Mercurial > vim
changeset 4801:3cd3cc1e9119 v7.3.1147
updated for version 7.3.1147
Problem: New regexp engine: regstart is only used to find the first match.
Solution: Use regstart whenever adding the start state.
| author | Bram Moolenaar <bram@vim.org> |
|---|---|
| date | Sat, 08 Jun 2013 14:38:27 +0200 |
| parents | 339cf2070eb7 |
| children | 8adbe7dae7fd |
| files | src/regexp_nfa.c src/version.c |
| diffstat | 2 files changed, 74 insertions(+), 21 deletions(-) [+] |
line wrap: on
line diff
--- a/src/regexp_nfa.c +++ b/src/regexp_nfa.c @@ -4153,6 +4153,7 @@ recursive_regmatch(state, prog, submatch } static int failure_chance __ARGS((nfa_state_T *state, int depth)); +static int skip_to_start __ARGS((int c, colnr_T *colp)); /* * Estimate the chance of a match with "state" failing. @@ -4305,6 +4306,31 @@ failure_chance(state, depth) } /* + * Skip until the char "c" we know a match must start with. + */ + static int +skip_to_start(c, colp) + int c; + colnr_T *colp; +{ + char_u *s; + + /* Used often, do some work to avoid call overhead. */ + if (!ireg_ic +#ifdef FEAT_MBYTE + && !has_mbyte +#endif + ) + s = vim_strbyte(regline + *colp, c); + else + s = cstrchr(regline + *colp, c); + if (s == NULL) + return FAIL; + *colp = (int)(s - regline); + return OK; +} + +/* * Main matching routine. * * Run NFA to determine whether it matches reginput. @@ -5449,12 +5475,50 @@ nfa_regmatch(prog, start, submatch, m) * the first MOPEN. */ if (toplevel) { - if (REG_MULTI) - m->norm.list.multi[0].start.col = + int add = TRUE; + int c; + + if (prog->regstart != NUL && clen != 0) + { + if (nextlist->n == 0) + { + colnr_T col = (colnr_T)(reginput - regline) + clen; + + /* Nextlist is empty, we can skip ahead to the + * character that must appear at the start. */ + if (skip_to_start(prog->regstart, &col) == FAIL) + break; +#ifdef ENABLE_LOG + fprintf(log_fd, " Skipping ahead %d bytes to regstart\n", + col - ((colnr_T)(reginput - regline) + clen)); +#endif + reginput = regline + col - clen; + } + else + { + /* Checking if the required start character matches is + * cheaper than adding a state that won't match. */ + c = PTR2CHAR(reginput + clen); + if (c != prog->regstart && (!ireg_ic || MB_TOLOWER(c) + != MB_TOLOWER(prog->regstart))) + { +#ifdef ENABLE_LOG + fprintf(log_fd, " Skipping start state, regstart does not match\n"); +#endif + add = FALSE; + } + } + } + + if (add) + { + if (REG_MULTI) + m->norm.list.multi[0].start.col = (colnr_T)(reginput - regline) + clen; - else - m->norm.list.line[0].start = reginput + clen; - addstate(nextlist, start->out, m, clen); + else + m->norm.list.line[0].start = reginput + clen; + addstate(nextlist, start->out, m, clen); + } } else addstate(nextlist, start, m, clen); @@ -5701,23 +5765,10 @@ nfa_regexec_both(line, startcol) return 0L; if (prog->regstart != NUL) - { - char_u *s; - - /* Skip until the char we know it must start with. - * Used often, do some work to avoid call overhead. */ - if (!ireg_ic -#ifdef FEAT_MBYTE - && !has_mbyte -#endif - ) - s = vim_strbyte(regline + col, prog->regstart); - else - s = cstrchr(regline + col, prog->regstart); - if (s == NULL) + /* Skip ahead until a character we know the match must start with. + * When there is none there is no match. */ + if (skip_to_start(prog->regstart, &col) == FAIL) return 0L; - col = (int)(s - regline); - } /* If the start column is past the maximum column: no need to try. */ if (ireg_maxcol > 0 && col >= ireg_maxcol)