ie_match_stat(): do not ignore skip-worktree bit with CE_MATCH_IGNORE_VALID
[git/git.git] / builtin-apply.c
1 /*
2 * apply.c
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 *
6 * This applies patches on top of some (arbitrary) version of the SCM.
7 *
8 */
9 #include "cache.h"
10 #include "cache-tree.h"
11 #include "quote.h"
12 #include "blob.h"
13 #include "delta.h"
14 #include "builtin.h"
15 #include "string-list.h"
16 #include "dir.h"
17 #include "parse-options.h"
18
19 /*
20 * --check turns on checking that the working tree matches the
21 * files that are being modified, but doesn't apply the patch
22 * --stat does just a diffstat, and doesn't actually apply
23 * --numstat does numeric diffstat, and doesn't actually apply
24 * --index-info shows the old and new index info for paths if available.
25 * --index updates the cache as well.
26 * --cached updates only the cache without ever touching the working tree.
27 */
28 static const char *prefix;
29 static int prefix_length = -1;
30 static int newfd = -1;
31
32 static int unidiff_zero;
33 static int p_value = 1;
34 static int p_value_known;
35 static int check_index;
36 static int update_index;
37 static int cached;
38 static int diffstat;
39 static int numstat;
40 static int summary;
41 static int check;
42 static int apply = 1;
43 static int apply_in_reverse;
44 static int apply_with_reject;
45 static int apply_verbosely;
46 static int no_add;
47 static const char *fake_ancestor;
48 static int line_termination = '\n';
49 static unsigned int p_context = UINT_MAX;
50 static const char * const apply_usage[] = {
51 "git apply [options] [<patch>...]",
52 NULL
53 };
54
55 static enum ws_error_action {
56 nowarn_ws_error,
57 warn_on_ws_error,
58 die_on_ws_error,
59 correct_ws_error,
60 } ws_error_action = warn_on_ws_error;
61 static int whitespace_error;
62 static int squelch_whitespace_errors = 5;
63 static int applied_after_fixing_ws;
64 static const char *patch_input_file;
65 static const char *root;
66 static int root_len;
67 static int read_stdin = 1;
68 static int options;
69
70 static void parse_whitespace_option(const char *option)
71 {
72 if (!option) {
73 ws_error_action = warn_on_ws_error;
74 return;
75 }
76 if (!strcmp(option, "warn")) {
77 ws_error_action = warn_on_ws_error;
78 return;
79 }
80 if (!strcmp(option, "nowarn")) {
81 ws_error_action = nowarn_ws_error;
82 return;
83 }
84 if (!strcmp(option, "error")) {
85 ws_error_action = die_on_ws_error;
86 return;
87 }
88 if (!strcmp(option, "error-all")) {
89 ws_error_action = die_on_ws_error;
90 squelch_whitespace_errors = 0;
91 return;
92 }
93 if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
94 ws_error_action = correct_ws_error;
95 return;
96 }
97 die("unrecognized whitespace option '%s'", option);
98 }
99
100 static void set_default_whitespace_mode(const char *whitespace_option)
101 {
102 if (!whitespace_option && !apply_default_whitespace)
103 ws_error_action = (apply ? warn_on_ws_error : nowarn_ws_error);
104 }
105
106 /*
107 * For "diff-stat" like behaviour, we keep track of the biggest change
108 * we've seen, and the longest filename. That allows us to do simple
109 * scaling.
110 */
111 static int max_change, max_len;
112
113 /*
114 * Various "current state", notably line numbers and what
115 * file (and how) we're patching right now.. The "is_xxxx"
116 * things are flags, where -1 means "don't know yet".
117 */
118 static int linenr = 1;
119
120 /*
121 * This represents one "hunk" from a patch, starting with
122 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
123 * patch text is pointed at by patch, and its byte length
124 * is stored in size. leading and trailing are the number
125 * of context lines.
126 */
127 struct fragment {
128 unsigned long leading, trailing;
129 unsigned long oldpos, oldlines;
130 unsigned long newpos, newlines;
131 const char *patch;
132 int size;
133 int rejected;
134 struct fragment *next;
135 };
136
137 /*
138 * When dealing with a binary patch, we reuse "leading" field
139 * to store the type of the binary hunk, either deflated "delta"
140 * or deflated "literal".
141 */
142 #define binary_patch_method leading
143 #define BINARY_DELTA_DEFLATED 1
144 #define BINARY_LITERAL_DEFLATED 2
145
146 /*
147 * This represents a "patch" to a file, both metainfo changes
148 * such as creation/deletion, filemode and content changes represented
149 * as a series of fragments.
150 */
151 struct patch {
152 char *new_name, *old_name, *def_name;
153 unsigned int old_mode, new_mode;
154 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */
155 int rejected;
156 unsigned ws_rule;
157 unsigned long deflate_origlen;
158 int lines_added, lines_deleted;
159 int score;
160 unsigned int is_toplevel_relative:1;
161 unsigned int inaccurate_eof:1;
162 unsigned int is_binary:1;
163 unsigned int is_copy:1;
164 unsigned int is_rename:1;
165 unsigned int recount:1;
166 struct fragment *fragments;
167 char *result;
168 size_t resultsize;
169 char old_sha1_prefix[41];
170 char new_sha1_prefix[41];
171 struct patch *next;
172 };
173
174 /*
175 * A line in a file, len-bytes long (includes the terminating LF,
176 * except for an incomplete line at the end if the file ends with
177 * one), and its contents hashes to 'hash'.
178 */
179 struct line {
180 size_t len;
181 unsigned hash : 24;
182 unsigned flag : 8;
183 #define LINE_COMMON 1
184 };
185
186 /*
187 * This represents a "file", which is an array of "lines".
188 */
189 struct image {
190 char *buf;
191 size_t len;
192 size_t nr;
193 size_t alloc;
194 struct line *line_allocated;
195 struct line *line;
196 };
197
198 /*
199 * Records filenames that have been touched, in order to handle
200 * the case where more than one patches touch the same file.
201 */
202
203 static struct string_list fn_table;
204
205 static uint32_t hash_line(const char *cp, size_t len)
206 {
207 size_t i;
208 uint32_t h;
209 for (i = 0, h = 0; i < len; i++) {
210 if (!isspace(cp[i])) {
211 h = h * 3 + (cp[i] & 0xff);
212 }
213 }
214 return h;
215 }
216
217 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
218 {
219 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
220 img->line_allocated[img->nr].len = len;
221 img->line_allocated[img->nr].hash = hash_line(bol, len);
222 img->line_allocated[img->nr].flag = flag;
223 img->nr++;
224 }
225
226 static void prepare_image(struct image *image, char *buf, size_t len,
227 int prepare_linetable)
228 {
229 const char *cp, *ep;
230
231 memset(image, 0, sizeof(*image));
232 image->buf = buf;
233 image->len = len;
234
235 if (!prepare_linetable)
236 return;
237
238 ep = image->buf + image->len;
239 cp = image->buf;
240 while (cp < ep) {
241 const char *next;
242 for (next = cp; next < ep && *next != '\n'; next++)
243 ;
244 if (next < ep)
245 next++;
246 add_line_info(image, cp, next - cp, 0);
247 cp = next;
248 }
249 image->line = image->line_allocated;
250 }
251
252 static void clear_image(struct image *image)
253 {
254 free(image->buf);
255 image->buf = NULL;
256 image->len = 0;
257 }
258
259 static void say_patch_name(FILE *output, const char *pre,
260 struct patch *patch, const char *post)
261 {
262 fputs(pre, output);
263 if (patch->old_name && patch->new_name &&
264 strcmp(patch->old_name, patch->new_name)) {
265 quote_c_style(patch->old_name, NULL, output, 0);
266 fputs(" => ", output);
267 quote_c_style(patch->new_name, NULL, output, 0);
268 } else {
269 const char *n = patch->new_name;
270 if (!n)
271 n = patch->old_name;
272 quote_c_style(n, NULL, output, 0);
273 }
274 fputs(post, output);
275 }
276
277 #define CHUNKSIZE (8192)
278 #define SLOP (16)
279
280 static void read_patch_file(struct strbuf *sb, int fd)
281 {
282 if (strbuf_read(sb, fd, 0) < 0)
283 die_errno("git apply: failed to read");
284
285 /*
286 * Make sure that we have some slop in the buffer
287 * so that we can do speculative "memcmp" etc, and
288 * see to it that it is NUL-filled.
289 */
290 strbuf_grow(sb, SLOP);
291 memset(sb->buf + sb->len, 0, SLOP);
292 }
293
294 static unsigned long linelen(const char *buffer, unsigned long size)
295 {
296 unsigned long len = 0;
297 while (size--) {
298 len++;
299 if (*buffer++ == '\n')
300 break;
301 }
302 return len;
303 }
304
305 static int is_dev_null(const char *str)
306 {
307 return !memcmp("/dev/null", str, 9) && isspace(str[9]);
308 }
309
310 #define TERM_SPACE 1
311 #define TERM_TAB 2
312
313 static int name_terminate(const char *name, int namelen, int c, int terminate)
314 {
315 if (c == ' ' && !(terminate & TERM_SPACE))
316 return 0;
317 if (c == '\t' && !(terminate & TERM_TAB))
318 return 0;
319
320 return 1;
321 }
322
323 /* remove double slashes to make --index work with such filenames */
324 static char *squash_slash(char *name)
325 {
326 int i = 0, j = 0;
327
328 while (name[i]) {
329 if ((name[j++] = name[i++]) == '/')
330 while (name[i] == '/')
331 i++;
332 }
333 name[j] = '\0';
334 return name;
335 }
336
337 static char *find_name(const char *line, char *def, int p_value, int terminate)
338 {
339 int len;
340 const char *start = line;
341
342 if (*line == '"') {
343 struct strbuf name = STRBUF_INIT;
344
345 /*
346 * Proposed "new-style" GNU patch/diff format; see
347 * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
348 */
349 if (!unquote_c_style(&name, line, NULL)) {
350 char *cp;
351
352 for (cp = name.buf; p_value; p_value--) {
353 cp = strchr(cp, '/');
354 if (!cp)
355 break;
356 cp++;
357 }
358 if (cp) {
359 /* name can later be freed, so we need
360 * to memmove, not just return cp
361 */
362 strbuf_remove(&name, 0, cp - name.buf);
363 free(def);
364 if (root)
365 strbuf_insert(&name, 0, root, root_len);
366 return squash_slash(strbuf_detach(&name, NULL));
367 }
368 }
369 strbuf_release(&name);
370 }
371
372 for (;;) {
373 char c = *line;
374
375 if (isspace(c)) {
376 if (c == '\n')
377 break;
378 if (name_terminate(start, line-start, c, terminate))
379 break;
380 }
381 line++;
382 if (c == '/' && !--p_value)
383 start = line;
384 }
385 if (!start)
386 return squash_slash(def);
387 len = line - start;
388 if (!len)
389 return squash_slash(def);
390
391 /*
392 * Generally we prefer the shorter name, especially
393 * if the other one is just a variation of that with
394 * something else tacked on to the end (ie "file.orig"
395 * or "file~").
396 */
397 if (def) {
398 int deflen = strlen(def);
399 if (deflen < len && !strncmp(start, def, deflen))
400 return squash_slash(def);
401 free(def);
402 }
403
404 if (root) {
405 char *ret = xmalloc(root_len + len + 1);
406 strcpy(ret, root);
407 memcpy(ret + root_len, start, len);
408 ret[root_len + len] = '\0';
409 return squash_slash(ret);
410 }
411
412 return squash_slash(xmemdupz(start, len));
413 }
414
415 static int count_slashes(const char *cp)
416 {
417 int cnt = 0;
418 char ch;
419
420 while ((ch = *cp++))
421 if (ch == '/')
422 cnt++;
423 return cnt;
424 }
425
426 /*
427 * Given the string after "--- " or "+++ ", guess the appropriate
428 * p_value for the given patch.
429 */
430 static int guess_p_value(const char *nameline)
431 {
432 char *name, *cp;
433 int val = -1;
434
435 if (is_dev_null(nameline))
436 return -1;
437 name = find_name(nameline, NULL, 0, TERM_SPACE | TERM_TAB);
438 if (!name)
439 return -1;
440 cp = strchr(name, '/');
441 if (!cp)
442 val = 0;
443 else if (prefix) {
444 /*
445 * Does it begin with "a/$our-prefix" and such? Then this is
446 * very likely to apply to our directory.
447 */
448 if (!strncmp(name, prefix, prefix_length))
449 val = count_slashes(prefix);
450 else {
451 cp++;
452 if (!strncmp(cp, prefix, prefix_length))
453 val = count_slashes(prefix) + 1;
454 }
455 }
456 free(name);
457 return val;
458 }
459
460 /*
461 * Does the ---/+++ line has the POSIX timestamp after the last HT?
462 * GNU diff puts epoch there to signal a creation/deletion event. Is
463 * this such a timestamp?
464 */
465 static int has_epoch_timestamp(const char *nameline)
466 {
467 /*
468 * We are only interested in epoch timestamp; any non-zero
469 * fraction cannot be one, hence "(\.0+)?" in the regexp below.
470 * For the same reason, the date must be either 1969-12-31 or
471 * 1970-01-01, and the seconds part must be "00".
472 */
473 const char stamp_regexp[] =
474 "^(1969-12-31|1970-01-01)"
475 " "
476 "[0-2][0-9]:[0-5][0-9]:00(\\.0+)?"
477 " "
478 "([-+][0-2][0-9][0-5][0-9])\n";
479 const char *timestamp = NULL, *cp;
480 static regex_t *stamp;
481 regmatch_t m[10];
482 int zoneoffset;
483 int hourminute;
484 int status;
485
486 for (cp = nameline; *cp != '\n'; cp++) {
487 if (*cp == '\t')
488 timestamp = cp + 1;
489 }
490 if (!timestamp)
491 return 0;
492 if (!stamp) {
493 stamp = xmalloc(sizeof(*stamp));
494 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
495 warning("Cannot prepare timestamp regexp %s",
496 stamp_regexp);
497 return 0;
498 }
499 }
500
501 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
502 if (status) {
503 if (status != REG_NOMATCH)
504 warning("regexec returned %d for input: %s",
505 status, timestamp);
506 return 0;
507 }
508
509 zoneoffset = strtol(timestamp + m[3].rm_so + 1, NULL, 10);
510 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
511 if (timestamp[m[3].rm_so] == '-')
512 zoneoffset = -zoneoffset;
513
514 /*
515 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
516 * (west of GMT) or 1970-01-01 (east of GMT)
517 */
518 if ((zoneoffset < 0 && memcmp(timestamp, "1969-12-31", 10)) ||
519 (0 <= zoneoffset && memcmp(timestamp, "1970-01-01", 10)))
520 return 0;
521
522 hourminute = (strtol(timestamp + 11, NULL, 10) * 60 +
523 strtol(timestamp + 14, NULL, 10) -
524 zoneoffset);
525
526 return ((zoneoffset < 0 && hourminute == 1440) ||
527 (0 <= zoneoffset && !hourminute));
528 }
529
530 /*
531 * Get the name etc info from the ---/+++ lines of a traditional patch header
532 *
533 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
534 * files, we can happily check the index for a match, but for creating a
535 * new file we should try to match whatever "patch" does. I have no idea.
536 */
537 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
538 {
539 char *name;
540
541 first += 4; /* skip "--- " */
542 second += 4; /* skip "+++ " */
543 if (!p_value_known) {
544 int p, q;
545 p = guess_p_value(first);
546 q = guess_p_value(second);
547 if (p < 0) p = q;
548 if (0 <= p && p == q) {
549 p_value = p;
550 p_value_known = 1;
551 }
552 }
553 if (is_dev_null(first)) {
554 patch->is_new = 1;
555 patch->is_delete = 0;
556 name = find_name(second, NULL, p_value, TERM_SPACE | TERM_TAB);
557 patch->new_name = name;
558 } else if (is_dev_null(second)) {
559 patch->is_new = 0;
560 patch->is_delete = 1;
561 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
562 patch->old_name = name;
563 } else {
564 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
565 name = find_name(second, name, p_value, TERM_SPACE | TERM_TAB);
566 if (has_epoch_timestamp(first)) {
567 patch->is_new = 1;
568 patch->is_delete = 0;
569 patch->new_name = name;
570 } else if (has_epoch_timestamp(second)) {
571 patch->is_new = 0;
572 patch->is_delete = 1;
573 patch->old_name = name;
574 } else {
575 patch->old_name = patch->new_name = name;
576 }
577 }
578 if (!name)
579 die("unable to find filename in patch at line %d", linenr);
580 }
581
582 static int gitdiff_hdrend(const char *line, struct patch *patch)
583 {
584 return -1;
585 }
586
587 /*
588 * We're anal about diff header consistency, to make
589 * sure that we don't end up having strange ambiguous
590 * patches floating around.
591 *
592 * As a result, gitdiff_{old|new}name() will check
593 * their names against any previous information, just
594 * to make sure..
595 */
596 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
597 {
598 if (!orig_name && !isnull)
599 return find_name(line, NULL, p_value, TERM_TAB);
600
601 if (orig_name) {
602 int len;
603 const char *name;
604 char *another;
605 name = orig_name;
606 len = strlen(name);
607 if (isnull)
608 die("git apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
609 another = find_name(line, NULL, p_value, TERM_TAB);
610 if (!another || memcmp(another, name, len))
611 die("git apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
612 free(another);
613 return orig_name;
614 }
615 else {
616 /* expect "/dev/null" */
617 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
618 die("git apply: bad git-diff - expected /dev/null on line %d", linenr);
619 return NULL;
620 }
621 }
622
623 static int gitdiff_oldname(const char *line, struct patch *patch)
624 {
625 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
626 return 0;
627 }
628
629 static int gitdiff_newname(const char *line, struct patch *patch)
630 {
631 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
632 return 0;
633 }
634
635 static int gitdiff_oldmode(const char *line, struct patch *patch)
636 {
637 patch->old_mode = strtoul(line, NULL, 8);
638 return 0;
639 }
640
641 static int gitdiff_newmode(const char *line, struct patch *patch)
642 {
643 patch->new_mode = strtoul(line, NULL, 8);
644 return 0;
645 }
646
647 static int gitdiff_delete(const char *line, struct patch *patch)
648 {
649 patch->is_delete = 1;
650 patch->old_name = patch->def_name;
651 return gitdiff_oldmode(line, patch);
652 }
653
654 static int gitdiff_newfile(const char *line, struct patch *patch)
655 {
656 patch->is_new = 1;
657 patch->new_name = patch->def_name;
658 return gitdiff_newmode(line, patch);
659 }
660
661 static int gitdiff_copysrc(const char *line, struct patch *patch)
662 {
663 patch->is_copy = 1;
664 patch->old_name = find_name(line, NULL, 0, 0);
665 return 0;
666 }
667
668 static int gitdiff_copydst(const char *line, struct patch *patch)
669 {
670 patch->is_copy = 1;
671 patch->new_name = find_name(line, NULL, 0, 0);
672 return 0;
673 }
674
675 static int gitdiff_renamesrc(const char *line, struct patch *patch)
676 {
677 patch->is_rename = 1;
678 patch->old_name = find_name(line, NULL, 0, 0);
679 return 0;
680 }
681
682 static int gitdiff_renamedst(const char *line, struct patch *patch)
683 {
684 patch->is_rename = 1;
685 patch->new_name = find_name(line, NULL, 0, 0);
686 return 0;
687 }
688
689 static int gitdiff_similarity(const char *line, struct patch *patch)
690 {
691 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
692 patch->score = 0;
693 return 0;
694 }
695
696 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
697 {
698 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
699 patch->score = 0;
700 return 0;
701 }
702
703 static int gitdiff_index(const char *line, struct patch *patch)
704 {
705 /*
706 * index line is N hexadecimal, "..", N hexadecimal,
707 * and optional space with octal mode.
708 */
709 const char *ptr, *eol;
710 int len;
711
712 ptr = strchr(line, '.');
713 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
714 return 0;
715 len = ptr - line;
716 memcpy(patch->old_sha1_prefix, line, len);
717 patch->old_sha1_prefix[len] = 0;
718
719 line = ptr + 2;
720 ptr = strchr(line, ' ');
721 eol = strchr(line, '\n');
722
723 if (!ptr || eol < ptr)
724 ptr = eol;
725 len = ptr - line;
726
727 if (40 < len)
728 return 0;
729 memcpy(patch->new_sha1_prefix, line, len);
730 patch->new_sha1_prefix[len] = 0;
731 if (*ptr == ' ')
732 patch->old_mode = strtoul(ptr+1, NULL, 8);
733 return 0;
734 }
735
736 /*
737 * This is normal for a diff that doesn't change anything: we'll fall through
738 * into the next diff. Tell the parser to break out.
739 */
740 static int gitdiff_unrecognized(const char *line, struct patch *patch)
741 {
742 return -1;
743 }
744
745 static const char *stop_at_slash(const char *line, int llen)
746 {
747 int i;
748
749 for (i = 0; i < llen; i++) {
750 int ch = line[i];
751 if (ch == '/')
752 return line + i;
753 }
754 return NULL;
755 }
756
757 /*
758 * This is to extract the same name that appears on "diff --git"
759 * line. We do not find and return anything if it is a rename
760 * patch, and it is OK because we will find the name elsewhere.
761 * We need to reliably find name only when it is mode-change only,
762 * creation or deletion of an empty file. In any of these cases,
763 * both sides are the same name under a/ and b/ respectively.
764 */
765 static char *git_header_name(char *line, int llen)
766 {
767 const char *name;
768 const char *second = NULL;
769 size_t len;
770
771 line += strlen("diff --git ");
772 llen -= strlen("diff --git ");
773
774 if (*line == '"') {
775 const char *cp;
776 struct strbuf first = STRBUF_INIT;
777 struct strbuf sp = STRBUF_INIT;
778
779 if (unquote_c_style(&first, line, &second))
780 goto free_and_fail1;
781
782 /* advance to the first slash */
783 cp = stop_at_slash(first.buf, first.len);
784 /* we do not accept absolute paths */
785 if (!cp || cp == first.buf)
786 goto free_and_fail1;
787 strbuf_remove(&first, 0, cp + 1 - first.buf);
788
789 /*
790 * second points at one past closing dq of name.
791 * find the second name.
792 */
793 while ((second < line + llen) && isspace(*second))
794 second++;
795
796 if (line + llen <= second)
797 goto free_and_fail1;
798 if (*second == '"') {
799 if (unquote_c_style(&sp, second, NULL))
800 goto free_and_fail1;
801 cp = stop_at_slash(sp.buf, sp.len);
802 if (!cp || cp == sp.buf)
803 goto free_and_fail1;
804 /* They must match, otherwise ignore */
805 if (strcmp(cp + 1, first.buf))
806 goto free_and_fail1;
807 strbuf_release(&sp);
808 return strbuf_detach(&first, NULL);
809 }
810
811 /* unquoted second */
812 cp = stop_at_slash(second, line + llen - second);
813 if (!cp || cp == second)
814 goto free_and_fail1;
815 cp++;
816 if (line + llen - cp != first.len + 1 ||
817 memcmp(first.buf, cp, first.len))
818 goto free_and_fail1;
819 return strbuf_detach(&first, NULL);
820
821 free_and_fail1:
822 strbuf_release(&first);
823 strbuf_release(&sp);
824 return NULL;
825 }
826
827 /* unquoted first name */
828 name = stop_at_slash(line, llen);
829 if (!name || name == line)
830 return NULL;
831 name++;
832
833 /*
834 * since the first name is unquoted, a dq if exists must be
835 * the beginning of the second name.
836 */
837 for (second = name; second < line + llen; second++) {
838 if (*second == '"') {
839 struct strbuf sp = STRBUF_INIT;
840 const char *np;
841
842 if (unquote_c_style(&sp, second, NULL))
843 goto free_and_fail2;
844
845 np = stop_at_slash(sp.buf, sp.len);
846 if (!np || np == sp.buf)
847 goto free_and_fail2;
848 np++;
849
850 len = sp.buf + sp.len - np;
851 if (len < second - name &&
852 !strncmp(np, name, len) &&
853 isspace(name[len])) {
854 /* Good */
855 strbuf_remove(&sp, 0, np - sp.buf);
856 return strbuf_detach(&sp, NULL);
857 }
858
859 free_and_fail2:
860 strbuf_release(&sp);
861 return NULL;
862 }
863 }
864
865 /*
866 * Accept a name only if it shows up twice, exactly the same
867 * form.
868 */
869 for (len = 0 ; ; len++) {
870 switch (name[len]) {
871 default:
872 continue;
873 case '\n':
874 return NULL;
875 case '\t': case ' ':
876 second = name+len;
877 for (;;) {
878 char c = *second++;
879 if (c == '\n')
880 return NULL;
881 if (c == '/')
882 break;
883 }
884 if (second[len] == '\n' && !memcmp(name, second, len)) {
885 return xmemdupz(name, len);
886 }
887 }
888 }
889 }
890
891 /* Verify that we recognize the lines following a git header */
892 static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
893 {
894 unsigned long offset;
895
896 /* A git diff has explicit new/delete information, so we don't guess */
897 patch->is_new = 0;
898 patch->is_delete = 0;
899
900 /*
901 * Some things may not have the old name in the
902 * rest of the headers anywhere (pure mode changes,
903 * or removing or adding empty files), so we get
904 * the default name from the header.
905 */
906 patch->def_name = git_header_name(line, len);
907 if (patch->def_name && root) {
908 char *s = xmalloc(root_len + strlen(patch->def_name) + 1);
909 strcpy(s, root);
910 strcpy(s + root_len, patch->def_name);
911 free(patch->def_name);
912 patch->def_name = s;
913 }
914
915 line += len;
916 size -= len;
917 linenr++;
918 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
919 static const struct opentry {
920 const char *str;
921 int (*fn)(const char *, struct patch *);
922 } optable[] = {
923 { "@@ -", gitdiff_hdrend },
924 { "--- ", gitdiff_oldname },
925 { "+++ ", gitdiff_newname },
926 { "old mode ", gitdiff_oldmode },
927 { "new mode ", gitdiff_newmode },
928 { "deleted file mode ", gitdiff_delete },
929 { "new file mode ", gitdiff_newfile },
930 { "copy from ", gitdiff_copysrc },
931 { "copy to ", gitdiff_copydst },
932 { "rename old ", gitdiff_renamesrc },
933 { "rename new ", gitdiff_renamedst },
934 { "rename from ", gitdiff_renamesrc },
935 { "rename to ", gitdiff_renamedst },
936 { "similarity index ", gitdiff_similarity },
937 { "dissimilarity index ", gitdiff_dissimilarity },
938 { "index ", gitdiff_index },
939 { "", gitdiff_unrecognized },
940 };
941 int i;
942
943 len = linelen(line, size);
944 if (!len || line[len-1] != '\n')
945 break;
946 for (i = 0; i < ARRAY_SIZE(optable); i++) {
947 const struct opentry *p = optable + i;
948 int oplen = strlen(p->str);
949 if (len < oplen || memcmp(p->str, line, oplen))
950 continue;
951 if (p->fn(line + oplen, patch) < 0)
952 return offset;
953 break;
954 }
955 }
956
957 return offset;
958 }
959
960 static int parse_num(const char *line, unsigned long *p)
961 {
962 char *ptr;
963
964 if (!isdigit(*line))
965 return 0;
966 *p = strtoul(line, &ptr, 10);
967 return ptr - line;
968 }
969
970 static int parse_range(const char *line, int len, int offset, const char *expect,
971 unsigned long *p1, unsigned long *p2)
972 {
973 int digits, ex;
974
975 if (offset < 0 || offset >= len)
976 return -1;
977 line += offset;
978 len -= offset;
979
980 digits = parse_num(line, p1);
981 if (!digits)
982 return -1;
983
984 offset += digits;
985 line += digits;
986 len -= digits;
987
988 *p2 = 1;
989 if (*line == ',') {
990 digits = parse_num(line+1, p2);
991 if (!digits)
992 return -1;
993
994 offset += digits+1;
995 line += digits+1;
996 len -= digits+1;
997 }
998
999 ex = strlen(expect);
1000 if (ex > len)
1001 return -1;
1002 if (memcmp(line, expect, ex))
1003 return -1;
1004
1005 return offset + ex;
1006 }
1007
1008 static void recount_diff(char *line, int size, struct fragment *fragment)
1009 {
1010 int oldlines = 0, newlines = 0, ret = 0;
1011
1012 if (size < 1) {
1013 warning("recount: ignore empty hunk");
1014 return;
1015 }
1016
1017 for (;;) {
1018 int len = linelen(line, size);
1019 size -= len;
1020 line += len;
1021
1022 if (size < 1)
1023 break;
1024
1025 switch (*line) {
1026 case ' ': case '\n':
1027 newlines++;
1028 /* fall through */
1029 case '-':
1030 oldlines++;
1031 continue;
1032 case '+':
1033 newlines++;
1034 continue;
1035 case '\\':
1036 continue;
1037 case '@':
1038 ret = size < 3 || prefixcmp(line, "@@ ");
1039 break;
1040 case 'd':
1041 ret = size < 5 || prefixcmp(line, "diff ");
1042 break;
1043 default:
1044 ret = -1;
1045 break;
1046 }
1047 if (ret) {
1048 warning("recount: unexpected line: %.*s",
1049 (int)linelen(line, size), line);
1050 return;
1051 }
1052 break;
1053 }
1054 fragment->oldlines = oldlines;
1055 fragment->newlines = newlines;
1056 }
1057
1058 /*
1059 * Parse a unified diff fragment header of the
1060 * form "@@ -a,b +c,d @@"
1061 */
1062 static int parse_fragment_header(char *line, int len, struct fragment *fragment)
1063 {
1064 int offset;
1065
1066 if (!len || line[len-1] != '\n')
1067 return -1;
1068
1069 /* Figure out the number of lines in a fragment */
1070 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1071 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1072
1073 return offset;
1074 }
1075
1076 static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
1077 {
1078 unsigned long offset, len;
1079
1080 patch->is_toplevel_relative = 0;
1081 patch->is_rename = patch->is_copy = 0;
1082 patch->is_new = patch->is_delete = -1;
1083 patch->old_mode = patch->new_mode = 0;
1084 patch->old_name = patch->new_name = NULL;
1085 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
1086 unsigned long nextlen;
1087
1088 len = linelen(line, size);
1089 if (!len)
1090 break;
1091
1092 /* Testing this early allows us to take a few shortcuts.. */
1093 if (len < 6)
1094 continue;
1095
1096 /*
1097 * Make sure we don't find any unconnected patch fragments.
1098 * That's a sign that we didn't find a header, and that a
1099 * patch has become corrupted/broken up.
1100 */
1101 if (!memcmp("@@ -", line, 4)) {
1102 struct fragment dummy;
1103 if (parse_fragment_header(line, len, &dummy) < 0)
1104 continue;
1105 die("patch fragment without header at line %d: %.*s",
1106 linenr, (int)len-1, line);
1107 }
1108
1109 if (size < len + 6)
1110 break;
1111
1112 /*
1113 * Git patch? It might not have a real patch, just a rename
1114 * or mode change, so we handle that specially
1115 */
1116 if (!memcmp("diff --git ", line, 11)) {
1117 int git_hdr_len = parse_git_header(line, len, size, patch);
1118 if (git_hdr_len <= len)
1119 continue;
1120 if (!patch->old_name && !patch->new_name) {
1121 if (!patch->def_name)
1122 die("git diff header lacks filename information (line %d)", linenr);
1123 patch->old_name = patch->new_name = patch->def_name;
1124 }
1125 patch->is_toplevel_relative = 1;
1126 *hdrsize = git_hdr_len;
1127 return offset;
1128 }
1129
1130 /* --- followed by +++ ? */
1131 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1132 continue;
1133
1134 /*
1135 * We only accept unified patches, so we want it to
1136 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1137 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1138 */
1139 nextlen = linelen(line + len, size - len);
1140 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1141 continue;
1142
1143 /* Ok, we'll consider it a patch */
1144 parse_traditional_patch(line, line+len, patch);
1145 *hdrsize = len + nextlen;
1146 linenr += 2;
1147 return offset;
1148 }
1149 return -1;
1150 }
1151
1152 static void check_whitespace(const char *line, int len, unsigned ws_rule)
1153 {
1154 char *err;
1155 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1156 if (!result)
1157 return;
1158
1159 whitespace_error++;
1160 if (squelch_whitespace_errors &&
1161 squelch_whitespace_errors < whitespace_error)
1162 ;
1163 else {
1164 err = whitespace_error_string(result);
1165 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1166 patch_input_file, linenr, err, len - 2, line + 1);
1167 free(err);
1168 }
1169 }
1170
1171 /*
1172 * Parse a unified diff. Note that this really needs to parse each
1173 * fragment separately, since the only way to know the difference
1174 * between a "---" that is part of a patch, and a "---" that starts
1175 * the next patch is to look at the line counts..
1176 */
1177 static int parse_fragment(char *line, unsigned long size,
1178 struct patch *patch, struct fragment *fragment)
1179 {
1180 int added, deleted;
1181 int len = linelen(line, size), offset;
1182 unsigned long oldlines, newlines;
1183 unsigned long leading, trailing;
1184
1185 offset = parse_fragment_header(line, len, fragment);
1186 if (offset < 0)
1187 return -1;
1188 if (offset > 0 && patch->recount)
1189 recount_diff(line + offset, size - offset, fragment);
1190 oldlines = fragment->oldlines;
1191 newlines = fragment->newlines;
1192 leading = 0;
1193 trailing = 0;
1194
1195 /* Parse the thing.. */
1196 line += len;
1197 size -= len;
1198 linenr++;
1199 added = deleted = 0;
1200 for (offset = len;
1201 0 < size;
1202 offset += len, size -= len, line += len, linenr++) {
1203 if (!oldlines && !newlines)
1204 break;
1205 len = linelen(line, size);
1206 if (!len || line[len-1] != '\n')
1207 return -1;
1208 switch (*line) {
1209 default:
1210 return -1;
1211 case '\n': /* newer GNU diff, an empty context line */
1212 case ' ':
1213 oldlines--;
1214 newlines--;
1215 if (!deleted && !added)
1216 leading++;
1217 trailing++;
1218 break;
1219 case '-':
1220 if (apply_in_reverse &&
1221 ws_error_action != nowarn_ws_error)
1222 check_whitespace(line, len, patch->ws_rule);
1223 deleted++;
1224 oldlines--;
1225 trailing = 0;
1226 break;
1227 case '+':
1228 if (!apply_in_reverse &&
1229 ws_error_action != nowarn_ws_error)
1230 check_whitespace(line, len, patch->ws_rule);
1231 added++;
1232 newlines--;
1233 trailing = 0;
1234 break;
1235
1236 /*
1237 * We allow "\ No newline at end of file". Depending
1238 * on locale settings when the patch was produced we
1239 * don't know what this line looks like. The only
1240 * thing we do know is that it begins with "\ ".
1241 * Checking for 12 is just for sanity check -- any
1242 * l10n of "\ No newline..." is at least that long.
1243 */
1244 case '\\':
1245 if (len < 12 || memcmp(line, "\\ ", 2))
1246 return -1;
1247 break;
1248 }
1249 }
1250 if (oldlines || newlines)
1251 return -1;
1252 fragment->leading = leading;
1253 fragment->trailing = trailing;
1254
1255 /*
1256 * If a fragment ends with an incomplete line, we failed to include
1257 * it in the above loop because we hit oldlines == newlines == 0
1258 * before seeing it.
1259 */
1260 if (12 < size && !memcmp(line, "\\ ", 2))
1261 offset += linelen(line, size);
1262
1263 patch->lines_added += added;
1264 patch->lines_deleted += deleted;
1265
1266 if (0 < patch->is_new && oldlines)
1267 return error("new file depends on old contents");
1268 if (0 < patch->is_delete && newlines)
1269 return error("deleted file still has contents");
1270 return offset;
1271 }
1272
1273 static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
1274 {
1275 unsigned long offset = 0;
1276 unsigned long oldlines = 0, newlines = 0, context = 0;
1277 struct fragment **fragp = &patch->fragments;
1278
1279 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1280 struct fragment *fragment;
1281 int len;
1282
1283 fragment = xcalloc(1, sizeof(*fragment));
1284 len = parse_fragment(line, size, patch, fragment);
1285 if (len <= 0)
1286 die("corrupt patch at line %d", linenr);
1287 fragment->patch = line;
1288 fragment->size = len;
1289 oldlines += fragment->oldlines;
1290 newlines += fragment->newlines;
1291 context += fragment->leading + fragment->trailing;
1292
1293 *fragp = fragment;
1294 fragp = &fragment->next;
1295
1296 offset += len;
1297 line += len;
1298 size -= len;
1299 }
1300
1301 /*
1302 * If something was removed (i.e. we have old-lines) it cannot
1303 * be creation, and if something was added it cannot be
1304 * deletion. However, the reverse is not true; --unified=0
1305 * patches that only add are not necessarily creation even
1306 * though they do not have any old lines, and ones that only
1307 * delete are not necessarily deletion.
1308 *
1309 * Unfortunately, a real creation/deletion patch do _not_ have
1310 * any context line by definition, so we cannot safely tell it
1311 * apart with --unified=0 insanity. At least if the patch has
1312 * more than one hunk it is not creation or deletion.
1313 */
1314 if (patch->is_new < 0 &&
1315 (oldlines || (patch->fragments && patch->fragments->next)))
1316 patch->is_new = 0;
1317 if (patch->is_delete < 0 &&
1318 (newlines || (patch->fragments && patch->fragments->next)))
1319 patch->is_delete = 0;
1320
1321 if (0 < patch->is_new && oldlines)
1322 die("new file %s depends on old contents", patch->new_name);
1323 if (0 < patch->is_delete && newlines)
1324 die("deleted file %s still has contents", patch->old_name);
1325 if (!patch->is_delete && !newlines && context)
1326 fprintf(stderr, "** warning: file %s becomes empty but "
1327 "is not deleted\n", patch->new_name);
1328
1329 return offset;
1330 }
1331
1332 static inline int metadata_changes(struct patch *patch)
1333 {
1334 return patch->is_rename > 0 ||
1335 patch->is_copy > 0 ||
1336 patch->is_new > 0 ||
1337 patch->is_delete ||
1338 (patch->old_mode && patch->new_mode &&
1339 patch->old_mode != patch->new_mode);
1340 }
1341
1342 static char *inflate_it(const void *data, unsigned long size,
1343 unsigned long inflated_size)
1344 {
1345 z_stream stream;
1346 void *out;
1347 int st;
1348
1349 memset(&stream, 0, sizeof(stream));
1350
1351 stream.next_in = (unsigned char *)data;
1352 stream.avail_in = size;
1353 stream.next_out = out = xmalloc(inflated_size);
1354 stream.avail_out = inflated_size;
1355 git_inflate_init(&stream);
1356 st = git_inflate(&stream, Z_FINISH);
1357 git_inflate_end(&stream);
1358 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1359 free(out);
1360 return NULL;
1361 }
1362 return out;
1363 }
1364
1365 static struct fragment *parse_binary_hunk(char **buf_p,
1366 unsigned long *sz_p,
1367 int *status_p,
1368 int *used_p)
1369 {
1370 /*
1371 * Expect a line that begins with binary patch method ("literal"
1372 * or "delta"), followed by the length of data before deflating.
1373 * a sequence of 'length-byte' followed by base-85 encoded data
1374 * should follow, terminated by a newline.
1375 *
1376 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1377 * and we would limit the patch line to 66 characters,
1378 * so one line can fit up to 13 groups that would decode
1379 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1380 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1381 */
1382 int llen, used;
1383 unsigned long size = *sz_p;
1384 char *buffer = *buf_p;
1385 int patch_method;
1386 unsigned long origlen;
1387 char *data = NULL;
1388 int hunk_size = 0;
1389 struct fragment *frag;
1390
1391 llen = linelen(buffer, size);
1392 used = llen;
1393
1394 *status_p = 0;
1395
1396 if (!prefixcmp(buffer, "delta ")) {
1397 patch_method = BINARY_DELTA_DEFLATED;
1398 origlen = strtoul(buffer + 6, NULL, 10);
1399 }
1400 else if (!prefixcmp(buffer, "literal ")) {
1401 patch_method = BINARY_LITERAL_DEFLATED;
1402 origlen = strtoul(buffer + 8, NULL, 10);
1403 }
1404 else
1405 return NULL;
1406
1407 linenr++;
1408 buffer += llen;
1409 while (1) {
1410 int byte_length, max_byte_length, newsize;
1411 llen = linelen(buffer, size);
1412 used += llen;
1413 linenr++;
1414 if (llen == 1) {
1415 /* consume the blank line */
1416 buffer++;
1417 size--;
1418 break;
1419 }
1420 /*
1421 * Minimum line is "A00000\n" which is 7-byte long,
1422 * and the line length must be multiple of 5 plus 2.
1423 */
1424 if ((llen < 7) || (llen-2) % 5)
1425 goto corrupt;
1426 max_byte_length = (llen - 2) / 5 * 4;
1427 byte_length = *buffer;
1428 if ('A' <= byte_length && byte_length <= 'Z')
1429 byte_length = byte_length - 'A' + 1;
1430 else if ('a' <= byte_length && byte_length <= 'z')
1431 byte_length = byte_length - 'a' + 27;
1432 else
1433 goto corrupt;
1434 /* if the input length was not multiple of 4, we would
1435 * have filler at the end but the filler should never
1436 * exceed 3 bytes
1437 */
1438 if (max_byte_length < byte_length ||
1439 byte_length <= max_byte_length - 4)
1440 goto corrupt;
1441 newsize = hunk_size + byte_length;
1442 data = xrealloc(data, newsize);
1443 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1444 goto corrupt;
1445 hunk_size = newsize;
1446 buffer += llen;
1447 size -= llen;
1448 }
1449
1450 frag = xcalloc(1, sizeof(*frag));
1451 frag->patch = inflate_it(data, hunk_size, origlen);
1452 if (!frag->patch)
1453 goto corrupt;
1454 free(data);
1455 frag->size = origlen;
1456 *buf_p = buffer;
1457 *sz_p = size;
1458 *used_p = used;
1459 frag->binary_patch_method = patch_method;
1460 return frag;
1461
1462 corrupt:
1463 free(data);
1464 *status_p = -1;
1465 error("corrupt binary patch at line %d: %.*s",
1466 linenr-1, llen-1, buffer);
1467 return NULL;
1468 }
1469
1470 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1471 {
1472 /*
1473 * We have read "GIT binary patch\n"; what follows is a line
1474 * that says the patch method (currently, either "literal" or
1475 * "delta") and the length of data before deflating; a
1476 * sequence of 'length-byte' followed by base-85 encoded data
1477 * follows.
1478 *
1479 * When a binary patch is reversible, there is another binary
1480 * hunk in the same format, starting with patch method (either
1481 * "literal" or "delta") with the length of data, and a sequence
1482 * of length-byte + base-85 encoded data, terminated with another
1483 * empty line. This data, when applied to the postimage, produces
1484 * the preimage.
1485 */
1486 struct fragment *forward;
1487 struct fragment *reverse;
1488 int status;
1489 int used, used_1;
1490
1491 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1492 if (!forward && !status)
1493 /* there has to be one hunk (forward hunk) */
1494 return error("unrecognized binary patch at line %d", linenr-1);
1495 if (status)
1496 /* otherwise we already gave an error message */
1497 return status;
1498
1499 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1500 if (reverse)
1501 used += used_1;
1502 else if (status) {
1503 /*
1504 * Not having reverse hunk is not an error, but having
1505 * a corrupt reverse hunk is.
1506 */
1507 free((void*) forward->patch);
1508 free(forward);
1509 return status;
1510 }
1511 forward->next = reverse;
1512 patch->fragments = forward;
1513 patch->is_binary = 1;
1514 return used;
1515 }
1516
1517 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1518 {
1519 int hdrsize, patchsize;
1520 int offset = find_header(buffer, size, &hdrsize, patch);
1521
1522 if (offset < 0)
1523 return offset;
1524
1525 patch->ws_rule = whitespace_rule(patch->new_name
1526 ? patch->new_name
1527 : patch->old_name);
1528
1529 patchsize = parse_single_patch(buffer + offset + hdrsize,
1530 size - offset - hdrsize, patch);
1531
1532 if (!patchsize) {
1533 static const char *binhdr[] = {
1534 "Binary files ",
1535 "Files ",
1536 NULL,
1537 };
1538 static const char git_binary[] = "GIT binary patch\n";
1539 int i;
1540 int hd = hdrsize + offset;
1541 unsigned long llen = linelen(buffer + hd, size - hd);
1542
1543 if (llen == sizeof(git_binary) - 1 &&
1544 !memcmp(git_binary, buffer + hd, llen)) {
1545 int used;
1546 linenr++;
1547 used = parse_binary(buffer + hd + llen,
1548 size - hd - llen, patch);
1549 if (used)
1550 patchsize = used + llen;
1551 else
1552 patchsize = 0;
1553 }
1554 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1555 for (i = 0; binhdr[i]; i++) {
1556 int len = strlen(binhdr[i]);
1557 if (len < size - hd &&
1558 !memcmp(binhdr[i], buffer + hd, len)) {
1559 linenr++;
1560 patch->is_binary = 1;
1561 patchsize = llen;
1562 break;
1563 }
1564 }
1565 }
1566
1567 /* Empty patch cannot be applied if it is a text patch
1568 * without metadata change. A binary patch appears
1569 * empty to us here.
1570 */
1571 if ((apply || check) &&
1572 (!patch->is_binary && !metadata_changes(patch)))
1573 die("patch with only garbage at line %d", linenr);
1574 }
1575
1576 return offset + hdrsize + patchsize;
1577 }
1578
1579 #define swap(a,b) myswap((a),(b),sizeof(a))
1580
1581 #define myswap(a, b, size) do { \
1582 unsigned char mytmp[size]; \
1583 memcpy(mytmp, &a, size); \
1584 memcpy(&a, &b, size); \
1585 memcpy(&b, mytmp, size); \
1586 } while (0)
1587
1588 static void reverse_patches(struct patch *p)
1589 {
1590 for (; p; p = p->next) {
1591 struct fragment *frag = p->fragments;
1592
1593 swap(p->new_name, p->old_name);
1594 swap(p->new_mode, p->old_mode);
1595 swap(p->is_new, p->is_delete);
1596 swap(p->lines_added, p->lines_deleted);
1597 swap(p->old_sha1_prefix, p->new_sha1_prefix);
1598
1599 for (; frag; frag = frag->next) {
1600 swap(frag->newpos, frag->oldpos);
1601 swap(frag->newlines, frag->oldlines);
1602 }
1603 }
1604 }
1605
1606 static const char pluses[] =
1607 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1608 static const char minuses[]=
1609 "----------------------------------------------------------------------";
1610
1611 static void show_stats(struct patch *patch)
1612 {
1613 struct strbuf qname = STRBUF_INIT;
1614 char *cp = patch->new_name ? patch->new_name : patch->old_name;
1615 int max, add, del;
1616
1617 quote_c_style(cp, &qname, NULL, 0);
1618
1619 /*
1620 * "scale" the filename
1621 */
1622 max = max_len;
1623 if (max > 50)
1624 max = 50;
1625
1626 if (qname.len > max) {
1627 cp = strchr(qname.buf + qname.len + 3 - max, '/');
1628 if (!cp)
1629 cp = qname.buf + qname.len + 3 - max;
1630 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
1631 }
1632
1633 if (patch->is_binary) {
1634 printf(" %-*s | Bin\n", max, qname.buf);
1635 strbuf_release(&qname);
1636 return;
1637 }
1638
1639 printf(" %-*s |", max, qname.buf);
1640 strbuf_release(&qname);
1641
1642 /*
1643 * scale the add/delete
1644 */
1645 max = max + max_change > 70 ? 70 - max : max_change;
1646 add = patch->lines_added;
1647 del = patch->lines_deleted;
1648
1649 if (max_change > 0) {
1650 int total = ((add + del) * max + max_change / 2) / max_change;
1651 add = (add * max + max_change / 2) / max_change;
1652 del = total - add;
1653 }
1654 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
1655 add, pluses, del, minuses);
1656 }
1657
1658 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
1659 {
1660 switch (st->st_mode & S_IFMT) {
1661 case S_IFLNK:
1662 if (strbuf_readlink(buf, path, st->st_size) < 0)
1663 return error("unable to read symlink %s", path);
1664 return 0;
1665 case S_IFREG:
1666 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
1667 return error("unable to open or read %s", path);
1668 convert_to_git(path, buf->buf, buf->len, buf, 0);
1669 return 0;
1670 default:
1671 return -1;
1672 }
1673 }
1674
1675 static void update_pre_post_images(struct image *preimage,
1676 struct image *postimage,
1677 char *buf,
1678 size_t len)
1679 {
1680 int i, ctx;
1681 char *new, *old, *fixed;
1682 struct image fixed_preimage;
1683
1684 /*
1685 * Update the preimage with whitespace fixes. Note that we
1686 * are not losing preimage->buf -- apply_one_fragment() will
1687 * free "oldlines".
1688 */
1689 prepare_image(&fixed_preimage, buf, len, 1);
1690 assert(fixed_preimage.nr == preimage->nr);
1691 for (i = 0; i < preimage->nr; i++)
1692 fixed_preimage.line[i].flag = preimage->line[i].flag;
1693 free(preimage->line_allocated);
1694 *preimage = fixed_preimage;
1695
1696 /*
1697 * Adjust the common context lines in postimage, in place.
1698 * This is possible because whitespace fixing does not make
1699 * the string grow.
1700 */
1701 new = old = postimage->buf;
1702 fixed = preimage->buf;
1703 for (i = ctx = 0; i < postimage->nr; i++) {
1704 size_t len = postimage->line[i].len;
1705 if (!(postimage->line[i].flag & LINE_COMMON)) {
1706 /* an added line -- no counterparts in preimage */
1707 memmove(new, old, len);
1708 old += len;
1709 new += len;
1710 continue;
1711 }
1712
1713 /* a common context -- skip it in the original postimage */
1714 old += len;
1715
1716 /* and find the corresponding one in the fixed preimage */
1717 while (ctx < preimage->nr &&
1718 !(preimage->line[ctx].flag & LINE_COMMON)) {
1719 fixed += preimage->line[ctx].len;
1720 ctx++;
1721 }
1722 if (preimage->nr <= ctx)
1723 die("oops");
1724
1725 /* and copy it in, while fixing the line length */
1726 len = preimage->line[ctx].len;
1727 memcpy(new, fixed, len);
1728 new += len;
1729 fixed += len;
1730 postimage->line[i].len = len;
1731 ctx++;
1732 }
1733
1734 /* Fix the length of the whole thing */
1735 postimage->len = new - postimage->buf;
1736 }
1737
1738 static int match_fragment(struct image *img,
1739 struct image *preimage,
1740 struct image *postimage,
1741 unsigned long try,
1742 int try_lno,
1743 unsigned ws_rule,
1744 int match_beginning, int match_end)
1745 {
1746 int i;
1747 char *fixed_buf, *buf, *orig, *target;
1748
1749 if (preimage->nr + try_lno > img->nr)
1750 return 0;
1751
1752 if (match_beginning && try_lno)
1753 return 0;
1754
1755 if (match_end && preimage->nr + try_lno != img->nr)
1756 return 0;
1757
1758 /* Quick hash check */
1759 for (i = 0; i < preimage->nr; i++)
1760 if (preimage->line[i].hash != img->line[try_lno + i].hash)
1761 return 0;
1762
1763 /*
1764 * Do we have an exact match? If we were told to match
1765 * at the end, size must be exactly at try+fragsize,
1766 * otherwise try+fragsize must be still within the preimage,
1767 * and either case, the old piece should match the preimage
1768 * exactly.
1769 */
1770 if ((match_end
1771 ? (try + preimage->len == img->len)
1772 : (try + preimage->len <= img->len)) &&
1773 !memcmp(img->buf + try, preimage->buf, preimage->len))
1774 return 1;
1775
1776 if (ws_error_action != correct_ws_error)
1777 return 0;
1778
1779 /*
1780 * The hunk does not apply byte-by-byte, but the hash says
1781 * it might with whitespace fuzz.
1782 */
1783 fixed_buf = xmalloc(preimage->len + 1);
1784 buf = fixed_buf;
1785 orig = preimage->buf;
1786 target = img->buf + try;
1787 for (i = 0; i < preimage->nr; i++) {
1788 size_t fixlen; /* length after fixing the preimage */
1789 size_t oldlen = preimage->line[i].len;
1790 size_t tgtlen = img->line[try_lno + i].len;
1791 size_t tgtfixlen; /* length after fixing the target line */
1792 char tgtfixbuf[1024], *tgtfix;
1793 int match;
1794
1795 /* Try fixing the line in the preimage */
1796 fixlen = ws_fix_copy(buf, orig, oldlen, ws_rule, NULL);
1797
1798 /* Try fixing the line in the target */
1799 if (sizeof(tgtfixbuf) > tgtlen)
1800 tgtfix = tgtfixbuf;
1801 else
1802 tgtfix = xmalloc(tgtlen);
1803 tgtfixlen = ws_fix_copy(tgtfix, target, tgtlen, ws_rule, NULL);
1804
1805 /*
1806 * If they match, either the preimage was based on
1807 * a version before our tree fixed whitespace breakage,
1808 * or we are lacking a whitespace-fix patch the tree
1809 * the preimage was based on already had (i.e. target
1810 * has whitespace breakage, the preimage doesn't).
1811 * In either case, we are fixing the whitespace breakages
1812 * so we might as well take the fix together with their
1813 * real change.
1814 */
1815 match = (tgtfixlen == fixlen && !memcmp(tgtfix, buf, fixlen));
1816
1817 if (tgtfix != tgtfixbuf)
1818 free(tgtfix);
1819 if (!match)
1820 goto unmatch_exit;
1821
1822 orig += oldlen;
1823 buf += fixlen;
1824 target += tgtlen;
1825 }
1826
1827 /*
1828 * Yes, the preimage is based on an older version that still
1829 * has whitespace breakages unfixed, and fixing them makes the
1830 * hunk match. Update the context lines in the postimage.
1831 */
1832 update_pre_post_images(preimage, postimage,
1833 fixed_buf, buf - fixed_buf);
1834 return 1;
1835
1836 unmatch_exit:
1837 free(fixed_buf);
1838 return 0;
1839 }
1840
1841 static int find_pos(struct image *img,
1842 struct image *preimage,
1843 struct image *postimage,
1844 int line,
1845 unsigned ws_rule,
1846 int match_beginning, int match_end)
1847 {
1848 int i;
1849 unsigned long backwards, forwards, try;
1850 int backwards_lno, forwards_lno, try_lno;
1851
1852 if (preimage->nr > img->nr)
1853 return -1;
1854
1855 /*
1856 * If match_begining or match_end is specified, there is no
1857 * point starting from a wrong line that will never match and
1858 * wander around and wait for a match at the specified end.
1859 */
1860 if (match_beginning)
1861 line = 0;
1862 else if (match_end)
1863 line = img->nr - preimage->nr;
1864
1865 if (line > img->nr)
1866 line = img->nr;
1867
1868 try = 0;
1869 for (i = 0; i < line; i++)
1870 try += img->line[i].len;
1871
1872 /*
1873 * There's probably some smart way to do this, but I'll leave
1874 * that to the smart and beautiful people. I'm simple and stupid.
1875 */
1876 backwards = try;
1877 backwards_lno = line;
1878 forwards = try;
1879 forwards_lno = line;
1880 try_lno = line;
1881
1882 for (i = 0; ; i++) {
1883 if (match_fragment(img, preimage, postimage,
1884 try, try_lno, ws_rule,
1885 match_beginning, match_end))
1886 return try_lno;
1887
1888 again:
1889 if (backwards_lno == 0 && forwards_lno == img->nr)
1890 break;
1891
1892 if (i & 1) {
1893 if (backwards_lno == 0) {
1894 i++;
1895 goto again;
1896 }
1897 backwards_lno--;
1898 backwards -= img->line[backwards_lno].len;
1899 try = backwards;
1900 try_lno = backwards_lno;
1901 } else {
1902 if (forwards_lno == img->nr) {
1903 i++;
1904 goto again;
1905 }
1906 forwards += img->line[forwards_lno].len;
1907 forwards_lno++;
1908 try = forwards;
1909 try_lno = forwards_lno;
1910 }
1911
1912 }
1913 return -1;
1914 }
1915
1916 static void remove_first_line(struct image *img)
1917 {
1918 img->buf += img->line[0].len;
1919 img->len -= img->line[0].len;
1920 img->line++;
1921 img->nr--;
1922 }
1923
1924 static void remove_last_line(struct image *img)
1925 {
1926 img->len -= img->line[--img->nr].len;
1927 }
1928
1929 static void update_image(struct image *img,
1930 int applied_pos,
1931 struct image *preimage,
1932 struct image *postimage)
1933 {
1934 /*
1935 * remove the copy of preimage at offset in img
1936 * and replace it with postimage
1937 */
1938 int i, nr;
1939 size_t remove_count, insert_count, applied_at = 0;
1940 char *result;
1941
1942 for (i = 0; i < applied_pos; i++)
1943 applied_at += img->line[i].len;
1944
1945 remove_count = 0;
1946 for (i = 0; i < preimage->nr; i++)
1947 remove_count += img->line[applied_pos + i].len;
1948 insert_count = postimage->len;
1949
1950 /* Adjust the contents */
1951 result = xmalloc(img->len + insert_count - remove_count + 1);
1952 memcpy(result, img->buf, applied_at);
1953 memcpy(result + applied_at, postimage->buf, postimage->len);
1954 memcpy(result + applied_at + postimage->len,
1955 img->buf + (applied_at + remove_count),
1956 img->len - (applied_at + remove_count));
1957 free(img->buf);
1958 img->buf = result;
1959 img->len += insert_count - remove_count;
1960 result[img->len] = '\0';
1961
1962 /* Adjust the line table */
1963 nr = img->nr + postimage->nr - preimage->nr;
1964 if (preimage->nr < postimage->nr) {
1965 /*
1966 * NOTE: this knows that we never call remove_first_line()
1967 * on anything other than pre/post image.
1968 */
1969 img->line = xrealloc(img->line, nr * sizeof(*img->line));
1970 img->line_allocated = img->line;
1971 }
1972 if (preimage->nr != postimage->nr)
1973 memmove(img->line + applied_pos + postimage->nr,
1974 img->line + applied_pos + preimage->nr,
1975 (img->nr - (applied_pos + preimage->nr)) *
1976 sizeof(*img->line));
1977 memcpy(img->line + applied_pos,
1978 postimage->line,
1979 postimage->nr * sizeof(*img->line));
1980 img->nr = nr;
1981 }
1982
1983 static int apply_one_fragment(struct image *img, struct fragment *frag,
1984 int inaccurate_eof, unsigned ws_rule)
1985 {
1986 int match_beginning, match_end;
1987 const char *patch = frag->patch;
1988 int size = frag->size;
1989 char *old, *new, *oldlines, *newlines;
1990 int new_blank_lines_at_end = 0;
1991 unsigned long leading, trailing;
1992 int pos, applied_pos;
1993 struct image preimage;
1994 struct image postimage;
1995
1996 memset(&preimage, 0, sizeof(preimage));
1997 memset(&postimage, 0, sizeof(postimage));
1998 oldlines = xmalloc(size);
1999 newlines = xmalloc(size);
2000
2001 old = oldlines;
2002 new = newlines;
2003 while (size > 0) {
2004 char first;
2005 int len = linelen(patch, size);
2006 int plen, added;
2007 int added_blank_line = 0;
2008
2009 if (!len)
2010 break;
2011
2012 /*
2013 * "plen" is how much of the line we should use for
2014 * the actual patch data. Normally we just remove the
2015 * first character on the line, but if the line is
2016 * followed by "\ No newline", then we also remove the
2017 * last one (which is the newline, of course).
2018 */
2019 plen = len - 1;
2020 if (len < size && patch[len] == '\\')
2021 plen--;
2022 first = *patch;
2023 if (apply_in_reverse) {
2024 if (first == '-')
2025 first = '+';
2026 else if (first == '+')
2027 first = '-';
2028 }
2029
2030 switch (first) {
2031 case '\n':
2032 /* Newer GNU diff, empty context line */
2033 if (plen < 0)
2034 /* ... followed by '\No newline'; nothing */
2035 break;
2036 *old++ = '\n';
2037 *new++ = '\n';
2038 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2039 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2040 break;
2041 case ' ':
2042 case '-':
2043 memcpy(old, patch + 1, plen);
2044 add_line_info(&preimage, old, plen,
2045 (first == ' ' ? LINE_COMMON : 0));
2046 old += plen;
2047 if (first == '-')
2048 break;
2049 /* Fall-through for ' ' */
2050 case '+':
2051 /* --no-add does not add new lines */
2052 if (first == '+' && no_add)
2053 break;
2054
2055 if (first != '+' ||
2056 !whitespace_error ||
2057 ws_error_action != correct_ws_error) {
2058 memcpy(new, patch + 1, plen);
2059 added = plen;
2060 }
2061 else {
2062 added = ws_fix_copy(new, patch + 1, plen, ws_rule, &applied_after_fixing_ws);
2063 }
2064 add_line_info(&postimage, new, added,
2065 (first == '+' ? 0 : LINE_COMMON));
2066 new += added;
2067 if (first == '+' &&
2068 added == 1 && new[-1] == '\n')
2069 added_blank_line = 1;
2070 break;
2071 case '@': case '\\':
2072 /* Ignore it, we already handled it */
2073 break;
2074 default:
2075 if (apply_verbosely)
2076 error("invalid start of line: '%c'", first);
2077 return -1;
2078 }
2079 if (added_blank_line)
2080 new_blank_lines_at_end++;
2081 else
2082 new_blank_lines_at_end = 0;
2083 patch += len;
2084 size -= len;
2085 }
2086 if (inaccurate_eof &&
2087 old > oldlines && old[-1] == '\n' &&
2088 new > newlines && new[-1] == '\n') {
2089 old--;
2090 new--;
2091 }
2092
2093 leading = frag->leading;
2094 trailing = frag->trailing;
2095
2096 /*
2097 * A hunk to change lines at the beginning would begin with
2098 * @@ -1,L +N,M @@
2099 * but we need to be careful. -U0 that inserts before the second
2100 * line also has this pattern.
2101 *
2102 * And a hunk to add to an empty file would begin with
2103 * @@ -0,0 +N,M @@
2104 *
2105 * In other words, a hunk that is (frag->oldpos <= 1) with or
2106 * without leading context must match at the beginning.
2107 */
2108 match_beginning = (!frag->oldpos ||
2109 (frag->oldpos == 1 && !unidiff_zero));
2110
2111 /*
2112 * A hunk without trailing lines must match at the end.
2113 * However, we simply cannot tell if a hunk must match end
2114 * from the lack of trailing lines if the patch was generated
2115 * with unidiff without any context.
2116 */
2117 match_end = !unidiff_zero && !trailing;
2118
2119 pos = frag->newpos ? (frag->newpos - 1) : 0;
2120 preimage.buf = oldlines;
2121 preimage.len = old - oldlines;
2122 postimage.buf = newlines;
2123 postimage.len = new - newlines;
2124 preimage.line = preimage.line_allocated;
2125 postimage.line = postimage.line_allocated;
2126
2127 for (;;) {
2128
2129 applied_pos = find_pos(img, &preimage, &postimage, pos,
2130 ws_rule, match_beginning, match_end);
2131
2132 if (applied_pos >= 0)
2133 break;
2134
2135 /* Am I at my context limits? */
2136 if ((leading <= p_context) && (trailing <= p_context))
2137 break;
2138 if (match_beginning || match_end) {
2139 match_beginning = match_end = 0;
2140 continue;
2141 }
2142
2143 /*
2144 * Reduce the number of context lines; reduce both
2145 * leading and trailing if they are equal otherwise
2146 * just reduce the larger context.
2147 */
2148 if (leading >= trailing) {
2149 remove_first_line(&preimage);
2150 remove_first_line(&postimage);
2151 pos--;
2152 leading--;
2153 }
2154 if (trailing > leading) {
2155 remove_last_line(&preimage);
2156 remove_last_line(&postimage);
2157 trailing--;
2158 }
2159 }
2160
2161 if (applied_pos >= 0) {
2162 if (ws_error_action == correct_ws_error &&
2163 new_blank_lines_at_end &&
2164 postimage.nr + applied_pos == img->nr) {
2165 /*
2166 * If the patch application adds blank lines
2167 * at the end, and if the patch applies at the
2168 * end of the image, remove those added blank
2169 * lines.
2170 */
2171 while (new_blank_lines_at_end--)
2172 remove_last_line(&postimage);
2173 }
2174
2175 /*
2176 * Warn if it was necessary to reduce the number
2177 * of context lines.
2178 */
2179 if ((leading != frag->leading) ||
2180 (trailing != frag->trailing))
2181 fprintf(stderr, "Context reduced to (%ld/%ld)"
2182 " to apply fragment at %d\n",
2183 leading, trailing, applied_pos+1);
2184 update_image(img, applied_pos, &preimage, &postimage);
2185 } else {
2186 if (apply_verbosely)
2187 error("while searching for:\n%.*s",
2188 (int)(old - oldlines), oldlines);
2189 }
2190
2191 free(oldlines);
2192 free(newlines);
2193 free(preimage.line_allocated);
2194 free(postimage.line_allocated);
2195
2196 return (applied_pos < 0);
2197 }
2198
2199 static int apply_binary_fragment(struct image *img, struct patch *patch)
2200 {
2201 struct fragment *fragment = patch->fragments;
2202 unsigned long len;
2203 void *dst;
2204
2205 /* Binary patch is irreversible without the optional second hunk */
2206 if (apply_in_reverse) {
2207 if (!fragment->next)
2208 return error("cannot reverse-apply a binary patch "
2209 "without the reverse hunk to '%s'",
2210 patch->new_name
2211 ? patch->new_name : patch->old_name);
2212 fragment = fragment->next;
2213 }
2214 switch (fragment->binary_patch_method) {
2215 case BINARY_DELTA_DEFLATED:
2216 dst = patch_delta(img->buf, img->len, fragment->patch,
2217 fragment->size, &len);
2218 if (!dst)
2219 return -1;
2220 clear_image(img);
2221 img->buf = dst;
2222 img->len = len;
2223 return 0;
2224 case BINARY_LITERAL_DEFLATED:
2225 clear_image(img);
2226 img->len = fragment->size;
2227 img->buf = xmalloc(img->len+1);
2228 memcpy(img->buf, fragment->patch, img->len);
2229 img->buf[img->len] = '\0';
2230 return 0;
2231 }
2232 return -1;
2233 }
2234
2235 static int apply_binary(struct image *img, struct patch *patch)
2236 {
2237 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2238 unsigned char sha1[20];
2239
2240 /*
2241 * For safety, we require patch index line to contain
2242 * full 40-byte textual SHA1 for old and new, at least for now.
2243 */
2244 if (strlen(patch->old_sha1_prefix) != 40 ||
2245 strlen(patch->new_sha1_prefix) != 40 ||
2246 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
2247 get_sha1_hex(patch->new_sha1_prefix, sha1))
2248 return error("cannot apply binary patch to '%s' "
2249 "without full index line", name);
2250
2251 if (patch->old_name) {
2252 /*
2253 * See if the old one matches what the patch
2254 * applies to.
2255 */
2256 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2257 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
2258 return error("the patch applies to '%s' (%s), "
2259 "which does not match the "
2260 "current contents.",
2261 name, sha1_to_hex(sha1));
2262 }
2263 else {
2264 /* Otherwise, the old one must be empty. */
2265 if (img->len)
2266 return error("the patch applies to an empty "
2267 "'%s' but it is not empty", name);
2268 }
2269
2270 get_sha1_hex(patch->new_sha1_prefix, sha1);
2271 if (is_null_sha1(sha1)) {
2272 clear_image(img);
2273 return 0; /* deletion patch */
2274 }
2275
2276 if (has_sha1_file(sha1)) {
2277 /* We already have the postimage */
2278 enum object_type type;
2279 unsigned long size;
2280 char *result;
2281
2282 result = read_sha1_file(sha1, &type, &size);
2283 if (!result)
2284 return error("the necessary postimage %s for "
2285 "'%s' cannot be read",
2286 patch->new_sha1_prefix, name);
2287 clear_image(img);
2288 img->buf = result;
2289 img->len = size;
2290 } else {
2291 /*
2292 * We have verified buf matches the preimage;
2293 * apply the patch data to it, which is stored
2294 * in the patch->fragments->{patch,size}.
2295 */
2296 if (apply_binary_fragment(img, patch))
2297 return error("binary patch does not apply to '%s'",
2298 name);
2299
2300 /* verify that the result matches */
2301 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2302 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
2303 return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)",
2304 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
2305 }
2306
2307 return 0;
2308 }
2309
2310 static int apply_fragments(struct image *img, struct patch *patch)
2311 {
2312 struct fragment *frag = patch->fragments;
2313 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2314 unsigned ws_rule = patch->ws_rule;
2315 unsigned inaccurate_eof = patch->inaccurate_eof;
2316
2317 if (patch->is_binary)
2318 return apply_binary(img, patch);
2319
2320 while (frag) {
2321 if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule)) {
2322 error("patch failed: %s:%ld", name, frag->oldpos);
2323 if (!apply_with_reject)
2324 return -1;
2325 frag->rejected = 1;
2326 }
2327 frag = frag->next;
2328 }
2329 return 0;
2330 }
2331
2332 static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
2333 {
2334 if (!ce)
2335 return 0;
2336
2337 if (S_ISGITLINK(ce->ce_mode)) {
2338 strbuf_grow(buf, 100);
2339 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(ce->sha1));
2340 } else {
2341 enum object_type type;
2342 unsigned long sz;
2343 char *result;
2344
2345 result = read_sha1_file(ce->sha1, &type, &sz);
2346 if (!result)
2347 return -1;
2348 /* XXX read_sha1_file NUL-terminates */
2349 strbuf_attach(buf, result, sz, sz + 1);
2350 }
2351 return 0;
2352 }
2353
2354 static struct patch *in_fn_table(const char *name)
2355 {
2356 struct string_list_item *item;
2357
2358 if (name == NULL)
2359 return NULL;
2360
2361 item = string_list_lookup(name, &fn_table);
2362 if (item != NULL)
2363 return (struct patch *)item->util;
2364
2365 return NULL;
2366 }
2367
2368 /*
2369 * item->util in the filename table records the status of the path.
2370 * Usually it points at a patch (whose result records the contents
2371 * of it after applying it), but it could be PATH_WAS_DELETED for a
2372 * path that a previously applied patch has already removed.
2373 */
2374 #define PATH_TO_BE_DELETED ((struct patch *) -2)
2375 #define PATH_WAS_DELETED ((struct patch *) -1)
2376
2377 static int to_be_deleted(struct patch *patch)
2378 {
2379 return patch == PATH_TO_BE_DELETED;
2380 }
2381
2382 static int was_deleted(struct patch *patch)
2383 {
2384 return patch == PATH_WAS_DELETED;
2385 }
2386
2387 static void add_to_fn_table(struct patch *patch)
2388 {
2389 struct string_list_item *item;
2390
2391 /*
2392 * Always add new_name unless patch is a deletion
2393 * This should cover the cases for normal diffs,
2394 * file creations and copies
2395 */
2396 if (patch->new_name != NULL) {
2397 item = string_list_insert(patch->new_name, &fn_table);
2398 item->util = patch;
2399 }
2400
2401 /*
2402 * store a failure on rename/deletion cases because
2403 * later chunks shouldn't patch old names
2404 */
2405 if ((patch->new_name == NULL) || (patch->is_rename)) {
2406 item = string_list_insert(patch->old_name, &fn_table);
2407 item->util = PATH_WAS_DELETED;
2408 }
2409 }
2410
2411 static void prepare_fn_table(struct patch *patch)
2412 {
2413 /*
2414 * store information about incoming file deletion
2415 */
2416 while (patch) {
2417 if ((patch->new_name == NULL) || (patch->is_rename)) {
2418 struct string_list_item *item;
2419 item = string_list_insert(patch->old_name, &fn_table);
2420 item->util = PATH_TO_BE_DELETED;
2421 }
2422 patch = patch->next;
2423 }
2424 }
2425
2426 static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
2427 {
2428 struct strbuf buf = STRBUF_INIT;
2429 struct image image;
2430 size_t len;
2431 char *img;
2432 struct patch *tpatch;
2433
2434 if (!(patch->is_copy || patch->is_rename) &&
2435 (tpatch = in_fn_table(patch->old_name)) != NULL && !to_be_deleted(tpatch)) {
2436 if (was_deleted(tpatch)) {
2437 return error("patch %s has been renamed/deleted",
2438 patch->old_name);
2439 }
2440 /* We have a patched copy in memory use that */
2441 strbuf_add(&buf, tpatch->result, tpatch->resultsize);
2442 } else if (cached) {
2443 if (read_file_or_gitlink(ce, &buf))
2444 return error("read of %s failed", patch->old_name);
2445 } else if (patch->old_name) {
2446 if (S_ISGITLINK(patch->old_mode)) {
2447 if (ce) {
2448 read_file_or_gitlink(ce, &buf);
2449 } else {
2450 /*
2451 * There is no way to apply subproject
2452 * patch without looking at the index.
2453 */
2454 patch->fragments = NULL;
2455 }
2456 } else {
2457 if (read_old_data(st, patch->old_name, &buf))
2458 return error("read of %s failed", patch->old_name);
2459 }
2460 }
2461
2462 img = strbuf_detach(&buf, &len);
2463 prepare_image(&image, img, len, !patch->is_binary);
2464
2465 if (apply_fragments(&image, patch) < 0)
2466 return -1; /* note with --reject this succeeds. */
2467 patch->result = image.buf;
2468 patch->resultsize = image.len;
2469 add_to_fn_table(patch);
2470 free(image.line_allocated);
2471
2472 if (0 < patch->is_delete && patch->resultsize)
2473 return error("removal patch leaves file contents");
2474
2475 return 0;
2476 }
2477
2478 static int check_to_create_blob(const char *new_name, int ok_if_exists)
2479 {
2480 struct stat nst;
2481 if (!lstat(new_name, &nst)) {
2482 if (S_ISDIR(nst.st_mode) || ok_if_exists)
2483 return 0;
2484 /*
2485 * A leading component of new_name might be a symlink
2486 * that is going to be removed with this patch, but
2487 * still pointing at somewhere that has the path.
2488 * In such a case, path "new_name" does not exist as
2489 * far as git is concerned.
2490 */
2491 if (has_symlink_leading_path(new_name, strlen(new_name)))
2492 return 0;
2493
2494 return error("%s: already exists in working directory", new_name);
2495 }
2496 else if ((errno != ENOENT) && (errno != ENOTDIR))
2497 return error("%s: %s", new_name, strerror(errno));
2498 return 0;
2499 }
2500
2501 static int verify_index_match(struct cache_entry *ce, struct stat *st)
2502 {
2503 if (S_ISGITLINK(ce->ce_mode)) {
2504 if (!S_ISDIR(st->st_mode))
2505 return -1;
2506 return 0;
2507 }
2508 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
2509 }
2510
2511 static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)
2512 {
2513 const char *old_name = patch->old_name;
2514 struct patch *tpatch = NULL;
2515 int stat_ret = 0;
2516 unsigned st_mode = 0;
2517
2518 /*
2519 * Make sure that we do not have local modifications from the
2520 * index when we are looking at the index. Also make sure
2521 * we have the preimage file to be patched in the work tree,
2522 * unless --cached, which tells git to apply only in the index.
2523 */
2524 if (!old_name)
2525 return 0;
2526
2527 assert(patch->is_new <= 0);
2528
2529 if (!(patch->is_copy || patch->is_rename) &&
2530 (tpatch = in_fn_table(old_name)) != NULL && !to_be_deleted(tpatch)) {
2531 if (was_deleted(tpatch))
2532 return error("%s: has been deleted/renamed", old_name);
2533 st_mode = tpatch->new_mode;
2534 } else if (!cached) {
2535 stat_ret = lstat(old_name, st);
2536 if (stat_ret && errno != ENOENT)
2537 return error("%s: %s", old_name, strerror(errno));
2538 }
2539
2540 if (to_be_deleted(tpatch))
2541 tpatch = NULL;
2542
2543 if (check_index && !tpatch) {
2544 int pos = cache_name_pos(old_name, strlen(old_name));
2545 if (pos < 0) {
2546 if (patch->is_new < 0)
2547 goto is_new;
2548 return error("%s: does not exist in index", old_name);
2549 }
2550 *ce = active_cache[pos];
2551 if (stat_ret < 0) {
2552 struct checkout costate;
2553 /* checkout */
2554 costate.base_dir = "";
2555 costate.base_dir_len = 0;
2556 costate.force = 0;
2557 costate.quiet = 0;
2558 costate.not_new = 0;
2559 costate.refresh_cache = 1;
2560 if (checkout_entry(*ce, &costate, NULL) ||
2561 lstat(old_name, st))
2562 return -1;
2563 }
2564 if (!cached && verify_index_match(*ce, st))
2565 return error("%s: does not match index", old_name);
2566 if (cached)
2567 st_mode = (*ce)->ce_mode;
2568 } else if (stat_ret < 0) {
2569 if (patch->is_new < 0)
2570 goto is_new;
2571 return error("%s: %s", old_name, strerror(errno));
2572 }
2573
2574 if (!cached && !tpatch)
2575 st_mode = ce_mode_from_stat(*ce, st->st_mode);
2576
2577 if (patch->is_new < 0)
2578 patch->is_new = 0;
2579 if (!patch->old_mode)
2580 patch->old_mode = st_mode;
2581 if ((st_mode ^ patch->old_mode) & S_IFMT)
2582 return error("%s: wrong type", old_name);
2583 if (st_mode != patch->old_mode)
2584 warning("%s has type %o, expected %o",
2585 old_name, st_mode, patch->old_mode);
2586 if (!patch->new_mode && !patch->is_delete)
2587 patch->new_mode = st_mode;
2588 return 0;
2589
2590 is_new:
2591 patch->is_new = 1;
2592 patch->is_delete = 0;
2593 patch->old_name = NULL;
2594 return 0;
2595 }
2596
2597 static int check_patch(struct patch *patch)
2598 {
2599 struct stat st;
2600 const char *old_name = patch->old_name;
2601 const char *new_name = patch->new_name;
2602 const char *name = old_name ? old_name : new_name;
2603 struct cache_entry *ce = NULL;
2604 struct patch *tpatch;
2605 int ok_if_exists;
2606 int status;
2607
2608 patch->rejected = 1; /* we will drop this after we succeed */
2609
2610 status = check_preimage(patch, &ce, &st);
2611 if (status)
2612 return status;
2613 old_name = patch->old_name;
2614
2615 if ((tpatch = in_fn_table(new_name)) &&
2616 (was_deleted(tpatch) || to_be_deleted(tpatch)))
2617 /*
2618 * A type-change diff is always split into a patch to
2619 * delete old, immediately followed by a patch to
2620 * create new (see diff.c::run_diff()); in such a case
2621 * it is Ok that the entry to be deleted by the
2622 * previous patch is still in the working tree and in
2623 * the index.
2624 */
2625 ok_if_exists = 1;
2626 else
2627 ok_if_exists = 0;
2628
2629 if (new_name &&
2630 ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
2631 if (check_index &&
2632 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
2633 !ok_if_exists)
2634 return error("%s: already exists in index", new_name);
2635 if (!cached) {
2636 int err = check_to_create_blob(new_name, ok_if_exists);
2637 if (err)
2638 return err;
2639 }
2640 if (!patch->new_mode) {
2641 if (0 < patch->is_new)
2642 patch->new_mode = S_IFREG | 0644;
2643 else
2644 patch->new_mode = patch->old_mode;
2645 }
2646 }
2647
2648 if (new_name && old_name) {
2649 int same = !strcmp(old_name, new_name);
2650 if (!patch->new_mode)
2651 patch->new_mode = patch->old_mode;
2652 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
2653 return error("new mode (%o) of %s does not match old mode (%o)%s%s",
2654 patch->new_mode, new_name, patch->old_mode,
2655 same ? "" : " of ", same ? "" : old_name);
2656 }
2657
2658 if (apply_data(patch, &st, ce) < 0)
2659 return error("%s: patch does not apply", name);
2660 patch->rejected = 0;
2661 return 0;
2662 }
2663
2664 static int check_patch_list(struct patch *patch)
2665 {
2666 int err = 0;
2667
2668 prepare_fn_table(patch);
2669 while (patch) {
2670 if (apply_verbosely)
2671 say_patch_name(stderr,
2672 "Checking patch ", patch, "...\n");
2673 err |= check_patch(patch);
2674 patch = patch->next;
2675 }
2676 return err;
2677 }
2678
2679 /* This function tries to read the sha1 from the current index */
2680 static int get_current_sha1(const char *path, unsigned char *sha1)
2681 {
2682 int pos;
2683
2684 if (read_cache() < 0)
2685 return -1;
2686 pos = cache_name_pos(path, strlen(path));
2687 if (pos < 0)
2688 return -1;
2689 hashcpy(sha1, active_cache[pos]->sha1);
2690 return 0;
2691 }
2692
2693 /* Build an index that contains the just the files needed for a 3way merge */
2694 static void build_fake_ancestor(struct patch *list, const char *filename)
2695 {
2696 struct patch *patch;
2697 struct index_state result = { NULL };
2698 int fd;
2699
2700 /* Once we start supporting the reverse patch, it may be
2701 * worth showing the new sha1 prefix, but until then...
2702 */
2703 for (patch = list; patch; patch = patch->next) {
2704 const unsigned char *sha1_ptr;
2705 unsigned char sha1[20];
2706 struct cache_entry *ce;
2707 const char *name;
2708
2709 name = patch->old_name ? patch->old_name : patch->new_name;
2710 if (0 < patch->is_new)
2711 continue;
2712 else if (get_sha1(patch->old_sha1_prefix, sha1))
2713 /* git diff has no index line for mode/type changes */
2714 if (!patch->lines_added && !patch->lines_deleted) {
2715 if (get_current_sha1(patch->new_name, sha1) ||
2716 get_current_sha1(patch->old_name, sha1))
2717 die("mode change for %s, which is not "
2718 "in current HEAD", name);
2719 sha1_ptr = sha1;
2720 } else
2721 die("sha1 information is lacking or useless "
2722 "(%s).", name);
2723 else
2724 sha1_ptr = sha1;
2725
2726 ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);
2727 if (!ce)
2728 die("make_cache_entry failed for path '%s'", name);
2729 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
2730 die ("Could not add %s to temporary index", name);
2731 }
2732
2733 fd = open(filename, O_WRONLY | O_CREAT, 0666);
2734 if (fd < 0 || write_index(&result, fd) || close(fd))
2735 die ("Could not write temporary index to %s", filename);
2736
2737 discard_index(&result);
2738 }
2739
2740 static void stat_patch_list(struct patch *patch)
2741 {
2742 int files, adds, dels;
2743
2744 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
2745 files++;
2746 adds += patch->lines_added;
2747 dels += patch->lines_deleted;
2748 show_stats(patch);
2749 }
2750
2751 printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
2752 }
2753
2754 static void numstat_patch_list(struct patch *patch)
2755 {
2756 for ( ; patch; patch = patch->next) {
2757 const char *name;
2758 name = patch->new_name ? patch->new_name : patch->old_name;
2759 if (patch->is_binary)
2760 printf("-\t-\t");
2761 else
2762 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
2763 write_name_quoted(name, stdout, line_termination);
2764 }
2765 }
2766
2767 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
2768 {
2769 if (mode)
2770 printf(" %s mode %06o %s\n", newdelete, mode, name);
2771 else
2772 printf(" %s %s\n", newdelete, name);
2773 }
2774
2775 static void show_mode_change(struct patch *p, int show_name)
2776 {
2777 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
2778 if (show_name)
2779 printf(" mode change %06o => %06o %s\n",
2780 p->old_mode, p->new_mode, p->new_name);
2781 else
2782 printf(" mode change %06o => %06o\n",
2783 p->old_mode, p->new_mode);
2784 }
2785 }
2786
2787 static void show_rename_copy(struct patch *p)
2788 {
2789 const char *renamecopy = p->is_rename ? "rename" : "copy";
2790 const char *old, *new;
2791
2792 /* Find common prefix */
2793 old = p->old_name;
2794 new = p->new_name;
2795 while (1) {
2796 const char *slash_old, *slash_new;
2797 slash_old = strchr(old, '/');
2798 slash_new = strchr(new, '/');
2799 if (!slash_old ||
2800 !slash_new ||
2801 slash_old - old != slash_new - new ||
2802 memcmp(old, new, slash_new - new))
2803 break;
2804 old = slash_old + 1;
2805 new = slash_new + 1;
2806 }
2807 /* p->old_name thru old is the common prefix, and old and new
2808 * through the end of names are renames
2809 */
2810 if (old != p->old_name)
2811 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
2812 (int)(old - p->old_name), p->old_name,
2813 old, new, p->score);
2814 else
2815 printf(" %s %s => %s (%d%%)\n", renamecopy,
2816 p->old_name, p->new_name, p->score);
2817 show_mode_change(p, 0);
2818 }
2819
2820 static void summary_patch_list(struct patch *patch)
2821 {
2822 struct patch *p;
2823
2824 for (p = patch; p; p = p->next) {
2825 if (p->is_new)
2826 show_file_mode_name("create", p->new_mode, p->new_name);
2827 else if (p->is_delete)
2828 show_file_mode_name("delete", p->old_mode, p->old_name);
2829 else {
2830 if (p->is_rename || p->is_copy)
2831 show_rename_copy(p);
2832 else {
2833 if (p->score) {
2834 printf(" rewrite %s (%d%%)\n",
2835 p->new_name, p->score);
2836 show_mode_change(p, 0);
2837 }
2838 else
2839 show_mode_change(p, 1);
2840 }
2841 }
2842 }
2843 }
2844
2845 static void patch_stats(struct patch *patch)
2846 {
2847 int lines = patch->lines_added + patch->lines_deleted;
2848
2849 if (lines > max_change)
2850 max_change = lines;
2851 if (patch->old_name) {
2852 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
2853 if (!len)
2854 len = strlen(patch->old_name);
2855 if (len > max_len)
2856 max_len = len;
2857 }
2858 if (patch->new_name) {
2859 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
2860 if (!len)
2861 len = strlen(patch->new_name);
2862 if (len > max_len)
2863 max_len = len;
2864 }
2865 }
2866
2867 static void remove_file(struct patch *patch, int rmdir_empty)
2868 {
2869 if (update_index) {
2870 if (remove_file_from_cache(patch->old_name) < 0)
2871 die("unable to remove %s from index", patch->old_name);
2872 }
2873 if (!cached) {
2874 if (S_ISGITLINK(patch->old_mode)) {
2875 if (rmdir(patch->old_name))
2876 warning("unable to remove submodule %s",
2877 patch->old_name);
2878 } else if (!unlink_or_warn(patch->old_name) && rmdir_empty) {
2879 remove_path(patch->old_name);
2880 }
2881 }
2882 }
2883
2884 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
2885 {
2886 struct stat st;
2887 struct cache_entry *ce;
2888 int namelen = strlen(path);
2889 unsigned ce_size = cache_entry_size(namelen);
2890
2891 if (!update_index)
2892 return;
2893
2894 ce = xcalloc(1, ce_size);
2895 memcpy(ce->name, path, namelen);
2896 ce->ce_mode = create_ce_mode(mode);
2897 ce->ce_flags = namelen;
2898 if (S_ISGITLINK(mode)) {
2899 const char *s = buf;
2900
2901 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
2902 die("corrupt patch for subproject %s", path);
2903 } else {
2904 if (!cached) {
2905 if (lstat(path, &st) < 0)
2906 die_errno("unable to stat newly created file '%s'",
2907 path);
2908 fill_stat_cache_info(ce, &st);
2909 }
2910 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
2911 die("unable to create backing store for newly created file %s", path);
2912 }
2913 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
2914 die("unable to add cache entry for %s", path);
2915 }
2916
2917 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
2918 {
2919 int fd;
2920 struct strbuf nbuf = STRBUF_INIT;
2921
2922 if (S_ISGITLINK(mode)) {
2923 struct stat st;
2924 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
2925 return 0;
2926 return mkdir(path, 0777);
2927 }
2928
2929 if (has_symlinks && S_ISLNK(mode))
2930 /* Although buf:size is counted string, it also is NUL
2931 * terminated.
2932 */
2933 return symlink(buf, path);
2934
2935 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
2936 if (fd < 0)
2937 return -1;
2938
2939 if (convert_to_working_tree(path, buf, size, &nbuf)) {
2940 size = nbuf.len;
2941 buf = nbuf.buf;
2942 }
2943 write_or_die(fd, buf, size);
2944 strbuf_release(&nbuf);
2945
2946 if (close(fd) < 0)
2947 die_errno("closing file '%s'", path);
2948 return 0;
2949 }
2950
2951 /*
2952 * We optimistically assume that the directories exist,
2953 * which is true 99% of the time anyway. If they don't,
2954 * we create them and try again.
2955 */
2956 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
2957 {
2958 if (cached)
2959 return;
2960 if (!try_create_file(path, mode, buf, size))
2961 return;
2962
2963 if (errno == ENOENT) {
2964 if (safe_create_leading_directories(path))
2965 return;
2966 if (!try_create_file(path, mode, buf, size))
2967 return;
2968 }
2969
2970 if (errno == EEXIST || errno == EACCES) {
2971 /* We may be trying to create a file where a directory
2972 * used to be.
2973 */
2974 struct stat st;
2975 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
2976 errno = EEXIST;
2977 }
2978
2979 if (errno == EEXIST) {
2980 unsigned int nr = getpid();
2981
2982 for (;;) {
2983 char newpath[PATH_MAX];
2984 mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
2985 if (!try_create_file(newpath, mode, buf, size)) {
2986 if (!rename(newpath, path))
2987 return;
2988 unlink_or_warn(newpath);
2989 break;
2990 }
2991 if (errno != EEXIST)
2992 break;
2993 ++nr;
2994 }
2995 }
2996 die_errno("unable to write file '%s' mode %o", path, mode);
2997 }
2998
2999 static void create_file(struct patch *patch)
3000 {
3001 char *path = patch->new_name;
3002 unsigned mode = patch->new_mode;
3003 unsigned long size = patch->resultsize;
3004 char *buf = patch->result;
3005
3006 if (!mode)
3007 mode = S_IFREG | 0644;
3008 create_one_file(path, mode, buf, size);
3009 add_index_file(path, mode, buf, size);
3010 }
3011
3012 /* phase zero is to remove, phase one is to create */
3013 static void write_out_one_result(struct patch *patch, int phase)
3014 {
3015 if (patch->is_delete > 0) {
3016 if (phase == 0)
3017 remove_file(patch, 1);
3018 return;
3019 }
3020 if (patch->is_new > 0 || patch->is_copy) {
3021 if (phase == 1)
3022 create_file(patch);
3023 return;
3024 }
3025 /*
3026 * Rename or modification boils down to the same
3027 * thing: remove the old, write the new
3028 */
3029 if (phase == 0)
3030 remove_file(patch, patch->is_rename);
3031 if (phase == 1)
3032 create_file(patch);
3033 }
3034
3035 static int write_out_one_reject(struct patch *patch)
3036 {
3037 FILE *rej;
3038 char namebuf[PATH_MAX];
3039 struct fragment *frag;
3040 int cnt = 0;
3041
3042 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
3043 if (!frag->rejected)
3044 continue;
3045 cnt++;
3046 }
3047
3048 if (!cnt) {
3049 if (apply_verbosely)
3050 say_patch_name(stderr,
3051 "Applied patch ", patch, " cleanly.\n");
3052 return 0;
3053 }
3054
3055 /* This should not happen, because a removal patch that leaves
3056 * contents are marked "rejected" at the patch level.
3057 */
3058 if (!patch->new_name)
3059 die("internal error");
3060
3061 /* Say this even without --verbose */
3062 say_patch_name(stderr, "Applying patch ", patch, " with");
3063 fprintf(stderr, " %d rejects...\n", cnt);
3064
3065 cnt = strlen(patch->new_name);
3066 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
3067 cnt = ARRAY_SIZE(namebuf) - 5;
3068 warning("truncating .rej filename to %.*s.rej",
3069 cnt - 1, patch->new_name);
3070 }
3071 memcpy(namebuf, patch->new_name, cnt);
3072 memcpy(namebuf + cnt, ".rej", 5);
3073
3074 rej = fopen(namebuf, "w");
3075 if (!rej)
3076 return error("cannot open %s: %s", namebuf, strerror(errno));
3077
3078 /* Normal git tools never deal with .rej, so do not pretend
3079 * this is a git patch by saying --git nor give extended
3080 * headers. While at it, maybe please "kompare" that wants
3081 * the trailing TAB and some garbage at the end of line ;-).
3082 */
3083 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
3084 patch->new_name, patch->new_name);
3085 for (cnt = 1, frag = patch->fragments;
3086 frag;
3087 cnt++, frag = frag->next) {
3088 if (!frag->rejected) {
3089 fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);
3090 continue;
3091 }
3092 fprintf(stderr, "Rejected hunk #%d.\n", cnt);
3093 fprintf(rej, "%.*s", frag->size, frag->patch);
3094 if (frag->patch[frag->size-1] != '\n')
3095 fputc('\n', rej);
3096 }
3097 fclose(rej);
3098 return -1;
3099 }
3100
3101 static int write_out_results(struct patch *list, int skipped_patch)
3102 {
3103 int phase;
3104 int errs = 0;
3105 struct patch *l;
3106
3107 if (!list && !skipped_patch)
3108 return error("No changes");
3109
3110 for (phase = 0; phase < 2; phase++) {
3111 l = list;
3112 while (l) {
3113 if (l->rejected)
3114 errs = 1;
3115 else {
3116 write_out_one_result(l, phase);
3117 if (phase == 1 && write_out_one_reject(l))
3118 errs = 1;
3119 }
3120 l = l->next;
3121 }
3122 }
3123 return errs;
3124 }
3125
3126 static struct lock_file lock_file;
3127
3128 static struct string_list limit_by_name;
3129 static int has_include;
3130 static void add_name_limit(const char *name, int exclude)
3131 {
3132 struct string_list_item *it;
3133
3134 it = string_list_append(name, &limit_by_name);
3135 it->util = exclude ? NULL : (void *) 1;
3136 }
3137
3138 static int use_patch(struct patch *p)
3139 {
3140 const char *pathname = p->new_name ? p->new_name : p->old_name;
3141 int i;
3142
3143 /* Paths outside are not touched regardless of "--include" */
3144 if (0 < prefix_length) {
3145 int pathlen = strlen(pathname);
3146 if (pathlen <= prefix_length ||
3147 memcmp(prefix, pathname, prefix_length))
3148 return 0;
3149 }
3150
3151 /* See if it matches any of exclude/include rule */
3152 for (i = 0; i < limit_by_name.nr; i++) {
3153 struct string_list_item *it = &limit_by_name.items[i];
3154 if (!fnmatch(it->string, pathname, 0))
3155 return (it->util != NULL);
3156 }
3157
3158 /*
3159 * If we had any include, a path that does not match any rule is
3160 * not used. Otherwise, we saw bunch of exclude rules (or none)
3161 * and such a path is used.
3162 */
3163 return !has_include;
3164 }
3165
3166
3167 static void prefix_one(char **name)
3168 {
3169 char *old_name = *name;
3170 if (!old_name)
3171 return;
3172 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
3173 free(old_name);
3174 }
3175
3176 static void prefix_patches(struct patch *p)
3177 {
3178 if (!prefix || p->is_toplevel_relative)
3179 return;
3180 for ( ; p; p = p->next) {
3181 if (p->new_name == p->old_name) {
3182 char *prefixed = p->new_name;
3183 prefix_one(&prefixed);
3184 p->new_name = p->old_name = prefixed;
3185 }
3186 else {
3187 prefix_one(&p->new_name);
3188 prefix_one(&p->old_name);
3189 }
3190 }
3191 }
3192
3193 #define INACCURATE_EOF (1<<0)
3194 #define RECOUNT (1<<1)
3195
3196 static int apply_patch(int fd, const char *filename, int options)
3197 {
3198 size_t offset;
3199 struct strbuf buf = STRBUF_INIT;
3200 struct patch *list = NULL, **listp = &list;
3201 int skipped_patch = 0;
3202
3203 /* FIXME - memory leak when using multiple patch files as inputs */
3204 memset(&fn_table, 0, sizeof(struct string_list));
3205 patch_input_file = filename;
3206 read_patch_file(&buf, fd);
3207 offset = 0;
3208 while (offset < buf.len) {
3209 struct patch *patch;
3210 int nr;
3211
3212 patch = xcalloc(1, sizeof(*patch));
3213 patch->inaccurate_eof = !!(options & INACCURATE_EOF);
3214 patch->recount = !!(options & RECOUNT);
3215 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
3216 if (nr < 0)
3217 break;
3218 if (apply_in_reverse)
3219 reverse_patches(patch);
3220 if (prefix)
3221 prefix_patches(patch);
3222 if (use_patch(patch)) {
3223 patch_stats(patch);
3224 *listp = patch;
3225 listp = &patch->next;
3226 }
3227 else {
3228 /* perhaps free it a bit better? */
3229 free(patch);
3230 skipped_patch++;
3231 }
3232 offset += nr;
3233 }
3234
3235 if (whitespace_error && (ws_error_action == die_on_ws_error))
3236 apply = 0;
3237
3238 update_index = check_index && apply;
3239 if (update_index && newfd < 0)
3240 newfd = hold_locked_index(&lock_file, 1);
3241
3242 if (check_index) {
3243 if (read_cache() < 0)
3244 die("unable to read index file");
3245 }
3246
3247 if ((check || apply) &&
3248 check_patch_list(list) < 0 &&
3249 !apply_with_reject)
3250 exit(1);
3251
3252 if (apply && write_out_results(list, skipped_patch))
3253 exit(1);
3254
3255 if (fake_ancestor)
3256 build_fake_ancestor(list, fake_ancestor);
3257
3258 if (diffstat)
3259 stat_patch_list(list);
3260
3261 if (numstat)
3262 numstat_patch_list(list);
3263
3264 if (summary)
3265 summary_patch_list(list);
3266
3267 strbuf_release(&buf);
3268 return 0;
3269 }
3270
3271 static int git_apply_config(const char *var, const char *value, void *cb)
3272 {
3273 if (!strcmp(var, "apply.whitespace"))
3274 return git_config_string(&apply_default_whitespace, var, value);
3275 return git_default_config(var, value, cb);
3276 }
3277
3278 static int option_parse_exclude(const struct option *opt,
3279 const char *arg, int unset)
3280 {
3281 add_name_limit(arg, 1);
3282 return 0;
3283 }
3284
3285 static int option_parse_include(const struct option *opt,
3286 const char *arg, int unset)
3287 {
3288 add_name_limit(arg, 0);
3289 has_include = 1;
3290 return 0;
3291 }
3292
3293 static int option_parse_p(const struct option *opt,
3294 const char *arg, int unset)
3295 {
3296 p_value = atoi(arg);
3297 p_value_known = 1;
3298 return 0;
3299 }
3300
3301 static int option_parse_z(const struct option *opt,
3302 const char *arg, int unset)
3303 {
3304 if (unset)
3305 line_termination = '\n';
3306 else
3307 line_termination = 0;
3308 return 0;
3309 }
3310
3311 static int option_parse_whitespace(const struct option *opt,
3312 const char *arg, int unset)
3313 {
3314 const char **whitespace_option = opt->value;
3315
3316 *whitespace_option = arg;
3317 parse_whitespace_option(arg);
3318 return 0;
3319 }
3320
3321 static int option_parse_directory(const struct option *opt,
3322 const char *arg, int unset)
3323 {
3324 root_len = strlen(arg);
3325 if (root_len && arg[root_len - 1] != '/') {
3326 char *new_root;
3327 root = new_root = xmalloc(root_len + 2);
3328 strcpy(new_root, arg);
3329 strcpy(new_root + root_len++, "/");
3330 } else
3331 root = arg;
3332 return 0;
3333 }
3334
3335 int cmd_apply(int argc, const char **argv, const char *unused_prefix)
3336 {
3337 int i;
3338 int errs = 0;
3339 int is_not_gitdir;
3340 int binary;
3341 int force_apply = 0;
3342
3343 const char *whitespace_option = NULL;
3344
3345 struct option builtin_apply_options[] = {
3346 { OPTION_CALLBACK, 0, "exclude", NULL, "path",
3347 "don't apply changes matching the given path",
3348 0, option_parse_exclude },
3349 { OPTION_CALLBACK, 0, "include", NULL, "path",
3350 "apply changes matching the given path",
3351 0, option_parse_include },
3352 { OPTION_CALLBACK, 'p', NULL, NULL, "num",
3353 "remove <num> leading slashes from traditional diff paths",
3354 0, option_parse_p },
3355 OPT_BOOLEAN(0, "no-add", &no_add,
3356 "ignore additions made by the patch"),
3357 OPT_BOOLEAN(0, "stat", &diffstat,
3358 "instead of applying the patch, output diffstat for the input"),
3359 { OPTION_BOOLEAN, 0, "allow-binary-replacement", &binary,
3360 NULL, "old option, now no-op",
3361 PARSE_OPT_HIDDEN | PARSE_OPT_NOARG },
3362 { OPTION_BOOLEAN, 0, "binary", &binary,
3363 NULL, "old option, now no-op",
3364 PARSE_OPT_HIDDEN | PARSE_OPT_NOARG },
3365 OPT_BOOLEAN(0, "numstat", &numstat,
3366 "shows number of added and deleted lines in decimal notation"),
3367 OPT_BOOLEAN(0, "summary", &summary,
3368 "instead of applying the patch, output a summary for the input"),
3369 OPT_BOOLEAN(0, "check", &check,
3370 "instead of applying the patch, see if the patch is applicable"),
3371 OPT_BOOLEAN(0, "index", &check_index,
3372 "make sure the patch is applicable to the current index"),
3373 OPT_BOOLEAN(0, "cached", &cached,
3374 "apply a patch without touching the working tree"),
3375 OPT_BOOLEAN(0, "apply", &force_apply,
3376 "also apply the patch (use with --stat/--summary/--check)"),
3377 OPT_FILENAME(0, "build-fake-ancestor", &fake_ancestor,
3378 "build a temporary index based on embedded index information"),
3379 { OPTION_CALLBACK, 'z', NULL, NULL, NULL,
3380 "paths are separated with NUL character",
3381 PARSE_OPT_NOARG, option_parse_z },
3382 OPT_INTEGER('C', NULL, &p_context,
3383 "ensure at least <n> lines of context match"),
3384 { OPTION_CALLBACK, 0, "whitespace", &whitespace_option, "action",
3385 "detect new or modified lines that have whitespace errors",
3386 0, option_parse_whitespace },
3387 OPT_BOOLEAN('R', "reverse", &apply_in_reverse,
3388 "apply the patch in reverse"),
3389 OPT_BOOLEAN(0, "unidiff-zero", &unidiff_zero,
3390 "don't expect at least one line of context"),
3391 OPT_BOOLEAN(0, "reject", &apply_with_reject,
3392 "leave the rejected hunks in corresponding *.rej files"),
3393 OPT__VERBOSE(&apply_verbosely),
3394 OPT_BIT(0, "inaccurate-eof", &options,
3395 "tolerate incorrectly detected missing new-line at the end of file",
3396 INACCURATE_EOF),
3397 OPT_BIT(0, "recount", &options,
3398 "do not trust the line counts in the hunk headers",
3399 RECOUNT),
3400 { OPTION_CALLBACK, 0, "directory", NULL, "root",
3401 "prepend <root> to all filenames",
3402 0, option_parse_directory },
3403 OPT_END()
3404 };
3405
3406 prefix = setup_git_directory_gently(&is_not_gitdir);
3407 prefix_length = prefix ? strlen(prefix) : 0;
3408 git_config(git_apply_config, NULL);
3409 if (apply_default_whitespace)
3410 parse_whitespace_option(apply_default_whitespace);
3411
3412 argc = parse_options(argc, argv, prefix, builtin_apply_options,
3413 apply_usage, 0);
3414
3415 if (apply_with_reject)
3416 apply = apply_verbosely = 1;
3417 if (!force_apply && (diffstat || numstat || summary || check || fake_ancestor))
3418 apply = 0;
3419 if (check_index && is_not_gitdir)
3420 die("--index outside a repository");
3421 if (cached) {
3422 if (is_not_gitdir)
3423 die("--cached outside a repository");
3424 check_index = 1;
3425 }
3426 for (i = 0; i < argc; i++) {
3427 const char *arg = argv[i];
3428 int fd;
3429
3430 if (!strcmp(arg, "-")) {
3431 errs |= apply_patch(0, "<stdin>", options);
3432 read_stdin = 0;
3433 continue;
3434 } else if (0 < prefix_length)
3435 arg = prefix_filename(prefix, prefix_length, arg);
3436
3437 fd = open(arg, O_RDONLY);
3438 if (fd < 0)
3439 die_errno("can't open patch '%s'", arg);
3440 read_stdin = 0;
3441 set_default_whitespace_mode(whitespace_option);
3442 errs |= apply_patch(fd, arg, options);
3443 close(fd);
3444 }
3445 set_default_whitespace_mode(whitespace_option);
3446 if (read_stdin)
3447 errs |= apply_patch(0, "<stdin>", options);
3448 if (whitespace_error) {
3449 if (squelch_whitespace_errors &&
3450 squelch_whitespace_errors < whitespace_error) {
3451 int squelched =
3452 whitespace_error - squelch_whitespace_errors;
3453 warning("squelched %d "
3454 "whitespace error%s",
3455 squelched,
3456 squelched == 1 ? "" : "s");
3457 }
3458 if (ws_error_action == die_on_ws_error)
3459 die("%d line%s add%s whitespace errors.",
3460 whitespace_error,
3461 whitespace_error == 1 ? "" : "s",
3462 whitespace_error == 1 ? "s" : "");
3463 if (applied_after_fixing_ws && apply)
3464 warning("%d line%s applied after"
3465 " fixing whitespace errors.",
3466 applied_after_fixing_ws,
3467 applied_after_fixing_ws == 1 ? "" : "s");
3468 else if (whitespace_error)
3469 warning("%d line%s add%s whitespace errors.",
3470 whitespace_error,
3471 whitespace_error == 1 ? "" : "s",
3472 whitespace_error == 1 ? "s" : "");
3473 }
3474
3475 if (update_index) {
3476 if (write_cache(newfd, active_cache, active_nr) ||
3477 commit_locked_index(&lock_file))
3478 die("Unable to write new index file");
3479 }
3480
3481 return !!errs;
3482 }