?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
string-list.[ch]: add a string_list_init_{nodup,dup}()
[git.git] / grep.c
blob8f91af1cb006f95ee25668f0c080ec09c745c141
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"
12
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);
16
17 static void std_output(struct grep_opt *opt, const void *buf, size_t size)
18 {
19 fwrite(buf, size, 1, stdout);
20 }
21
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,
37 },
38 .only_matching = 0,
39 .color = -1,
40 .output = std_output,
41 };
42
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",
53 };
54
55 static int parse_pattern_type_arg(const char *opt, const char *arg)
56 {
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);
68 }
69
70 define_list_config_array_extra(color_grep_slots, {"match"});
71
72 /*
73 * Read the configuration file once and store it in
74 * the grep_defaults template.
75 */
76 int grep_config(const char *var, const char *value, void *cb)
77 {
78 struct grep_opt *opt = &grep_defaults;
79 const char *slot;
80
81 if (userdiff_config(var, value) < 0)
82 return -1;
83
84 /*
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.
90 */
91
92 if (!strcmp(var, "grep.extendedregexp")) {
93 opt->extended_regexp_option = git_config_bool(var, value);
94 return 0;
95 }
96
97 if (!strcmp(var, "grep.patterntype")) {
98 opt->pattern_type_option = parse_pattern_type_arg(var, value);
99 return 0;
100 }
101
102 if (!strcmp(var, "grep.linenumber")) {
103 opt->linenum = git_config_bool(var, value);
104 return 0;
105 }
106 if (!strcmp(var, "grep.column")) {
107 opt->columnnum = git_config_bool(var, value);
108 return 0;
109 }
110
111 if (!strcmp(var, "grep.fullname")) {
112 opt->relative = !git_config_bool(var, value);
113 return 0;
114 }
115
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;
126
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);
133 }
134 return 0;
135 }
136
137 /*
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).
141 */
142 void grep_init(struct grep_opt *opt, struct repository *repo, const char *prefix)
143 {
144 *opt = grep_defaults;
145
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;
151 }
152
153 static void grep_set_pattern_type_option(enum grep_pattern_type pattern_type, struct grep_opt *opt)
154 {
155 /*
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.
161 *
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.
168 */
169 if (pattern_type != GREP_PATTERN_TYPE_ERE &&
170 opt->extended_regexp_option)
171 opt->extended_regexp_option = 0;
172
173 switch (pattern_type) {
174 case GREP_PATTERN_TYPE_UNSPECIFIED:
175 /* fall through */
176
177 case GREP_PATTERN_TYPE_BRE:
178 break;
179
180 case GREP_PATTERN_TYPE_ERE:
181 opt->extended_regexp_option = 1;
182 break;
183
184 case GREP_PATTERN_TYPE_FIXED:
185 opt->fixed = 1;
186 break;
187
188 case GREP_PATTERN_TYPE_PCRE:
189 opt->pcre2 = 1;
190 break;
191 }
192 }
193
194 void grep_commit_pattern_type(enum grep_pattern_type pattern_type, struct grep_opt *opt)
195 {
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)
201 /*
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!
205 */
206 grep_set_pattern_type_option(GREP_PATTERN_TYPE_ERE, opt);
207 }
208
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)
213 {
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;
222 }
223
224 static void do_append_grep_pat(struct grep_pat ***tail, struct grep_pat *p)
225 {
226 **tail = p;
227 *tail = &p->next;
228 p->next = NULL;
229
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;
242 }
243 }
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;
254 }
255 break;
256 default:
257 break;
258 }
259 }
260
261 void append_header_grep_pattern(struct grep_opt *opt,
262 enum grep_header_field field, const char *pat)
263 {
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);
269 }
270
271 void append_grep_pattern(struct grep_opt *opt, const char *pat,
272 const char *origin, int no, enum grep_pat_token t)
273 {
274 append_grep_pat(opt, pat, strlen(pat), origin, no, t);
275 }
276
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)
279 {
280 struct grep_pat *p = create_grep_pat(pat, patlen, origin, no, t, 0);
281 do_append_grep_pat(&opt->pattern_tail, p);
282 }
283
284 struct grep_opt *grep_opt_dup(const struct grep_opt *opt)
285 {
286 struct grep_pat *pat;
287 struct grep_opt *ret = xmalloc(sizeof(struct grep_opt));
288 *ret = *opt;
289
290 ret->pattern_list = NULL;
291 ret->pattern_tail = &ret->pattern_list;
292
293 for(pat = opt->pattern_list; pat != NULL; pat = pat->next)
294 {
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);
301 }
302
303 return ret;
304 }
305
306 static NORETURN void compile_regexp_failed(const struct grep_pat *p,
307 const char *error)
308 {
309 char where[1024];
310
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;
317
318 die("%s'%s': %s", where, p->pattern, error);
319 }
320
321 static int is_fixed(const char *s, size_t len)
322 {
323 size_t i;
324
325 for (i = 0; i < len; i++) {
326 if (is_regex_special(s[i]))
327 return 0;
328 }
329
330 return 1;
331 }
332
333 #ifdef USE_LIBPCRE2
334 #define GREP_PCRE2_DEBUG_MALLOC 0
335
336 static void *pcre2_malloc(PCRE2_SIZE size, MAYBE_UNUSED void *memory_data)
337 {
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;
344 }
345
346 static void pcre2_free(void *pointer, MAYBE_UNUSED void *memory_data)
347 {
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);
354 }
355
356 static void compile_pcre2_pattern(struct grep_pat *p, const struct grep_opt *opt)
357 {
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;
365
366 /*
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.
370 */
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");
375
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);
382 }
383 options |= PCRE2_CASELESS;
384 }
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);
388
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
394
395 p->pcre2_pattern = pcre2_compile((PCRE2_SPTR)p->pattern,
396 p->patternlen, options, &error, &erroffset,
397 p->pcre2_compile_context);
398
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);
406 }
407
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);
413
414 /*
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.
420 *
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)
428 */
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;
435 }
436 }
437 }
438
439 static int pcre2match(struct grep_pat *p, const char *line, const char *eol,
440 regmatch_t *match, int eflags)
441 {
442 int ret, flags = 0;
443 PCRE2_SIZE *ovector;
444 PCRE2_UCHAR errbuf[256];
445
446 if (eflags & REG_NOTBOL)
447 flags |= PCRE2_NOTBOL;
448
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);
457
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);
463 }
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];
469 }
470
471 return ret;
472 }
473
474 static void free_pcre2_pattern(struct grep_pat *p)
475 {
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);
485 }
486 #else /* !USE_LIBPCRE2 */
487 static void compile_pcre2_pattern(struct grep_pat *p, const struct grep_opt *opt)
488 {
489 die("cannot use Perl-compatible regexes when not compiled with USE_LIBPCRE");
490 }
491
492 static int pcre2match(struct grep_pat *p, const char *line, const char *eol,
493 regmatch_t *match, int eflags)
494 {
495 return 1;
496 }
497
498 static void free_pcre2_pattern(struct grep_pat *p)
499 {
500 }
501
502 static void compile_fixed_regexp(struct grep_pat *p, struct grep_opt *opt)
503 {
504 struct strbuf sb = STRBUF_INIT;
505 int err;
506 int regflags = 0;
507
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);
517 }
518 }
519 #endif /* !USE_LIBPCRE2 */
520
521 static void compile_regexp(struct grep_pat *p, struct grep_opt *opt)
522 {
523 int err;
524 int regflags = REG_NEWLINE;
525
526 p->word_regexp = opt->word_regexp;
527 p->ignore_case = opt->ignore_case;
528 p->fixed = opt->fixed;
529
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"));
532
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;
542 }
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 {
549 /*
550 * E.g. t7811-grep-open.sh relies on the
551 * pattern being restored.
552 */
553 char *old_pattern = p->pattern;
554 size_t old_patternlen = p->patternlen;
555 struct strbuf sb = STRBUF_INIT;
556
557 /*
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.
563 */
564 strbuf_add(&sb, "\\Q", 2);
565 strbuf_add(&sb, p->pattern, p->patternlen);
566 strbuf_add(&sb, "\\E", 2);
567
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);
574 }
575 #else /* !USE_LIBPCRE2 */
576 compile_fixed_regexp(p, opt);
577 #endif /* !USE_LIBPCRE2 */
578 return;
579 }
580
581 if (opt->pcre2) {
582 compile_pcre2_pattern(p, opt);
583 return;
584 }
585
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);
595 }
596 }
597
598 static struct grep_expr *compile_pattern_or(struct grep_pat **);
599 static struct grep_expr *compile_pattern_atom(struct grep_pat **list)
600 {
601 struct grep_pat *p;
602 struct grep_expr *x;
603
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;
625 }
626 }
627
628 static struct grep_expr *compile_pattern_not(struct grep_pat **list)
629 {
630 struct grep_pat *p;
631 struct grep_expr *x;
632
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);
649 }
650 }
651
652 static struct grep_expr *compile_pattern_and(struct grep_pat **list)
653 {
654 struct grep_pat *p;
655 struct grep_expr *x, *y, *z;
656
657 x = compile_pattern_not(list);
658 p = *list;
659 if (p && p->token == GREP_AND) {
660 if (!p->next)
661 die("--and not followed by pattern expression");
662 *list = p->next;
663 y = compile_pattern_and(list);
664 if (!y)
665 die("--and not followed by pattern expression");
666 CALLOC_ARRAY(z, 1);
667 z->node = GREP_NODE_AND;
668 z->u.binary.left = x;
669 z->u.binary.right = y;
670 return z;
671 }
672 return x;
673 }
674
675 static struct grep_expr *compile_pattern_or(struct grep_pat **list)
676 {
677 struct grep_pat *p;
678 struct grep_expr *x, *y, *z;
679
680 x = compile_pattern_and(list);
681 p = *list;
682 if (x && p && p->token != GREP_CLOSE_PAREN) {
683 y = compile_pattern_or(list);
684 if (!y)
685 die("not a pattern expression %s", p->pattern);
686 CALLOC_ARRAY(z, 1);
687 z->node = GREP_NODE_OR;
688 z->u.binary.left = x;
689 z->u.binary.right = y;
690 return z;
691 }
692 return x;
693 }
694
695 static struct grep_expr *compile_pattern_expr(struct grep_pat **list)
696 {
697 return compile_pattern_or(list);
698 }
699
700 static struct grep_expr *grep_true_expr(void)
701 {
702 struct grep_expr *z = xcalloc(1, sizeof(*z));
703 z->node = GREP_NODE_TRUE;
704 return z;
705 }
706
707 static struct grep_expr *grep_or_expr(struct grep_expr *left, struct grep_expr *right)
708 {
709 struct grep_expr *z = xcalloc(1, sizeof(*z));
710 z->node = GREP_NODE_OR;
711 z->u.binary.left = left;
712 z->u.binary.right = right;
713 return z;
714 }
715
716 static struct grep_expr *prep_header_patterns(struct grep_opt *opt)
717 {
718 struct grep_pat *p;
719 struct grep_expr *header_expr;
720 struct grep_expr *(header_group[GREP_HEADER_FIELD_MAX]);
721 enum grep_header_field fld;
722
723 if (!opt->header_list)
724 return NULL;
725
726 for (p = opt->header_list; p; p = p->next) {
727 if (p->token != GREP_PATTERN_HEAD)
728 BUG("a non-header pattern in grep header list.");
729 if (p->field < GREP_HEADER_FIELD_MIN ||
730 GREP_HEADER_FIELD_MAX <= p->field)
731 BUG("unknown header field %d", p->field);
732 compile_regexp(p, opt);
733 }
734
735 for (fld = 0; fld < GREP_HEADER_FIELD_MAX; fld++)
736 header_group[fld] = NULL;
737
738 for (p = opt->header_list; p; p = p->next) {
739 struct grep_expr *h;
740 struct grep_pat *pp = p;
741
742 h = compile_pattern_atom(&pp);
743 if (!h || pp != p->next)
744 BUG("malformed header expr");
745 if (!header_group[p->field]) {
746 header_group[p->field] = h;
747 continue;
748 }
749 header_group[p->field] = grep_or_expr(h, header_group[p->field]);
750 }
751
752 header_expr = NULL;
753
754 for (fld = 0; fld < GREP_HEADER_FIELD_MAX; fld++) {
755 if (!header_group[fld])
756 continue;
757 if (!header_expr)
758 header_expr = grep_true_expr();
759 header_expr = grep_or_expr(header_group[fld], header_expr);
760 }
761 return header_expr;
762 }
763
764 static struct grep_expr *grep_splice_or(struct grep_expr *x, struct grep_expr *y)
765 {
766 struct grep_expr *z = x;
767
768 while (x) {
769 assert(x->node == GREP_NODE_OR);
770 if (x->u.binary.right &&
771 x->u.binary.right->node == GREP_NODE_TRUE) {
772 x->u.binary.right = y;
773 break;
774 }
775 x = x->u.binary.right;
776 }
777 return z;
778 }
779
780 void compile_grep_patterns(struct grep_opt *opt)
781 {
782 struct grep_pat *p;
783 struct grep_expr *header_expr = prep_header_patterns(opt);
784
785 for (p = opt->pattern_list; p; p = p->next) {
786 switch (p->token) {
787 case GREP_PATTERN: /* atom */
788 case GREP_PATTERN_HEAD:
789 case GREP_PATTERN_BODY:
790 compile_regexp(p, opt);
791 break;
792 default:
793 opt->extended = 1;
794 break;
795 }
796 }
797
798 if (opt->all_match || header_expr)
799 opt->extended = 1;
800 else if (!opt->extended)
801 return;
802
803 p = opt->pattern_list;
804 if (p)
805 opt->pattern_expression = compile_pattern_expr(&p);
806 if (p)
807 die("incomplete pattern expression: %s", p->pattern);
808
809 if (!header_expr)
810 return;
811
812 if (!opt->pattern_expression)
813 opt->pattern_expression = header_expr;
814 else if (opt->all_match)
815 opt->pattern_expression = grep_splice_or(header_expr,
816 opt->pattern_expression);
817 else
818 opt->pattern_expression = grep_or_expr(opt->pattern_expression,
819 header_expr);
820 opt->all_match = 1;
821 }
822
823 static void free_pattern_expr(struct grep_expr *x)
824 {
825 switch (x->node) {
826 case GREP_NODE_TRUE:
827 case GREP_NODE_ATOM:
828 break;
829 case GREP_NODE_NOT:
830 free_pattern_expr(x->u.unary);
831 break;
832 case GREP_NODE_AND:
833 case GREP_NODE_OR:
834 free_pattern_expr(x->u.binary.left);
835 free_pattern_expr(x->u.binary.right);
836 break;
837 }
838 free(x);
839 }
840
841 void free_grep_patterns(struct grep_opt *opt)
842 {
843 struct grep_pat *p, *n;
844
845 for (p = opt->pattern_list; p; p = n) {
846 n = p->next;
847 switch (p->token) {
848 case GREP_PATTERN: /* atom */
849 case GREP_PATTERN_HEAD:
850 case GREP_PATTERN_BODY:
851 if (p->pcre2_pattern)
852 free_pcre2_pattern(p);
853 else
854 regfree(&p->regexp);
855 free(p->pattern);
856 break;
857 default:
858 break;
859 }
860 free(p);
861 }
862
863 if (!opt->extended)
864 return;
865 free_pattern_expr(opt->pattern_expression);
866 }
867
868 static char *end_of_line(char *cp, unsigned long *left)
869 {
870 unsigned long l = *left;
871 while (l && *cp != '\n') {
872 l--;
873 cp++;
874 }
875 *left = l;
876 return cp;
877 }
878
879 static int word_char(char ch)
880 {
881 return isalnum(ch) || ch == '_';
882 }
883
884 static void output_color(struct grep_opt *opt, const void *data, size_t size,
885 const char *color)
886 {
887 if (want_color(opt->color) && color && color[0]) {
888 opt->output(opt, color, strlen(color));
889 opt->output(opt, data, size);
890 opt->output(opt, GIT_COLOR_RESET, strlen(GIT_COLOR_RESET));
891 } else
892 opt->output(opt, data, size);
893 }
894
895 static void output_sep(struct grep_opt *opt, char sign)
896 {
897 if (opt->null_following_name)
898 opt->output(opt, "\0", 1);
899 else
900 output_color(opt, &sign, 1, opt->colors[GREP_COLOR_SEP]);
901 }
902
903 static void show_name(struct grep_opt *opt, const char *name)
904 {
905 output_color(opt, name, strlen(name), opt->colors[GREP_COLOR_FILENAME]);
906 opt->output(opt, opt->null_following_name ? "\0" : "\n", 1);
907 }
908
909 static int patmatch(struct grep_pat *p, char *line, char *eol,
910 regmatch_t *match, int eflags)
911 {
912 int hit;
913
914 if (p->pcre2_pattern)
915 hit = !pcre2match(p, line, eol, match, eflags);
916 else
917 hit = !regexec_buf(&p->regexp, line, eol - line, 1, match,
918 eflags);
919
920 return hit;
921 }
922
923 static int strip_timestamp(char *bol, char **eol_p)
924 {
925 char *eol = *eol_p;
926 int ch;
927
928 while (bol < --eol) {
929 if (*eol != '>')
930 continue;
931 *eol_p = ++eol;
932 ch = *eol;
933 *eol = '\0';
934 return ch;
935 }
936 return 0;
937 }
938
939 static struct {
940 const char *field;
941 size_t len;
942 } header_field[] = {
943 { "author ", 7 },
944 { "committer ", 10 },
945 { "reflog ", 7 },
946 };
947
948 static int match_one_pattern(struct grep_pat *p, char *bol, char *eol,
949 enum grep_context ctx,
950 regmatch_t *pmatch, int eflags)
951 {
952 int hit = 0;
953 int saved_ch = 0;
954 const char *start = bol;
955
956 if ((p->token != GREP_PATTERN) &&
957 ((p->token == GREP_PATTERN_HEAD) != (ctx == GREP_CONTEXT_HEAD)))
958 return 0;
959
960 if (p->token == GREP_PATTERN_HEAD) {
961 const char *field;
962 size_t len;
963 assert(p->field < ARRAY_SIZE(header_field));
964 field = header_field[p->field].field;
965 len = header_field[p->field].len;
966 if (strncmp(bol, field, len))
967 return 0;
968 bol += len;
969 switch (p->field) {
970 case GREP_HEADER_AUTHOR:
971 case GREP_HEADER_COMMITTER:
972 saved_ch = strip_timestamp(bol, &eol);
973 break;
974 default:
975 break;
976 }
977 }
978
979 again:
980 hit = patmatch(p, bol, eol, pmatch, eflags);
981
982 if (hit && p->word_regexp) {
983 if ((pmatch[0].rm_so < 0) ||
984 (eol - bol) < pmatch[0].rm_so ||
985 (pmatch[0].rm_eo < 0) ||
986 (eol - bol) < pmatch[0].rm_eo)
987 die("regexp returned nonsense");
988
989 /* Match beginning must be either beginning of the
990 * line, or at word boundary (i.e. the last char must
991 * not be a word char). Similarly, match end must be
992 * either end of the line, or at word boundary
993 * (i.e. the next char must not be a word char).
994 */
995 if ( ((pmatch[0].rm_so == 0) ||
996 !word_char(bol[pmatch[0].rm_so-1])) &&
997 ((pmatch[0].rm_eo == (eol-bol)) ||
998 !word_char(bol[pmatch[0].rm_eo])) )
999 ;
1000 else
1001 hit = 0;
1002
1003 /* Words consist of at least one character. */
1004 if (pmatch->rm_so == pmatch->rm_eo)
1005 hit = 0;
1006
1007 if (!hit && pmatch[0].rm_so + bol + 1 < eol) {
1008 /* There could be more than one match on the
1009 * line, and the first match might not be
1010 * strict word match. But later ones could be!
1011 * Forward to the next possible start, i.e. the
1012 * next position following a non-word char.
1013 */
1014 bol = pmatch[0].rm_so + bol + 1;
1015 while (word_char(bol[-1]) && bol < eol)
1016 bol++;
1017 eflags |= REG_NOTBOL;
1018 if (bol < eol)
1019 goto again;
1020 }
1021 }
1022 if (p->token == GREP_PATTERN_HEAD && saved_ch)
1023 *eol = saved_ch;
1024 if (hit) {
1025 pmatch[0].rm_so += bol - start;
1026 pmatch[0].rm_eo += bol - start;
1027 }
1028 return hit;
1029 }
1030
1031 static int match_expr_eval(struct grep_opt *opt, struct grep_expr *x, char *bol,
1032 char *eol, enum grep_context ctx, ssize_t *col,
1033 ssize_t *icol, int collect_hits)
1034 {
1035 int h = 0;
1036
1037 if (!x)
1038 die("Not a valid grep expression");
1039 switch (x->node) {
1040 case GREP_NODE_TRUE:
1041 h = 1;
1042 break;
1043 case GREP_NODE_ATOM:
1044 {
1045 regmatch_t tmp;
1046 h = match_one_pattern(x->u.atom, bol, eol, ctx,
1047 &tmp, 0);
1048 if (h && (*col < 0 || tmp.rm_so < *col))
1049 *col = tmp.rm_so;
1050 }
1051 break;
1052 case GREP_NODE_NOT:
1053 /*
1054 * Upon visiting a GREP_NODE_NOT, col and icol become swapped.
1055 */
1056 h = !match_expr_eval(opt, x->u.unary, bol, eol, ctx, icol, col,
1057 0);
1058 break;
1059 case GREP_NODE_AND:
1060 h = match_expr_eval(opt, x->u.binary.left, bol, eol, ctx, col,
1061 icol, 0);
1062 if (h || opt->columnnum) {
1063 /*
1064 * Don't short-circuit AND when given --column, since a
1065 * NOT earlier in the tree may turn this into an OR. In
1066 * this case, see the below comment.
1067 */
1068 h &= match_expr_eval(opt, x->u.binary.right, bol, eol,
1069 ctx, col, icol, 0);
1070 }
1071 break;
1072 case GREP_NODE_OR:
1073 if (!(collect_hits || opt->columnnum)) {
1074 /*
1075 * Don't short-circuit OR when given --column (or
1076 * collecting hits) to ensure we don't skip a later
1077 * child that would produce an earlier match.
1078 */
1079 return (match_expr_eval(opt, x->u.binary.left, bol, eol,
1080 ctx, col, icol, 0) ||
1081 match_expr_eval(opt, x->u.binary.right, bol,
1082 eol, ctx, col, icol, 0));
1083 }
1084 h = match_expr_eval(opt, x->u.binary.left, bol, eol, ctx, col,
1085 icol, 0);
1086 if (collect_hits)
1087 x->u.binary.left->hit |= h;
1088 h |= match_expr_eval(opt, x->u.binary.right, bol, eol, ctx, col,
1089 icol, collect_hits);
1090 break;
1091 default:
1092 die("Unexpected node type (internal error) %d", x->node);
1093 }
1094 if (collect_hits)
1095 x->hit |= h;
1096 return h;
1097 }
1098
1099 static int match_expr(struct grep_opt *opt, char *bol, char *eol,
1100 enum grep_context ctx, ssize_t *col,
1101 ssize_t *icol, int collect_hits)
1102 {
1103 struct grep_expr *x = opt->pattern_expression;
1104 return match_expr_eval(opt, x, bol, eol, ctx, col, icol, collect_hits);
1105 }
1106
1107 static int match_line(struct grep_opt *opt, char *bol, char *eol,
1108 ssize_t *col, ssize_t *icol,
1109 enum grep_context ctx, int collect_hits)
1110 {
1111 struct grep_pat *p;
1112 int hit = 0;
1113
1114 if (opt->extended)
1115 return match_expr(opt, bol, eol, ctx, col, icol,
1116 collect_hits);
1117
1118 /* we do not call with collect_hits without being extended */
1119 for (p = opt->pattern_list; p; p = p->next) {
1120 regmatch_t tmp;
1121 if (match_one_pattern(p, bol, eol, ctx, &tmp, 0)) {
1122 hit |= 1;
1123 if (!opt->columnnum) {
1124 /*
1125 * Without --column, any single match on a line
1126 * is enough to know that it needs to be
1127 * printed. With --column, scan _all_ patterns
1128 * to find the earliest.
1129 */
1130 break;
1131 }
1132 if (*col < 0 || tmp.rm_so < *col)
1133 *col = tmp.rm_so;
1134 }
1135 }
1136 return hit;
1137 }
1138
1139 static int match_next_pattern(struct grep_pat *p, char *bol, char *eol,
1140 enum grep_context ctx,
1141 regmatch_t *pmatch, int eflags)
1142 {
1143 regmatch_t match;
1144
1145 if (!match_one_pattern(p, bol, eol, ctx, &match, eflags))
1146 return 0;
1147 if (match.rm_so < 0 || match.rm_eo < 0)
1148 return 0;
1149 if (pmatch->rm_so >= 0 && pmatch->rm_eo >= 0) {
1150 if (match.rm_so > pmatch->rm_so)
1151 return 1;
1152 if (match.rm_so == pmatch->rm_so && match.rm_eo < pmatch->rm_eo)
1153 return 1;
1154 }
1155 pmatch->rm_so = match.rm_so;
1156 pmatch->rm_eo = match.rm_eo;
1157 return 1;
1158 }
1159
1160 static int next_match(struct grep_opt *opt, char *bol, char *eol,
1161 enum grep_context ctx, regmatch_t *pmatch, int eflags)
1162 {
1163 struct grep_pat *p;
1164 int hit = 0;
1165
1166 pmatch->rm_so = pmatch->rm_eo = -1;
1167 if (bol < eol) {
1168 for (p = opt->pattern_list; p; p = p->next) {
1169 switch (p->token) {
1170 case GREP_PATTERN: /* atom */
1171 case GREP_PATTERN_HEAD:
1172 case GREP_PATTERN_BODY:
1173 hit |= match_next_pattern(p, bol, eol, ctx,
1174 pmatch, eflags);
1175 break;
1176 default:
1177 break;
1178 }
1179 }
1180 }
1181 return hit;
1182 }
1183
1184 static void show_line_header(struct grep_opt *opt, const char *name,
1185 unsigned lno, ssize_t cno, char sign)
1186 {
1187 if (opt->heading && opt->last_shown == 0) {
1188 output_color(opt, name, strlen(name), opt->colors[GREP_COLOR_FILENAME]);
1189 opt->output(opt, "\n", 1);
1190 }
1191 opt->last_shown = lno;
1192
1193 if (!opt->heading && opt->pathname) {
1194 output_color(opt, name, strlen(name), opt->colors[GREP_COLOR_FILENAME]);
1195 output_sep(opt, sign);
1196 }
1197 if (opt->linenum) {
1198 char buf[32];
1199 xsnprintf(buf, sizeof(buf), "%d", lno);
1200 output_color(opt, buf, strlen(buf), opt->colors[GREP_COLOR_LINENO]);
1201 output_sep(opt, sign);
1202 }
1203 /*
1204 * Treat 'cno' as the 1-indexed offset from the start of a non-context
1205 * line to its first match. Otherwise, 'cno' is 0 indicating that we are
1206 * being called with a context line.
1207 */
1208 if (opt->columnnum && cno) {
1209 char buf[32];
1210 xsnprintf(buf, sizeof(buf), "%"PRIuMAX, (uintmax_t)cno);
1211 output_color(opt, buf, strlen(buf), opt->colors[GREP_COLOR_COLUMNNO]);
1212 output_sep(opt, sign);
1213 }
1214 }
1215
1216 static void show_line(struct grep_opt *opt, char *bol, char *eol,
1217 const char *name, unsigned lno, ssize_t cno, char sign)
1218 {
1219 int rest = eol - bol;
1220 const char *match_color = NULL;
1221 const char *line_color = NULL;
1222
1223 if (opt->file_break && opt->last_shown == 0) {
1224 if (opt->show_hunk_mark)
1225 opt->output(opt, "\n", 1);
1226 } else if (opt->pre_context || opt->post_context || opt->funcbody) {
1227 if (opt->last_shown == 0) {
1228 if (opt->show_hunk_mark) {
1229 output_color(opt, "--", 2, opt->colors[GREP_COLOR_SEP]);
1230 opt->output(opt, "\n", 1);
1231 }
1232 } else if (lno > opt->last_shown + 1) {
1233 output_color(opt, "--", 2, opt->colors[GREP_COLOR_SEP]);
1234 opt->output(opt, "\n", 1);
1235 }
1236 }
1237 if (!opt->only_matching) {
1238 /*
1239 * In case the line we're being called with contains more than
1240 * one match, leave printing each header to the loop below.
1241 */
1242 show_line_header(opt, name, lno, cno, sign);
1243 }
1244 if (opt->color || opt->only_matching) {
1245 regmatch_t match;
1246 enum grep_context ctx = GREP_CONTEXT_BODY;
1247 int ch = *eol;
1248 int eflags = 0;
1249
1250 if (opt->color) {
1251 if (sign == ':')
1252 match_color = opt->colors[GREP_COLOR_MATCH_SELECTED];
1253 else
1254 match_color = opt->colors[GREP_COLOR_MATCH_CONTEXT];
1255 if (sign == ':')
1256 line_color = opt->colors[GREP_COLOR_SELECTED];
1257 else if (sign == '-')
1258 line_color = opt->colors[GREP_COLOR_CONTEXT];
1259 else if (sign == '=')
1260 line_color = opt->colors[GREP_COLOR_FUNCTION];
1261 }
1262 *eol = '\0';
1263 while (next_match(opt, bol, eol, ctx, &match, eflags)) {
1264 if (match.rm_so == match.rm_eo)
1265 break;
1266
1267 if (opt->only_matching)
1268 show_line_header(opt, name, lno, cno, sign);
1269 else
1270 output_color(opt, bol, match.rm_so, line_color);
1271 output_color(opt, bol + match.rm_so,
1272 match.rm_eo - match.rm_so, match_color);
1273 if (opt->only_matching)
1274 opt->output(opt, "\n", 1);
1275 bol += match.rm_eo;
1276 cno += match.rm_eo;
1277 rest -= match.rm_eo;
1278 eflags = REG_NOTBOL;
1279 }
1280 *eol = ch;
1281 }
1282 if (!opt->only_matching) {
1283 output_color(opt, bol, rest, line_color);
1284 opt->output(opt, "\n", 1);
1285 }
1286 }
1287
1288 int grep_use_locks;
1289
1290 /*
1291 * This lock protects access to the gitattributes machinery, which is
1292 * not thread-safe.
1293 */
1294 pthread_mutex_t grep_attr_mutex;
1295
1296 static inline void grep_attr_lock(void)
1297 {
1298 if (grep_use_locks)
1299 pthread_mutex_lock(&grep_attr_mutex);
1300 }
1301
1302 static inline void grep_attr_unlock(void)
1303 {
1304 if (grep_use_locks)
1305 pthread_mutex_unlock(&grep_attr_mutex);
1306 }
1307
1308 static int match_funcname(struct grep_opt *opt, struct grep_source *gs, char *bol, char *eol)
1309 {
1310 xdemitconf_t *xecfg = opt->priv;
1311 if (xecfg && !xecfg->find_func) {
1312 grep_source_load_driver(gs, opt->repo->index);
1313 if (gs->driver->funcname.pattern) {
1314 const struct userdiff_funcname *pe = &gs->driver->funcname;
1315 xdiff_set_find_func(xecfg, pe->pattern, pe->cflags);
1316 } else {
1317 xecfg = opt->priv = NULL;
1318 }
1319 }
1320
1321 if (xecfg) {
1322 char buf[1];
1323 return xecfg->find_func(bol, eol - bol, buf, 1,
1324 xecfg->find_func_priv) >= 0;
1325 }
1326
1327 if (bol == eol)
1328 return 0;
1329 if (isalpha(*bol) || *bol == '_' || *bol == '$')
1330 return 1;
1331 return 0;
1332 }
1333
1334 static void show_funcname_line(struct grep_opt *opt, struct grep_source *gs,
1335 char *bol, unsigned lno)
1336 {
1337 while (bol > gs->buf) {
1338 char *eol = --bol;
1339
1340 while (bol > gs->buf && bol[-1] != '\n')
1341 bol--;
1342 lno--;
1343
1344 if (lno <= opt->last_shown)
1345 break;
1346
1347 if (match_funcname(opt, gs, bol, eol)) {
1348 show_line(opt, bol, eol, gs->name, lno, 0, '=');
1349 break;
1350 }
1351 }
1352 }
1353
1354 static int is_empty_line(const char *bol, const char *eol);
1355
1356 static void show_pre_context(struct grep_opt *opt, struct grep_source *gs,
1357 char *bol, char *end, unsigned lno)
1358 {
1359 unsigned cur = lno, from = 1, funcname_lno = 0, orig_from;
1360 int funcname_needed = !!opt->funcname, comment_needed = 0;
1361
1362 if (opt->pre_context < lno)
1363 from = lno - opt->pre_context;
1364 if (from <= opt->last_shown)
1365 from = opt->last_shown + 1;
1366 orig_from = from;
1367 if (opt->funcbody) {
1368 if (match_funcname(opt, gs, bol, end))
1369 comment_needed = 1;
1370 else
1371 funcname_needed = 1;
1372 from = opt->last_shown + 1;
1373 }
1374
1375 /* Rewind. */
1376 while (bol > gs->buf && cur > from) {
1377 char *next_bol = bol;
1378 char *eol = --bol;
1379
1380 while (bol > gs->buf && bol[-1] != '\n')
1381 bol--;
1382 cur--;
1383 if (comment_needed && (is_empty_line(bol, eol) ||
1384 match_funcname(opt, gs, bol, eol))) {
1385 comment_needed = 0;
1386 from = orig_from;
1387 if (cur < from) {
1388 cur++;
1389 bol = next_bol;
1390 break;
1391 }
1392 }
1393 if (funcname_needed && match_funcname(opt, gs, bol, eol)) {
1394 funcname_lno = cur;
1395 funcname_needed = 0;
1396 if (opt->funcbody)
1397 comment_needed = 1;
1398 else
1399 from = orig_from;
1400 }
1401 }
1402
1403 /* We need to look even further back to find a function signature. */
1404 if (opt->funcname && funcname_needed)
1405 show_funcname_line(opt, gs, bol, cur);
1406
1407 /* Back forward. */
1408 while (cur < lno) {
1409 char *eol = bol, sign = (cur == funcname_lno) ? '=' : '-';
1410
1411 while (*eol != '\n')
1412 eol++;
1413 show_line(opt, bol, eol, gs->name, cur, 0, sign);
1414 bol = eol + 1;
1415 cur++;
1416 }
1417 }
1418
1419 static int should_lookahead(struct grep_opt *opt)
1420 {
1421 struct grep_pat *p;
1422
1423 if (opt->extended)
1424 return 0; /* punt for too complex stuff */
1425 if (opt->invert)
1426 return 0;
1427 for (p = opt->pattern_list; p; p = p->next) {
1428 if (p->token != GREP_PATTERN)
1429 return 0; /* punt for "header only" and stuff */
1430 }
1431 return 1;
1432 }
1433
1434 static int look_ahead(struct grep_opt *opt,
1435 unsigned long *left_p,
1436 unsigned *lno_p,
1437 char **bol_p)
1438 {
1439 unsigned lno = *lno_p;
1440 char *bol = *bol_p;
1441 struct grep_pat *p;
1442 char *sp, *last_bol;
1443 regoff_t earliest = -1;
1444
1445 for (p = opt->pattern_list; p; p = p->next) {
1446 int hit;
1447 regmatch_t m;
1448
1449 hit = patmatch(p, bol, bol + *left_p, &m, 0);
1450 if (!hit || m.rm_so < 0 || m.rm_eo < 0)
1451 continue;
1452 if (earliest < 0 || m.rm_so < earliest)
1453 earliest = m.rm_so;
1454 }
1455
1456 if (earliest < 0) {
1457 *bol_p = bol + *left_p;
1458 *left_p = 0;
1459 return 1;
1460 }
1461 for (sp = bol + earliest; bol < sp && sp[-1] != '\n'; sp--)
1462 ; /* find the beginning of the line */
1463 last_bol = sp;
1464
1465 for (sp = bol; sp < last_bol; sp++) {
1466 if (*sp == '\n')
1467 lno++;
1468 }
1469 *left_p -= last_bol - bol;
1470 *bol_p = last_bol;
1471 *lno_p = lno;
1472 return 0;
1473 }
1474
1475 static int fill_textconv_grep(struct repository *r,
1476 struct userdiff_driver *driver,
1477 struct grep_source *gs)
1478 {
1479 struct diff_filespec *df;
1480 char *buf;
1481 size_t size;
1482
1483 if (!driver || !driver->textconv)
1484 return grep_source_load(gs);
1485
1486 /*
1487 * The textconv interface is intimately tied to diff_filespecs, so we
1488 * have to pretend to be one. If we could unify the grep_source
1489 * and diff_filespec structs, this mess could just go away.
1490 */
1491 df = alloc_filespec(gs->path);
1492 switch (gs->type) {
1493 case GREP_SOURCE_OID:
1494 fill_filespec(df, gs->identifier, 1, 0100644);
1495 break;
1496 case GREP_SOURCE_FILE:
1497 fill_filespec(df, null_oid(), 0, 0100644);
1498 break;
1499 default:
1500 BUG("attempt to textconv something without a path?");
1501 }
1502
1503 /*
1504 * fill_textconv is not remotely thread-safe; it modifies the global
1505 * diff tempfile structure, writes to the_repo's odb and might
1506 * internally call thread-unsafe functions such as the
1507 * prepare_packed_git() lazy-initializator. Because of the last two, we
1508 * must ensure mutual exclusion between this call and the object reading
1509 * API, thus we use obj_read_lock() here.
1510 *
1511 * TODO: allowing text conversion to run in parallel with object
1512 * reading operations might increase performance in the multithreaded
1513 * non-worktreee git-grep with --textconv.
1514 */
1515 obj_read_lock();
1516 size = fill_textconv(r, driver, df, &buf);
1517 obj_read_unlock();
1518 free_filespec(df);
1519
1520 /*
1521 * The normal fill_textconv usage by the diff machinery would just keep
1522 * the textconv'd buf separate from the diff_filespec. But much of the
1523 * grep code passes around a grep_source and assumes that its "buf"
1524 * pointer is the beginning of the thing we are searching. So let's
1525 * install our textconv'd version into the grep_source, taking care not
1526 * to leak any existing buffer.
1527 */
1528 grep_source_clear_data(gs);
1529 gs->buf = buf;
1530 gs->size = size;
1531
1532 return 0;
1533 }
1534
1535 static int is_empty_line(const char *bol, const char *eol)
1536 {
1537 while (bol < eol && isspace(*bol))
1538 bol++;
1539 return bol == eol;
1540 }
1541
1542 static int grep_source_1(struct grep_opt *opt, struct grep_source *gs, int collect_hits)
1543 {
1544 char *bol;
1545 char *peek_bol = NULL;
1546 unsigned long left;
1547 unsigned lno = 1;
1548 unsigned last_hit = 0;
1549 int binary_match_only = 0;
1550 unsigned count = 0;
1551 int try_lookahead = 0;
1552 int show_function = 0;
1553 struct userdiff_driver *textconv = NULL;
1554 enum grep_context ctx = GREP_CONTEXT_HEAD;
1555 xdemitconf_t xecfg;
1556
1557 if (!opt->status_only && gs->name == NULL)
1558 BUG("grep call which could print a name requires "
1559 "grep_source.name be non-NULL");
1560
1561 if (!opt->output)
1562 opt->output = std_output;
1563
1564 if (opt->pre_context || opt->post_context || opt->file_break ||
1565 opt->funcbody) {
1566 /* Show hunk marks, except for the first file. */
1567 if (opt->last_shown)
1568 opt->show_hunk_mark = 1;
1569 /*
1570 * If we're using threads then we can't easily identify
1571 * the first file. Always put hunk marks in that case
1572 * and skip the very first one later in work_done().
1573 */
1574 if (opt->output != std_output)
1575 opt->show_hunk_mark = 1;
1576 }
1577 opt->last_shown = 0;
1578
1579 if (opt->allow_textconv) {
1580 grep_source_load_driver(gs, opt->repo->index);
1581 /*
1582 * We might set up the shared textconv cache data here, which
1583 * is not thread-safe. Also, get_oid_with_context() and
1584 * parse_object() might be internally called. As they are not
1585 * currently thread-safe and might be racy with object reading,
1586 * obj_read_lock() must be called.
1587 */
1588 grep_attr_lock();
1589 obj_read_lock();
1590 textconv = userdiff_get_textconv(opt->repo, gs->driver);
1591 obj_read_unlock();
1592 grep_attr_unlock();
1593 }
1594
1595 /*
1596 * We know the result of a textconv is text, so we only have to care
1597 * about binary handling if we are not using it.
1598 */
1599 if (!textconv) {
1600 switch (opt->binary) {
1601 case GREP_BINARY_DEFAULT:
1602 if (grep_source_is_binary(gs, opt->repo->index))
1603 binary_match_only = 1;
1604 break;
1605 case GREP_BINARY_NOMATCH:
1606 if (grep_source_is_binary(gs, opt->repo->index))
1607 return 0; /* Assume unmatch */
1608 break;
1609 case GREP_BINARY_TEXT:
1610 break;
1611 default:
1612 BUG("unknown binary handling mode");
1613 }
1614 }
1615
1616 memset(&xecfg, 0, sizeof(xecfg));
1617 opt->priv = &xecfg;
1618
1619 try_lookahead = should_lookahead(opt);
1620
1621 if (fill_textconv_grep(opt->repo, textconv, gs) < 0)
1622 return 0;
1623
1624 bol = gs->buf;
1625 left = gs->size;
1626 while (left) {
1627 char *eol, ch;
1628 int hit;
1629 ssize_t cno;
1630 ssize_t col = -1, icol = -1;
1631
1632 /*
1633 * look_ahead() skips quickly to the line that possibly
1634 * has the next hit; don't call it if we need to do
1635 * something more than just skipping the current line
1636 * in response to an unmatch for the current line. E.g.
1637 * inside a post-context window, we will show the current
1638 * line as a context around the previous hit when it
1639 * doesn't hit.
1640 */
1641 if (try_lookahead
1642 && !(last_hit
1643 && (show_function ||
1644 lno <= last_hit + opt->post_context))
1645 && look_ahead(opt, &left, &lno, &bol))
1646 break;
1647 eol = end_of_line(bol, &left);
1648 ch = *eol;
1649 *eol = 0;
1650
1651 if ((ctx == GREP_CONTEXT_HEAD) && (eol == bol))
1652 ctx = GREP_CONTEXT_BODY;
1653
1654 hit = match_line(opt, bol, eol, &col, &icol, ctx, collect_hits);
1655 *eol = ch;
1656
1657 if (collect_hits)
1658 goto next_line;
1659
1660 /* "grep -v -e foo -e bla" should list lines
1661 * that do not have either, so inversion should
1662 * be done outside.
1663 */
1664 if (opt->invert)
1665 hit = !hit;
1666 if (opt->unmatch_name_only) {
1667 if (hit)
1668 return 0;
1669 goto next_line;
1670 }
1671 if (hit) {
1672 count++;
1673 if (opt->status_only)
1674 return 1;
1675 if (opt->name_only) {
1676 show_name(opt, gs->name);
1677 return 1;
1678 }
1679 if (opt->count)
1680 goto next_line;
1681 if (binary_match_only) {
1682 opt->output(opt, "Binary file ", 12);
1683 output_color(opt, gs->name, strlen(gs->name),
1684 opt->colors[GREP_COLOR_FILENAME]);
1685 opt->output(opt, " matches\n", 9);
1686 return 1;
1687 }
1688 /* Hit at this line. If we haven't shown the
1689 * pre-context lines, we would need to show them.
1690 */
1691 if (opt->pre_context || opt->funcbody)
1692 show_pre_context(opt, gs, bol, eol, lno);
1693 else if (opt->funcname)
1694 show_funcname_line(opt, gs, bol, lno);
1695 cno = opt->invert ? icol : col;
1696 if (cno < 0) {
1697 /*
1698 * A negative cno indicates that there was no
1699 * match on the line. We are thus inverted and
1700 * being asked to show all lines that _don't_
1701 * match a given expression. Therefore, set cno
1702 * to 0 to suggest the whole line matches.
1703 */
1704 cno = 0;
1705 }
1706 show_line(opt, bol, eol, gs->name, lno, cno + 1, ':');
1707 last_hit = lno;
1708 if (opt->funcbody)
1709 show_function = 1;
1710 goto next_line;
1711 }
1712 if (show_function && (!peek_bol || peek_bol < bol)) {
1713 unsigned long peek_left = left;
1714 char *peek_eol = eol;
1715
1716 /*
1717 * Trailing empty lines are not interesting.
1718 * Peek past them to see if they belong to the
1719 * body of the current function.
1720 */
1721 peek_bol = bol;
1722 while (is_empty_line(peek_bol, peek_eol)) {
1723 peek_bol = peek_eol + 1;
1724 peek_eol = end_of_line(peek_bol, &peek_left);
1725 }
1726
1727 if (match_funcname(opt, gs, peek_bol, peek_eol))
1728 show_function = 0;
1729 }
1730 if (show_function ||
1731 (last_hit && lno <= last_hit + opt->post_context)) {
1732 /* If the last hit is within the post context,
1733 * we need to show this line.
1734 */
1735 show_line(opt, bol, eol, gs->name, lno, col + 1, '-');
1736 }
1737
1738 next_line:
1739 bol = eol + 1;
1740 if (!left)
1741 break;
1742 left--;
1743 lno++;
1744 }
1745
1746 if (collect_hits)
1747 return 0;
1748
1749 if (opt->status_only)
1750 return opt->unmatch_name_only;
1751 if (opt->unmatch_name_only) {
1752 /* We did not see any hit, so we want to show this */
1753 show_name(opt, gs->name);
1754 return 1;
1755 }
1756
1757 xdiff_clear_find_func(&xecfg);
1758 opt->priv = NULL;
1759
1760 /* NEEDSWORK:
1761 * The real "grep -c foo *.c" gives many "bar.c:0" lines,
1762 * which feels mostly useless but sometimes useful. Maybe
1763 * make it another option? For now suppress them.
1764 */
1765 if (opt->count && count) {
1766 char buf[32];
1767 if (opt->pathname) {
1768 output_color(opt, gs->name, strlen(gs->name),
1769 opt->colors[GREP_COLOR_FILENAME]);
1770 output_sep(opt, ':');
1771 }
1772 xsnprintf(buf, sizeof(buf), "%u\n", count);
1773 opt->output(opt, buf, strlen(buf));
1774 return 1;
1775 }
1776 return !!last_hit;
1777 }
1778
1779 static void clr_hit_marker(struct grep_expr *x)
1780 {
1781 /* All-hit markers are meaningful only at the very top level
1782 * OR node.
1783 */
1784 while (1) {
1785 x->hit = 0;
1786 if (x->node != GREP_NODE_OR)
1787 return;
1788 x->u.binary.left->hit = 0;
1789 x = x->u.binary.right;
1790 }
1791 }
1792
1793 static int chk_hit_marker(struct grep_expr *x)
1794 {
1795 /* Top level nodes have hit markers. See if they all are hits */
1796 while (1) {
1797 if (x->node != GREP_NODE_OR)
1798 return x->hit;
1799 if (!x->u.binary.left->hit)
1800 return 0;
1801 x = x->u.binary.right;
1802 }
1803 }
1804
1805 int grep_source(struct grep_opt *opt, struct grep_source *gs)
1806 {
1807 /*
1808 * we do not have to do the two-pass grep when we do not check
1809 * buffer-wide "all-match".
1810 */
1811 if (!opt->all_match)
1812 return grep_source_1(opt, gs, 0);
1813
1814 /* Otherwise the toplevel "or" terms hit a bit differently.
1815 * We first clear hit markers from them.
1816 */
1817 clr_hit_marker(opt->pattern_expression);
1818 grep_source_1(opt, gs, 1);
1819
1820 if (!chk_hit_marker(opt->pattern_expression))
1821 return 0;
1822
1823 return grep_source_1(opt, gs, 0);
1824 }
1825
1826 int grep_buffer(struct grep_opt *opt, char *buf, unsigned long size)
1827 {
1828 struct grep_source gs;
1829 int r;
1830
1831 grep_source_init(&gs, GREP_SOURCE_BUF, NULL, NULL, NULL);
1832 gs.buf = buf;
1833 gs.size = size;
1834
1835 r = grep_source(opt, &gs);
1836
1837 grep_source_clear(&gs);
1838 return r;
1839 }
1840
1841 void grep_source_init(struct grep_source *gs, enum grep_source_type type,
1842 const char *name, const char *path,
1843 const void *identifier)
1844 {
1845 gs->type = type;
1846 gs->name = xstrdup_or_null(name);
1847 gs->path = xstrdup_or_null(path);
1848 gs->buf = NULL;
1849 gs->size = 0;
1850 gs->driver = NULL;
1851
1852 switch (type) {
1853 case GREP_SOURCE_FILE:
1854 gs->identifier = xstrdup(identifier);
1855 break;
1856 case GREP_SOURCE_OID:
1857 gs->identifier = oiddup(identifier);
1858 break;
1859 case GREP_SOURCE_BUF:
1860 gs->identifier = NULL;
1861 break;
1862 }
1863 }
1864
1865 void grep_source_clear(struct grep_source *gs)
1866 {
1867 FREE_AND_NULL(gs->name);
1868 FREE_AND_NULL(gs->path);
1869 FREE_AND_NULL(gs->identifier);
1870 grep_source_clear_data(gs);
1871 }
1872
1873 void grep_source_clear_data(struct grep_source *gs)
1874 {
1875 switch (gs->type) {
1876 case GREP_SOURCE_FILE:
1877 case GREP_SOURCE_OID:
1878 FREE_AND_NULL(gs->buf);
1879 gs->size = 0;
1880 break;
1881 case GREP_SOURCE_BUF:
1882 /* leave user-provided buf intact */
1883 break;
1884 }
1885 }
1886
1887 static int grep_source_load_oid(struct grep_source *gs)
1888 {
1889 enum object_type type;
1890
1891 gs->buf = read_object_file(gs->identifier, &type, &gs->size);
1892 if (!gs->buf)
1893 return error(_("'%s': unable to read %s"),
1894 gs->name,
1895 oid_to_hex(gs->identifier));
1896 return 0;
1897 }
1898
1899 static int grep_source_load_file(struct grep_source *gs)
1900 {
1901 const char *filename = gs->identifier;
1902 struct stat st;
1903 char *data;
1904 size_t size;
1905 int i;
1906
1907 if (lstat(filename, &st) < 0) {
1908 err_ret:
1909 if (errno != ENOENT)
1910 error_errno(_("failed to stat '%s'"), filename);
1911 return -1;
1912 }
1913 if (!S_ISREG(st.st_mode))
1914 return -1;
1915 size = xsize_t(st.st_size);
1916 i = open(filename, O_RDONLY);
1917 if (i < 0)
1918 goto err_ret;
1919 data = xmallocz(size);
1920 if (st.st_size != read_in_full(i, data, size)) {
1921 error_errno(_("'%s': short read"), filename);
1922 close(i);
1923 free(data);
1924 return -1;
1925 }
1926 close(i);
1927
1928 gs->buf = data;
1929 gs->size = size;
1930 return 0;
1931 }
1932
1933 static int grep_source_load(struct grep_source *gs)
1934 {
1935 if (gs->buf)
1936 return 0;
1937
1938 switch (gs->type) {
1939 case GREP_SOURCE_FILE:
1940 return grep_source_load_file(gs);
1941 case GREP_SOURCE_OID:
1942 return grep_source_load_oid(gs);
1943 case GREP_SOURCE_BUF:
1944 return gs->buf ? 0 : -1;
1945 }
1946 BUG("invalid grep_source type to load");
1947 }
1948
1949 void grep_source_load_driver(struct grep_source *gs,
1950 struct index_state *istate)
1951 {
1952 if (gs->driver)
1953 return;
1954
1955 grep_attr_lock();
1956 if (gs->path)
1957 gs->driver = userdiff_find_by_path(istate, gs->path);
1958 if (!gs->driver)
1959 gs->driver = userdiff_find_by_name("default");
1960 grep_attr_unlock();
1961 }
1962
1963 static int grep_source_is_binary(struct grep_source *gs,
1964 struct index_state *istate)
1965 {
1966 grep_source_load_driver(gs, istate);
1967 if (gs->driver->binary != -1)
1968 return gs->driver->binary;
1969
1970 if (!grep_source_load(gs))
1971 return buffer_is_binary(gs->buf, gs->size);
1972
1973 return 0;
1974 }
The Best VCS Around