apply.c 114.5 KB
Newer Older
L
Linus Torvalds 已提交
1 2 3 4 5 6 7 8 9
/*
 * apply.c
 *
 * Copyright (C) Linus Torvalds, 2005
 *
 * This applies patches on top of some (arbitrary) version of the SCM.
 *
 */
#include "cache.h"
10
#include "cache-tree.h"
11
#include "quote.h"
12
#include "blob.h"
J
Junio C Hamano 已提交
13
#include "delta.h"
P
Peter Eriksen 已提交
14
#include "builtin.h"
15
#include "string-list.h"
16
#include "dir.h"
17
#include "diff.h"
M
Miklos Vajna 已提交
18
#include "parse-options.h"
19 20
#include "xdiff-interface.h"
#include "ll-merge.h"
21
#include "rerere.h"
L
Linus Torvalds 已提交
22

23 24 25 26 27 28 29 30 31
/*
 *  --check turns on checking that the working tree matches the
 *    files that are being modified, but doesn't apply the patch
 *  --stat does just a diffstat, and doesn't actually apply
 *  --numstat does numeric diffstat, and doesn't actually apply
 *  --index-info shows the old and new index info for paths if available.
 *  --index updates the cache as well.
 *  --cached updates only the cache without ever touching the working tree.
 */
32 33
static const char *prefix;
static int prefix_length = -1;
34
static int newfd = -1;
35

36
static int unidiff_zero;
37
static int p_value = 1;
38
static int p_value_known;
39
static int check_index;
40
static int update_index;
41 42 43 44 45
static int cached;
static int diffstat;
static int numstat;
static int summary;
static int check;
46
static int apply = 1;
47
static int apply_in_reverse;
J
Junio C Hamano 已提交
48
static int apply_with_reject;
J
Junio C Hamano 已提交
49
static int apply_verbosely;
50
static int allow_overlap;
51
static int no_add;
52
static int threeway;
53
static const char *fake_ancestor;
54
static int line_termination = '\n';
M
Miklos Vajna 已提交
55 56
static unsigned int p_context = UINT_MAX;
static const char * const apply_usage[] = {
57
	N_("git apply [options] [<patch>...]"),
M
Miklos Vajna 已提交
58 59
	NULL
};
L
Linus Torvalds 已提交
60

61 62 63 64
static enum ws_error_action {
	nowarn_ws_error,
	warn_on_ws_error,
	die_on_ws_error,
65
	correct_ws_error
66
} ws_error_action = warn_on_ws_error;
67
static int whitespace_error;
68
static int squelch_whitespace_errors = 5;
69
static int applied_after_fixing_ws;
70 71 72

static enum ws_ignore {
	ignore_ws_none,
73
	ignore_ws_change
74 75 76
} ws_ignore_action = ignore_ws_none;


77
static const char *patch_input_file;
78 79
static const char *root;
static int root_len;
M
Miklos Vajna 已提交
80 81
static int read_stdin = 1;
static int options;
L
Linus Torvalds 已提交
82

83 84 85
static void parse_whitespace_option(const char *option)
{
	if (!option) {
86
		ws_error_action = warn_on_ws_error;
87 88 89
		return;
	}
	if (!strcmp(option, "warn")) {
90
		ws_error_action = warn_on_ws_error;
91 92
		return;
	}
J
Junio C Hamano 已提交
93
	if (!strcmp(option, "nowarn")) {
94
		ws_error_action = nowarn_ws_error;
J
Junio C Hamano 已提交
95 96
		return;
	}
97
	if (!strcmp(option, "error")) {
98
		ws_error_action = die_on_ws_error;
99 100 101
		return;
	}
	if (!strcmp(option, "error-all")) {
102
		ws_error_action = die_on_ws_error;
103 104 105
		squelch_whitespace_errors = 0;
		return;
	}
106 107
	if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
		ws_error_action = correct_ws_error;
108 109
		return;
	}
110
	die(_("unrecognized whitespace option '%s'"), option);
111 112
}

113 114 115 116 117 118 119 120 121 122 123 124
static void parse_ignorewhitespace_option(const char *option)
{
	if (!option || !strcmp(option, "no") ||
	    !strcmp(option, "false") || !strcmp(option, "never") ||
	    !strcmp(option, "none")) {
		ws_ignore_action = ignore_ws_none;
		return;
	}
	if (!strcmp(option, "change")) {
		ws_ignore_action = ignore_ws_change;
		return;
	}
125
	die(_("unrecognized whitespace ignore option '%s'"), option);
126 127
}

128 129
static void set_default_whitespace_mode(const char *whitespace_option)
{
130 131
	if (!whitespace_option && !apply_default_whitespace)
		ws_error_action = (apply ? warn_on_ws_error : nowarn_ws_error);
132 133
}

134 135 136 137 138 139 140
/*
 * For "diff-stat" like behaviour, we keep track of the biggest change
 * we've seen, and the longest filename. That allows us to do simple
 * scaling.
 */
static int max_change, max_len;

141 142 143 144 145
/*
 * Various "current state", notably line numbers and what
 * file (and how) we're patching right now.. The "is_xxxx"
 * things are flags, where -1 means "don't know yet".
 */
146
static int linenr = 1;
147

148 149 150 151 152 153 154
/*
 * This represents one "hunk" from a patch, starting with
 * "@@ -oldpos,oldlines +newpos,newlines @@" marker.  The
 * patch text is pointed at by patch, and its byte length
 * is stored in size.  leading and trailing are the number
 * of context lines.
 */
155
struct fragment {
156
	unsigned long leading, trailing;
157 158
	unsigned long oldpos, oldlines;
	unsigned long newpos, newlines;
159 160 161 162
	/*
	 * 'patch' is usually borrowed from buf in apply_patch(),
	 * but some codepaths store an allocated buffer.
	 */
163
	const char *patch;
164 165
	unsigned free_patch:1,
		rejected:1;
166
	int size;
167
	int linenr;
168 169 170
	struct fragment *next;
};

171 172 173 174 175 176 177 178 179
/*
 * When dealing with a binary patch, we reuse "leading" field
 * to store the type of the binary hunk, either deflated "delta"
 * or deflated "literal".
 */
#define binary_patch_method leading
#define BINARY_DELTA_DEFLATED	1
#define BINARY_LITERAL_DEFLATED 2

180 181 182 183 184
/*
 * This represents a "patch" to a file, both metainfo changes
 * such as creation/deletion, filemode and content changes represented
 * as a series of fragments.
 */
185
struct patch {
186
	char *new_name, *old_name, *def_name;
187
	unsigned int old_mode, new_mode;
188
	int is_new, is_delete;	/* -1 = unknown, 0 = false, 1 = true */
J
Junio C Hamano 已提交
189
	int rejected;
190
	unsigned ws_rule;
191
	int lines_added, lines_deleted;
192
	int score;
193
	unsigned int is_toplevel_relative:1;
194 195 196 197
	unsigned int inaccurate_eof:1;
	unsigned int is_binary:1;
	unsigned int is_copy:1;
	unsigned int is_rename:1;
198
	unsigned int recount:1;
199
	unsigned int conflicted_threeway:1;
200
	unsigned int direct_to_threeway:1;
201
	struct fragment *fragments;
202
	char *result;
203
	size_t resultsize;
204 205
	char old_sha1_prefix[41];
	char new_sha1_prefix[41];
206
	struct patch *next;
207 208 209

	/* three-way fallback result */
	unsigned char threeway_stage[3][20];
210
};
211

212
static void free_fragment_list(struct fragment *list)
213
{
214 215 216 217 218 219
	while (list) {
		struct fragment *next = list->next;
		if (list->free_patch)
			free((char *)list->patch);
		free(list);
		list = next;
220
	}
221 222 223 224 225
}

static void free_patch(struct patch *patch)
{
	free_fragment_list(patch->fragments);
226 227 228
	free(patch->def_name);
	free(patch->old_name);
	free(patch->new_name);
J
Junio C Hamano 已提交
229
	free(patch->result);
230 231 232 233 234 235 236 237 238
	free(patch);
}

static void free_patch_list(struct patch *list)
{
	while (list) {
		struct patch *next = list->next;
		free_patch(list);
		list = next;
239 240 241
	}
}

242 243 244 245 246 247 248 249 250
/*
 * A line in a file, len-bytes long (includes the terminating LF,
 * except for an incomplete line at the end if the file ends with
 * one), and its contents hashes to 'hash'.
 */
struct line {
	size_t len;
	unsigned hash : 24;
	unsigned flag : 8;
251
#define LINE_COMMON     1
252
#define LINE_PATCHED	2
253 254 255 256 257 258 259 260 261
};

/*
 * This represents a "file", which is an array of "lines".
 */
struct image {
	char *buf;
	size_t len;
	size_t nr;
262
	size_t alloc;
263 264 265 266
	struct line *line_allocated;
	struct line *line;
};

267 268 269 270 271
/*
 * Records filenames that have been touched, in order to handle
 * the case where more than one patches touch the same file.
 */

272
static struct string_list fn_table;
273

274 275 276 277 278 279 280 281 282 283 284 285
static uint32_t hash_line(const char *cp, size_t len)
{
	size_t i;
	uint32_t h;
	for (i = 0, h = 0; i < len; i++) {
		if (!isspace(cp[i])) {
			h = h * 3 + (cp[i] & 0xff);
		}
	}
	return h;
}

286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302
/*
 * Compare lines s1 of length n1 and s2 of length n2, ignoring
 * whitespace difference. Returns 1 if they match, 0 otherwise
 */
static int fuzzy_matchlines(const char *s1, size_t n1,
			    const char *s2, size_t n2)
{
	const char *last1 = s1 + n1 - 1;
	const char *last2 = s2 + n2 - 1;
	int result = 0;

	/* ignore line endings */
	while ((*last1 == '\r') || (*last1 == '\n'))
		last1--;
	while ((*last2 == '\r') || (*last2 == '\n'))
		last2--;

303 304 305 306 307 308 309
	/* skip leading whitespaces, if both begin with whitespace */
	if (s1 <= last1 && s2 <= last2 && isspace(*s1) && isspace(*s2)) {
		while (isspace(*s1) && (s1 <= last1))
			s1++;
		while (isspace(*s2) && (s2 <= last2))
			s2++;
	}
310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340
	/* early return if both lines are empty */
	if ((s1 > last1) && (s2 > last2))
		return 1;
	while (!result) {
		result = *s1++ - *s2++;
		/*
		 * Skip whitespace inside. We check for whitespace on
		 * both buffers because we don't want "a b" to match
		 * "ab"
		 */
		if (isspace(*s1) && isspace(*s2)) {
			while (isspace(*s1) && s1 <= last1)
				s1++;
			while (isspace(*s2) && s2 <= last2)
				s2++;
		}
		/*
		 * If we reached the end on one side only,
		 * lines don't match
		 */
		if (
		    ((s2 > last2) && (s1 <= last1)) ||
		    ((s1 > last1) && (s2 <= last2)))
			return 0;
		if ((s1 > last1) && (s2 > last2))
			break;
	}

	return !result;
}

341 342 343 344 345 346 347 348 349
static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
{
	ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
	img->line_allocated[img->nr].len = len;
	img->line_allocated[img->nr].hash = hash_line(bol, len);
	img->line_allocated[img->nr].flag = flag;
	img->nr++;
}

350 351 352 353 354
/*
 * "buf" has the file contents to be patched (read from various sources).
 * attach it to "image" and add line-based index to it.
 * "image" now owns the "buf".
 */
355 356 357 358 359
static void prepare_image(struct image *image, char *buf, size_t len,
			  int prepare_linetable)
{
	const char *cp, *ep;

360
	memset(image, 0, sizeof(*image));
361 362 363
	image->buf = buf;
	image->len = len;

364
	if (!prepare_linetable)
365 366 367 368 369 370 371 372 373 374
		return;

	ep = image->buf + image->len;
	cp = image->buf;
	while (cp < ep) {
		const char *next;
		for (next = cp; next < ep && *next != '\n'; next++)
			;
		if (next < ep)
			next++;
375
		add_line_info(image, cp, next - cp, 0);
376 377
		cp = next;
	}
378
	image->line = image->line_allocated;
379 380 381 382 383
}

static void clear_image(struct image *image)
{
	free(image->buf);
384 385
	free(image->line_allocated);
	memset(image, 0, sizeof(*image));
386 387
}

388 389
/* fmt must contain _one_ %s and no other substitution */
static void say_patch_name(FILE *output, const char *fmt, struct patch *patch)
J
Junio C Hamano 已提交
390
{
391 392
	struct strbuf sb = STRBUF_INIT;

J
Junio C Hamano 已提交
393 394
	if (patch->old_name && patch->new_name &&
	    strcmp(patch->old_name, patch->new_name)) {
395 396 397
		quote_c_style(patch->old_name, &sb, NULL, 0);
		strbuf_addstr(&sb, " => ");
		quote_c_style(patch->new_name, &sb, NULL, 0);
398
	} else {
J
Junio C Hamano 已提交
399 400 401
		const char *n = patch->new_name;
		if (!n)
			n = patch->old_name;
402
		quote_c_style(n, &sb, NULL, 0);
J
Junio C Hamano 已提交
403
	}
404 405 406
	fprintf(output, fmt, sb.buf);
	fputc('\n', output);
	strbuf_release(&sb);
J
Junio C Hamano 已提交
407 408
}

409
#define SLOP (16)
L
Linus Torvalds 已提交
410

411
static void read_patch_file(struct strbuf *sb, int fd)
L
Linus Torvalds 已提交
412
{
413
	if (strbuf_read(sb, fd, 0) < 0)
414
		die_errno("git apply: failed to read");
415 416 417 418 419 420

	/*
	 * Make sure that we have some slop in the buffer
	 * so that we can do speculative "memcmp" etc, and
	 * see to it that it is NUL-filled.
	 */
421 422
	strbuf_grow(sb, SLOP);
	memset(sb->buf + sb->len, 0, SLOP);
L
Linus Torvalds 已提交
423 424
}

425
static unsigned long linelen(const char *buffer, unsigned long size)
L
Linus Torvalds 已提交
426 427 428 429 430 431 432 433 434 435
{
	unsigned long len = 0;
	while (size--) {
		len++;
		if (*buffer++ == '\n')
			break;
	}
	return len;
}

436 437 438 439 440
static int is_dev_null(const char *str)
{
	return !memcmp("/dev/null", str, 9) && isspace(str[9]);
}

441 442
#define TERM_SPACE	1
#define TERM_TAB	2
443 444 445 446 447 448 449 450 451 452 453

static int name_terminate(const char *name, int namelen, int c, int terminate)
{
	if (c == ' ' && !(terminate & TERM_SPACE))
		return 0;
	if (c == '\t' && !(terminate & TERM_TAB))
		return 0;

	return 1;
}

454 455 456 457 458
/* remove double slashes to make --index work with such filenames */
static char *squash_slash(char *name)
{
	int i = 0, j = 0;

459 460 461
	if (!name)
		return NULL;

462 463 464 465 466 467 468 469 470
	while (name[i]) {
		if ((name[j++] = name[i++]) == '/')
			while (name[i] == '/')
				i++;
	}
	name[j] = '\0';
	return name;
}

471
static char *find_name_gnu(const char *line, const char *def, int p_value)
L
Linus Torvalds 已提交
472
{
473 474
	struct strbuf name = STRBUF_INIT;
	char *cp;
475

476 477
	/*
	 * Proposed "new-style" GNU patch/diff format; see
478
	 * http://marc.info/?l=git&m=112927316408690&w=2
479 480 481 482 483
	 */
	if (unquote_c_style(&name, line, NULL)) {
		strbuf_release(&name);
		return NULL;
	}
484

485 486 487 488 489
	for (cp = name.buf; p_value; p_value--) {
		cp = strchr(cp, '/');
		if (!cp) {
			strbuf_release(&name);
			return NULL;
490
		}
491 492 493 494 495 496 497 498 499
		cp++;
	}

	strbuf_remove(&name, 0, cp - name.buf);
	if (root)
		strbuf_insert(&name, 0, root, root_len);
	return squash_slash(strbuf_detach(&name, NULL));
}

500
static size_t sane_tz_len(const char *line, size_t len)
L
Linus Torvalds 已提交
501
{
502
	const char *tz, *p;
503

504 505 506 507 508 509 510 511 512 513 514 515 516 517
	if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
		return 0;
	tz = line + len - strlen(" +0500");

	if (tz[1] != '+' && tz[1] != '-')
		return 0;

	for (p = tz + 2; p != line + len; p++)
		if (!isdigit(*p))
			return 0;

	return line + len - tz;
}

518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535
static size_t tz_with_colon_len(const char *line, size_t len)
{
	const char *tz, *p;

	if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
		return 0;
	tz = line + len - strlen(" +08:00");

	if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
		return 0;
	p = tz + 2;
	if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
	    !isdigit(*p++) || !isdigit(*p++))
		return 0;

	return line + len - tz;
}

536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610
static size_t date_len(const char *line, size_t len)
{
	const char *date, *p;

	if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
		return 0;
	p = date = line + len - strlen("72-02-05");

	if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
	    !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
	    !isdigit(*p++) || !isdigit(*p++))	/* Not a date. */
		return 0;

	if (date - line >= strlen("19") &&
	    isdigit(date[-1]) && isdigit(date[-2]))	/* 4-digit year */
		date -= strlen("19");

	return line + len - date;
}

static size_t short_time_len(const char *line, size_t len)
{
	const char *time, *p;

	if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
		return 0;
	p = time = line + len - strlen(" 07:01:32");

	/* Permit 1-digit hours? */
	if (*p++ != ' ' ||
	    !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
	    !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
	    !isdigit(*p++) || !isdigit(*p++))	/* Not a time. */
		return 0;

	return line + len - time;
}

static size_t fractional_time_len(const char *line, size_t len)
{
	const char *p;
	size_t n;

	/* Expected format: 19:41:17.620000023 */
	if (!len || !isdigit(line[len - 1]))
		return 0;
	p = line + len - 1;

	/* Fractional seconds. */
	while (p > line && isdigit(*p))
		p--;
	if (*p != '.')
		return 0;

	/* Hours, minutes, and whole seconds. */
	n = short_time_len(line, p - line);
	if (!n)
		return 0;

	return line + len - p + n;
}

static size_t trailing_spaces_len(const char *line, size_t len)
{
	const char *p;

	/* Expected format: ' ' x (1 or more)  */
	if (!len || line[len - 1] != ' ')
		return 0;

	p = line + len;
	while (p != line) {
		p--;
		if (*p != ' ')
			return line + len - (p + 1);
611 612
	}

613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629
	/* All spaces! */
	return len;
}

static size_t diff_timestamp_len(const char *line, size_t len)
{
	const char *end = line + len;
	size_t n;

	/*
	 * Posix: 2010-07-05 19:41:17
	 * GNU: 2010-07-05 19:41:17.620000023 -0500
	 */

	if (!isdigit(end[-1]))
		return 0;

630 631 632
	n = sane_tz_len(line, end - line);
	if (!n)
		n = tz_with_colon_len(line, end - line);
633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658
	end -= n;

	n = short_time_len(line, end - line);
	if (!n)
		n = fractional_time_len(line, end - line);
	end -= n;

	n = date_len(line, end - line);
	if (!n)	/* No date.  Too bad. */
		return 0;
	end -= n;

	if (end == line)	/* No space before date. */
		return 0;
	if (end[-1] == '\t') {	/* Success! */
		end--;
		return line + len - end;
	}
	if (end[-1] != ' ')	/* No space before date. */
		return 0;

	/* Whitespace damage. */
	end -= trailing_spaces_len(line, end - line);
	return line + len - end;
}

659 660 661 662 663 664 665
static char *null_strdup(const char *s)
{
	return s ? xstrdup(s) : NULL;
}

static char *find_name_common(const char *line, const char *def,
			      int p_value, const char *end, int terminate)
666 667 668 669
{
	int len;
	const char *start = NULL;

670 671
	if (p_value == 0)
		start = line;
672
	while (line != end) {
673
		char c = *line;
674

675
		if (!end && isspace(c)) {
676 677 678 679 680
			if (c == '\n')
				break;
			if (name_terminate(start, line-start, c, terminate))
				break;
		}
681 682 683 684 685
		line++;
		if (c == '/' && !--p_value)
			start = line;
	}
	if (!start)
686
		return squash_slash(null_strdup(def));
687 688
	len = line - start;
	if (!len)
689
		return squash_slash(null_strdup(def));
690 691 692 693 694 695 696 697 698 699

	/*
	 * Generally we prefer the shorter name, especially
	 * if the other one is just a variation of that with
	 * something else tacked on to the end (ie "file.orig"
	 * or "file~").
	 */
	if (def) {
		int deflen = strlen(def);
		if (deflen < len && !strncmp(start, def, deflen))
700
			return squash_slash(xstrdup(def));
L
Linus Torvalds 已提交
701
	}
702

703 704 705 706 707
	if (root) {
		char *ret = xmalloc(root_len + len + 1);
		strcpy(ret, root);
		memcpy(ret + root_len, start, len);
		ret[root_len + len] = '\0';
708
		return squash_slash(ret);
709 710
	}

711
	return squash_slash(xmemdupz(start, len));
712 713
}

714 715 716 717 718 719 720 721 722 723 724 725 726
static char *find_name(const char *line, char *def, int p_value, int terminate)
{
	if (*line == '"') {
		char *name = find_name_gnu(line, def, p_value);
		if (name)
			return name;
	}

	return find_name_common(line, def, p_value, NULL, terminate);
}

static char *find_name_traditional(const char *line, char *def, int p_value)
{
727
	size_t len;
728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744
	size_t date_len;

	if (*line == '"') {
		char *name = find_name_gnu(line, def, p_value);
		if (name)
			return name;
	}

	len = strchrnul(line, '\n') - line;
	date_len = diff_timestamp_len(line, len);
	if (!date_len)
		return find_name_common(line, def, p_value, NULL, TERM_TAB);
	len -= date_len;

	return find_name_common(line, def, p_value, line + len, 0);
}

745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766
static int count_slashes(const char *cp)
{
	int cnt = 0;
	char ch;

	while ((ch = *cp++))
		if (ch == '/')
			cnt++;
	return cnt;
}

/*
 * Given the string after "--- " or "+++ ", guess the appropriate
 * p_value for the given patch.
 */
static int guess_p_value(const char *nameline)
{
	char *name, *cp;
	int val = -1;

	if (is_dev_null(nameline))
		return -1;
767
	name = find_name_traditional(nameline, NULL, 0);
768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789
	if (!name)
		return -1;
	cp = strchr(name, '/');
	if (!cp)
		val = 0;
	else if (prefix) {
		/*
		 * Does it begin with "a/$our-prefix" and such?  Then this is
		 * very likely to apply to our directory.
		 */
		if (!strncmp(name, prefix, prefix_length))
			val = count_slashes(prefix);
		else {
			cp++;
			if (!strncmp(cp, prefix, prefix_length))
				val = count_slashes(prefix) + 1;
		}
	}
	free(name);
	return val;
}

790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807
/*
 * Does the ---/+++ line has the POSIX timestamp after the last HT?
 * GNU diff puts epoch there to signal a creation/deletion event.  Is
 * this such a timestamp?
 */
static int has_epoch_timestamp(const char *nameline)
{
	/*
	 * We are only interested in epoch timestamp; any non-zero
	 * fraction cannot be one, hence "(\.0+)?" in the regexp below.
	 * For the same reason, the date must be either 1969-12-31 or
	 * 1970-01-01, and the seconds part must be "00".
	 */
	const char stamp_regexp[] =
		"^(1969-12-31|1970-01-01)"
		" "
		"[0-2][0-9]:[0-5][0-9]:00(\\.0+)?"
		" "
808 809
		"([-+][0-2][0-9]:?[0-5][0-9])\n";
	const char *timestamp = NULL, *cp, *colon;
810 811 812 813 814 815 816 817 818 819 820 821 822 823 824
	static regex_t *stamp;
	regmatch_t m[10];
	int zoneoffset;
	int hourminute;
	int status;

	for (cp = nameline; *cp != '\n'; cp++) {
		if (*cp == '\t')
			timestamp = cp + 1;
	}
	if (!timestamp)
		return 0;
	if (!stamp) {
		stamp = xmalloc(sizeof(*stamp));
		if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
825
			warning(_("Cannot prepare timestamp regexp %s"),
826 827 828 829 830 831 832 833
				stamp_regexp);
			return 0;
		}
	}

	status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
	if (status) {
		if (status != REG_NOMATCH)
834
			warning(_("regexec returned %d for input: %s"),
835 836 837 838
				status, timestamp);
		return 0;
	}

839 840 841 842 843
	zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
	if (*colon == ':')
		zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
	else
		zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862
	if (timestamp[m[3].rm_so] == '-')
		zoneoffset = -zoneoffset;

	/*
	 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
	 * (west of GMT) or 1970-01-01 (east of GMT)
	 */
	if ((zoneoffset < 0 && memcmp(timestamp, "1969-12-31", 10)) ||
	    (0 <= zoneoffset && memcmp(timestamp, "1970-01-01", 10)))
		return 0;

	hourminute = (strtol(timestamp + 11, NULL, 10) * 60 +
		      strtol(timestamp + 14, NULL, 10) -
		      zoneoffset);

	return ((zoneoffset < 0 && hourminute == 1440) ||
		(0 <= zoneoffset && !hourminute));
}

863
/*
J
Junio C Hamano 已提交
864
 * Get the name etc info from the ---/+++ lines of a traditional patch header
865
 *
866 867 868
 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
 * files, we can happily check the index for a match, but for creating a
 * new file we should try to match whatever "patch" does. I have no idea.
869
 */
870
static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
871 872 873
{
	char *name;

874 875
	first += 4;	/* skip "--- " */
	second += 4;	/* skip "+++ " */
876 877 878 879 880 881 882 883 884 885
	if (!p_value_known) {
		int p, q;
		p = guess_p_value(first);
		q = guess_p_value(second);
		if (p < 0) p = q;
		if (0 <= p && p == q) {
			p_value = p;
			p_value_known = 1;
		}
	}
886
	if (is_dev_null(first)) {
887 888
		patch->is_new = 1;
		patch->is_delete = 0;
889
		name = find_name_traditional(second, NULL, p_value);
890
		patch->new_name = name;
891
	} else if (is_dev_null(second)) {
892 893
		patch->is_new = 0;
		patch->is_delete = 1;
894
		name = find_name_traditional(first, NULL, p_value);
895
		patch->old_name = name;
896
	} else {
897 898 899 900
		char *first_name;
		first_name = find_name_traditional(first, NULL, p_value);
		name = find_name_traditional(second, first_name, p_value);
		free(first_name);
901 902 903 904 905 906 907 908 909
		if (has_epoch_timestamp(first)) {
			patch->is_new = 1;
			patch->is_delete = 0;
			patch->new_name = name;
		} else if (has_epoch_timestamp(second)) {
			patch->is_new = 0;
			patch->is_delete = 1;
			patch->old_name = name;
		} else {
910
			patch->old_name = name;
911
			patch->new_name = null_strdup(name);
912
		}
913 914
	}
	if (!name)
915
		die(_("unable to find filename in patch at line %d"), linenr);
916 917
}

918
static int gitdiff_hdrend(const char *line, struct patch *patch)
919 920 921 922
{
	return -1;
}

923 924 925 926 927 928 929 930 931
/*
 * We're anal about diff header consistency, to make
 * sure that we don't end up having strange ambiguous
 * patches floating around.
 *
 * As a result, gitdiff_{old|new}name() will check
 * their names against any previous information, just
 * to make sure..
 */
932 933 934 935
#define DIFF_OLD_NAME 0
#define DIFF_NEW_NAME 1

static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, int side)
936 937
{
	if (!orig_name && !isnull)
938
		return find_name(line, NULL, p_value, TERM_TAB);
939 940

	if (orig_name) {
941 942 943
		int len;
		const char *name;
		char *another;
944 945 946
		name = orig_name;
		len = strlen(name);
		if (isnull)
947
			die(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"), name, linenr);
948
		another = find_name(line, NULL, p_value, TERM_TAB);
949
		if (!another || memcmp(another, name, len + 1))
950 951 952
			die((side == DIFF_NEW_NAME) ?
			    _("git apply: bad git-diff - inconsistent new filename on line %d") :
			    _("git apply: bad git-diff - inconsistent old filename on line %d"), linenr);
953
		free(another);
954 955
		return orig_name;
	}
956 957 958
	else {
		/* expect "/dev/null" */
		if (memcmp("/dev/null", line, 9) || line[9] != '\n')
959
			die(_("git apply: bad git-diff - expected /dev/null on line %d"), linenr);
960 961
		return NULL;
	}
962 963
}

964
static int gitdiff_oldname(const char *line, struct patch *patch)
965
{
966
	char *orig = patch->old_name;
967 968
	patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name,
					      DIFF_OLD_NAME);
969 970
	if (orig != patch->old_name)
		free(orig);
971 972 973
	return 0;
}

974
static int gitdiff_newname(const char *line, struct patch *patch)
975
{
976
	char *orig = patch->new_name;
977 978
	patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name,
					      DIFF_NEW_NAME);
979 980
	if (orig != patch->new_name)
		free(orig);
981 982 983
	return 0;
}

984
static int gitdiff_oldmode(const char *line, struct patch *patch)
985
{
986
	patch->old_mode = strtoul(line, NULL, 8);
987 988 989
	return 0;
}

990
static int gitdiff_newmode(const char *line, struct patch *patch)
991
{
992
	patch->new_mode = strtoul(line, NULL, 8);
993 994 995
	return 0;
}

996
static int gitdiff_delete(const char *line, struct patch *patch)
997
{
998
	patch->is_delete = 1;
999 1000
	free(patch->old_name);
	patch->old_name = null_strdup(patch->def_name);
1001
	return gitdiff_oldmode(line, patch);
1002 1003
}

1004
static int gitdiff_newfile(const char *line, struct patch *patch)
1005
{
1006
	patch->is_new = 1;
1007 1008
	free(patch->new_name);
	patch->new_name = null_strdup(patch->def_name);
1009
	return gitdiff_newmode(line, patch);
1010 1011
}

1012
static int gitdiff_copysrc(const char *line, struct patch *patch)
1013
{
1014
	patch->is_copy = 1;
1015
	free(patch->old_name);
1016
	patch->old_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1017 1018 1019
	return 0;
}

1020
static int gitdiff_copydst(const char *line, struct patch *patch)
1021
{
1022
	patch->is_copy = 1;
1023
	free(patch->new_name);
1024
	patch->new_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1025 1026 1027
	return 0;
}

1028
static int gitdiff_renamesrc(const char *line, struct patch *patch)
1029
{
1030
	patch->is_rename = 1;
1031
	free(patch->old_name);
1032
	patch->old_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1033 1034 1035
	return 0;
}

1036
static int gitdiff_renamedst(const char *line, struct patch *patch)
1037
{
1038
	patch->is_rename = 1;
1039
	free(patch->new_name);
1040
	patch->new_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1041 1042 1043
	return 0;
}

1044
static int gitdiff_similarity(const char *line, struct patch *patch)
1045
{
1046 1047 1048
	unsigned long val = strtoul(line, NULL, 10);
	if (val <= 100)
		patch->score = val;
1049
	return 0;
L
Linus Torvalds 已提交
1050 1051
}

1052 1053
static int gitdiff_dissimilarity(const char *line, struct patch *patch)
{
1054 1055 1056
	unsigned long val = strtoul(line, NULL, 10);
	if (val <= 100)
		patch->score = val;
1057 1058 1059
	return 0;
}

1060 1061
static int gitdiff_index(const char *line, struct patch *patch)
{
1062 1063
	/*
	 * index line is N hexadecimal, "..", N hexadecimal,
1064 1065 1066 1067 1068 1069
	 * and optional space with octal mode.
	 */
	const char *ptr, *eol;
	int len;

	ptr = strchr(line, '.');
1070
	if (!ptr || ptr[1] != '.' || 40 < ptr - line)
1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083
		return 0;
	len = ptr - line;
	memcpy(patch->old_sha1_prefix, line, len);
	patch->old_sha1_prefix[len] = 0;

	line = ptr + 2;
	ptr = strchr(line, ' ');
	eol = strchr(line, '\n');

	if (!ptr || eol < ptr)
		ptr = eol;
	len = ptr - line;

1084
	if (40 < len)
1085 1086 1087 1088
		return 0;
	memcpy(patch->new_sha1_prefix, line, len);
	patch->new_sha1_prefix[len] = 0;
	if (*ptr == ' ')
1089
		patch->old_mode = strtoul(ptr+1, NULL, 8);
1090 1091 1092
	return 0;
}

1093 1094 1095 1096
/*
 * This is normal for a diff that doesn't change anything: we'll fall through
 * into the next diff. Tell the parser to break out.
 */
1097
static int gitdiff_unrecognized(const char *line, struct patch *patch)
1098 1099 1100 1101
{
	return -1;
}

1102 1103 1104 1105 1106
/*
 * Skip p_value leading components from "line"; as we do not accept
 * absolute paths, return NULL in that case.
 */
static const char *skip_tree_prefix(const char *line, int llen)
1107
{
1108
	int nslash;
1109 1110
	int i;

1111 1112 1113 1114
	if (!p_value)
		return (llen && line[0] == '/') ? NULL : line;

	nslash = p_value;
1115 1116
	for (i = 0; i < llen; i++) {
		int ch = line[i];
1117
		if (ch == '/' && --nslash <= 0)
1118
			return (i == 0) ? NULL : &line[i + 1];
1119 1120 1121 1122
	}
	return NULL;
}

1123 1124
/*
 * This is to extract the same name that appears on "diff --git"
1125 1126 1127 1128 1129 1130
 * line.  We do not find and return anything if it is a rename
 * patch, and it is OK because we will find the name elsewhere.
 * We need to reliably find name only when it is mode-change only,
 * creation or deletion of an empty file.  In any of these cases,
 * both sides are the same name under a/ and b/ respectively.
 */
1131
static char *git_header_name(const char *line, int llen)
1132
{
1133 1134
	const char *name;
	const char *second = NULL;
1135
	size_t len, line_len;
1136

1137 1138 1139 1140 1141
	line += strlen("diff --git ");
	llen -= strlen("diff --git ");

	if (*line == '"') {
		const char *cp;
1142 1143
		struct strbuf first = STRBUF_INIT;
		struct strbuf sp = STRBUF_INIT;
1144 1145 1146

		if (unquote_c_style(&first, line, &second))
			goto free_and_fail1;
1147

1148 1149 1150
		/* strip the a/b prefix including trailing slash */
		cp = skip_tree_prefix(first.buf, first.len);
		if (!cp)
1151
			goto free_and_fail1;
1152
		strbuf_remove(&first, 0, cp - first.buf);
1153

1154 1155
		/*
		 * second points at one past closing dq of name.
1156 1157 1158 1159 1160 1161
		 * find the second name.
		 */
		while ((second < line + llen) && isspace(*second))
			second++;

		if (line + llen <= second)
1162
			goto free_and_fail1;
1163
		if (*second == '"') {
1164 1165
			if (unquote_c_style(&sp, second, NULL))
				goto free_and_fail1;
1166 1167
			cp = skip_tree_prefix(sp.buf, sp.len);
			if (!cp)
1168
				goto free_and_fail1;
1169
			/* They must match, otherwise ignore */
1170
			if (strcmp(cp, first.buf))
1171 1172
				goto free_and_fail1;
			strbuf_release(&sp);
1173
			return strbuf_detach(&first, NULL);
1174 1175 1176
		}

		/* unquoted second */
1177 1178
		cp = skip_tree_prefix(second, line + llen - second);
		if (!cp)
1179
			goto free_and_fail1;
1180
		if (line + llen - cp != first.len ||
1181 1182
		    memcmp(first.buf, cp, first.len))
			goto free_and_fail1;
1183
		return strbuf_detach(&first, NULL);
1184 1185 1186 1187 1188

	free_and_fail1:
		strbuf_release(&first);
		strbuf_release(&sp);
		return NULL;
1189 1190
	}

1191
	/* unquoted first name */
1192 1193
	name = skip_tree_prefix(line, llen);
	if (!name)
1194
		return NULL;
1195

1196 1197
	/*
	 * since the first name is unquoted, a dq if exists must be
1198 1199 1200 1201
	 * the beginning of the second name.
	 */
	for (second = name; second < line + llen; second++) {
		if (*second == '"') {
1202
			struct strbuf sp = STRBUF_INIT;
1203
			const char *np;
1204 1205 1206 1207

			if (unquote_c_style(&sp, second, NULL))
				goto free_and_fail2;

1208 1209
			np = skip_tree_prefix(sp.buf, sp.len);
			if (!np)
1210 1211 1212 1213
				goto free_and_fail2;

			len = sp.buf + sp.len - np;
			if (len < second - name &&
1214 1215 1216
			    !strncmp(np, name, len) &&
			    isspace(name[len])) {
				/* Good */
1217
				strbuf_remove(&sp, 0, np - sp.buf);
1218
				return strbuf_detach(&sp, NULL);
1219
			}
1220 1221 1222 1223

		free_and_fail2:
			strbuf_release(&sp);
			return NULL;
1224 1225 1226
		}
	}

1227 1228 1229 1230
	/*
	 * Accept a name only if it shows up twice, exactly the same
	 * form.
	 */
1231 1232 1233 1234
	second = strchr(name, '\n');
	if (!second)
		return NULL;
	line_len = second - name;
1235
	for (len = 0 ; ; len++) {
1236
		switch (name[len]) {
1237 1238 1239
		default:
			continue;
		case '\n':
1240
			return NULL;
1241
		case '\t': case ' ':
1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253
			/*
			 * Is this the separator between the preimage
			 * and the postimage pathname?  Again, we are
			 * only interested in the case where there is
			 * no rename, as this is only to set def_name
			 * and a rename patch has the names elsewhere
			 * in an unambiguous form.
			 */
			if (!name[len + 1])
				return NULL; /* no postimage name */
			second = skip_tree_prefix(name + len + 1,
						  line_len - (len + 1));
1254 1255
			if (!second)
				return NULL;
1256 1257 1258 1259 1260 1261
			/*
			 * Does len bytes starting at "name" and "second"
			 * (that are separated by one HT or SP we just
			 * found) exactly match?
			 */
			if (second[len] == '\n' && !strncmp(name, second, len))
P
Pierre Habouzit 已提交
1262
				return xmemdupz(name, len);
1263 1264 1265 1266
		}
	}
}

L
Linus Torvalds 已提交
1267
/* Verify that we recognize the lines following a git header */
1268
static int parse_git_header(const char *line, int len, unsigned int size, struct patch *patch)
L
Linus Torvalds 已提交
1269
{
1270 1271 1272
	unsigned long offset;

	/* A git diff has explicit new/delete information, so we don't guess */
1273 1274
	patch->is_new = 0;
	patch->is_delete = 0;
1275

1276 1277 1278 1279 1280 1281
	/*
	 * Some things may not have the old name in the
	 * rest of the headers anywhere (pure mode changes,
	 * or removing or adding empty files), so we get
	 * the default name from the header.
	 */
1282
	patch->def_name = git_header_name(line, len);
1283 1284 1285 1286 1287 1288 1289
	if (patch->def_name && root) {
		char *s = xmalloc(root_len + strlen(patch->def_name) + 1);
		strcpy(s, root);
		strcpy(s + root_len, patch->def_name);
		free(patch->def_name);
		patch->def_name = s;
	}
1290

1291 1292 1293 1294 1295 1296
	line += len;
	size -= len;
	linenr++;
	for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
		static const struct opentry {
			const char *str;
1297
			int (*fn)(const char *, struct patch *);
1298 1299 1300 1301 1302 1303 1304 1305 1306 1307
		} optable[] = {
			{ "@@ -", gitdiff_hdrend },
			{ "--- ", gitdiff_oldname },
			{ "+++ ", gitdiff_newname },
			{ "old mode ", gitdiff_oldmode },
			{ "new mode ", gitdiff_newmode },
			{ "deleted file mode ", gitdiff_delete },
			{ "new file mode ", gitdiff_newfile },
			{ "copy from ", gitdiff_copysrc },
			{ "copy to ", gitdiff_copydst },
1308 1309
			{ "rename old ", gitdiff_renamesrc },
			{ "rename new ", gitdiff_renamedst },
L
Linus Torvalds 已提交
1310 1311
			{ "rename from ", gitdiff_renamesrc },
			{ "rename to ", gitdiff_renamedst },
1312
			{ "similarity index ", gitdiff_similarity },
1313
			{ "dissimilarity index ", gitdiff_dissimilarity },
1314
			{ "index ", gitdiff_index },
1315
			{ "", gitdiff_unrecognized },
1316 1317
		};
		int i;
L
Linus Torvalds 已提交
1318 1319

		len = linelen(line, size);
1320
		if (!len || line[len-1] != '\n')
L
Linus Torvalds 已提交
1321
			break;
1322
		for (i = 0; i < ARRAY_SIZE(optable); i++) {
1323 1324 1325 1326
			const struct opentry *p = optable + i;
			int oplen = strlen(p->str);
			if (len < oplen || memcmp(p->str, line, oplen))
				continue;
1327
			if (p->fn(line + oplen, patch) < 0)
1328
				return offset;
1329
			break;
1330
		}
L
Linus Torvalds 已提交
1331 1332
	}

1333
	return offset;
L
Linus Torvalds 已提交
1334 1335
}

1336
static int parse_num(const char *line, unsigned long *p)
1337 1338
{
	char *ptr;
1339 1340 1341 1342 1343 1344 1345 1346

	if (!isdigit(*line))
		return 0;
	*p = strtoul(line, &ptr, 10);
	return ptr - line;
}

static int parse_range(const char *line, int len, int offset, const char *expect,
1347
		       unsigned long *p1, unsigned long *p2)
1348
{
1349 1350 1351 1352 1353 1354 1355
	int digits, ex;

	if (offset < 0 || offset >= len)
		return -1;
	line += offset;
	len -= offset;

1356 1357
	digits = parse_num(line, p1);
	if (!digits)
1358 1359 1360 1361 1362 1363
		return -1;

	offset += digits;
	line += digits;
	len -= digits;

L
Linus Torvalds 已提交
1364
	*p2 = 1;
1365 1366 1367 1368 1369 1370 1371 1372 1373 1374
	if (*line == ',') {
		digits = parse_num(line+1, p2);
		if (!digits)
			return -1;

		offset += digits+1;
		line += digits+1;
		len -= digits+1;
	}

1375 1376 1377 1378 1379 1380 1381 1382 1383
	ex = strlen(expect);
	if (ex > len)
		return -1;
	if (memcmp(line, expect, ex))
		return -1;

	return offset + ex;
}

1384
static void recount_diff(const char *line, int size, struct fragment *fragment)
1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411
{
	int oldlines = 0, newlines = 0, ret = 0;

	if (size < 1) {
		warning("recount: ignore empty hunk");
		return;
	}

	for (;;) {
		int len = linelen(line, size);
		size -= len;
		line += len;

		if (size < 1)
			break;

		switch (*line) {
		case ' ': case '\n':
			newlines++;
			/* fall through */
		case '-':
			oldlines++;
			continue;
		case '+':
			newlines++;
			continue;
		case '\\':
1412
			continue;
1413
		case '@':
1414
			ret = size < 3 || !starts_with(line, "@@ ");
1415 1416
			break;
		case 'd':
1417
			ret = size < 5 || !starts_with(line, "diff ");
1418 1419 1420 1421 1422 1423
			break;
		default:
			ret = -1;
			break;
		}
		if (ret) {
1424
			warning(_("recount: unexpected line: %.*s"),
1425 1426 1427 1428 1429 1430 1431 1432 1433
				(int)linelen(line, size), line);
			return;
		}
		break;
	}
	fragment->oldlines = oldlines;
	fragment->newlines = newlines;
}

1434 1435 1436 1437
/*
 * Parse a unified diff fragment header of the
 * form "@@ -a,b +c,d @@"
 */
1438
static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1439 1440 1441 1442 1443 1444 1445
{
	int offset;

	if (!len || line[len-1] != '\n')
		return -1;

	/* Figure out the number of lines in a fragment */
1446 1447
	offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
	offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1448 1449 1450 1451

	return offset;
}

1452
static int find_header(const char *line, unsigned long size, int *hdrsize, struct patch *patch)
L
Linus Torvalds 已提交
1453 1454 1455
{
	unsigned long offset, len;

1456
	patch->is_toplevel_relative = 0;
1457 1458 1459 1460
	patch->is_rename = patch->is_copy = 0;
	patch->is_new = patch->is_delete = -1;
	patch->old_mode = patch->new_mode = 0;
	patch->old_name = patch->new_name = NULL;
1461
	for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
L
Linus Torvalds 已提交
1462 1463 1464 1465 1466 1467 1468 1469 1470
		unsigned long nextlen;

		len = linelen(line, size);
		if (!len)
			break;

		/* Testing this early allows us to take a few shortcuts.. */
		if (len < 6)
			continue;
1471 1472

		/*
1473
		 * Make sure we don't find any unconnected patch fragments.
1474 1475 1476 1477
		 * That's a sign that we didn't find a header, and that a
		 * patch has become corrupted/broken up.
		 */
		if (!memcmp("@@ -", line, 4)) {
1478 1479
			struct fragment dummy;
			if (parse_fragment_header(line, len, &dummy) < 0)
1480
				continue;
1481
			die(_("patch fragment without header at line %d: %.*s"),
1482
			    linenr, (int)len-1, line);
1483 1484
		}

L
Linus Torvalds 已提交
1485 1486 1487 1488 1489 1490 1491 1492
		if (size < len + 6)
			break;

		/*
		 * Git patch? It might not have a real patch, just a rename
		 * or mode change, so we handle that specially
		 */
		if (!memcmp("diff --git ", line, 11)) {
1493
			int git_hdr_len = parse_git_header(line, len, size, patch);
1494
			if (git_hdr_len <= len)
L
Linus Torvalds 已提交
1495
				continue;
1496 1497
			if (!patch->old_name && !patch->new_name) {
				if (!patch->def_name)
1498 1499 1500 1501 1502 1503
					die(Q_("git diff header lacks filename information when removing "
					       "%d leading pathname component (line %d)",
					       "git diff header lacks filename information when removing "
					       "%d leading pathname components (line %d)",
					       p_value),
					    p_value, linenr);
1504 1505
				patch->old_name = xstrdup(patch->def_name);
				patch->new_name = xstrdup(patch->def_name);
1506
			}
1507 1508 1509
			if (!patch->is_delete && !patch->new_name)
				die("git diff header lacks filename information "
				    "(line %d)", linenr);
1510
			patch->is_toplevel_relative = 1;
1511
			*hdrsize = git_hdr_len;
L
Linus Torvalds 已提交
1512 1513 1514
			return offset;
		}

1515
		/* --- followed by +++ ? */
L
Linus Torvalds 已提交
1516 1517 1518 1519 1520 1521
		if (memcmp("--- ", line,  4) || memcmp("+++ ", line + len, 4))
			continue;

		/*
		 * We only accept unified patches, so we want it to
		 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1522
		 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
L
Linus Torvalds 已提交
1523 1524 1525 1526 1527 1528
		 */
		nextlen = linelen(line + len, size - len);
		if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
			continue;

		/* Ok, we'll consider it a patch */
1529
		parse_traditional_patch(line, line+len, patch);
L
Linus Torvalds 已提交
1530
		*hdrsize = len + nextlen;
1531
		linenr += 2;
L
Linus Torvalds 已提交
1532 1533 1534 1535 1536
		return offset;
	}
	return -1;
}

1537
static void record_ws_error(unsigned result, const char *line, int len, int linenr)
1538
{
W
Wincent Colaiuta 已提交
1539
	char *err;
1540

W
Wincent Colaiuta 已提交
1541 1542
	if (!result)
		return;
1543 1544 1545 1546

	whitespace_error++;
	if (squelch_whitespace_errors &&
	    squelch_whitespace_errors < whitespace_error)
1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559
		return;

	err = whitespace_error_string(result);
	fprintf(stderr, "%s:%d: %s.\n%.*s\n",
		patch_input_file, linenr, err, len, line);
	free(err);
}

static void check_whitespace(const char *line, int len, unsigned ws_rule)
{
	unsigned result = ws_check(line + 1, len - 1, ws_rule);

	record_ws_error(result, line + 1, len - 2, linenr);
1560 1561
}

L
Linus Torvalds 已提交
1562
/*
1563 1564 1565 1566
 * Parse a unified diff. Note that this really needs to parse each
 * fragment separately, since the only way to know the difference
 * between a "---" that is part of a patch, and a "---" that starts
 * the next patch is to look at the line counts..
L
Linus Torvalds 已提交
1567
 */
1568
static int parse_fragment(const char *line, unsigned long size,
1569
			  struct patch *patch, struct fragment *fragment)
L
Linus Torvalds 已提交
1570
{
1571
	int added, deleted;
L
Linus Torvalds 已提交
1572
	int len = linelen(line, size), offset;
1573
	unsigned long oldlines, newlines;
1574
	unsigned long leading, trailing;
L
Linus Torvalds 已提交
1575

1576
	offset = parse_fragment_header(line, len, fragment);
L
Linus Torvalds 已提交
1577 1578
	if (offset < 0)
		return -1;
1579 1580
	if (offset > 0 && patch->recount)
		recount_diff(line + offset, size - offset, fragment);
1581 1582
	oldlines = fragment->oldlines;
	newlines = fragment->newlines;
1583 1584
	leading = 0;
	trailing = 0;
L
Linus Torvalds 已提交
1585 1586 1587 1588

	/* Parse the thing.. */
	line += len;
	size -= len;
1589
	linenr++;
1590
	added = deleted = 0;
1591 1592 1593
	for (offset = len;
	     0 < size;
	     offset += len, size -= len, line += len, linenr++) {
L
Linus Torvalds 已提交
1594 1595 1596 1597 1598 1599 1600 1601
		if (!oldlines && !newlines)
			break;
		len = linelen(line, size);
		if (!len || line[len-1] != '\n')
			return -1;
		switch (*line) {
		default:
			return -1;
1602
		case '\n': /* newer GNU diff, an empty context line */
L
Linus Torvalds 已提交
1603 1604 1605
		case ' ':
			oldlines--;
			newlines--;
1606 1607 1608
			if (!deleted && !added)
				leading++;
			trailing++;
L
Linus Torvalds 已提交
1609 1610
			break;
		case '-':
1611
			if (apply_in_reverse &&
1612
			    ws_error_action != nowarn_ws_error)
1613
				check_whitespace(line, len, patch->ws_rule);
1614
			deleted++;
L
Linus Torvalds 已提交
1615
			oldlines--;
1616
			trailing = 0;
L
Linus Torvalds 已提交
1617 1618
			break;
		case '+':
1619
			if (!apply_in_reverse &&
1620
			    ws_error_action != nowarn_ws_error)
1621
				check_whitespace(line, len, patch->ws_rule);
1622
			added++;
L
Linus Torvalds 已提交
1623
			newlines--;
1624
			trailing = 0;
L
Linus Torvalds 已提交
1625
			break;
1626

1627 1628
		/*
		 * We allow "\ No newline at end of file". Depending
1629 1630
                 * on locale settings when the patch was produced we
                 * don't know what this line looks like. The only
1631 1632 1633 1634
                 * thing we do know is that it begins with "\ ".
		 * Checking for 12 is just for sanity check -- any
		 * l10n of "\ No newline..." is at least that long.
		 */
1635
		case '\\':
1636
			if (len < 12 || memcmp(line, "\\ ", 2))
1637
				return -1;
1638
			break;
L
Linus Torvalds 已提交
1639 1640
		}
	}
L
Linus Torvalds 已提交
1641 1642
	if (oldlines || newlines)
		return -1;
1643 1644 1645
	fragment->leading = leading;
	fragment->trailing = trailing;

1646 1647
	/*
	 * If a fragment ends with an incomplete line, we failed to include
1648 1649 1650
	 * it in the above loop because we hit oldlines == newlines == 0
	 * before seeing it.
	 */
1651
	if (12 < size && !memcmp(line, "\\ ", 2))
1652 1653
		offset += linelen(line, size);

1654 1655
	patch->lines_added += added;
	patch->lines_deleted += deleted;
1656 1657

	if (0 < patch->is_new && oldlines)
1658
		return error(_("new file depends on old contents"));
1659
	if (0 < patch->is_delete && newlines)
1660
		return error(_("deleted file still has contents"));
L
Linus Torvalds 已提交
1661 1662 1663
	return offset;
}

1664 1665 1666 1667 1668 1669 1670 1671
/*
 * We have seen "diff --git a/... b/..." header (or a traditional patch
 * header).  Read hunks that belong to this patch into fragments and hang
 * them to the given patch structure.
 *
 * The (fragment->patch, fragment->size) pair points into the memory given
 * by the caller, not a copy, when we return.
 */
1672
static int parse_single_patch(const char *line, unsigned long size, struct patch *patch)
L
Linus Torvalds 已提交
1673 1674
{
	unsigned long offset = 0;
1675
	unsigned long oldlines = 0, newlines = 0, context = 0;
1676
	struct fragment **fragp = &patch->fragments;
L
Linus Torvalds 已提交
1677 1678

	while (size > 4 && !memcmp(line, "@@ -", 4)) {
1679 1680 1681
		struct fragment *fragment;
		int len;

1682
		fragment = xcalloc(1, sizeof(*fragment));
1683
		fragment->linenr = linenr;
1684
		len = parse_fragment(line, size, patch, fragment);
L
Linus Torvalds 已提交
1685
		if (len <= 0)
1686
			die(_("corrupt patch at line %d"), linenr);
1687 1688
		fragment->patch = line;
		fragment->size = len;
1689 1690 1691
		oldlines += fragment->oldlines;
		newlines += fragment->newlines;
		context += fragment->leading + fragment->trailing;
1692 1693 1694

		*fragp = fragment;
		fragp = &fragment->next;
L
Linus Torvalds 已提交
1695 1696 1697 1698 1699

		offset += len;
		line += len;
		size -= len;
	}
1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721

	/*
	 * If something was removed (i.e. we have old-lines) it cannot
	 * be creation, and if something was added it cannot be
	 * deletion.  However, the reverse is not true; --unified=0
	 * patches that only add are not necessarily creation even
	 * though they do not have any old lines, and ones that only
	 * delete are not necessarily deletion.
	 *
	 * Unfortunately, a real creation/deletion patch do _not_ have
	 * any context line by definition, so we cannot safely tell it
	 * apart with --unified=0 insanity.  At least if the patch has
	 * more than one hunk it is not creation or deletion.
	 */
	if (patch->is_new < 0 &&
	    (oldlines || (patch->fragments && patch->fragments->next)))
		patch->is_new = 0;
	if (patch->is_delete < 0 &&
	    (newlines || (patch->fragments && patch->fragments->next)))
		patch->is_delete = 0;

	if (0 < patch->is_new && oldlines)
1722
		die(_("new file %s depends on old contents"), patch->new_name);
1723
	if (0 < patch->is_delete && newlines)
1724
		die(_("deleted file %s still has contents"), patch->old_name);
1725
	if (!patch->is_delete && !newlines && context)
1726 1727 1728 1729
		fprintf_ln(stderr,
			   _("** warning: "
			     "file %s becomes empty but is not deleted"),
			   patch->new_name);
1730

L
Linus Torvalds 已提交
1731 1732 1733
	return offset;
}

1734 1735 1736 1737 1738 1739 1740 1741 1742 1743
static inline int metadata_changes(struct patch *patch)
{
	return	patch->is_rename > 0 ||
		patch->is_copy > 0 ||
		patch->is_new > 0 ||
		patch->is_delete ||
		(patch->old_mode && patch->new_mode &&
		 patch->old_mode != patch->new_mode);
}

1744 1745
static char *inflate_it(const void *data, unsigned long size,
			unsigned long inflated_size)
J
Junio C Hamano 已提交
1746
{
1747
	git_zstream stream;
1748 1749 1750 1751 1752 1753 1754 1755 1756
	void *out;
	int st;

	memset(&stream, 0, sizeof(stream));

	stream.next_in = (unsigned char *)data;
	stream.avail_in = size;
	stream.next_out = out = xmalloc(inflated_size);
	stream.avail_out = inflated_size;
1757 1758 1759
	git_inflate_init(&stream);
	st = git_inflate(&stream, Z_FINISH);
	git_inflate_end(&stream);
1760 1761 1762 1763 1764 1765 1766
	if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
		free(out);
		return NULL;
	}
	return out;
}

1767 1768 1769 1770 1771
/*
 * Read a binary hunk and return a new fragment; fragment->patch
 * points at an allocated memory that the caller must free, so
 * it is marked as "->free_patch = 1".
 */
1772 1773 1774 1775 1776
static struct fragment *parse_binary_hunk(char **buf_p,
					  unsigned long *sz_p,
					  int *status_p,
					  int *used_p)
{
1777 1778
	/*
	 * Expect a line that begins with binary patch method ("literal"
1779 1780 1781
	 * or "delta"), followed by the length of data before deflating.
	 * a sequence of 'length-byte' followed by base-85 encoded data
	 * should follow, terminated by a newline.
J
Junio C Hamano 已提交
1782 1783 1784 1785 1786 1787 1788 1789
	 *
	 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
	 * and we would limit the patch line to 66 characters,
	 * so one line can fit up to 13 groups that would decode
	 * to 52 bytes max.  The length byte 'A'-'Z' corresponds
	 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
	 */
	int llen, used;
1790 1791 1792 1793
	unsigned long size = *sz_p;
	char *buffer = *buf_p;
	int patch_method;
	unsigned long origlen;
J
Junio C Hamano 已提交
1794
	char *data = NULL;
1795 1796
	int hunk_size = 0;
	struct fragment *frag;
J
Junio C Hamano 已提交
1797

J
Junio C Hamano 已提交
1798 1799
	llen = linelen(buffer, size);
	used = llen;
1800 1801

	*status_p = 0;
J
Junio C Hamano 已提交
1802

1803
	if (starts_with(buffer, "delta ")) {
1804 1805
		patch_method = BINARY_DELTA_DEFLATED;
		origlen = strtoul(buffer + 6, NULL, 10);
J
Junio C Hamano 已提交
1806
	}
1807
	else if (starts_with(buffer, "literal ")) {
1808 1809
		patch_method = BINARY_LITERAL_DEFLATED;
		origlen = strtoul(buffer + 8, NULL, 10);
J
Junio C Hamano 已提交
1810 1811
	}
	else
1812 1813 1814
		return NULL;

	linenr++;
J
Junio C Hamano 已提交
1815
	buffer += llen;
J
Junio C Hamano 已提交
1816 1817 1818 1819 1820
	while (1) {
		int byte_length, max_byte_length, newsize;
		llen = linelen(buffer, size);
		used += llen;
		linenr++;
1821 1822 1823 1824
		if (llen == 1) {
			/* consume the blank line */
			buffer++;
			size--;
J
Junio C Hamano 已提交
1825
			break;
1826
		}
1827 1828
		/*
		 * Minimum line is "A00000\n" which is 7-byte long,
J
Junio C Hamano 已提交
1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847
		 * and the line length must be multiple of 5 plus 2.
		 */
		if ((llen < 7) || (llen-2) % 5)
			goto corrupt;
		max_byte_length = (llen - 2) / 5 * 4;
		byte_length = *buffer;
		if ('A' <= byte_length && byte_length <= 'Z')
			byte_length = byte_length - 'A' + 1;
		else if ('a' <= byte_length && byte_length <= 'z')
			byte_length = byte_length - 'a' + 27;
		else
			goto corrupt;
		/* if the input length was not multiple of 4, we would
		 * have filler at the end but the filler should never
		 * exceed 3 bytes
		 */
		if (max_byte_length < byte_length ||
		    byte_length <= max_byte_length - 4)
			goto corrupt;
1848
		newsize = hunk_size + byte_length;
J
Junio C Hamano 已提交
1849
		data = xrealloc(data, newsize);
1850
		if (decode_85(data + hunk_size, buffer + 1, byte_length))
J
Junio C Hamano 已提交
1851
			goto corrupt;
1852
		hunk_size = newsize;
J
Junio C Hamano 已提交
1853 1854 1855
		buffer += llen;
		size -= llen;
	}
1856 1857 1858

	frag = xcalloc(1, sizeof(*frag));
	frag->patch = inflate_it(data, hunk_size, origlen);
1859
	frag->free_patch = 1;
1860 1861 1862 1863 1864 1865 1866 1867 1868 1869
	if (!frag->patch)
		goto corrupt;
	free(data);
	frag->size = origlen;
	*buf_p = buffer;
	*sz_p = size;
	*used_p = used;
	frag->binary_patch_method = patch_method;
	return frag;

J
Junio C Hamano 已提交
1870
 corrupt:
J
Junio C Hamano 已提交
1871
	free(data);
1872
	*status_p = -1;
1873
	error(_("corrupt binary patch at line %d: %.*s"),
1874 1875 1876 1877 1878 1879
	      linenr-1, llen-1, buffer);
	return NULL;
}

static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
{
1880 1881
	/*
	 * We have read "GIT binary patch\n"; what follows is a line
1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901
	 * that says the patch method (currently, either "literal" or
	 * "delta") and the length of data before deflating; a
	 * sequence of 'length-byte' followed by base-85 encoded data
	 * follows.
	 *
	 * When a binary patch is reversible, there is another binary
	 * hunk in the same format, starting with patch method (either
	 * "literal" or "delta") with the length of data, and a sequence
	 * of length-byte + base-85 encoded data, terminated with another
	 * empty line.  This data, when applied to the postimage, produces
	 * the preimage.
	 */
	struct fragment *forward;
	struct fragment *reverse;
	int status;
	int used, used_1;

	forward = parse_binary_hunk(&buffer, &size, &status, &used);
	if (!forward && !status)
		/* there has to be one hunk (forward hunk) */
1902
		return error(_("unrecognized binary patch at line %d"), linenr-1);
1903 1904 1905 1906 1907 1908 1909 1910
	if (status)
		/* otherwise we already gave an error message */
		return status;

	reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
	if (reverse)
		used += used_1;
	else if (status) {
1911 1912
		/*
		 * Not having reverse hunk is not an error, but having
1913 1914 1915 1916 1917 1918 1919 1920 1921 1922
		 * a corrupt reverse hunk is.
		 */
		free((void*) forward->patch);
		free(forward);
		return status;
	}
	forward->next = reverse;
	patch->fragments = forward;
	patch->is_binary = 1;
	return used;
J
Junio C Hamano 已提交
1923 1924
}

1925
/*
1926
 * Read the patch text in "buffer" that extends for "size" bytes; stop
1927 1928 1929 1930 1931
 * reading after seeing a single patch (i.e. changes to a single file).
 * Create fragments (i.e. patch hunks) and hang them to the given patch.
 * Return the number of bytes consumed, so that the caller can call us
 * again for the next patch.
 */
1932
static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
L
Linus Torvalds 已提交
1933 1934
{
	int hdrsize, patchsize;
1935
	int offset = find_header(buffer, size, &hdrsize, patch);
L
Linus Torvalds 已提交
1936 1937 1938 1939

	if (offset < 0)
		return offset;

1940 1941 1942 1943
	patch->ws_rule = whitespace_rule(patch->new_name
					 ? patch->new_name
					 : patch->old_name);

1944 1945
	patchsize = parse_single_patch(buffer + offset + hdrsize,
				       size - offset - hdrsize, patch);
L
Linus Torvalds 已提交
1946

1947
	if (!patchsize) {
J
Junio C Hamano 已提交
1948
		static const char git_binary[] = "GIT binary patch\n";
1949 1950 1951
		int hd = hdrsize + offset;
		unsigned long llen = linelen(buffer + hd, size - hd);

J
Junio C Hamano 已提交
1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963
		if (llen == sizeof(git_binary) - 1 &&
		    !memcmp(git_binary, buffer + hd, llen)) {
			int used;
			linenr++;
			used = parse_binary(buffer + hd + llen,
					    size - hd - llen, patch);
			if (used)
				patchsize = used + llen;
			else
				patchsize = 0;
		}
		else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1964 1965 1966 1967 1968 1969
			static const char *binhdr[] = {
				"Binary files ",
				"Files ",
				NULL,
			};
			int i;
1970 1971 1972 1973
			for (i = 0; binhdr[i]; i++) {
				int len = strlen(binhdr[i]);
				if (len < size - hd &&
				    !memcmp(binhdr[i], buffer + hd, len)) {
J
Junio C Hamano 已提交
1974
					linenr++;
1975
					patch->is_binary = 1;
J
Junio C Hamano 已提交
1976
					patchsize = llen;
1977 1978 1979
					break;
				}
			}
J
Junio C Hamano 已提交
1980
		}
1981

J
Junio C Hamano 已提交
1982 1983 1984
		/* Empty patch cannot be applied if it is a text patch
		 * without metadata change.  A binary patch appears
		 * empty to us here.
1985 1986
		 */
		if ((apply || check) &&
J
Junio C Hamano 已提交
1987
		    (!patch->is_binary && !metadata_changes(patch)))
1988
			die(_("patch with only garbage at line %d"), linenr);
1989
	}
1990

L
Linus Torvalds 已提交
1991 1992 1993
	return offset + hdrsize + patchsize;
}

J
Johannes Schindelin 已提交
1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020
#define swap(a,b) myswap((a),(b),sizeof(a))

#define myswap(a, b, size) do {		\
	unsigned char mytmp[size];	\
	memcpy(mytmp, &a, size);		\
	memcpy(&a, &b, size);		\
	memcpy(&b, mytmp, size);		\
} while (0)

static void reverse_patches(struct patch *p)
{
	for (; p; p = p->next) {
		struct fragment *frag = p->fragments;

		swap(p->new_name, p->old_name);
		swap(p->new_mode, p->old_mode);
		swap(p->is_new, p->is_delete);
		swap(p->lines_added, p->lines_deleted);
		swap(p->old_sha1_prefix, p->new_sha1_prefix);

		for (; frag; frag = frag->next) {
			swap(frag->newpos, frag->oldpos);
			swap(frag->newlines, frag->oldlines);
		}
	}
}

2021 2022 2023 2024
static const char pluses[] =
"++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
static const char minuses[]=
"----------------------------------------------------------------------";
2025 2026 2027

static void show_stats(struct patch *patch)
{
2028
	struct strbuf qname = STRBUF_INIT;
2029 2030
	char *cp = patch->new_name ? patch->new_name : patch->old_name;
	int max, add, del;
2031

2032
	quote_c_style(cp, &qname, NULL, 0);
2033

2034 2035 2036 2037 2038 2039
	/*
	 * "scale" the filename
	 */
	max = max_len;
	if (max > 50)
		max = 50;
2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051

	if (qname.len > max) {
		cp = strchr(qname.buf + qname.len + 3 - max, '/');
		if (!cp)
			cp = qname.buf + qname.len + 3 - max;
		strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
	}

	if (patch->is_binary) {
		printf(" %-*s |  Bin\n", max, qname.buf);
		strbuf_release(&qname);
		return;
2052
	}
2053 2054 2055

	printf(" %-*s |", max, qname.buf);
	strbuf_release(&qname);
2056 2057 2058 2059

	/*
	 * scale the add/delete
	 */
2060
	max = max + max_change > 70 ? 70 - max : max_change;
2061 2062 2063
	add = patch->lines_added;
	del = patch->lines_deleted;

2064
	if (max_change > 0) {
2065
		int total = ((add + del) * max + max_change / 2) / max_change;
2066 2067 2068
		add = (add * max + max_change / 2) / max_change;
		del = total - add;
	}
2069 2070
	printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
		add, pluses, del, minuses);
2071 2072
}

2073
static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
2074 2075 2076
{
	switch (st->st_mode & S_IFMT) {
	case S_IFLNK:
2077
		if (strbuf_readlink(buf, path, st->st_size) < 0)
2078
			return error(_("unable to read symlink %s"), path);
2079
		return 0;
2080
	case S_IFREG:
2081
		if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2082
			return error(_("unable to open or read %s"), path);
2083
		convert_to_git(path, buf->buf, buf->len, buf, 0);
2084
		return 0;
2085 2086 2087 2088 2089
	default:
		return -1;
	}
}

2090 2091 2092 2093 2094 2095 2096
/*
 * Update the preimage, and the common lines in postimage,
 * from buffer buf of length len. If postlen is 0 the postimage
 * is updated in place, otherwise it's updated on a new buffer
 * of length postlen
 */

2097 2098 2099
static void update_pre_post_images(struct image *preimage,
				   struct image *postimage,
				   char *buf,
2100
				   size_t len, size_t postlen)
2101
{
2102
	int i, ctx, reduced;
2103 2104
	char *new, *old, *fixed;
	struct image fixed_preimage;
2105

2106 2107 2108 2109 2110 2111
	/*
	 * Update the preimage with whitespace fixes.  Note that we
	 * are not losing preimage->buf -- apply_one_fragment() will
	 * free "oldlines".
	 */
	prepare_image(&fixed_preimage, buf, len, 1);
2112 2113 2114 2115
	assert(postlen
	       ? fixed_preimage.nr == preimage->nr
	       : fixed_preimage.nr <= preimage->nr);
	for (i = 0; i < fixed_preimage.nr; i++)
2116 2117 2118
		fixed_preimage.line[i].flag = preimage->line[i].flag;
	free(preimage->line_allocated);
	*preimage = fixed_preimage;
2119

2120
	/*
2121
	 * Adjust the common context lines in postimage. This can be
2122 2123 2124 2125
	 * done in-place when we are shrinking it with whitespace
	 * fixing, but needs a new buffer when ignoring whitespace or
	 * expanding leading tabs to spaces.
	 *
2126 2127
	 * We trust the caller to tell us if the update can be done
	 * in place (postlen==0) or not.
2128
	 */
2129 2130 2131 2132 2133
	old = postimage->buf;
	if (postlen)
		new = postimage->buf = xmalloc(postlen);
	else
		new = old;
2134
	fixed = preimage->buf;
2135 2136

	for (i = reduced = ctx = 0; i < postimage->nr; i++) {
2137 2138 2139 2140 2141 2142 2143
		size_t len = postimage->line[i].len;
		if (!(postimage->line[i].flag & LINE_COMMON)) {
			/* an added line -- no counterparts in preimage */
			memmove(new, old, len);
			old += len;
			new += len;
			continue;
2144
		}
2145 2146 2147 2148 2149 2150 2151 2152 2153 2154

		/* a common context -- skip it in the original postimage */
		old += len;

		/* and find the corresponding one in the fixed preimage */
		while (ctx < preimage->nr &&
		       !(preimage->line[ctx].flag & LINE_COMMON)) {
			fixed += preimage->line[ctx].len;
			ctx++;
		}
2155 2156 2157 2158 2159 2160 2161 2162 2163

		/*
		 * preimage is expected to run out, if the caller
		 * fixed addition of trailing blank lines.
		 */
		if (preimage->nr <= ctx) {
			reduced++;
			continue;
		}
2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175

		/* and copy it in, while fixing the line length */
		len = preimage->line[ctx].len;
		memcpy(new, fixed, len);
		new += len;
		fixed += len;
		postimage->line[i].len = len;
		ctx++;
	}

	/* Fix the length of the whole thing */
	postimage->len = new - postimage->buf;
2176
	postimage->nr -= reduced;
2177 2178
}

2179 2180 2181
static int match_fragment(struct image *img,
			  struct image *preimage,
			  struct image *postimage,
2182
			  unsigned long try,
2183
			  int try_lno,
2184
			  unsigned ws_rule,
2185
			  int match_beginning, int match_end)
2186
{
2187
	int i;
2188
	char *fixed_buf, *buf, *orig, *target;
2189
	struct strbuf fixed;
2190
	size_t fixed_len, postlen;
2191
	int preimage_limit;
2192

2193 2194 2195 2196 2197 2198 2199 2200
	if (preimage->nr + try_lno <= img->nr) {
		/*
		 * The hunk falls within the boundaries of img.
		 */
		preimage_limit = preimage->nr;
		if (match_end && (preimage->nr + try_lno != img->nr))
			return 0;
	} else if (ws_error_action == correct_ws_error &&
2201
		   (ws_rule & WS_BLANK_AT_EOF)) {
2202
		/*
2203 2204 2205 2206 2207
		 * This hunk extends beyond the end of img, and we are
		 * removing blank lines at the end of the file.  This
		 * many lines from the beginning of the preimage must
		 * match with img, and the remainder of the preimage
		 * must be blank.
2208 2209 2210 2211 2212 2213 2214 2215
		 */
		preimage_limit = img->nr - try_lno;
	} else {
		/*
		 * The hunk extends beyond the end of the img and
		 * we are not removing blanks at the end, so we
		 * should reject the hunk at this position.
		 */
2216
		return 0;
2217
	}
2218 2219

	if (match_beginning && try_lno)
2220
		return 0;
2221

2222
	/* Quick hash check */
2223
	for (i = 0; i < preimage_limit; i++)
2224 2225
		if ((img->line[try_lno + i].flag & LINE_PATCHED) ||
		    (preimage->line[i].hash != img->line[try_lno + i].hash))
2226 2227
			return 0;

2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256 2257 2258 2259 2260 2261
	if (preimage_limit == preimage->nr) {
		/*
		 * Do we have an exact match?  If we were told to match
		 * at the end, size must be exactly at try+fragsize,
		 * otherwise try+fragsize must be still within the preimage,
		 * and either case, the old piece should match the preimage
		 * exactly.
		 */
		if ((match_end
		     ? (try + preimage->len == img->len)
		     : (try + preimage->len <= img->len)) &&
		    !memcmp(img->buf + try, preimage->buf, preimage->len))
			return 1;
	} else {
		/*
		 * The preimage extends beyond the end of img, so
		 * there cannot be an exact match.
		 *
		 * There must be one non-blank context line that match
		 * a line before the end of img.
		 */
		char *buf_end;

		buf = preimage->buf;
		buf_end = buf;
		for (i = 0; i < preimage_limit; i++)
			buf_end += preimage->line[i].len;

		for ( ; buf < buf_end; buf++)
			if (!isspace(*buf))
				break;
		if (buf == buf_end)
			return 0;
	}
2262

2263 2264 2265 2266 2267 2268 2269 2270 2271
	/*
	 * No exact match. If we are ignoring whitespace, run a line-by-line
	 * fuzzy matching. We collect all the line length information because
	 * we need it to adjust whitespace if we match.
	 */
	if (ws_ignore_action == ignore_ws_change) {
		size_t imgoff = 0;
		size_t preoff = 0;
		size_t postlen = postimage->len;
2272 2273 2274 2275
		size_t extra_chars;
		char *preimage_eof;
		char *preimage_end;
		for (i = 0; i < preimage_limit; i++) {
2276
			size_t prelen = preimage->line[i].len;
2277
			size_t imglen = img->line[try_lno+i].len;
2278

2279 2280
			if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,
					      preimage->buf + preoff, prelen))
2281 2282
				return 0;
			if (preimage->line[i].flag & LINE_COMMON)
2283 2284
				postlen += imglen - prelen;
			imgoff += imglen;
2285 2286 2287 2288
			preoff += prelen;
		}

		/*
2289 2290 2291
		 * Ok, the preimage matches with whitespace fuzz.
		 *
		 * imgoff now holds the true length of the target that
2292 2293 2294 2295 2296 2297
		 * matches the preimage before the end of the file.
		 *
		 * Count the number of characters in the preimage that fall
		 * beyond the end of the file and make sure that all of them
		 * are whitespace characters. (This can only happen if
		 * we are removing blank lines at the end of the file.)
2298
		 */
2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314
		buf = preimage_eof = preimage->buf + preoff;
		for ( ; i < preimage->nr; i++)
			preoff += preimage->line[i].len;
		preimage_end = preimage->buf + preoff;
		for ( ; buf < preimage_end; buf++)
			if (!isspace(*buf))
				return 0;

		/*
		 * Update the preimage and the common postimage context
		 * lines to use the same whitespace as the target.
		 * If whitespace is missing in the target (i.e.
		 * if the preimage extends beyond the end of the file),
		 * use the whitespace from the preimage.
		 */
		extra_chars = preimage_end - preimage_eof;
2315 2316 2317 2318
		strbuf_init(&fixed, imgoff + extra_chars);
		strbuf_add(&fixed, img->buf + try, imgoff);
		strbuf_add(&fixed, preimage_eof, extra_chars);
		fixed_buf = strbuf_detach(&fixed, &fixed_len);
2319
		update_pre_post_images(preimage, postimage,
2320
				fixed_buf, fixed_len, postlen);
2321 2322 2323
		return 1;
	}

2324 2325 2326
	if (ws_error_action != correct_ws_error)
		return 0;

2327
	/*
2328
	 * The hunk does not apply byte-by-byte, but the hash says
J
Junio C Hamano 已提交
2329
	 * it might with whitespace fuzz. We weren't asked to
2330 2331
	 * ignore whitespace, we were asked to correct whitespace
	 * errors, so let's try matching after whitespace correction.
2332 2333 2334 2335
	 *
	 * The preimage may extend beyond the end of the file,
	 * but in this loop we will only handle the part of the
	 * preimage that falls within the file.
2336
	 */
2337
	strbuf_init(&fixed, preimage->len + 1);
2338 2339
	orig = preimage->buf;
	target = img->buf + try;
2340
	postlen = 0;
2341
	for (i = 0; i < preimage_limit; i++) {
2342 2343
		size_t oldlen = preimage->line[i].len;
		size_t tgtlen = img->line[try_lno + i].len;
2344 2345
		size_t fixstart = fixed.len;
		struct strbuf tgtfix;
2346 2347 2348
		int match;

		/* Try fixing the line in the preimage */
2349
		ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2350 2351

		/* Try fixing the line in the target */
2352 2353
		strbuf_init(&tgtfix, tgtlen);
		ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364

		/*
		 * If they match, either the preimage was based on
		 * a version before our tree fixed whitespace breakage,
		 * or we are lacking a whitespace-fix patch the tree
		 * the preimage was based on already had (i.e. target
		 * has whitespace breakage, the preimage doesn't).
		 * In either case, we are fixing the whitespace breakages
		 * so we might as well take the fix together with their
		 * real change.
		 */
2365 2366 2367
		match = (tgtfix.len == fixed.len - fixstart &&
			 !memcmp(tgtfix.buf, fixed.buf + fixstart,
					     fixed.len - fixstart));
2368
		postlen += tgtfix.len;
2369

2370
		strbuf_release(&tgtfix);
2371 2372 2373 2374 2375
		if (!match)
			goto unmatch_exit;

		orig += oldlen;
		target += tgtlen;
2376 2377
	}

2378 2379 2380 2381 2382 2383 2384 2385

	/*
	 * Now handle the lines in the preimage that falls beyond the
	 * end of the file (if any). They will only match if they are
	 * empty or only contain whitespace (if WS_BLANK_AT_EOL is
	 * false).
	 */
	for ( ; i < preimage->nr; i++) {
2386
		size_t fixstart = fixed.len; /* start of the fixed preimage */
2387 2388 2389 2390
		size_t oldlen = preimage->line[i].len;
		int j;

		/* Try fixing the line in the preimage */
2391
		ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2392

2393 2394
		for (j = fixstart; j < fixed.len; j++)
			if (!isspace(fixed.buf[j]))
2395 2396 2397 2398 2399
				goto unmatch_exit;

		orig += oldlen;
	}

2400 2401 2402 2403 2404
	/*
	 * Yes, the preimage is based on an older version that still
	 * has whitespace breakages unfixed, and fixing them makes the
	 * hunk match.  Update the context lines in the postimage.
	 */
2405
	fixed_buf = strbuf_detach(&fixed, &fixed_len);
2406 2407
	if (postlen < postimage->len)
		postlen = 0;
2408
	update_pre_post_images(preimage, postimage,
2409
			       fixed_buf, fixed_len, postlen);
2410 2411 2412
	return 1;

 unmatch_exit:
2413
	strbuf_release(&fixed);
2414
	return 0;
2415 2416
}

2417 2418 2419 2420
static int find_pos(struct image *img,
		    struct image *preimage,
		    struct image *postimage,
		    int line,
2421
		    unsigned ws_rule,
2422
		    int match_beginning, int match_end)
2423
{
2424 2425 2426
	int i;
	unsigned long backwards, forwards, try;
	int backwards_lno, forwards_lno, try_lno;
2427

2428
	/*
2429
	 * If match_beginning or match_end is specified, there is no
2430 2431 2432 2433 2434 2435 2436 2437
	 * point starting from a wrong line that will never match and
	 * wander around and wait for a match at the specified end.
	 */
	if (match_beginning)
		line = 0;
	else if (match_end)
		line = img->nr - preimage->nr;

2438 2439 2440 2441 2442 2443
	/*
	 * Because the comparison is unsigned, the following test
	 * will also take care of a negative line number that can
	 * result when match_end and preimage is larger than the target.
	 */
	if ((size_t) line > img->nr)
2444 2445
		line = img->nr;

2446 2447 2448
	try = 0;
	for (i = 0; i < line; i++)
		try += img->line[i].len;
2449

2450 2451 2452 2453
	/*
	 * There's probably some smart way to do this, but I'll leave
	 * that to the smart and beautiful people. I'm simple and stupid.
	 */
2454 2455 2456 2457 2458
	backwards = try;
	backwards_lno = line;
	forwards = try;
	forwards_lno = line;
	try_lno = line;
2459

2460
	for (i = 0; ; i++) {
2461
		if (match_fragment(img, preimage, postimage,
2462
				   try, try_lno, ws_rule,
2463 2464
				   match_beginning, match_end))
			return try_lno;
2465 2466

	again:
2467
		if (backwards_lno == 0 && forwards_lno == img->nr)
2468
			break;
2469 2470

		if (i & 1) {
2471
			if (backwards_lno == 0) {
2472 2473
				i++;
				goto again;
2474
			}
2475 2476
			backwards_lno--;
			backwards -= img->line[backwards_lno].len;
2477
			try = backwards;
2478
			try_lno = backwards_lno;
2479
		} else {
2480
			if (forwards_lno == img->nr) {
2481 2482
				i++;
				goto again;
2483
			}
2484 2485
			forwards += img->line[forwards_lno].len;
			forwards_lno++;
2486
			try = forwards;
2487
			try_lno = forwards_lno;
2488 2489 2490
		}

	}
2491 2492 2493
	return -1;
}

2494
static void remove_first_line(struct image *img)
2495
{
2496 2497 2498 2499
	img->buf += img->line[0].len;
	img->len -= img->line[0].len;
	img->line++;
	img->nr--;
2500 2501
}

2502
static void remove_last_line(struct image *img)
2503
{
2504
	img->len -= img->line[--img->nr].len;
2505 2506
}

2507 2508 2509 2510 2511
/*
 * The change from "preimage" and "postimage" has been found to
 * apply at applied_pos (counts in line numbers) in "img".
 * Update "img" to remove "preimage" and replace it with "postimage".
 */
2512 2513 2514 2515
static void update_image(struct image *img,
			 int applied_pos,
			 struct image *preimage,
			 struct image *postimage)
2516
{
2517
	/*
2518 2519
	 * remove the copy of preimage at offset in img
	 * and replace it with postimage
2520
	 */
2521 2522 2523
	int i, nr;
	size_t remove_count, insert_count, applied_at = 0;
	char *result;
2524 2525 2526 2527 2528 2529 2530 2531 2532 2533 2534 2535 2536 2537
	int preimage_limit;

	/*
	 * If we are removing blank lines at the end of img,
	 * the preimage may extend beyond the end.
	 * If that is the case, we must be careful only to
	 * remove the part of the preimage that falls within
	 * the boundaries of img. Initialize preimage_limit
	 * to the number of lines in the preimage that falls
	 * within the boundaries.
	 */
	preimage_limit = preimage->nr;
	if (preimage_limit > img->nr - applied_pos)
		preimage_limit = img->nr - applied_pos;
2538

2539 2540 2541 2542
	for (i = 0; i < applied_pos; i++)
		applied_at += img->line[i].len;

	remove_count = 0;
2543
	for (i = 0; i < preimage_limit; i++)
2544 2545 2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557 2558 2559
		remove_count += img->line[applied_pos + i].len;
	insert_count = postimage->len;

	/* Adjust the contents */
	result = xmalloc(img->len + insert_count - remove_count + 1);
	memcpy(result, img->buf, applied_at);
	memcpy(result + applied_at, postimage->buf, postimage->len);
	memcpy(result + applied_at + postimage->len,
	       img->buf + (applied_at + remove_count),
	       img->len - (applied_at + remove_count));
	free(img->buf);
	img->buf = result;
	img->len += insert_count - remove_count;
	result[img->len] = '\0';

	/* Adjust the line table */
2560 2561
	nr = img->nr + postimage->nr - preimage_limit;
	if (preimage_limit < postimage->nr) {
2562
		/*
2563 2564
		 * NOTE: this knows that we never call remove_first_line()
		 * on anything other than pre/post image.
2565
		 */
2566 2567
		img->line = xrealloc(img->line, nr * sizeof(*img->line));
		img->line_allocated = img->line;
2568
	}
2569
	if (preimage_limit != postimage->nr)
2570
		memmove(img->line + applied_pos + postimage->nr,
2571 2572
			img->line + applied_pos + preimage_limit,
			(img->nr - (applied_pos + preimage_limit)) *
2573 2574 2575 2576
			sizeof(*img->line));
	memcpy(img->line + applied_pos,
	       postimage->line,
	       postimage->nr * sizeof(*img->line));
2577 2578 2579
	if (!allow_overlap)
		for (i = 0; i < postimage->nr; i++)
			img->line[applied_pos + i].flag |= LINE_PATCHED;
2580
	img->nr = nr;
2581 2582
}

2583 2584 2585 2586 2587
/*
 * Use the patch-hunk text in "frag" to prepare two images (preimage and
 * postimage) for the hunk.  Find lines that match "preimage" in "img" and
 * replace the part of "img" with "postimage" text.
 */
2588
static int apply_one_fragment(struct image *img, struct fragment *frag,
2589 2590
			      int inaccurate_eof, unsigned ws_rule,
			      int nth_fragment)
2591
{
2592
	int match_beginning, match_end;
2593
	const char *patch = frag->patch;
2594
	int size = frag->size;
2595 2596
	char *old, *oldlines;
	struct strbuf newlines;
2597
	int new_blank_lines_at_end = 0;
2598 2599
	int found_new_blank_lines_at_end = 0;
	int hunk_linenr = frag->linenr;
2600
	unsigned long leading, trailing;
2601 2602 2603
	int pos, applied_pos;
	struct image preimage;
	struct image postimage;
2604

2605 2606
	memset(&preimage, 0, sizeof(preimage));
	memset(&postimage, 0, sizeof(postimage));
2607
	oldlines = xmalloc(size);
2608
	strbuf_init(&newlines, size);
2609

2610
	old = oldlines;
2611
	while (size > 0) {
J
Johannes Schindelin 已提交
2612
		char first;
2613
		int len = linelen(patch, size);
2614
		int plen;
2615
		int added_blank_line = 0;
2616
		int is_blank_context = 0;
2617
		size_t start;
2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628

		if (!len)
			break;

		/*
		 * "plen" is how much of the line we should use for
		 * the actual patch data. Normally we just remove the
		 * first character on the line, but if the line is
		 * followed by "\ No newline", then we also remove the
		 * last one (which is the newline, of course).
		 */
2629
		plen = len - 1;
2630
		if (len < size && patch[len] == '\\')
2631
			plen--;
J
Johannes Schindelin 已提交
2632
		first = *patch;
2633
		if (apply_in_reverse) {
J
Johannes Schindelin 已提交
2634 2635 2636 2637 2638
			if (first == '-')
				first = '+';
			else if (first == '+')
				first = '-';
		}
2639

J
Johannes Schindelin 已提交
2640
		switch (first) {
2641 2642 2643 2644 2645
		case '\n':
			/* Newer GNU diff, empty context line */
			if (plen < 0)
				/* ... followed by '\No newline'; nothing */
				break;
2646
			*old++ = '\n';
2647
			strbuf_addch(&newlines, '\n');
2648 2649
			add_line_info(&preimage, "\n", 1, LINE_COMMON);
			add_line_info(&postimage, "\n", 1, LINE_COMMON);
2650
			is_blank_context = 1;
2651
			break;
2652
		case ' ':
2653 2654
			if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
			    ws_blank_line(patch + 1, plen, ws_rule))
2655
				is_blank_context = 1;
2656
		case '-':
2657 2658
			memcpy(old, patch + 1, plen);
			add_line_info(&preimage, old, plen,
2659
				      (first == ' ' ? LINE_COMMON : 0));
2660
			old += plen;
J
Johannes Schindelin 已提交
2661
			if (first == '-')
2662 2663 2664
				break;
		/* Fall-through for ' ' */
		case '+':
2665 2666 2667 2668
			/* --no-add does not add new lines */
			if (first == '+' && no_add)
				break;

2669
			start = newlines.len;
2670 2671 2672
			if (first != '+' ||
			    !whitespace_error ||
			    ws_error_action != correct_ws_error) {
2673
				strbuf_add(&newlines, patch + 1, plen);
2674 2675
			}
			else {
2676
				ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &applied_after_fixing_ws);
2677
			}
2678
			add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2679 2680
				      (first == '+' ? 0 : LINE_COMMON));
			if (first == '+' &&
2681 2682
			    (ws_rule & WS_BLANK_AT_EOF) &&
			    ws_blank_line(patch + 1, plen, ws_rule))
2683
				added_blank_line = 1;
2684 2685 2686 2687 2688
			break;
		case '@': case '\\':
			/* Ignore it, we already handled it */
			break;
		default:
2689
			if (apply_verbosely)
2690
				error(_("invalid start of line: '%c'"), first);
2691 2692
			return -1;
		}
2693 2694 2695
		if (added_blank_line) {
			if (!new_blank_lines_at_end)
				found_new_blank_lines_at_end = hunk_linenr;
2696
			new_blank_lines_at_end++;
2697
		}
2698 2699
		else if (is_blank_context)
			;
2700 2701
		else
			new_blank_lines_at_end = 0;
2702 2703
		patch += len;
		size -= len;
2704
		hunk_linenr++;
2705
	}
2706
	if (inaccurate_eof &&
2707
	    old > oldlines && old[-1] == '\n' &&
2708
	    newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2709
		old--;
2710
		strbuf_setlen(&newlines, newlines.len - 1);
2711
	}
2712 2713 2714

	leading = frag->leading;
	trailing = frag->trailing;
2715 2716

	/*
2717 2718
	 * A hunk to change lines at the beginning would begin with
	 * @@ -1,L +N,M @@
2719 2720
	 * but we need to be careful.  -U0 that inserts before the second
	 * line also has this pattern.
2721
	 *
2722 2723 2724 2725 2726
	 * And a hunk to add to an empty file would begin with
	 * @@ -0,0 +N,M @@
	 *
	 * In other words, a hunk that is (frag->oldpos <= 1) with or
	 * without leading context must match at the beginning.
2727
	 */
2728 2729
	match_beginning = (!frag->oldpos ||
			   (frag->oldpos == 1 && !unidiff_zero));
2730 2731 2732 2733 2734 2735 2736 2737

	/*
	 * A hunk without trailing lines must match at the end.
	 * However, we simply cannot tell if a hunk must match end
	 * from the lack of trailing lines if the patch was generated
	 * with unidiff without any context.
	 */
	match_end = !unidiff_zero && !trailing;
2738

2739
	pos = frag->newpos ? (frag->newpos - 1) : 0;
2740 2741
	preimage.buf = oldlines;
	preimage.len = old - oldlines;
2742 2743
	postimage.buf = newlines.buf;
	postimage.len = newlines.len;
2744 2745 2746
	preimage.line = preimage.line_allocated;
	postimage.line = postimage.line_allocated;

2747
	for (;;) {
2748

2749 2750
		applied_pos = find_pos(img, &preimage, &postimage, pos,
				       ws_rule, match_beginning, match_end);
2751 2752

		if (applied_pos >= 0)
2753 2754 2755 2756 2757
			break;

		/* Am I at my context limits? */
		if ((leading <= p_context) && (trailing <= p_context))
			break;
2758 2759
		if (match_beginning || match_end) {
			match_beginning = match_end = 0;
2760 2761
			continue;
		}
2762

2763 2764 2765 2766
		/*
		 * Reduce the number of context lines; reduce both
		 * leading and trailing if they are equal otherwise
		 * just reduce the larger context.
2767 2768
		 */
		if (leading >= trailing) {
2769 2770
			remove_first_line(&preimage);
			remove_first_line(&postimage);
2771 2772 2773 2774
			pos--;
			leading--;
		}
		if (trailing > leading) {
2775 2776
			remove_last_line(&preimage);
			remove_last_line(&postimage);
2777
			trailing--;
2778
		}
2779 2780
	}

2781
	if (applied_pos >= 0) {
2782
		if (new_blank_lines_at_end &&
2783
		    preimage.nr + applied_pos >= img->nr &&
2784 2785
		    (ws_rule & WS_BLANK_AT_EOF) &&
		    ws_error_action != nowarn_ws_error) {
2786 2787
			record_ws_error(WS_BLANK_AT_EOF, "+", 1,
					found_new_blank_lines_at_end);
2788 2789 2790 2791
			if (ws_error_action == correct_ws_error) {
				while (new_blank_lines_at_end--)
					remove_last_line(&postimage);
			}
2792
			/*
2793 2794 2795 2796 2797
			 * We would want to prevent write_out_results()
			 * from taking place in apply_patch() that follows
			 * the callchain led us here, which is:
			 * apply_patch->check_patch_list->check_patch->
			 * apply_data->apply_fragments->apply_one_fragment
2798
			 */
2799 2800
			if (ws_error_action == die_on_ws_error)
				apply = 0;
2801
		}
2802

2803 2804 2805 2806
		if (apply_verbosely && applied_pos != pos) {
			int offset = applied_pos - pos;
			if (apply_in_reverse)
				offset = 0 - offset;
2807 2808 2809 2810 2811
			fprintf_ln(stderr,
				   Q_("Hunk #%d succeeded at %d (offset %d line).",
				      "Hunk #%d succeeded at %d (offset %d lines).",
				      offset),
				   nth_fragment, applied_pos + 1, offset);
2812 2813
		}

2814 2815 2816 2817 2818 2819
		/*
		 * Warn if it was necessary to reduce the number
		 * of context lines.
		 */
		if ((leading != frag->leading) ||
		    (trailing != frag->trailing))
2820 2821 2822
			fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
					     " to apply fragment at %d"),
				   leading, trailing, applied_pos+1);
2823 2824 2825
		update_image(img, applied_pos, &preimage, &postimage);
	} else {
		if (apply_verbosely)
2826
			error(_("while searching for:\n%.*s"),
2827
			      (int)(old - oldlines), oldlines);
2828
	}
2829

2830
	free(oldlines);
2831
	strbuf_release(&newlines);
2832 2833 2834 2835
	free(preimage.line_allocated);
	free(postimage.line_allocated);

	return (applied_pos < 0);
2836 2837
}

2838
static int apply_binary_fragment(struct image *img, struct patch *patch)
J
Junio C Hamano 已提交
2839 2840
{
	struct fragment *fragment = patch->fragments;
2841 2842
	unsigned long len;
	void *dst;
J
Junio C Hamano 已提交
2843

2844
	if (!fragment)
2845
		return error(_("missing binary patch data for '%s'"),
2846 2847 2848 2849
			     patch->new_name ?
			     patch->new_name :
			     patch->old_name);

2850 2851 2852 2853 2854 2855 2856
	/* Binary patch is irreversible without the optional second hunk */
	if (apply_in_reverse) {
		if (!fragment->next)
			return error("cannot reverse-apply a binary patch "
				     "without the reverse hunk to '%s'",
				     patch->new_name
				     ? patch->new_name : patch->old_name);
2857
		fragment = fragment->next;
2858 2859
	}
	switch (fragment->binary_patch_method) {
J
Junio C Hamano 已提交
2860
	case BINARY_DELTA_DEFLATED:
2861
		dst = patch_delta(img->buf, img->len, fragment->patch,
2862 2863 2864
				  fragment->size, &len);
		if (!dst)
			return -1;
2865 2866 2867
		clear_image(img);
		img->buf = dst;
		img->len = len;
2868
		return 0;
J
Junio C Hamano 已提交
2869
	case BINARY_LITERAL_DEFLATED:
2870 2871
		clear_image(img);
		img->len = fragment->size;
2872
		img->buf = xmemdupz(fragment->patch, img->len);
2873
		return 0;
J
Junio C Hamano 已提交
2874
	}
2875
	return -1;
J
Junio C Hamano 已提交
2876 2877
}

2878 2879 2880 2881 2882 2883
/*
 * Replace "img" with the result of applying the binary patch.
 * The binary patch data itself in patch->fragment is still kept
 * but the preimage prepared by the caller in "img" is freed here
 * or in the helper function apply_binary_fragment() this calls.
 */
2884
static int apply_binary(struct image *img, struct patch *patch)
2885
{
J
Junio C Hamano 已提交
2886
	const char *name = patch->old_name ? patch->old_name : patch->new_name;
J
Junio C Hamano 已提交
2887
	unsigned char sha1[20];
J
Junio C Hamano 已提交
2888

2889 2890
	/*
	 * For safety, we require patch index line to contain
J
Junio C Hamano 已提交
2891 2892 2893 2894 2895 2896 2897 2898
	 * full 40-byte textual SHA1 for old and new, at least for now.
	 */
	if (strlen(patch->old_sha1_prefix) != 40 ||
	    strlen(patch->new_sha1_prefix) != 40 ||
	    get_sha1_hex(patch->old_sha1_prefix, sha1) ||
	    get_sha1_hex(patch->new_sha1_prefix, sha1))
		return error("cannot apply binary patch to '%s' "
			     "without full index line", name);
J
Junio C Hamano 已提交
2899

J
Junio C Hamano 已提交
2900
	if (patch->old_name) {
2901 2902
		/*
		 * See if the old one matches what the patch
J
Junio C Hamano 已提交
2903
		 * applies to.
J
Junio C Hamano 已提交
2904
		 */
2905
		hash_sha1_file(img->buf, img->len, blob_type, sha1);
J
Junio C Hamano 已提交
2906 2907 2908 2909 2910 2911 2912 2913
		if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
			return error("the patch applies to '%s' (%s), "
				     "which does not match the "
				     "current contents.",
				     name, sha1_to_hex(sha1));
	}
	else {
		/* Otherwise, the old one must be empty. */
2914
		if (img->len)
J
Junio C Hamano 已提交
2915 2916 2917
			return error("the patch applies to an empty "
				     "'%s' but it is not empty", name);
	}
J
Junio C Hamano 已提交
2918

J
Junio C Hamano 已提交
2919
	get_sha1_hex(patch->new_sha1_prefix, sha1);
D
David Rientjes 已提交
2920
	if (is_null_sha1(sha1)) {
2921
		clear_image(img);
J
Junio C Hamano 已提交
2922
		return 0; /* deletion patch */
J
Junio C Hamano 已提交
2923
	}
J
Junio C Hamano 已提交
2924

J
Junio C Hamano 已提交
2925
	if (has_sha1_file(sha1)) {
J
Junio C Hamano 已提交
2926
		/* We already have the postimage */
2927
		enum object_type type;
J
Junio C Hamano 已提交
2928
		unsigned long size;
2929
		char *result;
J
Junio C Hamano 已提交
2930

2931 2932
		result = read_sha1_file(sha1, &type, &size);
		if (!result)
J
Junio C Hamano 已提交
2933 2934 2935
			return error("the necessary postimage %s for "
				     "'%s' cannot be read",
				     patch->new_sha1_prefix, name);
2936 2937 2938
		clear_image(img);
		img->buf = result;
		img->len = size;
2939
	} else {
2940 2941
		/*
		 * We have verified buf matches the preimage;
J
Junio C Hamano 已提交
2942 2943
		 * apply the patch data to it, which is stored
		 * in the patch->fragments->{patch,size}.
J
Junio C Hamano 已提交
2944
		 */
2945
		if (apply_binary_fragment(img, patch))
2946
			return error(_("binary patch does not apply to '%s'"),
J
Junio C Hamano 已提交
2947
				     name);
J
Junio C Hamano 已提交
2948

J
Junio C Hamano 已提交
2949
		/* verify that the result matches */
2950
		hash_sha1_file(img->buf, img->len, blob_type, sha1);
J
Junio C Hamano 已提交
2951
		if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
2952
			return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
2953
				name, patch->new_sha1_prefix, sha1_to_hex(sha1));
J
Junio C Hamano 已提交
2954
	}
2955

J
Junio C Hamano 已提交
2956 2957 2958
	return 0;
}

2959
static int apply_fragments(struct image *img, struct patch *patch)
J
Junio C Hamano 已提交
2960 2961 2962
{
	struct fragment *frag = patch->fragments;
	const char *name = patch->old_name ? patch->old_name : patch->new_name;
2963 2964
	unsigned ws_rule = patch->ws_rule;
	unsigned inaccurate_eof = patch->inaccurate_eof;
2965
	int nth = 0;
J
Junio C Hamano 已提交
2966 2967

	if (patch->is_binary)
2968
		return apply_binary(img, patch);
J
Junio C Hamano 已提交
2969

2970
	while (frag) {
2971 2972
		nth++;
		if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule, nth)) {
2973
			error(_("patch failed: %s:%ld"), name, frag->oldpos);
J
Junio C Hamano 已提交
2974 2975 2976 2977
			if (!apply_with_reject)
				return -1;
			frag->rejected = 1;
		}
2978 2979
		frag = frag->next;
	}
2980
	return 0;
2981 2982
}

2983
static int read_blob_object(struct strbuf *buf, const unsigned char *sha1, unsigned mode)
2984
{
2985
	if (S_ISGITLINK(mode)) {
2986
		strbuf_grow(buf, 100);
2987
		strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(sha1));
2988 2989
	} else {
		enum object_type type;
2990 2991 2992
		unsigned long sz;
		char *result;

2993
		result = read_sha1_file(sha1, &type, &sz);
2994
		if (!result)
2995
			return -1;
2996 2997
		/* XXX read_sha1_file NUL-terminates */
		strbuf_attach(buf, result, sz, sz + 1);
2998 2999 3000 3001
	}
	return 0;
}

3002
static int read_file_or_gitlink(const struct cache_entry *ce, struct strbuf *buf)
3003 3004 3005 3006 3007 3008
{
	if (!ce)
		return 0;
	return read_blob_object(buf, ce->sha1, ce->ce_mode);
}

3009 3010
static struct patch *in_fn_table(const char *name)
{
3011
	struct string_list_item *item;
3012 3013 3014 3015

	if (name == NULL)
		return NULL;

3016
	item = string_list_lookup(&fn_table, name);
3017 3018 3019 3020 3021 3022
	if (item != NULL)
		return (struct patch *)item->util;

	return NULL;
}

3023 3024 3025 3026
/*
 * item->util in the filename table records the status of the path.
 * Usually it points at a patch (whose result records the contents
 * of it after applying it), but it could be PATH_WAS_DELETED for a
3027 3028 3029 3030 3031
 * path that a previously applied patch has already removed, or
 * PATH_TO_BE_DELETED for a path that a later patch would remove.
 *
 * The latter is needed to deal with a case where two paths A and B
 * are swapped by first renaming A to B and then renaming B to A;
3032
 * moving A to B should not be prevented due to presence of B as we
3033
 * will remove it in a later patch.
3034
 */
3035
#define PATH_TO_BE_DELETED ((struct patch *) -2)
3036 3037 3038 3039 3040 3041 3042 3043 3044 3045 3046 3047
#define PATH_WAS_DELETED ((struct patch *) -1)

static int to_be_deleted(struct patch *patch)
{
	return patch == PATH_TO_BE_DELETED;
}

static int was_deleted(struct patch *patch)
{
	return patch == PATH_WAS_DELETED;
}

3048 3049
static void add_to_fn_table(struct patch *patch)
{
3050
	struct string_list_item *item;
3051 3052 3053 3054 3055 3056 3057

	/*
	 * Always add new_name unless patch is a deletion
	 * This should cover the cases for normal diffs,
	 * file creations and copies
	 */
	if (patch->new_name != NULL) {
3058
		item = string_list_insert(&fn_table, patch->new_name);
3059 3060 3061 3062 3063 3064 3065 3066
		item->util = patch;
	}

	/*
	 * store a failure on rename/deletion cases because
	 * later chunks shouldn't patch old names
	 */
	if ((patch->new_name == NULL) || (patch->is_rename)) {
3067
		item = string_list_insert(&fn_table, patch->old_name);
3068 3069 3070 3071 3072 3073 3074 3075 3076 3077 3078 3079
		item->util = PATH_WAS_DELETED;
	}
}

static void prepare_fn_table(struct patch *patch)
{
	/*
	 * store information about incoming file deletion
	 */
	while (patch) {
		if ((patch->new_name == NULL) || (patch->is_rename)) {
			struct string_list_item *item;
3080
			item = string_list_insert(&fn_table, patch->old_name);
3081 3082 3083
			item->util = PATH_TO_BE_DELETED;
		}
		patch = patch->next;
3084 3085 3086
	}
}

3087 3088 3089 3090 3091 3092 3093 3094 3095 3096 3097 3098
static int checkout_target(struct cache_entry *ce, struct stat *st)
{
	struct checkout costate;

	memset(&costate, 0, sizeof(costate));
	costate.base_dir = "";
	costate.refresh_cache = 1;
	if (checkout_entry(ce, &costate, NULL) || lstat(ce->name, st))
		return error(_("cannot checkout %s"), ce->name);
	return 0;
}

3099 3100 3101 3102 3103 3104 3105 3106 3107 3108 3109 3110 3111 3112 3113 3114 3115 3116 3117 3118 3119
static struct patch *previous_patch(struct patch *patch, int *gone)
{
	struct patch *previous;

	*gone = 0;
	if (patch->is_copy || patch->is_rename)
		return NULL; /* "git" patches do not depend on the order */

	previous = in_fn_table(patch->old_name);
	if (!previous)
		return NULL;

	if (to_be_deleted(previous))
		return NULL; /* the deletion hasn't happened yet */

	if (was_deleted(previous))
		*gone = 1;

	return previous;
}

3120
static int verify_index_match(const struct cache_entry *ce, struct stat *st)
3121 3122 3123 3124 3125 3126 3127 3128 3129
{
	if (S_ISGITLINK(ce->ce_mode)) {
		if (!S_ISDIR(st->st_mode))
			return -1;
		return 0;
	}
	return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
}

3130 3131 3132
#define SUBMODULE_PATCH_WITHOUT_INDEX 1

static int load_patch_target(struct strbuf *buf,
3133
			     const struct cache_entry *ce,
3134 3135 3136 3137 3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148 3149 3150 3151 3152 3153 3154
			     struct stat *st,
			     const char *name,
			     unsigned expected_mode)
{
	if (cached) {
		if (read_file_or_gitlink(ce, buf))
			return error(_("read of %s failed"), name);
	} else if (name) {
		if (S_ISGITLINK(expected_mode)) {
			if (ce)
				return read_file_or_gitlink(ce, buf);
			else
				return SUBMODULE_PATCH_WITHOUT_INDEX;
		} else {
			if (read_old_data(st, name, buf))
				return error(_("read of %s failed"), name);
		}
	}
	return 0;
}

3155 3156 3157 3158 3159 3160 3161
/*
 * We are about to apply "patch"; populate the "image" with the
 * current version we have, from the working tree or from the index,
 * depending on the situation e.g. --cached/--index.  If we are
 * applying a non-git patch that incrementally updates the tree,
 * we read from the result of a previous diff.
 */
3162
static int load_preimage(struct image *image,
3163 3164
			 struct patch *patch, struct stat *st,
			 const struct cache_entry *ce)
3165
{
3166
	struct strbuf buf = STRBUF_INIT;
3167 3168
	size_t len;
	char *img;
3169 3170
	struct patch *previous;
	int status;
3171

3172 3173 3174 3175 3176
	previous = previous_patch(patch, &status);
	if (status)
		return error(_("path %s has been renamed/deleted"),
			     patch->old_name);
	if (previous) {
3177
		/* We have a patched copy in memory; use that. */
3178
		strbuf_add(&buf, previous->result, previous->resultsize);
3179 3180 3181 3182 3183 3184 3185 3186 3187 3188 3189 3190 3191 3192 3193
	} else {
		status = load_patch_target(&buf, ce, st,
					   patch->old_name, patch->old_mode);
		if (status < 0)
			return status;
		else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
			/*
			 * There is no way to apply subproject
			 * patch without looking at the index.
			 * NEEDSWORK: shouldn't this be flagged
			 * as an error???
			 */
			free_fragment_list(patch->fragments);
			patch->fragments = NULL;
		} else if (status) {
3194
			return error(_("read of %s failed"), patch->old_name);
3195 3196
		}
	}
3197

3198
	img = strbuf_detach(&buf, &len);
3199 3200 3201
	prepare_image(image, img, len, !patch->is_binary);
	return 0;
}
3202

3203 3204 3205 3206 3207 3208
static int three_way_merge(struct image *image,
			   char *path,
			   const unsigned char *base,
			   const unsigned char *ours,
			   const unsigned char *theirs)
{
3209 3210 3211
	mmfile_t base_file, our_file, their_file;
	mmbuffer_t result = { NULL };
	int status;
3212

3213 3214 3215 3216 3217 3218 3219 3220 3221 3222 3223 3224 3225 3226 3227 3228 3229
	read_mmblob(&base_file, base);
	read_mmblob(&our_file, ours);
	read_mmblob(&their_file, theirs);
	status = ll_merge(&result, path,
			  &base_file, "base",
			  &our_file, "ours",
			  &their_file, "theirs", NULL);
	free(base_file.ptr);
	free(our_file.ptr);
	free(their_file.ptr);
	if (status < 0 || !result.ptr) {
		free(result.ptr);
		return -1;
	}
	clear_image(image);
	image->buf = result.ptr;
	image->len = result.size;
3230

3231
	return status;
3232 3233
}

3234 3235 3236 3237 3238 3239 3240 3241 3242 3243 3244 3245 3246 3247 3248 3249 3250 3251 3252 3253 3254 3255 3256 3257 3258 3259 3260 3261 3262 3263 3264 3265 3266 3267 3268 3269 3270 3271 3272
/*
 * When directly falling back to add/add three-way merge, we read from
 * the current contents of the new_name.  In no cases other than that
 * this function will be called.
 */
static int load_current(struct image *image, struct patch *patch)
{
	struct strbuf buf = STRBUF_INIT;
	int status, pos;
	size_t len;
	char *img;
	struct stat st;
	struct cache_entry *ce;
	char *name = patch->new_name;
	unsigned mode = patch->new_mode;

	if (!patch->is_new)
		die("BUG: patch to %s is not a creation", patch->old_name);

	pos = cache_name_pos(name, strlen(name));
	if (pos < 0)
		return error(_("%s: does not exist in index"), name);
	ce = active_cache[pos];
	if (lstat(name, &st)) {
		if (errno != ENOENT)
			return error(_("%s: %s"), name, strerror(errno));
		if (checkout_target(ce, &st))
			return -1;
	}
	if (verify_index_match(ce, &st))
		return error(_("%s: does not match index"), name);

	status = load_patch_target(&buf, ce, &st, name, mode);
	if (status < 0)
		return status;
	else if (status)
		return -1;
	img = strbuf_detach(&buf, &len);
	prepare_image(image, img, len, !patch->is_binary);
3273 3274 3275
	return 0;
}

3276
static int try_threeway(struct image *image, struct patch *patch,
3277
			struct stat *st, const struct cache_entry *ce)
3278
{
3279 3280 3281
	unsigned char pre_sha1[20], post_sha1[20], our_sha1[20];
	struct strbuf buf = STRBUF_INIT;
	size_t len;
3282
	int status;
3283 3284 3285 3286
	char *img;
	struct image tmp_image;

	/* No point falling back to 3-way merge in these cases */
3287
	if (patch->is_delete ||
3288 3289
	    S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode))
		return -1;
3290

3291
	/* Preimage the patch was prepared for */
3292 3293 3294 3295
	if (patch->is_new)
		write_sha1_file("", 0, blob_type, pre_sha1);
	else if (get_sha1(patch->old_sha1_prefix, pre_sha1) ||
		 read_blob_object(&buf, pre_sha1, patch->old_mode))
3296
		return error("repository lacks the necessary blob to fall back on 3-way merge.");
3297 3298 3299

	fprintf(stderr, "Falling back to three-way merge...\n");

3300 3301 3302 3303 3304 3305 3306 3307 3308 3309 3310 3311
	img = strbuf_detach(&buf, &len);
	prepare_image(&tmp_image, img, len, 1);
	/* Apply the patch to get the post image */
	if (apply_fragments(&tmp_image, patch) < 0) {
		clear_image(&tmp_image);
		return -1;
	}
	/* post_sha1[] is theirs */
	write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, post_sha1);
	clear_image(&tmp_image);

	/* our_sha1[] is ours */
3312 3313 3314 3315 3316 3317 3318 3319 3320
	if (patch->is_new) {
		if (load_current(&tmp_image, patch))
			return error("cannot read the current contents of '%s'",
				     patch->new_name);
	} else {
		if (load_preimage(&tmp_image, patch, st, ce))
			return error("cannot read the current contents of '%s'",
				     patch->old_name);
	}
3321 3322 3323 3324
	write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, our_sha1);
	clear_image(&tmp_image);

	/* in-core three-way merge between post and our using pre as base */
3325 3326 3327 3328 3329 3330 3331 3332 3333
	status = three_way_merge(image, patch->new_name,
				 pre_sha1, our_sha1, post_sha1);
	if (status < 0) {
		fprintf(stderr, "Failed to fall back on three-way merge...\n");
		return status;
	}

	if (status) {
		patch->conflicted_threeway = 1;
3334 3335 3336 3337
		if (patch->is_new)
			hashclr(patch->threeway_stage[0]);
		else
			hashcpy(patch->threeway_stage[0], pre_sha1);
3338 3339 3340 3341 3342
		hashcpy(patch->threeway_stage[1], our_sha1);
		hashcpy(patch->threeway_stage[2], post_sha1);
		fprintf(stderr, "Applied patch to '%s' with conflicts.\n", patch->new_name);
	} else {
		fprintf(stderr, "Applied patch to '%s' cleanly.\n", patch->new_name);
3343 3344 3345 3346
	}
	return 0;
}

3347
static int apply_data(struct patch *patch, struct stat *st, const struct cache_entry *ce)
3348
{
3349 3350 3351 3352
	struct image image;

	if (load_preimage(&image, patch, st, ce) < 0)
		return -1;
3353

3354 3355
	if (patch->direct_to_threeway ||
	    apply_fragments(&image, patch) < 0) {
3356 3357
		/* Note: with --reject, apply_fragments() returns 0 */
		if (!threeway || try_threeway(&image, patch, st, ce) < 0)
3358 3359
			return -1;
	}
3360 3361
	patch->result = image.buf;
	patch->resultsize = image.len;
3362
	add_to_fn_table(patch);
3363
	free(image.line_allocated);
3364

3365
	if (0 < patch->is_delete && patch->resultsize)
3366
		return error(_("removal patch leaves file contents"));
3367

3368
	return 0;
3369 3370
}

3371 3372 3373 3374 3375 3376 3377 3378 3379 3380 3381
/*
 * If "patch" that we are looking at modifies or deletes what we have,
 * we would want it not to lose any local modification we have, either
 * in the working tree or in the index.
 *
 * This also decides if a non-git patch is a creation patch or a
 * modification to an existing empty file.  We do not check the state
 * of the current tree for a creation patch in this function; the caller
 * check_patch() separately makes sure (and errors out otherwise) that
 * the path the patch creates does not exist in the current tree.
 */
3382
static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)
3383 3384
{
	const char *old_name = patch->old_name;
3385 3386
	struct patch *previous = NULL;
	int stat_ret = 0, status;
3387
	unsigned st_mode = 0;
3388

3389 3390
	if (!old_name)
		return 0;
3391

3392
	assert(patch->is_new <= 0);
3393
	previous = previous_patch(patch, &status);
J
Junio C Hamano 已提交
3394

3395 3396 3397 3398
	if (status)
		return error(_("path %s has been renamed/deleted"), old_name);
	if (previous) {
		st_mode = previous->new_mode;
3399
	} else if (!cached) {
3400 3401
		stat_ret = lstat(old_name, st);
		if (stat_ret && errno != ENOENT)
3402
			return error(_("%s: %s"), old_name, strerror(errno));
3403
	}
J
Junio C Hamano 已提交
3404

3405
	if (check_index && !previous) {
3406 3407 3408 3409
		int pos = cache_name_pos(old_name, strlen(old_name));
		if (pos < 0) {
			if (patch->is_new < 0)
				goto is_new;
3410
			return error(_("%s: does not exist in index"), old_name);
3411 3412 3413
		}
		*ce = active_cache[pos];
		if (stat_ret < 0) {
3414
			if (checkout_target(*ce, st))
3415 3416 3417
				return -1;
		}
		if (!cached && verify_index_match(*ce, st))
3418
			return error(_("%s: does not match index"), old_name);
3419 3420 3421
		if (cached)
			st_mode = (*ce)->ce_mode;
	} else if (stat_ret < 0) {
3422
		if (patch->is_new < 0)
3423
			goto is_new;
3424
		return error(_("%s: %s"), old_name, strerror(errno));
3425
	}
3426

3427
	if (!cached && !previous)
3428 3429 3430 3431 3432 3433 3434
		st_mode = ce_mode_from_stat(*ce, st->st_mode);

	if (patch->is_new < 0)
		patch->is_new = 0;
	if (!patch->old_mode)
		patch->old_mode = st_mode;
	if ((st_mode ^ patch->old_mode) & S_IFMT)
3435
		return error(_("%s: wrong type"), old_name);
3436
	if (st_mode != patch->old_mode)
3437
		warning(_("%s has type %o, expected %o"),
3438
			old_name, st_mode, patch->old_mode);
3439
	if (!patch->new_mode && !patch->is_delete)
3440
		patch->new_mode = st_mode;
3441 3442 3443 3444 3445
	return 0;

 is_new:
	patch->is_new = 1;
	patch->is_delete = 0;
3446
	free(patch->old_name);
3447 3448 3449 3450
	patch->old_name = NULL;
	return 0;
}

3451 3452 3453 3454 3455

#define EXISTS_IN_INDEX 1
#define EXISTS_IN_WORKTREE 2

static int check_to_create(const char *new_name, int ok_if_exists)
3456 3457
{
	struct stat nst;
3458 3459 3460 3461 3462 3463 3464 3465

	if (check_index &&
	    cache_name_pos(new_name, strlen(new_name)) >= 0 &&
	    !ok_if_exists)
		return EXISTS_IN_INDEX;
	if (cached)
		return 0;

3466 3467 3468 3469 3470 3471 3472 3473 3474 3475 3476 3477 3478
	if (!lstat(new_name, &nst)) {
		if (S_ISDIR(nst.st_mode) || ok_if_exists)
			return 0;
		/*
		 * A leading component of new_name might be a symlink
		 * that is going to be removed with this patch, but
		 * still pointing at somewhere that has the path.
		 * In such a case, path "new_name" does not exist as
		 * far as git is concerned.
		 */
		if (has_symlink_leading_path(new_name, strlen(new_name)))
			return 0;

3479 3480
		return EXISTS_IN_WORKTREE;
	} else if ((errno != ENOENT) && (errno != ENOTDIR)) {
3481
		return error("%s: %s", new_name, strerror(errno));
3482
	}
3483 3484 3485
	return 0;
}

3486 3487 3488 3489
/*
 * Check and apply the patch in-core; leave the result in patch->result
 * for the caller to write it out to the final destination.
 */
3490
static int check_patch(struct patch *patch)
3491 3492 3493 3494 3495 3496
{
	struct stat st;
	const char *old_name = patch->old_name;
	const char *new_name = patch->new_name;
	const char *name = old_name ? old_name : new_name;
	struct cache_entry *ce = NULL;
3497
	struct patch *tpatch;
3498 3499 3500 3501 3502 3503 3504 3505 3506 3507
	int ok_if_exists;
	int status;

	patch->rejected = 1; /* we will drop this after we succeed */

	status = check_preimage(patch, &ce, &st);
	if (status)
		return status;
	old_name = patch->old_name;

3508 3509 3510 3511 3512 3513 3514 3515 3516
	/*
	 * A type-change diff is always split into a patch to delete
	 * old, immediately followed by a patch to create new (see
	 * diff.c::run_diff()); in such a case it is Ok that the entry
	 * to be deleted by the previous patch is still in the working
	 * tree and in the index.
	 *
	 * A patch to swap-rename between A and B would first rename A
	 * to B and then rename B to A.  While applying the first one,
3517
	 * the presence of B should not stop A from getting renamed to
3518 3519 3520 3521
	 * B; ask to_be_deleted() about the later rename.  Removal of
	 * B and rename from A to B is handled the same way by asking
	 * was_deleted().
	 */
3522
	if ((tpatch = in_fn_table(new_name)) &&
3523
	    (was_deleted(tpatch) || to_be_deleted(tpatch)))
3524 3525 3526 3527
		ok_if_exists = 1;
	else
		ok_if_exists = 0;

3528
	if (new_name &&
3529
	    ((0 < patch->is_new) || patch->is_rename || patch->is_copy)) {
3530 3531
		int err = check_to_create(new_name, ok_if_exists);

3532 3533 3534
		if (err && threeway) {
			patch->direct_to_threeway = 1;
		} else switch (err) {
3535 3536 3537
		case 0:
			break; /* happy */
		case EXISTS_IN_INDEX:
3538
			return error(_("%s: already exists in index"), new_name);
3539 3540 3541 3542 3543 3544
			break;
		case EXISTS_IN_WORKTREE:
			return error(_("%s: already exists in working directory"),
				     new_name);
		default:
			return err;
3545
		}
3546

3547
		if (!patch->new_mode) {
3548
			if (0 < patch->is_new)
3549 3550 3551 3552
				patch->new_mode = S_IFREG | 0644;
			else
				patch->new_mode = patch->old_mode;
		}
3553
	}
3554 3555 3556 3557 3558

	if (new_name && old_name) {
		int same = !strcmp(old_name, new_name);
		if (!patch->new_mode)
			patch->new_mode = patch->old_mode;
3559 3560 3561 3562 3563 3564 3565 3566 3567 3568 3569 3570
		if ((patch->old_mode ^ patch->new_mode) & S_IFMT) {
			if (same)
				return error(_("new mode (%o) of %s does not "
					       "match old mode (%o)"),
					patch->new_mode, new_name,
					patch->old_mode);
			else
				return error(_("new mode (%o) of %s does not "
					       "match old mode (%o) of %s"),
					patch->new_mode, new_name,
					patch->old_mode, old_name);
		}
3571
	}
3572

3573
	if (apply_data(patch, &st, ce) < 0)
3574
		return error(_("%s: patch does not apply"), name);
J
Junio C Hamano 已提交
3575
	patch->rejected = 0;
3576
	return 0;
3577 3578
}

3579
static int check_patch_list(struct patch *patch)
3580
{
3581
	int err = 0;
3582

3583
	prepare_fn_table(patch);
3584
	while (patch) {
J
Junio C Hamano 已提交
3585 3586
		if (apply_verbosely)
			say_patch_name(stderr,
3587
				       _("Checking patch %s..."), patch);
3588 3589
		err |= check_patch(patch);
		patch = patch->next;
3590
	}
3591
	return err;
3592 3593
}

3594 3595 3596 3597 3598 3599 3600 3601 3602 3603 3604 3605 3606 3607
/* This function tries to read the sha1 from the current index */
static int get_current_sha1(const char *path, unsigned char *sha1)
{
	int pos;

	if (read_cache() < 0)
		return -1;
	pos = cache_name_pos(path, strlen(path));
	if (pos < 0)
		return -1;
	hashcpy(sha1, active_cache[pos]->sha1);
	return 0;
}

3608 3609 3610 3611 3612 3613 3614 3615 3616 3617 3618 3619 3620 3621 3622 3623 3624 3625 3626 3627 3628 3629
static int preimage_sha1_in_gitlink_patch(struct patch *p, unsigned char sha1[20])
{
	/*
	 * A usable gitlink patch has only one fragment (hunk) that looks like:
	 * @@ -1 +1 @@
	 * -Subproject commit <old sha1>
	 * +Subproject commit <new sha1>
	 * or
	 * @@ -1 +0,0 @@
	 * -Subproject commit <old sha1>
	 * for a removal patch.
	 */
	struct fragment *hunk = p->fragments;
	static const char heading[] = "-Subproject commit ";
	char *preimage;

	if (/* does the patch have only one hunk? */
	    hunk && !hunk->next &&
	    /* is its preimage one line? */
	    hunk->oldpos == 1 && hunk->oldlines == 1 &&
	    /* does preimage begin with the heading? */
	    (preimage = memchr(hunk->patch, '\n', hunk->size)) != NULL &&
3630
	    starts_with(++preimage, heading) &&
3631 3632 3633 3634
	    /* does it record full SHA-1? */
	    !get_sha1_hex(preimage + sizeof(heading) - 1, sha1) &&
	    preimage[sizeof(heading) + 40 - 1] == '\n' &&
	    /* does the abbreviated name on the index line agree with it? */
3635
	    starts_with(preimage + sizeof(heading) - 1, p->old_sha1_prefix))
3636 3637 3638 3639 3640 3641
		return 0; /* it all looks fine */

	/* we may have full object name on the index line */
	return get_sha1_hex(p->old_sha1_prefix, sha1);
}

3642 3643
/* Build an index that contains the just the files needed for a 3way merge */
static void build_fake_ancestor(struct patch *list, const char *filename)
3644 3645
{
	struct patch *patch;
3646
	struct index_state result = { NULL };
3647
	int fd;
3648 3649 3650 3651 3652 3653

	/* Once we start supporting the reverse patch, it may be
	 * worth showing the new sha1 prefix, but until then...
	 */
	for (patch = list; patch; patch = patch->next) {
		unsigned char sha1[20];
3654
		struct cache_entry *ce;
3655 3656 3657
		const char *name;

		name = patch->old_name ? patch->old_name : patch->new_name;
3658
		if (0 < patch->is_new)
3659
			continue;
3660

3661
		if (S_ISGITLINK(patch->old_mode)) {
3662 3663 3664 3665
			if (!preimage_sha1_in_gitlink_patch(patch, sha1))
				; /* ok, the textual part looks sane */
			else
				die("sha1 information is lacking or useless for submoule %s",
3666 3667
				    name);
		} else if (!get_sha1_blob(patch->old_sha1_prefix, sha1)) {
3668 3669 3670 3671 3672 3673 3674 3675 3676 3677 3678
			; /* ok */
		} else if (!patch->lines_added && !patch->lines_deleted) {
			/* mode-only change: update the current */
			if (get_current_sha1(patch->old_name, sha1))
				die("mode change for %s, which is not "
				    "in current HEAD", name);
		} else
			die("sha1 information is lacking or useless "
			    "(%s).", name);

		ce = make_cache_entry(patch->old_mode, sha1, name, 0, 0);
3679
		if (!ce)
3680
			die(_("make_cache_entry failed for path '%s'"), name);
3681 3682
		if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
			die ("Could not add %s to temporary index", name);
3683
	}
3684 3685 3686 3687 3688 3689

	fd = open(filename, O_WRONLY | O_CREAT, 0666);
	if (fd < 0 || write_index(&result, fd) || close(fd))
		die ("Could not write temporary index to %s", filename);

	discard_index(&result);
3690 3691
}

3692 3693 3694 3695 3696 3697 3698 3699 3700 3701 3702
static void stat_patch_list(struct patch *patch)
{
	int files, adds, dels;

	for (files = adds = dels = 0 ; patch ; patch = patch->next) {
		files++;
		adds += patch->lines_added;
		dels += patch->lines_deleted;
		show_stats(patch);
	}

3703
	print_stat_summary(stdout, files, adds, dels);
3704 3705
}

J
Junio C Hamano 已提交
3706 3707 3708 3709
static void numstat_patch_list(struct patch *patch)
{
	for ( ; patch; patch = patch->next) {
		const char *name;
3710
		name = patch->new_name ? patch->new_name : patch->old_name;
3711 3712 3713
		if (patch->is_binary)
			printf("-\t-\t");
		else
3714 3715
			printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
		write_name_quoted(name, stdout, line_termination);
J
Junio C Hamano 已提交
3716 3717 3718
	}
}

3719 3720 3721 3722 3723 3724 3725 3726 3727 3728 3729 3730 3731 3732 3733 3734 3735 3736 3737 3738 3739 3740 3741 3742 3743 3744 3745 3746 3747 3748 3749 3750 3751 3752 3753 3754 3755 3756 3757 3758 3759 3760 3761 3762 3763
static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
{
	if (mode)
		printf(" %s mode %06o %s\n", newdelete, mode, name);
	else
		printf(" %s %s\n", newdelete, name);
}

static void show_mode_change(struct patch *p, int show_name)
{
	if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
		if (show_name)
			printf(" mode change %06o => %06o %s\n",
			       p->old_mode, p->new_mode, p->new_name);
		else
			printf(" mode change %06o => %06o\n",
			       p->old_mode, p->new_mode);
	}
}

static void show_rename_copy(struct patch *p)
{
	const char *renamecopy = p->is_rename ? "rename" : "copy";
	const char *old, *new;

	/* Find common prefix */
	old = p->old_name;
	new = p->new_name;
	while (1) {
		const char *slash_old, *slash_new;
		slash_old = strchr(old, '/');
		slash_new = strchr(new, '/');
		if (!slash_old ||
		    !slash_new ||
		    slash_old - old != slash_new - new ||
		    memcmp(old, new, slash_new - new))
			break;
		old = slash_old + 1;
		new = slash_new + 1;
	}
	/* p->old_name thru old is the common prefix, and old and new
	 * through the end of names are renames
	 */
	if (old != p->old_name)
		printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
3764
		       (int)(old - p->old_name), p->old_name,
3765 3766 3767 3768 3769 3770 3771 3772 3773 3774 3775 3776 3777 3778 3779 3780 3781 3782 3783 3784 3785 3786 3787 3788 3789 3790 3791 3792 3793 3794 3795 3796
		       old, new, p->score);
	else
		printf(" %s %s => %s (%d%%)\n", renamecopy,
		       p->old_name, p->new_name, p->score);
	show_mode_change(p, 0);
}

static void summary_patch_list(struct patch *patch)
{
	struct patch *p;

	for (p = patch; p; p = p->next) {
		if (p->is_new)
			show_file_mode_name("create", p->new_mode, p->new_name);
		else if (p->is_delete)
			show_file_mode_name("delete", p->old_mode, p->old_name);
		else {
			if (p->is_rename || p->is_copy)
				show_rename_copy(p);
			else {
				if (p->score) {
					printf(" rewrite %s (%d%%)\n",
					       p->new_name, p->score);
					show_mode_change(p, 0);
				}
				else
					show_mode_change(p, 1);
			}
		}
	}
}

3797 3798 3799 3800 3801 3802 3803
static void patch_stats(struct patch *patch)
{
	int lines = patch->lines_added + patch->lines_deleted;

	if (lines > max_change)
		max_change = lines;
	if (patch->old_name) {
3804 3805 3806
		int len = quote_c_style(patch->old_name, NULL, NULL, 0);
		if (!len)
			len = strlen(patch->old_name);
3807 3808 3809 3810
		if (len > max_len)
			max_len = len;
	}
	if (patch->new_name) {
3811 3812 3813
		int len = quote_c_style(patch->new_name, NULL, NULL, 0);
		if (!len)
			len = strlen(patch->new_name);
3814 3815 3816
		if (len > max_len)
			max_len = len;
	}
3817 3818
}

3819
static void remove_file(struct patch *patch, int rmdir_empty)
3820
{
3821
	if (update_index) {
3822
		if (remove_file_from_cache(patch->old_name) < 0)
3823
			die(_("unable to remove %s from index"), patch->old_name);
3824
	}
3825
	if (!cached) {
3826
		if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {
3827
			remove_path(patch->old_name);
3828 3829
		}
	}
3830 3831 3832 3833 3834 3835 3836 3837 3838
}

static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
{
	struct stat st;
	struct cache_entry *ce;
	int namelen = strlen(path);
	unsigned ce_size = cache_entry_size(namelen);

3839
	if (!update_index)
3840 3841
		return;

3842
	ce = xcalloc(1, ce_size);
3843 3844
	memcpy(ce->name, path, namelen);
	ce->ce_mode = create_ce_mode(mode);
3845 3846
	ce->ce_flags = create_ce_flags(0);
	ce->ce_namelen = namelen;
3847 3848 3849 3850
	if (S_ISGITLINK(mode)) {
		const char *s = buf;

		if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
3851
			die(_("corrupt patch for submodule %s"), path);
3852 3853 3854
	} else {
		if (!cached) {
			if (lstat(path, &st) < 0)
3855
				die_errno(_("unable to stat newly created file '%s'"),
3856
					  path);
3857 3858 3859
			fill_stat_cache_info(ce, &st);
		}
		if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
3860
			die(_("unable to create backing store for newly created file %s"), path);
3861
	}
3862
	if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
3863
		die(_("unable to add cache entry for %s"), path);
3864 3865
}

3866 3867
static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
{
3868
	int fd;
3869
	struct strbuf nbuf = STRBUF_INIT;
3870

3871 3872 3873 3874 3875 3876 3877
	if (S_ISGITLINK(mode)) {
		struct stat st;
		if (!lstat(path, &st) && S_ISDIR(st.st_mode))
			return 0;
		return mkdir(path, 0777);
	}

3878
	if (has_symlinks && S_ISLNK(mode))
3879 3880 3881
		/* Although buf:size is counted string, it also is NUL
		 * terminated.
		 */
3882
		return symlink(buf, path);
3883 3884 3885 3886 3887

	fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
	if (fd < 0)
		return -1;

3888 3889 3890
	if (convert_to_working_tree(path, buf, size, &nbuf)) {
		size = nbuf.len;
		buf  = nbuf.buf;
3891
	}
3892 3893
	write_or_die(fd, buf, size);
	strbuf_release(&nbuf);
3894

3895
	if (close(fd) < 0)
3896
		die_errno(_("closing file '%s'"), path);
3897 3898 3899
	return 0;
}

3900 3901 3902 3903 3904
/*
 * We optimistically assume that the directories exist,
 * which is true 99% of the time anyway. If they don't,
 * we create them and try again.
 */
3905
static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
3906
{
3907 3908
	if (cached)
		return;
3909 3910
	if (!try_create_file(path, mode, buf, size))
		return;
3911

3912
	if (errno == ENOENT) {
3913 3914
		if (safe_create_leading_directories(path))
			return;
3915 3916
		if (!try_create_file(path, mode, buf, size))
			return;
3917 3918
	}

J
Johannes Schindelin 已提交
3919
	if (errno == EEXIST || errno == EACCES) {
3920 3921 3922 3923
		/* We may be trying to create a file where a directory
		 * used to be.
		 */
		struct stat st;
3924
		if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
3925 3926 3927
			errno = EEXIST;
	}

3928 3929
	if (errno == EEXIST) {
		unsigned int nr = getpid();
3930

3931
		for (;;) {
3932 3933
			char newpath[PATH_MAX];
			mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
3934 3935 3936
			if (!try_create_file(newpath, mode, buf, size)) {
				if (!rename(newpath, path))
					return;
3937
				unlink_or_warn(newpath);
3938 3939 3940 3941
				break;
			}
			if (errno != EEXIST)
				break;
3942 3943
			++nr;
		}
3944
	}
3945
	die_errno(_("unable to write file '%s' mode %o"), path, mode);
3946 3947
}

3948 3949 3950 3951 3952 3953 3954 3955 3956 3957 3958 3959 3960 3961 3962 3963 3964 3965 3966
static void add_conflicted_stages_file(struct patch *patch)
{
	int stage, namelen;
	unsigned ce_size, mode;
	struct cache_entry *ce;

	if (!update_index)
		return;
	namelen = strlen(patch->new_name);
	ce_size = cache_entry_size(namelen);
	mode = patch->new_mode ? patch->new_mode : (S_IFREG | 0644);

	remove_file_from_cache(patch->new_name);
	for (stage = 1; stage < 4; stage++) {
		if (is_null_sha1(patch->threeway_stage[stage - 1]))
			continue;
		ce = xcalloc(1, ce_size);
		memcpy(ce->name, patch->new_name, namelen);
		ce->ce_mode = create_ce_mode(mode);
3967 3968
		ce->ce_flags = create_ce_flags(stage);
		ce->ce_namelen = namelen;
3969 3970 3971 3972 3973 3974
		hashcpy(ce->sha1, patch->threeway_stage[stage - 1]);
		if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
			die(_("unable to add cache entry for %s"), patch->new_name);
	}
}

3975 3976
static void create_file(struct patch *patch)
{
3977
	char *path = patch->new_name;
3978 3979 3980 3981 3982 3983
	unsigned mode = patch->new_mode;
	unsigned long size = patch->resultsize;
	char *buf = patch->result;

	if (!mode)
		mode = S_IFREG | 0644;
3984
	create_one_file(path, mode, buf, size);
3985 3986 3987 3988 3989

	if (patch->conflicted_threeway)
		add_conflicted_stages_file(patch);
	else
		add_index_file(path, mode, buf, size);
3990 3991
}

3992 3993
/* phase zero is to remove, phase one is to create */
static void write_out_one_result(struct patch *patch, int phase)
3994 3995
{
	if (patch->is_delete > 0) {
3996
		if (phase == 0)
3997
			remove_file(patch, 1);
3998 3999 4000
		return;
	}
	if (patch->is_new > 0 || patch->is_copy) {
4001 4002
		if (phase == 1)
			create_file(patch);
4003 4004 4005 4006 4007 4008
		return;
	}
	/*
	 * Rename or modification boils down to the same
	 * thing: remove the old, write the new
	 */
4009
	if (phase == 0)
4010
		remove_file(patch, patch->is_rename);
4011
	if (phase == 1)
J
Junio C Hamano 已提交
4012
		create_file(patch);
4013 4014
}

J
Junio C Hamano 已提交
4015 4016
static int write_out_one_reject(struct patch *patch)
{
4017 4018
	FILE *rej;
	char namebuf[PATH_MAX];
J
Junio C Hamano 已提交
4019
	struct fragment *frag;
4020
	int cnt = 0;
4021
	struct strbuf sb = STRBUF_INIT;
J
Junio C Hamano 已提交
4022

4023
	for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
J
Junio C Hamano 已提交
4024 4025
		if (!frag->rejected)
			continue;
4026 4027 4028
		cnt++;
	}

J
Junio C Hamano 已提交
4029 4030 4031
	if (!cnt) {
		if (apply_verbosely)
			say_patch_name(stderr,
4032
				       _("Applied patch %s cleanly."), patch);
4033
		return 0;
J
Junio C Hamano 已提交
4034
	}
4035 4036 4037 4038 4039

	/* This should not happen, because a removal patch that leaves
	 * contents are marked "rejected" at the patch level.
	 */
	if (!patch->new_name)
4040
		die(_("internal error"));
4041

J
Junio C Hamano 已提交
4042
	/* Say this even without --verbose */
4043 4044 4045 4046 4047 4048
	strbuf_addf(&sb, Q_("Applying patch %%s with %d reject...",
			    "Applying patch %%s with %d rejects...",
			    cnt),
		    cnt);
	say_patch_name(stderr, sb.buf, patch);
	strbuf_release(&sb);
J
Junio C Hamano 已提交
4049

4050 4051 4052
	cnt = strlen(patch->new_name);
	if (ARRAY_SIZE(namebuf) <= cnt + 5) {
		cnt = ARRAY_SIZE(namebuf) - 5;
4053
		warning(_("truncating .rej filename to %.*s.rej"),
4054 4055 4056 4057 4058 4059 4060
			cnt - 1, patch->new_name);
	}
	memcpy(namebuf, patch->new_name, cnt);
	memcpy(namebuf + cnt, ".rej", 5);

	rej = fopen(namebuf, "w");
	if (!rej)
4061
		return error(_("cannot open %s: %s"), namebuf, strerror(errno));
4062 4063

	/* Normal git tools never deal with .rej, so do not pretend
J
Justin Lebar 已提交
4064
	 * this is a git patch by saying --git or giving extended
4065 4066 4067 4068 4069
	 * headers.  While at it, maybe please "kompare" that wants
	 * the trailing TAB and some garbage at the end of line ;-).
	 */
	fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
		patch->new_name, patch->new_name);
4070
	for (cnt = 1, frag = patch->fragments;
4071 4072 4073
	     frag;
	     cnt++, frag = frag->next) {
		if (!frag->rejected) {
4074
			fprintf_ln(stderr, _("Hunk #%d applied cleanly."), cnt);
4075
			continue;
J
Junio C Hamano 已提交
4076
		}
4077
		fprintf_ln(stderr, _("Rejected hunk #%d."), cnt);
4078
		fprintf(rej, "%.*s", frag->size, frag->patch);
J
Junio C Hamano 已提交
4079
		if (frag->patch[frag->size-1] != '\n')
4080
			fputc('\n', rej);
J
Junio C Hamano 已提交
4081
	}
4082 4083
	fclose(rej);
	return -1;
4084 4085
}

4086
static int write_out_results(struct patch *list)
4087
{
4088
	int phase;
J
Junio C Hamano 已提交
4089 4090
	int errs = 0;
	struct patch *l;
4091
	struct string_list cpath = STRING_LIST_INIT_DUP;
4092 4093

	for (phase = 0; phase < 2; phase++) {
J
Junio C Hamano 已提交
4094
		l = list;
4095
		while (l) {
J
Junio C Hamano 已提交
4096 4097
			if (l->rejected)
				errs = 1;
4098
			else {
J
Junio C Hamano 已提交
4099
				write_out_one_result(l, phase);
4100 4101 4102 4103 4104 4105 4106 4107
				if (phase == 1) {
					if (write_out_one_reject(l))
						errs = 1;
					if (l->conflicted_threeway) {
						string_list_append(&cpath, l->new_name);
						errs = 1;
					}
				}
J
Junio C Hamano 已提交
4108
			}
4109 4110
			l = l->next;
		}
4111
	}
4112 4113 4114 4115 4116 4117 4118 4119

	if (cpath.nr) {
		struct string_list_item *item;

		sort_string_list(&cpath);
		for_each_string_list_item(item, &cpath)
			fprintf(stderr, "U %s\n", item->string);
		string_list_clear(&cpath, 0);
4120 4121

		rerere(0);
4122 4123
	}

J
Junio C Hamano 已提交
4124
	return errs;
4125 4126
}

4127
static struct lock_file lock_file;
4128

J
Junio C Hamano 已提交
4129 4130 4131 4132 4133 4134
static struct string_list limit_by_name;
static int has_include;
static void add_name_limit(const char *name, int exclude)
{
	struct string_list_item *it;

4135
	it = string_list_append(&limit_by_name, name);
J
Junio C Hamano 已提交
4136 4137
	it->util = exclude ? NULL : (void *) 1;
}
4138 4139 4140

static int use_patch(struct patch *p)
{
J
Jason Riedy 已提交
4141
	const char *pathname = p->new_name ? p->new_name : p->old_name;
J
Junio C Hamano 已提交
4142 4143 4144
	int i;

	/* Paths outside are not touched regardless of "--include" */
4145 4146 4147 4148 4149 4150
	if (0 < prefix_length) {
		int pathlen = strlen(pathname);
		if (pathlen <= prefix_length ||
		    memcmp(prefix, pathname, prefix_length))
			return 0;
	}
J
Junio C Hamano 已提交
4151 4152 4153 4154

	/* See if it matches any of exclude/include rule */
	for (i = 0; i < limit_by_name.nr; i++) {
		struct string_list_item *it = &limit_by_name.items[i];
4155
		if (!wildmatch(it->string, pathname, 0, NULL))
J
Junio C Hamano 已提交
4156 4157 4158 4159 4160 4161 4162 4163 4164
			return (it->util != NULL);
	}

	/*
	 * If we had any include, a path that does not match any rule is
	 * not used.  Otherwise, we saw bunch of exclude rules (or none)
	 * and such a path is used.
	 */
	return !has_include;
4165 4166
}

J
Junio C Hamano 已提交
4167

4168
static void prefix_one(char **name)
4169
{
4170 4171 4172 4173 4174
	char *old_name = *name;
	if (!old_name)
		return;
	*name = xstrdup(prefix_filename(prefix, prefix_length, *name));
	free(old_name);
4175 4176 4177 4178
}

static void prefix_patches(struct patch *p)
{
4179
	if (!prefix || p->is_toplevel_relative)
4180 4181
		return;
	for ( ; p; p = p->next) {
4182 4183
		prefix_one(&p->new_name);
		prefix_one(&p->old_name);
4184 4185 4186
	}
}

4187 4188 4189 4190
#define INACCURATE_EOF	(1<<0)
#define RECOUNT		(1<<1)

static int apply_patch(int fd, const char *filename, int options)
L
Linus Torvalds 已提交
4191
{
4192
	size_t offset;
4193
	struct strbuf buf = STRBUF_INIT; /* owns the patch text */
4194
	struct patch *list = NULL, **listp = &list;
4195
	int skipped_patch = 0;
L
Linus Torvalds 已提交
4196

4197
	patch_input_file = filename;
4198
	read_patch_file(&buf, fd);
L
Linus Torvalds 已提交
4199
	offset = 0;
4200
	while (offset < buf.len) {
4201 4202 4203
		struct patch *patch;
		int nr;

4204
		patch = xcalloc(1, sizeof(*patch));
4205 4206
		patch->inaccurate_eof = !!(options & INACCURATE_EOF);
		patch->recount =  !!(options & RECOUNT);
4207
		nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
L
Linus Torvalds 已提交
4208 4209
		if (nr < 0)
			break;
4210
		if (apply_in_reverse)
J
Johannes Schindelin 已提交
4211
			reverse_patches(patch);
4212 4213
		if (prefix)
			prefix_patches(patch);
4214 4215 4216 4217
		if (use_patch(patch)) {
			patch_stats(patch);
			*listp = patch;
			listp = &patch->next;
4218 4219
		}
		else {
4220
			free_patch(patch);
4221 4222
			skipped_patch++;
		}
L
Linus Torvalds 已提交
4223 4224
		offset += nr;
	}
4225

4226
	if (!list && !skipped_patch)
4227
		die(_("unrecognized input"));
4228

4229
	if (whitespace_error && (ws_error_action == die_on_ws_error))
4230 4231
		apply = 0;

4232 4233
	update_index = check_index && apply;
	if (update_index && newfd < 0)
4234 4235
		newfd = hold_locked_index(&lock_file, 1);

4236 4237
	if (check_index) {
		if (read_cache() < 0)
4238
			die(_("unable to read index file"));
4239 4240
	}

J
Junio C Hamano 已提交
4241 4242 4243
	if ((check || apply) &&
	    check_patch_list(list) < 0 &&
	    !apply_with_reject)
4244 4245
		exit(1);

4246 4247 4248 4249 4250 4251
	if (apply && write_out_results(list)) {
		if (apply_with_reject)
			exit(1);
		/* with --3way, we still need to write the index out */
		return 1;
	}
4252

4253 4254
	if (fake_ancestor)
		build_fake_ancestor(list, fake_ancestor);
4255

4256 4257
	if (diffstat)
		stat_patch_list(list);
4258

J
Junio C Hamano 已提交
4259 4260 4261
	if (numstat)
		numstat_patch_list(list);

4262 4263 4264
	if (summary)
		summary_patch_list(list);

4265
	free_patch_list(list);
4266
	strbuf_release(&buf);
4267
	string_list_clear(&fn_table, 0);
L
Linus Torvalds 已提交
4268 4269 4270
	return 0;
}

4271
static int git_apply_config(const char *var, const char *value, void *cb)
4272
{
4273 4274
	if (!strcmp(var, "apply.whitespace"))
		return git_config_string(&apply_default_whitespace, var, value);
4275 4276
	else if (!strcmp(var, "apply.ignorewhitespace"))
		return git_config_string(&apply_default_ignorewhitespace, var, value);
4277
	return git_default_config(var, value, cb);
4278 4279
}

M
Miklos Vajna 已提交
4280 4281 4282 4283 4284 4285 4286 4287 4288 4289 4290 4291 4292 4293 4294 4295 4296 4297 4298 4299 4300 4301 4302 4303 4304 4305 4306 4307 4308 4309 4310 4311 4312
static int option_parse_exclude(const struct option *opt,
				const char *arg, int unset)
{
	add_name_limit(arg, 1);
	return 0;
}

static int option_parse_include(const struct option *opt,
				const char *arg, int unset)
{
	add_name_limit(arg, 0);
	has_include = 1;
	return 0;
}

static int option_parse_p(const struct option *opt,
			  const char *arg, int unset)
{
	p_value = atoi(arg);
	p_value_known = 1;
	return 0;
}

static int option_parse_z(const struct option *opt,
			  const char *arg, int unset)
{
	if (unset)
		line_termination = '\n';
	else
		line_termination = 0;
	return 0;
}

4313 4314 4315 4316 4317 4318 4319 4320 4321 4322
static int option_parse_space_change(const struct option *opt,
			  const char *arg, int unset)
{
	if (unset)
		ws_ignore_action = ignore_ws_none;
	else
		ws_ignore_action = ignore_ws_change;
	return 0;
}

M
Miklos Vajna 已提交
4323 4324 4325 4326 4327 4328 4329 4330 4331 4332 4333 4334 4335 4336 4337 4338 4339 4340 4341 4342 4343 4344 4345
static int option_parse_whitespace(const struct option *opt,
				   const char *arg, int unset)
{
	const char **whitespace_option = opt->value;

	*whitespace_option = arg;
	parse_whitespace_option(arg);
	return 0;
}

static int option_parse_directory(const struct option *opt,
				  const char *arg, int unset)
{
	root_len = strlen(arg);
	if (root_len && arg[root_len - 1] != '/') {
		char *new_root;
		root = new_root = xmalloc(root_len + 2);
		strcpy(new_root, arg);
		strcpy(new_root + root_len++, "/");
	} else
		root = arg;
	return 0;
}
4346

4347
int cmd_apply(int argc, const char **argv, const char *prefix_)
L
Linus Torvalds 已提交
4348 4349
{
	int i;
J
Junio C Hamano 已提交
4350
	int errs = 0;
4351
	int is_not_gitdir = !startup_info->have_repository;
M
Miklos Vajna 已提交
4352
	int force_apply = 0;
4353

4354
	const char *whitespace_option = NULL;
L
Linus Torvalds 已提交
4355

M
Miklos Vajna 已提交
4356
	struct option builtin_apply_options[] = {
4357 4358
		{ OPTION_CALLBACK, 0, "exclude", NULL, N_("path"),
			N_("don't apply changes matching the given path"),
M
Miklos Vajna 已提交
4359
			0, option_parse_exclude },
4360 4361
		{ OPTION_CALLBACK, 0, "include", NULL, N_("path"),
			N_("apply changes matching the given path"),
M
Miklos Vajna 已提交
4362
			0, option_parse_include },
4363 4364
		{ OPTION_CALLBACK, 'p', NULL, NULL, N_("num"),
			N_("remove <num> leading slashes from traditional diff paths"),
M
Miklos Vajna 已提交
4365
			0, option_parse_p },
4366
		OPT_BOOL(0, "no-add", &no_add,
4367
			N_("ignore additions made by the patch")),
4368
		OPT_BOOL(0, "stat", &diffstat,
4369
			N_("instead of applying the patch, output diffstat for the input")),
R
René Scharfe 已提交
4370 4371
		OPT_NOOP_NOARG(0, "allow-binary-replacement"),
		OPT_NOOP_NOARG(0, "binary"),
4372
		OPT_BOOL(0, "numstat", &numstat,
4373
			N_("show number of added and deleted lines in decimal notation")),
4374
		OPT_BOOL(0, "summary", &summary,
4375
			N_("instead of applying the patch, output a summary for the input")),
4376
		OPT_BOOL(0, "check", &check,
4377
			N_("instead of applying the patch, see if the patch is applicable")),
4378
		OPT_BOOL(0, "index", &check_index,
4379
			N_("make sure the patch is applicable to the current index")),
4380
		OPT_BOOL(0, "cached", &cached,
4381
			N_("apply a patch without touching the working tree")),
4382
		OPT_BOOL(0, "apply", &force_apply,
4383
			N_("also apply the patch (use with --stat/--summary/--check)")),
4384
		OPT_BOOL('3', "3way", &threeway,
J
Junio C Hamano 已提交
4385
			 N_( "attempt three-way merge if a patch does not apply")),
4386
		OPT_FILENAME(0, "build-fake-ancestor", &fake_ancestor,
4387
			N_("build a temporary index based on embedded index information")),
M
Miklos Vajna 已提交
4388
		{ OPTION_CALLBACK, 'z', NULL, NULL, NULL,
4389
			N_("paths are separated with NUL character"),
M
Miklos Vajna 已提交
4390 4391
			PARSE_OPT_NOARG, option_parse_z },
		OPT_INTEGER('C', NULL, &p_context,
4392 4393 4394
				N_("ensure at least <n> lines of context match")),
		{ OPTION_CALLBACK, 0, "whitespace", &whitespace_option, N_("action"),
			N_("detect new or modified lines that have whitespace errors"),
M
Miklos Vajna 已提交
4395
			0, option_parse_whitespace },
4396
		{ OPTION_CALLBACK, 0, "ignore-space-change", NULL, NULL,
4397
			N_("ignore changes in whitespace when finding context"),
4398 4399
			PARSE_OPT_NOARG, option_parse_space_change },
		{ OPTION_CALLBACK, 0, "ignore-whitespace", NULL, NULL,
4400
			N_("ignore changes in whitespace when finding context"),
4401
			PARSE_OPT_NOARG, option_parse_space_change },
4402
		OPT_BOOL('R', "reverse", &apply_in_reverse,
4403
			N_("apply the patch in reverse")),
4404
		OPT_BOOL(0, "unidiff-zero", &unidiff_zero,
4405
			N_("don't expect at least one line of context")),
4406
		OPT_BOOL(0, "reject", &apply_with_reject,
4407
			N_("leave the rejected hunks in corresponding *.rej files")),
4408
		OPT_BOOL(0, "allow-overlap", &allow_overlap,
4409 4410
			N_("allow overlapping hunks")),
		OPT__VERBOSE(&apply_verbosely, N_("be verbose")),
M
Miklos Vajna 已提交
4411
		OPT_BIT(0, "inaccurate-eof", &options,
4412
			N_("tolerate incorrectly detected missing new-line at the end of file"),
M
Miklos Vajna 已提交
4413 4414
			INACCURATE_EOF),
		OPT_BIT(0, "recount", &options,
4415
			N_("do not trust the line counts in the hunk headers"),
M
Miklos Vajna 已提交
4416
			RECOUNT),
4417 4418
		{ OPTION_CALLBACK, 0, "directory", NULL, N_("root"),
			N_("prepend <root> to all filenames"),
M
Miklos Vajna 已提交
4419 4420 4421 4422
			0, option_parse_directory },
		OPT_END()
	};

4423
	prefix = prefix_;
4424
	prefix_length = prefix ? strlen(prefix) : 0;
4425
	git_config(git_apply_config, NULL);
4426 4427
	if (apply_default_whitespace)
		parse_whitespace_option(apply_default_whitespace);
4428 4429
	if (apply_default_ignorewhitespace)
		parse_ignorewhitespace_option(apply_default_ignorewhitespace);
4430

4431
	argc = parse_options(argc, argv, prefix, builtin_apply_options,
M
Miklos Vajna 已提交
4432
			apply_usage, 0);
4433

4434 4435 4436 4437 4438 4439 4440 4441 4442
	if (apply_with_reject && threeway)
		die("--reject and --3way cannot be used together.");
	if (cached && threeway)
		die("--cached and --3way cannot be used together.");
	if (threeway) {
		if (is_not_gitdir)
			die(_("--3way outside a repository"));
		check_index = 1;
	}
M
Miklos Vajna 已提交
4443 4444 4445 4446 4447
	if (apply_with_reject)
		apply = apply_verbosely = 1;
	if (!force_apply && (diffstat || numstat || summary || check || fake_ancestor))
		apply = 0;
	if (check_index && is_not_gitdir)
4448
		die(_("--index outside a repository"));
M
Miklos Vajna 已提交
4449 4450
	if (cached) {
		if (is_not_gitdir)
4451
			die(_("--cached outside a repository"));
M
Miklos Vajna 已提交
4452 4453 4454
		check_index = 1;
	}
	for (i = 0; i < argc; i++) {
L
Linus Torvalds 已提交
4455 4456 4457 4458
		const char *arg = argv[i];
		int fd;

		if (!strcmp(arg, "-")) {
4459
			errs |= apply_patch(0, "<stdin>", options);
4460
			read_stdin = 0;
L
Linus Torvalds 已提交
4461
			continue;
4462
		} else if (0 < prefix_length)
4463 4464
			arg = prefix_filename(prefix, prefix_length, arg);

L
Linus Torvalds 已提交
4465 4466
		fd = open(arg, O_RDONLY);
		if (fd < 0)
4467
			die_errno(_("can't open patch '%s'"), arg);
4468
		read_stdin = 0;
4469
		set_default_whitespace_mode(whitespace_option);
4470
		errs |= apply_patch(fd, arg, options);
L
Linus Torvalds 已提交
4471 4472
		close(fd);
	}
4473
	set_default_whitespace_mode(whitespace_option);
4474
	if (read_stdin)
4475
		errs |= apply_patch(0, "<stdin>", options);
4476 4477 4478 4479 4480
	if (whitespace_error) {
		if (squelch_whitespace_errors &&
		    squelch_whitespace_errors < whitespace_error) {
			int squelched =
				whitespace_error - squelch_whitespace_errors;
4481 4482 4483 4484
			warning(Q_("squelched %d whitespace error",
				   "squelched %d whitespace errors",
				   squelched),
				squelched);
4485
		}
4486
		if (ws_error_action == die_on_ws_error)
4487 4488 4489 4490
			die(Q_("%d line adds whitespace errors.",
			       "%d lines add whitespace errors.",
			       whitespace_error),
			    whitespace_error);
4491
		if (applied_after_fixing_ws && apply)
4492 4493
			warning("%d line%s applied after"
				" fixing whitespace errors.",
4494 4495
				applied_after_fixing_ws,
				applied_after_fixing_ws == 1 ? "" : "s");
4496
		else if (whitespace_error)
4497 4498 4499 4500
			warning(Q_("%d line adds whitespace errors.",
				   "%d lines add whitespace errors.",
				   whitespace_error),
				whitespace_error);
4501
	}
4502

4503
	if (update_index) {
4504
		if (write_cache(newfd, active_cache, active_nr) ||
B
Brandon Casey 已提交
4505
		    commit_locked_index(&lock_file))
4506
			die(_("Unable to write new index file"));
4507 4508
	}

J
Junio C Hamano 已提交
4509
	return !!errs;
L
Linus Torvalds 已提交
4510
}