]> git.ipfire.org Git - thirdparty/git.git/blob - grep.c
grep: work around UTF-8 related JIT bug in PCRE2 <= 10.34
[thirdparty/git.git] / grep.c
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 const char *color_grep_slots[] = {
23 [GREP_COLOR_CONTEXT] = "context",
24 [GREP_COLOR_FILENAME] = "filename",
25 [GREP_COLOR_FUNCTION] = "function",
26 [GREP_COLOR_LINENO] = "lineNumber",
27 [GREP_COLOR_COLUMNNO] = "column",
28 [GREP_COLOR_MATCH_CONTEXT] = "matchContext",
29 [GREP_COLOR_MATCH_SELECTED] = "matchSelected",
30 [GREP_COLOR_SELECTED] = "selected",
31 [GREP_COLOR_SEP] = "separator",
32 };
33
34 static int parse_pattern_type_arg(const char *opt, const char *arg)
35 {
36 if (!strcmp(arg, "default"))
37 return GREP_PATTERN_TYPE_UNSPECIFIED;
38 else if (!strcmp(arg, "basic"))
39 return GREP_PATTERN_TYPE_BRE;
40 else if (!strcmp(arg, "extended"))
41 return GREP_PATTERN_TYPE_ERE;
42 else if (!strcmp(arg, "fixed"))
43 return GREP_PATTERN_TYPE_FIXED;
44 else if (!strcmp(arg, "perl"))
45 return GREP_PATTERN_TYPE_PCRE;
46 die("bad %s argument: %s", opt, arg);
47 }
48
49 define_list_config_array_extra(color_grep_slots, {"match"});
50
51 /*
52 * Read the configuration file once and store it in
53 * the grep_defaults template.
54 */
55 int grep_config(const char *var, const char *value, void *cb)
56 {
57 struct grep_opt *opt = cb;
58 const char *slot;
59
60 if (userdiff_config(var, value) < 0)
61 return -1;
62
63 if (!strcmp(var, "grep.extendedregexp")) {
64 opt->extended_regexp_option = git_config_bool(var, value);
65 return 0;
66 }
67
68 if (!strcmp(var, "grep.patterntype")) {
69 opt->pattern_type_option = parse_pattern_type_arg(var, value);
70 return 0;
71 }
72
73 if (!strcmp(var, "grep.linenumber")) {
74 opt->linenum = git_config_bool(var, value);
75 return 0;
76 }
77 if (!strcmp(var, "grep.column")) {
78 opt->columnnum = git_config_bool(var, value);
79 return 0;
80 }
81
82 if (!strcmp(var, "grep.fullname")) {
83 opt->relative = !git_config_bool(var, value);
84 return 0;
85 }
86
87 if (!strcmp(var, "color.grep"))
88 opt->color = git_config_colorbool(var, value);
89 if (!strcmp(var, "color.grep.match")) {
90 if (grep_config("color.grep.matchcontext", value, cb) < 0)
91 return -1;
92 if (grep_config("color.grep.matchselected", value, cb) < 0)
93 return -1;
94 } else if (skip_prefix(var, "color.grep.", &slot)) {
95 int i = LOOKUP_CONFIG(color_grep_slots, slot);
96 char *color;
97
98 if (i < 0)
99 return -1;
100 color = opt->colors[i];
101 if (!value)
102 return config_error_nonbool(var);
103 return color_parse(value, color);
104 }
105 return 0;
106 }
107
108 void grep_init(struct grep_opt *opt, struct repository *repo)
109 {
110 struct grep_opt blank = GREP_OPT_INIT;
111 memcpy(opt, &blank, sizeof(*opt));
112
113 opt->repo = repo;
114 opt->pattern_tail = &opt->pattern_list;
115 opt->header_tail = &opt->header_list;
116 }
117
118 static struct grep_pat *create_grep_pat(const char *pat, size_t patlen,
119 const char *origin, int no,
120 enum grep_pat_token t,
121 enum grep_header_field field)
122 {
123 struct grep_pat *p = xcalloc(1, sizeof(*p));
124 p->pattern = xmemdupz(pat, patlen);
125 p->patternlen = patlen;
126 p->origin = origin;
127 p->no = no;
128 p->token = t;
129 p->field = field;
130 return p;
131 }
132
133 static void do_append_grep_pat(struct grep_pat ***tail, struct grep_pat *p)
134 {
135 **tail = p;
136 *tail = &p->next;
137 p->next = NULL;
138
139 switch (p->token) {
140 case GREP_PATTERN: /* atom */
141 case GREP_PATTERN_HEAD:
142 case GREP_PATTERN_BODY:
143 for (;;) {
144 struct grep_pat *new_pat;
145 size_t len = 0;
146 char *cp = p->pattern + p->patternlen, *nl = NULL;
147 while (++len <= p->patternlen) {
148 if (*(--cp) == '\n') {
149 nl = cp;
150 break;
151 }
152 }
153 if (!nl)
154 break;
155 new_pat = create_grep_pat(nl + 1, len - 1, p->origin,
156 p->no, p->token, p->field);
157 new_pat->next = p->next;
158 if (!p->next)
159 *tail = &new_pat->next;
160 p->next = new_pat;
161 *nl = '\0';
162 p->patternlen -= len;
163 }
164 break;
165 default:
166 break;
167 }
168 }
169
170 void append_header_grep_pattern(struct grep_opt *opt,
171 enum grep_header_field field, const char *pat)
172 {
173 struct grep_pat *p = create_grep_pat(pat, strlen(pat), "header", 0,
174 GREP_PATTERN_HEAD, field);
175 if (field == GREP_HEADER_REFLOG)
176 opt->use_reflog_filter = 1;
177 do_append_grep_pat(&opt->header_tail, p);
178 }
179
180 void append_grep_pattern(struct grep_opt *opt, const char *pat,
181 const char *origin, int no, enum grep_pat_token t)
182 {
183 append_grep_pat(opt, pat, strlen(pat), origin, no, t);
184 }
185
186 void append_grep_pat(struct grep_opt *opt, const char *pat, size_t patlen,
187 const char *origin, int no, enum grep_pat_token t)
188 {
189 struct grep_pat *p = create_grep_pat(pat, patlen, origin, no, t, 0);
190 do_append_grep_pat(&opt->pattern_tail, p);
191 }
192
193 struct grep_opt *grep_opt_dup(const struct grep_opt *opt)
194 {
195 struct grep_pat *pat;
196 struct grep_opt *ret = xmalloc(sizeof(struct grep_opt));
197 *ret = *opt;
198
199 ret->pattern_list = NULL;
200 ret->pattern_tail = &ret->pattern_list;
201
202 for(pat = opt->pattern_list; pat != NULL; pat = pat->next)
203 {
204 if(pat->token == GREP_PATTERN_HEAD)
205 append_header_grep_pattern(ret, pat->field,
206 pat->pattern);
207 else
208 append_grep_pat(ret, pat->pattern, pat->patternlen,
209 pat->origin, pat->no, pat->token);
210 }
211
212 return ret;
213 }
214
215 static NORETURN void compile_regexp_failed(const struct grep_pat *p,
216 const char *error)
217 {
218 char where[1024];
219
220 if (p->no)
221 xsnprintf(where, sizeof(where), "In '%s' at %d, ", p->origin, p->no);
222 else if (p->origin)
223 xsnprintf(where, sizeof(where), "%s, ", p->origin);
224 else
225 where[0] = 0;
226
227 die("%s'%s': %s", where, p->pattern, error);
228 }
229
230 static int is_fixed(const char *s, size_t len)
231 {
232 size_t i;
233
234 for (i = 0; i < len; i++) {
235 if (is_regex_special(s[i]))
236 return 0;
237 }
238
239 return 1;
240 }
241
242 #ifdef USE_LIBPCRE2
243 #define GREP_PCRE2_DEBUG_MALLOC 0
244
245 static void *pcre2_malloc(PCRE2_SIZE size, MAYBE_UNUSED void *memory_data)
246 {
247 void *pointer = malloc(size);
248 #if GREP_PCRE2_DEBUG_MALLOC
249 static int count = 1;
250 fprintf(stderr, "PCRE2:%p -> #%02d: alloc(%lu)\n", pointer, count++, size);
251 #endif
252 return pointer;
253 }
254
255 static void pcre2_free(void *pointer, MAYBE_UNUSED void *memory_data)
256 {
257 #if GREP_PCRE2_DEBUG_MALLOC
258 static int count = 1;
259 if (pointer)
260 fprintf(stderr, "PCRE2:%p -> #%02d: free()\n", pointer, count++);
261 #endif
262 free(pointer);
263 }
264
265 static void compile_pcre2_pattern(struct grep_pat *p, const struct grep_opt *opt)
266 {
267 int error;
268 PCRE2_UCHAR errbuf[256];
269 PCRE2_SIZE erroffset;
270 int options = PCRE2_MULTILINE;
271 int jitret;
272 int patinforet;
273 size_t jitsizearg;
274 int literal = !opt->ignore_case && (p->fixed || p->is_fixed);
275
276 /*
277 * Call pcre2_general_context_create() before calling any
278 * other pcre2_*(). It sets up our malloc()/free() functions
279 * with which everything else is allocated.
280 */
281 p->pcre2_general_context = pcre2_general_context_create(
282 pcre2_malloc, pcre2_free, NULL);
283 if (!p->pcre2_general_context)
284 die("Couldn't allocate PCRE2 general context");
285
286 if (opt->ignore_case) {
287 if (!opt->ignore_locale && has_non_ascii(p->pattern)) {
288 p->pcre2_tables = pcre2_maketables(p->pcre2_general_context);
289 p->pcre2_compile_context = pcre2_compile_context_create(p->pcre2_general_context);
290 pcre2_set_character_tables(p->pcre2_compile_context,
291 p->pcre2_tables);
292 }
293 options |= PCRE2_CASELESS;
294 }
295 if (!opt->ignore_locale && is_utf8_locale() && !literal)
296 options |= (PCRE2_UTF | PCRE2_UCP | PCRE2_MATCH_INVALID_UTF);
297
298 #ifndef GIT_PCRE2_VERSION_10_35_OR_HIGHER
299 /*
300 * Work around a JIT bug related to invalid Unicode character handling
301 * fixed in 10.35:
302 * https://github.com/PCRE2Project/pcre2/commit/c21bd977547d
303 */
304 options &= ~PCRE2_UCP;
305 #endif
306
307 #ifndef GIT_PCRE2_VERSION_10_36_OR_HIGHER
308 /* Work around https://bugs.exim.org/show_bug.cgi?id=2642 fixed in 10.36 */
309 if (PCRE2_MATCH_INVALID_UTF && options & (PCRE2_UTF | PCRE2_CASELESS))
310 options |= PCRE2_NO_START_OPTIMIZE;
311 #endif
312
313 p->pcre2_pattern = pcre2_compile((PCRE2_SPTR)p->pattern,
314 p->patternlen, options, &error, &erroffset,
315 p->pcre2_compile_context);
316
317 if (p->pcre2_pattern) {
318 p->pcre2_match_data = pcre2_match_data_create_from_pattern(p->pcre2_pattern, p->pcre2_general_context);
319 if (!p->pcre2_match_data)
320 die("Couldn't allocate PCRE2 match data");
321 } else {
322 pcre2_get_error_message(error, errbuf, sizeof(errbuf));
323 compile_regexp_failed(p, (const char *)&errbuf);
324 }
325
326 pcre2_config(PCRE2_CONFIG_JIT, &p->pcre2_jit_on);
327 if (p->pcre2_jit_on) {
328 jitret = pcre2_jit_compile(p->pcre2_pattern, PCRE2_JIT_COMPLETE);
329 if (jitret)
330 die("Couldn't JIT the PCRE2 pattern '%s', got '%d'\n", p->pattern, jitret);
331
332 /*
333 * The pcre2_config(PCRE2_CONFIG_JIT, ...) call just
334 * tells us whether the library itself supports JIT,
335 * but to see whether we're going to be actually using
336 * JIT we need to extract PCRE2_INFO_JITSIZE from the
337 * pattern *after* we do pcre2_jit_compile() above.
338 *
339 * This is because if the pattern contains the
340 * (*NO_JIT) verb (see pcre2syntax(3))
341 * pcre2_jit_compile() will exit early with 0. If we
342 * then proceed to call pcre2_jit_match() further down
343 * the line instead of pcre2_match() we'll either
344 * segfault (pre PCRE 10.31) or run into a fatal error
345 * (post PCRE2 10.31)
346 */
347 patinforet = pcre2_pattern_info(p->pcre2_pattern, PCRE2_INFO_JITSIZE, &jitsizearg);
348 if (patinforet)
349 BUG("pcre2_pattern_info() failed: %d", patinforet);
350 if (jitsizearg == 0) {
351 p->pcre2_jit_on = 0;
352 return;
353 }
354 }
355 }
356
357 static int pcre2match(struct grep_pat *p, const char *line, const char *eol,
358 regmatch_t *match, int eflags)
359 {
360 int ret, flags = 0;
361 PCRE2_SIZE *ovector;
362 PCRE2_UCHAR errbuf[256];
363
364 if (eflags & REG_NOTBOL)
365 flags |= PCRE2_NOTBOL;
366
367 if (p->pcre2_jit_on)
368 ret = pcre2_jit_match(p->pcre2_pattern, (unsigned char *)line,
369 eol - line, 0, flags, p->pcre2_match_data,
370 NULL);
371 else
372 ret = pcre2_match(p->pcre2_pattern, (unsigned char *)line,
373 eol - line, 0, flags, p->pcre2_match_data,
374 NULL);
375
376 if (ret < 0 && ret != PCRE2_ERROR_NOMATCH) {
377 pcre2_get_error_message(ret, errbuf, sizeof(errbuf));
378 die("%s failed with error code %d: %s",
379 (p->pcre2_jit_on ? "pcre2_jit_match" : "pcre2_match"), ret,
380 errbuf);
381 }
382 if (ret > 0) {
383 ovector = pcre2_get_ovector_pointer(p->pcre2_match_data);
384 ret = 0;
385 match->rm_so = (int)ovector[0];
386 match->rm_eo = (int)ovector[1];
387 }
388
389 return ret;
390 }
391
392 static void free_pcre2_pattern(struct grep_pat *p)
393 {
394 pcre2_compile_context_free(p->pcre2_compile_context);
395 pcre2_code_free(p->pcre2_pattern);
396 pcre2_match_data_free(p->pcre2_match_data);
397 #ifdef GIT_PCRE2_VERSION_10_34_OR_HIGHER
398 pcre2_maketables_free(p->pcre2_general_context, p->pcre2_tables);
399 #else
400 free((void *)p->pcre2_tables);
401 #endif
402 pcre2_general_context_free(p->pcre2_general_context);
403 }
404 #else /* !USE_LIBPCRE2 */
405 static void compile_pcre2_pattern(struct grep_pat *p, const struct grep_opt *opt)
406 {
407 die("cannot use Perl-compatible regexes when not compiled with USE_LIBPCRE");
408 }
409
410 static int pcre2match(struct grep_pat *p, const char *line, const char *eol,
411 regmatch_t *match, int eflags)
412 {
413 return 1;
414 }
415
416 static void free_pcre2_pattern(struct grep_pat *p)
417 {
418 }
419
420 static void compile_fixed_regexp(struct grep_pat *p, struct grep_opt *opt)
421 {
422 struct strbuf sb = STRBUF_INIT;
423 int err;
424 int regflags = 0;
425
426 basic_regex_quote_buf(&sb, p->pattern);
427 if (opt->ignore_case)
428 regflags |= REG_ICASE;
429 err = regcomp(&p->regexp, sb.buf, regflags);
430 strbuf_release(&sb);
431 if (err) {
432 char errbuf[1024];
433 regerror(err, &p->regexp, errbuf, sizeof(errbuf));
434 compile_regexp_failed(p, errbuf);
435 }
436 }
437 #endif /* !USE_LIBPCRE2 */
438
439 static void compile_regexp(struct grep_pat *p, struct grep_opt *opt)
440 {
441 int err;
442 int regflags = REG_NEWLINE;
443
444 if (opt->pattern_type_option == GREP_PATTERN_TYPE_UNSPECIFIED)
445 opt->pattern_type_option = (opt->extended_regexp_option
446 ? GREP_PATTERN_TYPE_ERE
447 : GREP_PATTERN_TYPE_BRE);
448
449 p->word_regexp = opt->word_regexp;
450 p->ignore_case = opt->ignore_case;
451 p->fixed = opt->pattern_type_option == GREP_PATTERN_TYPE_FIXED;
452
453 if (opt->pattern_type_option != GREP_PATTERN_TYPE_PCRE &&
454 memchr(p->pattern, 0, p->patternlen))
455 die(_("given pattern contains NULL byte (via -f <file>). This is only supported with -P under PCRE v2"));
456
457 p->is_fixed = is_fixed(p->pattern, p->patternlen);
458 #ifdef USE_LIBPCRE2
459 if (!p->fixed && !p->is_fixed) {
460 const char *no_jit = "(*NO_JIT)";
461 const int no_jit_len = strlen(no_jit);
462 if (starts_with(p->pattern, no_jit) &&
463 is_fixed(p->pattern + no_jit_len,
464 p->patternlen - no_jit_len))
465 p->is_fixed = 1;
466 }
467 #endif
468 if (p->fixed || p->is_fixed) {
469 #ifdef USE_LIBPCRE2
470 if (p->is_fixed) {
471 compile_pcre2_pattern(p, opt);
472 } else {
473 /*
474 * E.g. t7811-grep-open.sh relies on the
475 * pattern being restored.
476 */
477 char *old_pattern = p->pattern;
478 size_t old_patternlen = p->patternlen;
479 struct strbuf sb = STRBUF_INIT;
480
481 /*
482 * There is the PCRE2_LITERAL flag, but it's
483 * only in PCRE v2 10.30 and later. Needing to
484 * ifdef our way around that and dealing with
485 * it + PCRE2_MULTILINE being an error is more
486 * complex than just quoting this ourselves.
487 */
488 strbuf_add(&sb, "\\Q", 2);
489 strbuf_add(&sb, p->pattern, p->patternlen);
490 strbuf_add(&sb, "\\E", 2);
491
492 p->pattern = sb.buf;
493 p->patternlen = sb.len;
494 compile_pcre2_pattern(p, opt);
495 p->pattern = old_pattern;
496 p->patternlen = old_patternlen;
497 strbuf_release(&sb);
498 }
499 #else /* !USE_LIBPCRE2 */
500 compile_fixed_regexp(p, opt);
501 #endif /* !USE_LIBPCRE2 */
502 return;
503 }
504
505 if (opt->pattern_type_option == GREP_PATTERN_TYPE_PCRE) {
506 compile_pcre2_pattern(p, opt);
507 return;
508 }
509
510 if (p->ignore_case)
511 regflags |= REG_ICASE;
512 if (opt->pattern_type_option == GREP_PATTERN_TYPE_ERE)
513 regflags |= REG_EXTENDED;
514 err = regcomp(&p->regexp, p->pattern, regflags);
515 if (err) {
516 char errbuf[1024];
517 regerror(err, &p->regexp, errbuf, 1024);
518 compile_regexp_failed(p, errbuf);
519 }
520 }
521
522 static struct grep_expr *grep_not_expr(struct grep_expr *expr)
523 {
524 struct grep_expr *z = xcalloc(1, sizeof(*z));
525 z->node = GREP_NODE_NOT;
526 z->u.unary = expr;
527 return z;
528 }
529
530 static struct grep_expr *grep_binexp(enum grep_expr_node kind,
531 struct grep_expr *left,
532 struct grep_expr *right)
533 {
534 struct grep_expr *z = xcalloc(1, sizeof(*z));
535 z->node = kind;
536 z->u.binary.left = left;
537 z->u.binary.right = right;
538 return z;
539 }
540
541 static struct grep_expr *grep_or_expr(struct grep_expr *left, struct grep_expr *right)
542 {
543 return grep_binexp(GREP_NODE_OR, left, right);
544 }
545
546 static struct grep_expr *grep_and_expr(struct grep_expr *left, struct grep_expr *right)
547 {
548 return grep_binexp(GREP_NODE_AND, left, right);
549 }
550
551 static struct grep_expr *compile_pattern_or(struct grep_pat **);
552 static struct grep_expr *compile_pattern_atom(struct grep_pat **list)
553 {
554 struct grep_pat *p;
555 struct grep_expr *x;
556
557 p = *list;
558 if (!p)
559 return NULL;
560 switch (p->token) {
561 case GREP_PATTERN: /* atom */
562 case GREP_PATTERN_HEAD:
563 case GREP_PATTERN_BODY:
564 CALLOC_ARRAY(x, 1);
565 x->node = GREP_NODE_ATOM;
566 x->u.atom = p;
567 *list = p->next;
568 return x;
569 case GREP_OPEN_PAREN:
570 *list = p->next;
571 x = compile_pattern_or(list);
572 if (!*list || (*list)->token != GREP_CLOSE_PAREN)
573 die("unmatched parenthesis");
574 *list = (*list)->next;
575 return x;
576 default:
577 return NULL;
578 }
579 }
580
581 static struct grep_expr *compile_pattern_not(struct grep_pat **list)
582 {
583 struct grep_pat *p;
584 struct grep_expr *x;
585
586 p = *list;
587 if (!p)
588 return NULL;
589 switch (p->token) {
590 case GREP_NOT:
591 if (!p->next)
592 die("--not not followed by pattern expression");
593 *list = p->next;
594 x = compile_pattern_not(list);
595 if (!x)
596 die("--not followed by non pattern expression");
597 return grep_not_expr(x);
598 default:
599 return compile_pattern_atom(list);
600 }
601 }
602
603 static struct grep_expr *compile_pattern_and(struct grep_pat **list)
604 {
605 struct grep_pat *p;
606 struct grep_expr *x, *y;
607
608 x = compile_pattern_not(list);
609 p = *list;
610 if (p && p->token == GREP_AND) {
611 if (!x)
612 die("--and not preceded by pattern expression");
613 if (!p->next)
614 die("--and not followed by pattern expression");
615 *list = p->next;
616 y = compile_pattern_and(list);
617 if (!y)
618 die("--and not followed by pattern expression");
619 return grep_and_expr(x, y);
620 }
621 return x;
622 }
623
624 static struct grep_expr *compile_pattern_or(struct grep_pat **list)
625 {
626 struct grep_pat *p;
627 struct grep_expr *x, *y;
628
629 x = compile_pattern_and(list);
630 p = *list;
631 if (x && p && p->token != GREP_CLOSE_PAREN) {
632 y = compile_pattern_or(list);
633 if (!y)
634 die("not a pattern expression %s", p->pattern);
635 return grep_or_expr(x, y);
636 }
637 return x;
638 }
639
640 static struct grep_expr *compile_pattern_expr(struct grep_pat **list)
641 {
642 return compile_pattern_or(list);
643 }
644
645 static struct grep_expr *grep_true_expr(void)
646 {
647 struct grep_expr *z = xcalloc(1, sizeof(*z));
648 z->node = GREP_NODE_TRUE;
649 return z;
650 }
651
652 static struct grep_expr *prep_header_patterns(struct grep_opt *opt)
653 {
654 struct grep_pat *p;
655 struct grep_expr *header_expr;
656 struct grep_expr *(header_group[GREP_HEADER_FIELD_MAX]);
657 enum grep_header_field fld;
658
659 if (!opt->header_list)
660 return NULL;
661
662 for (p = opt->header_list; p; p = p->next) {
663 if (p->token != GREP_PATTERN_HEAD)
664 BUG("a non-header pattern in grep header list.");
665 if (p->field < GREP_HEADER_FIELD_MIN ||
666 GREP_HEADER_FIELD_MAX <= p->field)
667 BUG("unknown header field %d", p->field);
668 compile_regexp(p, opt);
669 }
670
671 for (fld = 0; fld < GREP_HEADER_FIELD_MAX; fld++)
672 header_group[fld] = NULL;
673
674 for (p = opt->header_list; p; p = p->next) {
675 struct grep_expr *h;
676 struct grep_pat *pp = p;
677
678 h = compile_pattern_atom(&pp);
679 if (!h || pp != p->next)
680 BUG("malformed header expr");
681 if (!header_group[p->field]) {
682 header_group[p->field] = h;
683 continue;
684 }
685 header_group[p->field] = grep_or_expr(h, header_group[p->field]);
686 }
687
688 header_expr = NULL;
689
690 for (fld = 0; fld < GREP_HEADER_FIELD_MAX; fld++) {
691 if (!header_group[fld])
692 continue;
693 if (!header_expr)
694 header_expr = grep_true_expr();
695 header_expr = grep_or_expr(header_group[fld], header_expr);
696 }
697 return header_expr;
698 }
699
700 static struct grep_expr *grep_splice_or(struct grep_expr *x, struct grep_expr *y)
701 {
702 struct grep_expr *z = x;
703
704 while (x) {
705 assert(x->node == GREP_NODE_OR);
706 if (x->u.binary.right &&
707 x->u.binary.right->node == GREP_NODE_TRUE) {
708 x->u.binary.right = y;
709 break;
710 }
711 x = x->u.binary.right;
712 }
713 return z;
714 }
715
716 void compile_grep_patterns(struct grep_opt *opt)
717 {
718 struct grep_pat *p;
719 struct grep_expr *header_expr = prep_header_patterns(opt);
720 int extended = 0;
721
722 for (p = opt->pattern_list; p; p = p->next) {
723 switch (p->token) {
724 case GREP_PATTERN: /* atom */
725 case GREP_PATTERN_HEAD:
726 case GREP_PATTERN_BODY:
727 compile_regexp(p, opt);
728 break;
729 default:
730 extended = 1;
731 break;
732 }
733 }
734
735 if (opt->all_match || opt->no_body_match || header_expr)
736 extended = 1;
737 else if (!extended)
738 return;
739
740 p = opt->pattern_list;
741 if (p)
742 opt->pattern_expression = compile_pattern_expr(&p);
743 if (p)
744 die("incomplete pattern expression: %s", p->pattern);
745
746 if (opt->no_body_match && opt->pattern_expression)
747 opt->pattern_expression = grep_not_expr(opt->pattern_expression);
748
749 if (!header_expr)
750 return;
751
752 if (!opt->pattern_expression)
753 opt->pattern_expression = header_expr;
754 else if (opt->all_match)
755 opt->pattern_expression = grep_splice_or(header_expr,
756 opt->pattern_expression);
757 else
758 opt->pattern_expression = grep_or_expr(opt->pattern_expression,
759 header_expr);
760 opt->all_match = 1;
761 }
762
763 static void free_pattern_expr(struct grep_expr *x)
764 {
765 switch (x->node) {
766 case GREP_NODE_TRUE:
767 case GREP_NODE_ATOM:
768 break;
769 case GREP_NODE_NOT:
770 free_pattern_expr(x->u.unary);
771 break;
772 case GREP_NODE_AND:
773 case GREP_NODE_OR:
774 free_pattern_expr(x->u.binary.left);
775 free_pattern_expr(x->u.binary.right);
776 break;
777 }
778 free(x);
779 }
780
781 void free_grep_patterns(struct grep_opt *opt)
782 {
783 struct grep_pat *p, *n;
784
785 for (p = opt->pattern_list; p; p = n) {
786 n = p->next;
787 switch (p->token) {
788 case GREP_PATTERN: /* atom */
789 case GREP_PATTERN_HEAD:
790 case GREP_PATTERN_BODY:
791 if (p->pcre2_pattern)
792 free_pcre2_pattern(p);
793 else
794 regfree(&p->regexp);
795 free(p->pattern);
796 break;
797 default:
798 break;
799 }
800 free(p);
801 }
802
803 if (!opt->pattern_expression)
804 return;
805 free_pattern_expr(opt->pattern_expression);
806 }
807
808 static const char *end_of_line(const char *cp, unsigned long *left)
809 {
810 unsigned long l = *left;
811 while (l && *cp != '\n') {
812 l--;
813 cp++;
814 }
815 *left = l;
816 return cp;
817 }
818
819 static int word_char(char ch)
820 {
821 return isalnum(ch) || ch == '_';
822 }
823
824 static void output_color(struct grep_opt *opt, const void *data, size_t size,
825 const char *color)
826 {
827 if (want_color(opt->color) && color && color[0]) {
828 opt->output(opt, color, strlen(color));
829 opt->output(opt, data, size);
830 opt->output(opt, GIT_COLOR_RESET, strlen(GIT_COLOR_RESET));
831 } else
832 opt->output(opt, data, size);
833 }
834
835 static void output_sep(struct grep_opt *opt, char sign)
836 {
837 if (opt->null_following_name)
838 opt->output(opt, "\0", 1);
839 else
840 output_color(opt, &sign, 1, opt->colors[GREP_COLOR_SEP]);
841 }
842
843 static void show_name(struct grep_opt *opt, const char *name)
844 {
845 output_color(opt, name, strlen(name), opt->colors[GREP_COLOR_FILENAME]);
846 opt->output(opt, opt->null_following_name ? "\0" : "\n", 1);
847 }
848
849 static int patmatch(struct grep_pat *p,
850 const char *line, const char *eol,
851 regmatch_t *match, int eflags)
852 {
853 int hit;
854
855 if (p->pcre2_pattern)
856 hit = !pcre2match(p, line, eol, match, eflags);
857 else
858 hit = !regexec_buf(&p->regexp, line, eol - line, 1, match,
859 eflags);
860
861 return hit;
862 }
863
864 static void strip_timestamp(const char *bol, const char **eol_p)
865 {
866 const char *eol = *eol_p;
867
868 while (bol < --eol) {
869 if (*eol != '>')
870 continue;
871 *eol_p = ++eol;
872 break;
873 }
874 }
875
876 static struct {
877 const char *field;
878 size_t len;
879 } header_field[] = {
880 { "author ", 7 },
881 { "committer ", 10 },
882 { "reflog ", 7 },
883 };
884
885 static int headerless_match_one_pattern(struct grep_pat *p,
886 const char *bol, const char *eol,
887 enum grep_context ctx,
888 regmatch_t *pmatch, int eflags)
889 {
890 int hit = 0;
891 const char *start = bol;
892
893 if ((p->token != GREP_PATTERN) &&
894 ((p->token == GREP_PATTERN_HEAD) != (ctx == GREP_CONTEXT_HEAD)))
895 return 0;
896
897 again:
898 hit = patmatch(p, bol, eol, pmatch, eflags);
899
900 if (hit && p->word_regexp) {
901 if ((pmatch[0].rm_so < 0) ||
902 (eol - bol) < pmatch[0].rm_so ||
903 (pmatch[0].rm_eo < 0) ||
904 (eol - bol) < pmatch[0].rm_eo)
905 die("regexp returned nonsense");
906
907 /* Match beginning must be either beginning of the
908 * line, or at word boundary (i.e. the last char must
909 * not be a word char). Similarly, match end must be
910 * either end of the line, or at word boundary
911 * (i.e. the next char must not be a word char).
912 */
913 if ( ((pmatch[0].rm_so == 0) ||
914 !word_char(bol[pmatch[0].rm_so-1])) &&
915 ((pmatch[0].rm_eo == (eol-bol)) ||
916 !word_char(bol[pmatch[0].rm_eo])) )
917 ;
918 else
919 hit = 0;
920
921 /* Words consist of at least one character. */
922 if (pmatch->rm_so == pmatch->rm_eo)
923 hit = 0;
924
925 if (!hit && pmatch[0].rm_so + bol + 1 < eol) {
926 /* There could be more than one match on the
927 * line, and the first match might not be
928 * strict word match. But later ones could be!
929 * Forward to the next possible start, i.e. the
930 * next position following a non-word char.
931 */
932 bol = pmatch[0].rm_so + bol + 1;
933 while (word_char(bol[-1]) && bol < eol)
934 bol++;
935 eflags |= REG_NOTBOL;
936 if (bol < eol)
937 goto again;
938 }
939 }
940 if (hit) {
941 pmatch[0].rm_so += bol - start;
942 pmatch[0].rm_eo += bol - start;
943 }
944 return hit;
945 }
946
947 static int match_one_pattern(struct grep_pat *p,
948 const char *bol, const char *eol,
949 enum grep_context ctx, regmatch_t *pmatch,
950 int eflags)
951 {
952 const char *field;
953 size_t len;
954
955 if (p->token == GREP_PATTERN_HEAD) {
956 assert(p->field < ARRAY_SIZE(header_field));
957 field = header_field[p->field].field;
958 len = header_field[p->field].len;
959 if (strncmp(bol, field, len))
960 return 0;
961 bol += len;
962
963 switch (p->field) {
964 case GREP_HEADER_AUTHOR:
965 case GREP_HEADER_COMMITTER:
966 strip_timestamp(bol, &eol);
967 break;
968 default:
969 break;
970 }
971 }
972
973 return headerless_match_one_pattern(p, bol, eol, ctx, pmatch, eflags);
974 }
975
976
977 static int match_expr_eval(struct grep_opt *opt, struct grep_expr *x,
978 const char *bol, const char *eol,
979 enum grep_context ctx, ssize_t *col,
980 ssize_t *icol, int collect_hits)
981 {
982 int h = 0;
983
984 switch (x->node) {
985 case GREP_NODE_TRUE:
986 h = 1;
987 break;
988 case GREP_NODE_ATOM:
989 {
990 regmatch_t tmp;
991 h = match_one_pattern(x->u.atom, bol, eol, ctx,
992 &tmp, 0);
993 if (h && (*col < 0 || tmp.rm_so < *col))
994 *col = tmp.rm_so;
995 }
996 if (x->u.atom->token == GREP_PATTERN_BODY)
997 opt->body_hit |= h;
998 break;
999 case GREP_NODE_NOT:
1000 /*
1001 * Upon visiting a GREP_NODE_NOT, col and icol become swapped.
1002 */
1003 h = !match_expr_eval(opt, x->u.unary, bol, eol, ctx, icol, col,
1004 0);
1005 break;
1006 case GREP_NODE_AND:
1007 h = match_expr_eval(opt, x->u.binary.left, bol, eol, ctx, col,
1008 icol, 0);
1009 if (h || opt->columnnum) {
1010 /*
1011 * Don't short-circuit AND when given --column, since a
1012 * NOT earlier in the tree may turn this into an OR. In
1013 * this case, see the below comment.
1014 */
1015 h &= match_expr_eval(opt, x->u.binary.right, bol, eol,
1016 ctx, col, icol, 0);
1017 }
1018 break;
1019 case GREP_NODE_OR:
1020 if (!(collect_hits || opt->columnnum)) {
1021 /*
1022 * Don't short-circuit OR when given --column (or
1023 * collecting hits) to ensure we don't skip a later
1024 * child that would produce an earlier match.
1025 */
1026 return (match_expr_eval(opt, x->u.binary.left, bol, eol,
1027 ctx, col, icol, 0) ||
1028 match_expr_eval(opt, x->u.binary.right, bol,
1029 eol, ctx, col, icol, 0));
1030 }
1031 h = match_expr_eval(opt, x->u.binary.left, bol, eol, ctx, col,
1032 icol, 0);
1033 if (collect_hits)
1034 x->u.binary.left->hit |= h;
1035 h |= match_expr_eval(opt, x->u.binary.right, bol, eol, ctx, col,
1036 icol, collect_hits);
1037 break;
1038 default:
1039 die("Unexpected node type (internal error) %d", x->node);
1040 }
1041 if (collect_hits)
1042 x->hit |= h;
1043 return h;
1044 }
1045
1046 static int match_expr(struct grep_opt *opt,
1047 const char *bol, const char *eol,
1048 enum grep_context ctx, ssize_t *col,
1049 ssize_t *icol, int collect_hits)
1050 {
1051 struct grep_expr *x = opt->pattern_expression;
1052 return match_expr_eval(opt, x, bol, eol, ctx, col, icol, collect_hits);
1053 }
1054
1055 static int match_line(struct grep_opt *opt,
1056 const char *bol, const char *eol,
1057 ssize_t *col, ssize_t *icol,
1058 enum grep_context ctx, int collect_hits)
1059 {
1060 struct grep_pat *p;
1061 int hit = 0;
1062
1063 if (opt->pattern_expression)
1064 return match_expr(opt, bol, eol, ctx, col, icol,
1065 collect_hits);
1066
1067 /* we do not call with collect_hits without being extended */
1068 for (p = opt->pattern_list; p; p = p->next) {
1069 regmatch_t tmp;
1070 if (match_one_pattern(p, bol, eol, ctx, &tmp, 0)) {
1071 hit |= 1;
1072 if (!opt->columnnum) {
1073 /*
1074 * Without --column, any single match on a line
1075 * is enough to know that it needs to be
1076 * printed. With --column, scan _all_ patterns
1077 * to find the earliest.
1078 */
1079 break;
1080 }
1081 if (*col < 0 || tmp.rm_so < *col)
1082 *col = tmp.rm_so;
1083 }
1084 }
1085 return hit;
1086 }
1087
1088 static int match_next_pattern(struct grep_pat *p,
1089 const char *bol, const char *eol,
1090 enum grep_context ctx,
1091 regmatch_t *pmatch, int eflags)
1092 {
1093 regmatch_t match;
1094
1095 if (!headerless_match_one_pattern(p, bol, eol, ctx, &match, eflags))
1096 return 0;
1097 if (match.rm_so < 0 || match.rm_eo < 0)
1098 return 0;
1099 if (pmatch->rm_so >= 0 && pmatch->rm_eo >= 0) {
1100 if (match.rm_so > pmatch->rm_so)
1101 return 1;
1102 if (match.rm_so == pmatch->rm_so && match.rm_eo < pmatch->rm_eo)
1103 return 1;
1104 }
1105 pmatch->rm_so = match.rm_so;
1106 pmatch->rm_eo = match.rm_eo;
1107 return 1;
1108 }
1109
1110 int grep_next_match(struct grep_opt *opt,
1111 const char *bol, const char *eol,
1112 enum grep_context ctx, regmatch_t *pmatch,
1113 enum grep_header_field field, int eflags)
1114 {
1115 struct grep_pat *p;
1116 int hit = 0;
1117
1118 pmatch->rm_so = pmatch->rm_eo = -1;
1119 if (bol < eol) {
1120 for (p = ((ctx == GREP_CONTEXT_HEAD)
1121 ? opt->header_list : opt->pattern_list);
1122 p; p = p->next) {
1123 switch (p->token) {
1124 case GREP_PATTERN_HEAD:
1125 if ((field != GREP_HEADER_FIELD_MAX) &&
1126 (p->field != field))
1127 continue;
1128 /* fall thru */
1129 case GREP_PATTERN: /* atom */
1130 case GREP_PATTERN_BODY:
1131 hit |= match_next_pattern(p, bol, eol, ctx,
1132 pmatch, eflags);
1133 break;
1134 default:
1135 break;
1136 }
1137 }
1138 }
1139 return hit;
1140 }
1141
1142 static void show_line_header(struct grep_opt *opt, const char *name,
1143 unsigned lno, ssize_t cno, char sign)
1144 {
1145 if (opt->heading && opt->last_shown == 0) {
1146 output_color(opt, name, strlen(name), opt->colors[GREP_COLOR_FILENAME]);
1147 opt->output(opt, "\n", 1);
1148 }
1149 opt->last_shown = lno;
1150
1151 if (!opt->heading && opt->pathname) {
1152 output_color(opt, name, strlen(name), opt->colors[GREP_COLOR_FILENAME]);
1153 output_sep(opt, sign);
1154 }
1155 if (opt->linenum) {
1156 char buf[32];
1157 xsnprintf(buf, sizeof(buf), "%d", lno);
1158 output_color(opt, buf, strlen(buf), opt->colors[GREP_COLOR_LINENO]);
1159 output_sep(opt, sign);
1160 }
1161 /*
1162 * Treat 'cno' as the 1-indexed offset from the start of a non-context
1163 * line to its first match. Otherwise, 'cno' is 0 indicating that we are
1164 * being called with a context line.
1165 */
1166 if (opt->columnnum && cno) {
1167 char buf[32];
1168 xsnprintf(buf, sizeof(buf), "%"PRIuMAX, (uintmax_t)cno);
1169 output_color(opt, buf, strlen(buf), opt->colors[GREP_COLOR_COLUMNNO]);
1170 output_sep(opt, sign);
1171 }
1172 }
1173
1174 static void show_line(struct grep_opt *opt,
1175 const char *bol, const char *eol,
1176 const char *name, unsigned lno, ssize_t cno, char sign)
1177 {
1178 int rest = eol - bol;
1179 const char *match_color = NULL;
1180 const char *line_color = NULL;
1181
1182 if (opt->file_break && opt->last_shown == 0) {
1183 if (opt->show_hunk_mark)
1184 opt->output(opt, "\n", 1);
1185 } else if (opt->pre_context || opt->post_context || opt->funcbody) {
1186 if (opt->last_shown == 0) {
1187 if (opt->show_hunk_mark) {
1188 output_color(opt, "--", 2, opt->colors[GREP_COLOR_SEP]);
1189 opt->output(opt, "\n", 1);
1190 }
1191 } else if (lno > opt->last_shown + 1) {
1192 output_color(opt, "--", 2, opt->colors[GREP_COLOR_SEP]);
1193 opt->output(opt, "\n", 1);
1194 }
1195 }
1196 if (!opt->only_matching) {
1197 /*
1198 * In case the line we're being called with contains more than
1199 * one match, leave printing each header to the loop below.
1200 */
1201 show_line_header(opt, name, lno, cno, sign);
1202 }
1203 if (opt->color || opt->only_matching) {
1204 regmatch_t match;
1205 enum grep_context ctx = GREP_CONTEXT_BODY;
1206 int eflags = 0;
1207
1208 if (opt->color) {
1209 if (sign == ':')
1210 match_color = opt->colors[GREP_COLOR_MATCH_SELECTED];
1211 else
1212 match_color = opt->colors[GREP_COLOR_MATCH_CONTEXT];
1213 if (sign == ':')
1214 line_color = opt->colors[GREP_COLOR_SELECTED];
1215 else if (sign == '-')
1216 line_color = opt->colors[GREP_COLOR_CONTEXT];
1217 else if (sign == '=')
1218 line_color = opt->colors[GREP_COLOR_FUNCTION];
1219 }
1220 while (grep_next_match(opt, bol, eol, ctx, &match,
1221 GREP_HEADER_FIELD_MAX, eflags)) {
1222 if (match.rm_so == match.rm_eo)
1223 break;
1224
1225 if (opt->only_matching)
1226 show_line_header(opt, name, lno, cno, sign);
1227 else
1228 output_color(opt, bol, match.rm_so, line_color);
1229 output_color(opt, bol + match.rm_so,
1230 match.rm_eo - match.rm_so, match_color);
1231 if (opt->only_matching)
1232 opt->output(opt, "\n", 1);
1233 bol += match.rm_eo;
1234 cno += match.rm_eo;
1235 rest -= match.rm_eo;
1236 eflags = REG_NOTBOL;
1237 }
1238 }
1239 if (!opt->only_matching) {
1240 output_color(opt, bol, rest, line_color);
1241 opt->output(opt, "\n", 1);
1242 }
1243 }
1244
1245 int grep_use_locks;
1246
1247 /*
1248 * This lock protects access to the gitattributes machinery, which is
1249 * not thread-safe.
1250 */
1251 pthread_mutex_t grep_attr_mutex;
1252
1253 static inline void grep_attr_lock(void)
1254 {
1255 if (grep_use_locks)
1256 pthread_mutex_lock(&grep_attr_mutex);
1257 }
1258
1259 static inline void grep_attr_unlock(void)
1260 {
1261 if (grep_use_locks)
1262 pthread_mutex_unlock(&grep_attr_mutex);
1263 }
1264
1265 static int match_funcname(struct grep_opt *opt, struct grep_source *gs,
1266 const char *bol, const char *eol)
1267 {
1268 xdemitconf_t *xecfg = opt->priv;
1269 if (xecfg && !xecfg->find_func) {
1270 grep_source_load_driver(gs, opt->repo->index);
1271 if (gs->driver->funcname.pattern) {
1272 const struct userdiff_funcname *pe = &gs->driver->funcname;
1273 xdiff_set_find_func(xecfg, pe->pattern, pe->cflags);
1274 } else {
1275 xecfg = opt->priv = NULL;
1276 }
1277 }
1278
1279 if (xecfg) {
1280 char buf[1];
1281 return xecfg->find_func(bol, eol - bol, buf, 1,
1282 xecfg->find_func_priv) >= 0;
1283 }
1284
1285 if (bol == eol)
1286 return 0;
1287 if (isalpha(*bol) || *bol == '_' || *bol == '$')
1288 return 1;
1289 return 0;
1290 }
1291
1292 static void show_funcname_line(struct grep_opt *opt, struct grep_source *gs,
1293 const char *bol, unsigned lno)
1294 {
1295 while (bol > gs->buf) {
1296 const char *eol = --bol;
1297
1298 while (bol > gs->buf && bol[-1] != '\n')
1299 bol--;
1300 lno--;
1301
1302 if (lno <= opt->last_shown)
1303 break;
1304
1305 if (match_funcname(opt, gs, bol, eol)) {
1306 show_line(opt, bol, eol, gs->name, lno, 0, '=');
1307 break;
1308 }
1309 }
1310 }
1311
1312 static int is_empty_line(const char *bol, const char *eol);
1313
1314 static void show_pre_context(struct grep_opt *opt, struct grep_source *gs,
1315 const char *bol, const char *end, unsigned lno)
1316 {
1317 unsigned cur = lno, from = 1, funcname_lno = 0, orig_from;
1318 int funcname_needed = !!opt->funcname, comment_needed = 0;
1319
1320 if (opt->pre_context < lno)
1321 from = lno - opt->pre_context;
1322 if (from <= opt->last_shown)
1323 from = opt->last_shown + 1;
1324 orig_from = from;
1325 if (opt->funcbody) {
1326 if (match_funcname(opt, gs, bol, end))
1327 comment_needed = 1;
1328 else
1329 funcname_needed = 1;
1330 from = opt->last_shown + 1;
1331 }
1332
1333 /* Rewind. */
1334 while (bol > gs->buf && cur > from) {
1335 const char *next_bol = bol;
1336 const char *eol = --bol;
1337
1338 while (bol > gs->buf && bol[-1] != '\n')
1339 bol--;
1340 cur--;
1341 if (comment_needed && (is_empty_line(bol, eol) ||
1342 match_funcname(opt, gs, bol, eol))) {
1343 comment_needed = 0;
1344 from = orig_from;
1345 if (cur < from) {
1346 cur++;
1347 bol = next_bol;
1348 break;
1349 }
1350 }
1351 if (funcname_needed && match_funcname(opt, gs, bol, eol)) {
1352 funcname_lno = cur;
1353 funcname_needed = 0;
1354 if (opt->funcbody)
1355 comment_needed = 1;
1356 else
1357 from = orig_from;
1358 }
1359 }
1360
1361 /* We need to look even further back to find a function signature. */
1362 if (opt->funcname && funcname_needed)
1363 show_funcname_line(opt, gs, bol, cur);
1364
1365 /* Back forward. */
1366 while (cur < lno) {
1367 const char *eol = bol, sign = (cur == funcname_lno) ? '=' : '-';
1368
1369 while (*eol != '\n')
1370 eol++;
1371 show_line(opt, bol, eol, gs->name, cur, 0, sign);
1372 bol = eol + 1;
1373 cur++;
1374 }
1375 }
1376
1377 static int should_lookahead(struct grep_opt *opt)
1378 {
1379 struct grep_pat *p;
1380
1381 if (opt->pattern_expression)
1382 return 0; /* punt for too complex stuff */
1383 if (opt->invert)
1384 return 0;
1385 for (p = opt->pattern_list; p; p = p->next) {
1386 if (p->token != GREP_PATTERN)
1387 return 0; /* punt for "header only" and stuff */
1388 }
1389 return 1;
1390 }
1391
1392 static int look_ahead(struct grep_opt *opt,
1393 unsigned long *left_p,
1394 unsigned *lno_p,
1395 const char **bol_p)
1396 {
1397 unsigned lno = *lno_p;
1398 const char *bol = *bol_p;
1399 struct grep_pat *p;
1400 const char *sp, *last_bol;
1401 regoff_t earliest = -1;
1402
1403 for (p = opt->pattern_list; p; p = p->next) {
1404 int hit;
1405 regmatch_t m;
1406
1407 hit = patmatch(p, bol, bol + *left_p, &m, 0);
1408 if (!hit || m.rm_so < 0 || m.rm_eo < 0)
1409 continue;
1410 if (earliest < 0 || m.rm_so < earliest)
1411 earliest = m.rm_so;
1412 }
1413
1414 if (earliest < 0) {
1415 *bol_p = bol + *left_p;
1416 *left_p = 0;
1417 return 1;
1418 }
1419 for (sp = bol + earliest; bol < sp && sp[-1] != '\n'; sp--)
1420 ; /* find the beginning of the line */
1421 last_bol = sp;
1422
1423 for (sp = bol; sp < last_bol; sp++) {
1424 if (*sp == '\n')
1425 lno++;
1426 }
1427 *left_p -= last_bol - bol;
1428 *bol_p = last_bol;
1429 *lno_p = lno;
1430 return 0;
1431 }
1432
1433 static int fill_textconv_grep(struct repository *r,
1434 struct userdiff_driver *driver,
1435 struct grep_source *gs)
1436 {
1437 struct diff_filespec *df;
1438 char *buf;
1439 size_t size;
1440
1441 if (!driver || !driver->textconv)
1442 return grep_source_load(gs);
1443
1444 /*
1445 * The textconv interface is intimately tied to diff_filespecs, so we
1446 * have to pretend to be one. If we could unify the grep_source
1447 * and diff_filespec structs, this mess could just go away.
1448 */
1449 df = alloc_filespec(gs->path);
1450 switch (gs->type) {
1451 case GREP_SOURCE_OID:
1452 fill_filespec(df, gs->identifier, 1, 0100644);
1453 break;
1454 case GREP_SOURCE_FILE:
1455 fill_filespec(df, null_oid(), 0, 0100644);
1456 break;
1457 default:
1458 BUG("attempt to textconv something without a path?");
1459 }
1460
1461 /*
1462 * fill_textconv is not remotely thread-safe; it modifies the global
1463 * diff tempfile structure, writes to the_repo's odb and might
1464 * internally call thread-unsafe functions such as the
1465 * prepare_packed_git() lazy-initializator. Because of the last two, we
1466 * must ensure mutual exclusion between this call and the object reading
1467 * API, thus we use obj_read_lock() here.
1468 *
1469 * TODO: allowing text conversion to run in parallel with object
1470 * reading operations might increase performance in the multithreaded
1471 * non-worktreee git-grep with --textconv.
1472 */
1473 obj_read_lock();
1474 size = fill_textconv(r, driver, df, &buf);
1475 obj_read_unlock();
1476 free_filespec(df);
1477
1478 /*
1479 * The normal fill_textconv usage by the diff machinery would just keep
1480 * the textconv'd buf separate from the diff_filespec. But much of the
1481 * grep code passes around a grep_source and assumes that its "buf"
1482 * pointer is the beginning of the thing we are searching. So let's
1483 * install our textconv'd version into the grep_source, taking care not
1484 * to leak any existing buffer.
1485 */
1486 grep_source_clear_data(gs);
1487 gs->buf = buf;
1488 gs->size = size;
1489
1490 return 0;
1491 }
1492
1493 static int is_empty_line(const char *bol, const char *eol)
1494 {
1495 while (bol < eol && isspace(*bol))
1496 bol++;
1497 return bol == eol;
1498 }
1499
1500 static int grep_source_1(struct grep_opt *opt, struct grep_source *gs, int collect_hits)
1501 {
1502 const char *bol;
1503 const char *peek_bol = NULL;
1504 unsigned long left;
1505 unsigned lno = 1;
1506 unsigned last_hit = 0;
1507 int binary_match_only = 0;
1508 unsigned count = 0;
1509 int try_lookahead = 0;
1510 int show_function = 0;
1511 struct userdiff_driver *textconv = NULL;
1512 enum grep_context ctx = GREP_CONTEXT_HEAD;
1513 xdemitconf_t xecfg;
1514
1515 if (!opt->status_only && gs->name == NULL)
1516 BUG("grep call which could print a name requires "
1517 "grep_source.name be non-NULL");
1518
1519 if (!opt->output)
1520 opt->output = std_output;
1521
1522 if (opt->pre_context || opt->post_context || opt->file_break ||
1523 opt->funcbody) {
1524 /* Show hunk marks, except for the first file. */
1525 if (opt->last_shown)
1526 opt->show_hunk_mark = 1;
1527 /*
1528 * If we're using threads then we can't easily identify
1529 * the first file. Always put hunk marks in that case
1530 * and skip the very first one later in work_done().
1531 */
1532 if (opt->output != std_output)
1533 opt->show_hunk_mark = 1;
1534 }
1535 opt->last_shown = 0;
1536
1537 if (opt->allow_textconv) {
1538 grep_source_load_driver(gs, opt->repo->index);
1539 /*
1540 * We might set up the shared textconv cache data here, which
1541 * is not thread-safe. Also, get_oid_with_context() and
1542 * parse_object() might be internally called. As they are not
1543 * currently thread-safe and might be racy with object reading,
1544 * obj_read_lock() must be called.
1545 */
1546 grep_attr_lock();
1547 obj_read_lock();
1548 textconv = userdiff_get_textconv(opt->repo, gs->driver);
1549 obj_read_unlock();
1550 grep_attr_unlock();
1551 }
1552
1553 /*
1554 * We know the result of a textconv is text, so we only have to care
1555 * about binary handling if we are not using it.
1556 */
1557 if (!textconv) {
1558 switch (opt->binary) {
1559 case GREP_BINARY_DEFAULT:
1560 if (grep_source_is_binary(gs, opt->repo->index))
1561 binary_match_only = 1;
1562 break;
1563 case GREP_BINARY_NOMATCH:
1564 if (grep_source_is_binary(gs, opt->repo->index))
1565 return 0; /* Assume unmatch */
1566 break;
1567 case GREP_BINARY_TEXT:
1568 break;
1569 default:
1570 BUG("unknown binary handling mode");
1571 }
1572 }
1573
1574 memset(&xecfg, 0, sizeof(xecfg));
1575 opt->priv = &xecfg;
1576
1577 try_lookahead = should_lookahead(opt);
1578
1579 if (fill_textconv_grep(opt->repo, textconv, gs) < 0)
1580 return 0;
1581
1582 bol = gs->buf;
1583 left = gs->size;
1584 while (left) {
1585 const char *eol;
1586 int hit;
1587 ssize_t cno;
1588 ssize_t col = -1, icol = -1;
1589
1590 /*
1591 * look_ahead() skips quickly to the line that possibly
1592 * has the next hit; don't call it if we need to do
1593 * something more than just skipping the current line
1594 * in response to an unmatch for the current line. E.g.
1595 * inside a post-context window, we will show the current
1596 * line as a context around the previous hit when it
1597 * doesn't hit.
1598 */
1599 if (try_lookahead
1600 && !(last_hit
1601 && (show_function ||
1602 lno <= last_hit + opt->post_context))
1603 && look_ahead(opt, &left, &lno, &bol))
1604 break;
1605 eol = end_of_line(bol, &left);
1606
1607 if ((ctx == GREP_CONTEXT_HEAD) && (eol == bol))
1608 ctx = GREP_CONTEXT_BODY;
1609
1610 hit = match_line(opt, bol, eol, &col, &icol, ctx, collect_hits);
1611
1612 if (collect_hits)
1613 goto next_line;
1614
1615 /* "grep -v -e foo -e bla" should list lines
1616 * that do not have either, so inversion should
1617 * be done outside.
1618 */
1619 if (opt->invert)
1620 hit = !hit;
1621 if (opt->unmatch_name_only) {
1622 if (hit)
1623 return 0;
1624 goto next_line;
1625 }
1626 if (hit && (opt->max_count < 0 || count < opt->max_count)) {
1627 count++;
1628 if (opt->status_only)
1629 return 1;
1630 if (opt->name_only) {
1631 show_name(opt, gs->name);
1632 return 1;
1633 }
1634 if (opt->count)
1635 goto next_line;
1636 if (binary_match_only) {
1637 opt->output(opt, "Binary file ", 12);
1638 output_color(opt, gs->name, strlen(gs->name),
1639 opt->colors[GREP_COLOR_FILENAME]);
1640 opt->output(opt, " matches\n", 9);
1641 return 1;
1642 }
1643 /* Hit at this line. If we haven't shown the
1644 * pre-context lines, we would need to show them.
1645 */
1646 if (opt->pre_context || opt->funcbody)
1647 show_pre_context(opt, gs, bol, eol, lno);
1648 else if (opt->funcname)
1649 show_funcname_line(opt, gs, bol, lno);
1650 cno = opt->invert ? icol : col;
1651 if (cno < 0) {
1652 /*
1653 * A negative cno indicates that there was no
1654 * match on the line. We are thus inverted and
1655 * being asked to show all lines that _don't_
1656 * match a given expression. Therefore, set cno
1657 * to 0 to suggest the whole line matches.
1658 */
1659 cno = 0;
1660 }
1661 show_line(opt, bol, eol, gs->name, lno, cno + 1, ':');
1662 last_hit = lno;
1663 if (opt->funcbody)
1664 show_function = 1;
1665 goto next_line;
1666 }
1667 if (show_function && (!peek_bol || peek_bol < bol)) {
1668 unsigned long peek_left = left;
1669 const char *peek_eol = eol;
1670
1671 /*
1672 * Trailing empty lines are not interesting.
1673 * Peek past them to see if they belong to the
1674 * body of the current function.
1675 */
1676 peek_bol = bol;
1677 while (is_empty_line(peek_bol, peek_eol)) {
1678 peek_bol = peek_eol + 1;
1679 peek_eol = end_of_line(peek_bol, &peek_left);
1680 }
1681
1682 if (match_funcname(opt, gs, peek_bol, peek_eol))
1683 show_function = 0;
1684 }
1685 if (show_function ||
1686 (last_hit && lno <= last_hit + opt->post_context)) {
1687 /* If the last hit is within the post context,
1688 * we need to show this line.
1689 */
1690 show_line(opt, bol, eol, gs->name, lno, col + 1, '-');
1691 }
1692
1693 next_line:
1694 bol = eol + 1;
1695 if (!left)
1696 break;
1697 left--;
1698 lno++;
1699 }
1700
1701 if (collect_hits)
1702 return 0;
1703
1704 if (opt->status_only)
1705 return opt->unmatch_name_only;
1706 if (opt->unmatch_name_only) {
1707 /* We did not see any hit, so we want to show this */
1708 show_name(opt, gs->name);
1709 return 1;
1710 }
1711
1712 xdiff_clear_find_func(&xecfg);
1713 opt->priv = NULL;
1714
1715 /* NEEDSWORK:
1716 * The real "grep -c foo *.c" gives many "bar.c:0" lines,
1717 * which feels mostly useless but sometimes useful. Maybe
1718 * make it another option? For now suppress them.
1719 */
1720 if (opt->count && count) {
1721 char buf[32];
1722 if (opt->pathname) {
1723 output_color(opt, gs->name, strlen(gs->name),
1724 opt->colors[GREP_COLOR_FILENAME]);
1725 output_sep(opt, ':');
1726 }
1727 xsnprintf(buf, sizeof(buf), "%u\n", count);
1728 opt->output(opt, buf, strlen(buf));
1729 return 1;
1730 }
1731 return !!last_hit;
1732 }
1733
1734 static void clr_hit_marker(struct grep_expr *x)
1735 {
1736 /* All-hit markers are meaningful only at the very top level
1737 * OR node.
1738 */
1739 while (1) {
1740 x->hit = 0;
1741 if (x->node != GREP_NODE_OR)
1742 return;
1743 x->u.binary.left->hit = 0;
1744 x = x->u.binary.right;
1745 }
1746 }
1747
1748 static int chk_hit_marker(struct grep_expr *x)
1749 {
1750 /* Top level nodes have hit markers. See if they all are hits */
1751 while (1) {
1752 if (x->node != GREP_NODE_OR)
1753 return x->hit;
1754 if (!x->u.binary.left->hit)
1755 return 0;
1756 x = x->u.binary.right;
1757 }
1758 }
1759
1760 int grep_source(struct grep_opt *opt, struct grep_source *gs)
1761 {
1762 /*
1763 * we do not have to do the two-pass grep when we do not check
1764 * buffer-wide "all-match".
1765 */
1766 if (!opt->all_match && !opt->no_body_match)
1767 return grep_source_1(opt, gs, 0);
1768
1769 /* Otherwise the toplevel "or" terms hit a bit differently.
1770 * We first clear hit markers from them.
1771 */
1772 clr_hit_marker(opt->pattern_expression);
1773 opt->body_hit = 0;
1774 grep_source_1(opt, gs, 1);
1775
1776 if (opt->all_match && !chk_hit_marker(opt->pattern_expression))
1777 return 0;
1778 if (opt->no_body_match && opt->body_hit)
1779 return 0;
1780
1781 return grep_source_1(opt, gs, 0);
1782 }
1783
1784 static void grep_source_init_buf(struct grep_source *gs,
1785 const char *buf,
1786 unsigned long size)
1787 {
1788 gs->type = GREP_SOURCE_BUF;
1789 gs->name = NULL;
1790 gs->path = NULL;
1791 gs->buf = buf;
1792 gs->size = size;
1793 gs->driver = NULL;
1794 gs->identifier = NULL;
1795 }
1796
1797 int grep_buffer(struct grep_opt *opt, const char *buf, unsigned long size)
1798 {
1799 struct grep_source gs;
1800 int r;
1801
1802 grep_source_init_buf(&gs, buf, size);
1803
1804 r = grep_source(opt, &gs);
1805
1806 grep_source_clear(&gs);
1807 return r;
1808 }
1809
1810 void grep_source_init_file(struct grep_source *gs, const char *name,
1811 const char *path)
1812 {
1813 gs->type = GREP_SOURCE_FILE;
1814 gs->name = xstrdup_or_null(name);
1815 gs->path = xstrdup_or_null(path);
1816 gs->buf = NULL;
1817 gs->size = 0;
1818 gs->driver = NULL;
1819 gs->identifier = xstrdup(path);
1820 }
1821
1822 void grep_source_init_oid(struct grep_source *gs, const char *name,
1823 const char *path, const struct object_id *oid,
1824 struct repository *repo)
1825 {
1826 gs->type = GREP_SOURCE_OID;
1827 gs->name = xstrdup_or_null(name);
1828 gs->path = xstrdup_or_null(path);
1829 gs->buf = NULL;
1830 gs->size = 0;
1831 gs->driver = NULL;
1832 gs->identifier = oiddup(oid);
1833 gs->repo = repo;
1834 }
1835
1836 void grep_source_clear(struct grep_source *gs)
1837 {
1838 FREE_AND_NULL(gs->name);
1839 FREE_AND_NULL(gs->path);
1840 FREE_AND_NULL(gs->identifier);
1841 grep_source_clear_data(gs);
1842 }
1843
1844 void grep_source_clear_data(struct grep_source *gs)
1845 {
1846 switch (gs->type) {
1847 case GREP_SOURCE_FILE:
1848 case GREP_SOURCE_OID:
1849 /* these types own the buffer */
1850 free((char *)gs->buf);
1851 gs->buf = NULL;
1852 gs->size = 0;
1853 break;
1854 case GREP_SOURCE_BUF:
1855 /* leave user-provided buf intact */
1856 break;
1857 }
1858 }
1859
1860 static int grep_source_load_oid(struct grep_source *gs)
1861 {
1862 enum object_type type;
1863
1864 gs->buf = repo_read_object_file(gs->repo, gs->identifier, &type,
1865 &gs->size);
1866 if (!gs->buf)
1867 return error(_("'%s': unable to read %s"),
1868 gs->name,
1869 oid_to_hex(gs->identifier));
1870 return 0;
1871 }
1872
1873 static int grep_source_load_file(struct grep_source *gs)
1874 {
1875 const char *filename = gs->identifier;
1876 struct stat st;
1877 char *data;
1878 size_t size;
1879 int i;
1880
1881 if (lstat(filename, &st) < 0) {
1882 err_ret:
1883 if (errno != ENOENT)
1884 error_errno(_("failed to stat '%s'"), filename);
1885 return -1;
1886 }
1887 if (!S_ISREG(st.st_mode))
1888 return -1;
1889 size = xsize_t(st.st_size);
1890 i = open(filename, O_RDONLY);
1891 if (i < 0)
1892 goto err_ret;
1893 data = xmallocz(size);
1894 if (st.st_size != read_in_full(i, data, size)) {
1895 error_errno(_("'%s': short read"), filename);
1896 close(i);
1897 free(data);
1898 return -1;
1899 }
1900 close(i);
1901
1902 gs->buf = data;
1903 gs->size = size;
1904 return 0;
1905 }
1906
1907 static int grep_source_load(struct grep_source *gs)
1908 {
1909 if (gs->buf)
1910 return 0;
1911
1912 switch (gs->type) {
1913 case GREP_SOURCE_FILE:
1914 return grep_source_load_file(gs);
1915 case GREP_SOURCE_OID:
1916 return grep_source_load_oid(gs);
1917 case GREP_SOURCE_BUF:
1918 return gs->buf ? 0 : -1;
1919 }
1920 BUG("invalid grep_source type to load");
1921 }
1922
1923 void grep_source_load_driver(struct grep_source *gs,
1924 struct index_state *istate)
1925 {
1926 if (gs->driver)
1927 return;
1928
1929 grep_attr_lock();
1930 if (gs->path)
1931 gs->driver = userdiff_find_by_path(istate, gs->path);
1932 if (!gs->driver)
1933 gs->driver = userdiff_find_by_name("default");
1934 grep_attr_unlock();
1935 }
1936
1937 static int grep_source_is_binary(struct grep_source *gs,
1938 struct index_state *istate)
1939 {
1940 grep_source_load_driver(gs, istate);
1941 if (gs->driver->binary != -1)
1942 return gs->driver->binary;
1943
1944 if (!grep_source_load(gs))
1945 return buffer_is_binary(gs->buf, gs->size);
1946
1947 return 0;
1948 }