Git 2.30.3
[git.git] / grep.c
blobefeb6dc58dbec7dad4ea715909d92780faa27ee0
1 #include "cache.h"
2 #include "config.h"
3 #include "grep.h"
4 #include "object-store.h"
5 #include "userdiff.h"
6 #include "xdiff-interface.h"
7 #include "diff.h"
8 #include "diffcore.h"
9 #include "commit.h"
10 #include "quote.h"
11 #include "help.h"
13 static int grep_source_load(struct grep_source *gs);
14 static int grep_source_is_binary(struct grep_source *gs,
15 struct index_state *istate);
17 static void std_output(struct grep_opt *opt, const void *buf, size_t size)
19 fwrite(buf, size, 1, stdout);
22 static struct grep_opt grep_defaults = {
23 .relative = 1,
24 .pathname = 1,
25 .max_depth = -1,
26 .pattern_type_option = GREP_PATTERN_TYPE_UNSPECIFIED,
27 .colors = {
28 [GREP_COLOR_CONTEXT] = "",
29 [GREP_COLOR_FILENAME] = "",
30 [GREP_COLOR_FUNCTION] = "",
31 [GREP_COLOR_LINENO] = "",
32 [GREP_COLOR_COLUMNNO] = "",
33 [GREP_COLOR_MATCH_CONTEXT] = GIT_COLOR_BOLD_RED,
34 [GREP_COLOR_MATCH_SELECTED] = GIT_COLOR_BOLD_RED,
35 [GREP_COLOR_SELECTED] = "",
36 [GREP_COLOR_SEP] = GIT_COLOR_CYAN,
38 .only_matching = 0,
39 .color = -1,
40 .output = std_output,
43 #ifdef USE_LIBPCRE2
44 static pcre2_general_context *pcre2_global_context;
46 static void *pcre2_malloc(PCRE2_SIZE size, MAYBE_UNUSED void *memory_data)
48 return malloc(size);
51 static void pcre2_free(void *pointer, MAYBE_UNUSED void *memory_data)
53 free(pointer);
55 #endif
57 static const char *color_grep_slots[] = {
58 [GREP_COLOR_CONTEXT] = "context",
59 [GREP_COLOR_FILENAME] = "filename",
60 [GREP_COLOR_FUNCTION] = "function",
61 [GREP_COLOR_LINENO] = "lineNumber",
62 [GREP_COLOR_COLUMNNO] = "column",
63 [GREP_COLOR_MATCH_CONTEXT] = "matchContext",
64 [GREP_COLOR_MATCH_SELECTED] = "matchSelected",
65 [GREP_COLOR_SELECTED] = "selected",
66 [GREP_COLOR_SEP] = "separator",
69 static int parse_pattern_type_arg(const char *opt, const char *arg)
71 if (!strcmp(arg, "default"))
72 return GREP_PATTERN_TYPE_UNSPECIFIED;
73 else if (!strcmp(arg, "basic"))
74 return GREP_PATTERN_TYPE_BRE;
75 else if (!strcmp(arg, "extended"))
76 return GREP_PATTERN_TYPE_ERE;
77 else if (!strcmp(arg, "fixed"))
78 return GREP_PATTERN_TYPE_FIXED;
79 else if (!strcmp(arg, "perl"))
80 return GREP_PATTERN_TYPE_PCRE;
81 die("bad %s argument: %s", opt, arg);
84 define_list_config_array_extra(color_grep_slots, {"match"});
87 * Read the configuration file once and store it in
88 * the grep_defaults template.
90 int grep_config(const char *var, const char *value, void *cb)
92 struct grep_opt *opt = &grep_defaults;
93 const char *slot;
95 if (userdiff_config(var, value) < 0)
96 return -1;
99 * The instance of grep_opt that we set up here is copied by
100 * grep_init() to be used by each individual invocation.
101 * When populating a new field of this structure here, be
102 * sure to think about ownership -- e.g., you might need to
103 * override the shallow copy in grep_init() with a deep copy.
106 if (!strcmp(var, "grep.extendedregexp")) {
107 opt->extended_regexp_option = git_config_bool(var, value);
108 return 0;
111 if (!strcmp(var, "grep.patterntype")) {
112 opt->pattern_type_option = parse_pattern_type_arg(var, value);
113 return 0;
116 if (!strcmp(var, "grep.linenumber")) {
117 opt->linenum = git_config_bool(var, value);
118 return 0;
120 if (!strcmp(var, "grep.column")) {
121 opt->columnnum = git_config_bool(var, value);
122 return 0;
125 if (!strcmp(var, "grep.fullname")) {
126 opt->relative = !git_config_bool(var, value);
127 return 0;
130 if (!strcmp(var, "color.grep"))
131 opt->color = git_config_colorbool(var, value);
132 if (!strcmp(var, "color.grep.match")) {
133 if (grep_config("color.grep.matchcontext", value, cb) < 0)
134 return -1;
135 if (grep_config("color.grep.matchselected", value, cb) < 0)
136 return -1;
137 } else if (skip_prefix(var, "color.grep.", &slot)) {
138 int i = LOOKUP_CONFIG(color_grep_slots, slot);
139 char *color;
141 if (i < 0)
142 return -1;
143 color = opt->colors[i];
144 if (!value)
145 return config_error_nonbool(var);
146 return color_parse(value, color);
148 return 0;
152 * Initialize one instance of grep_opt and copy the
153 * default values from the template we read the configuration
154 * information in an earlier call to git_config(grep_config).
156 * If using PCRE, make sure that the library is configured
157 * to use the same allocator as Git (e.g. nedmalloc on Windows).
159 * Any allocated memory needs to be released in grep_destroy().
161 void grep_init(struct grep_opt *opt, struct repository *repo, const char *prefix)
163 #if defined(USE_LIBPCRE2)
164 if (!pcre2_global_context)
165 pcre2_global_context = pcre2_general_context_create(
166 pcre2_malloc, pcre2_free, NULL);
167 #endif
169 #ifdef USE_LIBPCRE1
170 pcre_malloc = malloc;
171 pcre_free = free;
172 #endif
174 *opt = grep_defaults;
176 opt->repo = repo;
177 opt->prefix = prefix;
178 opt->prefix_length = (prefix && *prefix) ? strlen(prefix) : 0;
179 opt->pattern_tail = &opt->pattern_list;
180 opt->header_tail = &opt->header_list;
183 void grep_destroy(void)
185 #ifdef USE_LIBPCRE2
186 pcre2_general_context_free(pcre2_global_context);
187 #endif
190 static void grep_set_pattern_type_option(enum grep_pattern_type pattern_type, struct grep_opt *opt)
193 * When committing to the pattern type by setting the relevant
194 * fields in grep_opt it's generally not necessary to zero out
195 * the fields we're not choosing, since they won't have been
196 * set by anything. The extended_regexp_option field is the
197 * only exception to this.
199 * This is because in the process of parsing grep.patternType
200 * & grep.extendedRegexp we set opt->pattern_type_option and
201 * opt->extended_regexp_option, respectively. We then
202 * internally use opt->extended_regexp_option to see if we're
203 * compiling an ERE. It must be unset if that's not actually
204 * the case.
206 if (pattern_type != GREP_PATTERN_TYPE_ERE &&
207 opt->extended_regexp_option)
208 opt->extended_regexp_option = 0;
210 switch (pattern_type) {
211 case GREP_PATTERN_TYPE_UNSPECIFIED:
212 /* fall through */
214 case GREP_PATTERN_TYPE_BRE:
215 break;
217 case GREP_PATTERN_TYPE_ERE:
218 opt->extended_regexp_option = 1;
219 break;
221 case GREP_PATTERN_TYPE_FIXED:
222 opt->fixed = 1;
223 break;
225 case GREP_PATTERN_TYPE_PCRE:
226 #ifdef USE_LIBPCRE2
227 opt->pcre2 = 1;
228 #else
230 * It's important that pcre1 always be assigned to
231 * even when there's no USE_LIBPCRE* defined. We still
232 * call the PCRE stub function, it just dies with
233 * "cannot use Perl-compatible regexes[...]".
235 opt->pcre1 = 1;
236 #endif
237 break;
241 void grep_commit_pattern_type(enum grep_pattern_type pattern_type, struct grep_opt *opt)
243 if (pattern_type != GREP_PATTERN_TYPE_UNSPECIFIED)
244 grep_set_pattern_type_option(pattern_type, opt);
245 else if (opt->pattern_type_option != GREP_PATTERN_TYPE_UNSPECIFIED)
246 grep_set_pattern_type_option(opt->pattern_type_option, opt);
247 else if (opt->extended_regexp_option)
249 * This branch *must* happen after setting from the
250 * opt->pattern_type_option above, we don't want
251 * grep.extendedRegexp to override grep.patternType!
253 grep_set_pattern_type_option(GREP_PATTERN_TYPE_ERE, opt);
256 static struct grep_pat *create_grep_pat(const char *pat, size_t patlen,
257 const char *origin, int no,
258 enum grep_pat_token t,
259 enum grep_header_field field)
261 struct grep_pat *p = xcalloc(1, sizeof(*p));
262 p->pattern = xmemdupz(pat, patlen);
263 p->patternlen = patlen;
264 p->origin = origin;
265 p->no = no;
266 p->token = t;
267 p->field = field;
268 return p;
271 static void do_append_grep_pat(struct grep_pat ***tail, struct grep_pat *p)
273 **tail = p;
274 *tail = &p->next;
275 p->next = NULL;
277 switch (p->token) {
278 case GREP_PATTERN: /* atom */
279 case GREP_PATTERN_HEAD:
280 case GREP_PATTERN_BODY:
281 for (;;) {
282 struct grep_pat *new_pat;
283 size_t len = 0;
284 char *cp = p->pattern + p->patternlen, *nl = NULL;
285 while (++len <= p->patternlen) {
286 if (*(--cp) == '\n') {
287 nl = cp;
288 break;
291 if (!nl)
292 break;
293 new_pat = create_grep_pat(nl + 1, len - 1, p->origin,
294 p->no, p->token, p->field);
295 new_pat->next = p->next;
296 if (!p->next)
297 *tail = &new_pat->next;
298 p->next = new_pat;
299 *nl = '\0';
300 p->patternlen -= len;
302 break;
303 default:
304 break;
308 void append_header_grep_pattern(struct grep_opt *opt,
309 enum grep_header_field field, const char *pat)
311 struct grep_pat *p = create_grep_pat(pat, strlen(pat), "header", 0,
312 GREP_PATTERN_HEAD, field);
313 if (field == GREP_HEADER_REFLOG)
314 opt->use_reflog_filter = 1;
315 do_append_grep_pat(&opt->header_tail, p);
318 void append_grep_pattern(struct grep_opt *opt, const char *pat,
319 const char *origin, int no, enum grep_pat_token t)
321 append_grep_pat(opt, pat, strlen(pat), origin, no, t);
324 void append_grep_pat(struct grep_opt *opt, const char *pat, size_t patlen,
325 const char *origin, int no, enum grep_pat_token t)
327 struct grep_pat *p = create_grep_pat(pat, patlen, origin, no, t, 0);
328 do_append_grep_pat(&opt->pattern_tail, p);
331 struct grep_opt *grep_opt_dup(const struct grep_opt *opt)
333 struct grep_pat *pat;
334 struct grep_opt *ret = xmalloc(sizeof(struct grep_opt));
335 *ret = *opt;
337 ret->pattern_list = NULL;
338 ret->pattern_tail = &ret->pattern_list;
340 for(pat = opt->pattern_list; pat != NULL; pat = pat->next)
342 if(pat->token == GREP_PATTERN_HEAD)
343 append_header_grep_pattern(ret, pat->field,
344 pat->pattern);
345 else
346 append_grep_pat(ret, pat->pattern, pat->patternlen,
347 pat->origin, pat->no, pat->token);
350 return ret;
353 static NORETURN void compile_regexp_failed(const struct grep_pat *p,
354 const char *error)
356 char where[1024];
358 if (p->no)
359 xsnprintf(where, sizeof(where), "In '%s' at %d, ", p->origin, p->no);
360 else if (p->origin)
361 xsnprintf(where, sizeof(where), "%s, ", p->origin);
362 else
363 where[0] = 0;
365 die("%s'%s': %s", where, p->pattern, error);
368 static int is_fixed(const char *s, size_t len)
370 size_t i;
372 for (i = 0; i < len; i++) {
373 if (is_regex_special(s[i]))
374 return 0;
377 return 1;
380 #ifdef USE_LIBPCRE1
381 static void compile_pcre1_regexp(struct grep_pat *p, const struct grep_opt *opt)
383 const char *error;
384 int erroffset;
385 int options = PCRE_MULTILINE;
386 int study_options = 0;
388 if (opt->ignore_case) {
389 if (!opt->ignore_locale && has_non_ascii(p->pattern))
390 p->pcre1_tables = pcre_maketables();
391 options |= PCRE_CASELESS;
393 if (!opt->ignore_locale && is_utf8_locale() && has_non_ascii(p->pattern))
394 options |= PCRE_UTF8;
396 p->pcre1_regexp = pcre_compile(p->pattern, options, &error, &erroffset,
397 p->pcre1_tables);
398 if (!p->pcre1_regexp)
399 compile_regexp_failed(p, error);
401 #if defined(PCRE_CONFIG_JIT) && !defined(NO_LIBPCRE1_JIT)
402 pcre_config(PCRE_CONFIG_JIT, &p->pcre1_jit_on);
403 if (opt->debug)
404 fprintf(stderr, "pcre1_jit_on=%d\n", p->pcre1_jit_on);
406 if (p->pcre1_jit_on)
407 study_options = PCRE_STUDY_JIT_COMPILE;
408 #endif
410 p->pcre1_extra_info = pcre_study(p->pcre1_regexp, study_options, &error);
411 if (!p->pcre1_extra_info && error)
412 die("%s", error);
415 static int pcre1match(struct grep_pat *p, const char *line, const char *eol,
416 regmatch_t *match, int eflags)
418 int ovector[30], ret, flags = PCRE_NO_UTF8_CHECK;
420 if (eflags & REG_NOTBOL)
421 flags |= PCRE_NOTBOL;
423 ret = pcre_exec(p->pcre1_regexp, p->pcre1_extra_info, line,
424 eol - line, 0, flags, ovector,
425 ARRAY_SIZE(ovector));
427 if (ret < 0 && ret != PCRE_ERROR_NOMATCH)
428 die("pcre_exec failed with error code %d", ret);
429 if (ret > 0) {
430 ret = 0;
431 match->rm_so = ovector[0];
432 match->rm_eo = ovector[1];
435 return ret;
438 static void free_pcre1_regexp(struct grep_pat *p)
440 pcre_free(p->pcre1_regexp);
441 #ifdef PCRE_CONFIG_JIT
442 if (p->pcre1_jit_on)
443 pcre_free_study(p->pcre1_extra_info);
444 else
445 #endif
446 pcre_free(p->pcre1_extra_info);
447 pcre_free((void *)p->pcre1_tables);
449 #else /* !USE_LIBPCRE1 */
450 static void compile_pcre1_regexp(struct grep_pat *p, const struct grep_opt *opt)
452 die("cannot use Perl-compatible regexes when not compiled with USE_LIBPCRE");
455 static int pcre1match(struct grep_pat *p, const char *line, const char *eol,
456 regmatch_t *match, int eflags)
458 return 1;
461 static void free_pcre1_regexp(struct grep_pat *p)
464 #endif /* !USE_LIBPCRE1 */
466 #ifdef USE_LIBPCRE2
467 static void compile_pcre2_pattern(struct grep_pat *p, const struct grep_opt *opt)
469 int error;
470 PCRE2_UCHAR errbuf[256];
471 PCRE2_SIZE erroffset;
472 int options = PCRE2_MULTILINE;
473 int jitret;
474 int patinforet;
475 size_t jitsizearg;
477 assert(opt->pcre2);
479 p->pcre2_compile_context = NULL;
481 /* pcre2_global_context is initialized in append_grep_pattern */
482 if (opt->ignore_case) {
483 if (!opt->ignore_locale && has_non_ascii(p->pattern)) {
484 if (!pcre2_global_context)
485 BUG("pcre2_global_context uninitialized");
486 p->pcre2_tables = pcre2_maketables(pcre2_global_context);
487 p->pcre2_compile_context = pcre2_compile_context_create(NULL);
488 pcre2_set_character_tables(p->pcre2_compile_context,
489 p->pcre2_tables);
491 options |= PCRE2_CASELESS;
493 if (!opt->ignore_locale && is_utf8_locale() && has_non_ascii(p->pattern) &&
494 !(!opt->ignore_case && (p->fixed || p->is_fixed)))
495 options |= PCRE2_UTF;
497 p->pcre2_pattern = pcre2_compile((PCRE2_SPTR)p->pattern,
498 p->patternlen, options, &error, &erroffset,
499 p->pcre2_compile_context);
501 if (p->pcre2_pattern) {
502 p->pcre2_match_data = pcre2_match_data_create_from_pattern(p->pcre2_pattern, NULL);
503 if (!p->pcre2_match_data)
504 die("Couldn't allocate PCRE2 match data");
505 } else {
506 pcre2_get_error_message(error, errbuf, sizeof(errbuf));
507 compile_regexp_failed(p, (const char *)&errbuf);
510 pcre2_config(PCRE2_CONFIG_JIT, &p->pcre2_jit_on);
511 if (opt->debug)
512 fprintf(stderr, "pcre2_jit_on=%d\n", p->pcre2_jit_on);
513 if (p->pcre2_jit_on) {
514 jitret = pcre2_jit_compile(p->pcre2_pattern, PCRE2_JIT_COMPLETE);
515 if (jitret)
516 die("Couldn't JIT the PCRE2 pattern '%s', got '%d'\n", p->pattern, jitret);
519 * The pcre2_config(PCRE2_CONFIG_JIT, ...) call just
520 * tells us whether the library itself supports JIT,
521 * but to see whether we're going to be actually using
522 * JIT we need to extract PCRE2_INFO_JITSIZE from the
523 * pattern *after* we do pcre2_jit_compile() above.
525 * This is because if the pattern contains the
526 * (*NO_JIT) verb (see pcre2syntax(3))
527 * pcre2_jit_compile() will exit early with 0. If we
528 * then proceed to call pcre2_jit_match() further down
529 * the line instead of pcre2_match() we'll either
530 * segfault (pre PCRE 10.31) or run into a fatal error
531 * (post PCRE2 10.31)
533 patinforet = pcre2_pattern_info(p->pcre2_pattern, PCRE2_INFO_JITSIZE, &jitsizearg);
534 if (patinforet)
535 BUG("pcre2_pattern_info() failed: %d", patinforet);
536 if (jitsizearg == 0) {
537 p->pcre2_jit_on = 0;
538 if (opt->debug)
539 fprintf(stderr, "pcre2_jit_on=%d: (*NO_JIT) in regex\n",
540 p->pcre2_jit_on);
541 return;
546 static int pcre2match(struct grep_pat *p, const char *line, const char *eol,
547 regmatch_t *match, int eflags)
549 int ret, flags = 0;
550 PCRE2_SIZE *ovector;
551 PCRE2_UCHAR errbuf[256];
553 if (eflags & REG_NOTBOL)
554 flags |= PCRE2_NOTBOL;
556 if (p->pcre2_jit_on)
557 ret = pcre2_jit_match(p->pcre2_pattern, (unsigned char *)line,
558 eol - line, 0, flags, p->pcre2_match_data,
559 NULL);
560 else
561 ret = pcre2_match(p->pcre2_pattern, (unsigned char *)line,
562 eol - line, 0, flags, p->pcre2_match_data,
563 NULL);
565 if (ret < 0 && ret != PCRE2_ERROR_NOMATCH) {
566 pcre2_get_error_message(ret, errbuf, sizeof(errbuf));
567 die("%s failed with error code %d: %s",
568 (p->pcre2_jit_on ? "pcre2_jit_match" : "pcre2_match"), ret,
569 errbuf);
571 if (ret > 0) {
572 ovector = pcre2_get_ovector_pointer(p->pcre2_match_data);
573 ret = 0;
574 match->rm_so = (int)ovector[0];
575 match->rm_eo = (int)ovector[1];
578 return ret;
581 static void free_pcre2_pattern(struct grep_pat *p)
583 pcre2_compile_context_free(p->pcre2_compile_context);
584 pcre2_code_free(p->pcre2_pattern);
585 pcre2_match_data_free(p->pcre2_match_data);
586 free((void *)p->pcre2_tables);
588 #else /* !USE_LIBPCRE2 */
589 static void compile_pcre2_pattern(struct grep_pat *p, const struct grep_opt *opt)
592 * Unreachable until USE_LIBPCRE2 becomes synonymous with
593 * USE_LIBPCRE. See the sibling comment in
594 * grep_set_pattern_type_option().
596 die("cannot use Perl-compatible regexes when not compiled with USE_LIBPCRE");
599 static int pcre2match(struct grep_pat *p, const char *line, const char *eol,
600 regmatch_t *match, int eflags)
602 return 1;
605 static void free_pcre2_pattern(struct grep_pat *p)
609 static void compile_fixed_regexp(struct grep_pat *p, struct grep_opt *opt)
611 struct strbuf sb = STRBUF_INIT;
612 int err;
613 int regflags = 0;
615 basic_regex_quote_buf(&sb, p->pattern);
616 if (opt->ignore_case)
617 regflags |= REG_ICASE;
618 err = regcomp(&p->regexp, sb.buf, regflags);
619 if (opt->debug)
620 fprintf(stderr, "fixed %s\n", sb.buf);
621 strbuf_release(&sb);
622 if (err) {
623 char errbuf[1024];
624 regerror(err, &p->regexp, errbuf, sizeof(errbuf));
625 compile_regexp_failed(p, errbuf);
628 #endif /* !USE_LIBPCRE2 */
630 static void compile_regexp(struct grep_pat *p, struct grep_opt *opt)
632 int err;
633 int regflags = REG_NEWLINE;
635 p->word_regexp = opt->word_regexp;
636 p->ignore_case = opt->ignore_case;
637 p->fixed = opt->fixed;
639 if (memchr(p->pattern, 0, p->patternlen) && !opt->pcre2)
640 die(_("given pattern contains NULL byte (via -f <file>). This is only supported with -P under PCRE v2"));
642 p->is_fixed = is_fixed(p->pattern, p->patternlen);
643 #ifdef USE_LIBPCRE2
644 if (!p->fixed && !p->is_fixed) {
645 const char *no_jit = "(*NO_JIT)";
646 const int no_jit_len = strlen(no_jit);
647 if (starts_with(p->pattern, no_jit) &&
648 is_fixed(p->pattern + no_jit_len,
649 p->patternlen - no_jit_len))
650 p->is_fixed = 1;
652 #endif
653 if (p->fixed || p->is_fixed) {
654 #ifdef USE_LIBPCRE2
655 opt->pcre2 = 1;
656 if (p->is_fixed) {
657 compile_pcre2_pattern(p, opt);
658 } else {
660 * E.g. t7811-grep-open.sh relies on the
661 * pattern being restored.
663 char *old_pattern = p->pattern;
664 size_t old_patternlen = p->patternlen;
665 struct strbuf sb = STRBUF_INIT;
668 * There is the PCRE2_LITERAL flag, but it's
669 * only in PCRE v2 10.30 and later. Needing to
670 * ifdef our way around that and dealing with
671 * it + PCRE2_MULTILINE being an error is more
672 * complex than just quoting this ourselves.
674 strbuf_add(&sb, "\\Q", 2);
675 strbuf_add(&sb, p->pattern, p->patternlen);
676 strbuf_add(&sb, "\\E", 2);
678 p->pattern = sb.buf;
679 p->patternlen = sb.len;
680 compile_pcre2_pattern(p, opt);
681 p->pattern = old_pattern;
682 p->patternlen = old_patternlen;
683 strbuf_release(&sb);
685 #else /* !USE_LIBPCRE2 */
686 compile_fixed_regexp(p, opt);
687 #endif /* !USE_LIBPCRE2 */
688 return;
691 if (opt->pcre2) {
692 compile_pcre2_pattern(p, opt);
693 return;
696 if (opt->pcre1) {
697 compile_pcre1_regexp(p, opt);
698 return;
701 if (p->ignore_case)
702 regflags |= REG_ICASE;
703 if (opt->extended_regexp_option)
704 regflags |= REG_EXTENDED;
705 err = regcomp(&p->regexp, p->pattern, regflags);
706 if (err) {
707 char errbuf[1024];
708 regerror(err, &p->regexp, errbuf, 1024);
709 compile_regexp_failed(p, errbuf);
713 static struct grep_expr *compile_pattern_or(struct grep_pat **);
714 static struct grep_expr *compile_pattern_atom(struct grep_pat **list)
716 struct grep_pat *p;
717 struct grep_expr *x;
719 p = *list;
720 if (!p)
721 return NULL;
722 switch (p->token) {
723 case GREP_PATTERN: /* atom */
724 case GREP_PATTERN_HEAD:
725 case GREP_PATTERN_BODY:
726 x = xcalloc(1, sizeof (struct grep_expr));
727 x->node = GREP_NODE_ATOM;
728 x->u.atom = p;
729 *list = p->next;
730 return x;
731 case GREP_OPEN_PAREN:
732 *list = p->next;
733 x = compile_pattern_or(list);
734 if (!*list || (*list)->token != GREP_CLOSE_PAREN)
735 die("unmatched parenthesis");
736 *list = (*list)->next;
737 return x;
738 default:
739 return NULL;
743 static struct grep_expr *compile_pattern_not(struct grep_pat **list)
745 struct grep_pat *p;
746 struct grep_expr *x;
748 p = *list;
749 if (!p)
750 return NULL;
751 switch (p->token) {
752 case GREP_NOT:
753 if (!p->next)
754 die("--not not followed by pattern expression");
755 *list = p->next;
756 x = xcalloc(1, sizeof (struct grep_expr));
757 x->node = GREP_NODE_NOT;
758 x->u.unary = compile_pattern_not(list);
759 if (!x->u.unary)
760 die("--not followed by non pattern expression");
761 return x;
762 default:
763 return compile_pattern_atom(list);
767 static struct grep_expr *compile_pattern_and(struct grep_pat **list)
769 struct grep_pat *p;
770 struct grep_expr *x, *y, *z;
772 x = compile_pattern_not(list);
773 p = *list;
774 if (p && p->token == GREP_AND) {
775 if (!p->next)
776 die("--and not followed by pattern expression");
777 *list = p->next;
778 y = compile_pattern_and(list);
779 if (!y)
780 die("--and not followed by pattern expression");
781 z = xcalloc(1, sizeof (struct grep_expr));
782 z->node = GREP_NODE_AND;
783 z->u.binary.left = x;
784 z->u.binary.right = y;
785 return z;
787 return x;
790 static struct grep_expr *compile_pattern_or(struct grep_pat **list)
792 struct grep_pat *p;
793 struct grep_expr *x, *y, *z;
795 x = compile_pattern_and(list);
796 p = *list;
797 if (x && p && p->token != GREP_CLOSE_PAREN) {
798 y = compile_pattern_or(list);
799 if (!y)
800 die("not a pattern expression %s", p->pattern);
801 z = xcalloc(1, sizeof (struct grep_expr));
802 z->node = GREP_NODE_OR;
803 z->u.binary.left = x;
804 z->u.binary.right = y;
805 return z;
807 return x;
810 static struct grep_expr *compile_pattern_expr(struct grep_pat **list)
812 return compile_pattern_or(list);
815 static void indent(int in)
817 while (in-- > 0)
818 fputc(' ', stderr);
821 static void dump_grep_pat(struct grep_pat *p)
823 switch (p->token) {
824 case GREP_AND: fprintf(stderr, "*and*"); break;
825 case GREP_OPEN_PAREN: fprintf(stderr, "*(*"); break;
826 case GREP_CLOSE_PAREN: fprintf(stderr, "*)*"); break;
827 case GREP_NOT: fprintf(stderr, "*not*"); break;
828 case GREP_OR: fprintf(stderr, "*or*"); break;
830 case GREP_PATTERN: fprintf(stderr, "pattern"); break;
831 case GREP_PATTERN_HEAD: fprintf(stderr, "pattern_head"); break;
832 case GREP_PATTERN_BODY: fprintf(stderr, "pattern_body"); break;
835 switch (p->token) {
836 default: break;
837 case GREP_PATTERN_HEAD:
838 fprintf(stderr, "<head %d>", p->field); break;
839 case GREP_PATTERN_BODY:
840 fprintf(stderr, "<body>"); break;
842 switch (p->token) {
843 default: break;
844 case GREP_PATTERN_HEAD:
845 case GREP_PATTERN_BODY:
846 case GREP_PATTERN:
847 fprintf(stderr, "%.*s", (int)p->patternlen, p->pattern);
848 break;
850 fputc('\n', stderr);
853 static void dump_grep_expression_1(struct grep_expr *x, int in)
855 indent(in);
856 switch (x->node) {
857 case GREP_NODE_TRUE:
858 fprintf(stderr, "true\n");
859 break;
860 case GREP_NODE_ATOM:
861 dump_grep_pat(x->u.atom);
862 break;
863 case GREP_NODE_NOT:
864 fprintf(stderr, "(not\n");
865 dump_grep_expression_1(x->u.unary, in+1);
866 indent(in);
867 fprintf(stderr, ")\n");
868 break;
869 case GREP_NODE_AND:
870 fprintf(stderr, "(and\n");
871 dump_grep_expression_1(x->u.binary.left, in+1);
872 dump_grep_expression_1(x->u.binary.right, in+1);
873 indent(in);
874 fprintf(stderr, ")\n");
875 break;
876 case GREP_NODE_OR:
877 fprintf(stderr, "(or\n");
878 dump_grep_expression_1(x->u.binary.left, in+1);
879 dump_grep_expression_1(x->u.binary.right, in+1);
880 indent(in);
881 fprintf(stderr, ")\n");
882 break;
886 static void dump_grep_expression(struct grep_opt *opt)
888 struct grep_expr *x = opt->pattern_expression;
890 if (opt->all_match)
891 fprintf(stderr, "[all-match]\n");
892 dump_grep_expression_1(x, 0);
893 fflush(NULL);
896 static struct grep_expr *grep_true_expr(void)
898 struct grep_expr *z = xcalloc(1, sizeof(*z));
899 z->node = GREP_NODE_TRUE;
900 return z;
903 static struct grep_expr *grep_or_expr(struct grep_expr *left, struct grep_expr *right)
905 struct grep_expr *z = xcalloc(1, sizeof(*z));
906 z->node = GREP_NODE_OR;
907 z->u.binary.left = left;
908 z->u.binary.right = right;
909 return z;
912 static struct grep_expr *prep_header_patterns(struct grep_opt *opt)
914 struct grep_pat *p;
915 struct grep_expr *header_expr;
916 struct grep_expr *(header_group[GREP_HEADER_FIELD_MAX]);
917 enum grep_header_field fld;
919 if (!opt->header_list)
920 return NULL;
922 for (p = opt->header_list; p; p = p->next) {
923 if (p->token != GREP_PATTERN_HEAD)
924 BUG("a non-header pattern in grep header list.");
925 if (p->field < GREP_HEADER_FIELD_MIN ||
926 GREP_HEADER_FIELD_MAX <= p->field)
927 BUG("unknown header field %d", p->field);
928 compile_regexp(p, opt);
931 for (fld = 0; fld < GREP_HEADER_FIELD_MAX; fld++)
932 header_group[fld] = NULL;
934 for (p = opt->header_list; p; p = p->next) {
935 struct grep_expr *h;
936 struct grep_pat *pp = p;
938 h = compile_pattern_atom(&pp);
939 if (!h || pp != p->next)
940 BUG("malformed header expr");
941 if (!header_group[p->field]) {
942 header_group[p->field] = h;
943 continue;
945 header_group[p->field] = grep_or_expr(h, header_group[p->field]);
948 header_expr = NULL;
950 for (fld = 0; fld < GREP_HEADER_FIELD_MAX; fld++) {
951 if (!header_group[fld])
952 continue;
953 if (!header_expr)
954 header_expr = grep_true_expr();
955 header_expr = grep_or_expr(header_group[fld], header_expr);
957 return header_expr;
960 static struct grep_expr *grep_splice_or(struct grep_expr *x, struct grep_expr *y)
962 struct grep_expr *z = x;
964 while (x) {
965 assert(x->node == GREP_NODE_OR);
966 if (x->u.binary.right &&
967 x->u.binary.right->node == GREP_NODE_TRUE) {
968 x->u.binary.right = y;
969 break;
971 x = x->u.binary.right;
973 return z;
976 static void compile_grep_patterns_real(struct grep_opt *opt)
978 struct grep_pat *p;
979 struct grep_expr *header_expr = prep_header_patterns(opt);
981 for (p = opt->pattern_list; p; p = p->next) {
982 switch (p->token) {
983 case GREP_PATTERN: /* atom */
984 case GREP_PATTERN_HEAD:
985 case GREP_PATTERN_BODY:
986 compile_regexp(p, opt);
987 break;
988 default:
989 opt->extended = 1;
990 break;
994 if (opt->all_match || header_expr)
995 opt->extended = 1;
996 else if (!opt->extended && !opt->debug)
997 return;
999 p = opt->pattern_list;
1000 if (p)
1001 opt->pattern_expression = compile_pattern_expr(&p);
1002 if (p)
1003 die("incomplete pattern expression: %s", p->pattern);
1005 if (!header_expr)
1006 return;
1008 if (!opt->pattern_expression)
1009 opt->pattern_expression = header_expr;
1010 else if (opt->all_match)
1011 opt->pattern_expression = grep_splice_or(header_expr,
1012 opt->pattern_expression);
1013 else
1014 opt->pattern_expression = grep_or_expr(opt->pattern_expression,
1015 header_expr);
1016 opt->all_match = 1;
1019 void compile_grep_patterns(struct grep_opt *opt)
1021 compile_grep_patterns_real(opt);
1022 if (opt->debug)
1023 dump_grep_expression(opt);
1026 static void free_pattern_expr(struct grep_expr *x)
1028 switch (x->node) {
1029 case GREP_NODE_TRUE:
1030 case GREP_NODE_ATOM:
1031 break;
1032 case GREP_NODE_NOT:
1033 free_pattern_expr(x->u.unary);
1034 break;
1035 case GREP_NODE_AND:
1036 case GREP_NODE_OR:
1037 free_pattern_expr(x->u.binary.left);
1038 free_pattern_expr(x->u.binary.right);
1039 break;
1041 free(x);
1044 void free_grep_patterns(struct grep_opt *opt)
1046 struct grep_pat *p, *n;
1048 for (p = opt->pattern_list; p; p = n) {
1049 n = p->next;
1050 switch (p->token) {
1051 case GREP_PATTERN: /* atom */
1052 case GREP_PATTERN_HEAD:
1053 case GREP_PATTERN_BODY:
1054 if (p->pcre1_regexp)
1055 free_pcre1_regexp(p);
1056 else if (p->pcre2_pattern)
1057 free_pcre2_pattern(p);
1058 else
1059 regfree(&p->regexp);
1060 free(p->pattern);
1061 break;
1062 default:
1063 break;
1065 free(p);
1068 if (!opt->extended)
1069 return;
1070 free_pattern_expr(opt->pattern_expression);
1073 static char *end_of_line(char *cp, unsigned long *left)
1075 unsigned long l = *left;
1076 while (l && *cp != '\n') {
1077 l--;
1078 cp++;
1080 *left = l;
1081 return cp;
1084 static int word_char(char ch)
1086 return isalnum(ch) || ch == '_';
1089 static void output_color(struct grep_opt *opt, const void *data, size_t size,
1090 const char *color)
1092 if (want_color(opt->color) && color && color[0]) {
1093 opt->output(opt, color, strlen(color));
1094 opt->output(opt, data, size);
1095 opt->output(opt, GIT_COLOR_RESET, strlen(GIT_COLOR_RESET));
1096 } else
1097 opt->output(opt, data, size);
1100 static void output_sep(struct grep_opt *opt, char sign)
1102 if (opt->null_following_name)
1103 opt->output(opt, "\0", 1);
1104 else
1105 output_color(opt, &sign, 1, opt->colors[GREP_COLOR_SEP]);
1108 static void show_name(struct grep_opt *opt, const char *name)
1110 output_color(opt, name, strlen(name), opt->colors[GREP_COLOR_FILENAME]);
1111 opt->output(opt, opt->null_following_name ? "\0" : "\n", 1);
1114 static int patmatch(struct grep_pat *p, char *line, char *eol,
1115 regmatch_t *match, int eflags)
1117 int hit;
1119 if (p->pcre1_regexp)
1120 hit = !pcre1match(p, line, eol, match, eflags);
1121 else if (p->pcre2_pattern)
1122 hit = !pcre2match(p, line, eol, match, eflags);
1123 else
1124 hit = !regexec_buf(&p->regexp, line, eol - line, 1, match,
1125 eflags);
1127 return hit;
1130 static int strip_timestamp(char *bol, char **eol_p)
1132 char *eol = *eol_p;
1133 int ch;
1135 while (bol < --eol) {
1136 if (*eol != '>')
1137 continue;
1138 *eol_p = ++eol;
1139 ch = *eol;
1140 *eol = '\0';
1141 return ch;
1143 return 0;
1146 static struct {
1147 const char *field;
1148 size_t len;
1149 } header_field[] = {
1150 { "author ", 7 },
1151 { "committer ", 10 },
1152 { "reflog ", 7 },
1155 static int match_one_pattern(struct grep_pat *p, char *bol, char *eol,
1156 enum grep_context ctx,
1157 regmatch_t *pmatch, int eflags)
1159 int hit = 0;
1160 int saved_ch = 0;
1161 const char *start = bol;
1163 if ((p->token != GREP_PATTERN) &&
1164 ((p->token == GREP_PATTERN_HEAD) != (ctx == GREP_CONTEXT_HEAD)))
1165 return 0;
1167 if (p->token == GREP_PATTERN_HEAD) {
1168 const char *field;
1169 size_t len;
1170 assert(p->field < ARRAY_SIZE(header_field));
1171 field = header_field[p->field].field;
1172 len = header_field[p->field].len;
1173 if (strncmp(bol, field, len))
1174 return 0;
1175 bol += len;
1176 switch (p->field) {
1177 case GREP_HEADER_AUTHOR:
1178 case GREP_HEADER_COMMITTER:
1179 saved_ch = strip_timestamp(bol, &eol);
1180 break;
1181 default:
1182 break;
1186 again:
1187 hit = patmatch(p, bol, eol, pmatch, eflags);
1189 if (hit && p->word_regexp) {
1190 if ((pmatch[0].rm_so < 0) ||
1191 (eol - bol) < pmatch[0].rm_so ||
1192 (pmatch[0].rm_eo < 0) ||
1193 (eol - bol) < pmatch[0].rm_eo)
1194 die("regexp returned nonsense");
1196 /* Match beginning must be either beginning of the
1197 * line, or at word boundary (i.e. the last char must
1198 * not be a word char). Similarly, match end must be
1199 * either end of the line, or at word boundary
1200 * (i.e. the next char must not be a word char).
1202 if ( ((pmatch[0].rm_so == 0) ||
1203 !word_char(bol[pmatch[0].rm_so-1])) &&
1204 ((pmatch[0].rm_eo == (eol-bol)) ||
1205 !word_char(bol[pmatch[0].rm_eo])) )
1207 else
1208 hit = 0;
1210 /* Words consist of at least one character. */
1211 if (pmatch->rm_so == pmatch->rm_eo)
1212 hit = 0;
1214 if (!hit && pmatch[0].rm_so + bol + 1 < eol) {
1215 /* There could be more than one match on the
1216 * line, and the first match might not be
1217 * strict word match. But later ones could be!
1218 * Forward to the next possible start, i.e. the
1219 * next position following a non-word char.
1221 bol = pmatch[0].rm_so + bol + 1;
1222 while (word_char(bol[-1]) && bol < eol)
1223 bol++;
1224 eflags |= REG_NOTBOL;
1225 if (bol < eol)
1226 goto again;
1229 if (p->token == GREP_PATTERN_HEAD && saved_ch)
1230 *eol = saved_ch;
1231 if (hit) {
1232 pmatch[0].rm_so += bol - start;
1233 pmatch[0].rm_eo += bol - start;
1235 return hit;
1238 static int match_expr_eval(struct grep_opt *opt, struct grep_expr *x, char *bol,
1239 char *eol, enum grep_context ctx, ssize_t *col,
1240 ssize_t *icol, int collect_hits)
1242 int h = 0;
1244 if (!x)
1245 die("Not a valid grep expression");
1246 switch (x->node) {
1247 case GREP_NODE_TRUE:
1248 h = 1;
1249 break;
1250 case GREP_NODE_ATOM:
1252 regmatch_t tmp;
1253 h = match_one_pattern(x->u.atom, bol, eol, ctx,
1254 &tmp, 0);
1255 if (h && (*col < 0 || tmp.rm_so < *col))
1256 *col = tmp.rm_so;
1258 break;
1259 case GREP_NODE_NOT:
1261 * Upon visiting a GREP_NODE_NOT, col and icol become swapped.
1263 h = !match_expr_eval(opt, x->u.unary, bol, eol, ctx, icol, col,
1265 break;
1266 case GREP_NODE_AND:
1267 h = match_expr_eval(opt, x->u.binary.left, bol, eol, ctx, col,
1268 icol, 0);
1269 if (h || opt->columnnum) {
1271 * Don't short-circuit AND when given --column, since a
1272 * NOT earlier in the tree may turn this into an OR. In
1273 * this case, see the below comment.
1275 h &= match_expr_eval(opt, x->u.binary.right, bol, eol,
1276 ctx, col, icol, 0);
1278 break;
1279 case GREP_NODE_OR:
1280 if (!(collect_hits || opt->columnnum)) {
1282 * Don't short-circuit OR when given --column (or
1283 * collecting hits) to ensure we don't skip a later
1284 * child that would produce an earlier match.
1286 return (match_expr_eval(opt, x->u.binary.left, bol, eol,
1287 ctx, col, icol, 0) ||
1288 match_expr_eval(opt, x->u.binary.right, bol,
1289 eol, ctx, col, icol, 0));
1291 h = match_expr_eval(opt, x->u.binary.left, bol, eol, ctx, col,
1292 icol, 0);
1293 if (collect_hits)
1294 x->u.binary.left->hit |= h;
1295 h |= match_expr_eval(opt, x->u.binary.right, bol, eol, ctx, col,
1296 icol, collect_hits);
1297 break;
1298 default:
1299 die("Unexpected node type (internal error) %d", x->node);
1301 if (collect_hits)
1302 x->hit |= h;
1303 return h;
1306 static int match_expr(struct grep_opt *opt, char *bol, char *eol,
1307 enum grep_context ctx, ssize_t *col,
1308 ssize_t *icol, int collect_hits)
1310 struct grep_expr *x = opt->pattern_expression;
1311 return match_expr_eval(opt, x, bol, eol, ctx, col, icol, collect_hits);
1314 static int match_line(struct grep_opt *opt, char *bol, char *eol,
1315 ssize_t *col, ssize_t *icol,
1316 enum grep_context ctx, int collect_hits)
1318 struct grep_pat *p;
1319 int hit = 0;
1321 if (opt->extended)
1322 return match_expr(opt, bol, eol, ctx, col, icol,
1323 collect_hits);
1325 /* we do not call with collect_hits without being extended */
1326 for (p = opt->pattern_list; p; p = p->next) {
1327 regmatch_t tmp;
1328 if (match_one_pattern(p, bol, eol, ctx, &tmp, 0)) {
1329 hit |= 1;
1330 if (!opt->columnnum) {
1332 * Without --column, any single match on a line
1333 * is enough to know that it needs to be
1334 * printed. With --column, scan _all_ patterns
1335 * to find the earliest.
1337 break;
1339 if (*col < 0 || tmp.rm_so < *col)
1340 *col = tmp.rm_so;
1343 return hit;
1346 static int match_next_pattern(struct grep_pat *p, char *bol, char *eol,
1347 enum grep_context ctx,
1348 regmatch_t *pmatch, int eflags)
1350 regmatch_t match;
1352 if (!match_one_pattern(p, bol, eol, ctx, &match, eflags))
1353 return 0;
1354 if (match.rm_so < 0 || match.rm_eo < 0)
1355 return 0;
1356 if (pmatch->rm_so >= 0 && pmatch->rm_eo >= 0) {
1357 if (match.rm_so > pmatch->rm_so)
1358 return 1;
1359 if (match.rm_so == pmatch->rm_so && match.rm_eo < pmatch->rm_eo)
1360 return 1;
1362 pmatch->rm_so = match.rm_so;
1363 pmatch->rm_eo = match.rm_eo;
1364 return 1;
1367 static int next_match(struct grep_opt *opt, char *bol, char *eol,
1368 enum grep_context ctx, regmatch_t *pmatch, int eflags)
1370 struct grep_pat *p;
1371 int hit = 0;
1373 pmatch->rm_so = pmatch->rm_eo = -1;
1374 if (bol < eol) {
1375 for (p = opt->pattern_list; p; p = p->next) {
1376 switch (p->token) {
1377 case GREP_PATTERN: /* atom */
1378 case GREP_PATTERN_HEAD:
1379 case GREP_PATTERN_BODY:
1380 hit |= match_next_pattern(p, bol, eol, ctx,
1381 pmatch, eflags);
1382 break;
1383 default:
1384 break;
1388 return hit;
1391 static void show_line_header(struct grep_opt *opt, const char *name,
1392 unsigned lno, ssize_t cno, char sign)
1394 if (opt->heading && opt->last_shown == 0) {
1395 output_color(opt, name, strlen(name), opt->colors[GREP_COLOR_FILENAME]);
1396 opt->output(opt, "\n", 1);
1398 opt->last_shown = lno;
1400 if (!opt->heading && opt->pathname) {
1401 output_color(opt, name, strlen(name), opt->colors[GREP_COLOR_FILENAME]);
1402 output_sep(opt, sign);
1404 if (opt->linenum) {
1405 char buf[32];
1406 xsnprintf(buf, sizeof(buf), "%d", lno);
1407 output_color(opt, buf, strlen(buf), opt->colors[GREP_COLOR_LINENO]);
1408 output_sep(opt, sign);
1411 * Treat 'cno' as the 1-indexed offset from the start of a non-context
1412 * line to its first match. Otherwise, 'cno' is 0 indicating that we are
1413 * being called with a context line.
1415 if (opt->columnnum && cno) {
1416 char buf[32];
1417 xsnprintf(buf, sizeof(buf), "%"PRIuMAX, (uintmax_t)cno);
1418 output_color(opt, buf, strlen(buf), opt->colors[GREP_COLOR_COLUMNNO]);
1419 output_sep(opt, sign);
1423 static void show_line(struct grep_opt *opt, char *bol, char *eol,
1424 const char *name, unsigned lno, ssize_t cno, char sign)
1426 int rest = eol - bol;
1427 const char *match_color = NULL;
1428 const char *line_color = NULL;
1430 if (opt->file_break && opt->last_shown == 0) {
1431 if (opt->show_hunk_mark)
1432 opt->output(opt, "\n", 1);
1433 } else if (opt->pre_context || opt->post_context || opt->funcbody) {
1434 if (opt->last_shown == 0) {
1435 if (opt->show_hunk_mark) {
1436 output_color(opt, "--", 2, opt->colors[GREP_COLOR_SEP]);
1437 opt->output(opt, "\n", 1);
1439 } else if (lno > opt->last_shown + 1) {
1440 output_color(opt, "--", 2, opt->colors[GREP_COLOR_SEP]);
1441 opt->output(opt, "\n", 1);
1444 if (!opt->only_matching) {
1446 * In case the line we're being called with contains more than
1447 * one match, leave printing each header to the loop below.
1449 show_line_header(opt, name, lno, cno, sign);
1451 if (opt->color || opt->only_matching) {
1452 regmatch_t match;
1453 enum grep_context ctx = GREP_CONTEXT_BODY;
1454 int ch = *eol;
1455 int eflags = 0;
1457 if (opt->color) {
1458 if (sign == ':')
1459 match_color = opt->colors[GREP_COLOR_MATCH_SELECTED];
1460 else
1461 match_color = opt->colors[GREP_COLOR_MATCH_CONTEXT];
1462 if (sign == ':')
1463 line_color = opt->colors[GREP_COLOR_SELECTED];
1464 else if (sign == '-')
1465 line_color = opt->colors[GREP_COLOR_CONTEXT];
1466 else if (sign == '=')
1467 line_color = opt->colors[GREP_COLOR_FUNCTION];
1469 *eol = '\0';
1470 while (next_match(opt, bol, eol, ctx, &match, eflags)) {
1471 if (match.rm_so == match.rm_eo)
1472 break;
1474 if (opt->only_matching)
1475 show_line_header(opt, name, lno, cno, sign);
1476 else
1477 output_color(opt, bol, match.rm_so, line_color);
1478 output_color(opt, bol + match.rm_so,
1479 match.rm_eo - match.rm_so, match_color);
1480 if (opt->only_matching)
1481 opt->output(opt, "\n", 1);
1482 bol += match.rm_eo;
1483 cno += match.rm_eo;
1484 rest -= match.rm_eo;
1485 eflags = REG_NOTBOL;
1487 *eol = ch;
1489 if (!opt->only_matching) {
1490 output_color(opt, bol, rest, line_color);
1491 opt->output(opt, "\n", 1);
1495 int grep_use_locks;
1498 * This lock protects access to the gitattributes machinery, which is
1499 * not thread-safe.
1501 pthread_mutex_t grep_attr_mutex;
1503 static inline void grep_attr_lock(void)
1505 if (grep_use_locks)
1506 pthread_mutex_lock(&grep_attr_mutex);
1509 static inline void grep_attr_unlock(void)
1511 if (grep_use_locks)
1512 pthread_mutex_unlock(&grep_attr_mutex);
1515 static int match_funcname(struct grep_opt *opt, struct grep_source *gs, char *bol, char *eol)
1517 xdemitconf_t *xecfg = opt->priv;
1518 if (xecfg && !xecfg->find_func) {
1519 grep_source_load_driver(gs, opt->repo->index);
1520 if (gs->driver->funcname.pattern) {
1521 const struct userdiff_funcname *pe = &gs->driver->funcname;
1522 xdiff_set_find_func(xecfg, pe->pattern, pe->cflags);
1523 } else {
1524 xecfg = opt->priv = NULL;
1528 if (xecfg) {
1529 char buf[1];
1530 return xecfg->find_func(bol, eol - bol, buf, 1,
1531 xecfg->find_func_priv) >= 0;
1534 if (bol == eol)
1535 return 0;
1536 if (isalpha(*bol) || *bol == '_' || *bol == '$')
1537 return 1;
1538 return 0;
1541 static void show_funcname_line(struct grep_opt *opt, struct grep_source *gs,
1542 char *bol, unsigned lno)
1544 while (bol > gs->buf) {
1545 char *eol = --bol;
1547 while (bol > gs->buf && bol[-1] != '\n')
1548 bol--;
1549 lno--;
1551 if (lno <= opt->last_shown)
1552 break;
1554 if (match_funcname(opt, gs, bol, eol)) {
1555 show_line(opt, bol, eol, gs->name, lno, 0, '=');
1556 break;
1561 static int is_empty_line(const char *bol, const char *eol);
1563 static void show_pre_context(struct grep_opt *opt, struct grep_source *gs,
1564 char *bol, char *end, unsigned lno)
1566 unsigned cur = lno, from = 1, funcname_lno = 0, orig_from;
1567 int funcname_needed = !!opt->funcname, comment_needed = 0;
1569 if (opt->pre_context < lno)
1570 from = lno - opt->pre_context;
1571 if (from <= opt->last_shown)
1572 from = opt->last_shown + 1;
1573 orig_from = from;
1574 if (opt->funcbody) {
1575 if (match_funcname(opt, gs, bol, end))
1576 comment_needed = 1;
1577 else
1578 funcname_needed = 1;
1579 from = opt->last_shown + 1;
1582 /* Rewind. */
1583 while (bol > gs->buf && cur > from) {
1584 char *next_bol = bol;
1585 char *eol = --bol;
1587 while (bol > gs->buf && bol[-1] != '\n')
1588 bol--;
1589 cur--;
1590 if (comment_needed && (is_empty_line(bol, eol) ||
1591 match_funcname(opt, gs, bol, eol))) {
1592 comment_needed = 0;
1593 from = orig_from;
1594 if (cur < from) {
1595 cur++;
1596 bol = next_bol;
1597 break;
1600 if (funcname_needed && match_funcname(opt, gs, bol, eol)) {
1601 funcname_lno = cur;
1602 funcname_needed = 0;
1603 if (opt->funcbody)
1604 comment_needed = 1;
1605 else
1606 from = orig_from;
1610 /* We need to look even further back to find a function signature. */
1611 if (opt->funcname && funcname_needed)
1612 show_funcname_line(opt, gs, bol, cur);
1614 /* Back forward. */
1615 while (cur < lno) {
1616 char *eol = bol, sign = (cur == funcname_lno) ? '=' : '-';
1618 while (*eol != '\n')
1619 eol++;
1620 show_line(opt, bol, eol, gs->name, cur, 0, sign);
1621 bol = eol + 1;
1622 cur++;
1626 static int should_lookahead(struct grep_opt *opt)
1628 struct grep_pat *p;
1630 if (opt->extended)
1631 return 0; /* punt for too complex stuff */
1632 if (opt->invert)
1633 return 0;
1634 for (p = opt->pattern_list; p; p = p->next) {
1635 if (p->token != GREP_PATTERN)
1636 return 0; /* punt for "header only" and stuff */
1638 return 1;
1641 static int look_ahead(struct grep_opt *opt,
1642 unsigned long *left_p,
1643 unsigned *lno_p,
1644 char **bol_p)
1646 unsigned lno = *lno_p;
1647 char *bol = *bol_p;
1648 struct grep_pat *p;
1649 char *sp, *last_bol;
1650 regoff_t earliest = -1;
1652 for (p = opt->pattern_list; p; p = p->next) {
1653 int hit;
1654 regmatch_t m;
1656 hit = patmatch(p, bol, bol + *left_p, &m, 0);
1657 if (!hit || m.rm_so < 0 || m.rm_eo < 0)
1658 continue;
1659 if (earliest < 0 || m.rm_so < earliest)
1660 earliest = m.rm_so;
1663 if (earliest < 0) {
1664 *bol_p = bol + *left_p;
1665 *left_p = 0;
1666 return 1;
1668 for (sp = bol + earliest; bol < sp && sp[-1] != '\n'; sp--)
1669 ; /* find the beginning of the line */
1670 last_bol = sp;
1672 for (sp = bol; sp < last_bol; sp++) {
1673 if (*sp == '\n')
1674 lno++;
1676 *left_p -= last_bol - bol;
1677 *bol_p = last_bol;
1678 *lno_p = lno;
1679 return 0;
1682 static int fill_textconv_grep(struct repository *r,
1683 struct userdiff_driver *driver,
1684 struct grep_source *gs)
1686 struct diff_filespec *df;
1687 char *buf;
1688 size_t size;
1690 if (!driver || !driver->textconv)
1691 return grep_source_load(gs);
1694 * The textconv interface is intimately tied to diff_filespecs, so we
1695 * have to pretend to be one. If we could unify the grep_source
1696 * and diff_filespec structs, this mess could just go away.
1698 df = alloc_filespec(gs->path);
1699 switch (gs->type) {
1700 case GREP_SOURCE_OID:
1701 fill_filespec(df, gs->identifier, 1, 0100644);
1702 break;
1703 case GREP_SOURCE_FILE:
1704 fill_filespec(df, &null_oid, 0, 0100644);
1705 break;
1706 default:
1707 BUG("attempt to textconv something without a path?");
1711 * fill_textconv is not remotely thread-safe; it modifies the global
1712 * diff tempfile structure, writes to the_repo's odb and might
1713 * internally call thread-unsafe functions such as the
1714 * prepare_packed_git() lazy-initializator. Because of the last two, we
1715 * must ensure mutual exclusion between this call and the object reading
1716 * API, thus we use obj_read_lock() here.
1718 * TODO: allowing text conversion to run in parallel with object
1719 * reading operations might increase performance in the multithreaded
1720 * non-worktreee git-grep with --textconv.
1722 obj_read_lock();
1723 size = fill_textconv(r, driver, df, &buf);
1724 obj_read_unlock();
1725 free_filespec(df);
1728 * The normal fill_textconv usage by the diff machinery would just keep
1729 * the textconv'd buf separate from the diff_filespec. But much of the
1730 * grep code passes around a grep_source and assumes that its "buf"
1731 * pointer is the beginning of the thing we are searching. So let's
1732 * install our textconv'd version into the grep_source, taking care not
1733 * to leak any existing buffer.
1735 grep_source_clear_data(gs);
1736 gs->buf = buf;
1737 gs->size = size;
1739 return 0;
1742 static int is_empty_line(const char *bol, const char *eol)
1744 while (bol < eol && isspace(*bol))
1745 bol++;
1746 return bol == eol;
1749 static int grep_source_1(struct grep_opt *opt, struct grep_source *gs, int collect_hits)
1751 char *bol;
1752 char *peek_bol = NULL;
1753 unsigned long left;
1754 unsigned lno = 1;
1755 unsigned last_hit = 0;
1756 int binary_match_only = 0;
1757 unsigned count = 0;
1758 int try_lookahead = 0;
1759 int show_function = 0;
1760 struct userdiff_driver *textconv = NULL;
1761 enum grep_context ctx = GREP_CONTEXT_HEAD;
1762 xdemitconf_t xecfg;
1764 if (!opt->status_only && gs->name == NULL)
1765 BUG("grep call which could print a name requires "
1766 "grep_source.name be non-NULL");
1768 if (!opt->output)
1769 opt->output = std_output;
1771 if (opt->pre_context || opt->post_context || opt->file_break ||
1772 opt->funcbody) {
1773 /* Show hunk marks, except for the first file. */
1774 if (opt->last_shown)
1775 opt->show_hunk_mark = 1;
1777 * If we're using threads then we can't easily identify
1778 * the first file. Always put hunk marks in that case
1779 * and skip the very first one later in work_done().
1781 if (opt->output != std_output)
1782 opt->show_hunk_mark = 1;
1784 opt->last_shown = 0;
1786 if (opt->allow_textconv) {
1787 grep_source_load_driver(gs, opt->repo->index);
1789 * We might set up the shared textconv cache data here, which
1790 * is not thread-safe. Also, get_oid_with_context() and
1791 * parse_object() might be internally called. As they are not
1792 * currently thread-safe and might be racy with object reading,
1793 * obj_read_lock() must be called.
1795 grep_attr_lock();
1796 obj_read_lock();
1797 textconv = userdiff_get_textconv(opt->repo, gs->driver);
1798 obj_read_unlock();
1799 grep_attr_unlock();
1803 * We know the result of a textconv is text, so we only have to care
1804 * about binary handling if we are not using it.
1806 if (!textconv) {
1807 switch (opt->binary) {
1808 case GREP_BINARY_DEFAULT:
1809 if (grep_source_is_binary(gs, opt->repo->index))
1810 binary_match_only = 1;
1811 break;
1812 case GREP_BINARY_NOMATCH:
1813 if (grep_source_is_binary(gs, opt->repo->index))
1814 return 0; /* Assume unmatch */
1815 break;
1816 case GREP_BINARY_TEXT:
1817 break;
1818 default:
1819 BUG("unknown binary handling mode");
1823 memset(&xecfg, 0, sizeof(xecfg));
1824 opt->priv = &xecfg;
1826 try_lookahead = should_lookahead(opt);
1828 if (fill_textconv_grep(opt->repo, textconv, gs) < 0)
1829 return 0;
1831 bol = gs->buf;
1832 left = gs->size;
1833 while (left) {
1834 char *eol, ch;
1835 int hit;
1836 ssize_t cno;
1837 ssize_t col = -1, icol = -1;
1840 * look_ahead() skips quickly to the line that possibly
1841 * has the next hit; don't call it if we need to do
1842 * something more than just skipping the current line
1843 * in response to an unmatch for the current line. E.g.
1844 * inside a post-context window, we will show the current
1845 * line as a context around the previous hit when it
1846 * doesn't hit.
1848 if (try_lookahead
1849 && !(last_hit
1850 && (show_function ||
1851 lno <= last_hit + opt->post_context))
1852 && look_ahead(opt, &left, &lno, &bol))
1853 break;
1854 eol = end_of_line(bol, &left);
1855 ch = *eol;
1856 *eol = 0;
1858 if ((ctx == GREP_CONTEXT_HEAD) && (eol == bol))
1859 ctx = GREP_CONTEXT_BODY;
1861 hit = match_line(opt, bol, eol, &col, &icol, ctx, collect_hits);
1862 *eol = ch;
1864 if (collect_hits)
1865 goto next_line;
1867 /* "grep -v -e foo -e bla" should list lines
1868 * that do not have either, so inversion should
1869 * be done outside.
1871 if (opt->invert)
1872 hit = !hit;
1873 if (opt->unmatch_name_only) {
1874 if (hit)
1875 return 0;
1876 goto next_line;
1878 if (hit) {
1879 count++;
1880 if (opt->status_only)
1881 return 1;
1882 if (opt->name_only) {
1883 show_name(opt, gs->name);
1884 return 1;
1886 if (opt->count)
1887 goto next_line;
1888 if (binary_match_only) {
1889 opt->output(opt, "Binary file ", 12);
1890 output_color(opt, gs->name, strlen(gs->name),
1891 opt->colors[GREP_COLOR_FILENAME]);
1892 opt->output(opt, " matches\n", 9);
1893 return 1;
1895 /* Hit at this line. If we haven't shown the
1896 * pre-context lines, we would need to show them.
1898 if (opt->pre_context || opt->funcbody)
1899 show_pre_context(opt, gs, bol, eol, lno);
1900 else if (opt->funcname)
1901 show_funcname_line(opt, gs, bol, lno);
1902 cno = opt->invert ? icol : col;
1903 if (cno < 0) {
1905 * A negative cno indicates that there was no
1906 * match on the line. We are thus inverted and
1907 * being asked to show all lines that _don't_
1908 * match a given expression. Therefore, set cno
1909 * to 0 to suggest the whole line matches.
1911 cno = 0;
1913 show_line(opt, bol, eol, gs->name, lno, cno + 1, ':');
1914 last_hit = lno;
1915 if (opt->funcbody)
1916 show_function = 1;
1917 goto next_line;
1919 if (show_function && (!peek_bol || peek_bol < bol)) {
1920 unsigned long peek_left = left;
1921 char *peek_eol = eol;
1924 * Trailing empty lines are not interesting.
1925 * Peek past them to see if they belong to the
1926 * body of the current function.
1928 peek_bol = bol;
1929 while (is_empty_line(peek_bol, peek_eol)) {
1930 peek_bol = peek_eol + 1;
1931 peek_eol = end_of_line(peek_bol, &peek_left);
1934 if (match_funcname(opt, gs, peek_bol, peek_eol))
1935 show_function = 0;
1937 if (show_function ||
1938 (last_hit && lno <= last_hit + opt->post_context)) {
1939 /* If the last hit is within the post context,
1940 * we need to show this line.
1942 show_line(opt, bol, eol, gs->name, lno, col + 1, '-');
1945 next_line:
1946 bol = eol + 1;
1947 if (!left)
1948 break;
1949 left--;
1950 lno++;
1953 if (collect_hits)
1954 return 0;
1956 if (opt->status_only)
1957 return opt->unmatch_name_only;
1958 if (opt->unmatch_name_only) {
1959 /* We did not see any hit, so we want to show this */
1960 show_name(opt, gs->name);
1961 return 1;
1964 xdiff_clear_find_func(&xecfg);
1965 opt->priv = NULL;
1967 /* NEEDSWORK:
1968 * The real "grep -c foo *.c" gives many "bar.c:0" lines,
1969 * which feels mostly useless but sometimes useful. Maybe
1970 * make it another option? For now suppress them.
1972 if (opt->count && count) {
1973 char buf[32];
1974 if (opt->pathname) {
1975 output_color(opt, gs->name, strlen(gs->name),
1976 opt->colors[GREP_COLOR_FILENAME]);
1977 output_sep(opt, ':');
1979 xsnprintf(buf, sizeof(buf), "%u\n", count);
1980 opt->output(opt, buf, strlen(buf));
1981 return 1;
1983 return !!last_hit;
1986 static void clr_hit_marker(struct grep_expr *x)
1988 /* All-hit markers are meaningful only at the very top level
1989 * OR node.
1991 while (1) {
1992 x->hit = 0;
1993 if (x->node != GREP_NODE_OR)
1994 return;
1995 x->u.binary.left->hit = 0;
1996 x = x->u.binary.right;
2000 static int chk_hit_marker(struct grep_expr *x)
2002 /* Top level nodes have hit markers. See if they all are hits */
2003 while (1) {
2004 if (x->node != GREP_NODE_OR)
2005 return x->hit;
2006 if (!x->u.binary.left->hit)
2007 return 0;
2008 x = x->u.binary.right;
2012 int grep_source(struct grep_opt *opt, struct grep_source *gs)
2015 * we do not have to do the two-pass grep when we do not check
2016 * buffer-wide "all-match".
2018 if (!opt->all_match)
2019 return grep_source_1(opt, gs, 0);
2021 /* Otherwise the toplevel "or" terms hit a bit differently.
2022 * We first clear hit markers from them.
2024 clr_hit_marker(opt->pattern_expression);
2025 grep_source_1(opt, gs, 1);
2027 if (!chk_hit_marker(opt->pattern_expression))
2028 return 0;
2030 return grep_source_1(opt, gs, 0);
2033 int grep_buffer(struct grep_opt *opt, char *buf, unsigned long size)
2035 struct grep_source gs;
2036 int r;
2038 grep_source_init(&gs, GREP_SOURCE_BUF, NULL, NULL, NULL);
2039 gs.buf = buf;
2040 gs.size = size;
2042 r = grep_source(opt, &gs);
2044 grep_source_clear(&gs);
2045 return r;
2048 void grep_source_init(struct grep_source *gs, enum grep_source_type type,
2049 const char *name, const char *path,
2050 const void *identifier)
2052 gs->type = type;
2053 gs->name = xstrdup_or_null(name);
2054 gs->path = xstrdup_or_null(path);
2055 gs->buf = NULL;
2056 gs->size = 0;
2057 gs->driver = NULL;
2059 switch (type) {
2060 case GREP_SOURCE_FILE:
2061 gs->identifier = xstrdup(identifier);
2062 break;
2063 case GREP_SOURCE_OID:
2064 gs->identifier = oiddup(identifier);
2065 break;
2066 case GREP_SOURCE_BUF:
2067 gs->identifier = NULL;
2068 break;
2072 void grep_source_clear(struct grep_source *gs)
2074 FREE_AND_NULL(gs->name);
2075 FREE_AND_NULL(gs->path);
2076 FREE_AND_NULL(gs->identifier);
2077 grep_source_clear_data(gs);
2080 void grep_source_clear_data(struct grep_source *gs)
2082 switch (gs->type) {
2083 case GREP_SOURCE_FILE:
2084 case GREP_SOURCE_OID:
2085 FREE_AND_NULL(gs->buf);
2086 gs->size = 0;
2087 break;
2088 case GREP_SOURCE_BUF:
2089 /* leave user-provided buf intact */
2090 break;
2094 static int grep_source_load_oid(struct grep_source *gs)
2096 enum object_type type;
2098 gs->buf = read_object_file(gs->identifier, &type, &gs->size);
2099 if (!gs->buf)
2100 return error(_("'%s': unable to read %s"),
2101 gs->name,
2102 oid_to_hex(gs->identifier));
2103 return 0;
2106 static int grep_source_load_file(struct grep_source *gs)
2108 const char *filename = gs->identifier;
2109 struct stat st;
2110 char *data;
2111 size_t size;
2112 int i;
2114 if (lstat(filename, &st) < 0) {
2115 err_ret:
2116 if (errno != ENOENT)
2117 error_errno(_("failed to stat '%s'"), filename);
2118 return -1;
2120 if (!S_ISREG(st.st_mode))
2121 return -1;
2122 size = xsize_t(st.st_size);
2123 i = open(filename, O_RDONLY);
2124 if (i < 0)
2125 goto err_ret;
2126 data = xmallocz(size);
2127 if (st.st_size != read_in_full(i, data, size)) {
2128 error_errno(_("'%s': short read"), filename);
2129 close(i);
2130 free(data);
2131 return -1;
2133 close(i);
2135 gs->buf = data;
2136 gs->size = size;
2137 return 0;
2140 static int grep_source_load(struct grep_source *gs)
2142 if (gs->buf)
2143 return 0;
2145 switch (gs->type) {
2146 case GREP_SOURCE_FILE:
2147 return grep_source_load_file(gs);
2148 case GREP_SOURCE_OID:
2149 return grep_source_load_oid(gs);
2150 case GREP_SOURCE_BUF:
2151 return gs->buf ? 0 : -1;
2153 BUG("invalid grep_source type to load");
2156 void grep_source_load_driver(struct grep_source *gs,
2157 struct index_state *istate)
2159 if (gs->driver)
2160 return;
2162 grep_attr_lock();
2163 if (gs->path)
2164 gs->driver = userdiff_find_by_path(istate, gs->path);
2165 if (!gs->driver)
2166 gs->driver = userdiff_find_by_name("default");
2167 grep_attr_unlock();
2170 static int grep_source_is_binary(struct grep_source *gs,
2171 struct index_state *istate)
2173 grep_source_load_driver(gs, istate);
2174 if (gs->driver->binary != -1)
2175 return gs->driver->binary;
2177 if (!grep_source_load(gs))
2178 return buffer_is_binary(gs->buf, gs->size);
2180 return 0;