Merge branch 'en/keep-cwd'
[git/debian.git] / grep.c
blobfe847a0111a209279656c5a14318d2fa196df2ed
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 static const char *color_grep_slots[] = {
44 [GREP_COLOR_CONTEXT] = "context",
45 [GREP_COLOR_FILENAME] = "filename",
46 [GREP_COLOR_FUNCTION] = "function",
47 [GREP_COLOR_LINENO] = "lineNumber",
48 [GREP_COLOR_COLUMNNO] = "column",
49 [GREP_COLOR_MATCH_CONTEXT] = "matchContext",
50 [GREP_COLOR_MATCH_SELECTED] = "matchSelected",
51 [GREP_COLOR_SELECTED] = "selected",
52 [GREP_COLOR_SEP] = "separator",
55 static int parse_pattern_type_arg(const char *opt, const char *arg)
57 if (!strcmp(arg, "default"))
58 return GREP_PATTERN_TYPE_UNSPECIFIED;
59 else if (!strcmp(arg, "basic"))
60 return GREP_PATTERN_TYPE_BRE;
61 else if (!strcmp(arg, "extended"))
62 return GREP_PATTERN_TYPE_ERE;
63 else if (!strcmp(arg, "fixed"))
64 return GREP_PATTERN_TYPE_FIXED;
65 else if (!strcmp(arg, "perl"))
66 return GREP_PATTERN_TYPE_PCRE;
67 die("bad %s argument: %s", opt, arg);
70 define_list_config_array_extra(color_grep_slots, {"match"});
73 * Read the configuration file once and store it in
74 * the grep_defaults template.
76 int grep_config(const char *var, const char *value, void *cb)
78 struct grep_opt *opt = &grep_defaults;
79 const char *slot;
81 if (userdiff_config(var, value) < 0)
82 return -1;
85 * The instance of grep_opt that we set up here is copied by
86 * grep_init() to be used by each individual invocation.
87 * When populating a new field of this structure here, be
88 * sure to think about ownership -- e.g., you might need to
89 * override the shallow copy in grep_init() with a deep copy.
92 if (!strcmp(var, "grep.extendedregexp")) {
93 opt->extended_regexp_option = git_config_bool(var, value);
94 return 0;
97 if (!strcmp(var, "grep.patterntype")) {
98 opt->pattern_type_option = parse_pattern_type_arg(var, value);
99 return 0;
102 if (!strcmp(var, "grep.linenumber")) {
103 opt->linenum = git_config_bool(var, value);
104 return 0;
106 if (!strcmp(var, "grep.column")) {
107 opt->columnnum = git_config_bool(var, value);
108 return 0;
111 if (!strcmp(var, "grep.fullname")) {
112 opt->relative = !git_config_bool(var, value);
113 return 0;
116 if (!strcmp(var, "color.grep"))
117 opt->color = git_config_colorbool(var, value);
118 if (!strcmp(var, "color.grep.match")) {
119 if (grep_config("color.grep.matchcontext", value, cb) < 0)
120 return -1;
121 if (grep_config("color.grep.matchselected", value, cb) < 0)
122 return -1;
123 } else if (skip_prefix(var, "color.grep.", &slot)) {
124 int i = LOOKUP_CONFIG(color_grep_slots, slot);
125 char *color;
127 if (i < 0)
128 return -1;
129 color = opt->colors[i];
130 if (!value)
131 return config_error_nonbool(var);
132 return color_parse(value, color);
134 return 0;
138 * Initialize one instance of grep_opt and copy the
139 * default values from the template we read the configuration
140 * information in an earlier call to git_config(grep_config).
142 void grep_init(struct grep_opt *opt, struct repository *repo, const char *prefix)
144 *opt = grep_defaults;
146 opt->repo = repo;
147 opt->prefix = prefix;
148 opt->prefix_length = (prefix && *prefix) ? strlen(prefix) : 0;
149 opt->pattern_tail = &opt->pattern_list;
150 opt->header_tail = &opt->header_list;
153 static void grep_set_pattern_type_option(enum grep_pattern_type pattern_type, struct grep_opt *opt)
156 * When committing to the pattern type by setting the relevant
157 * fields in grep_opt it's generally not necessary to zero out
158 * the fields we're not choosing, since they won't have been
159 * set by anything. The extended_regexp_option field is the
160 * only exception to this.
162 * This is because in the process of parsing grep.patternType
163 * & grep.extendedRegexp we set opt->pattern_type_option and
164 * opt->extended_regexp_option, respectively. We then
165 * internally use opt->extended_regexp_option to see if we're
166 * compiling an ERE. It must be unset if that's not actually
167 * the case.
169 if (pattern_type != GREP_PATTERN_TYPE_ERE &&
170 opt->extended_regexp_option)
171 opt->extended_regexp_option = 0;
173 switch (pattern_type) {
174 case GREP_PATTERN_TYPE_UNSPECIFIED:
175 /* fall through */
177 case GREP_PATTERN_TYPE_BRE:
178 break;
180 case GREP_PATTERN_TYPE_ERE:
181 opt->extended_regexp_option = 1;
182 break;
184 case GREP_PATTERN_TYPE_FIXED:
185 opt->fixed = 1;
186 break;
188 case GREP_PATTERN_TYPE_PCRE:
189 opt->pcre2 = 1;
190 break;
194 void grep_commit_pattern_type(enum grep_pattern_type pattern_type, struct grep_opt *opt)
196 if (pattern_type != GREP_PATTERN_TYPE_UNSPECIFIED)
197 grep_set_pattern_type_option(pattern_type, opt);
198 else if (opt->pattern_type_option != GREP_PATTERN_TYPE_UNSPECIFIED)
199 grep_set_pattern_type_option(opt->pattern_type_option, opt);
200 else if (opt->extended_regexp_option)
202 * This branch *must* happen after setting from the
203 * opt->pattern_type_option above, we don't want
204 * grep.extendedRegexp to override grep.patternType!
206 grep_set_pattern_type_option(GREP_PATTERN_TYPE_ERE, opt);
209 static struct grep_pat *create_grep_pat(const char *pat, size_t patlen,
210 const char *origin, int no,
211 enum grep_pat_token t,
212 enum grep_header_field field)
214 struct grep_pat *p = xcalloc(1, sizeof(*p));
215 p->pattern = xmemdupz(pat, patlen);
216 p->patternlen = patlen;
217 p->origin = origin;
218 p->no = no;
219 p->token = t;
220 p->field = field;
221 return p;
224 static void do_append_grep_pat(struct grep_pat ***tail, struct grep_pat *p)
226 **tail = p;
227 *tail = &p->next;
228 p->next = NULL;
230 switch (p->token) {
231 case GREP_PATTERN: /* atom */
232 case GREP_PATTERN_HEAD:
233 case GREP_PATTERN_BODY:
234 for (;;) {
235 struct grep_pat *new_pat;
236 size_t len = 0;
237 char *cp = p->pattern + p->patternlen, *nl = NULL;
238 while (++len <= p->patternlen) {
239 if (*(--cp) == '\n') {
240 nl = cp;
241 break;
244 if (!nl)
245 break;
246 new_pat = create_grep_pat(nl + 1, len - 1, p->origin,
247 p->no, p->token, p->field);
248 new_pat->next = p->next;
249 if (!p->next)
250 *tail = &new_pat->next;
251 p->next = new_pat;
252 *nl = '\0';
253 p->patternlen -= len;
255 break;
256 default:
257 break;
261 void append_header_grep_pattern(struct grep_opt *opt,
262 enum grep_header_field field, const char *pat)
264 struct grep_pat *p = create_grep_pat(pat, strlen(pat), "header", 0,
265 GREP_PATTERN_HEAD, field);
266 if (field == GREP_HEADER_REFLOG)
267 opt->use_reflog_filter = 1;
268 do_append_grep_pat(&opt->header_tail, p);
271 void append_grep_pattern(struct grep_opt *opt, const char *pat,
272 const char *origin, int no, enum grep_pat_token t)
274 append_grep_pat(opt, pat, strlen(pat), origin, no, t);
277 void append_grep_pat(struct grep_opt *opt, const char *pat, size_t patlen,
278 const char *origin, int no, enum grep_pat_token t)
280 struct grep_pat *p = create_grep_pat(pat, patlen, origin, no, t, 0);
281 do_append_grep_pat(&opt->pattern_tail, p);
284 struct grep_opt *grep_opt_dup(const struct grep_opt *opt)
286 struct grep_pat *pat;
287 struct grep_opt *ret = xmalloc(sizeof(struct grep_opt));
288 *ret = *opt;
290 ret->pattern_list = NULL;
291 ret->pattern_tail = &ret->pattern_list;
293 for(pat = opt->pattern_list; pat != NULL; pat = pat->next)
295 if(pat->token == GREP_PATTERN_HEAD)
296 append_header_grep_pattern(ret, pat->field,
297 pat->pattern);
298 else
299 append_grep_pat(ret, pat->pattern, pat->patternlen,
300 pat->origin, pat->no, pat->token);
303 return ret;
306 static NORETURN void compile_regexp_failed(const struct grep_pat *p,
307 const char *error)
309 char where[1024];
311 if (p->no)
312 xsnprintf(where, sizeof(where), "In '%s' at %d, ", p->origin, p->no);
313 else if (p->origin)
314 xsnprintf(where, sizeof(where), "%s, ", p->origin);
315 else
316 where[0] = 0;
318 die("%s'%s': %s", where, p->pattern, error);
321 static int is_fixed(const char *s, size_t len)
323 size_t i;
325 for (i = 0; i < len; i++) {
326 if (is_regex_special(s[i]))
327 return 0;
330 return 1;
333 #ifdef USE_LIBPCRE2
334 #define GREP_PCRE2_DEBUG_MALLOC 0
336 static void *pcre2_malloc(PCRE2_SIZE size, MAYBE_UNUSED void *memory_data)
338 void *pointer = malloc(size);
339 #if GREP_PCRE2_DEBUG_MALLOC
340 static int count = 1;
341 fprintf(stderr, "PCRE2:%p -> #%02d: alloc(%lu)\n", pointer, count++, size);
342 #endif
343 return pointer;
346 static void pcre2_free(void *pointer, MAYBE_UNUSED void *memory_data)
348 #if GREP_PCRE2_DEBUG_MALLOC
349 static int count = 1;
350 if (pointer)
351 fprintf(stderr, "PCRE2:%p -> #%02d: free()\n", pointer, count++);
352 #endif
353 free(pointer);
356 static void compile_pcre2_pattern(struct grep_pat *p, const struct grep_opt *opt)
358 int error;
359 PCRE2_UCHAR errbuf[256];
360 PCRE2_SIZE erroffset;
361 int options = PCRE2_MULTILINE;
362 int jitret;
363 int patinforet;
364 size_t jitsizearg;
367 * Call pcre2_general_context_create() before calling any
368 * other pcre2_*(). It sets up our malloc()/free() functions
369 * with which everything else is allocated.
371 p->pcre2_general_context = pcre2_general_context_create(
372 pcre2_malloc, pcre2_free, NULL);
373 if (!p->pcre2_general_context)
374 die("Couldn't allocate PCRE2 general context");
376 if (opt->ignore_case) {
377 if (!opt->ignore_locale && has_non_ascii(p->pattern)) {
378 p->pcre2_tables = pcre2_maketables(p->pcre2_general_context);
379 p->pcre2_compile_context = pcre2_compile_context_create(p->pcre2_general_context);
380 pcre2_set_character_tables(p->pcre2_compile_context,
381 p->pcre2_tables);
383 options |= PCRE2_CASELESS;
385 if (!opt->ignore_locale && is_utf8_locale() && has_non_ascii(p->pattern) &&
386 !(!opt->ignore_case && (p->fixed || p->is_fixed)))
387 options |= (PCRE2_UTF | PCRE2_MATCH_INVALID_UTF);
389 #ifdef GIT_PCRE2_VERSION_10_36_OR_HIGHER
390 /* Work around https://bugs.exim.org/show_bug.cgi?id=2642 fixed in 10.36 */
391 if (PCRE2_MATCH_INVALID_UTF && options & (PCRE2_UTF | PCRE2_CASELESS))
392 options |= PCRE2_NO_START_OPTIMIZE;
393 #endif
395 p->pcre2_pattern = pcre2_compile((PCRE2_SPTR)p->pattern,
396 p->patternlen, options, &error, &erroffset,
397 p->pcre2_compile_context);
399 if (p->pcre2_pattern) {
400 p->pcre2_match_data = pcre2_match_data_create_from_pattern(p->pcre2_pattern, p->pcre2_general_context);
401 if (!p->pcre2_match_data)
402 die("Couldn't allocate PCRE2 match data");
403 } else {
404 pcre2_get_error_message(error, errbuf, sizeof(errbuf));
405 compile_regexp_failed(p, (const char *)&errbuf);
408 pcre2_config(PCRE2_CONFIG_JIT, &p->pcre2_jit_on);
409 if (p->pcre2_jit_on) {
410 jitret = pcre2_jit_compile(p->pcre2_pattern, PCRE2_JIT_COMPLETE);
411 if (jitret)
412 die("Couldn't JIT the PCRE2 pattern '%s', got '%d'\n", p->pattern, jitret);
415 * The pcre2_config(PCRE2_CONFIG_JIT, ...) call just
416 * tells us whether the library itself supports JIT,
417 * but to see whether we're going to be actually using
418 * JIT we need to extract PCRE2_INFO_JITSIZE from the
419 * pattern *after* we do pcre2_jit_compile() above.
421 * This is because if the pattern contains the
422 * (*NO_JIT) verb (see pcre2syntax(3))
423 * pcre2_jit_compile() will exit early with 0. If we
424 * then proceed to call pcre2_jit_match() further down
425 * the line instead of pcre2_match() we'll either
426 * segfault (pre PCRE 10.31) or run into a fatal error
427 * (post PCRE2 10.31)
429 patinforet = pcre2_pattern_info(p->pcre2_pattern, PCRE2_INFO_JITSIZE, &jitsizearg);
430 if (patinforet)
431 BUG("pcre2_pattern_info() failed: %d", patinforet);
432 if (jitsizearg == 0) {
433 p->pcre2_jit_on = 0;
434 return;
439 static int pcre2match(struct grep_pat *p, const char *line, const char *eol,
440 regmatch_t *match, int eflags)
442 int ret, flags = 0;
443 PCRE2_SIZE *ovector;
444 PCRE2_UCHAR errbuf[256];
446 if (eflags & REG_NOTBOL)
447 flags |= PCRE2_NOTBOL;
449 if (p->pcre2_jit_on)
450 ret = pcre2_jit_match(p->pcre2_pattern, (unsigned char *)line,
451 eol - line, 0, flags, p->pcre2_match_data,
452 NULL);
453 else
454 ret = pcre2_match(p->pcre2_pattern, (unsigned char *)line,
455 eol - line, 0, flags, p->pcre2_match_data,
456 NULL);
458 if (ret < 0 && ret != PCRE2_ERROR_NOMATCH) {
459 pcre2_get_error_message(ret, errbuf, sizeof(errbuf));
460 die("%s failed with error code %d: %s",
461 (p->pcre2_jit_on ? "pcre2_jit_match" : "pcre2_match"), ret,
462 errbuf);
464 if (ret > 0) {
465 ovector = pcre2_get_ovector_pointer(p->pcre2_match_data);
466 ret = 0;
467 match->rm_so = (int)ovector[0];
468 match->rm_eo = (int)ovector[1];
471 return ret;
474 static void free_pcre2_pattern(struct grep_pat *p)
476 pcre2_compile_context_free(p->pcre2_compile_context);
477 pcre2_code_free(p->pcre2_pattern);
478 pcre2_match_data_free(p->pcre2_match_data);
479 #ifdef GIT_PCRE2_VERSION_10_34_OR_HIGHER
480 pcre2_maketables_free(p->pcre2_general_context, p->pcre2_tables);
481 #else
482 free((void *)p->pcre2_tables);
483 #endif
484 pcre2_general_context_free(p->pcre2_general_context);
486 #else /* !USE_LIBPCRE2 */
487 static void compile_pcre2_pattern(struct grep_pat *p, const struct grep_opt *opt)
489 die("cannot use Perl-compatible regexes when not compiled with USE_LIBPCRE");
492 static int pcre2match(struct grep_pat *p, const char *line, const char *eol,
493 regmatch_t *match, int eflags)
495 return 1;
498 static void free_pcre2_pattern(struct grep_pat *p)
502 static void compile_fixed_regexp(struct grep_pat *p, struct grep_opt *opt)
504 struct strbuf sb = STRBUF_INIT;
505 int err;
506 int regflags = 0;
508 basic_regex_quote_buf(&sb, p->pattern);
509 if (opt->ignore_case)
510 regflags |= REG_ICASE;
511 err = regcomp(&p->regexp, sb.buf, regflags);
512 strbuf_release(&sb);
513 if (err) {
514 char errbuf[1024];
515 regerror(err, &p->regexp, errbuf, sizeof(errbuf));
516 compile_regexp_failed(p, errbuf);
519 #endif /* !USE_LIBPCRE2 */
521 static void compile_regexp(struct grep_pat *p, struct grep_opt *opt)
523 int err;
524 int regflags = REG_NEWLINE;
526 p->word_regexp = opt->word_regexp;
527 p->ignore_case = opt->ignore_case;
528 p->fixed = opt->fixed;
530 if (memchr(p->pattern, 0, p->patternlen) && !opt->pcre2)
531 die(_("given pattern contains NULL byte (via -f <file>). This is only supported with -P under PCRE v2"));
533 p->is_fixed = is_fixed(p->pattern, p->patternlen);
534 #ifdef USE_LIBPCRE2
535 if (!p->fixed && !p->is_fixed) {
536 const char *no_jit = "(*NO_JIT)";
537 const int no_jit_len = strlen(no_jit);
538 if (starts_with(p->pattern, no_jit) &&
539 is_fixed(p->pattern + no_jit_len,
540 p->patternlen - no_jit_len))
541 p->is_fixed = 1;
543 #endif
544 if (p->fixed || p->is_fixed) {
545 #ifdef USE_LIBPCRE2
546 if (p->is_fixed) {
547 compile_pcre2_pattern(p, opt);
548 } else {
550 * E.g. t7811-grep-open.sh relies on the
551 * pattern being restored.
553 char *old_pattern = p->pattern;
554 size_t old_patternlen = p->patternlen;
555 struct strbuf sb = STRBUF_INIT;
558 * There is the PCRE2_LITERAL flag, but it's
559 * only in PCRE v2 10.30 and later. Needing to
560 * ifdef our way around that and dealing with
561 * it + PCRE2_MULTILINE being an error is more
562 * complex than just quoting this ourselves.
564 strbuf_add(&sb, "\\Q", 2);
565 strbuf_add(&sb, p->pattern, p->patternlen);
566 strbuf_add(&sb, "\\E", 2);
568 p->pattern = sb.buf;
569 p->patternlen = sb.len;
570 compile_pcre2_pattern(p, opt);
571 p->pattern = old_pattern;
572 p->patternlen = old_patternlen;
573 strbuf_release(&sb);
575 #else /* !USE_LIBPCRE2 */
576 compile_fixed_regexp(p, opt);
577 #endif /* !USE_LIBPCRE2 */
578 return;
581 if (opt->pcre2) {
582 compile_pcre2_pattern(p, opt);
583 return;
586 if (p->ignore_case)
587 regflags |= REG_ICASE;
588 if (opt->extended_regexp_option)
589 regflags |= REG_EXTENDED;
590 err = regcomp(&p->regexp, p->pattern, regflags);
591 if (err) {
592 char errbuf[1024];
593 regerror(err, &p->regexp, errbuf, 1024);
594 compile_regexp_failed(p, errbuf);
598 static struct grep_expr *compile_pattern_or(struct grep_pat **);
599 static struct grep_expr *compile_pattern_atom(struct grep_pat **list)
601 struct grep_pat *p;
602 struct grep_expr *x;
604 p = *list;
605 if (!p)
606 return NULL;
607 switch (p->token) {
608 case GREP_PATTERN: /* atom */
609 case GREP_PATTERN_HEAD:
610 case GREP_PATTERN_BODY:
611 CALLOC_ARRAY(x, 1);
612 x->node = GREP_NODE_ATOM;
613 x->u.atom = p;
614 *list = p->next;
615 return x;
616 case GREP_OPEN_PAREN:
617 *list = p->next;
618 x = compile_pattern_or(list);
619 if (!*list || (*list)->token != GREP_CLOSE_PAREN)
620 die("unmatched parenthesis");
621 *list = (*list)->next;
622 return x;
623 default:
624 return NULL;
628 static struct grep_expr *compile_pattern_not(struct grep_pat **list)
630 struct grep_pat *p;
631 struct grep_expr *x;
633 p = *list;
634 if (!p)
635 return NULL;
636 switch (p->token) {
637 case GREP_NOT:
638 if (!p->next)
639 die("--not not followed by pattern expression");
640 *list = p->next;
641 CALLOC_ARRAY(x, 1);
642 x->node = GREP_NODE_NOT;
643 x->u.unary = compile_pattern_not(list);
644 if (!x->u.unary)
645 die("--not followed by non pattern expression");
646 return x;
647 default:
648 return compile_pattern_atom(list);
652 static struct grep_expr *compile_pattern_and(struct grep_pat **list)
654 struct grep_pat *p;
655 struct grep_expr *x, *y, *z;
657 x = compile_pattern_not(list);
658 p = *list;
659 if (p && p->token == GREP_AND) {
660 if (!x)
661 die("--and not preceded by pattern expression");
662 if (!p->next)
663 die("--and not followed by pattern expression");
664 *list = p->next;
665 y = compile_pattern_and(list);
666 if (!y)
667 die("--and not followed by pattern expression");
668 CALLOC_ARRAY(z, 1);
669 z->node = GREP_NODE_AND;
670 z->u.binary.left = x;
671 z->u.binary.right = y;
672 return z;
674 return x;
677 static struct grep_expr *compile_pattern_or(struct grep_pat **list)
679 struct grep_pat *p;
680 struct grep_expr *x, *y, *z;
682 x = compile_pattern_and(list);
683 p = *list;
684 if (x && p && p->token != GREP_CLOSE_PAREN) {
685 y = compile_pattern_or(list);
686 if (!y)
687 die("not a pattern expression %s", p->pattern);
688 CALLOC_ARRAY(z, 1);
689 z->node = GREP_NODE_OR;
690 z->u.binary.left = x;
691 z->u.binary.right = y;
692 return z;
694 return x;
697 static struct grep_expr *compile_pattern_expr(struct grep_pat **list)
699 return compile_pattern_or(list);
702 static struct grep_expr *grep_true_expr(void)
704 struct grep_expr *z = xcalloc(1, sizeof(*z));
705 z->node = GREP_NODE_TRUE;
706 return z;
709 static struct grep_expr *grep_or_expr(struct grep_expr *left, struct grep_expr *right)
711 struct grep_expr *z = xcalloc(1, sizeof(*z));
712 z->node = GREP_NODE_OR;
713 z->u.binary.left = left;
714 z->u.binary.right = right;
715 return z;
718 static struct grep_expr *prep_header_patterns(struct grep_opt *opt)
720 struct grep_pat *p;
721 struct grep_expr *header_expr;
722 struct grep_expr *(header_group[GREP_HEADER_FIELD_MAX]);
723 enum grep_header_field fld;
725 if (!opt->header_list)
726 return NULL;
728 for (p = opt->header_list; p; p = p->next) {
729 if (p->token != GREP_PATTERN_HEAD)
730 BUG("a non-header pattern in grep header list.");
731 if (p->field < GREP_HEADER_FIELD_MIN ||
732 GREP_HEADER_FIELD_MAX <= p->field)
733 BUG("unknown header field %d", p->field);
734 compile_regexp(p, opt);
737 for (fld = 0; fld < GREP_HEADER_FIELD_MAX; fld++)
738 header_group[fld] = NULL;
740 for (p = opt->header_list; p; p = p->next) {
741 struct grep_expr *h;
742 struct grep_pat *pp = p;
744 h = compile_pattern_atom(&pp);
745 if (!h || pp != p->next)
746 BUG("malformed header expr");
747 if (!header_group[p->field]) {
748 header_group[p->field] = h;
749 continue;
751 header_group[p->field] = grep_or_expr(h, header_group[p->field]);
754 header_expr = NULL;
756 for (fld = 0; fld < GREP_HEADER_FIELD_MAX; fld++) {
757 if (!header_group[fld])
758 continue;
759 if (!header_expr)
760 header_expr = grep_true_expr();
761 header_expr = grep_or_expr(header_group[fld], header_expr);
763 return header_expr;
766 static struct grep_expr *grep_splice_or(struct grep_expr *x, struct grep_expr *y)
768 struct grep_expr *z = x;
770 while (x) {
771 assert(x->node == GREP_NODE_OR);
772 if (x->u.binary.right &&
773 x->u.binary.right->node == GREP_NODE_TRUE) {
774 x->u.binary.right = y;
775 break;
777 x = x->u.binary.right;
779 return z;
782 void compile_grep_patterns(struct grep_opt *opt)
784 struct grep_pat *p;
785 struct grep_expr *header_expr = prep_header_patterns(opt);
787 for (p = opt->pattern_list; p; p = p->next) {
788 switch (p->token) {
789 case GREP_PATTERN: /* atom */
790 case GREP_PATTERN_HEAD:
791 case GREP_PATTERN_BODY:
792 compile_regexp(p, opt);
793 break;
794 default:
795 opt->extended = 1;
796 break;
800 if (opt->all_match || header_expr)
801 opt->extended = 1;
802 else if (!opt->extended)
803 return;
805 p = opt->pattern_list;
806 if (p)
807 opt->pattern_expression = compile_pattern_expr(&p);
808 if (p)
809 die("incomplete pattern expression: %s", p->pattern);
811 if (!header_expr)
812 return;
814 if (!opt->pattern_expression)
815 opt->pattern_expression = header_expr;
816 else if (opt->all_match)
817 opt->pattern_expression = grep_splice_or(header_expr,
818 opt->pattern_expression);
819 else
820 opt->pattern_expression = grep_or_expr(opt->pattern_expression,
821 header_expr);
822 opt->all_match = 1;
825 static void free_pattern_expr(struct grep_expr *x)
827 switch (x->node) {
828 case GREP_NODE_TRUE:
829 case GREP_NODE_ATOM:
830 break;
831 case GREP_NODE_NOT:
832 free_pattern_expr(x->u.unary);
833 break;
834 case GREP_NODE_AND:
835 case GREP_NODE_OR:
836 free_pattern_expr(x->u.binary.left);
837 free_pattern_expr(x->u.binary.right);
838 break;
840 free(x);
843 void free_grep_patterns(struct grep_opt *opt)
845 struct grep_pat *p, *n;
847 for (p = opt->pattern_list; p; p = n) {
848 n = p->next;
849 switch (p->token) {
850 case GREP_PATTERN: /* atom */
851 case GREP_PATTERN_HEAD:
852 case GREP_PATTERN_BODY:
853 if (p->pcre2_pattern)
854 free_pcre2_pattern(p);
855 else
856 regfree(&p->regexp);
857 free(p->pattern);
858 break;
859 default:
860 break;
862 free(p);
865 if (!opt->extended)
866 return;
867 free_pattern_expr(opt->pattern_expression);
870 static const char *end_of_line(const char *cp, unsigned long *left)
872 unsigned long l = *left;
873 while (l && *cp != '\n') {
874 l--;
875 cp++;
877 *left = l;
878 return cp;
881 static int word_char(char ch)
883 return isalnum(ch) || ch == '_';
886 static void output_color(struct grep_opt *opt, const void *data, size_t size,
887 const char *color)
889 if (want_color(opt->color) && color && color[0]) {
890 opt->output(opt, color, strlen(color));
891 opt->output(opt, data, size);
892 opt->output(opt, GIT_COLOR_RESET, strlen(GIT_COLOR_RESET));
893 } else
894 opt->output(opt, data, size);
897 static void output_sep(struct grep_opt *opt, char sign)
899 if (opt->null_following_name)
900 opt->output(opt, "\0", 1);
901 else
902 output_color(opt, &sign, 1, opt->colors[GREP_COLOR_SEP]);
905 static void show_name(struct grep_opt *opt, const char *name)
907 output_color(opt, name, strlen(name), opt->colors[GREP_COLOR_FILENAME]);
908 opt->output(opt, opt->null_following_name ? "\0" : "\n", 1);
911 static int patmatch(struct grep_pat *p,
912 const char *line, const char *eol,
913 regmatch_t *match, int eflags)
915 int hit;
917 if (p->pcre2_pattern)
918 hit = !pcre2match(p, line, eol, match, eflags);
919 else
920 hit = !regexec_buf(&p->regexp, line, eol - line, 1, match,
921 eflags);
923 return hit;
926 static void strip_timestamp(const char *bol, const char **eol_p)
928 const char *eol = *eol_p;
930 while (bol < --eol) {
931 if (*eol != '>')
932 continue;
933 *eol_p = ++eol;
934 break;
938 static struct {
939 const char *field;
940 size_t len;
941 } header_field[] = {
942 { "author ", 7 },
943 { "committer ", 10 },
944 { "reflog ", 7 },
947 static int headerless_match_one_pattern(struct grep_pat *p,
948 const char *bol, const char *eol,
949 enum grep_context ctx,
950 regmatch_t *pmatch, int eflags)
952 int hit = 0;
953 const char *start = bol;
955 if ((p->token != GREP_PATTERN) &&
956 ((p->token == GREP_PATTERN_HEAD) != (ctx == GREP_CONTEXT_HEAD)))
957 return 0;
959 again:
960 hit = patmatch(p, bol, eol, pmatch, eflags);
962 if (hit && p->word_regexp) {
963 if ((pmatch[0].rm_so < 0) ||
964 (eol - bol) < pmatch[0].rm_so ||
965 (pmatch[0].rm_eo < 0) ||
966 (eol - bol) < pmatch[0].rm_eo)
967 die("regexp returned nonsense");
969 /* Match beginning must be either beginning of the
970 * line, or at word boundary (i.e. the last char must
971 * not be a word char). Similarly, match end must be
972 * either end of the line, or at word boundary
973 * (i.e. the next char must not be a word char).
975 if ( ((pmatch[0].rm_so == 0) ||
976 !word_char(bol[pmatch[0].rm_so-1])) &&
977 ((pmatch[0].rm_eo == (eol-bol)) ||
978 !word_char(bol[pmatch[0].rm_eo])) )
980 else
981 hit = 0;
983 /* Words consist of at least one character. */
984 if (pmatch->rm_so == pmatch->rm_eo)
985 hit = 0;
987 if (!hit && pmatch[0].rm_so + bol + 1 < eol) {
988 /* There could be more than one match on the
989 * line, and the first match might not be
990 * strict word match. But later ones could be!
991 * Forward to the next possible start, i.e. the
992 * next position following a non-word char.
994 bol = pmatch[0].rm_so + bol + 1;
995 while (word_char(bol[-1]) && bol < eol)
996 bol++;
997 eflags |= REG_NOTBOL;
998 if (bol < eol)
999 goto again;
1002 if (hit) {
1003 pmatch[0].rm_so += bol - start;
1004 pmatch[0].rm_eo += bol - start;
1006 return hit;
1009 static int match_one_pattern(struct grep_pat *p,
1010 const char *bol, const char *eol,
1011 enum grep_context ctx, regmatch_t *pmatch,
1012 int eflags)
1014 const char *field;
1015 size_t len;
1017 if (p->token == GREP_PATTERN_HEAD) {
1018 assert(p->field < ARRAY_SIZE(header_field));
1019 field = header_field[p->field].field;
1020 len = header_field[p->field].len;
1021 if (strncmp(bol, field, len))
1022 return 0;
1023 bol += len;
1025 switch (p->field) {
1026 case GREP_HEADER_AUTHOR:
1027 case GREP_HEADER_COMMITTER:
1028 strip_timestamp(bol, &eol);
1029 break;
1030 default:
1031 break;
1035 return headerless_match_one_pattern(p, bol, eol, ctx, pmatch, eflags);
1039 static int match_expr_eval(struct grep_opt *opt, struct grep_expr *x,
1040 const char *bol, const char *eol,
1041 enum grep_context ctx, ssize_t *col,
1042 ssize_t *icol, int collect_hits)
1044 int h = 0;
1046 if (!x)
1047 die("Not a valid grep expression");
1048 switch (x->node) {
1049 case GREP_NODE_TRUE:
1050 h = 1;
1051 break;
1052 case GREP_NODE_ATOM:
1054 regmatch_t tmp;
1055 h = match_one_pattern(x->u.atom, bol, eol, ctx,
1056 &tmp, 0);
1057 if (h && (*col < 0 || tmp.rm_so < *col))
1058 *col = tmp.rm_so;
1060 break;
1061 case GREP_NODE_NOT:
1063 * Upon visiting a GREP_NODE_NOT, col and icol become swapped.
1065 h = !match_expr_eval(opt, x->u.unary, bol, eol, ctx, icol, col,
1067 break;
1068 case GREP_NODE_AND:
1069 h = match_expr_eval(opt, x->u.binary.left, bol, eol, ctx, col,
1070 icol, 0);
1071 if (h || opt->columnnum) {
1073 * Don't short-circuit AND when given --column, since a
1074 * NOT earlier in the tree may turn this into an OR. In
1075 * this case, see the below comment.
1077 h &= match_expr_eval(opt, x->u.binary.right, bol, eol,
1078 ctx, col, icol, 0);
1080 break;
1081 case GREP_NODE_OR:
1082 if (!(collect_hits || opt->columnnum)) {
1084 * Don't short-circuit OR when given --column (or
1085 * collecting hits) to ensure we don't skip a later
1086 * child that would produce an earlier match.
1088 return (match_expr_eval(opt, x->u.binary.left, bol, eol,
1089 ctx, col, icol, 0) ||
1090 match_expr_eval(opt, x->u.binary.right, bol,
1091 eol, ctx, col, icol, 0));
1093 h = match_expr_eval(opt, x->u.binary.left, bol, eol, ctx, col,
1094 icol, 0);
1095 if (collect_hits)
1096 x->u.binary.left->hit |= h;
1097 h |= match_expr_eval(opt, x->u.binary.right, bol, eol, ctx, col,
1098 icol, collect_hits);
1099 break;
1100 default:
1101 die("Unexpected node type (internal error) %d", x->node);
1103 if (collect_hits)
1104 x->hit |= h;
1105 return h;
1108 static int match_expr(struct grep_opt *opt,
1109 const char *bol, const char *eol,
1110 enum grep_context ctx, ssize_t *col,
1111 ssize_t *icol, int collect_hits)
1113 struct grep_expr *x = opt->pattern_expression;
1114 return match_expr_eval(opt, x, bol, eol, ctx, col, icol, collect_hits);
1117 static int match_line(struct grep_opt *opt,
1118 const char *bol, const char *eol,
1119 ssize_t *col, ssize_t *icol,
1120 enum grep_context ctx, int collect_hits)
1122 struct grep_pat *p;
1123 int hit = 0;
1125 if (opt->extended)
1126 return match_expr(opt, bol, eol, ctx, col, icol,
1127 collect_hits);
1129 /* we do not call with collect_hits without being extended */
1130 for (p = opt->pattern_list; p; p = p->next) {
1131 regmatch_t tmp;
1132 if (match_one_pattern(p, bol, eol, ctx, &tmp, 0)) {
1133 hit |= 1;
1134 if (!opt->columnnum) {
1136 * Without --column, any single match on a line
1137 * is enough to know that it needs to be
1138 * printed. With --column, scan _all_ patterns
1139 * to find the earliest.
1141 break;
1143 if (*col < 0 || tmp.rm_so < *col)
1144 *col = tmp.rm_so;
1147 return hit;
1150 static int match_next_pattern(struct grep_pat *p,
1151 const char *bol, const char *eol,
1152 enum grep_context ctx,
1153 regmatch_t *pmatch, int eflags)
1155 regmatch_t match;
1157 if (!headerless_match_one_pattern(p, bol, eol, ctx, &match, eflags))
1158 return 0;
1159 if (match.rm_so < 0 || match.rm_eo < 0)
1160 return 0;
1161 if (pmatch->rm_so >= 0 && pmatch->rm_eo >= 0) {
1162 if (match.rm_so > pmatch->rm_so)
1163 return 1;
1164 if (match.rm_so == pmatch->rm_so && match.rm_eo < pmatch->rm_eo)
1165 return 1;
1167 pmatch->rm_so = match.rm_so;
1168 pmatch->rm_eo = match.rm_eo;
1169 return 1;
1172 int grep_next_match(struct grep_opt *opt,
1173 const char *bol, const char *eol,
1174 enum grep_context ctx, regmatch_t *pmatch,
1175 enum grep_header_field field, int eflags)
1177 struct grep_pat *p;
1178 int hit = 0;
1180 pmatch->rm_so = pmatch->rm_eo = -1;
1181 if (bol < eol) {
1182 for (p = ((ctx == GREP_CONTEXT_HEAD)
1183 ? opt->header_list : opt->pattern_list);
1184 p; p = p->next) {
1185 switch (p->token) {
1186 case GREP_PATTERN_HEAD:
1187 if ((field != GREP_HEADER_FIELD_MAX) &&
1188 (p->field != field))
1189 continue;
1190 /* fall thru */
1191 case GREP_PATTERN: /* atom */
1192 case GREP_PATTERN_BODY:
1193 hit |= match_next_pattern(p, bol, eol, ctx,
1194 pmatch, eflags);
1195 break;
1196 default:
1197 break;
1201 return hit;
1204 static void show_line_header(struct grep_opt *opt, const char *name,
1205 unsigned lno, ssize_t cno, char sign)
1207 if (opt->heading && opt->last_shown == 0) {
1208 output_color(opt, name, strlen(name), opt->colors[GREP_COLOR_FILENAME]);
1209 opt->output(opt, "\n", 1);
1211 opt->last_shown = lno;
1213 if (!opt->heading && opt->pathname) {
1214 output_color(opt, name, strlen(name), opt->colors[GREP_COLOR_FILENAME]);
1215 output_sep(opt, sign);
1217 if (opt->linenum) {
1218 char buf[32];
1219 xsnprintf(buf, sizeof(buf), "%d", lno);
1220 output_color(opt, buf, strlen(buf), opt->colors[GREP_COLOR_LINENO]);
1221 output_sep(opt, sign);
1224 * Treat 'cno' as the 1-indexed offset from the start of a non-context
1225 * line to its first match. Otherwise, 'cno' is 0 indicating that we are
1226 * being called with a context line.
1228 if (opt->columnnum && cno) {
1229 char buf[32];
1230 xsnprintf(buf, sizeof(buf), "%"PRIuMAX, (uintmax_t)cno);
1231 output_color(opt, buf, strlen(buf), opt->colors[GREP_COLOR_COLUMNNO]);
1232 output_sep(opt, sign);
1236 static void show_line(struct grep_opt *opt,
1237 const char *bol, const char *eol,
1238 const char *name, unsigned lno, ssize_t cno, char sign)
1240 int rest = eol - bol;
1241 const char *match_color = NULL;
1242 const char *line_color = NULL;
1244 if (opt->file_break && opt->last_shown == 0) {
1245 if (opt->show_hunk_mark)
1246 opt->output(opt, "\n", 1);
1247 } else if (opt->pre_context || opt->post_context || opt->funcbody) {
1248 if (opt->last_shown == 0) {
1249 if (opt->show_hunk_mark) {
1250 output_color(opt, "--", 2, opt->colors[GREP_COLOR_SEP]);
1251 opt->output(opt, "\n", 1);
1253 } else if (lno > opt->last_shown + 1) {
1254 output_color(opt, "--", 2, opt->colors[GREP_COLOR_SEP]);
1255 opt->output(opt, "\n", 1);
1258 if (!opt->only_matching) {
1260 * In case the line we're being called with contains more than
1261 * one match, leave printing each header to the loop below.
1263 show_line_header(opt, name, lno, cno, sign);
1265 if (opt->color || opt->only_matching) {
1266 regmatch_t match;
1267 enum grep_context ctx = GREP_CONTEXT_BODY;
1268 int eflags = 0;
1270 if (opt->color) {
1271 if (sign == ':')
1272 match_color = opt->colors[GREP_COLOR_MATCH_SELECTED];
1273 else
1274 match_color = opt->colors[GREP_COLOR_MATCH_CONTEXT];
1275 if (sign == ':')
1276 line_color = opt->colors[GREP_COLOR_SELECTED];
1277 else if (sign == '-')
1278 line_color = opt->colors[GREP_COLOR_CONTEXT];
1279 else if (sign == '=')
1280 line_color = opt->colors[GREP_COLOR_FUNCTION];
1282 while (grep_next_match(opt, bol, eol, ctx, &match,
1283 GREP_HEADER_FIELD_MAX, eflags)) {
1284 if (match.rm_so == match.rm_eo)
1285 break;
1287 if (opt->only_matching)
1288 show_line_header(opt, name, lno, cno, sign);
1289 else
1290 output_color(opt, bol, match.rm_so, line_color);
1291 output_color(opt, bol + match.rm_so,
1292 match.rm_eo - match.rm_so, match_color);
1293 if (opt->only_matching)
1294 opt->output(opt, "\n", 1);
1295 bol += match.rm_eo;
1296 cno += match.rm_eo;
1297 rest -= match.rm_eo;
1298 eflags = REG_NOTBOL;
1301 if (!opt->only_matching) {
1302 output_color(opt, bol, rest, line_color);
1303 opt->output(opt, "\n", 1);
1307 int grep_use_locks;
1310 * This lock protects access to the gitattributes machinery, which is
1311 * not thread-safe.
1313 pthread_mutex_t grep_attr_mutex;
1315 static inline void grep_attr_lock(void)
1317 if (grep_use_locks)
1318 pthread_mutex_lock(&grep_attr_mutex);
1321 static inline void grep_attr_unlock(void)
1323 if (grep_use_locks)
1324 pthread_mutex_unlock(&grep_attr_mutex);
1327 static int match_funcname(struct grep_opt *opt, struct grep_source *gs,
1328 const char *bol, const char *eol)
1330 xdemitconf_t *xecfg = opt->priv;
1331 if (xecfg && !xecfg->find_func) {
1332 grep_source_load_driver(gs, opt->repo->index);
1333 if (gs->driver->funcname.pattern) {
1334 const struct userdiff_funcname *pe = &gs->driver->funcname;
1335 xdiff_set_find_func(xecfg, pe->pattern, pe->cflags);
1336 } else {
1337 xecfg = opt->priv = NULL;
1341 if (xecfg) {
1342 char buf[1];
1343 return xecfg->find_func(bol, eol - bol, buf, 1,
1344 xecfg->find_func_priv) >= 0;
1347 if (bol == eol)
1348 return 0;
1349 if (isalpha(*bol) || *bol == '_' || *bol == '$')
1350 return 1;
1351 return 0;
1354 static void show_funcname_line(struct grep_opt *opt, struct grep_source *gs,
1355 const char *bol, unsigned lno)
1357 while (bol > gs->buf) {
1358 const char *eol = --bol;
1360 while (bol > gs->buf && bol[-1] != '\n')
1361 bol--;
1362 lno--;
1364 if (lno <= opt->last_shown)
1365 break;
1367 if (match_funcname(opt, gs, bol, eol)) {
1368 show_line(opt, bol, eol, gs->name, lno, 0, '=');
1369 break;
1374 static int is_empty_line(const char *bol, const char *eol);
1376 static void show_pre_context(struct grep_opt *opt, struct grep_source *gs,
1377 const char *bol, const char *end, unsigned lno)
1379 unsigned cur = lno, from = 1, funcname_lno = 0, orig_from;
1380 int funcname_needed = !!opt->funcname, comment_needed = 0;
1382 if (opt->pre_context < lno)
1383 from = lno - opt->pre_context;
1384 if (from <= opt->last_shown)
1385 from = opt->last_shown + 1;
1386 orig_from = from;
1387 if (opt->funcbody) {
1388 if (match_funcname(opt, gs, bol, end))
1389 comment_needed = 1;
1390 else
1391 funcname_needed = 1;
1392 from = opt->last_shown + 1;
1395 /* Rewind. */
1396 while (bol > gs->buf && cur > from) {
1397 const char *next_bol = bol;
1398 const char *eol = --bol;
1400 while (bol > gs->buf && bol[-1] != '\n')
1401 bol--;
1402 cur--;
1403 if (comment_needed && (is_empty_line(bol, eol) ||
1404 match_funcname(opt, gs, bol, eol))) {
1405 comment_needed = 0;
1406 from = orig_from;
1407 if (cur < from) {
1408 cur++;
1409 bol = next_bol;
1410 break;
1413 if (funcname_needed && match_funcname(opt, gs, bol, eol)) {
1414 funcname_lno = cur;
1415 funcname_needed = 0;
1416 if (opt->funcbody)
1417 comment_needed = 1;
1418 else
1419 from = orig_from;
1423 /* We need to look even further back to find a function signature. */
1424 if (opt->funcname && funcname_needed)
1425 show_funcname_line(opt, gs, bol, cur);
1427 /* Back forward. */
1428 while (cur < lno) {
1429 const char *eol = bol, sign = (cur == funcname_lno) ? '=' : '-';
1431 while (*eol != '\n')
1432 eol++;
1433 show_line(opt, bol, eol, gs->name, cur, 0, sign);
1434 bol = eol + 1;
1435 cur++;
1439 static int should_lookahead(struct grep_opt *opt)
1441 struct grep_pat *p;
1443 if (opt->extended)
1444 return 0; /* punt for too complex stuff */
1445 if (opt->invert)
1446 return 0;
1447 for (p = opt->pattern_list; p; p = p->next) {
1448 if (p->token != GREP_PATTERN)
1449 return 0; /* punt for "header only" and stuff */
1451 return 1;
1454 static int look_ahead(struct grep_opt *opt,
1455 unsigned long *left_p,
1456 unsigned *lno_p,
1457 const char **bol_p)
1459 unsigned lno = *lno_p;
1460 const char *bol = *bol_p;
1461 struct grep_pat *p;
1462 const char *sp, *last_bol;
1463 regoff_t earliest = -1;
1465 for (p = opt->pattern_list; p; p = p->next) {
1466 int hit;
1467 regmatch_t m;
1469 hit = patmatch(p, bol, bol + *left_p, &m, 0);
1470 if (!hit || m.rm_so < 0 || m.rm_eo < 0)
1471 continue;
1472 if (earliest < 0 || m.rm_so < earliest)
1473 earliest = m.rm_so;
1476 if (earliest < 0) {
1477 *bol_p = bol + *left_p;
1478 *left_p = 0;
1479 return 1;
1481 for (sp = bol + earliest; bol < sp && sp[-1] != '\n'; sp--)
1482 ; /* find the beginning of the line */
1483 last_bol = sp;
1485 for (sp = bol; sp < last_bol; sp++) {
1486 if (*sp == '\n')
1487 lno++;
1489 *left_p -= last_bol - bol;
1490 *bol_p = last_bol;
1491 *lno_p = lno;
1492 return 0;
1495 static int fill_textconv_grep(struct repository *r,
1496 struct userdiff_driver *driver,
1497 struct grep_source *gs)
1499 struct diff_filespec *df;
1500 char *buf;
1501 size_t size;
1503 if (!driver || !driver->textconv)
1504 return grep_source_load(gs);
1507 * The textconv interface is intimately tied to diff_filespecs, so we
1508 * have to pretend to be one. If we could unify the grep_source
1509 * and diff_filespec structs, this mess could just go away.
1511 df = alloc_filespec(gs->path);
1512 switch (gs->type) {
1513 case GREP_SOURCE_OID:
1514 fill_filespec(df, gs->identifier, 1, 0100644);
1515 break;
1516 case GREP_SOURCE_FILE:
1517 fill_filespec(df, null_oid(), 0, 0100644);
1518 break;
1519 default:
1520 BUG("attempt to textconv something without a path?");
1524 * fill_textconv is not remotely thread-safe; it modifies the global
1525 * diff tempfile structure, writes to the_repo's odb and might
1526 * internally call thread-unsafe functions such as the
1527 * prepare_packed_git() lazy-initializator. Because of the last two, we
1528 * must ensure mutual exclusion between this call and the object reading
1529 * API, thus we use obj_read_lock() here.
1531 * TODO: allowing text conversion to run in parallel with object
1532 * reading operations might increase performance in the multithreaded
1533 * non-worktreee git-grep with --textconv.
1535 obj_read_lock();
1536 size = fill_textconv(r, driver, df, &buf);
1537 obj_read_unlock();
1538 free_filespec(df);
1541 * The normal fill_textconv usage by the diff machinery would just keep
1542 * the textconv'd buf separate from the diff_filespec. But much of the
1543 * grep code passes around a grep_source and assumes that its "buf"
1544 * pointer is the beginning of the thing we are searching. So let's
1545 * install our textconv'd version into the grep_source, taking care not
1546 * to leak any existing buffer.
1548 grep_source_clear_data(gs);
1549 gs->buf = buf;
1550 gs->size = size;
1552 return 0;
1555 static int is_empty_line(const char *bol, const char *eol)
1557 while (bol < eol && isspace(*bol))
1558 bol++;
1559 return bol == eol;
1562 static int grep_source_1(struct grep_opt *opt, struct grep_source *gs, int collect_hits)
1564 const char *bol;
1565 const char *peek_bol = NULL;
1566 unsigned long left;
1567 unsigned lno = 1;
1568 unsigned last_hit = 0;
1569 int binary_match_only = 0;
1570 unsigned count = 0;
1571 int try_lookahead = 0;
1572 int show_function = 0;
1573 struct userdiff_driver *textconv = NULL;
1574 enum grep_context ctx = GREP_CONTEXT_HEAD;
1575 xdemitconf_t xecfg;
1577 if (!opt->status_only && gs->name == NULL)
1578 BUG("grep call which could print a name requires "
1579 "grep_source.name be non-NULL");
1581 if (!opt->output)
1582 opt->output = std_output;
1584 if (opt->pre_context || opt->post_context || opt->file_break ||
1585 opt->funcbody) {
1586 /* Show hunk marks, except for the first file. */
1587 if (opt->last_shown)
1588 opt->show_hunk_mark = 1;
1590 * If we're using threads then we can't easily identify
1591 * the first file. Always put hunk marks in that case
1592 * and skip the very first one later in work_done().
1594 if (opt->output != std_output)
1595 opt->show_hunk_mark = 1;
1597 opt->last_shown = 0;
1599 if (opt->allow_textconv) {
1600 grep_source_load_driver(gs, opt->repo->index);
1602 * We might set up the shared textconv cache data here, which
1603 * is not thread-safe. Also, get_oid_with_context() and
1604 * parse_object() might be internally called. As they are not
1605 * currently thread-safe and might be racy with object reading,
1606 * obj_read_lock() must be called.
1608 grep_attr_lock();
1609 obj_read_lock();
1610 textconv = userdiff_get_textconv(opt->repo, gs->driver);
1611 obj_read_unlock();
1612 grep_attr_unlock();
1616 * We know the result of a textconv is text, so we only have to care
1617 * about binary handling if we are not using it.
1619 if (!textconv) {
1620 switch (opt->binary) {
1621 case GREP_BINARY_DEFAULT:
1622 if (grep_source_is_binary(gs, opt->repo->index))
1623 binary_match_only = 1;
1624 break;
1625 case GREP_BINARY_NOMATCH:
1626 if (grep_source_is_binary(gs, opt->repo->index))
1627 return 0; /* Assume unmatch */
1628 break;
1629 case GREP_BINARY_TEXT:
1630 break;
1631 default:
1632 BUG("unknown binary handling mode");
1636 memset(&xecfg, 0, sizeof(xecfg));
1637 opt->priv = &xecfg;
1639 try_lookahead = should_lookahead(opt);
1641 if (fill_textconv_grep(opt->repo, textconv, gs) < 0)
1642 return 0;
1644 bol = gs->buf;
1645 left = gs->size;
1646 while (left) {
1647 const char *eol;
1648 int hit;
1649 ssize_t cno;
1650 ssize_t col = -1, icol = -1;
1653 * look_ahead() skips quickly to the line that possibly
1654 * has the next hit; don't call it if we need to do
1655 * something more than just skipping the current line
1656 * in response to an unmatch for the current line. E.g.
1657 * inside a post-context window, we will show the current
1658 * line as a context around the previous hit when it
1659 * doesn't hit.
1661 if (try_lookahead
1662 && !(last_hit
1663 && (show_function ||
1664 lno <= last_hit + opt->post_context))
1665 && look_ahead(opt, &left, &lno, &bol))
1666 break;
1667 eol = end_of_line(bol, &left);
1669 if ((ctx == GREP_CONTEXT_HEAD) && (eol == bol))
1670 ctx = GREP_CONTEXT_BODY;
1672 hit = match_line(opt, bol, eol, &col, &icol, ctx, collect_hits);
1674 if (collect_hits)
1675 goto next_line;
1677 /* "grep -v -e foo -e bla" should list lines
1678 * that do not have either, so inversion should
1679 * be done outside.
1681 if (opt->invert)
1682 hit = !hit;
1683 if (opt->unmatch_name_only) {
1684 if (hit)
1685 return 0;
1686 goto next_line;
1688 if (hit) {
1689 count++;
1690 if (opt->status_only)
1691 return 1;
1692 if (opt->name_only) {
1693 show_name(opt, gs->name);
1694 return 1;
1696 if (opt->count)
1697 goto next_line;
1698 if (binary_match_only) {
1699 opt->output(opt, "Binary file ", 12);
1700 output_color(opt, gs->name, strlen(gs->name),
1701 opt->colors[GREP_COLOR_FILENAME]);
1702 opt->output(opt, " matches\n", 9);
1703 return 1;
1705 /* Hit at this line. If we haven't shown the
1706 * pre-context lines, we would need to show them.
1708 if (opt->pre_context || opt->funcbody)
1709 show_pre_context(opt, gs, bol, eol, lno);
1710 else if (opt->funcname)
1711 show_funcname_line(opt, gs, bol, lno);
1712 cno = opt->invert ? icol : col;
1713 if (cno < 0) {
1715 * A negative cno indicates that there was no
1716 * match on the line. We are thus inverted and
1717 * being asked to show all lines that _don't_
1718 * match a given expression. Therefore, set cno
1719 * to 0 to suggest the whole line matches.
1721 cno = 0;
1723 show_line(opt, bol, eol, gs->name, lno, cno + 1, ':');
1724 last_hit = lno;
1725 if (opt->funcbody)
1726 show_function = 1;
1727 goto next_line;
1729 if (show_function && (!peek_bol || peek_bol < bol)) {
1730 unsigned long peek_left = left;
1731 const char *peek_eol = eol;
1734 * Trailing empty lines are not interesting.
1735 * Peek past them to see if they belong to the
1736 * body of the current function.
1738 peek_bol = bol;
1739 while (is_empty_line(peek_bol, peek_eol)) {
1740 peek_bol = peek_eol + 1;
1741 peek_eol = end_of_line(peek_bol, &peek_left);
1744 if (match_funcname(opt, gs, peek_bol, peek_eol))
1745 show_function = 0;
1747 if (show_function ||
1748 (last_hit && lno <= last_hit + opt->post_context)) {
1749 /* If the last hit is within the post context,
1750 * we need to show this line.
1752 show_line(opt, bol, eol, gs->name, lno, col + 1, '-');
1755 next_line:
1756 bol = eol + 1;
1757 if (!left)
1758 break;
1759 left--;
1760 lno++;
1763 if (collect_hits)
1764 return 0;
1766 if (opt->status_only)
1767 return opt->unmatch_name_only;
1768 if (opt->unmatch_name_only) {
1769 /* We did not see any hit, so we want to show this */
1770 show_name(opt, gs->name);
1771 return 1;
1774 xdiff_clear_find_func(&xecfg);
1775 opt->priv = NULL;
1777 /* NEEDSWORK:
1778 * The real "grep -c foo *.c" gives many "bar.c:0" lines,
1779 * which feels mostly useless but sometimes useful. Maybe
1780 * make it another option? For now suppress them.
1782 if (opt->count && count) {
1783 char buf[32];
1784 if (opt->pathname) {
1785 output_color(opt, gs->name, strlen(gs->name),
1786 opt->colors[GREP_COLOR_FILENAME]);
1787 output_sep(opt, ':');
1789 xsnprintf(buf, sizeof(buf), "%u\n", count);
1790 opt->output(opt, buf, strlen(buf));
1791 return 1;
1793 return !!last_hit;
1796 static void clr_hit_marker(struct grep_expr *x)
1798 /* All-hit markers are meaningful only at the very top level
1799 * OR node.
1801 while (1) {
1802 x->hit = 0;
1803 if (x->node != GREP_NODE_OR)
1804 return;
1805 x->u.binary.left->hit = 0;
1806 x = x->u.binary.right;
1810 static int chk_hit_marker(struct grep_expr *x)
1812 /* Top level nodes have hit markers. See if they all are hits */
1813 while (1) {
1814 if (x->node != GREP_NODE_OR)
1815 return x->hit;
1816 if (!x->u.binary.left->hit)
1817 return 0;
1818 x = x->u.binary.right;
1822 int grep_source(struct grep_opt *opt, struct grep_source *gs)
1825 * we do not have to do the two-pass grep when we do not check
1826 * buffer-wide "all-match".
1828 if (!opt->all_match)
1829 return grep_source_1(opt, gs, 0);
1831 /* Otherwise the toplevel "or" terms hit a bit differently.
1832 * We first clear hit markers from them.
1834 clr_hit_marker(opt->pattern_expression);
1835 grep_source_1(opt, gs, 1);
1837 if (!chk_hit_marker(opt->pattern_expression))
1838 return 0;
1840 return grep_source_1(opt, gs, 0);
1843 static void grep_source_init_buf(struct grep_source *gs,
1844 const char *buf,
1845 unsigned long size)
1847 gs->type = GREP_SOURCE_BUF;
1848 gs->name = NULL;
1849 gs->path = NULL;
1850 gs->buf = buf;
1851 gs->size = size;
1852 gs->driver = NULL;
1853 gs->identifier = NULL;
1856 int grep_buffer(struct grep_opt *opt, const char *buf, unsigned long size)
1858 struct grep_source gs;
1859 int r;
1861 grep_source_init_buf(&gs, buf, size);
1863 r = grep_source(opt, &gs);
1865 grep_source_clear(&gs);
1866 return r;
1869 void grep_source_init_file(struct grep_source *gs, const char *name,
1870 const char *path)
1872 gs->type = GREP_SOURCE_FILE;
1873 gs->name = xstrdup_or_null(name);
1874 gs->path = xstrdup_or_null(path);
1875 gs->buf = NULL;
1876 gs->size = 0;
1877 gs->driver = NULL;
1878 gs->identifier = xstrdup(path);
1881 void grep_source_init_oid(struct grep_source *gs, const char *name,
1882 const char *path, const struct object_id *oid,
1883 struct repository *repo)
1885 gs->type = GREP_SOURCE_OID;
1886 gs->name = xstrdup_or_null(name);
1887 gs->path = xstrdup_or_null(path);
1888 gs->buf = NULL;
1889 gs->size = 0;
1890 gs->driver = NULL;
1891 gs->identifier = oiddup(oid);
1892 gs->repo = repo;
1895 void grep_source_clear(struct grep_source *gs)
1897 FREE_AND_NULL(gs->name);
1898 FREE_AND_NULL(gs->path);
1899 FREE_AND_NULL(gs->identifier);
1900 grep_source_clear_data(gs);
1903 void grep_source_clear_data(struct grep_source *gs)
1905 switch (gs->type) {
1906 case GREP_SOURCE_FILE:
1907 case GREP_SOURCE_OID:
1908 /* these types own the buffer */
1909 free((char *)gs->buf);
1910 gs->buf = NULL;
1911 gs->size = 0;
1912 break;
1913 case GREP_SOURCE_BUF:
1914 /* leave user-provided buf intact */
1915 break;
1919 static int grep_source_load_oid(struct grep_source *gs)
1921 enum object_type type;
1923 gs->buf = repo_read_object_file(gs->repo, gs->identifier, &type,
1924 &gs->size);
1925 if (!gs->buf)
1926 return error(_("'%s': unable to read %s"),
1927 gs->name,
1928 oid_to_hex(gs->identifier));
1929 return 0;
1932 static int grep_source_load_file(struct grep_source *gs)
1934 const char *filename = gs->identifier;
1935 struct stat st;
1936 char *data;
1937 size_t size;
1938 int i;
1940 if (lstat(filename, &st) < 0) {
1941 err_ret:
1942 if (errno != ENOENT)
1943 error_errno(_("failed to stat '%s'"), filename);
1944 return -1;
1946 if (!S_ISREG(st.st_mode))
1947 return -1;
1948 size = xsize_t(st.st_size);
1949 i = open(filename, O_RDONLY);
1950 if (i < 0)
1951 goto err_ret;
1952 data = xmallocz(size);
1953 if (st.st_size != read_in_full(i, data, size)) {
1954 error_errno(_("'%s': short read"), filename);
1955 close(i);
1956 free(data);
1957 return -1;
1959 close(i);
1961 gs->buf = data;
1962 gs->size = size;
1963 return 0;
1966 static int grep_source_load(struct grep_source *gs)
1968 if (gs->buf)
1969 return 0;
1971 switch (gs->type) {
1972 case GREP_SOURCE_FILE:
1973 return grep_source_load_file(gs);
1974 case GREP_SOURCE_OID:
1975 return grep_source_load_oid(gs);
1976 case GREP_SOURCE_BUF:
1977 return gs->buf ? 0 : -1;
1979 BUG("invalid grep_source type to load");
1982 void grep_source_load_driver(struct grep_source *gs,
1983 struct index_state *istate)
1985 if (gs->driver)
1986 return;
1988 grep_attr_lock();
1989 if (gs->path)
1990 gs->driver = userdiff_find_by_path(istate, gs->path);
1991 if (!gs->driver)
1992 gs->driver = userdiff_find_by_name("default");
1993 grep_attr_unlock();
1996 static int grep_source_is_binary(struct grep_source *gs,
1997 struct index_state *istate)
1999 grep_source_load_driver(gs, istate);
2000 if (gs->driver->binary != -1)
2001 return gs->driver->binary;
2003 if (!grep_source_load(gs))
2004 return buffer_is_binary(gs->buf, gs->size);
2006 return 0;