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