Merge branch 'jc/diff' into next
[git] / 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 <fnmatch.h>
10 #include "cache.h"
11 #include "cache-tree.h"
12 #include "quote.h"
13 #include "blob.h"
14
15 //  --check turns on checking that the working tree matches the
16 //    files that are being modified, but doesn't apply the patch
17 //  --stat does just a diffstat, and doesn't actually apply
18 //  --numstat does numeric diffstat, and doesn't actually apply
19 //  --index-info shows the old and new index info for paths if available.
20 //
21 static const char *prefix;
22 static int prefix_length = -1;
23
24 static int p_value = 1;
25 static int allow_binary_replacement = 0;
26 static int check_index = 0;
27 static int write_index = 0;
28 static int diffstat = 0;
29 static int numstat = 0;
30 static int summary = 0;
31 static int check = 0;
32 static int apply = 1;
33 static int no_add = 0;
34 static int show_index_info = 0;
35 static int line_termination = '\n';
36 static unsigned long p_context = -1;
37 static const char apply_usage[] =
38 "git-apply [--stat] [--numstat] [--summary] [--check] [--index] [--apply] [--no-add] [--index-info] [--allow-binary-replacement] [-z] [-pNUM] [-CNUM] [--whitespace=<nowarn|warn|error|error-all|strip>] <patch>...";
39
40 static enum whitespace_eol {
41         nowarn_whitespace,
42         warn_on_whitespace,
43         error_on_whitespace,
44         strip_whitespace,
45 } new_whitespace = warn_on_whitespace;
46 static int whitespace_error = 0;
47 static int squelch_whitespace_errors = 5;
48 static int applied_after_stripping = 0;
49 static const char *patch_input_file = NULL;
50
51 static void parse_whitespace_option(const char *option)
52 {
53         if (!option) {
54                 new_whitespace = warn_on_whitespace;
55                 return;
56         }
57         if (!strcmp(option, "warn")) {
58                 new_whitespace = warn_on_whitespace;
59                 return;
60         }
61         if (!strcmp(option, "nowarn")) {
62                 new_whitespace = nowarn_whitespace;
63                 return;
64         }
65         if (!strcmp(option, "error")) {
66                 new_whitespace = error_on_whitespace;
67                 return;
68         }
69         if (!strcmp(option, "error-all")) {
70                 new_whitespace = error_on_whitespace;
71                 squelch_whitespace_errors = 0;
72                 return;
73         }
74         if (!strcmp(option, "strip")) {
75                 new_whitespace = strip_whitespace;
76                 return;
77         }
78         die("unrecognized whitespace option '%s'", option);
79 }
80
81 static void set_default_whitespace_mode(const char *whitespace_option)
82 {
83         if (!whitespace_option && !apply_default_whitespace) {
84                 new_whitespace = (apply
85                                   ? warn_on_whitespace
86                                   : nowarn_whitespace);
87         }
88 }
89
90 /*
91  * For "diff-stat" like behaviour, we keep track of the biggest change
92  * we've seen, and the longest filename. That allows us to do simple
93  * scaling.
94  */
95 static int max_change, max_len;
96
97 /*
98  * Various "current state", notably line numbers and what
99  * file (and how) we're patching right now.. The "is_xxxx"
100  * things are flags, where -1 means "don't know yet".
101  */
102 static int linenr = 1;
103
104 struct fragment {
105         unsigned long leading, trailing;
106         unsigned long oldpos, oldlines;
107         unsigned long newpos, newlines;
108         const char *patch;
109         int size;
110         struct fragment *next;
111 };
112
113 struct patch {
114         char *new_name, *old_name, *def_name;
115         unsigned int old_mode, new_mode;
116         int is_rename, is_copy, is_new, is_delete, is_binary;
117         int lines_added, lines_deleted;
118         int score;
119         struct fragment *fragments;
120         char *result;
121         unsigned long resultsize;
122         char old_sha1_prefix[41];
123         char new_sha1_prefix[41];
124         struct patch *next;
125 };
126
127 #define CHUNKSIZE (8192)
128 #define SLOP (16)
129
130 static void *read_patch_file(int fd, unsigned long *sizep)
131 {
132         unsigned long size = 0, alloc = CHUNKSIZE;
133         void *buffer = xmalloc(alloc);
134
135         for (;;) {
136                 int nr = alloc - size;
137                 if (nr < 1024) {
138                         alloc += CHUNKSIZE;
139                         buffer = xrealloc(buffer, alloc);
140                         nr = alloc - size;
141                 }
142                 nr = xread(fd, buffer + size, nr);
143                 if (!nr)
144                         break;
145                 if (nr < 0)
146                         die("git-apply: read returned %s", strerror(errno));
147                 size += nr;
148         }
149         *sizep = size;
150
151         /*
152          * Make sure that we have some slop in the buffer
153          * so that we can do speculative "memcmp" etc, and
154          * see to it that it is NUL-filled.
155          */
156         if (alloc < size + SLOP)
157                 buffer = xrealloc(buffer, size + SLOP);
158         memset(buffer + size, 0, SLOP);
159         return buffer;
160 }
161
162 static unsigned long linelen(const char *buffer, unsigned long size)
163 {
164         unsigned long len = 0;
165         while (size--) {
166                 len++;
167                 if (*buffer++ == '\n')
168                         break;
169         }
170         return len;
171 }
172
173 static int is_dev_null(const char *str)
174 {
175         return !memcmp("/dev/null", str, 9) && isspace(str[9]);
176 }
177
178 #define TERM_SPACE      1
179 #define TERM_TAB        2
180
181 static int name_terminate(const char *name, int namelen, int c, int terminate)
182 {
183         if (c == ' ' && !(terminate & TERM_SPACE))
184                 return 0;
185         if (c == '\t' && !(terminate & TERM_TAB))
186                 return 0;
187
188         return 1;
189 }
190
191 static char * find_name(const char *line, char *def, int p_value, int terminate)
192 {
193         int len;
194         const char *start = line;
195         char *name;
196
197         if (*line == '"') {
198                 /* Proposed "new-style" GNU patch/diff format; see
199                  * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
200                  */
201                 name = unquote_c_style(line, NULL);
202                 if (name) {
203                         char *cp = name;
204                         while (p_value) {
205                                 cp = strchr(name, '/');
206                                 if (!cp)
207                                         break;
208                                 cp++;
209                                 p_value--;
210                         }
211                         if (cp) {
212                                 /* name can later be freed, so we need
213                                  * to memmove, not just return cp
214                                  */
215                                 memmove(name, cp, strlen(cp) + 1);
216                                 free(def);
217                                 return name;
218                         }
219                         else {
220                                 free(name);
221                                 name = NULL;
222                         }
223                 }
224         }
225
226         for (;;) {
227                 char c = *line;
228
229                 if (isspace(c)) {
230                         if (c == '\n')
231                                 break;
232                         if (name_terminate(start, line-start, c, terminate))
233                                 break;
234                 }
235                 line++;
236                 if (c == '/' && !--p_value)
237                         start = line;
238         }
239         if (!start)
240                 return def;
241         len = line - start;
242         if (!len)
243                 return def;
244
245         /*
246          * Generally we prefer the shorter name, especially
247          * if the other one is just a variation of that with
248          * something else tacked on to the end (ie "file.orig"
249          * or "file~").
250          */
251         if (def) {
252                 int deflen = strlen(def);
253                 if (deflen < len && !strncmp(start, def, deflen))
254                         return def;
255         }
256
257         name = xmalloc(len + 1);
258         memcpy(name, start, len);
259         name[len] = 0;
260         free(def);
261         return name;
262 }
263
264 /*
265  * Get the name etc info from the --/+++ lines of a traditional patch header
266  *
267  * NOTE! This hardcodes "-p1" behaviour in filename detection.
268  *
269  * FIXME! The end-of-filename heuristics are kind of screwy. For existing
270  * files, we can happily check the index for a match, but for creating a
271  * new file we should try to match whatever "patch" does. I have no idea.
272  */
273 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
274 {
275         char *name;
276
277         first += 4;     // skip "--- "
278         second += 4;    // skip "+++ "
279         if (is_dev_null(first)) {
280                 patch->is_new = 1;
281                 patch->is_delete = 0;
282                 name = find_name(second, NULL, p_value, TERM_SPACE | TERM_TAB);
283                 patch->new_name = name;
284         } else if (is_dev_null(second)) {
285                 patch->is_new = 0;
286                 patch->is_delete = 1;
287                 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
288                 patch->old_name = name;
289         } else {
290                 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
291                 name = find_name(second, name, p_value, TERM_SPACE | TERM_TAB);
292                 patch->old_name = patch->new_name = name;
293         }
294         if (!name)
295                 die("unable to find filename in patch at line %d", linenr);
296 }
297
298 static int gitdiff_hdrend(const char *line, struct patch *patch)
299 {
300         return -1;
301 }
302
303 /*
304  * We're anal about diff header consistency, to make
305  * sure that we don't end up having strange ambiguous
306  * patches floating around.
307  *
308  * As a result, gitdiff_{old|new}name() will check
309  * their names against any previous information, just
310  * to make sure..
311  */
312 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
313 {
314         if (!orig_name && !isnull)
315                 return find_name(line, NULL, 1, 0);
316
317         if (orig_name) {
318                 int len;
319                 const char *name;
320                 char *another;
321                 name = orig_name;
322                 len = strlen(name);
323                 if (isnull)
324                         die("git-apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
325                 another = find_name(line, NULL, 1, 0);
326                 if (!another || memcmp(another, name, len))
327                         die("git-apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
328                 free(another);
329                 return orig_name;
330         }
331         else {
332                 /* expect "/dev/null" */
333                 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
334                         die("git-apply: bad git-diff - expected /dev/null on line %d", linenr);
335                 return NULL;
336         }
337 }
338
339 static int gitdiff_oldname(const char *line, struct patch *patch)
340 {
341         patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
342         return 0;
343 }
344
345 static int gitdiff_newname(const char *line, struct patch *patch)
346 {
347         patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
348         return 0;
349 }
350
351 static int gitdiff_oldmode(const char *line, struct patch *patch)
352 {
353         patch->old_mode = strtoul(line, NULL, 8);
354         return 0;
355 }
356
357 static int gitdiff_newmode(const char *line, struct patch *patch)
358 {
359         patch->new_mode = strtoul(line, NULL, 8);
360         return 0;
361 }
362
363 static int gitdiff_delete(const char *line, struct patch *patch)
364 {
365         patch->is_delete = 1;
366         patch->old_name = patch->def_name;
367         return gitdiff_oldmode(line, patch);
368 }
369
370 static int gitdiff_newfile(const char *line, struct patch *patch)
371 {
372         patch->is_new = 1;
373         patch->new_name = patch->def_name;
374         return gitdiff_newmode(line, patch);
375 }
376
377 static int gitdiff_copysrc(const char *line, struct patch *patch)
378 {
379         patch->is_copy = 1;
380         patch->old_name = find_name(line, NULL, 0, 0);
381         return 0;
382 }
383
384 static int gitdiff_copydst(const char *line, struct patch *patch)
385 {
386         patch->is_copy = 1;
387         patch->new_name = find_name(line, NULL, 0, 0);
388         return 0;
389 }
390
391 static int gitdiff_renamesrc(const char *line, struct patch *patch)
392 {
393         patch->is_rename = 1;
394         patch->old_name = find_name(line, NULL, 0, 0);
395         return 0;
396 }
397
398 static int gitdiff_renamedst(const char *line, struct patch *patch)
399 {
400         patch->is_rename = 1;
401         patch->new_name = find_name(line, NULL, 0, 0);
402         return 0;
403 }
404
405 static int gitdiff_similarity(const char *line, struct patch *patch)
406 {
407         if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
408                 patch->score = 0;
409         return 0;
410 }
411
412 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
413 {
414         if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
415                 patch->score = 0;
416         return 0;
417 }
418
419 static int gitdiff_index(const char *line, struct patch *patch)
420 {
421         /* index line is N hexadecimal, "..", N hexadecimal,
422          * and optional space with octal mode.
423          */
424         const char *ptr, *eol;
425         int len;
426
427         ptr = strchr(line, '.');
428         if (!ptr || ptr[1] != '.' || 40 < ptr - line)
429                 return 0;
430         len = ptr - line;
431         memcpy(patch->old_sha1_prefix, line, len);
432         patch->old_sha1_prefix[len] = 0;
433
434         line = ptr + 2;
435         ptr = strchr(line, ' ');
436         eol = strchr(line, '\n');
437
438         if (!ptr || eol < ptr)
439                 ptr = eol;
440         len = ptr - line;
441
442         if (40 < len)
443                 return 0;
444         memcpy(patch->new_sha1_prefix, line, len);
445         patch->new_sha1_prefix[len] = 0;
446         if (*ptr == ' ')
447                 patch->new_mode = patch->old_mode = strtoul(ptr+1, NULL, 8);
448         return 0;
449 }
450
451 /*
452  * This is normal for a diff that doesn't change anything: we'll fall through
453  * into the next diff. Tell the parser to break out.
454  */
455 static int gitdiff_unrecognized(const char *line, struct patch *patch)
456 {
457         return -1;
458 }
459
460 static const char *stop_at_slash(const char *line, int llen)
461 {
462         int i;
463
464         for (i = 0; i < llen; i++) {
465                 int ch = line[i];
466                 if (ch == '/')
467                         return line + i;
468         }
469         return NULL;
470 }
471
472 /* This is to extract the same name that appears on "diff --git"
473  * line.  We do not find and return anything if it is a rename
474  * patch, and it is OK because we will find the name elsewhere.
475  * We need to reliably find name only when it is mode-change only,
476  * creation or deletion of an empty file.  In any of these cases,
477  * both sides are the same name under a/ and b/ respectively.
478  */
479 static char *git_header_name(char *line, int llen)
480 {
481         int len;
482         const char *name;
483         const char *second = NULL;
484
485         line += strlen("diff --git ");
486         llen -= strlen("diff --git ");
487
488         if (*line == '"') {
489                 const char *cp;
490                 char *first = unquote_c_style(line, &second);
491                 if (!first)
492                         return NULL;
493
494                 /* advance to the first slash */
495                 cp = stop_at_slash(first, strlen(first));
496                 if (!cp || cp == first) {
497                         /* we do not accept absolute paths */
498                 free_first_and_fail:
499                         free(first);
500                         return NULL;
501                 }
502                 len = strlen(cp+1);
503                 memmove(first, cp+1, len+1); /* including NUL */
504
505                 /* second points at one past closing dq of name.
506                  * find the second name.
507                  */
508                 while ((second < line + llen) && isspace(*second))
509                         second++;
510
511                 if (line + llen <= second)
512                         goto free_first_and_fail;
513                 if (*second == '"') {
514                         char *sp = unquote_c_style(second, NULL);
515                         if (!sp)
516                                 goto free_first_and_fail;
517                         cp = stop_at_slash(sp, strlen(sp));
518                         if (!cp || cp == sp) {
519                         free_both_and_fail:
520                                 free(sp);
521                                 goto free_first_and_fail;
522                         }
523                         /* They must match, otherwise ignore */
524                         if (strcmp(cp+1, first))
525                                 goto free_both_and_fail;
526                         free(sp);
527                         return first;
528                 }
529
530                 /* unquoted second */
531                 cp = stop_at_slash(second, line + llen - second);
532                 if (!cp || cp == second)
533                         goto free_first_and_fail;
534                 cp++;
535                 if (line + llen - cp != len + 1 ||
536                     memcmp(first, cp, len))
537                         goto free_first_and_fail;
538                 return first;
539         }
540
541         /* unquoted first name */
542         name = stop_at_slash(line, llen);
543         if (!name || name == line)
544                 return NULL;
545
546         name++;
547
548         /* since the first name is unquoted, a dq if exists must be
549          * the beginning of the second name.
550          */
551         for (second = name; second < line + llen; second++) {
552                 if (*second == '"') {
553                         const char *cp = second;
554                         const char *np;
555                         char *sp = unquote_c_style(second, NULL);
556
557                         if (!sp)
558                                 return NULL;
559                         np = stop_at_slash(sp, strlen(sp));
560                         if (!np || np == sp) {
561                         free_second_and_fail:
562                                 free(sp);
563                                 return NULL;
564                         }
565                         np++;
566                         len = strlen(np);
567                         if (len < cp - name &&
568                             !strncmp(np, name, len) &&
569                             isspace(name[len])) {
570                                 /* Good */
571                                 memmove(sp, np, len + 1);
572                                 return sp;
573                         }
574                         goto free_second_and_fail;
575                 }
576         }
577
578         /*
579          * Accept a name only if it shows up twice, exactly the same
580          * form.
581          */
582         for (len = 0 ; ; len++) {
583                 char c = name[len];
584
585                 switch (c) {
586                 default:
587                         continue;
588                 case '\n':
589                         return NULL;
590                 case '\t': case ' ':
591                         second = name+len;
592                         for (;;) {
593                                 char c = *second++;
594                                 if (c == '\n')
595                                         return NULL;
596                                 if (c == '/')
597                                         break;
598                         }
599                         if (second[len] == '\n' && !memcmp(name, second, len)) {
600                                 char *ret = xmalloc(len + 1);
601                                 memcpy(ret, name, len);
602                                 ret[len] = 0;
603                                 return ret;
604                         }
605                 }
606         }
607         return NULL;
608 }
609
610 /* Verify that we recognize the lines following a git header */
611 static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
612 {
613         unsigned long offset;
614
615         /* A git diff has explicit new/delete information, so we don't guess */
616         patch->is_new = 0;
617         patch->is_delete = 0;
618
619         /*
620          * Some things may not have the old name in the
621          * rest of the headers anywhere (pure mode changes,
622          * or removing or adding empty files), so we get
623          * the default name from the header.
624          */
625         patch->def_name = git_header_name(line, len);
626
627         line += len;
628         size -= len;
629         linenr++;
630         for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
631                 static const struct opentry {
632                         const char *str;
633                         int (*fn)(const char *, struct patch *);
634                 } optable[] = {
635                         { "@@ -", gitdiff_hdrend },
636                         { "--- ", gitdiff_oldname },
637                         { "+++ ", gitdiff_newname },
638                         { "old mode ", gitdiff_oldmode },
639                         { "new mode ", gitdiff_newmode },
640                         { "deleted file mode ", gitdiff_delete },
641                         { "new file mode ", gitdiff_newfile },
642                         { "copy from ", gitdiff_copysrc },
643                         { "copy to ", gitdiff_copydst },
644                         { "rename old ", gitdiff_renamesrc },
645                         { "rename new ", gitdiff_renamedst },
646                         { "rename from ", gitdiff_renamesrc },
647                         { "rename to ", gitdiff_renamedst },
648                         { "similarity index ", gitdiff_similarity },
649                         { "dissimilarity index ", gitdiff_dissimilarity },
650                         { "index ", gitdiff_index },
651                         { "", gitdiff_unrecognized },
652                 };
653                 int i;
654
655                 len = linelen(line, size);
656                 if (!len || line[len-1] != '\n')
657                         break;
658                 for (i = 0; i < ARRAY_SIZE(optable); i++) {
659                         const struct opentry *p = optable + i;
660                         int oplen = strlen(p->str);
661                         if (len < oplen || memcmp(p->str, line, oplen))
662                                 continue;
663                         if (p->fn(line + oplen, patch) < 0)
664                                 return offset;
665                         break;
666                 }
667         }
668
669         return offset;
670 }
671
672 static int parse_num(const char *line, unsigned long *p)
673 {
674         char *ptr;
675
676         if (!isdigit(*line))
677                 return 0;
678         *p = strtoul(line, &ptr, 10);
679         return ptr - line;
680 }
681
682 static int parse_range(const char *line, int len, int offset, const char *expect,
683                         unsigned long *p1, unsigned long *p2)
684 {
685         int digits, ex;
686
687         if (offset < 0 || offset >= len)
688                 return -1;
689         line += offset;
690         len -= offset;
691
692         digits = parse_num(line, p1);
693         if (!digits)
694                 return -1;
695
696         offset += digits;
697         line += digits;
698         len -= digits;
699
700         *p2 = 1;
701         if (*line == ',') {
702                 digits = parse_num(line+1, p2);
703                 if (!digits)
704                         return -1;
705
706                 offset += digits+1;
707                 line += digits+1;
708                 len -= digits+1;
709         }
710
711         ex = strlen(expect);
712         if (ex > len)
713                 return -1;
714         if (memcmp(line, expect, ex))
715                 return -1;
716
717         return offset + ex;
718 }
719
720 /*
721  * Parse a unified diff fragment header of the
722  * form "@@ -a,b +c,d @@"
723  */
724 static int parse_fragment_header(char *line, int len, struct fragment *fragment)
725 {
726         int offset;
727
728         if (!len || line[len-1] != '\n')
729                 return -1;
730
731         /* Figure out the number of lines in a fragment */
732         offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
733         offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
734
735         return offset;
736 }
737
738 static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
739 {
740         unsigned long offset, len;
741
742         patch->is_rename = patch->is_copy = 0;
743         patch->is_new = patch->is_delete = -1;
744         patch->old_mode = patch->new_mode = 0;
745         patch->old_name = patch->new_name = NULL;
746         for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
747                 unsigned long nextlen;
748
749                 len = linelen(line, size);
750                 if (!len)
751                         break;
752
753                 /* Testing this early allows us to take a few shortcuts.. */
754                 if (len < 6)
755                         continue;
756
757                 /*
758                  * Make sure we don't find any unconnected patch fragmants.
759                  * That's a sign that we didn't find a header, and that a
760                  * patch has become corrupted/broken up.
761                  */
762                 if (!memcmp("@@ -", line, 4)) {
763                         struct fragment dummy;
764                         if (parse_fragment_header(line, len, &dummy) < 0)
765                                 continue;
766                         error("patch fragment without header at line %d: %.*s", linenr, (int)len-1, line);
767                 }
768
769                 if (size < len + 6)
770                         break;
771
772                 /*
773                  * Git patch? It might not have a real patch, just a rename
774                  * or mode change, so we handle that specially
775                  */
776                 if (!memcmp("diff --git ", line, 11)) {
777                         int git_hdr_len = parse_git_header(line, len, size, patch);
778                         if (git_hdr_len <= len)
779                                 continue;
780                         if (!patch->old_name && !patch->new_name) {
781                                 if (!patch->def_name)
782                                         die("git diff header lacks filename information (line %d)", linenr);
783                                 patch->old_name = patch->new_name = patch->def_name;
784                         }
785                         *hdrsize = git_hdr_len;
786                         return offset;
787                 }
788
789                 /** --- followed by +++ ? */
790                 if (memcmp("--- ", line,  4) || memcmp("+++ ", line + len, 4))
791                         continue;
792
793                 /*
794                  * We only accept unified patches, so we want it to
795                  * at least have "@@ -a,b +c,d @@\n", which is 14 chars
796                  * minimum
797                  */
798                 nextlen = linelen(line + len, size - len);
799                 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
800                         continue;
801
802                 /* Ok, we'll consider it a patch */
803                 parse_traditional_patch(line, line+len, patch);
804                 *hdrsize = len + nextlen;
805                 linenr += 2;
806                 return offset;
807         }
808         return -1;
809 }
810
811 /*
812  * Parse a unified diff. Note that this really needs
813  * to parse each fragment separately, since the only
814  * way to know the difference between a "---" that is
815  * part of a patch, and a "---" that starts the next
816  * patch is to look at the line counts..
817  */
818 static int parse_fragment(char *line, unsigned long size, struct patch *patch, struct fragment *fragment)
819 {
820         int added, deleted;
821         int len = linelen(line, size), offset;
822         unsigned long oldlines, newlines;
823         unsigned long leading, trailing;
824
825         offset = parse_fragment_header(line, len, fragment);
826         if (offset < 0)
827                 return -1;
828         oldlines = fragment->oldlines;
829         newlines = fragment->newlines;
830         leading = 0;
831         trailing = 0;
832
833         if (patch->is_new < 0) {
834                 patch->is_new =  !oldlines;
835                 if (!oldlines)
836                         patch->old_name = NULL;
837         }
838         if (patch->is_delete < 0) {
839                 patch->is_delete = !newlines;
840                 if (!newlines)
841                         patch->new_name = NULL;
842         }
843
844         if (patch->is_new && oldlines)
845                 return error("new file depends on old contents");
846         if (patch->is_delete != !newlines) {
847                 if (newlines)
848                         return error("deleted file still has contents");
849                 fprintf(stderr, "** warning: file %s becomes empty but is not deleted\n", patch->new_name);
850         }
851
852         /* Parse the thing.. */
853         line += len;
854         size -= len;
855         linenr++;
856         added = deleted = 0;
857         for (offset = len; size > 0; offset += len, size -= len, line += len, linenr++) {
858                 if (!oldlines && !newlines)
859                         break;
860                 len = linelen(line, size);
861                 if (!len || line[len-1] != '\n')
862                         return -1;
863                 switch (*line) {
864                 default:
865                         return -1;
866                 case ' ':
867                         oldlines--;
868                         newlines--;
869                         if (!deleted && !added)
870                                 leading++;
871                         trailing++;
872                         break;
873                 case '-':
874                         deleted++;
875                         oldlines--;
876                         trailing = 0;
877                         break;
878                 case '+':
879                         /*
880                          * We know len is at least two, since we have a '+' and
881                          * we checked that the last character was a '\n' above.
882                          * That is, an addition of an empty line would check
883                          * the '+' here.  Sneaky...
884                          */
885                         if ((new_whitespace != nowarn_whitespace) &&
886                             isspace(line[len-2])) {
887                                 whitespace_error++;
888                                 if (squelch_whitespace_errors &&
889                                     squelch_whitespace_errors <
890                                     whitespace_error)
891                                         ;
892                                 else {
893                                         fprintf(stderr, "Adds trailing whitespace.\n%s:%d:%.*s\n",
894                                                 patch_input_file,
895                                                 linenr, len-2, line+1);
896                                 }
897                         }
898                         added++;
899                         newlines--;
900                         trailing = 0;
901                         break;
902
903                 /* We allow "\ No newline at end of file". Depending
904                  * on locale settings when the patch was produced we
905                  * don't know what this line looks like. The only
906                  * thing we do know is that it begins with "\ ".
907                  * Checking for 12 is just for sanity check -- any
908                  * l10n of "\ No newline..." is at least that long.
909                  */
910                 case '\\':
911                         if (len < 12 || memcmp(line, "\\ ", 2))
912                                 return -1;
913                         break;
914                 }
915         }
916         if (oldlines || newlines)
917                 return -1;
918         fragment->leading = leading;
919         fragment->trailing = trailing;
920
921         /* If a fragment ends with an incomplete line, we failed to include
922          * it in the above loop because we hit oldlines == newlines == 0
923          * before seeing it.
924          */
925         if (12 < size && !memcmp(line, "\\ ", 2))
926                 offset += linelen(line, size);
927
928         patch->lines_added += added;
929         patch->lines_deleted += deleted;
930         return offset;
931 }
932
933 static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
934 {
935         unsigned long offset = 0;
936         struct fragment **fragp = &patch->fragments;
937
938         while (size > 4 && !memcmp(line, "@@ -", 4)) {
939                 struct fragment *fragment;
940                 int len;
941
942                 fragment = xcalloc(1, sizeof(*fragment));
943                 len = parse_fragment(line, size, patch, fragment);
944                 if (len <= 0)
945                         die("corrupt patch at line %d", linenr);
946
947                 fragment->patch = line;
948                 fragment->size = len;
949
950                 *fragp = fragment;
951                 fragp = &fragment->next;
952
953                 offset += len;
954                 line += len;
955                 size -= len;
956         }
957         return offset;
958 }
959
960 static inline int metadata_changes(struct patch *patch)
961 {
962         return  patch->is_rename > 0 ||
963                 patch->is_copy > 0 ||
964                 patch->is_new > 0 ||
965                 patch->is_delete ||
966                 (patch->old_mode && patch->new_mode &&
967                  patch->old_mode != patch->new_mode);
968 }
969
970 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
971 {
972         int hdrsize, patchsize;
973         int offset = find_header(buffer, size, &hdrsize, patch);
974
975         if (offset < 0)
976                 return offset;
977
978         patchsize = parse_single_patch(buffer + offset + hdrsize, size - offset - hdrsize, patch);
979
980         if (!patchsize) {
981                 static const char *binhdr[] = {
982                         "Binary files ",
983                         "Files ",
984                         NULL,
985                 };
986                 int i;
987                 int hd = hdrsize + offset;
988                 unsigned long llen = linelen(buffer + hd, size - hd);
989
990                 if (!memcmp(" differ\n", buffer + hd + llen - 8, 8))
991                         for (i = 0; binhdr[i]; i++) {
992                                 int len = strlen(binhdr[i]);
993                                 if (len < size - hd &&
994                                     !memcmp(binhdr[i], buffer + hd, len)) {
995                                         patch->is_binary = 1;
996                                         break;
997                                 }
998                         }
999
1000                 /* Empty patch cannot be applied if:
1001                  * - it is a binary patch and we do not do binary_replace, or
1002                  * - text patch without metadata change
1003                  */
1004                 if ((apply || check) &&
1005                     (patch->is_binary
1006                      ? !allow_binary_replacement
1007                      : !metadata_changes(patch)))
1008                         die("patch with only garbage at line %d", linenr);
1009         }
1010
1011         return offset + hdrsize + patchsize;
1012 }
1013
1014 static const char pluses[] = "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1015 static const char minuses[]= "----------------------------------------------------------------------";
1016
1017 static void show_stats(struct patch *patch)
1018 {
1019         const char *prefix = "";
1020         char *name = patch->new_name;
1021         char *qname = NULL;
1022         int len, max, add, del, total;
1023
1024         if (!name)
1025                 name = patch->old_name;
1026
1027         if (0 < (len = quote_c_style(name, NULL, NULL, 0))) {
1028                 qname = xmalloc(len + 1);
1029                 quote_c_style(name, qname, NULL, 0);
1030                 name = qname;
1031         }
1032
1033         /*
1034          * "scale" the filename
1035          */
1036         len = strlen(name);
1037         max = max_len;
1038         if (max > 50)
1039                 max = 50;
1040         if (len > max) {
1041                 char *slash;
1042                 prefix = "...";
1043                 max -= 3;
1044                 name += len - max;
1045                 slash = strchr(name, '/');
1046                 if (slash)
1047                         name = slash;
1048         }
1049         len = max;
1050
1051         /*
1052          * scale the add/delete
1053          */
1054         max = max_change;
1055         if (max + len > 70)
1056                 max = 70 - len;
1057
1058         add = patch->lines_added;
1059         del = patch->lines_deleted;
1060         total = add + del;
1061
1062         if (max_change > 0) {
1063                 total = (total * max + max_change / 2) / max_change;
1064                 add = (add * max + max_change / 2) / max_change;
1065                 del = total - add;
1066         }
1067         if (patch->is_binary)
1068                 printf(" %s%-*s |  Bin\n", prefix, len, name);
1069         else
1070                 printf(" %s%-*s |%5d %.*s%.*s\n", prefix,
1071                        len, name, patch->lines_added + patch->lines_deleted,
1072                        add, pluses, del, minuses);
1073         if (qname)
1074                 free(qname);
1075 }
1076
1077 static int read_old_data(struct stat *st, const char *path, void *buf, unsigned long size)
1078 {
1079         int fd;
1080         unsigned long got;
1081
1082         switch (st->st_mode & S_IFMT) {
1083         case S_IFLNK:
1084                 return readlink(path, buf, size);
1085         case S_IFREG:
1086                 fd = open(path, O_RDONLY);
1087                 if (fd < 0)
1088                         return error("unable to open %s", path);
1089                 got = 0;
1090                 for (;;) {
1091                         int ret = xread(fd, buf + got, size - got);
1092                         if (ret <= 0)
1093                                 break;
1094                         got += ret;
1095                 }
1096                 close(fd);
1097                 return got;
1098
1099         default:
1100                 return -1;
1101         }
1102 }
1103
1104 static int find_offset(const char *buf, unsigned long size, const char *fragment, unsigned long fragsize, int line, int *lines)
1105 {
1106         int i;
1107         unsigned long start, backwards, forwards;
1108
1109         if (fragsize > size)
1110                 return -1;
1111
1112         start = 0;
1113         if (line > 1) {
1114                 unsigned long offset = 0;
1115                 i = line-1;
1116                 while (offset + fragsize <= size) {
1117                         if (buf[offset++] == '\n') {
1118                                 start = offset;
1119                                 if (!--i)
1120                                         break;
1121                         }
1122                 }
1123         }
1124
1125         /* Exact line number? */
1126         if (!memcmp(buf + start, fragment, fragsize))
1127                 return start;
1128
1129         /*
1130          * There's probably some smart way to do this, but I'll leave
1131          * that to the smart and beautiful people. I'm simple and stupid.
1132          */
1133         backwards = start;
1134         forwards = start;
1135         for (i = 0; ; i++) {
1136                 unsigned long try;
1137                 int n;
1138
1139                 /* "backward" */
1140                 if (i & 1) {
1141                         if (!backwards) {
1142                                 if (forwards + fragsize > size)
1143                                         break;
1144                                 continue;
1145                         }
1146                         do {
1147                                 --backwards;
1148                         } while (backwards && buf[backwards-1] != '\n');
1149                         try = backwards;
1150                 } else {
1151                         while (forwards + fragsize <= size) {
1152                                 if (buf[forwards++] == '\n')
1153                                         break;
1154                         }
1155                         try = forwards;
1156                 }
1157
1158                 if (try + fragsize > size)
1159                         continue;
1160                 if (memcmp(buf + try, fragment, fragsize))
1161                         continue;
1162                 n = (i >> 1)+1;
1163                 if (i & 1)
1164                         n = -n;
1165                 *lines = n;
1166                 return try;
1167         }
1168
1169         /*
1170          * We should start searching forward and backward.
1171          */
1172         return -1;
1173 }
1174
1175 static void remove_first_line(const char **rbuf, int *rsize)
1176 {
1177         const char *buf = *rbuf;
1178         int size = *rsize;
1179         unsigned long offset;
1180         offset = 0;
1181         while (offset <= size) {
1182                 if (buf[offset++] == '\n')
1183                         break;
1184         }
1185         *rsize = size - offset;
1186         *rbuf = buf + offset;
1187 }
1188
1189 static void remove_last_line(const char **rbuf, int *rsize)
1190 {
1191         const char *buf = *rbuf;
1192         int size = *rsize;
1193         unsigned long offset;
1194         offset = size - 1;
1195         while (offset > 0) {
1196                 if (buf[--offset] == '\n')
1197                         break;
1198         }
1199         *rsize = offset + 1;
1200 }
1201
1202 struct buffer_desc {
1203         char *buffer;
1204         unsigned long size;
1205         unsigned long alloc;
1206 };
1207
1208 static int apply_line(char *output, const char *patch, int plen)
1209 {
1210         /* plen is number of bytes to be copied from patch,
1211          * starting at patch+1 (patch[0] is '+').  Typically
1212          * patch[plen] is '\n'.
1213          */
1214         int add_nl_to_tail = 0;
1215         if ((new_whitespace == strip_whitespace) &&
1216             1 < plen && isspace(patch[plen-1])) {
1217                 if (patch[plen] == '\n')
1218                         add_nl_to_tail = 1;
1219                 plen--;
1220                 while (0 < plen && isspace(patch[plen]))
1221                         plen--;
1222                 applied_after_stripping++;
1223         }
1224         memcpy(output, patch + 1, plen);
1225         if (add_nl_to_tail)
1226                 output[plen++] = '\n';
1227         return plen;
1228 }
1229
1230 static int apply_one_fragment(struct buffer_desc *desc, struct fragment *frag)
1231 {
1232         char *buf = desc->buffer;
1233         const char *patch = frag->patch;
1234         int offset, size = frag->size;
1235         char *old = xmalloc(size);
1236         char *new = xmalloc(size);
1237         const char *oldlines, *newlines;
1238         int oldsize = 0, newsize = 0;
1239         unsigned long leading, trailing;
1240         int pos, lines;
1241
1242         while (size > 0) {
1243                 int len = linelen(patch, size);
1244                 int plen;
1245
1246                 if (!len)
1247                         break;
1248
1249                 /*
1250                  * "plen" is how much of the line we should use for
1251                  * the actual patch data. Normally we just remove the
1252                  * first character on the line, but if the line is
1253                  * followed by "\ No newline", then we also remove the
1254                  * last one (which is the newline, of course).
1255                  */
1256                 plen = len-1;
1257                 if (len < size && patch[len] == '\\')
1258                         plen--;
1259                 switch (*patch) {
1260                 case ' ':
1261                 case '-':
1262                         memcpy(old + oldsize, patch + 1, plen);
1263                         oldsize += plen;
1264                         if (*patch == '-')
1265                                 break;
1266                 /* Fall-through for ' ' */
1267                 case '+':
1268                         if (*patch != '+' || !no_add)
1269                                 newsize += apply_line(new + newsize, patch,
1270                                                       plen);
1271                         break;
1272                 case '@': case '\\':
1273                         /* Ignore it, we already handled it */
1274                         break;
1275                 default:
1276                         return -1;
1277                 }
1278                 patch += len;
1279                 size -= len;
1280         }
1281
1282 #ifdef NO_ACCURATE_DIFF
1283         if (oldsize > 0 && old[oldsize - 1] == '\n' &&
1284                         newsize > 0 && new[newsize - 1] == '\n') {
1285                 oldsize--;
1286                 newsize--;
1287         }
1288 #endif
1289
1290         oldlines = old;
1291         newlines = new;
1292         leading = frag->leading;
1293         trailing = frag->trailing;
1294         lines = 0;
1295         pos = frag->newpos;
1296         for (;;) {
1297                 offset = find_offset(buf, desc->size, oldlines, oldsize, pos, &lines);
1298                 if (offset >= 0) {
1299                         int diff = newsize - oldsize;
1300                         unsigned long size = desc->size + diff;
1301                         unsigned long alloc = desc->alloc;
1302
1303                         /* Warn if it was necessary to reduce the number
1304                          * of context lines.
1305                          */
1306                         if ((leading != frag->leading) || (trailing != frag->trailing))
1307                                 fprintf(stderr, "Context reduced to (%ld/%ld) to apply fragment at %d\n",
1308                                         leading, trailing, pos + lines);
1309
1310                         if (size > alloc) {
1311                                 alloc = size + 8192;
1312                                 desc->alloc = alloc;
1313                                 buf = xrealloc(buf, alloc);
1314                                 desc->buffer = buf;
1315                         }
1316                         desc->size = size;
1317                         memmove(buf + offset + newsize, buf + offset + oldsize, size - offset - newsize);
1318                         memcpy(buf + offset, newlines, newsize);
1319                         offset = 0;
1320
1321                         break;
1322                 }
1323
1324                 /* Am I at my context limits? */
1325                 if ((leading <= p_context) && (trailing <= p_context))
1326                         break;
1327                 /* Reduce the number of context lines
1328                  * Reduce both leading and trailing if they are equal
1329                  * otherwise just reduce the larger context.
1330                  */
1331                 if (leading >= trailing) {
1332                         remove_first_line(&oldlines, &oldsize);
1333                         remove_first_line(&newlines, &newsize);
1334                         pos--;
1335                         leading--;
1336                 }
1337                 if (trailing > leading) {
1338                         remove_last_line(&oldlines, &oldsize);
1339                         remove_last_line(&newlines, &newsize);
1340                         trailing--;
1341                 }
1342         }
1343
1344         free(old);
1345         free(new);
1346         return offset;
1347 }
1348
1349 static int apply_fragments(struct buffer_desc *desc, struct patch *patch)
1350 {
1351         struct fragment *frag = patch->fragments;
1352         const char *name = patch->old_name ? patch->old_name : patch->new_name;
1353
1354         if (patch->is_binary) {
1355                 unsigned char sha1[20];
1356
1357                 if (!allow_binary_replacement)
1358                         return error("cannot apply binary patch to '%s' "
1359                                      "without --allow-binary-replacement",
1360                                      name);
1361
1362                 /* For safety, we require patch index line to contain
1363                  * full 40-byte textual SHA1 for old and new, at least for now.
1364                  */
1365                 if (strlen(patch->old_sha1_prefix) != 40 ||
1366                     strlen(patch->new_sha1_prefix) != 40 ||
1367                     get_sha1_hex(patch->old_sha1_prefix, sha1) ||
1368                     get_sha1_hex(patch->new_sha1_prefix, sha1))
1369                         return error("cannot apply binary patch to '%s' "
1370                                      "without full index line", name);
1371
1372                 if (patch->old_name) {
1373                         unsigned char hdr[50];
1374                         int hdrlen;
1375
1376                         /* See if the old one matches what the patch
1377                          * applies to.
1378                          */
1379                         write_sha1_file_prepare(desc->buffer, desc->size,
1380                                                 blob_type, sha1, hdr, &hdrlen);
1381                         if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
1382                                 return error("the patch applies to '%s' (%s), "
1383                                              "which does not match the "
1384                                              "current contents.",
1385                                              name, sha1_to_hex(sha1));
1386                 }
1387                 else {
1388                         /* Otherwise, the old one must be empty. */
1389                         if (desc->size)
1390                                 return error("the patch applies to an empty "
1391                                              "'%s' but it is not empty", name);
1392                 }
1393
1394                 /* For now, we do not record post-image data in the patch,
1395                  * and require the object already present in the recipient's
1396                  * object database.
1397                  */
1398                 if (desc->buffer) {
1399                         free(desc->buffer);
1400                         desc->alloc = desc->size = 0;
1401                 }
1402                 get_sha1_hex(patch->new_sha1_prefix, sha1);
1403
1404                 if (memcmp(sha1, null_sha1, 20)) {
1405                         char type[10];
1406                         unsigned long size;
1407
1408                         desc->buffer = read_sha1_file(sha1, type, &size);
1409                         if (!desc->buffer)
1410                                 return error("the necessary postimage %s for "
1411                                              "'%s' does not exist",
1412                                              patch->new_sha1_prefix, name);
1413                         desc->alloc = desc->size = size;
1414                 }
1415
1416                 return 0;
1417         }
1418
1419         while (frag) {
1420                 if (apply_one_fragment(desc, frag) < 0)
1421                         return error("patch failed: %s:%ld",
1422                                      name, frag->oldpos);
1423                 frag = frag->next;
1424         }
1425         return 0;
1426 }
1427
1428 static int apply_data(struct patch *patch, struct stat *st)
1429 {
1430         char *buf;
1431         unsigned long size, alloc;
1432         struct buffer_desc desc;
1433
1434         size = 0;
1435         alloc = 0;
1436         buf = NULL;
1437         if (patch->old_name) {
1438                 size = st->st_size;
1439                 alloc = size + 8192;
1440                 buf = xmalloc(alloc);
1441                 if (read_old_data(st, patch->old_name, buf, alloc) != size)
1442                         return error("read of %s failed", patch->old_name);
1443         }
1444
1445         desc.size = size;
1446         desc.alloc = alloc;
1447         desc.buffer = buf;
1448         if (apply_fragments(&desc, patch) < 0)
1449                 return -1;
1450         patch->result = desc.buffer;
1451         patch->resultsize = desc.size;
1452
1453         if (patch->is_delete && patch->resultsize)
1454                 return error("removal patch leaves file contents");
1455
1456         return 0;
1457 }
1458
1459 static int check_patch(struct patch *patch)
1460 {
1461         struct stat st;
1462         const char *old_name = patch->old_name;
1463         const char *new_name = patch->new_name;
1464         const char *name = old_name ? old_name : new_name;
1465
1466         if (old_name) {
1467                 int changed;
1468                 int stat_ret = lstat(old_name, &st);
1469
1470                 if (check_index) {
1471                         int pos = cache_name_pos(old_name, strlen(old_name));
1472                         if (pos < 0)
1473                                 return error("%s: does not exist in index",
1474                                              old_name);
1475                         if (stat_ret < 0) {
1476                                 struct checkout costate;
1477                                 if (errno != ENOENT)
1478                                         return error("%s: %s", old_name,
1479                                                      strerror(errno));
1480                                 /* checkout */
1481                                 costate.base_dir = "";
1482                                 costate.base_dir_len = 0;
1483                                 costate.force = 0;
1484                                 costate.quiet = 0;
1485                                 costate.not_new = 0;
1486                                 costate.refresh_cache = 1;
1487                                 if (checkout_entry(active_cache[pos],
1488                                                    &costate,
1489                                                    NULL) ||
1490                                     lstat(old_name, &st))
1491                                         return -1;
1492                         }
1493
1494                         changed = ce_match_stat(active_cache[pos], &st, 1);
1495                         if (changed)
1496                                 return error("%s: does not match index",
1497                                              old_name);
1498                 }
1499                 else if (stat_ret < 0)
1500                         return error("%s: %s", old_name, strerror(errno));
1501
1502                 if (patch->is_new < 0)
1503                         patch->is_new = 0;
1504                 st.st_mode = ntohl(create_ce_mode(st.st_mode));
1505                 if (!patch->old_mode)
1506                         patch->old_mode = st.st_mode;
1507                 if ((st.st_mode ^ patch->old_mode) & S_IFMT)
1508                         return error("%s: wrong type", old_name);
1509                 if (st.st_mode != patch->old_mode)
1510                         fprintf(stderr, "warning: %s has type %o, expected %o\n",
1511                                 old_name, st.st_mode, patch->old_mode);
1512         }
1513
1514         if (new_name && (patch->is_new | patch->is_rename | patch->is_copy)) {
1515                 if (check_index && cache_name_pos(new_name, strlen(new_name)) >= 0)
1516                         return error("%s: already exists in index", new_name);
1517                 if (!lstat(new_name, &st))
1518                         return error("%s: already exists in working directory", new_name);
1519                 if (errno != ENOENT)
1520                         return error("%s: %s", new_name, strerror(errno));
1521                 if (!patch->new_mode) {
1522                         if (patch->is_new)
1523                                 patch->new_mode = S_IFREG | 0644;
1524                         else
1525                                 patch->new_mode = patch->old_mode;
1526                 }
1527         }
1528
1529         if (new_name && old_name) {
1530                 int same = !strcmp(old_name, new_name);
1531                 if (!patch->new_mode)
1532                         patch->new_mode = patch->old_mode;
1533                 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
1534                         return error("new mode (%o) of %s does not match old mode (%o)%s%s",
1535                                 patch->new_mode, new_name, patch->old_mode,
1536                                 same ? "" : " of ", same ? "" : old_name);
1537         }       
1538
1539         if (apply_data(patch, &st) < 0)
1540                 return error("%s: patch does not apply", name);
1541         return 0;
1542 }
1543
1544 static int check_patch_list(struct patch *patch)
1545 {
1546         int error = 0;
1547
1548         for (;patch ; patch = patch->next)
1549                 error |= check_patch(patch);
1550         return error;
1551 }
1552
1553 static inline int is_null_sha1(const unsigned char *sha1)
1554 {
1555         return !memcmp(sha1, null_sha1, 20);
1556 }
1557
1558 static void show_index_list(struct patch *list)
1559 {
1560         struct patch *patch;
1561
1562         /* Once we start supporting the reverse patch, it may be
1563          * worth showing the new sha1 prefix, but until then...
1564          */
1565         for (patch = list; patch; patch = patch->next) {
1566                 const unsigned char *sha1_ptr;
1567                 unsigned char sha1[20];
1568                 const char *name;
1569
1570                 name = patch->old_name ? patch->old_name : patch->new_name;
1571                 if (patch->is_new)
1572                         sha1_ptr = null_sha1;
1573                 else if (get_sha1(patch->old_sha1_prefix, sha1))
1574                         die("sha1 information is lacking or useless (%s).",
1575                             name);
1576                 else
1577                         sha1_ptr = sha1;
1578
1579                 printf("%06o %s ",patch->old_mode, sha1_to_hex(sha1_ptr));
1580                 if (line_termination && quote_c_style(name, NULL, NULL, 0))
1581                         quote_c_style(name, NULL, stdout, 0);
1582                 else
1583                         fputs(name, stdout);
1584                 putchar(line_termination);
1585         }
1586 }
1587
1588 static void stat_patch_list(struct patch *patch)
1589 {
1590         int files, adds, dels;
1591
1592         for (files = adds = dels = 0 ; patch ; patch = patch->next) {
1593                 files++;
1594                 adds += patch->lines_added;
1595                 dels += patch->lines_deleted;
1596                 show_stats(patch);
1597         }
1598
1599         printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
1600 }
1601
1602 static void numstat_patch_list(struct patch *patch)
1603 {
1604         for ( ; patch; patch = patch->next) {
1605                 const char *name;
1606                 name = patch->old_name ? patch->old_name : patch->new_name;
1607                 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
1608                 if (line_termination && quote_c_style(name, NULL, NULL, 0))
1609                         quote_c_style(name, NULL, stdout, 0);
1610                 else
1611                         fputs(name, stdout);
1612                 putchar('\n');
1613         }
1614 }
1615
1616 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
1617 {
1618         if (mode)
1619                 printf(" %s mode %06o %s\n", newdelete, mode, name);
1620         else
1621                 printf(" %s %s\n", newdelete, name);
1622 }
1623
1624 static void show_mode_change(struct patch *p, int show_name)
1625 {
1626         if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
1627                 if (show_name)
1628                         printf(" mode change %06o => %06o %s\n",
1629                                p->old_mode, p->new_mode, p->new_name);
1630                 else
1631                         printf(" mode change %06o => %06o\n",
1632                                p->old_mode, p->new_mode);
1633         }
1634 }
1635
1636 static void show_rename_copy(struct patch *p)
1637 {
1638         const char *renamecopy = p->is_rename ? "rename" : "copy";
1639         const char *old, *new;
1640
1641         /* Find common prefix */
1642         old = p->old_name;
1643         new = p->new_name;
1644         while (1) {
1645                 const char *slash_old, *slash_new;
1646                 slash_old = strchr(old, '/');
1647                 slash_new = strchr(new, '/');
1648                 if (!slash_old ||
1649                     !slash_new ||
1650                     slash_old - old != slash_new - new ||
1651                     memcmp(old, new, slash_new - new))
1652                         break;
1653                 old = slash_old + 1;
1654                 new = slash_new + 1;
1655         }
1656         /* p->old_name thru old is the common prefix, and old and new
1657          * through the end of names are renames
1658          */
1659         if (old != p->old_name)
1660                 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
1661                        (int)(old - p->old_name), p->old_name,
1662                        old, new, p->score);
1663         else
1664                 printf(" %s %s => %s (%d%%)\n", renamecopy,
1665                        p->old_name, p->new_name, p->score);
1666         show_mode_change(p, 0);
1667 }
1668
1669 static void summary_patch_list(struct patch *patch)
1670 {
1671         struct patch *p;
1672
1673         for (p = patch; p; p = p->next) {
1674                 if (p->is_new)
1675                         show_file_mode_name("create", p->new_mode, p->new_name);
1676                 else if (p->is_delete)
1677                         show_file_mode_name("delete", p->old_mode, p->old_name);
1678                 else {
1679                         if (p->is_rename || p->is_copy)
1680                                 show_rename_copy(p);
1681                         else {
1682                                 if (p->score) {
1683                                         printf(" rewrite %s (%d%%)\n",
1684                                                p->new_name, p->score);
1685                                         show_mode_change(p, 0);
1686                                 }
1687                                 else
1688                                         show_mode_change(p, 1);
1689                         }
1690                 }
1691         }
1692 }
1693
1694 static void patch_stats(struct patch *patch)
1695 {
1696         int lines = patch->lines_added + patch->lines_deleted;
1697
1698         if (lines > max_change)
1699                 max_change = lines;
1700         if (patch->old_name) {
1701                 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
1702                 if (!len)
1703                         len = strlen(patch->old_name);
1704                 if (len > max_len)
1705                         max_len = len;
1706         }
1707         if (patch->new_name) {
1708                 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
1709                 if (!len)
1710                         len = strlen(patch->new_name);
1711                 if (len > max_len)
1712                         max_len = len;
1713         }
1714 }
1715
1716 static void remove_file(struct patch *patch)
1717 {
1718         if (write_index) {
1719                 if (remove_file_from_cache(patch->old_name) < 0)
1720                         die("unable to remove %s from index", patch->old_name);
1721                 cache_tree_invalidate_path(active_cache_tree, patch->old_name);
1722         }
1723         unlink(patch->old_name);
1724 }
1725
1726 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
1727 {
1728         struct stat st;
1729         struct cache_entry *ce;
1730         int namelen = strlen(path);
1731         unsigned ce_size = cache_entry_size(namelen);
1732
1733         if (!write_index)
1734                 return;
1735
1736         ce = xcalloc(1, ce_size);
1737         memcpy(ce->name, path, namelen);
1738         ce->ce_mode = create_ce_mode(mode);
1739         ce->ce_flags = htons(namelen);
1740         if (lstat(path, &st) < 0)
1741                 die("unable to stat newly created file %s", path);
1742         fill_stat_cache_info(ce, &st);
1743         if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
1744                 die("unable to create backing store for newly created file %s", path);
1745         if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
1746                 die("unable to add cache entry for %s", path);
1747 }
1748
1749 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
1750 {
1751         int fd;
1752
1753         if (S_ISLNK(mode))
1754                 return symlink(buf, path);
1755         fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
1756         if (fd < 0)
1757                 return -1;
1758         while (size) {
1759                 int written = xwrite(fd, buf, size);
1760                 if (written < 0)
1761                         die("writing file %s: %s", path, strerror(errno));
1762                 if (!written)
1763                         die("out of space writing file %s", path);
1764                 buf += written;
1765                 size -= written;
1766         }
1767         if (close(fd) < 0)
1768                 die("closing file %s: %s", path, strerror(errno));
1769         return 0;
1770 }
1771
1772 /*
1773  * We optimistically assume that the directories exist,
1774  * which is true 99% of the time anyway. If they don't,
1775  * we create them and try again.
1776  */
1777 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
1778 {
1779         if (!try_create_file(path, mode, buf, size))
1780                 return;
1781
1782         if (errno == ENOENT) {
1783                 if (safe_create_leading_directories(path))
1784                         return;
1785                 if (!try_create_file(path, mode, buf, size))
1786                         return;
1787         }
1788
1789         if (errno == EEXIST) {
1790                 unsigned int nr = getpid();
1791
1792                 for (;;) {
1793                         const char *newpath;
1794                         newpath = mkpath("%s~%u", path, nr);
1795                         if (!try_create_file(newpath, mode, buf, size)) {
1796                                 if (!rename(newpath, path))
1797                                         return;
1798                                 unlink(newpath);
1799                                 break;
1800                         }
1801                         if (errno != EEXIST)
1802                                 break;
1803                         ++nr;
1804                 }
1805         }
1806         die("unable to write file %s mode %o", path, mode);
1807 }
1808
1809 static void create_file(struct patch *patch)
1810 {
1811         char *path = patch->new_name;
1812         unsigned mode = patch->new_mode;
1813         unsigned long size = patch->resultsize;
1814         char *buf = patch->result;
1815
1816         if (!mode)
1817                 mode = S_IFREG | 0644;
1818         create_one_file(path, mode, buf, size);
1819         add_index_file(path, mode, buf, size);
1820         cache_tree_invalidate_path(active_cache_tree, path);
1821 }
1822
1823 static void write_out_one_result(struct patch *patch)
1824 {
1825         if (patch->is_delete > 0) {
1826                 remove_file(patch);
1827                 return;
1828         }
1829         if (patch->is_new > 0 || patch->is_copy) {
1830                 create_file(patch);
1831                 return;
1832         }
1833         /*
1834          * Rename or modification boils down to the same
1835          * thing: remove the old, write the new
1836          */
1837         remove_file(patch);
1838         create_file(patch);
1839 }
1840
1841 static void write_out_results(struct patch *list, int skipped_patch)
1842 {
1843         if (!list && !skipped_patch)
1844                 die("No changes");
1845
1846         while (list) {
1847                 write_out_one_result(list);
1848                 list = list->next;
1849         }
1850 }
1851
1852 static struct cache_file cache_file;
1853
1854 static struct excludes {
1855         struct excludes *next;
1856         const char *path;
1857 } *excludes;
1858
1859 static int use_patch(struct patch *p)
1860 {
1861         const char *pathname = p->new_name ? p->new_name : p->old_name;
1862         struct excludes *x = excludes;
1863         while (x) {
1864                 if (fnmatch(x->path, pathname, 0) == 0)
1865                         return 0;
1866                 x = x->next;
1867         }
1868         if (0 < prefix_length) {
1869                 int pathlen = strlen(pathname);
1870                 if (pathlen <= prefix_length ||
1871                     memcmp(prefix, pathname, prefix_length))
1872                         return 0;
1873         }
1874         return 1;
1875 }
1876
1877 static int apply_patch(int fd, const char *filename)
1878 {
1879         int newfd;
1880         unsigned long offset, size;
1881         char *buffer = read_patch_file(fd, &size);
1882         struct patch *list = NULL, **listp = &list;
1883         int skipped_patch = 0;
1884
1885         patch_input_file = filename;
1886         if (!buffer)
1887                 return -1;
1888         offset = 0;
1889         while (size > 0) {
1890                 struct patch *patch;
1891                 int nr;
1892
1893                 patch = xcalloc(1, sizeof(*patch));
1894                 nr = parse_chunk(buffer + offset, size, patch);
1895                 if (nr < 0)
1896                         break;
1897                 if (use_patch(patch)) {
1898                         patch_stats(patch);
1899                         *listp = patch;
1900                         listp = &patch->next;
1901                 } else {
1902                         /* perhaps free it a bit better? */
1903                         free(patch);
1904                         skipped_patch++;
1905                 }
1906                 offset += nr;
1907                 size -= nr;
1908         }
1909
1910         newfd = -1;
1911         if (whitespace_error && (new_whitespace == error_on_whitespace))
1912                 apply = 0;
1913
1914         write_index = check_index && apply;
1915         if (write_index)
1916                 newfd = hold_index_file_for_update(&cache_file, get_index_file());
1917         if (check_index) {
1918                 if (read_cache() < 0)
1919                         die("unable to read index file");
1920         }
1921
1922         if ((check || apply) && check_patch_list(list) < 0)
1923                 exit(1);
1924
1925         if (apply)
1926                 write_out_results(list, skipped_patch);
1927
1928         if (write_index) {
1929                 if (write_cache(newfd, active_cache, active_nr) ||
1930                     commit_index_file(&cache_file))
1931                         die("Unable to write new cachefile");
1932         }
1933
1934         if (show_index_info)
1935                 show_index_list(list);
1936
1937         if (diffstat)
1938                 stat_patch_list(list);
1939
1940         if (numstat)
1941                 numstat_patch_list(list);
1942
1943         if (summary)
1944                 summary_patch_list(list);
1945
1946         free(buffer);
1947         return 0;
1948 }
1949
1950 static int git_apply_config(const char *var, const char *value)
1951 {
1952         if (!strcmp(var, "apply.whitespace")) {
1953                 apply_default_whitespace = strdup(value);
1954                 return 0;
1955         }
1956         return git_default_config(var, value);
1957 }
1958
1959
1960 int main(int argc, char **argv)
1961 {
1962         int i;
1963         int read_stdin = 1;
1964         const char *whitespace_option = NULL;
1965
1966         for (i = 1; i < argc; i++) {
1967                 const char *arg = argv[i];
1968                 char *end;
1969                 int fd;
1970
1971                 if (!strcmp(arg, "-")) {
1972                         apply_patch(0, "<stdin>");
1973                         read_stdin = 0;
1974                         continue;
1975                 }
1976                 if (!strncmp(arg, "--exclude=", 10)) {
1977                         struct excludes *x = xmalloc(sizeof(*x));
1978                         x->path = arg + 10;
1979                         x->next = excludes;
1980                         excludes = x;
1981                         continue;
1982                 }
1983                 if (!strncmp(arg, "-p", 2)) {
1984                         p_value = atoi(arg + 2);
1985                         continue;
1986                 }
1987                 if (!strcmp(arg, "--no-add")) {
1988                         no_add = 1;
1989                         continue;
1990                 }
1991                 if (!strcmp(arg, "--stat")) {
1992                         apply = 0;
1993                         diffstat = 1;
1994                         continue;
1995                 }
1996                 if (!strcmp(arg, "--allow-binary-replacement")) {
1997                         allow_binary_replacement = 1;
1998                         continue;
1999                 }
2000                 if (!strcmp(arg, "--numstat")) {
2001                         apply = 0;
2002                         numstat = 1;
2003                         continue;
2004                 }
2005                 if (!strcmp(arg, "--summary")) {
2006                         apply = 0;
2007                         summary = 1;
2008                         continue;
2009                 }
2010                 if (!strcmp(arg, "--check")) {
2011                         apply = 0;
2012                         check = 1;
2013                         continue;
2014                 }
2015                 if (!strcmp(arg, "--index")) {
2016                         check_index = 1;
2017                         continue;
2018                 }
2019                 if (!strcmp(arg, "--apply")) {
2020                         apply = 1;
2021                         continue;
2022                 }
2023                 if (!strcmp(arg, "--index-info")) {
2024                         apply = 0;
2025                         show_index_info = 1;
2026                         continue;
2027                 }
2028                 if (!strcmp(arg, "-z")) {
2029                         line_termination = 0;
2030                         continue;
2031                 }
2032                 if (!strncmp(arg, "-C", 2)) {
2033                         p_context = strtoul(arg + 2, &end, 0);
2034                         if (*end != '\0')
2035                                 die("unrecognized context count '%s'", arg + 2);
2036                         continue;
2037                 }
2038                 if (!strncmp(arg, "--whitespace=", 13)) {
2039                         whitespace_option = arg + 13;
2040                         parse_whitespace_option(arg + 13);
2041                         continue;
2042                 }
2043
2044                 if (check_index && prefix_length < 0) {
2045                         prefix = setup_git_directory();
2046                         prefix_length = prefix ? strlen(prefix) : 0;
2047                         git_config(git_apply_config);
2048                         if (!whitespace_option && apply_default_whitespace)
2049                                 parse_whitespace_option(apply_default_whitespace);
2050                 }
2051                 if (0 < prefix_length)
2052                         arg = prefix_filename(prefix, prefix_length, arg);
2053
2054                 fd = open(arg, O_RDONLY);
2055                 if (fd < 0)
2056                         usage(apply_usage);
2057                 read_stdin = 0;
2058                 set_default_whitespace_mode(whitespace_option);
2059                 apply_patch(fd, arg);
2060                 close(fd);
2061         }
2062         set_default_whitespace_mode(whitespace_option);
2063         if (read_stdin)
2064                 apply_patch(0, "<stdin>");
2065         if (whitespace_error) {
2066                 if (squelch_whitespace_errors &&
2067                     squelch_whitespace_errors < whitespace_error) {
2068                         int squelched =
2069                                 whitespace_error - squelch_whitespace_errors;
2070                         fprintf(stderr, "warning: squelched %d whitespace error%s\n",
2071                                 squelched,
2072                                 squelched == 1 ? "" : "s");
2073                 }
2074                 if (new_whitespace == error_on_whitespace)
2075                         die("%d line%s add%s trailing whitespaces.",
2076                             whitespace_error,
2077                             whitespace_error == 1 ? "" : "s",
2078                             whitespace_error == 1 ? "s" : "");
2079                 if (applied_after_stripping)
2080                         fprintf(stderr, "warning: %d line%s applied after"
2081                                 " stripping trailing whitespaces.\n",
2082                                 applied_after_stripping,
2083                                 applied_after_stripping == 1 ? "" : "s");
2084                 else if (whitespace_error)
2085                         fprintf(stderr, "warning: %d line%s add%s trailing"
2086                                 " whitespaces.\n",
2087                                 whitespace_error,
2088                                 whitespace_error == 1 ? "" : "s",
2089                                 whitespace_error == 1 ? "s" : "");
2090         }
2091         return 0;
2092 }