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
[] = "$Id: ex_subst.c,v 10.46 2000/09/02 13:14:15 skimo Exp $ (Berkeley) $Date: 2000/09/02 13:14:15 $";
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_T
**, size_t *, int *));
36 static int re_cscope_conv
__P((SCR
*, CHAR_T
**, size_t *, int *));
37 static int re_sub
__P((SCR
*,
38 CHAR_T
*, CHAR_T
**, size_t *, size_t *, regmatch_t
[10]));
39 static int re_tag_conv
__P((SCR
*, CHAR_T
**, size_t *, int *));
40 static int s
__P((SCR
*, EXCMD
*, CHAR_T
*, 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_T
*bp
, *p
, *ptrn
, *rep
, *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
) &&
144 re_compile(sp
, sp
->re
, sp
->re_len
,
145 NULL
, NULL
, &sp
->re_c
, SEARCH_CSEARCH
| SEARCH_MSG
))
151 * Compile the RE. Historic practice is that substitutes set
152 * the search direction as well as both substitute and search
153 * RE's. We compile the RE twice, as we don't want to bother
154 * ref counting the pattern string and (opaque) structure.
156 if (re_compile(sp
, ptrn
, t
- ptrn
, &sp
->re
,
157 &sp
->re_len
, &sp
->re_c
, SEARCH_CSEARCH
| SEARCH_MSG
))
159 if (re_compile(sp
, ptrn
, t
- ptrn
, &sp
->subre
,
160 &sp
->subre_len
, &sp
->subre_c
, SEARCH_CSUBST
| SEARCH_MSG
))
164 sp
->searchdir
= FORWARD
;
169 * Get the replacement string.
171 * The special character & (\& if O_MAGIC not set) matches the
172 * entire RE. No handling of & is required here, it's done by
175 * The special character ~ (\~ if O_MAGIC not set) inserts the
176 * previous replacement string into this replacement string.
177 * Count ~'s to figure out how much space we need. We could
178 * special case nonexistent last patterns or whether or not
179 * O_MAGIC is set, but it's probably not worth the effort.
183 * Only toss an escaping character if it escapes a delimiter or
184 * if O_MAGIC is set and it escapes a tilde.
187 * If the entire replacement pattern is "%", then use the last
188 * replacement pattern. This semantic was added to vi in System
189 * V and then percolated elsewhere, presumably around the time
190 * that it was added to their version of ed(1).
192 if (p
[0] == '\0' || p
[0] == delim
) {
195 if (sp
->repl
!= NULL
)
199 } else if (p
[0] == '%' && (p
[1] == '\0' || p
[1] == delim
))
200 p
+= p
[1] == delim
? 2 : 1;
202 for (rep
= p
, len
= 0;
203 p
[0] != '\0' && p
[0] != delim
; ++p
, ++len
)
206 GET_SPACE_RETW(sp
, bp
, blen
, len
);
207 for (t
= bp
, len
= 0, p
= rep
;;) {
208 if (p
[0] == '\0' || p
[0] == delim
) {
216 else if (p
[1] == '\\') {
219 } else if (p
[1] == '~') {
221 if (!O_ISSET(sp
, O_MAGIC
))
224 } else if (p
[0] == '~' && O_ISSET(sp
, O_MAGIC
)) {
226 MEMCPYW(t
, sp
->repl
, sp
->repl_len
);
234 if ((sp
->repl_len
= len
) != 0) {
235 if (sp
->repl
!= NULL
)
237 if ((sp
->repl
= malloc(len
)) == NULL
) {
238 msgq(sp
, M_SYSERR
, NULL
);
239 FREE_SPACEW(sp
, bp
, blen
);
242 MEMCPYW(sp
->repl
, bp
, len
);
244 FREE_SPACEW(sp
, bp
, blen
);
246 return (s(sp
, cmdp
, p
, re
, flags
));
251 * [line [,line]] & [cgr] [count] [#lp]]
253 * Substitute using the last substitute RE and replacement pattern.
255 * PUBLIC: int ex_subagain __P((SCR *, EXCMD *));
258 ex_subagain(sp
, cmdp
)
262 if (sp
->subre
== NULL
) {
263 ex_emsg(sp
, NULL
, EXM_NOPREVRE
);
266 if (!F_ISSET(sp
, SC_RE_SUBST
) &&
267 re_compile(sp
, sp
->subre
, sp
->subre_len
,
268 NULL
, NULL
, &sp
->subre_c
, SEARCH_CSUBST
| SEARCH_MSG
))
271 cmdp
, cmdp
->argc
? cmdp
->argv
[0]->bp
: NULL
, &sp
->subre_c
, 0));
276 * [line [,line]] ~ [cgr] [count] [#lp]]
278 * Substitute using the last RE and last substitute replacement pattern.
280 * PUBLIC: int ex_subtilde __P((SCR *, EXCMD *));
283 ex_subtilde(sp
, cmdp
)
287 if (sp
->re
== NULL
) {
288 ex_emsg(sp
, NULL
, EXM_NOPREVRE
);
291 if (!F_ISSET(sp
, SC_RE_SEARCH
) && re_compile(sp
, sp
->re
,
292 sp
->re_len
, NULL
, NULL
, &sp
->re_c
, SEARCH_CSEARCH
| SEARCH_MSG
))
295 cmdp
, cmdp
->argc
? cmdp
->argv
[0]->bp
: NULL
, &sp
->re_c
, 0));
300 * Do the substitution. This stuff is *really* tricky. There are lots of
301 * special cases, and general nastiness. Don't mess with it unless you're
304 * The nasty part of the substitution is what happens when the replacement
305 * string contains newlines. It's a bit tricky -- consider the information
306 * that has to be retained for "s/f\(o\)o/^M\1^M\1/". The solution here is
307 * to build a set of newline offsets which we use to break the line up later,
308 * when the replacement is done. Don't change it unless you're *damned*
311 #define NEEDNEWLINE(sp) { \
312 if (sp->newl_len == sp->newl_cnt) { \
313 sp->newl_len += 25; \
314 REALLOC(sp, sp->newl, size_t *, \
315 sp->newl_len * sizeof(size_t)); \
316 if (sp->newl == NULL) { \
323 #define BUILD(sp, l, len) { \
324 if (lbclen + (len) > lblen) { \
325 lblen += MAX(lbclen + (len), 256); \
326 REALLOC(sp, lb, CHAR_T *, lblen * sizeof(CHAR_T)); \
332 MEMCPYW(lb + lbclen, l, len); \
336 #define NEEDSP(sp, len, pnt) { \
337 if (lbclen + (len) > lblen) { \
338 lblen += MAX(lbclen + (len), 256); \
339 REALLOC(sp, lb, CHAR_T *, lblen * sizeof(CHAR_T)); \
349 s(sp
, cmdp
, s
, re
, flags
)
359 db_recno_t elno
, lno
, slno
;
361 regmatch_t match
[10];
362 size_t blen
, cnt
, last
, lbclen
, lblen
, len
, llen
;
363 size_t offset
, saved_offset
, scno
;
364 int cflag
, lflag
, nflag
, pflag
, rflag
;
365 int didsub
, do_eol_match
, eflags
, empty_ok
, eval
;
366 int linechanged
, matched
, quit
, rval
;
377 * Historically, the 'g' and 'c' suffices were always toggled as flags,
378 * so ":s/A/B/" was the same as ":s/A/B/ccgg". If O_EDCOMPATIBLE was
379 * not set, they were initialized to 0 for all substitute commands. If
380 * O_EDCOMPATIBLE was set, they were initialized to 0 only if the user
381 * specified substitute/replacement patterns (see ex_s()).
383 if (!O_ISSET(sp
, O_EDCOMPATIBLE
))
384 sp
->c_suffix
= sp
->g_suffix
= 0;
387 * Historic vi permitted the '#', 'l' and 'p' options in vi mode, but
388 * it only displayed the last change. I'd disallow them, but they are
389 * useful in combination with the [v]global commands. In the current
390 * model the problem is combining them with the 'c' flag -- the screen
391 * would have to flip back and forth between the confirm screen and the
392 * ex print screen, which would be pretty awful. We do display all
393 * changes, though, for what that's worth.
396 * Historic vi was fairly strict about the order of "options", the
397 * count, and "flags". I'm somewhat fuzzy on the difference between
398 * options and flags, anyway, so this is a simpler approach, and we
399 * just take it them in whatever order the user gives them. (The ex
400 * usage statement doesn't reflect this.)
402 cflag
= lflag
= nflag
= pflag
= rflag
= 0;
405 for (lno
= OOBLNO
; *s
!= '\0'; ++s
)
416 case '0': case '1': case '2': case '3': case '4':
417 case '5': case '6': case '7': case '8': case '9':
421 nret
= nget_uslong(sp
, &ul
, s
, &s
, 10);
423 if (*s
== '\0') /* Loop increment correction. */
425 if (nret
!= NUM_OK
) {
426 if (nret
== NUM_OVER
)
427 msgq(sp
, M_ERR
, "153|Count overflow");
428 else if (nret
== NUM_UNDER
)
429 msgq(sp
, M_ERR
, "154|Count underflow");
431 msgq(sp
, M_SYSERR
, NULL
);
435 * In historic vi, the count was inclusive from the
438 cmdp
->addr1
.lno
= cmdp
->addr2
.lno
;
439 cmdp
->addr2
.lno
+= lno
- 1;
440 if (!db_exist(sp
, cmdp
->addr2
.lno
) &&
441 db_last(sp
, &cmdp
->addr2
.lno
))
448 sp
->c_suffix
= !sp
->c_suffix
;
450 /* Ex text structure initialization. */
451 if (F_ISSET(sp
, SC_EX
)) {
452 memset(&tiq
, 0, sizeof(TEXTH
));
457 sp
->g_suffix
= !sp
->g_suffix
;
466 if (LF_ISSET(SUB_FIRST
)) {
468 "155|Regular expression specified; r flag meaningless");
471 if (!F_ISSET(sp
, SC_RE_SEARCH
)) {
472 ex_emsg(sp
, NULL
, EXM_NOPREVRE
);
482 if (*s
!= '\0' || !rflag
&& LF_ISSET(SUB_MUSTSETR
)) {
483 usage
: ex_emsg(sp
, cmdp
->cmd
->usage
, EXM_USAGE
);
487 noargs
: if (F_ISSET(sp
, SC_VI
) && sp
->c_suffix
&& (lflag
|| nflag
|| pflag
)) {
489 "156|The #, l and p flags may not be combined with the c flag in vi mode");
494 * bp: if interactive, line cache
495 * blen: if interactive, line cache length
496 * lb: build buffer pointer.
497 * lbclen: current length of built buffer.
498 * lblen; length of build buffer.
501 blen
= lbclen
= lblen
= 0;
503 /* For each line... */
504 lno
= cmdp
->addr1
.lno
== 0 ? 1 : cmdp
->addr1
.lno
;
505 for (matched
= quit
= 0,
506 elno
= cmdp
->addr2
.lno
; !quit
&& lno
<= elno
; ++lno
) {
508 /* Someone's unhappy, time to stop. */
513 if (db_get(sp
, lno
, DBG_FATAL
, &s
, &llen
))
517 * Make a local copy if doing confirmation -- when calling
518 * the confirm routine we're likely to lose the cached copy.
522 GET_SPACE_RETW(sp
, bp
, blen
, llen
);
524 ADD_SPACE_RETW(sp
, bp
, blen
, llen
);
525 MEMCPYW(bp
, s
, llen
);
529 /* Start searching from the beginning. */
533 /* Reset the build buffer offset. */
536 /* Reset empty match flag. */
540 * We don't want to have to do a setline if the line didn't
541 * change -- keep track of whether or not this line changed.
542 * If doing confirmations, don't want to keep setting the
543 * line if change is refused -- keep track of substitutions.
545 didsub
= linechanged
= 0;
547 /* New line, do an EOL match. */
550 /* It's not nul terminated, but we pretend it is. */
551 eflags
= REG_STARTEND
;
554 * The search area is from s + offset to the EOL.
556 * Generally, match[0].rm_so is the offset of the start
557 * of the match from the start of the search, and offset
558 * is the offset of the start of the last search.
560 nextmatch
: match
[0].rm_so
= 0;
561 match
[0].rm_eo
= len
;
563 /* Get the next match. */
564 eval
= regexec(re
, s
+ offset
, 10, match
, eflags
);
567 * There wasn't a match or if there was an error, deal with
568 * it. If there was a previous match in this line, resolve
569 * the changes into the database. Otherwise, just move on.
571 if (eval
== REG_NOMATCH
)
574 re_error(sp
, eval
, re
);
579 /* Only the first search can match an anchored expression. */
580 eflags
|= REG_NOTBOL
;
584 * It's possible to match 0-length strings -- for example, the
585 * command s;a*;X;, when matched against the string "aabb" will
586 * result in "XbXbX", i.e. the matches are "aa", the space
587 * between the b's and the space between the b's and the end of
588 * the string. There is a similar space between the beginning
589 * of the string and the a's. The rule that we use (because vi
590 * historically used it) is that any 0-length match, occurring
591 * immediately after a match, is ignored. Otherwise, the above
592 * example would have resulted in "XXbXbX". Another example is
593 * incorrectly using " *" to replace groups of spaces with one
596 * The way we do this is that if we just had a successful match,
597 * the starting offset does not skip characters, and the match
598 * is empty, ignore the match and move forward. If there's no
599 * more characters in the string, we were attempting to match
600 * after the last character, so quit.
602 if (!empty_ok
&& match
[0].rm_so
== 0 && match
[0].rm_eo
== 0) {
606 BUILD(sp
, s
+ offset
, 1)
612 /* Confirm change. */
615 * Set the cursor position for confirmation. Note,
616 * if we matched on a '$', the cursor may be past
619 from
.lno
= to
.lno
= lno
;
620 from
.cno
= match
[0].rm_so
+ offset
;
621 to
.cno
= match
[0].rm_eo
+ offset
;
623 * Both ex and vi have to correct for a change before
624 * the first character in the line.
627 from
.cno
= to
.cno
= 0;
628 if (F_ISSET(sp
, SC_VI
)) {
630 * Only vi has to correct for a change after
631 * the last character in the line.
634 * It would be nice to change the vi code so
635 * that we could display a cursor past EOL.
639 if (from
.cno
>= llen
)
644 if (vs_refresh(sp
, 1))
647 vs_update(sp
, msg_cat(sp
,
648 "169|Confirm change? [n]", NULL
), NULL
);
650 if (v_event_get(sp
, &ev
, 0, 0))
652 switch (ev
.e_event
) {
660 v_event_err(sp
, &ev
);
664 if (ex_print(sp
, cmdp
, &from
, &to
, 0) ||
665 ex_scprint(sp
, &from
, &to
))
667 if (ex_txt(sp
, &tiq
, 0, TXT_CR
))
669 ev
.e_c
= tiq
.cqh_first
->lb
[0];
678 BUILD(sp
, s
+offset
, match
[0].rm_eo
);
681 /* Set the quit/interrupted flags. */
683 F_SET(sp
->gp
, G_INTERRUPTED
);
686 * Resolve any changes, then return to (and
687 * exit from) the main loop.
694 * Set the cursor to the last position changed, converting
695 * from 1-based to 0-based.
698 sp
->cno
= match
[0].rm_so
;
700 /* Copy the bytes before the match into the build buffer. */
701 BUILD(sp
, s
+ offset
, match
[0].rm_so
);
703 /* Substitute the matching bytes. */
705 if (re_sub(sp
, s
+ offset
, &lb
, &lbclen
, &lblen
, match
))
708 /* Set the change flag so we know this line was modified. */
711 /* Move past the matched bytes. */
712 skip
: offset
+= match
[0].rm_eo
;
713 len
-= match
[0].rm_eo
;
715 /* A match cannot be followed by an empty pattern. */
719 * If doing a global change with confirmation, we have to
720 * update the screen. The basic idea is to store the line
721 * so the screen update routines can find it, and restart.
723 if (didsub
&& sp
->c_suffix
&& sp
->g_suffix
) {
725 * The new search offset will be the end of the
728 saved_offset
= lbclen
;
730 /* Copy the rest of the line. */
732 BUILD(sp
, s
+ offset
, len
)
734 /* Set the new offset. */
735 offset
= saved_offset
;
737 /* Store inserted lines, adjusting the build buffer. */
741 cnt
< sp
->newl_cnt
; ++cnt
, ++lno
, ++elno
) {
742 if (db_insert(sp
, lno
,
743 lb
+ last
, sp
->newl
[cnt
] - last
))
745 last
= sp
->newl
[cnt
] + 1;
746 ++sp
->rptlines
[L_ADDED
];
753 /* Store and retrieve the line. */
754 if (db_set(sp
, lno
, lb
+ last
, lbclen
))
756 if (db_get(sp
, lno
, DBG_FATAL
, &s
, &llen
))
758 ADD_SPACE_RETW(sp
, bp
, blen
, llen
)
759 MEMCPYW(bp
, s
, llen
);
763 /* Restart the build. */
765 BUILD(sp
, s
, offset
);
768 * If we haven't already done the after-the-string
769 * match, do one. Set REG_NOTEOL so the '$' pattern
776 eflags
|= REG_NOTEOL
;
784 * If at the end of the string, do a test for the after
785 * the string match. Set REG_NOTEOL so the '$' pattern
788 if (sp
->g_suffix
&& do_eol_match
) {
791 eflags
|= REG_NOTEOL
;
796 endmatch
: if (!linechanged
)
799 /* Copy any remaining bytes into the build buffer. */
801 BUILD(sp
, s
+ offset
, len
)
803 /* Store inserted lines, adjusting the build buffer. */
807 cnt
< sp
->newl_cnt
; ++cnt
, ++lno
, ++elno
) {
809 lno
, lb
+ last
, sp
->newl
[cnt
] - last
))
811 last
= sp
->newl
[cnt
] + 1;
812 ++sp
->rptlines
[L_ADDED
];
818 /* Store the changed line. */
819 if (db_set(sp
, lno
, lb
+ last
, lbclen
))
822 /* Update changed line counter. */
823 if (sp
->rptlchange
!= lno
) {
824 sp
->rptlchange
= lno
;
825 ++sp
->rptlines
[L_CHANGED
];
830 * Display as necessary. Historic practice is to only
831 * display the last line of a line split into multiple
834 if (lflag
|| nflag
|| pflag
) {
835 from
.lno
= to
.lno
= lno
;
836 from
.cno
= to
.cno
= 0;
838 (void)ex_print(sp
, cmdp
, &from
, &to
, E_C_LIST
);
840 (void)ex_print(sp
, cmdp
, &from
, &to
, E_C_HASH
);
842 (void)ex_print(sp
, cmdp
, &from
, &to
, E_C_PRINT
);
848 * Historically, vi attempted to leave the cursor at the same place if
849 * the substitution was done at the current cursor position. Otherwise
850 * it moved it to the first non-blank of the last line changed. There
851 * were some problems: for example, :s/$/foo/ with the cursor on the
852 * last character of the line left the cursor on the last character, or
853 * the & command with multiple occurrences of the matching string in the
854 * line usually left the cursor in a fairly random position.
856 * We try to do the same thing, with the exception that if the user is
857 * doing substitution with confirmation, we move to the last line about
858 * which the user was consulted, as opposed to the last line that they
859 * actually changed. This prevents a screen flash if the user doesn't
860 * change many of the possible lines.
862 if (!sp
->c_suffix
&& (sp
->lno
!= slno
|| sp
->cno
!= scno
)) {
864 (void)nonblank(sp
, sp
->lno
, &sp
->cno
);
868 * If not in a global command, and nothing matched, say so.
869 * Else, if none of the lines displayed, put something up.
873 if (!F_ISSET(sp
, SC_EX_GLOBAL
)) {
874 msgq(sp
, M_ERR
, "157|No match found");
877 } else if (!lflag
&& !nflag
&& !pflag
)
878 F_SET(cmdp
, E_AUTOPRINT
);
885 FREE_SPACEW(sp
, bp
, blen
);
895 * PUBLIC: int re_compile __P((SCR *,
896 * PUBLIC: CHAR_T *, size_t, CHAR_T **, size_t *, regex_t *, u_int));
899 re_compile(sp
, ptrn
, plen
, ptrnp
, lenp
, rep
, flags
)
901 CHAR_T
*ptrn
, **ptrnp
;
907 int reflags
, replaced
, rval
;
912 if (LF_ISSET(SEARCH_EXTEND
))
913 reflags
|= REG_EXTENDED
;
914 if (LF_ISSET(SEARCH_IC
))
915 reflags
|= REG_ICASE
;
916 if (LF_ISSET(SEARCH_LITERAL
))
917 reflags
|= REG_NOSPEC
;
918 if (!LF_ISSET(SEARCH_NOOPT
| SEARCH_CSCOPE
| SEARCH_TAG
)) {
919 if (O_ISSET(sp
, O_EXTENDED
))
920 reflags
|= REG_EXTENDED
;
921 if (O_ISSET(sp
, O_IGNORECASE
))
922 reflags
|= REG_ICASE
;
923 if (O_ISSET(sp
, O_ICLOWER
))
926 if (LF_ISSET(SEARCH_ICL
)) {
927 iclower
: for (p
= ptrn
, len
= plen
; len
> 0; ++p
, --len
)
931 reflags
|= REG_ICASE
;
934 /* If we're replacing a saved value, clear the old one. */
935 if (LF_ISSET(SEARCH_CSEARCH
) && F_ISSET(sp
, SC_RE_SEARCH
)) {
937 F_CLR(sp
, SC_RE_SEARCH
);
939 if (LF_ISSET(SEARCH_CSUBST
) && F_ISSET(sp
, SC_RE_SUBST
)) {
940 regfree(&sp
->subre_c
);
941 F_CLR(sp
, SC_RE_SUBST
);
945 * If we're saving the string, it's a pattern we haven't seen before,
946 * so convert the vi-style RE's to POSIX 1003.2 RE's. Save a copy for
947 * later recompilation. Free any previously saved value.
951 if (LF_ISSET(SEARCH_CSCOPE
)) {
952 if (re_cscope_conv(sp
, &ptrn
, &plen
, &replaced
))
956 * Currently, the match-any-<blank> expression used in
957 * re_cscope_conv() requires extended RE's. This may
958 * not be right or safe.
960 reflags
|= REG_EXTENDED
;
961 } else if (LF_ISSET(SEARCH_TAG
)) {
962 if (re_tag_conv(sp
, &ptrn
, &plen
, &replaced
))
964 } else if (!LF_ISSET(SEARCH_LITERAL
))
965 if (re_conv(sp
, &ptrn
, &plen
, &replaced
))
968 /* Discard previous pattern. */
969 if (*ptrnp
!= NULL
) {
977 * Copy the string into allocated memory.
980 * Regcomp isn't 8-bit clean, so the pattern is nul-terminated
981 * for now. There's just no other solution.
983 MALLOC(sp
, *ptrnp
, CHAR_T
*, (plen
+ 1) * sizeof(CHAR_T
));
984 if (*ptrnp
!= NULL
) {
985 MEMCPYW(*ptrnp
, ptrn
, plen
);
986 (*ptrnp
)[plen
] = '\0';
989 /* Free up conversion-routine-allocated memory. */
991 FREE_SPACEW(sp
, ptrn
, 0);
1001 * Regcomp isn't 8-bit clean, so we just lost if the pattern
1002 * contained a nul. Bummer!
1004 if ((rval
= regcomp(rep
, ptrn
, /* plen, */ reflags
)) != 0) {
1005 if (LF_ISSET(SEARCH_MSG
))
1006 re_error(sp
, rval
, rep
);
1010 if (LF_ISSET(SEARCH_CSEARCH
))
1011 F_SET(sp
, SC_RE_SEARCH
);
1012 if (LF_ISSET(SEARCH_CSUBST
))
1013 F_SET(sp
, SC_RE_SUBST
);
1020 * Convert vi's regular expressions into something that the
1021 * the POSIX 1003.2 RE functions can handle.
1023 * There are three conversions we make to make vi's RE's (specifically
1024 * the global, search, and substitute patterns) work with POSIX RE's.
1026 * 1: If O_MAGIC is not set, strip backslashes from the magic character
1027 * set (.[*~) that have them, and add them to the ones that don't.
1028 * 2: If O_MAGIC is not set, the string "\~" is replaced with the text
1029 * from the last substitute command's replacement string. If O_MAGIC
1030 * is set, it's the string "~".
1031 * 3: The pattern \<ptrn\> does "word" searches, convert it to use the
1035 * This doesn't exactly match the historic behavior of vi because we do
1036 * the ~ substitution before calling the RE engine, so magic characters
1037 * in the replacement string will be expanded by the RE engine, and they
1038 * weren't historically. It's a bug.
1041 re_conv(sp
, ptrnp
, plenp
, replacedp
)
1047 size_t blen
, len
, needlen
;
1052 * First pass through, we figure out how much space we'll need.
1053 * We do it in two passes, on the grounds that most of the time
1054 * the user is doing a search and won't have magic characters.
1055 * That way we can skip most of the memory allocation and copies.
1058 for (p
= *ptrnp
, len
= *plenp
, needlen
= 0; len
> 0; ++p
, --len
)
1066 needlen
+= RE_WSTART_LEN
;
1070 needlen
+= RE_WSTOP_LEN
;
1073 if (!O_ISSET(sp
, O_MAGIC
)) {
1075 needlen
+= sp
->repl_len
;
1081 if (!O_ISSET(sp
, O_MAGIC
)) {
1093 if (O_ISSET(sp
, O_MAGIC
)) {
1095 needlen
+= sp
->repl_len
;
1101 if (!O_ISSET(sp
, O_MAGIC
)) {
1116 /* Get enough memory to hold the final pattern. */
1118 GET_SPACE_RETW(sp
, bp
, blen
, needlen
);
1120 for (p
= *ptrnp
, len
= *plenp
, t
= bp
; len
> 0; ++p
, --len
)
1128 RE_WSTART
, RE_WSTART_LEN
- 1);
1129 t
+= RE_WSTART_LEN
- 1;
1133 RE_WSTOP
, RE_WSTOP_LEN
- 1);
1134 t
+= RE_WSTOP_LEN
- 1;
1137 if (O_ISSET(sp
, O_MAGIC
))
1141 sp
->repl
, sp
->repl_len
);
1148 if (O_ISSET(sp
, O_MAGIC
))
1160 if (O_ISSET(sp
, O_MAGIC
)) {
1161 MEMCPYW(t
, sp
->repl
, sp
->repl_len
);
1169 if (!O_ISSET(sp
, O_MAGIC
))
1185 * Convert a tags search path into something that the POSIX
1186 * 1003.2 RE functions can handle.
1189 re_tag_conv(sp
, ptrnp
, plenp
, replacedp
)
1201 /* Max memory usage is 2 times the length of the string. */
1203 GET_SPACE_RETW(sp
, bp
, blen
, len
* 2);
1208 /* If the last character is a '/' or '?', we just strip it. */
1209 if (len
> 0 && (p
[len
- 1] == '/' || p
[len
- 1] == '?'))
1212 /* If the next-to-last or last character is a '$', it's magic. */
1213 if (len
> 0 && p
[len
- 1] == '$') {
1219 /* If the first character is a '/' or '?', we just strip it. */
1220 if (len
> 0 && (p
[0] == '/' || p
[0] == '?')) {
1225 /* If the first or second character is a '^', it's magic. */
1232 * Escape every other magic character we can find, meanwhile stripping
1233 * the backslashes ctags inserts when escaping the search delimiter
1236 for (; len
> 0; --len
) {
1237 if (p
[0] == '\\' && (p
[1] == '/' || p
[1] == '?')) {
1240 } else if (strchr("^.[]$*", p
[0]))
1254 * Convert a cscope search path into something that the POSIX
1255 * 1003.2 RE functions can handle.
1258 re_cscope_conv(sp
, ptrnp
, plenp
, replacedp
)
1264 size_t blen
, len
, nspaces
;
1271 * Each space in the source line printed by cscope represents an
1272 * arbitrary sequence of spaces, tabs, and comments.
1274 #define CSCOPE_RE_SPACE "([ \t]|/\\*([^*]|\\*/)*\\*/)*"
1275 #define CSCOPE_LEN sizeof(CSCOPE_RE_SPACE) - 1
1276 CHAR2INT(sp
, CSCOPE_RE_SPACE
, CSCOPE_LEN
, wp
, wlen
);
1277 for (nspaces
= 0, p
= *ptrnp
, len
= *plenp
; len
> 0; ++p
, --len
)
1282 * Allocate plenty of space:
1283 * the string, plus potential escaping characters;
1284 * nspaces + 2 copies of CSCOPE_RE_SPACE;
1285 * ^, $, nul terminator characters.
1288 len
= (p
- *ptrnp
) * 2 + (nspaces
+ 2) * sizeof(CSCOPE_RE_SPACE
) + 3;
1289 GET_SPACE_RETW(sp
, bp
, blen
, len
);
1295 MEMCPYW(t
, wp
, wlen
);
1298 for (len
= *plenp
; len
> 0; ++p
, --len
)
1300 MEMCPYW(t
, wp
, wlen
);
1303 if (strchr("\\^.[]$*+?()|{}", *p
))
1308 MEMCPYW(t
, wp
, wlen
);
1319 * Report a regular expression error.
1321 * PUBLIC: void re_error __P((SCR *, int, regex_t *));
1324 re_error(sp
, errcode
, preg
)
1332 s
= regerror(errcode
, preg
, "", 0);
1333 if ((oe
= malloc(s
)) == NULL
)
1334 msgq(sp
, M_SYSERR
, NULL
);
1336 (void)regerror(errcode
, preg
, oe
, s
);
1337 msgq(sp
, M_ERR
, "RE error: %s", oe
);
1344 * Do the substitution for a regular expression.
1347 re_sub(sp
, ip
, lbp
, lbclenp
, lblenp
, match
)
1349 CHAR_T
*ip
; /* Input line. */
1351 size_t *lbclenp
, *lblenp
;
1352 regmatch_t match
[10];
1354 enum { C_NOTSET
, C_LOWER
, C_ONELOWER
, C_ONEUPPER
, C_UPPER
} conv
;
1355 size_t lbclen
, lblen
; /* Local copies. */
1356 size_t mlen
; /* Match length. */
1357 size_t rpl
; /* Remaining replacement length. */
1358 CHAR_T
*rp
; /* Replacement pointer. */
1360 int no
; /* Match replacement offset. */
1361 CHAR_T
*p
, *t
; /* Buffer pointers. */
1362 CHAR_T
*lb
; /* Local copies. */
1364 lb
= *lbp
; /* Get local copies. */
1371 * There are some special sequences that vi provides in the
1372 * replacement patterns.
1373 * & string the RE matched (\& if nomagic set)
1374 * \# n-th regular subexpression
1375 * \E end \U, \L conversion
1376 * \e end \U, \L conversion
1377 * \l convert the next character to lower-case
1378 * \L convert to lower-case, until \E, \e, or end of replacement
1379 * \u convert the next character to upper-case
1380 * \U convert to upper-case, until \E, \e, or end of replacement
1382 * Otherwise, since this is the lowest level of replacement, discard
1383 * all escaping characters. This (hopefully) matches historic practice.
1385 #define OUTCH(ch, nltrans) { \
1386 CHAR_T __ch = (ch); \
1387 u_int __value = KEY_VAL(sp, __ch); \
1388 if (nltrans && (__value == K_CR || __value == K_NL)) { \
1390 sp->newl[sp->newl_cnt++] = lbclen; \
1391 } else if (conv != C_NOTSET) { \
1397 if (isupper(__ch)) \
1398 __ch = tolower(__ch); \
1404 if (islower(__ch)) \
1405 __ch = toupper(__ch); \
1416 for (rp
= sp
->repl
, rpl
= sp
->repl_len
, p
= lb
+ lbclen
; rpl
--;) {
1417 switch (ch
= *rp
++) {
1419 if (O_ISSET(sp
, O_MAGIC
)) {
1431 if (!O_ISSET(sp
, O_MAGIC
)) {
1436 case '0': case '1': case '2': case '3': case '4':
1437 case '5': case '6': case '7': case '8': case '9':
1439 subzero
: if (match
[no
].rm_so
== -1 ||
1440 match
[no
].rm_eo
== -1)
1442 mlen
= match
[no
].rm_eo
- match
[no
].rm_so
;
1443 for (t
= ip
+ match
[no
].rm_so
; mlen
--; ++t
)
1475 *lbp
= lb
; /* Update caller's information. */