Code

fc6c7083f76043168bf8024a9fa50b7f767b99bc
[git.git] / builtin-apply.c
1 /*
2  * apply.c
3  *
4  * Copyright (C) Linus Torvalds, 2005
5  *
6  * This applies patches on top of some (arbitrary) version of the SCM.
7  *
8  */
9 #include "cache.h"
10 #include "cache-tree.h"
11 #include "quote.h"
12 #include "blob.h"
13 #include "delta.h"
14 #include "builtin.h"
15 #include "string-list.h"
16 #include "dir.h"
17 #include "parse-options.h"
19 /*
20  *  --check turns on checking that the working tree matches the
21  *    files that are being modified, but doesn't apply the patch
22  *  --stat does just a diffstat, and doesn't actually apply
23  *  --numstat does numeric diffstat, and doesn't actually apply
24  *  --index-info shows the old and new index info for paths if available.
25  *  --index updates the cache as well.
26  *  --cached updates only the cache without ever touching the working tree.
27  */
28 static const char *prefix;
29 static int prefix_length = -1;
30 static int newfd = -1;
32 static int unidiff_zero;
33 static int p_value = 1;
34 static int p_value_known;
35 static int check_index;
36 static int update_index;
37 static int cached;
38 static int diffstat;
39 static int numstat;
40 static int summary;
41 static int check;
42 static int apply = 1;
43 static int apply_in_reverse;
44 static int apply_with_reject;
45 static int apply_verbosely;
46 static int no_add;
47 static const char *fake_ancestor;
48 static int line_termination = '\n';
49 static unsigned int p_context = UINT_MAX;
50 static const char * const apply_usage[] = {
51         "git apply [options] [<patch>...]",
52         NULL
53 };
55 static enum ws_error_action {
56         nowarn_ws_error,
57         warn_on_ws_error,
58         die_on_ws_error,
59         correct_ws_error,
60 } ws_error_action = warn_on_ws_error;
61 static int whitespace_error;
62 static int squelch_whitespace_errors = 5;
63 static int applied_after_fixing_ws;
65 static enum ws_ignore {
66         ignore_ws_none,
67         ignore_ws_change,
68 } ws_ignore_action = ignore_ws_none;
71 static const char *patch_input_file;
72 static const char *root;
73 static int root_len;
74 static int read_stdin = 1;
75 static int options;
77 static void parse_whitespace_option(const char *option)
78 {
79         if (!option) {
80                 ws_error_action = warn_on_ws_error;
81                 return;
82         }
83         if (!strcmp(option, "warn")) {
84                 ws_error_action = warn_on_ws_error;
85                 return;
86         }
87         if (!strcmp(option, "nowarn")) {
88                 ws_error_action = nowarn_ws_error;
89                 return;
90         }
91         if (!strcmp(option, "error")) {
92                 ws_error_action = die_on_ws_error;
93                 return;
94         }
95         if (!strcmp(option, "error-all")) {
96                 ws_error_action = die_on_ws_error;
97                 squelch_whitespace_errors = 0;
98                 return;
99         }
100         if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
101                 ws_error_action = correct_ws_error;
102                 return;
103         }
104         die("unrecognized whitespace option '%s'", option);
107 static void parse_ignorewhitespace_option(const char *option)
109         if (!option || !strcmp(option, "no") ||
110             !strcmp(option, "false") || !strcmp(option, "never") ||
111             !strcmp(option, "none")) {
112                 ws_ignore_action = ignore_ws_none;
113                 return;
114         }
115         if (!strcmp(option, "change")) {
116                 ws_ignore_action = ignore_ws_change;
117                 return;
118         }
119         die("unrecognized whitespace ignore option '%s'", option);
122 static void set_default_whitespace_mode(const char *whitespace_option)
124         if (!whitespace_option && !apply_default_whitespace)
125                 ws_error_action = (apply ? warn_on_ws_error : nowarn_ws_error);
128 /*
129  * For "diff-stat" like behaviour, we keep track of the biggest change
130  * we've seen, and the longest filename. That allows us to do simple
131  * scaling.
132  */
133 static int max_change, max_len;
135 /*
136  * Various "current state", notably line numbers and what
137  * file (and how) we're patching right now.. The "is_xxxx"
138  * things are flags, where -1 means "don't know yet".
139  */
140 static int linenr = 1;
142 /*
143  * This represents one "hunk" from a patch, starting with
144  * "@@ -oldpos,oldlines +newpos,newlines @@" marker.  The
145  * patch text is pointed at by patch, and its byte length
146  * is stored in size.  leading and trailing are the number
147  * of context lines.
148  */
149 struct fragment {
150         unsigned long leading, trailing;
151         unsigned long oldpos, oldlines;
152         unsigned long newpos, newlines;
153         const char *patch;
154         int size;
155         int rejected;
156         int linenr;
157         struct fragment *next;
158 };
160 /*
161  * When dealing with a binary patch, we reuse "leading" field
162  * to store the type of the binary hunk, either deflated "delta"
163  * or deflated "literal".
164  */
165 #define binary_patch_method leading
166 #define BINARY_DELTA_DEFLATED   1
167 #define BINARY_LITERAL_DEFLATED 2
169 /*
170  * This represents a "patch" to a file, both metainfo changes
171  * such as creation/deletion, filemode and content changes represented
172  * as a series of fragments.
173  */
174 struct patch {
175         char *new_name, *old_name, *def_name;
176         unsigned int old_mode, new_mode;
177         int is_new, is_delete;  /* -1 = unknown, 0 = false, 1 = true */
178         int rejected;
179         unsigned ws_rule;
180         unsigned long deflate_origlen;
181         int lines_added, lines_deleted;
182         int score;
183         unsigned int is_toplevel_relative:1;
184         unsigned int inaccurate_eof:1;
185         unsigned int is_binary:1;
186         unsigned int is_copy:1;
187         unsigned int is_rename:1;
188         unsigned int recount:1;
189         struct fragment *fragments;
190         char *result;
191         size_t resultsize;
192         char old_sha1_prefix[41];
193         char new_sha1_prefix[41];
194         struct patch *next;
195 };
197 /*
198  * A line in a file, len-bytes long (includes the terminating LF,
199  * except for an incomplete line at the end if the file ends with
200  * one), and its contents hashes to 'hash'.
201  */
202 struct line {
203         size_t len;
204         unsigned hash : 24;
205         unsigned flag : 8;
206 #define LINE_COMMON     1
207 };
209 /*
210  * This represents a "file", which is an array of "lines".
211  */
212 struct image {
213         char *buf;
214         size_t len;
215         size_t nr;
216         size_t alloc;
217         struct line *line_allocated;
218         struct line *line;
219 };
221 /*
222  * Records filenames that have been touched, in order to handle
223  * the case where more than one patches touch the same file.
224  */
226 static struct string_list fn_table;
228 static uint32_t hash_line(const char *cp, size_t len)
230         size_t i;
231         uint32_t h;
232         for (i = 0, h = 0; i < len; i++) {
233                 if (!isspace(cp[i])) {
234                         h = h * 3 + (cp[i] & 0xff);
235                 }
236         }
237         return h;
240 /*
241  * Compare lines s1 of length n1 and s2 of length n2, ignoring
242  * whitespace difference. Returns 1 if they match, 0 otherwise
243  */
244 static int fuzzy_matchlines(const char *s1, size_t n1,
245                             const char *s2, size_t n2)
247         const char *last1 = s1 + n1 - 1;
248         const char *last2 = s2 + n2 - 1;
249         int result = 0;
251         if (n1 < 0 || n2 < 0)
252                 return 0;
254         /* ignore line endings */
255         while ((*last1 == '\r') || (*last1 == '\n'))
256                 last1--;
257         while ((*last2 == '\r') || (*last2 == '\n'))
258                 last2--;
260         /* skip leading whitespace */
261         while (isspace(*s1) && (s1 <= last1))
262                 s1++;
263         while (isspace(*s2) && (s2 <= last2))
264                 s2++;
265         /* early return if both lines are empty */
266         if ((s1 > last1) && (s2 > last2))
267                 return 1;
268         while (!result) {
269                 result = *s1++ - *s2++;
270                 /*
271                  * Skip whitespace inside. We check for whitespace on
272                  * both buffers because we don't want "a b" to match
273                  * "ab"
274                  */
275                 if (isspace(*s1) && isspace(*s2)) {
276                         while (isspace(*s1) && s1 <= last1)
277                                 s1++;
278                         while (isspace(*s2) && s2 <= last2)
279                                 s2++;
280                 }
281                 /*
282                  * If we reached the end on one side only,
283                  * lines don't match
284                  */
285                 if (
286                     ((s2 > last2) && (s1 <= last1)) ||
287                     ((s1 > last1) && (s2 <= last2)))
288                         return 0;
289                 if ((s1 > last1) && (s2 > last2))
290                         break;
291         }
293         return !result;
296 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
298         ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
299         img->line_allocated[img->nr].len = len;
300         img->line_allocated[img->nr].hash = hash_line(bol, len);
301         img->line_allocated[img->nr].flag = flag;
302         img->nr++;
305 static void prepare_image(struct image *image, char *buf, size_t len,
306                           int prepare_linetable)
308         const char *cp, *ep;
310         memset(image, 0, sizeof(*image));
311         image->buf = buf;
312         image->len = len;
314         if (!prepare_linetable)
315                 return;
317         ep = image->buf + image->len;
318         cp = image->buf;
319         while (cp < ep) {
320                 const char *next;
321                 for (next = cp; next < ep && *next != '\n'; next++)
322                         ;
323                 if (next < ep)
324                         next++;
325                 add_line_info(image, cp, next - cp, 0);
326                 cp = next;
327         }
328         image->line = image->line_allocated;
331 static void clear_image(struct image *image)
333         free(image->buf);
334         image->buf = NULL;
335         image->len = 0;
338 static void say_patch_name(FILE *output, const char *pre,
339                            struct patch *patch, const char *post)
341         fputs(pre, output);
342         if (patch->old_name && patch->new_name &&
343             strcmp(patch->old_name, patch->new_name)) {
344                 quote_c_style(patch->old_name, NULL, output, 0);
345                 fputs(" => ", output);
346                 quote_c_style(patch->new_name, NULL, output, 0);
347         } else {
348                 const char *n = patch->new_name;
349                 if (!n)
350                         n = patch->old_name;
351                 quote_c_style(n, NULL, output, 0);
352         }
353         fputs(post, output);
356 #define CHUNKSIZE (8192)
357 #define SLOP (16)
359 static void read_patch_file(struct strbuf *sb, int fd)
361         if (strbuf_read(sb, fd, 0) < 0)
362                 die_errno("git apply: failed to read");
364         /*
365          * Make sure that we have some slop in the buffer
366          * so that we can do speculative "memcmp" etc, and
367          * see to it that it is NUL-filled.
368          */
369         strbuf_grow(sb, SLOP);
370         memset(sb->buf + sb->len, 0, SLOP);
373 static unsigned long linelen(const char *buffer, unsigned long size)
375         unsigned long len = 0;
376         while (size--) {
377                 len++;
378                 if (*buffer++ == '\n')
379                         break;
380         }
381         return len;
384 static int is_dev_null(const char *str)
386         return !memcmp("/dev/null", str, 9) && isspace(str[9]);
389 #define TERM_SPACE      1
390 #define TERM_TAB        2
392 static int name_terminate(const char *name, int namelen, int c, int terminate)
394         if (c == ' ' && !(terminate & TERM_SPACE))
395                 return 0;
396         if (c == '\t' && !(terminate & TERM_TAB))
397                 return 0;
399         return 1;
402 /* remove double slashes to make --index work with such filenames */
403 static char *squash_slash(char *name)
405         int i = 0, j = 0;
407         if (!name)
408                 return NULL;
410         while (name[i]) {
411                 if ((name[j++] = name[i++]) == '/')
412                         while (name[i] == '/')
413                                 i++;
414         }
415         name[j] = '\0';
416         return name;
419 static char *find_name(const char *line, char *def, int p_value, int terminate)
421         int len;
422         const char *start = NULL;
424         if (p_value == 0)
425                 start = line;
427         if (*line == '"') {
428                 struct strbuf name = STRBUF_INIT;
430                 /*
431                  * Proposed "new-style" GNU patch/diff format; see
432                  * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
433                  */
434                 if (!unquote_c_style(&name, line, NULL)) {
435                         char *cp;
437                         for (cp = name.buf; p_value; p_value--) {
438                                 cp = strchr(cp, '/');
439                                 if (!cp)
440                                         break;
441                                 cp++;
442                         }
443                         if (cp) {
444                                 /* name can later be freed, so we need
445                                  * to memmove, not just return cp
446                                  */
447                                 strbuf_remove(&name, 0, cp - name.buf);
448                                 free(def);
449                                 if (root)
450                                         strbuf_insert(&name, 0, root, root_len);
451                                 return squash_slash(strbuf_detach(&name, NULL));
452                         }
453                 }
454                 strbuf_release(&name);
455         }
457         for (;;) {
458                 char c = *line;
460                 if (isspace(c)) {
461                         if (c == '\n')
462                                 break;
463                         if (name_terminate(start, line-start, c, terminate))
464                                 break;
465                 }
466                 line++;
467                 if (c == '/' && !--p_value)
468                         start = line;
469         }
470         if (!start)
471                 return squash_slash(def);
472         len = line - start;
473         if (!len)
474                 return squash_slash(def);
476         /*
477          * Generally we prefer the shorter name, especially
478          * if the other one is just a variation of that with
479          * something else tacked on to the end (ie "file.orig"
480          * or "file~").
481          */
482         if (def) {
483                 int deflen = strlen(def);
484                 if (deflen < len && !strncmp(start, def, deflen))
485                         return squash_slash(def);
486                 free(def);
487         }
489         if (root) {
490                 char *ret = xmalloc(root_len + len + 1);
491                 strcpy(ret, root);
492                 memcpy(ret + root_len, start, len);
493                 ret[root_len + len] = '\0';
494                 return squash_slash(ret);
495         }
497         return squash_slash(xmemdupz(start, len));
500 static int count_slashes(const char *cp)
502         int cnt = 0;
503         char ch;
505         while ((ch = *cp++))
506                 if (ch == '/')
507                         cnt++;
508         return cnt;
511 /*
512  * Given the string after "--- " or "+++ ", guess the appropriate
513  * p_value for the given patch.
514  */
515 static int guess_p_value(const char *nameline)
517         char *name, *cp;
518         int val = -1;
520         if (is_dev_null(nameline))
521                 return -1;
522         name = find_name(nameline, NULL, 0, TERM_SPACE | TERM_TAB);
523         if (!name)
524                 return -1;
525         cp = strchr(name, '/');
526         if (!cp)
527                 val = 0;
528         else if (prefix) {
529                 /*
530                  * Does it begin with "a/$our-prefix" and such?  Then this is
531                  * very likely to apply to our directory.
532                  */
533                 if (!strncmp(name, prefix, prefix_length))
534                         val = count_slashes(prefix);
535                 else {
536                         cp++;
537                         if (!strncmp(cp, prefix, prefix_length))
538                                 val = count_slashes(prefix) + 1;
539                 }
540         }
541         free(name);
542         return val;
545 /*
546  * Does the ---/+++ line has the POSIX timestamp after the last HT?
547  * GNU diff puts epoch there to signal a creation/deletion event.  Is
548  * this such a timestamp?
549  */
550 static int has_epoch_timestamp(const char *nameline)
552         /*
553          * We are only interested in epoch timestamp; any non-zero
554          * fraction cannot be one, hence "(\.0+)?" in the regexp below.
555          * For the same reason, the date must be either 1969-12-31 or
556          * 1970-01-01, and the seconds part must be "00".
557          */
558         const char stamp_regexp[] =
559                 "^(1969-12-31|1970-01-01)"
560                 " "
561                 "[0-2][0-9]:[0-5][0-9]:00(\\.0+)?"
562                 " "
563                 "([-+][0-2][0-9][0-5][0-9])\n";
564         const char *timestamp = NULL, *cp;
565         static regex_t *stamp;
566         regmatch_t m[10];
567         int zoneoffset;
568         int hourminute;
569         int status;
571         for (cp = nameline; *cp != '\n'; cp++) {
572                 if (*cp == '\t')
573                         timestamp = cp + 1;
574         }
575         if (!timestamp)
576                 return 0;
577         if (!stamp) {
578                 stamp = xmalloc(sizeof(*stamp));
579                 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
580                         warning("Cannot prepare timestamp regexp %s",
581                                 stamp_regexp);
582                         return 0;
583                 }
584         }
586         status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
587         if (status) {
588                 if (status != REG_NOMATCH)
589                         warning("regexec returned %d for input: %s",
590                                 status, timestamp);
591                 return 0;
592         }
594         zoneoffset = strtol(timestamp + m[3].rm_so + 1, NULL, 10);
595         zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
596         if (timestamp[m[3].rm_so] == '-')
597                 zoneoffset = -zoneoffset;
599         /*
600          * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
601          * (west of GMT) or 1970-01-01 (east of GMT)
602          */
603         if ((zoneoffset < 0 && memcmp(timestamp, "1969-12-31", 10)) ||
604             (0 <= zoneoffset && memcmp(timestamp, "1970-01-01", 10)))
605                 return 0;
607         hourminute = (strtol(timestamp + 11, NULL, 10) * 60 +
608                       strtol(timestamp + 14, NULL, 10) -
609                       zoneoffset);
611         return ((zoneoffset < 0 && hourminute == 1440) ||
612                 (0 <= zoneoffset && !hourminute));
615 /*
616  * Get the name etc info from the ---/+++ lines of a traditional patch header
617  *
618  * FIXME! The end-of-filename heuristics are kind of screwy. For existing
619  * files, we can happily check the index for a match, but for creating a
620  * new file we should try to match whatever "patch" does. I have no idea.
621  */
622 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
624         char *name;
626         first += 4;     /* skip "--- " */
627         second += 4;    /* skip "+++ " */
628         if (!p_value_known) {
629                 int p, q;
630                 p = guess_p_value(first);
631                 q = guess_p_value(second);
632                 if (p < 0) p = q;
633                 if (0 <= p && p == q) {
634                         p_value = p;
635                         p_value_known = 1;
636                 }
637         }
638         if (is_dev_null(first)) {
639                 patch->is_new = 1;
640                 patch->is_delete = 0;
641                 name = find_name(second, NULL, p_value, TERM_SPACE | TERM_TAB);
642                 patch->new_name = name;
643         } else if (is_dev_null(second)) {
644                 patch->is_new = 0;
645                 patch->is_delete = 1;
646                 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
647                 patch->old_name = name;
648         } else {
649                 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
650                 name = find_name(second, name, p_value, TERM_SPACE | TERM_TAB);
651                 if (has_epoch_timestamp(first)) {
652                         patch->is_new = 1;
653                         patch->is_delete = 0;
654                         patch->new_name = name;
655                 } else if (has_epoch_timestamp(second)) {
656                         patch->is_new = 0;
657                         patch->is_delete = 1;
658                         patch->old_name = name;
659                 } else {
660                         patch->old_name = patch->new_name = name;
661                 }
662         }
663         if (!name)
664                 die("unable to find filename in patch at line %d", linenr);
667 static int gitdiff_hdrend(const char *line, struct patch *patch)
669         return -1;
672 /*
673  * We're anal about diff header consistency, to make
674  * sure that we don't end up having strange ambiguous
675  * patches floating around.
676  *
677  * As a result, gitdiff_{old|new}name() will check
678  * their names against any previous information, just
679  * to make sure..
680  */
681 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
683         if (!orig_name && !isnull)
684                 return find_name(line, NULL, p_value, TERM_TAB);
686         if (orig_name) {
687                 int len;
688                 const char *name;
689                 char *another;
690                 name = orig_name;
691                 len = strlen(name);
692                 if (isnull)
693                         die("git apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
694                 another = find_name(line, NULL, p_value, TERM_TAB);
695                 if (!another || memcmp(another, name, len + 1))
696                         die("git apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
697                 free(another);
698                 return orig_name;
699         }
700         else {
701                 /* expect "/dev/null" */
702                 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
703                         die("git apply: bad git-diff - expected /dev/null on line %d", linenr);
704                 return NULL;
705         }
708 static int gitdiff_oldname(const char *line, struct patch *patch)
710         patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
711         return 0;
714 static int gitdiff_newname(const char *line, struct patch *patch)
716         patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
717         return 0;
720 static int gitdiff_oldmode(const char *line, struct patch *patch)
722         patch->old_mode = strtoul(line, NULL, 8);
723         return 0;
726 static int gitdiff_newmode(const char *line, struct patch *patch)
728         patch->new_mode = strtoul(line, NULL, 8);
729         return 0;
732 static int gitdiff_delete(const char *line, struct patch *patch)
734         patch->is_delete = 1;
735         patch->old_name = patch->def_name;
736         return gitdiff_oldmode(line, patch);
739 static int gitdiff_newfile(const char *line, struct patch *patch)
741         patch->is_new = 1;
742         patch->new_name = patch->def_name;
743         return gitdiff_newmode(line, patch);
746 static int gitdiff_copysrc(const char *line, struct patch *patch)
748         patch->is_copy = 1;
749         patch->old_name = find_name(line, NULL, 0, 0);
750         return 0;
753 static int gitdiff_copydst(const char *line, struct patch *patch)
755         patch->is_copy = 1;
756         patch->new_name = find_name(line, NULL, 0, 0);
757         return 0;
760 static int gitdiff_renamesrc(const char *line, struct patch *patch)
762         patch->is_rename = 1;
763         patch->old_name = find_name(line, NULL, 0, 0);
764         return 0;
767 static int gitdiff_renamedst(const char *line, struct patch *patch)
769         patch->is_rename = 1;
770         patch->new_name = find_name(line, NULL, 0, 0);
771         return 0;
774 static int gitdiff_similarity(const char *line, struct patch *patch)
776         if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
777                 patch->score = 0;
778         return 0;
781 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
783         if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
784                 patch->score = 0;
785         return 0;
788 static int gitdiff_index(const char *line, struct patch *patch)
790         /*
791          * index line is N hexadecimal, "..", N hexadecimal,
792          * and optional space with octal mode.
793          */
794         const char *ptr, *eol;
795         int len;
797         ptr = strchr(line, '.');
798         if (!ptr || ptr[1] != '.' || 40 < ptr - line)
799                 return 0;
800         len = ptr - line;
801         memcpy(patch->old_sha1_prefix, line, len);
802         patch->old_sha1_prefix[len] = 0;
804         line = ptr + 2;
805         ptr = strchr(line, ' ');
806         eol = strchr(line, '\n');
808         if (!ptr || eol < ptr)
809                 ptr = eol;
810         len = ptr - line;
812         if (40 < len)
813                 return 0;
814         memcpy(patch->new_sha1_prefix, line, len);
815         patch->new_sha1_prefix[len] = 0;
816         if (*ptr == ' ')
817                 patch->old_mode = strtoul(ptr+1, NULL, 8);
818         return 0;
821 /*
822  * This is normal for a diff that doesn't change anything: we'll fall through
823  * into the next diff. Tell the parser to break out.
824  */
825 static int gitdiff_unrecognized(const char *line, struct patch *patch)
827         return -1;
830 static const char *stop_at_slash(const char *line, int llen)
832         int nslash = p_value;
833         int i;
835         for (i = 0; i < llen; i++) {
836                 int ch = line[i];
837                 if (ch == '/' && --nslash <= 0)
838                         return &line[i];
839         }
840         return NULL;
843 /*
844  * This is to extract the same name that appears on "diff --git"
845  * line.  We do not find and return anything if it is a rename
846  * patch, and it is OK because we will find the name elsewhere.
847  * We need to reliably find name only when it is mode-change only,
848  * creation or deletion of an empty file.  In any of these cases,
849  * both sides are the same name under a/ and b/ respectively.
850  */
851 static char *git_header_name(char *line, int llen)
853         const char *name;
854         const char *second = NULL;
855         size_t len;
857         line += strlen("diff --git ");
858         llen -= strlen("diff --git ");
860         if (*line == '"') {
861                 const char *cp;
862                 struct strbuf first = STRBUF_INIT;
863                 struct strbuf sp = STRBUF_INIT;
865                 if (unquote_c_style(&first, line, &second))
866                         goto free_and_fail1;
868                 /* advance to the first slash */
869                 cp = stop_at_slash(first.buf, first.len);
870                 /* we do not accept absolute paths */
871                 if (!cp || cp == first.buf)
872                         goto free_and_fail1;
873                 strbuf_remove(&first, 0, cp + 1 - first.buf);
875                 /*
876                  * second points at one past closing dq of name.
877                  * find the second name.
878                  */
879                 while ((second < line + llen) && isspace(*second))
880                         second++;
882                 if (line + llen <= second)
883                         goto free_and_fail1;
884                 if (*second == '"') {
885                         if (unquote_c_style(&sp, second, NULL))
886                                 goto free_and_fail1;
887                         cp = stop_at_slash(sp.buf, sp.len);
888                         if (!cp || cp == sp.buf)
889                                 goto free_and_fail1;
890                         /* They must match, otherwise ignore */
891                         if (strcmp(cp + 1, first.buf))
892                                 goto free_and_fail1;
893                         strbuf_release(&sp);
894                         return strbuf_detach(&first, NULL);
895                 }
897                 /* unquoted second */
898                 cp = stop_at_slash(second, line + llen - second);
899                 if (!cp || cp == second)
900                         goto free_and_fail1;
901                 cp++;
902                 if (line + llen - cp != first.len + 1 ||
903                     memcmp(first.buf, cp, first.len))
904                         goto free_and_fail1;
905                 return strbuf_detach(&first, NULL);
907         free_and_fail1:
908                 strbuf_release(&first);
909                 strbuf_release(&sp);
910                 return NULL;
911         }
913         /* unquoted first name */
914         name = stop_at_slash(line, llen);
915         if (!name || name == line)
916                 return NULL;
917         name++;
919         /*
920          * since the first name is unquoted, a dq if exists must be
921          * the beginning of the second name.
922          */
923         for (second = name; second < line + llen; second++) {
924                 if (*second == '"') {
925                         struct strbuf sp = STRBUF_INIT;
926                         const char *np;
928                         if (unquote_c_style(&sp, second, NULL))
929                                 goto free_and_fail2;
931                         np = stop_at_slash(sp.buf, sp.len);
932                         if (!np || np == sp.buf)
933                                 goto free_and_fail2;
934                         np++;
936                         len = sp.buf + sp.len - np;
937                         if (len < second - name &&
938                             !strncmp(np, name, len) &&
939                             isspace(name[len])) {
940                                 /* Good */
941                                 strbuf_remove(&sp, 0, np - sp.buf);
942                                 return strbuf_detach(&sp, NULL);
943                         }
945                 free_and_fail2:
946                         strbuf_release(&sp);
947                         return NULL;
948                 }
949         }
951         /*
952          * Accept a name only if it shows up twice, exactly the same
953          * form.
954          */
955         for (len = 0 ; ; len++) {
956                 switch (name[len]) {
957                 default:
958                         continue;
959                 case '\n':
960                         return NULL;
961                 case '\t': case ' ':
962                         second = name+len;
963                         for (;;) {
964                                 char c = *second++;
965                                 if (c == '\n')
966                                         return NULL;
967                                 if (c == '/')
968                                         break;
969                         }
970                         if (second[len] == '\n' && !memcmp(name, second, len)) {
971                                 return xmemdupz(name, len);
972                         }
973                 }
974         }
977 /* Verify that we recognize the lines following a git header */
978 static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
980         unsigned long offset;
982         /* A git diff has explicit new/delete information, so we don't guess */
983         patch->is_new = 0;
984         patch->is_delete = 0;
986         /*
987          * Some things may not have the old name in the
988          * rest of the headers anywhere (pure mode changes,
989          * or removing or adding empty files), so we get
990          * the default name from the header.
991          */
992         patch->def_name = git_header_name(line, len);
993         if (patch->def_name && root) {
994                 char *s = xmalloc(root_len + strlen(patch->def_name) + 1);
995                 strcpy(s, root);
996                 strcpy(s + root_len, patch->def_name);
997                 free(patch->def_name);
998                 patch->def_name = s;
999         }
1001         line += len;
1002         size -= len;
1003         linenr++;
1004         for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
1005                 static const struct opentry {
1006                         const char *str;
1007                         int (*fn)(const char *, struct patch *);
1008                 } optable[] = {
1009                         { "@@ -", gitdiff_hdrend },
1010                         { "--- ", gitdiff_oldname },
1011                         { "+++ ", gitdiff_newname },
1012                         { "old mode ", gitdiff_oldmode },
1013                         { "new mode ", gitdiff_newmode },
1014                         { "deleted file mode ", gitdiff_delete },
1015                         { "new file mode ", gitdiff_newfile },
1016                         { "copy from ", gitdiff_copysrc },
1017                         { "copy to ", gitdiff_copydst },
1018                         { "rename old ", gitdiff_renamesrc },
1019                         { "rename new ", gitdiff_renamedst },
1020                         { "rename from ", gitdiff_renamesrc },
1021                         { "rename to ", gitdiff_renamedst },
1022                         { "similarity index ", gitdiff_similarity },
1023                         { "dissimilarity index ", gitdiff_dissimilarity },
1024                         { "index ", gitdiff_index },
1025                         { "", gitdiff_unrecognized },
1026                 };
1027                 int i;
1029                 len = linelen(line, size);
1030                 if (!len || line[len-1] != '\n')
1031                         break;
1032                 for (i = 0; i < ARRAY_SIZE(optable); i++) {
1033                         const struct opentry *p = optable + i;
1034                         int oplen = strlen(p->str);
1035                         if (len < oplen || memcmp(p->str, line, oplen))
1036                                 continue;
1037                         if (p->fn(line + oplen, patch) < 0)
1038                                 return offset;
1039                         break;
1040                 }
1041         }
1043         return offset;
1046 static int parse_num(const char *line, unsigned long *p)
1048         char *ptr;
1050         if (!isdigit(*line))
1051                 return 0;
1052         *p = strtoul(line, &ptr, 10);
1053         return ptr - line;
1056 static int parse_range(const char *line, int len, int offset, const char *expect,
1057                        unsigned long *p1, unsigned long *p2)
1059         int digits, ex;
1061         if (offset < 0 || offset >= len)
1062                 return -1;
1063         line += offset;
1064         len -= offset;
1066         digits = parse_num(line, p1);
1067         if (!digits)
1068                 return -1;
1070         offset += digits;
1071         line += digits;
1072         len -= digits;
1074         *p2 = 1;
1075         if (*line == ',') {
1076                 digits = parse_num(line+1, p2);
1077                 if (!digits)
1078                         return -1;
1080                 offset += digits+1;
1081                 line += digits+1;
1082                 len -= digits+1;
1083         }
1085         ex = strlen(expect);
1086         if (ex > len)
1087                 return -1;
1088         if (memcmp(line, expect, ex))
1089                 return -1;
1091         return offset + ex;
1094 static void recount_diff(char *line, int size, struct fragment *fragment)
1096         int oldlines = 0, newlines = 0, ret = 0;
1098         if (size < 1) {
1099                 warning("recount: ignore empty hunk");
1100                 return;
1101         }
1103         for (;;) {
1104                 int len = linelen(line, size);
1105                 size -= len;
1106                 line += len;
1108                 if (size < 1)
1109                         break;
1111                 switch (*line) {
1112                 case ' ': case '\n':
1113                         newlines++;
1114                         /* fall through */
1115                 case '-':
1116                         oldlines++;
1117                         continue;
1118                 case '+':
1119                         newlines++;
1120                         continue;
1121                 case '\\':
1122                         continue;
1123                 case '@':
1124                         ret = size < 3 || prefixcmp(line, "@@ ");
1125                         break;
1126                 case 'd':
1127                         ret = size < 5 || prefixcmp(line, "diff ");
1128                         break;
1129                 default:
1130                         ret = -1;
1131                         break;
1132                 }
1133                 if (ret) {
1134                         warning("recount: unexpected line: %.*s",
1135                                 (int)linelen(line, size), line);
1136                         return;
1137                 }
1138                 break;
1139         }
1140         fragment->oldlines = oldlines;
1141         fragment->newlines = newlines;
1144 /*
1145  * Parse a unified diff fragment header of the
1146  * form "@@ -a,b +c,d @@"
1147  */
1148 static int parse_fragment_header(char *line, int len, struct fragment *fragment)
1150         int offset;
1152         if (!len || line[len-1] != '\n')
1153                 return -1;
1155         /* Figure out the number of lines in a fragment */
1156         offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1157         offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1159         return offset;
1162 static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
1164         unsigned long offset, len;
1166         patch->is_toplevel_relative = 0;
1167         patch->is_rename = patch->is_copy = 0;
1168         patch->is_new = patch->is_delete = -1;
1169         patch->old_mode = patch->new_mode = 0;
1170         patch->old_name = patch->new_name = NULL;
1171         for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
1172                 unsigned long nextlen;
1174                 len = linelen(line, size);
1175                 if (!len)
1176                         break;
1178                 /* Testing this early allows us to take a few shortcuts.. */
1179                 if (len < 6)
1180                         continue;
1182                 /*
1183                  * Make sure we don't find any unconnected patch fragments.
1184                  * That's a sign that we didn't find a header, and that a
1185                  * patch has become corrupted/broken up.
1186                  */
1187                 if (!memcmp("@@ -", line, 4)) {
1188                         struct fragment dummy;
1189                         if (parse_fragment_header(line, len, &dummy) < 0)
1190                                 continue;
1191                         die("patch fragment without header at line %d: %.*s",
1192                             linenr, (int)len-1, line);
1193                 }
1195                 if (size < len + 6)
1196                         break;
1198                 /*
1199                  * Git patch? It might not have a real patch, just a rename
1200                  * or mode change, so we handle that specially
1201                  */
1202                 if (!memcmp("diff --git ", line, 11)) {
1203                         int git_hdr_len = parse_git_header(line, len, size, patch);
1204                         if (git_hdr_len <= len)
1205                                 continue;
1206                         if (!patch->old_name && !patch->new_name) {
1207                                 if (!patch->def_name)
1208                                         die("git diff header lacks filename information when removing "
1209                                             "%d leading pathname components (line %d)" , p_value, linenr);
1210                                 patch->old_name = patch->new_name = patch->def_name;
1211                         }
1212                         patch->is_toplevel_relative = 1;
1213                         *hdrsize = git_hdr_len;
1214                         return offset;
1215                 }
1217                 /* --- followed by +++ ? */
1218                 if (memcmp("--- ", line,  4) || memcmp("+++ ", line + len, 4))
1219                         continue;
1221                 /*
1222                  * We only accept unified patches, so we want it to
1223                  * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1224                  * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1225                  */
1226                 nextlen = linelen(line + len, size - len);
1227                 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1228                         continue;
1230                 /* Ok, we'll consider it a patch */
1231                 parse_traditional_patch(line, line+len, patch);
1232                 *hdrsize = len + nextlen;
1233                 linenr += 2;
1234                 return offset;
1235         }
1236         return -1;
1239 static void record_ws_error(unsigned result, const char *line, int len, int linenr)
1241         char *err;
1243         if (!result)
1244                 return;
1246         whitespace_error++;
1247         if (squelch_whitespace_errors &&
1248             squelch_whitespace_errors < whitespace_error)
1249                 return;
1251         err = whitespace_error_string(result);
1252         fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1253                 patch_input_file, linenr, err, len, line);
1254         free(err);
1257 static void check_whitespace(const char *line, int len, unsigned ws_rule)
1259         unsigned result = ws_check(line + 1, len - 1, ws_rule);
1261         record_ws_error(result, line + 1, len - 2, linenr);
1264 /*
1265  * Parse a unified diff. Note that this really needs to parse each
1266  * fragment separately, since the only way to know the difference
1267  * between a "---" that is part of a patch, and a "---" that starts
1268  * the next patch is to look at the line counts..
1269  */
1270 static int parse_fragment(char *line, unsigned long size,
1271                           struct patch *patch, struct fragment *fragment)
1273         int added, deleted;
1274         int len = linelen(line, size), offset;
1275         unsigned long oldlines, newlines;
1276         unsigned long leading, trailing;
1278         offset = parse_fragment_header(line, len, fragment);
1279         if (offset < 0)
1280                 return -1;
1281         if (offset > 0 && patch->recount)
1282                 recount_diff(line + offset, size - offset, fragment);
1283         oldlines = fragment->oldlines;
1284         newlines = fragment->newlines;
1285         leading = 0;
1286         trailing = 0;
1288         /* Parse the thing.. */
1289         line += len;
1290         size -= len;
1291         linenr++;
1292         added = deleted = 0;
1293         for (offset = len;
1294              0 < size;
1295              offset += len, size -= len, line += len, linenr++) {
1296                 if (!oldlines && !newlines)
1297                         break;
1298                 len = linelen(line, size);
1299                 if (!len || line[len-1] != '\n')
1300                         return -1;
1301                 switch (*line) {
1302                 default:
1303                         return -1;
1304                 case '\n': /* newer GNU diff, an empty context line */
1305                 case ' ':
1306                         oldlines--;
1307                         newlines--;
1308                         if (!deleted && !added)
1309                                 leading++;
1310                         trailing++;
1311                         break;
1312                 case '-':
1313                         if (apply_in_reverse &&
1314                             ws_error_action != nowarn_ws_error)
1315                                 check_whitespace(line, len, patch->ws_rule);
1316                         deleted++;
1317                         oldlines--;
1318                         trailing = 0;
1319                         break;
1320                 case '+':
1321                         if (!apply_in_reverse &&
1322                             ws_error_action != nowarn_ws_error)
1323                                 check_whitespace(line, len, patch->ws_rule);
1324                         added++;
1325                         newlines--;
1326                         trailing = 0;
1327                         break;
1329                 /*
1330                  * We allow "\ No newline at end of file". Depending
1331                  * on locale settings when the patch was produced we
1332                  * don't know what this line looks like. The only
1333                  * thing we do know is that it begins with "\ ".
1334                  * Checking for 12 is just for sanity check -- any
1335                  * l10n of "\ No newline..." is at least that long.
1336                  */
1337                 case '\\':
1338                         if (len < 12 || memcmp(line, "\\ ", 2))
1339                                 return -1;
1340                         break;
1341                 }
1342         }
1343         if (oldlines || newlines)
1344                 return -1;
1345         fragment->leading = leading;
1346         fragment->trailing = trailing;
1348         /*
1349          * If a fragment ends with an incomplete line, we failed to include
1350          * it in the above loop because we hit oldlines == newlines == 0
1351          * before seeing it.
1352          */
1353         if (12 < size && !memcmp(line, "\\ ", 2))
1354                 offset += linelen(line, size);
1356         patch->lines_added += added;
1357         patch->lines_deleted += deleted;
1359         if (0 < patch->is_new && oldlines)
1360                 return error("new file depends on old contents");
1361         if (0 < patch->is_delete && newlines)
1362                 return error("deleted file still has contents");
1363         return offset;
1366 static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
1368         unsigned long offset = 0;
1369         unsigned long oldlines = 0, newlines = 0, context = 0;
1370         struct fragment **fragp = &patch->fragments;
1372         while (size > 4 && !memcmp(line, "@@ -", 4)) {
1373                 struct fragment *fragment;
1374                 int len;
1376                 fragment = xcalloc(1, sizeof(*fragment));
1377                 fragment->linenr = linenr;
1378                 len = parse_fragment(line, size, patch, fragment);
1379                 if (len <= 0)
1380                         die("corrupt patch at line %d", linenr);
1381                 fragment->patch = line;
1382                 fragment->size = len;
1383                 oldlines += fragment->oldlines;
1384                 newlines += fragment->newlines;
1385                 context += fragment->leading + fragment->trailing;
1387                 *fragp = fragment;
1388                 fragp = &fragment->next;
1390                 offset += len;
1391                 line += len;
1392                 size -= len;
1393         }
1395         /*
1396          * If something was removed (i.e. we have old-lines) it cannot
1397          * be creation, and if something was added it cannot be
1398          * deletion.  However, the reverse is not true; --unified=0
1399          * patches that only add are not necessarily creation even
1400          * though they do not have any old lines, and ones that only
1401          * delete are not necessarily deletion.
1402          *
1403          * Unfortunately, a real creation/deletion patch do _not_ have
1404          * any context line by definition, so we cannot safely tell it
1405          * apart with --unified=0 insanity.  At least if the patch has
1406          * more than one hunk it is not creation or deletion.
1407          */
1408         if (patch->is_new < 0 &&
1409             (oldlines || (patch->fragments && patch->fragments->next)))
1410                 patch->is_new = 0;
1411         if (patch->is_delete < 0 &&
1412             (newlines || (patch->fragments && patch->fragments->next)))
1413                 patch->is_delete = 0;
1415         if (0 < patch->is_new && oldlines)
1416                 die("new file %s depends on old contents", patch->new_name);
1417         if (0 < patch->is_delete && newlines)
1418                 die("deleted file %s still has contents", patch->old_name);
1419         if (!patch->is_delete && !newlines && context)
1420                 fprintf(stderr, "** warning: file %s becomes empty but "
1421                         "is not deleted\n", patch->new_name);
1423         return offset;
1426 static inline int metadata_changes(struct patch *patch)
1428         return  patch->is_rename > 0 ||
1429                 patch->is_copy > 0 ||
1430                 patch->is_new > 0 ||
1431                 patch->is_delete ||
1432                 (patch->old_mode && patch->new_mode &&
1433                  patch->old_mode != patch->new_mode);
1436 static char *inflate_it(const void *data, unsigned long size,
1437                         unsigned long inflated_size)
1439         z_stream stream;
1440         void *out;
1441         int st;
1443         memset(&stream, 0, sizeof(stream));
1445         stream.next_in = (unsigned char *)data;
1446         stream.avail_in = size;
1447         stream.next_out = out = xmalloc(inflated_size);
1448         stream.avail_out = inflated_size;
1449         git_inflate_init(&stream);
1450         st = git_inflate(&stream, Z_FINISH);
1451         git_inflate_end(&stream);
1452         if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1453                 free(out);
1454                 return NULL;
1455         }
1456         return out;
1459 static struct fragment *parse_binary_hunk(char **buf_p,
1460                                           unsigned long *sz_p,
1461                                           int *status_p,
1462                                           int *used_p)
1464         /*
1465          * Expect a line that begins with binary patch method ("literal"
1466          * or "delta"), followed by the length of data before deflating.
1467          * a sequence of 'length-byte' followed by base-85 encoded data
1468          * should follow, terminated by a newline.
1469          *
1470          * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1471          * and we would limit the patch line to 66 characters,
1472          * so one line can fit up to 13 groups that would decode
1473          * to 52 bytes max.  The length byte 'A'-'Z' corresponds
1474          * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1475          */
1476         int llen, used;
1477         unsigned long size = *sz_p;
1478         char *buffer = *buf_p;
1479         int patch_method;
1480         unsigned long origlen;
1481         char *data = NULL;
1482         int hunk_size = 0;
1483         struct fragment *frag;
1485         llen = linelen(buffer, size);
1486         used = llen;
1488         *status_p = 0;
1490         if (!prefixcmp(buffer, "delta ")) {
1491                 patch_method = BINARY_DELTA_DEFLATED;
1492                 origlen = strtoul(buffer + 6, NULL, 10);
1493         }
1494         else if (!prefixcmp(buffer, "literal ")) {
1495                 patch_method = BINARY_LITERAL_DEFLATED;
1496                 origlen = strtoul(buffer + 8, NULL, 10);
1497         }
1498         else
1499                 return NULL;
1501         linenr++;
1502         buffer += llen;
1503         while (1) {
1504                 int byte_length, max_byte_length, newsize;
1505                 llen = linelen(buffer, size);
1506                 used += llen;
1507                 linenr++;
1508                 if (llen == 1) {
1509                         /* consume the blank line */
1510                         buffer++;
1511                         size--;
1512                         break;
1513                 }
1514                 /*
1515                  * Minimum line is "A00000\n" which is 7-byte long,
1516                  * and the line length must be multiple of 5 plus 2.
1517                  */
1518                 if ((llen < 7) || (llen-2) % 5)
1519                         goto corrupt;
1520                 max_byte_length = (llen - 2) / 5 * 4;
1521                 byte_length = *buffer;
1522                 if ('A' <= byte_length && byte_length <= 'Z')
1523                         byte_length = byte_length - 'A' + 1;
1524                 else if ('a' <= byte_length && byte_length <= 'z')
1525                         byte_length = byte_length - 'a' + 27;
1526                 else
1527                         goto corrupt;
1528                 /* if the input length was not multiple of 4, we would
1529                  * have filler at the end but the filler should never
1530                  * exceed 3 bytes
1531                  */
1532                 if (max_byte_length < byte_length ||
1533                     byte_length <= max_byte_length - 4)
1534                         goto corrupt;
1535                 newsize = hunk_size + byte_length;
1536                 data = xrealloc(data, newsize);
1537                 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1538                         goto corrupt;
1539                 hunk_size = newsize;
1540                 buffer += llen;
1541                 size -= llen;
1542         }
1544         frag = xcalloc(1, sizeof(*frag));
1545         frag->patch = inflate_it(data, hunk_size, origlen);
1546         if (!frag->patch)
1547                 goto corrupt;
1548         free(data);
1549         frag->size = origlen;
1550         *buf_p = buffer;
1551         *sz_p = size;
1552         *used_p = used;
1553         frag->binary_patch_method = patch_method;
1554         return frag;
1556  corrupt:
1557         free(data);
1558         *status_p = -1;
1559         error("corrupt binary patch at line %d: %.*s",
1560               linenr-1, llen-1, buffer);
1561         return NULL;
1564 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1566         /*
1567          * We have read "GIT binary patch\n"; what follows is a line
1568          * that says the patch method (currently, either "literal" or
1569          * "delta") and the length of data before deflating; a
1570          * sequence of 'length-byte' followed by base-85 encoded data
1571          * follows.
1572          *
1573          * When a binary patch is reversible, there is another binary
1574          * hunk in the same format, starting with patch method (either
1575          * "literal" or "delta") with the length of data, and a sequence
1576          * of length-byte + base-85 encoded data, terminated with another
1577          * empty line.  This data, when applied to the postimage, produces
1578          * the preimage.
1579          */
1580         struct fragment *forward;
1581         struct fragment *reverse;
1582         int status;
1583         int used, used_1;
1585         forward = parse_binary_hunk(&buffer, &size, &status, &used);
1586         if (!forward && !status)
1587                 /* there has to be one hunk (forward hunk) */
1588                 return error("unrecognized binary patch at line %d", linenr-1);
1589         if (status)
1590                 /* otherwise we already gave an error message */
1591                 return status;
1593         reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1594         if (reverse)
1595                 used += used_1;
1596         else if (status) {
1597                 /*
1598                  * Not having reverse hunk is not an error, but having
1599                  * a corrupt reverse hunk is.
1600                  */
1601                 free((void*) forward->patch);
1602                 free(forward);
1603                 return status;
1604         }
1605         forward->next = reverse;
1606         patch->fragments = forward;
1607         patch->is_binary = 1;
1608         return used;
1611 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1613         int hdrsize, patchsize;
1614         int offset = find_header(buffer, size, &hdrsize, patch);
1616         if (offset < 0)
1617                 return offset;
1619         patch->ws_rule = whitespace_rule(patch->new_name
1620                                          ? patch->new_name
1621                                          : patch->old_name);
1623         patchsize = parse_single_patch(buffer + offset + hdrsize,
1624                                        size - offset - hdrsize, patch);
1626         if (!patchsize) {
1627                 static const char *binhdr[] = {
1628                         "Binary files ",
1629                         "Files ",
1630                         NULL,
1631                 };
1632                 static const char git_binary[] = "GIT binary patch\n";
1633                 int i;
1634                 int hd = hdrsize + offset;
1635                 unsigned long llen = linelen(buffer + hd, size - hd);
1637                 if (llen == sizeof(git_binary) - 1 &&
1638                     !memcmp(git_binary, buffer + hd, llen)) {
1639                         int used;
1640                         linenr++;
1641                         used = parse_binary(buffer + hd + llen,
1642                                             size - hd - llen, patch);
1643                         if (used)
1644                                 patchsize = used + llen;
1645                         else
1646                                 patchsize = 0;
1647                 }
1648                 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1649                         for (i = 0; binhdr[i]; i++) {
1650                                 int len = strlen(binhdr[i]);
1651                                 if (len < size - hd &&
1652                                     !memcmp(binhdr[i], buffer + hd, len)) {
1653                                         linenr++;
1654                                         patch->is_binary = 1;
1655                                         patchsize = llen;
1656                                         break;
1657                                 }
1658                         }
1659                 }
1661                 /* Empty patch cannot be applied if it is a text patch
1662                  * without metadata change.  A binary patch appears
1663                  * empty to us here.
1664                  */
1665                 if ((apply || check) &&
1666                     (!patch->is_binary && !metadata_changes(patch)))
1667                         die("patch with only garbage at line %d", linenr);
1668         }
1670         return offset + hdrsize + patchsize;
1673 #define swap(a,b) myswap((a),(b),sizeof(a))
1675 #define myswap(a, b, size) do {         \
1676         unsigned char mytmp[size];      \
1677         memcpy(mytmp, &a, size);                \
1678         memcpy(&a, &b, size);           \
1679         memcpy(&b, mytmp, size);                \
1680 } while (0)
1682 static void reverse_patches(struct patch *p)
1684         for (; p; p = p->next) {
1685                 struct fragment *frag = p->fragments;
1687                 swap(p->new_name, p->old_name);
1688                 swap(p->new_mode, p->old_mode);
1689                 swap(p->is_new, p->is_delete);
1690                 swap(p->lines_added, p->lines_deleted);
1691                 swap(p->old_sha1_prefix, p->new_sha1_prefix);
1693                 for (; frag; frag = frag->next) {
1694                         swap(frag->newpos, frag->oldpos);
1695                         swap(frag->newlines, frag->oldlines);
1696                 }
1697         }
1700 static const char pluses[] =
1701 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1702 static const char minuses[]=
1703 "----------------------------------------------------------------------";
1705 static void show_stats(struct patch *patch)
1707         struct strbuf qname = STRBUF_INIT;
1708         char *cp = patch->new_name ? patch->new_name : patch->old_name;
1709         int max, add, del;
1711         quote_c_style(cp, &qname, NULL, 0);
1713         /*
1714          * "scale" the filename
1715          */
1716         max = max_len;
1717         if (max > 50)
1718                 max = 50;
1720         if (qname.len > max) {
1721                 cp = strchr(qname.buf + qname.len + 3 - max, '/');
1722                 if (!cp)
1723                         cp = qname.buf + qname.len + 3 - max;
1724                 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
1725         }
1727         if (patch->is_binary) {
1728                 printf(" %-*s |  Bin\n", max, qname.buf);
1729                 strbuf_release(&qname);
1730                 return;
1731         }
1733         printf(" %-*s |", max, qname.buf);
1734         strbuf_release(&qname);
1736         /*
1737          * scale the add/delete
1738          */
1739         max = max + max_change > 70 ? 70 - max : max_change;
1740         add = patch->lines_added;
1741         del = patch->lines_deleted;
1743         if (max_change > 0) {
1744                 int total = ((add + del) * max + max_change / 2) / max_change;
1745                 add = (add * max + max_change / 2) / max_change;
1746                 del = total - add;
1747         }
1748         printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
1749                 add, pluses, del, minuses);
1752 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
1754         switch (st->st_mode & S_IFMT) {
1755         case S_IFLNK:
1756                 if (strbuf_readlink(buf, path, st->st_size) < 0)
1757                         return error("unable to read symlink %s", path);
1758                 return 0;
1759         case S_IFREG:
1760                 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
1761                         return error("unable to open or read %s", path);
1762                 convert_to_git(path, buf->buf, buf->len, buf, 0);
1763                 return 0;
1764         default:
1765                 return -1;
1766         }
1769 /*
1770  * Update the preimage, and the common lines in postimage,
1771  * from buffer buf of length len. If postlen is 0 the postimage
1772  * is updated in place, otherwise it's updated on a new buffer
1773  * of length postlen
1774  */
1776 static void update_pre_post_images(struct image *preimage,
1777                                    struct image *postimage,
1778                                    char *buf,
1779                                    size_t len, size_t postlen)
1781         int i, ctx;
1782         char *new, *old, *fixed;
1783         struct image fixed_preimage;
1785         /*
1786          * Update the preimage with whitespace fixes.  Note that we
1787          * are not losing preimage->buf -- apply_one_fragment() will
1788          * free "oldlines".
1789          */
1790         prepare_image(&fixed_preimage, buf, len, 1);
1791         assert(fixed_preimage.nr == preimage->nr);
1792         for (i = 0; i < preimage->nr; i++)
1793                 fixed_preimage.line[i].flag = preimage->line[i].flag;
1794         free(preimage->line_allocated);
1795         *preimage = fixed_preimage;
1797         /*
1798          * Adjust the common context lines in postimage. This can be
1799          * done in-place when we are just doing whitespace fixing,
1800          * which does not make the string grow, but needs a new buffer
1801          * when ignoring whitespace causes the update, since in this case
1802          * we could have e.g. tabs converted to multiple spaces.
1803          * We trust the caller to tell us if the update can be done
1804          * in place (postlen==0) or not.
1805          */
1806         old = postimage->buf;
1807         if (postlen)
1808                 new = postimage->buf = xmalloc(postlen);
1809         else
1810                 new = old;
1811         fixed = preimage->buf;
1812         for (i = ctx = 0; i < postimage->nr; i++) {
1813                 size_t len = postimage->line[i].len;
1814                 if (!(postimage->line[i].flag & LINE_COMMON)) {
1815                         /* an added line -- no counterparts in preimage */
1816                         memmove(new, old, len);
1817                         old += len;
1818                         new += len;
1819                         continue;
1820                 }
1822                 /* a common context -- skip it in the original postimage */
1823                 old += len;
1825                 /* and find the corresponding one in the fixed preimage */
1826                 while (ctx < preimage->nr &&
1827                        !(preimage->line[ctx].flag & LINE_COMMON)) {
1828                         fixed += preimage->line[ctx].len;
1829                         ctx++;
1830                 }
1831                 if (preimage->nr <= ctx)
1832                         die("oops");
1834                 /* and copy it in, while fixing the line length */
1835                 len = preimage->line[ctx].len;
1836                 memcpy(new, fixed, len);
1837                 new += len;
1838                 fixed += len;
1839                 postimage->line[i].len = len;
1840                 ctx++;
1841         }
1843         /* Fix the length of the whole thing */
1844         postimage->len = new - postimage->buf;
1847 static int match_fragment(struct image *img,
1848                           struct image *preimage,
1849                           struct image *postimage,
1850                           unsigned long try,
1851                           int try_lno,
1852                           unsigned ws_rule,
1853                           int match_beginning, int match_end)
1855         int i;
1856         char *fixed_buf, *buf, *orig, *target;
1858         if (preimage->nr + try_lno > img->nr)
1859                 return 0;
1861         if (match_beginning && try_lno)
1862                 return 0;
1864         if (match_end && preimage->nr + try_lno != img->nr)
1865                 return 0;
1867         /* Quick hash check */
1868         for (i = 0; i < preimage->nr; i++)
1869                 if (preimage->line[i].hash != img->line[try_lno + i].hash)
1870                         return 0;
1872         /*
1873          * Do we have an exact match?  If we were told to match
1874          * at the end, size must be exactly at try+fragsize,
1875          * otherwise try+fragsize must be still within the preimage,
1876          * and either case, the old piece should match the preimage
1877          * exactly.
1878          */
1879         if ((match_end
1880              ? (try + preimage->len == img->len)
1881              : (try + preimage->len <= img->len)) &&
1882             !memcmp(img->buf + try, preimage->buf, preimage->len))
1883                 return 1;
1885         /*
1886          * No exact match. If we are ignoring whitespace, run a line-by-line
1887          * fuzzy matching. We collect all the line length information because
1888          * we need it to adjust whitespace if we match.
1889          */
1890         if (ws_ignore_action == ignore_ws_change) {
1891                 size_t imgoff = 0;
1892                 size_t preoff = 0;
1893                 size_t postlen = postimage->len;
1894                 for (i = 0; i < preimage->nr; i++) {
1895                         size_t prelen = preimage->line[i].len;
1896                         size_t imglen = img->line[try_lno+i].len;
1898                         if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,
1899                                               preimage->buf + preoff, prelen))
1900                                 return 0;
1901                         if (preimage->line[i].flag & LINE_COMMON)
1902                                 postlen += imglen - prelen;
1903                         imgoff += imglen;
1904                         preoff += prelen;
1905                 }
1907                 /*
1908                  * Ok, the preimage matches with whitespace fuzz.
1909                  *
1910                  * imgoff now holds the true length of the target that
1911                  * matches the preimage.  Update the preimage and
1912                  * the common postimage context lines to use the same
1913                  * whitespace as the target.
1914                  */
1915                 fixed_buf = xmalloc(imgoff);
1916                 memcpy(fixed_buf, img->buf + try, imgoff);
1917                 update_pre_post_images(preimage, postimage,
1918                                 fixed_buf, imgoff, postlen);
1919                 return 1;
1920         }
1922         if (ws_error_action != correct_ws_error)
1923                 return 0;
1925         /*
1926          * The hunk does not apply byte-by-byte, but the hash says
1927          * it might with whitespace fuzz. We haven't been asked to
1928          * ignore whitespace, we were asked to correct whitespace
1929          * errors, so let's try matching after whitespace correction.
1930          */
1931         fixed_buf = xmalloc(preimage->len + 1);
1932         buf = fixed_buf;
1933         orig = preimage->buf;
1934         target = img->buf + try;
1935         for (i = 0; i < preimage->nr; i++) {
1936                 size_t fixlen; /* length after fixing the preimage */
1937                 size_t oldlen = preimage->line[i].len;
1938                 size_t tgtlen = img->line[try_lno + i].len;
1939                 size_t tgtfixlen; /* length after fixing the target line */
1940                 char tgtfixbuf[1024], *tgtfix;
1941                 int match;
1943                 /* Try fixing the line in the preimage */
1944                 fixlen = ws_fix_copy(buf, orig, oldlen, ws_rule, NULL);
1946                 /* Try fixing the line in the target */
1947                 if (sizeof(tgtfixbuf) > tgtlen)
1948                         tgtfix = tgtfixbuf;
1949                 else
1950                         tgtfix = xmalloc(tgtlen);
1951                 tgtfixlen = ws_fix_copy(tgtfix, target, tgtlen, ws_rule, NULL);
1953                 /*
1954                  * If they match, either the preimage was based on
1955                  * a version before our tree fixed whitespace breakage,
1956                  * or we are lacking a whitespace-fix patch the tree
1957                  * the preimage was based on already had (i.e. target
1958                  * has whitespace breakage, the preimage doesn't).
1959                  * In either case, we are fixing the whitespace breakages
1960                  * so we might as well take the fix together with their
1961                  * real change.
1962                  */
1963                 match = (tgtfixlen == fixlen && !memcmp(tgtfix, buf, fixlen));
1965                 if (tgtfix != tgtfixbuf)
1966                         free(tgtfix);
1967                 if (!match)
1968                         goto unmatch_exit;
1970                 orig += oldlen;
1971                 buf += fixlen;
1972                 target += tgtlen;
1973         }
1975         /*
1976          * Yes, the preimage is based on an older version that still
1977          * has whitespace breakages unfixed, and fixing them makes the
1978          * hunk match.  Update the context lines in the postimage.
1979          */
1980         update_pre_post_images(preimage, postimage,
1981                                fixed_buf, buf - fixed_buf, 0);
1982         return 1;
1984  unmatch_exit:
1985         free(fixed_buf);
1986         return 0;
1989 static int find_pos(struct image *img,
1990                     struct image *preimage,
1991                     struct image *postimage,
1992                     int line,
1993                     unsigned ws_rule,
1994                     int match_beginning, int match_end)
1996         int i;
1997         unsigned long backwards, forwards, try;
1998         int backwards_lno, forwards_lno, try_lno;
2000         if (preimage->nr > img->nr)
2001                 return -1;
2003         /*
2004          * If match_begining or match_end is specified, there is no
2005          * point starting from a wrong line that will never match and
2006          * wander around and wait for a match at the specified end.
2007          */
2008         if (match_beginning)
2009                 line = 0;
2010         else if (match_end)
2011                 line = img->nr - preimage->nr;
2013         if (line > img->nr)
2014                 line = img->nr;
2016         try = 0;
2017         for (i = 0; i < line; i++)
2018                 try += img->line[i].len;
2020         /*
2021          * There's probably some smart way to do this, but I'll leave
2022          * that to the smart and beautiful people. I'm simple and stupid.
2023          */
2024         backwards = try;
2025         backwards_lno = line;
2026         forwards = try;
2027         forwards_lno = line;
2028         try_lno = line;
2030         for (i = 0; ; i++) {
2031                 if (match_fragment(img, preimage, postimage,
2032                                    try, try_lno, ws_rule,
2033                                    match_beginning, match_end))
2034                         return try_lno;
2036         again:
2037                 if (backwards_lno == 0 && forwards_lno == img->nr)
2038                         break;
2040                 if (i & 1) {
2041                         if (backwards_lno == 0) {
2042                                 i++;
2043                                 goto again;
2044                         }
2045                         backwards_lno--;
2046                         backwards -= img->line[backwards_lno].len;
2047                         try = backwards;
2048                         try_lno = backwards_lno;
2049                 } else {
2050                         if (forwards_lno == img->nr) {
2051                                 i++;
2052                                 goto again;
2053                         }
2054                         forwards += img->line[forwards_lno].len;
2055                         forwards_lno++;
2056                         try = forwards;
2057                         try_lno = forwards_lno;
2058                 }
2060         }
2061         return -1;
2064 static void remove_first_line(struct image *img)
2066         img->buf += img->line[0].len;
2067         img->len -= img->line[0].len;
2068         img->line++;
2069         img->nr--;
2072 static void remove_last_line(struct image *img)
2074         img->len -= img->line[--img->nr].len;
2077 static void update_image(struct image *img,
2078                          int applied_pos,
2079                          struct image *preimage,
2080                          struct image *postimage)
2082         /*
2083          * remove the copy of preimage at offset in img
2084          * and replace it with postimage
2085          */
2086         int i, nr;
2087         size_t remove_count, insert_count, applied_at = 0;
2088         char *result;
2090         for (i = 0; i < applied_pos; i++)
2091                 applied_at += img->line[i].len;
2093         remove_count = 0;
2094         for (i = 0; i < preimage->nr; i++)
2095                 remove_count += img->line[applied_pos + i].len;
2096         insert_count = postimage->len;
2098         /* Adjust the contents */
2099         result = xmalloc(img->len + insert_count - remove_count + 1);
2100         memcpy(result, img->buf, applied_at);
2101         memcpy(result + applied_at, postimage->buf, postimage->len);
2102         memcpy(result + applied_at + postimage->len,
2103                img->buf + (applied_at + remove_count),
2104                img->len - (applied_at + remove_count));
2105         free(img->buf);
2106         img->buf = result;
2107         img->len += insert_count - remove_count;
2108         result[img->len] = '\0';
2110         /* Adjust the line table */
2111         nr = img->nr + postimage->nr - preimage->nr;
2112         if (preimage->nr < postimage->nr) {
2113                 /*
2114                  * NOTE: this knows that we never call remove_first_line()
2115                  * on anything other than pre/post image.
2116                  */
2117                 img->line = xrealloc(img->line, nr * sizeof(*img->line));
2118                 img->line_allocated = img->line;
2119         }
2120         if (preimage->nr != postimage->nr)
2121                 memmove(img->line + applied_pos + postimage->nr,
2122                         img->line + applied_pos + preimage->nr,
2123                         (img->nr - (applied_pos + preimage->nr)) *
2124                         sizeof(*img->line));
2125         memcpy(img->line + applied_pos,
2126                postimage->line,
2127                postimage->nr * sizeof(*img->line));
2128         img->nr = nr;
2131 static int apply_one_fragment(struct image *img, struct fragment *frag,
2132                               int inaccurate_eof, unsigned ws_rule)
2134         int match_beginning, match_end;
2135         const char *patch = frag->patch;
2136         int size = frag->size;
2137         char *old, *new, *oldlines, *newlines;
2138         int new_blank_lines_at_end = 0;
2139         unsigned long leading, trailing;
2140         int pos, applied_pos;
2141         struct image preimage;
2142         struct image postimage;
2144         memset(&preimage, 0, sizeof(preimage));
2145         memset(&postimage, 0, sizeof(postimage));
2146         oldlines = xmalloc(size);
2147         newlines = xmalloc(size);
2149         old = oldlines;
2150         new = newlines;
2151         while (size > 0) {
2152                 char first;
2153                 int len = linelen(patch, size);
2154                 int plen, added;
2155                 int added_blank_line = 0;
2156                 int is_blank_context = 0;
2158                 if (!len)
2159                         break;
2161                 /*
2162                  * "plen" is how much of the line we should use for
2163                  * the actual patch data. Normally we just remove the
2164                  * first character on the line, but if the line is
2165                  * followed by "\ No newline", then we also remove the
2166                  * last one (which is the newline, of course).
2167                  */
2168                 plen = len - 1;
2169                 if (len < size && patch[len] == '\\')
2170                         plen--;
2171                 first = *patch;
2172                 if (apply_in_reverse) {
2173                         if (first == '-')
2174                                 first = '+';
2175                         else if (first == '+')
2176                                 first = '-';
2177                 }
2179                 switch (first) {
2180                 case '\n':
2181                         /* Newer GNU diff, empty context line */
2182                         if (plen < 0)
2183                                 /* ... followed by '\No newline'; nothing */
2184                                 break;
2185                         *old++ = '\n';
2186                         *new++ = '\n';
2187                         add_line_info(&preimage, "\n", 1, LINE_COMMON);
2188                         add_line_info(&postimage, "\n", 1, LINE_COMMON);
2189                         is_blank_context = 1;
2190                         break;
2191                 case ' ':
2192                         if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2193                             ws_blank_line(patch + 1, plen, ws_rule))
2194                                 is_blank_context = 1;
2195                 case '-':
2196                         memcpy(old, patch + 1, plen);
2197                         add_line_info(&preimage, old, plen,
2198                                       (first == ' ' ? LINE_COMMON : 0));
2199                         old += plen;
2200                         if (first == '-')
2201                                 break;
2202                 /* Fall-through for ' ' */
2203                 case '+':
2204                         /* --no-add does not add new lines */
2205                         if (first == '+' && no_add)
2206                                 break;
2208                         if (first != '+' ||
2209                             !whitespace_error ||
2210                             ws_error_action != correct_ws_error) {
2211                                 memcpy(new, patch + 1, plen);
2212                                 added = plen;
2213                         }
2214                         else {
2215                                 added = ws_fix_copy(new, patch + 1, plen, ws_rule, &applied_after_fixing_ws);
2216                         }
2217                         add_line_info(&postimage, new, added,
2218                                       (first == '+' ? 0 : LINE_COMMON));
2219                         new += added;
2220                         if (first == '+' &&
2221                             (ws_rule & WS_BLANK_AT_EOF) &&
2222                             ws_blank_line(patch + 1, plen, ws_rule))
2223                                 added_blank_line = 1;
2224                         break;
2225                 case '@': case '\\':
2226                         /* Ignore it, we already handled it */
2227                         break;
2228                 default:
2229                         if (apply_verbosely)
2230                                 error("invalid start of line: '%c'", first);
2231                         return -1;
2232                 }
2233                 if (added_blank_line)
2234                         new_blank_lines_at_end++;
2235                 else if (is_blank_context)
2236                         ;
2237                 else
2238                         new_blank_lines_at_end = 0;
2239                 patch += len;
2240                 size -= len;
2241         }
2242         if (inaccurate_eof &&
2243             old > oldlines && old[-1] == '\n' &&
2244             new > newlines && new[-1] == '\n') {
2245                 old--;
2246                 new--;
2247         }
2249         leading = frag->leading;
2250         trailing = frag->trailing;
2252         /*
2253          * A hunk to change lines at the beginning would begin with
2254          * @@ -1,L +N,M @@
2255          * but we need to be careful.  -U0 that inserts before the second
2256          * line also has this pattern.
2257          *
2258          * And a hunk to add to an empty file would begin with
2259          * @@ -0,0 +N,M @@
2260          *
2261          * In other words, a hunk that is (frag->oldpos <= 1) with or
2262          * without leading context must match at the beginning.
2263          */
2264         match_beginning = (!frag->oldpos ||
2265                            (frag->oldpos == 1 && !unidiff_zero));
2267         /*
2268          * A hunk without trailing lines must match at the end.
2269          * However, we simply cannot tell if a hunk must match end
2270          * from the lack of trailing lines if the patch was generated
2271          * with unidiff without any context.
2272          */
2273         match_end = !unidiff_zero && !trailing;
2275         pos = frag->newpos ? (frag->newpos - 1) : 0;
2276         preimage.buf = oldlines;
2277         preimage.len = old - oldlines;
2278         postimage.buf = newlines;
2279         postimage.len = new - newlines;
2280         preimage.line = preimage.line_allocated;
2281         postimage.line = postimage.line_allocated;
2283         for (;;) {
2285                 applied_pos = find_pos(img, &preimage, &postimage, pos,
2286                                        ws_rule, match_beginning, match_end);
2288                 if (applied_pos >= 0)
2289                         break;
2291                 /* Am I at my context limits? */
2292                 if ((leading <= p_context) && (trailing <= p_context))
2293                         break;
2294                 if (match_beginning || match_end) {
2295                         match_beginning = match_end = 0;
2296                         continue;
2297                 }
2299                 /*
2300                  * Reduce the number of context lines; reduce both
2301                  * leading and trailing if they are equal otherwise
2302                  * just reduce the larger context.
2303                  */
2304                 if (leading >= trailing) {
2305                         remove_first_line(&preimage);
2306                         remove_first_line(&postimage);
2307                         pos--;
2308                         leading--;
2309                 }
2310                 if (trailing > leading) {
2311                         remove_last_line(&preimage);
2312                         remove_last_line(&postimage);
2313                         trailing--;
2314                 }
2315         }
2317         if (applied_pos >= 0) {
2318                 if (new_blank_lines_at_end &&
2319                     preimage.nr + applied_pos == img->nr &&
2320                     (ws_rule & WS_BLANK_AT_EOF) &&
2321                     ws_error_action != nowarn_ws_error) {
2322                         record_ws_error(WS_BLANK_AT_EOF, "+", 1, frag->linenr);
2323                         if (ws_error_action == correct_ws_error) {
2324                                 while (new_blank_lines_at_end--)
2325                                         remove_last_line(&postimage);
2326                         }
2327                         /*
2328                          * We would want to prevent write_out_results()
2329                          * from taking place in apply_patch() that follows
2330                          * the callchain led us here, which is:
2331                          * apply_patch->check_patch_list->check_patch->
2332                          * apply_data->apply_fragments->apply_one_fragment
2333                          */
2334                         if (ws_error_action == die_on_ws_error)
2335                                 apply = 0;
2336                 }
2338                 /*
2339                  * Warn if it was necessary to reduce the number
2340                  * of context lines.
2341                  */
2342                 if ((leading != frag->leading) ||
2343                     (trailing != frag->trailing))
2344                         fprintf(stderr, "Context reduced to (%ld/%ld)"
2345                                 " to apply fragment at %d\n",
2346                                 leading, trailing, applied_pos+1);
2347                 update_image(img, applied_pos, &preimage, &postimage);
2348         } else {
2349                 if (apply_verbosely)
2350                         error("while searching for:\n%.*s",
2351                               (int)(old - oldlines), oldlines);
2352         }
2354         free(oldlines);
2355         free(newlines);
2356         free(preimage.line_allocated);
2357         free(postimage.line_allocated);
2359         return (applied_pos < 0);
2362 static int apply_binary_fragment(struct image *img, struct patch *patch)
2364         struct fragment *fragment = patch->fragments;
2365         unsigned long len;
2366         void *dst;
2368         /* Binary patch is irreversible without the optional second hunk */
2369         if (apply_in_reverse) {
2370                 if (!fragment->next)
2371                         return error("cannot reverse-apply a binary patch "
2372                                      "without the reverse hunk to '%s'",
2373                                      patch->new_name
2374                                      ? patch->new_name : patch->old_name);
2375                 fragment = fragment->next;
2376         }
2377         switch (fragment->binary_patch_method) {
2378         case BINARY_DELTA_DEFLATED:
2379                 dst = patch_delta(img->buf, img->len, fragment->patch,
2380                                   fragment->size, &len);
2381                 if (!dst)
2382                         return -1;
2383                 clear_image(img);
2384                 img->buf = dst;
2385                 img->len = len;
2386                 return 0;
2387         case BINARY_LITERAL_DEFLATED:
2388                 clear_image(img);
2389                 img->len = fragment->size;
2390                 img->buf = xmalloc(img->len+1);
2391                 memcpy(img->buf, fragment->patch, img->len);
2392                 img->buf[img->len] = '\0';
2393                 return 0;
2394         }
2395         return -1;
2398 static int apply_binary(struct image *img, struct patch *patch)
2400         const char *name = patch->old_name ? patch->old_name : patch->new_name;
2401         unsigned char sha1[20];
2403         /*
2404          * For safety, we require patch index line to contain
2405          * full 40-byte textual SHA1 for old and new, at least for now.
2406          */
2407         if (strlen(patch->old_sha1_prefix) != 40 ||
2408             strlen(patch->new_sha1_prefix) != 40 ||
2409             get_sha1_hex(patch->old_sha1_prefix, sha1) ||
2410             get_sha1_hex(patch->new_sha1_prefix, sha1))
2411                 return error("cannot apply binary patch to '%s' "
2412                              "without full index line", name);
2414         if (patch->old_name) {
2415                 /*
2416                  * See if the old one matches what the patch
2417                  * applies to.
2418                  */
2419                 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2420                 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
2421                         return error("the patch applies to '%s' (%s), "
2422                                      "which does not match the "
2423                                      "current contents.",
2424                                      name, sha1_to_hex(sha1));
2425         }
2426         else {
2427                 /* Otherwise, the old one must be empty. */
2428                 if (img->len)
2429                         return error("the patch applies to an empty "
2430                                      "'%s' but it is not empty", name);
2431         }
2433         get_sha1_hex(patch->new_sha1_prefix, sha1);
2434         if (is_null_sha1(sha1)) {
2435                 clear_image(img);
2436                 return 0; /* deletion patch */
2437         }
2439         if (has_sha1_file(sha1)) {
2440                 /* We already have the postimage */
2441                 enum object_type type;
2442                 unsigned long size;
2443                 char *result;
2445                 result = read_sha1_file(sha1, &type, &size);
2446                 if (!result)
2447                         return error("the necessary postimage %s for "
2448                                      "'%s' cannot be read",
2449                                      patch->new_sha1_prefix, name);
2450                 clear_image(img);
2451                 img->buf = result;
2452                 img->len = size;
2453         } else {
2454                 /*
2455                  * We have verified buf matches the preimage;
2456                  * apply the patch data to it, which is stored
2457                  * in the patch->fragments->{patch,size}.
2458                  */
2459                 if (apply_binary_fragment(img, patch))
2460                         return error("binary patch does not apply to '%s'",
2461                                      name);
2463                 /* verify that the result matches */
2464                 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2465                 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
2466                         return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)",
2467                                 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
2468         }
2470         return 0;
2473 static int apply_fragments(struct image *img, struct patch *patch)
2475         struct fragment *frag = patch->fragments;
2476         const char *name = patch->old_name ? patch->old_name : patch->new_name;
2477         unsigned ws_rule = patch->ws_rule;
2478         unsigned inaccurate_eof = patch->inaccurate_eof;
2480         if (patch->is_binary)
2481                 return apply_binary(img, patch);
2483         while (frag) {
2484                 if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule)) {
2485                         error("patch failed: %s:%ld", name, frag->oldpos);
2486                         if (!apply_with_reject)
2487                                 return -1;
2488                         frag->rejected = 1;
2489                 }
2490                 frag = frag->next;
2491         }
2492         return 0;
2495 static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
2497         if (!ce)
2498                 return 0;
2500         if (S_ISGITLINK(ce->ce_mode)) {
2501                 strbuf_grow(buf, 100);
2502                 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(ce->sha1));
2503         } else {
2504                 enum object_type type;
2505                 unsigned long sz;
2506                 char *result;
2508                 result = read_sha1_file(ce->sha1, &type, &sz);
2509                 if (!result)
2510                         return -1;
2511                 /* XXX read_sha1_file NUL-terminates */
2512                 strbuf_attach(buf, result, sz, sz + 1);
2513         }
2514         return 0;
2517 static struct patch *in_fn_table(const char *name)
2519         struct string_list_item *item;
2521         if (name == NULL)
2522                 return NULL;
2524         item = string_list_lookup(name, &fn_table);
2525         if (item != NULL)
2526                 return (struct patch *)item->util;
2528         return NULL;
2531 /*
2532  * item->util in the filename table records the status of the path.
2533  * Usually it points at a patch (whose result records the contents
2534  * of it after applying it), but it could be PATH_WAS_DELETED for a
2535  * path that a previously applied patch has already removed.
2536  */
2537  #define PATH_TO_BE_DELETED ((struct patch *) -2)
2538 #define PATH_WAS_DELETED ((struct patch *) -1)
2540 static int to_be_deleted(struct patch *patch)
2542         return patch == PATH_TO_BE_DELETED;
2545 static int was_deleted(struct patch *patch)
2547         return patch == PATH_WAS_DELETED;
2550 static void add_to_fn_table(struct patch *patch)
2552         struct string_list_item *item;
2554         /*
2555          * Always add new_name unless patch is a deletion
2556          * This should cover the cases for normal diffs,
2557          * file creations and copies
2558          */
2559         if (patch->new_name != NULL) {
2560                 item = string_list_insert(patch->new_name, &fn_table);
2561                 item->util = patch;
2562         }
2564         /*
2565          * store a failure on rename/deletion cases because
2566          * later chunks shouldn't patch old names
2567          */
2568         if ((patch->new_name == NULL) || (patch->is_rename)) {
2569                 item = string_list_insert(patch->old_name, &fn_table);
2570                 item->util = PATH_WAS_DELETED;
2571         }
2574 static void prepare_fn_table(struct patch *patch)
2576         /*
2577          * store information about incoming file deletion
2578          */
2579         while (patch) {
2580                 if ((patch->new_name == NULL) || (patch->is_rename)) {
2581                         struct string_list_item *item;
2582                         item = string_list_insert(patch->old_name, &fn_table);
2583                         item->util = PATH_TO_BE_DELETED;
2584                 }
2585                 patch = patch->next;
2586         }
2589 static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
2591         struct strbuf buf = STRBUF_INIT;
2592         struct image image;
2593         size_t len;
2594         char *img;
2595         struct patch *tpatch;
2597         if (!(patch->is_copy || patch->is_rename) &&
2598             (tpatch = in_fn_table(patch->old_name)) != NULL && !to_be_deleted(tpatch)) {
2599                 if (was_deleted(tpatch)) {
2600                         return error("patch %s has been renamed/deleted",
2601                                 patch->old_name);
2602                 }
2603                 /* We have a patched copy in memory use that */
2604                 strbuf_add(&buf, tpatch->result, tpatch->resultsize);
2605         } else if (cached) {
2606                 if (read_file_or_gitlink(ce, &buf))
2607                         return error("read of %s failed", patch->old_name);
2608         } else if (patch->old_name) {
2609                 if (S_ISGITLINK(patch->old_mode)) {
2610                         if (ce) {
2611                                 read_file_or_gitlink(ce, &buf);
2612                         } else {
2613                                 /*
2614                                  * There is no way to apply subproject
2615                                  * patch without looking at the index.
2616                                  */
2617                                 patch->fragments = NULL;
2618                         }
2619                 } else {
2620                         if (read_old_data(st, patch->old_name, &buf))
2621                                 return error("read of %s failed", patch->old_name);
2622                 }
2623         }
2625         img = strbuf_detach(&buf, &len);
2626         prepare_image(&image, img, len, !patch->is_binary);
2628         if (apply_fragments(&image, patch) < 0)
2629                 return -1; /* note with --reject this succeeds. */
2630         patch->result = image.buf;
2631         patch->resultsize = image.len;
2632         add_to_fn_table(patch);
2633         free(image.line_allocated);
2635         if (0 < patch->is_delete && patch->resultsize)
2636                 return error("removal patch leaves file contents");
2638         return 0;
2641 static int check_to_create_blob(const char *new_name, int ok_if_exists)
2643         struct stat nst;
2644         if (!lstat(new_name, &nst)) {
2645                 if (S_ISDIR(nst.st_mode) || ok_if_exists)
2646                         return 0;
2647                 /*
2648                  * A leading component of new_name might be a symlink
2649                  * that is going to be removed with this patch, but
2650                  * still pointing at somewhere that has the path.
2651                  * In such a case, path "new_name" does not exist as
2652                  * far as git is concerned.
2653                  */
2654                 if (has_symlink_leading_path(new_name, strlen(new_name)))
2655                         return 0;
2657                 return error("%s: already exists in working directory", new_name);
2658         }
2659         else if ((errno != ENOENT) && (errno != ENOTDIR))
2660                 return error("%s: %s", new_name, strerror(errno));
2661         return 0;
2664 static int verify_index_match(struct cache_entry *ce, struct stat *st)
2666         if (S_ISGITLINK(ce->ce_mode)) {
2667                 if (!S_ISDIR(st->st_mode))
2668                         return -1;
2669                 return 0;
2670         }
2671         return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
2674 static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)
2676         const char *old_name = patch->old_name;
2677         struct patch *tpatch = NULL;
2678         int stat_ret = 0;
2679         unsigned st_mode = 0;
2681         /*
2682          * Make sure that we do not have local modifications from the
2683          * index when we are looking at the index.  Also make sure
2684          * we have the preimage file to be patched in the work tree,
2685          * unless --cached, which tells git to apply only in the index.
2686          */
2687         if (!old_name)
2688                 return 0;
2690         assert(patch->is_new <= 0);
2692         if (!(patch->is_copy || patch->is_rename) &&
2693             (tpatch = in_fn_table(old_name)) != NULL && !to_be_deleted(tpatch)) {
2694                 if (was_deleted(tpatch))
2695                         return error("%s: has been deleted/renamed", old_name);
2696                 st_mode = tpatch->new_mode;
2697         } else if (!cached) {
2698                 stat_ret = lstat(old_name, st);
2699                 if (stat_ret && errno != ENOENT)
2700                         return error("%s: %s", old_name, strerror(errno));
2701         }
2703         if (to_be_deleted(tpatch))
2704                 tpatch = NULL;
2706         if (check_index && !tpatch) {
2707                 int pos = cache_name_pos(old_name, strlen(old_name));
2708                 if (pos < 0) {
2709                         if (patch->is_new < 0)
2710                                 goto is_new;
2711                         return error("%s: does not exist in index", old_name);
2712                 }
2713                 *ce = active_cache[pos];
2714                 if (stat_ret < 0) {
2715                         struct checkout costate;
2716                         /* checkout */
2717                         costate.base_dir = "";
2718                         costate.base_dir_len = 0;
2719                         costate.force = 0;
2720                         costate.quiet = 0;
2721                         costate.not_new = 0;
2722                         costate.refresh_cache = 1;
2723                         if (checkout_entry(*ce, &costate, NULL) ||
2724                             lstat(old_name, st))
2725                                 return -1;
2726                 }
2727                 if (!cached && verify_index_match(*ce, st))
2728                         return error("%s: does not match index", old_name);
2729                 if (cached)
2730                         st_mode = (*ce)->ce_mode;
2731         } else if (stat_ret < 0) {
2732                 if (patch->is_new < 0)
2733                         goto is_new;
2734                 return error("%s: %s", old_name, strerror(errno));
2735         }
2737         if (!cached && !tpatch)
2738                 st_mode = ce_mode_from_stat(*ce, st->st_mode);
2740         if (patch->is_new < 0)
2741                 patch->is_new = 0;
2742         if (!patch->old_mode)
2743                 patch->old_mode = st_mode;
2744         if ((st_mode ^ patch->old_mode) & S_IFMT)
2745                 return error("%s: wrong type", old_name);
2746         if (st_mode != patch->old_mode)
2747                 warning("%s has type %o, expected %o",
2748                         old_name, st_mode, patch->old_mode);
2749         if (!patch->new_mode && !patch->is_delete)
2750                 patch->new_mode = st_mode;
2751         return 0;
2753  is_new:
2754         patch->is_new = 1;
2755         patch->is_delete = 0;
2756         patch->old_name = NULL;
2757         return 0;
2760 static int check_patch(struct patch *patch)
2762         struct stat st;
2763         const char *old_name = patch->old_name;
2764         const char *new_name = patch->new_name;
2765         const char *name = old_name ? old_name : new_name;
2766         struct cache_entry *ce = NULL;
2767         struct patch *tpatch;
2768         int ok_if_exists;
2769         int status;
2771         patch->rejected = 1; /* we will drop this after we succeed */
2773         status = check_preimage(patch, &ce, &st);
2774         if (status)
2775                 return status;
2776         old_name = patch->old_name;
2778         if ((tpatch = in_fn_table(new_name)) &&
2779                         (was_deleted(tpatch) || to_be_deleted(tpatch)))
2780                 /*
2781                  * A type-change diff is always split into a patch to
2782                  * delete old, immediately followed by a patch to
2783                  * create new (see diff.c::run_diff()); in such a case
2784                  * it is Ok that the entry to be deleted by the
2785                  * previous patch is still in the working tree and in
2786                  * the index.
2787                  */
2788                 ok_if_exists = 1;
2789         else
2790                 ok_if_exists = 0;
2792         if (new_name &&
2793             ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
2794                 if (check_index &&
2795                     cache_name_pos(new_name, strlen(new_name)) >= 0 &&
2796                     !ok_if_exists)
2797                         return error("%s: already exists in index", new_name);
2798                 if (!cached) {
2799                         int err = check_to_create_blob(new_name, ok_if_exists);
2800                         if (err)
2801                                 return err;
2802                 }
2803                 if (!patch->new_mode) {
2804                         if (0 < patch->is_new)
2805                                 patch->new_mode = S_IFREG | 0644;
2806                         else
2807                                 patch->new_mode = patch->old_mode;
2808                 }
2809         }
2811         if (new_name && old_name) {
2812                 int same = !strcmp(old_name, new_name);
2813                 if (!patch->new_mode)
2814                         patch->new_mode = patch->old_mode;
2815                 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
2816                         return error("new mode (%o) of %s does not match old mode (%o)%s%s",
2817                                 patch->new_mode, new_name, patch->old_mode,
2818                                 same ? "" : " of ", same ? "" : old_name);
2819         }
2821         if (apply_data(patch, &st, ce) < 0)
2822                 return error("%s: patch does not apply", name);
2823         patch->rejected = 0;
2824         return 0;
2827 static int check_patch_list(struct patch *patch)
2829         int err = 0;
2831         prepare_fn_table(patch);
2832         while (patch) {
2833                 if (apply_verbosely)
2834                         say_patch_name(stderr,
2835                                        "Checking patch ", patch, "...\n");
2836                 err |= check_patch(patch);
2837                 patch = patch->next;
2838         }
2839         return err;
2842 /* This function tries to read the sha1 from the current index */
2843 static int get_current_sha1(const char *path, unsigned char *sha1)
2845         int pos;
2847         if (read_cache() < 0)
2848                 return -1;
2849         pos = cache_name_pos(path, strlen(path));
2850         if (pos < 0)
2851                 return -1;
2852         hashcpy(sha1, active_cache[pos]->sha1);
2853         return 0;
2856 /* Build an index that contains the just the files needed for a 3way merge */
2857 static void build_fake_ancestor(struct patch *list, const char *filename)
2859         struct patch *patch;
2860         struct index_state result = { NULL };
2861         int fd;
2863         /* Once we start supporting the reverse patch, it may be
2864          * worth showing the new sha1 prefix, but until then...
2865          */
2866         for (patch = list; patch; patch = patch->next) {
2867                 const unsigned char *sha1_ptr;
2868                 unsigned char sha1[20];
2869                 struct cache_entry *ce;
2870                 const char *name;
2872                 name = patch->old_name ? patch->old_name : patch->new_name;
2873                 if (0 < patch->is_new)
2874                         continue;
2875                 else if (get_sha1(patch->old_sha1_prefix, sha1))
2876                         /* git diff has no index line for mode/type changes */
2877                         if (!patch->lines_added && !patch->lines_deleted) {
2878                                 if (get_current_sha1(patch->new_name, sha1) ||
2879                                     get_current_sha1(patch->old_name, sha1))
2880                                         die("mode change for %s, which is not "
2881                                                 "in current HEAD", name);
2882                                 sha1_ptr = sha1;
2883                         } else
2884                                 die("sha1 information is lacking or useless "
2885                                         "(%s).", name);
2886                 else
2887                         sha1_ptr = sha1;
2889                 ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);
2890                 if (!ce)
2891                         die("make_cache_entry failed for path '%s'", name);
2892                 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
2893                         die ("Could not add %s to temporary index", name);
2894         }
2896         fd = open(filename, O_WRONLY | O_CREAT, 0666);
2897         if (fd < 0 || write_index(&result, fd) || close(fd))
2898                 die ("Could not write temporary index to %s", filename);
2900         discard_index(&result);
2903 static void stat_patch_list(struct patch *patch)
2905         int files, adds, dels;
2907         for (files = adds = dels = 0 ; patch ; patch = patch->next) {
2908                 files++;
2909                 adds += patch->lines_added;
2910                 dels += patch->lines_deleted;
2911                 show_stats(patch);
2912         }
2914         printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
2917 static void numstat_patch_list(struct patch *patch)
2919         for ( ; patch; patch = patch->next) {
2920                 const char *name;
2921                 name = patch->new_name ? patch->new_name : patch->old_name;
2922                 if (patch->is_binary)
2923                         printf("-\t-\t");
2924                 else
2925                         printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
2926                 write_name_quoted(name, stdout, line_termination);
2927         }
2930 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
2932         if (mode)
2933                 printf(" %s mode %06o %s\n", newdelete, mode, name);
2934         else
2935                 printf(" %s %s\n", newdelete, name);
2938 static void show_mode_change(struct patch *p, int show_name)
2940         if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
2941                 if (show_name)
2942                         printf(" mode change %06o => %06o %s\n",
2943                                p->old_mode, p->new_mode, p->new_name);
2944                 else
2945                         printf(" mode change %06o => %06o\n",
2946                                p->old_mode, p->new_mode);
2947         }
2950 static void show_rename_copy(struct patch *p)
2952         const char *renamecopy = p->is_rename ? "rename" : "copy";
2953         const char *old, *new;
2955         /* Find common prefix */
2956         old = p->old_name;
2957         new = p->new_name;
2958         while (1) {
2959                 const char *slash_old, *slash_new;
2960                 slash_old = strchr(old, '/');
2961                 slash_new = strchr(new, '/');
2962                 if (!slash_old ||
2963                     !slash_new ||
2964                     slash_old - old != slash_new - new ||
2965                     memcmp(old, new, slash_new - new))
2966                         break;
2967                 old = slash_old + 1;
2968                 new = slash_new + 1;
2969         }
2970         /* p->old_name thru old is the common prefix, and old and new
2971          * through the end of names are renames
2972          */
2973         if (old != p->old_name)
2974                 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
2975                        (int)(old - p->old_name), p->old_name,
2976                        old, new, p->score);
2977         else
2978                 printf(" %s %s => %s (%d%%)\n", renamecopy,
2979                        p->old_name, p->new_name, p->score);
2980         show_mode_change(p, 0);
2983 static void summary_patch_list(struct patch *patch)
2985         struct patch *p;
2987         for (p = patch; p; p = p->next) {
2988                 if (p->is_new)
2989                         show_file_mode_name("create", p->new_mode, p->new_name);
2990                 else if (p->is_delete)
2991                         show_file_mode_name("delete", p->old_mode, p->old_name);
2992                 else {
2993                         if (p->is_rename || p->is_copy)
2994                                 show_rename_copy(p);
2995                         else {
2996                                 if (p->score) {
2997                                         printf(" rewrite %s (%d%%)\n",
2998                                                p->new_name, p->score);
2999                                         show_mode_change(p, 0);
3000                                 }
3001                                 else
3002                                         show_mode_change(p, 1);
3003                         }
3004                 }
3005         }
3008 static void patch_stats(struct patch *patch)
3010         int lines = patch->lines_added + patch->lines_deleted;
3012         if (lines > max_change)
3013                 max_change = lines;
3014         if (patch->old_name) {
3015                 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
3016                 if (!len)
3017                         len = strlen(patch->old_name);
3018                 if (len > max_len)
3019                         max_len = len;
3020         }
3021         if (patch->new_name) {
3022                 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
3023                 if (!len)
3024                         len = strlen(patch->new_name);
3025                 if (len > max_len)
3026                         max_len = len;
3027         }
3030 static void remove_file(struct patch *patch, int rmdir_empty)
3032         if (update_index) {
3033                 if (remove_file_from_cache(patch->old_name) < 0)
3034                         die("unable to remove %s from index", patch->old_name);
3035         }
3036         if (!cached) {
3037                 if (S_ISGITLINK(patch->old_mode)) {
3038                         if (rmdir(patch->old_name))
3039                                 warning("unable to remove submodule %s",
3040                                         patch->old_name);
3041                 } else if (!unlink_or_warn(patch->old_name) && rmdir_empty) {
3042                         remove_path(patch->old_name);
3043                 }
3044         }
3047 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
3049         struct stat st;
3050         struct cache_entry *ce;
3051         int namelen = strlen(path);
3052         unsigned ce_size = cache_entry_size(namelen);
3054         if (!update_index)
3055                 return;
3057         ce = xcalloc(1, ce_size);
3058         memcpy(ce->name, path, namelen);
3059         ce->ce_mode = create_ce_mode(mode);
3060         ce->ce_flags = namelen;
3061         if (S_ISGITLINK(mode)) {
3062                 const char *s = buf;
3064                 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
3065                         die("corrupt patch for subproject %s", path);
3066         } else {
3067                 if (!cached) {
3068                         if (lstat(path, &st) < 0)
3069                                 die_errno("unable to stat newly created file '%s'",
3070                                           path);
3071                         fill_stat_cache_info(ce, &st);
3072                 }
3073                 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
3074                         die("unable to create backing store for newly created file %s", path);
3075         }
3076         if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
3077                 die("unable to add cache entry for %s", path);
3080 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
3082         int fd;
3083         struct strbuf nbuf = STRBUF_INIT;
3085         if (S_ISGITLINK(mode)) {
3086                 struct stat st;
3087                 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
3088                         return 0;
3089                 return mkdir(path, 0777);
3090         }
3092         if (has_symlinks && S_ISLNK(mode))
3093                 /* Although buf:size is counted string, it also is NUL
3094                  * terminated.
3095                  */
3096                 return symlink(buf, path);
3098         fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
3099         if (fd < 0)
3100                 return -1;
3102         if (convert_to_working_tree(path, buf, size, &nbuf)) {
3103                 size = nbuf.len;
3104                 buf  = nbuf.buf;
3105         }
3106         write_or_die(fd, buf, size);
3107         strbuf_release(&nbuf);
3109         if (close(fd) < 0)
3110                 die_errno("closing file '%s'", path);
3111         return 0;
3114 /*
3115  * We optimistically assume that the directories exist,
3116  * which is true 99% of the time anyway. If they don't,
3117  * we create them and try again.
3118  */
3119 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
3121         if (cached)
3122                 return;
3123         if (!try_create_file(path, mode, buf, size))
3124                 return;
3126         if (errno == ENOENT) {
3127                 if (safe_create_leading_directories(path))
3128                         return;
3129                 if (!try_create_file(path, mode, buf, size))
3130                         return;
3131         }
3133         if (errno == EEXIST || errno == EACCES) {
3134                 /* We may be trying to create a file where a directory
3135                  * used to be.
3136                  */
3137                 struct stat st;
3138                 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
3139                         errno = EEXIST;
3140         }
3142         if (errno == EEXIST) {
3143                 unsigned int nr = getpid();
3145                 for (;;) {
3146                         char newpath[PATH_MAX];
3147                         mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
3148                         if (!try_create_file(newpath, mode, buf, size)) {
3149                                 if (!rename(newpath, path))
3150                                         return;
3151                                 unlink_or_warn(newpath);
3152                                 break;
3153                         }
3154                         if (errno != EEXIST)
3155                                 break;
3156                         ++nr;
3157                 }
3158         }
3159         die_errno("unable to write file '%s' mode %o", path, mode);
3162 static void create_file(struct patch *patch)
3164         char *path = patch->new_name;
3165         unsigned mode = patch->new_mode;
3166         unsigned long size = patch->resultsize;
3167         char *buf = patch->result;
3169         if (!mode)
3170                 mode = S_IFREG | 0644;
3171         create_one_file(path, mode, buf, size);
3172         add_index_file(path, mode, buf, size);
3175 /* phase zero is to remove, phase one is to create */
3176 static void write_out_one_result(struct patch *patch, int phase)
3178         if (patch->is_delete > 0) {
3179                 if (phase == 0)
3180                         remove_file(patch, 1);
3181                 return;
3182         }
3183         if (patch->is_new > 0 || patch->is_copy) {
3184                 if (phase == 1)
3185                         create_file(patch);
3186                 return;
3187         }
3188         /*
3189          * Rename or modification boils down to the same
3190          * thing: remove the old, write the new
3191          */
3192         if (phase == 0)
3193                 remove_file(patch, patch->is_rename);
3194         if (phase == 1)
3195                 create_file(patch);
3198 static int write_out_one_reject(struct patch *patch)
3200         FILE *rej;
3201         char namebuf[PATH_MAX];
3202         struct fragment *frag;
3203         int cnt = 0;
3205         for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
3206                 if (!frag->rejected)
3207                         continue;
3208                 cnt++;
3209         }
3211         if (!cnt) {
3212                 if (apply_verbosely)
3213                         say_patch_name(stderr,
3214                                        "Applied patch ", patch, " cleanly.\n");
3215                 return 0;
3216         }
3218         /* This should not happen, because a removal patch that leaves
3219          * contents are marked "rejected" at the patch level.
3220          */
3221         if (!patch->new_name)
3222                 die("internal error");
3224         /* Say this even without --verbose */
3225         say_patch_name(stderr, "Applying patch ", patch, " with");
3226         fprintf(stderr, " %d rejects...\n", cnt);
3228         cnt = strlen(patch->new_name);
3229         if (ARRAY_SIZE(namebuf) <= cnt + 5) {
3230                 cnt = ARRAY_SIZE(namebuf) - 5;
3231                 warning("truncating .rej filename to %.*s.rej",
3232                         cnt - 1, patch->new_name);
3233         }
3234         memcpy(namebuf, patch->new_name, cnt);
3235         memcpy(namebuf + cnt, ".rej", 5);
3237         rej = fopen(namebuf, "w");
3238         if (!rej)
3239                 return error("cannot open %s: %s", namebuf, strerror(errno));
3241         /* Normal git tools never deal with .rej, so do not pretend
3242          * this is a git patch by saying --git nor give extended
3243          * headers.  While at it, maybe please "kompare" that wants
3244          * the trailing TAB and some garbage at the end of line ;-).
3245          */
3246         fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
3247                 patch->new_name, patch->new_name);
3248         for (cnt = 1, frag = patch->fragments;
3249              frag;
3250              cnt++, frag = frag->next) {
3251                 if (!frag->rejected) {
3252                         fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);
3253                         continue;
3254                 }
3255                 fprintf(stderr, "Rejected hunk #%d.\n", cnt);
3256                 fprintf(rej, "%.*s", frag->size, frag->patch);
3257                 if (frag->patch[frag->size-1] != '\n')
3258                         fputc('\n', rej);
3259         }
3260         fclose(rej);
3261         return -1;
3264 static int write_out_results(struct patch *list, int skipped_patch)
3266         int phase;
3267         int errs = 0;
3268         struct patch *l;
3270         if (!list && !skipped_patch)
3271                 return error("No changes");
3273         for (phase = 0; phase < 2; phase++) {
3274                 l = list;
3275                 while (l) {
3276                         if (l->rejected)
3277                                 errs = 1;
3278                         else {
3279                                 write_out_one_result(l, phase);
3280                                 if (phase == 1 && write_out_one_reject(l))
3281                                         errs = 1;
3282                         }
3283                         l = l->next;
3284                 }
3285         }
3286         return errs;
3289 static struct lock_file lock_file;
3291 static struct string_list limit_by_name;
3292 static int has_include;
3293 static void add_name_limit(const char *name, int exclude)
3295         struct string_list_item *it;
3297         it = string_list_append(name, &limit_by_name);
3298         it->util = exclude ? NULL : (void *) 1;
3301 static int use_patch(struct patch *p)
3303         const char *pathname = p->new_name ? p->new_name : p->old_name;
3304         int i;
3306         /* Paths outside are not touched regardless of "--include" */
3307         if (0 < prefix_length) {
3308                 int pathlen = strlen(pathname);
3309                 if (pathlen <= prefix_length ||
3310                     memcmp(prefix, pathname, prefix_length))
3311                         return 0;
3312         }
3314         /* See if it matches any of exclude/include rule */
3315         for (i = 0; i < limit_by_name.nr; i++) {
3316                 struct string_list_item *it = &limit_by_name.items[i];
3317                 if (!fnmatch(it->string, pathname, 0))
3318                         return (it->util != NULL);
3319         }
3321         /*
3322          * If we had any include, a path that does not match any rule is
3323          * not used.  Otherwise, we saw bunch of exclude rules (or none)
3324          * and such a path is used.
3325          */
3326         return !has_include;
3330 static void prefix_one(char **name)
3332         char *old_name = *name;
3333         if (!old_name)
3334                 return;
3335         *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
3336         free(old_name);
3339 static void prefix_patches(struct patch *p)
3341         if (!prefix || p->is_toplevel_relative)
3342                 return;
3343         for ( ; p; p = p->next) {
3344                 if (p->new_name == p->old_name) {
3345                         char *prefixed = p->new_name;
3346                         prefix_one(&prefixed);
3347                         p->new_name = p->old_name = prefixed;
3348                 }
3349                 else {
3350                         prefix_one(&p->new_name);
3351                         prefix_one(&p->old_name);
3352                 }
3353         }
3356 #define INACCURATE_EOF  (1<<0)
3357 #define RECOUNT         (1<<1)
3359 static int apply_patch(int fd, const char *filename, int options)
3361         size_t offset;
3362         struct strbuf buf = STRBUF_INIT;
3363         struct patch *list = NULL, **listp = &list;
3364         int skipped_patch = 0;
3366         /* FIXME - memory leak when using multiple patch files as inputs */
3367         memset(&fn_table, 0, sizeof(struct string_list));
3368         patch_input_file = filename;
3369         read_patch_file(&buf, fd);
3370         offset = 0;
3371         while (offset < buf.len) {
3372                 struct patch *patch;
3373                 int nr;
3375                 patch = xcalloc(1, sizeof(*patch));
3376                 patch->inaccurate_eof = !!(options & INACCURATE_EOF);
3377                 patch->recount =  !!(options & RECOUNT);
3378                 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
3379                 if (nr < 0)
3380                         break;
3381                 if (apply_in_reverse)
3382                         reverse_patches(patch);
3383                 if (prefix)
3384                         prefix_patches(patch);
3385                 if (use_patch(patch)) {
3386                         patch_stats(patch);
3387                         *listp = patch;
3388                         listp = &patch->next;
3389                 }
3390                 else {
3391                         /* perhaps free it a bit better? */
3392                         free(patch);
3393                         skipped_patch++;
3394                 }
3395                 offset += nr;
3396         }
3398         if (whitespace_error && (ws_error_action == die_on_ws_error))
3399                 apply = 0;
3401         update_index = check_index && apply;
3402         if (update_index && newfd < 0)
3403                 newfd = hold_locked_index(&lock_file, 1);
3405         if (check_index) {
3406                 if (read_cache() < 0)
3407                         die("unable to read index file");
3408         }
3410         if ((check || apply) &&
3411             check_patch_list(list) < 0 &&
3412             !apply_with_reject)
3413                 exit(1);
3415         if (apply && write_out_results(list, skipped_patch))
3416                 exit(1);
3418         if (fake_ancestor)
3419                 build_fake_ancestor(list, fake_ancestor);
3421         if (diffstat)
3422                 stat_patch_list(list);
3424         if (numstat)
3425                 numstat_patch_list(list);
3427         if (summary)
3428                 summary_patch_list(list);
3430         strbuf_release(&buf);
3431         return 0;
3434 static int git_apply_config(const char *var, const char *value, void *cb)
3436         if (!strcmp(var, "apply.whitespace"))
3437                 return git_config_string(&apply_default_whitespace, var, value);
3438         else if (!strcmp(var, "apply.ignorewhitespace"))
3439                 return git_config_string(&apply_default_ignorewhitespace, var, value);
3440         return git_default_config(var, value, cb);
3443 static int option_parse_exclude(const struct option *opt,
3444                                 const char *arg, int unset)
3446         add_name_limit(arg, 1);
3447         return 0;
3450 static int option_parse_include(const struct option *opt,
3451                                 const char *arg, int unset)
3453         add_name_limit(arg, 0);
3454         has_include = 1;
3455         return 0;
3458 static int option_parse_p(const struct option *opt,
3459                           const char *arg, int unset)
3461         p_value = atoi(arg);
3462         p_value_known = 1;
3463         return 0;
3466 static int option_parse_z(const struct option *opt,
3467                           const char *arg, int unset)
3469         if (unset)
3470                 line_termination = '\n';
3471         else
3472                 line_termination = 0;
3473         return 0;
3476 static int option_parse_space_change(const struct option *opt,
3477                           const char *arg, int unset)
3479         if (unset)
3480                 ws_ignore_action = ignore_ws_none;
3481         else
3482                 ws_ignore_action = ignore_ws_change;
3483         return 0;
3486 static int option_parse_whitespace(const struct option *opt,
3487                                    const char *arg, int unset)
3489         const char **whitespace_option = opt->value;
3491         *whitespace_option = arg;
3492         parse_whitespace_option(arg);
3493         return 0;
3496 static int option_parse_directory(const struct option *opt,
3497                                   const char *arg, int unset)
3499         root_len = strlen(arg);
3500         if (root_len && arg[root_len - 1] != '/') {
3501                 char *new_root;
3502                 root = new_root = xmalloc(root_len + 2);
3503                 strcpy(new_root, arg);
3504                 strcpy(new_root + root_len++, "/");
3505         } else
3506                 root = arg;
3507         return 0;
3510 int cmd_apply(int argc, const char **argv, const char *unused_prefix)
3512         int i;
3513         int errs = 0;
3514         int is_not_gitdir;
3515         int binary;
3516         int force_apply = 0;
3518         const char *whitespace_option = NULL;
3520         struct option builtin_apply_options[] = {
3521                 { OPTION_CALLBACK, 0, "exclude", NULL, "path",
3522                         "don't apply changes matching the given path",
3523                         0, option_parse_exclude },
3524                 { OPTION_CALLBACK, 0, "include", NULL, "path",
3525                         "apply changes matching the given path",
3526                         0, option_parse_include },
3527                 { OPTION_CALLBACK, 'p', NULL, NULL, "num",
3528                         "remove <num> leading slashes from traditional diff paths",
3529                         0, option_parse_p },
3530                 OPT_BOOLEAN(0, "no-add", &no_add,
3531                         "ignore additions made by the patch"),
3532                 OPT_BOOLEAN(0, "stat", &diffstat,
3533                         "instead of applying the patch, output diffstat for the input"),
3534                 { OPTION_BOOLEAN, 0, "allow-binary-replacement", &binary,
3535                   NULL, "old option, now no-op",
3536                   PARSE_OPT_HIDDEN | PARSE_OPT_NOARG },
3537                 { OPTION_BOOLEAN, 0, "binary", &binary,
3538                   NULL, "old option, now no-op",
3539                   PARSE_OPT_HIDDEN | PARSE_OPT_NOARG },
3540                 OPT_BOOLEAN(0, "numstat", &numstat,
3541                         "shows number of added and deleted lines in decimal notation"),
3542                 OPT_BOOLEAN(0, "summary", &summary,
3543                         "instead of applying the patch, output a summary for the input"),
3544                 OPT_BOOLEAN(0, "check", &check,
3545                         "instead of applying the patch, see if the patch is applicable"),
3546                 OPT_BOOLEAN(0, "index", &check_index,
3547                         "make sure the patch is applicable to the current index"),
3548                 OPT_BOOLEAN(0, "cached", &cached,
3549                         "apply a patch without touching the working tree"),
3550                 OPT_BOOLEAN(0, "apply", &force_apply,
3551                         "also apply the patch (use with --stat/--summary/--check)"),
3552                 OPT_FILENAME(0, "build-fake-ancestor", &fake_ancestor,
3553                         "build a temporary index based on embedded index information"),
3554                 { OPTION_CALLBACK, 'z', NULL, NULL, NULL,
3555                         "paths are separated with NUL character",
3556                         PARSE_OPT_NOARG, option_parse_z },
3557                 OPT_INTEGER('C', NULL, &p_context,
3558                                 "ensure at least <n> lines of context match"),
3559                 { OPTION_CALLBACK, 0, "whitespace", &whitespace_option, "action",
3560                         "detect new or modified lines that have whitespace errors",
3561                         0, option_parse_whitespace },
3562                 { OPTION_CALLBACK, 0, "ignore-space-change", NULL, NULL,
3563                         "ignore changes in whitespace when finding context",
3564                         PARSE_OPT_NOARG, option_parse_space_change },
3565                 { OPTION_CALLBACK, 0, "ignore-whitespace", NULL, NULL,
3566                         "ignore changes in whitespace when finding context",
3567                         PARSE_OPT_NOARG, option_parse_space_change },
3568                 OPT_BOOLEAN('R', "reverse", &apply_in_reverse,
3569                         "apply the patch in reverse"),
3570                 OPT_BOOLEAN(0, "unidiff-zero", &unidiff_zero,
3571                         "don't expect at least one line of context"),
3572                 OPT_BOOLEAN(0, "reject", &apply_with_reject,
3573                         "leave the rejected hunks in corresponding *.rej files"),
3574                 OPT__VERBOSE(&apply_verbosely),
3575                 OPT_BIT(0, "inaccurate-eof", &options,
3576                         "tolerate incorrectly detected missing new-line at the end of file",
3577                         INACCURATE_EOF),
3578                 OPT_BIT(0, "recount", &options,
3579                         "do not trust the line counts in the hunk headers",
3580                         RECOUNT),
3581                 { OPTION_CALLBACK, 0, "directory", NULL, "root",
3582                         "prepend <root> to all filenames",
3583                         0, option_parse_directory },
3584                 OPT_END()
3585         };
3587         prefix = setup_git_directory_gently(&is_not_gitdir);
3588         prefix_length = prefix ? strlen(prefix) : 0;
3589         git_config(git_apply_config, NULL);
3590         if (apply_default_whitespace)
3591                 parse_whitespace_option(apply_default_whitespace);
3592         if (apply_default_ignorewhitespace)
3593                 parse_ignorewhitespace_option(apply_default_ignorewhitespace);
3595         argc = parse_options(argc, argv, prefix, builtin_apply_options,
3596                         apply_usage, 0);
3598         if (apply_with_reject)
3599                 apply = apply_verbosely = 1;
3600         if (!force_apply && (diffstat || numstat || summary || check || fake_ancestor))
3601                 apply = 0;
3602         if (check_index && is_not_gitdir)
3603                 die("--index outside a repository");
3604         if (cached) {
3605                 if (is_not_gitdir)
3606                         die("--cached outside a repository");
3607                 check_index = 1;
3608         }
3609         for (i = 0; i < argc; i++) {
3610                 const char *arg = argv[i];
3611                 int fd;
3613                 if (!strcmp(arg, "-")) {
3614                         errs |= apply_patch(0, "<stdin>", options);
3615                         read_stdin = 0;
3616                         continue;
3617                 } else if (0 < prefix_length)
3618                         arg = prefix_filename(prefix, prefix_length, arg);
3620                 fd = open(arg, O_RDONLY);
3621                 if (fd < 0)
3622                         die_errno("can't open patch '%s'", arg);
3623                 read_stdin = 0;
3624                 set_default_whitespace_mode(whitespace_option);
3625                 errs |= apply_patch(fd, arg, options);
3626                 close(fd);
3627         }
3628         set_default_whitespace_mode(whitespace_option);
3629         if (read_stdin)
3630                 errs |= apply_patch(0, "<stdin>", options);
3631         if (whitespace_error) {
3632                 if (squelch_whitespace_errors &&
3633                     squelch_whitespace_errors < whitespace_error) {
3634                         int squelched =
3635                                 whitespace_error - squelch_whitespace_errors;
3636                         warning("squelched %d "
3637                                 "whitespace error%s",
3638                                 squelched,
3639                                 squelched == 1 ? "" : "s");
3640                 }
3641                 if (ws_error_action == die_on_ws_error)
3642                         die("%d line%s add%s whitespace errors.",
3643                             whitespace_error,
3644                             whitespace_error == 1 ? "" : "s",
3645                             whitespace_error == 1 ? "s" : "");
3646                 if (applied_after_fixing_ws && apply)
3647                         warning("%d line%s applied after"
3648                                 " fixing whitespace errors.",
3649                                 applied_after_fixing_ws,
3650                                 applied_after_fixing_ws == 1 ? "" : "s");
3651                 else if (whitespace_error)
3652                         warning("%d line%s add%s whitespace errors.",
3653                                 whitespace_error,
3654                                 whitespace_error == 1 ? "" : "s",
3655                                 whitespace_error == 1 ? "s" : "");
3656         }
3658         if (update_index) {
3659                 if (write_cache(newfd, active_cache, active_nr) ||
3660                     commit_locked_index(&lock_file))
3661                         die("Unable to write new index file");
3662         }
3664         return !!errs;