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.
12 #include <sys/types.h>
13 #include <sys/queue.h>
16 #include <bitstring.h>
25 #include "../common/common.h"
28 #define SUB_FIRST 0x01 /* The 'r' flag isn't reasonable. */
29 #define SUB_MUSTSETR 0x02 /* The 'r' flag is required. */
31 static int re_conv(SCR
*, CHAR_T
**, size_t *, int *);
32 static int re_cscope_conv(SCR
*, CHAR_T
**, size_t *, int *);
33 static int re_sub(SCR
*,
34 CHAR_T
*, CHAR_T
**, size_t *, size_t *, regmatch_t
[10]);
35 static int re_tag_conv(SCR
*, CHAR_T
**, size_t *, int *);
36 static int s(SCR
*, EXCMD
*, CHAR_T
*, regex_t
*, u_int
);
40 * [line [,line]] s[ubstitute] [[/;]pat[/;]/repl[/;] [cgr] [count] [#lp]]
42 * Substitute on lines matching a pattern.
44 * PUBLIC: int ex_s(SCR *, EXCMD *);
47 ex_s(SCR
*sp
, EXCMD
*cmdp
)
53 CHAR_T
*bp
, *p
, *ptrn
, *rep
, *t
;
56 * Skip leading white space.
59 * Historic vi allowed any non-alphanumeric to serve as the
60 * substitution command delimiter.
63 * If the arguments are empty, it's the same as &, i.e. we
64 * repeat the last substitution.
68 for (p
= cmdp
->argv
[0]->bp
,
69 len
= cmdp
->argv
[0]->len
; len
> 0; --len
, ++p
) {
74 subagain
: return (ex_subagain(sp
, cmdp
));
77 if (is09azAZ(delim
) || delim
== '\\')
78 return (s(sp
, cmdp
, p
, &sp
->subre_c
, SUB_MUSTSETR
));
82 * The full-blown substitute command reset the remembered
83 * state of the 'c' and 'g' suffices.
85 sp
->c_suffix
= sp
->g_suffix
= 0;
88 * Get the pattern string, toss escaping characters.
91 * Historic vi accepted any of the following forms:
93 * :s/abc/def/ change "abc" to "def"
94 * :s/abc/def change "abc" to "def"
95 * :s/abc/ delete "abc"
100 * Only toss an escaping character if it escapes a delimiter.
101 * This means that "s/A/\\\\f" replaces "A" with "\\f". It
102 * would be nice to be more regular, i.e. for each layer of
103 * escaping a single escaping character is removed, but that's
104 * not how the historic vi worked.
106 for (ptrn
= t
= p
;;) {
107 if (p
[0] == '\0' || p
[0] == delim
) {
112 * Nul terminate the pattern string -- it's passed
113 * to regcomp which doesn't understand anything else.
121 else if (p
[1] == '\\')
127 * If the pattern string is empty, use the last RE (not just the
128 * last substitution RE).
131 if (sp
->re
== NULL
) {
132 ex_emsg(sp
, NULL
, EXM_NOPREVRE
);
136 /* Re-compile the RE if necessary. */
137 if (!F_ISSET(sp
, SC_RE_SEARCH
) &&
138 re_compile(sp
, sp
->re
, sp
->re_len
,
139 NULL
, NULL
, &sp
->re_c
, RE_C_SEARCH
))
145 * Compile the RE. Historic practice is that substitutes set
146 * the search direction as well as both substitute and search
147 * RE's. We compile the RE twice, as we don't want to bother
148 * ref counting the pattern string and (opaque) structure.
150 if (re_compile(sp
, ptrn
, t
- ptrn
, &sp
->re
,
151 &sp
->re_len
, &sp
->re_c
, RE_C_SEARCH
))
153 if (re_compile(sp
, ptrn
, t
- ptrn
, &sp
->subre
,
154 &sp
->subre_len
, &sp
->subre_c
, RE_C_SUBST
))
158 sp
->searchdir
= FORWARD
;
163 * Get the replacement string.
165 * The special character & (\& if O_MAGIC not set) matches the
166 * entire RE. No handling of & is required here, it's done by
169 * The special character ~ (\~ if O_MAGIC not set) inserts the
170 * previous replacement string into this replacement string.
171 * Count ~'s to figure out how much space we need. We could
172 * special case nonexistent last patterns or whether or not
173 * O_MAGIC is set, but it's probably not worth the effort.
177 * Only toss an escaping character if it escapes a delimiter or
178 * if O_MAGIC is set and it escapes a tilde.
181 * If the entire replacement pattern is "%", then use the last
182 * replacement pattern. This semantic was added to vi in System
183 * V and then percolated elsewhere, presumably around the time
184 * that it was added to their version of ed(1).
186 if (p
[0] == '\0' || p
[0] == delim
) {
192 } else if (p
[0] == '%' && (p
[1] == '\0' || p
[1] == delim
))
193 p
+= p
[1] == delim
? 2 : 1;
195 for (rep
= p
, len
= 0;
196 p
[0] != '\0' && p
[0] != delim
; ++p
, ++len
)
199 GET_SPACE_RETW(sp
, bp
, blen
, len
);
200 for (t
= bp
, len
= 0, p
= rep
;;) {
201 if (p
[0] == '\0' || p
[0] == delim
) {
209 else if (p
[1] == '\\') {
212 } else if (p
[1] == '~') {
214 if (!O_ISSET(sp
, O_MAGIC
))
217 } else if (p
[0] == '~' && O_ISSET(sp
, O_MAGIC
)) {
219 MEMCPY(t
, sp
->repl
, sp
->repl_len
);
227 if ((sp
->repl_len
= len
) != 0) {
229 MALLOC(sp
, sp
->repl
, len
* sizeof(CHAR_T
));
230 if (sp
->repl
== NULL
) {
231 FREE_SPACEW(sp
, bp
, blen
);
234 MEMCPY(sp
->repl
, bp
, len
);
236 FREE_SPACEW(sp
, bp
, blen
);
238 return (s(sp
, cmdp
, p
, re
, flags
));
243 * [line [,line]] & [cgr] [count] [#lp]]
245 * Substitute using the last substitute RE and replacement pattern.
247 * PUBLIC: int ex_subagain(SCR *, EXCMD *);
250 ex_subagain(SCR
*sp
, EXCMD
*cmdp
)
252 if (sp
->subre
== NULL
) {
253 ex_emsg(sp
, NULL
, EXM_NOPREVRE
);
256 if (!F_ISSET(sp
, SC_RE_SUBST
) &&
257 re_compile(sp
, sp
->subre
, sp
->subre_len
,
258 NULL
, NULL
, &sp
->subre_c
, RE_C_SUBST
))
261 cmdp
, cmdp
->argc
? cmdp
->argv
[0]->bp
: NULL
, &sp
->subre_c
, 0));
266 * [line [,line]] ~ [cgr] [count] [#lp]]
268 * Substitute using the last RE and last substitute replacement pattern.
270 * PUBLIC: int ex_subtilde(SCR *, EXCMD *);
273 ex_subtilde(SCR
*sp
, EXCMD
*cmdp
)
275 if (sp
->re
== NULL
) {
276 ex_emsg(sp
, NULL
, EXM_NOPREVRE
);
279 if (!F_ISSET(sp
, SC_RE_SEARCH
) && re_compile(sp
, sp
->re
,
280 sp
->re_len
, NULL
, NULL
, &sp
->re_c
, RE_C_SEARCH
))
283 cmdp
, cmdp
->argc
? cmdp
->argv
[0]->bp
: NULL
, &sp
->re_c
, 0));
288 * Do the substitution. This stuff is *really* tricky. There are lots of
289 * special cases, and general nastiness. Don't mess with it unless you're
292 * The nasty part of the substitution is what happens when the replacement
293 * string contains newlines. It's a bit tricky -- consider the information
294 * that has to be retained for "s/f\(o\)o/^M\1^M\1/". The solution here is
295 * to build a set of newline offsets which we use to break the line up later,
296 * when the replacement is done. Don't change it unless you're *damned*
299 #define NEEDNEWLINE(sp) { \
300 if (sp->newl_len == sp->newl_cnt) { \
301 sp->newl_len += 25; \
302 REALLOC(sp, sp->newl, size_t *, \
303 sp->newl_len * sizeof(size_t)); \
304 if (sp->newl == NULL) { \
311 #define BUILD(sp, l, len) { \
312 if (lbclen + (len) > lblen) { \
313 lblen = p2roundup(MAX(lbclen + (len), 256)); \
314 REALLOC(sp, lb, CHAR_T *, lblen * sizeof(CHAR_T)); \
320 MEMCPY(lb + lbclen, l, len); \
324 #define NEEDSP(sp, len, pnt) { \
325 if (lbclen + (len) > lblen) { \
326 lblen = p2roundup(MAX(lbclen + (len), 256)); \
327 REALLOC(sp, lb, CHAR_T *, lblen * sizeof(CHAR_T)); \
337 s(SCR
*sp
, EXCMD
*cmdp
, CHAR_T
*s
, regex_t
*re
, u_int flags
)
341 TEXTH tiq
[] = {{ 0 }};
342 recno_t elno
, lno
, slno
;
344 regmatch_t match
[10];
345 size_t blen
, cnt
, last
, lbclen
, lblen
, len
, llen
;
346 size_t offset
, saved_offset
, scno
;
347 int cflag
, lflag
, nflag
, pflag
, rflag
;
348 int didsub
, do_eol_match
, eflags
, empty_ok
, eval
;
349 int linechanged
, matched
, quit
, rval
;
360 * Historically, the 'g' and 'c' suffices were always toggled as flags,
361 * so ":s/A/B/" was the same as ":s/A/B/ccgg". If O_EDCOMPATIBLE was
362 * not set, they were initialized to 0 for all substitute commands. If
363 * O_EDCOMPATIBLE was set, they were initialized to 0 only if the user
364 * specified substitute/replacement patterns (see ex_s()).
366 if (!O_ISSET(sp
, O_EDCOMPATIBLE
))
367 sp
->c_suffix
= sp
->g_suffix
= 0;
370 * Historic vi permitted the '#', 'l' and 'p' options in vi mode, but
371 * it only displayed the last change. I'd disallow them, but they are
372 * useful in combination with the [v]global commands. In the current
373 * model the problem is combining them with the 'c' flag -- the screen
374 * would have to flip back and forth between the confirm screen and the
375 * ex print screen, which would be pretty awful. We do display all
376 * changes, though, for what that's worth.
379 * Historic vi was fairly strict about the order of "options", the
380 * count, and "flags". I'm somewhat fuzzy on the difference between
381 * options and flags, anyway, so this is a simpler approach, and we
382 * just take it them in whatever order the user gives them. (The ex
383 * usage statement doesn't reflect this.)
385 cflag
= lflag
= nflag
= pflag
= rflag
= 0;
388 for (lno
= OOBLNO
; *s
!= '\0'; ++s
)
399 case '0': case '1': case '2': case '3': case '4':
400 case '5': case '6': case '7': case '8': case '9':
404 nret
= nget_uslong(&ul
, s
, &s
, 10);
406 if (*s
== '\0') /* Loop increment correction. */
408 if (nret
!= NUM_OK
) {
409 if (nret
== NUM_OVER
)
410 msgq(sp
, M_ERR
, "153|Count overflow");
411 else if (nret
== NUM_UNDER
)
412 msgq(sp
, M_ERR
, "154|Count underflow");
414 msgq(sp
, M_SYSERR
, NULL
);
418 * In historic vi, the count was inclusive from the
421 cmdp
->addr1
.lno
= cmdp
->addr2
.lno
;
422 cmdp
->addr2
.lno
+= lno
- 1;
423 if (!db_exist(sp
, cmdp
->addr2
.lno
) &&
424 db_last(sp
, &cmdp
->addr2
.lno
))
431 sp
->c_suffix
= !sp
->c_suffix
;
433 /* Ex text structure initialization. */
434 if (F_ISSET(sp
, SC_EX
))
438 sp
->g_suffix
= !sp
->g_suffix
;
447 if (LF_ISSET(SUB_FIRST
)) {
449 "155|Regular expression specified; r flag meaningless");
452 if (!F_ISSET(sp
, SC_RE_SEARCH
)) {
453 ex_emsg(sp
, NULL
, EXM_NOPREVRE
);
463 if (*s
!= '\0' || (!rflag
&& LF_ISSET(SUB_MUSTSETR
))) {
464 usage
: ex_emsg(sp
, cmdp
->cmd
->usage
, EXM_USAGE
);
468 noargs
: if (F_ISSET(sp
, SC_VI
) && sp
->c_suffix
&& (lflag
|| nflag
|| pflag
)) {
470 "156|The #, l and p flags may not be combined with the c flag in vi mode");
475 * bp: if interactive, line cache
476 * blen: if interactive, line cache length
477 * lb: build buffer pointer.
478 * lbclen: current length of built buffer.
479 * lblen; length of build buffer.
482 blen
= lbclen
= lblen
= 0;
484 /* For each line... */
485 lno
= cmdp
->addr1
.lno
== 0 ? 1 : cmdp
->addr1
.lno
;
486 for (matched
= quit
= 0,
487 elno
= cmdp
->addr2
.lno
; !quit
&& lno
<= elno
; ++lno
) {
489 /* Someone's unhappy, time to stop. */
494 if (db_get(sp
, lno
, DBG_FATAL
, &s
, &llen
))
498 * Make a local copy if doing confirmation -- when calling
499 * the confirm routine we're likely to lose the cached copy.
503 GET_SPACE_RETW(sp
, bp
, blen
, llen
);
505 ADD_SPACE_RETW(sp
, bp
, blen
, llen
);
510 /* Start searching from the beginning. */
514 /* Reset the build buffer offset. */
517 /* Reset empty match flag. */
521 * We don't want to have to do a setline if the line didn't
522 * change -- keep track of whether or not this line changed.
523 * If doing confirmations, don't want to keep setting the
524 * line if change is refused -- keep track of substitutions.
526 didsub
= linechanged
= 0;
528 /* New line, do an EOL match. */
531 /* It's not nul terminated, but we pretend it is. */
532 eflags
= REG_STARTEND
;
535 * The search area is from s + offset to the EOL.
537 * Generally, match[0].rm_so is the offset of the start
538 * of the match from the start of the search, and offset
539 * is the offset of the start of the last search.
541 nextmatch
: match
[0].rm_so
= 0;
542 match
[0].rm_eo
= len
;
544 /* Get the next match. */
545 eval
= regexec(re
, s
+ offset
, 10, match
, eflags
);
548 * There wasn't a match or if there was an error, deal with
549 * it. If there was a previous match in this line, resolve
550 * the changes into the database. Otherwise, just move on.
552 if (eval
== REG_NOMATCH
)
555 re_error(sp
, eval
, re
);
560 /* Only the first search can match an anchored expression. */
561 eflags
|= REG_NOTBOL
;
565 * It's possible to match 0-length strings -- for example, the
566 * command s;a*;X;, when matched against the string "aabb" will
567 * result in "XbXbX", i.e. the matches are "aa", the space
568 * between the b's and the space between the b's and the end of
569 * the string. There is a similar space between the beginning
570 * of the string and the a's. The rule that we use (because vi
571 * historically used it) is that any 0-length match, occurring
572 * immediately after a match, is ignored. Otherwise, the above
573 * example would have resulted in "XXbXbX". Another example is
574 * incorrectly using " *" to replace groups of spaces with one
577 * The way we do this is that if we just had a successful match,
578 * the starting offset does not skip characters, and the match
579 * is empty, ignore the match and move forward. If there's no
580 * more characters in the string, we were attempting to match
581 * after the last character, so quit.
583 if (!empty_ok
&& match
[0].rm_so
== 0 && match
[0].rm_eo
== 0) {
587 BUILD(sp
, s
+ offset
, 1)
593 /* Confirm change. */
596 * Set the cursor position for confirmation. Note,
597 * if we matched on a '$', the cursor may be past
600 from
.lno
= to
.lno
= lno
;
601 from
.cno
= match
[0].rm_so
+ offset
;
602 to
.cno
= match
[0].rm_eo
+ offset
;
604 * Both ex and vi have to correct for a change before
605 * the first character in the line.
608 from
.cno
= to
.cno
= 0;
609 if (F_ISSET(sp
, SC_VI
)) {
611 * Only vi has to correct for a change after
612 * the last character in the line.
615 * It would be nice to change the vi code so
616 * that we could display a cursor past EOL.
620 if (from
.cno
>= llen
)
625 if (vs_refresh(sp
, 1))
628 vs_update(sp
, msg_cat(sp
,
629 "169|Confirm change? [n]", NULL
), NULL
);
631 if (v_event_get(sp
, &ev
, 0, 0))
633 switch (ev
.e_event
) {
641 v_event_err(sp
, &ev
);
645 if (ex_print(sp
, cmdp
, &from
, &to
, 0) ||
646 ex_scprint(sp
, &from
, &to
))
648 if (ex_txt(sp
, tiq
, 0, TXT_CR
))
650 ev
.e_c
= TAILQ_FIRST(tiq
)->lb
[0];
659 BUILD(sp
, s
+offset
, match
[0].rm_eo
);
662 /* Set the quit/interrupted flags. */
664 F_SET(sp
->gp
, G_INTERRUPTED
);
667 * Resolve any changes, then return to (and
668 * exit from) the main loop.
675 * Set the cursor to the last position changed, converting
676 * from 1-based to 0-based.
679 sp
->cno
= match
[0].rm_so
;
681 /* Copy the bytes before the match into the build buffer. */
682 BUILD(sp
, s
+ offset
, match
[0].rm_so
);
684 /* Substitute the matching bytes. */
686 if (re_sub(sp
, s
+ offset
, &lb
, &lbclen
, &lblen
, match
))
689 /* Set the change flag so we know this line was modified. */
692 /* Move past the matched bytes. */
693 skip
: offset
+= match
[0].rm_eo
;
694 len
-= match
[0].rm_eo
;
696 /* A match cannot be followed by an empty pattern. */
700 * If doing a global change with confirmation, we have to
701 * update the screen. The basic idea is to store the line
702 * so the screen update routines can find it, and restart.
704 if (didsub
&& sp
->c_suffix
&& sp
->g_suffix
) {
706 * The new search offset will be the end of the
709 saved_offset
= lbclen
;
711 /* Copy the rest of the line. */
713 BUILD(sp
, s
+ offset
, len
)
715 /* Set the new offset. */
716 offset
= saved_offset
;
718 /* Store inserted lines, adjusting the build buffer. */
722 cnt
< sp
->newl_cnt
; ++cnt
, ++lno
, ++elno
) {
723 if (db_insert(sp
, lno
,
724 lb
+ last
, sp
->newl
[cnt
] - last
))
726 last
= sp
->newl
[cnt
] + 1;
727 ++sp
->rptlines
[L_ADDED
];
734 /* Store and retrieve the line. */
735 if (db_set(sp
, lno
, lb
+ last
, lbclen
))
737 if (db_get(sp
, lno
, DBG_FATAL
, &s
, &llen
))
739 ADD_SPACE_RETW(sp
, bp
, blen
, llen
)
744 /* Restart the build. */
746 BUILD(sp
, s
, offset
);
749 * If we haven't already done the after-the-string
750 * match, do one. Set REG_NOTEOL so the '$' pattern
757 eflags
|= REG_NOTEOL
;
765 * If at the end of the string, do a test for the after
766 * the string match. Set REG_NOTEOL so the '$' pattern
769 if (sp
->g_suffix
&& do_eol_match
) {
772 eflags
|= REG_NOTEOL
;
777 endmatch
: if (!linechanged
)
780 /* Copy any remaining bytes into the build buffer. */
782 BUILD(sp
, s
+ offset
, len
)
784 /* Store inserted lines, adjusting the build buffer. */
788 cnt
< sp
->newl_cnt
; ++cnt
, ++lno
, ++elno
) {
790 lno
, lb
+ last
, sp
->newl
[cnt
] - last
))
792 last
= sp
->newl
[cnt
] + 1;
793 ++sp
->rptlines
[L_ADDED
];
799 /* Store the changed line. */
800 if (db_set(sp
, lno
, lb
+ last
, lbclen
))
803 /* Update changed line counter. */
804 if (sp
->rptlchange
!= lno
) {
805 sp
->rptlchange
= lno
;
806 ++sp
->rptlines
[L_CHANGED
];
811 * Display as necessary. Historic practice is to only
812 * display the last line of a line split into multiple
815 if (lflag
|| nflag
|| pflag
) {
816 from
.lno
= to
.lno
= lno
;
817 from
.cno
= to
.cno
= 0;
819 (void)ex_print(sp
, cmdp
, &from
, &to
, E_C_LIST
);
821 (void)ex_print(sp
, cmdp
, &from
, &to
, E_C_HASH
);
823 (void)ex_print(sp
, cmdp
, &from
, &to
, E_C_PRINT
);
829 * Historically, vi attempted to leave the cursor at the same place if
830 * the substitution was done at the current cursor position. Otherwise
831 * it moved it to the first non-blank of the last line changed. There
832 * were some problems: for example, :s/$/foo/ with the cursor on the
833 * last character of the line left the cursor on the last character, or
834 * the & command with multiple occurrences of the matching string in the
835 * line usually left the cursor in a fairly random position.
837 * We try to do the same thing, with the exception that if the user is
838 * doing substitution with confirmation, we move to the last line about
839 * which the user was consulted, as opposed to the last line that they
840 * actually changed. This prevents a screen flash if the user doesn't
841 * change many of the possible lines.
843 if (!sp
->c_suffix
&& (sp
->lno
!= slno
|| sp
->cno
!= scno
)) {
845 (void)nonblank(sp
, sp
->lno
, &sp
->cno
);
849 * If not in a global command, and nothing matched, say so.
850 * Else, if none of the lines displayed, put something up.
854 if (!F_ISSET(sp
, SC_EX_GLOBAL
)) {
855 msgq(sp
, M_ERR
, "157|No match found");
858 } else if (!lflag
&& !nflag
&& !pflag
)
859 F_SET(cmdp
, E_AUTOPRINT
);
866 FREE_SPACEW(sp
, bp
, blen
);
875 * PUBLIC: int re_compile(SCR *,
876 * PUBLIC: CHAR_T *, size_t, CHAR_T **, size_t *, regex_t *, u_int);
879 re_compile(SCR
*sp
, CHAR_T
*ptrn
, size_t plen
, CHAR_T
**ptrnp
, size_t *lenp
, regex_t
*rep
, u_int flags
)
882 int reflags
, replaced
, rval
;
887 if (!LF_ISSET(RE_C_CSCOPE
| RE_C_TAG
)) {
888 if (O_ISSET(sp
, O_EXTENDED
))
889 reflags
|= REG_EXTENDED
;
890 if (O_ISSET(sp
, O_IGNORECASE
))
891 reflags
|= REG_ICASE
;
892 if (O_ISSET(sp
, O_ICLOWER
)) {
893 for (p
= ptrn
, len
= plen
; len
> 0; ++p
, --len
)
897 reflags
|= REG_ICASE
;
901 /* If we're replacing a saved value, clear the old one. */
902 if (LF_ISSET(RE_C_SEARCH
) && F_ISSET(sp
, SC_RE_SEARCH
)) {
904 F_CLR(sp
, SC_RE_SEARCH
);
906 if (LF_ISSET(RE_C_SUBST
) && F_ISSET(sp
, SC_RE_SUBST
)) {
907 regfree(&sp
->subre_c
);
908 F_CLR(sp
, SC_RE_SUBST
);
912 * If we're saving the string, it's a pattern we haven't seen before,
913 * so convert the vi-style RE's to POSIX 1003.2 RE's. Save a copy for
914 * later recompilation. Free any previously saved value.
918 if (LF_ISSET(RE_C_CSCOPE
)) {
919 if (re_cscope_conv(sp
, &ptrn
, &plen
, &replaced
))
923 * Currently, the match-any-<blank> expression used in
924 * re_cscope_conv() requires extended RE's. This may
925 * not be right or safe.
927 reflags
|= REG_EXTENDED
;
928 } else if (LF_ISSET(RE_C_TAG
)) {
929 if (re_tag_conv(sp
, &ptrn
, &plen
, &replaced
))
932 if (re_conv(sp
, &ptrn
, &plen
, &replaced
))
935 /* Discard previous pattern. */
943 * Copy the string into allocated memory.
946 * Regcomp isn't 8-bit clean, so the pattern is nul-terminated
947 * for now. There's just no other solution.
949 MALLOC(sp
, *ptrnp
, (plen
+ 1) * sizeof(CHAR_T
));
950 if (*ptrnp
!= NULL
) {
951 MEMCPY(*ptrnp
, ptrn
, plen
);
952 (*ptrnp
)[plen
] = '\0';
955 /* Free up conversion-routine-allocated memory. */
957 FREE_SPACEW(sp
, ptrn
, 0);
967 * Regcomp isn't 8-bit clean, so we just lost if the pattern
968 * contained a nul. Bummer!
970 if ((rval
= regcomp(rep
, ptrn
, /* plen, */ reflags
)) != 0) {
971 if (!LF_ISSET(RE_C_SILENT
))
972 re_error(sp
, rval
, rep
);
976 if (LF_ISSET(RE_C_SEARCH
))
977 F_SET(sp
, SC_RE_SEARCH
);
978 if (LF_ISSET(RE_C_SUBST
))
979 F_SET(sp
, SC_RE_SUBST
);
986 * Convert vi's regular expressions into something that the
987 * the POSIX 1003.2 RE functions can handle.
989 * There are three conversions we make to make vi's RE's (specifically
990 * the global, search, and substitute patterns) work with POSIX RE's.
992 * 1: If O_MAGIC is not set, strip backslashes from the magic character
993 * set (.[*~) that have them, and add them to the ones that don't.
994 * 2: If O_MAGIC is not set, the string "\~" is replaced with the text
995 * from the last substitute command's replacement string. If O_MAGIC
996 * is set, it's the string "~".
997 * 3: The pattern \<ptrn\> does "word" searches, convert it to use the
1001 * This doesn't exactly match the historic behavior of vi because we do
1002 * the ~ substitution before calling the RE engine, so magic characters
1003 * in the replacement string will be expanded by the RE engine, and they
1004 * weren't historically. It's a bug.
1007 re_conv(SCR
*sp
, CHAR_T
**ptrnp
, size_t *plenp
, int *replacedp
)
1009 size_t blen
, len
, needlen
;
1014 * First pass through, we figure out how much space we'll need.
1015 * We do it in two passes, on the grounds that most of the time
1016 * the user is doing a search and won't have magic characters.
1017 * That way we can skip most of the memory allocation and copies.
1020 for (p
= *ptrnp
, len
= *plenp
, needlen
= 0; len
> 0; ++p
, --len
)
1028 needlen
+= RE_WSTART_LEN
+ 1;
1032 needlen
+= RE_WSTOP_LEN
+ 1;
1035 if (!O_ISSET(sp
, O_MAGIC
)) {
1037 needlen
+= sp
->repl_len
;
1043 if (!O_ISSET(sp
, O_MAGIC
)) {
1055 if (O_ISSET(sp
, O_MAGIC
)) {
1057 needlen
+= sp
->repl_len
;
1063 if (!O_ISSET(sp
, O_MAGIC
)) {
1078 /* Get enough memory to hold the final pattern. */
1080 GET_SPACE_RETW(sp
, bp
, blen
, needlen
);
1082 for (p
= *ptrnp
, len
= *plenp
, t
= bp
; len
> 0; ++p
, --len
)
1090 RE_WSTART
, RE_WSTART_LEN
);
1095 RE_WSTOP
, RE_WSTOP_LEN
);
1099 if (O_ISSET(sp
, O_MAGIC
))
1103 sp
->repl
, sp
->repl_len
);
1110 if (O_ISSET(sp
, O_MAGIC
))
1122 if (O_ISSET(sp
, O_MAGIC
)) {
1123 MEMCPY(t
, sp
->repl
, sp
->repl_len
);
1131 if (!O_ISSET(sp
, O_MAGIC
))
1147 * Convert a tags search path into something that the POSIX
1148 * 1003.2 RE functions can handle.
1151 re_tag_conv(SCR
*sp
, CHAR_T
**ptrnp
, size_t *plenp
, int *replacedp
)
1159 /* Max memory usage is 2 times the length of the string. */
1161 GET_SPACE_RETW(sp
, bp
, blen
, len
* 2);
1166 /* If the last character is a '/' or '?', we just strip it. */
1167 if (len
> 0 && (p
[len
- 1] == '/' || p
[len
- 1] == '?'))
1170 /* If the next-to-last or last character is a '$', it's magic. */
1171 if (len
> 0 && p
[len
- 1] == '$') {
1177 /* If the first character is a '/' or '?', we just strip it. */
1178 if (len
> 0 && (p
[0] == '/' || p
[0] == '?')) {
1183 /* If the first or second character is a '^', it's magic. */
1190 * Escape every other magic character we can find, meanwhile stripping
1191 * the backslashes ctags inserts when escaping the search delimiter
1194 for (; len
> 0; --len
) {
1195 if (p
[0] == '\\' && (p
[1] == '/' || p
[1] == '?')) {
1198 } else if (STRCHR(L("^.[]$*"), p
[0]))
1212 * Convert a cscope search path into something that the POSIX
1213 * 1003.2 RE functions can handle.
1216 re_cscope_conv(SCR
*sp
, CHAR_T
**ptrnp
, size_t *plenp
, int *replacedp
)
1218 size_t blen
, len
, nspaces
;
1225 * Each space in the source line printed by cscope represents an
1226 * arbitrary sequence of spaces, tabs, and comments.
1228 #define CSCOPE_RE_SPACE "([ \t]|/\\*([^*]|\\*/)*\\*/)*"
1229 #define CSCOPE_LEN sizeof(CSCOPE_RE_SPACE) - 1
1230 CHAR2INT(sp
, CSCOPE_RE_SPACE
, CSCOPE_LEN
, wp
, wlen
);
1231 for (nspaces
= 0, p
= *ptrnp
, len
= *plenp
; len
> 0; ++p
, --len
)
1236 * Allocate plenty of space:
1237 * the string, plus potential escaping characters;
1238 * nspaces + 2 copies of CSCOPE_RE_SPACE;
1239 * ^, $, nul terminator characters.
1242 len
= (p
- *ptrnp
) * 2 + (nspaces
+ 2) * sizeof(CSCOPE_RE_SPACE
) + 3;
1243 GET_SPACE_RETW(sp
, bp
, blen
, len
);
1249 MEMCPY(t
, wp
, wlen
);
1252 for (len
= *plenp
; len
> 0; ++p
, --len
)
1254 MEMCPY(t
, wp
, wlen
);
1257 if (STRCHR(L("\\^.[]$*+?()|{}"), *p
))
1262 MEMCPY(t
, wp
, wlen
);
1273 * Report a regular expression error.
1275 * PUBLIC: void re_error(SCR *, int, regex_t *);
1278 re_error(SCR
*sp
, int errcode
, regex_t
*preg
)
1283 s
= regerror(errcode
, preg
, "", 0);
1286 (void)regerror(errcode
, preg
, oe
, s
);
1287 msgq(sp
, M_ERR
, "RE error: %s", oe
);
1294 * Do the substitution for a regular expression.
1299 CHAR_T
*ip
, /* Input line. */
1303 regmatch_t match
[10])
1305 enum { C_NOTSET
, C_LOWER
, C_ONELOWER
, C_ONEUPPER
, C_UPPER
} conv
;
1306 size_t lbclen
, lblen
; /* Local copies. */
1307 size_t mlen
; /* Match length. */
1308 size_t rpl
; /* Remaining replacement length. */
1309 CHAR_T
*rp
; /* Replacement pointer. */
1311 int no
; /* Match replacement offset. */
1312 CHAR_T
*p
, *t
; /* Buffer pointers. */
1313 CHAR_T
*lb
; /* Local copies. */
1315 lb
= *lbp
; /* Get local copies. */
1322 * There are some special sequences that vi provides in the
1323 * replacement patterns.
1324 * & string the RE matched (\& if nomagic set)
1325 * \# n-th regular subexpression
1326 * \E end \U, \L conversion
1327 * \e end \U, \L conversion
1328 * \l convert the next character to lower-case
1329 * \L convert to lower-case, until \E, \e, or end of replacement
1330 * \u convert the next character to upper-case
1331 * \U convert to upper-case, until \E, \e, or end of replacement
1333 * Otherwise, since this is the lowest level of replacement, discard
1334 * all escaping characters. This (hopefully) matches historic practice.
1336 #define OUTCH(ch, nltrans) { \
1337 ARG_CHAR_T __ch = (ch); \
1338 e_key_t __value = KEY_VAL(sp, __ch); \
1339 if (nltrans && (__value == K_CR || __value == K_NL)) { \
1341 sp->newl[sp->newl_cnt++] = lbclen; \
1342 } else if (conv != C_NOTSET) { \
1348 if (ISUPPER(__ch)) \
1349 __ch = TOLOWER(__ch); \
1355 if (ISLOWER(__ch)) \
1356 __ch = TOUPPER(__ch); \
1367 for (rp
= sp
->repl
, rpl
= sp
->repl_len
, p
= lb
+ lbclen
; rpl
--;) {
1368 switch (ch
= *rp
++) {
1370 if (O_ISSET(sp
, O_MAGIC
)) {
1382 if (!O_ISSET(sp
, O_MAGIC
)) {
1387 case '0': case '1': case '2': case '3': case '4':
1388 case '5': case '6': case '7': case '8': case '9':
1390 subzero
: if (match
[no
].rm_so
== -1 ||
1391 match
[no
].rm_eo
== -1)
1393 mlen
= match
[no
].rm_eo
- match
[no
].rm_so
;
1394 for (t
= ip
+ match
[no
].rm_so
; mlen
--; ++t
)
1429 *lbp
= lb
; /* Update caller's information. */