Merge branch 'maint' to synchronize with 1.5.3.6
[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
16 /*
17 * --check turns on checking that the working tree matches the
18 * files that are being modified, but doesn't apply the patch
19 * --stat does just a diffstat, and doesn't actually apply
20 * --numstat does numeric diffstat, and doesn't actually apply
21 * --index-info shows the old and new index info for paths if available.
22 * --index updates the cache as well.
23 * --cached updates only the cache without ever touching the working tree.
24 */
25 static const char *prefix;
26 static int prefix_length = -1;
27 static int newfd = -1;
28
29 static int unidiff_zero;
30 static int p_value = 1;
31 static int p_value_known;
32 static int check_index;
33 static int update_index;
34 static int cached;
35 static int diffstat;
36 static int numstat;
37 static int summary;
38 static int check;
39 static int apply = 1;
40 static int apply_in_reverse;
41 static int apply_with_reject;
42 static int apply_verbosely;
43 static int no_add;
44 static const char *fake_ancestor;
45 static int line_termination = '\n';
46 static unsigned long p_context = ULONG_MAX;
47 static const char apply_usage[] =
48 "git-apply [--stat] [--numstat] [--summary] [--check] [--index] [--cached] [--apply] [--no-add] [--index-info] [--allow-binary-replacement] [--reverse] [--reject] [--verbose] [-z] [-pNUM] [-CNUM] [--whitespace=<nowarn|warn|error|error-all|strip>] <patch>...";
49
50 static enum whitespace_eol {
51 nowarn_whitespace,
52 warn_on_whitespace,
53 error_on_whitespace,
54 strip_whitespace,
55 } new_whitespace = warn_on_whitespace;
56 static int whitespace_error;
57 static int squelch_whitespace_errors = 5;
58 static int applied_after_fixing_ws;
59 static const char *patch_input_file;
60
61 static void parse_whitespace_option(const char *option)
62 {
63 if (!option) {
64 new_whitespace = warn_on_whitespace;
65 return;
66 }
67 if (!strcmp(option, "warn")) {
68 new_whitespace = warn_on_whitespace;
69 return;
70 }
71 if (!strcmp(option, "nowarn")) {
72 new_whitespace = nowarn_whitespace;
73 return;
74 }
75 if (!strcmp(option, "error")) {
76 new_whitespace = error_on_whitespace;
77 return;
78 }
79 if (!strcmp(option, "error-all")) {
80 new_whitespace = error_on_whitespace;
81 squelch_whitespace_errors = 0;
82 return;
83 }
84 if (!strcmp(option, "strip")) {
85 new_whitespace = strip_whitespace;
86 return;
87 }
88 die("unrecognized whitespace option '%s'", option);
89 }
90
91 static void set_default_whitespace_mode(const char *whitespace_option)
92 {
93 if (!whitespace_option && !apply_default_whitespace) {
94 new_whitespace = (apply
95 ? warn_on_whitespace
96 : nowarn_whitespace);
97 }
98 }
99
100 /*
101 * For "diff-stat" like behaviour, we keep track of the biggest change
102 * we've seen, and the longest filename. That allows us to do simple
103 * scaling.
104 */
105 static int max_change, max_len;
106
107 /*
108 * Various "current state", notably line numbers and what
109 * file (and how) we're patching right now.. The "is_xxxx"
110 * things are flags, where -1 means "don't know yet".
111 */
112 static int linenr = 1;
113
114 /*
115 * This represents one "hunk" from a patch, starting with
116 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
117 * patch text is pointed at by patch, and its byte length
118 * is stored in size. leading and trailing are the number
119 * of context lines.
120 */
121 struct fragment {
122 unsigned long leading, trailing;
123 unsigned long oldpos, oldlines;
124 unsigned long newpos, newlines;
125 const char *patch;
126 int size;
127 int rejected;
128 struct fragment *next;
129 };
130
131 /*
132 * When dealing with a binary patch, we reuse "leading" field
133 * to store the type of the binary hunk, either deflated "delta"
134 * or deflated "literal".
135 */
136 #define binary_patch_method leading
137 #define BINARY_DELTA_DEFLATED 1
138 #define BINARY_LITERAL_DEFLATED 2
139
140 struct patch {
141 char *new_name, *old_name, *def_name;
142 unsigned int old_mode, new_mode;
143 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */
144 int rejected;
145 unsigned long deflate_origlen;
146 int lines_added, lines_deleted;
147 int score;
148 unsigned int is_toplevel_relative:1;
149 unsigned int inaccurate_eof:1;
150 unsigned int is_binary:1;
151 unsigned int is_copy:1;
152 unsigned int is_rename:1;
153 struct fragment *fragments;
154 char *result;
155 size_t resultsize;
156 char old_sha1_prefix[41];
157 char new_sha1_prefix[41];
158 struct patch *next;
159 };
160
161 static void say_patch_name(FILE *output, const char *pre, struct patch *patch, const char *post)
162 {
163 fputs(pre, output);
164 if (patch->old_name && patch->new_name &&
165 strcmp(patch->old_name, patch->new_name)) {
166 quote_c_style(patch->old_name, NULL, output, 0);
167 fputs(" => ", output);
168 quote_c_style(patch->new_name, NULL, output, 0);
169 } else {
170 const char *n = patch->new_name;
171 if (!n)
172 n = patch->old_name;
173 quote_c_style(n, NULL, output, 0);
174 }
175 fputs(post, output);
176 }
177
178 #define CHUNKSIZE (8192)
179 #define SLOP (16)
180
181 static void read_patch_file(struct strbuf *sb, int fd)
182 {
183 if (strbuf_read(sb, fd, 0) < 0)
184 die("git-apply: read returned %s", strerror(errno));
185
186 /*
187 * Make sure that we have some slop in the buffer
188 * so that we can do speculative "memcmp" etc, and
189 * see to it that it is NUL-filled.
190 */
191 strbuf_grow(sb, SLOP);
192 memset(sb->buf + sb->len, 0, SLOP);
193 }
194
195 static unsigned long linelen(const char *buffer, unsigned long size)
196 {
197 unsigned long len = 0;
198 while (size--) {
199 len++;
200 if (*buffer++ == '\n')
201 break;
202 }
203 return len;
204 }
205
206 static int is_dev_null(const char *str)
207 {
208 return !memcmp("/dev/null", str, 9) && isspace(str[9]);
209 }
210
211 #define TERM_SPACE 1
212 #define TERM_TAB 2
213
214 static int name_terminate(const char *name, int namelen, int c, int terminate)
215 {
216 if (c == ' ' && !(terminate & TERM_SPACE))
217 return 0;
218 if (c == '\t' && !(terminate & TERM_TAB))
219 return 0;
220
221 return 1;
222 }
223
224 static char *find_name(const char *line, char *def, int p_value, int terminate)
225 {
226 int len;
227 const char *start = line;
228
229 if (*line == '"') {
230 struct strbuf name;
231
232 /* Proposed "new-style" GNU patch/diff format; see
233 * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
234 */
235 strbuf_init(&name, 0);
236 if (!unquote_c_style(&name, line, NULL)) {
237 char *cp;
238
239 for (cp = name.buf; p_value; p_value--) {
240 cp = strchr(cp, '/');
241 if (!cp)
242 break;
243 cp++;
244 }
245 if (cp) {
246 /* name can later be freed, so we need
247 * to memmove, not just return cp
248 */
249 strbuf_remove(&name, 0, cp - name.buf);
250 free(def);
251 return strbuf_detach(&name, NULL);
252 }
253 }
254 strbuf_release(&name);
255 }
256
257 for (;;) {
258 char c = *line;
259
260 if (isspace(c)) {
261 if (c == '\n')
262 break;
263 if (name_terminate(start, line-start, c, terminate))
264 break;
265 }
266 line++;
267 if (c == '/' && !--p_value)
268 start = line;
269 }
270 if (!start)
271 return def;
272 len = line - start;
273 if (!len)
274 return def;
275
276 /*
277 * Generally we prefer the shorter name, especially
278 * if the other one is just a variation of that with
279 * something else tacked on to the end (ie "file.orig"
280 * or "file~").
281 */
282 if (def) {
283 int deflen = strlen(def);
284 if (deflen < len && !strncmp(start, def, deflen))
285 return def;
286 free(def);
287 }
288
289 return xmemdupz(start, len);
290 }
291
292 static int count_slashes(const char *cp)
293 {
294 int cnt = 0;
295 char ch;
296
297 while ((ch = *cp++))
298 if (ch == '/')
299 cnt++;
300 return cnt;
301 }
302
303 /*
304 * Given the string after "--- " or "+++ ", guess the appropriate
305 * p_value for the given patch.
306 */
307 static int guess_p_value(const char *nameline)
308 {
309 char *name, *cp;
310 int val = -1;
311
312 if (is_dev_null(nameline))
313 return -1;
314 name = find_name(nameline, NULL, 0, TERM_SPACE | TERM_TAB);
315 if (!name)
316 return -1;
317 cp = strchr(name, '/');
318 if (!cp)
319 val = 0;
320 else if (prefix) {
321 /*
322 * Does it begin with "a/$our-prefix" and such? Then this is
323 * very likely to apply to our directory.
324 */
325 if (!strncmp(name, prefix, prefix_length))
326 val = count_slashes(prefix);
327 else {
328 cp++;
329 if (!strncmp(cp, prefix, prefix_length))
330 val = count_slashes(prefix) + 1;
331 }
332 }
333 free(name);
334 return val;
335 }
336
337 /*
338 * Get the name etc info from the --/+++ lines of a traditional patch header
339 *
340 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
341 * files, we can happily check the index for a match, but for creating a
342 * new file we should try to match whatever "patch" does. I have no idea.
343 */
344 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
345 {
346 char *name;
347
348 first += 4; /* skip "--- " */
349 second += 4; /* skip "+++ " */
350 if (!p_value_known) {
351 int p, q;
352 p = guess_p_value(first);
353 q = guess_p_value(second);
354 if (p < 0) p = q;
355 if (0 <= p && p == q) {
356 p_value = p;
357 p_value_known = 1;
358 }
359 }
360 if (is_dev_null(first)) {
361 patch->is_new = 1;
362 patch->is_delete = 0;
363 name = find_name(second, NULL, p_value, TERM_SPACE | TERM_TAB);
364 patch->new_name = name;
365 } else if (is_dev_null(second)) {
366 patch->is_new = 0;
367 patch->is_delete = 1;
368 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
369 patch->old_name = name;
370 } else {
371 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
372 name = find_name(second, name, p_value, TERM_SPACE | TERM_TAB);
373 patch->old_name = patch->new_name = name;
374 }
375 if (!name)
376 die("unable to find filename in patch at line %d", linenr);
377 }
378
379 static int gitdiff_hdrend(const char *line, struct patch *patch)
380 {
381 return -1;
382 }
383
384 /*
385 * We're anal about diff header consistency, to make
386 * sure that we don't end up having strange ambiguous
387 * patches floating around.
388 *
389 * As a result, gitdiff_{old|new}name() will check
390 * their names against any previous information, just
391 * to make sure..
392 */
393 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
394 {
395 if (!orig_name && !isnull)
396 return find_name(line, NULL, p_value, TERM_TAB);
397
398 if (orig_name) {
399 int len;
400 const char *name;
401 char *another;
402 name = orig_name;
403 len = strlen(name);
404 if (isnull)
405 die("git-apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
406 another = find_name(line, NULL, p_value, TERM_TAB);
407 if (!another || memcmp(another, name, len))
408 die("git-apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
409 free(another);
410 return orig_name;
411 }
412 else {
413 /* expect "/dev/null" */
414 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
415 die("git-apply: bad git-diff - expected /dev/null on line %d", linenr);
416 return NULL;
417 }
418 }
419
420 static int gitdiff_oldname(const char *line, struct patch *patch)
421 {
422 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
423 return 0;
424 }
425
426 static int gitdiff_newname(const char *line, struct patch *patch)
427 {
428 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
429 return 0;
430 }
431
432 static int gitdiff_oldmode(const char *line, struct patch *patch)
433 {
434 patch->old_mode = strtoul(line, NULL, 8);
435 return 0;
436 }
437
438 static int gitdiff_newmode(const char *line, struct patch *patch)
439 {
440 patch->new_mode = strtoul(line, NULL, 8);
441 return 0;
442 }
443
444 static int gitdiff_delete(const char *line, struct patch *patch)
445 {
446 patch->is_delete = 1;
447 patch->old_name = patch->def_name;
448 return gitdiff_oldmode(line, patch);
449 }
450
451 static int gitdiff_newfile(const char *line, struct patch *patch)
452 {
453 patch->is_new = 1;
454 patch->new_name = patch->def_name;
455 return gitdiff_newmode(line, patch);
456 }
457
458 static int gitdiff_copysrc(const char *line, struct patch *patch)
459 {
460 patch->is_copy = 1;
461 patch->old_name = find_name(line, NULL, 0, 0);
462 return 0;
463 }
464
465 static int gitdiff_copydst(const char *line, struct patch *patch)
466 {
467 patch->is_copy = 1;
468 patch->new_name = find_name(line, NULL, 0, 0);
469 return 0;
470 }
471
472 static int gitdiff_renamesrc(const char *line, struct patch *patch)
473 {
474 patch->is_rename = 1;
475 patch->old_name = find_name(line, NULL, 0, 0);
476 return 0;
477 }
478
479 static int gitdiff_renamedst(const char *line, struct patch *patch)
480 {
481 patch->is_rename = 1;
482 patch->new_name = find_name(line, NULL, 0, 0);
483 return 0;
484 }
485
486 static int gitdiff_similarity(const char *line, struct patch *patch)
487 {
488 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
489 patch->score = 0;
490 return 0;
491 }
492
493 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
494 {
495 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
496 patch->score = 0;
497 return 0;
498 }
499
500 static int gitdiff_index(const char *line, struct patch *patch)
501 {
502 /* index line is N hexadecimal, "..", N hexadecimal,
503 * and optional space with octal mode.
504 */
505 const char *ptr, *eol;
506 int len;
507
508 ptr = strchr(line, '.');
509 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
510 return 0;
511 len = ptr - line;
512 memcpy(patch->old_sha1_prefix, line, len);
513 patch->old_sha1_prefix[len] = 0;
514
515 line = ptr + 2;
516 ptr = strchr(line, ' ');
517 eol = strchr(line, '\n');
518
519 if (!ptr || eol < ptr)
520 ptr = eol;
521 len = ptr - line;
522
523 if (40 < len)
524 return 0;
525 memcpy(patch->new_sha1_prefix, line, len);
526 patch->new_sha1_prefix[len] = 0;
527 if (*ptr == ' ')
528 patch->new_mode = patch->old_mode = strtoul(ptr+1, NULL, 8);
529 return 0;
530 }
531
532 /*
533 * This is normal for a diff that doesn't change anything: we'll fall through
534 * into the next diff. Tell the parser to break out.
535 */
536 static int gitdiff_unrecognized(const char *line, struct patch *patch)
537 {
538 return -1;
539 }
540
541 static const char *stop_at_slash(const char *line, int llen)
542 {
543 int i;
544
545 for (i = 0; i < llen; i++) {
546 int ch = line[i];
547 if (ch == '/')
548 return line + i;
549 }
550 return NULL;
551 }
552
553 /* This is to extract the same name that appears on "diff --git"
554 * line. We do not find and return anything if it is a rename
555 * patch, and it is OK because we will find the name elsewhere.
556 * We need to reliably find name only when it is mode-change only,
557 * creation or deletion of an empty file. In any of these cases,
558 * both sides are the same name under a/ and b/ respectively.
559 */
560 static char *git_header_name(char *line, int llen)
561 {
562 const char *name;
563 const char *second = NULL;
564 size_t len;
565
566 line += strlen("diff --git ");
567 llen -= strlen("diff --git ");
568
569 if (*line == '"') {
570 const char *cp;
571 struct strbuf first;
572 struct strbuf sp;
573
574 strbuf_init(&first, 0);
575 strbuf_init(&sp, 0);
576
577 if (unquote_c_style(&first, line, &second))
578 goto free_and_fail1;
579
580 /* advance to the first slash */
581 cp = stop_at_slash(first.buf, first.len);
582 /* we do not accept absolute paths */
583 if (!cp || cp == first.buf)
584 goto free_and_fail1;
585 strbuf_remove(&first, 0, cp + 1 - first.buf);
586
587 /* second points at one past closing dq of name.
588 * find the second name.
589 */
590 while ((second < line + llen) && isspace(*second))
591 second++;
592
593 if (line + llen <= second)
594 goto free_and_fail1;
595 if (*second == '"') {
596 if (unquote_c_style(&sp, second, NULL))
597 goto free_and_fail1;
598 cp = stop_at_slash(sp.buf, sp.len);
599 if (!cp || cp == sp.buf)
600 goto free_and_fail1;
601 /* They must match, otherwise ignore */
602 if (strcmp(cp + 1, first.buf))
603 goto free_and_fail1;
604 strbuf_release(&sp);
605 return strbuf_detach(&first, NULL);
606 }
607
608 /* unquoted second */
609 cp = stop_at_slash(second, line + llen - second);
610 if (!cp || cp == second)
611 goto free_and_fail1;
612 cp++;
613 if (line + llen - cp != first.len + 1 ||
614 memcmp(first.buf, cp, first.len))
615 goto free_and_fail1;
616 return strbuf_detach(&first, NULL);
617
618 free_and_fail1:
619 strbuf_release(&first);
620 strbuf_release(&sp);
621 return NULL;
622 }
623
624 /* unquoted first name */
625 name = stop_at_slash(line, llen);
626 if (!name || name == line)
627 return NULL;
628 name++;
629
630 /* since the first name is unquoted, a dq if exists must be
631 * the beginning of the second name.
632 */
633 for (second = name; second < line + llen; second++) {
634 if (*second == '"') {
635 struct strbuf sp;
636 const char *np;
637
638 strbuf_init(&sp, 0);
639 if (unquote_c_style(&sp, second, NULL))
640 goto free_and_fail2;
641
642 np = stop_at_slash(sp.buf, sp.len);
643 if (!np || np == sp.buf)
644 goto free_and_fail2;
645 np++;
646
647 len = sp.buf + sp.len - np;
648 if (len < second - name &&
649 !strncmp(np, name, len) &&
650 isspace(name[len])) {
651 /* Good */
652 strbuf_remove(&sp, 0, np - sp.buf);
653 return strbuf_detach(&sp, NULL);
654 }
655
656 free_and_fail2:
657 strbuf_release(&sp);
658 return NULL;
659 }
660 }
661
662 /*
663 * Accept a name only if it shows up twice, exactly the same
664 * form.
665 */
666 for (len = 0 ; ; len++) {
667 switch (name[len]) {
668 default:
669 continue;
670 case '\n':
671 return NULL;
672 case '\t': case ' ':
673 second = name+len;
674 for (;;) {
675 char c = *second++;
676 if (c == '\n')
677 return NULL;
678 if (c == '/')
679 break;
680 }
681 if (second[len] == '\n' && !memcmp(name, second, len)) {
682 return xmemdupz(name, len);
683 }
684 }
685 }
686 }
687
688 /* Verify that we recognize the lines following a git header */
689 static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
690 {
691 unsigned long offset;
692
693 /* A git diff has explicit new/delete information, so we don't guess */
694 patch->is_new = 0;
695 patch->is_delete = 0;
696
697 /*
698 * Some things may not have the old name in the
699 * rest of the headers anywhere (pure mode changes,
700 * or removing or adding empty files), so we get
701 * the default name from the header.
702 */
703 patch->def_name = git_header_name(line, len);
704
705 line += len;
706 size -= len;
707 linenr++;
708 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
709 static const struct opentry {
710 const char *str;
711 int (*fn)(const char *, struct patch *);
712 } optable[] = {
713 { "@@ -", gitdiff_hdrend },
714 { "--- ", gitdiff_oldname },
715 { "+++ ", gitdiff_newname },
716 { "old mode ", gitdiff_oldmode },
717 { "new mode ", gitdiff_newmode },
718 { "deleted file mode ", gitdiff_delete },
719 { "new file mode ", gitdiff_newfile },
720 { "copy from ", gitdiff_copysrc },
721 { "copy to ", gitdiff_copydst },
722 { "rename old ", gitdiff_renamesrc },
723 { "rename new ", gitdiff_renamedst },
724 { "rename from ", gitdiff_renamesrc },
725 { "rename to ", gitdiff_renamedst },
726 { "similarity index ", gitdiff_similarity },
727 { "dissimilarity index ", gitdiff_dissimilarity },
728 { "index ", gitdiff_index },
729 { "", gitdiff_unrecognized },
730 };
731 int i;
732
733 len = linelen(line, size);
734 if (!len || line[len-1] != '\n')
735 break;
736 for (i = 0; i < ARRAY_SIZE(optable); i++) {
737 const struct opentry *p = optable + i;
738 int oplen = strlen(p->str);
739 if (len < oplen || memcmp(p->str, line, oplen))
740 continue;
741 if (p->fn(line + oplen, patch) < 0)
742 return offset;
743 break;
744 }
745 }
746
747 return offset;
748 }
749
750 static int parse_num(const char *line, unsigned long *p)
751 {
752 char *ptr;
753
754 if (!isdigit(*line))
755 return 0;
756 *p = strtoul(line, &ptr, 10);
757 return ptr - line;
758 }
759
760 static int parse_range(const char *line, int len, int offset, const char *expect,
761 unsigned long *p1, unsigned long *p2)
762 {
763 int digits, ex;
764
765 if (offset < 0 || offset >= len)
766 return -1;
767 line += offset;
768 len -= offset;
769
770 digits = parse_num(line, p1);
771 if (!digits)
772 return -1;
773
774 offset += digits;
775 line += digits;
776 len -= digits;
777
778 *p2 = 1;
779 if (*line == ',') {
780 digits = parse_num(line+1, p2);
781 if (!digits)
782 return -1;
783
784 offset += digits+1;
785 line += digits+1;
786 len -= digits+1;
787 }
788
789 ex = strlen(expect);
790 if (ex > len)
791 return -1;
792 if (memcmp(line, expect, ex))
793 return -1;
794
795 return offset + ex;
796 }
797
798 /*
799 * Parse a unified diff fragment header of the
800 * form "@@ -a,b +c,d @@"
801 */
802 static int parse_fragment_header(char *line, int len, struct fragment *fragment)
803 {
804 int offset;
805
806 if (!len || line[len-1] != '\n')
807 return -1;
808
809 /* Figure out the number of lines in a fragment */
810 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
811 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
812
813 return offset;
814 }
815
816 static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
817 {
818 unsigned long offset, len;
819
820 patch->is_toplevel_relative = 0;
821 patch->is_rename = patch->is_copy = 0;
822 patch->is_new = patch->is_delete = -1;
823 patch->old_mode = patch->new_mode = 0;
824 patch->old_name = patch->new_name = NULL;
825 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
826 unsigned long nextlen;
827
828 len = linelen(line, size);
829 if (!len)
830 break;
831
832 /* Testing this early allows us to take a few shortcuts.. */
833 if (len < 6)
834 continue;
835
836 /*
837 * Make sure we don't find any unconnected patch fragments.
838 * That's a sign that we didn't find a header, and that a
839 * patch has become corrupted/broken up.
840 */
841 if (!memcmp("@@ -", line, 4)) {
842 struct fragment dummy;
843 if (parse_fragment_header(line, len, &dummy) < 0)
844 continue;
845 die("patch fragment without header at line %d: %.*s",
846 linenr, (int)len-1, line);
847 }
848
849 if (size < len + 6)
850 break;
851
852 /*
853 * Git patch? It might not have a real patch, just a rename
854 * or mode change, so we handle that specially
855 */
856 if (!memcmp("diff --git ", line, 11)) {
857 int git_hdr_len = parse_git_header(line, len, size, patch);
858 if (git_hdr_len <= len)
859 continue;
860 if (!patch->old_name && !patch->new_name) {
861 if (!patch->def_name)
862 die("git diff header lacks filename information (line %d)", linenr);
863 patch->old_name = patch->new_name = patch->def_name;
864 }
865 patch->is_toplevel_relative = 1;
866 *hdrsize = git_hdr_len;
867 return offset;
868 }
869
870 /** --- followed by +++ ? */
871 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
872 continue;
873
874 /*
875 * We only accept unified patches, so we want it to
876 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
877 * minimum
878 */
879 nextlen = linelen(line + len, size - len);
880 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
881 continue;
882
883 /* Ok, we'll consider it a patch */
884 parse_traditional_patch(line, line+len, patch);
885 *hdrsize = len + nextlen;
886 linenr += 2;
887 return offset;
888 }
889 return -1;
890 }
891
892 static void check_whitespace(const char *line, int len)
893 {
894 const char *err = "Adds trailing whitespace";
895 int seen_space = 0;
896 int i;
897
898 /*
899 * We know len is at least two, since we have a '+' and we
900 * checked that the last character was a '\n' before calling
901 * this function. That is, an addition of an empty line would
902 * check the '+' here. Sneaky...
903 */
904 if (isspace(line[len-2]))
905 goto error;
906
907 /*
908 * Make sure that there is no space followed by a tab in
909 * indentation.
910 */
911 err = "Space in indent is followed by a tab";
912 for (i = 1; i < len; i++) {
913 if (line[i] == '\t') {
914 if (seen_space)
915 goto error;
916 }
917 else if (line[i] == ' ')
918 seen_space = 1;
919 else
920 break;
921 }
922 return;
923
924 error:
925 whitespace_error++;
926 if (squelch_whitespace_errors &&
927 squelch_whitespace_errors < whitespace_error)
928 ;
929 else
930 fprintf(stderr, "%s.\n%s:%d:%.*s\n",
931 err, patch_input_file, linenr, len-2, line+1);
932 }
933
934
935 /*
936 * Parse a unified diff. Note that this really needs to parse each
937 * fragment separately, since the only way to know the difference
938 * between a "---" that is part of a patch, and a "---" that starts
939 * the next patch is to look at the line counts..
940 */
941 static int parse_fragment(char *line, unsigned long size, struct patch *patch, struct fragment *fragment)
942 {
943 int added, deleted;
944 int len = linelen(line, size), offset;
945 unsigned long oldlines, newlines;
946 unsigned long leading, trailing;
947
948 offset = parse_fragment_header(line, len, fragment);
949 if (offset < 0)
950 return -1;
951 oldlines = fragment->oldlines;
952 newlines = fragment->newlines;
953 leading = 0;
954 trailing = 0;
955
956 /* Parse the thing.. */
957 line += len;
958 size -= len;
959 linenr++;
960 added = deleted = 0;
961 for (offset = len;
962 0 < size;
963 offset += len, size -= len, line += len, linenr++) {
964 if (!oldlines && !newlines)
965 break;
966 len = linelen(line, size);
967 if (!len || line[len-1] != '\n')
968 return -1;
969 switch (*line) {
970 default:
971 return -1;
972 case '\n': /* newer GNU diff, an empty context line */
973 case ' ':
974 oldlines--;
975 newlines--;
976 if (!deleted && !added)
977 leading++;
978 trailing++;
979 break;
980 case '-':
981 if (apply_in_reverse &&
982 new_whitespace != nowarn_whitespace)
983 check_whitespace(line, len);
984 deleted++;
985 oldlines--;
986 trailing = 0;
987 break;
988 case '+':
989 if (!apply_in_reverse &&
990 new_whitespace != nowarn_whitespace)
991 check_whitespace(line, len);
992 added++;
993 newlines--;
994 trailing = 0;
995 break;
996
997 /* We allow "\ No newline at end of file". Depending
998 * on locale settings when the patch was produced we
999 * don't know what this line looks like. The only
1000 * thing we do know is that it begins with "\ ".
1001 * Checking for 12 is just for sanity check -- any
1002 * l10n of "\ No newline..." is at least that long.
1003 */
1004 case '\\':
1005 if (len < 12 || memcmp(line, "\\ ", 2))
1006 return -1;
1007 break;
1008 }
1009 }
1010 if (oldlines || newlines)
1011 return -1;
1012 fragment->leading = leading;
1013 fragment->trailing = trailing;
1014
1015 /* If a fragment ends with an incomplete line, we failed to include
1016 * it in the above loop because we hit oldlines == newlines == 0
1017 * before seeing it.
1018 */
1019 if (12 < size && !memcmp(line, "\\ ", 2))
1020 offset += linelen(line, size);
1021
1022 patch->lines_added += added;
1023 patch->lines_deleted += deleted;
1024
1025 if (0 < patch->is_new && oldlines)
1026 return error("new file depends on old contents");
1027 if (0 < patch->is_delete && newlines)
1028 return error("deleted file still has contents");
1029 return offset;
1030 }
1031
1032 static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
1033 {
1034 unsigned long offset = 0;
1035 unsigned long oldlines = 0, newlines = 0, context = 0;
1036 struct fragment **fragp = &patch->fragments;
1037
1038 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1039 struct fragment *fragment;
1040 int len;
1041
1042 fragment = xcalloc(1, sizeof(*fragment));
1043 len = parse_fragment(line, size, patch, fragment);
1044 if (len <= 0)
1045 die("corrupt patch at line %d", linenr);
1046 fragment->patch = line;
1047 fragment->size = len;
1048 oldlines += fragment->oldlines;
1049 newlines += fragment->newlines;
1050 context += fragment->leading + fragment->trailing;
1051
1052 *fragp = fragment;
1053 fragp = &fragment->next;
1054
1055 offset += len;
1056 line += len;
1057 size -= len;
1058 }
1059
1060 /*
1061 * If something was removed (i.e. we have old-lines) it cannot
1062 * be creation, and if something was added it cannot be
1063 * deletion. However, the reverse is not true; --unified=0
1064 * patches that only add are not necessarily creation even
1065 * though they do not have any old lines, and ones that only
1066 * delete are not necessarily deletion.
1067 *
1068 * Unfortunately, a real creation/deletion patch do _not_ have
1069 * any context line by definition, so we cannot safely tell it
1070 * apart with --unified=0 insanity. At least if the patch has
1071 * more than one hunk it is not creation or deletion.
1072 */
1073 if (patch->is_new < 0 &&
1074 (oldlines || (patch->fragments && patch->fragments->next)))
1075 patch->is_new = 0;
1076 if (patch->is_delete < 0 &&
1077 (newlines || (patch->fragments && patch->fragments->next)))
1078 patch->is_delete = 0;
1079 if (!unidiff_zero || context) {
1080 /* If the user says the patch is not generated with
1081 * --unified=0, or if we have seen context lines,
1082 * then not having oldlines means the patch is creation,
1083 * and not having newlines means the patch is deletion.
1084 */
1085 if (patch->is_new < 0 && !oldlines) {
1086 patch->is_new = 1;
1087 patch->old_name = NULL;
1088 }
1089 if (patch->is_delete < 0 && !newlines) {
1090 patch->is_delete = 1;
1091 patch->new_name = NULL;
1092 }
1093 }
1094
1095 if (0 < patch->is_new && oldlines)
1096 die("new file %s depends on old contents", patch->new_name);
1097 if (0 < patch->is_delete && newlines)
1098 die("deleted file %s still has contents", patch->old_name);
1099 if (!patch->is_delete && !newlines && context)
1100 fprintf(stderr, "** warning: file %s becomes empty but "
1101 "is not deleted\n", patch->new_name);
1102
1103 return offset;
1104 }
1105
1106 static inline int metadata_changes(struct patch *patch)
1107 {
1108 return patch->is_rename > 0 ||
1109 patch->is_copy > 0 ||
1110 patch->is_new > 0 ||
1111 patch->is_delete ||
1112 (patch->old_mode && patch->new_mode &&
1113 patch->old_mode != patch->new_mode);
1114 }
1115
1116 static char *inflate_it(const void *data, unsigned long size,
1117 unsigned long inflated_size)
1118 {
1119 z_stream stream;
1120 void *out;
1121 int st;
1122
1123 memset(&stream, 0, sizeof(stream));
1124
1125 stream.next_in = (unsigned char *)data;
1126 stream.avail_in = size;
1127 stream.next_out = out = xmalloc(inflated_size);
1128 stream.avail_out = inflated_size;
1129 inflateInit(&stream);
1130 st = inflate(&stream, Z_FINISH);
1131 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1132 free(out);
1133 return NULL;
1134 }
1135 return out;
1136 }
1137
1138 static struct fragment *parse_binary_hunk(char **buf_p,
1139 unsigned long *sz_p,
1140 int *status_p,
1141 int *used_p)
1142 {
1143 /* Expect a line that begins with binary patch method ("literal"
1144 * or "delta"), followed by the length of data before deflating.
1145 * a sequence of 'length-byte' followed by base-85 encoded data
1146 * should follow, terminated by a newline.
1147 *
1148 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1149 * and we would limit the patch line to 66 characters,
1150 * so one line can fit up to 13 groups that would decode
1151 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1152 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1153 */
1154 int llen, used;
1155 unsigned long size = *sz_p;
1156 char *buffer = *buf_p;
1157 int patch_method;
1158 unsigned long origlen;
1159 char *data = NULL;
1160 int hunk_size = 0;
1161 struct fragment *frag;
1162
1163 llen = linelen(buffer, size);
1164 used = llen;
1165
1166 *status_p = 0;
1167
1168 if (!prefixcmp(buffer, "delta ")) {
1169 patch_method = BINARY_DELTA_DEFLATED;
1170 origlen = strtoul(buffer + 6, NULL, 10);
1171 }
1172 else if (!prefixcmp(buffer, "literal ")) {
1173 patch_method = BINARY_LITERAL_DEFLATED;
1174 origlen = strtoul(buffer + 8, NULL, 10);
1175 }
1176 else
1177 return NULL;
1178
1179 linenr++;
1180 buffer += llen;
1181 while (1) {
1182 int byte_length, max_byte_length, newsize;
1183 llen = linelen(buffer, size);
1184 used += llen;
1185 linenr++;
1186 if (llen == 1) {
1187 /* consume the blank line */
1188 buffer++;
1189 size--;
1190 break;
1191 }
1192 /* Minimum line is "A00000\n" which is 7-byte long,
1193 * and the line length must be multiple of 5 plus 2.
1194 */
1195 if ((llen < 7) || (llen-2) % 5)
1196 goto corrupt;
1197 max_byte_length = (llen - 2) / 5 * 4;
1198 byte_length = *buffer;
1199 if ('A' <= byte_length && byte_length <= 'Z')
1200 byte_length = byte_length - 'A' + 1;
1201 else if ('a' <= byte_length && byte_length <= 'z')
1202 byte_length = byte_length - 'a' + 27;
1203 else
1204 goto corrupt;
1205 /* if the input length was not multiple of 4, we would
1206 * have filler at the end but the filler should never
1207 * exceed 3 bytes
1208 */
1209 if (max_byte_length < byte_length ||
1210 byte_length <= max_byte_length - 4)
1211 goto corrupt;
1212 newsize = hunk_size + byte_length;
1213 data = xrealloc(data, newsize);
1214 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1215 goto corrupt;
1216 hunk_size = newsize;
1217 buffer += llen;
1218 size -= llen;
1219 }
1220
1221 frag = xcalloc(1, sizeof(*frag));
1222 frag->patch = inflate_it(data, hunk_size, origlen);
1223 if (!frag->patch)
1224 goto corrupt;
1225 free(data);
1226 frag->size = origlen;
1227 *buf_p = buffer;
1228 *sz_p = size;
1229 *used_p = used;
1230 frag->binary_patch_method = patch_method;
1231 return frag;
1232
1233 corrupt:
1234 free(data);
1235 *status_p = -1;
1236 error("corrupt binary patch at line %d: %.*s",
1237 linenr-1, llen-1, buffer);
1238 return NULL;
1239 }
1240
1241 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1242 {
1243 /* We have read "GIT binary patch\n"; what follows is a line
1244 * that says the patch method (currently, either "literal" or
1245 * "delta") and the length of data before deflating; a
1246 * sequence of 'length-byte' followed by base-85 encoded data
1247 * follows.
1248 *
1249 * When a binary patch is reversible, there is another binary
1250 * hunk in the same format, starting with patch method (either
1251 * "literal" or "delta") with the length of data, and a sequence
1252 * of length-byte + base-85 encoded data, terminated with another
1253 * empty line. This data, when applied to the postimage, produces
1254 * the preimage.
1255 */
1256 struct fragment *forward;
1257 struct fragment *reverse;
1258 int status;
1259 int used, used_1;
1260
1261 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1262 if (!forward && !status)
1263 /* there has to be one hunk (forward hunk) */
1264 return error("unrecognized binary patch at line %d", linenr-1);
1265 if (status)
1266 /* otherwise we already gave an error message */
1267 return status;
1268
1269 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1270 if (reverse)
1271 used += used_1;
1272 else if (status) {
1273 /* not having reverse hunk is not an error, but having
1274 * a corrupt reverse hunk is.
1275 */
1276 free((void*) forward->patch);
1277 free(forward);
1278 return status;
1279 }
1280 forward->next = reverse;
1281 patch->fragments = forward;
1282 patch->is_binary = 1;
1283 return used;
1284 }
1285
1286 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1287 {
1288 int hdrsize, patchsize;
1289 int offset = find_header(buffer, size, &hdrsize, patch);
1290
1291 if (offset < 0)
1292 return offset;
1293
1294 patchsize = parse_single_patch(buffer + offset + hdrsize, size - offset - hdrsize, patch);
1295
1296 if (!patchsize) {
1297 static const char *binhdr[] = {
1298 "Binary files ",
1299 "Files ",
1300 NULL,
1301 };
1302 static const char git_binary[] = "GIT binary patch\n";
1303 int i;
1304 int hd = hdrsize + offset;
1305 unsigned long llen = linelen(buffer + hd, size - hd);
1306
1307 if (llen == sizeof(git_binary) - 1 &&
1308 !memcmp(git_binary, buffer + hd, llen)) {
1309 int used;
1310 linenr++;
1311 used = parse_binary(buffer + hd + llen,
1312 size - hd - llen, patch);
1313 if (used)
1314 patchsize = used + llen;
1315 else
1316 patchsize = 0;
1317 }
1318 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1319 for (i = 0; binhdr[i]; i++) {
1320 int len = strlen(binhdr[i]);
1321 if (len < size - hd &&
1322 !memcmp(binhdr[i], buffer + hd, len)) {
1323 linenr++;
1324 patch->is_binary = 1;
1325 patchsize = llen;
1326 break;
1327 }
1328 }
1329 }
1330
1331 /* Empty patch cannot be applied if it is a text patch
1332 * without metadata change. A binary patch appears
1333 * empty to us here.
1334 */
1335 if ((apply || check) &&
1336 (!patch->is_binary && !metadata_changes(patch)))
1337 die("patch with only garbage at line %d", linenr);
1338 }
1339
1340 return offset + hdrsize + patchsize;
1341 }
1342
1343 #define swap(a,b) myswap((a),(b),sizeof(a))
1344
1345 #define myswap(a, b, size) do { \
1346 unsigned char mytmp[size]; \
1347 memcpy(mytmp, &a, size); \
1348 memcpy(&a, &b, size); \
1349 memcpy(&b, mytmp, size); \
1350 } while (0)
1351
1352 static void reverse_patches(struct patch *p)
1353 {
1354 for (; p; p = p->next) {
1355 struct fragment *frag = p->fragments;
1356
1357 swap(p->new_name, p->old_name);
1358 swap(p->new_mode, p->old_mode);
1359 swap(p->is_new, p->is_delete);
1360 swap(p->lines_added, p->lines_deleted);
1361 swap(p->old_sha1_prefix, p->new_sha1_prefix);
1362
1363 for (; frag; frag = frag->next) {
1364 swap(frag->newpos, frag->oldpos);
1365 swap(frag->newlines, frag->oldlines);
1366 }
1367 }
1368 }
1369
1370 static const char pluses[] = "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1371 static const char minuses[]= "----------------------------------------------------------------------";
1372
1373 static void show_stats(struct patch *patch)
1374 {
1375 struct strbuf qname;
1376 char *cp = patch->new_name ? patch->new_name : patch->old_name;
1377 int max, add, del;
1378
1379 strbuf_init(&qname, 0);
1380 quote_c_style(cp, &qname, NULL, 0);
1381
1382 /*
1383 * "scale" the filename
1384 */
1385 max = max_len;
1386 if (max > 50)
1387 max = 50;
1388
1389 if (qname.len > max) {
1390 cp = strchr(qname.buf + qname.len + 3 - max, '/');
1391 if (!cp)
1392 cp = qname.buf + qname.len + 3 - max;
1393 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
1394 }
1395
1396 if (patch->is_binary) {
1397 printf(" %-*s | Bin\n", max, qname.buf);
1398 strbuf_release(&qname);
1399 return;
1400 }
1401
1402 printf(" %-*s |", max, qname.buf);
1403 strbuf_release(&qname);
1404
1405 /*
1406 * scale the add/delete
1407 */
1408 max = max + max_change > 70 ? 70 - max : max_change;
1409 add = patch->lines_added;
1410 del = patch->lines_deleted;
1411
1412 if (max_change > 0) {
1413 int total = ((add + del) * max + max_change / 2) / max_change;
1414 add = (add * max + max_change / 2) / max_change;
1415 del = total - add;
1416 }
1417 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
1418 add, pluses, del, minuses);
1419 }
1420
1421 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
1422 {
1423 switch (st->st_mode & S_IFMT) {
1424 case S_IFLNK:
1425 strbuf_grow(buf, st->st_size);
1426 if (readlink(path, buf->buf, st->st_size) != st->st_size)
1427 return -1;
1428 strbuf_setlen(buf, st->st_size);
1429 return 0;
1430 case S_IFREG:
1431 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
1432 return error("unable to open or read %s", path);
1433 convert_to_git(path, buf->buf, buf->len, buf);
1434 return 0;
1435 default:
1436 return -1;
1437 }
1438 }
1439
1440 static int find_offset(const char *buf, unsigned long size, const char *fragment, unsigned long fragsize, int line, int *lines)
1441 {
1442 int i;
1443 unsigned long start, backwards, forwards;
1444
1445 if (fragsize > size)
1446 return -1;
1447
1448 start = 0;
1449 if (line > 1) {
1450 unsigned long offset = 0;
1451 i = line-1;
1452 while (offset + fragsize <= size) {
1453 if (buf[offset++] == '\n') {
1454 start = offset;
1455 if (!--i)
1456 break;
1457 }
1458 }
1459 }
1460
1461 /* Exact line number? */
1462 if ((start + fragsize <= size) &&
1463 !memcmp(buf + start, fragment, fragsize))
1464 return start;
1465
1466 /*
1467 * There's probably some smart way to do this, but I'll leave
1468 * that to the smart and beautiful people. I'm simple and stupid.
1469 */
1470 backwards = start;
1471 forwards = start;
1472 for (i = 0; ; i++) {
1473 unsigned long try;
1474 int n;
1475
1476 /* "backward" */
1477 if (i & 1) {
1478 if (!backwards) {
1479 if (forwards + fragsize > size)
1480 break;
1481 continue;
1482 }
1483 do {
1484 --backwards;
1485 } while (backwards && buf[backwards-1] != '\n');
1486 try = backwards;
1487 } else {
1488 while (forwards + fragsize <= size) {
1489 if (buf[forwards++] == '\n')
1490 break;
1491 }
1492 try = forwards;
1493 }
1494
1495 if (try + fragsize > size)
1496 continue;
1497 if (memcmp(buf + try, fragment, fragsize))
1498 continue;
1499 n = (i >> 1)+1;
1500 if (i & 1)
1501 n = -n;
1502 *lines = n;
1503 return try;
1504 }
1505
1506 /*
1507 * We should start searching forward and backward.
1508 */
1509 return -1;
1510 }
1511
1512 static void remove_first_line(const char **rbuf, int *rsize)
1513 {
1514 const char *buf = *rbuf;
1515 int size = *rsize;
1516 unsigned long offset;
1517 offset = 0;
1518 while (offset <= size) {
1519 if (buf[offset++] == '\n')
1520 break;
1521 }
1522 *rsize = size - offset;
1523 *rbuf = buf + offset;
1524 }
1525
1526 static void remove_last_line(const char **rbuf, int *rsize)
1527 {
1528 const char *buf = *rbuf;
1529 int size = *rsize;
1530 unsigned long offset;
1531 offset = size - 1;
1532 while (offset > 0) {
1533 if (buf[--offset] == '\n')
1534 break;
1535 }
1536 *rsize = offset + 1;
1537 }
1538
1539 static int apply_line(char *output, const char *patch, int plen)
1540 {
1541 /* plen is number of bytes to be copied from patch,
1542 * starting at patch+1 (patch[0] is '+'). Typically
1543 * patch[plen] is '\n', unless this is the incomplete
1544 * last line.
1545 */
1546 int i;
1547 int add_nl_to_tail = 0;
1548 int fixed = 0;
1549 int last_tab_in_indent = -1;
1550 int last_space_in_indent = -1;
1551 int need_fix_leading_space = 0;
1552 char *buf;
1553
1554 if ((new_whitespace != strip_whitespace) || !whitespace_error ||
1555 *patch != '+') {
1556 memcpy(output, patch + 1, plen);
1557 return plen;
1558 }
1559
1560 if (1 < plen && isspace(patch[plen-1])) {
1561 if (patch[plen] == '\n')
1562 add_nl_to_tail = 1;
1563 plen--;
1564 while (0 < plen && isspace(patch[plen]))
1565 plen--;
1566 fixed = 1;
1567 }
1568
1569 for (i = 1; i < plen; i++) {
1570 char ch = patch[i];
1571 if (ch == '\t') {
1572 last_tab_in_indent = i;
1573 if (0 <= last_space_in_indent)
1574 need_fix_leading_space = 1;
1575 }
1576 else if (ch == ' ')
1577 last_space_in_indent = i;
1578 else
1579 break;
1580 }
1581
1582 buf = output;
1583 if (need_fix_leading_space) {
1584 int consecutive_spaces = 0;
1585 /* between patch[1..last_tab_in_indent] strip the
1586 * funny spaces, updating them to tab as needed.
1587 */
1588 for (i = 1; i < last_tab_in_indent; i++, plen--) {
1589 char ch = patch[i];
1590 if (ch != ' ') {
1591 consecutive_spaces = 0;
1592 *output++ = ch;
1593 } else {
1594 consecutive_spaces++;
1595 if (consecutive_spaces == 8) {
1596 *output++ = '\t';
1597 consecutive_spaces = 0;
1598 }
1599 }
1600 }
1601 fixed = 1;
1602 i = last_tab_in_indent;
1603 }
1604 else
1605 i = 1;
1606
1607 memcpy(output, patch + i, plen);
1608 if (add_nl_to_tail)
1609 output[plen++] = '\n';
1610 if (fixed)
1611 applied_after_fixing_ws++;
1612 return output + plen - buf;
1613 }
1614
1615 static int apply_one_fragment(struct strbuf *buf, struct fragment *frag, int inaccurate_eof)
1616 {
1617 int match_beginning, match_end;
1618 const char *patch = frag->patch;
1619 int offset, size = frag->size;
1620 char *old = xmalloc(size);
1621 char *new = xmalloc(size);
1622 const char *oldlines, *newlines;
1623 int oldsize = 0, newsize = 0;
1624 int new_blank_lines_at_end = 0;
1625 unsigned long leading, trailing;
1626 int pos, lines;
1627
1628 while (size > 0) {
1629 char first;
1630 int len = linelen(patch, size);
1631 int plen;
1632 int added_blank_line = 0;
1633
1634 if (!len)
1635 break;
1636
1637 /*
1638 * "plen" is how much of the line we should use for
1639 * the actual patch data. Normally we just remove the
1640 * first character on the line, but if the line is
1641 * followed by "\ No newline", then we also remove the
1642 * last one (which is the newline, of course).
1643 */
1644 plen = len-1;
1645 if (len < size && patch[len] == '\\')
1646 plen--;
1647 first = *patch;
1648 if (apply_in_reverse) {
1649 if (first == '-')
1650 first = '+';
1651 else if (first == '+')
1652 first = '-';
1653 }
1654
1655 switch (first) {
1656 case '\n':
1657 /* Newer GNU diff, empty context line */
1658 if (plen < 0)
1659 /* ... followed by '\No newline'; nothing */
1660 break;
1661 old[oldsize++] = '\n';
1662 new[newsize++] = '\n';
1663 break;
1664 case ' ':
1665 case '-':
1666 memcpy(old + oldsize, patch + 1, plen);
1667 oldsize += plen;
1668 if (first == '-')
1669 break;
1670 /* Fall-through for ' ' */
1671 case '+':
1672 if (first != '+' || !no_add) {
1673 int added = apply_line(new + newsize, patch,
1674 plen);
1675 newsize += added;
1676 if (first == '+' &&
1677 added == 1 && new[newsize-1] == '\n')
1678 added_blank_line = 1;
1679 }
1680 break;
1681 case '@': case '\\':
1682 /* Ignore it, we already handled it */
1683 break;
1684 default:
1685 if (apply_verbosely)
1686 error("invalid start of line: '%c'", first);
1687 return -1;
1688 }
1689 if (added_blank_line)
1690 new_blank_lines_at_end++;
1691 else
1692 new_blank_lines_at_end = 0;
1693 patch += len;
1694 size -= len;
1695 }
1696
1697 if (inaccurate_eof && oldsize > 0 && old[oldsize - 1] == '\n' &&
1698 newsize > 0 && new[newsize - 1] == '\n') {
1699 oldsize--;
1700 newsize--;
1701 }
1702
1703 oldlines = old;
1704 newlines = new;
1705 leading = frag->leading;
1706 trailing = frag->trailing;
1707
1708 /*
1709 * If we don't have any leading/trailing data in the patch,
1710 * we want it to match at the beginning/end of the file.
1711 *
1712 * But that would break if the patch is generated with
1713 * --unified=0; sane people wouldn't do that to cause us
1714 * trouble, but we try to please not so sane ones as well.
1715 */
1716 if (unidiff_zero) {
1717 match_beginning = (!leading && !frag->oldpos);
1718 match_end = 0;
1719 }
1720 else {
1721 match_beginning = !leading && (frag->oldpos == 1);
1722 match_end = !trailing;
1723 }
1724
1725 lines = 0;
1726 pos = frag->newpos;
1727 for (;;) {
1728 offset = find_offset(buf->buf, buf->len,
1729 oldlines, oldsize, pos, &lines);
1730 if (match_end && offset + oldsize != buf->len)
1731 offset = -1;
1732 if (match_beginning && offset)
1733 offset = -1;
1734 if (offset >= 0) {
1735 if (new_whitespace == strip_whitespace &&
1736 (buf->len - oldsize - offset == 0)) /* end of file? */
1737 newsize -= new_blank_lines_at_end;
1738
1739 /* Warn if it was necessary to reduce the number
1740 * of context lines.
1741 */
1742 if ((leading != frag->leading) ||
1743 (trailing != frag->trailing))
1744 fprintf(stderr, "Context reduced to (%ld/%ld)"
1745 " to apply fragment at %d\n",
1746 leading, trailing, pos + lines);
1747
1748 strbuf_splice(buf, offset, oldsize, newlines, newsize);
1749 offset = 0;
1750 break;
1751 }
1752
1753 /* Am I at my context limits? */
1754 if ((leading <= p_context) && (trailing <= p_context))
1755 break;
1756 if (match_beginning || match_end) {
1757 match_beginning = match_end = 0;
1758 continue;
1759 }
1760 /* Reduce the number of context lines
1761 * Reduce both leading and trailing if they are equal
1762 * otherwise just reduce the larger context.
1763 */
1764 if (leading >= trailing) {
1765 remove_first_line(&oldlines, &oldsize);
1766 remove_first_line(&newlines, &newsize);
1767 pos--;
1768 leading--;
1769 }
1770 if (trailing > leading) {
1771 remove_last_line(&oldlines, &oldsize);
1772 remove_last_line(&newlines, &newsize);
1773 trailing--;
1774 }
1775 }
1776
1777 if (offset && apply_verbosely)
1778 error("while searching for:\n%.*s", oldsize, oldlines);
1779
1780 free(old);
1781 free(new);
1782 return offset;
1783 }
1784
1785 static int apply_binary_fragment(struct strbuf *buf, struct patch *patch)
1786 {
1787 struct fragment *fragment = patch->fragments;
1788 unsigned long len;
1789 void *dst;
1790
1791 /* Binary patch is irreversible without the optional second hunk */
1792 if (apply_in_reverse) {
1793 if (!fragment->next)
1794 return error("cannot reverse-apply a binary patch "
1795 "without the reverse hunk to '%s'",
1796 patch->new_name
1797 ? patch->new_name : patch->old_name);
1798 fragment = fragment->next;
1799 }
1800 switch (fragment->binary_patch_method) {
1801 case BINARY_DELTA_DEFLATED:
1802 dst = patch_delta(buf->buf, buf->len, fragment->patch,
1803 fragment->size, &len);
1804 if (!dst)
1805 return -1;
1806 /* XXX patch_delta NUL-terminates */
1807 strbuf_attach(buf, dst, len, len + 1);
1808 return 0;
1809 case BINARY_LITERAL_DEFLATED:
1810 strbuf_reset(buf);
1811 strbuf_add(buf, fragment->patch, fragment->size);
1812 return 0;
1813 }
1814 return -1;
1815 }
1816
1817 static int apply_binary(struct strbuf *buf, struct patch *patch)
1818 {
1819 const char *name = patch->old_name ? patch->old_name : patch->new_name;
1820 unsigned char sha1[20];
1821
1822 /* For safety, we require patch index line to contain
1823 * full 40-byte textual SHA1 for old and new, at least for now.
1824 */
1825 if (strlen(patch->old_sha1_prefix) != 40 ||
1826 strlen(patch->new_sha1_prefix) != 40 ||
1827 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
1828 get_sha1_hex(patch->new_sha1_prefix, sha1))
1829 return error("cannot apply binary patch to '%s' "
1830 "without full index line", name);
1831
1832 if (patch->old_name) {
1833 /* See if the old one matches what the patch
1834 * applies to.
1835 */
1836 hash_sha1_file(buf->buf, buf->len, blob_type, sha1);
1837 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
1838 return error("the patch applies to '%s' (%s), "
1839 "which does not match the "
1840 "current contents.",
1841 name, sha1_to_hex(sha1));
1842 }
1843 else {
1844 /* Otherwise, the old one must be empty. */
1845 if (buf->len)
1846 return error("the patch applies to an empty "
1847 "'%s' but it is not empty", name);
1848 }
1849
1850 get_sha1_hex(patch->new_sha1_prefix, sha1);
1851 if (is_null_sha1(sha1)) {
1852 strbuf_release(buf);
1853 return 0; /* deletion patch */
1854 }
1855
1856 if (has_sha1_file(sha1)) {
1857 /* We already have the postimage */
1858 enum object_type type;
1859 unsigned long size;
1860 char *result;
1861
1862 result = read_sha1_file(sha1, &type, &size);
1863 if (!result)
1864 return error("the necessary postimage %s for "
1865 "'%s' cannot be read",
1866 patch->new_sha1_prefix, name);
1867 /* XXX read_sha1_file NUL-terminates */
1868 strbuf_attach(buf, result, size, size + 1);
1869 } else {
1870 /* We have verified buf matches the preimage;
1871 * apply the patch data to it, which is stored
1872 * in the patch->fragments->{patch,size}.
1873 */
1874 if (apply_binary_fragment(buf, patch))
1875 return error("binary patch does not apply to '%s'",
1876 name);
1877
1878 /* verify that the result matches */
1879 hash_sha1_file(buf->buf, buf->len, blob_type, sha1);
1880 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
1881 return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)",
1882 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
1883 }
1884
1885 return 0;
1886 }
1887
1888 static int apply_fragments(struct strbuf *buf, struct patch *patch)
1889 {
1890 struct fragment *frag = patch->fragments;
1891 const char *name = patch->old_name ? patch->old_name : patch->new_name;
1892
1893 if (patch->is_binary)
1894 return apply_binary(buf, patch);
1895
1896 while (frag) {
1897 if (apply_one_fragment(buf, frag, patch->inaccurate_eof)) {
1898 error("patch failed: %s:%ld", name, frag->oldpos);
1899 if (!apply_with_reject)
1900 return -1;
1901 frag->rejected = 1;
1902 }
1903 frag = frag->next;
1904 }
1905 return 0;
1906 }
1907
1908 static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
1909 {
1910 if (!ce)
1911 return 0;
1912
1913 if (S_ISGITLINK(ntohl(ce->ce_mode))) {
1914 strbuf_grow(buf, 100);
1915 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(ce->sha1));
1916 } else {
1917 enum object_type type;
1918 unsigned long sz;
1919 char *result;
1920
1921 result = read_sha1_file(ce->sha1, &type, &sz);
1922 if (!result)
1923 return -1;
1924 /* XXX read_sha1_file NUL-terminates */
1925 strbuf_attach(buf, result, sz, sz + 1);
1926 }
1927 return 0;
1928 }
1929
1930 static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
1931 {
1932 struct strbuf buf;
1933
1934 strbuf_init(&buf, 0);
1935 if (cached) {
1936 if (read_file_or_gitlink(ce, &buf))
1937 return error("read of %s failed", patch->old_name);
1938 } else if (patch->old_name) {
1939 if (S_ISGITLINK(patch->old_mode)) {
1940 if (ce) {
1941 read_file_or_gitlink(ce, &buf);
1942 } else {
1943 /*
1944 * There is no way to apply subproject
1945 * patch without looking at the index.
1946 */
1947 patch->fragments = NULL;
1948 }
1949 } else {
1950 if (read_old_data(st, patch->old_name, &buf))
1951 return error("read of %s failed", patch->old_name);
1952 }
1953 }
1954
1955 if (apply_fragments(&buf, patch) < 0)
1956 return -1; /* note with --reject this succeeds. */
1957 patch->result = strbuf_detach(&buf, &patch->resultsize);
1958
1959 if (0 < patch->is_delete && patch->resultsize)
1960 return error("removal patch leaves file contents");
1961
1962 return 0;
1963 }
1964
1965 static int check_to_create_blob(const char *new_name, int ok_if_exists)
1966 {
1967 struct stat nst;
1968 if (!lstat(new_name, &nst)) {
1969 if (S_ISDIR(nst.st_mode) || ok_if_exists)
1970 return 0;
1971 /*
1972 * A leading component of new_name might be a symlink
1973 * that is going to be removed with this patch, but
1974 * still pointing at somewhere that has the path.
1975 * In such a case, path "new_name" does not exist as
1976 * far as git is concerned.
1977 */
1978 if (has_symlink_leading_path(new_name, NULL))
1979 return 0;
1980
1981 return error("%s: already exists in working directory", new_name);
1982 }
1983 else if ((errno != ENOENT) && (errno != ENOTDIR))
1984 return error("%s: %s", new_name, strerror(errno));
1985 return 0;
1986 }
1987
1988 static int verify_index_match(struct cache_entry *ce, struct stat *st)
1989 {
1990 if (S_ISGITLINK(ntohl(ce->ce_mode))) {
1991 if (!S_ISDIR(st->st_mode))
1992 return -1;
1993 return 0;
1994 }
1995 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID);
1996 }
1997
1998 static int check_patch(struct patch *patch, struct patch *prev_patch)
1999 {
2000 struct stat st;
2001 const char *old_name = patch->old_name;
2002 const char *new_name = patch->new_name;
2003 const char *name = old_name ? old_name : new_name;
2004 struct cache_entry *ce = NULL;
2005 int ok_if_exists;
2006
2007 patch->rejected = 1; /* we will drop this after we succeed */
2008
2009 /*
2010 * Make sure that we do not have local modifications from the
2011 * index when we are looking at the index. Also make sure
2012 * we have the preimage file to be patched in the work tree,
2013 * unless --cached, which tells git to apply only in the index.
2014 */
2015 if (old_name) {
2016 int stat_ret = 0;
2017 unsigned st_mode = 0;
2018
2019 if (!cached)
2020 stat_ret = lstat(old_name, &st);
2021 if (check_index) {
2022 int pos = cache_name_pos(old_name, strlen(old_name));
2023 if (pos < 0)
2024 return error("%s: does not exist in index",
2025 old_name);
2026 ce = active_cache[pos];
2027 if (stat_ret < 0) {
2028 struct checkout costate;
2029 if (errno != ENOENT)
2030 return error("%s: %s", old_name,
2031 strerror(errno));
2032 /* checkout */
2033 costate.base_dir = "";
2034 costate.base_dir_len = 0;
2035 costate.force = 0;
2036 costate.quiet = 0;
2037 costate.not_new = 0;
2038 costate.refresh_cache = 1;
2039 if (checkout_entry(ce,
2040 &costate,
2041 NULL) ||
2042 lstat(old_name, &st))
2043 return -1;
2044 }
2045 if (!cached && verify_index_match(ce, &st))
2046 return error("%s: does not match index",
2047 old_name);
2048 if (cached)
2049 st_mode = ntohl(ce->ce_mode);
2050 } else if (stat_ret < 0)
2051 return error("%s: %s", old_name, strerror(errno));
2052
2053 if (!cached)
2054 st_mode = ntohl(ce_mode_from_stat(ce, st.st_mode));
2055
2056 if (patch->is_new < 0)
2057 patch->is_new = 0;
2058 if (!patch->old_mode)
2059 patch->old_mode = st_mode;
2060 if ((st_mode ^ patch->old_mode) & S_IFMT)
2061 return error("%s: wrong type", old_name);
2062 if (st_mode != patch->old_mode)
2063 fprintf(stderr, "warning: %s has type %o, expected %o\n",
2064 old_name, st_mode, patch->old_mode);
2065 }
2066
2067 if (new_name && prev_patch && 0 < prev_patch->is_delete &&
2068 !strcmp(prev_patch->old_name, new_name))
2069 /* A type-change diff is always split into a patch to
2070 * delete old, immediately followed by a patch to
2071 * create new (see diff.c::run_diff()); in such a case
2072 * it is Ok that the entry to be deleted by the
2073 * previous patch is still in the working tree and in
2074 * the index.
2075 */
2076 ok_if_exists = 1;
2077 else
2078 ok_if_exists = 0;
2079
2080 if (new_name &&
2081 ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
2082 if (check_index &&
2083 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
2084 !ok_if_exists)
2085 return error("%s: already exists in index", new_name);
2086 if (!cached) {
2087 int err = check_to_create_blob(new_name, ok_if_exists);
2088 if (err)
2089 return err;
2090 }
2091 if (!patch->new_mode) {
2092 if (0 < patch->is_new)
2093 patch->new_mode = S_IFREG | 0644;
2094 else
2095 patch->new_mode = patch->old_mode;
2096 }
2097 }
2098
2099 if (new_name && old_name) {
2100 int same = !strcmp(old_name, new_name);
2101 if (!patch->new_mode)
2102 patch->new_mode = patch->old_mode;
2103 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
2104 return error("new mode (%o) of %s does not match old mode (%o)%s%s",
2105 patch->new_mode, new_name, patch->old_mode,
2106 same ? "" : " of ", same ? "" : old_name);
2107 }
2108
2109 if (apply_data(patch, &st, ce) < 0)
2110 return error("%s: patch does not apply", name);
2111 patch->rejected = 0;
2112 return 0;
2113 }
2114
2115 static int check_patch_list(struct patch *patch)
2116 {
2117 struct patch *prev_patch = NULL;
2118 int err = 0;
2119
2120 for (prev_patch = NULL; patch ; patch = patch->next) {
2121 if (apply_verbosely)
2122 say_patch_name(stderr,
2123 "Checking patch ", patch, "...\n");
2124 err |= check_patch(patch, prev_patch);
2125 prev_patch = patch;
2126 }
2127 return err;
2128 }
2129
2130 /* This function tries to read the sha1 from the current index */
2131 static int get_current_sha1(const char *path, unsigned char *sha1)
2132 {
2133 int pos;
2134
2135 if (read_cache() < 0)
2136 return -1;
2137 pos = cache_name_pos(path, strlen(path));
2138 if (pos < 0)
2139 return -1;
2140 hashcpy(sha1, active_cache[pos]->sha1);
2141 return 0;
2142 }
2143
2144 /* Build an index that contains the just the files needed for a 3way merge */
2145 static void build_fake_ancestor(struct patch *list, const char *filename)
2146 {
2147 struct patch *patch;
2148 struct index_state result = { 0 };
2149 int fd;
2150
2151 /* Once we start supporting the reverse patch, it may be
2152 * worth showing the new sha1 prefix, but until then...
2153 */
2154 for (patch = list; patch; patch = patch->next) {
2155 const unsigned char *sha1_ptr;
2156 unsigned char sha1[20];
2157 struct cache_entry *ce;
2158 const char *name;
2159
2160 name = patch->old_name ? patch->old_name : patch->new_name;
2161 if (0 < patch->is_new)
2162 continue;
2163 else if (get_sha1(patch->old_sha1_prefix, sha1))
2164 /* git diff has no index line for mode/type changes */
2165 if (!patch->lines_added && !patch->lines_deleted) {
2166 if (get_current_sha1(patch->new_name, sha1) ||
2167 get_current_sha1(patch->old_name, sha1))
2168 die("mode change for %s, which is not "
2169 "in current HEAD", name);
2170 sha1_ptr = sha1;
2171 } else
2172 die("sha1 information is lacking or useless "
2173 "(%s).", name);
2174 else
2175 sha1_ptr = sha1;
2176
2177 ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);
2178 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
2179 die ("Could not add %s to temporary index", name);
2180 }
2181
2182 fd = open(filename, O_WRONLY | O_CREAT, 0666);
2183 if (fd < 0 || write_index(&result, fd) || close(fd))
2184 die ("Could not write temporary index to %s", filename);
2185
2186 discard_index(&result);
2187 }
2188
2189 static void stat_patch_list(struct patch *patch)
2190 {
2191 int files, adds, dels;
2192
2193 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
2194 files++;
2195 adds += patch->lines_added;
2196 dels += patch->lines_deleted;
2197 show_stats(patch);
2198 }
2199
2200 printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
2201 }
2202
2203 static void numstat_patch_list(struct patch *patch)
2204 {
2205 for ( ; patch; patch = patch->next) {
2206 const char *name;
2207 name = patch->new_name ? patch->new_name : patch->old_name;
2208 if (patch->is_binary)
2209 printf("-\t-\t");
2210 else
2211 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
2212 write_name_quoted(name, stdout, line_termination);
2213 }
2214 }
2215
2216 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
2217 {
2218 if (mode)
2219 printf(" %s mode %06o %s\n", newdelete, mode, name);
2220 else
2221 printf(" %s %s\n", newdelete, name);
2222 }
2223
2224 static void show_mode_change(struct patch *p, int show_name)
2225 {
2226 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
2227 if (show_name)
2228 printf(" mode change %06o => %06o %s\n",
2229 p->old_mode, p->new_mode, p->new_name);
2230 else
2231 printf(" mode change %06o => %06o\n",
2232 p->old_mode, p->new_mode);
2233 }
2234 }
2235
2236 static void show_rename_copy(struct patch *p)
2237 {
2238 const char *renamecopy = p->is_rename ? "rename" : "copy";
2239 const char *old, *new;
2240
2241 /* Find common prefix */
2242 old = p->old_name;
2243 new = p->new_name;
2244 while (1) {
2245 const char *slash_old, *slash_new;
2246 slash_old = strchr(old, '/');
2247 slash_new = strchr(new, '/');
2248 if (!slash_old ||
2249 !slash_new ||
2250 slash_old - old != slash_new - new ||
2251 memcmp(old, new, slash_new - new))
2252 break;
2253 old = slash_old + 1;
2254 new = slash_new + 1;
2255 }
2256 /* p->old_name thru old is the common prefix, and old and new
2257 * through the end of names are renames
2258 */
2259 if (old != p->old_name)
2260 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
2261 (int)(old - p->old_name), p->old_name,
2262 old, new, p->score);
2263 else
2264 printf(" %s %s => %s (%d%%)\n", renamecopy,
2265 p->old_name, p->new_name, p->score);
2266 show_mode_change(p, 0);
2267 }
2268
2269 static void summary_patch_list(struct patch *patch)
2270 {
2271 struct patch *p;
2272
2273 for (p = patch; p; p = p->next) {
2274 if (p->is_new)
2275 show_file_mode_name("create", p->new_mode, p->new_name);
2276 else if (p->is_delete)
2277 show_file_mode_name("delete", p->old_mode, p->old_name);
2278 else {
2279 if (p->is_rename || p->is_copy)
2280 show_rename_copy(p);
2281 else {
2282 if (p->score) {
2283 printf(" rewrite %s (%d%%)\n",
2284 p->new_name, p->score);
2285 show_mode_change(p, 0);
2286 }
2287 else
2288 show_mode_change(p, 1);
2289 }
2290 }
2291 }
2292 }
2293
2294 static void patch_stats(struct patch *patch)
2295 {
2296 int lines = patch->lines_added + patch->lines_deleted;
2297
2298 if (lines > max_change)
2299 max_change = lines;
2300 if (patch->old_name) {
2301 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
2302 if (!len)
2303 len = strlen(patch->old_name);
2304 if (len > max_len)
2305 max_len = len;
2306 }
2307 if (patch->new_name) {
2308 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
2309 if (!len)
2310 len = strlen(patch->new_name);
2311 if (len > max_len)
2312 max_len = len;
2313 }
2314 }
2315
2316 static void remove_file(struct patch *patch, int rmdir_empty)
2317 {
2318 if (update_index) {
2319 if (remove_file_from_cache(patch->old_name) < 0)
2320 die("unable to remove %s from index", patch->old_name);
2321 }
2322 if (!cached) {
2323 if (S_ISGITLINK(patch->old_mode)) {
2324 if (rmdir(patch->old_name))
2325 warning("unable to remove submodule %s",
2326 patch->old_name);
2327 } else if (!unlink(patch->old_name) && rmdir_empty) {
2328 char *name = xstrdup(patch->old_name);
2329 char *end = strrchr(name, '/');
2330 while (end) {
2331 *end = 0;
2332 if (rmdir(name))
2333 break;
2334 end = strrchr(name, '/');
2335 }
2336 free(name);
2337 }
2338 }
2339 }
2340
2341 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
2342 {
2343 struct stat st;
2344 struct cache_entry *ce;
2345 int namelen = strlen(path);
2346 unsigned ce_size = cache_entry_size(namelen);
2347
2348 if (!update_index)
2349 return;
2350
2351 ce = xcalloc(1, ce_size);
2352 memcpy(ce->name, path, namelen);
2353 ce->ce_mode = create_ce_mode(mode);
2354 ce->ce_flags = htons(namelen);
2355 if (S_ISGITLINK(mode)) {
2356 const char *s = buf;
2357
2358 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
2359 die("corrupt patch for subproject %s", path);
2360 } else {
2361 if (!cached) {
2362 if (lstat(path, &st) < 0)
2363 die("unable to stat newly created file %s",
2364 path);
2365 fill_stat_cache_info(ce, &st);
2366 }
2367 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
2368 die("unable to create backing store for newly created file %s", path);
2369 }
2370 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
2371 die("unable to add cache entry for %s", path);
2372 }
2373
2374 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
2375 {
2376 int fd;
2377 struct strbuf nbuf;
2378
2379 if (S_ISGITLINK(mode)) {
2380 struct stat st;
2381 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
2382 return 0;
2383 return mkdir(path, 0777);
2384 }
2385
2386 if (has_symlinks && S_ISLNK(mode))
2387 /* Although buf:size is counted string, it also is NUL
2388 * terminated.
2389 */
2390 return symlink(buf, path);
2391
2392 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
2393 if (fd < 0)
2394 return -1;
2395
2396 strbuf_init(&nbuf, 0);
2397 if (convert_to_working_tree(path, buf, size, &nbuf)) {
2398 size = nbuf.len;
2399 buf = nbuf.buf;
2400 }
2401 write_or_die(fd, buf, size);
2402 strbuf_release(&nbuf);
2403
2404 if (close(fd) < 0)
2405 die("closing file %s: %s", path, strerror(errno));
2406 return 0;
2407 }
2408
2409 /*
2410 * We optimistically assume that the directories exist,
2411 * which is true 99% of the time anyway. If they don't,
2412 * we create them and try again.
2413 */
2414 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
2415 {
2416 if (cached)
2417 return;
2418 if (!try_create_file(path, mode, buf, size))
2419 return;
2420
2421 if (errno == ENOENT) {
2422 if (safe_create_leading_directories(path))
2423 return;
2424 if (!try_create_file(path, mode, buf, size))
2425 return;
2426 }
2427
2428 if (errno == EEXIST || errno == EACCES) {
2429 /* We may be trying to create a file where a directory
2430 * used to be.
2431 */
2432 struct stat st;
2433 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
2434 errno = EEXIST;
2435 }
2436
2437 if (errno == EEXIST) {
2438 unsigned int nr = getpid();
2439
2440 for (;;) {
2441 const char *newpath;
2442 newpath = mkpath("%s~%u", path, nr);
2443 if (!try_create_file(newpath, mode, buf, size)) {
2444 if (!rename(newpath, path))
2445 return;
2446 unlink(newpath);
2447 break;
2448 }
2449 if (errno != EEXIST)
2450 break;
2451 ++nr;
2452 }
2453 }
2454 die("unable to write file %s mode %o", path, mode);
2455 }
2456
2457 static void create_file(struct patch *patch)
2458 {
2459 char *path = patch->new_name;
2460 unsigned mode = patch->new_mode;
2461 unsigned long size = patch->resultsize;
2462 char *buf = patch->result;
2463
2464 if (!mode)
2465 mode = S_IFREG | 0644;
2466 create_one_file(path, mode, buf, size);
2467 add_index_file(path, mode, buf, size);
2468 }
2469
2470 /* phase zero is to remove, phase one is to create */
2471 static void write_out_one_result(struct patch *patch, int phase)
2472 {
2473 if (patch->is_delete > 0) {
2474 if (phase == 0)
2475 remove_file(patch, 1);
2476 return;
2477 }
2478 if (patch->is_new > 0 || patch->is_copy) {
2479 if (phase == 1)
2480 create_file(patch);
2481 return;
2482 }
2483 /*
2484 * Rename or modification boils down to the same
2485 * thing: remove the old, write the new
2486 */
2487 if (phase == 0)
2488 remove_file(patch, patch->is_rename);
2489 if (phase == 1)
2490 create_file(patch);
2491 }
2492
2493 static int write_out_one_reject(struct patch *patch)
2494 {
2495 FILE *rej;
2496 char namebuf[PATH_MAX];
2497 struct fragment *frag;
2498 int cnt = 0;
2499
2500 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
2501 if (!frag->rejected)
2502 continue;
2503 cnt++;
2504 }
2505
2506 if (!cnt) {
2507 if (apply_verbosely)
2508 say_patch_name(stderr,
2509 "Applied patch ", patch, " cleanly.\n");
2510 return 0;
2511 }
2512
2513 /* This should not happen, because a removal patch that leaves
2514 * contents are marked "rejected" at the patch level.
2515 */
2516 if (!patch->new_name)
2517 die("internal error");
2518
2519 /* Say this even without --verbose */
2520 say_patch_name(stderr, "Applying patch ", patch, " with");
2521 fprintf(stderr, " %d rejects...\n", cnt);
2522
2523 cnt = strlen(patch->new_name);
2524 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
2525 cnt = ARRAY_SIZE(namebuf) - 5;
2526 fprintf(stderr,
2527 "warning: truncating .rej filename to %.*s.rej",
2528 cnt - 1, patch->new_name);
2529 }
2530 memcpy(namebuf, patch->new_name, cnt);
2531 memcpy(namebuf + cnt, ".rej", 5);
2532
2533 rej = fopen(namebuf, "w");
2534 if (!rej)
2535 return error("cannot open %s: %s", namebuf, strerror(errno));
2536
2537 /* Normal git tools never deal with .rej, so do not pretend
2538 * this is a git patch by saying --git nor give extended
2539 * headers. While at it, maybe please "kompare" that wants
2540 * the trailing TAB and some garbage at the end of line ;-).
2541 */
2542 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
2543 patch->new_name, patch->new_name);
2544 for (cnt = 1, frag = patch->fragments;
2545 frag;
2546 cnt++, frag = frag->next) {
2547 if (!frag->rejected) {
2548 fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);
2549 continue;
2550 }
2551 fprintf(stderr, "Rejected hunk #%d.\n", cnt);
2552 fprintf(rej, "%.*s", frag->size, frag->patch);
2553 if (frag->patch[frag->size-1] != '\n')
2554 fputc('\n', rej);
2555 }
2556 fclose(rej);
2557 return -1;
2558 }
2559
2560 static int write_out_results(struct patch *list, int skipped_patch)
2561 {
2562 int phase;
2563 int errs = 0;
2564 struct patch *l;
2565
2566 if (!list && !skipped_patch)
2567 return error("No changes");
2568
2569 for (phase = 0; phase < 2; phase++) {
2570 l = list;
2571 while (l) {
2572 if (l->rejected)
2573 errs = 1;
2574 else {
2575 write_out_one_result(l, phase);
2576 if (phase == 1 && write_out_one_reject(l))
2577 errs = 1;
2578 }
2579 l = l->next;
2580 }
2581 }
2582 return errs;
2583 }
2584
2585 static struct lock_file lock_file;
2586
2587 static struct excludes {
2588 struct excludes *next;
2589 const char *path;
2590 } *excludes;
2591
2592 static int use_patch(struct patch *p)
2593 {
2594 const char *pathname = p->new_name ? p->new_name : p->old_name;
2595 struct excludes *x = excludes;
2596 while (x) {
2597 if (fnmatch(x->path, pathname, 0) == 0)
2598 return 0;
2599 x = x->next;
2600 }
2601 if (0 < prefix_length) {
2602 int pathlen = strlen(pathname);
2603 if (pathlen <= prefix_length ||
2604 memcmp(prefix, pathname, prefix_length))
2605 return 0;
2606 }
2607 return 1;
2608 }
2609
2610 static void prefix_one(char **name)
2611 {
2612 char *old_name = *name;
2613 if (!old_name)
2614 return;
2615 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
2616 free(old_name);
2617 }
2618
2619 static void prefix_patches(struct patch *p)
2620 {
2621 if (!prefix || p->is_toplevel_relative)
2622 return;
2623 for ( ; p; p = p->next) {
2624 if (p->new_name == p->old_name) {
2625 char *prefixed = p->new_name;
2626 prefix_one(&prefixed);
2627 p->new_name = p->old_name = prefixed;
2628 }
2629 else {
2630 prefix_one(&p->new_name);
2631 prefix_one(&p->old_name);
2632 }
2633 }
2634 }
2635
2636 static int apply_patch(int fd, const char *filename, int inaccurate_eof)
2637 {
2638 size_t offset;
2639 struct strbuf buf;
2640 struct patch *list = NULL, **listp = &list;
2641 int skipped_patch = 0;
2642
2643 strbuf_init(&buf, 0);
2644 patch_input_file = filename;
2645 read_patch_file(&buf, fd);
2646 offset = 0;
2647 while (offset < buf.len) {
2648 struct patch *patch;
2649 int nr;
2650
2651 patch = xcalloc(1, sizeof(*patch));
2652 patch->inaccurate_eof = inaccurate_eof;
2653 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
2654 if (nr < 0)
2655 break;
2656 if (apply_in_reverse)
2657 reverse_patches(patch);
2658 if (prefix)
2659 prefix_patches(patch);
2660 if (use_patch(patch)) {
2661 patch_stats(patch);
2662 *listp = patch;
2663 listp = &patch->next;
2664 }
2665 else {
2666 /* perhaps free it a bit better? */
2667 free(patch);
2668 skipped_patch++;
2669 }
2670 offset += nr;
2671 }
2672
2673 if (whitespace_error && (new_whitespace == error_on_whitespace))
2674 apply = 0;
2675
2676 update_index = check_index && apply;
2677 if (update_index && newfd < 0)
2678 newfd = hold_locked_index(&lock_file, 1);
2679
2680 if (check_index) {
2681 if (read_cache() < 0)
2682 die("unable to read index file");
2683 }
2684
2685 if ((check || apply) &&
2686 check_patch_list(list) < 0 &&
2687 !apply_with_reject)
2688 exit(1);
2689
2690 if (apply && write_out_results(list, skipped_patch))
2691 exit(1);
2692
2693 if (fake_ancestor)
2694 build_fake_ancestor(list, fake_ancestor);
2695
2696 if (diffstat)
2697 stat_patch_list(list);
2698
2699 if (numstat)
2700 numstat_patch_list(list);
2701
2702 if (summary)
2703 summary_patch_list(list);
2704
2705 strbuf_release(&buf);
2706 return 0;
2707 }
2708
2709 static int git_apply_config(const char *var, const char *value)
2710 {
2711 if (!strcmp(var, "apply.whitespace")) {
2712 apply_default_whitespace = xstrdup(value);
2713 return 0;
2714 }
2715 return git_default_config(var, value);
2716 }
2717
2718
2719 int cmd_apply(int argc, const char **argv, const char *unused_prefix)
2720 {
2721 int i;
2722 int read_stdin = 1;
2723 int inaccurate_eof = 0;
2724 int errs = 0;
2725 int is_not_gitdir = 0;
2726
2727 const char *whitespace_option = NULL;
2728
2729 prefix = setup_git_directory_gently(&is_not_gitdir);
2730 prefix_length = prefix ? strlen(prefix) : 0;
2731 git_config(git_apply_config);
2732 if (apply_default_whitespace)
2733 parse_whitespace_option(apply_default_whitespace);
2734
2735 for (i = 1; i < argc; i++) {
2736 const char *arg = argv[i];
2737 char *end;
2738 int fd;
2739
2740 if (!strcmp(arg, "-")) {
2741 errs |= apply_patch(0, "<stdin>", inaccurate_eof);
2742 read_stdin = 0;
2743 continue;
2744 }
2745 if (!prefixcmp(arg, "--exclude=")) {
2746 struct excludes *x = xmalloc(sizeof(*x));
2747 x->path = arg + 10;
2748 x->next = excludes;
2749 excludes = x;
2750 continue;
2751 }
2752 if (!prefixcmp(arg, "-p")) {
2753 p_value = atoi(arg + 2);
2754 p_value_known = 1;
2755 continue;
2756 }
2757 if (!strcmp(arg, "--no-add")) {
2758 no_add = 1;
2759 continue;
2760 }
2761 if (!strcmp(arg, "--stat")) {
2762 apply = 0;
2763 diffstat = 1;
2764 continue;
2765 }
2766 if (!strcmp(arg, "--allow-binary-replacement") ||
2767 !strcmp(arg, "--binary")) {
2768 continue; /* now no-op */
2769 }
2770 if (!strcmp(arg, "--numstat")) {
2771 apply = 0;
2772 numstat = 1;
2773 continue;
2774 }
2775 if (!strcmp(arg, "--summary")) {
2776 apply = 0;
2777 summary = 1;
2778 continue;
2779 }
2780 if (!strcmp(arg, "--check")) {
2781 apply = 0;
2782 check = 1;
2783 continue;
2784 }
2785 if (!strcmp(arg, "--index")) {
2786 if (is_not_gitdir)
2787 die("--index outside a repository");
2788 check_index = 1;
2789 continue;
2790 }
2791 if (!strcmp(arg, "--cached")) {
2792 if (is_not_gitdir)
2793 die("--cached outside a repository");
2794 check_index = 1;
2795 cached = 1;
2796 continue;
2797 }
2798 if (!strcmp(arg, "--apply")) {
2799 apply = 1;
2800 continue;
2801 }
2802 if (!strcmp(arg, "--build-fake-ancestor")) {
2803 apply = 0;
2804 if (++i >= argc)
2805 die ("need a filename");
2806 fake_ancestor = argv[i];
2807 continue;
2808 }
2809 if (!strcmp(arg, "-z")) {
2810 line_termination = 0;
2811 continue;
2812 }
2813 if (!prefixcmp(arg, "-C")) {
2814 p_context = strtoul(arg + 2, &end, 0);
2815 if (*end != '\0')
2816 die("unrecognized context count '%s'", arg + 2);
2817 continue;
2818 }
2819 if (!prefixcmp(arg, "--whitespace=")) {
2820 whitespace_option = arg + 13;
2821 parse_whitespace_option(arg + 13);
2822 continue;
2823 }
2824 if (!strcmp(arg, "-R") || !strcmp(arg, "--reverse")) {
2825 apply_in_reverse = 1;
2826 continue;
2827 }
2828 if (!strcmp(arg, "--unidiff-zero")) {
2829 unidiff_zero = 1;
2830 continue;
2831 }
2832 if (!strcmp(arg, "--reject")) {
2833 apply = apply_with_reject = apply_verbosely = 1;
2834 continue;
2835 }
2836 if (!strcmp(arg, "-v") || !strcmp(arg, "--verbose")) {
2837 apply_verbosely = 1;
2838 continue;
2839 }
2840 if (!strcmp(arg, "--inaccurate-eof")) {
2841 inaccurate_eof = 1;
2842 continue;
2843 }
2844 if (0 < prefix_length)
2845 arg = prefix_filename(prefix, prefix_length, arg);
2846
2847 fd = open(arg, O_RDONLY);
2848 if (fd < 0)
2849 usage(apply_usage);
2850 read_stdin = 0;
2851 set_default_whitespace_mode(whitespace_option);
2852 errs |= apply_patch(fd, arg, inaccurate_eof);
2853 close(fd);
2854 }
2855 set_default_whitespace_mode(whitespace_option);
2856 if (read_stdin)
2857 errs |= apply_patch(0, "<stdin>", inaccurate_eof);
2858 if (whitespace_error) {
2859 if (squelch_whitespace_errors &&
2860 squelch_whitespace_errors < whitespace_error) {
2861 int squelched =
2862 whitespace_error - squelch_whitespace_errors;
2863 fprintf(stderr, "warning: squelched %d "
2864 "whitespace error%s\n",
2865 squelched,
2866 squelched == 1 ? "" : "s");
2867 }
2868 if (new_whitespace == error_on_whitespace)
2869 die("%d line%s add%s whitespace errors.",
2870 whitespace_error,
2871 whitespace_error == 1 ? "" : "s",
2872 whitespace_error == 1 ? "s" : "");
2873 if (applied_after_fixing_ws)
2874 fprintf(stderr, "warning: %d line%s applied after"
2875 " fixing whitespace errors.\n",
2876 applied_after_fixing_ws,
2877 applied_after_fixing_ws == 1 ? "" : "s");
2878 else if (whitespace_error)
2879 fprintf(stderr, "warning: %d line%s add%s whitespace errors.\n",
2880 whitespace_error,
2881 whitespace_error == 1 ? "" : "s",
2882 whitespace_error == 1 ? "s" : "");
2883 }
2884
2885 if (update_index) {
2886 if (write_cache(newfd, active_cache, active_nr) ||
2887 close(newfd) || commit_locked_index(&lock_file))
2888 die("Unable to write new index file");
2889 }
2890
2891 return !!errs;
2892 }