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