?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'ab/sequencer-unleak'
[git/debian.git] / grep.c
blob0690ed9b895612bc106efcb03254da5241361029
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 int pcre2_jit_functional(void)
266 {
267 static int jit_working = -1;
268 pcre2_code *code;
269 size_t off;
270 int err;
271
272 if (jit_working != -1)
273 return jit_working;
274
275 /*
276 * Try to JIT compile a simple pattern to probe if the JIT is
277 * working in general. It might fail for systems where creating
278 * memory mappings for runtime code generation is restricted.
279 */
280 code = pcre2_compile((PCRE2_SPTR)".", 1, 0, &err, &off, NULL);
281 if (!code)
282 return 0;
283
284 jit_working = pcre2_jit_compile(code, PCRE2_JIT_COMPLETE) == 0;
285 pcre2_code_free(code);
286
287 return jit_working;
288 }
289
290 static void compile_pcre2_pattern(struct grep_pat *p, const struct grep_opt *opt)
291 {
292 int error;
293 PCRE2_UCHAR errbuf[256];
294 PCRE2_SIZE erroffset;
295 int options = PCRE2_MULTILINE;
296 int jitret;
297 int patinforet;
298 size_t jitsizearg;
299 int literal = !opt->ignore_case && (p->fixed || p->is_fixed);
300
301 /*
302 * Call pcre2_general_context_create() before calling any
303 * other pcre2_*(). It sets up our malloc()/free() functions
304 * with which everything else is allocated.
305 */
306 p->pcre2_general_context = pcre2_general_context_create(
307 pcre2_malloc, pcre2_free, NULL);
308 if (!p->pcre2_general_context)
309 die("Couldn't allocate PCRE2 general context");
310
311 if (opt->ignore_case) {
312 if (!opt->ignore_locale && has_non_ascii(p->pattern)) {
313 p->pcre2_tables = pcre2_maketables(p->pcre2_general_context);
314 p->pcre2_compile_context = pcre2_compile_context_create(p->pcre2_general_context);
315 pcre2_set_character_tables(p->pcre2_compile_context,
316 p->pcre2_tables);
317 }
318 options |= PCRE2_CASELESS;
319 }
320 if (!opt->ignore_locale && is_utf8_locale() && !literal)
321 options |= (PCRE2_UTF | PCRE2_UCP | PCRE2_MATCH_INVALID_UTF);
322
323 #ifndef GIT_PCRE2_VERSION_10_36_OR_HIGHER
324 /* Work around https://bugs.exim.org/show_bug.cgi?id=2642 fixed in 10.36 */
325 if (PCRE2_MATCH_INVALID_UTF && options & (PCRE2_UTF | PCRE2_CASELESS))
326 options |= PCRE2_NO_START_OPTIMIZE;
327 #endif
328
329 p->pcre2_pattern = pcre2_compile((PCRE2_SPTR)p->pattern,
330 p->patternlen, options, &error, &erroffset,
331 p->pcre2_compile_context);
332
333 if (p->pcre2_pattern) {
334 p->pcre2_match_data = pcre2_match_data_create_from_pattern(p->pcre2_pattern, p->pcre2_general_context);
335 if (!p->pcre2_match_data)
336 die("Couldn't allocate PCRE2 match data");
337 } else {
338 pcre2_get_error_message(error, errbuf, sizeof(errbuf));
339 compile_regexp_failed(p, (const char *)&errbuf);
340 }
341
342 pcre2_config(PCRE2_CONFIG_JIT, &p->pcre2_jit_on);
343 if (p->pcre2_jit_on) {
344 jitret = pcre2_jit_compile(p->pcre2_pattern, PCRE2_JIT_COMPLETE);
345 if (jitret == PCRE2_ERROR_NOMEMORY && !pcre2_jit_functional()) {
346 /*
347 * Even though pcre2_config(PCRE2_CONFIG_JIT, ...)
348 * indicated JIT support, the library might still
349 * fail to generate JIT code for various reasons,
350 * e.g. when SELinux's 'deny_execmem' or PaX's
351 * MPROTECT prevent creating W|X memory mappings.
352 *
353 * Instead of faling hard, fall back to interpreter
354 * mode, just as if the pattern was prefixed with
355 * '(*NO_JIT)'.
356 */
357 p->pcre2_jit_on = 0;
358 return;
359 } else if (jitret) {
360 int need_clip = p->patternlen > 64;
361 int clip_len = need_clip ? 64 : p->patternlen;
362 die("Couldn't JIT the PCRE2 pattern '%.*s'%s, got '%d'%s",
363 clip_len, p->pattern, need_clip ? "..." : "", jitret,
364 pcre2_jit_functional()
365 ? "\nPerhaps prefix (*NO_JIT) to your pattern?"
366 : "");
367 }
368
369 /*
370 * The pcre2_config(PCRE2_CONFIG_JIT, ...) call just
371 * tells us whether the library itself supports JIT,
372 * but to see whether we're going to be actually using
373 * JIT we need to extract PCRE2_INFO_JITSIZE from the
374 * pattern *after* we do pcre2_jit_compile() above.
375 *
376 * This is because if the pattern contains the
377 * (*NO_JIT) verb (see pcre2syntax(3))
378 * pcre2_jit_compile() will exit early with 0. If we
379 * then proceed to call pcre2_jit_match() further down
380 * the line instead of pcre2_match() we'll either
381 * segfault (pre PCRE 10.31) or run into a fatal error
382 * (post PCRE2 10.31)
383 */
384 patinforet = pcre2_pattern_info(p->pcre2_pattern, PCRE2_INFO_JITSIZE, &jitsizearg);
385 if (patinforet)
386 BUG("pcre2_pattern_info() failed: %d", patinforet);
387 if (jitsizearg == 0) {
388 p->pcre2_jit_on = 0;
389 return;
390 }
391 }
392 }
393
394 static int pcre2match(struct grep_pat *p, const char *line, const char *eol,
395 regmatch_t *match, int eflags)
396 {
397 int ret, flags = 0;
398 PCRE2_SIZE *ovector;
399 PCRE2_UCHAR errbuf[256];
400
401 if (eflags & REG_NOTBOL)
402 flags |= PCRE2_NOTBOL;
403
404 if (p->pcre2_jit_on)
405 ret = pcre2_jit_match(p->pcre2_pattern, (unsigned char *)line,
406 eol - line, 0, flags, p->pcre2_match_data,
407 NULL);
408 else
409 ret = pcre2_match(p->pcre2_pattern, (unsigned char *)line,
410 eol - line, 0, flags, p->pcre2_match_data,
411 NULL);
412
413 if (ret < 0 && ret != PCRE2_ERROR_NOMATCH) {
414 pcre2_get_error_message(ret, errbuf, sizeof(errbuf));
415 die("%s failed with error code %d: %s",
416 (p->pcre2_jit_on ? "pcre2_jit_match" : "pcre2_match"), ret,
417 errbuf);
418 }
419 if (ret > 0) {
420 ovector = pcre2_get_ovector_pointer(p->pcre2_match_data);
421 ret = 0;
422 match->rm_so = (int)ovector[0];
423 match->rm_eo = (int)ovector[1];
424 }
425
426 return ret;
427 }
428
429 static void free_pcre2_pattern(struct grep_pat *p)
430 {
431 pcre2_compile_context_free(p->pcre2_compile_context);
432 pcre2_code_free(p->pcre2_pattern);
433 pcre2_match_data_free(p->pcre2_match_data);
434 #ifdef GIT_PCRE2_VERSION_10_34_OR_HIGHER
435 pcre2_maketables_free(p->pcre2_general_context, p->pcre2_tables);
436 #else
437 free((void *)p->pcre2_tables);
438 #endif
439 pcre2_general_context_free(p->pcre2_general_context);
440 }
441 #else /* !USE_LIBPCRE2 */
442 static void compile_pcre2_pattern(struct grep_pat *p, const struct grep_opt *opt)
443 {
444 die("cannot use Perl-compatible regexes when not compiled with USE_LIBPCRE");
445 }
446
447 static int pcre2match(struct grep_pat *p, const char *line, const char *eol,
448 regmatch_t *match, int eflags)
449 {
450 return 1;
451 }
452
453 static void free_pcre2_pattern(struct grep_pat *p)
454 {
455 }
456
457 static void compile_fixed_regexp(struct grep_pat *p, struct grep_opt *opt)
458 {
459 struct strbuf sb = STRBUF_INIT;
460 int err;
461 int regflags = 0;
462
463 basic_regex_quote_buf(&sb, p->pattern);
464 if (opt->ignore_case)
465 regflags |= REG_ICASE;
466 err = regcomp(&p->regexp, sb.buf, regflags);
467 strbuf_release(&sb);
468 if (err) {
469 char errbuf[1024];
470 regerror(err, &p->regexp, errbuf, sizeof(errbuf));
471 compile_regexp_failed(p, errbuf);
472 }
473 }
474 #endif /* !USE_LIBPCRE2 */
475
476 static void compile_regexp(struct grep_pat *p, struct grep_opt *opt)
477 {
478 int err;
479 int regflags = REG_NEWLINE;
480
481 if (opt->pattern_type_option == GREP_PATTERN_TYPE_UNSPECIFIED)
482 opt->pattern_type_option = (opt->extended_regexp_option
483 ? GREP_PATTERN_TYPE_ERE
484 : GREP_PATTERN_TYPE_BRE);
485
486 p->word_regexp = opt->word_regexp;
487 p->ignore_case = opt->ignore_case;
488 p->fixed = opt->pattern_type_option == GREP_PATTERN_TYPE_FIXED;
489
490 if (opt->pattern_type_option != GREP_PATTERN_TYPE_PCRE &&
491 memchr(p->pattern, 0, p->patternlen))
492 die(_("given pattern contains NULL byte (via -f <file>). This is only supported with -P under PCRE v2"));
493
494 p->is_fixed = is_fixed(p->pattern, p->patternlen);
495 #ifdef USE_LIBPCRE2
496 if (!p->fixed && !p->is_fixed) {
497 const char *no_jit = "(*NO_JIT)";
498 const int no_jit_len = strlen(no_jit);
499 if (starts_with(p->pattern, no_jit) &&
500 is_fixed(p->pattern + no_jit_len,
501 p->patternlen - no_jit_len))
502 p->is_fixed = 1;
503 }
504 #endif
505 if (p->fixed || p->is_fixed) {
506 #ifdef USE_LIBPCRE2
507 if (p->is_fixed) {
508 compile_pcre2_pattern(p, opt);
509 } else {
510 /*
511 * E.g. t7811-grep-open.sh relies on the
512 * pattern being restored.
513 */
514 char *old_pattern = p->pattern;
515 size_t old_patternlen = p->patternlen;
516 struct strbuf sb = STRBUF_INIT;
517
518 /*
519 * There is the PCRE2_LITERAL flag, but it's
520 * only in PCRE v2 10.30 and later. Needing to
521 * ifdef our way around that and dealing with
522 * it + PCRE2_MULTILINE being an error is more
523 * complex than just quoting this ourselves.
524 */
525 strbuf_add(&sb, "\\Q", 2);
526 strbuf_add(&sb, p->pattern, p->patternlen);
527 strbuf_add(&sb, "\\E", 2);
528
529 p->pattern = sb.buf;
530 p->patternlen = sb.len;
531 compile_pcre2_pattern(p, opt);
532 p->pattern = old_pattern;
533 p->patternlen = old_patternlen;
534 strbuf_release(&sb);
535 }
536 #else /* !USE_LIBPCRE2 */
537 compile_fixed_regexp(p, opt);
538 #endif /* !USE_LIBPCRE2 */
539 return;
540 }
541
542 if (opt->pattern_type_option == GREP_PATTERN_TYPE_PCRE) {
543 compile_pcre2_pattern(p, opt);
544 return;
545 }
546
547 if (p->ignore_case)
548 regflags |= REG_ICASE;
549 if (opt->pattern_type_option == GREP_PATTERN_TYPE_ERE)
550 regflags |= REG_EXTENDED;
551 err = regcomp(&p->regexp, p->pattern, regflags);
552 if (err) {
553 char errbuf[1024];
554 regerror(err, &p->regexp, errbuf, 1024);
555 compile_regexp_failed(p, errbuf);
556 }
557 }
558
559 static struct grep_expr *grep_not_expr(struct grep_expr *expr)
560 {
561 struct grep_expr *z = xcalloc(1, sizeof(*z));
562 z->node = GREP_NODE_NOT;
563 z->u.unary = expr;
564 return z;
565 }
566
567 static struct grep_expr *grep_binexp(enum grep_expr_node kind,
568 struct grep_expr *left,
569 struct grep_expr *right)
570 {
571 struct grep_expr *z = xcalloc(1, sizeof(*z));
572 z->node = kind;
573 z->u.binary.left = left;
574 z->u.binary.right = right;
575 return z;
576 }
577
578 static struct grep_expr *grep_or_expr(struct grep_expr *left, struct grep_expr *right)
579 {
580 return grep_binexp(GREP_NODE_OR, left, right);
581 }
582
583 static struct grep_expr *grep_and_expr(struct grep_expr *left, struct grep_expr *right)
584 {
585 return grep_binexp(GREP_NODE_AND, left, right);
586 }
587
588 static struct grep_expr *compile_pattern_or(struct grep_pat **);
589 static struct grep_expr *compile_pattern_atom(struct grep_pat **list)
590 {
591 struct grep_pat *p;
592 struct grep_expr *x;
593
594 p = *list;
595 if (!p)
596 return NULL;
597 switch (p->token) {
598 case GREP_PATTERN: /* atom */
599 case GREP_PATTERN_HEAD:
600 case GREP_PATTERN_BODY:
601 CALLOC_ARRAY(x, 1);
602 x->node = GREP_NODE_ATOM;
603 x->u.atom = p;
604 *list = p->next;
605 return x;
606 case GREP_OPEN_PAREN:
607 *list = p->next;
608 x = compile_pattern_or(list);
609 if (!*list || (*list)->token != GREP_CLOSE_PAREN)
610 die("unmatched parenthesis");
611 *list = (*list)->next;
612 return x;
613 default:
614 return NULL;
615 }
616 }
617
618 static struct grep_expr *compile_pattern_not(struct grep_pat **list)
619 {
620 struct grep_pat *p;
621 struct grep_expr *x;
622
623 p = *list;
624 if (!p)
625 return NULL;
626 switch (p->token) {
627 case GREP_NOT:
628 if (!p->next)
629 die("--not not followed by pattern expression");
630 *list = p->next;
631 x = compile_pattern_not(list);
632 if (!x)
633 die("--not followed by non pattern expression");
634 return grep_not_expr(x);
635 default:
636 return compile_pattern_atom(list);
637 }
638 }
639
640 static struct grep_expr *compile_pattern_and(struct grep_pat **list)
641 {
642 struct grep_pat *p;
643 struct grep_expr *x, *y;
644
645 x = compile_pattern_not(list);
646 p = *list;
647 if (p && p->token == GREP_AND) {
648 if (!x)
649 die("--and not preceded by pattern expression");
650 if (!p->next)
651 die("--and not followed by pattern expression");
652 *list = p->next;
653 y = compile_pattern_and(list);
654 if (!y)
655 die("--and not followed by pattern expression");
656 return grep_and_expr(x, y);
657 }
658 return x;
659 }
660
661 static struct grep_expr *compile_pattern_or(struct grep_pat **list)
662 {
663 struct grep_pat *p;
664 struct grep_expr *x, *y;
665
666 x = compile_pattern_and(list);
667 p = *list;
668 if (x && p && p->token != GREP_CLOSE_PAREN) {
669 y = compile_pattern_or(list);
670 if (!y)
671 die("not a pattern expression %s", p->pattern);
672 return grep_or_expr(x, y);
673 }
674 return x;
675 }
676
677 static struct grep_expr *compile_pattern_expr(struct grep_pat **list)
678 {
679 return compile_pattern_or(list);
680 }
681
682 static struct grep_expr *grep_true_expr(void)
683 {
684 struct grep_expr *z = xcalloc(1, sizeof(*z));
685 z->node = GREP_NODE_TRUE;
686 return z;
687 }
688
689 static struct grep_expr *prep_header_patterns(struct grep_opt *opt)
690 {
691 struct grep_pat *p;
692 struct grep_expr *header_expr;
693 struct grep_expr *(header_group[GREP_HEADER_FIELD_MAX]);
694 enum grep_header_field fld;
695
696 if (!opt->header_list)
697 return NULL;
698
699 for (p = opt->header_list; p; p = p->next) {
700 if (p->token != GREP_PATTERN_HEAD)
701 BUG("a non-header pattern in grep header list.");
702 if (p->field < GREP_HEADER_FIELD_MIN ||
703 GREP_HEADER_FIELD_MAX <= p->field)
704 BUG("unknown header field %d", p->field);
705 compile_regexp(p, opt);
706 }
707
708 for (fld = 0; fld < GREP_HEADER_FIELD_MAX; fld++)
709 header_group[fld] = NULL;
710
711 for (p = opt->header_list; p; p = p->next) {
712 struct grep_expr *h;
713 struct grep_pat *pp = p;
714
715 h = compile_pattern_atom(&pp);
716 if (!h || pp != p->next)
717 BUG("malformed header expr");
718 if (!header_group[p->field]) {
719 header_group[p->field] = h;
720 continue;
721 }
722 header_group[p->field] = grep_or_expr(h, header_group[p->field]);
723 }
724
725 header_expr = NULL;
726
727 for (fld = 0; fld < GREP_HEADER_FIELD_MAX; fld++) {
728 if (!header_group[fld])
729 continue;
730 if (!header_expr)
731 header_expr = grep_true_expr();
732 header_expr = grep_or_expr(header_group[fld], header_expr);
733 }
734 return header_expr;
735 }
736
737 static struct grep_expr *grep_splice_or(struct grep_expr *x, struct grep_expr *y)
738 {
739 struct grep_expr *z = x;
740
741 while (x) {
742 assert(x->node == GREP_NODE_OR);
743 if (x->u.binary.right &&
744 x->u.binary.right->node == GREP_NODE_TRUE) {
745 x->u.binary.right = y;
746 break;
747 }
748 x = x->u.binary.right;
749 }
750 return z;
751 }
752
753 void compile_grep_patterns(struct grep_opt *opt)
754 {
755 struct grep_pat *p;
756 struct grep_expr *header_expr = prep_header_patterns(opt);
757 int extended = 0;
758
759 for (p = opt->pattern_list; p; p = p->next) {
760 switch (p->token) {
761 case GREP_PATTERN: /* atom */
762 case GREP_PATTERN_HEAD:
763 case GREP_PATTERN_BODY:
764 compile_regexp(p, opt);
765 break;
766 default:
767 extended = 1;
768 break;
769 }
770 }
771
772 if (opt->all_match || opt->no_body_match || header_expr)
773 extended = 1;
774 else if (!extended)
775 return;
776
777 p = opt->pattern_list;
778 if (p)
779 opt->pattern_expression = compile_pattern_expr(&p);
780 if (p)
781 die("incomplete pattern expression: %s", p->pattern);
782
783 if (opt->no_body_match && opt->pattern_expression)
784 opt->pattern_expression = grep_not_expr(opt->pattern_expression);
785
786 if (!header_expr)
787 return;
788
789 if (!opt->pattern_expression)
790 opt->pattern_expression = header_expr;
791 else if (opt->all_match)
792 opt->pattern_expression = grep_splice_or(header_expr,
793 opt->pattern_expression);
794 else
795 opt->pattern_expression = grep_or_expr(opt->pattern_expression,
796 header_expr);
797 opt->all_match = 1;
798 }
799
800 static void free_pattern_expr(struct grep_expr *x)
801 {
802 switch (x->node) {
803 case GREP_NODE_TRUE:
804 case GREP_NODE_ATOM:
805 break;
806 case GREP_NODE_NOT:
807 free_pattern_expr(x->u.unary);
808 break;
809 case GREP_NODE_AND:
810 case GREP_NODE_OR:
811 free_pattern_expr(x->u.binary.left);
812 free_pattern_expr(x->u.binary.right);
813 break;
814 }
815 free(x);
816 }
817
818 void free_grep_patterns(struct grep_opt *opt)
819 {
820 struct grep_pat *p, *n;
821
822 for (p = opt->pattern_list; p; p = n) {
823 n = p->next;
824 switch (p->token) {
825 case GREP_PATTERN: /* atom */
826 case GREP_PATTERN_HEAD:
827 case GREP_PATTERN_BODY:
828 if (p->pcre2_pattern)
829 free_pcre2_pattern(p);
830 else
831 regfree(&p->regexp);
832 free(p->pattern);
833 break;
834 default:
835 break;
836 }
837 free(p);
838 }
839
840 if (!opt->pattern_expression)
841 return;
842 free_pattern_expr(opt->pattern_expression);
843 }
844
845 static const char *end_of_line(const char *cp, unsigned long *left)
846 {
847 unsigned long l = *left;
848 while (l && *cp != '\n') {
849 l--;
850 cp++;
851 }
852 *left = l;
853 return cp;
854 }
855
856 static int word_char(char ch)
857 {
858 return isalnum(ch) || ch == '_';
859 }
860
861 static void output_color(struct grep_opt *opt, const void *data, size_t size,
862 const char *color)
863 {
864 if (want_color(opt->color) && color && color[0]) {
865 opt->output(opt, color, strlen(color));
866 opt->output(opt, data, size);
867 opt->output(opt, GIT_COLOR_RESET, strlen(GIT_COLOR_RESET));
868 } else
869 opt->output(opt, data, size);
870 }
871
872 static void output_sep(struct grep_opt *opt, char sign)
873 {
874 if (opt->null_following_name)
875 opt->output(opt, "\0", 1);
876 else
877 output_color(opt, &sign, 1, opt->colors[GREP_COLOR_SEP]);
878 }
879
880 static void show_name(struct grep_opt *opt, const char *name)
881 {
882 output_color(opt, name, strlen(name), opt->colors[GREP_COLOR_FILENAME]);
883 opt->output(opt, opt->null_following_name ? "\0" : "\n", 1);
884 }
885
886 static int patmatch(struct grep_pat *p,
887 const char *line, const char *eol,
888 regmatch_t *match, int eflags)
889 {
890 int hit;
891
892 if (p->pcre2_pattern)
893 hit = !pcre2match(p, line, eol, match, eflags);
894 else
895 hit = !regexec_buf(&p->regexp, line, eol - line, 1, match,
896 eflags);
897
898 return hit;
899 }
900
901 static void strip_timestamp(const char *bol, const char **eol_p)
902 {
903 const char *eol = *eol_p;
904
905 while (bol < --eol) {
906 if (*eol != '>')
907 continue;
908 *eol_p = ++eol;
909 break;
910 }
911 }
912
913 static struct {
914 const char *field;
915 size_t len;
916 } header_field[] = {
917 { "author ", 7 },
918 { "committer ", 10 },
919 { "reflog ", 7 },
920 };
921
922 static int headerless_match_one_pattern(struct grep_pat *p,
923 const char *bol, const char *eol,
924 enum grep_context ctx,
925 regmatch_t *pmatch, int eflags)
926 {
927 int hit = 0;
928 const char *start = bol;
929
930 if ((p->token != GREP_PATTERN) &&
931 ((p->token == GREP_PATTERN_HEAD) != (ctx == GREP_CONTEXT_HEAD)))
932 return 0;
933
934 again:
935 hit = patmatch(p, bol, eol, pmatch, eflags);
936
937 if (hit && p->word_regexp) {
938 if ((pmatch[0].rm_so < 0) ||
939 (eol - bol) < pmatch[0].rm_so ||
940 (pmatch[0].rm_eo < 0) ||
941 (eol - bol) < pmatch[0].rm_eo)
942 die("regexp returned nonsense");
943
944 /* Match beginning must be either beginning of the
945 * line, or at word boundary (i.e. the last char must
946 * not be a word char). Similarly, match end must be
947 * either end of the line, or at word boundary
948 * (i.e. the next char must not be a word char).
949 */
950 if ( ((pmatch[0].rm_so == 0) ||
951 !word_char(bol[pmatch[0].rm_so-1])) &&
952 ((pmatch[0].rm_eo == (eol-bol)) ||
953 !word_char(bol[pmatch[0].rm_eo])) )
954 ;
955 else
956 hit = 0;
957
958 /* Words consist of at least one character. */
959 if (pmatch->rm_so == pmatch->rm_eo)
960 hit = 0;
961
962 if (!hit && pmatch[0].rm_so + bol + 1 < eol) {
963 /* There could be more than one match on the
964 * line, and the first match might not be
965 * strict word match. But later ones could be!
966 * Forward to the next possible start, i.e. the
967 * next position following a non-word char.
968 */
969 bol = pmatch[0].rm_so + bol + 1;
970 while (word_char(bol[-1]) && bol < eol)
971 bol++;
972 eflags |= REG_NOTBOL;
973 if (bol < eol)
974 goto again;
975 }
976 }
977 if (hit) {
978 pmatch[0].rm_so += bol - start;
979 pmatch[0].rm_eo += bol - start;
980 }
981 return hit;
982 }
983
984 static int match_one_pattern(struct grep_pat *p,
985 const char *bol, const char *eol,
986 enum grep_context ctx, regmatch_t *pmatch,
987 int eflags)
988 {
989 const char *field;
990 size_t len;
991
992 if (p->token == GREP_PATTERN_HEAD) {
993 assert(p->field < ARRAY_SIZE(header_field));
994 field = header_field[p->field].field;
995 len = header_field[p->field].len;
996 if (strncmp(bol, field, len))
997 return 0;
998 bol += len;
999
1000 switch (p->field) {
1001 case GREP_HEADER_AUTHOR:
1002 case GREP_HEADER_COMMITTER:
1003 strip_timestamp(bol, &eol);
1004 break;
1005 default:
1006 break;
1007 }
1008 }
1009
1010 return headerless_match_one_pattern(p, bol, eol, ctx, pmatch, eflags);
1011 }
1012
1013
1014 static int match_expr_eval(struct grep_opt *opt, struct grep_expr *x,
1015 const char *bol, const char *eol,
1016 enum grep_context ctx, ssize_t *col,
1017 ssize_t *icol, int collect_hits)
1018 {
1019 int h = 0;
1020
1021 switch (x->node) {
1022 case GREP_NODE_TRUE:
1023 h = 1;
1024 break;
1025 case GREP_NODE_ATOM:
1026 {
1027 regmatch_t tmp;
1028 h = match_one_pattern(x->u.atom, bol, eol, ctx,
1029 &tmp, 0);
1030 if (h && (*col < 0 || tmp.rm_so < *col))
1031 *col = tmp.rm_so;
1032 }
1033 if (x->u.atom->token == GREP_PATTERN_BODY)
1034 opt->body_hit |= h;
1035 break;
1036 case GREP_NODE_NOT:
1037 /*
1038 * Upon visiting a GREP_NODE_NOT, col and icol become swapped.
1039 */
1040 h = !match_expr_eval(opt, x->u.unary, bol, eol, ctx, icol, col,
1041 0);
1042 break;
1043 case GREP_NODE_AND:
1044 h = match_expr_eval(opt, x->u.binary.left, bol, eol, ctx, col,
1045 icol, 0);
1046 if (h || opt->columnnum) {
1047 /*
1048 * Don't short-circuit AND when given --column, since a
1049 * NOT earlier in the tree may turn this into an OR. In
1050 * this case, see the below comment.
1051 */
1052 h &= match_expr_eval(opt, x->u.binary.right, bol, eol,
1053 ctx, col, icol, 0);
1054 }
1055 break;
1056 case GREP_NODE_OR:
1057 if (!(collect_hits || opt->columnnum)) {
1058 /*
1059 * Don't short-circuit OR when given --column (or
1060 * collecting hits) to ensure we don't skip a later
1061 * child that would produce an earlier match.
1062 */
1063 return (match_expr_eval(opt, x->u.binary.left, bol, eol,
1064 ctx, col, icol, 0) ||
1065 match_expr_eval(opt, x->u.binary.right, bol,
1066 eol, ctx, col, icol, 0));
1067 }
1068 h = match_expr_eval(opt, x->u.binary.left, bol, eol, ctx, col,
1069 icol, 0);
1070 if (collect_hits)
1071 x->u.binary.left->hit |= h;
1072 h |= match_expr_eval(opt, x->u.binary.right, bol, eol, ctx, col,
1073 icol, collect_hits);
1074 break;
1075 default:
1076 die("Unexpected node type (internal error) %d", x->node);
1077 }
1078 if (collect_hits)
1079 x->hit |= h;
1080 return h;
1081 }
1082
1083 static int match_expr(struct grep_opt *opt,
1084 const char *bol, const char *eol,
1085 enum grep_context ctx, ssize_t *col,
1086 ssize_t *icol, int collect_hits)
1087 {
1088 struct grep_expr *x = opt->pattern_expression;
1089 return match_expr_eval(opt, x, bol, eol, ctx, col, icol, collect_hits);
1090 }
1091
1092 static int match_line(struct grep_opt *opt,
1093 const char *bol, const char *eol,
1094 ssize_t *col, ssize_t *icol,
1095 enum grep_context ctx, int collect_hits)
1096 {
1097 struct grep_pat *p;
1098 int hit = 0;
1099
1100 if (opt->pattern_expression)
1101 return match_expr(opt, bol, eol, ctx, col, icol,
1102 collect_hits);
1103
1104 /* we do not call with collect_hits without being extended */
1105 for (p = opt->pattern_list; p; p = p->next) {
1106 regmatch_t tmp;
1107 if (match_one_pattern(p, bol, eol, ctx, &tmp, 0)) {
1108 hit |= 1;
1109 if (!opt->columnnum) {
1110 /*
1111 * Without --column, any single match on a line
1112 * is enough to know that it needs to be
1113 * printed. With --column, scan _all_ patterns
1114 * to find the earliest.
1115 */
1116 break;
1117 }
1118 if (*col < 0 || tmp.rm_so < *col)
1119 *col = tmp.rm_so;
1120 }
1121 }
1122 return hit;
1123 }
1124
1125 static int match_next_pattern(struct grep_pat *p,
1126 const char *bol, const char *eol,
1127 enum grep_context ctx,
1128 regmatch_t *pmatch, int eflags)
1129 {
1130 regmatch_t match;
1131
1132 if (!headerless_match_one_pattern(p, bol, eol, ctx, &match, eflags))
1133 return 0;
1134 if (match.rm_so < 0 || match.rm_eo < 0)
1135 return 0;
1136 if (pmatch->rm_so >= 0 && pmatch->rm_eo >= 0) {
1137 if (match.rm_so > pmatch->rm_so)
1138 return 1;
1139 if (match.rm_so == pmatch->rm_so && match.rm_eo < pmatch->rm_eo)
1140 return 1;
1141 }
1142 pmatch->rm_so = match.rm_so;
1143 pmatch->rm_eo = match.rm_eo;
1144 return 1;
1145 }
1146
1147 int grep_next_match(struct grep_opt *opt,
1148 const char *bol, const char *eol,
1149 enum grep_context ctx, regmatch_t *pmatch,
1150 enum grep_header_field field, int eflags)
1151 {
1152 struct grep_pat *p;
1153 int hit = 0;
1154
1155 pmatch->rm_so = pmatch->rm_eo = -1;
1156 if (bol < eol) {
1157 for (p = ((ctx == GREP_CONTEXT_HEAD)
1158 ? opt->header_list : opt->pattern_list);
1159 p; p = p->next) {
1160 switch (p->token) {
1161 case GREP_PATTERN_HEAD:
1162 if ((field != GREP_HEADER_FIELD_MAX) &&
1163 (p->field != field))
1164 continue;
1165 /* fall thru */
1166 case GREP_PATTERN: /* atom */
1167 case GREP_PATTERN_BODY:
1168 hit |= match_next_pattern(p, bol, eol, ctx,
1169 pmatch, eflags);
1170 break;
1171 default:
1172 break;
1173 }
1174 }
1175 }
1176 return hit;
1177 }
1178
1179 static void show_line_header(struct grep_opt *opt, const char *name,
1180 unsigned lno, ssize_t cno, char sign)
1181 {
1182 if (opt->heading && opt->last_shown == 0) {
1183 output_color(opt, name, strlen(name), opt->colors[GREP_COLOR_FILENAME]);
1184 opt->output(opt, "\n", 1);
1185 }
1186 opt->last_shown = lno;
1187
1188 if (!opt->heading && opt->pathname) {
1189 output_color(opt, name, strlen(name), opt->colors[GREP_COLOR_FILENAME]);
1190 output_sep(opt, sign);
1191 }
1192 if (opt->linenum) {
1193 char buf[32];
1194 xsnprintf(buf, sizeof(buf), "%d", lno);
1195 output_color(opt, buf, strlen(buf), opt->colors[GREP_COLOR_LINENO]);
1196 output_sep(opt, sign);
1197 }
1198 /*
1199 * Treat 'cno' as the 1-indexed offset from the start of a non-context
1200 * line to its first match. Otherwise, 'cno' is 0 indicating that we are
1201 * being called with a context line.
1202 */
1203 if (opt->columnnum && cno) {
1204 char buf[32];
1205 xsnprintf(buf, sizeof(buf), "%"PRIuMAX, (uintmax_t)cno);
1206 output_color(opt, buf, strlen(buf), opt->colors[GREP_COLOR_COLUMNNO]);
1207 output_sep(opt, sign);
1208 }
1209 }
1210
1211 static void show_line(struct grep_opt *opt,
1212 const char *bol, const char *eol,
1213 const char *name, unsigned lno, ssize_t cno, char sign)
1214 {
1215 int rest = eol - bol;
1216 const char *match_color = NULL;
1217 const char *line_color = NULL;
1218
1219 if (opt->file_break && opt->last_shown == 0) {
1220 if (opt->show_hunk_mark)
1221 opt->output(opt, "\n", 1);
1222 } else if (opt->pre_context || opt->post_context || opt->funcbody) {
1223 if (opt->last_shown == 0) {
1224 if (opt->show_hunk_mark) {
1225 output_color(opt, "--", 2, opt->colors[GREP_COLOR_SEP]);
1226 opt->output(opt, "\n", 1);
1227 }
1228 } else if (lno > opt->last_shown + 1) {
1229 output_color(opt, "--", 2, opt->colors[GREP_COLOR_SEP]);
1230 opt->output(opt, "\n", 1);
1231 }
1232 }
1233 if (!opt->only_matching) {
1234 /*
1235 * In case the line we're being called with contains more than
1236 * one match, leave printing each header to the loop below.
1237 */
1238 show_line_header(opt, name, lno, cno, sign);
1239 }
1240 if (opt->color || opt->only_matching) {
1241 regmatch_t match;
1242 enum grep_context ctx = GREP_CONTEXT_BODY;
1243 int eflags = 0;
1244
1245 if (opt->color) {
1246 if (sign == ':')
1247 match_color = opt->colors[GREP_COLOR_MATCH_SELECTED];
1248 else
1249 match_color = opt->colors[GREP_COLOR_MATCH_CONTEXT];
1250 if (sign == ':')
1251 line_color = opt->colors[GREP_COLOR_SELECTED];
1252 else if (sign == '-')
1253 line_color = opt->colors[GREP_COLOR_CONTEXT];
1254 else if (sign == '=')
1255 line_color = opt->colors[GREP_COLOR_FUNCTION];
1256 }
1257 while (grep_next_match(opt, bol, eol, ctx, &match,
1258 GREP_HEADER_FIELD_MAX, eflags)) {
1259 if (match.rm_so == match.rm_eo)
1260 break;
1261
1262 if (opt->only_matching)
1263 show_line_header(opt, name, lno, cno, sign);
1264 else
1265 output_color(opt, bol, match.rm_so, line_color);
1266 output_color(opt, bol + match.rm_so,
1267 match.rm_eo - match.rm_so, match_color);
1268 if (opt->only_matching)
1269 opt->output(opt, "\n", 1);
1270 bol += match.rm_eo;
1271 cno += match.rm_eo;
1272 rest -= match.rm_eo;
1273 eflags = REG_NOTBOL;
1274 }
1275 }
1276 if (!opt->only_matching) {
1277 output_color(opt, bol, rest, line_color);
1278 opt->output(opt, "\n", 1);
1279 }
1280 }
1281
1282 int grep_use_locks;
1283
1284 /*
1285 * This lock protects access to the gitattributes machinery, which is
1286 * not thread-safe.
1287 */
1288 pthread_mutex_t grep_attr_mutex;
1289
1290 static inline void grep_attr_lock(void)
1291 {
1292 if (grep_use_locks)
1293 pthread_mutex_lock(&grep_attr_mutex);
1294 }
1295
1296 static inline void grep_attr_unlock(void)
1297 {
1298 if (grep_use_locks)
1299 pthread_mutex_unlock(&grep_attr_mutex);
1300 }
1301
1302 static int match_funcname(struct grep_opt *opt, struct grep_source *gs,
1303 const char *bol, const char *eol)
1304 {
1305 xdemitconf_t *xecfg = opt->priv;
1306 if (xecfg && !xecfg->find_func) {
1307 grep_source_load_driver(gs, opt->repo->index);
1308 if (gs->driver->funcname.pattern) {
1309 const struct userdiff_funcname *pe = &gs->driver->funcname;
1310 xdiff_set_find_func(xecfg, pe->pattern, pe->cflags);
1311 } else {
1312 xecfg = opt->priv = NULL;
1313 }
1314 }
1315
1316 if (xecfg) {
1317 char buf[1];
1318 return xecfg->find_func(bol, eol - bol, buf, 1,
1319 xecfg->find_func_priv) >= 0;
1320 }
1321
1322 if (bol == eol)
1323 return 0;
1324 if (isalpha(*bol) || *bol == '_' || *bol == '$')
1325 return 1;
1326 return 0;
1327 }
1328
1329 static void show_funcname_line(struct grep_opt *opt, struct grep_source *gs,
1330 const char *bol, unsigned lno)
1331 {
1332 while (bol > gs->buf) {
1333 const char *eol = --bol;
1334
1335 while (bol > gs->buf && bol[-1] != '\n')
1336 bol--;
1337 lno--;
1338
1339 if (lno <= opt->last_shown)
1340 break;
1341
1342 if (match_funcname(opt, gs, bol, eol)) {
1343 show_line(opt, bol, eol, gs->name, lno, 0, '=');
1344 break;
1345 }
1346 }
1347 }
1348
1349 static int is_empty_line(const char *bol, const char *eol);
1350
1351 static void show_pre_context(struct grep_opt *opt, struct grep_source *gs,
1352 const char *bol, const char *end, unsigned lno)
1353 {
1354 unsigned cur = lno, from = 1, funcname_lno = 0, orig_from;
1355 int funcname_needed = !!opt->funcname, comment_needed = 0;
1356
1357 if (opt->pre_context < lno)
1358 from = lno - opt->pre_context;
1359 if (from <= opt->last_shown)
1360 from = opt->last_shown + 1;
1361 orig_from = from;
1362 if (opt->funcbody) {
1363 if (match_funcname(opt, gs, bol, end))
1364 comment_needed = 1;
1365 else
1366 funcname_needed = 1;
1367 from = opt->last_shown + 1;
1368 }
1369
1370 /* Rewind. */
1371 while (bol > gs->buf && cur > from) {
1372 const char *next_bol = bol;
1373 const char *eol = --bol;
1374
1375 while (bol > gs->buf && bol[-1] != '\n')
1376 bol--;
1377 cur--;
1378 if (comment_needed && (is_empty_line(bol, eol) ||
1379 match_funcname(opt, gs, bol, eol))) {
1380 comment_needed = 0;
1381 from = orig_from;
1382 if (cur < from) {
1383 cur++;
1384 bol = next_bol;
1385 break;
1386 }
1387 }
1388 if (funcname_needed && match_funcname(opt, gs, bol, eol)) {
1389 funcname_lno = cur;
1390 funcname_needed = 0;
1391 if (opt->funcbody)
1392 comment_needed = 1;
1393 else
1394 from = orig_from;
1395 }
1396 }
1397
1398 /* We need to look even further back to find a function signature. */
1399 if (opt->funcname && funcname_needed)
1400 show_funcname_line(opt, gs, bol, cur);
1401
1402 /* Back forward. */
1403 while (cur < lno) {
1404 const char *eol = bol, sign = (cur == funcname_lno) ? '=' : '-';
1405
1406 while (*eol != '\n')
1407 eol++;
1408 show_line(opt, bol, eol, gs->name, cur, 0, sign);
1409 bol = eol + 1;
1410 cur++;
1411 }
1412 }
1413
1414 static int should_lookahead(struct grep_opt *opt)
1415 {
1416 struct grep_pat *p;
1417
1418 if (opt->pattern_expression)
1419 return 0; /* punt for too complex stuff */
1420 if (opt->invert)
1421 return 0;
1422 for (p = opt->pattern_list; p; p = p->next) {
1423 if (p->token != GREP_PATTERN)
1424 return 0; /* punt for "header only" and stuff */
1425 }
1426 return 1;
1427 }
1428
1429 static int look_ahead(struct grep_opt *opt,
1430 unsigned long *left_p,
1431 unsigned *lno_p,
1432 const char **bol_p)
1433 {
1434 unsigned lno = *lno_p;
1435 const char *bol = *bol_p;
1436 struct grep_pat *p;
1437 const char *sp, *last_bol;
1438 regoff_t earliest = -1;
1439
1440 for (p = opt->pattern_list; p; p = p->next) {
1441 int hit;
1442 regmatch_t m;
1443
1444 hit = patmatch(p, bol, bol + *left_p, &m, 0);
1445 if (!hit || m.rm_so < 0 || m.rm_eo < 0)
1446 continue;
1447 if (earliest < 0 || m.rm_so < earliest)
1448 earliest = m.rm_so;
1449 }
1450
1451 if (earliest < 0) {
1452 *bol_p = bol + *left_p;
1453 *left_p = 0;
1454 return 1;
1455 }
1456 for (sp = bol + earliest; bol < sp && sp[-1] != '\n'; sp--)
1457 ; /* find the beginning of the line */
1458 last_bol = sp;
1459
1460 for (sp = bol; sp < last_bol; sp++) {
1461 if (*sp == '\n')
1462 lno++;
1463 }
1464 *left_p -= last_bol - bol;
1465 *bol_p = last_bol;
1466 *lno_p = lno;
1467 return 0;
1468 }
1469
1470 static int fill_textconv_grep(struct repository *r,
1471 struct userdiff_driver *driver,
1472 struct grep_source *gs)
1473 {
1474 struct diff_filespec *df;
1475 char *buf;
1476 size_t size;
1477
1478 if (!driver || !driver->textconv)
1479 return grep_source_load(gs);
1480
1481 /*
1482 * The textconv interface is intimately tied to diff_filespecs, so we
1483 * have to pretend to be one. If we could unify the grep_source
1484 * and diff_filespec structs, this mess could just go away.
1485 */
1486 df = alloc_filespec(gs->path);
1487 switch (gs->type) {
1488 case GREP_SOURCE_OID:
1489 fill_filespec(df, gs->identifier, 1, 0100644);
1490 break;
1491 case GREP_SOURCE_FILE:
1492 fill_filespec(df, null_oid(), 0, 0100644);
1493 break;
1494 default:
1495 BUG("attempt to textconv something without a path?");
1496 }
1497
1498 /*
1499 * fill_textconv is not remotely thread-safe; it modifies the global
1500 * diff tempfile structure, writes to the_repo's odb and might
1501 * internally call thread-unsafe functions such as the
1502 * prepare_packed_git() lazy-initializator. Because of the last two, we
1503 * must ensure mutual exclusion between this call and the object reading
1504 * API, thus we use obj_read_lock() here.
1505 *
1506 * TODO: allowing text conversion to run in parallel with object
1507 * reading operations might increase performance in the multithreaded
1508 * non-worktreee git-grep with --textconv.
1509 */
1510 obj_read_lock();
1511 size = fill_textconv(r, driver, df, &buf);
1512 obj_read_unlock();
1513 free_filespec(df);
1514
1515 /*
1516 * The normal fill_textconv usage by the diff machinery would just keep
1517 * the textconv'd buf separate from the diff_filespec. But much of the
1518 * grep code passes around a grep_source and assumes that its "buf"
1519 * pointer is the beginning of the thing we are searching. So let's
1520 * install our textconv'd version into the grep_source, taking care not
1521 * to leak any existing buffer.
1522 */
1523 grep_source_clear_data(gs);
1524 gs->buf = buf;
1525 gs->size = size;
1526
1527 return 0;
1528 }
1529
1530 static int is_empty_line(const char *bol, const char *eol)
1531 {
1532 while (bol < eol && isspace(*bol))
1533 bol++;
1534 return bol == eol;
1535 }
1536
1537 static int grep_source_1(struct grep_opt *opt, struct grep_source *gs, int collect_hits)
1538 {
1539 const char *bol;
1540 const char *peek_bol = NULL;
1541 unsigned long left;
1542 unsigned lno = 1;
1543 unsigned last_hit = 0;
1544 int binary_match_only = 0;
1545 unsigned count = 0;
1546 int try_lookahead = 0;
1547 int show_function = 0;
1548 struct userdiff_driver *textconv = NULL;
1549 enum grep_context ctx = GREP_CONTEXT_HEAD;
1550 xdemitconf_t xecfg;
1551
1552 if (!opt->status_only && gs->name == NULL)
1553 BUG("grep call which could print a name requires "
1554 "grep_source.name be non-NULL");
1555
1556 if (!opt->output)
1557 opt->output = std_output;
1558
1559 if (opt->pre_context || opt->post_context || opt->file_break ||
1560 opt->funcbody) {
1561 /* Show hunk marks, except for the first file. */
1562 if (opt->last_shown)
1563 opt->show_hunk_mark = 1;
1564 /*
1565 * If we're using threads then we can't easily identify
1566 * the first file. Always put hunk marks in that case
1567 * and skip the very first one later in work_done().
1568 */
1569 if (opt->output != std_output)
1570 opt->show_hunk_mark = 1;
1571 }
1572 opt->last_shown = 0;
1573
1574 if (opt->allow_textconv) {
1575 grep_source_load_driver(gs, opt->repo->index);
1576 /*
1577 * We might set up the shared textconv cache data here, which
1578 * is not thread-safe. Also, get_oid_with_context() and
1579 * parse_object() might be internally called. As they are not
1580 * currently thread-safe and might be racy with object reading,
1581 * obj_read_lock() must be called.
1582 */
1583 grep_attr_lock();
1584 obj_read_lock();
1585 textconv = userdiff_get_textconv(opt->repo, gs->driver);
1586 obj_read_unlock();
1587 grep_attr_unlock();
1588 }
1589
1590 /*
1591 * We know the result of a textconv is text, so we only have to care
1592 * about binary handling if we are not using it.
1593 */
1594 if (!textconv) {
1595 switch (opt->binary) {
1596 case GREP_BINARY_DEFAULT:
1597 if (grep_source_is_binary(gs, opt->repo->index))
1598 binary_match_only = 1;
1599 break;
1600 case GREP_BINARY_NOMATCH:
1601 if (grep_source_is_binary(gs, opt->repo->index))
1602 return 0; /* Assume unmatch */
1603 break;
1604 case GREP_BINARY_TEXT:
1605 break;
1606 default:
1607 BUG("unknown binary handling mode");
1608 }
1609 }
1610
1611 memset(&xecfg, 0, sizeof(xecfg));
1612 opt->priv = &xecfg;
1613
1614 try_lookahead = should_lookahead(opt);
1615
1616 if (fill_textconv_grep(opt->repo, textconv, gs) < 0)
1617 return 0;
1618
1619 bol = gs->buf;
1620 left = gs->size;
1621 while (left) {
1622 const char *eol;
1623 int hit;
1624 ssize_t cno;
1625 ssize_t col = -1, icol = -1;
1626
1627 /*
1628 * look_ahead() skips quickly to the line that possibly
1629 * has the next hit; don't call it if we need to do
1630 * something more than just skipping the current line
1631 * in response to an unmatch for the current line. E.g.
1632 * inside a post-context window, we will show the current
1633 * line as a context around the previous hit when it
1634 * doesn't hit.
1635 */
1636 if (try_lookahead
1637 && !(last_hit
1638 && (show_function ||
1639 lno <= last_hit + opt->post_context))
1640 && look_ahead(opt, &left, &lno, &bol))
1641 break;
1642 eol = end_of_line(bol, &left);
1643
1644 if ((ctx == GREP_CONTEXT_HEAD) && (eol == bol))
1645 ctx = GREP_CONTEXT_BODY;
1646
1647 hit = match_line(opt, bol, eol, &col, &icol, ctx, collect_hits);
1648
1649 if (collect_hits)
1650 goto next_line;
1651
1652 /* "grep -v -e foo -e bla" should list lines
1653 * that do not have either, so inversion should
1654 * be done outside.
1655 */
1656 if (opt->invert)
1657 hit = !hit;
1658 if (opt->unmatch_name_only) {
1659 if (hit)
1660 return 0;
1661 goto next_line;
1662 }
1663 if (hit && (opt->max_count < 0 || count < opt->max_count)) {
1664 count++;
1665 if (opt->status_only)
1666 return 1;
1667 if (opt->name_only) {
1668 show_name(opt, gs->name);
1669 return 1;
1670 }
1671 if (opt->count)
1672 goto next_line;
1673 if (binary_match_only) {
1674 opt->output(opt, "Binary file ", 12);
1675 output_color(opt, gs->name, strlen(gs->name),
1676 opt->colors[GREP_COLOR_FILENAME]);
1677 opt->output(opt, " matches\n", 9);
1678 return 1;
1679 }
1680 /* Hit at this line. If we haven't shown the
1681 * pre-context lines, we would need to show them.
1682 */
1683 if (opt->pre_context || opt->funcbody)
1684 show_pre_context(opt, gs, bol, eol, lno);
1685 else if (opt->funcname)
1686 show_funcname_line(opt, gs, bol, lno);
1687 cno = opt->invert ? icol : col;
1688 if (cno < 0) {
1689 /*
1690 * A negative cno indicates that there was no
1691 * match on the line. We are thus inverted and
1692 * being asked to show all lines that _don't_
1693 * match a given expression. Therefore, set cno
1694 * to 0 to suggest the whole line matches.
1695 */
1696 cno = 0;
1697 }
1698 show_line(opt, bol, eol, gs->name, lno, cno + 1, ':');
1699 last_hit = lno;
1700 if (opt->funcbody)
1701 show_function = 1;
1702 goto next_line;
1703 }
1704 if (show_function && (!peek_bol || peek_bol < bol)) {
1705 unsigned long peek_left = left;
1706 const char *peek_eol = eol;
1707
1708 /*
1709 * Trailing empty lines are not interesting.
1710 * Peek past them to see if they belong to the
1711 * body of the current function.
1712 */
1713 peek_bol = bol;
1714 while (is_empty_line(peek_bol, peek_eol)) {
1715 peek_bol = peek_eol + 1;
1716 peek_eol = end_of_line(peek_bol, &peek_left);
1717 }
1718
1719 if (match_funcname(opt, gs, peek_bol, peek_eol))
1720 show_function = 0;
1721 }
1722 if (show_function ||
1723 (last_hit && lno <= last_hit + opt->post_context)) {
1724 /* If the last hit is within the post context,
1725 * we need to show this line.
1726 */
1727 show_line(opt, bol, eol, gs->name, lno, col + 1, '-');
1728 }
1729
1730 next_line:
1731 bol = eol + 1;
1732 if (!left)
1733 break;
1734 left--;
1735 lno++;
1736 }
1737
1738 if (collect_hits)
1739 return 0;
1740
1741 if (opt->status_only)
1742 return opt->unmatch_name_only;
1743 if (opt->unmatch_name_only) {
1744 /* We did not see any hit, so we want to show this */
1745 show_name(opt, gs->name);
1746 return 1;
1747 }
1748
1749 xdiff_clear_find_func(&xecfg);
1750 opt->priv = NULL;
1751
1752 /* NEEDSWORK:
1753 * The real "grep -c foo *.c" gives many "bar.c:0" lines,
1754 * which feels mostly useless but sometimes useful. Maybe
1755 * make it another option? For now suppress them.
1756 */
1757 if (opt->count && count) {
1758 char buf[32];
1759 if (opt->pathname) {
1760 output_color(opt, gs->name, strlen(gs->name),
1761 opt->colors[GREP_COLOR_FILENAME]);
1762 output_sep(opt, ':');
1763 }
1764 xsnprintf(buf, sizeof(buf), "%u\n", count);
1765 opt->output(opt, buf, strlen(buf));
1766 return 1;
1767 }
1768 return !!last_hit;
1769 }
1770
1771 static void clr_hit_marker(struct grep_expr *x)
1772 {
1773 /* All-hit markers are meaningful only at the very top level
1774 * OR node.
1775 */
1776 while (1) {
1777 x->hit = 0;
1778 if (x->node != GREP_NODE_OR)
1779 return;
1780 x->u.binary.left->hit = 0;
1781 x = x->u.binary.right;
1782 }
1783 }
1784
1785 static int chk_hit_marker(struct grep_expr *x)
1786 {
1787 /* Top level nodes have hit markers. See if they all are hits */
1788 while (1) {
1789 if (x->node != GREP_NODE_OR)
1790 return x->hit;
1791 if (!x->u.binary.left->hit)
1792 return 0;
1793 x = x->u.binary.right;
1794 }
1795 }
1796
1797 int grep_source(struct grep_opt *opt, struct grep_source *gs)
1798 {
1799 /*
1800 * we do not have to do the two-pass grep when we do not check
1801 * buffer-wide "all-match".
1802 */
1803 if (!opt->all_match && !opt->no_body_match)
1804 return grep_source_1(opt, gs, 0);
1805
1806 /* Otherwise the toplevel "or" terms hit a bit differently.
1807 * We first clear hit markers from them.
1808 */
1809 clr_hit_marker(opt->pattern_expression);
1810 opt->body_hit = 0;
1811 grep_source_1(opt, gs, 1);
1812
1813 if (opt->all_match && !chk_hit_marker(opt->pattern_expression))
1814 return 0;
1815 if (opt->no_body_match && opt->body_hit)
1816 return 0;
1817
1818 return grep_source_1(opt, gs, 0);
1819 }
1820
1821 static void grep_source_init_buf(struct grep_source *gs,
1822 const char *buf,
1823 unsigned long size)
1824 {
1825 gs->type = GREP_SOURCE_BUF;
1826 gs->name = NULL;
1827 gs->path = NULL;
1828 gs->buf = buf;
1829 gs->size = size;
1830 gs->driver = NULL;
1831 gs->identifier = NULL;
1832 }
1833
1834 int grep_buffer(struct grep_opt *opt, const char *buf, unsigned long size)
1835 {
1836 struct grep_source gs;
1837 int r;
1838
1839 grep_source_init_buf(&gs, buf, size);
1840
1841 r = grep_source(opt, &gs);
1842
1843 grep_source_clear(&gs);
1844 return r;
1845 }
1846
1847 void grep_source_init_file(struct grep_source *gs, const char *name,
1848 const char *path)
1849 {
1850 gs->type = GREP_SOURCE_FILE;
1851 gs->name = xstrdup_or_null(name);
1852 gs->path = xstrdup_or_null(path);
1853 gs->buf = NULL;
1854 gs->size = 0;
1855 gs->driver = NULL;
1856 gs->identifier = xstrdup(path);
1857 }
1858
1859 void grep_source_init_oid(struct grep_source *gs, const char *name,
1860 const char *path, const struct object_id *oid,
1861 struct repository *repo)
1862 {
1863 gs->type = GREP_SOURCE_OID;
1864 gs->name = xstrdup_or_null(name);
1865 gs->path = xstrdup_or_null(path);
1866 gs->buf = NULL;
1867 gs->size = 0;
1868 gs->driver = NULL;
1869 gs->identifier = oiddup(oid);
1870 gs->repo = repo;
1871 }
1872
1873 void grep_source_clear(struct grep_source *gs)
1874 {
1875 FREE_AND_NULL(gs->name);
1876 FREE_AND_NULL(gs->path);
1877 FREE_AND_NULL(gs->identifier);
1878 grep_source_clear_data(gs);
1879 }
1880
1881 void grep_source_clear_data(struct grep_source *gs)
1882 {
1883 switch (gs->type) {
1884 case GREP_SOURCE_FILE:
1885 case GREP_SOURCE_OID:
1886 /* these types own the buffer */
1887 free((char *)gs->buf);
1888 gs->buf = NULL;
1889 gs->size = 0;
1890 break;
1891 case GREP_SOURCE_BUF:
1892 /* leave user-provided buf intact */
1893 break;
1894 }
1895 }
1896
1897 static int grep_source_load_oid(struct grep_source *gs)
1898 {
1899 enum object_type type;
1900
1901 gs->buf = repo_read_object_file(gs->repo, gs->identifier, &type,
1902 &gs->size);
1903 if (!gs->buf)
1904 return error(_("'%s': unable to read %s"),
1905 gs->name,
1906 oid_to_hex(gs->identifier));
1907 return 0;
1908 }
1909
1910 static int grep_source_load_file(struct grep_source *gs)
1911 {
1912 const char *filename = gs->identifier;
1913 struct stat st;
1914 char *data;
1915 size_t size;
1916 int i;
1917
1918 if (lstat(filename, &st) < 0) {
1919 err_ret:
1920 if (errno != ENOENT)
1921 error_errno(_("failed to stat '%s'"), filename);
1922 return -1;
1923 }
1924 if (!S_ISREG(st.st_mode))
1925 return -1;
1926 size = xsize_t(st.st_size);
1927 i = open(filename, O_RDONLY);
1928 if (i < 0)
1929 goto err_ret;
1930 data = xmallocz(size);
1931 if (st.st_size != read_in_full(i, data, size)) {
1932 error_errno(_("'%s': short read"), filename);
1933 close(i);
1934 free(data);
1935 return -1;
1936 }
1937 close(i);
1938
1939 gs->buf = data;
1940 gs->size = size;
1941 return 0;
1942 }
1943
1944 static int grep_source_load(struct grep_source *gs)
1945 {
1946 if (gs->buf)
1947 return 0;
1948
1949 switch (gs->type) {
1950 case GREP_SOURCE_FILE:
1951 return grep_source_load_file(gs);
1952 case GREP_SOURCE_OID:
1953 return grep_source_load_oid(gs);
1954 case GREP_SOURCE_BUF:
1955 return gs->buf ? 0 : -1;
1956 }
1957 BUG("invalid grep_source type to load");
1958 }
1959
1960 void grep_source_load_driver(struct grep_source *gs,
1961 struct index_state *istate)
1962 {
1963 if (gs->driver)
1964 return;
1965
1966 grep_attr_lock();
1967 if (gs->path)
1968 gs->driver = userdiff_find_by_path(istate, gs->path);
1969 if (!gs->driver)
1970 gs->driver = userdiff_find_by_name("default");
1971 grep_attr_unlock();
1972 }
1973
1974 static int grep_source_is_binary(struct grep_source *gs,
1975 struct index_state *istate)
1976 {
1977 grep_source_load_driver(gs, istate);
1978 if (gs->driver->binary != -1)
1979 return gs->driver->binary;
1980
1981 if (!grep_source_load(gs))
1982 return buffer_is_binary(gs->buf, gs->size);
1983
1984 return 0;
1985 }
git for Debian