strbuf change: be sure ->buf is never ever NULL.
[git] / builtin-apply.c
1 /*
2  * apply.c
3  *
4  * Copyright (C) Linus Torvalds, 2005
5  *
6  * This applies patches on top of some (arbitrary) version of the SCM.
7  *
8  */
9 #include "cache.h"
10 #include "cache-tree.h"
11 #include "quote.h"
12 #include "blob.h"
13 #include "delta.h"
14 #include "builtin.h"
15
16 /*
17  *  --check turns on checking that the working tree matches the
18  *    files that are being modified, but doesn't apply the patch
19  *  --stat does just a diffstat, and doesn't actually apply
20  *  --numstat does numeric diffstat, and doesn't actually apply
21  *  --index-info shows the old and new index info for paths if available.
22  *  --index updates the cache as well.
23  *  --cached updates only the cache without ever touching the working tree.
24  */
25 static const char *prefix;
26 static int prefix_length = -1;
27 static int newfd = -1;
28
29 static int unidiff_zero;
30 static int p_value = 1;
31 static int p_value_known;
32 static int check_index;
33 static int update_index;
34 static int cached;
35 static int diffstat;
36 static int numstat;
37 static int summary;
38 static int check;
39 static int apply = 1;
40 static int apply_in_reverse;
41 static int apply_with_reject;
42 static int apply_verbosely;
43 static int no_add;
44 static int show_index_info;
45 static int line_termination = '\n';
46 static unsigned long p_context = ULONG_MAX;
47 static const char apply_usage[] =
48 "git-apply [--stat] [--numstat] [--summary] [--check] [--index] [--cached] [--apply] [--no-add] [--index-info] [--allow-binary-replacement] [--reverse] [--reject] [--verbose] [-z] [-pNUM] [-CNUM] [--whitespace=<nowarn|warn|error|error-all|strip>] <patch>...";
49
50 static enum whitespace_eol {
51         nowarn_whitespace,
52         warn_on_whitespace,
53         error_on_whitespace,
54         strip_whitespace,
55 } new_whitespace = warn_on_whitespace;
56 static int whitespace_error;
57 static int squelch_whitespace_errors = 5;
58 static int applied_after_fixing_ws;
59 static const char *patch_input_file;
60
61 static void parse_whitespace_option(const char *option)
62 {
63         if (!option) {
64                 new_whitespace = warn_on_whitespace;
65                 return;
66         }
67         if (!strcmp(option, "warn")) {
68                 new_whitespace = warn_on_whitespace;
69                 return;
70         }
71         if (!strcmp(option, "nowarn")) {
72                 new_whitespace = nowarn_whitespace;
73                 return;
74         }
75         if (!strcmp(option, "error")) {
76                 new_whitespace = error_on_whitespace;
77                 return;
78         }
79         if (!strcmp(option, "error-all")) {
80                 new_whitespace = error_on_whitespace;
81                 squelch_whitespace_errors = 0;
82                 return;
83         }
84         if (!strcmp(option, "strip")) {
85                 new_whitespace = strip_whitespace;
86                 return;
87         }
88         die("unrecognized whitespace option '%s'", option);
89 }
90
91 static void set_default_whitespace_mode(const char *whitespace_option)
92 {
93         if (!whitespace_option && !apply_default_whitespace) {
94                 new_whitespace = (apply
95                                   ? warn_on_whitespace
96                                   : nowarn_whitespace);
97         }
98 }
99
100 /*
101  * For "diff-stat" like behaviour, we keep track of the biggest change
102  * we've seen, and the longest filename. That allows us to do simple
103  * scaling.
104  */
105 static int max_change, max_len;
106
107 /*
108  * Various "current state", notably line numbers and what
109  * file (and how) we're patching right now.. The "is_xxxx"
110  * things are flags, where -1 means "don't know yet".
111  */
112 static int linenr = 1;
113
114 /*
115  * This represents one "hunk" from a patch, starting with
116  * "@@ -oldpos,oldlines +newpos,newlines @@" marker.  The
117  * patch text is pointed at by patch, and its byte length
118  * is stored in size.  leading and trailing are the number
119  * of context lines.
120  */
121 struct fragment {
122         unsigned long leading, trailing;
123         unsigned long oldpos, oldlines;
124         unsigned long newpos, newlines;
125         const char *patch;
126         int size;
127         int rejected;
128         struct fragment *next;
129 };
130
131 /*
132  * When dealing with a binary patch, we reuse "leading" field
133  * to store the type of the binary hunk, either deflated "delta"
134  * or deflated "literal".
135  */
136 #define binary_patch_method leading
137 #define BINARY_DELTA_DEFLATED   1
138 #define BINARY_LITERAL_DEFLATED 2
139
140 struct patch {
141         char *new_name, *old_name, *def_name;
142         unsigned int old_mode, new_mode;
143         int is_new, is_delete;  /* -1 = unknown, 0 = false, 1 = true */
144         int rejected;
145         unsigned long deflate_origlen;
146         int lines_added, lines_deleted;
147         int score;
148         unsigned int is_toplevel_relative:1;
149         unsigned int inaccurate_eof:1;
150         unsigned int is_binary:1;
151         unsigned int is_copy:1;
152         unsigned int is_rename:1;
153         struct fragment *fragments;
154         char *result;
155         unsigned long resultsize;
156         char old_sha1_prefix[41];
157         char new_sha1_prefix[41];
158         struct patch *next;
159 };
160
161 static void say_patch_name(FILE *output, const char *pre, struct patch *patch, const char *post)
162 {
163         fputs(pre, output);
164         if (patch->old_name && patch->new_name &&
165             strcmp(patch->old_name, patch->new_name)) {
166                 quote_c_style(patch->old_name, NULL, output, 0);
167                 fputs(" => ", output);
168                 quote_c_style(patch->new_name, NULL, output, 0);
169         } else {
170                 const char *n = patch->new_name;
171                 if (!n)
172                         n = patch->old_name;
173                 quote_c_style(n, NULL, output, 0);
174         }
175         fputs(post, output);
176 }
177
178 #define CHUNKSIZE (8192)
179 #define SLOP (16)
180
181 static void *read_patch_file(int fd, size_t *sizep)
182 {
183         struct strbuf buf;
184
185         strbuf_init(&buf, 0);
186         if (strbuf_read(&buf, fd, 0) < 0)
187                 die("git-apply: read returned %s", strerror(errno));
188
189         /*
190          * Make sure that we have some slop in the buffer
191          * so that we can do speculative "memcmp" etc, and
192          * see to it that it is NUL-filled.
193          */
194         strbuf_grow(&buf, SLOP);
195         memset(buf.buf + buf.len, 0, SLOP);
196         return strbuf_detach(&buf, sizep);
197 }
198
199 static unsigned long linelen(const char *buffer, unsigned long size)
200 {
201         unsigned long len = 0;
202         while (size--) {
203                 len++;
204                 if (*buffer++ == '\n')
205                         break;
206         }
207         return len;
208 }
209
210 static int is_dev_null(const char *str)
211 {
212         return !memcmp("/dev/null", str, 9) && isspace(str[9]);
213 }
214
215 #define TERM_SPACE      1
216 #define TERM_TAB        2
217
218 static int name_terminate(const char *name, int namelen, int c, int terminate)
219 {
220         if (c == ' ' && !(terminate & TERM_SPACE))
221                 return 0;
222         if (c == '\t' && !(terminate & TERM_TAB))
223                 return 0;
224
225         return 1;
226 }
227
228 static char *find_name(const char *line, char *def, int p_value, int terminate)
229 {
230         int len;
231         const char *start = line;
232
233         if (*line == '"') {
234                 struct strbuf name;
235
236                 /* Proposed "new-style" GNU patch/diff format; see
237                  * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
238                  */
239                 strbuf_init(&name, 0);
240                 if (!unquote_c_style(&name, line, NULL)) {
241                         char *cp;
242
243                         for (cp = name.buf; p_value; p_value--) {
244                                 cp = strchr(cp, '/');
245                                 if (!cp)
246                                         break;
247                                 cp++;
248                         }
249                         if (cp) {
250                                 /* name can later be freed, so we need
251                                  * to memmove, not just return cp
252                                  */
253                                 strbuf_remove(&name, 0, cp - name.buf);
254                                 free(def);
255                                 return strbuf_detach(&name, NULL);
256                         }
257                 }
258                 strbuf_release(&name);
259         }
260
261         for (;;) {
262                 char c = *line;
263
264                 if (isspace(c)) {
265                         if (c == '\n')
266                                 break;
267                         if (name_terminate(start, line-start, c, terminate))
268                                 break;
269                 }
270                 line++;
271                 if (c == '/' && !--p_value)
272                         start = line;
273         }
274         if (!start)
275                 return def;
276         len = line - start;
277         if (!len)
278                 return def;
279
280         /*
281          * Generally we prefer the shorter name, especially
282          * if the other one is just a variation of that with
283          * something else tacked on to the end (ie "file.orig"
284          * or "file~").
285          */
286         if (def) {
287                 int deflen = strlen(def);
288                 if (deflen < len && !strncmp(start, def, deflen))
289                         return def;
290                 free(def);
291         }
292
293         return xmemdupz(start, len);
294 }
295
296 static int count_slashes(const char *cp)
297 {
298         int cnt = 0;
299         char ch;
300
301         while ((ch = *cp++))
302                 if (ch == '/')
303                         cnt++;
304         return cnt;
305 }
306
307 /*
308  * Given the string after "--- " or "+++ ", guess the appropriate
309  * p_value for the given patch.
310  */
311 static int guess_p_value(const char *nameline)
312 {
313         char *name, *cp;
314         int val = -1;
315
316         if (is_dev_null(nameline))
317                 return -1;
318         name = find_name(nameline, NULL, 0, TERM_SPACE | TERM_TAB);
319         if (!name)
320                 return -1;
321         cp = strchr(name, '/');
322         if (!cp)
323                 val = 0;
324         else if (prefix) {
325                 /*
326                  * Does it begin with "a/$our-prefix" and such?  Then this is
327                  * very likely to apply to our directory.
328                  */
329                 if (!strncmp(name, prefix, prefix_length))
330                         val = count_slashes(prefix);
331                 else {
332                         cp++;
333                         if (!strncmp(cp, prefix, prefix_length))
334                                 val = count_slashes(prefix) + 1;
335                 }
336         }
337         free(name);
338         return val;
339 }
340
341 /*
342  * Get the name etc info from the --/+++ lines of a traditional patch header
343  *
344  * FIXME! The end-of-filename heuristics are kind of screwy. For existing
345  * files, we can happily check the index for a match, but for creating a
346  * new file we should try to match whatever "patch" does. I have no idea.
347  */
348 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
349 {
350         char *name;
351
352         first += 4;     /* skip "--- " */
353         second += 4;    /* skip "+++ " */
354         if (!p_value_known) {
355                 int p, q;
356                 p = guess_p_value(first);
357                 q = guess_p_value(second);
358                 if (p < 0) p = q;
359                 if (0 <= p && p == q) {
360                         p_value = p;
361                         p_value_known = 1;
362                 }
363         }
364         if (is_dev_null(first)) {
365                 patch->is_new = 1;
366                 patch->is_delete = 0;
367                 name = find_name(second, NULL, p_value, TERM_SPACE | TERM_TAB);
368                 patch->new_name = name;
369         } else if (is_dev_null(second)) {
370                 patch->is_new = 0;
371                 patch->is_delete = 1;
372                 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
373                 patch->old_name = name;
374         } else {
375                 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
376                 name = find_name(second, name, p_value, TERM_SPACE | TERM_TAB);
377                 patch->old_name = patch->new_name = name;
378         }
379         if (!name)
380                 die("unable to find filename in patch at line %d", linenr);
381 }
382
383 static int gitdiff_hdrend(const char *line, struct patch *patch)
384 {
385         return -1;
386 }
387
388 /*
389  * We're anal about diff header consistency, to make
390  * sure that we don't end up having strange ambiguous
391  * patches floating around.
392  *
393  * As a result, gitdiff_{old|new}name() will check
394  * their names against any previous information, just
395  * to make sure..
396  */
397 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
398 {
399         if (!orig_name && !isnull)
400                 return find_name(line, NULL, p_value, TERM_TAB);
401
402         if (orig_name) {
403                 int len;
404                 const char *name;
405                 char *another;
406                 name = orig_name;
407                 len = strlen(name);
408                 if (isnull)
409                         die("git-apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
410                 another = find_name(line, NULL, p_value, TERM_TAB);
411                 if (!another || memcmp(another, name, len))
412                         die("git-apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
413                 free(another);
414                 return orig_name;
415         }
416         else {
417                 /* expect "/dev/null" */
418                 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
419                         die("git-apply: bad git-diff - expected /dev/null on line %d", linenr);
420                 return NULL;
421         }
422 }
423
424 static int gitdiff_oldname(const char *line, struct patch *patch)
425 {
426         patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
427         return 0;
428 }
429
430 static int gitdiff_newname(const char *line, struct patch *patch)
431 {
432         patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
433         return 0;
434 }
435
436 static int gitdiff_oldmode(const char *line, struct patch *patch)
437 {
438         patch->old_mode = strtoul(line, NULL, 8);
439         return 0;
440 }
441
442 static int gitdiff_newmode(const char *line, struct patch *patch)
443 {
444         patch->new_mode = strtoul(line, NULL, 8);
445         return 0;
446 }
447
448 static int gitdiff_delete(const char *line, struct patch *patch)
449 {
450         patch->is_delete = 1;
451         patch->old_name = patch->def_name;
452         return gitdiff_oldmode(line, patch);
453 }
454
455 static int gitdiff_newfile(const char *line, struct patch *patch)
456 {
457         patch->is_new = 1;
458         patch->new_name = patch->def_name;
459         return gitdiff_newmode(line, patch);
460 }
461
462 static int gitdiff_copysrc(const char *line, struct patch *patch)
463 {
464         patch->is_copy = 1;
465         patch->old_name = find_name(line, NULL, 0, 0);
466         return 0;
467 }
468
469 static int gitdiff_copydst(const char *line, struct patch *patch)
470 {
471         patch->is_copy = 1;
472         patch->new_name = find_name(line, NULL, 0, 0);
473         return 0;
474 }
475
476 static int gitdiff_renamesrc(const char *line, struct patch *patch)
477 {
478         patch->is_rename = 1;
479         patch->old_name = find_name(line, NULL, 0, 0);
480         return 0;
481 }
482
483 static int gitdiff_renamedst(const char *line, struct patch *patch)
484 {
485         patch->is_rename = 1;
486         patch->new_name = find_name(line, NULL, 0, 0);
487         return 0;
488 }
489
490 static int gitdiff_similarity(const char *line, struct patch *patch)
491 {
492         if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
493                 patch->score = 0;
494         return 0;
495 }
496
497 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
498 {
499         if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
500                 patch->score = 0;
501         return 0;
502 }
503
504 static int gitdiff_index(const char *line, struct patch *patch)
505 {
506         /* index line is N hexadecimal, "..", N hexadecimal,
507          * and optional space with octal mode.
508          */
509         const char *ptr, *eol;
510         int len;
511
512         ptr = strchr(line, '.');
513         if (!ptr || ptr[1] != '.' || 40 < ptr - line)
514                 return 0;
515         len = ptr - line;
516         memcpy(patch->old_sha1_prefix, line, len);
517         patch->old_sha1_prefix[len] = 0;
518
519         line = ptr + 2;
520         ptr = strchr(line, ' ');
521         eol = strchr(line, '\n');
522
523         if (!ptr || eol < ptr)
524                 ptr = eol;
525         len = ptr - line;
526
527         if (40 < len)
528                 return 0;
529         memcpy(patch->new_sha1_prefix, line, len);
530         patch->new_sha1_prefix[len] = 0;
531         if (*ptr == ' ')
532                 patch->new_mode = patch->old_mode = strtoul(ptr+1, NULL, 8);
533         return 0;
534 }
535
536 /*
537  * This is normal for a diff that doesn't change anything: we'll fall through
538  * into the next diff. Tell the parser to break out.
539  */
540 static int gitdiff_unrecognized(const char *line, struct patch *patch)
541 {
542         return -1;
543 }
544
545 static const char *stop_at_slash(const char *line, int llen)
546 {
547         int i;
548
549         for (i = 0; i < llen; i++) {
550                 int ch = line[i];
551                 if (ch == '/')
552                         return line + i;
553         }
554         return NULL;
555 }
556
557 /* This is to extract the same name that appears on "diff --git"
558  * line.  We do not find and return anything if it is a rename
559  * patch, and it is OK because we will find the name elsewhere.
560  * We need to reliably find name only when it is mode-change only,
561  * creation or deletion of an empty file.  In any of these cases,
562  * both sides are the same name under a/ and b/ respectively.
563  */
564 static char *git_header_name(char *line, int llen)
565 {
566         const char *name;
567         const char *second = NULL;
568         size_t len;
569
570         line += strlen("diff --git ");
571         llen -= strlen("diff --git ");
572
573         if (*line == '"') {
574                 const char *cp;
575                 struct strbuf first;
576                 struct strbuf sp;
577
578                 strbuf_init(&first, 0);
579                 strbuf_init(&sp, 0);
580
581                 if (unquote_c_style(&first, line, &second))
582                         goto free_and_fail1;
583
584                 /* advance to the first slash */
585                 cp = stop_at_slash(first.buf, first.len);
586                 /* we do not accept absolute paths */
587                 if (!cp || cp == first.buf)
588                         goto free_and_fail1;
589                 strbuf_remove(&first, 0, cp + 1 - first.buf);
590
591                 /* second points at one past closing dq of name.
592                  * find the second name.
593                  */
594                 while ((second < line + llen) && isspace(*second))
595                         second++;
596
597                 if (line + llen <= second)
598                         goto free_and_fail1;
599                 if (*second == '"') {
600                         if (unquote_c_style(&sp, second, NULL))
601                                 goto free_and_fail1;
602                         cp = stop_at_slash(sp.buf, sp.len);
603                         if (!cp || cp == sp.buf)
604                                 goto free_and_fail1;
605                         /* They must match, otherwise ignore */
606                         if (strcmp(cp + 1, first.buf))
607                                 goto free_and_fail1;
608                         strbuf_release(&sp);
609                         return strbuf_detach(&first, NULL);
610                 }
611
612                 /* unquoted second */
613                 cp = stop_at_slash(second, line + llen - second);
614                 if (!cp || cp == second)
615                         goto free_and_fail1;
616                 cp++;
617                 if (line + llen - cp != first.len + 1 ||
618                     memcmp(first.buf, cp, first.len))
619                         goto free_and_fail1;
620                 return strbuf_detach(&first, NULL);
621
622         free_and_fail1:
623                 strbuf_release(&first);
624                 strbuf_release(&sp);
625                 return NULL;
626         }
627
628         /* unquoted first name */
629         name = stop_at_slash(line, llen);
630         if (!name || name == line)
631                 return NULL;
632         name++;
633
634         /* since the first name is unquoted, a dq if exists must be
635          * the beginning of the second name.
636          */
637         for (second = name; second < line + llen; second++) {
638                 if (*second == '"') {
639                         struct strbuf sp;
640                         const char *np;
641
642                         strbuf_init(&sp, 0);
643                         if (unquote_c_style(&sp, second, NULL))
644                                 goto free_and_fail2;
645
646                         np = stop_at_slash(sp.buf, sp.len);
647                         if (!np || np == sp.buf)
648                                 goto free_and_fail2;
649                         np++;
650
651                         len = sp.buf + sp.len - np;
652                         if (len < second - name &&
653                             !strncmp(np, name, len) &&
654                             isspace(name[len])) {
655                                 /* Good */
656                                 strbuf_remove(&sp, 0, np - sp.buf);
657                                 return strbuf_detach(&sp, NULL);
658                         }
659
660                 free_and_fail2:
661                         strbuf_release(&sp);
662                         return NULL;
663                 }
664         }
665
666         /*
667          * Accept a name only if it shows up twice, exactly the same
668          * form.
669          */
670         for (len = 0 ; ; len++) {
671                 switch (name[len]) {
672                 default:
673                         continue;
674                 case '\n':
675                         return NULL;
676                 case '\t': case ' ':
677                         second = name+len;
678                         for (;;) {
679                                 char c = *second++;
680                                 if (c == '\n')
681                                         return NULL;
682                                 if (c == '/')
683                                         break;
684                         }
685                         if (second[len] == '\n' && !memcmp(name, second, len)) {
686                                 return xmemdupz(name, len);
687                         }
688                 }
689         }
690         return NULL;
691 }
692
693 /* Verify that we recognize the lines following a git header */
694 static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
695 {
696         unsigned long offset;
697
698         /* A git diff has explicit new/delete information, so we don't guess */
699         patch->is_new = 0;
700         patch->is_delete = 0;
701
702         /*
703          * Some things may not have the old name in the
704          * rest of the headers anywhere (pure mode changes,
705          * or removing or adding empty files), so we get
706          * the default name from the header.
707          */
708         patch->def_name = git_header_name(line, len);
709
710         line += len;
711         size -= len;
712         linenr++;
713         for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
714                 static const struct opentry {
715                         const char *str;
716                         int (*fn)(const char *, struct patch *);
717                 } optable[] = {
718                         { "@@ -", gitdiff_hdrend },
719                         { "--- ", gitdiff_oldname },
720                         { "+++ ", gitdiff_newname },
721                         { "old mode ", gitdiff_oldmode },
722                         { "new mode ", gitdiff_newmode },
723                         { "deleted file mode ", gitdiff_delete },
724                         { "new file mode ", gitdiff_newfile },
725                         { "copy from ", gitdiff_copysrc },
726                         { "copy to ", gitdiff_copydst },
727                         { "rename old ", gitdiff_renamesrc },
728                         { "rename new ", gitdiff_renamedst },
729                         { "rename from ", gitdiff_renamesrc },
730                         { "rename to ", gitdiff_renamedst },
731                         { "similarity index ", gitdiff_similarity },
732                         { "dissimilarity index ", gitdiff_dissimilarity },
733                         { "index ", gitdiff_index },
734                         { "", gitdiff_unrecognized },
735                 };
736                 int i;
737
738                 len = linelen(line, size);
739                 if (!len || line[len-1] != '\n')
740                         break;
741                 for (i = 0; i < ARRAY_SIZE(optable); i++) {
742                         const struct opentry *p = optable + i;
743                         int oplen = strlen(p->str);
744                         if (len < oplen || memcmp(p->str, line, oplen))
745                                 continue;
746                         if (p->fn(line + oplen, patch) < 0)
747                                 return offset;
748                         break;
749                 }
750         }
751
752         return offset;
753 }
754
755 static int parse_num(const char *line, unsigned long *p)
756 {
757         char *ptr;
758
759         if (!isdigit(*line))
760                 return 0;
761         *p = strtoul(line, &ptr, 10);
762         return ptr - line;
763 }
764
765 static int parse_range(const char *line, int len, int offset, const char *expect,
766                         unsigned long *p1, unsigned long *p2)
767 {
768         int digits, ex;
769
770         if (offset < 0 || offset >= len)
771                 return -1;
772         line += offset;
773         len -= offset;
774
775         digits = parse_num(line, p1);
776         if (!digits)
777                 return -1;
778
779         offset += digits;
780         line += digits;
781         len -= digits;
782
783         *p2 = 1;
784         if (*line == ',') {
785                 digits = parse_num(line+1, p2);
786                 if (!digits)
787                         return -1;
788
789                 offset += digits+1;
790                 line += digits+1;
791                 len -= digits+1;
792         }
793
794         ex = strlen(expect);
795         if (ex > len)
796                 return -1;
797         if (memcmp(line, expect, ex))
798                 return -1;
799
800         return offset + ex;
801 }
802
803 /*
804  * Parse a unified diff fragment header of the
805  * form "@@ -a,b +c,d @@"
806  */
807 static int parse_fragment_header(char *line, int len, struct fragment *fragment)
808 {
809         int offset;
810
811         if (!len || line[len-1] != '\n')
812                 return -1;
813
814         /* Figure out the number of lines in a fragment */
815         offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
816         offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
817
818         return offset;
819 }
820
821 static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
822 {
823         unsigned long offset, len;
824
825         patch->is_toplevel_relative = 0;
826         patch->is_rename = patch->is_copy = 0;
827         patch->is_new = patch->is_delete = -1;
828         patch->old_mode = patch->new_mode = 0;
829         patch->old_name = patch->new_name = NULL;
830         for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
831                 unsigned long nextlen;
832
833                 len = linelen(line, size);
834                 if (!len)
835                         break;
836
837                 /* Testing this early allows us to take a few shortcuts.. */
838                 if (len < 6)
839                         continue;
840
841                 /*
842                  * Make sure we don't find any unconnected patch fragments.
843                  * That's a sign that we didn't find a header, and that a
844                  * patch has become corrupted/broken up.
845                  */
846                 if (!memcmp("@@ -", line, 4)) {
847                         struct fragment dummy;
848                         if (parse_fragment_header(line, len, &dummy) < 0)
849                                 continue;
850                         die("patch fragment without header at line %d: %.*s",
851                             linenr, (int)len-1, line);
852                 }
853
854                 if (size < len + 6)
855                         break;
856
857                 /*
858                  * Git patch? It might not have a real patch, just a rename
859                  * or mode change, so we handle that specially
860                  */
861                 if (!memcmp("diff --git ", line, 11)) {
862                         int git_hdr_len = parse_git_header(line, len, size, patch);
863                         if (git_hdr_len <= len)
864                                 continue;
865                         if (!patch->old_name && !patch->new_name) {
866                                 if (!patch->def_name)
867                                         die("git diff header lacks filename information (line %d)", linenr);
868                                 patch->old_name = patch->new_name = patch->def_name;
869                         }
870                         patch->is_toplevel_relative = 1;
871                         *hdrsize = git_hdr_len;
872                         return offset;
873                 }
874
875                 /** --- followed by +++ ? */
876                 if (memcmp("--- ", line,  4) || memcmp("+++ ", line + len, 4))
877                         continue;
878
879                 /*
880                  * We only accept unified patches, so we want it to
881                  * at least have "@@ -a,b +c,d @@\n", which is 14 chars
882                  * minimum
883                  */
884                 nextlen = linelen(line + len, size - len);
885                 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
886                         continue;
887
888                 /* Ok, we'll consider it a patch */
889                 parse_traditional_patch(line, line+len, patch);
890                 *hdrsize = len + nextlen;
891                 linenr += 2;
892                 return offset;
893         }
894         return -1;
895 }
896
897 static void check_whitespace(const char *line, int len)
898 {
899         const char *err = "Adds trailing whitespace";
900         int seen_space = 0;
901         int i;
902
903         /*
904          * We know len is at least two, since we have a '+' and we
905          * checked that the last character was a '\n' before calling
906          * this function.  That is, an addition of an empty line would
907          * check the '+' here.  Sneaky...
908          */
909         if (isspace(line[len-2]))
910                 goto error;
911
912         /*
913          * Make sure that there is no space followed by a tab in
914          * indentation.
915          */
916         err = "Space in indent is followed by a tab";
917         for (i = 1; i < len; i++) {
918                 if (line[i] == '\t') {
919                         if (seen_space)
920                                 goto error;
921                 }
922                 else if (line[i] == ' ')
923                         seen_space = 1;
924                 else
925                         break;
926         }
927         return;
928
929  error:
930         whitespace_error++;
931         if (squelch_whitespace_errors &&
932             squelch_whitespace_errors < whitespace_error)
933                 ;
934         else
935                 fprintf(stderr, "%s.\n%s:%d:%.*s\n",
936                         err, patch_input_file, linenr, len-2, line+1);
937 }
938
939
940 /*
941  * Parse a unified diff. Note that this really needs to parse each
942  * fragment separately, since the only way to know the difference
943  * between a "---" that is part of a patch, and a "---" that starts
944  * the next patch is to look at the line counts..
945  */
946 static int parse_fragment(char *line, unsigned long size, struct patch *patch, struct fragment *fragment)
947 {
948         int added, deleted;
949         int len = linelen(line, size), offset;
950         unsigned long oldlines, newlines;
951         unsigned long leading, trailing;
952
953         offset = parse_fragment_header(line, len, fragment);
954         if (offset < 0)
955                 return -1;
956         oldlines = fragment->oldlines;
957         newlines = fragment->newlines;
958         leading = 0;
959         trailing = 0;
960
961         /* Parse the thing.. */
962         line += len;
963         size -= len;
964         linenr++;
965         added = deleted = 0;
966         for (offset = len;
967              0 < size;
968              offset += len, size -= len, line += len, linenr++) {
969                 if (!oldlines && !newlines)
970                         break;
971                 len = linelen(line, size);
972                 if (!len || line[len-1] != '\n')
973                         return -1;
974                 switch (*line) {
975                 default:
976                         return -1;
977                 case '\n': /* newer GNU diff, an empty context line */
978                 case ' ':
979                         oldlines--;
980                         newlines--;
981                         if (!deleted && !added)
982                                 leading++;
983                         trailing++;
984                         break;
985                 case '-':
986                         if (apply_in_reverse &&
987                                         new_whitespace != nowarn_whitespace)
988                                 check_whitespace(line, len);
989                         deleted++;
990                         oldlines--;
991                         trailing = 0;
992                         break;
993                 case '+':
994                         if (!apply_in_reverse &&
995                                         new_whitespace != nowarn_whitespace)
996                                 check_whitespace(line, len);
997                         added++;
998                         newlines--;
999                         trailing = 0;
1000                         break;
1001
1002                 /* We allow "\ No newline at end of file". Depending
1003                  * on locale settings when the patch was produced we
1004                  * don't know what this line looks like. The only
1005                  * thing we do know is that it begins with "\ ".
1006                  * Checking for 12 is just for sanity check -- any
1007                  * l10n of "\ No newline..." is at least that long.
1008                  */
1009                 case '\\':
1010                         if (len < 12 || memcmp(line, "\\ ", 2))
1011                                 return -1;
1012                         break;
1013                 }
1014         }
1015         if (oldlines || newlines)
1016                 return -1;
1017         fragment->leading = leading;
1018         fragment->trailing = trailing;
1019
1020         /* If a fragment ends with an incomplete line, we failed to include
1021          * it in the above loop because we hit oldlines == newlines == 0
1022          * before seeing it.
1023          */
1024         if (12 < size && !memcmp(line, "\\ ", 2))
1025                 offset += linelen(line, size);
1026
1027         patch->lines_added += added;
1028         patch->lines_deleted += deleted;
1029
1030         if (0 < patch->is_new && oldlines)
1031                 return error("new file depends on old contents");
1032         if (0 < patch->is_delete && newlines)
1033                 return error("deleted file still has contents");
1034         return offset;
1035 }
1036
1037 static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
1038 {
1039         unsigned long offset = 0;
1040         unsigned long oldlines = 0, newlines = 0, context = 0;
1041         struct fragment **fragp = &patch->fragments;
1042
1043         while (size > 4 && !memcmp(line, "@@ -", 4)) {
1044                 struct fragment *fragment;
1045                 int len;
1046
1047                 fragment = xcalloc(1, sizeof(*fragment));
1048                 len = parse_fragment(line, size, patch, fragment);
1049                 if (len <= 0)
1050                         die("corrupt patch at line %d", linenr);
1051                 fragment->patch = line;
1052                 fragment->size = len;
1053                 oldlines += fragment->oldlines;
1054                 newlines += fragment->newlines;
1055                 context += fragment->leading + fragment->trailing;
1056
1057                 *fragp = fragment;
1058                 fragp = &fragment->next;
1059
1060                 offset += len;
1061                 line += len;
1062                 size -= len;
1063         }
1064
1065         /*
1066          * If something was removed (i.e. we have old-lines) it cannot
1067          * be creation, and if something was added it cannot be
1068          * deletion.  However, the reverse is not true; --unified=0
1069          * patches that only add are not necessarily creation even
1070          * though they do not have any old lines, and ones that only
1071          * delete are not necessarily deletion.
1072          *
1073          * Unfortunately, a real creation/deletion patch do _not_ have
1074          * any context line by definition, so we cannot safely tell it
1075          * apart with --unified=0 insanity.  At least if the patch has
1076          * more than one hunk it is not creation or deletion.
1077          */
1078         if (patch->is_new < 0 &&
1079             (oldlines || (patch->fragments && patch->fragments->next)))
1080                 patch->is_new = 0;
1081         if (patch->is_delete < 0 &&
1082             (newlines || (patch->fragments && patch->fragments->next)))
1083                 patch->is_delete = 0;
1084         if (!unidiff_zero || context) {
1085                 /* If the user says the patch is not generated with
1086                  * --unified=0, or if we have seen context lines,
1087                  * then not having oldlines means the patch is creation,
1088                  * and not having newlines means the patch is deletion.
1089                  */
1090                 if (patch->is_new < 0 && !oldlines) {
1091                         patch->is_new = 1;
1092                         patch->old_name = NULL;
1093                 }
1094                 if (patch->is_delete < 0 && !newlines) {
1095                         patch->is_delete = 1;
1096                         patch->new_name = NULL;
1097                 }
1098         }
1099
1100         if (0 < patch->is_new && oldlines)
1101                 die("new file %s depends on old contents", patch->new_name);
1102         if (0 < patch->is_delete && newlines)
1103                 die("deleted file %s still has contents", patch->old_name);
1104         if (!patch->is_delete && !newlines && context)
1105                 fprintf(stderr, "** warning: file %s becomes empty but "
1106                         "is not deleted\n", patch->new_name);
1107
1108         return offset;
1109 }
1110
1111 static inline int metadata_changes(struct patch *patch)
1112 {
1113         return  patch->is_rename > 0 ||
1114                 patch->is_copy > 0 ||
1115                 patch->is_new > 0 ||
1116                 patch->is_delete ||
1117                 (patch->old_mode && patch->new_mode &&
1118                  patch->old_mode != patch->new_mode);
1119 }
1120
1121 static char *inflate_it(const void *data, unsigned long size,
1122                         unsigned long inflated_size)
1123 {
1124         z_stream stream;
1125         void *out;
1126         int st;
1127
1128         memset(&stream, 0, sizeof(stream));
1129
1130         stream.next_in = (unsigned char *)data;
1131         stream.avail_in = size;
1132         stream.next_out = out = xmalloc(inflated_size);
1133         stream.avail_out = inflated_size;
1134         inflateInit(&stream);
1135         st = inflate(&stream, Z_FINISH);
1136         if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1137                 free(out);
1138                 return NULL;
1139         }
1140         return out;
1141 }
1142
1143 static struct fragment *parse_binary_hunk(char **buf_p,
1144                                           unsigned long *sz_p,
1145                                           int *status_p,
1146                                           int *used_p)
1147 {
1148         /* Expect a line that begins with binary patch method ("literal"
1149          * or "delta"), followed by the length of data before deflating.
1150          * a sequence of 'length-byte' followed by base-85 encoded data
1151          * should follow, terminated by a newline.
1152          *
1153          * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1154          * and we would limit the patch line to 66 characters,
1155          * so one line can fit up to 13 groups that would decode
1156          * to 52 bytes max.  The length byte 'A'-'Z' corresponds
1157          * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1158          */
1159         int llen, used;
1160         unsigned long size = *sz_p;
1161         char *buffer = *buf_p;
1162         int patch_method;
1163         unsigned long origlen;
1164         char *data = NULL;
1165         int hunk_size = 0;
1166         struct fragment *frag;
1167
1168         llen = linelen(buffer, size);
1169         used = llen;
1170
1171         *status_p = 0;
1172
1173         if (!prefixcmp(buffer, "delta ")) {
1174                 patch_method = BINARY_DELTA_DEFLATED;
1175                 origlen = strtoul(buffer + 6, NULL, 10);
1176         }
1177         else if (!prefixcmp(buffer, "literal ")) {
1178                 patch_method = BINARY_LITERAL_DEFLATED;
1179                 origlen = strtoul(buffer + 8, NULL, 10);
1180         }
1181         else
1182                 return NULL;
1183
1184         linenr++;
1185         buffer += llen;
1186         while (1) {
1187                 int byte_length, max_byte_length, newsize;
1188                 llen = linelen(buffer, size);
1189                 used += llen;
1190                 linenr++;
1191                 if (llen == 1) {
1192                         /* consume the blank line */
1193                         buffer++;
1194                         size--;
1195                         break;
1196                 }
1197                 /* Minimum line is "A00000\n" which is 7-byte long,
1198                  * and the line length must be multiple of 5 plus 2.
1199                  */
1200                 if ((llen < 7) || (llen-2) % 5)
1201                         goto corrupt;
1202                 max_byte_length = (llen - 2) / 5 * 4;
1203                 byte_length = *buffer;
1204                 if ('A' <= byte_length && byte_length <= 'Z')
1205                         byte_length = byte_length - 'A' + 1;
1206                 else if ('a' <= byte_length && byte_length <= 'z')
1207                         byte_length = byte_length - 'a' + 27;
1208                 else
1209                         goto corrupt;
1210                 /* if the input length was not multiple of 4, we would
1211                  * have filler at the end but the filler should never
1212                  * exceed 3 bytes
1213                  */
1214                 if (max_byte_length < byte_length ||
1215                     byte_length <= max_byte_length - 4)
1216                         goto corrupt;
1217                 newsize = hunk_size + byte_length;
1218                 data = xrealloc(data, newsize);
1219                 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1220                         goto corrupt;
1221                 hunk_size = newsize;
1222                 buffer += llen;
1223                 size -= llen;
1224         }
1225
1226         frag = xcalloc(1, sizeof(*frag));
1227         frag->patch = inflate_it(data, hunk_size, origlen);
1228         if (!frag->patch)
1229                 goto corrupt;
1230         free(data);
1231         frag->size = origlen;
1232         *buf_p = buffer;
1233         *sz_p = size;
1234         *used_p = used;
1235         frag->binary_patch_method = patch_method;
1236         return frag;
1237
1238  corrupt:
1239         free(data);
1240         *status_p = -1;
1241         error("corrupt binary patch at line %d: %.*s",
1242               linenr-1, llen-1, buffer);
1243         return NULL;
1244 }
1245
1246 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1247 {
1248         /* We have read "GIT binary patch\n"; what follows is a line
1249          * that says the patch method (currently, either "literal" or
1250          * "delta") and the length of data before deflating; a
1251          * sequence of 'length-byte' followed by base-85 encoded data
1252          * follows.
1253          *
1254          * When a binary patch is reversible, there is another binary
1255          * hunk in the same format, starting with patch method (either
1256          * "literal" or "delta") with the length of data, and a sequence
1257          * of length-byte + base-85 encoded data, terminated with another
1258          * empty line.  This data, when applied to the postimage, produces
1259          * the preimage.
1260          */
1261         struct fragment *forward;
1262         struct fragment *reverse;
1263         int status;
1264         int used, used_1;
1265
1266         forward = parse_binary_hunk(&buffer, &size, &status, &used);
1267         if (!forward && !status)
1268                 /* there has to be one hunk (forward hunk) */
1269                 return error("unrecognized binary patch at line %d", linenr-1);
1270         if (status)
1271                 /* otherwise we already gave an error message */
1272                 return status;
1273
1274         reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1275         if (reverse)
1276                 used += used_1;
1277         else if (status) {
1278                 /* not having reverse hunk is not an error, but having
1279                  * a corrupt reverse hunk is.
1280                  */
1281                 free((void*) forward->patch);
1282                 free(forward);
1283                 return status;
1284         }
1285         forward->next = reverse;
1286         patch->fragments = forward;
1287         patch->is_binary = 1;
1288         return used;
1289 }
1290
1291 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1292 {
1293         int hdrsize, patchsize;
1294         int offset = find_header(buffer, size, &hdrsize, patch);
1295
1296         if (offset < 0)
1297                 return offset;
1298
1299         patchsize = parse_single_patch(buffer + offset + hdrsize, size - offset - hdrsize, patch);
1300
1301         if (!patchsize) {
1302                 static const char *binhdr[] = {
1303                         "Binary files ",
1304                         "Files ",
1305                         NULL,
1306                 };
1307                 static const char git_binary[] = "GIT binary patch\n";
1308                 int i;
1309                 int hd = hdrsize + offset;
1310                 unsigned long llen = linelen(buffer + hd, size - hd);
1311
1312                 if (llen == sizeof(git_binary) - 1 &&
1313                     !memcmp(git_binary, buffer + hd, llen)) {
1314                         int used;
1315                         linenr++;
1316                         used = parse_binary(buffer + hd + llen,
1317                                             size - hd - llen, patch);
1318                         if (used)
1319                                 patchsize = used + llen;
1320                         else
1321                                 patchsize = 0;
1322                 }
1323                 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1324                         for (i = 0; binhdr[i]; i++) {
1325                                 int len = strlen(binhdr[i]);
1326                                 if (len < size - hd &&
1327                                     !memcmp(binhdr[i], buffer + hd, len)) {
1328                                         linenr++;
1329                                         patch->is_binary = 1;
1330                                         patchsize = llen;
1331                                         break;
1332                                 }
1333                         }
1334                 }
1335
1336                 /* Empty patch cannot be applied if it is a text patch
1337                  * without metadata change.  A binary patch appears
1338                  * empty to us here.
1339                  */
1340                 if ((apply || check) &&
1341                     (!patch->is_binary && !metadata_changes(patch)))
1342                         die("patch with only garbage at line %d", linenr);
1343         }
1344
1345         return offset + hdrsize + patchsize;
1346 }
1347
1348 #define swap(a,b) myswap((a),(b),sizeof(a))
1349
1350 #define myswap(a, b, size) do {         \
1351         unsigned char mytmp[size];      \
1352         memcpy(mytmp, &a, size);                \
1353         memcpy(&a, &b, size);           \
1354         memcpy(&b, mytmp, size);                \
1355 } while (0)
1356
1357 static void reverse_patches(struct patch *p)
1358 {
1359         for (; p; p = p->next) {
1360                 struct fragment *frag = p->fragments;
1361
1362                 swap(p->new_name, p->old_name);
1363                 swap(p->new_mode, p->old_mode);
1364                 swap(p->is_new, p->is_delete);
1365                 swap(p->lines_added, p->lines_deleted);
1366                 swap(p->old_sha1_prefix, p->new_sha1_prefix);
1367
1368                 for (; frag; frag = frag->next) {
1369                         swap(frag->newpos, frag->oldpos);
1370                         swap(frag->newlines, frag->oldlines);
1371                 }
1372         }
1373 }
1374
1375 static const char pluses[] = "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1376 static const char minuses[]= "----------------------------------------------------------------------";
1377
1378 static void show_stats(struct patch *patch)
1379 {
1380         struct strbuf qname;
1381         char *cp = patch->new_name ? patch->new_name : patch->old_name;
1382         int max, add, del;
1383
1384         strbuf_init(&qname, 0);
1385         quote_c_style(cp, &qname, NULL, 0);
1386
1387         /*
1388          * "scale" the filename
1389          */
1390         max = max_len;
1391         if (max > 50)
1392                 max = 50;
1393
1394         if (qname.len > max) {
1395                 cp = strchr(qname.buf + qname.len + 3 - max, '/');
1396                 if (!cp)
1397                         cp = qname.buf + qname.len + 3 - max;
1398                 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
1399         }
1400
1401         if (patch->is_binary) {
1402                 printf(" %-*s |  Bin\n", max, qname.buf);
1403                 strbuf_release(&qname);
1404                 return;
1405         }
1406
1407         printf(" %-*s |", max, qname.buf);
1408         strbuf_release(&qname);
1409
1410         /*
1411          * scale the add/delete
1412          */
1413         max = max + max_change > 70 ? 70 - max : max_change;
1414         add = patch->lines_added;
1415         del = patch->lines_deleted;
1416
1417         if (max_change > 0) {
1418                 int total = ((add + del) * max + max_change / 2) / max_change;
1419                 add = (add * max + max_change / 2) / max_change;
1420                 del = total - add;
1421         }
1422         printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
1423                 add, pluses, del, minuses);
1424 }
1425
1426 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
1427 {
1428         int fd;
1429
1430         switch (st->st_mode & S_IFMT) {
1431         case S_IFLNK:
1432                 strbuf_grow(buf, st->st_size);
1433                 if (readlink(path, buf->buf, st->st_size) != st->st_size)
1434                         return -1;
1435                 strbuf_setlen(buf, st->st_size);
1436                 return 0;
1437         case S_IFREG:
1438                 fd = open(path, O_RDONLY);
1439                 if (fd < 0)
1440                         return error("unable to open %s", path);
1441                 if (strbuf_read(buf, fd, st->st_size) < 0) {
1442                         close(fd);
1443                         return -1;
1444                 }
1445                 close(fd);
1446                 convert_to_git(path, buf->buf, buf->len, buf);
1447                 return 0;
1448         default:
1449                 return -1;
1450         }
1451 }
1452
1453 static int find_offset(const char *buf, unsigned long size, const char *fragment, unsigned long fragsize, int line, int *lines)
1454 {
1455         int i;
1456         unsigned long start, backwards, forwards;
1457
1458         if (fragsize > size)
1459                 return -1;
1460
1461         start = 0;
1462         if (line > 1) {
1463                 unsigned long offset = 0;
1464                 i = line-1;
1465                 while (offset + fragsize <= size) {
1466                         if (buf[offset++] == '\n') {
1467                                 start = offset;
1468                                 if (!--i)
1469                                         break;
1470                         }
1471                 }
1472         }
1473
1474         /* Exact line number? */
1475         if ((start + fragsize <= size) &&
1476             !memcmp(buf + start, fragment, fragsize))
1477                 return start;
1478
1479         /*
1480          * There's probably some smart way to do this, but I'll leave
1481          * that to the smart and beautiful people. I'm simple and stupid.
1482          */
1483         backwards = start;
1484         forwards = start;
1485         for (i = 0; ; i++) {
1486                 unsigned long try;
1487                 int n;
1488
1489                 /* "backward" */
1490                 if (i & 1) {
1491                         if (!backwards) {
1492                                 if (forwards + fragsize > size)
1493                                         break;
1494                                 continue;
1495                         }
1496                         do {
1497                                 --backwards;
1498                         } while (backwards && buf[backwards-1] != '\n');
1499                         try = backwards;
1500                 } else {
1501                         while (forwards + fragsize <= size) {
1502                                 if (buf[forwards++] == '\n')
1503                                         break;
1504                         }
1505                         try = forwards;
1506                 }
1507
1508                 if (try + fragsize > size)
1509                         continue;
1510                 if (memcmp(buf + try, fragment, fragsize))
1511                         continue;
1512                 n = (i >> 1)+1;
1513                 if (i & 1)
1514                         n = -n;
1515                 *lines = n;
1516                 return try;
1517         }
1518
1519         /*
1520          * We should start searching forward and backward.
1521          */
1522         return -1;
1523 }
1524
1525 static void remove_first_line(const char **rbuf, int *rsize)
1526 {
1527         const char *buf = *rbuf;
1528         int size = *rsize;
1529         unsigned long offset;
1530         offset = 0;
1531         while (offset <= size) {
1532                 if (buf[offset++] == '\n')
1533                         break;
1534         }
1535         *rsize = size - offset;
1536         *rbuf = buf + offset;
1537 }
1538
1539 static void remove_last_line(const char **rbuf, int *rsize)
1540 {
1541         const char *buf = *rbuf;
1542         int size = *rsize;
1543         unsigned long offset;
1544         offset = size - 1;
1545         while (offset > 0) {
1546                 if (buf[--offset] == '\n')
1547                         break;
1548         }
1549         *rsize = offset + 1;
1550 }
1551
1552 static int apply_line(char *output, const char *patch, int plen)
1553 {
1554         /* plen is number of bytes to be copied from patch,
1555          * starting at patch+1 (patch[0] is '+').  Typically
1556          * patch[plen] is '\n', unless this is the incomplete
1557          * last line.
1558          */
1559         int i;
1560         int add_nl_to_tail = 0;
1561         int fixed = 0;
1562         int last_tab_in_indent = -1;
1563         int last_space_in_indent = -1;
1564         int need_fix_leading_space = 0;
1565         char *buf;
1566
1567         if ((new_whitespace != strip_whitespace) || !whitespace_error ||
1568             *patch != '+') {
1569                 memcpy(output, patch + 1, plen);
1570                 return plen;
1571         }
1572
1573         if (1 < plen && isspace(patch[plen-1])) {
1574                 if (patch[plen] == '\n')
1575                         add_nl_to_tail = 1;
1576                 plen--;
1577                 while (0 < plen && isspace(patch[plen]))
1578                         plen--;
1579                 fixed = 1;
1580         }
1581
1582         for (i = 1; i < plen; i++) {
1583                 char ch = patch[i];
1584                 if (ch == '\t') {
1585                         last_tab_in_indent = i;
1586                         if (0 <= last_space_in_indent)
1587                                 need_fix_leading_space = 1;
1588                 }
1589                 else if (ch == ' ')
1590                         last_space_in_indent = i;
1591                 else
1592                         break;
1593         }
1594
1595         buf = output;
1596         if (need_fix_leading_space) {
1597                 int consecutive_spaces = 0;
1598                 /* between patch[1..last_tab_in_indent] strip the
1599                  * funny spaces, updating them to tab as needed.
1600                  */
1601                 for (i = 1; i < last_tab_in_indent; i++, plen--) {
1602                         char ch = patch[i];
1603                         if (ch != ' ') {
1604                                 consecutive_spaces = 0;
1605                                 *output++ = ch;
1606                         } else {
1607                                 consecutive_spaces++;
1608                                 if (consecutive_spaces == 8) {
1609                                         *output++ = '\t';
1610                                         consecutive_spaces = 0;
1611                                 }
1612                         }
1613                 }
1614                 fixed = 1;
1615                 i = last_tab_in_indent;
1616         }
1617         else
1618                 i = 1;
1619
1620         memcpy(output, patch + i, plen);
1621         if (add_nl_to_tail)
1622                 output[plen++] = '\n';
1623         if (fixed)
1624                 applied_after_fixing_ws++;
1625         return output + plen - buf;
1626 }
1627
1628 static int apply_one_fragment(struct strbuf *buf, struct fragment *frag, int inaccurate_eof)
1629 {
1630         int match_beginning, match_end;
1631         const char *patch = frag->patch;
1632         int offset, size = frag->size;
1633         char *old = xmalloc(size);
1634         char *new = xmalloc(size);
1635         const char *oldlines, *newlines;
1636         int oldsize = 0, newsize = 0;
1637         int new_blank_lines_at_end = 0;
1638         unsigned long leading, trailing;
1639         int pos, lines;
1640
1641         while (size > 0) {
1642                 char first;
1643                 int len = linelen(patch, size);
1644                 int plen;
1645                 int added_blank_line = 0;
1646
1647                 if (!len)
1648                         break;
1649
1650                 /*
1651                  * "plen" is how much of the line we should use for
1652                  * the actual patch data. Normally we just remove the
1653                  * first character on the line, but if the line is
1654                  * followed by "\ No newline", then we also remove the
1655                  * last one (which is the newline, of course).
1656                  */
1657                 plen = len-1;
1658                 if (len < size && patch[len] == '\\')
1659                         plen--;
1660                 first = *patch;
1661                 if (apply_in_reverse) {
1662                         if (first == '-')
1663                                 first = '+';
1664                         else if (first == '+')
1665                                 first = '-';
1666                 }
1667
1668                 switch (first) {
1669                 case '\n':
1670                         /* Newer GNU diff, empty context line */
1671                         if (plen < 0)
1672                                 /* ... followed by '\No newline'; nothing */
1673                                 break;
1674                         old[oldsize++] = '\n';
1675                         new[newsize++] = '\n';
1676                         break;
1677                 case ' ':
1678                 case '-':
1679                         memcpy(old + oldsize, patch + 1, plen);
1680                         oldsize += plen;
1681                         if (first == '-')
1682                                 break;
1683                 /* Fall-through for ' ' */
1684                 case '+':
1685                         if (first != '+' || !no_add) {
1686                                 int added = apply_line(new + newsize, patch,
1687                                                        plen);
1688                                 newsize += added;
1689                                 if (first == '+' &&
1690                                     added == 1 && new[newsize-1] == '\n')
1691                                         added_blank_line = 1;
1692                         }
1693                         break;
1694                 case '@': case '\\':
1695                         /* Ignore it, we already handled it */
1696                         break;
1697                 default:
1698                         if (apply_verbosely)
1699                                 error("invalid start of line: '%c'", first);
1700                         return -1;
1701                 }
1702                 if (added_blank_line)
1703                         new_blank_lines_at_end++;
1704                 else
1705                         new_blank_lines_at_end = 0;
1706                 patch += len;
1707                 size -= len;
1708         }
1709
1710         if (inaccurate_eof && oldsize > 0 && old[oldsize - 1] == '\n' &&
1711                         newsize > 0 && new[newsize - 1] == '\n') {
1712                 oldsize--;
1713                 newsize--;
1714         }
1715
1716         oldlines = old;
1717         newlines = new;
1718         leading = frag->leading;
1719         trailing = frag->trailing;
1720
1721         /*
1722          * If we don't have any leading/trailing data in the patch,
1723          * we want it to match at the beginning/end of the file.
1724          *
1725          * But that would break if the patch is generated with
1726          * --unified=0; sane people wouldn't do that to cause us
1727          * trouble, but we try to please not so sane ones as well.
1728          */
1729         if (unidiff_zero) {
1730                 match_beginning = (!leading && !frag->oldpos);
1731                 match_end = 0;
1732         }
1733         else {
1734                 match_beginning = !leading && (frag->oldpos == 1);
1735                 match_end = !trailing;
1736         }
1737
1738         lines = 0;
1739         pos = frag->newpos;
1740         for (;;) {
1741                 offset = find_offset(buf->buf, buf->len,
1742                                      oldlines, oldsize, pos, &lines);
1743                 if (match_end && offset + oldsize != buf->len)
1744                         offset = -1;
1745                 if (match_beginning && offset)
1746                         offset = -1;
1747                 if (offset >= 0) {
1748                         if (new_whitespace == strip_whitespace &&
1749                             (buf->len - oldsize - offset == 0)) /* end of file? */
1750                                 newsize -= new_blank_lines_at_end;
1751
1752                         /* Warn if it was necessary to reduce the number
1753                          * of context lines.
1754                          */
1755                         if ((leading != frag->leading) ||
1756                             (trailing != frag->trailing))
1757                                 fprintf(stderr, "Context reduced to (%ld/%ld)"
1758                                         " to apply fragment at %d\n",
1759                                         leading, trailing, pos + lines);
1760
1761                         strbuf_splice(buf, offset, oldsize, newlines, newsize);
1762                         offset = 0;
1763                         break;
1764                 }
1765
1766                 /* Am I at my context limits? */
1767                 if ((leading <= p_context) && (trailing <= p_context))
1768                         break;
1769                 if (match_beginning || match_end) {
1770                         match_beginning = match_end = 0;
1771                         continue;
1772                 }
1773                 /* Reduce the number of context lines
1774                  * Reduce both leading and trailing if they are equal
1775                  * otherwise just reduce the larger context.
1776                  */
1777                 if (leading >= trailing) {
1778                         remove_first_line(&oldlines, &oldsize);
1779                         remove_first_line(&newlines, &newsize);
1780                         pos--;
1781                         leading--;
1782                 }
1783                 if (trailing > leading) {
1784                         remove_last_line(&oldlines, &oldsize);
1785                         remove_last_line(&newlines, &newsize);
1786                         trailing--;
1787                 }
1788         }
1789
1790         if (offset && apply_verbosely)
1791                 error("while searching for:\n%.*s", oldsize, oldlines);
1792
1793         free(old);
1794         free(new);
1795         return offset;
1796 }
1797
1798 static int apply_binary_fragment(struct strbuf *buf, struct patch *patch)
1799 {
1800         struct fragment *fragment = patch->fragments;
1801         unsigned long len;
1802         void *dst;
1803
1804         /* Binary patch is irreversible without the optional second hunk */
1805         if (apply_in_reverse) {
1806                 if (!fragment->next)
1807                         return error("cannot reverse-apply a binary patch "
1808                                      "without the reverse hunk to '%s'",
1809                                      patch->new_name
1810                                      ? patch->new_name : patch->old_name);
1811                 fragment = fragment->next;
1812         }
1813         switch (fragment->binary_patch_method) {
1814         case BINARY_DELTA_DEFLATED:
1815                 dst = patch_delta(buf->buf, buf->len, fragment->patch,
1816                                   fragment->size, &len);
1817                 if (!dst)
1818                         return -1;
1819                 /* XXX patch_delta NUL-terminates */
1820                 strbuf_attach(buf, dst, len, len + 1);
1821                 return 0;
1822         case BINARY_LITERAL_DEFLATED:
1823                 strbuf_reset(buf);
1824                 strbuf_add(buf, fragment->patch, fragment->size);
1825                 return 0;
1826         }
1827         return -1;
1828 }
1829
1830 static int apply_binary(struct strbuf *buf, struct patch *patch)
1831 {
1832         const char *name = patch->old_name ? patch->old_name : patch->new_name;
1833         unsigned char sha1[20];
1834
1835         /* For safety, we require patch index line to contain
1836          * full 40-byte textual SHA1 for old and new, at least for now.
1837          */
1838         if (strlen(patch->old_sha1_prefix) != 40 ||
1839             strlen(patch->new_sha1_prefix) != 40 ||
1840             get_sha1_hex(patch->old_sha1_prefix, sha1) ||
1841             get_sha1_hex(patch->new_sha1_prefix, sha1))
1842                 return error("cannot apply binary patch to '%s' "
1843                              "without full index line", name);
1844
1845         if (patch->old_name) {
1846                 /* See if the old one matches what the patch
1847                  * applies to.
1848                  */
1849                 hash_sha1_file(buf->buf, buf->len, blob_type, sha1);
1850                 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
1851                         return error("the patch applies to '%s' (%s), "
1852                                      "which does not match the "
1853                                      "current contents.",
1854                                      name, sha1_to_hex(sha1));
1855         }
1856         else {
1857                 /* Otherwise, the old one must be empty. */
1858                 if (buf->len)
1859                         return error("the patch applies to an empty "
1860                                      "'%s' but it is not empty", name);
1861         }
1862
1863         get_sha1_hex(patch->new_sha1_prefix, sha1);
1864         if (is_null_sha1(sha1)) {
1865                 strbuf_release(buf);
1866                 return 0; /* deletion patch */
1867         }
1868
1869         if (has_sha1_file(sha1)) {
1870                 /* We already have the postimage */
1871                 enum object_type type;
1872                 unsigned long size;
1873                 char *result;
1874
1875                 result = read_sha1_file(sha1, &type, &size);
1876                 if (!result)
1877                         return error("the necessary postimage %s for "
1878                                      "'%s' cannot be read",
1879                                      patch->new_sha1_prefix, name);
1880                 /* XXX read_sha1_file NUL-terminates */
1881                 strbuf_attach(buf, result, size, size + 1);
1882         } else {
1883                 /* We have verified buf matches the preimage;
1884                  * apply the patch data to it, which is stored
1885                  * in the patch->fragments->{patch,size}.
1886                  */
1887                 if (apply_binary_fragment(buf, patch))
1888                         return error("binary patch does not apply to '%s'",
1889                                      name);
1890
1891                 /* verify that the result matches */
1892                 hash_sha1_file(buf->buf, buf->len, blob_type, sha1);
1893                 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
1894                         return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)",
1895                                 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
1896         }
1897
1898         return 0;
1899 }
1900
1901 static int apply_fragments(struct strbuf *buf, struct patch *patch)
1902 {
1903         struct fragment *frag = patch->fragments;
1904         const char *name = patch->old_name ? patch->old_name : patch->new_name;
1905
1906         if (patch->is_binary)
1907                 return apply_binary(buf, patch);
1908
1909         while (frag) {
1910                 if (apply_one_fragment(buf, frag, patch->inaccurate_eof)) {
1911                         error("patch failed: %s:%ld", name, frag->oldpos);
1912                         if (!apply_with_reject)
1913                                 return -1;
1914                         frag->rejected = 1;
1915                 }
1916                 frag = frag->next;
1917         }
1918         return 0;
1919 }
1920
1921 static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
1922 {
1923         if (!ce)
1924                 return 0;
1925
1926         if (S_ISGITLINK(ntohl(ce->ce_mode))) {
1927                 strbuf_grow(buf, 100);
1928                 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(ce->sha1));
1929         } else {
1930                 enum object_type type;
1931                 unsigned long sz;
1932                 char *result;
1933
1934                 result = read_sha1_file(ce->sha1, &type, &sz);
1935                 if (!result)
1936                         return -1;
1937                 /* XXX read_sha1_file NUL-terminates */
1938                 strbuf_attach(buf, result, sz, sz + 1);
1939         }
1940         return 0;
1941 }
1942
1943 static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
1944 {
1945         struct strbuf buf;
1946
1947         strbuf_init(&buf, 0);
1948         if (cached) {
1949                 if (read_file_or_gitlink(ce, &buf))
1950                         return error("read of %s failed", patch->old_name);
1951         } else if (patch->old_name) {
1952                 if (S_ISGITLINK(patch->old_mode)) {
1953                         if (ce) {
1954                                 read_file_or_gitlink(ce, &buf);
1955                         } else {
1956                                 /*
1957                                  * There is no way to apply subproject
1958                                  * patch without looking at the index.
1959                                  */
1960                                 patch->fragments = NULL;
1961                         }
1962                 } else {
1963                         if (read_old_data(st, patch->old_name, &buf))
1964                                 return error("read of %s failed", patch->old_name);
1965                 }
1966         }
1967
1968         if (apply_fragments(&buf, patch) < 0)
1969                 return -1; /* note with --reject this succeeds. */
1970         patch->result = strbuf_detach(&buf, &patch->resultsize);
1971
1972         if (0 < patch->is_delete && patch->resultsize)
1973                 return error("removal patch leaves file contents");
1974
1975         return 0;
1976 }
1977
1978 static int check_to_create_blob(const char *new_name, int ok_if_exists)
1979 {
1980         struct stat nst;
1981         if (!lstat(new_name, &nst)) {
1982                 if (S_ISDIR(nst.st_mode) || ok_if_exists)
1983                         return 0;
1984                 /*
1985                  * A leading component of new_name might be a symlink
1986                  * that is going to be removed with this patch, but
1987                  * still pointing at somewhere that has the path.
1988                  * In such a case, path "new_name" does not exist as
1989                  * far as git is concerned.
1990                  */
1991                 if (has_symlink_leading_path(new_name, NULL))
1992                         return 0;
1993
1994                 return error("%s: already exists in working directory", new_name);
1995         }
1996         else if ((errno != ENOENT) && (errno != ENOTDIR))
1997                 return error("%s: %s", new_name, strerror(errno));
1998         return 0;
1999 }
2000
2001 static int verify_index_match(struct cache_entry *ce, struct stat *st)
2002 {
2003         if (S_ISGITLINK(ntohl(ce->ce_mode))) {
2004                 if (!S_ISDIR(st->st_mode))
2005                         return -1;
2006                 return 0;
2007         }
2008         return ce_match_stat(ce, st, 1);
2009 }
2010
2011 static int check_patch(struct patch *patch, struct patch *prev_patch)
2012 {
2013         struct stat st;
2014         const char *old_name = patch->old_name;
2015         const char *new_name = patch->new_name;
2016         const char *name = old_name ? old_name : new_name;
2017         struct cache_entry *ce = NULL;
2018         int ok_if_exists;
2019
2020         patch->rejected = 1; /* we will drop this after we succeed */
2021
2022         /*
2023          * Make sure that we do not have local modifications from the
2024          * index when we are looking at the index.  Also make sure
2025          * we have the preimage file to be patched in the work tree,
2026          * unless --cached, which tells git to apply only in the index.
2027          */
2028         if (old_name) {
2029                 int stat_ret = 0;
2030                 unsigned st_mode = 0;
2031
2032                 if (!cached)
2033                         stat_ret = lstat(old_name, &st);
2034                 if (check_index) {
2035                         int pos = cache_name_pos(old_name, strlen(old_name));
2036                         if (pos < 0)
2037                                 return error("%s: does not exist in index",
2038                                              old_name);
2039                         ce = active_cache[pos];
2040                         if (stat_ret < 0) {
2041                                 struct checkout costate;
2042                                 if (errno != ENOENT)
2043                                         return error("%s: %s", old_name,
2044                                                      strerror(errno));
2045                                 /* checkout */
2046                                 costate.base_dir = "";
2047                                 costate.base_dir_len = 0;
2048                                 costate.force = 0;
2049                                 costate.quiet = 0;
2050                                 costate.not_new = 0;
2051                                 costate.refresh_cache = 1;
2052                                 if (checkout_entry(ce,
2053                                                    &costate,
2054                                                    NULL) ||
2055                                     lstat(old_name, &st))
2056                                         return -1;
2057                         }
2058                         if (!cached && verify_index_match(ce, &st))
2059                                 return error("%s: does not match index",
2060                                              old_name);
2061                         if (cached)
2062                                 st_mode = ntohl(ce->ce_mode);
2063                 } else if (stat_ret < 0)
2064                         return error("%s: %s", old_name, strerror(errno));
2065
2066                 if (!cached)
2067                         st_mode = ntohl(ce_mode_from_stat(ce, st.st_mode));
2068
2069                 if (patch->is_new < 0)
2070                         patch->is_new = 0;
2071                 if (!patch->old_mode)
2072                         patch->old_mode = st_mode;
2073                 if ((st_mode ^ patch->old_mode) & S_IFMT)
2074                         return error("%s: wrong type", old_name);
2075                 if (st_mode != patch->old_mode)
2076                         fprintf(stderr, "warning: %s has type %o, expected %o\n",
2077                                 old_name, st_mode, patch->old_mode);
2078         }
2079
2080         if (new_name && prev_patch && 0 < prev_patch->is_delete &&
2081             !strcmp(prev_patch->old_name, new_name))
2082                 /* A type-change diff is always split into a patch to
2083                  * delete old, immediately followed by a patch to
2084                  * create new (see diff.c::run_diff()); in such a case
2085                  * it is Ok that the entry to be deleted by the
2086                  * previous patch is still in the working tree and in
2087                  * the index.
2088                  */
2089                 ok_if_exists = 1;
2090         else
2091                 ok_if_exists = 0;
2092
2093         if (new_name &&
2094             ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
2095                 if (check_index &&
2096                     cache_name_pos(new_name, strlen(new_name)) >= 0 &&
2097                     !ok_if_exists)
2098                         return error("%s: already exists in index", new_name);
2099                 if (!cached) {
2100                         int err = check_to_create_blob(new_name, ok_if_exists);
2101                         if (err)
2102                                 return err;
2103                 }
2104                 if (!patch->new_mode) {
2105                         if (0 < patch->is_new)
2106                                 patch->new_mode = S_IFREG | 0644;
2107                         else
2108                                 patch->new_mode = patch->old_mode;
2109                 }
2110         }
2111
2112         if (new_name && old_name) {
2113                 int same = !strcmp(old_name, new_name);
2114                 if (!patch->new_mode)
2115                         patch->new_mode = patch->old_mode;
2116                 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
2117                         return error("new mode (%o) of %s does not match old mode (%o)%s%s",
2118                                 patch->new_mode, new_name, patch->old_mode,
2119                                 same ? "" : " of ", same ? "" : old_name);
2120         }
2121
2122         if (apply_data(patch, &st, ce) < 0)
2123                 return error("%s: patch does not apply", name);
2124         patch->rejected = 0;
2125         return 0;
2126 }
2127
2128 static int check_patch_list(struct patch *patch)
2129 {
2130         struct patch *prev_patch = NULL;
2131         int err = 0;
2132
2133         for (prev_patch = NULL; patch ; patch = patch->next) {
2134                 if (apply_verbosely)
2135                         say_patch_name(stderr,
2136                                        "Checking patch ", patch, "...\n");
2137                 err |= check_patch(patch, prev_patch);
2138                 prev_patch = patch;
2139         }
2140         return err;
2141 }
2142
2143 /* This function tries to read the sha1 from the current index */
2144 static int get_current_sha1(const char *path, unsigned char *sha1)
2145 {
2146         int pos;
2147
2148         if (read_cache() < 0)
2149                 return -1;
2150         pos = cache_name_pos(path, strlen(path));
2151         if (pos < 0)
2152                 return -1;
2153         hashcpy(sha1, active_cache[pos]->sha1);
2154         return 0;
2155 }
2156
2157 static void show_index_list(struct patch *list)
2158 {
2159         struct patch *patch;
2160
2161         /* Once we start supporting the reverse patch, it may be
2162          * worth showing the new sha1 prefix, but until then...
2163          */
2164         for (patch = list; patch; patch = patch->next) {
2165                 const unsigned char *sha1_ptr;
2166                 unsigned char sha1[20];
2167                 const char *name;
2168
2169                 name = patch->old_name ? patch->old_name : patch->new_name;
2170                 if (0 < patch->is_new)
2171                         sha1_ptr = null_sha1;
2172                 else if (get_sha1(patch->old_sha1_prefix, sha1))
2173                         /* git diff has no index line for mode/type changes */
2174                         if (!patch->lines_added && !patch->lines_deleted) {
2175                                 if (get_current_sha1(patch->new_name, sha1) ||
2176                                     get_current_sha1(patch->old_name, sha1))
2177                                         die("mode change for %s, which is not "
2178                                                 "in current HEAD", name);
2179                                 sha1_ptr = sha1;
2180                         } else
2181                                 die("sha1 information is lacking or useless "
2182                                         "(%s).", name);
2183                 else
2184                         sha1_ptr = sha1;
2185
2186                 printf("%06o %s ",patch->old_mode, sha1_to_hex(sha1_ptr));
2187                 if (line_termination && quote_c_style(name, NULL, NULL, 0))
2188                         quote_c_style(name, NULL, stdout, 0);
2189                 else
2190                         fputs(name, stdout);
2191                 putchar(line_termination);
2192         }
2193 }
2194
2195 static void stat_patch_list(struct patch *patch)
2196 {
2197         int files, adds, dels;
2198
2199         for (files = adds = dels = 0 ; patch ; patch = patch->next) {
2200                 files++;
2201                 adds += patch->lines_added;
2202                 dels += patch->lines_deleted;
2203                 show_stats(patch);
2204         }
2205
2206         printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
2207 }
2208
2209 static void numstat_patch_list(struct patch *patch)
2210 {
2211         for ( ; patch; patch = patch->next) {
2212                 const char *name;
2213                 name = patch->new_name ? patch->new_name : patch->old_name;
2214                 if (patch->is_binary)
2215                         printf("-\t-\t");
2216                 else
2217                         printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
2218                 write_name_quoted(name, stdout, line_termination);
2219         }
2220 }
2221
2222 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
2223 {
2224         if (mode)
2225                 printf(" %s mode %06o %s\n", newdelete, mode, name);
2226         else
2227                 printf(" %s %s\n", newdelete, name);
2228 }
2229
2230 static void show_mode_change(struct patch *p, int show_name)
2231 {
2232         if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
2233                 if (show_name)
2234                         printf(" mode change %06o => %06o %s\n",
2235                                p->old_mode, p->new_mode, p->new_name);
2236                 else
2237                         printf(" mode change %06o => %06o\n",
2238                                p->old_mode, p->new_mode);
2239         }
2240 }
2241
2242 static void show_rename_copy(struct patch *p)
2243 {
2244         const char *renamecopy = p->is_rename ? "rename" : "copy";
2245         const char *old, *new;
2246
2247         /* Find common prefix */
2248         old = p->old_name;
2249         new = p->new_name;
2250         while (1) {
2251                 const char *slash_old, *slash_new;
2252                 slash_old = strchr(old, '/');
2253                 slash_new = strchr(new, '/');
2254                 if (!slash_old ||
2255                     !slash_new ||
2256                     slash_old - old != slash_new - new ||
2257                     memcmp(old, new, slash_new - new))
2258                         break;
2259                 old = slash_old + 1;
2260                 new = slash_new + 1;
2261         }
2262         /* p->old_name thru old is the common prefix, and old and new
2263          * through the end of names are renames
2264          */
2265         if (old != p->old_name)
2266                 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
2267                        (int)(old - p->old_name), p->old_name,
2268                        old, new, p->score);
2269         else
2270                 printf(" %s %s => %s (%d%%)\n", renamecopy,
2271                        p->old_name, p->new_name, p->score);
2272         show_mode_change(p, 0);
2273 }
2274
2275 static void summary_patch_list(struct patch *patch)
2276 {
2277         struct patch *p;
2278
2279         for (p = patch; p; p = p->next) {
2280                 if (p->is_new)
2281                         show_file_mode_name("create", p->new_mode, p->new_name);
2282                 else if (p->is_delete)
2283                         show_file_mode_name("delete", p->old_mode, p->old_name);
2284                 else {
2285                         if (p->is_rename || p->is_copy)
2286                                 show_rename_copy(p);
2287                         else {
2288                                 if (p->score) {
2289                                         printf(" rewrite %s (%d%%)\n",
2290                                                p->new_name, p->score);
2291                                         show_mode_change(p, 0);
2292                                 }
2293                                 else
2294                                         show_mode_change(p, 1);
2295                         }
2296                 }
2297         }
2298 }
2299
2300 static void patch_stats(struct patch *patch)
2301 {
2302         int lines = patch->lines_added + patch->lines_deleted;
2303
2304         if (lines > max_change)
2305                 max_change = lines;
2306         if (patch->old_name) {
2307                 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
2308                 if (!len)
2309                         len = strlen(patch->old_name);
2310                 if (len > max_len)
2311                         max_len = len;
2312         }
2313         if (patch->new_name) {
2314                 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
2315                 if (!len)
2316                         len = strlen(patch->new_name);
2317                 if (len > max_len)
2318                         max_len = len;
2319         }
2320 }
2321
2322 static void remove_file(struct patch *patch, int rmdir_empty)
2323 {
2324         if (update_index) {
2325                 if (remove_file_from_cache(patch->old_name) < 0)
2326                         die("unable to remove %s from index", patch->old_name);
2327         }
2328         if (!cached) {
2329                 if (S_ISGITLINK(patch->old_mode)) {
2330                         if (rmdir(patch->old_name))
2331                                 warning("unable to remove submodule %s",
2332                                         patch->old_name);
2333                 } else if (!unlink(patch->old_name) && rmdir_empty) {
2334                         char *name = xstrdup(patch->old_name);
2335                         char *end = strrchr(name, '/');
2336                         while (end) {
2337                                 *end = 0;
2338                                 if (rmdir(name))
2339                                         break;
2340                                 end = strrchr(name, '/');
2341                         }
2342                         free(name);
2343                 }
2344         }
2345 }
2346
2347 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
2348 {
2349         struct stat st;
2350         struct cache_entry *ce;
2351         int namelen = strlen(path);
2352         unsigned ce_size = cache_entry_size(namelen);
2353
2354         if (!update_index)
2355                 return;
2356
2357         ce = xcalloc(1, ce_size);
2358         memcpy(ce->name, path, namelen);
2359         ce->ce_mode = create_ce_mode(mode);
2360         ce->ce_flags = htons(namelen);
2361         if (S_ISGITLINK(mode)) {
2362                 const char *s = buf;
2363
2364                 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
2365                         die("corrupt patch for subproject %s", path);
2366         } else {
2367                 if (!cached) {
2368                         if (lstat(path, &st) < 0)
2369                                 die("unable to stat newly created file %s",
2370                                     path);
2371                         fill_stat_cache_info(ce, &st);
2372                 }
2373                 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
2374                         die("unable to create backing store for newly created file %s", path);
2375         }
2376         if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
2377                 die("unable to add cache entry for %s", path);
2378 }
2379
2380 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
2381 {
2382         int fd;
2383         struct strbuf nbuf;
2384
2385         if (S_ISGITLINK(mode)) {
2386                 struct stat st;
2387                 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
2388                         return 0;
2389                 return mkdir(path, 0777);
2390         }
2391
2392         if (has_symlinks && S_ISLNK(mode))
2393                 /* Although buf:size is counted string, it also is NUL
2394                  * terminated.
2395                  */
2396                 return symlink(buf, path);
2397
2398         fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
2399         if (fd < 0)
2400                 return -1;
2401
2402         strbuf_init(&nbuf, 0);
2403         if (convert_to_working_tree(path, buf, size, &nbuf)) {
2404                 size = nbuf.len;
2405                 buf  = nbuf.buf;
2406         }
2407         write_or_die(fd, buf, size);
2408         strbuf_release(&nbuf);
2409
2410         if (close(fd) < 0)
2411                 die("closing file %s: %s", path, strerror(errno));
2412         return 0;
2413 }
2414
2415 /*
2416  * We optimistically assume that the directories exist,
2417  * which is true 99% of the time anyway. If they don't,
2418  * we create them and try again.
2419  */
2420 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
2421 {
2422         if (cached)
2423                 return;
2424         if (!try_create_file(path, mode, buf, size))
2425                 return;
2426
2427         if (errno == ENOENT) {
2428                 if (safe_create_leading_directories(path))
2429                         return;
2430                 if (!try_create_file(path, mode, buf, size))
2431                         return;
2432         }
2433
2434         if (errno == EEXIST || errno == EACCES) {
2435                 /* We may be trying to create a file where a directory
2436                  * used to be.
2437                  */
2438                 struct stat st;
2439                 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
2440                         errno = EEXIST;
2441         }
2442
2443         if (errno == EEXIST) {
2444                 unsigned int nr = getpid();
2445
2446                 for (;;) {
2447                         const char *newpath;
2448                         newpath = mkpath("%s~%u", path, nr);
2449                         if (!try_create_file(newpath, mode, buf, size)) {
2450                                 if (!rename(newpath, path))
2451                                         return;
2452                                 unlink(newpath);
2453                                 break;
2454                         }
2455                         if (errno != EEXIST)
2456                                 break;
2457                         ++nr;
2458                 }
2459         }
2460         die("unable to write file %s mode %o", path, mode);
2461 }
2462
2463 static void create_file(struct patch *patch)
2464 {
2465         char *path = patch->new_name;
2466         unsigned mode = patch->new_mode;
2467         unsigned long size = patch->resultsize;
2468         char *buf = patch->result;
2469
2470         if (!mode)
2471                 mode = S_IFREG | 0644;
2472         create_one_file(path, mode, buf, size);
2473         add_index_file(path, mode, buf, size);
2474 }
2475
2476 /* phase zero is to remove, phase one is to create */
2477 static void write_out_one_result(struct patch *patch, int phase)
2478 {
2479         if (patch->is_delete > 0) {
2480                 if (phase == 0)
2481                         remove_file(patch, 1);
2482                 return;
2483         }
2484         if (patch->is_new > 0 || patch->is_copy) {
2485                 if (phase == 1)
2486                         create_file(patch);
2487                 return;
2488         }
2489         /*
2490          * Rename or modification boils down to the same
2491          * thing: remove the old, write the new
2492          */
2493         if (phase == 0)
2494                 remove_file(patch, patch->is_rename);
2495         if (phase == 1)
2496                 create_file(patch);
2497 }
2498
2499 static int write_out_one_reject(struct patch *patch)
2500 {
2501         FILE *rej;
2502         char namebuf[PATH_MAX];
2503         struct fragment *frag;
2504         int cnt = 0;
2505
2506         for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
2507                 if (!frag->rejected)
2508                         continue;
2509                 cnt++;
2510         }
2511
2512         if (!cnt) {
2513                 if (apply_verbosely)
2514                         say_patch_name(stderr,
2515                                        "Applied patch ", patch, " cleanly.\n");
2516                 return 0;
2517         }
2518
2519         /* This should not happen, because a removal patch that leaves
2520          * contents are marked "rejected" at the patch level.
2521          */
2522         if (!patch->new_name)
2523                 die("internal error");
2524
2525         /* Say this even without --verbose */
2526         say_patch_name(stderr, "Applying patch ", patch, " with");
2527         fprintf(stderr, " %d rejects...\n", cnt);
2528
2529         cnt = strlen(patch->new_name);
2530         if (ARRAY_SIZE(namebuf) <= cnt + 5) {
2531                 cnt = ARRAY_SIZE(namebuf) - 5;
2532                 fprintf(stderr,
2533                         "warning: truncating .rej filename to %.*s.rej",
2534                         cnt - 1, patch->new_name);
2535         }
2536         memcpy(namebuf, patch->new_name, cnt);
2537         memcpy(namebuf + cnt, ".rej", 5);
2538
2539         rej = fopen(namebuf, "w");
2540         if (!rej)
2541                 return error("cannot open %s: %s", namebuf, strerror(errno));
2542
2543         /* Normal git tools never deal with .rej, so do not pretend
2544          * this is a git patch by saying --git nor give extended
2545          * headers.  While at it, maybe please "kompare" that wants
2546          * the trailing TAB and some garbage at the end of line ;-).
2547          */
2548         fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
2549                 patch->new_name, patch->new_name);
2550         for (cnt = 1, frag = patch->fragments;
2551              frag;
2552              cnt++, frag = frag->next) {
2553                 if (!frag->rejected) {
2554                         fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);
2555                         continue;
2556                 }
2557                 fprintf(stderr, "Rejected hunk #%d.\n", cnt);
2558                 fprintf(rej, "%.*s", frag->size, frag->patch);
2559                 if (frag->patch[frag->size-1] != '\n')
2560                         fputc('\n', rej);
2561         }
2562         fclose(rej);
2563         return -1;
2564 }
2565
2566 static int write_out_results(struct patch *list, int skipped_patch)
2567 {
2568         int phase;
2569         int errs = 0;
2570         struct patch *l;
2571
2572         if (!list && !skipped_patch)
2573                 return error("No changes");
2574
2575         for (phase = 0; phase < 2; phase++) {
2576                 l = list;
2577                 while (l) {
2578                         if (l->rejected)
2579                                 errs = 1;
2580                         else {
2581                                 write_out_one_result(l, phase);
2582                                 if (phase == 1 && write_out_one_reject(l))
2583                                         errs = 1;
2584                         }
2585                         l = l->next;
2586                 }
2587         }
2588         return errs;
2589 }
2590
2591 static struct lock_file lock_file;
2592
2593 static struct excludes {
2594         struct excludes *next;
2595         const char *path;
2596 } *excludes;
2597
2598 static int use_patch(struct patch *p)
2599 {
2600         const char *pathname = p->new_name ? p->new_name : p->old_name;
2601         struct excludes *x = excludes;
2602         while (x) {
2603                 if (fnmatch(x->path, pathname, 0) == 0)
2604                         return 0;
2605                 x = x->next;
2606         }
2607         if (0 < prefix_length) {
2608                 int pathlen = strlen(pathname);
2609                 if (pathlen <= prefix_length ||
2610                     memcmp(prefix, pathname, prefix_length))
2611                         return 0;
2612         }
2613         return 1;
2614 }
2615
2616 static void prefix_one(char **name)
2617 {
2618         char *old_name = *name;
2619         if (!old_name)
2620                 return;
2621         *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
2622         free(old_name);
2623 }
2624
2625 static void prefix_patches(struct patch *p)
2626 {
2627         if (!prefix || p->is_toplevel_relative)
2628                 return;
2629         for ( ; p; p = p->next) {
2630                 if (p->new_name == p->old_name) {
2631                         char *prefixed = p->new_name;
2632                         prefix_one(&prefixed);
2633                         p->new_name = p->old_name = prefixed;
2634                 }
2635                 else {
2636                         prefix_one(&p->new_name);
2637                         prefix_one(&p->old_name);
2638                 }
2639         }
2640 }
2641
2642 static int apply_patch(int fd, const char *filename, int inaccurate_eof)
2643 {
2644         unsigned long offset, size;
2645         char *buffer = read_patch_file(fd, &size);
2646         struct patch *list = NULL, **listp = &list;
2647         int skipped_patch = 0;
2648
2649         patch_input_file = filename;
2650         if (!buffer)
2651                 return -1;
2652         offset = 0;
2653         while (size > 0) {
2654                 struct patch *patch;
2655                 int nr;
2656
2657                 patch = xcalloc(1, sizeof(*patch));
2658                 patch->inaccurate_eof = inaccurate_eof;
2659                 nr = parse_chunk(buffer + offset, size, patch);
2660                 if (nr < 0)
2661                         break;
2662                 if (apply_in_reverse)
2663                         reverse_patches(patch);
2664                 if (prefix)
2665                         prefix_patches(patch);
2666                 if (use_patch(patch)) {
2667                         patch_stats(patch);
2668                         *listp = patch;
2669                         listp = &patch->next;
2670                 }
2671                 else {
2672                         /* perhaps free it a bit better? */
2673                         free(patch);
2674                         skipped_patch++;
2675                 }
2676                 offset += nr;
2677                 size -= nr;
2678         }
2679
2680         if (whitespace_error && (new_whitespace == error_on_whitespace))
2681                 apply = 0;
2682
2683         update_index = check_index && apply;
2684         if (update_index && newfd < 0)
2685                 newfd = hold_locked_index(&lock_file, 1);
2686
2687         if (check_index) {
2688                 if (read_cache() < 0)
2689                         die("unable to read index file");
2690         }
2691
2692         if ((check || apply) &&
2693             check_patch_list(list) < 0 &&
2694             !apply_with_reject)
2695                 exit(1);
2696
2697         if (apply && write_out_results(list, skipped_patch))
2698                 exit(1);
2699
2700         if (show_index_info)
2701                 show_index_list(list);
2702
2703         if (diffstat)
2704                 stat_patch_list(list);
2705
2706         if (numstat)
2707                 numstat_patch_list(list);
2708
2709         if (summary)
2710                 summary_patch_list(list);
2711
2712         free(buffer);
2713         return 0;
2714 }
2715
2716 static int git_apply_config(const char *var, const char *value)
2717 {
2718         if (!strcmp(var, "apply.whitespace")) {
2719                 apply_default_whitespace = xstrdup(value);
2720                 return 0;
2721         }
2722         return git_default_config(var, value);
2723 }
2724
2725
2726 int cmd_apply(int argc, const char **argv, const char *unused_prefix)
2727 {
2728         int i;
2729         int read_stdin = 1;
2730         int inaccurate_eof = 0;
2731         int errs = 0;
2732         int is_not_gitdir = 0;
2733
2734         const char *whitespace_option = NULL;
2735
2736         prefix = setup_git_directory_gently(&is_not_gitdir);
2737         prefix_length = prefix ? strlen(prefix) : 0;
2738         git_config(git_apply_config);
2739         if (apply_default_whitespace)
2740                 parse_whitespace_option(apply_default_whitespace);
2741
2742         for (i = 1; i < argc; i++) {
2743                 const char *arg = argv[i];
2744                 char *end;
2745                 int fd;
2746
2747                 if (!strcmp(arg, "-")) {
2748                         errs |= apply_patch(0, "<stdin>", inaccurate_eof);
2749                         read_stdin = 0;
2750                         continue;
2751                 }
2752                 if (!prefixcmp(arg, "--exclude=")) {
2753                         struct excludes *x = xmalloc(sizeof(*x));
2754                         x->path = arg + 10;
2755                         x->next = excludes;
2756                         excludes = x;
2757                         continue;
2758                 }
2759                 if (!prefixcmp(arg, "-p")) {
2760                         p_value = atoi(arg + 2);
2761                         p_value_known = 1;
2762                         continue;
2763                 }
2764                 if (!strcmp(arg, "--no-add")) {
2765                         no_add = 1;
2766                         continue;
2767                 }
2768                 if (!strcmp(arg, "--stat")) {
2769                         apply = 0;
2770                         diffstat = 1;
2771                         continue;
2772                 }
2773                 if (!strcmp(arg, "--allow-binary-replacement") ||
2774                     !strcmp(arg, "--binary")) {
2775                         continue; /* now no-op */
2776                 }
2777                 if (!strcmp(arg, "--numstat")) {
2778                         apply = 0;
2779                         numstat = 1;
2780                         continue;
2781                 }
2782                 if (!strcmp(arg, "--summary")) {
2783                         apply = 0;
2784                         summary = 1;
2785                         continue;
2786                 }
2787                 if (!strcmp(arg, "--check")) {
2788                         apply = 0;
2789                         check = 1;
2790                         continue;
2791                 }
2792                 if (!strcmp(arg, "--index")) {
2793                         if (is_not_gitdir)
2794                                 die("--index outside a repository");
2795                         check_index = 1;
2796                         continue;
2797                 }
2798                 if (!strcmp(arg, "--cached")) {
2799                         if (is_not_gitdir)
2800                                 die("--cached outside a repository");
2801                         check_index = 1;
2802                         cached = 1;
2803                         continue;
2804                 }
2805                 if (!strcmp(arg, "--apply")) {
2806                         apply = 1;
2807                         continue;
2808                 }
2809                 if (!strcmp(arg, "--index-info")) {
2810                         apply = 0;
2811                         show_index_info = 1;
2812                         continue;
2813                 }
2814                 if (!strcmp(arg, "-z")) {
2815                         line_termination = 0;
2816                         continue;
2817                 }
2818                 if (!prefixcmp(arg, "-C")) {
2819                         p_context = strtoul(arg + 2, &end, 0);
2820                         if (*end != '\0')
2821                                 die("unrecognized context count '%s'", arg + 2);
2822                         continue;
2823                 }
2824                 if (!prefixcmp(arg, "--whitespace=")) {
2825                         whitespace_option = arg + 13;
2826                         parse_whitespace_option(arg + 13);
2827                         continue;
2828                 }
2829                 if (!strcmp(arg, "-R") || !strcmp(arg, "--reverse")) {
2830                         apply_in_reverse = 1;
2831                         continue;
2832                 }
2833                 if (!strcmp(arg, "--unidiff-zero")) {
2834                         unidiff_zero = 1;
2835                         continue;
2836                 }
2837                 if (!strcmp(arg, "--reject")) {
2838                         apply = apply_with_reject = apply_verbosely = 1;
2839                         continue;
2840                 }
2841                 if (!strcmp(arg, "-v") || !strcmp(arg, "--verbose")) {
2842                         apply_verbosely = 1;
2843                         continue;
2844                 }
2845                 if (!strcmp(arg, "--inaccurate-eof")) {
2846                         inaccurate_eof = 1;
2847                         continue;
2848                 }
2849                 if (0 < prefix_length)
2850                         arg = prefix_filename(prefix, prefix_length, arg);
2851
2852                 fd = open(arg, O_RDONLY);
2853                 if (fd < 0)
2854                         usage(apply_usage);
2855                 read_stdin = 0;
2856                 set_default_whitespace_mode(whitespace_option);
2857                 errs |= apply_patch(fd, arg, inaccurate_eof);
2858                 close(fd);
2859         }
2860         set_default_whitespace_mode(whitespace_option);
2861         if (read_stdin)
2862                 errs |= apply_patch(0, "<stdin>", inaccurate_eof);
2863         if (whitespace_error) {
2864                 if (squelch_whitespace_errors &&
2865                     squelch_whitespace_errors < whitespace_error) {
2866                         int squelched =
2867                                 whitespace_error - squelch_whitespace_errors;
2868                         fprintf(stderr, "warning: squelched %d "
2869                                 "whitespace error%s\n",
2870                                 squelched,
2871                                 squelched == 1 ? "" : "s");
2872                 }
2873                 if (new_whitespace == error_on_whitespace)
2874                         die("%d line%s add%s whitespace errors.",
2875                             whitespace_error,
2876                             whitespace_error == 1 ? "" : "s",
2877                             whitespace_error == 1 ? "s" : "");
2878                 if (applied_after_fixing_ws)
2879                         fprintf(stderr, "warning: %d line%s applied after"
2880                                 " fixing whitespace errors.\n",
2881                                 applied_after_fixing_ws,
2882                                 applied_after_fixing_ws == 1 ? "" : "s");
2883                 else if (whitespace_error)
2884                         fprintf(stderr, "warning: %d line%s add%s whitespace errors.\n",
2885                                 whitespace_error,
2886                                 whitespace_error == 1 ? "" : "s",
2887                                 whitespace_error == 1 ? "s" : "");
2888         }
2889
2890         if (update_index) {
2891                 if (write_cache(newfd, active_cache, active_nr) ||
2892                     close(newfd) || commit_locked_index(&lock_file))
2893                         die("Unable to write new index file");
2894         }
2895
2896         return !!errs;
2897 }