Merge branch 'jc/maint-add-sync-stat'
[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 return NULL;
687 }
688
689 /* Verify that we recognize the lines following a git header */
690 static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
691 {
692 unsigned long offset;
693
694 /* A git diff has explicit new/delete information, so we don't guess */
695 patch->is_new = 0;
696 patch->is_delete = 0;
697
698 /*
699 * Some things may not have the old name in the
700 * rest of the headers anywhere (pure mode changes,
701 * or removing or adding empty files), so we get
702 * the default name from the header.
703 */
704 patch->def_name = git_header_name(line, len);
705
706 line += len;
707 size -= len;
708 linenr++;
709 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
710 static const struct opentry {
711 const char *str;
712 int (*fn)(const char *, struct patch *);
713 } optable[] = {
714 { "@@ -", gitdiff_hdrend },
715 { "--- ", gitdiff_oldname },
716 { "+++ ", gitdiff_newname },
717 { "old mode ", gitdiff_oldmode },
718 { "new mode ", gitdiff_newmode },
719 { "deleted file mode ", gitdiff_delete },
720 { "new file mode ", gitdiff_newfile },
721 { "copy from ", gitdiff_copysrc },
722 { "copy to ", gitdiff_copydst },
723 { "rename old ", gitdiff_renamesrc },
724 { "rename new ", gitdiff_renamedst },
725 { "rename from ", gitdiff_renamesrc },
726 { "rename to ", gitdiff_renamedst },
727 { "similarity index ", gitdiff_similarity },
728 { "dissimilarity index ", gitdiff_dissimilarity },
729 { "index ", gitdiff_index },
730 { "", gitdiff_unrecognized },
731 };
732 int i;
733
734 len = linelen(line, size);
735 if (!len || line[len-1] != '\n')
736 break;
737 for (i = 0; i < ARRAY_SIZE(optable); i++) {
738 const struct opentry *p = optable + i;
739 int oplen = strlen(p->str);
740 if (len < oplen || memcmp(p->str, line, oplen))
741 continue;
742 if (p->fn(line + oplen, patch) < 0)
743 return offset;
744 break;
745 }
746 }
747
748 return offset;
749 }
750
751 static int parse_num(const char *line, unsigned long *p)
752 {
753 char *ptr;
754
755 if (!isdigit(*line))
756 return 0;
757 *p = strtoul(line, &ptr, 10);
758 return ptr - line;
759 }
760
761 static int parse_range(const char *line, int len, int offset, const char *expect,
762 unsigned long *p1, unsigned long *p2)
763 {
764 int digits, ex;
765
766 if (offset < 0 || offset >= len)
767 return -1;
768 line += offset;
769 len -= offset;
770
771 digits = parse_num(line, p1);
772 if (!digits)
773 return -1;
774
775 offset += digits;
776 line += digits;
777 len -= digits;
778
779 *p2 = 1;
780 if (*line == ',') {
781 digits = parse_num(line+1, p2);
782 if (!digits)
783 return -1;
784
785 offset += digits+1;
786 line += digits+1;
787 len -= digits+1;
788 }
789
790 ex = strlen(expect);
791 if (ex > len)
792 return -1;
793 if (memcmp(line, expect, ex))
794 return -1;
795
796 return offset + ex;
797 }
798
799 /*
800 * Parse a unified diff fragment header of the
801 * form "@@ -a,b +c,d @@"
802 */
803 static int parse_fragment_header(char *line, int len, struct fragment *fragment)
804 {
805 int offset;
806
807 if (!len || line[len-1] != '\n')
808 return -1;
809
810 /* Figure out the number of lines in a fragment */
811 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
812 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
813
814 return offset;
815 }
816
817 static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
818 {
819 unsigned long offset, len;
820
821 patch->is_toplevel_relative = 0;
822 patch->is_rename = patch->is_copy = 0;
823 patch->is_new = patch->is_delete = -1;
824 patch->old_mode = patch->new_mode = 0;
825 patch->old_name = patch->new_name = NULL;
826 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
827 unsigned long nextlen;
828
829 len = linelen(line, size);
830 if (!len)
831 break;
832
833 /* Testing this early allows us to take a few shortcuts.. */
834 if (len < 6)
835 continue;
836
837 /*
838 * Make sure we don't find any unconnected patch fragments.
839 * That's a sign that we didn't find a header, and that a
840 * patch has become corrupted/broken up.
841 */
842 if (!memcmp("@@ -", line, 4)) {
843 struct fragment dummy;
844 if (parse_fragment_header(line, len, &dummy) < 0)
845 continue;
846 die("patch fragment without header at line %d: %.*s",
847 linenr, (int)len-1, line);
848 }
849
850 if (size < len + 6)
851 break;
852
853 /*
854 * Git patch? It might not have a real patch, just a rename
855 * or mode change, so we handle that specially
856 */
857 if (!memcmp("diff --git ", line, 11)) {
858 int git_hdr_len = parse_git_header(line, len, size, patch);
859 if (git_hdr_len <= len)
860 continue;
861 if (!patch->old_name && !patch->new_name) {
862 if (!patch->def_name)
863 die("git diff header lacks filename information (line %d)", linenr);
864 patch->old_name = patch->new_name = patch->def_name;
865 }
866 patch->is_toplevel_relative = 1;
867 *hdrsize = git_hdr_len;
868 return offset;
869 }
870
871 /** --- followed by +++ ? */
872 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
873 continue;
874
875 /*
876 * We only accept unified patches, so we want it to
877 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
878 * minimum
879 */
880 nextlen = linelen(line + len, size - len);
881 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
882 continue;
883
884 /* Ok, we'll consider it a patch */
885 parse_traditional_patch(line, line+len, patch);
886 *hdrsize = len + nextlen;
887 linenr += 2;
888 return offset;
889 }
890 return -1;
891 }
892
893 static void check_whitespace(const char *line, int len)
894 {
895 const char *err = "Adds trailing whitespace";
896 int seen_space = 0;
897 int i;
898
899 /*
900 * We know len is at least two, since we have a '+' and we
901 * checked that the last character was a '\n' before calling
902 * this function. That is, an addition of an empty line would
903 * check the '+' here. Sneaky...
904 */
905 if (isspace(line[len-2]))
906 goto error;
907
908 /*
909 * Make sure that there is no space followed by a tab in
910 * indentation.
911 */
912 err = "Space in indent is followed by a tab";
913 for (i = 1; i < len; i++) {
914 if (line[i] == '\t') {
915 if (seen_space)
916 goto error;
917 }
918 else if (line[i] == ' ')
919 seen_space = 1;
920 else
921 break;
922 }
923 return;
924
925 error:
926 whitespace_error++;
927 if (squelch_whitespace_errors &&
928 squelch_whitespace_errors < whitespace_error)
929 ;
930 else
931 fprintf(stderr, "%s.\n%s:%d:%.*s\n",
932 err, patch_input_file, linenr, len-2, line+1);
933 }
934
935
936 /*
937 * Parse a unified diff. Note that this really needs to parse each
938 * fragment separately, since the only way to know the difference
939 * between a "---" that is part of a patch, and a "---" that starts
940 * the next patch is to look at the line counts..
941 */
942 static int parse_fragment(char *line, unsigned long size, struct patch *patch, struct fragment *fragment)
943 {
944 int added, deleted;
945 int len = linelen(line, size), offset;
946 unsigned long oldlines, newlines;
947 unsigned long leading, trailing;
948
949 offset = parse_fragment_header(line, len, fragment);
950 if (offset < 0)
951 return -1;
952 oldlines = fragment->oldlines;
953 newlines = fragment->newlines;
954 leading = 0;
955 trailing = 0;
956
957 /* Parse the thing.. */
958 line += len;
959 size -= len;
960 linenr++;
961 added = deleted = 0;
962 for (offset = len;
963 0 < size;
964 offset += len, size -= len, line += len, linenr++) {
965 if (!oldlines && !newlines)
966 break;
967 len = linelen(line, size);
968 if (!len || line[len-1] != '\n')
969 return -1;
970 switch (*line) {
971 default:
972 return -1;
973 case '\n': /* newer GNU diff, an empty context line */
974 case ' ':
975 oldlines--;
976 newlines--;
977 if (!deleted && !added)
978 leading++;
979 trailing++;
980 break;
981 case '-':
982 if (apply_in_reverse &&
983 new_whitespace != nowarn_whitespace)
984 check_whitespace(line, len);
985 deleted++;
986 oldlines--;
987 trailing = 0;
988 break;
989 case '+':
990 if (!apply_in_reverse &&
991 new_whitespace != nowarn_whitespace)
992 check_whitespace(line, len);
993 added++;
994 newlines--;
995 trailing = 0;
996 break;
997
998 /* We allow "\ No newline at end of file". Depending
999 * on locale settings when the patch was produced we
1000 * don't know what this line looks like. The only
1001 * thing we do know is that it begins with "\ ".
1002 * Checking for 12 is just for sanity check -- any
1003 * l10n of "\ No newline..." is at least that long.
1004 */
1005 case '\\':
1006 if (len < 12 || memcmp(line, "\\ ", 2))
1007 return -1;
1008 break;
1009 }
1010 }
1011 if (oldlines || newlines)
1012 return -1;
1013 fragment->leading = leading;
1014 fragment->trailing = trailing;
1015
1016 /* If a fragment ends with an incomplete line, we failed to include
1017 * it in the above loop because we hit oldlines == newlines == 0
1018 * before seeing it.
1019 */
1020 if (12 < size && !memcmp(line, "\\ ", 2))
1021 offset += linelen(line, size);
1022
1023 patch->lines_added += added;
1024 patch->lines_deleted += deleted;
1025
1026 if (0 < patch->is_new && oldlines)
1027 return error("new file depends on old contents");
1028 if (0 < patch->is_delete && newlines)
1029 return error("deleted file still has contents");
1030 return offset;
1031 }
1032
1033 static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
1034 {
1035 unsigned long offset = 0;
1036 unsigned long oldlines = 0, newlines = 0, context = 0;
1037 struct fragment **fragp = &patch->fragments;
1038
1039 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1040 struct fragment *fragment;
1041 int len;
1042
1043 fragment = xcalloc(1, sizeof(*fragment));
1044 len = parse_fragment(line, size, patch, fragment);
1045 if (len <= 0)
1046 die("corrupt patch at line %d", linenr);
1047 fragment->patch = line;
1048 fragment->size = len;
1049 oldlines += fragment->oldlines;
1050 newlines += fragment->newlines;
1051 context += fragment->leading + fragment->trailing;
1052
1053 *fragp = fragment;
1054 fragp = &fragment->next;
1055
1056 offset += len;
1057 line += len;
1058 size -= len;
1059 }
1060
1061 /*
1062 * If something was removed (i.e. we have old-lines) it cannot
1063 * be creation, and if something was added it cannot be
1064 * deletion. However, the reverse is not true; --unified=0
1065 * patches that only add are not necessarily creation even
1066 * though they do not have any old lines, and ones that only
1067 * delete are not necessarily deletion.
1068 *
1069 * Unfortunately, a real creation/deletion patch do _not_ have
1070 * any context line by definition, so we cannot safely tell it
1071 * apart with --unified=0 insanity. At least if the patch has
1072 * more than one hunk it is not creation or deletion.
1073 */
1074 if (patch->is_new < 0 &&
1075 (oldlines || (patch->fragments && patch->fragments->next)))
1076 patch->is_new = 0;
1077 if (patch->is_delete < 0 &&
1078 (newlines || (patch->fragments && patch->fragments->next)))
1079 patch->is_delete = 0;
1080 if (!unidiff_zero || context) {
1081 /* If the user says the patch is not generated with
1082 * --unified=0, or if we have seen context lines,
1083 * then not having oldlines means the patch is creation,
1084 * and not having newlines means the patch is deletion.
1085 */
1086 if (patch->is_new < 0 && !oldlines) {
1087 patch->is_new = 1;
1088 patch->old_name = NULL;
1089 }
1090 if (patch->is_delete < 0 && !newlines) {
1091 patch->is_delete = 1;
1092 patch->new_name = NULL;
1093 }
1094 }
1095
1096 if (0 < patch->is_new && oldlines)
1097 die("new file %s depends on old contents", patch->new_name);
1098 if (0 < patch->is_delete && newlines)
1099 die("deleted file %s still has contents", patch->old_name);
1100 if (!patch->is_delete && !newlines && context)
1101 fprintf(stderr, "** warning: file %s becomes empty but "
1102 "is not deleted\n", patch->new_name);
1103
1104 return offset;
1105 }
1106
1107 static inline int metadata_changes(struct patch *patch)
1108 {
1109 return patch->is_rename > 0 ||
1110 patch->is_copy > 0 ||
1111 patch->is_new > 0 ||
1112 patch->is_delete ||
1113 (patch->old_mode && patch->new_mode &&
1114 patch->old_mode != patch->new_mode);
1115 }
1116
1117 static char *inflate_it(const void *data, unsigned long size,
1118 unsigned long inflated_size)
1119 {
1120 z_stream stream;
1121 void *out;
1122 int st;
1123
1124 memset(&stream, 0, sizeof(stream));
1125
1126 stream.next_in = (unsigned char *)data;
1127 stream.avail_in = size;
1128 stream.next_out = out = xmalloc(inflated_size);
1129 stream.avail_out = inflated_size;
1130 inflateInit(&stream);
1131 st = inflate(&stream, Z_FINISH);
1132 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1133 free(out);
1134 return NULL;
1135 }
1136 return out;
1137 }
1138
1139 static struct fragment *parse_binary_hunk(char **buf_p,
1140 unsigned long *sz_p,
1141 int *status_p,
1142 int *used_p)
1143 {
1144 /* Expect a line that begins with binary patch method ("literal"
1145 * or "delta"), followed by the length of data before deflating.
1146 * a sequence of 'length-byte' followed by base-85 encoded data
1147 * should follow, terminated by a newline.
1148 *
1149 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1150 * and we would limit the patch line to 66 characters,
1151 * so one line can fit up to 13 groups that would decode
1152 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1153 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1154 */
1155 int llen, used;
1156 unsigned long size = *sz_p;
1157 char *buffer = *buf_p;
1158 int patch_method;
1159 unsigned long origlen;
1160 char *data = NULL;
1161 int hunk_size = 0;
1162 struct fragment *frag;
1163
1164 llen = linelen(buffer, size);
1165 used = llen;
1166
1167 *status_p = 0;
1168
1169 if (!prefixcmp(buffer, "delta ")) {
1170 patch_method = BINARY_DELTA_DEFLATED;
1171 origlen = strtoul(buffer + 6, NULL, 10);
1172 }
1173 else if (!prefixcmp(buffer, "literal ")) {
1174 patch_method = BINARY_LITERAL_DEFLATED;
1175 origlen = strtoul(buffer + 8, NULL, 10);
1176 }
1177 else
1178 return NULL;
1179
1180 linenr++;
1181 buffer += llen;
1182 while (1) {
1183 int byte_length, max_byte_length, newsize;
1184 llen = linelen(buffer, size);
1185 used += llen;
1186 linenr++;
1187 if (llen == 1) {
1188 /* consume the blank line */
1189 buffer++;
1190 size--;
1191 break;
1192 }
1193 /* Minimum line is "A00000\n" which is 7-byte long,
1194 * and the line length must be multiple of 5 plus 2.
1195 */
1196 if ((llen < 7) || (llen-2) % 5)
1197 goto corrupt;
1198 max_byte_length = (llen - 2) / 5 * 4;
1199 byte_length = *buffer;
1200 if ('A' <= byte_length && byte_length <= 'Z')
1201 byte_length = byte_length - 'A' + 1;
1202 else if ('a' <= byte_length && byte_length <= 'z')
1203 byte_length = byte_length - 'a' + 27;
1204 else
1205 goto corrupt;
1206 /* if the input length was not multiple of 4, we would
1207 * have filler at the end but the filler should never
1208 * exceed 3 bytes
1209 */
1210 if (max_byte_length < byte_length ||
1211 byte_length <= max_byte_length - 4)
1212 goto corrupt;
1213 newsize = hunk_size + byte_length;
1214 data = xrealloc(data, newsize);
1215 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1216 goto corrupt;
1217 hunk_size = newsize;
1218 buffer += llen;
1219 size -= llen;
1220 }
1221
1222 frag = xcalloc(1, sizeof(*frag));
1223 frag->patch = inflate_it(data, hunk_size, origlen);
1224 if (!frag->patch)
1225 goto corrupt;
1226 free(data);
1227 frag->size = origlen;
1228 *buf_p = buffer;
1229 *sz_p = size;
1230 *used_p = used;
1231 frag->binary_patch_method = patch_method;
1232 return frag;
1233
1234 corrupt:
1235 free(data);
1236 *status_p = -1;
1237 error("corrupt binary patch at line %d: %.*s",
1238 linenr-1, llen-1, buffer);
1239 return NULL;
1240 }
1241
1242 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1243 {
1244 /* We have read "GIT binary patch\n"; what follows is a line
1245 * that says the patch method (currently, either "literal" or
1246 * "delta") and the length of data before deflating; a
1247 * sequence of 'length-byte' followed by base-85 encoded data
1248 * follows.
1249 *
1250 * When a binary patch is reversible, there is another binary
1251 * hunk in the same format, starting with patch method (either
1252 * "literal" or "delta") with the length of data, and a sequence
1253 * of length-byte + base-85 encoded data, terminated with another
1254 * empty line. This data, when applied to the postimage, produces
1255 * the preimage.
1256 */
1257 struct fragment *forward;
1258 struct fragment *reverse;
1259 int status;
1260 int used, used_1;
1261
1262 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1263 if (!forward && !status)
1264 /* there has to be one hunk (forward hunk) */
1265 return error("unrecognized binary patch at line %d", linenr-1);
1266 if (status)
1267 /* otherwise we already gave an error message */
1268 return status;
1269
1270 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1271 if (reverse)
1272 used += used_1;
1273 else if (status) {
1274 /* not having reverse hunk is not an error, but having
1275 * a corrupt reverse hunk is.
1276 */
1277 free((void*) forward->patch);
1278 free(forward);
1279 return status;
1280 }
1281 forward->next = reverse;
1282 patch->fragments = forward;
1283 patch->is_binary = 1;
1284 return used;
1285 }
1286
1287 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1288 {
1289 int hdrsize, patchsize;
1290 int offset = find_header(buffer, size, &hdrsize, patch);
1291
1292 if (offset < 0)
1293 return offset;
1294
1295 patchsize = parse_single_patch(buffer + offset + hdrsize, size - offset - hdrsize, patch);
1296
1297 if (!patchsize) {
1298 static const char *binhdr[] = {
1299 "Binary files ",
1300 "Files ",
1301 NULL,
1302 };
1303 static const char git_binary[] = "GIT binary patch\n";
1304 int i;
1305 int hd = hdrsize + offset;
1306 unsigned long llen = linelen(buffer + hd, size - hd);
1307
1308 if (llen == sizeof(git_binary) - 1 &&
1309 !memcmp(git_binary, buffer + hd, llen)) {
1310 int used;
1311 linenr++;
1312 used = parse_binary(buffer + hd + llen,
1313 size - hd - llen, patch);
1314 if (used)
1315 patchsize = used + llen;
1316 else
1317 patchsize = 0;
1318 }
1319 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1320 for (i = 0; binhdr[i]; i++) {
1321 int len = strlen(binhdr[i]);
1322 if (len < size - hd &&
1323 !memcmp(binhdr[i], buffer + hd, len)) {
1324 linenr++;
1325 patch->is_binary = 1;
1326 patchsize = llen;
1327 break;
1328 }
1329 }
1330 }
1331
1332 /* Empty patch cannot be applied if it is a text patch
1333 * without metadata change. A binary patch appears
1334 * empty to us here.
1335 */
1336 if ((apply || check) &&
1337 (!patch->is_binary && !metadata_changes(patch)))
1338 die("patch with only garbage at line %d", linenr);
1339 }
1340
1341 return offset + hdrsize + patchsize;
1342 }
1343
1344 #define swap(a,b) myswap((a),(b),sizeof(a))
1345
1346 #define myswap(a, b, size) do { \
1347 unsigned char mytmp[size]; \
1348 memcpy(mytmp, &a, size); \
1349 memcpy(&a, &b, size); \
1350 memcpy(&b, mytmp, size); \
1351 } while (0)
1352
1353 static void reverse_patches(struct patch *p)
1354 {
1355 for (; p; p = p->next) {
1356 struct fragment *frag = p->fragments;
1357
1358 swap(p->new_name, p->old_name);
1359 swap(p->new_mode, p->old_mode);
1360 swap(p->is_new, p->is_delete);
1361 swap(p->lines_added, p->lines_deleted);
1362 swap(p->old_sha1_prefix, p->new_sha1_prefix);
1363
1364 for (; frag; frag = frag->next) {
1365 swap(frag->newpos, frag->oldpos);
1366 swap(frag->newlines, frag->oldlines);
1367 }
1368 }
1369 }
1370
1371 static const char pluses[] = "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1372 static const char minuses[]= "----------------------------------------------------------------------";
1373
1374 static void show_stats(struct patch *patch)
1375 {
1376 struct strbuf qname;
1377 char *cp = patch->new_name ? patch->new_name : patch->old_name;
1378 int max, add, del;
1379
1380 strbuf_init(&qname, 0);
1381 quote_c_style(cp, &qname, NULL, 0);
1382
1383 /*
1384 * "scale" the filename
1385 */
1386 max = max_len;
1387 if (max > 50)
1388 max = 50;
1389
1390 if (qname.len > max) {
1391 cp = strchr(qname.buf + qname.len + 3 - max, '/');
1392 if (!cp)
1393 cp = qname.buf + qname.len + 3 - max;
1394 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
1395 }
1396
1397 if (patch->is_binary) {
1398 printf(" %-*s | Bin\n", max, qname.buf);
1399 strbuf_release(&qname);
1400 return;
1401 }
1402
1403 printf(" %-*s |", max, qname.buf);
1404 strbuf_release(&qname);
1405
1406 /*
1407 * scale the add/delete
1408 */
1409 max = max + max_change > 70 ? 70 - max : max_change;
1410 add = patch->lines_added;
1411 del = patch->lines_deleted;
1412
1413 if (max_change > 0) {
1414 int total = ((add + del) * max + max_change / 2) / max_change;
1415 add = (add * max + max_change / 2) / max_change;
1416 del = total - add;
1417 }
1418 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
1419 add, pluses, del, minuses);
1420 }
1421
1422 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
1423 {
1424 switch (st->st_mode & S_IFMT) {
1425 case S_IFLNK:
1426 strbuf_grow(buf, st->st_size);
1427 if (readlink(path, buf->buf, st->st_size) != st->st_size)
1428 return -1;
1429 strbuf_setlen(buf, st->st_size);
1430 return 0;
1431 case S_IFREG:
1432 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
1433 return error("unable to open or read %s", path);
1434 convert_to_git(path, buf->buf, buf->len, buf);
1435 return 0;
1436 default:
1437 return -1;
1438 }
1439 }
1440
1441 static int find_offset(const char *buf, unsigned long size, const char *fragment, unsigned long fragsize, int line, int *lines)
1442 {
1443 int i;
1444 unsigned long start, backwards, forwards;
1445
1446 if (fragsize > size)
1447 return -1;
1448
1449 start = 0;
1450 if (line > 1) {
1451 unsigned long offset = 0;
1452 i = line-1;
1453 while (offset + fragsize <= size) {
1454 if (buf[offset++] == '\n') {
1455 start = offset;
1456 if (!--i)
1457 break;
1458 }
1459 }
1460 }
1461
1462 /* Exact line number? */
1463 if ((start + fragsize <= size) &&
1464 !memcmp(buf + start, fragment, fragsize))
1465 return start;
1466
1467 /*
1468 * There's probably some smart way to do this, but I'll leave
1469 * that to the smart and beautiful people. I'm simple and stupid.
1470 */
1471 backwards = start;
1472 forwards = start;
1473 for (i = 0; ; i++) {
1474 unsigned long try;
1475 int n;
1476
1477 /* "backward" */
1478 if (i & 1) {
1479 if (!backwards) {
1480 if (forwards + fragsize > size)
1481 break;
1482 continue;
1483 }
1484 do {
1485 --backwards;
1486 } while (backwards && buf[backwards-1] != '\n');
1487 try = backwards;
1488 } else {
1489 while (forwards + fragsize <= size) {
1490 if (buf[forwards++] == '\n')
1491 break;
1492 }
1493 try = forwards;
1494 }
1495
1496 if (try + fragsize > size)
1497 continue;
1498 if (memcmp(buf + try, fragment, fragsize))
1499 continue;
1500 n = (i >> 1)+1;
1501 if (i & 1)
1502 n = -n;
1503 *lines = n;
1504 return try;
1505 }
1506
1507 /*
1508 * We should start searching forward and backward.
1509 */
1510 return -1;
1511 }
1512
1513 static void remove_first_line(const char **rbuf, int *rsize)
1514 {
1515 const char *buf = *rbuf;
1516 int size = *rsize;
1517 unsigned long offset;
1518 offset = 0;
1519 while (offset <= size) {
1520 if (buf[offset++] == '\n')
1521 break;
1522 }
1523 *rsize = size - offset;
1524 *rbuf = buf + offset;
1525 }
1526
1527 static void remove_last_line(const char **rbuf, int *rsize)
1528 {
1529 const char *buf = *rbuf;
1530 int size = *rsize;
1531 unsigned long offset;
1532 offset = size - 1;
1533 while (offset > 0) {
1534 if (buf[--offset] == '\n')
1535 break;
1536 }
1537 *rsize = offset + 1;
1538 }
1539
1540 static int apply_line(char *output, const char *patch, int plen)
1541 {
1542 /* plen is number of bytes to be copied from patch,
1543 * starting at patch+1 (patch[0] is '+'). Typically
1544 * patch[plen] is '\n', unless this is the incomplete
1545 * last line.
1546 */
1547 int i;
1548 int add_nl_to_tail = 0;
1549 int fixed = 0;
1550 int last_tab_in_indent = -1;
1551 int last_space_in_indent = -1;
1552 int need_fix_leading_space = 0;
1553 char *buf;
1554
1555 if ((new_whitespace != strip_whitespace) || !whitespace_error ||
1556 *patch != '+') {
1557 memcpy(output, patch + 1, plen);
1558 return plen;
1559 }
1560
1561 if (1 < plen && isspace(patch[plen-1])) {
1562 if (patch[plen] == '\n')
1563 add_nl_to_tail = 1;
1564 plen--;
1565 while (0 < plen && isspace(patch[plen]))
1566 plen--;
1567 fixed = 1;
1568 }
1569
1570 for (i = 1; i < plen; i++) {
1571 char ch = patch[i];
1572 if (ch == '\t') {
1573 last_tab_in_indent = i;
1574 if (0 <= last_space_in_indent)
1575 need_fix_leading_space = 1;
1576 }
1577 else if (ch == ' ')
1578 last_space_in_indent = i;
1579 else
1580 break;
1581 }
1582
1583 buf = output;
1584 if (need_fix_leading_space) {
1585 int consecutive_spaces = 0;
1586 /* between patch[1..last_tab_in_indent] strip the
1587 * funny spaces, updating them to tab as needed.
1588 */
1589 for (i = 1; i < last_tab_in_indent; i++, plen--) {
1590 char ch = patch[i];
1591 if (ch != ' ') {
1592 consecutive_spaces = 0;
1593 *output++ = ch;
1594 } else {
1595 consecutive_spaces++;
1596 if (consecutive_spaces == 8) {
1597 *output++ = '\t';
1598 consecutive_spaces = 0;
1599 }
1600 }
1601 }
1602 fixed = 1;
1603 i = last_tab_in_indent;
1604 }
1605 else
1606 i = 1;
1607
1608 memcpy(output, patch + i, plen);
1609 if (add_nl_to_tail)
1610 output[plen++] = '\n';
1611 if (fixed)
1612 applied_after_fixing_ws++;
1613 return output + plen - buf;
1614 }
1615
1616 static int apply_one_fragment(struct strbuf *buf, struct fragment *frag, int inaccurate_eof)
1617 {
1618 int match_beginning, match_end;
1619 const char *patch = frag->patch;
1620 int offset, size = frag->size;
1621 char *old = xmalloc(size);
1622 char *new = xmalloc(size);
1623 const char *oldlines, *newlines;
1624 int oldsize = 0, newsize = 0;
1625 int new_blank_lines_at_end = 0;
1626 unsigned long leading, trailing;
1627 int pos, lines;
1628
1629 while (size > 0) {
1630 char first;
1631 int len = linelen(patch, size);
1632 int plen;
1633 int added_blank_line = 0;
1634
1635 if (!len)
1636 break;
1637
1638 /*
1639 * "plen" is how much of the line we should use for
1640 * the actual patch data. Normally we just remove the
1641 * first character on the line, but if the line is
1642 * followed by "\ No newline", then we also remove the
1643 * last one (which is the newline, of course).
1644 */
1645 plen = len-1;
1646 if (len < size && patch[len] == '\\')
1647 plen--;
1648 first = *patch;
1649 if (apply_in_reverse) {
1650 if (first == '-')
1651 first = '+';
1652 else if (first == '+')
1653 first = '-';
1654 }
1655
1656 switch (first) {
1657 case '\n':
1658 /* Newer GNU diff, empty context line */
1659 if (plen < 0)
1660 /* ... followed by '\No newline'; nothing */
1661 break;
1662 old[oldsize++] = '\n';
1663 new[newsize++] = '\n';
1664 break;
1665 case ' ':
1666 case '-':
1667 memcpy(old + oldsize, patch + 1, plen);
1668 oldsize += plen;
1669 if (first == '-')
1670 break;
1671 /* Fall-through for ' ' */
1672 case '+':
1673 if (first != '+' || !no_add) {
1674 int added = apply_line(new + newsize, patch,
1675 plen);
1676 newsize += added;
1677 if (first == '+' &&
1678 added == 1 && new[newsize-1] == '\n')
1679 added_blank_line = 1;
1680 }
1681 break;
1682 case '@': case '\\':
1683 /* Ignore it, we already handled it */
1684 break;
1685 default:
1686 if (apply_verbosely)
1687 error("invalid start of line: '%c'", first);
1688 return -1;
1689 }
1690 if (added_blank_line)
1691 new_blank_lines_at_end++;
1692 else
1693 new_blank_lines_at_end = 0;
1694 patch += len;
1695 size -= len;
1696 }
1697
1698 if (inaccurate_eof && oldsize > 0 && old[oldsize - 1] == '\n' &&
1699 newsize > 0 && new[newsize - 1] == '\n') {
1700 oldsize--;
1701 newsize--;
1702 }
1703
1704 oldlines = old;
1705 newlines = new;
1706 leading = frag->leading;
1707 trailing = frag->trailing;
1708
1709 /*
1710 * If we don't have any leading/trailing data in the patch,
1711 * we want it to match at the beginning/end of the file.
1712 *
1713 * But that would break if the patch is generated with
1714 * --unified=0; sane people wouldn't do that to cause us
1715 * trouble, but we try to please not so sane ones as well.
1716 */
1717 if (unidiff_zero) {
1718 match_beginning = (!leading && !frag->oldpos);
1719 match_end = 0;
1720 }
1721 else {
1722 match_beginning = !leading && (frag->oldpos == 1);
1723 match_end = !trailing;
1724 }
1725
1726 lines = 0;
1727 pos = frag->newpos;
1728 for (;;) {
1729 offset = find_offset(buf->buf, buf->len,
1730 oldlines, oldsize, pos, &lines);
1731 if (match_end && offset + oldsize != buf->len)
1732 offset = -1;
1733 if (match_beginning && offset)
1734 offset = -1;
1735 if (offset >= 0) {
1736 if (new_whitespace == strip_whitespace &&
1737 (buf->len - oldsize - offset == 0)) /* end of file? */
1738 newsize -= new_blank_lines_at_end;
1739
1740 /* Warn if it was necessary to reduce the number
1741 * of context lines.
1742 */
1743 if ((leading != frag->leading) ||
1744 (trailing != frag->trailing))
1745 fprintf(stderr, "Context reduced to (%ld/%ld)"
1746 " to apply fragment at %d\n",
1747 leading, trailing, pos + lines);
1748
1749 strbuf_splice(buf, offset, oldsize, newlines, newsize);
1750 offset = 0;
1751 break;
1752 }
1753
1754 /* Am I at my context limits? */
1755 if ((leading <= p_context) && (trailing <= p_context))
1756 break;
1757 if (match_beginning || match_end) {
1758 match_beginning = match_end = 0;
1759 continue;
1760 }
1761 /* Reduce the number of context lines
1762 * Reduce both leading and trailing if they are equal
1763 * otherwise just reduce the larger context.
1764 */
1765 if (leading >= trailing) {
1766 remove_first_line(&oldlines, &oldsize);
1767 remove_first_line(&newlines, &newsize);
1768 pos--;
1769 leading--;
1770 }
1771 if (trailing > leading) {
1772 remove_last_line(&oldlines, &oldsize);
1773 remove_last_line(&newlines, &newsize);
1774 trailing--;
1775 }
1776 }
1777
1778 if (offset && apply_verbosely)
1779 error("while searching for:\n%.*s", oldsize, oldlines);
1780
1781 free(old);
1782 free(new);
1783 return offset;
1784 }
1785
1786 static int apply_binary_fragment(struct strbuf *buf, struct patch *patch)
1787 {
1788 struct fragment *fragment = patch->fragments;
1789 unsigned long len;
1790 void *dst;
1791
1792 /* Binary patch is irreversible without the optional second hunk */
1793 if (apply_in_reverse) {
1794 if (!fragment->next)
1795 return error("cannot reverse-apply a binary patch "
1796 "without the reverse hunk to '%s'",
1797 patch->new_name
1798 ? patch->new_name : patch->old_name);
1799 fragment = fragment->next;
1800 }
1801 switch (fragment->binary_patch_method) {
1802 case BINARY_DELTA_DEFLATED:
1803 dst = patch_delta(buf->buf, buf->len, fragment->patch,
1804 fragment->size, &len);
1805 if (!dst)
1806 return -1;
1807 /* XXX patch_delta NUL-terminates */
1808 strbuf_attach(buf, dst, len, len + 1);
1809 return 0;
1810 case BINARY_LITERAL_DEFLATED:
1811 strbuf_reset(buf);
1812 strbuf_add(buf, fragment->patch, fragment->size);
1813 return 0;
1814 }
1815 return -1;
1816 }
1817
1818 static int apply_binary(struct strbuf *buf, struct patch *patch)
1819 {
1820 const char *name = patch->old_name ? patch->old_name : patch->new_name;
1821 unsigned char sha1[20];
1822
1823 /* For safety, we require patch index line to contain
1824 * full 40-byte textual SHA1 for old and new, at least for now.
1825 */
1826 if (strlen(patch->old_sha1_prefix) != 40 ||
1827 strlen(patch->new_sha1_prefix) != 40 ||
1828 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
1829 get_sha1_hex(patch->new_sha1_prefix, sha1))
1830 return error("cannot apply binary patch to '%s' "
1831 "without full index line", name);
1832
1833 if (patch->old_name) {
1834 /* See if the old one matches what the patch
1835 * applies to.
1836 */
1837 hash_sha1_file(buf->buf, buf->len, blob_type, sha1);
1838 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
1839 return error("the patch applies to '%s' (%s), "
1840 "which does not match the "
1841 "current contents.",
1842 name, sha1_to_hex(sha1));
1843 }
1844 else {
1845 /* Otherwise, the old one must be empty. */
1846 if (buf->len)
1847 return error("the patch applies to an empty "
1848 "'%s' but it is not empty", name);
1849 }
1850
1851 get_sha1_hex(patch->new_sha1_prefix, sha1);
1852 if (is_null_sha1(sha1)) {
1853 strbuf_release(buf);
1854 return 0; /* deletion patch */
1855 }
1856
1857 if (has_sha1_file(sha1)) {
1858 /* We already have the postimage */
1859 enum object_type type;
1860 unsigned long size;
1861 char *result;
1862
1863 result = read_sha1_file(sha1, &type, &size);
1864 if (!result)
1865 return error("the necessary postimage %s for "
1866 "'%s' cannot be read",
1867 patch->new_sha1_prefix, name);
1868 /* XXX read_sha1_file NUL-terminates */
1869 strbuf_attach(buf, result, size, size + 1);
1870 } else {
1871 /* We have verified buf matches the preimage;
1872 * apply the patch data to it, which is stored
1873 * in the patch->fragments->{patch,size}.
1874 */
1875 if (apply_binary_fragment(buf, patch))
1876 return error("binary patch does not apply to '%s'",
1877 name);
1878
1879 /* verify that the result matches */
1880 hash_sha1_file(buf->buf, buf->len, blob_type, sha1);
1881 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
1882 return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)",
1883 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
1884 }
1885
1886 return 0;
1887 }
1888
1889 static int apply_fragments(struct strbuf *buf, struct patch *patch)
1890 {
1891 struct fragment *frag = patch->fragments;
1892 const char *name = patch->old_name ? patch->old_name : patch->new_name;
1893
1894 if (patch->is_binary)
1895 return apply_binary(buf, patch);
1896
1897 while (frag) {
1898 if (apply_one_fragment(buf, frag, patch->inaccurate_eof)) {
1899 error("patch failed: %s:%ld", name, frag->oldpos);
1900 if (!apply_with_reject)
1901 return -1;
1902 frag->rejected = 1;
1903 }
1904 frag = frag->next;
1905 }
1906 return 0;
1907 }
1908
1909 static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
1910 {
1911 if (!ce)
1912 return 0;
1913
1914 if (S_ISGITLINK(ntohl(ce->ce_mode))) {
1915 strbuf_grow(buf, 100);
1916 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(ce->sha1));
1917 } else {
1918 enum object_type type;
1919 unsigned long sz;
1920 char *result;
1921
1922 result = read_sha1_file(ce->sha1, &type, &sz);
1923 if (!result)
1924 return -1;
1925 /* XXX read_sha1_file NUL-terminates */
1926 strbuf_attach(buf, result, sz, sz + 1);
1927 }
1928 return 0;
1929 }
1930
1931 static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
1932 {
1933 struct strbuf buf;
1934
1935 strbuf_init(&buf, 0);
1936 if (cached) {
1937 if (read_file_or_gitlink(ce, &buf))
1938 return error("read of %s failed", patch->old_name);
1939 } else if (patch->old_name) {
1940 if (S_ISGITLINK(patch->old_mode)) {
1941 if (ce) {
1942 read_file_or_gitlink(ce, &buf);
1943 } else {
1944 /*
1945 * There is no way to apply subproject
1946 * patch without looking at the index.
1947 */
1948 patch->fragments = NULL;
1949 }
1950 } else {
1951 if (read_old_data(st, patch->old_name, &buf))
1952 return error("read of %s failed", patch->old_name);
1953 }
1954 }
1955
1956 if (apply_fragments(&buf, patch) < 0)
1957 return -1; /* note with --reject this succeeds. */
1958 patch->result = strbuf_detach(&buf, &patch->resultsize);
1959
1960 if (0 < patch->is_delete && patch->resultsize)
1961 return error("removal patch leaves file contents");
1962
1963 return 0;
1964 }
1965
1966 static int check_to_create_blob(const char *new_name, int ok_if_exists)
1967 {
1968 struct stat nst;
1969 if (!lstat(new_name, &nst)) {
1970 if (S_ISDIR(nst.st_mode) || ok_if_exists)
1971 return 0;
1972 /*
1973 * A leading component of new_name might be a symlink
1974 * that is going to be removed with this patch, but
1975 * still pointing at somewhere that has the path.
1976 * In such a case, path "new_name" does not exist as
1977 * far as git is concerned.
1978 */
1979 if (has_symlink_leading_path(new_name, NULL))
1980 return 0;
1981
1982 return error("%s: already exists in working directory", new_name);
1983 }
1984 else if ((errno != ENOENT) && (errno != ENOTDIR))
1985 return error("%s: %s", new_name, strerror(errno));
1986 return 0;
1987 }
1988
1989 static int verify_index_match(struct cache_entry *ce, struct stat *st)
1990 {
1991 if (S_ISGITLINK(ntohl(ce->ce_mode))) {
1992 if (!S_ISDIR(st->st_mode))
1993 return -1;
1994 return 0;
1995 }
1996 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID);
1997 }
1998
1999 static int check_patch(struct patch *patch, struct patch *prev_patch)
2000 {
2001 struct stat st;
2002 const char *old_name = patch->old_name;
2003 const char *new_name = patch->new_name;
2004 const char *name = old_name ? old_name : new_name;
2005 struct cache_entry *ce = NULL;
2006 int ok_if_exists;
2007
2008 patch->rejected = 1; /* we will drop this after we succeed */
2009
2010 /*
2011 * Make sure that we do not have local modifications from the
2012 * index when we are looking at the index. Also make sure
2013 * we have the preimage file to be patched in the work tree,
2014 * unless --cached, which tells git to apply only in the index.
2015 */
2016 if (old_name) {
2017 int stat_ret = 0;
2018 unsigned st_mode = 0;
2019
2020 if (!cached)
2021 stat_ret = lstat(old_name, &st);
2022 if (check_index) {
2023 int pos = cache_name_pos(old_name, strlen(old_name));
2024 if (pos < 0)
2025 return error("%s: does not exist in index",
2026 old_name);
2027 ce = active_cache[pos];
2028 if (stat_ret < 0) {
2029 struct checkout costate;
2030 if (errno != ENOENT)
2031 return error("%s: %s", old_name,
2032 strerror(errno));
2033 /* checkout */
2034 costate.base_dir = "";
2035 costate.base_dir_len = 0;
2036 costate.force = 0;
2037 costate.quiet = 0;
2038 costate.not_new = 0;
2039 costate.refresh_cache = 1;
2040 if (checkout_entry(ce,
2041 &costate,
2042 NULL) ||
2043 lstat(old_name, &st))
2044 return -1;
2045 }
2046 if (!cached && verify_index_match(ce, &st))
2047 return error("%s: does not match index",
2048 old_name);
2049 if (cached)
2050 st_mode = ntohl(ce->ce_mode);
2051 } else if (stat_ret < 0)
2052 return error("%s: %s", old_name, strerror(errno));
2053
2054 if (!cached)
2055 st_mode = ntohl(ce_mode_from_stat(ce, st.st_mode));
2056
2057 if (patch->is_new < 0)
2058 patch->is_new = 0;
2059 if (!patch->old_mode)
2060 patch->old_mode = st_mode;
2061 if ((st_mode ^ patch->old_mode) & S_IFMT)
2062 return error("%s: wrong type", old_name);
2063 if (st_mode != patch->old_mode)
2064 fprintf(stderr, "warning: %s has type %o, expected %o\n",
2065 old_name, st_mode, patch->old_mode);
2066 }
2067
2068 if (new_name && prev_patch && 0 < prev_patch->is_delete &&
2069 !strcmp(prev_patch->old_name, new_name))
2070 /* A type-change diff is always split into a patch to
2071 * delete old, immediately followed by a patch to
2072 * create new (see diff.c::run_diff()); in such a case
2073 * it is Ok that the entry to be deleted by the
2074 * previous patch is still in the working tree and in
2075 * the index.
2076 */
2077 ok_if_exists = 1;
2078 else
2079 ok_if_exists = 0;
2080
2081 if (new_name &&
2082 ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
2083 if (check_index &&
2084 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
2085 !ok_if_exists)
2086 return error("%s: already exists in index", new_name);
2087 if (!cached) {
2088 int err = check_to_create_blob(new_name, ok_if_exists);
2089 if (err)
2090 return err;
2091 }
2092 if (!patch->new_mode) {
2093 if (0 < patch->is_new)
2094 patch->new_mode = S_IFREG | 0644;
2095 else
2096 patch->new_mode = patch->old_mode;
2097 }
2098 }
2099
2100 if (new_name && old_name) {
2101 int same = !strcmp(old_name, new_name);
2102 if (!patch->new_mode)
2103 patch->new_mode = patch->old_mode;
2104 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
2105 return error("new mode (%o) of %s does not match old mode (%o)%s%s",
2106 patch->new_mode, new_name, patch->old_mode,
2107 same ? "" : " of ", same ? "" : old_name);
2108 }
2109
2110 if (apply_data(patch, &st, ce) < 0)
2111 return error("%s: patch does not apply", name);
2112 patch->rejected = 0;
2113 return 0;
2114 }
2115
2116 static int check_patch_list(struct patch *patch)
2117 {
2118 struct patch *prev_patch = NULL;
2119 int err = 0;
2120
2121 for (prev_patch = NULL; patch ; patch = patch->next) {
2122 if (apply_verbosely)
2123 say_patch_name(stderr,
2124 "Checking patch ", patch, "...\n");
2125 err |= check_patch(patch, prev_patch);
2126 prev_patch = patch;
2127 }
2128 return err;
2129 }
2130
2131 /* This function tries to read the sha1 from the current index */
2132 static int get_current_sha1(const char *path, unsigned char *sha1)
2133 {
2134 int pos;
2135
2136 if (read_cache() < 0)
2137 return -1;
2138 pos = cache_name_pos(path, strlen(path));
2139 if (pos < 0)
2140 return -1;
2141 hashcpy(sha1, active_cache[pos]->sha1);
2142 return 0;
2143 }
2144
2145 /* Build an index that contains the just the files needed for a 3way merge */
2146 static void build_fake_ancestor(struct patch *list, const char *filename)
2147 {
2148 struct patch *patch;
2149 struct index_state result = { 0 };
2150 int fd;
2151
2152 /* Once we start supporting the reverse patch, it may be
2153 * worth showing the new sha1 prefix, but until then...
2154 */
2155 for (patch = list; patch; patch = patch->next) {
2156 const unsigned char *sha1_ptr;
2157 unsigned char sha1[20];
2158 struct cache_entry *ce;
2159 const char *name;
2160
2161 name = patch->old_name ? patch->old_name : patch->new_name;
2162 if (0 < patch->is_new)
2163 continue;
2164 else if (get_sha1(patch->old_sha1_prefix, sha1))
2165 /* git diff has no index line for mode/type changes */
2166 if (!patch->lines_added && !patch->lines_deleted) {
2167 if (get_current_sha1(patch->new_name, sha1) ||
2168 get_current_sha1(patch->old_name, sha1))
2169 die("mode change for %s, which is not "
2170 "in current HEAD", name);
2171 sha1_ptr = sha1;
2172 } else
2173 die("sha1 information is lacking or useless "
2174 "(%s).", name);
2175 else
2176 sha1_ptr = sha1;
2177
2178 ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);
2179 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
2180 die ("Could not add %s to temporary index", name);
2181 }
2182
2183 fd = open(filename, O_WRONLY | O_CREAT, 0666);
2184 if (fd < 0 || write_index(&result, fd) || close(fd))
2185 die ("Could not write temporary index to %s", filename);
2186
2187 discard_index(&result);
2188 }
2189
2190 static void stat_patch_list(struct patch *patch)
2191 {
2192 int files, adds, dels;
2193
2194 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
2195 files++;
2196 adds += patch->lines_added;
2197 dels += patch->lines_deleted;
2198 show_stats(patch);
2199 }
2200
2201 printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
2202 }
2203
2204 static void numstat_patch_list(struct patch *patch)
2205 {
2206 for ( ; patch; patch = patch->next) {
2207 const char *name;
2208 name = patch->new_name ? patch->new_name : patch->old_name;
2209 if (patch->is_binary)
2210 printf("-\t-\t");
2211 else
2212 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
2213 write_name_quoted(name, stdout, line_termination);
2214 }
2215 }
2216
2217 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
2218 {
2219 if (mode)
2220 printf(" %s mode %06o %s\n", newdelete, mode, name);
2221 else
2222 printf(" %s %s\n", newdelete, name);
2223 }
2224
2225 static void show_mode_change(struct patch *p, int show_name)
2226 {
2227 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
2228 if (show_name)
2229 printf(" mode change %06o => %06o %s\n",
2230 p->old_mode, p->new_mode, p->new_name);
2231 else
2232 printf(" mode change %06o => %06o\n",
2233 p->old_mode, p->new_mode);
2234 }
2235 }
2236
2237 static void show_rename_copy(struct patch *p)
2238 {
2239 const char *renamecopy = p->is_rename ? "rename" : "copy";
2240 const char *old, *new;
2241
2242 /* Find common prefix */
2243 old = p->old_name;
2244 new = p->new_name;
2245 while (1) {
2246 const char *slash_old, *slash_new;
2247 slash_old = strchr(old, '/');
2248 slash_new = strchr(new, '/');
2249 if (!slash_old ||
2250 !slash_new ||
2251 slash_old - old != slash_new - new ||
2252 memcmp(old, new, slash_new - new))
2253 break;
2254 old = slash_old + 1;
2255 new = slash_new + 1;
2256 }
2257 /* p->old_name thru old is the common prefix, and old and new
2258 * through the end of names are renames
2259 */
2260 if (old != p->old_name)
2261 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
2262 (int)(old - p->old_name), p->old_name,
2263 old, new, p->score);
2264 else
2265 printf(" %s %s => %s (%d%%)\n", renamecopy,
2266 p->old_name, p->new_name, p->score);
2267 show_mode_change(p, 0);
2268 }
2269
2270 static void summary_patch_list(struct patch *patch)
2271 {
2272 struct patch *p;
2273
2274 for (p = patch; p; p = p->next) {
2275 if (p->is_new)
2276 show_file_mode_name("create", p->new_mode, p->new_name);
2277 else if (p->is_delete)
2278 show_file_mode_name("delete", p->old_mode, p->old_name);
2279 else {
2280 if (p->is_rename || p->is_copy)
2281 show_rename_copy(p);
2282 else {
2283 if (p->score) {
2284 printf(" rewrite %s (%d%%)\n",
2285 p->new_name, p->score);
2286 show_mode_change(p, 0);
2287 }
2288 else
2289 show_mode_change(p, 1);
2290 }
2291 }
2292 }
2293 }
2294
2295 static void patch_stats(struct patch *patch)
2296 {
2297 int lines = patch->lines_added + patch->lines_deleted;
2298
2299 if (lines > max_change)
2300 max_change = lines;
2301 if (patch->old_name) {
2302 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
2303 if (!len)
2304 len = strlen(patch->old_name);
2305 if (len > max_len)
2306 max_len = len;
2307 }
2308 if (patch->new_name) {
2309 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
2310 if (!len)
2311 len = strlen(patch->new_name);
2312 if (len > max_len)
2313 max_len = len;
2314 }
2315 }
2316
2317 static void remove_file(struct patch *patch, int rmdir_empty)
2318 {
2319 if (update_index) {
2320 if (remove_file_from_cache(patch->old_name) < 0)
2321 die("unable to remove %s from index", patch->old_name);
2322 }
2323 if (!cached) {
2324 if (S_ISGITLINK(patch->old_mode)) {
2325 if (rmdir(patch->old_name))
2326 warning("unable to remove submodule %s",
2327 patch->old_name);
2328 } else if (!unlink(patch->old_name) && rmdir_empty) {
2329 char *name = xstrdup(patch->old_name);
2330 char *end = strrchr(name, '/');
2331 while (end) {
2332 *end = 0;
2333 if (rmdir(name))
2334 break;
2335 end = strrchr(name, '/');
2336 }
2337 free(name);
2338 }
2339 }
2340 }
2341
2342 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
2343 {
2344 struct stat st;
2345 struct cache_entry *ce;
2346 int namelen = strlen(path);
2347 unsigned ce_size = cache_entry_size(namelen);
2348
2349 if (!update_index)
2350 return;
2351
2352 ce = xcalloc(1, ce_size);
2353 memcpy(ce->name, path, namelen);
2354 ce->ce_mode = create_ce_mode(mode);
2355 ce->ce_flags = htons(namelen);
2356 if (S_ISGITLINK(mode)) {
2357 const char *s = buf;
2358
2359 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
2360 die("corrupt patch for subproject %s", path);
2361 } else {
2362 if (!cached) {
2363 if (lstat(path, &st) < 0)
2364 die("unable to stat newly created file %s",
2365 path);
2366 fill_stat_cache_info(ce, &st);
2367 }
2368 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
2369 die("unable to create backing store for newly created file %s", path);
2370 }
2371 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
2372 die("unable to add cache entry for %s", path);
2373 }
2374
2375 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
2376 {
2377 int fd;
2378 struct strbuf nbuf;
2379
2380 if (S_ISGITLINK(mode)) {
2381 struct stat st;
2382 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
2383 return 0;
2384 return mkdir(path, 0777);
2385 }
2386
2387 if (has_symlinks && S_ISLNK(mode))
2388 /* Although buf:size is counted string, it also is NUL
2389 * terminated.
2390 */
2391 return symlink(buf, path);
2392
2393 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
2394 if (fd < 0)
2395 return -1;
2396
2397 strbuf_init(&nbuf, 0);
2398 if (convert_to_working_tree(path, buf, size, &nbuf)) {
2399 size = nbuf.len;
2400 buf = nbuf.buf;
2401 }
2402 write_or_die(fd, buf, size);
2403 strbuf_release(&nbuf);
2404
2405 if (close(fd) < 0)
2406 die("closing file %s: %s", path, strerror(errno));
2407 return 0;
2408 }
2409
2410 /*
2411 * We optimistically assume that the directories exist,
2412 * which is true 99% of the time anyway. If they don't,
2413 * we create them and try again.
2414 */
2415 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
2416 {
2417 if (cached)
2418 return;
2419 if (!try_create_file(path, mode, buf, size))
2420 return;
2421
2422 if (errno == ENOENT) {
2423 if (safe_create_leading_directories(path))
2424 return;
2425 if (!try_create_file(path, mode, buf, size))
2426 return;
2427 }
2428
2429 if (errno == EEXIST || errno == EACCES) {
2430 /* We may be trying to create a file where a directory
2431 * used to be.
2432 */
2433 struct stat st;
2434 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
2435 errno = EEXIST;
2436 }
2437
2438 if (errno == EEXIST) {
2439 unsigned int nr = getpid();
2440
2441 for (;;) {
2442 const char *newpath;
2443 newpath = mkpath("%s~%u", path, nr);
2444 if (!try_create_file(newpath, mode, buf, size)) {
2445 if (!rename(newpath, path))
2446 return;
2447 unlink(newpath);
2448 break;
2449 }
2450 if (errno != EEXIST)
2451 break;
2452 ++nr;
2453 }
2454 }
2455 die("unable to write file %s mode %o", path, mode);
2456 }
2457
2458 static void create_file(struct patch *patch)
2459 {
2460 char *path = patch->new_name;
2461 unsigned mode = patch->new_mode;
2462 unsigned long size = patch->resultsize;
2463 char *buf = patch->result;
2464
2465 if (!mode)
2466 mode = S_IFREG | 0644;
2467 create_one_file(path, mode, buf, size);
2468 add_index_file(path, mode, buf, size);
2469 }
2470
2471 /* phase zero is to remove, phase one is to create */
2472 static void write_out_one_result(struct patch *patch, int phase)
2473 {
2474 if (patch->is_delete > 0) {
2475 if (phase == 0)
2476 remove_file(patch, 1);
2477 return;
2478 }
2479 if (patch->is_new > 0 || patch->is_copy) {
2480 if (phase == 1)
2481 create_file(patch);
2482 return;
2483 }
2484 /*
2485 * Rename or modification boils down to the same
2486 * thing: remove the old, write the new
2487 */
2488 if (phase == 0)
2489 remove_file(patch, patch->is_rename);
2490 if (phase == 1)
2491 create_file(patch);
2492 }
2493
2494 static int write_out_one_reject(struct patch *patch)
2495 {
2496 FILE *rej;
2497 char namebuf[PATH_MAX];
2498 struct fragment *frag;
2499 int cnt = 0;
2500
2501 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
2502 if (!frag->rejected)
2503 continue;
2504 cnt++;
2505 }
2506
2507 if (!cnt) {
2508 if (apply_verbosely)
2509 say_patch_name(stderr,
2510 "Applied patch ", patch, " cleanly.\n");
2511 return 0;
2512 }
2513
2514 /* This should not happen, because a removal patch that leaves
2515 * contents are marked "rejected" at the patch level.
2516 */
2517 if (!patch->new_name)
2518 die("internal error");
2519
2520 /* Say this even without --verbose */
2521 say_patch_name(stderr, "Applying patch ", patch, " with");
2522 fprintf(stderr, " %d rejects...\n", cnt);
2523
2524 cnt = strlen(patch->new_name);
2525 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
2526 cnt = ARRAY_SIZE(namebuf) - 5;
2527 fprintf(stderr,
2528 "warning: truncating .rej filename to %.*s.rej",
2529 cnt - 1, patch->new_name);
2530 }
2531 memcpy(namebuf, patch->new_name, cnt);
2532 memcpy(namebuf + cnt, ".rej", 5);
2533
2534 rej = fopen(namebuf, "w");
2535 if (!rej)
2536 return error("cannot open %s: %s", namebuf, strerror(errno));
2537
2538 /* Normal git tools never deal with .rej, so do not pretend
2539 * this is a git patch by saying --git nor give extended
2540 * headers. While at it, maybe please "kompare" that wants
2541 * the trailing TAB and some garbage at the end of line ;-).
2542 */
2543 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
2544 patch->new_name, patch->new_name);
2545 for (cnt = 1, frag = patch->fragments;
2546 frag;
2547 cnt++, frag = frag->next) {
2548 if (!frag->rejected) {
2549 fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);
2550 continue;
2551 }
2552 fprintf(stderr, "Rejected hunk #%d.\n", cnt);
2553 fprintf(rej, "%.*s", frag->size, frag->patch);
2554 if (frag->patch[frag->size-1] != '\n')
2555 fputc('\n', rej);
2556 }
2557 fclose(rej);
2558 return -1;
2559 }
2560
2561 static int write_out_results(struct patch *list, int skipped_patch)
2562 {
2563 int phase;
2564 int errs = 0;
2565 struct patch *l;
2566
2567 if (!list && !skipped_patch)
2568 return error("No changes");
2569
2570 for (phase = 0; phase < 2; phase++) {
2571 l = list;
2572 while (l) {
2573 if (l->rejected)
2574 errs = 1;
2575 else {
2576 write_out_one_result(l, phase);
2577 if (phase == 1 && write_out_one_reject(l))
2578 errs = 1;
2579 }
2580 l = l->next;
2581 }
2582 }
2583 return errs;
2584 }
2585
2586 static struct lock_file lock_file;
2587
2588 static struct excludes {
2589 struct excludes *next;
2590 const char *path;
2591 } *excludes;
2592
2593 static int use_patch(struct patch *p)
2594 {
2595 const char *pathname = p->new_name ? p->new_name : p->old_name;
2596 struct excludes *x = excludes;
2597 while (x) {
2598 if (fnmatch(x->path, pathname, 0) == 0)
2599 return 0;
2600 x = x->next;
2601 }
2602 if (0 < prefix_length) {
2603 int pathlen = strlen(pathname);
2604 if (pathlen <= prefix_length ||
2605 memcmp(prefix, pathname, prefix_length))
2606 return 0;
2607 }
2608 return 1;
2609 }
2610
2611 static void prefix_one(char **name)
2612 {
2613 char *old_name = *name;
2614 if (!old_name)
2615 return;
2616 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
2617 free(old_name);
2618 }
2619
2620 static void prefix_patches(struct patch *p)
2621 {
2622 if (!prefix || p->is_toplevel_relative)
2623 return;
2624 for ( ; p; p = p->next) {
2625 if (p->new_name == p->old_name) {
2626 char *prefixed = p->new_name;
2627 prefix_one(&prefixed);
2628 p->new_name = p->old_name = prefixed;
2629 }
2630 else {
2631 prefix_one(&p->new_name);
2632 prefix_one(&p->old_name);
2633 }
2634 }
2635 }
2636
2637 static int apply_patch(int fd, const char *filename, int inaccurate_eof)
2638 {
2639 size_t offset;
2640 struct strbuf buf;
2641 struct patch *list = NULL, **listp = &list;
2642 int skipped_patch = 0;
2643
2644 strbuf_init(&buf, 0);
2645 patch_input_file = filename;
2646 read_patch_file(&buf, fd);
2647 offset = 0;
2648 while (offset < buf.len) {
2649 struct patch *patch;
2650 int nr;
2651
2652 patch = xcalloc(1, sizeof(*patch));
2653 patch->inaccurate_eof = inaccurate_eof;
2654 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
2655 if (nr < 0)
2656 break;
2657 if (apply_in_reverse)
2658 reverse_patches(patch);
2659 if (prefix)
2660 prefix_patches(patch);
2661 if (use_patch(patch)) {
2662 patch_stats(patch);
2663 *listp = patch;
2664 listp = &patch->next;
2665 }
2666 else {
2667 /* perhaps free it a bit better? */
2668 free(patch);
2669 skipped_patch++;
2670 }
2671 offset += nr;
2672 }
2673
2674 if (whitespace_error && (new_whitespace == error_on_whitespace))
2675 apply = 0;
2676
2677 update_index = check_index && apply;
2678 if (update_index && newfd < 0)
2679 newfd = hold_locked_index(&lock_file, 1);
2680
2681 if (check_index) {
2682 if (read_cache() < 0)
2683 die("unable to read index file");
2684 }
2685
2686 if ((check || apply) &&
2687 check_patch_list(list) < 0 &&
2688 !apply_with_reject)
2689 exit(1);
2690
2691 if (apply && write_out_results(list, skipped_patch))
2692 exit(1);
2693
2694 if (fake_ancestor)
2695 build_fake_ancestor(list, fake_ancestor);
2696
2697 if (diffstat)
2698 stat_patch_list(list);
2699
2700 if (numstat)
2701 numstat_patch_list(list);
2702
2703 if (summary)
2704 summary_patch_list(list);
2705
2706 strbuf_release(&buf);
2707 return 0;
2708 }
2709
2710 static int git_apply_config(const char *var, const char *value)
2711 {
2712 if (!strcmp(var, "apply.whitespace")) {
2713 apply_default_whitespace = xstrdup(value);
2714 return 0;
2715 }
2716 return git_default_config(var, value);
2717 }
2718
2719
2720 int cmd_apply(int argc, const char **argv, const char *unused_prefix)
2721 {
2722 int i;
2723 int read_stdin = 1;
2724 int inaccurate_eof = 0;
2725 int errs = 0;
2726 int is_not_gitdir = 0;
2727
2728 const char *whitespace_option = NULL;
2729
2730 prefix = setup_git_directory_gently(&is_not_gitdir);
2731 prefix_length = prefix ? strlen(prefix) : 0;
2732 git_config(git_apply_config);
2733 if (apply_default_whitespace)
2734 parse_whitespace_option(apply_default_whitespace);
2735
2736 for (i = 1; i < argc; i++) {
2737 const char *arg = argv[i];
2738 char *end;
2739 int fd;
2740
2741 if (!strcmp(arg, "-")) {
2742 errs |= apply_patch(0, "<stdin>", inaccurate_eof);
2743 read_stdin = 0;
2744 continue;
2745 }
2746 if (!prefixcmp(arg, "--exclude=")) {
2747 struct excludes *x = xmalloc(sizeof(*x));
2748 x->path = arg + 10;
2749 x->next = excludes;
2750 excludes = x;
2751 continue;
2752 }
2753 if (!prefixcmp(arg, "-p")) {
2754 p_value = atoi(arg + 2);
2755 p_value_known = 1;
2756 continue;
2757 }
2758 if (!strcmp(arg, "--no-add")) {
2759 no_add = 1;
2760 continue;
2761 }
2762 if (!strcmp(arg, "--stat")) {
2763 apply = 0;
2764 diffstat = 1;
2765 continue;
2766 }
2767 if (!strcmp(arg, "--allow-binary-replacement") ||
2768 !strcmp(arg, "--binary")) {
2769 continue; /* now no-op */
2770 }
2771 if (!strcmp(arg, "--numstat")) {
2772 apply = 0;
2773 numstat = 1;
2774 continue;
2775 }
2776 if (!strcmp(arg, "--summary")) {
2777 apply = 0;
2778 summary = 1;
2779 continue;
2780 }
2781 if (!strcmp(arg, "--check")) {
2782 apply = 0;
2783 check = 1;
2784 continue;
2785 }
2786 if (!strcmp(arg, "--index")) {
2787 if (is_not_gitdir)
2788 die("--index outside a repository");
2789 check_index = 1;
2790 continue;
2791 }
2792 if (!strcmp(arg, "--cached")) {
2793 if (is_not_gitdir)
2794 die("--cached outside a repository");
2795 check_index = 1;
2796 cached = 1;
2797 continue;
2798 }
2799 if (!strcmp(arg, "--apply")) {
2800 apply = 1;
2801 continue;
2802 }
2803 if (!strcmp(arg, "--build-fake-ancestor")) {
2804 apply = 0;
2805 if (++i >= argc)
2806 die ("need a filename");
2807 fake_ancestor = argv[i];
2808 continue;
2809 }
2810 if (!strcmp(arg, "-z")) {
2811 line_termination = 0;
2812 continue;
2813 }
2814 if (!prefixcmp(arg, "-C")) {
2815 p_context = strtoul(arg + 2, &end, 0);
2816 if (*end != '\0')
2817 die("unrecognized context count '%s'", arg + 2);
2818 continue;
2819 }
2820 if (!prefixcmp(arg, "--whitespace=")) {
2821 whitespace_option = arg + 13;
2822 parse_whitespace_option(arg + 13);
2823 continue;
2824 }
2825 if (!strcmp(arg, "-R") || !strcmp(arg, "--reverse")) {
2826 apply_in_reverse = 1;
2827 continue;
2828 }
2829 if (!strcmp(arg, "--unidiff-zero")) {
2830 unidiff_zero = 1;
2831 continue;
2832 }
2833 if (!strcmp(arg, "--reject")) {
2834 apply = apply_with_reject = apply_verbosely = 1;
2835 continue;
2836 }
2837 if (!strcmp(arg, "-v") || !strcmp(arg, "--verbose")) {
2838 apply_verbosely = 1;
2839 continue;
2840 }
2841 if (!strcmp(arg, "--inaccurate-eof")) {
2842 inaccurate_eof = 1;
2843 continue;
2844 }
2845 if (0 < prefix_length)
2846 arg = prefix_filename(prefix, prefix_length, arg);
2847
2848 fd = open(arg, O_RDONLY);
2849 if (fd < 0)
2850 usage(apply_usage);
2851 read_stdin = 0;
2852 set_default_whitespace_mode(whitespace_option);
2853 errs |= apply_patch(fd, arg, inaccurate_eof);
2854 close(fd);
2855 }
2856 set_default_whitespace_mode(whitespace_option);
2857 if (read_stdin)
2858 errs |= apply_patch(0, "<stdin>", inaccurate_eof);
2859 if (whitespace_error) {
2860 if (squelch_whitespace_errors &&
2861 squelch_whitespace_errors < whitespace_error) {
2862 int squelched =
2863 whitespace_error - squelch_whitespace_errors;
2864 fprintf(stderr, "warning: squelched %d "
2865 "whitespace error%s\n",
2866 squelched,
2867 squelched == 1 ? "" : "s");
2868 }
2869 if (new_whitespace == error_on_whitespace)
2870 die("%d line%s add%s whitespace errors.",
2871 whitespace_error,
2872 whitespace_error == 1 ? "" : "s",
2873 whitespace_error == 1 ? "s" : "");
2874 if (applied_after_fixing_ws)
2875 fprintf(stderr, "warning: %d line%s applied after"
2876 " fixing whitespace errors.\n",
2877 applied_after_fixing_ws,
2878 applied_after_fixing_ws == 1 ? "" : "s");
2879 else if (whitespace_error)
2880 fprintf(stderr, "warning: %d line%s add%s whitespace errors.\n",
2881 whitespace_error,
2882 whitespace_error == 1 ? "" : "s",
2883 whitespace_error == 1 ? "s" : "");
2884 }
2885
2886 if (update_index) {
2887 if (write_cache(newfd, active_cache, active_nr) ||
2888 close(newfd) || commit_locked_index(&lock_file))
2889 die("Unable to write new index file");
2890 }
2891
2892 return !!errs;
2893 }