The third batch
[git.git] / grep.c
blob0d50598acda6f1c41951d1d656ea701114473239
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 struct grep_opt grep_defaults;
19 static const char *color_grep_slots[] = {
20 [GREP_COLOR_CONTEXT] = "context",
21 [GREP_COLOR_FILENAME] = "filename",
22 [GREP_COLOR_FUNCTION] = "function",
23 [GREP_COLOR_LINENO] = "lineNumber",
24 [GREP_COLOR_COLUMNNO] = "column",
25 [GREP_COLOR_MATCH_CONTEXT] = "matchContext",
26 [GREP_COLOR_MATCH_SELECTED] = "matchSelected",
27 [GREP_COLOR_SELECTED] = "selected",
28 [GREP_COLOR_SEP] = "separator",
31 static void std_output(struct grep_opt *opt, const void *buf, size_t size)
33 fwrite(buf, size, 1, stdout);
36 static void color_set(char *dst, const char *color_bytes)
38 xsnprintf(dst, COLOR_MAXLEN, "%s", color_bytes);
42 * Initialize the grep_defaults template with hardcoded defaults.
43 * We could let the compiler do this, but without C99 initializers
44 * the code gets unwieldy and unreadable, so...
46 void init_grep_defaults(struct repository *repo)
48 struct grep_opt *opt = &grep_defaults;
49 static int run_once;
51 if (run_once)
52 return;
53 run_once++;
55 memset(opt, 0, sizeof(*opt));
56 opt->repo = repo;
57 opt->relative = 1;
58 opt->pathname = 1;
59 opt->max_depth = -1;
60 opt->pattern_type_option = GREP_PATTERN_TYPE_UNSPECIFIED;
61 color_set(opt->colors[GREP_COLOR_CONTEXT], "");
62 color_set(opt->colors[GREP_COLOR_FILENAME], "");
63 color_set(opt->colors[GREP_COLOR_FUNCTION], "");
64 color_set(opt->colors[GREP_COLOR_LINENO], "");
65 color_set(opt->colors[GREP_COLOR_COLUMNNO], "");
66 color_set(opt->colors[GREP_COLOR_MATCH_CONTEXT], GIT_COLOR_BOLD_RED);
67 color_set(opt->colors[GREP_COLOR_MATCH_SELECTED], GIT_COLOR_BOLD_RED);
68 color_set(opt->colors[GREP_COLOR_SELECTED], "");
69 color_set(opt->colors[GREP_COLOR_SEP], GIT_COLOR_CYAN);
70 opt->only_matching = 0;
71 opt->color = -1;
72 opt->output = std_output;
75 static int parse_pattern_type_arg(const char *opt, const char *arg)
77 if (!strcmp(arg, "default"))
78 return GREP_PATTERN_TYPE_UNSPECIFIED;
79 else if (!strcmp(arg, "basic"))
80 return GREP_PATTERN_TYPE_BRE;
81 else if (!strcmp(arg, "extended"))
82 return GREP_PATTERN_TYPE_ERE;
83 else if (!strcmp(arg, "fixed"))
84 return GREP_PATTERN_TYPE_FIXED;
85 else if (!strcmp(arg, "perl"))
86 return GREP_PATTERN_TYPE_PCRE;
87 die("bad %s argument: %s", opt, arg);
90 define_list_config_array_extra(color_grep_slots, {"match"});
93 * Read the configuration file once and store it in
94 * the grep_defaults template.
96 int grep_config(const char *var, const char *value, void *cb)
98 struct grep_opt *opt = &grep_defaults;
99 const char *slot;
101 if (userdiff_config(var, value) < 0)
102 return -1;
104 if (!strcmp(var, "grep.extendedregexp")) {
105 opt->extended_regexp_option = git_config_bool(var, value);
106 return 0;
109 if (!strcmp(var, "grep.patterntype")) {
110 opt->pattern_type_option = parse_pattern_type_arg(var, value);
111 return 0;
114 if (!strcmp(var, "grep.linenumber")) {
115 opt->linenum = git_config_bool(var, value);
116 return 0;
118 if (!strcmp(var, "grep.column")) {
119 opt->columnnum = git_config_bool(var, value);
120 return 0;
123 if (!strcmp(var, "grep.fullname")) {
124 opt->relative = !git_config_bool(var, value);
125 return 0;
128 if (!strcmp(var, "color.grep"))
129 opt->color = git_config_colorbool(var, value);
130 if (!strcmp(var, "color.grep.match")) {
131 if (grep_config("color.grep.matchcontext", value, cb) < 0)
132 return -1;
133 if (grep_config("color.grep.matchselected", value, cb) < 0)
134 return -1;
135 } else if (skip_prefix(var, "color.grep.", &slot)) {
136 int i = LOOKUP_CONFIG(color_grep_slots, slot);
137 char *color;
139 if (i < 0)
140 return -1;
141 color = opt->colors[i];
142 if (!value)
143 return config_error_nonbool(var);
144 return color_parse(value, color);
146 return 0;
150 * Initialize one instance of grep_opt and copy the
151 * default values from the template we read the configuration
152 * information in an earlier call to git_config(grep_config).
154 void grep_init(struct grep_opt *opt, struct repository *repo, const char *prefix)
156 struct grep_opt *def = &grep_defaults;
157 int i;
159 memset(opt, 0, sizeof(*opt));
160 opt->repo = repo;
161 opt->prefix = prefix;
162 opt->prefix_length = (prefix && *prefix) ? strlen(prefix) : 0;
163 opt->pattern_tail = &opt->pattern_list;
164 opt->header_tail = &opt->header_list;
166 opt->only_matching = def->only_matching;
167 opt->color = def->color;
168 opt->extended_regexp_option = def->extended_regexp_option;
169 opt->pattern_type_option = def->pattern_type_option;
170 opt->linenum = def->linenum;
171 opt->columnnum = def->columnnum;
172 opt->max_depth = def->max_depth;
173 opt->pathname = def->pathname;
174 opt->relative = def->relative;
175 opt->output = def->output;
177 for (i = 0; i < NR_GREP_COLORS; i++)
178 color_set(opt->colors[i], def->colors[i]);
181 static void grep_set_pattern_type_option(enum grep_pattern_type pattern_type, struct grep_opt *opt)
184 * When committing to the pattern type by setting the relevant
185 * fields in grep_opt it's generally not necessary to zero out
186 * the fields we're not choosing, since they won't have been
187 * set by anything. The extended_regexp_option field is the
188 * only exception to this.
190 * This is because in the process of parsing grep.patternType
191 * & grep.extendedRegexp we set opt->pattern_type_option and
192 * opt->extended_regexp_option, respectively. We then
193 * internally use opt->extended_regexp_option to see if we're
194 * compiling an ERE. It must be unset if that's not actually
195 * the case.
197 if (pattern_type != GREP_PATTERN_TYPE_ERE &&
198 opt->extended_regexp_option)
199 opt->extended_regexp_option = 0;
201 switch (pattern_type) {
202 case GREP_PATTERN_TYPE_UNSPECIFIED:
203 /* fall through */
205 case GREP_PATTERN_TYPE_BRE:
206 break;
208 case GREP_PATTERN_TYPE_ERE:
209 opt->extended_regexp_option = 1;
210 break;
212 case GREP_PATTERN_TYPE_FIXED:
213 opt->fixed = 1;
214 break;
216 case GREP_PATTERN_TYPE_PCRE:
217 #ifdef USE_LIBPCRE2
218 opt->pcre2 = 1;
219 #else
221 * It's important that pcre1 always be assigned to
222 * even when there's no USE_LIBPCRE* defined. We still
223 * call the PCRE stub function, it just dies with
224 * "cannot use Perl-compatible regexes[...]".
226 opt->pcre1 = 1;
227 #endif
228 break;
232 void grep_commit_pattern_type(enum grep_pattern_type pattern_type, struct grep_opt *opt)
234 if (pattern_type != GREP_PATTERN_TYPE_UNSPECIFIED)
235 grep_set_pattern_type_option(pattern_type, opt);
236 else if (opt->pattern_type_option != GREP_PATTERN_TYPE_UNSPECIFIED)
237 grep_set_pattern_type_option(opt->pattern_type_option, opt);
238 else if (opt->extended_regexp_option)
240 * This branch *must* happen after setting from the
241 * opt->pattern_type_option above, we don't want
242 * grep.extendedRegexp to override grep.patternType!
244 grep_set_pattern_type_option(GREP_PATTERN_TYPE_ERE, opt);
247 static struct grep_pat *create_grep_pat(const char *pat, size_t patlen,
248 const char *origin, int no,
249 enum grep_pat_token t,
250 enum grep_header_field field)
252 struct grep_pat *p = xcalloc(1, sizeof(*p));
253 p->pattern = xmemdupz(pat, patlen);
254 p->patternlen = patlen;
255 p->origin = origin;
256 p->no = no;
257 p->token = t;
258 p->field = field;
259 return p;
262 static void do_append_grep_pat(struct grep_pat ***tail, struct grep_pat *p)
264 **tail = p;
265 *tail = &p->next;
266 p->next = NULL;
268 switch (p->token) {
269 case GREP_PATTERN: /* atom */
270 case GREP_PATTERN_HEAD:
271 case GREP_PATTERN_BODY:
272 for (;;) {
273 struct grep_pat *new_pat;
274 size_t len = 0;
275 char *cp = p->pattern + p->patternlen, *nl = NULL;
276 while (++len <= p->patternlen) {
277 if (*(--cp) == '\n') {
278 nl = cp;
279 break;
282 if (!nl)
283 break;
284 new_pat = create_grep_pat(nl + 1, len - 1, p->origin,
285 p->no, p->token, p->field);
286 new_pat->next = p->next;
287 if (!p->next)
288 *tail = &new_pat->next;
289 p->next = new_pat;
290 *nl = '\0';
291 p->patternlen -= len;
293 break;
294 default:
295 break;
299 void append_header_grep_pattern(struct grep_opt *opt,
300 enum grep_header_field field, const char *pat)
302 struct grep_pat *p = create_grep_pat(pat, strlen(pat), "header", 0,
303 GREP_PATTERN_HEAD, field);
304 if (field == GREP_HEADER_REFLOG)
305 opt->use_reflog_filter = 1;
306 do_append_grep_pat(&opt->header_tail, p);
309 void append_grep_pattern(struct grep_opt *opt, const char *pat,
310 const char *origin, int no, enum grep_pat_token t)
312 append_grep_pat(opt, pat, strlen(pat), origin, no, t);
315 void append_grep_pat(struct grep_opt *opt, const char *pat, size_t patlen,
316 const char *origin, int no, enum grep_pat_token t)
318 struct grep_pat *p = create_grep_pat(pat, patlen, origin, no, t, 0);
319 do_append_grep_pat(&opt->pattern_tail, p);
322 struct grep_opt *grep_opt_dup(const struct grep_opt *opt)
324 struct grep_pat *pat;
325 struct grep_opt *ret = xmalloc(sizeof(struct grep_opt));
326 *ret = *opt;
328 ret->pattern_list = NULL;
329 ret->pattern_tail = &ret->pattern_list;
331 for(pat = opt->pattern_list; pat != NULL; pat = pat->next)
333 if(pat->token == GREP_PATTERN_HEAD)
334 append_header_grep_pattern(ret, pat->field,
335 pat->pattern);
336 else
337 append_grep_pat(ret, pat->pattern, pat->patternlen,
338 pat->origin, pat->no, pat->token);
341 return ret;
344 static NORETURN void compile_regexp_failed(const struct grep_pat *p,
345 const char *error)
347 char where[1024];
349 if (p->no)
350 xsnprintf(where, sizeof(where), "In '%s' at %d, ", p->origin, p->no);
351 else if (p->origin)
352 xsnprintf(where, sizeof(where), "%s, ", p->origin);
353 else
354 where[0] = 0;
356 die("%s'%s': %s", where, p->pattern, error);
359 static int is_fixed(const char *s, size_t len)
361 size_t i;
363 for (i = 0; i < len; i++) {
364 if (is_regex_special(s[i]))
365 return 0;
368 return 1;
371 static int has_null(const char *s, size_t len)
374 * regcomp cannot accept patterns with NULs so when using it
375 * we consider any pattern containing a NUL fixed.
377 if (memchr(s, 0, len))
378 return 1;
380 return 0;
383 #ifdef USE_LIBPCRE1
384 static void compile_pcre1_regexp(struct grep_pat *p, const struct grep_opt *opt)
386 const char *error;
387 int erroffset;
388 int options = PCRE_MULTILINE;
390 if (opt->ignore_case) {
391 if (has_non_ascii(p->pattern))
392 p->pcre1_tables = pcre_maketables();
393 options |= PCRE_CASELESS;
395 if (is_utf8_locale() && has_non_ascii(p->pattern))
396 options |= PCRE_UTF8;
398 p->pcre1_regexp = pcre_compile(p->pattern, options, &error, &erroffset,
399 p->pcre1_tables);
400 if (!p->pcre1_regexp)
401 compile_regexp_failed(p, error);
403 p->pcre1_extra_info = pcre_study(p->pcre1_regexp, GIT_PCRE_STUDY_JIT_COMPILE, &error);
404 if (!p->pcre1_extra_info && error)
405 die("%s", error);
407 #ifdef GIT_PCRE1_USE_JIT
408 pcre_config(PCRE_CONFIG_JIT, &p->pcre1_jit_on);
409 if (p->pcre1_jit_on == 1) {
410 p->pcre1_jit_stack = pcre_jit_stack_alloc(1, 1024 * 1024);
411 if (!p->pcre1_jit_stack)
412 die("Couldn't allocate PCRE JIT stack");
413 pcre_assign_jit_stack(p->pcre1_extra_info, NULL, p->pcre1_jit_stack);
414 } else if (p->pcre1_jit_on != 0) {
415 BUG("The pcre1_jit_on variable should be 0 or 1, not %d",
416 p->pcre1_jit_on);
418 #endif
421 static int pcre1match(struct grep_pat *p, const char *line, const char *eol,
422 regmatch_t *match, int eflags)
424 int ovector[30], ret, flags = 0;
426 if (eflags & REG_NOTBOL)
427 flags |= PCRE_NOTBOL;
429 #ifdef GIT_PCRE1_USE_JIT
430 if (p->pcre1_jit_on) {
431 ret = pcre_jit_exec(p->pcre1_regexp, p->pcre1_extra_info, line,
432 eol - line, 0, flags, ovector,
433 ARRAY_SIZE(ovector), p->pcre1_jit_stack);
434 } else
435 #endif
437 ret = pcre_exec(p->pcre1_regexp, p->pcre1_extra_info, line,
438 eol - line, 0, flags, ovector,
439 ARRAY_SIZE(ovector));
442 if (ret < 0 && ret != PCRE_ERROR_NOMATCH)
443 die("pcre_exec failed with error code %d", ret);
444 if (ret > 0) {
445 ret = 0;
446 match->rm_so = ovector[0];
447 match->rm_eo = ovector[1];
450 return ret;
453 static void free_pcre1_regexp(struct grep_pat *p)
455 pcre_free(p->pcre1_regexp);
456 #ifdef GIT_PCRE1_USE_JIT
457 if (p->pcre1_jit_on) {
458 pcre_free_study(p->pcre1_extra_info);
459 pcre_jit_stack_free(p->pcre1_jit_stack);
460 } else
461 #endif
463 pcre_free(p->pcre1_extra_info);
465 pcre_free((void *)p->pcre1_tables);
467 #else /* !USE_LIBPCRE1 */
468 static void compile_pcre1_regexp(struct grep_pat *p, const struct grep_opt *opt)
470 die("cannot use Perl-compatible regexes when not compiled with USE_LIBPCRE");
473 static int pcre1match(struct grep_pat *p, const char *line, const char *eol,
474 regmatch_t *match, int eflags)
476 return 1;
479 static void free_pcre1_regexp(struct grep_pat *p)
482 #endif /* !USE_LIBPCRE1 */
484 #ifdef USE_LIBPCRE2
485 static void compile_pcre2_pattern(struct grep_pat *p, const struct grep_opt *opt)
487 int error;
488 PCRE2_UCHAR errbuf[256];
489 PCRE2_SIZE erroffset;
490 int options = PCRE2_MULTILINE;
491 const uint8_t *character_tables = NULL;
492 int jitret;
493 int patinforet;
494 size_t jitsizearg;
496 assert(opt->pcre2);
498 p->pcre2_compile_context = NULL;
500 if (opt->ignore_case) {
501 if (has_non_ascii(p->pattern)) {
502 character_tables = pcre2_maketables(NULL);
503 p->pcre2_compile_context = pcre2_compile_context_create(NULL);
504 pcre2_set_character_tables(p->pcre2_compile_context, character_tables);
506 options |= PCRE2_CASELESS;
508 if (is_utf8_locale() && has_non_ascii(p->pattern))
509 options |= PCRE2_UTF;
511 p->pcre2_pattern = pcre2_compile((PCRE2_SPTR)p->pattern,
512 p->patternlen, options, &error, &erroffset,
513 p->pcre2_compile_context);
515 if (p->pcre2_pattern) {
516 p->pcre2_match_data = pcre2_match_data_create_from_pattern(p->pcre2_pattern, NULL);
517 if (!p->pcre2_match_data)
518 die("Couldn't allocate PCRE2 match data");
519 } else {
520 pcre2_get_error_message(error, errbuf, sizeof(errbuf));
521 compile_regexp_failed(p, (const char *)&errbuf);
524 pcre2_config(PCRE2_CONFIG_JIT, &p->pcre2_jit_on);
525 if (p->pcre2_jit_on == 1) {
526 jitret = pcre2_jit_compile(p->pcre2_pattern, PCRE2_JIT_COMPLETE);
527 if (jitret)
528 die("Couldn't JIT the PCRE2 pattern '%s', got '%d'\n", p->pattern, jitret);
531 * The pcre2_config(PCRE2_CONFIG_JIT, ...) call just
532 * tells us whether the library itself supports JIT,
533 * but to see whether we're going to be actually using
534 * JIT we need to extract PCRE2_INFO_JITSIZE from the
535 * pattern *after* we do pcre2_jit_compile() above.
537 * This is because if the pattern contains the
538 * (*NO_JIT) verb (see pcre2syntax(3))
539 * pcre2_jit_compile() will exit early with 0. If we
540 * then proceed to call pcre2_jit_match() further down
541 * the line instead of pcre2_match() we'll either
542 * segfault (pre PCRE 10.31) or run into a fatal error
543 * (post PCRE2 10.31)
545 patinforet = pcre2_pattern_info(p->pcre2_pattern, PCRE2_INFO_JITSIZE, &jitsizearg);
546 if (patinforet)
547 BUG("pcre2_pattern_info() failed: %d", patinforet);
548 if (jitsizearg == 0) {
549 p->pcre2_jit_on = 0;
550 return;
553 p->pcre2_jit_stack = pcre2_jit_stack_create(1, 1024 * 1024, NULL);
554 if (!p->pcre2_jit_stack)
555 die("Couldn't allocate PCRE2 JIT stack");
556 p->pcre2_match_context = pcre2_match_context_create(NULL);
557 if (!p->pcre2_match_context)
558 die("Couldn't allocate PCRE2 match context");
559 pcre2_jit_stack_assign(p->pcre2_match_context, NULL, p->pcre2_jit_stack);
560 } else if (p->pcre2_jit_on != 0) {
561 BUG("The pcre2_jit_on variable should be 0 or 1, not %d",
562 p->pcre1_jit_on);
566 static int pcre2match(struct grep_pat *p, const char *line, const char *eol,
567 regmatch_t *match, int eflags)
569 int ret, flags = 0;
570 PCRE2_SIZE *ovector;
571 PCRE2_UCHAR errbuf[256];
573 if (eflags & REG_NOTBOL)
574 flags |= PCRE2_NOTBOL;
576 if (p->pcre2_jit_on)
577 ret = pcre2_jit_match(p->pcre2_pattern, (unsigned char *)line,
578 eol - line, 0, flags, p->pcre2_match_data,
579 NULL);
580 else
581 ret = pcre2_match(p->pcre2_pattern, (unsigned char *)line,
582 eol - line, 0, flags, p->pcre2_match_data,
583 NULL);
585 if (ret < 0 && ret != PCRE2_ERROR_NOMATCH) {
586 pcre2_get_error_message(ret, errbuf, sizeof(errbuf));
587 die("%s failed with error code %d: %s",
588 (p->pcre2_jit_on ? "pcre2_jit_match" : "pcre2_match"), ret,
589 errbuf);
591 if (ret > 0) {
592 ovector = pcre2_get_ovector_pointer(p->pcre2_match_data);
593 ret = 0;
594 match->rm_so = (int)ovector[0];
595 match->rm_eo = (int)ovector[1];
598 return ret;
601 static void free_pcre2_pattern(struct grep_pat *p)
603 pcre2_compile_context_free(p->pcre2_compile_context);
604 pcre2_code_free(p->pcre2_pattern);
605 pcre2_match_data_free(p->pcre2_match_data);
606 pcre2_jit_stack_free(p->pcre2_jit_stack);
607 pcre2_match_context_free(p->pcre2_match_context);
609 #else /* !USE_LIBPCRE2 */
610 static void compile_pcre2_pattern(struct grep_pat *p, const struct grep_opt *opt)
613 * Unreachable until USE_LIBPCRE2 becomes synonymous with
614 * USE_LIBPCRE. See the sibling comment in
615 * grep_set_pattern_type_option().
617 die("cannot use Perl-compatible regexes when not compiled with USE_LIBPCRE");
620 static int pcre2match(struct grep_pat *p, const char *line, const char *eol,
621 regmatch_t *match, int eflags)
623 return 1;
626 static void free_pcre2_pattern(struct grep_pat *p)
629 #endif /* !USE_LIBPCRE2 */
631 static void compile_fixed_regexp(struct grep_pat *p, struct grep_opt *opt)
633 struct strbuf sb = STRBUF_INIT;
634 int err;
635 int regflags = 0;
637 basic_regex_quote_buf(&sb, p->pattern);
638 if (opt->ignore_case)
639 regflags |= REG_ICASE;
640 err = regcomp(&p->regexp, sb.buf, regflags);
641 if (opt->debug)
642 fprintf(stderr, "fixed %s\n", sb.buf);
643 strbuf_release(&sb);
644 if (err) {
645 char errbuf[1024];
646 regerror(err, &p->regexp, errbuf, sizeof(errbuf));
647 compile_regexp_failed(p, errbuf);
651 static void compile_regexp(struct grep_pat *p, struct grep_opt *opt)
653 int ascii_only;
654 int err;
655 int regflags = REG_NEWLINE;
657 p->word_regexp = opt->word_regexp;
658 p->ignore_case = opt->ignore_case;
659 ascii_only = !has_non_ascii(p->pattern);
662 * Even when -F (fixed) asks us to do a non-regexp search, we
663 * may not be able to correctly case-fold when -i
664 * (ignore-case) is asked (in which case, we'll synthesize a
665 * regexp to match the pattern that matches regexp special
666 * characters literally, while ignoring case differences). On
667 * the other hand, even without -F, if the pattern does not
668 * have any regexp special characters and there is no need for
669 * case-folding search, we can internally turn it into a
670 * simple string match using kws. p->fixed tells us if we
671 * want to use kws.
673 if (opt->fixed ||
674 has_null(p->pattern, p->patternlen) ||
675 is_fixed(p->pattern, p->patternlen))
676 p->fixed = !p->ignore_case || ascii_only;
678 if (p->fixed) {
679 p->kws = kwsalloc(p->ignore_case ? tolower_trans_tbl : NULL);
680 kwsincr(p->kws, p->pattern, p->patternlen);
681 kwsprep(p->kws);
682 return;
683 } else if (opt->fixed) {
685 * We come here when the pattern has the non-ascii
686 * characters we cannot case-fold, and asked to
687 * ignore-case.
689 compile_fixed_regexp(p, opt);
690 return;
693 if (opt->pcre2) {
694 compile_pcre2_pattern(p, opt);
695 return;
698 if (opt->pcre1) {
699 compile_pcre1_regexp(p, opt);
700 return;
703 if (p->ignore_case)
704 regflags |= REG_ICASE;
705 if (opt->extended_regexp_option)
706 regflags |= REG_EXTENDED;
707 err = regcomp(&p->regexp, p->pattern, regflags);
708 if (err) {
709 char errbuf[1024];
710 regerror(err, &p->regexp, errbuf, 1024);
711 compile_regexp_failed(p, errbuf);
715 static struct grep_expr *compile_pattern_or(struct grep_pat **);
716 static struct grep_expr *compile_pattern_atom(struct grep_pat **list)
718 struct grep_pat *p;
719 struct grep_expr *x;
721 p = *list;
722 if (!p)
723 return NULL;
724 switch (p->token) {
725 case GREP_PATTERN: /* atom */
726 case GREP_PATTERN_HEAD:
727 case GREP_PATTERN_BODY:
728 x = xcalloc(1, sizeof (struct grep_expr));
729 x->node = GREP_NODE_ATOM;
730 x->u.atom = p;
731 *list = p->next;
732 return x;
733 case GREP_OPEN_PAREN:
734 *list = p->next;
735 x = compile_pattern_or(list);
736 if (!*list || (*list)->token != GREP_CLOSE_PAREN)
737 die("unmatched parenthesis");
738 *list = (*list)->next;
739 return x;
740 default:
741 return NULL;
745 static struct grep_expr *compile_pattern_not(struct grep_pat **list)
747 struct grep_pat *p;
748 struct grep_expr *x;
750 p = *list;
751 if (!p)
752 return NULL;
753 switch (p->token) {
754 case GREP_NOT:
755 if (!p->next)
756 die("--not not followed by pattern expression");
757 *list = p->next;
758 x = xcalloc(1, sizeof (struct grep_expr));
759 x->node = GREP_NODE_NOT;
760 x->u.unary = compile_pattern_not(list);
761 if (!x->u.unary)
762 die("--not followed by non pattern expression");
763 return x;
764 default:
765 return compile_pattern_atom(list);
769 static struct grep_expr *compile_pattern_and(struct grep_pat **list)
771 struct grep_pat *p;
772 struct grep_expr *x, *y, *z;
774 x = compile_pattern_not(list);
775 p = *list;
776 if (p && p->token == GREP_AND) {
777 if (!p->next)
778 die("--and not followed by pattern expression");
779 *list = p->next;
780 y = compile_pattern_and(list);
781 if (!y)
782 die("--and not followed by pattern expression");
783 z = xcalloc(1, sizeof (struct grep_expr));
784 z->node = GREP_NODE_AND;
785 z->u.binary.left = x;
786 z->u.binary.right = y;
787 return z;
789 return x;
792 static struct grep_expr *compile_pattern_or(struct grep_pat **list)
794 struct grep_pat *p;
795 struct grep_expr *x, *y, *z;
797 x = compile_pattern_and(list);
798 p = *list;
799 if (x && p && p->token != GREP_CLOSE_PAREN) {
800 y = compile_pattern_or(list);
801 if (!y)
802 die("not a pattern expression %s", p->pattern);
803 z = xcalloc(1, sizeof (struct grep_expr));
804 z->node = GREP_NODE_OR;
805 z->u.binary.left = x;
806 z->u.binary.right = y;
807 return z;
809 return x;
812 static struct grep_expr *compile_pattern_expr(struct grep_pat **list)
814 return compile_pattern_or(list);
817 static void indent(int in)
819 while (in-- > 0)
820 fputc(' ', stderr);
823 static void dump_grep_pat(struct grep_pat *p)
825 switch (p->token) {
826 case GREP_AND: fprintf(stderr, "*and*"); break;
827 case GREP_OPEN_PAREN: fprintf(stderr, "*(*"); break;
828 case GREP_CLOSE_PAREN: fprintf(stderr, "*)*"); break;
829 case GREP_NOT: fprintf(stderr, "*not*"); break;
830 case GREP_OR: fprintf(stderr, "*or*"); break;
832 case GREP_PATTERN: fprintf(stderr, "pattern"); break;
833 case GREP_PATTERN_HEAD: fprintf(stderr, "pattern_head"); break;
834 case GREP_PATTERN_BODY: fprintf(stderr, "pattern_body"); break;
837 switch (p->token) {
838 default: break;
839 case GREP_PATTERN_HEAD:
840 fprintf(stderr, "<head %d>", p->field); break;
841 case GREP_PATTERN_BODY:
842 fprintf(stderr, "<body>"); break;
844 switch (p->token) {
845 default: break;
846 case GREP_PATTERN_HEAD:
847 case GREP_PATTERN_BODY:
848 case GREP_PATTERN:
849 fprintf(stderr, "%.*s", (int)p->patternlen, p->pattern);
850 break;
852 fputc('\n', stderr);
855 static void dump_grep_expression_1(struct grep_expr *x, int in)
857 indent(in);
858 switch (x->node) {
859 case GREP_NODE_TRUE:
860 fprintf(stderr, "true\n");
861 break;
862 case GREP_NODE_ATOM:
863 dump_grep_pat(x->u.atom);
864 break;
865 case GREP_NODE_NOT:
866 fprintf(stderr, "(not\n");
867 dump_grep_expression_1(x->u.unary, in+1);
868 indent(in);
869 fprintf(stderr, ")\n");
870 break;
871 case GREP_NODE_AND:
872 fprintf(stderr, "(and\n");
873 dump_grep_expression_1(x->u.binary.left, in+1);
874 dump_grep_expression_1(x->u.binary.right, in+1);
875 indent(in);
876 fprintf(stderr, ")\n");
877 break;
878 case GREP_NODE_OR:
879 fprintf(stderr, "(or\n");
880 dump_grep_expression_1(x->u.binary.left, in+1);
881 dump_grep_expression_1(x->u.binary.right, in+1);
882 indent(in);
883 fprintf(stderr, ")\n");
884 break;
888 static void dump_grep_expression(struct grep_opt *opt)
890 struct grep_expr *x = opt->pattern_expression;
892 if (opt->all_match)
893 fprintf(stderr, "[all-match]\n");
894 dump_grep_expression_1(x, 0);
895 fflush(NULL);
898 static struct grep_expr *grep_true_expr(void)
900 struct grep_expr *z = xcalloc(1, sizeof(*z));
901 z->node = GREP_NODE_TRUE;
902 return z;
905 static struct grep_expr *grep_or_expr(struct grep_expr *left, struct grep_expr *right)
907 struct grep_expr *z = xcalloc(1, sizeof(*z));
908 z->node = GREP_NODE_OR;
909 z->u.binary.left = left;
910 z->u.binary.right = right;
911 return z;
914 static struct grep_expr *prep_header_patterns(struct grep_opt *opt)
916 struct grep_pat *p;
917 struct grep_expr *header_expr;
918 struct grep_expr *(header_group[GREP_HEADER_FIELD_MAX]);
919 enum grep_header_field fld;
921 if (!opt->header_list)
922 return NULL;
924 for (p = opt->header_list; p; p = p->next) {
925 if (p->token != GREP_PATTERN_HEAD)
926 BUG("a non-header pattern in grep header list.");
927 if (p->field < GREP_HEADER_FIELD_MIN ||
928 GREP_HEADER_FIELD_MAX <= p->field)
929 BUG("unknown header field %d", p->field);
930 compile_regexp(p, opt);
933 for (fld = 0; fld < GREP_HEADER_FIELD_MAX; fld++)
934 header_group[fld] = NULL;
936 for (p = opt->header_list; p; p = p->next) {
937 struct grep_expr *h;
938 struct grep_pat *pp = p;
940 h = compile_pattern_atom(&pp);
941 if (!h || pp != p->next)
942 BUG("malformed header expr");
943 if (!header_group[p->field]) {
944 header_group[p->field] = h;
945 continue;
947 header_group[p->field] = grep_or_expr(h, header_group[p->field]);
950 header_expr = NULL;
952 for (fld = 0; fld < GREP_HEADER_FIELD_MAX; fld++) {
953 if (!header_group[fld])
954 continue;
955 if (!header_expr)
956 header_expr = grep_true_expr();
957 header_expr = grep_or_expr(header_group[fld], header_expr);
959 return header_expr;
962 static struct grep_expr *grep_splice_or(struct grep_expr *x, struct grep_expr *y)
964 struct grep_expr *z = x;
966 while (x) {
967 assert(x->node == GREP_NODE_OR);
968 if (x->u.binary.right &&
969 x->u.binary.right->node == GREP_NODE_TRUE) {
970 x->u.binary.right = y;
971 break;
973 x = x->u.binary.right;
975 return z;
978 static void compile_grep_patterns_real(struct grep_opt *opt)
980 struct grep_pat *p;
981 struct grep_expr *header_expr = prep_header_patterns(opt);
983 for (p = opt->pattern_list; p; p = p->next) {
984 switch (p->token) {
985 case GREP_PATTERN: /* atom */
986 case GREP_PATTERN_HEAD:
987 case GREP_PATTERN_BODY:
988 compile_regexp(p, opt);
989 break;
990 default:
991 opt->extended = 1;
992 break;
996 if (opt->all_match || header_expr)
997 opt->extended = 1;
998 else if (!opt->extended && !opt->debug)
999 return;
1001 p = opt->pattern_list;
1002 if (p)
1003 opt->pattern_expression = compile_pattern_expr(&p);
1004 if (p)
1005 die("incomplete pattern expression: %s", p->pattern);
1007 if (!header_expr)
1008 return;
1010 if (!opt->pattern_expression)
1011 opt->pattern_expression = header_expr;
1012 else if (opt->all_match)
1013 opt->pattern_expression = grep_splice_or(header_expr,
1014 opt->pattern_expression);
1015 else
1016 opt->pattern_expression = grep_or_expr(opt->pattern_expression,
1017 header_expr);
1018 opt->all_match = 1;
1021 void compile_grep_patterns(struct grep_opt *opt)
1023 compile_grep_patterns_real(opt);
1024 if (opt->debug)
1025 dump_grep_expression(opt);
1028 static void free_pattern_expr(struct grep_expr *x)
1030 switch (x->node) {
1031 case GREP_NODE_TRUE:
1032 case GREP_NODE_ATOM:
1033 break;
1034 case GREP_NODE_NOT:
1035 free_pattern_expr(x->u.unary);
1036 break;
1037 case GREP_NODE_AND:
1038 case GREP_NODE_OR:
1039 free_pattern_expr(x->u.binary.left);
1040 free_pattern_expr(x->u.binary.right);
1041 break;
1043 free(x);
1046 void free_grep_patterns(struct grep_opt *opt)
1048 struct grep_pat *p, *n;
1050 for (p = opt->pattern_list; p; p = n) {
1051 n = p->next;
1052 switch (p->token) {
1053 case GREP_PATTERN: /* atom */
1054 case GREP_PATTERN_HEAD:
1055 case GREP_PATTERN_BODY:
1056 if (p->kws)
1057 kwsfree(p->kws);
1058 else if (p->pcre1_regexp)
1059 free_pcre1_regexp(p);
1060 else if (p->pcre2_pattern)
1061 free_pcre2_pattern(p);
1062 else
1063 regfree(&p->regexp);
1064 free(p->pattern);
1065 break;
1066 default:
1067 break;
1069 free(p);
1072 if (!opt->extended)
1073 return;
1074 free_pattern_expr(opt->pattern_expression);
1077 static char *end_of_line(char *cp, unsigned long *left)
1079 unsigned long l = *left;
1080 while (l && *cp != '\n') {
1081 l--;
1082 cp++;
1084 *left = l;
1085 return cp;
1088 static int word_char(char ch)
1090 return isalnum(ch) || ch == '_';
1093 static void output_color(struct grep_opt *opt, const void *data, size_t size,
1094 const char *color)
1096 if (want_color(opt->color) && color && color[0]) {
1097 opt->output(opt, color, strlen(color));
1098 opt->output(opt, data, size);
1099 opt->output(opt, GIT_COLOR_RESET, strlen(GIT_COLOR_RESET));
1100 } else
1101 opt->output(opt, data, size);
1104 static void output_sep(struct grep_opt *opt, char sign)
1106 if (opt->null_following_name)
1107 opt->output(opt, "\0", 1);
1108 else
1109 output_color(opt, &sign, 1, opt->colors[GREP_COLOR_SEP]);
1112 static void show_name(struct grep_opt *opt, const char *name)
1114 output_color(opt, name, strlen(name), opt->colors[GREP_COLOR_FILENAME]);
1115 opt->output(opt, opt->null_following_name ? "\0" : "\n", 1);
1118 static int fixmatch(struct grep_pat *p, char *line, char *eol,
1119 regmatch_t *match)
1121 struct kwsmatch kwsm;
1122 size_t offset = kwsexec(p->kws, line, eol - line, &kwsm);
1123 if (offset == -1) {
1124 match->rm_so = match->rm_eo = -1;
1125 return REG_NOMATCH;
1126 } else {
1127 match->rm_so = offset;
1128 match->rm_eo = match->rm_so + kwsm.size[0];
1129 return 0;
1133 static int patmatch(struct grep_pat *p, char *line, char *eol,
1134 regmatch_t *match, int eflags)
1136 int hit;
1138 if (p->fixed)
1139 hit = !fixmatch(p, line, eol, match);
1140 else if (p->pcre1_regexp)
1141 hit = !pcre1match(p, line, eol, match, eflags);
1142 else if (p->pcre2_pattern)
1143 hit = !pcre2match(p, line, eol, match, eflags);
1144 else
1145 hit = !regexec_buf(&p->regexp, line, eol - line, 1, match,
1146 eflags);
1148 return hit;
1151 static int strip_timestamp(char *bol, char **eol_p)
1153 char *eol = *eol_p;
1154 int ch;
1156 while (bol < --eol) {
1157 if (*eol != '>')
1158 continue;
1159 *eol_p = ++eol;
1160 ch = *eol;
1161 *eol = '\0';
1162 return ch;
1164 return 0;
1167 static struct {
1168 const char *field;
1169 size_t len;
1170 } header_field[] = {
1171 { "author ", 7 },
1172 { "committer ", 10 },
1173 { "reflog ", 7 },
1176 static int match_one_pattern(struct grep_pat *p, char *bol, char *eol,
1177 enum grep_context ctx,
1178 regmatch_t *pmatch, int eflags)
1180 int hit = 0;
1181 int saved_ch = 0;
1182 const char *start = bol;
1184 if ((p->token != GREP_PATTERN) &&
1185 ((p->token == GREP_PATTERN_HEAD) != (ctx == GREP_CONTEXT_HEAD)))
1186 return 0;
1188 if (p->token == GREP_PATTERN_HEAD) {
1189 const char *field;
1190 size_t len;
1191 assert(p->field < ARRAY_SIZE(header_field));
1192 field = header_field[p->field].field;
1193 len = header_field[p->field].len;
1194 if (strncmp(bol, field, len))
1195 return 0;
1196 bol += len;
1197 switch (p->field) {
1198 case GREP_HEADER_AUTHOR:
1199 case GREP_HEADER_COMMITTER:
1200 saved_ch = strip_timestamp(bol, &eol);
1201 break;
1202 default:
1203 break;
1207 again:
1208 hit = patmatch(p, bol, eol, pmatch, eflags);
1210 if (hit && p->word_regexp) {
1211 if ((pmatch[0].rm_so < 0) ||
1212 (eol - bol) < pmatch[0].rm_so ||
1213 (pmatch[0].rm_eo < 0) ||
1214 (eol - bol) < pmatch[0].rm_eo)
1215 die("regexp returned nonsense");
1217 /* Match beginning must be either beginning of the
1218 * line, or at word boundary (i.e. the last char must
1219 * not be a word char). Similarly, match end must be
1220 * either end of the line, or at word boundary
1221 * (i.e. the next char must not be a word char).
1223 if ( ((pmatch[0].rm_so == 0) ||
1224 !word_char(bol[pmatch[0].rm_so-1])) &&
1225 ((pmatch[0].rm_eo == (eol-bol)) ||
1226 !word_char(bol[pmatch[0].rm_eo])) )
1228 else
1229 hit = 0;
1231 /* Words consist of at least one character. */
1232 if (pmatch->rm_so == pmatch->rm_eo)
1233 hit = 0;
1235 if (!hit && pmatch[0].rm_so + bol + 1 < eol) {
1236 /* There could be more than one match on the
1237 * line, and the first match might not be
1238 * strict word match. But later ones could be!
1239 * Forward to the next possible start, i.e. the
1240 * next position following a non-word char.
1242 bol = pmatch[0].rm_so + bol + 1;
1243 while (word_char(bol[-1]) && bol < eol)
1244 bol++;
1245 eflags |= REG_NOTBOL;
1246 if (bol < eol)
1247 goto again;
1250 if (p->token == GREP_PATTERN_HEAD && saved_ch)
1251 *eol = saved_ch;
1252 if (hit) {
1253 pmatch[0].rm_so += bol - start;
1254 pmatch[0].rm_eo += bol - start;
1256 return hit;
1259 static int match_expr_eval(struct grep_opt *opt, struct grep_expr *x, char *bol,
1260 char *eol, enum grep_context ctx, ssize_t *col,
1261 ssize_t *icol, int collect_hits)
1263 int h = 0;
1265 if (!x)
1266 die("Not a valid grep expression");
1267 switch (x->node) {
1268 case GREP_NODE_TRUE:
1269 h = 1;
1270 break;
1271 case GREP_NODE_ATOM:
1273 regmatch_t tmp;
1274 h = match_one_pattern(x->u.atom, bol, eol, ctx,
1275 &tmp, 0);
1276 if (h && (*col < 0 || tmp.rm_so < *col))
1277 *col = tmp.rm_so;
1279 break;
1280 case GREP_NODE_NOT:
1282 * Upon visiting a GREP_NODE_NOT, col and icol become swapped.
1284 h = !match_expr_eval(opt, x->u.unary, bol, eol, ctx, icol, col,
1286 break;
1287 case GREP_NODE_AND:
1288 h = match_expr_eval(opt, x->u.binary.left, bol, eol, ctx, col,
1289 icol, 0);
1290 if (h || opt->columnnum) {
1292 * Don't short-circuit AND when given --column, since a
1293 * NOT earlier in the tree may turn this into an OR. In
1294 * this case, see the below comment.
1296 h &= match_expr_eval(opt, x->u.binary.right, bol, eol,
1297 ctx, col, icol, 0);
1299 break;
1300 case GREP_NODE_OR:
1301 if (!(collect_hits || opt->columnnum)) {
1303 * Don't short-circuit OR when given --column (or
1304 * collecting hits) to ensure we don't skip a later
1305 * child that would produce an earlier match.
1307 return (match_expr_eval(opt, x->u.binary.left, bol, eol,
1308 ctx, col, icol, 0) ||
1309 match_expr_eval(opt, x->u.binary.right, bol,
1310 eol, ctx, col, icol, 0));
1312 h = match_expr_eval(opt, x->u.binary.left, bol, eol, ctx, col,
1313 icol, 0);
1314 if (collect_hits)
1315 x->u.binary.left->hit |= h;
1316 h |= match_expr_eval(opt, x->u.binary.right, bol, eol, ctx, col,
1317 icol, collect_hits);
1318 break;
1319 default:
1320 die("Unexpected node type (internal error) %d", x->node);
1322 if (collect_hits)
1323 x->hit |= h;
1324 return h;
1327 static int match_expr(struct grep_opt *opt, char *bol, char *eol,
1328 enum grep_context ctx, ssize_t *col,
1329 ssize_t *icol, int collect_hits)
1331 struct grep_expr *x = opt->pattern_expression;
1332 return match_expr_eval(opt, x, bol, eol, ctx, col, icol, collect_hits);
1335 static int match_line(struct grep_opt *opt, char *bol, char *eol,
1336 ssize_t *col, ssize_t *icol,
1337 enum grep_context ctx, int collect_hits)
1339 struct grep_pat *p;
1340 int hit = 0;
1342 if (opt->extended)
1343 return match_expr(opt, bol, eol, ctx, col, icol,
1344 collect_hits);
1346 /* we do not call with collect_hits without being extended */
1347 for (p = opt->pattern_list; p; p = p->next) {
1348 regmatch_t tmp;
1349 if (match_one_pattern(p, bol, eol, ctx, &tmp, 0)) {
1350 hit |= 1;
1351 if (!opt->columnnum) {
1353 * Without --column, any single match on a line
1354 * is enough to know that it needs to be
1355 * printed. With --column, scan _all_ patterns
1356 * to find the earliest.
1358 break;
1360 if (*col < 0 || tmp.rm_so < *col)
1361 *col = tmp.rm_so;
1364 return hit;
1367 static int match_next_pattern(struct grep_pat *p, char *bol, char *eol,
1368 enum grep_context ctx,
1369 regmatch_t *pmatch, int eflags)
1371 regmatch_t match;
1373 if (!match_one_pattern(p, bol, eol, ctx, &match, eflags))
1374 return 0;
1375 if (match.rm_so < 0 || match.rm_eo < 0)
1376 return 0;
1377 if (pmatch->rm_so >= 0 && pmatch->rm_eo >= 0) {
1378 if (match.rm_so > pmatch->rm_so)
1379 return 1;
1380 if (match.rm_so == pmatch->rm_so && match.rm_eo < pmatch->rm_eo)
1381 return 1;
1383 pmatch->rm_so = match.rm_so;
1384 pmatch->rm_eo = match.rm_eo;
1385 return 1;
1388 static int next_match(struct grep_opt *opt, char *bol, char *eol,
1389 enum grep_context ctx, regmatch_t *pmatch, int eflags)
1391 struct grep_pat *p;
1392 int hit = 0;
1394 pmatch->rm_so = pmatch->rm_eo = -1;
1395 if (bol < eol) {
1396 for (p = opt->pattern_list; p; p = p->next) {
1397 switch (p->token) {
1398 case GREP_PATTERN: /* atom */
1399 case GREP_PATTERN_HEAD:
1400 case GREP_PATTERN_BODY:
1401 hit |= match_next_pattern(p, bol, eol, ctx,
1402 pmatch, eflags);
1403 break;
1404 default:
1405 break;
1409 return hit;
1412 static void show_line_header(struct grep_opt *opt, const char *name,
1413 unsigned lno, ssize_t cno, char sign)
1415 if (opt->heading && opt->last_shown == 0) {
1416 output_color(opt, name, strlen(name), opt->colors[GREP_COLOR_FILENAME]);
1417 opt->output(opt, "\n", 1);
1419 opt->last_shown = lno;
1421 if (!opt->heading && opt->pathname) {
1422 output_color(opt, name, strlen(name), opt->colors[GREP_COLOR_FILENAME]);
1423 output_sep(opt, sign);
1425 if (opt->linenum) {
1426 char buf[32];
1427 xsnprintf(buf, sizeof(buf), "%d", lno);
1428 output_color(opt, buf, strlen(buf), opt->colors[GREP_COLOR_LINENO]);
1429 output_sep(opt, sign);
1432 * Treat 'cno' as the 1-indexed offset from the start of a non-context
1433 * line to its first match. Otherwise, 'cno' is 0 indicating that we are
1434 * being called with a context line.
1436 if (opt->columnnum && cno) {
1437 char buf[32];
1438 xsnprintf(buf, sizeof(buf), "%"PRIuMAX, (uintmax_t)cno);
1439 output_color(opt, buf, strlen(buf), opt->colors[GREP_COLOR_COLUMNNO]);
1440 output_sep(opt, sign);
1444 static void show_line(struct grep_opt *opt, char *bol, char *eol,
1445 const char *name, unsigned lno, ssize_t cno, char sign)
1447 int rest = eol - bol;
1448 const char *match_color = NULL;
1449 const char *line_color = NULL;
1451 if (opt->file_break && opt->last_shown == 0) {
1452 if (opt->show_hunk_mark)
1453 opt->output(opt, "\n", 1);
1454 } else if (opt->pre_context || opt->post_context || opt->funcbody) {
1455 if (opt->last_shown == 0) {
1456 if (opt->show_hunk_mark) {
1457 output_color(opt, "--", 2, opt->colors[GREP_COLOR_SEP]);
1458 opt->output(opt, "\n", 1);
1460 } else if (lno > opt->last_shown + 1) {
1461 output_color(opt, "--", 2, opt->colors[GREP_COLOR_SEP]);
1462 opt->output(opt, "\n", 1);
1465 if (!opt->only_matching) {
1467 * In case the line we're being called with contains more than
1468 * one match, leave printing each header to the loop below.
1470 show_line_header(opt, name, lno, cno, sign);
1472 if (opt->color || opt->only_matching) {
1473 regmatch_t match;
1474 enum grep_context ctx = GREP_CONTEXT_BODY;
1475 int ch = *eol;
1476 int eflags = 0;
1478 if (opt->color) {
1479 if (sign == ':')
1480 match_color = opt->colors[GREP_COLOR_MATCH_SELECTED];
1481 else
1482 match_color = opt->colors[GREP_COLOR_MATCH_CONTEXT];
1483 if (sign == ':')
1484 line_color = opt->colors[GREP_COLOR_SELECTED];
1485 else if (sign == '-')
1486 line_color = opt->colors[GREP_COLOR_CONTEXT];
1487 else if (sign == '=')
1488 line_color = opt->colors[GREP_COLOR_FUNCTION];
1490 *eol = '\0';
1491 while (next_match(opt, bol, eol, ctx, &match, eflags)) {
1492 if (match.rm_so == match.rm_eo)
1493 break;
1495 if (opt->only_matching)
1496 show_line_header(opt, name, lno, cno, sign);
1497 else
1498 output_color(opt, bol, match.rm_so, line_color);
1499 output_color(opt, bol + match.rm_so,
1500 match.rm_eo - match.rm_so, match_color);
1501 if (opt->only_matching)
1502 opt->output(opt, "\n", 1);
1503 bol += match.rm_eo;
1504 cno += match.rm_eo;
1505 rest -= match.rm_eo;
1506 eflags = REG_NOTBOL;
1508 *eol = ch;
1510 if (!opt->only_matching) {
1511 output_color(opt, bol, rest, line_color);
1512 opt->output(opt, "\n", 1);
1516 int grep_use_locks;
1519 * This lock protects access to the gitattributes machinery, which is
1520 * not thread-safe.
1522 pthread_mutex_t grep_attr_mutex;
1524 static inline void grep_attr_lock(void)
1526 if (grep_use_locks)
1527 pthread_mutex_lock(&grep_attr_mutex);
1530 static inline void grep_attr_unlock(void)
1532 if (grep_use_locks)
1533 pthread_mutex_unlock(&grep_attr_mutex);
1537 * Same as git_attr_mutex, but protecting the thread-unsafe object db access.
1539 pthread_mutex_t grep_read_mutex;
1541 static int match_funcname(struct grep_opt *opt, struct grep_source *gs, char *bol, char *eol)
1543 xdemitconf_t *xecfg = opt->priv;
1544 if (xecfg && !xecfg->find_func) {
1545 grep_source_load_driver(gs, opt->repo->index);
1546 if (gs->driver->funcname.pattern) {
1547 const struct userdiff_funcname *pe = &gs->driver->funcname;
1548 xdiff_set_find_func(xecfg, pe->pattern, pe->cflags);
1549 } else {
1550 xecfg = opt->priv = NULL;
1554 if (xecfg) {
1555 char buf[1];
1556 return xecfg->find_func(bol, eol - bol, buf, 1,
1557 xecfg->find_func_priv) >= 0;
1560 if (bol == eol)
1561 return 0;
1562 if (isalpha(*bol) || *bol == '_' || *bol == '$')
1563 return 1;
1564 return 0;
1567 static void show_funcname_line(struct grep_opt *opt, struct grep_source *gs,
1568 char *bol, unsigned lno)
1570 while (bol > gs->buf) {
1571 char *eol = --bol;
1573 while (bol > gs->buf && bol[-1] != '\n')
1574 bol--;
1575 lno--;
1577 if (lno <= opt->last_shown)
1578 break;
1580 if (match_funcname(opt, gs, bol, eol)) {
1581 show_line(opt, bol, eol, gs->name, lno, 0, '=');
1582 break;
1587 static int is_empty_line(const char *bol, const char *eol);
1589 static void show_pre_context(struct grep_opt *opt, struct grep_source *gs,
1590 char *bol, char *end, unsigned lno)
1592 unsigned cur = lno, from = 1, funcname_lno = 0, orig_from;
1593 int funcname_needed = !!opt->funcname, comment_needed = 0;
1595 if (opt->pre_context < lno)
1596 from = lno - opt->pre_context;
1597 if (from <= opt->last_shown)
1598 from = opt->last_shown + 1;
1599 orig_from = from;
1600 if (opt->funcbody) {
1601 if (match_funcname(opt, gs, bol, end))
1602 comment_needed = 1;
1603 else
1604 funcname_needed = 1;
1605 from = opt->last_shown + 1;
1608 /* Rewind. */
1609 while (bol > gs->buf && cur > from) {
1610 char *next_bol = bol;
1611 char *eol = --bol;
1613 while (bol > gs->buf && bol[-1] != '\n')
1614 bol--;
1615 cur--;
1616 if (comment_needed && (is_empty_line(bol, eol) ||
1617 match_funcname(opt, gs, bol, eol))) {
1618 comment_needed = 0;
1619 from = orig_from;
1620 if (cur < from) {
1621 cur++;
1622 bol = next_bol;
1623 break;
1626 if (funcname_needed && match_funcname(opt, gs, bol, eol)) {
1627 funcname_lno = cur;
1628 funcname_needed = 0;
1629 if (opt->funcbody)
1630 comment_needed = 1;
1631 else
1632 from = orig_from;
1636 /* We need to look even further back to find a function signature. */
1637 if (opt->funcname && funcname_needed)
1638 show_funcname_line(opt, gs, bol, cur);
1640 /* Back forward. */
1641 while (cur < lno) {
1642 char *eol = bol, sign = (cur == funcname_lno) ? '=' : '-';
1644 while (*eol != '\n')
1645 eol++;
1646 show_line(opt, bol, eol, gs->name, cur, 0, sign);
1647 bol = eol + 1;
1648 cur++;
1652 static int should_lookahead(struct grep_opt *opt)
1654 struct grep_pat *p;
1656 if (opt->extended)
1657 return 0; /* punt for too complex stuff */
1658 if (opt->invert)
1659 return 0;
1660 for (p = opt->pattern_list; p; p = p->next) {
1661 if (p->token != GREP_PATTERN)
1662 return 0; /* punt for "header only" and stuff */
1664 return 1;
1667 static int look_ahead(struct grep_opt *opt,
1668 unsigned long *left_p,
1669 unsigned *lno_p,
1670 char **bol_p)
1672 unsigned lno = *lno_p;
1673 char *bol = *bol_p;
1674 struct grep_pat *p;
1675 char *sp, *last_bol;
1676 regoff_t earliest = -1;
1678 for (p = opt->pattern_list; p; p = p->next) {
1679 int hit;
1680 regmatch_t m;
1682 hit = patmatch(p, bol, bol + *left_p, &m, 0);
1683 if (!hit || m.rm_so < 0 || m.rm_eo < 0)
1684 continue;
1685 if (earliest < 0 || m.rm_so < earliest)
1686 earliest = m.rm_so;
1689 if (earliest < 0) {
1690 *bol_p = bol + *left_p;
1691 *left_p = 0;
1692 return 1;
1694 for (sp = bol + earliest; bol < sp && sp[-1] != '\n'; sp--)
1695 ; /* find the beginning of the line */
1696 last_bol = sp;
1698 for (sp = bol; sp < last_bol; sp++) {
1699 if (*sp == '\n')
1700 lno++;
1702 *left_p -= last_bol - bol;
1703 *bol_p = last_bol;
1704 *lno_p = lno;
1705 return 0;
1708 static int fill_textconv_grep(struct repository *r,
1709 struct userdiff_driver *driver,
1710 struct grep_source *gs)
1712 struct diff_filespec *df;
1713 char *buf;
1714 size_t size;
1716 if (!driver || !driver->textconv)
1717 return grep_source_load(gs);
1720 * The textconv interface is intimately tied to diff_filespecs, so we
1721 * have to pretend to be one. If we could unify the grep_source
1722 * and diff_filespec structs, this mess could just go away.
1724 df = alloc_filespec(gs->path);
1725 switch (gs->type) {
1726 case GREP_SOURCE_OID:
1727 fill_filespec(df, gs->identifier, 1, 0100644);
1728 break;
1729 case GREP_SOURCE_FILE:
1730 fill_filespec(df, &null_oid, 0, 0100644);
1731 break;
1732 default:
1733 BUG("attempt to textconv something without a path?");
1737 * fill_textconv is not remotely thread-safe; it may load objects
1738 * behind the scenes, and it modifies the global diff tempfile
1739 * structure.
1741 grep_read_lock();
1742 size = fill_textconv(r, driver, df, &buf);
1743 grep_read_unlock();
1744 free_filespec(df);
1747 * The normal fill_textconv usage by the diff machinery would just keep
1748 * the textconv'd buf separate from the diff_filespec. But much of the
1749 * grep code passes around a grep_source and assumes that its "buf"
1750 * pointer is the beginning of the thing we are searching. So let's
1751 * install our textconv'd version into the grep_source, taking care not
1752 * to leak any existing buffer.
1754 grep_source_clear_data(gs);
1755 gs->buf = buf;
1756 gs->size = size;
1758 return 0;
1761 static int is_empty_line(const char *bol, const char *eol)
1763 while (bol < eol && isspace(*bol))
1764 bol++;
1765 return bol == eol;
1768 static int grep_source_1(struct grep_opt *opt, struct grep_source *gs, int collect_hits)
1770 char *bol;
1771 char *peek_bol = NULL;
1772 unsigned long left;
1773 unsigned lno = 1;
1774 unsigned last_hit = 0;
1775 int binary_match_only = 0;
1776 unsigned count = 0;
1777 int try_lookahead = 0;
1778 int show_function = 0;
1779 struct userdiff_driver *textconv = NULL;
1780 enum grep_context ctx = GREP_CONTEXT_HEAD;
1781 xdemitconf_t xecfg;
1783 if (!opt->output)
1784 opt->output = std_output;
1786 if (opt->pre_context || opt->post_context || opt->file_break ||
1787 opt->funcbody) {
1788 /* Show hunk marks, except for the first file. */
1789 if (opt->last_shown)
1790 opt->show_hunk_mark = 1;
1792 * If we're using threads then we can't easily identify
1793 * the first file. Always put hunk marks in that case
1794 * and skip the very first one later in work_done().
1796 if (opt->output != std_output)
1797 opt->show_hunk_mark = 1;
1799 opt->last_shown = 0;
1801 if (opt->allow_textconv) {
1802 grep_source_load_driver(gs, opt->repo->index);
1804 * We might set up the shared textconv cache data here, which
1805 * is not thread-safe.
1807 grep_attr_lock();
1808 textconv = userdiff_get_textconv(opt->repo, gs->driver);
1809 grep_attr_unlock();
1813 * We know the result of a textconv is text, so we only have to care
1814 * about binary handling if we are not using it.
1816 if (!textconv) {
1817 switch (opt->binary) {
1818 case GREP_BINARY_DEFAULT:
1819 if (grep_source_is_binary(gs, opt->repo->index))
1820 binary_match_only = 1;
1821 break;
1822 case GREP_BINARY_NOMATCH:
1823 if (grep_source_is_binary(gs, opt->repo->index))
1824 return 0; /* Assume unmatch */
1825 break;
1826 case GREP_BINARY_TEXT:
1827 break;
1828 default:
1829 BUG("unknown binary handling mode");
1833 memset(&xecfg, 0, sizeof(xecfg));
1834 opt->priv = &xecfg;
1836 try_lookahead = should_lookahead(opt);
1838 if (fill_textconv_grep(opt->repo, textconv, gs) < 0)
1839 return 0;
1841 bol = gs->buf;
1842 left = gs->size;
1843 while (left) {
1844 char *eol, ch;
1845 int hit;
1846 ssize_t cno;
1847 ssize_t col = -1, icol = -1;
1850 * look_ahead() skips quickly to the line that possibly
1851 * has the next hit; don't call it if we need to do
1852 * something more than just skipping the current line
1853 * in response to an unmatch for the current line. E.g.
1854 * inside a post-context window, we will show the current
1855 * line as a context around the previous hit when it
1856 * doesn't hit.
1858 if (try_lookahead
1859 && !(last_hit
1860 && (show_function ||
1861 lno <= last_hit + opt->post_context))
1862 && look_ahead(opt, &left, &lno, &bol))
1863 break;
1864 eol = end_of_line(bol, &left);
1865 ch = *eol;
1866 *eol = 0;
1868 if ((ctx == GREP_CONTEXT_HEAD) && (eol == bol))
1869 ctx = GREP_CONTEXT_BODY;
1871 hit = match_line(opt, bol, eol, &col, &icol, ctx, collect_hits);
1872 *eol = ch;
1874 if (collect_hits)
1875 goto next_line;
1877 /* "grep -v -e foo -e bla" should list lines
1878 * that do not have either, so inversion should
1879 * be done outside.
1881 if (opt->invert)
1882 hit = !hit;
1883 if (opt->unmatch_name_only) {
1884 if (hit)
1885 return 0;
1886 goto next_line;
1888 if (hit) {
1889 count++;
1890 if (opt->status_only)
1891 return 1;
1892 if (opt->name_only) {
1893 show_name(opt, gs->name);
1894 return 1;
1896 if (opt->count)
1897 goto next_line;
1898 if (binary_match_only) {
1899 opt->output(opt, "Binary file ", 12);
1900 output_color(opt, gs->name, strlen(gs->name),
1901 opt->colors[GREP_COLOR_FILENAME]);
1902 opt->output(opt, " matches\n", 9);
1903 return 1;
1905 /* Hit at this line. If we haven't shown the
1906 * pre-context lines, we would need to show them.
1908 if (opt->pre_context || opt->funcbody)
1909 show_pre_context(opt, gs, bol, eol, lno);
1910 else if (opt->funcname)
1911 show_funcname_line(opt, gs, bol, lno);
1912 cno = opt->invert ? icol : col;
1913 if (cno < 0) {
1915 * A negative cno indicates that there was no
1916 * match on the line. We are thus inverted and
1917 * being asked to show all lines that _don't_
1918 * match a given expression. Therefore, set cno
1919 * to 0 to suggest the whole line matches.
1921 cno = 0;
1923 show_line(opt, bol, eol, gs->name, lno, cno + 1, ':');
1924 last_hit = lno;
1925 if (opt->funcbody)
1926 show_function = 1;
1927 goto next_line;
1929 if (show_function && (!peek_bol || peek_bol < bol)) {
1930 unsigned long peek_left = left;
1931 char *peek_eol = eol;
1934 * Trailing empty lines are not interesting.
1935 * Peek past them to see if they belong to the
1936 * body of the current function.
1938 peek_bol = bol;
1939 while (is_empty_line(peek_bol, peek_eol)) {
1940 peek_bol = peek_eol + 1;
1941 peek_eol = end_of_line(peek_bol, &peek_left);
1944 if (match_funcname(opt, gs, peek_bol, peek_eol))
1945 show_function = 0;
1947 if (show_function ||
1948 (last_hit && lno <= last_hit + opt->post_context)) {
1949 /* If the last hit is within the post context,
1950 * we need to show this line.
1952 show_line(opt, bol, eol, gs->name, lno, col + 1, '-');
1955 next_line:
1956 bol = eol + 1;
1957 if (!left)
1958 break;
1959 left--;
1960 lno++;
1963 if (collect_hits)
1964 return 0;
1966 if (opt->status_only)
1967 return opt->unmatch_name_only;
1968 if (opt->unmatch_name_only) {
1969 /* We did not see any hit, so we want to show this */
1970 show_name(opt, gs->name);
1971 return 1;
1974 xdiff_clear_find_func(&xecfg);
1975 opt->priv = NULL;
1977 /* NEEDSWORK:
1978 * The real "grep -c foo *.c" gives many "bar.c:0" lines,
1979 * which feels mostly useless but sometimes useful. Maybe
1980 * make it another option? For now suppress them.
1982 if (opt->count && count) {
1983 char buf[32];
1984 if (opt->pathname) {
1985 output_color(opt, gs->name, strlen(gs->name),
1986 opt->colors[GREP_COLOR_FILENAME]);
1987 output_sep(opt, ':');
1989 xsnprintf(buf, sizeof(buf), "%u\n", count);
1990 opt->output(opt, buf, strlen(buf));
1991 return 1;
1993 return !!last_hit;
1996 static void clr_hit_marker(struct grep_expr *x)
1998 /* All-hit markers are meaningful only at the very top level
1999 * OR node.
2001 while (1) {
2002 x->hit = 0;
2003 if (x->node != GREP_NODE_OR)
2004 return;
2005 x->u.binary.left->hit = 0;
2006 x = x->u.binary.right;
2010 static int chk_hit_marker(struct grep_expr *x)
2012 /* Top level nodes have hit markers. See if they all are hits */
2013 while (1) {
2014 if (x->node != GREP_NODE_OR)
2015 return x->hit;
2016 if (!x->u.binary.left->hit)
2017 return 0;
2018 x = x->u.binary.right;
2022 int grep_source(struct grep_opt *opt, struct grep_source *gs)
2025 * we do not have to do the two-pass grep when we do not check
2026 * buffer-wide "all-match".
2028 if (!opt->all_match)
2029 return grep_source_1(opt, gs, 0);
2031 /* Otherwise the toplevel "or" terms hit a bit differently.
2032 * We first clear hit markers from them.
2034 clr_hit_marker(opt->pattern_expression);
2035 grep_source_1(opt, gs, 1);
2037 if (!chk_hit_marker(opt->pattern_expression))
2038 return 0;
2040 return grep_source_1(opt, gs, 0);
2043 int grep_buffer(struct grep_opt *opt, char *buf, unsigned long size)
2045 struct grep_source gs;
2046 int r;
2048 grep_source_init(&gs, GREP_SOURCE_BUF, NULL, NULL, NULL);
2049 gs.buf = buf;
2050 gs.size = size;
2052 r = grep_source(opt, &gs);
2054 grep_source_clear(&gs);
2055 return r;
2058 void grep_source_init(struct grep_source *gs, enum grep_source_type type,
2059 const char *name, const char *path,
2060 const void *identifier)
2062 gs->type = type;
2063 gs->name = xstrdup_or_null(name);
2064 gs->path = xstrdup_or_null(path);
2065 gs->buf = NULL;
2066 gs->size = 0;
2067 gs->driver = NULL;
2069 switch (type) {
2070 case GREP_SOURCE_FILE:
2071 gs->identifier = xstrdup(identifier);
2072 break;
2073 case GREP_SOURCE_OID:
2074 gs->identifier = oiddup(identifier);
2075 break;
2076 case GREP_SOURCE_BUF:
2077 gs->identifier = NULL;
2078 break;
2082 void grep_source_clear(struct grep_source *gs)
2084 FREE_AND_NULL(gs->name);
2085 FREE_AND_NULL(gs->path);
2086 FREE_AND_NULL(gs->identifier);
2087 grep_source_clear_data(gs);
2090 void grep_source_clear_data(struct grep_source *gs)
2092 switch (gs->type) {
2093 case GREP_SOURCE_FILE:
2094 case GREP_SOURCE_OID:
2095 FREE_AND_NULL(gs->buf);
2096 gs->size = 0;
2097 break;
2098 case GREP_SOURCE_BUF:
2099 /* leave user-provided buf intact */
2100 break;
2104 static int grep_source_load_oid(struct grep_source *gs)
2106 enum object_type type;
2108 grep_read_lock();
2109 gs->buf = read_object_file(gs->identifier, &type, &gs->size);
2110 grep_read_unlock();
2112 if (!gs->buf)
2113 return error(_("'%s': unable to read %s"),
2114 gs->name,
2115 oid_to_hex(gs->identifier));
2116 return 0;
2119 static int grep_source_load_file(struct grep_source *gs)
2121 const char *filename = gs->identifier;
2122 struct stat st;
2123 char *data;
2124 size_t size;
2125 int i;
2127 if (lstat(filename, &st) < 0) {
2128 err_ret:
2129 if (errno != ENOENT)
2130 error_errno(_("failed to stat '%s'"), filename);
2131 return -1;
2133 if (!S_ISREG(st.st_mode))
2134 return -1;
2135 size = xsize_t(st.st_size);
2136 i = open(filename, O_RDONLY);
2137 if (i < 0)
2138 goto err_ret;
2139 data = xmallocz(size);
2140 if (st.st_size != read_in_full(i, data, size)) {
2141 error_errno(_("'%s': short read"), filename);
2142 close(i);
2143 free(data);
2144 return -1;
2146 close(i);
2148 gs->buf = data;
2149 gs->size = size;
2150 return 0;
2153 static int grep_source_load(struct grep_source *gs)
2155 if (gs->buf)
2156 return 0;
2158 switch (gs->type) {
2159 case GREP_SOURCE_FILE:
2160 return grep_source_load_file(gs);
2161 case GREP_SOURCE_OID:
2162 return grep_source_load_oid(gs);
2163 case GREP_SOURCE_BUF:
2164 return gs->buf ? 0 : -1;
2166 BUG("invalid grep_source type to load");
2169 void grep_source_load_driver(struct grep_source *gs,
2170 struct index_state *istate)
2172 if (gs->driver)
2173 return;
2175 grep_attr_lock();
2176 if (gs->path)
2177 gs->driver = userdiff_find_by_path(istate, gs->path);
2178 if (!gs->driver)
2179 gs->driver = userdiff_find_by_name("default");
2180 grep_attr_unlock();
2183 static int grep_source_is_binary(struct grep_source *gs,
2184 struct index_state *istate)
2186 grep_source_load_driver(gs, istate);
2187 if (gs->driver->binary != -1)
2188 return gs->driver->binary;
2190 if (!grep_source_load(gs))
2191 return buffer_is_binary(gs->buf, gs->size);
2193 return 0;