ce_match_stat, run_diff_files: use symbolic constants for readability
[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 unsigned long 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 write_name_quoted(NULL, 0, patch->old_name, 1, output);
167 fputs(" => ", output);
168 write_name_quoted(NULL, 0, patch->new_name, 1, output);
169 }
170 else {
171 const char *n = patch->new_name;
172 if (!n)
173 n = patch->old_name;
174 write_name_quoted(NULL, 0, n, 1, output);
175 }
176 fputs(post, output);
177 }
178
179 #define CHUNKSIZE (8192)
180 #define SLOP (16)
181
182 static void *read_patch_file(int fd, unsigned long *sizep)
183 {
184 unsigned long size = 0, alloc = CHUNKSIZE;
185 void *buffer = xmalloc(alloc);
186
187 for (;;) {
188 ssize_t nr = alloc - size;
189 if (nr < 1024) {
190 alloc += CHUNKSIZE;
191 buffer = xrealloc(buffer, alloc);
192 nr = alloc - size;
193 }
194 nr = xread(fd, (char *) buffer + size, nr);
195 if (!nr)
196 break;
197 if (nr < 0)
198 die("git-apply: read returned %s", strerror(errno));
199 size += nr;
200 }
201 *sizep = size;
202
203 /*
204 * Make sure that we have some slop in the buffer
205 * so that we can do speculative "memcmp" etc, and
206 * see to it that it is NUL-filled.
207 */
208 if (alloc < size + SLOP)
209 buffer = xrealloc(buffer, size + SLOP);
210 memset((char *) buffer + size, 0, SLOP);
211 return buffer;
212 }
213
214 static unsigned long linelen(const char *buffer, unsigned long size)
215 {
216 unsigned long len = 0;
217 while (size--) {
218 len++;
219 if (*buffer++ == '\n')
220 break;
221 }
222 return len;
223 }
224
225 static int is_dev_null(const char *str)
226 {
227 return !memcmp("/dev/null", str, 9) && isspace(str[9]);
228 }
229
230 #define TERM_SPACE 1
231 #define TERM_TAB 2
232
233 static int name_terminate(const char *name, int namelen, int c, int terminate)
234 {
235 if (c == ' ' && !(terminate & TERM_SPACE))
236 return 0;
237 if (c == '\t' && !(terminate & TERM_TAB))
238 return 0;
239
240 return 1;
241 }
242
243 static char *find_name(const char *line, char *def, int p_value, int terminate)
244 {
245 int len;
246 const char *start = line;
247 char *name;
248
249 if (*line == '"') {
250 /* Proposed "new-style" GNU patch/diff format; see
251 * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
252 */
253 name = unquote_c_style(line, NULL);
254 if (name) {
255 char *cp = name;
256 while (p_value) {
257 cp = strchr(cp, '/');
258 if (!cp)
259 break;
260 cp++;
261 p_value--;
262 }
263 if (cp) {
264 /* name can later be freed, so we need
265 * to memmove, not just return cp
266 */
267 memmove(name, cp, strlen(cp) + 1);
268 free(def);
269 return name;
270 }
271 else {
272 free(name);
273 name = NULL;
274 }
275 }
276 }
277
278 for (;;) {
279 char c = *line;
280
281 if (isspace(c)) {
282 if (c == '\n')
283 break;
284 if (name_terminate(start, line-start, c, terminate))
285 break;
286 }
287 line++;
288 if (c == '/' && !--p_value)
289 start = line;
290 }
291 if (!start)
292 return def;
293 len = line - start;
294 if (!len)
295 return def;
296
297 /*
298 * Generally we prefer the shorter name, especially
299 * if the other one is just a variation of that with
300 * something else tacked on to the end (ie "file.orig"
301 * or "file~").
302 */
303 if (def) {
304 int deflen = strlen(def);
305 if (deflen < len && !strncmp(start, def, deflen))
306 return def;
307 }
308
309 name = xmalloc(len + 1);
310 memcpy(name, start, len);
311 name[len] = 0;
312 free(def);
313 return name;
314 }
315
316 static int count_slashes(const char *cp)
317 {
318 int cnt = 0;
319 char ch;
320
321 while ((ch = *cp++))
322 if (ch == '/')
323 cnt++;
324 return cnt;
325 }
326
327 /*
328 * Given the string after "--- " or "+++ ", guess the appropriate
329 * p_value for the given patch.
330 */
331 static int guess_p_value(const char *nameline)
332 {
333 char *name, *cp;
334 int val = -1;
335
336 if (is_dev_null(nameline))
337 return -1;
338 name = find_name(nameline, NULL, 0, TERM_SPACE | TERM_TAB);
339 if (!name)
340 return -1;
341 cp = strchr(name, '/');
342 if (!cp)
343 val = 0;
344 else if (prefix) {
345 /*
346 * Does it begin with "a/$our-prefix" and such? Then this is
347 * very likely to apply to our directory.
348 */
349 if (!strncmp(name, prefix, prefix_length))
350 val = count_slashes(prefix);
351 else {
352 cp++;
353 if (!strncmp(cp, prefix, prefix_length))
354 val = count_slashes(prefix) + 1;
355 }
356 }
357 free(name);
358 return val;
359 }
360
361 /*
362 * Get the name etc info from the --/+++ lines of a traditional patch header
363 *
364 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
365 * files, we can happily check the index for a match, but for creating a
366 * new file we should try to match whatever "patch" does. I have no idea.
367 */
368 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
369 {
370 char *name;
371
372 first += 4; /* skip "--- " */
373 second += 4; /* skip "+++ " */
374 if (!p_value_known) {
375 int p, q;
376 p = guess_p_value(first);
377 q = guess_p_value(second);
378 if (p < 0) p = q;
379 if (0 <= p && p == q) {
380 p_value = p;
381 p_value_known = 1;
382 }
383 }
384 if (is_dev_null(first)) {
385 patch->is_new = 1;
386 patch->is_delete = 0;
387 name = find_name(second, NULL, p_value, TERM_SPACE | TERM_TAB);
388 patch->new_name = name;
389 } else if (is_dev_null(second)) {
390 patch->is_new = 0;
391 patch->is_delete = 1;
392 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
393 patch->old_name = name;
394 } else {
395 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
396 name = find_name(second, name, p_value, TERM_SPACE | TERM_TAB);
397 patch->old_name = patch->new_name = name;
398 }
399 if (!name)
400 die("unable to find filename in patch at line %d", linenr);
401 }
402
403 static int gitdiff_hdrend(const char *line, struct patch *patch)
404 {
405 return -1;
406 }
407
408 /*
409 * We're anal about diff header consistency, to make
410 * sure that we don't end up having strange ambiguous
411 * patches floating around.
412 *
413 * As a result, gitdiff_{old|new}name() will check
414 * their names against any previous information, just
415 * to make sure..
416 */
417 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
418 {
419 if (!orig_name && !isnull)
420 return find_name(line, NULL, p_value, TERM_TAB);
421
422 if (orig_name) {
423 int len;
424 const char *name;
425 char *another;
426 name = orig_name;
427 len = strlen(name);
428 if (isnull)
429 die("git-apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
430 another = find_name(line, NULL, p_value, TERM_TAB);
431 if (!another || memcmp(another, name, len))
432 die("git-apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
433 free(another);
434 return orig_name;
435 }
436 else {
437 /* expect "/dev/null" */
438 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
439 die("git-apply: bad git-diff - expected /dev/null on line %d", linenr);
440 return NULL;
441 }
442 }
443
444 static int gitdiff_oldname(const char *line, struct patch *patch)
445 {
446 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
447 return 0;
448 }
449
450 static int gitdiff_newname(const char *line, struct patch *patch)
451 {
452 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
453 return 0;
454 }
455
456 static int gitdiff_oldmode(const char *line, struct patch *patch)
457 {
458 patch->old_mode = strtoul(line, NULL, 8);
459 return 0;
460 }
461
462 static int gitdiff_newmode(const char *line, struct patch *patch)
463 {
464 patch->new_mode = strtoul(line, NULL, 8);
465 return 0;
466 }
467
468 static int gitdiff_delete(const char *line, struct patch *patch)
469 {
470 patch->is_delete = 1;
471 patch->old_name = patch->def_name;
472 return gitdiff_oldmode(line, patch);
473 }
474
475 static int gitdiff_newfile(const char *line, struct patch *patch)
476 {
477 patch->is_new = 1;
478 patch->new_name = patch->def_name;
479 return gitdiff_newmode(line, patch);
480 }
481
482 static int gitdiff_copysrc(const char *line, struct patch *patch)
483 {
484 patch->is_copy = 1;
485 patch->old_name = find_name(line, NULL, 0, 0);
486 return 0;
487 }
488
489 static int gitdiff_copydst(const char *line, struct patch *patch)
490 {
491 patch->is_copy = 1;
492 patch->new_name = find_name(line, NULL, 0, 0);
493 return 0;
494 }
495
496 static int gitdiff_renamesrc(const char *line, struct patch *patch)
497 {
498 patch->is_rename = 1;
499 patch->old_name = find_name(line, NULL, 0, 0);
500 return 0;
501 }
502
503 static int gitdiff_renamedst(const char *line, struct patch *patch)
504 {
505 patch->is_rename = 1;
506 patch->new_name = find_name(line, NULL, 0, 0);
507 return 0;
508 }
509
510 static int gitdiff_similarity(const char *line, struct patch *patch)
511 {
512 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
513 patch->score = 0;
514 return 0;
515 }
516
517 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
518 {
519 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
520 patch->score = 0;
521 return 0;
522 }
523
524 static int gitdiff_index(const char *line, struct patch *patch)
525 {
526 /* index line is N hexadecimal, "..", N hexadecimal,
527 * and optional space with octal mode.
528 */
529 const char *ptr, *eol;
530 int len;
531
532 ptr = strchr(line, '.');
533 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
534 return 0;
535 len = ptr - line;
536 memcpy(patch->old_sha1_prefix, line, len);
537 patch->old_sha1_prefix[len] = 0;
538
539 line = ptr + 2;
540 ptr = strchr(line, ' ');
541 eol = strchr(line, '\n');
542
543 if (!ptr || eol < ptr)
544 ptr = eol;
545 len = ptr - line;
546
547 if (40 < len)
548 return 0;
549 memcpy(patch->new_sha1_prefix, line, len);
550 patch->new_sha1_prefix[len] = 0;
551 if (*ptr == ' ')
552 patch->new_mode = patch->old_mode = strtoul(ptr+1, NULL, 8);
553 return 0;
554 }
555
556 /*
557 * This is normal for a diff that doesn't change anything: we'll fall through
558 * into the next diff. Tell the parser to break out.
559 */
560 static int gitdiff_unrecognized(const char *line, struct patch *patch)
561 {
562 return -1;
563 }
564
565 static const char *stop_at_slash(const char *line, int llen)
566 {
567 int i;
568
569 for (i = 0; i < llen; i++) {
570 int ch = line[i];
571 if (ch == '/')
572 return line + i;
573 }
574 return NULL;
575 }
576
577 /* This is to extract the same name that appears on "diff --git"
578 * line. We do not find and return anything if it is a rename
579 * patch, and it is OK because we will find the name elsewhere.
580 * We need to reliably find name only when it is mode-change only,
581 * creation or deletion of an empty file. In any of these cases,
582 * both sides are the same name under a/ and b/ respectively.
583 */
584 static char *git_header_name(char *line, int llen)
585 {
586 int len;
587 const char *name;
588 const char *second = NULL;
589
590 line += strlen("diff --git ");
591 llen -= strlen("diff --git ");
592
593 if (*line == '"') {
594 const char *cp;
595 char *first = unquote_c_style(line, &second);
596 if (!first)
597 return NULL;
598
599 /* advance to the first slash */
600 cp = stop_at_slash(first, strlen(first));
601 if (!cp || cp == first) {
602 /* we do not accept absolute paths */
603 free_first_and_fail:
604 free(first);
605 return NULL;
606 }
607 len = strlen(cp+1);
608 memmove(first, cp+1, len+1); /* including NUL */
609
610 /* second points at one past closing dq of name.
611 * find the second name.
612 */
613 while ((second < line + llen) && isspace(*second))
614 second++;
615
616 if (line + llen <= second)
617 goto free_first_and_fail;
618 if (*second == '"') {
619 char *sp = unquote_c_style(second, NULL);
620 if (!sp)
621 goto free_first_and_fail;
622 cp = stop_at_slash(sp, strlen(sp));
623 if (!cp || cp == sp) {
624 free_both_and_fail:
625 free(sp);
626 goto free_first_and_fail;
627 }
628 /* They must match, otherwise ignore */
629 if (strcmp(cp+1, first))
630 goto free_both_and_fail;
631 free(sp);
632 return first;
633 }
634
635 /* unquoted second */
636 cp = stop_at_slash(second, line + llen - second);
637 if (!cp || cp == second)
638 goto free_first_and_fail;
639 cp++;
640 if (line + llen - cp != len + 1 ||
641 memcmp(first, cp, len))
642 goto free_first_and_fail;
643 return first;
644 }
645
646 /* unquoted first name */
647 name = stop_at_slash(line, llen);
648 if (!name || name == line)
649 return NULL;
650
651 name++;
652
653 /* since the first name is unquoted, a dq if exists must be
654 * the beginning of the second name.
655 */
656 for (second = name; second < line + llen; second++) {
657 if (*second == '"') {
658 const char *cp = second;
659 const char *np;
660 char *sp = unquote_c_style(second, NULL);
661
662 if (!sp)
663 return NULL;
664 np = stop_at_slash(sp, strlen(sp));
665 if (!np || np == sp) {
666 free_second_and_fail:
667 free(sp);
668 return NULL;
669 }
670 np++;
671 len = strlen(np);
672 if (len < cp - name &&
673 !strncmp(np, name, len) &&
674 isspace(name[len])) {
675 /* Good */
676 memmove(sp, np, len + 1);
677 return sp;
678 }
679 goto free_second_and_fail;
680 }
681 }
682
683 /*
684 * Accept a name only if it shows up twice, exactly the same
685 * form.
686 */
687 for (len = 0 ; ; len++) {
688 switch (name[len]) {
689 default:
690 continue;
691 case '\n':
692 return NULL;
693 case '\t': case ' ':
694 second = name+len;
695 for (;;) {
696 char c = *second++;
697 if (c == '\n')
698 return NULL;
699 if (c == '/')
700 break;
701 }
702 if (second[len] == '\n' && !memcmp(name, second, len)) {
703 char *ret = xmalloc(len + 1);
704 memcpy(ret, name, len);
705 ret[len] = 0;
706 return ret;
707 }
708 }
709 }
710 return NULL;
711 }
712
713 /* Verify that we recognize the lines following a git header */
714 static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
715 {
716 unsigned long offset;
717
718 /* A git diff has explicit new/delete information, so we don't guess */
719 patch->is_new = 0;
720 patch->is_delete = 0;
721
722 /*
723 * Some things may not have the old name in the
724 * rest of the headers anywhere (pure mode changes,
725 * or removing or adding empty files), so we get
726 * the default name from the header.
727 */
728 patch->def_name = git_header_name(line, len);
729
730 line += len;
731 size -= len;
732 linenr++;
733 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
734 static const struct opentry {
735 const char *str;
736 int (*fn)(const char *, struct patch *);
737 } optable[] = {
738 { "@@ -", gitdiff_hdrend },
739 { "--- ", gitdiff_oldname },
740 { "+++ ", gitdiff_newname },
741 { "old mode ", gitdiff_oldmode },
742 { "new mode ", gitdiff_newmode },
743 { "deleted file mode ", gitdiff_delete },
744 { "new file mode ", gitdiff_newfile },
745 { "copy from ", gitdiff_copysrc },
746 { "copy to ", gitdiff_copydst },
747 { "rename old ", gitdiff_renamesrc },
748 { "rename new ", gitdiff_renamedst },
749 { "rename from ", gitdiff_renamesrc },
750 { "rename to ", gitdiff_renamedst },
751 { "similarity index ", gitdiff_similarity },
752 { "dissimilarity index ", gitdiff_dissimilarity },
753 { "index ", gitdiff_index },
754 { "", gitdiff_unrecognized },
755 };
756 int i;
757
758 len = linelen(line, size);
759 if (!len || line[len-1] != '\n')
760 break;
761 for (i = 0; i < ARRAY_SIZE(optable); i++) {
762 const struct opentry *p = optable + i;
763 int oplen = strlen(p->str);
764 if (len < oplen || memcmp(p->str, line, oplen))
765 continue;
766 if (p->fn(line + oplen, patch) < 0)
767 return offset;
768 break;
769 }
770 }
771
772 return offset;
773 }
774
775 static int parse_num(const char *line, unsigned long *p)
776 {
777 char *ptr;
778
779 if (!isdigit(*line))
780 return 0;
781 *p = strtoul(line, &ptr, 10);
782 return ptr - line;
783 }
784
785 static int parse_range(const char *line, int len, int offset, const char *expect,
786 unsigned long *p1, unsigned long *p2)
787 {
788 int digits, ex;
789
790 if (offset < 0 || offset >= len)
791 return -1;
792 line += offset;
793 len -= offset;
794
795 digits = parse_num(line, p1);
796 if (!digits)
797 return -1;
798
799 offset += digits;
800 line += digits;
801 len -= digits;
802
803 *p2 = 1;
804 if (*line == ',') {
805 digits = parse_num(line+1, p2);
806 if (!digits)
807 return -1;
808
809 offset += digits+1;
810 line += digits+1;
811 len -= digits+1;
812 }
813
814 ex = strlen(expect);
815 if (ex > len)
816 return -1;
817 if (memcmp(line, expect, ex))
818 return -1;
819
820 return offset + ex;
821 }
822
823 /*
824 * Parse a unified diff fragment header of the
825 * form "@@ -a,b +c,d @@"
826 */
827 static int parse_fragment_header(char *line, int len, struct fragment *fragment)
828 {
829 int offset;
830
831 if (!len || line[len-1] != '\n')
832 return -1;
833
834 /* Figure out the number of lines in a fragment */
835 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
836 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
837
838 return offset;
839 }
840
841 static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
842 {
843 unsigned long offset, len;
844
845 patch->is_toplevel_relative = 0;
846 patch->is_rename = patch->is_copy = 0;
847 patch->is_new = patch->is_delete = -1;
848 patch->old_mode = patch->new_mode = 0;
849 patch->old_name = patch->new_name = NULL;
850 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
851 unsigned long nextlen;
852
853 len = linelen(line, size);
854 if (!len)
855 break;
856
857 /* Testing this early allows us to take a few shortcuts.. */
858 if (len < 6)
859 continue;
860
861 /*
862 * Make sure we don't find any unconnected patch fragments.
863 * That's a sign that we didn't find a header, and that a
864 * patch has become corrupted/broken up.
865 */
866 if (!memcmp("@@ -", line, 4)) {
867 struct fragment dummy;
868 if (parse_fragment_header(line, len, &dummy) < 0)
869 continue;
870 die("patch fragment without header at line %d: %.*s",
871 linenr, (int)len-1, line);
872 }
873
874 if (size < len + 6)
875 break;
876
877 /*
878 * Git patch? It might not have a real patch, just a rename
879 * or mode change, so we handle that specially
880 */
881 if (!memcmp("diff --git ", line, 11)) {
882 int git_hdr_len = parse_git_header(line, len, size, patch);
883 if (git_hdr_len <= len)
884 continue;
885 if (!patch->old_name && !patch->new_name) {
886 if (!patch->def_name)
887 die("git diff header lacks filename information (line %d)", linenr);
888 patch->old_name = patch->new_name = patch->def_name;
889 }
890 patch->is_toplevel_relative = 1;
891 *hdrsize = git_hdr_len;
892 return offset;
893 }
894
895 /** --- followed by +++ ? */
896 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
897 continue;
898
899 /*
900 * We only accept unified patches, so we want it to
901 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
902 * minimum
903 */
904 nextlen = linelen(line + len, size - len);
905 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
906 continue;
907
908 /* Ok, we'll consider it a patch */
909 parse_traditional_patch(line, line+len, patch);
910 *hdrsize = len + nextlen;
911 linenr += 2;
912 return offset;
913 }
914 return -1;
915 }
916
917 static void check_whitespace(const char *line, int len)
918 {
919 const char *err = "Adds trailing whitespace";
920 int seen_space = 0;
921 int i;
922
923 /*
924 * We know len is at least two, since we have a '+' and we
925 * checked that the last character was a '\n' before calling
926 * this function. That is, an addition of an empty line would
927 * check the '+' here. Sneaky...
928 */
929 if (isspace(line[len-2]))
930 goto error;
931
932 /*
933 * Make sure that there is no space followed by a tab in
934 * indentation.
935 */
936 err = "Space in indent is followed by a tab";
937 for (i = 1; i < len; i++) {
938 if (line[i] == '\t') {
939 if (seen_space)
940 goto error;
941 }
942 else if (line[i] == ' ')
943 seen_space = 1;
944 else
945 break;
946 }
947 return;
948
949 error:
950 whitespace_error++;
951 if (squelch_whitespace_errors &&
952 squelch_whitespace_errors < whitespace_error)
953 ;
954 else
955 fprintf(stderr, "%s.\n%s:%d:%.*s\n",
956 err, patch_input_file, linenr, len-2, line+1);
957 }
958
959
960 /*
961 * Parse a unified diff. Note that this really needs to parse each
962 * fragment separately, since the only way to know the difference
963 * between a "---" that is part of a patch, and a "---" that starts
964 * the next patch is to look at the line counts..
965 */
966 static int parse_fragment(char *line, unsigned long size, struct patch *patch, struct fragment *fragment)
967 {
968 int added, deleted;
969 int len = linelen(line, size), offset;
970 unsigned long oldlines, newlines;
971 unsigned long leading, trailing;
972
973 offset = parse_fragment_header(line, len, fragment);
974 if (offset < 0)
975 return -1;
976 oldlines = fragment->oldlines;
977 newlines = fragment->newlines;
978 leading = 0;
979 trailing = 0;
980
981 /* Parse the thing.. */
982 line += len;
983 size -= len;
984 linenr++;
985 added = deleted = 0;
986 for (offset = len;
987 0 < size;
988 offset += len, size -= len, line += len, linenr++) {
989 if (!oldlines && !newlines)
990 break;
991 len = linelen(line, size);
992 if (!len || line[len-1] != '\n')
993 return -1;
994 switch (*line) {
995 default:
996 return -1;
997 case '\n': /* newer GNU diff, an empty context line */
998 case ' ':
999 oldlines--;
1000 newlines--;
1001 if (!deleted && !added)
1002 leading++;
1003 trailing++;
1004 break;
1005 case '-':
1006 if (apply_in_reverse &&
1007 new_whitespace != nowarn_whitespace)
1008 check_whitespace(line, len);
1009 deleted++;
1010 oldlines--;
1011 trailing = 0;
1012 break;
1013 case '+':
1014 if (!apply_in_reverse &&
1015 new_whitespace != nowarn_whitespace)
1016 check_whitespace(line, len);
1017 added++;
1018 newlines--;
1019 trailing = 0;
1020 break;
1021
1022 /* We allow "\ No newline at end of file". Depending
1023 * on locale settings when the patch was produced we
1024 * don't know what this line looks like. The only
1025 * thing we do know is that it begins with "\ ".
1026 * Checking for 12 is just for sanity check -- any
1027 * l10n of "\ No newline..." is at least that long.
1028 */
1029 case '\\':
1030 if (len < 12 || memcmp(line, "\\ ", 2))
1031 return -1;
1032 break;
1033 }
1034 }
1035 if (oldlines || newlines)
1036 return -1;
1037 fragment->leading = leading;
1038 fragment->trailing = trailing;
1039
1040 /* If a fragment ends with an incomplete line, we failed to include
1041 * it in the above loop because we hit oldlines == newlines == 0
1042 * before seeing it.
1043 */
1044 if (12 < size && !memcmp(line, "\\ ", 2))
1045 offset += linelen(line, size);
1046
1047 patch->lines_added += added;
1048 patch->lines_deleted += deleted;
1049
1050 if (0 < patch->is_new && oldlines)
1051 return error("new file depends on old contents");
1052 if (0 < patch->is_delete && newlines)
1053 return error("deleted file still has contents");
1054 return offset;
1055 }
1056
1057 static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
1058 {
1059 unsigned long offset = 0;
1060 unsigned long oldlines = 0, newlines = 0, context = 0;
1061 struct fragment **fragp = &patch->fragments;
1062
1063 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1064 struct fragment *fragment;
1065 int len;
1066
1067 fragment = xcalloc(1, sizeof(*fragment));
1068 len = parse_fragment(line, size, patch, fragment);
1069 if (len <= 0)
1070 die("corrupt patch at line %d", linenr);
1071 fragment->patch = line;
1072 fragment->size = len;
1073 oldlines += fragment->oldlines;
1074 newlines += fragment->newlines;
1075 context += fragment->leading + fragment->trailing;
1076
1077 *fragp = fragment;
1078 fragp = &fragment->next;
1079
1080 offset += len;
1081 line += len;
1082 size -= len;
1083 }
1084
1085 /*
1086 * If something was removed (i.e. we have old-lines) it cannot
1087 * be creation, and if something was added it cannot be
1088 * deletion. However, the reverse is not true; --unified=0
1089 * patches that only add are not necessarily creation even
1090 * though they do not have any old lines, and ones that only
1091 * delete are not necessarily deletion.
1092 *
1093 * Unfortunately, a real creation/deletion patch do _not_ have
1094 * any context line by definition, so we cannot safely tell it
1095 * apart with --unified=0 insanity. At least if the patch has
1096 * more than one hunk it is not creation or deletion.
1097 */
1098 if (patch->is_new < 0 &&
1099 (oldlines || (patch->fragments && patch->fragments->next)))
1100 patch->is_new = 0;
1101 if (patch->is_delete < 0 &&
1102 (newlines || (patch->fragments && patch->fragments->next)))
1103 patch->is_delete = 0;
1104 if (!unidiff_zero || context) {
1105 /* If the user says the patch is not generated with
1106 * --unified=0, or if we have seen context lines,
1107 * then not having oldlines means the patch is creation,
1108 * and not having newlines means the patch is deletion.
1109 */
1110 if (patch->is_new < 0 && !oldlines) {
1111 patch->is_new = 1;
1112 patch->old_name = NULL;
1113 }
1114 if (patch->is_delete < 0 && !newlines) {
1115 patch->is_delete = 1;
1116 patch->new_name = NULL;
1117 }
1118 }
1119
1120 if (0 < patch->is_new && oldlines)
1121 die("new file %s depends on old contents", patch->new_name);
1122 if (0 < patch->is_delete && newlines)
1123 die("deleted file %s still has contents", patch->old_name);
1124 if (!patch->is_delete && !newlines && context)
1125 fprintf(stderr, "** warning: file %s becomes empty but "
1126 "is not deleted\n", patch->new_name);
1127
1128 return offset;
1129 }
1130
1131 static inline int metadata_changes(struct patch *patch)
1132 {
1133 return patch->is_rename > 0 ||
1134 patch->is_copy > 0 ||
1135 patch->is_new > 0 ||
1136 patch->is_delete ||
1137 (patch->old_mode && patch->new_mode &&
1138 patch->old_mode != patch->new_mode);
1139 }
1140
1141 static char *inflate_it(const void *data, unsigned long size,
1142 unsigned long inflated_size)
1143 {
1144 z_stream stream;
1145 void *out;
1146 int st;
1147
1148 memset(&stream, 0, sizeof(stream));
1149
1150 stream.next_in = (unsigned char *)data;
1151 stream.avail_in = size;
1152 stream.next_out = out = xmalloc(inflated_size);
1153 stream.avail_out = inflated_size;
1154 inflateInit(&stream);
1155 st = inflate(&stream, Z_FINISH);
1156 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1157 free(out);
1158 return NULL;
1159 }
1160 return out;
1161 }
1162
1163 static struct fragment *parse_binary_hunk(char **buf_p,
1164 unsigned long *sz_p,
1165 int *status_p,
1166 int *used_p)
1167 {
1168 /* Expect a line that begins with binary patch method ("literal"
1169 * or "delta"), followed by the length of data before deflating.
1170 * a sequence of 'length-byte' followed by base-85 encoded data
1171 * should follow, terminated by a newline.
1172 *
1173 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1174 * and we would limit the patch line to 66 characters,
1175 * so one line can fit up to 13 groups that would decode
1176 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1177 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1178 */
1179 int llen, used;
1180 unsigned long size = *sz_p;
1181 char *buffer = *buf_p;
1182 int patch_method;
1183 unsigned long origlen;
1184 char *data = NULL;
1185 int hunk_size = 0;
1186 struct fragment *frag;
1187
1188 llen = linelen(buffer, size);
1189 used = llen;
1190
1191 *status_p = 0;
1192
1193 if (!prefixcmp(buffer, "delta ")) {
1194 patch_method = BINARY_DELTA_DEFLATED;
1195 origlen = strtoul(buffer + 6, NULL, 10);
1196 }
1197 else if (!prefixcmp(buffer, "literal ")) {
1198 patch_method = BINARY_LITERAL_DEFLATED;
1199 origlen = strtoul(buffer + 8, NULL, 10);
1200 }
1201 else
1202 return NULL;
1203
1204 linenr++;
1205 buffer += llen;
1206 while (1) {
1207 int byte_length, max_byte_length, newsize;
1208 llen = linelen(buffer, size);
1209 used += llen;
1210 linenr++;
1211 if (llen == 1) {
1212 /* consume the blank line */
1213 buffer++;
1214 size--;
1215 break;
1216 }
1217 /* Minimum line is "A00000\n" which is 7-byte long,
1218 * and the line length must be multiple of 5 plus 2.
1219 */
1220 if ((llen < 7) || (llen-2) % 5)
1221 goto corrupt;
1222 max_byte_length = (llen - 2) / 5 * 4;
1223 byte_length = *buffer;
1224 if ('A' <= byte_length && byte_length <= 'Z')
1225 byte_length = byte_length - 'A' + 1;
1226 else if ('a' <= byte_length && byte_length <= 'z')
1227 byte_length = byte_length - 'a' + 27;
1228 else
1229 goto corrupt;
1230 /* if the input length was not multiple of 4, we would
1231 * have filler at the end but the filler should never
1232 * exceed 3 bytes
1233 */
1234 if (max_byte_length < byte_length ||
1235 byte_length <= max_byte_length - 4)
1236 goto corrupt;
1237 newsize = hunk_size + byte_length;
1238 data = xrealloc(data, newsize);
1239 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1240 goto corrupt;
1241 hunk_size = newsize;
1242 buffer += llen;
1243 size -= llen;
1244 }
1245
1246 frag = xcalloc(1, sizeof(*frag));
1247 frag->patch = inflate_it(data, hunk_size, origlen);
1248 if (!frag->patch)
1249 goto corrupt;
1250 free(data);
1251 frag->size = origlen;
1252 *buf_p = buffer;
1253 *sz_p = size;
1254 *used_p = used;
1255 frag->binary_patch_method = patch_method;
1256 return frag;
1257
1258 corrupt:
1259 free(data);
1260 *status_p = -1;
1261 error("corrupt binary patch at line %d: %.*s",
1262 linenr-1, llen-1, buffer);
1263 return NULL;
1264 }
1265
1266 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1267 {
1268 /* We have read "GIT binary patch\n"; what follows is a line
1269 * that says the patch method (currently, either "literal" or
1270 * "delta") and the length of data before deflating; a
1271 * sequence of 'length-byte' followed by base-85 encoded data
1272 * follows.
1273 *
1274 * When a binary patch is reversible, there is another binary
1275 * hunk in the same format, starting with patch method (either
1276 * "literal" or "delta") with the length of data, and a sequence
1277 * of length-byte + base-85 encoded data, terminated with another
1278 * empty line. This data, when applied to the postimage, produces
1279 * the preimage.
1280 */
1281 struct fragment *forward;
1282 struct fragment *reverse;
1283 int status;
1284 int used, used_1;
1285
1286 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1287 if (!forward && !status)
1288 /* there has to be one hunk (forward hunk) */
1289 return error("unrecognized binary patch at line %d", linenr-1);
1290 if (status)
1291 /* otherwise we already gave an error message */
1292 return status;
1293
1294 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1295 if (reverse)
1296 used += used_1;
1297 else if (status) {
1298 /* not having reverse hunk is not an error, but having
1299 * a corrupt reverse hunk is.
1300 */
1301 free((void*) forward->patch);
1302 free(forward);
1303 return status;
1304 }
1305 forward->next = reverse;
1306 patch->fragments = forward;
1307 patch->is_binary = 1;
1308 return used;
1309 }
1310
1311 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1312 {
1313 int hdrsize, patchsize;
1314 int offset = find_header(buffer, size, &hdrsize, patch);
1315
1316 if (offset < 0)
1317 return offset;
1318
1319 patchsize = parse_single_patch(buffer + offset + hdrsize, size - offset - hdrsize, patch);
1320
1321 if (!patchsize) {
1322 static const char *binhdr[] = {
1323 "Binary files ",
1324 "Files ",
1325 NULL,
1326 };
1327 static const char git_binary[] = "GIT binary patch\n";
1328 int i;
1329 int hd = hdrsize + offset;
1330 unsigned long llen = linelen(buffer + hd, size - hd);
1331
1332 if (llen == sizeof(git_binary) - 1 &&
1333 !memcmp(git_binary, buffer + hd, llen)) {
1334 int used;
1335 linenr++;
1336 used = parse_binary(buffer + hd + llen,
1337 size - hd - llen, patch);
1338 if (used)
1339 patchsize = used + llen;
1340 else
1341 patchsize = 0;
1342 }
1343 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1344 for (i = 0; binhdr[i]; i++) {
1345 int len = strlen(binhdr[i]);
1346 if (len < size - hd &&
1347 !memcmp(binhdr[i], buffer + hd, len)) {
1348 linenr++;
1349 patch->is_binary = 1;
1350 patchsize = llen;
1351 break;
1352 }
1353 }
1354 }
1355
1356 /* Empty patch cannot be applied if it is a text patch
1357 * without metadata change. A binary patch appears
1358 * empty to us here.
1359 */
1360 if ((apply || check) &&
1361 (!patch->is_binary && !metadata_changes(patch)))
1362 die("patch with only garbage at line %d", linenr);
1363 }
1364
1365 return offset + hdrsize + patchsize;
1366 }
1367
1368 #define swap(a,b) myswap((a),(b),sizeof(a))
1369
1370 #define myswap(a, b, size) do { \
1371 unsigned char mytmp[size]; \
1372 memcpy(mytmp, &a, size); \
1373 memcpy(&a, &b, size); \
1374 memcpy(&b, mytmp, size); \
1375 } while (0)
1376
1377 static void reverse_patches(struct patch *p)
1378 {
1379 for (; p; p = p->next) {
1380 struct fragment *frag = p->fragments;
1381
1382 swap(p->new_name, p->old_name);
1383 swap(p->new_mode, p->old_mode);
1384 swap(p->is_new, p->is_delete);
1385 swap(p->lines_added, p->lines_deleted);
1386 swap(p->old_sha1_prefix, p->new_sha1_prefix);
1387
1388 for (; frag; frag = frag->next) {
1389 swap(frag->newpos, frag->oldpos);
1390 swap(frag->newlines, frag->oldlines);
1391 }
1392 }
1393 }
1394
1395 static const char pluses[] = "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1396 static const char minuses[]= "----------------------------------------------------------------------";
1397
1398 static void show_stats(struct patch *patch)
1399 {
1400 const char *prefix = "";
1401 char *name = patch->new_name;
1402 char *qname = NULL;
1403 int len, max, add, del, total;
1404
1405 if (!name)
1406 name = patch->old_name;
1407
1408 if (0 < (len = quote_c_style(name, NULL, NULL, 0))) {
1409 qname = xmalloc(len + 1);
1410 quote_c_style(name, qname, NULL, 0);
1411 name = qname;
1412 }
1413
1414 /*
1415 * "scale" the filename
1416 */
1417 len = strlen(name);
1418 max = max_len;
1419 if (max > 50)
1420 max = 50;
1421 if (len > max) {
1422 char *slash;
1423 prefix = "...";
1424 max -= 3;
1425 name += len - max;
1426 slash = strchr(name, '/');
1427 if (slash)
1428 name = slash;
1429 }
1430 len = max;
1431
1432 /*
1433 * scale the add/delete
1434 */
1435 max = max_change;
1436 if (max + len > 70)
1437 max = 70 - len;
1438
1439 add = patch->lines_added;
1440 del = patch->lines_deleted;
1441 total = add + del;
1442
1443 if (max_change > 0) {
1444 total = (total * max + max_change / 2) / max_change;
1445 add = (add * max + max_change / 2) / max_change;
1446 del = total - add;
1447 }
1448 if (patch->is_binary)
1449 printf(" %s%-*s | Bin\n", prefix, len, name);
1450 else
1451 printf(" %s%-*s |%5d %.*s%.*s\n", prefix,
1452 len, name, patch->lines_added + patch->lines_deleted,
1453 add, pluses, del, minuses);
1454 free(qname);
1455 }
1456
1457 static int read_old_data(struct stat *st, const char *path, char **buf_p, unsigned long *alloc_p, unsigned long *size_p)
1458 {
1459 int fd;
1460 unsigned long got;
1461 unsigned long nsize;
1462 char *nbuf;
1463 unsigned long size = *size_p;
1464 char *buf = *buf_p;
1465
1466 switch (st->st_mode & S_IFMT) {
1467 case S_IFLNK:
1468 return readlink(path, buf, size) != size;
1469 case S_IFREG:
1470 fd = open(path, O_RDONLY);
1471 if (fd < 0)
1472 return error("unable to open %s", path);
1473 got = 0;
1474 for (;;) {
1475 ssize_t ret = xread(fd, buf + got, size - got);
1476 if (ret <= 0)
1477 break;
1478 got += ret;
1479 }
1480 close(fd);
1481 nsize = got;
1482 nbuf = convert_to_git(path, buf, &nsize);
1483 if (nbuf) {
1484 free(buf);
1485 *buf_p = nbuf;
1486 *alloc_p = nsize;
1487 *size_p = nsize;
1488 }
1489 return got != size;
1490 default:
1491 return -1;
1492 }
1493 }
1494
1495 static int find_offset(const char *buf, unsigned long size, const char *fragment, unsigned long fragsize, int line, int *lines)
1496 {
1497 int i;
1498 unsigned long start, backwards, forwards;
1499
1500 if (fragsize > size)
1501 return -1;
1502
1503 start = 0;
1504 if (line > 1) {
1505 unsigned long offset = 0;
1506 i = line-1;
1507 while (offset + fragsize <= size) {
1508 if (buf[offset++] == '\n') {
1509 start = offset;
1510 if (!--i)
1511 break;
1512 }
1513 }
1514 }
1515
1516 /* Exact line number? */
1517 if ((start + fragsize <= size) &&
1518 !memcmp(buf + start, fragment, fragsize))
1519 return start;
1520
1521 /*
1522 * There's probably some smart way to do this, but I'll leave
1523 * that to the smart and beautiful people. I'm simple and stupid.
1524 */
1525 backwards = start;
1526 forwards = start;
1527 for (i = 0; ; i++) {
1528 unsigned long try;
1529 int n;
1530
1531 /* "backward" */
1532 if (i & 1) {
1533 if (!backwards) {
1534 if (forwards + fragsize > size)
1535 break;
1536 continue;
1537 }
1538 do {
1539 --backwards;
1540 } while (backwards && buf[backwards-1] != '\n');
1541 try = backwards;
1542 } else {
1543 while (forwards + fragsize <= size) {
1544 if (buf[forwards++] == '\n')
1545 break;
1546 }
1547 try = forwards;
1548 }
1549
1550 if (try + fragsize > size)
1551 continue;
1552 if (memcmp(buf + try, fragment, fragsize))
1553 continue;
1554 n = (i >> 1)+1;
1555 if (i & 1)
1556 n = -n;
1557 *lines = n;
1558 return try;
1559 }
1560
1561 /*
1562 * We should start searching forward and backward.
1563 */
1564 return -1;
1565 }
1566
1567 static void remove_first_line(const char **rbuf, int *rsize)
1568 {
1569 const char *buf = *rbuf;
1570 int size = *rsize;
1571 unsigned long offset;
1572 offset = 0;
1573 while (offset <= size) {
1574 if (buf[offset++] == '\n')
1575 break;
1576 }
1577 *rsize = size - offset;
1578 *rbuf = buf + offset;
1579 }
1580
1581 static void remove_last_line(const char **rbuf, int *rsize)
1582 {
1583 const char *buf = *rbuf;
1584 int size = *rsize;
1585 unsigned long offset;
1586 offset = size - 1;
1587 while (offset > 0) {
1588 if (buf[--offset] == '\n')
1589 break;
1590 }
1591 *rsize = offset + 1;
1592 }
1593
1594 struct buffer_desc {
1595 char *buffer;
1596 unsigned long size;
1597 unsigned long alloc;
1598 };
1599
1600 static int apply_line(char *output, const char *patch, int plen)
1601 {
1602 /* plen is number of bytes to be copied from patch,
1603 * starting at patch+1 (patch[0] is '+'). Typically
1604 * patch[plen] is '\n', unless this is the incomplete
1605 * last line.
1606 */
1607 int i;
1608 int add_nl_to_tail = 0;
1609 int fixed = 0;
1610 int last_tab_in_indent = -1;
1611 int last_space_in_indent = -1;
1612 int need_fix_leading_space = 0;
1613 char *buf;
1614
1615 if ((new_whitespace != strip_whitespace) || !whitespace_error ||
1616 *patch != '+') {
1617 memcpy(output, patch + 1, plen);
1618 return plen;
1619 }
1620
1621 if (1 < plen && isspace(patch[plen-1])) {
1622 if (patch[plen] == '\n')
1623 add_nl_to_tail = 1;
1624 plen--;
1625 while (0 < plen && isspace(patch[plen]))
1626 plen--;
1627 fixed = 1;
1628 }
1629
1630 for (i = 1; i < plen; i++) {
1631 char ch = patch[i];
1632 if (ch == '\t') {
1633 last_tab_in_indent = i;
1634 if (0 <= last_space_in_indent)
1635 need_fix_leading_space = 1;
1636 }
1637 else if (ch == ' ')
1638 last_space_in_indent = i;
1639 else
1640 break;
1641 }
1642
1643 buf = output;
1644 if (need_fix_leading_space) {
1645 int consecutive_spaces = 0;
1646 /* between patch[1..last_tab_in_indent] strip the
1647 * funny spaces, updating them to tab as needed.
1648 */
1649 for (i = 1; i < last_tab_in_indent; i++, plen--) {
1650 char ch = patch[i];
1651 if (ch != ' ') {
1652 consecutive_spaces = 0;
1653 *output++ = ch;
1654 } else {
1655 consecutive_spaces++;
1656 if (consecutive_spaces == 8) {
1657 *output++ = '\t';
1658 consecutive_spaces = 0;
1659 }
1660 }
1661 }
1662 fixed = 1;
1663 i = last_tab_in_indent;
1664 }
1665 else
1666 i = 1;
1667
1668 memcpy(output, patch + i, plen);
1669 if (add_nl_to_tail)
1670 output[plen++] = '\n';
1671 if (fixed)
1672 applied_after_fixing_ws++;
1673 return output + plen - buf;
1674 }
1675
1676 static int apply_one_fragment(struct buffer_desc *desc, struct fragment *frag, int inaccurate_eof)
1677 {
1678 int match_beginning, match_end;
1679 char *buf = desc->buffer;
1680 const char *patch = frag->patch;
1681 int offset, size = frag->size;
1682 char *old = xmalloc(size);
1683 char *new = xmalloc(size);
1684 const char *oldlines, *newlines;
1685 int oldsize = 0, newsize = 0;
1686 int new_blank_lines_at_end = 0;
1687 unsigned long leading, trailing;
1688 int pos, lines;
1689
1690 while (size > 0) {
1691 char first;
1692 int len = linelen(patch, size);
1693 int plen;
1694 int added_blank_line = 0;
1695
1696 if (!len)
1697 break;
1698
1699 /*
1700 * "plen" is how much of the line we should use for
1701 * the actual patch data. Normally we just remove the
1702 * first character on the line, but if the line is
1703 * followed by "\ No newline", then we also remove the
1704 * last one (which is the newline, of course).
1705 */
1706 plen = len-1;
1707 if (len < size && patch[len] == '\\')
1708 plen--;
1709 first = *patch;
1710 if (apply_in_reverse) {
1711 if (first == '-')
1712 first = '+';
1713 else if (first == '+')
1714 first = '-';
1715 }
1716
1717 switch (first) {
1718 case '\n':
1719 /* Newer GNU diff, empty context line */
1720 if (plen < 0)
1721 /* ... followed by '\No newline'; nothing */
1722 break;
1723 old[oldsize++] = '\n';
1724 new[newsize++] = '\n';
1725 break;
1726 case ' ':
1727 case '-':
1728 memcpy(old + oldsize, patch + 1, plen);
1729 oldsize += plen;
1730 if (first == '-')
1731 break;
1732 /* Fall-through for ' ' */
1733 case '+':
1734 if (first != '+' || !no_add) {
1735 int added = apply_line(new + newsize, patch,
1736 plen);
1737 newsize += added;
1738 if (first == '+' &&
1739 added == 1 && new[newsize-1] == '\n')
1740 added_blank_line = 1;
1741 }
1742 break;
1743 case '@': case '\\':
1744 /* Ignore it, we already handled it */
1745 break;
1746 default:
1747 if (apply_verbosely)
1748 error("invalid start of line: '%c'", first);
1749 return -1;
1750 }
1751 if (added_blank_line)
1752 new_blank_lines_at_end++;
1753 else
1754 new_blank_lines_at_end = 0;
1755 patch += len;
1756 size -= len;
1757 }
1758
1759 if (inaccurate_eof && oldsize > 0 && old[oldsize - 1] == '\n' &&
1760 newsize > 0 && new[newsize - 1] == '\n') {
1761 oldsize--;
1762 newsize--;
1763 }
1764
1765 oldlines = old;
1766 newlines = new;
1767 leading = frag->leading;
1768 trailing = frag->trailing;
1769
1770 /*
1771 * If we don't have any leading/trailing data in the patch,
1772 * we want it to match at the beginning/end of the file.
1773 *
1774 * But that would break if the patch is generated with
1775 * --unified=0; sane people wouldn't do that to cause us
1776 * trouble, but we try to please not so sane ones as well.
1777 */
1778 if (unidiff_zero) {
1779 match_beginning = (!leading && !frag->oldpos);
1780 match_end = 0;
1781 }
1782 else {
1783 match_beginning = !leading && (frag->oldpos == 1);
1784 match_end = !trailing;
1785 }
1786
1787 lines = 0;
1788 pos = frag->newpos;
1789 for (;;) {
1790 offset = find_offset(buf, desc->size,
1791 oldlines, oldsize, pos, &lines);
1792 if (match_end && offset + oldsize != desc->size)
1793 offset = -1;
1794 if (match_beginning && offset)
1795 offset = -1;
1796 if (offset >= 0) {
1797 int diff;
1798 unsigned long size, alloc;
1799
1800 if (new_whitespace == strip_whitespace &&
1801 (desc->size - oldsize - offset == 0)) /* end of file? */
1802 newsize -= new_blank_lines_at_end;
1803
1804 diff = newsize - oldsize;
1805 size = desc->size + diff;
1806 alloc = desc->alloc;
1807
1808 /* Warn if it was necessary to reduce the number
1809 * of context lines.
1810 */
1811 if ((leading != frag->leading) ||
1812 (trailing != frag->trailing))
1813 fprintf(stderr, "Context reduced to (%ld/%ld)"
1814 " to apply fragment at %d\n",
1815 leading, trailing, pos + lines);
1816
1817 if (size > alloc) {
1818 alloc = size + 8192;
1819 desc->alloc = alloc;
1820 buf = xrealloc(buf, alloc);
1821 desc->buffer = buf;
1822 }
1823 desc->size = size;
1824 memmove(buf + offset + newsize,
1825 buf + offset + oldsize,
1826 size - offset - newsize);
1827 memcpy(buf + offset, newlines, newsize);
1828 offset = 0;
1829
1830 break;
1831 }
1832
1833 /* Am I at my context limits? */
1834 if ((leading <= p_context) && (trailing <= p_context))
1835 break;
1836 if (match_beginning || match_end) {
1837 match_beginning = match_end = 0;
1838 continue;
1839 }
1840 /* Reduce the number of context lines
1841 * Reduce both leading and trailing if they are equal
1842 * otherwise just reduce the larger context.
1843 */
1844 if (leading >= trailing) {
1845 remove_first_line(&oldlines, &oldsize);
1846 remove_first_line(&newlines, &newsize);
1847 pos--;
1848 leading--;
1849 }
1850 if (trailing > leading) {
1851 remove_last_line(&oldlines, &oldsize);
1852 remove_last_line(&newlines, &newsize);
1853 trailing--;
1854 }
1855 }
1856
1857 if (offset && apply_verbosely)
1858 error("while searching for:\n%.*s", oldsize, oldlines);
1859
1860 free(old);
1861 free(new);
1862 return offset;
1863 }
1864
1865 static int apply_binary_fragment(struct buffer_desc *desc, struct patch *patch)
1866 {
1867 unsigned long dst_size;
1868 struct fragment *fragment = patch->fragments;
1869 void *data;
1870 void *result;
1871
1872 /* Binary patch is irreversible without the optional second hunk */
1873 if (apply_in_reverse) {
1874 if (!fragment->next)
1875 return error("cannot reverse-apply a binary patch "
1876 "without the reverse hunk to '%s'",
1877 patch->new_name
1878 ? patch->new_name : patch->old_name);
1879 fragment = fragment->next;
1880 }
1881 data = (void*) fragment->patch;
1882 switch (fragment->binary_patch_method) {
1883 case BINARY_DELTA_DEFLATED:
1884 result = patch_delta(desc->buffer, desc->size,
1885 data,
1886 fragment->size,
1887 &dst_size);
1888 free(desc->buffer);
1889 desc->buffer = result;
1890 break;
1891 case BINARY_LITERAL_DEFLATED:
1892 free(desc->buffer);
1893 desc->buffer = data;
1894 dst_size = fragment->size;
1895 break;
1896 }
1897 if (!desc->buffer)
1898 return -1;
1899 desc->size = desc->alloc = dst_size;
1900 return 0;
1901 }
1902
1903 static int apply_binary(struct buffer_desc *desc, struct patch *patch)
1904 {
1905 const char *name = patch->old_name ? patch->old_name : patch->new_name;
1906 unsigned char sha1[20];
1907
1908 /* For safety, we require patch index line to contain
1909 * full 40-byte textual SHA1 for old and new, at least for now.
1910 */
1911 if (strlen(patch->old_sha1_prefix) != 40 ||
1912 strlen(patch->new_sha1_prefix) != 40 ||
1913 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
1914 get_sha1_hex(patch->new_sha1_prefix, sha1))
1915 return error("cannot apply binary patch to '%s' "
1916 "without full index line", name);
1917
1918 if (patch->old_name) {
1919 /* See if the old one matches what the patch
1920 * applies to.
1921 */
1922 hash_sha1_file(desc->buffer, desc->size, blob_type, sha1);
1923 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
1924 return error("the patch applies to '%s' (%s), "
1925 "which does not match the "
1926 "current contents.",
1927 name, sha1_to_hex(sha1));
1928 }
1929 else {
1930 /* Otherwise, the old one must be empty. */
1931 if (desc->size)
1932 return error("the patch applies to an empty "
1933 "'%s' but it is not empty", name);
1934 }
1935
1936 get_sha1_hex(patch->new_sha1_prefix, sha1);
1937 if (is_null_sha1(sha1)) {
1938 free(desc->buffer);
1939 desc->alloc = desc->size = 0;
1940 desc->buffer = NULL;
1941 return 0; /* deletion patch */
1942 }
1943
1944 if (has_sha1_file(sha1)) {
1945 /* We already have the postimage */
1946 enum object_type type;
1947 unsigned long size;
1948
1949 free(desc->buffer);
1950 desc->buffer = read_sha1_file(sha1, &type, &size);
1951 if (!desc->buffer)
1952 return error("the necessary postimage %s for "
1953 "'%s' cannot be read",
1954 patch->new_sha1_prefix, name);
1955 desc->alloc = desc->size = size;
1956 }
1957 else {
1958 /* We have verified desc matches the preimage;
1959 * apply the patch data to it, which is stored
1960 * in the patch->fragments->{patch,size}.
1961 */
1962 if (apply_binary_fragment(desc, patch))
1963 return error("binary patch does not apply to '%s'",
1964 name);
1965
1966 /* verify that the result matches */
1967 hash_sha1_file(desc->buffer, desc->size, blob_type, sha1);
1968 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
1969 return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)", name, patch->new_sha1_prefix, sha1_to_hex(sha1));
1970 }
1971
1972 return 0;
1973 }
1974
1975 static int apply_fragments(struct buffer_desc *desc, struct patch *patch)
1976 {
1977 struct fragment *frag = patch->fragments;
1978 const char *name = patch->old_name ? patch->old_name : patch->new_name;
1979
1980 if (patch->is_binary)
1981 return apply_binary(desc, patch);
1982
1983 while (frag) {
1984 if (apply_one_fragment(desc, frag, patch->inaccurate_eof)) {
1985 error("patch failed: %s:%ld", name, frag->oldpos);
1986 if (!apply_with_reject)
1987 return -1;
1988 frag->rejected = 1;
1989 }
1990 frag = frag->next;
1991 }
1992 return 0;
1993 }
1994
1995 static int read_file_or_gitlink(struct cache_entry *ce, char **buf_p,
1996 unsigned long *size_p)
1997 {
1998 if (!ce)
1999 return 0;
2000
2001 if (S_ISGITLINK(ntohl(ce->ce_mode))) {
2002 *buf_p = xmalloc(100);
2003 *size_p = snprintf(*buf_p, 100,
2004 "Subproject commit %s\n", sha1_to_hex(ce->sha1));
2005 } else {
2006 enum object_type type;
2007 *buf_p = read_sha1_file(ce->sha1, &type, size_p);
2008 if (!*buf_p)
2009 return -1;
2010 }
2011 return 0;
2012 }
2013
2014 static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
2015 {
2016 char *buf;
2017 unsigned long size, alloc;
2018 struct buffer_desc desc;
2019
2020 size = 0;
2021 alloc = 0;
2022 buf = NULL;
2023 if (cached) {
2024 if (read_file_or_gitlink(ce, &buf, &size))
2025 return error("read of %s failed", patch->old_name);
2026 alloc = size;
2027 } else if (patch->old_name) {
2028 if (S_ISGITLINK(patch->old_mode)) {
2029 if (ce)
2030 read_file_or_gitlink(ce, &buf, &size);
2031 else {
2032 /*
2033 * There is no way to apply subproject
2034 * patch without looking at the index.
2035 */
2036 patch->fragments = NULL;
2037 size = 0;
2038 }
2039 }
2040 else {
2041 size = xsize_t(st->st_size);
2042 alloc = size + 8192;
2043 buf = xmalloc(alloc);
2044 if (read_old_data(st, patch->old_name,
2045 &buf, &alloc, &size))
2046 return error("read of %s failed",
2047 patch->old_name);
2048 }
2049 }
2050
2051 desc.size = size;
2052 desc.alloc = alloc;
2053 desc.buffer = buf;
2054
2055 if (apply_fragments(&desc, patch) < 0)
2056 return -1; /* note with --reject this succeeds. */
2057
2058 /* NUL terminate the result */
2059 if (desc.alloc <= desc.size)
2060 desc.buffer = xrealloc(desc.buffer, desc.size + 1);
2061 desc.buffer[desc.size] = 0;
2062
2063 patch->result = desc.buffer;
2064 patch->resultsize = desc.size;
2065
2066 if (0 < patch->is_delete && patch->resultsize)
2067 return error("removal patch leaves file contents");
2068
2069 return 0;
2070 }
2071
2072 static int check_to_create_blob(const char *new_name, int ok_if_exists)
2073 {
2074 struct stat nst;
2075 if (!lstat(new_name, &nst)) {
2076 if (S_ISDIR(nst.st_mode) || ok_if_exists)
2077 return 0;
2078 /*
2079 * A leading component of new_name might be a symlink
2080 * that is going to be removed with this patch, but
2081 * still pointing at somewhere that has the path.
2082 * In such a case, path "new_name" does not exist as
2083 * far as git is concerned.
2084 */
2085 if (has_symlink_leading_path(new_name, NULL))
2086 return 0;
2087
2088 return error("%s: already exists in working directory", new_name);
2089 }
2090 else if ((errno != ENOENT) && (errno != ENOTDIR))
2091 return error("%s: %s", new_name, strerror(errno));
2092 return 0;
2093 }
2094
2095 static int verify_index_match(struct cache_entry *ce, struct stat *st)
2096 {
2097 if (S_ISGITLINK(ntohl(ce->ce_mode))) {
2098 if (!S_ISDIR(st->st_mode))
2099 return -1;
2100 return 0;
2101 }
2102 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID);
2103 }
2104
2105 static int check_patch(struct patch *patch, struct patch *prev_patch)
2106 {
2107 struct stat st;
2108 const char *old_name = patch->old_name;
2109 const char *new_name = patch->new_name;
2110 const char *name = old_name ? old_name : new_name;
2111 struct cache_entry *ce = NULL;
2112 int ok_if_exists;
2113
2114 patch->rejected = 1; /* we will drop this after we succeed */
2115
2116 /*
2117 * Make sure that we do not have local modifications from the
2118 * index when we are looking at the index. Also make sure
2119 * we have the preimage file to be patched in the work tree,
2120 * unless --cached, which tells git to apply only in the index.
2121 */
2122 if (old_name) {
2123 int stat_ret = 0;
2124 unsigned st_mode = 0;
2125
2126 if (!cached)
2127 stat_ret = lstat(old_name, &st);
2128 if (check_index) {
2129 int pos = cache_name_pos(old_name, strlen(old_name));
2130 if (pos < 0)
2131 return error("%s: does not exist in index",
2132 old_name);
2133 ce = active_cache[pos];
2134 if (stat_ret < 0) {
2135 struct checkout costate;
2136 if (errno != ENOENT)
2137 return error("%s: %s", old_name,
2138 strerror(errno));
2139 /* checkout */
2140 costate.base_dir = "";
2141 costate.base_dir_len = 0;
2142 costate.force = 0;
2143 costate.quiet = 0;
2144 costate.not_new = 0;
2145 costate.refresh_cache = 1;
2146 if (checkout_entry(ce,
2147 &costate,
2148 NULL) ||
2149 lstat(old_name, &st))
2150 return -1;
2151 }
2152 if (!cached && verify_index_match(ce, &st))
2153 return error("%s: does not match index",
2154 old_name);
2155 if (cached)
2156 st_mode = ntohl(ce->ce_mode);
2157 } else if (stat_ret < 0)
2158 return error("%s: %s", old_name, strerror(errno));
2159
2160 if (!cached)
2161 st_mode = ntohl(ce_mode_from_stat(ce, st.st_mode));
2162
2163 if (patch->is_new < 0)
2164 patch->is_new = 0;
2165 if (!patch->old_mode)
2166 patch->old_mode = st_mode;
2167 if ((st_mode ^ patch->old_mode) & S_IFMT)
2168 return error("%s: wrong type", old_name);
2169 if (st_mode != patch->old_mode)
2170 fprintf(stderr, "warning: %s has type %o, expected %o\n",
2171 old_name, st_mode, patch->old_mode);
2172 }
2173
2174 if (new_name && prev_patch && 0 < prev_patch->is_delete &&
2175 !strcmp(prev_patch->old_name, new_name))
2176 /* A type-change diff is always split into a patch to
2177 * delete old, immediately followed by a patch to
2178 * create new (see diff.c::run_diff()); in such a case
2179 * it is Ok that the entry to be deleted by the
2180 * previous patch is still in the working tree and in
2181 * the index.
2182 */
2183 ok_if_exists = 1;
2184 else
2185 ok_if_exists = 0;
2186
2187 if (new_name &&
2188 ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
2189 if (check_index &&
2190 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
2191 !ok_if_exists)
2192 return error("%s: already exists in index", new_name);
2193 if (!cached) {
2194 int err = check_to_create_blob(new_name, ok_if_exists);
2195 if (err)
2196 return err;
2197 }
2198 if (!patch->new_mode) {
2199 if (0 < patch->is_new)
2200 patch->new_mode = S_IFREG | 0644;
2201 else
2202 patch->new_mode = patch->old_mode;
2203 }
2204 }
2205
2206 if (new_name && old_name) {
2207 int same = !strcmp(old_name, new_name);
2208 if (!patch->new_mode)
2209 patch->new_mode = patch->old_mode;
2210 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
2211 return error("new mode (%o) of %s does not match old mode (%o)%s%s",
2212 patch->new_mode, new_name, patch->old_mode,
2213 same ? "" : " of ", same ? "" : old_name);
2214 }
2215
2216 if (apply_data(patch, &st, ce) < 0)
2217 return error("%s: patch does not apply", name);
2218 patch->rejected = 0;
2219 return 0;
2220 }
2221
2222 static int check_patch_list(struct patch *patch)
2223 {
2224 struct patch *prev_patch = NULL;
2225 int err = 0;
2226
2227 for (prev_patch = NULL; patch ; patch = patch->next) {
2228 if (apply_verbosely)
2229 say_patch_name(stderr,
2230 "Checking patch ", patch, "...\n");
2231 err |= check_patch(patch, prev_patch);
2232 prev_patch = patch;
2233 }
2234 return err;
2235 }
2236
2237 /* This function tries to read the sha1 from the current index */
2238 static int get_current_sha1(const char *path, unsigned char *sha1)
2239 {
2240 int pos;
2241
2242 if (read_cache() < 0)
2243 return -1;
2244 pos = cache_name_pos(path, strlen(path));
2245 if (pos < 0)
2246 return -1;
2247 hashcpy(sha1, active_cache[pos]->sha1);
2248 return 0;
2249 }
2250
2251 /* Build an index that contains the just the files needed for a 3way merge */
2252 static void build_fake_ancestor(struct patch *list, const char *filename)
2253 {
2254 struct patch *patch;
2255 struct index_state result = { 0 };
2256 int fd;
2257
2258 /* Once we start supporting the reverse patch, it may be
2259 * worth showing the new sha1 prefix, but until then...
2260 */
2261 for (patch = list; patch; patch = patch->next) {
2262 const unsigned char *sha1_ptr;
2263 unsigned char sha1[20];
2264 struct cache_entry *ce;
2265 const char *name;
2266
2267 name = patch->old_name ? patch->old_name : patch->new_name;
2268 if (0 < patch->is_new)
2269 continue;
2270 else if (get_sha1(patch->old_sha1_prefix, sha1))
2271 /* git diff has no index line for mode/type changes */
2272 if (!patch->lines_added && !patch->lines_deleted) {
2273 if (get_current_sha1(patch->new_name, sha1) ||
2274 get_current_sha1(patch->old_name, sha1))
2275 die("mode change for %s, which is not "
2276 "in current HEAD", name);
2277 sha1_ptr = sha1;
2278 } else
2279 die("sha1 information is lacking or useless "
2280 "(%s).", name);
2281 else
2282 sha1_ptr = sha1;
2283
2284 ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);
2285 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
2286 die ("Could not add %s to temporary index", name);
2287 }
2288
2289 fd = open(filename, O_WRONLY | O_CREAT, 0666);
2290 if (fd < 0 || write_index(&result, fd) || close(fd))
2291 die ("Could not write temporary index to %s", filename);
2292
2293 discard_index(&result);
2294 }
2295
2296 static void stat_patch_list(struct patch *patch)
2297 {
2298 int files, adds, dels;
2299
2300 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
2301 files++;
2302 adds += patch->lines_added;
2303 dels += patch->lines_deleted;
2304 show_stats(patch);
2305 }
2306
2307 printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
2308 }
2309
2310 static void numstat_patch_list(struct patch *patch)
2311 {
2312 for ( ; patch; patch = patch->next) {
2313 const char *name;
2314 name = patch->new_name ? patch->new_name : patch->old_name;
2315 if (patch->is_binary)
2316 printf("-\t-\t");
2317 else
2318 printf("%d\t%d\t",
2319 patch->lines_added, patch->lines_deleted);
2320 if (line_termination && quote_c_style(name, NULL, NULL, 0))
2321 quote_c_style(name, NULL, stdout, 0);
2322 else
2323 fputs(name, stdout);
2324 putchar(line_termination);
2325 }
2326 }
2327
2328 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
2329 {
2330 if (mode)
2331 printf(" %s mode %06o %s\n", newdelete, mode, name);
2332 else
2333 printf(" %s %s\n", newdelete, name);
2334 }
2335
2336 static void show_mode_change(struct patch *p, int show_name)
2337 {
2338 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
2339 if (show_name)
2340 printf(" mode change %06o => %06o %s\n",
2341 p->old_mode, p->new_mode, p->new_name);
2342 else
2343 printf(" mode change %06o => %06o\n",
2344 p->old_mode, p->new_mode);
2345 }
2346 }
2347
2348 static void show_rename_copy(struct patch *p)
2349 {
2350 const char *renamecopy = p->is_rename ? "rename" : "copy";
2351 const char *old, *new;
2352
2353 /* Find common prefix */
2354 old = p->old_name;
2355 new = p->new_name;
2356 while (1) {
2357 const char *slash_old, *slash_new;
2358 slash_old = strchr(old, '/');
2359 slash_new = strchr(new, '/');
2360 if (!slash_old ||
2361 !slash_new ||
2362 slash_old - old != slash_new - new ||
2363 memcmp(old, new, slash_new - new))
2364 break;
2365 old = slash_old + 1;
2366 new = slash_new + 1;
2367 }
2368 /* p->old_name thru old is the common prefix, and old and new
2369 * through the end of names are renames
2370 */
2371 if (old != p->old_name)
2372 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
2373 (int)(old - p->old_name), p->old_name,
2374 old, new, p->score);
2375 else
2376 printf(" %s %s => %s (%d%%)\n", renamecopy,
2377 p->old_name, p->new_name, p->score);
2378 show_mode_change(p, 0);
2379 }
2380
2381 static void summary_patch_list(struct patch *patch)
2382 {
2383 struct patch *p;
2384
2385 for (p = patch; p; p = p->next) {
2386 if (p->is_new)
2387 show_file_mode_name("create", p->new_mode, p->new_name);
2388 else if (p->is_delete)
2389 show_file_mode_name("delete", p->old_mode, p->old_name);
2390 else {
2391 if (p->is_rename || p->is_copy)
2392 show_rename_copy(p);
2393 else {
2394 if (p->score) {
2395 printf(" rewrite %s (%d%%)\n",
2396 p->new_name, p->score);
2397 show_mode_change(p, 0);
2398 }
2399 else
2400 show_mode_change(p, 1);
2401 }
2402 }
2403 }
2404 }
2405
2406 static void patch_stats(struct patch *patch)
2407 {
2408 int lines = patch->lines_added + patch->lines_deleted;
2409
2410 if (lines > max_change)
2411 max_change = lines;
2412 if (patch->old_name) {
2413 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
2414 if (!len)
2415 len = strlen(patch->old_name);
2416 if (len > max_len)
2417 max_len = len;
2418 }
2419 if (patch->new_name) {
2420 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
2421 if (!len)
2422 len = strlen(patch->new_name);
2423 if (len > max_len)
2424 max_len = len;
2425 }
2426 }
2427
2428 static void remove_file(struct patch *patch, int rmdir_empty)
2429 {
2430 if (update_index) {
2431 if (remove_file_from_cache(patch->old_name) < 0)
2432 die("unable to remove %s from index", patch->old_name);
2433 cache_tree_invalidate_path(active_cache_tree, patch->old_name);
2434 }
2435 if (!cached) {
2436 if (S_ISGITLINK(patch->old_mode)) {
2437 if (rmdir(patch->old_name))
2438 warning("unable to remove submodule %s",
2439 patch->old_name);
2440 } else if (!unlink(patch->old_name) && rmdir_empty) {
2441 char *name = xstrdup(patch->old_name);
2442 char *end = strrchr(name, '/');
2443 while (end) {
2444 *end = 0;
2445 if (rmdir(name))
2446 break;
2447 end = strrchr(name, '/');
2448 }
2449 free(name);
2450 }
2451 }
2452 }
2453
2454 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
2455 {
2456 struct stat st;
2457 struct cache_entry *ce;
2458 int namelen = strlen(path);
2459 unsigned ce_size = cache_entry_size(namelen);
2460
2461 if (!update_index)
2462 return;
2463
2464 ce = xcalloc(1, ce_size);
2465 memcpy(ce->name, path, namelen);
2466 ce->ce_mode = create_ce_mode(mode);
2467 ce->ce_flags = htons(namelen);
2468 if (S_ISGITLINK(mode)) {
2469 const char *s = buf;
2470
2471 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
2472 die("corrupt patch for subproject %s", path);
2473 } else {
2474 if (!cached) {
2475 if (lstat(path, &st) < 0)
2476 die("unable to stat newly created file %s",
2477 path);
2478 fill_stat_cache_info(ce, &st);
2479 }
2480 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
2481 die("unable to create backing store for newly created file %s", path);
2482 }
2483 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
2484 die("unable to add cache entry for %s", path);
2485 }
2486
2487 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
2488 {
2489 int fd;
2490 char *nbuf;
2491
2492 if (S_ISGITLINK(mode)) {
2493 struct stat st;
2494 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
2495 return 0;
2496 return mkdir(path, 0777);
2497 }
2498
2499 if (has_symlinks && S_ISLNK(mode))
2500 /* Although buf:size is counted string, it also is NUL
2501 * terminated.
2502 */
2503 return symlink(buf, path);
2504
2505 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
2506 if (fd < 0)
2507 return -1;
2508
2509 nbuf = convert_to_working_tree(path, buf, &size);
2510 if (nbuf)
2511 buf = nbuf;
2512
2513 while (size) {
2514 int written = xwrite(fd, buf, size);
2515 if (written < 0)
2516 die("writing file %s: %s", path, strerror(errno));
2517 if (!written)
2518 die("out of space writing file %s", path);
2519 buf += written;
2520 size -= written;
2521 }
2522 if (close(fd) < 0)
2523 die("closing file %s: %s", path, strerror(errno));
2524 if (nbuf)
2525 free(nbuf);
2526 return 0;
2527 }
2528
2529 /*
2530 * We optimistically assume that the directories exist,
2531 * which is true 99% of the time anyway. If they don't,
2532 * we create them and try again.
2533 */
2534 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
2535 {
2536 if (cached)
2537 return;
2538 if (!try_create_file(path, mode, buf, size))
2539 return;
2540
2541 if (errno == ENOENT) {
2542 if (safe_create_leading_directories(path))
2543 return;
2544 if (!try_create_file(path, mode, buf, size))
2545 return;
2546 }
2547
2548 if (errno == EEXIST || errno == EACCES) {
2549 /* We may be trying to create a file where a directory
2550 * used to be.
2551 */
2552 struct stat st;
2553 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
2554 errno = EEXIST;
2555 }
2556
2557 if (errno == EEXIST) {
2558 unsigned int nr = getpid();
2559
2560 for (;;) {
2561 const char *newpath;
2562 newpath = mkpath("%s~%u", path, nr);
2563 if (!try_create_file(newpath, mode, buf, size)) {
2564 if (!rename(newpath, path))
2565 return;
2566 unlink(newpath);
2567 break;
2568 }
2569 if (errno != EEXIST)
2570 break;
2571 ++nr;
2572 }
2573 }
2574 die("unable to write file %s mode %o", path, mode);
2575 }
2576
2577 static void create_file(struct patch *patch)
2578 {
2579 char *path = patch->new_name;
2580 unsigned mode = patch->new_mode;
2581 unsigned long size = patch->resultsize;
2582 char *buf = patch->result;
2583
2584 if (!mode)
2585 mode = S_IFREG | 0644;
2586 create_one_file(path, mode, buf, size);
2587 add_index_file(path, mode, buf, size);
2588 cache_tree_invalidate_path(active_cache_tree, path);
2589 }
2590
2591 /* phase zero is to remove, phase one is to create */
2592 static void write_out_one_result(struct patch *patch, int phase)
2593 {
2594 if (patch->is_delete > 0) {
2595 if (phase == 0)
2596 remove_file(patch, 1);
2597 return;
2598 }
2599 if (patch->is_new > 0 || patch->is_copy) {
2600 if (phase == 1)
2601 create_file(patch);
2602 return;
2603 }
2604 /*
2605 * Rename or modification boils down to the same
2606 * thing: remove the old, write the new
2607 */
2608 if (phase == 0)
2609 remove_file(patch, patch->is_rename);
2610 if (phase == 1)
2611 create_file(patch);
2612 }
2613
2614 static int write_out_one_reject(struct patch *patch)
2615 {
2616 FILE *rej;
2617 char namebuf[PATH_MAX];
2618 struct fragment *frag;
2619 int cnt = 0;
2620
2621 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
2622 if (!frag->rejected)
2623 continue;
2624 cnt++;
2625 }
2626
2627 if (!cnt) {
2628 if (apply_verbosely)
2629 say_patch_name(stderr,
2630 "Applied patch ", patch, " cleanly.\n");
2631 return 0;
2632 }
2633
2634 /* This should not happen, because a removal patch that leaves
2635 * contents are marked "rejected" at the patch level.
2636 */
2637 if (!patch->new_name)
2638 die("internal error");
2639
2640 /* Say this even without --verbose */
2641 say_patch_name(stderr, "Applying patch ", patch, " with");
2642 fprintf(stderr, " %d rejects...\n", cnt);
2643
2644 cnt = strlen(patch->new_name);
2645 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
2646 cnt = ARRAY_SIZE(namebuf) - 5;
2647 fprintf(stderr,
2648 "warning: truncating .rej filename to %.*s.rej",
2649 cnt - 1, patch->new_name);
2650 }
2651 memcpy(namebuf, patch->new_name, cnt);
2652 memcpy(namebuf + cnt, ".rej", 5);
2653
2654 rej = fopen(namebuf, "w");
2655 if (!rej)
2656 return error("cannot open %s: %s", namebuf, strerror(errno));
2657
2658 /* Normal git tools never deal with .rej, so do not pretend
2659 * this is a git patch by saying --git nor give extended
2660 * headers. While at it, maybe please "kompare" that wants
2661 * the trailing TAB and some garbage at the end of line ;-).
2662 */
2663 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
2664 patch->new_name, patch->new_name);
2665 for (cnt = 1, frag = patch->fragments;
2666 frag;
2667 cnt++, frag = frag->next) {
2668 if (!frag->rejected) {
2669 fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);
2670 continue;
2671 }
2672 fprintf(stderr, "Rejected hunk #%d.\n", cnt);
2673 fprintf(rej, "%.*s", frag->size, frag->patch);
2674 if (frag->patch[frag->size-1] != '\n')
2675 fputc('\n', rej);
2676 }
2677 fclose(rej);
2678 return -1;
2679 }
2680
2681 static int write_out_results(struct patch *list, int skipped_patch)
2682 {
2683 int phase;
2684 int errs = 0;
2685 struct patch *l;
2686
2687 if (!list && !skipped_patch)
2688 return error("No changes");
2689
2690 for (phase = 0; phase < 2; phase++) {
2691 l = list;
2692 while (l) {
2693 if (l->rejected)
2694 errs = 1;
2695 else {
2696 write_out_one_result(l, phase);
2697 if (phase == 1 && write_out_one_reject(l))
2698 errs = 1;
2699 }
2700 l = l->next;
2701 }
2702 }
2703 return errs;
2704 }
2705
2706 static struct lock_file lock_file;
2707
2708 static struct excludes {
2709 struct excludes *next;
2710 const char *path;
2711 } *excludes;
2712
2713 static int use_patch(struct patch *p)
2714 {
2715 const char *pathname = p->new_name ? p->new_name : p->old_name;
2716 struct excludes *x = excludes;
2717 while (x) {
2718 if (fnmatch(x->path, pathname, 0) == 0)
2719 return 0;
2720 x = x->next;
2721 }
2722 if (0 < prefix_length) {
2723 int pathlen = strlen(pathname);
2724 if (pathlen <= prefix_length ||
2725 memcmp(prefix, pathname, prefix_length))
2726 return 0;
2727 }
2728 return 1;
2729 }
2730
2731 static void prefix_one(char **name)
2732 {
2733 char *old_name = *name;
2734 if (!old_name)
2735 return;
2736 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
2737 free(old_name);
2738 }
2739
2740 static void prefix_patches(struct patch *p)
2741 {
2742 if (!prefix || p->is_toplevel_relative)
2743 return;
2744 for ( ; p; p = p->next) {
2745 if (p->new_name == p->old_name) {
2746 char *prefixed = p->new_name;
2747 prefix_one(&prefixed);
2748 p->new_name = p->old_name = prefixed;
2749 }
2750 else {
2751 prefix_one(&p->new_name);
2752 prefix_one(&p->old_name);
2753 }
2754 }
2755 }
2756
2757 static int apply_patch(int fd, const char *filename, int inaccurate_eof)
2758 {
2759 unsigned long offset, size;
2760 char *buffer = read_patch_file(fd, &size);
2761 struct patch *list = NULL, **listp = &list;
2762 int skipped_patch = 0;
2763
2764 patch_input_file = filename;
2765 if (!buffer)
2766 return -1;
2767 offset = 0;
2768 while (size > 0) {
2769 struct patch *patch;
2770 int nr;
2771
2772 patch = xcalloc(1, sizeof(*patch));
2773 patch->inaccurate_eof = inaccurate_eof;
2774 nr = parse_chunk(buffer + offset, size, patch);
2775 if (nr < 0)
2776 break;
2777 if (apply_in_reverse)
2778 reverse_patches(patch);
2779 if (prefix)
2780 prefix_patches(patch);
2781 if (use_patch(patch)) {
2782 patch_stats(patch);
2783 *listp = patch;
2784 listp = &patch->next;
2785 }
2786 else {
2787 /* perhaps free it a bit better? */
2788 free(patch);
2789 skipped_patch++;
2790 }
2791 offset += nr;
2792 size -= nr;
2793 }
2794
2795 if (whitespace_error && (new_whitespace == error_on_whitespace))
2796 apply = 0;
2797
2798 update_index = check_index && apply;
2799 if (update_index && newfd < 0)
2800 newfd = hold_locked_index(&lock_file, 1);
2801
2802 if (check_index) {
2803 if (read_cache() < 0)
2804 die("unable to read index file");
2805 }
2806
2807 if ((check || apply) &&
2808 check_patch_list(list) < 0 &&
2809 !apply_with_reject)
2810 exit(1);
2811
2812 if (apply && write_out_results(list, skipped_patch))
2813 exit(1);
2814
2815 if (fake_ancestor)
2816 build_fake_ancestor(list, fake_ancestor);
2817
2818 if (diffstat)
2819 stat_patch_list(list);
2820
2821 if (numstat)
2822 numstat_patch_list(list);
2823
2824 if (summary)
2825 summary_patch_list(list);
2826
2827 free(buffer);
2828 return 0;
2829 }
2830
2831 static int git_apply_config(const char *var, const char *value)
2832 {
2833 if (!strcmp(var, "apply.whitespace")) {
2834 apply_default_whitespace = xstrdup(value);
2835 return 0;
2836 }
2837 return git_default_config(var, value);
2838 }
2839
2840
2841 int cmd_apply(int argc, const char **argv, const char *unused_prefix)
2842 {
2843 int i;
2844 int read_stdin = 1;
2845 int inaccurate_eof = 0;
2846 int errs = 0;
2847 int is_not_gitdir = 0;
2848
2849 const char *whitespace_option = NULL;
2850
2851 prefix = setup_git_directory_gently(&is_not_gitdir);
2852 prefix_length = prefix ? strlen(prefix) : 0;
2853 git_config(git_apply_config);
2854 if (apply_default_whitespace)
2855 parse_whitespace_option(apply_default_whitespace);
2856
2857 for (i = 1; i < argc; i++) {
2858 const char *arg = argv[i];
2859 char *end;
2860 int fd;
2861
2862 if (!strcmp(arg, "-")) {
2863 errs |= apply_patch(0, "<stdin>", inaccurate_eof);
2864 read_stdin = 0;
2865 continue;
2866 }
2867 if (!prefixcmp(arg, "--exclude=")) {
2868 struct excludes *x = xmalloc(sizeof(*x));
2869 x->path = arg + 10;
2870 x->next = excludes;
2871 excludes = x;
2872 continue;
2873 }
2874 if (!prefixcmp(arg, "-p")) {
2875 p_value = atoi(arg + 2);
2876 p_value_known = 1;
2877 continue;
2878 }
2879 if (!strcmp(arg, "--no-add")) {
2880 no_add = 1;
2881 continue;
2882 }
2883 if (!strcmp(arg, "--stat")) {
2884 apply = 0;
2885 diffstat = 1;
2886 continue;
2887 }
2888 if (!strcmp(arg, "--allow-binary-replacement") ||
2889 !strcmp(arg, "--binary")) {
2890 continue; /* now no-op */
2891 }
2892 if (!strcmp(arg, "--numstat")) {
2893 apply = 0;
2894 numstat = 1;
2895 continue;
2896 }
2897 if (!strcmp(arg, "--summary")) {
2898 apply = 0;
2899 summary = 1;
2900 continue;
2901 }
2902 if (!strcmp(arg, "--check")) {
2903 apply = 0;
2904 check = 1;
2905 continue;
2906 }
2907 if (!strcmp(arg, "--index")) {
2908 if (is_not_gitdir)
2909 die("--index outside a repository");
2910 check_index = 1;
2911 continue;
2912 }
2913 if (!strcmp(arg, "--cached")) {
2914 if (is_not_gitdir)
2915 die("--cached outside a repository");
2916 check_index = 1;
2917 cached = 1;
2918 continue;
2919 }
2920 if (!strcmp(arg, "--apply")) {
2921 apply = 1;
2922 continue;
2923 }
2924 if (!strcmp(arg, "--build-fake-ancestor")) {
2925 apply = 0;
2926 if (++i >= argc)
2927 die ("need a filename");
2928 fake_ancestor = argv[i];
2929 continue;
2930 }
2931 if (!strcmp(arg, "-z")) {
2932 line_termination = 0;
2933 continue;
2934 }
2935 if (!prefixcmp(arg, "-C")) {
2936 p_context = strtoul(arg + 2, &end, 0);
2937 if (*end != '\0')
2938 die("unrecognized context count '%s'", arg + 2);
2939 continue;
2940 }
2941 if (!prefixcmp(arg, "--whitespace=")) {
2942 whitespace_option = arg + 13;
2943 parse_whitespace_option(arg + 13);
2944 continue;
2945 }
2946 if (!strcmp(arg, "-R") || !strcmp(arg, "--reverse")) {
2947 apply_in_reverse = 1;
2948 continue;
2949 }
2950 if (!strcmp(arg, "--unidiff-zero")) {
2951 unidiff_zero = 1;
2952 continue;
2953 }
2954 if (!strcmp(arg, "--reject")) {
2955 apply = apply_with_reject = apply_verbosely = 1;
2956 continue;
2957 }
2958 if (!strcmp(arg, "-v") || !strcmp(arg, "--verbose")) {
2959 apply_verbosely = 1;
2960 continue;
2961 }
2962 if (!strcmp(arg, "--inaccurate-eof")) {
2963 inaccurate_eof = 1;
2964 continue;
2965 }
2966 if (0 < prefix_length)
2967 arg = prefix_filename(prefix, prefix_length, arg);
2968
2969 fd = open(arg, O_RDONLY);
2970 if (fd < 0)
2971 usage(apply_usage);
2972 read_stdin = 0;
2973 set_default_whitespace_mode(whitespace_option);
2974 errs |= apply_patch(fd, arg, inaccurate_eof);
2975 close(fd);
2976 }
2977 set_default_whitespace_mode(whitespace_option);
2978 if (read_stdin)
2979 errs |= apply_patch(0, "<stdin>", inaccurate_eof);
2980 if (whitespace_error) {
2981 if (squelch_whitespace_errors &&
2982 squelch_whitespace_errors < whitespace_error) {
2983 int squelched =
2984 whitespace_error - squelch_whitespace_errors;
2985 fprintf(stderr, "warning: squelched %d "
2986 "whitespace error%s\n",
2987 squelched,
2988 squelched == 1 ? "" : "s");
2989 }
2990 if (new_whitespace == error_on_whitespace)
2991 die("%d line%s add%s whitespace errors.",
2992 whitespace_error,
2993 whitespace_error == 1 ? "" : "s",
2994 whitespace_error == 1 ? "s" : "");
2995 if (applied_after_fixing_ws)
2996 fprintf(stderr, "warning: %d line%s applied after"
2997 " fixing whitespace errors.\n",
2998 applied_after_fixing_ws,
2999 applied_after_fixing_ws == 1 ? "" : "s");
3000 else if (whitespace_error)
3001 fprintf(stderr, "warning: %d line%s add%s whitespace errors.\n",
3002 whitespace_error,
3003 whitespace_error == 1 ? "" : "s",
3004 whitespace_error == 1 ? "s" : "");
3005 }
3006
3007 if (update_index) {
3008 if (write_cache(newfd, active_cache, active_nr) ||
3009 close(newfd) || commit_locked_index(&lock_file))
3010 die("Unable to write new index file");
3011 }
3012
3013 return !!errs;
3014 }