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 10.53 2011/12/21 20:40:35 zy Exp $";
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 *));
51 ex_s(SCR
*sp
, EXCMD
*cmdp
)
57 CHAR_T
*bp
, *p
, *ptrn
, *rep
, *t
;
60 * Skip leading white space.
63 * Historic vi allowed any non-alphanumeric to serve as the
64 * substitution command delimiter.
67 * If the arguments are empty, it's the same as &, i.e. we
68 * repeat the last substitution.
72 for (p
= cmdp
->argv
[0]->bp
,
73 len
= cmdp
->argv
[0]->len
; len
> 0; --len
, ++p
) {
78 subagain
: return (ex_subagain(sp
, cmdp
));
81 if (!isascii(delim
) || isalnum(delim
) || delim
== '\\')
82 return (s(sp
, cmdp
, p
, &sp
->subre_c
, SUB_MUSTSETR
));
86 * The full-blown substitute command reset the remembered
87 * state of the 'c' and 'g' suffices.
89 sp
->c_suffix
= sp
->g_suffix
= 0;
92 * Get the pattern string, toss escaping characters.
95 * Historic vi accepted any of the following forms:
97 * :s/abc/def/ change "abc" to "def"
98 * :s/abc/def change "abc" to "def"
99 * :s/abc/ delete "abc"
100 * :s/abc delete "abc"
104 * Only toss an escaping character if it escapes a delimiter.
105 * This means that "s/A/\\\\f" replaces "A" with "\\f". It
106 * would be nice to be more regular, i.e. for each layer of
107 * escaping a single escaping character is removed, but that's
108 * not how the historic vi worked.
110 for (ptrn
= t
= p
;;) {
111 if (p
[0] == '\0' || p
[0] == delim
) {
116 * Nul terminate the pattern string -- it's passed
117 * to regcomp which doesn't understand anything else.
125 else if (p
[1] == '\\')
131 * If the pattern string is empty, use the last RE (not just the
132 * last substitution RE).
135 if (sp
->re
== NULL
) {
136 ex_emsg(sp
, NULL
, EXM_NOPREVRE
);
140 /* Re-compile the RE if necessary. */
141 if (!F_ISSET(sp
, SC_RE_SEARCH
) &&
142 re_compile(sp
, sp
->re
, sp
->re_len
,
143 NULL
, NULL
, &sp
->re_c
, RE_C_SEARCH
))
149 * Compile the RE. Historic practice is that substitutes set
150 * the search direction as well as both substitute and search
151 * RE's. We compile the RE twice, as we don't want to bother
152 * ref counting the pattern string and (opaque) structure.
154 if (re_compile(sp
, ptrn
, t
- ptrn
, &sp
->re
,
155 &sp
->re_len
, &sp
->re_c
, RE_C_SEARCH
))
157 if (re_compile(sp
, ptrn
, t
- ptrn
, &sp
->subre
,
158 &sp
->subre_len
, &sp
->subre_c
, RE_C_SUBST
))
162 sp
->searchdir
= FORWARD
;
167 * Get the replacement string.
169 * The special character & (\& if O_MAGIC not set) matches the
170 * entire RE. No handling of & is required here, it's done by
173 * The special character ~ (\~ if O_MAGIC not set) inserts the
174 * previous replacement string into this replacement string.
175 * Count ~'s to figure out how much space we need. We could
176 * special case nonexistent last patterns or whether or not
177 * O_MAGIC is set, but it's probably not worth the effort.
181 * Only toss an escaping character if it escapes a delimiter or
182 * if O_MAGIC is set and it escapes a tilde.
185 * If the entire replacement pattern is "%", then use the last
186 * replacement pattern. This semantic was added to vi in System
187 * V and then percolated elsewhere, presumably around the time
188 * that it was added to their version of ed(1).
190 if (p
[0] == '\0' || p
[0] == delim
) {
193 if (sp
->repl
!= NULL
)
197 } else if (p
[0] == '%' && (p
[1] == '\0' || p
[1] == delim
))
198 p
+= p
[1] == delim
? 2 : 1;
200 for (rep
= p
, len
= 0;
201 p
[0] != '\0' && p
[0] != delim
; ++p
, ++len
)
204 GET_SPACE_RETW(sp
, bp
, blen
, len
);
205 for (t
= bp
, len
= 0, p
= rep
;;) {
206 if (p
[0] == '\0' || p
[0] == delim
) {
214 else if (p
[1] == '\\') {
217 } else if (p
[1] == '~') {
219 if (!O_ISSET(sp
, O_MAGIC
))
222 } else if (p
[0] == '~' && O_ISSET(sp
, O_MAGIC
)) {
224 MEMCPY(t
, sp
->repl
, sp
->repl_len
);
232 if ((sp
->repl_len
= len
) != 0) {
233 if (sp
->repl
!= NULL
)
235 MALLOC(sp
, sp
->repl
, CHAR_T
*, len
* sizeof(CHAR_T
));
236 if (sp
->repl
== NULL
) {
237 FREE_SPACEW(sp
, bp
, blen
);
240 MEMCPY(sp
->repl
, bp
, len
);
242 FREE_SPACEW(sp
, bp
, blen
);
244 return (s(sp
, cmdp
, p
, re
, flags
));
249 * [line [,line]] & [cgr] [count] [#lp]]
251 * Substitute using the last substitute RE and replacement pattern.
253 * PUBLIC: int ex_subagain __P((SCR *, EXCMD *));
256 ex_subagain(SCR
*sp
, EXCMD
*cmdp
)
258 if (sp
->subre
== NULL
) {
259 ex_emsg(sp
, NULL
, EXM_NOPREVRE
);
262 if (!F_ISSET(sp
, SC_RE_SUBST
) &&
263 re_compile(sp
, sp
->subre
, sp
->subre_len
,
264 NULL
, NULL
, &sp
->subre_c
, RE_C_SUBST
))
267 cmdp
, cmdp
->argc
? cmdp
->argv
[0]->bp
: NULL
, &sp
->subre_c
, 0));
272 * [line [,line]] ~ [cgr] [count] [#lp]]
274 * Substitute using the last RE and last substitute replacement pattern.
276 * PUBLIC: int ex_subtilde __P((SCR *, EXCMD *));
279 ex_subtilde(SCR
*sp
, EXCMD
*cmdp
)
281 if (sp
->re
== NULL
) {
282 ex_emsg(sp
, NULL
, EXM_NOPREVRE
);
285 if (!F_ISSET(sp
, SC_RE_SEARCH
) && re_compile(sp
, sp
->re
,
286 sp
->re_len
, NULL
, NULL
, &sp
->re_c
, RE_C_SEARCH
))
289 cmdp
, cmdp
->argc
? cmdp
->argv
[0]->bp
: NULL
, &sp
->re_c
, 0));
294 * Do the substitution. This stuff is *really* tricky. There are lots of
295 * special cases, and general nastiness. Don't mess with it unless you're
298 * The nasty part of the substitution is what happens when the replacement
299 * string contains newlines. It's a bit tricky -- consider the information
300 * that has to be retained for "s/f\(o\)o/^M\1^M\1/". The solution here is
301 * to build a set of newline offsets which we use to break the line up later,
302 * when the replacement is done. Don't change it unless you're *damned*
305 #define NEEDNEWLINE(sp) { \
306 if (sp->newl_len == sp->newl_cnt) { \
307 sp->newl_len += 25; \
308 REALLOC(sp, sp->newl, size_t *, \
309 sp->newl_len * sizeof(size_t)); \
310 if (sp->newl == NULL) { \
317 #define BUILD(sp, l, len) { \
318 if (lbclen + (len) > lblen) { \
319 lblen = p2roundup(MAX(lbclen + (len), 256)); \
320 REALLOC(sp, lb, CHAR_T *, lblen * sizeof(CHAR_T)); \
326 MEMCPY(lb + lbclen, l, len); \
330 #define NEEDSP(sp, len, pnt) { \
331 if (lbclen + (len) > lblen) { \
332 lblen = p2roundup(MAX(lbclen + (len), 256)); \
333 REALLOC(sp, lb, CHAR_T *, lblen * sizeof(CHAR_T)); \
343 s(SCR
*sp
, EXCMD
*cmdp
, CHAR_T
*s
, regex_t
*re
, u_int flags
)
347 TEXTH tiq
[] = {{ 0 }};
348 recno_t elno
, lno
, slno
;
350 regmatch_t match
[10];
351 size_t blen
, cnt
, last
, lbclen
, lblen
, len
, llen
;
352 size_t offset
, saved_offset
, scno
;
353 int cflag
, lflag
, nflag
, pflag
, rflag
;
354 int didsub
, do_eol_match
, eflags
, empty_ok
, eval
;
355 int linechanged
, matched
, quit
, rval
;
366 * Historically, the 'g' and 'c' suffices were always toggled as flags,
367 * so ":s/A/B/" was the same as ":s/A/B/ccgg". If O_EDCOMPATIBLE was
368 * not set, they were initialized to 0 for all substitute commands. If
369 * O_EDCOMPATIBLE was set, they were initialized to 0 only if the user
370 * specified substitute/replacement patterns (see ex_s()).
372 if (!O_ISSET(sp
, O_EDCOMPATIBLE
))
373 sp
->c_suffix
= sp
->g_suffix
= 0;
376 * Historic vi permitted the '#', 'l' and 'p' options in vi mode, but
377 * it only displayed the last change. I'd disallow them, but they are
378 * useful in combination with the [v]global commands. In the current
379 * model the problem is combining them with the 'c' flag -- the screen
380 * would have to flip back and forth between the confirm screen and the
381 * ex print screen, which would be pretty awful. We do display all
382 * changes, though, for what that's worth.
385 * Historic vi was fairly strict about the order of "options", the
386 * count, and "flags". I'm somewhat fuzzy on the difference between
387 * options and flags, anyway, so this is a simpler approach, and we
388 * just take it them in whatever order the user gives them. (The ex
389 * usage statement doesn't reflect this.)
391 cflag
= lflag
= nflag
= pflag
= rflag
= 0;
394 for (lno
= OOBLNO
; *s
!= '\0'; ++s
)
405 case '0': case '1': case '2': case '3': case '4':
406 case '5': case '6': case '7': case '8': case '9':
410 nret
= nget_uslong(&ul
, s
, &s
, 10);
412 if (*s
== '\0') /* Loop increment correction. */
414 if (nret
!= NUM_OK
) {
415 if (nret
== NUM_OVER
)
416 msgq(sp
, M_ERR
, "153|Count overflow");
417 else if (nret
== NUM_UNDER
)
418 msgq(sp
, M_ERR
, "154|Count underflow");
420 msgq(sp
, M_SYSERR
, NULL
);
424 * In historic vi, the count was inclusive from the
427 cmdp
->addr1
.lno
= cmdp
->addr2
.lno
;
428 cmdp
->addr2
.lno
+= lno
- 1;
429 if (!db_exist(sp
, cmdp
->addr2
.lno
) &&
430 db_last(sp
, &cmdp
->addr2
.lno
))
437 sp
->c_suffix
= !sp
->c_suffix
;
439 /* Ex text structure initialization. */
440 if (F_ISSET(sp
, SC_EX
))
444 sp
->g_suffix
= !sp
->g_suffix
;
453 if (LF_ISSET(SUB_FIRST
)) {
455 "155|Regular expression specified; r flag meaningless");
458 if (!F_ISSET(sp
, SC_RE_SEARCH
)) {
459 ex_emsg(sp
, NULL
, EXM_NOPREVRE
);
469 if (*s
!= '\0' || (!rflag
&& LF_ISSET(SUB_MUSTSETR
))) {
470 usage
: ex_emsg(sp
, cmdp
->cmd
->usage
, EXM_USAGE
);
474 noargs
: if (F_ISSET(sp
, SC_VI
) && sp
->c_suffix
&& (lflag
|| nflag
|| pflag
)) {
476 "156|The #, l and p flags may not be combined with the c flag in vi mode");
481 * bp: if interactive, line cache
482 * blen: if interactive, line cache length
483 * lb: build buffer pointer.
484 * lbclen: current length of built buffer.
485 * lblen; length of build buffer.
488 blen
= lbclen
= lblen
= 0;
490 /* For each line... */
491 lno
= cmdp
->addr1
.lno
== 0 ? 1 : cmdp
->addr1
.lno
;
492 for (matched
= quit
= 0,
493 elno
= cmdp
->addr2
.lno
; !quit
&& lno
<= elno
; ++lno
) {
495 /* Someone's unhappy, time to stop. */
500 if (db_get(sp
, lno
, DBG_FATAL
, &s
, &llen
))
504 * Make a local copy if doing confirmation -- when calling
505 * the confirm routine we're likely to lose the cached copy.
509 GET_SPACE_RETW(sp
, bp
, blen
, llen
);
511 ADD_SPACE_RETW(sp
, bp
, blen
, llen
);
516 /* Start searching from the beginning. */
520 /* Reset the build buffer offset. */
523 /* Reset empty match flag. */
527 * We don't want to have to do a setline if the line didn't
528 * change -- keep track of whether or not this line changed.
529 * If doing confirmations, don't want to keep setting the
530 * line if change is refused -- keep track of substitutions.
532 didsub
= linechanged
= 0;
534 /* New line, do an EOL match. */
537 /* It's not nul terminated, but we pretend it is. */
538 eflags
= REG_STARTEND
;
541 * The search area is from s + offset to the EOL.
543 * Generally, match[0].rm_so is the offset of the start
544 * of the match from the start of the search, and offset
545 * is the offset of the start of the last search.
547 nextmatch
: match
[0].rm_so
= 0;
548 match
[0].rm_eo
= len
;
550 /* Get the next match. */
551 eval
= regexec(re
, s
+ offset
, 10, match
, eflags
);
554 * There wasn't a match or if there was an error, deal with
555 * it. If there was a previous match in this line, resolve
556 * the changes into the database. Otherwise, just move on.
558 if (eval
== REG_NOMATCH
)
561 re_error(sp
, eval
, re
);
566 /* Only the first search can match an anchored expression. */
567 eflags
|= REG_NOTBOL
;
571 * It's possible to match 0-length strings -- for example, the
572 * command s;a*;X;, when matched against the string "aabb" will
573 * result in "XbXbX", i.e. the matches are "aa", the space
574 * between the b's and the space between the b's and the end of
575 * the string. There is a similar space between the beginning
576 * of the string and the a's. The rule that we use (because vi
577 * historically used it) is that any 0-length match, occurring
578 * immediately after a match, is ignored. Otherwise, the above
579 * example would have resulted in "XXbXbX". Another example is
580 * incorrectly using " *" to replace groups of spaces with one
583 * The way we do this is that if we just had a successful match,
584 * the starting offset does not skip characters, and the match
585 * is empty, ignore the match and move forward. If there's no
586 * more characters in the string, we were attempting to match
587 * after the last character, so quit.
589 if (!empty_ok
&& match
[0].rm_so
== 0 && match
[0].rm_eo
== 0) {
593 BUILD(sp
, s
+ offset
, 1)
599 /* Confirm change. */
602 * Set the cursor position for confirmation. Note,
603 * if we matched on a '$', the cursor may be past
606 from
.lno
= to
.lno
= lno
;
607 from
.cno
= match
[0].rm_so
+ offset
;
608 to
.cno
= match
[0].rm_eo
+ offset
;
610 * Both ex and vi have to correct for a change before
611 * the first character in the line.
614 from
.cno
= to
.cno
= 0;
615 if (F_ISSET(sp
, SC_VI
)) {
617 * Only vi has to correct for a change after
618 * the last character in the line.
621 * It would be nice to change the vi code so
622 * that we could display a cursor past EOL.
626 if (from
.cno
>= llen
)
631 if (vs_refresh(sp
, 1))
634 vs_update(sp
, msg_cat(sp
,
635 "169|Confirm change? [n]", NULL
), NULL
);
637 if (v_event_get(sp
, &ev
, 0, 0))
639 switch (ev
.e_event
) {
647 v_event_err(sp
, &ev
);
651 if (ex_print(sp
, cmdp
, &from
, &to
, 0) ||
652 ex_scprint(sp
, &from
, &to
))
654 if (ex_txt(sp
, tiq
, 0, TXT_CR
))
656 ev
.e_c
= TAILQ_FIRST(tiq
)->lb
[0];
665 BUILD(sp
, s
+offset
, match
[0].rm_eo
);
668 /* Set the quit/interrupted flags. */
670 F_SET(sp
->gp
, G_INTERRUPTED
);
673 * Resolve any changes, then return to (and
674 * exit from) the main loop.
681 * Set the cursor to the last position changed, converting
682 * from 1-based to 0-based.
685 sp
->cno
= match
[0].rm_so
;
687 /* Copy the bytes before the match into the build buffer. */
688 BUILD(sp
, s
+ offset
, match
[0].rm_so
);
690 /* Substitute the matching bytes. */
692 if (re_sub(sp
, s
+ offset
, &lb
, &lbclen
, &lblen
, match
))
695 /* Set the change flag so we know this line was modified. */
698 /* Move past the matched bytes. */
699 skip
: offset
+= match
[0].rm_eo
;
700 len
-= match
[0].rm_eo
;
702 /* A match cannot be followed by an empty pattern. */
706 * If doing a global change with confirmation, we have to
707 * update the screen. The basic idea is to store the line
708 * so the screen update routines can find it, and restart.
710 if (didsub
&& sp
->c_suffix
&& sp
->g_suffix
) {
712 * The new search offset will be the end of the
715 saved_offset
= lbclen
;
717 /* Copy the rest of the line. */
719 BUILD(sp
, s
+ offset
, len
)
721 /* Set the new offset. */
722 offset
= saved_offset
;
724 /* Store inserted lines, adjusting the build buffer. */
728 cnt
< sp
->newl_cnt
; ++cnt
, ++lno
, ++elno
) {
729 if (db_insert(sp
, lno
,
730 lb
+ last
, sp
->newl
[cnt
] - last
))
732 last
= sp
->newl
[cnt
] + 1;
733 ++sp
->rptlines
[L_ADDED
];
740 /* Store and retrieve the line. */
741 if (db_set(sp
, lno
, lb
+ last
, lbclen
))
743 if (db_get(sp
, lno
, DBG_FATAL
, &s
, &llen
))
745 ADD_SPACE_RETW(sp
, bp
, blen
, llen
)
750 /* Restart the build. */
752 BUILD(sp
, s
, offset
);
755 * If we haven't already done the after-the-string
756 * match, do one. Set REG_NOTEOL so the '$' pattern
763 eflags
|= REG_NOTEOL
;
771 * If at the end of the string, do a test for the after
772 * the string match. Set REG_NOTEOL so the '$' pattern
775 if (sp
->g_suffix
&& do_eol_match
) {
778 eflags
|= REG_NOTEOL
;
783 endmatch
: if (!linechanged
)
786 /* Copy any remaining bytes into the build buffer. */
788 BUILD(sp
, s
+ offset
, len
)
790 /* Store inserted lines, adjusting the build buffer. */
794 cnt
< sp
->newl_cnt
; ++cnt
, ++lno
, ++elno
) {
796 lno
, lb
+ last
, sp
->newl
[cnt
] - last
))
798 last
= sp
->newl
[cnt
] + 1;
799 ++sp
->rptlines
[L_ADDED
];
805 /* Store the changed line. */
806 if (db_set(sp
, lno
, lb
+ last
, lbclen
))
809 /* Update changed line counter. */
810 if (sp
->rptlchange
!= lno
) {
811 sp
->rptlchange
= lno
;
812 ++sp
->rptlines
[L_CHANGED
];
817 * Display as necessary. Historic practice is to only
818 * display the last line of a line split into multiple
821 if (lflag
|| nflag
|| pflag
) {
822 from
.lno
= to
.lno
= lno
;
823 from
.cno
= to
.cno
= 0;
825 (void)ex_print(sp
, cmdp
, &from
, &to
, E_C_LIST
);
827 (void)ex_print(sp
, cmdp
, &from
, &to
, E_C_HASH
);
829 (void)ex_print(sp
, cmdp
, &from
, &to
, E_C_PRINT
);
835 * Historically, vi attempted to leave the cursor at the same place if
836 * the substitution was done at the current cursor position. Otherwise
837 * it moved it to the first non-blank of the last line changed. There
838 * were some problems: for example, :s/$/foo/ with the cursor on the
839 * last character of the line left the cursor on the last character, or
840 * the & command with multiple occurrences of the matching string in the
841 * line usually left the cursor in a fairly random position.
843 * We try to do the same thing, with the exception that if the user is
844 * doing substitution with confirmation, we move to the last line about
845 * which the user was consulted, as opposed to the last line that they
846 * actually changed. This prevents a screen flash if the user doesn't
847 * change many of the possible lines.
849 if (!sp
->c_suffix
&& (sp
->lno
!= slno
|| sp
->cno
!= scno
)) {
851 (void)nonblank(sp
, sp
->lno
, &sp
->cno
);
855 * If not in a global command, and nothing matched, say so.
856 * Else, if none of the lines displayed, put something up.
860 if (!F_ISSET(sp
, SC_EX_GLOBAL
)) {
861 msgq(sp
, M_ERR
, "157|No match found");
864 } else if (!lflag
&& !nflag
&& !pflag
)
865 F_SET(cmdp
, E_AUTOPRINT
);
872 FREE_SPACEW(sp
, bp
, blen
);
882 * PUBLIC: int re_compile __P((SCR *,
883 * PUBLIC: CHAR_T *, size_t, CHAR_T **, size_t *, regex_t *, u_int));
886 re_compile(SCR
*sp
, CHAR_T
*ptrn
, size_t plen
, CHAR_T
**ptrnp
, size_t *lenp
, regex_t
*rep
, u_int flags
)
889 int reflags
, replaced
, rval
;
894 if (!LF_ISSET(RE_C_CSCOPE
| RE_C_TAG
)) {
895 if (O_ISSET(sp
, O_EXTENDED
))
896 reflags
|= REG_EXTENDED
;
897 if (O_ISSET(sp
, O_IGNORECASE
))
898 reflags
|= REG_ICASE
;
899 if (O_ISSET(sp
, O_ICLOWER
)) {
900 for (p
= ptrn
, len
= plen
; len
> 0; ++p
, --len
)
904 reflags
|= REG_ICASE
;
908 /* If we're replacing a saved value, clear the old one. */
909 if (LF_ISSET(RE_C_SEARCH
) && F_ISSET(sp
, SC_RE_SEARCH
)) {
911 F_CLR(sp
, SC_RE_SEARCH
);
913 if (LF_ISSET(RE_C_SUBST
) && F_ISSET(sp
, SC_RE_SUBST
)) {
914 regfree(&sp
->subre_c
);
915 F_CLR(sp
, SC_RE_SUBST
);
919 * If we're saving the string, it's a pattern we haven't seen before,
920 * so convert the vi-style RE's to POSIX 1003.2 RE's. Save a copy for
921 * later recompilation. Free any previously saved value.
925 if (LF_ISSET(RE_C_CSCOPE
)) {
926 if (re_cscope_conv(sp
, &ptrn
, &plen
, &replaced
))
930 * Currently, the match-any-<blank> expression used in
931 * re_cscope_conv() requires extended RE's. This may
932 * not be right or safe.
934 reflags
|= REG_EXTENDED
;
935 } else if (LF_ISSET(RE_C_TAG
)) {
936 if (re_tag_conv(sp
, &ptrn
, &plen
, &replaced
))
939 if (re_conv(sp
, &ptrn
, &plen
, &replaced
))
942 /* Discard previous pattern. */
943 if (*ptrnp
!= NULL
) {
951 * Copy the string into allocated memory.
954 * Regcomp isn't 8-bit clean, so the pattern is nul-terminated
955 * for now. There's just no other solution.
957 MALLOC(sp
, *ptrnp
, CHAR_T
*, (plen
+ 1) * sizeof(CHAR_T
));
958 if (*ptrnp
!= NULL
) {
959 MEMCPY(*ptrnp
, ptrn
, plen
);
960 (*ptrnp
)[plen
] = '\0';
963 /* Free up conversion-routine-allocated memory. */
965 FREE_SPACEW(sp
, ptrn
, 0);
975 * Regcomp isn't 8-bit clean, so we just lost if the pattern
976 * contained a nul. Bummer!
978 if ((rval
= regcomp(rep
, ptrn
, /* plen, */ reflags
)) != 0) {
979 if (!LF_ISSET(RE_C_SILENT
))
980 re_error(sp
, rval
, rep
);
984 if (LF_ISSET(RE_C_SEARCH
))
985 F_SET(sp
, SC_RE_SEARCH
);
986 if (LF_ISSET(RE_C_SUBST
))
987 F_SET(sp
, SC_RE_SUBST
);
994 * Convert vi's regular expressions into something that the
995 * the POSIX 1003.2 RE functions can handle.
997 * There are three conversions we make to make vi's RE's (specifically
998 * the global, search, and substitute patterns) work with POSIX RE's.
1000 * 1: If O_MAGIC is not set, strip backslashes from the magic character
1001 * set (.[*~) that have them, and add them to the ones that don't.
1002 * 2: If O_MAGIC is not set, the string "\~" is replaced with the text
1003 * from the last substitute command's replacement string. If O_MAGIC
1004 * is set, it's the string "~".
1005 * 3: The pattern \<ptrn\> does "word" searches, convert it to use the
1009 * This doesn't exactly match the historic behavior of vi because we do
1010 * the ~ substitution before calling the RE engine, so magic characters
1011 * in the replacement string will be expanded by the RE engine, and they
1012 * weren't historically. It's a bug.
1015 re_conv(SCR
*sp
, CHAR_T
**ptrnp
, size_t *plenp
, int *replacedp
)
1017 size_t blen
, len
, needlen
;
1022 * First pass through, we figure out how much space we'll need.
1023 * We do it in two passes, on the grounds that most of the time
1024 * the user is doing a search and won't have magic characters.
1025 * That way we can skip most of the memory allocation and copies.
1028 for (p
= *ptrnp
, len
= *plenp
, needlen
= 0; len
> 0; ++p
, --len
)
1036 needlen
+= RE_WSTART_LEN
+ 1;
1040 needlen
+= RE_WSTOP_LEN
+ 1;
1043 if (!O_ISSET(sp
, O_MAGIC
)) {
1045 needlen
+= sp
->repl_len
;
1051 if (!O_ISSET(sp
, O_MAGIC
)) {
1063 if (O_ISSET(sp
, O_MAGIC
)) {
1065 needlen
+= sp
->repl_len
;
1071 if (!O_ISSET(sp
, O_MAGIC
)) {
1086 /* Get enough memory to hold the final pattern. */
1088 GET_SPACE_RETW(sp
, bp
, blen
, needlen
);
1090 for (p
= *ptrnp
, len
= *plenp
, t
= bp
; len
> 0; ++p
, --len
)
1098 RE_WSTART
, RE_WSTART_LEN
);
1103 RE_WSTOP
, RE_WSTOP_LEN
);
1107 if (O_ISSET(sp
, O_MAGIC
))
1111 sp
->repl
, sp
->repl_len
);
1118 if (O_ISSET(sp
, O_MAGIC
))
1130 if (O_ISSET(sp
, O_MAGIC
)) {
1131 MEMCPY(t
, sp
->repl
, sp
->repl_len
);
1139 if (!O_ISSET(sp
, O_MAGIC
))
1155 * Convert a tags search path into something that the POSIX
1156 * 1003.2 RE functions can handle.
1159 re_tag_conv(SCR
*sp
, CHAR_T
**ptrnp
, size_t *plenp
, int *replacedp
)
1167 /* Max memory usage is 2 times the length of the string. */
1169 GET_SPACE_RETW(sp
, bp
, blen
, len
* 2);
1174 /* If the last character is a '/' or '?', we just strip it. */
1175 if (len
> 0 && (p
[len
- 1] == '/' || p
[len
- 1] == '?'))
1178 /* If the next-to-last or last character is a '$', it's magic. */
1179 if (len
> 0 && p
[len
- 1] == '$') {
1185 /* If the first character is a '/' or '?', we just strip it. */
1186 if (len
> 0 && (p
[0] == '/' || p
[0] == '?')) {
1191 /* If the first or second character is a '^', it's magic. */
1198 * Escape every other magic character we can find, meanwhile stripping
1199 * the backslashes ctags inserts when escaping the search delimiter
1202 for (; len
> 0; --len
) {
1203 if (p
[0] == '\\' && (p
[1] == '/' || p
[1] == '?')) {
1206 } else if (STRCHR(L("^.[]$*"), p
[0]))
1220 * Convert a cscope search path into something that the POSIX
1221 * 1003.2 RE functions can handle.
1224 re_cscope_conv(SCR
*sp
, CHAR_T
**ptrnp
, size_t *plenp
, int *replacedp
)
1226 size_t blen
, len
, nspaces
;
1233 * Each space in the source line printed by cscope represents an
1234 * arbitrary sequence of spaces, tabs, and comments.
1236 #define CSCOPE_RE_SPACE "([ \t]|/\\*([^*]|\\*/)*\\*/)*"
1237 #define CSCOPE_LEN sizeof(CSCOPE_RE_SPACE) - 1
1238 CHAR2INT(sp
, CSCOPE_RE_SPACE
, CSCOPE_LEN
, wp
, wlen
);
1239 for (nspaces
= 0, p
= *ptrnp
, len
= *plenp
; len
> 0; ++p
, --len
)
1244 * Allocate plenty of space:
1245 * the string, plus potential escaping characters;
1246 * nspaces + 2 copies of CSCOPE_RE_SPACE;
1247 * ^, $, nul terminator characters.
1250 len
= (p
- *ptrnp
) * 2 + (nspaces
+ 2) * sizeof(CSCOPE_RE_SPACE
) + 3;
1251 GET_SPACE_RETW(sp
, bp
, blen
, len
);
1257 MEMCPY(t
, wp
, wlen
);
1260 for (len
= *plenp
; len
> 0; ++p
, --len
)
1262 MEMCPY(t
, wp
, wlen
);
1265 if (STRCHR(L("\\^.[]$*+?()|{}"), *p
))
1270 MEMCPY(t
, wp
, wlen
);
1281 * Report a regular expression error.
1283 * PUBLIC: void re_error __P((SCR *, int, regex_t *));
1286 re_error(SCR
*sp
, int errcode
, regex_t
*preg
)
1291 s
= regerror(errcode
, preg
, "", 0);
1292 MALLOC(sp
, oe
, char *, s
);
1294 (void)regerror(errcode
, preg
, oe
, s
);
1295 msgq(sp
, M_ERR
, "RE error: %s", oe
);
1302 * Do the substitution for a regular expression.
1307 CHAR_T
*ip
, /* Input line. */
1311 regmatch_t match
[10])
1313 enum { C_NOTSET
, C_LOWER
, C_ONELOWER
, C_ONEUPPER
, C_UPPER
} conv
;
1314 size_t lbclen
, lblen
; /* Local copies. */
1315 size_t mlen
; /* Match length. */
1316 size_t rpl
; /* Remaining replacement length. */
1317 CHAR_T
*rp
; /* Replacement pointer. */
1319 int no
; /* Match replacement offset. */
1320 CHAR_T
*p
, *t
; /* Buffer pointers. */
1321 CHAR_T
*lb
; /* Local copies. */
1323 lb
= *lbp
; /* Get local copies. */
1330 * There are some special sequences that vi provides in the
1331 * replacement patterns.
1332 * & string the RE matched (\& if nomagic set)
1333 * \# n-th regular subexpression
1334 * \E end \U, \L conversion
1335 * \e end \U, \L conversion
1336 * \l convert the next character to lower-case
1337 * \L convert to lower-case, until \E, \e, or end of replacement
1338 * \u convert the next character to upper-case
1339 * \U convert to upper-case, until \E, \e, or end of replacement
1341 * Otherwise, since this is the lowest level of replacement, discard
1342 * all escaping characters. This (hopefully) matches historic practice.
1344 #define OUTCH(ch, nltrans) { \
1345 ARG_CHAR_T __ch = (ch); \
1346 e_key_t __value = KEY_VAL(sp, __ch); \
1347 if (nltrans && (__value == K_CR || __value == K_NL)) { \
1349 sp->newl[sp->newl_cnt++] = lbclen; \
1350 } else if (conv != C_NOTSET) { \
1356 if (ISUPPER(__ch)) \
1357 __ch = TOLOWER(__ch); \
1363 if (ISLOWER(__ch)) \
1364 __ch = TOUPPER(__ch); \
1375 for (rp
= sp
->repl
, rpl
= sp
->repl_len
, p
= lb
+ lbclen
; rpl
--;) {
1376 switch (ch
= *rp
++) {
1378 if (O_ISSET(sp
, O_MAGIC
)) {
1390 if (!O_ISSET(sp
, O_MAGIC
)) {
1395 case '0': case '1': case '2': case '3': case '4':
1396 case '5': case '6': case '7': case '8': case '9':
1398 subzero
: if (match
[no
].rm_so
== -1 ||
1399 match
[no
].rm_eo
== -1)
1401 mlen
= match
[no
].rm_eo
- match
[no
].rm_so
;
1402 for (t
= ip
+ match
[no
].rm_so
; mlen
--; ++t
)
1437 *lbp
= lb
; /* Update caller's information. */