git

1
#include "git-compat-util.h"
2
#include "config.h"
3
#include "gettext.h"
4
#include "grep.h"
5
#include "hex.h"
6
#include "object-store-ll.h"
7
#include "pretty.h"
8
#include "userdiff.h"
9
#include "xdiff-interface.h"
10
#include "diff.h"
11
#include "diffcore.h"
12
#include "quote.h"
13
#include "help.h"
14

15
static int grep_source_load(struct grep_source *gs);
16
static int grep_source_is_binary(struct grep_source *gs,
17
				 struct index_state *istate);
18

19
static void std_output(struct grep_opt *opt UNUSED, const void *buf, size_t size)
20
{
21
	fwrite(buf, size, 1, stdout);
22
}
23

24
static const char *color_grep_slots[] = {
25
	[GREP_COLOR_CONTEXT]	    = "context",
26
	[GREP_COLOR_FILENAME]	    = "filename",
27
	[GREP_COLOR_FUNCTION]	    = "function",
28
	[GREP_COLOR_LINENO]	    = "lineNumber",
29
	[GREP_COLOR_COLUMNNO]	    = "column",
30
	[GREP_COLOR_MATCH_CONTEXT]  = "matchContext",
31
	[GREP_COLOR_MATCH_SELECTED] = "matchSelected",
32
	[GREP_COLOR_SELECTED]	    = "selected",
33
	[GREP_COLOR_SEP]	    = "separator",
34
};
35

36
static int parse_pattern_type_arg(const char *opt, const char *arg)
37
{
38
	if (!strcmp(arg, "default"))
39
		return GREP_PATTERN_TYPE_UNSPECIFIED;
40
	else if (!strcmp(arg, "basic"))
41
		return GREP_PATTERN_TYPE_BRE;
42
	else if (!strcmp(arg, "extended"))
43
		return GREP_PATTERN_TYPE_ERE;
44
	else if (!strcmp(arg, "fixed"))
45
		return GREP_PATTERN_TYPE_FIXED;
46
	else if (!strcmp(arg, "perl"))
47
		return GREP_PATTERN_TYPE_PCRE;
48
	die("bad %s argument: %s", opt, arg);
49
}
50

51
define_list_config_array_extra(color_grep_slots, {"match"});
52

53
/*
54
 * Read the configuration file once and store it in
55
 * the grep_defaults template.
56
 */
57
int grep_config(const char *var, const char *value,
58
		const struct config_context *ctx, void *cb)
59
{
60
	struct grep_opt *opt = cb;
61
	const char *slot;
62

63
	if (userdiff_config(var, value) < 0)
64
		return -1;
65

66
	if (!strcmp(var, "grep.extendedregexp")) {
67
		opt->extended_regexp_option = git_config_bool(var, value);
68
		return 0;
69
	}
70

71
	if (!strcmp(var, "grep.patterntype")) {
72
		opt->pattern_type_option = parse_pattern_type_arg(var, value);
73
		return 0;
74
	}
75

76
	if (!strcmp(var, "grep.linenumber")) {
77
		opt->linenum = git_config_bool(var, value);
78
		return 0;
79
	}
80
	if (!strcmp(var, "grep.column")) {
81
		opt->columnnum = git_config_bool(var, value);
82
		return 0;
83
	}
84

85
	if (!strcmp(var, "grep.fullname")) {
86
		opt->relative = !git_config_bool(var, value);
87
		return 0;
88
	}
89

90
	if (!strcmp(var, "color.grep"))
91
		opt->color = git_config_colorbool(var, value);
92
	if (!strcmp(var, "color.grep.match")) {
93
		if (grep_config("color.grep.matchcontext", value, ctx, cb) < 0)
94
			return -1;
95
		if (grep_config("color.grep.matchselected", value, ctx, cb) < 0)
96
			return -1;
97
	} else if (skip_prefix(var, "color.grep.", &slot)) {
98
		int i = LOOKUP_CONFIG(color_grep_slots, slot);
99
		char *color;
100

101
		if (i < 0)
102
			return -1;
103
		color = opt->colors[i];
104
		if (!value)
105
			return config_error_nonbool(var);
106
		return color_parse(value, color);
107
	}
108
	return 0;
109
}
110

111
void grep_init(struct grep_opt *opt, struct repository *repo)
112
{
113
	struct grep_opt blank = GREP_OPT_INIT;
114
	memcpy(opt, &blank, sizeof(*opt));
115

116
	opt->repo = repo;
117
	opt->pattern_tail = &opt->pattern_list;
118
	opt->header_tail = &opt->header_list;
119
}
120

121
static struct grep_pat *create_grep_pat(const char *pat, size_t patlen,
122
					const char *origin, int no,
123
					enum grep_pat_token t,
124
					enum grep_header_field field)
125
{
126
	struct grep_pat *p = xcalloc(1, sizeof(*p));
127
	p->pattern = xmemdupz(pat, patlen);
128
	p->patternlen = patlen;
129
	p->origin = origin;
130
	p->no = no;
131
	p->token = t;
132
	p->field = field;
133
	return p;
134
}
135

136
static void do_append_grep_pat(struct grep_pat ***tail, struct grep_pat *p)
137
{
138
	**tail = p;
139
	*tail = &p->next;
140
	p->next = NULL;
141

142
	switch (p->token) {
143
	case GREP_PATTERN: /* atom */
144
	case GREP_PATTERN_HEAD:
145
	case GREP_PATTERN_BODY:
146
		for (;;) {
147
			struct grep_pat *new_pat;
148
			size_t len = 0;
149
			char *cp = p->pattern + p->patternlen, *nl = NULL;
150
			while (++len <= p->patternlen) {
151
				if (*(--cp) == '\n') {
152
					nl = cp;
153
					break;
154
				}
155
			}
156
			if (!nl)
157
				break;
158
			new_pat = create_grep_pat(nl + 1, len - 1, p->origin,
159
						  p->no, p->token, p->field);
160
			new_pat->next = p->next;
161
			if (!p->next)
162
				*tail = &new_pat->next;
163
			p->next = new_pat;
164
			*nl = '\0';
165
			p->patternlen -= len;
166
		}
167
		break;
168
	default:
169
		break;
170
	}
171
}
172

173
void append_header_grep_pattern(struct grep_opt *opt,
174
				enum grep_header_field field, const char *pat)
175
{
176
	struct grep_pat *p = create_grep_pat(pat, strlen(pat), "header", 0,
177
					     GREP_PATTERN_HEAD, field);
178
	if (field == GREP_HEADER_REFLOG)
179
		opt->use_reflog_filter = 1;
180
	do_append_grep_pat(&opt->header_tail, p);
181
}
182

183
void append_grep_pattern(struct grep_opt *opt, const char *pat,
184
			 const char *origin, int no, enum grep_pat_token t)
185
{
186
	append_grep_pat(opt, pat, strlen(pat), origin, no, t);
187
}
188

189
void append_grep_pat(struct grep_opt *opt, const char *pat, size_t patlen,
190
		     const char *origin, int no, enum grep_pat_token t)
191
{
192
	struct grep_pat *p = create_grep_pat(pat, patlen, origin, no, t, 0);
193
	do_append_grep_pat(&opt->pattern_tail, p);
194
}
195

196
struct grep_opt *grep_opt_dup(const struct grep_opt *opt)
197
{
198
	struct grep_pat *pat;
199
	struct grep_opt *ret = xmalloc(sizeof(struct grep_opt));
200
	*ret = *opt;
201

202
	ret->pattern_list = NULL;
203
	ret->pattern_tail = &ret->pattern_list;
204

205
	for(pat = opt->pattern_list; pat != NULL; pat = pat->next)
206
	{
207
		if(pat->token == GREP_PATTERN_HEAD)
208
			append_header_grep_pattern(ret, pat->field,
209
						   pat->pattern);
210
		else
211
			append_grep_pat(ret, pat->pattern, pat->patternlen,
212
					pat->origin, pat->no, pat->token);
213
	}
214

215
	return ret;
216
}
217

218
static NORETURN void compile_regexp_failed(const struct grep_pat *p,
219
		const char *error)
220
{
221
	char where[1024];
222

223
	if (p->no)
224
		xsnprintf(where, sizeof(where), "In '%s' at %d, ", p->origin, p->no);
225
	else if (p->origin)
226
		xsnprintf(where, sizeof(where), "%s, ", p->origin);
227
	else
228
		where[0] = 0;
229

230
	die("%s'%s': %s", where, p->pattern, error);
231
}
232

233
static int is_fixed(const char *s, size_t len)
234
{
235
	size_t i;
236

237
	for (i = 0; i < len; i++) {
238
		if (is_regex_special(s[i]))
239
			return 0;
240
	}
241

242
	return 1;
243
}
244

245
#ifdef USE_LIBPCRE2
246
#define GREP_PCRE2_DEBUG_MALLOC 0
247

248
static void *pcre2_malloc(PCRE2_SIZE size, MAYBE_UNUSED void *memory_data)
249
{
250
	void *pointer = malloc(size);
251
#if GREP_PCRE2_DEBUG_MALLOC
252
	static int count = 1;
253
	fprintf(stderr, "PCRE2:%p -> #%02d: alloc(%lu)\n", pointer, count++, size);
254
#endif
255
	return pointer;
256
}
257

258
static void pcre2_free(void *pointer, MAYBE_UNUSED void *memory_data)
259
{
260
#if GREP_PCRE2_DEBUG_MALLOC
261
	static int count = 1;
262
	if (pointer)
263
		fprintf(stderr, "PCRE2:%p -> #%02d: free()\n", pointer, count++);
264
#endif
265
	free(pointer);
266
}
267

268
static int pcre2_jit_functional(void)
269
{
270
	static int jit_working = -1;
271
	pcre2_code *code;
272
	size_t off;
273
	int err;
274

275
	if (jit_working != -1)
276
		return jit_working;
277

278
	/*
279
	 * Try to JIT compile a simple pattern to probe if the JIT is
280
	 * working in general. It might fail for systems where creating
281
	 * memory mappings for runtime code generation is restricted.
282
	 */
283
	code = pcre2_compile((PCRE2_SPTR)".", 1, 0, &err, &off, NULL);
284
	if (!code)
285
		return 0;
286

287
	jit_working = pcre2_jit_compile(code, PCRE2_JIT_COMPLETE) == 0;
288
	pcre2_code_free(code);
289

290
	return jit_working;
291
}
292

293
static void compile_pcre2_pattern(struct grep_pat *p, const struct grep_opt *opt)
294
{
295
	int error;
296
	PCRE2_UCHAR errbuf[256];
297
	PCRE2_SIZE erroffset;
298
	int options = PCRE2_MULTILINE;
299
	int jitret;
300
	int patinforet;
301
	size_t jitsizearg;
302
	int literal = !opt->ignore_case && (p->fixed || p->is_fixed);
303

304
	/*
305
	 * Call pcre2_general_context_create() before calling any
306
	 * other pcre2_*(). It sets up our malloc()/free() functions
307
	 * with which everything else is allocated.
308
	 */
309
	p->pcre2_general_context = pcre2_general_context_create(
310
		pcre2_malloc, pcre2_free, NULL);
311
	if (!p->pcre2_general_context)
312
		die("Couldn't allocate PCRE2 general context");
313

314
	if (opt->ignore_case) {
315
		if (!opt->ignore_locale && has_non_ascii(p->pattern)) {
316
			p->pcre2_tables = pcre2_maketables(p->pcre2_general_context);
317
			p->pcre2_compile_context = pcre2_compile_context_create(p->pcre2_general_context);
318
			pcre2_set_character_tables(p->pcre2_compile_context,
319
							p->pcre2_tables);
320
		}
321
		options |= PCRE2_CASELESS;
322
	}
323
	if (!opt->ignore_locale && is_utf8_locale() && !literal)
324
		options |= (PCRE2_UTF | PCRE2_UCP | PCRE2_MATCH_INVALID_UTF);
325

326
#ifndef GIT_PCRE2_VERSION_10_35_OR_HIGHER
327
	/*
328
	 * Work around a JIT bug related to invalid Unicode character handling
329
	 * fixed in 10.35:
330
	 * https://github.com/PCRE2Project/pcre2/commit/c21bd977547d
331
	 */
332
	options &= ~PCRE2_UCP;
333
#endif
334

335
#ifndef GIT_PCRE2_VERSION_10_36_OR_HIGHER
336
	/* Work around https://bugs.exim.org/show_bug.cgi?id=2642 fixed in 10.36 */
337
	if (PCRE2_MATCH_INVALID_UTF && options & (PCRE2_UTF | PCRE2_CASELESS))
338
		options |= PCRE2_NO_START_OPTIMIZE;
339
#endif
340

341
	p->pcre2_pattern = pcre2_compile((PCRE2_SPTR)p->pattern,
342
					 p->patternlen, options, &error, &erroffset,
343
					 p->pcre2_compile_context);
344

345
	if (p->pcre2_pattern) {
346
		p->pcre2_match_data = pcre2_match_data_create_from_pattern(p->pcre2_pattern, p->pcre2_general_context);
347
		if (!p->pcre2_match_data)
348
			die("Couldn't allocate PCRE2 match data");
349
	} else {
350
		pcre2_get_error_message(error, errbuf, sizeof(errbuf));
351
		compile_regexp_failed(p, (const char *)&errbuf);
352
	}
353

354
	pcre2_config(PCRE2_CONFIG_JIT, &p->pcre2_jit_on);
355
	if (p->pcre2_jit_on) {
356
		jitret = pcre2_jit_compile(p->pcre2_pattern, PCRE2_JIT_COMPLETE);
357
		if (jitret == PCRE2_ERROR_NOMEMORY && !pcre2_jit_functional()) {
358
			/*
359
			 * Even though pcre2_config(PCRE2_CONFIG_JIT, ...)
360
			 * indicated JIT support, the library might still
361
			 * fail to generate JIT code for various reasons,
362
			 * e.g. when SELinux's 'deny_execmem' or PaX's
363
			 * MPROTECT prevent creating W|X memory mappings.
364
			 *
365
			 * Instead of faling hard, fall back to interpreter
366
			 * mode, just as if the pattern was prefixed with
367
			 * '(*NO_JIT)'.
368
			 */
369
			p->pcre2_jit_on = 0;
370
			return;
371
		} else if (jitret) {
372
			int need_clip = p->patternlen > 64;
373
			int clip_len = need_clip ? 64 : p->patternlen;
374
			die("Couldn't JIT the PCRE2 pattern '%.*s'%s, got '%d'%s",
375
			    clip_len, p->pattern, need_clip ? "..." : "", jitret,
376
			    pcre2_jit_functional()
377
			    ? "\nPerhaps prefix (*NO_JIT) to your pattern?"
378
			    : "");
379
		}
380

381
		/*
382
		 * The pcre2_config(PCRE2_CONFIG_JIT, ...) call just
383
		 * tells us whether the library itself supports JIT,
384
		 * but to see whether we're going to be actually using
385
		 * JIT we need to extract PCRE2_INFO_JITSIZE from the
386
		 * pattern *after* we do pcre2_jit_compile() above.
387
		 *
388
		 * This is because if the pattern contains the
389
		 * (*NO_JIT) verb (see pcre2syntax(3))
390
		 * pcre2_jit_compile() will exit early with 0. If we
391
		 * then proceed to call pcre2_jit_match() further down
392
		 * the line instead of pcre2_match() we'll either
393
		 * segfault (pre PCRE 10.31) or run into a fatal error
394
		 * (post PCRE2 10.31)
395
		 */
396
		patinforet = pcre2_pattern_info(p->pcre2_pattern, PCRE2_INFO_JITSIZE, &jitsizearg);
397
		if (patinforet)
398
			BUG("pcre2_pattern_info() failed: %d", patinforet);
399
		if (jitsizearg == 0) {
400
			p->pcre2_jit_on = 0;
401
			return;
402
		}
403
	}
404
}
405

406
static int pcre2match(struct grep_pat *p, const char *line, const char *eol,
407
		regmatch_t *match, int eflags)
408
{
409
	int ret, flags = 0;
410
	PCRE2_SIZE *ovector;
411
	PCRE2_UCHAR errbuf[256];
412

413
	if (eflags & REG_NOTBOL)
414
		flags |= PCRE2_NOTBOL;
415

416
	if (p->pcre2_jit_on)
417
		ret = pcre2_jit_match(p->pcre2_pattern, (unsigned char *)line,
418
				      eol - line, 0, flags, p->pcre2_match_data,
419
				      NULL);
420
	else
421
		ret = pcre2_match(p->pcre2_pattern, (unsigned char *)line,
422
				  eol - line, 0, flags, p->pcre2_match_data,
423
				  NULL);
424

425
	if (ret < 0 && ret != PCRE2_ERROR_NOMATCH) {
426
		pcre2_get_error_message(ret, errbuf, sizeof(errbuf));
427
		die("%s failed with error code %d: %s",
428
		    (p->pcre2_jit_on ? "pcre2_jit_match" : "pcre2_match"), ret,
429
		    errbuf);
430
	}
431
	if (ret > 0) {
432
		ovector = pcre2_get_ovector_pointer(p->pcre2_match_data);
433
		ret = 0;
434
		match->rm_so = (int)ovector[0];
435
		match->rm_eo = (int)ovector[1];
436
	}
437

438
	return ret;
439
}
440

441
static void free_pcre2_pattern(struct grep_pat *p)
442
{
443
	pcre2_compile_context_free(p->pcre2_compile_context);
444
	pcre2_code_free(p->pcre2_pattern);
445
	pcre2_match_data_free(p->pcre2_match_data);
446
#ifdef GIT_PCRE2_VERSION_10_34_OR_HIGHER
447
	pcre2_maketables_free(p->pcre2_general_context, p->pcre2_tables);
448
#else
449
	free((void *)p->pcre2_tables);
450
#endif
451
	pcre2_general_context_free(p->pcre2_general_context);
452
}
453
#else /* !USE_LIBPCRE2 */
454
static void compile_pcre2_pattern(struct grep_pat *p UNUSED,
455
				  const struct grep_opt *opt UNUSED)
456
{
457
	die("cannot use Perl-compatible regexes when not compiled with USE_LIBPCRE");
458
}
459

460
static int pcre2match(struct grep_pat *p UNUSED, const char *line UNUSED,
461
		      const char *eol UNUSED, regmatch_t *match UNUSED,
462
		      int eflags UNUSED)
463
{
464
	return 1;
465
}
466

467
static void free_pcre2_pattern(struct grep_pat *p UNUSED)
468
{
469
}
470

471
static void compile_fixed_regexp(struct grep_pat *p, struct grep_opt *opt)
472
{
473
	struct strbuf sb = STRBUF_INIT;
474
	int err;
475
	int regflags = 0;
476

477
	basic_regex_quote_buf(&sb, p->pattern);
478
	if (opt->ignore_case)
479
		regflags |= REG_ICASE;
480
	err = regcomp(&p->regexp, sb.buf, regflags);
481
	strbuf_release(&sb);
482
	if (err) {
483
		char errbuf[1024];
484
		regerror(err, &p->regexp, errbuf, sizeof(errbuf));
485
		compile_regexp_failed(p, errbuf);
486
	}
487
}
488
#endif /* !USE_LIBPCRE2 */
489

490
static void compile_regexp(struct grep_pat *p, struct grep_opt *opt)
491
{
492
	int err;
493
	int regflags = REG_NEWLINE;
494

495
	if (opt->pattern_type_option == GREP_PATTERN_TYPE_UNSPECIFIED)
496
		opt->pattern_type_option = (opt->extended_regexp_option
497
					    ? GREP_PATTERN_TYPE_ERE
498
					    : GREP_PATTERN_TYPE_BRE);
499

500
	p->word_regexp = opt->word_regexp;
501
	p->ignore_case = opt->ignore_case;
502
	p->fixed = opt->pattern_type_option == GREP_PATTERN_TYPE_FIXED;
503

504
	if (opt->pattern_type_option != GREP_PATTERN_TYPE_PCRE &&
505
	    memchr(p->pattern, 0, p->patternlen))
506
		die(_("given pattern contains NULL byte (via -f <file>). This is only supported with -P under PCRE v2"));
507

508
	p->is_fixed = is_fixed(p->pattern, p->patternlen);
509
#ifdef USE_LIBPCRE2
510
       if (!p->fixed && !p->is_fixed) {
511
	       const char *no_jit = "(*NO_JIT)";
512
	       const int no_jit_len = strlen(no_jit);
513
	       if (starts_with(p->pattern, no_jit) &&
514
		   is_fixed(p->pattern + no_jit_len,
515
			    p->patternlen - no_jit_len))
516
		       p->is_fixed = 1;
517
       }
518
#endif
519
	if (p->fixed || p->is_fixed) {
520
#ifdef USE_LIBPCRE2
521
		if (p->is_fixed) {
522
			compile_pcre2_pattern(p, opt);
523
		} else {
524
			/*
525
			 * E.g. t7811-grep-open.sh relies on the
526
			 * pattern being restored.
527
			 */
528
			char *old_pattern = p->pattern;
529
			size_t old_patternlen = p->patternlen;
530
			struct strbuf sb = STRBUF_INIT;
531

532
			/*
533
			 * There is the PCRE2_LITERAL flag, but it's
534
			 * only in PCRE v2 10.30 and later. Needing to
535
			 * ifdef our way around that and dealing with
536
			 * it + PCRE2_MULTILINE being an error is more
537
			 * complex than just quoting this ourselves.
538
			*/
539
			strbuf_add(&sb, "\\Q", 2);
540
			strbuf_add(&sb, p->pattern, p->patternlen);
541
			strbuf_add(&sb, "\\E", 2);
542

543
			p->pattern = sb.buf;
544
			p->patternlen = sb.len;
545
			compile_pcre2_pattern(p, opt);
546
			p->pattern = old_pattern;
547
			p->patternlen = old_patternlen;
548
			strbuf_release(&sb);
549
		}
550
#else /* !USE_LIBPCRE2 */
551
		compile_fixed_regexp(p, opt);
552
#endif /* !USE_LIBPCRE2 */
553
		return;
554
	}
555

556
	if (opt->pattern_type_option == GREP_PATTERN_TYPE_PCRE) {
557
		compile_pcre2_pattern(p, opt);
558
		return;
559
	}
560

561
	if (p->ignore_case)
562
		regflags |= REG_ICASE;
563
	if (opt->pattern_type_option == GREP_PATTERN_TYPE_ERE)
564
		regflags |= REG_EXTENDED;
565
	err = regcomp(&p->regexp, p->pattern, regflags);
566
	if (err) {
567
		char errbuf[1024];
568
		regerror(err, &p->regexp, errbuf, 1024);
569
		compile_regexp_failed(p, errbuf);
570
	}
571
}
572

573
static struct grep_expr *grep_not_expr(struct grep_expr *expr)
574
{
575
	struct grep_expr *z = xcalloc(1, sizeof(*z));
576
	z->node = GREP_NODE_NOT;
577
	z->u.unary = expr;
578
	return z;
579
}
580

581
static struct grep_expr *grep_binexp(enum grep_expr_node kind,
582
				     struct grep_expr *left,
583
				     struct grep_expr *right)
584
{
585
	struct grep_expr *z = xcalloc(1, sizeof(*z));
586
	z->node = kind;
587
	z->u.binary.left = left;
588
	z->u.binary.right = right;
589
	return z;
590
}
591

592
static struct grep_expr *grep_or_expr(struct grep_expr *left, struct grep_expr *right)
593
{
594
	return grep_binexp(GREP_NODE_OR, left, right);
595
}
596

597
static struct grep_expr *grep_and_expr(struct grep_expr *left, struct grep_expr *right)
598
{
599
	return grep_binexp(GREP_NODE_AND, left, right);
600
}
601

602
static struct grep_expr *compile_pattern_or(struct grep_pat **);
603
static struct grep_expr *compile_pattern_atom(struct grep_pat **list)
604
{
605
	struct grep_pat *p;
606
	struct grep_expr *x;
607

608
	p = *list;
609
	if (!p)
610
		return NULL;
611
	switch (p->token) {
612
	case GREP_PATTERN: /* atom */
613
	case GREP_PATTERN_HEAD:
614
	case GREP_PATTERN_BODY:
615
		CALLOC_ARRAY(x, 1);
616
		x->node = GREP_NODE_ATOM;
617
		x->u.atom = p;
618
		*list = p->next;
619
		return x;
620
	case GREP_OPEN_PAREN:
621
		*list = p->next;
622
		x = compile_pattern_or(list);
623
		if (!*list || (*list)->token != GREP_CLOSE_PAREN)
624
			die("unmatched ( for expression group");
625
		*list = (*list)->next;
626
		return x;
627
	default:
628
		return NULL;
629
	}
630
}
631

632
static struct grep_expr *compile_pattern_not(struct grep_pat **list)
633
{
634
	struct grep_pat *p;
635
	struct grep_expr *x;
636

637
	p = *list;
638
	if (!p)
639
		return NULL;
640
	switch (p->token) {
641
	case GREP_NOT:
642
		if (!p->next)
643
			die("--not not followed by pattern expression");
644
		*list = p->next;
645
		x = compile_pattern_not(list);
646
		if (!x)
647
			die("--not followed by non pattern expression");
648
		return grep_not_expr(x);
649
	default:
650
		return compile_pattern_atom(list);
651
	}
652
}
653

654
static struct grep_expr *compile_pattern_and(struct grep_pat **list)
655
{
656
	struct grep_pat *p;
657
	struct grep_expr *x, *y;
658

659
	x = compile_pattern_not(list);
660
	p = *list;
661
	if (p && p->token == GREP_AND) {
662
		if (!x)
663
			die("--and not preceded by pattern expression");
664
		if (!p->next)
665
			die("--and not followed by pattern expression");
666
		*list = p->next;
667
		y = compile_pattern_and(list);
668
		if (!y)
669
			die("--and not followed by pattern expression");
670
		return grep_and_expr(x, y);
671
	}
672
	return x;
673
}
674

675
static struct grep_expr *compile_pattern_or(struct grep_pat **list)
676
{
677
	struct grep_pat *p;
678
	struct grep_expr *x, *y;
679

680
	x = compile_pattern_and(list);
681
	p = *list;
682
	if (x && p && p->token != GREP_CLOSE_PAREN) {
683
		y = compile_pattern_or(list);
684
		if (!y)
685
			die("not a pattern expression %s", p->pattern);
686
		return grep_or_expr(x, y);
687
	}
688
	return x;
689
}
690

691
static struct grep_expr *compile_pattern_expr(struct grep_pat **list)
692
{
693
	return compile_pattern_or(list);
694
}
695

696
static struct grep_expr *grep_true_expr(void)
697
{
698
	struct grep_expr *z = xcalloc(1, sizeof(*z));
699
	z->node = GREP_NODE_TRUE;
700
	return z;
701
}
702

703
static struct grep_expr *prep_header_patterns(struct grep_opt *opt)
704
{
705
	struct grep_pat *p;
706
	struct grep_expr *header_expr;
707
	struct grep_expr *(header_group[GREP_HEADER_FIELD_MAX]);
708
	enum grep_header_field fld;
709

710
	if (!opt->header_list)
711
		return NULL;
712

713
	for (p = opt->header_list; p; p = p->next) {
714
		if (p->token != GREP_PATTERN_HEAD)
715
			BUG("a non-header pattern in grep header list.");
716
		if (p->field < GREP_HEADER_FIELD_MIN ||
717
		    GREP_HEADER_FIELD_MAX <= p->field)
718
			BUG("unknown header field %d", p->field);
719
		compile_regexp(p, opt);
720
	}
721

722
	for (fld = 0; fld < GREP_HEADER_FIELD_MAX; fld++)
723
		header_group[fld] = NULL;
724

725
	for (p = opt->header_list; p; p = p->next) {
726
		struct grep_expr *h;
727
		struct grep_pat *pp = p;
728

729
		h = compile_pattern_atom(&pp);
730
		if (!h || pp != p->next)
731
			BUG("malformed header expr");
732
		if (!header_group[p->field]) {
733
			header_group[p->field] = h;
734
			continue;
735
		}
736
		header_group[p->field] = grep_or_expr(h, header_group[p->field]);
737
	}
738

739
	header_expr = NULL;
740

741
	for (fld = 0; fld < GREP_HEADER_FIELD_MAX; fld++) {
742
		if (!header_group[fld])
743
			continue;
744
		if (!header_expr)
745
			header_expr = grep_true_expr();
746
		header_expr = grep_or_expr(header_group[fld], header_expr);
747
	}
748
	return header_expr;
749
}
750

751
static struct grep_expr *grep_splice_or(struct grep_expr *x, struct grep_expr *y)
752
{
753
	struct grep_expr *z = x;
754

755
	while (x) {
756
		assert(x->node == GREP_NODE_OR);
757
		if (x->u.binary.right &&
758
		    x->u.binary.right->node == GREP_NODE_TRUE) {
759
			x->u.binary.right = y;
760
			break;
761
		}
762
		x = x->u.binary.right;
763
	}
764
	return z;
765
}
766

767
void compile_grep_patterns(struct grep_opt *opt)
768
{
769
	struct grep_pat *p;
770
	struct grep_expr *header_expr = prep_header_patterns(opt);
771
	int extended = 0;
772

773
	for (p = opt->pattern_list; p; p = p->next) {
774
		switch (p->token) {
775
		case GREP_PATTERN: /* atom */
776
		case GREP_PATTERN_HEAD:
777
		case GREP_PATTERN_BODY:
778
			compile_regexp(p, opt);
779
			break;
780
		default:
781
			extended = 1;
782
			break;
783
		}
784
	}
785

786
	if (opt->all_match || opt->no_body_match || header_expr)
787
		extended = 1;
788
	else if (!extended)
789
		return;
790

791
	p = opt->pattern_list;
792
	if (p)
793
		opt->pattern_expression = compile_pattern_expr(&p);
794
	if (p)
795
		die("incomplete pattern expression group: %s", p->pattern);
796

797
	if (opt->no_body_match && opt->pattern_expression)
798
		opt->pattern_expression = grep_not_expr(opt->pattern_expression);
799

800
	if (!header_expr)
801
		return;
802

803
	if (!opt->pattern_expression)
804
		opt->pattern_expression = header_expr;
805
	else if (opt->all_match)
806
		opt->pattern_expression = grep_splice_or(header_expr,
807
							 opt->pattern_expression);
808
	else
809
		opt->pattern_expression = grep_or_expr(opt->pattern_expression,
810
						       header_expr);
811
	opt->all_match = 1;
812
}
813

814
static void free_pattern_expr(struct grep_expr *x)
815
{
816
	switch (x->node) {
817
	case GREP_NODE_TRUE:
818
	case GREP_NODE_ATOM:
819
		break;
820
	case GREP_NODE_NOT:
821
		free_pattern_expr(x->u.unary);
822
		break;
823
	case GREP_NODE_AND:
824
	case GREP_NODE_OR:
825
		free_pattern_expr(x->u.binary.left);
826
		free_pattern_expr(x->u.binary.right);
827
		break;
828
	}
829
	free(x);
830
}
831

832
static void free_grep_pat(struct grep_pat *pattern)
833
{
834
	struct grep_pat *p, *n;
835

836
	for (p = pattern; p; p = n) {
837
		n = p->next;
838
		switch (p->token) {
839
		case GREP_PATTERN: /* atom */
840
		case GREP_PATTERN_HEAD:
841
		case GREP_PATTERN_BODY:
842
			if (p->pcre2_pattern)
843
				free_pcre2_pattern(p);
844
			else
845
				regfree(&p->regexp);
846
			free(p->pattern);
847
			break;
848
		default:
849
			break;
850
		}
851
		free(p);
852
	}
853
}
854

855
void free_grep_patterns(struct grep_opt *opt)
856
{
857
	free_grep_pat(opt->pattern_list);
858
	free_grep_pat(opt->header_list);
859

860
	if (opt->pattern_expression)
861
		free_pattern_expr(opt->pattern_expression);
862
}
863

864
static const char *end_of_line(const char *cp, unsigned long *left)
865
{
866
	unsigned long l = *left;
867
	while (l && *cp != '\n') {
868
		l--;
869
		cp++;
870
	}
871
	*left = l;
872
	return cp;
873
}
874

875
static int word_char(char ch)
876
{
877
	return isalnum(ch) || ch == '_';
878
}
879

880
static void output_color(struct grep_opt *opt, const void *data, size_t size,
881
			 const char *color)
882
{
883
	if (want_color(opt->color) && color && color[0]) {
884
		opt->output(opt, color, strlen(color));
885
		opt->output(opt, data, size);
886
		opt->output(opt, GIT_COLOR_RESET, strlen(GIT_COLOR_RESET));
887
	} else
888
		opt->output(opt, data, size);
889
}
890

891
static void output_sep(struct grep_opt *opt, char sign)
892
{
893
	if (opt->null_following_name)
894
		opt->output(opt, "\0", 1);
895
	else
896
		output_color(opt, &sign, 1, opt->colors[GREP_COLOR_SEP]);
897
}
898

899
static void show_name(struct grep_opt *opt, const char *name)
900
{
901
	output_color(opt, name, strlen(name), opt->colors[GREP_COLOR_FILENAME]);
902
	opt->output(opt, opt->null_following_name ? "\0" : "\n", 1);
903
}
904

905
static int patmatch(struct grep_pat *p,
906
		    const char *line, const char *eol,
907
		    regmatch_t *match, int eflags)
908
{
909
	int hit;
910

911
	if (p->pcre2_pattern)
912
		hit = !pcre2match(p, line, eol, match, eflags);
913
	else
914
		hit = !regexec_buf(&p->regexp, line, eol - line, 1, match,
915
				   eflags);
916

917
	return hit;
918
}
919

920
static void strip_timestamp(const char *bol, const char **eol_p)
921
{
922
	const char *eol = *eol_p;
923

924
	while (bol < --eol) {
925
		if (*eol != '>')
926
			continue;
927
		*eol_p = ++eol;
928
		break;
929
	}
930
}
931

932
static struct {
933
	const char *field;
934
	size_t len;
935
} header_field[] = {
936
	{ "author ", 7 },
937
	{ "committer ", 10 },
938
	{ "reflog ", 7 },
939
};
940

941
static int headerless_match_one_pattern(struct grep_pat *p,
942
					const char *bol, const char *eol,
943
					enum grep_context ctx,
944
					regmatch_t *pmatch, int eflags)
945
{
946
	int hit = 0;
947
	const char *start = bol;
948

949
	if ((p->token != GREP_PATTERN) &&
950
	    ((p->token == GREP_PATTERN_HEAD) != (ctx == GREP_CONTEXT_HEAD)))
951
		return 0;
952

953
 again:
954
	hit = patmatch(p, bol, eol, pmatch, eflags);
955

956
	if (hit && p->word_regexp) {
957
		if ((pmatch[0].rm_so < 0) ||
958
		    (eol - bol) < pmatch[0].rm_so ||
959
		    (pmatch[0].rm_eo < 0) ||
960
		    (eol - bol) < pmatch[0].rm_eo)
961
			die("regexp returned nonsense");
962

963
		/* Match beginning must be either beginning of the
964
		 * line, or at word boundary (i.e. the last char must
965
		 * not be a word char).  Similarly, match end must be
966
		 * either end of the line, or at word boundary
967
		 * (i.e. the next char must not be a word char).
968
		 */
969
		if ( ((pmatch[0].rm_so == 0) ||
970
		      !word_char(bol[pmatch[0].rm_so-1])) &&
971
		     ((pmatch[0].rm_eo == (eol-bol)) ||
972
		      !word_char(bol[pmatch[0].rm_eo])) )
973
			;
974
		else
975
			hit = 0;
976

977
		/* Words consist of at least one character. */
978
		if (pmatch->rm_so == pmatch->rm_eo)
979
			hit = 0;
980

981
		if (!hit && pmatch[0].rm_so + bol + 1 < eol) {
982
			/* There could be more than one match on the
983
			 * line, and the first match might not be
984
			 * strict word match.  But later ones could be!
985
			 * Forward to the next possible start, i.e. the
986
			 * next position following a non-word char.
987
			 */
988
			bol = pmatch[0].rm_so + bol + 1;
989
			while (word_char(bol[-1]) && bol < eol)
990
				bol++;
991
			eflags |= REG_NOTBOL;
992
			if (bol < eol)
993
				goto again;
994
		}
995
	}
996
	if (hit) {
997
		pmatch[0].rm_so += bol - start;
998
		pmatch[0].rm_eo += bol - start;
999
	}
1000
	return hit;
1001
}
1002

1003
static int match_one_pattern(struct grep_pat *p,
1004
			     const char *bol, const char *eol,
1005
			     enum grep_context ctx, regmatch_t *pmatch,
1006
			     int eflags)
1007
{
1008
	const char *field;
1009
	size_t len;
1010

1011
	if (p->token == GREP_PATTERN_HEAD) {
1012
		assert(p->field < ARRAY_SIZE(header_field));
1013
		field = header_field[p->field].field;
1014
		len = header_field[p->field].len;
1015
		if (strncmp(bol, field, len))
1016
			return 0;
1017
		bol += len;
1018

1019
		switch (p->field) {
1020
		case GREP_HEADER_AUTHOR:
1021
		case GREP_HEADER_COMMITTER:
1022
			strip_timestamp(bol, &eol);
1023
			break;
1024
		default:
1025
			break;
1026
		}
1027
	}
1028

1029
	return headerless_match_one_pattern(p, bol, eol, ctx, pmatch, eflags);
1030
}
1031

1032

1033
static int match_expr_eval(struct grep_opt *opt, struct grep_expr *x,
1034
			   const char *bol, const char *eol,
1035
			   enum grep_context ctx, ssize_t *col,
1036
			   ssize_t *icol, int collect_hits)
1037
{
1038
	int h = 0;
1039

1040
	switch (x->node) {
1041
	case GREP_NODE_TRUE:
1042
		h = 1;
1043
		break;
1044
	case GREP_NODE_ATOM:
1045
		{
1046
			regmatch_t tmp;
1047
			h = match_one_pattern(x->u.atom, bol, eol, ctx,
1048
					      &tmp, 0);
1049
			if (h && (*col < 0 || tmp.rm_so < *col))
1050
				*col = tmp.rm_so;
1051
		}
1052
		if (x->u.atom->token == GREP_PATTERN_BODY)
1053
			opt->body_hit |= h;
1054
		break;
1055
	case GREP_NODE_NOT:
1056
		/*
1057
		 * Upon visiting a GREP_NODE_NOT, col and icol become swapped.
1058
		 */
1059
		h = !match_expr_eval(opt, x->u.unary, bol, eol, ctx, icol, col,
1060
				     0);
1061
		break;
1062
	case GREP_NODE_AND:
1063
		h = match_expr_eval(opt, x->u.binary.left, bol, eol, ctx, col,
1064
				    icol, 0);
1065
		if (h || opt->columnnum) {
1066
			/*
1067
			 * Don't short-circuit AND when given --column, since a
1068
			 * NOT earlier in the tree may turn this into an OR. In
1069
			 * this case, see the below comment.
1070
			 */
1071
			h &= match_expr_eval(opt, x->u.binary.right, bol, eol,
1072
					     ctx, col, icol, 0);
1073
		}
1074
		break;
1075
	case GREP_NODE_OR:
1076
		if (!(collect_hits || opt->columnnum)) {
1077
			/*
1078
			 * Don't short-circuit OR when given --column (or
1079
			 * collecting hits) to ensure we don't skip a later
1080
			 * child that would produce an earlier match.
1081
			 */
1082
			return (match_expr_eval(opt, x->u.binary.left, bol, eol,
1083
						ctx, col, icol, 0) ||
1084
				match_expr_eval(opt, x->u.binary.right, bol,
1085
						eol, ctx, col, icol, 0));
1086
		}
1087
		h = match_expr_eval(opt, x->u.binary.left, bol, eol, ctx, col,
1088
				    icol, 0);
1089
		if (collect_hits)
1090
			x->u.binary.left->hit |= h;
1091
		h |= match_expr_eval(opt, x->u.binary.right, bol, eol, ctx, col,
1092
				     icol, collect_hits);
1093
		break;
1094
	default:
1095
		die("Unexpected node type (internal error) %d", x->node);
1096
	}
1097
	if (collect_hits)
1098
		x->hit |= h;
1099
	return h;
1100
}
1101

1102
static int match_expr(struct grep_opt *opt,
1103
		      const char *bol, const char *eol,
1104
		      enum grep_context ctx, ssize_t *col,
1105
		      ssize_t *icol, int collect_hits)
1106
{
1107
	struct grep_expr *x = opt->pattern_expression;
1108
	return match_expr_eval(opt, x, bol, eol, ctx, col, icol, collect_hits);
1109
}
1110

1111
static int match_line(struct grep_opt *opt,
1112
		      const char *bol, const char *eol,
1113
		      ssize_t *col, ssize_t *icol,
1114
		      enum grep_context ctx, int collect_hits)
1115
{
1116
	struct grep_pat *p;
1117
	int hit = 0;
1118

1119
	if (opt->pattern_expression)
1120
		return match_expr(opt, bol, eol, ctx, col, icol,
1121
				  collect_hits);
1122

1123
	/* we do not call with collect_hits without being extended */
1124
	for (p = opt->pattern_list; p; p = p->next) {
1125
		regmatch_t tmp;
1126
		if (match_one_pattern(p, bol, eol, ctx, &tmp, 0)) {
1127
			hit |= 1;
1128
			if (!opt->columnnum) {
1129
				/*
1130
				 * Without --column, any single match on a line
1131
				 * is enough to know that it needs to be
1132
				 * printed. With --column, scan _all_ patterns
1133
				 * to find the earliest.
1134
				 */
1135
				break;
1136
			}
1137
			if (*col < 0 || tmp.rm_so < *col)
1138
				*col = tmp.rm_so;
1139
		}
1140
	}
1141
	return hit;
1142
}
1143

1144
static int match_next_pattern(struct grep_pat *p,
1145
			      const char *bol, const char *eol,
1146
			      enum grep_context ctx,
1147
			      regmatch_t *pmatch, int eflags)
1148
{
1149
	regmatch_t match;
1150

1151
	if (!headerless_match_one_pattern(p, bol, eol, ctx, &match, eflags))
1152
		return 0;
1153
	if (match.rm_so < 0 || match.rm_eo < 0)
1154
		return 0;
1155
	if (pmatch->rm_so >= 0 && pmatch->rm_eo >= 0) {
1156
		if (match.rm_so > pmatch->rm_so)
1157
			return 1;
1158
		if (match.rm_so == pmatch->rm_so && match.rm_eo < pmatch->rm_eo)
1159
			return 1;
1160
	}
1161
	pmatch->rm_so = match.rm_so;
1162
	pmatch->rm_eo = match.rm_eo;
1163
	return 1;
1164
}
1165

1166
int grep_next_match(struct grep_opt *opt,
1167
		    const char *bol, const char *eol,
1168
		    enum grep_context ctx, regmatch_t *pmatch,
1169
		    enum grep_header_field field, int eflags)
1170
{
1171
	struct grep_pat *p;
1172
	int hit = 0;
1173

1174
	pmatch->rm_so = pmatch->rm_eo = -1;
1175
	if (bol < eol) {
1176
		for (p = ((ctx == GREP_CONTEXT_HEAD)
1177
			   ? opt->header_list : opt->pattern_list);
1178
			  p; p = p->next) {
1179
			switch (p->token) {
1180
			case GREP_PATTERN_HEAD:
1181
				if ((field != GREP_HEADER_FIELD_MAX) &&
1182
				    (p->field != field))
1183
					continue;
1184
				/* fall thru */
1185
			case GREP_PATTERN: /* atom */
1186
			case GREP_PATTERN_BODY:
1187
				hit |= match_next_pattern(p, bol, eol, ctx,
1188
							  pmatch, eflags);
1189
				break;
1190
			default:
1191
				break;
1192
			}
1193
		}
1194
	}
1195
	return hit;
1196
}
1197

1198
static void show_line_header(struct grep_opt *opt, const char *name,
1199
			     unsigned lno, ssize_t cno, char sign)
1200
{
1201
	if (opt->heading && opt->last_shown == 0) {
1202
		output_color(opt, name, strlen(name), opt->colors[GREP_COLOR_FILENAME]);
1203
		opt->output(opt, "\n", 1);
1204
	}
1205
	opt->last_shown = lno;
1206

1207
	if (!opt->heading && opt->pathname) {
1208
		output_color(opt, name, strlen(name), opt->colors[GREP_COLOR_FILENAME]);
1209
		output_sep(opt, sign);
1210
	}
1211
	if (opt->linenum) {
1212
		char buf[32];
1213
		xsnprintf(buf, sizeof(buf), "%d", lno);
1214
		output_color(opt, buf, strlen(buf), opt->colors[GREP_COLOR_LINENO]);
1215
		output_sep(opt, sign);
1216
	}
1217
	/*
1218
	 * Treat 'cno' as the 1-indexed offset from the start of a non-context
1219
	 * line to its first match. Otherwise, 'cno' is 0 indicating that we are
1220
	 * being called with a context line.
1221
	 */
1222
	if (opt->columnnum && cno) {
1223
		char buf[32];
1224
		xsnprintf(buf, sizeof(buf), "%"PRIuMAX, (uintmax_t)cno);
1225
		output_color(opt, buf, strlen(buf), opt->colors[GREP_COLOR_COLUMNNO]);
1226
		output_sep(opt, sign);
1227
	}
1228
}
1229

1230
static void show_line(struct grep_opt *opt,
1231
		      const char *bol, const char *eol,
1232
		      const char *name, unsigned lno, ssize_t cno, char sign)
1233
{
1234
	int rest = eol - bol;
1235
	const char *match_color = NULL;
1236
	const char *line_color = NULL;
1237

1238
	if (opt->file_break && opt->last_shown == 0) {
1239
		if (opt->show_hunk_mark)
1240
			opt->output(opt, "\n", 1);
1241
	} else if (opt->pre_context || opt->post_context || opt->funcbody) {
1242
		if (opt->last_shown == 0) {
1243
			if (opt->show_hunk_mark) {
1244
				output_color(opt, "--", 2, opt->colors[GREP_COLOR_SEP]);
1245
				opt->output(opt, "\n", 1);
1246
			}
1247
		} else if (lno > opt->last_shown + 1) {
1248
			output_color(opt, "--", 2, opt->colors[GREP_COLOR_SEP]);
1249
			opt->output(opt, "\n", 1);
1250
		}
1251
	}
1252
	if (!opt->only_matching) {
1253
		/*
1254
		 * In case the line we're being called with contains more than
1255
		 * one match, leave printing each header to the loop below.
1256
		 */
1257
		show_line_header(opt, name, lno, cno, sign);
1258
	}
1259
	if (opt->color || opt->only_matching) {
1260
		regmatch_t match;
1261
		enum grep_context ctx = GREP_CONTEXT_BODY;
1262
		int eflags = 0;
1263

1264
		if (opt->color) {
1265
			if (sign == ':')
1266
				match_color = opt->colors[GREP_COLOR_MATCH_SELECTED];
1267
			else
1268
				match_color = opt->colors[GREP_COLOR_MATCH_CONTEXT];
1269
			if (sign == ':')
1270
				line_color = opt->colors[GREP_COLOR_SELECTED];
1271
			else if (sign == '-')
1272
				line_color = opt->colors[GREP_COLOR_CONTEXT];
1273
			else if (sign == '=')
1274
				line_color = opt->colors[GREP_COLOR_FUNCTION];
1275
		}
1276
		while (grep_next_match(opt, bol, eol, ctx, &match,
1277
				       GREP_HEADER_FIELD_MAX, eflags)) {
1278
			if (match.rm_so == match.rm_eo)
1279
				break;
1280

1281
			if (opt->only_matching)
1282
				show_line_header(opt, name, lno, cno, sign);
1283
			else
1284
				output_color(opt, bol, match.rm_so, line_color);
1285
			output_color(opt, bol + match.rm_so,
1286
				     match.rm_eo - match.rm_so, match_color);
1287
			if (opt->only_matching)
1288
				opt->output(opt, "\n", 1);
1289
			bol += match.rm_eo;
1290
			cno += match.rm_eo;
1291
			rest -= match.rm_eo;
1292
			eflags = REG_NOTBOL;
1293
		}
1294
	}
1295
	if (!opt->only_matching) {
1296
		output_color(opt, bol, rest, line_color);
1297
		opt->output(opt, "\n", 1);
1298
	}
1299
}
1300

1301
int grep_use_locks;
1302

1303
/*
1304
 * This lock protects access to the gitattributes machinery, which is
1305
 * not thread-safe.
1306
 */
1307
pthread_mutex_t grep_attr_mutex;
1308

1309
static inline void grep_attr_lock(void)
1310
{
1311
	if (grep_use_locks)
1312
		pthread_mutex_lock(&grep_attr_mutex);
1313
}
1314

1315
static inline void grep_attr_unlock(void)
1316
{
1317
	if (grep_use_locks)
1318
		pthread_mutex_unlock(&grep_attr_mutex);
1319
}
1320

1321
static int match_funcname(struct grep_opt *opt, struct grep_source *gs,
1322
			  const char *bol, const char *eol)
1323
{
1324
	xdemitconf_t *xecfg = opt->priv;
1325
	if (xecfg && !xecfg->find_func) {
1326
		grep_source_load_driver(gs, opt->repo->index);
1327
		if (gs->driver->funcname.pattern) {
1328
			const struct userdiff_funcname *pe = &gs->driver->funcname;
1329
			xdiff_set_find_func(xecfg, pe->pattern, pe->cflags);
1330
		} else {
1331
			xecfg = opt->priv = NULL;
1332
		}
1333
	}
1334

1335
	if (xecfg) {
1336
		char buf[1];
1337
		return xecfg->find_func(bol, eol - bol, buf, 1,
1338
					xecfg->find_func_priv) >= 0;
1339
	}
1340

1341
	if (bol == eol)
1342
		return 0;
1343
	if (isalpha(*bol) || *bol == '_' || *bol == '$')
1344
		return 1;
1345
	return 0;
1346
}
1347

1348
static void show_funcname_line(struct grep_opt *opt, struct grep_source *gs,
1349
			       const char *bol, unsigned lno)
1350
{
1351
	while (bol > gs->buf) {
1352
		const char *eol = --bol;
1353

1354
		while (bol > gs->buf && bol[-1] != '\n')
1355
			bol--;
1356
		lno--;
1357

1358
		if (lno <= opt->last_shown)
1359
			break;
1360

1361
		if (match_funcname(opt, gs, bol, eol)) {
1362
			show_line(opt, bol, eol, gs->name, lno, 0, '=');
1363
			break;
1364
		}
1365
	}
1366
}
1367

1368
static int is_empty_line(const char *bol, const char *eol);
1369

1370
static void show_pre_context(struct grep_opt *opt, struct grep_source *gs,
1371
			     const char *bol, const char *end, unsigned lno)
1372
{
1373
	unsigned cur = lno, from = 1, funcname_lno = 0, orig_from;
1374
	int funcname_needed = !!opt->funcname, comment_needed = 0;
1375

1376
	if (opt->pre_context < lno)
1377
		from = lno - opt->pre_context;
1378
	if (from <= opt->last_shown)
1379
		from = opt->last_shown + 1;
1380
	orig_from = from;
1381
	if (opt->funcbody) {
1382
		if (match_funcname(opt, gs, bol, end))
1383
			comment_needed = 1;
1384
		else
1385
			funcname_needed = 1;
1386
		from = opt->last_shown + 1;
1387
	}
1388

1389
	/* Rewind. */
1390
	while (bol > gs->buf && cur > from) {
1391
		const char *next_bol = bol;
1392
		const char *eol = --bol;
1393

1394
		while (bol > gs->buf && bol[-1] != '\n')
1395
			bol--;
1396
		cur--;
1397
		if (comment_needed && (is_empty_line(bol, eol) ||
1398
				       match_funcname(opt, gs, bol, eol))) {
1399
			comment_needed = 0;
1400
			from = orig_from;
1401
			if (cur < from) {
1402
				cur++;
1403
				bol = next_bol;
1404
				break;
1405
			}
1406
		}
1407
		if (funcname_needed && match_funcname(opt, gs, bol, eol)) {
1408
			funcname_lno = cur;
1409
			funcname_needed = 0;
1410
			if (opt->funcbody)
1411
				comment_needed = 1;
1412
			else
1413
				from = orig_from;
1414
		}
1415
	}
1416

1417
	/* We need to look even further back to find a function signature. */
1418
	if (opt->funcname && funcname_needed)
1419
		show_funcname_line(opt, gs, bol, cur);
1420

1421
	/* Back forward. */
1422
	while (cur < lno) {
1423
		const char *eol = bol, sign = (cur == funcname_lno) ? '=' : '-';
1424

1425
		while (*eol != '\n')
1426
			eol++;
1427
		show_line(opt, bol, eol, gs->name, cur, 0, sign);
1428
		bol = eol + 1;
1429
		cur++;
1430
	}
1431
}
1432

1433
static int should_lookahead(struct grep_opt *opt)
1434
{
1435
	struct grep_pat *p;
1436

1437
	if (opt->pattern_expression)
1438
		return 0; /* punt for too complex stuff */
1439
	if (opt->invert)
1440
		return 0;
1441
	for (p = opt->pattern_list; p; p = p->next) {
1442
		if (p->token != GREP_PATTERN)
1443
			return 0; /* punt for "header only" and stuff */
1444
	}
1445
	return 1;
1446
}
1447

1448
static int look_ahead(struct grep_opt *opt,
1449
		      unsigned long *left_p,
1450
		      unsigned *lno_p,
1451
		      const char **bol_p)
1452
{
1453
	unsigned lno = *lno_p;
1454
	const char *bol = *bol_p;
1455
	struct grep_pat *p;
1456
	const char *sp, *last_bol;
1457
	regoff_t earliest = -1;
1458

1459
	for (p = opt->pattern_list; p; p = p->next) {
1460
		int hit;
1461
		regmatch_t m;
1462

1463
		hit = patmatch(p, bol, bol + *left_p, &m, 0);
1464
		if (!hit || m.rm_so < 0 || m.rm_eo < 0)
1465
			continue;
1466
		if (earliest < 0 || m.rm_so < earliest)
1467
			earliest = m.rm_so;
1468
	}
1469

1470
	if (earliest < 0) {
1471
		*bol_p = bol + *left_p;
1472
		*left_p = 0;
1473
		return 1;
1474
	}
1475
	for (sp = bol + earliest; bol < sp && sp[-1] != '\n'; sp--)
1476
		; /* find the beginning of the line */
1477
	last_bol = sp;
1478

1479
	for (sp = bol; sp < last_bol; sp++) {
1480
		if (*sp == '\n')
1481
			lno++;
1482
	}
1483
	*left_p -= last_bol - bol;
1484
	*bol_p = last_bol;
1485
	*lno_p = lno;
1486
	return 0;
1487
}
1488

1489
static int fill_textconv_grep(struct repository *r,
1490
			      struct userdiff_driver *driver,
1491
			      struct grep_source *gs)
1492
{
1493
	struct diff_filespec *df;
1494
	char *buf;
1495
	size_t size;
1496

1497
	if (!driver || !driver->textconv)
1498
		return grep_source_load(gs);
1499

1500
	/*
1501
	 * The textconv interface is intimately tied to diff_filespecs, so we
1502
	 * have to pretend to be one. If we could unify the grep_source
1503
	 * and diff_filespec structs, this mess could just go away.
1504
	 */
1505
	df = alloc_filespec(gs->path);
1506
	switch (gs->type) {
1507
	case GREP_SOURCE_OID:
1508
		fill_filespec(df, gs->identifier, 1, 0100644);
1509
		break;
1510
	case GREP_SOURCE_FILE:
1511
		fill_filespec(df, null_oid(), 0, 0100644);
1512
		break;
1513
	default:
1514
		BUG("attempt to textconv something without a path?");
1515
	}
1516

1517
	/*
1518
	 * fill_textconv is not remotely thread-safe; it modifies the global
1519
	 * diff tempfile structure, writes to the_repo's odb and might
1520
	 * internally call thread-unsafe functions such as the
1521
	 * prepare_packed_git() lazy-initializator. Because of the last two, we
1522
	 * must ensure mutual exclusion between this call and the object reading
1523
	 * API, thus we use obj_read_lock() here.
1524
	 *
1525
	 * TODO: allowing text conversion to run in parallel with object
1526
	 * reading operations might increase performance in the multithreaded
1527
	 * non-worktreee git-grep with --textconv.
1528
	 */
1529
	obj_read_lock();
1530
	size = fill_textconv(r, driver, df, &buf);
1531
	obj_read_unlock();
1532
	free_filespec(df);
1533

1534
	/*
1535
	 * The normal fill_textconv usage by the diff machinery would just keep
1536
	 * the textconv'd buf separate from the diff_filespec. But much of the
1537
	 * grep code passes around a grep_source and assumes that its "buf"
1538
	 * pointer is the beginning of the thing we are searching. So let's
1539
	 * install our textconv'd version into the grep_source, taking care not
1540
	 * to leak any existing buffer.
1541
	 */
1542
	grep_source_clear_data(gs);
1543
	gs->buf = buf;
1544
	gs->size = size;
1545

1546
	return 0;
1547
}
1548

1549
static int is_empty_line(const char *bol, const char *eol)
1550
{
1551
	while (bol < eol && isspace(*bol))
1552
		bol++;
1553
	return bol == eol;
1554
}
1555

1556
static int grep_source_1(struct grep_opt *opt, struct grep_source *gs, int collect_hits)
1557
{
1558
	const char *bol;
1559
	const char *peek_bol = NULL;
1560
	unsigned long left;
1561
	unsigned lno = 1;
1562
	unsigned last_hit = 0;
1563
	int binary_match_only = 0;
1564
	unsigned count = 0;
1565
	int try_lookahead = 0;
1566
	int show_function = 0;
1567
	struct userdiff_driver *textconv = NULL;
1568
	enum grep_context ctx = GREP_CONTEXT_HEAD;
1569
	xdemitconf_t xecfg;
1570

1571
	if (!opt->status_only && gs->name == NULL)
1572
		BUG("grep call which could print a name requires "
1573
		    "grep_source.name be non-NULL");
1574

1575
	if (!opt->output)
1576
		opt->output = std_output;
1577

1578
	if (opt->pre_context || opt->post_context || opt->file_break ||
1579
	    opt->funcbody) {
1580
		/* Show hunk marks, except for the first file. */
1581
		if (opt->last_shown)
1582
			opt->show_hunk_mark = 1;
1583
		/*
1584
		 * If we're using threads then we can't easily identify
1585
		 * the first file.  Always put hunk marks in that case
1586
		 * and skip the very first one later in work_done().
1587
		 */
1588
		if (opt->output != std_output)
1589
			opt->show_hunk_mark = 1;
1590
	}
1591
	opt->last_shown = 0;
1592

1593
	if (opt->allow_textconv) {
1594
		grep_source_load_driver(gs, opt->repo->index);
1595
		/*
1596
		 * We might set up the shared textconv cache data here, which
1597
		 * is not thread-safe. Also, get_oid_with_context() and
1598
		 * parse_object() might be internally called. As they are not
1599
		 * currently thread-safe and might be racy with object reading,
1600
		 * obj_read_lock() must be called.
1601
		 */
1602
		grep_attr_lock();
1603
		obj_read_lock();
1604
		textconv = userdiff_get_textconv(opt->repo, gs->driver);
1605
		obj_read_unlock();
1606
		grep_attr_unlock();
1607
	}
1608

1609
	/*
1610
	 * We know the result of a textconv is text, so we only have to care
1611
	 * about binary handling if we are not using it.
1612
	 */
1613
	if (!textconv) {
1614
		switch (opt->binary) {
1615
		case GREP_BINARY_DEFAULT:
1616
			if (grep_source_is_binary(gs, opt->repo->index))
1617
				binary_match_only = 1;
1618
			break;
1619
		case GREP_BINARY_NOMATCH:
1620
			if (grep_source_is_binary(gs, opt->repo->index))
1621
				return 0; /* Assume unmatch */
1622
			break;
1623
		case GREP_BINARY_TEXT:
1624
			break;
1625
		default:
1626
			BUG("unknown binary handling mode");
1627
		}
1628
	}
1629

1630
	memset(&xecfg, 0, sizeof(xecfg));
1631
	opt->priv = &xecfg;
1632

1633
	try_lookahead = should_lookahead(opt);
1634

1635
	if (fill_textconv_grep(opt->repo, textconv, gs) < 0)
1636
		return 0;
1637

1638
	bol = gs->buf;
1639
	left = gs->size;
1640
	while (left) {
1641
		const char *eol;
1642
		int hit;
1643
		ssize_t cno;
1644
		ssize_t col = -1, icol = -1;
1645

1646
		/*
1647
		 * look_ahead() skips quickly to the line that possibly
1648
		 * has the next hit; don't call it if we need to do
1649
		 * something more than just skipping the current line
1650
		 * in response to an unmatch for the current line.  E.g.
1651
		 * inside a post-context window, we will show the current
1652
		 * line as a context around the previous hit when it
1653
		 * doesn't hit.
1654
		 */
1655
		if (try_lookahead
1656
		    && !(last_hit
1657
			 && (show_function ||
1658
			     lno <= last_hit + opt->post_context))
1659
		    && look_ahead(opt, &left, &lno, &bol))
1660
			break;
1661
		eol = end_of_line(bol, &left);
1662

1663
		if ((ctx == GREP_CONTEXT_HEAD) && (eol == bol))
1664
			ctx = GREP_CONTEXT_BODY;
1665

1666
		hit = match_line(opt, bol, eol, &col, &icol, ctx, collect_hits);
1667

1668
		if (collect_hits)
1669
			goto next_line;
1670

1671
		/* "grep -v -e foo -e bla" should list lines
1672
		 * that do not have either, so inversion should
1673
		 * be done outside.
1674
		 */
1675
		if (opt->invert)
1676
			hit = !hit;
1677
		if (opt->unmatch_name_only) {
1678
			if (hit)
1679
				return 0;
1680
			goto next_line;
1681
		}
1682
		if (hit && (opt->max_count < 0 || count < opt->max_count)) {
1683
			count++;
1684
			if (opt->status_only)
1685
				return 1;
1686
			if (opt->name_only) {
1687
				show_name(opt, gs->name);
1688
				return 1;
1689
			}
1690
			if (opt->count)
1691
				goto next_line;
1692
			if (binary_match_only) {
1693
				opt->output(opt, "Binary file ", 12);
1694
				output_color(opt, gs->name, strlen(gs->name),
1695
					     opt->colors[GREP_COLOR_FILENAME]);
1696
				opt->output(opt, " matches\n", 9);
1697
				return 1;
1698
			}
1699
			/* Hit at this line.  If we haven't shown the
1700
			 * pre-context lines, we would need to show them.
1701
			 */
1702
			if (opt->pre_context || opt->funcbody)
1703
				show_pre_context(opt, gs, bol, eol, lno);
1704
			else if (opt->funcname)
1705
				show_funcname_line(opt, gs, bol, lno);
1706
			cno = opt->invert ? icol : col;
1707
			if (cno < 0) {
1708
				/*
1709
				 * A negative cno indicates that there was no
1710
				 * match on the line. We are thus inverted and
1711
				 * being asked to show all lines that _don't_
1712
				 * match a given expression. Therefore, set cno
1713
				 * to 0 to suggest the whole line matches.
1714
				 */
1715
				cno = 0;
1716
			}
1717
			show_line(opt, bol, eol, gs->name, lno, cno + 1, ':');
1718
			last_hit = lno;
1719
			if (opt->funcbody)
1720
				show_function = 1;
1721
			goto next_line;
1722
		}
1723
		if (show_function && (!peek_bol || peek_bol < bol)) {
1724
			unsigned long peek_left = left;
1725
			const char *peek_eol = eol;
1726

1727
			/*
1728
			 * Trailing empty lines are not interesting.
1729
			 * Peek past them to see if they belong to the
1730
			 * body of the current function.
1731
			 */
1732
			peek_bol = bol;
1733
			while (is_empty_line(peek_bol, peek_eol)) {
1734
				peek_bol = peek_eol + 1;
1735
				peek_eol = end_of_line(peek_bol, &peek_left);
1736
			}
1737

1738
			if (peek_bol >= gs->buf + gs->size ||
1739
			    match_funcname(opt, gs, peek_bol, peek_eol))
1740
				show_function = 0;
1741
		}
1742
		if (show_function ||
1743
		    (last_hit && lno <= last_hit + opt->post_context)) {
1744
			/* If the last hit is within the post context,
1745
			 * we need to show this line.
1746
			 */
1747
			show_line(opt, bol, eol, gs->name, lno, col + 1, '-');
1748
		}
1749

1750
	next_line:
1751
		bol = eol + 1;
1752
		if (!left)
1753
			break;
1754
		left--;
1755
		lno++;
1756
	}
1757

1758
	if (collect_hits)
1759
		return 0;
1760

1761
	if (opt->status_only)
1762
		return opt->unmatch_name_only;
1763
	if (opt->unmatch_name_only) {
1764
		/* We did not see any hit, so we want to show this */
1765
		show_name(opt, gs->name);
1766
		return 1;
1767
	}
1768

1769
	xdiff_clear_find_func(&xecfg);
1770
	opt->priv = NULL;
1771

1772
	/* NEEDSWORK:
1773
	 * The real "grep -c foo *.c" gives many "bar.c:0" lines,
1774
	 * which feels mostly useless but sometimes useful.  Maybe
1775
	 * make it another option?  For now suppress them.
1776
	 */
1777
	if (opt->count && count) {
1778
		char buf[32];
1779
		if (opt->pathname) {
1780
			output_color(opt, gs->name, strlen(gs->name),
1781
				     opt->colors[GREP_COLOR_FILENAME]);
1782
			output_sep(opt, ':');
1783
		}
1784
		xsnprintf(buf, sizeof(buf), "%u\n", count);
1785
		opt->output(opt, buf, strlen(buf));
1786
		return 1;
1787
	}
1788
	return !!last_hit;
1789
}
1790

1791
static void clr_hit_marker(struct grep_expr *x)
1792
{
1793
	/* All-hit markers are meaningful only at the very top level
1794
	 * OR node.
1795
	 */
1796
	while (1) {
1797
		x->hit = 0;
1798
		if (x->node != GREP_NODE_OR)
1799
			return;
1800
		x->u.binary.left->hit = 0;
1801
		x = x->u.binary.right;
1802
	}
1803
}
1804

1805
static int chk_hit_marker(struct grep_expr *x)
1806
{
1807
	/* Top level nodes have hit markers.  See if they all are hits */
1808
	while (1) {
1809
		if (x->node != GREP_NODE_OR)
1810
			return x->hit;
1811
		if (!x->u.binary.left->hit)
1812
			return 0;
1813
		x = x->u.binary.right;
1814
	}
1815
}
1816

1817
int grep_source(struct grep_opt *opt, struct grep_source *gs)
1818
{
1819
	/*
1820
	 * we do not have to do the two-pass grep when we do not check
1821
	 * buffer-wide "all-match".
1822
	 */
1823
	if (!opt->all_match && !opt->no_body_match)
1824
		return grep_source_1(opt, gs, 0);
1825

1826
	/* Otherwise the toplevel "or" terms hit a bit differently.
1827
	 * We first clear hit markers from them.
1828
	 */
1829
	clr_hit_marker(opt->pattern_expression);
1830
	opt->body_hit = 0;
1831
	grep_source_1(opt, gs, 1);
1832

1833
	if (opt->all_match && !chk_hit_marker(opt->pattern_expression))
1834
		return 0;
1835
	if (opt->no_body_match && opt->body_hit)
1836
		return 0;
1837

1838
	return grep_source_1(opt, gs, 0);
1839
}
1840

1841
static void grep_source_init_buf(struct grep_source *gs,
1842
				 const char *buf,
1843
				 unsigned long size)
1844
{
1845
	gs->type = GREP_SOURCE_BUF;
1846
	gs->name = NULL;
1847
	gs->path = NULL;
1848
	gs->buf = buf;
1849
	gs->size = size;
1850
	gs->driver = NULL;
1851
	gs->identifier = NULL;
1852
}
1853

1854
int grep_buffer(struct grep_opt *opt, const char *buf, unsigned long size)
1855
{
1856
	struct grep_source gs;
1857
	int r;
1858

1859
	grep_source_init_buf(&gs, buf, size);
1860

1861
	r = grep_source(opt, &gs);
1862

1863
	grep_source_clear(&gs);
1864
	return r;
1865
}
1866

1867
void grep_source_init_file(struct grep_source *gs, const char *name,
1868
			   const char *path)
1869
{
1870
	gs->type = GREP_SOURCE_FILE;
1871
	gs->name = xstrdup_or_null(name);
1872
	gs->path = xstrdup_or_null(path);
1873
	gs->buf = NULL;
1874
	gs->size = 0;
1875
	gs->driver = NULL;
1876
	gs->identifier = xstrdup(path);
1877
}
1878

1879
void grep_source_init_oid(struct grep_source *gs, const char *name,
1880
			  const char *path, const struct object_id *oid,
1881
			  struct repository *repo)
1882
{
1883
	gs->type = GREP_SOURCE_OID;
1884
	gs->name = xstrdup_or_null(name);
1885
	gs->path = xstrdup_or_null(path);
1886
	gs->buf = NULL;
1887
	gs->size = 0;
1888
	gs->driver = NULL;
1889
	gs->identifier = oiddup(oid);
1890
	gs->repo = repo;
1891
}
1892

1893
void grep_source_clear(struct grep_source *gs)
1894
{
1895
	FREE_AND_NULL(gs->name);
1896
	FREE_AND_NULL(gs->path);
1897
	FREE_AND_NULL(gs->identifier);
1898
	grep_source_clear_data(gs);
1899
}
1900

1901
void grep_source_clear_data(struct grep_source *gs)
1902
{
1903
	switch (gs->type) {
1904
	case GREP_SOURCE_FILE:
1905
	case GREP_SOURCE_OID:
1906
		/* these types own the buffer */
1907
		free((char *)gs->buf);
1908
		gs->buf = NULL;
1909
		gs->size = 0;
1910
		break;
1911
	case GREP_SOURCE_BUF:
1912
		/* leave user-provided buf intact */
1913
		break;
1914
	}
1915
}
1916

1917
static int grep_source_load_oid(struct grep_source *gs)
1918
{
1919
	enum object_type type;
1920

1921
	gs->buf = repo_read_object_file(gs->repo, gs->identifier, &type,
1922
					&gs->size);
1923
	if (!gs->buf)
1924
		return error(_("'%s': unable to read %s"),
1925
			     gs->name,
1926
			     oid_to_hex(gs->identifier));
1927
	return 0;
1928
}
1929

1930
static int grep_source_load_file(struct grep_source *gs)
1931
{
1932
	const char *filename = gs->identifier;
1933
	struct stat st;
1934
	char *data;
1935
	size_t size;
1936
	int i;
1937

1938
	if (lstat(filename, &st) < 0) {
1939
	err_ret:
1940
		if (errno != ENOENT)
1941
			error_errno(_("failed to stat '%s'"), filename);
1942
		return -1;
1943
	}
1944
	if (!S_ISREG(st.st_mode))
1945
		return -1;
1946
	size = xsize_t(st.st_size);
1947
	i = open(filename, O_RDONLY);
1948
	if (i < 0)
1949
		goto err_ret;
1950
	data = xmallocz(size);
1951
	if (st.st_size != read_in_full(i, data, size)) {
1952
		error_errno(_("'%s': short read"), filename);
1953
		close(i);
1954
		free(data);
1955
		return -1;
1956
	}
1957
	close(i);
1958

1959
	gs->buf = data;
1960
	gs->size = size;
1961
	return 0;
1962
}
1963

1964
static int grep_source_load(struct grep_source *gs)
1965
{
1966
	if (gs->buf)
1967
		return 0;
1968

1969
	switch (gs->type) {
1970
	case GREP_SOURCE_FILE:
1971
		return grep_source_load_file(gs);
1972
	case GREP_SOURCE_OID:
1973
		return grep_source_load_oid(gs);
1974
	case GREP_SOURCE_BUF:
1975
		return gs->buf ? 0 : -1;
1976
	}
1977
	BUG("invalid grep_source type to load");
1978
}
1979

1980
void grep_source_load_driver(struct grep_source *gs,
1981
			     struct index_state *istate)
1982
{
1983
	if (gs->driver)
1984
		return;
1985

1986
	grep_attr_lock();
1987
	if (gs->path)
1988
		gs->driver = userdiff_find_by_path(istate, gs->path);
1989
	if (!gs->driver)
1990
		gs->driver = userdiff_find_by_name("default");
1991
	grep_attr_unlock();
1992
}
1993

1994
static int grep_source_is_binary(struct grep_source *gs,
1995
				 struct index_state *istate)
1996
{
1997
	grep_source_load_driver(gs, istate);
1998
	if (gs->driver->binary != -1)
1999
		return gs->driver->binary;
2000

2001
	if (!grep_source_load(gs))
2002
		return buffer_is_binary(gs->buf, gs->size);
2003

2004
	return 0;
2005
}
2006