2 * Copyright (c) 1992, 1993, 1994
3 * The Regents of the University of California. All rights reserved.
4 * Copyright (c) 1992, 1993, 1994, 1995, 1996
5 * Keith Bostic. All rights reserved.
7 * See the LICENSE file for redistribution information.
13 static const char sccsid[] = "@(#)ex_subst.c 10.37 (Berkeley) 9/15/96";
16 #include <sys/types.h>
17 #include <sys/queue.h>
20 #include <bitstring.h>
29 #include "../common/common.h"
32 #define SUB_FIRST 0x01 /* The 'r' flag isn't reasonable. */
33 #define SUB_MUSTSETR 0x02 /* The 'r' flag is required. */
35 static int re_conv __P((SCR *, char **, size_t *, int *));
36 static int re_cscope_conv __P((SCR *, char **, size_t *, int *));
37 static int re_sub __P((SCR *,
38 char *, char **, size_t *, size_t *, regmatch_t [10]));
39 static int re_tag_conv __P((SCR *, char **, size_t *, int *));
40 static int s __P((SCR *, EXCMD *, char *, regex_t *, u_int));
44 * [line [,line]] s[ubstitute] [[/;]pat[/;]/repl[/;] [cgr] [count] [#lp]]
46 * Substitute on lines matching a pattern.
48 * PUBLIC: int ex_s __P((SCR *, EXCMD *));
59 char *bp, *ptrn, *rep, *p, *t;
62 * Skip leading white space.
65 * Historic vi allowed any non-alphanumeric to serve as the
66 * substitution command delimiter.
69 * If the arguments are empty, it's the same as &, i.e. we
70 * repeat the last substitution.
74 for (p = cmdp->argv[0]->bp,
75 len = cmdp->argv[0]->len; len > 0; --len, ++p) {
80 subagain: return (ex_subagain(sp, cmdp));
83 if (isalnum(delim) || delim == '\\')
84 return (s(sp, cmdp, p, &sp->subre_c, SUB_MUSTSETR));
88 * The full-blown substitute command reset the remembered
89 * state of the 'c' and 'g' suffices.
91 sp->c_suffix = sp->g_suffix = 0;
94 * Get the pattern string, toss escaping characters.
97 * Historic vi accepted any of the following forms:
99 * :s/abc/def/ change "abc" to "def"
100 * :s/abc/def change "abc" to "def"
101 * :s/abc/ delete "abc"
102 * :s/abc delete "abc"
106 * Only toss an escaping character if it escapes a delimiter.
107 * This means that "s/A/\\\\f" replaces "A" with "\\f". It
108 * would be nice to be more regular, i.e. for each layer of
109 * escaping a single escaping character is removed, but that's
110 * not how the historic vi worked.
112 for (ptrn = t = p;;) {
113 if (p[0] == '\0' || p[0] == delim) {
118 * Nul terminate the pattern string -- it's passed
119 * to regcomp which doesn't understand anything else.
127 else if (p[1] == '\\')
133 * If the pattern string is empty, use the last RE (not just the
134 * last substitution RE).
137 if (sp->re == NULL) {
138 ex_emsg(sp, NULL, EXM_NOPREVRE);
142 /* Re-compile the RE if necessary. */
143 if (!F_ISSET(sp, SC_RE_SEARCH) && re_compile(sp,
144 sp->re, sp->re_len, NULL, NULL, &sp->re_c, RE_C_SEARCH))
150 * Compile the RE. Historic practice is that substitutes set
151 * the search direction as well as both substitute and search
152 * RE's. We compile the RE twice, as we don't want to bother
153 * ref counting the pattern string and (opaque) structure.
155 if (re_compile(sp, ptrn, t - ptrn,
156 &sp->re, &sp->re_len, &sp->re_c, RE_C_SEARCH))
158 if (re_compile(sp, ptrn, t - ptrn,
159 &sp->subre, &sp->subre_len, &sp->subre_c, RE_C_SUBST))
163 sp->searchdir = FORWARD;
168 * Get the replacement string.
170 * The special character & (\& if O_MAGIC not set) matches the
171 * entire RE. No handling of & is required here, it's done by
174 * The special character ~ (\~ if O_MAGIC not set) inserts the
175 * previous replacement string into this replacement string.
176 * Count ~'s to figure out how much space we need. We could
177 * special case nonexistent last patterns or whether or not
178 * O_MAGIC is set, but it's probably not worth the effort.
182 * Only toss an escaping character if it escapes a delimiter or
183 * if O_MAGIC is set and it escapes a tilde.
186 * If the entire replacement pattern is "%", then use the last
187 * replacement pattern. This semantic was added to vi in System
188 * V and then percolated elsewhere, presumably around the time
189 * that it was added to their version of ed(1).
191 if (p[0] == '\0' || p[0] == delim) {
194 if (sp->repl != NULL)
198 } else if (p[0] == '%' && (p[1] == '\0' || p[1] == delim))
199 p += p[1] == delim ? 2 : 1;
201 for (rep = p, len = 0;
202 p[0] != '\0' && p[0] != delim; ++p, ++len)
205 GET_SPACE_RET(sp, bp, blen, len);
206 for (t = bp, len = 0, p = rep;;) {
207 if (p[0] == '\0' || p[0] == delim) {
215 else if (p[1] == '\\') {
218 } else if (p[1] == '~') {
220 if (!O_ISSET(sp, O_MAGIC))
223 } else if (p[0] == '~' && O_ISSET(sp, O_MAGIC)) {
225 memcpy(t, sp->repl, sp->repl_len);
233 if ((sp->repl_len = len) != 0) {
234 if (sp->repl != NULL)
236 if ((sp->repl = malloc(len)) == NULL) {
237 msgq(sp, M_SYSERR, NULL);
238 FREE_SPACE(sp, bp, blen);
241 memcpy(sp->repl, bp, len);
243 FREE_SPACE(sp, bp, blen);
245 return (s(sp, cmdp, p, re, flags));
250 * [line [,line]] & [cgr] [count] [#lp]]
252 * Substitute using the last substitute RE and replacement pattern.
254 * PUBLIC: int ex_subagain __P((SCR *, EXCMD *));
257 ex_subagain(sp, cmdp)
261 if (sp->subre == NULL) {
262 ex_emsg(sp, NULL, EXM_NOPREVRE);
265 if (!F_ISSET(sp, SC_RE_SUBST) && re_compile(sp,
266 sp->subre, sp->subre_len, NULL, NULL, &sp->subre_c, RE_C_SUBST))
269 cmdp, cmdp->argc ? cmdp->argv[0]->bp : NULL, &sp->subre_c, 0));
274 * [line [,line]] ~ [cgr] [count] [#lp]]
276 * Substitute using the last RE and last substitute replacement pattern.
278 * PUBLIC: int ex_subtilde __P((SCR *, EXCMD *));
281 ex_subtilde(sp, cmdp)
285 if (sp->re == NULL) {
286 ex_emsg(sp, NULL, EXM_NOPREVRE);
289 if (!F_ISSET(sp, SC_RE_SEARCH) && re_compile(sp,
290 sp->re, sp->re_len, NULL, NULL, &sp->re_c, RE_C_SEARCH))
293 cmdp, cmdp->argc ? cmdp->argv[0]->bp : NULL, &sp->re_c, 0));
298 * Do the substitution. This stuff is *really* tricky. There are lots of
299 * special cases, and general nastiness. Don't mess with it unless you're
302 * The nasty part of the substitution is what happens when the replacement
303 * string contains newlines. It's a bit tricky -- consider the information
304 * that has to be retained for "s/f\(o\)o/^M\1^M\1/". The solution here is
305 * to build a set of newline offsets which we use to break the line up later,
306 * when the replacement is done. Don't change it unless you're *damned*
309 #define NEEDNEWLINE(sp) { \
310 if (sp->newl_len == sp->newl_cnt) { \
311 sp->newl_len += 25; \
312 REALLOC(sp, sp->newl, size_t *, \
313 sp->newl_len * sizeof(size_t)); \
314 if (sp->newl == NULL) { \
321 #define BUILD(sp, l, len) { \
322 if (lbclen + (len) > lblen) { \
323 lblen += MAX(lbclen + (len), 256); \
324 REALLOC(sp, lb, char *, lblen); \
330 memcpy(lb + lbclen, l, len); \
334 #define NEEDSP(sp, len, pnt) { \
335 if (lbclen + (len) > lblen) { \
336 lblen += MAX(lbclen + (len), 256); \
337 REALLOC(sp, lb, char *, lblen); \
347 s(sp, cmdp, s, re, flags)
357 recno_t elno, lno, slno;
358 regmatch_t match[10];
359 size_t blen, cnt, last, lbclen, lblen, len, llen;
360 size_t offset, saved_offset, scno;
361 int cflag, lflag, nflag, pflag, rflag;
362 int didsub, do_eol_match, eflags, empty_ok, eval;
363 int linechanged, matched, quit, rval;
373 * Historically, the 'g' and 'c' suffices were always toggled as flags,
374 * so ":s/A/B/" was the same as ":s/A/B/ccgg". If O_EDCOMPATIBLE was
375 * not set, they were initialized to 0 for all substitute commands. If
376 * O_EDCOMPATIBLE was set, they were initialized to 0 only if the user
377 * specified substitute/replacement patterns (see ex_s()).
379 if (!O_ISSET(sp, O_EDCOMPATIBLE))
380 sp->c_suffix = sp->g_suffix = 0;
383 * Historic vi permitted the '#', 'l' and 'p' options in vi mode, but
384 * it only displayed the last change. I'd disallow them, but they are
385 * useful in combination with the [v]global commands. In the current
386 * model the problem is combining them with the 'c' flag -- the screen
387 * would have to flip back and forth between the confirm screen and the
388 * ex print screen, which would be pretty awful. We do display all
389 * changes, though, for what that's worth.
392 * Historic vi was fairly strict about the order of "options", the
393 * count, and "flags". I'm somewhat fuzzy on the difference between
394 * options and flags, anyway, so this is a simpler approach, and we
395 * just take it them in whatever order the user gives them. (The ex
396 * usage statement doesn't reflect this.)
398 cflag = lflag = nflag = pflag = rflag = 0;
401 for (lno = OOBLNO; *s != '\0'; ++s)
412 case '0': case '1': case '2': case '3': case '4':
413 case '5': case '6': case '7': case '8': case '9':
417 lno = strtoul(s, &s, 10);
418 if (*s == '\0') /* Loop increment correction. */
420 if (errno == ERANGE) {
422 msgq(sp, M_ERR, "153|Count overflow");
423 else if (lno == LONG_MIN)
424 msgq(sp, M_ERR, "154|Count underflow");
426 msgq(sp, M_SYSERR, NULL);
430 * In historic vi, the count was inclusive from the
433 cmdp->addr1.lno = cmdp->addr2.lno;
434 cmdp->addr2.lno += lno - 1;
435 if (!db_exist(sp, cmdp->addr2.lno) &&
436 db_last(sp, &cmdp->addr2.lno))
443 sp->c_suffix = !sp->c_suffix;
445 /* Ex text structure initialization. */
446 if (F_ISSET(sp, SC_EX)) {
447 memset(&tiq, 0, sizeof(TEXTH));
452 sp->g_suffix = !sp->g_suffix;
461 if (LF_ISSET(SUB_FIRST)) {
463 "155|Regular expression specified; r flag meaningless");
466 if (!F_ISSET(sp, SC_RE_SEARCH)) {
467 ex_emsg(sp, NULL, EXM_NOPREVRE);
477 if (*s != '\0' || !rflag && LF_ISSET(SUB_MUSTSETR)) {
478 usage: ex_emsg(sp, cmdp->cmd->usage, EXM_USAGE);
482 noargs: if (F_ISSET(sp, SC_VI) && sp->c_suffix && (lflag || nflag || pflag)) {
484 "156|The #, l and p flags may not be combined with the c flag in vi mode");
489 * bp: if interactive, line cache
490 * blen: if interactive, line cache length
491 * lb: build buffer pointer.
492 * lbclen: current length of built buffer.
493 * lblen; length of build buffer.
496 blen = lbclen = lblen = 0;
498 /* For each line... */
499 for (matched = quit = 0, lno = cmdp->addr1.lno,
500 elno = cmdp->addr2.lno; !quit && lno <= elno; ++lno) {
502 /* Someone's unhappy, time to stop. */
507 if (db_get(sp, lno, DBG_FATAL, &s, &llen))
511 * Make a local copy if doing confirmation -- when calling
512 * the confirm routine we're likely to lose the cached copy.
516 GET_SPACE_RET(sp, bp, blen, llen);
518 ADD_SPACE_RET(sp, bp, blen, llen);
523 /* Start searching from the beginning. */
527 /* Reset the build buffer offset. */
530 /* Reset empty match flag. */
534 * We don't want to have to do a setline if the line didn't
535 * change -- keep track of whether or not this line changed.
536 * If doing confirmations, don't want to keep setting the
537 * line if change is refused -- keep track of substitutions.
539 didsub = linechanged = 0;
541 /* New line, do an EOL match. */
544 /* It's not nul terminated, but we pretend it is. */
545 eflags = REG_STARTEND;
548 * The search area is from s + offset to the EOL.
550 * Generally, match[0].rm_so is the offset of the start
551 * of the match from the start of the search, and offset
552 * is the offset of the start of the last search.
554 nextmatch: match[0].rm_so = 0;
555 match[0].rm_eo = len;
557 /* Get the next match. */
558 eval = regexec(re, (char *)s + offset, 10, match, eflags);
561 * There wasn't a match or if there was an error, deal with
562 * it. If there was a previous match in this line, resolve
563 * the changes into the database. Otherwise, just move on.
565 if (eval == REG_NOMATCH)
568 re_error(sp, eval, re);
573 /* Only the first search can match an anchored expression. */
574 eflags |= REG_NOTBOL;
578 * It's possible to match 0-length strings -- for example, the
579 * command s;a*;X;, when matched against the string "aabb" will
580 * result in "XbXbX", i.e. the matches are "aa", the space
581 * between the b's and the space between the b's and the end of
582 * the string. There is a similar space between the beginning
583 * of the string and the a's. The rule that we use (because vi
584 * historically used it) is that any 0-length match, occurring
585 * immediately after a match, is ignored. Otherwise, the above
586 * example would have resulted in "XXbXbX". Another example is
587 * incorrectly using " *" to replace groups of spaces with one
590 * The way we do this is that if we just had a successful match,
591 * the starting offset does not skip characters, and the match
592 * is empty, ignore the match and move forward. If there's no
593 * more characters in the string, we were attempting to match
594 * after the last character, so quit.
596 if (!empty_ok && match[0].rm_so == 0 && match[0].rm_eo == 0) {
600 BUILD(sp, s + offset, 1)
606 /* Confirm change. */
609 * Set the cursor position for confirmation. Note,
610 * if we matched on a '$', the cursor may be past
613 from.lno = to.lno = lno;
614 from.cno = match[0].rm_so + offset;
615 to.cno = match[0].rm_eo + offset;
617 * Both ex and vi have to correct for a change before
618 * the first character in the line.
621 from.cno = to.cno = 0;
622 if (F_ISSET(sp, SC_VI)) {
624 * Only vi has to correct for a change after
625 * the last character in the line.
628 * It would be nice to change the vi code so
629 * that we could display a cursor past EOL.
633 if (from.cno >= llen)
638 if (vs_refresh(sp, 1))
641 vs_update(sp, msg_cat(sp,
642 "169|Confirm change? [n]", NULL), NULL);
644 if (v_event_get(sp, &ev, 0, 0))
646 switch (ev.e_event) {
654 v_event_err(sp, &ev);
658 if (ex_print(sp, cmdp, &from, &to, 0) ||
659 ex_scprint(sp, &from, &to))
661 if (ex_txt(sp, &tiq, 0, TXT_CR))
663 ev.e_c = tiq.cqh_first->lb[0];
672 BUILD(sp, s +offset, match[0].rm_eo);
675 /* Set the quit/interrupted flags. */
677 F_SET(sp->gp, G_INTERRUPTED);
680 * Resolve any changes, then return to (and
681 * exit from) the main loop.
688 * Set the cursor to the last position changed, converting
689 * from 1-based to 0-based.
692 sp->cno = match[0].rm_so;
694 /* Copy the bytes before the match into the build buffer. */
695 BUILD(sp, s + offset, match[0].rm_so);
697 /* Substitute the matching bytes. */
699 if (re_sub(sp, s + offset, &lb, &lbclen, &lblen, match))
702 /* Set the change flag so we know this line was modified. */
705 /* Move past the matched bytes. */
706 skip: offset += match[0].rm_eo;
707 len -= match[0].rm_eo;
709 /* A match cannot be followed by an empty pattern. */
713 * If doing a global change with confirmation, we have to
714 * update the screen. The basic idea is to store the line
715 * so the screen update routines can find it, and restart.
717 if (didsub && sp->c_suffix && sp->g_suffix) {
719 * The new search offset will be the end of the
722 saved_offset = lbclen;
724 /* Copy the rest of the line. */
726 BUILD(sp, s + offset, len)
728 /* Set the new offset. */
729 offset = saved_offset;
731 /* Store inserted lines, adjusting the build buffer. */
735 cnt < sp->newl_cnt; ++cnt, ++lno, ++elno) {
736 if (db_insert(sp, lno,
737 lb + last, sp->newl[cnt] - last))
739 last = sp->newl[cnt] + 1;
740 ++sp->rptlines[L_ADDED];
747 /* Store and retrieve the line. */
748 if (db_set(sp, lno, lb + last, lbclen))
750 if (db_get(sp, lno, DBG_FATAL, &s, &llen))
752 ADD_SPACE_RET(sp, bp, blen, llen)
757 /* Restart the build. */
759 BUILD(sp, s, offset);
762 * If we haven't already done the after-the-string
763 * match, do one. Set REG_NOTEOL so the '$' pattern
770 eflags |= REG_NOTEOL;
778 * If at the end of the string, do a test for the after
779 * the string match. Set REG_NOTEOL so the '$' pattern
782 if (sp->g_suffix && do_eol_match) {
785 eflags |= REG_NOTEOL;
790 endmatch: if (!linechanged)
793 /* Copy any remaining bytes into the build buffer. */
795 BUILD(sp, s + offset, len)
797 /* Store inserted lines, adjusting the build buffer. */
801 cnt < sp->newl_cnt; ++cnt, ++lno, ++elno) {
803 lno, lb + last, sp->newl[cnt] - last))
805 last = sp->newl[cnt] + 1;
806 ++sp->rptlines[L_ADDED];
812 /* Store the changed line. */
813 if (db_set(sp, lno, lb + last, lbclen))
816 /* Update changed line counter. */
817 if (sp->rptlchange != lno) {
818 sp->rptlchange = lno;
819 ++sp->rptlines[L_CHANGED];
824 * Display as necessary. Historic practice is to only
825 * display the last line of a line split into multiple
828 if (lflag || nflag || pflag) {
829 from.lno = to.lno = lno;
830 from.cno = to.cno = 0;
832 (void)ex_print(sp, cmdp, &from, &to, E_C_LIST);
834 (void)ex_print(sp, cmdp, &from, &to, E_C_HASH);
836 (void)ex_print(sp, cmdp, &from, &to, E_C_PRINT);
842 * Historically, vi attempted to leave the cursor at the same place if
843 * the substitution was done at the current cursor position. Otherwise
844 * it moved it to the first non-blank of the last line changed. There
845 * were some problems: for example, :s/$/foo/ with the cursor on the
846 * last character of the line left the cursor on the last character, or
847 * the & command with multiple occurrences of the matching string in the
848 * line usually left the cursor in a fairly random position.
850 * We try to do the same thing, with the exception that if the user is
851 * doing substitution with confirmation, we move to the last line about
852 * which the user was consulted, as opposed to the last line that they
853 * actually changed. This prevents a screen flash if the user doesn't
854 * change many of the possible lines.
856 if (!sp->c_suffix && (sp->lno != slno || sp->cno != scno)) {
858 (void)nonblank(sp, sp->lno, &sp->cno);
862 * If not in a global command, and nothing matched, say so.
863 * Else, if none of the lines displayed, put something up.
867 if (!F_ISSET(sp, SC_EX_GLOBAL)) {
868 msgq(sp, M_ERR, "157|No match found");
871 } else if (!lflag && !nflag && !pflag)
872 F_SET(cmdp, E_AUTOPRINT);
879 FREE_SPACE(sp, bp, blen);
889 * PUBLIC: int re_compile __P((SCR *,
890 * PUBLIC: char *, size_t, char **, size_t *, regex_t *, u_int));
893 re_compile(sp, ptrn, plen, ptrnp, lenp, rep, flags)
901 int reflags, replaced, rval;
906 if (!LF_ISSET(RE_C_CSCOPE | RE_C_TAG)) {
907 if (O_ISSET(sp, O_EXTENDED))
908 reflags |= REG_EXTENDED;
909 if (O_ISSET(sp, O_IGNORECASE))
910 reflags |= REG_ICASE;
911 if (O_ISSET(sp, O_ICLOWER)) {
912 for (p = ptrn, len = plen; len > 0; ++p, --len)
916 reflags |= REG_ICASE;
920 /* If we're replacing a saved value, clear the old one. */
921 if (LF_ISSET(RE_C_SEARCH) && F_ISSET(sp, SC_RE_SEARCH)) {
923 F_CLR(sp, SC_RE_SEARCH);
925 if (LF_ISSET(RE_C_SUBST) && F_ISSET(sp, SC_RE_SUBST)) {
926 regfree(&sp->subre_c);
927 F_CLR(sp, SC_RE_SUBST);
931 * If we're saving the string, it's a pattern we haven't seen before,
932 * so convert the vi-style RE's to POSIX 1003.2 RE's. Save a copy for
933 * later recompilation. Free any previously saved value.
936 if (LF_ISSET(RE_C_CSCOPE)) {
937 if (re_cscope_conv(sp, &ptrn, &plen, &replaced))
941 * Currently, the match-any-<blank> expression used in
942 * re_cscope_conv() requires extended RE's. This may
943 * not be right or safe.
945 reflags |= REG_EXTENDED;
946 } else if (LF_ISSET(RE_C_TAG)) {
947 if (re_tag_conv(sp, &ptrn, &plen, &replaced))
950 if (re_conv(sp, &ptrn, &plen, &replaced))
953 /* Discard previous pattern. */
954 if (*ptrnp != NULL) {
962 * Copy the string into allocated memory.
965 * Regcomp isn't 8-bit clean, so the pattern is nul-terminated
966 * for now. There's just no other solution.
968 MALLOC(sp, *ptrnp, char *, plen + 1);
969 if (*ptrnp != NULL) {
970 memcpy(*ptrnp, ptrn, plen);
971 (*ptrnp)[plen] = '\0';
974 /* Free up conversion-routine-allocated memory. */
976 FREE_SPACE(sp, ptrn, 0);
986 * Regcomp isn't 8-bit clean, so we just lost if the pattern
987 * contained a nul. Bummer!
989 if ((rval = regcomp(rep, ptrn, /* plen, */ reflags)) != 0) {
990 if (!LF_ISSET(RE_C_SILENT))
991 re_error(sp, rval, rep);
995 if (LF_ISSET(RE_C_SEARCH))
996 F_SET(sp, SC_RE_SEARCH);
997 if (LF_ISSET(RE_C_SUBST))
998 F_SET(sp, SC_RE_SUBST);
1005 * Convert vi's regular expressions into something that the
1006 * the POSIX 1003.2 RE functions can handle.
1008 * There are three conversions we make to make vi's RE's (specifically
1009 * the global, search, and substitute patterns) work with POSIX RE's.
1011 * 1: If O_MAGIC is not set, strip backslashes from the magic character
1012 * set (.[*~) that have them, and add them to the ones that don't.
1013 * 2: If O_MAGIC is not set, the string "\~" is replaced with the text
1014 * from the last substitute command's replacement string. If O_MAGIC
1015 * is set, it's the string "~".
1016 * 3: The pattern \<ptrn\> does "word" searches, convert it to use the
1020 * This doesn't exactly match the historic behavior of vi because we do
1021 * the ~ substitution before calling the RE engine, so magic characters
1022 * in the replacement string will be expanded by the RE engine, and they
1023 * weren't historically. It's a bug.
1026 re_conv(sp, ptrnp, plenp, replacedp)
1032 size_t blen, len, needlen;
1037 * First pass through, we figure out how much space we'll need.
1038 * We do it in two passes, on the grounds that most of the time
1039 * the user is doing a search and won't have magic characters.
1040 * That way we can skip most of the memory allocation and copies.
1043 for (p = *ptrnp, len = *plenp, needlen = 0; len > 0; ++p, --len)
1051 needlen += sizeof(RE_WSTART);
1055 needlen += sizeof(RE_WSTOP);
1058 if (!O_ISSET(sp, O_MAGIC)) {
1060 needlen += sp->repl_len;
1066 if (!O_ISSET(sp, O_MAGIC)) {
1078 if (O_ISSET(sp, O_MAGIC)) {
1080 needlen += sp->repl_len;
1086 if (!O_ISSET(sp, O_MAGIC)) {
1101 /* Get enough memory to hold the final pattern. */
1103 GET_SPACE_RET(sp, bp, blen, needlen);
1105 for (p = *ptrnp, len = *plenp, t = bp; len > 0; ++p, --len)
1113 RE_WSTART, sizeof(RE_WSTART) - 1);
1114 t += sizeof(RE_WSTART) - 1;
1118 RE_WSTOP, sizeof(RE_WSTOP) - 1);
1119 t += sizeof(RE_WSTOP) - 1;
1122 if (O_ISSET(sp, O_MAGIC))
1126 sp->repl, sp->repl_len);
1133 if (O_ISSET(sp, O_MAGIC))
1145 if (O_ISSET(sp, O_MAGIC)) {
1146 memcpy(t, sp->repl, sp->repl_len);
1154 if (!O_ISSET(sp, O_MAGIC))
1170 * Convert a tags search path into something that the POSIX
1171 * 1003.2 RE functions can handle.
1174 re_tag_conv(sp, ptrnp, plenp, replacedp)
1186 /* Max memory usage is 2 times the length of the string. */
1188 GET_SPACE_RET(sp, bp, blen, len * 2);
1193 /* If the last character is a '/' or '?', we just strip it. */
1194 if (len > 0 && (p[len - 1] == '/' || p[len - 1] == '?'))
1197 /* If the next-to-last or last character is a '$', it's magic. */
1198 if (len > 0 && p[len - 1] == '$') {
1204 /* If the first character is a '/' or '?', we just strip it. */
1205 if (len > 0 && (p[0] == '/' || p[0] == '?')) {
1210 /* If the first or second character is a '^', it's magic. */
1217 * Escape every other magic character we can find, meanwhile stripping
1218 * the backslashes ctags inserts when escaping the search delimiter
1221 for (; len > 0; --len) {
1222 if (p[0] == '\\' && (p[1] == '/' || p[1] == '?')) {
1225 } else if (strchr("^.[]$*", p[0]))
1239 * Convert a cscope search path into something that the POSIX
1240 * 1003.2 RE functions can handle.
1243 re_cscope_conv(sp, ptrnp, plenp, replacedp)
1249 size_t blen, len, nspaces;
1253 * Each space in the source line printed by cscope represents an
1254 * arbitrary sequence of spaces, tabs, and comments.
1256 #define CSCOPE_RE_SPACE "([ \t]|/\\*([^*]|\\*/)*\\*/)*"
1257 for (nspaces = 0, p = *ptrnp, len = *plenp; len > 0; ++p, --len)
1262 * Allocate plenty of space:
1263 * the string, plus potential escaping characters;
1264 * nspaces + 2 copies of CSCOPE_RE_SPACE;
1265 * ^, $, nul terminator characters.
1268 len = (p - *ptrnp) * 2 + (nspaces + 2) * sizeof(CSCOPE_RE_SPACE) + 3;
1269 GET_SPACE_RET(sp, bp, blen, len);
1275 memcpy(t, CSCOPE_RE_SPACE, sizeof(CSCOPE_RE_SPACE) - 1);
1276 t += sizeof(CSCOPE_RE_SPACE) - 1;
1278 for (len = *plenp; len > 0; ++p, --len)
1280 memcpy(t, CSCOPE_RE_SPACE, sizeof(CSCOPE_RE_SPACE) - 1);
1281 t += sizeof(CSCOPE_RE_SPACE) - 1;
1283 if (strchr("\\^.[]$*+?()|{}", *p))
1288 memcpy(t, CSCOPE_RE_SPACE, sizeof(CSCOPE_RE_SPACE) - 1);
1289 t += sizeof(CSCOPE_RE_SPACE) - 1;
1299 * Report a regular expression error.
1301 * PUBLIC: void re_error __P((SCR *, int, regex_t *));
1304 re_error(sp, errcode, preg)
1312 s = regerror(errcode, preg, "", 0);
1313 if ((oe = malloc(s)) == NULL)
1314 msgq(sp, M_SYSERR, NULL);
1316 (void)regerror(errcode, preg, oe, s);
1317 msgq(sp, M_ERR, "RE error: %s", oe);
1324 * Do the substitution for a regular expression.
1327 re_sub(sp, ip, lbp, lbclenp, lblenp, match)
1329 char *ip; /* Input line. */
1331 size_t *lbclenp, *lblenp;
1332 regmatch_t match[10];
1334 enum { C_NOTSET, C_LOWER, C_ONELOWER, C_ONEUPPER, C_UPPER } conv;
1335 size_t lbclen, lblen; /* Local copies. */
1336 size_t mlen; /* Match length. */
1337 size_t rpl; /* Remaining replacement length. */
1338 char *rp; /* Replacement pointer. */
1340 int no; /* Match replacement offset. */
1341 char *p, *t; /* Buffer pointers. */
1342 char *lb; /* Local copies. */
1344 lb = *lbp; /* Get local copies. */
1351 * There are some special sequences that vi provides in the
1352 * replacement patterns.
1353 * & string the RE matched (\& if nomagic set)
1354 * \# n-th regular subexpression
1355 * \E end \U, \L conversion
1356 * \e end \U, \L conversion
1357 * \l convert the next character to lower-case
1358 * \L convert to lower-case, until \E, \e, or end of replacement
1359 * \u convert the next character to upper-case
1360 * \U convert to upper-case, until \E, \e, or end of replacement
1362 * Otherwise, since this is the lowest level of replacement, discard
1363 * all escaping characters. This (hopefully) matches historic practice.
1365 #define OUTCH(ch, nltrans) { \
1366 CHAR_T __ch = (ch); \
1367 u_int __value = KEY_VAL(sp, __ch); \
1368 if (nltrans && (__value == K_CR || __value == K_NL)) { \
1370 sp->newl[sp->newl_cnt++] = lbclen; \
1371 } else if (conv != C_NOTSET) { \
1377 if (isupper(__ch)) \
1378 __ch = tolower(__ch); \
1384 if (islower(__ch)) \
1385 __ch = toupper(__ch); \
1396 for (rp = sp->repl, rpl = sp->repl_len, p = lb + lbclen; rpl--;) {
1397 switch (ch = *rp++) {
1399 if (O_ISSET(sp, O_MAGIC)) {
1411 if (!O_ISSET(sp, O_MAGIC)) {
1416 case '0': case '1': case '2': case '3': case '4':
1417 case '5': case '6': case '7': case '8': case '9':
1419 subzero: if (match[no].rm_so == -1 ||
1420 match[no].rm_eo == -1)
1422 mlen = match[no].rm_eo - match[no].rm_so;
1423 for (t = ip + match[no].rm_so; mlen--; ++t)
1455 *lbp = lb; /* Update caller's information. */