Merge branch 'pc/remove-warn' into maint
[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
65 static enum ws_ignore {
66 ignore_ws_none,
67 ignore_ws_change,
68 } ws_ignore_action = ignore_ws_none;
69
70
71 static const char *patch_input_file;
72 static const char *root;
73 static int root_len;
74 static int read_stdin = 1;
75 static int options;
76
77 static void parse_whitespace_option(const char *option)
78 {
79 if (!option) {
80 ws_error_action = warn_on_ws_error;
81 return;
82 }
83 if (!strcmp(option, "warn")) {
84 ws_error_action = warn_on_ws_error;
85 return;
86 }
87 if (!strcmp(option, "nowarn")) {
88 ws_error_action = nowarn_ws_error;
89 return;
90 }
91 if (!strcmp(option, "error")) {
92 ws_error_action = die_on_ws_error;
93 return;
94 }
95 if (!strcmp(option, "error-all")) {
96 ws_error_action = die_on_ws_error;
97 squelch_whitespace_errors = 0;
98 return;
99 }
100 if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
101 ws_error_action = correct_ws_error;
102 return;
103 }
104 die("unrecognized whitespace option '%s'", option);
105 }
106
107 static void parse_ignorewhitespace_option(const char *option)
108 {
109 if (!option || !strcmp(option, "no") ||
110 !strcmp(option, "false") || !strcmp(option, "never") ||
111 !strcmp(option, "none")) {
112 ws_ignore_action = ignore_ws_none;
113 return;
114 }
115 if (!strcmp(option, "change")) {
116 ws_ignore_action = ignore_ws_change;
117 return;
118 }
119 die("unrecognized whitespace ignore option '%s'", option);
120 }
121
122 static void set_default_whitespace_mode(const char *whitespace_option)
123 {
124 if (!whitespace_option && !apply_default_whitespace)
125 ws_error_action = (apply ? warn_on_ws_error : nowarn_ws_error);
126 }
127
128 /*
129 * For "diff-stat" like behaviour, we keep track of the biggest change
130 * we've seen, and the longest filename. That allows us to do simple
131 * scaling.
132 */
133 static int max_change, max_len;
134
135 /*
136 * Various "current state", notably line numbers and what
137 * file (and how) we're patching right now.. The "is_xxxx"
138 * things are flags, where -1 means "don't know yet".
139 */
140 static int linenr = 1;
141
142 /*
143 * This represents one "hunk" from a patch, starting with
144 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
145 * patch text is pointed at by patch, and its byte length
146 * is stored in size. leading and trailing are the number
147 * of context lines.
148 */
149 struct fragment {
150 unsigned long leading, trailing;
151 unsigned long oldpos, oldlines;
152 unsigned long newpos, newlines;
153 const char *patch;
154 int size;
155 int rejected;
156 int linenr;
157 struct fragment *next;
158 };
159
160 /*
161 * When dealing with a binary patch, we reuse "leading" field
162 * to store the type of the binary hunk, either deflated "delta"
163 * or deflated "literal".
164 */
165 #define binary_patch_method leading
166 #define BINARY_DELTA_DEFLATED 1
167 #define BINARY_LITERAL_DEFLATED 2
168
169 /*
170 * This represents a "patch" to a file, both metainfo changes
171 * such as creation/deletion, filemode and content changes represented
172 * as a series of fragments.
173 */
174 struct patch {
175 char *new_name, *old_name, *def_name;
176 unsigned int old_mode, new_mode;
177 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */
178 int rejected;
179 unsigned ws_rule;
180 unsigned long deflate_origlen;
181 int lines_added, lines_deleted;
182 int score;
183 unsigned int is_toplevel_relative:1;
184 unsigned int inaccurate_eof:1;
185 unsigned int is_binary:1;
186 unsigned int is_copy:1;
187 unsigned int is_rename:1;
188 unsigned int recount:1;
189 struct fragment *fragments;
190 char *result;
191 size_t resultsize;
192 char old_sha1_prefix[41];
193 char new_sha1_prefix[41];
194 struct patch *next;
195 };
196
197 /*
198 * A line in a file, len-bytes long (includes the terminating LF,
199 * except for an incomplete line at the end if the file ends with
200 * one), and its contents hashes to 'hash'.
201 */
202 struct line {
203 size_t len;
204 unsigned hash : 24;
205 unsigned flag : 8;
206 #define LINE_COMMON 1
207 };
208
209 /*
210 * This represents a "file", which is an array of "lines".
211 */
212 struct image {
213 char *buf;
214 size_t len;
215 size_t nr;
216 size_t alloc;
217 struct line *line_allocated;
218 struct line *line;
219 };
220
221 /*
222 * Records filenames that have been touched, in order to handle
223 * the case where more than one patches touch the same file.
224 */
225
226 static struct string_list fn_table;
227
228 static uint32_t hash_line(const char *cp, size_t len)
229 {
230 size_t i;
231 uint32_t h;
232 for (i = 0, h = 0; i < len; i++) {
233 if (!isspace(cp[i])) {
234 h = h * 3 + (cp[i] & 0xff);
235 }
236 }
237 return h;
238 }
239
240 /*
241 * Compare lines s1 of length n1 and s2 of length n2, ignoring
242 * whitespace difference. Returns 1 if they match, 0 otherwise
243 */
244 static int fuzzy_matchlines(const char *s1, size_t n1,
245 const char *s2, size_t n2)
246 {
247 const char *last1 = s1 + n1 - 1;
248 const char *last2 = s2 + n2 - 1;
249 int result = 0;
250
251 if (n1 < 0 || n2 < 0)
252 return 0;
253
254 /* ignore line endings */
255 while ((*last1 == '\r') || (*last1 == '\n'))
256 last1--;
257 while ((*last2 == '\r') || (*last2 == '\n'))
258 last2--;
259
260 /* skip leading whitespace */
261 while (isspace(*s1) && (s1 <= last1))
262 s1++;
263 while (isspace(*s2) && (s2 <= last2))
264 s2++;
265 /* early return if both lines are empty */
266 if ((s1 > last1) && (s2 > last2))
267 return 1;
268 while (!result) {
269 result = *s1++ - *s2++;
270 /*
271 * Skip whitespace inside. We check for whitespace on
272 * both buffers because we don't want "a b" to match
273 * "ab"
274 */
275 if (isspace(*s1) && isspace(*s2)) {
276 while (isspace(*s1) && s1 <= last1)
277 s1++;
278 while (isspace(*s2) && s2 <= last2)
279 s2++;
280 }
281 /*
282 * If we reached the end on one side only,
283 * lines don't match
284 */
285 if (
286 ((s2 > last2) && (s1 <= last1)) ||
287 ((s1 > last1) && (s2 <= last2)))
288 return 0;
289 if ((s1 > last1) && (s2 > last2))
290 break;
291 }
292
293 return !result;
294 }
295
296 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
297 {
298 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
299 img->line_allocated[img->nr].len = len;
300 img->line_allocated[img->nr].hash = hash_line(bol, len);
301 img->line_allocated[img->nr].flag = flag;
302 img->nr++;
303 }
304
305 static void prepare_image(struct image *image, char *buf, size_t len,
306 int prepare_linetable)
307 {
308 const char *cp, *ep;
309
310 memset(image, 0, sizeof(*image));
311 image->buf = buf;
312 image->len = len;
313
314 if (!prepare_linetable)
315 return;
316
317 ep = image->buf + image->len;
318 cp = image->buf;
319 while (cp < ep) {
320 const char *next;
321 for (next = cp; next < ep && *next != '\n'; next++)
322 ;
323 if (next < ep)
324 next++;
325 add_line_info(image, cp, next - cp, 0);
326 cp = next;
327 }
328 image->line = image->line_allocated;
329 }
330
331 static void clear_image(struct image *image)
332 {
333 free(image->buf);
334 image->buf = NULL;
335 image->len = 0;
336 }
337
338 static void say_patch_name(FILE *output, const char *pre,
339 struct patch *patch, const char *post)
340 {
341 fputs(pre, output);
342 if (patch->old_name && patch->new_name &&
343 strcmp(patch->old_name, patch->new_name)) {
344 quote_c_style(patch->old_name, NULL, output, 0);
345 fputs(" => ", output);
346 quote_c_style(patch->new_name, NULL, output, 0);
347 } else {
348 const char *n = patch->new_name;
349 if (!n)
350 n = patch->old_name;
351 quote_c_style(n, NULL, output, 0);
352 }
353 fputs(post, output);
354 }
355
356 #define CHUNKSIZE (8192)
357 #define SLOP (16)
358
359 static void read_patch_file(struct strbuf *sb, int fd)
360 {
361 if (strbuf_read(sb, fd, 0) < 0)
362 die_errno("git apply: failed to read");
363
364 /*
365 * Make sure that we have some slop in the buffer
366 * so that we can do speculative "memcmp" etc, and
367 * see to it that it is NUL-filled.
368 */
369 strbuf_grow(sb, SLOP);
370 memset(sb->buf + sb->len, 0, SLOP);
371 }
372
373 static unsigned long linelen(const char *buffer, unsigned long size)
374 {
375 unsigned long len = 0;
376 while (size--) {
377 len++;
378 if (*buffer++ == '\n')
379 break;
380 }
381 return len;
382 }
383
384 static int is_dev_null(const char *str)
385 {
386 return !memcmp("/dev/null", str, 9) && isspace(str[9]);
387 }
388
389 #define TERM_SPACE 1
390 #define TERM_TAB 2
391
392 static int name_terminate(const char *name, int namelen, int c, int terminate)
393 {
394 if (c == ' ' && !(terminate & TERM_SPACE))
395 return 0;
396 if (c == '\t' && !(terminate & TERM_TAB))
397 return 0;
398
399 return 1;
400 }
401
402 /* remove double slashes to make --index work with such filenames */
403 static char *squash_slash(char *name)
404 {
405 int i = 0, j = 0;
406
407 if (!name)
408 return NULL;
409
410 while (name[i]) {
411 if ((name[j++] = name[i++]) == '/')
412 while (name[i] == '/')
413 i++;
414 }
415 name[j] = '\0';
416 return name;
417 }
418
419 static char *find_name(const char *line, char *def, int p_value, int terminate)
420 {
421 int len;
422 const char *start = NULL;
423
424 if (p_value == 0)
425 start = line;
426
427 if (*line == '"') {
428 struct strbuf name = STRBUF_INIT;
429
430 /*
431 * Proposed "new-style" GNU patch/diff format; see
432 * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
433 */
434 if (!unquote_c_style(&name, line, NULL)) {
435 char *cp;
436
437 for (cp = name.buf; p_value; p_value--) {
438 cp = strchr(cp, '/');
439 if (!cp)
440 break;
441 cp++;
442 }
443 if (cp) {
444 /* name can later be freed, so we need
445 * to memmove, not just return cp
446 */
447 strbuf_remove(&name, 0, cp - name.buf);
448 free(def);
449 if (root)
450 strbuf_insert(&name, 0, root, root_len);
451 return squash_slash(strbuf_detach(&name, NULL));
452 }
453 }
454 strbuf_release(&name);
455 }
456
457 for (;;) {
458 char c = *line;
459
460 if (isspace(c)) {
461 if (c == '\n')
462 break;
463 if (name_terminate(start, line-start, c, terminate))
464 break;
465 }
466 line++;
467 if (c == '/' && !--p_value)
468 start = line;
469 }
470 if (!start)
471 return squash_slash(def);
472 len = line - start;
473 if (!len)
474 return squash_slash(def);
475
476 /*
477 * Generally we prefer the shorter name, especially
478 * if the other one is just a variation of that with
479 * something else tacked on to the end (ie "file.orig"
480 * or "file~").
481 */
482 if (def) {
483 int deflen = strlen(def);
484 if (deflen < len && !strncmp(start, def, deflen))
485 return squash_slash(def);
486 free(def);
487 }
488
489 if (root) {
490 char *ret = xmalloc(root_len + len + 1);
491 strcpy(ret, root);
492 memcpy(ret + root_len, start, len);
493 ret[root_len + len] = '\0';
494 return squash_slash(ret);
495 }
496
497 return squash_slash(xmemdupz(start, len));
498 }
499
500 static int count_slashes(const char *cp)
501 {
502 int cnt = 0;
503 char ch;
504
505 while ((ch = *cp++))
506 if (ch == '/')
507 cnt++;
508 return cnt;
509 }
510
511 /*
512 * Given the string after "--- " or "+++ ", guess the appropriate
513 * p_value for the given patch.
514 */
515 static int guess_p_value(const char *nameline)
516 {
517 char *name, *cp;
518 int val = -1;
519
520 if (is_dev_null(nameline))
521 return -1;
522 name = find_name(nameline, NULL, 0, TERM_SPACE | TERM_TAB);
523 if (!name)
524 return -1;
525 cp = strchr(name, '/');
526 if (!cp)
527 val = 0;
528 else if (prefix) {
529 /*
530 * Does it begin with "a/$our-prefix" and such? Then this is
531 * very likely to apply to our directory.
532 */
533 if (!strncmp(name, prefix, prefix_length))
534 val = count_slashes(prefix);
535 else {
536 cp++;
537 if (!strncmp(cp, prefix, prefix_length))
538 val = count_slashes(prefix) + 1;
539 }
540 }
541 free(name);
542 return val;
543 }
544
545 /*
546 * Does the ---/+++ line has the POSIX timestamp after the last HT?
547 * GNU diff puts epoch there to signal a creation/deletion event. Is
548 * this such a timestamp?
549 */
550 static int has_epoch_timestamp(const char *nameline)
551 {
552 /*
553 * We are only interested in epoch timestamp; any non-zero
554 * fraction cannot be one, hence "(\.0+)?" in the regexp below.
555 * For the same reason, the date must be either 1969-12-31 or
556 * 1970-01-01, and the seconds part must be "00".
557 */
558 const char stamp_regexp[] =
559 "^(1969-12-31|1970-01-01)"
560 " "
561 "[0-2][0-9]:[0-5][0-9]:00(\\.0+)?"
562 " "
563 "([-+][0-2][0-9][0-5][0-9])\n";
564 const char *timestamp = NULL, *cp;
565 static regex_t *stamp;
566 regmatch_t m[10];
567 int zoneoffset;
568 int hourminute;
569 int status;
570
571 for (cp = nameline; *cp != '\n'; cp++) {
572 if (*cp == '\t')
573 timestamp = cp + 1;
574 }
575 if (!timestamp)
576 return 0;
577 if (!stamp) {
578 stamp = xmalloc(sizeof(*stamp));
579 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
580 warning("Cannot prepare timestamp regexp %s",
581 stamp_regexp);
582 return 0;
583 }
584 }
585
586 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
587 if (status) {
588 if (status != REG_NOMATCH)
589 warning("regexec returned %d for input: %s",
590 status, timestamp);
591 return 0;
592 }
593
594 zoneoffset = strtol(timestamp + m[3].rm_so + 1, NULL, 10);
595 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
596 if (timestamp[m[3].rm_so] == '-')
597 zoneoffset = -zoneoffset;
598
599 /*
600 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
601 * (west of GMT) or 1970-01-01 (east of GMT)
602 */
603 if ((zoneoffset < 0 && memcmp(timestamp, "1969-12-31", 10)) ||
604 (0 <= zoneoffset && memcmp(timestamp, "1970-01-01", 10)))
605 return 0;
606
607 hourminute = (strtol(timestamp + 11, NULL, 10) * 60 +
608 strtol(timestamp + 14, NULL, 10) -
609 zoneoffset);
610
611 return ((zoneoffset < 0 && hourminute == 1440) ||
612 (0 <= zoneoffset && !hourminute));
613 }
614
615 /*
616 * Get the name etc info from the ---/+++ lines of a traditional patch header
617 *
618 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
619 * files, we can happily check the index for a match, but for creating a
620 * new file we should try to match whatever "patch" does. I have no idea.
621 */
622 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
623 {
624 char *name;
625
626 first += 4; /* skip "--- " */
627 second += 4; /* skip "+++ " */
628 if (!p_value_known) {
629 int p, q;
630 p = guess_p_value(first);
631 q = guess_p_value(second);
632 if (p < 0) p = q;
633 if (0 <= p && p == q) {
634 p_value = p;
635 p_value_known = 1;
636 }
637 }
638 if (is_dev_null(first)) {
639 patch->is_new = 1;
640 patch->is_delete = 0;
641 name = find_name(second, NULL, p_value, TERM_SPACE | TERM_TAB);
642 patch->new_name = name;
643 } else if (is_dev_null(second)) {
644 patch->is_new = 0;
645 patch->is_delete = 1;
646 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
647 patch->old_name = name;
648 } else {
649 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
650 name = find_name(second, name, p_value, TERM_SPACE | TERM_TAB);
651 if (has_epoch_timestamp(first)) {
652 patch->is_new = 1;
653 patch->is_delete = 0;
654 patch->new_name = name;
655 } else if (has_epoch_timestamp(second)) {
656 patch->is_new = 0;
657 patch->is_delete = 1;
658 patch->old_name = name;
659 } else {
660 patch->old_name = patch->new_name = name;
661 }
662 }
663 if (!name)
664 die("unable to find filename in patch at line %d", linenr);
665 }
666
667 static int gitdiff_hdrend(const char *line, struct patch *patch)
668 {
669 return -1;
670 }
671
672 /*
673 * We're anal about diff header consistency, to make
674 * sure that we don't end up having strange ambiguous
675 * patches floating around.
676 *
677 * As a result, gitdiff_{old|new}name() will check
678 * their names against any previous information, just
679 * to make sure..
680 */
681 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
682 {
683 if (!orig_name && !isnull)
684 return find_name(line, NULL, p_value, TERM_TAB);
685
686 if (orig_name) {
687 int len;
688 const char *name;
689 char *another;
690 name = orig_name;
691 len = strlen(name);
692 if (isnull)
693 die("git apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
694 another = find_name(line, NULL, p_value, TERM_TAB);
695 if (!another || memcmp(another, name, len + 1))
696 die("git apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
697 free(another);
698 return orig_name;
699 }
700 else {
701 /* expect "/dev/null" */
702 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
703 die("git apply: bad git-diff - expected /dev/null on line %d", linenr);
704 return NULL;
705 }
706 }
707
708 static int gitdiff_oldname(const char *line, struct patch *patch)
709 {
710 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
711 return 0;
712 }
713
714 static int gitdiff_newname(const char *line, struct patch *patch)
715 {
716 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
717 return 0;
718 }
719
720 static int gitdiff_oldmode(const char *line, struct patch *patch)
721 {
722 patch->old_mode = strtoul(line, NULL, 8);
723 return 0;
724 }
725
726 static int gitdiff_newmode(const char *line, struct patch *patch)
727 {
728 patch->new_mode = strtoul(line, NULL, 8);
729 return 0;
730 }
731
732 static int gitdiff_delete(const char *line, struct patch *patch)
733 {
734 patch->is_delete = 1;
735 patch->old_name = patch->def_name;
736 return gitdiff_oldmode(line, patch);
737 }
738
739 static int gitdiff_newfile(const char *line, struct patch *patch)
740 {
741 patch->is_new = 1;
742 patch->new_name = patch->def_name;
743 return gitdiff_newmode(line, patch);
744 }
745
746 static int gitdiff_copysrc(const char *line, struct patch *patch)
747 {
748 patch->is_copy = 1;
749 patch->old_name = find_name(line, NULL, 0, 0);
750 return 0;
751 }
752
753 static int gitdiff_copydst(const char *line, struct patch *patch)
754 {
755 patch->is_copy = 1;
756 patch->new_name = find_name(line, NULL, 0, 0);
757 return 0;
758 }
759
760 static int gitdiff_renamesrc(const char *line, struct patch *patch)
761 {
762 patch->is_rename = 1;
763 patch->old_name = find_name(line, NULL, 0, 0);
764 return 0;
765 }
766
767 static int gitdiff_renamedst(const char *line, struct patch *patch)
768 {
769 patch->is_rename = 1;
770 patch->new_name = find_name(line, NULL, 0, 0);
771 return 0;
772 }
773
774 static int gitdiff_similarity(const char *line, struct patch *patch)
775 {
776 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
777 patch->score = 0;
778 return 0;
779 }
780
781 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
782 {
783 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
784 patch->score = 0;
785 return 0;
786 }
787
788 static int gitdiff_index(const char *line, struct patch *patch)
789 {
790 /*
791 * index line is N hexadecimal, "..", N hexadecimal,
792 * and optional space with octal mode.
793 */
794 const char *ptr, *eol;
795 int len;
796
797 ptr = strchr(line, '.');
798 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
799 return 0;
800 len = ptr - line;
801 memcpy(patch->old_sha1_prefix, line, len);
802 patch->old_sha1_prefix[len] = 0;
803
804 line = ptr + 2;
805 ptr = strchr(line, ' ');
806 eol = strchr(line, '\n');
807
808 if (!ptr || eol < ptr)
809 ptr = eol;
810 len = ptr - line;
811
812 if (40 < len)
813 return 0;
814 memcpy(patch->new_sha1_prefix, line, len);
815 patch->new_sha1_prefix[len] = 0;
816 if (*ptr == ' ')
817 patch->old_mode = strtoul(ptr+1, NULL, 8);
818 return 0;
819 }
820
821 /*
822 * This is normal for a diff that doesn't change anything: we'll fall through
823 * into the next diff. Tell the parser to break out.
824 */
825 static int gitdiff_unrecognized(const char *line, struct patch *patch)
826 {
827 return -1;
828 }
829
830 static const char *stop_at_slash(const char *line, int llen)
831 {
832 int nslash = p_value;
833 int i;
834
835 for (i = 0; i < llen; i++) {
836 int ch = line[i];
837 if (ch == '/' && --nslash <= 0)
838 return &line[i];
839 }
840 return NULL;
841 }
842
843 /*
844 * This is to extract the same name that appears on "diff --git"
845 * line. We do not find and return anything if it is a rename
846 * patch, and it is OK because we will find the name elsewhere.
847 * We need to reliably find name only when it is mode-change only,
848 * creation or deletion of an empty file. In any of these cases,
849 * both sides are the same name under a/ and b/ respectively.
850 */
851 static char *git_header_name(char *line, int llen)
852 {
853 const char *name;
854 const char *second = NULL;
855 size_t len;
856
857 line += strlen("diff --git ");
858 llen -= strlen("diff --git ");
859
860 if (*line == '"') {
861 const char *cp;
862 struct strbuf first = STRBUF_INIT;
863 struct strbuf sp = STRBUF_INIT;
864
865 if (unquote_c_style(&first, line, &second))
866 goto free_and_fail1;
867
868 /* advance to the first slash */
869 cp = stop_at_slash(first.buf, first.len);
870 /* we do not accept absolute paths */
871 if (!cp || cp == first.buf)
872 goto free_and_fail1;
873 strbuf_remove(&first, 0, cp + 1 - first.buf);
874
875 /*
876 * second points at one past closing dq of name.
877 * find the second name.
878 */
879 while ((second < line + llen) && isspace(*second))
880 second++;
881
882 if (line + llen <= second)
883 goto free_and_fail1;
884 if (*second == '"') {
885 if (unquote_c_style(&sp, second, NULL))
886 goto free_and_fail1;
887 cp = stop_at_slash(sp.buf, sp.len);
888 if (!cp || cp == sp.buf)
889 goto free_and_fail1;
890 /* They must match, otherwise ignore */
891 if (strcmp(cp + 1, first.buf))
892 goto free_and_fail1;
893 strbuf_release(&sp);
894 return strbuf_detach(&first, NULL);
895 }
896
897 /* unquoted second */
898 cp = stop_at_slash(second, line + llen - second);
899 if (!cp || cp == second)
900 goto free_and_fail1;
901 cp++;
902 if (line + llen - cp != first.len + 1 ||
903 memcmp(first.buf, cp, first.len))
904 goto free_and_fail1;
905 return strbuf_detach(&first, NULL);
906
907 free_and_fail1:
908 strbuf_release(&first);
909 strbuf_release(&sp);
910 return NULL;
911 }
912
913 /* unquoted first name */
914 name = stop_at_slash(line, llen);
915 if (!name || name == line)
916 return NULL;
917 name++;
918
919 /*
920 * since the first name is unquoted, a dq if exists must be
921 * the beginning of the second name.
922 */
923 for (second = name; second < line + llen; second++) {
924 if (*second == '"') {
925 struct strbuf sp = STRBUF_INIT;
926 const char *np;
927
928 if (unquote_c_style(&sp, second, NULL))
929 goto free_and_fail2;
930
931 np = stop_at_slash(sp.buf, sp.len);
932 if (!np || np == sp.buf)
933 goto free_and_fail2;
934 np++;
935
936 len = sp.buf + sp.len - np;
937 if (len < second - name &&
938 !strncmp(np, name, len) &&
939 isspace(name[len])) {
940 /* Good */
941 strbuf_remove(&sp, 0, np - sp.buf);
942 return strbuf_detach(&sp, NULL);
943 }
944
945 free_and_fail2:
946 strbuf_release(&sp);
947 return NULL;
948 }
949 }
950
951 /*
952 * Accept a name only if it shows up twice, exactly the same
953 * form.
954 */
955 for (len = 0 ; ; len++) {
956 switch (name[len]) {
957 default:
958 continue;
959 case '\n':
960 return NULL;
961 case '\t': case ' ':
962 second = name+len;
963 for (;;) {
964 char c = *second++;
965 if (c == '\n')
966 return NULL;
967 if (c == '/')
968 break;
969 }
970 if (second[len] == '\n' && !memcmp(name, second, len)) {
971 return xmemdupz(name, len);
972 }
973 }
974 }
975 }
976
977 /* Verify that we recognize the lines following a git header */
978 static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
979 {
980 unsigned long offset;
981
982 /* A git diff has explicit new/delete information, so we don't guess */
983 patch->is_new = 0;
984 patch->is_delete = 0;
985
986 /*
987 * Some things may not have the old name in the
988 * rest of the headers anywhere (pure mode changes,
989 * or removing or adding empty files), so we get
990 * the default name from the header.
991 */
992 patch->def_name = git_header_name(line, len);
993 if (patch->def_name && root) {
994 char *s = xmalloc(root_len + strlen(patch->def_name) + 1);
995 strcpy(s, root);
996 strcpy(s + root_len, patch->def_name);
997 free(patch->def_name);
998 patch->def_name = s;
999 }
1000
1001 line += len;
1002 size -= len;
1003 linenr++;
1004 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
1005 static const struct opentry {
1006 const char *str;
1007 int (*fn)(const char *, struct patch *);
1008 } optable[] = {
1009 { "@@ -", gitdiff_hdrend },
1010 { "--- ", gitdiff_oldname },
1011 { "+++ ", gitdiff_newname },
1012 { "old mode ", gitdiff_oldmode },
1013 { "new mode ", gitdiff_newmode },
1014 { "deleted file mode ", gitdiff_delete },
1015 { "new file mode ", gitdiff_newfile },
1016 { "copy from ", gitdiff_copysrc },
1017 { "copy to ", gitdiff_copydst },
1018 { "rename old ", gitdiff_renamesrc },
1019 { "rename new ", gitdiff_renamedst },
1020 { "rename from ", gitdiff_renamesrc },
1021 { "rename to ", gitdiff_renamedst },
1022 { "similarity index ", gitdiff_similarity },
1023 { "dissimilarity index ", gitdiff_dissimilarity },
1024 { "index ", gitdiff_index },
1025 { "", gitdiff_unrecognized },
1026 };
1027 int i;
1028
1029 len = linelen(line, size);
1030 if (!len || line[len-1] != '\n')
1031 break;
1032 for (i = 0; i < ARRAY_SIZE(optable); i++) {
1033 const struct opentry *p = optable + i;
1034 int oplen = strlen(p->str);
1035 if (len < oplen || memcmp(p->str, line, oplen))
1036 continue;
1037 if (p->fn(line + oplen, patch) < 0)
1038 return offset;
1039 break;
1040 }
1041 }
1042
1043 return offset;
1044 }
1045
1046 static int parse_num(const char *line, unsigned long *p)
1047 {
1048 char *ptr;
1049
1050 if (!isdigit(*line))
1051 return 0;
1052 *p = strtoul(line, &ptr, 10);
1053 return ptr - line;
1054 }
1055
1056 static int parse_range(const char *line, int len, int offset, const char *expect,
1057 unsigned long *p1, unsigned long *p2)
1058 {
1059 int digits, ex;
1060
1061 if (offset < 0 || offset >= len)
1062 return -1;
1063 line += offset;
1064 len -= offset;
1065
1066 digits = parse_num(line, p1);
1067 if (!digits)
1068 return -1;
1069
1070 offset += digits;
1071 line += digits;
1072 len -= digits;
1073
1074 *p2 = 1;
1075 if (*line == ',') {
1076 digits = parse_num(line+1, p2);
1077 if (!digits)
1078 return -1;
1079
1080 offset += digits+1;
1081 line += digits+1;
1082 len -= digits+1;
1083 }
1084
1085 ex = strlen(expect);
1086 if (ex > len)
1087 return -1;
1088 if (memcmp(line, expect, ex))
1089 return -1;
1090
1091 return offset + ex;
1092 }
1093
1094 static void recount_diff(char *line, int size, struct fragment *fragment)
1095 {
1096 int oldlines = 0, newlines = 0, ret = 0;
1097
1098 if (size < 1) {
1099 warning("recount: ignore empty hunk");
1100 return;
1101 }
1102
1103 for (;;) {
1104 int len = linelen(line, size);
1105 size -= len;
1106 line += len;
1107
1108 if (size < 1)
1109 break;
1110
1111 switch (*line) {
1112 case ' ': case '\n':
1113 newlines++;
1114 /* fall through */
1115 case '-':
1116 oldlines++;
1117 continue;
1118 case '+':
1119 newlines++;
1120 continue;
1121 case '\\':
1122 continue;
1123 case '@':
1124 ret = size < 3 || prefixcmp(line, "@@ ");
1125 break;
1126 case 'd':
1127 ret = size < 5 || prefixcmp(line, "diff ");
1128 break;
1129 default:
1130 ret = -1;
1131 break;
1132 }
1133 if (ret) {
1134 warning("recount: unexpected line: %.*s",
1135 (int)linelen(line, size), line);
1136 return;
1137 }
1138 break;
1139 }
1140 fragment->oldlines = oldlines;
1141 fragment->newlines = newlines;
1142 }
1143
1144 /*
1145 * Parse a unified diff fragment header of the
1146 * form "@@ -a,b +c,d @@"
1147 */
1148 static int parse_fragment_header(char *line, int len, struct fragment *fragment)
1149 {
1150 int offset;
1151
1152 if (!len || line[len-1] != '\n')
1153 return -1;
1154
1155 /* Figure out the number of lines in a fragment */
1156 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1157 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1158
1159 return offset;
1160 }
1161
1162 static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
1163 {
1164 unsigned long offset, len;
1165
1166 patch->is_toplevel_relative = 0;
1167 patch->is_rename = patch->is_copy = 0;
1168 patch->is_new = patch->is_delete = -1;
1169 patch->old_mode = patch->new_mode = 0;
1170 patch->old_name = patch->new_name = NULL;
1171 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
1172 unsigned long nextlen;
1173
1174 len = linelen(line, size);
1175 if (!len)
1176 break;
1177
1178 /* Testing this early allows us to take a few shortcuts.. */
1179 if (len < 6)
1180 continue;
1181
1182 /*
1183 * Make sure we don't find any unconnected patch fragments.
1184 * That's a sign that we didn't find a header, and that a
1185 * patch has become corrupted/broken up.
1186 */
1187 if (!memcmp("@@ -", line, 4)) {
1188 struct fragment dummy;
1189 if (parse_fragment_header(line, len, &dummy) < 0)
1190 continue;
1191 die("patch fragment without header at line %d: %.*s",
1192 linenr, (int)len-1, line);
1193 }
1194
1195 if (size < len + 6)
1196 break;
1197
1198 /*
1199 * Git patch? It might not have a real patch, just a rename
1200 * or mode change, so we handle that specially
1201 */
1202 if (!memcmp("diff --git ", line, 11)) {
1203 int git_hdr_len = parse_git_header(line, len, size, patch);
1204 if (git_hdr_len <= len)
1205 continue;
1206 if (!patch->old_name && !patch->new_name) {
1207 if (!patch->def_name)
1208 die("git diff header lacks filename information when removing "
1209 "%d leading pathname components (line %d)" , p_value, linenr);
1210 patch->old_name = patch->new_name = patch->def_name;
1211 }
1212 patch->is_toplevel_relative = 1;
1213 *hdrsize = git_hdr_len;
1214 return offset;
1215 }
1216
1217 /* --- followed by +++ ? */
1218 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1219 continue;
1220
1221 /*
1222 * We only accept unified patches, so we want it to
1223 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1224 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1225 */
1226 nextlen = linelen(line + len, size - len);
1227 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1228 continue;
1229
1230 /* Ok, we'll consider it a patch */
1231 parse_traditional_patch(line, line+len, patch);
1232 *hdrsize = len + nextlen;
1233 linenr += 2;
1234 return offset;
1235 }
1236 return -1;
1237 }
1238
1239 static void record_ws_error(unsigned result, const char *line, int len, int linenr)
1240 {
1241 char *err;
1242
1243 if (!result)
1244 return;
1245
1246 whitespace_error++;
1247 if (squelch_whitespace_errors &&
1248 squelch_whitespace_errors < whitespace_error)
1249 return;
1250
1251 err = whitespace_error_string(result);
1252 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1253 patch_input_file, linenr, err, len, line);
1254 free(err);
1255 }
1256
1257 static void check_whitespace(const char *line, int len, unsigned ws_rule)
1258 {
1259 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1260
1261 record_ws_error(result, line + 1, len - 2, linenr);
1262 }
1263
1264 /*
1265 * Parse a unified diff. Note that this really needs to parse each
1266 * fragment separately, since the only way to know the difference
1267 * between a "---" that is part of a patch, and a "---" that starts
1268 * the next patch is to look at the line counts..
1269 */
1270 static int parse_fragment(char *line, unsigned long size,
1271 struct patch *patch, struct fragment *fragment)
1272 {
1273 int added, deleted;
1274 int len = linelen(line, size), offset;
1275 unsigned long oldlines, newlines;
1276 unsigned long leading, trailing;
1277
1278 offset = parse_fragment_header(line, len, fragment);
1279 if (offset < 0)
1280 return -1;
1281 if (offset > 0 && patch->recount)
1282 recount_diff(line + offset, size - offset, fragment);
1283 oldlines = fragment->oldlines;
1284 newlines = fragment->newlines;
1285 leading = 0;
1286 trailing = 0;
1287
1288 /* Parse the thing.. */
1289 line += len;
1290 size -= len;
1291 linenr++;
1292 added = deleted = 0;
1293 for (offset = len;
1294 0 < size;
1295 offset += len, size -= len, line += len, linenr++) {
1296 if (!oldlines && !newlines)
1297 break;
1298 len = linelen(line, size);
1299 if (!len || line[len-1] != '\n')
1300 return -1;
1301 switch (*line) {
1302 default:
1303 return -1;
1304 case '\n': /* newer GNU diff, an empty context line */
1305 case ' ':
1306 oldlines--;
1307 newlines--;
1308 if (!deleted && !added)
1309 leading++;
1310 trailing++;
1311 break;
1312 case '-':
1313 if (apply_in_reverse &&
1314 ws_error_action != nowarn_ws_error)
1315 check_whitespace(line, len, patch->ws_rule);
1316 deleted++;
1317 oldlines--;
1318 trailing = 0;
1319 break;
1320 case '+':
1321 if (!apply_in_reverse &&
1322 ws_error_action != nowarn_ws_error)
1323 check_whitespace(line, len, patch->ws_rule);
1324 added++;
1325 newlines--;
1326 trailing = 0;
1327 break;
1328
1329 /*
1330 * We allow "\ No newline at end of file". Depending
1331 * on locale settings when the patch was produced we
1332 * don't know what this line looks like. The only
1333 * thing we do know is that it begins with "\ ".
1334 * Checking for 12 is just for sanity check -- any
1335 * l10n of "\ No newline..." is at least that long.
1336 */
1337 case '\\':
1338 if (len < 12 || memcmp(line, "\\ ", 2))
1339 return -1;
1340 break;
1341 }
1342 }
1343 if (oldlines || newlines)
1344 return -1;
1345 fragment->leading = leading;
1346 fragment->trailing = trailing;
1347
1348 /*
1349 * If a fragment ends with an incomplete line, we failed to include
1350 * it in the above loop because we hit oldlines == newlines == 0
1351 * before seeing it.
1352 */
1353 if (12 < size && !memcmp(line, "\\ ", 2))
1354 offset += linelen(line, size);
1355
1356 patch->lines_added += added;
1357 patch->lines_deleted += deleted;
1358
1359 if (0 < patch->is_new && oldlines)
1360 return error("new file depends on old contents");
1361 if (0 < patch->is_delete && newlines)
1362 return error("deleted file still has contents");
1363 return offset;
1364 }
1365
1366 static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
1367 {
1368 unsigned long offset = 0;
1369 unsigned long oldlines = 0, newlines = 0, context = 0;
1370 struct fragment **fragp = &patch->fragments;
1371
1372 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1373 struct fragment *fragment;
1374 int len;
1375
1376 fragment = xcalloc(1, sizeof(*fragment));
1377 fragment->linenr = linenr;
1378 len = parse_fragment(line, size, patch, fragment);
1379 if (len <= 0)
1380 die("corrupt patch at line %d", linenr);
1381 fragment->patch = line;
1382 fragment->size = len;
1383 oldlines += fragment->oldlines;
1384 newlines += fragment->newlines;
1385 context += fragment->leading + fragment->trailing;
1386
1387 *fragp = fragment;
1388 fragp = &fragment->next;
1389
1390 offset += len;
1391 line += len;
1392 size -= len;
1393 }
1394
1395 /*
1396 * If something was removed (i.e. we have old-lines) it cannot
1397 * be creation, and if something was added it cannot be
1398 * deletion. However, the reverse is not true; --unified=0
1399 * patches that only add are not necessarily creation even
1400 * though they do not have any old lines, and ones that only
1401 * delete are not necessarily deletion.
1402 *
1403 * Unfortunately, a real creation/deletion patch do _not_ have
1404 * any context line by definition, so we cannot safely tell it
1405 * apart with --unified=0 insanity. At least if the patch has
1406 * more than one hunk it is not creation or deletion.
1407 */
1408 if (patch->is_new < 0 &&
1409 (oldlines || (patch->fragments && patch->fragments->next)))
1410 patch->is_new = 0;
1411 if (patch->is_delete < 0 &&
1412 (newlines || (patch->fragments && patch->fragments->next)))
1413 patch->is_delete = 0;
1414
1415 if (0 < patch->is_new && oldlines)
1416 die("new file %s depends on old contents", patch->new_name);
1417 if (0 < patch->is_delete && newlines)
1418 die("deleted file %s still has contents", patch->old_name);
1419 if (!patch->is_delete && !newlines && context)
1420 fprintf(stderr, "** warning: file %s becomes empty but "
1421 "is not deleted\n", patch->new_name);
1422
1423 return offset;
1424 }
1425
1426 static inline int metadata_changes(struct patch *patch)
1427 {
1428 return patch->is_rename > 0 ||
1429 patch->is_copy > 0 ||
1430 patch->is_new > 0 ||
1431 patch->is_delete ||
1432 (patch->old_mode && patch->new_mode &&
1433 patch->old_mode != patch->new_mode);
1434 }
1435
1436 static char *inflate_it(const void *data, unsigned long size,
1437 unsigned long inflated_size)
1438 {
1439 z_stream stream;
1440 void *out;
1441 int st;
1442
1443 memset(&stream, 0, sizeof(stream));
1444
1445 stream.next_in = (unsigned char *)data;
1446 stream.avail_in = size;
1447 stream.next_out = out = xmalloc(inflated_size);
1448 stream.avail_out = inflated_size;
1449 git_inflate_init(&stream);
1450 st = git_inflate(&stream, Z_FINISH);
1451 git_inflate_end(&stream);
1452 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1453 free(out);
1454 return NULL;
1455 }
1456 return out;
1457 }
1458
1459 static struct fragment *parse_binary_hunk(char **buf_p,
1460 unsigned long *sz_p,
1461 int *status_p,
1462 int *used_p)
1463 {
1464 /*
1465 * Expect a line that begins with binary patch method ("literal"
1466 * or "delta"), followed by the length of data before deflating.
1467 * a sequence of 'length-byte' followed by base-85 encoded data
1468 * should follow, terminated by a newline.
1469 *
1470 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1471 * and we would limit the patch line to 66 characters,
1472 * so one line can fit up to 13 groups that would decode
1473 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1474 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1475 */
1476 int llen, used;
1477 unsigned long size = *sz_p;
1478 char *buffer = *buf_p;
1479 int patch_method;
1480 unsigned long origlen;
1481 char *data = NULL;
1482 int hunk_size = 0;
1483 struct fragment *frag;
1484
1485 llen = linelen(buffer, size);
1486 used = llen;
1487
1488 *status_p = 0;
1489
1490 if (!prefixcmp(buffer, "delta ")) {
1491 patch_method = BINARY_DELTA_DEFLATED;
1492 origlen = strtoul(buffer + 6, NULL, 10);
1493 }
1494 else if (!prefixcmp(buffer, "literal ")) {
1495 patch_method = BINARY_LITERAL_DEFLATED;
1496 origlen = strtoul(buffer + 8, NULL, 10);
1497 }
1498 else
1499 return NULL;
1500
1501 linenr++;
1502 buffer += llen;
1503 while (1) {
1504 int byte_length, max_byte_length, newsize;
1505 llen = linelen(buffer, size);
1506 used += llen;
1507 linenr++;
1508 if (llen == 1) {
1509 /* consume the blank line */
1510 buffer++;
1511 size--;
1512 break;
1513 }
1514 /*
1515 * Minimum line is "A00000\n" which is 7-byte long,
1516 * and the line length must be multiple of 5 plus 2.
1517 */
1518 if ((llen < 7) || (llen-2) % 5)
1519 goto corrupt;
1520 max_byte_length = (llen - 2) / 5 * 4;
1521 byte_length = *buffer;
1522 if ('A' <= byte_length && byte_length <= 'Z')
1523 byte_length = byte_length - 'A' + 1;
1524 else if ('a' <= byte_length && byte_length <= 'z')
1525 byte_length = byte_length - 'a' + 27;
1526 else
1527 goto corrupt;
1528 /* if the input length was not multiple of 4, we would
1529 * have filler at the end but the filler should never
1530 * exceed 3 bytes
1531 */
1532 if (max_byte_length < byte_length ||
1533 byte_length <= max_byte_length - 4)
1534 goto corrupt;
1535 newsize = hunk_size + byte_length;
1536 data = xrealloc(data, newsize);
1537 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1538 goto corrupt;
1539 hunk_size = newsize;
1540 buffer += llen;
1541 size -= llen;
1542 }
1543
1544 frag = xcalloc(1, sizeof(*frag));
1545 frag->patch = inflate_it(data, hunk_size, origlen);
1546 if (!frag->patch)
1547 goto corrupt;
1548 free(data);
1549 frag->size = origlen;
1550 *buf_p = buffer;
1551 *sz_p = size;
1552 *used_p = used;
1553 frag->binary_patch_method = patch_method;
1554 return frag;
1555
1556 corrupt:
1557 free(data);
1558 *status_p = -1;
1559 error("corrupt binary patch at line %d: %.*s",
1560 linenr-1, llen-1, buffer);
1561 return NULL;
1562 }
1563
1564 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1565 {
1566 /*
1567 * We have read "GIT binary patch\n"; what follows is a line
1568 * that says the patch method (currently, either "literal" or
1569 * "delta") and the length of data before deflating; a
1570 * sequence of 'length-byte' followed by base-85 encoded data
1571 * follows.
1572 *
1573 * When a binary patch is reversible, there is another binary
1574 * hunk in the same format, starting with patch method (either
1575 * "literal" or "delta") with the length of data, and a sequence
1576 * of length-byte + base-85 encoded data, terminated with another
1577 * empty line. This data, when applied to the postimage, produces
1578 * the preimage.
1579 */
1580 struct fragment *forward;
1581 struct fragment *reverse;
1582 int status;
1583 int used, used_1;
1584
1585 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1586 if (!forward && !status)
1587 /* there has to be one hunk (forward hunk) */
1588 return error("unrecognized binary patch at line %d", linenr-1);
1589 if (status)
1590 /* otherwise we already gave an error message */
1591 return status;
1592
1593 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1594 if (reverse)
1595 used += used_1;
1596 else if (status) {
1597 /*
1598 * Not having reverse hunk is not an error, but having
1599 * a corrupt reverse hunk is.
1600 */
1601 free((void*) forward->patch);
1602 free(forward);
1603 return status;
1604 }
1605 forward->next = reverse;
1606 patch->fragments = forward;
1607 patch->is_binary = 1;
1608 return used;
1609 }
1610
1611 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1612 {
1613 int hdrsize, patchsize;
1614 int offset = find_header(buffer, size, &hdrsize, patch);
1615
1616 if (offset < 0)
1617 return offset;
1618
1619 patch->ws_rule = whitespace_rule(patch->new_name
1620 ? patch->new_name
1621 : patch->old_name);
1622
1623 patchsize = parse_single_patch(buffer + offset + hdrsize,
1624 size - offset - hdrsize, patch);
1625
1626 if (!patchsize) {
1627 static const char *binhdr[] = {
1628 "Binary files ",
1629 "Files ",
1630 NULL,
1631 };
1632 static const char git_binary[] = "GIT binary patch\n";
1633 int i;
1634 int hd = hdrsize + offset;
1635 unsigned long llen = linelen(buffer + hd, size - hd);
1636
1637 if (llen == sizeof(git_binary) - 1 &&
1638 !memcmp(git_binary, buffer + hd, llen)) {
1639 int used;
1640 linenr++;
1641 used = parse_binary(buffer + hd + llen,
1642 size - hd - llen, patch);
1643 if (used)
1644 patchsize = used + llen;
1645 else
1646 patchsize = 0;
1647 }
1648 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1649 for (i = 0; binhdr[i]; i++) {
1650 int len = strlen(binhdr[i]);
1651 if (len < size - hd &&
1652 !memcmp(binhdr[i], buffer + hd, len)) {
1653 linenr++;
1654 patch->is_binary = 1;
1655 patchsize = llen;
1656 break;
1657 }
1658 }
1659 }
1660
1661 /* Empty patch cannot be applied if it is a text patch
1662 * without metadata change. A binary patch appears
1663 * empty to us here.
1664 */
1665 if ((apply || check) &&
1666 (!patch->is_binary && !metadata_changes(patch)))
1667 die("patch with only garbage at line %d", linenr);
1668 }
1669
1670 return offset + hdrsize + patchsize;
1671 }
1672
1673 #define swap(a,b) myswap((a),(b),sizeof(a))
1674
1675 #define myswap(a, b, size) do { \
1676 unsigned char mytmp[size]; \
1677 memcpy(mytmp, &a, size); \
1678 memcpy(&a, &b, size); \
1679 memcpy(&b, mytmp, size); \
1680 } while (0)
1681
1682 static void reverse_patches(struct patch *p)
1683 {
1684 for (; p; p = p->next) {
1685 struct fragment *frag = p->fragments;
1686
1687 swap(p->new_name, p->old_name);
1688 swap(p->new_mode, p->old_mode);
1689 swap(p->is_new, p->is_delete);
1690 swap(p->lines_added, p->lines_deleted);
1691 swap(p->old_sha1_prefix, p->new_sha1_prefix);
1692
1693 for (; frag; frag = frag->next) {
1694 swap(frag->newpos, frag->oldpos);
1695 swap(frag->newlines, frag->oldlines);
1696 }
1697 }
1698 }
1699
1700 static const char pluses[] =
1701 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1702 static const char minuses[]=
1703 "----------------------------------------------------------------------";
1704
1705 static void show_stats(struct patch *patch)
1706 {
1707 struct strbuf qname = STRBUF_INIT;
1708 char *cp = patch->new_name ? patch->new_name : patch->old_name;
1709 int max, add, del;
1710
1711 quote_c_style(cp, &qname, NULL, 0);
1712
1713 /*
1714 * "scale" the filename
1715 */
1716 max = max_len;
1717 if (max > 50)
1718 max = 50;
1719
1720 if (qname.len > max) {
1721 cp = strchr(qname.buf + qname.len + 3 - max, '/');
1722 if (!cp)
1723 cp = qname.buf + qname.len + 3 - max;
1724 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
1725 }
1726
1727 if (patch->is_binary) {
1728 printf(" %-*s | Bin\n", max, qname.buf);
1729 strbuf_release(&qname);
1730 return;
1731 }
1732
1733 printf(" %-*s |", max, qname.buf);
1734 strbuf_release(&qname);
1735
1736 /*
1737 * scale the add/delete
1738 */
1739 max = max + max_change > 70 ? 70 - max : max_change;
1740 add = patch->lines_added;
1741 del = patch->lines_deleted;
1742
1743 if (max_change > 0) {
1744 int total = ((add + del) * max + max_change / 2) / max_change;
1745 add = (add * max + max_change / 2) / max_change;
1746 del = total - add;
1747 }
1748 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
1749 add, pluses, del, minuses);
1750 }
1751
1752 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
1753 {
1754 switch (st->st_mode & S_IFMT) {
1755 case S_IFLNK:
1756 if (strbuf_readlink(buf, path, st->st_size) < 0)
1757 return error("unable to read symlink %s", path);
1758 return 0;
1759 case S_IFREG:
1760 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
1761 return error("unable to open or read %s", path);
1762 convert_to_git(path, buf->buf, buf->len, buf, 0);
1763 return 0;
1764 default:
1765 return -1;
1766 }
1767 }
1768
1769 /*
1770 * Update the preimage, and the common lines in postimage,
1771 * from buffer buf of length len. If postlen is 0 the postimage
1772 * is updated in place, otherwise it's updated on a new buffer
1773 * of length postlen
1774 */
1775
1776 static void update_pre_post_images(struct image *preimage,
1777 struct image *postimage,
1778 char *buf,
1779 size_t len, size_t postlen)
1780 {
1781 int i, ctx;
1782 char *new, *old, *fixed;
1783 struct image fixed_preimage;
1784
1785 /*
1786 * Update the preimage with whitespace fixes. Note that we
1787 * are not losing preimage->buf -- apply_one_fragment() will
1788 * free "oldlines".
1789 */
1790 prepare_image(&fixed_preimage, buf, len, 1);
1791 assert(fixed_preimage.nr == preimage->nr);
1792 for (i = 0; i < preimage->nr; i++)
1793 fixed_preimage.line[i].flag = preimage->line[i].flag;
1794 free(preimage->line_allocated);
1795 *preimage = fixed_preimage;
1796
1797 /*
1798 * Adjust the common context lines in postimage. This can be
1799 * done in-place when we are just doing whitespace fixing,
1800 * which does not make the string grow, but needs a new buffer
1801 * when ignoring whitespace causes the update, since in this case
1802 * we could have e.g. tabs converted to multiple spaces.
1803 * We trust the caller to tell us if the update can be done
1804 * in place (postlen==0) or not.
1805 */
1806 old = postimage->buf;
1807 if (postlen)
1808 new = postimage->buf = xmalloc(postlen);
1809 else
1810 new = old;
1811 fixed = preimage->buf;
1812 for (i = ctx = 0; i < postimage->nr; i++) {
1813 size_t len = postimage->line[i].len;
1814 if (!(postimage->line[i].flag & LINE_COMMON)) {
1815 /* an added line -- no counterparts in preimage */
1816 memmove(new, old, len);
1817 old += len;
1818 new += len;
1819 continue;
1820 }
1821
1822 /* a common context -- skip it in the original postimage */
1823 old += len;
1824
1825 /* and find the corresponding one in the fixed preimage */
1826 while (ctx < preimage->nr &&
1827 !(preimage->line[ctx].flag & LINE_COMMON)) {
1828 fixed += preimage->line[ctx].len;
1829 ctx++;
1830 }
1831 if (preimage->nr <= ctx)
1832 die("oops");
1833
1834 /* and copy it in, while fixing the line length */
1835 len = preimage->line[ctx].len;
1836 memcpy(new, fixed, len);
1837 new += len;
1838 fixed += len;
1839 postimage->line[i].len = len;
1840 ctx++;
1841 }
1842
1843 /* Fix the length of the whole thing */
1844 postimage->len = new - postimage->buf;
1845 }
1846
1847 static int match_fragment(struct image *img,
1848 struct image *preimage,
1849 struct image *postimage,
1850 unsigned long try,
1851 int try_lno,
1852 unsigned ws_rule,
1853 int match_beginning, int match_end)
1854 {
1855 int i;
1856 char *fixed_buf, *buf, *orig, *target;
1857 int preimage_limit;
1858
1859 if (preimage->nr + try_lno <= img->nr) {
1860 /*
1861 * The hunk falls within the boundaries of img.
1862 */
1863 preimage_limit = preimage->nr;
1864 if (match_end && (preimage->nr + try_lno != img->nr))
1865 return 0;
1866 } else if (ws_error_action == correct_ws_error &&
1867 (ws_rule & WS_BLANK_AT_EOF)) {
1868 /*
1869 * This hunk extends beyond the end of img, and we are
1870 * removing blank lines at the end of the file. This
1871 * many lines from the beginning of the preimage must
1872 * match with img, and the remainder of the preimage
1873 * must be blank.
1874 */
1875 preimage_limit = img->nr - try_lno;
1876 } else {
1877 /*
1878 * The hunk extends beyond the end of the img and
1879 * we are not removing blanks at the end, so we
1880 * should reject the hunk at this position.
1881 */
1882 return 0;
1883 }
1884
1885 if (match_beginning && try_lno)
1886 return 0;
1887
1888 /* Quick hash check */
1889 for (i = 0; i < preimage_limit; i++)
1890 if (preimage->line[i].hash != img->line[try_lno + i].hash)
1891 return 0;
1892
1893 if (preimage_limit == preimage->nr) {
1894 /*
1895 * Do we have an exact match? If we were told to match
1896 * at the end, size must be exactly at try+fragsize,
1897 * otherwise try+fragsize must be still within the preimage,
1898 * and either case, the old piece should match the preimage
1899 * exactly.
1900 */
1901 if ((match_end
1902 ? (try + preimage->len == img->len)
1903 : (try + preimage->len <= img->len)) &&
1904 !memcmp(img->buf + try, preimage->buf, preimage->len))
1905 return 1;
1906 } else {
1907 /*
1908 * The preimage extends beyond the end of img, so
1909 * there cannot be an exact match.
1910 *
1911 * There must be one non-blank context line that match
1912 * a line before the end of img.
1913 */
1914 char *buf_end;
1915
1916 buf = preimage->buf;
1917 buf_end = buf;
1918 for (i = 0; i < preimage_limit; i++)
1919 buf_end += preimage->line[i].len;
1920
1921 for ( ; buf < buf_end; buf++)
1922 if (!isspace(*buf))
1923 break;
1924 if (buf == buf_end)
1925 return 0;
1926 }
1927
1928 /*
1929 * No exact match. If we are ignoring whitespace, run a line-by-line
1930 * fuzzy matching. We collect all the line length information because
1931 * we need it to adjust whitespace if we match.
1932 */
1933 if (ws_ignore_action == ignore_ws_change) {
1934 size_t imgoff = 0;
1935 size_t preoff = 0;
1936 size_t postlen = postimage->len;
1937 size_t extra_chars;
1938 char *preimage_eof;
1939 char *preimage_end;
1940 for (i = 0; i < preimage_limit; i++) {
1941 size_t prelen = preimage->line[i].len;
1942 size_t imglen = img->line[try_lno+i].len;
1943
1944 if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,
1945 preimage->buf + preoff, prelen))
1946 return 0;
1947 if (preimage->line[i].flag & LINE_COMMON)
1948 postlen += imglen - prelen;
1949 imgoff += imglen;
1950 preoff += prelen;
1951 }
1952
1953 /*
1954 * Ok, the preimage matches with whitespace fuzz.
1955 *
1956 * imgoff now holds the true length of the target that
1957 * matches the preimage before the end of the file.
1958 *
1959 * Count the number of characters in the preimage that fall
1960 * beyond the end of the file and make sure that all of them
1961 * are whitespace characters. (This can only happen if
1962 * we are removing blank lines at the end of the file.)
1963 */
1964 buf = preimage_eof = preimage->buf + preoff;
1965 for ( ; i < preimage->nr; i++)
1966 preoff += preimage->line[i].len;
1967 preimage_end = preimage->buf + preoff;
1968 for ( ; buf < preimage_end; buf++)
1969 if (!isspace(*buf))
1970 return 0;
1971
1972 /*
1973 * Update the preimage and the common postimage context
1974 * lines to use the same whitespace as the target.
1975 * If whitespace is missing in the target (i.e.
1976 * if the preimage extends beyond the end of the file),
1977 * use the whitespace from the preimage.
1978 */
1979 extra_chars = preimage_end - preimage_eof;
1980 fixed_buf = xmalloc(imgoff + extra_chars);
1981 memcpy(fixed_buf, img->buf + try, imgoff);
1982 memcpy(fixed_buf + imgoff, preimage_eof, extra_chars);
1983 imgoff += extra_chars;
1984 update_pre_post_images(preimage, postimage,
1985 fixed_buf, imgoff, postlen);
1986 return 1;
1987 }
1988
1989 if (ws_error_action != correct_ws_error)
1990 return 0;
1991
1992 /*
1993 * The hunk does not apply byte-by-byte, but the hash says
1994 * it might with whitespace fuzz. We haven't been asked to
1995 * ignore whitespace, we were asked to correct whitespace
1996 * errors, so let's try matching after whitespace correction.
1997 *
1998 * The preimage may extend beyond the end of the file,
1999 * but in this loop we will only handle the part of the
2000 * preimage that falls within the file.
2001 */
2002 fixed_buf = xmalloc(preimage->len + 1);
2003 buf = fixed_buf;
2004 orig = preimage->buf;
2005 target = img->buf + try;
2006 for (i = 0; i < preimage_limit; i++) {
2007 size_t fixlen; /* length after fixing the preimage */
2008 size_t oldlen = preimage->line[i].len;
2009 size_t tgtlen = img->line[try_lno + i].len;
2010 size_t tgtfixlen; /* length after fixing the target line */
2011 char tgtfixbuf[1024], *tgtfix;
2012 int match;
2013
2014 /* Try fixing the line in the preimage */
2015 fixlen = ws_fix_copy(buf, orig, oldlen, ws_rule, NULL);
2016
2017 /* Try fixing the line in the target */
2018 if (sizeof(tgtfixbuf) > tgtlen)
2019 tgtfix = tgtfixbuf;
2020 else
2021 tgtfix = xmalloc(tgtlen);
2022 tgtfixlen = ws_fix_copy(tgtfix, target, tgtlen, ws_rule, NULL);
2023
2024 /*
2025 * If they match, either the preimage was based on
2026 * a version before our tree fixed whitespace breakage,
2027 * or we are lacking a whitespace-fix patch the tree
2028 * the preimage was based on already had (i.e. target
2029 * has whitespace breakage, the preimage doesn't).
2030 * In either case, we are fixing the whitespace breakages
2031 * so we might as well take the fix together with their
2032 * real change.
2033 */
2034 match = (tgtfixlen == fixlen && !memcmp(tgtfix, buf, fixlen));
2035
2036 if (tgtfix != tgtfixbuf)
2037 free(tgtfix);
2038 if (!match)
2039 goto unmatch_exit;
2040
2041 orig += oldlen;
2042 buf += fixlen;
2043 target += tgtlen;
2044 }
2045
2046
2047 /*
2048 * Now handle the lines in the preimage that falls beyond the
2049 * end of the file (if any). They will only match if they are
2050 * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2051 * false).
2052 */
2053 for ( ; i < preimage->nr; i++) {
2054 size_t fixlen; /* length after fixing the preimage */
2055 size_t oldlen = preimage->line[i].len;
2056 int j;
2057
2058 /* Try fixing the line in the preimage */
2059 fixlen = ws_fix_copy(buf, orig, oldlen, ws_rule, NULL);
2060
2061 for (j = 0; j < fixlen; j++)
2062 if (!isspace(buf[j]))
2063 goto unmatch_exit;
2064
2065 orig += oldlen;
2066 buf += fixlen;
2067 }
2068
2069 /*
2070 * Yes, the preimage is based on an older version that still
2071 * has whitespace breakages unfixed, and fixing them makes the
2072 * hunk match. Update the context lines in the postimage.
2073 */
2074 update_pre_post_images(preimage, postimage,
2075 fixed_buf, buf - fixed_buf, 0);
2076 return 1;
2077
2078 unmatch_exit:
2079 free(fixed_buf);
2080 return 0;
2081 }
2082
2083 static int find_pos(struct image *img,
2084 struct image *preimage,
2085 struct image *postimage,
2086 int line,
2087 unsigned ws_rule,
2088 int match_beginning, int match_end)
2089 {
2090 int i;
2091 unsigned long backwards, forwards, try;
2092 int backwards_lno, forwards_lno, try_lno;
2093
2094 /*
2095 * If match_beginning or match_end is specified, there is no
2096 * point starting from a wrong line that will never match and
2097 * wander around and wait for a match at the specified end.
2098 */
2099 if (match_beginning)
2100 line = 0;
2101 else if (match_end)
2102 line = img->nr - preimage->nr;
2103
2104 /*
2105 * Because the comparison is unsigned, the following test
2106 * will also take care of a negative line number that can
2107 * result when match_end and preimage is larger than the target.
2108 */
2109 if ((size_t) line > img->nr)
2110 line = img->nr;
2111
2112 try = 0;
2113 for (i = 0; i < line; i++)
2114 try += img->line[i].len;
2115
2116 /*
2117 * There's probably some smart way to do this, but I'll leave
2118 * that to the smart and beautiful people. I'm simple and stupid.
2119 */
2120 backwards = try;
2121 backwards_lno = line;
2122 forwards = try;
2123 forwards_lno = line;
2124 try_lno = line;
2125
2126 for (i = 0; ; i++) {
2127 if (match_fragment(img, preimage, postimage,
2128 try, try_lno, ws_rule,
2129 match_beginning, match_end))
2130 return try_lno;
2131
2132 again:
2133 if (backwards_lno == 0 && forwards_lno == img->nr)
2134 break;
2135
2136 if (i & 1) {
2137 if (backwards_lno == 0) {
2138 i++;
2139 goto again;
2140 }
2141 backwards_lno--;
2142 backwards -= img->line[backwards_lno].len;
2143 try = backwards;
2144 try_lno = backwards_lno;
2145 } else {
2146 if (forwards_lno == img->nr) {
2147 i++;
2148 goto again;
2149 }
2150 forwards += img->line[forwards_lno].len;
2151 forwards_lno++;
2152 try = forwards;
2153 try_lno = forwards_lno;
2154 }
2155
2156 }
2157 return -1;
2158 }
2159
2160 static void remove_first_line(struct image *img)
2161 {
2162 img->buf += img->line[0].len;
2163 img->len -= img->line[0].len;
2164 img->line++;
2165 img->nr--;
2166 }
2167
2168 static void remove_last_line(struct image *img)
2169 {
2170 img->len -= img->line[--img->nr].len;
2171 }
2172
2173 static void update_image(struct image *img,
2174 int applied_pos,
2175 struct image *preimage,
2176 struct image *postimage)
2177 {
2178 /*
2179 * remove the copy of preimage at offset in img
2180 * and replace it with postimage
2181 */
2182 int i, nr;
2183 size_t remove_count, insert_count, applied_at = 0;
2184 char *result;
2185 int preimage_limit;
2186
2187 /*
2188 * If we are removing blank lines at the end of img,
2189 * the preimage may extend beyond the end.
2190 * If that is the case, we must be careful only to
2191 * remove the part of the preimage that falls within
2192 * the boundaries of img. Initialize preimage_limit
2193 * to the number of lines in the preimage that falls
2194 * within the boundaries.
2195 */
2196 preimage_limit = preimage->nr;
2197 if (preimage_limit > img->nr - applied_pos)
2198 preimage_limit = img->nr - applied_pos;
2199
2200 for (i = 0; i < applied_pos; i++)
2201 applied_at += img->line[i].len;
2202
2203 remove_count = 0;
2204 for (i = 0; i < preimage_limit; i++)
2205 remove_count += img->line[applied_pos + i].len;
2206 insert_count = postimage->len;
2207
2208 /* Adjust the contents */
2209 result = xmalloc(img->len + insert_count - remove_count + 1);
2210 memcpy(result, img->buf, applied_at);
2211 memcpy(result + applied_at, postimage->buf, postimage->len);
2212 memcpy(result + applied_at + postimage->len,
2213 img->buf + (applied_at + remove_count),
2214 img->len - (applied_at + remove_count));
2215 free(img->buf);
2216 img->buf = result;
2217 img->len += insert_count - remove_count;
2218 result[img->len] = '\0';
2219
2220 /* Adjust the line table */
2221 nr = img->nr + postimage->nr - preimage_limit;
2222 if (preimage_limit < postimage->nr) {
2223 /*
2224 * NOTE: this knows that we never call remove_first_line()
2225 * on anything other than pre/post image.
2226 */
2227 img->line = xrealloc(img->line, nr * sizeof(*img->line));
2228 img->line_allocated = img->line;
2229 }
2230 if (preimage_limit != postimage->nr)
2231 memmove(img->line + applied_pos + postimage->nr,
2232 img->line + applied_pos + preimage_limit,
2233 (img->nr - (applied_pos + preimage_limit)) *
2234 sizeof(*img->line));
2235 memcpy(img->line + applied_pos,
2236 postimage->line,
2237 postimage->nr * sizeof(*img->line));
2238 img->nr = nr;
2239 }
2240
2241 static int apply_one_fragment(struct image *img, struct fragment *frag,
2242 int inaccurate_eof, unsigned ws_rule)
2243 {
2244 int match_beginning, match_end;
2245 const char *patch = frag->patch;
2246 int size = frag->size;
2247 char *old, *new, *oldlines, *newlines;
2248 int new_blank_lines_at_end = 0;
2249 unsigned long leading, trailing;
2250 int pos, applied_pos;
2251 struct image preimage;
2252 struct image postimage;
2253
2254 memset(&preimage, 0, sizeof(preimage));
2255 memset(&postimage, 0, sizeof(postimage));
2256 oldlines = xmalloc(size);
2257 newlines = xmalloc(size);
2258
2259 old = oldlines;
2260 new = newlines;
2261 while (size > 0) {
2262 char first;
2263 int len = linelen(patch, size);
2264 int plen, added;
2265 int added_blank_line = 0;
2266 int is_blank_context = 0;
2267
2268 if (!len)
2269 break;
2270
2271 /*
2272 * "plen" is how much of the line we should use for
2273 * the actual patch data. Normally we just remove the
2274 * first character on the line, but if the line is
2275 * followed by "\ No newline", then we also remove the
2276 * last one (which is the newline, of course).
2277 */
2278 plen = len - 1;
2279 if (len < size && patch[len] == '\\')
2280 plen--;
2281 first = *patch;
2282 if (apply_in_reverse) {
2283 if (first == '-')
2284 first = '+';
2285 else if (first == '+')
2286 first = '-';
2287 }
2288
2289 switch (first) {
2290 case '\n':
2291 /* Newer GNU diff, empty context line */
2292 if (plen < 0)
2293 /* ... followed by '\No newline'; nothing */
2294 break;
2295 *old++ = '\n';
2296 *new++ = '\n';
2297 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2298 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2299 is_blank_context = 1;
2300 break;
2301 case ' ':
2302 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2303 ws_blank_line(patch + 1, plen, ws_rule))
2304 is_blank_context = 1;
2305 case '-':
2306 memcpy(old, patch + 1, plen);
2307 add_line_info(&preimage, old, plen,
2308 (first == ' ' ? LINE_COMMON : 0));
2309 old += plen;
2310 if (first == '-')
2311 break;
2312 /* Fall-through for ' ' */
2313 case '+':
2314 /* --no-add does not add new lines */
2315 if (first == '+' && no_add)
2316 break;
2317
2318 if (first != '+' ||
2319 !whitespace_error ||
2320 ws_error_action != correct_ws_error) {
2321 memcpy(new, patch + 1, plen);
2322 added = plen;
2323 }
2324 else {
2325 added = ws_fix_copy(new, patch + 1, plen, ws_rule, &applied_after_fixing_ws);
2326 }
2327 add_line_info(&postimage, new, added,
2328 (first == '+' ? 0 : LINE_COMMON));
2329 new += added;
2330 if (first == '+' &&
2331 (ws_rule & WS_BLANK_AT_EOF) &&
2332 ws_blank_line(patch + 1, plen, ws_rule))
2333 added_blank_line = 1;
2334 break;
2335 case '@': case '\\':
2336 /* Ignore it, we already handled it */
2337 break;
2338 default:
2339 if (apply_verbosely)
2340 error("invalid start of line: '%c'", first);
2341 return -1;
2342 }
2343 if (added_blank_line)
2344 new_blank_lines_at_end++;
2345 else if (is_blank_context)
2346 ;
2347 else
2348 new_blank_lines_at_end = 0;
2349 patch += len;
2350 size -= len;
2351 }
2352 if (inaccurate_eof &&
2353 old > oldlines && old[-1] == '\n' &&
2354 new > newlines && new[-1] == '\n') {
2355 old--;
2356 new--;
2357 }
2358
2359 leading = frag->leading;
2360 trailing = frag->trailing;
2361
2362 /*
2363 * A hunk to change lines at the beginning would begin with
2364 * @@ -1,L +N,M @@
2365 * but we need to be careful. -U0 that inserts before the second
2366 * line also has this pattern.
2367 *
2368 * And a hunk to add to an empty file would begin with
2369 * @@ -0,0 +N,M @@
2370 *
2371 * In other words, a hunk that is (frag->oldpos <= 1) with or
2372 * without leading context must match at the beginning.
2373 */
2374 match_beginning = (!frag->oldpos ||
2375 (frag->oldpos == 1 && !unidiff_zero));
2376
2377 /*
2378 * A hunk without trailing lines must match at the end.
2379 * However, we simply cannot tell if a hunk must match end
2380 * from the lack of trailing lines if the patch was generated
2381 * with unidiff without any context.
2382 */
2383 match_end = !unidiff_zero && !trailing;
2384
2385 pos = frag->newpos ? (frag->newpos - 1) : 0;
2386 preimage.buf = oldlines;
2387 preimage.len = old - oldlines;
2388 postimage.buf = newlines;
2389 postimage.len = new - newlines;
2390 preimage.line = preimage.line_allocated;
2391 postimage.line = postimage.line_allocated;
2392
2393 for (;;) {
2394
2395 applied_pos = find_pos(img, &preimage, &postimage, pos,
2396 ws_rule, match_beginning, match_end);
2397
2398 if (applied_pos >= 0)
2399 break;
2400
2401 /* Am I at my context limits? */
2402 if ((leading <= p_context) && (trailing <= p_context))
2403 break;
2404 if (match_beginning || match_end) {
2405 match_beginning = match_end = 0;
2406 continue;
2407 }
2408
2409 /*
2410 * Reduce the number of context lines; reduce both
2411 * leading and trailing if they are equal otherwise
2412 * just reduce the larger context.
2413 */
2414 if (leading >= trailing) {
2415 remove_first_line(&preimage);
2416 remove_first_line(&postimage);
2417 pos--;
2418 leading--;
2419 }
2420 if (trailing > leading) {
2421 remove_last_line(&preimage);
2422 remove_last_line(&postimage);
2423 trailing--;
2424 }
2425 }
2426
2427 if (applied_pos >= 0) {
2428 if (new_blank_lines_at_end &&
2429 preimage.nr + applied_pos >= img->nr &&
2430 (ws_rule & WS_BLANK_AT_EOF) &&
2431 ws_error_action != nowarn_ws_error) {
2432 record_ws_error(WS_BLANK_AT_EOF, "+", 1, frag->linenr);
2433 if (ws_error_action == correct_ws_error) {
2434 while (new_blank_lines_at_end--)
2435 remove_last_line(&postimage);
2436 }
2437 /*
2438 * We would want to prevent write_out_results()
2439 * from taking place in apply_patch() that follows
2440 * the callchain led us here, which is:
2441 * apply_patch->check_patch_list->check_patch->
2442 * apply_data->apply_fragments->apply_one_fragment
2443 */
2444 if (ws_error_action == die_on_ws_error)
2445 apply = 0;
2446 }
2447
2448 /*
2449 * Warn if it was necessary to reduce the number
2450 * of context lines.
2451 */
2452 if ((leading != frag->leading) ||
2453 (trailing != frag->trailing))
2454 fprintf(stderr, "Context reduced to (%ld/%ld)"
2455 " to apply fragment at %d\n",
2456 leading, trailing, applied_pos+1);
2457 update_image(img, applied_pos, &preimage, &postimage);
2458 } else {
2459 if (apply_verbosely)
2460 error("while searching for:\n%.*s",
2461 (int)(old - oldlines), oldlines);
2462 }
2463
2464 free(oldlines);
2465 free(newlines);
2466 free(preimage.line_allocated);
2467 free(postimage.line_allocated);
2468
2469 return (applied_pos < 0);
2470 }
2471
2472 static int apply_binary_fragment(struct image *img, struct patch *patch)
2473 {
2474 struct fragment *fragment = patch->fragments;
2475 unsigned long len;
2476 void *dst;
2477
2478 /* Binary patch is irreversible without the optional second hunk */
2479 if (apply_in_reverse) {
2480 if (!fragment->next)
2481 return error("cannot reverse-apply a binary patch "
2482 "without the reverse hunk to '%s'",
2483 patch->new_name
2484 ? patch->new_name : patch->old_name);
2485 fragment = fragment->next;
2486 }
2487 switch (fragment->binary_patch_method) {
2488 case BINARY_DELTA_DEFLATED:
2489 dst = patch_delta(img->buf, img->len, fragment->patch,
2490 fragment->size, &len);
2491 if (!dst)
2492 return -1;
2493 clear_image(img);
2494 img->buf = dst;
2495 img->len = len;
2496 return 0;
2497 case BINARY_LITERAL_DEFLATED:
2498 clear_image(img);
2499 img->len = fragment->size;
2500 img->buf = xmalloc(img->len+1);
2501 memcpy(img->buf, fragment->patch, img->len);
2502 img->buf[img->len] = '\0';
2503 return 0;
2504 }
2505 return -1;
2506 }
2507
2508 static int apply_binary(struct image *img, struct patch *patch)
2509 {
2510 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2511 unsigned char sha1[20];
2512
2513 /*
2514 * For safety, we require patch index line to contain
2515 * full 40-byte textual SHA1 for old and new, at least for now.
2516 */
2517 if (strlen(patch->old_sha1_prefix) != 40 ||
2518 strlen(patch->new_sha1_prefix) != 40 ||
2519 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
2520 get_sha1_hex(patch->new_sha1_prefix, sha1))
2521 return error("cannot apply binary patch to '%s' "
2522 "without full index line", name);
2523
2524 if (patch->old_name) {
2525 /*
2526 * See if the old one matches what the patch
2527 * applies to.
2528 */
2529 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2530 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
2531 return error("the patch applies to '%s' (%s), "
2532 "which does not match the "
2533 "current contents.",
2534 name, sha1_to_hex(sha1));
2535 }
2536 else {
2537 /* Otherwise, the old one must be empty. */
2538 if (img->len)
2539 return error("the patch applies to an empty "
2540 "'%s' but it is not empty", name);
2541 }
2542
2543 get_sha1_hex(patch->new_sha1_prefix, sha1);
2544 if (is_null_sha1(sha1)) {
2545 clear_image(img);
2546 return 0; /* deletion patch */
2547 }
2548
2549 if (has_sha1_file(sha1)) {
2550 /* We already have the postimage */
2551 enum object_type type;
2552 unsigned long size;
2553 char *result;
2554
2555 result = read_sha1_file(sha1, &type, &size);
2556 if (!result)
2557 return error("the necessary postimage %s for "
2558 "'%s' cannot be read",
2559 patch->new_sha1_prefix, name);
2560 clear_image(img);
2561 img->buf = result;
2562 img->len = size;
2563 } else {
2564 /*
2565 * We have verified buf matches the preimage;
2566 * apply the patch data to it, which is stored
2567 * in the patch->fragments->{patch,size}.
2568 */
2569 if (apply_binary_fragment(img, patch))
2570 return error("binary patch does not apply to '%s'",
2571 name);
2572
2573 /* verify that the result matches */
2574 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2575 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
2576 return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)",
2577 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
2578 }
2579
2580 return 0;
2581 }
2582
2583 static int apply_fragments(struct image *img, struct patch *patch)
2584 {
2585 struct fragment *frag = patch->fragments;
2586 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2587 unsigned ws_rule = patch->ws_rule;
2588 unsigned inaccurate_eof = patch->inaccurate_eof;
2589
2590 if (patch->is_binary)
2591 return apply_binary(img, patch);
2592
2593 while (frag) {
2594 if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule)) {
2595 error("patch failed: %s:%ld", name, frag->oldpos);
2596 if (!apply_with_reject)
2597 return -1;
2598 frag->rejected = 1;
2599 }
2600 frag = frag->next;
2601 }
2602 return 0;
2603 }
2604
2605 static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
2606 {
2607 if (!ce)
2608 return 0;
2609
2610 if (S_ISGITLINK(ce->ce_mode)) {
2611 strbuf_grow(buf, 100);
2612 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(ce->sha1));
2613 } else {
2614 enum object_type type;
2615 unsigned long sz;
2616 char *result;
2617
2618 result = read_sha1_file(ce->sha1, &type, &sz);
2619 if (!result)
2620 return -1;
2621 /* XXX read_sha1_file NUL-terminates */
2622 strbuf_attach(buf, result, sz, sz + 1);
2623 }
2624 return 0;
2625 }
2626
2627 static struct patch *in_fn_table(const char *name)
2628 {
2629 struct string_list_item *item;
2630
2631 if (name == NULL)
2632 return NULL;
2633
2634 item = string_list_lookup(name, &fn_table);
2635 if (item != NULL)
2636 return (struct patch *)item->util;
2637
2638 return NULL;
2639 }
2640
2641 /*
2642 * item->util in the filename table records the status of the path.
2643 * Usually it points at a patch (whose result records the contents
2644 * of it after applying it), but it could be PATH_WAS_DELETED for a
2645 * path that a previously applied patch has already removed.
2646 */
2647 #define PATH_TO_BE_DELETED ((struct patch *) -2)
2648 #define PATH_WAS_DELETED ((struct patch *) -1)
2649
2650 static int to_be_deleted(struct patch *patch)
2651 {
2652 return patch == PATH_TO_BE_DELETED;
2653 }
2654
2655 static int was_deleted(struct patch *patch)
2656 {
2657 return patch == PATH_WAS_DELETED;
2658 }
2659
2660 static void add_to_fn_table(struct patch *patch)
2661 {
2662 struct string_list_item *item;
2663
2664 /*
2665 * Always add new_name unless patch is a deletion
2666 * This should cover the cases for normal diffs,
2667 * file creations and copies
2668 */
2669 if (patch->new_name != NULL) {
2670 item = string_list_insert(patch->new_name, &fn_table);
2671 item->util = patch;
2672 }
2673
2674 /*
2675 * store a failure on rename/deletion cases because
2676 * later chunks shouldn't patch old names
2677 */
2678 if ((patch->new_name == NULL) || (patch->is_rename)) {
2679 item = string_list_insert(patch->old_name, &fn_table);
2680 item->util = PATH_WAS_DELETED;
2681 }
2682 }
2683
2684 static void prepare_fn_table(struct patch *patch)
2685 {
2686 /*
2687 * store information about incoming file deletion
2688 */
2689 while (patch) {
2690 if ((patch->new_name == NULL) || (patch->is_rename)) {
2691 struct string_list_item *item;
2692 item = string_list_insert(patch->old_name, &fn_table);
2693 item->util = PATH_TO_BE_DELETED;
2694 }
2695 patch = patch->next;
2696 }
2697 }
2698
2699 static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
2700 {
2701 struct strbuf buf = STRBUF_INIT;
2702 struct image image;
2703 size_t len;
2704 char *img;
2705 struct patch *tpatch;
2706
2707 if (!(patch->is_copy || patch->is_rename) &&
2708 (tpatch = in_fn_table(patch->old_name)) != NULL && !to_be_deleted(tpatch)) {
2709 if (was_deleted(tpatch)) {
2710 return error("patch %s has been renamed/deleted",
2711 patch->old_name);
2712 }
2713 /* We have a patched copy in memory use that */
2714 strbuf_add(&buf, tpatch->result, tpatch->resultsize);
2715 } else if (cached) {
2716 if (read_file_or_gitlink(ce, &buf))
2717 return error("read of %s failed", patch->old_name);
2718 } else if (patch->old_name) {
2719 if (S_ISGITLINK(patch->old_mode)) {
2720 if (ce) {
2721 read_file_or_gitlink(ce, &buf);
2722 } else {
2723 /*
2724 * There is no way to apply subproject
2725 * patch without looking at the index.
2726 */
2727 patch->fragments = NULL;
2728 }
2729 } else {
2730 if (read_old_data(st, patch->old_name, &buf))
2731 return error("read of %s failed", patch->old_name);
2732 }
2733 }
2734
2735 img = strbuf_detach(&buf, &len);
2736 prepare_image(&image, img, len, !patch->is_binary);
2737
2738 if (apply_fragments(&image, patch) < 0)
2739 return -1; /* note with --reject this succeeds. */
2740 patch->result = image.buf;
2741 patch->resultsize = image.len;
2742 add_to_fn_table(patch);
2743 free(image.line_allocated);
2744
2745 if (0 < patch->is_delete && patch->resultsize)
2746 return error("removal patch leaves file contents");
2747
2748 return 0;
2749 }
2750
2751 static int check_to_create_blob(const char *new_name, int ok_if_exists)
2752 {
2753 struct stat nst;
2754 if (!lstat(new_name, &nst)) {
2755 if (S_ISDIR(nst.st_mode) || ok_if_exists)
2756 return 0;
2757 /*
2758 * A leading component of new_name might be a symlink
2759 * that is going to be removed with this patch, but
2760 * still pointing at somewhere that has the path.
2761 * In such a case, path "new_name" does not exist as
2762 * far as git is concerned.
2763 */
2764 if (has_symlink_leading_path(new_name, strlen(new_name)))
2765 return 0;
2766
2767 return error("%s: already exists in working directory", new_name);
2768 }
2769 else if ((errno != ENOENT) && (errno != ENOTDIR))
2770 return error("%s: %s", new_name, strerror(errno));
2771 return 0;
2772 }
2773
2774 static int verify_index_match(struct cache_entry *ce, struct stat *st)
2775 {
2776 if (S_ISGITLINK(ce->ce_mode)) {
2777 if (!S_ISDIR(st->st_mode))
2778 return -1;
2779 return 0;
2780 }
2781 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
2782 }
2783
2784 static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)
2785 {
2786 const char *old_name = patch->old_name;
2787 struct patch *tpatch = NULL;
2788 int stat_ret = 0;
2789 unsigned st_mode = 0;
2790
2791 /*
2792 * Make sure that we do not have local modifications from the
2793 * index when we are looking at the index. Also make sure
2794 * we have the preimage file to be patched in the work tree,
2795 * unless --cached, which tells git to apply only in the index.
2796 */
2797 if (!old_name)
2798 return 0;
2799
2800 assert(patch->is_new <= 0);
2801
2802 if (!(patch->is_copy || patch->is_rename) &&
2803 (tpatch = in_fn_table(old_name)) != NULL && !to_be_deleted(tpatch)) {
2804 if (was_deleted(tpatch))
2805 return error("%s: has been deleted/renamed", old_name);
2806 st_mode = tpatch->new_mode;
2807 } else if (!cached) {
2808 stat_ret = lstat(old_name, st);
2809 if (stat_ret && errno != ENOENT)
2810 return error("%s: %s", old_name, strerror(errno));
2811 }
2812
2813 if (to_be_deleted(tpatch))
2814 tpatch = NULL;
2815
2816 if (check_index && !tpatch) {
2817 int pos = cache_name_pos(old_name, strlen(old_name));
2818 if (pos < 0) {
2819 if (patch->is_new < 0)
2820 goto is_new;
2821 return error("%s: does not exist in index", old_name);
2822 }
2823 *ce = active_cache[pos];
2824 if (stat_ret < 0) {
2825 struct checkout costate;
2826 /* checkout */
2827 memset(&costate, 0, sizeof(costate));
2828 costate.base_dir = "";
2829 costate.refresh_cache = 1;
2830 if (checkout_entry(*ce, &costate, NULL) ||
2831 lstat(old_name, st))
2832 return -1;
2833 }
2834 if (!cached && verify_index_match(*ce, st))
2835 return error("%s: does not match index", old_name);
2836 if (cached)
2837 st_mode = (*ce)->ce_mode;
2838 } else if (stat_ret < 0) {
2839 if (patch->is_new < 0)
2840 goto is_new;
2841 return error("%s: %s", old_name, strerror(errno));
2842 }
2843
2844 if (!cached && !tpatch)
2845 st_mode = ce_mode_from_stat(*ce, st->st_mode);
2846
2847 if (patch->is_new < 0)
2848 patch->is_new = 0;
2849 if (!patch->old_mode)
2850 patch->old_mode = st_mode;
2851 if ((st_mode ^ patch->old_mode) & S_IFMT)
2852 return error("%s: wrong type", old_name);
2853 if (st_mode != patch->old_mode)
2854 warning("%s has type %o, expected %o",
2855 old_name, st_mode, patch->old_mode);
2856 if (!patch->new_mode && !patch->is_delete)
2857 patch->new_mode = st_mode;
2858 return 0;
2859
2860 is_new:
2861 patch->is_new = 1;
2862 patch->is_delete = 0;
2863 patch->old_name = NULL;
2864 return 0;
2865 }
2866
2867 static int check_patch(struct patch *patch)
2868 {
2869 struct stat st;
2870 const char *old_name = patch->old_name;
2871 const char *new_name = patch->new_name;
2872 const char *name = old_name ? old_name : new_name;
2873 struct cache_entry *ce = NULL;
2874 struct patch *tpatch;
2875 int ok_if_exists;
2876 int status;
2877
2878 patch->rejected = 1; /* we will drop this after we succeed */
2879
2880 status = check_preimage(patch, &ce, &st);
2881 if (status)
2882 return status;
2883 old_name = patch->old_name;
2884
2885 if ((tpatch = in_fn_table(new_name)) &&
2886 (was_deleted(tpatch) || to_be_deleted(tpatch)))
2887 /*
2888 * A type-change diff is always split into a patch to
2889 * delete old, immediately followed by a patch to
2890 * create new (see diff.c::run_diff()); in such a case
2891 * it is Ok that the entry to be deleted by the
2892 * previous patch is still in the working tree and in
2893 * the index.
2894 */
2895 ok_if_exists = 1;
2896 else
2897 ok_if_exists = 0;
2898
2899 if (new_name &&
2900 ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
2901 if (check_index &&
2902 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
2903 !ok_if_exists)
2904 return error("%s: already exists in index", new_name);
2905 if (!cached) {
2906 int err = check_to_create_blob(new_name, ok_if_exists);
2907 if (err)
2908 return err;
2909 }
2910 if (!patch->new_mode) {
2911 if (0 < patch->is_new)
2912 patch->new_mode = S_IFREG | 0644;
2913 else
2914 patch->new_mode = patch->old_mode;
2915 }
2916 }
2917
2918 if (new_name && old_name) {
2919 int same = !strcmp(old_name, new_name);
2920 if (!patch->new_mode)
2921 patch->new_mode = patch->old_mode;
2922 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
2923 return error("new mode (%o) of %s does not match old mode (%o)%s%s",
2924 patch->new_mode, new_name, patch->old_mode,
2925 same ? "" : " of ", same ? "" : old_name);
2926 }
2927
2928 if (apply_data(patch, &st, ce) < 0)
2929 return error("%s: patch does not apply", name);
2930 patch->rejected = 0;
2931 return 0;
2932 }
2933
2934 static int check_patch_list(struct patch *patch)
2935 {
2936 int err = 0;
2937
2938 prepare_fn_table(patch);
2939 while (patch) {
2940 if (apply_verbosely)
2941 say_patch_name(stderr,
2942 "Checking patch ", patch, "...\n");
2943 err |= check_patch(patch);
2944 patch = patch->next;
2945 }
2946 return err;
2947 }
2948
2949 /* This function tries to read the sha1 from the current index */
2950 static int get_current_sha1(const char *path, unsigned char *sha1)
2951 {
2952 int pos;
2953
2954 if (read_cache() < 0)
2955 return -1;
2956 pos = cache_name_pos(path, strlen(path));
2957 if (pos < 0)
2958 return -1;
2959 hashcpy(sha1, active_cache[pos]->sha1);
2960 return 0;
2961 }
2962
2963 /* Build an index that contains the just the files needed for a 3way merge */
2964 static void build_fake_ancestor(struct patch *list, const char *filename)
2965 {
2966 struct patch *patch;
2967 struct index_state result = { NULL };
2968 int fd;
2969
2970 /* Once we start supporting the reverse patch, it may be
2971 * worth showing the new sha1 prefix, but until then...
2972 */
2973 for (patch = list; patch; patch = patch->next) {
2974 const unsigned char *sha1_ptr;
2975 unsigned char sha1[20];
2976 struct cache_entry *ce;
2977 const char *name;
2978
2979 name = patch->old_name ? patch->old_name : patch->new_name;
2980 if (0 < patch->is_new)
2981 continue;
2982 else if (get_sha1(patch->old_sha1_prefix, sha1))
2983 /* git diff has no index line for mode/type changes */
2984 if (!patch->lines_added && !patch->lines_deleted) {
2985 if (get_current_sha1(patch->new_name, sha1) ||
2986 get_current_sha1(patch->old_name, sha1))
2987 die("mode change for %s, which is not "
2988 "in current HEAD", name);
2989 sha1_ptr = sha1;
2990 } else
2991 die("sha1 information is lacking or useless "
2992 "(%s).", name);
2993 else
2994 sha1_ptr = sha1;
2995
2996 ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);
2997 if (!ce)
2998 die("make_cache_entry failed for path '%s'", name);
2999 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
3000 die ("Could not add %s to temporary index", name);
3001 }
3002
3003 fd = open(filename, O_WRONLY | O_CREAT, 0666);
3004 if (fd < 0 || write_index(&result, fd) || close(fd))
3005 die ("Could not write temporary index to %s", filename);
3006
3007 discard_index(&result);
3008 }
3009
3010 static void stat_patch_list(struct patch *patch)
3011 {
3012 int files, adds, dels;
3013
3014 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
3015 files++;
3016 adds += patch->lines_added;
3017 dels += patch->lines_deleted;
3018 show_stats(patch);
3019 }
3020
3021 printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
3022 }
3023
3024 static void numstat_patch_list(struct patch *patch)
3025 {
3026 for ( ; patch; patch = patch->next) {
3027 const char *name;
3028 name = patch->new_name ? patch->new_name : patch->old_name;
3029 if (patch->is_binary)
3030 printf("-\t-\t");
3031 else
3032 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
3033 write_name_quoted(name, stdout, line_termination);
3034 }
3035 }
3036
3037 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
3038 {
3039 if (mode)
3040 printf(" %s mode %06o %s\n", newdelete, mode, name);
3041 else
3042 printf(" %s %s\n", newdelete, name);
3043 }
3044
3045 static void show_mode_change(struct patch *p, int show_name)
3046 {
3047 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
3048 if (show_name)
3049 printf(" mode change %06o => %06o %s\n",
3050 p->old_mode, p->new_mode, p->new_name);
3051 else
3052 printf(" mode change %06o => %06o\n",
3053 p->old_mode, p->new_mode);
3054 }
3055 }
3056
3057 static void show_rename_copy(struct patch *p)
3058 {
3059 const char *renamecopy = p->is_rename ? "rename" : "copy";
3060 const char *old, *new;
3061
3062 /* Find common prefix */
3063 old = p->old_name;
3064 new = p->new_name;
3065 while (1) {
3066 const char *slash_old, *slash_new;
3067 slash_old = strchr(old, '/');
3068 slash_new = strchr(new, '/');
3069 if (!slash_old ||
3070 !slash_new ||
3071 slash_old - old != slash_new - new ||
3072 memcmp(old, new, slash_new - new))
3073 break;
3074 old = slash_old + 1;
3075 new = slash_new + 1;
3076 }
3077 /* p->old_name thru old is the common prefix, and old and new
3078 * through the end of names are renames
3079 */
3080 if (old != p->old_name)
3081 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
3082 (int)(old - p->old_name), p->old_name,
3083 old, new, p->score);
3084 else
3085 printf(" %s %s => %s (%d%%)\n", renamecopy,
3086 p->old_name, p->new_name, p->score);
3087 show_mode_change(p, 0);
3088 }
3089
3090 static void summary_patch_list(struct patch *patch)
3091 {
3092 struct patch *p;
3093
3094 for (p = patch; p; p = p->next) {
3095 if (p->is_new)
3096 show_file_mode_name("create", p->new_mode, p->new_name);
3097 else if (p->is_delete)
3098 show_file_mode_name("delete", p->old_mode, p->old_name);
3099 else {
3100 if (p->is_rename || p->is_copy)
3101 show_rename_copy(p);
3102 else {
3103 if (p->score) {
3104 printf(" rewrite %s (%d%%)\n",
3105 p->new_name, p->score);
3106 show_mode_change(p, 0);
3107 }
3108 else
3109 show_mode_change(p, 1);
3110 }
3111 }
3112 }
3113 }
3114
3115 static void patch_stats(struct patch *patch)
3116 {
3117 int lines = patch->lines_added + patch->lines_deleted;
3118
3119 if (lines > max_change)
3120 max_change = lines;
3121 if (patch->old_name) {
3122 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
3123 if (!len)
3124 len = strlen(patch->old_name);
3125 if (len > max_len)
3126 max_len = len;
3127 }
3128 if (patch->new_name) {
3129 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
3130 if (!len)
3131 len = strlen(patch->new_name);
3132 if (len > max_len)
3133 max_len = len;
3134 }
3135 }
3136
3137 static void remove_file(struct patch *patch, int rmdir_empty)
3138 {
3139 if (update_index) {
3140 if (remove_file_from_cache(patch->old_name) < 0)
3141 die("unable to remove %s from index", patch->old_name);
3142 }
3143 if (!cached) {
3144 if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {
3145 remove_path(patch->old_name);
3146 }
3147 }
3148 }
3149
3150 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
3151 {
3152 struct stat st;
3153 struct cache_entry *ce;
3154 int namelen = strlen(path);
3155 unsigned ce_size = cache_entry_size(namelen);
3156
3157 if (!update_index)
3158 return;
3159
3160 ce = xcalloc(1, ce_size);
3161 memcpy(ce->name, path, namelen);
3162 ce->ce_mode = create_ce_mode(mode);
3163 ce->ce_flags = namelen;
3164 if (S_ISGITLINK(mode)) {
3165 const char *s = buf;
3166
3167 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
3168 die("corrupt patch for subproject %s", path);
3169 } else {
3170 if (!cached) {
3171 if (lstat(path, &st) < 0)
3172 die_errno("unable to stat newly created file '%s'",
3173 path);
3174 fill_stat_cache_info(ce, &st);
3175 }
3176 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
3177 die("unable to create backing store for newly created file %s", path);
3178 }
3179 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
3180 die("unable to add cache entry for %s", path);
3181 }
3182
3183 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
3184 {
3185 int fd;
3186 struct strbuf nbuf = STRBUF_INIT;
3187
3188 if (S_ISGITLINK(mode)) {
3189 struct stat st;
3190 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
3191 return 0;
3192 return mkdir(path, 0777);
3193 }
3194
3195 if (has_symlinks && S_ISLNK(mode))
3196 /* Although buf:size is counted string, it also is NUL
3197 * terminated.
3198 */
3199 return symlink(buf, path);
3200
3201 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
3202 if (fd < 0)
3203 return -1;
3204
3205 if (convert_to_working_tree(path, buf, size, &nbuf)) {
3206 size = nbuf.len;
3207 buf = nbuf.buf;
3208 }
3209 write_or_die(fd, buf, size);
3210 strbuf_release(&nbuf);
3211
3212 if (close(fd) < 0)
3213 die_errno("closing file '%s'", path);
3214 return 0;
3215 }
3216
3217 /*
3218 * We optimistically assume that the directories exist,
3219 * which is true 99% of the time anyway. If they don't,
3220 * we create them and try again.
3221 */
3222 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
3223 {
3224 if (cached)
3225 return;
3226 if (!try_create_file(path, mode, buf, size))
3227 return;
3228
3229 if (errno == ENOENT) {
3230 if (safe_create_leading_directories(path))
3231 return;
3232 if (!try_create_file(path, mode, buf, size))
3233 return;
3234 }
3235
3236 if (errno == EEXIST || errno == EACCES) {
3237 /* We may be trying to create a file where a directory
3238 * used to be.
3239 */
3240 struct stat st;
3241 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
3242 errno = EEXIST;
3243 }
3244
3245 if (errno == EEXIST) {
3246 unsigned int nr = getpid();
3247
3248 for (;;) {
3249 char newpath[PATH_MAX];
3250 mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
3251 if (!try_create_file(newpath, mode, buf, size)) {
3252 if (!rename(newpath, path))
3253 return;
3254 unlink_or_warn(newpath);
3255 break;
3256 }
3257 if (errno != EEXIST)
3258 break;
3259 ++nr;
3260 }
3261 }
3262 die_errno("unable to write file '%s' mode %o", path, mode);
3263 }
3264
3265 static void create_file(struct patch *patch)
3266 {
3267 char *path = patch->new_name;
3268 unsigned mode = patch->new_mode;
3269 unsigned long size = patch->resultsize;
3270 char *buf = patch->result;
3271
3272 if (!mode)
3273 mode = S_IFREG | 0644;
3274 create_one_file(path, mode, buf, size);
3275 add_index_file(path, mode, buf, size);
3276 }
3277
3278 /* phase zero is to remove, phase one is to create */
3279 static void write_out_one_result(struct patch *patch, int phase)
3280 {
3281 if (patch->is_delete > 0) {
3282 if (phase == 0)
3283 remove_file(patch, 1);
3284 return;
3285 }
3286 if (patch->is_new > 0 || patch->is_copy) {
3287 if (phase == 1)
3288 create_file(patch);
3289 return;
3290 }
3291 /*
3292 * Rename or modification boils down to the same
3293 * thing: remove the old, write the new
3294 */
3295 if (phase == 0)
3296 remove_file(patch, patch->is_rename);
3297 if (phase == 1)
3298 create_file(patch);
3299 }
3300
3301 static int write_out_one_reject(struct patch *patch)
3302 {
3303 FILE *rej;
3304 char namebuf[PATH_MAX];
3305 struct fragment *frag;
3306 int cnt = 0;
3307
3308 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
3309 if (!frag->rejected)
3310 continue;
3311 cnt++;
3312 }
3313
3314 if (!cnt) {
3315 if (apply_verbosely)
3316 say_patch_name(stderr,
3317 "Applied patch ", patch, " cleanly.\n");
3318 return 0;
3319 }
3320
3321 /* This should not happen, because a removal patch that leaves
3322 * contents are marked "rejected" at the patch level.
3323 */
3324 if (!patch->new_name)
3325 die("internal error");
3326
3327 /* Say this even without --verbose */
3328 say_patch_name(stderr, "Applying patch ", patch, " with");
3329 fprintf(stderr, " %d rejects...\n", cnt);
3330
3331 cnt = strlen(patch->new_name);
3332 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
3333 cnt = ARRAY_SIZE(namebuf) - 5;
3334 warning("truncating .rej filename to %.*s.rej",
3335 cnt - 1, patch->new_name);
3336 }
3337 memcpy(namebuf, patch->new_name, cnt);
3338 memcpy(namebuf + cnt, ".rej", 5);
3339
3340 rej = fopen(namebuf, "w");
3341 if (!rej)
3342 return error("cannot open %s: %s", namebuf, strerror(errno));
3343
3344 /* Normal git tools never deal with .rej, so do not pretend
3345 * this is a git patch by saying --git nor give extended
3346 * headers. While at it, maybe please "kompare" that wants
3347 * the trailing TAB and some garbage at the end of line ;-).
3348 */
3349 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
3350 patch->new_name, patch->new_name);
3351 for (cnt = 1, frag = patch->fragments;
3352 frag;
3353 cnt++, frag = frag->next) {
3354 if (!frag->rejected) {
3355 fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);
3356 continue;
3357 }
3358 fprintf(stderr, "Rejected hunk #%d.\n", cnt);
3359 fprintf(rej, "%.*s", frag->size, frag->patch);
3360 if (frag->patch[frag->size-1] != '\n')
3361 fputc('\n', rej);
3362 }
3363 fclose(rej);
3364 return -1;
3365 }
3366
3367 static int write_out_results(struct patch *list, int skipped_patch)
3368 {
3369 int phase;
3370 int errs = 0;
3371 struct patch *l;
3372
3373 if (!list && !skipped_patch)
3374 return error("No changes");
3375
3376 for (phase = 0; phase < 2; phase++) {
3377 l = list;
3378 while (l) {
3379 if (l->rejected)
3380 errs = 1;
3381 else {
3382 write_out_one_result(l, phase);
3383 if (phase == 1 && write_out_one_reject(l))
3384 errs = 1;
3385 }
3386 l = l->next;
3387 }
3388 }
3389 return errs;
3390 }
3391
3392 static struct lock_file lock_file;
3393
3394 static struct string_list limit_by_name;
3395 static int has_include;
3396 static void add_name_limit(const char *name, int exclude)
3397 {
3398 struct string_list_item *it;
3399
3400 it = string_list_append(name, &limit_by_name);
3401 it->util = exclude ? NULL : (void *) 1;
3402 }
3403
3404 static int use_patch(struct patch *p)
3405 {
3406 const char *pathname = p->new_name ? p->new_name : p->old_name;
3407 int i;
3408
3409 /* Paths outside are not touched regardless of "--include" */
3410 if (0 < prefix_length) {
3411 int pathlen = strlen(pathname);
3412 if (pathlen <= prefix_length ||
3413 memcmp(prefix, pathname, prefix_length))
3414 return 0;
3415 }
3416
3417 /* See if it matches any of exclude/include rule */
3418 for (i = 0; i < limit_by_name.nr; i++) {
3419 struct string_list_item *it = &limit_by_name.items[i];
3420 if (!fnmatch(it->string, pathname, 0))
3421 return (it->util != NULL);
3422 }
3423
3424 /*
3425 * If we had any include, a path that does not match any rule is
3426 * not used. Otherwise, we saw bunch of exclude rules (or none)
3427 * and such a path is used.
3428 */
3429 return !has_include;
3430 }
3431
3432
3433 static void prefix_one(char **name)
3434 {
3435 char *old_name = *name;
3436 if (!old_name)
3437 return;
3438 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
3439 free(old_name);
3440 }
3441
3442 static void prefix_patches(struct patch *p)
3443 {
3444 if (!prefix || p->is_toplevel_relative)
3445 return;
3446 for ( ; p; p = p->next) {
3447 if (p->new_name == p->old_name) {
3448 char *prefixed = p->new_name;
3449 prefix_one(&prefixed);
3450 p->new_name = p->old_name = prefixed;
3451 }
3452 else {
3453 prefix_one(&p->new_name);
3454 prefix_one(&p->old_name);
3455 }
3456 }
3457 }
3458
3459 #define INACCURATE_EOF (1<<0)
3460 #define RECOUNT (1<<1)
3461
3462 static int apply_patch(int fd, const char *filename, int options)
3463 {
3464 size_t offset;
3465 struct strbuf buf = STRBUF_INIT;
3466 struct patch *list = NULL, **listp = &list;
3467 int skipped_patch = 0;
3468
3469 /* FIXME - memory leak when using multiple patch files as inputs */
3470 memset(&fn_table, 0, sizeof(struct string_list));
3471 patch_input_file = filename;
3472 read_patch_file(&buf, fd);
3473 offset = 0;
3474 while (offset < buf.len) {
3475 struct patch *patch;
3476 int nr;
3477
3478 patch = xcalloc(1, sizeof(*patch));
3479 patch->inaccurate_eof = !!(options & INACCURATE_EOF);
3480 patch->recount = !!(options & RECOUNT);
3481 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
3482 if (nr < 0)
3483 break;
3484 if (apply_in_reverse)
3485 reverse_patches(patch);
3486 if (prefix)
3487 prefix_patches(patch);
3488 if (use_patch(patch)) {
3489 patch_stats(patch);
3490 *listp = patch;
3491 listp = &patch->next;
3492 }
3493 else {
3494 /* perhaps free it a bit better? */
3495 free(patch);
3496 skipped_patch++;
3497 }
3498 offset += nr;
3499 }
3500
3501 if (whitespace_error && (ws_error_action == die_on_ws_error))
3502 apply = 0;
3503
3504 update_index = check_index && apply;
3505 if (update_index && newfd < 0)
3506 newfd = hold_locked_index(&lock_file, 1);
3507
3508 if (check_index) {
3509 if (read_cache() < 0)
3510 die("unable to read index file");
3511 }
3512
3513 if ((check || apply) &&
3514 check_patch_list(list) < 0 &&
3515 !apply_with_reject)
3516 exit(1);
3517
3518 if (apply && write_out_results(list, skipped_patch))
3519 exit(1);
3520
3521 if (fake_ancestor)
3522 build_fake_ancestor(list, fake_ancestor);
3523
3524 if (diffstat)
3525 stat_patch_list(list);
3526
3527 if (numstat)
3528 numstat_patch_list(list);
3529
3530 if (summary)
3531 summary_patch_list(list);
3532
3533 strbuf_release(&buf);
3534 return 0;
3535 }
3536
3537 static int git_apply_config(const char *var, const char *value, void *cb)
3538 {
3539 if (!strcmp(var, "apply.whitespace"))
3540 return git_config_string(&apply_default_whitespace, var, value);
3541 else if (!strcmp(var, "apply.ignorewhitespace"))
3542 return git_config_string(&apply_default_ignorewhitespace, var, value);
3543 return git_default_config(var, value, cb);
3544 }
3545
3546 static int option_parse_exclude(const struct option *opt,
3547 const char *arg, int unset)
3548 {
3549 add_name_limit(arg, 1);
3550 return 0;
3551 }
3552
3553 static int option_parse_include(const struct option *opt,
3554 const char *arg, int unset)
3555 {
3556 add_name_limit(arg, 0);
3557 has_include = 1;
3558 return 0;
3559 }
3560
3561 static int option_parse_p(const struct option *opt,
3562 const char *arg, int unset)
3563 {
3564 p_value = atoi(arg);
3565 p_value_known = 1;
3566 return 0;
3567 }
3568
3569 static int option_parse_z(const struct option *opt,
3570 const char *arg, int unset)
3571 {
3572 if (unset)
3573 line_termination = '\n';
3574 else
3575 line_termination = 0;
3576 return 0;
3577 }
3578
3579 static int option_parse_space_change(const struct option *opt,
3580 const char *arg, int unset)
3581 {
3582 if (unset)
3583 ws_ignore_action = ignore_ws_none;
3584 else
3585 ws_ignore_action = ignore_ws_change;
3586 return 0;
3587 }
3588
3589 static int option_parse_whitespace(const struct option *opt,
3590 const char *arg, int unset)
3591 {
3592 const char **whitespace_option = opt->value;
3593
3594 *whitespace_option = arg;
3595 parse_whitespace_option(arg);
3596 return 0;
3597 }
3598
3599 static int option_parse_directory(const struct option *opt,
3600 const char *arg, int unset)
3601 {
3602 root_len = strlen(arg);
3603 if (root_len && arg[root_len - 1] != '/') {
3604 char *new_root;
3605 root = new_root = xmalloc(root_len + 2);
3606 strcpy(new_root, arg);
3607 strcpy(new_root + root_len++, "/");
3608 } else
3609 root = arg;
3610 return 0;
3611 }
3612
3613 int cmd_apply(int argc, const char **argv, const char *unused_prefix)
3614 {
3615 int i;
3616 int errs = 0;
3617 int is_not_gitdir;
3618 int binary;
3619 int force_apply = 0;
3620
3621 const char *whitespace_option = NULL;
3622
3623 struct option builtin_apply_options[] = {
3624 { OPTION_CALLBACK, 0, "exclude", NULL, "path",
3625 "don't apply changes matching the given path",
3626 0, option_parse_exclude },
3627 { OPTION_CALLBACK, 0, "include", NULL, "path",
3628 "apply changes matching the given path",
3629 0, option_parse_include },
3630 { OPTION_CALLBACK, 'p', NULL, NULL, "num",
3631 "remove <num> leading slashes from traditional diff paths",
3632 0, option_parse_p },
3633 OPT_BOOLEAN(0, "no-add", &no_add,
3634 "ignore additions made by the patch"),
3635 OPT_BOOLEAN(0, "stat", &diffstat,
3636 "instead of applying the patch, output diffstat for the input"),
3637 { OPTION_BOOLEAN, 0, "allow-binary-replacement", &binary,
3638 NULL, "old option, now no-op",
3639 PARSE_OPT_HIDDEN | PARSE_OPT_NOARG },
3640 { OPTION_BOOLEAN, 0, "binary", &binary,
3641 NULL, "old option, now no-op",
3642 PARSE_OPT_HIDDEN | PARSE_OPT_NOARG },
3643 OPT_BOOLEAN(0, "numstat", &numstat,
3644 "shows number of added and deleted lines in decimal notation"),
3645 OPT_BOOLEAN(0, "summary", &summary,
3646 "instead of applying the patch, output a summary for the input"),
3647 OPT_BOOLEAN(0, "check", &check,
3648 "instead of applying the patch, see if the patch is applicable"),
3649 OPT_BOOLEAN(0, "index", &check_index,
3650 "make sure the patch is applicable to the current index"),
3651 OPT_BOOLEAN(0, "cached", &cached,
3652 "apply a patch without touching the working tree"),
3653 OPT_BOOLEAN(0, "apply", &force_apply,
3654 "also apply the patch (use with --stat/--summary/--check)"),
3655 OPT_FILENAME(0, "build-fake-ancestor", &fake_ancestor,
3656 "build a temporary index based on embedded index information"),
3657 { OPTION_CALLBACK, 'z', NULL, NULL, NULL,
3658 "paths are separated with NUL character",
3659 PARSE_OPT_NOARG, option_parse_z },
3660 OPT_INTEGER('C', NULL, &p_context,
3661 "ensure at least <n> lines of context match"),
3662 { OPTION_CALLBACK, 0, "whitespace", &whitespace_option, "action",
3663 "detect new or modified lines that have whitespace errors",
3664 0, option_parse_whitespace },
3665 { OPTION_CALLBACK, 0, "ignore-space-change", NULL, NULL,
3666 "ignore changes in whitespace when finding context",
3667 PARSE_OPT_NOARG, option_parse_space_change },
3668 { OPTION_CALLBACK, 0, "ignore-whitespace", NULL, NULL,
3669 "ignore changes in whitespace when finding context",
3670 PARSE_OPT_NOARG, option_parse_space_change },
3671 OPT_BOOLEAN('R', "reverse", &apply_in_reverse,
3672 "apply the patch in reverse"),
3673 OPT_BOOLEAN(0, "unidiff-zero", &unidiff_zero,
3674 "don't expect at least one line of context"),
3675 OPT_BOOLEAN(0, "reject", &apply_with_reject,
3676 "leave the rejected hunks in corresponding *.rej files"),
3677 OPT__VERBOSE(&apply_verbosely),
3678 OPT_BIT(0, "inaccurate-eof", &options,
3679 "tolerate incorrectly detected missing new-line at the end of file",
3680 INACCURATE_EOF),
3681 OPT_BIT(0, "recount", &options,
3682 "do not trust the line counts in the hunk headers",
3683 RECOUNT),
3684 { OPTION_CALLBACK, 0, "directory", NULL, "root",
3685 "prepend <root> to all filenames",
3686 0, option_parse_directory },
3687 OPT_END()
3688 };
3689
3690 prefix = setup_git_directory_gently(&is_not_gitdir);
3691 prefix_length = prefix ? strlen(prefix) : 0;
3692 git_config(git_apply_config, NULL);
3693 if (apply_default_whitespace)
3694 parse_whitespace_option(apply_default_whitespace);
3695 if (apply_default_ignorewhitespace)
3696 parse_ignorewhitespace_option(apply_default_ignorewhitespace);
3697
3698 argc = parse_options(argc, argv, prefix, builtin_apply_options,
3699 apply_usage, 0);
3700
3701 if (apply_with_reject)
3702 apply = apply_verbosely = 1;
3703 if (!force_apply && (diffstat || numstat || summary || check || fake_ancestor))
3704 apply = 0;
3705 if (check_index && is_not_gitdir)
3706 die("--index outside a repository");
3707 if (cached) {
3708 if (is_not_gitdir)
3709 die("--cached outside a repository");
3710 check_index = 1;
3711 }
3712 for (i = 0; i < argc; i++) {
3713 const char *arg = argv[i];
3714 int fd;
3715
3716 if (!strcmp(arg, "-")) {
3717 errs |= apply_patch(0, "<stdin>", options);
3718 read_stdin = 0;
3719 continue;
3720 } else if (0 < prefix_length)
3721 arg = prefix_filename(prefix, prefix_length, arg);
3722
3723 fd = open(arg, O_RDONLY);
3724 if (fd < 0)
3725 die_errno("can't open patch '%s'", arg);
3726 read_stdin = 0;
3727 set_default_whitespace_mode(whitespace_option);
3728 errs |= apply_patch(fd, arg, options);
3729 close(fd);
3730 }
3731 set_default_whitespace_mode(whitespace_option);
3732 if (read_stdin)
3733 errs |= apply_patch(0, "<stdin>", options);
3734 if (whitespace_error) {
3735 if (squelch_whitespace_errors &&
3736 squelch_whitespace_errors < whitespace_error) {
3737 int squelched =
3738 whitespace_error - squelch_whitespace_errors;
3739 warning("squelched %d "
3740 "whitespace error%s",
3741 squelched,