X-Git-Url: https://perl5.git.perl.org/perl5.git/blobdiff_plain/92e82afa16f5f1aa1b3e163f6d4656d14c44a4d2..2f131c9612f9b25cb3f97870dce95aad0057d181:/regexec.c diff --git a/regexec.c b/regexec.c index eb5e346..ced20bd 100644 --- a/regexec.c +++ b/regexec.c @@ -122,12 +122,26 @@ #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim)) /* these are unrolled below in the CCC_TRY_XXX defined */ -#define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \ - if (!CAT2(PL_utf8_,class)) { bool ok; ENTER; save_re_context(); ok=CAT2(is_utf8_,class)((const U8*)str); assert(ok); LEAVE; } } STMT_END +#ifdef EBCDIC + /* Often 'str' is a hard-coded utf8 string instead of utfebcdic. so just + * skip the check on EBCDIC platforms */ +# define LOAD_UTF8_CHARCLASS(class,str) LOAD_UTF8_CHARCLASS_NO_CHECK(class) +#else +# define LOAD_UTF8_CHARCLASS(class,str) STMT_START { \ + if (!CAT2(PL_utf8_,class)) { \ + bool ok; \ + ENTER; save_re_context(); \ + ok=CAT2(is_utf8_,class)((const U8*)str); \ + assert(ok); assert(CAT2(PL_utf8_,class)); LEAVE; } } STMT_END +#endif /* Doesn't do an assert to verify that is correct */ #define LOAD_UTF8_CHARCLASS_NO_CHECK(class) STMT_START { \ - if (!CAT2(PL_utf8_,class)) { bool throw_away; ENTER; save_re_context(); throw_away = CAT2(is_utf8_,class)((const U8*)" "); LEAVE; } } STMT_END + if (!CAT2(PL_utf8_,class)) { \ + bool throw_away PERL_UNUSED_DECL; \ + ENTER; save_re_context(); \ + throw_away = CAT2(is_utf8_,class)((const U8*)" "); \ + LEAVE; } } STMT_END #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS(alnum,"a") #define LOAD_UTF8_CHARCLASS_DIGIT() LOAD_UTF8_CHARCLASS(digit,"0") @@ -684,11 +698,26 @@ Perl_re_intuit_start(pTHX_ REGEXP * const rx, SV *sv, char *strpos, (IV)prog->check_end_shift); }); - if (flags & REXEC_SCREAM) { + if ((flags & REXEC_SCREAM) && SvSCREAM(sv)) { I32 p = -1; /* Internal iterator of scream. */ I32 * const pp = data ? data->scream_pos : &p; + const MAGIC *mg; + bool found = FALSE; + + assert(SvMAGICAL(sv)); + mg = mg_find(sv, PERL_MAGIC_study); + assert(mg); + + if (mg->mg_private == 1) { + found = ((U8 *)mg->mg_ptr)[BmRARE(check)] != (U8)~0; + } else if (mg->mg_private == 2) { + found = ((U16 *)mg->mg_ptr)[BmRARE(check)] != (U16)~0; + } else { + assert (mg->mg_private == 4); + found = ((U32 *)mg->mg_ptr)[BmRARE(check)] != (U32)~0; + } - if (PL_screamfirst[BmRARE(check)] >= 0 + if (found || ( BmRARE(check) == '\n' && (BmPREVIOUS(check) == SvCUR(check) - 1) && SvTAIL(check) )) @@ -1171,8 +1200,8 @@ uvc, charid, foldlen, foldbuf, uniflags) STMT_START { \ uscan += len; \ len=0; \ } else { \ - uvc = utf8n_to_uvuni( (U8*)uc, UTF8_MAXLEN, &len, uniflags ); \ - uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \ + uvc = to_utf8_fold( (U8 *) uc, foldbuf, &foldlen ); \ + len = UTF8SKIP(uc); \ foldlen -= UNISKIP( uvc ); \ uscan = foldbuf + UNISKIP( uvc ); \ } \ @@ -1434,23 +1463,11 @@ S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s, folder = foldEQ_latin1; /* /a, except the sharp s one which */ goto do_exactf_non_utf8; /* isn't dealt with by these */ - case EXACTFU: - if (UTF_PATTERN || utf8_target) { - utf8_fold_flags = 0; - goto do_exactf_utf8; - } - fold_array = PL_fold_latin1; - folder = foldEQ_latin1; - /* XXX This uses the full utf8 fold because if the pattern contains - * 'ss' it could match LATIN_SMALL_LETTER SHARP_S in the string. - * There could be a new node type, say EXACTFU_SS, which is - * generated by regcomp only if there is an 'ss', and then every - * other case could goto do_exactf_non_utf8;*/ - goto do_exactf_utf8; - case EXACTF: if (UTF_PATTERN || utf8_target) { - utf8_fold_flags = 0; + + /* regcomp.c already folded this if pattern is in UTF-8 */ + utf8_fold_flags = (UTF_PATTERN) ? FOLDEQ_S2_ALREADY_FOLDED : 0; goto do_exactf_utf8; } fold_array = PL_fold; @@ -1464,6 +1481,19 @@ S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s, } fold_array = PL_fold_locale; folder = foldEQ_locale; + goto do_exactf_non_utf8; + + case EXACTFU: + if (UTF_PATTERN || utf8_target) { + utf8_fold_flags = (UTF_PATTERN) ? FOLDEQ_S2_ALREADY_FOLDED : 0; + goto do_exactf_utf8; + } + + /* Any 'ss' in the pattern should have been replaced by regcomp, + * so we don't have to worry here about this single special case + * in the Latin1 range */ + fold_array = PL_fold_latin1; + folder = foldEQ_latin1; /* FALL THROUGH */ @@ -1478,6 +1508,11 @@ S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s, pat_string = STRING(c); ln = STR_LEN(c); /* length to match in octets/bytes */ + /* We know that we have to match at least 'ln' bytes (which is the + * same as characters, since not utf8). If we have to match 3 + * characters, and there are only 2 availabe, we know without + * trying that it will fail; so don't start a match past the + * required minimum number from the far end */ e = HOP3c(strend, -((I32)ln), s); if (!reginfo && e < s) { @@ -1495,6 +1530,9 @@ S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s, break; do_exactf_utf8: + { + unsigned expansion; + /* If one of the operands is in utf8, we can't use the simpler * folding above, due to the fact that many different characters @@ -1507,12 +1545,35 @@ S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s, ? utf8_length((U8 *) pat_string, (U8 *) pat_end) : ln; + /* We have 'lnc' characters to match in the pattern, but because of + * multi-character folding, each character in the target can match + * up to 3 characters (Unicode guarantees it will never exceed + * this) if it is utf8-encoded; and up to 2 if not (based on the + * fact that the Latin 1 folds are already determined, and the + * only multi-char fold in that range is the sharp-s folding to + * 'ss'. Thus, a pattern character can match as little as 1/3 of a + * string character. Adjust lnc accordingly, rounding up, so that + * if we need to match at least 4+1/3 chars, that really is 5. */ + expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2; + lnc = (lnc + expansion - 1) / expansion; + + /* As in the non-UTF8 case, if we have to match 3 characters, and + * only 2 are left, it's guaranteed to fail, so don't start a + * match that would require us to go beyond the end of the string + */ e = HOP3c(strend, -((I32)lnc), s); if (!reginfo && e < s) { e = s; /* Due to minlen logic of intuit() */ } + /* XXX Note that we could recalculate e to stop the loop earlier, + * as the worst case expansion above will rarely be met, and as we + * go along we would usually find that e moves further to the left. + * This would happen only after we reached the point in the loop + * where if there were no expansion we should fail. Unclear if + * worth the expense */ + while (s <= e) { char *my_strend= (char *)strend; if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target, @@ -1521,9 +1582,10 @@ S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s, { goto got_it; } - s += UTF8SKIP(s); + s += (utf8_target) ? UTF8SKIP(s) : 1; } break; + } case BOUNDL: PL_reg_flags |= RF_tainted; FBC_BOUND(isALNUM_LC, @@ -1594,7 +1656,7 @@ S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s, case NALNUMU: REXEC_FBC_CSCAN_PRELOAD( LOAD_UTF8_CHARCLASS_ALNUM(), - swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target), + !swash_fetch(PL_utf8_alnum,(U8*)s, utf8_target), ! isWORDCHAR_L1((U8) *s) ); break; @@ -2282,7 +2344,7 @@ Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, register char *stre dontbother = end_shift; strend = HOPc(strend, -dontbother); while ( (s <= last) && - ((flags & REXEC_SCREAM) + ((flags & REXEC_SCREAM) && SvSCREAM(sv) ? (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, (back_min<0 ? strbeg : strend)), @@ -2361,7 +2423,7 @@ Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, register char *stre utf8_target ? to_utf8_substr(prog) : to_byte_substr(prog); float_real = utf8_target ? prog->float_utf8 : prog->float_substr; - if (flags & REXEC_SCREAM) { + if ((flags & REXEC_SCREAM) && SvSCREAM(sv)) { last = screaminstr(sv, float_real, s - strbeg, end_shift, &scream_pos, 1); /* last one */ if (!last) @@ -2555,7 +2617,7 @@ S_regtry(pTHX_ regmatch_info *reginfo, char **startpos) /* This is safe against NULLs: */ ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); /* PM_reg_curpm owns a reference to this regexp. */ - ReREFCNT_inc(rx); + (void)ReREFCNT_inc(rx); #endif PM_SETRE(PL_reg_curpm, rx); PL_reg_oldcurpm = PL_curpm; @@ -3601,7 +3663,7 @@ S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog) case EXACTFU: folder = foldEQ_latin1; fold_array = PL_fold_latin1; - fold_utf8_flags = 0; + fold_utf8_flags = (UTF_PATTERN) ? FOLDEQ_S1_ALREADY_FOLDED : 0; goto do_exactf; case EXACTFA: @@ -3613,7 +3675,7 @@ S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog) case EXACTF: folder = foldEQ; fold_array = PL_fold; - fold_utf8_flags = 0; + fold_utf8_flags = (UTF_PATTERN) ? FOLDEQ_S1_ALREADY_FOLDED : 0; do_exactf: s = STRING(scan); @@ -3625,20 +3687,9 @@ S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog) char *e = PL_regeol; if (! foldEQ_utf8_flags(s, 0, ln, cBOOL(UTF_PATTERN), - l, &e, 0, utf8_target, fold_utf8_flags)) { - /* One more case for the sharp s: - * pack("U0U*", 0xDF) =~ /ss/i, - * the 0xC3 0x9F are the UTF-8 - * byte sequence for the U+00DF. */ - - if (!(utf8_target && - toLOWER(s[0]) == 's' && - ln >= 2 && - toLOWER(s[1]) == 's' && - (U8)l[0] == 0xC3 && - e - l >= 2 && - (U8)l[1] == 0x9F)) - sayNO; + l, &e, 0, utf8_target, fold_utf8_flags)) + { + sayNO; } locinput = e; nextchr = UCHARAT(locinput); @@ -3674,7 +3725,10 @@ S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog) case NBOUNDU: case NBOUNDA: /* was last char in word? */ - if (utf8_target && FLAGS(scan) != REGEX_ASCII_RESTRICTED_CHARSET) { + if (utf8_target + && FLAGS(scan) != REGEX_ASCII_RESTRICTED_CHARSET + && FLAGS(scan) != REGEX_ASCII_MORE_RESTRICTED_CHARSET) + { if (locinput == PL_bostr) ln = '\n'; else { @@ -3721,6 +3775,7 @@ S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog) n = isALNUM(nextchr); break; case REGEX_ASCII_RESTRICTED_CHARSET: + case REGEX_ASCII_MORE_RESTRICTED_CHARSET: ln = isWORDCHAR_A(ln); n = isWORDCHAR_A(nextchr); break; @@ -3820,17 +3875,17 @@ S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog) L* (L | LVT T* | V V* T* | LV V* T*) That means that if we have seen any L's at all we can quit - there, but if the next character is a LVT, a V or and LV we + there, but if the next character is an LVT, a V, or an LV we should keep going. There is a subtlety with Prepend* which showed up in testing. Note that the Begin, and only the Begin is required in: | Prepend* Begin Extend* - Also, Begin contains '! Control'. A Prepend must be a '! - Control', which means it must be a Begin. What it comes down to - is that if we match Prepend* and then find no suitable Begin - afterwards, that if we backtrack the last Prepend, that one will - be a suitable Begin. + Also, Begin contains '! Control'. A Prepend must be a + '! Control', which means it must also be a Begin. What it + comes down to is that if we match Prepend* and then find no + suitable Begin afterwards, that if we backtrack the last + Prepend, that one will be a suitable Begin. */ if (locinput >= PL_regeol) @@ -3840,7 +3895,8 @@ S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog) /* Match either CR LF or '.', as all the other possibilities * require utf8 */ locinput++; /* Match the . or CR */ - if (nextchr == '\r' + if (nextchr == '\r' /* And if it was CR, and the next is LF, + match the LF */ && locinput < PL_regeol && UCHARAT(locinput) == '\n') locinput++; } @@ -4202,6 +4258,12 @@ S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog) PL_op = (OP_4tree*)rexi->data->data[n]; DEBUG_STATE_r( PerlIO_printf(Perl_debug_log, " re_eval 0x%"UVxf"\n", PTR2UV(PL_op)) ); + /* wrap the call in two SAVECOMPPADs. This ensures that + * when the save stack is eventually unwound, all the + * accumulated SAVEt_CLEARSV's will be processed with + * interspersed SAVEt_COMPPAD's to ensure that lexicals + * are cleared in the right pad */ + SAVECOMPPAD(); PAD_SAVE_LOCAL(old_comppad, (PAD*)rexi->data->data[n + 2]); PL_regoffs[0].end = PL_reg_magic->mg_len = locinput - PL_bostr; @@ -4222,6 +4284,7 @@ S_regmatch(pTHX_ regmatch_info *reginfo, regnode *prog) Copy(&saved_state, &PL_reg_state, 1, struct re_save_state); PL_op = oop; + SAVECOMPPAD(); PAD_RESTORE_LOCAL(old_comppad); PL_curcop = ocurcop; PL_regeol = saved_regeol; @@ -5412,7 +5475,7 @@ NULL rex = (struct regexp *)SvANY(rex_sv); rexi = RXi_GET(rex); cur_curlyx = cur_eval->u.eval.prev_curlyx; - ReREFCNT_inc(rex_sv); + (void)ReREFCNT_inc(rex_sv); st->u.eval.cp = regcppush(0); /* Save *all* the positions. */ /* rex was changed so update the pointer in PL_reglastparen and PL_reglastcloseparen */ @@ -5660,6 +5723,8 @@ NULL #define CASE_CLASS(nAmE) \ case nAmE: \ + if (locinput >= PL_regeol) \ + sayNO; \ if ((n=is_##nAmE(locinput,utf8_target))) { \ locinput += n; \ nextchr = UCHARAT(locinput); \ @@ -5667,6 +5732,8 @@ NULL sayNO; \ break; \ case N##nAmE: \ + if (locinput >= PL_regeol) \ + sayNO; \ if ((n=is_##nAmE(locinput,utf8_target))) { \ sayNO; \ } else { \ @@ -5941,7 +6008,7 @@ S_regrepeat(pTHX_ const regexp *prog, const regnode *p, I32 max, int depth) /* Here, the string is utf8, and the pattern char is different * in utf8 than not, so can't compare them directly. Outside the - * loop, find find the two utf8 bytes that represent c, and then + * loop, find the two utf8 bytes that represent c, and then * look for those in sequence in the utf8 string */ U8 high = UTF8_TWO_BYTE_HI(c); U8 low = UTF8_TWO_BYTE_LO(c); @@ -5968,7 +6035,7 @@ S_regrepeat(pTHX_ const regexp *prog, const regnode *p, I32 max, int depth) case EXACTF: case EXACTFU: - utf8_flags = 0; + utf8_flags = (UTF_PATTERN) ? FOLDEQ_S2_ALREADY_FOLDED : 0; /* The comments for the EXACT case above apply as well to these fold * ones */ @@ -6628,31 +6695,19 @@ S_reginclass(pTHX_ const regexp * const prog, register const regnode * const n, else if (flags & ANYOF_LOC_NONBITMAP_FOLD) { /* Here, we need to test if the fold of the target string - * matches. In the case of a multi-char fold that is - * caught by regcomp.c, it has stored all such folds into - * 'av'; we linearly check to see if any match the target - * string (folded). We know that the originals were each - * one character, but we don't currently know how many - * characters/bytes each folded to, except we do know that - * there are small limits imposed by Unicode. XXX A - * performance enhancement would be to have regcomp.c store - * the max number of chars/bytes that are in an av entry, - * as, say the 0th element. Even better would be to have a - * hash of the few characters that can start a multi-char - * fold to the max number of chars of those folds. - * - * Further down, if there isn't a - * match in the av, we will check if there is another - * fold-type match. For that, we also need the fold, but - * only the first character. No sense in folding it twice, - * so we do it here, even if there isn't any multi-char - * fold, so we always fold at least the first character. - * If the node is a straight ANYOF node, or there is only - * one character available in the string, or if there isn't - * any av, that's all we have to fold. In the case of a - * multi-char fold, we do have guarantees in Unicode that - * it can only expand up to so many characters and so many - * bytes. We keep track so don't exceed either. + * matches. The non-multi char folds have all been moved to + * the compilation phase, and the multi-char folds have + * been stored by regcomp into 'av'; we linearly check to + * see if any match the target string (folded). We know + * that the originals were each one character, but we don't + * currently know how many characters/bytes each folded to, + * except we do know that there are small limits imposed by + * Unicode. XXX A performance enhancement would be to have + * regcomp.c store the max number of chars/bytes that are + * in an av entry, as, say the 0th element. Even better + * would be to have a hash of the few characters that can + * start a multi-char fold to the max number of chars of + * those folds. * * If there is a match, we will need to advance (if lenp is * specified) the match pointer in the target string. But @@ -6662,28 +6717,34 @@ S_reginclass(pTHX_ const regexp * const prog, register const regnode * const n, * create a map so that we know how many bytes in the * source to advance given that we have matched a certain * number of bytes in the fold. This map is stored in - * 'map_fold_len_back'. The first character in the fold - * has array element 1 contain the number of bytes in the - * source that folded to it; the 2nd is the cumulative - * number to match it; ... */ - U8 map_fold_len_back[UTF8_MAX_FOLD_CHAR_EXPAND+1] = { 0 }; + * 'map_fold_len_back'. Let n mean the number of bytes in + * the fold of the first character that we are folding. + * Then map_fold_len_back[n] is set to the number of bytes + * in that first character. Similarly let m be the + * corresponding number for the second character to be + * folded. Then map_fold_len_back[n+m] is set to the + * number of bytes occupied by the first two source + * characters. ... */ + U8 map_fold_len_back[UTF8_MAXBYTES_CASE+1] = { 0 }; U8 folded[UTF8_MAXBYTES_CASE+1]; STRLEN foldlen = 0; /* num bytes in fold of 1st char */ - STRLEN foldlen_for_av; /* num bytes in fold of all chars */ + STRLEN total_foldlen = 0; /* num bytes in fold of all + chars */ if (OP(n) == ANYOF || maxlen == 1 || ! lenp || ! av) { /* Here, only need to fold the first char of the target - * string */ + * string. It the source wasn't utf8, is 1 byte long */ to_utf8_fold(utf8_p, folded, &foldlen); - foldlen_for_av = foldlen; - map_fold_len_back[1] = UTF8SKIP(utf8_p); + total_foldlen = foldlen; + map_fold_len_back[foldlen] = (utf8_target) + ? UTF8SKIP(utf8_p) + : 1; } else { /* Here, need to fold more than the first char. Do so * up to the limits */ - UV which_char = 0; U8* source_ptr = utf8_p; /* The source for the fold is the regex target string */ @@ -6691,8 +6752,10 @@ S_reginclass(pTHX_ const regexp * const prog, register const regnode * const n, U8* e = utf8_p + maxlen; /* Can't go beyond last available byte in the target string */ - while (which_char < UTF8_MAX_FOLD_CHAR_EXPAND - && source_ptr < e) + U8 i; + for (i = 0; + i < UTF8_MAX_FOLD_CHAR_EXPAND && source_ptr < e; + i++) { /* Fold the next character */ @@ -6710,50 +6773,52 @@ S_reginclass(pTHX_ const regexp * const prog, register const regnode * const n, break; } - /* Save the first character's folded length, in - * case we have to use it later */ - if (! foldlen) { - foldlen = this_char_foldlen; - } - - /* Here, add the fold of this character */ + /* Add the fold of this character */ Copy(this_char_folded, folded_ptr, this_char_foldlen, U8); - which_char++; - map_fold_len_back[which_char] = - map_fold_len_back[which_char - 1] - + UTF8SKIP(source_ptr); - folded_ptr += this_char_foldlen; source_ptr += UTF8SKIP(source_ptr); + folded_ptr += this_char_foldlen; + total_foldlen = folded_ptr - folded; + + /* Create map from the number of bytes in the fold + * back to the number of bytes in the source. If + * the source isn't utf8, the byte count is just + * the number of characters so far */ + map_fold_len_back[total_foldlen] + = (utf8_target) + ? source_ptr - utf8_p + : i + 1; } *folded_ptr = '\0'; - foldlen_for_av = folded_ptr - folded; } /* Do the linear search to see if the fold is in the list - * of multi-char folds. (Useless to look if won't be able - * to store that it is a multi-char fold in *lenp) */ - if (lenp && av) { + * of multi-char folds. */ + if (av) { I32 i; for (i = 0; i <= av_len(av); i++) { SV* const sv = *av_fetch(av, i, FALSE); STRLEN len; const char * const s = SvPV_const(sv, len); - if (len <= foldlen_for_av && memEQ(s, - (char*)folded, - len)) + + if (len <= total_foldlen + && memEQ(s, (char*)folded, len) + + /* If 0, means matched a partial char. See + * [perl #90536] */ + && map_fold_len_back[len]) { /* Advance the target string ptr to account for * this fold, but have to translate from the * folded length to the corresponding source - * length. The array is indexed by how many - * characters in the match */ - *lenp = map_fold_len_back[ - utf8_length(folded, folded + len)]; + * length. */ + if (lenp) { + *lenp = map_fold_len_back[len]; + } match = TRUE; break; } @@ -6773,6 +6838,10 @@ S_reginclass(pTHX_ const regexp * const prog, register const regnode * const n, STATIC U8 * S_reghop3(U8 *s, I32 off, const U8* lim) { + /* return the position 'off' UTF-8 characters away from 's', forward if + * 'off' >= 0, backwards if negative. But don't go outside of position + * 'lim', which better be < s if off < 0 */ + dVAR; PERL_ARGS_ASSERT_REGHOP3; @@ -6892,16 +6961,16 @@ S_to_utf8_substr(pTHX_ register regexp *prog) prog->substrs->data[i].utf8_substr = sv; sv_utf8_upgrade(sv); if (SvVALID(prog->substrs->data[i].substr)) { - const U8 flags = BmFLAGS(prog->substrs->data[i].substr); - if (flags & FBMcf_TAIL) { + if (SvTAIL(prog->substrs->data[i].substr)) { /* Trim the trailing \n that fbm_compile added last time. */ SvCUR_set(sv, SvCUR(sv) - 1); /* Whilst this makes the SV technically "invalid" (as its buffer is no longer followed by "\0") when fbm_compile() adds the "\n" back, a "\0" is restored. */ - } - fbm_compile(sv, flags); + fbm_compile(sv, FBMcf_TAIL); + } else + fbm_compile(sv, 0); } if (prog->substrs->data[i].substr == prog->check_substr) prog->check_utf8 = sv; @@ -6923,15 +6992,14 @@ S_to_byte_substr(pTHX_ register regexp *prog) SV* sv = newSVsv(prog->substrs->data[i].utf8_substr); if (sv_utf8_downgrade(sv, TRUE)) { if (SvVALID(prog->substrs->data[i].utf8_substr)) { - const U8 flags - = BmFLAGS(prog->substrs->data[i].utf8_substr); - if (flags & FBMcf_TAIL) { + if (SvTAIL(prog->substrs->data[i].utf8_substr)) { /* Trim the trailing \n that fbm_compile added last time. */ SvCUR_set(sv, SvCUR(sv) - 1); - } - fbm_compile(sv, flags); - } + fbm_compile(sv, FBMcf_TAIL); + } else + fbm_compile(sv, 0); + } } else { SvREFCNT_dec(sv); sv = &PL_sv_undef;