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