterative, and to save state onto the heap rather - * than the stack. Essentially whereever regmatch() used to be called, it - * pushes the current state, notes where to return, then jumps back into - * the main loop. - * - * Originally the structure of this function used to look something like - S_regmatch() { - int a = 1, b = 2; +regmatch() - main matching routine + +This is basically one big switch statement in a loop. We execute an op, +set 'next' to point the next op, and continue. If we come to a point which +we may need to backtrack to on failure such as (A|B|C), we push a +backtrack state onto the backtrack stack. On failure, we pop the top +state, and re-enter the loop at the state indicated. If there are no more +states to pop, we return failure. + +Sometimes we also need to backtrack on success; for example /A+/, where +after successfully matching one A, we need to go back and try to +match another one; similarly for lookahead assertions: if the assertion +completes successfully, we backtrack to the state just before the assertion +and then carry on. In these cases, the pushed state is marked as +'backtrack on success too'. This marking is in fact done by a chain of +pointers, each pointing to the previous 'yes' state. On success, we pop to +the nearest yes state, discarding any intermediate failure-only states. +Sometimes a yes state is pushed just to force some cleanup code to be +called at the end of a successful match or submatch; e.g. (??{$re}) uses +it to free the inner regex. + +Note that failure backtracking rewinds the cursor position, while +success backtracking leaves it alone. + +A pattern is complete when the END op is executed, while a subpattern +such as (?=foo) is complete when the SUCCESS op is executed. Both of these +ops trigger the "pop to last yes state if any, otherwise return true" +behaviour. + +A common convention in this function is to use A and B to refer to the two +subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is +the subpattern to be matched possibly multiple times, while B is the entire +rest of the pattern. Variable and state names reflect this convention. + +The states in the main switch are the union of ops and failure/success of +substates associated with with that op. For example, IFMATCH is the op +that does lookahead assertions /(?=A)B/ and so the IFMATCH state means +'execute IFMATCH'; while IFMATCH_A is a state saying that we have just +successfully matched A and IFMATCH_A_fail is a state saying that we have +just failed to match A. Resume states always come in pairs. The backtrack +state we push is marked as 'IFMATCH_A', but when that is popped, we resume +at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking +on success or failure. + +The struct that holds a backtracking state is actually a big union, with +one variant for each major type of op. The variable st points to the +top-most backtrack struct. To make the code clearer, within each +block of code we #define ST to alias the relevant union. + +Here's a concrete example of a (vastly oversimplified) IFMATCH +implementation: + + switch (state) { + .... + +#define ST st->u.ifmatch + + case IFMATCH: // we are executing the IFMATCH op, (?=A)B + ST.foo = ...; // some state we wish to save ... - while (scan != NULL) { - a++; // do stuff with a and b - ... - switch (OP(scan)) { - case FOO: { - int local = 3; - ... - if (regmatch(...)) // recurse - goto yes; - } - ... - } - } - yes: - return 1; - } + // push a yes backtrack state with a resume value of + // IFMATCH_A/IFMATCH_A_fail, then continue execution at the + // first node of A: + PUSH_YES_STATE_GOTO(IFMATCH_A, A); + // NOTREACHED + + case IFMATCH_A: // we have successfully executed A; now continue with B + next = B; + bar = ST.foo; // do something with the preserved value + break; + + case IFMATCH_A_fail: // A failed, so the assertion failed + ...; // do some housekeeping, then ... + sayNO; // propagate the failure - * Now it looks something like this: +#undef ST - typedef struct { - int a, b, local; - int resume_state; - } regmatch_state; + ... + } + +For any old-timers reading this who are familiar with the old recursive +approach, the code above is equivalent to: - S_regmatch() { - regmatch_state *st = new(); - int depth=0; - st->a++; // do stuff with a and b + case IFMATCH: // we are executing the IFMATCH op, (?=A)B + { + int foo = ... ... - while (scan != NULL) { - ... - switch (OP(scan)) { - case FOO: { - st->local = 3; - ... - st->scan = scan; - scan = ...; - st->resume_state = resume_FOO; - goto start_recurse; // recurse - - resume_point_FOO: - if (result) - goto yes; - } - ... - } - start_recurse: - st = new(); push a new state - st->a = 1; st->b = 2; - depth++; - } - yes: - result = 1; - if (depth--) { - st = pop(); - switch (resume_state) { - case resume_FOO: - goto resume_point_FOO; - ... - } + if (regmatch(A)) { + next = B; + bar = foo; + break; } - return result + ...; // do some housekeeping, then ... + sayNO; // propagate the failure } - - * WARNING: this means that any line in this function that contains a - * REGMATCH() or TRYPAREN() is actually simulating a recursive call to - * regmatch() using gotos instead. Thus the values of any local variables - * not saved in the regmatch_state structure will have been lost when - * execution resumes on the next line . - * - * States (ie the st pointer) are allocated in slabs of about 4K in size. - * PL_regmatch_state always points to the currently active state, and - * PL_regmatch_slab points to the slab currently containing PL_regmatch_state. - * The first time regmatch is called, the first slab is allocated, and is - * never freed until interpreter desctruction. When the slab is full, - * a new one is allocated chained to the end. At exit from regmatch, slabs - * allocated since entry are freed. - */ + +The topmost backtrack state, pointed to by st, is usually free. If you +want to claim it, populate any ST.foo fields in it with values you wish to +save, then do one of + + PUSH_STATE_GOTO(resume_state, node); + PUSH_YES_STATE_GOTO(resume_state, node); + +which sets that backtrack state's resume value to 'resume_state', pushes a +new free entry to the top of the backtrack stack, then goes to 'node'. +On backtracking, the free slot is popped, and the saved state becomes the +new free state. An ST.foo field in this new top state can be temporarily +accessed to retrieve values, but once the main loop is re-entered, it +becomes available for reuse. + +Note that the depth of the backtrack stack constantly increases during the +left-to-right execution of the pattern, rather than going up and down with +the pattern nesting. For example the stack is at its maximum at Z at the +end of the pattern, rather than at X in the following: + + /(((X)+)+)+....(Y)+....Z/ + +The only exceptions to this are lookahead/behind assertions and the cut, +(?>A), which pop all the backtrack states associated with A before +continuing. + +Bascktrack state structs are allocated in slabs of about 4K in size. +PL_regmatch_state and st always point to the currently active state, +and PL_regmatch_slab points to the slab currently containing +PL_regmatch_state. The first time regmatch() is called, the first slab is +allocated, and is never freed until interpreter destruction. When the slab +is full, a new one is allocated and chained to the end. At exit from +regmatch(), slabs allocated since entry are freed. + +*/ + +#define DEBUG_STATE_pp(pp) \ + DEBUG_STATE_r({ \ + DUMP_EXEC_POS(locinput, scan, do_utf8); \ + PerlIO_printf(Perl_debug_log, \ + " %*s"pp" %s\n", \ + depth*2, "", \ + reg_name[st->resume_state] ); \ + }); + + #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1) -#ifdef DEBUGGING -STATIC void -S_dump_exec_pos(pTHX_ const char *locinput, const regnode *scan, const bool do_utf8) +#ifdef DEBUGGING + +STATIC void +S_debug_start_match(pTHX_ const regexp *prog, const bool do_utf8, + const char *start, const char *end, const char *blurb) +{ + const bool utf8_pat= prog->reganch & ROPT_UTF8 ? 1 : 0; + if (!PL_colorset) + reginitcolors(); + { + RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0), + prog->precomp, prog->prelen, 60); + + RE_PV_QUOTED_DECL(s1, do_utf8, PERL_DEBUG_PAD_ZERO(1), + start, end - start, 60); + + PerlIO_printf(Perl_debug_log, + "%s%s REx%s %s against %s\n", + PL_colors[4], blurb, PL_colors[5], s0, s1); + + if (do_utf8||utf8_pat) + PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n", + utf8_pat ? "pattern" : "", + utf8_pat && do_utf8 ? " and " : "", + do_utf8 ? "string" : "" + ); + } +} + +STATIC void +S_dump_exec_pos(pTHX_ const char *locinput, + const regnode *scan, + const char *loc_regeol, + const char *loc_bostr, + const char *loc_reg_starttry, + const bool do_utf8) { - const int docolor = *PL_colors[0]; + const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4]; const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */ - int l = (PL_regeol - locinput) > taill ? taill : (PL_regeol - locinput); + int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput); /* The part of the string before starttry has one color (pref0_len chars), between starttry and current position another one (pref_len - pref0_len chars), after the current position the third one. We assume that pref0_len <= pref_len, otherwise we decrease pref0_len. */ - int pref_len = (locinput - PL_bostr) > (5 + taill) - l - ? (5 + taill) - l : locinput - PL_bostr; + int pref_len = (locinput - loc_bostr) > (5 + taill) - l + ? (5 + taill) - l : locinput - loc_bostr; int pref0_len; while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len))) pref_len++; - pref0_len = pref_len - (locinput - PL_reg_starttry); - if (l + pref_len < (5 + taill) && l < PL_regeol - locinput) - l = ( PL_regeol - locinput > (5 + taill) - pref_len - ? (5 + taill) - pref_len : PL_regeol - locinput); + pref0_len = pref_len - (locinput - loc_reg_starttry); + if (l + pref_len < (5 + taill) && l < loc_regeol - locinput) + l = ( loc_regeol - locinput > (5 + taill) - pref_len + ? (5 + taill) - pref_len : loc_regeol - locinput); while (do_utf8 && UTF8_IS_CONTINUATION(*(U8*)(locinput + l))) l--; if (pref0_len < 0) @@ -2680,42 +2481,57 @@ S_dump_exec_pos(pTHX_ const char *locinput, const regnode *scan, const bool do_u if (pref0_len > pref_len) pref0_len = pref_len; { - const char * const s0 = - do_utf8 && OP(scan) != CANY ? - pv_uni_display(PERL_DEBUG_PAD(0), (U8*)(locinput - pref_len), - pref0_len, 60, UNI_DISPLAY_REGEX) : - locinput - pref_len; - const int len0 = do_utf8 ? (int)strlen(s0) : pref0_len; - const char * const s1 = do_utf8 && OP(scan) != CANY ? - pv_uni_display(PERL_DEBUG_PAD(1), - (U8*)(locinput - pref_len + pref0_len), - pref_len - pref0_len, 60, UNI_DISPLAY_REGEX) : - locinput - pref_len + pref0_len; - const int len1 = do_utf8 ? (int)strlen(s1) : pref_len - pref0_len; - const char * const s2 = do_utf8 && OP(scan) != CANY ? - pv_uni_display(PERL_DEBUG_PAD(2), (U8*)locinput, - PL_regeol - locinput, 60, UNI_DISPLAY_REGEX) : - locinput; - const int len2 = do_utf8 ? (int)strlen(s2) : l; - PerlIO_printf(Perl_debug_log, - "%4"IVdf" <%s%.*s%s%s%.*s%s%s%s%.*s%s>%*s|", - (IV)(locinput - PL_bostr), - PL_colors[4], + const int is_uni = (do_utf8 && OP(scan) != CANY) ? 1 : 0; + + RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0), + (locinput - pref_len),pref0_len, 60, 4, 5); + + RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1), + (locinput - pref_len + pref0_len), + pref_len - pref0_len, 60, 2, 3); + + RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2), + locinput, loc_regeol - locinput, 10, 0, 1); + + const STRLEN tlen=len0+len1+len2; + PerlIO_printf(Perl_debug_log, + "%4"IVdf" <%.*s%.*s%s%.*s>%*s|", + (IV)(locinput - loc_bostr), len0, s0, - PL_colors[5], - PL_colors[2], len1, s1, - PL_colors[3], (docolor ? "" : "> <"), - PL_colors[0], len2, s2, - PL_colors[1], - 15 - l - pref_len + 1, + (int)(tlen > 19 ? 0 : 19 - tlen), ""); } } + #endif +/* reg_check_named_buff_matched() + * Checks to see if a named buffer has matched. The data array of + * buffer numbers corresponding to the buffer is expected to reside + * in the regexp->data->data array in the slot stored in the ARG() of + * node involved. Note that this routine doesn't actually care about the + * name, that information is not preserved from compilation to execution. + * Returns the index of the leftmost defined buffer with the given name + * or 0 if non of the buffers matched. + */ +STATIC I32 +S_reg_check_named_buff_matched(pTHX_ const regexp *rex, const regnode *scan) { + I32 n; + SV *sv_dat=(SV*)rex->data->data[ ARG( scan ) ]; + I32 *nums=(I32*)SvPVX(sv_dat); + for ( n=0; n= nums[n] && + PL_regendp[nums[n]] != -1) + { + return nums[n]; + } + } + return 0; +} + STATIC I32 /* 0 failure, 1 success */ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) { @@ -2737,22 +2553,37 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) /* cache heavy used fields of st in registers */ register regnode *scan; register regnode *next; - register I32 n = 0; /* initialize to shut up compiler warning */ + register I32 n = 0; /* general value; init to avoid compiler warning */ + register I32 ln = 0; /* len or last; init to avoid compiler warning */ register char *locinput = PL_reginput; - - /* these variables are NOT saved during a recusive RFEGMATCH: */ register I32 nextchr; /* is always set to UCHARAT(locinput) */ - bool result; /* return value of S_regmatch */ - regnode *inner; /* Next node in internal branch. */ - int depth = 0; /* depth of recursion */ - regmatch_state *newst; /* when pushing a state, this is the new one */ + + bool result = 0; /* return value of S_regmatch */ + int depth = 0; /* depth of backtrack stack */ + int nochange_depth = 0; /* depth of RECURSE recursion with nochange*/ regmatch_state *yes_state = NULL; /* state to pop to on success of subpattern */ - + regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */ + struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */ + U32 state_num; + + /* these three flags are set by various ops to signal information to + * the very next op. They have a useful lifetime of exactly one loop + * iteration, and are not preserved or restored by state pushes/pops + */ + bool sw = 0; /* the condition value in (?(cond)a|b) */ + bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */ + int logical = 0; /* the following EVAL is: + 0: (?{...}) + 1: (?(?{...})X|Y) + 2: (??{...}) + or the following IFMATCH/UNLESSM is: + false: plain (?=foo) + true: used as a condition: (?(?=foo)) + */ + #ifdef DEBUGGING - SV *re_debug_flags = NULL; - GET_RE_DEBUG_FLAGS; - PL_regindent++; + GET_RE_DEBUG_FLAGS_DECL; #endif /* on first ever call to regmatch, allocate first slab */ @@ -2773,11 +2604,6 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) if (st > SLAB_LAST(PL_regmatch_slab)) st = PL_regmatch_state = S_push_slab(aTHX); - st->minmod = 0; - st->sw = 0; - st->logical = 0; - st->unwind = 0; - st->cc = NULL; /* Note that nextchr is a byte even in UTF */ nextchr = UCHARAT(locinput); scan = prog; @@ -2785,21 +2611,25 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) DEBUG_EXECUTE_r( { SV * const prop = sv_newmortal(); - dump_exec_pos( locinput, scan, do_utf8 ); + regnode *rnext=regnext(scan); + DUMP_EXEC_POS( locinput, scan, do_utf8 ); regprop(rex, prop, scan); PerlIO_printf(Perl_debug_log, "%3"IVdf":%*s%s(%"IVdf")\n", - (IV)(scan - rex->program), PL_regindent*2, "", + (IV)(scan - rex->program), depth*2, "", SvPVX_const(prop), - PL_regkind[OP(scan)] == END ? 0 : (IV)(regnext(scan) - rex->program)); + (PL_regkind[OP(scan)] == END || !rnext) ? + 0 : (IV)(rnext - rex->program)); }); next = scan + NEXT_OFF(scan); if (next == scan) next = NULL; + state_num = OP(scan); - switch (OP(scan)) { + reenter_switch: + switch (state_num) { case BOL: if (locinput == PL_bostr) { @@ -2868,6 +2698,25 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) else nextchr = UCHARAT(++locinput); break; + +#undef ST +#define ST st->u.trie + case TRIEC: + /* In this case the charclass data is available inline so + we can fail fast without a lot of extra overhead. + */ + if (scan->flags == EXACT || !do_utf8) { + if(!ANYOF_BITMAP_TEST(scan, *locinput)) { + DEBUG_EXECUTE_r( + PerlIO_printf(Perl_debug_log, + "%*s %sfailed to match trie start class...%s\n", + REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]) + ); + sayNO_SILENT; + /* NOTREACHED */ + } + } + /* FALL THROUGH */ case TRIE: { /* what type of TRIE am I? (utf8 makes this contextual) */ @@ -2880,7 +2729,7 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) reg_trie_data * const trie = (reg_trie_data*)rex->data->data[ ARG( scan ) ]; U32 state = trie->startstate; - + if (trie->bitmap && trie_type != trie_utf8_fold && !TRIE_BITMAP_TEST(trie,*locinput) ) { @@ -2888,115 +2737,99 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, "%*s %smatched empty string...%s\n", - REPORT_CODE_OFF+PL_regindent*2, "", PL_colors[4], PL_colors[5]) + REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]) ); break; } else { DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, - "%*s %sfailed to match start class...%s\n", - REPORT_CODE_OFF+PL_regindent*2, "", PL_colors[4], PL_colors[5]) + "%*s %sfailed to match trie start class...%s\n", + REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]) ); sayNO_SILENT; } } - { - /* - traverse the TRIE keeping track of all accepting states - we transition through until we get to a failing node. - */ + { U8 *uc = ( U8* )locinput; - U16 charid = 0; - U32 base = 0; - UV uvc = 0; + STRLEN len = 0; STRLEN foldlen = 0; U8 *uscan = (U8*)NULL; STRLEN bufflen=0; SV *sv_accept_buff = NULL; + U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ]; + + ST.accepted = 0; /* how many accepting states we have seen */ + ST.B = next; + ST.jump = trie->jump; + +#ifdef DEBUGGING + ST.me = scan; +#endif + + - st->u.trie.accepted = 0; /* how many accepting states we have seen */ - result = 0; + /* + traverse the TRIE keeping track of all accepting states + we transition through until we get to a failing node. + */ while ( state && uc <= (U8*)PL_regeol ) { + U32 base = trie->states[ state ].trans.base; + UV uvc = 0; + U16 charid; + /* We use charid to hold the wordnum as we don't use it + for charid until after we have done the wordnum logic. + We define an alias just so that the wordnum logic reads + more naturally. */ - if (trie->states[ state ].wordnum) { - if (!st->u.trie.accepted ) { +#define got_wordnum charid + got_wordnum = trie->states[ state ].wordnum; + + if ( got_wordnum ) { + if ( ! ST.accepted ) { ENTER; SAVETMPS; bufflen = TRIE_INITAL_ACCEPT_BUFFLEN; sv_accept_buff=newSV(bufflen * sizeof(reg_trie_accepted) - 1); - SvCUR_set(sv_accept_buff, - sizeof(reg_trie_accepted)); + SvCUR_set(sv_accept_buff, 0); SvPOK_on(sv_accept_buff); sv_2mortal(sv_accept_buff); - st->u.trie.accept_buff = + SAVETMPS; + ST.accept_buff = (reg_trie_accepted*)SvPV_nolen(sv_accept_buff ); } - else { - if (st->u.trie.accepted >= bufflen) { + do { + if (ST.accepted >= bufflen) { bufflen *= 2; - st->u.trie.accept_buff =(reg_trie_accepted*) + ST.accept_buff =(reg_trie_accepted*) SvGROW(sv_accept_buff, bufflen * sizeof(reg_trie_accepted)); } SvCUR_set(sv_accept_buff,SvCUR(sv_accept_buff) + sizeof(reg_trie_accepted)); - } - st->u.trie.accept_buff[st->u.trie.accepted].wordnum = trie->states[state].wordnum; - st->u.trie.accept_buff[st->u.trie.accepted].endpos = uc; - ++st->u.trie.accepted; - } - base = trie->states[ state ].trans.base; + + ST.accept_buff[ST.accepted].wordnum = got_wordnum; + ST.accept_buff[ST.accepted].endpos = uc; + ++ST.accepted; + } while (trie->nextword && (got_wordnum= trie->nextword[got_wordnum])); + } +#undef got_wordnum DEBUG_TRIE_EXECUTE_r({ - dump_exec_pos( (char *)uc, scan, do_utf8 ); + DUMP_EXEC_POS( (char *)uc, scan, do_utf8 ); PerlIO_printf( Perl_debug_log, - "%*s %sState: %4"UVxf", Base: %4"UVxf", Accepted: %4"UVxf" ", - 2+PL_regindent * 2, "", PL_colors[4], - (UV)state, (UV)base, (UV)st->u.trie.accepted ); + "%*s %sState: %4"UVxf" Accepted: %4"UVxf" ", + 2+depth * 2, "", PL_colors[4], + (UV)state, (UV)ST.accepted ); }); if ( base ) { - switch (trie_type) { - case trie_utf8_fold: - if ( foldlen>0 ) { - uvc = utf8n_to_uvuni( uscan, UTF8_MAXLEN, &len, uniflags ); - foldlen -= len; - uscan += len; - len=0; - } else { - U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ]; - uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); - uvc = to_uni_fold( uvc, foldbuf, &foldlen ); - foldlen -= UNISKIP( uvc ); - uscan = foldbuf + UNISKIP( uvc ); - } - break; - case trie_utf8: - uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, - &len, uniflags ); - break; - case trie_plain: - uvc = (UV)*uc; - len = 1; - } - - if (uvc < 256) { - charid = trie->charmap[ uvc ]; - } - else { - charid = 0; - if (trie->widecharmap) { - SV** const svpp = hv_fetch(trie->widecharmap, - (char*)&uvc, sizeof(UV), 0); - if (svpp) - charid = (U16)SvIV(*svpp); - } - } + REXEC_TRIE_READ_CHAR(trie_type, trie, uc, uscan, len, + uvc, charid, foldlen, foldbuf, uniflags); if (charid && (base + charid > trie->uniquecharcount ) @@ -3019,116 +2852,136 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) } DEBUG_TRIE_EXECUTE_r( PerlIO_printf( Perl_debug_log, - "Charid:%3x CV:%4"UVxf" After State: %4"UVxf"%s\n", + "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n", charid, uvc, (UV)state, PL_colors[5] ); ); } - if (!st->u.trie.accepted ) + if (!ST.accepted ) sayNO; + DEBUG_EXECUTE_r( + PerlIO_printf( Perl_debug_log, + "%*s %sgot %"IVdf" possible matches%s\n", + REPORT_CODE_OFF + depth * 2, "", + PL_colors[4], (IV)ST.accepted, PL_colors[5] ); + ); + }} + + /* FALL THROUGH */ + + case TRIE_next_fail: /* we failed - try next alterative */ + + if ( ST.accepted == 1 ) { + /* only one choice left - just continue */ + DEBUG_EXECUTE_r({ + reg_trie_data * const trie + = (reg_trie_data*)rex->data->data[ ARG(ST.me) ]; + SV ** const tmp = RX_DEBUG(reginfo->prog) + ? av_fetch( trie->words, ST.accept_buff[ 0 ].wordnum-1, 0 ) + : NULL; + PerlIO_printf( Perl_debug_log, + "%*s %sonly one match left: #%d <%s>%s\n", + REPORT_CODE_OFF+depth*2, "", PL_colors[4], + ST.accept_buff[ 0 ].wordnum, + tmp ? SvPV_nolen_const( *tmp ) : "not compiled under -Dr", + PL_colors[5] ); + }); + PL_reginput = (char *)ST.accept_buff[ 0 ].endpos; + /* in this case we free tmps/leave before we call regmatch + as we wont be using accept_buff again. */ + FREETMPS; + LEAVE; + locinput = PL_reginput; + nextchr = UCHARAT(locinput); + + if ( !ST.jump ) + scan = ST.B; + else + scan = ST.B - ST.jump[ST.accept_buff[0].wordnum]; + + continue; /* execute rest of RE */ + } + + if (!ST.accepted-- ) { + FREETMPS; + LEAVE; + sayNO; + } + /* - There was at least one accepting state that we - transitioned through. Presumably the number of accepting - states is going to be low, typically one or two. So we - simply scan through to find the one with lowest wordnum. - Once we find it, we swap the last state into its place - and decrement the size. We then try to match the rest of - the pattern at the point where the word ends, if we - succeed then we end the loop, otherwise the loop - eventually terminates once all of the accepting states - have been tried. - */ + There are at least two accepting states left. Presumably + the number of accepting states is going to be low, + typically two. So we simply scan through to find the one + with lowest wordnum. Once we find it, we swap the last + state into its place and decrement the size. We then try to + match the rest of the pattern at the point where the word + ends. If we succeed, control just continues along the + regex; if we fail we return here to try the next accepting + state + */ - if ( st->u.trie.accepted == 1 ) { - DEBUG_EXECUTE_r({ - SV ** const tmp = RX_DEBUG(reginfo->prog) - ? av_fetch( trie->words, st->u.trie.accept_buff[ 0 ].wordnum-1, 0 ) - : NULL; + { + U32 best = 0; + U32 cur; + for( cur = 1 ; cur <= ST.accepted ; cur++ ) { + DEBUG_TRIE_EXECUTE_r( PerlIO_printf( Perl_debug_log, - "%*s %sonly one match : #%d <%s>%s\n", - REPORT_CODE_OFF+PL_regindent*2, "", PL_colors[4], - st->u.trie.accept_buff[ 0 ].wordnum, - tmp ? SvPV_nolen_const( *tmp ) : "not compiled under -Dr", - PL_colors[5] ); - }); - PL_reginput = (char *)st->u.trie.accept_buff[ 0 ].endpos; - /* in this case we free tmps/leave before we call regmatch - as we wont be using accept_buff again. */ - FREETMPS; - LEAVE; - /* do we need this? why dont we just do a break? */ - REGMATCH(scan + NEXT_OFF(scan), TRIE1); - /*** all unsaved local vars undefined at this point */ - } else { - DEBUG_EXECUTE_r( - PerlIO_printf( Perl_debug_log,"%*s %sgot %"IVdf" possible matches%s\n", - REPORT_CODE_OFF + PL_regindent * 2, "", PL_colors[4], (IV)st->u.trie.accepted, - PL_colors[5] ); - ); - while ( !result && st->u.trie.accepted-- ) { - U32 best = 0; - U32 cur; - for( cur = 1 ; cur <= st->u.trie.accepted ; cur++ ) { - DEBUG_TRIE_EXECUTE_r( - PerlIO_printf( Perl_debug_log, - "%*s %sgot %"IVdf" (%d) as best, looking at %"IVdf" (%d)%s\n", - REPORT_CODE_OFF + PL_regindent * 2, "", PL_colors[4], - (IV)best, st->u.trie.accept_buff[ best ].wordnum, (IV)cur, - st->u.trie.accept_buff[ cur ].wordnum, PL_colors[5] ); - ); - - if (st->u.trie.accept_buff[cur].wordnum < - st->u.trie.accept_buff[best].wordnum) - best = cur; - } - DEBUG_EXECUTE_r({ - reg_trie_data * const trie = (reg_trie_data*) - rex->data->data[ARG(scan)]; - SV ** const tmp = RX_DEBUG(reginfo->prog) - ? av_fetch( trie->words, st->u.trie.accept_buff[ best ].wordnum - 1, 0 ) - : NULL; - PerlIO_printf( Perl_debug_log, "%*s %strying alternation #%d <%s> at node #%d %s\n", - REPORT_CODE_OFF+PL_regindent*2, "", PL_colors[4], - st->u.trie.accept_buff[best].wordnum, - tmp ? SvPV_nolen_const( *tmp ) : "not compiled under -Dr", REG_NODE_NUM(scan), - PL_colors[5] ); - }); - if ( bestu.trie.accepted ) { - reg_trie_accepted tmp = st->u.trie.accept_buff[ best ]; - st->u.trie.accept_buff[ best ] = st->u.trie.accept_buff[ st->u.trie.accepted ]; - st->u.trie.accept_buff[ st->u.trie.accepted ] = tmp; - best = st->u.trie.accepted; - } - PL_reginput = (char *)st->u.trie.accept_buff[ best ].endpos; - - /* - as far as I can tell we only need the SAVETMPS/FREETMPS - for re's with EVAL in them but I'm leaving them in for - all until I can be sure. - */ - SAVETMPS; - REGMATCH(scan + NEXT_OFF(scan), TRIE2); - /*** all unsaved local vars undefined at this point */ - FREETMPS; - } - FREETMPS; - LEAVE; + "%*s %sgot %"IVdf" (%d) as best, looking at %"IVdf" (%d)%s\n", + REPORT_CODE_OFF + depth * 2, "", PL_colors[4], + (IV)best, ST.accept_buff[ best ].wordnum, (IV)cur, + ST.accept_buff[ cur ].wordnum, PL_colors[5] ); + ); + + if (ST.accept_buff[cur].wordnum < + ST.accept_buff[best].wordnum) + best = cur; } - - if (result) { - sayYES; - } else { - sayNO; + + DEBUG_EXECUTE_r({ + reg_trie_data * const trie + = (reg_trie_data*)rex->data->data[ ARG(ST.me) ]; + SV ** const tmp = RX_DEBUG(reginfo->prog) + ? av_fetch( trie->words, ST.accept_buff[ best ].wordnum - 1, 0 ) + : NULL; + regnode *nextop=!ST.jump ? + ST.B : + ST.B - ST.jump[ST.accept_buff[best].wordnum]; + PerlIO_printf( Perl_debug_log, + "%*s %strying alternation #%d <%s> at node #%d %s\n", + REPORT_CODE_OFF+depth*2, "", PL_colors[4], + ST.accept_buff[best].wordnum, + tmp ? SvPV_nolen_const( *tmp ) : "not compiled under -Dr", + REG_NODE_NUM(nextop), + PL_colors[5] ); + }); + + if ( bestln = STR_LEN(scan); + ln = STR_LEN(scan); if (do_utf8 != UTF) { /* The target and the pattern have differing utf8ness. */ char *l = locinput; - const char * const e = s + st->ln; + const char * const e = s + ln; if (do_utf8) { /* The target is utf8, the pattern is not utf8. */ @@ -3166,11 +3019,11 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) /* Inline the first character, for speed. */ if (UCHARAT(s) != nextchr) sayNO; - if (PL_regeol - locinput < st->ln) + if (PL_regeol - locinput < ln) sayNO; - if (st->ln > 1 && memNE(s, locinput, st->ln)) + if (ln > 1 && memNE(s, locinput, ln)) sayNO; - locinput += st->ln; + locinput += ln; nextchr = UCHARAT(locinput); break; } @@ -3179,14 +3032,14 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) /* FALL THROUGH */ case EXACTF: { char * const s = STRING(scan); - st->ln = STR_LEN(scan); + ln = STR_LEN(scan); if (do_utf8 || UTF) { /* Either target or the pattern are utf8. */ const char * const l = locinput; char *e = PL_regeol; - if (ibcmp_utf8(s, 0, st->ln, (bool)UTF, + if (ibcmp_utf8(s, 0, ln, (bool)UTF, l, &e, 0, do_utf8)) { /* One more case for the sharp s: * pack("U0U*", 0xDF) =~ /ss/i, @@ -3194,7 +3047,7 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) * byte sequence for the U+00DF. */ if (!(do_utf8 && toLOWER(s[0]) == 's' && - st->ln >= 2 && + ln >= 2 && toLOWER(s[1]) == 's' && (U8)l[0] == 0xC3 && e - l >= 2 && @@ -3213,13 +3066,13 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) UCHARAT(s) != ((OP(scan) == EXACTF) ? PL_fold : PL_fold_locale)[nextchr]) sayNO; - if (PL_regeol - locinput < st->ln) + if (PL_regeol - locinput < ln) sayNO; - if (st->ln > 1 && (OP(scan) == EXACTF - ? ibcmp(s, locinput, st->ln) - : ibcmp_locale(s, locinput, st->ln))) + if (ln > 1 && (OP(scan) == EXACTF + ? ibcmp(s, locinput, ln) + : ibcmp_locale(s, locinput, ln))) sayNO; - locinput += st->ln; + locinput += ln; nextchr = UCHARAT(locinput); break; } @@ -3228,7 +3081,7 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) STRLEN inclasslen = PL_regeol - locinput; if (!reginclass(rex, scan, (U8*)locinput, &inclasslen, do_utf8)) - sayNO_ANYOF; + goto anyof_fail; if (locinput >= PL_regeol) sayNO; locinput += inclasslen ? inclasslen : UTF8SKIP(locinput); @@ -3239,13 +3092,13 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) if (nextchr < 0) nextchr = UCHARAT(locinput); if (!REGINCLASS(rex, scan, (U8*)locinput)) - sayNO_ANYOF; + goto anyof_fail; if (!nextchr && locinput >= PL_regeol) sayNO; nextchr = UCHARAT(++locinput); break; } - no_anyof: + anyof_fail: /* If we might have the case of the German sharp s * in a casefolding Unicode character class. */ @@ -3311,35 +3164,35 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) /* was last char in word? */ if (do_utf8) { if (locinput == PL_bostr) - st->ln = '\n'; + ln = '\n'; else { const U8 * const r = reghop3((U8*)locinput, -1, (U8*)PL_bostr); - st->ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags); + ln = utf8n_to_uvchr(r, UTF8SKIP(r), 0, uniflags); } if (OP(scan) == BOUND || OP(scan) == NBOUND) { - st->ln = isALNUM_uni(st->ln); + ln = isALNUM_uni(ln); LOAD_UTF8_CHARCLASS_ALNUM(); n = swash_fetch(PL_utf8_alnum, (U8*)locinput, do_utf8); } else { - st->ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(st->ln)); + ln = isALNUM_LC_uvchr(UNI_TO_NATIVE(ln)); n = isALNUM_LC_utf8((U8*)locinput); } } else { - st->ln = (locinput != PL_bostr) ? + ln = (locinput != PL_bostr) ? UCHARAT(locinput - 1) : '\n'; if (OP(scan) == BOUND || OP(scan) == NBOUND) { - st->ln = isALNUM(st->ln); + ln = isALNUM(ln); n = isALNUM(nextchr); } else { - st->ln = isALNUM_LC(st->ln); + ln = isALNUM_LC(ln); n = isALNUM_LC(nextchr); } } - if (((!st->ln) == (!n)) == (OP(scan) == BOUND || + if (((!ln) == (!n)) == (OP(scan) == BOUND || OP(scan) == BOUNDL)) sayNO; break; @@ -3461,22 +3314,42 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) locinput++; nextchr = UCHARAT(locinput); break; + + case NREFFL: + { + char *s; + char type; + PL_reg_flags |= RF_tainted; + /* FALL THROUGH */ + case NREF: + case NREFF: + type = OP(scan); + n = reg_check_named_buff_matched(rex,scan); + + if ( n ) { + type = REF + ( type - NREF ); + goto do_ref; + } else { + sayNO; + } + /* unreached */ case REFFL: PL_reg_flags |= RF_tainted; /* FALL THROUGH */ case REF: - case REFF: { - char *s; + case REFF: n = ARG(scan); /* which paren pair */ - st->ln = PL_regstartp[n]; + type = OP(scan); + do_ref: + ln = PL_regstartp[n]; PL_reg_leftiter = PL_reg_maxiter; /* Void cache */ - if ((I32)*PL_reglastparen < n || st->ln == -1) + if ((I32)*PL_reglastparen < n || ln == -1) sayNO; /* Do not match unless seen CLOSEn. */ - if (st->ln == PL_regendp[n]) + if (ln == PL_regendp[n]) break; - s = PL_bostr + st->ln; - if (do_utf8 && OP(scan) != REF) { /* REF can do byte comparison */ + s = PL_bostr + ln; + if (do_utf8 && type != REF) { /* REF can do byte comparison */ char *l = locinput; const char *e = PL_bostr + PL_regendp[n]; /* @@ -3484,7 +3357,7 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) * in the 8-bit case (no pun intended) because in Unicode we * have to map both upper and title case to lower case. */ - if (OP(scan) == REFF) { + if (type == REFF) { while (s < e) { STRLEN ulen1, ulen2; U8 tmpbuf1[UTF8_MAXBYTES_CASE+1]; @@ -3507,32 +3380,64 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) /* Inline the first character, for speed. */ if (UCHARAT(s) != nextchr && - (OP(scan) == REF || - (UCHARAT(s) != ((OP(scan) == REFF - ? PL_fold : PL_fold_locale)[nextchr])))) + (type == REF || + (UCHARAT(s) != (type == REFF + ? PL_fold : PL_fold_locale)[nextchr]))) sayNO; - st->ln = PL_regendp[n] - st->ln; - if (locinput + st->ln > PL_regeol) + ln = PL_regendp[n] - ln; + if (locinput + ln > PL_regeol) sayNO; - if (st->ln > 1 && (OP(scan) == REF - ? memNE(s, locinput, st->ln) - : (OP(scan) == REFF - ? ibcmp(s, locinput, st->ln) - : ibcmp_locale(s, locinput, st->ln)))) + if (ln > 1 && (type == REF + ? memNE(s, locinput, ln) + : (type == REFF + ? ibcmp(s, locinput, ln) + : ibcmp_locale(s, locinput, ln)))) sayNO; - locinput += st->ln; + locinput += ln; nextchr = UCHARAT(locinput); break; - } - + } case NOTHING: case TAIL: break; case BACK: break; - case EVAL: + +#undef ST +#define ST st->u.eval { SV *ret; + regexp *re; + regnode *startpoint; + + case SRECURSE: + case RECURSE: /* /(...(?1))/ */ + if (cur_eval && cur_eval->locinput==locinput) { + if (cur_eval->u.eval.close_paren == ARG(scan)) + Perl_croak(aTHX_ "Infinite recursion in RECURSE in regexp"); + if ( ++nochange_depth > MAX_RECURSE_EVAL_NOCHANGE_DEPTH ) + Perl_croak(aTHX_ "RECURSE without pos change exceeded limit in regexp"); + } else { + nochange_depth = 0; + } + re = rex; + (void)ReREFCNT_inc(rex); + if (OP(scan)==RECURSE) { + startpoint = scan + ARG2L(scan); + ST.close_paren = ARG(scan); + } else { + startpoint = re->program+1; + ST.close_paren = 0; + } + goto eval_recurse_doit; + /* NOTREACHED */ + case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */ + if (cur_eval && cur_eval->locinput==locinput) { + if ( ++nochange_depth > MAX_RECURSE_EVAL_NOCHANGE_DEPTH ) + Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regexp"); + } else { + nochange_depth = 0; + } { /* execute the code in the {...} */ dSP; @@ -3559,14 +3464,14 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) PL_op = oop; PAD_RESTORE_LOCAL(old_comppad); PL_curcop = ocurcop; - if (!st->logical) { + if (!logical) { /* /(?{...})/ */ sv_setsv(save_scalar(PL_replgv), ret); break; } } - if (st->logical == 2) { /* Postponed subexpression: /(??{...})/ */ - regexp *re; + if (logical == 2) { /* Postponed subexpression: /(??{...})/ */ + logical = 0; { /* extract RE object from returned value; compiling if * necessary */ @@ -3594,7 +3499,7 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) Zero(&pm, 1, PMOP); if (DO_UTF8(ret)) pm.op_pmdynflags |= PMdf_DYN_UTF8; - re = CALLREGCOMP(aTHX_ (char*)t, (char*)t + len, &pm); + re = CALLREGCOMP((char*)t, (char*)t + len, &pm); if (!(SvFLAGS(ret) & (SVs_TEMP | SVs_PADTMP | SVf_READONLY | SVs_GMG))) @@ -3603,50 +3508,89 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) PL_regsize = osize; } } - + DEBUG_EXECUTE_r( + debug_start_match(re, do_utf8, locinput, PL_regeol, + "Matching embedded"); + ); + startpoint = re->program + 1; + ST.close_paren = 0; /* only used for RECURSE */ + /* borrowed from regtry */ + if (PL_reg_start_tmpl <= re->nparens) { + PL_reg_start_tmpl = re->nparens*3/2 + 3; + if(PL_reg_start_tmp) + Renew(PL_reg_start_tmp, PL_reg_start_tmpl, char*); + else + Newx(PL_reg_start_tmp, PL_reg_start_tmpl, char*); + } + + eval_recurse_doit: /* Share code with RECURSE below this line */ /* run the pattern returned from (??{...}) */ - - DEBUG_EXECUTE_r( - PerlIO_printf(Perl_debug_log, - "Entering embedded \"%s%.60s%s%s\"\n", - PL_colors[0], - re->precomp, - PL_colors[1], - (strlen(re->precomp) > 60 ? "..." : "")) - ); - - st->u.eval.cp = regcppush(0); /* Save *all* the positions. */ - REGCP_SET(st->u.eval.lastcp); + ST.cp = regcppush(0); /* Save *all* the positions. */ + REGCP_SET(ST.lastcp); + + PL_regstartp = re->startp; /* essentially NOOP on RECURSE */ + PL_regendp = re->endp; /* essentially NOOP on RECURSE */ + *PL_reglastparen = 0; *PL_reglastcloseparen = 0; PL_reginput = locinput; + PL_regsize = 0; /* XXXX This is too dramatic a measure... */ PL_reg_maxiter = 0; - st->logical = 0; - st->u.eval.toggleutf = ((PL_reg_flags & RF_utf8) != 0) ^ - ((re->reganch & ROPT_UTF8) != 0); - if (st->u.eval.toggleutf) PL_reg_flags ^= RF_utf8; - st->u.eval.prev_rex = rex; - rex = re; - - /* resume to current state on success */ - st->u.yes.prev_yes_state = yes_state; - yes_state = st; - PUSH_STATE(newst, resume_EVAL); - st = newst; + ST.toggle_reg_flags = PL_reg_flags; + if (re->reganch & ROPT_UTF8) + PL_reg_flags |= RF_utf8; + else + PL_reg_flags &= ~RF_utf8; + ST.toggle_reg_flags ^= PL_reg_flags; /* diff of old and new */ - /* now continue from first node in postoned RE */ - next = re->program + 1; - break; + ST.prev_rex = rex; + ST.prev_curlyx = cur_curlyx; + rex = re; + cur_curlyx = NULL; + ST.B = next; + ST.prev_eval = cur_eval; + cur_eval = st; + /* now continue from first node in postoned RE */ + PUSH_YES_STATE_GOTO(EVAL_AB, startpoint); /* NOTREACHED */ } - /* /(?(?{...})X|Y)/ */ - st->sw = SvTRUE(ret); - st->logical = 0; + /* logical is 1, /(?(?{...})X|Y)/ */ + sw = (bool)SvTRUE(ret); + logical = 0; break; } + + case EVAL_AB: /* cleanup after a successful (??{A})B */ + /* note: this is called twice; first after popping B, then A */ + PL_reg_flags ^= ST.toggle_reg_flags; + ReREFCNT_dec(rex); + rex = ST.prev_rex; + regcpblow(ST.cp); + cur_eval = ST.prev_eval; + cur_curlyx = ST.prev_curlyx; + /* XXXX This is too dramatic a measure... */ + PL_reg_maxiter = 0; + sayYES; + + + case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */ + /* note: this is called twice; first after popping B, then A */ + PL_reg_flags ^= ST.toggle_reg_flags; + ReREFCNT_dec(rex); + rex = ST.prev_rex; + PL_reginput = locinput; + REGCP_UNWIND(ST.lastcp); + regcppop(rex); + cur_eval = ST.prev_eval; + cur_curlyx = ST.prev_curlyx; + /* XXXX This is too dramatic a measure... */ + PL_reg_maxiter = 0; + sayNO_SILENT; +#undef ST + case OPEN: n = ARG(scan); /* which paren pair */ PL_reg_start_tmp[n] = locinput; @@ -3660,14 +3604,28 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) if (n > (I32)*PL_reglastparen) *PL_reglastparen = n; *PL_reglastcloseparen = n; + if (cur_eval && cur_eval->u.eval.close_paren == (U32)n) { + goto fake_end; + } break; case GROUPP: n = ARG(scan); /* which paren pair */ - st->sw = ((I32)*PL_reglastparen >= n && PL_regendp[n] != -1); + sw = (bool)((I32)*PL_reglastparen >= n && PL_regendp[n] != -1); break; + case NGROUPP: + /* reg_check_named_buff_matched returns 0 for no match */ + sw = (bool)(0 < reg_check_named_buff_matched(rex,scan)); + break; + case RECURSEP: + n = ARG(scan); + sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n)); + break; + case DEFINEP: + sw = 0; + break; case IFTHEN: PL_reg_leftiter = PL_reg_maxiter; /* Void cache */ - if (st->sw) + if (sw) next = NEXTOPER(NEXTOPER(scan)); else { next = scan + ARG(scan); @@ -3676,214 +3634,199 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) } break; case LOGICAL: - st->logical = scan->flags; + logical = scan->flags; break; + /******************************************************************* - cc points to the regmatch_state associated with the most recent CURLYX. - This struct contains info about the innermost (...)* loop (an - "infoblock"), and a pointer to the next outer cc. - - Here is how Y(A)*Z is processed (if it is compiled into CURLYX/WHILEM): - - 1) After matching Y, regnode for CURLYX is processed; - - 2) This regnode populates cc, and calls regmatch() recursively - with the starting point at WHILEM node; - - 3) Each hit of WHILEM node tries to match A and Z (in the order - depending on the current iteration, min/max of {min,max} and - greediness). The information about where are nodes for "A" - and "Z" is read from cc, as is info on how many times "A" - was already matched, and greediness. - - 4) After A matches, the same WHILEM node is hit again. - - 5) Each time WHILEM is hit, cc is the infoblock created by CURLYX - of the same pair. Thus when WHILEM tries to match Z, it temporarily - resets cc, since this Y(A)*Z can be a part of some other loop: - as in (Y(A)*Z)*. If Z matches, the automaton will hit the WHILEM node - of the external loop. - - Currently present infoblocks form a tree with a stem formed by st->cc - and whatever it mentions via ->next, and additional attached trees - corresponding to temporarily unset infoblocks as in "5" above. - - In the following picture, infoblocks for outer loop of - (Y(A)*?Z)*?T are denoted O, for inner I. NULL starting block - is denoted by x. The matched string is YAAZYAZT. Temporarily postponed - infoblocks are drawn below the "reset" infoblock. - - In fact in the picture below we do not show failed matches for Z and T - by WHILEM blocks. [We illustrate minimal matches, since for them it is - more obvious *why* one needs to *temporary* unset infoblocks.] - - Matched REx position InfoBlocks Comment - (Y(A)*?Z)*?T x - Y(A)*?Z)*?T x <- O - Y (A)*?Z)*?T x <- O - Y A)*?Z)*?T x <- O <- I - YA )*?Z)*?T x <- O <- I - YA A)*?Z)*?T x <- O <- I - YAA )*?Z)*?T x <- O <- I - YAA Z)*?T x <- O # Temporary unset I - I - - YAAZ Y(A)*?Z)*?T x <- O - I - - YAAZY (A)*?Z)*?T x <- O - I - - YAAZY A)*?Z)*?T x <- O <- I - I - - YAAZYA )*?Z)*?T x <- O <- I - I - - YAAZYA Z)*?T x <- O # Temporary unset I - I,I - - YAAZYAZ )*?T x <- O - I,I - - YAAZYAZ T x # Temporary unset O - O - I,I - - YAAZYAZT x - O - I,I - *******************************************************************/ - - case CURLYX: { - /* No need to save/restore up to this paren */ - I32 parenfloor = scan->flags; - - /* Dave says: - - CURLYX and WHILEM are always paired: they're the moral - equivalent of pp_enteriter anbd pp_iter. - - The only time next could be null is if the node tree is - corrupt. This was mentioned on p5p a few days ago. - - See http://www.xray.mpe.mpg.de/mailing-lists/perl5-porters/2006-04/msg00556.html - So we'll assert that this is true: - */ - assert(next); - if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */ - next += ARG(next); - /* XXXX Probably it is better to teach regpush to support - parenfloor > PL_regsize... */ - if (parenfloor > (I32)*PL_reglastparen) - parenfloor = *PL_reglastparen; /* Pessimization... */ - - st->u.curlyx.cp = PL_savestack_ix; - st->u.curlyx.outercc = st->cc; - st->cc = st; - /* these fields contain the state of the current curly. - * they are accessed by subsequent WHILEMs; - * cur and lastloc are also updated by WHILEM */ - st->u.curlyx.parenfloor = parenfloor; - st->u.curlyx.cur = -1; /* this will be updated by WHILEM */ - st->u.curlyx.min = ARG1(scan); - st->u.curlyx.max = ARG2(scan); - st->u.curlyx.scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS; - st->u.curlyx.lastloc = 0; - /* st->next and st->minmod are also read by WHILEM */ - PL_reginput = locinput; - REGMATCH(PREVOPER(next), CURLYX); /* start on the WHILEM */ - /*** all unsaved local vars undefined at this point */ - regcpblow(st->u.curlyx.cp); - st->cc = st->u.curlyx.outercc; - saySAME(result); - } +The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/ +pattern, where A and B are subpatterns. (For simple A, CURLYM or +STAR/PLUS/CURLY/CURLYN are used instead.) + +A*B is compiled as + +On entry to the subpattern, CURLYX is called. This pushes a CURLYX +state, which contains the current count, initialised to -1. It also sets +cur_curlyx to point to this state, with any previous value saved in the +state block. + +CURLYX then jumps straight to the WHILEM op, rather than executing A, +since the pattern may possibly match zero times (i.e. it's a while {} loop +rather than a do {} while loop). + +Each entry to WHILEM represents a successful match of A. The count in the +CURLYX block is incremented, another WHILEM state is pushed, and execution +passes to A or B depending on greediness and the current count. + +For example, if matching against the string a1a2a3b (where the aN are +substrings that match /A/), then the match progresses as follows: (the +pushed states are interspersed with the bits of strings matched so far): + + + + a1 + a1 a2 + a1 a2 a3 + a1 a2 a3 b + +(Contrast this with something like CURLYM, which maintains only a single +backtrack state: + + a1 + a1 a2 + a1 a2 a3 + a1 a2 a3 b +) + +Each WHILEM state block marks a point to backtrack to upon partial failure +of A or B, and also contains some minor state data related to that +iteration. The CURLYX block, pointed to by cur_curlyx, contains the +overall state, such as the count, and pointers to the A and B ops. + +This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx +must always point to the *current* CURLYX block, the rules are: + +When executing CURLYX, save the old cur_curlyx in the CURLYX state block, +and set cur_curlyx to point the new block. + +When popping the CURLYX block after a successful or unsuccessful match, +restore the previous cur_curlyx. + +When WHILEM is about to execute B, save the current cur_curlyx, and set it +to the outer one saved in the CURLYX block. + +When popping the WHILEM block after a successful or unsuccessful B match, +restore the previous cur_curlyx. + +Here's an example for the pattern (AI* BI)*BO +I and O refer to inner and outer, C and W refer to CURLYX and WHILEM: + +cur_ +curlyx backtrack stack +------ --------------- +NULL +CO +CI ai +CO ai bi +NULL ai bi bo + +At this point the pattern succeeds, and we work back down the stack to +clean up, restoring as we go: + +CO ai bi +CI ai +CO +NULL + +*******************************************************************/ + +#define ST st->u.curlyx + + case CURLYX: /* start of /A*B/ (for complex A) */ + { + /* No need to save/restore up to this paren */ + I32 parenfloor = scan->flags; + + assert(next); /* keep Coverity happy */ + if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */ + next += ARG(next); + + /* XXXX Probably it is better to teach regpush to support + parenfloor > PL_regsize... */ + if (parenfloor > (I32)*PL_reglastparen) + parenfloor = *PL_reglastparen; /* Pessimization... */ + + ST.prev_curlyx= cur_curlyx; + cur_curlyx = st; + ST.cp = PL_savestack_ix; + + /* these fields contain the state of the current curly. + * they are accessed by subsequent WHILEMs */ + ST.parenfloor = parenfloor; + ST.min = ARG1(scan); + ST.max = ARG2(scan); + ST.A = NEXTOPER(scan) + EXTRA_STEP_2ARGS; + ST.B = next; + ST.minmod = minmod; + minmod = 0; + ST.count = -1; /* this will be updated by WHILEM */ + ST.lastloc = NULL; /* this will be updated by WHILEM */ + + PL_reginput = locinput; + PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next)); /* NOTREACHED */ - case WHILEM: { - /* - * This is really hard to understand, because after we match - * what we're trying to match, we must make sure the rest of - * the REx is going to match for sure, and to do that we have - * to go back UP the parse tree by recursing ever deeper. And - * if it fails, we have to reset our parent's current state - * that we can try again after backing off. - */ + } - /* Dave says: - - st->cc gets initialised by CURLYX ready for use by WHILEM. - So again, unless somethings been corrupted, st->cc cannot - be null at that point in WHILEM. - - See http://www.xray.mpe.mpg.de/mailing-lists/perl5-porters/2006-04/msg00556.html - So we'll assert that this is true: - */ - assert(st->cc); - st->u.whilem.lastloc = st->cc->u.curlyx.lastloc; /* Detection of 0-len. */ - st->u.whilem.cache_offset = 0; - st->u.whilem.cache_bit = 0; - - n = st->cc->u.curlyx.cur + 1; /* how many we know we matched */ - PL_reginput = locinput; + case CURLYX_end: /* just finished matching all of A*B */ + regcpblow(ST.cp); + cur_curlyx = ST.prev_curlyx; + sayYES; + /* NOTREACHED */ - DEBUG_EXECUTE_r( - PerlIO_printf(Perl_debug_log, - "%*s %ld out of %ld..%ld cc=%"UVxf"\n", - REPORT_CODE_OFF+PL_regindent*2, "", - (long)n, (long)st->cc->u.curlyx.min, - (long)st->cc->u.curlyx.max, PTR2UV(st->cc)) - ); + case CURLYX_end_fail: /* just failed to match all of A*B */ + regcpblow(ST.cp); + cur_curlyx = ST.prev_curlyx; + sayNO; + /* NOTREACHED */ - /* If degenerate scan matches "", assume scan done. */ - - if (locinput == st->cc->u.curlyx.lastloc && n >= st->cc->u.curlyx.min) { - st->u.whilem.savecc = st->cc; - st->cc = st->cc->u.curlyx.outercc; - if (st->cc) - st->ln = st->cc->u.curlyx.cur; - DEBUG_EXECUTE_r( - PerlIO_printf(Perl_debug_log, - "%*s empty match detected, try continuation...\n", - REPORT_CODE_OFF+PL_regindent*2, "") - ); - REGMATCH(st->u.whilem.savecc->next, WHILEM1); - /*** all unsaved local vars undefined at this point */ - st->cc = st->u.whilem.savecc; - if (result) - sayYES; - if (st->cc->u.curlyx.outercc) - st->cc->u.curlyx.outercc->u.curlyx.cur = st->ln; - sayNO; - } - /* First just match a string of min scans. */ - - if (n < st->cc->u.curlyx.min) { - st->cc->u.curlyx.cur = n; - st->cc->u.curlyx.lastloc = locinput; - REGMATCH(st->cc->u.curlyx.scan, WHILEM2); - /*** all unsaved local vars undefined at this point */ - if (result) - sayYES; - st->cc->u.curlyx.cur = n - 1; - st->cc->u.curlyx.lastloc = st->u.whilem.lastloc; - sayNO; - } +#undef ST +#define ST st->u.whilem + + case WHILEM: /* just matched an A in /A*B/ (for complex A) */ + { + /* see the discussion above about CURLYX/WHILEM */ + I32 n; + assert(cur_curlyx); /* keep Coverity happy */ + n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */ + ST.save_lastloc = cur_curlyx->u.curlyx.lastloc; + ST.cache_offset = 0; + ST.cache_mask = 0; + + PL_reginput = locinput; + + DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, + "%*s whilem: matched %ld out of %ld..%ld\n", + REPORT_CODE_OFF+depth*2, "", (long)n, + (long)cur_curlyx->u.curlyx.min, + (long)cur_curlyx->u.curlyx.max) + ); + + /* First just match a string of min A's. */ + + if (n < cur_curlyx->u.curlyx.min) { + cur_curlyx->u.curlyx.lastloc = locinput; + PUSH_STATE_GOTO(WHILEM_A_pre, cur_curlyx->u.curlyx.A); + /* NOTREACHED */ + } + + /* If degenerate A matches "", assume A done. */ + + if (locinput == cur_curlyx->u.curlyx.lastloc) { + DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, + "%*s whilem: empty match detected, trying continuation...\n", + REPORT_CODE_OFF+depth*2, "") + ); + goto do_whilem_B_max; + } + + /* super-linear cache processing */ + + if (scan->flags) { - if (scan->flags) { - /* Check whether we already were at this position. - Postpone detection until we know the match is not - *that* much linear. */ if (!PL_reg_maxiter) { + /* start the countdown: Postpone detection until we + * know the match is not *that* much linear. */ PL_reg_maxiter = (PL_regeol - PL_bostr + 1) * (scan->flags>>4); /* possible overflow for long strings and many CURLYX's */ if (PL_reg_maxiter < 0) PL_reg_maxiter = I32_MAX; PL_reg_leftiter = PL_reg_maxiter; } + if (PL_reg_leftiter-- == 0) { - const I32 size = (PL_reg_maxiter + 7 + POSCACHE_START)/8; + /* initialise cache */ + const I32 size = (PL_reg_maxiter + 7)/8; if (PL_reg_poscache) { if ((I32)PL_reg_poscache_size < size) { Renew(PL_reg_poscache, size, char); @@ -3895,369 +3838,396 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) PL_reg_poscache_size = size; Newxz(PL_reg_poscache, size, char); } - DEBUG_EXECUTE_r( - PerlIO_printf(Perl_debug_log, - "%sDetected a super-linear match, switching on caching%s...\n", - PL_colors[4], PL_colors[5]) - ); + DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, + "%swhilem: Detected a super-linear match, switching on caching%s...\n", + PL_colors[4], PL_colors[5]) + ); } + if (PL_reg_leftiter < 0) { - st->u.whilem.cache_offset = locinput - PL_bostr; - - st->u.whilem.cache_offset = (scan->flags & 0xf) - 1 + POSCACHE_START - + st->u.whilem.cache_offset * (scan->flags>>4); - st->u.whilem.cache_bit = st->u.whilem.cache_offset % 8; - st->u.whilem.cache_offset /= 8; - if (PL_reg_poscache[st->u.whilem.cache_offset] & (1<u.whilem.cache_bit)) { - DEBUG_EXECUTE_r( - PerlIO_printf(Perl_debug_log, - "%*s already tried at this position...\n", - REPORT_CODE_OFF+PL_regindent*2, "") + /* have we already failed at this position? */ + I32 offset, mask; + offset = (scan->flags & 0xf) - 1 + + (locinput - PL_bostr) * (scan->flags>>4); + mask = 1 << (offset % 8); + offset /= 8; + if (PL_reg_poscache[offset] & mask) { + DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, + "%*s whilem: (cache) already tried at this position...\n", + REPORT_CODE_OFF+depth*2, "") ); - if (PL_reg_poscache[0] & (1<cc->minmod) { - st->u.whilem.savecc = st->cc; - st->cc = st->cc->u.curlyx.outercc; - if (st->cc) - st->ln = st->cc->u.curlyx.cur; - st->u.whilem.cp = regcppush(st->u.whilem.savecc->u.curlyx.parenfloor); - REGCP_SET(st->u.whilem.lastcp); - REGMATCH(st->u.whilem.savecc->next, WHILEM3); - /*** all unsaved local vars undefined at this point */ - st->cc = st->u.whilem.savecc; - if (result) { - regcpblow(st->u.whilem.cp); - CACHEsayYES; /* All done. */ - } - REGCP_UNWIND(st->u.whilem.lastcp); - regcppop(rex); - if (st->cc->u.curlyx.outercc) - st->cc->u.curlyx.outercc->u.curlyx.cur = st->ln; - - if (n >= st->cc->u.curlyx.max) { /* Maximum greed exceeded? */ - if (ckWARN(WARN_REGEXP) && n >= REG_INFTY - && !(PL_reg_flags & RF_warned)) { - PL_reg_flags |= RF_warned; - Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded", - "Complex regular subexpression recursion", - REG_INFTY - 1); - } - CACHEsayNO; - } + /* Prefer B over A for minimal matching. */ - DEBUG_EXECUTE_r( - PerlIO_printf(Perl_debug_log, - "%*s trying longer...\n", - REPORT_CODE_OFF+PL_regindent*2, "") - ); - /* Try scanning more and see if it helps. */ - PL_reginput = locinput; - st->cc->u.curlyx.cur = n; - st->cc->u.curlyx.lastloc = locinput; - st->u.whilem.cp = regcppush(st->cc->u.curlyx.parenfloor); - REGCP_SET(st->u.whilem.lastcp); - REGMATCH(st->cc->u.curlyx.scan, WHILEM4); - /*** all unsaved local vars undefined at this point */ - if (result) { - regcpblow(st->u.whilem.cp); - CACHEsayYES; - } - REGCP_UNWIND(st->u.whilem.lastcp); - regcppop(rex); - st->cc->u.curlyx.cur = n - 1; - st->cc->u.curlyx.lastloc = st->u.whilem.lastloc; - CACHEsayNO; - } + if (cur_curlyx->u.curlyx.minmod) { + ST.save_curlyx = cur_curlyx; + cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx; + ST.cp = regcppush(ST.save_curlyx->u.curlyx.parenfloor); + REGCP_SET(ST.lastcp); + PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B); + /* NOTREACHED */ + } - /* Prefer scan over next for maximal matching. */ - - if (n < st->cc->u.curlyx.max) { /* More greed allowed? */ - st->u.whilem.cp = regcppush(st->cc->u.curlyx.parenfloor); - st->cc->u.curlyx.cur = n; - st->cc->u.curlyx.lastloc = locinput; - REGCP_SET(st->u.whilem.lastcp); - REGMATCH(st->cc->u.curlyx.scan, WHILEM5); - /*** all unsaved local vars undefined at this point */ - if (result) { - regcpblow(st->u.whilem.cp); - CACHEsayYES; - } - REGCP_UNWIND(st->u.whilem.lastcp); - regcppop(rex); /* Restore some previous $s? */ - PL_reginput = locinput; - DEBUG_EXECUTE_r( - PerlIO_printf(Perl_debug_log, - "%*s failed, try continuation...\n", - REPORT_CODE_OFF+PL_regindent*2, "") - ); - } - if (ckWARN(WARN_REGEXP) && n >= REG_INFTY - && !(PL_reg_flags & RF_warned)) { + /* Prefer A over B for maximal matching. */ + + if (n < cur_curlyx->u.curlyx.max) { /* More greed allowed? */ + ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor); + cur_curlyx->u.curlyx.lastloc = locinput; + REGCP_SET(ST.lastcp); + PUSH_STATE_GOTO(WHILEM_A_max, cur_curlyx->u.curlyx.A); + /* NOTREACHED */ + } + goto do_whilem_B_max; + } + /* NOTREACHED */ + + case WHILEM_B_min: /* just matched B in a minimal match */ + case WHILEM_B_max: /* just matched B in a maximal match */ + cur_curlyx = ST.save_curlyx; + sayYES; + /* NOTREACHED */ + + case WHILEM_B_max_fail: /* just failed to match B in a maximal match */ + cur_curlyx = ST.save_curlyx; + cur_curlyx->u.curlyx.lastloc = ST.save_lastloc; + cur_curlyx->u.curlyx.count--; + CACHEsayNO; + /* NOTREACHED */ + + case WHILEM_A_min_fail: /* just failed to match A in a minimal match */ + REGCP_UNWIND(ST.lastcp); + regcppop(rex); + /* FALL THROUGH */ + case WHILEM_A_pre_fail: /* just failed to match even minimal A */ + cur_curlyx->u.curlyx.lastloc = ST.save_lastloc; + cur_curlyx->u.curlyx.count--; + CACHEsayNO; + /* NOTREACHED */ + + case WHILEM_A_max_fail: /* just failed to match A in a maximal match */ + REGCP_UNWIND(ST.lastcp); + regcppop(rex); /* Restore some previous $s? */ + PL_reginput = locinput; + DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, + "%*s whilem: failed, trying continuation...\n", + REPORT_CODE_OFF+depth*2, "") + ); + do_whilem_B_max: + if (cur_curlyx->u.curlyx.count >= REG_INFTY + && ckWARN(WARN_REGEXP) + && !(PL_reg_flags & RF_warned)) + { + PL_reg_flags |= RF_warned; + Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded", + "Complex regular subexpression recursion", + REG_INFTY - 1); + } + + /* now try B */ + ST.save_curlyx = cur_curlyx; + cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx; + PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B); + /* NOTREACHED */ + + case WHILEM_B_min_fail: /* just failed to match B in a minimal match */ + cur_curlyx = ST.save_curlyx; + REGCP_UNWIND(ST.lastcp); + regcppop(rex); + + if (cur_curlyx->u.curlyx.count >= cur_curlyx->u.curlyx.max) { + /* Maximum greed exceeded */ + if (cur_curlyx->u.curlyx.count >= REG_INFTY + && ckWARN(WARN_REGEXP) + && !(PL_reg_flags & RF_warned)) + { PL_reg_flags |= RF_warned; - Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s limit (%d) exceeded", - "Complex regular subexpression recursion", - REG_INFTY - 1); + Perl_warner(aTHX_ packWARN(WARN_REGEXP), + "%s limit (%d) exceeded", + "Complex regular subexpression recursion", + REG_INFTY - 1); } - - /* Failed deeper matches of scan, so see if this one works. */ - st->u.whilem.savecc = st->cc; - st->cc = st->cc->u.curlyx.outercc; - if (st->cc) - st->ln = st->cc->u.curlyx.cur; - REGMATCH(st->u.whilem.savecc->next, WHILEM6); - /*** all unsaved local vars undefined at this point */ - st->cc = st->u.whilem.savecc; - if (result) - CACHEsayYES; - if (st->cc->u.curlyx.outercc) - st->cc->u.curlyx.outercc->u.curlyx.cur = st->ln; - st->cc->u.curlyx.cur = n - 1; - st->cc->u.curlyx.lastloc = st->u.whilem.lastloc; + cur_curlyx->u.curlyx.count--; CACHEsayNO; } + + DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, + "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "") + ); + /* Try grabbing another A and see if it helps. */ + PL_reginput = locinput; + cur_curlyx->u.curlyx.lastloc = locinput; + ST.cp = regcppush(cur_curlyx->u.curlyx.parenfloor); + REGCP_SET(ST.lastcp); + PUSH_STATE_GOTO(WHILEM_A_min, ST.save_curlyx->u.curlyx.A); /* NOTREACHED */ - case BRANCHJ: + +#undef ST +#define ST st->u.branch + + case BRANCHJ: /* /(...|A|...)/ with long next pointer */ next = scan + ARG(scan); if (next == scan) next = NULL; - inner = NEXTOPER(NEXTOPER(scan)); - goto do_branch; - case BRANCH: - inner = NEXTOPER(scan); - do_branch: - { - const I32 type = OP(scan); - if (!next || OP(next) != type) /* No choice. */ - next = inner; /* Avoid recursion. */ - else { - const I32 lastparen = *PL_reglastparen; - /* Put unwinding data on stack */ - const I32 unwind1 = SSNEWt(1,re_unwind_branch_t); - re_unwind_branch_t * const uw = SSPTRt(unwind1,re_unwind_branch_t); - - uw->prev = st->unwind; - st->unwind = unwind1; - uw->type = ((type == BRANCH) - ? RE_UNWIND_BRANCH - : RE_UNWIND_BRANCHJ); - uw->lastparen = lastparen; - uw->next = next; - uw->locinput = locinput; - uw->nextchr = nextchr; - uw->minmod = st->minmod; -#ifdef DEBUGGING - uw->regindent = ++PL_regindent; -#endif + scan = NEXTOPER(scan); + /* FALL THROUGH */ - REGCP_SET(uw->lastcp); + case BRANCH: /* /(...|A|...)/ */ + scan = NEXTOPER(scan); /* scan now points to inner node */ + if (!next || (OP(next) != BRANCH && OP(next) != BRANCHJ)) + /* last branch; skip state push and jump direct to node */ + continue; + ST.lastparen = *PL_reglastparen; + ST.next_branch = next; + REGCP_SET(ST.cp); + PL_reginput = locinput; - /* Now go into the first branch */ - next = inner; - } - } - break; + /* Now go into the branch */ + PUSH_STATE_GOTO(BRANCH_next, scan); + /* NOTREACHED */ + + case BRANCH_next_fail: /* that branch failed; try the next, if any */ + REGCP_UNWIND(ST.cp); + for (n = *PL_reglastparen; n > ST.lastparen; n--) + PL_regendp[n] = -1; + *PL_reglastparen = n; + /*dmq: *PL_reglastcloseparen = n; */ + scan = ST.next_branch; + /* no more branches? */ + if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) + sayNO; + continue; /* execute next BRANCH[J] op */ + /* NOTREACHED */ + case MINMOD: - st->minmod = 1; + minmod = 1; break; - case CURLYM: - { - st->u.curlym.l = st->u.curlym.matches = 0; - - /* We suppose that the next guy does not need - backtracking: in particular, it is of constant non-zero length, - and has no parenths to influence future backrefs. */ - st->ln = ARG1(scan); /* min to match */ - n = ARG2(scan); /* max to match */ - st->u.curlym.paren = scan->flags; - if (st->u.curlym.paren) { - if (st->u.curlym.paren > PL_regsize) - PL_regsize = st->u.curlym.paren; - if (st->u.curlym.paren > (I32)*PL_reglastparen) - *PL_reglastparen = st->u.curlym.paren; - } + +#undef ST +#define ST st->u.curlym + + case CURLYM: /* /A{m,n}B/ where A is fixed-length */ + + /* This is an optimisation of CURLYX that enables us to push + * only a single backtracking state, no matter now many matches + * there are in {m,n}. It relies on the pattern being constant + * length, with no parens to influence future backrefs + */ + + ST.me = scan; scan = NEXTOPER(scan) + NODE_STEP_REGNODE; - if (st->u.curlym.paren) + + /* if paren positive, emulate an OPEN/CLOSE around A */ + if (ST.me->flags) { + I32 paren = ST.me->flags; + if (paren > PL_regsize) + PL_regsize = paren; + if (paren > (I32)*PL_reglastparen) + *PL_reglastparen = paren; scan += NEXT_OFF(scan); /* Skip former OPEN. */ + } + ST.A = scan; + ST.B = next; + ST.alen = 0; + ST.count = 0; + ST.minmod = minmod; + minmod = 0; + ST.c1 = CHRTEST_UNINIT; + REGCP_SET(ST.cp); + + if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */ + goto curlym_do_B; + + curlym_do_A: /* execute the A in /A{m,n}B/ */ PL_reginput = locinput; - st->u.curlym.maxwanted = st->minmod ? st->ln : n; - while (PL_reginput < PL_regeol && st->u.curlym.matches < st->u.curlym.maxwanted) { - /* resume to current state on success */ - st->u.yes.prev_yes_state = yes_state; - yes_state = st; - REGMATCH(scan, CURLYM1); - yes_state = st->u.yes.prev_yes_state; - /*** all unsaved local vars undefined at this point */ - if (!result) - break; - /* on first match, determine length, u.curlym.l */ - if (!st->u.curlym.matches++) { - if (PL_reg_match_utf8) { - char *s = locinput; - while (s < PL_reginput) { - st->u.curlym.l++; - s += UTF8SKIP(s); - } - } - else { - st->u.curlym.l = PL_reginput - locinput; - } - if (st->u.curlym.l == 0) { - st->u.curlym.matches = st->u.curlym.maxwanted; - break; + PUSH_YES_STATE_GOTO(CURLYM_A, ST.A); /* match A */ + /* NOTREACHED */ + + case CURLYM_A: /* we've just matched an A */ + locinput = st->locinput; + nextchr = UCHARAT(locinput); + + ST.count++; + /* after first match, determine A's length: u.curlym.alen */ + if (ST.count == 1) { + if (PL_reg_match_utf8) { + char *s = locinput; + while (s < PL_reginput) { + ST.alen++; + s += UTF8SKIP(s); } } - locinput = PL_reginput; + else { + ST.alen = PL_reginput - locinput; + } + if (ST.alen == 0) + ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me); } + DEBUG_EXECUTE_r( + PerlIO_printf(Perl_debug_log, + "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n", + (int)(REPORT_CODE_OFF+(depth*2)), "", + (IV) ST.count, (IV)ST.alen) + ); - PL_reginput = locinput; - if (st->u.curlym.matches < st->ln) { - st->minmod = 0; + locinput = PL_reginput; + + if (cur_eval && cur_eval->u.eval.close_paren && + cur_eval->u.eval.close_paren == ST.me->flags) + goto fake_end; + + if ( ST.count < (ST.minmod ? ARG1(ST.me) : ARG2(ST.me)) ) + goto curlym_do_A; /* try to match another A */ + goto curlym_do_B; /* try to match B */ + + case CURLYM_A_fail: /* just failed to match an A */ + REGCP_UNWIND(ST.cp); + + if (ST.minmod || ST.count < ARG1(ST.me) /* min*/ + || (cur_eval && cur_eval->u.eval.close_paren && + cur_eval->u.eval.close_paren == ST.me->flags)) sayNO; + + curlym_do_B: /* execute the B in /A{m,n}B/ */ + PL_reginput = locinput; + if (ST.c1 == CHRTEST_UNINIT) { + /* calculate c1 and c2 for possible match of 1st char + * following curly */ + ST.c1 = ST.c2 = CHRTEST_VOID; + if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) { + regnode *text_node = ST.B; + if (! HAS_TEXT(text_node)) + FIND_NEXT_IMPT(text_node); + if (HAS_TEXT(text_node) + && PL_regkind[OP(text_node)] != REF) + { + ST.c1 = (U8)*STRING(text_node); + ST.c2 = + (OP(text_node) == EXACTF || OP(text_node) == REFF) + ? PL_fold[ST.c1] + : (OP(text_node) == EXACTFL || OP(text_node) == REFFL) + ? PL_fold_locale[ST.c1] + : ST.c1; + } + } } DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, - "%*s matched %"IVdf" times, len=%"IVdf"...\n", - (int)(REPORT_CODE_OFF+PL_regindent*2), "", - (IV) st->u.curlym.matches, (IV)st->u.curlym.l) - ); + "%*s CURLYM trying tail with matches=%"IVdf"...\n", + (int)(REPORT_CODE_OFF+(depth*2)), + "", (IV)ST.count) + ); + if (ST.c1 != CHRTEST_VOID + && UCHARAT(PL_reginput) != ST.c1 + && UCHARAT(PL_reginput) != ST.c2) + { + /* simulate B failing */ + state_num = CURLYM_B_fail; + goto reenter_switch; + } - /* calculate c1 and c1 for possible match of 1st char - * following curly */ - st->u.curlym.c1 = st->u.curlym.c2 = CHRTEST_VOID; - if (HAS_TEXT(next) || JUMPABLE(next)) { - regnode *text_node = next; - if (! HAS_TEXT(text_node)) - FIND_NEXT_IMPT(text_node); - if (HAS_TEXT(text_node) - && PL_regkind[OP(text_node)] != REF) - { - st->u.curlym.c1 = (U8)*STRING(text_node); - st->u.curlym.c2 = - (OP(text_node) == EXACTF || OP(text_node) == REFF) - ? PL_fold[st->u.curlym.c1] - : (OP(text_node) == EXACTFL || OP(text_node) == REFFL) - ? PL_fold_locale[st->u.curlym.c1] - : st->u.curlym.c1; + if (ST.me->flags) { + /* mark current A as captured */ + I32 paren = ST.me->flags; + if (ST.count) { + PL_regstartp[paren] + = HOPc(PL_reginput, -ST.alen) - PL_bostr; + PL_regendp[paren] = PL_reginput - PL_bostr; + /*dmq: *PL_reglastcloseparen = paren; */ } + else + PL_regendp[paren] = -1; + if (cur_eval && cur_eval->u.eval.close_paren && + cur_eval->u.eval.close_paren == ST.me->flags) + { + if (ST.count) + goto fake_end; + else + sayNO; + } } + + PUSH_STATE_GOTO(CURLYM_B, ST.B); /* match B */ + /* NOTREACHED */ - REGCP_SET(st->u.curlym.lastcp); + case CURLYM_B_fail: /* just failed to match a B */ + REGCP_UNWIND(ST.cp); + if (ST.minmod) { + if (ST.count == ARG2(ST.me) /* max */) + sayNO; + goto curlym_do_A; /* try to match a further A */ + } + /* backtrack one A */ + if (ST.count == ARG1(ST.me) /* min */) + sayNO; + ST.count--; + locinput = HOPc(locinput, -ST.alen); + goto curlym_do_B; /* try to match B */ + +#undef ST +#define ST st->u.curly + +#define CURLY_SETPAREN(paren, success) \ + if (paren) { \ + if (success) { \ + PL_regstartp[paren] = HOPc(locinput, -1) - PL_bostr; \ + PL_regendp[paren] = locinput - PL_bostr; \ + *PL_reglastcloseparen = paren; \ + } \ + else \ + PL_regendp[paren] = -1; \ + } - st->u.curlym.minmod = st->minmod; - st->minmod = 0; - while (st->u.curlym.matches >= st->ln - && (st->u.curlym.matches <= n - /* for REG_INFTY, ln could overflow to negative */ - || (n == REG_INFTY && st->u.curlym.matches >= 0))) - { - /* If it could work, try it. */ - if (st->u.curlym.c1 == CHRTEST_VOID || - UCHARAT(PL_reginput) == st->u.curlym.c1 || - UCHARAT(PL_reginput) == st->u.curlym.c2) - { - DEBUG_EXECUTE_r( - PerlIO_printf(Perl_debug_log, - "%*s trying tail with matches=%"IVdf"...\n", - (int)(REPORT_CODE_OFF+PL_regindent*2), - "", (IV)st->u.curlym.matches) - ); - if (st->u.curlym.paren) { - if (st->u.curlym.matches) { - PL_regstartp[st->u.curlym.paren] - = HOPc(PL_reginput, -st->u.curlym.l) - PL_bostr; - PL_regendp[st->u.curlym.paren] = PL_reginput - PL_bostr; - } - else - PL_regendp[st->u.curlym.paren] = -1; - } - /* resume to current state on success */ - st->u.yes.prev_yes_state = yes_state; - yes_state = st; - REGMATCH(next, CURLYM2); - yes_state = st->u.yes.prev_yes_state; - /*** all unsaved local vars undefined at this point */ - if (result) - /* XXX tmp sayYES; */ - sayYES_FINAL; - REGCP_UNWIND(st->u.curlym.lastcp); - } - /* Couldn't or didn't -- move forward/backward. */ - if (st->u.curlym.minmod) { - PL_reginput = locinput; - /* resume to current state on success */ - st->u.yes.prev_yes_state = yes_state; - yes_state = st; - REGMATCH(scan, CURLYM3); - yes_state = st->u.yes.prev_yes_state; - /*** all unsaved local vars undefined at this point */ - if (result) { - st->u.curlym.matches++; - locinput = PL_reginput; - } - else - sayNO; - } - else { - st->u.curlym.matches--; - locinput = HOPc(locinput, -st->u.curlym.l); - PL_reginput = locinput; - } + case STAR: /* /A*B/ where A is width 1 */ + ST.paren = 0; + ST.min = 0; + ST.max = REG_INFTY; + scan = NEXTOPER(scan); + goto repeat; + case PLUS: /* /A+B/ where A is width 1 */ + ST.paren = 0; + ST.min = 1; + ST.max = REG_INFTY; + scan = NEXTOPER(scan); + goto repeat; + case CURLYN: /* /(A){m,n}B/ where A is width 1 */ + ST.paren = scan->flags; /* Which paren to set */ + if (ST.paren > PL_regsize) + PL_regsize = ST.paren; + if (ST.paren > (I32)*PL_reglastparen) + *PL_reglastparen = ST.paren; + ST.min = ARG1(scan); /* min to match */ + ST.max = ARG2(scan); /* max to match */ + if (cur_eval && cur_eval->u.eval.close_paren && + cur_eval->u.eval.close_paren == ST.paren) { + ST.min=1; + ST.max=1; } - sayNO; - /* NOTREACHED */ - break; - } - case CURLYN: - st->u.plus.paren = scan->flags; /* Which paren to set */ - if (st->u.plus.paren > PL_regsize) - PL_regsize = st->u.plus.paren; - if (st->u.plus.paren > (I32)*PL_reglastparen) - *PL_reglastparen = st->u.plus.paren; - st->ln = ARG1(scan); /* min to match */ - n = ARG2(scan); /* max to match */ scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE); goto repeat; - case CURLY: - st->u.plus.paren = 0; - st->ln = ARG1(scan); /* min to match */ - n = ARG2(scan); /* max to match */ + case CURLY: /* /A{m,n}B/ where A is width 1 */ + ST.paren = 0; + ST.min = ARG1(scan); /* min to match */ + ST.max = ARG2(scan); /* max to match */ scan = NEXTOPER(scan) + NODE_STEP_REGNODE; - goto repeat; - case STAR: - st->ln = 0; - n = REG_INFTY; - scan = NEXTOPER(scan); - st->u.plus.paren = 0; - goto repeat; - case PLUS: - st->ln = 1; - n = REG_INFTY; - scan = NEXTOPER(scan); - st->u.plus.paren = 0; repeat: /* * Lookahead to avoid useless match attempts * when we know what character comes next. - */ - - /* + * * Used to only do .*x and .*?x, but now it allows * for )'s, ('s and (?{ ... })'s to be in the way * of the quantifier and the EXACT-like node. -- japhy */ + if (ST.min > ST.max) /* XXX make this a compile-time check? */ + sayNO; if (HAS_TEXT(next) || JUMPABLE(next)) { U8 *s; regnode *text_node = next; @@ -4266,21 +4236,21 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) FIND_NEXT_IMPT(text_node); if (! HAS_TEXT(text_node)) - st->u.plus.c1 = st->u.plus.c2 = CHRTEST_VOID; + ST.c1 = ST.c2 = CHRTEST_VOID; else { if (PL_regkind[OP(text_node)] == REF) { - st->u.plus.c1 = st->u.plus.c2 = CHRTEST_VOID; + ST.c1 = ST.c2 = CHRTEST_VOID; goto assume_ok_easy; } else s = (U8*)STRING(text_node); if (!UTF) { - st->u.plus.c2 = st->u.plus.c1 = *s; + ST.c2 = ST.c1 = *s; if (OP(text_node) == EXACTF || OP(text_node) == REFF) - st->u.plus.c2 = PL_fold[st->u.plus.c1]; + ST.c2 = PL_fold[ST.c1]; else if (OP(text_node) == EXACTFL || OP(text_node) == REFFL) - st->u.plus.c2 = PL_fold_locale[st->u.plus.c1]; + ST.c2 = PL_fold_locale[ST.c1]; } else { /* UTF */ if (OP(text_node) == EXACTF || OP(text_node) == REFF) { @@ -4290,187 +4260,266 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) to_utf8_lower((U8*)s, tmpbuf1, &ulen1); to_utf8_upper((U8*)s, tmpbuf2, &ulen2); - - st->u.plus.c1 = utf8n_to_uvuni(tmpbuf1, UTF8_MAXBYTES, 0, - uniflags); - st->u.plus.c2 = utf8n_to_uvuni(tmpbuf2, UTF8_MAXBYTES, 0, - uniflags); +#ifdef EBCDIC + ST.c1 = utf8n_to_uvchr(tmpbuf1, UTF8_MAXLEN, 0, + ckWARN(WARN_UTF8) ? + 0 : UTF8_ALLOW_ANY); + ST.c2 = utf8n_to_uvchr(tmpbuf2, UTF8_MAXLEN, 0, + ckWARN(WARN_UTF8) ? + 0 : UTF8_ALLOW_ANY); +#else + ST.c1 = utf8n_to_uvuni(tmpbuf1, UTF8_MAXBYTES, 0, + uniflags); + ST.c2 = utf8n_to_uvuni(tmpbuf2, UTF8_MAXBYTES, 0, + uniflags); +#endif } else { - st->u.plus.c2 = st->u.plus.c1 = utf8n_to_uvchr(s, UTF8_MAXBYTES, 0, + ST.c2 = ST.c1 = utf8n_to_uvchr(s, UTF8_MAXBYTES, 0, uniflags); } } } } else - st->u.plus.c1 = st->u.plus.c2 = CHRTEST_VOID; + ST.c1 = ST.c2 = CHRTEST_VOID; assume_ok_easy: + + ST.A = scan; + ST.B = next; PL_reginput = locinput; - if (st->minmod) { - st->minmod = 0; - if (st->ln && regrepeat(rex, scan, st->ln) < st->ln) + if (minmod) { + minmod = 0; + if (ST.min && regrepeat(rex, ST.A, ST.min) < ST.min) sayNO; + ST.count = ST.min; locinput = PL_reginput; - REGCP_SET(st->u.plus.lastcp); - if (st->u.plus.c1 != CHRTEST_VOID) { - st->u.plus.old = locinput; - st->u.plus.count = 0; - - if (n == REG_INFTY) { - st->u.plus.e = PL_regeol - 1; - if (do_utf8) - while (UTF8_IS_CONTINUATION(*(U8*)st->u.plus.e)) - st->u.plus.e--; - } - else if (do_utf8) { - int m = n - st->ln; - for (st->u.plus.e = locinput; - m >0 && st->u.plus.e + UTF8SKIP(st->u.plus.e) <= PL_regeol; m--) - st->u.plus.e += UTF8SKIP(st->u.plus.e); + REGCP_SET(ST.cp); + if (ST.c1 == CHRTEST_VOID) + goto curly_try_B_min; + + ST.oldloc = locinput; + + /* set ST.maxpos to the furthest point along the + * string that could possibly match */ + if (ST.max == REG_INFTY) { + ST.maxpos = PL_regeol - 1; + if (do_utf8) + while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos)) + ST.maxpos--; + } + else if (do_utf8) { + int m = ST.max - ST.min; + for (ST.maxpos = locinput; + m >0 && ST.maxpos + UTF8SKIP(ST.maxpos) <= PL_regeol; m--) + ST.maxpos += UTF8SKIP(ST.maxpos); + } + else { + ST.maxpos = locinput + ST.max - ST.min; + if (ST.maxpos >= PL_regeol) + ST.maxpos = PL_regeol - 1; + } + goto curly_try_B_min_known; + + } + else { + ST.count = regrepeat(rex, ST.A, ST.max); + locinput = PL_reginput; + if (ST.count < ST.min) + sayNO; + if ((ST.count > ST.min) + && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL)) + { + /* A{m,n} must come at the end of the string, there's + * no point in backing off ... */ + ST.min = ST.count; + /* ...except that $ and \Z can match before *and* after + newline at the end. Consider "\n\n" =~ /\n+\Z\n/. + We may back off by one in this case. */ + if (UCHARAT(PL_reginput - 1) == '\n' && OP(ST.B) != EOS) + ST.min--; + } + REGCP_SET(ST.cp); + goto curly_try_B_max; + } + /* NOTREACHED */ + + + case CURLY_B_min_known_fail: + /* failed to find B in a non-greedy match where c1,c2 valid */ + if (ST.paren && ST.count) + PL_regendp[ST.paren] = -1; + + PL_reginput = locinput; /* Could be reset... */ + REGCP_UNWIND(ST.cp); + /* Couldn't or didn't -- move forward. */ + ST.oldloc = locinput; + if (do_utf8) + locinput += UTF8SKIP(locinput); + else + locinput++; + ST.count++; + curly_try_B_min_known: + /* find the next place where 'B' could work, then call B */ + { + int n; + if (do_utf8) { + n = (ST.oldloc == locinput) ? 0 : 1; + if (ST.c1 == ST.c2) { + STRLEN len; + /* set n to utf8_distance(oldloc, locinput) */ + while (locinput <= ST.maxpos && + utf8n_to_uvchr((U8*)locinput, + UTF8_MAXBYTES, &len, + uniflags) != (UV)ST.c1) { + locinput += len; + n++; + } } else { - st->u.plus.e = locinput + n - st->ln; - if (st->u.plus.e >= PL_regeol) - st->u.plus.e = PL_regeol - 1; - } - while (1) { - /* Find place 'next' could work */ - if (!do_utf8) { - if (st->u.plus.c1 == st->u.plus.c2) { - while (locinput <= st->u.plus.e && - UCHARAT(locinput) != st->u.plus.c1) - locinput++; - } else { - while (locinput <= st->u.plus.e - && UCHARAT(locinput) != st->u.plus.c1 - && UCHARAT(locinput) != st->u.plus.c2) - locinput++; - } - st->u.plus.count = locinput - st->u.plus.old; - } - else { - if (st->u.plus.c1 == st->u.plus.c2) { - STRLEN len; - /* count initialised to - * utf8_distance(old, locinput) */ - while (locinput <= st->u.plus.e && - utf8n_to_uvchr((U8*)locinput, - UTF8_MAXBYTES, &len, - uniflags) != (UV)st->u.plus.c1) { - locinput += len; - st->u.plus.count++; - } - } else { - /* count initialised to - * utf8_distance(old, locinput) */ - while (locinput <= st->u.plus.e) { - STRLEN len; - const UV c = utf8n_to_uvchr((U8*)locinput, - UTF8_MAXBYTES, &len, - uniflags); - if (c == (UV)st->u.plus.c1 || c == (UV)st->u.plus.c2) - break; - locinput += len; - st->u.plus.count++; - } - } - } - if (locinput > st->u.plus.e) - sayNO; - /* PL_reginput == old now */ - if (locinput != st->u.plus.old) { - st->ln = 1; /* Did some */ - if (regrepeat(rex, scan, st->u.plus.count) < st->u.plus.count) - sayNO; + /* set n to utf8_distance(oldloc, locinput) */ + while (locinput <= ST.maxpos) { + STRLEN len; + const UV c = utf8n_to_uvchr((U8*)locinput, + UTF8_MAXBYTES, &len, + uniflags); + if (c == (UV)ST.c1 || c == (UV)ST.c2) + break; + locinput += len; + n++; } - /* PL_reginput == locinput now */ - TRYPAREN(st->u.plus.paren, st->ln, locinput, PLUS1); - /*** all unsaved local vars undefined at this point */ - PL_reginput = locinput; /* Could be reset... */ - REGCP_UNWIND(st->u.plus.lastcp); - /* Couldn't or didn't -- move forward. */ - st->u.plus.old = locinput; - if (do_utf8) - locinput += UTF8SKIP(locinput); - else - locinput++; - st->u.plus.count = 1; } } - else - while (n >= st->ln || (n == REG_INFTY && st->ln > 0)) { /* ln overflow ? */ - UV c; - if (st->u.plus.c1 != CHRTEST_VOID) { - if (do_utf8) - c = utf8n_to_uvchr((U8*)PL_reginput, - UTF8_MAXBYTES, 0, - uniflags); - else - c = UCHARAT(PL_reginput); - /* If it could work, try it. */ - if (c == (UV)st->u.plus.c1 || c == (UV)st->u.plus.c2) { - TRYPAREN(st->u.plus.paren, st->ln, PL_reginput, PLUS2); - /*** all unsaved local vars undefined at this point */ - REGCP_UNWIND(st->u.plus.lastcp); - } - } - /* If it could work, try it. */ - else if (st->u.plus.c1 == CHRTEST_VOID) { - TRYPAREN(st->u.plus.paren, st->ln, PL_reginput, PLUS3); - /*** all unsaved local vars undefined at this point */ - REGCP_UNWIND(st->u.plus.lastcp); + else { + if (ST.c1 == ST.c2) { + while (locinput <= ST.maxpos && + UCHARAT(locinput) != ST.c1) + locinput++; } - /* Couldn't or didn't -- move forward. */ - PL_reginput = locinput; - if (regrepeat(rex, scan, 1)) { - st->ln++; - locinput = PL_reginput; + else { + while (locinput <= ST.maxpos + && UCHARAT(locinput) != ST.c1 + && UCHARAT(locinput) != ST.c2) + locinput++; } - else + n = locinput - ST.oldloc; + } + if (locinput > ST.maxpos) + sayNO; + /* PL_reginput == oldloc now */ + if (n) { + ST.count += n; + if (regrepeat(rex, ST.A, n) < n) sayNO; } + PL_reginput = locinput; + CURLY_SETPAREN(ST.paren, ST.count); + if (cur_eval && cur_eval->u.eval.close_paren && + cur_eval->u.eval.close_paren == ST.paren) { + goto fake_end; + } + PUSH_STATE_GOTO(CURLY_B_min_known, ST.B); } - else { - n = regrepeat(rex, scan, n); + /* NOTREACHED */ + + + case CURLY_B_min_fail: + /* failed to find B in a non-greedy match where c1,c2 invalid */ + if (ST.paren && ST.count) + PL_regendp[ST.paren] = -1; + + REGCP_UNWIND(ST.cp); + /* failed -- move forward one */ + PL_reginput = locinput; + if (regrepeat(rex, ST.A, 1)) { + ST.count++; locinput = PL_reginput; - if ((st->ln < n) && (PL_regkind[OP(next)] == EOL) && - (OP(next) != MEOL || OP(next) == SEOL || OP(next) == EOS)) + if (ST.count <= ST.max || (ST.max == REG_INFTY && + ST.count > 0)) /* count overflow ? */ { - st->ln = n; /* why back off? */ - /* ...because $ and \Z can match before *and* after - newline at the end. Consider "\n\n" =~ /\n+\Z\n/. - We should back off by one in this case. */ - if (UCHARAT(PL_reginput - 1) == '\n' && OP(next) != EOS) - st->ln--; - } - REGCP_SET(st->u.plus.lastcp); - { - while (n >= st->ln) { - UV c = 0; - if (st->u.plus.c1 != CHRTEST_VOID) { - if (do_utf8) - c = utf8n_to_uvchr((U8*)PL_reginput, - UTF8_MAXBYTES, 0, - uniflags); - else - c = UCHARAT(PL_reginput); - } - /* If it could work, try it. */ - if (st->u.plus.c1 == CHRTEST_VOID || c == (UV)st->u.plus.c1 || c == (UV)st->u.plus.c2) { - TRYPAREN(st->u.plus.paren, n, PL_reginput, PLUS4); - /*** all unsaved local vars undefined at this point */ - REGCP_UNWIND(st->u.plus.lastcp); - } - /* Couldn't or didn't -- back up. */ - n--; - PL_reginput = locinput = HOPc(locinput, -1); - } + curly_try_B_min: + CURLY_SETPAREN(ST.paren, ST.count); + if (cur_eval && cur_eval->u.eval.close_paren && + cur_eval->u.eval.close_paren == ST.paren) { + goto fake_end; + } + PUSH_STATE_GOTO(CURLY_B_min, ST.B); } } sayNO; - break; + /* NOTREACHED */ + + + curly_try_B_max: + /* a successful greedy match: now try to match B */ + { + UV c = 0; + if (ST.c1 != CHRTEST_VOID) + c = do_utf8 ? utf8n_to_uvchr((U8*)PL_reginput, + UTF8_MAXBYTES, 0, uniflags) + : (UV) UCHARAT(PL_reginput); + /* If it could work, try it. */ + if (ST.c1 == CHRTEST_VOID || c == (UV)ST.c1 || c == (UV)ST.c2) { + CURLY_SETPAREN(ST.paren, ST.count); + if (cur_eval && cur_eval->u.eval.close_paren && + cur_eval->u.eval.close_paren == ST.paren) { + goto fake_end; + } + PUSH_STATE_GOTO(CURLY_B_max, ST.B); + /* NOTREACHED */ + } + } + /* FALL THROUGH */ + case CURLY_B_max_fail: + /* failed to find B in a greedy match */ + if (ST.paren && ST.count) + PL_regendp[ST.paren] = -1; + + REGCP_UNWIND(ST.cp); + /* back up. */ + if (--ST.count < ST.min) + sayNO; + PL_reginput = locinput = HOPc(locinput, -1); + goto curly_try_B_max; + +#undef ST + + case END: + fake_end: + if (cur_eval) { + /* we've just finished A in /(??{A})B/; now continue with B */ + I32 tmpix; + + + st->u.eval.toggle_reg_flags + = cur_eval->u.eval.toggle_reg_flags; + PL_reg_flags ^= st->u.eval.toggle_reg_flags; + + st->u.eval.prev_rex = rex; /* inner */ + rex = cur_eval->u.eval.prev_rex; /* outer */ + cur_curlyx = cur_eval->u.eval.prev_curlyx; + ReREFCNT_inc(rex); + st->u.eval.cp = regcppush(0); /* Save *all* the positions. */ + REGCP_SET(st->u.eval.lastcp); + PL_reginput = locinput; + + /* Restore parens of the outer rex without popping the + * savestack */ + tmpix = PL_savestack_ix; + PL_savestack_ix = cur_eval->u.eval.lastcp; + regcppop(rex); + PL_savestack_ix = tmpix; + + st->u.eval.prev_eval = cur_eval; + cur_eval = cur_eval->u.eval.prev_eval; + DEBUG_EXECUTE_r( + PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %x\n", + REPORT_CODE_OFF+depth*2, "",(int)cur_eval);); + PUSH_YES_STATE_GOTO(EVAL_AB, + st->u.eval.prev_eval->u.eval.B); /* match B */ + } + if (locinput < reginfo->till) { DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n", @@ -4478,40 +4527,43 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) (long)(locinput - PL_reg_starttry), (long)(reginfo->till - PL_reg_starttry), PL_colors[5])); - sayNO_FINAL; /* Cannot match: too short. */ + sayNO_SILENT; /* Cannot match: too short. */ } PL_reginput = locinput; /* put where regtry can find it */ - sayYES_FINAL; /* Success! */ + sayYES; /* Success! */ case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */ DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, "%*s %ssubpattern success...%s\n", - REPORT_CODE_OFF+PL_regindent*2, "", PL_colors[4], PL_colors[5])); + REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])); PL_reginput = locinput; /* put where regtry can find it */ - sayYES_FINAL; /* Success! */ + sayYES; /* Success! */ - case SUSPEND: /* (?>FOO) */ - st->u.ifmatch.wanted = 1; +#undef ST +#define ST st->u.ifmatch + + case SUSPEND: /* (?>A) */ + ST.wanted = 1; PL_reginput = locinput; goto do_ifmatch; - case UNLESSM: /* -ve lookaround: (?!FOO), or with flags, (?u.ifmatch.wanted = 0; + case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?u.ifmatch.wanted = 1; + case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */ + ST.wanted = 1; ifmatch_trivial_fail_test: if (scan->flags) { char * const s = HOPBACKc(locinput, scan->flags); if (!s) { /* trivial fail */ - if (st->logical) { - st->logical = 0; - st->sw = 1 - (bool)st->u.ifmatch.wanted; + if (logical) { + logical = 0; + sw = 1 - (bool)ST.wanted; } - else if (st->u.ifmatch.wanted) + else if (ST.wanted) sayNO; next = scan + ARG(scan); if (next == scan) @@ -4524,13 +4576,35 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) PL_reginput = locinput; do_ifmatch: - /* resume to current state on success */ - st->u.yes.prev_yes_state = yes_state; - yes_state = st; - PUSH_STATE(newst, resume_IFMATCH); - st = newst; - next = NEXTOPER(NEXTOPER(scan)); - break; + ST.me = scan; + ST.logical = logical; + /* execute body of (?...A) */ + PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan))); + /* NOTREACHED */ + + case IFMATCH_A_fail: /* body of (?...A) failed */ + ST.wanted = !ST.wanted; + /* FALL THROUGH */ + + case IFMATCH_A: /* body of (?...A) succeeded */ + if (ST.logical) { + sw = (bool)ST.wanted; + } + else if (!ST.wanted) + sayNO; + + if (OP(ST.me) == SUSPEND) + locinput = PL_reginput; + else { + locinput = PL_reginput = st->locinput; + nextchr = UCHARAT(locinput); + } + scan = ST.me + ARG(ST.me); + if (scan == ST.me) + scan = NULL; + continue; /* execute B */ + +#undef ST case LONGJMP: next = scan + ARG(scan); @@ -4543,48 +4617,36 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) Perl_croak(aTHX_ "regexp memory corruption"); } - reenter: scan = next; continue; /* NOTREACHED */ - /* simulate recursively calling regmatch(), but without actually - * recursing - ie save the current state on the heap rather than on - * the stack, then re-enter the loop. This avoids complex regexes - * blowing the processor stack */ - - start_recurse: + push_yes_state: + /* push a state that backtracks on success */ + st->u.yes.prev_yes_state = yes_state; + yes_state = st; + /* FALL THROUGH */ + push_state: + /* push a new regex state, then continue at scan */ { - /* push new state */ - regmatch_state *oldst = st; + regmatch_state *newst; + DEBUG_STATE_pp("push"); depth++; + st->locinput = locinput; + newst = st+1; + if (newst > SLAB_LAST(PL_regmatch_slab)) + newst = S_push_slab(aTHX); + PL_regmatch_state = newst; - /* grab the next free state slot */ - st++; - if (st > SLAB_LAST(PL_regmatch_slab)) - st = S_push_slab(aTHX); - PL_regmatch_state = st; - - oldst->next = next; - oldst->n = n; - oldst->locinput = locinput; - - st->cc = oldst->cc; locinput = PL_reginput; - nextchr = UCHARAT(locinput); - st->minmod = 0; - st->sw = 0; - st->logical = 0; - st->unwind = 0; -#ifdef DEBUGGING - PL_regindent++; -#endif + nextchr = UCHARAT(locinput); + st = newst; + continue; + /* NOTREACHED */ } } - - /* * We get here only if there's trouble -- normally "case END" is * the terminating point. @@ -4593,19 +4655,22 @@ S_regmatch(pTHX_ const regmatch_info *reginfo, regnode *prog) /*NOTREACHED*/ sayNO; -yes_final: - +yes: if (yes_state) { /* we have successfully completed a subexpression, but we must now * pop to the state marked by yes_state and continue from there */ - - /*XXX tmp for CURLYM*/ - regmatch_slab * const oslab = PL_regmatch_slab; - regmatch_state * const ost = st; - regmatch_state * const oys = yes_state; - int odepth = depth; - assert(st != yes_state); +#ifdef DEBUGGING + while (st != yes_state) { + st--; + if (st < SLAB_FIRST(PL_regmatch_slab)) { + PL_regmatch_slab = PL_regmatch_slab->prev; + st = SLAB_LAST(PL_regmatch_slab); + } + DEBUG_STATE_pp("pop (yes)"); + depth--; + } +#else while (yes_state < SLAB_FIRST(PL_regmatch_slab) || yes_state > SLAB_LAST(PL_regmatch_slab)) { @@ -4615,265 +4680,52 @@ yes_final: st = SLAB_LAST(PL_regmatch_slab); } depth -= (st - yes_state); - DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "POP STATE TO (%d)\n", depth)); +#endif st = yes_state; yes_state = st->u.yes.prev_yes_state; PL_regmatch_state = st; - switch (st->resume_state) { - case resume_EVAL: - if (st->u.eval.toggleutf) - PL_reg_flags ^= RF_utf8; - ReREFCNT_dec(rex); - rex = st->u.eval.prev_rex; - /* XXXX This is too dramatic a measure... */ - PL_reg_maxiter = 0; - /* Restore parens of the caller without popping the - * savestack */ - { - const I32 tmp = PL_savestack_ix; - PL_savestack_ix = st->u.eval.lastcp; - regcppop(rex); - PL_savestack_ix = tmp; - } - PL_reginput = locinput; - /* continue at the node following the (??{...}) */ - next = st->next; - goto reenter; - - case resume_IFMATCH: - if (st->logical) { - st->logical = 0; - st->sw = (bool)st->u.ifmatch.wanted; - } - else if (!st->u.ifmatch.wanted) - sayNO; - - if (OP(st->scan) == SUSPEND) - locinput = PL_reginput; - else { - locinput = PL_reginput = st->locinput; - nextchr = UCHARAT(locinput); - } - next = st->scan + ARG(st->scan); - if (next == st->scan) - next = NULL; - goto reenter; - - /* XXX tmp don't handle yes_state yet */ - case resume_CURLYM1: - case resume_CURLYM2: - case resume_CURLYM3: - PL_regmatch_slab =oslab; - st = ost; - PL_regmatch_state = st; - depth = odepth; - yes_state = oys; - DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "XXX revering a CURLYM\n")); - goto yes; - - default: - Perl_croak(aTHX_ "unexpected yes reume state"); - } + state_num = st->resume_state; + goto reenter_switch; } DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n", PL_colors[4], PL_colors[5])); -yes: -#ifdef DEBUGGING - PL_regindent--; -#endif result = 1; - /* XXX this is duplicate(ish) code to that in the do_no section. - * eventually a yes should just pop the stack back to the current - * yes_state */ - if (depth) { - /* restore previous state and re-enter */ - POP_STATE; - - switch (st->resume_state) { - case resume_TRIE1: - goto resume_point_TRIE1; - case resume_TRIE2: - goto resume_point_TRIE2; - case resume_CURLYX: - goto resume_point_CURLYX; - case resume_WHILEM1: - goto resume_point_WHILEM1; - case resume_WHILEM2: - goto resume_point_WHILEM2; - case resume_WHILEM3: - goto resume_point_WHILEM3; - case resume_WHILEM4: - goto resume_point_WHILEM4; - case resume_WHILEM5: - goto resume_point_WHILEM5; - case resume_WHILEM6: - goto resume_point_WHILEM6; - case resume_CURLYM1: - goto resume_point_CURLYM1; - case resume_CURLYM2: - goto resume_point_CURLYM2; - case resume_CURLYM3: - goto resume_point_CURLYM3; - case resume_PLUS1: - goto resume_point_PLUS1; - case resume_PLUS2: - goto resume_point_PLUS2; - case resume_PLUS3: - goto resume_point_PLUS3; - case resume_PLUS4: - goto resume_point_PLUS4; - - case resume_IFMATCH: - case resume_EVAL: - default: - Perl_croak(aTHX_ "regexp resume memory corruption"); - } - } goto final_exit; no: DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log, - "%*s %sfailed...%s\n", - REPORT_CODE_OFF+PL_regindent*2, "", PL_colors[4], PL_colors[5]) + "%*s %sfailed...%s\n", + REPORT_CODE_OFF+depth*2, "", + PL_colors[4], PL_colors[5]) ); - goto do_no; -no_final: -do_no: - if (st->unwind) { - re_unwind_t * const uw = SSPTRt(st->unwind,re_unwind_t); - - switch (uw->type) { - case RE_UNWIND_BRANCH: - case RE_UNWIND_BRANCHJ: - { - re_unwind_branch_t * const uwb = &(uw->branch); - const I32 lastparen = uwb->lastparen; - - REGCP_UNWIND(uwb->lastcp); - for (n = *PL_reglastparen; n > lastparen; n--) - PL_regendp[n] = -1; - *PL_reglastparen = n; - scan = next = uwb->next; - st->minmod = uwb->minmod; - if ( !scan || - OP(scan) != (uwb->type == RE_UNWIND_BRANCH - ? BRANCH : BRANCHJ) ) { /* Failure */ - st->unwind = uwb->prev; -#ifdef DEBUGGING - PL_regindent--; -#endif - goto do_no; - } - /* Have more choice yet. Reuse the same uwb. */ - if ((n = (uwb->type == RE_UNWIND_BRANCH - ? NEXT_OFF(next) : ARG(next)))) - next += n; - else - next = NULL; /* XXXX Needn't unwinding in this case... */ - uwb->next = next; - next = NEXTOPER(scan); - if (uwb->type == RE_UNWIND_BRANCHJ) - next = NEXTOPER(next); - locinput = uwb->locinput; - nextchr = uwb->nextchr; -#ifdef DEBUGGING - PL_regindent = uwb->regindent; -#endif - - goto reenter; - } - /* NOTREACHED */ - default: - Perl_croak(aTHX_ "regexp unwind memory corruption"); - } - /* NOTREACHED */ - } - -#ifdef DEBUGGING - PL_regindent--; -#endif - result = 0; +no_silent: if (depth) { /* there's a previous state to backtrack to */ - POP_STATE; - switch (st->resume_state) { - case resume_TRIE1: - goto resume_point_TRIE1; - case resume_TRIE2: - goto resume_point_TRIE2; - case resume_EVAL: - /* we have failed an (??{...}). Restore state to the outer re - * then re-throw the failure */ - if (st->u.eval.toggleutf) - PL_reg_flags ^= RF_utf8; - ReREFCNT_dec(rex); - rex = st->u.eval.prev_rex; - yes_state = st->u.yes.prev_yes_state; - - /* XXXX This is too dramatic a measure... */ - PL_reg_maxiter = 0; + st--; + if (st < SLAB_FIRST(PL_regmatch_slab)) { + PL_regmatch_slab = PL_regmatch_slab->prev; + st = SLAB_LAST(PL_regmatch_slab); + } + PL_regmatch_state = st; + locinput= st->locinput; + nextchr = UCHARAT(locinput); - PL_reginput = locinput; - REGCP_UNWIND(st->u.eval.lastcp); - regcppop(rex); - goto do_no; - - case resume_CURLYX: - goto resume_point_CURLYX; - case resume_WHILEM1: - goto resume_point_WHILEM1; - case resume_WHILEM2: - goto resume_point_WHILEM2; - case resume_WHILEM3: - goto resume_point_WHILEM3; - case resume_WHILEM4: - goto resume_point_WHILEM4; - case resume_WHILEM5: - goto resume_point_WHILEM5; - case resume_WHILEM6: - goto resume_point_WHILEM6; - case resume_CURLYM1: - goto resume_point_CURLYM1; - case resume_CURLYM2: - goto resume_point_CURLYM2; - case resume_CURLYM3: - goto resume_point_CURLYM3; - case resume_IFMATCH: + DEBUG_STATE_pp("pop"); + depth--; + if (yes_state == st) yes_state = st->u.yes.prev_yes_state; - if (st->logical) { - st->logical = 0; - st->sw = !st->u.ifmatch.wanted; - } - else if (st->u.ifmatch.wanted) - sayNO; - assert(OP(scan) != SUSPEND); /* XXX DAPM tmp */ - locinput = PL_reginput = st->locinput; - nextchr = UCHARAT(locinput); - next = scan + ARG(scan); - if (next == scan) - next = NULL; - goto reenter; - - case resume_PLUS1: - goto resume_point_PLUS1; - case resume_PLUS2: - goto resume_point_PLUS2; - case resume_PLUS3: - goto resume_point_PLUS3; - case resume_PLUS4: - goto resume_point_PLUS4; - default: - Perl_croak(aTHX_ "regexp resume memory corruption"); - } + state_num = st->resume_state + 1; /* failure = success + 1 */ + goto reenter_switch; } + result = 0; -final_exit: + final_exit: /* restore original high-water mark */ PL_regmatch_slab = orig_slab; @@ -4891,7 +4743,6 @@ final_exit: } return result; - } /* @@ -5128,12 +4979,11 @@ S_regrepeat(pTHX_ const regexp *prog, const regnode *p, I32 max) PL_reginput = scan; DEBUG_r({ - SV *re_debug_flags = NULL; - SV * const prop = sv_newmortal(); - GET_RE_DEBUG_FLAGS; + GET_RE_DEBUG_FLAGS_DECL; DEBUG_EXECUTE_r({ - regprop(prog, prop, p); - PerlIO_printf(Perl_debug_log, + SV * const prop = sv_newmortal(); + regprop(prog, prop, p); + PerlIO_printf(Perl_debug_log, "%*s %s can match %"IVdf" times out of %"IVdf"...\n", REPORT_CODE_OFF+1, "", SvPVX_const(prop),(IV)c,(IV)max); }); @@ -5217,7 +5067,7 @@ S_reginclass(pTHX_ const regexp *prog, register const regnode *n, register const c = utf8n_to_uvchr(p, UTF8_MAXBYTES, &len, (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV) | UTF8_CHECK_ONLY); /* see [perl #37836] for UTF8_ALLOW_ANYUV */ - if (len == (STRLEN)-1) + if (len == (STRLEN)-1) Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)"); } @@ -5337,20 +5187,47 @@ S_reghop3(U8 *s, I32 off, const U8* lim) } } else { - while (off++) { - if (s > lim) { - s--; - if (UTF8_IS_CONTINUED(*s)) { - while (s > (U8*)lim && UTF8_IS_CONTINUATION(*s)) - s--; - } - /* XXX could check well-formedness here */ + while (off++ && s > lim) { + s--; + if (UTF8_IS_CONTINUED(*s)) { + while (s > lim && UTF8_IS_CONTINUATION(*s)) + s--; } + /* XXX could check well-formedness here */ } } return s; } +#ifdef XXX_dmq +/* there are a bunch of places where we use two reghop3's that should + be replaced with this routine. but since thats not done yet + we ifdef it out - dmq +*/ +STATIC U8 * +S_reghop4(U8 *s, I32 off, const U8* llim, const U8* rlim) +{ + dVAR; + if (off >= 0) { + while (off-- && s < rlim) { + /* XXX could check well-formedness here */ + s += UTF8SKIP(s); + } + } + else { + while (off++ && s > llim) { + s--; + if (UTF8_IS_CONTINUED(*s)) { + while (s > llim && UTF8_IS_CONTINUATION(*s)) + s--; + } + /* XXX could check well-formedness here */ + } + } + return s; +} +#endif + STATIC U8 * S_reghopmaybe3(U8* s, I32 off, const U8* lim) { @@ -5364,17 +5241,13 @@ S_reghopmaybe3(U8* s, I32 off, const U8* lim) return NULL; } else { - while (off++) { - if (s > lim) { - s--; - if (UTF8_IS_CONTINUED(*s)) { - while (s > (U8*)lim && UTF8_IS_CONTINUATION(*s)) - s--; - } - /* XXX could check well-formedness here */ + while (off++ && s > lim) { + s--; + if (UTF8_IS_CONTINUED(*s)) { + while (s > lim && UTF8_IS_CONTINUATION(*s)) + s--; } - else - break; + /* XXX could check well-formedness here */ } if (off <= 0) return NULL;